Toric Picard group

Intuition [Beginner]

The Picard group of a variety records the line bundles on it, with multiplication by tensor product. On a toric variety $X_{\Sigma }$ built from a fan $\Sigma$ in the previous unit, this group has a beautifully combinatorial description. Each one-dimensional cone (ray) of the fan contributes a torus-invariant divisor, and every line bundle on $X_{\Sigma }$ comes from an integer combination of these ray divisors. Two such combinations give the same line bundle if and only if they differ by a "principal" combination coming from a character of the torus. So the Picard group is the free abelian group on the rays, quotiented by the characters.

Why is this useful? Computing line bundles on a general variety is hard — you have to track invertible sheaves modulo isomorphism, and there is no canonical recipe. On a toric variety the recipe is finite and combinatorial. The Picard group has rank equal to the number of rays minus the dimension of the fan (when $X_{\Sigma }$ is smooth and complete), and it has explicit generators given by the toric divisors $D_{\rho }$. The cones of "positive" line bundles — the ones giving projective embeddings or non-collapsing maps to projective space — are themselves polyhedral cones inside the Picard group, computable directly from the fan.

The classical examples are sharp. Projective space ${\mathbb{P}}^n$ has Picard group $\mathbb{Z}$, generated by the hyperplane class. The product ${\mathbb{P}}^1\times {\mathbb{P}}^1$ has ${\mathbb{Z}}^2$, one factor for each projective line. The Hirzebruch surface $F_a$ also has ${\mathbb{Z}}^2$, generated by a fibre class and a section class. The weighted projective space $\mathbb{P}(1,2,3)$, which is singular, has Picard group $\mathbb{Z}$ but a larger divisor class group $\mathbb{Z}$ generated by the rational class $D/6$. Every case is read off from the fan in a few lines.

Visual [Beginner]

A schematic of the toric Picard exact sequence, illustrating how the character lattice $M$ maps into the free abelian group on the rays, with cokernel the Picard group. Left side: the character lattice $M={\mathbb{Z}}^n$ pictured as a regular grid in $n$-space. Centre: the free abelian group ${⨁}_{\rho }\mathbb{Z}\cdot D_{\rho }$ pictured as a hypercube grid with one axis per ray of the fan. Right side: the Picard group of $X_{\Sigma }$ pictured as the quotient — the hypercube with the image of $M$ collapsed to the origin.

The diagram captures the central recipe: characters of the torus give principal divisor combinations, ray divisors generate all line bundles, and the Picard group is what is left over. The picture also previews the rank formula — rank of the Picard group equals the number of generators (one per ray, $|\Sigma (1)|$) minus the rank of the image of $M$ (equal to $n$ in the smooth complete case).

Worked example [Beginner]

Compute the Picard group of projective space ${\mathbb{P}}^2$. The fan $\Sigma$ of ${\mathbb{P}}^2$ in ${\mathbb{R}}^2$ has three rays with primitive generators $v_0=-e_1-e_2$, $v_1=e_1$, $v_2=e_2$, with the single relation $v_0+v_1+v_2=0$.

Step 1. Identify the building blocks. Each ray ${\rho }_i$ contributes a toric divisor $D_i$, so the free abelian group on the rays is $\mathbb{Z}{D}_0\oplus \mathbb{Z}{D}_1\oplus \mathbb{Z}{D}_2\cong {\mathbb{Z}}^3$. The character lattice is $M={\mathbb{Z}}^2$, with dual basis $e_1^{\ast },e_2^{\ast }$.

Step 2. Compute the character map. The map $M\to {\mathbb{Z}}^3$ sends $m\mapsto (⟨m,v_0⟩,⟨m,v_1⟩,⟨m,v_2⟩)$. Evaluate on basis: $e_1^{\ast }\mapsto (-1,1,0)$ and $e_2^{\ast }\mapsto (-1,0,1)$. So the image is the subgroup of ${\mathbb{Z}}^3$ spanned by these two vectors.

Step 3. Compute the cokernel. The Picard group is ${\mathbb{Z}}^3$ modulo the relations $D_1-D_0=0$ and $D_2-D_0=0$, i.e., $D_0=D_1=D_2$ in $\mathrm{Pic}({\mathbb{P}}^2)$. Set $H=D_0=D_1=D_2$. Then $\mathrm{Pic}({\mathbb{P}}^2)=\mathbb{Z}\cdot H$, the free abelian group on the hyperplane class. Rank check: $|\Sigma (1)|-n=3-2=1$, matching.

What this tells us. The Picard group of ${\mathbb{P}}^2$ is $\mathbb{Z}$, generated by the hyperplane class $H$. Every line bundle on ${\mathbb{P}}^2$ has the form $\mathcal{O}(d)={\mathcal{O}}_{{\mathbb{P}}^2}(d\cdot H)$ for some integer $d$, with $d>0$ ample (very ample for $d\ge 1$), $d=0$ the identity line bundle, and $d<0$ anti-ample. The whole picture comes out of three rays in ${\mathbb{R}}^2$ and one short calculation.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $\Sigma$ be a fan in $N_{\mathbb{R}}$ with $N$ a lattice of rank $n$ and dual lattice $M={\mathrm{Hom}}_{\mathbb{Z}}(N,\mathbb{Z})$. Let $\Sigma (1)$ denote the set of rays of $\Sigma$ (one-dimensional cones), and for each $\rho \in \Sigma (1)$ let $v_{\rho }\in N$ denote the primitive generator (the unique lattice point on $\rho$ such that $\rho ={\mathbb{R}}_{\ge 0}\cdot v_{\rho }$).

Definition (toric divisor). For each ray $\rho \in \Sigma (1)$, the corresponding toric divisor $D_{\rho }\subseteq X_{\Sigma }$ is the closure of the codimension-one torus orbit $O_{\rho }$ (an orbit of $T$ of dimension $n-1$ corresponding to $\rho$ under the orbit-cone correspondence of the next unit [04.11.05]). The divisors $D_{\rho }$ are torus-invariant, prime, and Weil. The free abelian group on these is denoted $$ \mathrm{Div}T(X\Sigma) := \bigoplus_{\rho \in \Sigma(1)} \mathbb{Z} \cdot D_\rho, $$ the group of torus-invariant Weil divisors.

Definition (character map). For each character $m\in M$, the rational function ${\chi }^m$ on the dense torus $T\subseteq X_{\Sigma }$ extends to a rational function on $X_{\Sigma }$ with divisor $$ \mathrm{div}(\chi^m) = \sum_{\rho \in \Sigma(1)} \langle m, v_\rho\rangle \cdot D_\rho \in \mathrm{Div}T(X\Sigma). $$ The assignment $m\mapsto \mathrm{div}({\chi }^m)$ is a $\mathbb{Z}$-linear map $\mathrm{div}:M\to {\mathrm{Div}}_T(X_{\Sigma })$.

Definition (divisor class group). The divisor class group of $X_{\Sigma }$ is $$ \mathrm{Cl}(X_\Sigma) := \mathrm{Div}(X_\Sigma) / \mathrm{div}(K(X_\Sigma)^*), $$ the group of Weil divisors on $X_{\Sigma }$ modulo principal divisors. For a normal variety with a $T$-action, every Weil divisor is linearly equivalent to a torus-invariant one (by averaging over the torus), and the principal divisors of interest are exactly the $\mathrm{div}({\chi }^m)$ for $m\in M$. Hence the toric description: $$ \mathrm{Cl}(X_\Sigma) = \mathrm{coker}\bigl(\mathrm{div} : M \to \mathrm{Div}T(X\Sigma)\bigr). $$

Definition (Picard group). The Picard group $\mathrm{Pic}(X_{\Sigma })$ is the group of isomorphism classes of invertible sheaves on $X_{\Sigma }$ under tensor product, equivalently the group of Cartier divisors modulo principal Cartier divisors. There is a natural inclusion $\mathrm{Pic}(X_{\Sigma })\hookrightarrow \mathrm{Cl}(X_{\Sigma })$ sending a Cartier divisor to its Weil-divisor class. For smooth $X_{\Sigma }$ — equivalently, every cone in $\Sigma$ generated by part of a $\mathbb{Z}$-basis of $N$ (the smoothness criterion of [04.11.04]) — this inclusion is an equality: $\mathrm{Pic}(X_{\Sigma })=\mathrm{Cl}(X_{\Sigma })$.

Convention. For singular toric varieties, $\mathrm{Pic}(X_{\Sigma })$ sits as a (potentially proper) subgroup of $\mathrm{Cl}(X_{\Sigma })$ — the Weil divisors that are locally given by one equation on each affine chart $U_{\sigma }$. The next subsection makes this distinction precise via support functions.

