Hook monomial immanants are Kazhdan-Lusztig positive

Abstract

Stembridge conjectured that every monomial immanant is non-negative on totally non-negative matrices, and Haiman proposed a stronger statement which at $q=1$ says that every monomial immanant is a non-negative integer combination of the Kazhdan-Lusztig immanants of Rhoades and Skandera. We prove this for every hook shape $(a,1^{n-a})$ and every $n$. More precisely, each of the per-vertex functions obtained by splitting Stembridge's formula for hook characters is Kazhdan-Lusztig positive. The main tool is a positivity statement for products: a product of Kazhdan-Lusztig positive immanants of complementary submatrices is Kazhdan-Lusztig positive, which we deduce at $q=1$ from Braden's hyperbolic localization, following Billey and Braden. With Haiman's theorem for irreducible characters, it shows that every class function whose Frobenius characteristic lies in the semiring generated by Schur functions and hook monomial symmetric functions is Kazhdan-Lusztig positive; for example, Skandera's chain of normalised hook immanants is Kazhdan-Lusztig positive for every $n$. We place these results in a comparison of three cones of class functions, and show that the remaining obstructions to Haiman's conjecture lie at singular Schubert varieties that are not reached by an iterated doubling identity of Abreu and Nigro, for which we give an explicit criterion. Finally, on the full span of permutation monomials, an element that is non-negative on totally non-negative matrices can have a negative coefficient at the Kazhdan-Lusztig immanant $\mathrm{Imm}_w$ if and only if the Schubert variety $X_w$ is singular; Skandera proved the smooth half.