Skip to main content

Section 2 The Eichler-Selberg trace formula

Subsection 2.1 The OG approach

What does it do? It calculates the trace of the \(m\)th Hecke operator \(T(m)\) on \(S_k\) the space of holomorphic modular forms of weight \(k\) level 1.

Input \(m \in \ZZ_{\ge 0}\text{,}\) \(k\) weight.

Remark 2.1

It is more general, there is an (Eichler-)Selberg trace formula for general level \(N\text{.}\)

Even more generally there is a Selberg trace formula for Maass forms of arbitrary level.

Even more general Arthur-Selberg trace formula for automorphic representations on any group.

Recall

\begin{equation*} T(m) f(z) = m^{k-1} \sum_{ad=m,\, b \pmod d} d^{-k} f\left(\frac {az+b}{d} \right) \end{equation*}

\(S_k\) cusp forms of weight \(k\ge 2\) even level 1.

\begin{equation*} \trace_{S_k} T(m) = -\frac12 \sum_{t\in \ZZ,\,t^2 - 4m \lt 0} P_k(t,m) H(4 m -t^2) \end{equation*}
\begin{equation*} -\frac12 \sum_{dd' = m} \min(d,d')^{k-1} +\begin{cases} \frac{k-1}{2} m^{k/2 - 1} \amp m\in \ZZ^2 \\ 0 \end{cases}\text{.} \end{equation*}

Where

\begin{equation*} P_k(t,m) = \frac{\rho^k - \bar \rho^k}{\rho - \bar\rho} \end{equation*}

for \(\{\rho,\bar \rho\}\) solutions to \(X^2 - tX + m = 0\text{.}\) \(H(n)\) is the “Hurwitz class number”

\begin{equation*} = \#\{Q \text{ pos. def. integral quad. form}: \disc Q = -N\}/\SL_2(\ZZ) \end{equation*}

each one counted with multiplicity 1 unless it is equivalent to \(x^2+y^2 \to \frac 12\text{,}\) \(x^2+xy+y^2 \to \frac 13\) We will prove this and examine some consequences.

How would we calculate terms in this? It can be hard, the Hurwitz class number requires \(\sqrt D\) time for \(N = f^2D\text{,}\) \(D\) a fundamental discriminant. To get the trace we can integrate \(\int K(x,x)\) for the kernel which gives us \(T(m)\text{.}\)

Towards the trace formula

We will follow Zagier: Modular forms whose coefficients are Dirichlet series.

Normalization: Let \(\{f_1, \ldots, f_r\}\) be a basis for \(S_k\text{.}\) We will assume that they are all Hecke eigenfunctions and they are normalized by \(a_1^i = 1\) (Hecke normalized).

\begin{equation*} f_i(z) = \sum_{i=1}^\infty a_n^i e(nx) \end{equation*}
Note 2.2

\(f_i\)s are pairwise orthogonal with respect to \(\pair{-}{-}_{\Pet}\)

\begin{equation*} \pair{f_i}{f_j}_{\Pet} = \int_{\Gamma\backslash \HH} f_i\bar f_j y^k \frac{\diff x\diff y}{y^2} \end{equation*}

why: say \(a^i_m \ne 0\) for some \(m\) then

\begin{equation*} \pair{f_i}{f_j}_{\Pet} = \frac{1}{a_m^i} \pair{T(m) f_i}{f_j}_\Pet \end{equation*}
\begin{equation*} = \frac{1}{a_m^i} \pair{ f_i}{T(m)f_j}_\Pet \end{equation*}
\begin{equation*} = \frac{a_m^j}{a_m^i} \pair{ f_i}{f_j}_\Pet \end{equation*}
Kernel function
Note 2.4

Let's set \(j_k(\gamma,z) = (cz+d)^{-k}\text{,}\)

\begin{equation*} \gamma = \begin{pmatrix} a \amp b\\ c \amp d\end{pmatrix} \in \SL_2(\ZZ) \end{equation*}
\begin{equation*} F_k(z,z') = (z+z')^{-k} \end{equation*}

so

\begin{equation*} h_m(z,z') = \sum_{\gamma\in \Mat_2(\ZZ),\,\det \gamma = m} j_k(\gamma,z) F_k(\gamma z, z')\text{.} \end{equation*}