Counterexamples to common slips

• "Every Weil divisor on a toric variety is Cartier." False on singular toric. Example: on $\mathbb{P}(1,2,3)$, the divisor $D_0$ at the singular fixed point is Weil but not Cartier; only multiples by $6=\mathrm{lcm}(1,2,3)$ are Cartier. The distinction is the difference between $\mathrm{Cl}$ (all Weil) and $\mathrm{Pic}$ (Cartier only).

• "The character map $\mathrm{div}:M\to {\mathrm{Div}}_T(X_{\Sigma })$ is always injective." It is injective when $|\Sigma (1)|$ spans $N_{\mathbb{R}}$ (equivalently when the rays are not contained in a proper hyperplane of $N_{\mathbb{R}}$) — in particular, always for complete fans. For non-complete fans where the rays span a proper subspace, the kernel of $\mathrm{div}$ is the dual of the lineality space of the rays.

• "The Picard group is free abelian." True for smooth complete toric varieties (the cokernel of an injective map from a free abelian group to a free abelian group). False on singular toric: $\mathrm{Cl}$ can have torsion, and even $\mathrm{Pic}$ can be a proper subgroup of finite index in $\mathrm{Cl}$. The torsion arises from the failure of the character map to have saturated image.

Key theorem with proof [Intermediate+]

The signature theorem of this unit is the toric Picard exact sequence of Fulton §3.4.

Theorem (toric Picard exact sequence; Demazure 1970, Fulton 1993). Let $\Sigma$ be a fan in $N_{\mathbb{R}}$ whose rays $\Sigma (1)$ span $N_{\mathbb{R}}$ (in particular, any complete fan satisfies this). Then there is a short exact sequence of abelian groups $$ 0 \to M \xrightarrow{\mathrm{div}} \bigoplus_{\rho \in \Sigma(1)} \mathbb{Z} \cdot D_\rho \to \mathrm{Cl}(X_\Sigma) \to 0, $$ with the first map $\mathrm{div}(m)={\sum }_{\rho }⟨m,v_{\rho }⟩D_{\rho }$ and the second the natural quotient. When $X_{\Sigma }$ is smooth, this becomes $$ 0 \to M \xrightarrow{\mathrm{div}} \bigoplus_{\rho \in \Sigma(1)} \mathbb{Z} \cdot D_\rho \to \mathrm{Pic}(X_\Sigma) \to 0, $$ with $\mathrm{Pic}(X_{\Sigma })=\mathrm{Cl}(X_{\Sigma })$ a free abelian group of rank $$ \mathrm{rank},\mathrm{Pic}(X_\Sigma) = |\Sigma(1)| - n. $$

Proof.

(Surjectivity of the quotient.) Every Weil divisor on $X_{\Sigma }$ is linearly equivalent to a $T$-invariant one. Reason: given a Weil divisor $D$, the orbit $D\cdot T$ under the dense torus action gives a family of linearly equivalent divisors parametrised by $T$; by completeness of the torus and a standard averaging argument (Fulton §3.3, using that the linear-equivalence class is a single closed orbit in the Hilbert scheme), one can find a $T$-fixed representative. So the natural map ${\mathrm{Div}}_T(X_{\Sigma })\to \mathrm{Cl}(X_{\Sigma })$ is surjective.

(Image of $\mathrm{div}$ is the kernel of the quotient.) A torus-invariant divisor $D={\sum }_{\rho }{a}_{\rho }{D}_{\rho }$ is principal — i.e., zero in $\mathrm{Cl}(X_{\Sigma })$ — if and only if $D=\mathrm{div}(f)$ for some rational function $f$ on $X_{\Sigma }$. By $T$-equivariance, if $D$ is $T$-invariant and principal, $f$ can be chosen $T$-semi-invariant, i.e., $f=c\cdot {\chi }^m$ for some $c\in {\mathbb{C}}^{\ast }$ and $m\in M$. The divisor of ${\chi }^m$ on $X_{\Sigma }$ is computed locally on each chart $U_{\sigma }$: the rational function ${\chi }^m\in \mathbb{C}[M]$ has order of vanishing along the divisor $D_{\rho }\cap U_{\sigma }$ (for $\rho \subseteq \sigma$ a ray) equal to $⟨m,v_{\rho }⟩$, since the local equation of $D_{\rho }$ in $U_{\sigma }$ is ${\chi }^{m_{\rho }}$ where $m_{\rho }\in {\sigma }^{\vee }\cap M$ generates the face of ${\sigma }^{\vee }$ dual to $\rho$, and ${\chi }^m/{\chi }^{⟨m,v_{\rho }⟩\cdot m_{\rho }}$ extends to a unit on a neighbourhood of $D_{\rho }$. Hence $$ \mathrm{div}(\chi^m) = \sum_\rho \langle m, v_\rho\rangle D_\rho, $$ confirming the formula. Conversely, every principal torus-invariant divisor is of this form.

(Injectivity of $\mathrm{div}$ when rays span $N_{\mathbb{R}}$.) Suppose $\mathrm{div}(m)=0$, i.e., $⟨m,v_{\rho }⟩=0$ for every $\rho \in \Sigma (1)$. Since the $v_{\rho }$ span $N_{\mathbb{R}}$ by hypothesis, the linear functional $m:N_{\mathbb{R}}\to \mathbb{R}$ vanishes on a spanning set, hence $m=0$ as a functional, hence $m=0$ in $M$ (since $M\hookrightarrow N_{\mathbb{R}}^{\ast }$). So $\mathrm{div}$ is injective.

(Equality $\mathrm{Pic}=\mathrm{Cl}$ in the smooth case.) If $X_{\Sigma }$ is smooth, every Weil divisor is Cartier (the local-rings of $X_{\Sigma }$ are regular at every point, so every height-one prime is principal locally). Hence $\mathrm{Pic}(X_{\Sigma })=\mathrm{Cl}(X_{\Sigma })$.

(Rank formula.) In the smooth case, the exact sequence $0\to M\to {\mathbb{Z}}^{|\Sigma (1)|}\to \mathrm{Pic}(X_{\Sigma })\to 0$ realises $\mathrm{Pic}(X_{\Sigma })$ as the cokernel of an injective map from a free abelian group of rank $n$ to one of rank $|\Sigma (1)|$. The cokernel is free abelian (no torsion; see Counterexamples slip above) of rank $|\Sigma (1)|-n$. $\square$

Bridge. The toric Picard exact sequence builds toward the polytope-fan correspondence of [04.11.10] and appears again in [04.11.12] as the basic linear-relation block in the Chow-ring presentation of a smooth complete toric variety. The central insight is that the geometry of line bundles on $X_{\Sigma }$ is completely captured by the combinatorial data of the rays — every line bundle is a finite integer combination of toric divisors $D_{\rho }$, with the redundancy supplied by characters.

This is exactly the foundational reason that toric Picard groups are tractable: where a general variety's Picard group is computed by a difficult cohomological argument (Hodge theory, the Néron-Severi group, or the Albanese variety), the toric Picard group reduces to a single short exact sequence of free abelian groups. Putting these together with the projectivity criterion of [04.11.04] (projective iff $\Sigma$ is the normal fan of a polytope) and the polytope-fan dictionary of [04.11.10], the toric Picard exact sequence is the bridge between fan combinatorics and divisor / line-bundle geometry. The pattern recurs: the same exact sequence supplies the grading of the Cox ring in [04.11.15] and identifies $X_{\Sigma }$ with a GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))/G$ for $G=\mathrm{Hom}(\mathrm{Pic},{\mathbb{C}}^{\ast })$, generalises in the equivariant setting to $T$-equivariant Picard groups, and is dual to the cocharacter-vs-character duality $N\leftrightarrow M$ from [04.11.01].

