CENTERS AND COXETER ELEMENTS W. G. DWYER AND C. W. WILKERSON Abstract. Suppose that G is a connected compact Lie group. We show that simple numerical information about the Weyl group of G can be used to obtain bounds, often sharp, on the size of the center of G. These bounds are obtained with the help of certain Coxeter elements in the Weyl group. Variants of the method use generalized Coxeter elements and apply to p-compact groups; in this case a splitting theorem emerges. The Lie group results are mostly known, but our arguments have a conceptual appeal. 1. Introduction Suppose that G is a connected compact Lie group. The center ZG of G is the subgroup consisting of elements which commute with every- thing else in G. In this paper we develop some simple general methods for obtaining information about ZG . The most striking technique ap- plies if G is simply connected and gives a formula for the order of ZG in terms of the exponents of the Weyl group W of G; the formula is derived by studying certain special elements in W called Coxeter ele- ments. Many of our results apply in one way or another to p-compact groups. Much of what we do is well-known, but does not appear in the standard textbooks; we are particularly interested in minimizing the amount of case-by-case consideration and in eliminating analytical arguments. Before stating our results we give some background on compact Lie groups. See [13 ] for a geometric approach to the structure theorems we use. 1.1. Product decompositions. A connected compact Lie group G has a finite covering group G" isomorphic to a product T x K, where T is a torus and K is a simply connected compact Lie group. The group G is obtained from G" by forming a quotient G"=A, where A is a finite subgroup of ZG", and the center of G is isomorphic to ZG"=A. This isomorphism indicates that in order to understand ZG it is enough to ____________ Date: January 21, 2000. The authors were supported in part by the National Science Foundation. 1 2 W. G. DWYER AND C. W. WILKERSON understand ZG"together with the subgroup A of ZG"which is used in constructing G from G". Since the torus T is abelian, ZG" is just T x ZK . We will mostly concentrate on studying ZK , which is known to be a finite abelian group. In principle this can be done once and for all. Say that K is almost simple if the quotient K=ZK is simple (in the group-theoretic sense of having no nontrivial normal subgroups). Then anyQsimply- connected compact Lie group K is isomorphic to a productQ iKi of simply connected almost simple factors, so that ZK ~= iZKi. The possible factors in this product decomposition are known and can be studied one-by-one; most of these possible factors have standard de- scriptions in terms of matrices. To be explicit, up to isomorphism a simply connected almost simple compact Lie group falls into one of four families or is one of five exceptional examples. The families are An (more concretely SU n+1), Bn (Spin 2n+1), Cn (Sp n), and Dn (Spin 2n); the exceptional examples are G2, F4, E6, E7, and E8. The groups are often paired up with the Dynkin diagrams of the same names. 1.2. Maximal tori and Weyl groups. A compact Lie group G has a maximal torus T unique up to conjugacy; we will usually pick one out and denote it TG (or just T if G is understood). The dimension of T is a numerical invariant of G called the rank `G (or `) of G; the integer subscripts which appear above in the conventional designations for the simply connected almost simple groups correspond to the ranks of the groups, so that, for instance, E7 has rank 7. The maximal torus T is of finite index in its normalizer NG (T ), and the quotient NG (T )=T is called the Weyl group WG (or W ) of G. Let LG denote ss1(TG ) (this is the Weyl lattice of G) and let VG denote Q LG . Then LG ~= Z`, VG ~= Q`. The conjugation action of W on T gives a homomorphism W ! Aut (T ) ! Aut (LG ) ! Aut (VG ) : If G is connected this homomorphism is faithful, and its image is a re- flection subgroup (x2) of Aut (VG ) ~=GL `(Q), i.e., a subgroup generated by the reflection matrices it contains. The group G is almost simple if and only if the action of W on VG affords an irreducible representation of W ; the fundamental group ss1(G) is finite, or equivalently, G has a finite center, if and only if (VG )W = 0. 1.3. Calculating centers. We can now describe our results about the center of a connected compact Lie group. 1.4. Theorem. Suppose that G is a simply connected compact Lie group with maximal torus T and Weyl group W . Then the center ZG is exactly the fixed point set T W. CENTERS 3 1.5. Remark. There is a general formula for ZG in [13 , S8]; it turns out that for an arbitrary connected compact Lie group, ZG differs from T W by an elementary abelian 2-group. Surprisingly, it turns out that in order to compute T W in the simply- connected case it is enough to focus on a single special kind of element in WG . Suppose that G is a compact Lie group of rank `, and that (LG )W = 0; as mentioned above, this will always be the case if G is simply connected. Coxeter [7] [6] proves that it is possible to find ` reflections s1; : :;:s` which generate W ; he goes on to show that, up to conjugacy in W , the product s1s2 . .s.`does not depend either on the choice of generating reflections or on the order in which they are written down. We will call such a product of generating reflections a Coxeter element of W , and denote it wCox. Let #(ZG ) denote the order of the center of G. 1.6. Theorem. Suppose that G is a simply connected compact Lie group with maximal torus T , Weyl group W , and Coxeter element wCox 2 W . Then the center ZG is exactly the fixed point set T , and #(ZG ) = | det(I - wCox)|. If the order of wCox is prime to p, then #(ZG ) is also prime to p. Here I denotes the identity matrix of rank `G , we identify wCox with its image in Aut (LG ) ~=GL `(Z), and det(I -wCox) is the determinant of the matrix (I-wCox). Theorem 1.6 was observed in the 50's, apparently on the basis of case-by-case inspection. Our proof (which uses ideas communicated to us by R. Steinberg) is more conceptual. The real impact of 1.6 comes from the fact that the eigenvaluesQ{ffi} of a Coxeter element wCox, and hence the number det(I -wCox) = i(1-ffi), can be determined (3.3) from the list of exponents of the reflection group WG . This reduces the calculation of the order of ZG to routine arithmetic (see 5.5). Say that an element w 2 WG is a generalized Coxeter element if det(I - w) 6= 0, i.e., if w has no nonzero fixed vectors in its action on VG . Any Coxeter element of WG is a generalized Coxeter element. It turns out that whereas Coxeter elements determine the order of the center of a simply connected group G, generalized Coxeter elements give bounds for the order of the center of an arbitrary connected group. 