A Mathematical Journey
Where the parallel postulate fails, triangles contain less than 180°, and an infinite world hides inside a finite disk.
In 1830, Nikolai Lobachevsky and János Bolyai independently discovered a geometry where Euclid's fifth postulate is not merely wrong — but richly, consistently, beautifully wrong.
In hyperbolic space, through any point not on a given line there pass infinitely many lines that never meet it. Space curves away from itself at every point. Triangles contain strictly less than 180°. And the farther you travel, the faster the room around you expands.
The Poincaré disk model compresses the entire infinite hyperbolic plane into the open unit disk. Geodesics appear as circular arcs meeting the boundary at right angles — or as diameters. The boundary circle is not really there; it represents points at infinite hyperbolic distance.
The hyperbolic plane ℍ² has constant Gaussian curvature K = −1. The Poincaré disk metric is ds² = 4(dx²+dy²)/(1−|z|²)². Distances are distorted: equal hyperbolic lengths appear shorter the closer they are to the boundary.
Click two points inside the disk to draw a geodesic. Switch to "Show parallels" to see infinitely many lines through a third point that never meet the first geodesic.
Click inside the disk to place points. All arcs are exact geodesics (hyperbolic straight lines).
Riemannian geometry offers exactly three homogeneous, isotropic, simply connected surfaces of constant curvature. Each satisfies its own version of triangle geometry and the parallel postulate.
No parallel lines — every pair of great circles meets. Triangle angle sum exceeds π.
α + β + γ = π + AreaExactly one parallel. Triangle angle sum is exactly π for every triangle.
α + β + γ = πInfinitely many parallels. Triangle angle sum is less than π. Deficit = area.
α + β + γ = π − AreaThe most striking fact in hyperbolic geometry: the area of a triangle is completely determined by its three angles alone — no side lengths needed.
This is the Gauss–Bonnet theorem in action. The angle deficit directly measures area. As a vertex approaches the boundary circle (a point at infinity), its angle approaches zero.
An ideal triangle — all three vertices on the boundary — has all angles equal to zero and achieves the maximum possible area of π. Every ideal triangle is congruent to every other ideal triangle.
In Euclidean geometry, similar non-congruent triangles exist. In hyperbolic geometry this is impossible: equal angles force equal sides. There is an absolute scale — no zoom-in or zoom-out symmetry.
Drag the orange vertices on the Poincaré disk. Watch the angles and area update in real time.
Drag the vertices. Area equals π minus the angle sum.
A regular {p, q} tessellation fills the hyperbolic plane with regular p-gons, q meeting at every vertex. It is hyperbolic when (p−2)(q−2) > 4. While the Euclidean plane admits only three regular tilings, the hyperbolic plane admits infinitely many. M.C. Escher, guided by the geometer H.S.M. Coxeter, based his Circle Limit woodcut series on these patterns.
In Euclidean space, a circle of radius r has circumference 2πr. In hyperbolic space:
For large r, sinh(r) ≈ eʳ/2, so circumference grows exponentially. The area of a hyperbolic disk is A(r) = 2π(cosh r − 1), also asymptotically exponential.
This explains the Poincaré disk picture: each ring of hyperbolic "unit length" must fit exponentially more tiles than the last, so tiles appear exponentially smaller near the boundary.
This exponential growth is the geometric heart of Gromov hyperbolicity, and explains why hyperbolic groups have linear Dehn functions and solvable word problems. The internet, phylogenetic trees, and knowledge graphs all embed naturally in hyperbolic space.
Fixing the topology of a surface — say, a closed genus-3 surface — there are still infinitely many distinct hyperbolic geometries it can carry. The Teichmüller space Tg is the space of all such geometries, carefully organized.
Cut a genus-g surface along 3g − 3 disjoint simple closed geodesics. It decomposes into 2g − 2 pairs of pants — each a sphere with three geodesic boundary circles. Two real numbers parametrize each seam: its length ℓᵢ > 0 (how wide the neck is) and a twist θᵢ ∈ ℝ (by how much we rotate before regluing). This gives the Fenchel–Nielsen coordinates:
For genus 3: 3g−3 = 6 seam geodesics, giving T₃ ≅ ℝ¹² (6 lengths + 6 twists). The surface at right has 4 pairs of pants arranged in a Y: a central junction P₀ with 3 arms, each arm ending in a handle (torus). Seams γ₁, γ₂, γ₃ are the neck circles connecting the central piece to each handle; seams γ₄, γ₅, γ₆ are the circles where each handle is sewn to itself.
The Weil–Petersson symplectic form takes the strikingly clean form ωWP = Σᵢ dℓᵢ ∧ dθᵢ in Fenchel–Nielsen coordinates. Tg becomes a symplectic manifold — the same structure as classical mechanics. Mirzakhani's volume computations integrate polynomials in ℓᵢ² over this symplectic structure.
Drag the sliders below. The top three (ℓ₁–ℓ₃) control the neck widths — as ℓᵢ → 0 the arm pinches off into a cusp. The bottom three (ℓ₄–ℓ₆) control the hole sizes in each handle — as ℓⱼ → 0 the handle collapses, as ℓⱼ grows the torus opens up. The six twists θ₁–θ₆ (not shown) rotate the handles without changing the shape visible here.
Genus-3 surface: 4 pairs of pants · 6 seam geodesics (γ₁…γ₆) · T₃ ≅ ℝ¹² (setting all twists = 0)
Teichmüller space Tg remembers too much: it distinguishes surfaces that are geometrically identical but labeled differently. The moduli space Mg = Tg / Modg quotients by the mapping class group Modg — all topological symmetries of the surface.
