Stacks of Langlands Parameters
🔗Strongly Continuous Representations of a Locally Profinite Group
We fix
- $\Gamma$ a locally profinite group (Hausdorff topological group and identity element has a basis consisting of compact open subgroups. A locally profinite group is compact if and only if it is profinite.);
- $H$ a smooth affine group scheme over $\mathbb{Z}_ {\ell }$.
We have a functor $$ \begin{split} {}^{\text{cl}}R_ {\Gamma,H}: \mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}^\heartsuit &\rightarrow \mathrm{Sets},\ A &\mapsto \mathrm{Hom}_ {\text{cts}}(\Gamma,H(A)), \end{split} $$ where by ``cts’’, we mean the following definition.
Definition 1. A homomorphism $\rho:\Gamma\rightarrow H(A)$ is called ``strongly’’ continuous if * for one (and therefore every) faithful representation $H\rightarrow\mathrm{GL}(M)$, where $M$ is finite projective $\mathbb{Z}_ {\ell }$-module, every $m\in M\otimes A$, every $\Gamma_ 0\subseteq \Gamma$ an open compact subgroup, the $\mathbb{Z}_ {\ell }$-module $N\subseteq M\otimes A$ spanned by $\rho(\Gamma_ 0)m$ is a finite $\mathbb{Z}_ {\ell }$-module and the resulting representation of $\Gamma_ 0$ on $N$ is continuous.
Remark 2.
- If $A$ is over $\mathbb{Z}_ {\ell }/{\ell }^n $, this is equivalent to endow $H(A)$ with the dsicrete topology and to require that $\rho:\Gamma\rightarrow H(A)$ is continuous.
- If $A$ is finite over $\mathbb{Z}_ {\ell }$, the definition above is reduced to the usual $\ell$-adic topology (since in this case $M\otimes A$ is finite over $\mathbb{Z}_ {\ell}$). Also works for $A$ to be finite over $\mathbb{Q}_ {\ell }$.
- If $A=\overline{\mathbb{Q}_ {\ell }}$, and $\Gamma$ is locally finite (every finitely generated subgroup is finite), then by definition $\rho:\Gamma\rightarrow H(\overline{\mathbb{Q}_ {\ell }})$ factors through $\Gamma\rightarrow H(E)$ for some finite extension $E$ of $\mathbb{Q}_ {\ell }$.
The functor ${}^{\text{cl}}R_ {\Gamma,H}$ admits a derived enhancement $$ R_ {\Gamma,H}:\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}\rightarrow \mathrm{Ani}. $$ A group homomorphism $\rho:\Gamma\rightarrow H(A)$ is the same thing as a morphism of cosimplicial sets $$ \rho^*: \mathbb{Z}_ {\ell }[H^\bullet]\rightarrow C(\Gamma^\bullet,A). $$ Now we put continuity condition on it. For a locally profinite set $S$, we define $C_ {\text{cts}}(S,A)$ to be the set of continuous maps from $S$ to $A$, where $A$ is equipped with ind-$\ell$-adic topology, i.e. write $A=\varinjlim_ {i}M_ i$ where $M_ i$ are finite $\mathbb{Z}_ {\ell }$-modules.
Remark 3. This topology does not make $A$ a topological ring (the multiplication is not necessarily continuous).
Then (strongly) continuous homomorphism $\rho:\Gamma\rightarrow H(A)$ now corresponds to $\rho^*:\mathbb{Z}_ {\ell }[H^\bullet]\rightarrow C_ {\text{cts}}(\Gamma^\bullet,A)$.
Now for a profinite set $S$, we have a functor $$ C_ {\text{cts}}(S,-):\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}^\heartsuit\rightarrow\mathrm{CAlg}^\heartsuit, $$ which preserves sifted colimits (if $S$ is profinite). Then we can extend it to (non-abelian derived categories) $$ C_ {\text{cts}}(S,-):\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}\rightarrow \mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}. $$ Then $C_ {\text{cts}}(\Gamma^\bullet,-):\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}\rightarrow\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}^\Delta$, where the target is the category of cosimplicial animated rings over $\mathbb{Z}_ {\ell }$.
