Branched coverings of Riemann surfaces

Intuition [Beginner]

A branched cover is what you get when you wrap one Riemann surface around another a fixed number of times, allowing the wrapping to pinch together at finitely many points. Think of a stack of identical pancakes pressed onto a base disc: away from the centre, every base point has the same number of pancakes above it; at the centre, the pancakes fuse. The fusion points are the branch points; the integer counting how many pancakes meet at one fused point is the ramification index.

The right way to think about all the covers of a fixed base curve $X$ is as a single category. Objects are pairs (a top curve $Y$, a wrapping map $f:Y\to X$). Morphisms are wrapping-preserving maps between covers. Riemann's existence theorem is the discovery that this category has three completely different but equivalent descriptions: as covers, as subgroups of a fundamental group, and as field extensions of the rational functions on $X$. One geometric object, three equivalent vocabularies.

Branched covers are the bridge between three different worlds: the topology of $X$ (its loops), the analysis on $X$ (its meromorphic functions), and the algebra of $X$ (its function field). Riemann's existence theorem says the bridge is an isomorphism.

Visual [Beginner]

A schematic of a degree-3 branched cover $Y\to X$. Three sheets of $Y$ sit above the base $X$. At one branch point of $X$, two of the three sheets pinch together (ramification index 2 with a third unramified preimage); at another branch point, all three sheets meet (ramification index 3, totally ramified).

The local picture at every branch point is the same as the model map $z\mapsto z^e$ near the origin in $\mathbb{C}$.

Worked example [Beginner]

The cleanest branched cover is $\pi :{\mathbb{P}}^1\to {\mathbb{P}}^1$, $z\mapsto z^n$, for a fixed positive integer $n$. The map has degree $n$: a generic base point $w\ne 0,\infty$ has the $n$ distinct preimages $w^{1/n},{\zeta }_n{w}^{1/n},{\zeta }_n^2{w}^{1/n},\dots$, where ${\zeta }_n=e^{2\pi i/n}$.

There are exactly two branch points, $0$ and $\infty$. Above $w=0$, the unique preimage is $z=0$, and the local model is literally $z\mapsto z^n$, so the ramification index is $n$. Above $w=\infty$, the same calculation in the chart at $\infty$ gives ramification index $n$.

Hurwitz check: $2g_Y-2=n\cdot (-2)+(n-1)+(n-1)=-2n+2(n-1)=-2$, so $g_Y=0$. The cover is the sphere again, consistent with the source curve ${\mathbb{P}}^1$ being genus 0.

What this tells us: a branched cover is determined by its base, its degree, and the ramification at each branch point — and Hurwitz checks that these data are consistent with the genus of the source curve. The same data also determine the cover up to isomorphism, by Riemann's existence theorem.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ and $Y$ be compact Riemann surfaces. A branched covering is a non-constant holomorphic map $f:Y\to X$. Such a map is automatically open, surjective, and finite-to-one; the function-field extension $\mathcal{M}(Y)/f^{\ast }\mathcal{M}(X)$ has finite degree, called the degree of $f$ and denoted $n=\mathrm{deg}f$.

Definition (ramification index). For $y\in Y$ with $x=f(y)\in X$, choose a chart $\psi$ at $y$ with $\psi (y)=0$ and a chart $\phi$ at $x$ with $\phi (x)=0$. The map $\phi \circ f\circ {\psi }^{-1}$ is holomorphic and vanishes at $0$; after a holomorphic change of source coordinate it has the unique normal form $$ z ;\longmapsto; z^{e_y}, \qquad e_y \geq 1. $$ The integer $e_y$ is the ramification index of $f$ at $y$; it is independent of chart choices. The point $y$ is unramified iff $e_y=1$, ramified iff $e_y>1$. The ramification points are $\{y:e_y>1\}$, a finite subset of $Y$. Their images under $f$ are the branch points of $f$, a finite subset $B\subset X$ called the branch locus. Above each $x\in X$, ${\sum }_{y\in f^{-1}(x)}{e}_y=n$.

