JOVANA
Explore Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Positive Curvature: Bonnet-Myers, Synge & the Bochner Technique

Positive curvature forces a manifold to be small: this is where local bending becomes a global verdict. We turn the index form from Guide 2 into the Bonnet-Myers diameter bound and Synge's theorem, then meet the Bochner technique, a wholly different engine that proves topology vanishes by integrating a clever identity.

The thread from Guide 2: the index form is the weapon

Guides 1 and 2 of this rung built the two tools we now fire. Guide 1 gave completeness and Hopf-Rinow: in a complete manifold any two points are joined by a minimizing geodesic, so distances are realized, not merely approached. Guide 2 gave the second variation of arc length — the index form I(V, V) — together with Jacobi fields and conjugate points: a geodesic stops being length-minimizing the moment it passes a conjugate point, because beyond there the index form goes negative and a nearby curve is genuinely shorter. The whole story of positive curvature is this: positivity makes the index form negative quickly, which forces conjugate points to appear soon, which caps how long a minimizing geodesic can be.

Recall the exact form of the weapon. Along a unit-speed geodesic gamma, for a vector field V vanishing at both endpoints, the index form is I(V, V) = integral of ( |nabla_t V|^2 - g(R(V, gamma')gamma', V) ) dt, where the curvature term carries the geometry. The sign convention here is that positive sectional curvature K makes the term g(R(V, gamma')gamma', V) positive — so it is SUBTRACTED, dragging the index form down. Sources differ on the sign of R; we fix ours so that the round sphere has K = +1 and the curvature term helps create conjugate points. State your convention before quoting any inequality, because the opposite choice flips every sign in what follows.

Bonnet-Myers: a Ricci lower bound makes the manifold small

Here is the first great theorem. Bonnet-Myers says: if (M, g) is complete and its Ricci curvature satisfies Ric >= (n-1)/r^2 g for some constant r > 0 — a uniform positive lower bound — then M is compact, its diameter is at most pi r, and its fundamental group pi_1(M) is finite. Three conclusions from one hypothesis. Notice it is Ricci, not full sectional, curvature that is bounded below: Ricci is an AVERAGE of sectional curvatures over directions, so the hypothesis is weaker and the theorem correspondingly stronger. The slogan 'positive curvature means compact' is only correct with this precise Ricci bound; drop the uniformity and the conclusion fails, as flat R^n (Ric = 0, noncompact) shows.

Why is it true? Run the index form on a minimizing geodesic gamma of length L. Build n-1 test fields by parallel-transporting an orthonormal frame perpendicular to gamma' and scaling each by sin(pi t / L), so they vanish at both ends. Summing I over those n-1 fields, the |nabla_t V|^2 terms add up to (n-1)(pi/L)^2 times an integral of cos^2, while the curvature terms sum to exactly the Ricci curvature Ric(gamma', gamma') in the direction of travel, weighted by sin^2. The Ricci hypothesis bounds that sum from below; if L exceeds pi r the total second variation becomes NEGATIVE, meaning gamma was not minimizing after all — contradiction. So no minimizing geodesic is longer than pi r, hence diam(M) <= pi r, and a complete bounded manifold is compact.

Synge: positive sectional curvature constrains topology by parity

Synge's theorem uses the index form on a CLOSED geodesic instead of a minimizing arc, and its conclusion is sensitive to dimension parity — a genuinely surprising twist. The statement: if M is compact with positive sectional curvature K > 0, then if dim M is even and M is orientable, M is simply connected (pi_1 = 0); and if dim M is odd, M is orientable. Two clean topological verdicts squeezed out of one curvature sign, split by whether the dimension is even or odd. The even sphere S^2 (simply connected) and the odd real projective space RP^3 (orientable) are exactly the boundary cases this theorem is built to explain.

The proof is a small gem. Suppose pi_1 were nontrivial; in a compact manifold each free homotopy class of loops contains a shortest one, which must be a closed geodesic gamma. Now take parallel transport once around gamma: it is a rotation of the perpendicular space (an orthogonal map of an (n-1)-dimensional space) preserving gamma'. When n is even and M orientable, this rotation of an ODD-dimensional perpendicular space must fix some nonzero vector V — a parallel field along gamma. Feed that V into the index form: |nabla_t V|^2 = 0 because V is parallel, while the curvature term -g(R(V, gamma')gamma', V) is strictly negative because K > 0. So I(V, V) < 0 — the closed geodesic can be shortened, contradicting that it was shortest in its class. Hence no nontrivial class exists: pi_1 = 0.

