Book 11A

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

Course 11Book 11A: The Ricci Flow: Existence and Maximum PrinciplesChapter 7

Ricci Flow on Surfaces

Uniformization by Ricci flow, and Hamilton’s entropy.

17 min read · Updated Oct 3, 2026

Read with Chow and Knopf's The Ricci Flow: An Introduction, chapter 5 (the Ricci flow on surfaces: the conformal factor, Hamilton's entropy and Harnack inequality, convergence). Hamilton's "The Ricci flow on surfaces" (Contemporary Mathematics 71, 1988) is the original. 5A.5 Uniformization and the Two-Dimensional Ricci Flow met the equation first.

In this chapter · 7 sections
  1. 7.1Surfaces in the computer
  2. 7.2The conformal factor
  3. 7.3Convergence
  4. 7.4Hamilton's entropy
  5. 7.5The Harnack inequality and the cigar
  6. 7.6History
  7. 7.7Exercises

On a surface, the Ricci flow is a scalar equation in disguise. Since Ric⁡=Kg=R2g\operatorname{Ric} = Kg = \frac R2g, the flow

∂tg=−Rg\partial_tg = -Rg

moves the metric only by conformal factors, and it reduces to a nonlinear heat equation for one function. It was the first case Hamilton solved after 1982. The answer is the uniformization theorem, proved by a flow: every metric on a closed surface flows, after normalisation, to a metric of constant curvature. This chapter proves the easy cases, explains the hard one (the sphere), and introduces two tools that reappear in Perelman's work in more elaborate form: Hamilton's entropy and the Harnack inequality. It also meets the cigar, the first steady soliton.

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

  • write the two-dimensional flow as an equation for the conformal factor, and derive ∂tR=ΔR+R2\partial_tR = \Delta R + R^2;
  • prove convergence of the normalised flow when χ(M)<0\chi(M) < 0, and describe what happens on the sphere;
  • define Hamilton's surface entropy and explain why it is monotone;
  • state the two-dimensional Harnack inequality and check it on the cigar soliton;
  • describe the ancient solutions on the sphere.

Surfaces in the computer

In the world In use Discrete surface Ricci flow

In geometry processing, the uniformization theorem is a tool. A surface mesh, such as a scanned face or the folded surface of the cerebral cortex, can be mapped conformally onto a sphere, a disc or a plane region, which makes it possible to compare, register or texture surfaces. Discrete surface Ricci flow computes such maps. The mesh is given a discrete conformal structure (circle packings or vertex scalings), the target curvature is prescribed at the vertices, and a discrete flow drives the vertex curvatures to the target, much as the smooth flow drives KK to a constant. Miao Jin, Junho Kim, Feng Luo and Xianfeng Gu described the method in "Discrete surface Ricci flow" (IEEE Transactions on Visualization and Computer Graphics, 2008), building on Bennett Chow and Feng Luo's combinatorial Ricci flow (2003). Gu, Shing-Tung Yau and collaborators have applied it in brain-mapping research. It is a discrete analogue with its own convergence theory, and the smooth theorems of this chapter are its model.

The conformal factor

Write g(t)=e2u(t)g0g(t) = e^{2u(t)}g_0. Since the flow only rescales gg pointwise, the conformal class is preserved, and with K=e−2u(K0−Δ0u)K = e^{-2u}(K_0 - \Delta_0u) (5A.5 Uniformization and the Two-Dimensional Ricci Flow),

∂tu=−12R=−e−2u(K0−Δ0u)=e−2uΔ0u−e−2uK0.\partial_tu = -\tfrac12R = -e^{-2u}\big(K_0 - \Delta_0u\big) = e^{-2u}\Delta_0u - e^{-2u}K_0.