Definition (the category $\mathrm{BCov}(X,B)$). Fix a compact Riemann surface $X$ and a finite subset $B\subset X$. The category of branched covers of $X$ with branch locus contained in $B$, denoted $\mathrm{BCov}(X,B)$, has:

• Objects: pairs $(Y,f)$ with $Y$ a compact Riemann surface and $f:Y\to X$ a branched cover whose branch locus is contained in $B$.
• Morphisms: a morphism $(Y_1,f_1)\to (Y_2,f_2)$ is a holomorphic map $g:Y_1\to Y_2$ commuting with the structure maps, $f_2\circ g=f_1$.

A branched cover is connected iff $Y$ is connected, Galois (synonyms: normal, regular) iff $\mathrm{Aut}(Y/X):=\{g:Y\to Y\text{ holomorphic},f\circ g=f\}$ acts transitively on each generic fibre.

Theorem (Hurwitz formula, recalled from 04.04.02). For a connected branched cover $f:Y\to X$ of degree $n$, $$ 2 g_Y - 2 ;=; n(2 g_X - 2) ;+; \deg R, \qquad R = \sum_{y \in Y} (e_y - 1), y. $$

Hurwitz constrains $\mathrm{BCov}(X,B)$ numerically; Riemann's existence theorem identifies the category outright.

Counterexamples to common slips.

• The branch locus is on $X$, the ramification points are on $Y$. The two sets live on different curves and have different cardinalities in general; a single branch point can have several ramified preimages.
• Holomorphic does not imply unramified. The local model $z\mapsto z^e$ for $e>1$ is holomorphic but ramified at $0$. Smoothness of the source and target does not eliminate ramification.
• Galois is not generic. A random degree-$n$ cover for $n\ge 3$ has Galois closure of degree $n!$ or a strictly smaller imprimitive value; only special covers are themselves Galois.
• The branch locus is intrinsic. Two non-isomorphic covers can share a branch locus but have different monodromy; the branch locus alone does not determine the cover.

Key theorem with proof [Intermediate+]

Theorem (Riemann existence theorem). Let $X$ be a compact Riemann surface, $B\subset X$ a finite subset, and $x_0\in X\setminus B$ a chosen base point. Then there is an equivalence of categories $$ \mathrm{BCov}_{\mathrm{conn}}(X, B) ;\xrightarrow{;\sim;}; \big{\text{conjugacy classes of finite-index subgroups of } \pi_1(X \setminus B, x_0)\big}, $$ sending a connected branched cover $(Y,f)$ to the stabiliser of a fixed lift ${\tilde{x}}_0\in f^{-1}(x_0)$ under the monodromy action of ${\pi }_1(X\setminus B,x_0)$ on the generic fibre.

Proof. The proof has three steps: (i) restrict to the punctured base, where the cover is unbranched and topological covering-space theory applies; (ii) extend uniquely across the punctures using the local model $z\mapsto z^e$ and finite local monodromy; (iii) verify functoriality, full faithfulness, and essential surjectivity.

Step 1 — restriction to the unbranched cover. Set $X^{\ast }=X\setminus B$ and $Y^{\ast }=f^{-1}(X^{\ast })$. By definition of the branch locus, $f^{\ast }:=f{|}_{Y^{\ast }}:Y^{\ast }\to X^{\ast }$ is a holomorphic local biholomorphism at every point, hence an unbranched topological covering of degree $n$. Standard covering-space theory ^{[Forster §5]} then identifies connected unbranched degree-$n$ covers of $X^{\ast }$, up to isomorphism, with conjugacy classes of index-$n$ subgroups of ${\pi }_1(X^{\ast },x_0)$. The functor sends $(Y^{\ast },f^{\ast })$ to the stabiliser of a fixed lift of $x_0$ under the monodromy action.

