Physics 251 Spring 2019

The Cartan-Killing Form

1. The Cartan metric tensor

We begin by introducing a metric tensor and a scalar product on a real Lie algebra g.
Suppose that X, Y ∈ g. We first define a linear operator, adX : g → g, such that
 
adX (Y ) = X, Y . (1)

Then, the scalar product of two elements of the Lie algebra, called the Killing form (also called
the Cartan-Killing form in the literature), is defined as


(X, Y ) = Tr adX adY . (2)

One can evaluate (X, Y ) explicitly by choosing a basis for the Lie algebra, {Ai }, which satisfies
the commutation relations, 
Ai , Aj ] = fijk Ak , (3)
where the structure constants fijk are real and i, j, k = 1, 2, . . . , dim g.1 Any element of X ∈ g
is a real linear combination of the basis vectors, Ak , i.e., X = xi Ai with xi ∈ R. With respect
to the basis, {Ai }, the matrix elements of the linear operator adAi are easily obtained by
noting that, 
adAi (Aj ) = Ai , Aj ] = fijk Ak , (4)
from which it follows that the matrix elements of adAi are given by,2

(adAi )k j = fijk . (5)


Hence, it follows that adX (Aj ) = [X, Aj ] = xi Ai , Aj ] = fijk xi Ak , which yields

(adX )k j = fijk xi . (6)

We can not compute the trace in eq. (2),

X j i ℓ
(X, Y ) = (adX )k j (adY )j k = fijk fℓk xy . (7)
jk

1
That is, the dimension of the Lie algebra g (denoted by dim g) is equal to the maximal number of linearly
independent basis vectors spanning the Lie algebra. P
2
Recall that using bra and ket notation, one can write T |ji = k |ki hk| T |ji. Equivalently, T |ji = T k j |ki
where the T k j ≡ hk| T |ji are the matrix elements of the operator T with respect to the basis {|ii} and there
is an implicit sum over the repeated index k.

1
 We now introduce the Cartan metric tensor of the Lie algebra, giℓ . For X = xi Ai and
Y = y i Ai , we define the metric tensor in terms of the scalar product in the usual way,

(X, Y ) = giℓ xi y ℓ . (8)

Comparing eqs. (7) and (8) then yields an explicit expression for the Cartan metric tensor,
j
giℓ = fijk fℓk . (9)

Equivalently, one can write,

giℓ = (Ai , Aℓ ) = Tr adAi , adAℓ , (10)

where the {Ak } are a basis for the Lie algebra g. Note that the metric tensor is a symmetric
covariant tensor, since
gℓi = fℓjk fikj = fℓk
j k
fij = giℓ , (11)
after relabeling j → k and k → j.
The expression for the Cartan metric tensor is basis-dependent. Recall that given a Lie
group G, the basis vectors of the corresponding Lie algebra g are defined by


∂A ~a(t) 
Ai = , (12)
∂ai t=0
where the coordinates of the Lie group element, A ∈ G are specified by ~a. The analytic curve,
~a(t), passes through the identity element at t = 0, which corresponds to the origin of the
coordinate system on the Lie group manifold. If we make a change of coordinates, ~a′ = ~a′ (~a)
on the Lie group manifold, then the basis vectors of g are changed to


′ ∂A ~a′ (t) 
Ai = , (13)
∂a′ i t=0
Employing the chain rule,


∂A ~a′ ∂A ~a′ (~a) ∂aj
= . (14)
∂a′ i ∂aj ∂a′ i
Setting t = 0, it follows that  
∂aj
A′i = Aj . (15)
∂a′ i t=0
Using
 ′ ′eq. (3)′ kand the corresponding commutation relations of the transformed basis vectors,
′
Ai , Aj ] = fij Ak , it follows that
 ′k ℓ m
′k ∂a ∂a ∂a
fij = fn . (16)
∂an ∂a′ i ∂a′ j t=0 ℓm

where ∂aℓ /∂a′ i is the inverse Jacobian defined by,

∂a′ i ∂aℓ
= δnℓ . (17)
∂an ∂a′ i
2
In particular, the structure constants change if one performs a linear transformation,

