Moments of characteristic polynomials and their derivatives for $SO(2N)$ and $USp(2N)$ and their application to one-level density in families of elliptic curve $L$-functions

Abstract

Using the ratios theorems, we calculate the leading order terms in $N$ for the following averages of the characteristic polynomial and its derivative: $\left< \left|\Lambda_A(1 )\right| ^{r} \frac{ \Lambda_A'(\mathrm{e}^{\mathrm{i} \phi}) }{ \Lambda_A(\mathrm{e}^{\mathrm{i} \phi})} \right>_{SO(2N)}$ and $\left< \left|\Lambda_A(1 )\right| ^{r} \frac{ \Lambda_A'(\mathrm{e}^{\mathrm{i} \phi}) }{ \Lambda_A(\mathrm{e}^{\mathrm{i} \phi})} \right>_{USp(2N)}$. Our expression, derived for integer $r$, permits analytic continuation in $r$ and we conjecture that this agrees with the above averages for non-integer exponents. We use this result to obtain an expression for the one level density of the `excised ensemble', a subensemble of $SO(2N)$, to next-to-leading order in $N$. We then present the analogous calculation for the one level density of quadratic twists of elliptic curve $L$-functions, taking into account a number theoretical bound on the central values of the $L$-functions. The method we use to calculate the above random matrix averages uses the contour integral form of the ratios theorems, which are a key tool in the growing literature on averages of characteristic polynomials and their derivatives, and as we evaluate the next-to-leading term for large matrix size $N$, this leads to some multi-dimensional contour integrals that are slightly asymmetric in the integration variables, which might be useful in other work.