Sei sulla pagina 1di 4

MS-E2140 Linear Programming Thu 01.10.

2015
Exercise 7 Week 4 U344
This weeks homework https://mycourses.aalto.fi/mod/folder/view.php?id=39963 is due no
later than Tuesday 13.10.2015 23:55.

Exercise 7.1 Unboundedness and duality

Consider the LP:

(P ) min c0 x
s.t. Ax = b
x0

where A Rmn and b Rm . Show that if (P ) has a finite optimal solution, then the new problem (P )
cannot be unbounded
obtained from (P ) by replacing the right hand side vector b with another one b,
no matter what value the components of b can take.

Solution

The dual of P is

(D) max p0 b
s.t. p0 A c
p Rm

while the dual of (P ) is

(D) max p0 b

s.t. p0 A c
p Rm

By strong duality, if (P ) has finite optimal solution, then (D) also has finite optimal solution. Since (D)
and (D) have the same feasible region and (D) is feasible (D) must also be feasible. By weak duality,
since (D) is feasible (P ) cannot be unbounded.

Exercise 7.2 Dual Simplex

(a) Solve the problem below by using the dual Simplex method. Report both the primal and dual
optimal solutions x and p associated with the optimal basis.
(b) Write the dual and use duality to verify that x and p are optimal.

min 2x1 + x3
s.t. 1/4x1 1/2x2 3/4
8x1 + 12x2 20
x1 + 1/2x2 x3 1/2
9x1 + 3x2 6
x1 , x2 , x3 0

Solution

1
MS-E2140 Linear Programming Thu 01.10.2015
Exercise 7 Week 4 U344

After transforming the problem in standard form and multiplying constraint 4 by -1 we obtain the
equivalent problem

min 2x1 + x3
s.t. 1/4x1 1/2x2 + x4 = 3/4
8x1 + 12x2 + x5 = 20
x1 + 1/2x2 x3 + x6 = 1/2
9x1 3x2 + x7 = 6
x1 , . . . , x 7 0
0
A basis matrix is B = (A4 , A5 , A6 , A7 ). The corresponding basic variables vector is xB = (x4 , x5 , x6 , x7 ) =
0
(3/4, 20, 1/2, 6) and cB = (c4 , c5 , c6 , c7 ) = (0, 0, 0, 0).

Since c0B = 0 the reduced cost of any variable xj is cj = cj c0B B1 Aj = cj . The reduced costs
c0 = (2, 0, 1, 0, 0, 0, 0). All reduced costs are non negative: B is dual-feasible and can be
vector is thus
used to start the dual Simplex.

Initial Simplex tableau

x1 x2 x3 x4 x5 x6 x7
0 2 0 1 0 0 0 0
x4 = -3/4 -1/4 -1/2 0 1 0 0 0
x5 = 20 8 12 0 0 1 0 0
x6 = -1/2 1 1/2 -1 0 0 1 0
x7 = 6 -9 -3 0 0 0 0 1
0 2
We select x4 as the leaving variable, and x2 is the entering variable since 1/2 < 1/4 .

x1 x2 x3 x4 x5 x6 x7
0 2 0 1 0 0 0 0
x2 = 3/2 1/2 1 0 -2 0 0 0
x5 = 2 2 0 0 24 1 0 0
x6 = -5/4 3/4 0 -1 1 0 1 0
x7 = 21/2 -15/2 0 0 -6 0 0 1

x1 x2 x3 x4 x5 x6 x7
-5/4 11/4 0 0 1 0 1 0
x2 = 3/2 1/2 1 0 -2 0 0 0
x5 = 2 2 0 0 24 1 0 0
x3 = 5/4 -3/4 0 1 -1 0 -1 0
x7 = 21/2 -15/2 0 0 -6 0 0 1
The optimal primal solution is x 0 = (0, 3/2, 5/4, 0, 2, 0, 21/2). Since columns A4 , A5 , A6 , A7 of vari-
ables x4 , x5 , x6 , x7 (in that order) form the identity matrix in the initial tableau and the corresponding
variables have cost 0, the reduced costs

cj = cj c0B B1 Aj = cj p0 Aj = p0 Aj

of x4 , x5 , x6 , x7 in the optimal tableau correspond to the dual variables p1 , p2 , p3 , p4 , respectively, changed


by sign. For example, c4 = p0 A4 = p0 e1 = p1 and thus p1 = c4 = 1. Therefore, the optimal
dual solution is p 0 = ( c4 ,
p1 , p2 , p3 , p4 ) = ( c5 ,
c6 ,
c7 ) = (1, 0, 1, 0).

(b) The primal (after multiplying constraint 4 by -1) and the corresponding dual is:

2
MS-E2140 Linear Programming Thu 01.10.2015
Exercise 7 Week 4 U344

min 2x1 + x3
max 3/4p1 + 20p2 1/2p3 + 6p4
s.t. 1/4x1 1/2x2 3/4
s.t. 1/4p1 + 8p2 + p3 9p4 2
8x1 + 12x2 20
1/2p1 + 12p2 + 1/2p3 3p4 0
x1 + 1/2x2 x3 1/2
p3 1
9x1 3x2 6
p1 , p2 , p3 , p4 0
x1 , x2 , x3 0

We can use the corollary to the weak duality to verify that x and p are optimal by verifying that
0 = (1, 0, 1, 0) is feasible (constraints -3/4 2, 0 0
they are both feasible and have the same cost. p
0 b = 3/4 + 1/2 = 5/4. The primal is feasible and has the same cost 5/4.
and 1 1). The dual cost is p
Thus x and p
are optimal.

