arXiv:0906.0563v2 [math.GR] 11 Aug 2013
THE z-CLASSES OF ISOMETRIES
KRISHNENDU GONGOPADHYAY AND RAVI S. KULKARNI
Abstract. Let G be a group. Two elements x, y are said to be in the
same z-class if their centralizers are conjugate in G. Let V be a vector
space of dimension n over a field F of characteristic different from 2.
Let B be a non-degenerate symmetric, or skew-symmetric, bilinear
form on V. Let I(V, B) denote the group of isometries of (V, B). We
show that the number of z-classes in I(V, B) is finite when F is perfect
and has the property that it has only finitely many field extensions of
degree at most n.
1. Introduction
Let G be a group. We define an equivalence relation ∼ on G as follows:
for x, y in G, x ∼ y if the centralizers ZG (x) and ZG (y) are conjugate in
G. The equivalence class of x is called the z-class of x in G. The z-classes
are pairwise disjoint and give a partition of the group G. This provides
important information about the internal structure of the group, see [15] for
further details. The structure of each z-class can be expressed as a certain
set theoretic fibration, see [15, Theorem 2.1 ]. In general, a group may be
infinite and it may have infinitely many conjugacy classes, but the number
of z-classes is often finite. For example, if G is a compact Lie group, then it
is implicit in Weyl’s structure theory see, [21], Borel-de Siebenthal [2], that
the number of z-classes in G is finite. Analogously, Steinberg [19, p.107] has
remarked on the finiteness of z-classes in reductive algebraic groups over an
algebraically closed field of good characteristic. In [15], Kulkarni proposed
to interpret the z-classes as an internal ingredient in a group G that can
be used to make precise the intuitive notion of “dynamical types” in the
G-action on any set X. The Fibration Theorem, see [15, Theorem 2.1],
gives a set-theoretic fibration of the z-class of x with base the homogeneous
Date: November 5, 2018.
2000 Mathematics Subject Classification. Primary 20E45; Secondary 20G15, 15A63,
Key words and phrases. conjugacy classes, centralizers, z-classes, orthogonal and symplectic groups.
Gongopadhyay acknowledges the support of SERC-DST FAST grant SR/FTP/MS004/2010.
1
space G/N (x), where N (x) is the normalizer of ZG (x) in G, and a fiber
consists of the elements y in the center of ZG (x) such that ZG (x) = ZG (y).
For example, in classical geometries over R, C or H, it is observed that the
“dynamical types” that our mind can perceive are just finite in number and
this finiteness of “dynamical types” can be interpreted as a phenomenon
related to the finiteness of the z-classes in the corresponding group of the
geometry. With this motivation, the z-classes in the isometry group of the ndimensional real hyperbolic space were classified and counted in [9]. It is also
an interesting problem to classify the z-classes in other linear groups that
appear as isometry group in rank one symmetric spaces of non-compact type.
The z-classes in the isometry group Sp(n, 1) of the n-dimensional quternionic
hyperbolic space have been classified and counted in [8]. Classification of
the z-classes in U(n, 1), the isometry group of the n-dimensional complex
hyperbolic space, has been obtained in [5], also see [8, Appendix]. Recently,
z-classes have also been used in the context of classifying the isometries in
hyperbolic geometries, see, [4, 6, 7].
In addition to these, it is of independent algebraic interest to parametrize
both the conjugacy and the z-classes in a group. For example, the problem
can be asked for finite groups of Lie type; classical groups or exceptional
groups. The conjugacy classes, z-classes and the set of operators themselves
of the general linear groups and the affine groups have been parametrized by
Kulkarni [14]. This has been extended to linear operators over division rings
by Gouraige [11]. In an attempt to understand the z-classes in exceptional
groups, Singh [17] has proved a finiteness result for the z-classes in the
compact real form G2 .
Let F be a field of characteristic different from 2. Let V be a vector space
of dimension n over F. Let V be equipped with a non-degenerate symmetric
or skew-symmetric bilinear form B. The group of isometries of (V, B) is
denoted by I(V, B; F), or simply I(V, B) when the underlying field is fixed.
When B is symmetric, resp. skew-symmetric, I(V, B) is the orthogonal,
resp. symplectic group. In this paper we ask for the z-classes in I(V, B).
Our main theorem is the following.
Theorem 1.1. If F is perfect and has the property that it has only finitely
many field extensions of degree at most dim V, then the number of z-classes
in I(V, B) is finite. This holds for example when the field F is algebraically
closed, the field of real numbers, or a local field.