Exercises [Intermediate+]

Lean formalization [Intermediate+]

The toric Picard exact sequence requires the Fan formalism from [04.11.04] and the ToricVariety construction together with the toric-divisor formalism (currently pending). Schematically:

import Mathlib.Algebra.Module.LinearMap.Defs
import Mathlib.LinearAlgebra.FreeModule.Basic
import Mathlib.Algebra.Exact
import Mathlib.AlgebraicGeometry.Scheme

namespace Codex.AlgGeom.Toric.PicardGroup

universe u

/-- Placeholder for the lattice `N` of cocharacters and `M` of characters
of an algebraic torus. Once `Codex.AlgGeom.Toric.Lattices` ships these as
`FreeAbelianGroup`s of finite rank, replace these `variable` declarations
with imports. -/
variable {N : Type u} [AddCommGroup N] [Module.Free ℤ N] [Module.Finite ℤ N]
variable {M : Type u} [AddCommGroup M] [Module.Free ℤ M] [Module.Finite ℤ M]
variable (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ)  -- the perfect pairing M × N → ℤ

/-- A primitive ray generator of a rational polyhedral cone in `N_ℝ`,
modelled as an element of `N`. The primitivity is a side condition not
recorded in the type for this skeleton. -/
structure PrimitiveRay (N : Type u) [AddCommGroup N] where
  generator : N

/-- A *fan* in `N` is (a placeholder for) a finite collection of cones
together with face-closure and intersection-as-face axioms; here we only
record the finite ray-set `rays` and the dimension `n` of the ambient
free abelian group, which is what the Picard-group exact sequence uses.
The full `Fan` structure lives in `04.11.04`. -/
structure Fan (N : Type u) [AddCommGroup N] [Module.Free ℤ N] where
  rays : Finset (PrimitiveRay N)
  /-- The rays span the rationalisation `N_ℝ`, i.e., the image of
  `rays.image (·.generator)` in `N ⊗ ℝ` is the entire space. Equivalent
  to the completeness condition on the fan (or any fan whose rays
  span `N_ℝ`). -/
  rays_span : True

/-- The free abelian group on the rays: `ZSigmaOne := ⊕_{ρ ∈ Σ(1)} ℤ · D_ρ`.
Each generator is the formal divisor class of a torus-invariant prime divisor.
We model it via `FreeAbelianGroup` on the finite ray set. -/
def ZSigmaOne (Σ : Fan N) : Type u :=
  Σ.rays → ℤ  -- formal ℤ-linear combinations of rays

instance (Σ : Fan N) : AddCommGroup (ZSigmaOne Σ) :=
  Pi.addCommGroup

instance (Σ : Fan N) : Module ℤ (ZSigmaOne Σ) :=
  Pi.module _ _ _

/-- The character map `div : M → ZSigmaOne Σ` sending `m ↦ ∑_ρ ⟨m, v_ρ⟩ D_ρ`.
The pairing `⟨m, v_ρ⟩` uses the perfect pairing `M × N → ℤ` and the
primitive generator of `ρ`. -/
noncomputable def divMap (Σ : Fan N) (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ) :
    M →ₗ[ℤ] ZSigmaOne Σ := by
  refine ⟨⟨fun m => fun ρ => pair m ρ.val.generator, ?_⟩, ?_⟩
  · intro m m'
    funext ρ
    simp [map_add]
  · intro c m
    funext ρ
    simp [map_smul]

/-- The toric Picard / divisor-class group: the cokernel of `divMap`.
For a smooth complete toric variety this is `Pic(X_Σ)`; for a general
fan whose rays span `N_ℝ` it is `Cl(X_Σ)` (Weil divisors mod principal),
with `Pic(X_Σ) ↪ Cl(X_Σ)` of finite index in the singular case. -/
noncomputable def PicardGroup (Σ : Fan N) (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ) :
    Type u :=
  (ZSigmaOne Σ) ⧸ LinearMap.range (divMap Σ pair)

instance (Σ : Fan N) (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ) :
    AddCommGroup (PicardGroup Σ pair) := by
  unfold PicardGroup
  infer_instance

/-! ## The toric Picard exact sequence (Demazure 1970, Fulton §3.4)

The key theorem of this unit: a short exact sequence

  0 → M --div--> ⊕_{ρ ∈ Σ(1)} ℤ · D_ρ → Pic(X_Σ) → 0

with `div` injective whenever the rays span `N_ℝ` (in particular, for any
complete fan). When `X_Σ` is smooth, this is the full Picard group; in
the singular case, the same exact sequence with `Cl(X_Σ)` on the right
remains valid, and `Pic(X_Σ) ↪ Cl(X_Σ)` is a finite-index subgroup.
-/

/-- The toric Picard exact sequence: `M → ZSigmaOne → Pic` is exact in
the middle, with `div` injective (when rays span `N_ℝ`) and the quotient
surjective. Stated as an exactness predicate on the composed maps. -/
theorem toric_picard_exact_sequence (Σ : Fan N) (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ) :
    Function.Injective (divMap Σ pair) ∧
      Function.Exact (divMap Σ pair)
        (LinearMap.range (divMap Σ pair)).mkQ := by
  -- (i) Injectivity: rays span N_ℝ implies the pairing M → ⊕_ρ ℤ is
  -- injective by non-degeneracy of the perfect pairing M × N → ℤ on
  -- rationalisation.
  -- (ii) Exactness in the middle: by definition of mkQ as the cokernel.
  sorry

/-- The rank formula for a smooth complete toric variety:
`rank Pic(X_Σ) = |Σ(1)| - n`. Established as the Euler characteristic of
the short exact sequence `0 → M → ZSigmaOne → Pic → 0` together with
freeness of the cokernel under smoothness. -/
theorem toric_picard_rank (Σ : Fan N) (pair : M →ₗ[ℤ] N →ₗ[ℤ] ℤ)
    (n : ℕ) [Module.Finite ℤ M] (hrank : Module.finrank ℤ M = n)
    (smooth : True)   -- placeholder: smoothness criterion on Σ
    (complete : True) -- placeholder: completeness of Σ
    :
    -- The Picard group is free abelian of rank `|Σ(1)| − n`.
    True := by
  -- (1) Picard exact sequence + injectivity → rank-additivity.
  -- (2) Smoothness: Smith normal form of `divMap` has all elementary
  --     divisors equal to 1 (since on each maximal cone, the primitive
  --     ray generators form a ℤ-basis of N, so the restricted matrix
  --     is in GL_n(ℤ)).
  -- (3) Conclude: cokernel is torsion-free of rank `|Σ(1)| − n`.
  sorry

/-- Worked numerical case: `Pic(P^n) ≅ ℤ`. The fan of P^n has `n + 1` rays
in `R^n`, so the rank is `(n + 1) − n = 1`. -/
theorem pic_projective_space (n : ℕ) :
    -- Symbolic placeholder: ranks of the relevant lattices.
    (n + 1 : ℕ) - n = 1 := by
  omega

/-- Worked numerical case: `Pic(P^1 × P^1) ≅ ℤ²`. The fan has 4 rays in
`R^2`, so the rank is `4 − 2 = 2`. -/
theorem pic_p1_times_p1 :
    (4 : ℕ) - 2 = 2 := by
  decide

/-- Worked numerical case: `Pic(F_a) ≅ ℤ²` for Hirzebruch surface `F_a`.
Independent of `a`, since the rank formula `|Σ(1)| - n = 4 - 2 = 2`. -/
theorem pic_hirzebruch (a : ℤ) :
    (4 : ℕ) - 2 = 2 := by
  decide

/-! ## Ample line bundles and the ample cone

A torus-invariant Cartier divisor `D = ∑_ρ a_ρ D_ρ` on a smooth complete
toric `X_Σ` is ample iff its associated piecewise-linear support function
`ψ_D : |Σ| → ℝ`, defined by `ψ_D(v_ρ) = -a_ρ` and extended linearly on
each maximal cone, is **strictly convex**: at every wall `τ = σ ∩ σ'`,
the two linear pieces of `ψ_D` on `σ` and `σ'` do not agree across `τ`.
-/

/-- The support function `ψ_D : |Σ| → ℝ` associated to a torus-invariant
divisor `D = ∑_ρ a_ρ D_ρ` on `X_Σ`. Placeholder structure; the full
piecewise-linearity and continuity is captured in `04.11.10`. -/
structure SupportFunction (Σ : Fan N) where
  pieces : Σ.rays → ℤ  -- the values `-a_ρ` at each ray

/-- Strict convexity of a support function: at every wall between two
maximal cones, the two linear pieces strictly disagree on the wall's
extension. Placeholder predicate. -/
def SupportFunction.StrictlyConvex (Σ : Fan N) (ψ : SupportFunction Σ) :
    Prop := True  -- placeholder

/-- Ample iff strictly convex: the ampleness criterion of Cox-Little-Schenck
§6.1 Theorem 6.1.14. Placeholder statement; the full proof needs the
toric vanishing theorem and the polytope-fan dictionary. -/
theorem ample_iff_strictly_convex (Σ : Fan N) (ψ : SupportFunction Σ)
    (smooth : True) (complete : True) :
    -- ample(ψ) ↔ StrictlyConvex(ψ)
    SupportFunction.StrictlyConvex Σ ψ ↔ SupportFunction.StrictlyConvex Σ ψ := by
  exact Iff.rfl

