Castelnuovo–Mumford Regularity¶

Locate the least projective twist whose diagonal higher-cohomology vanishings persist, yielding one integer bound on global generation, Hilbert-function stabilization, and graded syzygy degrees.

Version: v3 · 2026-09-06 · History
Domain-specific #: 1439
Origin domain: algebraic geometry
Subdomain: projective sheaf cohomology and graded resolutions
Aliases: Castelnuovo-Mumford regularity, CM regularity

Core Idea¶

Castelnuovo–Mumford regularity is an integer that marks when the higher-cohomology and syzygy behavior of a projective object has entered a controlled linear range. Fix a field k, projective space P^n_k, its twisting sheaf O(1), and a coherent sheaf F. The sheaf is m-regular when.

H^i(P^n_k, F(m-i)) = 0 for every i=1,...,n.

Its regularity is the least such m when that minimum exists; the zero sheaf is commonly assigned -infinity. This diagonal pattern—not merely eventual vanishing—is the identity. The Stacks Project gives the definition and proves that m-regularity persists upward: an m-regular sheaf is (m+1)-regular.^[1] Mumford's foundational treatment introduced the technique in the construction-oriented setting of projective algebraic geometry.^[2]

One integer then controls several phenomena. If F is m-regular, F(m) is globally generated, multiplication of sections by linear forms is surjective in the stable range, and higher cohomology vanishes for sufficiently positive twists with an explicit threshold derived from m.^[1] For d>=m, all H^i(F(d)) with i>0 vanish, so.

dim_k H^0(P^n_k,F(d)) = chi(F(d)) = P_F(d),

where P_F is the Hilbert polynomial. Regularity therefore supplies a sufficient onset bound for the sheaf Hilbert function to agree with the polynomial; it need not be the sharp onset of every consequence.

There is an algebraic face of the same mechanism. Let S=k[x_0,...,x_n] with the standard grading and let a finitely generated graded S-module M have minimal graded free resolution.

... -> F_i -> ... -> F_0 -> M -> 0, with F_i = direct_sum_j S(-b_(i,j)).

Then.

reg(M) = max_(i,j) { b_(i,j) - i }.

Thus the generators in homological position i occur in degrees at most reg(M)+i. Equivalently, sufficiently high truncations are generated in their initial degree and have linear free resolutions. Eisenbud develops the cohomological, local-cohomological, and resolution formulations as one theory of syzygies; the official Macaulay2 documentation implements the shift-minus-homological-degree convention.^[3]^[4]

The sheaf, ideal, quotient, and module values must be named. For a closed subscheme X of projective space, “regularity of X” often means the regularity of its ideal sheaf I_X or saturated homogeneous ideal I_X. The coordinate ring S/I_X commonly has regularity one less for a nonzero proper ideal. A statement such as “the regularity is two” is incomplete until it says which object and grading convention produced the two.

Structural Signature¶

Sig role-phrases:

the projective frame — a projective embedding or chosen ample twisting object, ordinarily P^n_k with O(1), relative to which degrees and twists are measured
the target object — a coherent sheaf, ideal sheaf, saturated homogeneous ideal, finitely generated graded module, coordinate ring, or explicitly stated extension of the classical setting
the integer twist parameter — m, with the convention for F(t) or M(t) fixed before comparisons are made
the diagonal vanishing test — H^i(F(m-i))=0 for every positive cohomological degree in the sheaf formulation
the graded-resolution test — b_(i,j)-i<=m for every summand S(-b_(i,j)) in a minimal graded free resolution of the module
the least-threshold rule — regularity is the minimum qualifying integer, not one convenient upper bound or one checked twist
the persistence theorem — once m qualifies, every larger integer qualifies, turning isolated vanishings into an upward-closed stable range
the consequence package — global generation, surjective multiplication, generator/syzygy degree bounds, linear truncations, and Hilbert-polynomial agreement in the appropriate range
the convention bridge — the declared correspondence and possible offsets among F, I_X, S/I_X, sheafification, saturation, and finite-length irrelevant-ideal torsion

The locked recognition test requires a projective/graded frame, one precisely typed target object, and either the complete diagonal cohomology-vanishing test or an equivalent standard graded resolution/local-cohomology test. The reported value must be least, conventions must be explicit, and at least one consequence or diagnostic must depend on that value. A degree bound called “regularity” without this bridge is not Castelnuovo–Mumford regularity.