a′ j = M j k ak , (18)

where M is a nonsingular matrix (j labels the rows and k labels the columns). In this case, it
follows that
A′i = (M −1 )j i Aj , fij′ k = M k n (M −1 )ℓ i (M −1 )m j fℓm
n
. (19)
′j
Using eqs. (9) and (19), it follows that giℓ′ = fij′ k fℓk is related to giℓ as expected for a rank two
covariant tensor,
giℓ′ = (M −1 )j i (M −1 )m ℓ gjm . (20)
If we define the matrix S ≡ M −1 , then we can rewrite eq. (20) in matrix form,

G ′ = S T GS , (21)

where G is a real symmetric matrix whose matrix elements are given by gjm . Using Sylvester’s
theorem (see Appendix A), a real invertible matrix S exists such that,

G ′ = S T GS = diag(1, 1, . . . , 1 , −1, −1, . . . , −1 , 0, 0, . . . , 0) , (22)

| {z } | {z } | {z }
r s t

where r, s and t are non-negative integers such that r + s + t = dim g. Moreover, if G is

positive definite then s = t = 0 and (S T GS)ij = δij . Likewise, if G is negative definite then
r = t = 0 and (S T GS)ij = −δij .

2. Properties of the Killing form

Recall that for X, Y ∈ g,3

(exp adX )Y = eX Y x−X ∈ g . (23)

We therefore introduce the notation,

Adg ≡ exp adX , where g ≡ eX . (24)

Since g = eX ∈ G, this allows us to rewrite eq. (23) as,

Adg (Y ) = gY g −1 . (25)

The following identity, which is equivalent to the Jacobi identity, is notable. For X, Y ∈ g,

ad[X,Y ] = [adX , adY ] , (26)

To prove this result, consider the action of both sides of eq. (26) on Z ∈ g.
   
ad[X,Y ] (Z) = [X, Y ], Z = Z, [Y, X] , (27)
3
See Theorem 1 in the class handout entitled, Results for Matrix Exponentials and Logarithms.

3
and
[adX , adY ](Z) = adX adY − adY adX )(Z) = adX ([Y, Z]) − adY ([X, Z])
   
= X, [Y, Z] − Y, [X, Z]
   
= − X, [Z, Y ] − Y, [X, Z] , (28)

after using the antisymmetry of the commutator. Hence, eqs. (26)–(28) yields,
     
Z, [Y, X] + X, [Z, Y ] + Y, [X, Z] = 0 , (29)

which is the Jacobi identity. Note that if we put X = Ai and Y = Aj in eq. (26), we obtain

[adAi , adAj ] = ad[Ai ,Aj ] = fijk adAk , (30)

after using eq. (3).4 Hence, {adX | X ∈ g} constitutes the adjoint representation of g, with
basis vectors {adAi }.
Finally, we prove one additional identity,

adAdg (X) = Adg adX (Adg )−1 . (31)

First, we note that Adg Adg−1 (Y ) = Adg (g −1Y g) = g(g −1Y g)g −1 = Y, which implies that

(Adg )−1 = Adg−1 . (32)

Then, by using the definitions of the operators ad and Ad, it follows that for X, Y ∈ g and
g ∈ G,


Adg adX Adg−1 (Y ) = Adg adX (g −1 Y g) = Adg [X, g −1 Y g] = g([X, g −1Y g]g −1
= gXg −1Y − Y gXg −1 = [gXg −1, Y ] = adgXg−1 (Y ) = adAdg (X) (Y ) . (33)

Combining eqs. (32) and (33) yields eq. (31).

