G2 holonomy manifolds are superconformal

Lázaro O. Rodríguez Díaz∗.

arXiv:1606.09534v1 [math.QA] 30 Jun 2016

Abstract
We study the chiral de Rham complex (CDR) over a manifold M with
holonomy G2 . We prove that the vertex algebra of global sections of the
CDR associated to M contains two commuting copies of the Shatashvili-
Vafa G2 superconformal algebra. Our proof is a tour de force, based on
explicit computations.

Contents
1 Introduction 3

2 Vertex superalgebras 6
2.1 The Shatashvili-Vafa G2 superconformal algebra . . . . . . . . . 11
2.2 The Shatashvili-Vafa Spin(7) superconformal algebra . . . . . . . 12

3 The Chiral de Rham Complex 13

4 G2 holonomy manifolds 16

5 The Chiral de Rham Complex over a G2 -manifold 18

5.1 G2 holonomy manifolds are superconformal . . . . . . . . . . . . 19
5.2 Proof of the Conjecture . . . . . . . . . . . . . . . . . . . . . . . 25
5.2.1 Linear λ-brackets . . . . . . . . . . . . . . . . . . . . . . 26
5.2.2 Non linear λ-brackets . . . . . . . . . . . . . . . . . . . . 29
5.2.3 Checking the relation . . . . . . . . . . . . . . . . . . . . 32

6 Computations 32
6.1 [Φ+ λ Φ+ ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.2 X+ (2) Φ+ and X+ (1) Φ+ . . . . . . . . . . . . . . . . . . . . . . 37
6.3 [Φ± λ K± ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3.1 (Φ+ (1) K+ ) + (Φ− (1) K− ) . . . . . . . . . . . . . . . . . 44
6.3.2 Φ± (1) K± . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.4 [G+ λ G− ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
∗
IMECC-UNICAMP, São Paulo. Supported by São Paulo State Research Council (Fapesp) grant
2014/13357-7.

1
 6.4.1 Coefficient of λ2 . . . . . . . . . . . . . . . . . . . . . . 53
6.4.2 Coefficient of λ . . . . . . . . . . . . . . . . . . . . . . . 53
6.4.3 Coefficient of λ0 . . . . . . . . . . . . . . . . . . . . . . 54
6.5 [G+ λ Φ− ] and [G− λ Φ+ ] . . . . . . . . . . . . . . . . . . . . . . . 58
6.5.1 Coefficient of λ . . . . . . . . . . . . . . . . . . . . . . . 62
6.5.2 Coefficient of λ0 . . . . . . . . . . . . . . . . . . . . . . 62
6.6 [G+ λ Φ+ ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.6.1 Coefficient of λ2 . . . . . . . . . . . . . . . . . . . . . . 64
6.6.2 Coefficient of λ . . . . . . . . . . . . . . . . . . . . . . . 65
6.7 [L+ λ Φ+ ] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2
1 Introduction
The chiral de Rham complex (CDR) was introduced in [MSV99] by Malikov,
Schechtman and Vaintrob. It is a sheaf of supersymmetric vertex algebras over
a smooth manifold M . Locally, over a coordinate chart, it is simply n copies of the
bc − βγ system (n is the dimension of M ), i.e., a tensor product of n copies of the
Clifford vertex algebra and n copies of the Weyl vertex algebra, and then extended
to M by gluing on the intersections of these coordinate charts.
One of the most important facts about the CDR proved in the seminal work
[MSV99] is that there exists an embedding of the N = 2 superconformal vertex
algebra into the vertex algebra of global sections of CDR when M is a Calabi-
Yau manifold. This idea of looking for special vertex subalgebras of the vertex
algebra of global sections of the CDR was further investigated in [BZHS08] where
it was proved that when M is a hyperkälher manifold, the N = 4 superconformal
vertex algebra appears as a subalgebra of global sections of CDR. Subsequently in
[Hel09] it was shown that in fact there are two commuting copies of the N = 2
superconformal algebra (N = 4 superconformal algebra) of half the central charge
when M is Calabi-Yau (respectively hyperkälher).
It is by now well known in the physics literature ([Oda89], [HP91], [HP93],
[SV95]) that even though we can define a sigma model on an arbritary target space,
in order for the theory to be supersymmetric, the target space manifold must be of
special holonomy. This implies the existence of covariantly constant p-forms, and
the existence of such forms on the target space manifold implies the existence of
extra elements in the chiral algebra.
In [EHKZ13] a program was launched in order to understand the facts men-
tioned above as a relation between special holonomy of M and the existence of
certain subalgebras of CDR. To pursue this objective they introduced an embed-
ding (Theorem 3.2) different than the one in [MSV99], of the space of differential
forms Ω∗ (M ) into global sections of CDR (in fact, they introduced two different
embeddings). When the manifold M has special holonomy it admits covariantly
constant forms, so they obtain corresponding sections of CDR and the subalgebra
generated by them. In particular, they recover the result of [BZHS08, Hel09] when
M is a Calabi-Yau manifold or hyperkälher. They also constructed two commuting
copies of the Odake algebra on the space of global sections of CDR of a Calabi-Yau
threefold, and conjectured a similar result for G2 holonomy manifolds.
More precisely they conjectured [EHKZ13, Conjecture 7.3] (Conjecture 5.1)
that: if M is a manifold with G2 holonomy, the vertex algebra of global sections
of CDR contains two commuting copies of the Shatashvili-Vafa G2 superconformal
algebra, each of these copies should be generated by the global section that comes
from the covariantly constant 3-form (that defines the geometry) using the two
different embeddings of Ω∗ (M ) that they defined.
The Shatashvili-Vafa G2 superconformal algebra appeared as the chiral algebra
associated to the sigma model with target a manifold with G2 holonomy in [SV95],
its classical counterpart had been studied in [HP93]. It is a superconformal vertex