Even more is true \(h_m(z,z') = h_1(T_m(z), z')\text{.}\)

\begin{equation*} f\ast h_m(z') = \frac{c_k}{m^{k-1}} T_m(f)(z') \end{equation*}

\(c_k = (-1)^{k/2} \pi/2^{k-3}(k-1)\text{.}\) set \(m = 1\) (this is enough because of our earlier observation).

\begin{equation*} f\ast h_1( z') = \int_{\Gamma \backslash \HH} f(z) \overline{h_1(z,-\bar z')} y^k \diff \mu(z) \end{equation*}
\begin{equation*} \int_{\Gamma \backslash\HH} f(z) \left\{ \sum_{\gamma} j_k(\gamma,z) (-z' + \gamma z)^{-k} \right\} y^k \diff \mu(y) \end{equation*}

\(f\) is a modular form of weight \(k\) implies

\begin{equation*} f(\gamma z) = (cz+d)^k f(z) \end{equation*}
\begin{equation*} \Im(\gamma z ) = \frac{y}{|cz+d|^2} \end{equation*}

so

\begin{equation} \overline{(cz+d)^{-k}} f(z) y^k = f(\gamma z) \Im(\gamma z )^k\text{.}\tag{2.1} \end{equation}

this implies

\begin{equation*} \int_{\Gamma\backslash \HH} f(\gamma z) \Im('g z)^i (-z' + \overline{\gamma z})^{-k} \frac{\diff x\diff y}{y^2} \end{equation*}
\begin{equation*} = 2 \int_0^\infty \int_{-\infty}^{\infty} f(z) y^k (-z' + \bar z) ^{-k} \frac{\diff x \diff y}{y^2} \end{equation*}

Cauchy integral formula implies

\begin{equation*} \int_{-\infty}^\infty \frac{f(x+iy)}{(x -iy -z')^k} \diff x \end{equation*}
\begin{equation*} 2 \pi i/(k-1)! f^{(k-1)}(2\pi i y) \end{equation*}
\begin{equation*} 4\pi i /(k-1)! \int _0^\infty f^{(k-1)} (2iy + z') y^{k-2} \diff y \end{equation*}

inner part is

\begin{equation*} 1/(2i)^{k-2} \frac{\diff^{k-2}}{\diff t^{k-2}} f' (2 i ty + z') |_{t=1} \end{equation*}
\begin{equation*} 4 \pi i/(k-1)! 1/(2i)^{k-2}\frac{\diff^{k-2}}{\diff t^{k-2}} \int_0^\infty f'(2ity + z') \diff y |_{t=1} \end{equation*}
\begin{equation*} = -4\pi i /(k-1)!(2i)^{k-1} \frac{\diff^{k-2}}{\diff t^{k-2}} \frac1t f(z')|_{t=1} \end{equation*}
\begin{equation*} = (i/2)^{k-1}4\pi i /(k-1) f(z') = c_kf(z')\text{.} \end{equation*}

Expand \(h_m(z,z')\) with respect to \(\{f_1, \ldots, f_r\}\text{.}\) \(h_m(z,z')\) is a cusp form in both variables, we will show this later!

\begin{equation*} h_m(z,z') = \sum_{1\le i,j\le r} \alpha_{i,j} f_i(z) f_j(z'),\, \alpha_{i,j} \in \CC \end{equation*}

By the theorem

\begin{equation*} \frac{c\inv_k}{m^{k-1}} f\ast h_m(z') = T_m(f)(z') \end{equation*}

let's apply this to \(f_{i_0}\) for any \(i_0\) .. Next time.

Summary:

Aim: to establish the Eichler-Selberg trace formula for the trace of \(T_m \acts S_k\text{,}\) for \(T_m\) the \(m\)th Hecke operator

\begin{equation*} (T_m f )(z ) = m^k \sum_{j=1}^r j_k(\gamma ,z ) f(\gamma_j z) \end{equation*}

where \(\gamma_j\) are coset representatives for

\begin{equation*} \Gamma\backslash \{ x\in \Mat_2(\ZZ) : \det(x) = m\}/\Gamma = \bigsqcup \Gamma \gamma_i \end{equation*}

and \(j_k(\gamma z ) = (cz+ d)^{-k}\text{,}\)

\begin{equation*} \gamma = \begin{pmatrix} a \amp b\\ c \amp d\end{pmatrix} \in \SL_2(\ZZ)\text{.} \end{equation*}
\begin{equation*} h_m(z,z') = \sum_{ad-bc = m }(czz' + dz' + az+ b)^{-k} \end{equation*}

