Computation of Hecke eigenvalues (mod $p$) via quaternions

Take $p$ to be 11. The quaternion algebra $D$ is then $D=\left(\frac{-1,-11}{\mathbb{Q}}\right)$. In this example we will work with level $N=1$ as doing so greatly reduces the size of $\Omega(N)$. We will then compute the Hecke eigenvalues on the space of functions on $\Omega:=U\backslash D^{\times}(\mathbb{A}_{f})/D^{\times}(\mathbb{Q})$, where $U:=\mathcal{O}_{p}^{\times}\times\prod\limits_{l\neq p}\mathcal{O}_{\ell}^{\times}$ is not quite $U(1)$ as we replace $\mathcal{O}_{p}^{\times}(1)$ with $\mathcal{O}_{p}^{\times}$. Note that this is the same as computing the Hecke eigenvalues on $\Omega(1)$ of weight $k\equiv 0\mod p^{2}-1$. Indeed, the modularity condition tells us that these are functions on $\Omega(1)$ that are invariant under the action of $\mathcal{O}_{p}^{\times}$, and so can be identified with functions on $\Omega$. The computation in this section is based on notes of Buzzard [Buz].

Firstly, let’s understand the set $\Omega$. By the same argument that $\Omega(N)$ corresponds to (isomorphism classes of) invertible left $\mathcal{O}$-ideals $I$ with $\pi N$-structure, we see that $\Omega$ corresponds just to the invertible left $\mathcal{O}$-ideals $I$. Running the following in MAGMA [BCP97], we compute a maximal order $\mathcal{O}$ of $D$, and then find that its left ideal class set has order 2, whose elements we call $I_{1}$ and $I_{2}$. So $\Omega$ has two elements. We can also compute integer bases for $\mathcal{O}$, $I_{1}$ and $I_{2}$.

>>ΨD := QuaternionAlgebra< RationalField() | -1, -11>;
>>ΨO := MaximalOrder(D);
>>ΨBasis(O);
>>ΨClasses := LeftIdealClasses(O);
>>Ψ#Classes;
>>ΨI1 := Classes[1];
>>ΨBasis(I1);
>>ΨI2 := Classes[2];
>>ΨBasis(I2);

which outputs:

[ 1, i, 1/2*i + 1/2*k, 1/2 + 1/2*j ]
2
[ 1, -i, -1/2*i - 1/2*k, 1/2 - 1/2*j ]
[ 2, -2*i, 1 - 3/2*i - 1/2*k, 1/2 - i - 1/2*j ]

We now want to rewrite $I_{1}$ and $I_{2}$ as adelic points in $\Omega$. For any invertible left $\mathcal{O}$-ideal $I$, $I\otimes\mathbb{Z}_{\ell}$ is a principal $\mathcal{O}_{\ell}$-ideal, generated by any nonzero element $\alpha_{\ell}$ whose reduced norm has minimal $\ell$-adic valuation. To see this, we refer to Corollary 16.6.12 in Voight [Voi21] that any invertible semi-order (lattice that contains 1 and has reduced norm equal to the ring $R=\mathbb{Z}_{\ell}$) is an order, so that $(I\otimes\mathbb{Z}_{\ell})\alpha_{\ell}^{-1}$ is an invertible semi-order, which must then by $\mathcal{O}_{\ell}$. So if we are given a $\mathbb{Z}$-basis for $I$, then $I\otimes\mathbb{Z}_{\ell}$ is generated by a basis element whose reduced norm has minimal $\ell$-adic valuation. Computing the reduced norms of the given basis elements of $I_{1}$ and $I_{2}$ we get $[1,1,3,3]$ and $[4,4,6,4]$ respectively. So we see that $I_{1}=\mathcal{O}$ (which was obvious anyway) and $I_{2}\otimes\mathbb{Z}_{\ell}=\mathcal{O}\otimes\mathbb{Z}_{\ell}$ for all $\ell\neq 2$. Moreover, $I_{2}\otimes\mathbb{Z}_{2}=\mathcal{O}\otimes\mathbb{Z}_{2}\cdot(1-\frac{3}{2}i-\frac{1}{2}ij)$ as 6 has minimal 2-adic valuation. Thus $I_{1}$ corresponds to $w^{1}:=[1,1,\dots]\in\Omega$ and $I_{2}$ corresponds to $w^{2}:=[1-\frac{3}{2}i-\frac{1}{2}ij,1,1,\dots]\in\Omega$ (we use superscripts here to avoid overloading the subscripts, which we want to use for the places). An obvious choice of basis for the vector space of $\overline{\mathbb{F}}_{p}$ valued functions on $\Omega$ are the characteristic functions $\mathbbm{1}_{w^{1}}$ and $\mathbbm{1}_{w^{2}}$.

