Bourbaki

§ 2. Affine Weyl group

In this paragraph (except in no. 5), we denote by RR a reduced root system in a real vector space VV. We denote by WW the Weyl group of RR; we identify it with a group of automorphisms of the dual V∗V^* of VV (§ 1, no. 1), and we provide V∗V^* with a scalar product invariant under WW. Let EE be the affine space underlying V∗V^*; for v∈V∗v \in V^*, we denote by t(v)t(v) the translation of EE by the vector vv. Finally, we denote by PP (resp. QQ) the group of translations t(v)t(v) whose vector vv belongs to the group of weights P(R∼)P(R^{\sim}) (resp. to the group of radical weights Q(R∼)Q(R^{\sim})) of the inverse root system R∼R^{\sim} of RR.

1. Affine Weyl group

For α∈R\alpha \in \mathbf{R} and k∈Zk \in \mathbf{Z}, let Lα,kL_{\alpha,k} be the hyperplane of E defined by:

Lα,k={x∈E∣⟨α,x⟩=k} L_{\alpha,k} = \{ x \in E \mid \langle \alpha, x \rangle = k \}

and let sα,ks_{\alpha,k} be the orthogonal reflection with respect to Lα,kL_{\alpha,k}. Then

sα,k(x)=x−(⟨α,x⟩−k)α∼=sα,0(x)+kα∼ s_{\alpha,k}(x) = x - (\langle \alpha, x \rangle - k)\alpha^\sim = s_{\alpha,0}(x) + k\alpha^\sim

for all x∈Ex \in E. In other words,

sα,k=t(kα∼)∘sα(1) s_{\alpha,k} = t(k\alpha^\sim) \circ s_\alpha \tag{1}

where sαs_\alpha is the orthogonal reflection with respect to the hyperplane Lα=Lα,0L_\alpha = L_{\alpha,0}, i.e. the reflection associated to the root α\alpha.

Formula (1) shows that sα,ks_{\alpha,k} does not depend on the choice of scalar product.

Definition 1

The group of affine transformations of E generated by the reflections sα,ks_{\alpha,k} for α∈R\alpha \in \mathbf{R} and k∈Zk \in \mathbf{Z} is called the affine Weyl group of the root system R and denoted by Wa(R)W_a(\mathbf{R}) (or simply by WaW_a).

Proposition 1

The group WaW_a is the semi-direct product of W by Q.

Since W is generated by the reflections sαs_\alpha, it is contained in WaW_a. On the other hand, t(α∼)=sα,1∘sαt(\alpha^\sim) = s_{\alpha,1} \circ s_\alpha if α∈R\alpha \in \mathbf{R}, which shows that Q⊆WaQ \subseteq W_a.

Since W leaves Q(R∼)Q(\mathbf{R}^\sim) stable (§ 1, no. 9), the group G of affine transformations generated by W and Q is the semi-direct product of W by Q. Now G⊆WaG \subseteq W_a from above and sα,k∈Gs_{\alpha,k} \in G for all α∈R\alpha \in \mathbf{R} and k∈Zk \in \mathbf{Z} by (1). It follows that Wa=GW_a = G.

Proposition 2

The group WaW_a, with the discrete topology, acts properly on E and permutes the hyperplanes Lα,kL_{\alpha,k} (for α∈R\alpha \in \mathbf{R} and k∈Zk \in \mathbf{Z}).

Since Q(R∼)Q(\mathbf{R}^\sim) is a discrete subgroup of V∗V^*, the group Q acts properly on E. Hence, so does Wa=W.QW_a = W.Q, since W is finite. Moreover, for α,β∈R\alpha, \beta \in \mathbf{R} and k∈Zk \in \mathbf{Z}:

sβ(Lα,k)=Lγ,kwithγ=sβ(α)∈R, s_\beta(L_{\alpha,k}) = L_{\gamma,k} \quad \text{with} \quad \gamma = s_\beta(\alpha) \in \mathbf{R},
t(β∼)(Lα,k)=Lα,k+n(α,β), t(\beta^\sim)(L_{\alpha,k}) = L_{\alpha,k+n(\alpha,\beta)},

where n(α,β)=⟨β∼,α⟩n(\alpha,\beta) = \langle \beta^\sim, \alpha \rangle is an integer, hence the second assertion.

