Linear Quotients in Monomial Ideals

• Linear quotients are defined by ordering the generators of a monomial ideal so that each colon ideal is generated by linear forms (variables), simplifying homological computations.
• They imply a linear resolution for equigenerated ideals and connect directly with homological invariants like projective dimension and depth through explicit colon-ideal filtrations.
• Their characterization via shellability, chordality, and lcm-lattice grading provides key insights for applications in graph theory, path ideals, and combinatorial constructions.

Linear quotients are an order-theoretic property of ideals, most commonly studied for monomial ideals in a polynomial ring. If $I\subseteq S=K[x_1,\dots,x_n]$ is generated by monomials $m_1,\dots,m_r$, then $I$ has linear quotients if the generators can be ordered so that each colon ideal $(m_1,\dots,m_{i-1}):m_i$ is generated by linear forms; in the monomial setting this means generated by variables. For equigenerated ideals, the property is strictly stronger than linear resolution, but it is often more tractable because it reduces homological questions to the existence of a “good ordering” of generators and to explicit colon-ideal computations (Đào et al., 19 Sep 2025).

1. Definition and foundational criterion

The standard definition is order-dependent. A homogeneous ideal $I$ with generators $f_1,\dots,f_m$ has linear quotients if there exists an ordering such that, for every $i>1$, the colon ideal $\langle f_1,\dots,f_{i-1}\rangle:f_i$ is generated by linear forms (Liu et al., 2016). For monomial ideals, one usually writes the minimal generators as $\mathcal G(I)=\{m_1,\dots,m_r\}$, and the condition becomes that each $(m_1,\dots,m_{i-1}):m_i$ is generated by variables (Varshavsky, 31 Jul 2025).

This formulation is frequently recast in terms of a linear divisor: a monomial $m_1,\dots,m_r$0 is a linear divisor of an ideal $m_1,\dots,m_r$1 if $m_1,\dots,m_r$2 is generated by variables. An ordering is then a linear quotient ordering precisely when each successive generator is a linear divisor for the ideal generated by the previous ones (Dochtermann, 2018). The terminology emphasizes that linear quotients are not an intrinsic property of the unordered generating set; the existence of a suitable order is essential.

In graph-theoretic and fixed-degree settings, the criterion is especially concrete. For powers of edge ideals, for example, one repeatedly constructs an order $m_1,\dots,m_r$3 on minimal generators and verifies that each colon ideal $m_1,\dots,m_r$4 is variable-generated. The resulting proofs are often entirely combinatorial, with the main difficulty concentrated in designing the ordering rather than in computing a full resolution (Erey et al., 2024).

2. Homological consequences and resolution-theoretic structure

For ideals generated in a single degree, linear quotients imply a linear resolution. A direct proof uses the filtration $m_1,\dots,m_r$5, the short exact sequences

$m_1,\dots,m_r$6

and repeated applications of the Horseshoe Lemma. If all generators have degree $m_1,\dots,m_r$7 and $m_1,\dots,m_r$8 has linear quotients, then $m_1,\dots,m_r$9 has a $I$0-linear resolution, and if $I$1 denotes the number of linear forms in a minimal generating set of $I$2, with $I$3, then $I$4 (Liu et al., 2016).

In the equigenerated monomial case, the ordering criterion admits a useful reformulation. For an ordering $I$5, the following are equivalent: $I$6 has linear quotients with respect to that order; each initial ideal $I$7 has a linear resolution; and each $I$8 is linearly presented. This equivalence is one of the principal mechanisms by which explicit generator orders are converted into homological information (Đào et al., 19 Sep 2025).

The property also interacts with finer invariants. If $I$9 is a monomial ideal with linear quotients and $(m_1,\dots,m_{i-1}):m_i$0, then $(m_1,\dots,m_{i-1}):m_i$1; if $(m_1,\dots,m_{i-1}):m_i$2 is squarefree, then $(m_1,\dots,m_{i-1}):m_i$3 as well (Bordianu et al., 2021). In a different direction, when a monomial ideal has linear quotients with respect to an admissible order of increasing support-degree, it is componentwise support-linear, and this connects linear quotients with support-graded analogues of componentwise linearity (Shen, 2014).

These consequences explain why linear quotients are stronger than linear resolution but often preferred in explicit constructions: they provide not only existence of a linear strand but also a generator-by-generator filtration controlling projective dimension, depth-related invariants, and inductive resolution building.

