Bounding the maximum likelihood degree (Q530615): Difference between revisions

Let \(X\) be a closed irreducible subvariety of dimension \(d\) in the complex torus \((\mathbb C^*)^n\) with coordinates \((p_1, \ldots, p_n)\). By definition, the maximum likelihood degree of \(X\), denoted by \(\text{MLdeg}(X)\), is equal to the number of non-singular points of \(X\), where the differential 1-form \(\lambda_1d\log p_1+\ldots +\lambda_nd\log p_n\) degenerates for a generic point \(\lambda\in \mathbb C^n\) (see [\textit{F. Catanese} et al., Am. J. Math. 128, No. 3, 671--697 (2006; Zbl 1123.13019)]). In this note the authors prove that \((-1)^d\chi(\text{IC}(\mathbb C_X)) \geqslant \text{MLdeg}(X)\), where \(\chi(\text{IC}(\mathbb C_X))\) is the Euler characteristic of the intersection cohomology complex of \(X\). Thus, they obtain a bound for the maximum likelihood degree which is similar to hypothetical one from [\textit{J. Huh} and \textit{B. Sturmfels}, Lect. Notes Math. 2108, 63--117 (2014; Zbl 1328.14004)], where the corresponding conjecture is formulated in terms of ordinary Euler characteristic of \(X\). Finally, the authors construct a few counterexamples to the latter conjecture.