Bound on difference of expectations for two close distributions

324 Views Asked by At

This problem is similar to Upper bound on difference of expected values between two close distributions, and we use the same notations. Let $p$ and $q$ be two distributions over the same finite set $X$ and the total variation distance $d_{TV}(p,q)$ is bounded by $\epsilon$.

Q: For some function $f(x): X \to \mathbb{R}$, is the difference of expectations $\left|\sum_x f(x)(p(x)-q(x))\right|$ can be bounded by the expectation of $p$ (or $q$), i.e., $\left|\sum_x f(x)p(x)\right|$?

In the original post, @Adam gives a bound for the difference of expectations with respect to the $\max |f(x)|$, I wonder whether there exists a tighter one by using the expectation of $p$ (or $q$).