From conformal to quasiconformal: bounding the distortion
Guide 4 introduced Teichmuller space T(S) as the space of marked complex structures on a surface S, and we said two points are close when one surface can be deformed into the other by a map that barely distorts angles. It is time to make 'barely distorts' precise. The key object is the quasiconformal map. A holomorphic (conformal) map sends infinitesimal circles to infinitesimal circles — it preserves angles exactly. A quasiconformal map relaxes this: it sends infinitesimal circles to infinitesimal ellipses, but with the eccentricity of those ellipses uniformly bounded across the whole surface. The bound is the quantitative measure of how far the map is from being conformal.
Write a map f of the plane in real coordinates and form its two Wirtinger derivatives, the holomorphic derivative f_z = (f_x - i f_y)/2 and the antiholomorphic derivative f_zbar = (f_x + i f_y)/2. A map is conformal exactly when f_zbar = 0 (that is the Cauchy-Riemann equations in disguise). The single complex function that records the failure of conformality is the Beltrami coefficient mu = f_zbar / f_z. Its modulus |mu(p)| at a point measures the local stretching: the image ellipse has axis ratio K(p) = (1 + |mu(p)|) / (1 - |mu(p)|), called the dilatation, and its argument records the direction of maximal stretch. The map is K-quasiconformal when |mu| is at most (K-1)/(K+1) everywhere, i.e. the dilatation never exceeds a fixed K.
conformal: f_zbar = 0 circles -> circles (K = 1) quasiconformal: |f_zbar| <= k |f_z| circles -> ellipses (K = (1+k)/(1-k)) Beltrami coefficient: mu = f_zbar / f_z , |mu| < 1 dilatation at p: K(p) = (1 + |mu(p)|) / (1 - |mu(p)|) ellipse axis ratio = K(p) , major axis along arg(mu)/2 direction
The measurable Riemann mapping theorem: solving for the surface
Here is the theorem that makes the whole machine run. The classical Riemann mapping theorem says: every simply connected proper domain is conformally a disk. The measurable Riemann mapping theorem (Morrey, Ahlfors-Bers) is the staggering upgrade: given ANY measurable Beltrami coefficient mu on the plane with the L-infinity norm of mu strictly less than 1, there exists a quasiconformal homeomorphism f of the plane whose complex dilatation is exactly that mu — and f is unique up to post-composition with a conformal map. In symbols, the Beltrami equation f_zbar = mu f_z can always be solved, for ANY bounded measurable right-hand side.
Why this is the heart of Teichmuller theory: a complex structure on a surface is, locally, a recipe for which infinitesimal ellipses count as 'circles.' A Beltrami coefficient mu is precisely such a recipe — a measurable field of infinitesimal ellipses (technically a (-1,1)-form, mu times dzbar/dz, invariant under the right transformations). So a bounded Beltrami coefficient on a fixed reference surface X_0 IS a deformation of its complex structure, and the measurable Riemann mapping theorem hands you back the actual deformed Riemann surface X_mu together with the quasiconformal map X_0 -> X_mu realizing it. The space of Beltrami coefficients, modulo the equivalences from marking and conformal change, is Teichmuller space. This is why T(S) is a manifold and not just a set: it inherits the linear structure of the space of Beltrami differentials.
The Teichmuller distance falls out immediately and beautifully. Define the distance between two marked surfaces as (1/2) log of the smallest dilatation K achievable by any quasiconformal map in the correct homotopy class between them. Teichmuller's theorem then says the minimizer always exists and is essentially unique: the optimal map has constant |mu| = k and its Beltrami coefficient has the form mu = k (qbar / |q|) for a holomorphic quadratic differential q on the surface. These extremal maps stretch along the horizontal trajectories of q by a factor and compress along the vertical by the reciprocal — affine stretches in the natural coordinate of q. That single picture, an affine stretch read in the coordinate of a quadratic differential, is the geometric content of the Teichmuller metric.
From Teichmuller space to moduli: quotient by the mapping class group
Recall from Guide 4 that the mapping class group Mod(S) — orientation-preserving diffeomorphisms modulo isotopy — acts on T(S) by changing the marking. The honest reason Teichmuller space exists at all is to UNFOLD the moduli space: forgetting the marking, the quotient M(S) = T(S) / Mod(S) is the moduli space of curves, the space of Riemann surfaces of genus g up to biholomorphism, with no marking remembered. Teichmuller space is the universal cover-like object that is a smooth manifold (in fact a cell, homeomorphic to R^(6g-6)), while the moduli space below it is an orbifold, carrying cone-like singularities exactly at the surfaces with extra symmetry.