Definition 4. $$ \begin{split} R_ {\Gamma,H}:\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}&\rightarrow \mathrm{Grpd},\ A &\mapsto \mathrm{Maps}_ {\mathrm{CAlg}_ {\mathbb{Z}_ {\ell }}^\Delta}(\mathbb{Z}_ {\ell }[H^\bullet],C_ {\text{cts}}(\Gamma^\bullet,A)). \end{split} $$
Remark 5. Notation: by $\mathrm{Grpd}$, of course we mean infinity groupoids, or denoted by $\mathrm{Ani}$ or $\mathrm{Spc}$ in other literature.
Facts:
- If $V$ is a representation of $H$, $V\otimes \mathbb{Z}_ {\ell }[H^\bullet]$ is a module over $\mathbb{Z}_ {\ell }[H^\bullet]$, and base change along $\rho$, we get $C_ {\text{cts}}(\Gamma^\bullet,V\otimes A)$. Where $A$ is classical, this is the cochain complex computing continuous group homology of $(\Gamma, V\otimes A)$.
- $R_ {\Gamma,H}$ admits tangent complex and $\mathbb{T}_ {\rho}R_ {\Gamma,H}/H=C_ {\text{cts}}(\Gamma, \mathrm{Ad}_ \rho)[-1]$, where $\mathrm{Ad}_ \rho$ is $\Gamma\xrightarrow{\rho}H$ acts on $\mathrm{Lie}(H)$ via Ad action. This is a theorem requiring some effort to prove.
- $R_ {\Gamma,H}$ is represented by an ind-scheme over $\mathbb{Z}_ {\ell }$ in general.
Example 6. $\Gamma=\hat{\mathbb{Z}}$ and $H=\mathbb{G}_ m$ over $\Lambda=\mathbb{Z}_ {\ell }$, then $R_ {\Gamma,H}=\cup_ {Z\subseteq \mathbb{G}_ m,Z/\mathbb{Z}_ {\ell }\text{ finite}}Z^\wedge$. By ${}^\wedge$ we mean formal completion.
Example 7. $\Gamma=\hat{\mathbb{Z}}$, $H$ reductive, then $R_ {\Gamma,H}=H\times_ {H//H}(H//H)^{\wedge}$, where $$ (H//H)^{\wedge}=\cup_ {Z\subseteq H//H, Z/\mathbb{Z}_ {\ell }\text{ finite}}(H//H)_ Z^{\wedge}, $$ and $H//H$ is the GIT quotient.
Example 8. $\Gamma=I_ F^t=\prod_ {(l,p)=1}\mathbb{Z}_ {\ell }(1)$ tame inertia and $p$ is the characteristic of $F$. $H=\mathbb{G}_ m$. $$ R_ {I_ F^t,H}\subseteq R_ {\hat{\mathbb{Z}},H}, $$ and $$ R_ {I_ F^t,H}=\bigcup_ {Z\subseteq \mathbb{G}_ m,\text{ finite over $\mathbb{Z}_ {\ell }$, }Z\otimes \mathbb{F}_ l \text{prime to $p$}}Z^\wedge. $$
🔗The Stack of Langlands Parameters
Fix
- $F$ a non-archimedean local field,
- $G$ reductive group over $F$, with $(\hat{G}/\mathbb{Z}_ {\ell }, \hat{B},\hat{T},\hat{e})$, where
such that $\hat{e}|_ {\hat{U_ \alpha}}:\hat{U_ \alpha}\xrightarrow{\cong}{\mathbb{G}_ a}$.
Recall that $\mathrm{Aut}(\hat{G},\hat{B},\hat{T},\hat{e})\cong\mathrm{Aut}(\text{root datum of }G)$, which bears an action of $\Gamma_ F=\mathrm{Gal}(\overline{F}/F)\twoheadrightarrow \Gamma_ {\widetilde{F}/F}$, where $\widetilde{F}$ is a finite extension of $F$. We denote this Galois action by $\xi$.
$\rho_ {\text{ad}}:=$ half sum of all positive coroots of $\hat{G}$, i.e. $$ \rho_ {\text{ad}}:\mathbb{G}_ m\rightarrow (\hat{G})_ {\text{ad}}. $$ Note that $$ \begin{split} \mathrm{Ad}_ {\rho_ {\text{ad}}}:\hat{B}&\xrightarrow{\cong}{\hat{B}}\ \hat{T}&\xrightarrow{\cong}{\hat{T}} \end{split} $$ fixing Borel and torus, while change the pinning by $\mathrm{Ad}_ {\rho_ {\text{ad}}(\lambda)}\hat{e}=\lambda\hat{e}$.