3
algebra with six generators {L, G, Φ, K, X, M }. It is an extension of the N = 1
superconformal algebra of central charge c = 21/2 (formed by the super-partners
{L, G}) by two fields Φ and K, primary of conformal weight 32 and 2 respectively,
and their superpartners X and M (of conformal weight 2 and 52 respectively). Their
OPEs can be found in subsection 2.1 in the language of lambda brackets of [DK98].
This algebra is a member of a two-parameter family SW ( 32 , 32 , 2) of non-linear
W -algebras previously studied in [Blu92], it is parametrized by (c, ε) (c is the
central charge and ε the coupling constant). The Shatashvili-Vafa G2 algebra is a
quotient of SW ( 23 , 32 , 2) with c = 21
2 and ε = 0, in other words is the only one
among this family which has central charge c = 21 2 and contains the tri-critical
Ising model as a subalgebra.
The above conjecture was checked in [EHKZ13] in the case when the manifold
M = R7 is the flat space, and when M = CY3 × S 1 , where CY3 is a compact
Calabi-Yau threefold and S 1 is a circle, using the above mentioned result about the
Odake algebra.
In this paper we prove the conjecture when M is an arbitrary non-flat G2 man-
ifold. Our approach is a tour de force, based on explicit computations, some of
them really long, and some abstract algebraic manipulations. But the beauty of
these after all is that we need to use many of the known identities in G2 geometry,
all the computations are tied in a non-trivial way to the geometry of the manifold.
To perform the computations we have used the Mathematica package OPEdefs
created by Kris Thielemans [Thi91] for symbolic computation of operator product
expansion, and the computer algebra system Cadabra [Pee07] created by Kasper
Peeters.
In section 2 we recap the basic facts about vertex superalgebras and we intro-
duce the main examples that are used in the paper. In section 3 we review the
construction of the CDR as well as the main tool (Theorem 3.2) used in the paper:
the embedding of the space of differential forms Ω∗ (M ) into global sections of
CDR, introduced by [EHKZ13]. In section 4 we recall some background material
on G2 geometry as well as many of the known identities necessary for the compu-
tations in the next section. In section 5 we prove the main result of this paper:
Theorem. Let M be a G2 holonomy manifold, then the space of global sections of
the CDR associated to M contains two commuting copies of the Shatashvili-Vafa
G2 superconformal algebra of central charge 21
2 .

The reader is referred to Theorem 5.5 for a more explicit description of the
generators of these pairs of algebras. It is important to note that the Conjecture 5.1
provides the explicit candidates (global sections of the CDR) for the generators,
the challenge is to verify that they actually generate the Shatashvili-Vafa G2 super-
conformal algebra and that the two copies commute.
We develop the proof of Theorem 5.5 in four steps. First we prove in Theo-
rem 5.2 that the space of global sections of CDR contains two commuting pairs of
N = 1 superconformal algebras, at central charge 21/2 and 7/10 respectively, the
last one is precisely the tri-critical Ising model. Secondly in subsection 5.2.1 we

