0% found this document useful (0 votes)
90 views132 pagesKanti Wali Book Aadi Adhuri 3,4,5,10,11,17,21,27
Uploaded by
AkhilCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
90 views132 pagesKanti Wali Book Aadi Adhuri 3,4,5,10,11,17,21,27
Uploaded by
AkhilCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 132
4
PREREQUISITE
012. Do the following vectors form a basis for RY?
@ 420.02) m= 17009, — ap= (12)
@ a= (10, 9 = BOM = G21)
(ay = (57h 92 = 14,06) 3 = {1.0.0}.
0:12. HYPERPLANES AND HYPERSPHERES
ining two points % = (x,""
Definition 1 (Line segment in R®). The line segment jo
and X_, = (%% XY evoy XA), is the collection of all points X = (Xp Xzy +++ Xy), Where
x= Ax + (1-A)x2 for all values of d such that 0 SS 1.
Definition 2 (Hyperplane). In R', the set of points X = (Sy Xy --~ x,) satisfying
cya + na tone + Sahn = Z
defines a hyperplane for given values of ¢;’s and 2.
Remarks 1. If z = 0, then the hyperplane is said to pass through the ori
written as
{not all cj = 0)
eit + pty tone t Gay 20 OF Ox =O.
‘This implies that ¢ is orthogonal to every vector x in the hyperplane.
2. If z # 0 and P (xj), Q(X») are any two points on the hyperplane, then
ex =z and oy
This yields e (&1 ~ X,) = 0. which implies that ¢ is normal to the vector Pi.
in the direction of the normal to the hyperplane.
4X2 — yy and is thus
In general, ¢ is called the normal to the hyperplane and + rl are called unit normals.
10 be parallel, if they have the
Definition 3 (Parallel hyperplanes). Two hyperplanes are said t
jarallel, if c= Ae for some
same unit normals. The hyperplanes ¢X = 2, and ©X = 2 are pi
a#0.
Definition 4 (Closed and Open half space). In R’, every hyperplane ex = z determines 1wo pen
half spaces and two closed half spaces. The sets defined by
He = {x:ex > z) and H2 = (xe 0 is defined
to be the set of points
S$ ={x:|x-al=e}.
and
Thus, the equation for hypersphere in R" is
Scanned with CamScanner 2 OPERATIONS RESEARCH
0:13. CONVEX SETS AND THEIR PROPERTIES
Definition 1 (Convex s A subset S © R", is said to be convex, if for any two points X,, x, in §
the line segment joining the points x, and x, is also contained in S.
In other words, a subset SC R" is conver,
if and only if
MAZE S => Ax +(1-AxQE S: O0 and xy x © R", is a convex
set.
Definition 2 (Extreme point or Vertex of a convex set). An extreme point (vertex) of a conver
set is a point of the set which does not lie on any segment joining two other points of the set
Scanned with CamScanner 13
PREREQUISITE
Thus, a point x of a convex set S is an extreme point of the set. if there does not exist any pa
of points x). 3, in S, such that pair
x= Am + (1 - Ax o 0, then AAx > 0 for non-neg
a (2 ) m
4 . 2 Oand Ax > 0, then A [Ax + (1 — 2) x0) > 0.
efinition ; i
oh ei ee pee ones hull of a set). Give @ subset Y CR’, the smallest convex set containing
i : convex hull of Y. and is denoted by .
can eas , . ; ;
‘an cusily be seen that the convex hull of a set is the intersection of all convex sets containing Y.
Example. Let A = (Xp x}. Then the line segment joining x), xz is a convex set, Also if S is
any convex set containing A, then it must contain the line segment joining x, and x.
os = {xix = Ax) +(1-A)x, O be its convex hull, Let S be the set of
all convex combinations of vectors in A.
‘Then, we know that $ is a convex set containing A; and hence c S.
Again contains A. We shall show that this implies that > S. We use finite induction
on &, the number of vectors in A, to prove this result.
For k = 2, let x), X) be two vectors in A. Then < A> coni
; x = Ax, + (1 -A)x, osast
empty subset CCR", is called a cone if for each x in C, and
fying Ax > 0 is a convex
1c A. It is convex because
SX. Xp3 and is a convex set.
is also in .
. All possible convex combinations of x), x, are in .
‘Assume that, for any positive integer k, the result is true for a set of at most & ~ 1 vectors.
Xp Xoy vs Xe Then is a convex set containing A, and in
Consider now the set A
particular { xj, Xp e+ Xo1 }- Let oo .
=! :
‘Saas w20 Zu= th
X= {x[x
Thus, by induction hypothesis > X1-
Now contains X,, as well as { : }- "
all segments joining x, to 2 point of Xy
fet
Bx, + = 8) Em
vw Conti
or
is a point in . .
Now
where B; = (1 - 8)By for
Scanned with CamScanner PREREQUISITE, ‘5
since, pj 2 0 for each i, and 0S 5S 1, + B20 forj= 1,2...
fl
Also +0 - 8) Ew
= x= £ Bx, is a convex combination of the vectors x), Xp... X and is in . ie.
2 5.
We conclude that =S.
In view of the above theorem, the convex polyhedron may equivalently be defined as the convex
hull of a set point S. consisting of a finite number of points in R".
Definition 7 (Simplex). A simplex in n-dimension is @ convex polyhedron having exactly (n + 1)
vertices.
A simplex in zero
itis a triangle, in three
ension is a point, in one dimension itis a fine segment, in two
a tetrahydron, and so on.
0:14. SUPPORTING AND SEPARATING HYPERPLANES
Definition (Supporting hyperplane). Let § CR" be any closed convex set and w € S be a
boundary point. Then a hyperplane cx = z is called a supporting hyperplane of S at w, if
(@) ew=z and (i) SCH, or SCH
where H, = {x :¢ex2z} and H_={x: cx Sz}.
Remarks I. The supporting hyperplane need not be ui
2. S may intersect the supporting hyperplane in more thay one boundary points.
Theorem 0-5. Let S be a closed. convex set wh ii bounded from below. Then Shas extreme
points in every supporting hyperplane.
Provf. Let w be a boundary point of a closed convex set S.
Let ex = z be a supporting hyperplane at we S. Let B= S..{x]ex =
Then B is a closed convex set and B.# @, for w ¢ B.
We claim that every extreme point of B is also an extreme point of S.
Let us assume to the contrary that an extreme point b of B, is not an extreme point of S. Then
there exist x, x5 € S, such that
bean +(-Au O lw -y) +A@-wP21W-¥P
a 22 |u ~ wi? + |w - yl? + 220W yu w) 2 [wo ?
=> 22 w- wl? + 2k - y)(u - w) 20.
Letting 4 0, and ¢ = (w — y) such that |e] # 0; we get
(w-y)(@-w)20 or cu-w 20
cr e@-y) 2 ew -y) = lel?
a cu >'ey.
Setting cy = z, we find that cu > z
Thus y lies on the hyperplane ex = z and for all u'€ S, ev > z.
This completes the proof.
PROBLEMS
013. Given the hyperplane 3x; + 2%) + 4xy + 6ry = 7.
= {6, 1. 7. 2}.
O14. Prove that the set x = { On 2) [ar +
015. Examine whether the following set is convex or not :
GS = (Gp 9) 2 xy = 1x 20 20}
Gi) S = {Gy 9) ¢ Sey + Bey 2 10, ey + Sey 2 10}
(i) S={ Gy. atta Shan tyet}
016. Show that $= ( (xj. X33): 2ty — t+ ay $4} © RBis a conver set.
Find in which half space is the point
* <4} is a convex set.
017. Show that in R3, the closed ball xj? + 13? + 4? $ 1 is a convex set.
018, Let $= {0 ) 1-2 S41 + e-3-m SB)
Find all the extreme points of S and represent x = (2, 1) as convex combination of the extreme points
019. Determine the convex hull of the following seis :
(A= Cy alae tar = UP
(i) A= {Gn}
Scanned with CamScanner PREREQUISITE
17
020, Graph the convex hull of the points
(0, Oy ON. Cy 2) (Le 1, 4. 0).
Which of these points is an interior point of the convex hull? Express it as a convex combination of
mbination of the
extreme points.
0:15. CONVEX FUNCTIONS
Definition 1 (Convex function). Let S be
‘said to be convex if for any two vectors x, and Xz in S
Fx, + (1 = Ax) SAFO) + (1 - ASO), OS AST
Definition 2 (Strictly convex function). Let $ be a non-empty convex subset of R". A function
f(s) on S is said to be strictly convex if for any two different vectors X, and Xin s
fldx, + (1 — Am) < Af) + (1 - Df) O x,, but no
Fig. 0.7
ple, f(x) = x is such a function
t strictly convex for x < X-
Scanned with CamScanner >,
7 OPERATIONS RESEARCH
The following results are the
(@ A linear function z = ¢
(concave).
Gi) The sum of convex
is strictly convex (,
70%e; In what follows we shall deal with convex functions only. However, all the results remain valid
if we deal with concave funetions ins
immediate consequences of the above definitions :
XX © R" is a convex (concave) function but not strictly convey
(concave) functions is convex (concave) and if at least one of the functions
concave) then so is their sum.
316. LOCAL AND GLOBAL EXTREMA
In the problems of constrained o
0 Ptimization, one is interested in determining a vector x that minimises
the function f(x) fo
¥ maximises —F(x)] subject to the ‘constraints’ g,(x) < 0 (i +). The
set of the vectors x satisfying these constraints is usually called the ‘feasible region’. These terms
Shall be defined in a more formal way later
Definition 1 (Global minima
%o Hf F(%) S F(X) for all x in the
Example. Function f(x) = x,2, Subject to the constraint x, 2 0, has a minimum at x; = 0.
Definition 2 (Local minima). A local minimum S(%) of function f(x) is said to be attained at
Xo if there exists a positive € such that f(%,) < f(x) for all x in the feasible region which also satisfy
the condition |x, ~ x| < ¢.
. A global minimum of the function f(x) is said t0 be attained at
feasible region,
Example. The function f(x) = x;
~ 21° subject to the constraint x, 2 0, has a local minimum at
(with €
2/3). Note that f(x) has no global minimum at all.
Note. The word extremum (plural extrema) is used to indicate either maximum or minimum.
