Inventiones math.

25, 159-198 (1974)

9 by Springer-Veriag1974

Regular Elements of Finite Reflection Groups

T.A. Springer (Utrecht)

Introduction
If G is a finite reflection group in a finite dimensional vector space V
then ve V is called regular if no nonidentity element of G fixes v. An
element g e G is regular if it has a regular eigenvector (a familiar example
of such an element is a Coxeter element in a Weyl group). The main
theme of this paper is the study of the properties and applications of the
regular elements. A review of the contents follows.
In w 1 we recall some known facts about the invariant theory of finite
linear groups. Then we discuss in w2 some, more or less familiar, facts
about finite reflection groups and their invariant theory. w3 deals with
the problem of finding the eigenvalues of the elements of a given finite
linear group G. It turns out that the explicit knowledge of the algebra R
of invariants of G implies a solution to this problem. If G is a reflection
group, R has a very simple structure, which enables one to obtain precise
results about the eigenvalues and their multiplicities (see 3.4). These
results are established by using some standard facts from algebraic
geometry. They can also be used to give a proof of the formula for the
order of finite Chevalley groups. We shall not go into this here. In the
case of eigenvalues of regular elements one can go further, this is discussed
in w4. One obtains, for example, a formula for the order of the centralizer
of a regular element (see 4.2) and a formula for the eigenvalues of a
regular element in an irreducible representation (see 4.5).
In w5 we give, using a case by case analysis, the regular elements of
the various irreducible Coxeter groups. The results ofw167 3, 4 are extended
in w to the "twisted" case, where certain outer automorphisms of
reflection groups come into play. This is used in w7 to establish properties
of "twisted Coxeter elements" of Weyl groups, partly via a case by case
discussion.
The main result of w8 is a reduction theorem (8.4) which can be used
to obtain properties of arbitrary elements of Weyi groups from those of
regular elements. We use it to give proofs of two results (8.5 and 8.7)
which so far were established only by using elaborate case by case check-
ings. In [10] Kostant has given a method to set up a connection between
the class of regular nilpotent elements of a complex semisimple Lie
160 T.A. Springer

algebra and the Coxeter class of the corresponding Weyl group. His
method is extended in w9 to obtain in a few other cases a connection
between a nilpotent class of a simple Lie algebra and a regular Weyl
group class. One can verify that among the nilpotent classes which occur
are the subregular ones of the exceptional Lie algebras (these subregular
classes were studied by Brieskorn, Steinberg and Tits, a brief report can
be found in Brieskorn's paper in the proceedings of the Nice congress).
Moreover two other cases occur in type E s. A closer study of these two
extra cases might be interesting.
In this paper we have not made a thorough study of regular elements
of complex reflection groups which are not Coxeter groups. That such
elements may be useful is shown by a result of A.M. Cohen, who in a
systematic study of the classification of irreducible complex reflection
groups has made use of theorems of w to obtain results of [15] about
degrees of the generators of the algebra of invariants.

The author is grateful to [Link], [Link] and [Link] for various

discussions about the topics of this paper.

1. Invariants of Finite Linear Groups

1.1. Let k be a field, let V be a finite dimensional vector space over k.
We put n = d i m V. Denote by S the symmetric algebra of the dual V'
of V. From the definition of S it follows that there exists an algebra
homomorphism ~t of S into the k-algebra of k-valued functions on V.
We call the elements of aS polynomial functions on E Ifk is infinite then
is injective. Below we shall mainly be interested in the case that k is the
field of complex numbers. Then, and also in the case of an infinite k, we
shall identify S with aS, and simply speak of S as the algebra of poly-
nomial functions on V.
The algebra S is (non-canonically) isomorphic to the polynomial
algebra k [ T 1. . . . . T~] in n indeterminates. There is a canonical grading
s = LI si,
i>o

corresponding to the usual grading of the polynomial algebra.

1.2. Let G be a group of nonsingular linear transformations of V. The
action of G on V carries over to an action on V', S and aS. I f f is a poly-
nomial function on V, then the action of g e G on f is given by
(g.f)(v)--f(g-t.v) (v~V).
We denote by R the subalgebra of S formed by the G-invariant elements,
i.e.
R = {s~SJg . s = s for all g e G } .
 Regular Elements of Finite Reflection Groups 161

We next recall without p r o o f some k n o w n results a b o u t the case of a

finite g r o u p G.
1.3. Proposition. Let G be finite.
(i) S is an R-module of finite type.
(ii) R is a k-algebra of finite type.
This is classical. (ii) goes back (at least) to E. Noether [13].
In the situation of 1.2, G acts on the set of prime ideals of S.
1.4. L e m m a . Let G be finite, let P and Q be two prime ideals orS. The
following properties are equivalent:
(i) There exists geG with g - P = Q .
(ii) P c ~ R = Q ~ R .
T h e following result is a consequence of 1.4.

1.5. Proposition. Let G be finite, let v, we V. The following properties

are equivalent:
(i) There exists g~G with g. v=w.
(ii) For all f e R we have (c~f)(v)=(~f)(w).
If we take k to be an algebraically closed field, then V can be viewed
as an affine algebraic variety over k and 1.5 then means that we can put
a structure of affine algebraic variety on the orbit space G \ V, whose
algebra is R.

2. Finite Reflection Groups, Recollections and Auxiliary Results

W e keep the notations of 1.1 and 1.2.
2.1. We say that a linear transformation of V is a reflection, if it is
diagonalizable and if all but one of its eigenvalues are equal to 1 (what
we call reflection is called pseudo-reflection in [4, p. 66]). G is called a
reflection group (in V) if G is generated by reflections. A reflection sub-
group of G is a s u b g r o u p of G which, viewed as a group of linear trans-
formations of V, is a reflection group.
T h e reflection group G is called a Coxeter group if the following
holds: (a) k is the field II~ of complex numbers and there is a structure
on V over the field IR of real numbers which is G-stable (this means that
there is a G-stable IR-subspace Vo of V, such that the canonical m a p
Vo(g~ll?---~ V is bijective), (b) G is generated by reflections of order 2.
Observe that (b) is a consequence of (a) if G is finite.
A finite Coxeter group G is called a Weft group if, with the a b o v e
notations, there is a lattice L in Vo which is stabilized by G. Then there is
a structure on V over the field Q of rational numbers which is G-stable.
162 T.A. Springer

From now on we shall assume the base field k to be infinite. As

announced in 1.1, we then identify S with the algebra of polynomial
functions on V, so that we shall drop the homomorphism cc In the sequel
the letter G will stand for a finite group, which most of the times is a
reflection group.
We next recall some basic results about the invariant theory of finite
reflection groups.

2.2. Theorem. Let G be a finite reflection group. I f char(k)=0 then R

is generated as a k-algebra by n = d i m V algebraically independent homo-
geneous polynomial functions f l .... , f,.

This is due to Chevalley [-7], see also I-4, p. 107].

The f being as in 2.2, let di=deg(fi) be the degree. The family
{d~ ..... d,} is then uniquely determined, i.e. is independent of the parti-
cular choice of the f~ (subject to the conditions of 2.2). This is also proved
in [4]. It also follows that the order IGI of G equals I=l di. We shall give
a different proof of the invariance statement, i=

2.3. Proposition. Assume that R is generated over k by n algebraically

independent homogeneous polynomial functions f : , ... , f,. Let d i =deg(f/)
and assume the ordering to be such that dl <=d2< "'" < d , .
(i) di is the minimal degree of a homogeneous element of R which is
a Igebraicatly independent of f l . . . . . fg- 1.
(ii) Let gl , .-., g, be n algebraically independent homogeneous elements
of R, let ei=deg(gi). Assume that el <e2 < . . . < e ~ . Then di <__ei ( l < i < n ) .
(iii) I f in the situation of (ii) we have di = ei (i <=i < n) then the gl generate
the k-algebra R.
Let f s R be homogeneous of degree d. There exists F ~ k [ T l .... , T,]
such that f = F ( f a . . . . . f~). The algebraic independence of the f then
implies that for any monomial Tf' ... T,e" which occurs in F with a non-
zero coefficient we must have e~ dl + e 2 d 2 + "'" + e , d , = d . Hence if ej=t=O
for some j > i, we have d>di. This implies (i). Observe that for i = 1 the
assertion of (i) means that dl is the smallest degree of a nonconstant
homogeneous element of R.
To prove (ii) we use induction on i. That dl < e~ follows from what we
just observed. Now assume that d i < e j (1 < j < i - 1 ) . There exist poly-
nomials F ~ k [-711. . . . . T.] such that g~ = Fj(fl . . . . . f.) (1 =j___ n). Because
of algebraic independence, the F~ with j < i cannot all involve the in-
determinates T~..... T~_~ only. Hence there is j < i such that Fj involves
an indeterminate Th with h>i. It follows that ej>=dh. Hence e~>ej>
d, > dg, which implies (ii).
 Regular Elements of Finite Reflection Groups 163

Assume that we have d~ = e i (1 =<i =<n). We then prove that there exist
polynomials Giek[T1 . . . . . T,] such that f~=Gi(gl . . . . . g,), which will
prove (iii). Assume that the existence of the G~has already been established
for all i with d~<d. We shall then prove the same for the i with d~=d
(the induction starts with d = 1). The F~ being as before we have, for any
monomial Tp ... Te" occurring in Fi with nonzero coefficient, that
el dl + . - - + e, d, = di. If di = d, then such monomials of degree > 1 involve
only Tj with dj<d. By our inductive assumption it follows that if di=d
there exist a~jek such that
gi = ~ aljf~+h~,
do,= d
where hiek[g I ..... g,] involves only g~ with dj<d. The algebraic in-
dependence of the & implies that the matrix (ao) is a non-singular square
matrix. This implies the existence of the G~ for all i with d~=d, finishing
the proof of 2.3.
Application of 2.3 in the situation of 2.2 shows that then the family
{dl .... , d,} is independent of the choice of the generators f~. We call the
integers d~ the degrees of the reflection group G. They are only defined
if char (k) = 0.
The next theorem is a slight refinement of a result of Shephard and
Todd [15, p. 288].
2.4. Theorem. Assume that k = 117,let G be a finite group. Assume that
.t"i. . . . . f, are n = dim V algebraically independent homogeneous elements
of R, let di = deg(f/).
(i) IGl~dld 2 ... dn.
(ii) If equality holds in (i) then G is a reflection group, R is generated
by the fi and the di are the degrees of G.
(iii) If the fl generate the k-algebra R, then equality holds in (i) and G
is a reflection group.
Let R~=R c~ S~ be the space of homogeneous invariants of degree i
and put a~=dim R i. We then have the following identity of formal
power series:
IGJ-1 E ( d e t ( 1 - g T))-I = ~ a, T' (1)
geO i=O
(see [4, p. 110-111]). It is clear that the power series

~ ai ti
i=0
converges for all real t with 0__<t< l.
Put bi = dim (k [f~, ..., f,] c~ &), then b~< ai. Moreover we have (see
[[Link]., p. 103])
b, V= (I (1-
i=0 i=1
164 T.A. Springer

It follows that

~ ai ti> [I (i --td') -1,

i=0 i=1

i f O < t < 1. Multiplying by (1 - t ) " and letting t tend to 1, we infer from (1)
that (i) holds.
If equality holds in (i), it follows from (1) that