1.7. Theorem. Suppose that G is a connected compact Lie group with maximal torus T , Weyl group W , and generalized Coxeter element w 2 W . Then the center ZG is finite, and #(ZG ) divides | det(I - w)|. If the order of w is prime to p, then #(ZG ) is also prime to p. 4 W. G. DWYER AND C. W. WILKERSON One source of generalized Coxeter elements is the center of the Weyl group itself. It is possible to do a little bit better than 1.7 in this case. 1.8. Theorem. Suppose that G is a simply connected almost simple compact Lie group with Weyl group W , and that ZW is nontrivial. Then ZW is Z=2, and the nonidentity element w 2 ZW is a general- ized Coxeter element which acts on TG by inversion. The determinant det(I - w) is 2`, and ZG is isomorphic to (Z=2)s for some s `. 1.9. Remark. Suppose that G is a simply connected almost simple com- pact Lie group. Then the center of WG is nontrivial as long as G is not E6, An for n > 1, or Dn+1. This fact can be read off from the list of degrees of WG (see 2.7). Finally, there is one additional class of examples (which only partially overlaps the class of simply connected groups) for which the inclusion 5.2 is an equality. 1.10. Proposition. Suppose that G is a connected compact Lie group of rank `, with Weyl group W Aut (LG ) ~= GL `(Z). If reduction mod 2 gives a monomorphism W ! GL `(Z=2), then ZG = (TG )W . 1.11. Remark. Every finite subgroup of the kernel of the reduction map GL `(Z) ! GL `(Z=2) is an elementary abelian 2-group. Thus the con- clusion of 1.10 applies to G whenever the Weyl group of G has no normal elementary abelian 2-subgroups. The proof of 1.10 shows that the theorem remains true under the weaker hypothesis that the kernel of the mod 2 reduction map contains none of the reflections in W . Thus among the almost simple types, except for some forms of A1, Bn, and Cn, ZG = (TG )W . 1.12. Remark. Combining the above theorems in a straightforward way with tables of Weyl group degrees gives the following information about the the centers of the simply connected almost simple compact Lie groups (x6): the center of An has order (n + 1), the centers of Bn and Cn are cyclic of order 2, the center of D2n is (Z=2)2, the center of D2n+1 is of order 4, the centers of G2, F4 and E8 are trivial, the center of E6 is Z=3, and the center of E7 is Z=2. Of course in some cases it is possible to obtain these results and more by inspection (for instance it is elementary to check that the center of An ~=SU n+1 is Z=(n + 1)), but in other cases the inspection is not so easy. 1.13. Fundamental groups. Suppose that G is a connected compact Lie group with finite fundamental group. In this case the universal cover of G is a simply connected almost simple compact Lie group K, and G is isomorphic to K=A for some finite subgroup A of ZK . In particular, CENTERS 5 ss1(G) ~=A is isomorphic to a subgroup of ZK . Since TG ~= TK =A, it is easy to see that `G = `K , and that WG is conjugate to WK as a subgroup of GL `(Q). It follows that Theorem 1.7 gives bounds for #(ss1G) in terms of determinants involving generalized Coxeter elements of WG ; we leave it to the reader to write this out. Similarly, Theorem 1.8 guarantees that if K is almost simple and the center of WG is nontrivial, then ss1G is isomorphic to (Z=2)s for s `G . 1.14. Generalizing to p-compact groups. A p-compact group X is a homotopy theoretic object which has the essential homotopy theo- retic structure of the p-completion of a compact Lie group; see [10 ], [17 ], [18 ], or i[8]. Technically, a p-compact group is a pair (X; BX), where X is a p-complete space with finite mod p homology and B X is a classify- ing space for X (that is, the loop space BX is equivalent to X). For technical reasons it is also convenient to require that ss0(X) ~=ss1(B X) be a finite p-group. Like a compact Lie group, a p-compact group X has a center ZX (which is another p-compact group), a fundamental group ss1(X), a rank ` = `X , a maximal (p-complete) torus TX , and a Weyl group W = WX . The group W acts on TX by a kind of conjugation. Let LX = ss1TX and VX = Q LX . Then LX ~= (Zp)` and VX ~= (Qp)`, where Zp is the ring of p-adic integers and Qp is its quotient field. The action of W on TX induces a homomorphism W ! Aut (LX ) ! Aut (VX ) : If X is connected this homomorphism is injective and its image is a (generalized) reflection subgroup of Aut (VX ) ~= GL `(Qp), i.e., a finite subgroup generated by the (generalized) reflections it contains (x2). Say that X is almost simple if VX affords an irreducible representation of W . Then any connected p-compact groupQX has a finite cover X" which can be written as a product T x iYi, where T is a p-complete torus and each Yiis a simply connected almost simple p-compact group. Moreover, X is obtained in a certain sense as a quotient X"=A, where A is a finite subgroup of the center of X". We give versions of the above results for p-compact groups; these versions are somewhat stronger if p is odd and weaker if p = 2. In any case they lead to a proof of the following theorem. Say that X is of Lie type G, where G is a connected compact Lie group, if WX GL `(Qp) is obtained up to conjugacy by taking the image of WG GL `(Q) under the usual inclusion GL `(Q) GL `(Qp). 1.15. Theorem. If X is a simply connected almost simple p-compact group which is not of Lie type, then ZX is trivial. 6 W. G. DWYER AND C. W. WILKERSON In view of the product decomposition theorem [12 ] for p-compact groups, this gives the following splitting result. 1.16. Theorem.Q Any connected p-compact group can be written as a product X x iYi, where X is a p-compact group of Lie type, and each Yi is a simply connected almost simple p-compact group which is not of Lie type. 1.17. Organization of the paper. Section 2 discusses (generalized) reflection groups and their centers, and in particular indicates how to determine the center of an irreducible reflection group from its list of degrees. Section 3 describes the key properties of Coxeter elements, and x4 moves on to generalized Coxeter elements. Section 5 has proofs of 1.4-1.10, and the next section draws conclusions about centers of particular groups. Section 7 extends our results to p-compact groups, while x8 presents the information about p-adic reflection groups which is combined with these extended results in x9 to prove 1.15 and 1.16. The authors wish to thank James E. Humphreys, Bertram Kostant, Robert Steinberg, Louis Solomon, and T. A. Springer for interesting correspondence and conversations about Coxeter elements. Individ- ually and collectively they provided answers to our sometimes naive questions and conjectures. 