4
check the linear λ-brackets that were not already computed in the way of proving
Theorem 5.2. In third place we verify the non-linear λ-brackets under the assump-
tion a non-linear identity (2.13) among the generators is satisfied. Finally in 5.2.3
we check the non-linear identity (2.13), see Remark 5.7. Along the way of proving
Theorem 5.5 we perform a lot of computations in local coordinates, for this we
make extensive use of properties of the manifold M such as: the Ricci flatness, the
contractions between the 3-form and its Hodge dual (see page 18), the 3-form is
parallel, the symmetries of the Riemann curvature tensor, etc. To avoid unneces-
sary calculations we take advantage of the space of global sections of the CDR is
a vertex algebra, to use the axioms as well as the identities satisfied by a vertex
algebra, e.g., the Jacobi identity, the Borcherds identity, etc, to derive λ-brackets
from the ones already computed.
It is remarkable that Borcherds identity was proved very useful computing the
non-linear λ-brackets in subsection 5.2.2. It is also engaging the appearance of
the first Pontryagin class p1 (M ) of the manifold M in the proof of the non-linear
identity (2.13) (see Remark 5.8), taking into account that the only oriented charac-
teristic class of interest for a G2 -manifold is p1 (M ) [CHNP15].
A new feature of the Shatashvili-Vafa G2 superconformal algebra not shared
by their cousins the N = 2 superconformal algebra (the N = 4 superconformal al-
gebra) in the Calabi-Yau manifold case (respectively in the hyperkälher case) is the
existence of non-linear λ-brackets among the generators, i.e., in some λ-brackets
appear the normally ordered product of generators. This is a direct consequence of
the geometry of the manifolds, in an almost Hermitian manifold even if the metric
and the complex structure satisfied some compatibility condition they are essen-
tially independent, though in the G2 case the metric and the cross product are both
determined in a highly nonlinear way from the 3-form.
It is worth to mention that our proof of the conjecture is more tortuous than
the proofs of previous results for Calabi-Yau and hyperkälher manifolds [BZHS08,
Hel09, EHKZ13], this is due to the lack of ‘good’ coordinates systems in the G2 -
holonomy case, it would be interesting to explore if there exist some coordinate
system in which the quantum corrections to the fields (3.4) and (3.5) get simplified
or even vanish as happens when M is Calabi-Yau or hyperkälher.
Another related question is the following. There is a proposal of how to define a
topological twist of the Shatashvili-Vafa G2 superconformal algebra in [SV95], in
fact, there is a more recent approach [dBNS08]. There are known a few necessary
topological conditions a manifold with holonomy G2 should satisfied, but we are
far from knowing sufficient conditions. Question: can we put the ‘topological’
Shatashvili-Vafa G2 superconformal algebra locally inside the bc − βγ systems in
such a way the obstruction to be globally well defined as a subalgebra of the CDR
be sufficient conditions for having holonomy G2 ? In the Calabi-Yau case this is
exactly what happens, the vanishing of the first Chern class of the manifold is the
obstruction to having the topological N = 2 superconformal algebra globally well
defined [MSV99, Section 4].
Among the special holonomy manifolds we are missing the Spin(7) case in

5
dimension 8. In view of the present work is reasonable to expect that an analogous
theorem can be proved, i.e., the space of global sections of the CDR associated to
a Spin(7)-holonomy manifold contains two commuting copies of the Shatashvili-
Vafa Spin(7) superconformal algebra (subsection 2.2) of central charge 12 (see
Remark 5.3). It follows easily from our result that for a Spin(7)-manifold of the
form M × R where M is a G2 -holonomy manifold the above claim is true.
Acknowledgements: The author would like to thank Reimundo Heluani for
generously sharing his insights and ideas and for the encouragement despite the
long computations; the present paper is influenced by his views about the subject.

2 Vertex superalgebras
In this section we recall some facts about vertex superalgebras, for details the
reader is referred to [Kac98, DSK06].
Let V be a vector superspace, i.e., a vector space with a decomposition V =
V0̄ ⊕ V1̄ for 0̄, 1̄ ∈ Z/2Z. We call V0 the even space and V1 the odd space. If
v ∈ Vα , one writes p(v) = α and calls it the parity of v.
The algebra EndV acquires a Z/2Z grading by letting

(EndV )ᾱ = {A ∈ EndV : A(Vβ̄ ) ⊂ Vᾱ+β̄ }

for ᾱ, β̄ ∈ Z/2Z.
An EndV -valued field is a formal distribution of the form
X
a(z) = a(n) z −n−1 , a(n) ∈ EndV
n∈Z

such that for every v ∈ V , we have a(n) v = 0 for large enough n.

We now proceed to give two different definitions of a vertex algebra that we
use freely through the text, for a proof of the equivalence of these definitions we
refer to [DSK06].

Definition 2.1. A vertex superalgebra consists of the data of a vector superspace

V (the space of states) , |0i ∈ V0̄ is a vector (the vacuum vector), an even endomor-
phism T , and a parity preserving linear map a → Y (a, z) from V to EndV -valued
fields (the state-field correspondence). This data should satisfy the following set of
axioms:

• Vacuum axioms:

Y (|0i , z) = Id, Y (a, z) |0i |z=0 = a, T |0i = 0.

• Translation invariance:

[T, Y (a, z)] = ∂z Y (a, z).

