On the convergence of partial differential equations of parabolic type with rapidly oscillating coefficients to stochastic partial differential equations (Q1124209): Difference between revisions

Consider partial differential equations of the form \[ D_ tu^{\epsilon}(t,x)=\sum^{d}_{i,j=1}a_{i,j}(t/\epsilon^ 2,x)D_{x_ i,x_ j}u^{\epsilon}(t,x)+ \] \[ +\sum^{d}_{i=1}b_ i(t/\epsilon^ 2,x)D_{x_ i}u^{\epsilon}(t,x)+\epsilon^{- 1}\sum^{d}_{i=1}c_ i(t/\epsilon^ 2,x)D_{x_ i}u^{\epsilon}(t,x) \] on \({\mathbb{R}}^ d\) with random coefficients, \(a_{i,j}\) and \(b_ i\) being stationary in time and \(c_ i\) in both variables. Under assumptions entailing the existence and uniqueness of solutions, \(u^{\epsilon}\) can be regarded as continuous processes with values in the space of Schwartz-distributions \({\mathcal S}'.\) Their laws \(Q^{\epsilon}\) on C([0,T];\({\mathcal S}')\) are proven to be tight and converge (as \(\epsilon\) \(\to 0)\) to the solution of a martingale problem. The martingale problem formally corresponds to stochastic partial differential equations with coefficients obtained by averaging the \(a_{i,j}\) and \(b_ i\).