Step 2 — extension across the punctures. The substantive direction is essential surjectivity: every finite-index subgroup of ${\pi }_1(X^{\ast },x_0)$ comes from a branched cover of $X$, not just an unbranched cover of $X^{\ast }$. Given a connected unbranched cover $f^{\ast }:Y^{\ast }\to X^{\ast }$ of degree $n$, we extend it to a branched cover $f:Y\to X$ as follows. For each $b\in B$, choose a small punctured disc $D_b^{\ast }=D_b\setminus \{b\}$ around $b$. The restriction of $f^{\ast }$ to $D_b^{\ast }$ is a finite cover of a punctured disc, hence locally a disjoint union of covers of the form $z\mapsto z^e$ (these are the only connected covers of the punctured disc, up to isomorphism, classified by ${\pi }_1(D_b^{\ast })=\mathbb{Z}$ and its finite-index subgroups). Each such piece extends uniquely across $b$ by adjoining one point above which the map becomes $z\mapsto z^e$ in coordinates. Glue these extensions back to $Y^{\ast }$ to obtain $Y$ and $f:Y\to X$. Compactness of $X$ and finiteness of $B$ ensure $Y$ is compact; connectedness of $Y^{\ast }$ implies connectedness of $Y$. The resulting $(Y,f)$ has branch locus contained in $B$, by construction.

Step 3 — functoriality and equivalence. The assignment $(Y,f)\mapsto Y^{\ast }$ is a functor ${\mathrm{BCov}}_{\mathrm{conn}}(X,B)\to {\mathrm{Cov}}_{\mathrm{conn}}(X^{\ast })$, fully faithful (a holomorphic morphism of branched covers restricts to a continuous morphism of unbranched covers, and conversely a continuous morphism of unbranched covers extends uniquely across the punctures by the same local model). Step 2 shows it is essentially surjective. Composing with the standard covering-space equivalence ${\mathrm{Cov}}_{\mathrm{conn}}(X^{\ast })\simeq \{\text{conjugacy classes of finite-index }H\subseteq {\pi }_1(X^{\ast },x_0)\}$ completes the proof. $\square$

The cover is Galois iff the corresponding subgroup is normal in ${\pi }_1(X\setminus B,x_0)$, in which case $\mathrm{Aut}(Y/X)\cong {\pi }_1(X\setminus B,x_0)/H$, the deck-transformation group identified with the quotient.

Bridge. The Riemann existence theorem in the form above is the topological half of a triangle of equivalences. The other two vertices are: the function-field correspondence — every connected branched cover $f:Y\to X$ is the normalisation of $X$ in a finite separable extension of $\mathcal{M}(X)$, and conversely; and Belyi's reformulation — a curve over $\mathbb{C}$ is defined over $\overline{\mathbb{Q}}$ iff it admits a branched cover of ${\mathbb{P}}^1$ ramified only over $\{0,1,\infty \}$. The combinatorics of these covers — counting them with prescribed monodromy — produces Hurwitz numbers, computed by character theory of the symmetric group and, via the ELSV formula, by integration on ${\overline{\mathcal{M}}}_{g,n}$. Each link is a different perspective on the same underlying object: a finite branched cover of a compact Riemann surface.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not currently formalise the category $\mathrm{BCov}(X,B)$ or the Riemann existence equivalence. A target signature, in Lean 4 / Mathlib syntax, sketches the statement:

[object Promise]

The proof depends on names that do not yet exist in Mathlib (the category of compact Riemann surfaces, the local-form-of-holomorphic-map normalisation $z\mapsto z^e$, the branched-extension theorem, and the topological fundamental group of a compact Riemann surface minus a finite set). Each is a candidate Mathlib contribution; until then this unit ships with lean_status: none.

