Assume $\lambda(P)$ is the first Dirichlet eigenvalue of a regular polygon $P$. Let $u$ be the corresponding eigenfunction, normalized by $\|u\|_{L^2(P)}=1$, and $\partial_{\nu}u$ be its normal derivative on the boundary. Is the following estimate correct: \begin{eqnarray*} \lambda(P)\geq \|\partial_{\nu} u\|^2_{L^{\infty}(\partial P)}\vert P\vert\quad? \end{eqnarray*} Here $\vert P\vert$ denotes the area of $P$.
-
$\begingroup$ Since you can multiply an eigenfunction by a non zero constant and still have an eigenfunction, your estimate cannot hold (by making this constant arbitrary large). $\endgroup$– HéhéhéCommented May 24, 2022 at 11:51
-
$\begingroup$ The eigenfunction is normalized, i.e. $\| u\|_{L^2(P)}=1$ $\endgroup$– guest61Commented May 24, 2022 at 11:53
-
1$\begingroup$ You should edit your question then because it does not appear. By the way, I guess that you are talking about eigenvalue of the laplacian operator but you didn't write it down. Moreover, it should be $\| \partial_\nu u \|$ rather than $\| \nabla u\|$, shoudn't it? $\endgroup$– HéhéhéCommented May 24, 2022 at 11:56
-
$\begingroup$ @Héhéhé: If we're talking about Dirichlet eigenvalues the tangential derivative of the eigenfunction is zero, since it is constant in the tangential direction. The norm of the gradient or the normal derivative gives the same thing. $\endgroup$– Beni BogoselCommented Dec 10 at 13:22
3 Answers
Ignoring the normalization, your inequality writes $$\lambda(P)\|u\|_{L^2(P)}^2\ge\|\partial_\nu u\|_{L^\infty(\partial P)}^2|P|\qquad?$$ Good news: this is scaling invariant.
Bad news: this is false for a square $P=(-\pi/2,\pi/2)^2$. Then $u(x,y)=\cos x\cos y$,$\lambda(P)=2$, $|P|=\pi^2$ and $$\|u\|_{L^2}^2=\left(\int_{-\pi/2}^{\pi/2}\cos^2 x\,dx\right)^2=\frac{\pi^2}4\,,\qquad\|\partial_\nu u\|_{L^\infty}^2=1.$$
Perhaps your inequality is correct with an extra constant factor.Perhaps also you should think to an inequality of the form $$\lambda(P)\|u\|_{L^2(P)}\ge c\|\partial_\nu u\|_{L^\infty(\partial P)},$$ which is scaling invariant too. For instance the limit case of the unit disk gives $u(x,y)=a(r)$ with $$a''+\frac1r\,a'=-\lambda a \quad a'(0)=0,\quad a(1)=0.$$ Then \begin{eqnarray*} \|\partial_\nu u\|_{L^\infty(\partial P)} & = & |a'(1)|=\left|\int_0^1(ra'(r)'dr\right| \\ & = & \lambda\left|\int_0^1ra(r)dr\right|\le\frac\lambda{2\sqrt\pi}\,\|u\|_{L^2(P)}. \end{eqnarray*}
-
$\begingroup$ Many thanks for the detailed answer and the counter example. Unfortunately I need the constant $1$ and the power $2$. For the disk the desired estimate holds with equality. Maybe there is some hope that it holds for a regular $N$ -polygon for $N$ large enough. Otherwise the reverse inequality may have a chance for all $N$. $\endgroup$– guest61Commented May 24, 2022 at 14:23
-
$\begingroup$ @guest61: the estimate you want holds in the other sense, as indicated in my answer. $\endgroup$ Commented Dec 10 at 13:30
There is a more general reason why any such statement will fail: If you consider a $P$ consisting of $N$ separate copies of the same basic region $P_0$, then $|P|=N|P_0|$, while everything else in your inequality is independent of $N$, so the inequality will fail for large $N$. (This problem has a degenerate ground state, but of course you could address this by slightly changing the shapes.)
If you connect the components by thin tubes, then you now have a connected $P$ and are still approximately in the situation described above.
The quantity $\lambda(P)|P|$ is scale invariant since $\lambda(tP)=\frac{1}{t^2}\lambda(P)$. Therefore, its derivative with perturbations preserving the regularity of the polygon equals zero.
The derivative of the eigenvalue can be found using the shape derivative formula:
$$\lambda'(P)(\theta) = -\int_{\partial P} (\partial_n u)^2 (\theta \cdot n)$$
It is immediate to notice that a vector field which corresponds to dilations verifies $\theta \cdot n = 1$.
In the same case the shape derivative of the area gives
$$ |P|'(\theta) = \int_{\partial P} \theta \cdot n.$$
This gives $\lambda(P)|P|'=-|P|\lambda'(P)(\theta) = |P|\int_{\partial P}(\partial_n u)^2$.
In the end: $\lambda(P)=|P| \frac{1}{|\partial P|}\int_{\partial P}(\partial_n u)^2$.
If my reasoning is correct, this is an equality (always better than an inequality :) )
Therefore the inequality the OP wants goes in the other sense (the average is smaller than the $L^\infty$ norm).