We can thus apply the results of Chapter V, § 3 to WaW_a acting on E. To avoid any confusion with the chambers of the Weyl group W in V∗V^*, we shall call the chambers determined by the system of hyperplanes Lα,kL_{\alpha,k} (for α∈R\alpha \in \mathbf{R} and k∈Zk \in \mathbf{Z}) in E alcoves. The group WaW_a thus acts simply transitively on the set of alcoves and the closure of an alcove is a fundamental domain for WaW_a acting on E (Chap. V, § 3, no. 2, Th. 1 and no. 3, Th. 2). It is clear that the Weyl group W is identified with the canonical image U(Wa)U(W_a) of WaW_a in the orthogonal group of V* (cf. Chap. V, § 3, no. 6). It follows that W_a is essential (Chap. V, § 3, no. 7) and that W_a is irreducible if and only if the root system R is (§ 1, no. 2, Cor. of Prop. 5). If R is irreducible, every alcove is an open simplex (Chap. V, § 3, no. 9, Prop. 8). In the general case, the canonical product decomposition of the affine space E (Chap. V, § 3, no. 8) corresponds to the decomposition of R into irreducible components. In particular, the alcoves are products of open simplexes.

Note also that the Cor. of Th. 1 of Chap. V, § 3, no. 2 shows that the s_{\alpha,k} are the only reflections in W_a.

2. Weights and special weights

Proposition 3

The special points (Chap. V, § 3, no. 10, Def. 1) of W_a are the weights of R'.

Let x_0 \in E and let \alpha \in R. The hyperplane L parallel to Ker \alpha and passing through x_0 has equation \langle \alpha, x \rangle = \langle \alpha, x_0 \rangle. To be equal to some L_{\beta,k}, it is necessary on the one hand that \alpha and \beta are proportional, and so, since R is reduced, that \beta = \pm \alpha, and on the other hand that \langle \alpha, x_0 \rangle is an integer. It follows immediately that x_0 is a special point of W_a if and only if \langle \alpha, x_0 \rangle \in \mathbf{Z} for all \alpha \in R, in other words, if and only if x_0 \in P(R') (§ 1, no. 9).

Corollary

(i) If \overline{\omega} \in P(R'), there exists an alcove C such that \overline{\omega} is an extremal point of \overline{C}. (ii) If C is an alcove, \overline{C} \cap Q(R') reduces to a single point and this is an extremal point of \overline{C}.

This follows from Prop. 3, in view of the Cor. of Prop. 11 of Chap. V, § 3, no. 10 and Prop. 12 of Chap. V, § 3, no. 10.

Proposition 4

Let C' be a chamber of R'.

(i) There exists a unique alcove C contained in C' such that 0 \in \overline{C}. (ii) The union of the w(\overline{C}) for w \in W is a neighbourhood of 0 in E. (iii) Every wall of C' is a wall of C.

This follows from Prop. 11 of Chap. V, § 3, no. 10.

Assume now that R is irreducible. Let (\alpha_i)_{i \in I} be a basis of R (§ 1, no. 5, Def. 2), and let (\overline{\omega}_i)_{i \in I} be the dual basis. The \overline{\omega}_i are the fundamental weights of R' for the chamber D' of R' corresponding to the basis (\alpha_i). Let

α~=∑i∈Iniαi \tilde{\alpha} = \sum_{i \in I} n_i \alpha_i

be the highest root of R (§ 1, no. 8), and let J be the set of i \in I such that n_i = 1.

Proposition 5

Let C be the alcove contained in C' and containing 0 in its closure (Prop. 4).

(i) CC is the set of x∈Ex \in E such that ⟨αi,x⟩>0\langle \alpha_i, x \rangle > 0 for all i∈Ii \in I and ⟨α~,x⟩<1\langle \tilde{\alpha}, x \rangle < 1.

(ii) The set C‾∩P(R−)\overline{C} \cap P(R^-) consists of 0 and the ω‾i\overline{\omega}_i for i∈Ji \in J.