2. Reflection groups and their centers In this section we recall some basic properties of finite reflection groups over fields of characteristic zero, and describe how the center of an irreducible reflection group can be determined by inspecting a finite collection of integers called the set of degrees of the reflection group. 2.1. Reflection groups. If k is a field of characteristic zero, an ele- ment w in GL `(k) of finite order is said to be a (generalized) reflection if w is conjugate to a diagonal matrix which differs from the identity matrix in only one position. A finite subgroup W of GL `(k) is said to be a reflection subgroup if it is generated by the reflections it contains. Let V = k`. The reflection sugroup is essential if V W = 0, and irre- ducible if W is nontrivial and V affords an irreducible representation of W . The number ` is called the rank of W . 2.2. Remark. If R is an integral domain with quotient field k, an el- ement of GL `(R) is said to be a reflection if it is a reflection when it is considered as an element of GL `(k); similarly, a finite subgroup of GL `(R) is said to be an (essential, irreducible) reflection subgroup if it has the corresponding properties as a subgroup of GL `(k). CENTERS 7 If G is a connected compact Lie group of rank `, the Weyl group WG can be viewed as a finite reflection subgroup of GL `(Z); WG is essential if and only if the fundamental group (equivalently, the center) of G is finite, and irreducible if and only if in addition the universal cover of G is compact and almost simple (equivalently, if and only if the universal cover of G cannot be split in a nontrivial way as a product of compact Lie groups). If X is a p-compact group of rank `, the Weyl group WX can be viewed as a finite reflection subgroup of GL `(Zp). Again, WX is essen- tial if and only if the fundamental group (equivalently, the center) of X is a finite p-group, and irreducible if and only if the universal cover of X cannot be split in a nontrivial way as a product of p-compact groups. 2.3. Remark. If W is an essential finite reflectionQsubgroupQof Aut (V ), then there are product decompositions V ~= jVj and W ~= j Wj, such that Wj acts as an irreducible reflection group on Vj and Wj acts trivially on Vi for i 6= j. We are particularly interested in finite reflec- tion groups over Q (Weyl groups of connected compact Lie groups), over R (Coxeter groups), over C (complex reflection groups), and over Qp (Weyl groups of p-compact groups). A convenient source of infor- mation on irreducible reflection groups is [5]. 2.4. Invariants and degrees. Suppose that V = k` and that W is a finite reflection subgroup of Aut (V ). Let V # be the k-dual of V . The action of W on V induces an action of W on V # and hence an action on the symmetric algebra Sym (V #). We grade this symmetric algebra so that V # itself is in degree one; the action of W then preserves the grading. It is a classical fact that the ring of invariants (Sym (V #))W is itself a graded polynomial algebra over k with ` generators [4] [19 ] [15 ]. The collection {d1; : :;:d`} of degrees of a set of homogeneous polynomial generators is well-defined, and is called the set of degrees of W . The numbers mj = dj- 1 are referred to as the exponents of W . 2.5. Remark. The degrees of the irreducible finite reflection groups of various sorts are listed in [5]. We take these degrees as the basic nu- merical invariants of a reflection group. If R is an integral domain of characteristic zero, the degrees of a reflection group over R are com- puted by first extending scalars to the quotient field of R and then determining the degrees as above. 2.6. Centers. If W is a finite reflection group with degrees {dj}, we let ffi(W ) denote gcd{dj}. The following theorem determines the center of an irreducible finite irreducible reflection group W in terms of ffi(W ). 8 W. G. DWYER AND C. W. WILKERSON 2.7. Theorem. Suppose that k is a field of characteristic zero, and that W is an irreducible finite reflection group of rank ` over k. Let ffi = ffi(W ). Then the center ZW is a cyclic group of order ffi, and acts on k` by scalar diagonal matrices. In particular, k contains the primitive roots of unity of order ffi, and any generator of ZW acts on k` by multiplication by such a primitive root. 2.8. Remark. If k is the field of complex numbers, there is a convenient geometric interpretation of the center of the reflection group as the ker- nel of the homomorphism from W to the associated collineation group W 0acting on the corresponding complex projective space. This con- nection is exploited in the Shephard-Todd classification scheme, [19 ]. The proof of 2.7 proceeds with a series of lemmas. Recall that an irreducible representation of a finite group W over a field k is said to be absolutely irreducible if the representation remains irreducible after extending scalars to the algebraic closure of k. 2.9. Lemma. Suppose that k is a field of characteristic 0, and that V is a finite dimensional vector space over k which affords an absolutely irreducible representation of the finite group W . Then the center of W is cyclic and and acts on V by scalar diagonal matrices. Proof. Extend scalars to obtain V = k k V , where k is the algebraic closure of k. Action by an element of the center of W on V is a W -map and hence by Schur's lemma a scalar unit. So ZW is embedded in the units of k, and hence, being finite, is cyclic. Let w be a generator. In Aut(V ), w = . I, for some in k. But then trace(c) = dim (V ) 2 k, __ not just in k. Since char(k) = 0, trace(c)= dim(V ) = 2 k also. |__| 2.10. Lemma. Let k be a field of characteristic zero and W GL `(k) = Aut (V ) an irreducible finite reflection group. Then the action of W on V affords an absolutely irreducible representation of W over k. Proof. Product decompositions of an essential reflection representation correspond to decompositions of the reflection group itself into products of subgroups generated by reflections (2.3, [12 , 7.1]). Since the action of W on V is irreducible, W cannot be written as a product of subgroups generated by reflections. Conversely, since W cannot be written as such a product group, the induced reflection representation of W on k k V is irreducible. |___| 2.11. Lemma. Let k be a field of characteristic zero and V a finite dimensional k-vector space. Denote