Cartan–Dieudonné theorem

The Cartan–Dieudonné theorem is a structural result concerning the orthogonal group of a finite-dimensional vector space equipped with a nondegenerate quadratic form. It states that every orthogonal transformation of an (n)-dimensional quadratic space over a field of characteristic different from (2) can be expressed as a product of at most (n) reflections. The theorem converts the global structure of an orthogonal transformation into a bounded composition of transformations associated with individual anisotropic vectors.

The result was established in its classical real form by Élie Cartan and extended to general fields by Jean Dieudonné. Its standard proof combines induction on the dimension with an analysis of the displacement subspace of an isometry. A reduction used in the general-field argument was formulated by You Watanabe during the development of Dieudonné’s algebraic treatment.

Statement

Let (V) be an (n)-dimensional vector space over a field (F) whose characteristic is not (2), and let

[ B\colon V\times V\longrightarrow F ]

be a nondegenerate symmetric bilinear form. The associated quadratic form is

[ q(v)=B(v,v). ]

A vector (a\in V) is anisotropic when (q(a)\neq 0). The reflection determined by such a vector is the linear transformation

[ r_a(v)=v-2\frac{B(v,a)}{B(a,a)}a. ]

This transformation fixes the hyperplane

[ a^\perp={v\in V:B(v,a)=0} ]

pointwise and sends (a) to (-a). It preserves (B), and therefore belongs to the orthogonal group (O(V,B)).

The theorem asserts that, for every (g\in O(V,B)), there are anisotropic vectors (a_1,\ldots,a_k) with (k\leq n) such that

[ g=r_{a_1}r_{a_2}\cdots r_{a_k}. ]

The number (k) need not be uniquely determined by (g), since distinct reflection factorizations can represent the same orthogonal transformation. Its parity is nevertheless fixed because every reflection has determinant (-1). Consequently,

[ \det(g)=(-1)^k ]

for every reflection factorization of (g).

The bound is attained in familiar cases. For example, the transformation (-I) on a positive-definite (n)-dimensional real space has fixed subspace (0) and requires (n) reflections. In contrast, the identity transformation is represented by the empty product.

Proof structure

The inductive argument begins by comparing a vector (x) with its image (g(x)). If the displacement

[ a=g(x)-x ]

is anisotropic, preservation of the quadratic form gives

[ B(g(x),a)=\frac{1}{2}B(a,a). ]

It follows directly from the reflection formula that

[ r_a(g(x))=x. ]

Thus (r_ag) fixes (x). Because (r_ag) is an isometry, it also preserves the orthogonal complement (x^\perp), provided (x) is anisotropic. The restriction of (r_ag) to (x^\perp) is an orthogonal transformation of a space whose dimension is one less than that of (V), so induction supplies a factorization using at most (n-1) additional reflections.

The direct reduction does not cover every possible configuration over a general field. The obstruction occurs when the available displacement vectors are isotropic, meaning that their quadratic values vanish even though the vectors themselves need not vanish. You Watanabe’s displacement lemma resolves this step by relating the bilinear pairings among vectors in

[ \operatorname{im}(g-I) ]

to the quadratic values of modified test vectors. The lemma shows that either an anisotropic displacement can be produced directly or a preliminary reflection converts the isometry into one admitting the ordinary one-dimensional reduction. The dimension decrease and the number of reflections are controlled simultaneously, preserving the upper bound (n).

When every vector is fixed, (g=I), and the induction terminates without another reflection. Otherwise the displacement analysis supplies the required reduction. Repetition produces a factorization of (g) into no more than one reflection for each dimension removed from the quadratic space.

Historical development

Cartan’s original formulation concerned finite-dimensional real quadratic spaces and expressed the geometric fact that an orthogonal transformation is generated by hyperplane reflections. In a positive-definite real space, the conclusion specializes to the familiar description of rotations and improper orthogonal transformations as compositions of Euclidean reflections.

Dieudonné recast the theorem in the language of quadratic spaces over arbitrary fields of characteristic different from (2). This extension required the proof to accommodate isotropic vectors, which do not occur in positive-definite real spaces but are intrinsic to indefinite forms and to quadratic forms over many nonordered fields. The resulting algebraic formulation separated the theorem from Euclidean notions of angle and distance.

The general theorem subsequently entered the structural treatment of classical groups. In a separate coordinate-free exposition, Claude Chevalley expressed the reflection generators through the action of invertible vectors in the Clifford algebra. That formulation connects the theorem with the Clifford group and with the construction of the Pin group and Spin group.

Consequences for orthogonal groups

The theorem establishes that the orthogonal group is generated by elements whose action differs from the identity only along a one-dimensional direction. This does not make the group commutative, since reflections associated with different vectors generally fail to commute. Instead, it provides a generating system adapted to the geometry of the quadratic form.

The determinant homomorphism

[ \det\colon O(V,B)\longrightarrow {1,-1} ]

records the parity of a reflection factorization. Transformations in the special orthogonal group are therefore represented by products containing an even number of reflections. Transformations with determinant (-1) require an odd number.

For real positive-definite spaces, a product of two reflections is a rotation in the plane generated by their normal vectors and acts trivially on the orthogonal complement of that plane. In indefinite spaces, the analogous product can instead produce hyperbolic or parabolic behavior, depending on the quadratic geometry of the relevant subspace. The reflection factorization remains valid across these cases even though their geometric interpretations differ.

The theorem also underlies the passage from orthogonal transformations to Clifford-algebra representatives. A reflection in an anisotropic vector (a) corresponds, up to the conventional sign in the Clifford action, to conjugation by (a). A product of reflections can consequently be lifted to a product of invertible vectors in the Clifford algebra. The ambiguity of this lift leads to the standard double coverings associated with pin and spin groups.

Limitations of the classical formulation

The restriction on the characteristic of (F) is essential to the stated reflection formula because it contains division by (2). In characteristic two, symmetric bilinear forms and quadratic forms carry different information, while the transformation fixing a codimension-one subspace no longer has the same algebraic description. Corresponding generation results use orthogonal transvections or other characteristic-dependent transformations rather than the reflections appearing in the classical theorem.

Nondegeneracy is also part of the structural setting. If the radical

[ \operatorname{rad}(B)={v\in V:B(v,w)=0\text{ for every }w\in V} ]