3. Endoscopic Groups

H H is an unramified endoscopic group of G G if it is an unramified reductive group over F F whose classifying data has the form

( X , X , Φ H , Φ H , σ H ) . (X^*,X_*,\Phi _H,\Phi _H^{\vee },\sigma _H).

The first two entries are the same for G G as for H H . To distinguish the data for H H from that for G G , we add subscripts H H or G G , as needed. The data for H H is subject to the constraints that there exists an element s Hom ( X , C × ) s \in \operatorname {Hom}(X_*,\mathbb {C}^\times ) and a Weyl group element w W ( Φ G ) w\in W(\Phi _G) such that

  • Φ H = { α Φ G   |   s ( α ) = 1 } \Phi _H^{\vee }= \{\alpha \in \Phi _G^{\vee } {\ \vert \ }s(\alpha )=1\} ,

  • σ H = w σ G \sigma _H = w\circ \sigma _G , and

  • σ H ( s ) = s \sigma _H(s) = s .

3.1. Endoscopic groups for S L ( 2 ) SL(2)

As an example, we determine the unramified endoscopic groups of G = S L ( 2 ) G = SL(2) . The character group X X^* can be identified with Z \mathbb {Z} , where n Z n\in \mathbb {Z} is identified with the character on the diagonal torus given by

( t 0 0 t 1 ) t n . \begin{pmatrix} t&0\\ 0&t^{-1} \end{pmatrix} \mapsto t^n.

The set Φ \Phi can be identified with the subset { ± 2 } \{\pm 2\} of Z \mathbb {Z} :

( t 0 0 t 1 ) t ± 2 . \begin{pmatrix} t&0\\ 0&t^{-1} \end{pmatrix} \mapsto t^{\pm 2}.

The cocharacter group X X_* is also identified with Z \mathbb {Z} , where n Z n\in \mathbb {Z} is identified with

t ( t n 0 0 t n ) . t\mapsto \begin{pmatrix} t^n&0\\ 0&t^{-n} \end{pmatrix}.

Under this identification Φ = { ± 1 } \Phi ^{\vee } = \{\pm 1\} . Since the group is split, σ = 1 \sigma = 1 .

We get an unramified endoscopic group by selecting s Hom ( X , C × ) C × s\in \operatorname {Hom}(X_*,\mathbb {C}^\times )\cong \mathbb {C}^\times and w W ( Φ ) w\in W(\Phi ) .

(3.0.1) Φ H = { α   |   s ( α ) = 1 } = { n { ± 1 }   |   s n = 1 } = if     ( s = 1 )     then     Φ G     else     . \begin{equation} \begin{array}{lll} \Phi _H^{\vee } = \{\alpha {\ \vert \ }s(\alpha )=1\} &= \{n\in \{\pm 1\}{\ \vert \ }s^n = 1\} \\&= \text{if}\ \ (s = 1)\ \ \text{then}\ \ \Phi _G^{\vee }\ \ \text{else}\ \ \emptyset . \end{array} \tag{3.0.1} \end{equation}

We consider two cases, according as w w is nontrivial or trivial. If w w is the nontrivial reflection, then σ H = w \sigma _H = w acts by negation on Z \mathbb {Z} . Thus,

( σ H ( s ) = s )         ( s 1 = s )     ( s = ± 1 ) . (\sigma _H(s) = s)\ \ \implies \ \ ( s^{-1} = s)\ \ \implies (s = \pm 1).

If s = 1 s=1 , then σ H \sigma _H does not fix a set of simple roots as required. So s = 1 s=-1 and Φ H = \Phi _H^{\vee } = \emptyset . Thus, the root data of H H is

( Z , Z , , , w ) (\mathbb {Z},\mathbb {Z},\emptyset ,\emptyset ,w)

This determines H H up to isomorphism as H = U E ( 1 ) H=U_E(1) , a 1 1 -dimensional torus split by an unramified quadratic extension E / F E/F .

