where $B(y^p,1/p,1-1/p)$ and $B(1/p,1-1/p)$ stand for the incomplete and complete Beta functions, respectively. See, e.g., Eq. (2.15) in [1]. Here, $\pi_p = \frac{2\pi}{p \sin(\pi/p)}$.
Consider the function $\sin_p(\pi_p x/2)$ for $x \in (0,1)$ which is then defined as the inverse of
\[\frac{B(y^p,1/p,1-1/p)}{B(1/p,1-1/p)}, \quad y \in (0,1).\]Using the relation between the derivatives (wrt a parameter) of a function and its inverse, and by launching, say, Mathematica, we can obtain the following figures. (Of course, if I didn’t mess up with the code). The blue graphs are the graphs of $\sin_p(\pi_p x/2)$ for $x \in (0,1)$, and the orange ones are the graphs of the corresponding derivative wrt $p$.
Last modified: 23-Mar-22
Bushell, P. J., Edmunds, D. E. (2012). Remarks on generalized trigonometric functions. The Rocky Mountain Journal of Mathematics, 25-57. ↩
where $u = u^+ - u^-$, and $u^\pm := \max{\pm u, 0}$.
In [1] it is proved that the first nontrivial curve of the Fucik spectrum can be described as a set of points $(s + c(s), c(s))$, where $s \in \mathbb{R}$ and $c(s)$ defined by
\[c(s) = \inf_{\gamma \in \Gamma} \max_{u \in \gamma[-1,1]} \left(\int_{\Omega}|\nabla u|^p \, dx - s \int_{\Omega}|u^+|^p \, dx \right).\]Here
\[\Gamma := \{\gamma \in C([-1,1], S):~ \gamma(-1) = -\varphi_1,~ \gamma(1) = \varphi_1 \},\]where $S :=\{w \in W_0^{1,p}:~ \|w\|_{L^p}=1\}$ and $\varphi_1$ is the first eigenfunction.
There is another characterization of the first nontrivial curve of the Fucik spectrum. Namely, consider
\[\alpha^*(\beta) := \inf\left\{ \frac{\int_{\Omega}|\nabla u^-|^p \, dx}{\int_{\Omega}|u^-|^p \, dx}:~ u \in W_0^{1,p},~ u^\pm \not\equiv 0,~ \frac{\int_{\Omega}|\nabla u^+|^p \, dx}{\int_{\Omega}|u^+|^p \, dx} = \beta \right\}.\]Note that the admissible set for this minimization problem is nonempty for all $\beta > \lambda_1(p)$. This definition is, in essence, the same as of Theorem 1.2 in [2] for the linear case $p=2$ (see also [3]), and it was pointed out in that works that for $p>1$ this definition is also ok. Let us prove this fact explicitly.
Proposition. The set of points $(\alpha^*(\beta), \beta)$ is the first nontrivial curve of the Fucik spectrum.
Proof. The main idea is to switch between the parametrizations: $c(s)$ parametrized by diagonals, while $\alpha^*(\beta)$ is parametrized by horizontal lines. Note that $c(s)$ is strictly decreasing (see Propositions 4.1 in [1]), i.e., $c(s) > c(s’)$ whenever $s < s’$; moreover, $c(s) \to \lambda_1(p)$ as $s \to +\infty$, see Proposition 4.4 in [1]. Thus, for each $\beta > \lambda_1(p)$ there exists unique $s \in \mathbb{R}$ such that $\beta = c(s)$. (See figure below). Notice that the $c(s)$ is constructed in [1] only for $s \geq 0$ and then the constructed part is reflected with respect to the bisector $\alpha = \beta$. However, it doesn’t cause troubles.
Let us show now that $\alpha^*(c(s)) = s + c(s)$ for any $c(s) = \beta > \lambda_1(p)$. Note first that the eigenfunction which corresponds to $(\alpha, \beta) = (s+c(s), c(s))$ is always an admissible point for $\alpha^*(c(s))$, and hence $\alpha^*(c(s)) \leq s + c(s)$. Suppose, by contradiction, that $\alpha^*(c(s)) < s + c(s)$ for some $s$. Then, by definition of $\alpha^*(c(s))$, there have to exist a function $u \in W_0^{1,p}$ such that
\[\alpha^*(c(s)) \leq \frac{\int_{\Omega}|\nabla u^-|^p \, dx}{\int_{\Omega}|u^-|^p \, dx} < s+ c(s) \quad \text{and} \quad \frac{\int_{\Omega}|\nabla u^+|^p \, dx}{\int_{\Omega}|u^+|^p \, dx} = \beta = c(s).\]Due to the continuity and monotonicity of $c(s)$ (see Proposition 4.1 in [1]), there exists $s_0$ such that
\[\frac{\int_{\Omega}|\nabla u^-|^p \, dx}{\int_{\Omega}|u^-|^p \, dx} = s_0 + c(s) < s_0 + c(s_0) \quad \text{and} \quad \frac{\int_{\Omega}|\nabla u^+|^p \, dx}{\int_{\Omega}|u^+|^p \, dx} = \beta < c(s_0),\]or, equivalently,
\[\int_{\Omega}|\nabla u^-|^p \, dx < (s_0 + c(s_0))\int_{\Omega}|u^-|^p \, dx \quad \text{and} \quad \int_{\Omega}|\nabla u^+|^p \, dx < c(s_0) \int_{\Omega}|u^+|^p \, dx,\]which is, in fact, the main contradictory assumption in the proof of Theorem 3.1 in [1] (see also the proof of Lemma 5.3, (5.10) in [1]). Thus, proceeding exactly as in the proof of Theorem 3.1 in [1], we obtain a contradiction to the definition of $c(s_0)$.
