Lecture note on Clifford algebra

arXiv:2205.09509v1 [hep-th] 18 May 2022

Jeong-Hyuck Park

Department of Physics, Sogang University, 35 Baekbeom-ro, Mapo-gu, Seoul 04107, Korea

[email protected]

Abstract
This lecture note surveys the gamma matrices in general dimensions with arbitrary signa-
tures, the study of which is essential to understand the supersymmetry in the corresponding
spacetime. The contents supplement the lecture presented by the author at Modave Summer
School in Mathematical Physics, Belgium, june, 2005.

May 20, 2022

1
1 Preliminary
Where do we see Clifford algebra?

• Dirac equation, for sure.

• Supersymmetry algebra.

• Non-anti-commutative superspace.

• Division algebra, R, C, H, O.

• Atiyah–Drinfeld–Hitchin–Manin construction of Yang–Mills instantons, F = ± ∗ F .

The gamma matrices in the Euclidean two-dimensions provide the fermionic oscillators,

f2 = 0 , f¯2 = 0 , {f, f¯} = 1 , (1.1)

where f = 21 (γ 1 + iγ 2 ), f¯ = 12 (γ 1 − iγ 2 ). Consequently, the irreducible representation is given

uniquely by 2 × 2 matrices acting on two dimensional spinors, |+i and |−i,
   
0 0 ¯ 0 1
f = |−ih+| = , f = |+ih−| = . (1.2)
1 0 0 0

Higher dimensional gamma matrices are then constructed by the direct products of them.

2 Gamma Matrix
We start with the following Theorem on linear algebra.

Theorem
Any matrix, M, satisfying M 2 = λ2 6= 0, λ ∈ C is diagonalizable, and furthermore if there is
another invertible matrix, N, which anti-commutes with M, {N, M} = 0, then M is 2n × 2n
matrix of the form  
λ 0
M =S S −1 . (2.1)
0 −λ
In particular, trM = 0. See Sec. A for our proof.

1
2.1 In Even Dimensions
In even d = t + s dimensions, with metric1

η µν = diag(+ · · · +} −
| +{z | −{z
· · · −}) , (2.2)
t s

gamma matrices, γ µ , satisfy the Clifford algebra

γ µ γ ν + γ ν γ µ = 2η µν . (2.3)

With2
γ µ1 µ2 ···µm = γ [µ1 γ µ2 · · · γ µm ] , (2.4)
we define ΓM , M = 1, 2, · · · 2d by assigning numbers to independent γ µ1 µ2 ···µm , e.g. imposing
µ1 < µ2 < · · · < µm ,

ΓM = (1, γ µ , γ µν , · · · , γ µ1 µ2 ···µm , · · · , γ 12···d ) . (2.5)

Then {ΓM }/Z2 forms a group:

ΓM ΓN = ΩM N ΓL , ΩM N = ±1 , (2.6)

where L is a fuction of M, N and ΩM N = ±1 does not depend on the specific choice of repre-
sentation of the gamma matrices.
Theorem (2.1) implies
1
tr(ΓM ΓN ) = ΩM N δ M N , (2.7)
2n
which shows the linear independence of {ΓM } so that any gamma matrix should not be smaller
than 2d/2 × 2d/2 .

In two-dimensions, one can take the Pauli sigma matrices, σ 1 , σ 2 as gamma matrices with a
possible factor, i, depending on the signature. In general, one can construct d + 2 dimensional
gamma matrices from d dimensional gamma matrices by taking tensor products as

(γ µ ⊗ σ 1 , 1 ⊗ σ 2 , 1 ⊗ σ 3 ) : up to a factor i . (2.8)
1
Note that throughout the lecture note we adopt the field theorists’ convention rather than string theorists such
that the time directions have the positive signature. The conversion is straightforward.
2
“[ ]” means the standard anti-symmetrization with “strength one”.

2
Thus, the smallest size of irreducible representations is 2d/2 × 2d/2 and {ΓM } forms a basis of
2d/2 × 2d/2 matrices.

By induction on the dimensions, from eq.(2.8), we may require gamma matrices to satisfy the
hermiticity condition 
 +γ µ for time-like µ
µ†
γ = γµ = . (2.9)
 µ
−γ for space-like µ
With this choice of gamma matrices we define γ (d+1) as
q
(d+1) t−s
γ = (−1) 2 γ 1 γ 2 · · · γ d , (2.10)

satisfying
γ (d+1) = (γ (d+1) )−1 = γ (d+1)† ,
(2.11)
µ (d+1)
{γ , γ } = 0.
For two sets of irreducible gamma matrices, γ µ , γ ′µ which are 2n×2n, 2n′ ×2n′ respectively,
we consider a matrix X
S= Γ′M T (ΓM )−1 , (2.12)
M
′
where T , is an arbitrary 2n × 2n matrix.
This matrix satisfies for any N from eq.(2.6)

Γ′N S = SΓN . (2.13)

By Schur’s Lemmas, it should be either S = 0 or n = n′ , det S 6= 0. Furthermore, S is unique

up to constant, although T is arbitrary. This implies the uniqueness of the irreducible 2d/2 × 2d/2
gamma matrices in even d dimensions, up to the similarity transformations. These similarity
transformations are also unique up to constant. Consequently there exist similarity transforma-
tions which relate γ µ to γ µ† , γ µ∗ , γ µT since the latter form also representations of the Clifford
algebra. By combining γ (d+1) with the similarity transformations, from eq.(2.11), we may acquire
the opposite sign, −γ µ† , −γ µ∗ , −γ µT as well.
Explicitly we define3 q
t(t−1)
A= (−1) 2 γ1γ2 · · · γt , (2.14)
3
Alternatively, one can construct C± explicitly out of the gamma matrices in a certain representation [1].

3
satisfying

A = A−1 = A† , (2.15)

γ µ† = (−1)t+1 Aγ µ A−1 . (2.16)

If we write
±γ µ∗ = B± γ µ B±
−1
, (2.17)
then from
γ µ = (γ µ∗ )∗ = B±
∗
B± γ µ (B±
∗
B± )−1 , (2.18)
one can normalize B± to satisfy [2, 3]
1
∗
B± B± = ε± 1 , ε± = (−1) 8 (s−t)(s−t±2) , (2.19)

†
B± B± = 1 , (2.20)

T
B± = ε± B± , (2.21)

where the unitarity follows from

† µ∗ † −1 † †
γ µ = 㵆 = (±B±
−1 ∗
γµ B± )† = ±B± γ (B± ) = B± B± γ µ (B± B± )−1 , (2.22)
†
and the positive definiteness of B± B± . The calculation of ε± is essentially counting the dimen-
sions of symmetric and anti-symmetric matrices [2, 3]4 .
What is worthy of notice is the case ε± = +1. As we see later in (4.4), (4.5), if ε+ = +1, the
gamma matrices can be chosen to real, i.e. B+ = 1, while if ε− = +1, the gamma matrices
can be chosen to pure imaginary, i.e. B− = 1. Especially when the gamma matrices are real
we say they are in the Majorana representation.

1
From (2.24) we have (C± γ µ1 µ2 ···µn )T = χn± C± γ µ1 µ2 ···µn , χn± := ε± (±1)t+n (−1)n+ 2 (t+n)(t+n−1) (2.29).
4

Thus, one can obtain the dimension of the symmetric 2d/2 × 2d/2 matrices as

  d
d/2−1 d/2
X
1 d!
2 2 +1 = 2 (1 + χn± ) .
n=0
n!(d − n)!

From this one can obtain the value of ε± (2.19).

4
 The charge conjugation matrix, C± , given by
T
C± = B± A, (2.23)

satisfies5 from the properties of A and B±

C± γ µ C±−1 = ζγ µT , ζ = ±(−1)t+1 , (2.24)

C±† C± = 1 , (2.25)

1 1
C±T = (−1) 8 d(d−ζ2) C± = ε± (±1)t (−1) 2 t(t−1) C± , (2.26)

1
ζ t (−1) 2 t(t−1) AT = B± AB±
−1
= C± AC±−1 . (2.27)

The sign factors ε± and ζ are related by

1 1
ε± = ζ t (−1) 2 t(t−1)+ 8 d(d−ζ2) . (2.28)

Eqs.(2.24, 2.26) imply

1 1
(C± γ µ1 µ2 ···µn )T = ζ n (−1) 8 d(d−ζ2)+ 2 n(n−1) C± γ µ1 µ2 ···µn
(2.29)
t+n n+ 21 (t+n)(t+n−1) µ1 µ2 ···µn
= ε± (±1) (−1) C± γ .

