Book 11A

© 2026 NeckPinch · www.neckpinch.com · All rights reserved.

Course 11Book 11A: The Ricci Flow: Existence and Maximum PrinciplesChapter 1

The Equation and Its First Solutions

A heat equation for the metric: spheres, cylinders and scaling.

16 min read · Updated Oct 3, 2026

Read with Topping's Lectures on the Ricci Flow, chapter 1 (introduction, the equation, first examples), and Chow and Knopf's The Ricci Flow: An Introduction, chapter 1 (special solutions). Skip both books' introductions to Riemannian geometry: that is Book 9A.

In this chapter · 7 sections
  1. 1.1Where the equation also appears
  2. 1.2A heat equation for the metric
  3. 1.3Symmetries
  4. 1.4Exact solutions
  5. 1.5The normalised flow
  6. 1.6History
  7. 1.7Exercises

This is the equation the whole Path has been preparing:

∂tg=−2Ric⁡(g).\partial_tg = -2\operatorname{Ric}(g).

A Riemannian metric g(t)g(t) moves in the direction of minus twice its Ricci curvature. Richard Hamilton introduced it in 1982 and used it at once to prove a theorem about 3-manifolds (11A.6 Hamilton’s 1982 Theorem). Twenty years later Perelman completed Hamilton's program and proved the Poincaré and geometrization conjectures with it.

This first chapter of Book 11A explains why the equation is a heat equation for the metric, what its symmetries are, and how to solve it exactly when the metric is symmetric enough: Einstein metrics shrink or expand homothetically, products flow factor by factor, and the round cylinder shrinks into a line. These exact solutions are the reference points for everything later. Their curvature evolutions check the formulas of 11A.2 How Curvature Evolves, and their shapes are the models of the singularities of 11B.4 Singularities.

By the end of this chapter you will be able to:

  • explain, through harmonic coordinates, why the Ricci flow is a nonlinear heat equation for the metric, and why the constant is −2-2;
  • use the scaling and diffeomorphism symmetries of the equation;
  • write the exact solutions for Einstein metrics, products, cylinders and flat metrics;
  • define the volume-normalised flow and relate it to the unnormalised one;
  • say where Ricci flow appears in physics and computing, and how those uses differ from the smooth flow.

Where the equation also appears

In the world Model Renormalisation in physics

In quantum field theory, the effective strength of an interaction depends on the length scale at which you probe it, and the renormalisation group describes how the couplings change with scale. For a two-dimensional nonlinear sigma model, a field theory of maps from a surface into a Riemannian manifold, the coupling is the target metric itself. Daniel Friedan showed in 1980 (Physical Review Letters) that, to lowest order in perturbation theory, the renormalisation group flow of the target metric is proportional to its Ricci tensor. So changing the scale at which the theory is viewed moves the target metric by a Ricci flow, to that order. The equation appeared in physics and in Hamilton's geometry at nearly the same time, for different reasons. In string theory, conformal invariance at one loop requires the target metric to be Ricci-flat.

In the world In use Ricci flow in computing

Discrete analogues of the flow are used in computing. Discrete surface Ricci flow on triangle meshes computes conformal parametrisations (5A.5 Uniformization and the Two-Dimensional Ricci Flow, 11A.7 Ricci Flow on Surfaces). On graphs, a discrete Ricci curvature defined from optimal transport, and a Ricci-flow-like reweighting of edges, have been used to detect communities in networks (Chien-Chun Ni, Yu-Yao Lin, Feng Luo and Jie Gao, Scientific Reports, 2019). The same discrete curvature has been used to diagnose "bottlenecks" that limit message passing in graph neural networks (Jake Topping and colleagues, ICLR 2022). These are analogues, defined and studied on their own terms; the theorems of this book are about the smooth flow. The site's Applications page collects such uses with their references.

