Bernstein-Sato ideals for free hyperplane arrangements

Let $f=(f_1,\dots ,f_r)$ be a complete factorization of a central hyperplane arrangement $D$ in $X={\mathbb{C}}^n$. For a monoid ideal $K\subseteq {\mathbb{N}}^r$ we study the Bernstein-Sato ideal $B_f^K$ of $f$ along $K$, that is, the $\mathbb{C}[s]$-annihilator of ${\mathcal{D}}_X[s]f^s/{\sum }_{m\in K}{\mathcal{D}}_X[s]f^{s+m}$. When $D$ is free we compute two families of these ideals with the help of AI. For the unit shift $K=⟨e_i⟩$ we prove that $B_f^{-e_i}$ is generated by an explicit product of linear forms indexed by the dense edges of $D$ contained in $D_i$. This determines all the Bernstein-Sato ideals $B_f^{a,b}={\mathrm{Ann}}_{\mathbb{C}[s]}{\mathcal{D}}_X[s]f^{s-a}/{\mathcal{D}}_X[s]f^{s-b}$, $a\ge b$, of a free arrangement, generalizing formulas of Maisonobe (2016) and Bath (2020). The main new ingredient identifies the multiplicities of the relative characteristic cycle of ${\mathcal{D}}_X[s]f^s/{\mathcal{D}}_X[s]f^{s+e_i}$ along the conormal bundle of the origin with the coefficients of the Hilbert series of an Artinian complete intersection attached to a generic Ziegler restriction of $D$; the total multiplicity computed in Wu (2022) then forces all the resulting coefficientwise upper bounds to be equalities. For the coordinate monoid ideal $K=⟨e_1,\dots ,e_r⟩$ we show that $B_f^K$ is generated by one Euler relation for each irreducible factor of the essential quotient of $D$. Finally, we show that the zero locus of a Bernstein-Sato ideal along a monoid ideal need not be a finite union of translated linear subvarieties, even for a reduced free arrangement in ${\mathbb{C}}^2$: for $f=(x,y,x+y,x+2y)$ and $K=⟨3e_1,3e_2⟩$ we compute $B_f^K$ exactly and find an irreducible quadric component. This disproves a conjecture due to Budur.