Two Laplacians that disagree, and the disagreement is curvature
The previous guide handed you the Hodge Laplacian Delta = d d* + d* d acting on differential forms — a beautiful, intrinsically-defined, nonnegative operator whose kernel is the harmonic forms and hence the cohomology. But there is a second, more naive Laplacian sitting on the same forms: the connection Laplacian (or rough Laplacian) nabla* nabla, built by covariantly differentiating twice and tracing. On functions these two agree (up to sign), but on 1-forms and higher they do NOT, and the entire payoff of this guide lives in that gap. The Bochner-Weitzenböck formula measures it precisely, and the measurement is a curvature term.
On 1-forms (equivalently, dual vector fields), the Weitzenbock identity reads:
Delta_Hodge = nabla* nabla + Ric
d d* + d* d = (rough Laplacian) + (Ricci curvature, acting as a symmetric endomorphism)
For a function f (a 0-form) the two Laplacians already agree, but on df it gives:
Delta(df) = d(Delta f), and pairing with df and integrating yields
(1/2) Delta |df|^2 = |Hess f|^2 + g( grad f , grad Delta f ) + Ric(grad f, grad f)
The last term, Ric(grad f, grad f), is the whole point: curvature has entered for free.Why must curvature appear here, with the same inevitability you saw in the second variation last rung? The reason is identical. The rough Laplacian and the Hodge Laplacian both differentiate twice, but in a different ORDER — one traces the second covariant derivative, the other antisymmetrizes through d and d*. Reconciling the two orders forces you to commute a pair of covariant derivatives, and the commutator of covariant derivatives is, by definition, the Riemann curvature tensor. On 1-forms the relevant trace of that tensor is exactly the Ricci curvature. So Ric is not bolted on; it is the unavoidable price of asking two natural Laplacians to be the same.
The Bochner technique: how an identity becomes a theorem
The formula by itself is just algebra; the Bochner technique is the recipe that squeezes geometry out of it. The move is a maximum-principle argument that becomes especially clean by integration on a closed manifold. Suppose omega is a harmonic 1-form, so Delta_Hodge omega = 0. Plug that into the Weitzenböck identity, pair with omega, and integrate over M. The Hodge term integrates to zero; integration by parts turns the rough-Laplacian term into the nonnegative quantity integral over M of |nabla omega|^2; and the curvature term becomes integral over M of Ric(omega, omega). You are left with a single equation balancing two integrals.
- Start from the identity on a CLOSED (compact, no boundary) manifold M: for a harmonic 1-form, 0 = integral over M of g(Delta_Hodge omega, omega) = integral over M of [ |nabla omega|^2 + Ric(omega, omega) ].
- Impose the curvature hypothesis Ric >= 0. Then both integrands are nonnegative, yet their sum integrates to zero — so each must vanish pointwise: |nabla omega|^2 = 0 AND Ric(omega, omega) = 0 everywhere.
- Read off the conclusion: nabla omega = 0 means omega is PARALLEL. A closed manifold with Ric >= 0 therefore admits only parallel harmonic 1-forms, and the number of independent ones is at most n = dim M.
- Tighten it to Ric > 0 (strictly): now Ric(omega, omega) = 0 forces omega = 0 outright. So there are NO nonzero harmonic 1-forms — and by the de Rham theorem the first Betti number b_1(M) = 0.
Stand back and admire what just happened: a hypothesis about CURVATURE (a local, analytic-geometric quantity) forced a conclusion about TOPOLOGY (the Betti number, a global homotopy invariant). This is the first Bochner vanishing theorem, and it is the prototype for a whole industry — by running the same argument on harmonic spinors, harmonic forms of higher degree, or holomorphic sections, one converts positivity of a curvature into the vanishing of a cohomology group. Lichnerowicz's vanishing for the Dirac operator under positive scalar curvature, and even the Kodaira vanishing theorem in complex geometry, are Weitzenböck arguments wearing different clothes.