Advanced results [Master]

The function-field form. A compact Riemann surface $X$ has a meromorphic function field $\mathcal{M}(X)$, finitely generated of transcendence degree 1 over $\mathbb{C}$. For $X={\mathbb{P}}^1$, $\mathcal{M}(X)=\mathbb{C}(z)$. Riemann's function-field theorem identifies the category ${\mathrm{BCov}}_{\mathrm{conn}}(X)$ — connected branched covers of $X$, no fixed branch locus — with the opposite of the category of finite separable extensions of $\mathcal{M}(X)$. Concretely: a connected branched cover $f:Y\to X$ produces $\mathcal{M}(Y)$, a finite extension of $f^{\ast }\mathcal{M}(X)$; conversely, the integral closure of $\mathcal{M}(X)\subset K$ inside ${\mathcal{O}}_X$ is a sheaf of ${\mathcal{O}}_X$-algebras whose relative Spec is the cover $Y\to X$. This is the classical normalisation construction; for $X={\mathbb{P}}^1$ it identifies finite covers of the sphere with finite extensions of $\mathbb{C}(z)$, the analytic incarnation of the fundamental theorem of arithmetic for function fields of curves.

Belyi's theorem and dessins d'enfants. A Belyi map is a holomorphic $\beta :Y\to {\mathbb{P}}^1$ with branch locus contained in $\{0,1,\infty \}$. Belyi 1979 ^{[Belyi 1979]} proved that a smooth projective curve $Y$ over $\mathbb{C}$ admits a Belyi map iff $Y$ is defined over $\overline{\mathbb{Q}}$. The "iff" is striking in both directions: if $Y$ is defined over $\overline{\mathbb{Q}}$, an explicit construction (involving repeated polynomial moves of the branch locus into $\{0,1,\infty \}$) produces a Belyi map; conversely, if a Belyi map exists, the cover is determined by combinatorial data on ${\mathbb{P}}^1\setminus \{0,1,\infty \}$, hence cannot involve transcendental moduli. Grothendieck 1984 ^{[Grothendieck 1984]} interpreted the data of a Belyi map combinatorially as a dessin d'enfant: a bipartite graph embedded in $Y$ whose vertices are the preimages of $0$ and $1$, and whose edges are the preimages of the segment $[0,1]\subset {\mathbb{P}}^1(\mathbb{R})$. The dessin determines the cover up to isomorphism, and the action of the absolute Galois group $\mathrm{Gal}(\overline{\mathbb{Q}}/\mathbb{Q})$ on $\overline{\mathbb{Q}}$-points of the moduli of Belyi pairs lifts to a faithful action on dessins, giving a combinatorial handle on $\mathrm{Gal}(\overline{\mathbb{Q}}/\mathbb{Q})$.

Hurwitz numbers and character theory. The Hurwitz number $H_{g,\mu }(X)$ counts (with automorphism weights) connected branched covers of $X$ of fixed degree, branch locus, and partition-typed monodromy $\mu$. For $X={\mathbb{P}}^1$, the simple Hurwitz number $h_{g,n}$ counts connected genus-$g$ covers of degree $n$ branched at the standard simple pattern: one totally ramified point with cyclic monodromy and $r=2g+n-2$ simple ramification points, each with monodromy a transposition. Burnside's formula gives the closed-form $$ h_{g, n} ;=; \frac{r!}{n!} \sum_{\lambda \vdash n} \chi^\lambda(n) \cdot \left(\frac{f^\lambda \cdot |T(\lambda)|}{n!}\right)^r \cdot \frac{1}{f^\lambda}, $$ where ${\chi }^{\lambda }$ is an irreducible character of $S_n$, $f^{\lambda }={\chi }^{\lambda }(\mathrm{id})$, and $T(\lambda )$ is the conjugacy class of transpositions. The double Hurwitz numbers (with two prescribed ramification patterns plus simple ramification) and their generating function admit a remarkable matrix-integral interpretation, related to Witten's conjecture and the KdV / KP integrable hierarchy.