For the proof set \(g_k(z,z') = (z+z')^{-k}\text{.}\) This implies that

\begin{equation*} h_m(z,z') = \sum_{\gamma\in \Mat_z(\ZZ),\det(\gamma) =m} j_k(\gamma; z) g_k(\gamma z ,z') \end{equation*}
\begin{equation*} = \sum_{\gamma\in \Mat_z(\ZZ),\det(\gamma) =m} j_k(\begin{pmatrix} 0 \amp 1 \\ 1 \amp 0 \end{pmatrix}^\transpose \gamma\begin{pmatrix} 0 \amp 1 \\ 1 \amp 0 \end{pmatrix}; z) g_k(z, \begin{pmatrix} 0 \amp 1 \\ 1 \amp 0 \end{pmatrix}^\transpose \gamma\begin{pmatrix} 0 \amp 1 \\ 1 \amp 0 \end{pmatrix}z') \end{equation*}
\begin{equation*} = \sum_{\gamma_1 \in \Mat_2(\ZZ), \det(\gamma_1) =m} j_k (\gamma z; z') g_k(z; \gamma_1 z') \end{equation*}

which implies that the integral is well defined.

  1. Reduce part 1. to the case of \(T_1\text{.}\)
    \begin{equation*} T_m(h_1(z, z')) = \sum_{\gamma \in \Gamma} T_m(j_k(\gamma; z) g_k(z; z')) \end{equation*}
    \begin{equation*} \sum_{\gamma\in \Gamma} m^{k-1} \sum_{j = 1}^r j_k(\gamma; \gamma_jz) g_k(\gamma\gamma_jz,z') j_k(\gamma_j, z) \end{equation*}
    \begin{equation*} \sum_{\gamma\in \Gamma} m^{k-1} \sum_{j = 1}^r j_k(\gamma\gamma_j; z) g_k(\gamma\gamma_jz,z') \end{equation*}
    \begin{equation*} m^{k-1} \sum_{\gamma_1 \in \Mat_2(\ZZ), \det(\gamma_1) =m} j_k(\gamma; z) g_k(\gamma z,z') = m^{k-1} h_m(z,z')\text{.} \end{equation*}
    For \(h_1\) our calculation last time established
    \begin{equation*} c_k\inv \int_{\Gamma\backslash \HH} f(z) \overline{h_1 (z, - \bar z')} y^k \diff \mu(z) = f(z')\text{.} \end{equation*}
  2. Fix \(z'\text{.}\) Implies
    \begin{equation} c_k\inv m^{k-1} h_m(z,z') = \sum_{j=1}^{\dim(S_k)} \frac {\alpha_j (z')}{\pair{f_j}{f_j}_{\Pet}}\label{eqn-pet-proof-1}\tag{2.2} \end{equation}
    by the first part
    \begin{equation} c_k\inv m^{k-1} \pair{f_k}{h(\cdot, z')}_\Pet = T_m(f_k)(z') = a_k(m) f_k(z')\label{eqn-pet-proof-2}\tag{2.3} \end{equation}
    as \(f_k\) is an eigenform. On the other hand by (2.2)
    \begin{equation*} c_k\inv m^{k-1} \pair{f_k}{h_m(\cdot, -\bar z ')} = c_k\inv m^{k-1} \sum_{j=1}^{\dim S_k} \pair{f_k}{\frac{\alpha_j(-\bar z ') f_j}{\pair{f_j}{f_j}_{\Pet}}}_{\Pet} \end{equation*}
    \begin{equation*} = c_k\inv m^{k-1} \sum_{j=1}^{\dim S_k} \frac{\overline{\alpha_j(-\bar z ') f_j}}{\pair{f_j}{f_j}_\Pet} \pair{f_k}{f_j}_\Pet \end{equation*}
    \begin{equation*} = c_k\inv m^{k-1} \sum_{j=1}^{\dim S_k} \overline{\alpha_j(-\bar z ')} \frac{\pair{f_k}{f_k}_{\Pet}}{\pair{f_k}{f_k}_{\Pet}} \end{equation*}
    combined with (2.3) this implies
    \begin{equation*} a_k(m) \overline{f_k(z')} = c_k\inv m^{k-1}\alpha_k(-\bar z') \end{equation*}
    Note:
    \begin{equation*} \overline{f_k(z')}=\overline{\sum_{n=1}^\infty a_k(n) e(nz)} \end{equation*}
    \begin{equation*} = \sum_{n=1}^\infty a_k(n) \overline{e(nz)} \end{equation*}
    \begin{equation*} = \sum_{n=1}^\infty a_k(n) e(-n \overline{z}') \end{equation*}
    implies
    \begin{equation*} a_k(m) f_k(z') = c_k\inv m^{k-1} \alpha_k (z') \end{equation*}
    so
    \begin{equation*} c_k\inv m^{k-1} h_m(z,z') = \sum_{j=1}^{\dim S_k} \frac{a_j(m)}{\pair{f_j}{f_j}_\Pet} f_j(z)f_j(z')\text{.} \end{equation*}
\begin{equation*} \int_{\Gamma\backslash \HH} \sum_{j=1}^{\dim S_k} \frac{a_j(m)}{\pair{f_j}{f_j}_\Pet} f_j(z)\overline{f_j(z)} y^k \diff \mu(z) \end{equation*}
\begin{equation*} = \sum_{j=1}^{\dim(S_k)} a_j(m) \frac{\pair{f_j}{f_j}_\Pet}{\pair{f_j}{f_j}_\Pet} = \trace(T(m))\text{.} \end{equation*}

