Hamilton-Jacobi Equations

Core Idea

Hamilton-Jacobi (HJ) equations are first-order nonlinear PDEs of the form u_t + H(x, ∇u) = 0, where H is the Hamiltonian. They arise in classical mechanics (Hamilton-Jacobi theory connects the action principle to wave-like evolution), optimal control (the value function satisfies an HJ equation), and geometric optics (wavefront propagation). Classical solutions typically break down in finite time as characteristics cross, and viscosity solutions provide the correct framework for global-in-time solutions. The Hopf-Lax formula gives an explicit representation for convex Hamiltonians.

Explainer

Hamilton-Jacobi equations sit at the intersection of classical mechanics, optimal control, and PDE theory. In mechanics, the Hamilton-Jacobi equation u_t + H(x, ∇u) = 0 describes the evolution of the action function, and its characteristics are the trajectories of Hamilton's equations (the equations of motion). Solving the HJ equation is equivalent to finding all trajectories simultaneously—a powerful reformulation that transforms particle mechanics into wave mechanics and lies at the historical foundation of quantum mechanics.

In optimal control, the value function V(x,t) measuring the minimum cost to reach a target from state x at time t satisfies the Hamilton-Jacobi-Bellman (HJB) equation. The Hamiltonian H(x,p) encodes the optimization: H(x,p) = min_a{f(x,a)·p + L(x,a)}, where f is the dynamics and L is the running cost. The optimal control is recovered from ∇V via the minimizing argument in H. This connection, formalized by Bellman's dynamic programming principle, makes HJ equations central to robotics, economics, and engineering.

Classical solutions of HJ equations break down because characteristics cross. For the equation u_t + ½|∇u|² = 0 with initial data u₀(x) = -|x|, the characteristics emanating from the origin fan out while those from far away converge, and the solution develops a corner (non-differentiable point) in finite time. The viscosity solution framework resolves this: it selects the physically correct solution by requiring that the equation is satisfied in an appropriate limiting sense. The Hopf-Lax formula u(x,t) = min_y{u₀(y) + tL((x-y)/t)} provides an explicit representation for convex Hamiltonians.

The eikonal equation |∇u| = f(x), the stationary HJ equation, describes wavefront propagation in geometric optics. The solution is the arrival time of a wavefront moving with speed 1/f(x). Its viscosity solution can be computed efficiently by the Fast Marching Method (O(N log N)), making it practical for applications in computational geometry, image segmentation, and path planning. The theory of HJ equations continues to develop: weak KAM theory (Fathi) connects HJ equations to dynamical systems and ergodic theory, while mean-field games couple HJ equations with Fokker-Planck equations to model large populations of interacting rational agents.