6
 • Locality:
(z − w)n [Y (a, z), Y (b, w)] = 0 n ≫ 0.

Given a vertex superalgebra V and a ∈ V we expand the fields

X
Y (a, z) = a(n) z −n−1 (2.1)
n∈Z

and we call the endomorphisms a(n) the Fourier modes of Y (a, z).
In any vertex algebra we have the following identity, known as the Borcherds
identity:
  
X
j n  
n p(a)p(b)

(−1) a(m+n−j) b(k+j) c − (−1) (−1) b(n+k−j) a(m+j) c
j
j∈Z+
X m  
= a(n+j) b c, (2.2)
j (m+k−j)
j∈Z+

for all a, b, c ∈ V , m, n, k ∈ Z. In fact this identity appeared in the original

Borcherds definition of vertex algebras [Bor86].
Given a vertex algebra V we define two operations:
X λj
[aλ b] = a(j) b, : ab := a(−1) b. (2.3)
j!
j≥0

The first is called the λ-bracket and the second is called the normally ordered prod-
uct. These operations, the vacuum vector |0i and the derivation T determine the
structure of the vertex algebra as follows from the next definition of vertex algebra.
We need to introduce the definition of a Lie conformal algebra.

Definition 2.2. A Lie conformal superalgebra R = R0̄ ⊕ R1̄ is a Z/2Z-graded

C[∂]-module endowed with a parity preserving C-bilinear λ-bracket

[.λ .] : R ⊗ R → R ⊗ C[λ] (2.4)

which satisfies:

(i) sesquilinearity:

[∂aλ b] = −λ[aλ b], [aλ ∂b] = (λ + ∂)[aλ b], (2.5)

(ii) skewsymmetry:
[bλ a] = −(−1)p(a)p(b) [a−∂−λ b], (2.6)

(iii) Jacobi identity:

[aλ [bµ c]] = (−1)p(a)p(b) [bµ [aλ c]] + [[aλ b]λ+µ c], (2.7)

7
 Definition 2.3. A vertex algebra is a quintuple (V, |0i , ∂, [.λ .], ::) which satisfies
the following three properties:
(i) (V, ∂, [.λ .]) is a Lie conformal algebra,
(ii) (V, |0i , ∂, ::) is a unital differential algebra satisfying

• quasi-commutativity:
Z 0
p(a)p(b)
: ab : −(−1) : ba := [aλ b]dλ (2.8)
−∂

for any a, b ∈ V,
• quasi-associativity:
! !
Z ∂ Z ∂
p(a)p(b)
:: ab : c : − : a : bc ::=: dλa [bλ c] : +(−1) : dλb [aλ c] :
0 0
(2.9)

for any a, b, c ∈ V,

The integrals in the expression above should be interpreted in the following manner.
First, expand the λ-bracket. Second, put the powers of λ on the left, under the
sign of integral. Finally, take the definite integral by the usual rules inside the
parenthesis. The binary operation :: is called the normally ordered product of V .
(iii) The λ-bracket and the normally ordered product are related by the non-
commutative Wick formula:

Z λ
p(a)p(b)
[aλ : bc :] =: [aλ b]c : +(−1) : b[aλ c] : + [[aλ b]µ c]dµ, (2.10)
0

for any a, b, c ∈ V.

In this paper we are concerned with a special class of vertex superalgebras,

known as conformal vertex algebras. We say that a vertex algebra V is conformal if
there is a L ∈ V such that the Fourier modes of the corresponding field Y (L, z) =
−n−2 satisfy:
P
n∈Z Ln z

(i) the operators Ln form the Virasoro algebra

c
[Ln , Lm ] = (m − n)Lm+n + (m3 − m)δm,−n IdV (2.11)
12
where c ∈ C is called the central charge,

(ii) L−1 = T ,

(iii) L0 is a diagonalizable operator with eigenvalues bounded below.

8
Remark 2.4. Using (2.3) note that (2.11) is equivalent to:

cλ3
[Lλ L] = (∂ + 2λ)L + . (2.12)
12
A field L(z) satisfying this is called a Virasoro field. Note also that the condition
(ii) above implies that L(z) is an even field.
If V is conformal vertex algebra and a ∈ V is an eigenvector of L0 , its eigen-
value is called the conformal weight of a and is denoted by ∆(a) = ∆. Moreover
a has conformal weight ∆ if and only if

[Lλ a] = (∂ + ∆λ)a + O(λ2 ).