The ELSV formula. Ekedahl-Lando-Shapiro-Vainshtein 2001 ^{[ELSV 2001]} expressed Hurwitz numbers as integrals on the moduli space ${\overline{\mathcal{M}}}_{g,n}$ of stable curves. For a partition $\mu =({\mu }_1,\dots ,{\mu }_n)⊢d$ of weight $d$, $$ H_{g, \mu} ;=; \frac{(2g - 2 + n + d)!}{|\mathrm{Aut}(\mu)|} \prod_{i = 1}^n \frac{\mu_i^{\mu_i}}{\mu_i!} \int_{\overline{\mathcal{M}}{g, n}} \frac{1 - \lambda_1 + \lambda_2 - \cdots \pm \lambda_g}{\prod{i = 1}^n (1 - \mu_i \psi_i)}, $$ where ${\lambda }_i$ are Chern classes of the Hodge bundle and ${\psi }_i$ are the cotangent classes at the marked points. The ELSV formula bridges the combinatorial Hurwitz-number side and the geometry of the moduli space of curves, and is the prototype of the Gromov-Witten / Witten-Kontsevich correspondence.

Stein structure on the punctured base. The punctured base $X\setminus B$ is a non-compact Riemann surface, hence Stein by Behnke-Stein 06.09.01, so $H^q(X\setminus B,\mathcal{F})=0$ for $q\ge 1$ and any coherent analytic sheaf $\mathcal{F}$. The construction of branched covers from monodromy data uses this vanishing: the obstruction to a global lift of a holomorphic section across the punctures lives in $H^1(X^{\ast },\mathcal{O})$, which is zero, so local data assemble into a global cover. The Stein structure also explains why every holomorphic line bundle on $X^{\ast }$ is holomorphically isomorphic to ${\mathcal{O}}_{X^{\ast }}$ — a fact that simplifies the cohomological analysis of branched covers radically.

Modular curves as branched covers. The modular curves $X(N),X_0(N),X_1(N)$ are compactifications of quotients of the upper half plane $\mathbb{H}$ by congruence subgroups of ${\mathrm{SL}}_2(\mathbb{Z})$. The natural map $X(N)\to X(1)={\mathbb{P}}_j^1$ is a Galois branched cover with deck group ${\mathrm{PSL}}_2(\mathbb{Z}/N)$, branched only over the elliptic points $j=0,1728$ and the cusp $j=\infty$. The genus of $X(N)$ is computed by Hurwitz from the index, the number of elliptic points fixed by elements of order 2 and 3, and the number of cusps. These calculations are foundational to the modular-forms approach to class field theory, the Eichler-Shimura relations, and the Langlands programme over function fields.

Synthesis. A branched cover of compact Riemann surfaces is a single mathematical object, but it admits four equivalent descriptions: (i) geometric — a non-constant holomorphic finite-degree map locally modelled on $z\mapsto z^e$; (ii) topological — a finite-index subgroup of ${\pi }_1$ of the punctured base; (iii) algebraic — a finite separable extension of the function field; (iv) combinatorial — a partition-decorated graph (the dessin), in the Belyi case. Riemann's existence theorem is the master equivalence linking (i) and (ii); function-field theory links (i) and (iii); Belyi's theorem and dessins link (iii) and (iv) over $\overline{\mathbb{Q}}$. Each equivalence converts one type of question into another: existence questions become subgroup questions become field-extension questions become combinatorial graph questions. The Hurwitz-number / ELSV side adds a fifth perspective: the count of covers with prescribed monodromy is itself an integral on ${\overline{\mathcal{M}}}_{g,n}$, embedding the branched-cover combinatorics into the Gromov-Witten universe. The geometric content of the unit is exactly that all five perspectives describe the same mathematics; the master tier consists in moving fluently between them.

Full proof set [Master]

Lemma (local form of a non-constant holomorphic map). For $f:Y\to X$ a non-constant holomorphic map of Riemann surfaces and $y\in Y$, there exist charts $\psi$ at $y$ and $\phi$ at $f(y)$ such that $\phi \circ f\circ {\psi }^{-1}(z)=z^e$ for a unique integer $e\ge 1$.