Diagnostics follow the roles. A one-unit disagreement often means one computation used I_X and another S/I_X. A larger discrepancy after sheafification points to saturation or irrelevant-ideal torsion. A negative answer is not automatically erroneous: twisting shifts regularity. A result that changes under a projective re-embedding may be correct because the chosen O(1) changed. A Betti-table computation from a nonminimal resolution gives only an upper bound unless cancellations are removed.

Interventions are specific: state the ring, grading, embedding, and object; saturate when the intended object is a projective subscheme; compute a minimal graded resolution or local cohomology; read every nonzero Betti shift rather than generator degrees alone; cross-check one convention through the ideal–quotient exact sequence; test the diagonal vanishings for a sheaf calculation; and distinguish a certified upper bound from an exact minimum.

What It Is Not¶

Not regularity of a ring or scheme in the nonsingularity sense. A singular projective scheme has Castelnuovo–Mumford regularity, and a smooth one can have large regularity.
Not a regular sequence. Regular sequences concern non-zero-divisor behavior; they may simplify a resolution but are not the same invariant.
Not statistical or optimization regularization. No penalty parameter is added to a fitting objective.
Not projective dimension. Projective dimension records resolution length; regularity records the largest degree shift minus homological position.
Not Krull dimension. Dimension counts geometric or algebraic degrees of freedom, not the twist at which cohomology and syzygies stabilize.
Not the degree of the generators alone. Later syzygies can force a larger regularity than every initial generator degree.
Not a Hilbert polynomial. The polynomial describes eventual graded dimensions; regularity bounds an onset and syzygy complexity but does not encode all polynomial coefficients.
Not a complete Betti table. Taking max(b_(i,j)-i) discards the number and distribution of generators and syzygies below the extremal diagonal.
Not intrinsically embedding-free. Classical regularity is relative to the chosen projective embedding or polarization.
Not automatically nonnegative. For example, reg(O(d))=-d on projective space; the zero-object convention may be -infinity.
Not a Riemann–Roch theorem. Riemann–Roch computes Euler characteristics from geometric classes; regularity supplies vanishing thresholds that can turn an Euler characteristic into h^0.

Scope of Application¶

Coherent sheaves on projective space. The original cohomological definition supplies vanishing and global-generation thresholds.^[1]
Projective subschemes and ideal sheaves. Regularity bounds degrees needed to generate equations and the degrees of their successive relations.
Finitely generated graded modules. Minimal resolutions, Betti tables, local cohomology, and linear truncations give computable equivalent formulations in the standard graded setting.^[5]
Hilbert and Quot constructions. Uniform regularity bounds for quotients with fixed Hilbert polynomial convert sheaf families into finite-dimensional section data; the Stacks Project proves the relevant boundedness step.^[1]
Computational algebraic geometry. Regularity bounds are tied to Gröbner-basis and syzygy computation; Bayer and Mumford explicitly study regularity as a complexity measure for effective algebraic geometry.^[6]
Projective curves, points, hypersurfaces, and arrangements. The invariant packages equation degree, higher relations, postulation, and embedding data into a common comparison scale.
Operations on sheaves and modules. Twists, direct sums, exact sequences, hyperplane sections, tensor products under hypotheses, powers of ideals, and changes of field motivate bounds and diagnostics.
Named extensions. Multigraded, toric, noncommutative, relative, and complex-valued variants preserve part of the role map but alter the regularity region or the equivalence theorems. They require their own definition and are not silently identical to the classical integer invariant.

Classical scope ends without a declared standard grading/projective polarization, a coherent or finitely generated target, and the corresponding cohomological or resolution theory.

Clarity¶

Write reg(F), reg(I_X), or reg(S/I_X), never an untyped reg(X) when an offset matters. Many texts define the regularity of a subscheme by its ideal sheaf. Software may accept an ideal and return the ideal module's regularity, while a resolution command applied to the quotient returns the coordinate ring's value. Macaulay2 documents this distinction explicitly.^[4]

Fix the shift convention. Here S(-b) has its generator in degree b, so a summand in homological degree i contributes b-i. Under this convention, reg(M(a))=reg(M)-a; an opposite notation for shifts can reverse the displayed sign while describing the same module.

