A Serre weight conjecture for geometric Hilbert modular forms in characteristic \(p\) (Q6057870)

Serre's conjecture (1975) asserts that every odd, continuous irreducible representation \(\varrho\) of \(\mathrm{Gal}(\overline{\mathbb Q}/\mathbb Q)\) over a finite field is \textit{modular}, i.e. is isomorphic to the mod \(p\) Galois representation \(\varrho_f\) associated to a modular eigenform \(f\). The strong version of the conjecture describes the smallest weight \(k \geq 2\) such that \(\varrho \cong \varrho_f\) for a modular eigenform \(f\) of weight \(k\). Prior to the proof of the full conjecture due to Khare-Wintenberger and Kisin, Edixhoven proved the \textit{weight part of the Serre's conjecture}: assuming that \(\varrho\) is modular, he proved that Serre's recipe for the smallest weight \(k \geq 2\) is correct. \textit{B. Edixhoven} [Invent. Math. 109, No. 3, 563--594 (1992; Zbl 0777.11013)] furthermore refined the conjecture to predict the minimal weight \(k \geq 1\) such that \(\varrho \cong \varrho_f\) for a modular form \(f\) of weight \(k\). The main subtelty is that not every weight \(1\) mod \(p\) modular form is the reduction of a characteristic 0 modular form. Relatedly, for \(k \geq 2\), mod \(p\) modular forms give rise to classes in the cohomology \(H^1(X, V)\) of a mod \(p\) local system \(V\) on an appropriate modular curve \(X\). On the other hand, weight one modular forms do not contribute to the cohomology groups \(H^1(X, V)\), but they do define section of certain line bundles \(\mathcal E\) on \(X\); symbolically, \(f \in H^0(X, \mathcal E)\). Accordingly, the weights \(k \geq 2\) are called \textit{algebraic} all the weights \(k \geq 1\) are called \textit{geometric}. The goal of the paper under review is to state a Serre weight conjecture for \textit{geometric} Hilbert modular forms. Compared to the story above, the field \(\mathbb Q\) is replaced by a totally real field \(F\) of degree \(d\) (in which \(p\) is unramified), modular forms of weight \(k\) are replaced by Hilbert modular forms of weight \((k_1, \ldots, k_d)\), and the modular curves are replaced by Hilbert modular surfaces (non-PEL type Shimura varieties for the group \(G =\mathrm{Res}_{F/\mathbb Q} \mathrm{GL}_2\)). A weight \((k_1, \ldots, k_d)\) is \textit{algebraic} if \(k_i \geq 2\) and \(k_1 \equiv k_2 \equiv \cdots \equiv k_d \bmod 2\) (the second condition is called \textit{paritious}); otherwise it is \textit{geometric}. For algebraic weights, the Serre weight conjecture was generalized by \textit{K. Buzzard} et al. [Duke Math. J. 155, No. 1, 105--161 (2010; Zbl 1227.11070)], and the goal of the present paper is state a geometric version of it. The first result of the paper is to associate Galois representations \(\varrho_f\) to mod \(p\) Hilbert modular eigenforms \(f\) of any geometric weights, without the parity assumption present in previous work of Emerton-Reduzzi-Xiao and \textit{W. Goldring} and \textit{J.-S. Koskivirta} [Invent. Math. 217, No. 3, 887--984 (2019; Zbl 1431.11071)]. Then, assuming that a Galois representation \(\varrho\) of \(\mathrm{Gal}(\overline F/F)\) over a finite field is geometrically modular of some weight, the main conjecture of the paper gives a recipe for the smallest geometric weight from which \(\varrho\) may arise, à la Edixhoven. We refer to Section 1.3 of the introduction for more details. On top of proving the aforementioned theorem and stating the conjecture, the paper establishes many results which are crucial to understanding mod \(p\) Hilbert modular forms. First of all, it redevelops the foundational work of \textit{F. Andreatta} and \textit{E. Z. Goren} [Hilbert modular forms: mod \(p\) and \(p\)-adic aspects. Providence, RI: American Mathematical Society (AMS) (2005; Zbl 1071.11023)], including the construction of Hecke operators and partial Hasse invariants, in the non-PEL Hilbert modular variety setting. It also studies partial \(\Theta\)-operators and \(q\)-expansions. This analysis culminates in some beautiful results which deal with new phenomena for \(d = 2\) and \(p\) inert in \(F\). When \(F = \mathbb Q\), it is known that Galois representations which are unramified at \(p\) arise from mod \(p\) eigenforms of weight \(1\). For \([F : \mathbb Q] = 2\), the analogous statement is more subtle because there is an infinite set of \\textit{partial weight one} forms, namely the weights \((1, k)\) and \((k, 1)\) for any \(k \geq 1\) (see also [Andreatta and Goren, loc. cit.] and [\textit{M. Dimitrov} and \textit{G. Wiese}, J. Inst. Math. Jussieu 19, No. 2, 281--306 (2020; Zbl 1437.11081)]. Using techniques from integral \(p\)-adic Hodge theory the authors describe \(\varrho\) for which the conjectural minimal weight is one of the above. They then use the theory developed in the paper to transfer modularity results between algebraic weights and non-algebraic weights. This general strategy is also carried out in \textit{H. Wiersema}'s thesis [J. Théor. Nombres Bordx. 33, No. 1, 107--124 (2021; Zbl 1477.11113)].