Proof. Choose any chart $(\tilde{\psi },U)$ at $y$ with $\tilde{\psi }(y)=0$ and a chart $(\phi ,V)$ at $f(y)$ with $\phi (f(y))=0$. The function $g:=\phi \circ f\circ {\tilde{\psi }}^{-1}$ is holomorphic on a neighbourhood of $0$ in $\mathbb{C}$ and vanishes at $0$. Write $g(w)=w^e\cdot h(w)$ where $e\ge 1$ is the order of vanishing and $h$ is holomorphic with $h(0)\ne 0$. Choose $\eta$ with ${\eta }^e=h$ on a small neighbourhood (possible because $h(0)\ne 0$ and $\mathbb{C}$ has $e$-th roots locally), and set $z=w\eta (w)$. Then $z^e=w^e{\eta }^e=w^{e}h=g(w)=\phi (f({\tilde{\psi }}^{-1}(w)))$, so the change of coordinate $\psi (p)=\tilde{\psi }(p)\cdot \eta (\tilde{\psi }(p))$ gives $\phi \circ f\circ {\psi }^{-1}(z)=z^e$. Uniqueness of $e$ follows because it equals the order of the zero of $f^{\ast }\phi$ at $y$, an invariant of the map. $\square$

Lemma (extension of a finite cover across a puncture). Let $f^ : Y^* \to D^$beaconnectedfiniteholomorphiccoverofthepunctureddisc$D^ = {0 < |z| < 1}$.Then$f^$extendsuniquelytoabranchedcover$f : Y \to D$ofthedisc,with$f$givennearthepunctureby$z \mapsto z^e$where$e$ is the index of the cover.

Proof. The fundamental group ${\pi }_1(D^{\ast })=\mathbb{Z}$, so connected finite covers of $D^{\ast }$ are classified by index-$n$ subgroups, all of the form $n\mathbb{Z}$. The corresponding cover is the standard map $\tilde{f}:{\tilde{D}}^{\ast }\to D^{\ast }$, $w\mapsto w^n$, where ${\tilde{D}}^{\ast }=\{0<|w|<1\}$. Any connected $n$-fold cover $f^{\ast }:Y^{\ast }\to D^{\ast }$ is isomorphic to $\tilde{f}$, so there is a biholomorphism $\Phi :{\tilde{D}}^{\ast }\to Y^{\ast }$ with $f^{\ast }\circ \Phi =\tilde{f}$. Adjoining one point $\tilde{0}$ to ${\tilde{D}}^{\ast }$ gives the disc $\tilde{D}=\{|w|<1\}$ with $\tilde{f}$ extending continuously (and holomorphically) to $\tilde{f}(0)=0$. Pull this back via $\Phi$: $Y=Y^{\ast }\cup \{\tilde{0}\}$ with the unique complex structure making $\Phi$ extend to a biholomorphism, and $f=\tilde{f}\circ {\Phi }^{-1}$ extends $f^{\ast }$ across the puncture. Uniqueness of the extension follows from the Riemann removable-singularity theorem applied to the extension of $f$. $\square$

Theorem (Riemann existence, full statement and proof). As stated and proved in the Intermediate-tier Key theorem section, using the two lemmas above. $\square$

Corollary (Galois case). Under the equivalence, Galois branched covers correspond to normal finite-index subgroups of ${\pi }_1(X\setminus B,x_0)$, and for such a cover $\mathrm{Aut}(Y/X)\cong {\pi }_1(X\setminus B,x_0)/H$.

Proof. A finite cover $Y^{\ast }\to X^{\ast }$ is regular (i.e., the deck group acts transitively on each fibre) iff the corresponding subgroup $H\subseteq {\pi }_1(X^{\ast })$ is normal; this is a standard fact in covering-space theory ^{[Forster §6]}. Step 2 of Riemann existence preserves this property: the local extension $z\mapsto z^e$ across each puncture is itself Galois with cyclic deck group, so the global cover is Galois iff its restriction to $X^{\ast }$ is. The automorphism group identification follows from the universal-cover construction. $\square$

Corollary (function-field form). The functor $(Y,f)\mapsto \mathcal{M}(Y)$ is an equivalence between ${\mathrm{BCov}}_{\mathrm{conn}}(X)$ and the opposite of the category of finite separable extensions of $\mathcal{M}(X)$.