by Sym (V #) the algebra of poly- nomial functions on V . Suppose that W is a finite subgroup of Aut(V ) which acts irreducibly on V . If Sym (V #)W is concentrated in degrees CENTERS 9 divisible by ffi, then W contains a scalar diagonal matrix . I, where is a primitive root of unity in k of order ffi. Proof. This is similar to an argument of Humphreys, [15 , p. 82]. Tem- porarily replacing k by a larger field, we assume that k contains the ffi'th roots of unity. Let S be the fraction field of Sym (V #) and K that of Sym (V #)W . Then K ! S is a Galois field extension with Galois group W . Given an ffi'th root of unity 2 k, define a field automorphism OE of S by setting OE(v) = v for v 2 V # and then extending multiplicatively to Sym (V #) and the fraction field. Since Sym (V #)W is concentrated in degrees divisible by ffi, and OE(xN ) = N x = (ffi)N=ffix = x if ffi|N, we have that Sym (V #)W is fixed by OE. Since K ! S is a Galois field extension, and hence normal, OE must be in the Galois group of S over K, namely W . Hence W contains the scalar diagonal matrix . I and the element is in the original field. Notice that . I is a central__ element in W . |__| __ Proof of 2.7.This is a combination of the previous three lemmas. |__| 2.12. Cohomology bounds. Theorem 2.7 has an interesting impli- cation for Weyl group cohomology. 2.13. Theorem. Suppose that R = Z or R = Zp, and that W is an irreducible reflection subgroup of GL `(R) = Aut (L). Let ffi = ffi(W ). Then R contains a primitive root of unity of order ffi, and for M = L or M = (Q L)=L, (1 - ) . H*(W ; M) = 0 : 2.14. Example. We apply this theorem in the following way. If R = Z or R = Z2, then the group of roots of unity in R is cyclic of order 2 and ffi(W ) is either 1 or 2. In the first case 2.13 gives no information; in the second case = -1, (1 - ) = 2, and the conclusion is that H *(W ; M) has exponent 2. If R = Zp for p odd, then the group of roots of unity in R is cyclic of order (p - 1). If ffi(W ) = 1 then 2.13 gives no information, otherwise is a nontrivial primitive root of unity, (1 - ) is a unit in R, and the conclusion is that H *(W ; M) = 0. 2.15. Remark. If W is the Weyl group of a connected compact Lie group G, then the cohomology group H 3(W ; LG ) is the home of the k-invariant or extension class that determines the group extension 1 ! TG ! NG (TG ) ! W ! 1 : Thus 2.13 and 2.14 enable us to recover the result of Tits [24 ] that this k-invariant has order 2, although only for the cases (1.9) in which 10 W. G. DWYER AND C. W. WILKERSON G has type Bn, Cn, D2n, G2, F4, E7 or E8. Related results for p- compact groups can be obtained by the same method. See [1] for a comprehensive statement. Proof of 2.13. By 2.7, the quotient field of R contains the required primitive root of unity ; since R is integrally closed in its field of fractions, 2 R. Again by 2.7, there is an element w 2 ZW which acts on L by multiplication by . For M = L or M = (Q L)=L, write H *(W ; M) = Ext *Z[W](Z; M) : The element w acts on Z by the identity, and on M by multiplication by . These actions commute with the action of W , and so give two actions of w on H *(W ; M). By a naturality argument, these actions __ must be the same. See for example [14 ]. The result follows. |__| 3. Coxeter elements In this section we study Coxeter elements in real reflection groups; these groups are also known as Coxeter groups. Any integral or rational reflection group (e.g., the Weyl group of a connnected compact Lie group) can be treated as a Coxeter group via the inclusion Q R. A finite Coxeter group W is a reflection subgroup of GL `(R). One can assume that the matrices in W belong to the orthogonal group, but this is not always convenient or necessary. Let V denote R`. If the action of W is essential, then it is possible to find ` distinct reflections {s1; : :;:s`} in W which generate W . The product wCox = s1s2 : :s:`is then called a Coxeter element. Note that wCox depends on the choice of the reflections {s1; : :;:s`} and on the order in which these reflections are written down in the product. These Coxeter elements have amazing properties, some of which will come up in this section. We are mainly interested in proving the follow- ing theorem (which relates Coxeter elements to subreflection groups) and in recalling the two subsequent statements (which in particular give information about the eigenvalues of a Coxeter element). 3.1. Theorem. (R. Steinberg, E-mail) Suppose that W is a finite ir- reducible Coxeter group. Then no Coxeter element of W is contained in any proper reflection subgroup of W . 3.2. Theorem. [15 , 3.17-19] Suppose that W is a finite essential Cox- eter group. 1. The various Coxeter elements wCox 2 W are all conjugate in W . 2. V = {0}. That is, 1 is not an eigenvalue of wCox. CENTERS 11 3. If W is irreducible and order(wCox) = h, then wCox has a one- dimensional eigenspace in C V for the eigenvalue j = e2ssi=h. 4. If W is irreducible, and v 2 C V is a nonzero wCox-eigenvector for j, then no reflection in W carries v to itself. If W is a finitePCoxeter groupPwith degrees {dj} and exponents {mj}, then the sum j(2dj - 1) = j(2mj + 1) is called the dimension dim (W ) of W . There is some fascinating numerology involving these quantities. 3.3. Theorem. [15 , x3] Let W be a finite Coxeter group with degrees {d1; : :;:d`} and exponents {m1; : :;:m`}. Let h denote the order of a Coxeter element wCox 2 W , and let j = e2ssi=h. 1. |W | = d1d2 : :d:n. P 2. W contains exactly N = j mj reflections. 3. If W is the Weyl group of a Lie group G with finite fundamental group, then dim(G) = dim (W ). P 4. If W is irreducible, then h = (2 jmj)=` = max {dj}. 5. If W is irreducible, then wCox has eigenvaluesQ{jmj }. 6. If W is irreducible, then | det(1 - wCox)| = j(1 - jmj ). Remark. Note that if W is a finite Coxeter group of rank `, the number of reflections in W is (dim (W ) - `)=2. This implies that if W 0 W is a subreflection group with dim (W 0) = dim (W ), then one must have W 0= W , since the number of reflections in each group is the same. 3.4. Remark. If W is not irreducible, but V W = 0, then V is a direct sum of irreducible W -spaces {Vi}, and W splits as a product of subre- flection groups {Wi} such that Wi acts irreducibly on Vi (2.3). In this case, the Coxeter element for W is the product of the Coxeter elements of the Wi. We leave it to the reader to extend the last three statements in 3.3 to cover this more general case. 3.5. Example. If W is the Weyl group of type An, then W = n+1 and the reflection representation is the reduced standard representation.P The set of exponents is {1; 2; : :;:n}, so that h = 2( ii)=n = (n + 1). The Coxeter element wCox can be taken to be the standard (n + 1)- cycle in n+1. 