Power-logconcavity of the Laplacian ground state

1. Problem Statement and Context

The paper addresses a fundamental question in the spectral theory of partial differential equations and convex geometry regarding the concavity properties of the first Dirichlet eigenfunction (the ground state) of the Laplacian.

• Background: The classical result by Brascamp and Lieb (1976) establishes that the first eigenfunction $u$ of the Dirichlet Laplacian on a bounded convex domain $\Omega \subset \mathbb{R}^n$ is log-concave (i.e., $\log u$ is concave). This property is crucial for understanding the fundamental gap, concentration of measure, and various geometric inequalities.
• The Open Question: Recent work by Ishige, Salani, and Takatsu introduced the concept of $\alpha$-logconcavity (or power-logconcavity). A function $u$ is $\alpha$-logconcave if the function $L_\alpha(\kappa u) = -(-\log(\kappa u))^\alpha$ is concave for some $\kappa > 0$.
  • It was known that the heat flow preserves $\alpha$-logconcavity for $\alpha \geq 1/2$.
  • Specifically, the heat kernel is $1/2$-logconcave, and this property arises spontaneously along the heat flow.
  • The Gap: It remained an open question whether the ground state $u$ itself possesses this stronger property of $(1/2)$-logconcavity.
• The Challenge: The standard limiting argument used by Brascamp-Lieb to deduce log-concavity of the ground state from the heat flow fails for $\alpha = 1/2$. The parameter $\kappa$ required to satisfy the concavity condition for the heat flow solution might degenerate to zero as time $t \to \infty$, preventing a direct limit to the eigenfunction.

2. Methodology

The authors employ a novel combination of the convex envelope method and improved gradient estimates to overcome the failure of standard comparison principles.

• Transformation: They define a transformed function $w_\kappa = (-\log(\kappa u))^{1/2}$. The goal is to prove that $w_\kappa$ is convex (which is equivalent to $u$ being $(1/2)$-logconcave).
• The Equation: The function $w_\kappa$ satisfies a specific nonlinear elliptic PDE:
  $$-\Delta w + \frac{1}{w} \left[ (2w^2 - 1)|\nabla w|^2 + \frac{\lambda_1(\Omega)}{2} \right] = 0$$
  This equation does not satisfy the structural conditions required for classical concavity maximum principles (e.g., Korevaar's method) or standard comparison principles.
• The Convex Envelope Approach: The authors analyze the convex envelope $w_\kappa^{**}$ of $w_\kappa$. They aim to show that $w_\kappa^{**}$ is a viscosity supersolution to the PDE satisfied by $w_\kappa$. If $w_\kappa^{**}$ is a supersolution and $w_\kappa$ is a solution, and given the boundary behavior, one can deduce $w_\kappa = w_\kappa^{**}$, implying convexity.
• Key Innovation: The standard convex envelope method fails here because the nonlinearity prevents a direct comparison. The authors circumvent this by:
  1. Utilizing the Andrews-Clutterbuck inequality (an improved log-concavity estimate for the ground state) to derive a lower bound on the gradient of the convex envelope at points where it differs from the function.
  2. Using this gradient estimate to prove that the convex envelope is indeed a supersolution, provided the scaling parameter $\kappa$ is below a specific threshold.
  3. Employing a variational argument based on the simplicity of the first eigenvalue to conclude that the convex envelope must coincide with the function itself.

3. Key Contributions and Results

Theorem 1: Global $(1/2)$-Logconcavity

The main result establishes that the ground state is $(1/2)$-logconcave, provided it is scaled by a specific factor $\kappa(\Omega)$.

• Statement: Let $u$ be the first Dirichlet eigenfunction normalized such that $\max_\Omega u = 1$. Then, for any $\kappa \in (0, \kappa(\Omega)]$, the function $L_{1/2}(\kappa u) = -(-\log(\kappa u))^{1/2}$ is concave in $\Omega$.
• The Threshold $\kappa(\Omega)$:
  $$\kappa(\Omega) := \exp\left[ -\frac{3}{2} \left( \frac{\lambda_1(\Omega) D_\Omega^2}{\pi^2} - 1 \right) \right]$$
  where $D_\Omega$ is the diameter of $\Omega$ and $\lambda_1(\Omega)$ is the first eigenvalue.
• Properties of $\kappa(\Omega)$:
  • It is maximized for balls (where it is approximately $0.133$ for $n=2$).
  • It degenerates to $0$ for "thinning" convex domains.
  • For specific domains like balls or parallelepipeds, the threshold can be improved to $\kappa=1$ (i.e., $(-\log u)^{1/2}$ is convex), but the general bound depends on the geometry.

Theorem 3: Local $(1/2)$-Logconcavity

The authors prove that even if the global threshold $\kappa(\Omega)$ is exceeded, the $(1/2)$-logconcavity holds locally near the maximum of the eigenfunction.

• Statement: For any $\kappa \in (0, 1)$, there exists a level set $\Omega_\kappa = \{x \in \Omega : u(x) > u_\kappa\}$ (where $u_\kappa \to 1$ as $\kappa \to 1^-$) such that $w_\kappa$ is convex on $\Omega_\kappa$.
• Significance: This confirms that the strongest form of concavity holds in a neighborhood of the peak of the eigenfunction, regardless of the domain's aspect ratio.

4. Technical Highlights of the Proof

• Gradient Estimate (Proposition 7): A critical step involves showing that if the convex envelope $w_\kappa^{**}$ is strictly less than $w_\kappa$ at a point $x$, the gradient of $w_\kappa$ at the supporting points must be large:
  $$|\nabla w_\kappa|^2 \geq \frac{\pi^2}{2 D_\Omega^2}$$
  This is derived using the Andrews-Clutterbuck inequality (tangent function modulus of concavity).
• Supersolution Property (Proposition 8): The authors show that under the condition $\kappa \leq \kappa(\Omega)$, the convex envelope satisfies the PDE inequality required to be a viscosity supersolution. This relies on the convexity of a specific auxiliary function $\phi_a(s) = s/(s^2+a)$ and the gradient lower bound.
• Variational Argument (Proof of Theorem 1): Since $w_\kappa^{**}$ is a supersolution, the corresponding function $u_\kappa$ (derived from the envelope) satisfies $-\Delta u_\kappa \leq \lambda_1 u_\kappa$. By the variational characterization of the first eigenvalue and the simplicity of the first eigenfunction, $u_\kappa$ must be a multiple of $u$. Since both are normalized to 1, $u_\kappa = u$, implying $w_\kappa = w_\kappa^{**}$.

5. Significance

• Resolution of an Open Problem: The paper definitively answers the question raised by Ishige, Salani, and Takatsu regarding the $(1/2)$-logconcavity of the Laplacian ground state.
• Optimality of Concavity: It confirms that $\alpha = 1/2$ is the critical threshold for power-logconcavity in this context, aligning with the properties of the heat kernel.
• Methodological Advancement: The paper introduces a robust technique for handling concavity properties in PDEs where standard comparison principles fail. By combining the convex envelope method with sharp gradient estimates derived from improved log-concavity inequalities, the authors provide a template for analyzing other nonlinear eigenvalue problems.
• Geometric Implications: The results deepen the understanding of how the geometry of the domain (specifically the ratio of the eigenvalue to the diameter squared) influences the shape and concentration of the ground state.

In summary, Crasta and Fragalà prove that the ground state of the Dirichlet Laplacian on a convex domain is not just log-concave, but possesses a stronger, "square-root" log-concavity property, provided the function is appropriately scaled. This result bridges the gap between the properties of the heat flow and the stationary eigenfunction.