Along the way we parametrize the z-classes of the semisimple elements,
see, Theorem 4.1.
2
A first step in understanding of the z-classes is to classify the conjugacy
classes. There has been a considerable amount of work on the conjugacy
problem in orthogonal and symplectic groups, see, Asai [1], Burgoyne and
Cushman [3], Kiehm [13], Milnor [16], Springer-Steinberg [18], Wall [20]
and Williamson [22]. A common theme of these works is to reduce the
conjugacy problem to the equivalence problem for Hermitian forms. Our
conjugacy classification has a similar flavor. However, a notable feature of
our classification is that, in the “generic” case when the minimal polynomial
of an element in I(V, B) is a prime-power, it gives an explicit parametrization
of the conjugacy classes, see, Theorem 3.6. Consequently, we also obtain a
parametrization of the z-classes in this case, see, Theorem 3.7. As we shall
see, the z-classification depends on the equivalence problem of Hermitian
forms over arbitrary fields. The equivalence problem of Hermitian forms
was solved by Kiehm [13] and Wall [20]. We will not get into the equivalence
problem of the Hermitian spaces in this paper. However, our classification
of the z-classes is enough to prove our main result, Theorem 1.1.
2. Preliminaries
2.1. Self-dual polynomial. Let F[x] be the ring of polynomials over F.
For a polynomial g(x) let ck (g) denote the coefficient of xk in g(x). Let F̄
denote the algebraic closure of F.
Let f (x) be a monic polynomial of degree n over F such that 0, 1 and −1
are not its roots. Over F̄ let
f (x) = (x − c1 )(x − c2 )....(x − cn ).
Then the polynomial
−1
f ∗ (x) = (x − c1−1 )(x − c−1
2 )....(x − cn )
is said to be the dual to f (x). It is easy to see that
f ∗ (x) = f (0)−1 xn f (x−1 ).
Clearly, ck (f ∗ ) = f (0)−1 cn−k .
Definition 2.1. Let f (x) be a monic polynomial over F such that −1, 0,
1 are not its roots. The polynomial f (x) is called reciprocal, or self-dual, if
f (x) = f ∗ (x).
Thus if f (x) self-dual, then the degree n of f (x) is even, and for all k,
ck (f ) = cn−k (f ).
3
2.2. Decomposition of the space relative to an isometry. Suppose T :
V → V is an element in I(V, B). Let mT (x) denote the minimal polynomial
of T . Suppose p1 (x), . . . , pl (x) are irreducible polynomials over F such that
mT (x) = p1 (x)d1 . . . pl (x)dl , where for i 6= j, pi (x) 6= pj (x). Suppose degree
of mT (x) is m. The integer di is called the exponent, or the multiplicity, of
the prime factor pi (x).
Let E = F[x]/(mT (x)). The image of the indeterminate x in E is denoted
by t. There is a canonical algebra structure on E defined by tv = T v. The
F-algebra E = F[t] is spanned by {1, t, t2 , · · · , tm−1 }. In particular, if the
minimal polynomial is irreducible, then E is an extension field of F. The
following lemma follows from Lemma 4.1 in [10].
Lemma 2.2. (i) The minimal polynomial of an element T in I(V, B) is
self-dual.
(ii)There is a unique automorphism e → ē of E over F which carries t to
−1
t .
Thus an irreducible factor p(x) of the minimal polynomial can be one of
the following three types:
(i) p(x) is self-dual.
(ii) p(x) = x − 1, or, x + 1.
(iii) p(x) is not self-dual. In this case there is an irreducible factor p∗ (x)
of the minimal polynomial such that p∗ (x) is dual to p(x).
Among the irreducible factors of mT (x), suppose pi (x) is self-dual for
i = 1, 2, ..., k1 . Let the other irreducible factors be pj (x), p∗j (x) for j =
1, 2, ..., k2 with pj (x) 6= p∗j (x). For a prime-power polynomial p(x)d , let
Vp = ker p(T )d . Let ⊕ denote the orthogonal sum, and + denote the usual
sum of subspaces. It can be seen easily that there is a primary decomposition
of V (with respect to T ) into T -invariant non-degenerate subspaces:
(2.1)
1
Vi
V = ⊕ki=1
M
2
Vj
⊕kj=1
where for i = 1, 2, ..., k1 , pi (x) is self-dual, Vi = Vpi , and B|Vi is nondegenerate; for j = 1, 2, ..., k2 , Vj = Vpj + Vp∗j , B|Vpj = 0 = B|Vp∗ , here
j
pj (x) 6= p∗j (x). Let Tl denote the restriction of T to Vl . Then mTi (x) =
pi (x)di for i = 1, 2, ..., k1 , and mTj (x) = pj (x)dj p∗j (x)dj for j = 1, 2, ..., k2 .
