Divisors, Line Bundles, and the Picard Group

⚡ Key Takeaways

One-line statement: Divisors are formal sums of codimension-one subvarieties; line bundles are locally free sheaves of rank one; on a sufficiently nice variety these two notions coincide, and the resulting isomorphism classes form an abelian group — the Picard group $\operatorname{Pic}(X)$ .

Key result in words: On an integral, separated, Noetherian, locally factorial scheme, the divisor class group $\operatorname{Cl}(X)$ is isomorphic to the Picard group $\operatorname{Pic}(X) = H^1(X,\mathcal{O}_X^\times)$ . This dictionary lets one translate geometry (subvarieties, intersection) into algebra (cohomology, group structure).

Where it applies: Algebraic geometry, complex geometry, number theory (Arakelov theory, arithmetic of curves), and string theory (D-brane charges, moduli of bundles).

Who/when: Riemann and Roch (1850s–60s) for curves; Picard, Severi, the Italian school (1880s–1920s); Weil, Grothendieck, Serre (1940s–60s) for the modern scheme-theoretic formulation.

💡 The Core Insight

"Codimension-one data is rank-one data." A subvariety of codimension one is locally the zero set of one equation; a line bundle is locally trivialized by one nowhere-vanishing section. Both are governed by the transition data of how local generators glue, which lives in $\mathcal{O}_X^\times$ . The Picard group is the universal repository of this gluing, and it is simultaneously:

a quotient of divisors by principal ones (geometry),
the group $H^1(X,\mathcal{O}_X^\times)$ of $\mathcal{O}_X^\times$ -torsors (cohomology),
the moduli of invertible sheaves under $\otimes$ (sheaf theory).

DefinitionWeil Divisor and Divisor Class Group

Let $X$ be a Noetherian, integral, separated scheme that is regular in codimension one (i.e. $\mathcal{O}_{X,\xi}$ is a DVR for every codimension-one point $\xi$ ). A prime divisor is an integral closed subscheme of codimension one. A Weil divisor is an element of the free abelian group $\operatorname{Div}(X)$ on prime divisors:

1

\quad D = \sum_{Y} n_Y\,[Y], \qquad n_Y \in \mathbb{Z},\ \ n_Y = 0 \text{ for all but finitely many } Y.

Derivation hint: the free generation is the definition; finiteness is required so that $D$ is a genuine element of a direct sum $\bigoplus_Y \mathbb{Z}$ . For each prime divisor $Y$ the local ring $\mathcal{O}_{X,\eta_Y}$ at the generic point $\eta_Y$ is a DVR with valuation $v_Y$ . The divisor of a rational function $f \in K(X)^\times$ is

2

\quad \operatorname{div}(f) = \sum_{Y} v_Y(f)\,[Y].

Derivation hint: $v_Y(f)$ counts the order of vanishing (or pole) of $f$ along $Y$ ; finiteness follows because $f$ is regular and nonzero on a dense open set whose complement has finitely many codimension-one components. Divisors of the form $\operatorname{div}(f)$ are principal, forming a subgroup $\operatorname{Prin}(X)$ . The divisor class group is

3

\quad \operatorname{Cl}(X) = \operatorname{Div}(X)/\operatorname{Prin}(X).

DefinitionCartier Divisor

Let $\mathcal{K}_X$ denote the sheaf of total quotient rings (for integral $X$ , the constant sheaf $K(X)$ ), and $\mathcal{K}_X^\times$ its units. A Cartier divisor is a global section of the quotient sheaf $\mathcal{K}_X^\times/\mathcal{O}_X^\times$ :

4

\quad \operatorname{CaDiv}(X) = H^0\!\left(X,\ \mathcal{K}_X^\times/\mathcal{O}_X^\times\right).

Derivation hint: concretely a Cartier divisor is a collection $\{(U_i, f_i)\}$ with $f_i \in K(X)^\times$ and $f_i/f_j \in \mathcal{O}_X^\times(U_i\cap U_j)$ — local equations agreeing up to units. It is principal if it lies in the image of $K(X)^\times = H^0(X,\mathcal{K}_X^\times)$ . The quotient is the Cartier class group $\operatorname{CaCl}(X)$ .

