Classification of Three-Dimensional Real Lie

Algebras
Adam Bowers

29 April, 2005∗

1 Introduction
In this paper, we will classify (up to isomorphism) all real Lie algebras of
three-dimensions. To that end, we will need the following proposition.

Proposition 1 Let g1 and g2 be two Lie algebras of dimension n < ∞.

Suppose each has a basis with respect to which the structure constants are
the same. Then g1 and g2 are isomorphic.

Proof. By assumption, a basis {X1 , . . . , Xn } for g1 and a basis {Y1 , . . . , Yn }

for g2 may be chosen so that the structure constants are the same; that is,
n
X
[Xi , Xj ] = Cijk Xk for all i, j = 1, . . . , n,
k=1

and
n
X
[Yi , Yj ] = C̃ijk Yk for all i, j = 1, . . . , n,
k=1

and

Cijk = C̃ijk for all i, j, k = 1, . . . , n.

Define a linear map φ : g1 → g2 by mapping basis elements to basis

elements,
∗
Revised March 2010. Originally written as coursework in a class taught by Maria
Gordina at the University of Connecticut in Spring Semester 2005.

1
 φ(Xi ) = Yi for all i = 1, . . . , n,

and linearly extending to g1 . By construction, φ is a bijection. It only

remains to check that φ is a homomorphism.
By linearity, it suffices to check the homomorphism property on basis
elements. Let i, j ∈ {1, . . . , n}. Then
n n n
!
  X X X
φ [Xi , Xj ] = φ Cijk Xk = Cijk φ(Xk ) = Cijk Yk .
k=1 k=1 k=1

But the structure constants are the same, and so

n
X n
X
Cijk Yk = C̃ijk Yk = [Yi , Yj ] = [φ(Xi ), φ(Xj )].
k=1 k=1

Therefore, the map φ is indeed a homomorphism, and hence an isomor-

phism. 

Using this result, we can classify a given Lie algebra by finding a suitable
basis so that its Lie bracket multiplication table will be in a “canonical”
form. These canonical Lie algebras will be shown to be non-isomorphic by
considering certain invariants of isomorphic mappings. One such invariant
is the dimension of the derived algebra.

Definition 1 Let g be a Lie algebra. The derived algebra of g is the set

of all linear combinations of Lie brackets in g. It is denoted [g, g] or g(1) .

Claim 2 The derived algebra of the Lie algebra g is a subalgebra of g.

Proof. By definition, g(1) is a vector space. Let X, Y ∈ g(1) . Then

[X, Y ] ∈ [g, g] = g(1) ,

so g(1) is closed under the Lie bracket. 

Corollary 3 If two Lie algebras are isomorphic, then their derived algebras
have the same dimension.

This is the only use we will be making of derived algebras, but the idea
can be extended to something much more general and useful. For more

2
information on this topic (including some insight into the notation), see [4]
or [6].
The classification presented here (which parallels that of [2], but for real
Lie algebras instead of complex) makes heavy use of Corollary 3. We will
refer to the dimension of the derived algebra as the rank of the Lie bracket,
which is somewhat an abuse of terminology (see [2]).

2 Lie Algebras of 1 and 2 Dimensions

Although the purpose of this paper is to classify the Lie algebras of dimension
three, it will be helpful to begin with those of dimensions one and two.
First, suppose g is a one-dimensional Lie algebra. Then any basis will
consist of one element, e1 . Since any element of g is a multiple of e1 and
[e1 , e1 ] = 0, the rank of the Lie bracket is zero. We call such a Lie alge-
bra abelian. It is a trivial consequence of Proposition 1 that there is only
one abelian Lie algebra of any dimension. We denote the one-dimensional
abelian Lie algebra by L(1,0) (see Section 6).
Now suppose g is a two-dimensional Lie algebra. Let g have any basis,
say {X1 , X2 }. Then

[X1 , X2 ] = αX1 + βX2 for some α, β ∈ R.