Let Z(T ) denote the centralizer of T in I(V, B). We observe that the decomposition (2.1) is in fact invariant under Z(T ). Moreover we have a canonical
decomposition
2
1
Z(Tj ).
Z(Ti ) × Πkj=1
Z(T ) = Πki=1
4
Thus the conjugacy classes and the z-classes of T are determined by the
restriction of T to each of the primary subspaces. Hence it is enough to
determine the conjugacy and the z-classes of an isometry T : V → V with
minimal polynomial mT (x) = p(x)d , where p(x) is one of the types (i), (ii),
(iii) above.
Finally note the following lemma. For a proof of the lemma, see, [10,
Lemma 4.2].
Lemma 2.3. Let T be an element in I(V, B). Suppose T : V → V is such
that the minimal polynomial is one of the types (i), (ii) above. Suppose
mT (x) = p(x)d . There is an orthogonal decomposition V = ⊕ki=1 Vdi , where
1 ≤ d1 < · · · < dk = d, and for each i = 1, ..., k, Vdi is free over the algebra
F[x]/(p(x)di ). For each i, the summand Vdi corresponds to the elementary
divisor p(x)di of T .
Remark 2.4. In the above lemma, suppose deg p(x) = m. Then
dimF F[x]/(p(x)di ) = mdi . Suppose Vdi has dimension li as a free module
over F[x]/(p(x)di ). Thus dimF Vdi = mdi li . This gives us a secondary
Pk
n
= i=1 di li .
partition π : m
We end this section with the following definition.
Definition 2.5. Let R be a commutative ring with involution e 7→ ē. Let
ǫ = 1 or −1. An ǫ-Hermitian form on an R-module M is a sesquilinear
mapping s : M × M → R such that for all x, y ∈ M ,
s(x, y) = ǫs(y, x).
That is for ǫ = 1, s is Hermitian; for ǫ = −1, s is skew-Hermitian.
3. The induced form and the conjugacy classes
3.1. The minimal polynomial is prime-power.
Lemma 3.1. (Springer-Steinberg [18]) Let T : V → V in I(V, B) be
such that mT (x) = p(x)d , where p(x) is an irreducible polynomial over F.
Assume that p(x) is either self-dual, or, x − 1. If p(x) = x − 1, then assume
d > 1. Consider the cyclic F-algebra ETd = F[x]/(p(x)d ). We simply denote
it by ET when there is no confusion about d. The ET -module V is denoted
by VT . Then we have the following.
(i) There is a unique automorphism e → ē of ET over F which carries t
to t−1 .
(ii) There exists an F-linear function hT : ET → F such that the symmetric bilinear map h̄T : (a, b) 7→ hT (ab) on ET × ET is non-degenerate. Also
5
there exists c ∈ ET such that for all e ∈ ET , hT (ē) = hT (ce). Moreover, if
p(x) 6= x − 1, we can take c = 1. If p(x) = x − 1, then c = (−1)d−1 .
For a proof of the above lemma cf. Springer-Steinberg [18, p.254]. For
a proof when the field extension Fd = F[x]/(p(x)) is separable, cf. Asai
[1, p.329]. Applying the above lemma we have the following theorem. The
theorem is implicit in the work of Springer-Steinberg [18]. Milnor [16] gave
a version of the following theorem when T is semisimple. We have given a
detailed proof in the general case. The proof is essentially imitating Milnor’s
proof in the semisimple case.
Lemma 3.2. The module V over ET admits a unique ǫ-Hermitian form
H T (u, v) = ǫH T (v, u), ET -linear in the first variable, and is related to the
original F-valued inner product by the identity
(3.1)
B(u, v) = hT (H T (u, v)).
Proof. For u, v in V, consider the linear map L : ET → F given by L(e) =
B(eu, v). There exists a unique e′ in ET such that hT (ee′ ) = L(e). We define
H T (u, v) to be this element e′ . That is, H T (u, v) is defined as follows:
for all e in ET , and for u, v in V, hT (eH T (u, v)) = B(eu, v).