In the world Analogy Curvature diffuses

Heat evens out temperature: hot spots cool, cold spots warm, and the temperature tends to its average. The Ricci flow evens out curvature in the same way, at least at first. In two dimensions the scalar curvature satisfies ∂tR=ΔR+R2\partial_tR = \Delta R + R^2 (11A.7 Ricci Flow on Surfaces), a heat equation with an extra term.

Where the picture breaks

The analogy breaks in three ways. The equation is nonlinear: the reaction terms in the evolution of curvature (11A.2 How Curvature Evolves) can drive curvature up, not just spread it out, so regions of high positive curvature can blow up in finite time. It is only weakly parabolic: the diffeomorphism invariance below means the equation is degenerate in some directions, and existence needs DeTurck's trick (11A.3 Short-Time Existence and Uniqueness). And it forms singularities, necks pinching and caps forming, which no linear heat equation does (11B.4 Singularities).

A heat equation for the metric

Why should −2Ric⁡-2\operatorname{Ric} be a Laplacian? In harmonic coordinates, coordinates xkx^k with Δxk=0\Delta x^k = 0, which exist near any point, the Ricci tensor has the form

Rij=−12gkl∂k∂lgij+Qij(g,∂g),R_{ij} = -\tfrac12g^{kl}\partial_k\partial_lg_{ij} + Q_{ij}(g, \partial g),

where QQ is quadratic in the first derivatives of gg (Petersen; Topping, chapter 1). So, in such coordinates,

∂tgij=gkl∂k∂lgij−2Qij(g,∂g):\partial_tg_{ij} = g^{kl}\partial_k\partial_lg_{ij} - 2Q_{ij}(g, \partial g):

each component of the metric satisfies a heat equation, coupled to the others through lower-order terms. The factor 22 makes the leading coefficient exactly 11, and the minus sign makes it a forward heat equation, well-posed forward in time (6A.1 What a PDE Is). The catch, that the coordinates themselves depend on the metric, is the subject of 11A.3 Short-Time Existence and Uniqueness.

The sign also explains the geometry. Where Ric⁡\operatorname{Ric} is positive, as on a sphere, the metric shrinks; where it is negative, as on a hyperbolic manifold, it expands. Since Ricci curvature measures how volume of small balls falls short of Euclidean (9A.4 Curvature and What It Means), the flow shrinks directions that are "too curved" and expands those that are not curved enough.

Symmetries

Scaling. If g(t)g(t) is a Ricci flow, so is

gλ(t)=λ g(tλ),λ>0,g_\lambda(t) = \lambda\,g\big(\tfrac t\lambda\big), \qquad \lambda > 0,

because Ric⁡\operatorname{Ric} is unchanged by constant scaling (9A.4 Curvature and What It Means) while ∂t\partial_t picks up a factor 1λ\frac1\lambda. Distances scale by λ\sqrt\lambda and time by λ\lambda: parabolic scaling, as for the heat equation (6A.1 What a PDE Is). Curvature scales by λ−1\lambda^{-1}. Blowing up a flow near a singularity, with λ=∣Rm⁡∣\lambda = |\operatorname{Rm}| at a point of large curvature, uses exactly this (11B.4 Singularities).

Diffeomorphism invariance. If ϕ\phi is a diffeomorphism, then ϕ∗g(t)\phi^*g(t) is also a Ricci flow, because Ric⁡(ϕ∗g)=ϕ∗Ric⁡(g)\operatorname{Ric}(\phi^*g) = \phi^*\operatorname{Ric}(g) (8A.6 Flows and the Lie Derivative). Two consequences come later. The equation is not strictly parabolic, which DeTurck's trick repairs (11A.3 Short-Time Existence and Uniqueness). And solutions that change only by diffeomorphisms and scaling, the Ricci solitons, are the natural self-similar solutions (11B.1 Ricci Solitons).

Exact solutions

Einstein metrics. If Ric⁡(g0)=λg0\operatorname{Ric}(g_0) = \lambda g_0, then g(t)=(1−2λt)g0g(t) = (1 - 2\lambda t)g_0 (9A.4 Curvature and What It Means).

