Inductions
๐Spectral Parabolic Induction
Let $P\subseteq G$ be a parabolic with levi $M$. Then on the dual group side, we have the corresponding ${}^cG\supseteq P\twoheadrightarrow {}^cM$.
Then we obtain the following correspondence
Facts:
- $\mathcal{X}_ { {}^cG,F}$ and $\mathcal{X}_ { {}^cM,F}$ are classical algebraic stacks, while $\mathcal{X}_ { {}^cP,F}$ is in general derived.
- $\mathcal{X}_ { {}^cP, F}$ is quasi-smooth, i.e. its cotangent complex is perfect and concentrated on degree $[-1,1]$ (This can be viewed as a derived version of l.c.i, or intuitively visualized as a space cut out by equations inside a smooth space.).
- The map $r$ is schematic, proper and hence we have $r_ *:\mathrm{Coh}(\mathcal{X}_ { {}^cP,F})\rightarrow \mathrm{Coh}(\mathcal{X}_ { {}^cG,F})$.
- The map $\pi$ is quasi-smooth (It means that the relative cotangent complex is perfect and concentrated on degree $[-1,1]$.) and hence we have a functor $\pi^!:\mathrm{Coh}(\mathcal{X}_ { {}^cM,F})\rightarrow\mathrm{Coh}(\mathcal{X}_ { {}^cP,F})$.
Definition 1 (Spectral parabolic induction). We define spectral parabolic induction (with respect to parabolic $P$) to be $$r_ *\pi^!:\mathrm{Coh}(\mathcal{X}_ {{}^cM,F})\rightarrow\mathrm{Coh}(\mathcal{X}_ {{}^cG,F}).$$
Now we assume that $G$ splits over a tamely ramified extension. The spectral parabolic induction naturally restricts to the tame part, i.e. we have correspondence
Now we take parabolic $P$ to be Borel, i.e. ${}^cP={}^cB$ and ${}^cM={}^cT=\hat{T}\rtimes(\mathbb{G}_ m\times\Gamma_ {\widetilde{F}/F})$.
Definition 2. We define $$\mathrm{CohSpr}^{\textrm{tame}}:=r_ *\pi^!\mathcal{O}_ {\mathcal{X}_ {{}^cT,F}^{\textrm{tame}}},$$ and $$\mathrm{CohSpr}^{\textrm{unip}}:=r_ *\pi^!\mathcal{O}_ {\mathcal{X}_ {{}^cT,F}^{\textrm{unip}}}=r_ *\pi^!\mathcal{O}_ {\mathcal{X}_ {{}^cT,F}^{\textrm{unr}}}.$$
Note that unipotent part for $T$ is the same as unramified part, bacause requiring $I_ T$ to map into unipotent part of $T$ is the same as requiring $I_ T$ maps to 1 (any unipotent subgroup of $T$ is trivial).
In particular, $\mathcal{X}_ {{}^cT,F}^{\textrm{unr}}=\hat{T}\overline{\sigma}/\hat{T}\hookrightarrow\mathcal{X}_ {{}^cT,F}^{\textrm{tame}}$.
Theorem 3. Let coefficient ring $\Lambda=\overline{\mathbb{Q}_ {\ell}}$ (This assumption is expected to be removed.).
- The sheaf $\mathrm{CohSpr}^{\textrm{unip}}$ is an honest coherent sheaf on $\mathcal{X}_ { {}^cG,F}$.
- We have an algebra isomorphism $R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}}(\mathrm{CohSpr}^{\textrm{unip}})^{\textrm{op}}\cong H_ I$ the Iwahori Hecke algebra of $G$ (put at degree 0).
Remark 4. In particular, we have a functor $$\begin{split} H_ I\textrm{-Mod}&\rightarrow \mathrm{IndCoh}(\mathcal{X}_ {{}^cG,F})\ M &\mapsto \mathrm{CohSpr}^{\textrm{unip}}\otimes_ {H_ I}M \end{split}$$
Let $G$ be a reductive group over $F$ as before. Assume that $G$ is unramified, i.e. $G$ has an integral model over $\mathcal{O}_ F$. Then we define $H_ {\textrm{sph}}:=C_ c(G(\mathcal{O})\backslash G(F)/G(\mathcal{O}))$. We have $\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}\cong\hat{G}\overline{\sigma}/\hat{G}$.
Theorem 5. Let coefficient ring $\Lambda=\overline{\mathbb{Q}_ {\ell}}$. Then $$R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}}(\mathcal{O}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}})\cong H_ {\textrm{sph}}.$$
Remark 6. An exercise: prove this by hand for $G=\mathrm{PGL}_ 2$, $\hat{G}=\mathrm{SL}_ 2$.