In particular taking e = 1 we have
hT (H T (u, v)) = B(u, v)
Now we see that for u1 , u2 , v in V,
hT (e(H T (u1 , v) + H T (u2 , v)))
= hT (eH T (u1 , v)) + hT (eH T (u2 , v))
= B(eu1 , v) + B(eu2 , v)
= B(eu1 + eu2 , v)
= B(e(u1 + u2 ), v) = hT (H T (u1 + u2 , v))
(3.2)
⇒ H T (u1 , v) + H T (u2 , v) = H T (u1 + u2 , v)
Now for all e′ in ET we have
hT (e′ eH T (u, v)) =
=
(3.3)
B(e′ eu, v)
B(e′ (eu), v) = hT (e′ H T (eu, v))
⇒ eH T (u, v) = H T (eu, v)
This shows that hT is ET -linear in the first variable.
Given any Hermitian form H(u, v) satisfying (3.1) we see that
hT (eH(u, v)) = hT (H(eu, v)) = B(eu, v).
6
Therefore H T (u, v) is unique.
Further, for all e in ET ,
hT (e(H T (u, v)))
= ǫhT (ēH T (u, v)), using part (ii) of Lemma 3.1
= ǫB(ēu, v)
= ǫB(ev, u) = hT (eǫH T (v, u))
(3.4)
⇒ H T (u, v) = ǫH T (v, u)
This proves the theorem.
Remark 3.3. Let S : V → V and T : V → V be two isometries such that
mS (x) = p(x)d , mT (x) = q(x)d , where p(x), q(x) are irreducible and selfdual, deg p(x) = deg q(x) and ES and ET are F-isomorphic. Let s and t
are images of S and T in ES and ET respectively. Let f : ES → ET be an
F-isomorphism such that f (s) = t. Let hS : ES → F be the linear map as in
Lemma 3.1. Then hT = hS ◦ f −1 is such a linear map on ET , and this map
induces a Hermitian form H ′ on VT . Since such a Hermitian form is unique,
hence we must have H ′ = H T . Thus for u, v in VS , hS (H S (u, v)) = B(u, v),
and for u′ , v ′ in VT , hT (H T (u′ , v ′ )) = hS ◦ f −1 (H T (u′ , v ′ )).
Definition 3.4. Suppose E and E′ are isomorphic modules over F, and let
f : E → E′ be an isomorphism. Let H be an E-valued Hermitian form on V
and let H ′ be an E′ -valued Hermitian form on V′ . Then (V, H) and (V′ , H ′ )
are equivalent if there exists an F-isomorphism T : V → V′ such that for all
u, v in V and for all e in E the following conditions are satisfied.
(i) T (ev) = f (e)T (v), and
(ii) H ′ (T (u), T (v)) = f (H(u, v)).
When E = E′ , we take f to be the identity in the definition.
Theorem 3.5. Suppose S and T are isometries of (V, B). Let the minimal
polynomial of both S and T be (x − 1)d or, p(x)d , where p(x) is monic,
self-dual, and, irreducible over F. Let H S and H T be the Hermitian form
induced by S and T respectively.
(i) Then S and T are conjugate in I(V, B) if and only if H S and H T
are equivalent.
(ii) Let Z(T ) be the centralizer of T in I(V, B). Then an isometry C is
in Z(T ) if and only if C preserves H T , i.e. Z(T ) = U (VT , H T ).
Proof. Suppose S is conjugate to T in I(V, B). Let C in I(V, B) be such
that T = CSC −1 . Then C : VS → VT is an F-isomorphism. For l ≥ 1, and
7
v in VS ,
C(sl v) = C ◦ S l (v) = T l ◦ C(v) = tl C(v) = f (sl )C(v).
It follows that, for all e in ES , and v in VS , C(ev) = f (e)C(v). For u, v in
VS , note that
hS (f −1 (H T (C(u), C(v)))
= hS ◦ f −1 (H T (C(u), C(v)))
= hT (H T (C(u), C(v))
= B(C(u), C(v)))
= B(u, v) = hS (H S (u, v)).
Hence, by the uniqueness of H S we have, f −1 (H T (C(u), C(v))) = H S (u, v),
i.e. H T (C(u), C(v)) = f (H S (u, v)). This shows that H S and H T are equivalent.
Conversely, suppose H S and H T are equivalent. Let C : VS → VT be an
F-isomorphism such that (i) and (ii) in Definition 3.4 hold. We have for v
in V,
CS(v) =
C(sv)
=
f (s)C(v)
=
tC(v) = T C(v).
that is, CSC −1 = T . Further, for x, y in V,
B(C(x), C(y))
=
hT (H T (C(x), C(y)))
=
hT (f (H S (x, y)))
=
hS (H S (x, y)) = B(x, y).
