---
title: 'Permanental Ideal: Algebraic and Combinatorial Insights'
url: https://www.emergentmind.com/topics/permanental-ideal
type: topic
---

# Permanental Ideal: Algebraic and Combinatorial Insights

A permanental ideal is an ideal generated by permanents of fixed-size submatrices of a matrix, or by corresponding permanent-type expressions in structured arrays such as symmetric matrices, Hankel matrices, and hypermatrices. For a square matrix \(A=(a_{ij})\), the permanent is
\[
\operatorname{perm}(A)=\sum_{\sigma\in S_r}\prod_{i=1}^r a_{i,\sigma(i)},
\]
formally analogous to the determinant but with all signs equal to \(+1\). In a polynomial ring on the entries of a generic matrix, the size-\(r\) permanental ideal is generated by all \(r\times r\) subpermanents. Recent work treats permanental ideals through codimension, minimal primes, Gröbner bases, Hilbert–Poincaré series, saturation, and colon invariants, and repeatedly emphasizes that their behavior differs sharply from the determinantal case [2402.17839][2505.03367].

## 1. General definition and basic constructions

Let \(M=(x_{ij})\) be a generic matrix. The corresponding permanental ideal of size \(r\) is the ideal generated by all \(r\times r\) permanents of submatrices of \(M\). In the maximal-row-size case for a generic \(k\times n\) matrix, the associated affine permanental variety is
\[
P_{k,n}=\{\,\operatorname{prk}(M)\le k-1\,\}\subset F^{k\times n},
\]
where \(\operatorname{prk}(M)\) is the permanental rank, defined as the largest integer \(r\) such that \(M\) has an \(r\times r\) submatrix with nonzero permanent. Ideal-theoretically,
\[
I(P_{k,n})=\big\langle \operatorname{perm}(M_{R,C}) \;\big|\; R\subset [k],\ C\subset [n],\ |R|=|C|=k \big\rangle.
\]
For a square \(k\times k\) matrix, the permanental hypersurface
\[
P=\{\operatorname{perm}(M)=0\}
\]
has singular locus
\[
\operatorname{Sing}(P)=\{\,\operatorname{prk}(M)\le k-2\,\},
\]
because the first derivatives of the permanent are \((k-1)\times (k-1)\) permanents [2402.17839].

The most frequently studied low-degree family is the \(2\times 2\) case. If \(X=(x_{ij})\) is an \(m\times n\) matrix, then
\[
P_2(X)=\big(x_{ij}x_{k\ell}+x_{i\ell}x_{kj}\;\big|\; 1\le i<k\le m,\;1\le j<\ell\le n\big).
\]
This family appears for generic matrices, generic symmetric matrices, and generic Hankel matrices, and it is the setting in which the most detailed calculations of primary decomposition, depth, Gröbner bases, and \(v\)-numbers are presently available [2605.11621].

## 2. Complete graphs and the \(2\times n\) permanental ideal

A particularly explicit permanental ideal arises from the \(2\times n\) matrix
\[
\begin{pmatrix} x_1 & x_2 & \cdots & x_n\\ y_1 & y_2 & \cdots & y_n \end{pmatrix}.
\]
Its \(2\times 2\) permanents are
\[
x_i y_j + x_j y_i \qquad (1\le i<j\le n),
\]
and the ideal they generate is equivalent, when \(\operatorname{char}(k)\neq 2\), to the parity binomial edge ideal of the complete graph \(K_n\). In the standard graded ring
\[
R=k[x_1,\dots,x_n,y_1,\dots,y_n],
\]
the complete-graph parity binomial edge ideal is
\[
I_{K_n}=(g_{ij}\mid 1\le i<j\le n), \qquad g_{ij}=x_i x_j-y_i y_j,
\]
and the linear change of coordinates
\[
x_i \mapsto x_i-y_i,\qquad y_i\mapsto x_i+y_i
\]
transforms \(I_{K_n}\) into the permanental edge ideal
\[
\bigl(x_i y_j + x_j y_i \mid 1\le i<j\le n\bigr).
\]
The paper studying this case explicitly switches to the parity-binomial presentation because the permanental ideal can contain monomials and its combinatorics are more opaque in that form [2003.00977].

