Skip to content

The Heat Equation – Intuition

"The Ricci flow is – roughly – the heat equation for the Riemannian metric. Whoever sees heat spreading sees curvature being evened out."

The heat equation is the simplest parabolic partial differential equation. Its qualitative properties – smoothing, maximum principle, energy decay – are the didactic blueprint for everything that happens in the Ricci flow.

1. The equation

On \(\mathbb{R}^n\) or a Riemannian manifold \((M, g)\): $$ \partial_t u = \Delta u, \qquad u(0, x) = u_0(x). $$ Here \(u(t, x)\) is, e.g., the temperature at point \(x\) at time \(t\), and \(\Delta\) is the Laplacian (see Vector Calculus in a Nutshell).

The equation says: the time derivative of \(u\) equals the Laplacian of \(u\) – i.e. the deviation of \(u\) from its local average. If \(u\) is higher in the surroundings than at the point, \(u\) grows; if it is lower, \(u\) falls.

2. Three basic properties

Smoothing. Even if \(u_0\) is merely continuous (or \(L^2\)), \(u(t, \cdot)\) is infinitely differentiable for \(t > 0\). The heat equation creates regularity from nothing.

Maximum principle. On a compact domain without sources, $$ \min_x u_0 \le u(t, x) \le \max_x u_0 \qquad \forall t > 0. $$ Intuitively: hot does not get hotter, cold does not get colder, unless heat enters from outside.

Energy decay. On a compact \(M\) without boundary: $$ \frac{\mathrm{d}}{\mathrm{d}t} \int_M u^2\, \mathrm{d}V = 2\int_M u\, \Delta u\, \mathrm{d}V = -2\int_M |\nabla u|^2\, \mathrm{d}V \le 0. $$ The \(L^2\)-norm decreases monotonically; the solution "spreads out".

3. Heat kernel on \(\mathbb{R}^n\)

The fundamental solution with initial datum \(\delta_y\) is the heat kernel $$ K(t, x, y) = (4\pi t)^{-n/2}\, \exp!\Big(-\frac{|x - y|^2}{4t}\Big). $$ Every solution with sufficiently nice \(u_0\) can be written as $$ u(t, x) = \int_{\mathbb{R}^n} K(t, x, y)\, u_0(y)\, \mathrm{d}y. $$ \(K\) is a Gaussian whose standard deviation grows like \(\sqrt{t}\) – the characteristic parabolic scaling \(x \sim \sqrt{t}\), which also appears in Ricci-flow blow-up (Act 2, Article 04).

4. On a manifold

On \((M, g)\) one replaces \(\Delta\) by the Laplace–Beltrami operator \(\Delta_g\). Smoothing and the maximum principle remain valid. The heat kernel exists too and encodes geometry: on \(S^n\) and on hyperbolic spaces \(K_M(t, x, y)\) is known explicitly; in general its asymptotics yield the heat-kernel expansion with curvature invariants as coefficients – the bridge between spectral and differential geometry.

5. Why "Ricci flow = heat equation for the metric"

Linearising the Ricci-flow equation \(\partial_t g = -2\mathrm{Ric}\) around a flat solution in a suitable gauge (e.g. via DeTurck's trick) yields $$ \partial_t h_{ij} \approx \Delta_g h_{ij} + \text{curvature terms}. $$ Up to gauge corrections the Ricci flow is a heat equation for the metric tensor. Its qualitative properties – smoothing, maximum principle on scalar curvature (Act 3, Article 02), energy monotonicity of the functionals \(\mathcal{F}\), \(\mathcal{W}\) (Act 2, Article 05) – are direct relatives of the three heat-equation properties listed above.

6. Conjugate heat equation

For a time-dependent metric \(g(t)\) following the Ricci flow, the conjugate heat equation $$ \partial_\tau u = -\Delta_g u + R\, u, \qquad \tau = T - t, $$ is the natural counterpart. It appears in the definition of Perelman's entropy and the reduced length (Act 2, Article 07).

Cross-references

Sources

  • Evans, Lawrence C. (2010). Partial Differential Equations. AMS GSM 19, 2nd ed. Ch. 2.3.
  • John, Fritz (1991). Partial Differential Equations. Springer, 4th ed.
  • Grigor'yan, Alexander (2009). Heat Kernel and Analysis on Manifolds. AMS/IP.
  • Topping, Peter (2006). Lectures on the Ricci Flow. Cambridge University Press, Ch. 1.