3.6. Example. If W is the Weyl group of E8, the set of exponents is {1; 7; 11; 13; 17; 19; 23; 29}, so h = 30. In general, W can have many generalized Coxeter elements, i.e., elements w with the property that det(I - w) 6= 0. The following theorem of Carter suggests that these arise as the Coxeter elements 12 W. G. DWYER AND C. W. WILKERSON for maximal rank subreflection groups. Carter [2] proves this if W is a rational reflection group. It is possible to extend this to arbitrary Coxeter groups by checking the additional irreducible real cases H3, H4, and I2(m) [15 , x2], which of course do not correspond to compact Lie groups. We state this extended theorem. 3.7. Theorem. Let W be a finite essential Coxeter group of rank `. 1. [2] Any w 2 W can be written as a product of ` or fewer reflections in W . The minimal number required is n - dim(V ). 2. An element w 2 W has det(1 - w) 6= 0 if and only if there exist a set of ` reflections {s1; : :;:s`} in W such that w = s1 : :s:`and the root spaces {image (I - sj)} span V . 3. if det(1 - w) 6= 0, then w is a Coxeter element for the subgroup W 0of W generated by the {sj} from (2). Proof. For the "if" part of (2), note that because each sj is a reflection, each subspace im (I - sj) has dimension 1, generated by some vector pj. Under the given assumptions on {sj}, the set {pj} forms a basis for V and sj(pj) = -pj. Now the argument is the same as that of Carter, [3, 10.5.6]. For the "only if" direction of (2), observe that by (1) the element w is the product of at most ` reflections. If the associated (one-dimensional) root spaces fail to span V , it is easy to argue that V 6= 0, which is assumed untrue. Therefore there must be exactly ` reflections in the product decomposition of w, and their root spaces must span V . Part (4) is immediate from the definition of Coxeter __ element. |__| Proof of 3.1.We thank R. Steinberg for his help with this argument. Let w be a Coxeter element of W of order h, and let j = e2ssi=h. By (3.2)(4), the j-eigenspace of w acting on C V is spanned by a vector v which is not fixed by any reflection in W . Suppose that w 2 W 0 for some proper reflection subgroup. Suppose further to begin with that W 0is a non-trivial product of reflection groups, W = W1 x W2, acting on V ~= V1 x V2. Then in C V , v = (v1; v2) and w(v) = (w(v1); w(v2)) = jv = (jv1; jv2). Thus both v1 and v2 are in the j- eigenspace for j. Hence one is zero, say v2. But W2 acts trivially on V1, so the reflections in W2 fix v = (v1; 0). Hence W2 must be the trivial group. We have concluded that W 0is irreducible, and we must now show that W 0= W . By 3.7(2), w is the product of ` reflections {sff} in W 0with rootspaces spanning V . If W 00is the subreflection group generated by the {sff}, then by 3.7(3) the element w is a Coxeter element for W 00. One reasons as before that W 00cannot be written as a nontrivial product. So W 00must be irreducible and w is a Coxeter CENTERS 13 element for it. Hence dim (W 00) = (h + 1)` = dim (W ), so W = W 0= __ W 00(see 3). |__| 4. Generalized Coxeter elements We move on to consider a weaker notion of Coxeter element which is useful in dealing with reflection groups which are not necessarily real. 4.1. Definition. Let R be Z or Zp. If W is a finite subgroup of GL `(R) (in our applications a reflection subgroup), an element w 2 W is said to be a generalized Coxeter element if det(I - w) 6= 0. If R is as above and W is a subgroup of GL `(R), we let L be the module R` on which W acts, V the vector space Q L, and T the quotient module V =L. 4.2. Remark. In the case in which R = Z and W is the Weyl group of a connected compact Lie group of rank `, then T is naturally isomorphic as a W -module to the group of elements of finite order in TG . If R = Zp and W is the Weyl group of a connected p-compact group of rank `, then T is naturally isomorphic as a W -module to the p-discrete torus TX [10 ]. 4.3. Theorem. Suppose that R is Z or Zp, that W is a finite subgroup of GL `(R), and that w 2 W is a generalized Coxeter element of order k. Then the group T W is finite, and its order divides det(I - w) in R. Moreover k . T W = 0. Proof. Let C = , so that C has order k. The definition of a gen- eralized Coxeter element implies that V C = 0, so from the long exact group cohomology sequence for the coefficients 0 ! L ! V ! T ! 0, it follows that T C = H 0(C; T) ~= H 1(C; L). Hence these groups are finite and have exponent at most k. Since T W T C, T W is also finite, and k . T W = 0. Let = 1 + w + w2 + . .+.wk-1 2 Z[C]. Then . L . V V C = 0, so that H 1(C; L) = ker( : L ! L)=(I - w)L = L=(I - w)L : By the theory of elementary divisors the quotient group on the right has order | det(I - w)| if R = Z case and has order given by the p-th __ part of this if R = Zp. |__| 4.4. Example. In the Weyl group of E8, the Coxeter element wCox has order 30 (3.6). Since the exponents {mj} are each relatively prime to 30, the 15-th power and the 10-th power of the Coxeter element are generalized Coxeter elements of order 2 and 3, respectively. By 4.3, 14 W. G. DWYER AND C. W. WILKERSON H 0(W ; T) is annihilated by 2 and 3. Hence this group is zero. One can also calculate from 3.3(5) that | det(1 - wCox)| = 1. 5. Centers of Lie groups In this section we give various descriptions of the center of a con- nected simply connected compact Lie group, and then various order or exponent bounds on the centers of other compact Lie groups. 5.1. Simply connected compact groups. Suppose first that G is a connected compact Lie group. It is a classical fact that every element of G is contained inside some maximal torus. Since all maximal tori in G are conjugate, it follows immediately that the center of G is contained inside the center of the normalizer of TG . In other words, (5.2) ZG (TG )WG : This inclusion is not in general an equality. See [13 , x8] for a detailed description of ZG in terms of the group structure of NG (TG ); in general, the quotient (T W)=ZG is an elementary abelian 2-group. However, the inclusion 5.2 is an equality if G is simply connected (Theorem 1.4). The proof of this depends upon the following result of Borel. 