If vp(t)=1 we say that f is unramified at p. Otherwise p is a ramification point of f, in which case we say that q is a branch point.
The ramification divisor of f is
R=p∈Y∑(vp(t)−1)⋅p.
Implicit in this definition is the fact that f has finitely many ramification points. Our first theorem is the Riemann-Hurwitz formula.
Theorem: In this situation in characteristic 0,
KY=f∗KX+R
and in particular
2g(Y)−2=d(2g(X)−2)+p∈Y∑(vp(t)−1)
where d=deg(t).
Remark: We say that f is separable if the induced field extension K[Y]/K[X] is separable. In positive characteristic
If f is not separable then it is ramified along all of Y
Even if it is separable the formula might fail if the ramification is not tame (meaning the ramification index is divisible by the characteristic of the base field).
Proof.(Sketch of an analytic proof.) Let ω be a meromorphic 1-form on X the target curve. Pulling it back by f gives us a 1-form on Y. Away from branch points, f is a local isomorphism (here we have a drawing of a stack of analytic disks in Y over an analytic disk around the point q in X) and thus any zero or pole of ω is given d zeros or d poles on Y.
At a ramification point p, f is given analytic locally be z↦zvp(t). On the other hand, ω is locally given by dt=dzvp(t)=(vp(t))zvp(t)−1dz on Y.
□
We can see where this fails when the ramification is not tame; if the characteristic of the base field divides vp(t) then ω vanishes on Y.
Cor: There is no nontrivial étale cover Y→P1.
Proof. Recall that étale = flat + unramified. Any finite surjective map of nonsingular varieties is flat, but a cover f:Y→P1 will always fail to be unramified. Indeed, suppose f is étale. Riemann-Hurwitz says that
2g(Y)−2=−2⋅d+deg(R)=−2d
which implies g(Y)=0,d=1. The important part is that deg(f)=d=1, hence f is an isomorphism on function fields. hence f is an isomorphism. The only étale map is therefore trivial.
□
Cor: If f:Y→X is a cover then g(Y)≥g(X).
Proof.
If g(X)=0 then there is nothing to show. Otherwise, we get by Riemann-Hurwitz that
If g(X)>0 then (d−1)(g(X)−1)≥0 and 2deg(R)≥0 so the claim follows.
□
Cor:deg(R) is even.
Cor: If g≥2 and g(X)=g(Y)=g then any cover f:Y→X is an isomorphism.
Proof.2g−2=d⋅(2g−2)+deg(R) which implies d=1 and hence f is an isomorphism.
□
Example: Take f:P12:1P1. Then
−2=2⋅(−2)+deg(R)⟹deg(R)=2
and so f is ramified at two points with ramification index 2: [x:y]↦[x2:y2].
Cor:(Hurwitz bound). Suppose X has genus ≥2 and finitely many automorphisms. Then ∣Aut(X)∣≤84(g−1).
Remark:
This bound is sharp in the sense that it is achieved for infinitely many g, but not all of them. For example, up to g=6 it is achieved only in g=3 by the Klein quartic:
{x3y+y3z+z3x=0}⊆P2
The upshot is that if you fix g, there is a bound on the automorphisms of ALL curves of that genus. A priori, it might be the case that for fixed genus g you could find curves of that genus whose automorphism group has arbitrarily large size.
The Hurwitz bound fails in positive characteristic.
Proof. You can check that X/Aut(X)=Y is again a curve. So you get a covering X→Y of degree Aut(X)∣. Ramification points of this cover are the points with non-trivial stabilizers.
Denote by ri,j the ramification points lying over a branch point pi for 1≤i≤b. (A picture of a line representing the base curve Y with a tick for point pi lying in Y. Above pi is a loopy curved line representing X so that each point ri,j is a ramification point. Each ri,1,ri,2,ri,3 is a place where the fiber is "tangent to the curve X")
Over pi all points in the fiber f−1(pi) have the same ramification index fi. If ri=∣f−1(pi)∣ then fi⋅ri=∣Aut(X)∣. Then Riemann-Hurwitz says