Exercise (Rudin, R&CA, no. 3.25). Suppose $\mu$ is a positive measure on the space $X$ and let $f \colon X \to (0,+\infty)$ be such that $\int_X f \, d\mu=1$.

Then for every $E \subset X$ with $0<\mu(E)<\infty$ we have $$ \int_E \log f \, d\mu \le \mu(E) \log \frac{1}{\mu(E)}. $$

This is not homework, it is self-studying. I think I should use Jensen's inequality, but I cannot get it.

I thought considering the positive probability measure $\nu$ given by $\nu:=f\mu$ but I do not see which are the right functions to play Jensen's inequality with. Since $\log$ is concave, I would have $$ \int_E f \log f \, d\mu \le \log \int_E f^2 \, d\mu $$ but this is not helpful. Than I can try $\log \frac{1}{x}$ which is convex and I would have $$ \int_E f \log\left( \frac{1}{f}\right) \, d\mu \ge \log \int_E 1\, d\mu = \log \mu(E) $$ which looks nicer since it gives $$ \int_E f \log f \, d\mu \le \log\left(\frac{1}{\mu(E)}\right) $$ but now I do not know how to handle the LHS.

Any hint, please? Thanks in advance.


Write the inequality as

$$\frac{1}{\mu(E)} \int_E \log f\,d\mu \leqslant \log \frac{1}{\mu(E)}.$$

Jensen's inequality gives you

$$\exp \left(\frac{1}{\mu(E)}\int_E \log f\,d\mu\right) \leqslant \frac{1}{\mu(E)}\int_E f\,d\mu.$$

The remaining part should be clear.

  • $\begingroup$ Now we have to take $\log$ of both sides and from $\int_E f d\mu \le 1$ we conclude, right? Thank you very much for your help. $\endgroup$ – Romeo Jul 23 '14 at 15:27
  • $\begingroup$ Yes, exactly so. $\endgroup$ – Daniel Fischer Jul 23 '14 at 15:28

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.