Why one surface carries many geometries
By now the previous guides in this rung have trained one reflex: a geometry on a manifold is a (G,X)-structure, packaged by a developing map and read off by a holonomy representation of pi_1 into the symmetry group G. Guide 3 then handed you a clean dichotomy in dimension 3 via Mostow rigidity: a closed hyperbolic n-manifold for n at least 3 has a hyperbolic structure that is unique up to isometry, so its holonomy is rigid. The headline of this guide is that dimension 2 is exactly where that rigidity fails, and the failure is not a defect but a rich new space to study. A closed orientable surface of genus g at least 2 admits an entire continuum of inequivalent hyperbolic structures, and organizing that continuum is the whole subject.
Why does the dimension matter so sharply? The uniformization theorem tells you that every Riemann surface of genus at least 2 is a quotient H^2 / Gamma of the hyperbolic plane by a discrete torsion-free subgroup Gamma of PSL(2, R), so its universal cover is rigidly H^2. The freedom does not live in the model space X = H^2; it lives in the holonomy. You can deform the discrete group Gamma — stretch one part of the surface, shrink another, twist along a curve — and as long as the deformed group stays discrete and torsion-free you get a genuinely different hyperbolic metric on the same underlying topological surface. Mostow rigidity says no such room exists in dimension 3 because PSL(2, C) is too rigid; in dimension 2 the room is exactly what we will measure.
What Teichmüller space is, exactly
Here is the definition done carefully, because the careful version is what makes the space well-behaved. Fix the topological surface S. A marked hyperbolic structure is a pair (X, f) where X is a hyperbolic surface and f: S -> X is a diffeomorphism — the map f is the marking, a chosen identification telling you which loop of X corresponds to which loop of S. Two pairs (X_1, f_1) and (X_2, f_2) are declared equivalent when there is an isometry h: X_1 -> X_2 with h composed with f_1 isotopic to f_2. The Teichmüller space Teich(S) is the set of these equivalence classes. The marking is the whole trick: it remembers the labelling, so a Dehn twist that scrambles the loops produces a genuinely different point, even when the underlying unmarked surfaces are isometric.
There is a second, equivalent face of the same space that connects straight back to holonomy, and it is worth holding both pictures at once. Via uniformization, a marked hyperbolic structure is the same data as a discrete faithful representation rho: pi_1(S) -> PSL(2, R) up to conjugation. So Teich(S) sits inside the representation variety Hom(pi_1(S), PSL(2, R)) / conjugation as the connected component of discrete faithful representations — the Fricke description. This is the deformation space of the holonomy from Guide 1, now studied as a space in its own right rather than a single point. The two pictures, marked metrics and discrete faithful reps, are the same Teichmüller space seen through geometry and through algebra.
Counting the dimensions: Fenchel-Nielsen coordinates
The cleanest proof that Teich(S) is a continuum, and the cleanest set of coordinates on it, both come from cutting the surface into simple pieces. Choose a maximal collection of disjoint simple closed curves on S that cut it into pairs of pants — three-holed spheres. A genus-g surface decomposes into 2g - 2 pants joined along 3g - 3 curves. The miracle that powers everything is that a hyperbolic pair of pants is rigidly determined by the three lengths of its boundary geodesics: pick three positive numbers, and there is exactly one hyperbolic pair of pants with those cuff lengths, no freedom left over. So a hyperbolic structure on S is built by choosing a length for each of the 3g - 3 cutting curves, then gluing the pants back together.