The simplest non-trivial example: the moduli space of flat tori. Every complex torus is ℂ / (ℤ + τℤ) for some τ in the upper half-plane ℍ. Two values τ, τ′ give the same torus if and only if they are related by a Möbius transformation in PSL(2, ℤ): τ′ = (aτ+b)/(cτ+d).
The fundamental domain — the region {|τ| ≥ 1, |Re τ| ≤ ½} — contains exactly one representative of each distinct torus shape. Drag the gold point to see how τ controls the lattice shape of the torus.
The cusp of this fundamental domain (Im τ → ∞) corresponds to tori with very long thin necks — surfaces approaching degeneration. This "compactification" behavior of moduli space is crucial in Mirzakhani's work.
The mapping class group for a genus-g surface is a rich, finitely presented group. For the torus it is PSL(2, ℤ), generated by two elements: a Dehn twist along the longitude and along the meridian.
The fundamental domain of PSL(2, ℤ) acting on ℍ — the moduli space ℳ1,1 of complex tori.
On a fixed hyperbolic surface Σ, how many simple closed geodesics (non-self-intersecting closed curves, pulled taut) have length at most L? For large L, Mirzakhani proved:
The exponent 6g − 6 is exactly the dimension of Teichmüller space — a deep coincidence. The constant c(Σ) depends on the geometry of the surface.
Compare: counting all closed geodesics gives exponential growth eL/L (the prime geodesic theorem). The simple ones are far rarer, growing only polynomially.
Mirzakhani's key idea: instead of fixing one surface, average over all surfaces. The moduli space Mg,n(b₁,…,bₙ) parametrizes hyperbolic surfaces of genus g with n geodesic boundary circles of prescribed lengths b₁,…,bₙ. She computed its volume with respect to the Weil–Petersson symplectic form.
Mirzakhani proved a topological recursion: Vg,n can be expressed in terms of Vg′,n′ for smaller (g′, n′). The key tool is the McShane identity — a remarkable formula saying that a certain sum over simple geodesics on a hyperbolic surface with one cusp equals exactly 1.
In 1990, physicist Edward Witten proposed a conjecture about 2-dimensional quantum gravity: the generating function of intersection numbers ⟨τd₁ ··· τdₙ⟩g on the moduli space of Riemann surfaces satisfies an integrable system called the KdV hierarchy.
Here, ⟨τd₁ ··· τdₙ⟩g is a rational number measuring how many times certain natural geometric cycles on Mg,n intersect — a purely algebraic-geometric quantity.
Kontsevich proved the conjecture in 1992 using matrix models and ribbon graphs. Mirzakhani gave a completely different proof in 2007: her volume recursion, when decoded algebraically, implies exactly the KdV equations. A bridge from counting geodesics on random hyperbolic surfaces to the equations of integrable systems in mathematical physics.
The generating function F(t₀, t₁, …) = Σg,n ℏg−1/n! Σ ⟨τd₁···τdₙ⟩g ∏tᵢ satisfies (∂/∂t₀)³F = (∂/∂t₁)F + ½[(∂/∂t₀)²F]² + 1⁄12ℏ (∂/∂t₀)⁴F and infinitely many more KdV equations. Mirzakhani's proof makes this concrete and geometric.
Imagine being born inside a hyperbolic universe. Your visual experience would be profoundly, structurally different — not merely distorted, but governed by different exponential laws at every scale.
Your complete visual field is the Poincaré disk. Every direction you can look corresponds to a point in the open unit disk. The glowing boundary circle represents the visual horizon — not a wall at finite distance, but the limit of every geodesic ray as it reaches infinitely far. Objects at large hyperbolic distance r appear compressed toward this boundary:
All tiles in the animated tessellation are congruent in hyperbolic geometry — same shape, same angles, same area. Only your Euclidean eye sees them shrink. The central tile and a tile near the rim are geometrically identical. You are always at the center of your own infinite world.
The boundary circle is the Gromov boundary of ℍ² — the space of "directions at infinity," homeomorphic to S¹ with a natural Hölder structure. Isometries of ℍ² extend to Möbius maps on this circle. This is the mathematical foundation of conformal field theory boundaries and AdS/CFT holography.
The animation shows a gentle figure-8 walk through the {7,3} tessellation. Your Möbius position in the disk corresponds exactly to your physical location in hyperbolic space.
Below is a first-person view from inside ℍ³. The floor is tiled with congruent H² squares — but each successive row of tiles requires exponentially more to fill the same visual width. Standing trees are equally tall, equally spaced in H², but shrink exponentially faster than in Euclidean space (angular size ∝ 1/sinh(d) instead of 1/d). The horizon is not a wall at finite distance but the ideal boundary ∂ℍ³ ≅ S² — every direction extends infinitely, yet the entire infinite forest crowds into a thin strip near the skyline.
Two geodesic rays leaving your eye at angle θ apart are separated at hyperbolic distance d by:
Compare Euclidean Δ(d) ≈ θ·d. At d = 5: hyperbolic separation is sinh(5)/5 ≈ 14.8× greater. Two walls that appear nearly parallel at d = 1 are exponentially far apart by d = 3.
An object of size s at hyperbolic distance d subtends apparent angle:
Euclidean: α = s/d. At d = 5, a friend appears not 1/5 of her original angular size but e−5/2 ≈ 0.7% — she vanishes almost immediately.
The bright night sky. In ℍ³, the volume of a ball of radius r grows as V(r) = π(sinh(2r) − 2r) ≈ πe²ʳ. Stars in the shell [r, r+1] number ~e²ʳ more than the innermost shell. Each appears e−2r dimmer (apparent size squared). Their total contribution to sky brightness stays constant per shell — a hyperbolic Olbers' paradox. The night sky would be uniformly luminous: packed at every angle with exponentially crowded, exponentially faint stars. This is exactly what the star field in the street view above begins to show.