The Bochner Technique: How Ricci Curvature Kills Harmonic Forms

The precise statement

Theorem (Bochner, 1946). Let (Mⁿ, g) be a compact (closed, boundaryless) connected Riemannian manifold whose Ricci curvature is positive definite everywhere: Ric(X, X) > 0 for all nonzero tangent vectors X. Then M carries no nonzero harmonic 1-form. Consequently, by the Hodge theorem, the first de Rham cohomology vanishes, H¹(M; ℝ) ≅ ℋ¹(M) = 0, so the first Betti number b₁(M) = 0.

Here a form ω is harmonic if Δω = 0, where Δ = dδ + δd is the Hodge–de Rham Laplacian, d is exterior derivative, and δ = d* its adjoint. On a closed manifold Δω = 0 is equivalent to the pair dω = 0 and δω = 0 (both hold simultaneously). Hodge theory identifies ℋᵖ(M), the space of harmonic p-forms, with Hᵖ(M; ℝ); Bochner's result forces this space to be trivial for p = 1 when Ric > 0. The weaker hypothesis Ric ≥ 0 already forces every harmonic 1-form to be parallel.

The picture: curvature as a potential well

A harmonic 1-form is a kind of steady, curl-free and divergence-free flow on the manifold — think of the constant-velocity flows on a flat torus, which are exactly its harmonic 1-forms. Positive Ricci curvature refuses to let such a flow persist. Intuitively, Ric > 0 means geodesics that start parallel converge (this is the same focusing that drives the Bonnet–Myers diameter bound), and this focusing acts like a restoring force that damps any nontrivial equilibrium flow down to zero.

The rigorous encoding of this intuition is the identity ½Δ_LB‖ω‖² = ‖∇ω‖² + Ric(ω♯, ω♯) (Δ_LB the Laplace–Beltrami operator, = −(dδ + δd) on functions), valid whenever ω is harmonic (ω♯ is the vector field metric-dual to ω). Read it as an energy balance: the average of the left side over a closed manifold is zero, so the two nonnegative-when-Ric≥0 terms on the right must each average to zero. Curvature is the ingredient that makes the second term have a definite sign, converting an averaged identity into a pointwise obstruction.

Key idea of the proof: the Weitzenböck identity

The engine is the Bochner–Weitzenböck formula on 1-forms: Δ = ∇*∇ + Ric, where Δ is the Hodge Laplacian, ∇*∇ is the (nonnegative) connection/rough Laplacian, and Ric acts on 1-forms via the metric. The two Laplacians differ by exactly the curvature term — that is the entire content. Now suppose Δω = 0. Pair the identity with ω and integrate over the closed manifold:

0 = ⟨Δω, ω⟩ = ⟨∇*∇ω, ω⟩ + ⟨Ric ω, ω⟩. Integration by parts turns ∫⟨∇*∇ω, ω⟩ into ∫‖∇ω‖² ≥ 0 (no boundary term — this is where compactness enters). Hence ∫ ‖∇ω‖² + ∫ Ric(ω♯, ω♯) = 0. If Ric ≥ 0 both integrands are ≥ 0, forcing ∇ω = 0 (ω is parallel) and Ric(ω♯, ω♯) = 0 pointwise. If Ric > 0 strictly, the second condition already forces ω♯ = 0 everywhere, so ω = 0. The pointwise version reads ½Δ_LB‖ω‖² = ‖∇ω‖² + Ric(ω♯, ω♯), where Δ_LB is the (analyst's) Laplace–Beltrami operator, equal to −(dδ + δd) on functions; equivalently −½(dδ + δd)‖ω‖² = ‖∇ω‖² + Ric(ω♯, ω♯). It follows from the same identity plus Δ_LB‖ω‖² = −2⟨Δω, ω⟩ + 2‖∇ω‖² with Δ = dδ + δd (the two Laplacians differ in sign on functions).

Canonical special case: the flat torus and the round sphere

Two examples pin down the sharpness. On the flat torus Tⁿ = ℝⁿ/ℤⁿ we have Ric ≡ 0, so Bochner gives only the borderline conclusion: harmonic 1-forms exist but must be parallel. Indeed ℋ¹(Tⁿ) is spanned by dx₁, …, dxₙ, the constant-coefficient forms, and b₁ = n. Every one is parallel, matching the Ric ≥ 0 rigidity statement exactly — you cannot do better than 'parallel' when curvature only vanishes.