Update: A closely related characterization is provided in [Horak, J., & Reichel, W. (2003). Analytical and numerical results for the Fučı́k spectrum of the Laplacian. Journal of computational and applied mathematics, 161(2), 313-338.]
The PDF-version of this post can be found here.
Last modified: 27-Oct-22
Cuesta, M., De Figueiredo, D., Gossez, J. P. (1999). The beginning of the Fucik spectrum for the p-Laplacian. Journal of Differential Equations, 159(1), 212-238. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8
Conti, M., Terracini, S., Verzini, G. (2005). On a class of optimal partition problems related to the Fucik spectrum and to the monotonicity formulae. Calculus of Variations and Partial Differential Equations, 22(1), 45-72. ↩
Molle, R., Passaseo, D. (2015). Variational properties of the first curve of the Fucik spectrum for elliptic operators. Calculus of Variations and Partial Differential Equations, 54(4), 3735-3752. ↩
Here $\Omega \subset \mathbb{R}^N$ is a bounded domain whose boundary $\partial \Omega$ consists of two connected components: the outer one - $\Gamma_0$, and the inner one - $\partial \Omega \setminus \Gamma_0$.
Let $\Omega^\#$ be the spherical shell (i.e., difference of two balls centred at the same point) of the same measure as $\Omega$ and such that the outer boundary of $\Omega^\#$ has the same $(N-1)$-measure as $\Gamma_0$.
It was proved by Payne & Weinberger [1] that in the planar case $N=2$, $\nu_1(\Omega)$ satisfies the following reversed Faber-Krahn inequality:
\[\nu_1(\Omega) \leq \nu_1(\Omega^\#).\]This result was generalized to the $N$-dimensional case, as well as to the $p$-Laplacian version of the problem, by Anoop & Kumar [2], under the assumption that $\Gamma_0$ is a sphere.
In Section 4 of the same paper [2], it is asked whether there exists a universal constant $C>0$ such that the following lower bound for $\nu_1(\Omega)$ takes place:
\[C \nu_1(\Omega^\#) \leq \nu_1(\Omega).\]Let us show that such $C$ does not exists, in general. To this end, we will construct a sequence of equimeasurable domains $\Omega_n$ with equimeasurable outer boundary, such that $\nu_1(\Omega_n) \to 0$.
We construct $\Omega_n$ as follows. Let the outer boundary $\Gamma_0$ of $\Omega_n$ be a sphere $\partial B$. Now we consider a smaller concentric spherical shell $\Theta_n$ with a cylindrical hole $H_n$ of radius $1/n$ in it, as depicted in the following figure:
Clearly, the boundary of $\Theta_n \setminus H_n$ is connected. Moreover, we assume that the inner boundary of $\Theta_n$ is of a fixed radius. We set $\Omega_n = B \setminus (\Theta_n \setminus H_n)$, and we assume the width of $\Theta_n$ to be such that $\Omega_n$ is of a constant measure with respect to $n$.
Note that $\nu_1(\Omega)$ has the following variational characterization:
\[\nu_1(\Omega) = \inf\left\{ \frac{\int_\Omega |\nabla u|^2 \, dx}{\int_\Omega |u|^2 \, dx}: u \in W^{1,2}(\Omega) \setminus \{0\},~ u=0 ~\text{on}~ \Gamma_0 \right\}.\]Let us take the following admissible function for this minimization problem. We set $u_n = 0$ in the “outer” part of $\Omega_n$, and $u_n=C=const$ in the “inner” part of $\Omega_n$. In the cylinder $H_n$ we assume $u_n$ to be, e.g., a linear function like $u_n(x_1,\dots,x_N) = ax_1+b$, where $a$ and $b$ are adjusted in such a way that $u_n$ is continuous. See the figure below.
Thus, we get
\[\nu_1(\Omega_n) \leq \frac{\int_{\Omega_n} |\nabla u_n|^2 \, dx}{\int_{\Omega_n} |u_n|^2 \, dx} = \frac{a^2 |H_n|}{\int_{H_n} |ax_1 + b|^2 \, dx + C^2 \times (\text{measure of the "inner" part of}~ \Omega_n)}.\]Clearly, the numerator goes to zero, while the denominator is bounded away from zero. Thus, the desired result follows.
Last modified: 11-Aug-20
Payne, L. E., & Weinberger, H. F. (1961). Some isoperimetric inequalities for membrane frequencies and torsional rigidity. Journal of Mathematical Analysis and Applications, 2(2), 210-216. ↩
Anoop, T. V., & Kumar, K. A. (2020). On reverse Faber-Krahn inequalities. Journal of Mathematical Analysis and Applications, 485(1), 123766. ↩ ↩2
upon domain perturbations driven by a family of diffeomorphisms
\[\Phi_t(x) = x + t R(x), \quad R \in C^1(\mathbb{R}^N, \mathbb{R}^N), \quad |t|<\delta.\]Here $F$ satisfies certain rather classical conditions.
Let us take an arbitrary minimizer $v_0$ of $E$ over the Nehari manifold $\mathcal{N}(\Omega)$ and consider a function $v_t(y) := v_0(\Phi_t^{-1}(y))$, $y \in \Omega_t$. One of the main results of our paper is the following Hadamard-type formula:
\[\left. \frac{\partial E[\alpha(v_t) v_t]}{\partial t} \right|_{t=0} = - \frac {p-1} p \int_{\partial \Omega} \left| \frac{\partial v_0}{\partial n} \right|^p \left<R, n\right> \, d\sigma,\]where $\alpha(v_t) \in \mathbb{R}$ is a normalization coefficient such that $\alpha(v_t) v_t \in \mathcal{N}(\Phi_t(\Omega))$, and $n$ is the outward unit normal vector to $\partial \Omega$.
