New invariants of \(G_{2}\)-structures (Q888893)

Metrics with holonomy \(G_2\) on a smooth, closed \(7\)-manifold \(M\) correspond to torsion-free \(G_2\) structures \(\varphi\in\Omega^3(M)\). Forgetting the holonomy reduction's integrability condition \(d\varphi=0=d^*\varphi\), one can ask about deformations of such \(\varphi\). If two \(G_2\) structures cannot be deformed into one another, certainly also the corresponding metrics will be inequivalent. This remarkable paper develops two tools \(\nu,\xi\) and techniques for distinguishing connected components of the \(G_2\) moduli space \[ \mathcal{M}=\{\mathrm{Riemannian metrics with holonomy}= G_2\}/\mathrm{Diff}(M). \] Whilst the local structure of \(\mathcal{M}\) is that of an orbifold of dimension \(b_3(M)\), the global properties are still rather mysterious. The authors produce the first examples where \(\mathcal{M}\) is disconnected, in the spirit of Donaldson and Thomas's higher-dimensional gauge theory. Assume \(M\) is \(2\)-connected and \(H^4(M)\) has no torsion, and let \(d(M)\) be the largest integer dividing the Pontrjagin class \(p_1(M)\). A theorem of \textit{D. L. Wilkens} [Bull. Lond. Math. Soc. 4, 27--31 (1972; Zbl 0241.57018)] guarantees that the topological type is determined by \(b_3\) and \(d(M)\), and there exist \(\mathrm{GCD}(28, \mathrm{numerator}(\frac{d(M)}{8}))\) inequivalent smooth structures on \(M\) (so the diffeo-type is unique if, e.g., \(8\) does not divide \(p_1\)). By identifying homotopy classes of \(G_2\) structures with homotopy classes of unit spinors, the space of \(G_2\) structures modulo homotopy is shown to be infinite cyclic (affinely). To study \(\pi_0(\mathcal{M})\) one still needs to understand how diffeomorphisms act on homotopy classes. In order to distinguish \(G_2\) structures modulo homotopy and diffeomorphism, the authors define the quantity \[ \nu(\varphi) : = \chi(W) - 3\sigma(W)\quad \mathrm{mod}\;48 \] where \(\sigma(W)\) and \(\chi(W)\) are the signature and Euler characteristic of a compact \(\mathrm{Spin}(7)\) manifold \(W^8\) co-bounding \(M\) (the spin bordism group \(\Omega_7^{\mathrm{Spin}}\) is trivial). This number does not depend on \(W\) nor on its spin structure, but only on the spin geometry induced on the boundary \(M\). It is \(\mathbb{Z}_{48}\)-valued, homotopy invariant, and closely resembles Milnor's invariant for exotic spheres. On a fixed \(M\) it takes exactly the \(24\) values permitted by the parity constraint \(\nu\equiv \chi_\mathbb{Q}(M):=\sum_{i=0}^7 b_i(M)\) mod \(2\). Although \(\nu\) alone is not, in general, capable of telling apart classes of \(G_2\) structures up to homotopy and diffeomorphism, it does so if \(d(M)\) divides \(7\cdot 32\), in which case there are exactly \(24\) different types. The invariant is computed for twisted connected sum of Fano \(3\)-folds [\textit{A. Kovalev}, J. Reine Angew. Math. 565, 125--160 (2003; Zbl 1043.53041)] and `weak' Fanos [\textit{A. Corti} et al., Duke Math. J. 164, No. 10, 1971--2092 (2015; Zbl 1343.53044)], where the associated torsion-free \(\varphi\) always takes the value \(\nu=24\). Some of Joyce's compact examples, viewed as `extra-twisted' connected sums, have odd \(\nu\). One issue is that \(\nu\) requires the explicit knowledge of a spin bordism \(W\). Overcoming this, [the authors and \textit{S. Goette}, ``An analytic invariant of \(G_2\) manifolds'', Preprint, \url{arXiv:1505.02734}] introduces an analytic refinement of \(\nu\) in terms of \(\eta\) invariants and Mathai-Quillen currents: this is invariant under diffeomorphisms and deformations through torsion-free \(G_2\) structures (yet not arbitrary homotopies), it distinguishes components of \(\mathcal{M}\) even when the \(G_2\) structures are homotopic, and is often more viable than \(\nu\). Another homotopy invariant is introduced in the article: \[ \xi(\varphi):=7\chi(W)+\frac{3p_W^2-45\sigma(W)}{2}\in\mathbb{Q}. \] When the spin characteristic class \(p_M=p_1(M)/2\) is a torsion class, \(\xi\) and \(\mu(M):=(p_W^2 -\sigma(W))/8\) determine \(\nu\). This second number \(\xi\) takes all possible values with the proviso that \(\xi(\varphi) = 7 \chi_\mathbb{Q}(M) + 12\mu(M)\) mod \(14\mathbb{Z}\). The point is that if \(M\) is \(2\)-connected then \((\nu,\xi)\) is a complete set of invariants of \(\pi_0(\mathcal{M})\) (cf. the diffeomorphism classification of [the authors, ``The classification of \(2\)-connected \(7\)-manifolds'', Preprint, \url{arXiv:1406.2226}]). As an interesting aside, one might relax torsion-freeness and consider co-closed \(G_2\) structures: for these the authors prove a parametric h-principle, namely that the space of coclosed \(G_2\) structures is homotopy-equivalent to the space of \(G_2\) structures. Consequently any smooth, spin \(7\)-manifold admits smooth co-closed \(G_2\) structures.