Bourbaki

§ 2. Tits systems

In this paragraph, the letters G, B, N, S, T, W have the meaning indicated in no. 1 below.

1. Definitions and first properties

Let G be a group and B a subgroup of G. The group B×BB \times B acts on G by (b,b′).g=bgb′−1(b, b').g = bg{b'}^{-1} for b,b′∈Bb, b' \in B and g∈Gg \in G. The orbits of B×BB \times B on G are the sets BgBBgB for g∈Gg \in G, and are called the double cosets of G with respect to B. They form a partition of G; the corresponding quotient is denoted by B\G/BB \backslash G / B. If C and C′C' are double cosets, CC′CC' is a union of double cosets.

Definition 1

A Tits system is a quadruple (G,B,N,S)(G, B, N, S), where G is a group, B and N are two subgroups of G and S is a subset of N/(B∩N)N/(B \cap N), satisfying the following axioms:

(T1) The set B∪NB \cup N generates G and B∩NB \cap N is a normal subgroup of N. (T2) The set S generates the group W=N/(B∩N)W = N/(B \cap N) and consists of elements of order 2. (T3) sBw⊂BwB∪BswBsBw \subset BwB \cup BswB for s∈Ss \in S and w∈Ww \in W. (T4) For all s∈Ss \in S, sBs∉BsBs \notin B.

The group W=N/(B∩N)W = N/(B \cap N) is sometimes called the Weyl group of the Tits system (G,B,N,S)(G, B, N, S).

5 Every element of W is a coset modulo B∩NB \cap N, and is thus a subset of G; hence products such as BwBBwB make sense. More generally, for any subset A of W, we denote by BAB the subset ⋃w∈ABwB\bigcup_{w \in A} BwB.

Remark 1

We shall see in no. 5 (Cor. of Th. 3) that, if (G,B,N)(G, B, N) is given, there exists at most one subset SS of N/(B∩N)N/(B \cap N) such that (G,B,N,S)(G, B, N, S) is a Tits system.

Remark 2

Let (G,B,N,S)(G, B, N, S) be a Tits system, and let ZZ be a normal subgroup of GG contained in BB. Let G′=G/Z,B′=B/Z,N′=N/(Z∩N)G' = G/Z, B' = B/Z, N' = N/(Z \cap N), and let S′S' be the image of SS in N′/(B′∩N′)N'/(B' \cap N'). Then one sees immediately that (G′,B′,N′,S′)(G', B', N', S') is a Tits system.

Throughout this paragraph, with (G,B,N,S)(G, B, N, S) denoting a Tits system, we set T=B∩NT = B \cap N and W=N/TW = N/T. A double coset means a double coset of GG with respect to BB. For any w∈Ww \in W, we set C(w)=BwBC(w) = BwB; this is a double coset.

We are going to deduce several elementary consequences of the axioms (T1) to (T4). We denote by w,w′,…w, w', \ldots elements of WW and by s,s′,…s, s', \ldots elements of SS. The following relations are clear:

C(1)=B,C(ww′)⊂C(w).C(w′),C(w−1)=C(w)−1. C(1) = B, \quad C(ww') \subset C(w).C(w'), \quad C(w^{-1}) = C(w)^{-1}.

Axiom (T3) can also be written in the form

C(s).C(w)⊂C(w)∪C(sw). C(s).C(w) \subset C(w) \cup C(sw).

Moreover, since C(sw)⊂C(s).C(w)C(sw) \subset C(s).C(w) by (1) and since C(s).C(w)C(s).C(w) is a union of double cosets, there are only two possibilities:

C(s).C(w)={C(sw)if C(w)⊄C(s).C(w)C(w)∪C(sw)if C(w)⊂C(s).C(w). C(s).C(w) = \begin{cases} C(sw) & \text{if } C(w) \not\subset C(s).C(w) \\ C(w) \cup C(sw) & \text{if } C(w) \subset C(s).C(w). \end{cases}

By (T4), B≠C(s).C(s)B \neq C(s).C(s); putting w=sw = s in (3) and using the relation s2=1s^2 = 1, we obtain

C(s).C(s)=B∪C(s). C(s).C(s) = B \cup C(s).

This formula shows that B∪C(s)B \cup C(s) is a subgroup of GG. Multiplying both sides of (4) on the right by C(w)C(w), and using formula (3) and the relation

B.C(w)=C(w), B.C(w) = C(w),

we obtain

C(s).C(s).C(w)=C(w)∪C(sw). C(s).C(s).C(w) = C(w) \cup C(sw).