Theorem 0-7. Let f(x) be a convex function of a convex set S. If ‘£(%) has @ local minimum on S,
then this local minimum is also a global minimum on S.
Proof. Let a local minimum f(x,) be attained at x,. Then, there exists at least one x in
Sx, # x) such that f(x) < f(x). Since, f(x) is a convex function on S, we have
Flhxy + (I~ A) SAFO) +. — FE)
Also APO) + (1 DF) < AFD + 2) F05) = F(R)
‘Thus FIAX, + (1 = 2) x) Sf).
Now, for any € > 0, we observe that
Ix, + ~2)x61 - 61 = Alay = gL <8,
50 long as 4 < € < |x; ~ X5["' and obviously such a 2 < I does exist whenever € < |x,
Thus, Ax, + (1 — A)x, will give a smaller value for f(x)
2. is so chosen as desired above, This contradicts the fact that
Hence x, is a global minima.
1. If a function f(x) has a local minimum on a convex set S on wi
enthetocal is is also a global minimum on that set.
single point.
Proof. Exercise for the reader.
tion on a convex set $, Then th
1m 0-8, Let f(x) be a convex func
wich 0 takes on its global minimum, is a convex set,
= xh
the €-neigbourhood of x,, whenever
(%) takes on a local minimum at x,.
‘hich it is strictly convex,
This global minimum is attained at a
le Set of points in Sat
Scanned with CamScanner pREREQUISITE 19
ous if the global-minimum is attained at just a single point. Let us
Proof. The result is obvi s
‘mum is attained at two different points x, and x; of S. Then
assume that the global mini
fx) = £02
Now, since f(x) is convex,
flax, + (1 Am S ALD + (L- AFOW) = FO) Bern
slaxy + (1 = WR] SSO) = FO.
flaxz + (= Ax) S FO).
corresponds to a global minima, The set of all such x is,
>
Thus, every point x = Ax; + (I - 2)x
obviously. a convex set.
Corollary. If the global minimum is attainable at vo different points
an infinite number of points of S.
Proof. Exercise for the reader.
Theorem 0-9. Let f(x) be differentiable in its domain. If f(s) és defined on an open convex set S,
then f(x) is convex if and only if
4x2)
of S, then it iy attainable at
Uy) 2 (= IT VIED
for all Xj) € 5.
Proof. (Sufficiency). Here, we shall prove that if
(83) ~ fH) 2 Oy ~ MIT VLC)
then f(x) is convex.
Since xy X € SX = AM + (1 -A)x, fros ASI implies that x, € S.
Using the above condition for x, and X,, we have
F(0) = f(%) 2 OH = HIT VIC). oli)
larly, for xp and Xp
$3) ~ fe) B (8 ~ Xo) VLC). wlll)
Multiplying (ii) by 2 and (i) by (1 ~ A) and then adding, we get
GOs) + 1 -— WLOa) 2 [O%0) 2 LO) + Tax? + (1 = MT] VS(8) ~ Xo" VSO)
Using the definition of x, this yields
‘AS (xa) + (1 = ASO) 2 fxg + = 2)
which implies that f(x) is convex.
(Necessity). We now prove that if f(x) is convex, we must have
Fo) ~ £41) 2 y= AT VSO)
For 0 < 2 <1, we have for x1, x; € S that
A(x) + CL A)S() 2 Sm + (I - 2)
he, Df xa) — fem) 2 fly + (I - 2m) - S04)
bs Sx) = Fey) = Lf fxg + (1 = ar} - FOV.
‘Takidlg the limit of the right hand side as 20, we obtain
S42) ~ $4) 2 (2 - x07 VFO).
This completes the proof.
Theorem 0-10. Let " i i i
2 Temeem 010 LI 36 Bie nea
nn (x = x) VF (%q) 2 0.
f. (Necessary condition). Let f(x) be a conve)
such that f(x,) $ f(x) for all x € S. ee S is open, ote SS arcase aa
Scanned with CamScanner 20 OPERATIONS RESE ARG,
~ Fo) S fIAX + (1 - AX)
| for all x in § and
Osasi
For A > 0,
£l% + A(x - x)] = F(%) 20
Taking limit as 2 approaches zero, we get ‘
7 ea for all xin S, by converity of f(x), ye
FO) ~ F(a) 2 (x ~ x) VF (%) 2 0.
in and x, minimizes f(x) over S.
Scometric point of view, the above theorem states that x, is a minimum point if the
© vectors V F(x.) and (x — X0)" for all x in S is less than or equal to n/2.
Hence, f(x) > 5 (XQ) for all x
Remark. From a
angle between th
0:17, QUADRATIC FORMS
Definition 1 (Quadratic
form). Ler A =
matrix x = (x, x,
so UA), the function
(a) be an n-rowed square matrix. Then given an nx )
Example. The quadratic form
Q& Sah + Qn? — Ie? - ane + Gxx3 - Sex,
a? — Tay? — arya
1-2 3]fa
fxn a-2 2 -srly |,
30-52 -7 Jl
Note, The matrix A can always be assumed symmetic, since each element of every pair of coefficients
a4 and ay (i # 3) can be replaced by (ay + aj)/2 without changing the value of Q (x),
can be written as
Properties of a Quadratic Form
(© Positive definite, The quadratic form Q (x) is positive definite if and only if Q(x) > 0 (po
)
for all x # 0, ‘ oe
For example, Q(x) = x? + 5x? + Tx? is Positive definite as it is Positive for all values of
yy ay Ay except when x, = x = x = 0.
in oe pa ee ly it
ii) Positi definite, The quadratic form Q(x) is positive semi-definite if and on
ane eee are for all x and there exist at least one non-zero vector x # 0 for which
Qm@=0, ¢
For example, Q(x) = (x) ~ =F + 2x; is positive semi-definite as it is zero for 4 = and
‘or example, 2x,
43 = O and positive for all values of x, x2, x5 except when x, = x, = x, = 0,
J
‘ . ; one ce) 8c ) is
_ i idratic form Q(x) is negative definite, if and only if, -Q(x)
aie deta Theiss 6) is negative definite when Q (x) is negative when x + 0.
positive .
Scanned with CamScanner —_—_— _
PREREQUISITE a
(iv) Negative semi-definite. The quadratic form Q (x) is negative semi-definite, if and only if,
13) is positive semi-definite. That is, Q (x) is negative semi-definite when Q(x) < 0 for all x and
gero for at least one x # 0.
ros example, if Q(X) = x2 + 2m — a = — Gy? - 2m +22) =O — Pe then
Q(@) < 0 for all x but zero for x, =
(0) Indefinite, The quadratic form Q (x) is indefinite if Q(x) is positive for some x and negative
for some other x.
For example, Q(x) = x7 ~ 3x, can be positive for x # 0 and x, = 0 and negative for x, = 0
and x # 0.
PROBLEMS
021. Show that the following functions are convex over RI:
( FO) = 3x2 (Df) = and (HH F) = [x
022. Show that the function f(x) = 2x)? + 22? is a convex function over all of R®.
023, Write the following quadratic forms in the form xTAx
(i) x2 + Bxyxy + 1GQ? - 3x37 (ii), xj? = Gxyxy + Dey + 2ey? + GeaKy - Sx?
024, Write down the quadratic forms whose associated matrices are :
2-301 12 4
@{-3 4 2 @ | 2 6 -2
1 2 -6 42 4
025. Determine whether the quadratic form
xy + Day? + Dey? - 2eyry — Leary
is positive definite.
026. Determine whether each of the following quadratic forms is positive definite or negative definite:
(a) 2x? + 6x2? - 6xypx, and (0) nay? = 2p? = dey? + yt - eats,
MULTIPLE CHOICE QUESTIONS
1, The matrix A + B is defined, if
(a) the matrix A has 3 rows and matrix B has 2 rows.
(B) the matrix A has 2 columns and matrix B has 4 columns.
(©) the matrix A has 3 rows and the matrix B has 4 columns.
(d) both A and B are square matrices of the same order.
(1 23 mismo (72 3),
neae() 23) me 2-(2 2 Sj meame( 3 2 3)esme
@ 3A - 2B, (® 3B - 2A, (©) 2A + 3B, (d)_none of these.
3.00
3. The matix A =|0 2 O|isa
oot
(a) diagonal matrix, (b) scalar matrix, (©) unit matrix, (d)_ none of these.
1 10
4 For A = =
aa(h sjemene(s 3)
(@) AB = BA, () AB + BA, (©) AB = BA (d)_ none of these.
Scanned with CamScanner Muy Sylobus GR
Ni Thp “ey Tube
‘S..Linear Programming Problem
—Graphical Solution and Extension
= “Graph is the best tool 0 visualise any cancept”
oe By deel sign Varvol o5
ion variables can easily be solved by graphical
Methoa-
constraints of the problem and identify the feasible region
‘imtersection of all the regions represented by the constraints
quadrant only. t
ed in step 3 may be bounded or unbounded. Compute the
asible region.
e function « at each comer (solution) point determined in
, (maximizes or minimizes) the value of the objective
Amro Method.
ach, involves the following steps :
fifth step we choose a convenient profit
le region. uNG PROBLEM—GRAPHIGAL SOLUTION AND EXTENSION er
compute the c-values corresponding to the extreme points
(x. ™) z= 4x, + 3x
0, 0) 0
(400, 0) 1600
(400, 200) 2200
(200, 600) 2600 <—— nibximnum
(100, 700) 2500
(0, 700) 2100
ion is that extreme point for which the objective function has the
optimum soluti
He the optimum solution occurs at the point R, ic., x = 200 and x, = 600 with the
value of Rs. 2600.
» maximize profit, the company should produce 200 belts of type A and 600 belts of
4 $35 = 1200,
jon as a straight line in the feasible region shown in Fig. 3.1(c). This line is
2x, + x= 1000 Direction of
x, = 400 maximum 2
Fig. 3.1(c)
ective value function) is a aeeibbe ies on nite
the origin. We observe that one of
ible region. This i:
" of the solution space. 88 OPERATIONS Réscy,
not accept an average rate of return below 12% or a risk factor above 6. Hence, you they f,,
important question. How much should you invest in each portfolio? 1m
Formulate this as a Linear Programming Problem and solve it by Graphic Method.
