Metric geometry
Metric geometry is the study of geometric properties that depend solely on distance — not on angles, coordinates, or smooth structure. Where differential geometry demands that spaces be infinitely differentiable manifolds, and where algebraic geometry studies the zero sets of polynomial equations, metric geometry asks a more primitive question: what can we know about a space if all we are given is a way to measure the distance between any two points? The answer, it turns out, is far more than one might expect. Entire theories of curvature, dimension, volume, and convergence can be built from the metric alone, without ever invoking derivatives, tangent spaces, or charts.
The field was given its modern form by Mikhail Gromov in the 1980s, though its roots extend back to the work of Karl Menger, Georges Bouligand, and Alexandr Alexandrov on length spaces and convex surfaces. Gromov's central insight was that many of the deepest theorems of Riemannian geometry — the Bishop-Gromov volume comparison, the splitting theorem, the compactness theorem — are not really theorems about smooth manifolds at all. They are theorems about metric spaces that satisfy certain synthetic conditions, and the smooth structure is merely one way of producing those conditions. This reframing turned metric geometry from a collection of inequalities into a unified research program: to reconstruct the theorems of classical geometry on the minimal possible analytic foundation.
The Gromov-Hausdorff Distance
The foundational tool of metric geometry is the Gromov-Hausdorff distance, which measures how far two metric spaces are from being isometric. Unlike the Hausdorff distance, which compares subsets of a common ambient space, the Gromov-Hausdorff distance compares abstract metric spaces directly by embedding them isometrically into a third space and measuring the Hausdorff distance between the images. Two spaces have distance zero if and only if they are isometric; a sequence of spaces converges in the Gromov-Hausdorff sense if they become arbitrarily close to a limiting space.
This notion of convergence is extraordinarily powerful. It allows sequences of smooth Riemannian manifolds to converge to singular spaces — spaces with corners, cones, and creases that are not manifolds at all. Cheeger-Colding theory shows that when a sequence of Riemannian manifolds with uniform lower Ricci curvature bounds converges in the Gromov-Hausdorff topology, the limit space retains remarkable analytic and geometric structure: it has a well-defined dimension, a measure, and a cotangent structure that supports a first-order calculus. The smooth structure may have collapsed, but the metric structure survives.
Alexandrov Spaces and Synthetic Curvature
A second pillar of metric geometry is the theory of Alexandrov spaces — metric spaces that satisfy curvature bounds in the sense of triangle comparison. In a Riemannian manifold with sectional curvature bounded below by κ, every sufficiently small geodesic triangle is fatter than its comparison triangle in the model space of constant curvature κ. Alexandrov's radical move was to turn this comparison property into a definition: a metric space is an Alexandrov space if every geodesic triangle satisfies the appropriate comparison inequality, regardless of whether the space is a manifold.
This definition captures curvature without smoothness. The space of directions at a point in an Alexandrov space — the metric space of geodesic directions — is itself an Alexandrov space of one lower dimension, and this recursive structure gives Alexandrov spaces a rich local theory. Perelman's proof of the geometrization conjecture relied crucially on Alexandrov space techniques: the Ricci flow with surgery produces singular limit spaces that are not manifolds but are Alexandrov spaces, and the analysis of these spaces is what makes the surgery argument rigorous.
Metric Measure Spaces and Optimal Transport
The most recent and perhaps most surprising development in metric geometry is the theory of metric measure spaces with synthetic Ricci curvature bounds. In 2006, John Lott and Cédric Villani, and independently Karl-Theodor Sturm, defined what it means for a metric measure space to have Ricci curvature bounded below — using not differential geometry but the theory of optimal transport. A space has Ricci curvature bounded below by K if the entropy functional along geodesics in the space of probability measures satisfies a certain convexity inequality. This definition makes sense on any metric measure space, including discrete spaces, fractals, and infinite-dimensional spaces.
The resulting theory, synthetic Ricci curvature, has proved extraordinarily robust. The Lott-Villani-Sturm definition implies sharp geometric inequalities — Brunn-Minkowski, Bishop-Gromov, spectral gap — that were previously known only for smooth Riemannian manifolds. And it has found applications far outside geometry: in probability theory, in the study of Markov chain mixing times, in machine learning through Wasserstein geometry, and even in economics, where optimal transport provides a natural framework for matching problems.
The triumph of metric geometry is not that it generalizes Riemannian geometry to worse spaces. It is that it reveals Riemannian geometry to have been a special case all along — a special case whose smooth structure obscured the deeper metric principles that governed it. The insistence that geometry requires differentiability is not a mathematical truth but a historical accident: we had the machinery of calculus before we had the machinery of metric comparison, and so we built geometry on the foundations we knew. Metric geometry is the correction of that historical bias. The claim that a space must be smooth to have curvature is not mathematics — it is a failure of imagination disguised as rigor.