Taking the inverses of the sets entering into formulas (2), (3) and (5) and then replacing ww by w−1w^{-1}, we obtain the formulas

C(w).C(s)⊂C(w)∪C(ws) C(w).C(s) \subset C(w) \cup C(ws)
C(w).C(s)={C(ws)if C(w)⊄C(w).C(s)C(w)∪C(ws)if C(w)⊂C(w).C(s) C(w).C(s) = \begin{cases} C(ws) & \text{if } C(w) \not\subset C(w).C(s) \\ C(w) \cup C(ws) & \text{if } C(w) \subset C(w).C(s) \end{cases}
C(w).C(s).C(s)=C(w)∪C(ws). C(w).C(s).C(s) = C(w) \cup C(ws).

Lemma 1

Let s1,…,sq∈Ss_1, \ldots, s_q \in S and let w∈Ww \in W. We have

C(s1…sq).C(w)⊂⋃(i1,…,ip)C(si1…sipw), C(s_1 \ldots s_q).C(w) \subset \bigcup_{(i_1, \ldots, i_p)} C(s_{i_1} \ldots s_{i_p} w),
where (i1,…,ip)(i_1, \ldots, i_p) denotes the set of strictly increasing sequences of integers in the interval [1,q][1, q].

We argue by induction on qq, the case q=0q = 0 being trivial. If q⩾1q \geqslant 1, we have C(s1…sq).C(w)⊂C(s1).C(s2…sq).C(w)C(s_1 \ldots s_q).C(w) \subset C(s_1).C(s_2 \ldots s_q).C(w). By the induction hypothesis, C(s2…sq).C(w)C(s_2 \ldots s_q).C(w) is contained in the union of the C(sj1…sjpw)C(s_{j_1} \ldots s_{j_p} w), where

2≤j1<⋯<jp≤q. 2 \leq j_1 < \cdots < j_p \leq q.
By (T3), the set C(s1).C(sj1…sjpw)C(s_1).C(s_{j_1} \ldots s_{j_p} w) is contained in the union of the sets C(s1sj1…sjpw)C(s_1 s_{j_1} \ldots s_{j_p} w) and C(sj1…sjpw)C(s_{j_1} \ldots s_{j_p} w). This proves the lemma.

2. An example

Let kk be a field, nn an integer ≥0\geq 0, and (ei)(e_i) the canonical basis of knk^n. Let G=GL(n,k)G = \mathbf{GL}(n, k), let BB be the upper triangular subgroup of GG, and let NN be the subgroup of GG consisting of the matrices having exactly one non-zero element in each row and column. An element of NN permutes the lines keike_i; this gives rise to a surjective homomorphism N→SnN \to \mathscr{S}_n whose kernel is the subgroup T=B∩NT = B \cap N of diagonal matrices, and allows us to identify W=N/TW = N/T with Sn\mathscr{S}_n. We denote by sjs_j (1≤j≤n−11 \leq j \leq n-1) the element of WW corresponding to the transposition of jj and j+1j+1; let SS be the set of sjs_j. The quadruple (G,B,N,S)(G, B, N, S) is a Tits system. Indeed:

Axiom (T1) follows from Cor. 2 of Prop. 14 of Algebra, Chap. II, § 10, no. 13.

Axiom (T2) is proved in Algebra, Chap. I, Correction to p. 97. Axiom (T4) is immediate. It remains to verify axiom (T3), i.e.

sjBw⊂BwB∪BsjwBfor 1≤j≤n−1,w∈W, s_j B w \subset B w B \cup B s_j w B \quad \text{for } 1 \leq j \leq n-1, w \in W,
or equivalently,
sjB⊂BB′∪BsjB′,with B′=wBw−1. s_j B \subset B B' \cup B s_j B', \quad \text{with } B' = w B w^{-1}.
Let GjG_j be the subgroup of GG consisting of the elements that fix the eie_i for i≠j,j+1i \neq j, j+1 and stabilize the plane spanned by eje_j and ej+1e_{j+1}; this group is isomorphic to GL(2,k)\mathbf{GL}(2, k). One checks that GjB=BGjG_j B = B G_j. Since sj∈Gjs_j \in G_j, we have sjB⊂BGjs_j B \subset B G_j, and it suffices to prove that
Gj⊂(B∩Gj)(B′∩Gj)∪(∩Gj)sj(B′∩Gj). G_j \subset (B \cap G_j)(B' \cap G_j) \cup (\cap G_j) s_j (B' \cap G_j).
Identify GjG_j with GL(2,k)\mathbf{GL}(2, k); the group B∩GjB \cap G_j is then identified with the upper triangular subgroup B2B_2 of GL(2,k)\mathbf{GL}(2, k), while the group B′∩GjB' \cap G_j is identified with B2B_2 when w(j)<w(j+1)w(j) < w(j+1) and with the lower triangular subgroup B2−B_2^- otherwise. In the first case, the formula to be proved can be written

