Ricci Flow and Geometrisation

Tier: R (Research). Builds on Riemannian geometry, curvature and geodesics. The Ricci flow is a heat equation for curvature, and in twenty years it went from Hamilton’s proposal to a proof of the Poincare conjecture a century after it was posed. This page covers the mechanism, the singularity problem that makes it hard, and the research questions still open.

Recall prompts

Attempt these from memory before reading.

Define the Ricci curvature tensor and the scalar curvature. How do they relate to the Riemann tensor?
State the heat equation. What does the heat equation do to a temperature distribution?
State the uniformisation theorem for surfaces.
What is a 3-manifold? What does “simply connected” mean?

Motivation: a heat equation for geometry

The uniformisation theorem (1907) says every closed surface admits a metric of constant Gaussian curvature: positive on the sphere, zero on the torus, negative everywhere else. Every surface is geometrically one of three types.

The natural question for a century: is the same true in three dimensions? The answer is no — 3-manifolds admit eight model geometries, not three — but the stronger and more famous question is whether every simply connected closed 3-manifold is the 3-sphere. That is the Poincare conjecture, posed 1904, open for 99 years.

Hamilton’s idea (1982): the uniformisation theorem is proved by a heat flow on metrics, so run a heat flow on metrics in three dimensions and see where it converges. The flow is

\frac{\partial}{\partial t} g(t) = -2 \operatorname{Ric}(g(t)),

where $\operatorname{Ric}$ is the Ricci tensor. Positive curvature decreases under the flow, exactly as heat diffuses from hot to cold.

Why it is not just a heat equation. The heat equation is parabolic because the second derivative in space is elliptic. The Ricci flow is only weakly parabolic: the operator has a kernel caused by the diffeomorphism invariance of the equation — pulling back by a diffeomorphism gives the same metric geometrically but a different point in the space of metrics. DeTurck’s trick fixes the gauge and makes the flow parabolic, which is what gives short-time existence and uniqueness.

The mechanism

In words. Curvature measures how volumes deviate from Euclidean. Positive Ricci curvature makes volumes grow more slowly than in Euclidean space, so the flow shrinks them. The flow is designed so that the shrinking is fastest where the curvature is largest, which is what lets it smooth out irregularities.

As equations. In local coordinates,

\frac{\partial g_{ij}}{\partial t} = -2 R_{ij},

and the scalar curvature satisfies

\frac{\partial R}{\partial t} = \Delta R + 2|\operatorname{Ric}|^2,

where $\Delta$ is the Laplace–Beltrami operator. That is a genuine heat equation with a reaction term $2|\operatorname{Ric}|^2 \ge 0$ — which is why the flow focuses rather than merely diffuses.

As a picture.

   a lumpy metric        under Ricci flow        a round metric
   ..............        ..................      ............
  .   /\   ___   .      .   ....   ____   .     .            .
 .   /  \ /   \  .  →   .  ....   ....   .  →   .     ....    .
 .   \__/ \___/  .      .   ....   ___   .      .            .
   ............          ..................      ............
  high curvature         curvature evening out   constant curvature

Hamilton’s theorem (A, 1982). If a closed 3-manifold admits a metric with positive Ricci curvature, then it admits a metric of constant positive sectional curvature; i.e. it is a quotient of the 3-sphere.

That is the 3-dimensional analogue of the positive-curvature case of uniformisation, and it was the first sign the flow could work.

Proofs

Perelman’s full argument is roughly a thousand pages and cannot be reproduced here. Two results can, and they are not arbitrary choices: the first is the opening step of Hamilton’s theorem, and the second is the computation that made Perelman’s contribution possible. Both are self-contained.

Theorem (maximum principle for the scalar curvature). Let $g(t)$ flow by Ricci flow on a closed manifold. If $R(\cdot, 0) \geq c > 0$ everywhere, then $R(\cdot, t) \geq c$ for all $t$ . More generally, if $R(\cdot, 0) \geq 0$ then $R(\cdot, t) \geq 0$ for all $t$ .

Proof. The evolution equation is $\partial_t R = \Delta R + 2|\operatorname{Ric}|^2$ , a heat equation with a non-negative reaction term, so $R$ is a supersolution of the heat equation: $\partial_t R - \Delta R \geq 0$ .

Handle the constant case first, by a damping trick. Fix $\varepsilon > 0$ and put $v = R - c - \varepsilon t$ . Then $\partial_t v - \Delta v \geq \varepsilon > 0$ . Suppose $v$ becomes negative somewhere, and let $(x_0, t_0)$ be the first space-time point where it does, in the sense that $v \geq 0$ on $M \times [0, t_0)$ and $v(x_0, t_0) < 0$ at the earliest such time. At $(x_0, t_0)$ the value is falling through zero, so $\partial_t v \leq 0$ , and it is a spatial minimum, so $\Delta v \geq 0$ ; hence

\partial_t v - \Delta v \leq 0,

contradicting $\partial_t v - \Delta v \geq \varepsilon$ . Letting $\varepsilon \to 0$ gives $R \geq c$ ; and for the weaker hypothesis take $c = 0$ . $\blacksquare$

Two hypotheses are doing work, and neither is decoration. Closedness of $M$ is what makes the spatial minimum exist and what kills the boundary term behind $\Delta v \geq 0$ at a minimum. And the reaction term $2|\operatorname{Ric}|^2 \geq 0$ is what makes the principle useful rather than merely true: for a pure heat equation the maximum of a solution decays, so positive curvature would diffuse away, while here the reaction term pushes $R$ upward, so positive scalar curvature is preserved and amplified. That amplification is what lets Hamilton conclude, and it is the reason the flow was designed with $-2\operatorname{Ric}$ rather than $+2\operatorname{Ric}$ .

