04.08.02 · algebraic-geometry / differentials

Canonical sheaf

shipped3 tiersLean: partial

Anchor (Master): Hartshorne §II.8, §III.7; Vakil §21; Iitaka *Algebraic Geometry*

Intuition [Beginner]

The canonical sheaf $ω_{X}$ is the line bundle of "top-degree differential forms" on a smooth variety. On a smooth curve, it is the line bundle of holomorphic 1-forms; on a smooth surface, it is the line bundle of holomorphic 2-forms; and so on. Its sections are the natural "volume forms" of algebraic geometry.

Bernhard Riemann implicitly introduced the canonical divisor in his 1857 paper Theorie der Abelschen Functionen: he showed that on a Riemann surface of genus $g$ , the space of everywhere-holomorphic 1-forms has dimension exactly $g$ . This dimension equality — genus = dimension of holomorphic 1-forms — is one of the foundational identities of algebraic geometry. Grothendieck made the construction functorial in EGA: $ω_{X}$ is the determinant of the cotangent sheaf, $det Ω_{X}^{1} = ⋀^{n} Ω_{X}^{1}$ for smooth $X$ of dimension $n$ .

The canonical sheaf governs three deep structures: Serre duality (the duality on cohomology pairs against $ω_{X}$ ), the canonical embedding of curves of genus $\geq 3$ , and the Kodaira dimension of a variety (a birational invariant measuring how positive $ω_{X}$ is).

Visual [Beginner]

A smooth curve with everywhere-holomorphic 1-forms; the canonical sheaf assembles all such forms into a single line bundle.

Worked example [Beginner]

The canonical sheaf of $P_{k}^{1}$ is $ω_{P^{1}} = O_{P^{1}} (- 2)$ — the line bundle of degree $- 2$ . There are no global sections: $H^{0} (P^{1}; ω) = 0$ . Geometrically: there are no everywhere-holomorphic 1-forms on the Riemann sphere. This matches the formula $g = dim H^{0} (ω) = 0$ for the genus-0 case.

For the genus-1 elliptic curve $E$ (e.g., $y^{2} = x^{3} + a x + b$ ), the canonical sheaf is $ω_{E} ≅ O_{E}$ — the structure-sheaf line bundle of degree 0. There is exactly one global holomorphic 1-form (up to scalar): $ω = d x / y$ . Indeed, $g = 1$ matches the dimension of $H^{0} (ω) = k$ .

For a curve $C$ of genus $g \geq 2$ : $ω_{C}$ has degree $2 g - 2 > 0$ , and $dim H^{0} (ω_{C}) = g$ . The map $C \to P^{g - 1}$ given by these $g$ holomorphic 1-forms — the canonical map — is an embedding for non-hyperelliptic curves, the canonical embedding of classical algebraic geometry.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a smooth variety of dimension $n$ over a field $k$ . The canonical sheaf $ω_{X}$ is the determinant line bundle of the cotangent sheaf:

ω_{X} := det Ω_{X / k}^{1} = ⋀ n Ω_{X / k}^{1} .

Since $Ω_{X / k}^{1}$ is locally free of rank $n$ (smooth case), the top exterior power is locally free of rank 1 — a line bundle on $X$ . Its associated divisor class $K_{X} \in Cl (X) = Pic (X)$ is the canonical divisor class.

Local description. On a coordinate chart $U \subset X$ with local parameters $x_{1}, \dots, x_{n}$ trivialising $Ω^{1}$ , $ω_{X} ∣_{U}$ is the free $O_{U}$ -module generated by the single element

d x_{1} \land d x_{2} \land \dots \land d x_{n} .

A change of coordinates $(y_{1}, \dots, y_{n}) = ϕ (x_{1}, \dots, x_{n})$ transforms the generator by the Jacobian determinant: $d y_{1} \land \dots \land d y_{n} = det (\partial y_{i} / \partial x_{j}) \cdot d x_{1} \land \dots \land d x_{n}$ .

Sections. A global section of $ω_{X}$ is an everywhere-regular volume form: a top-degree differential form whose representation in local coordinates has no poles. The space of global sections $H^{0} (X; ω_{X})$ is finite-dimensional for $X$ proper.