ICA, Final (May) joo
Pea ) 1995)
Step I, The appropriate mathematical formulation of the linear programming problem \,
‘Maximize z = 0.10r + 0.20x2 subject to the constraints
Se x1 + 1 S 1,00,000, x = 75,000, x2 < 75,000
Oto, + 0.20%, = 0.12( + 2) OF 0.02% + 0.081, = 0
i an, + 9m, < 6c +) OF 24, + 3 SO
i oa x) 20 and x 20
where x, = amount invested in portfolio J, and x = amount invested in portfolio 2
Step 2. The first constraint x, + x; S 1,00,000 can be graphed by plotting the straight line?
Roe eh ean [, This cuts a x-intercept and 2;intercept of 1,00,000 each. The area below
jis line represents the feasible area in respect of this constraint. Similarly, the other constraints are
depicted by plotting the straight lines corresponding to the equations x, = 75,000, x5 = 75,000,
= 2m, + 3m = 0, and —0,02x, + 0.08, = 0. Here, the area below the first three lines and beyond the
fourth line gives the feasible region in respect of these four constraints,
Thus, the feasible region in respect of the given problem is as shown in Fig. 3.2
: Sd Ces eonoorant points are : O = A = (75,000, 18,750), g PROBLEM—GRAPHICAL SOLUTION AND EXTENSION
4s, say, A and B. One unit of product A contains 36 units of X. 3 un
et ‘One unit of product B contains 6 units of X, 12 units of ¥ and 10 units of Z
of 2 mnt of X, ¥ and Z is 108 units, 36 units aud 100 units respectively. Product A
a it and product B Rs. 40 per unit
Daye above as a linear programming problem ro minimize she so1al cost, and solve the
‘graphic method. ICA. Final (May) 2002}
J. The data of the given problem can be summarised as follows
‘Mutrient content in product Minimum amount
of musrient
06 108
12 36
100
p Gry 2 108, 3x, + 12xy 2 36, 2Or, + Ir) = 100, and xy, ay 2 0,
f units of product A, and x, = number of units of product B.
W a set of cartesian rectangular axis OX,X, in the plane, As each point has the
4%), any point satisfying the conditions x, > 0 and x3 = 0 lies in the first
the given problem are plotted as described earlier by treating them as OPEPATIONS rH,
Points CAN be evaluated 4, Follow
= 201, + 401,
. n0
2H0
160 «
240
A and 2 units of PFOCUEE Hin orgy
are to be assigned to Natty
d per hour day. Grade |
Hey Of 98%. Grade 2 dispecioy,
5%, The wage rate for rude 1
Each time an evror iy canted
ht grade J and ten grade )
Anspectors 10 minimise total
(Antnamalai 1.8.4. 2002) OPERATIONS nes,
EAR CY PHICAL SOLUTION AND E
JON
pgrammes ought to be on TV and the number of radio programmes mx
dicate that a single TV programme reaches 7,50,000 customer
0 in target audience B. One radio programme reaches 40,000 i»
in target audience B.
inear programming problem and determine the media mix to maximi
a [Dethi M.A. (Oct.) 2010)
tepresent the number of programmes to be released on TV
eal formulation of the LPP is :
2 = TV programme viewers + Radio programme viewers
| __—_-= (750,000 + 1,50,000), + (40,000 + 2,60,000) x,
3 = 9,00,000.x, + 3,00,000x,
the constraints <
radio,
50,000, + 20,000.x, < 2,00,000 (Budget constraint)
423 and <5 (Programune constraints)
20 and x20 (Non-negative restriction)
ly, we first consider the inequalities as equations and draw the
Now, consider the inequalities and shade the feasible region
iz., the region ABC consisting of the extreme points A, B and C. OPERATIONS ese,
Be ila HB aailable foe 10 hou, stung
"© get maximum prog,
Lieu
mix Of AM-FM ras hy
Delhi HEBLA. (Kon) iy
tonnes per day,
at least 2 tonnes fang
hours production tne
Possible number of
- 80 per tonne of X and
M profit and t calcula
9; CA. Final (Nor) say EM—GRAPHICAL SOLUTION AND EXTENSION
€ to produce 9 units of R and 7 units of S. The newer proces: :
i amits of crude C to produce 5 units of R and 9 units of S petrol, Becau:
fhe refinery must produce at least 500 units of R and at least 300 units of § for th
i 1500 units of crude A, 1900 units of crude B and 1000 units of crude C. For each unit
“60, while for each unit of $ it receives Rs, 90. Write down the linear programm
BR as to maximize the revenue and solve it by graphical method!
‘an oil refinery must decide on the optimal mix of two possible blending proces
ts per production run are as follows :
Input (units)
eerie)
‘Grade A. Grade B
3
4 5 4 4
Sms available of erudes A and B are 200 units and 150 units respectively. M
Mieast 100 units of gasoline X and 80 units of gasoline Y must be produced. The profit
1 and process 2 are Rs. 300 and Rs. 400 respectively. Solve the LP problem by the
[Rajasthan M.B.A, 2008)
aler has received an order from a customer for at least 2,000 kilograms of scrap metal.
‘at least 1,000 kilograms of the shipment of metal must be high quality copper that ean
‘copper tubings. Furthermore, the customer will not accept delivery of the
175 kilograms of metal that he deems unfit for commercial use, ie., metal that
| of impurities and cannot be melted down and refined profitably.
‘scrap metal from two different supplies in unlimited quantitics with the following
igh quality copper and unfit scrap.
. Supplier A Supplier B
e 25% 15%
‘5% 10%
‘metal purchase from supplier A and supplier B are Re. | and Rs. 4 respectively. The
‘graphic method of LP technique) the optimum quantities of metal for the dealer
lu ce {Allahabad M.B.A_ 2009]
Isr, and
18 kg. of rj and 12 kg. of r, must be used daily. Also at mast 34 hours of
-are needed for each model X and | kg. of r, for each model Y, For each model
3 hours to manufacture a model X and 2 hours to manufacture a model
Rs. 30 for each model ¥. How many units of each model should be
model. {Gujarat M.B.A. 2008]
inspectors | and 2, who are to be assigned for a quality control inspection.
p 8-hour day. Grade 1 inspector can check picces at the rate of
tor checks at the rate of 30 pieces per hour wil
; while that of a Grade 2 inspector is Rs. 4 per hour.
. There are only nine Grade 1 inspectors and eleven
‘work fo the available inspectors 50 as
near programming model and solve it by
Panjab B.Com. 2008} 76
than 4 kg. of By and vainimum of 2 kg. of By. Since the demand for the produ
of the brick, find out graphically minimum cost of the brick satisfying the above
322. Find the maximum value of z = 50x; + 60 subject 10 the const
2x) + 3m $1500, Ay + 2; 5 1500, 0S x 5 400, 0
75x, + 1.50e) subject to the constraints :
AW, + Sry $320, dy, + Suz 5 200, OS 4, 215 OF
324. Find the maximum value of z = 7x; + 3x2 subject to the constraints:
x #2y 23, 4+ 54,05" 552, 05m
323. Maximize
325. Show graphically that the maximum or minimum values of the obj
problem are same :
Maximize (or Minimize) ¢ = Sr; + 3% subject to the constrains :
+56 24 +3n 23 055,22 05423,
326, Find the maximum value of z= Sx, + 3x. subject to the constraints:
4 S624 43n 26 055,54 05.453
327. Solve graphically the following LPP. :
ls ie Maxi ake = 3%, + 2p subject to the constra
STATS acy me ea
328, Find the minimum or/and maximum value of 2 = 3x, + Sxy
(@) "233 #41 2 12, 2, — 2-2 0)
13m 212 42420 2220,
Subject 10 the constraints :
#54, 2n <6,
B+ x39. x
3:3, SOME EXCEPTIONAL CASES
In the preceding sections, we discussed som {it
‘well-behaved’ problems. In each case, 3 i
ei sone hes Ieee “while in
finally obtained. It should not be taken as a
or for which the only soluti te is
real situations, it will be an omission, if.
seal | Nate
This section considers the following thre
(ie.
assume
Iternative PROBLEM—GRAPHICAL SOLUTION AND EXTENSION
serap metal has two types of serap available to them. The first type of s
Tee of impurity and 20% of metal B by weight, The second type of serap has 30" ines
59 of metal B by weight. The company requires atleast 120 kg. of metal A, a
react 90 Ke. of metal B. The prices for the (wo seraps are Rs. 200 and Rs 300 per ky
imum quantities of the wo scraps to be purchased so that the feclisinenis of te
‘ction on impurity are satisfied at minimum cost {Nagpur M.A, 19927]
firm has a long history of production troubles. It produces two products A and B. which
Mpecently, the company has entered into contract to supply 40 units of A and 20 units of
pany, The technology of the chemical process implies that production of A must always
iB. There are two raw material constraints to be satisfied: E
SA + 8B < 400 and SSA + SOK 50.
LINEAR PROGRAMMING PROBLEM
the LPP. in the general context, that is, when the number of variables is more
i
| Linear Programming Problem). Let < be a linear funetion bn RY defined by
2 c1% + Oy +. + Gate S
nts, Let (ay) be an m x n real ‘matrix and 1By, By, ++» By} be @ set of constants
Pays + airy +--+ inh 2 Or S or = by
Hayy + dap +. + Opty 2 OF S OF = bo
ts :
in 1 + Gu82 + -~ + Brey B OF S OF = Bw
ons (¢) is usuaily known as the set of non-negative
s which satisfies the constraints
. OPERATION
of foodstuffs available and m different types of nuttien,
f units of nutrient ¢ in one unit of foodstuff
J 1
umber of units of food j in the desired diet. Then, the jou
U and xy” > 0. Thus
lem become :
Bie 24s, + x, = 4
+ 2%) + ay’ - 13”) - x = 5
2x, + 3(e;’ — x3") + x5 = 2
it by
= ay = Ary + 4x5” + O44 + O.xs + 0.x subject to the constraints =
+4 ee
2 4 34,5,"
¢ LPP to its standard ae
zt = 3x, + 4x; + Oxy subject to the constraints 2
By +2 + 2g 26, 3) + 2p = 8,
x) 20, x22 0 and_xz_wnrestricted in sign.
third constraints are inequalities with 2 sign, we introduce surplus /
“respective inequalities.
i ices. < ay — x)", where xy/ 2 0 and 2x3” = Og Seb
LPP are: 29 =~ ‘Lony, many decision vaisbe, yy
Bl
of equality type ; (4)
Linear Programming Problem
—Simplex Method
“One step for better is good enough”
. INTRODUCTION
As discussed earlier, the Simplex Method, also called the ‘Simplex Technique’ or the ‘Simplex
Algorithm’ is an iterative procedure for solving a linear programming problem in a finite number of
steps. The method provides an algorithm which consists in moving from one vertex of the region of
feasible solutions to another in such a manner that the value of the objective function at the
succeeding vertex is less (or.more, as the case may be) than at the preceding vertex. This procedure
of jumping from one vertex to another is then repeated. Since the number of vertices is finite, the
method leads to an optimal vertex in a finite number of steps or indicates the existence of an
unbounded solution.