IGI -~ ~ ( d e t ( 1 - t g ) ) - l >
g~l
I•I ( 1 - t d , ) - a - l G l - l ( 1 - )
i=1
t -n ,

if O<=t<l (observe the left-hand side is a real number). Multiplying

by ( l - t ) "-~ and letting t tend to 1, it follows from an argument of
Shephard-Todd (see [4, p. 111-112]) that the number of reflections in G
n

is at least ~ (d~- t). Let G' be the subgroup of G generated by the reflec-
i=i
tions contained in G. Let e~..... e, be the degrees of the reflection group G'.
By 2.3(ii) we may assume that e~<di ( l < i < n ) . Since the number of
reflections in G' is ~ (ei-1) [[Link]., p. 111], we find that
i=l
n

(ei - 1)>= ~" (di- 1).

i=l i=1

But then we must have ei=d~. Since G' is a reflection group, we have
[G'I= [I di [[Link]., p. 110-], hence G'= G. This establishes (i).
i=l n
If the f generate R, then the right-hand side of(l) equals ~ (1 - Td')-~.
i=1
Multiplying both sides of (1) by ( 1 - T ) " and putting T = I we obtain
equality in (i). Then (iii) follows from (ii).
2.5. In the remainder of this section we assume that G is a finite
reflection group over k = ~ , with degrees di. The integers pi=di-1
(1 <i<=n) are called the exponents of G. The result 2.6 to be discussed
now is a known one (although not stated in the literature). It was first
pointed out to me by Macdonald. Let I be the ideal in the algebra S
of polynomial functions generated by the invariants of strictly positive
degree, it is a G-stable graded ideal. Hence the quotient algebra S/I
is a graded G-module. By a theorem of Chevalley one knows that the
representation of G in S/I is equivalent to the regular representation
(see [7, p. 779]). Let Z be an irreducible character of G, i.e. the character
of an irreducible complex representation p of G. Then p occurs Z(1)
times in the representation of G in S/I.
 R e g u l a r E l e m e n t s o f Finite Reflection G r o u p s 165

The grading gives a decomposition

s/z = IA (s/I)t
t>O

into a finite number of G-stable subspaces. Let (Pj(X))l<j-<xm be the set

of degrees i such that p occurs in (S/I) i, each i occurring in this set with
a multiplicity equal to that of p in (S/l)t. The P~(X) are called the X-
exponents of G. If Z is the trivial character there is only one, viz. 0.
Define the polynomial s ~ 7Z[ T ] by
z(l)
fz(T) = ~ T p~'x~.
j=!

We then have the following result.

2.6. Proposition. Let G be a complex reflection group with degrees
di (1 <i<n). Let Z be an irreducible character of G. Then

IGI-t Z z(g)( det(1 - g T)) -t = f z ( T ) 1~ (1 - Te') -t. (2)

g~G i= 1

If Z is the trivial character, (2) reduces to a well-known formula which

follows from (1) (see [4, p. 111]). We use the notations of 2.5. Let A be a
graded G-stable subspace of S such that S is the direct sum of A and !
(the existence of A follows by the complete reducibility of the action of G
on the homogeneous components Si). One then proves, by a straight-
forward induction on degrees, that the product map of S induces a
bijection of G-modules A |162R ~ S. (3)

Now observe that by [[Link].], the left-hand side of (2) can be written
as a formal power series
~ at(z) T i,
i=0

where at(z) is the multiplicity of p in the representation of G in St. The

isomorphism (3) allows one to determine these multiplicities, leading
to (2). The argument is much the same as that used to deal with the
special case X= 1.
2.7. If one assumes that G acts irreducibly in V then one may take
for )~ the character of the corresponding representation of G. The left-
hand side of (2) can then be computed, by using a formula of Solomon
[16], which shows that we then have

L(T) = ~ T~'-',
t=l

so that we now recover the original exponents of G.

166 T.A. Springer

Another special case is that of a character Z of G of degree 1. Then

there is only one exponent PI(Z). The following result, which is an
immediate consequence of the isomorphism (3), shows how this exponent
can be found.
2.8. Corollary. Let Z be a character of degree 1 of G. Its exponent
Pl(Z) is the minimal degree of a nonzero homogeneous f ~S such that

g "f=z(g)f (geG).

Let 6 be the character of degree 1 of G defined by

6(g) = (det g).

If X is an irreducible character of G, then 6Z is also one, of the same

degree. Denote by 2 the complex conjugate of Z, so that ~,(g)=z(g-1).
We then have a connection between the exponents of Z and those of 6~.

2.9. Lemma. Let N = ~ Pl. The family of 62-exponents of G is

(N_pi(z))t<i<=z(t)" i=l

This follows by replacing T by T - i in (2).

2.9 implies that, for any Z, the x-exponents are =<N. It also follows
that a z-exponent equals N if and only if ;~=6. In that case there is a
welt-known polynomial f with the property of 2.8 (see [4, p. 113], our
notations are slightly different).

3. Eigenvalues of Elements of Linear Groups,

in Particular of Reflection Groups
3.1. We keep the notations of 1.1 and 1.2. We assume the field k to
be algebraically closed. We shall use in this section some facts from
algebraic geometry. In particular, we shall view V as an affine algebraic
variety over k.
Fix a finite set of homogeneous invariants f l . . . . . fm which generate
the k-algebra R (such a set exists by 1.3(ii)). We put d i = d e g f and we
assume, as we may, that all di are >0. I f k = C and G is a reflection group
we take f as in 2.2. In particular, we then have m = n.
Conversely, if k = ~ and if we have a set of generators f with m=n
then G is a reflection group. For if m = n the f~ must be algebraically
independent because of 1.3 (i), so that 2.4 (iii) applies.
We denote by Hi the algebraic hypersurface in V defined by f , i.e.

_~, = {re viA(v)=o}.

 Regular Elements of Finite Reflection Groups 167

Fix a root of unity ~ek*, of order d. Ifge G denote by V(g, () the eigenspace
of V corresponding to the eigenvalue ~, i.e.
v(g, O={ve Vlg. v=~v}.
We say that V(g, () is maximal if it is not property contained in another
V(h, O.
3.2. Proposition. (i) We have

U v(g, 0 = N u,.
g~ G d~d,

(ii) The irreducible components of the algebraic set n Hi are the

maximal V(g, ~). ,~gd,
It follows from 1.5 that, given v~V, there exists gEG with g - v = ~ v
if and only if f ( v ) = 0 for all f ~ R which are homogeneous of a degree
not divisible by d. It then suffices to require that f~(v)=0 for all i such
that d does not divide di. This proves (i). It follows that (-] H i is the
dgdi
union of the distinct maximal V(g, (). Since each V(g, 0 is an irreducible
algebraic variety, (ii) now follows from a standard elementary result in
algebraic geometry [12, p. 15].
3.3. Corollary. Assume moreover that G acts irreducibly in V. Let
p=char(k) and denote by e the part prime to p of the greatest common
divisor of d 1. . . . . dm (e equals this g. c.d, if p=0). Then the center of G
is cyclic of order e.
The center of G consists of scalar multiplications by Schur's lemma.
If scalar multiplication by ( lies in this center, then (using the previous
notations) the set 0 Hi must be all of Vby 3.2(i). This means that all di
d~d~
are divisible by d. Conversely, if this is so then (-id lies in the center
of G, by 3.2 (i), (ii). Then 3.3 follows by observing that, if p > 0, the order d
has to be prime to p.
Remark. If G is a Weyl group then 3.3 gives a well-known result
[4, p .112].
From now on we assume that k = ~ and that G is a finite reflection
group. So m = n and the di are the degrees of G. For each integer d, we
denote by a(d) the number of di divisible by d. We can now sharpen 3.2
considerably.
3.4. Theorem. Let G be a complex reflection group.
(i) max dim V(g, ()=a(d). In particular, there exists geG with eigen-
geG
value ( if and only if d divides a degree di.
168 T.A. Springer

(ii) For any g~G there exists hEG such that V(h,~) has maximal
dimension a (d) and that V(g, ( ) c V(h, ().
(iii) If dim V(g,()=dim V(g',()=a(d) then there exists h~G with
h. V(g, ()= V(g', ~).
It follows from 1.5, with w=0, that ~ H i= {0}. Hence, if Ci is an
i=l
irreducible component of H i (an irreducible hypersurface in V) we also
have ~ Ci= {0}. It follows that the n irreducible hypersurfaces Ci
i=1
intersect properly in the sense of [22, p. 146]. But then for any subset A
of { 1. . . . . n} the C~ with i~A also intersect properly, so that each irreduci-
ble component of (-] C~ has dimension IAI(n - 1 ) - ( [ A I - 1) n = n -IAI.
i~A
Consequently, each irreducible component of a partial intersection
Hi has dimension n - [A}. Applying this to the case that A = {ild,(dl}
i~A
and using 3.2(ii) we obtain (i). (ii) is also a consequence of 3.2(ii).
We come now to the proof of (iii). Let dim V(g, ~)=a(d). Number
the f/such that fl ..... f, (where a =a(d)) are those with degree divisible
by d. We claim that the restrictions of the (fib __<i<__~to V(g, ~) are algebraic-
ally independent polynomial functions on V(g, (). To show this, consider
the morphism of algebraic varieties ~b of V(g, () to {E" with

n
qb(v)=( f , (v), ... , f,(v)).
From ~ Hi={0} one concludes that the fibre r consists of {0}
i=1
only (observe that the f with i>a all vanish on V(g, ()).
But by a known result [12, p. 92] this fibre has dimension at least
a-dim r ~)). It follows that dim qS(V(g, ~))=a, which can only be
if, as we claimed, the restrictions of the (fi)t =<i~_ato V(g, () are algebraically
independent.
Let U c V(g, ~) be the set of those elements which are not contained
in a maximal space V(h, ~) distinct from V(g, ~). It is a Zariski-open
subset of V(g, ~).
Doing the same things for V(g', () we obtain a morphism ~b': V(g', ()---,IE"
and a Zariski-open subset U' of V(g',(). Now ~b(U) and 4/(U') both
contain a nonempty open subset of C ~, hence have a nonempty intersec-
tion. This means that there exist vsV(g,~) and v'~V(g',~) such that
(a) v' is not contained in a maximal V(h, () distinct from V(g', ~), (b)fi(v)=
f~(v') (1 <i<n). By 1.5, it follows from (b) that there is h~G with h. v=v'.
But then v' e h . V(g, ~) = V(h g h- 1, ~). By (a) this can only be if h. V(g, () =
V(g', (), which proves (iii).
Now fix g e G such that dim V(g, 0 = a ( d ) . Let H be the stabilizer of
V(g, ~) in G and let H' be the normal subgroup of H whose elements
 Regular Elements of Finite Reflection Groups 169

fix V(g, ~) elementwise. Let H 0 be the group of restrictions of elements

of H to V(g, ~'). Clearly Ho is isomorphic to H/H'.
3.5. Proposition. (i) We have ]Hof =< ]7I d~.
did,
(ii) I f equality holds in (i) then H o is a complex reflection group in
V(g, ~) whose algebra of invariants is generated by the restrictions to
V(g, ~) of the f with d] di.
(iii) I f no nontrivial element of G fixes V(g, ~) elementwise then
equality holds in (i) and H is isomorphic to H o .
We have seen in the proof of 3.4(iii) that the restrictions to V(g, ~)
of the f~ with d]di are algebraically independent. (i) and 0it now follow
from 2.4(i) and 2.4(ii), respectively.
In the proof of (iii), we denote by V' the projective space defined
by V and by H~ the projective hypersurface defined by Hi (1 < inn). It
follows from 3.4(iii) that G acts transitively on the irreducible components
of the intersection ~ H~. By the definition of H, the number of these
d,~a,
components is [G[ [HI -1.
Since the Hi with dXdi intersect properly in V, the corresponding
projective hypersurfaces H'i intersect properly in V' and hence an
intersection multiplicity is defined for each irreducible component of the
latter intersection (see [22, p. 2001). Because of the transitivity of G, the
multiplicity is the same for all components. Let # be this common
multiplicity.
By Bezout's theorem we have that the total number of irreducible
components, each counted with its multiplicity, equals the product of
the degrees of the intersecting projective hypersurfaces (for Bezout's
theorem see [14, p. 107]).
It follows that
I-[ di=~lGIIHI -~

Since ]GI = YI dl, we obtain

i=l
lur = 1--[
did,

Using (i) we find that /z<[H']. Under the hypothesis of (iii) we have
IH'I-1, w h e n c e / , = 1 and ]H]=IHoF = I~ di. This proves (iii).