For X, Y, Z ∈ g and g ∈ G, the Killing form, (X, Y ) = Tr adX adY , satisfies the following
four properties:

1. bilinearity: (αX + βY, Z) = α(X, Z) + β(Y, Z) , (34)

2. symmetry: (X, Y ) = (Y, X) , (35)



3. antisymmetric: adZ (X), Y = − X, adZ (Y ) , (36)


4. orthogonal: Adg (X), Adg (Y ) = (X, Y ) , (37)

where α, β ∈ R.
The proofs of eqs. (34)–(37) are straightforward. Bilinearity of the Killing form follows from
the linearity of the operator ad and the linearity of the trace. The symmetry of the Killing
form follows from the identity Tr(AB) = Tr(BA). Eq. (36) is a consequence of the Jacobi
4
Observe that ad kX = k adX for k ∈ R, since [cf. eq. (1)], ad kX (Y ) = [kX, Y ] = k[X, Y ] = k adX (Y ), for
X, Y ∈ g and k ∈ R.

4
identity [cf. eqs. (26)–(29)]. To show this, consider the identity, [A, BC] = [A, B]C + B[A, C].
it then follows from eq. (26) that
[adZ , adX adY ] = [adZ , adX ] adY + adX [adZ , adY ]
= ad[Z,X] adY + adX ad[Z,Y ] . (38)
Taking the trace of both sides of eq. (38) and using the fact that the trace of a commutator
vanishes, one obtains


[Z, X], Y + X, [Z, Y ] = 0 , (39)
which is equivalent to eq. (36) in light of eq. (1). Finally, we use eq. (31) to compute,

 
Adg (X), Adg (Y ) = Tr adAdg (X) adAdg (Y ) = Tr Adg adX (Adg )−1 Adg adY (Adg )−1


= Tr adX adY = (X, Y ) , (40)
using the invariance of a trace under a cyclic permutation of its arguments. Thus, we have
confirmed eq. (37).
It is instructive to examine eqs. (36) and (37) by expressing X = xk Ak and y = y k Ak with
respect to a basis {Ak } of the Lie algebra g. We introduce the matrix elements of the operators
Adg and adX , which are henceforth denoted by Aj k and B j k , respectively [cf. footnote 2],
Adg (Ak ) = gAk g −1 = Aj k Aj , (41)
adX (Ak ) = [X, Ak ] = B j k Aj . (42)
Writing X = xk Ak ∈ g, it follows that
Adg (X) = xk Aj k Aj , ad(X) = xk B j k Aj . (43)
Then, eqs. (8) and (37) yield,
gkℓ xk y ℓ = gij xk y ℓ Ai k Ajℓ . (44)
Since eq. (44) is valid for any choice of X, Y ∈ g, it follows that
gkℓ = gij Ai k Aj ℓ . (45)
In matrix notation, where G is a real symmetric matrix whose matrix elements are given by
gjm and A ≡ Adg , we can rewrite eq. (45) as
G = AT GA . (46)
Similarly, eqs. (8) and (36) yield,
gij B i ℓ xℓ y j = −gℓk xℓ B k j y j . (47)
Since eq. (47) is valid for any choice of X, Y ∈ g, it follows that
gij B i ℓ = −gℓk B k j . (48)
In matrix form, eq. (48) can be written as GB = −(GB)T = −B T G, after using G = G T .
Equivalently, one can write,
B = −G −1 B T G . (49)