Hence C : V → V is an isometry. This completes the proof of (i).
(ii) Note that an invertible linear transformation C : V → V is ET -linear
if and only if CT = T C. Now replacing S by T , and f by identity in the
proof of (i) the theorem follows.
3.2. Conjugacy classes.
Theorem 3.6. Let T be an element of I(V, B). Let the minimal polynomial
of T be p(x)d , where p(x) = x − 1 or p(x) is monic, self-dual and irreducible
over F.
(1) The conjugacy class of T in I(V, B) is determined by the following
data.
(i) The elementary divisors of T .
8
(ii) The finite sequence of equivalence classes of Hermitian spaces
{(VTd1 , HdT1 ), · · · , (VTdk , HdTk )},
where 1 ≤ d1 < d2 < · · · < dk = d, and for each i, HdTi takes
values in the cyclic algebra Edi = F[x]/(p(x)di ).
(2) The centralizer of T is the direct product
U (VTd1 , HdT1 ) × · · · × U (VTdk , HdTk ).
Proof. Suppose S : V → V and T : V → V are two isometries. If S and
T are conjugate in I(V, B), then by the structure theory of linear operators
and Theorem 3.5, it is clear that they have the same data.
Conversely, suppose S and T have the same data. The elementary divisors
of S and T determine orthogonal decompositions of V as
(3.5)
V = VSd1 ⊕ · · · ⊕ VSdk ,
(3.6)
V = VTd1 ⊕ · · · ⊕ VTdk ,
where 1 ≤ d1 < · · · < dk = d, and for each i, VSdi , resp. VTdi is free when
considered as a module over ESdi , resp. ETdi . Since ESdi and ETdi are isomorphic,
wthout loss of generality, we identify them with Edi = F[x]/(p(x)di ). Since
S and T have the same set of elementary divisors, VSdi is isomorphic to VTdi
as a free module over Edi , for i = 1, 2, ..., k. For each i = 1, 2, · · · , k, since
(VSdi , HdSi ) is equivalent to (VTdi , HdTi ), by Theorem 3.5, S|VSd is conjugate to
i
T |VTd . Hence S is conjugate to T .
i
The description of Z(T ) is clear from the orthogonal decomposition of V
and part (2) of Theorem 3.5.
3.3. The z-classes.
Theorem 3.7. Let T : V → V be an element in I(V, B) such that mT (x) =
p(x)d , where p(x) is self-dual and irreducible over F. The z-class of T is
determined by the following data.
(i) The degree m of p(x).
(ii) A non-decreasing sequence of integers (d1 , . . . , dk ) which corresponds
n
= Σki=1 di li .
to the secondary partition π : m
(iii) A sequence (Ed1 , . . . , Edk ) of isomorphism classes of cyclic algebras
over F, where for each i = 1, 2, . . . , k, Edi is isomorphic to F[x]/(p(x)di ).
(iv) A finite sequence of equivalence classes of Hermitian forms
(Hd1 , . . . , Hdk ), where each Hdi takes values in Edi .
9
Proof. Let S and T be two isometries of (V, B) with same data (i)−(iv). We
use the same notations as in the previous theorem. Let mS (x) = p(x)d , and
mT (x) = q(x)d , degree of p(x) = degree of q(x) = m. For each i = 1, . . . , k,
ESdi and ETdi are isomorphic. Let fi : ESdi → ETdi be one such isomorphism.
For simplicity, for each i, we identify ESdi , and ETdi with Edi . Moreover,
following Remark 3.3 assume hS = hT .
Since the Hermitian forms HdSi and HdTi are equivalent, let Fi : (VSdi , HdSi ) →
T
(Vdi , HdTi ) be an equivalence of the Hermitian spaces. We see that, for
u, v ∈ VSdi ,
B(Fi u, Fi v) =
hT (HdTi (Fi u, Fi v))
=
hT ◦ fi (HdSi (u, v))
=
hS (HdSi (u, v)), see, Remark 3.3,
=
B(u, v).
Thus Fi is an isometry with respect to B. Further, Fi conjugates Z(S|VSd ) =
i
U (VSdi , HdSi ) and Z(T |VTd ) = U (VTdi , HdTi ). Thus, F = F1 ⊕ F2 ⊕ · · · ⊕ Fk is
i
an isometry of (V, B) and F conjugates Z(S) and Z(T ).
