Vous êtes sur la page 1sur 10

THE RIEMANN-HURWITZ FORMULA

JUSTIN LANIER
Abstract. In this paper, we discuss the Riemann-Hurwitz formula, first as a theorem
about topological surfaces, and then as a theorem about Riemann surfaces. We then prove
the result as a theorem about algebraic curves over an arbitrary field as a consequence of
the Riemann-Roch theorem. To conclude we discuss extensions of the Riemann-Hurwitz
formula to the case of graphs. We give illustrative examples throughout.

1. Covering Maps
We begin in a topological setting. You have probably wrapped a rubber band around
your finger before. By doing so, you took a circle and then doubled or tripled it up. If you
looked at any little section of your looping, it would look like you might have used two or
three little rubber bands all stacked up, rather than a single long rubber band wrapped
multiple times. Depictions of these scenarios are given in Figure 1.
A multiple covering of a circle by a circle is one of the simplest examples of a covering
map, a map from one topological space to another that is a local homeomorphism and
where the inverse image of any point is a collection of points of cardinality n. This n is
called the degree of the map.
In this paper we will mostly be concerned with the case when the spaces involved are
orientable surfaces, that is, orientable manifolds locally homeomorphic to R2 . Such surfaces
are uniquely determined by their genus, which can be thought of as the number of holes

Figure 1. A triple cover of a circle by a circle. A triple cover of a circle


by three circles.
1

JUSTIN LANIER

Figure 2. A triple cover of one surface by another. What is the genus of


each surfaces? How do these numbers relate?
in a surface. In Figure 2 we have a first example of a covering map of surfaces that is very
similar to our rubber band example above. Imagine triple-wrapping the one with more holes
around the one with fewer holes, just as we did before with the circles. This is a covering
map. What can be said about how the number of holes in the two surfaces compare? The
result that is the full answer to this question is know as the Riemann-Hurwitz formula,
which we will build to in what follows.
In our example, we might at first think that since we have a triple cover, the number of
holes in the first surface ought to be three times the number of holes in the second surface.
Indeed, for each of the holes going around the loop downstairs, there are three holes going
around the loop upstairs. However, this does not take into account the holes in the centers
of the two surfaces! The upstairs surfaces has genus g = 7, rather than 6, and the surface
downstairs has genus g 0 = 3, not 2. So instead of the relation g = n g 0 , where n is the
degree of the covering maps, it appears that we should have instead,
g 1 = n (g 0 1).
Does this formula hold for covering maps of surfaces in general, or is it an accident
of our particular example? To proceed, we will do the same kind of count in terms of
Euler characteristic rather than genus. The Euler characteristic of a manifold is the
alternating sum of the number of n-cells in a cell decomposition of the manifold; this
number is independent of which cell decomposition is taken. In the case of surfaces, we
can also define genus in terms of Euler characteristic: = 2 2g.
Using Euler characteristic instead of genus is convenient because Euler characteristic
plays nicely with covering maps. While we had to make an adjustment when trying to
relate the genus of the two surfaces in our example, such an adjustment is not necessary
when we consider Euler characteristic. To begin, take any cell decomposition of the target

THE RIEMANN-HURWITZ FORMULA

Figure 3. A covering map of a genus 7 surface onto a genus 3 surface; a


cell decomposition of each of the surfaces; a single hexagonal cell.
surface. Then pull back this cell decomposition to the upstairs surface via the covering
map. In a covering map, the preimage of any cell must be n copies of that cell. Factoring
out this n from the Euler characteristic summation, we have = n0 . If we now substitute
in our definition of genus in terms of Euler characteristic and rearrange algebraically, we
recover our conjectured formula: g 1 = n (g 1). As we will see, this formula is a special
case of the Riemann-Hurwitz formula. See Figure 3 for an illustration of computing Euler
characteristic in the case of our earlier example.
In this case we can decompose our target surface into eight hexagons. This gives a total
of 48 edges before gluing and 24 after gluing, since the gluing happens in pairs. Similarly,
there are 48 vertices before gluing and 12 vertices after gluing, since four hexagons meet
at every vertex. Thus 0 = 8 24 + 12 = 4. A similar computation shows that = 12,
and we have = 3 0 , as expected.
The important point here is when counting holes before, we didnt exactly have that
for every hole downstairs there were three holes upstairs. But with the cell decompositions,
we do exactly have this correspondenceevery cell downstairs has three corresponding cells
upstairs. This is what we meant when we said that Euler characteristic plays nicely with
covering mapsthe correspondence it produces is clean and clear.
Much of the background for this discussion about covering spaces and Euler characteristic
can be found in Farb-Margalit [3] and Hatcher [4].
2. Branched covering maps
There are many natural maps between surfaces that are close to being covering maps
but fail to be so at finitely many points. For instance, this occurs with group actions on
surfaces where a point of the surface is fixed, as in a rotation. The special points upstairs
are called branch points, and their images under the covering map are called ramification
points. Consider the map between a genus 5 surface and a genus 1 surface depicted in

JUSTIN LANIER

Figure 4. A branched cover from a surface of genus 5 to one of genus 1.


