Professional Documents
Culture Documents
STUDY182@396977
STUDY182@396977
Duality Theorems
Prof. V.S. Verma
Department of Mathematics and Statistics
D.D.U. Gorakhpur University, Gorakhpur
Now, our aim is to construct the dual of the dual (ii). For this, we shall
first express the above dual problem (ii) in the symmetric form of the
primal by multiplying throughout by – 1. Thus, we obtain the following
LPP :
Maximize (– Zw) = – b1w1 – b2w2 – b3w3 ... – bmwm
subject to – a11w1 – a21w1 – ... – am1wm – c1
– a12w1 – a22w2 – ... – am2wm – c2
: : : : …(iii)
– a1nw1 – a2nw2 – ... – amnwm – cn
and satisfying w1 0, w2 0, ..., wm 0
Dual of Dual. Now, constructing the dual of (iii), we obtain the following LPP :
Minimize Zv = – c1v1 – c2v2 – ... – cnvn
subject to – a11v1 – a12v2 – ... – a1nv1 – b1
– a21v1 – a22v2 – ... – a2nvn – b2
: : : :...(iv)
– am1v1 – am2v2 – ... – amnvn – bm
and satisfying v1 0, v2 0, ..., vn 0
Now, multiplying (iv) throughout by – 1, we get the following LPP:
Maximize (Zv) = c1v1 + c2v2 + ... + cnvn where Zv = – Zv
subject toa11v1 + a12v2 + ... + a1nvnb1
a21v1 + a22v2 + ... + a2nvnb2...(v)
: : : :
am1v1 + am2v2 + ... + amnvnbm
and satisfying v1 0, v2 0, ..., vn 0
which is identical to the primal (i).
This completes the proof of the theorem.
Note 1. Of course, the variables v1, v2, ...,vn in (v) will be equal to x1, x2,
..., xn respectively.
3
(c) Both problems may be infeasible, i.e. if the primal has no solution,
then the dual will also have no solution.
Proof (a) : We first assume that the primal is feasible and that a maximum
feasible solution has been obtained by simplex method.
The primal dual pair of problems in matrix form may be taken as
follows :
Primal : Max. Zx = CX, subject to AX b, X 0.
Dual : Min. Zw = Wb, subject to WA C, W 0.
Now, let the primal problem has a finite optimal solution given by
XB = B–1b. ...(i)
Then, the corresponding value of primal objective function is given by
Zx* = CBXB = CB (B–1b) ...(ii)
and the accompanying conditions for optimality are given by
{Zj – cj} 0 or {CBB–1 A – C, CBB–1} 0 or CBB–1 A – C 0, CBB–1> 0 ...(iii)
Now, to verify that the associated optimal dual solution is given by
WB = CBB–1 ...(iv)
we must only observe that the condition (iii) then, demonstrates the
feasibility condition or in other words, the dual problem is satisfied.
To show Max. Zx = Min. Zw :
For the optimal dual solution (iv) i.e. WB = CBB–1, the
corresponding value of the dual objective function is given by
Zw* = WBb = (CBB–1) b = CB (B–1b) = CBXB = Max. Zx
Now, it only remains to show that Zw* is also the optimal value for
the dual objective function.
Since, by Theorem 2, we have Zx Zw for all feasible X and W, extremes
of primal and dual are related by max. Zx min. Zw.
Using (iii), we have Zw* min. Zw.
9
But Zw* can never be less than min. Zw. Hence, we have max. Zx= min Zw.
Similarly, starting with the finite optimal solution to the dual problem, if it
exists, we can prove that the solution to the primal is feasible and max. Z x
= min. Zw holds. To do so, we shall use the fact that 'dual of a dual is the
original primal'.
This completes the first part of the theorem.
Note : It is also important to note that if the primary constraints are
equalities, then (iv) continues to hold, but WB, no longer has to be non-
negative.
(b) To prove second part of the theorem, we suppose that for any value of
the primal objective function, no matter however large, there exists a
feasible solution yielding this value. Then, the dual problem can have no
feasible solution (W) whose components are all finite. If it did,
acontradiction would arise. By the result [(max. Zx Zw) or ( Wb)], the
dual solution would provide an upper bound on the value of the primal
objective function for any feasible primal solution. But, we have assumed
that the primal objective function can be made arbitrarily large.
