November 27, 2006

2000 Mathematics Subject Classification. Primary 11G05.
<ph f="cmbx">Open conditions for infinite multiplicity eigenvalues on elliptic curves</ph>

Bo-Hae Im and Michael Larsen

Department of Mathematics, University of Utah, Salt Lake City, Utah 84112, USA E-mail address : im@math.utah.edu Department of Mathematics, Indiana University, Bloomington, Indiana 47405, USA E-mail address : larsen@math.indiana.edu

1 Introduction

Let K   be a number field, K ¯   an algebraic closure of K   , and G K : = G a l ( K ¯ / K )   the absolute Galois group of K ¯   over K   . Let E   be an elliptic curve defined over K   . There is a natural continuous action of G K   on the countably infinite-dimensional complex vector space V E : = E ( K ¯ ) C   . The resulting representation decomposes as a direct sum of finite-dimensional irreducible representations in each of which G K   acts through a finite quotient group.
In particular, the action of every σ G K   on V E   is diagonalizable, with all eigenvalues roots of unity. In [3, the first-named author showed that for generic σ   , every root of unity appears as an eigenvalue of countably infinite multiplicity. This is true both in terms of measure and of Baire category. However, there exist σ   for which the spectrum is quite different: trivially, the identity and complex conjugation elements; less trivially, examples which can be constructed for an arbitrary set S   of primes, such that ζ   is an eigenvalue if and only if every prime factor of its order lies in S   .
Throughout this paper, we will write Σ ζ   for the subset of G K   consisting of elements σ   acting as ζ   on an infinite-dimensional subspace of V E   ( E   and K   being fixed). For ζ = 1   , a good deal is known. In [2, it is proved that whenever 1   appears as an eigenvalue of σ   at all, we have σ Σ 1   . It follows that Σ 1   is open. By [4, when K = Q   , Σ 1   is all of G K   , and quite possibly this may be true without restriction on K   . We have already observed that Σ ζ G K   for ζ 1   . We can still hope for positive answers to the following progression of increasingly optimistic questions:
Question 1.1. Does Σ ζ   have non-empty interior for all ζ   ?
Question 1.2. Is Σ ζ   open for all ζ   ?
Question 1.3. Do all eigenvalues of σ   acting on V E   appear with infinite multiplicity?
In this paper, we give an affirmative answer to Question  1.1 for all ζ   and an affirmative answer to all three questions for ζ = 1   .
The difficulty in proving such theorems is that placing σ   in a basic open subset U   of G K   amounts to specifying the action of σ   on a finite Galois extension L   of K   . By the Mordell-Weil theorem, E ( L ) C   cannot provide an infinite eigenspace for ζ   . Thus, theintersection of eigenspaces σ U V E σ ζ   is finite-dimensional. Thus, the behavior of a finite collection of rational points must be enough to guarantee the existence of infinitely many linearly independent points on the curve with specified σ   -action.
We would like to thank L. Moret-Bailly for correcting an error in the proof of Proposition  3.1 in an earlier draft of this manuscript.

