Variation of Jensen's inequality

43 Views Asked by At

I have a more general question: can we always say that, given a convex function $g$ and some $h$ s.t. $\sum_{x\in\mathcal{X}}h(x)=1$, $A\subset\mathcal{X}$ $$\sum_{x\in A} g(x)h(x)\geq g\left(\sum_{x\in A} xh(x)\right)$$ even if we do not sum on all the $x$? Why does it hold? I have seen it done a few times (practical question), like here: $$\sum_{x\in A} \left(\frac{P(x)}{Q(x)}\right)^c\cdot\frac{1}{P(A)} P(x) \geq \sum_{x\in A} \left(\frac{P(x)}{Q(x)}\cdot\frac{1}{P(A)} P(x)\right)^c$$ where $c\geq 0$ and $P,Q$ are probability measures on $\mathcal{X}$. Is it correct? Can you give me some hints on the proof? Sorry if it is silly but I just can't see it!