Remark 7. We have $$R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}}(\mathcal{O}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}})\cong R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{tame}}}(\mathcal{O}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}})$$ and in characteristic 0, we further have $$R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{tame}}}(\mathcal{O}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}})\cong R\mathrm{End}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unip}}}(\mathcal{O}_ {\mathcal{X}_ {{}^cG,F}^{\textrm{unr}}}).$$
๐A Magical Correspondence
Consider the following general framework.
- $\phi_ x:\mathcal{X}\rightarrow \mathcal{X}$ and $\phi_ y:\mathcal{Y}\rightarrow \mathcal{Y}$ two endomorphisms of two spaces.
- $f:\mathcal{X}\rightarrow \mathcal{Y}$ a morphism compatible with $\phi_ x,\phi_ y$, i.e. $\phi_ y\circ f=f\circ\phi_ x$.
Note that $\mathcal{L}_ {\phi}\mathcal{Y}:=\mathcal{Y}\times_ {{}^{\Gamma_ {\phi_ y}}\mathcal{Y}\times \mathcal{Y}^\Delta} \mathcal{Y}$ is the (derived) $\phi_ y$-fixed points
Then we have a fundamental correspondence:
Let me explain the construction in detail. First, note that $\mathcal{X}\times_ {\mathcal{Y}}\mathcal{X} = (\mathcal{X}\times \mathcal{X})\times_ {\mathcal{Y}\times \mathcal{Y}^\Delta}\mathcal{Y}$, where
Next, we define
Then $f\times \textrm{id}: \mathcal{X}\times_ {\mathcal{Y}}\mathcal{L}_ \phi\mathcal{Y}\rightarrow \mathcal{L}_ \phi\mathcal{Y}$ gives the definition of $\pi$ in the correspondence above.
Finally, Since $f\circ\phi_ x=\phi_ y\circ f$, then $({\textrm{id}_ \mathcal{X}},\phi_ x)\times\textrm{id}_ {\mathcal{Y}}:\mathcal{X}\times_ {\mathcal{Y}\times\mathcal{Y}^\Delta}\mathcal{Y}\rightarrow (\mathcal{X}\times\mathcal{X})\times_ {\mathcal{Y}\times\mathcal{Y}^\Delta}\mathcal{Y}$ gives the definition of $\delta$ in the correspondence.
Remark 8. This seems abstract but this underlies many considerations in representation theory, e.g. Deligne-Lusztig theory.
๐Spectral Deligne-Lusztig Induction
Back to our situation:
- take $\mathcal{X}=\mathcal{X}_ {{}^cB.\breve{F}}^{\textrm{tame}}={\rho: I_ F^t\rightarrow {}^cB,d\circ\rho=(\textrm{cycl}^{-1},\textrm{pr})}/\hat{B}$. Fix a lifting of $\mathrm{Frob}$ to $W_ F^t$, denoted by $\phi$.
- Take $\mathcal{Y}=\mathcal{X}_ {{}^cG,\breve{F}}^{\textrm{tame}}={\rho:I_ F^t\rightarrow {}^cG,d\circ\rho=(\textrm{cycl}^{-1},\textrm{pr})}/\hat{G}$, also acted by $\phi$.
- There is a natural $\phi$-equivariant map (induced by ${}^cB\rightarrow {}^cG$) $$f:\mathcal{X}\rightarrow \mathcal{Y}.$$
Using Prop. 2.9, we know that $\mathcal{L}_ \phi\mathcal{Y}=\mathcal{X}_ {{}^cG,F}^{\textrm{tame}}$. Now apply the general construction above, and we obtain
where
- $\widetilde{\mathcal{X}_ {{}^cG,F}^{\textrm{tame}}}:=\mathcal{X}_ {{}^cB,\breve{F}}^{\textrm{tame}}\times_ {\mathcal{X}_ {{}^cG,\breve{F}}^{\textrm{tame}}}\mathcal{X}_ {{}^cG,F}^{\textrm{tame}}$.
- $S_ {{}^cG,F}:=\mathcal{X}\times_ \mathcal{Y}\mathcal{X}$ is called Steinberg stack.
Fact:
- Steinberg stack $S_ {{}^cG,F}$ is classical.
- $\delta$ is quasi-smooth.
- $\pi$ is (ind-)proper.
Then we can make the following definition.
Definition 9 (Spectral Deligne-Lusztig Induction). We define spectral Deligne-Lusztig induction to be $$\pi_ *\circ\delta^!:\mathrm{Coh}(S_ {{}^cG,F})\rightarrow \mathrm{Coh}(\mathcal{X}_ {{}^cG,F}^{\textrm{tame}})$$
Remark 10. It is still open whether or not the above holds when removing the tame assumption.
Fix $\tau\in I_ F^t$ a topological generator. Let $\hat{\mathbb{Z}}^p:=\prod_ {(\ell,p)=1}\mathbb{Z}_ \ell$ denote the prime-to-$p$ part. Then $I_ F^t\cong \hat{\mathbb{Z}}^p$.