By linearity and skew-symmetry, the non-zero structure constants are deter-

mined by α and β. Therefore the derived algebra can only have dimension
zero or one.
If α = 0 and β = 0 (i.e., the dimension of the derived algebra is zero),
then g is the unique two-dimensional abelian Lie algebra, denoted by L(2,0).
Suppose α2 + β 2 6= 0 (so the derived algebra is one-dimensional). We will
show that g has a basis {e1 , e2 } such that [e1 , e2 ] = e1 . It will follow from
Proposition 1 that any two such Lie algebras are isomorphic, thus completing
the classification of two-dimensional Lie algebras.
At least one of α, β is not zero. If α 6= 0, then let
β 1
e1 = X1 + X2 and e2 = X2 .
α α
This is a basis for g, and
h i
[e1 , e2 ] = X1 + αβ X2 , α1 X2 = 1
α [X1 , X2 ]

1
= α (αX1 + βX2 ) = e1 .

3
Therefore, [e1 , e2 ] = e1 , as required.
Now suppose α = 0. Then β 6= 0, so let
1
e1 = X2 and e2 = − X1 .
β
This is a basis for g and
h i
[e1 , e2 ] = X2 , − β1 X1 = 1
β [X1 , X2 ]

1
= β (βX2 ) = e1 .

So, once again, [e1 , e2 ] = e1 , as required.

As claimed, there is only one non-abelian two-dimensional Lie algebra.
This algebra will be called L(2,1).

3 Dimension 3, Ranks 0 and 1

Now we turn our attention to Lie algebras of three dimensions, the main
topic of this paper. As indicated above, we will be considering the dimension
of the derived algebra as a starting point. In this section, we consider the
cases when the derived algebra is of dimension zero or one.
If the dimension of the derived algebra is zero, then the Lie algebra in
question is abelian. As before, it is a trivial consequence of Proposition 1
that there is only one abelian Lie algebra of dimension three. This Lie
algebra will be called L(3,0).
Now suppose the derived algebra has dimension 1. Then there exits
some non-zero X1 ∈ g such that g(1) = span{X1 }. Extend this to a basis
{X1 , X2 , X3 } for g. Then there exist scalars α, β, γ ∈ R (not all zero) such
that

[X1 , X2 ] = αX1 ,
[X1 , X3 ] = βX1 ,
[X2 , X3 ] = γX1 .

Suppose α 6= 0. Then construct a new basis, as follows:

e1 = X1 ,
1
e2 = X2 ,
α
e3 = αX3 − βX2 + γX1 .

4
Since α 6= 0, by assumption, this is a basis for the Lie algebra g. Let us
calculate the Lie brackets for this basis:
[e1 , e2 ] = X1 , α1 X2 = X1 = e1 ,
 

[e1 , e3 ] = [X1 , αX3 − βX2 + γX1 ] = αβX1 − βαX1 = 0,

1 
[e2 , e3 ] = α X2 , αX 3 − βX2 + γX1 = γX1 − γX1 = 0.

This Lie algebra is seen to be the direct sum of two Lie algebras, L(2,1) ⊕
L(1,0), and will be denoted L(3,−1). (The minus indicates the Lie algebra
can be decomposed as a direct sum of other Lie algebras.)
If β 6= 0, then we can switch the labels of X2 and X3 and we will have
the same Lie brackets as before, so we might as well assume α = 0, β = 0,
and γ 6= 0. In this case let e1 = X2 , e2 = X3 , and e3 = γX1 . Then the Lie
brackets are:

[e1 , e2 ] = [X2 , X3 ] = γX1 = e3 ,

[e1 , e3 ] = [X2 , γX1 ] = 0,

[e2 , e3 ] = [X3 , γX1 ] = 0.

This Lie algebra will be denoted L(3,1).

4 Dimension 3, Rank 2
In the case when the dimension of g(1) is two, we can find a basis {Y, Z}
for the derived algebra, and extend this to a basis for g, say {X, Y, Z}. By
assumption,
[Y, Z] = αY + βZ
for some scalars α, β ∈ R.
Consider the adjoint map, ad : g → gl(g), defined by

adX 0 (Y 0 ) = [X 0 , Y 0 ], X 0 , Y 0 ∈ g.

For any fixed X 0 ∈ g, the map adX 0 takes values only in g(1) , so the restriction
of adX 0 to the derived algebra is well-defined. In particular, we want to
consider
adY : g(1) → g(1) ,
where Y is the element in our basis, chosen above.

