All possible actions of a cyclic group on an algebraic curve

Written on May 24, 2012

I will present here a small program to compute (a superset of) all possible faithful actions of a cyclic group on a smooth algebraic curve (Riemann surface) of genus g, given the information on a part of the ramification. In other words, the program computes all possible cyclic subgroups of Aut(F) for some curve F of genus g. The actual possibilities may be fewer than the ones computed, because in this first version of the program the only ingredient is Riemann-Hurwitz formula, that is a necessary condition, but not sufficient.

The code is hosted on github.

Let \( F \) be a smooth algebraic curve of genus \( g \), and \( C_r \) a cyclic group of order \( r \geq 2 \) acting faithfully on \( F \). Let \( C = F/G \) be the quotient curve, of genus \( h \), and \( \pi\colon F \to C \) the quotient morphism. Riemann-Hurwitz formula tells us that \[ 2 g - 2 = r(2h - 2) + \sum_{p \in F} (|{\rm Stab}_{C_r}(p)| - 1). \]

In other words, the “nice” formula \( 2g-2 = r(2h-2) \) must be amended with a term coming from the points in which the morphism \( \pi \) is not étale. We can simplify the discrepancy using the fact that \( |{\rm Stab}_{C_r}(p)| \) is the same for all points \( p \) in the same orbit (that is, for all points \( p \) over the same point \( q \in C \)), and we have a total contribution of \[ (|{\rm Stab}_{C_r}(p)| - 1) \cdot |C_r \cdot p| = \left( \frac{r}{|C_r \cdot p|} - 1\right) \cdot |C_r \cdot p| = r - |C_r \cdot p| = r - |\pi^{-1}(q)| \] from the points over \( q \). Writing \( n_q = |\pi^{-1}(q)| \), we can rewrite Riemann-Hurwitz as \[ 2 g - 2 = r(2h - 2) + \sum_{q \in C} (r - n_q). \]

The input data for the program are the genus \( g \) of \( F \) and the information on the counterimage on some branch points of \( C \), that is, some points in which the counterimage fails to have \( r \) points. For example, we may ask for all cyclic actions on a curve of genus \( 4 \) such that there are two points with \( n_q = 1 \) and one point with \( n_q = 2 \).

The first observation is that \( r \) is forced to be a multiple of the lcm of all prescribed \( n_q \): if there are \( n_q \) points in a orbit, an element of \( C_r \) must have order exactly \( n_q \), and so \( n_q \vert r \).

The second observation is that we need to have at least \( g = 2 \) or \( g = 1 \) and one prescribed point of ramification to have a limit on \( r \).

Let \( Q’ \) be the discrepancy in Riemann-Hurwitz formula. The third observation is that we can factor out the prescribed ramification as in \[ Q’ = Q + \sum_{q}(r - n_q) = Q + Nr - c,\] where \( q \) varies among the points with prescribed ramification, \( N \geq 0 \) is the number of points of \( C \) with prescribed ramification, \( c > 0 \) is a constant, and \( Q \geq 0 \) is the (potential) additional contribution from the ramification.

The program is now straightforward. We try every \( r \) that is a multiple of the lcm, starting from \( 2 \) upwards. For every such \( r \) we can express the genus of \( C \) as \[ h = \frac{(2-N)r + (2g - 2 + c) - Q}{2r} = \frac{(2-N)r + c’ - Q}{2r}. \] Since \( h \) is a non-negative integer and the only parameter that can vary is the non-negative \( Q \), there are only a finite number of values for \( h \). For every such values, we need to find out if there is a configuration of branch points that can realize that contribution. To do so, we compute the possible contributions to the discrepancy for a branch point in \( C \), that are of the form \( (i - 1) \cdot r / i \) for a divisor \( i \) of \( r \), and we can recursively find out all linear combinations of those discrepancies that sum to \( Q \).

The only thing still open is to find an upper limit for \( r \). As before, solving for \( h \) gives us \[ h = \frac{(2-N)r + c’ - Q}{2r}. \] Since \( h \geq 0 \) and \( r \geq 2 \), the numerator is non-negative. On the other hand, if we solve for \( Q \) we obtain \[ Q = (2-N)r + c’ - 2 r h \] that is again non-negative.

We obtain three cases depending on the sign of the coefficient of \( r \), that is, depending on the sign of \( 2-N \).

Case \( 2-N < 0 \). In this case, we obtain \[ r \leq \mathrm{min} \left\{ \frac{c’ - 2 r h}{N-2}, \frac{c’ - Q}{N-2} \right\} \Longrightarrow r \leq \frac{c’}{N-2}, \] since both \( h \) and \( Q \) are non-negative.

Case \( 2-N = 0 \). Since \( Q \geq 0 \), we obtain that either \( h = 0 \) or \( r \leq c’/2h \), so we need to care only for a bound when \( h = 0 \) (where, \( Q = c’ \) is constant). Note that for any \( r \), the smallest non-zero contribution to \( Q \) is \( (i-1)\cdot r/i \) for the smallest non-trivial divisor \( i \) of \( r \). In particular, the smallest contribution is at least \( r/2 \). Therefore, we can use as a limit \( r \leq 2c’ \).

Case \( 2-N > 0 \). Of course \( 2-N \leq 2 \); we have \( Q = (2-N-2h)r + c’ \). If \( 2-N-2h \leq 0 \) there are no problems since \( Q \) is bounded and we can limit \( r \) to twice the maximum \( Q \) as in the previous case. If \( 2-N-2h > 0 \), then \( Q = r + c’ \) or \( Q = 2r + c’ \), and it seems that \( r \) cannot be bounded. What happens is that \( c’ \), the difference between \( Q \) and the nearest multiple of \( r \), become relatively smaller and smaller as \( r \) increases, and then \( Q \) cannot be realized with the possible contributions. It is hard enough to estimate a bound on \( r \) based on this, but it is much easier to limit \( r \) using a result of Wiman in 1895 that states that the maximum order of an automorphism of a Riemann surface is \( 2(2g + 1) \).

If you liked this post, please share it with your friends: