ProofComplete

Proof of Hurwitz Formula

The Hurwitz formula relates the genera of two smooth projective curves connected by a finite morphism. It is one of the most fundamental results in curve theory, governing the topology and arithmetic of branched covers. This page gives a complete algebraic proof via canonical divisors, a topological proof sketch over C\mathbb{C}, the generalization to wild ramification, and extensive examples.


Statement of the Theorem

Theorem4.6Hurwitz Formula

Let f:C→Df: C \to D be a finite separable morphism of degree nn between smooth projective curves over an algebraically closed field kk. Then

2gCβˆ’2=n(2gDβˆ’2)+deg⁑R,2g_C - 2 = n(2g_D - 2) + \deg R,

where gCg_C and gDg_D are the genera of CC and DD respectively, and the ramification divisor is

R=βˆ‘P∈C(ePβˆ’1)β‹…P,R = \sum_{P \in C} (e_P - 1) \cdot P,

with ePe_P the ramification index of ff at PP. In particular, deg⁑R=βˆ‘P∈C(ePβˆ’1)β‰₯0\deg R = \sum_{P \in C} (e_P - 1) \geq 0.

RemarkEquivalent formulations

The Hurwitz formula can be written in several equivalent ways:

  • In terms of the Euler characteristic: Ο‡(C)=nβ‹…Ο‡(D)βˆ’deg⁑R\chi(C) = n \cdot \chi(D) - \deg R, where Ο‡=2βˆ’2g\chi = 2 - 2g.
  • In terms of canonical divisors: KC∼fβˆ—KD+RK_C \sim f^*K_D + R, so deg⁑KC=nβ‹…deg⁑KD+deg⁑R\deg K_C = n \cdot \deg K_D + \deg R.
  • In terms of branch points: if Q1,…,Qr∈DQ_1, \ldots, Q_r \in D are the branch points and the fiber over QjQ_j consists of points Pj,1,…,Pj,sjP_{j,1}, \ldots, P_{j,s_j} with ramification indices ej,1,…,ej,sje_{j,1}, \ldots, e_{j,s_j}, then deg⁑R=βˆ‘j=1rβˆ‘i=1sj(ej,iβˆ’1)\deg R = \sum_{j=1}^r \sum_{i=1}^{s_j} (e_{j,i} - 1).

Proof via Canonical Divisors

The proof proceeds in four steps: pullback of differentials, local computation of ramification, passage to canonical divisors, and degree counting.

Step 1: Pullback of Differentials

Proof

Step 1: The pullback map fβˆ—:Ξ©D1β†’Ξ©C1f^*: \Omega^1_D \to \Omega^1_C.

Let f:Cβ†’Df: C \to D be a finite separable morphism of degree nn. The morphism ff induces an inclusion of function fields fβˆ—:k(D)β†ͺk(C)f^*: k(D) \hookrightarrow k(C), which is a finite separable field extension of degree nn.

By functoriality of Kahler differentials, there is a pullback map on the sheaves of differentials:

fβˆ—:Ξ©D/k1β†’Ξ©C/k1.f^*: \Omega^1_{D/k} \to \Omega^1_{C/k}.

More precisely, for an open set UβŠ†DU \subseteq D and a regular differential Ο‰βˆˆΞ©D1(U)\omega \in \Omega^1_D(U), we obtain fβˆ—Ο‰βˆˆΞ©C1(fβˆ’1(U))f^*\omega \in \Omega^1_C(f^{-1}(U)) by composition with fβˆ—f^* on functions.

At the level of function fields, if Ο‰=g ds\omega = g \, ds for some g,s∈k(D)g, s \in k(D), then fβˆ—Ο‰=fβˆ—(g) d(fβˆ—s)f^*\omega = f^*(g) \, d(f^*s), where we view fβˆ—(g)f^*(g) and fβˆ—(s)f^*(s) as elements of k(C)k(C).

Key point: Since ff is separable, the map fβˆ—:Ξ©k(D)/k1β†’Ξ©k(C)/k1f^*: \Omega^1_{k(D)/k} \to \Omega^1_{k(C)/k} is injective. The pullback fβˆ—Ο‰f^*\omega is nonzero whenever Ο‰\omega is nonzero, so the pullback of a nonzero rational differential on DD gives a nonzero rational differential on CC.

β– 

Step 2: Local Computation of Ramification

Proof

Step 2: Local analysis of fβˆ—Ο‰f^*\omega near a ramification point.

Let P∈CP \in C be a closed point with Q=f(P)∈DQ = f(P) \in D, and let e=ePe = e_P be the ramification index. Choose a local uniformizer ss at QQ (so vQ(s)=1v_Q(s) = 1) and a local uniformizer tt at PP (so vP(t)=1v_P(t) = 1).

Since ff has ramification index ee at PP, we have vP(fβˆ—s)=ev_P(f^*s) = e. Therefore we can write

fβˆ—s=uβ‹…tef^*s = u \cdot t^e

where uu is a unit in the local ring OC,P\mathcal{O}_{C,P} (i.e., vP(u)=0v_P(u) = 0).