$\mathrm{Ad}_ {\rho_ {\text{ad}}}$ and $\Gamma_ {\widetilde{F}/F}$ action commutes with each other and we define $$ {}^cG:=\hat{G}\rtimes_ {\mathrm{Ad}_ {\rho_ {\text{ad}}}\times\xi}(\mathbb{G}_ m\times \Gamma_ {\widetilde{F}/F}). $$ Recall that $$ {}^LG:=\hat{G}\rtimes_ {\xi} \Gamma_ {\widetilde{F}/F} $$
Definition 9. The stack of $L$-parameters $\mathcal{X}_ {G,F}:=\mathcal{X}_ {G,F}^\square/\hat{G}$, where $\mathcal{X}_ {G,F}^\square:=R_ {W_ F,{}^cG}\times_ {R_ {W_ F,\mathbb{G}_ m,\times \Gamma_ {\widetilde{F}/F}}}{\text{cycl}^{-1},\text{pr}}$, where $\text{cycl}:W_ F\rightarrow W_ F/I_ F\cong\mathbb{Z}\xrightarrow{\cong} q\mathbb{Z}$ is the cyclotomic character.
Denote $(\text{cycl}^{-1},\text{pr})$ by $\widetilde{\text{pr}}$.
🔗Main Theorem
Theorem 10. The stack $\mathcal{X}_ {{}^cG,F}:=R_ {W_ F,{}^cG}\times_ {R_ {W_ F,\mathbb{G}_ m\times \Gamma_ {\widetilde{F}/F}}}{\widetilde{\text{pr}}}/\hat{G}=\mathcal{X}_ {{}^cG,F}^\square/\hat{G}$.
$\mathcal{X}_ {{}^cG,F}^\square$ is represented by a disjoint union $$\mathcal{X}_ {{}^cG,F}^\square=\bigcup_ {L/F\text{Galois}, L\supseteq F^t\widetilde{F},[L:F^t]<\infty} \mathcal{X}_ {{}^cG,L/F}^\square.$$ Each $\mathcal{X}_ {{}^cG,L/F}^\square$ is an affine scheme of finite type over $\mathbb{Z}_ {\ell }$, flat of dimension $\dim\hat{G}$, l.c.i, and (base change to $\mathbb{Q}_ {\ell}$) reduced,
where
$F^t$ is the maximal tame extension of $F$, and
and $\text{cycl}:W_ F\rightarrow q^\mathbb{Z}\subseteq \mathbb{G}_ m(\mathbb{Z}_ {\ell })$.
Remark 11. Can define $\mathcal{X}_ {^{L}G,F}=(R_ {W_ F,{}^LG})\times_ {R_ {\Gamma_ {W_ F,\widetilde{F}/F}}}{\widetilde{\text{pr}}}/\hat{G}$, where ${}^LG=\hat{G}\rtimes \Gamma_ {\widetilde{F}/F}$. Fix $\sqrt{q}\in\overline{\mathbb{Q}_ {\ell }}$. Then $$ \mathcal{X}_ {{}^LG,F}\otimes\mathbb{Z}_ {\ell }[\sqrt{q}^{\pm1}]\xrightarrow{\cong}\mathcal{X}_ {{}^cG,F}\otimes\mathbb{Z}_ {\ell }[\sqrt{q}^{\pm1}]. $$ The isomorphism is given as follows: given $\varphi=(\varphi_ 0,\text{pr})\in {}^LG=\hat{G}\rtimes \Gamma_ {\widetilde{F}/F}$, we map it to $$\widetilde{\varphi}(\gamma):=(\varphi_ 0(\gamma)2\rho\sqrt{\text{cycl}(\gamma)},\text{cycl}^{-1}(\gamma),\text{pr}(\gamma)).$$
Remark 12. Can define $\mathcal{X}_ {{}^cG,\breve{F}}$ with $W_ F$ replaced by $I_ F$ and $\mathrm{Gal}(\overline{\breve{F}}/\breve{F})=I_ F$ (i.e. $\breve{F}$ is the completion of the maximal unramified extension of $F$). Choose $\sigma\in W_ F$ a lifting of $\mathrm{Frob}$ and $\overline{\sigma}=\widetilde{pr}(\sigma)\in\mathbb{G}_ m\rtimes\Gamma_ {\widetilde{F}/F}$. Then action of $\sigma$ on $I_ F$ gives an action of $\phi$ on $\mathcal{X}_ {{}^cG,\breve{F}}$.