/-! ## Mori dream space

Every smooth complete toric variety is a **Mori dream space** in the sense
of Hu-Keel 2000: the Cox ring is finitely generated (= polynomial), the
moveable cone has a finite chamber decomposition into nef cones of GIT
models, and the small Q-factorial modifications are themselves toric.
-/

/-- Hu-Keel 2000: every smooth complete toric variety has a finitely
generated Cox ring. Placeholder statement. -/
theorem smooth_complete_toric_is_mds (Σ : Fan N)
    (smooth : True) (complete : True) :
    True := by
  trivial

end Codex.AlgGeom.Toric.PicardGroup

The skeleton above declares the toric Picard exact sequence, the rank formula, numerical witnesses for ${\mathbb{P}}^n$, ${\mathbb{P}}^1\times {\mathbb{P}}^1$, and Hirzebruch surfaces, the strictly-convex-support-function characterisation of ampleness, and the Mori-dream-space classification. The sorry-stubbed theorems are reachable from current Mathlib once the Fan formalism and toric-divisor infrastructure of [04.11.04] ship: the Picard exact sequence reduces to the cokernel calculation, the rank formula uses Smith normal form, the ampleness criterion uses the toric vanishing theorem, and the Mori-dream-space classification uses Cox's 1995 GIT presentation. The decidable numerical witnesses for ${\mathbb{P}}^n,{\mathbb{P}}^1\times {\mathbb{P}}^1$, and $F_a$ already compile against Nat and Int arithmetic in Mathlib.

Advanced results [Master]

§A. The toric Picard exact sequence and the divisor class group

Theorem (toric Picard exact sequence — refined form; Fulton §3.4, Cox-Little-Schenck Theorem 4.2.1). Let $\Sigma$ be a fan in $N_{\mathbb{R}}$ whose rays span $N_{\mathbb{R}}$. There is a commutative diagram of exact sequences: $$ \begin{array}{ccccccccc} 0 & \to & M & \xrightarrow{\mathrm{div}} & \mathrm{CDiv}T(X\Sigma) & \to & \mathrm{Pic}(X_\Sigma) & \to & 0 \ & & | & & \downarrow & & \downarrow & & \ 0 & \to & M & \xrightarrow{\mathrm{div}} & \mathrm{Div}T(X\Sigma) = \bigoplus_\rho \mathbb{Z} D_\rho & \to & \mathrm{Cl}(X_\Sigma) & \to & 0 \end{array} $$ where ${\mathrm{CDiv}}_T(X_{\Sigma })\subseteq {\mathrm{Div}}_T(X_{\Sigma })$ denotes the torus-invariant Cartier divisors, characterised by the local-Cartier condition on each affine chart $U_{\sigma }$. The vertical inclusion $\mathrm{Pic}\hookrightarrow \mathrm{Cl}$ has finite index if $X_{\Sigma }$ is $\mathbb{Q}$-factorial (every Weil divisor has a Cartier multiple) and is an equality iff $X_{\Sigma }$ is smooth.

The full refinement (Cox-Little-Schenck Theorem 4.2.1) distinguishes the Cartier-divisor description (giving the Picard group) from the Weil-divisor description (giving the class group). The torus-invariant Cartier condition on a divisor $D={\sum }_{\rho }{a}_{\rho }{D}_{\rho }$ is: for every maximal cone $\sigma \in \Sigma$, there exists $m_{\sigma }\in M$ such that $⟨m_{\sigma },v_{\rho }⟩=-a_{\rho }$ for every ray $\rho \subseteq \sigma$. The collection $\{m_{\sigma }{\}}_{\sigma \in {\Sigma }_{\max }}$ is the local-data of the Cartier divisor — equivalently, a piecewise-linear function on $|\Sigma |$ with integer slopes (the support function discussed in §C).

The diagram presents two parallel exact sequences, with the upper one defining $\mathrm{Pic}$ and the lower defining $\mathrm{Cl}$. The vertical inclusion ${\mathrm{CDiv}}_T\hookrightarrow {\mathrm{Div}}_T$ is the identity-on-elements but with the Cartier sub-condition, and the induced inclusion $\mathrm{Pic}\hookrightarrow \mathrm{Cl}$ is the natural map "Cartier divisor → Weil divisor class". For singular toric $X_{\Sigma }$, the inclusion is strict; for smooth toric, it is equality. The exact sequence also makes the divisor class group $\mathrm{Cl}(X_{\Sigma })$ accessible to combinatorial computation in the general (singular) case, where the Picard group itself is harder to describe directly.

§B. The rank formula and the smooth-vs-singular dichotomy

Theorem (rank of $\mathrm{Cl}(X_{\Sigma })$; Fulton §3.4 Proposition). For any fan $\Sigma$ in $N_{\mathbb{R}}$ with rays spanning $N_{\mathbb{R}}$, the divisor class group satisfies $$ \mathrm{rank},\mathrm{Cl}(X_\Sigma) = |\Sigma(1)| - n, $$ with torsion equal to the torsion of ${\mathbb{Z}}^{|\Sigma (1)|}/\mathrm{image}(\mathrm{div})$, computable as the cokernel-torsion via Smith normal form of the ray-matrix.

Theorem (rank of $\mathrm{Pic}(X_{\Sigma })$ for smooth complete; Demazure 1970). If $X_{\Sigma }$ is smooth and complete, then $\mathrm{Pic}(X_{\Sigma })=\mathrm{Cl}(X_{\Sigma })$ is free abelian of rank exactly $|\Sigma (1)|-n$, with basis any complement of the image of $M$.

The rank formula for $\mathrm{Cl}$ is universal across all toric varieties (with the mild ray-spanning hypothesis), but the Picard group equality $\mathrm{Pic}=\mathrm{Cl}$ requires smoothness. In the singular case, $\mathrm{Pic}$ sits as a sublattice of finite index inside $\mathrm{Cl}$, and the index records the failure of Weil divisors to be Cartier.

The smooth-vs-singular dichotomy is starkly visible in weighted projective space. For $\mathbb{P}(a_0,\dots ,a_n)$ with $\gcd (a_0,\dots ,a_n)=1$, the divisor class group is $\mathrm{Cl}(\mathbb{P}(a_0,\dots ,a_n))=\mathbb{Z}$ (free of rank 1), but the Picard group is $\mathrm{Pic}(\mathbb{P}(a_0,\dots ,a_n))=\mathrm{lcm}(a_0,\dots ,a_n)\cdot \mathbb{Z}\subseteq \mathbb{Z}$, a finite-index subgroup. The index $\mathrm{lcm}(a_i)$ records the worst quotient-singularity along the toric-fixed locus, where Cartier-ness imposes divisibility constraints on the divisor's coefficients.

For non-complete fans the rank formula still applies to $\mathrm{Cl}$, but $\mathrm{Pic}$ can vanish in further interesting ways. For $X_{\Sigma }={\mathbb{C}}^n$ (the fan with one maximal cone), $|\Sigma (1)|=n$ and the character map $M\to {\mathbb{Z}}^n$ is the identity (in a suitable basis), with cokernel zero: $\mathrm{Cl}({\mathbb{C}}^n)=\mathrm{Pic}({\mathbb{C}}^n)=0$. This recovers the standard fact that affine space has vanishing Picard group, confirmed via the combinatorial recipe.

§C. Ample and nef cones via piecewise-linear support functions

Theorem (ampleness via strict convexity; Fulton §3.4, Cox-Little-Schenck Theorem 6.1.14). Let $X_{\Sigma }$ be a smooth complete toric variety. A torus-invariant Cartier divisor $D={\sum }_{\rho }{a}_{\rho }{D}_{\rho }$ is ample iff its associated piecewise-linear support function ${\psi }_D:|\Sigma |\to \mathbb{R}$ (defined by ${\psi }_D(v_{\rho })=-a_{\rho }$ and extended linearly on maximal cones) is strictly convex.

The support function dictionary translates the cohomological / projective-embedding criterion for ampleness (Cartan-Serre-Grothendieck, [04.05.05]) into a purely combinatorial criterion on the fan. The mechanism is the toric vanishing theorem of Demazure 1970 (refined by Mavlyutov 2000 Compositio Math. 124): the higher cohomology $H^i(X_{\Sigma },\mathcal{O}(D))$ decomposes into character spaces indexed by lattice points of $M$, and these vanish for $i\ge 1$ exactly when ${\psi }_D$ is strictly convex on the relevant chambers.

Theorem (nef cone as closed cone of convex support functions; Reid 1983, Cox-Little-Schenck §6.3). Let $X_{\Sigma }$ be a smooth complete toric variety. The nef cone $\mathrm{Nef}(X_{\Sigma })\subseteq \mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$ is the closure of the ample cone $\mathrm{Amp}(X_{\Sigma })$. Explicitly, $\mathrm{Nef}(X_{\Sigma })$ is the set of classes whose support function is convex (not strictly convex), which forms a closed full-dimensional rational polyhedral cone inside $\mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$.

Theorem (effective cone and Mori cone; Reid 1983). The effective cone $\mathrm{Eff}(X_{\Sigma })\subseteq \mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$ is the closed cone generated by the classes of the toric divisors $D_{\rho }$ for $\rho \in \Sigma (1)$. The Mori cone $\overline{\mathrm{NE}}(X_{\Sigma })$ of effective curve classes is the cone generated by the classes of the closures of one-dimensional torus orbits (equivalently, by walls of $\Sigma$); $\overline{\mathrm{NE}}(X_{\Sigma })$ is dual to $\mathrm{Nef}(X_{\Sigma })$ under the intersection pairing $\mathrm{Pic}\times {\mathrm{N}}_1\to \mathbb{Z}$.

The four cones $\mathrm{Amp}\subseteq \mathrm{Nef}\subseteq \mathrm{Eff}$ inside ${\mathrm{Pic}}_{\mathbb{R}}$, together with the Mori cone $\overline{\mathrm{NE}}$ in ${\mathrm{N}}_1(X_{\Sigma }{)}_{\mathbb{R}}$, are the foundational positivity cones of birational geometry. For toric varieties, all four are rational polyhedral cones with explicit generators: the effective cone is generated by toric divisors, the Mori cone by toric curves, and the nef/ample cones by their duality. This makes toric birational geometry fully algorithmic — Mori program steps reduce to subdivision operations on the fan.

§D. Mori dream spaces and the Cox-ring presentation

Theorem (toric → Mori dream space; Hu-Keel 2000). Every smooth complete toric variety $X_{\Sigma }$ is a Mori dream space: the Cox ring $$ \mathrm{Cox}(X_\Sigma) := \bigoplus_{[D] \in \mathrm{Pic}(X_\Sigma)} H^0(X_\Sigma, \mathcal{O}(D)) $$ is finitely generated as a $\mathbb{C}$-algebra, in fact equals the polynomial ring $\mathbb{C}[x_{\rho }:\rho \in \Sigma (1)]$ graded by $\mathrm{Pic}(X_{\Sigma })$ via the toric Picard exact sequence.

Hu-Keel 2000 introduced Mori dream spaces as the class of varieties for which the Minimal Model Program runs in finitely many steps with all intermediate models geometric. Their motivating examples were smooth projective toric varieties, where the Cox ring is manifestly polynomial. The grading is given by the dual of the Picard exact sequence: a monomial ${\prod }_{\rho }{x}_{\rho }^{a_{\rho }}$ has degree the class of ${\sum }_{\rho }{a}_{\rho }{D}_{\rho }$ in $\mathrm{Pic}(X_{\Sigma })$, and the $\mathrm{Pic}$-graded ring structure recovers ${⨁}_D{H}^0(X_{\Sigma },\mathcal{O}(D))$ as the sum of degree-$D$ pieces.

Theorem (Cox quotient; Cox 1995 J. Algebraic Geom. 4). Let $X_{\Sigma }$ be a toric variety with $\mathrm{Pic}(X_{\Sigma })$ free abelian (e.g., $X_{\Sigma }$ smooth, or more generally if the rays span $N_{\mathbb{R}}$ and the cokernel is torsion-free). Set $G := \mathrm{Hom}(\mathrm{Pic}(X_\Sigma), \mathbb{C}^)$,a$\mathrm{rank},\mathrm{Pic}$-dimensional algebraic torus. Then* $$ X_\Sigma = \bigl(\mathbb{C}^{\Sigma(1)} \setminus Z(\Sigma)\bigr) / G, $$ where $G$ acts on ${\mathbb{C}}^{\Sigma (1)}$ via the exact sequence $0\to M\to {\mathbb{Z}}^{\Sigma (1)}\to \mathrm{Pic}(X_{\Sigma })\to 0$ dualised to a sequence of tori, and $Z(\Sigma )$ is the irrelevant locus — the union of coordinate subspaces $V(\{x_{\rho }:\rho \in C{\}}^c)$ for $C\subseteq \Sigma (1)$ such that $\{C\}$ does not generate a cone of $\Sigma$.

Cox's 1995 paper unified the toric Picard group, the line-bundle theory of $X_{\Sigma }$, the global sections via lattice-point bases, and the GIT-quotient construction into a single $\mathrm{Pic}$-graded polynomial ring. The "homogeneous coordinate ring" perspective generalises the standard graded ring $\mathbb{C}[x_0,\dots ,x_n]$ of ${\mathbb{P}}^n$ (graded by $\mathrm{Pic}({\mathbb{P}}^n)=\mathbb{Z}$ via degree) to arbitrary toric varieties (graded by the larger Picard group). The Cox quotient presentation has been the foundation for computational toric algorithms in Macaulay2, Polymake, SageMath, and the toric Cox ring is now the standard data structure for representing toric varieties.

Theorem (chamber decomposition of the moveable cone; Hu-Keel 2000, Reid 1983). Let $X_{\Sigma }$ be a smooth complete toric variety. The moveable cone $\overline{\mathrm{Mov}}(X_{\Sigma })\subseteq \mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$ admits a finite decomposition into closures of nef cones of $\mathbb{Q}$-factorial small modifications of $X_{\Sigma }$ — equivalently, of toric varieties $X_{{\Sigma }^{\prime }}$ for ${\Sigma }^{\prime }$ a rational simplicial fan with the same rays as $\Sigma$. The decomposition records the GIT chambers of the Cox quotient under varying linearisations.

The chamber decomposition is the toric realisation of the Hu-Keel theorem that Mori dream spaces have finite GIT chambers — the various "small modifications" (flips, flops) of a smooth complete toric variety are themselves toric, and the chamber structure on $\overline{\mathrm{Mov}}$ enumerates them combinatorially. This is the toric Minimal Model Program of Reid 1983: every step reduces to a fan operation (subdivision, contraction, refinement), and the program terminates in finitely many steps to a minimal model in a finite chamber.

Synthesis. The toric Picard group is the foundational reason that line-bundle and divisor geometry on a toric variety reduces to a single short exact sequence of free abelian groups, and the central insight is that the entire positivity calculus — ample cone, nef cone, effective cone, Mori cone, Cox ring — flows from the same combinatorial data: the rays $\Sigma (1)$, the character lattice $M$, and the piecewise-linear support functions on the fan. Putting these together with the polytope-fan correspondence of [04.11.10], the toric Picard exact sequence is the bridge between the lattice combinatorics of $\Sigma$ and the global algebraic geometry of $X_{\Sigma }$.

This is exactly the structural fact that identifies toric geometry with algorithmic algebraic geometry: where a general variety's birational geometry requires Mori program steps that can be obstructed by infinitely many flips, on a toric variety the program terminates in a finite chamber decomposition of $\overline{\mathrm{Mov}}$ by Hu-Keel 2000. The bridge is the Cox-ring presentation: every smooth complete toric $X_{\Sigma }$ is a GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))/G$ where $G=\mathrm{Hom}(\mathrm{Pic},{\mathbb{C}}^{\ast })$, and the chamber decomposition records the GIT linearisations.

The pattern recurs across three generalisations. To non-smooth toric varieties, the same exact sequence describes $\mathrm{Cl}(X_{\Sigma })$ but the Picard group sits as a sublattice of finite index — the $\mathbb{Q}$-factorial case is when the index is well-behaved, and the simplicial case is when every cone has $\dim$-many rays (generalising smooth = simplicial-unimodular). To toric stacks (Borisov-Chen-Smith 2005), the Picard group is replaced by the stacky Picard group, which captures the full Cartier-Weil discrepancy and recovers the smooth case at the level of the underlying scheme but with extra information from the stabiliser groups at the singular fixed points. To logarithmic / equivariant toric (Kato 1989, Vistoli 1989, Gillet 1984), the Picard group enriches with logarithmic / equivariant data, and the exact sequence becomes a long exact sequence of equivariant cohomology groups. Each generalisation preserves the cone-to-Picard dictionary developed here, identifies $X_{\Sigma }$ with the appropriate refined quotient, and generalises the rank formula by replacing $|\Sigma (1)|$ with the appropriate generator count.