4. Regular Elements in Reflection Groups

In this section, G is a complex reflection group. The notations are
as before.
170 T.A. Springer

A vector v~ V is called regular if v is not contained in any of the

reflecting hyperplanes of G. The following known result, proved in [18],
provides a characterization of the regular vectors.
4.1. Proposition. Let w V, let G~ be the stabilizer of v in G.
(i) v is regutar if and only if G~={1}.
(ii) G, is generated by the reflections which it contains.
As in no. 3 we denote by ~ a primitive d-th root of unity.
4.2. Theorem. Let g~G be such that the eigenspace V(g, ~) contains a
regular vector. Then we have the following:
(i) ga= 1.
(ii) dim V(g, ~)=a(d).
(iii) The centralizer of g in G is isomorphic to a reflection group in
v(g, (), whose degrees are the di divisible by d and whose order is [I di.
d[di
(iv) The elements of G with the property (ii) form a single conjugacy
class.
(v) The eige~values of g are ~-P', where the Pi are the exponents of G.
Let v~ V(g, 0 be regular. Then gd fixes v, hence is the identity by 4.1 (i),
which proves (i). By 3A(ii) there is h e G with dim V(h,~)=a(d),
V(g, ( ) c V(h, (). Then h-~g fixes v, whence, again by 4.1(i), h=g. This
proves (ii). A similar argument, using 3.4(iii), proves (iv). (iii) follows
from 3.5(iii). The proof of (v) is similar to the one given in [4, p. 122]
for the case of a Coxeter element in a Weyl group. We briefly sketch
the argument. Let (ei)l_<~_<,be a basis of V consisting of eigenvectors of g,
with el ~ V(g, ~) regular. By (i) the eigenvalues of g are d-th roots of unity,
let ~h, be the one corresponding to ei. Let D~ be the derivation of S
defined by ei, and let
J=det(Dif~h~i,j<=,.
By [4, p. 113] we have J(el):#O. Consequently, there is a permutation s
of { 1.... , n} such that
Ds(i)f(el):~O (l__<i__<n).

It follows that the polynomial f xj ej must involve a monomial

J
xf '-~ xs(~). The invariance of f then gives that ( d ' - l + h ~ ' = l , which
implies (v).
We shall say that g~G is regular if g has a regular eigenvector.
By 4.2(i) the order of the corresponding eigenvalue equals the order of g,
hence is uniquely determined by g. By 4.2(ii), (iv) the regular elements
of a given order are conjugate.
 R e g u l a r E l e m e n t s o f Finite Reflection G r o u p s 171

4.3. Corollary. Let g~G be a regular element of order d, such that

there is only one di divisible by d. Then the centralizer of g is cyclic of
order d.
This follows from 4.2(iii).
Now Iet G be a Weyl group, with an irreducible root system. Let
ceG be Coxeter element. For this notion and for the properties of
Coxeter elements we refer to [4, p. 116-123].
4.4. Corollary. The centralizer of c in G is the cyclic group generated
byc.
Let the degrees di be ordered in increasing magnitude. One knows
that the smallest degree dl equals 2 and that d 2> 2 (because of the
irreducibility assumption). Also we have that di+d,+l_i is constant,
say h + 2. Then d, = h and d,_ 1 < h. Now a Coxeter element has a regular
eigenvector, whose eigenvatue is a h-th root of unity. 4.4 then follows
from 4.3. For another proof of 4.4 see [5, p. 35-37].
We next discuss miscellaneous results about regular elements.
4.5. Proposition. Let g be as in 4.2. Let p be an irreducible complex
representation of G, with character X. Then the eigenvalues of g in the
representation p are the ~-p,~x), where the Pi(Z) are the z-exponents of G
defined in 2.5.
From formula (2) of no. 2 we obtain
n
[GI -I ~ z ( h ) ( d e t ( 1 - ( - l h T ) ) - ' = L ( ( - 1 T ) l ~ ( 1 - ( - a ' T a') 1. (1)
heG i= 1

Multiply both sides of (1) by ( t - T) a~d) and then put 7"= 1. In the left-
hand side of (1) we get nonzero contributions only from the h with
dim V(h,~)=a(d), all of which are conjugate to g by 4.2(iv). Using
4.2 (iii), (v) a simple computation then gives that

z(g)---fz((- 1).

Since any power gl also satisfies the hypothesis of 4.2, with (i instead of (,
we have z(gi)=~((--i).

This determines the restriction of Z to the cyclic group generated by g.

The assertion then follows from the connection between fx and the
z-exponents (see 2.5).
4.5 is somewhat similar to a result proved for Coxeter elements of
Weyl groups by Kostant in [11, p. 399].
4.6. Proposition. Assume that G is a .finite Coxeter group. Let g ~ 1
be a regular element of G which is not an involution.
12 Inventiones math., Vol. 25
 72 T. A, Springer

(i) There exist two involutions s, t e G such that g = s t.

(ii) I f sl, tt is a second pair of such involutions then there exists h e G
centralizing g such that sl = s h- 1, t~ = h t.
We may assume that there is a real vector space Vo such that
V= Vo|162 and that G stabilizes Vo| 1. Let v~---,~be the semilinear map
of V with V o | 1 7 4 (the last bar denoting complex conjugation).
We have g - v = g . ~ if g e G . Moreover we may assume that the poly-
nomials Ji are real, i.e. satisfy f i ( v ) = f ( ~ ) ( 1 < i < n).
Let v e V(g, () be regular. Since g is not an involution, the order d
of ~ is > 2, so that ( is not a real number.
From g. v = ( v it follows that g - i . ~ =(~, whence V(g, () = V(g -1, ~).
Now an argument like that used in the proof of 3.4(ii) shows that there
exists a regular v~V(g,~) with f~(v)~lR for all i. This implies that
f ( v ) = f ( ~ ) ( l < i < n ) , so that by 1.5 there is t ~ G with t . v = ~ . Then
t 2. v = v , whence t z = 1. It also follows that

( t - l g t ) 9 V = ~ -1 v = g -1 "V,

which implies that s = g t is also an involution. Hence g = s t, as we claimed

in (i).
We come now to the proof of (ii). Let v as before (with all fi(v)elR)
and put v ~ = t l . v . Then g - l . v l = ~ v l , so vl~V(g-l,~). Also fi(v~)=
fi(v)=fi(~). Let H be the centralizer ofg in G, let H 0 be its restriction to
V (g- 1, ~).
By 3.5 (ii), (iii)/4o is a reflection group, whose algebra of invariants
is generated by the restrictions to V(g-1, () of the f with d[d,. 1.5 now
implies that there exists h e H with v~ = h - ~ . Hence t~ 9 v = ( h t ) . v , whence
t I = h t. Then s~ = h- ~. This finishes the proof of (ii).
Remark. 4.6 (i) was proved by Carter [5, p. 45] for all elements of
Weyl groups, using a case by case discussion. We shall give in no. 8
another proof of Carter's result, based on 4.6.

4.7. Proposition. Assume that G is a Weyl group. Let g be a regular

etement o f order d. Then for each i prime to d we have that gi is conjugate
to g in G.
Since G is a Weyt group, G stabilizes a structure over the field of
rationals in V. It follows that ifg has a regular eigenvector with eigenvalue
{, a primitive d-th root of unity, then so does gi. The assertion now
follows from 4.2 (iv).
4.8. Corollary. Let G and g be as in 4.7. Let Z be the character of a
complex representation o f G. Then x(g) is a rational integer.
 Regular Elements of Finite Reflection Groups 173

This is clear. Again, this is true for any element of a Weyl group. We
Shall return in no. 8 to the more general result.
4.9. From now on we assume that G is a Weyl group. Let R be its
root system. The roots are elements in V, they span a real vector space
Voc K such that V ~ V0|
We fix a positive definite symmetric bilinear form on V0 x V0 which
is G-invariant. We extend it to a positive definite hermitian form (,)
on V x V
An ordering on Vo defines an ordering of R, which in turn fixes a
basis B of R (in the sense of [4, p. 153]) and a set of generators S of G.
The elements of S are reflections of order 2. One can speak of the length
1s(g) of an element of G with respect to S.
If R1 c R is a closed subsystem of R [4, p. 160], the reflections defined
by the roots in R1 generate a subgroup G1 of G, called a Weft subgroup.
The rank of G1 is the dimension of the subspace of V generated by the
roots of R1.
4.10. Proposition. Let geG be a regular element of order d.
(i) The elements (~f R are permuted by g in orbits of length d.
(ii) There exists an ordering of R such that, with respect to the set S
of generators of G defined by it, g has length d -1 IR[.
(iii) I f g has no eigenvalue 1, the length of g with respect to any set S
of generators, defined by an ordering of R, is at least d -I [el.
Let v be a regular vector in V(g, ~), { being a primitive d-th root of
unity. Suppose that c~eR, gS. a = a . Then
(v, ~) = ~(v, ~).
Since v is regular, we have (v, c 0 * 0 for all c~sR (otherwise v would lie
in the reflecting hyperplane defined by c0. Hence ~s= 1 and s is divisible
by d, which implies (i) (this argument is due to Kostant, see [10, p. 1021]).
Since G is a Weyl group, together with ~ all its conjugates over the
rationals occur as eigenvalues ofg. Hence we may assume that ~ = ez~a-'.
Multiplying v by a suitable nonzero scalar, we may and shall assume that
the real parts Re(v, a) are nonzero for all ~eR. We define an ordering
on R by declaring e~R to be positive if Re(v, a ) > 0 and negative other-
wise.
Observe that Re(v, g- ~) = Re (~-1 (v, ~)). Let {or,ga,..., ga-~. ~} be
a g-orbit in R. The d complex numbers (v,g~a)0~<a are located at the
vertices of a regular d-gon in the complex plane.
The action of g on the orbit determines a rotation over an angle
- 2 g d - t of the d-gon. It follows that there is exactly one i with O < i < d
such that Re(v,g ~- a)>0, Re(v,g i+~. ~)<0. From (i) we conclude that
12"
174 T.A. Springer

the number of roots a < 0 with g. c~<0 equals d -1 [RI. One knows that
this number equals the length of g, see [21, p. 7], whence (ii). The argu-
ment is similar to the one given by Steinberg in [19, p. 57-58]. If g has
no eigenvalue 1, the sum of the roots in a g-orbit must be 0, so that each
g-orbit contains at least one ~ > 0 with g. ~ < 0 . So (iii) is a consequence
of (i).
We terminate this section by two results which can be used to decide
whether a given element of a Weyl group is regular.