5
 With respect to our chosen basis for g(1) , {Y, Z}, the matrix associated
with adY (restricted to g(1) ) is as follows:
 
0 α
[adY ]{Y,Z} = .
0 β
 
Claim 4 For any Y ∈ g(1) , trace adY = 0.

Proof. By assumption, there exist Y1 , Y2 ∈ g such that [Y1 , Y2 ] = Y . Thus

adY = ad[Y1 ,Y2 ] .

But
ad[Y1 ,Y2 ] = [adY1 , adY2 ] = adY1 adY2 − adY2 adY1 ,
and so
     
trace adY = trace adY1 adY2 − trace adY2 adY1 = 0.

This proves the claim. 

Applying the above claim to

 
0 α
[adY ]{Y,Z} = ,
0 β

we see that β = 0. Similarly,

 
−α 0
[adZ ]{Y,Z} = ,
−β 0

so that, again using the claim, α = 0.

Having shown that both α = 0 and β = 0, we see that [Y, Z] = 0.
But the derived algebra is two dimensional, so the image of adX must be
two-dimensional. Therefore,

adX : g(1) → g(1)

must be an isomorphism. It is the properties of this isomorphism that we

will exploit for our classification.

6
Proposition 5 Let g1 and g2 be two isomorphic Lie algebras of dimension
3, with Lie brackets of rank 2. Let X1 ∈ g1 and X2 ∈ g2 be chosen as in the
preceding paragraph. The matrices representing adX1 and adX2 restricted to
the derived algebras are similar matrices.

Proof. Since the two Lie algebras are isomorphic, we can find a change of
basis so that they have the same Lie brackets, hence the same ad matrices.
A change of basis is represented by conjugation of the ad matrix, so the
matrices are similar. 

Because adX is an isomorphism, it must have two non-zero eigenvalues.

Since similar matrices have the same eigenvalues, we can, using Proposi-
tion 5, distinguish between three separate cases:

1. adX has two distinct real eigenvalues, or

2. adX has one real eigenvalue of multiplicity two, or

3. adX has two complex conjugate eigenvalues.

For the final item in this list, we remark that, if adX has one complex
eigenvalue λ, then the complex conjugate λ is also an eigenvalue. This
is clear if we consider the characteristic polynomial, P (x) for adX . Any
eigenvalue of adX will be a zero of the characteristic polynomial, so

P (λ) = P (λ) = 0 = 0.

Case 1: Two distinct real eigenvalues.

Suppose adX has two distinct real eigenvalues, λ1 and λ2 . Then adX is
diagonalizable over R, so
 
λ1 0
[adX ]{Y,Z} ∼ .
0 λ2

Therefore, there exits a basis of eigenvectors for g(1) , call it {e1 , e2 }, such
that
[X, e1 ] = λ1 e1 , [X, e2 ] = λ2 e2 , [e1 , e2 ] = 0.
1 λ2
Let e3 = − X and let x = . Then
λ1 λ1
[e1 , e3 ] = e1 , [e2 , e3 ] = xe2 , [e1 , e2 ] = 0. (1)

7
 The structure equations in (1) describe a continuously varying family
of Lie algebras, which will be called L(3,2,x), where x is the parameter
appearing in (1). Observe that x 6= 0, because neither of the eigenvalues
was 0.

Theorem 6 Two Lie algebras L(3,2,x) and L(3,2,x0 ) corresponding to two

different scalars in the structure equations in (1) are isomorphic if and only
if either x = x0 or x = 1/x0

Proof. I will prove only that L(3,2,x) ∼ = L(3,2,1/x). The other direction,
while not difficult, is a bit technical and time consuming.
Consider the Lie algebra L(3,2,1/x) with basis {e1 , e2 , e3 } and structure
equations
1
[e1 , e3 ] = e1 , [e2 , e3 ] = e2 , [e1 , e2 ] = 0,
x
where x ∈ R is nonzero. Let

e˜1 = e2 , e˜2 = e1 , e˜3 = xe3 .

Then
[e˜1 , e˜3 ] = [e2 , xe3 ] = e2 = e˜1 ,

[e˜2 , e˜3 ] = [e1 , xe3 ] = xe1 = xe˜2 ,

[e˜1 , e˜2 ] = [e2 , e1 ] = 0.

This change of basis gives the Lie brackets for the Lie algebra L(3,2,x), as
required. 