2 Multiplicity of the eigenvalue -1

In this section, we answer Questions  1.2  and  1.3 for ζ = 1 .  
Proposition 2.1. Let E / K   be an elliptic curve over K   . Suppose 1   is an eigenvalue of the action of σ G K   on V E   . Then the 1   -eigenspace of σ   is infinite-dimensional.
  • Proof. As 1   is an eigenvalue of σ   acting on V E   , it is an eigenvalue of σ   acting on E ( K ¯ ) Q   . Clearing denominators, there exists a non-torsion P E ( K ¯ )   such that σ ( P ) + P E ( K ¯ ) t o r   . Replacing P   by a suitable positive integral multiple, σ ( P ) = P   .
    Let y 2 = f ( x )   be a fixed Weierstrass equation of E / K   . Let P = ( α , f ( α ) )   . As σ ( P ) = P   , we have α K ¯ σ   but σ ( f ( α ) ) = f ( α )   so f ( α ) / K ¯ σ   . Then, f ( α ) / K ( α )   , since K ( α ) K ¯ σ   Note that f ( α ) K ( α ) K ¯ σ   . Let c = f ( α ) K ( α )   . We still have σ   Gal ( K ¯ / K ( α ) )   and σ ( c ) = c   .
    Let E / K ( α )   denote the twist y 2 = c f ( x )   . Then, E   has a rational point P = ( α , f ( α ) )   over K ( α )   . The K ¯   -isomorphism φ : E E   mapping ( x , y ) ( x , f ( α ) y )   sends P   to P   , so P   is of infinite order on E   . By ([2,Theorem5.3, E ( K ¯ σ )   has infinite rank. Let { P i = ( x i , c f ( x i ) ) } i = 1   be an infinite sequence of linearly independent points of E   generating the infinite dimensional eigenspace of 1   of σ   in E ( K ¯ ) C   . Then, σ ( x i ) = x i   and σ ( f ( x i ) ) = f ( x i )   for all i   , since σ ( c ) = c   .
    Let P i = φ 1 ( P i ) = ( x i , f ( x i ) )   . These are points of the given elliptic curve E   such that σ ( P i ) = P i   for all i   , since σ ( x i ) = x i   and σ ( f ( x i ) ) = f ( x i )   .
    The points P i   are linearly independent because the P i   are so. Therefore, { P i 1 } i = 1   generates an infinite dimensional subspace of the 1   -eigenspace of σ   on V E   . This completes the proof.
Theorem 2.2. Let E / K   be an elliptic curve over K   . Then, Σ 1   is open.
  • Proof. We have already seen that if σ Σ 1   , we can choose a point P E ( K ¯ )   of infinite order such that σ ( P ) = P   . By Proposition  2.1 , τ ( P ) = P   implies τ Σ 1   It follows that Σ 1   contains the open neighborhood { τ G K | τ ( P ) = σ ( P ) }   of σ   .
Remark 2.3. The same argument shows that Questions  1.2  and  1.3 have an affirmative answer for ζ = ω   (resp. ζ = i   ) when E   has complex multiplication by Z [ ω ]   (resp. Z [ i ]   ).