5.3. Theorem. [22 ] [23 ] If G is a simply connected compact Lie group and g 2 G then the centralizer of g in G is connected. Proof of 1.4.As above, ZG T W. Suppose that t 2 T W . Consider the centralizer CG (t). By 5.3, CG (t) is connected. It is clear that T is a maximal torus for CG (t) and that the Weyl group WCG(t) is the subgroup of W that fixes t. By the choice of t, then, WCG(t) = W . The dimension of G is determined by W and the dimension of CG (t) by WCG(t)(see 3.3(3)). So dim (CG (t)) = dim (G). Since G and CG (t) are compact connected manifolds, we must have G = CG (t). Hence __ t 2 ZG . This is true for all t 2 T W, so CG = T W. |__| The Weyl group of a connected compact Lie group G is a finite Coxeter group, and so has a unique conjugacy class of elements called Coxeter elements (see x3). From 5.2 it is immediately clear that for any Coxeter element wCox 2 WG there is an inclusion (5.4) ZG (TG ): It is somewhat surprising that even this inclusion is an equality if G is simply connected (Theorem 1.6). Proof of 1.6.Let t 2 TG be fixed by wCox. The centralizer CG (t) is connected and hence its Weyl group is a subreflection group of the CENTERS 15 Weyl group W of G. Since t is fixed by wCox, certainly wCox 2 WCG(t). By 3.1, WCG(t)= W . As in the above proof of 1.4, this gives CG (t) = G. Let L = LG , T = TG , and C = . It is easy to see that T ~= (R L)=L as a module over W . Since wCox is a Coxeter element, (R L)C = 0, so a long exact cohomology sequence (cf. proof of 4.3) gives an isomorphism T C ~=H 1(C; L) : The order of the group on the right is calculated in the proof of 4.3 as_ given by | det(I - wCox)|. |__| 5.5. Remark. Theorem 1.6 gives a simple algorithm for calculating the order of the center of a simply connected almost simple compact Lie group G. It consists of the following steps. 1. Determine the exponents (2.4) of WG by consulting a table. 2. Combine this list of exponents with 3.3(5) to obtain the eigenval- ues of the action of a Coxeter element wCox 2 WG on VG . 3. Use 1.6 to calculate #(ZG ) = | det(1 - wCox)|. A little Galois theory shows that since wCox is represented by a rational matrix in the reflection representation of WG , the power p of a prime p which divides | det(1 - wCox)| can be determined as follows. Let {mj} bePthe set of exponents for WG , let ` be the rank of G, and let h = (2 mj)=` be the order of wCox. Let cibe the number of exponents mj such that the denominatorPof mj=h (in lowest terms) is equal to pi. Then p = (p=(p - 1))c1 + i>1ci. 5.6. Other connected groups. We now prove some general results about centers which are simpler to work with than the somewhat cum- bersome formula from [13 , S8]. Proof of 1.7.The inclusion 5.2 is valid for any connected compact Lie group G, as is the analogue of 5.4, with wCox replaced by any element __ of WG . Thus the second half of the above proof of 1.6 also gives 1.7. |__| Proof of 1.8.By 2.14, the center of WG is of order 2, generated by an element w which acts on L = LG by (-1) and hence on TG = (RL)=L by inversion. Then det(1 - w) = 2`, so that w is a generalized Coxeter element. The center of G is a subgroup of (TG ) ~=(Z=2)` and so is __ of the indicated form. |__| Proof of 1.10. We will use the notation of [13 , 8.1]. It follows from [13 , 8.2] that the ZG = (TG )W if for each reflection s in WG there is an equality oe(s) = F (s). As is clear from [13 , 8.7], it is enough to 16 W. G. DWYER AND C. W. WILKERSON show that H(s) = F (s). In the notation of [13 , 8.7], consider the exact sequence {0} ! H(s) ! T ! C ! {0} and the induced exact sequence of elements of exponent 2 {0} ! 2H(s) ! 2T ! 2C ! {0} ; where 2C ~= Z=2 because C is a circle group. (This induced sequence is exact because H(s), as a torus, is divisible and hence injective as an abelian group.) By assumption the action of s on 2T ~= Z=2 ss1(T ) is nontrivial, and by construction the action of s on 2H(s) is trivial. It follows immediately that (2T ) = 2H(s). Now because of the exact sequence displayed in [13 , 8.7] and the fact that H(s) is injective, F (s) = T is isomorphic as an abelian group either to H(s) or to H(s) Z=2. Since we have shown that 2F (s) = 2H(s), it must __ be that F (s) = H(s). |__| 6. Centers of specific groups In this section we apply the results of the previous section to obtain information about the Weyl groups of the simply connected almost simple compact Lie groups. We assume that for these groups the list of exponents of the corresponding Weyl group is known. We will work out the case An in more detail than the others in order to illustrate the kinds of calculations that come up. The Weyl group is the symmetric group n+1, its exponents are {1; 2; 3; : :;:n}, the order of wCox is h = n + 1, and the eigenvalues of wCox are e2ssij=(n+1), j = 1; : :;:n (3.3). These eigenvalues are exactly the roots of the polynomial Yn n+1 1 - x 2 n x - e2ssij=(n+1)= _________= 1 + x + x + . .+.x : j=1 1 - x The determinant det(1 - wCox) is then the value of this polynomial at x = 1, namely, (n + 1). This implies that the order of the center of An is (n + 1). In fact this center is a cyclic group (An = SU n+1) but our numerical methods don't give this directly. A similar calculation shows that for the Weyl groups of G2, F4, and E8, | det(I - wCox)| = 1. Hence these Lie groups have trivial center. The corresponding determinants for E6 and E7 have absolute values 3 and 2, respectively. Thus the center of E6 is cyclic of order 3, and the center of E7 is cyclic of order 2. For Bn and Cn the determinants have absolute value 2 and so these groups have center Z=2. CENTERS 17 The groups Dn are left; the determinant det(1 - wCox) has absolute value 4 in all cases. The degrees associated to the Weyl group of D2k are even, so this Weyl group contains a central element of order 2 (2.7) and the center of D2k is an elementary abelian 2-group (1.8). Since this center has order 4, we conclude that for n even the center of Dn is (Z=2)2. Our numerical methods do not determine the group structure of center of D2k+1; for that purpose it is necessary to look more closely at the explicit reflection representation, or to examine the group Spin4k+2. It turns out that the center of D2k+1 is isomorphic to Z=4. 7. Centers of p-compact groups Suppose that X is a p-compact group. Associated to the maximal torus TX and the center ZX are certain discrete groups TX and ZX called the discrete torus and discrete center respectively. The group TX is isomorphic to (Zp1 )`X, while ZX TX is isomorphic to the product of a finite abelian p-group and a group of the form (Zp1 )k, where k `X and k = 0 if ss1(X) is finite. The group ZX effectively determines the p-compact group ZX , since B ZX is the p-completion of BZX [11 ]. In this section we write down various estimates for the discrete center of a p-compact group X. These are somewhat different from the Lie group estimates in x5, for two reasons: first of all, we do not have a version of Theorem 5.3 for p-compact groups; secondly, we do not have a theory of Coxeter elements for non-rational p-adic reflection groups. Suppose that X is a connected p-compact group. The Weyl group WX acts on the discrete torus TX ; in fact, TX is isomorphic to VX =LX as a WX -module. By [11 ] there is an inclusion ZX (TX )WX analogous to 5.2. 