But gluing has its own freedom: when you reattach two pants along a cuff of fixed length, you may rotate one side relative to the other before gluing, and different rotation amounts give non-isometric marked surfaces. That rotation is the twist parameter, one real number per gluing curve. Pairing each curve's length with its twist gives the Fenchel-Nielsen coordinates: 3g - 3 length parameters ranging over the positive reals and 3g - 3 twist parameters ranging over all of R. They are global coordinates, a homeomorphism onto their range. Counting them settles the dimension once and for all.
surface S, genus g >= 2:
pants decomposition: 2g - 2 pairs of pants, 3g - 3 cutting curves
each gluing curve --> ( length L_i > 0 , twist tau_i in R )
Fenchel-Nielsen: ( L_1,...,L_{3g-3} , tau_1,...,tau_{3g-3} )
Teich(S) ~= (R_{>0})^{3g-3} x R^{3g-3} ~= R^{6g-6}
dim Teich(S) = 6g - 6 (g = 2: dimension 6)Two consequences are worth pausing on. First, Teich(S) is homeomorphic to R^{6g-6} — it is a cell, contractible, with no holes and no boundary, which is what makes it such a friendly parameter space despite cataloguing infinitely many geometries. For a genus-2 surface that is already six real dimensions of hyperbolic structures on a single fixed surface. Second, the choice of pants decomposition was arbitrary; a different cutting system gives different but equally valid global coordinates, and the change of coordinates between two such systems is a genuinely intricate map. The intrinsic object is Teich(S); the coordinates are charts on it.
The mapping class group: symmetries of the surface
We built Teich(S) by remembering the marking. The price of remembering is that genuinely symmetric pictures get counted as different points: a self-diffeomorphism of S that permutes the loops sends one marked structure to another, even when the unmarked geometries coincide. The group recording exactly these self-symmetries is the mapping class group Mod(S), defined as the group of orientation-preserving diffeomorphisms of S modulo isotopy — that is, pi_0 of the diffeomorphism group, two maps being the same element if one can be smoothly deformed into the other. It is the natural symmetry group of a surface, the topological analogue of the symmetry group of a polygon, only infinite and far richer.
What does an element of Mod(S) actually look like? The basic move is a Dehn twist along a simple closed curve c: cut S open along c, rotate one side by a full turn, and reglue. Picture a thickened annulus around c; the twist fixes everything outside the annulus and shears the annulus by 360 degrees, dragging every loop crossing c along with it. A single Dehn twist is invisible to homology in some directions but utterly reshuffles the marking. The decisive structural theorem, due to Dehn and Lickorish, is that finitely many Dehn twists generate the entire mapping class group — every symmetry of the surface, however complicated, is a product of these elementary twists.
Mod(S) acts on Teich(S) in the most natural way imaginable: a mapping class [phi] sends a marked structure (X, f) to (X, f composed with phi-inverse), simply changing the marking by the symmetry. This action is by isometries of the natural Teichmüller metric and, crucially, it is properly discontinuous — orbits do not accumulate, so the action is well-behaved enough to take a quotient. There is one honest subtlety: the action is not quite free, because surfaces with extra symmetry have nontrivial stabilizers, which is exactly why the quotient will have mild singularities rather than being a smooth manifold.
Dividing out: moduli space
Now combine the two halves. Teichmüller space remembered the marking; the mapping class group is exactly the ambiguity in the marking. Forgetting the marking therefore means quotienting Teich(S) by the Mod(S) action, and the result is the moduli space of Riemann surfaces, M(S) = Teich(S) / Mod(S). A point of M(S) is an unmarked hyperbolic surface — a shape, with no labelling of its loops. This is the answer to the question we have been circling since uniformization: how many genus-g Riemann surfaces are there, and how do they vary? They form a space of complex dimension 3g - 3, the single most studied object in the geometry of curves.
The relationship is worth saying as a slogan, then immediately qualifying it. Teich(S) is the smooth, contractible universal cover R^{6g-6}; Mod(S) is the deck group acting properly discontinuously; M(S) is the quotient. The slogan would be "moduli space is Teichmüller space mod the mapping class group," and that is correct, but two honesty points keep you from over-reading it. First, because the action is not free, M(S) is an orbifold rather than a manifold — it has cone-like points coming from surfaces with automorphisms, and pretending otherwise will eventually bite you. Second, M(S) is not compact: surfaces can degenerate by pinching a curve, sending its Fenchel-Nielsen length to zero and running off to infinity, which is precisely why a compactification (the Deligne-Mumford compactification) had to be invented.