3. Characterizations via shellability, chordality, and the lcm-lattice

One major theme in the theory is that linear quotients can be recognized through combinatorial or topological structures attached to the ideal. In Stanley–Reisner theory, the Alexander dual viewpoint is classical: for a simplicial complex $(m_1,\dots,m_{i-1}):m_i$4, $(m_1,\dots,m_{i-1}):m_i$5 has linear quotients if and only if $(m_1,\dots,m_{i-1}):m_i$6 is shellable (Varshavsky, 31 Jul 2025). In squarefree settings this is one reason linear quotients are tightly connected to shellability phenomena.

For edge ideals of complements of graphs, chordality is the central criterion. If $(m_1,\dots,m_{i-1}):m_i$7 is a graph and $(m_1,\dots,m_{i-1}):m_i$8 is the edge ideal of the complement graph, then $(m_1,\dots,m_{i-1}):m_i$9 has linear quotients if and only if $I$0 is chordal (Dochtermann, 2018). The same paper identifies the local combinatorial move underlying this equivalence: an edge $I$1 is exposed in $I$2 if and only if the monomial $I$3 is a linear divisor for $I$4. This turns sequences of exposed-edge removals into linear quotient orderings, and the graph-theoretic characterization of chordal graphs becomes an algebraic statement about successive colon ideals (Dochtermann, 2018).

A more global characterization uses the lcm-lattice. If $I$5 is generated in a single degree $I$6, then $I$7 has linear quotients if and only if its lcm-lattice $I$8 is $I$9-degree graded and CL-shellable; by contrast, $f_1,\dots,f_m$0 has a $f_1,\dots,f_m$1-linear resolution if and only if $f_1,\dots,f_m$2 is $f_1,\dots,f_m$3-degree graded and Cohen–Macaulay (Varshavsky, 31 Jul 2025). This sharply separates the two notions: the passage from linear resolution to linear quotients corresponds, at the lattice level, to strengthening Cohen–Macaulayness to CL-shellability. The same paper emphasizes that this lattice-theoretic strengthening is genuinely stronger than ordinary shellability, even though the Alexander dual viewpoint uses ordinary shellability (Varshavsky, 31 Jul 2025).

These characterizations show that linear quotients occupy a distinctive position between local divisibility data and global combinatorial structure. They are verified by colon ideals, reflected in shellings, and completely encoded by the graded and shellable structure of the lcm-lattice.

4. Coincidence and separation from linear resolutions

In general, linear quotients imply linear resolution, but the converse fails. Counterexamples are known, and Reisner’s example is cited as a classical small example of a monomial ideal with linear resolution but without linear quotients (Đào et al., 19 Sep 2025). Much of the subject therefore concerns identifying classes where the gap disappears.

Several such classes are now well understood. For quadratic monomial ideals, the theorem of Herzog, Hibi, and Zheng gives an equivalence among three conditions: the ideal has a linear resolution, it has linear quotients, and all of its powers have linear resolutions (Basser et al., 2024). For edge ideals this interacts with Fröberg’s theorem, so chordality or co-chordality criteria can be read as linear quotient criteria in degree $f_1,\dots,f_m$4.

In three variables, the equivalence becomes strikingly broad. If $f_1,\dots,f_m$5 is equigenerated, then linear resolution implies linear quotients; the proof uses a combinatorial characterization of linear resolution via connectedness of induced dual graphs together with the absence of bad configurations, and then constructs a “tree ordering” of generators that yields linear quotients (Đào et al., 19 Sep 2025). The same paper summarizes the resulting special-case collapse as

$f_1,\dots,f_m$6

for equigenerated monomial ideals in three variables (Đào et al., 19 Sep 2025).

Lexsegment ideals provide another class where the distinction disappears completely. For a lexsegment ideal $f_1,\dots,f_m$7, the following are equivalent: $f_1,\dots,f_m$8 has a linear resolution, $f_1,\dots,f_m$9 has linear quotients, all powers $i>1$0 have linear quotients, and all powers $i>1$1 have linear resolutions (Ene et al., 2010). The proofs divide into completely lexsegment and non-completely lexsegment cases, but the final outcome is a four-way equivalence.

These coincidence results are structurally important. They isolate settings in which linear quotients are not merely a strengthening of linear resolution but an equivalent and often more constructive reformulation of it.

