Dimension of Lie groups

Yan Gobeil

March 2017

We show how to find the dimension of the most common Lie groups (number of free real parameters
in a generic matrix in the group) and we discuss the agreement with their algebras.

1 Orthogonal groups

1.1 O(n) and SO(n)

The group O(n) is composed of n × n real matrices that are orthogonal, so that satisfy OT O = I. In
general a n × n matrix has n2 elements, but the constraint of orthogonality adds some relation between them
and decreases the number of independent elements. To find exactly by how much the number of elements is
reduced, we have to note that the constraint equation satisfies (OT O)T = OT O so that it is symmetric and
has only n(n+1)
2 independent components (n on the diagonal and n(n−1) 2 on the upper triangle). Equating
n(n+1)
this matrix to the identity then creates 2 constraints that involve the elements of O and it reduces the
number of independent elements to

n(n + 1) n(n − 1)
dim[O(n)] = n2 − = .
2 2

To consider the group SO(n), we add the requirement that the matrix has a determinant of 1. This seems
to be an additional constraint, but in fact taking the determinant of the orthogonality constraint tells us
that det(O)2 = 1 ⇒ det(O) = ±1. Asking for unit determinant is then just some choice of sign and doesn’t
add an extra constraint, which leads to

n(n − 1)
dim[SO(n)] = dim[O(n)] = .
2

 
a b
As an example, consider the 2 × 2 matrix O = . The orthogonality constraint is written as
c d
  2
a + c2 ab + cd
    
a c a b 1 0
OT O = = =
b d c d ab + cd b2 + d2 0 1

1 = a2 + c2
⇒ 0 = ab + cd .
1 = b2 + d2
There are four unknowns and 3 equations so there is one free variable, which agrees with the formula for the
dimension of O(2). If we solve this system and compute det(O) = ad − bc, we will get that it is either +1

1
or −1. Working with SO(2) then just means to ignore the solutions which have a determinant of −1, so it
doesn’t reduce the number of free elements. Actually a matrix in O(2) can be written as
 
cos θ sin θ
O=
∓ sin θ ± cos θ

and to get an element of SO(2) we pick the upper sign.

1.2 o(n) and so(n)

When we talk about Lie algebras, we ask for matrices that, when exponentiated, give an element of a Lie
group. For the Lie algebra o(n), we are looking for matrices A such that eA (defined by the series expansion
of the exponential) is orthogonal. Note that the orthogonality condition can be rewritten as O−1 = OT . It’s
P∞ m 2
easy to see from the expansion eA = m=0 Am! = 1 + A + A2 + ... that the inverse of eA is e−A and its
T
transpose is eA . This means that for O to be orthogonal we need A to be anti-symmetric AT = −A.
Let’s check the number of independent parameters in an anti-symmetric n×n matrix to make sure that it
agrees with the dimension of O(n). The first condition coming from the anti-symmetry is that the diagonal
elements must vanish because Aaa = −Aaa ⇒ Aaa = 0. Next, the elements in the lower triangle are just
2
the negative of the ones in the upper triangle, so we are left with only n 2−n independent elements, meaning
that
n(n − 1)
dim[o(n)] = ,
2
the same as for O(n).
To study the algebra of SO(n), we need to remember (or discover) the identity

det(eA ) = eT r A .

This is easy to see for a diagonal matrix since ediag(a1 ,...,an ) = diag(ea1 , ..., ean ). The case of a general matrix
follows since we can just find a basis in which it is diagonal. Using this, the requirement that the matrix
O has determinant 1 is equivalent to asking that the matrix A has a trace of zero. This should be a new
requirement for a matrix in so(n), but actually an anti-symmetric matrix is already traceless so there is no
additional constraint and
n(n − 1)
dim[so(n)] = dim[o(n)] = .
2

It’s interesting to note that we don’t see the difference between SO(n) and O(n) at the level of the
algebra. This can be understood from the fact that the part of O(n) with a negative determinant can’t be
expressed as an exponential because eA must have a positive determinant, so the only part that is present
in the algebra o(n) is exactly the one that corresponds to so(n).

2 Unitary groups

2.1 U (n) and SU (n)

To consider the group U (n), we look at complex n × n matrices that are unitary, so they satisfy U † U = I
(remember that the hermitian conjugate of a matrix is the transpose of its complex conjugate). Originally a
complex matrix has n2 complex elements, which means that it has 2n2 free real parameters. The constraints
are coming from the unitarity condition, which is in fact hermitian since (U † U )† = U † U . This means that
we need to know the number of real parameters in a hermitian matrix and substract that from 2n2 . The
diagonal of a hermitian matrix must be real since it must be equal to its complex conjugate. This gives n
parameters. Hermiticity fixes the elements on the lower triangle in terms of the ones in the upper triangle

2
 2
since they must be complex conjugates of each other. This leaves us with n 2−n complex parameters on the
upper triangle, which means n(n − 1) real ones. Overall this sums to n + n(n − 1) = n2 real parameters
in a hermitian matrix, which is then the number of constraints on the matrix U coming from the unitarity
constraint. The number of real parameters in a general unitary matrix is then found to be

dim[U (n)] = 2n2 − n2 = n2 .

When we talk about the group SU (n), we ask that the determinant of the matrix be 1. Taking the
determinant of the hermiticity condition gives |det(U )|2 = 1 ⇒ det(U ) = eiθ . There is still a continuum of
possible values for the determinant of a unitary matrix, so fixing it adds a new constraint. Basically we ask
for θ = 0, which can be expressed in terms of the elements of U if we want. The result is then

dim[SU (n)] = dim[U (n)] − 1 = n2 − 1 .

For an example,

2.2 u(n) and su(n)

Like before we want to write an element of the Lie group as eH . The unitarity requirement can be
†
rewritten as U −1 = U † and it’s easy to see again that (eH )−1 = e−H and (eH )† = eH so that unitarity asks
for the elements of the Lie algebra to be anti-hermitian matrices satisfying H † = −H. To see how many free
real parameters there are in such a matrix, first note that the elements of the diagonal must be imaginary
because they are equal to minus their complex conjugate. This gives n parameters. The elements of the
lower triangle are fixed as minus the complex conjugate of the elements of the upper triangle so there are
2
only n 2−n other complex parameters, which gives n(n − 1) real ones. The total is then

dim[u(n)] = n + n(n − 1) = n2 ,

agreeing with U (n).

3 Linear groups

3.1 SL(n, R) and SL(n, C)

The groups of the SL(n, .) family are simply general non-degenerate matrices (non-zero determinant)
that have unit determinant. For SL(n, R) we consider real matrices, which start with n2 parameters. The
condition that the determinant is 1 is the only constraints so that we get

dim[SL(n, R)] = n2 − 1 .

For SL(n, C) we care about complex matrices so we start with 2n2 real parameters. The unit determinant
condition actually gives 2 constraints since the determinant could be any complex number and we fix its real
and imaginary parts. The result is
dim[SL(n, C)] = 2n2 − 2 .

3
3.2 sl(n, R) and sl(n, C)

4 Summary
Name of the group Properties Dimension Algebra
n(n−1)
O(n) OT O = I 2 anti-symmetric matrices
T n(n−1)
SO(n) O O = I and det(O) = 1 2 anti-symmetric matrices
U (n) U †U = I n 2
anti-hermitian matrices
†
SU (n) U U = I and det(U ) = 1 n2 − 1 traceless anti-hermitian matrices
SL(n, R) det(L) = 1 n2 − 1 traceless real matrices
SL(n, C) det(L) = 1 2(n2 − 1) traceless complex matrices