Case 2: One real eigenvalue of multiplicity two.

Suppose adX has one real eigenvalue, λ, of multiplicity 2. If adX has
two distinct eigenvectors, then we can use the same argument as in the first
case and we end up with the Lie algebra L(3,2,x = 1). We may suppose,
then, that adX does not have two distinct eigenvectors. Then
 
λ 1
[adX ]{Y,Z} ∼ ,
0 λ

8
so there exists a basis for g(1) of so-called generalized eigenvectors, call it
{e1 , e2 }, such that

[X, e1 ] = λe1 , [X, e2 ] = e1 + λe2 , [e1 , e2 ] = 0.

1 1
Let e3 = − X and replace e1 with e1 . Then
λ λ
[e1 , e3 ] = e1 , [e2 , e3 ] = e1 + e2 , [e1 , e2 ] = 0. (2)

The structure equations in (2) describe a Lie algebra which will be called
L(3,3).
For more information on the theory of generalized eigenvectors and gen-
eralized eigenspaces, and how this leads to the Jordan form of a matrix, see
[1] for a delightful discussion.

Case 3: Two complex conjugate eigenvalues.

Suppose adX has two complex conjugate eigenvalues, λ and λ. Then
adX is diagonalizable over C, and so
 
λ 0
[adX ]{Y,Z} ∼ .
0 λ

Therefore, there exit vectors e1 , e2 ∈ g(1) such that

adX (e1 + ie2 ) = λ(e1 + ie2 ). (3)

Let λ = α + iβ for α, β ∈ R. Without loss of generality, we may assume

sign(β) = sign(α) (otherwise, we may replace λ with λ and have our desired
property). Observe that β 6= 0, since λ 6∈ R, by assumption.
By linearity of the adjoint map,

adX (e1 + ie2 ) = adX (e1 ) + iadX (e2 ).

Since

λ(e1 + ie2 ) = (α + iβ)e1 + i(α + iβ)e2

= (αe1 − βe2 ) + i(βe1 + αe2 ),

we can equate real and imaginary parts in (3), whence

adX (e1 ) = αe1 − βe2 , adX (e2 ) = βe1 + αe2 .

9
 1
Observe, if we replace X with β X, we have
α α
ad 1 X (e1 ) = e1 − e2 , ad 1 X (e2 ) = e1 + e2 .
β β β β
1 α
Let e3 = − X and define x = . Then
β β
[e1 , e3 ] = xe1 − e2 , [e2 , e3 ] = e1 + xe2 , [e1 , e2 ] = 0. (4)
The structure equations in (4) describe another continuously varying
family of Lie algebras, which will be called L(3,4,x), where x is the parameter
appearing in (4). Note, by construction, x ≥ 0.

Remark 1. The adjoint matrix from Case 1 was diagonalizable over R,

whereas the one in Case 3 was not. We could therefore conclude that they
would lead to non-isomorphic Lie algebras (by Proposition 5). If we were
considering complex Lie algebras instead of real ones, we could not make
such a distinction. In fact, the same arguments used in Case 1 would apply
to Case 3, so there would be only one family of Lie algebras instead of two.
Indeed, for a fixed x,
 
L(3, 4, x) ∼ x−i
= L 3, 2, x+i .

To see this isomorphism, construct a new basis for L(3,4,x) as follows:

e01 = e1 + ie2 ,
e02 = e1 − ie2 ,
1
e03 = e3 .
x+1

x−i
Note that x ∈ R, by construction, and so = 1.
x+i
For L(3,2,x), we saw that distinct values of x could lead to isomorphic
Lie algebras. The same is not true for L(3,4,x), a fact made precise in the
following theorem, given without proof.

Theorem 7 Two Lie algebras L(3,4,x) and L(3,4,x0 ) corresponding to two

different scalars in the structure equations in (4) are isomorphic if and only
if x = x0 .

Remark 2. If we had allowed the choice of complex eigenvalue with real

and imaginary parts of different sign, we would have ended up with x < 0.
In this case we would have L(3,4,−x) ∼ = L(3,4,x).

