Power counting in the spectral action matrix model

We derive power counting formulas for ribbon graph amplitudes that were recently independently discovered in two contexts, namely as a generalization of the Kontsevich model, and as corresponding to a matrix model approach to the spectral action. The Feynman rules are based on divided difference functions of eigenvalues of an abstract Dirac operator. We obtain formulas for the order of divergence, depending on the spectral dimension $d$, the order of decay of the test function $f$ of the spectral action, and the graph properties. Several consequences are discussed, such as the fact that all graphs with maximal order of divergence (at a given loop order and number of external vertices) are planar. To derive our main results we establish two-sided bounds for divided differences, and in particular generalize Hunter's positivity theorem to a larger class of functions.