aa r X i v : . [ m a t h . K T ] J u l FORMS OVER FIELDS AND WITT’S LEMMA
DAVID SPREHN and NATHALIE WAHL
Abstract
We give an overview of the general framework of forms of Bak, Tits and Wall, when restrictingto vector spaces over fields, and describe its relationship to the classical notions of Hermitian,alternating and quadratic forms. We then prove a version of Witt’s lemma in this context,showing in particular that the action of the group of isometries of a space equipped with aform is transitive on isometric subspaces.
1. Introduction
Let F be a (commutative) field, and σ : F → F an involution, that is a fieldisomorphism squaring to the identity. To simplify notations, we will write ¯ x for σ ( x ) in analogy with complex conjugation. Let E be an F –vector space. Recallthat a map f : E × E → F is called sesquilinear if it is biadditive and satisfiesthat f ( av, bw ) = ¯ af ( v, w ) b for all a, b ∈ F and v, w ∈ E . If σ is the identity, sucha map is just a bilinear form.The “classical groups” over F are the automorphism groups ( isometry groups )of different types of forms on F –vector spaces:(1) A symplectic group is the group of isometries of a vector space E equippedwith an alternating form , that is a bilinear map ω : E × E → F such that ω ( v, v ) = 0 . (2) A unitary group is the group of isometries of a vector space E equipped witha Hermitian form , that is a map ω : E × E → F that is sesquilinear with respectto a (non-trivial) involution σ : F → F , and such that ω ( v, w ) = ω ( w, v ) . (3) An orthogonal group is the group of isometries of a vector space E equippedwith a quadratic form , that is a set map Q : E → F such that Q ( cv ) = c Q ( v ) for all c ∈ F , v ∈ E and such that the associated map B Q ( v, w ) = Q ( v + w ) − Q ( v ) − Q ( w ) is bilinear. These examples of “forms” all fit into a common framework, developed inparticular by Bak, Tits, Wall (see eg. [2, 15, 16, 17]) and later by Magurn-Vaserstein-van der Kallen [10, Sec 4]. This framework allows to treat all threecases at once, including when the field has characteristic 2, and symmetric andanti-symmetric forms are indistinguishable. The framework makes sense in thecontext of rings with anti-involutions, and has turned out very useful for examplefor proving homological stability of symplectic, unitary and orthogonal groupsin one go, see eg. [5, 11, 13] in the context of rings, and [14] in the special caseof fields, or in the study of certain classes of subgroups of general linear groups,see eg. [12].In the present paper, we give an overview of what this general frameworkbecomes under the assumption that we work only with fields instead of generalrings. In particular, we assemble and complete results from the literature to showthat in this case, alternating, Hermitian and quadratic forms are essentially theonly existing flavors of forms arising in this context. In the second part of thepaper, we prove a general version of the classical Witt’s Lemma [18], showingthat the isometry group acts transitively on isometric subspaces. We allow thecase of degenerate forms that is often not treated in the literature, and providea relative version. This last result is used by our companion paper [14], whichproves a homological stability result for such isometry groups of vector spacesequipped with forms.We describe our main results in more detail now.Fix a pair ( F , σ ) with F a field and σ an involution (possibly the identity).The generalized definition of a form over an F –vector space E , which we describenow, depends on the further choice of an element ε ∈ F such that ε ¯ ε = 1, and acertain additive subgroup 0 ≤ Λ ≤ F . In the present situation, with F a field, thegroup Λ is almost always determined by the triple ( F , σ, ε ) (see Proposition A.1in the Appendix).Given such a tuple ( F , σ, ε, Λ) and a sesquilinear form q : E × E → F , we canconstruct two new maps: define ω q : E × E → F and Q q : E → F / Λ by setting ω q ( v, w ) = q ( v, w ) + εq ( w, v ) and Q q ( v ) = q ( v, v ) + Λ ∈ F / Λ . The map ω q is an “ ε -skew symmetric sesquilinear form” by construction, and Q q is to be thought of as a “quadratic refinement” of ω q . Theorem 2.5 shows that ω q and Q q are always related by the equation Q q ( v + w ) − Q q ( v ) − Q q ( w ) = ω q ( v, w ) ∈ F / Λ . In particular, if σ = id, ε = 1 and Λ = 0, then Q q is precisely a quadratic form asdescribed above, with ω q its associated bilinear form. But if σ = id with ε = 1, ORMS OVER FIELDS AND WITT’S LEMMA 3 then ω q is a Hermitian form. To get alternating forms, we will take σ = id and ε = − q ( ω q , Q q ), from sesquilinear maps to such pairs of maps,is not injective. Let X ( E, σ, ε, Λ) ≤ Sesq σ ( E ) denote its kernel, where Sesq σ ( E )denotes the vector space of sesquilinear forms on E . (See Definition 2.1 fora direct description of X .) The group of all ( σ, ε, Λ) –quadratic forms on E isdefined to be the quotientForm( E, σ, ε,
Λ) := Sesq σ ( E ) /X ( E, σ, ε, Λ) . By definition, the form q is thus always determined by the pair ( ω q , Q q ) andin fact, Proposition 2.7 shows that only one of ω q or Q q is always enough todetermine q , but which one of the two depends on the chosen Λ. Hence, insteadof working sometimes with forms like ω q , sometimes with forms like Q q , we canalways just work with equivalence classes of sesquilinear forms q . This leaves uswith the question of what “forms like ω q or Q q ” arise in this framework? Thisis answered by our first main result, which says that, for almost all choices ofparameters ( σ, ε, Λ), the general framework of forms simply specializes to theclassical Hermitian, alternating and quadratic forms:
Theorem A . (1) When σ = id , there is a natural isomorphism Form(
E, σ, ε, Λ) ∼ = Herm σ ( E ) . (2) When σ = id , ε = − and Λ = F , there is a natural isomorphism Form(
E, σ, ε, Λ) ∼ = Alt ( E ) . (3) When σ = id , ε = 1 and Λ = 0 , there is a natural isomorphism
Form(