Now consider the differential Ο‰=ds\omega = ds on DD near QQ. Its pullback is:

fβˆ—Ο‰=fβˆ—(ds)=d(fβˆ—s)=d(uβ‹…te)=uβ‹…eβ‹…teβˆ’1 dt+te du.f^*\omega = f^*(ds) = d(f^*s) = d(u \cdot t^e) = u \cdot e \cdot t^{e-1} \, dt + t^e \, du.

Since uu is a unit and du=u′ dtdu = u' \, dt for some regular function uβ€²u', we get

fβˆ—(ds)=(eβ‹…uβ‹…teβˆ’1+teβ‹…uβ€²) dt=teβˆ’1(eu+tuβ€²) dt.f^*(ds) = (e \cdot u \cdot t^{e-1} + t^e \cdot u') \, dt = t^{e-1}(eu + tu') \, dt.

The factor (eu+tuβ€²)(eu + tu') is a unit at PP because uu is a unit and we are in characteristic 00 or gcd⁑(e,char⁑k)=1\gcd(e, \operatorname{char} k) = 1 (the tamely ramified case), so eβ‹…ue \cdot u is a unit.

Therefore: vP(fβˆ—(ds))=eβˆ’1v_P(f^*(ds)) = e - 1.

Conclusion: The pullback of the differential dsds acquires a zero of order exactly ePβˆ’1e_P - 1 at each ramification point PP.

β– 

Step 3: The Key Identity KC=fβˆ—KD+RK_C = f^*K_D + R

Proof

Step 3: From local data to the canonical divisor identity.

Let Ο‰D\omega_D be a nonzero rational differential on DD, so div⁑(Ο‰D)=KD\operatorname{div}(\omega_D) = K_D is a canonical divisor on DD. The pullback fβˆ—Ο‰Df^*\omega_D is a nonzero rational differential on CC.

We compute div⁑(fβˆ—Ο‰D)\operatorname{div}(f^*\omega_D) by combining the pullback of divisors with the ramification contribution.

Pullback of divisors: For any point Q∈DQ \in D and P∈fβˆ’1(Q)P \in f^{-1}(Q), the pullback of divisors satisfies vP(fβˆ—KD)=ePβ‹…vQ(KD)v_P(f^*K_D) = e_P \cdot v_Q(K_D). More precisely, for the divisor fβˆ—KD=βˆ‘Q∈DvQ(KD)β‹…fβˆ—(Q)f^*K_D = \sum_{Q \in D} v_Q(K_D) \cdot f^*(Q) where fβˆ—(Q)=βˆ‘P↦QePβ‹…Pf^*(Q) = \sum_{P \mapsto Q} e_P \cdot P.

Ramification contribution: From Step 2, the pullback of differentials introduces an additional factor of tePβˆ’1t^{e_P - 1} at each point PP, contributing ePβˆ’1e_P - 1 to the order of vanishing.

Combining these two effects:

vP(fβˆ—Ο‰D)=ePβ‹…vQ(Ο‰D)+(ePβˆ’1)v_P(f^*\omega_D) = e_P \cdot v_Q(\omega_D) + (e_P - 1)

for each P∈CP \in C with Q=f(P)Q = f(P). In terms of divisors:

div⁑(fβˆ—Ο‰D)=fβˆ—(div⁑(Ο‰D))+R=fβˆ—KD+R,\operatorname{div}(f^*\omega_D) = f^*(\operatorname{div}(\omega_D)) + R = f^*K_D + R,

where R=βˆ‘P∈C(ePβˆ’1)β‹…PR = \sum_{P \in C}(e_P - 1) \cdot P is the ramification divisor.

Since div⁑(fβˆ—Ο‰D)\operatorname{div}(f^*\omega_D) is a canonical divisor on CC, we obtain:

KC∼fβˆ—KD+R.K_C \sim f^*K_D + R.

β– 

Step 4: Taking Degrees

Proof

Step 4: Degree computation yields the Hurwitz formula.

Taking degrees of both sides of KC∼fβˆ—KD+RK_C \sim f^*K_D + R:

deg⁑KC=deg⁑(fβˆ—KD)+deg⁑R.\deg K_C = \deg(f^*K_D) + \deg R.

We use two standard facts:

Fact 1: For a smooth projective curve XX of genus gg, deg⁑KX=2gβˆ’2\deg K_X = 2g - 2.

Fact 2: For a finite morphism f:Cβ†’Df: C \to D of degree nn and any divisor EE on DD, deg⁑(fβˆ—E)=nβ‹…deg⁑E\deg(f^*E) = n \cdot \deg E. This follows from fβˆ—(Q)=βˆ‘P↦QePβ‹…Pf^*(Q) = \sum_{P \mapsto Q} e_P \cdot P and βˆ‘P↦QeP=n\sum_{P \mapsto Q} e_P = n for each QQ.

Applying these:

2gCβˆ’2=n(2gDβˆ’2)+deg⁑R.2g_C - 2 = n(2g_D - 2) + \deg R.