Exercise 7.3 Lamp Production


Course book Exercise 5.6
Company A has agreed to supply the following quantities of special lamps to Company B during the
next 4 months:
Month January February March April
Units 150 160 225 180

Company A can produce a maximum of 160 lamps per month at a cost of 35 euros per unit. Additional
lamps can be purchased from company C at a cost of 50 euros per lamp. Company A incurs an inventory
holding cost of 5 euros per month for each lamp held in inventory.

(a) Formulate the problem that Company A is facing as a linear program.


(b) Solve the problem using a linear programming package.
(c) Company A is considering some preventive maintenance during one of the first three months. If
maintenance is scheduled for January, the company can manufacture only 151 units (instead of
160); similarly, the maximum possible production if maintenance is scheduled for February or
March is 153 and 155 units, respectively. Which maintenance schedule would you recommend and
why?
(d) Company D has offered to supply up to 50 lamps (total) to company A during either January,
February or March. Company D charges 45 euros per lamp. Should Company A buy lamps from
Company D? If yes, when and how many lamps should Company A purchase, and what is the
impact of this decision on the total cost?
(e) Company C has offered to lower the price of units supplied to Company A during February. What
is the minimum decrease that would make this offer attractive to Company A?

(f) Because of anticipated increases in interest rates, the holding cost per lamp is expected to increase
to 8 euros per unit in February. How does this change affect the total cost and the optimal solution?
(g) Company B has just informed Company A that is requires only 90 units in January (instead of 150
requested previously). Calculate the upper and lower bounds for the impact of this order in the
optimal cost using information from the optimal solution of the original problem.

Solution

(a) Let di be the demand for month i. Then the decision variables are

3
MS-E2140 Linear Programming Thu 01.10.2015
Exercise 7 Week 4 U344

xi Manufactured lamps in month i = 1, . . . , 4


yi Lamps purchased from Company C in month i = 1, . . . , 4
zi Inventory in month i = 1, . . . , 4

4
X 4
X 4
X
min 35 xi + 50 yi + 5 zi
i=1 i=1 i=1
s.t. x1 + y1 z1 = d1 (1)
x2 + y2 + z1 z2 = d2 (2)
x3 + y3 + z2 z3 = d3 (3)
x4 + y4 + z3 z4 = d4 (4)
xi 160 (5) i = 1, . . . , 4
xi , yi , zi 0 i = 1, . . . , 4

(b) The optimal solution is xi = 160 for all i, y1 = y2 = 0, y3 = 55, y4 = 20, z1 = z2 = 10 and
z3 = z4 = 0, which give the optimal cost of 26250.
(c) Let us consider the dual variables, that is shadow prices, associated with constraints (5). These
are 5, 10, 15 and 15 and their interpretation is that if we could produce one unit less in the
corresponding month, the total cost would decrease by the amount of the shadow price (assuming
that the basis remains optimal). The proposed preventive maintenance will decrease capacity from
160 to 151 (January), 153 (February) and 155 (March), which are all greater than the allowable
lower bounds (150 for January, 150 for February, 0 for March). Thus, the dual prices are valid.
The cost of the maintenance is thus in January 5 (9) = 45, in February 10 (7) = 70
and in March 15 5 = 75, wherefore the maintenance is best scheduled in January.
(d) In the optimal solution the production levels x1 = x2 = 160, and the demand for January and
February are 150 and 160 units respectively. Since producing the lamps costs only 35 it is not
wise to buy from company D during the first two months. However, in March y3 = 55, that is, 55
lamps are bought from company C at a unit price 50. Since the last chance to buy (a maximum
of 50) lamps from D is in March with unit price 45, it makes sense to buy the 50 first lamps from
company D and the rest 5 from company C. This way the optimal solution remains the same, and
we have saved 50 (50 45) = 250.
(e) Currently, the company does not buy lamps from Company C in February, which is indicated by
that at optimum, y2 = 0. The reduced cost of this variable is 5, and if this reduced cost would
decrease by more than 5, the company should start buying lamps in February from Company C
(because then all of the reduced costs would not be non-negative, indicating a non-optimal solution,
where y2 should be taken into the basis because it is the only variable with a negative reduced cost).
Since y2 is non-basic, a discount of more than 5/lamp in the cost coefficient is thus sufficient.
(f) Since the allowable upper bound for the cost coefficient of z2 is 10, and the expected cost increases
by 3 from 5 to 8, the current solution remains optimal. Thus, the optimal solution remains
the same, but the cost increases by 310 = 30 (z2 = 10 units and 3 increase per unit).
(g) The January demand (i.e., the rhs value of constraint 1) decreases by 60 units from 150 to 90 units,
whereas the allowable lower bound for this value is 95 units. Thus, the current solution does not
remain optimal after the decrease. The optimal cost as a function of the rhs value is convex. It is
also increasing as the more lamps we produce, the higher the cost. The dual price for this constraint
is 40, wherefore a lower bound for the cost is thus 262506040= 23850. Since the maximum
allowable decrease is 55 units (from 150 to 95), an upper bound is 2625055 40= 24050. If
we solve the problem optimally with 90 instead of 150 (wherefore the basis changes), we find that
the new cost is 23875, which lies between the bounds we derived using the dual prices.