The Bochner technique: a different engine entirely

Bonnet-Myers and Synge are variational — they argue by perturbing geodesics. The Bochner technique reaches conclusions of the same flavor (positive curvature kills topology) by an utterly different route: it integrates a pointwise identity. The starting point is the Bochner-Weitzenbock formula, which relates two natural Laplacians on a tensor or form. For a 1-form omega it reads, schematically, (Hodge Laplacian) omega = (rough/connection Laplacian) omega + Ric(omega), where the curvature enters as the Ricci term acting algebraically. Two second-order operators differ by exactly a curvature term — that is the whole content, and it is an identity, true at every point, no geodesics in sight.

Bochner-Weitzenbock for a 1-form omega on (M^n, g):

    Delta_H omega  =  nabla* nabla omega  +  Ric(omega)

    Delta_H        = Hodge Laplacian  (d delta + delta d)
    nabla* nabla    = connection (rough) Laplacian, >= 0 after integrating
    Ric(omega)     = Ricci curvature acting on omega  (algebraic, 0th order)

Pair with omega and integrate over a CLOSED M:

    integral |nabla omega|^2  +  integral Ric(omega, omega)  =  integral |Delta_H-part|

If omega is harmonic (Delta_H omega = 0) and Ric > 0:
    0  =  integral |nabla omega|^2  +  integral Ric(omega,omega)
    both terms >= 0  ==>  both = 0  ==>  omega = 0
The Bochner-Weitzenbock identity splits the Hodge Laplacian into a nonnegative connection term plus a Ricci term; integrating against a harmonic form forces both to vanish when Ric > 0.

Watch the engine turn. Suppose omega is a harmonic 1-form on a compact M, so Delta_H omega = 0. Pair the Bochner formula with omega and integrate over M. The Hodge term vanishes (harmonic), and integration by parts turns the connection-Laplacian term into integral |nabla omega|^2 >= 0, a non-negative quantity. What is left is 0 = integral |nabla omega|^2 + integral Ric(omega, omega). If Ric > 0 everywhere, the second integral is strictly positive unless omega = 0 — but the sum is zero and the first term is also non-negative, so BOTH must vanish, forcing omega = 0. Conclusion: a compact manifold with Ric > 0 has no nonzero harmonic 1-forms, hence its first Betti number b_1 = 0 (by Hodge theory). The topology is gone, proven without ever moving a curve.

Two engines, one moral — and honest limits

Step back and compare the two machines, because the contrast is the lesson. The variational engine (Bonnet-Myers, Synge) perturbs geodesics and reads off the SECOND variation; positive curvature makes the index form negative, so candidate minimizers fail and topology is squeezed out geometrically. The Bochner engine integrates a curvature identity against harmonic objects; positive curvature makes a sum of squares plus a positive term equal zero, so the harmonic object must vanish and topology drops out analytically. Same moral — positive curvature is a strong constraint that forbids large fundamental groups and rich cohomology — reached by two independent routes. Knowing both is what lets you choose the right tool for a given theorem.

Now the honest caveats, the ones easy to lose in the excitement. First, these are upper-bound results on topology, not classifications: Bonnet-Myers says pi_1 is finite, not which finite group; Bochner says b_1 = 0, not the whole cohomology. Second, the hypotheses are sharp and unforgiving — Bonnet-Myers needs a UNIFORM Ricci lower bound (pointwise Ric > 0 is not enough; the paraboloid has positive curvature everywhere yet is noncompact because the curvature decays), and Synge needs both compactness and the parity-orientability package. Third, what counts as 'positive curvature' splits: Ricci-positive is much weaker than sectional-positive, and theorems are not interchangeable between them. The famous open ground beyond this guide — which manifolds admit positive sectional curvature, still largely unknown beyond a short list — is a reminder that we are charting the easy implications, not the full classification.

  1. Reach for Bonnet-Myers when you have a uniform Ricci lower bound and want compactness, a diameter cap of pi r, or finiteness of pi_1 — the proof is the index form summed over a perpendicular frame.
  2. Reach for Synge when curvature is positive SECTIONAL and the manifold is compact, and you care about pi_1 (even dim, orientable) or orientability (odd dim) — the proof is the index form on a shortest closed geodesic.
  3. Reach for the Bochner technique when you want to kill harmonic forms (vanishing Betti numbers) under a Ricci or curvature-operator sign — the proof integrates the Bochner-Weitzenbock identity against a harmonic object.
  4. Guide 4 flips every sign to the nonpositive side (Cartan-Hadamard, where curvature K <= 0 makes the universal cover diffeomorphic to R^n), and Guide 5 builds the full comparison machinery — Rauch, Toponogov, the sphere theorem, and splitting.