Colored \(\mathrm{sl}(\mathrm{N})\) homology and \(\mathrm{SU}(\mathrm{N})\) representations (Q6979947)

Khovanov-Rozansky homology is a package of link invariants which assigns to a link \(\mathcal{L}\) a bigraded vector space. There are various parameters which can be chosen in this package, including a (positive integer) rank \(N\) and a choice of color \(0 < a_c < N\) for each link component \(\mathcal{L}_c\). For fixed rank \(N\) and choice of colors, the resulting bigraded vector space \(KR_N(\mathcal{L})\) is called the colored \(\mathfrak{sl}_N\) homology of \(\mathcal{L}\). The (uncolored) Khovanov homology of \(\mathcal{L}\) is recovered when \(N = 2\) and all colors are \(1\).\N\NComputing \(KR_N(\mathcal{L})\) is challenging. Even for uncolored Khovanov homology, while computer programs are able to handle knots with about a dozen crossings, it remains a major open problem to compute this invariant for positive torus knots. These computations only grow in difficulty as the rank and color increase. One particularly thorny issue is the presence of torsion when working over the integers, demonstrating rich patterns and many open conjectures already at the level of uncolored Khovanov homology.\N\NIn the paper under review, Wang computes the colored \(\mathfrak{sl}_N\) homology of the Hopf link and the positive trefoil for arbitrary rank and color data. These are the first complete computations of colored Khovanov-Rozansky homology for nontrivial links. The resulting invariant is shown to be isomorphic to the cohomology of the \(SU(N)\) representation space \(\mathcal{R}_N(\mathcal{L})\), defined as a space of homomorphisms from \(\pi_1(\mathbb{R}^3 \backslash \mathcal{L})\) to \(SU(N)\) which satisfy certain conditions. This agreement is shown on several levels. First, \(H^*(\mathcal{R}_N(\mathcal{L}))\) carries a natural bigrading, and Wang shows that a certain regrading identifies \(H^*(\mathcal{R}_N(\mathcal{L}))\) and \(KR_N(\mathcal{L})\) as bigraded vector spaces. Each space also carries a natural module structure over \(H^*(Gr(N, a))\), and Wang shows an isomorphism between the two modules (and between certain reduced invariants defined using these actions). Finally, colored \(\mathfrak{sl}_N\) homology also admits an equivariant refinement, and Wang shows (again for the Hopf link and the positive trefoil) that this invariant agrees with the \(U(N)\)-equivariant cohomology of \(\mathcal{R}_N(\mathcal{L})\).\N\NTo prove these results, Wang leverages recent work of \textit{M. Hogancamp} et al. [``A skein relation for singular Soergel bimodules'', Preprint, \url{arXiv:2107.08117}] to provide explicit simplifications of the chain complexes assigned to the Hopf link and the trefoil. The latter authors work over a field. Adapting their results over the integers presents a major challenge and, in light of the presence of interesting torsion, a necessary one. Section 3 is dedicated to this argument and relies on a technical tour de force of delicate perturbation theory and explicit computations. (Section 2 contains a review of colored Khovanov-Rozansky homology and its equivariant version, mostly following the exposition of \textit{M. Ehrig} et al. [Proc. Lond. Math. Soc. (3) 117, No. 5, 996--1040 (2018; Zbl 1414.57010)].)\N\NSection 4 contains an analysis of the \(SU(N)\) representation spaces for the Hopf link and the trefoil using the theory of principle angles and a slight refinement (in Proposition 4.9) of a theorem of Gugenheim-May concerning the collapse of a certain spectral sequence [\textit{V. K. A. M. Gugenheim} and \textit{J. P. May}, On the theory and applications of differential torsion products. Providence, RI: American Mathematical Society (AMS) (1974; Zbl 0292.55019)]. More traditional algebro-topological material of this sort is rarely found in the link homology literature, and this section provides a welcome bridge between the two worlds.\N\NThe main results of the paper are proved in Section 5. The proof goes by constructing an explicit isomorphism between the state spaces assigned to certain \(\mathfrak{sl}_N\) webs, appearing in the simplifications of Section 3, with the equivariant homology of certain partial flag varieties, appearing as connected components of representation spaces in Section 4, followed by some analysis of the respective module actions on the two sides. The correspondence between direct summands of complexes on the link homology side and connected components on the topological side is particularly striking.