If w w is trivial, then there are two further cases, according as Φ H \Phi _H is empty or not:

  • The endoscopic group G m \mathbb {G}_m has root data ( Z , Z , , , 1 ) . (\mathbb {Z},\mathbb {Z},\emptyset ,\emptyset ,1).

  • The endoscopic group H = S L ( 2 ) H=SL(2) has root data ( Z , Z , { ± 2 } , { ± 1 } , 1 ) . (\mathbb {Z},\mathbb {Z},\{\pm 2\},\{\pm 1\},1).

In summary, the three unramified endoscopic groups of S L ( 2 ) SL(2) are U E ( 1 ) U_E(1) , G m \mathbb {G}_m , and S L ( 2 ) SL(2) itself.

3.2. Endoscopic groups for P G L ( 2 ) PGL(2)

As a second complete example, we determine the endoscopic groups of P G L ( 2 ) PGL(2) . The group P G L ( 2 ) PGL(2) is dual to S L ( 2 ) SL(2) in the sense that the coroots of one group can be identified with the roots of the other group. The root data for P G L ( 2 ) PGL(2) is

( Z , Z , { ± 1 } , { ± 2 } , 1 ) . (\mathbb {Z},\mathbb {Z},\{\pm 1\},\{\pm 2\},1).

When the Weyl group element is trivial, then the calculation is almost identical to the calculation for S L ( 2 ) SL(2) . We find that there are again two cases, according as Φ H \Phi _H is empty or not:

  • The endoscopic group G m \mathbb {G}_m has root data ( Z , Z , , , 1 ) . (\mathbb {Z},\mathbb {Z},\emptyset ,\emptyset ,1).

  • The endoscopic group H = P G L ( 2 ) H=PGL(2) has root data ( Z , Z , { ± 1 } , { ± 2 } , 1 ) . (\mathbb {Z},\mathbb {Z},\{\pm 1\},\{\pm 2\},1).

When the Weyl group element w w is nontrivial, then s { ± 1 } s\in \{\pm 1\} , as in the S L ( 2 ) SL(2) calculation.

(3.0.2) Φ H = { α   |   s ( α ) = 1 } = { n { ± 2 }   |   s n = 1 } = Φ G . \begin{equation} \Phi _H^{\vee } = \{\alpha {\ \vert \ }s(\alpha )=1\} = \{n \in \{\pm 2\}{\ \vert \ }s^n = 1\} = \Phi _G^{\vee }. \tag{3.0.2} \end{equation}

From this, we see that picking w w to be nontrivial is incompatible with the requirement that σ H = w \sigma _H=w must fix a set of simple roots. Thus, there are no endoscopic groups with w w nontrivial.

In summary, the two endoscopic groups of P G L ( 2 ) PGL(2) are G m \mathbb {G}_m and P G L ( 2 ) PGL(2) itself.

3.3. Elliptic Endoscopic groups

Definition 3.1

An unramified endoscopic group H H is said to be elliptic, if

( R Φ G ) W ( Φ H ) σ H = ( 0 ) . (\mathbb {R}\Phi _G)^{W(\Phi _H)\rtimes \langle \sigma _H\rangle } = (0).

That is, the span of the set of roots of G G has no invariant vectors under the Weyl group of H H and the automorphism σ H \sigma _H .

The origin of the term elliptic is the following. We will see below that each Cartan subgroup of H H is isomorphic to a Cartan subgroup of G G . (Here and elsewhere, when we speak of an isomorphic between algebraic groups defined over F F , we mean an isomorphism over F F .) The condition on H H for it to be elliptic is precisely the condition that is needed for some Cartan subgroup of H H to be isomorphic to an elliptic Cartan subgroup of G G .

