Documentation

LeanPool.BrillNoetherGraphs.Bananas.Theta.ThetaInvTauCorrection

The exact genus-two inversion formula, with its correction term #

ThetaGenusTwoCornerSum.lean reduces inv_k(τ_D) to three finite degree slices and then evaluates them under the rigidity hypothesis u + v ≁ K_G, which makes the residual degree-two product vanish pointwise. This file evaluates that residual product instead of discarding it, and so proves the paper's Lemma 4.10 (lem:invtau) as stated:

inv_k(τ_D) = # {[D'] ∈ T¹_D : |D'| ≠ ∅} + δ(0 ∈ T⁰_D and u + v ∼ K_G).

The correction term is invTauCorrection. Two facts pin it down: each factor of the residual product is the principality indicator of a degree-zero divisor, and exact torsion makes at most one residue in a fundamental period principal. The upshot is that the whole cyclic sum of residual products is 0 or 1, and is 1 exactly under the paper's conjunction.

The converse half of thm:kgtThetas (Theorem 4.8) is then immediate: apply the formula to D = w for a vertex w and read the bound inv_k ≤ g = 2 backwards.

Generic divisor bookkeeping #

Everything in this section is stated for an abstract {G : CFGraph} with abstract divisors, so that no downstream use has to run divisor algebra on a concrete banana.

theorem Bananas.fixedDegreeTwist_eq_degreeTwistInt (G : CFGraph) (u v : G.V) (D : CFDiv G) (d b : ℤ) :
fixedDegreeTwist G u v D d b = degreeTwistInt (mark G u v) D d b

fixedDegreeTwist and degreeTwistInt are the same twist family, the former carrying its marks as separate arguments.

theorem Bananas.fixedDegreeTwist_two_eq_next_zero_add_uv (G : CFGraph) (u v : G.V) (D : CFDiv G) (b : ℤ) :
fixedDegreeTwist G u v D 2 b = fixedDegreeTwist G u v D 0 (b + 1) + oneChip u + oneChip v

The degree-two twist is the next degree-zero twist with both marks added back.

theorem Bananas.degreeTwistInt_one_chip_one (G : CFGraph) (u v w : G.V) (b : ℤ) :
degreeTwistInt (mark G u v) (oneChip w) 1 b = oneChip w + b • (oneChip u - oneChip v)

The degree-one twists of a single chip are its marked-difference translates.

theorem Bananas.degreeTwistInt_emod_linearEquiv {M : TwiceMarked} {k : ℕ} (hk : TorsionWitness M k) (D : CFDiv M.graph) (d b : ℤ) :
linearEquiv M.graph (degreeTwistInt M D d b) (degreeTwistInt M D d (b % ↑k))

A degree twist may be replaced by its Euclidean residue index modulo any torsion witness.

theorem Bananas.degreeTwistInt_linearEquiv_zero_of_emod_eq {M : TwiceMarked} {k : ℕ} (hk : TorsionWitness M k) (D : CFDiv M.graph) (d b c : ℤ) (hmod : b % ↑k = c % ↑k) (hc : linearEquiv M.graph (degreeTwistInt M D d c) 0) :

Principality of a degree twist depends only on the index modulo a torsion witness.

theorem Bananas.twist_eq_degreeTwistInt (M : TwiceMarked) (D : CFDiv M.graph) (a b d : ℤ) (h : CFDiv.degree D + a - b = d) :
D + a • oneChip M.u - b • oneChip M.v = degreeTwistInt M D d b

Every twist D + a·u - b·v of degree d is the degree-d twist family member at index b. Together with the next lemma and degreeTwistInt_injective_on_fundamental_period this says that the finite family degreeTwistInt M D d · on Fin k is a system of representatives for the paper's set T^d_D of degree-d twist classes.

theorem Bananas.exists_fin_degreeTwistInt_linearEquiv {M : TwiceMarked} {k : ℕ} (hk : TorsionWitness M k) (D : CFDiv M.graph) (d b : ℤ) :
∃ (c : Fin k), linearEquiv M.graph (degreeTwistInt M D d b) (degreeTwistInt M D d ↑↑c)

Every degree-d twist index is represented in one fundamental period.

With A principal, the canonical complement of A + u + v is principal exactly when the marked pair is canonical. This is the pointwise content of the paper's correction term.

theorem Bananas.rankPlusOne_mul_eq_one_of_linearEquiv_zero {G : CFGraph} (A Y : CFDiv G) (hA : CFDiv.degree A = 0) (hY : CFDiv.degree Y = 0) (h1 : linearEquiv G A 0) (h2 : linearEquiv G Y 0) :

A product of two degree-zero multiplicities is one when both divisors are principal and zero otherwise.

If either factor of a degree-zero multiplicity product is nonprincipal, the product vanishes.

The correction term #

noncomputable def Bananas.invTauCorrection (M : TwiceMarked) (D : CFDiv M.graph) :

Paper source: the second summand of lem:invtau (Lemma 4.10), δ(0 ∈ T⁰_D and u + v ∼ K_G).

0 ∈ T⁰_D says some degree-zero twist of D is principal; the second conjunct is the failure of the paper's rigidity condition (in genus two, r(u+v) = 0 is equivalent to u + v ≁ K_G).

Equations
  • One or more equations did not get rendered due to their size.
Instances For
    theorem Bananas.invTauCorrection_mark (G : CFGraph) (u v : G.V) (D : CFDiv G) :
    invTauCorrection (mark G u v) D = if (∃ (b : ℤ), linearEquiv G (degreeTwistInt (mark G u v) D 0 b) 0) ∧ linearEquiv G (oneChip u + oneChip v) (canonicalDivisor G) then 1 else 0

    rfl-wrapper: the correction term at an explicitly marked graph, stated without the TwiceMarked projections so that rw fires on it.

    Paper source: lem:invtau (Lemma 4.10), in full, with its correction term.

    For a theta graph with any two marks, an exact torsion order k and any divisor D admitting a k-affine transmission permutation, the number of k-inversions is the number of effective degree-one twists of D in one torsion period, plus δ(0 ∈ T⁰_D and u + v ∼ K_G).

    The converse half of Theorem 4.8 #

    theorem Bananas.eq_of_ncard_le_two_of_mem_three {α : Type u_1} [Finite α] {S : Set α} (h : S.ncard ≤ 2) {z n m : α} (hz : z ∈ S) (hn : n ∈ S) (hm : m ∈ S) (hnz : n ≠ z) (hmz : m ≠ z) :
    n = m

    A finite set of size at most two containing a distinguished element has at most one other element.

    The effective degree-one twists of a single chip are exactly the non-recurrence indices.

    Paper source: thm:kgtThetas (Theorem 4.8), the "only if" direction.

    For a rigidly marked theta graph, k-general transmission forces the marked difference class to be non-recurrent. The proof is the paper's: apply the exact formula of Lemma 4.10 to D = w for each vertex w, note that the zero residue is always effective, and read the inversion bound inv_k ≤ 2 backwards.