5
3. Killing form of a semisimple Lie algebra

In this section, we prove two theorems concerning the Killing form of a semisimple Lie
algebra g.
Theorem 1: If g is a semisimple Lie algebra, then the Killing form is nondegenerate and
conversely.
Proof: Suppose that g is non-semisimple. Then, g possesses a nonzero abelian ideal a. By
definition, an ideal a satisfies the condition that for all X ∈ a and Y ∈ g, we have [X, Y ] ∈ a.
If {An } is a basis for g, then a subset of {An } will span the ideal a. Suppose that Ai ∈ a and
Aj ∈ g. Then, we can conclude that [Ai , Aj ] ∈ a. Since [Ai , Aj ] = fijk Ak ∈ a, it follows that
fijk = 0 unless Ak is in the set of basis vectors that span a. Moreover, if both Ai and Aj are in
the set of basis vectors that span a then [Ai , Aj ] ∈ a must vanish (since a is abelian), which
implies that fijk = 0 for all k.
j
Consider the Cartan metric tensor, giℓ = fijk fℓk , in the case where its indices i and ℓ run
over values corresponding to the basis vectors of a. The argument presented above implies
that because a is an ideal, the index k runs only over values corresponding to the basis vectors
of a (since otherwise fijk = 0). Finally, because a is an abelian ideal, it follows that when the
j
indices ℓ and k run over values corresponding to the basis vectors of a, then fℓk = 0. Hence,
we conclude that giℓ = 0, when the indices i and ℓ run over values corresponding to the basis
vectors of a. Consequently, det G = 0, or equivalently, the Killing form is degenerate.
Conversely, if the Killing form is degenerate, then {X | (X, Y ) = 0 for all Y ∈ g} is a
nonzero solvable ideal of g. Thus, g is non-semisimple.5 Details can be found in the references.
Theorem 2: If g is a compact semisimple Lie algebra, then for any nonzero X ∈ g, it follows
that (X, X) < 0 and conversely.
Proof: Since Adg (for g ∈ G) is the adjoint representation of the Lie group G, it follows
that the matrix elements of the adjoint representation satisfy eq. (46). Moreover, in light of
eq. (22), one is always free to choose the coordinates on the Lie group manifold such that
G = diag(1, 1, . . . , 1 , −1, −1, . . . , −1) , (50)
| {z } | {z }
r s

where n = r + s is the dimension of the Lie algebra g. Note that in light of Theorem 1
above, the Killing form is nondegenerate, which implies that det G 6= 0 and hence t = 0 in the
notation of eq. (22). Recall that the matrix Lie group O(r, s) is defined by6
O(r, s) = {A ∈ Mr+s (R) | ATG = GA−1 } , (51)
5
Given a Lie algebra a, one can define a sequence of derived ideals, a(1) = [a, a], a(2) = [a(1) , a1 ], etc. By
induction and with the aid of the Jacobi identity, one can show that a(n) is an ideal of a for any value of n.
Then a is solvable if a positive integer k exists such that a(k) = [a(k−1) , a(k−1) ] = {~
0}. It follows that a(k−1) is
an abelian ideal of a. Hence, if g is non-semisimple then it possesses a nonzero solvable ideal, which implies
that it also possesses a nonzero abelian ideal.
6
See, e.g., the class handout providing a table of the real Lie algebras corresponding to the classical matrix
Lie groups, taken from J.F. Cornwell, Group Theory in Physics: An Introduction (Academic Press Inc., San
Diego, CA, 1997).

6
where Mr+s (R) is the set of all (r + s) × (r + s) real matrices and G is given in eq. (50). Thus,
it follows from eq. (46) that the adjoint group corresponding to the matrices of the adjoint
representation of G must be a subgroup of O(r, s).
Likewise, since adX is the adjoint representation of the Lie algebra g, it follows that the
matrix elements of the adjoint representation satisfy eq. (49). Recall that the matrix Lie
algebra so(r, s) is defined by [cf. footnote 6],

so(r, s) = {B ∈ Mr+s (R) | B TG = −GB} . (52)

Thus, it follows from eq. (49) that the matrices of the adjoint representation of g must be a
subalgebra of so(r, s).
Consider the special case where G = ±I, where I is the n × n identity matrix. In this case,
eqs. (46) and (49) yield,
AT A = I , B T = −B . (53)
In particular, the adjoint group corresponding to the matrices of the adjoint representation of
G constitute a subgroup of O(n) and the matrices of the adjoint representation of g constitute
a subalgebra of so(n). Moreover, for X ∈ g,7

X
(X, X) = Tr adX adX = TrB 2 = −Tr(BB T ) = − Bij Bij < 0 , (54)
i,j

unless B = 0 (or equivalently, unless adX = 0). In particular, (Ai , Ai ) < 0 [no sum over i],
which in light of eq. (10) means that gii < 0. Thus, we conclude that in the special case under
consideration, G = −I. Hence, a negative definite Killing form implies that the adjoint group
corresponding to the matrices of the adjoint representation of G constitutes a closed subgroup
of O(n), which implies that the Lie group G is compact. Conversely, an indefinite Killing form
implies that the adjoint group constitutes a subgroup of O(r, s) where neither r nor s is zero.
In this case, G is a noncompact group.
Note that the analysis presented above is based on the assumption that g is a real Lie
algebra. In the case of a complex Lie group, we showed in class that any compact complex Lie
group is abelian. Thus, any compact semisimple Lie group (and its corresponding Lie algebra)
must be real. That is, Theorem 2 is only relevant for real Lie algebras. In contrast, Theorem 1
applies both to real and complex Lie algebras.

