A convexity property of expectations under exponential weights

Abstract

Take a random variable X with some finite exponential moments. Define an exponentially weighted expectation by E^t(f) = E(e^{tX}f)/E(e^{tX}) for admissible values of the parameter t. Denote the weighted expectation of X itself by r(t) = E^t(X), with inverse function t(r). We prove that for a convex function f the expectation E^{t(r)}(f) is a convex function of the parameter r. Along the way we develop correlation inequalities for convex functions. Motivation for this result comes from equilibrium investigations of some stochastic interacting systems with stationary product distributions. In particular, convexity of the hydrodynamic flux function follows in some cases.

After completion of this manuscript we learned that our main results can be obtained as a special case of some propositions in Karlin: Total Positivity, Vol.1

References in corpus (1)

• Growth fluctuations in a class of deposition models

Cited by in corpus (7)

• Random walk of second class particles in product shock measures
• Microscopic concavity and fluctuation bounds in a class of deposition processes
• Supercriticality conditions for asymmetric zero-range process with sitewise disorder
• Fluctuation bounds in the exponential bricklayers process
• Q-zero range has random walking shocks
• Hydrodynamics in a condensation regime: the disordered asymmetric zero-range process
• Unifying particle-based and continuum models of hillslope evolution with a probabilistic scaling technique