Full proof set [Master]

Proposition (toric Picard exact sequence), proof. Given in the key theorem. Surjectivity of ${\mathrm{Div}}_T\to \mathrm{Cl}$ uses the fact that every Weil divisor on a normal $T$-variety is linearly equivalent to a torus-invariant one (averaging over the torus). The image of $\mathrm{div}$ is exactly the principal torus-invariant divisors, coming from characters ${\chi }^m$ for $m\in M$. Injectivity of $\mathrm{div}$ when the rays span $N_{\mathbb{R}}$ uses non-degeneracy of the pairing $M\times N\to \mathbb{Z}$. Smoothness implies $\mathrm{Pic}=\mathrm{Cl}$ because regular local rings are UFDs, so every height-one prime is principal. $\square$

Proposition (rank formula for smooth complete), proof. Given in Exercise 7. Smith normal form of the matrix $(⟨e_i^{\ast },v_{{\rho }_j}⟩)$ has elementary divisors all equal to 1 on the part restricted to any maximal cone, because the primitive ray generators of a smooth cone form a $\mathbb{Z}$-basis of $N$. Hence the cokernel is torsion-free, and rank-additivity gives $|\Sigma (1)|-n$. $\square$

Proposition (Pic of $\mathbb{P}(1,2,3)$ singular case), proof. Given in Exercise 4. The character map has matrix $\left(\begin{array}{cc}1& 0\\ 0& 1\\ -2& -3\end{array}\right)$, with Smith normal form $\mathrm{diag}(1,1)$ truncated by zero, giving cokernel $\mathbb{Z}$. The Cartier condition restricts to multiples of $\mathrm{lcm}(2,3)=6$ at the singular fixed points, so $\mathrm{Pic}=6\mathbb{Z}$ inside $\mathrm{Cl}=\mathbb{Z}$. $\square$

Proposition (ampleness ↔ strict convexity), proof sketch. On a smooth complete toric $X_{\Sigma }$, a torus-invariant divisor $D={\sum }_{\rho }{a}_{\rho }{D}_{\rho }$ is ample iff its support function ${\psi }_D$ is strictly convex on $|\Sigma |$.

Proof. ($\Rightarrow$) Ample $D$ implies via the Cartan-Serre-Grothendieck criterion ([04.05.05]) that $H^i(X_{\Sigma },\mathcal{O}(nD))=0$ for $i\ge 1$ and $n\gg 0$. The toric vanishing theorem of Demazure 1970 (refined by Mavlyutov 2000) decomposes $H^i(X_{\Sigma },\mathcal{O}(D))={⨁}_{m\in M}{H}^i(\Sigma ,{\mathcal{F}}_{D,m})$ where ${\mathcal{F}}_{D,m}$ is a combinatorial sheaf on the cone complex of $\Sigma$ determined by the relative position of $m$ to the polytope-of-sections. Vanishing of these for every $i\ge 1$ forces ${\psi }_D$ to be strictly convex (the only $m$ contributing to $i=0$ are interior lattice points of the polytope, and the polytope has full dimension iff ${\psi }_D$ is strictly convex).

($\Leftarrow$) Conversely, strict convexity of ${\psi }_D$ gives a lattice polytope $P_D=\{m\in M_{\mathbb{R}}:⟨m,v⟩\ge {\psi }_D(v)\text{ for all }v\in |\Sigma |\}$ with $\Sigma$ as its normal fan. By the projectivity criterion of [04.11.04], $X_{\Sigma }$ is projective and $\mathcal{O}(D)=L_{P_D}$ is the polarisation associated to $P_D$. Lattice points of $P_D$ form a basis of $H^0(X_{\Sigma },\mathcal{O}(D))$, the associated rational map is a closed embedding for $P_D$ sufficiently scaled, and $D$ is therefore ample. $\square$

Proposition (Mori dream space), proof sketch. Every smooth complete toric variety $X_{\Sigma }$ has finitely generated Cox ring.

Proof. The Cox ring of $X_{\Sigma }$ is $\mathrm{Cox}(X_{\Sigma })={⨁}_{[D]\in \mathrm{Pic}(X_{\Sigma })}{H}^0(X_{\Sigma },\mathcal{O}(D))$. For toric $X_{\Sigma }$, $H^0(X_{\Sigma },\mathcal{O}(D))$ has basis the lattice points of the polytope $P_D$ (when $D$ is nef; for non-nef $D$, $H^0=0$). Cox 1995 shows the direct sum equals the polynomial ring $\mathbb{C}[x_{\rho }:\rho \in \Sigma (1)]$ in $|\Sigma (1)|$ variables, with the $\mathrm{Pic}$-grading from the toric Picard exact sequence. Polynomial rings in finitely many variables are finitely generated by definition, completing the proof. $\square$

Proposition (Cox quotient), proof sketch. For $X_{\Sigma }$ with $\mathrm{Pic}(X_{\Sigma })$ free abelian, $X_{\Sigma }=({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))/G$ with $G = \mathrm{Hom}(\mathrm{Pic}, \mathbb{C}^)$.*

Proof. The polynomial ring $S:=\mathbb{C}[x_{\rho }:\rho \in \Sigma (1)]$ is naturally graded by $\mathrm{Pic}(X_{\Sigma })$ via the toric Picard exact sequence: a monomial ${\prod }_{\rho }{x}_{\rho }^{a_{\rho }}$ has $\mathrm{Pic}$-degree the class of ${\sum }_{\rho }{a}_{\rho }{D}_{\rho }$. Dualising $0\to M\to {\mathbb{Z}}^{\Sigma (1)}\to \mathrm{Pic}\to 0$ gives the exact sequence of tori $1\to G\to ({\mathbb{C}}^{\ast }{)}^{\Sigma (1)}\to T\to 1$, with $G=\mathrm{Hom}(\mathrm{Pic},{\mathbb{C}}^{\ast })$ acting on ${\mathbb{C}}^{\Sigma (1)}=\mathrm{Spec}S$ through the dual of the grading.

The irrelevant locus $Z(\Sigma )$ is defined as the union of coordinate-subspaces $V(\{x_{\rho }:\rho \in C\})$ where $C\subseteq \Sigma (1)$ is a "non-cone" — a subset of rays not contained in any single cone of $\Sigma$. By construction, ${\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma )$ is the open subset of points whose support (= set of nonzero coordinates) corresponds to a cone of $\Sigma$.

The GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))//G$ in the stability sense identifies orbits under $G$, and the quotient inherits a $T$-action with a dense open orbit. Sumihiro's classification ([04.11.04]) recovers a fan ${\Sigma }^{\prime }$ for the quotient; explicit verification on each chart shows ${\Sigma }^{\prime }=\Sigma$. The quotient is therefore $X_{\Sigma }$. $\square$

Proposition (chamber decomposition of moveable cone), proof sketch. For smooth complete toric $X_{\Sigma }$, the moveable cone $\overline{\mathrm{Mov}}(X_{\Sigma })\subseteq {\mathrm{Pic}}_{\mathbb{R}}$ has a finite decomposition into closures of nef cones of small $\mathbb{Q}$-factorial modifications.

Proof. The Cox-quotient presentation identifies $X_{\Sigma }$ as a GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))//G$. Varying the GIT linearisation $\theta \in \mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$ produces a family of GIT quotients $X_{\theta }:=({\mathbb{C}}^{\Sigma (1)}{)}^{\theta \text{-ss}}//G$. The unstable locus depends on $\theta$ in a piecewise-constant way: as $\theta$ varies in $\mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$, the unstable locus stays constant on chambers and jumps at walls. The chamber decomposition of $\overline{\mathrm{Mov}}(X_{\Sigma })$ into these GIT chambers is finite (Hu-Keel 2000, Theorem 3.7) because the set of possible unstable loci is finite (parametrised by subsets of $\Sigma (1)$).

In each chamber, the GIT quotient is a smooth $\mathbb{Q}$-factorial toric variety (different from $X_{\Sigma }$ but birational to it via the small modifications corresponding to wall-crossings). The interior of each chamber is the ample cone of the corresponding GIT quotient; the closures tile $\overline{\mathrm{Mov}}$. $\square$

Connections [Master]

• Fan and toric variety $X_{\Sigma }$ 04.11.04. The structural prerequisite. The toric Picard exact sequence is a statement about the variety $X_{\Sigma }$ constructed from a fan $\Sigma$ in the previous unit; every ingredient — the rays $\Sigma (1)$, the character lattice $M$, the cocharacter lattice $N$, the maximal cones supplying affine charts — comes from that construction. The smoothness criterion ($\mathrm{Pic}=\mathrm{Cl}$) and the completeness criterion (rays span $N_{\mathbb{R}}$, so $\mathrm{div}$ is injective) are conditions on the fan inherited directly from [04.11.04].

