Bourbaki

Theorem 1 (Witt)

Tag 0DKZ, permanent. Cite it as [Bourbaki, Tag 0DKZ].

Algebra, chapter IX, § 4. Sous-espaces totalement isotropes. Théorème de Witt

Let E and E′E' be two vector spaces of finite dimensions, provided respectively with two non-degenerate ε\varepsilon-hermitian forms Φ\Phi and Φ′\Phi' satisfying condition (T) of No. 2 (resp. with two non-degenerate quadratic forms Q and Q'), and isomorphic for these structures. Given an arbitrary subspace F of E, every injective metric homomorphism from F into E' extends to a metric isomorphism from E onto E'.

Using the given isomorphism from E onto E', one sees that it suffices to show that every injective metric homomorphism u from F into E extends to a metric automorphism of E. Let us remark that if, for i = 1, 2, F_i is a subspace of E and u_i a metric homomorphism from F_i into E, such that F_1 ∩ F_2 = {0} and that Φ(u_1(x_1), u_2(x_2)) = Φ(x_1, x_2) for x_i ∈ F_i (i = 1, 2), then the homomorphism ψ : x_1 + x_2 → u_1(x_1) + u_2(x_2) from F_1 + F_2 into E which extends u_1 and u_2 is metric: indeed, whatever x_i, y_i in F_i (i = 1, 2), the expansion of each of the expressions Φ(x_1 + x_2, y_1 + y_2) and Φ(u_1(x_1) + u_2(x_2), u_1(y_1) + u_2(y_2)) (resp. Q(x_1 + x_2) and Q(u_1(x_1) + u_2(x_2))) contains four (resp. three) terms equal each to each according to the assumptions made. Moreover, if u_1 and u_2 are injective and if u_1(F_1) ∩ u_2(F_2) = {0}, then ψ is injective.

1) We first prove the Witt theorem in the case where the set of elements invariant by u is a hyperplane U of F. The set of vectors of the form u(x) − x with x ∈ F is then a right D. If F' is a subspace orthogonal to D such that F' ∩ F = F' ∩ u(F) = {0}, we shall have Φ(u(x), y) = Φ(x, y) for x ∈ F and y ∈ F' ; our initial remark therefore applies to u and to the identical mapping of F' into E, showing that u extends to F + F' by leaving the points of F' fixed; the set of vectors of the form u(x) − x (x ∈ F + F') is still the right D. Now we have, for x ∈ F, y ∈ F

(5) Φ(u(x),u(y)−y)=Φ(u(x),u(y))−Φ(u(x),y)=Φ(x−u(x),y)\Phi(u(x), u(y) - y) = \Phi(u(x), u(y)) - \Phi(u(x), y) = \Phi(x - u(x), y),

which, when x ∈ U (that is to say when u(x) = x), shows that x ∈ D^0; in other words we have U ⊂ D^0. We shall distinguish two cases:

a) F⊄D0F \not\subset D^0. Formula (5) shows that u(F)u(F) is not contained in D0D^0, hence F∩D0=u(F)∩D0=UF \cap D^0 = u(F) \cap D^0 = U. One may then take for F' a supplementary subspace of U in D^0; since F + F' contains the hyperplane D^0 and is distinct from it, we have F + F' = E, and in this case we have found the required extension of u to E.

b) F ⊂ D^0. Formula (5) shows that u(F) ⊂ D^0, and therefore that

D ⊂ D^0 ; the right D is therefore isotropic (resp. singular, since we have Q(u(x) − x) = Q(u(x)) − Φ(x, u(x)) + Q(x) = 2Q(x) − Φ(x, x) = 0 for x ∈ F). We shall show that, under these conditions, there exists a subspace F' of D^0 which is supplementary to F and u(F) in D^0. This is immediate if F = u(F). Otherwise, let x and y be vectors such that x ∈ F, x ∉ U, y ∈ u(F), y ∉ U; we then have F = U + Ax, u(F) = U + Ay, and F does not contain x + y otherwise y = (x + y) − x would belong to F ∩ u(F) = U; one sees analogously that x + y does not belong to u(F); thus the right A(x + y) is supplementary to F and u(F) in the subspace F + u(F); it suffices then to put F' = A(x + y) + G where G is supplementary to F + u(F) in D^0. This being so, we have F + F' = u(F) + F' = D^0, and, in this case, what was said at the beginning of 1) shows that there exists an extension of u to the hyperplane D^0 of E, and that D^0 is stable for this extension.

We are therefore reduced to the case where F is the hyperplane D^0 and where u is an automorphism of F. We prove that, for every z ∈ E, there exists z' ∈ E such that

(6)

Φ(u(x),z′)=Φ(x,z) \Phi(u(x), z') = \Phi(x, z)
for every x ∈ F; indeed the linear form x → Φ(u^{-1}(x), z) on F is the restriction of a linear form on E, a form which is of the type x → Φ(x, z') since Φ is non-degenerate; hence (6) holds. Moreover, if z ∉ F, there exists a vector z' ∈ E satisfying (6) and such that Φ(z', z') = Φ(z, z) (resp. Q(z') = Q(z)): indeed formula (6) remains valid if one adds to z' an element u(y) − y (y ∈ F) of D since F = D^0, and Lemma 1 of No. 2 allows one to conclude since z is not orthogonal to D. Our initial remark then shows that there exists a metric homomorphism ϕ from F + Az = E into E which extends u and which transforms z into z'. Since Φ is non-degenerate, ϕ is the required metric automorphism of E.

2) In the general case, we reason by induction on r = dim F. The case r = 0 is trivial. Let then r > 0, that is to say F ≠ {0}, and let U be a hyperplane of F. The restriction u_0 of u to U extends, by the induction hypothesis, to a metric automorphism ϕ_0 of E. If ϕ_0 extends u, the theorem is proved. Otherwise U is the set of elements invariant by v_0^{-1}u, and there exists, according to 1), a metric automorphism v_1 of E extending v_0^{-1}u. The automorphism v_0v_1 is then the required extension of u. QED.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Nothing.

Cites

Nothing.

Translated with machine help, not yet reviewed by a person.