In the present chapter, we shall introduce and explain the computational procedure of the simplex
method. The theory behind the method will be first developed and then the computational techniques
explained and illustrated, The following important definitions are necessary for understanding and
developing the theory that follows :
Befinition (Basic solution). Given a system of m simultaneous linear equations in n unknowns
(m l3t
‘Thus, basic solution to the given problem is
(Basic) x) = 1, xy =
(Non-basic) 13 = 0.
Scanned with CamScanner UNEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 89
Similarly, the other two solutions are ;
(Basic) xy = 5/2, y= 1/3: (Non-basie) 44 © 0
and (Basic) y= 1 y= 0: (Non-basie) yy © O
In each of the two basic solutions, at least one of the basic variables is zero. Hence, Wwo of the
pasic solutions are degenerate solutions
Definition (Basic feasible solution). A feasible solution to an L.P.P., which is also a basic
solution to the problem is called a basic feasible solution to the L.P.P.
Mustrations
1, In sample problem 40/ observe that [5, 0, —1] is not a feasible solution, Only basic feasible
solutions are :
() xy = 5/3 and y=
i) oy
2, In sample problem 402, the non-degenerate solution [0, 5
basic feasible degenerate solutions are :
() m= 1 and ay = 0; (i) xy = 1 and yy = 0,
Definition (Associated cost vector). Let xq be a basic feasible solution 10 the L.P.P. :
Maximize 2 = ex subject to: Ax = b and x 20. .
2 and
= 1/3} is not feasible, Only
L
Then, the vector
Sp = (emp Crys CBs
where cy, are components of ¢ associated with the basic variables, is called the cost vector associated
with the basic feasible solution Xp.
It is obvious that the value of the objective function for the basic feasible solution Xp, is
to = Cake
© Definition (Improved basic feasible solution). Let xp and Xy be two feasible solutions to the
standard L.P.P. Then Sy is said to be an improved basic feasible solution, as compared t0 Xp if
Sahn 2 care
where by is constituted of cost components corresponding to Xp.
given by
Definition (Optimum basic feasible solution). A basic feasible solution xq 10 the LP.P.
Maximize z= ex subject to: Ax = b and x20
is called an optimum basic feasible solution if z, = ¢pXq 2 2* where z* is the value of objective
function for any feasible solution.
PROBLEMS
4403. Find all the basic feasible solutions of the equations
v Dy + 64 +2 + y= 3
Gr, + dry + dey + Ory = 2
404, Find all the basic feasible solutions ofthe following LPP. without using the simplex alogorithm :
Maximize <= 2x1 + 3ny + dry + Try subject to:
dey + By tay + Ay = 8
ay - ty + Gy — Ig = 3
420121234
and choose that one which maximises
M08. If x, = 0.47 = 0,45 = Og = Tes
3x
12, x4 = 10 is an extreme point solution of the set of equations
way ty tye
wt +4 tay = 12
may, + Bry + Bry tay = 10s
find two more extreme point solutions of the above problem.
Scanned with CamScanner 90 OPERATIONS RESEARG}
4:2, FUNDAMENTAL PROPERTIES OF SOLUTIONS
: ocedure for obtaining
We are now well set to discuss the theory behind the computatic al ries of the solutions ve 2
optimum tusie feasible solution to a given L.P.P. The fundamental propertes Tee i
L.P.P., that now follow, will allow us to arrive at the much awaited s
F Solution). [f an L.P.P, hy,
Theorem 4-1 (Reduction of a Feasible Solution to a Basic Feasible Soly fee
@ Feasible solution, then it also has a basic feasible solution.
Proof, Let the L.P.P. be to determine x so as to maximize
Te R!
axle ‘
box 2 Oz where A is an mxn real matrix and B, © are m1 ang
x= b, : Ais
8s respectively. Let p (A) = m.
Since there does exist a feasible solution, we must have
pia, b) = p(A) and m m,
@, } be a set of constants (not all zero) such that
May + aya, +
+ Ga, = 0.
Suppose that for any index 1, a, # 0. Then a, = $ %y
vj a,
e ea
yt (£2s).- :
or
Thus, we have a solutio
are positive, we shall choo:
ZeFO components. To ensure that these
O
4-20 forall jay
Scanned with CamScanner pr
| near PROGRAMMING PRORLEMSIMPLEX METHOD ”
|
‘This requires that either 0 = 0 or
\
2 Caso “
gta yao and sy
‘Thus, if we select a, such that
+ ig eo
% \
ew min} 2+ a> 0},
aN Lay,
a, ‘
a is non-negative, and so we have a feasible solution with
then for each of the p~ I variables x, ~ x,
jot more than p~ 1 non-zero component
Consider now this new feasible solution with not more than p~ 1 non-zero components. If the
corresponding set of p= 1 columns of A is linearly independent, ease (1) applies and we have arrived
gta basic feasible solution, If this set is again linearly dependent, we may repeat the process to arrive
ta feasible solution with not more than p-2 non-zero components. ‘The argument can be repeated.
Unimately, we get a feasible solution with associated set of column vectors of A linearly independent.
The discussion of case (i) then applies and we do get a basic feasible solution.
This completes the proof.
Corollary. There exists only finite number of basic feasible solutions to an LPP.
Proof, Exercise for the reader.
Theorem 4-2 (Extreme Point Correspondence). A basic feasible solution to an L.P.P, must
correspond to an extreme point of the set of all feasible solutions and conversely.
Proof. Let the L.P.P. be :
Maximize z= elx, x € Rt subject to the constraints:
Ax=b x20
where A, band ¢ are real mxn, mx 1 and nx 1 matrices respectively. Let (A)
Let $ be the set of all feasible solutions to the L.P.P. Also suppose that x is a basic feasible
solution
-[7]
0
where Xp is an mx 1 vector, such that for a non-singular sub-matrix B of A, Bx,
If possible, let x be a point of S, such that there exist x), xp € S such that x, # xp and
xeAx+(l-Axy OCACL
3 [i) =]
t asl, fm ely
Where uj, ug are mx 1 vectors and ¥,, V2 are (n—m) x1 vectors. Then
[sll ]-e-af3]
m.
a Xp = Au + (1 - A)uy
and O= dy, +(1- A) o 0, we can write
Eyyte £ b.
aya, b.
yee a
’
or JE tema = b (some a, #
‘Thus, the two different points
xl) = (xy 4,
+My FE Gy 0, Oy sore 0)
and xO) (uy Bay, ay pr E Oy 0, 0, sev OD
satisfy the constraints Ax” = b.
Also, since x, > 0, by selecting € such that
Oce {| 1,2,
: Therefore, any x € Scan be expressed as a convex
combination of the extreme points, say
k
x= E ayy: 20, Bay =I forall j
a |
Let ty = max. (6%) f= 1s 2 ou Kh Then for any x € 5,
d= ex = FM (Lo;x) = Dayj(e™H) S ZOjz%— = %
& 25% foray xe S.
‘Thus, the maximum value of z is attained only at one of the extreme points of S. That is, at least
one extreme point.of 5 yields an optimal solution.
Now, since each extreme point of S corresponds to a basic feasible solution of the L.P-P.
therefore, at least one basic feasible solution is optimal. This completes the proof.
Remark. It is easily seen that if the optimal value of z is attained at more than one extreme points of
'S, then every convex combination of such extreme points also provides an optimal solution to the L.P.P.
Thus, the optimal solution to an L.P.P. is either unique or infinite in number,
SAMPLE PROBLEM
$066 Let xy = 2 x2 = 4 and x5 = 1 be a feasible solution to the system of equations
2xy = xy + 2xy = 2
xy + 4x2 = 18:
Reduce the given feasible solution to a basic feasible solution.
Solution. The given system of equations can be written as Ax = b, where
ax(04 9) at ofa]
Scanned with CamScanner 94
vectors of A be denoted by it), st; and ay respectively, then
Gyay + Um, + Kya = O
-Hethen ay = Ha, + ap,
where a,'s are scalars not all zero. If we let 01
2a, - a; and 0 = Oy + 40. Solving theve egy
This gives 2
-2/9. ' ;
min | beg 21.9
Now, a, mim gy > OF = min | aig) = |
This indicates that vector a, can be removed. Hence, the new sol
ution With 8 sige
non-zero variables is given by
a oes eaten eae eee
he2-Gxp 0. a4 - 3x yy
4 1
he1-2xen=B
Thus, a basic feasible solution to the given L.P.P. is x = 9/2 and
(non-basic).
Note. If we let ay = 1, another basic feasible solution is obtained as
26/9 and 4 = 34/9 with x, = 0 (roon-basic),
PROBLEMS
is a feasible solution to the system of equations
4) +4 + 2y = 4
Wyant ye?
407. The column vector (1,
Reduce the given feasible solution to a basic feasible solution,
408. Consider the system of equations
ay tet ay t x27
Dey ay + Bey = Oy = 4.
Here x = 1. x2 = 1,
different basic feasible solutions.
409, If xy = 2. x3
find the basic feasible solution
Maximize =
(Meerut M.Sc. Wath.) 1998; Debit B Se.
3
= x1 + 2ty + 4x5 subject to the constraints :
Hye dey IN Bey ty + Sey = 14,
Fn 20.
61 is not a basic feasible solution to the system
tht ye?
410. Show that the vector (1, 0, 1,
Nn eye?
. Bey ry + ay ay oO.