A similar argument will show that the primal has no solution, when the
dual has an unbounded solution.
Note : It must be carefully noted that "if the primal has no solution then,
the dual has an unbounded solution" is never true. But, it will be true that
if the primal has no solution, then the dual will also have no solution. This
point has been explained by an example.
(c) The following example will be sufficient to reach the desired result of
third part of fundamental duality theorem :
Consider the following linear programming problem as a primal
problem :
Max. Zx = 3x1 + 4x3, subject to the constraints x1 – x2 1, – x1 + x2
0 and x1 0, x2 0.
This primal has no feasible solution, since the constraints
are inconsistent as seen from Fig.1.
10
𝑥2 𝑤2
𝑥1
−𝑤1
− 𝑥2
−𝑥1 − 𝑤2
≤ −1
+ 𝑥2 ≥4
≤ −1
𝑤1
− 𝑤2
≥3
𝑥1 𝑤1
O O
Fig. 1 Fig. 2
Primal Problem
Dual Problem
Feasible solution No feasible solution
Feasible solution Optimal solution Dual unbounded
No feasible solution Primal unbounded May occur
Complementary Slackness Theorem
In this section, we shall make a further effort to relate the primal and the
dual formulations of linear programming problem. The following theorem
(called the complementary slackenss theorem) shows the intimate relation
between the primal and the dual problems :
Theorem 11.(Complementary Slackness Theorem) For the optimal
feasible solutions of the primal and the dual system, we have
(i)if the inequality occurs in the ith relation of either system (the
corresponding slack or surplus variable xn + i is positive), then the ith
variable wi of its dual is zero, i.e. wi = 0 and
(ii)if the jth variable xj is positive in either system, then the jth dual
constraint holds as a strict equality (i.e. the corresponding slack or surplus
variable wm + j is zero).
Proof : The primal objective function can be written as :
c1x1 + c2 x2 + ... + bm wm = Zw (max.) ...(i)
and the dual objective function can be written as :
b1w1 + b2w2 + ... + bmwm = Zw (min.) ...(ii)
Also, the primal-inequalities can be written as equations after
introducing the non-negative slack variables xn + 1, xn + 2, ...,xn + m as
follows :
𝑎11 𝑥1 + 𝑎12 𝑥2 + . . . + 𝑎1𝑛 𝑥𝑛 + 𝑥𝑛 + 1 = 𝑏1
𝑎21 𝑥1 + 𝑎22 𝑥2 + . . . + 𝑎2𝑛 𝑥𝑛 + 𝑥𝑛 + 2 = b2 }...(iii)
∶ ∶ ∶ ∶
𝑎𝑚1 𝑥1 + 𝑎𝑚2 𝑥2 + . . . + 𝑎𝑚𝑛 𝑥𝑛 + 𝑥𝑛 + 𝑚 = 𝑏𝑚
15
The primal problem in standard simplex form is: The dual problem in standard simplex form is:
Max. Zx = 40x1 + 50x2 + 0x3 + 0x4 Max Zw = – 3w1 – 5w2 + 0w3 + 0w4 + 0w5 + 0w6
subject to the constraints where Zw = – Zw subject to the constraints
2x1 + 3x2 + x3 + 0x4 = 3
2w1 + 8w2 – w3 + 0w4 + w5 + 0w6 = 40
5x1 + 4x2 + 0x3 + x4 = 5
3w1 + 4w2 + 0w3 – w4 + 0w5 + w6 = 50
andx10, x20, x30, x40.
andw1 0, w2 0, w3 0, w4 0, w5 0, w6 0.
PHASE 2
Basic CB XB X1 X2 X3 X4 Basic CB WB W1 W2 W3 W4
Variables Variable
x2 50 7/8 0 1 1/2 1/8 w2 –5 5/4 3 1
0 1 −
16 8
→ x1 40 3/16 1 0 – 1/4 – 3/4 w1 –3 15 1 1
1 0 −
4 2
x3=x4=0 Zx = 205/4
0 0 15 5/4 w3=w4= 0 Zw = – 205/4 0 0 3/16 7/8
primal linear programming problem can provide a solution to its dual. The
relationship described by arrows should be carefully checked and grasped.
