Dirichlet series (Q1131281)

The work on this paper was done shortly before the author's untimely death in 1957. The editor states that ``while the paper is largely expository and while the principal object of the author was not achieved, the material discussed is significant and seems worth presenting to the mathematical public in order to stimulate further research.'' Suppose \(k > 1\) and \(\chi\) is a Dirichlet character function (mod \(k\)) with \(\chi(-1) = 1\). Then the \(L\)-function \(L(s,\chi) = \sum_{n=1}^\infty \chi(n)n^{-s}\) is the transform of \(\psi(\tau,\chi) = \sum_{-\infty}^\infty \chi(n) e^{i\pi n^2\tau/k}\) in the sense that for \(\text{Re}(s) > 1\) \[ 2\Gamma\left(\frac{s}{2}\right)\left(\frac{k}{\pi}\right)^{s/2} L(s,\chi)= \int_0^\infty u^{s/2-1} \psi(iu,\chi)\,du. \] \textit{H. Hamburger} [Math. Z. 10, 240--254 (1921); 11, 224--245 (1921; JFM 48.1210.03); 13, 283--311 (1922; JFM 48.1213.02)], \textit{C. L. Siegel} [Math. Ann. 86, 276--279 (1922; JFM 48.1216.01)] and \textit{E. Hecke} [Math. Ann. 112, 664--699 (1936; Zbl 0014.01601); 119, 266--287 (1944; Zbl 0028.40503)] have done work in connection with function-theoretic definitions of the functions \(\psi\). In the present paper, dealing with the case \(k=p\) an odd prime, it is shown that if \((\alpha\tau+\beta)/(\gamma\tau+\delta)\) is in the group \(H(p)\) of modular-transformations with\(\alpha+\delta\equiv \beta+\gamma\equiv \pmod 2\) and \(\gamma\equiv\beta\equiv 0\pmod p\), then, the \((p-1)/2\) functions \(\psi(\tau,\chi)\) satisfy the functional equations \[ \psi((\alpha\tau+\beta)/(\gamma\tau+\delta),\chi) = S' \chi(\delta) (\gamma\tau+\delta)^{1/2}\psi(\tau,\chi) \] where \(S'\) is a specified function of \(\alpha\), \(\beta\), \(\gamma\), \(\delta\) and \(p\). It is found more convenient to put the problem in terms of the subgroup \(\Gamma(p)\) of \(H(p)\) in which \(\alpha\equiv \delta\equiv = \pm 1 \pmod p\). The question is asked: ``If \(\psi(\tau)\) is regular in \(\text{Im}(\tau) > 0\) where it satisfies \[ \psi((\alpha\tau+\beta)/(\gamma\tau+\delta)) = S'(\gamma\tau + \delta)^{1/2} \psi(\tau)\quad\text{in}\;\Gamma(p) \tag{9} \] and the order relation \[ |\psi(x+ iy)| \leq A(1+y^{-c}),\tag{10} \] must \(\psi(\tau)\) be a linear combination of the \((p+1)/2\) functions \(\varphi(\tau)\), \(\psi(\tau,\chi)\) ?'' It had already been observed that the \(\psi(\tau,\chi)\) satisfy (10) and that \(\varphi(\tau) = \sum_\infty^\infty e^{i\pi n^2\tau/p}\) also satisfies (9) and (10). Observing that the functions \(1\), \(\psi(\tau,\chi)/\varphi(\tau)\) are automorphic under the group \(\Gamma(p)\) and regular in its fundamental domain \(D_\Gamma (p)\) except for certain poles, it is shown that there is a divisor \(Q'\) on \(D_\Gamma (p)\) consisting of poles of total multiplicity \((p^2-1) (p-1)/16\) such that if \(\psi(\tau)\) satisfies (9) and (10) then \(\psi(\tau)/\varphi(\tau)\) is a multiple of \(Q'\). If the functions \(1\), \(\psi(\tau,\chi)/\varphi(\tau)\) can be shown to form a complete set of linearly independent multiples of \(Q'\) then the above question can be answered in the affirmative. Apparently the author had verified this in the cases \(p=3\) and \(5\) but he did not succeed in completing a general proof. It is indicated that the problem is connected with that of finding irreducible representations of the group \(H(1)/\Gamma(p)\) and with that of the number of representations of an integer as a sum of three or more squares.