10
5 Dimension 3, Rank 3
Finally, we consider the case where the Lie bracket has full rank, that is,
the derived algebra is three-dimensional. For the moment, let the Lie al-
gebra g have as basis {X 0 , Y 0 , Z 0 }. Since the derived algebra g(1) is three-
dimensional, each bracket [X 0 , Y 0 ], [X 0 , Z 0 ], and [Y 0 , Z 0 ] must be non-zero.
Since each basis element appears twice, the adjoint maps adX 0 , adY 0 , and
adZ 0 have rank 2. Therefore, for any non-zero H ∈ g, the adjoint map adH
must also have rank 2. Fix such an H ∈ g. As in the previous section, we
will consider the eigenvalues of adH . Either

1. adH has two real eigenvalues, or

2. adH has two distinct complex conjugate eigenvalues.

We can be certain that these are the only possibilities, because, as we saw
before, if adH has one complex eigenvalue λ, then the complex conjugate λ
is also an eigenvalue.

Case 1: Two real eigenvalues.

We begin by making a claim.

Claim 8 There is an element in g such that the corresponding adjoint map

has a non-zero eigenvalue.

Proof. Let H be the element we picked in g before making the claim. If

adH does not have a non-zero eigenvalue, then, since it’s eigenvalues are
real, it is nilpotent, and so there is some basis B such that
 
0 α β
[adH ]B =  0 0 γ  .
0 0 0

Neither α nor γ is zero in this matrix, since the rank of adH is two. We see
that there is some vector H 0 such that adH (H 0 ) = αH. But then

adH 0 (H) = −αH,

and so H is an eigenvector for adH 0 with non-zero eigenvalue. 

By the above claim, we may assume, without loss of generality, that H is

such that adH has a non-zero eigenvalue. Let α be the non-zero eigenvalue

11
and let X be the corresponding eigenvector. Since g(1) = g, H is a commu-
tator, whence the trace of adH is zero (see Claim 4). Then adH must have
another eigenvector, Y , with eigenvalue −α. So we have

[H, X] = αX, [H, Y ] = −αY.

It only remains to find the bracket of X and Y . By the Jacobi identity,

[H, [X, Y ]] = − [X, [Y, H]] − [Y, [H, X]]

= − [X, αY ] − [Y, αX] = 0.

Since adH has rank two, ker(adH ) = span{H}. Therefore,

[X, Y ] = βH,

for some β 6= 0. We make a slight change of basis, for aesthetic purposes:

1 1
X0 = X, Y 0 = Y, H 0 = H.
αβ α
Now our structure equations are

[H 0 , X 0 ] = X 0 , [H 0 , Y 0 ] = −Y 0 , [X 0 , Y 0 ] = H 0 . (5)

Let
e1 = Y 0 , e2 = H 0 , e3 = 2X 0 .
Then we have our final structure equations,

[e1 , e2 ] = [Y 0 , H 0 ] = Y0 = e1 ,

[e1 , e3 ] = [Y 0 , 2X 0 ] = −2H 0 = −2e2 ,

[e2 , e3 ] = [H 0 , 2X 0 ] = 2X 0 = e3 .

This Lie algebra will be called L(3,5).

Case 2: Two distinct complex conjugate eigenvalues.

In this, our last case, the map we started with, adH , has two distinct
complex conjugate eigenvalues, λ and λ. By Claim 4, the trace of adH is
zero, and so the eigenvalues must be purely imaginary. Let λ = iα for some

12
non-zero α ∈ R. We may assume α > 0, since otherwise we can consider the
conjugate, λ. There exists some vector X + iY such that
adH (X + iY ) = iα(X + iY ).
Equating real and imaginary parts in this expression, we have
[H, X] = −αY, [H, Y ] = αX.
In a calculation similar to the one in the previous case, we find, using the
Jacobi identity,
[X, Y ] = βH,
for some β 6= 0.
Let
1 1 1
X0 = p X, Y 0 = − p Y, H 0 = H.
|αβ| |αβ| α
Now, our structure equations are
β 0
[H 0 , X 0 ] = Y 0 , [H 0 , Y 0 ] = −X 0 , [X 0 , Y 0 ] = − H. (6)
|β|
If β > 0 in (6), then [X 0 , Y 0 ] = −H 0 . Let
e1 = X 0 − H 0 , e2 = Y 0 , e3 = X 0 + H 0 .
Then
[e1 , e2 ] = [X 0 − H 0 , Y 0 ] = −H 0 + X 0 = e1 ,