Subtle point (in general): This integral converges, and gives a manageable expression.

Subsection 2.2 Zagier's approach

\begin{equation*} I(s) = \int_{\Gamma\backslash \HH} h_m(z,-\bar z') E(z; s) y^k \diff \mu(z) \end{equation*}

for \(\Re(s) \gt 1\text{.}\) (This “goes around the convergence issues”). Implies

\begin{equation*} \Res_{s=1 } I(s) = \trace( T_m) \frac{c_k}{m^{k-1}} E(z; s) = \frac12 \zeta(2s) \pi^s \Gamma(s) \sum_{\gamma \in \Gamma_\infty \backslash \Gamma} \Im(\gamma z)^s\text{.} \end{equation*}
Calculation of \(I(s)\)

First step:

Recall that there exists a one to one correspondence between (for fixed \(m,t\))

\begin{equation*} \{A \in \Mat_{2\times 2 }(\ZZ) : \trace(A) = t,\,\det(A) =m\} \end{equation*}
\begin{equation*} \updownarrow \end{equation*}
\begin{equation*} \{\phi(u,v) = au^2 + buv + cv^2 : \Delta_\phi = | \phi| = b^2- 4ac= t^2 - 4m\} \end{equation*}
\begin{equation*} \begin{pmatrix} a\amp b \\ c\amp d \end{pmatrix} \mapsto \phi(u,v) = cu^2 + (d-a) uv - b^2v^2 \end{equation*}
\begin{equation*} \phi(u,v) = au^2 + buv + cv^2 \mapsto \begin{pmatrix} \frac12 (t-b)\amp -c \\ a\amp \frac12 (t+b)\end{pmatrix}\text{.} \end{equation*}

Using this

\begin{equation*} y^k h_m(z,-\bar z) = \sum_{a,b,c,d\in \ZZ,\,ad-bc = m} \frac{y^k}{(c|z|^2 + d\bar z - az - b)^k} \end{equation*}
\begin{equation*} = \sum_{t\in \ZZ} \sum_{ad-bc = m,\, a+d = t} \frac{y^k}{(c|z|^2 + d\bar z - az - b)^k} \end{equation*}
\begin{equation*} = \sum_{t\in \ZZ} \sum_{\phi,\, |\phi | = t^2 - 4m} R_\phi(z,t) \end{equation*}

where

\begin{equation*} R_\phi(z,t) = \frac{y^k}{(a|z|^2 + bx + c - ity)^k}\text{.} \end{equation*}

Exercise: substitute \(a= \frac 12 (t-b), b= -c , c = a, d= \frac12 (t+b)\text{.}\)

Remark 2.10

It will turn out that the \(t\)-sum is finite.

Recall that \(\SL_2(\ZZ)\) acts on binary quadratic forms \(\Phi\) by

\begin{equation*} \gamma \phi(u,v) = \phi(au + cv, bu + dv)= \phi(\gamma^\transpose \begin{pmatrix}u\\v\end{pmatrix})\text{.} \end{equation*}
Remark 2.12

If \(\Delta = \disc(K/\QQ)\) then

\begin{equation*} \zeta(s, \Delta) = \zeta_K(s)\text{.} \end{equation*}

Then we have

\begin{equation*} \sum_t\Res_{s= 1} \zeta_K(s) \end{equation*}

where \(K = \QQ(\sqrt{t^2 - 4m})\text{.}\)

  1. A slick proof involves noting that

    \begin{equation*} n(a) = \#\{\phi \in \Phi ,\,(m,n) \in \ZZ^2 : \phi(m,n) = 0, |\phi| = \Delta,\, \gcd(m,n) = 1\} \end{equation*}

    comes from the theory of quadratic forms.

