Paper 3, Section II, 21H

Optimization | Part IB, 2012

For given positive real numbers (cij:i,j{1,2,3})\left(c_{i j}: i, j \in\{1,2,3\}\right), consider the linear program

P:minimizei=13j=13cijxij,P: \operatorname{minimize} \sum_{i=1}^{3} \sum_{j=1}^{3} c_{i j} x_{i j},

subject to i=13xij1\sum_{i=1}^{3} x_{i j} \leqslant 1 for all j,j=13xij1j, \quad \sum_{j=1}^{3} x_{i j} \geqslant 1 for all ii,

and xij0x_{i j} \geqslant 0 for all i,ji, j.

(i) Consider the feasible solution xx in which x11=x12=x22=x23=x31=x33=1/2x_{11}=x_{12}=x_{22}=x_{23}=x_{31}=x_{33}=1 / 2 and xij=0x_{i j}=0 otherwise. Write down two basic feasible solutions of PP, one of which you can be sure is at least as good as xx. Are either of these basic feasible solutions of PP degenerate?

(ii) Starting from a general definition of a Lagrangian dual problem show that the dual of PP can be written as

D:maximizeλi0,μi0i=13(λiμi) subject to λiμjcij for all i,j.D: \operatorname{maximize}_{\lambda_{i} \geqslant 0, \mu_{i} \geqslant 0} \sum_{i=1}^{3}\left(\lambda_{i}-\mu_{i}\right) \quad \text { subject to } \lambda_{i}-\mu_{j} \leqslant c_{i j} \text { for all } i, j .

What happens to the optimal value of this problem if the constraints λi0\lambda_{i} \geqslant 0 and μi0\mu_{i} \geqslant 0 are removed?

Prove that x11=x22=x33=1x_{11}=x_{22}=x_{33}=1 is an optimal solution to PP if and only if there exist λ1,λ2,λ3\lambda_{1}, \lambda_{2}, \lambda_{3} such that

λiλjcijcjj, for all i,j\lambda_{i}-\lambda_{j} \leqslant c_{i j}-c_{j j}, \quad \text { for all } i, j

[You may use any facts that you know from the general theory of linear programming provided that you state them.]

Typos? Please submit corrections to this page on GitHub.