Wild Galois representations: elliptic curves over a $2$-adic field with non-abelian inertia action
Nirvana Coppola

TL;DR
This paper characterizes the Galois representations of elliptic curves over 2-adic fields with non-abelian inertia images, providing a basis for explicit computational algorithms in four distinct cases.
Contribution
It classifies possible non-abelian inertia images for elliptic curves over 2-adic fields and details how to explicitly compute the associated Galois representations.
Findings
Inertia image can be $Q_8$ or $SL_2(\mathbb{F}_3)$.
Distinction based on whether inertia degree over $\mathbb{Q}_2$ is even or odd.
Provides an algorithmic framework for explicit Galois representation computation.
Abstract
In this paper we present a description of the Galois representation attached to an elliptic curve defined over a -adic field , in the case where the image of inertia is non-abelian. There are two possibilities for the image of inertia, namely and , and in each case we need to distinguish whether the inertia degree of over is even or odd. The result presented here can be implemented in an algorithm to compute explicitly the Galois representation in these four cases.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
\catchline
Wild Galois representations: elliptic curves over a -adic field with non-abelian inertia action
Nirvana Coppola111 University of Bristol
School of Mathematics, University of Bristol, Fry Building,
University Walk, Bristol, BS8 1UG, United Kingdom
Abstract
In this paper we present a description of the -adic Galois representation attached to an elliptic curve defined over a -adic field , in the case where the image of inertia is non-abelian. There are two possibilities for the image of inertia, namely and , and in each case we need to distinguish whether the inertia degree of over is even or odd. The result presented here are being implemented in an algorithm to compute explicitly the Galois representation in these four cases.
keywords:
Elliptic curves, local fields, wild ramification, Galois representations
{history}
\ccode
Mathematics Subject Classification 2010: 11G07, 11F80
1 Introduction
Let be a -adic field (i.e. a finite extension of or, equivalently, a non-archimedean local field with characteristic [math] and residue characteristic ) and let be an elliptic curve. Since we can always assume that is in short Weierstrass form, , for . Let be the residue field of and let , i.e. is the inertia degree of . Suppose that has potential good reduction, that is it has additive reduction and there exists a finite extension of where acquires good reduction.
Let be a fixed algebraic closure of and let be the absolute Galois group of , which acts on the points of . This induces a representation on the -adic Tate module , which is independent of the prime as long as , in the sense of [8, §2 Theorem 2.ii].
More precisely we will denote by or simply the following representation:
{G_{K}}$${Aut(T_{\ell}(E)\otimes\overline{\mathbb{Q}}_{\ell}),}
which is a -dimensional representation over , so after fixing a basis for we can identify with . Let us consider the restriction of to the inertia subgroup of (here is the maximal unramified extension of ). If is the minimal extension of where acquires good reduction, which exists by [8, §2 Corollary 3], then and the image of inertia is isomorphic to , as a consequence of the Criterion of Néron-Ogg-Shafarevich (see [9, VII §7 Theorem 7.1]). Moreover, it is proved in [6, Theorems 2,3] that the image of inertia can only be one of the following:
[TABLE]
For a more explicit approach, see also [5, Part IV §11,12]. In this paper we focus on the cases where is non-abelian (equivalently non-cyclic), hence it is either or . In [6, Theorem 3], there is a criterion to check whether this holds.
Recall that the quotient is isomorphic to the absolute Galois group of the residue field, which is pro-cyclic and generated by the Frobenius element, that acts as where . We call an Arithmetic Frobenius of , and denote it by , any fixed choice of an element of that reduces to the Frobenius element modulo . In order to compute explicitly the elements in the image of , let us fix an embedding ; in particular we will identify the element of with .
We will prove the following result. We refer to [4] for the notation used for group names and character tables; in particular we denote each conjugacy class by the order of its elements, followed by a letter if there is more than one class with the same order.
Theorem 1.1**.**
Let be an elliptic curve with potential good reduction over a -adic field, let be a prime different from and let be the -adic Galois representation attached to . Suppose that is non-abelian. Let be the discriminant of a (not necessarily minimal) equation for and let be the inertia degree of . Then factors as
[TABLE]
where is the unramified character mapping the (Arithmetic) Frobenius of to , and is the irreducible -dimensional representation of the group given as follows.
