The branch of Riemannian and metric geometry that deduces geometric and topological properties of a space from upper or lower curvature bounds by comparing it to model spaces of constant sectional curvature.

Principle

If a space satisfies pointwise or integral curvature bounds (upper or lower), then metric and topological invariants can be estimated by corresponding invariants in constant-curvature model spaces via comparison inequalities.

Demonstration

A classical demonstration is the Toponogov triangle comparison: triangles in a manifold with sectional curvature bounded below by κ are no thinner than triangles in the model space of constant curvature κ, which yields control of geodesic divergence and diameter.

Misapplication

Applying comparison theorems when curvature bounds fail globally or only hold in an averaged sense can produce incorrect conclusions about geodesic convexity or injectivity radius.

Consequence

Correct use yields quantitative control over distances, angles, volume growth, and topological finiteness results (for example diameter or fundamental group bounds) derived from the model comparisons.

Reversal

The reversal is considering spaces with no curvature bounds or with arbitrarily oscillatory curvature; then model comparison estimates break down and local geometry can behave like many incompatible models.

Boundary

Scope excludes structures without a metric notion of curvature (purely topological spaces), and excludes uses where curvature is defined only distributionally unless the comparison framework is extended to that setting.

Semantic Tension

Tension arises between synthetic comparison approaches (Alexandrov spaces, CAT(κ)) that use triangle-comparison axioms and analytic approaches that use sectional/Ricci curvature inequalities; both aim to control geometry but differ in hypotheses and conclusions.

Synthesis

Comparison geometry unifies analytic curvature bounds and synthetic model comparisons to translate curvature constraints into concrete metric, volumetric, and topological estimates by contrasting the given space with constant-curvature models.