Toric foliations — lattice, wall data, foliated length, and -bundle data
DefinitionToricFoliationsdefinition_name: ToricFoliations
definition_title: Toric foliations — lattice, wall data, foliated length, and -bundle data
Define the toric combinatorial core used in the proof of Theorem 1.3 of Fujino–Sato.
Fix a dimension . The lattice is and its complexification is ; a toric foliation is represented, via Pang's correspondence, by a -subspace of rank .
For a torus-invariant curve of a complete simplicial fan, let
be the primitive vectors appearing in Reid's description
of the wall and of the two adjacent maximal cones . The
type WallData n packages:
- the vectors and the integers satisfying the wall relation of equation (3.1), with , with the ordering , and with (stated as a Bézout identity);
- the multiplicities , assumed positive with and .
divisorIntersection implements the intersection numbers from equation (3.1):
curveLength is the foliated length
; ExtremalRay and
rayLength package an abstract extremal ray and the minimum
;
IsProjectiveSpaceBundle encodes the conclusion of Theorem 1.3(2): the tail
vectors sum to zero, span a saturated rank- sublattice, and give a lattice
splitting — the toric data of a
-bundle.
Formalization Note. Lattice n := Fin n → ℤ, Complexified n := Fin n → ℂ,
V is a Submodule ℂ (Complexified n), and the rank is
Module.finrank ℂ V. Division by zero is total in ℚ, so no side conditions
are needed inside divisorIntersection.
import Mathlib
open scoped BigOperators
/-!
# Toric foliations: combinatorial core of Fujino--Sato, "A remark on toric foliations"
This file formalizes the *combinatorial core* of Theorem 1.3 of
Osamu Fujino and Hiroshi Sato,
*A remark on toric foliations*,
Arch. Math. 122 (2024), 621--627.
https://doi.org/10.1007/s00013-024-01991-1
The full statement of Theorem 1.3 lives in toric Mori theory (projective
`Q`-factorial toric varieties, saturated torus-equivariant subsheaves of the
tangent sheaf, the Mori cone, extremal contraction morphisms). Mathlib does
not yet contain that machinery, so this mission fixes the combinatorial data
the proof of Section 3 actually uses:
* a lattice `N = Z^n` together with its complexification `N_C = C^n`;
* a wall `W` of a complete simplicial fan, i.e. a relation
`sum_i a_i v_i = 0` among the `n+1` primitive vectors `v_1,...,v_{n+1}`
appearing in Reid's description of toric extremal contractions;
* a toric foliation, represented (via Pang's correspondence) by a complex
subspace `V <= N_C`;
* the intersection numbers `D_{v_i} . C` computed by equation (3.1) of the
paper, which build the foliated length `-K_F . C`.
The geometric wrapper (sheaves, Mori cone, contraction morphisms, relative
tangent sheaf) is deliberately left as future work; the goal theorem is the
faithful lattice/fan content of Theorem 1.3.
-/
namespace ToricFoliations
/-- The lattice `N = Z^n` of a toric variety of dimension `n`. -/
abbrev Lattice (n : ℕ) := Fin n → ℤ
/-- The complexification `N_C = C^n`. -/
abbrev Complexified (n : ℕ) := Fin n → ℂ
/-- The natural inclusion `N -> N_C`. -/
def toComplex {n : ℕ} (x : Lattice n) : Complexified n :=
fun i => (x i : ℂ)
/-- The index of `v_{n+1}` among the `n+1` vectors `v_1,...,v_{n+1}`.
(We use 0-based `Fin (n+1)` indices, so `lastIndex n = n`.) -/
def lastIndex (n : ℕ) : Fin (n + 1) :=
⟨n, Nat.lt_succ_self n⟩
/-- The index of `v_n` among the `n+1` vectors (0-based index `n-1`). -/
def wallIndex (n : ℕ) (hn : 0 < n) : Fin (n + 1) :=
⟨n - 1, by omega⟩
/--
The wall data of a torus-invariant curve `C = V(W)` in a complete simplicial
fan, following equations (3.1)--(3.3) and Proposition 14-1-5 of Matsuki's
book as used in the proof of Theorem 1.3.
`v : Fin (n+1) -> N` lists the primitive vectors; `a : Fin (n+1) -> Z` are the
integers in the wall relation `sum_i a_i v_i = 0`; `mWall`, `mSigma`,
`mSigma'` are the multiplicities of the wall `W` and of the two maximal cones
`sigma`, `sigma'` adjacent to `W`.
-/
structure WallData (n : ℕ) where
hn : 0 < n
v : Fin (n + 1) → Lattice n
a : Fin (n + 1) → ℤ
/-- The wall relation (3.1): `sum_i a_i v_i = 0`. -/
relation : (∑ i : Fin (n + 1), a i • v i) = 0
/-- `a_n > 0` (paper index `n`, Lean index `n-1`). -/
an_pos : 0 < a (wallIndex n hn)
/-- `a_{n+1} > 0`. -/
aLast_pos : 0 < a (lastIndex n)
/-- After reordering, (3.2): `a_1 <= ... <= a_{n+1}`. -/
sorted : ∀ i j : Fin (n + 1), i.val ≤ j.val → a i ≤ a j
/-- Multiplicity of the wall `W`. -/
mWall : ℕ
/-- Multiplicity of the maximal cone `sigma`. -/
mSigma : ℕ
/-- Multiplicity of the maximal cone `sigma'`. -/
mSigma' : ℕ
mWall_pos : 0 < mWall
mSigma_pos : 0 < mSigma
mSigma'_pos : 0 < mSigma'
/-- `mult(W) | mult(sigma)`, used for `D . C <= 1`. -/
wall_dvd_sigma : mWall ∣ mSigma
/-- `mult(W) | mult(sigma')`, used for `D . C <= 1`. -/
wall_dvd_sigma' : mWall ∣ mSigma'
/-- `gcd(a_1,...,a_{n+1}) = 1` via a Bézout identity. -/
gcd_one : ∃ b : Fin (n + 1) → ℤ, (∑ i : Fin (n + 1), b i * a i) = 1
/--
The intersection number `D_{v_i} . C` from equation (3.1) of the paper:
for `i = n+1` it is `mult(W)/mult(sigma')`, and for every other index it is
`a_i * mult(W) / (a_n * mult(sigma))`. Division by zero is total in `Q`,
so no side conditions are needed in the definition.