- •
If is even and is a cube in , then is the representation of with character
[TABLE]
- •
If is even and is not a cube in , then is the representation of with character
[TABLE]
Moreover the image of inertia is if is a cube in and otherwise.
- •
If is odd and is a cube in (equivalently the image of inertia is ), then is the representation of with character
[TABLE]
- •
If is odd and is not a cube in (equivalently the image of inertia is ), then is the representation of with character
[TABLE]
In the last two cases a generator for the class can be described explicitly (it is in the proof of Theorem 3.9).
This theorem is almost completely proved in [3, §5]. In particular the cases where is even are already known, and here we present a proof for completeness. The cases where is odd are more subtle. Although it can be easily proved that the representation can only be either the one described above or the one which has the same character values for every conjugacy class except for the classes and , which are swapped, it is not trivial to identify which of these two is equal to . In this work we prove that, with the definition of made in the statement of Theorem 1.1, only one of the two possible cases occur for elliptic curves. The method of proof consists of describing explicitly a generator of the class and computing the trace of on it.
2 The good model
In the following, is an elliptic curve over a -adic field , with potential good reduction, such that the Galois action attached to it has non-abelian inertia image .
Lemma 2.1**.**
Let be the field obtained from by adjoining the coordinates of one point of exact order and a cube root of the discriminant of . Then acquires good reduction over and it reduces to on the residue field.
Proof 2.2**.**
Let be a non-trivial -torsion point with coordinates in and let be the slope of the tangent line at . Then after applying the change of coordinates
[TABLE]
we get an equation for over with the same discriminant , of the form
[TABLE]
with (for a detailed computation see [7, §2, Proposition 2.22 and Corollary 2.23]).
Next we prove that . Note that the discriminant of equation (6) is given by
[TABLE]
since and are cubes in , we have that also 27B-A^{3}=27B\big{(}1-\frac{A^{3}}{27B}\big{)} is a cube in . If we show that the quantity is a cube in , then also is. To prove this claim, it is sufficient to show that the valuation in of is strictly positive. Then we conclude using Hensel’s Lemma that the polynomial z^{3}-\big{(}1-\frac{A^{3}}{27B}\big{)} has a root in .
We know by [8, §2 Corollary 2] that acquires good reduction over the field . We write for the normalized valuation on this field, and for the normalized valuation on . As shown in the proof of Theorem 2 in [8], the image of inertia under the Galois action injects into , therefore by the classification of the automorphisms of an elliptic curve over a field of characteristic (see [9, III §10, Theorem 10.1]) it can be non-abelian only if , where is the -invariant of the curve, and therefore we have .
Assume by contradiction that v_{F}\big{(}\frac{A^{3}}{27B}\big{)}\leq 0, or equivalently . By direct computation,
[TABLE]
so we have that the valuation of the numerator is , and the valuation of the denominator is at least . Now
[TABLE]
contradicting the fact that .
Therefore is a cube in and the following is a well-defined change of variables over the field .
[TABLE]
After applying this transformation to the curve (6), we get the model , with . By the computation above, , so and the valuation of the discriminant is . Therefore this model reduces to on the residue field of , and in particular acquires good reduction over .
Computationally it is possible to find the values using the following modified version of the -division polynomial, whose roots are precisely the slopes of all tangent lines at the non-trivial -torsion points (for a proof, see [2, Theorem 1]):
[TABLE]
If is a root of , then the corresponding point has coordinates , .
Let be the maximal unramified extension of , which is equal to the compositum of and . Note that is the minimal extension of where the curve acquires good reduction. Indeed if is such extension then by [8, §2 Corollary ], we have that and so it clearly contains the coordinates of any -torsion point and any cube root of , which by an easy computation can be expressed in terms of these coordinates, so (see [7, §2 Lemma 2.20]). On the other hand E does acquire good reduction over , hence on , so by minimality. Also note that and so the representation factors through and the representation induced here is injective.
We have that , and since we are assuming that is non-abelian then is either or , so . This occurs precisely when the extension given by adjoining the coordinates of is totally ramified of degree , i.e. when the polynomial defined above is irreducible over .
There are several cases to consider:
- •
if is a cube in , then the degree of is exactly ;
- •
if is a cube in but not in , then and ;
- •
if is not a cube in , then .