E, σ, ε, Λ) ∼ = Quad ( E ) . Here Herm σ ( E ), Alt( E ) and Quad( E ) are the groups of Hermitian, alternatingand quadratic forms as described above. (See also Definition 2.4.) When F is afinite field, there is only one isomorphism class of non-degenerate such form ineach dimension, except in the orthogonal case (3) (see eg. [4, Chap II]). Formsover infinite fields is on the other hand a vast subject, see eg. the book [9] whichis concerned with quadratic forms over fields of characteristic not 2. Remark . (i) When σ = id, the result shows that the group of formsdepends neither on Λ, which is a consequence of the fact that there is only onepossible Λ in this case by Proposition A.1, nor on ε .(ii) When σ = id, there are two main possibilities: the alternating case (2), andthe quadratic case (3). If the field is not of characteristic 2, cases (2) and (3)are distinguished by whether ε = 1 or ε = − F has characteristic 2, then 1 = − DAVID SPREHN and NATHALIE WAHL are distinguished by the choice of Λ. This is in fact the one situation wherethere is a freedom of choice for Λ and cases (2) and (3) correspond to the twopossible extremes, the smallest and largest possible Λ. Note also that, givenΛ, Proposition A.1 shows that ε is in fact always redundant information when σ = id. We chose to leave the ε in the statement as it is informative.(iii) The only cases not covered by the theorem are thus in characteristic 2 with σ = id and 0 < Λ < F . Proposition 2.11 shows that under the additionalassumption that F is a perfect field, also these cases are actually covered by (3).More generally, combining Proposition 2.7 (2) and Lemma 2.9 (2), we get thata “mod Λ” version of (3) holds under the weaker assumption that σ = id andΛ < F , giving the remaining cases.The above results are all proved in Section 2.The second part of the paper is concerned with isometries: Given a vectorspace E equipped with a form q , called here a formed space ( E, q ), an isometry of E is a linear map respecting the form. For appropriate choices of q , the classicalgroups Sp n ( F ), O n ( F ), O n,n ( F ), U n ( F ) and U n,n ( F ) are all such isometry groups(see Example 2.8). Restricting the form q to subspaces of E , we can talk aboutisometries between subspaces.The classical groups just mentioned are all automorphism groups of non-degenerate formed spaces, that is with trivial kernel, where the kernel K ( E ) of( E, q ) is defined to be the kernel of the map E → σ E ∗ , v ω q ( − , v ) , where σ E ∗ denotes the vector space of σ -skew-linear maps E → F . Degenerateformed spaces do however also naturally occur (see eg. Example 3.3 for examplesarising from Coxeter groups, or, in the context of rings, [6, Sec 5] for examplescoming from intersections of immersed spheres on manifolds). In the presentpaper, we do allow forms with non-trivial kernels.Our main result in the second part of the paper is the following: Theorem
B (Witt’s Lemma (Theorem 3.4)) . Let ( E, q ) be a formed space.Suppose that f : U → W is a bijective isometry between two subspaces of E suchthat f (cid:0) U ∩ K ( E ) (cid:1) = W ∩ K ( E ) . Then f can be extended to a bijective isometryof E . Theorem 3.7 in the paper gives in addition a relative version of Witt’s lemma,where a subspace may be assumed to be fixed by the constructed isometry.Witt originally considered in [18] the case of non-degenerate symmetric formsin characteristic not 2. His result was generalized by many authors (see eg. [15,Prop 2] and [3, p21,22,36], [10, Cor 8.3], [1, Thm 3.9], [12, Thm 1] and the paper
ORMS OVER FIELDS AND WITT’S LEMMA 5 [7] for many references to the existing literature). We were however not able tofind the result in the above generality, as needed by our companion paper [14].The proofs of Witt’s lemma and its relative version are given in Section 3.
2. Forms
Let ( F , σ ), as above, be a field with a chosen involution. Examples we have inmind are F any field with σ = id, or F = C with σ the conjugation, or F = F p r a finite field with σ = ( − ) r . As before, we write ¯ c = σ ( c ).Pick ε ∈ F satisfying ε ¯ ε = 1. Taking ε = ± ε = i also works when ( F , σ ) is the complex numbers with conjugation.Set Λ min = { a − ε ¯ a | a ∈ F } Λ max = { a ∈ F | a + ε ¯ a = 0 } . These are additive subgroups of F . Let Λ be an abelian group such thatΛ min ≤ Λ ≤ Λ max . Proposition A.1 in the appendix shows that for most triples ( F , σ, ε ), we actuallyhave Λ min = Λ max so that there is a unique Λ determined by the triple. Typicalvalues of Λ are 0, the fixed points F σ of the involution, or the whole field F . Definition . A ( σ, ε,
Λ)- quadratic form on E (or just a form for short) isan element of the quotientForm( E, σ, ε,
Λ) := Sesq σ ( E ) /X ( E, σ, ε, Λ) , of the vector space Sesq σ ( E ) of sesquilinear forms on E by its the additivesubgroup X ( E, σ, ε,
Λ) of all those f satisfying f ( v, v ) ∈ Λ and f ( w, v ) = − εf ( v, w )for all v, w ∈ E . A formed space ( E, q ) is a finite-dimensional vector space E equipped with a form q .We will see in Theorem 2.5 below that this agrees with the definition given inthe introduction, where X was defined as the kernel of a map. Remark . A typical example of an element in X ( E, σ, ε,
Λ) has the fol-lowing form: let q ∈ Sesq σ ( E ) be a sesquilinear form, and let f ( v, w ) = q ( v, w ) − εq ( w, v ) . Then − εf ( v, w ) = − εq ( v, w ) + ε ( εq ( w, v )) = f ( w, v ) given that ε ¯ ε = 1, while f ( v, v ) ∈ Λ min ≤ Λ. In particular, X ( E, σ, ε,
Λ) is always non-trivial.
DAVID SPREHN and NATHALIE WAHL
Example . (1) Let E = F n . The Eucledian form on E is the form q E defined by q E ( v, w ) = n X i =1 v i w i . (2) Let E = F n . The hyperbolic form on E is the form q H defined by q H ( v, w ) = n X i =1 v i − w i . If E is finite dimensional, say of dimension n , the choice of a basis gives aone-to-one correspondence between sesquilinear forms on E and ( n × n )–matriceswith coefficients in F , with the ( i, j )–th entry of the matrix being the value ofthe form on the i th and j th basis vectors. Describing the quotient of sesquilinearforms by X ( E, σ, ε,
Λ) is of course much more tricky.The set of forms on E is contravariantly functorial in E : from a linear map f : E → F and a form q on F , we produce a form on E as f ∗ q ( v, w ) = q ( f ( v ) , f ( w ))for v, w ∈ E . This is indeed well-defined as f ∗ q is again sesquilinear, and f ∗ takes X ( F, σ, ε,
Λ) to X ( E, σ, ε,
Λ).The goal of this section is to relate forms in the sense of Definition 2.1 to themore classical notions of Hermitian, alternating, and quadratic forms.