7.1. Theorem. Suppose that X is a connected p-compact group. If p is odd then ZX = (TX )WX . If p = 2 and reduction mod 2 gives a mononomorphism (7.2) WX ! GL `(Z2) ! GL `(Z=2) then the same conclusion holds. 7.3. Remark. The first statement of 7.1 parallel to 1.4 but a bit stronger in that it does not require simple connectivity. The second statement is parallel to 1.10. Since every finite subgroup of the kernel of the reduc- tion map GL `(Z2) ! GL `(Z=2) is an elementary abelian 2-group, the 18 W. G. DWYER AND C. W. WILKERSON conclusion of 7.1 holds for any connected 2-compact group X with the property that the Weyl group of X has no normal elementary abelian 2-subgroups. Proof of 7.1.The statement for p odd is in the remark after [11 , 7.7]. The second statement is proved by the arguments in the proof of 1.10, __ but with references to [13 , x8] replaced by references to [11 , x7]. |__| 7.4. Proposition. Suppose that X is a connected p-compact group with a generalized Coxeter element w 2 WX . Then ZX is finite, and #(ZX ) divides | det(1 - w)| in Zp. If the order of w is prime to p, then ZX is trivial. __ Proof. This is very similar to the proof of 1.7. |__| 7.5. Proposition. Suppose that X is a connected p-compact group of rank ` such that WX is an irreducible reflection group with a nontrivial center. Then if p is odd, ZX is trivial, while if p = 2, ZX is isomorphic to (Z=2)s for some s `. Proof. This is very similar to the proof of 1.8. The difference between_ p = 2 and p odd is discussed in 2.14. |__| 8. Irreducible p-adic reflection groups The Weyl group of a connected p-compact group is naturally pre- sented as a finite p-adic reflection group. Any such group is a product of finite irreducible p-adic reflection groups, and the list of such fi- nite irreducible p-adic reflection groups has been known for some time. This list (which depends on p) was given by Clark and Ewing [5]; they started with the list of irreducible finite complex reflection groups compiled by Shephard and Todd [19 ] and then determined, for each p, which of the groups in the Shephard-Todd list have realizations over the p-adic integers. We are motivated by 7.4 to search for generalized Coxeter elements in Weyl groups of p-compact groups. In the case that the reflection group can be generated by exactly n reflections, 5.4 [19 ] verifies that there is a choice of product of the generating reflections with properi- ties similar to the classical Coxeter elements. Unfortunately, there is a rather substantial list of complex reflection groups which are not gener- ated by n reflections. On the other hand, the theorem of Solomon [20 ] gives the existence of a large number of elements with no eigenvalue +1 if the action is essential. The regular elements of Springer [21 ] are a related generalization of Coxeter elements. CENTERS 19 However, for our needs, it is most convenient to analyse directly four cases : 1. Rational reflection groups (Weyl groups of Lie groups). 2. Sporadic non-rational reflection groups over Qp, p odd. 3. Infinite families of non-rational reflection group over Qp, p odd. 4. Non-rational reflection groups over Q2. Clark and Ewing list these groups in these various classes and their degrees (which were determined by Shephard and Todd). Hardly any of our work directly depends on the field of definition. 8.1. Rational reflection groups. These were already treated in x6. We give the list of ordered pairs (G; d) where G is the standard desig- nation for the a compact simply connected almost simple compact Lie group and d = | det(1 - wCox)|: (An; n + 1), (Bn; 2), (Cn; 2), (Dn; 4), (G2; 1), (F4; 1), (E6; 3), (E7; 2), (E8; 1). Note that the Weyl groups of Bn and Cn are isomorphic as rational reflection groups. 8.2. Sporadic non-rational reflection groups. If W is one of these reflection groups, then W contains either a central element of order 2, or in the case of Type 25 [5] a central element of order 3. This follows from glancing over the degrees of these groups and applying 2.7. One can also calculate the order of ZW as the ratio of the columns labeled g and g0 in Table VII, page 302 of [19 ]. Note that if the prime p > 2 divides the order of ZW , then the representation is not realizable over Qp. 8.3. Infinite familes of non-rational reflection groups. The cal- culations are along the lines of [1, x3] but because of the nature of what has to be proved, they are simpler. There are three infinite fam- ilies which contain non-rational complex reflection groups: Family 2a, Family 2b, and Family 3. We treat these families in reverse order. If W belongs to Family 3 and is non-rational, then #(W ) = m > 2 and W is cyclic. Any generator of W is a central generalized Coxeter element w of order m, with det(1 - w) = 1 - m. If W is realizable over Qp with p odd, then m divides (p - 1), and m is prime to p. If W belongs to Family 2b, then W is a dihedral group of order 2m, and contains a generalized Coxeter element w of order m. If W is non-rational and realizable over Qp, then m > 1 and m divides either (p - 1) or (p + 1) and hence, since p > 2, m is relatively prime to p. If W belongs to Family 2a and has rank n, then W is one of the groups G(m; r; n) of Shephard-Todd. Here r must divide m. Given a 20 W. G. DWYER AND C. W. WILKERSON basis {ej} of C` and = e2ssi=m, G(m; r; n) is generated by thePoper- ations ei 7! i eoe(i), oe 2 n, subject to the constraint that ii = 0 mod r. If m > 1, the action is essential. G(m; r; n) has degrees {m; : :;:(n - 1)m; (m=r)n} and order (m=r)mn-1n!. We show below that in all cases for which m > 1, there is a generalized Coxeter element in G(m; r; n) with order dividing 2m. The order of the center is d = (m=r)gcd(r; n). If d > 1, the center is generated by the scalar diagonal matrix D = aIn, where a = m=d. Thus D is a generalized Coxeter element of order d dividing m. If d = 1 then r = m and gcd(r; n) = 1. Hence, there exists a number b with 0 < b < r so that bn = 1 mod r. If b > 1, define the non-scalar diagonal matrix M = (b; : :;:b; b-1) in G(m; r; n). The matrix M has no +1 eigenvalues and has order m. If b = 1, take M = (; : :;:; [0; ; 1; 0]), where the notation specifies a matrix which is diagonal except for a 2x2 block in the lower right-hand corner. Now