In the case when [Lλ a] = (∂ + ∆λ)a the vector a and the corresponding vertex
operator Y (a, z) are called primary. This is equivalent to Ln a = δn,0 ∆a for all
n ≥ 0.
Remark 2.5. An important feature of defining vertex algebras using Lie conformal
algebras is that: vertex algebras are to conformal Lie algebras what associative
algebras are to Lie algebras in the following sense. For any Lie conformal algebra
R there exists a vertex algebra U (R) with an embedding of conformal algebras π :
R ֒→ U (R) satisfying the usual universal property: for any other vertex algebra V
and a map f : R ֒→ V , there exists a morphism of vertex algebras g : U (R) ֒→ V
such that f = g ◦ π. Moreover, the algebra U (R) is constructed very similar as
in the Lie algebra situation, any vector of U (R) can be obtained by products of
elements of R.
This parallel with the Lie algebra case justifies the presentation of the exam-
ples below. When we say that a certain set A of vectors satisfying some prescribed
λ-brackets generate a vertex algebra, we first construct the corresponding Lie con-
formal algebra and then we consider its universal enveloping vertex algebra, in
particular any vector of this algebra is a combination of products of elements of
A and their derivatives. There are other situations (see subsections 2.1 and 2.2 )
when the λ-bracket of elements in A is not linear in the elements of A but can be
expressed as combinations of products of elements of A and their derivatives. In
this case we say that the vertex algebra is non-linearly generated [DSK05].
Remark 2.6. Note that the λ-bracket notation encodes the same information as the
singular part of the operator product expansion (OPE) of fields inP
a quantum field
N −1 λj
theory in dimension two. In other words when we write [Aλ B] = j=0 j! A(j) B,
in physics notation this means that:
N −1
X (A(j) B)(w)
A(z)B(w) ∼ ,
(z − w)j+1
j=0

where A(z), B(w) and (A(j) B)(w) are the fields corresponding to the vectors A,
B and A(j) B under the state-field correspondence.

9
Example 2.7 ([Kac98]). The N = 1 (Neveu-Schwarz) superconformal vertex al-
gebra.
The N = 1 superconformal vertex algebra of central charge c is an extension of
the Virasoro algebra of central charge c by an odd primary field G of conformal
weight 3/2 i.e it is generated by L and G with λ-brackets:
c 3 3 c
[Lλ L] = (∂ + 2λ)L + λ , [Lλ G] = (∂ + λ)G, [Gλ G] = 2L + λ2 .
12 2 3
Example 2.8 ([Kac98]). The N = 2 superconformal vertex algebra.
The N = 2 superconformal vertex algebra of central charge c is generated by the
Virasoro field L with λ-bracket (2.12), an even primary field J of conformal weight
1, and two odd primary fields G± of conformal weight 23 , with the λ-brackets given
by:
 
3
[Lλ J] = (∂ + λ) J, [Lλ G ] = ∂ + λ G± ,
±
2
c
[Jλ G± ] = ±G± , [Jλ J] = λ, [G± λ G± ] = 0,
3
1 c
[G+ λ G− ] = L + ∂J + Jλ + λ2 .
2 6
Example 2.9 ([KW04]). The N = 4 superconformal vertex algebra.
The N = 4 superconformal vertex algebra is generated by a Virasoro field L, three
primary even fields J 0 , J + and J − of conformal weight 1, and four primary odd
fields G± , G¯± of conformal weight 23 . The remaining non-zero λ-brackets are:
c c
[J 0 λ J ± ] = ±2J ± ,
[J 0 λ J 0 ] = λ, [J + λ J − ] = J 0 + λ,
3 6
[J 0 λ G± ] = ±G± , [J 0 λ Ḡ± ] = ±Ḡ± , [J + λ G− ] = G+ ,
[J − λ G+ ] = G− , [J + λ Ḡ− ] = −Ḡ+ , [J − λ Ḡ+ ] = −Ḡ− ,
1 c
[G± λ Ḡ± ] = (∂ + 2λ) J ± , [G± λ Ḡ∓ ] = L ± ∂J 0 ± J 0 λ + λ2 .
2 6
Example 2.10. The bc − βγ system.
This vertex algebra is generated by four fields: b and c are odd fields, β and γ are
even fields, and the non-trivial λ-brackets between the generators are:

[βλ γ] = 1, [bλ c] = 1.