Let DD be the set of x∈Ex \in E such that ⟨α~,x⟩<1\langle \tilde{\alpha}, x \rangle < 1 and set C1=C′∩DC_1 = C' \cap D. Since 0∈C‾0 \in \overline{C}, we have C⊆DC \subseteq D and hence C⊆C1C \subseteq C_1. We are going to show that, for all α∈R\alpha \in R and all k∈Zk \in \mathbf{Z}, the sets CC and C1C_1 are on the same side of the hyperplane Lα,kL_{\alpha,k}. This will prove that C1⊆CC_1 \subseteq C and so will establish assertion (i). If k=0k = 0, the whole of the chamber C′C' is on one side of Lα,0L_{\alpha,0}, which establishes our assertion in this case. If k≠0k \neq 0, we may, by replacing α\alpha by −α-\alpha, assume that k>0k > 0. Then ⟨α,x⟩<k\langle \alpha, x \rangle < k on CC, since 0∈C‾0 \in \overline{C}. On the other hand, α~−α\tilde{\alpha} - \alpha is positive on C′C' (§ 1, no. 8, Prop. 25). Thus, for y∈C1y \in C_1, we have ⟨α,y⟩≤⟨α~,y⟩<1≤k\langle \alpha, y \rangle \leq \langle \tilde{\alpha}, y \rangle < 1 \leq k. Consequently, CC and C1C_1 are on the same side of Lα,kL_{\alpha,k}.

Now let ω‾∈P(R−)\overline{\omega} \in P(R^-). Then ω‾=∑ipiω‾i\overline{\omega} = \sum_i p_i \overline{\omega}_i, with pi∈Zp_i \in \mathbf{Z} (§ 1, no. 10), and ω‾∈C′‾\overline{\omega} \in \overline{C'} if and only if the integers pip_i are positive. If ω‾∈C′‾\overline{\omega} \in \overline{C'}, then ω‾∈C‾\overline{\omega} \in \overline{C} if and only if ⟨α~,ω‾⟩=∑inipi\langle \tilde{\alpha}, \overline{\omega} \rangle = \sum_i n_i p_i is ≤1\leq 1, hence (ii).

Corollary

The alcove CC is an open simplex with vertices 0 and the ω‾i/ni,i∈I\overline{\omega}_i / n_i, i \in I.

This follows from (i).

3. Normaliser of WaW_a

In this no., we assume that the chosen scalar product on VV is invariant not only under WW but under the whole of the group A(R)A(R). We identify A(R)A(R) and A(R−)A(R^-).

Let GG be the normaliser of WaW_a in the group of displacements of the affine euclidean space EE. If gg is a displacement of EE, and ss is the orthogonal reflection with respect to a hyperplane LL, the displacement gsg−1g s g^{-1} is the orthogonal reflection with respect to the hyperplane g(L)g(L). It follows that GG is the set of displacements of EE that permute the hyperplanes Lα,kL_{\alpha,k} (for α∈R\alpha \in R and k∈Zk \in \mathbf{Z}).

Now, the group of automorphisms of EE is the semi-direct product of the orthogonal group UU of V∗V^* and the group TT of translations. If u∈Uu \in U and v∈V∗v \in V^*, the hyperplane Lα,kL_{\alpha,k} is transformed by g=u∘t(v)g = u \circ t(v) into the hyperplane with equation

⟨tu−1(α),x⟩=k+⟨α,v⟩. \langle {}^t u^{-1}(\alpha), x \rangle = k + \langle \alpha, v \rangle.
Consequently, g∈Gg \in G if and only if, on the one hand tu{}^t u permutes the roots, in other words belongs to A(R)A(R), and on the other hand ⟨α,v⟩∈Z\langle \alpha, v \rangle \in \mathbf{Z} for all α∈R\alpha \in R, that is v∈P(R−)v \in P(R^-). In other words, the group GG is the semi-direct product of A(R)A(R) by PP. Since Q⊆PQ \subseteq P and W⊆A(R)W \subseteq A(R), the quotient group G/WaG/W_a is the semi-direct product of A(R)/WA(R)/W by P(R−)/Q(R−)P(R^-)/Q(R^-); it is immediately checked that the corresponding action of A(R)/WA(R)/W on P(R−)/Q(R−)P(R^-)/Q(R^-) is the canonical action (§ 1, no. 9).

We denote by Wa′W'_a the subgroup of G formed by the semi-direct product of W by P. This is a normal subgroup of G, and G/Wa′G/W'_a is canonically isomorphic to A(R)/WA(R)/W; moreover, the canonical map from P(R−)P(R^-) to Wa′/WaW'_a/W_a gives by passing to the quotient an isomorphism from P(R−)/Q(R−)P(R^-)/Q(R^-) to Wa′/WaW'_a/W_a.

