Sensitivity Analysis: An Example

Consider the linear program:
Maximize z = −5x
1
+5x
2
+13x
3
Subject to:
−x
1
+x
2
+3x
3
≤ 20 (1)
12x
1
+4x
2
+10x
3
≤ 90 (2)
x
1
, x
2
, x
3
≥ 0 .
After introducing two slack variables s
1
and s
2
and executing the Simplex algorithm to
optimality, we obtain the following final set of equations:
z +2x
3
+5s
1
= 100 , (0)
−x
1
+x
2
+3x
3
+s
1
= 20 , (1)
16x
1
−2x
3
−4s
1
+s
2
= 10 . (2)
Our task is to conduct sensitivity analysis by independently investigating each of a set of
nine changes (detailed below) in the original problem. For each change, we will use the
fundamental insight to revise the final set of equations (in tableau form) to identify a new
solution and to test the new solution for feasibility and (if applicable) optimality.
We will first recast the above equation systems into the following pair of initial and final
tableaus.
Initial Tableau: Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 5 −5 −13 0 0 0
s
1
0 −1 1 3 1 0 20
s
2
0 12 4 10 0 1 90
Final Tableau: Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 0 0 2 5 0 100
x
2
0 −1 1 3 1 0 20
s
2
0 16 0 −2 −4 1 10
The basic variables associated with this final tableau are x
2
and s
2
; therefore, the current
basic feasible solution is (x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 20, 0, 0, 10), which has an objective-
function value of 100.
An inspection of the initial tableau shows that the columns associated with z, s
1
, and s
2
form a 3 × 3 identity matrix. Therefore, the P matrix will come from the corresponding
columns in the final tableau. That is, we have
P =
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_;
1
and the final tableau equals the matrix product of this P and the initial tableau, i.e.,
T
F
= P×T
I
.
Our basic approach for dealing with parameter changes in the original problem is in two
steps. In the first step, we will revise the final tableau by multiplying the same P to the
new initial tableau; in other words, despite a revision in T
I
, we intend to follow the original
sequence of pivots. After producing a revised T
F
, we will, in the second step, take the
revised T
F
as the starting point and initiate any necessary further analysis of the revised
problem.
We now begin a detailed sensitivity analysis of this problem.
(a) Change the right-hand side of constraint (1) to 30.
Denote the right-hand-side constants in the original constraints as b
1
and b
2
. Then, the
proposed change is to revise b
1
from 20 to 30, while retaining the original value of b
2
at 90.
With this change, the RHS column in the initial tableau becomes
_
¸
_
0
30
90
_
¸
_.
Since the rest of the columns in the initial tableau stays the same, the only necessary revision
in T
F
will be in the RHS column. To determine this new RHS column, we multiply P to
the above new column to obtain:
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
0
30
90
_
¸
_ =
_
¸
_
150
30
−30
_
¸
_.
Since the basic variables in the final tableau are x
2
and s
2
, the solution associated with the
revised T
F
is (x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 30, 0, 0, −30). With a negative value for s
2
, this
(basic) solution is not feasible.
Geometrically speaking, increasing the value of b
1
from 20 to 30 means that we are relaxing
the first inequality constraint. Relaxing a constraint is tantamount to enlarging the feasible
set; therefore, one would expect an improved optimal objective-function value. The fact
that the revised solution above is not feasible is not a contradiction to this statement. It
only means that additional work is necessary to determine the new optimal solution.
What causes the infeasibility of the new solution? Recall that the original optimal solution is
(x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 20, 0, 0, 10). Since x
1
, x
3
, and s
1
are serving as nonbasic variables,
the defining equations for this solution are: x
1
= 0, x
3
= 0, and −x
1
+x
2
+3x
3
= 20. Now,
2
imagine an attempt to increase the RHS constant of the last equation from 20 to 20 + δ
(say) while maintaining these three equalities. As we increase δ (from 0), we will trace out
a family of solutions. That the new solution is infeasible simply means that if δ is made
sufficiently large (in this case, δ = 10), then this family of solutions will eventually exit the
feasible set.