5. Graph ideals, connected ideals, and path ideals

Graph-associated ideals supply a large family of linear quotient problems beyond edge ideals. For a graph $i>1$2 and $i>1$3, the $i>1$4-connected ideal

$i>1$5

generalizes edge ideals and path ideals: $i>1$6, while $i>1$7 agrees with the usual path ideal $i>1$8 (Ananthnarayan et al., 2024). For trees, there is an exact criterion: $i>1$9 has linear quotients if and only if $\langle f_1,\dots,f_{i-1}\rangle:f_i$0 has a linear resolution if and only if $\langle f_1,\dots,f_{i-1}\rangle:f_i$1 is $\langle f_1,\dots,f_{i-1}\rangle:f_i$2-gap-free (Ananthnarayan et al., 2024). For general graphs, a strong sufficient condition is also available: if $\langle f_1,\dots,f_{i-1}\rangle:f_i$3 is gap-free and $\langle f_1,\dots,f_{i-1}\rangle:f_i$4-claw-free, then $\langle f_1,\dots,f_{i-1}\rangle:f_i$5 has linear quotients (Ananthnarayan et al., 2024).

The $\langle f_1,\dots,f_{i-1}\rangle:f_i$6-path ideal of a tree admits an even finer classification. For a tree $\langle f_1,\dots,f_{i-1}\rangle:f_i$7 and $\langle f_1,\dots,f_{i-1}\rangle:f_i$8, $\langle f_1,\dots,f_{i-1}\rangle:f_i$9 has linear quotients if and only if it has linear resolution, and this occurs precisely under an induced-subgraph exclusion criterion involving $\mathcal G(I)=\{m_1,\dots,m_r\}$0 and trees $\mathcal G(I)=\{m_1,\dots,m_r\}$1, with the exceptional forbidden tree $\mathcal G(I)=\{m_1,\dots,m_r\}$2 when $\mathcal G(I)=\{m_1,\dots,m_r\}$3 (Chau et al., 6 Jun 2025). The proof reduces arbitrary trees satisfying the forbidden-structure condition to caterpillar trees via a trimming operation preserving $\mathcal G(I)=\{m_1,\dots,m_r\}$4, then orders the generators lexicographically and proves that the relevant colon ideals are generated by variables (Chau et al., 6 Jun 2025).

A further refinement is provided by homological linear quotients. If $\mathcal G(I)=\{m_1,\dots,m_r\}$5 denotes the $\mathcal G(I)=\{m_1,\dots,m_r\}$6-th homological shift ideal appearing in the minimal free resolution of a monomial ideal $\mathcal G(I)=\{m_1,\dots,m_r\}$7, then $\mathcal G(I)=\{m_1,\dots,m_r\}$8 has homological linear quotients if every $\mathcal G(I)=\{m_1,\dots,m_r\}$9 has linear quotients (Chau et al., 14 Mar 2025). For edge ideals $(m_1,\dots,m_{i-1}):m_i$0, co-chordality is necessary but not sufficient: the family $(m_1,\dots,m_{i-1}):m_i$1 consists of co-chordal graphs for which $(m_1,\dots,m_{i-1}):m_i$2 has the base-level behavior expected of edge ideals with linear resolution, but $(m_1,\dots,m_{i-1}):m_i$3 does not have linear quotients (Chau et al., 14 Mar 2025). The paper conjectures that homological linear quotients are characterized by being co-chordal and $(m_1,\dots,m_{i-1}):m_i$4-free, and proves the obstruction direction (Chau et al., 14 Mar 2025).

Taken together, these results show that graph-derived ideals support a hierarchy of linear quotient phenomena: exact equivalences for trees, sufficient conditions for broader graph classes, and higher-resolution variants where the property must persist across homological shift ideals.

6. Powers, persistence, and eventual behavior

The behavior of powers is one of the most active directions in the theory. For quadratic monomial ideals that already have linear quotients, there is an explicit constructive theorem: if $(m_1,\dots,m_{i-1}):m_i$5 admits a linear quotient ordering, then for every $(m_1,\dots,m_{i-1}):m_i$6 the minimal generators of $(m_1,\dots,m_{i-1}):m_i$7 can be ordered by reverse lexicographic order with respect to edge decompositions, after removing repeated representatives, and the resulting order is again a linear quotient ordering (Basser et al., 2024). This recovers the quadratic case of the Herzog–Hibi–Zheng theorem but adds a direct and usable ordering construction.