Proof. Combine Riemann existence with the standard normalisation construction: for any finite extension $K/\mathcal{M}(X)$, the integral closure $\tilde{\mathcal{O}}$ of ${\mathcal{O}}_X$ in $K$ is a coherent ${\mathcal{O}}_X$-algebra, and ${\mathrm{Spec}}_X\tilde{\mathcal{O}}$ is a connected branched cover of $X$ with function field $K$ ^{[Forster §8]}. The branch locus of this cover equals the set of $x\in X$ where ${\tilde{\mathcal{O}}}_x$ is genuinely ramified over ${\mathcal{O}}_{X,x}$, a finite set. Inverse functoriality of normalisation and meromorphic-function-field-extraction (Forster §8) concludes. $\square$

Theorem (Belyi, 1979). A smooth projective curve $Y$ over $\mathbb{C}$ admits a Belyi map $\beta :Y\to {\mathbb{P}}^1$ iff $Y$ is defined over $\overline{\mathbb{Q}}$.

Proof sketch. The "$\Leftarrow$" direction is Belyi's iterative construction: starting with any non-constant rational function $f:Y\to {\mathbb{P}}^1$ with branch locus contained in $\overline{\mathbb{Q}}$, repeatedly post-compose with polynomial maps that move branch points into $\{0,1,\infty \}$ while preserving the $\overline{\mathbb{Q}}$-structure; the construction terminates after finitely many steps. The "$\Rightarrow$" direction uses that branched covers of ${\mathbb{P}}^1\setminus \{0,1,\infty \}$ are classified by combinatorial data (subgroups of the free group on two generators), so any such cover is determined by data definable over $\overline{\mathbb{Q}}$, and hence so is $Y$ itself. The full proof appears in Belyi 1979 ^{[Belyi 1979]}; modern accounts are in Schneps 1994 ^[Schneps]. $\square$

Connections [Master]

• Hurwitz formula 04.04.02. The numerical compatibility constraint between the genus of source and target curves of a branched cover. Riemann's existence theorem, taken with Hurwitz, is the existence-and-numerical-data toolkit for branched covers: existence is categorical (subgroup of ${\pi }_1$), numerics is Hurwitz.

• Holomorphic line bundle on a Riemann surface 06.05.02. Cyclic branched covers of degree $n$ are constructed from line bundles $L$ on $X$ with $L^n=\mathcal{O}(D)$ for an effective divisor $D$: the cover is ${\mathrm{Spec}}_X({⨁}_{i=0}^{n-1}{L}^{-i})$, with $D$ the branch divisor. This is the Kummer / superelliptic construction, and it is functorial in $L$.

• Riemann-Hurwitz on Riemann surfaces 06.05.03. The analytic version of the Hurwitz formula on compact Riemann surfaces; the statement is identical, the proof passes through differential forms instead of canonical divisors. Both versions consume the same ramification data.

• Branch point and ramification 06.02.01. The local model $z\mapsto z^e$ analysed at the level of a single branch point. The category of branched coverings is the global assembly of these local models.

• Riemann surface 06.03.01. The objects of the category $\mathrm{BCov}(X,B)$ are compact Riemann surfaces equipped with a non-constant holomorphic map to $X$; the entire theory rests on the foundational definition of a Riemann surface.

• Genus of a Riemann surface 06.03.02. The genus enters Riemann existence through Hurwitz: given a target genus $g_X$, branch locus, and ramification data, the genus $g_Y$ of any cover with those data is fixed.

• Stein Riemann surfaces 06.09.01. The punctured base $X\setminus B$ is Stein, so $H^q(X^{\ast },\mathcal{F})=0$ for $q\ge 1$. This vanishing is the analytic input to constructing branched covers from local monodromy data — the obstruction to gluing local extensions is itself a coherent-cohomology class on the Stein base.

