1
$\begingroup$

How can I prove the following Liouville theorem without using the mean value property?

If $u$ is harmonic on $\mathbb{R}^n$ and $\int_{\mathbb{R}^n}|\nabla u|^2 dx \leq C$ for some $C > 0$, then $u$ is constant.

The proof that I know indeed uses the mean value property for harmonic functions.


From the comments: is it rigorous to do it like this: $-\Delta u = 0 \implies \int_{\mathbb R^n} |\nabla u|^2 = 0$ (integrating by parts, hence $u$ is constant? It seems to easy, probably I'm missing something.

$\endgroup$
9
  • 4
    $\begingroup$ If you are comfortable with tempered distributions and Fourier transform, you can write a short proof of a much more general fact: any harmonic tempered distribution $u$ is a (harmonic) polynomial. Indeed, $u$ being harmonic implies $|\xi|^2 \hat u(\xi)$ being equal to zero, which easily implies that $\hat u$ is a distribution supported at $\{0\}$, thus the Fourier transform of a polynomial. $\endgroup$ Commented Jan 11, 2021 at 18:45
  • $\begingroup$ @MateuszKwaśnicki Thank you! This is very interesting. But actually I'd like to avoid the Fourier transform too in the proof. $\endgroup$
    – Lao
    Commented Jan 11, 2021 at 19:53
  • 1
    $\begingroup$ You could use the Bochner formula and integration by parts to show the Hessian of $u$ identically vanishes. This means $u$ is affine and hence constant. $\endgroup$
    – RBega2
    Commented Jan 11, 2021 at 21:19
  • 1
    $\begingroup$ The suggested Bochner formula states that $\Delta \lvert Du \rvert^2 = 2 \lvert D^2 u \rvert^2$ for a harmonic function $u$. You could then for example integrate this against a suitable sequence of test functions $(\varphi_i \mid i \in \mathbf{N})$. In this approach the assumed bound would be used via a Cauchy-Schwarz inequality. $\endgroup$
    – Leo Moos
    Commented Jan 11, 2021 at 22:59
  • 1
    $\begingroup$ @Lao The problem with your comment is that you forgot the boundary term in your integration by parts. $\endgroup$
    – Leo Moos
    Commented Jan 11, 2021 at 23:27

2 Answers 2

3
$\begingroup$

Below we will prove, by purely energy methods, the following sharper statement: Let $u$ be a harmonic function which satisfies $$\liminf_{r \to \infty} \frac1{r^2} \frac{1}{|B_r|}\int_{B_r} |u|^2 = 0.$$ Then $u$ is constant.

All Liouville theorems of this sort are soft/qualitative versions of harder/more quantitative regularity statements. In this case, the Liouville statement follows from the interior gradient $L^\infty$ bound for harmonic functions (as we will see below). So another version of your question could be: Is there an energy methods proof of the interior $L^\infty$ bound for harmonic functions which in particular does not use the mean value property?

The classical energy methods proof of the pointwise estimates for harmonic functions uses only two ingredients: the Caccioppoli inequality (the most basic energy estimate), and the Sobolev inequality (because you have to get pointwise bounds from $L^2$ bounds somehow!).

The Caccioppoli inequality says that \begin{equation*} \int_{B_r} |\nabla u|^2 \leq \frac{C}{r^2} \int_{B_{2r}} u^2 \,. \end{equation*} You get this by testing the equation with $\varphi^2u$ where $\varphi$ is an appropriate cutoff function.

Iterating the Caccioppoli inequality many times, we get \begin{equation*} \int_{B_r} |\nabla^k u|^2 \leq \frac{C}{r^{2k}} \int_{B_{2^kr}} u^2 \,. \end{equation*} If $k > 1+\frac d2$, then we can the Sobolev inequality to get \begin{equation*} \| \nabla u \|_{L^\infty(B_r)}^2 \leq \frac{C}{r^2} \frac{1}{|B_r|} \int_{B_{2^kr}} u^2\,. \end{equation*} Now send $r\to \infty$ along a good subsequence for the liminf, and you discover that $\nabla u=0$, which means $u$ is constant.

$\endgroup$
0
5
$\begingroup$

Here is a more thorough write up of my comment.

Fix a non-negative smooth function $\phi$ which is identically $1$ on $B_1$ and vanishes identically outside $B_2$. Pick $M$ so $|\Delta \phi| \leq M$. Set $\phi_R(x)=\phi(x/R)$. We have $|\Delta \phi_R(x)|\leq M/R^2$.

By the Bochner identity $$ \Delta \frac{1}{2} |\nabla u|^2= \nabla u \cdot \nabla \Delta u + |\nabla^2 u|^2=|\nabla^2 u|^2 $$

We have $$ \int_{B_R} |\nabla^2 u|^2\leq \int_{\mathbb{R}^n} \phi_R |\nabla^2 u|^2=\frac{1}{2} \int_{\mathbb{R}^n} \phi_R \Delta |\nabla u|^2\leq \frac{M}{2R^2} \int_{B_{2R}} |\nabla u|^2 \leq \frac{CM}{2 R^2}. $$ Sending $R\to \infty$ implies $\int_{\mathbb{R}^n}|\nabla^2 u|^2=0$ so $\nabla^2 u$ vanishes identically.

This means $u(x)=\mathbf{a}\cdot x +b$ but finite energy forces $\mathbf{a}=0$.

$\endgroup$
14
  • $\begingroup$ Thank you. Two questions: (1) At which point was the assumption $\int_{\mathbb{R}^n}|\nabla u|^2 dx \leq C$ used? (2) Why the simpler observation $-\Delta u = 0 \implies \int_{\mathbb R^n} |\nabla u|^2 = 0$ does not work? $\endgroup$
    – Lao
    Commented Jan 11, 2021 at 23:19
  • $\begingroup$ (1): Errr... finite energy was used when RBega writes "finite energy forces"! (2) try $u(x)=x$ in dimension 1. Surely your "proof" is wrong! $\endgroup$ Commented Jan 11, 2021 at 23:28
  • $\begingroup$ @leomonsaingeon (1) Thank you: I missed the last line. (2) Indeed it must be wrong, but I cannot pinpoint the flaw: do you see where the mistake is? $\endgroup$
    – Lao
    Commented Jan 11, 2021 at 23:31
  • 1
    $\begingroup$ @Lao: well, for us to try pinpointing your mistake you must first give us your proof! But this is a classical mistake, in fact: you cannot integrate by parts in the whole space, that's all $\endgroup$ Commented Jan 11, 2021 at 23:33
  • 1
    $\begingroup$ @Lao I used finite energy twice. First, for the final inequality in the string of inequalities and secondly to rule out affine functions. $\endgroup$
    – RBega2
    Commented Jan 12, 2021 at 0:03

You must log in to answer this question.

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