7 Theorem 4-4 (Replacement of a Basls Vector). Let an LIP.P. have a basic feasible
we drop one of the basis vectors and introduce a non-basis vector in the Busts set.
solution obtained is also a basic feasible solution,
Proof. Let us consider the L.P.P. of maximizing z = ex subject to the constris!
X2 O: where x7, € € RY and A, b are mxn and m1 real matrices respectively.
Let p(A) = m.
- ” Scanned with CamScanner
Since, rank (A) = 2, there are only two linearly independent columns of A ‘fay. le,
= 1 and 44 = 0 is a feasible solution. Reduce this feesible solu
fag
(Sis 8
| be a feasible solution of the following Linear Programming prion tet
issn §
en te
ss ase
A NEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 98
Suppose that there does exist a basic feasible solution xp. so that
Bxy = b, xp 20
e€ A, we
set for the column vectors of A. Clearly, for any column vector
By to + Yin Dn = BYy
ay = yyby + yyy
where b, € Band y, are suitable scalars.
Now, we know that if the basis vector b,, for which the coefficient y;, is non-zero,
4,€ Acthen the new set of vectors also forms a basis
Now, for y,, #0. we can write
replaced by
and therefore (1) gives
‘Thus, the new basic solution is Xp having as its components, the variables
Snr
Sa rm — tae t = 2
We shall now show that Xp fi
non-negative, Two cases do arise :
Case (i). xp, = 0. In this case the new set o!
have assumed the existence of a basic feasible solution xp.
Case (Ii). xq, # 0. In this case we must have yy, > 0. This
O for i # ror
and Snr
”
0
F ; “
feasible also, that is, the new basic variables Xp, are also
{basic variables is obviously non-negative, since we
requires that for the remaining
yy (i # 0) either y
Sai
tm ye, for yy >0 and ier and —*s for yj <0 and ir
Mw i
$0, if we select the index (vy # 0) in such a way that
i >OQ ier
a or
%
a a
then, the new set of basic variables are non-negative; and hence the basic solution X, is feasible.
”
0
Remark, If after replacement of basis vector, the new basis matrix is
: B= ch, i,
where b, = bj fori # rand by. then, the new basic feasible solution is
Sys 'b
Me; 2 Xr
Moja and %q-= 2 are the basic variables.
Ya Yi
_ Definition (Net evaluation). Let xp be a basic feasible solution to the LP.P. :
Maximize z= cx, where Ax=b and x20.
Let ¢y be the cost vector corresponding 10 Xx For each column vector a; in A, which is not a
column vector of B, let
where 3p; = Xn) ~ tar
Scanned with CamScanner = OPERATIONS RESEAR,
Then the number
=F, ey
is _ ¢ is called the net evaluat
is called the evaluation corresponding to a and the number 3 — 15 ation
corresponding 10 a. sont ln
7 basic feasible solution to 1h,
\/ Theorem 4-5 (Improved Basic Feasible Solution). Let Xx bea te
LP.P.: Maximize z= ex subject 10 Ax = b, x 2 0. /
. intir I olumn vect h
Let %q be another basic feasible solution obtained by admiting a non basis ¢ oF & in
ihe basis, for A ” J
ean ative. x proved basic feasible
thich the net evaluati is jive. Then Xp is anil db
for wi ve ee
solution to the problem, that is pee
aXe a XB r :
imi = : R" subject to
Proof. The L.P.P. is to determine x, so as to Maximize z= “x ae rj the
constraints; Ax = b and x 2 0, b’ € R", where A is an m xz real a
We are given that Xp is a basic feasible solution. Let 2 = ¢p Xa
Let 4, be the column vector introduced in the basis, such that z ~ ¢j < 0. /
Let b be vector removed from the basis and let Sy be the new basic feasible solution, then
; A 2B
id Sy, = A
and fae = 5%
. dy
Sei = X01 - er
ai = Xai Xr,
Thus, the new value of the objective function is
B= 2) Cade
z Y
2B nf sy tet) + bay
Sjea(m- ait) + %
z yi
=, ¢ai| x01 - Sor | + 5”
ml ot) 1 Yq
Lv xBr/yq > 1
Hence, the new basic feasible solution Xp gives an improved value of the objective function.
Remark. If the given basic solution happens to be a non-degenerate one, then x5, /y,j > 0 and a
definite increase in the value of z, occurs. If the given feasible solution happens to be a degenerate one,
then an increase in the value of z, depends on xg,/y,;. In any case, zj ~ ¢j < O implies that ¢ can never
be less than z,.
‘orollary. If z ~ cj = 0 for at least one j for which yy > 0, i = 1, 2, ..., m; then another basic
feasible solution is obtained which gives an unchanged value of the objective function.
Proof. It follows from the above theorem that
* os
Pent Go 2ty, >70
wa-y-g 4 My >
ie = a since 3 ~ ¢ = 0.
Vv ?
for eee ae Soin), Let there exist a basic feasible solution to a given LP.P. If
; | for which yy = 1,2, sym), Eos : Ee
exist any optimum solution to this L.P.P. ‘m), and z - cy is negative, then there does
Scanned with CamScanner »
'
near PROGRAMMING PROBLEM—SIMPLEX METHOD 7
proof. Let the given L.P.P. be to determine x so as to
Maximize z= ex; ¢ x" @ R" subject to the constraints
Ax=b and x20, b’e R"
here A is an mXn real matrix.
Let a basic feasible solution to this problem be xp, so that
Bxp=b and xp 20
with the value of the objective function
2 = pkg = E carpy
Now, we can write
Bxy + G4) - Ey a ¢ A, & is a salar
E smb + Bay ~ 8 Fay by
E, Gn - boy) bs + Ba
If & > 0, then (xg; — Ey,) 2 0 since yy < 0. This shows that there exists a feasible solution
whose (m + 1) components may be strictly positive. But, in general, it may not be a basic solution.
The value of the objective function for these (m + 1) variables is given by
B= E enilens— Erp) + 8G = 2, enxn- 80 2, cai ~ D
=m-5G- 4.
But y-¢<0 and &>0
*. fate a bEote,
Hence, there is no limit to the optimum value of z and, hence there exists an unbounded section
to the given L.P.P.
Remarks 1. If for some ay for which zj~ cj > 0 and yy $0, ¢= 1, 2, 0 for all j such tha
where 2* is the value of the objective fun
Hence, 2 is an optimum for that basic feasible sol
ae B.
ay
Corollary. A necessary and sufficient condition for a basic feasible solution to an L.P-P. to be an
optimum (maximum) is that z ~ Gy 2 0 for all j for which a @ B.
Proof. It follows immediately from theorems 4-5 and 4-6 that the condition is necessary,
Theorem 4-7 proves the sufficiency of the condition.
\Theorem 4-8. Any convex combination of k different optimum solutions to an L.P.P. is again an
‘optimum solution t0 the problem,
Proof. Let xj, X2, ..., %; be & different optimum solutions to the L.P.P. :
Maximize z= ex; ex" RY subject to the constraints :
Ax=b and x20
where A and b are mn and mx I real matrices respectively,
Let z, be the optimum value of z.
Obviously, x, = ex; = ex; =
where x) 20, x 20... x20.
Now, any combination of xy. xy. eu) Xy
y= AX, = ou = Any
ex and b
‘i 4 :
can be written as x= a,x, where o's are positive
scalars such that 3, a, = 1.
: ‘ ‘ ‘
Ax= ACE ayx) = 2 aAxy = 2,4b=b za
=b (since Za; = 1).
Further, since x; 2.0 and ct, > 0 for all i = 1, 2,
on-negative.
Thus x is a feasible solution to the LPP, Y
+ sn k therefore every component of x will be
a
Scanned with CamScanner a
LINEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 99
x & a
Again, x= e( Sax) Egy = Eqs ss
Hence, a convex combination of & different optimum solutions to an L.P.P. is also an optimum
solution to the L.P.P.
4:3. THE COMPUTATIONAL PROCEDURE
‘The nwo fundamental conditions on which the simplex method is based are :
(i Condition of feasibility. assumes that if the initial (starting) solution is basic feasible, then
during computation only basic feasible solutions will be obtained.
(ii) Condition of optimatity. It guarantees that only better solutions will be encountered.
‘The computational procedure of the simplex method requires the construction of the simplex table.
‘The initial simplex table is constructed by writing out the coefficients and the constraints of L.P.P. in
a systematie tabular form.
Following is a specimen of the simplex table which conveniently displays the basic feasible
solution xq, the associated cost vector ¢y, the basic variables, column vectors y;'s corresponding to
s well as non-basic variables, the corresponding cost coefficients and the net evaluations in a
e & fo
Yr Xp Yn
Yau Xs Yin
Yen
You no.
The optimal solution to a General L.P.P. (when it exists) is obtained in the following major steps :
Step 1. Select an initial (starting) basic feasible solution to initiate the algorithm. (Use theorem
4-1 for the existence of a basic solution.)
Step 2. Check the objective function to see whether there is some non-basic variable that would
improve the objective function if brought in the basis. If such a variable exists. go to step 3, otherwise
stop. (Use Theorems 4-5 and 4-6 for the check.)
Step 3. Determine how large the variable found in step 2 can be made until one of the basic
variables in the current solution becomes zero. Eliminate the latter variable and let the next trial
solution contain the newly found variable instead. (Use Theorem 4-4.)
Step 4. Check for optimality the current solution. (Use Theorem 4-7.)
Step 5. Continue the iterations until either an optimum solution is attained or there is an
indication that an unbounded solution exists.
Note. To simplify the computations, we usually consider the starting basis matrix B as an identity
matrix I, ie., B= 1 Therefore yj = Bola; = Ia,
rm[o tM)
xq = Bb = Ib = Ib = b, etc.
