Model spaces, necks, Lie groups and the curvature operator.
29 min read · Updated Oct 3, 2026
Read with Lee, Introduction to Riemannian Manifolds (2nd edition), chapter 7 (the Ricci decomposition, the Weyl tensor, conformal changes) and the computations for the model spaces and warped products in chapters 3 and 8 and their problems. Petersen's Riemannian Geometry, chapter 4 (examples: rotationally symmetric metrics, doubly warped products, Berger spheres), is the second voice. Milnor's paper "Curvatures of left invariant metrics on Lie groups" (Advances in Mathematics, 1976) is the source for the Lie group section.
This chapter is a workshop. It computes the curvature of the examples that the Ricci flow needs, in a form you can reuse: the model spaces, products, rotationally symmetric metrics (necks, caps, cones and dumbbells), conformal changes, and left-invariant metrics on three-dimensional Lie groups. It then takes the curvature tensor apart into its Ricci decomposition. In dimension three that decomposition shows that the Ricci tensor determines the whole curvature tensor, which is why Hamilton's three-dimensional theory can work with Ricci curvature alone.
There are fewer real-world boxes here than elsewhere, deliberately. The payoff is the next three books: every curvature value quoted later for a neck, a cigar, a Berger sphere or a Bryant soliton comes from the formulas below. Every formula in this chapter has been checked numerically by the figures script for the chapter (in the guide's tools/ folder), on the examples given.
By the end of this chapter you will be able to:
compute the curvature of products and of warped products dr2+φ(r)2gSn−1;
read off the geometry of a neck from its profile;
transform curvature under a conformal change of metric;
compute the Ricci curvature of left-invariant metrics on three-dimensional Lie groups, including Berger spheres, Nil and Sol;
state the Ricci decomposition, explain why W=0 in dimension 3, and describe the curvature operator.
On a Riemannian product (M1×M2,g1⊕g2), the Levi-Civita connection and the curvature split: vectors tangent to the two factors do not interact. So Rm=Rm1⊕Rm2, Ric=Ric1⊕Ric2 and R=R1+R2, and a mixed plane, spanned by X1∈TM1 and X2∈TM2, has K(X1,X2)=0.
S2(ρ)×R, the round cylinder in dimension 3: Ric has eigenvalues ρ21,ρ21,0, the zero in the axis direction, and R=ρ22. Its sectional curvatures range over [0,ρ21].
H2×R: Ric has eigenvalues −1,−1,0.
S2×S2 (equal radii 1): Einstein, with Ric=g, but not of constant curvature; sectional curvatures range over [0,1].
The cylinder is the most important of these. It is the model of a neck in a Ricci flow, and under the flow it shrinks by ρ(t)2=ρ02−2t (the last exercise), pinching off at t=2ρ02 everywhere at once.
For the warped product g=dr2+φ(r)2gSn−1 of 9A.1 Riemannian Metrics and Model Spaces, use an orthonormal frame e1=∂r and e2,…,en tangent to the spheres. The curvature operator is diagonal in this frame, and there are only two kinds of planes:
Why. The radial planes: the surface swept out by the geodesics in the plane of ∂r and ei is dr2+φ(r)2dθ2, totally geodesic, and its Gauss curvature is −φφ′′ by the formula of 8A.9 The Curvature of Surfaces (Exercise 5.1). The tangential planes: the sphere {r=r0} is a round sphere of radius φ(r0), with intrinsic curvature φ21, sitting in M with second fundamental form φφ′ times its metric; the Gauss equation (9A.8 Submanifolds and Minimal Surfaces) subtracts (φφ′)2. Petersen works through the full computation for rotationally symmetric metrics.
Checks. The sphere (φ=sinr): Krad=Ktan=1. Hyperbolic space (φ=sinhr): both −1. The cylinder (φ=ρ): Krad=0, Ktan=ρ21. The cone (φ=ar, 0<a<1): Krad=0 and Ktan=a2r21−a2>0, blowing up at the tip. So in dimension n≥3 a cone over a small sphere has positive curvature blowing up at the vertex, unlike the 2D cone, which is flat away from its tip. These formulas make the model spaces of 9A.1 Riemannian Metrics and Model Spaces honest: Rn, Sn and Hn have constant curvature 0, 1, −1.
A neck.Figure 5.1 shows the two curvatures along a dumbbell metric on S3 whose neck has radius 0.2. In the middle of the neck, Ktan=25 and Krad=1, so the neck is nearly a round cylinder S2×R of radius 0.2, with scalar curvature R=2(2Krad+Ktan)=54, close to the cylinder's 0.222=50. At the shoulders, where the neck widens quickly, φ′′ is large and positive, and both curvatures become strongly negative, Ktan wherever ∣φ′∣>1. That mix of large positive and negative curvature is what Angenent and Knopf had to control to prove that such necks pinch (11B.4 Singularities).