Definition . Let E be a vector space over F . Define Herm σ,ε ( E ) : the vector space of sesquilinear forms ω : E × E → F satis-fying that ω ( v, w ) = εω ( w, v )for all v, w ∈ E , considered here as a group. We write Herm σ ( E ) =Herm σ, ( E ) for the classical Hermitian forms. Alt(E) ≤ Herm id, − ( E ) : the subspace of alternating forms, i.e. those suchthat ω ( v, v ) = 0for all v ∈ E . Set ( E, F / Λ) : the group of set maps Q : E → F / Λ. Quad ( E ) ≤ Set ( E, F ) : the subgroup of classical quadratic forms, i.e. themaps Q : E → F such that Q ( av ) = a Q ( v ) ORMS OVER FIELDS AND WITT’S LEMMA 7 for all v ∈ E , a ∈ F and such that B Q : E × E → F defined by B Q ( v, w ) = Q ( v + w ) − Q ( v ) − Q ( w )is bilinear.Just like Form( E, σ, ε,
Λ), the groups Herm σ,ε ( E ), Alt(E), Set σ ( E, F / Λ) andQuad( E, F ) assemble to give contravariant functors from the category of vectorspaces and linear maps to the category of abelian groups. Theorem . (see also [16, Thm 1] ) Let E be a vector space. For q ∈ Form(
E, σ, ε, Λ) , setting ω q ( v, w ) = q ( v, w ) + εq ( w, v ) and Q q ( v ) = q ( v, v ) + Λ ∈ F / Λ , defines a map χ = ( ω ( − ) , Q ( − ) ) : Form( E, σ, ε, Λ) −→ Herm σ,ε ( E ) × Set ( E, F / Λ) which is an injective homomorphism, natural with respect to linear maps E → E ′ .In particular, if ( E, q ) and ( E ′ , q ′ ) are formed spaces, and f : E → E ′ a linearmap, then f is an isometry if and only if ω q ′ ( f ( v ) , f ( w )) = ω q ( v, w ) and Q q ′ ( f ( v )) = Q q ( v ) for all v, w ∈ E . Moreover, for any q ∈ Form(
E, σ, ε, Λ) , the maps ω q and Q q are related by the following equation: Q q ( v + w ) − Q q ( v ) − Q q ( w ) = ω q ( v, w ) ∈ F / Λ . (2.1) Example . (1) When σ = id and ε = 1, the form Q q E associated to theEucledian form of Example 2.3 is the standard quadratic form Q q E ( v ) = P ni =1 v i on F n , with ω q E = 2 q E its associated symmetric bilinear form. Note that ω q E iszero in characteristic 2!(2) When σ = id and ε = −
1, the form q q H associated to the hyperbolic form ofExample 2.3 is (up to reordering of the basis vectors) the standard symplecticform on F n , namely ω q H ( v, w ) = P ni =1 v i − w i − w i − v i . Proof of Theorem 2.5.
The map χ is well-defined and injective since ω f =0 and Q f ( v ) ∈ Λ if and only if f ∈ X ( E, σ, ε,
Λ). Also, for any q ∈ Form(
E, σ, ε,
Λ),we have that ω ( v, w ) = εω ( w, v ) for all v, w ∈ E , so the first component of χ has image in Herm σ,ε ( E ) as claimed. One checks that the assignment is agroup homomorphism that is natural with respect to vector space homomor-phisms. The characterization of the isometries follows directly from natural-ity and injectivity of χ . Finally, the last equation follows from the fact that q ( w, v ) − εq ( w, v ) ∈ Λ min . DAVID SPREHN and NATHALIE WAHL
The group of linear automorphisms GL( E ) acts on both the domain andcodomain of χ , and naturality implies that χ commutes with this action. Theisotropy group of an element q ∈ Form(
E, σ, ε,
Λ) is the group G ( E, q ) of bijectiveisometries of (
E, q ). It follows from the above result that G ( E, q ) = Aut(
E, ω q ) ∩ Aut(
E, Q q )is exactly the group of maps preserving both ω q and Q q . The following re-sult show that in fact one of those two conditions is always redundant and theautomorphism group G ( E, q ) identifies with either Aut(
E, ω q ) or Aut( E, Q q ): Proposition . (1) If Λ = Λ max , then Q q is determined by ω q , in thesense that the assignment ω ( − ) : Form( E, σ, ε, Λ) −→ Herm σ,ε ( E ) of Theorem 2.5 is injective. In particular, G ( E, q ) = Aut(
E, ω q ) as subgroups of GL( E ) in this case.(2) If Λ = F , then ω q is determined by Q q , in the sense that the assignment Q ( − ) : Form( E, σ, ε, Λ) −→ Set ( E, F / Λ) of Theorem 2.5 is injective. In particular, G ( E, q ) = Aut(
E, Q q ) as subgroupsof GL( E ) in this case. Example . Recall the Eucledian and hyperbolic forms q E and q H of Ex-ample 2.3.(1) The group G ( F n , q H ) for different choices of parameters ( σ, ε, Λ) identifieswith classical groups as follows:( σ, ε,
Λ) = (id , − , F ) G ( F n , q H ) = Aut( F n , ω q H ) = Sp n ( F )( σ, ε, Λ) = ( = id , , F σ ) G ( F n , q H ) = Aut( F n , Q q H ) = Aut( F n , ω q H ) = U n,n ( F )( σ, ε, Λ) = (id , , G ( F n , q H ) = Aut( F n , Q q H ) = O n,n ( F )(2) The group G ( F n , q E ) for different choices of parameters ( σ, ε, Λ) identifieswith classical groups as follows:( σ, ε,
Λ) = ( = id , , F σ ) G ( F n , q E ) = Aut( F n , Q q E ) = Aut( F n , ω q E ) = U n ( F )( σ, ε, Λ) = (id , , G ( F n , q E ) = Aut( F n , Q q E ) = O n ( F )the latter under the additional assumption that char( F ) = 2 for the form q E to benon-degenerate. The “symplectic version” of the Euclidean groups would be thegeneral linear group GL( E ) because the form q E is trivial with the parameters( σ, ε, Λ) = (id , − , F ). Proof of Proposition 2.7.
