On diffusion approximation with discountinuous coefficients. (Q2574527)

The main results of the paper concern the problem of diffusion approximation, when the limiting process satisfies a stochastic differential equation whose coefficients may be discontinuous. More specifically, the limiting diffusion process satisfies the equation \[ x_t= x_0 +\int _0^t b(s,x_s)\,ds +\int _0^t \sqrt {a(s,x_s)}\,dw_s, \] the respective ``coefficients of semimartingales \((x^n)\)'' converge to \(b\) and \(a\) on a large set, and certain growth and nondegeneracy conditions are satisfied. No assumption about any control \(b\) and \(a\) as \(| x| \to \infty \) has been made, but instead it is supposed that the distributions \(Q^n\) of the semimartingales \((x^n)\) in the Skorokhod space \(\mathcal D([0, \infty ))\) converge weakly to a limiting distribution. Both coefficients \(b(t,\cdot )\) and \(a(t,\cdot ) \) may be discontinuous on a set \(G_t\) of Lebesgue measure zero. An example is given where the diffusion approximation of a queueing model yields a diffusion process with discontinuous drift and diffusion coefficients.