  • The round sphere of radius r0r_0 in dimension nn has λ=n−1r02\lambda = \frac{n - 1}{r_0^2}, so r(t)2=r02−2(n−1)tr(t)^2 = r_0^2 - 2(n - 1)t. It shrinks to a point at T=r022(n−1)T = \frac{r_0^2}{2(n - 1)}, staying round.
  • Hyperbolic metrics (λ=−(n−1)\lambda = -(n - 1)) expand forever: g(t)=(1+2(n−1)t)g0g(t) = (1 + 2(n - 1)t)g_0.
  • Flat metrics do not move.

Products. The Ricci tensor of a product is the product of the Ricci tensors (9A.5 Computing Curvature), so each factor flows independently. The round cylinder Sn−1×RS^{n-1}\times\mathbb{R} with sphere radius ρ0\rho_0 evolves by

ρ(t)2=ρ02−2(n−2)t,\rho(t)^2 = \rho_0^2 - 2(n - 2)t,

pinching everywhere at once at T=ρ022(n−2)T = \frac{\rho_0^2}{2(n - 2)}, while the line factor does not move (Figure 1.1, Figure 1.2). For n=3n = 3: S2(ρ02−2t)×RS^2(\sqrt{\rho_0^2 - 2t})\times\mathbb{R}. This is the neck, the most important singularity model of the three-dimensional flow.

Figure 1.1. Radius squared against time for the round S3S^3 (r2=1−4tr^2 = 1 - 4t, extinct at t=14t = \frac14) and for the S2S^2 factor of the cylinder S2×RS^2\times\mathbb{R} (ρ2=1−2t\rho^2 = 1 - 2t, pinching at t=12t = \frac12), both starting with radius 11. Both are straight lines: the squared length scale changes linearly, the hallmark of a Type I singularity (11B.4 Singularities).
Figure 1.2. The round cylinder S2×RS^2\times\mathbb{R} (drawn one dimension down) at t=0t = 0, 14\frac14 and 716\frac{7}{16} for ρ0=1\rho_0 = 1 (computed radii 11, 0.710.71, 0.350.35). The sphere factor shrinks; the line factor does not move.

The normalised flow

On a closed manifold, the volume changes by ddtVol⁡=−∫R dV\frac{d}{dt}\operatorname{Vol} = -\int R\,dV (8A.7 Tensors and Index Notation), so a sphere shrinks to nothing. Hamilton also used the normalised Ricci flow

∂tg=−2Ric⁡+2nr g,r=∫MR dV∫MdV,\partial_tg = -2\operatorname{Ric} + \frac2nr\,g, \qquad r = \frac{\int_MR\,dV}{\int_MdV},

which preserves volume (Exercise 1.3). It is the unnormalised flow, rescaled at each time to keep volume fixed, with a reparametrised time. So the two have the same geometry up to scaling. On the round sphere the normalised flow is stationary, which is why Hamilton's theorem (11A.6 Hamilton’s 1982 Theorem) is stated as "the normalised flow converges to a round metric".

Where this goes What comes next

11A.2 How Curvature Evolves derives how curvature evolves under the flow, checking every formula on the solutions of this chapter. 11A.3 Short-Time Existence and Uniqueness proves that a solution exists for a short time from any initial metric on a closed manifold, and is unique. 11A.4 Maximum Principles under Ricci Flow develops the maximum principles that control it, and 11A.6 Hamilton’s 1982 Theorem uses them to prove Hamilton's 1982 theorem.

History

Hamilton introduced the Ricci flow in "Three-manifolds with positive Ricci curvature" (Journal of Differential Geometry, 1982). Friedan's paper on nonlinear sigma models appeared in Physical Review Letters in 1980, with a fuller account in 1985. DeTurck's simplified existence proof followed in 1983. Discrete Ricci flows for surfaces were introduced by Bennett Chow and Feng Luo in 2003, and graph versions in the 2010s.

