---
title: Lowest Discriminant Subvariety
url: https://www.emergentmind.com/topics/lowest-discriminant-subvariety
type: topic
---

# Lowest Discriminant Subvariety

In contemporary algebraic and geometric literature, the lowest discriminant subvariety is the first nonempty discriminant stratum obtained when discriminant loci are ordered by increasing degeneracy. In the most literal formulation, for a Cayley–Hamilton Hopf algebra \((H,C,\mathrm{tr})\) that is module-finite over a central Hopf subalgebra, it is the zero locus \(\mathcal V_\ell=\mathcal V(D_\ell(H/C;\mathrm{tr}))\) of the lowest discriminant ideal, where \(\ell\) is the smallest index for which the discriminant locus is nonempty. Closely related constructions occur in the discriminant hypersurface of univariate polynomials, in subdiscriminant varieties of matrices, and in several algebro-geometric settings where the same organizing idea appears as a deepest singular stratum, a minimal-degree discriminant-type invariant, or a locus of minimal discriminant behavior [2307.15477, 2509.25820].

## 1. General discriminant-stratification framework

For a module-finite algebra with trace \((A,C,\mathrm{tr})\), the \(k\)-discriminant ideal and modified \(k\)-discriminant ideal are defined by
\[
D_k(A/C,\mathrm{tr})=\left\langle \det(\mathrm{tr}(a_i a_j))_{i,j=1}^k \right\rangle,\qquad
MD_k(A/C,\mathrm{tr})=\left\langle \det(\mathrm{tr}(a_i b_j))_{i,j=1}^k \right\rangle.
\]
In the Cayley–Hamilton setting, their zero loci coincide and admit the Brown–Yakimov description
\[
\mathcal V_k=\{\mathfrak m\in \operatorname{MaxSpec} C \mid \sum_{V\in \operatorname{Irr}(A/\mathfrak mA)}(\dim V)^2<k\}.
\]
There is then a unique integer \(\ell\) such that
\[
\varnothing=\mathcal V_1=\cdots=\mathcal V_{\ell-1}\subsetneq \mathcal V_\ell,
\]
and \(D_\ell\) is the lowest discriminant ideal while \(\mathcal V_\ell\) is the lowest discriminant subvariety. In this sense, the locus records the most degenerate fibers, in the opposite direction from the Azumaya locus [2307.15477].

An analogous nested hierarchy appears in the discriminant hypersurface of a monic univariate polynomial
\[
F=x^n+a_{n-1}x^{n-1}+\cdots+a_0.
\]
There one defines
\[
Z_k=\{(a_0,\ldots,a_{n-1})\in \mathbb C^n: F \text{ has at most } k \text{ distinct roots}\},
\]
giving the chain
\[
\varnothing=Z_0\subsetneq Z_1\subsetneq \cdots \subsetneq Z_{n-1}=V(D)\subsetneq Z_n=\mathbb C^n.
\]
The stratum \(S_i=Z_i\setminus Z_{i-1}\) consists of polynomials with exactly \(i\) distinct complex roots. Here the “lowest” object is the terminal nonempty stratum \(Z_1\), the locus of polynomials of the form \((x-r)^n\) up to translation [2509.25820].

## 2. Deepest stratum of the univariate polynomial discriminant

For the discriminant hypersurface of a monic univariate polynomial, the classical subdiscriminant stratification is
\[
Z_k=V(\{D_0,\ldots,D_{n-k-1}\}),
\]
where \(D_0=D\) is the discriminant and \(D_1,\ldots,D_{n-1}\) are the subdiscriminants. The paper "Stratifying Discriminant Hypersurface" replaces this description by two constructions that use only the discriminant itself [2509.25820].

The first construction is local-algebraic. If \(\gamma\) is a coefficient vector and \(\operatorname{ord}D(\gamma)\) is the order of vanishing of \(D\) at \(\gamma\), then
\[
Z_k=O_k:=\{\gamma\in\mathbb C^n:\operatorname{ord}D(\gamma)\ge n-k\},
\]
equivalently,
\[
Z_k=V\bigl(\{\partial_\delta D:\ |\delta|\le n-k-1\}\bigr).
\]
Thus a polynomial has at most \(k\) distinct roots if and only if the discriminant vanishes to order at least \(n-k\) at the corresponding coefficient point. The stratification is therefore simultaneously a stratification by number of distinct roots and a stratification by order of singularity.

