HDS

Exercise 5.12: Gaussian Contraction Inequality

chapter 5

Let Xϕ(θ)=ϕ(θ),w, where wN(0,Id) is a shared standard normal random vector, so that G(ϕ(T))=E[supθXϕ(θ)]. By linearity of expectation E[Xϕ(θ)]=0, and the associated squared metric E[(Xϕ(θ)Xϕ(θ))2]=ϕ(θ)ϕ(θ)22 is upper bounded by θθ22 due to the 1-Lipschitz assumption. Hence for any finite GT (1)E[supθGϕ(θ),w]E[supθGθ,w], by the Sudakov-Fernique inequality (Theorem 5.27). Applying the monotone convergence theorem, first on the r.h.s., then on the l.h.s., yields the result.

Published on 1 March 2021.