Myers' Theorem: Positive Curvature Forces a Compact, Small Universe

Precise statement: what Myers' Theorem claims

Let (Mⁿ, g) be a connected, complete Riemannian manifold of dimension n ≥ 2. Suppose there is a constant k > 0 with a lower bound on the Ricci curvature:

Ric(v, v) ≥ (n−1)k · g(v, v)   for all tangent vectors v.

Then the diameter satisfies diam(M) ≤ π/√k. In particular, by the Hopf–Rinow theorem, a complete manifold with finite diameter is compact (closed bounded sets are compact, and M itself is one). Moreover the fundamental group π₁(M) is finite.

The normalization (n−1)k is chosen so the model case is exact: the round n-sphere of radius r = 1/√k has constant sectional curvature k, hence Ric = (n−1)k g, and diameter exactly π r = π/√k. So the theorem's bound is sharp, attained by the sphere. Note the hypothesis is on Ricci — an average of sectional curvatures — not on every sectional curvature, which makes Myers' 1941 result strictly stronger than Bonnet's 1855 surface theorem.

The picture: positive curvature makes geodesics reconverge

On a sphere, two geodesics leaving the north pole in different directions spread apart, slow, stop spreading, and then reconverge at the south pole. Positive curvature is precisely the tendency of nearby geodesics to bend back toward each other. If you follow a single geodesic too far, neighboring geodesics have already refocused (a conjugate point appears), and beyond that focal point the geodesic can no longer be the shortest path.

Myers turns this focusing into a hard length limit. Ricci curvature controls the average refocusing over the n−1 directions transverse to a geodesic. A uniform lower bound Ric ≥ (n−1)k forces a conjugate point to appear within arc length π/√k along every geodesic. Since any two points are joined by a minimizing geodesic (completeness), and a minimizing geodesic cannot extend past a conjugate point, no two points can be farther apart than π/√k. The manifold simply runs out of room to be large.

Key idea of the proof: the second variation of arc length

Fix a unit-speed geodesic γ: [0, L] → M that minimizes length between its endpoints p = γ(0) and q = γ(L). Minimality means the second variation of arc length is nonnegative for every variation fixing the endpoints. The mechanism is to feed the geodesic a cleverly chosen family of test variations and extract a contradiction if L is too large.

Pick a parallel orthonormal frame e₁(t),…,e_{n−1}(t) along γ, orthogonal to γ′. For each i use the variation field Vᵢ(t) = sin(πt/L) eᵢ(t). The second-variation (index) formula gives

I(Vᵢ, Vᵢ) = ∫₀ᴸ [ (π/L)²cos²(πt/L) − sin²(πt/L) · K(γ′, eᵢ) ] dt ≥ 0.

Now sum over i = 1,…,n−1. The sectional curvatures sum to Ricci: Σᵢ K(γ′, eᵢ) = Ric(γ′, γ′) ≥ (n−1)k. Summing and using ∫₀ᴸ cos² = ∫₀ᴸ sin² = L/2 yields 0 ≤ Σᵢ I(Vᵢ,Vᵢ) ≤ (n−1)(L/2)[ (π/L)² − k ]. Nonnegativity forces (π/L)² ≥ k, i.e. L ≤ π/√k. Summing is exactly the trick that converts a sectional hypothesis into a Ricci one.

Canonical example: the round sphere, and why the constant is exact

Take Mⁿ = Sⁿ(r), the sphere of radius r in ℝⁿ⁺¹, with its induced round metric. Its sectional curvature is constant K = 1/r² everywhere, so setting k = 1/r² we get Ric = (n−1)(1/r²) g, saturating the hypothesis. The theorem predicts diam ≤ π/√k = π r, and indeed the farthest two points are antipodes, at distance π r along a great circle. The bound is achieved, so no universal constant smaller than π works.

Trace the proof on S²(1): a great circle γ has length 2π, but the segment from a point to its antipode (length π) is the longest a minimizing geodesic can be. The variation field sin(πt/π) e(t) = sin(t) e(t) is exactly a Jacobi field vanishing at both endpoints — the geodesic from north pole to south pole has its antipode as a conjugate point at distance π = π/√k. Beyond it, great-circle arcs stop minimizing. The test field in Myers' proof is precisely this borderline Jacobi field, which is why the estimate is tight.

Why the hypotheses are essential — counterexamples and rigidity

Completeness is not optional. An open geodesic ball in Sⁿ(1) of radius slightly less than π/2 has curvature 1 everywhere yet is non-compact and can be extended; it violates the conclusion because it is incomplete. Without Hopf–Rinow you cannot guarantee minimizing geodesics between points, and the whole argument collapses.

Strict positivity is essential. If you only assume Ric ≥ 0, the flat plane ℝ² (Ric ≡ 0) is complete, non-compact, with infinite diameter and trivial (finite) π₁ — it defeats the diameter and compactness conclusions but not π₁-finiteness; the flat cylinder S¹ × ℝ has Ric ≡ 0, is complete, non-compact, and has π₁ = ℤ, infinite — this is the example showing finiteness of π₁ also fails at k = 0. So neither the diameter bound nor finiteness of π₁ survives at k = 0. (The k = 0 borderline is governed instead by the deeper Cheeger–Gromoll splitting theorem.)

Rigidity. Cheng's Maximal Diameter Theorem (1975) says equality diam = π/√k forces M to be isometric to the round sphere Sⁿ(1/√k) — the bound is not just sharp but rigid, with the sphere as the unique extremizer.

Applications and significance: topology from a curvature pinch

Myers' Theorem is the archetype of a curvature-controls-topology result and a workhorse in global Riemannian geometry. The finiteness of π₁ has an immediate corollary: the universal cover M̃ inherits the same Ricci bound (curvature is local), so M̃ is itself compact by Myers, and a compact cover of M can only wrap finitely many times — hence π₁(M) is finite. This kills, for instance, any hope of a flat torus or hyperbolic manifold carrying a metric of uniformly positive Ricci curvature.

It underpins sphere theorems, Gromov's precompactness and Bishop–Gromov volume comparison (same Ricci hypothesis, comparison geometry), and structural results in general relativity via the Lorentzian Hawking–Penrose singularity theorems, whose second-variation focusing arguments are direct descendants of Myers'. In Ricci-flow and Einstein-manifold theory, a positive lower Ricci bound instantly yields compactness and diameter control. Whenever you need to promote a pointwise curvature estimate to a global topological conclusion, Bonnet–Myers is usually the first tool reached for.

Which hypotheses buy which conclusions in the Bonnet–Myers circle of results

Hypothesis on curvature	Completeness?	Conclusion
Sectional curvature K ≥ k > 0 (Bonnet, surfaces & higher)	Yes	diam ≤ π/√k, compact, π₁ finite
Ricci Ric ≥ (n−1)k g, k > 0 (Myers)	Yes	diam ≤ π/√k, compact, π₁ finite (weaker hypothesis, same conclusion)
Ric ≥ 0 only (nonnegative)	Yes	No diameter bound; ℝⁿ, cylinders are counterexamples
Ric ≥ (n−1)k > 0 but M incomplete	No	Fails: open geodesic ball in the sphere has any diameter < π/√k but is not compact
Ric ≥ (n−1)k g, equality diam = π/√k (Cheng 1975)	Yes	Rigidity: M is isometric to the round sphere of radius 1/√k