“F is m-regular” means m is an admissible bound, not necessarily reg(F)=m. Since regularity persists, a sheaf of exact regularity two is also 3-regular, 4-regular, and so on. Exactness requires both proof of 2-regularity and failure of 1-regularity.

The Hilbert statement also needs type discipline. For a sheaf, m-regularity implies higher-cohomology vanishing at twists d>=m, hence h^0(F(d))=P_F(d) there. For a raw graded module, finite-length pieces supported at the irrelevant ideal can change low-degree Hilbert data without changing the associated sheaf. Saturation and local cohomology determine whether the sheaf and module statement being compared are actually equivalent.

Manages Complexity¶

A coherent sheaf presents a triangular array of groups H^i(F(d)) over cohomological degree and twist. A minimal resolution presents a second array of Betti numbers over homological position and internal degree. Regularity draws one diagonal boundary through either array: above that boundary the required higher cohomology vanishes, and every free summand lies on or below the corresponding degree-shift line.

This compression makes downstream reasoning finite. Instead of checking global generation at every twist, one proves one regularity bound and invokes persistence. Instead of listing all generator and syzygy degrees, one number certifies b_(i,j)<=m+i. Instead of waiting vaguely for a Hilbert function to become polynomial, one obtains a sufficient twist.

The number is particularly valuable in families. A uniform bound says all members can be represented using sections in one degree, supporting finite-dimensional parameter-space constructions. In computation, a bound limits how far graded search, truncation, or Gröbner machinery may need to proceed. Bayer and Mumford stress the link between regularity bounds and computational complexity.^[6]

Compression has a cost. Two modules with the same regularity may have entirely different Betti tables, depth, dimension, singularities, or Hilbert polynomials. The invariant manages the latest obstructing diagonal, not the entire algebraic object.

Abstract Reasoning¶

Cohomology audit. For each i>0, compute or prove vanishing of H^i(F(m-i)). To prove exact regularity m, also exhibit a nonzero group on the preceding diagonal or use an equivalent lower-bound witness.

Betti-diagonal audit. From a minimal resolution, record every degree b_(i,j) in homological position i, subtract i, and take the maximum. A nonminimal complex can contain cancellable equal shifts and must not be treated as exact evidence without minimization.

Twist audit. Replacing F by F(a) translates the cohomology table, giving reg(F(a))=reg(F)-a. Compare values only after aligning polarizations and twist notation.

Ideal–quotient audit. Use 0 -> I -> S -> S/I -> 0. For a nonzero proper homogeneous ideal in the usual setting, the ideal-module value is typically reg(I)=reg(S/I)+1. Verify hypotheses and object types rather than applying the offset to the zero or unit ideal.

Exact-sequence bounds. For 0 -> A -> B -> C -> 0, long exact cohomology or resolution arguments give the standard inequalities reg(A)<=max(reg(B),reg(C)+1), reg(B)<=max(reg(A),reg(C)), and reg(C)<=max(reg(B),reg(A)-1).^[3] Use them to bound a difficult term from two easier ones, then seek a lower-bound witness if equality matters.

Sheafification audit. Remove or account for finite-length irrelevant-ideal torsion and compare a module with its saturation or section module. Sheafification can erase graded information, so equal associated sheaves do not force equal raw-module regularity.

Embedding audit. If X is re-embedded by another very ample line bundle, rebuild the graded ring and twist scale. A changed result can express changed polarization rather than faulty computation.

Knowledge Transfer¶

The invariant transfers literally among coherent sheaves, saturated ideals, coordinate modules, and minimal free resolutions only through declared equivalence conditions. The role map provides the bridge: identify the projective frame, target object, diagonal vanishings, resolution shifts, least threshold, and convention offset. This lets a geometric vanishing theorem become an algebraic syzygy bound or a Betti-table computation become a global-generation statement.