Since each ePβ‰₯1e_P \geq 1, we have ePβˆ’1β‰₯0e_P - 1 \geq 0 with equality at unramified points, so deg⁑Rβ‰₯0\deg R \geq 0. This completes the proof.

β– 

Wild Ramification: Characteristic pp

RemarkThe different in characteristic p

When char⁑(k)=p>0\operatorname{char}(k) = p > 0 and p∣ePp \mid e_P (wild ramification), the computation in Step 2 breaks down: the factor eβ‹…ue \cdot u may vanish since e≑0(modp)e \equiv 0 \pmod{p}. In this case, vP(fβˆ—(ds))v_P(f^*(ds)) is strictly greater than ePβˆ’1e_P - 1.

The correct generalization replaces the naive ramification divisor with the different:

KC∼fβˆ—KD+D,K_C \sim f^*K_D + \mathfrak{D},

where the different D=βˆ‘P∈CdPβ‹…P\mathfrak{D} = \sum_{P \in C} d_P \cdot P satisfies dPβ‰₯ePβˆ’1d_P \geq e_P - 1, with equality if and only if p∀ePp \nmid e_P (tame ramification).

The generalized Hurwitz formula becomes:

2gCβˆ’2=n(2gDβˆ’2)+βˆ‘P∈CdP.2g_C - 2 = n(2g_D - 2) + \sum_{P \in C} d_P.

Bounds on the different: For wild ramification at PP, we have dPβ‰₯ePβˆ’1+1=ePd_P \geq e_P - 1 + 1 = e_P when p∣ePp \mid e_P. More precisely, if eP=paβ‹…me_P = p^a \cdot m with gcd⁑(m,p)=1\gcd(m, p) = 1, then dPβ‰₯ePβˆ’1+ad_P \geq e_P - 1 + a (though the exact value depends on the higher ramification groups).

Artin conductor: The integer dP+1d_P + 1 is related to the Artin conductor at PP, which measures the "depth" of wild ramification via the filtration of the inertia group by higher ramification groups G0βŠƒG1βŠƒG2βŠƒβ‹―G_0 \supset G_1 \supset G_2 \supset \cdots.

ExampleArtin-Schreier covers (wild ramification)

Let kk have characteristic p>0p > 0. Consider the Artin--Schreier cover f:Cβ†’P1f: C \to \mathbb{P}^1 defined by ypβˆ’y=xmy^p - y = x^m where gcd⁑(m,p)=1\gcd(m, p) = 1 and m>0m > 0.

This is a degree-pp cover, totally ramified at ∞\infty with a single point P∞P_\infty above ∞\infty. The ramification index is eP∞=pe_{P_\infty} = p.

The different at P∞P_\infty is dP∞=(pβˆ’1)(m+1)d_{P_\infty} = (p-1)(m+1), which exceeds ePβˆžβˆ’1=pβˆ’1e_{P_\infty} - 1 = p - 1 by (pβˆ’1)m(p-1)m.

By the generalized Hurwitz formula:

2gCβˆ’2=p(2β‹…0βˆ’2)+(pβˆ’1)(m+1)=βˆ’2p+(pβˆ’1)(m+1).2g_C - 2 = p(2 \cdot 0 - 2) + (p-1)(m+1) = -2p + (p-1)(m+1).

So gC=(pβˆ’1)(mβˆ’1)2g_C = \frac{(p-1)(m-1)}{2}. For example, with p=2p = 2 and m=3m = 3: gC=1g_C = 1 (an elliptic curve).


Topological Proof Sketch over C\mathbb{C}

Proof

Topological proof using Euler characteristics.

When working over C\mathbb{C}, the Hurwitz formula admits a purely topological proof via Euler characteristics. Let f:C→Df: C \to D be a branched cover of compact Riemann surfaces of degree nn.

Step A: Remove branch points. Let B={Q1,…,Qr}βŠ‚DB = \{Q_1, \ldots, Q_r\} \subset D be the set of branch points. Set D∘=Dβˆ–BD^\circ = D \setminus B and C∘=fβˆ’1(D∘)=Cβˆ–fβˆ’1(B)C^\circ = f^{-1}(D^\circ) = C \setminus f^{-1}(B). Then f∣C∘:Cβˆ˜β†’D∘f|_{C^\circ}: C^\circ \to D^\circ is an honest (unramified) covering of degree nn.

Step B: Euler characteristic of the unbranched cover. For an unramified degree-nn cover, Ο‡(C∘)=nβ‹…Ο‡(D∘)\chi(C^\circ) = n \cdot \chi(D^\circ).

We compute Ο‡(D∘)=Ο‡(D)βˆ’r=2βˆ’2gDβˆ’r\chi(D^\circ) = \chi(D) - r = 2 - 2g_D - r (removing rr points from DD decreases Ο‡\chi by rr, since each puncture removes a cell from a CW decomposition).