The second construction is intrinsic and geometric:
\[
\operatorname{Sing} Z_k=Z_{k-1}\qquad (k=n-1,n-2,\ldots,2),
\]
hence
\[
Z_k=\underbrace{\operatorname{Sing}\cdots\operatorname{Sing}}_{n-1-k\text{ times}}V(D).
\]
Starting from the full discriminant hypersurface \(V(D)=Z_{n-1}\), successive singular loci recover \(Z_{n-2},Z_{n-3},\ldots,Z_1\). Each difference \(Z_k\setminus Z_{k-1}\) is smooth, and \(Z_1\) itself is smooth.

The deepest nonempty stratum is
\[
Z_1=\underbrace{\operatorname{Sing}\cdots\operatorname{Sing}}_{n-2\text{ times}}V(D),
\]
consisting of polynomials with one distinct root. The paper identifies \(Z_1\) explicitly as an affine line parametrized by
\[
t\mapsto\left((-1)^1\binom{n}{1}t,(-1)^2\binom{n}{2}t^2,\ldots,(-1)^n\binom{n}{n}t^n\right).
\]
If “lowest discriminant subvariety” is interpreted as the deepest nonempty singular stratum of the discriminant hypersurface, then \(Z_1\) is exactly that object [2509.25820].

## 3. Lowest discriminant ideals in Cayley–Hamilton Hopf algebras

The paper "The lowest discriminant ideal of a Cayley-Hamilton Hopf algebra" studies a Cayley–Hamilton Hopf algebra \((H,C,\operatorname{tr})\) with central Hopf subalgebra \(C\subseteq H\), \(H\) module-finite over \(C\), and basic identity fiber \(H/m_\varepsilon H\) [2307.15477]. In this setting the identity fiber determines a finite group
\[
G_0:=G\bigl((H/m_\varepsilon H)^\circ\bigr)\cong \operatorname{Irr}(H/m_\varepsilon H),
\]
and \(G_0\) acts on each fiber \(\operatorname{Irr}(H/mH)\) by tensoring:
\[
x\cdot V=x\otimes V.
\]

For \(V\in \operatorname{Irr}(H/mH)\), the stabilizer
\[
\operatorname{Stab}_{G_0}(V)=\{x\in G_0\mid x\otimes V\cong V\}
\]
satisfies
\[
|\operatorname{Stab}_{G_0}(V)|\le \dim(V)^2.
\]
A module is maximally stable when equality holds. The paper proves several equivalent characterizations:
\[
V\text{ maximally stable} \iff V\otimes V^* \text{ is a direct sum of pairwise nonisomorphic 1-dimensional modules},
\]
and
\[
V\otimes V^*=\bigoplus_{x\in \operatorname{Stab}_{G_0}(V)}x.
\]
In that case the primitive quotient is a twisted group algebra
\[
H/\operatorname{Ann}_H(V)\cong k^{\gamma_V}\operatorname{Stab}_{G_0}(V).
\]

The discriminant threshold at a point \(m\) is given by
\[
\min\{k\in \mathbb Z_{>0}\mid D_k(H/C,\operatorname{tr})(m)=0\}
=
\frac{|G_0|\,\dim(V)^2}{|\operatorname{Stab}_{G_0}(V)|}.
\]
From this, the level of the lowest discriminant ideal is
\[
\ell=|G_0|+1.
\]
Moreover, for \(m\in \operatorname{MaxSpec} C\), the following are equivalent: \(m\) lies in the zero set of the lowest discriminant ideal; there exists \(V\in \operatorname{Irr}(H/mH)\) that is maximally stable; every \(V\in \operatorname{Irr}(H/mH)\) is maximally stable. The lowest discriminant subvariety is therefore the locus where the fiber representation theory is maximally stable under tensoring with the identity-fiber character group [2307.15477].

## 4. Chevalley-property refinements and subgroup rigidity

The subsequent Chevalley-property theory replaces the basicity hypothesis by a representation-theoretic condition on the identity fiber algebra \(H/\mathfrak m_{\overline\varepsilon}H\). Under that assumption, any nonempty zero locus of a discriminant ideal contains the orbit of the identity under left or right winding automorphisms,
\[
W_l(G(H^\circ))(\mathfrak m_{\overline\varepsilon})
=
W_r(G(H^\circ))(\mathfrak m_{\overline\varepsilon}),
\]
and the level of the lowest discriminant ideal becomes
\[
\ell=\mathrm{FPdim}\big(\mathrm{Gr}(H/\mathfrak m_{\overline\varepsilon}H)\big)+1.
\]
The lowest locus can then be characterized by complete reducibility of tensor squares: \(\mathfrak m\) lies in the zero locus of the lowest discriminant ideal if and only if, for every irreducible \(H/\mathfrak mH\)-module \(W\), the module \(W\otimes W^*\) is completely reducible; equivalently,
\[
W\otimes W^* \cong \bigoplus_{i=1}^m V_i^{\oplus \dim_\Bbbk \mathrm{Hom}_H(V_i\otimes W,W)},
\]
where \(\{[V_1],\dots,[V_m]\}=\mathrm{Irr}(H/\mathfrak m_{\overline\varepsilon}H)\) [2506.21879].