[e1 , e3 ] = [X 0 − H 0 , X 0 + H 0 ] = −2Y 0 = −2e2 ,

[e2 , e3 ] = [Y 0 , X 0 + H 0 ] = H0 + X0 = e3 .
Checking back to the previous case, we see these are once again the structure
equations for the Lie algebra we called L(3,5).
Now suppose β < 0 in (6). Then [X 0 , Y 0 ] = H 0 . Let
e1 = X 0 , e 2 = Y 0 , e 3 = H 0 .
This time, our structure equations become
[e1 , e2 ] = [X 0 , Y 0 ] = H0 = e3 ,

[e1 , e3 ] = [X 0 , H 0 ] = −Y 0 = −e2 ,

[e2 , e3 ] = [Y 0 , H 0 ] = X0 = e1 .

13
This, our final Lie algebra, will be called L(3,6).

Remark 3. Consider the structure equations in (6) above. The choice

between β > 0 and β < 0 changed only the sign of the bracket [X 0 , Y 0 ].
This simple change in sign produced two different (non-isomorphic) real Lie
algebras. It may come as no surprise that these two Lie algebras, L(3,5) and
L(3,6), are isomorphic as complex Lie algebras. One can simply multiply
both X 0 and Y 0 by i in (6) and change the sign from negative to positive.

6 The Winternitz Classification

The choice of canonical basis for each Lie algebra in this paper, as well as the
name given to each Lie algebra, was taken from a classification scheme intro-
duced by Dr. Pavel Winternetz, of the Département de mathématiques et de
statistique, Université de Montréal, in the early 2000’s. To my knowledge,
his classification has never been published.1
Below is listed the Lie bracket multiplication table for each of the non-
abelian Lie algebras discussed in this paper.

Table 1: Nonabelian Lie Algebras of Dimensions 2, 3

e1 e2
L(2,1) e1 0 e1
e2 −e1 0

e1 e2 e3
e1 0 0 0
L(3,1)
e2 0 0 e1
e3 0 −e1 0

Continued on next page.

1
As of 2010, this classification still has not been published. However, we refer the
reader to [5] and [7].

14
 Table 1: Continued.

e1 e2 e3
e1 0 0 e1
L(3,2,x) 0 < |x| ≤ 1
e2 0 0 xe2
e3 −e1 −xe2 0

e1 e2 e3
e1 0 0 e1
L(3,3)
e2 0 0 e1 + e2
e3 −e1 −e1 − e2 0

e1 e2 e3
e1 0 0 xe1 − e2
L(3,4,x) x≥0
e2 0 0 e1 + xe2
e3 −xe1 + e2 −e1 − xe2 0

e1 e2 e3
e1 0 e1 −2e2
L(3,5)
e2 −e1 0 e3
e3 2e2 −e3 0

e1 e2 e3
e1 0 e3 −e2
L(3,6)
e2 −e3 0 e1
e3 e2 −e1 0

e1 e2 e3
e1 0 e1 0
L(3,-1) L(2,1) ⊕ L(1,1)
e2 −e1 0 0
e3 0 0 0

15
7 Examples
In this, the final section, we look at some examples of each nonabelian three-
dimensional Lie algebra isomorphism class.

7.-1 L(3,-1)
Consider the following basis for the vector space t(2) of 2×2 upper-triangular
real matrices:
     
0 1 −1 0 1 0
e1 = , e2 = , e3 = .
0 0 0 0 0 1
With this basis, the structure equations are
[e1 , e2 ] = e1 , [e1 , e3 ] = 0, [e2 , e3 ] = 0.

7.1 L(3,1)
For this example, we consider the vector space n(3) of 3 × 3 strictly upper-
triangular real matrices, which is the Lie algebra of the Heisenberg group.
If we use the basis
     
0 0 1 0 1 0 0 0 0
e1 =  0 0 0  , e2 =  0 0 0  , e3 =  0 0 1  ,
0 0 0 0 0 0 0 0 0
then the structure equations for n3 are
[e1 , e2 ] = 0, [e1 , e3 ] = 0, [e2 , e3 ] = e1 .

