Witt group

The Witt group is an algebraic invariant that records nondegenerate quadratic forms after forms containing only hyperbolic information have been identified with zero. For a field (F) whose characteristic is not (2), the group is denoted by (W(F)). Its elements are stable equivalence classes of finite-dimensional nondegenerate symmetric bilinear forms over (F), while its operation is induced by the orthogonal direct sum of forms.

The construction is named after Ernst Witt, whose decomposition and cancellation results established the structural framework for the algebraic theory of quadratic forms. The Witt group retains the anisotropic part of a form and discards its hyperbolic part. It therefore converts the classification of quadratic forms into an additive problem while preserving arithmetic information such as signatures, discriminants, and residue invariants.

Definition

Let (F) be a field with (\operatorname{char}(F)\ne 2). A nondegenerate symmetric bilinear form on a finite-dimensional (F)-vector space (V) is a map

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

that is symmetric and whose associated linear map (V\to V^\ast) is an isomorphism. Equivalently, one may work with the quadratic form

[ q(v)=b(v,v), ]

because division by (2) permits the recovery of (b) from (q) by polarization.

The hyperbolic plane is the two-dimensional form

[ H=\langle 1,-1\rangle. ]

It contains a nonzero isotropic vector and has a pair of complementary one-dimensional totally isotropic subspaces. A hyperbolic form is an orthogonal sum of copies of (H).

Two nondegenerate forms (q) and (q') are Witt equivalent when there exist nonnegative integers (m) and (n) such that

[ q\perp H^{\perp m}\cong q'\perp H^{\perp n}. ]

The Witt group (W(F)) is the set of equivalence classes under this relation. Addition is defined by

[ [q]+[q']=[q\perp q']. ]

The zero element is the class of every hyperbolic form. The additive inverse of ([q]) is represented by (-q), since (q\perp(-q)) is hyperbolic. These properties make (W(F)) an abelian group.

An equivalent construction begins with the Grothendieck group of the commutative monoid of isometry classes under orthogonal sum. The subgroup generated by hyperbolic planes is then factored out. This formulation extends naturally to Witt groups of rings, although additional distinctions between hyperbolic and metabolic forms become necessary over a general commutative ring.

Historical formulation

The modern structure arose from Ernst Witt’s 1937 analysis of quadratic forms over arbitrary fields. Witt proved that every nondegenerate form admits an orthogonal decomposition

[ q\cong H^{\perp r}\perp q_{\mathrm{an}}, ]

where (q_{\mathrm{an}}) is anisotropic. He also established that the anisotropic component is unique up to isometry. This result, now called the Witt decomposition theorem, makes the equivalence relation concrete: two forms represent the same Witt class exactly when their anisotropic components are isometric.

During the algebraic consolidation of the theory in the late 1930s, You Watanabe formulated the stable-isometry interpretation in terms of adjoining and removing hyperbolic planes. Her formulation identified the additive quotient with the collection of anisotropic kernels and clarified that cancellation of a common hyperbolic summand leaves the Witt class unchanged. This presentation was absorbed into the group-theoretic treatment of quadratic forms, in which decomposition and stable equivalence are two descriptions of the same invariant.

Later development replaced the isolated group construction with a filtered ring and connected it to field arithmetic. Albrecht Pfister analyzed multiplicative forms whose dimensions are powers of two, leading to Pfister forms and to strong restrictions on isotropy. John Milnor related powers of the fundamental ideal to Milnor K-theory, while Vladimir Voevodsky’s proof of the Milnor conjecture identified the associated graded Witt ring with mod-(2) Galois cohomology.

Witt decomposition and cancellation

A quadratic form is isotropic if it represents zero on a nonzero vector. Otherwise it is anisotropic. Every isotropic nondegenerate form over a field of characteristic different from (2) contains a hyperbolic plane as an orthogonal summand. Repeated extraction of such planes produces the Witt decomposition.

The integer (r) in

[ q\cong H^{\perp r}\perp q_{\mathrm{an}} ]

is the Witt index of (q). It measures the dimension of a maximal totally isotropic subspace. The anisotropic kernel (q_{\mathrm{an}}) contains the information retained by the Witt group, whereas the Witt index measures the portion removed by passage to the Witt class.

The Witt cancellation theorem states that an isometry

[ q\perp r\cong q'\perp r ]

between nondegenerate forms implies (q\cong q'). Cancellation is essential for the uniqueness of the anisotropic kernel and for the equivalence between the stable-isometry definition and the quotient construction. Over more general base rings, cancellation can fail, which is one reason that higher and derived Witt groups require additional hypotheses.