γ (d+1) satisfies
γ (d+1)† = (−1)t A± γ (d+1) A−1
± = γ
(d+1)
,
t−s −1
γ (d+1)∗ = (−1) 2 B± γ (d+1) B± , (2.30)
t+s
γ (d+1)T = (−1) 2 C± γ (d+1) C±−1 ,
where {A+ , A− } = {A, γ (d+1) A}.

In stead of eq.(2.8) one can construct d + 2 dimensional gamma matrices from d dimensional
gamma matrices by taking tensor products as

(γ µ ⊗ σ 1 , γ (d+1) ⊗ σ 1 , 1 ⊗ σ 2 ) : up to a factor i . (2.31)

5
Essentially all the properties of the charge conjugation matrix, C± depends only on d and ζ. However it is useful
here to have expression in terms of the signature to dicuss the Majorana supersymmetry later.

5
Therefore the gamma matrices in even dimensions can be chosen to have the “off-block diagonal”
form    
µ 0 σµ (d+1) 1 0
γ = , γ = , (2.32)
σ̃ µ 0 0 −1
d d
where the 2 2 −1 × 2 2 −1 matrices, σ µ , σ̃ µ satisfy
σ µ σ̃ ν + σ ν σ̃ µ = 2η µν , (2.33)

σ µ† = σ̃µ . (2.34)
In this choice of gamma matrices, from eq.(2.30), A± , B± , C± are either “block diagonal” or
“off-block diagonal” depending on whether t, t−s 2
, t+s
2
are even or odd respectively.
In particular, in the case of odd t, we write from eqs.(2.14, 2.15) A as
  q
0 a t(t−1)
A= , a = (−1) 2 σ 1 σ̃ 2 · · · σ t = ㆠ= ã−1 , (2.35)
ã 0
t+s
and in the case of odd we write from eq.(2.26) C± as
2
 
0 c t(t−1)
C± = , c = ε+ (−1) 2 c̃T = (c† )−1 , (2.36)
±c̃ 0
where a, ã, c, c̃ satisfy from eqs.(2.16, 2.24)
σ µ† = ãσ µ ã , σ̃ µ† = aσ̃ µ a ,
(2.37)
µT t+1 µ −1 µT t+1 µ −1
σ = (−1) c̃σ c , σ̃ = (−1) cσ̃ c̃ .
t+s
If both of t and 2
are odd then from eq.(2.27)
t−1 t−1
aT = (−1) 2 c̃ a c−1 , ãT = (−1) 2 c ã c̃−1 . (2.38)

2.2 In Odd Dimensions

The gamma matrices in odd d + 1 = t + s dimensions are constructed by combining a set of
even d dimensional gamma matrices with either ±γ (d+1) or ±iγ (d+1) depending on the signature
of even d dimensions. This way of construction is general, since γ (d+1) serves the role of γ d+1
−γ µ = γ d+1 γ µ (γ d+1 )−1 , for µ = 1, 2, · · · , d ,
(2.39)
d+1 2
(γ ) = ±1 ,

6
and such a matrix is unique in irreducible representations up to sign.

However, contrary to the even dimensional Clifford algebra, in odd dimensions two different
choices of the signs in γ d+1 bring two irreducible representations for the Clifford algebra, which
can not be mapped to each other6 by similarity transformations

γ µ = (γ 1 , γ 2 , · · · , γ d+1 ) and γ ′µ = (γ 1 , γ 2 , · · · , γ d , −γ d+1 ) . (2.40)

If there were a similarity transformation between these two, it should have been identity up to
constant because of the uniqueness of the similarity transformation in even dimensions. Clearly
this would be a contradiction due to the presence of the two opposite signs in γ d+1 .

In general one can put7


 ±γ 12···d for t − s ≡ 1 mod 4 ,
d+1
γ = (2.41)

±iγ 12···d for t − s ≡ 3 mod 4 .

2d/2 × 2d/2 gamma matrices in odd d + 1 dimensions, γ µ , µ = 1, 2, · · · , d + 1, induce the

following basis of 2d/2 × 2d/2 matrices, Γ̃M

Γ̃M = (1, γ µ , γ µν , · · · , γ µ1 µ2 ···µd/2 ) , M = 1, 2, · · · 2d . (2.42)

From eq.(2.41)
Γ̃M Γ̃N = Ω̃M N Γ̃L ,

 ±1 for t − s ≡ 1 mod 4 , (2.43)
Ω̃M N =

±1, ±i For t − s ≡ 3 mod 4 .
Here, contrary to the even dimensional case, Ω̃M N depends on each particular choice of the
representations due to the arbitrary sign factor in γ d+1 . This is why eq.(2.13) does not hold
in odd dimensions. Therefore it is not peculiar that not all of ±γ µ† , ±γ µ∗ , ±γ µT are related
to γ µ by similarity transformations. In fact, if it were true, say for ±γ µ∗ , then the similarity
Nevertheless, this can be cured by the following transformation. Under xµ = (x1 , x2 , · · · , xd+1 ) → x′µ =
6

(x , x2 , · · · , −xd+1 ), we transform the Dirac field ψ(x) as ψ(x)

1
→ ψ ′ (x′ ) = ψ(x) , to get ψ̄(x)γ ·
′ ′ ′ ′ ′ ′
∂ψ(x) → ψ̄ (x )γ · ∂ ψ (x ) = ψ̄(x)γ · ∂ψ(x) . Hence those two representations are equivalent describing
the same physical system.
7
Our results (2.41-2.50) do not depend on the choice of the signature in d dimensions, i.e. they hold for either
increasing the time dimensions, d = (t − 1) + s or the space dimensions, d = t + (s − 1).

7
transformation should have been B± (2.17) by the uniqueness of the similarity transformations in
even dimensions, but this would be a contradiction to eq.(2.30), where the sign does not alternate
under the change of B+ ↔ B− . Thus, in odd dimensions, only the half of ±γ µ† , ±γ µ∗ , ±γ µT are
related to γ µ by similarity transformations and hence from eq.(2.30) there exist three similarity
transformations, A, B, C such that
(−1)t+1 γ µ† = Aγ µ A−1 , (2.44)

t−s−1
(−1) 2 γ µ∗ = Bγ µ B −1 , (2.45)

t+s−1
(−1) 2 γ µT = Cγ µ C −1 . (2.46)
A, B, C are all unitary and satisfy
A = A−1 = A† , C = BT A , (2.47)

1
B ∗ B = ε 1 = (−1) 8 (t−s+1)(t−s−1) 1 , (2.48)

ts 1
B T = εB , C T = ε(−1) 2 C = (−1) 8 (t+s+1)(t+s−1) C , (2.49)

ts
(−1) 2 AT = BAB −1 = CAC −1 . (2.50)
In particular, A is given by eq.(2.14).

2.3 Lorentz Transformations

Lorentz transformations, L can be represented by the following action on gamma matrices in a
standard way
L−1 γ µ L = Lµν γ ν , (2.51)
where L and L are given by
µν 1 µν
L = ewµν M , L = e 2 wµν γ ,
(2.52)
(M µν )λρ = η µλ δ νρ − η νλ δ µρ .

For even d, if a 2d/2 × 2d/2 matrix, M µ1 µ2 ···µn , is totally anti-symmetric over the n spacetime
indices
M µ1 µ2 ···µn = M [µ1 µ2 ···µn ] , (2.53)

8
and transforms covariantly under Lorentz transformations in d or d + 1 dimensions as
n
Y
−1 µ1 µ2 ···µn
L M L= Lµi νi M ν1 ν2 ···νn , (2.54)
i=1

then for 0 ≤ n ≤ max(d/2, 2), the general forms of M µ1 µ2 ···µn are


 (1 + cγ (d+1) )γ µ1 µ2 ···µn In even d dimensions ,
M µ1 µ2 ···µn = (2.55)
 µ1 µ2 ···µn
γ In odd d + 1 dimensions ,

where c is a constant.

To show this, one may first expand M µ1 µ2 ···µn in terms of γν1 ν2 ···νm , γ (d+1) γν1 ν2 ···νm or γν1 ν2 ···νm
depending on the dimensions, d or d + 1, with 0 ≤ m ≤ d/2. Then eq.(2.54) implies that the
coefficients of them, say T µ1 µ2 ···µm+n , are Lorentz invariant tensors satisfying
m+n
Y
Lµνii T ν1 ν2 ···νm+n = T µ1 µ2 ···µm+n (2.56)
i=1

Finally one can recall the well known fact [4] that the general forms of Lorentz invariant tensors
are multi-products of the metric, η µν , and the totally antisymmetric tensor, ǫµ1 µ2 ··· , which verifies
eq.(2.55).

2.4 Crucial Identities for Super Yang-Mills

The following identities are crucial to show the existence of the non-Abelian super Yang-Mills in
THREE, FOUR, SIX and TEN dimensions.

