🔢 mathematics

A Heat Kernel Expectation Approach to Boundary-Corrected Li--Yau Estimates for the Dirichlet Heat Equation

This paper establishes an explicit boundary-corrected Li--Yau gradient estimate for positive solutions of the Dirichlet heat equation on the Euclidean half-space by utilizing the reflection structure of the heat kernel to derive a hyperbolic correction term and prove a refined lower bound involving an inverse-square distance term.

Original authors: Li-Chang Hung

Published 2026-08-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Li-Chang Hung

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, invisible fog that drifts and spreads out over time. This isn't just any fog; it's the "heat equation," a mathematical rule that describes how heat, or even the spread of a drop of ink in water, moves through space. Scientists have been studying this for decades because it helps them understand everything from how stars cool down to how information travels through a network. A famous rule in this field, called the Li–Yau inequality, acts like a speed limit sign for this diffusion. It tells us exactly how fast the "fog" can change its shape and how steep its edges can get as it spreads through empty space. But what happens when that fog hits a wall? In the real world, things don't just float forever; they hit boundaries, like a cup of coffee hitting the side of a mug or a scent hitting a closed door. When the heat equation meets a wall that absorbs everything (a "Dirichlet boundary"), the old rules change. The fog doesn't just bounce off; it gets squished and distorted in a very specific, tricky way. Understanding exactly how this distortion works is crucial for anyone trying to predict how things move in confined spaces, from microchips to biological cells.

This paper, written by Li-Chang Hung, tackles that exact problem: what happens to the "speed limit" of heat diffusion when it's trapped in a half-space, like a room with only one wall? The author discovers a new, precise formula that corrects the old rules to account for that wall. Instead of just guessing or using complicated, abstract logic, Hung uses a clever trick involving "expectations" (a fancy word for averages) and a special kind of mirror.

Here's the core idea: When heat hits a wall that absorbs it, mathematicians often imagine a "ghost" version of the heat source on the other side of the wall to calculate the result. This is called the "reflection method." Hung realized that this ghost isn't just a mathematical trick; it creates a hidden, hyperbolic structure in the math, similar to how a mirror creates a specific kind of optical illusion. By treating the heat distribution as a probability game—where the heat kernel acts like a weighted die—the author shows that the wall introduces a new force. This force is an "inverse-square correction," which means the closer you get to the wall, the stronger the effect becomes, dropping off sharply as you move away.

The paper proves that for any positive solution to this heat equation, the change in the logarithm of the heat (a way of measuring its steepness) is always greater than or equal to a specific value. This value has two parts: the first part is the classic, universal rule for how heat spreads in empty space ( $-\frac{n}{2t}$ ), and the second part is the new, wall-specific correction ( $-\frac{1}{x_n^2}$ ). Here, $x_n$ is simply the distance from the point to the wall. The paper shows that this correction term is not just a guess; it is a hard, proven fact derived directly from the structure of the heat kernel itself. The author uses a mathematical tool called Jensen's inequality (which, in simple terms, says that the average of a curve is always higher than the curve of the average) to prove that this wall effect is unavoidable and follows a strict pattern.

Interestingly, this new formula connects to older work by a mathematician named Hamilton, who had found a similar rule for a one-dimensional line (a single straight path) using a very different, more complex method. Hung's paper shows that Hamilton's result is actually just a special case of this broader, more general rule. The paper also explores the "paths" the heat takes. It turns out that the wall acts like a repulsive force, pushing the heat away. If you were to trace the "best" path the heat could take to get from point A to point B while respecting the wall, it wouldn't be a straight line; it would curve away from the wall, as if the wall were a magnet pushing it back. This curvature is described by a specific equation that looks like a particle moving under the influence of a repulsive force that gets infinitely strong as you get closer to the wall.

The paper is careful to note that while this formula is proven for a flat, half-space (like an infinite room with one floor), it suggests that similar rules might apply to more complex shapes, like curved surfaces or irregular rooms, where the "distance to the wall" would simply be replaced by the shortest distance to the boundary. The author doesn't claim to have solved the problem for every possible shape, but the method provides a clear, new way of looking at how boundaries shape diffusion. By turning a difficult calculus problem into a story about averages and probabilities, the paper reveals that the "ghost" of the reflected heat isn't just a calculation tool—it's the physical reason why the math changes near a wall. The result is a sharper, more accurate map for understanding how things spread when they are confined, separating the universal laws of diffusion from the specific, local influence of the boundaries that hold them in.

