---
title: Binomial Edge Ideals Overview
url: https://www.emergentmind.com/topics/binomial-edge-ideals
type: topic
---

# Binomial Edge Ideals Overview

A binomial edge ideal is the quadratic binomial ideal attached to a simple graph \(G\) on \([n]=\{1,\dots,n\}\), defined in the polynomial ring
\[
S=K[x_1,\dots,x_n,y_1,\dots,y_n]
\]
by
\[
J_G=\bigl(x_i y_j-x_j y_i:\{i,j\}\in E(G),\ i<j\bigr).
\]
Its generators are the \(2\times 2\) minors of the generic \(2\times n\) matrix
\[
\begin{pmatrix}
x_1 & x_2 & \cdots & x_n\\
y_1 & y_2 & \cdots & y_n
\end{pmatrix},
\]
restricted to the edge set of \(G\). The subject lies at the interface of determinantal algebra, Gröbner theory, combinatorics of cut sets and clique complexes, and homological methods. The classical theory has also become the \(m=2\) case of several broader constructions, including generalized binomial edge ideals and ideals attached to a pair of graphs [2009.03156; 1210.7960; 1203.2775].

## 1. Determinantal origin and basic structure

The determinantal viewpoint is built into the definition. When \(G=K_n\), the ideal \(J_G\) is the full ideal of \(2\)-minors of the generic \(2\times n\) matrix, so binomial edge ideals interpolate between edge ideals of graphs and classical determinantal ideals. This positioning explains why techniques from Gröbner bases, Rees algebras, linkage, and Frobenius methods all enter the theory.

A central foundational fact is that binomial edge ideals are radical. This is already visible from their Gröbner-theoretic behavior: for suitable term orders, their initial ideals are squarefree monomial ideals. The radicality statement extends naturally to Rauh’s generalized binomial edge ideals. If \(X_0\) is a finite label set of size \(d_0>1\), \(X\) is the vertex set of a graph, and
\[
S=\Bbbk[p_{ix}:i\in X_0,\ x\in X],
\]
then the generalized binomial edge ideal is
\[
I_G=\bigl\langle f^{ij}_{xy}: i,j\in X_0,\ (x,y)\in E(G)\bigr\rangle,
\qquad
f^{ij}_{xy}=p_{ix}p_{jy}-p_{iy}p_{jx}.
\]
For \(d_0=2\), this recovers the classical binomial edge ideal. The generalized setting retains radicality and admits an explicit path-indexed Gröbner basis [1210.7960].

A further enlargement is the binomial edge ideal of a pair of graphs. If \(G_1\) is on \([m]\), \(G_2\) is on \([n]\), and \(X=(x_{ij})\) is an \(m\times n\) matrix of indeterminates, then
\[
J_{G_1,G_2}=\bigl(P_{e,f}:e\in E(G_1),\ f\in E(G_2)\bigr),
\]
where \(P_{e,f}\) is the \(2\)-minor associated to the edge pair \((e,f)\). This framework simultaneously contains classical binomial edge ideals, generalized binomial edge ideals, and ideals of adjacent minors [1203.2775].

## 2. Minimal primes, cut sets, and geometry

The minimal-prime theory of \(J_G\) is governed by vertex deletion. For a subset \(S\subseteq[n]\), let \(G\setminus S\) have connected components \(G_1,\dots,G_{c(S)}\), and let \(\widetilde G_i\) denote the complete graph on the vertex set of \(G_i\). Then the associated prime is
\[
P_S(G)= (x_i,y_i : i\in S)+J_{\widetilde G_1}+\cdots+J_{\widetilde G_{c(S)}}.
\]
The ideal \(J_G\) is the intersection of such primes, and \(P_S(G)\) is minimal exactly when \(S\) is a cut-point set, equivalently a cut set in the sense that every vertex of \(S\) is essential for the induced disconnection. For connected \(G\),
\[
\operatorname{height} P_S(G)=n-c(S)+|S|.
\]
This dictionary between minimal primes and cut sets is one of the basic structural features of the subject [2009.03156; 2012.14904].

Unmixedness is encoded numerically by the same combinatorics. For connected \(G\), the standard criterion is
\[
J_G \text{ is unmixed } \Longleftrightarrow c_G(S)=|S|+1 \text{ for every cut set } S.
\]
This condition recurs throughout the literature because it is the entry point for Cohen–Macaulayness, Serre conditions, and dual-graph connectedness arguments [1205.0475; 2306.17076].

