I know the expectation preserves the concavity (or convexity), but I was wondering is it still true that the expectation of log-concave function still log-concave; to be more precise,

Let $g(x,Y)$ be log-concave function in $x$ where $Y$ be discrete-time random variable with density $f_Y$. Is it true that $$ E[g(x,Y)] $$be still log-concave in $x$?

I noticed there is one result called Prekopa theorem, which states that if $g(x,y): \mathbb{R}^{n+m} \to \mathbb{R}$ be (jointly) log-concave, then $$ h(x) = \int g(x,y) dy $$is log-concave. But I'm not sure how to apply properly on the expectation case, since I have to deal with the log-concavity of integrand function $g(x,y) f_Y(y)$ first; i.e., $$ E[g(x,Y)] = \int g(x,y) f_Y(y)dy $$

Any suggestion is appreciated. Thanks


Not necessarily. Consider $Y$ such that $Pr[Y=0]=Pr[Y=1]=1/2$. Define $g(x,Y)=e^{Yx}$. Then $g(x,Y)$ is log concave in $x$ because $\log g(x,Y) = Yx$ is linear. But: $$ E[g(x,Y)] = \frac{1 + e^x}{2} $$ and $\log E[g(x,Y)] = \log(1/2) + \log(1 + e^x)$, which is no longer concave.

  • 1
    $\begingroup$ Thanks @Michael !! very helpful example. $\endgroup$ – Fianra Jun 28 '15 at 1:15

It is true under stricter hypothesis.

Proposition. Let $g(x,y)$ be jointly log-concave and $Y$ be a log-concave random variable with density $f(y)$. Then $\mathbb{E}[g(x,Y)]$ is log-concave.

Proof. The function $g(x,y)f(y)$ is log-concave because is logarithm $\log g(x+y) + \log f(y)$ is the sum of two concave functions. Therefore, because log-concavity is preserved under marginalization, $$ \mathbb{E}[g(x,Y)] = \int g(x,y)f(y)\,dy $$ is log-concave.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.