Theorem 6.3, p.49 — and for cubic tensor modules of stable classical groups
OpenGroupIsoPoly.Supplement.theorem_6_3Let be a finite field of characteristic and let be a product of derived classical groups , , , over the common defining field , in their natural representations and in the stable range (natural dimension , Witt index for symplectic, unitary and orthogonal factors). Let
where each is a natural, dual, or Frobenius-twisted module of one factor of , and let be its pairing submodule (Definition 16). Then
These assertions hold
- with fixed multiplicities: for , every ;
- for every central-product image of , given by a surjection with central kernel through which the action on factors;
- after prime-field descent: for the prime-field tensor module , where is viewed over , and its pairing submodule formed over .
Consequently every 1-cocycle equals for a unique , and for any generating set of this is the unique solution of the linear equations , .
In the paper this removes the affine obstruction in cubic tensor modules: after quotienting by invariant pairings, an affine action can be moved to a linear one by a computable change of origin.
Formalization Note The conclusions are packaged in QuotCohVanishes, asserted over and over for every matrix realization of the restricted modules. Factors may repeat and the may come from different factors.
Differences from the source:
- reading: "each Ti is a natural, dual, or Frobenius-twisted module of one factor" → Frobenius twists of the natural and of the dual module are both allowed; reason: Definition 16 does not restrict the twist; Theorem 6.1(iv) says "a Frobenius twist of either"; the broader reading is the stronger statement, and the prime-field descent clause involves Galois twists of duals anyway
- reading: "The assertions hold after prime-field descent" → for every F_p-matrix realization T'_k of the restriction of scalars of T_k to F_p, the same conclusions for T' = T'1 ⊗{F_p} T'2 ⊗{F_p} T'_3 and its pairing submodule J(T') formed over F_p; reason: Theorem 6.2 applies its assertions to "a prime-field tensor module"; the other reading (T/J(T) viewed over F_p) is clause 1 itself, since fixed vectors and cocycles do not depend on the scalar field
- reading: "with fixed multiplicities" → (T/J(T)) ⊗ F^m for every m ≥ 0, i.e. m copies; reason: multiplicity spaces carry the trivial action
- reading: "for a central-product image given with its quotient map" → every surjective φ : D → D̄ with ker φ ≤ Z(D) such that ker φ acts trivially on T/J(T): H¹(D̄, T/J(T)) = 0; reason: the image must act on T/J(T); its fixed vectors are those of D since φ is onto
- reading: "a unique b … computable by solving these linear equations on generators" → b is unique modulo J and, for every generating set S of D, every solution of (s−1)b′ = c(s) (s ∈ S) equals b modulo J; reason: no running time is claimed; the content is that the generator equations determine b
- encoding: T/J(T), T = T_1 ⊗_K T_2 ⊗_K T_3 → coordinate arrays modulo the submodule J(T); cocycles and coboundaries as congruences modulo J; reason: a function to the quotient always lifts; same instances
- encoding: J(T) = im J_T → span of the insertions κ_{i,η}(v), η ∈ B_i; reason: J_T is linear and P(T) is spanned by pure tensors v ⊗ η
- encoding: D a product of derived classical groups in the stable range → D = ∏ classicalGroup (dat i) over the common field K: SL, Sp(A), SU(σ, H), Ω(Q) = [O(Q), O(Q)] in their natural representations with any nondegenerate Gram data, n_i ≥ 64 and Witt index ≥ 32 (IsStableDerived); reason: §6.1 and Remark 1; Ω(Q) = [O(Q), O(Q)] is the paper's identification in the stable range
import Mathlib import Definitions.Def_GroupIsoPoly_Supplement_IsStableDerived import Definitions.Def_GroupIsoPoly_Supplement_factorMat import Definitions.Def_GroupIsoPoly_Supplement_QuotCohVanishes set_option autoImplicit false
namespace GroupIsoPoly
namespace Supplement
/-- **Theorem 6.3 (Cohomology after quotienting by invariant pairings).**
Qiao, Zhang, *Testing isomorphism of finite groups in time polynomial in the group order*, preprint,
Nov. 3, 2026 (held version sha256 eb9a3355…), Theorem 6.3, p.49 (PDF p.53): "Let D and T satisfy the
hypotheses of Theorem 6.2. Then (T/J(T))^D = 0, H1(D, T/J(T)) = 0. The assertions hold after
prime-field descent, with fixed multiplicities, and for a central-product image given with its
quotient map. Consequently a cocycle c : D → T/J(T) specified by its values on the generators has a
unique b ∈ T/J(T) with c(g) = (g − 1)b, computable by solving these linear equations on generators."
The hypotheses of Theorem 6.2 (p.48, PDF p.52) are those of Definition 16 with "D a product of
classical derived groups in the stable range above" (§6.1, Remark 1, p.45).
Let `K` be a finite field of characteristic `p`, and `D = ∏_{i<t} G_i` a product of derived classical
groups `G_i ∈ {SL, Sp, SU, Ω}` over the common defining field `K`, each in its natural representation
on `K^{n_i}` and in the stable range. Let `T = T₁ ⊗_K T₂ ⊗_K T₃` where `T_k` is the natural, dual or
Frobenius-twisted module (of kind `κ k`) of the factor `f k`. Then:
1. over `K`, `QuotCohVanishes` holds for `T/J(T)`: `(T/J(T))^D = 0`, `H¹(D, T/J(T)) = 0`, both with
every multiplicity, the unique coboundary solution `b` is the unique solution of the equations on
any generating set, and `H¹` vanishes for every central-product image;
2. after **prime-field descent**: for any matrix realizations `ρ k` over `F_p` of the restrictions of
scalars of `T_k` to `F_p`, the same holds for the prime-field tensor module
`T' = T'₁ ⊗_{F_p} T'₂ ⊗_{F_p} T'₃` and its pairing submodule `J(T')` formed over `F_p`.
**Formalization Note.** Factors may repeat and the `T_k` may belong to different factors. A twist of
the dual module is allowed. See `QuotCohVanishes` for the encoding of the quotient. -/
theorem theorem_6_3 (p : ℕ) [Fact p.Prime] {K : Type} [Field K] [Fintype K] [CharP K p]
{t : ℕ} (n : Fin t → ℕ) (dat : (i : Fin t) → ClassicalDatum K (n i))
(hst : ∀ i, IsStableDerived (dat i)) (f : Fin 3 → Fin t) (κ : Fin 3 → ModKind) :
QuotCohVanishes (F := K) (factorMat (K := fun _ => K) dat (f 0) (κ 0))
(factorMat (K := fun _ => K) dat (f 1) (κ 1))
(factorMat (K := fun _ => K) dat (f 2) (κ 2)) ∧
∀ (a : Fin 3 → ℕ)
(ρ : (k : Fin 3) → ClProd (K := fun _ => K) dat → Matrix (Fin (a k)) (Fin (a k)) (ZMod p)),
(∀ k, IsRestrictionOf (K := fun _ => K) dat (f k) (κ k) (ρ k)) →
QuotCohVanishes (ρ 0) (ρ 1) (ρ 2) := by sorry
end Supplement
end GroupIsoPoly