GL(2,k)=B2∪B2sB2where s=(0110); \mathbf{GL}(2, k) = B_2 \cup B_2 s B_2 \quad \text{where } s = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix};

this follows for example from the fact that B2B_2 is the stabilizer of a point for the action of GL(2,k)\mathbf{GL}(2, k) on the projective line P1(k)\mathbf{P}_1(k), and acts transitively on the complement of this point. In the second case, the formula to be proved can be written

GL(2,k)=B2B2−∪B2sB2−; \mathbf{GL}(2, k) = B_2 B_2^- \cup B_2 s B_2^-;

since B2−=sB2sB_2^- = s B_2 s, this follows from the preceding formula by multiplying on the right by ss.

3. Decomposition of g into double cosets

Theorem 1

We have G=BWBG = \mathrm{BWB}. The map w↦C(w)w \mapsto C(w) is a bijection from WW to the set B\G/BB \backslash G / B of double cosets of GG with respect to BB.

It is clear that BWB\mathrm{BWB} is stable under x↦x−1x \mapsto x^{-1}, and Lemma 1 shows that it is stable under the product. Since it contains BB and NN, it is equal to GG.

It remains to prove that C(w)≠C(w′)C(w) \neq C(w') if w≠w′w \neq w'. For this, we shall prove by induction on the integer qq the following assertion:

(A_q) If ww and w′w' are distinct elements of WW such that lS(w)≥lS(w′)=ql_S(w) \geq l_S(w') = q, then C(w)≠C(w′)C(w) \neq C(w').

(For the definition of lS(w)l_S(w), see § 1, no. 1.)

This assertion is clear for q=0q = 0, since then w′=1w' = 1 and w≠1w \neq 1, hence C(w′)=BC(w') = B and C(w)≠BC(w) \neq B.

Assume that q≥1q \geq 1 and that w,w′w, w' satisfy the hypotheses of (A_q). There exists s∈Ss \in S such that sw′sw' is of length q−1q - 1. We have

lS(w)>lS(sw′) l_S(w) > l_S(sw')

hence w≠sw′w \neq sw'. Moreover, sw≠sw′sw \neq sw'; by formula (3) of § 1, no. 1, we have

lS(sw)≥lS(w)−1≥lS(sw′)=q−1. l_S(sw) \geq l_S(w) - 1 \geq l_S(sw') = q - 1.

By the induction hypothesis, C(sw′)C(sw') is distinct from C(w)C(w) and from C(sw)C(sw); from formula (2) it follows that

C(sw′)∩C(s).C(w)=∅. C(sw') \cap C(s).C(w) = \varnothing.

Since C(sw′)⊂C(s).C(w′)C(sw') \subset C(s).C(w'), we have finally that C(w)≠C(w′)C(w) \neq C(w').

Remark

Axiom (T4) was not used in the preceding proof.

4. Relations with Coxeter systems

Theorem 2

The pair (W,S)(W, S) is a Coxeter system. Moreover, for s∈Ss \in S and w∈Ww \in W, the relations C(sw)=C(s).C(w)C(sw) = C(s).C(w) and lS(sw)>lS(w)l_S(sw) > l_S(w) are equivalent.

For any s∈Ss \in S, let PsP_s be the set of elements w∈Ww \in W such that

C(s).C(w)=C(sw). C(s).C(w) = C(sw).
We are going to verify that the PsP_s satisfy conditions (A'), (B') and (C) of § 1, no. 7; the two assertions of the theorem will then follow from Prop. 6 of § 1, no. 7.

Condition (A') is clear.

We verify (B'). If PsP_s and sPssP_s had an element ww in common, we would have w∈Psw \in P_s and sw∈Pssw \in P_s, and hence

C(s).C(w)=C(sw),C(s).C(sw)=C(w). C(s).C(w) = C(sw), \qquad C(s).C(sw) = C(w).
It would follow that C(s).C(s).C(w)=C(w)C(s).C(s).C(w) = C(w) and, by formula (5), this would imply that C(w)=C(sw)C(w) = C(sw), which would contradict Th. 1.

