Linear Resolutions in Commutative Algebra

• Linear resolutions are graded module resolutions where all nonzero syzygies occur at minimal degrees, precisely along the diagonal j = d + i.
• They play a pivotal role in homological and combinatorial commutative algebra, linking concepts such as linear quotients, initially linear syzygies, and lcm-lattice properties.
• Recent studies investigate how linearity behaves under products, powers, and ambient ring changes, impacting applications in toric geometry, edge ideal theory, and beyond.

Searching arXiv for the primary and supporting papers on linear resolutions to ground the article in current metadata and citations. Linear resolution is the condition that a graded module generated in one degree has all syzygies in the minimal possible degrees. If $I\subset S=K[x_1,\dots,x_n]$ is generated in degree $d$, this means that in the minimal graded free resolution of $I$, the graded Betti numbers satisfy $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$; equivalently, $\operatorname{reg}(I)=d$. This notion sits at the intersection of homological algebra, combinatorial commutative algebra, toric geometry, and the study of singular quotients, and it underlies current work on products and powers of ideals, Stanley–Reisner theory, edge and hypergraph ideals, and Gorenstein algebras (Conca et al., 2017).

1. Homological formulation

Let $M$ be a finitely generated graded module over a standard graded polynomial ring $S$. Its minimal graded free resolution has the form

$$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$

If $M$ is generated in a single degree $d$, then $d$0 has a $d$1-linear resolution precisely when all nonzero $d$2 lie on the diagonal $d$3. The same condition can be expressed in terms of Castelnuovo–Mumford regularity: $d$4 For ideals this is the standard numerical formulation of linearity, and for quotient rings $d$5 it becomes $d$6 when $d$7 is generated in degree $d$8 (Sköldberg, 2011).

Over quotient rings $d$9, one uses relative regularity

$I$0

If $I$1 is generated in degree $I$2, then $I$3 has a linear resolution over $I$4 if and only if $I$5, equivalently $I$6 for $I$7. This distinction between absolute regularity over $I$8 and relative regularity over $I$9 becomes essential on singular ambient rings, especially hypersurfaces, where minimal free resolutions are typically infinite (Conca et al., 2017).

A useful weakening is partial linearity. If $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$0 is generated in degree $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$1, condition $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$2 requires

$\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$3

where $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$4 is the largest degree of a minimal generator in homological degree $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$5. Thus $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$6 means linear presentation, $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$7 with $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$8 means a full linear resolution, and $\beta_{i,j}(I)\neq 0\Rightarrow j=d+i$9 is the “almost linear” case, linear in all steps except possibly the last (Dao et al., 2022).

2. Structural criteria and equivalent languages

Several distinct frameworks characterize linear resolutions, often translating a homological condition into Gröbner, lattice-theoretic, or topological data.

Framework	Hypothesis	Consequence
Initially linear syzygies	Initial module of syzygies generated by terms $\operatorname{reg}(I)=d$0	If equigenerated, the module has a linear resolution
Linear quotients	$\operatorname{reg}(I)=d$1 generated by variables	For equigenerated monomial ideals, linear quotients imply a linear resolution
lcm-lattice	$\operatorname{reg}(I)=d$2 is $\operatorname{reg}(I)=d$3-degree graded and Cohen–Macaulay	$\operatorname{reg}(I)=d$4 has a $\operatorname{reg}(I)=d$5-linear resolution
Alexander dual	$\operatorname{reg}(I)=d$6 squarefree, $\operatorname{reg}(I)=d$7 pure and Cohen–Macaulay	$\operatorname{reg}(I)=d$8 has a linear resolution

The class of modules with initially linear syzygies, introduced through Gröbner theory, contains ideals with linear quotients. Its central consequence is that if such a module is generated in one degree, then its minimal resolution is linear; more generally, these modules are componentwise linear, and in crit-monotone cases their differentials admit Eliahou–Kervaire-type formulas (Sköldberg, 2011).

For monomial ideals, linear quotients are a stronger condition than linear resolution in general, but they become equivalent in several special settings. In three variables, every equigenerated monomial ideal with a linear resolution has linear quotients, and this is further equivalent to all powers having linear resolutions (Đào et al., 19 Sep 2025). Under graph-theoretic restrictions on the linear syzygy graph $\operatorname{reg}(I)=d$9, the same equivalence holds: if $M$0 is a tree or a cycle, then linear resolution, linear quotients, and variable-decomposability coincide (Manouchehri et al., 2018).

A poset-theoretic reformulation is given by the lcm-lattice $M$1. A monomial ideal $M$2 has a $M$3-linear resolution if and only if $M$4 is $M$5-degree graded and Cohen–Macaulay, and $M$6 has linear quotients if and only if $M$7 is $M$8-degree graded and CL-shellable (Varshavsky, 31 Jul 2025). In the squarefree case, this matches the classical Alexander-dual descriptions: $M$9 has a linear resolution exactly when $S$0 is pure and Cohen–Macaulay, while linear quotients correspond to shellability of $S$1 (Varshavsky, 31 Jul 2025).