We first prove (1). Since we know q χ ( q ) =( ω q , Q q ) is additive and injective, it suffices to check that ω q = 0 implies Q q = 0.Indeed, for v ∈ E , 0 = ω q ( v, v ) = q ( v, v ) + εq ( v, v ) , ORMS OVER FIELDS AND WITT’S LEMMA 9 so q ( v, v ) ∈ { a ∈ F | a + εa = 0 } = Λ max = Λ , meaning Q q ( v ) = q ( v, v ) + Λ = 0 ∈ F / Λ, proving (1).For statement (2), we assume Λ = F is a proper additive subgroup. By theadditivity and injectivity of χ , it suffices now to check that Q q = 0 implies ω q = 0. The condition Q q = 0 is equivalent to q ( v, v ) ∈ Λ for every v ∈ E . Itfollows that q ( u, v ) = q ( u + v, u + v ) − q ( u, u ) − q ( v, v ) − q ( v, u )= − q ( v, u ) mod Λ= − εq ( v, u ) mod Λ , where the last equality holds as q ( v, u ) − εq ( v, u ) ∈ Λ min ≤ Λ. Hence ω ( u, v ) ∈ Λfor every u, v ∈ E . Scaling v by a any element in F defines an ideal of Λ generatedby ω ( u, v ). As Λ has no non-trivial ideal, it follows that ω ( u, v ) = 0, which proves(2).Finally, we prove Theorem A. The proof will use two lemmas, which we stateand prove now. Lemma . (1) The assignment ω ( − ) : Form( E, σ, ε, Λ) −→ Herm σ,ε ( E ) of Theorem 2.5 is surjective unless σ = id and F has is a field of characteristic , in which case the image is the subspace Alt ( E ) of alternating bilinear forms.(2) If σ = id the assignment Q ( − ) : Form( E, σ, ε, Λ) −→ Set ( E, F / Λ) of Theorem 2.5 has image the mod Λ quadratic forms , i.e. the set maps Q : E → F / Λ such that Q ( av ) = a Q ( v ) for all v ∈ E , a ∈ F and B Q ( v, w ) = Q ( v + w ) − Q ( v ) − Q ( w ) is a bilinear function. In particular, if in addition Λ = 0 , then Q ( − ) has image Quad ( E ) . Proof.
To prove (1), note first that in the exceptional case ( σ = id in char-acteristic 2), the form ω q is indeed alternating, since we necessarily have ε = − ω q ( v, v ) = q ( v, v ) + εq ( v, v ) = 0 . Now, returning to the general case, let ω be any element of Herm σ,ε ( E ), assumedalternating in the special case. Arbitrarily pick a basis { x i } of E , and define q ( x i , x j ) = ω ( x i , x j ) if i < j,a i if i = j, i > j, extending sesquilinearly to obtain an element q ∈ Sesq σ ( E ). Here a i ∈ F is ascalar chosen so that a i + ε ¯ a i = ω ( x i , x i ) . This is possible in the special case by setting a i = 0. In the other cases, applyingProposition A.1 we getΛ min ( F , σ, − ε ) = { a + ε ¯ a | a ∈ F } = (cid:8) b ∈ F b = ε ¯ b (cid:9) = Λ max ( F , σ, − ε )and ω ( x i , x i ) lies in the right hand side, so such an a i does exists. The form q satisfies that ω q = ω in all cases.For (2), we now assume that σ = id. Given a formed space ( E, q ), we firstcheck that Q q is a mod Λ quadratic form, i.e. that B Q q ( v, w ) := Q q ( v + w ) − Q q ( v ) − Q q ( w )is bilinear. But B Q q ( x, y ) = ω q mod Λ, which is bilinear under the assumptionthat σ = id.Conversely, assume that Q : E → F / Λ is a quadratic form (mod Λ). We wantto show that it is in the image of Q ( − ) . Arbitrarily pick a basis { x i } of E , andpick b ij , c i ∈ F such that B Q ( x i , x j ) = b ij mod Λ and Q ( x i ) = c i mod Λ. Thendefine q ( x i , x j ) = b ij if i < j,c i if i = j, i > j, extending bilinearly to obtain an element q ∈ Sesq id ( E ). For each i , we have Q q ( x i ) = Q ( x i ) by construction. Now ω q ( x i , x j ) = q ( x i , x j ) + εq ( x j , x i ) = q ( x i , x j ) + q ( x j , x i ) mod Λas q ( x j , x i ) = εq ( x j , x i ) mod Λ min . Hence ω q ( x i , x j ) = B Q ( x i , x j ) mod Λ if i = j , while ω q ( x i , x i ) = 2 q ( x i , x i ) = 2 Q ( x i ) mod Λand B Q ( x i , x i ) = Q (2 x i ) − Q ( x i ) = 2 Q ( x i ) . ORMS OVER FIELDS AND WITT’S LEMMA 11
It follows that ω q = B Q , while Q q and Q coincide on a basis. This implies that Q q = Q because Q ( v + w ) = Q ( v ) + Q ( w ) + B Q ( v, w ) and likewise Q q ( v + w ) = Q q ( v ) + Q q ( w ) + ω q ( v, w ) mod Λ, finishing the proof. Lemma . If σ = id , there exists an a = a ( ε ) ∈ F × such that multiplicationby a induces an isomorphism Form(
E, σ, ε, Λ) ∼ = −→ Form(
E, σ, , a Λ) . Note that G ( E, q ) = G ( E, aq ) as subgroups of GL( E ). Proof.
Multiplication by a ∈ F takes X ( V, σ, ε,
Λ) to X ( V, σ, a ¯ a − ε, a Λ). Sowe need to check that there exists an a ∈ F such that ε = a − ¯ a . The existence ofsuch an element is given by applying Hilbert’s Theorem 90 to the field extension F of F σ . Proof of Theorem A.
We start by proving (1). So assume σ = id. Thenby Proposition A.1, we must have Λ = Λ max . Now by Lemma 2.7(1), the map ω ( − ) is injective and by Lemma 2.9 it is also surjective. Statement (1) thenfollows from Lemma 2.10.To prove (2), we now assume that σ = id, ε = − F . In particular,Λ = Λ max in all cases by Proposition A.1. Applying Lemma 2.7(1) again, wehave that the map ω ( − ) is injective and by Lemma 2.9 it is also surjective ontoHerm id , − ( E ) unless F has characteristic 2 in which case the image is the sub-group of alternating forms Alt( E ). Now statement (2) follows from the fact thatHerm id , − ( E ) and Alt( E ) are actually isomorphic if the field is not of character-istic 2, as ω ∈ Herm id , − ( E ) satisfies that ω ( v, v ) = − ω ( v, v ).For (3), we now assume that σ = id, ε = 1 and Λ = 0. By Proposition 2.7(2),the map Q ( − ) is injective, and by Lemma 2.9 it surjects onto the quadratic formsQuad( E ), which proves the result.Theorem A gives a description of almost all the possible formed space in thebetter-known terms of Hermitian, alternating or quadratic forms. As we saw inRemark 1.1, left is the case σ = id, p = 2, and 0 < Λ < F . We show now thatat least for F finite, this case is in fact redundant, producing the same isometrygroups as if Λ = 0: Proposition . Assume F is a perfect field of characteristic 2. Let σ = id and Λ < F be proper. Then X ( E, id , , Λ) = X ( E, id , , . Proof.