More formally, suppose the original RHS column is revised to
_
¸
_
0
20 + δ
90
_
¸
_; or alternatively, to
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
δ
0
_
¸
_.
Then, after premultiplying this new column by P, we obtain
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
0
20 + δ
90
_
¸
_
=
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
_
_
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
δ
0
_
¸
_
_
_
_
=
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
0
δ
0
_
¸
_
=
_
¸
_
100
20
10
_
¸
_ +
_
¸
_

δ
−4δ
_
¸
_
=
_
¸
_
100 + 5δ
20 + δ
10 −4δ
_
¸
_.
Hence, with δ = 10, we indeed have s
2
= −30, which means that the original inequality
constraint 12x
1
+ 4x
2
+ 10x
3
≤ 90 is violated. Moreover, this calculation also shows that
in order for 10 − 4δ to remain nonnegative, δ cannot exceed 5/2. In other words, at
δ = 5/2, the family of solutions (0, 20 + δ, 0, 0, 10 − 4δ) “hits” the constraint equation
12x
1
+ 4x
2
+ 10x
3
= 90; and therefore, progressing further will produce solutions that are
outside the feasible set.
Interestingly, our analysis above holds even if we allow δ to assume a negative value. Such a
case corresponds to a tightening of the constraint −x
1
+x
2
+3x
3
≤ 20. A quick inspection
of
_
¸
_
100 + 5δ
20 + δ
10 −4δ
_
¸
_
3
shows that x
2
is reduced to 0 when δ reaches −20. It follows that in order to maintain
feasibility, and hence optimality (since the optimality test is not affected by a change in the
RHS column), of solutions of the form (0, 20 + δ, 0, 0, 10 − 4δ), the value of δ must stay
within the range [−20, 5/2].
Another important observation regarding the above calculation is that the optimal objective-
function value will increase from 100 to 100 + 5δ, provided that δ is sufficiently small (so
that we remain within the feasible set). If we interpret the value of b
1
as the availability of
a resource, then this observation implies that for every additional unit of this resource, the
optimal objective-function value will increase by 5. Thus, from an economics viewpoint, we
will be unwilling to pay more than 5 (dollars) for an additional unit of this resource. For
this reason, the value 5 is called the shadow price of this resource.
It is interesting to note that the shadow price of the first resource (5, in this case) can be
read directly from the top entry in the second column of P.
It is possible to derive a new optimal solution for the proposed new problem with δ = 10.
The standard approach for doing this is to start from the revised final tableau and apply
what is called the dual Simplex algorithm. As this algorithm is more advanced, we will not
attempt to solve this new problem to optimality.
(b) Change the right-hand side of constraint (2) to 70.
Since the original value of b
2
is 90, this is an attempt to reduce the availability of the second
resource by 20. The analysis is similar to that in part (a). Again, we will write the new
RHS column in the initial tableau as
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
0
δ
_
¸
_,
where δ is targeted to assume the value −20. After premultiplying this new column by P,
we obtain
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
_
_
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
0
δ
_
¸
_
_
_
_
=
_
¸
_
100
20
10
_
¸
_ +
_
¸
_
0
0
δ
_
¸
_
=
_
¸
_
100
20
10 + δ
_
¸
_.
4
Hence, for all δ within the range [−10, ∞), solutions of the form (0, 20, 0, 0, 10 + δ) will
remain optimal.
With the particular choice of δ = −20, we have
_
¸
_
100
20
10 + δ
_
¸
_ =
_
¸
_
100
20
−10
_
¸
_.
It follows that the new solution (0, 20, 0, 0, −10) is infeasible. As in part (a), we will not
attempt to derive a new optimal solution.
The shadow price of the second resource can be read directly from the top entry in the third
column of P. In this case, it is given by 0. That the shadow price of the second resource is
equal to 0 is expected. It is a consequence of the fact that in the current optimal solution,
we have s
2
= 10 and hence there is already an excess in the supply of the second resource.