    Instead note that \(\Gamma \acts \Phi\) via \(\gamma \cdot \phi =(u,v) = \phi(\gamma^\transpose \begin{pmatrix} u\\v \end{pmatrix})\text{.}\) We also have \(\Gamma\acts \ZZ^2\) via \(\gamma \cdot (m,n) = \gamma \begin{pmatrix} m\\n \end{pmatrix}\text{.}\)

    Note that

    \begin{equation*} (\gamma ^\transpose)\inv = \begin{pmatrix} 0 \amp 1 \\ -1 \amp 0 \end{pmatrix} \gamma \begin{pmatrix} 0 \amp -1 \\ 1 \amp 0\end{pmatrix} \end{equation*}

    (check!).

    Define

    \begin{equation*} \pair{-}{-}\colon \Phi \times X \end{equation*}
    \begin{equation*} \pair{\phi}{(m,n)} = \phi(n,-m) \end{equation*}
    \begin{equation*} = \phi \left( \begin{pmatrix} 0 \amp 1 \\ -1 \amp 0 \end{pmatrix} \begin{pmatrix} m \\n \end{pmatrix} \right)\text{.} \end{equation*}

    Observe \(\pair{-}{-}\) is \(\Gamma\)-invariant! i.e.

    \begin{equation*} \pair{\phi}{(m,n)} = \pair{\gamma \phi }{ \gamma(m,n)} \end{equation*}

    for all \(\gamma \in \Gamma\text{.}\) Then

    \begin{equation*} \zeta(s, \Delta) = \sum_{\phi \pmod \Gamma,\, |\phi| = \Delta} \sum_{(m,n) \in X/\Aut(\phi),\,\phi(m,n) \gt 0} \frac{1}{\phi(m,n)^s},\,\Re(s) \gt 1 \end{equation*}
    \begin{equation*} = \sum_{\phi \in \Phi_\Delta/\Gamma} \sum_{x \in X/\Aut(\phi)} \frac{1}{\pair{\phi}{x}^s} \end{equation*}
    \begin{equation*} = \sum_{(\phi,x) \in (\Phi_\Delta \times X)/\Gamma} \frac{1}{\pair{\phi}{x}^s} \end{equation*}
    \begin{equation} = \sum_{x\in X/\Gamma} \sum_{\phi\in \Phi_\Delta/\Aut(x)} \frac{1}{\pair{\phi}{x}^s}\label{eqn-zeta-delta-calc}\tag{2.4} \end{equation}

    we are using Fubini to swap the order of summation as the pairing is invariant. Finally note that

    \begin{equation*} x= \pm(m,n)\sim_{\Gamma} \pm(0,\gcd(m,n)),\, r = \gcd(m,n) \gt 0 \end{equation*}

    so (2.4)

    \begin{equation*} = \sum_{r= 1}^\infty \sum_{\phi \in \Phi_\Delta/\Aut((r,0))} \frac{1}{\phi(r,0)^s} \end{equation*}

    what is \(\Aut((r,0))\text{?}\) It is

    \begin{equation*} \Gamma_\infty^\transpose = \left\{\begin{pmatrix} 1 \amp 0 \\ \ast \amp 1 \end{pmatrix}\right\} \text{ as }\begin{pmatrix} 1 \amp 0 \\ \ast \amp 1 \end{pmatrix}\begin{pmatrix} 0\\ r \end{pmatrix} = \begin{pmatrix} 0\\ r \end{pmatrix} \end{equation*}

    continuing we have

    \begin{equation*} \zeta(2s) \sum_{a= 1}^\infty \frac{1}{a^s} \left\{ \sum_{b \pmod{2a},\, b^2 \equiv \Delta \pmod{4a}} 1\right\} \end{equation*}

    Exercise, show this is the sum from before. Hint, consider, explicitly, the action of \(\gamma\in \Gamma_\infty^\transpose\) on \(\phi(u,v)\text{.}\)

Back to the proof of the trace formula.