On the round sphere Sⁿ (n ≥ 2), Ric = (n−1)g > 0. Bochner predicts b₁ = 0, and indeed H¹(Sⁿ; ℝ) = 0 — spheres are simply connected. More strikingly, the round metric has positive curvature operator, so the strong Bochner conclusion kills all intermediate harmonic forms: bₚ(Sⁿ) = 0 for 0 < p < n, recovering the homology of a sphere purely from curvature. These two cases bracket the phenomenon: strict positivity ⇒ vanishing; flatness ⇒ parallel forms of maximal dimension.

Why the hypotheses are essential

Compactness is not optional. Drop it and the integration by parts acquires a boundary/decay term that need not vanish, so the argument collapses. Concretely, the hyperbolic plane ℍ² is complete with Ric = −g < 0, yet its space of L² harmonic 1-forms is infinite-dimensional (Dodziuk, Donnelly): completeness alone gives no vanishing without curvature control and integrability. (For ℍⁿ this infinite-dimensionality happens only in the middle degree p = n/2, so among 1-forms only the n = 2 case is affected.) Even with Ric ≥ 0, noncompact manifolds like ℝⁿ have nonparallel harmonic forms once you drop the L² condition. On complete noncompact manifolds one instead needs a Ric bound plus volume/integrability hypotheses (Yau, Gromov).

The curvature sign is essential. Bochner controls only the case Ric ≥ 0; with negative Ricci the term ∫Ric(ω♯,ω♯) becomes ≤ 0 and cannot be trapped, so harmonic forms are unconstrained — every compact hyperbolic manifold has b₁ ≥ 0 with no bound from below. The method generalizes by replacing Ric with the appropriate Weitzenböck curvature operator on p-forms (Gallot–Meyer, Lichnerowicz on spinors), and its sign, not merely Ric, is what governs vanishing for higher-degree forms.

Significance and what it unlocks

Bochner's technique launched an entire industry of vanishing theorems: assume a curvature positivity, write the relevant Weitzenböck identity, integrate, and conclude the vanishing of a cohomology or of the kernel of a natural elliptic operator. Its descendants are load-bearing across geometry. In complex geometry, the Bochner–Kodaira–Nakano identity yields the Kodaira vanishing theorem Hᵍ(X, K_X ⊗ L) = 0 for all q > 0 when L is a positive line bundle — the cornerstone of algebraic-geometric embedding results. In spin geometry, Lichnerowicz's formula D² = ∇*∇ + ¼Scal shows that a spin manifold with positive scalar curvature has no harmonic spinors, forcing the Â-genus to vanish — the origin of all scalar-curvature obstructions and, via the index theorem, of Gromov–Lawson and Schoen–Yau.

The circle of ideas also yields Bonnet–Myers-style finiteness of π₁, Gallot–Meyer's homology-sphere theorem, and Bérard–Besson–Gallot estimates. Every time you see 'positive curvature ⇒ topology is simple,' a Bochner-type integration is usually doing the work underneath.

How the curvature hypothesis controls harmonic forms and topology on a compact manifold (Mⁿ, g)

Hypothesis	Weitzenböck curvature term	Conclusion on harmonic forms	Topological payoff
Ric > 0 (strictly)	Ric(ω♯,ω♯) > 0 unless ω = 0	No nonzero harmonic 1-forms	b₁ = 0; π₁ finite (Myers)
Ric ≥ 0	Ric(ω♯,ω♯) ≥ 0	Every harmonic 1-form is parallel (∇ω = 0)	b₁ ≤ n; equality forces a flat torus (Bochner–Yano)
Curvature operator > 0	Weitzenböck 𝒲 on p-forms > 0	No nonzero harmonic p-forms, 1 ≤ p ≤ n−1	Real homology sphere; bₚ = 0 for 0 < p < n
Scalar curv > 0, spin	¼ Scal on spinors (Lichnerowicz)	No nonzero harmonic spinors	Â-genus = 0; no PSC obstruction met
No curvature sign	Term uncontrolled	Harmonic forms may be abundant	bₚ = Betti number of the manifold (Hodge)