The Simplex, Algorithm ‘
For the solution of any
x y L.P.P. by simplex algorithm, the existence of an initial basic feasible solution
assumed. The steps for the computation of an optimum solution are as follows :
Scanned with CamScanner OPERATIONS RESEAy
3 RCH
pp, bn to be maximized OF minimized, y
Step 1. Chek whether the objective fuetion of the glven LPP ato re A
itis to be minimized then we convert it into a problem of mastiniz
Minimum ze Maximum 2). Hany one of b, is nega
Step 2. Cheek whether all by (© $y Qe vor. m) are non neque TC AAT
multiply the corresponding inequation of the constraints by ts
meee 0 atroducing slack andio
‘ep 4. Convert all the inequations of the constralnts io equations by id 8 andlor
surplus variables in the constraints, Put the coxts of Hreve Yarht eee ed wai
Step 4. Obtain an initial basic feasible solutlon to the problem
in the first column of the simple table, ee Parken ee
Step 5. Compute the net evaluations <= (J % te 2 vw
8) = 6 = py, — cy Where yy © Bly
Examine the sign z) = ¢j oe
(i) Hall ~ 6) 2 0 then the inital basie feasible solution xy is an optimum basic feasible solution,
(ii) If at least one (zj = ¢) < 0, proceed on to the next step.
Sep 6, IF there are more than one negative z ~ cp then choose the most negative of them. Let i
be z= ¢, for some j =r. :
(i) Wally, $ O(/ = 1, 2, s+ m), then there is an unbounded solution to the given problem,
(ii) Watleast one yj, > O(/ = 1y 2, va m) then the corresponding vector y, enters the basis yp.
‘Step 7. Compute the ratios {+ y, > 0, Fe 1, 2s sm} and choose the minimum of them, Let
Ye
the minimum of these ratios be xy/y4,. Then the vector y, will level the basis yg. The common
clement y4, which is in the Ath row and the rth column is known as the leading element (or pivotal
element) of the table,
Step 8 Convent the leading element to unity by dividing its row by the leading element itself and
all other elements in its column to zeroes by making use of the relations :
a Ms
Sys iy ne Pek Qumehiek
+ my
and yy = Me ve '
wy, FEO Mz wat
‘Step 9. Go to Step 5 and repeat the computational procedure until either an optimum solution is
obtained or there is an indication of an unbounded solution,
. SAMPLE PROBLEMS
. wn Use simplex method to solve the following LP.P.
Maximize 2 dey + 1c subject to the constraints :
Dey + Ay S50, Day + SxS 100, 2x, + 90;
Solution. : ia ei
Step 1. By introdueiny
constraints of the given L
ap 20 and x20.
1B slack variables 520,
4,20 and 5) 2 0 respectively, the set of
.P.P, are converted into a i " ne
the system of equation:
1
2110 0)/m] [50
250 1 offs |e] s00
230.0 iy 90
%
Scanned with CamScanner _
LINEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 101
‘The modified objective function is to maximize
= day + Ur + 0.5) + O.sy + 0.53.
ep 2. An obvious initial basic feasible solution is given by xp = B-'b.
_ IG steht)
50
B+, and b ws)
90
‘Step 3. Compute y; and (z; — ¢) as follows :
100 2
=|o 10 + ]
oo ijl} 2
10 oft] fi
Bla, =|0 1 el]. :]
oo ifs} [3
ya = Bele; =e), yy = Brley = ep and ys = Ble; = €
25 — €5 = CpY5 ~ C5 = (0, 0, 0)
2
= cays — c= (0, 0, of}]-4 a4
2
1
o-aran-a=00o/3]- 10=
3
1
s-oren 9-003] o=0
0
0
weacena-eoo)i] o=0
0
1
Step 4. The initial basic feasible solution is now represented in the simplex table below :
“ 4 10 0 0 o
£5 Ye Xp on Ys ys ¥s
0 Ys 50 2. 1 1 0 0
0 v4 100 2 © 0 1 0
0 Ys 90 2 3 0 0 1
zy 0 0 0 0
yg 2 0) 4 10 0 0 0
From the tableau, it is apparent that there are two z; — c) which are negative. We choose the most
negative of these, 10, The corresponding column vector y2 enters the basis.
| Step 5. Since all the entries of y, are positive, we compute min. {38-22 9}:
hr
400. This occurs for the element ys. = (= 5). Thus the vector y4 will leave the
basis ‘Yg and the common element y,, becomes the leading element for the first iteration.
Scanned with CamScanner =
102
OPERATIONS FESEAR,
Step 6. Convert the leading clement y,, t0 Unity and all other elements of y; t0 Zer0es by makin
use of the following transformation : :
$y = yy - hn: CET AN
YAY yy rat 2
and J2ON~AKGS
a or 20,etc,
= 30
= 30
da * Ys
¥3 mye Js.
1 -US 0
0 us 0
0 as 1
0 2 0
y-g z= 20) 0 0 2 0
The above simplex table yields a new basic feasible solution with increased value of z. Moreover,
possible, since all z; ~ ¢ 2 0. Hence, we have attained
4, and s3. The maximal basic feasible solution
no further improvement in the value of 2
our desired optimum solution with basic variables
to the given L.P.P., therefore, is
x
Note. It is evident from the net evaluations of the opti
to non-basic variable 1, is zero.
This is an indication that an alternative basic solution exists. Thus, we can bring y, into the basis
in place of y, or ys, the resulting new basic feasible solution will also be an optimum one.
by introducing y, into the basis in place of y,, the new optimum simplex table is as follows
ee
20 with maximum z = 200,
jum table that the net evaluation corresponding
0. ay
Now,
: Yo v4 % v4
4 v1 15078 1 sie “18
10 Y 252 0 us us
o Ys 15 Ours -1n ce
2 4 oO 2
_ yog 2 (=200) o 0 0 2 o
We observe from this table that the basic feasible solution has been changed but the optimum
value of z remains the same, : |
If two basic (optimum) feasible solutions are known, an infinite number of basic (éptimum)
feasible solutions can be obtained by taking any weighted average of the two solutions. Thus, if
: |
Scanned with CamScanner NEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 103
° 130/8
20 25/2
xy =[0] and x; =| “0° | are wo BFSs, then 9 new BFS is =
0
0 Is
(1 = 4) 150/8 (1 = A) 1150/8
. 20K + (1 - 4) 25/2 QS - 15A)/2
xe = dx (m= 302 = 30, «|, OS ASI
0 0
30A + (1 - AVIS 15 + 15%
it can be verified that the solution x” will always give the same value 200 of z, for all
osasth
~~ Remark. Interpretation of the final simplex table. The simplex table helps 10 predict the effect of
cchanges in the resources and the profit mar;
(a) Opportunity cost, The terms in z;-row indicate the opportunity cost. If we do not ul ji:
of ay: the loss of profit is 4 units and non-utilization of one wnit of x2 costs 10 units.
(b) Shadow cost, The coefficient of non-basic variables, i.e, of s and s2 are called ‘shadow costs’.
‘They represent the decrease in the optimum value of the objective function resulting from unit
increase in a non-basic variable.
one unit
412-Use simplex method to
Minimize z= xz ~ 3x; + 2x5 subject to the constraints :
Bry - xy + ts $7, -2m + dey S12,
xy + Ixy + Bes $10; Xp 20, xy 20 and xy 2 0.
[Garhwal M.Sc. (Math.) 2002]
Solution, Introducing slack variables s, 2 0, sp 2 O and s, 2 O in the respective inequalities; and
converting the objective function into that of maximization; the linear programming problem is:
Maximize zo = - (cq ~ 3x5 + 213) + 0. + Ousg + Os) subject to the constraints +
Bry - 4 + 2s + 5) = 7
a 2ip + day + Oats + 53 = 12
= 4r + 3xy + Bs + 55 = 10
Ape Xp Kye Sty Sp. Sy BO
‘The set of constraints can be written in matrix form as
- (x
a
12100
4001 0ff3l=
3 8001)"
cy
5
3-12 100
where A=|-2 40 0 1 0) x=[a 45 5 % 5] and b= [7 12 10)
438001
‘An obvious initial basic feasible solution is x» = B-'b, where B = Ij, and
Xp = basic variables corresponding to coloumns of basis matrix B (= 1).
sacs BRST IEEE F TetE
Scanned with CamScanner 104
The iterative simplex tables are :
Initial Heration. Introduce y and drop ys.
9 9 °
a Yo Xp % ¥s 5
0 Ya 7 1 6 as
0 Ys 12 6 1 ;
0 ¥. 10 Fi
2 0 :
At least one zj ~ cy viz. zz ~ Ca is negative and therefore the current basic feasible wolution
; ;
optimum, We choose the column corresponding t0 z ~ cy ie, column vector y, exten the ve
2 cit,
Yin? OF is P= 3), current basis »
Jn
(since at least one yj > 0). Further, since minimum {
Ys leaves the bass. This gives Yo (= 4) as the leading element. Now using E-Row operation, vx
convert the leading element into unity and all other elements of the entering column vector y; 19 ze,
We get the improved basic feasible solution as shown in the next simplex table, 7
First Iteration. Introduce y, and drop yy.
o Ya Xn yt na ys Ms 35 Xe
Ys 10 @ 0 2 1 us o
M2 3 “IR 1 o 0 us 9
Ye 1 -52 0 8 0 34 1
z 9 =I 0 2 °o 34 6
Observe that z) — c, is negative and thus the current basic feasible solution is not optimum. Te
column vector corresponding to z, - cy enters the next basis yg (since y,, > 0). Further, since oxly
Jui > O both yz < O and y,3 < 0); current basis vector y, leaves the basis. This gives y,, (= as the
leading element. Using E-Row operations, we convert the leading element into unity and i! other
entries in its column y, to zero. The improved basic feasible solution is obtained as shown in the next
simplex table.
Final Iteration. Optimum Solution,
oe Ya Xp yu x2 ¥3 ya ¥s Xe
= 4 1 0 Is ws 10 a
3 Yo 5 0 1 2S us 3n0 10
0 6 ul 0 0 10 G\ -n a
z i o 0 12S us 8/0 9
Since all z; - ¢, 2 0, an optimal basic feasible solution has been attained. Thus, the optimex
solution to the given L.P.P. is
Minimum z = - Maximum z+ = -11 with x) = 4, x5 = 5 and x5 = 0.
413: Find the maximum value of 2 = 107x, + x; + 2x; subject to the constraints :
Y dx; + Xz - Ox; + Ixy Ox, + xz ~ 6x3 SS, 3x, - xy - 2 SOF
Xp Xp xy 1420.
Solution. By introducing slack variables s, > 0 and s, > 0 respectively, the set of constraints of
the given linear programming problem is converted into the system of equations :
Al
Scanned with CamScanner LINEAR PROGRAMMING PROBLEM—SIMPLEX METHOD 105
4/73 173 2 1 0 0
1 60 10
3-1 -10 041
‘An obvious initial basic feasible solution is xp = B-'b, where B =
(7/3 5 O}.
