Rank one elliptic \(A\)-modules and \(A\)-harmonic series (Q1325141)

This paper is a tour de force on the part of the author. It represents a high point of a trend started by Drinfeld: Apply ideas from the theory of partial differential equations to the arithmetic of function fields over finite fields; in particular, basic to the author's work is the concept of a ``soliton'' -- an idea from the theory of water waves and the famous nonlinear Korteweg-de Vries equation (KdV): \[ {{\partial\varphi} \over {\partial t}} = {{\partial^ 3\varphi} \over {\partial x^ 3}} +6\varphi{{\partial\varphi} \over {\partial x}}. \] (A ``soliton'' is -- roughly -- a periodic or quasi-periodic solution to KdV.) Let \(X/\mathbb{F}_ q\) be a smooth, projective, geometrically irreducible curve and let \(\infty\in X\) be a rational point; one then sets \(A\) to be the affine ring of \(X-\infty\) and \(k\) to be the quotient field of \(A\). By fiat the field \(k\) contains exactly one place not in \(\text{Spec} (A)\) thus mimicking the classical case of \(\mathbb{Z}\subset \mathbb{Q}\). The field \(k\) forms a ``bottom'' for the theory, exactly as \(\mathbb{Q}\) is a bottom in algebraic number theory. In this situation we can define characteristic \(p\) analogs of classical \(L\)-functions and \(\Gamma\)-functions [see the reviewer in `The arithmetic of function fields', de Gruyter, 313-402 (1992)]. In order to make these definitions ``good'' analytic functions (in a straightforward sense) one always ``groups according to degree''. For instance, an \(L\)-series \(\sum c_ a a^{-s}\) is analytically continued by grouping the terms of the same degree; \[ \sum c_ a a^{- s}= \sum_ j \Bigl( \sum_{\deg a=j} c_ a a^{-s} \Bigr), \] etc. Thus each such function \(f\) is made up of a sequence \(f_ k\) of ``degree parts''. The author seeks to study such \(f\) via an ``interpolation formula'': One seeks an algebraic function \(\varphi\) of two variables \(\{x,t\}\) such that \(f_ k\) (or, perhaps, \(f_{k+1}/ f_ k\) or \(f_{k+2} f_ k/ f^ 2_{k+1}\)) is recovered by specializing \(x\) and \(t\) (for instance, if \(A= \mathbb{F}_ q [T]\) then \(x=T\) and \(t= T^{q^ k}\)). The point being that such a \(\varphi\) is found by using soliton methods in characteristic \(p\) obtained by essentially replacing differentiation with the \(q\)-th power mapping, an idea that may seem far-fetched from afar but becomes more and more natural the closer it is examined. For instance, set \(\delta y:= y^ q- y\). One then finds trivially that \(\delta\) satisfies the derivation-identity: \[ \delta (xy)= x\delta y+ y^ q\delta x. \] In the case \(A= \mathbb{F}_ q[T]\) an ad-hoc version of this procedure was used to find \(\varphi\) [see \textit{G. W. Anderson} in `The arithmetic of function fields', 51-73 (1992; Zbl 0797.11056)]. The main tool in the author's derivation of \(\varphi\) is his Theorem 4.1.1: one maps \(Y\times Y\) to \(J\) and looks at the locus \(Z\) of the pullback of the theta-divisor. One expresses \(Z\) as a sum over certain correspondences coming from graphs of (composites) of the Frobenius and elements of \(\Aut(Y/X)\) -- so one is interested in the multiplicities that appear. The author establishes the deep fact that these multiplicities may be read off of an \(L\)-function evaluator (i.e., a formal power series with coefficients in \(\mathbb{Z}[ \Aut (Y/X)]\) used to define classical -- characteristic 0 -- \(L\)-series). The main result of the author is then established: A version for general sign-normalized rank one Drinfeld \(A\)-modules of the basic classical identity: \(\exp(- \sum_{n=1}^ \infty {{z^ n} \over n})= 1-z\). This result was inspired by some remarkable class number one calculations of \textit{D. Thakur} [Int. Math. Res. Not. 1992, No. 9, 185-197 (1992; Zbl 0756.11015)]. Motivated by the present paper, in a recent preprint the author is able to construct regulators for zeta-functions (in the above sense).