Now let C be an alcove of E, and let GCG_C be the subgroup consisting of the elements g∈Gg \in G such that g(C)=Cg(C) = C. Since WaW_a is simply-transitive on the alcoves, the group G is the semi-direct product of GCG_C by WaW_a. The corresponding isomorphism from G/WaG/W_a to GCG_C gives rise in particular to a canonical isomorphism from P(R−)/Q(R−)P(R^-)/Q(R^-) to the group ΓC=GC∩Wa′\Gamma_C = G_C \cap W'_a.

Assume that R is irreducible, and retain the notation of Prop. 5 of no. 2. Put R0=RR_0 = R, and let RiR_i (i∈Ii \in I) be the root system generated by the aja_j, for j≠ij \neq i. For i=0i = 0 (resp. i∈Ii \in I), let wiw_i be the unique element of W(Ri)W(R_i) (identified with a subgroup of W) which transforms the positive roots of RiR_i relative to the basis (αj)j≠i(\alpha_j)_{j \neq i} into negative roots (§\S 1, no. 6, Cor. 3 of Prop. 17).

Proposition 6

For all i∈Ji \in J, the element γi=t(ω‾i)wiw0\gamma_i = t(\overline{\omega}_i)w_iw_0 belongs to ΓC\Gamma_C and the map i↦γii \mapsto \gamma_i is a bijection from J to ΓC−{1}\Gamma_C - \{1\}.

We remark first of all that the root wi(α~)w_i(\tilde{\alpha}) is of the form

niαi+∑j≠ibijαj, n_i \alpha_i + \sum_{j \neq i} b_{ij} \alpha_j,
and hence is positive.

We show that, if i∈Ji \in J, then γi∈ΓC\gamma_i \in \Gamma_C. Indeed, let a∈Ca \in C and b=γi(a)b = \gamma_i(a). For 1≤j≤l1 \leq j \leq l and j≠ij \neq i,

⟨b,αj⟩=⟨ω‾i+wiw0(a),αj⟩=⟨w0(a),wi(αj)⟩>0 \langle b, \alpha_j \rangle = \langle \overline{\omega}_i + w_iw_0(a), \alpha_j \rangle \\ = \langle w_0(a), w_i(\alpha_j) \rangle > 0
since w0(a)∈−C′w_0(a) \in -C' and wi(αj)w_i(\alpha_j) is negative. On the other hand,
⟨b,αi⟩=1+⟨w0(a),wi(αi)⟩≥1+⟨w0(a),α~⟩>0 \langle b, \alpha_i \rangle = 1 + \langle w_0(a), w_i(\alpha_i) \rangle \geq 1 + \langle w_0(a), \tilde{\alpha} \rangle > 0
since w0(a)∈−C′w_0(a) \in -C', α~−wi(αi)\tilde{\alpha} - w_i(\alpha_i) takes negative values on −C′-C', and ⟨w0(a),α~⟩>−1\langle w_0(a), \tilde{\alpha} \rangle > -1. Finally,
⟨b,α~⟩=ni+⟨w0(a),wi(α~)⟩=1+⟨w0(a),wi(α~)⟩<1 \langle b, \tilde{\alpha} \rangle = n_i + \langle w_0(a), w_i(\tilde{\alpha}) \rangle = 1 + \langle w_0(a), w_i(\tilde{\alpha}) \rangle < 1
since w0(a)∈−C′w_0(a) \in -C' and wi(α~)w_i(\tilde{\alpha}) is a positive root. The relations (2), (3) and (4) then imply that b∈Cb \in C, and hence that γi∈ΓC\gamma_i \in \Gamma_C. It is clear that the map i↦γii \mapsto \gamma_i is injective, since γi(0)=ω‾i\gamma_i(0) = \overline{\omega}_i. Finally, let γ∈ΓC\gamma \in \Gamma_C with γ≠1\gamma \neq 1, and put γ=tw\gamma = tw with t∈Pt \in P and w∈Ww \in W. Then t≠1t \neq 1 since ΓC∩W={1}\Gamma_C \cap W = \{1\}. On the other hand, t(0)=γ(0)∈C‾∩P(R−)t(0) = \gamma(0) \in \overline{C} \cap P(R^-) and Prop. 5 implies that there exists i∈Ji \in J such that t(0)=ω‾it(0) = \overline{\omega}_i. Then γi−1γ(0)=0\gamma_i^{-1}\gamma(0) = 0, hence γ=γi\gamma = \gamma_i since ΓC∩W={1}\Gamma_C \cap W = \{1\}. This completes the proof.

