the Rauch comparison theorem
/ ROWK /
If you know your curved space is everywhere curving at least as much as a sphere, geodesics in it must spread no faster than they do on the sphere; if it curves at most as much as a flat plane, they spread at least as fast as in the plane. The Rauch comparison theorem makes this 'more curvature, slower spreading' principle into a precise inequality. It compares the growth of a Jacobi field on your manifold against the growth of the corresponding Jacobi field on a constant-curvature model space, and so converts a curvature bound directly into a bound on how distances behave.
Precisely, in its cleanest form: let M and M-bar be Riemannian manifolds and gamma, gamma-bar unit-speed geodesics in each, with Jacobi fields J, J-bar that both vanish at the start (J(0) = 0) and have the same initial derivative speed |J'(0)| = |J-bar'(0)|. Suppose the sectional curvatures along the geodesics satisfy K_M >= K_{M-bar} (M is more curved). Suppose also no conjugate point of gamma-bar has yet occurred on the interval. Then |J(t)| <= |J-bar(t)| for all t up to the first conjugate point — more curvature means slower-growing Jacobi fields, hence faster reconvergence and earlier conjugate points. The proof compares the index forms via a clever monotonicity of the ratio <J', J> / <J, J>, a Riccati/Sturm comparison in disguise. Specializing M-bar to a model of constant curvature k gives explicit bounds: |J(t)| <= |J'(0)| times s_k(t), where s_k is sin, t, or sinh according to the sign of k.
Rauch is the analytic heart of comparison geometry — almost every global theorem in this field passes through it. It implies conjugate points come no later under a lower curvature bound (driving Bonnet-Myers), and combined with integration it yields Bishop-Gromov volume comparison and, in its triangle-shaped consequence, Toponogov's theorem. Honest caveats. First, the inequality is one-directional and requires the no-conjugate-point hypothesis on the model side up to t; past a conjugate point the comparison breaks. Second, the cleanest statements assume the lower or upper bound holds along the geodesic, not just at a point — a pointwise bound at the start is not enough. Third, sign conventions for K and R must match between the two manifolds, or the inequality flips.
Compare a Jacobi field on a manifold M with K_M >= 1 against the unit sphere (K = 1). On the sphere the model field with J(0) = 0, |J'(0)| = 1 is |J-bar(t)| = sin t. Rauch then forces |J(t)| <= sin t on M up to t = pi, so the field on M must vanish (a conjugate point) no later than the model's vanishing at t = pi. That is exactly the Bonnet-Myers diameter conclusion in miniature: K >= 1 forces a conjugate point by distance pi, hence diameter <= pi. Flip the bound — if K_M <= 0, compare against flat R^n where |J-bar(t)| = t, giving |J(t)| >= t, so Jacobi fields grow at least linearly and never vanish: no conjugate points, the Cartan-Hadamard input.
Rauch turns a curvature bound into a Jacobi-field bound: K up forces |J| down, conjugate points earlier.
Rauch compares Jacobi fields, not distances directly. Going from the field inequality to a triangle (Toponogov) or volume (Bishop-Gromov) statement requires extra integration; the raw theorem is local-to-a-geodesic, not global.