  1. Note \(R_{\gamma\phi}(t,z) = R_\phi (t, \gamma^\transpose z)\) which means that the integral is well defined (check this).
  2. \begin{equation*} \int_{\Gamma\backslash \HH} \left( \sum_{|\phi| = \Delta} R_\phi (t,z)\right) E(z; s) \diff \mu (z) \end{equation*}
    \begin{equation*} = \int_{\Gamma\backslash \HH} \sum_{|\phi| = \Delta} \sum_{(m,n) \in \ZZ^2 \smallsetminus \{(0,0)\}} R_\phi(t, z) \frac{y^s}{|mz+n|^{2s}} \diff \mu(z) \end{equation*}
    unfold whenever you can!
  3. Recall \(\Gamma \acts \Phi_\Delta \times X\) fixing \(\pair{\phi}{(m,n)} = \phi(m,n)\text{.}\) We can break the sum into \(\phi(m,n) \gt 0,\, \phi(m,n) \lt 0\text{.}\)
  4. Fact: The action of \(\Gamma\) on the above is free, (exercise: prove this fact, hint: first take \((m,n) \to (0,r)\text{,}\) and analyse the action of the stabiliser on \((0,r)\)).
  5. \begin{equation*} \int_{\Gamma \backslash \HH} \sum_{|\phi| = \Delta} \sum_{(m,n) \in X,\, \phi(m,n) \gt 0} R_\phi(t,z) \frac{y^s}{|mz+n|^s} \diff \mu(z) \end{equation*}
    \begin{equation*} \sum_{(\Phi_\Delta \times X)/\Gamma} \int_\HH R_\phi (t,z) \frac{y^s}{|mz+n|^{2s}} \diff \mu(z)\text{.} \end{equation*}
  6. Check
    \begin{equation*} z\mapsto \frac{nx - \frac 12 bn +cm}{-mz + an - \frac 12 bm} \end{equation*}
    finishes the proof.
  7. To check other cases use
    \begin{equation*} R_{-\phi} (z,t) = R_\phi(z,-t) \end{equation*}
    and
    \begin{equation*} \pair{\phi}{(m,n)} = 0 \end{equation*}
    needs to be considered separately.

We really need to check some convergence here, but lets instead use the trace formula for something useful, to determine its value.

Application to spaces of modular forms

Recall that

\begin{equation*} \trace(T_{S_k}(m)) \end{equation*}

is given by the Eichler-Selberg trace formula. In particular

\begin{equation*} \trace(T_{S_k}(m)) = \dim(S_k) \end{equation*}
\begin{equation*} = -\frac 12 \sum_{t^2 - 4 \le 0} P_k(t,1) H(4-t^2) - \frac 12 \end{equation*}

we have

\begin{equation*} H(0) = \frac{-1}{12} \end{equation*}
\begin{equation*} H(3) = \frac{1}{3} \end{equation*}
\begin{equation*} H(4) = \frac{1}{2} \end{equation*}

to find the \(P_k(t,1)\)

\begin{equation*} \rho_t, \bar \rho_t= \frac{t \pm \sqrt{t^2 - 4}}{2} = e^{i\theta}, e^{-i\theta} \end{equation*}

so \(\tan \theta = \frac{\sqrt{4-t^2}}{t}\) note then

\begin{equation*} \frac{\rho_t^{k-1} - \bar \rho_t ^{k-1}}{\rho_t - \bar \rho_t} = \frac{\rho_{-t}^{k-1} - \bar \rho_{-t} ^{k-1}}{\rho_{-t} - \bar \rho_{-t}} = \frac{\sin((k-1)\theta)}{\sin(\theta)} \end{equation*}

so \(P_k(t,1)\) is as in the table

\(t\) \(\theta\) \(P_k(t,1)\)
\(0\) \(\pi/2 \) \(k\mod 4 - 1\)
\(1\) \(\pi/3 \) \((k\mod 6 - 4)/2)\)
\(2\) \(0 \) \((k-1) \)
Table 2.16 Values of \(P_k(t,1)\text{.}\)

Now if \(k \equiv 0\pmod{12}\)

\begin{equation*} -\left( \frac{-1}{4} - \frac {1}{3} - \frac{k-1}{12}\right) - \frac12 = \frac{k}{12} = \left\lfloor \frac{k}{12}\right \rfloor= \cdots\text{.} \end{equation*}
Another application: Equidistribution of Hecke eigenvalues.