The same paper proves that if \(H\) has the Chevalley property, then all discriminant ideals are trivial:
\[
D_k(H/C;\mathrm{tr})=MD_k(H/C;\mathrm{tr})=
\begin{cases}
C,& 1\le k\le \ell-1,\\
0,& k\ge \ell.
\end{cases}
\]
Equivalently, the discriminant filtration has only one jump:
\[
\varnothing=\mathcal V(D_1)=\cdots=\mathcal V(D_{\ell-1})\subsetneq \mathcal V(D_\ell)=\operatorname{maxSpec}C.
\]
In this regime, the lowest discriminant subvariety is the whole base [2506.21879].

A further rigidity theorem shows that, assuming the identity fiber algebra has the Chevalley property, the lowest discriminant subvariety
\[
\mathcal V_\ell=\mathcal V(D_\ell(H/C;\mathrm{tr}))
\]
is a closed subgroup of the affine algebraic group \(\operatorname{maxSpec}C\). In particular, it is smooth and equidimensional. The same work establishes that, for an irreducible \(H\)-module \(V\), the following are equivalent: \(V\) is tensor-reducible; \(V\) is left tensor-reducible; \(V\) is right tensor-reducible; \(V\) is annihilated by the lowest discriminant ideal; \(V\otimes V^*\) is completely reducible. It also proves that \(H\) has the Chevalley property if and only if the identity fiber algebra has the Chevalley property and all the discriminant ideals are trivial [2604.15986].

## 5. Explicit Hopf-algebra realizations

A complete calculation is available for group algebras of central extensions of Abelian groups. In the setting
\[
1\to M\to G\to A\times B\to 1,
\]
with \(A\) and \(B\) finitely generated Abelian, \(M\) finite Abelian, \(\Bbbk\) algebraically closed, \(\operatorname{char}(\Bbbk)\notin [1,m]\), \(m=[H:C]\), and \(H=\Bbbk[G]\), the paper proves
\[
V(D_k(H/C,\mathrm{tr}))=
\begin{cases}
\varnothing,& k\le m,\\
\operatorname{MaxSpec}(C),& k>m.
\end{cases}
\]
The fiber \(H/\mathfrak mH\) is simple for every maximal ideal \(\mathfrak m\), its irreducible representations are tensor products of representations of algebras
\[
R_i \cong \frac{\langle x_i,y_i\rangle}{(x_i^{l_i}-1,\; y_i^{l_i}-1,\; x_i y_i-\xi^{\,n/l_i} y_i x_i)},
\]
and every irreducible representation is maximally stable. In this class, the lowest discriminant subvariety is therefore the entire \(\operatorname{MaxSpec}(C)\) [2404.10366].

The later Hopf-algebra literature supplies more structured examples. For the big quantized Borel algebra \(\mathcal U_\epsilon^{\ge 0}(\mathfrak g)\), with central Hopf subalgebra \(\mathcal C_\epsilon^{\ge 0}(\mathfrak g)\), the identity fiber is basic and the lowest discriminant variety equals the winding orbit of the identity. For irreducible \(\mathcal U_\epsilon^{\ge 0}(\mathfrak g)\)-modules, tensor-reducible, left tensor-reducible, right tensor-reducible, \(1\)-dimensional, and maximally stable are equivalent conditions. For the generalized Liu algebras \(\mathcal B(n,w,\xi)\), where \(C=\mathbb k[x^{\pm 1}]\) and \(\operatorname{maxSpec}C\cong \mathbb k^\times\), the lowest discriminant subvariety is the finite cyclic subgroup
\[
\mathcal V(D_{n+1}(H/C;\mathrm{tr}_{\mathrm{reg}}))
=
\{\alpha\in \mathbb k^\times\mid \alpha^w=1\}.
\]
More generally, for a prime affine Cayley–Hamilton Hopf algebra of GK-dimension one with Chevalley identity fiber, either \(H\) is commutative or \(\mathcal V_\ell\) is a finite cyclic subgroup of \(\operatorname{maxSpec}C\) [2604.15986].

