TheoremComplete

Cheeger-Gromoll Splitting Theorem

The splitting theorem shows that a complete manifold with non-negative Ricci curvature containing a geodesic line must split isometrically as a product. It is a fundamental rigidity result in Riemannian geometry.


Statement

Definition9.6Geodesic line

A geodesic line in a Riemannian manifold is a geodesic γ:RM\gamma: \mathbb{R} \to M that is minimizing between any two of its points: d(γ(s),γ(t))=std(\gamma(s), \gamma(t)) = |s - t| for all s,tRs, t \in \mathbb{R}.

Theorem9.8Cheeger-Gromoll splitting theorem

Let (Mn,g)(M^n, g) be a complete Riemannian manifold with Ric0\mathrm{Ric} \geq 0. If MM contains a geodesic line, then MM is isometric to Nn1×RN^{n-1} \times \mathbb{R}, where (N,gN)(N, g_N) is a complete Riemannian manifold with Ric0\mathrm{Ric} \geq 0.


Proof Outline

Proof

Step 1: Busemann functions. Let γ:RM\gamma: \mathbb{R} \to M be a geodesic line. Define the Busemann functions:

b+(x)=limt+(td(x,γ(t))),b(x)=limt+(td(x,γ(t))).b^+(x) = \lim_{t \to +\infty} (t - d(x, \gamma(t))), \quad b^-(x) = \lim_{t \to +\infty} (t - d(x, \gamma(-t))).

These limits exist by the triangle inequality (the functions ttd(x,γ(t))t \mapsto t - d(x, \gamma(t)) are bounded and non-decreasing). Both b+b^+ and bb^- are Lipschitz with b±1|\nabla b^\pm| \leq 1.

Step 2: Subharmonicity. The key analytical step: b+b^+ and bb^- are superharmonic in the barrier sense (Δb±0\Delta b^\pm \leq 0) when Ric0\mathrm{Ric} \geq 0. This follows from the Laplacian comparison theorem: for a distance function r(x)=d(x,p)r(x) = d(x, p) on a manifold with Ric0\mathrm{Ric} \geq 0, Δr(n1)/r\Delta r \leq (n-1)/r. Taking limits as p=γ(t)p = \gamma(t) \to \infty gives Δb+0\Delta b^+ \leq 0.

Step 3: Harmonicity. Since γ\gamma is a line, b++b=0b^+ + b^- = 0 along γ\gamma, and by the triangle inequality b++b0b^+ + b^- \leq 0 everywhere. Since both b+b^+ and bb^- are superharmonic, b++bb^+ + b^- is superharmonic and 0\leq 0. By the maximum principle, b++b=0b^+ + b^- = 0 on all of MM. Therefore b+b^+ is both super- and sub-harmonic, hence harmonic: Δb+=0\Delta b^+ = 0.

Step 4: Splitting. Since b+b^+ is harmonic with b+1|\nabla b^+| \leq 1, and equality b+=1|\nabla b^+| = 1 holds along γ\gamma (hence everywhere by the Bochner formula and Ric0\mathrm{Ric} \geq 0), we have b+=1|\nabla b^+| = 1 on all of MM.

The level sets {b+=c}\{b^+ = c\} are smooth hypersurfaces, and b+\nabla b^+ is a unit-length parallel gradient field (2b+=0\nabla^2 b^+ = 0 follows from the Bochner formula 0=Δb+2=22b+2+2Ric(b+,b+)22b+20 = \Delta |\nabla b^+|^2 = 2|\nabla^2 b^+|^2 + 2\mathrm{Ric}(\nabla b^+, \nabla b^+) \geq 2|\nabla^2 b^+|^2). The flow of b+\nabla b^+ gives the isometric splitting MN×RM \cong N \times \mathbb{R}, where N=(b+)1(0)N = (b^+)^{-1}(0). \blacksquare


Applications

ExampleConsequences of the splitting theorem
  1. Flat manifolds: A compact flat manifold (K=0K = 0, Ric=0\mathrm{Ric} = 0) has universal cover Rn\mathbb{R}^n (apply splitting iteratively).
  2. Structure of non-negative Ricci: Any complete manifold with Ric0\mathrm{Ric} \geq 0 splits as MN×RkM \cong N \times \mathbb{R}^k where NN has no line. If MM is compact, there are no lines, so no splitting occurs.
  3. Topological consequences: A complete non-compact manifold with Ric>0\mathrm{Ric} > 0 (strictly positive) cannot contain a line, hence cannot split. This constrains its topology.
RemarkGeneralizations

The splitting theorem has been generalized to:

  • Lorentzian setting (Eschenburg, Galloway): Timelike geodesic lines and non-negative timelike Ricci curvature yield a product splitting, relevant in general relativity.
  • Metric measure spaces (Gigli): The theorem holds for RCD(0,N)\mathrm{RCD}(0, N) spaces, extending to non-smooth settings.
  • Almost splitting (Cheeger-Colding): Manifolds with Ricϵ\mathrm{Ric} \geq -\epsilon and long "almost-lines" are Gromov-Hausdorff close to products.