Generalisation to non-smooth $X$ . For a variety $X$ that is not smooth but is Cohen-Macaulay (or even more generally), there is a dualising sheaf $ω_{X}$ extending the smooth case. For local complete intersections, $ω_{X}$ is still a line bundle (computed by an adjunction formula); for general $X$ , $ω_{X}$ may only be a coherent sheaf. The general construction is via Grothendieck's Residues and Duality (Hartshorne LNM 20, 1966).

Key formulas.

(C1) Smooth projective curves. $ω_{C} ≅ Ω_{C / k}^{1}$ (line bundle of holomorphic 1-forms), with $de g (ω_{C}) = 2 g - 2$ and $dim H^{0} (ω_{C}) = g$ — the genus formula.

(C2) Projective space. $ω_{P^{n}} = O_{P^{n}} (- (n + 1))$ , derived from the Euler sequence and computing the determinant.

(C3) Adjunction formula. For a smooth divisor $D \subset X$ on a smooth variety:

ω_{D} = (ω_{X} \otimes O_{X} (D)) ∣_{D} .

This is Adjunction: the canonical bundle of a hypersurface is the canonical bundle of the ambient variety twisted by the divisor class, restricted to the hypersurface.

(C4) Product formula. $ω_{X \times Y} = p_{1}^{*} ω_{X} \otimes p_{2}^{*} ω_{Y}$ .

(C5) Smooth complete intersection in $P^{n}$ . For $X = V (f_{1}, \dots, f_{r}) \subset P^{n}$ a smooth complete intersection of degrees $d_{1}, \dots, d_{r}$ :

ω_{X} = O_{X} (d_{1} + \dots + d_{r} - n - 1) .

In particular: a smooth quintic threefold in $P^{4}$ has $ω_{X} = O_{X} (5 - 5) = O_{X}$ — Calabi-Yau.

Kodaira dimension. The asymptotic growth of $dim H^{0} (X; ω_{X}^{\otimes m})$ as $m \to \infty$ defines the Kodaira dimension $κ (X) \in {- \infty, 0, 1, \dots, n}$ :

κ (X) = m \to \infty lim sup \frac{lo g dim H ^{0} ( ω _{X}^{\otimes m} )}{lo g m} .

This is a birational invariant. $κ (X) = - \infty$ corresponds to uniruled / Fano-like varieties; $κ (X) = 0$ to Calabi-Yau; $κ (X) = n$ to general type. Iitaka's classification of algebraic varieties is organised by Kodaira dimension.

Key theorem with proof [Intermediate+]

Theorem (canonical divisor degree on a smooth projective curve). Let $C$ be a smooth projective curve of genus $g$ over an algebraically closed field $k$ . Then the canonical sheaf $ω_{C}$ has degree $de g (ω_{C}) = 2 g - 2$ and $dim_{k} H^{0} (C; ω_{C}) = g$ .

Proof. Step 1 — relating $ω_{C}$ to $Ω_{C}^{1}$ . On a smooth curve (dimension 1), $Ω_{C / k}^{1}$ is locally free of rank 1, so $ω_{C} = det Ω_{C}^{1} = Ω_{C}^{1}$ — the canonical sheaf and the cotangent sheaf coincide.

Step 2 — defining the genus. The geometric genus of a smooth projective curve $C$ is defined as $g := dim_{k} H^{0} (C; Ω_{C}^{1}) = dim_{k} H^{0} (C; ω_{C})$ . This matches Riemann's 1857 definition: the dimension of the space of everywhere-holomorphic 1-forms.

Step 3 — Riemann-Roch with $D = K_{C}$ . Apply the Riemann-Roch theorem 04.04.01 to the divisor $K_{C}$ :

dim H^{0} (O (K_{C})) - dim H^{0} (O (K_{C} - K_{C})) = de g (K_{C}) - g + 1.

The left side: $dim H^{0} (O (K_{C})) = dim H^{0} (ω_{C}) = g$ by definition of genus. The second term: $dim H^{0} (O (K_{C} - K_{C})) = dim H^{0} (O_{C}) = 1$ (only constant functions on a connected projective curve).

So: $g - 1 = de g (K_{C}) - g + 1$ , giving $de g (K_{C}) = 2 g - 2$ .