We verify (C). Let s,s′∈Ss, s' \in S and w,w′∈Ww, w' \in W with w′=ws′w' = ws'. The assumption that w∈Psw \in P_s and w′∉Psw' \notin P_s implies that

C(sw)=C(s).C(w)(9) C(sw) = C(s).C(w) \tag{9}
C(w′)⊂C(s).C(w′)(10) C(w') \subset C(s).C(w') \tag{10}
by (3).

From (9) and the relation w=w′s′w = w's', it follows that

C(s)w′s′B=C(sw).(11) C(s)w's'B = C(sw). \tag{11}
By formula (2'), C(w′).C(s′)⊂C(w′)∪C(w′s′)C(w').C(s') \subset C(w') \cup C(w's'), which immediately implies that
C(w′)s′B⊂C(ws′)∪C(w).(12) C(w')s'B \subset C(ws') \cup C(w). \tag{12}
Since C(w′)C(w') is a union of left cosets gBgB and since
C(s).C(w′)=C(s)w′B, C(s).C(w') = C(s)w'B,
formula (10) shows that C(s)w′C(s)w' meets C(w′)C(w') and a fortiori that C(s)w′s′BC(s)w's'B meets C(w′)s′BC(w')s'B. It follows from formulas (11) and (12) that the double coset C(sw)C(sw) is equal to one of the double cosets C(ws′)C(ws') and C(w)C(w); since sw≠wsw \neq w, Th. 1 allows us to conclude that sw=ws′sw = ws'.

Corollary 1

Let* w1,…,wq∈Ww_1, \ldots, w_q \in W and let w=w1…wqw = w_1 \ldots w_q. If

lS(w)=lS(w1)+⋯+lS(wq), l_S(w) = l_S(w_1) + \cdots + l_S(w_q),
then
C(w)=C(w1)…C(wq). C(w) = C(w_1) \ldots C(w_q).

On taking reduced decompositions of the wiw_i, one is reduced to the case of a reduced decomposition

w=s1…sq,with si∈S. w = s_1 \ldots s_q, \quad \text{with } s_i \in S.

If u=s2…squ = s_2 \ldots s_q, then w=s1uw = s_1 u and lS(s1u)>lS(u)l_S(s_1 u) > l_S(u), so C(w)=C(s1).C(u)C(w) = C(s_1).C(u) by the theorem. The required formula follows from this by induction on qq.

Corollary 2

Let w∈Ww \in W and let TwT_w be the subset of WW associated to ww by the procedure of Lemma 2 of § 1, no. 4. If t∈Twt \in T_w, then

C(t)⊂C(w).C(w−1). C(t) \subset C(w).C(w^{-1}).

If t∈Twt \in T_w, there exist by definition elements w′,w′′∈Ww', w'' \in W and s∈Ss \in S such that

w=w′sw′′,lS(w)=lS(w′)+lS(w′′)+1andt=w′sw′−1. w = w' s w'', \quad l_S(w) = l_S(w') + l_S(w'') + 1 \quad \text{and} \quad t = w' s {w'}^{-1}.

By Cor. 1,

C(w).C(w−1)=C(w′).C(s).C(w′′).C(w′′−1).C(s).C(w′−1). C(w).C(w^{-1}) = C(w').C(s).C(w'').C({w''}^{-1}).C(s).C({w'}^{-1}).

Hence,

C(w).C(w−1)⊃C(w′).C(s).C(s).C(w′−1). C(w).C(w^{-1}) \supset C(w').C(s).C(s).C({w'}^{-1}).

By (4), C(s)⊂C(s).C(s)C(s) \subset C(s).C(s). Hence,

C(w).C(w−1)⊃C(w′).C(s).C(w′−1)⊃C(t). C(w).C(w^{-1}) \supset C(w').C(s).C({w'}^{-1}) \supset C(t).

Corollary 3

Let w∈Ww \in W and let HwH_w be the subgroup of GG generated by C(w).C(w−1)C(w).C(w^{-1}). Then:

a) For any reduced decomposition (s1,…,sq)(s_1, \ldots, s_q) of ww,

C(sj)⊂Hwfor 1≤j≤q. C(s_j) \subset H_w \quad \text{for } 1 \leq j \leq q.

b) The group HwH_w contains C(w)C(w) and is generated by C(w)C(w).

We prove a) by induction on jj. Assume that C(sk)C(s_k) is contained in HwH_w for k<jk < j. Let

t=(s1…sj−1)sj(s1…sj−1)−1. t = (s_1 \ldots s_{j-1}) s_j (s_1 \ldots s_{j-1})^{-1}.

