M-stationarity Concept for a Class of Stochastic MPCC Problems

Speaker’s Name: Arnab Sur

Speaker’s Institution: Industrial Engineering and Operations Research (IEOR), Indian Institute of Technology Bombay

In this talk we are going to discuss the importance of M-stationary conditions for a special class of one-stage stochastic mathematical programming problem with complementarity constraints (SMPCC, for short). M-stationarity concept is well known for deterministic MPCC problems. Now using the results of deterministic MPCC problems we can easily derive the M-stationarity for SMPCC problems under some well known constraint qualifications. It is well observed that under MPCC-linear independence constraint qualification we obtain strong stationarity conditions at a local minimum, which is a stronger notion than M-stationarity. Same result cab be derived for SMPCC problems under SMPCC-LICQ. Then the question that will arise is: What is the importance to study M-stationarity under the assumption of SMPCC-LICQ. To answer this question we have to discuss sample average approximation (SAA) method, which is a common technique to solve stochastic optimization problems. Here one has to discretize the underlying probability space and then using the strong Law of Large Numbers one has to approximate the expectation functionals. Now the main result of this discussion as follows: If we consider a sequence of M-type Fritz John points of the SAA problems then any accumulation point of this sequence will be an M-stationarity point under SMPCC-LICQ. But this kind of result, in general, does not hold for strong stationarity conditions.

Seminar Convenors: Matthew Tam

AGR Contacts: Andrew Danson, David Allingham