The central homological result is a closed formula for the Hilbert–Poincaré polynomial:
\[
P_{R/I_{K_n}}(t)
=
2(1-t)^n+
\left(
-1+3t+\frac{n^2-n-6}{2}t^2+\frac{n^2-3n+2}{2}t^3
\right)(1-t)^{2n-3}.
\]
Equivalently,
\[
\operatorname{HP}_{R/I_{K_n}}(t)=
\frac{
2(1-t)^n+
\left(
-1+3t+\frac{n^2-n-6}{2}t^2+\frac{n^2-3n+2}{2}t^3
\right)(1-t)^{2n-3}
}{(1-t)^{2n}}.
\]
From this formula one obtains
\[
\dim(R/I_{K_n})=n,\qquad
\depth(R/I_{K_n})=3,\qquad
\reg(R/I_{K_n})=3,\qquad
\pdim(R/I_{K_n})=2n-3
\]
for \(n\ge 3\), together with the extremal Betti number
\[
\beta_{2n-3,\,2n}(R/I_{K_n})=\frac{n^2-3n+2}{2}.
\]
The notable feature is that depth and Castelnuovo–Mumford regularity are independent of \(n\), whereas projective dimension grows linearly [2003.00977].

The proof is organized around the explicit primary decomposition
\[
I_{K_n}=J_{K_n}\cap \mathfrak{p}^+\cap \mathfrak{p}^-\cap \bigcap_{1\le i<j\le n}P_{ij},
\]
where \(J_{K_n}\) is a saturation, \(\mathfrak p^\pm=(x_i\pm y_i\mid i\in[n])\), and
\[
P_{ij}=(g_{ij})+\mathfrak{m}_{[n]\setminus\{i,j\}}.
\]
The inductive argument introduces ideals \(I_0:=I_{K_n}\) and \(I_k:=I_{k-1}+(f_{kn})\), with \(f_{ij}=x_i y_j-x_j y_i\), and repeatedly uses colon computations such as
\[
I_{k-1}:f_{kn}=P_{kn}
\]
and short exact sequences
\[
0\to R/(I_{k-1}:f_{kn})(-2)\to R/I_{k-1}\to R/I_k\to 0.
\]
This case has become a standard benchmark because it is one of the rare permanental families with an exact homological description [2003.00977].

## 3. Codimension, maximal permanents, and permanental varieties

For maximal permanents of a generic \(k\times n\) matrix, a basic estimate is
\[
n-k+1 \le \operatorname{codim}(P_{k,n}) \le n.
\]
In characteristic \(0\), if \(2\le k\le 5\) and \(n\ge k+1\), then
\[
\operatorname{codim}(P_{k,n})=n.
\]
Equivalently, for \(2\le k\le 5\), the ideal generated by all maximal \(k\times k\) permanents of a generic \(k\times n\) matrix has height \(n\). In particular, for \(2\le k\le 4\), \(P_{k,k+1}\) is a complete intersection. The central reduction principle states that if \(P_{h,h+1}\) has codimension \(h+1\) for all \(h\le k\), then \(P_{k,n}\) has codimension \(n\) for all \(n\ge k+1\). This reduces the general maximal-permanent codimension problem to the borderline case \(n=k+1\) [2402.17839].

The same work studies the singular locus of the square permanental hypersurface. For a generic \(k\times k\) matrix over any field,
\[
\operatorname{codim} \operatorname{Sing}(P)\ge 4,
\]
and this implies that \(\operatorname{perm}(M)\) is irreducible. For \(k\ge 3\), a theorem of von zur Gathen gives
\[
5 \le \operatorname{codim}\operatorname{Sing}(P)\le 2k,
\]
and for \(k\ge 6\) this lower bound is improved to
\[
6 \le \operatorname{codim}\operatorname{Sing}(P)\le 2k.
\]
The improved lower bound is one of the main geometric results presently available for permanental varieties [2402.17839].