The element tt belongs to the subset TwT_w of WW defined in Lemma 2 of § 1, no. 4. By Cor. 2, C(t)⊂HwC(t) \subset H_w, and hence C(sj)⊂HwC(s_j) \subset H_w.

Since C(w)=C(s1)…C(sq)C(w) = C(s_1) \ldots C(s_q), cf. Cor. 1, we have C(w)⊂HwC(w) \subset H_w, and b) follows.

Example

Th. 2, applied to the Tits system described in no. 2, shows that the symmetric group Sn\mathfrak{S}_n, with the set of transpositions of consecutive elements, is a Coxeter group.

5. Subgroups of g containing b

For any subset X of S, we denote by W_X the subgroup of W generated by X (cf. §1, no. 8) and by G_X the union BW_XB of the double cosets C(w), w ∈ W_X. We have G_∅ = B and G_S = G.

Theorem 3

a) For any subset X of S, the set G_X is a subgroup of G, generated by ⋃s∈XC(s)\bigcup_{s \in X} C(s).

b) The map X ↦ G_X is a bijection from P(S)\mathscr{P}(S) to the set of subgroups of G containing B.

c) Let (X_i)_{i \in I} be a family of subsets of X. If X=⋂i∈IXiX = \bigcap_{i \in I} X_i, then GX=⋂i∈IGXiG_X = \bigcap_{i \in I} G_{X_i}.

d) Let X and Y be two subsets of S. Then GX⊂GYG_X \subset G_Y (resp. GX=GYG_X = G_Y) if and only if X⊂YX \subset Y (resp. X=YX = Y).

It is clear that GX=(GX)−1G_X = (G_X)^{-1}; Lemma 1 of no. 1 shows that GX.GX⊂GXG_X . G_X \subset G_X; and hence a) follows, taking into account Cor. 1 of Th. 2.

The injectivity of X ↦ G_X follows from that of X ↦ W_X (§1, no. 8, Th. 2). Conversely, let H be a subgroup of G containing B. Let U be the set of w∈Ww \in W such that C(w)⊂HC(w) \subset H. We have H=BUBH = BUB since H is a union of double cosets. Let X=U∩SX = U \cap S; we show that H=GXH = G_X. Clearly, GX⊂HG_X \subset H. On the other hand, let u∈Uu \in U and let (s1,…,sq)(s_1, \ldots, s_q) be a reduced decomposition of u. Cor. 3 of Th. 2 implies that C(sj)⊂HC(s_j) \subset H, and hence that sj∈Xs_j \in X for 1≤j≤q1 \leq j \leq q. Thus, u∈WXu \in W_X, and since H is the union of the C(u)C(u) for u∈Uu \in U, we have H⊂GXH \subset G_X, which proves b).

Assertions c) and d) follow from analogous properties of W_X (§ 1, no. 8, Th. 2).

Corollary

The set S consists of the elements w∈Ww \in W such that w≠1w \neq 1 and B∪C(w)B \cup C(w) is a subgroup of G.

The elements w∈Ww \in W such that B∪C(w)B \cup C(w) is a subgroup of G are those for which there exists X⊂SX \subset S with WX={1,w}W_X = \{1, w\}. Moreover, if w≠1w \neq 1, we necessarily have Card(X) = 1 , i.e. w∈Sw \in S.

Remark 1

The above corollary shows that S is determined by (G, B, N); for this reason, we sometimes allow ourselves to say that (G, B, N) is a Tits system, or that (B, N) is a Tits system in G.

Proposition 1

Let X be a subset of S and N' a subgroup of N whose image in W is equal to W_X. Then, (G_X, B, N', X) is a Tits system.

We have GX=BWXB=BN′BG_X = BW_XB = BN'B, which shows that GXG_X is generated by B∪N′B \cup N'. The verification of the axioms (T1) to (T4) is now immediate.

Proposition 2

Let X, Y ⊂ S and w∈Ww \in W. We have

GXwGY=BWXwWYB. G_X w G_Y = BW_X w W_Y B.

Let s1,…,sq∈Xs_1, \ldots, s_q \in X and t1,…,tq∈Yt_1, \ldots, t_q \in Y. Lemma 1 shows that

C(s1…sq).C(w).C(t1…tq)⊂BWXwWYB, C(s_1 \ldots s_q).C(w).C(t_1 \ldots t_q) \subset BW_X w W_Y B,
and hence that
GXwGY⊂BWXwWYB. G_X w G_Y \subset BW_X w W_Y B.
The opposite inclusion is obvious.

