Convex Optimization · Problem 2.15

Convex functions — Boyd & Vandenberghe

Mathematics — Optimization · sets-of-probability-distributions

Problem

Some sets of probability distributions. Let xx be a real-valued random variable with prob(x=ai)=pi\mathbf{prob}(x = a_i) = p_i, i=1,…,ni = 1, \ldots, n, where a1<a2<⋯<ana_1 < a_2 < \cdots < a_n. Of course p∈Rnp \in \mathbf{R}^n lies in the standard probability simplex P={p∣1Tp=1, p⪰0}P = \{p \mid \mathbf{1}^T p = 1,\, p \succeq 0\}. Which of the following conditions are convex in pp? (That is, for which of the following conditions is the set of p∈Pp \in P that satisfy the condition convex?) (a) α≤Ef(x)≤β\alpha \leq \mathbf{E} f(x) \leq \beta, where Ef(x)\mathbf{E} f(x) is the expected value of f(x)f(x), i.e., Ef(x)=∑i=1npif(ai)\mathbf{E} f(x) = \sum_{i=1}^n p_i f(a_i). (The function f:R→Rf : \mathbf{R} \to \mathbf{R} is given.) (b) prob(x>α)≤β\mathbf{prob}(x > \alpha) \leq \beta. (c) E∣x3∣≤αE∣x∣\mathbf{E}|x^3| \leq \alpha \mathbf{E}|x|. (d) Ex2≤α\mathbf{E} x^2 \leq \alpha. (e) Ex2≥α\mathbf{E} x^2 \geq \alpha. (f) var(x)≤α\mathbf{var}(x) \leq \alpha, where var(x)=E(x−Ex)2\mathbf{var}(x) = \mathbf{E}(x - \mathbf{E} x)^2 is the variance of xx. (g) var(x)≥α\mathbf{var}(x) \geq \alpha. (h) quartile(x)≥α\mathbf{quartile}(x) \geq \alpha, where quartile(x)=inf⁡{β∣prob(x≤β)≥0.25}\mathbf{quartile}(x) = \inf\{\beta \mid \mathbf{prob}(x \leq \beta) \geq 0.25\}. (i) quartile(x)≤α\mathbf{quartile}(x) \leq \alpha.

Related theory

Browse the problem library