10
$\begingroup$

This could be asked in more generality, but let me stick to a concrete case.

Usually one considers a fixed domain $E \subset \mathbb{C}$ and attaches to it the equilibrium probability measure $\nu_E$, the one that minimizes the energy integral $$ I_E(\nu) := \int_E \int_E -\log{|z-w|} \, d\nu(z) \, d\nu(w). $$ I want to look at the opposite problem. Fix an infinite measure $\mu$ on $\mathbb{C}$, say the Lebesgue measure $dx\, dy$, and look for the measurable set $E \subset \mathbb{C}$ of unit measure $\mu(E) = 1$ for which the energy $I_E(\mu)$ is a maximum.

Is there a literature available on this type of question? In particular, what is the answer when $\mu$ is the Lebesgue measure? I am fine with assuming $E$ is a 'nice' domain of unit area rather than a general measurable set.

The same question could be posed in Euclidean space of any dimension, or a more general Riemannian manifold. For dimension one and the Lebesgue measure on $\mathbb{R}$ the maximum is clearly attained for a unit interval, with energy $$ \int_0^1\int_0^1 -\log{|x-y|} \, dx \, dy = 3/2. $$

$\endgroup$
  • 2
    $\begingroup$ It ought to be a circle (or generally a Euclidean sphere) of unit area (volume), right? $\endgroup$ – Noam D. Elkies May 21 '15 at 14:31
  • 1
    $\begingroup$ Doesn't Steiner symmetrization prove this? $\endgroup$ – Boris Bukh May 21 '15 at 14:38
  • 1
    $\begingroup$ @NoamD.Elkies: That is what I thought (the ball of unit volume), but I didn't see how to prove this. $\endgroup$ – Vesselin Dimitrov May 21 '15 at 16:31
  • $\begingroup$ @BorisBukh: Thanks for the key word, I'll think about this. $\endgroup$ – Vesselin Dimitrov May 21 '15 at 17:28
4
$\begingroup$

That a circular disk maximizes the energy for the Lebesgue measure follows immediately from Riesz's inequality: $I(f,g,h)\le I(f^*,g^*,h^*)$,where $I(f,g,h) = \langle f, g*h\rangle$, and $f^*$ is the monotonically decreasing radial function whose superlevel sets have the same measures as those of $f$. In your case, take $f=g=1_E$ and $h=h^*=-log(|z|)$.

$\endgroup$
  • 2
    $\begingroup$ This is probably only a small technical point, but the quoted result doesn't immediately apply because our $h$ is not non-negative. (I don't think this can be a real problem because it's silly to enter regions where $h<-100$, say, if we want to maximize, so for all practical purposes, $h$ is bounded below.) $\endgroup$ – Christian Remling May 22 '15 at 1:28
  • 2
    $\begingroup$ @ChristianRemling, that's a good point. One way to work around it is, for a given $E$, to cut off $h$ at the diameter $d$ of $E$, $h(r>d)=h(d)$, and add a constant for positivity. You still get that the modified energy for $E$, which is the same as the real energy for $E$ plus a constant, is less than the modified energy for the disk, which is the same as the real energy for the disk plus the same constant. $\endgroup$ – Yoav Kallus May 22 '15 at 1:55

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.