4. Complex semisimple Lie algebras and their real forms

In general, the structure constants of a complex Lie algebra are complex. However, in the
case of a complex semisimple Lie algebra, one can always transform the original basis to a
new basis in which the structure constants are real. Conversely, starting from eq. (3) [where
the structure constants are real], one can complexify a real Lie algebra g by writing X = xi Ai
with xi ∈ C. This process is called complexification. We shall denote the resulting complex
Lie algebra by gC .
7
Note that when gij ∝ δij , then there is no distinction between covariant and contravariant indices, and
one can write all tensor quantities with lowered indices.

7
 Given a complex semisimple Lie algebra, gC , with basis vectors satisfying [Ai , Aj ] = fijk Ak
with fijk ∈ R, one can restrict xi ∈ R to obtain a real semisimple Lie algebra, gR , with the
same commutation relations.8 The Lie algebra gR is an example of a real form of gC . More
generally, a real form of gC is defined as any real Lie algebra whose complexification yields gC .
Note that dimC gC = dimR gR . That is, given a complex Lie algebra of complex dimension
n, the corresponding real forms have real dimension n. In particular, the maximal number of
linearly independent basis vectors is n for both gC and gR .
If gC is semisimple, then its Killing form is non-degenerate (in light of Theorem 1). Hence,
the corresponding real form, which has the same Cartan metric tensor, is also semisimple.
Indeed, one can always transform the original basis of g to a new basis in which the Cartan
metric tensor is of the form given by eq. (50). However, this new basis is not unique. For
example, one can always choose to multiply some of the basis vectors by i as long as all
the structure constants remain real. In particular, it is always possible to find a basis for a
semisimple gC such that G = −I, in which case the corresponding real form is compact.
Indeed, there exists a simple algorithm to derive various possible real forms of gC . The
algorithm consists of first choosing a basis for gC such that the Cartan metric tensor is of the
form given by eq. (50). Next, we consider all possible ways of multiplying a subset of the
basis vectors {Ak } of gC each by i such that fijk ∈ R. The resulting commutation relations
will differ in some cases from the original one by some signs. Nevertheless, complex linear
combinations of the new basis vectors, xi Ai (xi ∈ C), still generate gC . After all possibilities
for the {Ai } are considered, one now restricts the xi to be real in each case in order to
obtain the corresponding real forms. Note that among all possible real forms obtained in this
way, only one of them is compact, corresponding to the basis in which G = −I. Other real
forms obtained by the algorithm above necessarily possess an indefinite Killing form, and thus
correspond to noncompact real Lie algebras.
A simple example demonstrates the procedure for finding real forms of gC . Consider the
commutation relations of sl(2, C),
[Ai , Aj ] = ǫijk Ak , (55)
where the indices i, j, k run over 1,2 and 3. The corresponding Cartan metric tensor is negative
definite,9
giℓ = ǫkij ǫjℓk = −2δiℓ . (56)
If we restrict x ∈ R, then we obtain the real compact Lie algebra, su(2) ∼
i
= so(3). Consider
what happens if we redefine A1 → iA1 and A2 → iA2 , while leaving A3 unchanged. The new
commutation relations of sl(2, C) are now given by,
[A1 , A2 ] = −A3 , [A2 , A3 ] = A1 , [A3 , A1 ] = A2 . (57)
In this case, the Cartan metric is now indefinite,
j
giℓ = fijk fℓk = diag(2, 2, −2) . (58)
8
Not every complex non-semisimple Lie algebra has a real form. See, e.g., Example 5.1.24 on p. 88 of Ref. 7
or Example 5 on p. 19 of Ref. 8.
9
We can normalize the basis vectors appropriately such that giℓ = −δiℓ (although it is not necessary for
this discussion).