Moreover the Galois closure of is given by , where is a primitive -rd root of unity; since if it generates a degree unramified extension, we have that is not Galois if and only if the inertia degree of over is odd. Note that this cannot occur if is a cube in but not in , otherwise the extension would be unramified and not cyclic.
3 Proof of the main theorem
We will use the same notation as in Section 2. Since is non-abelian, then the group is also non-abelian. By [3, §2 Lemma 1], the representation factors as , where is the following character:
[TABLE]
and factors through the finite group , which is either or if is even, or if is odd. As a -representation, is irreducible and faithful, and it is given by . The definition of is suggested by the following lemma.
Lemma 3.1**.**
Let be the Arithmetic Frobenius of ; then the eigenvalues of are ; in particular these are real and equal if is even, complex conjugate if is odd.
Proof 3.2**.**
Suppose that . By Lemma 2.1 we can compute the trace of via point-counting on the reduced curve , getting
[TABLE]
Then by [9, V §2 Proposition 2.3], the characteristic polynomial of is
[TABLE]
with roots . For general , the eigenvalues of are the -th powers of the roots of the polynomial above, hence for odd we get , , and for even there is only one double eigenvalue .
We have that for even , and so is central in the group , so it acts as a scalar matrix, with eigenvalue given by Lemma 3.1. Moreover, for any , if , then and have the same residue field and so ; in this case . Otherwise, the unramified part of the extension is given by and therefore has degree , so . In particular .
Suppose first that is even and that . Then we have the following.
Theorem 3.3**.**
If is a -adic field with even inertia degree over , then Theorem 1.1 is true for any elliptic curve with potential good reduction such that the image of inertia under is non-abelian and .
Proof 3.4**.**
Since is even, is equal to its inertia subgroup since either or . As noticed above, acts as the multiplication by a scalar with eigenvalue , therefore is given by the representation restricted to inertia, hence it is a faithful, irreducible -dimensional representation of (which is either or ). Moreover by [8, §2 Theorem 2.ii], the character of this representation has values in . By inspecting the character tables of and on [4], we deduce that each of these groups only has one such representation, the one given in the statement.
For the case , the image of inertia is strictly smaller than , so the argument that the character values are in does not apply directly. However it is still possible to compute , getting a result surprisingly similar to the one in Theorem 3.3.
Theorem 3.5**.**
If is a -adic field with even inertia degree over and is an elliptic curve with potential good reduction over such that the image of inertia under is non-abelian and , then Theorem 1.1 holds for .
Proof 3.6**.**
The difference between and its inertia subgroup is determined by . We will show that the trace of is integer and so the result will follow from the proof of Theorem 3.3.
Recall that is the unramified character given by ; then since the inertia degree of is we have , therefore using the relation and the fact that is a scalar:
[TABLE]
so the eigenvalues of are -rd roots of unity (not necessarily primitive) in ; moreover the order of is exactly since is faithful as a representation of , so not both the eigenvalues can be . Computing the determinant on both sides we obtain that , therefore the eigenvalues of can only be the two distinct primitive -rd roots of unity, with trace . Hence the representation of is the one given in the statement.
From this moment on, we assume that is odd or equivalently that is not Galois. Then is an irreducible faithful representation of dimension of , which is either if is a cube in , or otherwise. Again by looking at the character tables of these two groups on [4], we obtain two possible such representations, both of which extend the representation of inertia described in the proof of Theorem 3.3. These two representations only differ for the character value on the elements of order . So we need a more explicit description of the action of this group to deduce which one is the correct representation. Note that we will only concentrate on the wild subgroup of , so we may assume for simplicity that the whole group is . If the wild subgroup does not change, since this Galois group differs from the previous one by a cubic totally ramified (hence tame) field extension, and the parity of is not affected.
First, we need to describe explicitly this wild group. Recall that if is the reduced curve of the good model for over then there is an injection of the image of inertia into , that is . This injection is obtained as follows: fix an element of inertia, and a point on the reduced curve, then lift it to a point of , which has coordinates in , apply to each coordinate, and then reduce to another point which again lies on . The group contains a copy of the image of inertia and an extra element of order ; applying the same construction, we see that acts as Frobenius on the reduced curve. Now fix and consider the representation which is the -adic representation modulo . This is the Galois representation induced by on ; after fixing a basis for as a -vector space, takes values in . In [7, §4 Figure 4.2] there is a visual interpretation of this action. Note that the two representations described above are identical, since they are both induced by the action of the Galois group on elements of . We will use both interpretations to find the character of the generators of the group under .