DefinitionLine Bundle / Invertible Sheaf and the Picard Group

A line bundle (invertible sheaf) is an $\mathcal{O}_X$ -module $\mathcal{L}$ that is locally free of rank one: there is an open cover $\{U_i\}$ with $\mathcal{L}|_{U_i} \cong \mathcal{O}_{U_i}$ . The Picard group is the set of isomorphism classes under tensor product:

5

\quad \operatorname{Pic}(X) = \{\text{invertible sheaves}\}/\!\cong,\qquad [\mathcal{L}]\cdot[\mathcal{M}] = [\mathcal{L}\otimes_{\mathcal{O}_X}\mathcal{M}],\quad [\mathcal{L}]^{-1} = [\mathcal{L}^\vee].

Derivation hint: $\mathcal{L}\otimes\mathcal{L}^\vee \cong \mathcal{H}om(\mathcal{L},\mathcal{L}) \cong \mathcal{O}_X$ via evaluation, giving inverses; associativity and identity ( $\mathcal{O}_X$ ) are inherited from $\otimes$ . Cohomologically, gluing the trivializations by transition functions $g_{ij} =$ (change of frame) $\in \mathcal{O}_X^\times(U_i\cap U_j)$ satisfying the cocycle condition $g_{ij}g_{jk}=g_{ik}$ identifies

6

\quad \operatorname{Pic}(X) \;\cong\; H^1\!\left(X,\ \mathcal{O}_X^\times\right).

TheoremThe Divisor–Line Bundle Dictionary

Let $X$ be integral, separated, and Noetherian. 1. There is a natural isomorphism of groups

7

\quad \operatorname{CaCl}(X) \;\xrightarrow{\ \sim\ }\; \operatorname{Pic}(X),\qquad D \longmapsto \mathcal{O}_X(D),

where $\mathcal{O}_X(D)$ has local generators $f_i^{-1}$ on $U_i$ for $D=\{(U_i,f_i)\}$ . Derivation hint: the map sends a Cartier class to its associated invertible subsheaf of $\mathcal{K}_X$ ; surjectivity uses that any line bundle has a rational section, injectivity that two such sections differ by a global rational function. 2. There is a natural homomorphism $\operatorname{CaDiv}(X)\to\operatorname{Div}(X)$ sending $\{(U_i,f_i)\}\mapsto \sum_Y v_Y(f_i)[Y]$ (well-defined since the $f_i$ agree up to units). It induces

8

\quad \operatorname{Pic}(X)\;\cong\;\operatorname{CaCl}(X)\ \longrightarrow\ \operatorname{Cl}(X),

which is an isomorphism when $X$ is locally factorial (every local ring is a UFD — e.g. $X$ regular).

ProofThe exponential/units sequence and

\operatorname{Pic} = H^1(\mathcal{O}^\times)

Step 1 — Line bundles are torsors. Choose a trivializing cover $\{U_i\}$ of $\mathcal{L}$ with isomorphisms $\varphi_i:\mathcal{L}|_{U_i}\xrightarrow{\sim}\mathcal{O}_{U_i}$ . On overlaps $\varphi_i\varphi_j^{-1}$ is an automorphism of $\mathcal{O}_{U_i\cap U_j}$ , i.e. multiplication by a unit $g_{ij}\in\mathcal{O}_X^\times(U_{ij})$ .

Step 2 — Cocycle condition. $\varphi_i\varphi_j^{-1}\cdot\varphi_j\varphi_k^{-1}=\varphi_i\varphi_k^{-1}$ gives

9

\quad g_{ij}\,g_{jk} = g_{ik}\quad\text{on } U_i\cap U_j\cap U_k.

Derivation hint: compose the three trivialization comparisons; cancellation of $\varphi_j$ yields the multiplicative cocycle identity, the defining relation of a Čech 1-cocycle.

Step 3 — Coboundaries = isomorphisms. Changing trivializations $\varphi_i\mapsto h_i\varphi_i$ with $h_i\in\mathcal{O}_X^\times(U_i)$ replaces $g_{ij}\mapsto h_i g_{ij} h_j^{-1}$ , a coboundary. Hence $[\mathcal{L}]$ depends only on the class $[g_{ij}]\in \check H^1(\{U_i\},\mathcal{O}_X^\times)$ .