7.2 L(3,2,x = −1)

The Poincaré group, P(1,1) – sometimes denoted ISO(2) – is the group of
affine transformations of R2 which preserve the Lorentz metric (see [3]). It
is isomorphic to the group of 3 × 3 matrices of the form
 
A x
,
0 1

where A ∈ O(1, 1) and x ∈ R2 . The associated Lie algebra, p(1, 1) or iso(2),

has as a basis
     
0 0 1 0 0 1 0 1 0
e1 =  0 0 −1  , e2 =  0 0 1  , e3 =  1 0 0  .
0 0 0 0 0 0 0 0 0

16
The structure equations with respect to this basis are

[e1 , e2 ] = 0, [e1 , e3 ] = e1 , [e2 , e3 ] = −e2 .

7.3 L(3,3)
Consider the Lie algebra lg(3) with basis
     
0 0 1 0 0 1 0 1 0
e1 =  0 0 0  , e2 =  0 0 −1  , e3 =  0 0 1  .
0 0 0 0 0 0 0 0 1

The structure equations with respect to this basis are

[e1 , e2 ] = 0, [e1 , e3 ] = e1 , [e2 , e3 ] = e1 + e2 .

This Lie algebra is the so-called “lucky guess” algebra, so named because
the author does not know of any standard examples of this Lie algebra in
the literature and found this one by chance.

7.4 L(3,4,x = 0)
The Euclidean group, E(2), is the group of affine transformations of R2
which preserve the Euclidean metric (see [3]). It is isomorphic to the group
of 3 × 3 matrices of the form
 
A x
,
0 1

where A ∈ O(2) and x ∈ R2 . The associated Lie algebra, e(2), has as a basis
     
0 0 1 0 0 0 0 −1 0
e1 =  0 0 0  , e 2 =  0 0 1  , e 3 =  1 0 0  .
0 0 0 0 0 0 0 0 0

The structure equations with respect to this basis are

[e1 , e2 ] = 0, [e1 , e3 ] = −e2 , [e2 , e3 ] = e1 .

17
7.5 L(3,5)
For this example, we consider the well-known Lie algebra of the special linear
group, sl(2), the Lie algebra of 2 × 2 trace-free real matrices. Let sl(2) have
the following basis:
     
0 0 1 1 0 0 1
e1 = , e2 = , e3 = .
1 0 2 0 −1 0 0

With this basis, the structure equations are

[e1 , e2 ] = e1 , [e1 , e3 ] = −2e2 , [e2 , e3 ] = 0.

This is not the standard basis for sl(2), which is the primary example
of this Lie algebra. The reasons for this choice of basis are not clear to
the author; however, in the words of Professor Ian Anderson (the author’s
master’s thesis advisor) of Utah State University, it likely is “well understood
by those who understand it well.”

7.6 L(3,6)
Finally, we consider the Lie algebra of the special orthogonal group, so(3),
the Lie algebra of 3 × 3 trace-free, skew-symmetric real matrices. Let so(3)
have the following basis:
     
0 0 1 0 0 0 0 1 0
e1 =  0 0 0  , e2 =  0 0 −1  , e3 =  −1 0 0  .
−1 0 0 0 1 0 0 0 0

With this basis, the structure equations are

[e1 , e2 ] = e3 , [e1 , e3 ] = −e2 , [e2 , e3 ] = e1 .

Fin

References
[1] S. Friedberg, A. Insel, and L. Spence, Linear Algebra, Upper Saddle
River, New Jersey: Prentice-Hall, Inc., third ed., 1997.
[2] W. Fulton and J. Harris, Representation Theory: A First Course, Springer-
Verlag New York Inc., 1991.

18
[3] B. C. Hall, Lie Groups, Lie Algebras, and Representations: an elementary
introduction, Springer-Verlag New York Inc., 2003.
[4] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory,
Springer-Verlag New York Inc., 1972.
[5] J. Patera, R. T. Sharp, P. Winternitz, and H. Zassenhaus, Invariants
of real low dimension Lie algebras, J. Mathematical Phys., 17 (1976), pp. 986–
994.
[6] D. Sattinger and O. Weaver, Lie Groups and Algebras with Applications
to Physics, Geometry, and Mechanics, Springer-Verlag New York Inc., 1986.
[7] L. Snobl and P. Winternitz, Classification and identification of Lie alge-
bras. To be published.

19