Differentiation formulas for probability functions: The transformation method (Q1363426): Difference between revisions

The paper deals with the transformation technique for the evaluation of derivatives of probability functions, such as \(P(x)= P(y(a(\omega),x)\geq 0)\), with respect to components of the multidimensional parameter \(x\). The main step is on integral transformation which leads to a region of integration independent of the considered component of \(x\); this allows then to exploit the standard approaches for differentiation of integrals dependent on parameters. An extension is the stochastic completion method which applies also under relaxed assumptions concerning the existence and the properties of the density of the random vector \(a(\omega)\). For the numerical computations, one can estimate the derivatives by sampling techniques applied to their mean value representation. Moreover, in case of transformed normal random vectors \(a(\omega)\), an asymptotic expansion method can be exploited. The presented applications of the method concern the probabilistic (chance constrained) programming and structural reliability problems.