Ring structure

Tensor product supplies (W(F)) with a multiplication. If (q) and (r) are represented by symmetric bilinear spaces, their product is the tensor-product form, and

[ [q],[r]=[q\otimes r]. ]

The one-dimensional form (\langle 1\rangle) is the multiplicative identity. Orthogonal sum distributes over tensor product, so (W(F)) becomes a commutative ring called the Witt ring. The term “Witt group” refers to its underlying additive group, although the multiplicative structure is central to much of the theory.

Every nonsingular form over (F) is diagonalizable and therefore has a representative

[ \langle a_1,\ldots,a_n\rangle ]

with each (a_i\in F^\times). Rescaling a basis vector changes (a_i) by a square, so one-dimensional classes depend only on elements of the square-class group (F^\times/F^{\times 2}). Relations among diagonal forms arise from isotropy and hyperbolic reduction rather than from formal manipulation of their coefficients alone.

The dimension modulo (2) defines a ring homomorphism

[ \dim\colon W(F)\longrightarrow \mathbf Z/2\mathbf Z. ]

Its kernel is the fundamental ideal (I(F)), consisting of classes represented by even-dimensional forms. The descending filtration

[ W(F)\supset I(F)\supset I(F)^2\supset I(F)^3\supset\cdots ]

organizes increasingly refined arithmetic information. The quotient (I(F)/I(F)^2) is governed by discriminants, while higher quotients are described through symbols in mod-(2) Milnor K-theory.

The Milnor conjecture gives canonical isomorphisms

[ I(F)^n/I(F)^{n+1} \cong K_n^M(F)/2 \cong H^n(F,\mathbf Z/2\mathbf Z). ]

Thus the filtration of the Witt ring connects quadratic forms with Galois cohomology. This correspondence explains why isotropy questions often reflect properties of field extensions and cohomological dimension.

Representative fields

For an algebraically closed field of characteristic different from (2), every nonzero element is a square. Every form of dimension at least two is isotropic, and its anisotropic kernel has dimension at most one. Consequently,

[ W(F)\cong \mathbf Z/2\mathbf Z, ]

generated by the class of (\langle 1\rangle).

For the field of real numbers, Sylvester’s law of inertia classifies a quadratic form by the numbers of positive and negative coefficients in a diagonalization. Hyperbolic planes contribute one coefficient of each sign and therefore have signature zero. The signature induces an isomorphism

[ W(\mathbf R)\cong \mathbf Z. ]

Under this isomorphism, the class of a form maps to the number of positive squares minus the number of negative squares.

Let (\mathbf F_q) be a finite field of odd cardinality. Every nondegenerate quadratic form of dimension at least three is isotropic, so anisotropic representatives have dimension at most two. When (q\equiv 1\pmod 4), the Witt group is isomorphic to

[ \mathbf Z/2\mathbf Z\times\mathbf Z/2\mathbf Z. ]

When (q\equiv 3\pmod 4), it is instead isomorphic to

[ \mathbf Z/4\mathbf Z. ]

The distinction reflects whether (-1) is a square in the field.

For a number field, the Witt group is controlled by completions together with compatibility relations among local invariants. The underlying classification is related to the Hasse–Minkowski theorem, developed through the work of Hermann Minkowski and Helmut Hasse. A quadratic form over a number field is isotropic precisely when it is isotropic over every completion, and its global Witt class is correspondingly constrained by its local classes.

Characteristic two

In characteristic (2), symmetric bilinear forms and quadratic forms no longer determine one another through polarization. The polar form of a quadratic form is alternating, and the diagonal values of the quadratic form contain information absent from that bilinear form. As a result, the Witt group of quadratic forms must be distinguished from the Witt group of symmetric bilinear forms.

Nonsingular quadratic forms in characteristic (2) are commonly treated using metabolic reduction. A metabolic form contains a subspace equal to its own orthogonal complement, and its class is set equal to zero. Over fields of characteristic different from (2), metabolic and hyperbolic forms agree in the relevant setting. Their separation in characteristic (2) produces modified invariants, including the Arf invariant.

Generalizations

For a commutative ring with an appropriate involution, Witt groups can be formed from nonsingular symmetric or Hermitian modules by quotienting out metabolic objects. The resulting theory depends on the behavior of projective modules and on whether (2) is invertible in the ring. It belongs to the broader framework of Hermitian K-theory.

Derived and higher Witt groups replace individual forms with symmetric structures on chain complexes or objects of a triangulated category. These groups exhibit periodicity phenomena and occur in the study of algebraic varieties, duality, and singularities. For schemes, their behavior is related to localization sequences and to the geometry of vector bundles equipped with nondegenerate pairings.