To compute the Hecke operator $T_{\ell_{0}}$, we will need to work with matrices at the $\ell_{0}$-place. Let’s compute $T_{2}$ and $T_{3}$ with respect to the basis $\{\mathbbm{1}_{w^{1}},\mathbbm{1}_{w^{2}}\}$. We will need isomorphisms $\mathcal{O}\otimes\mathbb{Z}_{2}\cong\mathrm{M}_{2}(\mathbb{Z}_{2})$ and $\mathcal{O}\otimes\mathbb{Z}_{3}\cong\mathrm{M}_{2}(\mathbb{Z}_{3})$. In general, we have an isomorphism

for $x,y\in\mathbb{Q}_{\ell}$ such that $x^{2}+y^{2}=-11$, when $\ell\neq 11$. For $\mathrm{M}_{2}(\mathbb{Z}_{3})$, note that $\mathcal{O}\otimes\mathbb{Z}_{3}$ has $\mathbb{Z}_{3}$-basis $\{1,i,j,ij\}$, by using the basis for $\mathcal{O}$ computed above and noticing that 2 is invertible in $\mathbb{Z}_{3}$. So for any $x,y\in\mathbb{Z}_{3}$ with $x^{2}+y^{2}=-11$, the above isomorphism restricts to an injection $\mathcal{O}\otimes\mathbb{Z}_{3}\hookrightarrow\mathrm{M}_{2}(\mathbb{Z}_{3})$, which must in fact be an isomorphism by maximality. We could take $(x,y)=(\sqrt{-11},0)$ for $\sqrt{-11}$ a root of $x^{2}+11=0$ in $\mathbb{Z}_{3}$, choosing for example the root which is $1\mod 3$ by Hensel’s lemma.

For $\mathrm{M}_{2}(\mathbb{Z}_{2})$ we need to be a little more careful because we need $1,i,\frac{1}{2}i+\frac{1}{2}ij,\frac{1}{2}+\frac{1}{2}j$ to all map to elements in $\mathrm{M}_{2}(\mathbb{Z}_{2})$, rather than just $1,i,j,ij$. If we take $(x,y)=(\sqrt{-15},2)$ for $\sqrt{-15}$ a root of $x^{2}+15=0$ in $\mathbb{Z}_{2}$ (taking for example the root congruent to $1\mod 4$ by Hensel’s lemma), then we see that $\frac{1}{2}+\frac{1}{2}j$ maps to $\begin{pmatrix}\frac{1+\sqrt{-15}}{2}&1\\ 1&\frac{1-\sqrt{-15}}{2}\end{pmatrix}$, which is in $\mathrm{M}_{2}(\mathbb{Z}_{2})$. This gives us our desired isomorphism $\mathcal{O}\otimes\mathbb{Z}_{2}\cong\mathrm{M}_{2}(\mathbb{Z}_{2})$. Note that if we took instead $(x,y)=(2,\sqrt{-15})$, we would then map $\frac{1}{2}+\frac{1}{2}j$ to $\begin{pmatrix}\frac{3}{2}&\frac{\sqrt{-15}}{2}\\ \frac{\sqrt{-15}}{2}&-\frac{1}{2}\end{pmatrix}$, which is not in $\mathrm{M}_{2}(\mathbb{Z}_{2})$.

We now begin our computation of $T_{2}$, starting with the value of $T_{2}(\mathbbm{1}_{w^{1}})(w^{1})$. By definition,

so we reduce to checking, for example, whether $\begin{pmatrix}1&0\\ 0&2\end{pmatrix}\cdot[1,1,\dots]=\left[\begin{pmatrix}1&0\\ 0&2\end{pmatrix},1,1,\dots\right]$ is the same as $w^{1}=[1,1,\dots]$ or $w^{2}=[1-\frac{3}{2}i-\frac{1}{2}ij,1,1,\dots]$ as an element of $\Omega$.

