Pairwise products along diagonal chains #
Several invariants of a diagonal quadratic form ⟨a₁, …, aₙ⟩ are products over the pairs of
distinct coefficients, ∏_{i<j} F aᵢ aⱼ, for a pairing F of units with values in a commutative
monoid: the Hasse invariant ∏_{i<j} [(aᵢ, aⱼ)] in the Brauer group, the local Hasse invariant
built from the {±1}-valued Hilbert symbol, and the second Stiefel–Whitney class. By Witt's chain
theorem, such a product is an invariant of the isometry class of the form as soon as it is
unchanged by the two kinds of elementary steps of a diagonal chain. This file proves that
invariance once, for an abstract pairing:
- a permutation step does not change the product when
Fis symmetric; - a binary step does not change the product when
Fis multiplicative in its first argument and takes equal values on the coefficients of isometric binary forms.
The second hypothesis already forces symmetry, because ⟨a, b⟩ ≅ ⟨b, a⟩. It also forces
F (a * b) x = F (c * d) x whenever ⟨a, b⟩ ≅ ⟨c, d⟩, because their discriminants differ by a
square, so rescaling the first coordinate gives ⟨a * b, x⟩ ≅ ⟨c * d, x⟩. Together with
multiplicativity, this makes the cross terms between the changed pair and the unchanged
coefficients agree.
Main results #
TauCeti.PermutationStep.prod_prod_Ioi_eq: invariance under a permutation step.TauCeti.BinaryStep.prod_prod_Ioi_eq: invariance under a binary step.
Which invariant each source means #
The sources give the name Hasse invariant to two different products of symbols, and the name
Witt invariant to a third invariant that is not a pairwise product at all. Below,
q ≅ ⟨a₁, …, aₙ⟩ and (a, b) is a symbol: the quaternion algebra up to isomorphism, its class
in the Brauer group, or the {±1}-valued Hilbert symbol over a local field.
- Lam's
s(Lam V.3.17) is∏_{i<j} (aᵢ, aⱼ), the product over pairs of distinct indices treated in this file, with the empty product in ranks0and1. The Hasse invariants of Tau Ceti follow this convention:TauCeti.RegularFormClass.hasseInvariantwith values in the Brauer group, andTauCeti.RegularFormClass.localHassewith Hilbert symbols over a nonarchimedean local field. - Serre's
ε(Serre, A Course in Arithmetic, IV.2.1) is the same product∏_{i<j} (aᵢ, aⱼ)of Hilbert symbols overℚ_porℝ, so it agrees with Lam'ss; overℚ_pit islocalHasse. It has nothing to do with Serre's signε(u) = (u - 1)/2 mod 2of a unituofℤ_2(Serre II.3.3), which isTauCeti.serreEps. - O'Meara's Hasse symbol, the invariant of his classification over local fields
(O'Meara 63:20), is
∏_{i≤j} (aᵢ, aⱼ), which also takes the diagonal pairs. Since(a, a) = (a, -1), it is Lam'ss(q)times(d, -1), whered = a₁ ⋯ aₙis the unsigned discriminant (TauCeti.RegularFormClass.omearaHasseSymbol_eqandTauCeti.RegularFormClass.omearaLocalHasseSymbol_eq). The two conventions differ already in rank one: for⟨-1⟩overℝ, Lam'ssis the empty product1, while O'Meara's symbol is(-1, -1) = -1. - Lam's
c, the Witt invariant (Lam V.3.12), is the class of the Clifford algebra ofqwhennis even, and of its even Clifford algebra whennis odd (TauCeti.RegularFormClass.cliffordInvariant). It is not a product over pairs, but Lam V.3.20 expresses it throughs(q), the discriminant and the rank (TauCeti.RegularFormClass.cliffordInvariant_eq_hasseInvariant_mul).
⚠ Lam (p. 120, Caution) records that the formula relating c and s published by
C. T. C. Wall is incorrect; TauCeti.RegularFormClass.cliffordInvariant_eq_hasseInvariant_mul
is Lam's formula, so a comparison with another source should start from it.
References #
- T. Y. Lam, Introduction to Quadratic Forms over Fields, Graduate Studies in Mathematics 67, American Mathematical Society (2005), Chapter I, Theorem 5.2, and Chapter V, §3, in particular 3.12, 3.17, 3.18, 3.20 and the Caution on p. 120.
- J.-P. Serre, A Course in Arithmetic, Graduate Texts in Mathematics 7, Springer (1973), Chapter II, §3.3, and Chapter IV, §2.1.
- O. T. O'Meara, Introduction to Quadratic Forms, Springer (1963), §63:20.
Pairwise products are invariant under a permutation step. For a symmetric pairing F,
reordering the coefficients does not change ∏_{i<j} F (w i) (w j).
Pairwise products are invariant under a binary step. Let F be multiplicative in its first
argument and take equal values on the coefficients of isometric binary forms. Then replacing two
coefficients of a diagonal form by the coefficients of an isometric binary form does not change
∏_{i<j} F (w i) (w j).