Let me clarify the setting I'm thinking.

For any totally bounded metric space $(Y,d_Y)$ and $\varepsilon>0$, the $\textit{metric entropy}$ $N_M(\varepsilon,Y)$ is the smallest number of closed ball $B(y_i,\varepsilon):=\{y\in Y:d_Y(y,y_i)\leq\varepsilon\}$, $i=1,\dots,n$, covering whole $Y$. In other words, $N_M(\varepsilon,Y)$ is the cardinality of the smallest $\varepsilon$-net of $Y$.

Now, consider a compact metric space $(X,d_X)$ and a finite borel measure $\mu_X$ on it. Consider the unit closed ball $B(0,1)\subset L^2(\mu_X)$. My questions are the following:

Q1: What is the growth rate of $N_M(\varepsilon, B(0,1))$ as $\varepsilon$ goes to $0$?

Q2: Do we have $\int_0^1 \log N_M(\varepsilon, B(0,1))\,d\varepsilon<\infty$ in general? if not, then under what kinds of condition for $(X,d_X,\mu_X)$ can we guarantee the finiteness?

I'm not at all familiar with these topics about the metric entropy of Hilbert space. I tried some googling but it was not successful. But it feels like natural question. So, are there any references answer my above questions? Or, are there any textbooks/papers dealing with some related contents?


1 Answer 1


A ball $B$ of radius $r$ in the Hilbert space can not be covered by finitely many balls of radius less than $r$. Indeed, if it can, than iterating this covering we cover $B$ by finitely many balls of arbitrarily small radius. This would imply that $B$ is precompact (totally bounded), but this implies that the space is finite-dimensional.

  • $\begingroup$ note that the 𝐿2(𝜇) will be finite dimensional if the measure is atomic, and this is an "iff" with the appropriate definition of "atomic" $\endgroup$
    – am70
    Jun 10, 2022 at 9:57

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.