Similarly, ∣fβˆ’1(B)∣=βˆ‘j=1r∣fiberΒ overΒ Qj∣|f^{-1}(B)| = \sum_{j=1}^r |\text{fiber over } Q_j|. Over QjQ_j, there are sjs_j points with ramification indices ej,1,…,ej,sje_{j,1}, \ldots, e_{j,s_j} where βˆ‘iej,i=n\sum_i e_{j,i} = n. So the total number of preimage points is βˆ‘jsj\sum_j s_j.

Thus Ο‡(C∘)=Ο‡(C)βˆ’βˆ‘jsj=2βˆ’2gCβˆ’βˆ‘jsj\chi(C^\circ) = \chi(C) - \sum_j s_j = 2 - 2g_C - \sum_j s_j.

Step C: Combine. Setting Ο‡(C∘)=nβ‹…Ο‡(D∘)\chi(C^\circ) = n \cdot \chi(D^\circ):

2βˆ’2gCβˆ’βˆ‘jsj=n(2βˆ’2gDβˆ’r).2 - 2g_C - \sum_j s_j = n(2 - 2g_D - r).

Rearranging:

2βˆ’2gC=n(2βˆ’2gD)βˆ’nr+βˆ‘jsj=n(2βˆ’2gD)βˆ’βˆ‘j(nβˆ’sj).2 - 2g_C = n(2 - 2g_D) - nr + \sum_j s_j = n(2 - 2g_D) - \sum_j (n - s_j).

Since nβˆ’sj=βˆ‘iej,iβˆ’sj=βˆ‘i(ej,iβˆ’1)=deg⁑Rjn - s_j = \sum_i e_{j,i} - s_j = \sum_i (e_{j,i} - 1) = \deg R_j (the ramification over QjQ_j), we get:

2βˆ’2gC=n(2βˆ’2gD)βˆ’deg⁑R,2 - 2g_C = n(2 - 2g_D) - \deg R,

which is exactly the Hurwitz formula 2gCβˆ’2=n(2gDβˆ’2)+deg⁑R2g_C - 2 = n(2g_D - 2) + \deg R.

β– 

Examples Verifying the Formula

Example 1: Covers of P1\mathbb{P}^1 by P1\mathbb{P}^1

ExampleDegree-n map between rational curves

Let f:P1β†’P1f: \mathbb{P}^1 \to \mathbb{P}^1 be f(t)=tnf(t) = t^n (a degree-nn map). Here gC=gD=0g_C = g_D = 0.

Ramification: The map ramifies at t=0t = 0 (with e0=ne_0 = n, since t↦tnt \mapsto t^n) and at t=∞t = \infty (with e∞=ne_\infty = n). All other points are unramified.

deg⁑R=(nβˆ’1)+(nβˆ’1)=2(nβˆ’1).\deg R = (n - 1) + (n - 1) = 2(n - 1).

Hurwitz check: 2(0)βˆ’2=n(2(0)βˆ’2)+2(nβˆ’1)2(0) - 2 = n(2(0) - 2) + 2(n - 1), i.e., βˆ’2=βˆ’2n+2nβˆ’2=βˆ’2-2 = -2n + 2n - 2 = -2. Verified.

Example 2: Hyperelliptic Curves

ExampleHyperelliptic double cover

Let C:y2=f(x)C: y^2 = f(x) where ff has degree 2g+12g + 1 or 2g+22g + 2 with distinct roots, defining a double cover Ο€:Cβ†’P1\pi: C \to \mathbb{P}^1 via (x,y)↦x(x, y) \mapsto x.

Case deg⁑f=2g+1\deg f = 2g + 1: The branch points are the 2g+12g + 1 roots of ff in A1\mathbb{A}^1 plus ∞\infty (since the cover is totally ramified at infinity when deg⁑f\deg f is odd). Total branch points: 2g+22g + 2. Each has e=2e = 2, so deg⁑R=(2g+2)β‹…1=2g+2\deg R = (2g + 2) \cdot 1 = 2g + 2.

Case deg⁑f=2g+2\deg f = 2g + 2: The 2g+22g + 2 roots of ff are the branch points (the point at infinity is unramified). Again deg⁑R=2g+2\deg R = 2g + 2.

Hurwitz check: 2gCβˆ’2=2(2β‹…0βˆ’2)+2g+2=βˆ’4+2g+2=2gβˆ’22g_C - 2 = 2(2 \cdot 0 - 2) + 2g + 2 = -4 + 2g + 2 = 2g - 2, so gC=gg_C = g. The genus matches the definition.

Example 3: Smooth Plane Quartic

ExampleProjection from a smooth plane quartic

Let CβŠ‚P2C \subset \mathbb{P}^2 be a smooth plane quartic (g=3g = 3). Projection from a point Pβˆ‰CP \notin C gives a map Ο€P:Cβ†’P1\pi_P: C \to \mathbb{P}^1 of degree 44.

Hurwitz formula: 2(3)βˆ’2=4(2(0)βˆ’2)+deg⁑R2(3) - 2 = 4(2(0) - 2) + \deg R, so 4=βˆ’8+deg⁑R4 = -8 + \deg R, giving deg⁑R=12\deg R = 12.