This is a quasilinear heat equation for uu, with diffusion coefficient e−2ue^{-2u}: it is strictly parabolic, so in two dimensions the DeTurck trick is unnecessary. The scalar curvature evolves by

∂tR=ΔR+R2,\partial_tR = \Delta R + R^2,

since 2∣Ric⁡∣2=2⋅2(R2)2=R22|\operatorname{Ric}|^2 = 2\cdot2\big(\frac R2\big)^2 = R^2 (11A.2 How Curvature Evolves). The normalised flow, which preserves area AA, is

∂tg=(r−R)g,r=∫R dAA=4πχ(M)A,\partial_tg = (r - R)g, \qquad r = \frac{\int R\,dA}{A} = \frac{4\pi\chi(M)}{A},

where rr is constant in time by Gauss–Bonnet (8A.9 The Curvature of Surfaces). Under it, ∂tR=ΔR+R(R−r)\partial_tR = \Delta R + R(R - r).

Convergence

Theorem 7.1 Hamilton (1988), Chow (1991)

On a closed surface, the normalised Ricci flow from any metric exists for all time and converges smoothly to a metric of constant curvature.

When χ(M)<0\chi(M) < 0. Here r<0r < 0, and the reaction term R(R−r)R(R - r) pushes RR towards rr. Comparing with the ODE ρ′=ρ(ρ−r)\rho' = \rho(\rho - r), whose solutions converge to rr exponentially when ρ\rho starts between the roots, the maximum principle gives bounds on RR, and a more careful argument (Hamilton's potential function ff with Δf=R−r\Delta f = R - r) gives ∣R−r∣≤Cert|R - r| \leq Ce^{rt} (Exercise 7.4). The case χ=0\chi = 0 is similar, with polynomial convergence.

The sphere. Here r>0r > 0, the reaction term is destabilising, and the proof is much harder. Hamilton proved convergence when R>0R > 0 initially, using the entropy and Harnack inequality below. Bennett Chow (1991) showed that any metric on S2S^2 develops R>0R > 0 in finite time. Hamilton's original argument used the uniformization theorem along the way, through the Kazdan–Warner identity, which rules out nontrivial solitons on S2S^2. Xiuxiong Chen, Peng Lu and Gang Tian (2006) removed that dependence, so the flow gives an independent proof of uniformization for the sphere.

Figure 7.1 shows an elongated sphere becoming round.

Figure 7.1. An elongated sphere under the normalised two-dimensional Ricci flow (computed by solving the conformal factor equation for a rotationally symmetric metric with area 4π4\pi; the area is renormalised at each step to remove discretisation drift). The Gauss curvature, between 0.270.27 and 1.571.57 at t=0t = 0, converges to 11: the sphere becomes round.

Hamilton's entropy

Proposition 7.2 Hamilton's surface entropy

On a closed surface with R>0R > 0, under the normalised flow, the entropy

N(t)=∫MRlog⁡R dAN(t) = \int_MR\log R\,dA

is nonincreasing.

The proof differentiates NN using ∂tR=ΔR+R(R−r)\partial_tR = \Delta R + R(R - r) and ∂t dA=(r−R) dA\partial_t\,dA = (r - R)\,dA, and finds dNdt=−∫∣∇R∣2R dA+∫(R−r)2 dA\frac{dN}{dt} = -\int\frac{|\nabla R|^2}{R}\,dA + \int(R - r)^2\,dA. The second term has the wrong sign. Hamilton controls it with a potential function, rewriting dNdt\frac{dN}{dt} as minus the integral of a square (Exercise 7.5 does the first part). Its form, an integral of Rlog⁡RR\log R that decreases, is the ancestor of Perelman's F\mathcal F- and W\mathcal W-functionals (12A.2 Ricci Flow as a Gradient Flow, 12A.3 The 𝓦-Entropy), which do the same for the full Ricci flow in every dimension, and of the entropy of the heat equation (6A.10 Entropy, Information and Diffusion). For the round sphere of area 4π4\pi, R=2R = 2 and N=8πlog⁡2≈17.42N = 8\pi\log2 \approx 17.42, its minimum value (Figure 7.2).