Then $\mathcal{X}_ {{}^cB,\breve{F}}^{\textrm{tame}}=(\hat{B}\overline{\tau}/\hat{B})\times_ {\hat{T}\overline{\tau}/\hat{T}}(\hat{T}\overline{\tau}/\hat{T})^{\wedge,p}\hookrightarrow\hat{B}\overline{\tau}/\hat{B}$ induces isomorphisms on tangent spaces.
Example 11. Take $\overline{\tau}=1$ and $\Lambda$ a field.
Similarly, we have maps $$\mathcal{X}_ {{}^cG,\breve{F}}^{\textrm{tame}}\hookrightarrow \hat{G}\overline{\tau}/\hat{G}.$$
We have Grothendieck-Springer alteration $$\hat{B}\overline{\tau}/\hat{B}\rightarrow \hat{G}\overline{\tau}/\hat{G},$$ and
Example 12. Take $\overline{\tau}=1$, and we really recover Grothendieck-Springer resolution. As in the proof of Thm. 2.7, we use an open dense subgroup $\iota: \Gamma_ q\rightarrow I_ F^t$ to describe the stacks, and we denoted the corresponding stacks with an extra subscript $\iota$.
Proposition 13. The stack $S$ is a classical stack of dimension 0, l.c.i. and irreducible components are parameterized by the Weyl group $W$.
In general, we always have
Proposition 14. The stack $S_ {\hat{G}\overline{\tau}}$ is classical, l.c.i. of dimension $0$ over $\Lambda$, whose irreducible components are parameterized by $W_ 0:=W^{\overline{\tau}}$.
Proof.
Note that $$S_ {\hat{G}\overline{\tau}}=S_ {\hat{G}\overline{\tau}}^\square/\hat{G},$$ and $$S_ {\hat{G}\overline{\tau}}^\square=\widetilde{\hat{G}\overline{\tau}}\times_ {\hat{G}\overline{\tau}} \widetilde{\hat{G}\overline{\tau}}$$ where
Example 15. For simplicity, consider $\overline{\tau}=1$. Then $$S_ {\hat{G}}^\square={g\in\hat{G},\hat{B_ 1},\hat{B_ 2}\textrm{ two Borels}:g\in \hat{B_ 1}\cap \hat{B_ 2}}.$$ Then $$f^{-1}(\mathcal{O}(w))={(g,\hat{B_ 1},\hat{B_ 2}):(\hat{B_ 1},\hat{B_ 2})\in\mathcal{O}(w),g\in \hat{B_ 1}\cap\hat{B_ 2}}.$$ Note that
- $\dim\mathcal{O}(w)=\dim \hat{G}/\hat{B}+l(w)$;
- $\dim (\hat{B_ 1}\cap\hat{B_ 2})=\dim (\hat{B}\cap w\hat{B}w^{-1})=\dim \hat{B}-l(w)$;
and therefore, $\dim f^{-1}(\mathcal{O}_ w)=\dim\hat{G}$. Therefore, $\dim f^{-1}(\mathcal{O}_ w)$ has the correct dimension, and stable under $\hat{G}$-action, i.e. $f^{-1}(\mathcal{O}_ w)/\hat{G}$ has dimension 0.
For general $\overline{\tau}$, the spirit is the same but with messier computation.
We define $$S_ {\hat{G}\tau,w}^\square:=\textrm{closure of }f^{-1}(\mathcal{O}_ w).$$ Then we have $$\overline{S_ {\hat{G}\tau,w}}\subseteq \bigcup_ {wโ\leq w} S_ {\hat{G}\overline{\tau},wโ},$$ which is not equality in general.
But for $w=e$ identity, $S_ {\hat{G}\overline{\tau},e}=\hat{B}\overline{\tau}/\hat{B}$ is closed in $S_ {\hat{G},\overline{\tau}}$, so the equality holds for $w=e$.
For simplicity, assume that $\overline{\tau}=1$ (the general case is in the same spirit but more complicated in computation). Then unwinding definitions, we have
where $\widetilde{\mathcal{X}^{\textrm{tame}}_ {{}^cG, F,w}}$ is defined by the pullback square and is called spectral Deligne-Lusztig stack. We have the following diagram
When $w=1$, the condition $\rho(\overline{\sigma})\hat{B}\rho(\sigma^{-1})=\hat{B}$ is same as $\rho(\overline{\sigma})\in\hat{B}$. So, ${\widetilde{\mathcal{X}_ {{}^cG, F,w}^{\textrm{tame}}}}=\mathcal{X}_ {{}^cB,F}^{\textrm{tame}}$ and spectral Deligne-Lusztig induction recovers spectral parabolic induction, similar to the classical Deligne-Lusztig induction.