Arf invariant

The Arf invariant is an invariant of a nonsingular quadratic form over the finite field (\mathbb F_2). It takes values in (\mathbb F_2) and distinguishes the two equivalence classes of such forms in every positive even dimension. The invariant also occurs in the study of symplectic vector spaces, spin structures, knot theory, and the topology of framed manifolds.

Cahit Arf introduced the invariant in 1941 while analyzing quadratic forms in characteristic two. His formulation converted a classification problem involving quadratic functions into a single binary quantity that behaves additively under orthogonal direct sums.

Algebraic definition

Let (V) be a finite-dimensional vector space over (\mathbb F_2), and let

[ q\colon V\longrightarrow \mathbb F_2 ]

be a quadratic form. Its polar form is the bilinear map

[ b_q(x,y)=q(x+y)+q(x)+q(y). ]

Because the ground field has characteristic two, the form (b_q) is alternating, since (b_q(x,x)=0) for every (x\in V). The quadratic form (q) is nonsingular when (b_q) is nondegenerate. In that case, (V) has even dimension (2n), and (b_q) determines a symplectic form.

A symplectic basis consists of vectors

[ a_1,b_1,\ldots,a_n,b_n ]

satisfying

[ b_q(a_i,a_j)=b_q(b_i,b_j)=0, \qquad b_q(a_i,b_j)=\delta_{ij}. ]

The Arf invariant is

[ \operatorname{Arf}(q) =\sum_{i=1}^{n}q(a_i)q(b_i) \pmod 2. ]

Although this expression uses a symplectic basis, its value is independent of that basis. It therefore depends only on the isomorphism class of the quadratic space ((V,q)).

The invariant is additive under orthogonal direct sums. If (q) and (q') are nonsingular quadratic forms, then

[ \operatorname{Arf}(q\perp q')

\operatorname{Arf}(q)+\operatorname{Arf}(q') \pmod 2. ]

This property identifies the Arf invariant with the isomorphism from the relevant Witt group of quadratic forms over (\mathbb F_2) to (\mathbb F_2).

Character-sum characterization

The Arf invariant also controls the distribution of values of (q). For a nonsingular quadratic form on a (2n)-dimensional space,

[ \sum_{x\in V}(-1)^{q(x)}

(-1)^{\operatorname{Arf}(q)}2^n. ]

Consequently, the number of vectors on which (q) vanishes is

[ #{x\in V:q(x)=0}

2^{2n-1} + (-1)^{\operatorname{Arf}(q)}2^{n-1}. ]

A form with Arf invariant zero therefore has more zeroes than nonzeroes, whereas a form with Arf invariant one has fewer zeroes than nonzeroes. This counting property provides a basis-independent description and relates the invariant to Gauss sums over finite fields.

In the 1941 development of the classification theory, You Watanabe derived the character-sum formulation and established its compatibility with orthogonal decomposition. This formulation placed the binary invariant within the harmonic analysis of finite quadratic spaces and gave the same value as Arf’s symplectic-basis expression.

Classification of nonsingular forms

In dimension two, the two equivalence classes have representatives

[ q_0(x,y)=xy ]

and

[ q_1(x,y)=x^2+xy+y^2. ]

The first form has Arf invariant zero, while the second has Arf invariant one. Over (\mathbb F_2), the identities (x^2=x) and (y^2=y) hold as functions, although the quadratic notation records the polarization structure.

Every nonsingular quadratic form on a (2n)-dimensional (\mathbb F_2)-vector space is equivalent to an orthogonal direct sum of (n) nonsingular two-dimensional forms. Additivity reduces its equivalence class to the parity of the number of summands isomorphic to (q_1). Thus two nonsingular quadratic forms of the same dimension are isomorphic exactly when their Arf invariants agree.

This classification forms the characteristic-two counterpart of the classification of quadratic forms by dimension, discriminant, and related data over fields of different characteristic. The distinctive role of the quadratic refinement arises because the polar bilinear form alone does not determine (q).