Corollary

The (ω‾i)i∈J(\overline{\omega}_i)_{i \in J} form a system of representatives in P(R−)P(R^-) of the non-zero elements of P(R−)/Q(R−)P(R^-)/Q(R^-).

Indeed, if we identify Γc\Gamma_c with P(R−)/Q(R−)P(R^-)/Q(R^-), the element γi\gamma_i is identified with the class of ω‾i\overline{\omega}_i mod. Q(R−)Q(R^-).

Remark 1

The map γ↦γ(0)\gamma \mapsto \gamma(0) is a bijection from ΓC\Gamma_C to C‾∩P(R−)\overline{C} \cap P(R^-).

Remark 2

The group GG is also the normaliser of WaW_a in the group of automorphisms of EE provided only with its affine structure (cf. Exerc. 3).

4. Application: order of the Weyl group

Lemma 1

Let XX be a locally compact space countable at infinity, GG a group acting continuously and properly on XX, μ\mu a measure on XX invariant under GG, G′G' a subgroup of GG, UU and U′U' two open subsets of XX with finite non-zero measure. Assume that the sUsU for s∈Gs \in G (resp. the s′U′s'U' for s′∈G′s' \in G') are pairwise disjoint and that their union is of negligible complement. Then G′G' is of finite index in GG and (G:G′)=μ(U′)/μ(U)(G : G') = \mu(U')/\mu(U).

Let (sλ)λ∈Λ(s_\lambda)_{\lambda \in \Lambda} be a family of representatives of the right cosets of G′G' in GG. Let U1U_1 be the union of the sλUs_\lambda U. Then the s′U1s'U_1, for s′∈G′s' \in G', are pairwise disjoint and have union M=⋃s∈GsUM = \bigcup_{s \in G} sU. Let M′=⋃s′∈G′s′U′M' = \bigcup_{s' \in G'} s'U'. The union of U′U' (resp. U1U_1) and a suitable subset of X−M′X - M' (resp. X−MX - M) is a fundamental domain, evidently μ\mu-measurable, for G′G'. By Integration, Chap. VII, § 2, no. 10, Cor. of Th. 4, μ(U′)=μ(U1)\mu(U') = \mu(U_1). This proves that Card Λ=(G:G′)\Lambda = (G : G') is finite, and that μ(U′)=(Card Λ)μ(U)\mu(U') = (\text{Card } \Lambda)\mu(U).

Proposition 7

Assume that RR is irreducible. Let B={α1,…,αl}B = \{ \alpha_1, \ldots, \alpha_l \} be a basis of RR, ff the connection index of RR (§ 1, no. 9) and α~=n1α1+⋯+nlαl\tilde{\alpha} = n_1 \alpha_1 + \cdots + n_l \alpha_l the highest root of RR (for the order defined by BB). Then the order of WW is equal to

(l!)n1n2…nlf. (l!) n_1 n_2 \ldots n_l f.

Let (ω‾1,…,ω‾l)(\overline{\omega}_1, \ldots, \overline{\omega}_l) be the basis of P(R−)P(R^-) dual to BB. By the Cor. of Prop. 5, the open simplex CC with vertices 0,n1−1ω‾1,…,nl−1ω‾l0, n_1^{-1} \overline{\omega}_1, \ldots, n_l^{-1} \overline{\omega}_l is an alcove of EE. Choose a Haar measure μ\mu on the additive group V∗V^*. Let AA be the set of elements of V∗V^* of the form ξ1ω‾1+⋯+ξlω‾l\xi_1 \overline{\omega}_1 + \cdots + \xi_l \overline{\omega}_l, with 0<ξi<10 < \xi_i < 1 for i=1,…,li = 1, \ldots, l. By Cor. 2 of Prop. 15 of Integration, Chap. VII, § 1, no. 10,