A major methodological contribution is the introduction of a \(\mathbb C^*\)-action on matrix space. For \(Y=P_{k,k+1}\subset \mathbb C^{k\times(k+1)}\), scaling the first row preserves the vanishing of maximal permanents. If \(X\subset Y\) is an irreducible component, the paper proves a Białynicki-Birula-type description
\[
X=T^1_{X,Y},
\]
and for a general fixed point \(p\in X^T\),
\[
\dim X=\dim X^T+\dim \ker(B_1(A_p)),
\qquad
\operatorname{codim} X=\operatorname{codim}_{V^T}(X^T)+\operatorname{rk}(B_1(A_p)).
\]
For the singular locus, a related torus action scaling the first two rows gives
\[
T_{Y,p}=V^T\oplus \ker(L_p)^{\oplus 2},
\]
hence
\[
\operatorname{codim}_V X \ge \operatorname{codim}_{V^T}(X^T)+2\,\operatorname{rk}(L_p).
\]
This suggests that codimension in permanental geometry can sometimes be split into a fixed-locus term and a rank term, even though no determinantal analogue of Eagon–Northcott is available [2402.17839].

## 4. \(2\times 2\) permanental ideals of hypermatrices

The hypermatrix generalization replaces matrix indices by
\[
N=[r_1]\times\cdots\times [r_n]
\]
and works in the ring
\[
R=k[x_a : a\in N].
\]
For \(L\subseteq [n]\) and \(a,b\in N\), the switch function \(s(L,a,b)\) replaces the coordinates of \(a\) in positions \(L\) by those of \(b\). The generalized permanent-type binomials are
\[
g_{K,a,b}=x_a x_b + x_{s(K,a,b)}x_{s(K,b,a)}.
\]
When \(d(a,b)=2\) and \(a_i\neq b_i\), the element
\[
g_{i,a,b}=x_a x_b + x_{s(i,a,b)}x_{s(i,b,a)}
\]
is called a slice permanent. For \(t\in[n]\), the principal ideal family is
\[
J^{\langle t\rangle}
=
\bigl(g_{i,a,b} : a,b\in N,\ d(a,b)=2,\ i\in [t],\ a_i\neq b_i\bigr).
\]
The standing assumption is \(\operatorname{char}(k)\neq 2\), since otherwise permanents and determinants coincide at the level of signs [1110.2716].

The key combinatorial notions are \(t\)-switchability and \(t\)-signedness. A subset \(S\subseteq N\) is \(t\)-switchable if switching in any of the first \(t\) coordinates preserves membership whenever two elements differ in exactly two coordinates. A \(t\)-switchable set is \(t\)-signed if each connected component satisfies at least one of three parity-compatible conditions: all elements have the same first \(t\) coordinates; any two elements differ in at most one component; or the parity of path length between any two elements is independent of the path. The passage from switchability to signedness is forced by a parity obstruction: if two connected elements admit paths of different parity, then local permanental relations produce monomials, which changes the prime structure [1110.2716].

For a \(t\)-signed set \(S\), the associated prime candidate is
\[
Q_S^{\langle t\rangle}
=
\mathrm{Var}_S^{\langle t\rangle} + \widetilde J_S^{\langle t\rangle},
\]
where \(\mathrm{Var}_S^{\langle t\rangle}\) is the variable ideal generated by variables outside \(S\), and \(\widetilde J_S^{\langle t\rangle}\) is generated by signed binomials
\[
h_{S,K,a,b}
=
x_a x_b
-
(-1)^{\#K\cdot \mathrm{pl}_S(a,b)}
x_{s(K,a,b)}x_{s(K,b,a)}.
\]
The main structural results are that \(\widetilde G_S^{\langle t\rangle}\) is a Gröbner basis for \(\widetilde J_S^{\langle t\rangle}\), that both \(\widetilde J_S^{\langle t\rangle}\) and \(Q_S^{\langle t\rangle}\) are prime when \(S\) is \(t\)-signed, and that the minimal primes of \(J^{\langle t\rangle}\) are exactly
\[
\{Q_S^{\langle t\rangle} : S \text{ is a maximal } t\text{-signed set}\}.
\]
This is the permanental analogue of the determinantal classification by maximal \(t\)-switchable sets, but with the crucial replacement of switchability by signedness [1110.2716].