The geometric decomposition has an especially transparent form for generalized binomial edge ideals. Writing a point of the ambient affine space as a matrix whose columns are \(\tilde p_x\), the defining equations impose proportionality of columns along edges. For any subset \(Y\subseteq X\), one considers the locus \(V_{G,Y}\) where columns outside \(Y\) vanish and columns within the same connected component of \(G_Y\) are proportional. Then
\[
V_G=\bigcup_{Y\subseteq X} V_{G,Y},
\]
each \(V_{G,Y}\) is the variety of a prime ideal \(I_{G,Y}\), and the irreducible components are rational. This gives a geometric counterpart to the combinatorial primary decomposition [1210.7960].

The same cut-set combinatorics also controls the generic initial ideal. For the \(\mathbb Z^n\)-graded generic initial ideal of \(J_G\), the minimal primes are indexed by pairs \((S,W)\), where \(S\) is a cut set and \(W\) is a transversal of \(G-S\). This description is central in the combinatorial analysis of Serre’s condition \((S_2)\) [2306.17076].

## 3. Gröbner bases, admissible paths, and special graph classes

Gröbner theory for binomial edge ideals is path-theoretic. For the classical ideal \(J_G\), the reduced Gröbner basis with respect to the lexicographic order
\[
x_1>\cdots>x_n>y_1>\cdots>y_n
\]
is indexed by admissible paths \(\pi:i=i_0,i_1,\dots,i_r=j\), together with monomials \(u_\pi\), and consists of the elements
\[
u_\pi f_{ij}.
\]
A standard consequence is that \(\operatorname{in}_{<}(J_G)\) is squarefree, hence \(J_G\) is radical [1402.2041; 1412.3542].

The generalized theory has a parallel but უფრო elaborate path combinatorics. There the reduced Gröbner basis is indexed by admissible paths together with strictly \(\pi\)-antitone labelings \(\kappa\), and the basis elements are
\[
u_\pi^\kappa\, f^{\kappa(y)\kappa(x)}_{xy}.
\]
Their initial monomials are squarefree, which again yields radicality [1210.7960].

A decisive graph class is the class of closed graphs. A graph is closed if it admits a labeling for which the natural generators of \(J_G\) form a quadratic Gröbner basis with respect to lex order. For such graphs,
\[
\operatorname{in}_{<}(J_G)=(x_i y_j:\{i,j\}\in E(G),\, i<j),
\]
which is the edge ideal of a bipartite graph. Closed graphs therefore occupy a privileged position at the intersection of quadratic Gröbner theory, normal torsion-freeness of initial ideals, and explicit Betti calculations [2009.03156; 2208.10749].

Weakly closed graphs form a broader class. They are precisely the co-comparability graphs, and they admit an algebraic characterization in terms of Knutson ideals for a specific polynomial
\[
f=y_1 f_{12}f_{23}\cdots f_{n-1,n}x_n.
\]
More precisely, after a suitable labeling,
\[
G \text{ is weakly closed } \iff J_G \in \mathcal C_f.
\]
This places weakly closed graphs between the quadratic Gröbner-basis world of closed graphs and the positive-characteristic theory of \(F\)-purity [2208.10749].

## 4. Unmixedness, Cohen–Macaulayness, and Serre-type conditions

A large part of the theory asks when graph-theoretic constraints force strong depth properties. One basic mechanism is gluing along a free vertex. If \(G=G_1\cup G_2\) with \(V(G_1)\cap V(G_2)=\{u\}\) and \(u\) free in both clique complexes, then
\[
\operatorname{depth} S/J_G=\operatorname{depth} S_1/J_{G_1}+\operatorname{depth} S_2/J_{G_2}-2,
\]
and
\[
J_G \text{ is Cohen--Macaulay } \Longleftrightarrow J_{G_1}\text{ and }J_{G_2}\text{ are Cohen--Macaulay}.
\]
This extends to tree-like gluings of several graphs and yields a construction toolkit for Cohen–Macaulay binomial edge ideals [1205.0475].

The cone construction behaves differently. If \(G=\operatorname{cone}(v,H)\) with \(H\) connected and \(J_H\) unmixed, then \(J_G\) is unmixed if and only if \(H\) is complete; if this holds, then \(J_G\) is Cohen–Macaulay. For a cone over two connected components, unmixedness is equivalent to unmixedness of both components, and Cohen–Macaulayness follows when both components are Cohen–Macaulay [1205.0475].

For bipartite graphs there is a full classification. The connected bipartite graphs whose binomial edge ideals are Cohen–Macaulay are exactly those obtained from the basic blocks \(F_m\) by the gluing operations \(*\) and \(\circ\). In this class, the connectedness of the dual graph is equivalent to Cohen–Macaulayness, giving a converse to Hartshorne’s connectedness theorem for connected bipartite binomial edge ideals [1704.00152].