Recall Where we stand

The Ricci flow ∂tg=−2Ric⁡\partial_tg = -2\operatorname{Ric} is, in harmonic coordinates, a heat equation for the metric components with quadratic lower-order terms; the factor 22 normalises the Laplacian and the sign makes it forward-parabolic. It is invariant under parabolic scaling λg(t/λ)\lambda g(t/\lambda) and under diffeomorphisms. Einstein metrics evolve homothetically: spheres shrink to points at T=r022(n−1)T = \frac{r_0^2}{2(n - 1)}, hyperbolic metrics expand, flat ones stay. Products flow factor by factor, so the cylinder Sn−1×RS^{n-1}\times\mathbb{R} pinches at T=ρ022(n−2)T = \frac{\rho_0^2}{2(n - 2)}. The normalised flow keeps volume fixed and differs only by scaling. 11A.2 How Curvature Evolves computes how curvature evolves.

Exercises

Exercise 1.1 The sphere and the cylinder

Verify that g(t)=(r02−2(n−1)t)gSng(t) = (r_0^2 - 2(n - 1)t)g_{S^n} and g(t)=(ρ02−2(n−2)t)gSn−1+dz2g(t) = (\rho_0^2 - 2(n - 2)t)g_{S^{n-1}} + dz^2 are Ricci flows, using Ric⁡(gSm)=(m−1)gSm\operatorname{Ric}(g_{S^m}) = (m - 1)g_{S^m} for the unit sphere and scale invariance of Ric⁡\operatorname{Ric}. Compute the scalar curvature of each and show it blows up like cT−t\frac{c}{T - t}.

Solution

∂t((r02−2(n−1)t)gSn)=−2(n−1)gSn=−2Ric⁡(gSn)\partial_t((r_0^2 - 2(n - 1)t)g_{S^n}) = -2(n - 1)g_{S^n} = -2\operatorname{Ric}(g_{S^n}), and Ric⁡\operatorname{Ric} of a rescaled metric is unchanged. Likewise for the cylinder, with m=n−1m = n - 1. Scalar curvature: R=n(n−1)r2=n(n−1)2(n−1)(T−t)=n2(T−t)R = \frac{n(n - 1)}{r^2} = \frac{n(n - 1)}{2(n - 1)(T - t)} = \frac{n}{2(T - t)} for the sphere, and R=(n−1)(n−2)ρ2=n−12(T−t)R = \frac{(n - 1)(n - 2)}{\rho^2} = \frac{n - 1}{2(T - t)} for the cylinder.

Exercise 1.2 Parabolic scaling

Show directly that if g(t)g(t) solves the Ricci flow on [0,T)[0, T), then gλ(t)=λg(t/λ)g_\lambda(t) = \lambda g(t/\lambda) solves it on [0,λT)[0, \lambda T). How does ∣Rm⁡∣|\operatorname{Rm}| transform? If a flow has ∣Rm⁡∣≤K|\operatorname{Rm}| \leq K on [0,T)[0, T), what bound does gλg_\lambda have, and on what interval?

Solution

∂tgλ(t)=λ⋅1λ(∂tg)(t/λ)=−2Ric⁡(g(t/λ))=−2Ric⁡(gλ(t))\partial_tg_\lambda(t) = \lambda\cdot\frac1\lambda(\partial_tg)(t/\lambda) = -2\operatorname{Ric}(g(t/\lambda)) = -2\operatorname{Ric}(g_\lambda(t)). The (1,3)(1, 3) curvature tensor is scale invariant and its norm involves the metric so that ∣Rm⁡λg∣λg=λ−1∣Rm⁡g∣g|\operatorname{Rm}_{\lambda g}|_{\lambda g} = \lambda^{-1}|\operatorname{Rm}_g|_g. So gλg_\lambda has ∣Rm⁡∣≤Kλ|\operatorname{Rm}| \leq \frac K\lambda on [0,λT)[0, \lambda T): choosing λ=K\lambda = K normalises the curvature bound to 11.