The rigidity clause — that a metric with $R \geq 0$ everywhere and $R = 0$ at one point splits off a factor, forcing constant curvature — is where this argument earns its “strong” prefix. It requires the equality case of the maximum principle, which needs more machinery than the strict version and is not proved here; saying so is part of saying what has been proved.

Theorem ( $\mathcal{F}$ -monotonicity, Perelman 2002). Let

\mathcal{F}(g, f) = \int_M \left(R + |\nabla f|^2\right) e^{-f}\, dV,

and flow the pair $(g, f)$ by $\partial_t g = -2\operatorname{Ric}$ , $\partial_t f = -\Delta f - R$ . Then

\frac{d}{dt}\mathcal{F} = 2\int_M \left|\operatorname{Ric} + \nabla^2 f\right|^2 e^{-f}\, dV \geq 0,

with equality iff the metric is a gradient shrinking soliton.

Proof. Three variation formulas, each obtained by differentiating the standard identities along the flow:

\partial_t R = \Delta R + 2|\operatorname{Ric}|^2, \qquad \partial_t(dV) = -R\,dV, \qquad \partial_t|\nabla f|^2 = 2\langle\operatorname{Ric}(\nabla f), \nabla f\rangle + 2\langle\nabla f, \nabla(\partial_t f)\rangle.

The first is the scalar curvature evolution already stated. The second is the first variation of volume for the variation field $-\operatorname{Ric}$ . The third is the compatibility of the Levi-Civita connection with the flow. Differentiating under the integral and substituting $\partial_t f = -\Delta f - R$ :

\frac{d}{dt}\mathcal{F} = \int\Big(\partial_t R + \partial_t|\nabla f|^2\Big)e^{-f}dV + \int\big(R + |\nabla f|^2\big)e^{-f}\big(-\partial_t f - R\big)dV

= \int\Big(\Delta R + 2|\operatorname{Ric}|^2\Big)e^{-f}dV + 2\int\langle\operatorname{Ric}\nabla f, \nabla f\rangle e^{-f}dV + 2\int\langle\nabla f, \nabla(\Delta f + R)\rangle e^{-f}dV + \int\big(R + |\nabla f|^2\big)\Delta f\; e^{-f}dV,

the $-R$ from $\partial_t dV$ cancelling the $-R$ from $-\partial_t f$ . Three integrations by parts. For the $\Delta R$ term, move the Laplacian onto $e^{-f}$ : $\int \Delta R\, e^{-f} = \int R\,\Delta e^{-f} = \int R(|\nabla f|^2 - \Delta f)e^{-f}$ . For the gradient term, move $\nabla$ onto $e^{-f}$ the same way. Collecting, all the $R\Delta f$ and $|\nabla f|^2$ terms cancel in pairs, leaving

\frac{d}{dt}\mathcal{F} = \int\Big(2|\operatorname{Ric}|^2 + 2\langle\operatorname{Ric}, \nabla^2 f\rangle + 2|\nabla^2 f|^2\Big)e^{-f}dV = 2\int|\operatorname{Ric} + \nabla^2 f|^2\, e^{-f}dV \geq 0.

Equality forces $\operatorname{Ric} + \nabla^2 f$ to vanish as a $(0,2)$ -tensor, which under the normalisation $\int e^{-f}dV = (4\pi t)^{3/2}$ is the gradient shrinking soliton equation. $\blacksquare$

Why this is the pivot of the subject. A flow can say nothing about its singularities without a monotone quantity: monotonicity is a coordinate-free measurement that can only increase, so it compares the geometry before a singularity with the geometry after and rules out most ways of degenerating. Perelman’s $\mathcal{W}$ functional adds a scale parameter to $\mathcal{F}$ and uses its monotonicity to prove no local collapsing — that a region of small volume cannot persist with bounded curvature. That is the step separating the geometrisation theorem from everything before it, and the computation above is its engine, written out.

Singularity formation: why the problem is hard

The flow develops singularities in finite time. On a sphere, the metric shrinks to a point in finite time $T$ , with curvature blowing up like $1/(T-t)$ : and at $T$ you can rescale and the manifold is the round sphere, so the flow has done its job.