Serre’s condition \((S_2)\) also has a purely graph-theoretic formulation. A graph \(G\) is called accessible when \(J_G\) is unmixed and the family of cut sets forms an accessible set system. The main theorem in this direction is
\[
R/J_G \text{ satisfies } (S_2)\ \Longleftrightarrow\ G \text{ is accessible}.
\]
A key intermediate fact is that, for unmixed binomial edge ideals, accessibility of the cut-set family is equivalent to strong accessibility [2306.17076].

At the lower end of the depth spectrum, the poset-topological approach gives a sharp classification:
\[
\operatorname{depth} S/J_G =4
\quad\Longleftrightarrow\quad
G=G' * 2K_1
\]
for some graph \(G'\), provided \(n\ge 4\). More generally, if \(G\) has at least three vertices, then \(\operatorname{depth} S/J_G\ge 4\) [2012.14904].

## 5. Regularity, Betti theory, projective dimension, and linkage

The homological profile of \(J_G\) reflects graph structure with unusual precision. A fundamental classification says that, for a graph \(G\) with no isolated vertices,
\[
J_G \text{ has a pure resolution}
\iff
G \text{ is a complete graph, a complete bipartite graph, or a disjoint union of paths.}
\]
The same work introduced free cut edge switching, an operation preserving all graded Betti numbers, and hence preserving projective dimension and regularity [1401.4817].

For block graphs, depth and regularity admit especially explicit descriptions. If \(G\) is a block graph with \(c\) connected components, then
\[
\operatorname{depth}(S/J_G)=\operatorname{depth}\bigl(S/\operatorname{in}_{<}(J_G)\bigr)=n+c.
\]
For the subclass of \(C_\ell\)-graphs,
\[
\operatorname{reg}(S/J_G)=\operatorname{reg}\bigl(S/\operatorname{in}_{<}(J_G)\bigr)=\ell.
\]
As a consequence, for a tree \(T\) with longest induced path of length \(\ell\),
\[
\operatorname{reg}(S/J_T)=\ell
\quad\Longleftrightarrow\quad
T \text{ is a caterpillar.}
\]
These statements link extremal regularity directly to clique-chain structure [1402.2041].

For generalized block graphs, the number \(m(G)\) of minimal cut sets enters the last Betti corner. If \(G\) is connected and indecomposable, then
\[
\beta_{p(G),\,p(G)+m(G)+1}^S(S/J_G)
\]
is an extremal Betti number, so \(\operatorname{reg}(S/J_G)\ge m(G)+1\). Moreover, \(S/J_G\) has a unique extremal Betti number exactly when no induced flower graph \(F_{h,k}(v)\) with \(h+k\ge 3\) occurs; in that case
\[
\operatorname{reg}(S/J_G)=m(G)+1.
\]
This gives a sharp obstruction theory for the Betti table of generalized block graphs [1910.06787].

Projective dimension can also be controlled combinatorially in specific families. For the cycle \(C_n\) with \(n\ge 4\),
\[
\operatorname{pd}(S/J_{C_n})=\bigheight(J_{C_n})=n,
\]
while for the crown graph \(C_{n,n}\),
\[
\operatorname{pd}(S/J_{C_{n,n}})=\bigheight(J_{C_{n,n}})=4n-6.
\]
For complete multipartite graphs \(K_{n_1,\dots,n_k}\), the equality \(\operatorname{pd}(S/J_G)=\bigheight(J_G)\) holds exactly when \(n_1=2\) [2507.03889].

Linkage imposes an even stronger rigidity. For a connected graph \(G\), \((J_G)_m\) is licci if and only if \(G\) is either a path graph or a triangle with possibly some paths attached to some of its vertices. In the chordal case, the Cohen–Macaulay hypothesis in this criterion can be weakened to unmixedness [1910.03612].

## 6. Koszulness and positive-characteristic properties

Because \(J_G\) is always generated by quadrics, Koszulness has long been a central question. Earlier work established the implication
\[
G \text{ closed } \implies S/J_G \text{ is Koszul } \implies G \text{ is chordal and claw-free},
\]
showed that gluing along a free vertex preserves Koszulness, characterized cone graphs by
\[
\operatorname{cone}(v,G) \text{ is Koszul }
\iff
\operatorname{cone}(v,G) \text{ is closed }
\iff
\operatorname{Ind}(G)\text{ is triangle-free and }G\text{ is chordal},
\]
and computed the quadratic dual of \(S/J_G\) explicitly. The first two Betti numbers in the minimal free resolution of the residue field over \(S/J_G\) are
\[
\beta_1=2n,\qquad \beta_2=\binom{2n}{2}+|E(G)|,
\]
and these first two syzygies are always linear [1412.3542].