4.11. Lemma. An element g of a Weyl group whose eigenvalues are

distinct primitive d-th roots of unity is regular.
Let ~ be one of the eigenvalues. If ve V(g, ~) is nonregular, there exists
e e R such that v lies in the hyperplane H defined by ~. There is a G-stable
Q-structure on V We may assume that v is defined over some Galois
extension k/ff~ containing ~. It then follows that for all s in the Galois-
group of k/Q we have that sveH. But our assumptions imply that these
sv generate V, which leads to a contradiction.

4.12. Lemma. Let G be a Weyl group. If v e V(g, ~) is nonzero and non-

regular, there exists a Weft subgroup G1 of G whose rank is strictly smaller
than dim V, such that the degree d of ~ divides one of the degrees of G1.
The same is true for Coxeter groups.
We use the notations of 4.9. Ifg has an eigenvalue 1, g fixes an element
of V0 | 1, and the assertion follows from 4.t (ii).
Assume now that g has no eigenvalue 1. By 4.1 (ii) there is a e R such
that the reflection r~ defined by a fixes v. Let w = ( g - 1 ) -~ -a, then
g. w = w + ~ . It follows that 2(w, ~)+(~, ~ ) = 0 , which implies that
r,. w = w+c~. But then r,. g has an eigenvalue 1, and ve V(r~. g, ~), so that
we are in the case considered first. The proof for Coxeter groups is quite
similar, it is left to the reader.

5. Regular Elements in Coxeter Groups

As an illustration and application oft he results of the preceding section
we shall discuss now the regular elements of an irreducible Coxeter
group G. We denote by geG a regular element of order d>=2. If G is a
Weyl group, the conjugacy class ofg is uniquely determined by d (see 4.2).
The d =>2 which occur as orders of regular elements are called the regular
numbers of G.
We first deal with the case o f a Weyl group G. Let R be its root system.
We discuss now the various types of irreducible root systems R. We refer
to [4] for the properties of root systems to be used below.
 Regular Elements of Finite Reflection Groups 175

5.1. Type A, (n > 1). We take

V={x=(xl, 9..,x.+l) ~C"+1 ,+i )

i~=lx,=O~.
Let (ei)l _<i_<,+1 be the canonical basis o f C "+I. The root system R consists
of the vectors e i - e j (i#:j), The Weyl group G is the symmetric group
6 , + 1, its action on Vbeing the restriction of the action on C "+1 defined by
S" ei=esq ~ (s~.+l).

The degrees of G are given by d i = i + l ( l < i < n ) . We have a{d)=

[d -1 (n+ 1)].
N o w if s s ~ , + l has e(j) orbits of length j in {1 . . . . . n + l } then its
characteristic polynomial P~(T) in V is
I-I(T j - I)~O)(T- 1) -1
j~l
as one easily sees. By 4.2(v) we find that P~(T) is divisible by (T a - l)atdl,
so that g has [ d - ~(n + I)] orbits of length d. If [ d - i (n + 1)] d < n, we would
find from 4.2(v) that g had an eigenvalue ( - 1 with multiplicity >a(d),
which is impossible. It follows that there are two possibilities:
(a) d] n + t and g has d-1 (n + 1) orbits of length d,
(b) dfn, g fixes one element and permutes the others in d - i n orbits
of length d.
In the first case g is a power of a Coxeter element of G, in the second
case it is a power of a Coxeter element ofa Weyl subgroup of type A,_ 1.
Since such Coxeter elements have a regular eigenvector in V (as one
easily sees) and since obviously a power of a regular element is also regular,
it follows that the elements of types (a) and (b) are indeed regular.
Thus the regular numbers are the divisors > 2 of n and n + 1.
5.2. Types B,, C. (n>2). We have V=~2". Now G consists of the group
of all linear transformations t of V with
[Link]=gie~(o, (1)
ti = _ 1, se 6 , . As before, (e i) is the canonical basis. G is the Weyl group
of the two root systems B. and C., the first one consisting of the vectors
++_ei+_-ej (i#j) and + e i, the second one of the vectors +e~+_e~ (i4-j) and
++_2ei.
T h e degrees of G are the integers 2,4 . . . . . 2n. It follows that a(d)=
[d -~ n] ifd is odd and a(d)=[2d -~ n] ifd is even. Let t be as above. An
orbit {i, s. i. . . . . s t-~. i} of s in {1, ..., n} is called a positive (negative)
orbit o f t if t-
I]~, ,= +1(-1),
j=0
176 T.A. Springer

see [5, p. 25]. If t has e+ (j) positive orbits of length j and e_ (j) negative
orbits of that length, then its characteristic polynomial in V is

P,(V)= [I(X ~- 1)e+~(X~+ 1)e - o

j->l

Using 4.2 (v) one finds, in a manner similar to that of 5.1, that there are
the following possibilities:
(a) d is an odd divisor of n, g has d-1 n positive orbits of length d,
(b) d is an even divisor of 2n, g has 2 d - t n negative orbits of length
89
The Coxeter elements of G are those with a negative orbit of length n.
One easily sees that each element g with the properties of (a) or (b) is
a power of a Coxeter element.
Since a Coxeter is regular (in the present case this can also easily be
checked directly) it follows that (a) and (b) describe the regular elements
of G.
The regular numbers are the divisors > 2 of 2 n.
5.3. Type D~ (n>4). Again, V=C". The Weyl group G is the group of
n

all linear transformations of the form (1), with l-[ e~= 1. Hence G is the
subgroup of the group considered in 5.2 consisting of the elements with
an even number of negative orbits.
R consists of the vectors +e~++_ej (i4:j). The degrees of G are the
integers 2,4 .... , 2 n - 2 , n. Put 6 ( x ) = l if x is an integer and 6 ( x ) = 0
otherwise. Then
a(d)=[d-l(n - 1)] + 3 ( d - i n ) ifd is odd,
a (d) = [d- 1 (2 n - 2)] + ~ (d- l n) if d is even.

We find the following possibilities:

(a) d is an odd divisor of n, alI orbits ofg are positive of length d,
(b) d is an odd divisor of n - 1, g has one positive orbit of length 1
and d -1 ( n - 1) positive orbits of length d,
(c) n is even, d is an even divisor of n, all orbits of g are negative of
length 89d.
(d) d is an even divisor of 2 n - 2 , g has d -1 ( 2 n - 2 ) negative orbits
of length 89 and moreover another orbit of length 1, which is positive
or negative according as d-1(2 n - 2 ) is even or odd.
The Coxeter elements are those of type (d), with d = 2 n - 2 . The
powers of the Coxeter elements are those of the form (b) and (d). An element
of the form (a) or (c) is a power of an element with one positive orbit of
 Regular Elements of Finite Reflection Groups 177

length n or of one with 2 negative orbits of lengths 89n. The regular

numbers are the divisors > 2 of 2 n - 2 and n.
5.4. The Exceptional Types. Here we shall give the results in the form
of tables. We have listed in the tables below for the exceptional types the
possible regular numbers d and for each d the degrees of the centralizer
ofg (which, by 4.2 (iii), is isomorphic to a reflection group) and the charac-
teristic polynomial of g. We denote by q5s the cyclotomic polynomial
whose roots are the primitive s-th roots of unity.
Table 1. Type E 6 (degrees 2, 5, 6, 8, 9, 12) Table2. Type E7 (degrees 2,6.8,10,12,14, lg)
d Degrees Characteristic d Degrees Characteristic
centralizer polynomial Centralizer polynomial
4 2
2 2,6,8,12 4)24)1 2 2,6,8, 10, 12, 14, 18 4)7
3 6,9, 12 4)3 3 6, 12, 18 4)~ 4)2
4 8, 12 4)2 4)~ 6 6, 12, 18 4)3 4)2
6 6, 12 4)6z 4)3 7 14 4)14 4)1
8 8 4)8 4)2 4)1 9 18 4)o 4)1
9 9 4)o 14 14 4)~4 4)2
12 12 4)12 4)3 18 18 4)1s cb2

Table 3. Type E8 Table 4. Type G

(degrees 2, 8, 12, 14, 18, 20, 24, 30) (degrees 2, 6, 8, 12)

Degrees Characteristic d Degrees Characteristic

centralizer polynomial centralizer polynomial

2 2,8,12,14,18,20,24,30 4)~ 2 2,6,8, 12 4)~

3 12, 18, 24, 30 4)~ 3 6, 12 4)32
4 8, 12, 20, 24 4)~ 4 8, 12 4)2
5 20, 30 4)~ 6 6, 12 4)62
6 12, 18, 24 4)4 8 8 4)s
8 8, 24 4)s2 12 12 4)iz
10 20, 30 4)~o
12 12, 24 4)~2
15 30 4)1s
20 20 4)20
24 24 4)24
30 30 q)3o

Finally, in type G 2 there are three regular classes, viz. those of the
Coxeter elements and their powers + 1, so that the regular numbers
are 2, 3, 6.
To establish the results contained in these tables, one first observes
that the regular numbers d must be divisors of degrees di, by 3.4(i). The
maximal value of d is the order of a Coxeter element, which is regular.
Using that a power of a regular element is regular one then finds already
quite a few of the regular numbers.
178 T.A. Springer

To illustrate the arguments we shall discuss the case of type E 8.

The Coxeter element then gives that 2, 3, 5, 6, 10, t5, 30 are regular
numbers. The remaining ones occur among the numbers 4, 7, 8, 9, 12, 14,
18, 20, 24. Using 4.11 one sees that 20 and 24 are regular numbers. It
follows that 4, 8, 12 are also regular. There remain the numbers 7, 9, 14, 18
which must be ruled out. If d = 18 were regular, then 4.2 (v) would give
the eigenvalues of the corresponding element. However one sees that
the eigenvalues given by 4.2 (v) cannot be those of a linear transformation
over the field of rationals, so d = 18 is nonregular.
For 7, 9, 14 the argument is similar.
Most of the results of the tables are established by this kind of
argument. In a few cases an additional reasoning is needed:
(a) T o prove that 8 is a regular number for E 6 o n e observes that 4
is one (since the order of a Coxeter element is 12). If an element g having
a primitive 8th root of unity as eigenvalue were nonregular then an
argument like that of the proof of 4.11 gives that there is a root orthogonal
to the space spanned by the eigenvectors for such eigenvalues. But since
g2 is regular, this space contains a regular vector, which is a contradiction.
(bl To prove that 14 is a regular number for E 7 o n e uses 4.12, noticing
that all degrees of Weyl groups of rank < 6 are less than 14.
5.5. The Other Coxeter Groups. The remaining finite irreducible
Coxeter groups are either dihedral or are of types H3,//4 (see [4, p. 194]).
The dihedral case is easily dealt with and is left to the reader. The Tables 5
and 6 give the regular numbers for H 3 and H4. For the degrees of H 3
a n d / / 4 see [8, p. 772].

