Second-order differentiability of probability functions (Q2361140)

The authors study second-order differentiability of the probability function \(\varphi :\mathbb{R}^{n}\rightarrow \left[ 0,1\right] \) defined by \(\varphi \left( x\right) :=\mathbb{P[}g\left( x,\xi\right) \leq 0]\), for all \(x\in \mathbb{R}^{n}\) where \(g:\mathbb{R} ^{n}\times \mathbb{R}^{m}\rightarrow \mathbb{R}^{k}\) is a continuously differentiable function, convex with respect to the second argument, and \( \xi \in \mathbb{R}^{m}\mathbb{\;}\)is a multi-variate non-degenerate Gaussian (or Student) random vector in \(\mathbb{R}^{m}\). \noindent They prove that \(\varphi \) is twice continuously differentiable around a trial point \(\bar{x}\) under growth conditions on \(\nabla ^{2}g\left( x\right) \) the Hessian of \(g\) at \(x.\) \noindent Also, they derive a general expression of \(\nabla ^{2}\varphi \left( x\right) \), as an integral with respect to the uniform distribution on the sphere. \noindent Finally, the authors illustrate their second-order differentiability result on some special examples: for chi-squares random vectors and for lognormal vectors.