8
Thus, if we now restrict xi ∈ R, the resulting real Lie algebra is noncompact. Indeed, the
commutation relations given in eq. (57) correspond to those of sl(2, R) ∼ = su(1, 1) ∼
= so(2, 1).
Note that any other redefinition of a subset of the basis vectors via multiplication by i, such
that the structure constants remain real, will yield Lie algebras that are isomorphic to one of
the two possible classes of real forms identified above.
It is straightforward to show that real forms of a simple Lie algebra are simple. However,
the complexification of a real simple Lie algebra can yield either a simple or a semisimple real
algebra. For example, the complexification of the real simple Lie algebra so(3, 1) yields the
semisimple complex Lie algebra so(4, C) ∼ = sl(2, C) ⊕ sl(2, C).
There is another way of obtaining a real Lie algebra from a complex Lie algebra called
realification. Starting from the basis vectors, {Ak }, of the complex Lie algebra, the elements
of gC are of the form xk Ak , with xk ∈ C. One can now construct a real Lie algebra that is
given by real linear combinations of the basis vectors {Ak , iAℓ }. We shall denote this real
Lie algebra by (gC )R .10 Note that if dimC gC = n, then dimR (gC )R = 2n. As an example of
this construction, the realification of sl(2, C) yields sl(2, C)R ∼
= so(3, 1). This isomorphism is
important in the study of the Lie algebra of the Lorentz group.
One can now consider the complexification of the 2n-dimensional real Lie algebra, (gC )R ,
which will be denoted by (gC )∗R . Note that dimC (gC )∗R = 2n. One can easily verify that,
(gC )∗R ∼
= gC ⊕ gC . (59)
Thus, in our previous example, the complexification of sl(2, C)R is sl(2, C) ⊕ sl(2, C). Since
the complexification of so(3, 1) is so(4, C), it then follows that so(4, C) ∼ = sl(2, C) ⊕ sl(2, C).
In particular, sl(2, C)R [or equivalently, so(3, 1)] is one of real forms of so(4, C).

5. A completely antisymmetric third rank tensor

Using the Cartan metric tensor, one can construct a completely antisymmetric third rank
tensor that is related to the structure constants of the Lie algebra g,
fijk ≡ gkℓ fijℓ . (60)
To prove that fijk is a completely antisymmetric third rank tensor, we employ eqs. (9) and
(30) to write,




fijk = fijℓ Tr adAk adAℓ = Tr fijℓ adAk adAℓ = Tr adAk [adAi , adAj ] , (61)
after using the linearity of the trace. That is,


fijk = Tr adAk adAi adAj − adAk adAj adAi , (62)
which is manifestly antisymmetric under the interchange of any pair of indices i, j and k due
to the invariance of the trace under a cyclic permutation of its arguments.
Note that in the special case where g is a compact semisimple Lie algebra, we can choose
a basis in which gij = −δij . For this basis choice, it follows that the structure constants of g,
fijk = −fijk , are completely antisymmetric under the interchange of any pair of indices.
10
In the literature, the subscript R is sometimes omitted. One then says that the complex Lie algebra g of
complex dimension n can be regarded as a real Lie algebra of real dimension 2n.

9
APPENDIX A: Sylvester’s Theorem

In this appendix we prove the following theorem, often called Sylvester’s law of inertia.11
Sylvester’s Theorem: Consider an n × n hermitian matrix M = M † . Then there exists an
invertible matrix S such that

S † MS = diag(1, 1, . . . , 1 , −1, −1, . . . , −1 , 0, 0, . . . , 0) , (63)

| {z } | {z } | {z }
r s t

where r, s and t are non-negative integers such that r + s + t = n. That is, S † MS is a

diagonal matrix whose elements consist of 1 repeated r times, −1 repeated s times and 0
repeated t times along the diagonal. Moreover, if M is positive definite then s = t = 0 and
(S T MS)ij = δij . Likewise, if M is negative definite then r = t = 0 and (S T MS)ij = −δij .
Proof: Since M is hermitian, it follows that the eigenvalues of M are real. Moreover, M is
unitarily diagonalizable. That is, a unitary matrix U exists such that

