the smoothing property

Start the heat equation with the ugliest temperature profile you can draw — full of jagged corners, jumps, even a profile that is not differentiable anywhere. Wait any positive instant. The temperature is now perfectly smooth: infinitely differentiable, all corners gone, all jumps healed. Diffusion does not merely fail to create roughness — it actively destroys roughness it was handed, and it does so instantly. This is the smoothing property, the single most characteristic behaviour of parabolic equations.

How does it work? The solution of the heat equation is the initial data convolved with the heat kernel, a smooth Gaussian bump that spreads with time. Convolving anything — however rough — with a smooth function produces a smooth result, because the derivatives can be moved onto the smooth kernel. Equivalently, in Fourier language, each mode of frequency xi decays like e^(-k xi^2 t): the higher the frequency, the more violently it is suppressed, and high frequencies are exactly what encode sharp features. After any t > 0 the high-frequency content has been crushed, so the solution is smooth. Elliptic equations smooth in a static way (a harmonic function is automatically analytic inside); parabolic equations smooth in time, turning rough initial data smooth the moment evolution begins.

Two honest consequences. First, smoothing has a price: it is irreversible. You cannot run it backward, because reconstructing the original jagged data from the smoothed-out result would require amplifying high frequencies by e^(+k xi^2 t), which blows up — this is precisely why the backward heat equation is ill-posed. Second, smoothing is tied to infinite propagation speed: the same instantaneous mixing that smooths also spreads influence everywhere at once. Wave-type (hyperbolic) equations do the opposite — they preserve singularities and have finite speed, so a hyperbolic solution is no smoother than its data.