• Riemann-Roch on compact Riemann surfaces 06.04.01. Riemann-Roch applied to a branched cover relates the dimensions of section spaces of line bundles on $Y$ to those on $X$, via the projection-formula identity $f_{\ast }(L)\otimes M\cong f_{\ast }(L\otimes f^{\ast }M)$. The Castelnuovo-Severi inequality follows.

• Sheaf cohomology 04.03.01. The cohomological obstruction to extending an unbranched cover across a puncture is a class in $H^1$ of a coherent sheaf on the punctured neighbourhood; vanishing of this $H^1$ on the Stein base is what makes the extension possible.

• Fundamental group of the punctured curve. A finitely generated group: free of rank $r-1$ for $X={\mathbb{P}}^1$ punctured at $r$ points, and a one-relator group with $2g+r-1$ generators for a genus-$g$ surface punctured at $r$ points. The combinatorial / group-theoretic side of Riemann existence is controlled by these explicit presentations.

Historical & philosophical context [Master]

Bernhard Riemann introduced branched covers in his 1851 dissertation ^{[Riemann 1857]} and developed the theory systematically in Theorie der Abel'schen Functionen (J. Reine Angew. Math. 54, 1857). His motivating problem was the analytic continuation of multivalued algebraic functions $w(z)$ defined by polynomial equations $P(z,w)=0$: each value of $z$ in general position has ${\mathrm{deg}}_{w}P$ values of $w$, but these values are not single-valued functions of $z$. Riemann's solution was to replace the domain $\mathbb{C}$ (or ${\mathbb{P}}^1$) by a new geometric object — the Riemann surface — on which $w$ becomes single-valued. The construction is exactly a branched cover of ${\mathbb{P}}^1$ branched over the discriminant locus of $P$. The formal categorical equivalence between "compact Riemann surfaces over ${\mathbb{P}}^1$" and "finite extensions of $\mathbb{C}(z)$" is therefore due to Riemann.

The modern categorical formulation appears in Otto Forster's Lectures on Riemann Surfaces (1981) ^[Forster], which presents branched covers as a category from §4 onwards and proves Riemann's existence theorem in §8 via the analytic-sheaf-cohomology framework of Cartan-Serre. Forster's approach is in the line of the Behnke-Stein school: covers of a non-compact Riemann surface are classified analytically through Stein theory and the vanishing of $H^1$ for coherent sheaves on Stein spaces. Alexander Grothendieck's SGA 1 (1960–61) ^[SGA1] generalised the entire framework to schemes via the étale fundamental group ${\pi }_1^{\text{eˊt}}$, recovering Riemann's correspondence as the special case of curves over $\mathbb{C}$ and identifying ${\pi }_1^{\text{eˊt}}$ with the profinite completion of the topological ${\pi }_1$ in that case (Riemann existence over $\mathbb{C}$).

Belyi's 1979 paper ^{[Belyi 1979]} proved the surprising characterisation of curves over $\overline{\mathbb{Q}}$ via covers branched only over $\{0,1,\infty \}$. Grothendieck's Esquisse d'un programme (1984) ^{[Grothendieck 1984]} interpreted Belyi maps combinatorially as dessins d'enfants and proposed a programme — still partly open — to study the absolute Galois group $\mathrm{Gal}(\overline{\mathbb{Q}}/\mathbb{Q})$ through its faithful action on dessins. The Schneps volume The Grothendieck Theory of Dessins d'Enfants (1994) ^[Schneps] is the canonical reference. The Hurwitz-number / ELSV thread originates with Hurwitz's 1891 paper on counting branched covers and was reformulated as an integral on ${\overline{\mathcal{M}}}_{g,n}$ by Ekedahl-Lando-Shapiro-Vainshtein in 2001 ^{[ELSV 2001]}, embedding the combinatorics of branched covers into the Gromov-Witten / Witten-Kontsevich framework.

Bibliography [Master]

[object Promise]