U † MU = diag(λ1 , . . . , λr , λr+1, . . . , λr+s , 0, 0, . . . , 0) , (64)

| {z }
t

where the eigenvalues λ1 , . . . , λr are all positive real numbers, the eigenvalues λr+1 , . . . , λr+s
are all negative real numbers, and there are zero eigenvalues with multiplicity t. We can
therefore define a diagonal matrix,
p p p p
D ≡ diag( λ1 , . . . , λr , −λr+1 , . . . , −λr+s , 1, 1, . . . , 1) . (65)
| {z }
t

Setting S = UD −1 , it follows that,

S † MS = diag(1, 1, . . . , 1 , −1, −1, . . . , −1 , 0, 0, . . . , 0) , (66)

| {z } | {z } | {z }
r s t

which completes the proof. Recall that a positive (negative) definite matrix is an hermitian
matrix whose eigenvalues are positive (negative). Thus, if M is positive definite, it follows
that s = t = 0 and (S T MS)ij = δij . Likewise, if M is negative definite then r = t = 0 and
(S T MS)ij = −δij .
Corollary: Consider an n × n real symmetric matrix G = GT . Then there exists an invertible
real matrix S such that

S T GS = diag(1, 1, . . . , 1 , −1, −1, . . . , −1 , 0, 0, . . . , 0) , (67)

| {z } | {z } | {z }
r s t

where r, s and t are non-negative integers such that r + s + t = n. Moreover, if G is positive

definite then s = t = 0 and (S T GS)ij = δij . Likewise, if G is negative definite then r = t = 0
and (S T GS)ij = −δij .
11
See, e.g., Howard Eves, Elementary Matrix Theory (Dover Publications, Inc., Mineola, NY, 1980) pp. 237–
245.

10
Proof: Since G is a real symmetric matrix, its eigenvalues are all real and it is diagonalizable
by a real orthogonal matrix. That is, a real orthogonal matrix Q exists such that

QT GQ = diag(λ1 , . . . , λr , λr+1, . . . , λr+s , 0, 0, . . . , 0) , (68)

| {z }
t

where the eigenvalues, λ1 , . . . , λr [λr+1 , . . . , λr+s ], are all positive [negative] real numbers and
the zero eigenvalues of G have multiplicity t. We again define the diagonal matrix D by
eq. (65). Setting S = QD −1 (which is a real matrix) then establishes eq. (67).

References

I have found the following references useful in preparing these note.

1. Robert Gilmore, Lie Groups, Lie Algebras, and Some of Their Applications (Dover
Publications, Inc., Mineola, NY, 2005).

2. J.F. Cornwell, Group Theory in Physics, Volume 2 (Academic Press, London, UK, 1984).

3. Alexey P. Isaev and Valery A. Rubakov, Theory of Groups and Symmetries (World
Scientific Publishing Company, Singapore, 2018).

4. D.H. Sattinger and O.L. Weaver, Lie Groups and Algebras with Applications to Physics,
Geometry, and Mechanics (Springer-Verlag, New York, NY, 1986). In particular, see
Chapter 9.

5. Barry Simon, Representations of Finite and Compact Groups (American Mathematical

Society, Providence, RI, 1996). In particular, see Chapter VIII.2.

6. Karin Erdmann and Mark J. Wildon, Introduction to Lie Algebras (Springer-Verlag,

London, UK, 2006).

7. Joachim Hilgert and Karl-Hermann Neeb, Structure and Geometry of Lie Groups (Springer
Science, New York, NY, 2012).

8. Arkady L. Onishchik, Lectures on Real Semisimple Lie Algebras and their Representa-
tions (European Mathematical Society, Zurich, Switzerland, 2004).

11