• Ample and very ample line bundle 04.05.05. The toric ampleness criterion of §C specialises the general Cartan-Serre-Grothendieck criterion from [04.05.05] to the toric setting. The strict-convexity condition on the support function ${\psi }_D$ is the combinatorial signature of ampleness; the toric vanishing theorem of Demazure 1970 (refined by Mavlyutov 2000) supplies the higher-cohomology-vanishing input that Cartan-Serre-Grothendieck demands. The ample cone, nef cone, and Mori cone defined in [04.05.05] become rational polyhedral cones in the toric Picard group $\mathrm{Pic}(X_{\Sigma }{)}_{\mathbb{R}}$, all of which are computable combinatorially from the fan.

• Toric divisor and support function 04.11.08. The pending sibling unit (in this batch). Defines $D_{\rho }$ as the closure of a one-dimensional torus orbit, develops the Cartier-vs-Weil distinction in detail, and treats the support-function ${\psi }_D$ formalism that powers the ampleness criterion of §C. The toric Picard exact sequence is the next step after the divisor definitions: once one has ${\mathrm{Div}}_T(X_{\Sigma })={⨁}_{\rho }\mathbb{Z}{D}_{\rho }$ as the free abelian group on toric divisors, the Picard group is the cokernel of the character map.

• Polytope-fan dictionary 04.11.10. The downstream unit. Every ample divisor on a projective toric variety $X_{\Sigma }$ has a polytope $P_D\subseteq M_{\mathbb{R}}$ whose normal fan is $\Sigma$, and the dictionary $P\leftrightarrow (X_{\Sigma },L_P)$ is the projective-toric special case of the toric Picard group described here. The lattice points $P_D\cap M$ form a basis of $H^0(X_{\Sigma },\mathcal{O}(D))$, making the cohomology of polarised toric varieties manifestly combinatorial.

• Algebraic torus and character/cocharacter lattices 04.11.01. The duality $M\leftrightarrow N$ underpinning the character map. The character map $\mathrm{div}:M\to {\mathrm{Div}}_T(X_{\Sigma })$ uses the perfect pairing $M\times N\to \mathbb{Z}$ to evaluate characters on ray generators. Non-degeneracy of the pairing is the foundational reason that $\mathrm{div}$ is injective when the rays span $N_{\mathbb{R}}$.

• Orbit-cone correspondence 04.11.06. The sibling unit. The toric divisor $D_{\rho }$ corresponding to a ray $\rho$ is the closure of the codimension-one torus orbit $O_{\rho }$ under the orbit-cone correspondence: orbits of $T$ on $X_{\Sigma }$ are in inclusion-reversing bijection with cones of $\Sigma$, with orbit dimension equal to $n-\dim \sigma$. Divisors are exactly the orbits of dimension $n-1$, i.e., the rays.

• Smoothness and completeness via fans 04.11.05. The sibling unit supplies the global smoothness and completeness criteria for $X_{\Sigma }$ that gate the Picard-group statements proved here. Smoothness ($\mathrm{Pic}(X_{\Sigma })=\mathrm{Cl}(X_{\Sigma })$ free abelian of rank $|\Sigma (1)|-n$) requires every cone simplicial unimodular by the smoothness criterion of [04.11.05]; completeness ($\mathrm{div}:M\to {⨁}_{\rho }\mathbb{Z}{D}_{\rho }$ injective, so $\mathrm{Pic}$ is the full cokernel) requires $|\Sigma |=N_{\mathbb{R}}$ by the completeness criterion. The Picard exact sequence is sharpest when both criteria hold.

• Toric resolution of singularities 04.11.07. The sibling unit gives a procedure to upgrade an arbitrary toric variety into a smooth one via fan refinement, and the Picard-group behaviour under this procedure is structural: each star subdivision adds one ray $\rho$, hence one toric divisor $D_{\rho }$, hence one $\mathbb{Z}$-summand to $\mathrm{Pic}(X_{{\Sigma }^{\prime }})/\mathrm{Pic}(X_{\Sigma })=\mathbb{Z}$ — the exceptional divisor class of the blow-up. The pullback map on Picard groups under a toric resolution is computable purely from the fan data developed here.

• Toric cohomology and the Chow ring 04.11.12. The downstream unit. The Chow ring $A^{\ast }(X_{\Sigma })$ of a smooth complete toric variety is the Stanley-Reisner ring modulo the linear relations from the toric Picard exact sequence. The toric Picard group provides the first-grade piece $A^1(X_{\Sigma })=\mathrm{Pic}(X_{\Sigma })$, and the Stanley-Reisner relations describe the higher products. The intersection theory of toric varieties is fully combinatorial because the Chow ring is finitely presented from fan data.

• Cox-ring and GIT quotient 04.11.15 pending. The toric variety $X_{\Sigma }$ presented as the GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))/G$ for $G=\mathrm{Hom}(\mathrm{Pic}(X_{\Sigma }),{\mathbb{C}}^{\ast })$, with the polynomial ring $\mathbb{C}[x_{\rho }:\rho \in \Sigma (1)]$ as the homogeneous coordinate ring graded by $\mathrm{Pic}$. The toric Picard exact sequence supplies the grading; the chamber decomposition of the moveable cone supplies the variation-of-GIT-quotient structure. Together these give the full Cox-ring presentation of toric geometry.

• Mikhalkin correspondence theorem 04.12.05. The toric surface $X_{\Sigma }$ from a Newton polygon $\Delta \subseteq {\mathbb{R}}^2$ has Picard group ${\mathbb{Z}}^{|\Sigma (1)|-2}$ — one less generator than the number of edges of $\Delta$ — and the ample line bundle on $X_{\Sigma }$ for an enumerative count of curves of Newton polygon $\Delta$ is the polarisation $L_{\Delta }$ whose Newton polytope is $\Delta$. The Picard group's rank controls the number of constraints in the tropical enumeration: a generic configuration of $r=|\Delta \cap {\mathbb{Z}}^2|-1-g$ points fixes the configuration up to a finite count, with $r$ matching the genus-modified dimension via Picard-group considerations.

• Hirzebruch-Riemann-Roch on toric 04.11.13. The downstream unit. The Hirzebruch-Riemann-Roch theorem for a line bundle $\mathcal{O}(D)$ on a smooth complete toric $X_{\Sigma }$ produces an integer-valued Euler characteristic computable from the polytope-lattice-point count, with the toric Picard group providing the line-bundle classification. Brion-Vergne 1997 (Invent. Math. 128) gave the explicit toric Hirzebruch-Riemann-Roch formula via equivariant cohomology, with the Picard group's combinatorial description making the formula effective.

• Picard group of a general variety 04.05.02. The toric Picard group is a sharp combinatorial special case of the general Picard-group theory of [04.05.02]. Where the general Picard group requires Hodge theory, the Néron-Severi group, and the Albanese variety to control, the toric Picard group reduces to a single short exact sequence of free abelian groups. The toric setting is therefore the canonical testing ground for understanding Picard-group computations and a benchmark for the more elaborate machinery needed in the non-toric case.

Historical & philosophical context [Master]

The toric Picard group in its modern form is due to Michel Demazure in Sous-groupes algébriques de rang maximum du groupe de Cremona (Annales scientifiques de l'École normale supérieure (4) 3, 1970, pp. 507-588) ^{[Demazure 1970]}, appearing alongside the cone-fan-variety construction reviewed in [04.11.04]. Demazure's paper established the exact sequence $0\to M\to {\mathrm{Div}}_T\to \mathrm{Pic}\to 0$ for smooth complete toric varieties, the rank formula $|\Sigma (1)|-n$, the smoothness criterion expressed in fan-theoretic language, and the description of ample line bundles via strictly convex piecewise-linear support functions. The terminology "groupe de Picard d'un espace torique" ("Picard group of a toric space") is Demazure's coinage.

The exact sequence in its Cartier-Weil refined form (distinguishing $\mathrm{Pic}$ from $\mathrm{Cl}$ on singular toric) was systematised in Toroidal Embeddings I by Kempf, Knudsen, Mumford, and Saint-Donat (Lecture Notes in Mathematics 339, Springer 1973) ^[pending], where the polytope-and-fan formalism was developed as an exposition of Demazure's results within the broader scheme-theoretic toric framework. KKMS Chapter I §2 contains the first systematic exposition of the toric divisor class group on singular toric (the lower exact sequence of §A of this unit).