In the matrix case \(N=[r_1]\times[r_2]\), this framework recovers the classical description of the ideal of \(2\times 2\) permanents of a generic matrix. When \(r_1,r_2>2\), the maximal \(t\)-signed sets are all \(2\times 2\) submatrices, all \(1\times r_2\) submatrices, and all \(r_1\times 1\) submatrices. The introduction recalls that for a generic \(m\times n\) matrix, the ideal of \(2\times2\) permanents has
\[
\binom m2 \binom n2 + m+n
\]
minimal components and one embedded component when \(m,n\ge 3\), in sharp contrast to the prime determinantal ideal generated by \(2\times 2\) minors [1110.2716].

## 5. Symmetric matrices

For a symmetric matrix
\[
X=(x_{ij})_{1\le i,j\le n}, \qquad x_{ij}=x_{ji},
\]
over a field \(K\) with \(\operatorname{char}(K)\neq 2\), the ideal
\[
P_2(X)\subseteq R=K[x_{ij}\mid 1\le i\le j\le n]
\]
is generated by all \(2\times 2\) permanents of symmetric \(2\times 2\) submatrices. Basic generators include
\[
x_{ii}x_{jj}+x_{ij}^2
\qquad\text{and}\qquad
x_{ii}x_{jk}+x_{ij}x_{ik}.
\]
A distinctive feature of the symmetric case is that, because \(2\) is invertible, the ideal also contains monomials such as
\[
x_{ij}x_{kl}\qquad (i,j,k,l\text{ distinct},\ i<j,\ k<l).
\]
This already separates the symmetric permanental ideal from the corresponding determinantal ideal [2505.03367].

With respect to any lexicographic diagonal monomial order, \(P_2(X)\) has a reduced Gröbner basis consisting of the quadratic families
\[
(1a)\; x_{ii}x_{jj}+x_{ij}^2,\qquad
(1b)\; x_{ii}x_{jk}+x_{ij}x_{ik},\qquad
(1c)\; x_{ij}x_{kl},
\]
together with cubic monomials of types \((2a)\), \((2b)\), \((2c)\), \((3a)\), \((3b)\), \((3c)\), and degree-\(4\) monomials
\[
(6a)\; x_{ik}^3x_{jj},\qquad
(6b)\; x_{ik}^2x_{jj}^2.
\]
The cardinality of this Gröbner basis is
\[
\binom{n}{2}+11\binom{n}{3}+7\binom{n}{4}.
\]
The quotient has
\[
\dim(R/P_2(X))=2
\]
and
\[
\depth(R/P_2(X))=
\begin{cases}
2 & \text{if } n=2,\\
0 & \text{if } n>2.
\end{cases}
\]
Thus \(R/P_2(X)\) is Cohen–Macaulay only in the \(n=2\) case [2505.03367].

The minimal primes are exactly the ideals
\[
P_{ij}=
\big(x_{ii}x_{jj}+x_{ij}^2,\ \text{all }x_{kl}\text{ with }\{k,l\}\not\subseteq\{i,j\}\big),
\qquad i<j.
\]
Hence \(P_2(X)\) has exactly \(\binom{n}{2}\) minimal primes. Its radical is
\[
\sqrt{P_2(X)}
=
P_2(X)
+
\big(x_{ij}x_{kl}\mid i\neq j,\ k\neq l,\ (i,j)\neq (k,l)\big)
+
\big(x_{ij}x_{kk}\mid i,j,k\text{ distinct}\big),
\]
so \(P_2(X)\) is radical if and only if \(n=2\). For \(n\ge 3\), the irredundant primary decomposition is
\[
P_2(X) =
\left(\bigcap_{1\le i<j\le n}P_{ij}\right)
\cap
\left(\bigcap_{k=1}^n Q_k\right)
\cap
\Big(P_2(X)+ (x_{ij}^3\mid 1\le i<j\le n)+(x_{kk}^2\mid k\in[n])\Big),
\]
where each \(Q_k\) is \(P_k\)-primary for
\[
P_k=(x_{ij}\mid (i,j)\neq (k,k)).
\]
The associated primes are therefore the \(\binom{n}{2}\) minimal primes \(P_{ij}\), the embedded primes \(P_1,\dots,P_n\), and the homogeneous maximal ideal \(\mathfrak m\), for a total of
\[
\binom{n}{2}+n+1
\]
associated primes [2505.03367].