What we must show is that, if q ( v, v ) ∈ Λ for all v ∈ V , then in fact q ( v, v ) = 0 for all v . For any c ∈ F × , we have q ( c − v, c − v ) = c − q ( v, v ) ∈ Λ . Now, if q ( v, v ) is nonzero, then every element of F is of that form for some c ,because the squaring function on F is bijective. That contradicts Λ being aproper subset of F . Remark . The proposition is false for imperfect fields of characteristictwo. For instance, for F = F ( t ), we have proper inclusions X ( F , id , , < X ( F , id , , F ( t )) < X ( F , id , , F ( t )) . Indeed, the forms q ( a, b ) = ab and q ′ ( a, b ) = tab demonstrate properness of thetwo inclusions.
3. Witt’s Lemma
In this section, we give a proof of Witt’s lemma in the context of formed spaces,stated as Theorem 3.4, and its relative version, stated as Theorem 3.7.Given a vector space E , recall that σ E ∗ denotes the vector space of σ -skew-linear maps E → F , that is, additive maps f : E → F such that f ( av ) = ¯ af ( v ),with vector space structure defined pointwise from that of F . Recall also themaps ω q and Q q of Theorem 2.5 associated to a form q . Definition . Let E = ( E, q ) be a formed space.(1) The kernel of E is defined to be the kernel K ( E ) = ker( ♭ q ) of the associatedlinear map ♭ q : E → σ E ∗ , v ω q ( − , v ) . We say that (
E, q ) is non-degenerate if K ( E ) = 0.(2) The orthogonal complement U ⊥ of a subspace U ≤ E is the subspace con-sisting of all v ∈ E such that ω ( v, U ) = 0. That is, U ⊥ is the kernel of thecomposition E ♭ q −→ σ E ∗ incl ∗ −−−→ σ U ∗ . (3) The radical of E is the set R ( E ) = { v ∈ K ( E ) | Q q ( v ) = 0 } . (4) A subspace U of E is called isotropic if q | U = 0 (or equivalently, ω q | U = 0and Q q | U = 0). ORMS OVER FIELDS AND WITT’S LEMMA 13
Remark . (1) The radical R ( E ) is in fact a subspace of E , because Q q is additive on K ( E ), and satisfies Q q ( cv ) = c Q q ( v ) for c ∈ F , v ∈ E . If thecharacteristic of F is not 2, then R ( E ) = K ( E ), since equation ((2.1)) in thatcase gives that ω q | K ( E ) = 0 implies Q q | K ( E ) = 0.(2) Note that orthogonality is a symmetric relation: ω q ( v, w ) = 0 if and only if ω q ( w, v ) = 0, but note also that the orthogonal complement U ⊥ defined abovewill usually not be a complement in the sense of vector spaces. In fact, if U isisotropic, we actually have U ⊂ U ⊥ ! Example . To a Coxeter graph on n vertices, one associates an ( n × n )–real symmetric matrix (see eg., [8, Sec 2.3]), with corresponding symmetricbilinear form on R n . Hence we get an associated formed space ( R n , q ) over R with the parameters ( σ, ε, Λ) = (id , , q is non-degenerate if the group is finite, but can be degenerated when the group isinfinite; see eg., [8, Sec 2.5] for examples of Coxeter graphs whose associatedsymmetric bilinear forms are degenerate.The following result is a version of Witt’s Lemma. Witt originally consid-ered non-degenerate symmetric forms in characteristic not 2 [18]. His resultwas generalize by eg., Tits [15, Prop 2] or Artin [1, Thm 3.9], who consid-ered non-degenerate alternating and quadratic forms, Dieudonn´e [3, p21,22 and35], who considers non-degenerate hermitian and quadratic forms, and Magurn-Vaserstein-van der Kallen [10, Cor 8.3] or Petrov [12, Thm 1] who both workin the more general context of rings and have a different set of assumptions,including an assumption on the index of the form, which counts the numberof hyperbolic summands it contains. The paper [7] of Huang considers flags ofsubspaces in the case of non-degenerate symmetric forms in characteristic not 2. Theorem . Let ( E, q ) be a formed space. Suppose that f : U → W is a bijective isometry between two subspaces of E such that f (cid:0) U ∩K ( E ) (cid:1) = W ∩ K ( E ) . Then f can be extended to a bijective isometry of E . Inparticular, the group G ( E, q ) of bijective isometries of ( E, q ) acts transitively onthe set of isotropic subspaces U with given values of dim U and dim U ∩ R ( E ) . Regarding the last statement, observe that when U is isotropic, U ∩ R ( E ) = U ∩ K ( E ). Note also that for general U, W and f as in the statement, we havethat f ( U ∩ R ( E )) = W ∩ R ( E ) as f is an isometry that takes U ∩ K ( E ) to W ∩ K ( E ). The result implies that the group G ( E, q ) acts transitively on the elements ofany fixed rank of the building of isotropic subspaces P ( E ) = { W < E | W isotropic , R ( E ) < W } . This transitivity is used in [14] to study the homology of the group G ( E, q ).The proof of the theorem will use the following basic result about vectorspaces:
Lemma . Let
A, B ≤ V be any two subspaces of the same dimension. Thenthere exists a subspace L which is simultaneously complementary to both A and B . Notice this would be false with three subspaces instead of two in e.g. V = F . Proof.
By modding out A ∩ B from V , we may assume without loss ofgenerality that A ∩ B = 0. By restricting to A + B < V , we may assume that A + B = V . Pick any linear isomorphism f : A → B , and define g : A → V bysetting g ( u ) = u + f ( u ). We claim that L = Im( g ) ≤ V is the desired complement.Now, A ∩ L = 0 because f is injective and A, B are disjoint. Similarly B ∩ L = 0.Also B + L = V since it contains A , and A + L = V since it contains Im( f ) = B ,which finishes the proof.In the proof of Witt’s Lemma, we will repeatedly use the following construc-tion: Lemma . Suppose
A, B, C ≤ E are subspaces of a formed space ( E, q ) suchthat C ≤ A ⊥ ∩ B ⊥ and A ∩ C = 0 = B ∩ C . Then for any isometry f : A → B ,the map f ⊕ id : A ⊕ C → B ⊕ C is also an isometry. If A = h a i , B = h b i and f ( a ) = b , the same holds under the weaker assumption that C ≤ ( a − b ) ⊥ . Proof.
In both cases, it is easy to see that f ⊕ id preserves ω q . By The-orem 2.5, we thus just have to check that f preserves Q q . In the first case,we have Q q ( f ( a ) + c ) = Q q ( f ( a )) + Q q ( c ) + ω q ( f ( a ) , c ) = Q q ( a ) + Q q ( c ).Similarly, Q q ( a + c ) = Q q ( a ) + Q q ( c ). In the second case, we have instead Q q ( f ( a )+ c ) = Q q ( b )+ Q q ( c )+ ω q ( b, c ) = Q q ( a )+ Q q ( c )+ ω q ( a, c ) = Q q ( a )+ Q q ( c ).This proves the result in both cases. Proof of Theorem 3.4.