How many preimages do the two bold points have?
Figure 4. You may think of the downstairs surface as given by a quotient map of a rotation
by 1/5 of a turn. Fundamental domains of the group action are drawn in upstairs.
First, note that the target surface is in fact a closed surface of genus 1, as the arcs on
the left are identified with the arcs on the right by the group action. In the figure, two
special points in the target surface are highlighted, one as a white dot and the other as
a grey dot. All other points in the target have five preimagesone in each fundamental
domainbut these two special points each have only a single preimage.
Instead of considering preimages of individual points, we can instead consider again
preimages of some cell decomposition of the target. Regardless of the cell decomposition
chosen, each vertex, edge, and face will have five preimagesexcept for the two special
points. If we do a cell-by-cell Euler characteristic count as we did in our earlier example,
we will find that = 5 0 holds only approximately. For each of the two branch points, we
only have a single corresponding vertex upstairs instead of the expected five, and so our
Euler characteristic count falls a total of 4+4 short of the expected 5 0 . We need to
add these back in for our calculation to be a true equality: = 5 0 +8. If we also again
make the substitution = 2 2g and rearrange, we have in our example
2(5 1) = 8 = 5 2(1 1) + (4 + 4)
Doing this kind of count in generalmaking a correction term that adds back missing vertices lost due to ramificationyields the Riemann-Hurwitz formula for topological
surfaces.
Theorem (Riemann-Hurwitz for topological surfaces). Given two topological surfaces Sg
and Sg0 and a branched covering map f : Sg Sg0 of degree n, we have
X
2(g 1) = n 2(g 0 1) +
(mp 1)
pSg

where mp is the ramification index at p Sg .

THE RIEMANN-HURWITZ FORMULA

Figure 5. An example of a 3-fold branched covering map that does not


come from a group action. The map is from a surface of genus 3 to a sphere.
The diagram is by Winarski [7].
Note that all of the (branched) covering maps described so far have been given via group
actionsZ/3Z and Z/5Z, respectively. Group actions are indeed a rich source of covering
maps of surfaces, but not all covering maps occur in this way. For example, consider the
branched covering map depicted in Figure 5 and discussed by Winarski [7]. This is a 3-fold
cover of a genus 3 surface over a sphere that does not come from a group action. We can
describe the cover as follows. Cut the upstairs surface along the three gamma curves, each
of which will descend to the gamma curve in the target. Of the four pieces obtained, the
top one descends to the upper hemisphere of the sphere, and the bottom one descends to
the lower hemisphere. The two remaining pieces are each quotiented by a half-turn rotation
about a horizontal axis to obtain disks with respectively eight and two branch points, and
then each disk is mapped to a hemisphere of the sphere, with the boundary of the disk
mapping to gamma, the equator. There are a total of ten branch points that arise as a
result of these quotientings. Despite the fact that this branched cover does not come from
a group action, the Riemann-Hurwitz formula still applies. Indeed, we may calculate:
2(3 1) = 4 = 6 + 10 = 3 2(0 1) +

10
X

1,

since there are ten ramification points, all with ramification index 2.

JUSTIN LANIER

Figure 6. An illustration of the branched cover f : C C given by


f (z) = z 3 at z = 0.

3. Riemann-Hurwitz for Riemann surfaces


We now carry over our discussion from the realm of topology to the realm of complex
geometry. We wish to equip our topological surfaces with complex structures and then
show that the natural maps between these enriched surfaces are branched covering maps.
By doing so, we will show that the Riemann-Hurwitz theorem also holds for natural maps
(i.e. holomorphic maps) between Riemann surfaces.
Since a surface is a manifold that is locally homeomorphic to R2 , we may think of
a surface as a collection of patches of R2 with continuous transition maps between the
patches. For instance, we can construct a torus along these lines by taking the usual
identification of a square torus and covering it with disks: one patch that covers the
interior, one patch for each edge that covers the edges interior, and one patch that covers
the single vertex. Similarly, we can construct a Riemann surface as patches of C along
with holomorphic transition maps between the patches.
It is a basic fact from complex analysis that any holomorphic function from one Riemann
surface to another may be locally modelled by some power of z. If this power of z is 1, then
locally we have a covering map. If this power is some n greater than 1, then locally we
have a branched covering map of degree n. Thus any holomorphic map from one Riemann
surface to another has an underlying branched cover of topological surfaces where the
ramification indices are given by the exponents of the local models. We may thus appeal
to our proof of the Riemann-Hurwitz theorem for topological surfaces and recover the result
for holomorphic maps of Riemann surfaces.