Step 4 — Tensor = sum. The transition cocycle of $\mathcal{L}\otimes\mathcal{M}$ is $g_{ij}^{\mathcal{L}}\,g_{ij}^{\mathcal{M}}$ , so the bijection $\operatorname{Pic}(X)\cong H^1(X,\mathcal{O}_X^\times)$ is a group isomorphism. $\qquad\blacksquare$

∎

ProofThe divisor sequence and the class group

Step 1 — Short exact sequence of sheaves. On $X$ integral there is

10

\quad 1 \longrightarrow \mathcal{O}_X^\times \longrightarrow \mathcal{K}_X^\times \longrightarrow \mathcal{K}_X^\times/\mathcal{O}_X^\times \longrightarrow 1 .

Derivation hint: $\mathcal{O}_X^\times$ sits inside the constant sheaf of nonzero rational functions $\mathcal{K}_X^\times$ ; the cokernel is the sheaf whose global sections are Cartier divisors.

Step 2 — Long exact cohomology. Taking cohomology and using that $\mathcal{K}_X^\times$ is a constant (flasque) sheaf, so $H^1(X,\mathcal{K}_X^\times)=0$ , gives

11

\quad K(X)^\times \xrightarrow{\operatorname{div}} \operatorname{CaDiv}(X) \longrightarrow H^1(X,\mathcal{O}_X^\times) \longrightarrow 0.

Derivation hint: the connecting map sends a Cartier divisor $\{(U_i,f_i)\}$ to the cocycle $f_i/f_j\in\mathcal{O}_X^\times$ , exactly the transition functions of $\mathcal{O}_X(D)$ .

Step 3 — Identify cokernel. Exactness yields $\operatorname{CaDiv}(X)/\operatorname{im}(\operatorname{div}) = \operatorname{CaCl}(X)\cong H^1(X,\mathcal{O}_X^\times)=\operatorname{Pic}(X)$ , proving [7].

Step 4 — Factoriality. On a locally factorial $X$ every height-one prime $\mathfrak p\subset\mathcal{O}_{X,x}$ is principal; thus each prime Weil divisor is locally cut out by one equation, making $\operatorname{Div}(X)\to\operatorname{CaDiv}(X)$ an isomorphism, hence [8] is an isomorphism. $\qquad\blacksquare$

∎

Figure

The three equivalent languages on overlaps

U_i\cap U_j

mathematician · Émile Picard | 1856–1941

Picard, working in the analytic theory of algebraic surfaces, studied the group of divisor classes attached to a surface and the role of regular differentials — what we now isolate as $\operatorname{Pic}^0$ and the Néron–Severi group. His name attached to the group of line bundles after the abstract formulation crystallized in the 1950s. Picard's deeper legacy is the recognition that the transcendental part of this group (an abelian variety, the Picard variety) and its discrete part (the Néron–Severi group $\operatorname{NS}(X)$ ) carry complementary geometric information — a structure later proven in full generality by Severi, Lefschetz, and made functorial by Grothendieck via the Picard scheme $\underline{\operatorname{Pic}}_{X}$ .

🧠 Intuition

Think of a line bundle as a consistent choice of "scale" that varies smoothly over $X$ but cannot necessarily be globalized to a single scale. A divisor records where you allow poles and zeros: $\mathcal{O}_X(D)$ is the sheaf of rational functions $f$ with $\operatorname{div}(f)+D\ge 0$ , i.e. functions whose poles are bounded by $D$ . Tensoring bundles = adding divisors = composing scale-changes — all the same operation. The Picard group measures the obstruction to trivializing: $\operatorname{Pic}(X)=0$ precisely when every codimension-one subvariety is globally a principal divisor, the geometric face of unique factorization. On a curve, $\operatorname{Pic}$ is governed by degree and the Jacobian; on affine space everything is principal so $\operatorname{Pic}=0$ ; on projective space $\operatorname{Pic}(\mathbb{P}^n)=\mathbb{Z}$ , generated by the hyperplane class $\mathcal{O}(1)$ .

ExampleProjective space:

\operatorname{Cl}(\mathbb{P}^n_k)\cong\mathbb{Z}

Given: $X=\mathbb{P}^n_k$ with homogeneous coordinate ring $k[x_0,\dots,x_n]$ .