## 6. Analogues in matrix theory, toric geometry, and plane-curve discriminants

The matrix-theoretic analogue is the variety \(E_k\) of real symmetric \(n\times n\) matrices with at most \(n-k-1\) distinct eigenvalues. It is the zero set of the \(k\)-subdiscriminant:
\[
E_k=\{A\in M:\mathrm{sDisc}_k(A)=0\}.
\]
Up to a nonzero scalar, \(\mathrm{sDisc}_k\) is the only \(\mathrm{SO}_n\)-invariant homogeneous polynomial of degree
\[
(n-k)(n-k-1)
\]
vanishing on \(E_k\), and there is no such invariant polynomial of smaller degree. For general matrices, the same uniqueness statement holds with \(\mathrm{GL}_n\) in place of \(\mathrm{SO}_n\). The paper also proves that the minimal degree of any nonzero polynomial vanishing on \(E_k\) is
\[
\frac{(n-k)(n-k-1)}{2}.
\]
In this setting, the subdiscriminant is the canonical lowest-degree discriminant-type invariant defining the bounded-eigenvalue locus [1206.2358].

For a smooth toric variety \(X\) with base point free line bundle \(L\), the discriminant \(\mathcal D(X,L)\subset |L|\) is the subvariety of singular sections. The discriminant defect is
\[
k(X,L)=\dim |L|-1-\dim \mathcal D(X,L).
\]
The toric paper treats “lowest discriminant” behavior as the regime of smallest discriminant dimension, equivalently largest defect. It proves bounds such as
\[
k\le \min\{n-f,N\},
\]
classifies the extremal cases \(k=n\), \(k=n-1\), \(k=N\), and \(k=N-1\), and shows that when the discriminant is a hypersurface with isolated general singularities its degree is
\[
\deg \mathcal D(X,L)=c_n(J_1(L)).
\]
This is a discriminant-minimality theory rather than a theory of a single canonical lowest stratum [1911.08851].

For separable plane curves \(f\in \mathbb K[x,y]\), the paper "Plane Curves With Minimal Discriminant" defines minimality by
\[
\deg_x\Delta_y(f)=d_y-1
\]
for irreducible \(f\). This is equivalent to the closure \(C\subset \mathbb P^1\times \mathbb P^1\) being rational, having a unique place over \(x=\infty\), and being smooth outside that place. In the monic irreducible case, minimal discriminant is equivalent to being a coordinate polynomial. A plausible implication is that this paper uses “minimal discriminant” to single out a rigid geometric stratum of curves rather than a separate subvariety defined intrinsically in parameter space [1507.01091].

## 7. Cubic-surface discriminantal coverings and distinguished special loci

In the arithmetic geometry of cubic surfaces, the nearest analogue to a lowest discriminant subvariety is not the vanishing locus \(A=0\) of the discriminant itself, but a distinguished subvariety inside the discriminantal covering of the pentahedral parameter space. For a cubic surface in Sylvester’s pentahedral normal form
\[
S(a_0,\dots,a_4)\subset \mathbf P^4:\quad a_0x_0^3+\cdots+a_4x_4^3=0,\qquad x_0+\cdots+x_4=0,
\]
the discriminantal double cover is
\[
w^2=-3\,A(a_0,\dots,a_4).
\]
Inside the transformed parameter space, the paper identifies a smooth quadric surface \(Q\subset \mathbf P^4\) defined by \(l=q=0\) such that on \(Q\)
\[
(-3)A(x_0,\dots,x_4)=\big[(x_0-x_1)(x_0-x_2)(x_1-x_2)(x_3-x_4)\big]^2.
\]
Hence the discriminantal cover splits over \(Q\), and rational points on \(Q\) lift to rational points on the cover [1006.0721].

This quadric is an accumulating subvariety in the precise arithmetic sense used in the paper: it is a smooth quadric surface, rational points are much denser there than expected from the ambient threefold, and among smooth quadrics satisfying the splitting and tangency conditions it is unique up to permutation of coordinates. The cubic-surface setting therefore uses a different but related notion: a distinguished locus on which the discriminant becomes maximally special by turning into a square, rather than a lowest stratum in a chain of vanishing loci [1006.0721].

Across these settings, a common pattern emerges. The lowest discriminant subvariety is the endpoint of a hierarchy ordered by degeneracy: the first nonempty discriminant zero locus for Cayley–Hamilton Hopf algebras, the deepest singular stratum \(Z_1\) of the polynomial discriminant hypersurface, the locus cut out by the first nontrivial subdiscriminant for matrices, or a special minimal-discriminant locus in algebraic geometry. This suggests that the phrase is best understood not as a universally fixed object, but as the terminal or first nontrivial piece in a discriminant stratification adapted to a given category of geometric or representation-theoretic problems.

Source: https://www.emergentmind.com/topics/lowest-discriminant-subvariety