If we define G =: cβ : + : (∂γ)b :, L =: (∂γ)β : − 12 : c(∂b) : + 21 : (∂c)b :,

{L, G} generate an N = 1 superconformal algebra of central charge 32 (example
2.7). With respect to the Virasoro L the conformal weight of the generators is as
follows:
1 1
∆b = , ∆c = , ∆γ = 0, ∆β = 1.
2 2

10
2.1 The Shatashvili-Vafa G2 superconformal algebra
The Shatashvili-Vafa G2 superconformal algebra [SV95] is an extension of the
N = 1 superconformal algebra {L, G} (Example 2.7) by four fields {Φ, K, X, M }
such that: Φ and M are odd, K and X are even, ∆(Φ) = 23 , ∆(K) = 2, ∆(X) = 2
and ∆(M ) = 25 . Furthermore Φ and K are primary fields. The λ-brackets are
given by:
7 15 5
[Φλ Φ] = (− )λ2 + 6X, [Φλ X] = − Φλ − ∂Φ,
2 2 2
35 3
[Xλ X] = λ − 10Xλ − 5∂X, [Gλ Φ] = K,
24
1
[Gλ X] = − Gλ + M, [Gλ K] = 3Φλ + ∂Φ,
2
 
7 3 1
[Gλ M ] = − λ + (L + 4X) λ + ∂X, [Φλ K] = −3Gλ − 3 M + ∂G ,
12 2
 
9 5
[Φλ M ] = Kλ − 3 : GΦ : − ∂K , [Xλ K] = −3Kλ + 3 (: GΦ : −∂K) ,
2 2
   
9 2 9 7 3 2
[Xλ M ] = − Gλ − 5M + ∂G λ + 4 : GX : − ∂M − ∂ G ,
4 4 2 4
21 3
[Kλ K] = − λ + 6 (X − L) λ + 3∂ (X − L) ,
6
15 11
[Kλ M ] = − Φλ2 − ∂Φλ + 3 (: GK : −2 : LΦ :) ,
2 2
  
35 4 1 2 9 3 2
[Mλ M ] = − λ + (20X − 9L) λ + 10∂X − ∂L λ + ∂ X
24 2 2 2

3
− ∂ 2 L − 4 : GM : +8 : LX : ,
2
7 3 1 5
[Lλ X] = − λ + 2Xλ + ∂X, [Lλ M ] = − Gλ2 + M λ + ∂M.
24 4 2
And the generators satisfy the following relation:

0 = 4 : GX : −2 : ΦK : −4∂M − ∂ 2 G. (2.13)

Remark 2.11. Unlike the previous examples the right hand side of some λ-brackets
is non-linear in the generators. This is an important feature of the algebra.
This superconformal algebra appeared as the chiral algebra associated to the
sigma model with target a manifold with G2 holonomy in [SV95] Its classical
counterpart had been studied by Howe and Papadopoulos in [HP93]. In fact this
algebra is a member of a two-parameter family SW ( 23 , 32 , 2) previously studied in
[Blu92] where the author found the family of all superconformal algebras which are

11
extension of the super-Virasoro algebra, i.e., the N = 1 superconformal algebra,
by two primary supercurrents of conformal weights 32 and 2 respectively. It is a
family parametrized by (c, ε) (c is the central charge and ε the coupling constant)
of non-linear W -algebras.
The Shatashvili-Vafa G2 algebra is a quotient of SW ( 23 , 32 , 2) with c = 21
2 and
ε = 0, the relation (2.13) is precisely the one imposed by the quotient.
In [SV95], this algebra was obtained as a free field realization in terms of seven
free Bosons and seven free Fermions, and the relation (2.13) was trivially satis-
fied. The first to note that we should impose this relation if we define the algebra
abstractly (it is necessary for the Jacobi identities to be checked) was Figueroa-
O’Farrill [FO97].
The Shatashvili-Vafa G2 superconformal
algebra can be obtained also as a
quantum Hamiltonian reduction of osp 4|2 [HRD15].
Remark 2.12. Note that if we define X̃ := − 15 X and Φ̃ := √i , then {X̃, Φ̃} gen-
15
7
erate an N = 1 superconformal algebra of central charge c = 10 . This N = 1 su-
perconformal algebra at this value of the central charge is known as the tri-critical
Ising model. Therefore the Shatashvili-Vafa algebra contains two N = 1 super-
conformal subalgebras, the original one generated by {L, G} and the subalgebra
generated by {X, Φ}.
Remark 2.13. Note that all the generators of the Shatashvili-Vafa G2 algebra can
be obtained as the λ-brackets of the generators in conformal weight 23 i.e., Φ and
G.
The Shatashvili-Vafa G2 superconformal algebra has an automorphism given
by L → L, G → G, Φ → −Φ, K → −K, X → X, M → M . It is interesting to
note that the fixed vectors {L, G, X, M } generate a closed subalgebra, this algebra
is a member of a one parameter family SW ( 32 , 2) of superconformal algebras, this
one corresponds to the parameter 21 2 . Another member of this family is the next
example.