For a general projection, the map has 1212 simple ramification points (each with e=2e = 2), consistent with deg⁑R=12\deg R = 12. These correspond to the tangent lines from PP to CC, of which there are generically 1212 by the class formula for a degree-44 plane curve.

Example 4: Elliptic Curve Double Cover

ExampleElliptic curve as double cover of P^1

An elliptic curve E:y2=x3+Ax+BE: y^2 = x^3 + Ax + B over a field with char⁑≠2\operatorname{char} \neq 2 admits a degree-22 map Eβ†’P1E \to \mathbb{P}^1 via (x,y)↦x(x, y) \mapsto x.

The branch points are the three roots of x3+Ax+Bx^3 + Ax + B plus ∞\infty, giving 44 branch points, each with e=2e = 2.

deg⁑R=4β‹…(2βˆ’1)=4.\deg R = 4 \cdot (2 - 1) = 4.

Hurwitz check: 2(1)βˆ’2=2(2(0)βˆ’2)+42(1) - 2 = 2(2(0) - 2) + 4, i.e., 0=βˆ’4+4=00 = -4 + 4 = 0. Verified.

Example 5: Cyclic Triple Cover

ExampleCyclic triple cover of P^1

Consider f:Cβ†’P1f: C \to \mathbb{P}^1 defined by y3=(xβˆ’a1)(xβˆ’a2)β‹―(xβˆ’ar)y^3 = (x - a_1)(x - a_2) \cdots (x - a_r) where the aia_i are distinct and 3∀r3 \nmid r (so the cover is totally ramified at ∞\infty as well).

Each root aia_i gives a totally ramified point with e=3e = 3, contributing eβˆ’1=2e - 1 = 2 to deg⁑R\deg R. If 3∀r3 \nmid r, then ∞\infty is also totally ramified with e=3e = 3, contributing 22 more.

deg⁑R=2r+2(3∀r),deg⁑R=2r(3∣r).\deg R = 2r + 2 \quad (3 \nmid r), \qquad \deg R = 2r \quad (3 \mid r).

Hurwitz: 2gCβˆ’2=3(βˆ’2)+deg⁑R2g_C - 2 = 3(-2) + \deg R.

  • If r=4r = 4 (so 3∀43 \nmid 4): deg⁑R=10\deg R = 10, 2gCβˆ’2=βˆ’6+10=42g_C - 2 = -6 + 10 = 4, so gC=3g_C = 3.
  • If r=3r = 3: deg⁑R=6\deg R = 6, 2gCβˆ’2=βˆ’6+6=02g_C - 2 = -6 + 6 = 0, so gC=1g_C = 1 (an elliptic curve).
  • If r=6r = 6: deg⁑R=12\deg R = 12, 2gCβˆ’2=βˆ’6+12=62g_C - 2 = -6 + 12 = 6, so gC=4g_C = 4.

Example 6: The Fermat Curve

ExampleFermat curve projection

The Fermat curve C:xn+yn=znC: x^n + y^n = z^n in P2\mathbb{P}^2 is smooth of genus g=(nβˆ’1)(nβˆ’2)2g = \frac{(n-1)(n-2)}{2}.

The projection Ο€:Cβ†’P1\pi: C \to \mathbb{P}^1 via [x:y:z]↦[x:z][x:y:z] \mapsto [x:z] has degree nn (for a fixed ratio x/z=Ξ»x/z = \lambda, we solve yn=znβˆ’xny^n = z^n - x^n).

The map Ο€\pi ramifies at the nn-th roots of unity Ξ»n=1\lambda^n = 1 (where yn=0y^n = 0, giving a single point with e=ne = n) and at ∞\infty (where xn+yn=0x^n + y^n = 0, giving nn points with e=1e = 1 if kk contains nn-th roots of βˆ’1-1). Let us verify more carefully: over Ξ»\lambda with Ξ»n=1\lambda^n = 1, the equation yn=zn(1βˆ’Ξ»n)=0y^n = z^n(1 - \lambda^n) = 0 gives y=0y = 0, a single preimage with e=ne = n. There are nn such branch points.

deg⁑R=nβ‹…(nβˆ’1)=n(nβˆ’1).\deg R = n \cdot (n - 1) = n(n-1).

Hurwitz check: 2β‹…(nβˆ’1)(nβˆ’2)2βˆ’2=n(βˆ’2)+n(nβˆ’1)2 \cdot \frac{(n-1)(n-2)}{2} - 2 = n(-2) + n(n-1), i.e., (nβˆ’1)(nβˆ’2)βˆ’2=βˆ’2n+n2βˆ’n=n2βˆ’3n(n-1)(n-2) - 2 = -2n + n^2 - n = n^2 - 3n. The left side is n2βˆ’3n+2βˆ’2=n2βˆ’3nn^2 - 3n + 2 - 2 = n^2 - 3n. Verified.

Example 7: Frobenius Morphism

ExampleFrobenius morphism (purely inseparable)

Let k=Fpβ€Ύk = \overline{\mathbb{F}_p} and CC a smooth curve of genus gg. The absolute Frobenius F:Cβ†’CF: C \to C sends (x,y)↦(xp,yp)(x, y) \mapsto (x^p, y^p). This is a purely inseparable morphism of degree pp.