Table 5. Type H 3 (degrees 2, 6, 10) Table 6. Type H 4 (degrees 2, 12, 20, 30)

d Degrees N u m b e r of d Degrees N u m b e r of
centralizer classes centralizer classes

2 2,6,10 1 2 2, 12, 20, 30 1

3 6 1 3 12, 30 1
5 10 2 4 12, 30 1
6 6 1 5 20, 30 2
10 I0 2 6 12,30 1
10 20, 30 2
12 t2 1
15 30 2
20 20 2
30 30 2

The results about the regular numbers are established in a way

similar to that followed before. One first uses the existence of Coxeter
elements and also an appropriate extension of 4.11 (taking into account
 Regular Elements of Finite Reflection Groups 179

that H 3 and H 4 are defined over Q(I/5)). Some trouble is given by d = 6

for type H3. There one can establish regularity by using the geometric
description of H 3 as a group of orthogonal transformations in IR3
leaving invariant an icosahedrou.
For any of the above d there exists an element in the group G of
t y p e H 3 or H4 having an eigenvalue ~ which is a primitive d-th root of
unity. For different ~, these elements need not be conjugate. Using 4.2 (v)
one can see how many classes there are which such eigenvalues. The
result is given in the tables.
5.6. The results of this section lead to certain complex reflection
groups. Consider, for example, a regular element g of order l0 in a Weyl
group of type Es. From Table 3 one sees that the centralizer of g is
isomorphic to a 2-dimensional complex reflection group H in a suitable
space V(g, ~), with degrees 20 and 30. One easily sees that H acts irre-
ducibly. By 3.3 the centre Z of H is cyclic of order 10. Then H / Z is iso-
morphic to the icosahedral group (see [-15, p. 286]). The discussion in
[Link].] of H and of similar groups is bases on the properties of Klein's
polyhedral groups. One can proceed the other way round and derive the
main results about polyhedral groups from the theory of Weyl groups,
making use of the results of this paper. We shall not go into this matter
here.
6. An Extension
Let G be a reflection group in the complex vector space V, let f~ . . . . . f,
be, as before, algebraically independent generators of the algebra of
invariants of G. We put di = deg(f~). We denote by cra linear transformation
of V of finite order such that a G a -~ = G. In this section we shall extend
a number of results, obtained in the previous sections about eigenvalues
of elements of G, to eigenvalues of elements of G~r.
6.1. Lemma. (i) The f~ may be chosen such that or. fi = ~i f (1 < i < n),
with suitable roots of unity ei.
(ii) The set of pairs (di, ~i)1 <-i<-, is independent of the choice of the fi.
Denote by R~ the space of homogeneous invariants of degree j. Then
~r acts on R~ so that Rj may be decomposed into eigenspaces for a. Define
inductively f / t o be an eigenvector of a of degree dg which is algebraically
independent of fx .... ,f~-l- By 2.3 (iii) these f will be as required. (ii)
follows by the same kind of argument as was used to prove 2.3 (iii).
We call the ~ the factors of a. Fix a primitive d-th root of unity ~.
Let a(d, a) be the number of i such that
~i~ai-----1 (l<i<n).
We denote by V(ga, ~) the eigenspace of ga for the eigenvalue ft.
 180 T, A. S p r i n g e r

6.2. Theorem. (i) max dim V(g a, ()=a(d, a). In particular, there exists
geG
gEG such that ga has an eigenvalue ( if and only if ei(d'= 1 for some i.
(ii) For any geG there exists h~G such that V(h~r,() has maximal
dimension a(d, a) and that V(g cr, ~)~ V(hcr, ().
(iii) If dim V(gc~,()=dimV(g'a,()=a(d,a) then there exists hEG
such that h. V(ga, ~) = V(g'~r, ~).
The proof runs parallel to that of 3.4 and will be left to the reader.
Fix g e G such that V(ga, ~) has maximal dimension a(d, a). Let H be
the stabilizer of V(ga, ~) in G and let H' be the normal subgroup of H
whose elements fix V(ga, ~) elementwise. Let H 0 be the group of restric-
tions of elements of H to V(ga, ~).
6.3. Proposition. (i) We have ]Ho]=
< l-I dl.
e, ~d, = 1
(ii) If equality holds in (i) then H o is a reflection group in V(ga, ~),
whose algebra of invariants is generated by the restrictions to V(ga, (,)
of the f with el ~a, = 1.
(iii) If no nontrivial element of G fixes V(ga, ~) elementwise, then
equality holds in (i) and H is isomorphic to H o .
This is an extension of 3.5. The proof is as before. We deduce from
it the following extension of 4.2.
6.4. Theorem. Let g ~ G be such that V(g a, ~) contains a regular vector.
Then we have the following:
(i) If crn= 1 then g a has order d.
(ii) dim V(ga, ~) =a(d, a).
(iii) The centralizer of g~r in G is isomorphic to a reflection group in
V(g a, ~), whose degrees are the di with r = 1, its order being the product
of these d~.
( i v ) / f dim V(ga, ~)=dim V(g'a,~)=a(d, cr) then ga and g'a are
conjugate by an element of G.
(v) The eigenvalues of g~r are the e[l ~-p, ( l < i < n ) , where the Pl
are the exponents of G.
The proof is like that of 4.2.
If ga has a regular eigenvector we say that ga is a regular element
of Go. There is an interesting particular case (which becomes trivial in
the situation of no. 4).
6.5. Corollary. Assume that a has a regular eigenvector with eigenvalue I ;
(i) The eigenvalues of a are the ~? 1 (1 < i < n).
(ii) The centralizer G~ of ~r in G is a reflection group in V(a, 1), whose
degrees are the dl such that ei = 1.
 Regular Elements of Finite Reflection Groups 181

Remark. If, in the situation of 6.5, G is a Coxeter group, so that V has

a structure over IR and if a is defined over IR, then G~ is also a Coxeter
group. Similarly for a Weyl group.
6.6. Now let G be a Weyl group, with root system R. Assume that there
is an ordering of R such that a stabilizes the set of all positive roots. Put
2p= ~ .
Then p is a regular vector, as is well-known, which is fixed by o. So 6.5
applies.
This situation was discussed by Steinberg in [21]. 6.5 (i) and the first
part of 6.5 (ii) are proved in [[Link]., p. 22 and p. 15]. The last assertion
of 6.5 (ii) is stated in [loc. cit., p. 22] as an observation.
We next give extensions of 4.6 and 4.10(i). The proofs are as before.
6.7. Proposition. Assume that G is a Coxeter group, that aZ~G and
that g a is a regular element 4= l which is not an involution.
(i) There exists an involution s e G a and an involution t e G such that
go=st.
(ill I f sl , tl is a second pair of such involutions then there exists h e G
centralizing g a such that s 1 = s h - 1, tl = h t.
6.8. Lemma. Let G be a Weyl group with root system R. Assume that a
f i x e s R. Let g a have a regular eigenvector whose eigenvalue is a primitive
d-th root of unity.
(i) g a permutes the roots of R in orbits o f length d.
(ii) There exists an ordering of R such that each orbit of g o contains
exactly one ~ > 0 with g o 9c~<0.
Let G be an irreducible Weyl group with root system R. The non-
trivial possibilities for a are discussed in [21 ]. We shall review them below,
and find the regular elements of G a in these cases. The discussion will
be quite similar to that of no. 5. g a denotes a regular element, the
corresponding eigenvalue being a primitive d-th root of unity ~, with
d>l.
6.9. Type A, (n>2). Let the notations be as in 5.1. There is only one
possible coset Go, viz. - ~ , + 1 - So we may take a = - l . We have
d j = j + 1, s j = ( - - 1)j+l (1 <=j<=n). It follows that

a(d,a)=[ 89 1)1 if d is odd,

=[d-l(n+l)] if d - O (rood4),
=[2d l(n+l)] if d-=2 (mod4), d > 2 ,
=n if d = 2 .
182 T.A. Springer

Proceeding as in 5.1 we find the following possibilities for g a:

(a) n and d are odd, g has 89d- ~(n + 1) orbits of length 2 d,
(b) n is even, d is odd, g has 8 9 orbits of length 2d and one of
length 1;
(c) n = 3 (mod 4), d = 0 (mod 4), g has d -~ (n+ 1) orbits of length d;
(d) n = d = - O (mod 4), g has d -1 orbits of length d and one of length 1,
(e) d = 2 (rood 4), g has 2 d- 1(n + 1) orbits of length 89d,

(fl d = 2 (rood 4), g has 2 d - I n orbits of length d and one of length 1.

The maximal value o f d is 2 n + 2 ifn is even and 2n ifn is odd (given
by the element of type (e) and (f), respectively).
6.10. Types B2, C2. Notations being as in 5.2, with n = 2 , we may
take a to be given by
a- el = 2 - § +e2),
~r'e2 = 2-~(el +ca).

We have d~ =2, d2 =4, ex = l, e 2 = - 1. One checks that the possibilities

for a regular g a are:
(a) d = 2 , g = 1,
(b) d = 4, g a Coxeter element of G.
6.11. Type D, (n>4). In this case there is ar of order 2, which (with
the notations of 5.3) we can take to be given by cr.e~=e~ (1 < i < n - 1 ) ,
a . G = - e,. W e ha ve di = 2 i, ei = l ( l <_ i <_n - 1 ) , d, = n, ~. = - I . It follows
that
a(d, a ) = [ d - l ( n - 1)] if d is odd,
=[d-l(2n-2)]+6(d -1 n+ 89 i f d is even,
6 being as in 5.3.
go can now be considered to be an element of the Weyl group of
type B. and must then have an odd number of negative orbits. One
finds the following possibilities:
(a) d is an odd divisor of n - 1 , g a has d - l ( n - 1) positive orbits of
length d and one negative one of length 1,
(b) d is an even divisor of 2 n - 2 , g a has 2 d - ~ ( n - 1 ) negative orbits
of length 89d and one orbit of length d,
(c) d > 2 is an even divisor of 2n such that 2 d - l n is odd, g a has
2 d- ~n negative orbits of length 89d.
The maximal possible value of d is 2 n, given by a g ~r which has one
negative orbit of length n.
 Regular Elements of Finite Reflection Groups 183

6.t2. Exceptional Types. In the remaining cases we say that (G, tr) is
of exceptional type. They are the following ones: G is of type D~ and a
has order 3, G is of type E 6 , F4, G 2 and a has order 2. The factors are
given in [21, p. 81-82].
The results are given in the tables below, which are similar to the
ones in no. 5.

Table 7. Type D 4 (degrees 2, 4, 4, 6; Table 8. Type E 6 (degrees 2, 5, 6, 8, 9, 12;

factors 1, ee"'3,e4~'
3, 1) factors 1, 1, 1, 1, - 1, 1 )

d Degrees Characteristic d Degrees Characteristic

centralizer polynomial centralizer polynomial

2 2,5,6,8,9, i2 q'~
3 4,6 ~] 3 6, 12 q% r
6 4,6 q~ 4 8, 12 q~a
a q~
12 4 c~12 6 6,9,12 q~63
8 8 ~P8 ~P2 opt
12 i2 ~btz q56
18 9 ~t8

Table 9. Type F 4 (degrees 2, 6, 8, 12; factors 1, - 1, 1, - 1)

d Degrees centralizer Characteristic polynomial

2 2, 8 ~ ~
4 6,8 4~4
~
8 8, 12 (T 2 - 2 ~ T + 1)2
12 6 ~12
24 12 (T2-2cos~2 T+ I) (Ta-2cos72T+ l)

In type G2 the degrees are 2 and 6, with factors 1, - 1 . There is a

regular element with d = 2 , centralizer of order 2 and characteristic
polynomial q)2 4'i and one with d = 12, centralizer of order 6 and

characteristic polynomial ( T 2 - 2 c o s 6 T + I ) .
These results are derived in much the same manner as in no. 5. We
also use another lemma. To state it, we recall that ~r induces a permuta-
tion p of the root system R, see [21, p. 73].
6.13. Lemma. Assume that there is a basis B of R which is stabilized
by p and such that the p-orbits in B consist of mutually orthogonal roots.
Then there exists a Coxeter element of G which commutes with ~r.
Let el . . . . . co, be an ordering of the roots of B such that those in a
p-orbit are consecutive. Let r,6G be the reflection defined by e s R .
Then c = r,,.., r,, is a Coxeter element with the required property.
184 T.A. Springer

That in the case of D 4 w e find d = 12 is proved by an argument like

that of 4.11. Taking powers we can deal with d = 3, [Link] case d = 2 is
given by the element -cr.
Table 8 comes from Table 1, using that we may take a = - 1 if G is
of type E6.
The case d = 2 4 in T a b l e 9 is taken care of by the argument of 4.11.
It then also follows that d = 8 occurs in Table9. The case d = 12 comes
from 6.13, and gives d = 4 . Finally, d = 2 is given again by - o .
We restrict ourselves to these indications, leaving the details to the
reader.