Lemma 3.7**.**
There exists a basis of where the matrix representing the image of Frobenius modulo is \left(\begin{array}[]{cc}1&0\\ 0&2\end{array}\right).
Proof 3.8**.**
Let be as in the proof of Lemma 2.1. Then and are the only points of exact order with coordinates in . Otherwise if is another point of order and , then the coordinates of every other point of order would be rational functions with rational coefficients of since , hence these coordinates would be in , thus would be Galois, contradiction.
We know that the good model for reduces to , and by direct computation this curve has the following points of exact order :
[TABLE]
where is a third root of unity in .
After applying the change of coordinates described in Lemma 2.1, reduces to and to . Let be the -torsion point of reducing to . Then under , Frobenius acts trivially on and maps to , that is to the only point that has the same abscissa of , which is in . Therefore if we complete to the basis of with as above, the matrix expressing the Frobenius in this basis is \left(\begin{array}[]{cc}1&0\\ 0&2\end{array}\right), as claimed.
Theorem 3.9**.**
If is a -adic field with odd inertia degree over , then Theorem 1.1 is true for any elliptic curve with potential good reduction such that the image of inertia under is non-abelian.
Proof 3.10**.**
We will denote by the matrix \overline{\rho}(\phi)=\left(\begin{array}[]{cc}1&0\\ 0&2\end{array}\right)\in\operatorname{GL}_{2}(\mathbb{F}_{3}). Now let us choose an element in the inertia subgroup, of order , for example
[TABLE]
It exists since is contained in the image of inertia under , therefore every element of with determinant and -power order is in the image of inertia. Then is given by the matrix \left(\begin{array}[]{cc}2&1\\ 1&1\end{array}\right). The element is an element of order of the group and so if we determine , we determine the irreducible representation . To compute this trace, we look at the trace of . Let be the reduction of modulo . Then a=\left(\begin{array}[]{cc}2&1\\ 2&2\end{array}\right), with trace . This means that . Note that , with the relations generate as a subgroup of (see the presentation of in [4]).
By looking at the character table of the group in [4], we deduce that is either or , so
[TABLE]
Only one of this two numbers is congruent to modulo , namely the one we obtain if . Therefore we have the following character for (note that only the generators of the conjugacy classes of elements outside inertia, which identify the correct representation, are explicitly written).
[TABLE]
Similarly if the inertia image is , we get the following character for :
[TABLE]
as stated.
4 Notes on the implementation
As explained in [1], the representation described here is the dual of the representation on the étale cohomology of . In particular the function GaloisRepresentation implemented in MAGMA computes the Galois representation on the étale cohomology. Concretely, the two only differ by the character value of on the elements of the two conjugacy classes and . The function is currently being improved implementing the result presented here.
Theorem 1.1 and Theorems 3.1 and 3.2 of [1] give a method to describe completely the -adic Galois representation of an elliptic curve with potential good reduction and non-abelian inertia action.
Acknowledgments
The author thanks her supervisor Tim Dokchitser for the useful conversations and corrections. This work was supported by EPSRC.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] N. Coppola, Wild Galois representations: Elliptic curves over a 3-adic field, ar Xiv e-prints (2018) ar Xiv:1812.05651.
- 2[2] T. Dokchitser, Ranks of elliptic curves in cubic extensions, Acta Arithmetica 126 (4) (2007) 357–360.
- 3[3] T. Dokchitser V. Dokchitser, Root numbers of elliptic curves in residue characteristic 2, Bulletin of the London Mathematical Society 40 (3) (2008) 516–524.
- 4[4] T. Dokchitser Group names, (groupnames.org).
- 5[5] N. Freitas A. Kraus, On the symplectic type of isomorphisms of the p-torsion of elliptic curves, Memoirs of AMS (to appear)
- 6[6] A. Kraus, Sur le défaut de semi-stabilité des courbes elliptiques à réduction additive, Manuscripta mathematica 69 (4) (1990) 353–386.
- 7[7] W. Robson, Connections between torsion points of elliptic curves and reduction over local fields, (unpublished M Sc thesis, University of Bristol, 2017).
- 8[8] J.P. Serre J. Tate, Good reduction of abelian varieties, Annals of Mathematics 88 (3) (1968) 492–517.