In the particular case $F(u) = |u|^q$, $q \in [1, p^*)$, it can easily be checked that the finding of the least critical level of $E$ can be restated as the finding of minimum for the problem
As a corollary of the proof of the Hadamard formula above, we obtain the following fact. If $u_0$ is a minimizer of $\mu_q(\Omega)$ normalized such that $|u_0|_{L^q(\Omega)} = 1$, and $u_t(y) := u_0(\Phi_t^{-1}(y))$, $y \in \Omega_t$, then
\begin{equation}\label{eq:had2} \left. \frac{\partial J(u_t)}{\partial t} \right|_{t=0} = - (p-1) \int_{\partial \Omega} \left| \frac{\partial u_0}{\partial n} \right|^p \left<R, n\right> \, d\sigma. \end{equation}
Short historical remark. If $q=p$, then $\mu_p(\Omega)$ is the first eigenvalue of the $p$-Laplacian, and the formula \eqref{eq:had2} (in fact, the actual derivative of $\mu_p(\Omega)$) was established by Garcia Melian & Sabina de Lis [2] in (2001) for $p>1$, and by Lamberti [3] in (2003) for $p \geq 2$. In the case $p=2$, this formula goes back to Hadamard (1908), see, e.g., [4].
However it appeared recently that the nonlinear case has been studied much earlier in a work of Roppongi [5] who established the Hadamard formula \eqref{eq:had2} for $p \geq 2$ and $q \geq p$ in (1994). In fact, the article of Roppongi is build upon the article of Osawa [6] who considered the case $p=2$, $q>2$. Moreover, Osawa also treated the Hadamard formula under the Robin boundary condition and the Neumann boundary condition. The reader will be able to find easily a few more related works of this Japanese group by looking at citations of Osawa’s article. Later, in (2017), the Hadamard formula \eqref{eq:had2} was independently stated in [7] by referring to the arguments from [2].
It is quite a pity that we didn’t know these references before and didn’t include them in our paper. Moreover, it seems that these works of the Japanese group are forgotten and have not being cited in related contemporary research. It makes sense to revive them.
The PDF-version of this post can be found here.
Last modified: 25-Apr-20
Bobkov, V., & Kolonitskii, S. (2020). On qualitative properties of solutions for elliptic problems with the $p$-Laplacian through domain perturbations. Communications in Partial Differential Equations, 45(3), 230-252. ↩
Garcia Melian, J., & Sabina de Lis, J. (2001). On the perturbation of eigenvalues for the $p$-Laplacian. Comptes Rendus de l’Acad'emie des Sciences-Series I-Mathematics, 332(10), 893-898. ↩ ↩2
Lamberti, P. D. (2003). A differentiability result for the first eigenvalue of the $p$-Laplacian upon domain perturbation. Nonlinear analysis and applications: to V. Lakshmikantham on his 80th birthday, 1(2), 741-754. ↩
Garabedian, P. R., & Schiffer, M. (1952). Convexity of domain functionals. Journal d’Analyse Math'ematique, 2(2), 281-368. ↩
Roppongi, S. (1994). The Hadamard variation of the ground state value of some quasi-linear elliptic equations. Kodai Mathematical Journal, 17(2), 214-227. ↩
Osawa, T. (1992). The Hadamard variational formula for the ground state value of $-\Delta u = \lambda |u|^{p-1} u$. Kodai Mathematical Journal, 15(2), 258-278. ↩
Carroll, T., Fall, M. M., & Ratzkin, J. (2017). On the rate of change of the sharp constant in the Sobolev-Poincare inequality. Mathematische Nachrichten, 290(14-15), 2185-2197. ↩
Theorem. Let $1 < q < p < \infty$ and $\alpha, \beta > 0$. Assume that $u>0$ and $\varphi \geq 0$ are some differentiable functions in a domain $\Omega$. Then
\[|\nabla u|^{p-2} \nabla u \nabla \left(\frac{\varphi^p}{\alpha u^{p-1} + \beta u^{q-1}}\right) \leq \frac{1}{\alpha C} |\nabla \varphi|^p\]and
\[|\nabla u|^{q-2} \nabla u \nabla \left(\frac{\varphi^p}{\alpha u^{p-1} + \beta u^{q-1}}\right) \leq \frac{1}{\beta}|\nabla (\varphi^{p/q})|^q,\]where $C = 1$ if $p \leq q+1$, and $C= \frac{(q-1)^{p-2} (p-q)}{(p-2)^{p-2}}$ if $p \geq q+1$.
