No label defined (Q4940289)

The author considers asymptotic properties of distribution of functionals on the space with a smooth measure. He gives an approach to estimate the upper and the lower bounds of the large derivations rate for the smooth measure and the distribution of its vector logarithmic derivative based on the differential structure of the corresponding probability space, namely the changing of variables in the Ramer-type formula for the smooth measure. Using this formula the author gives an explicit expression for exponential moments of the logarithmic derivative of the measure and obtains exact estimations for large deviations. The author applies his results to the general large deviations theory. For example, the following assertion is proved: Let the family of random vectors \(\{\xi_{\varepsilon}\}\) in a separable Banach space \(X\) satisfy the following conditions: 1) there exists a lower semicontinuous function \(\varphi: X^{\ast}\to R\) such that \(\forall x^{\ast}\in X^{\ast}\), \[ \varepsilon^2\ln E\exp[\langle \xi_{\varepsilon}, x^{\ast}\rangle]\to \varphi(x^{\ast}),\;\varepsilon \to 0; \] 2) the family of distributions of \(\{\xi_{\varepsilon}\}\) is exponentially tight, i.e. for every \(C>0\) there exists a compact \(K_{C}\) such that \[ \limsup_{\varepsilon\to 0}\varepsilon^2\ln P(\xi_{\varepsilon}\notin K_{C})\leq -C. \] Then the function \(\varphi^{\ast}:X\ni x\to \sup_{x^{\ast}\in X^{\ast}}[\langle x,x^{\ast} \rangle >-\varphi(x^{\ast})]\) gives the upper bounds for the large deviations of the family of distributions of \(X_{\varepsilon}\) with the rate function \(\varepsilon^2\), i.e. for \(A\in {\mathcal B}(X)\) \[ \lim\limits_{\varepsilon\to 0}\varepsilon^2\ln P\{\xi_{\varepsilon}\in A\}\leq -\inf\limits_{x\in \text{cl}(A)}\varphi^{\ast}(x). \]