Following Serre '97, very readable article.

Equidistribution:

Definition 2.17

Let \((\Omega, \mu)\) be a measure space

\begin{equation*} \Omega\text{ compact} \end{equation*}
\begin{equation*} \mu\text{ positive Radon measure} \end{equation*}
\begin{equation*} \int_\Omega \diff \mu = 1 \end{equation*}
Example 2.18
\begin{equation*} [-1,1],\,\frac{\diff x}{2} = \diff \mu \end{equation*}
Note 2.19

\(\mu\) gives a linear functional.

\begin{equation*} f\mapsto \int_\Omega f(x) \diff \mu(x)\text{,} \end{equation*}

notated

\begin{equation*} \pair{f}{\mu}\text{.} \end{equation*}
Definition 2.20 \(\delta\) measures

Let \(L\) be a sequence of indices, going to \(\infty\) (this will be the indexing set for the weight \(k\)). For \(\lambda \in L\) let \(I_\lambda\) be a non-empty finite set of cardinality

\begin{equation*} \#I_\lambda = d_\lambda \end{equation*}

(this will be the set of eigenvalues for each \(k\)). Let

\begin{equation*} \mathbf X _\lambda = (x_{i,\lambda}),\,i\in I_\lambda\text{.} \end{equation*}

We define

\begin{equation*} \delta _{\mathbf X_\lambda} = \frac{1}{d_\lambda} \sum_{i\in I_\lambda} \delta_{x_{i,\lambda}} \end{equation*}

note

\begin{equation*} \pair{f}{\delta _{\mathbf X_\lambda}} = \frac{1}{d_\lambda} \sum_{i\in I_\lambda} f(x_{i,\lambda})\text{.} \end{equation*}

The family \(\mathbf X_\lambda\) for \(\lambda \in L\) is called \(\mu\)-equidistributed (or equidistributed with respect to the measure \(\mu\)) if

\begin{equation*} \lim_{\lambda \to \infty} \delta_{\mathbf X_\lambda} = \mu \end{equation*}

the limit is in the weak-\(\ast\) sense i.e.

\begin{equation*} \lim_{\lambda \to \infty} \pair{f}{\delta_{\mathbf X_\lambda}} = \pair{f}{\mu},\,\forall f \in C(\Omega, \RR)\text{.} \end{equation*}

So far these are all general statements. Let us put ourselves in the following situation:

For each \(\lambda \in L\) we have a linear operator

\begin{equation*} H_\lambda,\,\rank (H_\lambda) = d_\lambda \lt \infty \end{equation*}

whose eigenvalues are in \(\Omega\text{.}\) And \(I_\lambda = \{ 1,2,\ldots, d_\lambda\}\text{,}\) \(\mathbf X_\lambda = \{x_{1,\lambda}, \ldots, x_{d_\lambda,\lambda}\}\) the eigenvalues of \(H_\lambda\) suitably normalised counted with multiplicity.

Exercise.

Now we will choose a special set of polynomials to work with.

Symmetric polynomials:

Let \(\Omega= \lb -2 , 2 \rb \leftrightarrow 2 \cos \lb -\pi, \pi\rb = \trace(\specialunitary(2,\CC))\text{,}\) this is really Satake parameters. For \(x\in \lb -2, 2\rb\) we have a corresponding \(2\cos (\phi)\text{.}\) Let

\begin{equation*} X_n (x) = e^{in\phi} + e^{i(n-2)\phi} + \cdots + e^{-in\phi} = \trace (\Sym^n(U)),\,x = \trace(u)\text{.} \end{equation*}

Clebesh-Gordon: (Decomposition of tensor powers of irreducible representations of \(\specialunitary(2, \CC)\)).

\begin{equation*} X_nX_m = \sum_{0\le r \le \min\{n,m\}} X_{n+m - 2r} = x_{n+m} + x_{n+m-2} +\cdots + x_{|n-m|}\text{.} \end{equation*}

Prove this.