In particular, if $\mu>0$, then
\[(|\nabla u|^{p-2} + \mu |\nabla u|^{q-2}) \nabla u \nabla \left(\frac{\varphi^p}{\alpha u^{p-1} + \beta u^{q-1}}\right) \leq \frac{1}{\alpha C} |\nabla \varphi|^p + \frac{\mu}{\beta}|\nabla (\varphi^{p/q})|^q.\]Proof. First, by standard calculations we get
\begin{align} \notag & |\nabla u|^{p-2} \nabla u \nabla \left( \frac{\varphi^p}{\alpha u^{p-1} + \beta u^{q-1}} \right)\newline \notag & = p |\nabla u|^{p-2} \nabla u \nabla \varphi \frac{\varphi^{p-1}}{\alpha u^{p-1} + \beta u^{q-1}} - |\nabla u|^{p} \varphi^p \frac{\alpha (p-1) u^{p-2} + \beta (q-1) u^{q-2}}{\left(\alpha u^{p-1} + \beta u^{q-1}\right)^2}\newline \label{eq:picone:1} & \leq p |\nabla u|^{p-1} |\nabla \varphi| \frac{\varphi^{p-1}}{\alpha u^{p-1} + \beta u^{q-1}} - |\nabla u|^{p} \varphi^p \frac{\alpha (p-1) u^{p-2} + \beta (q-1) u^{q-2}}{\left(\alpha u^{p-1} + \beta u^{q-1}\right)^2}. \end{align}
Applying to the first term Young’s inequality
\[ab =\frac{a}{\rho^{\frac{p-1}{p}}}\,\rho^{\frac{p-1}{p}} b \leq \frac{|a|^p}{p\rho^{p-1}} + \frac{\rho (p-1) |b|^{\frac{p}{p-1}}}{p}\]with $a= |\nabla \varphi|$, $b= |\nabla u|^{p-1} \frac{\varphi^{p-1}}{\alpha u^{p-1} + \beta u^{q-1}}$, and any $\rho>0$, we obtain
\begin{align} \notag \eqref{eq:picone:1} & \leq \frac{|\nabla \varphi|^p}{\rho^{p-1}} + \frac{\rho(p-1) |\nabla u|^{p} \varphi^p}{\left(\alpha u^{p-1} + \beta u^{q-1}\right)^{\frac{p}{p-1}}} - |\nabla u|^{p} \varphi^p \frac{\alpha (p-1) u^{p-2} + \beta (q-1) u^{q-2}}{\left(\alpha u^{p-1} + \beta u^{q-1}\right)^2}\newline \notag & = \frac{|\nabla \varphi|^p}{\rho^{p-1}} + \frac{\rho(p-1) |\nabla u|^{p} \varphi^p}{\left( \alpha u^{p-1} + \beta u^{q-1} \right)^{\frac{p}{p-1}}} \left[ 1 - \frac{\alpha (p-1) + \beta (q-1) u^{q-p}}{\rho(p-1) \left(\alpha + \beta u^{q-p}\right)^{\frac{p-2}{p-1}}} \right]. \end{align}
Analysing the function
\[g(u) = \frac{\alpha (p-1) + \beta (q-1) u^{q-p}}{\rho(p-1) \left(\alpha + \beta u^{q-p}\right)^{\frac{p-2}{p-1}}},\]we see that $g(u) \geq 1$ for all $u > 0$ in the following cases:
\begin{align} \notag &1)\quad p \leq q+1 \quad \text{and} \quad \rho \leq \alpha^\frac{1}{p-1},\newline \notag &2) \quad p \geq q+1 \quad \text{and} \quad \rho \leq \frac{(q-1)^\frac{p-2}{p-1}(p-q)^\frac{1}{p-1}}{(p-2)^\frac{p-2}{p-1}}\alpha^\frac{1}{p-1}. \end{align}
Under the assumptions 1) or 2) we get
\[|\nabla u|^{p-2} \nabla u \nabla \left( \frac{\varphi^p}{\alpha u^{p-1} + \beta u^{q-1}} \right) \leq \frac{|\nabla \varphi|^p}{\rho^{p-1}}.\]Thus, taking $\rho$ as the maximal admissible value in the cases 1) and 2), we obtain the constat $C_1$.
Arguing in the similar way, for $\psi=\varphi^{p/q}$ (note that $p/q>1$ and $\varphi^p=\psi^q$) we obtain
\begin{align} \notag & |\nabla u|^{q-2} \nabla u \nabla \left( \frac{\psi^q}{\alpha u^{p-1} + \beta u^{q-1}} \right) \newline \notag &\le \frac{|\nabla \psi|^q}{\rho^{q-1}} + \frac{\rho(q-1) |\nabla u|^{q} \psi^q}{\left( \alpha u^{p-1} + \beta u^{q-1} \right)^{\frac{q}{q-1}}} \left[ 1 - \frac{\alpha (p-1) u^{p-q} + \beta (q-1)}{\rho(q-1) \left(\alpha u^{p-q} + \beta \right)^{\frac{q-2}{q-1}}} \right]. \end{align}
Analysing the function
\[g(u) = \frac{\alpha (p-1) u^{p-q} + \beta (q-1)}{\rho(q-1) \left(\alpha u^{p-q} + \beta \right)^{\frac{q-2}{q-1}}},\]we see that $g(u) \geq 1$ for all $u > 0$ provided $\rho \leq \beta^\frac{1}{q-1}$. Thus, taking $\rho = \beta^\frac{1}{q-1}$, we get
\[|\nabla u|^{q-2} \nabla u \nabla \left( \frac{\psi^q}{\alpha u^{p-1} + \beta u^{q-1}} \right) \le \frac{|\nabla \psi|^q}{\rho^{q-1}} = \frac{|\nabla \psi|^q}{\beta}.\]Combining the above inequalities, we complete the proof.
Allegretto, W., & Huang, X. Y. (1998). A Picone’s identity for the $p$-Laplacian and applications. Nonlinear Analysis: Theory, Methods & Applications, 32(7), 819-830. ↩
Bobkov, V., & Tanaka, M. (2015). On positive solutions for $(p,q)$-Laplace equations with two parameters. Calculus of Variations and Partial Differential Equations, 54(3), 3277-3301. ↩ ↩2
Let $u = u(x,y) = u(r,\theta)$, where $r>0$ and $\theta \in (-\pi, \pi)$. Then we have
\begin{align} \notag \Delta_p u(x,y) &= \left(u_r^2 + \frac{u_\theta^2}{r^2} \right)^\frac{p-4}{2} \newline \notag &\times \left( (p-1) u_r^2 u_{rr} + \frac{u_r^3}{r} + \frac{2(p-2) u_r u_\theta u_{r\theta}}{r^2} + \frac{u_r^2 u_{\theta \theta}}{r^2} + \frac{u_{rr} u_\theta^2}{r^2} - \frac{(p-3) u_r u_\theta^2}{r^3} + \frac{(p-1) u_\theta^2 u_{\theta \theta}}{r^4} \right). \end{align}
In particular, if $u = u(x,y) = u(r)$, then $u_\theta = u_{\theta \theta}=0$, and we get the usual
\[\Delta_p u(x,y) = (p-1) |u_r|^{p-2} u_{rr} + \frac{|u_r|^{p-2} u_r}{r}.\]To obtain the claimed expression, let us write all necessary partial derivatives in polar coordinates:
\[u_{x} = u_r \cos \theta - u_\theta \frac{\sin \theta}{r},\] \[u_y = u_r \sin \theta + u_\theta \frac{\cos \theta}{r},\] \[u_{xx} = u_{rr} \cos^2 \theta - u_{r\theta} \frac{2 \cos \theta \sin \theta}{r} + u_{\theta \theta} \frac{\sin^2 \theta}{r^2} + u_r \frac{\sin^2 \theta}{r} + u_{\theta} \frac{2 \cos \theta \sin \theta}{r^2},\] \[u_{yy} = u_{rr} \sin^2 \theta + u_{r\theta} \frac{2 \cos \theta \sin \theta}{r} + u_{\theta \theta} \frac{\cos^2 \theta}{r^2} + u_r \frac{\cos^2 \theta}{r} - u_{\theta} \frac{2 \cos \theta \sin \theta}{r^2},\] \[u_{xy} = u_{rr} \cos \theta \sin \theta + u_{r\theta} \frac{\cos^2 \theta - \sin^2 \theta}{r} - u_{\theta \theta} \frac{\cos \theta \sin \theta}{r^2} - u_r \frac{\cos \theta \sin \theta}{r} - u_{\theta} \frac{\cos^2 \theta - \sin^2 \theta}{r^2}.\]Then, we substitute these expressions into the pointwise formula of the $p$-Laplacian in $2D$ (see, e.g., here):
\[\Delta_p u = \text{div}\left(|\nabla u|^{p-2} \nabla u\right) = |\nabla u|^{p-4} \left( |\nabla u|^2 \Delta u + (p-2) u_x^2 u_{xx} + (p-2) u_y^2 u_{yy} + 2(p-2) u_{x}u_y u_{xy}. \right).\]After some standard simplifications, we arrive at the desired expression.
]]>It is known that $H(n) < n^{\eta}$, where $\eta = 1.37779\dots$ is the unique positive real zero of $\left(1-\frac{1}{2^s}\right) \zeta(s)=2$, see Theorem 5 in [1]. Our aim is to show that this upper bound is optimal.
Lemma. For any $\varepsilon>0$ there exist infinitely many odd $n$ such that $H(n) > n^{\eta-\varepsilon}$.
Proof. We will argue along the same lines as in Section 3 of [2], where the similar optimality was obtained for all (even) numbers. First, we notice that
\[\left(1-\frac{1}{2^s}\right) \zeta(s) = \sum_{m=1,~ m ~\text{odd}}^{\infty} \frac{1}{m^s}.\]Since this expression is a decreasing function of $s$, we take any $\varepsilon>0$ and find that
\[\sum_{m=1,~ m ~\text{odd}}^{\infty} \frac{1}{m^{\eta-\varepsilon}} > 2,\]and there exists some odd $b$ such that
\[\sum_{m=1,~ m ~\text{odd}}^{b} \frac{1}{m^{\eta-\varepsilon}} > 2.\]Recalling the monotonicity property and the definition of $\eta$, we can find some $\gamma \in (\eta-\varepsilon, \eta)$ such that
\[\sum_{m=1,~ m ~\text{odd}}^{b} \frac{1}{m^{\gamma}} = 2,\]which we rewrite as
\[\sum_{m=3,~ m ~\text{odd}}^{b} \frac{1}{m^{\gamma}} = 1.\]Let us now consider for each odd $m \in [3,b]$ and sufficiently large $t>0$ the function
\[c_m = c_m(t) = \text{floor}\left(\frac{t}{m^\gamma}\right).\]Notice that
\[\frac{t}{m^\gamma} -1 \leq c_m \leq \frac{t}{m^\gamma},\]which yields, in particular, that
\begin{equation}\label{eq:summ} \left(\frac{\sum_m c_m}{t}\right)^{\sum_m c_m} \geq \left(\sum_{m=3,~ m ~\text{odd}}^{b} \frac{1}{m^{\gamma}}-\frac{b}{t}\right)^{\sum_m c_m} = \left(1-\frac{b}{t}\right)^{\sum_m c_m} \geq \left(1-\frac{b}{t}\right)^{t} = e^{-b} + o(1). \end{equation}
This inequality will be used slightly below.
Let us define
\[n = \prod_{m=3,~ m ~\text{odd}}^b m^{c_m}.\]Since $m$ is odd, we see that $n$ is odd. Moreover, clearly, $H(n)$ is bounded from below by the number of multiset permutations
\[v(n) = \frac{(\sum_m c_m)!}{\prod_m c_m!}.\]Let us estimate $v(n)$ by Stirling’s bounds $\sqrt{2\pi} n^{n+1/2} e^{-n} \leq n! \leq e n^{n+1/2} e^{-n}$:
\[v(n) \geq \frac{\sqrt{2\pi}}{e^b} \sqrt{\frac{\sum_m c_m}{\prod_m c_m}} \prod_i\left(\frac{\sum_m c_m}{c_i}\right)^{c_i}.\]First, recalling that $c_i \leq \frac{t}{i^\gamma}$ and using \eqref{eq:summ}, we get
\[\prod_i \left(\frac{\sum_m c_m}{c_i}\right)^{c_i} \geq \prod_i \left(\frac{\sum_m c_m}{t}\right)^{c_i} \left(\prod_i i^{c_i}\right)^\gamma = \left(\frac{\sum_m c_m}{t}\right)^{\sum_m c_m} n^\gamma \geq (e^{-b} + o(1)) n^\gamma.\]Second, arguing exactly as in [2], we deduce that
\[\sqrt{\frac{\sum_m c_m}{\prod_m c_m}} \geq c_b \frac{1}{(\ln n)^\frac{b-5}{4}}\]for some $c_b > 0$.