Technical Summary: A Heat Kernel Expectation Approach to Boundary-Corrected Li–Yau Estimates

Problem Statement
The paper addresses the problem of establishing gradient estimates for positive solutions of the Dirichlet heat equation on the Euclidean half-space $\mathbb{R}^n_+ = \{x = (x_1, \dots, x_n) \in \mathbb{R}^n : x_n > 0\}$ . While the classical Li–Yau inequality provides sharp control over the space-time behavior of diffusion on complete Riemannian manifolds without boundary, the presence of an absorbing boundary fundamentally alters the heat kernel structure. Previous work by Hamilton established that the Dirichlet boundary introduces an inverse-square correction term, but the analytic origin of this correction within the heat kernel structure was not directly visible, as Hamilton's proof relied on a nonlinear maximum principle argument. This paper seeks to provide a direct kernel-based interpretation of this boundary correction.

Methodology
The author employs a "heat kernel expectation approach" to derive the estimates. The core methodology involves three steps:

Kernel Representation: The solution $w(x,t)$ is represented using the Dirichlet heat kernel $K_D(x, y, t)$ , which is constructed via the method of reflection:
$K_D(x, y, t) = \frac{1}{(4\pi t)^{n/2}} \left( e^{-\frac{|x'-y'|^2 + (x_n-y_n)^2}{4t}} - e^{-\frac{|x'-y'|^2 + (x_n+y_n)^2}{4t}} \right).$
Normalized Probability Measure: A normalized kernel-induced probability measure is introduced for fixed $(x, t)$ :
$d\mu_{x,t}(y) = \frac{K_D(x, y, t)g(y)}{w(x, t)} dy,$
where $g(y)$ is the non-negative initial datum. This transformation allows logarithmic derivatives of the solution to be expressed as expectations of explicit kernel quantities.
Hyperbolic Analysis and Jensen's Inequality: The logarithmic derivatives of the kernel are computed explicitly. The reflected Gaussian term generates a hyperbolic structure involving $\coth\left(\frac{x_n y_n}{2t}\right)$ . By applying Jensen's inequality and utilizing the sharp elementary estimate $0 < z \operatorname{csch} z < 1$ for $z > 0$ , the author bounds the resulting expectations to derive the gradient estimate.

Key Results
The primary result is an explicit boundary-corrected Li–Yau type gradient estimate. For any positive solution $w$ of the Dirichlet heat equation on $\mathbb{R}^n_+$ , the following inequality holds:
$\Delta \log w(x, t) \geq -\frac{n}{2t} - \frac{1}{x_n^2}.$
Equivalently, in terms of the Laplacian and gradient:
$\frac{\Delta w}{w} - \frac{|\nabla w|^2}{w^2} \geq -\frac{n}{2t} - \frac{1}{x_n^2}.$

The paper further derives the corresponding integrated Harnack inequality. For a smooth curve $\gamma(t) = (x(t), t)$ in the half-space:
$\log \frac{w(x(t_2), t_2)}{w(x(t_1), t_1)} \geq -\frac{1}{4} \int_{t_1}^{t_2} |x'(t)|^2 dt - \frac{n}{2} \log \frac{t_2}{t_1} - \int_{t_1}^{t_2} \frac{1}{x_n(t)^2} dt.$

Significance and Contributions
The paper claims several specific contributions to the understanding of boundary effects in diffusion:

Direct Kernel Interpretation: Unlike Hamilton's approach, which used a nonlinear maximum principle, this work demonstrates that the boundary correction arises intrinsically from the reflected Gaussian structure of the Dirichlet heat kernel. The correction term is shown to be a direct consequence of the hyperbolic identity $1 - \coth^2 z = -\operatorname{csch}^2 z$ .
Geometric Separation: The estimate explicitly separates the universal diffusion scaling ( $-\frac{n}{2t}$ ) from the boundary contribution ( $-\frac{1}{x_n^2}$ ). The boundary term is identified geometrically as $-\frac{1}{d(x, \partial \mathbb{R}^n_+)^2}$ , indicating that the influence of the absorbing boundary decays quadratically with distance.
Recovery of Hamilton's Estimate: The paper shows that in the one-dimensional case ( $n=1$ ), the derived inequality reduces to the simplified Li–Yau estimate obtained by Hamilton ( $l_{xx} + \frac{1}{2t} + \frac{1}{x^2} \geq 0$ ). Furthermore, it provides a probabilistic interpretation of Hamilton's auxiliary variable $v$ as a normalized expectation of the boundary interaction term.
Sharpness: The paper argues that the inverse-square correction is of optimal order. Near the boundary, the logarithmic Hessian of the heat kernel necessarily exhibits an inverse-square singularity due to the vanishing Dirichlet condition, implying the correction term cannot be removed or generally improved in scaling.
Variational Interpretation: The results suggest that the boundary correction modifies the geometry of optimal diffusion paths. The Harnack inequality leads to a variational problem where the optimal path minimizes an action functional containing a repulsive potential $V(x_n) = \frac{1}{x_n^2}$ , preventing paths from crossing the boundary.

The paper concludes by suggesting that this kernel-based approach offers a framework for extending boundary-corrected estimates to more general domains and manifolds with boundary, where the Euclidean distance function would replace $x_n$ .

More like this