Surgery on \(\mathrm{Aut}(F_2)\) (Q6137346)

The main results of the paper concern the so-called Brady complex \(X_0\), which was introduced in the 90's by \textit{T. Brady} [Arch. Math. 63, No. 2, 97--102 (1994; Zbl 0811.20037)] in order to give a proof of the automaticity of \(\Aut(F_2)\) that does not rely on the automatic structure of braid groups. The authors provide a new perspective on \(X_0\) by constructing a family of locally \(\mathrm{CAT}(0)\)-spaces \(B_n\) whose fundamental groups \(G_n\) are finite-index subgroup of \(\Aut(F_2)\) and whose universal covers \(X_n\) are isometrically isomorphic to \(X_0\). In order to obtain these isomorphisms, the authors prove a rigidity result in Theorem 4.3, according to which every \(\mathrm{CAT}(0)\)-space that locally resembles the geometry of \(X_0\) must be isometric to \(X_0\). Here the condition of locally resembling the geometry is slightly more restrictive than requiring the existence of an abstract isometry between the links in that it moreover requires to preserve the local type, which is introduced in Section 4. This further restriction is indeed necessary as the authors show in Section 7 that there is a \(\mathrm{CAT}(0)\) \(2\)-complex that is locally isometric to the Brady complex, but not isometrically isomorphic to it. So far, rigidity results of this spirit are quite rare, which suggests that the Brady complex is a natural and interesting object to study. Indeed, even in the case of such nice spaces as Bruhat-Tits buildings for \(\mathrm{SL}_d(\mathbb{F}_p((t)))\), \(d \geq 4\), it is known that such rigidity results do not always hold [\textit{M. de la Salle} and \textit{R. Tessera}, Ill. J. Math. 60, No. 3--4, 641--654 (2016; Zbl 1376.20033)]. In order to obtain the spaces \(B_n\), the authors start with a torus of size \(6 \times n\), tessellated by lozenges, for some fixed integer \(n \geq 1\). The complex \(B_n\) is then obtained by identifying pairs of vertices in the torus and filling triangles in the identification space. In Section 6, the authors show that this procedure does not come out of the blue but rather results from the study of cobordism categories that can be used to construct groups acting on complexes of a given local type.