CLASSICAL GROUPS

DAVID VOGAN

1. Orthogonal groups
These notes are about “classical groups.” That term is used in various
ways by various people; I’ll try to say a little about that as I go along. Basi-
cally these are groups of matrices with entries in fields or division algebras.
To warm up, I’ll recall a definition of the orthogonal group.
Definition 1.1. Suppose n ≥ 1 is an integer. The real orthogonal group
O(n) is
O(n) = all n × n real matrices g such that t gg = I .


Here if A is an m × n matrix, t A is the n × n matrix whose (j, i) entry is

equal to the (i, j) entry of A.
Here is another way to formulate the definition. Write
(1.2a) Rn = {column vectors of size n} = {n × 1 matrices}
and identify
(1.2b) Hom(Rn , Rm ) ' {m × n real matrices A} =def Mm×n (R)
by matrix multiplication on the left. For the (most important) case of square
matrices, we will write
(1.2c) Mn (R) =def Mn×n (R);
this is a real algebra under matrix multiplication, and a real Lie algebra
under commutator of matrices:
(1.2d) [A, B] = AB − BA (A, B ∈ Mn (R).
Write
(1.2e) (, ) = (, )n , (v, w) = t v · w, (v, w ∈ Rn )
for the standard inner product on Rn . Then
(Av, w)m = v, t Aw n (v ∈ Rn , w ∈ Rm , A ∈ Mm×n (R).


(1.2f)
With this notation, it is very easy to check that
(1.3) O(n) = {linear transformations of Rn preserving (, )n } .
1
2 DAVID VOGAN

The orthogonal group is the first classical group. The zeroth classical
group is
GL(n, R) = {all invertible n × n matrices}
(1.4)
= {invertible linear transformations of Rn } .
This is a Lie group or a real algebraic group; from either of these perspec-
tives, we compute
Lie(GL(n, R)) =def gl(n, R) = Mn (R),
(1.5)
dimR (GL(n, R)) = n2 ;
the Lie algebra structure is the commutator of matrices defined above.
The descriptions above make it clear that O(n) is a (closed) Lie subgroup
of GL(n, R), and a real algebraic subgroup. We can compute
Lie(O(n)) =def o(n) = {X ∈ Mn (R) | t X = −X},
(1.6)
dimR (O(n)) = n(n − 1)/2.
The goal of these notes is to write down descriptions like Definition 1.1,
(1.3), and (1.6) of other classical groups.
Already in this case, one can see where ambiguity might arise about what
constitutes a classical group. There is a short exact sequence (recall that
n ≥ 1)
det
(1.7) 1 → SO(n) −→ O(n) −→ {±1} → 1
which is the definition of the special orthogonal group SO(n). It is the
identity component of O(n), and therefore has the same dimension and the
same Lie algebra.
Because there are lots of nice theorems about connected compact Lie
groups, some people prefer SO(n) to O(n), and like to call SO(n) a classical
group. Others are interested in simply connected groups, or only in the Lie
algebra, and so like to call the double cover Spin(n) of SO(n) a classical
group. But there are some subtle theorems about O(n) that actually fail for
SO(n). An example is the fact that O(n) acts irreducibly on the space of
complex harmonic polynomials of degree m whenever this space is not zero.
There are similar difficulties with Spin(n). I will therefore take the second
point of view and regard O(n) as the only compact classical orthogonal
group.

2. Unitary groups
To get another family of groups “like” the orthogonal groups, we can
replace the field R by C. Write
Cn = {complex column vectors of size n}
(2.1a)
= {n × 1 complex matrices)}
 CLASSICAL GROUPS 3

and identify
(2.1b) HomC (Cn , Cm ) ' {m × n complex matrices A} =def Mm×n (C)
by matrix multiplication on the left. For square matrices, we write
(2.1c) Mn (C) =def Mn×n (C);
this is a complex algebra under matrix multiplication, and a complex Lie
algebra under commutator of matrices:
(2.1d) [A, B] = AB − BA (A, B ∈ Mn (C).
In this setting we have a complex Lie group, or complex algebraic group
GL(n, C) = {all invertible n × n complex matrices}
(2.1e)
= {invertible linear transformations of Cn } .
We compute
Lie(GL(n, C)) =def gl(n, C) = Mn (C),
(2.1f)
dimC (GL(n, C)) = n2 ;
the Lie algebra structure is the commutator of matrices. The ordinary real
Lie algebra is the same vector space with the complex structure forgotten;
so
(2.1g) dimR (GL(n, C)) = 2n2 .
Write
(2.1h) h, i = h, in , hv, wi = t v · w, (v, w ∈ Cn )
for the standard Hermitian form on Cn . Then
(2.1i) hAv, wim = hv, t Awin (v ∈ Cn , w ∈ Cm , A ∈ Mm×n (C)).
It is convenient to define
(2.1j) Ah =def t A (A ∈ Mm×n (C)),
the Hermitian transpose of A.
Definition 2.2. Suppose n ≥ 1 is an integer. The unitary group U (n) is
n o
U (n) = all n × n complex matrices g such that g h g = I .
Equivalently,
U (n) = {linear transformations of Cn preserving h, in } .
It is clear from the definition that U (n) is a closed real Lie subgroup, or
a real algebraic subgroup, of GL(n, C). Because of the complex conjugates
appearing in the definition, U (n) is not a complex Lie subgroup or a complex
algebraic group. We calculate
(2.3a) Lie(U (n)) =def u(n) = {X ∈ Mn (C) | X h = −X},
the set of skew-Hermitian complex matrices. The condition on X is that
each diagonal entry is purely imaginary, and the entries above the diagonal
4 DAVID VOGAN