2.2 The Shatashvili-Vafa Spin(7) superconformal algebra

The Shatashvili-Vafa Spin(7) superconformal algebra [SV95] is an extension of
the N = 1 superconformal algebra {L, G} (Example 2.7) by an even field X̄ of
conformal weight 2 (not primary) and an odd field M̄ of conformal weight 52 (not
primary). The λ-bracket are given by:

8 1 3
[X̄λ X̄] = λ3 + 16X̄λ + 8∂ X̄, [Lλ X̄] = λ + (∂ + 2λ) X̄,
3 3
1 2
[Gλ X̄] = Gλ + M̄ , [Gλ M̄ ] = λ3 − L − 4X̄ λ + ∂ X̄,


2 3
 
15 15 11 5
[X̄λ M̄ ] = − Gλ2 − ∂G − 8M̄ λ + ∂ M̄ − ∂ 2 G − 6 : GX̄ :,
4 4 2 4

12
    
8 4 15 2 15
[M̄λ M̄ ] = − λ − L + 16X̄ λ − ∂L + 16∂ X̄ λ
3 2 2
 
5 2 5 2
− ∂ X̄ + ∂ L + 12 : LX̄ : −6 : GM̄ : .
2 2
Similar to the example above this algebra appeared as the chiral algebra associated
to the sigma model with target a manifold with Spin(7) holonomy in [SV95] and its
classical counterpart had been studied by Howe and Papadopoulos in [HP93]. This
algebra also belongs to a family SW ( 32 , 2) of superconformal algebras with one
parameter c (the central charge), the Shatashvili-Vafa Spin(7) algebra corresponds
to c = 12, see ([FOS91], [FOS92], [FO97]).

3 The Chiral de Rham Complex

In this section we review the construction of the chiral de Rham complex [MSV99]
of a manifold M and remember some of its properties. We also recall a theorem
from [EHKZ13] that provides two different embeddings of the space of differential
forms Ω∗ (M ) into global sections of CDR in the case when M is a smooth Rie-
mannian manifold. The chiral de Rham complex can be defined in the algebraic,
analytic and smooth category, in this paper we are only concerned with the smooth
setting.
Let M be a C ∞ manifold, the CDR of M is a sheaf of vertex algebras, let us
begin with a local description. On a coordinate chart of M with coordinates {xi }
the sections of CDR are described as follows: for each coordinate xi we have an
∂
even field γ i , corresponding to each vector field ∂x i and each differential form dx
i

we have an odd fields bi and ci respectively, we also have an even field βi for each
γ i , and {γ i , βi , bi , ci } form a bc − βγ system (see example 2.10).
Now we need to take care of what happens if we change coordinates. A crucial
observation in [MSV99] is that for any change of coordinates xi → y i (xj ) there
exists an automorphism of CDR on the intersection of the coordinate charts. Us-
ing this it is possible to glue on intersections and construct a global sheaf. More
precisely if γ̃ i , β̃i , c̃i and b̃i are the fields associated to the coordinates y i , they are
expressed in terms of the fields in the coordinates xi as follows:

γ̃ i =y i (γ),
∂y i
c̃i = (γ)cj ,
∂xj
∂xj
b̃i = i (y(γ))bj , (3.1)
∂y
∂xj ∂ 2 xk ∂y l
β̃i =βj i (y(γ)) + i l (y(γ)) r cr bk .
∂y ∂y ∂y ∂x
We see that the γ i transform as coordinates do, the bi transform as vector fields, the
ci change as differential forms, however the βi change in a non-tensorial manner. In

13
the remaining of this section we collect some results about the existence of global
sections of CDR.
Remark 3.1. The first thing to note is that the multiplication (normally ordered
product) in CDR is neither associative nor commutative by the own nature of CDR:
CDR is a sheaf of vertex algebras. Even though we can define a multiplication map
OM × CDR → CDR, it is not associative. This implies that it is very difficult to
construct global sections of CDR.
From (3.1) follows that functions and vector fields of M give rise to global sections.
However, trying to construct sections of CDR from other tensors on M is not trivial
because of the terms on the RHS of the quasi-associativity (2.9) appearing under a
change of coordinates.
In [BZHS08] it was noticed that one can use the Levi-Civita connection on M to
counteract these quasi-associativity terms in order to construct sections of CDR
associated to differential two-forms. In [EHKZ13] this was generalized to higher
order forms.
We can define in a local chart the following field G = (∂γ i )bi + βi ci (see
example 2.10) and ask if this field is globally defined. The answer [MSV99] is
the following, if we change coordinates xi → y i (xj ), and let G̃ be the field in the
coordinates y i we obtain:
  
