Toponogov's comparison theorem
/ toh-poh-NOH-gof /
Draw a triangle whose sides are shortest paths on a curved surface. On a positively curved surface like a sphere the triangle looks 'fat' — its angles add up to more than 180 degrees and it bulges; on a flat plane it is the usual straight triangle. Toponogov's theorem makes this comparison rigorous and global: it pins the geometry of any geodesic triangle on your manifold against the corresponding triangle drawn in a constant-curvature model, turning a curvature lower bound into hard inequalities on angles and on the distances between points.
Precisely: suppose a complete manifold M has sectional curvature K >= k everywhere. Take a geodesic triangle in M with vertices p, q, r and side lengths a, b, c (the sides being minimizing geodesics). Build the comparison triangle with the same three side lengths in the model space of constant curvature k (the sphere of radius 1/sqrt(k) if k > 0, the plane if k = 0, hyperbolic space if k < 0). Then each angle of the M-triangle is at least the corresponding angle of the model triangle. Equivalently (the 'hinge' version), if you fix two sides and the angle between them, the opposite side in M is no longer than the model's: lower curvature bound means triangles are at least as fat as the model. The proof globalizes the infinitesimal Rauch estimate by an arc-by-arc argument along the sides, using completeness to keep minimizing geodesics available.
Toponogov is the global triangle backbone of comparison geometry and the technical engine behind the sphere theorem and the structure theory of nonnegatively curved spaces. Its great virtue over Rauch is that it is a statement about actual distances and angles of finite-size triangles, not infinitesimal Jacobi fields, so it survives the passage to limits — which is exactly why it generalizes to Alexandrov spaces with curvature bounded below, the singular-space version studied in metric geometry. Honest caveats. First, the theorem needs a global lower bound K >= k along the whole region, not a pointwise estimate, and completeness to guarantee minimizing sides. Second, there is a dual upper-bound version (sides thinner than the model) but it requires extra convexity/injectivity-radius hypotheses and is closer to the CAT(k) story; do not assume the fat and thin statements are symmetric.
On the unit sphere (K = 1 >= 1, so model = sphere) consider a triangle with all three vertices and right angles — e.g. one vertex at the north pole and two on the equator a quarter-circle apart. Its angle sum is 270 degrees, fatter than the 180 of any flat triangle, exactly as Toponogov predicts for K >= 0 (compare against k = 0): every angle of a triangle on a nonnegatively curved space is at least the angle of the flat triangle with the same side lengths. Concretely, for the hinge form: fix two unit-length sides meeting at a right angle; on the sphere the third side is SHORTER than the sqrt(2) it would be in the plane, because positive curvature pulls the far ends together.
Lower curvature bound = fat triangles: on the sphere a right-angle hinge closes tighter than in the plane.
Toponogov is a GLOBAL statement requiring K >= k everywhere along the region; it is not a pointwise fact. Its survival under Gromov-Hausdorff limits is what makes Alexandrov geometry possible — but that singular-space theory lives in metric geometry, not here.