Homepage Solution manuals Dimitris Bertsimas Introduction to Linear Optimization Exercise 7.30 (The primal-dual method as steepest dual ascent)

Exercise 7.30 (The primal-dual method as steepest dual ascent)

Consider the dual ascent algorithm. Show that the choice of the set S in the primal-dual method maximizes ( d s ) b over all sets S for which d s is a feasible direction.

Answers

We content ourselves with a high-level description of the proof. Choose an arbitrary feasible direction, say D. We can transform it to S with two moves that doesn’t make b d D smaller. Let

Q : = { v | b v < 0 } S ¯ D .

We delete all of Q and any vertex that has a path in the graph(with balanced edges to keep it feasible. Obviously Q + : = { v | b v > 0 } S ¯ D is also removed. Because every vertex of Q + has a path to a vertex of Q . Note that all the outputed flow of Q + goes through Q . So v Q + b v = f Q + out f Q in = f Q out v Q | b v | , the first equality comes from the fact that Q weren’t markes though there is a edge from s to them. Notice the change in b d D , it increases with v Q | b v | v Q + b v 0 .
In a similar manner we can define: Q : = { v | b v < 0 } D ¯ Q + : = { v | b v < 0 } D ¯ . And add all of Q + (and other necessary vertices for feasibility) to D. Similarly

v Q + b v f Q + out f Q in = f Q out = v Q | b v | v Q + b v v Q | b v | 0 .

Taking these two steps, we turned D to S and only made b d D bigger. So S is optimal.

User profile picture
2025-01-24 12:47
Comments