The condition that $\left[\begin{pmatrix}1&0\\ 0&2\end{pmatrix},1,1,\dots\right]=[1,1,\dots]$ in $\Omega$ is equivalent to saying that the $\mathcal{O}$-ideal $\mathcal{J}$ is principal, where $\mathcal{J}$ is defined by $\mathcal{J}\otimes\mathbb{Z}_{2}=(\mathcal{O}\otimes\mathbb{Z}_{2})\cdot\begin{pmatrix}1&0\\ 0&2\end{pmatrix}$ and $\mathcal{J}\otimes\mathbb{Z}_{\ell}=\mathcal{O}\otimes\mathbb{Z}_{\ell}$ for all $\ell\neq 2$. If this was the case, then we see that $\mathcal{J}$ must be generated by an element $\alpha$ of $\mathcal{O}$ (since $\mathcal{J}\otimes\mathbb{Z}_{\ell}\subset\mathcal{O}\otimes\mathbb{Z}_{\ell}$ for all $\ell$) of reduced norm $2=\mathrm{det}\begin{pmatrix}1&0\\ 0&2\end{pmatrix}$. We have previously computed a basis $[1,i,\frac{1}{2}i+\frac{1}{2}ij,\frac{1}{2}+\frac{1}{2}j]$ for $\mathcal{O}$. Writing $\alpha=t+x\cdot i+y\cdot(\frac{1}{2}i+\frac{1}{2}ij)+z\cdot(\frac{1}{2}+\frac{1}{2}j)$, for $t,x,y,z\in\mathbb{Z}$, we compute

For this to be equal to 2, we must have $y=z=0$ and $t^{2}=x^{2}=1$. So $\mathcal{J}$ must be $\mathcal{O}\cdot(1\pm i)$. This satisfies $\mathcal{J}\otimes\mathbb{Z}_{\ell}=\mathcal{O}\otimes\mathbb{Z}_{\ell}$ for $\ell\neq 2$, so we only need to check whether this works at 2. In other words, we need to check if

or

where our isomorphism $\mathcal{O}\otimes\mathbb{Z}_{2}\cong\mathrm{M}_{2}(\mathbb{Z}_{2})$ sends $1+i$ to $\begin{pmatrix}1&-1\\ 1&1\end{pmatrix}$, and $1-i$ to $\begin{pmatrix}1&1\\ -1&1\end{pmatrix}$. Equivalently, we need to check if

We see that neither is true. So we have computed one of the terms in $T_{2}(\mathbbm{1}_{w^{1}})(w^{1})$, namely

As a sanity check, we verify that $\left[\begin{pmatrix}1&0\\ 0&2\end{pmatrix},1,1,\dots\right]=[1-\frac{3}{2}i-\frac{1}{2}ij,1,\dots]=w^{2}\in\Omega$, so that

as we would expect. The isomorphism $\mathcal{O}\otimes\mathbb{Z}_{2}\cong\mathrm{M}_{2}(\mathbb{Z}_{2})$ sends $1-\frac{3}{2}i-\frac{1}{2}ij$ to $\begin{pmatrix}2&\frac{3-\sqrt{-15}}{2}\\ \frac{-3-\sqrt{-15}}{2}&0\end{pmatrix}$. Let $\mathcal{J}$ be the $\mathcal{O}$-ideal generated by the adelic point $\left(\begin{pmatrix}1&0\\ 0&2\end{pmatrix},1,1,\dots\right)$, and $\mathcal{I}=I_{2}$ the ideal generated by $(1-\frac{3}{2}i-\frac{1}{2}ij,1,\dots)$. Then the condition that they correspond to the same element of $\Omega$ is equivalent to the existence of some $\alpha\in D^{\times}(\mathbb{Q})$ such that $\mathcal{J}=\mathcal{I}\alpha$. Checking this locally, this means $\alpha\in(\mathcal{O}\otimes\mathbb{Z}_{\ell})^{\times}$ for $\ell\neq 2$, and at $\ell=2$ we have

We can rewrite this as

so we see that $2\alpha\in\mathrm{M}_{2}(\mathbb{Z}_{2})$, by inverting the matrix and clearing the denominator. It follows that if $\alpha$ exists, we must have $2\alpha\in\mathcal{O}$, and $\mathrm{nrd}(\alpha)=1$, by checking locally. There are finitely many possibilities for $2\alpha$, namely $\pm 2,\pm 2i,\pm i\pm\frac{1}{2}\pm\frac{1}{2}j,\pm 1\pm\frac{1}{2}i\pm\frac{1}{2}ij$. We need to check if any of these satisfies (1) when we replace $\alpha$ with the corresponding matrix in $\frac{1}{2}\mathrm{M}_{2}(\mathbb{Z}_{2})$. A computation shows that one can take $\alpha=\frac{1}{2}(i+\frac{1}{2}(1+j))$, where we need that actually $\sqrt{-15}\equiv 1\mod 8$ for our choice of square root.