Combining the lattter two inequalities, we deduce that
\[H(n) \geq v(n) \geq C \frac{n^\gamma}{(\ln n)^\frac{b-5}{4}} > n^{\eta-\varepsilon}\]for some $C>0$ and all sufficiently large $t$. The proof is complete.
In the linear case $p=2$, the well-known Courant nodal domain theorem says that $\mu(\varphi_n) \leq n$ for all $n \geq 1$. Its generalization to the nonlinear case $p \neq 2$ obtained in [1] asserts that
\[\mu(\varphi_n) \leq 2n-2 \quad \text{for all } n \geq 2,\]which implies
\[\limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq 2.\]On the other hand, in the linear case $p=2$, there is a result of Pleijel [2] on the following asymptotic refinement of the Courant nodal domain theorem:
\begin{equation}\label{eq:P} \limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{4}{j_{0,1}^2} = 0.69166\ldots, \end{equation}
see, e.g., this post for a discussion.
The aim of the present post is to generalize the result of Pleijel to the $p$-Laplacian settings. Pleijel’s approach is purely variational and consists of two main ingredients: the Faber-Krahn inequality and the Weyl law.
where $B_1$ is a unit ball in $\mathbb{R}^N$; see, e.g., the discussion here. Therefore, noting that $\lambda_n(\Omega) = \lambda_1(\Omega_i)$ for any $i=1..\mu(\varphi_n)$ where $\Omega_i$ is a nodal domain of $\varphi_n$, we get
Equivalently,
\begin{equation}\label{eq:FKP} \mu(\varphi_n) \leq \frac{|\Omega| \lambda_n(\Omega)^\frac{N}{p}}{|B_1| \lambda_1(B_1)^\frac{N}{p}}. \end{equation}
Second, by the variational characterization of $\lambda_n(\Omega)$, we can estimate
\[\lambda_n(\Omega) \leq \lambda_1(Q_{h_n}),\]where $h_n$ is such that there are $n$ disjoint cubes $Q_{h_n}$ inscribed in $\Omega$. We can assume that $h_n$ is maximal.
Third, we know that
\[\lambda_1(Q_h) = \lambda_1(Q_1) h^{-p}.\]Combining the previous three facts, we get
\begin{equation}\label{eq:WP} \lambda_n(\Omega) \leq \lambda_1(Q_{h_n}) = \lambda_1(Q_1) h_n^{-p} \approx \lambda_1(Q_1) \left(\frac{n}{|\Omega|}\right)^\frac{p}{N} \quad \text{as } n \to \infty. \end{equation}
Finally, mixing \eqref{eq:FKP} and \eqref{eq:WP}, we deduce that
\begin{equation}\label{eq:Plp} \boxed{\limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{1}{|B_1|} \left(\frac{\lambda_1(Q_1)}{\lambda_1(B_1)} \right)^\frac{N}{p}.} \end{equation}
Notice that this upper bound does not depend on $\Omega$. Below, we will discuss a possible way how to improve this bound.
All we need now is to get a ‘‘good’’ upper bound for $\lambda_1(Q_1)$ and a ‘‘good’’ lower bound for $\lambda_1(B_1)$.
Let us start with an upper bound for $\lambda_1(Q_1)$. From Proposition 2.7 of [5] we know that
\[\lambda_1(Q_1) \leq \widetilde{\pi}_p^p N \quad \text{for} \quad p<2\]and
\[\lambda_1(Q_1) \leq \widetilde{\pi}_p^p N^\frac{p}{2} \quad \text{for} \quad p>2,\]where
\[\widetilde{\pi}_p = (p-1)^\frac{1}{p} \frac{2 \pi}{p \sin(\pi /p)} \equiv 2 (p-1)^\frac{1}{p} \int_0^1 \frac{ds}{(1-s^p)^\frac{1}{p}}.\]As lower estimates for $\lambda_1(B_1)$, we use the estimate
\[\lambda_1(B_1) \geq N \left(\frac{p}{p-1}\right)^{p-1} \quad \text{for} \quad p<2,\] \[\lambda_1(B_1) \geq Np \quad \text{for} \quad p>2,\]see [8] and, in general, this post for a discussion of lower bounds.
Thus, substituting all these things into \eqref{eq:Plp}, we get
\begin{equation}\label{eq:Pp<2} \limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{\Gamma\left(\frac{N}{2}+1\right)\pi^\frac{N}{2} 2^N (p-1)^N}{p^\frac{(2p-1)N}{p} \sin(\pi /p)^N} \quad \text{for} \quad p<2 \end{equation}
and
\begin{equation}\label{eq:Pp>2} \limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{\Gamma\left(\frac{N}{2}+1\right)\pi^\frac{N}{2} 2^N N^\frac{(p-2)N}{2p} (p-1)^\frac{N}{p}}{p^\frac{(p+1)N}{p} \sin(\pi /p)^N} \quad \text{for} \quad p>2. \end{equation}
The corresponding plot is depicted below by the increasing line. We see that these upper bounds does not give us a Pleijel constant smaller than $1$ even in the dimension $N=2$, which is quite sad. Note that if $p \to 1$, then the bound \eqref{eq:Pp<2} approaches $\frac{4}{\pi}=1.2732\dots$, while if $p \to \infty$, then the bound \eqref{eq:Pp>2} approaches $\frac{8}{\pi}=2.5464\dots$, see the blue line on figure below.