M has no +1 eigenvalues and M has order 2m. For example, G(2; 2; 2n + 1) is the Weyl group of type D2n+1. Finally, if G(m; r; n) is realizable over Qp, then by [5] m is a divisor of (p - 1). Hence in all cases G(m; r; n) has a generalized Coxeter element of order prime to p. 8.4. Non-rational reflection groups over Q2. There is only one non- rational reflection group W over Q2. This is the reflection group of Type 24 on the Clark-Ewing list [5], i.e., the Weyl group of the 2- compact group DI(4) constructed in [9]. The group W is abstractly isomorphic to Z=2 x GL 3(Z=2). We show that W has a generalized Coxeter element w such that det(1 - w) is a unit in Z2. Let V = (Q2)3 be the vector space which affords the reflection representation of W . According to the constructions in [9], there is a 2-adic integral lattice L V such that L is preserved by W and such that the natural map W ! Aut (L) ! Aut (L=2L) ~=GL 3(Z=2) can be identified with the obvious surjection Z=2 x GL 3(Z=2) ! GL 3(Z=2) : Let M be a matrix in GL 3(Z=2) such that det(I - M) 6= 0, and let w be a lift of M to W . Then det(I - w) is a unit in Z2. 8.5. Remark. By using more information about the fields of definitions of the particular complex reflection groups, one can replace some of the theory of this section with calculations. For example, if the order of W is prime to p, then any p-compact group realizing W has trivial center. However, these considerations do not cover all the examples, e.g. the group denoted by Type 29 at the prime 5. CENTERS 21 9. Centers of specific p-compact groups The following result includes 1.15. 9.1. Theorem. Let X be a simply connected almost simple p-compact group with Weyl group W . If W is a rational reflection roup, i.e. X is of Lie type G for some simply connected almost simple compact Lie group G, then ZX is a finite abelian p-group whose order divides the order of the center of G . If W is a non-rational reflection group, then ZX is trivial. 9.2. Remark. Taken in combination with the list in 8.1, Theorem 9.1 implies that if p > 3 and X is a connected p-compact group with an irreducible Weyl group, then the center of X is trivial unless X is of Lie type An, i.e., WX = n. Proof of 9.1.If W is rational, the result follows from 1.6 and 7.4; the point is that any Coxeter element of W gives a formula for the order of the center of G and, as a generalized Coxeter element, a bound for the order of the center of X. Suppose that W is non-rational. If p > 2, then according to the calculations in x8, W contains a generalized Coxeter element of order prime to p. By 7.2 and 4.3, the center of X is trivial. If p = 2 then W is the Weyl group of DI(4), and W contains a generalized Coxeter element w such that det(1 - w) is a unit in Z2 (see 8.4). Again by 7.2 __ and 4.3, the center of X is trivial. |__| Proof of 1.16. This follows from the results in [12 ], [11 ],Qand [16 ]. As described in 1.14, X has a finite cover X" of the form T x iYi, where each Yi is simply connected and almost simple, and X = X"=A for some finite subgroup of ZX". Let R be the product of those Yi's of Lie type (those with rational Weyl groups) and let N be the product of the remaining Yi's. By 9.1, ZN = 0, so that A ZR and X = (T x R x N)=A = ((T x R)=A) x N : Since (T x R)=A is of Lie type, this gives the desired product decom-__ position of X. |__| References 1.Kasper K. S. Andersen, The normalizer splitting conjecture for p-compact groups, Fund. Math. 161 (1999), no. 1-2, 1-16, Algebraic topology (Kazimierz Dolny, 1997). 2.R. W. Carter, Conjugacy classes in the Weyl group, Compositio Math. 25 (1972), 1-59. 22 W. G. DWYER AND C. W. WILKERSON 3.Roger W. Carter, Simple groups of Lie type, John Wiley & Sons Inc., New York, 1989, Reprint of the 1972 original, A Wiley-Interscience Publication. 4.Claude Chevalley, Invariants of finite groups generated by reflections, Amer* *. J. Math. 77 (1955), 778-782. 5.Allan Clark and John Ewing, The realization of polynomial algebras as coho- mology rings, Pacific J. Math. 50 (1974), 425-434. 6.H. S. M. Coxeter, Discrete groups generated by reflections, Annals of Math. (1933). 7.______, The product of the generators of a finite group generated by reflect* *ions, Duke Math. J. 18 (1951), 765-782. 8.W. G. Dwyer, Lie groups and p-compact groups, Proceedings of the Interna- tional Congress of Mathematicians, Vol. II (Berlin, 1998), vol. 1998, pp. 43* *3-442 (electronic). 9.W. G. Dwyer and C. W. Wilkerson, A new finite loop space at the prime two, J. Amer. Math. Soc. 6 (1993), no. 1, 37-64. 10.______, Homotopy fixed-point methods for Lie groups and finite loop spaces, Ann. of Math. (2) 139 (1994), no. 2, 395-442. 11.______, The center of a p-compact group, The Cech centennial (Boston, MA, 1993), Amer. Math. Soc., Providence, RI, 1995, pp. 119-157. 12.______, Product splittings for p-compact groups, Fund. Math. 147 (1995), no.* * 3, 279-300. 13.______, The elementary geometric structure of compact Lie groups, Bull. Lon- don Math. Soc. 30 (1998), no. 4, 337-364. 14.William G. Dwyer, Vanishing homology over nilpotent groups, Proc. Amer. Math. Soc. 49 (1975), 8-12. 15.James E. Humphreys, Reflection groups and Coxeter groups, Cambridge Uni- versity Press, Cambridge, 1990. 16.J. M. Moller and D. Notbohm, Centers and finite coverings of finite loop spa* *ces, J. Reine Angew. Math. 456 (1994), 99-133. 17.Jesper M. Moller, Homotopy Lie groups, Bull. Amer. Math. Soc. (N.S.) 32 (1995), no. 4, 413-428. 18.D. Notbohm, Classifying spaces of compact Lie groups and finite loop spaces, Handbook of algebraic topology, North-Holland, Amsterdam, 1995, pp. 1049- 1094. 19.G. C. Shephard and J. A. Todd, Finite unitary reflection groups, Canadian J. Math. 6 (1954), 274-304. 20.Louis Solomon, Invariants of finite reflection groups, Nagoya Math. J. 22 (1963), 57-64. 21.T. A. Springer, Regular elements of finite reflection groups, Invent. Math. * *25 (1974), 159-198. 22.T. A. Springer and R. Steinberg, Conjugacy classes, Seminar on Algebraic Groups and Related Finite Groups (The Institute for Advanced Study, Prince- ton, N.J., 1968/69), Springer, Berlin, 1970, Lecture Notes in Mathematics, V* *ol. 131, pp. 167-266. 23.Robert Steinberg, Endomorphisms of linear algebraic groups, American Mathe- matical Society, Providence, R.I., 1968, Memoirs of the American Mathematical Society, No. 80. 24.J. Tits, Normalisateurs de tores. I. Groupes de Coxeteretendus, J. Algebra 4 (1966), 96-116. CENTERS 23 University of Notre Dame, Notre Dame, Indiana 46556 Purdue University, West Lafayette, Indiana 47907 E-mail address: dwyer.1@nd.edu E-mail address: wilker@math.purdue.edu