Returning to our computation of $T_{2}(\mathbbm{1}_{w^{1}})(w^{1})$, we need to compute the remaining terms $\mathbbm{1}_{w^{1}}\left(\begin{pmatrix}1&1\\ 0&2\end{pmatrix}\cdot[1,1,\dots]\right)$ and $\mathbbm{1}_{w^{1}}\left(\begin{pmatrix}2&0\\ 0&1\end{pmatrix}\cdot[1,1,\dots]\right)$. By the same argument as above, it suffices to check whether

for the first term, and whether

for the second. The first two are true and the last two are not. So we deduce that

Therefore

We can also deduce that $2T_{2}(\mathbbm{1}_{w^{2}})(w^{1})=3-1=2$.

One similarly shows that $2T_{2}(\mathbbm{1}_{w^{1}})(w^{2})=3$ and so $2T_{2}(\mathbbm{1}_{w^{2}})(w^{2})=0$. The required computation is to show that the $\mathcal{O}$-ideals generated by $\left(g\cdot\begin{pmatrix}2&\frac{3-\sqrt{-15}}{2}\\ \frac{-3-\sqrt{-15}}{2}&0\end{pmatrix},1,1,\dots\right)$, for $g=\begin{pmatrix}1&0\\ 0&2\end{pmatrix},\begin{pmatrix}1&1\\ 0&2\end{pmatrix}$ and $\begin{pmatrix}2&0\\ 0&1\end{pmatrix}$, are principal $\mathcal{O}$-ideals. The generators can be taken to be $\alpha=-i+\frac{1}{2}-\frac{1}{2}j,i+\frac{1}{2}-\frac{1}{2}j$ and $2$ respectively.

It follows that $2T_{2}(\mathbbm{1}_{w^{1}})=\mathbbm{1}_{w^{1}}+3\mathbbm{1}_{w^{2}}$ and $2T_{2}(\mathbbm{1}_{w^{2}})=2\mathbbm{1}_{w^{1}}$, so $T_{2}$ has matrix $\frac{1}{2}\begin{pmatrix}1&2\\ 3&0\end{pmatrix}$ with respect to the basis $\{\mathbbm{1}_{w^{1}},\mathbbm{1}_{w^{2}}\}$. The eigenvalues are $\lambda=\frac{-2}{2},\frac{3}{2}\in\overline{\mathbb{F}}_{11}$.

The Hecke operator $T_{3}$ can be computed in a similar way. We give one example evaluating the summand

of $3T_{3}(\mathbbm{1}_{w^{2}})(w^{2})$. One could do this by checking whether the $\mathcal{O}$-ideal $\mathcal{J}$ generated by the adelic point in the square brackets is principal. If it is, then this value of $\mathbbm{1}_{w^{2}}$ is $0$. This only works because $\Omega$ has two elements. In general, we need to check if $\mathcal{J}$ is in the same left ideal class as the $\mathcal{O}$-ideal $\mathcal{I}=I_{2}$ generated by the adelic point $\left(\begin{pmatrix}2&\frac{3-\sqrt{-15}}{2}\\ \frac{-3-\sqrt{-15}}{2}&0\end{pmatrix},1,1,\dots\right)$. So we need to check if there exists $\alpha\in D^{\times}$ such that $\mathcal{J}=\mathcal{I}\alpha$. Checking this locally, we need (for the two isomorphisms $D\otimes\mathbb{Q}_{2}\cong\mathrm{M}_{2}(\mathbb{Q}_{2})$ and $D\otimes\mathbb{Q}_{3}\cong\mathrm{M}_{2}(\mathbb{Q}_{3})$ specified)

and

and $\alpha\in(\mathcal{O}\otimes\mathbb{Z}_{\ell})^{\times}$ for $\ell\neq 2,3$. The first two conditions imply that $2\alpha\in\mathcal{O}\otimes\mathbb{Z}_{2}$ and $\alpha\in\mathcal{O}\otimes\mathbb{Z}_{3}$. So $2\alpha\in\mathcal{O}$, and $\mathrm{nrd}(\alpha)=3$. There are finitely many possibilities for $2\alpha$.

