Iterated and mixed discriminants (Q6166674)

In the paper under review, the authors define and studies properties of the multivariate iterated discriminants. Let \(K\) be an algebraically closed field of characteristic zero and we denote by \(A \subset \mathbb{Z}^{n}\) a finite lattice subset. A polynomial \(p = \sum_{a \in A}c_{a}x^{a} \in K[x_{1}, \dots, x_{n}]\) with the support on the point configuration \(A\) is called an \(A\)-polynomial. Denote by \(X_{A} \subset \mathbb{P}^{|A|-1}\) the projective variety defined as the closed image of the monomial embedding given by the \(A\)-monomials. The dual variety \(X_{A}^{\nu}\) is the closure of the coefficient vectors of the \(A\)-polynomials \(p\) whose zero locus has a singular point \(x \in (\mathbb{K}^{*})^{n}\) with non-zero coordinates. The expected codimension of \(X_{A}^{\nu}\) is one and when this is the case we say that \(A\) is non-defective. When \(A\) is non-defective, the irreducible polynomial \(D_{a} \in \mathbb{Z}[(c_{a})_{a\in a}]\) defining the dual variety \(X_{A}^{\nu} = V(D_{A})\) is called the \(A\)-discriminant. Now, for a given \(r+1\) finite configurations \(A_{0},\dots, A_{r} \subset \mathbb{Z}^{n}\) and a system of \(A_{i}\)-polynomials \(p_{0}, \dots, p_{r}\), namely \[ (\star): \quad p_{0} = \dots = p_{r}=0, \quad p_{i} = \sum_{a\in A_{i}}c_{i,a}x^{a} \] we call an isolated solution \(x \in (K^{*})^{n}\) a non-degenerate multiple root for the system \((\star)\) if the \(r+1\) gradient vectors \(\nabla_{x}p_{i}(x)\), for \(i \in \{0, \dots, r\}\) are linearly dependent, but any subset of \(r\) of them is linearly independent. The associated mixed discriminantal variety is the closure of the locus of coefficients for which the system has a non-degenerate multiple root. If this variety is a hypersurface, it is defined by a single irreducible polynomial which we call the mixed discriminant, denoted by \(\mathrm{MD}_{A_{0}, \dots, A_{r}}\). If it is not a hypersurface, then the system is called defective and one sets \(\mathrm{MD}_{A_{0}, \dots, A_{r}} = 1\). Given \(A_{0}, \dots, A_{r}\), the associated Cayley configuration \(C = C(A_{0}, \dots, A_{r}) \subset \mathbb{Z}^{n+r}\) is the union of the lifted configurations \(e_{i} \times A_{i} \in \mathbb{Z}^{n+r}\) for \(i \in \{0, \dots, r\}\), where \(e_{0}=0\) and \(e_{i}\)'s are the standard basis vectors in \(\mathbb{Z}^{r}\). Now one introduces \(r+1\) new variables \(\lambda_{0}, \dots, \lambda_{r}\) and encodes the initial system by one \(C\)-polynomial \[ P_{\lambda} = \lambda_{0}p_{0} + \dots + \lambda_{r}p_{r} \in K[\lambda_{0}, \dots, \lambda_{r},x_{1}, \dots, x_{n}]. \] We will denote the polynomial and its tuple of coefficients by \(P_{\lambda}\), where \(\lambda=(\lambda_{0}, \dots, \lambda_{r})\). The authors show that if \(C\) is non-defective, then \[ \mathrm{MD}(A_{0}, \dots, A_{r})(p_{0}, \dots, p_{r}) = D_{C}\bigg(\sum_{i=0}^{r}\lambda_{i}p_{i}\bigg), \] where \(D_{C}\) denotes the \(C\)-discriminant. This characterization leads to the following definition of multivariate iterated discriminant of order \(r\). In this paper the authors study the case when \(A_{0} = \dots = A_{r} = A\) and due to this reason they use the notation \(\mathrm{MD}_{r,A} =\mathrm{MD}_{A, \dots, A}\). We notice that the polynomial \(D_{A}(P_{\lambda})\) is a homogeneous polynomial of degree \(\deg(D_{A})\) in \(\lambda_{0}, \dots, \lambda_{r}\). Main Definition. Given \(A \subset \mathbb{Z}^{n}\) non-defective, denote by \(d\) the codimension of the singular locus of the dual variety \(X_{A}^{\nu}\). Given \(r \geq 0\), the multivariate iterated discriminant of order \(r\) is the polynomial \(\mathrm{ID}_{r,A}\) on the coefficients of \((r+1)\) \(A\)-polynomials \(p_{0}, \dots, p_{r}\) defined by \[ \mathrm{ID}_{r,A}(p_{0}, \dots, p_{r}) := D_{\delta_{A}\triangle_{r}}(D_{A}(P_{\lambda})) \text{ if } d\geq r, \] or \[ \mathrm{ID}_{r,A}(p_{0}, \dots, p_{r}) = 0 \text{ otherwise}. \] In order to formulate the main result of the paper, we need to do some final preparation. Let \(A\) be non-defective, we denote by \(\mathrm{Sing}(X_{A}^{\nu})\) the subscheme of \(X_{A}^{\nu}\) defined by the ideal generated by the partial derivative of \(D_{A}\). Denote by \(\mathrm{Ch}_{Y_{k}}\) the Chow forms of the higher dimensional irreducible components of the schematic singular locus of the dual variety \(X_{A}^{\nu}\). The main result of the paper under review can be formulated as follows. Theorem. Assume that \(A \subset \mathbb{Z}^{n}\) is non-defective and let \(r \in \mathbb{Z}\) with \( 0 \leq r \leq\dim(X_{A})\). Then the mixed discriminant \(\mathrm{MD}_{r,A}\) always divides the iterated discriminant \(\mathrm{ID}_{r,A}\). Moreover, \begin{itemize} \item[i)] If \(\mathrm{codim}_{X_{A}^{\nu}}(\mathrm{Sing}(X_{A}^{\nu}))>r\), then \(\mathrm{ID}_{r,A}=\mathrm{MD}_{r,A}.\) \item[ii)] If \(\mathrm{codim}_{X_{A}^{\nu}}(\mathrm{Sing}(X_{A}^{\nu})) = r\), then \(\mathrm{ID}_{r,A} = \mathrm{MD}_{r,A}\prod_{i=k}^{l} \mathrm{Ch}_{Y_{k}^{\mu_{k}}}\), where \(Y_{1}, \dots, Y_{l}\) are the irreducible components of \(\mathrm{Sing}(X_{A}^{\nu})\) of codimension \(r\) with respect to multiplicities \(\mu_{k}\geq 2\). \item[iii)] If \(\mathrm{codim}_{X_{A}^{\nu}}(\mathrm{Sing}(X_{A}^{\nu})) < r\), then \(\mathrm{ID}_{r,A} = 0\). \end{itemize}