7. Twisted Coxeter Elements

7.1. Let G be a Weyl group in V, with root system R. Let a be a
linear transformation of V of finite order, with o ~ G, a G o - 1 = G. Assume
that there is a basis B of R which is fixed by o (so that ~- fixes R). In that
situation one can introduce a "twisted" version of Coxeter e~ements,
as was observed by Steinberg (unpublished). We shall show how the
results of no. 6 can be used to establish the properties of the twisted
Coxeter elements.
As a preliminary we give below, in the cases where R is irreducible
the maximal possible order h~ of the eigenvalue of an element g o-. We
also have listed the dimension no of the space of vectors of V fixed by o.
It is readily seen that n~ equals the number of ~--orbits in B.

Table 10

Type h~ no

A. (n even) 2n+2 89
A. (n odd and >1) 2n ~(n+l)
D. (~2= 1) 2n n- 1
D4 (~3= 1) 12 2
E6 18 4

7.2. Lemma. (i) We have ho n, = JR[.

(ii) An element g a with an eigenvalue of order ho is regular.
(i) follows from Table I0 by inspection. (iil follows from the results
of no. 6.
7.3. Let r~eG be the reflection defined by a e R . For each orbit of a
in B pick out a root a in that orbit. Let g be the product of the cor-
responding r~, in any order, and put e = g a . We call c a twisted Coxeter
element of (G, o-). We shall show presently that they possess properties
similar to those of the usual Coxeter elements of Weyl groups. We first
 Regular Elements of Finite Reflection Groups 185

establish some auxiliary results. We denote by ( , ) a positive definite

hermitian form on V • V, which is invariant for G and a.
7.4. Lemma. Assume that G fixes no nonzero vector of V. Let c=gc7
be a twisted Coxeter element.
(i) c fixes no nonzero vector of V.
(ii) I f g' is a product of n~ distinct reflections r= with ~eB, then g' ~
has no eigenvalue 1 if and only if the c~ lie in different a-orbits.
Let ve V, c. v=v. We then find that

a . v = v + Y c,~, (1)

where c~ runs through a set S of representatives of the a-orbits in B.

Assume that ~re=l. Applying ~r,~r2,...,~r ~-* to (1) and adding the
results we find from (1) that
d~=0,
ateB

where we have d==e, q with e=>0 if ~ S . It follows that c==0 for all
~reS, hence a v = v and g v = v . Now g is a Coxeter element o f a Weyl
subgroup of G. Using that a Coxeter element of an irreducible Weyl
group has no eigenvalue 1 [4, p. 118], we see that (v,~)=0 for all a s S .
From o . v=v it then follows that (v, ~ ) = 0 for all e e B . This proves (i).
Let g' be, as in (ii), the product of n, reflections r=, e running through
a subset S of B, with [Sl=n~. If not all a-orbits are represented in S,
there is a proper a-invariant subspace W of V, containing S. Let W' be
its orthogonal complement. Clearly g' acts trivially on W'. Since the
space of fixed points of a is not contained in W, there must be a nonzero
vector in W' fixed by both a and g'. This establishes (ii).
The following lemma is a generalization of [4, L e m m a 1, p. 117].
It is due to Steinberg.
7.5. Lemma. Let X be a finite forest, let c~ be an automorphism of X.
Let x~--~g~ be a map of X into a group F, such that gx and gy commute
whenever x and y are not joined in X. Let ~ be an automorphism of F
with g,t.(xl=~(gx) ( x e X ) . Let Y be the set of ~b-orbits in X, let Y be the
set of total orderings of Y. For each section s: Y---,X and each ~ "
denote by p~,r the product in F of the sequence (g~(y))y~r, defined by ~.
Then if p,,r and Pt,, are two such elements there exists g e F with p,,,=
gp~,r
The p r o o f is by induction on n = [XI. The case n = 1 is trivial, assume
n > 2 . Let a be an endpoint of X, let b e X - { a } be joined to a if such
a point exists; if not let b be an arbitrary point of X - { a } .
186 T.A. Springer

We first show: for each p~, r there exists g ~ F such that g Ps, ~(~ g)- ~=
Pt, ~ where t Y contains a and b.
That we may achieve a ~ t Y follows from the trivial formula

p~,~ p~,~q,(p~,~)=p~o~,~. (2)

Now if s is such that a ~ s Y, bCs Y, we have that Ps, e is a product of g,

and some gx ( x e X - {a}) which commute with g,. We can then establish
the result by using once more (2), but now for the forest X - {a}.
To prove 7.5, it now suffices to consider only the p~,~ with a, b ~ s Z
The proof then goes along the lines of that of the lemma in [4] quoted
before. We leave the details to the reader.
The next theorem gives the properties of a twisted Coxeter element.

7.6. Theorem. Assume that R is irreducible. Let c be a twisted Coxeter

element.
(i) The G-conjugacy class of c in Gtx is uniquely determined.
(ii) c has a regular eigenvector whose eigenvalue is a primitive h,-th
root of unity, with multiplicity 1.
(iii) The eigenvalues of c are ~ f l e-Z,i~j:h, (1 < j < n ) , no eigenvalue
equals 1.
(iv) c permutes the roots in n, orbits of length h,.
(v) The centralizer of c in G is cyclic and consists of the powers of c
which tie in G.
(i) follows from 7.5 and the fact (easily checked) that two bases of R
which are fixed by ~ are conjugate by an element of G which commutes
with cr.
N o w let c' = g t r be a regular element with an eigenvalue of order h~,
such elements exist by 7.2 (ii). By 6.8 (ii) we may take an ordering on R
such that each orbit of c' in R contains exactly one ~ > 0 with c'. ct<0.
We m a y (replacing ~ by a conjugate) assume that ~ keeps positive roots
positive.
It then follows that there are n~ roots ~ with ct>0, g. ~ < 0 , hence g
has length nr Using 6.4 (v) a case by case inspection shows that c has
no eigenvalue 1. But then it follows from 7.4(ii) that c' is a Coxeter
element, hence conjugate to c by (i). (ii) now also follows. (iii), (iv), (v)
are consequences of 6.4 (v), 6.8 and 6.4(iii), respectively.

7.7. Remarks. We have deduced here the properties of twisted Coxeter

elements from the results of no. 6. This involves a case by case discussion.
If o"2 -~ 1 one may, however, also prove the regularity of twisted Coxeter
 Regular Elements of Finite Reflection Groups 187

elements (which is the essential point) along the lines of Steinberg's

proof for the usual Coxeter elements (see [-4, p. 118-121]).

8. A Reduction Theorem for Elements of Weyl Groups

In this section we prove a result about elements of Weyl groups (8.4)
which will allow us to extend 4.6 and 4.8 to nonregular elements.
8.1. Let G be a Weyl group in V. We use the notations of 4.9. Denote
by Q the lattice in V spanned by the roots c ~ R and by P the weight
lattice of R, which is the set of v~ V with

for all ~ R . Then P/Q is a finite abelian group. We denote its order by e.
8.2. Proposition. Let g~G.
(i) We have (g - 1) P c Q.
(ii) Let 1 be a prime number. I f v ~ P is such that ( g - 1 ) v ~ l Q , then g is
a product of reflections r, (~ ~ R) qf G such that (r~ - 1) v ~ l Q.
Since the reflection r,e G defined by a e R is given by
r~(x)= x - 2(x, cO(~, c~)-a ~,
(i) is true if g is a reflection. If g is arbitrary, write it as a product of
reflections g = r,1 ... r~ and put g' = r,2 ... r , . Then
( g - 1) x = r ~ ( g ' - t ) x + ( r = l - 1)x,

from which (i) follows by induction on h.

To prove (ii), we have to use the affine Weyl group Go of the dual root
system. This is the group of affine transformations of V generated by
the elements of G and the translations defined by the elements of Q, it
is isomorphic to the semi-direct product of G and Q. For a discussion of
affine Weyl groups we refer to [-4, p. 173] (we are working here in a
complex vector space, whereas in [[Link].] one is over the reals, but this
is an immaterial difference).
Now let v be as in (ii) and put w = 1 - 1 v. The assumption made in (ii)
then implies that w is fixed by an element t of Go. According to a theorem
of Steinberg [21, p. 10-11] we can conclude that t is a product of re-
flections contained in Go and fixing w. These affine reflections are the
affine transformations r~,k (c~ R, k ~ ;g) with
r~.k(X) = X -- 2 ((~, X) -- k)(c~, c~)-i ~.
Now r,.k(w)=w means that ( r ~ - I ) v ~ l Q . The fact that h is a product of
such r,. k implies that g is a stated in (ii).
13a ]nventionesmath.,Vol,25
188 T.A. Springer

8.3. Corollary. Let g~G.

(i) d e t ( g - 1) is an integer, which is divisible by e.
(ii) I f det(g-1):l: _+e there is a proper Weyl subgroup of G which
contafns g.
(i) is a direct consequence of 10.2(i). If det(g-1)4= +e, then either
d e t ( g - 1)=0, in which case g fixes some nonzero vector ve V and (ii)
follows from 4.1 or there is a prime number l such that d e t ( g - 1 ) is
divisible by le. In the latter case there exists v e P - I P such that
( g - 1) velQ.
Denote by R~ the subset of R consisting of the roots ~ with

2 (v, ~) (~, ~)- 1 _ 0 (rood 1).

One checks that R, is a root system, let G, be its Weyl group. By 10.2 (ii)
we have g e G , . Since v r we have R, + R and G1 #G. This proves (ii).
8.4. Theorem, Let G be the Weft group of an irreducible root system R,
let g ~ G. There are the following (non-exclusive) possibilities:
(a) g is contained in a proper Weyl subgroup of G,
(b) g is regular,
(cl - t eG and - g is in case (a) or (b),
(d) R is of type D, and g is an element with only 2 orbits, which are
both negative.
The notations in case (d) are those of 5.3.
By 9.3 (ii), it suffices to prove this for elements g with d e t ( g - 1 ) = + e,
where e is as before. We shall check the various irreducible root systems,
using the results of no. 5.
Type A,. The description of characteristic polynomials made in 5.1
shows that d e t ( g - 1 ) : ~ 0 if and only if g is a Coxeter element, hence
regular.
Type B,, C,. We have e = 2. From 5.2 it follows that det ( g - 1)= + 2
if and only ifg has only one negative orbit, of length n. Then g is a Coxeter
element.
Type D,. Now e=4. A similar argument shows that d e t ( g - 1 ) = _ + 4
if and only if g is as in case (d).
in dealing with the exceptional types we use the fact that the charac-
teristic polynomial fg of g is of the form

i
 Regular Elements of Finite Reflection Groups 189

(where 4"h is the cyclotomic polynomial of primitive h-th roots of unity),

the ej being such that the e 2"~!~ are eigenvalues of g, each ~ej occurring
with the appropriate multiplicity. Hence

det(g-1)=+l-[4"ej(1). (1)
J
It is well-known that

4"h(1)= 1 if h is not a prime power,

=1 if h = l ~, k > l .