In fact, we will have an over supply as long as the availability of the second resource is no
less than 80 (which corresponds to δ = −10).
(c) Change b
1
and b
2
to 10 and 100, respectively.
Again, we will first consider a revision of the RHS column in T
I
of the form:
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
δ
1
δ
2
_
¸
_,
where δ
1
and δ
2
are two independent changes. After premultiplying this new column by P,
we obtain
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
_
_
_
¸
_
0
20
90
_
¸
_ +
_
¸
_
0
δ
1
δ
2
_
¸
_
_
_
_ =
_
¸
_
100 + 5δ
1
20 + δ
1
10 −4δ
1
+ δ
2
_
¸
_.
With δ
1
= −10 and δ
2
= 10, the new RHS column in T
F
is:
_
¸
_
50
10
60
_
¸
_.
Since the new solution (x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 10, 0, 0, 60) is feasible, it is also optimal.
The new optimal objective-function value is 50.
(d) Change the coefficient of x
3
in the objective function to c
3
= 8 (from c
3
= 13).
5
Consider a revision in the value of c
3
by δ; that is, let c
3
= 13 + δ. Then, the x
3
-column in
T
I
is revised to
_
¸
_
−13 −δ
3
10
_
¸
_; or alternatively, to
_
¸
_
−13
3
10
_
¸
_ +
_
¸
_
−δ
0
0
_
¸
_.
From the fundamental insight, the corresponding revision in the x
3
-column in T
F
is
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
_
_
_
¸
_
−13
3
10
_
¸
_ +
_
¸
_
−δ
0
0
_
¸
_
_
_
_ =
_
¸
_
2
3
−2
_
¸
_ +
_
¸
_
−δ
0
0
_
¸
_
=
_
¸
_
2 −δ
3
−2
_
¸
_.
Therefore, if δ = −5, which corresponds to c
3
= 8, then the new x
3
-column in T
F
is
explicitly given by
_
¸
_
2 −δ
3
−2
_
¸
_ =
_
¸
_
2 −(−5)
3
−2
_
¸
_ =
_
¸
_
7
3
−2
_
¸
_.
Observe that the x
3
-column is the only column in T
F
that requires a revision, the variable
x
3
is nonbasic, and the coefficient of x
3
in the revised R
0
is positive (7, that is). It follows
that the original optimal solution (x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 20, 0, 0, 10) remains optimal.
More generally, an inspection of the top entry in the new x
3
-column,
_
¸
_
2 −δ
3
−2
_
¸
_,
reveals that the original optimal solution will remain optimal for all δ such that 2 −δ ≥ 0,
i.e., for all δ in the range (−∞, 2].
(e) Change c
1
to −2, a
11
to 0, and a
21
to 5.
This means that the x
1
-column in T
I
is revised from
_
¸
_
5
−1
12
_
¸
_ to
_
¸
_
2
0
5
_
¸
_.
6
Since the corresponding new column in T
F
is
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
2
0
5
_
¸
_ =
_
¸
_
2
0
5
_
¸
_,
where the top entry, 2, is positive, and since x
1
is nonbasic in T
F
, we see that the original
optimal solution remains optimal.
(f ) Change c
2
to 6, a
12
to 2, and a
22
to 5.
This means that the x
2
-column in T
I
is revised from
_
¸
_
−5
1
4
_
¸
_ to
_
¸
_
−6
2
5
_
¸
_.
The fundamental insight implies that the corresponding new x
2
-column in T
F
is
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
−6
2
5
_
¸
_ =
_
¸
_
4
2
−3
_
¸
_.
The fact that this new column is no longer of the form
_
¸
_
0
1
0
_
¸
_
indicates that x
2
cannot serve as a basic variable in R
1
. It follows that a pivot in the
x
2
-column is needed to restore x
2
back to the status of a basic variable. More explicitly,
the revised final tableau is
Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 0 4 2 5 0 100
− 0 −1 2 3 1 0 20
s
2
0 16 −3 −2 −4 1 10
and we will execute a pivot with the x
2
-column as the pivot column and R
1
as the pivot
row. After this pivot, we obtain
Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 2 0 −4 3 0 60
x
2
0 −1/2 1 3/2 1/2 0 10
s
2
0 29/2 0 5/2 −5/2 1 40
.