Proposition 13. We have a Cartesian square
Sketch of proof of Proposition 2.13.
It is easy to see such a commutative diagram exists. To check if it is an isomorphism, * check classical points; * compare tangent complexes.
Idea of the proof of Thm 2.10.
We sketch the strategy of the proof.
-
Find a dense discrete subgroup in $W_ F$. We always have $$ 1\rightarrow \mathcal{P}_ F\rightarrow W_ F\rightarrow W_ F^t\rightarrow 1, $$ where $\mathcal{P}_ F$ is the wild inertia and $$ 1\rightarrow I_ F^t\rightarrow W_ F^t\rightarrow\mathbb{Z}\rightarrow 1. $$ where tame inertia $I_ F^t\cong\prod_ {(l,p)=1}\mathbb{Z}_ {\ell }(1)$.
Definition: $q$-tame group $\Gamma_ q=\langle\sigma,\tau|\sigma\tau\sigma^{-1}=\tau^q\rangle$
Fact:
- There exists a splitting $W_ F^t\rightarrow W_ F$.
- $\iota:\Gamma_ q\hookrightarrow W_ F^t$ is dense, where $\iota$ is defined by $$ \begin{split} \tau&\mapsto \text{(topological) generator of }I_ F^t,\ \sigma&\mapsto\text{lifting of }\mathrm{Frob}. \end{split} $$
- The group $\Gamma_ L=\mathrm{Gal}(\overline{F}/L)$ is a normal subgroup of $\mathcal{P}_ F$ wild inertia, and the finitely generated discrete group $\Gamma’:=\mathcal{P}_ F/\Gamma_ L\rtimes \Gamma_ q\hookrightarrow W_ F/\Gamma_ L$.
Lemma: $R_ {L/F,{}^cG}:=R_ {W_ F/\Gamma_ L,{}^cG}\xrightarrow{\cong}R_ {\Gamma’,{}^cG}$.
The proof of this lemma again checks points and tangent complexes.
- $Q:=\mathcal{P}_ F/\Gamma_ L$ is a normal subgroup of $\Gamma’$ and is actually a finite $p$-group. The map $\varphi:\Gamma’\rightarrow {}^cG$ restricts to $\varphi|_ Q:Q\rightarrow {}^cG$. Then $$ {}^cG\supseteq N_ {{}^cG}(\varphi|_ Q)\supseteq Z_ {{}^cG}(\varphi|_ Q)\supseteq Z_ {{}^cG}(\varphi|_ Q)^\circ, $$ where $N_ {{}^cG}(\varphi|_ Q)$ is smooth and $Z_ {{}^cG}(\varphi|_ Q)^\circ$ is reductive. Need to show that $R_ {\Gamma_ q,N_ {{}^cG}(\varphi|_ Q)}$ has good properties.
- Suppose that $H$ is a smooth group and $H^\circ$ reductive. Consider $R_ {\Gamma_ q,H}={(g,h)\in H^2:ghg^{-1}=h^q}$.
Proposition: This space is flat over $\mathbb{Z}_ {\ell }$, l.c.i. reduced of dimension $\dim H$.
Recall that $$ R_ {\Gamma_ q,H}={(g,h)\in H^2: ghg^{-1}=h^q}, $$ which is claimed to be an affine scheme, dimension $\dim H$, and hence l.c.i and flat over $\Lambda:=\mathbb{Z}[\frac{1}{q}],$ and reduced.
Now we prove this claim. We have
where
and $H^{[q]}$ is defined to be the pullback.
Lemma 14. We have the following.
- $(H//H)^{[q]}$ is finite over $\Lambda$.