Type E 6. We have e = 3 . By 3.4(i), the ej occur among the integers

2, 3, 4, 5, 6, 8, 9, 12. If d e t ( g - l ) = +3, the only possibilities for the
characteristic polynomial of g are 4"9, q'624"3,45124"3. All of these possi-
bilities lead to regular elements, by Table 1 in 5.4.
Type E 7 . Now e = 2. The ej occur among the integers 2, 3, 4, 5, 6, 7,
8, 9, 10, 12, 14, 18. I f d e t ( g - 1)= + 2 , s must have a factor 4"2,4"4 or 4"8.
But since the degree of 45h is even for all h 4=2 in this list of integers, a
factor 452 must occur. So, if d e t ( g - 1)= +2, we have that g is in case (c),
since -- 1 9 G.
In the remaining cases E s , F4, G2 we have e = 1. We shall only discuss
the most complicated case E 8, the discussion of the other two is left to
the reader.
Type Es. The ej occur among the integers 2, 3, 4, 5, 6, 7, 8, 9, 10, 12,
15, 18, 20, 24, 30. If d e t ( g - 1)= -+ 1, we need only consider those e i for
which ~e,(1)= 1. Also, if some el is twice a power of an odd prime, then
Idet(1 +g)l > i from which it follows (since - 1 eG) that then g is in case
(c). Hence it suffices to consider the case that the ej occur among the
integers 12, 15, 20, 24, 30. If one of the last four occurs, then g is regular
by Table 3 in 5.4. There only remains the case that g has characteristic
polynomial 4"22, in which case this table shows that g is again regular.
We now discuss some applications of 8.4.
8.5. Theorem. Let G be a Weyl group, let Z be the character of a complex
representation of G. Then )~ takes rational integral values.
We use induction on the order 16I. The starting case [GI = 2 is trivial.
Assume that 8.5 is true for all Weyl groups of order less than [G[. We may
assume G to be irreducible.
Let g 9 G. The induction assumption and 4.8 imply that )c(g)9 unless
we are in the cases (c) and (d) of 8.4. If - 1 9 G and if Z is an irreducible
character, then z ( - g ) = _+)~(g). This observation implies that z ( g ) e Z
also in case (c).
190 T.A. Springer

Finally, if we are in case (d), an easy verification shows that all powers
g J, with j prime to the order g, are conjugate to g, from which it again
follows that X(g)e7A This proves 8.5.
8.6. Remark. 8.5 is a known result. It can be deduced by inspection
from the known character tables of the irreducible Weyl groups. For
references to the literature we refer to [-1]. The present proof seems
somewhat more satisfying. It would be interesting to prove by an argu-
ment similar to the above, the sharper result that all representations of
a Weyl group are rational (see [[Link].]).
The next result was proved in the case of Weyl groups by Carter
[5, p. 45].
8.7. Theorem. An element + 1 of a finite Coxeter group is a product
of at most two involutions.
Let G be a finite Coxeter group, let g~G. For the proof of 8.7 it
suffices to assume that G is irreducible and that gZ % 1. First let G be a
Weyl group. Using 8.4 and 4.6, an argument like that given in the proof
of 8.5 establishes 8.7.
In the case (d) of 8.4 one uses that then there is an involution i in G
such that ig i = g - 1 (which is easily checked).
To complete the proof, we have to deal with the cases that G is
dihedral or is of t y p e / / 3 , / / 4 . The dihedral case is easy (and the assertion
is, of course, well-known in that case).
Type H 3 . From 3.3, using that the degrees are 2, 6, 10, it follows that
- 1 e G. Since G is a group of orthogonal transformations in an odd-
dimensional vector space V, it is welt-known that any g e G has an eigen-
value + 1. Hence + g fixes a vector in V, and an application of 4.1 (ii)
shows that we can reduce the situation to that of a 2-dimensional Coxeter
group, which is a dihedral or a Weyl group.
Type H 4. We have - 1 e G. From 3.4, using that the degrees are now
2, 12, 20, 30, we obtain that the eigenvalues of g e G are primitive d-th
roots of unity, where d occurs among the following integers: 2, 3, 4, 5, 6,
10, 12, 15, 20, 30. If one of the last four integers occurs then one reads
off from Table 6 that g is regular, and 4.6 shows that g has the required
property. If g has an eigenvalue - 1, then - g has an eigenvalue 1 and
4.1 (ii) shows that we are reduced to a lower-dimensional situation,
which has already been discussed. So we may assume that g has no
eigenvalue - 1.
Now from Table 6 and 4.6 (v) it follows that if g is not regular, we
may assume that its characteristic polynomial is a product of two
distinct real quadratic polynomials, whose roots are d-th roots of unity,
where d occurs the integers 3, 4, 5, 6, 10. If g has two eigenvalues with
 Regular Elements of" Finite Reflection Groups 191

relatively prime orders, we can write g as a product of two commuting

elements, each fixing elementwise a 2-dimensional subspace. We then
are again reduced to a lower-dimensional situation. Since we are free
to replace g by - g , it follows, finally, that we have only to consider the
case that the eigenvalues of g are primitive roots of unity of orders 3, 6
and 5, 10 respectively. In both cases, g2 is reguIar, as follows from Table 6
and 4.2 (ii). Also, 4.2 (iii) implies that the centralizer H of gZ is isomorphic
to a 2-dimensional complex reflection group with degrees 12, 30 and
20, 30 respectively, acting in a suitable space V(g2, ~).
Let HcSL2(q2) be the binary icosahedral group. Denoting by ~e
the group of scalar multiplications in (r 2 by d-th roots of unity it follows
from the results of [-15] that H is isomorphic to H-#3 or //.kL 5. But
then g would lead to an element of H . ge whose eigenvalues are a d-th
and a 2d-th primitive root of unity (d=3, 5). Taking determinants one
sees that this is impossible. This establishes 8.7 for type//4 and finishes
the proof of 8.7.
Question (posed in [5, p. 45]). Can one prove 8.7 directly from the
abstract definition of finite Coxeter groups?

9. Certain Nilpotent Elements of Semisimple Lie Algebras

In a well-known paper [10], Kostant has exhibited an explicit
connexion between the regular nilpotent elements of a complex semi-
simple Lie algebra (called "principal nilpotent elements" in [[Link].])
and the Coxeter elements of the corresponding Weyl group. In this
section that result will be generalized, so as to obtain a similar connexion
for certain other nilpotent elements. We first fix notations and recall
some known facts.
9.1. Let G be a complex connected semisimple Lie group, let g be its
Lie algebra. I f X e g we denote by Z(X) its centralizer in G, i.e.

Z ( X ) = {g~GlAd(g) X=X}

(Ad denoting the adjoint representation). We denote by Z(X) ~ the

connected centralizer of X in G (i.e. the identity component of Z(X))
and by 3(X) the centralizer of X in g, which coincides with the Lie
algebra of Z(X).
For results to be recalled see [-3, EIII, w Let A be a nonzero nil-
potent element of g. By the theorem of Jacobson-Morozov there exist
B, H ~ g such that
[H, A] = 2 A , [/4, B] = - 2 B , [A, B] = H .
13b Inventionesmath.,VoI. 25
192 T.A. Springer

If i t 7Z,,put
g(i)= {x~gIEH, x] =ix}.
Then g is the direct sum of the 9(0, moreover we have

Eg(i), g(j)] r
so that the g(i) define a structure of graded Lie algebra on g. All this is
unique up to an automorphism of g induced by an element of Z(A) ~
F r o m now on we make the following assumptions on A :
(a) 3(A) consists of nilpotent elements,
(b) A is even, i.e. 9 (i)= 0 if i is odd.
Actually, (b) is a consequence of (a), as can be checked from the
classification of nilpotent conjugacy classes (given in m3, E IV] for the
simple Lie algebras of classical type and in I-9] for those of exceptional
type). This matter will be discussed in a forthcoming paper by Bala and
Carter.
9.2. Lemma. Assumption (a) holds if and only if 3(A~m9(i)= {0} Jot
i<O.
This follows from [3, p. 238].
Let a be the smallest integer such that g(i)=0 for i>2a. Since
d i m g ( i ) = d i m g ( - i ) we then have g ( - 2 a ) + {0}.

9.3. Lemma. Let M ~g ( - 2 a) and put C = A + M.

(i) C is nilpotent if and only if M = 0 .
(ii) 3 (C) ~ g (0) = {0}.
If C is nilpotent, the Jacobson-Morozov theorem shows that there
exists X ~ with IX, C ] = 2 C . Write X = X X I , with XiEg(2i ). Looking
at homogeneous components, we obtain

IX o, A] = 2 A , I X o, M] = 2 M .

From the first formula we obtain, using 9.2 that Xo = H. The second one
then gives (2a + 2) M = 0, whence M = 0. This proves (i).
Now let Xeg(0)c~3(C ). We find, similarly, that [X, A] =0, whence
X = 0 by 9.2.
Let S be the 1-dimensional torus in G whose Lie algebra is CH.
Let y be the character of S with

Ad(s) A =s~A (s~S).

We then have, ifX~g(2i),
Ad(s) X =si~ X. (1)
 Regular Elements of Fimte Reflection Groups 193

Fix a primitive (a+ 1)-th root of unity ~ and choose c e S such that
c'P= ~. The eigenspaces of Ad(c) are

V ~ = g ( 2 i ) + g ( 2 i - 2 a - 2) (O<i<=a),

the eigenvalue for V~being ~i. In particular, the fixed point set of Ad(c)
is 9 (o).
Let M#:O be in g ( - 2 a ) and put C = A + M , as before.