Remark 2

Denote by GX\G/GYG_X \backslash G / G_Y the set of subsets of GG of the form GXgGYG_X g G_Y, g∈Gg \in G; and define WX\W/WYW_X \backslash W / W_Y analogously. The preceding proposition shows that the canonical bijection w↦C(w)w \mapsto C(w) from WW to B\G/BB \backslash G / B defines by passage to the quotient a bijection WX\W/WY→GX\G/GYW_X \backslash W / W_Y \to G_X \backslash G / G_Y.

Proposition 3

Let X⊂SX \subset S and g∈Gg \in G. The relation gBg−1⊂GXg B g^{-1} \subset G_X implies that g∈GXg \in G_X.

Let w∈Ww \in W be such that g∈C(w)g \in C(w). Since BB is a subgroup of GXG_X, the hypothesis gBg−1⊂GXg B g^{-1} \subset G_X implies that C(w).C(w−1)⊂GXC(w).C(w^{-1}) \subset G_X, and hence that C(w)⊂GXC(w) \subset G_X by Cor. 3 of Th. 2, so gg belongs to GXG_X.

6. Parabolic subgroups

Definition 2

A subgroup of GG is said to be parabolic if it contains a conjugate of BB.

It is clear that every subgroup that contains a parabolic subgroup is parabolic.

Proposition 4

Let PP be a subgroup of GG.

a) PP is parabolic if and only if there exists a subset XX of SS such that PP is conjugate to GXG_X (cf. no. 5 for the definition of GXG_X).

b) Let X,X′⊂SX, X' \subset S and g,g′∈Gg, g' \in G be such that P=gGXg−1=g′GX′g′−1P = g G_X g^{-1} = g' G_{X'} {g'}^{-1}. Then, X=X′X = X' and g′g−1∈Pg' g^{-1} \in P.

Assertion a) follows from Th. 3, b). Under the hypotheses of b), we have

g−1g′Bg′−1g⊂g−1g′GX′g′−1=GX, g^{-1} g' B {g'}^{-1} g \subset g^{-1} g' G_{X'} {g'}^{-1} = G_X,
and Prop. 3 shows that g−1g′∈GXg^{-1} g' \in G_X. Hence, GX′=GXG_{X'} = G_X and X′=XX' = X by Th. 3, b). Finally,
g′g−1=g.g−1g′.g−1∈gGXg−1, g' g^{-1} = g . g^{-1} g' . g^{-1} \in g G_X g^{-1},
which proves b).

If the parabolic subgroup PP is conjugate to GXG_X, where X⊂SX \subset S, then PP is said to be of type X.

Theorem 4

(i) Let P1P_1 and P2P_2 be two parabolic subgroups of GG whose intersection is parabolic and let g∈Gg \in G be such that gP1g−1⊂P2gP_1g^{-1} \subset P_2. Then g∈P2g \in P_2 and P1⊂P2P_1 \subset P_2.

(ii) Two parabolic subgroups whose intersection is parabolic are not conjugate.

(iii) Let Q1Q_1 and Q2Q_2 be two parabolic subgroups of GG contained in a subgroup QQ of GG. Then any g∈Gg \in G such that gQ1g−1=Q2gQ_1g^{-1} = Q_2 belongs to QQ.

(iv) Every parabolic subgroup is its own normaliser6^6.

Assertion (i) follows from Props. 3 and 4, and implies (ii). Under the hypotheses of (iii), we have gQ1g−1⊂QgQ_1g^{-1} \subset Q, which implies that g∈Qg \in Q by (i). Finally, (iv) follows from (iii) by taking Q1=Q2=QQ_1 = Q_2 = Q.

Proposition 5

Let P1P_1 and P2P_2 be two parabolic subgroups of GG. Then P1∩P2P_1 \cap P_2 contains a conjugate of TT.

By first transforming P1P_1 and P2P_2 by an inner automorphism of GG, we may assume that B⊂P1B \subset P_1. Let g∈Gg \in G be such that gB2g−1⊂P2gB_2g^{-1} \subset P_2. By Th. 1, there exist n∈Nn \in N and b,b′∈Bb, b' \in B such that g=bnb′g = bn b'. Since TT is normal in NN,

P2⊃gB2g−1=bnBn−1b−1⊃bnTn−1b−1=bTb−1 P_2 \supset gB_2g^{-1} = bnBn^{-1}b^{-1} \supset bnTn^{-1}b^{-1} = bTb^{-1}

and

P1⊃B⊃bTb−1, P_1 \supset B \supset bTb^{-1},

which proves the proposition.