Step 4 — alternative derivation via Riemann-Hurwitz. For the projective line $P^{1}$ (genus 0): $de g (ω_{P^{1}}) = 2 \cdot 0 - 2 = - 2$ , and indeed $ω_{P^{1}} = O (- 2)$ by the Euler sequence. For higher genus, the Riemann-Hurwitz formula applied to a degree- $n$ map $C \to P^{1}$ recovers $de g (ω_{C}) = 2 g - 2$ from the ramification data. $□$

The identity $de g (ω_{C}) = 2 g - 2$ is the foundational computation linking the topological invariant (genus = number of holes) to the algebraic invariant (canonical divisor degree). Both Riemann's original 1857 and Roch's 1865 papers exploit this identification.

Bridge. The construction here builds toward later units of the strand, where the same pattern is taken up at higher structure. The defining pattern appears again in those units in a sharpened form, where the local data is glued or quotiented. Putting these together, the foundational insight is that the data of this unit gives the structural signature that the rest of the strand reads off.

Exercises [Intermediate+]

Exercise 4 (medium, proof).

Prove the adjunction formula: for a smooth divisor $D \subset X$ on a smooth variety, $ω_{D} = (ω_{X} \otimes O (D)) ∣_{D}$ .

Hint

Use the conormal exact sequence and take determinants.

Answer

The conormal exact sequence for the smooth divisor $D ↪ X$ :

0 \to N_{D / X}^{\lor} \to Ω_{X}^{1} ∣_{D} \to Ω_{D}^{1} \to 0.

Here $N_{D / X}^{\lor} = O_{X} (- D) ∣_{D}$ — the conormal sheaf of a Cartier divisor is the dual of $O (D)$ , restricted.

Taking determinants of a short exact sequence: $det Ω_{X}^{1} ∣_{D} = det N_{D / X}^{\lor} \otimes det Ω_{D}^{1}$ . So:

ω_{X} ∣_{D} = N_{D / X}^{\lor} \otimes ω_{D} .

Tensoring by $O (D) ∣_{D} = N_{D / X}$ (the dual of $N_{D / X}^{\lor}$ ) gives:

(ω_{X} \otimes O (D)) ∣_{D} = ω_{D} .

This is the adjunction formula. $□$

The geometric content: on the divisor $D$ , the "missing" cotangent direction (the normal direction to $D$ in $X$ ) contributes $O (D) ∣_{D}$ to upgrade $ω_{X} ∣_{D}$ to $ω_{D}$ .

Exercise 6 (hard, proof).

State and prove the canonical embedding theorem: for a smooth projective non-hyperelliptic curve $C$ of genus $g \geq 3$ , the linear system $∣ K_{C} ∣$ defines an embedding $C ↪ P^{g - 1}$ .

Hint

Use Riemann-Roch and the very-ample criterion: $ℓ (K - P - Q) = ℓ (K) - 2 = g - 2$ for non-hyperelliptic generic $P, Q$ .

Answer

Canonical embedding theorem. Let $C$ be a smooth projective curve of genus $g \geq 2$ over an algebraically closed field. The linear system $∣ K_{C} ∣$ has dimension $g - 1$ , so it defines a morphism $ϕ_{K} : C \to P^{g - 1}$ . If $C$ is non-hyperelliptic (i.e., admits no degree-2 morphism to $P^{1}$ ), then $ϕ_{K}$ is a closed embedding, called the canonical embedding.

Proof. The morphism is defined by the $g$ global sections of $ω_{C} = O (K_{C})$ , i.e., by a basis of $H^{0} (C; ω_{C})$ , giving a map to $P^{g - 1}$ .

For $ϕ_{K}$ to be a closed embedding, we need:

Separates points: for any two points $P, Q \in C$ , there is a section of $ω_{C}$ vanishing at $P$ but not $Q$ . By Riemann-Roch:

ℓ (K - P) = ℓ (P) + de g (K) - de g (P) - g + 1 - g + 1 = \dots = g - 1.

So sections of $ω_{C}$ vanishing at $P$ form a hyperplane (codimension 1), and points $P \neq = Q$ correspond to distinct hyperplanes — separating points iff $ℓ (K - P - Q) = g - 2$ .

Separates tangent vectors: $ℓ (K - 2 P) = g - 2$ .