Conversely, suppose S and T are in the same z-class. Replacing S by its
conjugate, we may assume, Z(S) = Z(T ). Hence by part (2) of Theorem 3.2
we see that S and T have isomorphic decompositions (3.5) and (3.6). After renaming the indices, if necessary, we may assume further that for
i = 1, 2, ..., k, (VSdi , HdSi ) and (VTdi , HdTi ) are equivalent. In particular, ESdi
and ETdi are isomorphic, and their common dimension over F is mdi . This
implies deg mS (x) = deg mT (x). Consequently we attach the partition (see,
n
= Σki=1 di li to the z-class and it follows that S and T
Remark 2.4) π : m
have the same data (i) − (iv).
This completes the proof.
3.4. The minimal polynomial is (x + 1)d .
Note that, two isometries S and T are conjugate if and only if −S and
−T are conjugate. Now, suppose T is an isometry with minimal polynomial
(x − 1)d . Then −T : V → V is also an isometry, and m−T (x) = (x + 1)d .
Conversely, if T is unipotent, then −T has minimal polynomial (x + 1)d .
Thus this case is reduced to the unipotent case, and the parametrization of
the conjugacy and the z-classes of T are similar to that of −T .
3.5. The minimal polynomial is a product of pairwise dual polynomials. Suppose T : V → V is an element in I(V, B) such that mT (x) =
q(x)d q ∗ (x)d , where q(x), q ∗ (x) are irreducible polynomials over F of degree
10
m and are dual to each-other. For our purpose, it is enough to consider
the case when T is semisimple. So assume, d = 1. Let Vq = ker q(T ),
Vq∗ = ker q ∗ (T ). We have
(3.7)
V = Vq + Vq ∗ ,
and B|Vq = 0 = B|Vq∗ , dim Vq = dim Vq∗ . Since B is non-degenerate, we
can choose a basis {e1 , ...., em , f1 , ..., fm } of V such that for each i, ei ∈ Vq ,
fi ∈ Vq∗ , and for all i, j = 1, . . . m,
B(ei , ei ) = 0 = B(fi , fi ), B(ei , fj ) = δij or − δij .
For each w∗ ∈ Vq∗ , define the linear map w∗ : v → B(v, w). These maps
enable us to identify Vq∗ with the dual of Vq . Thus (V, B) is a standard
space, see, [10, Section-2.1], and T = TL + TL∗ , where TL , the restriction of
T to Vq , is an element of GL(Vq ). Conversely, given an element in GL(Vq ),
it can be extended to an isometry of (V, B). Hence the conjugacy classes in
I(V, B) are parametrized by the usual structure theory of linear maps.
Define a form HT on V as follows: For u, v ∈ V, HT (u, v) = B(T u, v).
Clearly, if S in I(V, B) commutes with T , then
HT (Su, Sv) = B(T Su, Sv) = B(ST u, Sv) = B(T u, v) = HT (u, v).
Conversely, if S preserves HT , then HT (Su, Sv) = HT (u, v) implies that for
u, v ∈ V, B(ST u, Sv) = B(T Su, Sv). By the non-degeneracy of B, it follows
that S commutes with T .
Now, suppose E is the splitting field of q(x) (hence of q ∗ (x) also). Let
α1 , · · · , αk be distinct roots of q(x) in E. There is a unique automorphism
e 7→ ē which maps αi → α−1
i . Further V over E has a decomposition into
eigenspaces:
k
M
(Vαi + Vα−1 ).
V=
i
i=1
Without loss of generality, assume V = Vα + Vα−1 . Then HT defines an
E-valued Hermitian form on V: when u ∈ Vα and v ∈ Vα−1 , we have
HT (u, v) = HT (v, u). Thus, Z(T ) = U(V, HT ). We have now proved the
following lemma.
Lemma 3.8. Let dim V be even. Let T be a semisimple element in I(V, B)
such that mT (x) = q(x)q ∗ (x), where q(x), q ∗ (x) are irreducible polynomials
over F and they are dual to each-other. Let E be the splitting field of the
minimal polynomial of T . Then the z-class of T is determined by
(i) The degree of q(x).
(ii) Equivalence class of E-valued Hermitian forms HT on V.