(i) The following identity holds only in THREE or FOUR dimensions with arbitrary signature

0 = (γ µ C −1 )αβ (γµ C −1 )γδ + cyclic permutations of α, β, γ (2.57)

To verify the identity in even dimensions we contract (γ µ C −1 )αβ (γµ )γδ with (Cγ ν1 ν2 ···νn )βα and
take cyclic permutations of α, β, γ to get
1 1
0 = 2d/2 δ1n + (d − 2n)(ζ + ζ n (−1) 2 n(n−1) )(−1)n+ 8 d(d−ζ2) (2.58)

9
This equation must be satisfied for all 0 ≤ n ≤ d, which is valid only in d = 4, ζ = −1.
Similar analysis can be done for the d+1 odd dimensions by adding (γ (d+1) C −1 )αβ (γ (d+1) C −1 )γδ
term into eq.(2.57). We get
1 1
0 = 2d/2 (δ1n + δdn ) + (d − 2n + 1)(ζ + ζ n (−1) 2 n(n−1) )(−1)n+ 8 d(d−ζ2) , ζ = (−1)d/2
(2.59)
Only in d = 2 and hence three dimensions, this equation is satisfied for all 0 ≤ n ≤ d.

(ii) The following identity holds only in TWO, FOUR or SIX dimensions with arbitrary signature

0 = (σ µ )αβ (σµ )γδ + (σ µ )γβ (σµ )αδ (2.60)

To verify this identity we take d dimensional sigma matrices from f = d − 2 dimensional gamma
matrices as in eq.(2.31)
σ µ = (γ µ , γ (f +1) , i) (2.61)
to get
(σ µ )αβ (σµ )γδ = (γ µ )αβ (γµ )γδ + (γ (f +1) )αβ (γ (f +1) )γδ − δαβ δγδ (2.62)
Again this expression is valid for any signature, (t, s). Now we contract this equation with
(γ ν1 ν2 ···νn C+−1 )βδ . From eqs.(2.24, 2.30) in the case of odd t we get
 f

(−1)n (f − 2n) + (−1) 2 +n − 1 (γ ν1 ν2 ···νn C+−1 )αγ (2.63)

To satisfy eq.(2.60) this expression must be anti-symmetric over α ↔ γ for any 0 ≤ n ≤ f .

f
Thus from eq.(2.29) we must require 0 = (−1)n (f − 2n) + (−1) 2 +n − 1 for all n satis-
1 1
fying (−1) 8 f (f −2)+ 2 n(n−1) = 1. This condition is satisfied only in f = 0, 2, 4 and hence
d = 2, 4, 6 (f = 6 case is excluded by choosing n = 6 and f ≥ 8 cases are excluded by
choosing either n = 0 or n = 3).

(iii) The following identity holds only in TWO or TEN dimensions with arbitrary signature

0 = (σ µ c−1 )αβ (σµ c−1 )γδ + cyclic permutations of α, β, γ (2.64)

10
3 Spinors
3.1 Weyl Spinor
In any even d dimensions, Weyl spinor, ψ, satisfies

γ (d+1) ψ = ψ (3.1)

and so ψ̄ = ψ † A satisfies from eq.(2.30)

t−s
ψ̄γ (d+1) = (−1)t ψ̄ γ (d+1) C±−1 ψ̄ T = (−1) 2 C±−1 ψ̄ T (3.2)

3.2 Majorana Spinor

By definition Majorana spinor satisfies

ψ̄ = ψ T C± or ψ̄ = ψ T C (3.3)

depending on the dimensions, even or odd. This is possible only if ε± , ε = 1 and so from
eqs.(2.19, 2.48)
η = +1 : t − s = 0, 1, 2 mod 8
(3.4)
η = −1 : t − s = 0, 6, 7 mod 8
where η is the sign factor, ±1, occuring in eq.(2.17) or eq.(2.45)8 .

3.3 Majorana-Weyl Spinor

Majorana-Weyl spinor satisfies both of the two conditions above

γ (d+1) ψ = ψ ψ̄ = ψ T C± (3.5)

Majorana-Weyl Spinor exists only if

η = +1 : t − s = 0 mod 8
(3.6)
η = −1 : t − s = 0 mod 8

8
In [2], η = −1 case is called Majorana and η = +1 case is called pseudo-Majorana.

11
4 Majorana Representation and SO(8)
Fact 1:
Consider a finite dimensional vector space, V with the unitary and symmetric matrix, B = BT ,
BB† = 1. For every |vi ∈ V if B|vi∗ ∈ V then there exists an orthonormal “semi-real ” basis,
V = {|li, l = 1, 2, · · ·} such that B|li∗ = |li.

Proof
Start with an arbitrary orthonormal bais, {|vl i, l = 1, 2, · · ·} and let |1i ∝ |v1 i + B|v1 i∗ . After
the normalization, h1|1i = 1, we can take a new orthonormal basis, {|1i, |2′i, |3′i, · · ·}. Now we
assume that {|1i, |2i, · · · |k − 1i, |k ′ i, |(k + 1)′i, · · ·} is an orhonormal basis such that B|ji∗ = |ji
for 1 ≤ j ≤ k − 1. To construct the k th such a vector, |ki we set |ki ∝ |k ′ i + B|k ′ i∗ with the
normalization. We check this is orthogonal to |ji, 1 ≤ j ≤ k − 1

hj|( |k ′ i + B|k ′ i∗ ) = 0 + hk|ji = 0 . (4.1)

In this way one can construct the desired basis.

In the spacetime which admits Majorana spinor from Eq.(3.4)

η = +1 : t − s = 0, 1, 2 mod 8
(4.2)
η = −1 : t − s = 0, 6, 7 mod 8 ,
more explicitly in the even dimensions having ε+ = 1 (or ε− = 1) where B+ (or B− ) is symmetric
and also in the odd dimensions of ε = 1 where B is symmetric, from Fact 1 above we can choose
an “semi-real ” orthonormal basis such that Bη−1 |li∗ = |li (here it is Bη−1 that plays the role of B
in Fact 1). In the basis, we write the gamma matrices
X µ
γµ = Rlm |lihm| . (4.3)

From η γ µ∗ = Bη γ µ Bη−1 and the property of the semi-real basis, Bη−1 |li∗ = |li we get
µ ∗ µ
(Rlm ) = ηRlm . (4.4)

Since Rµ is also a representation of the gamma matrix

Rµ Rν + Rν Rµ = 2η µν , (4.5)

adopting the true real basis, we conclude that there exists a Majorana represention where the
gamma matrices are real, η = + or pure imaginary, η = − in any spacetime admitting

12
Majorana spinors.

Furthermore from Eq.(2.30), in the even dimension of t − s ≡ 0 mod 8, ε± = 1 and γ (d+1)∗ =

Bγ (d+1) B −1 (here we omit the subscript index ± or η for simplicity.). The action, |vi → B † |vi∗
preserves the chirality, and from the fact 1 above we can choose an orthonormal semi-real basis
for the chiral and anti-chiral spinor spaces, V = V+ + V− , V± = {|l± i} such that
hl± |m± i = δlm , hl± |m∓ i = 0 , γ (d+1) |l± i = ±|l± i , B † |l± i∗ = |l± i . (4.6)
With the semi-real basis  
(d+1) 1 0
γ = , (4.7)
0 −1
and the gamma matrices are in the Majorana representation
 µ

0 r
γµ = , r µ ∈ O(2d/2−1 ) , r µ r ν T + r ν r µT = 2δ µν . (4.8)
rµT 0

From Eq.(6.8) any two sets of semi-real basis, say {|l± i} and {|˜l± i} are connected by an
O((2d/2−1 )) transformation
X X
|˜l± i = Λ±ml |m± i , Λ±lm Λ±nm = δln . (4.9)
m m

If we define X
Λ± = Λ±lm |l± ihm± | , (4.10)
l,m

then |˜l± i = Λ± |l± i and from the definition of the semi-real basis

Λ± = B † Λ∗± B = Λ± P± = P± Λ± , Λ± Λ†± = P± . (4.11)

We write ∞
M±
X 1
Λ± = e , M± ≡ (−1)n+1 (Λ± − P± )n = ln Λ± . (4.12)
n=1
n
Thus for Λ± such that the infinity sum converges we have
M± = −M±† = B † M±∗ B = M± P± = P± M± . (4.13)
This gives a strong constraint when we express M± by the gamma matrix products. For the
Eucledean eight dimensions only the SO(8) generators for the spinors survive in the expansion!
M± = 12 wab γ ab P± . (4.14)

13
Namely we find an isomorphism between the two SO(8)’s, one for the semi-real vectors and the
other for the spinors in the conventional sense. Alternatively this can be seen from
 [a b] T 
ab r r 0
γ = , (4.15)
0 r [a T r b]
where the each block diagonal is a generator of SO(D) while the dimension of the chiral space
is 2d/2−1 . Only in d = 8 both coincide leading to the “so(8) triolity” among sov (8), soc (8) and
soc̄ (8).