Rules for Obtaining Dual Optimal Solution from That of Primal and
Conversely
The following rules are useful for obtaining the dual optimal solution from
that of primal and conversely :
Rule 1. The optimal value of primal objective function is equal to the
optimal value of the dual objective function, i.e. max. Zx = min. Zw.
Proof : This rule has already been proved in Duality Theorem.
Rule 2. Thej's(wherej = Zj – cj) for the slack vector in optimal (final)
simplex table for the primal are the values of the corresponding optimal
dual variables in the final simplex table for the dual.
20
Advantages of Duality
The knowledge of the dual of a primal LPP is important for the following
main reasons :
(i) As seen from above worked examples, the solution may be easier to
obtain through its dual than through the original (primal) problem. This is
true of cases in which the number of original variables in the primal
problem is considerably less than the number of slack or surplus variables.
(ii)Duality is not only restricted to linear programming problems but
frequently occurs in Economics, Physics and other fields of Mathematics
also.
(a) In Economics, it is used in the formulation of the input and output
systems. The economic interpretation of the dual is found useful in
making future decisions in the activities being programmed.
(b) In Physics, it is used in the parallel circuit and series circuit theory.
(iii)In game theory having two players A and B, it is used to find the
optimal strategies of the player B when he minimizes his losses. Then, by
duality, we can change the player A's problem into player B's problem and
vice-versa. If we solve the problem for one player, the solution to the other
can be easily read-off by using duality properties.
23
ILLUSTRATIVE EXAMPLES
Example 1. Solve the following primal LPP and its dual by simplex
method :
Max. Zx = 5x1 + 12x2 + 4x3, subject to the following constraints :
x1 + 2x2 + x3 5, 2x1 – x2 + 3x3 = 2 and x1, x2, x3 0.
Verify that the solution of primal can be read from the optimal table of the
dual and vice versa. Also, state the several steps involved in this process.
Solution.Optimal Solution of Primal Problem : By introducing the
slack variable x4 and the artificial variable x5 in the constraints of the
above problem and then applying Big-M method, we get the following
optimal primal simplex table :
Optimal Primal Simplex Table
cj→ 5 12 4 0 –M
Basic Variables CB XB X1 X2 X3 X4 X5
Here, we note that the starting basic variables are the slack and artificial variables
respectively, while both the optimal basic variables are original variables.
Now, the dual of given primal LPP is :
Minimize Zw = 5w1 + 2w2, subject to the following constraints :
w1 + 2w2 5, 2w1 – w2 12, w1 + 3w2 4, and w1 0, w2 is unrestricted.
Optimal Solution of Dual Problem : Since w2 is unrestricted, therefore,we
replace w2 byw2 – w2, where w2 0, w2 0. Introducing the surplus variables
w3, w4, w5 and the artificial variables w6, w7, w8 in the dual constraints, we shall
arrive at the following optimal dual simplex table :
24
An inspection of the optimal simplex table for the primal and dual reveals
the following interesting results :
(i)The basic variables in the starting solution of the primal are x4 and x5.
(ii)The dual variables corresponding to the primal constraint equations
containing x4 and x5 are w1 and w2 respectively.
(iii) Consider the net evaluations (j's) corresponding to x4 and x5 in the
optimal primal simplex table as follows :
Basic variables in starting primal solution : x4 x5
Corresponding optimal net evaluations : 4 = 29/5 5 = – 2/5 + M
Corresponding dual decision variables : w1 – 0 = w1w2 – (– M) = w2 + M
Example 2. Write down the dual of the following primal LPP and then
solve it :
Max. Zx = 4x1 + 2x2, subject to the following constraints :
– x1 – x2 – 3, – x1 + x2 – 2 and x1 0, x2 0
Hence or otherwise, write down the solution of the primal LPP.