Several measures:

  1. Sato-Tate:
    \begin{equation*} \mu_\infty = \frac 1 \pi \sqrt{1-\frac{x^2}{4}} \diff x \end{equation*}
    \begin{equation*} = \frac 2 \pi \sin^2(\phi) \diff \phi \end{equation*}
    properties
    \begin{equation*} \int_{\Omega} \diff \mu_\infty = 1 \end{equation*}
    \begin{equation*} \pair{X_n}{ \mu_\infty} = \begin{cases} 1\amp n = 0,\\0\amp\text{otw}\end{cases} \end{equation*}
    \begin{equation*} \pair{X_nX_m}{ \mu_\infty} = \delta_{n,m}\text{.} \end{equation*}
    Exercise: prove this, with orthogonality of characters.
  2. \(p\)-adic Plancherel measure: Let \(q \in \RR_{\gt 1}\) and define
    \begin{equation*} f_q(x) = \sum_{n=0}^\infty q^{-n} X_{2n}(x) = \frac{q+1}{(q^{1/2} + q^{-1/2})^2 - x^2} \end{equation*}
    \begin{equation*} \mu_q = f_q \mu_\infty \end{equation*}
    note:
    \begin{equation*} \lim_{q\to 1^+} \mu_q = \frac{\diff \phi}{\pi} \end{equation*}
    \begin{equation*} \lim_{q \to \infty} \mu_q = \mu_\infty \end{equation*}
    \(\mu_q\) is positive and has total mass 1. note:
    \begin{equation*} \pair{X_n}{\mu_q} = \pair{X_n}{\sum_{m=0}^\infty q^{-m} X_{2m} \mu_\infty} \end{equation*}
    \begin{equation*} = \sum_{m=0}^\infty q^{-m} \pair{X_nX_{2m} }{\mu_\infty} \end{equation*}
    \begin{equation*} = \begin{cases} q^{-n/2} \amp n\equiv 0 \pmod 2 \\ 0 \amp \text{otw}\end{cases}\text{.} \end{equation*}

Claim:

\begin{equation*} \lim_{k\to \infty,k\equiv 0 \pmod 2} \frac{\tr(T_k'(n))}{s(k)} \end{equation*}
\begin{equation*} = \lim_{k\to \infty,k\equiv 0 \pmod 2} \frac{\tr(T_k'(n))}{(k-1)/12} = \begin{cases} n^{-1/2} \amp \text{ if } n \text{ is a square}\\ 0 \amp\text{otw} \end{cases} \end{equation*}

Assume the claim for now: Note that

\begin{equation*} T_k'(p^m) = X_m(T_k'(p)) \end{equation*}

as

\begin{equation*} \sum_{m=0}^\infty T_k'(p^m)t^m = \frac{1}{1- T_k'(p)t + t^2} \end{equation*}
\begin{equation*} \sum_{m=0}^\infty X_m(T_k'(p))t^m \end{equation*}
\begin{equation*} = \lim_{k\to \infty,k\equiv 0 \pmod 2} \frac{\tr(T_k'(n))}{s(k)} = \begin{cases} p^{-m/2} \amp m \equiv 0 \pmod 2\\ 0 \amp\text{otw} \end{cases} \end{equation*}
\begin{equation*} = \pair{X_m}{\mu_p} \end{equation*}

by the claim and previous calculation.

Proof of the claim: Recall that the Eichler-Selberg trace formula says that

\begin{gather} \tr(T_k(n)) = \frac{k-1}{12} n^{k/2 -1} \delta_{n= \square}\label{eqn-serre-equi-1}\tag{2.6}\\ - \frac12 \sum_{t^2 \lt 4n} P_k(t,n) H(4n - t^2)\label{eqn-serre-equi-2}\tag{2.7}\\ - \frac 12 \sum_{dd' = n,\,d,d' \gt 0} \min(d,d')^{k-1}\label{eqn-serre-equi-3}\tag{2.8} \end{gather}

Sublemma: (2.7) \(= O(n^{k/2})\)

Proof: (2.7) \(= O_n(\sum_{t^2 \lt 4n} P_k(t,n) )\)

\begin{equation*} P_k(t,n) = \frac{\rho^{k-1} - \bar \rho^{k-1}}{\rho - \bar \rho},\, \rho = \frac{t + i \sqrt{4n - t^2}}{2} \end{equation*}
\begin{equation*} | \rho^{k-1} - \bar \rho^{k-1}| \le 2|\rho^{k-1}| = O((4n- t^2)^{(k-1)/2}) \end{equation*}
\begin{equation*} \rho - \bar \rho = \sqrt{n - t^2} \end{equation*}

Sublemma: (2.8) \(= O(n^{(k-1)/2})\) Proof: exercise.

The equations above combined with the fact \(n^{k-1/2} \to n^{-1/2} \delta_{n=\square}\text{.}\)