Weinstein's theorem
/ WINE-stine /
On a positively curved space, a rigid motion (an isometry) cannot wander too freely: if it preserves orientation and the dimension is right, it is forced to leave at least one point exactly where it was. Weinstein's theorem is this fixed-point guarantee. It is the companion to Synge's theorem, born from the same second-variation-of-energy mechanism, and it shows that positive curvature constrains not just the manifold's shape but the behaviour of every symmetry the manifold can carry.
Precisely: let M be a compact, oriented Riemannian manifold with positive sectional curvature K > 0, and let f : M -> M be an isometry. If M is even-dimensional and f preserves orientation, then f has a fixed point; if M is odd-dimensional and f reverses orientation, then f has a fixed point. The proof is the same gem that powers Synge. Suppose f has no fixed point; then the displacement function p -> d(p, f(p)) attains a positive minimum at some point q, and the geodesic gamma from q to f(q) is a shortest connector. Parallel transport along gamma composed with the differential of f is an isometry of the tangent space whose orientation behaviour (in the even+orientation-preserving or odd+orientation-reversing case) forces a fixed unit vector; the second variation of energy of gamma in that direction is strictly negative because K > 0, producing a strictly shorter displacement nearby and contradicting minimality. So a fixed point must exist.
Weinstein's theorem is a clean illustration of how curvature controls dynamics and symmetry, and historically it sits beside Synge's theorem as the second half of the 'positive curvature is rigid' package. Honest caveats. First, every hypothesis is sharp and the parity pairing is essential — an orientation-preserving isometry of an ODD-dimensional positively curved manifold need NOT have a fixed point (a rotation of the round sphere S^3 with no fixed point is orientation-preserving), and likewise the even-dimensional orientation-reversing case can be fixed-point-free (the antipodal map on S^2 reverses orientation and has no fixed point). Second, K > 0 (sectional) is needed; weaker curvature bounds do not suffice. Third, it is a compact-manifold statement — on noncompact spaces the displacement minimum may not be attained, and the argument collapses.
Any orientation-preserving isometry of the round 2-sphere S^2 (even-dimensional, K > 0) must fix a point — a rotation of the sphere about an axis fixes the two poles, as predicted. The sharpness of the parity is shown by the antipodal map x -> -x on S^2: it IS an isometry of a positively curved even-dimensional manifold but it REVERSES orientation (in even dimension the antipodal map is orientation-reversing), so it falls outside the even+orientation-preserving hypothesis — and indeed it has no fixed point. On S^3 (odd), a free orientation-preserving rotation likewise escapes the theorem and has no fixed point.
Even + orientation-preserving + K>0 forces a fixed point; the antipodal map (orientation-reversing) escapes and has none.
The parity-orientation pairing is not interchangeable: even needs orientation-PRESERVING, odd needs orientation-REVERSING. Swap them and the theorem is false, as free rotations of spheres show.