Revisiting Rademacher's formula for the partition function \(p(n)\) (Q5931574)

Let \(p(n)\) be the number of unrestricted partitions of \(n\in \mathbb Z^+\), with order irrelevant. Euler observed that \(f(x)= \prod_{m=1}^\infty (1-x^m)^{-1}= \sum_{n=0}^\infty p(n)x^n\), \(|x|<1\), where \(p(0)=1\). It is familiar that \(\eta(\tau)= e^{\pi i\tau/12} f(e^{2\pi i\tau})\), \(\tau\in H=\) upper half-plane, is a modular cusp form on \(\Gamma(1)\), the full modular group. Almost as well known is the fact that an exact formula for \(p(n)\) can be derived from a suitable form of the circle method if one uses the fact that \(\eta(\tau)^{-1}\) is a modular form of negative weight (weight \(-1/2)\) with no singularities in \(H\) and known growth at \(i\infty\). This stands in sharp contrast to the situation 65 years ago when \textit{H. Rademacher} [Proc. Lond. Math. Soc. (2) 43, 241--254 (1937; Zbl 0017.05503)] surprised everyone (including himself) with his success in modifying the circle method to obtain a (rapidly convergent) series representation for \(p(n)\). Another approach to obtaining this representation (and, indeed, exact formulas for the Fourier coefficients of modular forms of negative weight on \(\Gamma(1)\) -- and similar groups) is that of Petersson, who employs the duality between modular forms of weight \(k\) with multiplier systems (MS) \(v\), and modular forms of weight \(k\) with MS \(\overline{v}\) (in the form of his generalized Riemann-Roch theorem). Petersson's approach yields an exact formula for \(p(n)\) since there are no nontrivial cusp forms on \(\Gamma(1)\) of weight \(5/2\) and MS \(v_\eta\). (Note that \(\eta(\tau)^{-1}\) has MS \(\overline{v}_\eta\).) The article under review provides a very interesting third approach, closer in spirit to Petersson's than it is to Rademacher's. (Interestingly, both the author and Douglas Niebur, whose 1968 dissertation forms the background for the author's treatment of \(p(n)\), are mathematical great-grandchildren of Rademacher.) Pribitkin begins by studying \(H(\tau\mid s)\), a nonanalytic (in \(\tau\)) version of a ``pseudo-Poincaré series'' first encountered by Niebur. In the end he proves a modular transformation law for \(H(\tau\mid 0)\) and this shows that \(H(\tau\mid 0)\) is a modular form on \(\Gamma(1)\) with exactly the same transformation properties as those of \(\eta(\tau)^{-1}\). As in Petersson's work, the fact that there are not nontrivial cusp forms of weight \(5/2\) and MS \(\overline{v}_\eta\) is required to complete the derivation of the exact formula for \(p(n)\). Essential to Pribitkin's proof is a treatment of the integral \[ {\mathcal I}(A,B,D)= \int_{-\infty}^\infty \frac{e^{-i Du}} {(u+i)^A (u-i)^B} \,du,\tag \(*\) \] where \(D\) is real and \(\text{Re}(A+B)> 1\). The integral \((*)\) has long been studied because of its importance in the theory of Eisenstein and Poincaré series (particularly those of small weight). The author does a public service by giving us an elegant new approach. It deserves to be mentioned as well that he provides interesting and insightful historical comments.