Example 3.2

We calculate the elliptic unramified endoscopic subgroups of S L ( 2 ) SL(2) . We may identify R Φ \mathbb {R}\Phi with R { ± 2 } \mathbb {R}\{\pm 2\} and hence with R \mathbb {R} . An unramified endoscopic group is elliptic precisely when W ( Φ H ) W(\Phi _H) or σ H \langle \sigma _H\rangle contains the nontrivial reflection x x x\mapsto -x . When H = S L ( 2 ) H=SL(2) , the Weyl group contains the nontrivial reflection. When H = U E ( 1 ) H=U_E(1) , the element σ H \sigma _H is the nontrivial reflection. But when H = G m H=\mathbb {G}_m , both W ( Φ H ) W(\Phi _H) and σ H \langle \sigma _H\rangle are trivial. Thus, H = S L ( 2 ) H=SL(2) and H = U E ( 1 ) H=U_E(1) are elliptic, but H = G m H=\mathbb {G}_m is not.

3.4. An exercise: elliptic endoscopic groups of unitary groups

This exercise is a calculation of the elliptic unramified endoscopic groups of U ( n , J ) U(n,J) . We assume that J J is a cross-diagonal matrix with units along the cross-diagonal as in Section 1.6.1. We give a few facts about the endoscopic groups of U ( n , J ) U(n,J) and leave it as an exercise to fill in the details.

Let T = { diag ( t 1 , . . . , t n ) } T = \{\operatorname {diag}(t_1,\mathinner {\ldotp \ldotp \ldotp },t_n)\} be the group of diagonal n n by n n matrices. The character group X X^* can be identified with Z n \mathbb {Z}^n in such a way that the character

diag ( t 1 , . . . , t n ) t 1 k 1 t n k n \operatorname {diag}(t_1,\mathinner {\ldotp \ldotp \ldotp },t_n) \mapsto t_1^{k_1}\cdots t_n^{k_n}

is identified with ( k 1 , . . . , k n ) (k_1,\mathinner {\ldotp \ldotp \ldotp },k_n) .

The cocharacter group can be identified with Z n \mathbb {Z}^n in such a way that the cocharacter

t diag ( t k 1 , . . . , t k n ) t\mapsto \operatorname {diag}(t^{k_1},\mathinner {\ldotp \ldotp \ldotp },t^{k_n})

is identified with ( k 1 , . . . , k n ) (k_1,\mathinner {\ldotp \ldotp \ldotp },k_n) .

Let e i e_i be the basis vector of Z n \mathbb {Z}^n whose j j -th entry is Kronecker δ i j \delta _{ij} . The set of roots can be identified with

Φ = { e i e j   |   i j } . \Phi = \{e_i-e_j {\ \vert \ }i\ne j\}.

The set of coroots Φ \Phi ^{\vee } can be identified with the set of roots Φ \Phi under the isomorphism X Z n X X_* \cong \mathbb {Z}^n \cong X^* .

We may identify Hom ( X , C × ) \operatorname {Hom}(X_*,\mathbb {C}^\times ) with Hom ( Z n , C × ) = ( C × ) n \operatorname {Hom}(\mathbb {Z}^n,\mathbb {C}^\times )= (\mathbb {C}^\times )^n . Thus, we take the element s s in the definition of endoscopic group to have the form s = ( s 1 , . . . , s n ) ( C × ) n s=(s_1,\mathinner {\ldotp \ldotp \ldotp },s_n)\in (\mathbb {C}^\times )^n . The element σ = σ G \sigma = \sigma _G acts on characters and cocharacters by

σ ( k 1 , . . . , k n ) = ( k n , . . . , k 1 ) . \sigma (k_1,\mathinner {\ldotp \ldotp \ldotp },k_n) = (-k_n,\mathinner {\ldotp \ldotp \ldotp },-k_1).

Let I = { 1 , . . . , n } I = \{1,\mathinner {\ldotp \ldotp \ldotp },n\} . Show that if H H is an elliptic unramified endoscopic group, then there is a partition

I = I 1 I 2 I = I_1\coprod I_2

with s i = 1 s_i = 1 for i I 1 i\in I_1 and s i = 1 s_i = -1 otherwise. The elliptic endoscopic group is a product of two smaller unitary groups H = U ( n 1 ) × U ( n 2 ) H = U(n_1)\times U(n_2) , where n i = # I i n_i = \#I_i , for i = 1 , 2 i=1,2 .