-/
noncomputable def divisorIntersection {n : ℕ} (C : WallData n) (i : Fin (n + 1)) : ℚ :=
if i = lastIndex n then
(C.mWall : ℚ) / (C.mSigma' : ℚ)
else
(C.a i : ℚ) * (C.mWall : ℚ) / ((C.a (wallIndex n C.hn) : ℚ) * (C.mSigma : ℚ))
/--
The foliated length `-K_F . C = sum_{v_i in V} D_{v_i} . C` of a
torus-invariant curve `C`, for the toric foliation represented by the complex
subspace `V` (see equation (1.1)--(1.2) and the proof of Theorem 1.3).
-/
noncomputable def curveLength {n : ℕ} (C : WallData n)
(V : Submodule ℂ (Complexified n)) : ℚ := by
classical
exact ∑ i : Fin (n + 1), if toComplex (C.v i) ∈ V then divisorIntersection C i else 0
/--
An abstract extremal ray of the Mori cone: a set of torus-invariant curves
plus an (intentionally opaque) extremality predicate. Making `isExtremal`
concrete is part of the mission's future work of connecting this combinatorial
core to toric Mori theory.
-/
structure ExtremalRay (n : ℕ) where
/-- The torus-invariant curves whose classes span the ray. -/
curves : Finset (WallData n)
nonempty : curves.Nonempty
isExtremal : Prop
/--
The length `l_F(R) = min_{[C] in R} (-K_F . C)` of an extremal ray with
respect to the toric foliation represented by `V`.
-/
noncomputable def rayLength {n : ℕ} (R : ExtremalRay n)
(V : Submodule ℂ (Complexified n)) : ℚ := by
classical
exact R.curves.inf' R.nonempty (fun C : WallData n => curveLength C V)
/--
The index of the paper's `v_{n-r+j}` (1-based) inside `Fin (n+1)`: for
`0 <= j <= r` this is `n-r+j`. These are the tail indices that appear in
equations (3.5)--(3.6) once the long-ray case has been analyzed.
-/
def tailIndex {n r : ℕ} (hr : r ≤ n) (j : Fin (r + 1)) : Fin (n + 1) :=
⟨n - r + j.val, by
have hj : j.val ≤ r := Nat.le_of_lt_succ j.isLt
omega⟩
/--
Combinatorial encoding of the conclusion of Theorem 1.3(2): the extremal
contraction is a `P^r`-bundle. In toric geometry this is equivalent to the
lattice splitting `N = N1 (+) N2` with `rank N1 = r`, where the tail vectors
`v_{n-r+1},...,v_{n+1}` span the saturated sublattice `N1` and satisfy
`v_{n-r+1} + ... + v_{n+1} = 0` (equations (3.10)--(3.11) and Fulton's
exercise quoted in the paper).
-/
structure IsProjectiveSpaceBundle {n r : ℕ} (hr : r ≤ n) (C : WallData n) : Prop where
/-- `v_{n-r+1} + ... + v_{n+1} = 0` (equation (3.10)). -/
tail_sum_zero : (∑ j : Fin (r + 1), C.v (tailIndex hr j)) = 0
/-- The tail vectors span a rank-`r` saturated sublattice (3.11). -/
tail_span :
Nonempty
((Submodule.span ℤ (Set.range fun j : Fin (r + 1) => C.v (tailIndex hr j)) :
Submodule ℤ (Lattice n)) ≃ₗ[ℤ] (Fin r → ℤ))
/-- The splitting `N = N1 (+) N2` giving the `P^r`-bundle structure. -/
lattice_split : Nonempty (Lattice n ≃ₗ[ℤ] ((Fin r → ℤ) × (Fin (n - r) → ℤ)))
end ToricFoliationsRead-back
What the Lean code literally says, in plain math · claude-haiku-4-5
Fix . The module defines the toric lattice as functions , its complexification , and the coordinatewise inclusion . The index inside is called the last index (the -st vector, ); under the hypothesis , is the "wall index" (the paper's -th vector). A structure consists of a proof , vectors , integers , the wall relation in , and , monotonicity whenever (so ), natural multiplicities , all , divisibilities and , and integers with . Because is a field, is empty. For such and index , the intersection number is if , and otherwise in ; the stated hypotheses make the latter denominator nonzero. For a complex subspace , the foliated curve length is over indices with the complexification of lying in , and for the rest. An is a nonempty finite set of such wall data together with an opaque proposition ; its ray length is the minimum of the curve lengths over that finite set. With and , the tail index is inside ; note yields only index , and yields . Finally, asserts: (i) the tail sum in ; (ii) the -span of those tail vectors is -linearly equivalent to ; (iii) there is a -linear equivalence .
WallData leaves mWall, mSigma, mSigma' as free naturals unrelated to the vectors vi, does not require the v_i to be primitive or nonzero, and ExtremalRay.isExtremal is an arbitrary Prop. The paper's argument (cones sigmak, curves V(muk,n+1),mult(muk,n+1)=mult(sigmak), primitivity of vn−r+1) cannot be expressed over this data. The dependent statements below are false in this abstraction.