On a general 3-manifold, singularities can be worse: necks can pinch, and the topology can change. Two strategies:

Surgery (Hamilton–Perelman). Cut out singular regions before they form, cap the holes, restart the flow. If the singular regions are precisely the pieces whose topology you understand, the flow is performing a topological decomposition of the manifold.
Ricci flow with bubbles. Study the rescaled limits at singularities (blow-up analysis) and classify them: the κ-solutions and ancient solutions. This is Perelman’s route.

Perelman’s contribution (R, 2002–03). Three preprints on the arXiv. The key insight is a monotone quantity; Perelman’s entropy

\mathcal{W}(g, f) = \int \left( R|\nabla f|^2 + \frac{f - 2n}{2(T-t)} \right) (4\pi(T-t))^{-n/2} e^{-f}\, dV

which is non-decreasing under the coupled flow of the metric and the function $f$ . Monotonicity gives non-collapsing (a lower volume bound at singularities), which is what makes the blow-up limits non-trivial and classifiable.

From there: singular regions are either caps or round necks, surgery is well-defined, and after finitely many surgeries the manifold is a union of pieces each admitting a geometric structure. That proves the Geometrisation Conjecture (Thurston), of which the Poincare conjecture is the simply connected case: a simply connected manifold with a geometric structure and no boundary must be the 3-sphere.

Why the proof matters beyond the conjecture

Tool	What it gives	Where else it is used
Perelman’s reduced volume	non-collapsing at singularities	convergence of Ricci flow, Ricci solitons
κ-solutions	classification of blow-up limits	singularity analysis in any dimension
Surgery	topology change under flow	the proof structure for geometrisation
Pseudolocality	small-scale almost-Euclidean behaviour	regularity of the flow

The pseudolocality theorem is the one with the widest application: it says that even if the initial metric is very irregular, the flow is almost Euclidean on small scales for a short time. That is the parabolic analogue of the local regularity theory for the heat equation, and it is used everywhere Ricci flow appears.

Counterexamples

Claim	True?	Counterexample
Ricci flow exists for all time	no	the round sphere shrinks to a point in finite time
Ricci flow preserves isometry class	no	it deforms the metric, only the diffeomorphism type is preserved
Ricci solitons are static	no	they shrink or expand by a diffeomorphism; the Gaussian soliton expands
Every 3-manifold admits positive Ricci curvature	no	Thurston’s geometrisation gives eight geometries, most with non-positive curvature
Ricci flow smooths any metric in any dimension	no	higher dimensions admit worse singularities; the flow can develop non-spherical singularities

Where the research frontier is

Differentiable sphere theorem. Does Ricci flow distinguish smooth structures on the sphere? In dimension 4 and above there are exotic spheres, and whether the flow can detect them is open.
Ricci solitons. Fixed points of the flow up to scaling and diffeomorphism. Classifying them is ongoing, and the noncompact case is wide open. The Bryant soliton is the standard example of a steady soliton.
Ricci flow in higher dimensions. Böhm constructed examples in dimensions 5 and higher where the flow develops singularities of non-spherical type, and the surgery theory is dimension-dependent.
Kähler–Ricci flow. The same flow restricted to Kähler manifolds, where it simplifies dramatically. The Yau–Tian–Donaldson conjecture relates the flow’s long-time existence to algebro-geometric stability, and was proved in important cases by Chen–Donaldson–Sun.

More counterexamples

Object	Looks like	Is actually
Ricci flow	a pure smoothing equation	weakly parabolic; the diffeomorphism kernel must be fixed
Singularities	always a shrinking sphere	necks pinch; the topology can change
Perelman’s entropy	a technical tool	the monotone quantity that makes the whole proof work
Geometrisation	a 3-dimensional curiosity	the model for how geometric flows decompose manifolds

Recall prompts: attempt after reading

State the Ricci flow equation. What does it do to positive Ricci curvature?
Why is the flow only weakly parabolic, and what fixes it?
What is Perelman’s entropy, and what does its monotonicity give?
State Hamilton’s theorem for positive Ricci curvature in dimension 3.
What does geometrisation assert, and how does Poincare follow?

Interleaving problems

Mix these with material from Curvature, Geodesics, Measure Theory, Functional Analysis and Stochastic Calculus.

(Curvature.) Show that on a surface, $\operatorname{Ric} = K g$ where $K$ is the Gaussian curvature, so the Ricci flow on surfaces is $\partial_t g = -2Kg$ .
(Geodesics.) Show that the round sphere’s metric shrinks under Ricci flow by a factor depending on time, and compute the finite extinction time.
(Measure theory.) Show that Perelman’s entropy involves an integral against the measure $(4\pi(T-t))^{-n/2} e^{-f} dV$ , and identify this as a conjugation of the heat kernel measure.
(Functional analysis.) Show the linearisation of the Ricci flow at the round sphere has a kernel consisting of infinitesimal diffeomorphisms, which is the origin of the weak parabolicity.
(Stochastic analysis.) The stochastic Ricci flow adds a noise term to the equation. Explain the regularity problem the noise introduces and why it is harder than for the stochastic heat equation.

Ricci Flow and Geometrisation