7
Since x
3
now has a negative coefficient in R
0
, indicating that the new solution is not optimal,
the Simplex algorithm should be restarted to derive a new optimal solution (if any).
(g) Introduce a new variable x
4
with c
4
= 10, a
14
= 3, and a
24
= 5.
This means that we need to introduce the new x
4
-column
_
¸
_
−10
3
5
_
¸
_
into the initial tableau. (The precise location of this new column is not important.) The
corresponding new column in the final tableau will be
_
¸
_
1 5 0
0 1 0
0 −4 1
_
¸
_ ×
_
¸
_
−10
3
5
_
¸
_ =
_
¸
_
5
3
−7
_
¸
_.
Since this column has a positive entry at the top and since x
4
is nonbasic, the current
optimal solution remains optimal. In an application, this means that there is insufficient
incentive to engage in the new “activity” x
4
.
(h) Introduce a new constraint 2x
1
+ 3x
2
+ 5x
3
≤ 50.
After adding a new slack variable s
3
, this inequality constraint becomes 2x
1
+3x
2
+5x
3
+s
3
=
50. Next, we incorporate this equation into the final tableau to obtain
Basic z x
1
x
2
x
3
s
1
s
2
s
3
Variable 1 0 0 2 5 0 0 100
− 0 −1 1 3 1 0 0 20
s
2
0 16 0 −2 −4 1 0 10
s
3
0 2 3 5 0 0 1 50
.
Observe that x
2
participates in the new equation and, therefore, cannot serve as the basic
variable for R
1
. To rectify this situation, we will execute the row operation (−3) ×R
1
+R
3
.
This yields
Basic z x
1
x
2
x
3
s
1
s
2
s
3
Variable 1 0 0 2 5 0 0 100
x
2
0 −1 1 3 1 0 0 20
s
2
0 16 0 −2 −4 1 0 10
s
3
0 5 0 −4 −3 0 1 −10
.
8
With s
3
= −10, the new basic solution is not feasible. We will not attempt to continue the
solution of this new problem (as it is now necessary to apply the dual Simplex algorithm).
(i) Change constraint (2) to 10x
1
+ 5x
2
+ 10x
3
≤ 100.
With this revision, the initial tableau becomes
Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 5 −5 −13 0 0 0
s
1
0 −1 1 3 1 0 20
s
2
0 10 5 10 0 1 100
.
After premultiplying this by P, we obtain the revised final tableau below.
Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 0 0 2 5 0 100
− 0 −1 1 3 1 0 20
s
2
0 14 1 −2 −4 1 20
Observe that x
2
participates in R
2
and, therefore, cannot serve as the basic variable for R
1
.
To rectify this situation, we will execute the row operation (−1) ×R
1
+ R
2
. This yields
Basic z x
1
x
2
x
3
s
1
s
2
Variable 1 0 0 2 5 0 100
x
2
0 −1 1 3 1 0 20
s
2
0 15 0 −5 −5 1 0
.
Therefore, the new optimal solution is (x
1
, x
2
, x
3
, s
1
, s
2
) = (0, 20, 0, 0, 0).
