\section*{Which distance means satisfy the volume bound?} The geometric mean in Erd\H{o}s's question belongs to a one-parameter family. Given the finite marks and positive weights of Theorem~\ref{thm:volume}, define \[ M_p(x)= \begin{cases} \bigl(\sum_j w_j |x-a_j|^p\bigr)^{1/p}, & p\ne0,\\ \prod_j |x-a_j|^{w_j}, & p=0 . \end{cases} \] For $p<0$ we set $M_p=0$ at a mark, which is the limit of $M_p(x)$ as $x$ approaches the mark. For fixed $x$, the weighted power mean $M_p(x)$ is nondecreasing in $p$. We show that $2-d$, the exponent of the Newtonian kernel, is the smallest exponent for which the volume bound of Theorem~\ref{thm:volume} holds for every configuration. In three dimensions this exponent is $-1$. The geometric mean ($p=0$), which is the mean in Erd\H{o}s's question, satisfies the volume bound. For $p<-1$, two equally weighted marks that are sufficiently close give, at $R=1$, a volume greater than that of the unit ball. \begin{theorem}\label{thm:power} Fix $d\ge3$ and $p\in\R$. The inequality \[ |\{M_p\le R\}|\le\omega_dR^d \] holds for all finite configurations, positive weights of sum one and $R>0$ if and only if $p\ge 2-d$. In this range, for each $R>0$, equality holds exactly when all the marks coincide. For every $p<2-d$, two equally weighted marks suffice to violate the inequality. \end{theorem} \subsection*{Consequences of earlier inequalities} For $-d
\lambda\}|\le\omega_d\lambda^{-d/(d-s)}$ for $\lambda>0$, where $V$ is the potential of a probability density with the kernel $|x|^{s-d}$, without a normalizing constant. To apply the estimate to finitely many marks, replace the point mass at each mark $a_j$ by $w_j$ times a smooth probability density supported in a small ball about $a_j$, and let the radii of the balls tend to zero. The sum of these densities is a probability density. Away from the marks, the resulting potentials converge to the potential of the marks, $V_{\rm at}(x)=\sum_j w_j|x-a_j|^p$. By Fatou's lemma, the same bound holds for the strict superlevel sets $\{V_{\rm at}>\lambda\}$. As positive thresholds increase to $\lambda$, these strict superlevel sets decrease to $\{V_{\rm at}\ge\lambda\}$, and their volumes are finite, so the bound holds for $\{V_{\rm at}\ge\lambda\}$ as well. For $s=2$ this is the volume bound for $M_{2-d}$. Since $M_p$ is nondecreasing in $p$, the bound follows for every $p\ge2-d$. For $2-d
\lambda\}$.
The inequality of part (iii) includes $p=2-d$, so the ball with the same
capacity as $K$ has radius at most $R$. At this exponent, however,
equality in part (iii) at a mark says only that the equilibrium potential
of $K$ at the mark equals the equilibrium energy of $K$.
Each mark is an interior point of $K$, and the equilibrium potential
takes this value at every interior point of $K$. So this equality
holds at every mark even when the marks are distinct, and
it cannot by itself force the marks to coincide.
To show that the marks coincide we use instead the equality case of the
isocapacitary inequality~\cite[\S1.1]{capacity}.
The level set $\{U=\lambda\}$ has volume zero, because $U$ is analytic
away from the marks. If $|K|=\omega_dR^d$, then
$|\Omega|=\omega_dR^d$ as well. Let $B_R$ be a ball of radius $R$. The
isocapacitary inequality and the monotonicity of Newtonian capacity give
\[
\operatorname{cap}(B_R)\le\operatorname{cap}(\Omega)
\le\operatorname{cap}(K)\le\operatorname{cap}(B_R).
\]
Hence, by the equality case, $\Omega$ agrees with a ball of radius $R$ up
to a set of volume zero.
This argument applies even when $\lambda$ is a critical value of $U$.