7. Simplicity theorems

Lemma 2

Let HH be a normal subgroup of GG. There exists a subset XX of SS such that BH=GXBH = G_X and such that every element of XX commutes with every element of S−XS - X.

Since BHBH is a subgroup of GG containing BB, there exists a unique subset XX of SS such that BH=GXBH = G_X (Th. 3).

Let s1∈Xs_1 \in X and s2∈S−Xs_2 \in S - X; let n1n_1 and n2n_2 be representatives in NN of s1s_1 and s2s_2, respectively. Then n1∈GX=BHn_1 \in G_X = BH and there exists b∈Bb \in B such that bn1∈Hbn_1 \in H. Since HH is normal in GG, the element h=n2bn1n2−1h = n_2 b n_1 n_2^{-1} of GG belongs to HH. This means that

h∈C(s2).C(s1).C(s2). h \in C(s_2).C(s_1).C(s_2).

If the length of s2s1s2s_2 s_1 s_2 is equal to 3, Cor. 1 of Th. 2 implies that

C(s2).C(s1).C(s2)=C(s2s1s2), C(s_2).C(s_1).C(s_2) = C(s_2 s_1 s_2),

(6) If HH is a subgroup of a group GG, the normaliser of HH in GG is the subgroup N(H)\mathfrak{N}(H) consisting of the elements gg of GG such that gHg−1=HgHg^{-1} = H. A subgroup H′H' is said to normalise HH if H′⊂N(H)H' \subset \mathfrak{N}(H), in which case HH′=H′HHH' = H'H is a subgroup of GG in which HH is normal.

and hence that h∈H∩C(s2s1s2)h \in H \cap C(s_2 s_1 s_2). Since H∩C(s2s1s2)H \cap C(s_2 s_1 s_2) is non-empty, s2s1s2∈WXs_2 s_1 s_2 \in W_X. As (s2,s1,s2)(s_2, s_1, s_2) is a reduced decomposition, it follows that s2∈Xs_2 \in X, contrary to our assumption.

Thus lS(s2s1s2)≤2l_S(s_2 s_1 s_2) \leq 2; if lS(s2s1s2)=1l_S(s_2 s_1 s_2) = 1, then s1s2∈Ss_1 s_2 \in S and so (s1s2)2=1(s_1 s_2)^2 = 1, or s1s2=s2s1s_1 s_2 = s_2 s_1. If lS(s2s1s2)=2l_S(s_2 s_1 s_2) = 2, property (E) of §1, no. 5 implies that s2s1=s1s2s_2 s_1 = s_1 s_2, since s1≠s2s_1 \neq s_2. Q.E.D.

The following property of a group U enters into Th. 5 below:

(R) For any normal subgroup V of U distinct from U, the commutator subgroup (cf. Algebra, Chap. I, §6, no. 8) of U/V is distinct from U/V.

Every soluble group satisfies (R); in particular, every abelian group satisfies (R). It can be shown that the symmetric group Sn\mathfrak{S}_n satisfies (R) (cf. Exerc. 29).

Theorem 5

Let Z be the intersection of the conjugates of B, let U be a subgroup of B and let G_1 be the subgroup generated by the conjugates of U in G. We make the following assumptions:

(1) U is normal in B and B = UT. (2) U has property (R). (3) G1 is equal to its commutator subgroup*. (4) The Coxeter system (W, S) is irreducible (cf. §1, no. 9).

Then every subgroup H of G normalised by G1 is either contained in Z or contains G_1*.

First we show that G = G1 T*. The group G1TG_1 T contains B and hence is its own normaliser (Th. 4); but as N normalises G1G_1 and T, it also normalises G1TG_1 T, so N⊂G1TN \subset G_1 T. Since G is generated by B and N, it follows that G=G1TG = G_1 T.

Next, set

G′=G1H,B′=B∩G′,N′=N∩G′,T′=T∩G′=B′∩N′andW′=N′/T′. G' = G_1 H, \quad B' = B \cap G', \quad N' = N \cap G', T' = T \cap G' = B' \cap N' \quad \text{and} \quad W' = N'/T'.

We have G=G′TG = G'T since G′G' contains G1G_1, and hence N=N′TN = N'T. The inclusion of N′N' into N thus defines, on passing to the quotient, an isomorphism α:W′→W\alpha : W' \to W. Let S′=α−1(S)S' = \alpha^{-1}(S).