Fact 2: Relation to octonions.

In Euclidean eight dimensions, the 16×16 gamma matrices can be taken of the off-block diagonal
form,  
0 ra
γa = , ra rbT + rb raT = 2δab , (4.16)
raT 0
where the 8 × 8 real matrices, ra , 1 ≤ a ≤ 8, give the multiplication of the octonions, oa ,
oa ob = (ra )b c oc . (4.17)

Fact 3:
Consider an arbitrary real self-dual or anti-self-dual four form in D = 8
±
Tabcd = ± 4!1 ǫabcdef gh T ±ef gh . (4.18)
Using the SO(8) rotations one can transform the four form into the canonical form where the non-
± ± ± ± ± ± ±
vanishing components are T1234 , T1256 , T1278 , T1357 , T1368 , T1458 , T1467 and their dual counter
parts only.

Proof
We start with the seven linearly independent traceless Hermitian matrices
E±1 = γ 2341 P± , E±2 = γ 2561 P± , E±3 = γ 2781 P± , E±4 = γ 1357 P± ,
(4.19)
E±5 = γ 3681 P± , E±6 = γ 4581 P± , E±7 = γ 4671 P± .
As they commute with each other, there exists a basis V± = {|l± i} diagonalizing the seven
quantities X
E±r = λrl |l± ihl± | , (λrl )2 = 1 . (4.20)
l

14
Further, since C|l± i∗ is also an eigenvector of the same eigenvalues, from the fact 1 we can im-
pose the semi-reality condition without loss of generality, C|l± i∗ = |l± i.

Now for the self-dual four form we let

T ± = 14 Tabcd
±
γ abcd . (4.21)

Since T ± is Hermitian and C(T ± )∗ C † = T ± , one can diagonalize T ± with a semi-real basis
X
T± = λl |˜l± ih˜l± | , C|˜l± i∗ = |˜l± i . (4.22)
l

For the two semi-real basis above we define a transformation matrix

O± = |l± ih˜l± | . (4.23)

Then, since T ± is traceless, O± T ± O±† can be written in terms of E±i ’s. Finally the fact O± gives
a spinorial SO(8) rotation completes our proof.

Some useful formulae are

±P± = E±1 E±2 E±3 = E±1 E±4 E±5 = E±1 E±6 E±7 = E±2 E±4 E±6
(4.24)
= E±2 E±5 E±7 = E±3 E±4 E±7 = E±3 E±5 E±6 .

For an arbitrary self-dual or anti-self-dual four form tensor in D = 8, from

± ±
Tacde T ±bcde = ( 4!1 )2 ǫacdef ghi ǫbcdejklm T ±f ghi Tjklm
(4.25)
1 ± ±
= 4
δa b Tcdef T ±cdef − Tacde T ±bcde ,

we obtain an identity
±
Tacde T ±bcde = 1
8
±
δa b Tcdef T ±cdef . (4.26)

15
5 Superalgebra
5.1 Graded Lie Algebra
Supersymmetry algebra is a Ẑ2 graded Lie algebra, g = {Ta }, which is an algebra with commu-
tation and anti-commutation relations [5, 6]
c
[Ta , Tb } = Cab Tc (5.1)
c
where Cab is the structure constant and
[Ta , Tb } = Ta Tb − (−1)#a#b Tb Ta (5.2)
with #a, the Ẑ2 grading of Ta ,

0 for bosonic a
#a = (5.3)
1 for fermionic a
The generalized Jacobi identity is
[Ta , [Tb , Tc }} − (−1)#a#b [Tb , [Ta , Tc }} = [[Ta , Tb }, Tc } (5.4)
which implies
(−1)#a#c Cab
d e
Cdc + (−1)#b#a Cbc
d e
Cda + (−1)#c#b Cca
d e
Cdb =0 (5.5)
For a graded Lie algebra we consider
g(z) = exp(z a Ta ) (5.6)
where z a is a superspace coordinate component which has the same bosonic or fermionic property
as Ta and hence z a Ta is bosonic.
In the general case of non-commuting objects, say A and B, the Baker-Campbell-Haussdorff
formula gives !
X∞
eA eB = exp Cn (A, B) (5.7)
n=0

where Cn (A, B) involves n commutators. The first three of these are

C0 (A, B) = A + B

C1 (A, B) = 12 [A, B] (5.8)

1 1
C2 (A, B) = 12
[[A, B], B] + 12
[A, [A, B]]

16
Since for the graded algebra

[z a Ta , z b Tb ] = z b z a [Ta , Tb } = z b z a Cab
c
Tc (5.9)

the Baker-Campbell-Haussdorff formula (5.7) implies that g(z) forms a group, the graded Lie
group. Hence we may define a function on superspace, f a (w, z), by

g(w)g(z) = g(f (w, z)) (5.10)

Since g(0) = e, the identity, we have f (0, z) = z, f (w, 0) = w and further we assume that
f (w, z) has a Taylor expansion in the neighbourhood of w = z = 0.
Associativity of the group multiplication requires f (w, z) to satisfy

f (f (u, w), z) = f (u, f (w, z)) (5.11)

5.2 Left & Right Invariant Derivatives

For a graded Lie group, left and right invariant derivatives, La , Ra are defined by

La g(z) = g(z)Ta (5.12)

Ra g(z) = −Ta g(z) (5.13)

Explicitly we have

b b ∂f b (z, u)
La = La (z)∂b La (z) = (5.14)
∂ua u=0

∂f b (u, z)
Ra = Ra b (z)∂b Ra b (z) = − (5.15)
∂ua u=0

∂
where ∂b = ∂z b.

It is easy to see that La is invariant under left action, g(z) → hg(z), and Ra is invariant under
right action, g(z) → g(z)h.
From eqs.(5.12, 5.13) we get
c
[La , Lb } = Cab Lc (5.16)

c
[Ra , Rb } = Cab Rc (5.17)

17
and from eqs.(5.12, 5.13) we can also easily show

[La , Rb } = 0 (5.18)

Thus, La (z), Ra (z) form representations of the graded Lie algebra separately. For the supersym-
metry algebra, the left invariant derivatives become covariant derivatives, while the right invariant
derivatives become the generators of the supersymmetry algebra acting on superfields.

5.3 Superspace & Supermatrices

In general a superspace may be denoted by Rp|q , where p, q are the number of real commuting
(bosonic) and anti-commuting (fermionic) variables respectively. A supermatrix which takes
Rp|q → Rp|q may be represented by a (p + q) × (p + q) matrix, M, of the form
 
a b
M= (5.19)
c d

where a, d are p × p, q × q matrices of Grassmanian even or bosonic variables and b, c are p × q,

q × p matrices of Grassmanian odd or fermionic variables respectively.
The inverse of M can be expressed as
 
−1 (a − bd−1 c)−1 −a−1 b(d − ca−1 b)−1
M = (5.20)
−d−1 c(a − bd−1 c)−1 (d − ca−1 b)−1

where we may write

∞
X
−1 −1 −1
(a − bd c) =a + (a−1 bd−1 c)n a−1 (5.21)
n=1

Note that due to the fermionic property of b, c, the power series terminates at n ≤ pq + 1.
The supertrace and the superdeterminant of M are defined as

str M = tr a − tr d (5.22)

sdet M = det(a − bd−1 c)/ det d = det a/ det(d − ca−1 b) (5.23)

The last equality comes from

det(1 − a−1 bd−1 c) = det−1 (1 − d−1 ca−1 b) (5.24)

18
which may be shown using
∞
!
X 1 n
det(1 − a) = exp − tr a (5.25)
n=1
n

and observing
tr (a−1 bd−1 c)n = −tr (d−1 ca−1 b)n (5.26)
From eq.(5.23) we note that sdet M 6= 0 implies the existence of M −1 . Thus the set of superma-
trices for sdet M 6= 0 forms the supergroup, Gl(p|q). If sdet M = 1 then
M ∈ Sl(p|q).
The supertrace and the superdeterminant have the properties

str (M1 M2 ) = str (M2 M1 ) (5.27)

sdet (M1 M2 ) = sdet M1 sdet M2 (5.28)

We may define the transpose of the supermatrix, M, either as

 t 
t a ct
M = (5.29)
−bt dt
or as  
t′ at −ct
M = (5.30)
bt dt
where at , bt , ct , dt are the ordinary transposes of a, b, c, d respectively.
We note that
′ ′ ′
(M1 M2 )t = M2t M1t (M1 M2 )t = M2t M1t (5.31)

′ ′
(M t )t = (M t )t = M (5.32)

19
6 Super Yang-Mills
6.1 (3 + 1)D N = 1 super Yang-Mills
In four-dimensional Minkowskian spacetime of the metric, η = diag(− + ++), the 4 × 4 gamma
matrices satisfy with µ = 0, 1, 2, 3,
Γµ† = Γµ = −AΓµ A† , A = Γt = −A† ,

Γµ∗ = +BΓµ B † , BT = B , B † = B −1 , (6.1)

ΓµT = −CΓµ C † , C = −C T = BΓt , C † = C −1 .