Figure 7.2. Hamilton's entropy N=∫Rlog⁡R dAN = \int R\log R\,dA along the flow of Figure 7.1 (computed), decreasing to the round value 8πlog⁡2≈17.428\pi\log2 \approx 17.42 (dashed). Monotone quantities like this are how Hamilton, and later Perelman, controlled the flow.

The Harnack inequality and the cigar

For the unnormalised flow on a surface with R>0R > 0, Hamilton proved the Harnack inequality

∂tlog⁡R−∣∇log⁡R∣2≥−1t,\partial_t\log R - |\nabla\log R|^2 \geq -\frac1t,

equivalently Δlog⁡R+R≥−1t\Delta\log R + R \geq -\frac1t, since ∂tlog⁡R=Δlog⁡R+∣∇log⁡R∣2+R\partial_t\log R = \Delta\log R + |\nabla\log R|^2 + R. Like the Li–Yau inequality for the heat equation (9B.7 The Heat Equation on a Manifold), integrating it along paths compares curvature at different points and times. Its matrix version in all dimensions (11B.2 Ancient Solutions and the Harnack Inequality) is one of the main tools of Book 11B.

The cigar soliton is the complete metric on R2\mathbb{R}^2

g=dx2+dy21+x2+y2,R=41+x2+y2,g = \frac{dx^2 + dy^2}{1 + x^2 + y^2}, \qquad R = \frac{4}{1 + x^2 + y^2},

which in geodesic polar coordinates is ds2+tanh⁡2s dθ2ds^2 + \tanh^2s\,d\theta^2 (9B.3 Collapsing and Noncollapsing). It is a steady soliton: under the Ricci flow it moves only by diffeomorphisms, flowing along a radial vector field (11B.1 Ricci Solitons). It satisfies the Harnack inequality with equality in the limit: Δlog⁡R+R=0\Delta\log R + R = 0 (Exercise 7.6).

Ancient solutions on the sphere. A solution is ancient if it exists for all t∈(−∞,T)t \in (-\infty, T). Shrinking round spheres are ancient, and so is the King–Rosenau solution, a "sausage" that looks, far in the past, like two cigars joined end to end, and becomes round as it shrinks. John King (1993) and Philip Rosenau (1995) found it as a solution of the logarithmic fast diffusion equation, which the conformal factor satisfies. Physicists Fateev, Onofri and Zamolodchikov found it independently in 1993, as the "sausage" sigma model. Panagiota Daskalopoulos, Hamilton and Nataša Šešum (2012) proved that these are the only compact ancient solutions on S2S^2, apart from quotients.

Where this goes Two dimensions as a rehearsal

The surface case contains, in miniature, the themes of the three-dimensional theory: monotone entropies (12A.3 The 𝓦-Entropy), the Harnack inequality (11B.2 Ancient Solutions and the Harnack Inequality), solitons as models (11B.1 Ricci Solitons), and ancient solutions as the possible limits of blow-ups (12B.1 κ-Solutions). The cigar plays a special role: it is the steady soliton that Perelman's noncollapsing theorem excludes as a singularity model in three dimensions (12A.4 κ-Noncollapsing).

History

Hamilton's "The Ricci flow on surfaces" appeared in 1988, Chow's "The Ricci flow on the 2-sphere" in 1991, and Chen–Lu–Tian's note in 2006. Hamilton introduced the surface entropy and the Harnack inequality in the 1988 paper. The cigar appeared there as well, and in physics as Witten's two-dimensional black hole (1991).