3 Interior Points

In this section, we show that for every root of unity ζ   , the set Σ ζ   contains a non-empty open subset. We assume that the order of ζ   is n 3   , the case n = 1   having been treated in [2, and the case n = 2   in Theorem  2.2 .
Our strategy will be to find points Q i E ( K ¯ )   such that the σ   -orbit of Q i   has length n   . For each such point Q i   , we set
R i : = i = 0 n 1 σ i ( Q i ) ζ i (1)
and observe that R i   is a ζ   -eigenvector of σ   provided that it is non-zero.
We therefore begin with the following proposition:
Proposition 3.1. Let X   be a Riemann surface of genus g > 1   with an automorphism σ   of order n   . Then X   contains a non-empty open set U   such that x U   implies that i = 0 n 1 [ σ i x ] ζ i 0   in P i c X C   .
  • Proof. We can regard X   as the group of complex points of a non-singular projective curve whose Picard scheme has complex locus P i c X   . Then P i c X Z [ ζ ]   is the group of complex points of a group scheme whose identity component P i c 0 X Z [ ζ ]   is isomorphic to the φ ( n )   th power of the Jacobian variety of this curve. The action of σ   on X   defines an action on P i c X   , and the map ψ : P i c X P i c X Z [ ζ ]   given by ψ ( y ) = i = 0 n 1 σ i y ζ i   then comes from a morphism of group schemes. The image of ψ   actually lies in P i c 0 X Z [ ζ ]   , and its kernel P ζ 0   is Zariski-closed in P i c X   .
    The set P ζ   of y   such that ψ ( y )   maps to 0 in P i c X C   is the union of all translates of P ζ 0   by torsion points of P i c X   . Applying Raynaud's theorem [7(i.e., the proof of theManin-Mumford conjecture) to the image of X   in P i c X / P ζ 0   , the intersection X P ζ   is finite whenever P ζ 0   does not contain the identity component of P i c X   . It therefore suffices to prove that the Lie algebra of P ζ 0   is a proper subspace of the Lie algebra of P i c X   . We identify the latter Lie algebra in the usual way [1,Ch. 2, §6with H 1 ( X , O X ) = H 0 , 1 ( X )   .
    As V E = ( E ( K ¯ ) Q ) Q C   , the eigenvalues of σ   form a Galois-invariant set. Therefore, it suffices to show some primitive n   th root of unity is an eigenvalue of σ   acting on H 0 , 1   .
    The Hodge decomposition H 1 ( X ( C ) , C ) = H 1 , 0 H 1 , 0 ¯   implies that it suffices to show that ζ   is an eigenvalue of σ   acting on H 1 ( X ( C ) , C )   .
    Let R C ( G )   denote the ring of complex (virtual) representations of G   . For any subgroup H   of G : = σ   , let R G / H   denote the regular representation of G / H   regarded as an element of R C ( G )   , and let I H : = R G R G / H   . In particular, I { 1 } = 0   . Regarded as an element of R C ( G )   , the G   -representation H 1 ( X ( C ) , C ) )   is 2 g + ( 2 h 2 ) I G + [ x ] X / G I S t a b G ( x ) ,   where h   is the genus of X / G   , and S t a b G ( x )   is the stabilizer of any element of X   representing the G   -orbit [ x ]   . This is worked out in the case that h = 0   in [6,Prop. 2.2), but the method (in which the character of H 1 ( X ( C ) , C ) )   as a representation of G   is deduced from the Hurwitz formula and the Lefschetz trace formula) works in general.
    The multiplicity of ζ   as an eigenvalue of σ   on H 1 ( X ( C ) , C ) )   is therefore 2 g 2   plus the number of ramification points of the cover X X / G   . This is positive except in two cases: the cyclic cover P 1 P 1   of degree n   (necessarily ramified over two points) and a degree n   isogeny of elliptic of elliptic curves. As g 2   , neither case applies.
Theorem 3.2. Let E / K   be an elliptic curve over a number field K   . For each root of unity ζ   , there exists a nonempty open subset Σ ζ   of Gal ( K ¯ )   suchthat the multiplicity of the eigenvalue ζ   for σ Σ ζ   acting on E ( K ¯ ) C   is infinite.
  • Proof. Let ζ   be an n   th root of unity. Let λ 1 , λ 2 , λ 3 ,   be the ramification points of a double cover E P 1   , and let λ   denote the cross-ratio of ( λ 1 , λ 2 , λ 3 , )   . Choose a , b K ¯   such that the ordered quadruple ( a , b , ζ a , ζ b   ) satisfies ( ζ a a ) ( ζ b b ) ( ζ b a ) ( ζ a b ) = λ   This is always possible; for instance, setting a = 1   , we get a non-trivial quadratic equation for b   , and since λ   is not 1   or   , we have b , ζ b { a , ζ a }   . Thus the elliptic curves X i : y 2 = ( x ζ i 1 a ) ( x ζ i 1 b ) ( x ζ i a ) ( x ζ i b ) , for i = 1 , , n .   all have the same j   -invariant as E   .