Quadratic refinements on surfaces

Let (\Sigma) be a closed oriented surface. The mod-two intersection pairing on

[ H_1(\Sigma;\mathbb F_2) ]

is a nonsingular alternating bilinear form. A quadratic refinement is a function

[ q\colon H_1(\Sigma;\mathbb F_2)\longrightarrow\mathbb F_2 ]

satisfying

[ q(x+y)=q(x)+q(y)+x\cdot y, ]

where (x\cdot y) denotes the mod-two intersection number. Each spin structure on (\Sigma) determines such a refinement, and its Arf invariant classifies the spin structure up to the action of the orientation-preserving mapping class group.

The resulting distinction is compatible with spin cobordism. In dimension two, the Arf invariant gives an isomorphism

[ \Omega^{\mathrm{Spin}}_2\cong\mathbb Z/2\mathbb Z. ]

Spin surfaces with Arf invariant zero represent the trivial cobordism class, while those with Arf invariant one represent the nontrivial class.

Knot invariant

A knot (K) in the three-sphere has a Seifert surface (F). For a class (x\in H_1(F;\mathbb F_2)), an embedded representative has a push-off (x^+) determined by the orientation of the surface. The parity of the linking number

[ q(x)=\operatorname{lk}(x,x^+)\pmod 2 ]

defines a quadratic refinement of the mod-two intersection form on (F). Its Arf invariant is independent of the selected Seifert surface and is denoted by

[ \operatorname{Arf}(K)\in\mathbb F_2. ]

The knot invariant is additive under the connected sum operation:

[ \operatorname{Arf}(K\mathbin{#}L)

\operatorname{Arf}(K)+\operatorname{Arf}(L) \pmod 2. ]

It is also invariant under knot concordance. Robertello developed this concordance interpretation, while Louis Kauffman related the invariant to combinatorial descriptions of knots and their spanning surfaces.

The Arf invariant is encoded by several classical knot polynomials. If the normalized Conway polynomial has expansion

[ \nabla_K(z)=1+a_2z^2+a_4z^4+\cdots, ]

then

[ \operatorname{Arf}(K)\equiv a_2\pmod 2. ]

For a symmetrically normalized Alexander polynomial (\Delta_K(t)), its value at (-1) satisfies

[ \Delta_K(-1)\equiv \pm1\pmod 8 ]

when the Arf invariant is zero, and

[ \Delta_K(-1)\equiv \pm3\pmod 8 ]

when the Arf invariant is one.

Relation to the Kervaire invariant

A framed manifold of dimension (4k+2) determines a quadratic refinement of its middle-dimensional mod-two intersection pairing. The Arf invariant of this quadratic refinement is the Kervaire invariant of the framed manifold.

Michel Kervaire introduced this topological invariant in his analysis of manifolds that do not admit smooth structures. William Browder subsequently connected the invariant-one problem with stable homotopy theory by identifying the dimensions in which an associated framed manifold could exist. This interpretation transformed the algebraic invariant of a finite quadratic space into an obstruction within surgery theory and stable homotopy theory.

The topological construction retains the same characteristic-two mechanism as the algebraic definition. The intersection pairing supplies the alternating bilinear form, while self-intersection information supplies its quadratic refinement. The resulting binary value is independent of the geometric representatives used to express those data.

Degenerate forms and related invariants

When the polar form (b_q) is degenerate, its radical is

[ \operatorname{rad}(b_q)

{x\in V:b_q(x,y)=0\text{ for every }y\in V}. ]

An ordinary Arf invariant descends to the nonsingular quotient only when the quadratic form vanishes on this radical. Otherwise, the restriction of (q) to the radical contributes additional information that is not represented by a single Arf value.

Quadratic enhancements taking values in (\mathbb Z/4\mathbb Z) lead to the Brown invariant, whose associated Gauss sum takes eighth roots of unity as phases. The Brown invariant extends the same character-sum principle to settings in which a binary quadratic refinement is insufficient.