Figure 5.1. Sectional curvatures of a dumbbell metric dr2+φ(r)2gS2 on S3 with a long neck of radius 0.2, computed from the formulas above. The dashed curve is the profile φ (scaled by 30). Across the neck the tangential curvature is 0.221=25 and the radial curvature about 1: the neck is nearly a cylinder. At the shoulders both dip sharply negative.
with Δ, ∇ and norms taken in g (Lee, chapter 7). In dimension 2 this is K~=e−2u(K−Δu), the equation behind uniformization and the two-dimensional Ricci flow (5A.5 Uniformization and the Two-Dimensional Ricci Flow). In dimension n≥3 it is the starting point of the Yamabe problem: find a conformal metric of constant scalar curvature. Writing e2u=v4/(n−2) turns the scalar curvature equation into the semilinear equation −n−24(n−1)Δv+Rv=R~v(n+2)/(n−2) (the critical exponent of the Sobolev embedding, 4A.10 Sobolev Embeddings and Critical Exponents).
As a check, the Poincaré ball metric (1−∣y∣2)24∣dy∣2 is e2u times the flat metric with u=log2−log(1−∣y∣2), and the formula gives R~=−n(n−1) (Exercise 5.3).
5.4Left-invariant metrics on three-dimensional Lie groups
A left-invariant metric on a Lie group G is determined by an inner product on the Lie algebra g (8A.5 Lie Groups and Group Actions). Its curvature is a computation in linear algebra, and its Ricci flow is a system of ODEs (11A.8 Homogeneous Flows). John Milnor (1976) showed that for a three-dimensional unimodular Lie group (one with tradX=0 for all X, which includes every group with a compact quotient), any left-invariant metric has an orthonormal basis e1,e2,e3 of g with
The signs of (λ1,λ2,λ3) identify the group: (+,+,+) is SU(2), so S3; (+,+,−) is SL(2,R); (+,+,0) is the universal cover of the Euclidean group of the plane; (+,−,0) is Sol; (+,0,0) is the Heisenberg group, Nil; (0,0,0) is R3. These carry five of Thurston's eight geometries (10A.5 Thurston’s Eight Geometries).
The round S3. For SU(2), the unit quaternions, the left-invariant fields X1=qi, X2=qj, X3=qk are orthonormal for the round metric of radius 1, and [X1,X2]=2X3 cyclically. So λi=2, μi=1 and Ric=2g, as it must be.
Berger spheres. Rescale X1, the direction of the Hopf circles, to have length ε: the orthonormal basis is e1=X1/ε, e2=X2, e3=X3, with λ=(2ε,ε2,ε2). Then μ1=ε2−ε and μ2=μ3=ε, so
Ric(e1,e1)=2ε2,Ric(e2,e2)=Ric(e3,e3)=4−2ε2,
and the sectional curvatures of the coordinate planes are K(e1,e2)=K(e1,e3)=ε2 and K(e2,e3)=4−3ε2 (Exercise 5.4). As ε→0 the curvatures stay between 0 and 4 while the Hopf circles shrink: collapse with bounded curvature (9B.3 Collapsing and Noncollapsing). Ricci curvature is positive exactly for ε<2 (Figure 5.2).
Nil, λ=(1,0,0): μ=(−21,21,21) and Ric has eigenvalues 21,−21,−21; R=−21.
Sol, λ=(1,−1,0): μ=(−1,1,0) and Ric has eigenvalues 0,0,−2.
Figure 5.2. Curvatures of the Berger spheres, the round S3 with the Hopf circles rescaled to length 2πε, from Milnor's formulas. At ε=1 everything equals the round values (K=1, Ric=2). Ricci curvature is positive for ε<2; as ε→0 the curvature stays bounded while the volume goes to zero.
The Kulkarni–Nomizu product of two symmetric 2-tensors is the 4-tensor
(h◯∧k)ijkl=hilkjk+hjkkil−hikkjl−hjlkik,
which has the symmetries of a curvature tensor. In this notation a constant curvature tensor is 2κg◯∧g. Every curvature tensor splits orthogonally as
Rm=W+n−21Ric˚◯∧g+2n(n−1)Rg◯∧g,
where Ric˚=Ric−nRg is the trace-free Ricci tensor and the Weyl tensorW is the part with all traces zero (Lee, chapter 7). The three pieces are the scalar part (what constant curvature has), the trace-free Ricci part (what Einstein metrics lack), and the Weyl part, the curvature that Ricci curvature does not see. The Weyl tensor is conformally invariant: W(e2ug)=e2uW(g) as a (0,4)-tensor. In general relativity it is the curvature of empty space, the tidal shape change of 9A.4 Curvature and What It Means.
