Rational Pontryagin classes of Euclidean fiber bundles (Q2078914)

Let \(\text{TOP}(n)\) denote the (topological or simplicial) group of homeomorphisms of \(\mathbb R^n\). Classical yet deep results by Sullivan and Kirby-Siebenmann imply that the canonical map \(BO \to B\text{TOP} = \text{colim}_{n \to \infty} B \text{TOP}(n)\) is a rational equivalence, hence the Pontryagin classes \(p_i \in \text{H}^{\ast}(B\text{TOP}; \mathbb Q)\) are well-defined. The main result of this paper states that for sufficiently big \(n\) and all \(k\) in some non-trivial range that grows linearly with \(n\), \(p_{n+k} \in \text{H}^{4(n+k)}(B\text{TOP}; \mathbb Q)\) remains non-zero when pulled back to \(\text{H}^{4(n+k)}(B\text{TOP}(2n); \mathbb Q)\). This is in stark contrast to the case of classifying spaces of linear \(\mathbb R^n\)-bundles where it is known that \(0 = p_{n+k} \in \text{H}^{4(n+k)}(BO(2n); \mathbb Q)\) for all \(k > 0\). The high-level strategy to prove this result is to combine work by \textit{S. Galatius} and \textit{O. Randal-Williams} on the cohomology of diffeomorphism groups (usually referenced by the keyword \emph{moduli spaces of manifolds}, cf. [Acta Math. 212, No. 2, 257--377 (2014; Zbl 1377.55012)]), with manifold calculus invented by \textit{T. G. Goodwillie} and \textit{M. Weiss} (also known as embedding calculus, cf. [Geom. Topol. 3, 103--118 (1999; Zbl 0927.57028)]), to construct a certain fiber bundle. This bundle \(E \to M\) is such that \(M\) is a \((2n+4k)\)-dimensional smooth closed stably parallelizable manifold, the fibers are homeomorphic to \(W_g = \#^g S^n \times S^n\), and the total space \(E\) has vanishing decomposable Pontryagin classes, but non-zero signature. Deducing the main result from the existence of such a bundle is uncomplicated. In the final chapter, it is proven that the `surreal' Pontryagin numbers appearing in the main theorem evaluate non-trivially against the image of the Hurewicz. On the more technical side, the paper starts with a computational, homotopically flavored analysis of certain \(\kappa\)-classes whose non-vanishing for \(W_g\) follows from the work by Galatius and Randal-Williams alluded to above. Then manifold/embedding calculus is used to study configuration categories (cf. [\textit{P. B. De Brito} and \textit{M. Weiss}, J. Topol. 11, No. 1, 65--143 (2018; Zbl 1390.57018)]) which in turn are helpful to understand spaces of self-embeddings. The essential geometric property that the manifolds \(W_{g,1} = W_g \backslash \mathring{D^{2n}}\) have and that is needed for this part of the paper is that their homotopical dimension is \(n\), whereas their geometric dimension is \(2n\). Thus for \(2n \geq 6\) the difference of the two is at least \(3\) and hence embedding calculus is applicable. By combining both approaches using rational homotopy theory, the existence of the aforementioned fiber bundle becomes evident. As a side-product, a fiber sequence \[B\text{Diff}_{\partial}(D^{2n}) \to B\text{Diff}_{\partial}(W_{g,1}) \to B\text{Diff}_{\partial}(W_{g,1+\varepsilon})\] is produced, where \(W_{g,1+\varepsilon}\) is obtained from \(W_{g,1}\) by removing a point in the boundary. The third term in the sequence is equivalent to the space of self-embeddings of \(W_{g,1}\) fixing half of the boundary; so this space is amenable to the methods of manifold calculus. The aforementioned fiber sequence was used with great success by Kupers (cf. [\textit{A. Kupers}, Geom. Topol. 23, No. 5, 2277--2333 (2019; Zbl 1437.57035)]) and others to study \(B\text{Diff}_{\partial}(D^{2n})\) and is now known by the name \emph{Weiss fiber sequence} in the literature. Hirzebruch's signature theorem is used multiple times; at one point, the non-vanishing of certain particular coefficients of Hirzebruch's \(\mathcal L\)-polynomials is needed. This was shown directly in a preliminary version of the paper, but in the meantime a much more general result was proven by \textit{A. Berglund} and \textit{J. Bergström} [Math. Ann. 372, No. 1--2, 125--137 (2018; Zbl 1406.55008)]. Without doubt, this paper is one of the biggest breakthroughs in homotopical geometric topology in recent years. The main result is rather surprising -- the author states he tried to prove the opposite statement for a long time -- and probably even more important are the new methods he developed to prove it. There are several preliminary versions available on the arXiv [\url{https://arxiv.org/abs/1507.00153}] also worth looking at. In the peculiar section \emph{Acknowledgements and Apologies} at the end of the introduction, it is explained how the paper grew out of some notes that were written to clarify a story told by the author at a satellite conference of the ICM 2014 in Dalian (China).