μ(A)/μ(C)=(l!)n1n2…nl.(5) \mu(A)/\mu(C) = (l!) n_1 n_2 \ldots n_l. \tag{5}
On the other hand, let A′A' be the set of elements of V∗V^* of the form
ξ1α1~+⋯+ξlαl~, \xi_1 \alpha_{\tilde{1}} + \cdots + \xi_l \alpha_{\tilde{l}},
with 0<ξi<10 < \xi_i < 1 for i=1,…,li = 1, \ldots, l. Since (α1~,…,αl~)(\alpha_{\tilde{1}}, \ldots, \alpha_{\tilde{l}}) is a basis of the Z\mathbf{Z}-module Q(R−)Q(R^-), we can apply Lemma 1 with X=V∗X = V^*, G=WaG = W_a, G′=QG' = Q, U=CU = C and U′=A′U' = A'. This gives
μ(A′)/μ(C)=(Wa:Q)=Card W.(6) \mu(A')/\mu(C) = (W_a : Q) = \text{Card } W. \tag{6}

Finally, Lemma 1 can be applied again, taking X = V*, G = P, G' = Q, U = A and U' = A'. This gives

μ(A′)/μ(A)=(P:Q)=(P(R−):Q(R−))=f. \mu(A')/\mu(A) = (P : Q) = (P(R^-) : Q(R^-)) = f.

The proposition now follows by comparing formulas (5), (6) and (7).

5. Root systems and groups generated by reflections

Proposition 8

Let F be a real Hilbert space of finite dimension l. Let H\mathfrak{H} be the set of affine hyperplanes of F and G the group generated by the orthogonal reflections sHs_H with respect to the hyperplanes H∈HH \in \mathfrak{H}. Assume that the conditions of Chap. V, § 3 are satisfied (i.e. that g(H)∈Hg(H) \in \mathfrak{H} for all H∈HH \in \mathfrak{H} and g∈Gg \in G, and that G acts properly on F). Assume also that 0 is a special point for G and that the group T of translations belonging to G is of rank l. Then there exists a unique reduced root system R in V=F∗V = F^* such that the canonical isomorphism from F to V∗V^* transforms G to the affine Weyl group WaW_a of R.

We remark first of all that the assumption on T implies that G is essential: otherwise, the affine space F would decompose into a product F0×F1F_0 \times F_1, with dim⁡F1<l\dim F_1 < l, the group G being identified with a group of displacements acting properly on F1F_1 (Chap. V, § 3, no. 8, Prop. 6), and T would not be of rank l.

Let H0\mathfrak{H}_0 be the set of H∈HH \in \mathfrak{H} such that 0∈H0 \in H. For H∈H0H \in \mathfrak{H}_0, let HH\mathfrak{H}_H be the set of elements of H\mathfrak{H} parallel to H. Since 0 is a special point, H\mathfrak{H} is the union of the HH\mathfrak{H}_H for H∈H0H \in \mathfrak{H}_0. Since T is of rank l, there exists a v∈Fv \in F such that the translation by the vector v belongs to T and v∉Hv \notin H. The hyperplanes H+kvH + kv for k∈Zk \in \mathbf{Z} are pairwise distinct and belong to HH\mathfrak{H}_H. Now let a be a unit vector of F orthogonal to H: then H+(v∣a)a∈HHH + (v|a)a \in \mathfrak{H}_H and since H\mathfrak{H} is locally finite (Chap. V, § 1, no. 1, Lemma 1), there is a smallest real number λ>0\lambda > 0 such that H+λa∈HH + \lambda a \in \mathfrak{H}. We are going to show that HH\mathfrak{H}_H is the set of hyperplanes H+kλaH + k\lambda a for k∈Zk \in \mathbf{Z}. Indeed,

H′=H+λa∈HH H' = H + \lambda a \in \mathfrak{H}_H

and the element sH′∘sHs_{H'} \circ s_H of G is the translation by the vector 2λa2\lambda a (Chap. V, § 2, no. 4, Prop. 5). Consequently, H+2nλa=(sH′sH)n(H)H + 2n\lambda a = (s_{H'} s_H)^n(H) and H+(2n+1)λa=(sH′sH)n(H′)H + (2n + 1)\lambda a = (s_{H'} s_H)^n(H') belongs to HH\mathfrak{H}_H. On the other hand, if L∈HHL \in \mathfrak{H}_H, there exists ξ∈R\xi \in \mathbf{R} such that L=H+ξλaL = H + \xi \lambda a and there exists an integer n such that

either 2n<ξ≤2n+12n < \xi \leq 2n + 1, or 2n−1<ξ≤2n2n - 1 < \xi \leq 2n.