Vs
°
x] [i 0 o]' [73
eo sie 10 5
$2 oot 0
Using now simplex method, the iterative simplex tables are :
Xp = [xy 5 sgh and
1"
Initial eration. Introduce y, and drop Yq .
_ 107 L 2 oo 0
& Yo Xn v1 Yo ys ya ys ¥
a) 7 73 143 1B 2 1 0 0
0 Ys 5 16 1 6 0 1 0
0 Yo 0 ® = ot = Unenl
5 0 0 0 0 0 0
y-g 2-0) 107 2 0 o 0
It is apparent from the table that there are three z ~ cj which are negative. We choose the mo:
negative of these, viz., - 107. The corresponding column vector y,, therefore, enters the basis yp (since
all yy > 0). Further, since minimum {= Ya> of
it
vector yg leaves the basis yy and the element ys, becomes the leading element, Redu
element into unity all other entries of y, to zero, we get
Final Iteration. Unbounded solution.
$ which occurs for the element yx (= 3)
g the leading
So Ye Xp yi yo v3 Ys ¥5 Yo
0 Ye 73 0 179) =419 1 0 149
0 Ys 5 0 19/3, mn 0 1 -1683
107 n 0 1 -3 <3 0 0 1B
3 107-1073 -107/3, 0 0 10783
y-g 20) 0 =110/3_-113/3 0 0 1078
It is apparent from the table that the net evaluations z - cz and x ~ cy are negative, 23 - ¢s
beirig the most negative one. The column vector corresponding to z3 ~ ¢3 is ys Now y3 will not enter
the basis yg, since yj (i = 1, 2, 3) are non-positive. Instead, this is an indication that there is an
unbounded solution to the given L.P.P.
PROBLEMS
414. Use simplex method to solve the L.P.P. :
(a) Maximize z = 3x, + 2ry subject to the constraints :
0 Ky tH S4 4-52 420 420. Mas 1992}
« +E $6 2+ S6 MeO
(b) Maximize + 3xp_ subject to the constraints :
At Sd nn tay Sd and xy + 2g SS, 4, 20, 4 20.
Scanned with CamScanner ee
OPERATIQy
NS Reg,
106
415,
Use simplex method to maximise
2 = Sey + 3xy_ subject to the constraints Ray
(a
HHH S2 Sy +24 S10 O HS4 O53 H+ Im S48
3H + By S12 aay 20. H+ $9 tym 20,
[Purvanchal M.CA. 1996} Wagariuna src,
Use simplex method to maximize z= xy + 2xy + Ary subject 10 the constraints hy
ny + ey t By S10, a Hay SS. Ay tnay 20.
Use simplex method to
Maximize z= 2x) ~ xy + xy subject to the constraints :
416,
417,
Br tay + ay S60, ay y+ 2G S 10,
Ay ty ay $20 and ay ty 5 20.
418. Use simplex method to
Maximize z = 3x, + 2vy + Sxy subject to the constraints :
Ay + ty + ty S430, Buy + ey $460, ory + 44S 420
419. Use simplex method to solve the L.P.P. :
Maximize z = 2x, + dry +3 + x4 subject to the constraints :
Ay 4 20,
Art RD Hay S 4, Oy ty $3 ay tay Hay SH Ae ty SHE Madras Maps, 19%,
420. plex method to
Hy + ey + Oey + dry 5 BO, 2ey + 2m Hy S60
Bay + By Hay ty S80, ay yy ay 2.
421. Using simplex method solve the L.P.P, :
Maximize z= 4x, + Sty + 9x5 + 11x subject to the constraints :
Hr + 445 + 44S IS, Tey + Sty + Buy + Dey S120
Bay + Sky + IO, + 15ty $100, xy pay ay 2 0. Wethi B.S, (Stat) 1999)
422. Solve the following problem by the Simplex Metho
Maximize 15x, + 6x2 + 9x3 + 2xy subject to the constraints :
Dey +, + Sty + 0.64510, 34, +5 + 34) + 0.2544 5 12
Tay +44 S35, 420 for j= 12,34,
423. Show that the L.P. P.
Maximize z = 4x; + 42 + 31s + xq subject to the constraints :
dry — Gy ~ 515 ~ 4g 2-20, 3x, = 26 + ey + ay < 10
Be) 32 +3 +S, dy ay 20
has.an unbounded solution. [Parvanchal MCA. 198)
. USE OF ARTIFICIAL VARIABLES
3
‘ra have seen thet ional I simplex method, it is most convenient
have the slack variables as the starting (nti!) basic variables. Thus, if the original constant 2
haven ori ofthe type @) we may no lager have 8 ready starting basic feasible solution. ad
jer to obtain an initial basic feasible solution, we first put the oi .P.P. into its stan
form ad then ‘a non-negative variable is added tothe left side of each ee asa ee the much
jed starting basic variables. The so-added variable is called an artificial variable and plays
je as a slack variable in providing the initial basic feasible solution. However, since =
variables have no physical meaning from the standpoint of the original problem, the meth
| needi
same rol
} ‘il be valid only if we are able to force these variables to be
Scanned with CamScanner Duality in Linear Programming
“Man is simply 10 recognise that there are different approaches for solving problems”
5:1. INTRODUCTION
Associated with every linear programming problem (maximization or minimization) there always exist
another Linear Programming problem which is based upon the same data and having the same
solution. The original problem is called the primal problem while the associated one is called its
dual problem. It is important to note that either of the two linear programming problems can be
tweated as primal and the other as its dual. The two problems, thus, constitute a primal-dual pair.
The concept of duality is based on the fact that any linear programming problem must be first put
in its standard form before solving the problem by simplex method. Since, all the primal-dual
computations are obtained directly from the simplex table, it is logical that we define the dual that
may be constituent with the standard form of the primal.
($2 GENERAL PRIMAL-DUAL PAIR
Based on the standard form of primal, there are two important primal-duial pairs :
Definition 1. (Standard primal problem)
Maximize z= c)x; + €2%z +... + CyXq subject to the constraints +
xy + Qk, + oon + Uinky = By
4205
Dual Problem
Minimize 2* = byw, + byw; +. + by Wy subject to the constraints :
ayy + yyy + one + Oy Wy 2 655
w; (= 1, 2... m) unrestricted
are the primal variables, w,’s the dual variables and the other constants have their
Note that
usual meanings.
Definition 2. (Standard primal problem)
Minimize z= ¢)X) +,¢2%z +. + Cy%q Subject to the constraints :
bs i
FFL Bown
aya, + apr, + on. + dint
a
Dual Problem 3
Maximize 2 = bjw; + b2W, +... + DyWy subject t0 the constraints :
yw, + dyyWy + oa. + yy Wy S C3
w; (i = 1, 2, ....m) unrestricted.
Scanned with CamScanner ee
q
130 OPERATIONS Res:
ARCH
Remark. From the above definitions one may easily observe the following:
there is a dual variable.
(a) For every primal constraint
(b) For every primal variable, there is a dual constraint
(6) The coefficients of the dual variables in the const
primal variables except that they are transposed, i... columns im pri
rows in the dual coefficient matt,
(a) The number of dual constraints is exactly equal to the number of primal variables. whereas the
number of dual variables is exactly equal to the number of primal constraints.
(e) The objective coefficients of the primal variables become the right-hand side cont tte of dual
vsasinte whereas the right-hand side constants of the dual constraints become the objective
coefficients of the primal problem.
The information regarding the primal-dual objective,
variables may be summarized in the following table :
Itints are the same as the coefficients of the
nal coefficient matrix become the
type of constraints and the signs of the dul
Standard primal ~ ‘Dual
objective Objective Constraints Variables
‘Maximization ‘Minimization 2 Unrestricted
Minimization Maximization s Unrestricted
f
. FORMULATING A DUAL PROBLEM
Various steps involved in the formulation of a primal-dual pair are >
Step 1. Put the given linear programming problem into its standard form. Consider it as the
primal problem.
Step 2. Identify the variables to be used in the dual probl
equals the number of constraint equations in the primal.
Step 3. Write down the objective function of the dual, using the right-h
primal constraints.
If the primal problem is of maximization type,
vice-versa.
Step 4. Making use of dual variable identified in Step 2, write the constraints for the dus!
problem.
(a) If the primal is a maximization prob’ m, the dual constraints must be all of “2° type. Ife
primal is a minimization pfoblem, the dual constraints must be all of ‘<" type.
“(b) The column coefficients of the primal constraints become the row coefficients of the du
constraints.
(c) The coefficients of the primal objective function becomes the right-hand side constants of
dual constraints.
(d) The dual variables are defined to be unrestricted in sign.
Shep 5. Using steps 3 and 4, write down the dual of the given L.P.P.
lem. The number of these variables
and side constants of the
the dual will be a minimization problem and
yf the:
‘Note : ‘The dual constraint corresponding to an artificial variable in the standard form of the primal is
always redundant, heuce it is never necessary to consider the dual constraint associated with an
artificial variable.
Remarks, 1. If the given linear programming problem is in its eanonical form, the primal-dual pat is
said to be symmetric. ;
Fit the given linear programming problem is in its standard form, the primal-dual pair is said to be
unsymumetric.
Scanned with CamScanner DUALITY IN LINEAR PROGRAMMING
131
5:4, PRIMAL-DUAL PAIR IN MATRIX FORM
ndard Primal Problem
Definition 1. (Standard primal probl ind x7 © Re
La ne contra fem). Find xT © RY so as to maximize z= ex, ¢€ RY
Ax=b and x20, bT Rm
where A isan mxn real matrix,
ual Problem. Find w? eR” ae .
D ind WT © R so as.t0 minimize 2x = blw, be RM subject to the constraints :
a Alw 2 cl ce RY ‘
where AT is the transpose of an mn real matrix A and w is unrestricted in sign,
Definition 2. (Standard primal problem). Find x7 € R" so as to minimize
subject to the constraints : = ex, ce RY
; Ax=b and x20, ble RM
where A is an mXn real matrix,
ind wT @ Re ;
Dual Problem. Find w" € RY, so as to maximize 2 = bTw, b € RM subject to the constraints:
uv i Alw scl ce R"
where AT is the transpose of an mn real matrix A and w is unrestricted in sign.