The workflow also transfers within projective algebraic geometry. For a hypersurface, the defining degree determines the simple resolution. For a curve, later syzygies may set the maximum. For a bounded family, one seeks a uniform rather than object-specific value. For a computation, one checks minimality and saturation before interpreting the number geometrically.

Multigraded and noncommutative theories transfer the organizing idea but not always the integer. A multigraded theory may produce a region of regular degrees rather than one least m; a different ample sequence changes which cohomology groups must vanish. These are recognized extensions only after restating persistence and the consequence package.

Outside algebraic geometry and graded commutative algebra, “the onset of a stable range” is only a general threshold or complexity analogy. Without sheaf twists, cohomological degrees, and syzygy shifts, it is not Castelnuovo–Mumford regularity.

Examples¶

Canonical¶

A plane conic and the ideal-versus-quotient offset. Let S=k[x_0,x_1,x_2], let q be a nonzero homogeneous quadratic, and let C=V(q) in P^2. The homogeneous ideal I_C=(q) is the free graded module S(-2), so its minimal resolution has a degree-two generator in homological degree zero and reg(I_C)=2. The quotient has resolution

0 -> S(-2) -> S -> S/I_C -> 0,

so reg(S/I_C)=max(0,2-1)=1.

Sheafifying gives the ideal sheaf I_C=O_P2(-2). The Stacks Project calculation says O(d) is m-regular exactly when d>=-m, hence reg(I_C)=2.^[1] At m=2, I_C(2)=O is globally generated. At m=1, the diagonal test fails because H^2(P^2,O(-3)) is nonzero. Thus two is the least value, not merely an upper bound.

Mapped back: P^2 with O(1) is the projective frame; the ideal sheaf, graded ideal I_C, and quotient S/I_C are three explicitly distinguished target objects; m is the twist parameter; the line-bundle cohomology supplies the diagonal vanishing test and lower-bound witness; the two minimal resolutions supply the graded-resolution test; upward persistence supplies the least-threshold rule; global generation is part of the consequence package; and the values two versus one instantiate the convention bridge.

Applied / In Practice¶

Reading the twisted cubic from a Betti table. In S=k[a,b,c,d], the twisted cubic has saturated ideal

I=(b^2-ac, bc-ad, c^2-bd).

Its minimal ideal resolution is

0 -> S(-3)^2 -> S(-2)^3 -> I -> 0.

Therefore reg(I)=max(2,3-1)=2. Adding S in homological degree zero gives the coordinate-ring resolution and reg(S/I)=max(0,2-1,3-2)=1. The official Macaulay2 documentation uses this ideal to demonstrate the same quotient value one and ideal value two.^[4]

Because the ideal is saturated, the ideal sheaf is 2-regular in the standard projective embedding. Its degree-two twist is globally generated, matching the three quadratic generators; the two cubic first syzygies lie exactly on the allowed m+1=3 boundary. The coordinate ring's regularity one likewise bounds the onset of its stable graded behavior. The example shows why reading only “quadratic generators” is insufficient in general: the full resolution must confirm that no later shift exceeds the regularity diagonal.

Mapped back: the standard embedding of the twisted cubic is the projective frame; I and S/I are the distinguished target objects; the nonzero shifts (2,0) and (3,1) implement the graded-resolution test; their maximum establishes the least-threshold rule; ideal saturation supports the sheaf bridge; quadrics and cubic syzygies instantiate the consequence package; and the two-versus-one output is audited by the convention bridge.