We will prove the theorem by induction on thedimension of U , starting with the case U ≤ K := K ( E ) . By our hypothesis, it follows that W ≤ K as well. By Lemma 3.5, we can finda subspace L ≤ E such that U ⊕ L = E = W ⊕ L. ORMS OVER FIELDS AND WITT’S LEMMA 15
We extend f to a bijection of E by setting f to be identity on L . As U ⊥ = W ⊥ = E , Lemma 3.6 shows that this is an isometry.Now, we may assume that U K. Then we can pick a codimension-one subspace
H < U , such that U ∩ K ≤ H. By induction hypothesis we may find an isometry of E which coincides with f on H ; by composing f with the inverse of such an isometry, we reduce to thecase where f | H = id H . Write U = H ⊕ h a i and W = H ⊕ h b i with b = f ( a ) , noting that a, b / ∈ K by our choice of H . The subspace ( b − a ) ⊥ ≤ E hascodimension at most 1 since it is the kernel of a linear map with 1-dimensionalcodomain. Case 1: ( b − a ) ⊥ = E . Then, by applying Lemma 3.5 again to the nonzerovectors a and b in the vector space E/H , we obtain a hyperplane M with H ≤ M < E , which is complementary to both h a i and h b i . By Lemma 3.6, f extendsto a bijective isometry on all of E by declaring it to be the identity on M as M ≤ ( b − a ) ⊥ . Case 2: ( b − a ) ⊥ < E is a hyperplane. Note that H ≤ ( b − a ) ⊥ , since ω q ( h, a ) = ω q ( h, b ) for all h ∈ H . Also, we have that a ∈ ( b − a ) ⊥ if andonly if b ∈ ( b − a ) ⊥ because ω q ( b, b ) = ω q ( a, a ), so ω q ( b, b − a ) = ω q ( b, b ) − ω q ( b, a ) = ω q ( a, a ) − ω q ( b, a ) = ω q ( a − b, a ) . Case 2.1: a, b / ∈ ( b − a ) ⊥ . This means that( b − a ) ⊥ ⊕ h a i = E = ( b − a ) ⊥ ⊕ h b i . As in case one, we can extend f by defining it to be the identity on ( b − a ) ⊥ using again Lemma 3.6 as this space is by definition orthogonal to b − a . Case 2.2: a, b ∈ ( b − a ) ⊥ . We first check that this implies that Q q ( b − a ) = 0 . Indeed, if the characteristic of F is two, this is true because Q q ( b − a ) = Q q ( b ) + Q q ( a ) − ω q ( b, a )= 2 Q q ( b ) − ω q ( a, a ) = q ( a, a ) − εq ( a, a ) = 0 mod Λ . If the characteristic of F is not 2, instead we observe that ω q ( b − a, b − a ) = 0 , and then recall that, since Λ min = Λ = Λ max in this case by Proposition A.1,the vanishing of ω q ( b − a, b − a ) implies that of Q q ( b − a ), as in the proof ofLemma 2.7(1).We have H, K ≤ ( b − a ) ⊥ and by the additional assumption also U = H ⊕h a i , W = H ⊕h b i are also subspaces of ( b − a ) ⊥ . Recall also that K ∩ U = K ∩ H = K ∩ W by our choice of H . Let L ≤ ( b − a ) ⊥ by a common complement of U and W inside ( b − a ) ⊥ that contains K . (Such a complement can be constructedby first picking a complement N of H ∩ K inside K and then using Lemma 3.5on ( b − a ) ⊥ /N .) Setting M = H ⊕ L ≤ ( b − a ) ⊥ , we have( b − a ) ⊥ = M ⊕ a = M ⊕ b, and by construction K, H ≤ M .As M is orthogonal to b − a , we can again extend f to an isometry of ( b − a ) ⊥ by setting it to be the identity map on M . We need to extend f to the whole of E . Recall that a, b / ∈ K , so both a ⊥ , b ⊥ < E are hyperplanes. We claim that neither of them contains M ⊥ . Assume to thecontrary that M ⊥ ⊂ a ⊥ . That implies that a ∈ a ⊥⊥ ≤ M ⊥⊥ = M + K = M , a contradiction since a isnot in M . Similar logic applies to b . So M ⊥ is contained in neither a ⊥ nor b ⊥ .So, M ⊥ ∩ a ⊥ , M ⊥ ∩ b ⊥ < M ⊥ have codimension one. Using Lemma 3.5 again, there is some vector v ∈ M ⊥ which is in neither of them. That is, v not orthogonal to either a or b . We breakinto two cases depending on whether or not v is orthogonal to b − a . Case 2.2.1: v / ∈ ( b − a ) ⊥ . We intend to extend f to all of E by setting f ( v ) = c ( b − a ) + dv ORMS OVER FIELDS AND WITT’S LEMMA 17 for some scalars c, d ∈ F with d = 0. This will be bijective, so we are done if wecan choose scalars yielding an isometry. Note that f ( v ) − v ∈ M ⊥ so ω q ( m, v ) = ω q ( m, f ( v ))for all m ∈ M . Now, ω q ( b, f ( v )) = dω q ( b, v ) . Since v is not orthogonal to a or b , ω q ( b, v ) , ω q ( a, v ) = 0 , so we can set d = ω q ( a, v ) ω q ( b, v ) = 0 to get ω q ( b, f ( v )) = ω q ( a, v ) as needed.Lastly, Q q ( f ( v )) = c ¯ cQ q ( b − a ) + d ¯ dQ q ( v ) + c ¯ dω q ( v, b − a ) . The first term is zero. In the last term, both d and ω q ( v, b − a ) are nonzero (sincewe have assumed that v is not orthogonal to b − a . So we can pick c = (1 − d ¯ d ) Q q ( v )¯ dω q ( v, b − a ) to get Q q ( f ( v )) = Q q ( v ) . This shows that f is an isometry of E . Case 2.2.2: v ∈ ( b − a ) ⊥ . Choose a complement z / ∈ ( b − a ) ⊥ . This time we intend to extend f to all of V by setting f ( z ) = z + c ( b − a ) + dv for some scalars c, d ∈ F . This will be bijective, so we are done if we can choosescalars yielding an isometry. Again, f ( z ) − z ∈ M ⊥ so ω q ( m, z ) = ω q ( m, f ( z ))for all m ∈ M . Now, ω q ( b, f ( z )) = ω q ( b, z ) + dω q ( b, v ) . Since v is not orthogonal to b , we can set d = ω q ( a, z ) − ω q ( b, z ) ω q ( b, v ) to get ω q ( b, f ( z )) = ω q ( a, z ) as needed.Lastly, Q q ( f ( z ))= Q q ( z ) + c ¯ cQ q ( b − a ) + d ¯ dQ q ( v ) + cω q ( z, b − a ) + dω q ( z, v ) + ¯ cdω q ( b − a, v )= Q q ( z ) + d ¯ dQ q ( v ) + cω q ( z, b − a ) + dω q ( z, v ) , since we have assumed that v is orthogonal to b − a . But z is not orthogonal to b − a , so we can set c = − d ¯ dQ q ( v ) + dω q ( z, v ) ω q ( z, b − a ) to get Q q ( f ( z )) = Q q ( z ) . Using Theorem 2.5, this shows that f is an isometry of E , which finishes theproof.The following result is a relative variant of Witt’s Lemma, which tells us whenwe can extend an isometry between two subspaces to an isometry of the ambientspace which is the identity on a fixed subspace A : Theorem . Let ( E, q ) be a formed space and A ≤ E a subspace. Suppose that f : U → W is a bijective isometry betweentwo subspaces of E such that(1) U ∩ A = W ∩ A and f | U ∩ A = id ;(2) f = id mod A ⊥ , i.e. f ( u ) − u ∈ A ⊥ for any u ∈ U ;(3) f (cid:0) U ∩ K ( E ) (cid:1) = W ∩ K ( E ) .Then f can be extended to a bijective isometry of E that restricts to the identityon A . The following result shows that the assumption that f is the identity modulo A ⊥ is actually necessary. Lemma . Let ( E, q ) be a formed space and A ≤ E any subspace. Anyisometry of E fixing A pointwise preserves A ⊥ and is the identity modulo A ⊥ . Proof.