To see why this determines the marks, suppose more generally that
$\{U>\lambda\}$ agrees with an open ball $B(z,b)$ up to a set of volume
zero. Then every mark lies in the open ball. Indeed, a neighborhood of
each mark lies in $\{U>\lambda\}$, whereas every neighborhood of a point
on or outside the sphere $|x-z|=b$ contains a set of positive volume
outside the ball. Next, $U=\lambda$ at each point of the sphere. Indeed,
if $U>\lambda$ at such a point, then by continuity $U>\lambda$ on a
neighborhood of it, which contains a set of positive volume outside the
ball; and if $U<\lambda$ there, then $U<\lambda$ on a neighborhood, which
contains a set of positive volume inside the ball. Since all the marks
are inside the ball, $U$ is harmonic outside the ball, equals $\lambda$
on the sphere, and tends to zero at infinity. By uniqueness for the
exterior Dirichlet problem,
\begin{equation}\label{eq:exterior-ball}
U(x)=\lambda\,b^{d-2}|x-z|^{2-d}\qquad (|x-z|>b).
\end{equation}
Both sides are harmonic on the complement of $z$ and the marks, which is
connected, so by analytic continuation the identity holds on that whole
set. If a mark lay at a point other than $z$, the left side would tend to
infinity at that mark, because its weight is positive, while the right
side would stay bounded. Thus all marks
coincide at $z$. Taking $b=R$ completes the deduction.
Now let $p<2-d$ and put $q=-p>d-2$. Consider $n$ equally weighted marks
$a_1,\ldots,a_n$ in the closed unit ball. For each placement of the
marks, take the minimum over all points $x$ of that ball of
$n^{-1}\sum_j|x-a_j|^{-q}$, and let $P_n$ be the supremum of this minimum
over all placements. Erd\'elyi and Saff~\cite[Theorem~3.2]{polarization}
show that $P_n\to\infty$ when $q\ge d$, and that for $d-21 .
\]
To see that this quotient exceeds one, note that it equals one at
$q=d-2$ and that its derivative in $q$ is positive on $d-2
(d-q)/2$. Thus in both ranges there are a number $n$
and a placement of $n$ equally weighted marks for which the minimum above
is some number $P>1$. For this placement, $M_{-q}\le P^{-1/q}$ throughout
the unit ball. Put $R=P^{-1/q}<1$. Then the set $\{M_{-q}\le R\}$
contains the unit ball, so its volume is at least $\omega_d$, which is
greater than $\omega_d R^d$.
\begin{proof}[Direct flux proof and two-mark counterexamples]
Use the Newtonian potential $U$, the superlevel volume $W(\lambda)$, and
the equal-volume radius $r(\lambda)$ from the flux proof of
Theorem~\ref{thm:volume}. That argument proves that
$r(\lambda)^{2-d}-\lambda$ is nondecreasing and has limit zero as $\lambda$ tends to zero.
At the endpoint $p=2-d$ we have $M_{2-d}=U^{1/(2-d)}$, so
$\{M_{2-d}\le R\}=\{U\ge R^{2-d}\}$. To prove the strict inequality at
this exponent we examine when equality holds in the isoperimetric inequality.
Suppose that $\lambda>0$ is a regular value of $U$ and that equality
holds in the isoperimetric inequality for $\{U>\lambda\}$. Then
$\{U>\lambda\}$ is a ball up to a set of volume zero. By the
identity~\eqref{eq:exterior-ball} and the argument that follows it,
every mark then lies at the center of that ball.
Consequently, if the marks do not all coincide, the isoperimetric
inequality is strict at every regular value. It follows that $(r^{2-d})'>1$ almost
everywhere on $(0,\infty)$. By local absolute continuity,
$r(\lambda)^{2-d}-\lambda$ is strictly increasing there. Its limit at zero is
zero, so $r(\lambda)^{2-d}>\lambda$ for every $\lambda>0$. Set $\lambda=R^{2-d}$ to obtain
$|\{M_{2-d}\le R\}|<\omega_d R^d$. Since $\{M_p\le R\}$ is contained
in $\{M_{2-d}\le R\}$ for $p\ge2-d$, the same strict bound holds for
every $p\ge2-d$. If all marks coincide at $a$,
then $M_p(x)=|x-a|$ and the set $\{M_p\le R\}$ is the ball of radius $R$ about $a$.
For the converse, fix $p<2-d$, take $R=1$, and put equal weights
at $\pm t e_1$ with $0