The Majorana spinor, ψ satisfies then

ψ̄ = ψ † Γt = ψ T C ⇐⇒ ψ ∗ = Bψ . (6.2)

The four-dimensional super Yang-Mills Lagrangian reads

L4D = tr − 41 Fµν F µν − i 21 ψ̄Γµ Dµ ψ . (6.3)

The supersymmetry transformations are

δAµ = iε̄Γµ ψ = −iψ̄Γµ ε , δψ = − 12 Fµν Γµν ε . (6.4)

6.2 (5 + 1)D (1, 0) super Yang-Mills

In six-dimensional Minkowskian spacetime of the metric, η = diag(− + + + ++), the 8 × 8
gamma matrices satisfy with M = 0, 1, 2, 3, 4, 5,
ΓM † = ΓM = AΓM A† , A := Γ12345 = A† = A−1 ,

ΓM T = CΓM C † , C T = −C , C † = C −1 , (6.5)

ΓM ∗ = BΓM B † , B = CA = −B T , B † = B −1 .

The gamma “seven” is given by Γ(7) = Γ012345 to satisfy Γ(7) = Γ(7)† = Γ(7)−1 and

ΓLM N = 16 ǫLM N P QR ΓP QR Γ(7) , (6.6)

where ǫ012345 = +1.

20
 The su(2) Majorana-Weyl spinor, ψi , i = 1, 2, satisfies then
Γ(7) ψi = +ψi , ψ̄ i Γ(7) = −ψ̄ i : chiral ,
(6.7)
i † ij T
ψ̄ = (ψi ) A = ǫ (ψj ) C : su(2) Majorana ,

where ǫij is the usual 2×2 skew-symmetric unimodular matrix. It is worth to note that ψ̄ i ΓM1 M2 ···M2n ρi =
0 and
 †
tr(iψ̄ i ΓM1 M2 ···M2n+1 ρi ) = tr(iψ̄ i ΓM1 M2 ···M2n+1 ρi ) = −(−1)n tr(iρ̄i ΓM1 M2 ···M2n+1 ψi ) , (6.8)
where ψi , ρi are two arbitrary Lie algebra valued su(2) Majorana-Weyl spinors.

The six-dimensional super Yang-Mills Lagrangian reads

L6D = tr − 41 FLM F LM − i 12 ψ̄ i ΓL DL ψi , (6.9)
where all the fields are in the adjoint representation of the gauge group such that, with the Her-
mitian Lie algebra valued gauge field, AM ,
DL ψi = ∂L ψi − i[AL , ψi ] , FLM = ∂L AM − ∂M AL − i[AL , AM ] . (6.10)
From (6.8) the action is real valued.

The supersymmetry transformations are given by with a su(2) Majorana-Weyl supersymmetry

parameter, εi ,
δAM = +iε̄i ΓM ψi = −iψ̄ i ΓM εi , δψi = − 12 FM N ΓM N εi , (6.11)
so that, in particular, δ ψ̄ i = + 12 FM N ε̄i ΓM N . The crucial Fierz identity for the supersymmetry
invariance is with the chiral projection matrix, P := 21 (1 + Γ(7) ),



ΓL P αβ (ΓL P )γδ + ΓL P γβ (ΓL P )αδ = 0 , (6.12)

which ensures the vanishing of the terms cubic in ψi ,

tr ψ̄ i ΓL [δAL , ψi ] = tr ψ̄ i ΓL [ iε̄j ΓL ψj , ψi ] = 0 . (6.13)

The equations of motion are

DL F LM + ψ̄ i ΓM ψi = 0 , ΓM DM ψi = 0 . (6.14)

21
6.3 6D super Yang-Mills in the spacetime of arbitrary signature
With
T
ΓM = ±C± ΓM C±−1 , C±T = ∓C± , (6.15)
we have
T
C± ΓM = −C± ΓM . (6.16)
We introduce a pair of Weyl spinors of the same chirality,

(ψ1 , ψ2 ) , Γ(7) ψi = sψi , s2 = 1 , (6.17)

and define the charge conjugate spinor by

ψ̄ci := ǫ−1ij ψjT C± . (6.18)

The super Yang-Mills Lagrangian reads

1


L6D = tr F F MN
4 MN
+ 21 ψ̄ci ΓM DM ψi , (6.19)

and the supersymmetry transformations are given by

δAM = ε̄ic ΓM ψi = −ψ̄ci ΓM εi ,

(6.20)
δψi = − 21 FM N ΓM N εi ,

so that, in particular, δ ψ̄ci = + 12 FM N ε̄ic ΓM N . The Lagrangian transforms as, from (2.60),


δL6D = ∂M tr F M N δAN − 21 ψ̄ci ΓM δψi . (6.21)
∗
Only if B± B± = −1, as in the Minkowskian signature, one can impose the pseudo-Majorana
condition,
ψ̄ci = ψ̄D
i
:= (ψi )† A . (6.22)

22
6.4 (9 + 1)D SYM, its reduction, and 4D superconformal symmetry
• Conventions for (9 + 1)D gamma matrices
Spacetime signature : η = diag(− + + · · · +), mostly plus signature.
32 × 32 Gamma matrices:
i) Hermitian conjugate,

(ΓM )† = ΓM = −Γ0 ΓM Γ0 = AΓM A† ,

A = Γ12···9 = A† = A−1 , (6.23)

1
(AΓM1 M2 ···Mn )† = (−1) 2 n(n−1) AΓM1 M2 ···Mn ,

ii) Complex conjugate,

†
(ΓM )∗ = ±B± ΓM B± ,
(6.24)
T † −1
B± = B± = (B± ) ,

iii) Transpose,
(ΓM )T = ±C± ΓM C±† ,

T
C± = B± A = ±C±T = (C±† )−1 , (6.25)
1
(C+ ΓM1 M2 ···Mn )T = (−1) 2 n(n−1) C+ ΓM1 M2 ···Mn .

Let the spinorial indices be located as

(ΓM )α β , (A)α β , (B± )αβ = (B± )βα , (C± )αβ = ±(C± )βα . (6.26)

Define
Γ(10) = Γ012···9 = (Γ(10) )† = (Γ(10) )−1 = −C+† (Γ(10) )T C+ . (6.27)

The crucial identity for the super Yang-Mills action is

(C+ ΓM Γ± )(αβ (C+ ΓM Γ± )γ)δ = 0 (6.28)

where Γ± = 21 (1 ± Γ(10) ) is either the chiral or the anti-chiral projector, and α, β, γ are
symmetrized. Note also the symmetric property, (C+ ΓM Γ± )T = C+ ΓM Γ± .

23
 For spinors we set
ψ̄ = ψ † A . (6.29)

Majorana-Weyl Spinor, ψ, satisfies

Γ(10) ψ = +ψ : Weyl condition , (6.30)

ψ ∗ = B+ ψ : Majorana condition , (6.31)

or equivalently,
ψ̄Γ(10) = −ψ̄ : opposite chirality ,
(6.32)
T
ψ̄ = ψ C+ .
Hence for the fermionic Majorana-Weyl spinors,

ψ̄1 ΓM1 M2 ···M2n ψ2 = 0 , (6.33)

and9

ψ̄1 ΓM1 M2 ···M2n+1 ψ2 = (−1)n+1 ψ̄2 ΓM1 M2 ···M2n+1 ψ1 = −(ψ̄1 ΓM1 M2 ···M2n+1 ψ2 )† : imaginary .
(6.34)

We can further set

 
M 0 γ̃ M
Γ = , γ M γ̃ N + γ N γ̃ M = 2η M N , η = diag(− + + + · · · +) .
γM 0
(6.35)
M N
Namely, (γ , γ̃ ) are the real 16 × 16 matrices appearing in the off block-diagonal parts
of the 32 × 32 gamma matrices,
satisfying10
(γ M )∗ = γ M , (γ M )T = γ̃ 0 γ M γ̃ 0 = γ̃M ,
(6.36)
0 1 2 9 0 1 2 9
γ̃ γ γ̃ · · · γ = +1 , γ γ̃ γ · · · γ̃ = −1 .
9
When the spinor is Lie algebra valued, Eq.(6.34) does not hold in general.
10
From γ̃ M = (γM )−1 it also follows that γ̃ M γ N + γ̃ N γ M = 2η MN . One may further impose the symmetric
property, (γ M )T = γ M , but it is not necessary in our paper.