This comparison also isolates a common misconception. Linear quotients always imply linear resolution for equigenerated monomial ideals, but the converse fails in general; the gap is measured combinatorially by the difference between Cohen–Macaulayness and CL-shellability of the lcm-lattice (Varshavsky, 31 Jul 2025).

3. Powers, products, and persistent linearity

A major theme is whether linearity survives under products and powers. For a family $S$2 of homogeneous ideals in a polynomial ring, “linear products” means that every finite product $S$3 with $S$4 has a linear resolution. Bruns and Conca show that this is equivalent to vanishing of the $S$5-th partial regularity of the multi-Rees algebra: $S$6 They analyze three families with linear products: polymatroidal ideals, ideals generated by linear forms, and northeast determinantal ideals $S$7 of maximal minors. In each case, products have linear resolutions, admit multiplicative-intersection-type decompositions, and the associated multi-Rees algebras are normal and Cohen–Macaulay; in the determinantal case they are also defined by Gröbner bases of quadrics (Bruns et al., 2016).

This perspective recovers the classical theorem that any product of ideals generated by linear forms in a polynomial ring has a linear resolution, and it places that theorem alongside broader patterns involving Gröbner bases, SAGBI deformations, and toric degenerations (Bruns et al., 2016). A related geometric instance occurs for line arrangements in $S$8: if $S$9 is a rank-three line arrangement, then for every $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$0, the ideal $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$1 generated by all $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$2-fold products of the $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$3 has a linear graded free resolution (Tohaneanu, 2019).

For powers, the behavior is subtler. Lexsegment ideals provide one stable class: a lexsegment ideal has a linear resolution if and only if it has linear quotients, and this is equivalent to saying that all powers have linear quotients and hence linear resolutions (Ene et al., 2010). Edge ideals furnish another stable class in higher powers: if $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$4 is gap-free and diamond-free, then

$$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$5

so every higher power has a linear minimal free resolution, even when $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$6 itself may have regularity $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$7 rather than $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$8 (Erey, 2017).

For squarefree monomial ideals generated in degree $$\cdots \to \bigoplus_j S(-j)^{\beta_{2,j}(M)} \to \bigoplus_j S(-j)^{\beta_{1,j}(M)} \to \bigoplus_j S(-j)^{\beta_{0,j}(M)} \to M \to 0.$$9, a sharp degree classification is now known. The following are equivalent: the class of squarefree ideals with $M$0-linear resolution is independent of the field, linear resolution coincides with linear powers, and $M$1. In these degrees, linear resolution is equivalent to linear quotients, and all powers have linear quotients; for $M$2, there exist fully supported squarefree ideals whose linear resolution depends on the characteristic, and also fully supported squarefree ideals with a linear resolution but with $M$3 not linearly resolved (Ficarra et al., 14 Aug 2025).

The same contrast appears in low dimension. In $M$4, equigenerated monomial ideals satisfy

$M$5

but this three-way equivalence is specific to three variables and is established through a combinatorial criterion involving the degree-$M$6 simplex and the absence of “bad configurations” (Đào et al., 19 Sep 2025).

4. Quotients, hypersurfaces, and singular ambient rings

Passing from polynomial rings to singular quotients changes the problem substantially. Over a polynomial ring, products of ideals of linear forms have finite linear resolutions. Over a hypersurface $M$7, resolutions are usually infinite, and linearity over $M$8 does not by itself control linearity over $M$9. The quadric case is exceptional. If $d$0, then every product of ideals generated by linear forms in $d$1 has a linear resolution over $d$2; equivalently, every quadric hypersurface is universally$d$3 Koszul (Conca et al., 2017).

The mechanism is Tor-linearity. For $d$4 a product of ideals generated by linear forms in $d$5, the mixed regularity satisfies the exact formula

$d$6

When $d$7 is quadratic, this yields Tor-linearity with respect to $d$8, and hence universally$d$9 Koszulness of $d$00 (Conca et al., 2017). The proof uses a flexible version of Derksen–Sidman approximation systems, replacing surjective approximations by $d$01-approximations that control both kernels and cokernels and remain stable under tensor products and Tor (Conca et al., 2017).

The theorem is specific to products of linear ideals. A counterexample shows that one cannot replace “product of ideals generated by linear forms” by “arbitrary ideal with a linear resolution over $d$02”: there exists an ideal $d$03 with $d$04 such that $d$05 does not have a linear resolution over $d$06 (Conca et al., 2017). This sharply separates ambient-ring effects from intrinsic linearity over the polynomial ring.

