Parabolic refined invariants and Macdonald polynomials (Q2339205)

The authors of the paper under review deduce string-theoretically the conjecture of \textit{T. Hausel} et al. [Duke Math. J. 160, No. 2, 323--400 (2011; Zbl 1246.14063)] on the cohomology of character varieties with marked points (\textit{HLRV conjecture} for short). In doing so, they provide strong evidence for this conjecture. We start by briefly reviewing the HLRV conjecture. In broadest terms, the HLRV conjecture expresses a weighted sum of mixed Poincaré polynomials of character varieties as the \(\log\) of a generating function made up of modified MacDonald polynomials. As this may appear unmotivated, the present paper provides strong physical reasons why it should be so. We review the definition of character varieties. Denote by \(C\) a smooth genus \(g\geq 0\) projective curve over \(\mathbb{C}\) and let \(D = p_1 + \cdots + p_k\) be a divisor of marked points, which are assumed to be distinct and reduced. Consider the fundamental group \(\pi_1(C\backslash D)\). Each marked point \(p_i\) determines a generator \(\gamma_i\) of \(\pi_1(C\backslash D)\). Let \(r\geq 1\). For any non-empty partition \(\mu=(\mu^1,\dots,\mu^l)\) of \(r\), we choose a semisimple conjugacy class \(C_\mu\) in \(GL(r,\mathbb{C})\) with the property that the eigenvalues of any matrix in \(C_\mu\) have multiplicities given by \(\left\{\mu^1,\dots,\mu^l\right\}\). Consider a collection \(\boldsymbol{\mu}=(\mu_1,\dots,\mu_k)\) of partitions of \(r\). The moduli problem consists in parametrizing conjugacy classes of representations \[ f : \pi_1(C\backslash D) \to GL(r,\mathbb{C}), \] with the property that \(f(\gamma_i)\in C_{\mu_i}\). The moduli space of such representations yields the \textit{character variety} \(\mathcal{C}(C,D;\boldsymbol{\mu})\). The dependence on the choice of eigenvalues is suppressed in the notation because the topological invariants below do not depend on these choices. Via a weight filtration, the compactly supported cohomology of \(\mathcal{C}(C,D;\boldsymbol{\mu})\) is packaged into a mixed Poincaré polynomial \(P_c(\mathcal{C}(C,D;\boldsymbol{\mu});u,t)\). To each marked point \(p_i\) is associated an infinite collection of formal variables \(\boldsymbol{x}_i=(\boldsymbol{x}_{i,1},\boldsymbol{x}_{i,2},\dots)\). Summing over \(n\geq 1\) and over all all partitions of \(n\) of an appropriate weighting of \(P_c(\mathcal{C}(C,D;\boldsymbol{\mu});u,t)\) in the variables \(\boldsymbol{x}_i\) yields a generating function \(F_{HLRV}(z,w,\boldsymbol{x})\). The authors consider another generating function \(Z_{HLRV}(z,w,\boldsymbol{x})\) made up of modified MacDonald polynomials. Then the HLRV conjecture claims that \[ (1) \; \; Z_{HLRV}(z,w,\boldsymbol{x}) = \exp(F_{HLRV}(z,w,\boldsymbol{x})). \] The strategy of the authors to derive (1) is as follows. For each curve \(C\), they construct a local Calabi-Yau orbifold curve \(\widetilde{Y}\) and consider its generating function \(Z^{ref}_{\widetilde{Y}}(q,\boldsymbol{x},y)\) of refined stable pair invariants. They conjecture that it agrees with \(Z_{HLRV}\) after a change of variables, namely that: \[ (2) \; \; Z^{ref}_{\widetilde{Y}}(z^{-1}w,\boldsymbol{x},z^{-1}w^{-1})=Z_{HLRV}(z,w,\boldsymbol{x}). \] The authors make use of the \textit{parabolic \(P=W\) conjecture}, which is as follows. Motivated by the fact that the character variety \(\mathcal{C}(C,D;\boldsymbol{\mu})\) is diffeomorphic to a moduli space of strongly parabolic Higgs bundles on \(C\), it conjectures that the mixed Poincaré polynomial \(P_c(\mathcal{C}(C,D;\boldsymbol{\mu});u,t)\) is identified with a certain mixed Poincaré polynomial for Higgs bundles. Assuming the parabolic \(P=W\) conjecture and the conjecture of (2), the authors prove using geometric engineering that the same change of variables as in (2) turns (1) into the \textit{refined} stable pair/\textit{refined} Gopakumar-Vafa expansion. The latter is an enumerative correspondence widely expected to hold true. Its unrefined version (taking the Euler characteristic) was proven for toric Calabi-Yau 3-folds in genus 0. Throughout the paper, the authors give ample evidence for the various involved conjectures to hold true. They review the various objects that are involved in great detail, contributing to the ease of reading.