$u$ satisfies "mean value property locally " on $\Omega$ if for every $x\in \Omega \exists \delta=\delta (x)>0 $ such that

$u(x) \le \frac {1}{|\mu(B(x,r)|}\int_{\partial B(x,r)} u(y) dS_y$

for all $r\le \delta(x)$

Does this imply that if $u\in C^2(\Omega)$ and satisfies mean value property locally in $\Omega$ then $u$ is subharmonic ?? Any hints will be nice.

  • $\begingroup$ Yes, I am sure this is covered in standard texts on potential theory, such as Helms ("Potential Theory"). In most of potential theory, we don't require $C^2$, just upper or lower semi-continuity. $\endgroup$
    – Old John
    Commented Jul 17, 2012 at 12:00
  • $\begingroup$ @Old John : What is the basic argument to extend local property to the global property ? $\endgroup$
    – Theorem
    Commented Jul 17, 2012 at 12:03
  • $\begingroup$ Hint: use the 2nd order Taylor expansion of u at x. The first derivatives cancel out in integration, the second derivatives contribute the Laplacian, and nothing else matters. $\endgroup$
    – user31373
    Commented Jul 17, 2012 at 12:06
  • $\begingroup$ From memory, I think it is a connectedness argument, but I think you might have to adjust the statement of the problem above, as I think it would fail if $\Omega$ were not connected. $\endgroup$
    – Old John
    Commented Jul 17, 2012 at 12:06
  • $\begingroup$ @LeonidKovalev : I will try and will try to get back with an answer . $\endgroup$
    – Theorem
    Commented Jul 17, 2012 at 12:09

2 Answers 2


We first claim that if a $C^0(\bar{\Omega})$ function $u$ satisfies the following local mean value inequality

$$u(y)\leq\frac{1}{n\omega_nR^{n-1}}\int_{\partial B_R(y)}u ds \text{ for all } R\leq \delta$$

then maximal principle applies, i.e. $u$ attains its maximum at $\partial \Omega$.

Assume our claim now. Let $\tilde{u}$ be the harmonic function defined by the boundary value $u|_{\partial B}$ on the boundary of an arbitrary ball $B\subset\subset \Omega$. Then $\tilde{u}$ satisfies mean value equality in $B$ since it is harmonic in $B$. Then $u-\tilde{u}$ satisfies the above mean value inequality. Then by our claim (maximal principle), we have $\sup_{x\in B}(u-\tilde{u})=\sup_{x\in\partial B}(u-\tilde{u})$. But $u-\tilde{u}=0$ on $\partial B$. Hence we have $u-\tilde{u}\leq 0$. $u\leq \tilde{u}$ in $B$.

Now we show that $u$ is subharmonic, given an harmonic function $h$ in $B\subset \subset \Omega$, such that $u\leq h$ on $\partial B$. We need to show that $u\leq h$ in B. Since we know $u\leq\tilde{u}$ in $B$. It suffices to show that $\tilde{u}\leq h$ in $B$. To see this one uses the Possion kernel $$\tilde{u}(x)=\int_{\partial B}K(x,y)u(y)ds_y$$ $$h(x)=\int_{\partial B}K(x,y)h(y)ds_y$$
Since $u(y)\leq h(y)$ on $\partial B$ and $K(x,y)\geq 0$. We have $\tilde{u}(x)\leq h(x)$.

Now we show our claim that the local mean value inequality $u(y)\leq\frac{1}{n\omega_nR^{n-1}}\int_{\partial B_R(y)}u ds $

implies maximal principle.

We use the standard topological argument. Define a set $S:=\{q\in\bar{\Omega}| u(q)=\sup_{x\in\bar{\Omega}}u(x)\}$. If $u$ attains maximum at a point $p\in\Omega$ then $S$ is nonempty since $p\in S$. Note that $S$ is closed since every sequence in $S$ converges to a point in $S$ since $u$ is continuous. Next if $p\in S\cap\Omega$, then $u(p)\leq\frac{1}{n\omega_nR^{n-1}}\int_{\partial B_R(p)}u ds $ and $u(p)\geq u(x)$ implies $u(x)=u(p)$ for all $x\in B_R(p)$ since $u$ is continuous. Hence $B_R(p)\in S$ Hence $S$ is open. Now $S$ is open and closed and nonempty, hence we have $S=\bar{\Omega}$. This proves our claim and finished our proof.

  • $\begingroup$ Thanks for your editing, Byron. $\endgroup$ Commented Feb 2, 2017 at 19:43
  • $\begingroup$ Shouldn't we be proving that $- \Delta u \leq 0$? Why do you show that $u \leq h$ instead? $\endgroup$ Commented Dec 15, 2020 at 18:33

No, a function satisfying a mean-value inequality is not necessarily $C^2$ (think of the norm function $\mathbb R^d\supseteq\Omega\ni\vec x\mapsto|\vec x|\in\mathbb R$) but if it satisfies a mean-value equality it is harmonic and therefore $C^\infty$ (in fact, analytic).

Yes, a function's being (sub/super)mean-valued is (sub/super)harmonic, but you need to interpret the Laplacian in a generalized sense. For example, in my example above the Laplacian is a dirac-delta which is a positive distribution (testing against positive test functions yields a positive output).


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .