Alternating groups and moduli space
lifting Invariants:
Published paper
hf-can0611591.pdf
Abstract: The genus of a
curve discretely separates decidely different algebraic
relations in two variables to focus us on the connected moduli space Mg. Yet, modern
applications also require a data variable (function) on the curve. The
function gives a morphism from the curve to the Riemann sphere P1
(projective 1-space). The
resulting spaces are versions, depending on our needs for this data
variable, of Hurwitz spaces.
The Main Theorem describes all components of Hurwitz spaces
of 3-cycle covers.
It combines with the much easier Clebsch case of 2-cycle
covers to give a target result for many connectedness results. This
combining – discussed in CFPV-Thm.html
– explains much of the nuance of the Conway-Fried-Parker-Voelklein
result on connected components. The nuance has two parts:
- The appearance of the Schur multiplier of the covering group in
distinguishing components.
- The necessity of repeating the conjugacy classes sufficiently
many times to get a clean
result.
The 3-cycle case is simple enough to see both these contingencies,
while it is hard enough to illustrate what are likely the full set of
issues any like-minded connectedness result will face. The 3-cycle case
also implies many other connectedness results, for which §VI is a
continuation.
[Fr1, §5 and §6] concentrates on
applications directly related to the 3-cycle case, since this result
touches on issues – under the heading of half-canonical classes – that hark
back to Riemann's generalization of Abel's Theorem. We continue
Riemann's program, by seeing how the combination of
works of Artibani and Pirola [AP], Fay [Fa],
Serre [Se], and Shimura [Sh] fill
in aspects of producing automorphic functions on these Hurwitz spaces.
Table of Contents of this Page:
I. A correspondence
between Hurwitz
spaces and braid orbits:
II. The M(ain)R(esult) on a
Specific
Nielsen Class of Covers and Hurwitz space cusps:
III. Illustration of
the Inner/Absolute
Result of Fried-Voelklein:
IV. Illustration
of Conway-Fried-Parker-Voelklein:
V. Inverse Galois and
Theta
Function applications:
V.1. The structure of GQ :
V.2. Production of
automorphic
functions on Hurwitz spaces:
VI. Application to
Modular
Towers:
I. A correspondence
between Hurwitz
spaces and braid orbits: Data for a Nielsen
class consists of r
≥
3 conjugacy classes C in the
data variable monodromy G. Unadorned
notation Ni(G,C) indicates r tuples of generators (g1,…,gr) of G lying in any order in the
classes comprising C,
satisfying the product-one
condition: g1…gr =1 (product in order
of appearance). Usually, an
equivalence relation enters naturally. That interprets an
equivalence
of covers of P1= P1z , with z denoting the variable from 1st
year complex variables, in a given Nielsen class. This paper uses the
two
equivalences, absolute
and inner, standard now for
dealing with classical results and also for the inverse Galois problem.
A Nielsen class generalizes the
genus as an appropriate invariant for a space of covers of the
sphere. For it there are stable-compactification theorems
by which to complete the Hurwitz space H(G,C) of
covers in the Nielsen class (different spaces for different
equivalences). Some Nielsen classes define connected
spaces. To detect,
however, the components of others requires further subtler
invariants, as in #1 above. Identifying components of
Hurwitz spaces, for any Nielsen
class, is equivalent to computing a braid
orbit on the corresponding Nielsen class.
Problems that distinguish between covers in the same connected
component are of two types: number theoretic (say, asking about
definition fields of covers) or
those that tack on an additional geometric property (do some surfaces
also
appear as representatives of another, related, Nielsen class). Neither
can proceed without distinguishing the connected components in a
Nielsen class.
II. The M(ain)R(esult) on
a Specific
Nielsen Class of Covers and Hurwitz space cusps: In [Fr1, Thms 1.2 and 1.3], G=An
(the alternating group), r counts
the data variable branch
points and C=C3r is r repetitions of the 3-cycle
conjugacy class. This Nielsen class defines a
space of degree n, genus g=r-(n-1) ≥ 0 covers we call H(An,C3r)abs,
the absolute Hurwitz space.
When g=0, H(An,C3r)abs
has exactly one component. When g
≥ 1, the parity of the Fried-Serre spin invariant precisely identifies
its
two components H ±(An,C3r)abs.
The spin invariant is a special case of a general lift invariant. It arises whenever you have a central extension H→G where the kernel is prime to the orders of entries in the r-tuple g=(g1,…,gr) in some Nielsen class. Then, each entry gi lifts uniquely to gi* in H having the same order as gi. The lift invariant is the product (in the same order) of these gi*s. Since the corresponding gis have product 1, the lift invariant is in the kernel of H→G. The case we use here is when H is the unique non-split extension of An with kernel Z/2.
The notation for Clebsh's (1872) 2-cycle case would be H(Sn,C2r)abs
(r even):
It has one component, consisting of
genus g=r/2-(n-1) ≥ 0 covers of P1z. Much complication
in the 3-cycle case appears in the long [Fr1, §3.3] where n = 4. The two distinct conjugacy
classes of
3-cycles in A4 is
part of the trouble. Yet, even here there is something
conceptual at the heart of the Hurwitz space approach to
classical spaces.
The compactification of reduced Hurwitz spaces has a
boundary consisting of cusps.
There is a natural pairing – called sh-incidence
– on these cusps from which we can read off much data about the spaces.
It reveals the components, cusp ramification and the
nature of elliptic fixed points. In particular, [Fr1, Prop. 3.5]
produces this pairing for n=r=4,
and both the absolute and inner spaces. It uses this information to
compute their genuses (they are zero) and to prove the spaces aren't
modular curves, despite their
close resemblance to them.
III. Illustration of
the Inner/Absolute
Result of Fried-Voelklein: The other half of [Fr1, Thms 1.2 and 1.3] is similar to
§II,
except it changes the degree n
covers to their Galois closures equipped with an isomorphism of the
covering group to An. These are the inner Hurwitz spaces of covers H(An,C3r)in.
[inv_gal, App.] says
there is a general
relation between inner and absolute
spaces defined by (G, C): Each component of the former is
a natural Galois cover of a component of the latter.
Further, the
covering group is a subgroup of the outer automorphism group of G. That covering group may change
as the components change. The outer automorphism group for
Anis Sn/An and
degree 2 is the largest expected covering degree. [Fr1, Thms 1.2 and 1.3] says that is
exactly what you get for all components of H(An,C3r)in
any n or allowable r.
This same result about inner/absolute space coverings applies also to reduced Hurwitz spaces (mod out by
a natural PGL2(₢) action – ₢
denoting the complex numbers). This allows comparison with modular curves: Replace An
by the dihedral group Dn of order 2n (take n odd), and replace C3r by C2r
with r even, denoting r repetitions of the involution
conjugacy class. The result in this case is that H(Dn,C2r)in
→ H(Dn,C2r)abs
is a Galois cover with group (Z/n)*/{±1},
* indicating invertible integers modulo n. For r=4, the reduced spaces then give
the modular curve cover X1(n) → X0(n).
Here is an example – [Prop.
1.5, twoorbit], described in Ex. 1.5 of this paper – that shows why
figuring the connected components of the inner spaces is a stronger result.
For n≡1 mod 4, and d(n)=(n+1)/2, consider the spaces H(An,Cd(n)4)abs
and H(An,Cd(n)4)in:
the conjugacy classes consist of 4 repetitions of a d(n)-cycle.
In all cases, the absolute spaces are connected. The inner spaces,
however, have two components, defined over a quadratic extension of Q, for n≡1 mod 8, because you
can't braid the outer automorphism of An. There is but one component for
n≡5
mod 8. The result for absolute spaces is a special case of [Thm. 5.5, LO].
IV. Illustration
of Conway-Fried-Parker-Voelklein:
[CFPV-Thm] explains,
and gives examples of, a result that describes exactly the components
of the Hurwitz space, inner or absolute, so long as this condition
holds:
(*) Each conjugacy class appearing in C appears suitably often (condition #2
above). [CFPV-Thm] also has
a brief history of the result.
For example, in our case, since the Schur multiplier of An is Z/2
and the orders of the elements in
C3r is
prime to 2 (the order of the Schur multiplier), CFPV says, for r large, both H(An,C3r)in
and H(An,C3r)abs
would have two components. [Fr1, Thms 1.2 and 1.3] is
much more explicit,
giving precisely the components for each n and r.
The example of this result, with some provisions on G that
don't need to concern us here, appearing in [inv_gal, App.]
has all
conjugacy classes of G appear
in C –
under the same suitably-often condition as in #2 above. The conclusion
is that the
Hurwitz space has but one component. That result gives no
bounds to indicate what suitably often would be.
[CFPV-Thm] explains how you
can combine the Clebsch and 3-cycle results to get the correct
result for any conjugacy classes (modulo condition #2) that includes
the 2-cycle
and 3-cycle cases where there is just one conjugacy class appearing.
In lieu of #1 above, since Sn has the Klein 4-group
as its Schur multiplier, from
a bigger viewpoint Clebsch's easy
result is surprising. The trick, then, in combining it with
the 3-cycle result is to recognize what happened in the 2-cycle case to
the Schur multiplier.
V. Inverse Galois
and Theta Function applications: We combine [Fr1, Thms 1.2 and 1.3]
with [AP].
This gives applications of Hurwitz spaces to properties of Mg, through
half-canonical classes.
V.1. The structure of GQ: We consider howMg
affects a special
case of a generalization of a
famous conjecture of Shafarevich (see [GQpresentation]).
The special case considers the composite Qalt
of all Galois extensions of Q with
group some alternating
group. If Qalt is a
P(pseudo-)A(lgebraically)C(losed) (the name is due to Frey, who showed
that the nilpotent closure of Q is
not PAC [FJ,
Cor. 11.5.7] field), then [FV]: the canonical
map GQ →G(Qalt/Q) has pro-free kernel. This would
establish a valuable example of the conjecture, where – unlike those in
[FV] – the field extension Qalt
has a canonical construction.
A basic result is that a characteristic 0 field K is PAC if and only if each curve
over Q has a K point. The result [Fr1, Thm. 6.15] says that "most"
curves of any given genus appear in one of the families of [Fr1, Thms 1.2 and 1.3], and further,
these families are defined over Q.
That seems to bode well for each curve over Q (or at least suitably general such
curves) having a cover over Q
with odd order branching. That would establish that Qalt is
PAC. Yet, the paper shows such odd order branching
covers would also have a half-canonical divisor class over Q. So, that hope is dashed by [Fr1, Prop. 5.11]. Yet, the
topic generates these continuing questions:
- Can you use other Nielsen classes, not given by odd order
branching, to conclude Qalt
is PAC?
- Though the moduli space, Mg,
of curves of genus g (g >> 0) is known not to be
uni-rational, might there still be some Hilbertian property invocable for
all of its many Q points
relative
to a cover of a Zariski open set of the moduli space?
The
point of #3 is that the PAC property is the only method that has worked
to get such presentations of GQ and
it would be news if it didn't work here. The point of #4 is that Mg must still be
considered an odd duck of a space whose many rational subvarieties
should be usable for diophantine results without the conniptions that I
go through in proving Prop. 5.11.
V.2. Production of
automorphic
functions on Hurwitz spaces: Based on a generalization of
automorphic functions on the
Siegel
upper half-space, we define a Hurwitz-Torelli
automorphic function on a Hurwitz space. The reduced versions of
any Hurwitz space defined by odd order conjugacy classes supports the
analytic continuation of a canonical ϑ function.
As a corollary of [Se], we can decide when this ϑ is even. That is the
case, for example, for the reduced spaces H+(An,C3r)abs,rd (resp. H-(An,C3r)abs,rd) exactly when r (≥n) is even (resp. odd). Here is how it works. For g an odd k-cycle in An, define w(g) to be (k2-1)/8. Extend this definition additively to any product of odd order disjoint cycles.
In [Se, Thm. 2] we take the special case X=P1z, so [Se, exp. (17)] applies. The result is that, whether ϑ is even or odd depends only on the Nielsen class and the value of the spin invariant (II for a branch cycle description of the covers in the orbit. In more detail. For g=(g1,…,gr) in the braid orbit of a Nielsen class Ni(G,C), we get even for the ϑ exactly when the product of the spin invariant g and (-1)Σi=1r w(gi) is 1: Serre's formula written multiplicatively. In the 3-cycle absolute case each w(gi) is 1. For, however, the inner case, each w(gi) is n!/6, which is even (for n≥ 4). So, the ϑ on the spaces H+(An,C3r)in,rd (resp. H-(An,C3r)in,rd) is always even (resp. odd).
We don't know for sure when the corresponding ϑ -null is nonzero,
though Thm. 6.15 shows it is so in many cases; in general if the genus of the covers in the Nielsen class is 1, or if 13n ≥ 12r+16. Whenever it is nonzero, an
explicit power of this ϑ -null is a Hurwitz-Torelli automorphic
function.
The definitions and results use a combination of [Fa]
and [Sh]. The former is closest to where we start,
for that works with moduli spaces of curves as do we. Still, we switch
to the latter of necessity, for the production of automorphic
functions, for that works with global moduli as do we, though our
spaces are reduced Hurwitz spaces, not Siegel upper half-spaces.
VI. Application to
Modular Towers:
A general application of [Fr1, Thms 1.2 and 1.3]
appears in [twoorbit].
This uses the Fried-Serre Lifting invariant formula (Invariance Cor.
2.3)
and the results of this paper to draw a number of detailed conclusions
on Hurwitz spaces for more general – pure-cycle – Nielsen
classes.
As a
special case of its one-orbit result on pure-cycle absolute Nielsen
classes, [LO] suggests there are no distinguishing properties of
these spaces
as n varies. Yet, [twoorbit,
Prop. 5.15] shows the cusp structure of the Modular Towers for the
prime p=2 are strikingly
different between the two cases n≡1 mod 8 and n≡5 mod 8
listed in §III.
The difference shows in the following statement: In the latter case,
starting from level 1, the cusp tree of the tower has a subtree
isomorphic to the cusp tree of a modular tower. We have named such a
subtree a spire [twoorbit, Cor. 5.17]
It
proves the Main Conjecture of Modular
Towers – that high tower levels have general type– in these cases. The
idea: Connectedness results allow drawing conclusions on the
type of cusps on the boundary of the Hurwitz space compactification.
The Main Conjecture then follows from a previous result about the
types of cusps necessary to prove this.
The sh-incidence cusp pairing mentioned in §II has become the whole
story in [twoorbit]
where it is applied to an infinite number of
distinct Modular Towers, each as rich in its own way as the towers of
all modular curves. Since this pairing seems unknown in
the case of modular curves, the paper also includes a memorable
reconstruction of the modular curve cusp tower using this it.
[AP] M. Artebani and P. Pirola, Algebraic
functions with even monodromy, PAMS April 2004.
[Fa] J. Fay, Theta Functions on Riemann Surfaces,
Lecture notes in Mathematics 352, Springer Verlag, Heidelberg, 1973.
[LO] F. Liu and B. Osserman, The Irreducibility of
Certain Pure-cycle
Hurwitz Spaces, Amer. J. Math. # 6, vol. 130 (2008).
[Fr1] M. Fried, Alternating
groups and Moduli Space
Lifting Invariants, to appear Israel Journal.
[Fr2] M. Fried, The Main Conjecture of
Modular Towers
and its higher rank generalization, in Groupes de Galois (Lum.
2004;
eds. Bertrand and Debes), Congres 13 (2006), 165–233.
[FV] M. Fried and H. Voelklein, The embedding problem over an
Hilbertian-PAC field, Annals of Math 135 (1992), 469–481.
[Se] J.-P. Serre, Revêtements a ramification impaire et thêta-caractéristiques, C. R. Acad. Sci. Paris 311 (1990), 547–552.
[Sh] G. Shimura, Abelian Varieties with Complex Multiplication and Modular Functions, Princeton U. Press, 1998.