The ideaWhy three dimensions is special
The Weyl tensor has the symmetries of a curvature tensor and all its traces vanish. In dimension 3, a curvature tensor has 6 independent components and the Ricci tensor also has 6, so no room is left: W=0 in dimension 3, and the Ricci tensor determines the whole curvature tensor,
This is why Hamilton's 1982 theorem (11A.6 Hamilton’s 1982 Theorem) can be stated and proved with Ricci curvature: in dimension 3, controlling Ric controls everything. It is also why the three-dimensional Ricci flow is so much more tractable than the four-dimensional one, where the Weyl tensor has 10 components of its own.
Since Rm is antisymmetric in each pair of slots, it defines a symmetric linear map on 2-vectors, the curvature operatorR:Λ2TpM→Λ2TpM, by
⟨R(X∧Y),Z∧W⟩=Rm(X,Y,W,Z),
where Λ2 has the inner product that makes ei∧ej (i<j) orthonormal. With this normalisation ⟨R(X∧Y),X∧Y⟩=K(X,Y)∣X∧Y∣2, and the unit sphere has R=id. (Other books scale R by a factor of 2 or change its sign; test on the unit sphere.) Positive curvature operator, R>0, implies positive sectional curvature, but is stronger in dimension 4 and above, because not every 2-vector is a plane X∧Y.
In dimension 3, every 2-vector is decomposable, Λ2≅TM (by ei∧ej↦ek for (i,j,k) cyclic), and if Ric has eigenvalues r1,r2,r3 in an orthonormal eigenbasis, then R is diagonal in the basis e2∧e3,e3∧e1,e1∧e2 with eigenvalues
2r2+r3−r1,2r3+r1−r2,2r1+r2−r3,
which are the sectional curvatures of the three coordinate planes (Exercise 5.5). So in dimension 3, R≥0 if and only if all sectional curvatures are ≥0, and the Hamilton–Ivey pinching estimate (11A.5 Hamilton–Ivey Pinching) is a statement about the smallest eigenvalue of R.
Where this goesCurvature as a moving target
Under the Ricci flow, the curvature operator evolves by a heat equation with a quadratic reaction term. In Hamilton's normalisation M=2R, it reads ∂tM=ΔM+M2+M#, where M# is a quadratic expression in M (11A.2 How Curvature Evolves). Hamilton's maximum principle for systems (11A.4 Maximum Principles under Ricci Flow) reduces many questions about the flow to the ODE dtdM=M2+M#, which in dimension 3 is an ODE for the three eigenvalues computed above (doubled).
Rotationally symmetric metrics have been the first examples since Riemann. Milnor's paper on left-invariant metrics appeared in 1976. The Weyl tensor was introduced by Hermann Weyl in 1918; the Kulkarni–Nomizu product is named after Ravindra Kulkarni and Katsumi Nomizu. The Yamabe problem was posed by Hidehiko Yamabe in 1960 and solved by the combined work of Neil Trudinger (1968), Thierry Aubin (1976) and Richard Schoen (1984). Berger spheres are named after Marcel Berger.
RecallWhere we stand
Products split curvature, with mixed planes flat; the cylinder S2×R has Ricci eigenvalues (ρ21,ρ21,0). Warped products dr2+φ2gSn−1 have only two sectional curvatures, −φφ′′ and φ21−φ′2; a thin neck has a large tangential and small radial curvature, nearly a cylinder, and its shoulders have negative curvature. Conformal change transforms Ric and R by explicit formulas generalising K~=e−2u(K−Δu). Milnor's formulas give the Ricci curvature of left-invariant metrics on unimodular 3D Lie groups: Berger spheres have Ricci >0 for ε<2, Nil has (21,−21,−21), Sol (0,0,−2). Curvature splits into scalar, trace-free Ricci and Weyl parts; in dimension 3, W=0 and Ricci determines everything, and the curvature operator has eigenvalues 21(rj+rk−ri). 9A.6 The Laplacian and the Bochner Formula builds the calculus (Laplacian, Bochner formula, commuting derivatives) in which the flow's evolution equations are written.
For the surface dr2+φ(r)2dθ2, use the formula K=−G(G)rr for a metric dr2+Gdθ2 (8A.9 The Curvature of Surfaces) to show K=−φφ′′. Check the sphere, the hyperbolic plane and the cone.
Solution
G=φ, so K=−φφ′′. Sphere: −sinr−sinr=1; hyperbolic: −sinhrsinhr=−1; cone φ=ar: φ′′=0, so K=0.
Exercise 5.2Products and cylinders
Compute Ric and R for S2(ρ)×R, H2×R, S2×S1 and S2×S2. Which are Einstein? Which have Ric≥0?
Solution
S2(ρ)×R and S2(ρ)×S1: Ric=ρ21gS2(ρ)⊕0, R=ρ22, Ric≥0, not Einstein. H2×R: Ric=−gH2⊕0, R=−2. S2×S2 with equal radii 1: Ric=g, Einstein, R=4. (In 9A.7 Jacobi Fields and Curvature versus Topology you prove that a compact manifold with Ric>0 has finite fundamental group, so S2×S1, with infinite π1, has no metric with Ric>0, though it has one with Ric≥0.)
Exercise 5.3Hyperbolic space by conformal change
With u=log2−log(1−∣y∣2) on the unit ball in Rn and the flat background (Ric=0), compute ∇u, ∣∇u∣2 and Δu, and show R~=−n(n−1).
Solution
With ρ=∣y∣: ∇u=1−ρ22y, ∣∇u∣2=(1−ρ2)24ρ2, Δu=1−ρ22n+(1−ρ2)24ρ2. Then R~=4(1−ρ2)2(n−1)(−1−ρ24n−(1−ρ2)28ρ2−(1−ρ2)24(n−2)ρ2)=4n−1(−4n(1−ρ2)−4nρ2)=−n(n−1).
Exercise 5.4Berger spheres
Verify the brackets λ=(2ε,ε2,ε2) for the Berger basis, and the Ricci curvatures. Using Ric(ei,ei)=∑j=iK(ei,ej) (the curvature operator is diagonal in Milnor's frame), find the three coordinate sectional curvatures. For which ε is the scalar curvature positive?
Solution
[e2,e3]=[X2,X3]=2X1=2εe1; [e3,e1]=ε1[X3,X1]=ε2X2=ε2e2; similarly [e1,e2]=ε2e3. Then 21∑λ=ε+ε2, so μ1=ε2−ε, μ2=μ3=ε, and Ric is (2ε2,4−2ε2,4−2ε2). Writing Kij for the coordinate planes: K12+K13=2ε2, K12+K23=4−2ε2=K13+K23, so K12=K13=ε2 and K23=4−3ε2. R=8−2ε2>0 for ε<2.
Exercise 5.5Dimension three
In an orthonormal basis diagonalising Ric with eigenvalues r1,r2,r3, use the formula for Rijkl in the box above to show K(e2,e3)=R2332=2r2+r3−r1, and that the components Rijkl with {i,j}={k,l} vanish. Check the round S3 and S2×R.
Solution
R2332=R22g33+R33g22−R23g32−R32g23−2R(g22g33−g23g32)=r2+r3−2r1+r2+r3. If {i,j}={k,l}, every term of the formula contains an off-diagonal g or an off-diagonal R, which vanish. Round S3: ri=2 gives K=1. S2×R with r=(1,1,0) (e3 the axis): K(e1,e2)=21+1−0=1, K(e2,e3)=21+0−1=0.
Exercise 5.6Nil
For the Heisenberg group with [e2,e3]=e1 and the other brackets of the basis zero, compute μ, Ric and R, and the sectional curvatures of the coordinate planes. Why can no left-invariant metric on Nil have constant curvature?
Solution
λ=(1,0,0), 21∑λ=21, μ=(−21,21,21); Ric=(2⋅41,2⋅(−41),2⋅(−41))=(21,−21,−21), R=−21. Coordinate planes: K23=2r2+r3−r1=−43, K12=K13=41. Any left-invariant metric on Nil can be put in Milnor's form with λ=(λ1,0,0), λ1>0, and then Ric has eigenvalues of both signs, so it is not a multiple of g.
Exercise 5.7Rehearsal: the shrinking neck
(a) For the cylinder Sn−1(ρ)×R, show that g(t)=ρ(t)2gSn−1+dz2 is a Ricci flow when ρ(t)2=ρ02−2(n−2)t, so it becomes singular at T=2(n−2)ρ02 with R(t)=ρ02−2(n−2)t(n−1)(n−2). (b) Check that (T−t)R(t) is constant: curvature blows up like T−t1, a Type I singularity (11B.4 Singularities). (c) Why does the R factor not move?
Solution
(a) Ric(g(t))=(n−2)gSn−1⊕0 (scale invariance, 9A.4 Curvature and What It Means), so ∂t(ρ2)=−2(n−2). R=ρ2(n−1)(n−2). (b) ρ(t)2=2(n−2)(T−t), so (T−t)R=2n−1. (c) Ric vanishes in the axis direction, so ∂tg(∂z,∂z)=0.