Let us now discuss a possible improvement of \eqref{eq:Plp} which concerns an improvement of the Weyl-type upper bound. For simplicity, let us fix $N=2$. First, we can inscribe in $\Omega$ not a square tiling, but a hexagonal tiling. If $H_r$ stands for a hexagon with the inradius $r$, and if $r \to 0$, then the number $m$ of $H_r$’s disjointly inscribed in $\Omega$ is given by
\[m \approx \frac{|\Omega|}{2 \sqrt{3} r^2}.\]Therefore, analogously to \eqref{eq:WP} we get
\[\lambda_n(\Omega) \leq \lambda_1(H_{r_n}) = \lambda_1(H_1) r_n^{-p} \approx \lambda_1(H_1) \left(\frac{2\sqrt{3} n}{|\Omega|}\right)^\frac{p}{2} \quad \text{as } n \to \infty,\]and hence, from \eqref{eq:FKP},
\begin{equation}\label{eq:Plp1} \boxed{\limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{2\sqrt{3}}{|B_1|} \left(\frac{\lambda_1(H_1)}{\lambda_1(B_1)} \right)^\frac{2}{p}.} \end{equation}
Noting that $B_1 \subset H_1$, we get $\lambda_1(H_1) \leq \lambda_1(B_1)$, which yields
\[\limsup_{n \to \infty} \frac{\mu(\varphi_n)}{n} \leq \frac{2\sqrt{3}}{\pi} = 1.1026\dots\]Moreover, if $p \to \infty$, then, by the known result from [9], $\lambda_1(H_1)^\frac{1}{p} \to 1$ and $\lambda_1(B_1)^\frac{1}{p} \to 1$, i.e., this upper estimate of the upper estimate \eqref{eq:Plp1} is sharp for $p \to \infty$. See the green line on the figure above.
Thus, unfortunately, even if $n \to \infty$, we cannot show that $\mu(\varphi_n) \leq n$ for all $p>1$ without getting a substantial improvement of the Weyl-type upper bound for $\lambda_n(\Omega)$. Such an improvement is clearly a prominent problem which needs to be studied much closer.
The PDF-version of this post can be found here.
Last modified: 18-Feb-20
Drábek, P., & Robinson, S. B. (2002). On the generalization of the Courant nodal domain theorem. Journal of Differential Equations, 181(1), 58-71. ↩
Pleijel, Å. (1956). Remarks on Courant’s nodal line theorem. Communications on pure and applied mathematics, 9(3), 543-550. ↩
Friedlander, L. (1989). Asymptotic behaviour of the eigenvalues of the $p$-laplacian. Communications in Partial Differential Equations, 14(8-9), 1059-1069. ↩
Azorero, J. G., & Peral Alonso, I. (1988). Comportement asymptotique des valeurs propres du $p$-laplacien. Comptes rendus de l’Académie des sciences. Série 1, Mathématique, 307(2), 75-78. ↩
Bonder, J. F., & Pinasco, J. P. (2008). Estimates for eigenvalues of quasilinear elliptic systems. Part II. Journal of Differential Equations, 245(4), 875-891. ↩
Bueno, H., Ercole, G., & Zumpano, A. (2009). Positive solutions for the $p$-Laplacian and bounds for its first eigenvalue. Advanced Nonlinear Studies, 9(2), 313-338. ↩
Benedikt, J., & Drábek, P. (2013). Asymptotics for the principal eigenvalue of the $p$-Laplacian on the ball as p approaches 1. Nonlinear Analysis: Theory, Methods & Applications, 93, 23-29. ↩
Benedikt, J., & Drábek, P. (2012). Estimates of the principal eigenvalue of the $p$-Laplacian. Journal of Mathematical Analysis and Applications, 393(1), 311-315. ↩
Juutinen, P., Lindqvist, P., & Manfredi, J. J. (1999). The $\infty$-eigenvalue problem. Archive for rational mechanics and analysis, 148(2), 89-105. ↩
where $\xi_1 \in W^{1,p}(\Omega)$.
The second variation (if exists) is also easy to compute:
\begin{align} \notag D^2 \left(\int_\Omega |\nabla u|^p \, dx\right) (\xi_1, \xi_2) &= p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_2) \, dx \newline \notag &+ p \int_\Omega |\nabla u|^{p-2} (\nabla \xi_1, \nabla \xi_2) \, dx, \end{align}
where $\xi_2 \in W^{1,p}(\Omega)$.
Let us make some effort to calculate the third variation:
\begin{align} \notag D^3 \left(\int_\Omega |\nabla u|^p \, dx\right) (\xi_1, \xi_2, \xi_3) &= p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_2) (\nabla u, \nabla \xi_3) \, dx \newline \notag &+ p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla u, \nabla \xi_1) (\nabla \xi_2, \nabla \xi_3) \, dx \newline \notag &+ p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla u, \nabla \xi_2) (\nabla \xi_1, \nabla \xi_3) \, dx \newline \notag &+ p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla u, \nabla \xi_3) (\nabla \xi_1, \nabla \xi_2) \, dx, \end{align}
where $\xi_3 \in W^{1,p}(\Omega)$.
We already start seeing some structure. So, let us now try to derive a general formula for the $n$-th variation of the energy functional. Our main result is the following one.