THE RIEMANN-HURWITZ FORMULA

4. Riemann-Hurwitz for algebraic curves


There is a correspondance between Riemann surfaces and algebraic curves over C. In
what follows, we give a proof of the Riemann-Hurwitz formula that holds over an arbitrary
field k. We follow the proof as given by Osserman [6]. We take as our starting point the
Riemann-Roch formula.
Theorem (Riemann-Roch). Let X be a curve of genus g and D a divisor on X of degree
d. Then
l(d) dimk (D) = d + 1 g.
A corollary of Riemann-Roch is that for a non-zero rational differential form on a
curve X of genus g, D() has degree 2g 2.
Since we are now operating over an arbitrary fieldpossibly of non-zero characteristic
a new possibility arises regarding ramification. The characteristic of the field may divide
a ramification index, and so a contribution to our sum of ramification indices could get
zeroed out. This can turn the equality in the Riemann-Hurwitz formula into an inequality
instead. Such a zeroing out ramification is called a wild ramification. A ramification
where this does not happen is called tame.
With these definitions and results in hand, we are prepared to prove a version of the
Riemann-Hurwitz for algebraic curves over an arbitrary field.
Theorem (Riemann-Hurwitz for algebraic curves). Let : X Y be a separable morphism of curves of degree d. Let gX and gY be the genus of X and Y respectively. Then
X
2gX 2 d (2gY 2) +
(eP 1)
P X

with equality if and only if is tamely ramified.


Proof. Start by taking to be a nonzero rational differential form on Y , such that the
pullback is also nonzero. The existence of such an is guaranteed by separability. By
the corollary to Riemann-Roch mentioned above, we have that D() has degree 2gY 2
and D( ) has degree 2gX 2. These are two of the major terms in the statement of
the theorem and have taken on the role that Euler characteristic played in the topological
setting. The proof would be complete if we had the following inequality,
X
D( ) D() +
(eP 1)[P ],
P X

for then we could take degrees. That is, we could then plug in our 2gY 2 and 2gX 2
terms, and since the degree of (eP 1)[P ] is just (eP 1), this yields the theorem.
To obtain this inequality, then, consider any P in X. Let s be a local coordinate at P ,

and let t be a local coordinate at (P ). Write = f tn dt for some f in O(P


),Y , where
n is by definition the coefficient of [P ] in D(). By further unravelling of definitions, we
have that the coefficient of [P ] in = ( f tn ) dt is its order of vanishing at P . This is
the sum of P f tn and the order of vanishing of dt at P . But by the definition of the
ramification index, we have that P f tn = nP t = eP n, and this last expression is equal

JUSTIN LANIER

to the coefficient of [P ] in both D(f tn ) and D(). Finally, we have equality if and
only if is tamely ramified, for otherwise a term in the summation may be zeroed out, as
noted above.

5. Riemann-Hurwitz for graphs
A prevalent trend in certain areas of mathematics is taking more classical results about
surfaces and translating them over to the realm of graphs. For instance, much progress has
been made in analyzing the group of outer automorphisms of free groups by considering
mapping class groups of graphs, in analogy with mapping class groups of surfaces. Another
example is how the study of the moduli space of tropical curvesthat is, metric graphs
arose in analogy with with the moduli space of curvesthat is, surfaces. In much the same
spirit, there have been translations in recent years of the Riemann-Hurwitz formula to the
case of graph coverings. We present here one such effort as pursued by Mednykh [5].
First, we define the genus of a graph X to be the rank of H1 (X; Z) or, informally, the
number of independent loops in X. (Note that this diverges from the definition of genus
for surfaces; the number of independent loops in a surface is twice its genus.) Now, since
surfaces are manifolds, they are topologically homogeneousno point can be topologically
distinguished from any other. Graphs do not enjoy this propertya point along an edge
is different from a point at a vertex. As such we must modify the definition of ramification
somewhat for graphs, as we may get ramification along a graph edge.
Mednykh restricts to the case where the covering map is induced by a graph automorphism and so an element of a finite group G. Mednykh lays out several possibilities as to
the allowed fixed sets of these automorphisms. We may have that G acts freely on the edges
and without edge inversionsso that no edge xy is taken to edge yx. This case, treated
by Baker-Norine [1] and Corry [2], yields the following analogue to the Riemann-Hurwitz
formula,
g 1 = |G|(g 0 1) +

(|Gx | 1),

xV (X)

|Gx |

where
denotes the size of the stabilizer of the vertex x of the graph X. Note that the
formula matches up exactly with the formula for surfaces, once we take into account the
change in definition of genus.
We will now look at an example of a graph, a graph automorphism that acts freely
on edges, and how it satisfies this Riemann-Hurwitz formula. We take the graph X of
genus 4 depicted in Figure 7 and its quotient X/G by the cyclic group G = Z/4Z. The
quotient graph has genus 1. When we consider the five vertices of X, the hub vertex is fixed
by all four graph automorphisms, while the spoke vertices are fixed only by the identity
automorphism. Hence we have
4 1 = 3 = 4(1 1) + (4 1).
In his paper, Mednykh address the more general case where automorphisms are allowed
to fix edges and invert edges. Mednykh points out that the main difficulty is defining the

THE RIEMANN-HURWITZ FORMULA

Figure 7. A graph X and its quotient X/G under rotation by /2.


quotient graph X/G in the case of inverted edges. One possibility is to turn an inverted edge
into a loop. A second is to replace an inverted edge with a new kind of graph component
called a semi-edge. A third possibility is to drop the inverted edge from the quotient graph
altogether.
The exact statement of a Riemann-Hurwitz formula for graphs where edge inversions
are allowed depends on how these inverted edges are resolved. In fact, Mednykh states and
proves three different versions. We give here the formula proven as Theorem 2 in the case
when an inverted edge is replaced by a semi-edge; the other results are similar. In this case
we have
g 1 = |G|(g 0 1) +

X
vV (X)

(|Gv | 1)

(|Ge | 1) +

eE(X)

|Ge |

eE inv (X)

where |Gv | and |Ge | denote the size of the stabilizer of either a vertex or edge of the graph
X and E inv (X) denotes the set of edges that are inverted by the group action.
In Figure 8 we depict a graph and its quotient under an action of G = Z/2Z Z/2Z.
The first factor swaps the left path and the right path, and the second factor reflects the
graph along a horizontal line. Note that the central vertical edge is inverted under this
latter action and that this is the only edge that is inverted under G. We will now compute
the inputs to Mednykhs Riemann-Hurwitz formula in this example. The genus of X is 4
and the genus of X/G is 1. The cardinality of the group G is 4. Each of the four vertices
is fixed by exactly two elements of G, and so each contribute a 1 to the sum. The four side
edges are fixed only by the identity of G, while the two loop edges are fixed by two elements
of G. The central edge is fixed by two elements and is also inverted by two elements. Thus
we have that
4 1 = 3 = 4(0) + (1 + 1 + 1 + 1) (0 + 0 + 0 + 0 + 1 + 1 + 1) + 2.

10

JUSTIN LANIER

Figure 8. A graph X and its quotient X/G under a Z/2Z action. The
white dot indicates a semi-edge in the quotient.
The proof that Mednykh produces for Riemann-Hurwitz for graphs with inversions and
fixed edges allow is bootstrapped from the results of Baker-Norine and Corry. The idea
of the proof is to first subdivide the graph by adding a vertex, and then it is a matter of
careful bookkeeping.
6. Summing up
We have now seen the Riemann-Hurwitz formula appear in different guises across a
variety of different contexts: surface topology, complex analysis, algebraic geometry, and
graph theory. The formula does not so much link these different areas of math as it stands
as a signpost for how these fields overlap and share common themes, tools, and insights.
References
[1] Matthew Baker and Serguei Norine. Harmonic morphisms and hyperelliptic graphs.
Int. Math. Res. Not. IMRN, (15):29142955, 2009.
[2] Scott Corry. Genus bounds for harmonic group actions on finite graphs. Int. Math.
Res. Not. IMRN, (19):45154533, 2011.
[3] Benson Farb and Dan Margalit. A Primer on Mapping Class Groups. Princeton University Press, 2011.
[4] Allen Hatcher. Algebraic topology. Cambridge University Press, Cambridge, 2002.
[5] A. D. Mednykh. On the Riemann-Hurwitz formula for graph coverings, 2015.
[6] Brian Osserman. The Riemann-Roch and Riemann-Hurwitz theorems. Available at
https://www.math.ucdavis.edu/ osserman/classes/248A-F13/riemann.pdf, 2013.
[7] Rebecca R. Winarski. Symmetry, isotopy, and irregular covers. Geom. Dedicata,
177:213227, 2015.

Vous aimerez peut-être aussi