9

we will revise the final tableau by multiplying the same P to the new initial tableau. Now. and s1 are serving as nonbasic variables. s1 . 10). (a) Change the right-hand side of constraint (1) to 30. we will. With a negative value for s2 .. 2      . x3 . 20.e. and −x1 + x2 + 3x3 = 20. x3 . in the second step. TF = P × TI . one would expect an improved optimal objective-function value. the defining equations for this solution are: x1 = 0. the solution associated with the revised TF is (x1 . s1 . What causes the infeasibility of the new solution? Recall that the original optimal solution is (x1 . 0. x2 . s2 ) = (0. With this change. We now begin a detailed sensitivity analysis of this problem. −30). we intend to follow the original sequence of pivots. in other words. s2 ) = (0. 90 Since the rest of the columns in the initial tableau stays the same. Geometrically speaking. In the first step. increasing the value of b1 from 20 to 30 means that we are relaxing the first inequality constraint. we multiply P to the above new column to obtain:   150 0 1 5 0       1 0  ×  30  =  30  . To determine this new RHS column. Then. It only means that additional work is necessary to determine the new optimal solution. the RHS column in the initial tableau becomes  0    30  . 0.  0 −30 90 0 −4 1 Since the basic variables in the final tableau are x2 and s2 . the proposed change is to revise b1 from 20 to 30. Relaxing a constraint is tantamount to enlarging the feasible set. x3 . Since x1 . x3 = 0. the only necessary revision in TF will be in the RHS column. Denote the right-hand-side constants in the original constraints as b1 and b2 . Our basic approach for dealing with parameter changes in the original problem is in two steps. x2 . i. take the revised TF as the starting point and initiate any necessary further analysis of the revised problem. The fact that the revised solution above is not feasible is not a contradiction to this statement.and the final tableau equals the matrix product of this P and the initial tableau. 0. while retaining the original value of b2 at 90. 30. this (basic) solution is not feasible. despite a revision in TI . After producing a revised TF . therefore. 0.

the family of solutions (0. 0. Interestingly. More formally. A quick inspection of   100 + 5δ   20 + δ   10 − 4δ 3                        . then this family of solutions will eventually exit the feasible set. and therefore. 0. 10 − 4δ Hence. δ cannot exceed 5/2. 90 90 0 Then. we will trace out a family of solutions. In other words. Such a case corresponds to a tightening of the constraint −x1 + x2 + 3x3 ≤ 20. to  20  +  δ  . δ = 10). we obtain       1 5 0 0     1 0  ×  20 + δ   0 0 −4 1 90 1 5 0 0 0       1 0  ×  20  +  δ  =  0 0 −4 1 90 0 0 1 5 0 0 1 5 0         1 0 × δ  1 0  ×  20  +  0 =  0 0 0 −4 1 90 0 −4 1 5δ 100     δ  =  20  +  −4δ 10 100 + 5δ   =  20 + δ  . or alternatively. after premultiplying this new column by P. That the new solution is infeasible simply means that if δ is made sufficiently large (in this case. progressing further will produce solutions that are outside the feasible set. suppose the original RHS column is revised to  0 0 0        20 + δ  . we indeed have s2 = −30. which means that the original inequality constraint 12x1 + 4x2 + 10x3 ≤ 90 is violated. our analysis above holds even if we allow δ to assume a negative value. with δ = 10. 10 − 4δ) “hits” the constraint equation 12x1 + 4x2 + 10x3 = 90. Moreover. 20 + δ. As we increase δ (from 0). at δ = 5/2. this calculation also shows that in order for 10 − 4δ to remain nonnegative.imagine an attempt to increase the RHS constant of the last equation from 20 to 20 + δ (say) while maintaining these three equalities.

we will be unwilling to pay more than 5 (dollars) for an additional unit of this resource. 20 + δ. δ 90 where δ is targeted to assume the value −20. we obtain     1 5 0 0 0       1 0  ×  20  +  0   0 0 −4 1 90 δ 0 100     =  20  +  0  δ 10 100   20  .shows that x2 is reduced to 0 when δ reaches −20. (b) Change the right-hand side of constraint (2) to 70. we will not attempt to solve this new problem to optimality. It is possible to derive a new optimal solution for the proposed new problem with δ = 10. Another important observation regarding the above calculation is that the optimal objectivefunction value will increase from 100 to 100 + 5δ. and hence optimality (since the optimality test is not affected by a change in the RHS column). 0. It is interesting to note that the shadow price of the first resource (5. this is an attempt to reduce the availability of the second resource by 20. For this reason. Since the original value of b2 is 90. =  10 + δ 4            . provided that δ is sufficiently small (so that we remain within the feasible set). The analysis is similar to that in part (a). from an economics viewpoint. As this algorithm is more advanced. 5/2]. Again. 10 − 4δ). the optimal objective-function value will increase by 5. then this observation implies that for every additional unit of this resource. The standard approach for doing this is to start from the revised final tableau and apply what is called the dual Simplex algorithm. we will write the new RHS column in the initial tableau as  0 0      20  +  0  . in this case) can be read directly from the top entry in the second column of P. It follows that in order to maintain feasibility. 0. the value 5 is called the shadow price of this resource. of solutions of the form (0. If we interpret the value of b1 as the availability of a resource. the value of δ must stay within the range [−20. Thus. After premultiplying this new column by P.