Hideyasu Sumihiro's Equivariant completion (Journal of Mathematics of Kyoto University 14, 1974, pp. 1-28) ^{[Sumihiro 1974]} supplied the foundational equivariant-covering theorem that makes the toric-divisor theory work globally: every normal $T$-variety admits a $T$-equivariant affine cover. Without Sumihiro's theorem, one would only have a local affine-toric divisor calculus; with it, the calculus extends to global $X_{\Sigma }$ and the Picard group becomes computable.

Tadao Oda's Convex Bodies and Algebraic Geometry (Ergebnisse 15, Springer 1988) ^[pending] is the canonical mid-1980s monograph, with §2.1–§2.3 the standard reference for the toric Picard group, divisor class group, and ample line bundles. Oda's exposition emphasises the polytope-fan dictionary and was the main pre-Fulton reference. Vasily Danilov's earlier 1978 survey The geometry of toric varieties (Russian Mathematical Surveys 33(2), 97-154) ^{[Danilov 1978]} gave the first English-language treatment of the toric Picard group and developed the Stanley-Reisner perspective on the Chow ring.

William Fulton's Introduction to Toric Varieties (Princeton University Press 1993) ^{[Fulton 1993]} §3.3–§3.4 is the canonical short textbook reference for the toric Picard exact sequence, the rank formula, and the ample-line-bundle / strictly-convex-support-function dictionary. Fulton's exposition is the standard pedagogical anchor for the material of this unit. David Cox, John Little, and Henry Schenck's Toric Varieties (American Mathematical Society 2011) ^[pending] is the modern thousand-page treatment, extending Fulton's exposition with extensive material on the Cox ring (Chapter 5), nef and Mori cones (Chapter 6), and the GIT-quotient perspective (Chapter 14).

David Cox's 1995 paper The homogeneous coordinate ring of a toric variety (Journal of Algebraic Geometry 4, 17-50) ^{[Cox 1995]} reframed the toric Picard group as the grading of the Cox ring $\mathbb{C}[x_{\rho }:\rho \in \Sigma (1)]$, presenting $X_{\Sigma }$ as a GIT quotient $({\mathbb{C}}^{\Sigma (1)}\setminus Z(\Sigma ))/G$. The Cox-ring perspective unifies projective and toric geometry under a single functorial framework and has been the foundation for modern computational toric algorithms in Macaulay2, Polymake, and SageMath, as well as the toric instances of the Minimal Model Program.

Yi Hu and Sean Keel's Mori dream spaces and GIT (Michigan Mathematical Journal 48, 2000, pp. 331-348) ^[pending] introduced the class of Mori dream spaces — varieties whose Cox ring is finitely generated and whose moveable cone admits a finite GIT chamber decomposition. Their motivating examples were smooth projective toric varieties, where the Cox ring is manifestly polynomial. Hu-Keel's theorem made the toric Picard group the foundational example of well-behaved birational geometry: every smooth complete toric variety is a Mori dream space, with the Minimal Model Program terminating in finitely many fan-modification steps.

Anvar Mavlyutov's Semiample hypersurfaces in toric varieties (Compositio Mathematica 124(1), 2000, pp. 7-24) ^[pending] and the subsequent series of papers refined the toric vanishing theorem and the ample / nef cone descriptions for singular toric varieties, extending Demazure's smooth-case results to the simplicial and $\mathbb{Q}$-factorial settings. Mavlyutov's work is the canonical reference for the toric Picard-Weil discrepancy in singular cases.

Miles Reid's Decomposition of toric morphisms (in Arithmetic and Geometry II, Progress in Mathematics 36, Birkhäuser 1983, pp. 395-418) ^[pending] established the toric Mori-cone description and the contraction theorem for extremal rays in the toric setting, anticipating the general Minimal Model Program by several years. Reid's toric analysis was the proof-of-concept that motivated the broader development of Mori theory under Kawamata, Kollár, Mori, Reid, and Shokurov in the late 1980s.

Friedrich Hirzebruch's introduction of the surfaces $F_a$ in Über eine Klasse von einfach-zusammenhängenden komplexen Mannigfaltigkeiten (Mathematische Annalen 124, 1951, pp. 77-86) ^{[Hirzebruch 1951]} predates the toric formalism by two decades but supplied the prototype example of a rank-2 toric Picard group, with the explicit basis $\{F,S\}$ of fibre and section classes recovered toric-geometrically by Demazure 1970. The Hirzebruch surfaces remain the standard pedagogical example of a substantively-twisted toric Picard group beyond ${\mathbb{P}}^n$ and ${\mathbb{P}}^a\times {\mathbb{P}}^b$.

Bibliography [Master]

@article{Demazure1970,
  author  = {Demazure, Michel},
  title   = {Sous-groupes alg{\'e}briques de rang maximum du groupe de Cremona},
  journal = {Annales scientifiques de l'{\'E}cole normale sup{\'e}rieure (4)},
  volume  = {3},
  year    = {1970},
  pages   = {507--588}
}

@book{KempfKnudsenMumfordSaintDonat1973,
  author    = {Kempf, George and Knudsen, Finn Faye and Mumford, David and Saint-Donat, Bernard},
  title     = {Toroidal Embeddings I},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {339},
  year      = {1973}
}

@article{Sumihiro1974,
  author  = {Sumihiro, Hideyasu},
  title   = {Equivariant completion},
  journal = {Journal of Mathematics of Kyoto University},
  volume  = {14},
  year    = {1974},
  pages   = {1--28}
}

@article{Danilov1978,
  author  = {Danilov, Vasily I.},
  title   = {The geometry of toric varieties},
  journal = {Russian Mathematical Surveys},
  volume  = {33},
  number  = {2},
  year    = {1978},
  pages   = {97--154}
}

@book{OdaConvexBodies,
  author    = {Oda, Tadao},
  title     = {Convex Bodies and Algebraic Geometry: An Introduction to the Theory of Toric Varieties},
  publisher = {Springer-Verlag},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete},
  volume    = {15},
  year      = {1988}
}

@book{FultonToric,
  author    = {Fulton, William},
  title     = {Introduction to Toric Varieties},
  publisher = {Princeton University Press},
  series    = {Annals of Mathematics Studies},
  volume    = {131},
  year      = {1993}
}

@book{CoxLittleSchenck,
  author    = {Cox, David A. and Little, John B. and Schenck, Henry K.},
  title     = {Toric Varieties},
  publisher = {American Mathematical Society},
  series    = {Graduate Studies in Mathematics},
  volume    = {124},
  year      = {2011}
}

@article{Cox1995,
  author  = {Cox, David A.},
  title   = {The homogeneous coordinate ring of a toric variety},
  journal = {Journal of Algebraic Geometry},
  volume  = {4},
  year    = {1995},
  pages   = {17--50}
}

@article{HuKeel2000,
  author  = {Hu, Yi and Keel, Sean},
  title   = {Mori dream spaces and {GIT}},
  journal = {Michigan Mathematical Journal},
  volume  = {48},
  year    = {2000},
  pages   = {331--348}
}

@article{Mavlyutov2000,
  author  = {Mavlyutov, Anvar R.},
  title   = {Semiample hypersurfaces in toric varieties},
  journal = {Compositio Mathematica},
  volume  = {124},
  number  = {1},
  year    = {2000},
  pages   = {7--24}
}

@incollection{Reid1983,
  author    = {Reid, Miles},
  title     = {Decomposition of toric morphisms},
  booktitle = {Arithmetic and Geometry, Volume II},
  publisher = {Birkh{\"a}user},
  series    = {Progress in Mathematics},
  volume    = {36},
  year      = {1983},
  pages     = {395--418}
}

@article{Hirzebruch1951,
  author  = {Hirzebruch, Friedrich},
  title   = {{\"U}ber eine {K}lasse von einfach-zusammenh{\"a}ngenden komplexen {M}annigfaltigkeiten},
  journal = {Mathematische Annalen},
  volume  = {124},
  year    = {1951},
  pages   = {77--86}
}

@article{BrionVergne1997,
  author  = {Brion, Michel and Vergne, Mich{\`e}le},
  title   = {An equivariant {R}iemann-{R}och theorem for complete, simplicial toric varieties},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {482},
  year    = {1997},
  pages   = {67--92}
}

@article{FultonSturmfels1997,
  author  = {Fulton, William and Sturmfels, Bernd},
  title   = {Intersection theory on toric varieties},
  journal = {Topology},
  volume  = {36},
  year    = {1997},
  pages   = {335--353}
}