By Riemann-Roch, both conditions reduce to $ℓ (P + Q) \leq 1$ (resp. $ℓ (2 P) \leq 1$ ). On a non-hyperelliptic curve, $ℓ (P + Q) = 1$ for all (generic) $P, Q$ — the only function with poles only at $P, Q$ and degree $\leq 2$ is the constant. (If $ℓ (P + Q) = 2$ , then there's a non-constant function with $\leq$ simple poles at $P, Q$ , giving a degree-2 morphism to $P^{1}$ — contradicting non-hyperellipticity.)

So for non-hyperelliptic $C$ , $ϕ_{K}$ is a closed embedding. $□$

For genus 2, every curve is hyperelliptic, and $ϕ_{K} : C \to P^{1}$ is the 2-to-1 hyperelliptic cover. For genus 3, non-hyperelliptic curves embed canonically in $P^{2}$ as smooth quartics. For genus 4, non-hyperelliptic curves are intersections of a quadric and a cubic in $P^{3}$ . The canonical embedding gives the canonical model of the curve, foundational in algebraic curves theory.

Exercise 7 (hard, conceptual).

Explain the Kodaira dimension and how it stratifies smooth projective varieties.

Hint

Kodaira dimension measures asymptotic growth of $H^{0} (ω_{X}^{\otimes m})$ .

Answer

The Kodaira dimension $κ (X) \in {- \infty, 0, 1, \dots, dim X}$ measures the asymptotic positivity of $ω_{X}$ :

κ (X) = m \to \infty lim sup \frac{lo g dim H ^{0} ( X ; ω _{X}^{\otimes m} )}{lo g m} .

Equivalently, $κ (X) = dim_{C} ϕ_{∣ m K ∣} (X)$ for $m ≫ 0$ , where $ϕ_{∣ m K ∣}$ is the rational map defined by $∣ m K_{X} ∣$ . The Kodaira dimension is a birational invariant — invariant under blowups and birational transformations.

Stratification.

$κ (X) = - \infty$ : no positive power of $ω_{X}$ has global sections. Examples: $P^{n}$ , Fano varieties (where $- K_{X}$ is ample), uniruled varieties.
$κ (X) = 0$ : $∣ m K_{X} ∣$ has finite (often 1-dimensional) sections for all $m$ . Examples: abelian varieties, K3 surfaces, Calabi-Yau threefolds (quintic, etc.).
$κ (X) \in {1, \dots, n - 1}$ : intermediate Kodaira dimension. The Iitaka fibration $ϕ : X ⇢ Y$ has $dim Y = κ (X)$ , and general fibres have $κ = 0$ .
$κ (X) = n$ : $ω_{X}$ is big — $∣ m K_{X} ∣$ defines a birational embedding for $m ≫ 0$ . Varieties of general type.

Iitaka conjecture. For a fibration $f : X \to Y$ , the Kodaira dimensions satisfy $κ (X) \geq κ (Y) + κ (F)$ (general fibre). Proven in many cases; open in general.

Minimal model program. Mori's program (1980s) seeks to find a minimal birational model $X_{m i n}$ in each Kodaira-dimension stratum. For $κ = - \infty$ : terminate at a Mori fibration. For $κ \geq 0$ : terminate at a minimal model with $ω$ nef. The Birkar-Cascini-Hacon-McKernan theorem (2010) establishes existence of minimal models for varieties of general type ( $κ = n$ ).

The Kodaira dimension is the primary birational classifier of higher-dimensional varieties — the higher-dimensional analogue of the genus for curves. $□$

Lean formalization [Intermediate+]

lean_status: partial — Mathlib has Kähler differentials and exterior powers; the named canonical sheaf $ω_{X} = det Ω_{X}^{1}$ on a smooth scheme is partial.

[object Promise]

Advanced results [Master]

Riemann-Roch and Serre duality intertwined. The canonical sheaf $ω_{X}$ is the dualising sheaf in Serre duality: $H^{i} (X; F) ≅ H^{n - i} (X; F^{\lor} \otimes ω_{X})^{\lor}$ for locally free $F$ on smooth projective $X$ of dimension $n$ . The Riemann-Roch formula on a curve uses this directly: $dim H^{0} (L) - dim H^{0} (ω \otimes L^{- 1}) = de g L + 1 - g$ .