are the complex conjugates of the corresponding entries below the diagonal.
We calculate
(2.3b) dimR (U (n)) = n + 2(n(n − 1)/2) = n2 .
The group U (n) is a classical group. Also interesting (but not for us a clas-
sical group) is the special unitary group, defined by the short exact sequence
det
(2.3c) 1 → SU (n) −→ U (n) −→ U (1) → 1;
here U (1) is the circle group of complex numbers of absolute value 1. The
(connected) subgroup SU (n) is the commutator subgroup of U (n). We can
calculate
(2.3d) Lie(SU (n)) =def su(n) = {X ∈ Mn (C) | X h = −X, tr(X) = 0},
the Lie algebra of skew Hermitian matrices of trace zero. Consequently
(2.3e) dimR (SU (n)) = n2 − 1.

3. Quaternion groups
For the last family of compact classical groups, we replace the field by the
division algebra H. Write
Hn = {quaternion column vectors of size n}
(3.1a)
= {n × 1 quaternion matrices)} .
We regard Hn as a right vector space over H; scalar multiplication is on the
right, one coordinate at a time. Then and identify
(3.1b) HomH (Hn , Hm ) ' {m × n quaternion matrices A} =def Mm×n (H)
by matrix multiplication on the left. For square matrices, we write
(3.1c) Mn (C) =def Mn×n (C);
this is a real algebra under matrix multiplication, and a real Lie algebra
under commutator of matrices:
(3.1d) [A, B] = AB − BA (A, B ∈ Mn (H).
(There is no obvious notion of a Lie algebra over a division ring; square
matrices over a division ring form an algebra or a Lie algebra only over the
center of the division ring, which in this case is R.) In this setting we have
a real Lie group, or real algebraic group
GL(n, H) = {all invertible n × n quaternion matrices}
(3.1e)
= {invertible linear transformations of Hn } .
We compute
Lie(GL(n, H)) =def gl(n, H) = Mn (H),
(3.1f)
dimR (GL(n, H)) = 4n2 ;
the Lie algebra structure is the commutator of matrices.
 CLASSICAL GROUPS 5

Recall the definition of conjugation on H:

(3.1g) a + bi + cj + dk = a − bi − cj − dk, h1 h2 = h2 h1 .
That is, conjugation is an antiautomorphism of the real algebra H.
Write
(3.1h) h, i = h, in , hv, wi = t v · w, (v, w ∈ Hn )
for the standard Hermitian form on Hn . Then
(3.1i) hAv, wim = hv, t Awin (v ∈ Hn , w ∈ Cm , A ∈ Mm×n (C)).
It is convenient to define
(3.1j) Ah =def t A (A ∈ Mm×n (H)),
the Hermitian transpose of A.
Definition 3.2. Suppose n ≥ 1 is an integer. The quaternionic unitary
group Sp(n) is
n o
Sp(n) = all n × n quaternion matrices g such that g h g = I .
Equivalently,
Sp(n) = {linear transformations of Hn preserving h, in } .
Another natural and commonly used notation for this group is U (n, H).
It is clear from the definition that Sp(n) is a closed real Lie subgroup, or
a real algebraic subgroup, of GL(n, H).
(3.3a) Lie(Sp(n)) =def sp(n) = {X ∈ Mn (H) | X h = −X},
the set of skew-Hermitian quaternion matrices. The condition on X is that
each diagonal entry is purely imaginary (that is, has real part zero), and the
entries above the diagonal are the conjugates of the corresponding entries
below the diagonal. We calculate
(3.3b) dimR (Sp(n)) = 3n + 4(n(n − 1)/2) = 2n2 + n.
The group Sp(n) is a classical group. There is no H-valued “determinant”
for quaternionic matrices; all that can be defined is the real determinant
of the matrix regarded as a real-linear transformation of a 4n-dimensional
space. This determinant is automatically one on Sp(n), so there is no smaller
“quaternionic special unitary group.”

4. Natural inclusions
The inclusions R ⊂ C ⊂ H define group inclusions
(4.1) O(n) ,→ U (n) ,→ Sp(n),
just by regarding a matrix over a small field as a matrix over a larger field
(or division algebra). Similarly, the identifications H ' C2 ' R4 and C ' R2
define inclusions
(4.2) Sp(n) ,→ U (2n) ,→ O(4n), U (n) ,→ O(2n).
6 DAVID VOGAN

These inclusions actually take values in the determinant one subgroups:

(4.3) Sp(n) ,→ SU (2n) ,→ SO(4n), U (n) ,→ SO(2n).
In case n = 1, we find
(4.4a) dim Sp(1) = 2 · 12 + 1 = 3 = 22 − 1 = dim SU (2);
It follows that the first inclusion above is an isomorphism
 
a + bi −c − di
(4.4b) Sp(1) ' SU (2), a + bi + cj + dk 7→ .
c − di a − bi
Another way to write this isomorphism explicitly is
(4.4c)
Sp(1) = unit quaternions = {a + bi + cj + dk | a2 + b2 + c2 + d2 = 1} ' S 3 ,
the unit sphere in R4 . Similarly
  
z −w 2 2
(4.4d) SU (2) = | z, w ∈ C, |z| + |w| = 1 .
w z
The isomorphism of SU (2) with Sp(1) sends the complex numbers (z, w) to
the quaternion z + wj.