The Hurwitz formula does not directly apply in the naive form since FF is inseparable (not separable). However, we can factor the pp-power map P1β†’P1\mathbb{P}^1 \to \mathbb{P}^1 via t↦tpt \mapsto t^p as a purely inseparable map. In this case, every point has ramification index pp but the differential d(tp)=pβ‹…tpβˆ’1dt=0d(t^p) = p \cdot t^{p-1} dt = 0, so the formula via differentials gives fβˆ—Ο‰=0f^*\omega = 0 and we need the generalized framework.

For the relative Frobenius FC/k:Cβ†’C(p)F_{C/k}: C \to C^{(p)}, the different is D=(pβˆ’1)KC+(contributionΒ fromΒ theΒ inseparableΒ structure)\mathfrak{D} = (p-1)K_C + (\text{contribution from the inseparable structure}). This is why separability is assumed in the standard Hurwitz formula.

Example 8: Map Between Elliptic Curves

ExampleIsogeny between elliptic curves

Let Ο•:E1β†’E2\phi: E_1 \to E_2 be a separable isogeny of degree nn between elliptic curves (gE1=gE2=1g_{E_1} = g_{E_2} = 1).

Hurwitz formula: 2(1)βˆ’2=n(2(1)βˆ’2)+deg⁑R2(1) - 2 = n(2(1) - 2) + \deg R, giving 0=0+deg⁑R0 = 0 + \deg R, so deg⁑R=0\deg R = 0.

This means Ο•\phi is unramified everywhere. Every separable isogeny between elliptic curves is an unramified (etale) cover. This is consistent with the group-theoretic viewpoint: an isogeny is a group homomorphism, and its fibers are cosets of ker⁑(Ο•)\ker(\phi), which are all translates of each other and hence uniformly distributed.

Example 9: Genus 2 to Genus 0

ExampleCovers from genus-2 curves to P^1

Any genus-22 curve CC admits a degree-22 map f:C→P1f: C \to \mathbb{P}^1 (every genus-22 curve is hyperelliptic).

Hurwitz: 2(2)βˆ’2=2(2(0)βˆ’2)+deg⁑R2(2) - 2 = 2(2(0) - 2) + \deg R, so 2=βˆ’4+deg⁑R2 = -4 + \deg R, giving deg⁑R=6\deg R = 6.

Since n=2n = 2, each ramification point has e=2e = 2, contributing 11 to deg⁑R\deg R. So there are exactly 66 branch points in P1\mathbb{P}^1. These are precisely the Weierstrass points of CC (their images under ff), and their preimages are the 66 Weierstrass points of CC.

Example 10: Belyi Maps

ExampleBelyi maps (three branch points)

A Belyi map is a finite morphism f:Cβ†’P1f: C \to \mathbb{P}^1 ramified over at most three points (conventionally 0,1,∞0, 1, \infty). By Belyi's theorem, a smooth curve over Qβ€Ύ\overline{\mathbb{Q}} admits a Belyi map if and only if it is defined over Qβ€Ύ\overline{\mathbb{Q}}.

For a Belyi map of degree nn on a curve of genus gg, the Hurwitz formula reads:

2gβˆ’2=βˆ’2n+deg⁑R.2g - 2 = -2n + \deg R.

Concrete example: Consider the "dessin d'enfant" given by a degree-44 Belyi map f:Cβ†’P1f: C \to \mathbb{P}^1 with ramification type (4)(4) over 00, (2,2)(2, 2) over 11, and (3,1)(3, 1) over ∞\infty. Then:

deg⁑R=(4βˆ’1)+(2βˆ’1)+(2βˆ’1)+(3βˆ’1)+(1βˆ’1)=3+1+1+2+0=7.\deg R = (4-1) + (2-1) + (2-1) + (3-1) + (1-1) = 3 + 1 + 1 + 2 + 0 = 7.

So 2gβˆ’2=βˆ’8+7=βˆ’12g - 2 = -8 + 7 = -1, giving g=1/2g = 1/2. This is impossible! Therefore, no such Belyi map exists. The ramification data must satisfy deg⁑R≑0(mod2)\deg R \equiv 0 \pmod{2} (since 2gβˆ’2+2n2g - 2 + 2n is always even).

Valid example: Degree 33, type (3)(3) over 00, (2,1)(2, 1) over 11, (3)(3) over ∞\infty: deg⁑R=2+1+0+2=5\deg R = 2 + 1 + 0 + 2 = 5, 2gβˆ’2=βˆ’6+5=βˆ’12g - 2 = -6 + 5 = -1... still fractional. Let us try type (3)(3) over 00, (2,1)(2, 1) over 11, (2,1)(2, 1) over ∞\infty: deg⁑R=2+1+1=4\deg R = 2 + 1 + 1 = 4, 2gβˆ’2=βˆ’6+4=βˆ’22g - 2 = -6 + 4 = -2, g=0g = 0. This is valid, and describes a rational Belyi function.