Iitaka's classification. Iitaka stratifies smooth projective varieties by Kodaira dimension into birational equivalence classes. For surfaces (dim 2), the Enriques-Kodaira classification (Enriques 1914, Kodaira 1960s) refines this: $κ = - \infty$ surfaces are ruled or rational; $κ = 0$ are K3, abelian surfaces, Enriques, hyperelliptic; $κ = 1$ are properly elliptic; $κ = 2$ are of general type.

Kodaira-Spencer and deformation theory. Infinitesimal deformations of a smooth proper variety $X$ are classified by $H^{1} (X; T_{X})$ where $T_{X} = H o m (Ω_{X}^{1}, O_{X})$ is the tangent sheaf. By Serre duality, this equals $H^{n - 1} (X; ω_{X} \otimes Ω_{X}^{1})^{\lor}$ , connecting deformation theory to canonical-sheaf cohomology. Kodaira-Spencer (1958) developed the analytic deformation theory; Schlessinger and Lichtenbaum-Schlessinger (1967) the algebraic.

Canonical models and minimal models. For a variety of general type $X$ , the canonical ring $R (X, ω_{X}) = ⨁_{m \geq 0} H^{0} (X; ω_{X}^{\otimes m})$ is finitely generated (Birkar-Cascini-Hacon-McKernan 2010, J. Amer. Math. Soc.). The canonical model $X_{can} = Proj R (X, ω_{X})$ is a (mildly singular) variety birational to $X$ , with $ω_{X_{can}}$ ample. This is the higher-dimensional analogue of the canonical embedding for curves.

Calabi-Yau and mirror symmetry. Compact Kähler manifolds with $ω_{X} ≅ O_{X}$ are Calabi-Yau. Yau's theorem (1978, Comm. Pure Appl. Math) — solving the Calabi conjecture — proves they admit Ricci-flat Kähler metrics. Mirror symmetry pairs Calabi-Yau threefolds in mirror pairs $(X, X^{\lor})$ with swapped Hodge numbers. The canonical sheaf governs the duality.

Anticanonical and Fano. A Fano variety is a smooth projective $X$ with $- K_{X}$ ample (i.e., $ω_{X}^{- 1}$ ample). Examples: $P^{n}$ , smooth quadrics, del Pezzo surfaces. Fano varieties have abundant rational curves and are central to birational geometry.

Synthesis. This construction generalises the pattern fixed in 04.08.01 (sheaf of differentials), with the symmetric data replaced by its skew or twisted analogue. Read in the opposite direction, the construction is dual to the metric story: complements and orthogonality are taken with respect to the bilinear datum of this unit, not a metric, and the resulting category of subobjects is the one the rest of the strand classifies. The central insight is that this datum identifies algebra with geometry: functions become vector fields, subspaces become quotients, and invariants become cohomology classes — and that identification is the engine driving every theorem downstream.

Full proof set [Master]

The genus formula $de g (ω_{C}) = 2 g - 2$ is proved in the formal-definition section. The adjunction formula is proved in Exercise 4. The canonical embedding theorem is proved in Exercise 6 with the Riemann-Roch separation criterion. Iitaka's classification and Birkar-Cascini-Hacon-McKernan are stated without proof — see Iitaka Algebraic Geometry GTM 76 1982; BCHM Existence of minimal models for varieties of log general type, J. Amer. Math. Soc. 23 (2010), 405–468.

Connections [Master]

Sheaf of differentials 04.08.01 — $ω_{X} = det Ω_{X}^{1}$ , the top exterior power.
Line bundle 04.05.03 — $ω_{X}$ is a line bundle on smooth $X$ ; one of the most important line bundles in algebraic geometry.
Picard group 04.05.02 — $K_{X} = [ω_{X}] \in Pic (X)$ , the canonical divisor class.
Riemann-Roch theorem for curves 04.04.01 — uses $ω_{C}$ via $dim H^{0} (ω \otimes L^{- 1})$ as the correction term.
Serre duality 04.08.03 — $ω_{X}$ is the dualising sheaf in Serre duality.
Hodge decomposition 04.09.01 — $H^{q} (X; ω_{X}) = H^{n, q} (X)$ in the Hodge decomposition.
Kodaira vanishing 04.09.02 — vanishing $H^{i} (X; ω_{X} \otimes L) = 0$ for $L$ ample, $i > 0$ .
Moduli of curves 04.10.01 — the canonical line bundle on $\overline{M_{g}}$ relates to tautological classes $κ, λ$ .