Related hypersurface rigidity appears in toric settings. For the edge ring $d$07 of a finite connected simple bipartite graph, if $d$08 has a $d$09-linear resolution with $d$10, then $d$11 is a hypersurface (Tsuchiya, 2019). The proof combines the description of the toric ideal by even cycles, the fact that a $d$12-linear resolution forces generators to come from cycles of length $d$13, and lower bounds on the Ehrhart degree of the edge polytope; the conclusion is that there can be exactly one such cycle, so the toric ideal is principal (Tsuchiya, 2019).

5. Combinatorial realizations

In the squarefree world, linear resolutions are closely tied to graphs, simplicial complexes, and hypergraphs. For quadratic squarefree ideals $d$14, linear resolution is equivalent to chordality of the complement graph $d$15; in the same degree, all powers have linear quotients whenever $d$16 is chordal (Ficarra et al., 14 Aug 2025). In degree $d$17, every squarefree ideal is a complementary edge ideal $d$18, and linear resolution is equivalent to the associated graph having exactly one nontrivial connected component; in that case every power has linear quotients (Ficarra et al., 14 Aug 2025).

The graph of first linear syzygies $d$19 gives another viewpoint. Its vertices are the minimal generators of a squarefree monomial ideal, and two generators are adjacent when they differ by one variable, equivalently when there is a first linear syzygy $d$20. If $d$21 is a tree or a cycle, then linear resolution, linear quotients, and variable-decomposability are equivalent. In the tree case, linearity is also equivalent to the connectivity of every induced graph $d$22, and in the cycle case a linear resolution forces a rigid support pattern in which each generator omits exactly two consecutive variables (Manouchehri et al., 2018).

Degree $d$23 squarefree ideals exhibit both rigidity and obstruction. For 3-uniform clutters, several operations preserve regularity of the circuit ideal of the complement: deleting a simplicial submaximal circuit, performing a flip, or applying a vertex expansion-contraction move all leave regularity unchanged. These moves generate large classes of ideals with 3-linear resolutions (Morales et al., 2012). At the same time, if a clutter comes from a triangulation of the $d$24-sphere, then the circuit ideal of the complement does not have a 3-linear resolution, while every proper subclutter does; triangulations of $d$25 therefore produce minimal obstructions to linearity (Morales et al., 2012).

Partial linearity also admits local combinatorial tests. For monomial ideals, condition $d$26 is determined by restrictions to subsets of at most $d$27 variables, and for squarefree cubic ideals linear presentation can be characterized by forbidding two explicit disconnected configurations on at most six variables (Dao et al., 2022). In the $d$28-primary almost-linear case $d$29, the missing degree-$d$30 monomials form disjoint degree-$d$31 shadows of pairwise $d$32-separated saturated monomials, a description that leads to Sierpiński-type families of ideals with few generators but controlled regularity (Dao et al., 2022).

6. Gorenstein-linear resolutions and universal constructions

Artinian Gorenstein algebras with linear resolutions form a particularly rigid class. Fix $d$33 and let $d$34 be a grade-$d$35 Gorenstein ideal generated in degree $d$36. If $d$37 has a linear resolution, then its socle degree is $d$38, its Hilbert function is symmetric, and the Macaulay inverse system is represented by an element $d$39 such that the map

$d$40

is an isomorphism (Khoury et al., 2013). For fixed $d$41 and $d$42, there exists a universal bigraded ring $d$43 and a complex $d$44 of free $d$45-modules with the property that every such $d$46 is obtained by specialization, and the specialized complex is its minimal homogeneous free resolution (Khoury et al., 2013).

The later explicit construction refines this substantially. The minimal homogeneous resolution $d$47 of a Gorenstein-linear Artinian algebra decomposes into Schur and Weyl modules associated to hook partitions, and its quotient by a distinguished linear form,

$d$48

depends only on the pair $d$49. This skeleton is the mapping cone of the zero map $d$50, where $d$51 is a Buchsbaum–Eisenbud resolution and $d$52 is its dual (Khoury et al., 2014). The full complex inherits a natural self-duality, and every nonzero element of $d$53 is a weak Lefschetz element (Khoury et al., 2014).

These results show that linear resolution can be either highly flexible or highly rigid, depending on context. For monomial ideals, the decisive structure may be an lcm-lattice, an Alexander dual complex, or a syzygy graph. For products and powers, the decisive structure may be a multi-Rees algebra or a Betti splitting. For hypersurfaces, mixed regularity and Tor-linearity control the passage from $d$54 to $d$55. For Artinian Gorenstein algebras, the entire minimal resolution is functorially encoded by the inverse system. The current boundary of the subject reflects that diversity: universally Koszul versus universally$d$56 Koszul beyond quadrics remains open in general (Conca et al., 2017); shellability versus CL-shellability of lcm-lattices is not yet fully matched to an algebraic property (Varshavsky, 31 Jul 2025); and for squarefree ideals in degrees $d$57, linear resolution is neither field-independent nor equivalent to linear powers (Ficarra et al., 14 Aug 2025).