Let f : E → E be an isometry fixing A pointwise. If v ∈ E and a ∈ A , then ω q ( f ( v ) , a ) = ω q ( f ( v ) , f ( a )) = ω q ( v, a ) , so f ( v ) − v ∈ A ⊥ , showing both that f preserves A ⊥ and that it is the identitymodulo A ⊥ . Proof of Theorem 3.7.
Let A be a complement of U ∩ A = W ∩ A inside A . We claim that the map ¯ f = f ⊕ id : U ⊕ A → W ⊕ A is an isometry.Indeed, ¯ f restricts to an isometry on both U and A . Let u ∈ U and a ∈ A .By assumption, we have f ( u ) = u + v with v ∈ A ⊥ . Hence ω q ( f ( u ) , f ( a )) = ω q ( u + v, a ) = ω q ( u, a ), so ¯ f preserves ω q . Also, Q q ( ¯ f ( u + a )) = Q q (( u + v )+ a ) = Q q ( f ( v )) + Q q ( a ) + ω q ( u + v, a ) = Q q ( v ) + Q q ( a ) + ω q ( u, a ) = Q q ( u + a ) wherewe used that f is an isometry. Hence, by Theorem 2.5, ¯ f is an isometry. As¯ f (( U ⊕ A ) ∩ K ( E )) = f ( U ∩ K ( E )) ⊕ ( A ∩ K ( E )) = ( W ⊕ A ) ∩ K ( E ), we canapply Theorem 3.4 to extend ¯ f to a bijective isometry ˆ f of E that takes U ⊕ A to W ⊕ A . By construction, ˆ f extends f and restricts to the identity on A ,which proves the result.The following corollary of the relative Witt’s Lemma was our motivation forwriting the result: Corollary . Let ( E, q ) be a formed space, and U, W , W isotropic sub-spaces containing R ( E ) . Suppose dim W = dim W and W + U ⊥ = W + U ⊥ = E . Then there is a bijective isometry of E sending W to W , and restricting tothe identity on U . Furthermore, an isomorphism f : W → W restricting to theidentity on R ( E ) can be extended to an isometry of E fixing U pointwise if andonly if f is the identity map modulo U ⊥ . For an isotropic subspace U containing the radical R ( E ), define the relativebuilding to be P ( E, U ) = (cid:8)
W < E W isotropic , R ( E ) < W, W + U ⊥ = E (cid:9) . The corollary implies in particular that the group A ( E, U ) := { g ∈ G ( E, q ) | g | U = id U } of the bijective isometries of ( E, q ) fixing U pointwise acts transitively on theelements of any given fixed rank. This is used in [14] to study the homology ofthe groups A ( E, U ).The proof of the corollary will use the following:
Lemma . Let ( E, q ) be a formed space and let U, W ≤ E be subspaces. If U is isotropic and W + U ⊥ = E , then U ∩ W ⊥ ≤ R ( E ) . In particular, if W isalso isotropic, then U ∩ W ≤ R ( E ) . Proof of Lemma 3.10.
Suppose that U ∩ W ⊥ = A . As A ⊂ W ⊥ , we alsohave W ⊂ A ⊥ . And A ⊂ U implies that U ⊥ ⊂ A ⊥ . So E = W + U ⊥ ⊂ A ⊥ . Thismeans that A ⊂ K ( E ). But A ⊂ U is isotropic, so Q q | A = 0 and we conclude A ≤ R ( E ) as claimed. The last claim follows from the fact that W ⊂ W ⊥ when W is isotropic. Proof of Corollary 3.9.
We want to apply Theorem 3.7. For this, weneed to construct a bijective isometry f : W → W satisfying the assumptionsof the theorem. As W and W are isotropic, any isomorphism will give a bijectiveisometry. Note now that, by Lemma 3.10, W i ∩ U ≤ R ( E ), and hence W i ∩ U = R ( E ) as R ( E ) is assumed to be contained in U, W and W . This gives condition(1) in the theorem. As W + U ⊥ = E = W + U ⊥ , we can pick a isomorphism f which is the identity modulo U ⊥ , and hence satisfies condition (2). As W ⊃R ( E ) ⊂ W , we can choose such an f which is the identity on R ( E ). This shows that condition (3) is also satisfied as W i ∩ K ( E ) = W i ∩ R ( E ) = R ( E ) as each W i is isotropic. Hence there exists a bijective isometry of E taking W to W and restricting to the identity on U .For the last statement, Lemma 3.8 shows that any isometry of E that fixes U necessarily is the identity modulo U ⊥ , which gives one direction. On the otherhand, if f is the identity on R ( E ) and the identity modulo U ⊥ , we see that itsatisfies the conditions of Theorem 3.7 as it is automatically an isometry, satisfiesconditions (2) and (3) by assumption, and condition (1) is satisfied by the aboveargument, which does not depend on f . Appendix A. The possible values of Λ Let ( F , σ ) be as above a field with chosen involution, where we again write¯ c = σ ( c ). If σ is not the identity, then F ⊃ F σ is a field extension of degree 2, where F σ denotes the fixed points of the involution.We fix as before a scalar ε ∈ F satisfying ε ¯ ε = 1.Recall thatΛ min = { a − ε ¯ a | a ∈ F } and Λ max = { a ∈ F | a + ε ¯ a = 0 } . Note first that Λ min ≤ Λ max , because ( a − ε ¯ a ) + ε ( a − ε ¯ a ) = 0.The following result gives the values of Λ min and Λ max in the case of fields,depending on whether the involution is trivial or not, the characteristic of thefield is even or not, and whether ε is equal to 1 , − Proposition
A.1 . Let F be a field of characteristic p , σ an involution on F and ε ∈ F such that ε ¯ ε = 1 . Then the values of Λ min and Λ max are given by thefollowing table: σ = id? p = 2? ε Λ min Λ max yes no yes no − F F yes yes F no − F σ F σ no = − ε ) F σ (1 + ε ) F σ . Note in particular that in most cases Λ min = Λ max in which case there isonly one possible choice for Λ min ≤ Λ ≤ Λ max . The only situation whereΛ min = Λ max is when σ is the identity and the characteristic is equal to 2, inwhich case Λ can be any additive subgroup of F . ORMS OVER FIELDS AND WITT’S LEMMA 21
Remark
A.2 . In the Hermitian case, we take σ = id and ε = 1. If char( F ) =2, then ε = 1 = − F σ , while if char( F ) = 2, then ε = 1 = − F σ , which also identifies with F σ as 2 is invertible in thatcase. Proof.