9.4. Lemma. (i) Ad(c) stabilizes ,3(C) and has no nonzero fixed point
in 3(C).
(ii) c normalizes Z(C) ~ and centralizes no element of Z(C) ~ outside
the center of G.
(iii) Z(C) ~ is a connected solvable linear algebraic group.
Since A d ( c ) C = ~ C , it follows that Ad(c) stabilizes 3(C) and that c
normalizes Z(C) ~ (i) now follows from 9.3 (ii). Then (ii) is a consequence
of known results [-2, p. 229] and the same is true for (iii) [-21, p. 71].
9.5. Lemma. The following properties are equivalent:
(i) Z(C) ~ is a maximal torus of G.
(ii) C is regular semisimpte.
(iii) C is semisimpte.
M can be chosen such that C has these properties if and only if the
following assumption is verified:
(c) V~ contains a regular semisimple element.
The implications (i)~ (ii)~ (iii) being known, it suffices for the proof
of the first part to show that (iii) implies (i). If (iii) holds, the connected
centralizer of C is on the one hand reductive (by a general result) and on
the other hand solvable (by 9.4 (iii)).
It must then be a maximal torus, which proves (i).
The "only if" part of the last statement follows from the first one,
since C s V~. Now assume that (c) holds. It is known that there is an
Ad(G)-invariant polynomial function F on 9 such that X e g is regular
semisimple if and only if F(X),t= O.
Let L = Z ( H ) ~ One knows that the orbit Ad(L)A is a Zariski-open
subset of 9(2). It then follows from (c) that there are YsAd(L)A and
Z E g ( - 2 a ) with Z4:0 such that F(Y+Z)@O. Choose lEL such that
Y=Ad(I) A and put M =Ad(l) -t Z. Then C = A + M is as required.
9.6. Lemma. Assumption (c) of 9.5 is a consequence of the .following
one"
(d) dim g (4)-- dim g (2)- 1.
194 T.A. Springer

We shall prove, more precisely, that if (d) holds any C = A + M with

nonzero M e g ( - 2 a ) is semisimple. This will imply 9.6 (by 9.5). Consider
such a C and let C, be its nilpotent part [2, p. 151]. The uniqueness of
the Jordan decomposition in g [[Link].] shows that A d ( c ) C , = ~ C , ,
whence C,e V1. Let C , = X 2 + X_2, be the decomposition in its homo-
geneous components. Since C~ commutes with C, we have

EX2 3r X_ 2a, A + M] = 0 ,

from which it follows that [A, X 2 ] = 0 . Now X~--~[A, X] is a surjective

linear map of 9(2) onto 9(4) ( a similar result is in [3, p. 240]). If(d) holds,
its kernel must be CA. Then X 2 is a multiple of A. Using 9.3 (i) it follows
that (7, is a multiple of either A or M. Using that [C,, C] = 0 it follows
in both cases from 9.2 that (7, =0. Hence C is semisimple, as we claimed.
Assume that (c) holds. We write T = Z ( C ) ~ this is a maximal torus
of G. Its Lie algebra t=3(C) is a Cartan subalgebra of g. We denote by
W the Weyl group of G with respect to T, which is a reflection group in t.
Since c normalizes T by 9.4(ii), it defines an element we W.
9.7. Proposition. (i) w is a regular element of W, it has C as a regular
eigenvector, with eigenvalue ~. No eigenvalue of w equals 1.
(ii) I f (d) holds, the eigenvaIue ~ of w has multiplicity 1.
A regular semisimple element of 1 is regular in the sense of no. 4
for W,, as is well-known. This proves the first part of (i). The second one
follows from 9.3 (ii).
An argument like that used in the proof of 9.6 shows that if
X2 + X_2aet ~ Vt, the assumption (d) implies that X z is a multiple of A.
On the other hand we have t ~ g ( - 2 a ) = { 0 } , since the elements of
g ( - 2 a ) are nilpotent (which follows from (1)). These observations imply
(ii).
Let n = d i m t and denote by {dl, ..., d,} the degrees of the reflection
group W. Let 2 N be the number of roots of the root system R of G with
respect to T(or of W). Then N = ~ (di- 1).
i=1

9.8. Corollary. (i) W contains a regular element of order a + i.

(ii) If(d) holds there is exactly one di which is divisible by a+ 1.
(i) follows from 9.7 (i) and 4.2 (i), (ii) from 9.7 (ii) and 3.4 (i).
We still assume (c).
9.9. Lemma. dim g ( 0 ) = d i m g ( 2 ) = 2(a+ 1) -1N.
If c~eR, denote X,~9 a corresponding root vector. By 9.7 and 4.10,
all orbits ofw in R have length a + 1. Now i f ~ s R is fixed we can normalize
 Regular Elements of Finite Reflection Groups 195

the root vectors for the roots in its orbit such that
Ad(c)g.x~=xwi.~ ( 0 < i < a + 1).
We then have
Ad (c)a + 1. X~ = ~X~, (2)

where e = _+ l (this follows by taking the X= to belong to a Chevalley

basis, see e.g. [20]).
If e,= 1, then
X~+ X~.~+ ... + X~o.~

is an eigenvector of Ad(c) for the eigenvalue 1, and all such eigenvectors

are linear combinations of those of this form, for suitable ~. If ~ = - 1
in (2), put 0 = e ~i'~"+1/ ~. Then
X~ +O Xw.~+ ... + O~Xwo.~

is an eigenvector of Ad(c) whose eigenvalue is not an (a+ 1)-th root of

unity, which is impossible. Hence e.= 1 always, from which it follows
that dim g(0)equals the number 2 ( a + l ) - l N of w-orbits in R. That
dim g (0)= dim g (2)is a known fact which follows from [3, p. 240] and 9.2.
9.10. Regular Nilpotent Elements. We shall now indicate how the
preceding results may be used to establish Kostant's connection between
regular nilpotent elements of g and Coxeter elements in the correspond-
ing Weyl group (the method used is Kostant's original one). We assume
G to be simple. Let To be a fixed maximal torus of G, let R0 be the root
system of G with respect to To. Let Bo be a basis of Ro. If ~eRo, we
denote by X~Eg a corresponding root vector. We put
A= ~X~.
~e Bo

9.11. Proposition. (i) A is a nilpotent element which satisfies (a), (b)

and (d).
(ii) The corresponding element w is a Coxeter element of W.
It is known that Z(A) ~ is unipotent, which implies (a). It is also known
that (b) holds and that (with the notations of 9.1) we have that for i > 0
the space g(2i) consists of the linear combinations of the X, where c~
has height i, with respect to the basis B0. For all this see for example [17].
It readily follows that (d) holds. Let n be the rank of G. By 9.9 we have that
the element w has order 2 n- ~N. By [4, p. 119] it is then a Coxeter element.
The conjugacy class of A is uniquely determined, its elements are
those nilpotents whose centralizer has dimension n. See [17].
9.12. We now consider the more interesting case that A is not regular
and satisfies (a), (b), (d). By 9.7 we must then look for non-Coxeter ele-
196 T.A. Springer

ments w in Weyt groups, without eigenvalue I which have an eigenvalue

whose eigenvectors are regular with multiplicity 1. Assume that the root
system R is irreducible. The results of no. 5 show which elements w and
which orders a + 1 can occur. In the classical types A,, Bn, C,, D, there
are none. The results for the other types are contained in the first two
columns of Table 11.
T a b l e 11

Type a+ 1 d i m g(O) Dynkin diagram

E6 9 8 ~ - ~

E7 14 9 ~ •

Es 24 10 ~ ~ •

E8 20 12 ~ ~ • o
6

E8 15 16 o o ~ - ~ o
•
F4 8 6 o-~--~-~- ~
G2 3 4

We do not yet know that these possibilities can all be realized as

coming from a nilpotent A satisfying (a), (b), (d). First observe that if
this is the case, dim g (0) must be as stated in the third column of Table 11
(by 9.9).
Put L=Z(H) ~ as in the proof of 9.9. This is a reductive group and
dim L = dim g (0). There is a parabolic subgroup P of G, whose Lie algebra
is [_I g(i), which has Las a Levi subgroup, see [3, p. 240].
i>0
Consider the Dynkin diagram of A, defined in [3, p. 243]. This is
the Dynkin graph D of W,, together with a distribution of the numbers
0 and 2 on the vertices of D [in the case of a nilpotent satisfying (b)).
The Dynkin graph of the semisimple part of L can be found from D by
omitting the vertices to which the number 2 is attached and the segments
ending in one of these vertices [loc. cit.].
In [14, Table 23, p. 190] one finds for the exceptional types a list of
all possible Dynkin diagrams of nilpotents satisfying (a). It is easy to
check which ones lead to the values of a + 1 and g(0) of Table I1. They
are listed in the last column of that table. We have put a cross at the nodes
to which the number 0 is attached. In all these cases we have dim g(4)=
d i m g ( 2 ) - l , and they also come from nilpotents A satisfying (a), as
follows from [loc. cit.]. Hence all Weyl group classes containing elements
 Regular Elements of Finite Reflection Groups 197

with the properties enunciated in the first paragraph of 9.12 arise from
nilpotents A satisfying (a), (b), (d).
9.13. Remarks. (1) In [6] a bijective map is described from a certain
set of nilpotent conjugacy classes of 9 to a set of conjugacy classes of the
Weyl group of G. From the results discussed above one can deduce a
similar map. It can be checked that these two maps give the same image
of a class on which both are defined.
(2) In this no. we have assumed that we worked over C. This is not
really necessary. The same results are obtained if one works over an
algebraically closed field K of characteristic p, where p is sufficiently
large. We refer to [3, E Ill, w4] for a discussion of the restrictions to be
imposed on p. The results of no. 3 and no. 4 remain valid in this more
general situation, as follows by using [4, p. 107, Th. 3]. It should also be
remarked that the proof of 9.11 indicated above works in characteristic p,
if p is bigger than the order of a Coxeter element (Kostant's original
proof only works in characteristic 0).

References
1. Benard, M.: On the Schur indices of characters of the exceptional Weyl groups. Ann. of
Math. 94, 89-t07 (1971)
2. Borel, A.:Linear algebraic groups. New York:Benjamin 1969
3. Borel, A.: Seminar in algebraic groups and related finite groups. Lecture Notes in
Mathematics 131. Berlin-Heidelberg-New York: Springer 1970
4. Bourbaki, N.: Groupes et alg6bres de Lie. Chap. tV, V, VI. Paris: Hermann 1968
5. Carter, R.W.: Conjugacy classes in the Weyl groups. Comp. Math. 25, 1-59 (1972)
6. Carter, R.W., Elkington, G.B.: A note on the parametrization of conjugacy classes.
J. Alg. 20, 350-354 (1972)
7. Chevalley, C.: Invariants of finite groups generated by reflections. Am. J. Math. 77,
778-782 (1955)
8. Coxeter, H.S.M.: The product of the generators of a finite group generated by reflec-
tions. Duke Math. J. 18, 765-782 (19511
9. Dynkin, E.B.: Semisimple subalgebras of semisimple Lie algebras. Am. Math. Soc.
Transl. Ser. 2, 6 (1957), 111-245 (= Math. Sbornik N.S. 30, 349 462 (1952))
10. Kostant, B.: The principal three-dimensional subgroup and the Betti numbers of a
complex simple Lie group. Am. J. Math. 81,973-1032 (I959)
11. Kostant, B.: Lie group representations in polynomiaI rings. Am. J. Math. 85, 327-404
(1963)
12. Mumford, D.: Introduction to algebraic geometry (mimeographed)
13. Noether, E.: Der Endlichkeitssatz der Invarianten endlicher Gruppen. Math. Ann. 77,
89-92 (1916)
14. Samuel, P.: M~thodes d'alg6bre abstraite en g6om6trie alg6brique. Erg. d. Math. 4.
Berlin-Heidelberg-New York: Springer 1955
15. Shephard, G.C., Todd, J.A.: Finite unitary reflection groups. Can. J. Math. 6, 274-304
(1954)
16. Solomon, L.: Invariants of finite reflection groups. Nagoya Math. J. 22, 57-64 (1963)
17. Springer, T.A.: Some arithmetical results on semi-simple Lie algebras. PuN. Math.
I.H.E.S. 30, 115-141 (1966)
198 T.A. Springer

18. Steinberg, R.: Differential equations invariant under finite reflection groups. Trans.
Am. Math. Soc. 112, 392-400 (1964)
19. Steinberg, R.: Regular elements of semisimple algebraic groups. PuN. Math. I.H.E.S.
25, 49-80 (1965)
20. Steinberg, R.: Lectures on Chevalley groups. Yale University, 1967
21. Steinberg, R.: Endomorphisms of linear algebraic groups. Mem. Am. Math. Soc. 80
(I968)
22. Well, A.: Foundations of algebraic geometry. Am. Math. Soc. Colloq. Pub1.29 (2nd ed.),
1962
T.A. Springer
Mathematisch lnstituut
Rijksuniversiteit Utrecht
Utrecht/The Netherlands

(Received December 7, i973)