ProbXiv
Explorersign in
Problem archiveProblem record

Statement

For XX and YY i.i.d.\ log-concave random variables, what is

max⁡λ∈[0,1]h(λX+1−λY)?\max_{\lambda \in [0,1]} h(\sqrt{\lambda} X + \sqrt{1 - \lambda} Y)?

We expect that the maximum is achieved at λ=1/2\lambda = 1/2.

It is known that this is not true without log-cavity. The counterexample is given by X,YX,Y with density x2ϕ(x)x^2 \phi(x) where ϕ(⋅)\phi(\cdot) is the standard Gaussian density, which is a bi-modal density. Another counterexample exists when XX and YY are uniform on [a,b]∪[−b,−a][a,b] \cup [-b, -a]. For certain choices of aa and bb, the maximum is achieved at λ\lambda arbitrarily close to 00.

We finally note that the Shepp-Olkin conjecture is relevant here. For parameters p1,p2,…,pn≜p‾p_1, p_2, \ldots, p_n \triangleq \underline{p}, with random variable

X(p‾)=Y1+Y2+…+Yn, where Yi are independent Bern(pi).X(\underline{p}) = Y_1 + Y_2 + \ldots + Y_n, \quad \text{ where } \quad Y_i \text{ are independent Bern}(p_i).

Then H(X(p‾))H(X(\underline{p})) is concave in p‾\underline{p}. The open question is whether H(X(p‾))H(X(\underline{p})) iincreases in p‾\underline{p} for p‾∈[0,1/2)n\underline{p} \in [0, 1/2)^n.

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.