Anticycles demonstrate that powers may behave better than the original ideal. For the anticycle $(m_1,\dots,m_{i-1}):m_i$8, the edge ideal $(m_1,\dots,m_{i-1}):m_i$9 does not have linear quotients, but $m_1,\dots,m_r$00 has linear quotients for all $m_1,\dots,m_r$01 under an explicit order $m_1,\dots,m_r$02 (Basser et al., 2024). This resolves a question of Hoefel and Whieldon and gives the first class of gap-free graphs whose edge ideals fail linear quotients at level $m_1,\dots,m_r$03 but acquire them on all higher powers (Basser et al., 2024). Earlier work had already provided an explicit linear quotient ordering for all powers of the antipath and for the square of the edge ideal of the anticycle, thereby strengthening prior linear-resolution results by supplying actual colon-ideal orderings (Hoefel et al., 2011).

The asymptotic problem for gap-free graphs is formulated most sharply in terms of eventual linear quotients. Nevo and Peeva conjectured that if $m_1,\dots,m_r$04 is gap-free, then $m_1,\dots,m_r$05 has linear resolution for all sufficiently large $m_1,\dots,m_r$06. A stronger conjecture proposes that if $m_1,\dots,m_r$07 has linear quotients for some $m_1,\dots,m_r$08, then $m_1,\dots,m_r$09 has linear quotients for all $m_1,\dots,m_r$10 (Erey et al., 2024). The paper develops an “efficient ordering” of generators of successive powers and proves a finite-propagation theorem: if $m_1,\dots,m_r$11 has linear quotients and $m_1,\dots,m_r$12 have linear quotients with compatible orders for some $m_1,\dots,m_r$13, then $m_1,\dots,m_r$14 also has linear quotients. Combined with further propositions, this reduces the conjectural eventual behavior to finitely many initial checks in the compatible-order setting (Erey et al., 2024).

The same work constructs several infinite families of gap-free graphs for which all powers from $m_1,\dots,m_r$15 onward have linear quotients, despite containing induced subgraphs that previously appeared as obstructions in linear-resolution results. These include $m_1,\dots,m_r$16 itself, pentagon-based graphs with additional vertices or inner edges, CDCC graphs containing cricket, diamond, $m_1,\dots,m_r$17, and $m_1,\dots,m_r$18 as induced subgraphs, and clique-duplication families such as $m_1,\dots,m_r$19 (Erey et al., 2024). A key technical device is that duplicating a vertex in the specified non-adjacent sense preserves linear quotients, enabling inductive construction of larger examples (Erey et al., 2024).

The study of powers therefore reveals two complementary patterns: direct permanence for several structured classes, and genuinely new higher-power linear quotient behavior that is absent in degree $m_1,\dots,m_r$20.

7. Variants, refinements, and terminological extensions

Linear quotients admit several refinements internal to monomial-ideal theory. One is componentwise linear quotients: if $m_1,\dots,m_r$21 denotes the degree-$m_1,\dots,m_r$22 component ideal, then $m_1,\dots,m_r$23 has componentwise linear quotients when every $m_1,\dots,m_r$24 has linear quotients. Linear quotients always imply componentwise linear quotients, but the converse is open in general. Two classes where the converse does hold are known: componentwise polymatroidal ideals in $m_1,\dots,m_r$25, and componentwise polymatroidal ideals with the strong exchange property (Bandari et al., 2021).

Another refinement is recursive. The class of variable decomposable monomial ideals was introduced as an ideal-theoretic analogue of vertex decomposable simplicial complexes, and in the squarefree case the correspondence is exact. Variable decomposable ideals have linear quotients, and the construction is designed to capture skeleton-stability properties paralleling those of vertex decomposable complexes (Shen, 2014). The same paper shows that when linear quotients are realized by an admissible order of increasing support-degree, one obtains componentwise support-linearity, linking the order itself to a support-graded homological theory (Shen, 2014).

The phrase linear quotient also has unrelated meanings outside monomial-ideal theory. In transformation-group topology, quotients of spheres by linear torus actions, such as $m_1,\dots,m_r$26, are called linear quotients because the action is linear and orthogonal; their topology is described in terms of the Tutte polynomial of an associated matroid (Hughes et al., 2012). In positive-characteristic birational geometry, “linear $m_1,\dots,m_r$27-quotients” refers to quotient singularities arising from linear $m_1,\dots,m_r$28-actions on affine space, with singularity thresholds and stringy invariants controlled by the block data of the action (Posva et al., 7 Mar 2026). These usages are terminologically independent of linear quotients of ideals.

Within commutative algebra, however, the dominant meaning remains the colon-ideal ordering property. Its significance lies in the way it converts free-resolution questions into explicit divisibility problems, interfaces with shellability and lattice theory, and persists across a growing range of graph, path, lexsegment, and polymatroidal constructions.