26
It is now, clear form the optimal table that the coefficient of M in eachjis
non-negative and a non-zero artificial vector appears in the basis.
Therefore, the dual problem does not possess any optimal basic feasible
solution. Consequently, there exists no finite optimal solution to the given
problem.
27
= s2 = 0
s1 0 36 1 2 1 1 0 0 36
Simplex Table I
Since all j are 0, therefore, the final simplex table will give the
optimalsolution :x = 20, y = 0, z = 5 and max. P = Rs. 1050.
30
Solution. Introducing the slack variables s1 and s2, the given problem
becomes :
Max. Zx = 30x1 + 23x2 + 29x3 + 0s1 + 0s2,
subject to the following constraints : 6x1 + 5x2 + 3x3 + 1s1 + 0s2 = 26
4x1 + 2x2 + 5x3+ 0s1 + 1s2 = 7
and satisfying x1 0, x2 0, x3 0, s1 0, s2 0.
We now give below the successive simplex tables for the solution of this
problem :
Simplex Table
Basic cj→ 30 23 29 0 0 Min Ratio
Variables XB/Xk,Xk > 0
CB XB X1 X2 X3S1S2
s1 0 26 6 5 3 1 0 26/6
Thus, the optimal solution from the final simplex table is given by
x1 = 0, x2 = 7/2, x3 = 0 and Z*x = 161/2 i.e. max. Zx = 161/2.
\
32
Reading of solution to the dual of the given primal problem from the
final simplex table :
According to the rules, the optimal solution to the dual of the given
primal problem will be
w1 = 4 = 0, w2 = 5 = 23/2 and Z*x = Z*w = 161/2.
In this way, we can find the solution to the dual without actually
solving it.
Example 6. Use duality to solve the linear programming problem : Min
Zx= x1 – x2, subject to the constraints 2x1 + x2 2, – x1 – x2 1 and x1 0, x2 0.
We now construct the simplex table for solution of the auxiliary LPP in
phase I as follows :
Since all j 0 in the last simplex table and no artificial variable appears in
the basis matrix, therefore, an optimal basic feasible solution to the
auxiliary LPP has been attained. Hence, we proceed to phase II to get an
optimal basic feasible solution to the above dual problem.
Phase II. To find optimal solution of the Dual of Original LPP
We shall obtain the following simplex table for the solution during the
simplex procedure.
s2 0 1 2 –1 0 1 0 1/(–1), neg.
s2 0 2 1 0 –1 1 1 No need to calcu-
From the above table, we see that 3 is negative and all elements of vector
S1 (not in the basis) are negative, this indicates that the problem under
Basic cj→ 2 1 0 0 Min. Ratio
Variables CB WB W1W2S1 S2 WB/Wk,Wk > 0
w1 = s1 = 0 Zw = CBXB = 1 –3 0 –1 0↓ j=CBWB–cj
Simplex Table
Basic cj→ 1 2 0 0 Min. Ratio
Variables CB WB WB/Wk,Wk > 0
W1 W2 S1 S2
S1 0 3 1 2 1 0 3/2
s2 0 1 1 3 0 1 1/3, min.
w1 = w2 = 0 Zw = CBWB = 0 –1 – 2↑ 0 0↓ j=CBWj–cj
s1 0 2 0 –1 1 –1 No need to calcul-
From the above simplex table, we see that all j 0, therefore, we
have arrived at the optimal solution.
Thus, the optimal solution is :x1 = 4, x2 = 9 and max. profit = Rs. 48.
Using duality, the optimal solution to the dual problem can be read
directly from the optimal simplex table as given by :w1 = Rs. 1/4 or Rs.
0.25 per hour, w2 = Rs. 3/2 or Rs. 1.50 per hour and min. cost = Rs. 48.
Economic Interpretation of Dual Problem :
As discussed above, shadow prices are the opportunity costs that indicate
the potential profit that is lost by not having an additional unit of the
respective right hand side (resource) assuming that all right hand side
values are used optimally. Thus w1 = 0.25 and w2 = 1.50 means that
additional processing hours on machine M1 and M2 will increase the profit
by Rs. 0.25 and Rs. 1.50, respectively.
37