Recall Where we stand

On a surface, ∂tg=−Rg\partial_tg = -Rg keeps the conformal class, and the conformal factor solves a strictly parabolic equation; ∂tR=ΔR+R2\partial_tR = \Delta R + R^2, and the normalised flow has ∂tR=ΔR+R(R−r)\partial_tR = \Delta R + R(R - r) with r=4πχAr = \frac{4\pi\chi}{A}. Every metric on a closed surface converges under the normalised flow to constant curvature (Hamilton for χ≤0\chi \leq 0 and for R>0R > 0 on S2S^2, Chow for all metrics on S2S^2, Chen–Lu–Tian without circularity). With R>0R > 0, Hamilton's entropy ∫Rlog⁡R\int R\log R decreases, and the Harnack inequality Δlog⁡R+R≥−1t\Delta\log R + R \geq -\frac1t holds, with equality on the cigar soliton. Compact ancient solutions on S2S^2 are round spheres and the King–Rosenau sausage. 11A.8 Homogeneous Flows studies homogeneous flows in three dimensions, which reduce to ODEs.

Exercises

Exercise 7.3 The conformal factor equation

For g=e2ug0g = e^{2u}g_0, derive ∂tu=−12R\partial_tu = -\frac12R from ∂tg=−Rg\partial_tg = -Rg, and substitute R=2e−2u(K0−Δ0u)R = 2e^{-2u}(K_0 - \Delta_0u). Show that the normalised flow becomes ∂tu=r2−e−2u(K0−Δ0u)\partial_tu = \frac r2 - e^{-2u}(K_0 - \Delta_0u).

Solution

∂t(e2ug0)=2(∂tu)e2ug0=−Re2ug0\partial_t(e^{2u}g_0) = 2(\partial_tu)e^{2u}g_0 = -Re^{2u}g_0 gives ∂tu=−R2=−e−2u(K0−Δ0u)\partial_tu = -\frac R2 = -e^{-2u}(K_0 - \Delta_0u). For the normalised flow, 2∂tu=r−R2\partial_tu = r - R, so ∂tu=r2−e−2u(K0−Δ0u)\partial_tu = \frac r2 - e^{-2u}(K_0 - \Delta_0u).

Exercise 7.4 Negative Euler characteristic (the ODE part)

Under the normalised flow, Rmax⁡R_{\max} satisfies ddtRmax⁡≤Rmax⁡(Rmax⁡−r)\frac{d}{dt}R_{\max} \leq R_{\max}(R_{\max} - r) in the sense of 9B.7 The Heat Equation on a Manifold. With r<0r < 0 and r<Rmax⁡(0)<0r < R_{\max}(0) < 0, compare with the ODE ρ′=ρ(ρ−r)\rho' = \rho(\rho - r) to show Rmax⁡(t)−r≤CertR_{\max}(t) - r \leq Ce^{rt}. (The general case, with RR positive somewhere, uses Hamilton's potential function.)

Solution

The ODE ρ′=ρ(ρ−r)\rho' = \rho(\rho - r) with r<0r < 0 and r<ρ(0)<0r < \rho(0) < 0 has the solution ρ(t)=r1−(1−rρ(0))ert\rho(t) = \frac{r}{1 - \big(1 - \frac{r}{\rho(0)}\big)e^{rt}}, which tends to rr with ρ−r=O(ert)\rho - r = O(e^{rt}) (since r<0r < 0). The maximum principle gives Rmax⁡(t)≤ρ(t)R_{\max}(t) \leq \rho(t) with ρ(0)=Rmax⁡(0)\rho(0) = R_{\max}(0), hence Rmax⁡−r≤CertR_{\max} - r \leq Ce^{rt}. A symmetric argument bounds Rmin⁡R_{\min} from below once it is known to be greater than some value, and the potential function handles positive values of RR.

Exercise 7.5 The derivative of the entropy

Using ∂tR=ΔR+R(R−r)\partial_tR = \Delta R + R(R - r) and ∂t dA=(r−R) dA\partial_t\,dA = (r - R)\,dA, show that

ddt∫Rlog⁡R dA=−∫∣∇R∣2R dA+∫R(R−r) dA,\frac{d}{dt}\int R\log R\,dA = -\int\frac{|\nabla R|^2}{R}\,dA + \int R(R - r)\,dA,

and that ∫R(R−r) dA=∫(R−r)2 dA\int R(R - r)\,dA = \int(R - r)^2\,dA. (Use ∫(R−r) dA=0\int(R - r)\,dA = 0.)