Before we go through all the possibilities for $2\alpha$ and check whether they satisfy the two equations, we remark that for equation (2) we only really need to work with $2\alpha$ and the matrix $\begin{pmatrix}2&\frac{3-\sqrt{-15}}{2}\\ \frac{-3-\sqrt{-15}}{2}&0\end{pmatrix}$ modulo some power of 2 (and for equation (3) we only need $2\alpha$ modulo some power of 3). Indeed, using the fact that our choice of $\sqrt{-15}$ is congruent to $1\mod 8$, we have

Suppose $2\alpha$ has matrix $\begin{pmatrix}a_{2}&b_{2}\\ c_{2}&d_{2}\end{pmatrix}\in\mathrm{M}_{2}(\mathbb{Z}_{2})$. Then equation (2) tells us (after inverting the matrix) that we need

Note that the norm condition on $\alpha$ will then guarantee that we land in $4\mathrm{GL}_{2}(\mathbb{Z}_{2})$ not just $4\mathrm{M}_{2}(\mathbb{Z}_{2})$. Working modulo 4, we need

Expanding this out, this is equivalent to

Similarly, if $2\alpha$ has matrix $\begin{pmatrix}a_{3}&b_{3}\\ c_{3}&d_{3}\end{pmatrix}\in\mathrm{M}_{2}(\mathbb{Z}_{3})$, then equation (3) is equivalent to

So now we look through $2\alpha\in\mathcal{O}$ of reduced norm 12. Write

for $t,x,y,z\in\mathbb{Z}$. We also compute the images of the basis elements $1,i,\frac{1}{2}(i+ij),\frac{1}{2}(1+j)$ in $\mathrm{M}_{2}(\mathbb{Z}_{2})$ and $\mathrm{M}_{2}(\mathbb{Z}_{3})$, then reduce them modulo 4 and 3 respectively:

It follows that

Hence equations (4) and (5) tell us that there exists $2\alpha\in\mathcal{O}$ of reduced norm 12 satisfying equations (2) and (3), if and only if we can find $t,x,y,z\in\mathbb{Z}$ such that

where the last equation tells us the reduced norm is 12. We can check that there are no such solutions. For example, we see that $z$ must be even, and then from the norm condition we see that $\lvert z\rvert\leq 2$. Trying $z=2$, the norm condition tells us that we must have $(t,x,y)=(0,0,0),(-2,0,0)$ or $(-1,\pm 1,0)$, none of which satisfy the congruence conditions. For $z=-2$ we similarly must have $(t,x,y)=(0,0,0),(2,0,0)$ or $(1,\pm 1,0)$ which also does not work. And finally for $z=0$ we need $t^{2}+(x+\frac{1}{2}y)^{2}+\frac{11}{4}y^{2}=12$. Since $3\mid t$, in one case we have $t^{2}=9$, and then in $(x+\frac{1}{2}y)^{2}+\frac{11}{4}y^{2}=3$ there are no solutions with $4\mid x+y$ and $3\mid x$. Otherwise we have $t=0$ and $(x+\frac{1}{2}y)^{2}+\frac{11}{4}y^{2}=12$ with $2\mid y$, $4\mid x+y$ and $3\mid x$, which has no solutions, checking $y=0,\pm 2$.

We deduce that

Performing the remaining calculations, using the relevant mod 3 and 4 matrices we have already written down above, we find that $T_{3}$ has matrix $\frac{1}{3}\begin{pmatrix}2&2\\ 3&1\end{pmatrix}$ with respect to the basis $\{\mathbbm{1}_{w^{1}},\mathbbm{1}_{w^{2}}\}$. We compute the eigenvalues to be $\frac{-1}{3},\frac{4}{3}\in\overline{\mathbb{F}}_{11}$.

Comparing the matrices for $T_{2}$ and $T_{3}$, they are simultaneously diagonalisable (as expected) with eigenvectors $\scriptsize\begin{pmatrix}2\\ -3\end{pmatrix}$ and $\scriptsize\begin{pmatrix}1\\ 1\end{pmatrix}$. The corresponding eigenvalues are $(a_{2},a_{3})=\left(\frac{-2}{2},\frac{-1}{3}\right)=(-1,-4)$ and $(a_{2},a_{3})=\left(\frac{3}{2},\frac{4}{3}\right)=(7,5)$, viewing $a_{2},a_{3}$ as elements of $\overline{\mathbb{F}}_{11}$. In view of Serre’s Theorem 1.1, one might wonder what mod 11 modular forms give rise to these eigenvalues. The first comes from $\theta^{9}(\Delta)=q-q^{2}-4q^{3}+\dots\mod 11$, which has minimal weight filtration 120 (meaning that 120 is the lowest weight at which a mod 11 modular form has this $q$-expansion), and the second from the Hasse invariant $E_{10}$.