In the first case, (sHsH′)n(L)=H+(ξ−2n)λa(s_H s_{H'})^n(L) = H + (\xi - 2n)\lambda a with

0<(ξ−2n)λ≤λ 0 < (\xi - 2n)\lambda \leq \lambda

and the definition of λ\lambda implies that ξ=2n+1\xi = 2n + 1; in the second case,

sH(sH′sH′′)n(L)=H+(2n−ξ)λawith0≤(2n−ξ)λ<λ s_H(s_{H'} s_{H''})^n(L) = H + (2n - \xi)\lambda a \quad \text{with} \quad 0 \leq (2n - \xi)\lambda < \lambda
and the definition of λ\lambda implies that ξ=2n\xi = 2n.

It follows that if αH\alpha_H is the linear form on FF such that

H′={x∈F∣⟨αH,x⟩=1}, H' = \{ x \in F \mid \langle \alpha_H, x \rangle = 1 \},
the set HH\mathscr{H}_H is the set of hyperplanes LαH,k={x∈F∣⟨αH,x⟩=k}L_{\alpha_H, k} = \{ x \in F \mid \langle \alpha_H, x \rangle = k \} for k∈Zk \in \mathbf{Z}, and αH\alpha_H and −αH-\alpha_H are the only linear forms with this property.

Consequently, the proposition will be proved if we show that the set R of elements of V of the form ±αH\pm \alpha_H is a reduced root system in V.

a) We prove condition (RS_I): it is clear that R is finite (since H0\mathscr{H}_0 is finite) and does not contain 0. Moreover, R generates V. Indeed, if x∈Fx \in F is orthogonal to R, then x∈Hx \in H for all H∈H0H \in \mathscr{H}_0 and the translation by the vector xx commutes with every element of G. Since G is essential, this implies that x=0x = 0.

b) We prove (RS_{II}). For v∈Vv \in V and r∈Rr \in \mathbf{R}, put Lv,r={x∈F∣⟨v,x⟩=r}L_{v, r} = \{ x \in F \mid \langle v, x \rangle = r \} as above; if α∈R\alpha \in R, put Hα=Lα,0H_\alpha = L_{\alpha, 0}, and let sαs_\alpha be the transpose of sHαs_{H_\alpha}. There exists a unique element α∼∈F\alpha^\sim \in F orthogonal to HαH_\alpha and such that ⟨α∼,α⟩=2\langle \alpha^\sim, \alpha \rangle = 2. Then sHα=sα∼,αs_{H_\alpha} = s_{\alpha^\sim, \alpha} and sα=sα,α∼s_\alpha = s_{\alpha, \alpha^\sim}. For β∈R\beta \in R,

Lsα(β),1=sHα(Lβ,1)∈H L_{s_\alpha(\beta), 1} = s_{H_\alpha}(L_{\beta, 1}) \in \mathscr{H}
and there exist γ∈R\gamma \in R and n∈N∗n \in \mathbf{N}^* such that Lsα(β),1=Lγ,nL_{s_\alpha(\beta), 1} = L_{\gamma, n}. Then
sHα(Lγ,1)=Lβ,1/n s_{H_\alpha}(L_{\gamma, 1}) = L_{\beta, 1/n}
and so 1/n∈Z1/n \in \mathbf{Z}. Thus n=1n = 1 and sα(β)=γ∈Rs_\alpha(\beta) = \gamma \in R. This proves (RS_{II}).

c) We prove (RS_{III}). Let α∈R\alpha \in R and set Hα′=Lα,1H'_\alpha = L_{\alpha, 1}. Then

Hα′=Hα+(1/2)α∼; H'_\alpha = H_\alpha + (1/2)\alpha^\sim;
since the translation t(α∼)t(\alpha^\sim) by the vector α∼\alpha^\sim is the product sHα′sHαs_{H'_\alpha} s_{H_\alpha} (Chap. V, § 2, no. 4, Prop. 5), it belongs to T and α∼=t(α∼)(0)\alpha^\sim = t(\alpha^\sim)(0) is a special point for G. Consequently, for all β∈R\beta \in R, there exists a hyperplane Lβ,kL_{\beta, k} passing through α∼\alpha^\sim, with kk an integer, which shows that ⟨β,α∼⟩∈Z\langle \beta, \alpha^\sim \rangle \in \mathbf{Z}, and proves (RS_{III}).

d) Finally, it is clear that R is reduced, for if H,H′∈H0,H≠H′H, H' \in \mathscr{H}_0, H \neq H', the linear forms αH\alpha_H and αH′\alpha_{H'} are not proportional.