- Fix any $\xi\in (H//H)^{[q]}(K)$, where $K$ is an algebraically closed field, then $\text{char}^{-1}(\xi)\subseteq H^{[q]}$ has only finitely many conjugacy classes.
Fix $\mathcal{O}\subseteq \text{char}^{-1}(\xi)$ an $H$-conjugacy class. Let $h\in\mathcal{O}$. Then $\text{res}^{-1}(h)$ is a torsor under $Z_ H(h)$. In particular, $R_ {\Gamma_ q,H}$ is generically reduced.
$(\text{char}\circ \text{res})^{-1}(\xi)$ has dimension $\dim H$ or ($-\infty$). By finiteness of choices of $\xi$ and $\mathcal{O}$, we see $$ \dim R_ {q,H}=\dim H. $$
Corollary 15. Irreducible components of $R_ {\Gamma_ q,H}$ are classified by $(\xi,\mathcal{O},\pi_ 0Z_ H(h)/\sim)$.
Then by miraculous flatness theorem (fiberwisely correct dimension over a Cohen-Maucauly base), we know that $R_ {\Gamma_ q, H}$ is flat. We know it is generically reduced. Since it is l.c.i, it is actually reduced.
🔗Coherent Sheaves
Let $\mathcal{X}_ {{}^cG,F}$ be the stack of $L$-parameters over $\mathbb{Z}_ {\ell }$ as before. Then $$ \begin{split} \mathrm{Coh}(\mathcal{X}_ {{}^cG,F}) &= \mathrm{Coh}^{\hat{G}}(\mathcal{X}_ {{}^cG,F}^\square)\ &=\varinjlim_ {L}\mathrm{Coh}^{\hat{G}}(\mathcal{X}_ {{}^cG,L/F}^\square). \end{split} $$
🔗Stacks of Tame and Unipotent Parameters
Suppose that $G$ splits over a tamely ramified extension, i.e. $\widetilde{F}/F$ is tame. So $F^t\widetilde{F}=F^t$. Take $L=F^t$. So $W(L/F)=W_ F^t$ the tame quotient of $W_ F$, i.e. $$ 1\rightarrow I_ F^t\rightarrow W_ F^t\rightarrow \mathbb{Z}\rightarrow 1, $$ and $I_ F^t\cong\prod_ {(l,q)=1}\mathbb{Z}_ {\ell }(1)$, $\Gamma_ q\hookrightarrow W_ F^t$ open dense subgroup.
Definition 16. We define $\mathcal{X}_ {{}^cG,F}^{\text{tame},\square}:=\mathcal{X}_ {{}^cG,L/F}^\square\xrightarrow[\text{fix }\iota:\Gamma_ q\hookrightarrow W_ F]{\cong}{g,h \in {}^cG: ghg^{-1}=h^q, d(g)=(q^{-1},\overline{\sigma}), d(h)=(1,\overline{\tau})}$, where $d:{}^cG\rightarrow\mathbb{G}_ m\times \Gamma_ {\widetilde{F}/F}$ is the natural projection, and $\text{pr}:\Gamma_ q=\langle\sigma,\tau|\sigma\tau\sigma^{-1}=\tau^q\rangle\rightarrow \Gamma_ {\widetilde{F}/F}$, and $\overline{\sigma}=\text{pr}({\sigma})$, $\overline{\tau}=\text{pr}(\tau)$.
We have a restriction map $$ \mathcal{X}_ {{}^cG,F}^{\text{tame}}\xrightarrow{\text{res}}\hat{G}\overline{\tau}/\hat{G}\rightarrow \hat{G}\overline{\tau}//\hat{G} $$ induced by $(g,h)\mapsto h$, and note that $d(h)=(1,\overline{\tau})$.
Note that $\overline{\tau}=1$ if and only if $G$ splits over an unramified extension since $\overline{\tau}$ is a (topological) generator of the (tame) inertia. Let $\hat{U}\subseteq \hat{G}$ the unipotent variety (consisting of all unipotent elements).
Definition 17 (A temporary definition). We define using the following diagram:
Lemma 18. Over $\mathbb{Q}_ {\ell }$, $\mathcal{X}_ {{}^cG,F}^{\text{unip}}$ is a connected component of $\mathcal{X}_ {{}^cG,F}^{\text{tame}}$.