Proof of a modular relation between 1-, 2- and 3-loop Feynman diagrams on a torus (Q1627955)

Motivated by considerations in string perturbation theory the authors introduced modular graph functions in [``Matching the \(D^6 R^4\) interaction at two-loops'', J. High Energy Phys. 2015, Article ID 31, 37 p. (2015; \url{doi:10.1007/JHEP01(2015)031})]. These are non-holomorphic modular functions for the group \(\operatorname{SL}_2(\mathbb Z)\) defined as lattice sums associated to Feynman diagrams with \(L\) loops on a complex torus with four marked points. In the simplest case \(L=1\) modular graph functions are simply non-holomorphic Eisenstein series \(E_k(\tau)\) for an integer \(k\geq 2\). Modular graph functions associated to diagrams with several loops are more complicated and, apparently, do not appear in the classical modular forms literature. The case \(L=2\) was extensively studied in [loc. cit.], where it was found that the respective modular graph functions satisfy Laplace eigenvalue equations whose inhomogeneous parts are quadratic polynomials in non-holomorphic Eisenstein series. Knowing in addition the asymptotic behaviour at the cusp then determines the Fourier expansion of the modular graph function in question. In the present paper under review the authors study one of the simplest examples of modular graph functions for \(L=3\), namely \[ D_4(\tau)=\sum_{m_i,n_i} \prod_{i=1}^4\left(\frac{\operatorname{Im}(\tau)}{\pi|m_i\tau+n_i|^2}\right), \] where the sum is over all tuples of integers \((m_1,\ldots,m_4,n_1,\ldots,n_4)\) such that \(m_1+\ldots+m_4=n_1+\ldots+n_4=0\), as well as \((m_i,n_i) \neq (0,0)\). Their main result is that \[ D_4(\tau)=24C_{2,1,1}(\tau)+3E_2(\tau)^2-18E_4(\tau), \] where \(C_{2,1,1}(\tau)\) is a certain modular graph function associated to a diagram with two loops. In particular, one sees that \(D_4(\tau)\), whose diagram has three loops, can in fact be expressed as a polynomial in modular graph functions for \(\leq 2\) loops. The idea of proof is to express the difference of both sides of (1) as a Laurent polynomial in \(\pi\operatorname{Im}(\tau)\) whose coefficients are real parts of certain holomorphic functions, \(\varphi_k(q)\), for \(k=0,\ldots,7\), on the open unit disc \(|q|<1\), where \(q=e^{2\pi i\tau}\). A direct computation shows that these vanish for \(k\neq 4,5,6\). On the other hand, proving vanishing of the remaining coefficients is much more involved. First, the authors show that the \(\varphi_k\) have real Fourier coefficients and are invariant, up to a sign, under inversion \(q\mapsto q^{-1}\) (establishing these properties is the most complicated part of the argument). Using this, the authors then go on to show that for \(k=4\) (respectively for \(k=5\), \(6\)), the function \(\varphi_k(q)\) is in fact a modular form for \(\operatorname{SL}_2(\mathbb Z)\) of weight \(0\) (respectively \(-2\), \(-4\)). This immediately implies the vanishing for \(k=5,6\) and the remaining case \(k=4\) follows by showing that its constant term vanishes. An alternative proof of the vanishing of \(\varphi_k\) for \(k=4,5,6\), due to Deligne, is sketched in an appendix.