Characteristic \(2\) is exceptional. In that case
\[
ad+bc=ad-bc,
\]
so the \(2\times 2\) permanents coincide with the \(2\times 2\) minors and \(P_2(X)=I_2(X)\). The non-\(2\) theory is therefore genuinely permanental rather than determinantal [2505.03367].

## 6. Invariants, colon structure, and related rank-based viewpoints

A recent line of work studies the \(v\)-number of permanental ideals. If \(R=K[x_1,\dots,x_n]\) is standard graded and \(I\subseteq R\) is a graded ideal, then
\[
v(I)\coloneqq \min\{d \mid \exists f\in R_d \text{ and } p\in \operatorname{Ass}(I)\text{ such that } I:f=p\}.
\]
For \(2\times 2\) permanental ideals over a field of characteristic \(\neq 2\), the results are complete in three matrix classes. For a generic \(m\times n\) matrix \(X\),
\[
v(P_2(X))=
\begin{cases}
0,& (m,n)=(2,2),\\
2,& m=2,\ n\ge 3,\\
3,& m,n\ge 3;
\end{cases}
\]
for a generic symmetric matrix \(Y\),
\[
v(P_2(Y))=
\begin{cases}
0,& n=2,\\
3,& n\ge 3;
\end{cases}
\]
and for a generic Hankel matrix \(H\),
\[
v(P_2(H))=
\begin{cases}
0,& (m,n)=(2,2),\\
1,& m\ge 3 \text{ and } m+n\ge 10,\\
1,& (m,n)\in\{(3,6),(4,5)\},\\
2,& m=2 \text{ and } n\ge 4,\\
2,& (m,n)\in\{(3,4),(3,5),(4,4)\},\\
3,& (m,n)\in\{(2,3),(3,3)\}.
\end{cases}
\]
The proofs rely on associated-prime descriptions, explicit Gröbner bases, and colon identities such as
\[
P_2(X):x_{11}x_{12}x_{13} = (x_{ij}\mid 2\le i\le m,\ 1\le j\le n)
\]
in the generic case [2605.11621].

A related matrix-theoretic viewpoint studies vanishing of subpermanents through permanental rank and permanental nullity. For a square matrix \(A\), the permanental polynomial is
\[
\pi(A,x)=\per(xI-A),
\]
and the permanental nullity \(\eta_{\per}(A)\) is the multiplicity of \(0\) as a root of \(\pi(A,x)\). The coefficient identity
\[
c_{n-k}=(-1)^k\sum_{|S|=k}\per(A[S,S])
\]
for
\[
\pi(A,x)=\sum_{i=0}^n c_{n-i}x^{n-i}
\]
immediately yields the general inequality
\[
\rho_{\per}(A)+\eta_{\per}(A)\ge n.
\]
Equality is proved for three classes: non-negative symmetric matrices, positive semi-definite matrices, and balanced signed symmetric matrices:
\[
\rho_{\per}(A)+\eta_{\per}(A)=n.
\]
The paper also shows that equality can fail for arbitrary square matrices and for symmetric matrices with mixed signs. This suggests that positivity and sign-structure are decisive when one tries to relate vanishing of permanents to polynomial invariants or to the geometry of permanent-vanishing loci [2507.01344].

Taken together, these developments indicate a recurring pattern. Exact formulas are currently available for highly structured permanental ideals—complete-graph \(2\times n\) ideals, maximal-permanent varieties in small row number, symmetric \(P_2\)-ideals, and hypermatrix slice-permanent ideals—while more general permanental geometry remains fragmentary. A plausible implication is that permanental ideals admit tractable theories chiefly when combinatorial structure, characteristic restrictions, or positivity eliminate sign-cancellation and make colon ideals or fixed-point stratifications accessible [2003.00977][2402.17839].

Source: https://www.emergentmind.com/topics/permanental-ideal