Step 1 — Degree map. Every prime divisor is the zero locus of an irreducible homogeneous polynomial of some degree $d>0$ . Define $\deg\big(\sum n_Y[Y]\big)=\sum n_Y\,\deg(Y)$ , giving a surjection $\operatorname{Div}\to\mathbb{Z}$ .

Step 2 — Kernel is principal. A rational function on $\mathbb{P}^n$ is a ratio $F/G$ of homogeneous polynomials of equal degree, so $\operatorname{div}(F/G)$ has degree $0$ . Conversely any degree-zero divisor equals $\operatorname{div}(F/G)$ for suitable $F,G$ .

Step 3 — Conclude. The degree map descends to an isomorphism $\operatorname{Cl}(\mathbb{P}^n)\xrightarrow{\sim}\mathbb{Z}$ . Since $\mathbb{P}^n$ is regular, $\operatorname{Pic}(\mathbb{P}^n)\cong\operatorname{Cl}(\mathbb{P}^n)$ , generated by the hyperplane bundle $\mathcal{O}(1)$ .

\boxed{\operatorname{Pic}(\mathbb{P}^n_k)\cong\mathbb{Z}\cdot[\mathcal{O}(1)],\qquad \mathcal{O}(d)\otimes\mathcal{O}(e)\cong\mathcal{O}(d+e).}

ExampleA nodal/singular quadric cone: nontrivial torsion

Given: $X=\operatorname{Spec} k[x,y,z]/(xy-z^2)$ , the affine quadric cone (a normal surface singular at the origin).

Step 1 — A non-principal divisor. Let $L$ be the line $\{x=z=0\}$ (a ruling). Its ideal is not principal: $L$ is cut out by the height-one prime $(x,z)$ , which requires two generators.

Step 2 — Compute orders. $\operatorname{div}(x)=2[L]$ since $x=z^2/y$ near $L$ and the valuation of $x$ along $L$ is $2$ (the cone is singular). Thus $2[L]=\operatorname{div}(x)$ is principal, but $[L]$ itself is not.

Step 3 — Class group. One shows $\operatorname{Cl}(X)\cong\mathbb{Z}/2\mathbb{Z}$ , generated by $[L]$ with $2[L]=0$ .

\boxed{\operatorname{Cl}(\text{quadric cone})\cong\mathbb{Z}/2\mathbb{Z}.}

Caveat: because $X$ is not locally factorial at the cone point, $\operatorname{Pic}(X)\subsetneq\operatorname{Cl}(X)$ . In fact $\operatorname{Pic}(X)=0$ (the cone is affine and the only Cartier divisors are principal near the vertex), so $[L]$ is Weil-but-not-Cartier — the prototypical failure of [8].

⚠️ Common mistake

"Weil = Cartier always." False. The equality holds only when $X$ is locally factorial. On normal but non-factorial varieties (the quadric cone above) $\operatorname{Pic}(X)\subsetneq\operatorname{Cl}(X)$ .
"Divisors are subvarieties." A divisor is a formal $\mathbb{Z}$ -combination; coefficients can be negative (poles). The associated sheaf $\mathcal{O}_X(D)$ is what is geometric.
" $\mathcal{O}_X(D)$ depends on the divisor, not the class." It depends only on the linear-equivalence class: $D\sim D'\iff\mathcal{O}_X(D)\cong\mathcal{O}_X(D')$ .
" $\operatorname{Pic}=H^1(\mathcal{O}^\times)$ needs the analytic topology." No — the isomorphism is Zariski/étale-sheaf-theoretic; in the analytic category one further uses the exponential sequence to relate it to $H^1(X,\mathcal{O})$ and $H^2(X,\mathbb{Z})$ .

CounterexampleSections do not detect the class faithfully

A line bundle $\mathcal{L}$ may have no global sections yet be nontrivial: on a genus- $g$ curve $C$ , $\mathcal{O}_C(D)$ with $\deg D<0$ has $H^0=0$ but $[\mathcal{O}_C(D)]\ne 0$ unless $D\sim 0$ . Conversely, two nonisomorphic line bundles can share the same $h^0$ . The number of independent sections is governed by Riemann–Roch, not by the class alone:

\dim H^0(C,\mathcal{O}(D)) - \dim H^1(C,\mathcal{O}(D)) = \deg D + 1 - g.

