the spectral gap

A diffusion left alone settles toward a steady state — but how fast? Intuitively, the bumpiest, most wiggly parts of a profile flatten out first, while the gentlest large-scale imbalance lingers longest. The slowest-decaying ingredient sets the overall rate of return to equilibrium, and that slowest mode is governed by the smallest nonzero eigenvalue of the spatial operator. The spectral gap is the size of that smallest eigenvalue (or the distance from the slowest mode to the rest of the spectrum); it is the number that quantifies how quickly the system forgets its initial state.

Concretely, expand the solution in eigenfunctions of the elliptic operator. If A phi_k = lambda_k phi_k with 0 < lambda_1 <= lambda_2 <= ... (for the heat IBVP these are the positive Dirichlet eigenvalues), then writing u_0 = sum c_k phi_k gives u(t) = e^{-tA} u_0 = sum c_k e^{-lambda_k t} phi_k. Each mode decays at its own rate e^{-lambda_k t}, and the slowest of all is the first, e^{-lambda_1 t}. The spectral gap is lambda_1 (the gap above zero), and it yields the exponential decay estimate norm of u(t) <= e^{-lambda_1 t} norm of u_0: the solution converges to zero (or to its steady state, after subtracting it) at the guaranteed exponential rate set by the gap. When a nonzero equilibrium exists, you measure the gap of the linearization around it.

This is the engine of long-time behaviour. A positive spectral gap means exponential convergence to the steady state, which is precisely what makes a steady state a stable attractor. The gap appears under different names across the subject — it is the Poincare constant (the best constant in the Poincare inequality equals lambda_1), the Rayleigh-quotient minimum, the principal eigenvalue — and a strictly positive value is what separates 'returns to equilibrium fast' from 'drifts forever'. On unbounded domains the spectrum can reach down to zero (no gap), and then decay is only algebraic, not exponential — a genuinely different and slower regime.