11
4. Classification over a perfect field
Let F be a perfect field of characteristic different from two. The group of
isometries I(V, B) consists of F-points of a linear algebraic group defined over
F. Thus each T in I(V, B) has the Jordan decomposition T = Ts Tu , where
Ts is semisimple (that is, every Ts -invariant subspace has a Ts -invariant
complement) and Tu is unipotent. Moreover Ts , Tu are also elements of
I(V, B), they commute with each other, and they are polynomials in T (see,
[12, Chapter 15]). Moreover we have Z(T ) = Z(Ts )∩Z(Tu ). To some extent,
the Jordan decomposition reduces the study of conjugacy and z-classes in
I(V, B) to the study of conjugacy and z-classes of semisimple and unipotent
elements. Suppose T : V → V is a semisimple isometry with prime and
self-dual minimal polynomial. Suppose E = F[x]/(p(x)). Then E is a finite
simple field extension of F, [E : F] = degree of p(x). Thus the underlying
cyclic algebras in Theorem 3.7 (or, Theorem 3.8) are isomorphic to the field
E, and the Hermitian forms Hdi are E-valued.
Now suppose T is an arbitrary semisimple isometry, and let its minimal
polynomial be a product of pairwise distinct prime polynomials over F. Let
mT (x) = (x − 1)e (x + 1)f Πki=1 pi (x) Πlj=1 qj (x)qj∗ (x),
where e, f = 0 or 1, p1 (x), ..., pk (x) are self-dual, and for j = 1, 2, ..., l, qj (x)
is dual to qj∗ (x). Suppose for each i, the degree of pi (x) is 2mi , and for each
j, degree of qj (x) is lj . Let the characteristic polynomial of T be
χT (x) = (x − 1)l (x + 1)m Πki=1 pi (x)di Πlj=1 qj (x)ej qj∗ (x)ej .
The primary decomposition of V with respect to T is determined by the minimal and the characteristic polynomial of T . We get the following orthogonal
decomposition of V into T -invariant subspaces:
M
(4.1)
V = V1 ⊕ V−1 ⊕ki=1 Vi
⊕lj=1 (Wj + W∗j ),
where V1 = ker (T −I)l , V−1 = ker (T +I)m , for each i = 1, ..., k, Vi = ker pi (T ),
and for each j = 1, 2, ..., l, Wj = ker qj (T ), W∗j = ker qj∗ (T ). We have,
dim Vi = 2mi di , and dim Wj = lj ej . Let Ei be the field isomorphic to
F[x]/(pi (x)), and let Kj be the field isomorphic to F[x]/(qj (x)). As a vector
space over Ei , Vi is the direct sum of di copies of Ei . The z-class of T is
determined by the z-classes of the restrictions of T to each component in
the primary decomposition (4.1). Note that T |V1 = I, T |V−1 = −I. Since I
and −I belong to the center of the group, the z-class of T restricted to V1
or V−1 is determined by dim V1 or dim V−1 . Now, the following theorem
follows from Theorem 3.7 and Lemma 3.8.
12
Theorem 4.1. Suppose F is perfect. Let T : V → V be a semisimple element
in I(V, B). The z-class of T is determined by
(i) A finite sequence of integers (l, m; m1 , ..., mk1 ; l1 , ..., lk2 ).
2
1
lj e j .
mi di + 2Σkj=1
(ii) A partition of n, π : n = l + m + 2Σki=1
(iii) Field extensions Ei , 1 ≤ i ≤ k1 of F, [Ei : F] = 2mi , and Kj ,
1 ≤ j ≤ k2 , [Kj : F] = lj .
(iv) Equivalence classes of Ei -valued Hermitian forms Hi , 1 ≤ i ≤ k1 ,
and Kj -valued Hermitian forms Hj′ , 1 ≤ j ≤ k2 .
5. Finiteness of the z-classes: Proof of Theorem 1.1
If there are only finitely many z-classes of semisimple and unipotent elements, it follows from the Jordan decomposition that there are only finitely
many z-classes. So it is enough to show the finiteness of z-classes of semisimple and unipotent elements respectively.
Suppose F is a perfect field that has only finitely many field extensions
of degree at most n. Then the number of distinct equivalence classes of
quadratic forms of rank at most n is finite. Hence the number of equivalence
classes of Hermitian forms of rank at most n over an extension field of F is
finite. Combining this fact with Theorem 4.1, and the fact that there are
only finitely many partitions of n, we have that there are only finitely many
z-classes of semisimple elements.
Similarly, it follows from Theorem 3.6 that there are only finitely many
conjugacy classes of unipotent elements; hence there are only finitely many
z-classes of unipotent elements.
This completes the proof of Theorem 1.1.
