While the low-energy part of the hadronic light-by-light (HLbL) tensor can be constrained from data using dispersion relations, for a full evaluation of its contribution to the anomalous magnetic moment of the muon (g − 2) µ also mixed-and high-energy regions need to be estimated. Both can be addressed within the operator product expansion (OPE), either for configurations where all photon virtualities become large or one of them remains finite. Imposing such short-distance constraints (SDCs) on the HLbL tensor is thus a major aspect of a model-independent approach towards HLbL scattering. Here, we focus on longitudinal SDCs, which concern the amplitudes containing the pseudoscalar-pole contributions from π 0 , η, η . Since these conditions cannot be fulfilled by a finite number of pseudoscalar poles, we consider a tower of excited pseudoscalars, constraining their masses and transition form factors from Regge theory, the OPE, and phenomenology. Implementing a matching of the resulting expressions for the HLbL tensor onto the perturbative QCD quark loop, we are able to further constrain our calculation and significantly reduce its model dependence. We find that especially for the π 0 the corresponding increase of the HLbL contribution is much smaller than previous prescriptions in the literature would imply. Overall, we estimate that longitudinal SDCs increase the HLbL contribution by ∆a LSDC µ = 13(6) × 10 −11 . A Anomalous Pseudoscalar-Vector-Vector Coupling 46 B Alternative model for pion, η, and η transition form factors 48 1 Introduction Current Standard Model (SM) evaluations of the anomalous magnetic moment of the muon, a µ = (g − 2) µ /2, differ from the value measured at the Brookhaven National Laboratory [1]a exp µ = 116 592 089(63) × 10 −11 , (1.1) by around 3.5 σ. In the near future, the new Fermilab E989 experiment [2] will be able to reduce the experimental uncertainty by a factor 4, and the E34 experiment at J-PARC [3] will provide an important cross check, see ref.[4] for a comparison of the experimental methods. Therefore, the theoretical calculation of a µ needs to be improved accordingly. The uncertainty of the SM prediction mainly stems from hadronic contributions, such as hadronic vacuum polarization (HVP), see figure 1 (a), and HLbL scattering, see figure 1 (b). Since the HVP contribution can be systematically calculated with a data-driven dispersive approach [5-9], lattice QCD [10][11][12][13][14][15][16], and potentially be accessed independently by the proposed MUonE experiment [17,18], which aims to measure the space-like finestructure constant α(t) in elastic electron-muon scattering, the HLbL contribution may end up dominating the theoretical error. 1 Apart from lattice QCD [27][28][29], recent data-driven approaches towards HLbL scattering are again rooted in dispersion theory, either for the HLbL tensor [30][31][32][33][34][35], the Pauli 1 Note that higher-order insertions of HVP [5,19,20] and HLbL [21] are already under sufficient control, as are hadronic corrections in the anomalou...