    Let L = K ( a , b , ζ )   . Fix q K   such that L ( q n )   is a Galois Z / n Z   -extension of L   . We claim that Σ ζ   contains the open set U ζ : = { σ G a l ( K ¯ / L ) | σ ( q n ) = ζ q n } .   Let M = L ( q n )   . For N   any number field containing M   , let C N   denote the affine curve over N   S p e c N [ x , y 1 , , y n ] / ( P 1 ( x , y 1 ) , , P n ( x , y n ) , y 1 y n ( x n a n ) ( x n b n ) )   where P i ( x , y ) = y 2 ( x ζ i 1 a ) ( x ζ i a ) ( x ζ i 1 b ) ( x ζ i b ) .   Note that the equation y 1 y n ( x n a n ) ( x n b n ) = 0   merely selects one of the two irreducible components of the 1-dimensional affine scheme cut out by the other equations.
    Let X   denote the compact Riemann surface which is the compactification of C N ( C )   .
    By the Hurwitz genus formula, the genus of X   is ( n 2 ) 2 n 2 + 1   , which is greater than 1   since n 3   . For any n   -tuple ( k 1 , , k n ) { 0 , 1 } n   with even sum, the map
    ( x , y 1 , , y n ) ( ζ x , ( 1 ) k 1 ζ 2 y n , ( 1 ) k 2 ζ 2 y 1 , ( 1 ) k 3 ζ 2 y 2 , , ( 1 ) k n ζ 2 y n 1 ) (2)
    defines an automorphism σ   of C N   and therefore of X   . As the k i   have even sum, σ   is of order n   . If x q n L *   and σ U ζ   , then σ ( x ) = ζ x   , so σ ( y i ) 2 = ζ 4 y i 1 2 ,   and so there exists an n   -tuple ( k 1 , , k n )   with even coordinate sum such that σ   acts on Q : = ( x , y 1 , , y n )   by ( 2 ). By Proposition  3.1 , for all but finitely many values of x   , R : = i = 0 n 1 σ i ( Q ) ζ i   is a non-zero eigenvector of σ   with eigenvalue ζ   .
    Assume now that N   is a finite Galois extension of M   . Consider the morphism from C N   to the affine line over M   given by ( x , y 1 , , y n ) x   . This is a branched Galois cover with Galois group G a l ( N / M ) × ( Z / 2 Z ) n 1   . There exists a Hilbert set of values t M   such that the geometric points lying over x = q n t   in C M   consists of a single G a l ( K ¯ / M )   -orbit or, equivalently, G a l ( M ( y 1 , , y n ) / M ) = ( Z / 2 Z ) n 1   and M ( y 1 , , y n )   is linearly disjoint from N   over M   . As a Hilbert set of a finite extension of L   always contains some Hilbert set of L   ([5,Ch. 9, Prop. 3.3), it follows that there exists t L   such that setting x = q n t   , relative to M   , the extension M ( y 1 , , y n )   is linearly disjoint from N   and has Galois group ( Z / 2 Z ) n 1   . We can therefore iteratively construct a sequence t 1 , t 2 , L *   such that the extensions
    M i : = M ( ( q n t i a ) ( q n t i b ) ( q n t i ζ a ) ( q n t i ζ b ) , , ( q n t i ζ n 1 a ) ( q n t i ζ n 1 b ) ( q n t i a ) ( q n t i b ) )  
    are all linearly disjoint over M   . Let Q i   be a point with x   -coordinate q n t i   , and R i   the corresponding ζ   -eigenvector of σ   given by ( 1 ). We claim that the R i   span a space of infinite dimension. The Q i   do so by [2,Lemma3.12, and as the ζ j   are linearly independent over Q   , it follows that the R i   do so as well.
We conclude with a question that does not seem to be directly amenable to the methods of this paper:
Question 3.3. Does the set ζ C t o r * Σ ζ   of elements of G K   having generic spectrum on V E   always have an interior point?
References

  1. P. Griffiths and J. Harris: Principles of Algebraic Geometry, John Wiley & Sons, New York, 1978.
  2. B. Im: Mordell-Weil groups and the rank over large fields of elliptic curves over large fields, arXiv: math.NT/0411533, to appear in Canadian J. Math.
  3. B. Im: Infinite multiplicity of roots of unity of the Galois group in the representation on elliptic curves, preprint, 2003.
  4. B. Im: Heegner points and Mordell-Weil groups of elliptic curves over large fields, preprint, arXiv: math.NT/0411534.
  5. S. Lang: Fundamentals of Diophantine Geometry, Springer-Verlag, New York, 1983.
  6. M. Larsen: Rigid lattices are Mordell-Weil, preprint, arXiv: math.NT/0401209.
  7. M. Raynaud: Courbes sur une variété abélienne et points de torsion. Invent. Math. 71 (1983), no. 1, 207–233.

Department of Mathematics, University of Utah, Salt Lake City, Utah 84112, USA E-mail address : im@math.utah.edu Department of Mathematics, Indiana University, Bloomington, Indiana 47405, USA E-mail address : larsen@math.indiana.edu