SAMPLE PROBLEMS
$0 Formulate the dual of the following linear programming problem :
v Maximize z = 5x, + 3x, subject to the constraints :
3xy + Sky $ IS, Sxy + 2xz $10, x) 20 and xy 20.
Solution. Standard Primal, Introducing slack variables s, 2 0 and s 2 0, the standard linear
programming problem i
Maximize z = Sx, + 3x + 0's, + O's, subject to the constraints :
Bry + Say + 51 + Og = 15, Sxy + 2ey + O51 + 52 = 10, xp, xy 51. 52 20
Dual. Let w, and w, be the dual variables corresponding to the primal constraints. Then, the dual
problem will be :
Minimize z* = 15w, + 10w, subject to the constraints :
Bw, + Sw 25, Sw) + 2) 23
Ww, + Ow, 20 o
jy, 8 ae Op w, 20 and. w = 5
w and w, unrestricted (redundant).
‘The dual variables “Ww, and w, unrestricted” are dominated by w, 2 0 and w, > 0, Eliminating
redundancy, the restricted variables are w, 2 0 and w, 2 0.
(2. Write the dual of the L.P.P. : :
Minimize. z = 4%) + 6x, + 18%; subject to the constraints
xy + 3xq 23, Xp + 2p ZS-and xp, xy xy 2 0. [Delhi B.Sc. (Stat,) 2006]
Solution. Standard Primal. Introducing surplus variables s, 2 0 and s,°2 0, the standard form of
the L.P.P. is:
Minimize z
xy + Bry — 5) = 3s Oxy + ap + ty - =
Ax, + 6x + 18x, + 0-5, + 0-5, subject to the constraints
Xp Age Be Sy 82:20
ATTN A
Scanned with CamScanner
E s 32 OPERATIONS RESEARCH
Dual. If, and wy be the dual variables corresponding to each primal constraint. the dual
problem will be
3w, + Sw subject to the constraints +
Maximize z*
wy + Oy $4, Buy + wy $6, Owe + 22 S 1B:
wy, + Oo $0, and Ow - w2 5%
w, and wy unrestricted (redundant).
Eliminating redundancy, the dual problem is :
Maximize zt = 31, + 502
wy $4, 3wy + wy $6, 2og S185
. aa ramming problem:
503, Write the dual ofthe following lear Pre Te constants:
/S Minimize 2 = 3x, ~ 2x2
xy + Sry + 43 27 Oxy + 32+ Ses 24
Ixy ~ 2x — 3 $10, ay — 2x + SH 2S
fxy + 7xq — 2x3 22, 4 2H 4g 20, 120
Solution. Standard Primal. Introducing surplus variables 1 20, 2 20 42
slack variable s, > 0, the standard form of L-P.P. is : : =
Pye 3x, — Dey + Any + 05, + OS + Oe + Os subject to the constraints :
aint :
subject to the constr
0.
w, 20 and 22
0, ss 2 0 anda
Minimize z
3x + 5m + 45-1 = 7
6+ mt3g-2=4
10
Tx -2n- t=
ay - 2a + Sty - 54 = 3
Ary + Trg - 2x3 - 55 = 2
20.
is Xp 236 Sts Sos Sh St $5
ponding to the five primal constraints
3, 4, 5) are the dual variables corres}
standard primal L.P.P. is :
+ 2ws subject to the constraints :
Dual, If w,(j = 1, 2
in the given order, the dual of the st
Maximize zt = 7w, + 4w, + 10w, + 3¥%
Bw, + wy + Tws + wy + dws $3
Sw + Wz — 2ws ~ 2Wg + Tws S -2
4wy + 32 - Wy + Sy — 2ws S 4
wy $0, =, $0, w $0, — my $0, ws $0
w= 1 23,4, 5) ate unrestricted in sign.
Eliminating redundancy, the dual variables are : w, > 0, w, 2 0, ws $0, ws 2 O and ws 2 0.
Note. In the primal, if we multiply the third constraint throughout by —1, the dual variable w3 will be 2 0
insteadof < 0.
/;
504/ Obtain the dual of the following LP.P. :
_/ Maximize % = 2x, + 3x, + x; subject to the constraints :
dry + 3iy +45 = 6 a 42mg + SR = 45 Xp Xp 520 Madras BE. 19)
Solution. Clearly, the given linear programming problem is in stai idering i
standard primal, its dual is : indard form, Considering it as
Minimize 2*
4wy + Wz 22, 30, + 2wy 23, wy + Sw2 21; wy and wy are unrestricted,
where w, and w, are dual variables. ,
6w, + 4 subject to the constraints :
Scanned with CamScanner SW ~
DUALITY IN LINEAR PROGRAMMING 133
505, Obtain the dual problem of the following primal problem :
Minimize z = x; ~ 3x, ~ 2x, subject to the constraints :
Bxy - x2 + 2x3 S 7, 2ey - dey 2 12, 4x) + 3xy + Bxy = 10
X42 20 and x; is unrestricted, [Delhi B.Sc, (Stat.) 1995}
Solution. Standard Primal. Introducing a slack variable s, 2 0 and a surplus variable 5, 2 0, the
primal problem is restated as
Minimize z = ex subject to the constraints : Ax
where x = [ty 254" sy he = (1, -3, -2, 2,0, 0), b
3-1 221 «0
2 -4 0 0 0 -1}, when x3 = ay - 345”.
4°38 80 0
Dual. If w = (wy, wz, ws) are the dual variables, the dual problem of the prima
Maximize z* =
b and x20,
(7, 12,10] and
A
Tw + 12Ww, + 10w; subject to the constraints :
Biv, + 2wy - 403 S 1
. = Wy = dey + By S—3 > wy + 4g — 3s 23
2m, + 8wy S-2
=2w, - 83S 2
wy $0
-w, 50
wi. w2 and w3 unrestricted.
Eliminating redundancy, dual variables are w, < 0, w, > 0 and ws unrestricted. This is re-written
as follows :
} 3 -2m ay = 2
Jomso and w) > 0
Maximize z* = 7w, + 12w + 10w, subject to the cons
Buy + avy — 4¥y S 1, wy + 4g — 303 2 3, —2wy - Bw,
ts:
wy $0 and wz 20, wy unrestricted.
Note. In the primal, if we multiply the first inequation by -1; the dual variable w; will be 2 0 instead
of $0.
PROBLEMS
Formulate dual of the following L.P.P.
(@) Maximize z= 4x, + 2x, subject to the constraints :
x) +23 4-42-25 20 and 20. (Annamalai MBA. (Nov.) 2009)
(6) Maximize z = 2000x + 3000y subject to the constraints :
Gx + 9y $100, 2 + y $20; x20 and y20, (Osmania M.BA. (Sept.) 2001)
50J-Formulate the dual of the following L.P.P. :
Maximize z = 10x, + 8x2 subject to the constraints : ‘
ay + 2ey 2S, 2p EIA Ky HID e A
x, 20 and x; is unrestricted, [Osmania M.B.A. 1999)
‘S08. Obtain the dual problem of the following L.P.P.
Maximize f(x) = 2x) + Sxq + Gry. subject to the constraints :
Sey + 6%; — ay $3, -24 ty tae $4, Sy +H AQ SL,
3k = By + Try $6 ay ty ay 20.
Scanned with CamScanner 134
/ OPERATIONS egg,
Find the dual of the following Lp.p, :
Maximize
te lys A subject to the constraints
NSH SI ay 4 ay
0, ‘ 4 unrestricted
510. Obtain the ys dy sm: ge and 4, unreste
in the dual of the following linear Programming problem
Maximize
2416 Bey + gy subject ta the constraints
1 MOM, Hy Oy Ody Sy
1, Obtain the dual problem of the following LPP,
Maximize z= xj ~ 2x + Bey subject tthe constraints:
= 2 +
420.
Madre HE, yy
tay 2 dey ede 6 ey BE tae ty 2
S12. Obtain the dual of the following finear programming problem :
Minimize z= xy + ty + xy subject to the constraints :
ay = ny 6 Any Sy 2g 3, Beh
44. % 2:0 andy unrestricted. [Garkwal M.Sc. (Mat) rp,
513. Give the dual of the following L.P.P. +
Minimize z= 2x; + 3xq + 4ry subject to the constraints =
2 tI S22 Ite Iya 4 4 4g + OH SS
yy 422.0 and xy is unrestricted.
S14. Maximize z = Gx, + Gr +43 + Tg + Sts subject (0 :
fey # ey + ty 6 Seg ry 2 Dey Hay t Udy + Ot + 9 =
sy Ap ty My 20, 5 unrestricted,
5 DUALITY THEOREMS
Theorém S41. The dual of the dual is the primal.
Proof. Let the primal L.P.P, be to determine x7 € R" so as to
Maximize f(x) = ex, ¢ € RY subject to the constraints :
Ax=b and x20, ble R"
where A is an m x1 real matrix.
The dual of this primal is the L.P.P. of determining w” € RM" so as to
Minimize f(w) = b’w, b’ € R” subject to the constraints :
ATw 2 c7, w unrestricted, ¢ € RY,
Now, introduce surplus variables s 2 0 in the constraints of the dual and write w = w; -™%>
where w, 2 0 and w, 2 0.
The standard form of dual then is to
Minimize g (w) = bT(w, ~ w;
bY € R™ subject to the constraints
AT(w, - 2) - 1,8 = 07, ce R"
wy, wz and s 2 0,
Considering this linear programming problem as our standard primal, the associated dual proble™
will be to
Maximize f(y) = ey, ¢ € R" subject to the constraints :
(ANTY SDT), <(AT YY s -bhy, :
-y $0 (=> yz 0) and y unrestricted,
ad
Scanned with CamScanner 135
DUALITY IN LINEAR PROGRAMMING
Eliminating redundancy, the dual problem may be re-written as :
Maximize f(y) = cy, ¢ € R" subject to the constraints =
Ay sb
aye) = Ay=b, bre R”
y20.
‘This problem, which is the dual of the dual problem, is just the primal problem we had started
with.
‘This completes the proof.
aid 52 (Weak Duality Theorem). Let x, be a feasible solution to the primal problem
Maximize f(x) = ex subject to: Ax