That the shadow price of the second resource is equal to 0 is expected. we obtain      0 1 5 0      1 0  ×  20  +   0 90 0 −4 1 After premultiplying this new column by P. 0. 5 . 0.  (d) Change the coefficient of x3 in the objective function to c3 = 8 (from c3 = 13). 20. δ2 10 − 4δ 1 + δ 2       With δ 1 = −10 and δ 2 = 10. we will have an over supply as long as the availability of the second resource is no less than 80 (which corresponds to δ = −10). In fact. −10) is infeasible. we have s2 = 10 and hence there is already an excess in the supply of the second resource. x3 . 20. s1 . 60 Since the new solution (x1 .    (c) Change b1 and b2 to 10 and 100. As in part (a). 0.  10 + δ −10 It follows that the new solution (0. ∞). we will first consider a revision of the RHS column in TI of the form:  0 0      20  +  δ 1  . we will not attempt to derive a new optimal solution. it is also optimal. In this case. it is given by 0. It is a consequence of the fact that in the current optimal solution. 0. 0. 100 + 5δ 1 0    δ 1  =  20 + δ 1  . 0. The new optimal objective-function value is 50. we have  100 100     20  =  20  . s2 ) = (0. The shadow price of the second resource can be read directly from the top entry in the third column of P. respectively. solutions of the form (0.Hence. With the particular choice of δ = −20. 60) is feasible. 90 δ2 where δ 1 and δ 2 are two independent changes. for all δ within the range [−10. the new RHS column in TF is:  50    10  . 10 + δ) will remain optimal. Again. x2 . 10.

5 12 6    .  10 10 0 From the fundamental insight. and a21 to 5. an inspection of the top entry in the new x3 -column. 10) remains optimal. for all δ in the range (−∞. 2]. This means that the x1 -column in TI is revised from  2 5      −1  to  0  .  −2 reveals that the original optimal solution will remain optimal for all δ such that 2 − δ ≥ 0.Consider a revision in the value of c3 by δ. then the new x3 -column in TF is explicitly given by             7 2 − (−5) 2−δ       3  =  3 . that is).e. Then. which corresponds to c3 = 8. and the coefficient of x3 in the revised R0 is positive (7. s2 ) = (0.       2−δ   3 . 20. the variable x3 is nonbasic. i. if δ = −5. let c3 = 13 + δ. a11 to 0. 0. It follows that the original optimal solution (x1 . More generally. or alternatively. x2 . =  −2 Therefore.. s1 . the x3 -column in TI is revised to  −δ −13 − δ −13       3  . to  3  +  0 . 0.  (e) Change c1 to −2. x3 . that is. the corresponding revision in the x3 -column in TF is       1 5 0 −13 −δ 2 −δ           0 1 0  ×  3  +  0  =  3  +  0   0 −4 1 10 0 −2 0 2−δ   3 . 3 =  −2 −2 −2 Observe that the x3 -column is the only column in TF that requires a revision.