24
• Lagrangian.
Let the gauge group be su(N) or u(N).
Lie algebra valued fields,

AM = ApM Tp , Ψ = Ψp Tp , (Tp )† = Tp . (6.37)

Field strength and the covariant derivative are

FM N = ∂M AN − ∂N AM − i[AM , AN ] , DM Ψ = ∂M Ψ − i[AM , Ψ] . (6.38)

Bianchi identity reads

DL FM N + DM FN L + DN FLM = 0 . (6.39)

The gauge symmetry is given by, for g † = g −1 ,

AM → gAM g −1 + ig∂M g −1 , FM N → gFM N g −1 , Ψ → gΨg −1 . (6.40)

The Lagrangian of 10D super Yang-Mills theory reads

 
L = tr − 41 FM N F M N − i 21 Ψ̄ΓM DM Ψ
  (6.41)
= tr − 14 FM N F M N − i 21 ψ̄γ M DM ψ ,

where Ψ ≡ (ψ 0)T and ψ α is a sixteen component spinor and ψ̄ := ψ T γ̃ 0 .

Under arbitrary infinitesimal transformations, δAM , δΨ,


  
δL = tr DL F LM + Ψ̄ΓM Ψ δAM − iΨ̄ΓM DM δΨ + ∂N tr F M N δAM − i 12 δ Ψ̄ΓN Ψ .
(6.42)

• Summary of supersymmetry in D ≤ 10.

The ordinary supersymmetry and kinetic supersymmetry are given by

δAM = iΨ̄ΓM ξ+ = −iξ¯+ ΓM Ψ , δΨ = 12 FM N ΓM N ξ+ + ξ+

′
1N×N , (6.43)

25
so that
δ Ψ̄ = − 21 ξ¯+ FM N ΓM N + ξ¯+
′
1N×N , (6.44)
′
where ξ+ and ξ+ are constant Majornana-Weyl spinors corresponding to the ordinary and
kinetic supersymmetry parameters. + denotes the chirality. The above is the symmetry of
the (9 + 1)D and also any dimensionally reduced super Yang-Mills action.

In four-dimensions of either Minkowskian or Euclidean signature, the supersymmetry gets

enhanced to the superconformal symmetry as
¯
δAM = iΨ̄ΓM E(x) = −iE(x)Γ MΨ , δΨ = 21 FM N ΓM N E(x) − 2Φa Γa ξ− + ξ+′
1N×N ,
(6.45)
where m is for the four-dimensions and a is for the rest. ξ− is a constant Majornana-Weyl
spinor of the opposite chirality corresponding to the special superconformal symmetry pa-
rameter, and
E(x) = xm Γm ξ− + ξ+ . (6.46)

In any case, the conserved supercurrent is of the universal form,

J M = −itr Ψ̄ΓM δΨ = +itr δ Ψ̄ΓM Ψ . (6.47)

In Appendix C, we present the derivation.

26
 • Superconformal symmetry in 4D of arbitrary signature.
The 32 supersymmetries in 4D super Yang-Mills which consist of ordinary supersymmetry
and special superconformal symmetry read
¯ + xm Γm )ΓM Ψ ,
δAM = iΨ̄ΓM (1 + xm Γm )ξ = −iξ(1
1 
δΨ = 21 (1 + Γ(10) ) 2
FM N Γ MN
(1 + xm
Γ m ) − 2Φa Γ a
ξ, (6.48)
 
δ Ψ̄ = ξ¯ − 12 (1 + xm Γm )FM N ΓM N − 2Φa Γa 21 (1 − Γ(10) ) ,

where ξ is a 32 component Majorana spinor,

ξ ∗ = B+ ξ . (6.49)

The chiral decomposition of the spinor gives the ordinary supersymmetry and special su-
perconformal symmetry,11

ξ = ξ+ + ξ− , ξ± = 21 (1 ± Γ(10) )ξ . (6.50)

The 32 component Majorana supercurrent is of the form,

¯ M,
J M = +iQ̄M ξ = −iξQ
 1


QM = tr 2
(1 + xm Γm )FKL ΓKL + 2Φa Γa ΓM Ψ ,


Q̄M = tr Ψ̄ΓM − 12 FKL ΓKL (1 + xm Γm ) + 2Φa Γa = (QM )† A = (QM )T C+ .
(6.51)
The supercharge is given by Z
Q= d3 x Q0 . (6.52)

11
Note also E(x) = 12 (1 + Γ(10) )(1 + xm Γm )ξ .

27
 APPENDIX
A Proof of the Theorem (2.1)
Theorem (2.1):
Any N × N matrix, M, satisfying M 2 = λ2 1N×N , λ 6= 0, is diagonalizable.

Proof
Suppose for some K, 1 ≤ K ≤ N, we have found a basis,

{ea , vr : 1 ≤ a ≤ K, 1 ≤ r ≤ N − K} (A.1)

such that
Mea = λa ea , for 1 ≤ a ≤ K ,
(A.2)
s a
Mvr = P r vs + h r ea , for K + 1 ≤ r, s ≤ N .
From M 2 = λ2 1N×N ,
λ2a = λ2 ,
(A.3)
2 2 s a a
λ vr = (P ) r vs + [(hP ) r + λa h r ] ea ,
and hence,
P 2 = λ2 1(N−K)×(N−K) ,
(A.4)
a a
(hP ) r + λa h r = 0.
The assumption holds for K = 1 surely. In order to construct eK+1 we first consider an eigenvec-
tor of the (N − K) × (N − K) matrix, P ,

P r s cs = λK+1 cr , λ2K+1 = λ2 , (A.5)

and set
v = cr vr , ha = ha r cr ,
(A.6)
a
Mv = λK+1v + h ea .
Consequently
(λK+1 + λa )ha = 0 : not a sum , (A.7)
so that
ha = 0 if λK+1 + λa 6= 0 . (A.8)

28
We construct eK+1 , with K unknown coefficients, da , as

eK+1 = v + da ea . (A.9)

From
MeK+1 = λK+1eK+1 + [ha + (λa − λK+1 )da ] ea , (A.10)
we determine 
 ha

 if λK+1 6= λa ,
λK+1 − λa
da = (A.11)


 any number if λK+1 = λa .
From (A.8) and λ2K+1 = λ2a = λ2 6= 0, we have

MeK+1 = λK+1eK+1 . (A.12)

This completes our proof.

If we set a N × N invertible matrix, S, by

(S)b a = (ea )b , Mea = λa ea , 1 ≤ a, b ≤ N , (A.13)

then
S −1 MS = diag(λ1 , λ2 , · · · , λN ) . (A.14)

29
B Gamma matrices in 4,6,10,12 dimensions
Our conventions are such that
γ̂ m : m = 0, 1, 2, 3 for 1 + 3D ,

γµ : µ = 1, 2, · · · , 6 for 2 + 4D ,

γa : a = 7, 8, · · · , 12 for 0 + 6D , (B.1)

ΓM : M = 0, 1, 2, 3, 7, · · · , 12 for 1 + 9D ,

ΓM : M = 1, 2, · · · , 12 for 2 + 10D .

B.1 Four dimensions

In Minkowskian four dimension of the metric, η̂ = diag(− + ++), the gamma matrices satisfy

γ̂ m γ̂ n + γ̂ n γ̂ m = 2η̂ mn , (γ̂ m )† = γ̂m , (B.2)

where m, n = 0, 1, 2, 3. The chiral matrix reads

γ̂ (5) = −iγ̂ 0123 = (γ̂ (5) )−1 = (γ̂ (5) )† . (B.3)

The three pairs of unitary matrices, ± , B̂± , Ĉ± , relate the hermitain conjugate, complex
conjugate, and the transpose of the gamma matrices,

±(γ̂ m )† = ± γ̂ m †± , †± ± = 1 ,

† †
±(γ̂ m )∗ = B̂± γ̂ m B̂± , B̂± B̂± = 1 , (B.4)

±(γ̂ m )T = Ĉ± γ̂ m Ĉ±† , Ĉ±† Ĉ± = 1 .

Especially in Minkowskian four dimensions, they can be chosen further to satisfy

Â+ = −iγ 123 , Â− = −γ̂ 0 , Â− = Â+ γ̂ (5) ,

∗ T
B̂± B̂± = ±1 , B̂± = ±B̂± , B̂− = B̂+ γ̂ (5) , (B.5)

T T
Ĉ± = B̂+ ± = B̂± Â+ , Ĉ±T = −Ĉ± , Ĉ− = Ĉ+ γ̂ (5) .