Thus $\operatorname{Pic}$ classifies bundles, while cohomology of sections is a finer (and harder) invariant.

connection

The Picard group is the codimension-one piece of a richer hierarchy. The Néron–Severi group $\operatorname{NS}(X)=\operatorname{Pic}(X)/\operatorname{Pic}^0(X)$ is finitely generated (Theorem of the Base), and $\operatorname{Pic}^0(X)$ is an abelian variety (the Jacobian Variety for curves). Intersection of divisors connects to Intersection Theory and the Chow Ring, where $\operatorname{Pic}(X)=\operatorname{CH}^1(X)$ for smooth $X$ . Cohomologically, the first Chern class $c_1:\operatorname{Pic}(X)\to H^2(X,\mathbb{Z})$ links to Chern Classes and the Hodge Conjecture (which divisor classes are algebraic). Ampleness of a line bundle controls embeddings into projective space — the bridge to Ample and Very Ample Line Bundles and the Kodaira Embedding Theorem. Over number fields, the analog is the Ideal Class Group: $\operatorname{Cl}(\operatorname{Spec}\mathcal{O}_K)$ is exactly the ideal class group of the ring of integers.

🔬 Where this appears

⏳ History

- 1851–1857 — Riemann. Introduces "Abelian functions," the Riemann surface, and the inequality

\dim H^0(D)\ge \deg D - g + 1

, the seed of divisor theory.

- 1865 — Roch. Completes Riemann's inequality to the exact Riemann–Roch equality on curves.

- 1882 — Dedekind & Weber. Algebraic theory of function fields; divisors as products of prime ideals, paralleling number theory.

- 1880s–1900s — Picard, Poincaré. Study divisor classes on surfaces; introduce the Picard variety via integrals of the first kind.

- 1906–1910 — Severi. Defines the Néron–Severi group; proves finite generation (Theorem of the Base) over

\mathbb{C}

.

- 1937–1940 — Weil, Zariski. Foundations of abstract algebraic geometry; rigorous theory of valuations and divisors on general varieties.

- 1950s — Serre (FAC, 1955). Coherent sheaves;

\operatorname{Pic}=H^1(\mathcal{O}^\times)

; the GAGA principle linking analytic and algebraic line bundles.

- 1957–1962 — Grothendieck. Schemes, Cartier divisors, and the representability of the Picard functor — the Picard scheme

\underline{\operatorname{Pic}}_{X/S}

(FGA).

- 1958–1965 — Néron, Chevalley, Mumford. Néron models, generic structure of

\operatorname{Pic}^0

as an abelian variety; modern functorial Picard theory.

open-problem

Representability of the Picard functor is delicate: $\underline{\operatorname{Pic}}_{X/S}$ is representable by a scheme under projectivity/flatness hypotheses (Grothendieck, Murre, Oort), but in general only by an algebraic space; the obstruction lives in the relative Brauer group $\operatorname{Br}(S)$ .
The Néron–Severi rank $\rho(X)$ (the Picard number) is computable in principle but hard in practice; bounding $\rho$ for K3 surfaces and the distribution of Picard numbers in families remain active.
Hodge / Tate conjectures ask whether $c_1$ surjects onto the appropriate "algebraic" classes; for divisors this is the Lefschetz $(1,1)$ -theorem over $\mathbb{C}$ (a theorem), but the integral and arithmetic ( $\ell$ -adic Tate) analogues for divisors are subtle — the integral Hodge conjecture fails in codimension $\ge 2$ but holds for divisors.
Brauer–Picard interplay: the exact sequence relating $\operatorname{Pic}$ , $\operatorname{Br}$ , and Galois cohomology controls how Picard classes descend along field extensions; effective descent of line bundles is governed by $H^1(\mathrm{Gal},\operatorname{Pic})$ and the Brauer obstruction.
Positivity in higher dimensions: characterizing which classes in $\operatorname{NS}(X)_\mathbb{R}$ are ample (the ample cone) and nef (the nef cone) leads to open conjectures (e.g. the SQM/Morrison–Kawamata cone conjecture).

Divisors, Line Bundles, and the Picard Group

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also