Witt invariants of quaternionic forms

Nicolas Garrel · 2025

Read Witt invariants of quaternionic forms by Nicolas Garrel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-08920.txt.

Introduction

Nicolas Garrel, a solo author, describes all Witt invariants of quaternionic forms, in a 2025 math paper. These forms are anti-hermitian forms over a quaternion algebra with its standard involution. Witt invariants are special functions on such forms that stay unchanged under certain equivalences.

Garrel shows the invariants are built from defined lambda-power operations, much as in the classic theory of quadratic forms. Unlike that classic case, the module of invariants is not free over those operations, so relations among them matter. The proof extends scalars to a field where the algebra splits and tracks residues at special points. The result also covers orthogonal groups of related algebras.

Catalog record

CREATORNicolas Garrel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-08920

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Witt invariants of quaternionic forms Nicolas Garrel Abstract Introduction In the seminal [ 2 ] , Serre starts the study of cohomological invariants, in particuliar of algebraic groups. Let k be a base field, which in this article we will assume to be of characteristic not 2 . Then if \mathbf{Field}_{/k} is the category of field extensions of k (it is enough to consider finitely generated extensions, in case one wants to work in an essentially small category), and F:\mathbf{Field}_{/k}\to\mathbf{Set} and A:\mathbf{Field}_{/k}\to\mathbf{Ab} are functors, then an invariant of F with values in A is simply a natural transformation from F to A (seen as a functor to \mathbf{Set} ). We write \Inv(F,A) for the set of such invariants; this is clearly an abelian group for the pointwise addition coming from the abelian group structure of A(K) for any extension K/k . If A(K)=H^{i}(K,M) for some Galois module M defined over k , we speak of cohomological invariants with values in M . If F(K)=H^{1}(K,G) for some algebraic group G defined over k , we speak of "invariants of G " (a more explicit terminology would be "invariants of G -torsors"). If G is a classical group, one recovers some familiar functors; in particular, if G=O_{n} is the classical orthogonal group, the corresponding F is isomorphic to the functor \mathbf{Quad}_{n} of non-singular n -dimensional quadratic forms. Taking M=\mathbb{Z}/2\mathbb{Z} with trivial Galois action, the classical invariants of quadratic forms q such as the determinant \det(q)\in K^{\times}/(K^{\times})^{2}\simeq H^{1}(K,\mathbb{Z}/2\mathbb{Z}) and the Hasse invariant in \Br(K)[2]\simeq H^{2}(K,\mathbb{Z}/2\mathbb{Z}) give examples of cohomological invariants of O_{n} . In [ 2 ] Serre gives a complete description of mod 2 cohomological invariants of O_{n} , proving that they form a free module over H^{*}(k,\mathbb{Z}/2\mathbb{Z}) with basis given by the Stiefel-Whitney invariants w_{i} (the determinant and Hasse invariant are then recovered as, respectively, w_{1} and w_{2} ). Serre also introduces in [ 2 ] the notion of Witt invariants (which amounts to taking A(K)=W(K) the Witt group of K ), noticing that the formal properties of the Witt group regarding residues and specialization with respect to discrete valuations are similar enough to cohomology groups that a similar theory can be developped. For Witt invariants of O_{n} , he proves that they form a free W(k) -module generated by the \lambda -operations \lambda^{i} (for 0\leqslant i\leqslant n ). Although there is clearly much more literature devoted to cohomological invariants, in particular due to their use for rationality problems and computations of essential dimensions of algebraic groups, the solution of the Milnor Conjecture by Voevodsky gives a direct connexion between cohomological and Witt invariants which can motivate a closer interest in those. Explicitly, if we can define a Witt invariant \alpha\in\Inv(F,W) with values in I^{n}\subset W (where I^{n} is the n th power of the fundamental ideal of the Witt group), then composing with the canonical isomorphism I^{n}(K)/I^{n+1}(K)\stackrel{{\scriptstyle\sim}}{{\rightarrow}}H^{2}(K,\mathbb{Z}/2\mathbb{Z}) defines a cohomological invariant in \Inv^{n}(F,\mathbb{Z}/2\mathbb{Z}) . It is not difficult to realize that one recovers all cohomological invariants of O_{n} in this way (see [ 3 , Section 9] for instance) Now we can ask the question of invariants of G=O(A,\sigma) where (A,\sigma) is a central simple algebra with orthogonal involution. This algebraic group is a form of O_{n} (which corresponds to the case where A=M_{n}(k) and \sigma is the adjoint involution of the form \langle 1,\dots,1\rangle ), so we might expect a similar description of invariants. The pointed set H^{1}(K,G) is this time is a canonical bijection with the set of isometry classes of nondegenerate 1 -dimensional hermitian forms over (A,\sigma) (the base point being \langle 1\rangle_{\sigma} ). We address in this article the question of Witt invariants of O(A,\sigma) when A has index \leqslant 2 and degree 2r (which is a class stable by scalar extension). If Q is the quaternion algebra Brauer-equivalent to A , we can choose a Morita equivalence between (A,\sigma) and (Q,\gamma) where \gamma is the canonical symplectic involution of Q , and this induces an isomorphism between the functor H^{1}(-,G) and H_{Q}^{(r)} , where for any extension K/k , H_{Q}^{(r)}(K) is the set of isometry classes of nondegenerate anti-hermitian forms of reduced dimension 2r over (Q_{K},\gamma_{K}) . Our main result is then Theorem 4.7 , which states that the Witt invariants of H_{Q}^{(r)}(K) (and therefore of G ) are again generated by \lambda -operations (this time in the sense of [ 4 ] ), but the coefficients have to be taken not only in W(k) but in the mixed Witt ring \widetilde{W}^{-1}(Q,\gamma) introduced in [ 5 ] . Furthermore, such a decomposition is not exactly unique, the norm form n_{Q} being the obstruction. Notations and conventions If Q is a quaternion algebra, \gamma is its canonical symplectic involution, Q_{0} is the space of pure quaternions, and Q_{0}^{\times} the set of invertible pure quaternions. 1 Mixed Witt rings 1.1 General case Let (A,\sigma) be an Azumaya algebra with involution of the first kind over k , and \varepsilon\in\mu_{2}(k) . Then SW^{\varepsilon}(A,\sigma) is defined as the monoid of \varepsilon -hermitian forms over (A,\sigma) , with orthogonal sums, GW^{\varepsilon}(A,\sigma) is the Grothendieck group of SW^{\varepsilon}(A,\sigma) , and W^{\varepsilon}(A,\sigma) is the quotient of GW^{\varepsilon}(A,\sigma) by the subgroup of hermitian forms. For any a\in A^{\times} such that \sigma(a)=\varepsilon a , the elementary diagonal form \langle a\rangle_{\sigma} is \begin{array}[]{rrcl}\langle a\rangle_{\sigma}:&A\times A&\longrightarrow&A\\ &(x,y)&\longmapsto&\sigma(x)ay.\end{array} (1) The diagonal form \langle a_{1},\dots,a_{r}\rangle_{\sigma} is the orthogonal sum \langle a_{1},\dots,a_{r}\rangle_{\sigma}=\langle a_{1}\rangle_{\sigma}\perp\dots\perp\langle a_{r}\rangle_{\sigma}. (2) When (A,\sigma)=(k,\Id) and \varepsilon=1 we just write SW(k) , GW(k) and W(k) . Note that W^{\varepsilon}(A,\sigma) is naturally a W(k) -module. We define a \mathbb{Z}/2\mathbb{Z} -graded GW(k) -module \widetilde{GW}^{\varepsilon}(A,\sigma)=GW(k)\oplus GW^{\varepsilon}(A,\sigma) (3) and a \mathbb{Z}/2\mathbb{Z} -graded W(k) -module \widetilde{W}^{\varepsilon}(A,\sigma)=W(k)\oplus W^{\varepsilon}(A,\sigma). (4) If (A,\sigma) and (B,\tau) are Azumaya algebra with involution of the first kind over k , and \varepsilon_{0}\in\mu_{2}(k) , we write (B,\tau)\stackrel{{\scriptstyle(V,h)}}{{\rightsquigarrow}}(A,\sigma) (5) if V is a B - A -bimodule, balanced over k , such that B\simeq\End_{A}(V) for this action, and h is an \varepsilon_{0} -hermitian form over (A,\sigma) , and \tau is the adjoint involution of h . We then say that ( 5 ) is an \varepsilon_{0} -hermitian Morita equivalence, or simply a Morita equivalence. Such an equivalence induces graded isomorphisms h_{*} which fit in this commutative diagram {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon}(B,\tau)} {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon\varepsilon_{0}}(A,\sigma)} {\lx@inpgf@ignorespaces\widetilde{W}^{\varepsilon}(B,\tau)} {\lx@inpgf@ignorespaces\widetilde{W}^{\varepsilon\varepsilon_{0}}(A,\sigma)} \scriptstyle{\lx@inpgf@ignorespaces h_{*}} \scriptstyle{\lx@inpgf@ignorespaces h_{*}} (6) such that h_{*} is the identity on the even components GW(k) and W(k) . In [ 5 ] , a graded commutative ring structure is defined on \widetilde{GW}^{\varepsilon}(A,\sigma) and \widetilde{W}^{\varepsilon}(A,\sigma) such that ( 6 ) is a commutative diagram of rings. The product is characterized by \langle a\rangle_{\sigma}\cdot\langle b\rangle_{\sigma}\simeq T_{\sigma,a,b}\in SW(k) (7) where T_{\sigma,a,b} is the twisted involution trace form defined as \begin{array}[]{rrcl}T_{\sigma,a,b}:&A\times A&\longrightarrow&k\\ &(x,y)&\longmapsto&\Trd_{A}(\sigma(x)ay\sigma(b)).\end{array} (8) Furthermore, if K/k is a field extension, then the scalar extension maps induce a commutative diagram {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon}(B,\tau)} {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon\varepsilon_{0}}(A,\sigma)} {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon}(B_{K},\tau_{K})} {\lx@inpgf@ignorespaces\widetilde{GW}^{\varepsilon\varepsilon_{0}}(A_{K},\sigma_{K})} \scriptstyle{\lx@inpgf@ignorespaces h_{*}} \scriptstyle{\lx@inpgf@ignorespaces(h_{K})_{*}} (9) and similarly for mixed Witt rings. If (A,\sigma)=(k,\Id) and \varepsilon=1 , the canonical \mathbb{Z}/2\mathbb{Z} -graded W(k) -module isomorphism \widetilde{W}^{1}(k,\Id)=W(k)\oplus W(k)\simeq W(k)[\mathbb{Z}/2\mathbb{Z}] (10) is a graded W(k) -algebra isomorphism. Let us write \delta:\widetilde{W}^{1}(k,\Id)\to W(k) (11) for the map W(k)\oplus W(k)\to W(k) given by the sum of components. It is a W(k) -algebra morphism. 1.2 Quaternion algebras We consider the case of a quaternion algebra Q with its canonical involution \gamma , and \varepsilon=-1 . Then

Read the complete text