References
[1] T. Asai, The conjugacy classes in the unitary, symplectic and orthogonal groups over
an algebraic number field, J. Math. Kyoto. Univ, 16-2 (1976), 325 - 350.
[2] A. Borel and J. de Siebenthal, Les sous-groupes de rang maximum des groupes de
Lie clos (French), Comment. Math. Helv. 23 (1949), 200 - 221.
[3] N. Burgoyne and R. Cushman, Conjugacy classes in linear groups, J. Algebra 44
(1977), 339 - 362.
[4] W. S. Cao and K. Gongopadhyay, Algebraic characterization of isometries of the
complex and the quaternionic hyperbolic planes, Geom. Dedicata, Volume 157, Number 1 (2012), 23 - 39.
[5] W. S. Cao and K. Gongopadhyay, Commuting isometries of the complex hyperbolic
space, Proc. Amer. Math. Soc. 139 (2011), 3317 - 3326.
[6] K. Gongopadhyay, Algebraic characterization of the isometries of the hyperbolic 5space, Geom. Dedicata, Volume 144, Number 1 (2010), Page 157 - 170.
13
[7] K. Gongopadhyay, Algebraic characterization of isometries of the hyperbolic 4-space,
ISRN Geometry (2012), Article ID 757489, doi: 10.5402/2012/757489.
[8] K. Gongopadhyay, The z-classes of quaternionic hyperbolic isometries, to appear in
J. Group Theory.
[9] K. Gongopadhyay and R. S. Kulkarni, z-Classes of isometries of the hyperbolic space,
Conform. Geom. Dyn. 13 (2009), 91 - 109.
[10] K. Gongopadhyay and R. S. Kulkarni, On the existance of invariant non-degenerate
bilinear form under a linear map. Linear Algebra Appl. 434 (2011), Issue 1, 89 - 103.
[11] R. Gouraige, z-classes of elements in central simple algebras, Thesis, City University
of New York (2006).
[12] J. E. Humphreys, Linear Algebraic Groups, Graduate Texts in Mathematics, No. 21.
Springer-Verlag, New York-Heidelberg, 1975.
[13] C. Kiehm, Equivalence of bilinear forms, J. Algebra 31 (1974), 45 - 66.
[14] R. S. Kulkarni, Dynamics of linear and affine maps, Asian J. Math. 12 (2008), no.
3, 321 - 344.
[15] R. S. Kulkarni, Dynamical types and conjugacy classes of centralizers in groups, J.
Ramanujan Math. Soc. 22 (2007), no. 1, 35 - 56.
[16] J. Milnor, On isometries of inner product spaces. Invent. Math. 8 (1969), 83 - 97.
[17] A. Singh, Conjugacy Classes of Centralizers in G2 , J. Ramanujan Math. Soc. 23
(2008), no. 4, 327 - 336.
[18] T. A. Springer and R. Steinberg, Conjugacy classes, Seminar on Algebraic Groups
and Related Finite Groups (Princeton 1968/69), 167 - 266, Lecture Notes in Mathematics 131, Springer-Verlag (1970).
[19] R. Steinberg, Conjugacy Classes in Algebraic Groups, notes by V. Deodhar, Lecture
Notes in Mathematics 366, Springer-Verlag (1974).
[20] G. E. Wall, On the conjugacy classes in the unitary, symplectic and orthogonal
groups, J. Austral. Math. Soc. 3 (1963), 1 - 63.
[21] H. Weyl, Theorie der Darstellung kontinuierlicher halb-einfacher Gruppen durch lineare Transformationen I, II, III and Nachtrag (German), I: Math Z. 23 (1925), no.
1, 271 - 309, II: Math. Z. 24 (1926), no. 1, 328 - 376, III: Math. Z. 24 (1926), no. 1,
377 - 395, Nachtrag: Math. Z. 24 (1926), no. 1, 789 - 791.
[22] J. Williamson, Normal matrices over an arbitrary field of characterrstrc zero, Amer.
J. Math 61 (1939), 335 - 356.
Department of Mathematical Sciences, Indian Institute of Science Education
and Research (IISER) Mohali, Knowledge city, Sector 81, S.A.S. Nagar, P.O.
Manauli 140306, India
E-mail address: krishnendu@iisermohali.ac.in, krishnendug@gmail.com
Bhaskaracharya Pratishthana, 56/14, Erandavane, Damle Path, Off Law College Road, Pune 411 004, India
E-mail address: punekulk@gmail.com
14