Remark 1

The assumption that T is of rank l is satisfied in particular when G is irreducible and infinite. Indeed, the vector space generated by vectors corresponding to the translations in T is invariant under the canonical image of G in the linear group of F. It is different from {0}\{0\} if G is infinite and is thus equal to the whole of F if G is infinite and irreducible.

A finite group generated by reflections is not always the Weyl group of a root system. More precisely:

Proposition 9

Let V be a real vector space of finite dimension l, and let G be a finite subgroup of GL(V)\mathbf{GL}(V), generated by reflections and essential. Give V a scalar product invariant under G. The following conditions are equivalent:

(i) There exists a discrete subgroup of V of rank l that is stable under G. (ii) There exists a Q\mathbf{Q}-structure on V (Algebra, Chap. II, § 8, no. 1, Def. 1) invariant under G. (iii) There exists a root system in V whose Weyl group is G. (iv) There exists a discrete group G′G' of displacements of V, acting properly on V, and generated by reflections, such that G′G' is the semi-direct product of G and a group of translations of rank l.

(ii) ⟹\Longrightarrow (i): let V′⊆VV' \subseteq V be a Q\mathbf{Q}-structure on V invariant under G. Let A be a finite subset of V′V' generating the Q\mathbf{Q}-vector space V′V'. Replacing A by ⋃s∈Gs(A)\bigcup_{s \in G} s(A), we can assume that A is stable under G. Let B be the subgroup of V generated by A. Then B is stable under G, of finite type and torsion-free, so has a basis over Z\mathbf{Z} which is both a basis of V′V' over Q\mathbf{Q} and a basis of V over R\mathbf{R}.

(iii) ⟹\Longrightarrow (ii): this follows from Prop. 1 of § 1, no. 1, for example. (iv) ⟹\Longrightarrow (iii): let G′G' be a group satisfying condition (iv). The group of translations of G′G' is of rank l, and 0 is a special point for G′G' by Prop. 9 of Chap. V, § 3, no. 10. Prop. 6 shows that there exists a reduced root system R0R_0 in V∗V^* such that G′G' is identified with Wα(R0)W_\alpha(R_0); the group G is then the Weyl group of the inverse root system of R0R_0.

(i) ⟹\Longrightarrow (iv): assume that G leaves stable a discrete subgroup M of V, of rank l. For any reflection s∈Gs \in G, s(x)−x∈Ms(x) - x \in M for all x∈Mx \in M, so the line DsD_s orthogonal to HsH_s meets M; let αs,−αs\alpha_s, -\alpha_s be the generators of the cyclic group Ds∩MD_s \cap M; the set A of the αs\alpha_s and −αs-\alpha_s is stable under G, hence generates a subgroup M′M' of M stable under G; the discrete group M′M' is of rank l because G is essential. Let G′G' be the group of affine transformations of V that is the semi-direct product of G and the group of translations whose vectors belong to M′M'. Let G1′G'_1 be the subgroup of G′G' generated by the reflections of G′G'. We shall show that G1′=G′G'_1 = G', which will complete the proof. First, G1′⊇GG'_1 \supseteq G since G is generated by reflections. On the other hand, for any reflection s of G, let tst_s be the translation with vector αs\alpha_s. The transformation s∘tss \circ t_s is a reflection, and s∘ts∈G′s \circ t_s \in G'; thus tst_s is a product of two reflections of G′G'; this being true for every reflection s of G, the translations whose vector belongs to M′M' are all in G1′G'_1.

Definition 2

A group G satisfying the equivalent conditions of Prop. 9 is called a crystallographic group.

Remark 2

Let G be a finite group generated by reflections and essential. Then G is crystallographic if and only if every element of its Coxeter matrix is one of the integers 1,2,3,4,61, 2, 3, 4, 6. Indeed, this condition is necessary by Remark 3) of § 1, no. 5. The fact that it is sufficient will follow from the classification of finite Coxeter groups given in § 4 (for a direct proof, see Chap. V, § 4, Exerc. 6).

Exercises

We denote by RR a reduced root system in a real vector space VV.

See the exercises for § 2.

Transcribed from the printed volume. Not translated.