Historical & philosophical context [Master]

Bernhard Riemann's 1857 Theorie der Abelschen Functionen (J. reine angew. Math. 54) introduced the concept implicitly, computing the dimension of the space of "everywhere-finite integrals of the first kind" on a Riemann surface — what would become $H^{0} (C; ω_{C})$ . He proved this dimension equals the genus of the surface: a topological invariant equal to the maximum number of independent closed curves cuttable without disconnecting. This identity — dim of holomorphic 1-forms = genus — is the foundational identification of algebraic geometry, deeply connecting topology and complex analysis.

Riemann's term Geschlecht (genus) was new; his proof of the dimension formula used the Dirichlet principle — minimisation of an energy functional — which Weierstrass later showed required additional rigour (provided by Hilbert in 1900 via Hilbert space methods). Riemann's direct intuition: a surface of genus $g$ has $g$ independent "holes," and each hole supports an independent closed 1-form of period 1 around it.

Felix Klein and Henri Poincaré extended Riemann's analytic perspective in the 1880s. Klein's Über Riemann's Theorie der algebraischen Functionen und ihrer Integrale (1882) was the first systematic exposition of Riemann's work in modernised language. The algebraic / scheme-theoretic perspective came much later: Severi, Castelnuovo, and Enriques developed the Italian school's birational classification in the early 20th century, with the canonical class $K_{X}$ as a central tool.

Grothendieck systematised the construction in EGA: $ω_{X} = det Ω_{X}^{1}$ is the determinant line bundle of the cotangent sheaf, functorially. For non-smooth $X$ , the dualising sheaf extends the construction via Grothendieck's 1966 Residues and Duality (LNM 20, written by Hartshorne from Grothendieck's notes). The dualising sheaf is the unique coherent sheaf making Serre duality hold for arbitrary proper $X$ over a field — for Cohen-Macaulay $X$ , it is a line bundle; in general it is a complex (the dualising complex $ω_{X}^{∙}$ ).

Kunihiko Kodaira's 1953–1958 work established the canonical sheaf's role in classifying complex surfaces — the Enriques-Kodaira classification. Iitaka's 1971 paper on the Kodaira dimension introduced the asymptotic invariant $κ (X)$ as the primary birational classifier in higher dimensions. The minimal model program (Reid, Mori, Kollár, Birkar-Cascini-Hacon-McKernan) places $ω_{X}$ at the centre of higher-dimensional birational geometry: minimisation of $ω$ via blowdowns and flips.

The canonical sheaf is one of the most-studied objects in algebraic geometry. Riemann-Roch, Serre duality, Hodge theory, Calabi-Yau structures, mirror symmetry, the Mori program — all are organised by the behaviour of $ω_{X}$ . Riemann's 1857 insight that holomorphic 1-forms count the genus started a thread that continues into 21st-century mathematics.

Bibliography [Master]

Riemann, B., Theorie der Abelschen Functionen, J. reine angew. Math. 54 (1857), 115–155.
Roch, G., Über die Anzahl der willkürlichen Constanten in algebraischen Functionen, J. reine angew. Math. 64 (1865), 372–376.
Hartshorne, R., Algebraic Geometry, Springer 1977 — §II.8, §III.7.
Hartshorne, R., Residues and Duality, Springer LNM 20, 1966.
Vakil, R., The Rising Sea: Foundations of Algebraic Geometry — §21.
Iitaka, S., Algebraic Geometry: An Introduction to Birational Geometry of Algebraic Varieties, Springer GTM 76, 1982.
Mori, S., Threefolds whose canonical bundles are not numerically effective, Annals of Mathematics 116 (1982), 133–176.
Birkar, C., Cascini, P., Hacon, C., McKernan, J., Existence of minimal models for varieties of log general type, J. Amer. Math. Soc. 23 (2010), 405–468.
Yau, S.-T., On the Ricci curvature of a compact Kähler manifold and the complex Monge-Ampère equation I, Comm. Pure Appl. Math. 31 (1978), 339–411.
Kodaira, K., On compact analytic surfaces I–III, Annals of Math 71 (1960), 77 (1963), 78 (1963).