TD3 - Simplex Algorithm
TD3 - Simplex Algorithm
TD3 - Simplex Algorithm
max cT x
xB
I A =b
xN
x≥0
(a) Show that if x is a solution of the system (resp. of the system without the non-
negativity constraints) then x is still a solution after the new solution of the sys-
tem after the pivot operation (resp. of the system without the non-negativity con-
straints).
i.e. x satisfies the constraint before the pivot iff x satisfies the constraint after the
pivot.
(b) Prove the same for the objective function. Deduce that the value of the optimal
solution is not modified by a pivoting rule.
(c) Show that after a pivot operation, the vector b is still non negative.
(d) Deduce that if the Simplex algorithm starts from a BFS, then the current solution is
a BFS at any step.
1
Optimization and Approximation ENS Lyon, 2018-2019
(c) Let c∗ be the objective function in an optimal tableau. Prove that if a LP has sev-
eral solutions then a non-basic variable MUST have coefficient 0 in the objection
function c∗ .
(d) Show that the above condition is sufficient if the LP is non degenerate.
(e) (***) Is it sufficient in general?
(a) Prove that if the ratio does not give any constraint, then the LP has an arbitrarily
optimal value.
max z = 2x1 + x2 max z = 3x1 + 2x2
subject to x1 + x2 ≤ 10
x1 − x2 ≤ 10 x2 ≤ 6
(b) 2x1 − x2 ≤ 40 3x1 + 2x2 = 18
x1 , x2 ≥ 0 x1 , x2 ≥ 0
Solve geometrically and with the Sim- Solve geometrically and with the Sim-
plex algorithm this LP. plex algorithm this LP.
2
Optimization and Approximation ENS Lyon, 2018-2019
The entering variable will be the one with the most negative coefficient in the objective
vector. The leaving variable will be the candidate for departing which has the smallest
subscript. Prove that there is a cycle !
(a) Prove that any basic feasible solution for the LP has the following properties:
• Each xi is in the basis.
• Exactly one of {rj , sj } is in the basis for each j = 1, 2, . . . , d.
(b) So we can associate to each bfs, the subset S of indices j in {1, 2, . . . , d} such that
rj is in the basis. We denote by xS the corresponding (uniquely) bfs. Prove that
if d ∈ S but d ∈ / S 0 , then for the corresponding basic feasible solutions’ we have
0 0
xd > xd . Moreover, if S = S 0 ∪ {d}, then xSd = 1 − xSd .
S S
Note that in particular each of these basis gives a different vertex of the polytope.
(c) Conclude that the Simplex algorithm might need an exponential number of steps.
3
Optimization and Approximation ENS Lyon, 2018-2019
max x1 + x2
subject to
ax1 + bx2 ≤ c
x1 , x2 ≥ 0
(a) Let P be a polyhedron and H be a open halfspace (i.e. a set satisfying H = {x/ct x >
b} for some c, b). How can you test if P ∩ H is empty? How can you find a point of
P ∩ H if it exists?
(b) Answer the initial question (if necessary, several calls to the Simplex Algorithm can
me made).
4 Simplex algorithm
Exercise 11 - Phase I/II. (*)
Using Phase I/II method, solve the following LPs:
4
Optimization and Approximation ENS Lyon, 2018-2019