Dynkin's formula under the \(G\)-expectation (Q2267636): Difference between revisions

Motivated by modelisation of dynamic risk measures in finance, Shige Peng recently introduced the notion of \(G\)-normally distributed random variables -- a normal distribution under the nonlinear, so-called \(G\)-expectation \(\widehat{E}[.]\), and the related notion of a \(G\)-Brownian motion \(B=(B_t)_{t\geq 0}\). In a sequel of works he developed a stochastic analysis related with these notions. A remarkable property of the quadratic variation process \(\langle B\rangle\) of \(B\) is, that it's not a deterministic function anymore as in the case of a classical Broznian motion. The author of the present paper studies the Dynkin formula in the framework of this \(G\)-stochastic analysis: Given the unique solution of the one-dimensional diffusion equation \(dX_t=a(X_t)dt+b(X_t)d\langle B\rangle_t+c(X_t)dB_t\), \(X_0=x\), he proves that the Dynkin operator \(A_G\) defined by the formula \(\widehat{E}[f(X_t)]=f(x)+\widehat{E}[\int_0^tA_Gf(X_s)\,ds]\), \(t\geq 0,\) is of the form \(A_Gf(x)=a(x)f'(x)+ \sup_{\underline{\sigma}\leq\gamma\leq\overline{\sigma}} (\gamma^2(b(x)f'(x)+\frac{1}{2}c(x)^2f''(x))).\)