We now show that (G', B', N', S') is a Tits system. Since G=BNBG = BNB and B=TU=UTB = TU = UT, we have G=UNUG = UNU. Since U is a subgroup of G′G', it follows that G′=UN′UG' = UN'U; since U⊂B′U \subset B', this proves (T1). Axiom (T2) is satisfied since α\alpha is an isomorphism. Let w∈Ww \in W and let w′=α−1(w)w' = \alpha^{-1}(w) be the corresponding element of W′W'. We have

BwB=BwB′=Bw′B′,since B=B′T. BwB = BwB' = Bw'B', \quad \text{since } B = B'T.

From this we conclude that G′∩BwB=B′w′B′G' \cap BwB = B'w'B', which means that the inclusion of G′G' into G defines on passing to the quotient a bijection from

B'\setminus G'/B' to B\setminus G/B. Axiom (T3) follows immediately. Axiom (T4) follows from B = B'T.

The subgroup H is normal in G'. By Lemma 2 applied to (G', B', N', S'), there exists a subset X' of S' such that B'H = G'_X', and every element of S' - X' commutes with every element of X'. In view of assumption (4), there are only two possibilities:

a) X' = \varnothing, i.e. B'H = B', so H \subset B' \subset B. If g \in G, then g = g_1 t with g_1 \in G_1, t \in T, and H \subset g_1 B g_1^{-1} since G_1 normalises H. Thus H \subset g B g^{-1}, and since Z is the intersection of the g B g^{-1}, we have H \subset Z.

b) X' = S', i.e. B'H = G'. Since G = G'T, we have

G=B′HT=HB′T=HB. G = B'HT = HB'T = HB.

As B normalises U, every conjugate of U is of the form hUh^{-1} with h \in H. Such a subgroup is contained in the group UH, hence G_1 \subset UH by the definition of G_1. Thus, we have the isomorphisms

U/(U∩H)≅UH/H=G1H/H≅G1/(G1∩H). U/(U \cap H) \cong UH/H = G_1 H/H \cong G_1/(G_1 \cap H).

By assumption (3), G_1/(G_1 \cap H) is equal to its commutator subgroup. Assumption (2) now shows that the group U/(U \cap H), which is isomorphic to G_1/(G_1 \cap H), reduces to the identity element. Hence G_1 \cap H = G_1 and G_1 \subset H, which completes the proof.

Corollary

Under the assumptions of Th. 5, the group G_1/(G_1 \cap Z) is either simple non-abelian or reduces to the identity element.

Th. 5 shows that G_1/(G_1 \cap Z) is simple or reduces to the identity element. On the other hand, assumption (3) implies that it is equal to its commutator subgroup. Hence the corollary.

Remark 1

Assumptions (2), (3), (4) were not used in the proof that (G', B', N', S') is a Tits system.

Remark 2

Suppose that Z \cap U = \{1\}. Since Z and U are normal in B, it follows that every element of Z commutes with every element of U, and hence with every element of G_1. In view of the preceding corollary, it follows that G_1 \cap Z is the centre of G_1.

Remark 3

Assumption (3) is implied by the following condition:

(3') U is generated by the commutators b^{-1} u^{-1} b u with u \in U and b \in B \cap G_1.

Example 1

Let k be a field, n an integer \geqslant 0, G = \mathbf{GL}(n, k), and let (G, B, N, S) be the Tits system described in no. 2. Let U be the strictly upper triangular subgroup of G, i.e. the subgroup of B consisting of the matrices whose diagonal entries are equal to 1. Condition (1) in Th. 5 is immediate, and so is (2) since U is soluble. Condition (4) is satisfied if n≥2n \geq 2. One can show (cf. Algebra, Chap. II, § 10, Exerc. 13) that (3) is satisfied if n≥3n \geq 3 and Card(k)≥4(k) \geq 4. Under these conditions, we conclude that G1/(G1∩Z)G_1 / (G_1 \cap Z) is simple and that G1∩ZG_1 \cap Z is the centre of G1G_1 (cf. Remark 2).

When kk is commutative, G1=SL(n,k)G_1 = \mathbf{SL}(n, k) (cf. Algebra, Chap. III, § 8, no. 9).

Example 2

Let gg be a simple Lie algebra over C\mathbf{C}, and let GG be the group of inner automorphisms of gg (cf. Chap. III, § 6, no. 2, Prop. 2). By using Th. 5, one can show that GG is simple non-abelian.

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.