Example 11: Covering of an Elliptic Curve by a Higher-Genus Curve

ExampleDouble cover of an elliptic curve

Let EE be an elliptic curve and f:C→Ef: C \to E a degree-22 cover.

Hurwitz: 2gCβˆ’2=2(2β‹…1βˆ’2)+deg⁑R=0+deg⁑R2g_C - 2 = 2(2 \cdot 1 - 2) + \deg R = 0 + \deg R, so gC=1+deg⁑R2g_C = 1 + \frac{\deg R}{2}.

Since each ramification point for a double cover has e=2e = 2 (contributing 11), the number of branch points equals deg⁑R\deg R. This must be even (by the Hurwitz formula parity constraint), so:

  • deg⁑R=0\deg R = 0: gC=1g_C = 1. Unramified double cover; CC is again an elliptic curve.
  • deg⁑R=2\deg R = 2: gC=2g_C = 2. A genus-22 curve double-covering EE with 22 branch points.
  • deg⁑R=4\deg R = 4: gC=3g_C = 3. A genus-33 curve double-covering EE with 44 branch points.
  • deg⁑R=2k\deg R = 2k: gC=1+kg_C = 1 + k.

Example 12: Degree-nn Unramified Cover

ExampleUnramified covers

If f:Cβ†’Df: C \to D is unramified (etale), then deg⁑R=0\deg R = 0 and the Hurwitz formula gives:

2gCβˆ’2=n(2gDβˆ’2),2g_C - 2 = n(2g_D - 2),

i.e., gCβˆ’1=n(gDβˆ’1)g_C - 1 = n(g_D - 1).

Consequences:

  • If gD=0g_D = 0: then gCβˆ’1=βˆ’n<0g_C - 1 = -n < 0, impossible for n>1n > 1. So P1\mathbb{P}^1 has no nontrivial etale covers (it is simply connected).
  • If gD=1g_D = 1: then gC=1g_C = 1 for any nn. Etale covers of elliptic curves are elliptic curves (the nn-isogenies).
  • If gDβ‰₯2g_D \geq 2: then gC=n(gDβˆ’1)+1β‰₯n+1g_C = n(g_D - 1) + 1 \geq n + 1. A degree-22 unramified cover of a genus-22 curve has genus 33.

Applications

Constraining Genus

RemarkHurwitz formula as an obstruction

The Hurwitz formula provides powerful obstructions:

  1. Lower bound on genus of covers: If f:Cβ†’Df: C \to D has degree nn, then gCβ‰₯n(gDβˆ’1)+1g_C \geq n(g_D - 1) + 1 (since deg⁑Rβ‰₯0\deg R \geq 0). Equality holds iff ff is unramified.

  2. Parity constraint: Since deg⁑R=2gCβˆ’2βˆ’n(2gDβˆ’2)\deg R = 2g_C - 2 - n(2g_D - 2), and deg⁑Rβ‰₯0\deg R \geq 0, we need 2gCβˆ’2β‰₯n(2gDβˆ’2)2g_C - 2 \geq n(2g_D - 2), with equality iff the cover is etale.

  3. Finiteness of automorphisms (Hurwitz bound): If GG is a finite group of automorphisms of a curve CC of genus gβ‰₯2g \geq 2, then f:Cβ†’C/Gf: C \to C/G has degree ∣G∣|G|, and the Hurwitz formula gives ∣Gβˆ£β‰€84(gβˆ’1)|G| \leq 84(g - 1). This is the Hurwitz bound, achieved by the Klein quartic (g=3g = 3, ∣G∣=168|G| = 168).

  4. Ruling out covers: There is no degree-22 map P1β†’C\mathbb{P}^1 \to C if gCβ‰₯1g_C \geq 1, since βˆ’2=2(βˆ’2)+deg⁑R-2 = 2(-2) + \deg R would give deg⁑R=2<0\deg R = 2 < 0 if gC=1g_C = 1... actually βˆ’2=2(2gCβˆ’2)+deg⁑R-2 = 2(2g_C - 2) + \deg R is wrong; the map goes the other direction. The correct obstruction: if f:Cβ†’P1f: C \to \mathbb{P}^1 has degree nn, then gC=deg⁑R2βˆ’n+1g_C = \frac{\deg R}{2} - n + 1, so deg⁑R=2gC+2nβˆ’2\deg R = 2g_C + 2n - 2, which is always achievable.

The Hurwitz Bound

Theorem4.7Hurwitz Automorphism Bound

Let CC be a smooth projective curve of genus gβ‰₯2g \geq 2 over an algebraically closed field of characteristic 00 (or characteristic pp not dividing ∣G∣|G|). Then

∣Aut⁑(C)βˆ£β‰€84(gβˆ’1).|\operatorname{Aut}(C)| \leq 84(g - 1).

Proof

Derivation from the Hurwitz formula.

Let G=Aut⁑(C)G = \operatorname{Aut}(C) and D=C/GD = C/G. The quotient map f:Cβ†’Df: C \to D has degree ∣G∣|G|. Let gDg_D be the genus of DD and suppose the branch points Q1,…,QrQ_1, \ldots, Q_r have ramification indices e1,…,ere_1, \ldots, e_r (where eje_j is the common ramification index above QjQ_j, well-defined since GG acts transitively on fibers).