30
B.2 Four to six dimensions
Using the four dimensional gamma matrices above, one can construct the six dimensional gamma
matrices in the off-block diagonal form,
 
µ 0 ρµ
γ = , µ = 1, 2, · · · , 6 , ρµ ρ̄ν + ρν ρ̄µ = 2η µν . (B.6)
ρ̄µ 0
With the relevant choice of the metric,

η = diag(− − + + ++) , (B.7)

we require ρ̄µ = (ρµ )† and set

γ 1 = U (−iτ2 ⊗ 1) U † , γ m+2 = U (τ1 ⊗ γ̂ m ) U † ,

  (B.8)
6 (5)

† Ĉ+ 0
γ = U τ1 ⊗ γ̂ U , U= .
0 1
Explicitly with (B.3), (B.5)

ρ1 = −Ĉ+ , ρm+2 = Ĉ+ γ̂ m , ρ6 = Ĉ− ,

(B.9)
1
ρ̄ = +Ĉ+−1 , m+2
ρ̄ = γ̂ m
Ĉ+−1 , ρ̄ =6
Ĉ−−1 .
Note  
(7) 1 2 6 1 0
γ = iγ γ · · · γ = , (B.10)
0 −1
and especially the anti-symmetric property of the 4 × 4 matrices,

(ρµ )αβ = −(ρµ )βα , (ρ̄µ )αβ = − 21 ǫαβγδ (ρµ )γδ . (B.11)

The spinorial indices, α, β = 1, 2, 3, 4, denote the fundamental representation of su(2, 2). It

follows that {ρµ } and {ρ̄µ } separately form bases for the anti-symmetric 4 × 4 matrices with the
completeness relation,

tr(ρµ ρ̄ν ) = 4δ µν , (ρµ )αβ (ρ̄µ )γδ = 2(δα δ δβ γ − δβ δ δα γ ) . (B.12)

On the other hand, Eq.(B.8) implies that12

ρ[µ ρ̄ν ρλ] = +i 61 ǫµνλστ κ ρ[σ ρ̄τ ρκ] , ρ̄[µ ρν ρ̄λ] = −i 16 ǫµνλστ κ ρ̄[σ ρτ ρ̄κ] , (B.13)
12
We put ǫ123456 = 1 and “[ ]” denotes the standard anti-symmetrization with “strength one”.

31
so each of the sets ρ[µ ρ̄ν ρλ] ≡ ρµνλ or ρ̄[µ ρν ρ̄λ] ≡ ρ̄µνλ has only 10 independent components and
forms a basis for symmetric 4 × 4 matrices,
[µ λ]
tr(ρµνλ ρ̄στ κ ) = −i4 ǫµνλ στ κ − 24δ σ δ ντ δ κ ,
(B.14)
µνλ γδ γ δ γ δ
(ρ )αβ (ρ̄µνλ ) = −24(δα δβ + δβ δα ) .
Finally, {ρµν ≡ 21 (ρµ ρ̄ν − ρν ρ̄µ )} or {ρ̄µν ≡ 12 (ρ̄µ ρν − ρ̄ν ρµ )} forms an orthonormal basis for the
general 4 × 4 traceless matrices,
tr(ρµν ρλκ ) = 4(δ µ κ δ ν λ − δ ν κ δ µ λ ) , − 18 (ρµν )α β (ρµν )γ δ + 41 δα β δγ δ = δα δ δγ β , (B.15)
satisfying
(ρ̄µν )α β = −(ρµν )β α . (B.16)

B.3 Six dimensions

The result above can be straightforwardly generalized to other signatures in six dimensions. In
Euclidean six dimensions, gamma matrices satisfy
γ a γ b + γ b γ a = 2δ ab , (B.17)
where we set a, b run from 7 to 12, instead of 1 to 6, as the latter have been reserved for so(2, 4).
With the choice,  
(7) 7 8 12 1 0
γ = iγ γ · · · γ = , (B.18)
0 −1
the six dimensional gamma matrices are in the block diagonal form,
 
a 0 ρa
γ = , (B.19)
ρ̄a 0
satisfying the hermiticity conditions,
ρ̄a = (ρa )† . (B.20)
a a
We can further set all the 4 × 4 matrices, ρ , ρ̄ to be anti-symmetric [7]
(ρa )α̇β̇ = −(ρa )β̇ α̇ , (ρ̄a )α̇β̇ = − 12 ǫα̇β̇ γ̇ δ̇ (ρa )γ̇ δ̇ , (B.21)

which makes the relation, su(4) ≡ so(6), manifest. That is, the indices, α̇, β̇ = 1, 2, 3, 4, denote
the fundamental representation of su(4).

Note that precisely the same equations as (B.12)-(B.16) hold for the so(6) gamma matrices,
{ρ , ρ̄b } after replacing µ, ν, α, β by a, b, α̇, β̇, etc.
a

32
B.4 Ten dimensions again
Using the four and six dimensional gamma matrices above, we write the ten dimensional gamma
matrices,
Γm = γ̂ m ⊗ γ (7) for m = 0, 1, 2, 3
(B.22)
a a
Γ = 1 ⊗γ for a = 7, 8, 9, 10, 11, 12 .
In the above choice of gamma matrices, we have from (6.27), (B.3), (B.18)

Γ(10) = γ̂ (5) ⊗ γ (7) , (B.23)

and
A = Â+ ⊗ 1 , B± = C± A ,
   
0 +1 0 +1
B+ = B̂− ⊗ , B− = B̂+ ⊗ ,
−1 0 +1 0 (B.24)
   
0 −1 0 +1
C+ = Ĉ− ⊗ , C− = Ĉ+ ⊗ .
+1 0 +1 0
Majorana spinor is now of the form,
 α

ψ+ α̇
−1 ∗ † β α̇
Ψ = B+ Ψ = , (ψ+ )α α̇ = (B̂− )αβ ψ− , (B.25)
αα̇
ψ−

where α is the so(1, 3) spinor index and ± denote the so(6) chirality.

Further to have 10 dimensional Majorana-Weyl spinor, imposing the chirality condition, Γ(10) Ψ =
Ψ, we also have
γ̂ (5) ψ± = ±ψ± . (B.26)
For the later convenience, we define ψαα̇ , ψ̄ αα̇ by
β
ψαα̇ = i(Ĉ+ )αβ ψ+ α̇ , ψ̄ αα̇ = ψ−
αα̇
. (B.27)

The Majorana condition is equivalent to

ψ̄ αα̇ = Aα β (ψ † )β α̇ , A = iÂ− = A† = A−1 . (B.28)

33
B.5 Twelve dimensions
In order to make the SO(2, 4) × SO(6) isometry of AdS5 × S 5 geometry manifest, it is convenient
to employ the twelve dimensional gamma matrices of spacetime signature, (− − + + + + + +
+ + ++), and write them in terms of two sets of six dimensional gamma matrices, {γ µ }, {γ a },
which we reviewed above,
Γµ = γ µ ⊗ γ (7) for µ = 1, 2, 3, 4, 5, 6
(B.29)
Γa = 1 ⊗ γ a for a = 7, 8, 9, 10, 11, 12 .
In the above choice of gamma matrices, the twelve dimensional charge conjugation matrices,
C± , are given by
   
M T M −1 0 1 0 1
±(Γ ) = C± Γ C± , M = 1, 2, · · · , 12, C± = ⊗ ,
±1 0 ∓1 0
(B.30)
while the complex conjugate matrices, A± , read
 t   
A 0 1 0
A± = ⊗ , A = −iρ̄12 = −iγ̂ 0 = iÂ− = A† = A−1 , (B.31)
0 ∓A 0 ±1
satisfying
±(ΓM )† = A± ΓM A−1
± . (B.32)
In particular, for µ = 1, 2, · · · , 6, we have
(ρµ )† = −Aρ̄µ At = ρ̄µ , (ρ̄µ )† = −At ρµ A = ρµ . (B.33)
Now if we define the twelve dimensional chirality operator as
Γ(13) ≡ γ (7) ⊗ γ (7) , (B.34)
then
{Γ(13) , ΓM } = 0 , C− = C+ Γ(13) , A− = A+ Γ(13) . (B.35)
In 2+10 dimensions it is possible to impose the Majorana-Weyl condition on spinors to have
sixteen independent complex components which coincides with the number of supercharges in
the AdS5 × S 5 superalgebra, su(2, 2|4). Up to the redefinition of the spinor by a phase factor,
there are essentially two choices for the Majorana-Weyl condition depending on the chirality,
Ψ = ±Γ(13) Ψ , and Ψ̄ = Ψ† A+ = ΨT C+ . (B.36)
Our choice will be the plus sign so that the 2 + 10 dimensional Weyl spinor carries the same chiral
indices for su(2, 2) and su(4), i.e. Ψ = (ψαα̇ , ψ̄ αα̇ )T , while the Majorana condition relates them
as ψ̄ αα̇ = Aα β (ψ † )β α̇ which is identical to (B.28). Hence, the Majorana-Weyl spinor in 2 + 10
dimenisons can be identified as the Majorana spinor in 1 + 9 dimensions.

