Serre polynomials of \(\mathrm{SL}_n\)- and \(\mathrm{PGL}_n\)-character varieties of free groups (Q2223745)

The \(E\)-polynomial \(E(X)=E(X;u,v)\) of the quasi-projective variety \(X\) is the specialization of the mixed Hodge polynomial of \(X\) namely \(\mu(X)=\mu(X;t,u,v)=\sum_{k,p,q}h^{k,p,q}t^ku^pv^q \in \mathbb{N}_0[t,u,v]\) by substituting \(t=-1\). They are used to compute the characteristic classes of the mixed Hodge modules (complexes) of \(X\) taking values in the Grothendieck group of certain categories. One can also get the Poincare polynomial of \(X\) by setting \(u=v=1\). The mixed Hodge and the \(E\)-polynomial are multiplicative. Also the \(E\)-polynomial is additive w.r.t stratifications by locally closed strata. In this context the pre-image of these polynomials should be the Grothendieck ring of all varieties over \(\mathbb{C}\). One has \(\chi(X)=E(X;1,1)= \mu(X;-1,1,1)\). The \(E\)-polynomial has many applications in mirror symmetry, Langlands duality and Physics [\textit{C. Florentino} et al., ``On Hodge polynomials of singular character varieties'', in: Proceedings of the 12th ISAAC conference Portugal: University of Aveiro (2019); ``Generating series for the Hodge-Euler polynomials of \(\mathrm{GL}(n, \mathbb{C})\)-character varieties'', Preprint, \url{arXiv:1902.06837}; \textit{O. García-Prada} et al., J. Eur. Math. Soc. (JEMS) 16, No. 12, 2617--2668 (2014; Zbl 1316.14060); \textit{ González-Prieto} et al., Bull. Sci. Math. 161, Article ID 102871, 33 p. (2020; Zbl 1441.57031); \textit{T. Hausel} and \textit{F. Rodriguez-Villegas}, Invent. Math. 174, No. 3, 555--624 (2008; Zbl 1213.14020); \textit{T. Hausel} and \textit{M. Thaddeus}, Invent. Math. 153, No. 1, 197--229 (2003; Zbl 1043.14011); \textit{C. A. M. Peters} and \textit{J. H. M. Steenbrink}, Mixed Hodge structures. Berlin: Springer (2008; Zbl 1138.14002); \textit{M. Thaddeus}, Int. Math. Res. Not. 2001, No. 22, 1169--1193 (2001; Zbl 1074.14511)]. An interesting property of the \(E\)-polynomial appear in a fibration \(F \to X \to B\) of quasi-projective varieties. One has \(E(X)=E(F)E(B)\) [Theorem 2.3]. The behaviour of the polynomials \(E\) and \(\mu\) is very much like that of characteristic classes of objects in abelian categories. In this context a combinatorial framework of defining characteristic classes and Adam's operations can be used to explain them. The paper is specially interested to the fibrations by special groups \(H\), i.e. principal \(H\)-bundles \(\pi: X \to B\) where \(H\) is special [see Section 2.2]. In the very situation of the paper a finite group \(W\) also acts on \(X\) fiberwise. In this case one can define the \(W\)-equivariant \(E\)-polynomial and above multiplicativity becomes \(E^W(X)=E^W(H)E(B)\) where \(E^W(.) \in R(W)[u,v]\) and \(R(W)\) is the Grothendieck ring of \(W\) (Corollary 2.5). Let \(\Gamma\) be a finitely presented group and \(G\) a complex connected reductive algebraic group. The \(G\)-character variety of \(\Gamma\) is defined as the GIT quotient \(\chi_{\Gamma}G=\text{Hom}(\Gamma,G)//G\) where \(\text{Hom}(\Gamma,G)\) stands for the algebraic variety of representations of \(\Gamma\) in \(G\) and \(G\) acts by conjugation on a representation. The GIT quotient identifies the orbits whose closures have a non-empty intersection. The quotient is described by the unique closed point in each equivalence class called polystable representations, [\textit{D. Baraglia} and \textit{P. Hekmati}, Proc. Lond. Math. Soc. (3) 114, No. 2, 293--332 (2017; Zbl 1391.14018); \textit{V. V. Batyrev} and \textit{D. I. Dais}, Topology 35, No. 4, 901--929 (1996; Zbl 0864.14022); \textit{A. Casimiro} et al., Forum Math. 28, No. 2, 275--294 (2016; Zbl 1349.14187); \textit{S. Cavazos} and \textit{S. Lawton}, Int. J. Math. 25, No. 6, Article ID 1450058, 27 p. (2014; Zbl 1325.14065); \textit{C. Florentino} and \textit{S. Lawton}, Math. Ann. 345, No. 2, 453--489 (2009; Zbl 1200.14093)]. One may use the stratification by polystable type to compute the \(E\)-polynomial. In the paper the case of interest is for \(G=\mathrm{GL}_n, \mathrm{SL}_n, \mathrm{PGL}_n\), where one has a basic decomposition \[ \chi_{\Gamma}\mathrm{GL}_n= \bigsqcup_{[k] \in \mathcal{P}_n}\chi_{\Gamma}^{[k]}\mathrm{GL}_n \] where \(\chi_{\Gamma}^{[k]}\mathrm{GL}_n\) is the \([k]\)-polystable representations [Proposition 4.3]. The partition \(n=\sum_j jk_j\) is denoted by \([k]=[k_1,...,k_n]\). In this special case one calls a representation \(\rho\) a \([k]\)-polystable representation if it is conjugated to \(\bigoplus \rho_j\) where each \(\rho_j\) is a direct sum of \(k_j\) irreducible representations in \(Hom(\Gamma , \mathrm{GL}_j)\) [Definition 4.1]. The representation \(\rho\) is irreducible if it is polystable and the orbit map denoted by \(\psi_{\rho}\) of the action of \(G\) is proper [Definition 3.1]. Similar stratifications exist for \(\mathrm{SL}_n, \mathrm{PGL}_n\) [Proposition 4.5], [\textit{D. Baraglia} and \textit{P. Hekmati}, Proc. Lond. Math. Soc. (3) 114, No. 2, 293--332 (2017; Zbl 1391.14018); \textit{C. Florentino} and \textit{S. Lawton}, Math. Ann. 345, No. 2, 453--489 (2009; Zbl 1200.14093); \textit{C. Florentino} et al., ``On Hodge polynomials of singular character varieties'', in: Proceedings of the 12th ISAAC conference Portugal: University of Aveiro (2019); ``Generating series for the Hodge-Euler polynomials of \(\mathrm{GL}(n, \mathbb{C})\)-character varieties'', Preprint, \url{arXiv:1902.06837}; \textit{O. García-Prada} et al., J. Eur. Math. Soc. (JEMS) 16, No. 12, 2617--2668 (2014; Zbl 1316.14060); \textit{S. Mozgovoy} and \textit{M. Reineke}, Int. J. Math. 26, No. 12, Article ID 1550100, 19 p. (2015; Zbl 1346.14115); \textit{A. S. Sikora}, Trans. Am. Math. Soc. 364, No. 10, 5173--5208 (2012; Zbl 1291.14022)]. The rest of the paper concerns the case of \(\Gamma=F_r\) the free group on \(r\) generators, and the comparison of the \(E\)-polynomials of the \([k]\)-strata of the \(G\)-character variety for these 3 groups [notation; \(\chi_{F_r}^{[k]}\mathrm{GL}_n=: \chi_{r}^{[k]}\mathrm{GL}_n\)]. Also the authors replace \(e(X,x)=E(X,\sqrt{x},\sqrt{x})\) when the \(E\)-polynomial is a polynomial on the variable \(x=uv\) [Eq. 4.1]. A basic \(\mathbb{C}^*\)-fibration argument [Proposition 4.4] gives \[ e(\chi_{r}^{[k]}\mathrm{GL}_n) =(x-1)^r e(\chi_{r}^{[k]}\mathrm{PGL}_n) \] The comparison of the \(E\)-polynomials of the \([k]\)-strata of \(\chi_{r}^{[k]}\mathrm{SL}_n\) and \(\chi_{r}^{[k]}\mathrm{PGL}_n\) is more difficult in the proofs. The author's strategy which is the the major result of the paper, is first to prove the equality \(e(\chi_{r}^{[k]}\mathrm{GL}_n) =(x-1)^r e(\chi_{r}^{[k]}\mathrm{SL}_n)\) when \(\mathrm{length}([k])>1\) [Theorem 4.6]. In the remaining irreducible strata they propose to prove \[ H^*(\chi_{r}^{irr}\mathrm{SL}_n)= H^*(\chi_{r}^{irr}\mathrm{PGL}_n) \qquad \text{as MHS} \] which implies the equality of \(E\)-polynomials for this strata [Main Theorem 4.8]. The proof of Eq. (3) involves the whole of Section 5. The main difference in the comparison of the \(E\)-polynomials of \(\chi_r \mathrm{SL}_n\) and \(\chi_r \mathrm{PGL}_n\) is that in the fibration \[ (\mathbb{Z}/n)^r \to \chi_r \mathrm{SL}_n \to \chi_r\mathrm{PGL}_n \] the fibres are not connected. So we can not use the fibration results mentioned directly. In Section 5 the authors use a compact version of the definition of character varieties, namely compact representation spaces. That is they consider similar fibrations for the compact Lie groups \(U_n, \mathrm{SU}_n\) and \(PU_n\), where they are able to use similar stratifications to obtain the result for these groups [Proposition 5.2 and Proposition 5.10]. The major consequence in Section 5 is the Theorem 5.11. The proof is based on the fact that For every reductive group \(G\), there is a strong deformation retraction from \(\chi_rG\) to the orbit space Hom\((F_r, K)/K\), where \(K\) is a maximal compact subgroup of \(G\) [\textit{C. Florentino} and \textit{S. Lawton}, Math. Ann. 345, No. 2, 453--489 (2009; Zbl 1200.14093)]. This implies \[ H^*(\chi_r \mathrm{SL}_n) =H^*(\chi_r \mathrm{SU}_n) =H^*(\chi_rPU_n) =H^*(\chi_r\mathrm{PGL}_n) \] With the aid of a cohomological argument for stratifications it completes the proof. Explicit computations are given in Section 6.