Proposition. Let $u \in W^{1,p}(\Omega)$. If for a natural $n \geq 1$ there exists $n$-th variation of the $p$-Direchlet energy of $u$ in direction $(\xi_1,\dots,\xi_n) \in (W^{1,p}(\Omega))^n$, then
\begin{align} \notag &D^n \left(\int_\Omega |\nabla u|^p \, dx\right) (\xi_1, \dots, \xi_n) \newline \notag &= \int_\Omega \left( \sum\limits_{i=0}^{\lfloor \frac{n}{2} \rfloor} |\nabla u|^{p-2(n-i)} \prod\limits_{j=0}^{n-i-1} (p-2j) \left[\sum\limits_{\sigma \in B(n,n-2i)} \prod\limits_{k=1}^{n-2i} (\nabla u, \nabla \xi_{\sigma(k)}) \left(\sum\limits_{\omega \in P(n,\sigma)} \prod\limits_{l=1}^{i} (\nabla \xi_{\omega(l,1)}, \nabla \xi_{\omega(l,2)}) \right) \right] \right) dx, \end{align}
where
$B(n,n-2i)$ is the set of all possible $(n-2i)$-combinations of $\{1,2,\dots,n\}$ such that the ordering inside each $\sigma \in B(n,n-2i)$ is immaterial. Evidently, the cardinality of $B(n,n-2i)$ is ${n \choose n-2i}$. In particular, if $i=0$, then $card(B(n,n-2i)) = 1$.
$P(n,\sigma)$ is the set of all possible partitions of the set $\{1,2,\dots,n\} \setminus \sigma$ into pairs such that the ordering of pairs and inside a pair is immaterial. Note that $card(\sigma)=n-2i$, and hence the number of pairs in each $\omega \in P(\sigma)$ is $i$. It is not hard to see that the cardinality of $P(\sigma)$ is $\frac{(2i)!}{2^i i!}$. We represent $\omega$ as a $i \times 2$-matrix $(\omega(s,t))_{s=1..i,~t=1,2}$. For instance, if $n=6$ and $\sigma = \{1,2\}$, then
Let us also calculate the fourth variation (just for fun) - either by straightforward computation, or by application of our general formula:
\begin{align}
\notag
D^4 &\left(\int_\Omega |\nabla u|^p \, dx\right) (\xi_1, \xi_2, \xi_3)
\
\notag
&=
p(p-2)(p-4)(p-6) \int_\Omega |\nabla u|^{p-8} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_2) (\nabla u, \nabla \xi_3) (\nabla u, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_2) (\nabla \xi_3, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_3) (\nabla \xi_2, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_1) (\nabla u, \nabla \xi_4) (\nabla \xi_2, \nabla \xi_3) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_2) (\nabla u, \nabla \xi_3) (\nabla \xi_1, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_2) (\nabla u, \nabla \xi_4) (\nabla \xi_1, \nabla \xi_3) \, dx
\
\notag
&+
p(p-2)(p-4) \int_\Omega |\nabla u|^{p-6} (\nabla u, \nabla \xi_3) (\nabla u, \nabla \xi_4) (\nabla \xi_1, \nabla \xi_2) \, dx
\
\notag
&+
p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla \xi_1, \nabla \xi_2) (\nabla \xi_3, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla \xi_1, \nabla \xi_3) (\nabla \xi_2, \nabla \xi_4) \, dx
\
\notag
&+
p(p-2) \int_\Omega |\nabla u|^{p-4} (\nabla \xi_1, \nabla \xi_4) (\nabla \xi_2, \nabla \xi_3) \, dx.
\end{align}
where $\xi_4 \in W^{1,p}(\Omega)$.
Visually, it could be easier to present this result as an $n$-th directional derivative of the $p$-th power of the norm of a vector. Namely, if $A, B_i \in \mathbb{R}^N$, then for any natural $n \geq 1$, we have
\begin{align} \notag &D^n \left(|A|^p\right) (B_1, \dots, B_n) \newline \notag &= \sum\limits_{i=0}^{\lfloor \frac{n}{2} \rfloor} |A|^{p-2(n-i)} \prod\limits_{j=0}^{n-i-1} (p-2j) \left[\sum\limits_{\sigma \in B(n,n-2i)} \prod\limits_{k=1}^{n-2i} (A, B_{\sigma(k)}) \left(\sum\limits_{\omega \in P(n,\sigma)} \prod\limits_{l=1}^{i} (B_{\omega(l,1)}, B_{\omega(l,2)}) \right) \right]. \end{align}
The PDF-version of this post can be found here.
Last modified: 08-Nov-18
In fact, this inequality easily follows from the general Sobolev inequality just by applying the latter one to each term on the lhs of the former one. (I’m almost sure there could be some investigation of an inequality of this type in the literature, but I was not able to find it.)
There are a couple of things about this inequality:
1) It is just beautiful that the exponents are permuted)
2) If one considers the optimization problem for the constant $C$, then it is possible to see that a corresponding minimizer of
\[\inf_{u \in W_0^{1,p}(\Omega) \setminus \{0\}} \frac{\left(\int_\Omega |\nabla u|^p \, dx \right)^{\gamma-q}}{\left(\int_\Omega |u|^q \, dx \right)^{\gamma-p} \left(\int_\Omega |u|^\gamma \, dx \right)^{p-q}}\]exists and satisfies the convex-concave equation
\[-\Delta_p u = \lambda |u|^{q-2} u + |u|^{\gamma-2} u.\]In fact, such a minimization problem was explicitly considered in connection with the above equation in [1].
Il’yasov, Y. (2005). On nonlocal existence results for elliptic equations with convex–concave nonlinearities. Nonlinear Analysis: Theory, Methods & Applications, 61(1-2), 211-236. ↩