The Heat Equation - Applications

The heat equation and its variants appear throughout mathematics, physics, and probability theory, with profound connections to other areas.

TheoremConnection to Brownian Motion

Let $B_t$ be standard Brownian motion starting at $x$ . Then the solution to the heat equation: $u_t = \frac{1}{2}\nabla^2 u, \quad u(x,0) = f(x)$ can be represented probabilistically as: $u(x,t) = \mathbb{E}[f(x + B_t)]$

This Feynman-Kac formula connects PDEs to stochastic processes, enabling probabilistic methods for solving PDEs and vice versa.

ExampleApplications Across Disciplines

Physics: Heat conduction, diffusion of particles, neutron transport
Finance: Black-Scholes equation for option pricing is a heat equation in disguise
Image processing: Gaussian blur and diffusion filters for noise reduction
Geometry: Ricci flow and mean curvature flow involve parabolic equations
Probability: The transition density of diffusion processes satisfies the heat equation

TheoremAsymptotic Behavior

For the heat equation on $\mathbb{R}^n$ with integrable initial data $\int f(x)\,dx = M$ : $u(x,t) \sim \frac{M}{(4\pi kt)^{n/2}}e^{-|x|^2/(4kt)} \quad \text{as } t \to \infty$

For bounded domains with zero boundary conditions: $u(x,t) \sim b_1\phi_1(x)e^{-k\lambda_1 t} \quad \text{as } t \to \infty$ where $\phi_1, \lambda_1$ are the first eigenfunction and eigenvalue of $-\nabla^2$ .

Remark

The long-time behavior shows universal decay patterns: Gaussian spreading on $\mathbb{R}^n$ and exponential decay to zero on bounded domains. The decay rate $\lambda_1$ (the first Dirichlet eigenvalue) has geometric significance—it measures how "round" the domain is (lower $\lambda_1$ means more elongated domain).

TheoremBackward Heat Equation (Ill-Posed)

The backward heat equation $u_t = -k\nabla^2 u$ with final data $u(x,T) = g(x)$ is ill-posed: solutions are not unique and do not depend continuously on data.

For example, $u_n(x,t) = n^{-2}e^{n^2k(t-T)}\sin(nx)$ satisfies the backward equation with $u_n(x,T) = n^{-2}\sin(nx) \to 0$ as $n \to \infty$ , but $u_n(x,0) = n^{-2}e^{-n^2kT}\sin(nx)$ explodes.

This ill-posedness reflects the irreversibility of diffusion and poses challenges for inverse problems like determining the initial temperature from later measurements. Regularization techniques are needed for practical applications.