Solution

ddt∫Rlog⁡R dA=∫(log⁡R+1)∂tR dA+∫Rlog⁡R (r−R) dA\frac{d}{dt}\int R\log R\,dA = \int(\log R + 1)\partial_tR\,dA + \int R\log R\,(r - R)\,dA. Substituting, the terms ∫log⁡R⋅R(R−r)\int\log R\cdot R(R - r) cancel, leaving ∫(log⁡R+1)ΔR+∫R(R−r)\int(\log R + 1)\Delta R + \int R(R - r). Integrating by parts, ∫log⁡R ΔR=−∫∣∇R∣2R\int\log R\,\Delta R = -\int\frac{|\nabla R|^2}{R} and ∫ΔR=0\int\Delta R = 0. Finally ∫R(R−r)=∫(R−r)2+r∫(R−r)=∫(R−r)2\int R(R - r) = \int(R - r)^2 + r\int(R - r) = \int(R - r)^2.

Exercise 7.6 The cigar

For g=dx2+dy21+x2+y2g = \frac{dx^2 + dy^2}{1 + x^2 + y^2}, use K=−12e−2uΔE(2u)K = -\frac12e^{-2u}\Delta_E(2u) for g=e2u∣dx∣2g = e^{2u}|dx|^2 to show R=41+r2R = \frac{4}{1 + r^2} (r2=x2+y2r^2 = x^2 + y^2). Then show Δglog⁡R=−R\Delta_g\log R = -R, so that Δlog⁡R+R=0\Delta\log R + R = 0.

Solution

e2u=(1+r2)−1e^{2u} = (1 + r^2)^{-1}, 2u=−log⁡(1+r2)2u = -\log(1 + r^2), and ΔElog⁡(1+r2)=f′′+f′r=2−2r2(1+r2)2+21+r2=4(1+r2)2\Delta_E\log(1 + r^2) = f'' + \frac{f'}{r} = \frac{2 - 2r^2}{(1 + r^2)^2} + \frac{2}{1 + r^2} = \frac{4}{(1 + r^2)^2}. So K=12(1+r2)4(1+r2)2=21+r2K = \frac12(1 + r^2)\frac{4}{(1 + r^2)^2} = \frac{2}{1 + r^2} and R=2KR = 2K. Then log⁡R=log⁡4−log⁡(1+r2)\log R = \log4 - \log(1 + r^2), Δg=(1+r2)ΔE\Delta_g = (1 + r^2)\Delta_E, and Δglog⁡R=−(1+r2)4(1+r2)2=−R\Delta_g\log R = -(1 + r^2)\frac{4}{(1 + r^2)^2} = -R.

Exercise 7.7 Rehearsal: the round sphere and the Harnack quantity

On the unnormalised flow of the round sphere, g(t)=(1−2t)gS2g(t) = (1 - 2t)g_{S^2}, compute R(t)R(t), and verify ∂tR=ΔR+R2\partial_tR = \Delta R + R^2. Compute the Harnack quantity ∂tlog⁡R−∣∇log⁡R∣2\partial_t\log R - |\nabla\log R|^2 and compare with −1t-\frac1t. Which inequality is "tight" for the sphere, and which for the cigar?

Solution

R=21−2tR = \frac{2}{1 - 2t}, R′=4(1−2t)2=R2R' = \frac{4}{(1 - 2t)^2} = R^2, with ΔR=0\Delta R = 0. The Harnack quantity is ∂tlog⁡R=R>0>−1t\partial_t\log R = R > 0 > -\frac1t: far from tight. For the cigar it is 00, the value for an eternal (steady) solution, while the bound −1t-\frac1t is attained only asymptotically by expanding solutions that emerge from a point. The sphere is compact and shrinking; the inequality is designed to be sharp on solitons.

© 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.