Exercise 1.3 The normalised flow preserves volume

Using ∂t dV=12tr⁡g(∂tg) dV\partial_t\,dV = \frac12\operatorname{tr}_g(\partial_tg)\,dV (9B.7 The Heat Equation on a Manifold), show that the normalised flow satisfies ddtVol⁡=∫(−R+r) dV=0\frac{d}{dt}\operatorname{Vol} = \int(-R + r)\,dV = 0.

Solution

12tr⁡(−2Ric⁡+2nrg)=−R+r\frac12\operatorname{tr}(-2\operatorname{Ric} + \frac2nrg) = -R + r, so ddtVol⁡=∫(r−R) dV=rVol⁡−∫R dV=0\frac{d}{dt}\operatorname{Vol} = \int(r - R)\,dV = r\operatorname{Vol} - \int R\,dV = 0 by the definition of rr.

Exercise 1.4 Hyperbolic metrics expand

Show that a hyperbolic metric expands linearly, and that its normalised flow is stationary. Show more generally that any Einstein metric is a fixed point of the normalised flow.

Solution

Ric⁡=−(n−1)g0\operatorname{Ric} = -(n - 1)g_0 gives g(t)=(1+2(n−1)t)g0g(t) = (1 + 2(n - 1)t)g_0. For an Einstein metric, Ric⁡=λg\operatorname{Ric} = \lambda g, R=nλR = n\lambda, r=nλr = n\lambda, and −2Ric⁡+2nrg=−2λg+2λg=0-2\operatorname{Ric} + \frac2nrg = -2\lambda g + 2\lambda g = 0.

Exercise 1.5 Rehearsal: from the unnormalised to the normalised flow

Let g(t)g(t) be a Ricci flow on a closed nn-manifold. Define ψ(t)=(Vol⁡(g(0))Vol⁡(g(t)))2/n\psi(t) = \big(\frac{\operatorname{Vol}(g(0))}{\operatorname{Vol}(g(t))}\big)^{2/n}, g~=ψg\tilde g = \psi g and t~=∫0tψ(s) ds\tilde t = \int_0^t\psi(s)\,ds. Show dψdt=2nψ r(t)\frac{d\psi}{dt} = \frac2n\psi\,r(t), where rr is the average scalar curvature of g(t)g(t), and deduce that g~\tilde g solves the normalised flow in the time t~\tilde t. (Use R~=ψ−1R\tilde R = \psi^{-1}R and Ric⁡~=Ric⁡\widetilde{\operatorname{Ric}} = \operatorname{Ric}.)

Solution

ddtlog⁡Vol⁡=−r\frac{d}{dt}\log\operatorname{Vol} = -r, so ddtlog⁡ψ=−2nddtlog⁡Vol⁡=2nr\frac{d}{dt}\log\psi = -\frac2n\frac{d}{dt}\log\operatorname{Vol} = \frac2nr. Then ∂g~∂t~=1ψddt(ψg)=ψ′ψg+∂tg=2nrg−2Ric⁡(g)\frac{\partial\tilde g}{\partial\tilde t} = \frac{1}{\psi}\frac{d}{dt}(\psi g) = \frac{\psi'}{\psi}g + \partial_tg = \frac2nrg - 2\operatorname{Ric}(g). Now Ric⁡(g~)=Ric⁡(g)\operatorname{Ric}(\tilde g) = \operatorname{Ric}(g) and r~=ψ−1r\tilde r = \psi^{-1}r, so 2nrg=2nr~ψg=2nr~g~\frac2nrg = \frac2n\tilde r\psi g = \frac2n\tilde r\tilde g. Hence ∂t~g~=−2Ric⁡(g~)+2nr~g~\partial_{\tilde t}\tilde g = -2\operatorname{Ric}(\tilde g) + \frac2n\tilde r\tilde g.

© 2026 NeckPinch (www.neckpinch.com). All content in the guidebook (text, mathematics, figures and exercises) is protected by copyright. All rights reserved. No part may be copied, republished or redistributed without written permission.