Eigenvalue estimates: hearing curvature in the spectrum
The very same Bochner identity, applied not to a harmonic form but to an eigenfunction, controls the SPECTRUM of the Laplace-Beltrami operator. On a closed M the Laplacian Delta f = -div(grad f) has a discrete sequence of nonnegative eigenvalues 0 = lambda_0 < lambda_1 <= lambda_2 <= ..., the resonant frequencies of the manifold thought of as a drum. The first nonzero eigenvalue lambda_1 is the fundamental tone, and curvature bounds pin it down. The Lichnerowicz estimate is the cleanest: if Ric >= (n-1) k g for a constant k > 0, then lambda_1 >= n k.
The proof is the Bochner recipe again, run on an eigenfunction. Take Delta f = lambda_1 f, feed it into the pointwise identity (1/2) Delta |grad f|^2 = |Hess f|^2 + g(grad f, grad Delta f) + Ric(grad f, grad f), and integrate over M. The left side integrates to zero (it is a divergence). The cross term gives -lambda_1 integral of |grad f|^2 from the eigenvalue equation. Bound the Hessian term from below using the Cauchy-Schwarz inequality |Hess f|^2 >= (Delta f)^2 / n, and substitute Ric >= (n-1) k. Rearranging the surviving inequality is exactly lambda_1 >= n k. The fundamental frequency of the drum cannot dip below a floor set by Ricci curvature.
Pointing the Laplacian forward in time: the heat flow
Everything so far has been STATIC — one fixed metric, one fixed operator. The final idea of the rung sets things in motion. The geometric heat flow is the evolution equation (d/dt) u = -Delta u (with the analyst's sign, so it smooths), the manifold version of the ordinary heat equation. Decompose any initial function in the eigenbasis of the Laplacian; under the flow each eigenmode decays like e^(-lambda t), so high-frequency wrinkles (large lambda) die fast and the solution relaxes toward its average. This is why the eigenvalue estimates of the last section matter dynamically: lambda_1 is precisely the slowest decay rate, the rate at which a manifold forgets its initial conditions and returns to equilibrium.
The same heat-flow philosophy upgrades harmonic maps from the earlier guide into a dynamical method. The Eells-Sampson theorem flows an arbitrary map f: M -> N down its Dirichlet energy by the harmonic map heat flow (d/dt) f = tau(f), where tau is the tension field (the gradient of energy). Provided the target N has nonpositive sectional curvature, a Bochner-type identity for the energy density keeps the flow from blowing up, and it converges as t -> infinity to a genuine harmonic map homotopic to the original. Note the curvature hypothesis on the TARGET — drop it and the flow can develop singularities (bubbling), an honest limitation, not a technicality.
Flowing the metric itself: a first look at Ricci flow
The boldest step is to flow not a function or a map but the METRIC. Hamilton's Ricci flow is the equation (d/dt) g_ij = -2 Ric_ij — let the metric evolve in the direction opposite its own Ricci curvature. Read it as a heat equation for the metric: in harmonic coordinates the leading term of -2 Ric is essentially the Laplacian of g, so curvature diffuses, hot spots (high curvature) cool, and the geometry tries to homogenize. On a round sphere the flow shrinks it uniformly to a point; on a surface it conformally deforms any metric toward a constant-curvature one, giving a flow-based proof of the uniformization theorem.
The dream is to use the flow to simplify topology: start with any metric on a 3-manifold, run the flow, and let it relax the geometry into recognizable constant-curvature pieces, reading off the geometrization conjecture and with it the Poincaré conjecture. The obstruction is that the flow develops SINGULARITIES — a neck pinch, where a thin tube shrinks to zero radius in finite time and curvature blows up. Perelman's monumental work analyzes these singularities, performs surgery to cut them out and restart the flow, and tracks a monotone quantity (the Perelman entropy) that controls the process — completing Hamilton's program.