Structural Tensions¶

T1: One-number compression vs. Betti-table information. Regularity preserves the extremal diagonal but forgets how many generators and syzygies occur elsewhere. Diagnostic: compare full Betti tables whenever multiplicity, depth, or resolution shape matters rather than ranking objects by regularity alone.
T2: Cohomological definition vs. computational resolution. Sheaf vanishings express geometry while minimal resolutions support calculation; the equivalence needs a correct graded/sheaf bridge. Diagnostic: cross-check one example through both diagonal cohomology and b-i before trusting a pipeline.
T3: Sheaf invariance vs. module torsion. Sheafification discards irrelevant-ideal torsion that can affect raw-module regularity. Diagnostic: inspect H^0 local cohomology and saturation when identical sheaves yield different module values.
T4: Ideal convention vs. quotient convention. Both values are called the regularity of a projective scheme, often differing by one. Diagnostic: require every reported number to carry an object label and verify it through 0 -> I -> S -> S/I -> 0.
T5: Invariant language vs. embedding dependence. The integer is stable under coordinate changes in a fixed standard frame but can change under re-embedding. Diagnostic: record the chosen O(1) or homogeneous coordinate ring with every comparison.
T6: Guaranteed onset vs. sharp onset. Regularity gives sufficient global-generation and Hilbert-function thresholds, but one consequence may stabilize earlier. Diagnostic: distinguish “by degree m” from “first occurs at degree m” and seek separate lower-bound witnesses.
T7: Effective bound vs. computational growth. A regularity bound can make an algorithm finite while itself being large or expensive to determine. Diagnostic: separate theoretical worst-case bounds, computed exact values, and empirical stopping degrees.
T8: Autonomy vs. reduction. Castelnuovo–Mumford regularity is autonomously useful because the diagonal vanishing, minimal threshold, Betti shifts, persistence, and convention bridge form one theorem-backed invariant. Reduction to complexity preserves only scalar intricacy and loses the projective grading and every validity condition. Diagnostic: if twist/cohomology or an equivalent graded-resolution structure is absent, route the claim to a broader complexity or threshold abstraction.

Structural–Framed Character¶

Castelnuovo–Mumford Regularity is structural-leaning. Once the field, projective frame, target object, grading, and shift convention are fixed, the admissible integers and their minimum are mathematical facts. Proof, cohomology, or a minimal resolution decides the result.

Evaluative weight: none. A lower value may be computationally convenient, but desirability does not determine whether an object is m-regular.

Human-practice-bound: none. Mathematicians choose which object to study, but the invariant's defining vanishings and resolution shifts do not depend on institutional roles or practical judgment.

Institutional origin: none. Textual conventions determine notation and whether an author says reg(X) for an ideal or quotient, but once the object is typed, no authority assigns its value.

Vocabulary travels: partly. “Regularity,” “twist,” “coherent sheaf,” “syzygy,” “Betti table,” and “Hilbert polynomial” are specialist. The least-stable-threshold skeleton travels, but the full identity does not.

Import vs. recognize: recognize in its home domain. A valid projective/graded object displays the defining pattern whether or not a practitioner computes it. Calling a nonalgebraic stabilization point “Castelnuovo–Mumford regularity” would import the frame without its semantics.

The portable skeleton is: arrange obstructions by level and shift, find an upward-closed diagonal beyond which all disappear, and use the least qualifying diagonal as a complexity bound. Its character: a formal projective complexity invariant whose specialist cohomology–syzygy equivalence is the substance, not decorative terminology.

Structural Core vs. Domain Accent¶

Skeletal structural core. A family of obstructions is indexed in two directions; a moving diagonal clears them; admissibility persists; and the least clear diagonal compresses the stable range into one integer.

Domain-bound accent. Coherent sheaves on projective space, O(1) twists, higher sheaf cohomology, Hilbert polynomials, standard graded polynomial rings, minimal free resolutions, Betti shifts, saturation, and ideal–quotient offsets determine the actual invariant. Removing them eliminates the tests that distinguish a correct regularity value from an arbitrary bound.

Why it is not a prime. The generic “least stable threshold” recurs elsewhere, but not with the same diagnostics and interventions. Cross-domain reduction lands in Complexity or a threshold-like schema and loses cohomological and graded validity. The identity therefore recurs broadly inside algebraic geometry and commutative algebra, not literally across three unrelated substrates.

prime:complexity — instantiates. Regularity is a strict algebraic-geometric complexity measure: it compresses the latest degrees of generators and syzygies and bounds effective computation. This is the proposed minimal direct parent; Bayer and Mumford explicitly connect regularity with computational complexity.^[6]
prime:dimension — related comparison. Projective dimension is resolution length and Krull dimension describes the underlying algebraic object. Neither subsumes regularity's degree-minus-position invariant.
domain_specific:homological_stability — related stable-range pattern. Homological Stability concerns maps in a sequence of groups or spaces becoming isomorphisms at fixed homological degree. Regularity concerns twist-indexed vanishings for one sheaf/module and is not a variant of that node.
domain_specific:riemann_roch_type_theorem — related consequence bridge. Riemann–Roch-type results calculate Euler characteristics; regularity supplies vanishing conditions under which Euler characteristic equals the dimension of global sections. Neither is a type of the other.
prime:closure — weakly related persistence. Admissible regularity values form an upward-closed set, but Closure does not capture the cohomological object or the least threshold.

