# Differential geometry Differential geometry turns the tools of [[Calculus]] loose on curved spaces: it equips a [[Manifold]] with a metric [[Tensor]], differentiates along it with a connection, and reads off curvature — the single invariant from which most of the subject's power flows. Founded on Gauss's study of surfaces (1827) and Riemann's leap to n dimensions (1854), it supplies the [[Geodesic|geodesics]] and curvature tensors of [[General_relativity]], the [[Symplectic_manifold|symplectic manifolds]] beneath [[Hamiltonian_mechanics]], the [[Lie_group|Lie groups]] of continuous symmetry, the Fisher metric that turns families of [[Probability_distribution|probability distributions]] into geometric objects, and the curvature-aware [[Mathematical_optimization|optimization]] now routine in [[Machine_learning]]. Its recurring lesson is local-to-global: infinitesimal data — how fast neighboring [[Geodesic|geodesics]] spread — integrates into global facts about shape, [[Topology]], and dynamics. ## Gauss's egregious theorem: curvature is intrinsic Gauss defined the curvature K of a surface from how it bends in space, then proved the *Theorema Egregium* (1827): K is computable entirely from distances measured *within* the surface. An ant with a tape measure can detect its world's curvature — no outside view required. A sphere of radius R has K = 1/R² everywhere; a triangle drawn on any surface has angle sum π plus the integral of K over its interior, so geodesic triangles bulge fat on spheres and pinch thin on saddles. One corollary reaches every atlas and [[Geographic_information_system]]: the plane has K = 0 and the sphere does not, so no flat map of Earth preserves distances — every projection must cheat somewhere. The theorem's intrinsic turn — geometry as measured from inside, not as embedded from outside — is the founding move of the modern subject, and of [[General_relativity|relativity]] after it. ## Riemann's leap: metrics on n-dimensional manifolds Riemann's 1854 habilitation lecture generalized Gauss to any dimension: a [[Manifold]] — a space locally chartable like ℝⁿ, per the framework [[Topology]] later made precise — carries a metric [[Tensor]] g giving each tangent space an inner product from [[Linear_algebra]], hence lengths ds² = g_ij dxⁱdxʲ, angles, and volumes. Christoffel (1869) and Levi-Civita (1917) supplied the connection: a rule for parallel-transporting vectors and differentiating fields, from which [[Geodesic|geodesics]] emerge as the locally shortest paths — extremals of length in the sense of the [[Calculus_of_variations]]. Curvature then measures the failure of parallel transport around a loop to return a vector unchanged, and the Jacobi equation — a linear [[Ordinary_differential_equation|differential equation]] along each geodesic — makes it observable: positive curvature focuses nearby geodesics together, negative curvature spreads them exponentially. All of it is coordinate-free — statements about the [[Geometry|geometric]] object, not about anyone's chart. ## Where curvature meets topology The Gauss–Bonnet theorem is the subject's cleanest local-to-global statement: for a closed [[Surface_(topology)|surface]], ∫K dA = 2πχ, where the Euler characteristic χ = 2 − 2g depends only on the [[Genus_(mathematics)|genus]] g. Total curvature is a topological constant — deform a sphere however you like, and the bumps and dents cancel to exactly 4π. Chern extended the identity to higher dimensions (1944), and the theme now organizes whole fields: curvature bounds constrain [[Topology|topological]] type, and geometric flows drive metrics toward canonical form. Hamilton's Ricci flow (1982), a [[Partial_differential_equation]] that evolves the metric by its own curvature the way heat [[Diffusion|diffuses]], culminated in Perelman's 2003 proof of the Poincaré conjecture — the century-old question [[Henri_Poincaré]] posed in 1904 — by flowing arbitrary 3-manifolds until their geometric pieces became recognizable. ## The physics dividend: relativity and mechanics Einstein's field equations (1915) made curvature physical: matter and energy determine the curvature of spacetime, and freely falling bodies follow its [[Geodesic|geodesics]] — [[Gravity]] is geometry, not force. Schwarzschild's solution (1916) described the [[Gravitational_field|field]] outside a spherical mass and, pushed inward, the [[Black_hole]]; geodesic deviation *is* tidal force; and GPS receivers correct for roughly 38 μs per day of net relativistic clock drift, curvature bookkeeping running silently in every phone. Mechanics is equally geometric: [[Hamiltonian_mechanics]] lives on a [[Symplectic_manifold]] — [[Phase_space]] with its canonical 2-form — where flows preserve volume by [[Liouville's_theorem_(Hamiltonian)]], and continuous symmetries under a [[Lie_group]] action yield conserved quantities by Noether's theorem (1918). Even [[James_Clerk_Maxwell|Maxwell's]] electromagnetism re-enters as geometry: the field is the curvature of a connection, the template for the gauge theories of [[Mathematical_physics]]. ## Curvature as a driver of dynamics Differential geometry and [[Dynamical_system|dynamical systems]] share a workhorse: the geodesic flow, motion by pure inertia on a curved space, unfolding in [[Phase_space]]. On negatively curved surfaces the flow is the archetype of [[Chaos_theory|chaos]] — Hadamard saw the sensitivity in 1898, and Anosov (1967) made such flows the model hyperbolic systems: nearby trajectories diverge exponentially at rates set by curvature, mixing follows, and time averages converge to space averages. That is [[Ludwig_Boltzmann|Boltzmann's]] ergodic hypothesis given honest geometric hypotheses, the bridge by which [[Statistical_mechanics]] borrows theorems from [[Geometry]]. The [[Nonlinear_system|nonlinear]] stability toolkit runs the same circuit in reverse: curvature-like quantities certify contraction, and Lyapunov-style arguments measure how perturbations grow along flows — geometry supplying the estimates, dynamics spending them. ## The geometry of information and control Statistics acquired a metric in 1945, when C. R. Rao observed that Fisher information makes a family of [[Probability_distribution|probability distributions]] a Riemannian [[Manifold]]: distinguishability *is* distance, and the Kullback–Leibler divergence of [[Information_theory]] is locally one-half the squared Fisher length. Amari's natural gradient (1998) descends loss surfaces along this metric rather than a naive coordinate one, an idea threaded through modern [[Machine_learning]] and [[Neural_network_(machine_learning)|neural-network]] training, where the manifold hypothesis treats data itself as concentrated near low-dimensional submanifolds. [[Control_theory]] speaks the same language: control-affine systems steer along vector fields, and the Chow–Rashevskii theorem (late 1930s) says brackets of available directions can span the rest — the theorem behind parallel-parking maneuvers in [[Robotics]] — while the Pontryagin maximum principle (1956) runs [[Optimal_control]] through Hamiltonian geometry, closing the loop back to mechanics. **On the spine:** [[Manifold]] · [[Geodesic]] · [[Tensor]] · [[Symplectic_manifold]] · [[General_relativity]]. ## Wikipedia : Wikitube **Strict pair:** [Wikipedia](https://en.wikipedia.org/wiki/Differential_geometry) : [Wikitube](https://en.wikitube.io/wiki/Differential_geometry) ## Previous hub tags Hubs: `Systems`. Portals: [[PORTAL_Graph_theory]], [[PORTAL_Decision_theory]], [[PORTAL_Information_theory]], [[PORTAL_Control_theory]], [[PORTAL_Operations_research]]. --- *Repopulated 2026-08-12 · redlink fill · 0 deletions.*