Suppose first that σ = id. This implies that ε = ± . (In characteristic 2, this gives only one possibility.) If ε = 1, then Λ min = 0.Otherwise, (1 − ε ) is invertible and Λ min = F . This gives Λ min in the first threecases in the table. If ε = −
1, we have that Λ max = F . Otherwise, (1 + ε ) isinvertible and Λ max = 0. This completes the proof of the first three cases in thetable.Now consider the case when σ is nontrivial. The set Λ min is the image of themap F → F , a a − ε ¯ a This map is F σ -linear, so its image is an F σ -subspace of the two-dimensionalvector space F (over F σ ). We claim that this map has rank 1.First assume that ε = −
1. Then its kernel contains the element (1 + ε ), soit is not injective. However, it cannot be identically zero, because (by applyingthe map to 1 ∈ F σ ) that would imply ε = 1 and σ = id , contradiction ourassumption. Next assume that ε = −
1. Then our map a a + ¯ a is the tracemap of the field extension. This map is well-known to be nonzero from charactertheory. It cannot be surjective since its image lies in F σ . In either case, weconclude that Λ min is a one-dimensional subspace of F over F σ .We will compute Λ max to identify Λ min . By definition, Λ max is the kernel ofthe map F → F σ , a a + ε ¯ a. Since this is the same map as before but with ε replaced by − ε , we have alreadyshowed that it has rank 1 as an F σ -linear endomorphism of F . So its kernel is aone-dimensional subspace. Because it must contain Λ min , we haveΛ max = Λ min = ( F σ if ε = − , (1 + ε ) F σ if ε = − . as subsets of F . Acknowledgements
The authors were both supported by the Danish National Research Founda-tion through the Centre for Symmetry and Deformation (DNRF92). The sec-ond author was also supported by the European Research Council (ERC) underthe European Union’s Horizon 2020 research and innovation programme (grant agreement No. 772960), the Heilbronn Institute for Mathematical Research, andwould like to thank the Isaac Newton Institute for Mathematical Sciences, Cam-bridge, for support and hospitality during the programme Homotopy Harness-ing Higher Structures, where this paper was finalized (EPSRC grant numbersEP/K032208/1 and EP/R014604/1).
References
1. Artin, E.,
Geometric algebra , Wiley Classics Library, John Wiley & Sons, Inc., New York,1988, Reprint of the 1957 original, A Wiley-Interscience Publication.2. Bak, A.,
On modules with quadratic forms , in “Algebraic K-Theory and its GeometricApplications (Conf., Hull, 1969)”, Springer, Berlin, 1969, pp. 55–66.3. Dieudonn´e, J. A.,
La g´eom´etrie des groupes classiques , Springer-Verlag, Berlin-New York,1971, Troisi`eme ´edition, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 5.4. Fiedorowicz, Z. and Priddy, S.,
Homology of classical groups over finite fields and theirassociated infinite loop spaces , Lecture Notes in Mathematics, vol. 674, Springer, Berlin,1978.5. Friedrich, N.,
Homological stability of automorphism groups of quadratic modules andmanifolds , Doc. Math. 22 (2017), 1729–1774.6. Galatius, S. and Randal-Williams, O.,
Homological stability for moduli spaces of highdimensional manifolds. I , J. Amer. Math. Soc. 31 (2018), no. 1, 215–264.7. Huang, H.,
Some extensions of Witt’s theorem , Linear Multilinear Algebra 57 (2009), no. 4,321–344.8. Humphreys, J. E.,
Reflection groups and Coxeter groups , Cambridge Studies in AdvancedMathematics, vol. 29, Cambridge University Press, Cambridge, 1990.9. Lam, T. Y.,
Introduction to quadratic forms over fields , Graduate Studies in Mathematics,vol. 67, American Mathematical Society, Providence, RI, 2005.10. Magurn, B. A., van der Kallen, W., and Vaserstein, L. N.,
Absolute stable rank and Wittcancellation for noncommutative rings , Invent. Math. 91 (1988), no. 3, 525–542.11. Mirzaii, B.,
Homology stability for unitary groups. II , K -Theory 36 (2005), no. 3-4, 305–326(2006).12. Petrov, V., Overgroups of unitary groups , K -Theory 29 (2003), no. 3, 147–174.13. Randal-Williams, O. and Wahl, N., Homological stability for automorphism groups , Adv.Math. 318 (2017), 534–626.14. Sprehn, D. and Wahl, N.,
Homological stability for classical groups , preprintarXiv:1812.08742, 2018.15. Tits, J.,
Formes quadratiques, groupes orthogonaux et alg`ebres de Clifford , Invent. Math.5 (1968), 19–41.16. Wall, C. T. C.,
On the axiomatic foundations of the theory of Hermitian forms , Proc.Cambridge Philos. Soc. 67 (1970), 243–250.17. Wall, C. T. C.,
On the classification of Hermitian forms. II. Semisimple rings , Invent.Math. 18 (1972), 119–141.18. Witt, E.,
Theorie der quadratischen Formen in beliebigen K¨orpern , J. Reine Angew. Math.176 (1937), 31–44.
ORMS OVER FIELDS AND WITT’S LEMMA 23
Department of MathematicsUniversity of CopenhagenUniversitetsparken 52100 CopenhagenDenmark
E-mail : [email protected]
Department of MathematicsUniversity of CopenhagenUniversitetsparken 52100 CopenhagenDenmark