Only prime:complexity is proposed as a direct DAC parent. The others are contrastive or mechanism-level neighbors, avoiding semantically false multiple inheritance.

Relationships to Other Abstractions¶

Current abstraction Castelnuovo–Mumford Regularity Domain-specific

Parents (1) — more general patterns this builds on

Castelnuovo–Mumford Regularity is a kind of Complexity Prime

prime:complexity — instantiates. Regularity is a strict algebraic-geometric complexity measure: it compresses the latest degrees of generators and syzygies and bounds effective computation.

Hierarchy path (1) — routes to 1 parentless root

Castelnuovo–Mumford Regularity → Complexity

Neighborhood in Abstraction Space¶

Castelnuovo–Mumford Regularity sits in a sparse region of the domain-specific corpus (74^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Algebraic Geometry & Bundle Structure (14 abstractions)

Nearest neighbors

Euler sequence — 0.88
Linear fractional transformation — 0.84
Structure Theorem for Finitely Generated Modules over a Principal Ideal Domain — 0.83
Ringed Space — 0.83
Algebraic stack — 0.82

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Regular local ring / regular scheme. These encode nonsingularity through local dimension and generators. Tell: Castelnuovo–Mumford regularity is a projective twist/syzygy integer and can be computed for singular schemes.
Regular sequence. This is a sequence of non-zero-divisors. Tell: ask whether the object is a sequence or an integer read from cohomology/resolution shifts.
Regularization. Optimization regularization penalizes model complexity. Tell: it has a tunable penalty coefficient, not H^i(F(m-i)) vanishings.
Projective dimension. Both are read from a free resolution. Tell: projective dimension is the last nonzero homological position; regularity is the maximum internal degree minus position.
Hilbert polynomial. Both govern eventual graded behavior. Tell: the polynomial is a function of degree; regularity is an integer bound on the onset and syzygies.
Homological Stability. Both use “stable range.” Tell: Homological Stability asks when structure maps become isomorphisms across a growing sequence; regularity asks when twisted higher cohomology vanishes.
Riemann–Roch-Type Theorem. Both relate cohomology and numerical data. Tell: Riemann–Roch computes an Euler characteristic; regularity does not compute characteristic classes.
Ideal regularity vs. coordinate-ring regularity. These are linked but not interchangeable. Tell: an ideal input ordinarily returns a value one greater than its proper quotient's value in the standard nonzero setting.^[4]
Multigraded regularity. This is a genuine extension, often a region rather than one integer. Tell: inspect the grading group and whether there is a unique least twist.

References¶

[1] The Stacks Project Authors, Section 33.35, Coherent Sheaves on Projective Space, especially Definition 33.35.7 and Lemmas 33.35.10–12 and 33.35.18, Tag 089X. registry ↩a ↩b ↩c ↩d ↩e

[2] David Mumford, Lectures on Curves on an Algebraic Surface, Annals of Mathematics Studies 59, Princeton University Press (1966), publisher DOI record. registry ↩

[3] David Eisenbud, The Geometry of Syzygies: A Second Course in Algebraic Geometry and Commutative Algebra, Graduate Texts in Mathematics 229, chapter 4, Springer (2005), publisher record. registry ↩a ↩b

[4] The Macaulay2 Team, regularity — compute the Castelnuovo–Mumford regularity, official documentation, Macaulay2 1.25 documentation. registry ↩a ↩b ↩c ↩d

[5] David Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Texts in Mathematics 150, Section 20.5, Springer (1995), publisher record. registry ↩

[6] Dave Bayer and David Mumford, “What Can Be Computed in Algebraic Geometry?” in Computational Algebraic Geometry and Commutative Algebra (1993), author preprint, arXiv:alg-geom/9304003. registry ↩a ↩b ↩c