34
C Looking for the general odd symmetry
With a Majorana-Weyl spinor, E, ∆Ψ , which may depend on xM , we focus on the following
transformations,
¯ MΨ ,
δAM = iΨ̄ΓM E = −iEΓ δΨ = 21 FM N ΓM N E + ∆Ψ , (C.1)

so that
δ Ψ̄ = − 12 ĒFM N ΓM N + ∆Ψ . (C.2)
Note that ∆Ψ is Lie algebra valued, while E is not.

From

Ψpα Ψqβ Ψrγ tr(Tp Tq Tr ) = Ψpγ Ψqα Ψrβ tr(Tp Tq Tr ) = Ψpβ Ψqγ Ψrα tr(Tp Tq Tr ) , (C.3)

and the identity (6.28), we note that the second term in (6.42) vanishes


tr Ψ̄ΓM Ψ Ψ̄ΓM E = 0 . (C.4)

We also get, using the Bianchi identity (6.39),

Ψ̄ΓM DM δΨ = 21 DL FM N Ψ̄(ΓLM N + 2η LM ΓN )E + 12 Ψ̄ΓL ΓM N ∂L EFM N + Ψ̄ΓL DL ∆Ψ

= −iDM F M N δAN + 21 Ψ̄ΓL ΓM N ∂L EFM N + Ψ̄ΓL DL ∆Ψ .

(C.5)

Thus, semi-finally, we obtain

   
δL = −i tr 12 FM N Ψ̄ΓL ΓM N ∂L E + Ψ̄ΓL DL ∆Ψ + ∂N tr F M N δAM + i 12 Ψ̄ΓN δΨ . (C.6)

We first note that constant E, and constant ∆Ψ which is central in the Lie algebra lead to the
ordinary and kinetic supersymmetries

E, ∆Ψ : constant and ∆Ψ ∝ 1N×N . (C.7)

Henceforth, keeping the dimensional reduction either to Minkowskian d-dimensions, 0 ≤

m ≤ d − 1, d ≤ a ≤ 9, or Euclidean d-dimensions, 1 ≤ m ≤ d, a = 0, d + 1 ≤ a ≤ 9, we set

35
Aa = Φa , “ ∂a ≡ 0 ”, and look for some possibilities of more general symmetries.

Since

FM N ΓL ΓM N ∂L E = Fmn Γl Γmn + 2Dm Φb Γl Γmb + Da Φb Γl Γab ∂l E , (C.8)
we first require
Γl Γmn ∂l E = 0 , (C.9)
or equivalently
Γmn Γl ∂l E = 2Γm ∂ n E − 2Γn ∂ m E . (C.10)
It follows after multiplying Γnm without m, n summing,

Γl ∂l E = 2Γm ∂m E + 2Γn ∂n E : no sum for m 6= n . (C.11)

Eqs.(C.9), (C.10), (C.11) are trivial when d = 0, 1. For d ≥ 2, summing over m 6= n in (C.11) we
get
(d − 1)(d − 4)Γl ∂l E = 0 . (C.12)
Hence, for d = 2, 3, d ≥ 5,

Γm ∂m E = −Γn ∂n E : no sum and m 6= n . (C.13)

• For d = 3, d ≥ 5, we easily conclude ∂m E = 0, i.e. constant parameter, E.

• When d = 2, we get
∂m E = −Γmn ∂ n E : for d = 2 , (C.14)
so that
∂ m ∂m E = 0 . (C.15)
Let σ 6= τ be the two different spacetime indices in d = 2 case. Eq.(C.9) is simply
equivalent to
(∂σ + Γσ τ ∂τ )E = 0 . (C.16)
This can be solved easily in the diagonal basis of Γσ τ . In the Minkowskian two-dimensions,
as Γ0 1 is hermitian, the solution is given by the left and right modes, σ ± τ . On the other
hand, in the Euclidean two-dimensions, Γ1 2 is anti-hermitian and the solution involves
holomorphic functions, σ ± iτ .

• For d = 4 we have for any m,

∂m E = Γm ξ− , ξ− = 41 Γl ∂l E . (C.17)

36
 From ∂[m ∂n] E = 0 we get an essenitally same relation as (C.13),
Γm ∂m ξ− = −Γn ∂n ξ− : no sum and m 6= n . (C.18)
Hence, ξ− is a constant spinor, and
E = xm Γm ξ− + ξ+ , (C.19)
where ξ+ , ξ− are constant Majorana-Weyl spinors of the opposite chiralities, corresponding
to the ordinary supersymmetry and special superconformal symmetry, respectively.
Provided the above solutions for (C.9), we are ready for the full analysis.
1. When d = 0 : IKKT matrix model.
Eq.(C.8) becomes trivial, and we natually require
Γa [Φa , ∆Ψ ] = 0 . (C.20)
We need to find the algebraic solution for ∆Ψ in terms of the Lie algebra valued fields, Φa ,
d ≤ a ≤ 9. Clearly, the kinetic supersymmetry transformation, i.e. ∆Ψ ∝ 1N×N , satisfies
the above equation. In fact, we can show that this is the most general solution.

Proof
We consider the special case, Φa = 0, d ≤ a ≤ 7. Eq.(C.20) gives
[Φ8 , Γ8 ∆Ψ ] + [Φ9 , Γ9 ∆Ψ ] = 0 . (C.21)
Multiplying Φ8 and taking the u(N) trace we get
tr ([Φ8 , Φ9 ]∆Ψ ) = 0 . (C.22)
Since the commutator, [Φ8 , Φ9 ], can be arbitrary except 1N×N , we conclude that ∆Ψ ∝
1N×N . This completes our proof.

Therefore, when d = 0, E and ∆Ψ are simply constant Majorana-Weyl spinors correspond-

ing to the ordinary and the kinetic supersymmetries.
2. When d = 1 : BFSS matrix model.
Eq.(C.9) is trivial, and with the coordinate, τ for d = 1, From Eq.(C.6) we require
0 = 12 FM N ΓL ΓM N ∂L E + ΓL DL ∆Ψ
(C.23)
τ τa b 1
= Γ Dτ (∆Ψ + Φa Γ ∂τ E) + Γ Db (∆Ψ − Φ Γτ a ∂τ E)
2 a
a τ
− Φa Γ ∂ ∂τ E .
The only possible algebraic solutions are (C.7) corresponding to the ordinary and the kinetic
supersymmetries.

37
 3. When d = 2.
From (C.6) we require, using (C.14), (C.15),

0 = 21 FM N ΓL ΓM N ∂L E + ΓL DL ∆Ψ
(C.24)
= Γm Dm ∆Ψ + Γa Da ∆Ψ + 2(D τ Φa − D σ Φa Γσ τ )Γa ∂τ E .

We conclude again that the only possible algebraic solutions are (C.7) corresponding to the
ordinary and the kinetic supersymmetries.

4. When d = 3, d ≥ 5.
Since E is constant, the only possible algebraic solutions are (C.7) corresponding to the
ordinary and the kinetic supersymmetries.

5. When d = 4.
From (C.6) we require, using (C.19),

0 = 12 FM N ΓL ΓM N ∂L E + ΓL DL ∆Ψ
(C.25)
L a
= Γ DL (∆Ψ + 2Φa Γ ξ− ) .

Thus the algebraic solution reads

∆Ψ + 2Φa Γa ξ− ∝ 1N×N . (C.26)

Acknowledgments
This work is supported by Basic Science Research Program through the National Research Foun-
dation of Korea Grants, NRF-2016R1D1A1B01015196 and NRF-2020R1A6A1A03047877 (Cen-
ter for Quantum Space Time).

38
References
[1] J. Strathdee. Extended Poincaré Supersymmetry. Int. J. Mod. Phys. A, 2: 273, 1987.

[2] T. Kugo and P. Townsend. Supersymmetry and the Division Algebras. Nucl. Phys.,
B221: 357, 1983.

[3] J. Scherk F. Gliozzi and D. Olive. Supersymmetry, Supergravity Theories and the Dual Spinor
Model. Nucl. Phys., B122: 253, 1977.

[4] H. Weyl. The Classical Groups. Princeton University Press, 1946.

[5] B. Dewitt. Supermanifolds. Cambridge University Press, 1984.

[6] J. F. Cornwell. Group Theory in Physics. Academic Press, 1989. See volume III.

[7] Jeong-Hyuck Park. Superconformal symmetry in six-dimensions and its reduction to four-
dimensions. Nucl. Phys. B, 539:599–642, 1999.

39