a12 to 2. we see that the original optimal solution remains optimal. . It follows that a pivot in the x2 -column is needed to restore x2 back to the status of a basic variable. and a22 to 5. (f ) Change c2 to 6. This means that the x2 -column in TI is revised from       −5 −6      1  to  2  . After this pivot. 2.  −3 5 0 −4 1 The fact that this new column is no longer of the form       0    1  0 indicates that x2 cannot serve as a basic variable in R1 .Since the corresponding new column in TF is  2 1 5 0 2       0 1 0  ×  0  =  0 . we obtain Basic Variable x2 s2 z 1 0 0 x1 2 −1/2 29/2 x2 0 1 0 7 x3 −4 3/2 5/2 s1 3 1/2 −5/2 s2 0 0 1 60 10 40 .  0 −4 1 5 5 where the top entry. More explicitly. the revised final tableau is Basic Variable − s2 z 1 0 0 x1 0 −1 16 x2 4 2 −3 x3 2 3 −2 s1 5 1 −4 s2 0 0 1  100 20 10 and we will execute a pivot with the x2 -column as the pivot column and R1 as the pivot row. 4 5 The fundamental insight implies that the corresponding new x2 -column in TF is     4 −6 1 5 0       0 1 0  ×  2  =  2 . and since x1 is nonbasic in TF . is positive.

Next. indicating that the new solution is not optimal. this inequality constraint becomes 2x1 +3x2 +5x3 +s3 = 50. (g) Introduce a new variable x4 with c4 = 10. we incorporate this equation into the final tableau to obtain Basic Variable − s2 s3 z 1 0 0 0 x1 0 −1 16 2 x2 0 1 0 3 x3 2 3 −2 5 s1 5 1 −4 0 s2 0 0 1 0 s3 0 0 0 1 100 20 10 50 . After adding a new slack variable s3 . therefore. (The precise location of this new column is not important. a14 = 3. Observe that x2 participates in the new equation and.) The corresponding new column in the final tableau will be   1 5 0 −10 5       0 1 0 × 3  =  3 . we will execute the row operation (−3) × R1 + R3 . This means that we need to introduce the new x4 -column  −10   3   5 into the initial tableau. 8 .  0 −4 1 5 −7 Since this column has a positive entry at the top and since x4 is nonbasic. To rectify this situation. and a24 = 5. cannot serve as the basic variable for R1 . this means that there is insufficient incentive to engage in the new “activity” x4 .Since x3 now has a negative coefficient in R0 .      (h) Introduce a new constraint 2x1 + 3x2 + 5x3 ≤ 50. This yields Basic Variable x2 s2 s3 z 1 0 0 0 x1 0 −1 16 5 x2 0 1 0 0 x3 2 3 −2 −4 s1 5 1 −4 −3 s2 0 0 1 0 s3 0 0 0 1 100 20 10 −10 . the current optimal solution remains optimal. In an application. the Simplex algorithm should be restarted to derive a new optimal solution (if any).

the new optimal solution is (x1 . the initial tableau becomes Basic Variable s1 s2 z 1 0 0 x1 5 −1 10 x2 −5 1 5 x3 −13 3 10 s1 0 1 0 s2 0 0 1 0 20 100 . we will execute the row operation (−1) × R1 + R2 . To rectify this situation. x3 . This yields Basic Variable x2 s2 z 1 0 0 x1 0 −1 15 x2 0 1 0 x3 2 3 −5 s1 5 1 −5 s2 0 0 1 100 20 0 . 0). With this revision.With s3 = −10. 20. x2 . the new basic solution is not feasible. we obtain the revised final tableau below. (i) Change constraint (2) to 10x1 + 5x2 + 10x3 ≤ 100. After premultiplying this by P. 0. 0. s2 ) = (0. s1 . therefore. 9 . We will not attempt to continue the solution of this new problem (as it is now necessary to apply the dual Simplex algorithm). cannot serve as the basic variable for R1 . Basic Variable − s2 z 1 0 0 x1 0 −1 14 x2 0 1 1 x3 2 3 −2 s1 5 1 −4 s2 0 0 1 100 20 20 Observe that x2 participates in R2 and. Therefore.

Sign up to vote on this title
UsefulNot useful