Guide 4 gave you global coordinates for all of this: the Fenchel-Nielsen coordinates. Cut a genus-g surface along 3g-3 disjoint simple closed curves (a pants decomposition) and you split it into 2g-2 pairs of pants, each a hyperbolic surface determined by its three boundary lengths. The 3g-3 length parameters l_i, together with the 3g-3 twist parameters tau_i recording how much you rotate before regluing each cuff, give 6g-6 real coordinates — exactly the dimension of T(S). The mapping class group acts on these coordinates in a controlled, if complicated, way: a Dehn twist about the i-th curve shifts tau_i by l_i. Passing to the moduli space means taking the quotient by all of these moves at once, which is why M(S) is so much subtler than the smooth T(S) above it.
The Weil-Petersson metric: a Kahler geometry on moduli
The Teichmuller metric of section 2 is genuinely natural but analytically harsh — it is only a Finsler metric (its unit balls are not ellipsoids), so you cannot do Riemannian calculus with it. The Weil-Petersson metric is the smooth Riemannian alternative, and it comes from a single honest idea: the tangent space to Teichmuller space at a surface X is the space of harmonic Beltrami differentials on X, and a Beltrami differential is paired against a holomorphic quadratic differential. Equip the cotangent space (the quadratic differentials q) with the L^2 inner product, integrating q times qbar against the hyperbolic area form of X, and dualize. That L^2 pairing of quadratic differentials is the Weil-Petersson metric.
Three remarkable facts follow, and each one earns its place. First, the Weil-Petersson metric is Kahler — its associated 2-form is closed — which is no small thing, because it means all the Kahler machinery from the complex-geometry track applies: Hodge theory, the relation between curvature and the complex structure, the works. Second, in Fenchel-Nielsen coordinates Wolpert's formula gives the Kahler form an astonishingly clean shape: omega_WP = (1/2) sum over i of d l_i wedge d tau_i. The length-and-twist coordinates are Darboux coordinates for the symplectic form, so the moduli space is a symplectic manifold with a completely explicit symplectic structure.
Third, and most consequential for geometry: the Weil-Petersson metric has negative sectional curvature, but it is incomplete. Incompleteness is geometric, not a defect — the missing points are exactly the noded surfaces where a curve has been pinched to zero length (l_i -> 0), the boundary along which moduli space gets its Deligne-Mumford compactification. So a Weil-Petersson geodesic can run off the space in finite time by pinching a curve. The negative curvature is not uniformly bounded away from zero, so you cannot just quote Cartan-Hadamard from the comparison rung; but enough of the negatively-curved philosophy survives that Teichmuller space with the WP metric behaves like a (singular, incomplete) nonpositively curved space, with a unique geodesic between most pairs of points.
Why moduli is central: rigidity, dynamics, and physics
Step back and see what this rung has built. Guides 1 and 2 put rigid (G,X)-geometries on manifolds via developing maps and holonomy; Guide 3 used Mostow rigidity to show that in dimension three and above a hyperbolic structure is unique, so no moduli at all — the geometry is rigid. The two-dimensional case is the glorious exception: a genus-g surface has a (6g-6)-dimensional moduli space of hyperbolic structures, and quasiconformal maps are exactly the flexibility that dimension two permits and dimension three forbids. Mostow rigidity, seen this way, is precisely the statement that the quasiconformal deformations which fill out Teichmuller space in dimension two collapse to a point in dimension three.
The mapping class group acting on Teichmuller space gives a rich dynamical story (Thurston's classification of surface diffeomorphisms into periodic, reducible, and pseudo-Anosov types is the surface analogue of the Nielsen-Thurston theory), and the Weil-Petersson geodesic flow on moduli space is a celebrated example of a non-uniformly hyperbolic, ergodic flow. Wolpert's symplectic formula omega_WP = (1/2) sum d l_i wedge d tau_i is the doorway to symplectic geometry on moduli, and Mirzakhani's Fields-Medal work used exactly this structure to compute Weil-Petersson volumes of moduli spaces by an elegant integration over the length-twist coordinates.
And the reach extends well past pure geometry. In string theory the moduli space of curves is the space of worldsheets, and the Weil-Petersson form (with its Mumford-class corrections) governs the measure one integrates over to compute string amplitudes — the same space, the same Kahler form, in physics. The uniformization theorem underwrites all of it by guaranteeing that every point of moduli is a genuine hyperbolic surface, so the hyperbolic length-and-twist picture is never just a model. Resist the hype, though: these connections are deep and real, but each — Mirzakhani's volumes, Thurston dynamics, string amplitudes — is a research program, surveyed here and proved elsewhere. What you should carry away is the architecture: quasiconformal maps measure deformation, the measurable Riemann mapping theorem realizes it, and the Weil-Petersson metric makes the resulting moduli space a curved geometric object you can actually do Riemannian and symplectic geometry on.