By Hurwitz: 2gβˆ’2=∣G∣(2gDβˆ’2)+∣Gβˆ£βˆ‘j=1r(1βˆ’1/ej)2g - 2 = |G|(2g_D - 2) + |G| \sum_{j=1}^r (1 - 1/e_j).

So 2gβˆ’2=∣G∣(2gDβˆ’2+βˆ‘j=1r(1βˆ’1/ej))2g - 2 = |G|\left(2g_D - 2 + \sum_{j=1}^r (1 - 1/e_j)\right).

We need ∣G∣|G| to be bounded, so we need the factor ΞΌ=2gDβˆ’2+βˆ‘j(1βˆ’1/ej)>0\mu = 2g_D - 2 + \sum_j(1 - 1/e_j) > 0, and then ∣G∣=(2gβˆ’2)/ΞΌ|G| = (2g-2)/\mu.

To maximize ∣G∣|G|, we minimize μ>0\mu > 0. The minimum positive value of μ\mu is achieved at gD=0g_D = 0, r=3r = 3, (e1,e2,e3)=(2,3,7)(e_1, e_2, e_3) = (2, 3, 7):

ΞΌ=βˆ’2+(1βˆ’1/2)+(1βˆ’1/3)+(1βˆ’1/7)=βˆ’2+1/2+2/3+6/7=1/42.\mu = -2 + (1 - 1/2) + (1 - 1/3) + (1 - 1/7) = -2 + 1/2 + 2/3 + 6/7 = 1/42.

Therefore ∣Gβˆ£β‰€42(2gβˆ’2)=84(gβˆ’1)|G| \leq 42(2g - 2) = 84(g - 1).

β– 
ExampleKlein quartic achieves the Hurwitz bound

The Klein quartic C:x3y+y3z+z3x=0C: x^3 y + y^3 z + z^3 x = 0 in P2\mathbb{P}^2 is a smooth curve of genus 33.

Its automorphism group is G=PSL⁑(2,F7)G = \operatorname{PSL}(2, \mathbb{F}_7) of order 168=84(3βˆ’1)=84β‹…2168 = 84(3 - 1) = 84 \cdot 2.

The quotient C/G≅P1C/G \cong \mathbb{P}^1, and the quotient map has three branch points with ramification types (2,3,7)(2, 3, 7), achieving the minimum μ=1/42\mu = 1/42. This is the unique curve of genus 33 that achieves the Hurwitz bound.

Riemann-Hurwitz and the Riemann Existence Theorem

RemarkWhich ramification data are realizable?

Over C\mathbb{C}, the Riemann Existence Theorem states that any compatible ramification data can be realized by an actual cover. More precisely, given:

  • a Riemann surface DD of genus gDg_D,
  • branch points Q1,…,Qr∈DQ_1, \ldots, Q_r \in D,
  • a monodromy representation ρ:Ο€1(Dβˆ–{Q1,…,Qr})β†’Sn\rho: \pi_1(D \setminus \{Q_1, \ldots, Q_r\}) \to S_n,

there exists a unique (up to isomorphism) degree-nn branched cover f:C→Df: C \to D realizing this data. The genus gCg_C is then determined by Hurwitz.

The Hurwitz formula thus provides the necessary condition (the genus must be a non-negative integer), while the Riemann Existence Theorem provides sufficiency (any combinatorially valid data is realized).

Hurwitz's counting problem: The number of degree-nn covers of P1\mathbb{P}^1 with prescribed ramification is a classical enumerative problem. When all ramification is simple (each e=2e = 2), the count is given by the Hurwitz numbers, which are connected to the representation theory of SnS_n via the ELSV formula.


Summary of the Proof

RemarkLogical structure

The proof of the Hurwitz formula follows this chain:

  1. Functoriality of differentials: a finite separable morphism f:C→Df: C \to D pulls back rational differentials on DD to rational differentials on CC.

  2. Local calculation: at a ramification point with index ee, the pullback fβˆ—(ds)=eβ‹…uβ‹…teβˆ’1dtf^*(ds) = e \cdot u \cdot t^{e-1} dt acquires a zero of order eβˆ’1e - 1 (in the tame case).

  3. Global identity: assembling the local data gives KC∼fβˆ—KD+RK_C \sim f^*K_D + R.

  4. Degree formula: taking degrees, deg⁑KC=2gCβˆ’2\deg K_C = 2g_C - 2, deg⁑(fβˆ—KD)=n(2gDβˆ’2)\deg(f^*K_D) = n(2g_D - 2), and deg⁑R=βˆ‘(ePβˆ’1)\deg R = \sum(e_P - 1), yielding the Hurwitz formula.

The formula is remarkably powerful: it constrains which covers can exist, bounds automorphism groups, computes genera of explicit curves, and underlies the combinatorics of Hurwitz numbers and the topology of moduli spaces.