A complete graph-theoretic classification is now known:
\[
J_G \text{ is Koszul }
\iff
G \text{ is strongly chordal and claw-free}.
\]
Equivalently, \(J_G\) is Koszul if and only if \(G\) is chordal, claw-free, and tent-free. This shows that Koszulness is strictly weaker than having a quadratic Gröbner basis; closed graphs form a proper subclass, and the net is the smallest obstruction separating the two notions [2601.15243].

In positive characteristic, weakly closed graphs are a natural sufficient class for \(F\)-purity. Indeed, if \(G\) is weakly closed, then \(R/J_G\) is \(F\)-pure in positive characteristic; the same extension holds for generalized binomial edge ideals [2208.10749].

More recently, Matsuda’s characteristic-\(2\) conjecture was resolved:
\[
R_G \text{ is F-pure in characteristic }2
\iff
G \text{ is weakly closed}.
\]
At the same time, the “eventual \(F\)-purity” conjecture was disproved in a strong form: if \(G\) contains an asteroidal triple, then \(R_G\) is not \(F\)-pure in any positive characteristic. For chordal graphs, \(F\)-purity in all characteristics, \(F\)-purity in some positive characteristic, AT-freeness, and weak closedness are equivalent. The same work also notes that every \(R_G\) is \(F\)-injective because \(J_G\) has a squarefree initial ideal for suitable term orders [2601.15403].

## 7. Powers, symbolic powers, and major generalizations

The symbolic-power problem asks when
\[
J_G^{(k)}=J_G^k
\qquad (k\ge 1).
\]
For a homogeneous ideal with squarefree initial ideal, a transfer principle can relate symbolic powers of the ideal to those of its initial ideal. Applied to binomial edge ideals, this yields the theorem that if \(G\) is connected and \(\operatorname{in}_{<}(J_G)\) is normally torsion-free, then
\[
J_G^{(k)}=J_G^k \quad \text{for all } k\ge 1.
\]
Since closed graphs have
\[
\operatorname{in}_{<}(J_G)=(x_i y_j:\{i,j\}\in E(G),\, i<j),
\]
the edge ideal of a bipartite graph, one obtains
\[
G \text{ closed } \Longrightarrow J_G^{(k)}=J_G^k \text{ for all }k\ge 1.
\]
The same analysis exhibits both a positive and a negative boundary case: for \(C_4\), the equality holds for all \(k\); for \(C_5\), the initial ideal is not normally torsion-free because \(\operatorname{in}_{<}(J_{C_5})^{(2)}\ne \operatorname{in}_{<}(J_{C_5})^2\), even though \(J_{C_5}^{(2)}=J_{C_5}^2\) still holds [2009.03156].

The homology of powers has also begun to stabilize in explicit closed forms. If \(G\) is closed and \(K_4\)-free, with \(e=|E(G)|\) and \(t\) the number of triangles, then for every \(m\ge 1\) and \(i\ge 0\),
\[
\beta_{i,\,2m+i}(J_G^m)
=
\beta_{i,\,2m+i}\bigl(\init_{\lex}(J_G^m)\bigr)
=
\binom{e+m-i-1}{m-i}\binom{2t}{i}.
\]
Thus the entire \(2m\)-linear strand of each power depends only on \(e\), \(t\), and \(m\), and agrees with that of the lexicographic initial ideal [2601.10842].

Among higher-row analogues, generalized binomial edge ideals have recently been shown to be Cartwright–Sturmfels for all \(m\). If \(I_G(m)\) denotes the ideal of \(2\)-minors in the columns indexed by the edges of \(G\), then \(I_G(m)\) is Cartwright–Sturmfels, and its \(\mathbb Z^n\)-graded generic initial ideal is generated by monomials
\[
\prod_{j\in A} x_{i_j j}
\]
where \(G_A\) is connected and
\[
\sum_{j\in A} i_j \le m(|A|-1).
\]
This places generalized binomial edge ideals inside a very rigid multigraded class and sharpens the older radicality theorem [2512.22012].

Two other related families show how special the classical theory is. For ideals \(J_{G_1,G_2}\) attached to a pair of graphs, radicality is equivalent to one of the graphs being complete, and a quadratic Gröbner basis occurs exactly when one graph is complete and the other is closed [1203.2775]. For parity binomial edge ideals,
\[
I_G=\bigl(x_i x_j-y_i y_j:\{i,j\}\in E(G)\bigr),
\]
the Gröbner basis is generally not squarefree, and
\[
I_G \text{ is radical } \iff \bigl(\operatorname{char}(\Bbbk)\neq 2\bigr)\ \text{or}\ G\ \text{is bipartite}.
\]
On bipartite graphs, however, parity and classical binomial edge ideals coincide up to a variable swap, which clarifies both the analogy and the divergence between the two theories [1503.00584].

Source: https://www.emergentmind.com/topics/binomial-edge-ideals