A variational formula for risk-sensitive control.
Wednesday, May 9, 2018 - 9:00am - 9:50am
This talk will describe a variational formula for risk-sensitive reward. This extends the Donsker-Varadhan characterization of principal eigenvalue of a non-negative matrix in discrete case and an elliptic operator in the continuous case. One application to linear and dynamic programming approaches for risk-sensitive control of finite Markov chains without the irreducibility assumption will be described. This is joint work with V. Anantharam (UC Berkeley), A. Arapostathis (UT Austin) and K. Suresh Kumar (IIT Bombay).