A link invariant with values in the Witt ring (Q2259625)

The \textit{signature} \(\sigma(L)\in \mathbb{Z}\) of an oriented link \(L \subset S^3\) is an isotopy invariant, which can be defined as the signature of the symmetrization \(V+ V^{t}\) of a Seifert matrix \(V\) of \(L\). For any integer \(n \geq 2\), one can consider the map \[ f_n: B_n \longrightarrow \mathbb{Z}, \;\beta \longmapsto \sigma(\widehat{\beta}), \] where \(\widehat{\beta}\) denotes the closure of an \(n\)-strand braid \(\beta\), and one can ask how far \(f_n\) is from being a group homomorphism. This question has been fully answered by \textit{J.-M. Gambaudo} and \textit{É. Ghys} [Bull. Soc. Math. Fr. 133, No. 4, 541--579 (2006; Zbl 1103.57001)] who proved that, for any \(\beta, \gamma \in B_n\), \[ (*)\quad f_n(\beta\gamma)-f_n(\beta)-f_n(\gamma) = \tau(\ell_{1}, \ell_{\beta},\ell_{\beta\gamma}). \] Here \(\ell_\kappa\) denotes a Lagrangian subspace of a certain symplectic vector space which can be associated to any \(\kappa \in B_n\) using the Burau representation, and \(\tau\) denotes the Maslov index. Recall now the following two facts: (i) the Maslov index \(\tau(\ell_1,\ell_2,\ell_3)\in \mathbb{Z}\) of three Lagrangian subspaces \(\ell_1, \ell_2,\ell_3\) in a symplectic vector space is defined as the signature of a certain non-degenerate symmetric bilinear form over \(\mathbb{R}\); (ii) the signature of non-degenerate symmetric bilinear forms over \(\mathbb{R}\) realizes an isomorphism between the Witt ring \(\text{W}(\mathbb{R})\) and the ring of integers \(\mathbb{Z}\). Thus the identity \((*)\) can be regarded as taking place in \(\text{W}(\mathbb{R})\). In fact, Gambaudo and Ghys proved their identity \((*)\) in the more general situation where the signature of links \(\sigma\) is replaced by the Levine--Tristram signature \(\sigma_\omega\) at a non-trivial root of unity \(\omega\): then the same considerations apply using the Hermitian Witt ring of the field \(\mathbb{C}\) equipped with the complex conjugation. Very much inspired by this work, the authors construct link invariants with values in Hermitian Witt rings. Specifically, let \(K\) be a field of characteristic different from \(2\), let \(\sigma: K \to K\) be an involutive ring homomorphism and let \(\text{WH}(K,\sigma)\) be the corresponding Hermitian Witt ring. Having choosen a ring homomorphism \(\alpha: \mathbb{Z}[t^{\pm1}] \to K\) such that \(\alpha(t^{-1}) =\sigma (\alpha(t))\), the authors consider for any integer \(n\geq 2\) a certain skew-Hermitian form \(h_n:V_n \times V_n \to K\) on an \(n\)-dimensional \(K\)-vector space \(V_n\), which is essentially the twisted intersection form of the \(n\)-punctured disk. If \(B_n\) is interpreted as the mapping class group of the \(n\)-punctured disk, one gets a unitary representation of the braid group \(r_n:B_n \to \mathbf{U}(V_n,h_n)\), which can be recovered from the Burau representation of \(B_n\) using \(\alpha\). Then the authors prove that there is a unique map \(f_n: B_n \to \text{WH}(K,\sigma)\) which vanishes on any elementary braid and satisfies \[ f_n(\beta\gamma)-f_n(\beta)-f_n(\gamma) = \tau\big(\Gamma_{1}, \Gamma_{r_n(\beta)},\Gamma_{r_n(\beta\gamma)}\big) \] where, for any \(\varphi \in \mathbf{U}(V_n,h_n)\), \(\Gamma_\varphi \subset V_n \oplus V_n\) denotes the graph of \(\varphi\) and \(\tau\) denotes the Maslov index in the skew-Hermitian space \((V_n,h_n) \oplus (V_n, -h_n)\). Furthermore, it is shown that the collection of functions \(\{f_n\}_{n \geq 2}\) satisfies the Markov moves: this results in a \(\text{WH}(K,\sigma)\)-valued link invariant \(\Theta_K\), which is defined by \(\Theta_K\big(\widehat{\beta}\big) := f_n(\beta)\) for any \(n\geq 2\) and any \(\beta \in B_n\). This link invariant recovers several classical invariants. When \(K=\mathbb{C}\), \(\sigma\) is the complex conjugation and \(\omega:=\alpha(t)\) is a non-trivial root of unity, it follows directly from the results of Gambaudo and Ghys that \[ \text{sgn} \big(\Theta_K(L)\big) = \sigma_\omega(L). \] When \(K=\mathbb{Q}(t)\) is the field of rational fractions, \(\sigma(t)=t^{-1}\) and \(\alpha(t)=t\), some computations strongly suggest that \[ \det \Theta_K(L) \equiv \Delta(L) \] where \(\Delta(L)\) is an appropriate sign-normalization of the Alexander polynomial of a knot \(L\), and the symbol \(\equiv\) means the identity up to multiplication by Laurent polynomials of the form \(r(t)\, r(t^{-1})\) for some \(r(t) \in \mathbb{Z}[t^{\pm 1}]\). The authors also discuss some other (potentially ``new'') link invariants arising from their construction by considering Witt rings of other fields, such as finite fields. Finally, they explain how to slightly modify their construction in order to derive a topological invariant from the character of the Weil representation: the resulting invariant is actually determined by the Alexander polynomial and the signature of links. The paper is very well written, and its results are illustrated by explicit computations of \(\Theta_K(L)\) for two examples of knots \(L\). Several questions about this knot invariant would deserve further studies. For instance, which ``parts'' of \(\Theta_K\) are invariant under concordance of knots? Or, how is \(\Theta_K\) related to the Blanchfield pairing of a knot? See also the Ph.D. dissertation of \textit{M. Maxime Bourrigan} in this connection [ENS Lyon, 2013; available from {\url www.math.ens.fr/\(\sim\)bourrigan/}].