!
i
 ∂ ∂x  r 
G̃ = G + ∂  r T r log c̃  .
∂y ∂y j

Note that if M is orientable G is globally defined. In general the field G is not

globally defined however the Fourier modes G(0) is globally defined because G̃
differs from G by a derivative of a field. Moreover we have (G(0) )2 = ∂. This
endomorphism G(0) is called the supersymmetric generator, and it is an odd deriva-
tion of all the n-products i.e., G(0) (a(n) b) = (G(0) a)(n) b + (−1)p(a) a(n) (G(0) b),
where a, b are two sections of the CDR.
In [EHKZ13] were produced two different embeddings of the space of differ-
ential forms Ω∗ (M ) into global sections of CDR in the case when M is a smooth
Riemannian manifold. These embeddings depend on a choice of a metric and are
explicitly given in terms of the corresponding Levi-Civita connection. We proceed
to recall their construction:
Let g be a Riemannian metric on M and let {xi } be a local coordinate system.
Define local sections ei± by the equations:

i gij bj − ci


i gij bj + ci i
e+ := √ , ie− := √ .
2 2
Note that

[ei+ λ ej+ ] = g ij , [ei− λ ej− ] = −gij , [ei+ λ ej− ] = 0. (3.2)

14
 Define

i1 ...ik
F±(0) := 1, F+(k) := ei+1 . . . ei+k , i1 ...ik
F−(k) := ik ei−1 . . . ei−k .
(Here and further we adopt the following convention: when a product of fields
appears without any colon or parenthesis, we read the normal product from right
to left, recalling that it is not associative.)
Define G±(n,n) = F±(n) and for each 1 ≤ s ≤ ⌊ n2 ⌋ we define

i i ...i
Gi±(n,n−2s)
1 ...in
= Γik11 l1 gi2 k1 ∂γ l1 . . . Γk2s−1
2s−1 l2s−1
gi2s k2s−1 ∂γ l2s−1 F±(n−2s)
2s+1 n
,

where Γijk are the Christoffel symbols of the Levi-Civita connection associated to
g, ⌊·⌋ denotes the integer part, the subscript between parenthesis denotes how many
e’s are present in the expression.
Define the numbers Tr,s as the coefficients of the Bessel polynomials:
r r
X
s
X (r + s)!
yr (x) = Tr,s x = xs ,
(r − s)!s!2s
s=0 s=0

and let Tr,s := 0 when s < 0 or s > t. The following theorem was proved in
[EHKZ13].

Theorem 3.2. [EHKZ13, Theorem 6.1] Let (M, g) be a Riemannian manifold. For
any differential form w ∈ Ω∗ (M ) locally described by wi1 . . . in dxi1 ∧ · · · ∧ dxin
define
1 i1 ...in
J±q := wi1 ...in E±(n) ,
n!
n
[2]
X
E±(n) := Tn−s,s G±(n,n−2s) .
s=0

Then J±q are well defined sections of CDR.

Corollary 3.3. G(0) J±q are also well defined sections of CDR.

This follows from the discussion above about the supersymmetric generator
G(0) . In fact this corollary was implicit in the original form of the Theorem 3.2
in [EHKZ13] where the authors consider the CDR as a supersymmetric sheaf of
vertex superalgebras.
As an exemplification of the Theorem 3.2 and because they will be used in
section 5 we write the sections of CDR that correspond to 2, 3 and 4 forms:
if w is a 2-form then
1 1
J±q = ± wij ei± ej± + Γijk gjl wil ∂γ k , (3.3)
2 2

15
if w is a 3-form then
1 1
J+q = wijk ei+ ej+ ek+ + wijk Γimn g jm ∂γ n ek+ , (3.4)
6 2

(−i) i
J−q = wijk ei− ej− ek− + wijk Γimn gjm ∂γ n ek− ,
6 2
if w is a 4-form then
1 1
J+q = wijkl ei+ ej+ ek+ el+ + wijkl Γimn gjm ∂γ n ek+ el+
24 4
1
+ wijkl Γim1 n1 gjm1 ∂γ n1 Γkm2 n2 glm2 ∂γ n2 , (3.5)
8

1 (−1)
J−q = wijkl ei− ej− ek− el− + wijkl Γimn g jm ∂γ n ek− el−
24 4
1
+ wijkl Γim1 n1 g jm1 ∂γ n1 Γkm2 n2 glm2 ∂γ n2 .
8
Theorem 3.4. [EHKZ13, section 7.1 and 7.2] Let M be a Calabi-Yau manifold,
ω the Kähler form, let J±q be the global sections given by Theorem 3.2 cor-
responding to ω, then the two sets of sections of CDR {J±q , G(0) (J±q )} gen-
erate two commuting copies of the N = 2 superconformal algebra of central
charge 23 dim M . Furthermore if we assume that M is a Calabi-Yau three-
fold, let Ω and Ω̄ denote the holomorphic volume form and its complex conju-
gate respectively. Let X± and X̄± be the global sections given by Theorem 3.2
corresponding to Ω and Ω̄ respectively, then the two sets of sections of CDR
{J±q , G(0) (J±q ), X± , G(0) (X± ), X̄± , G(0) (X̄± )} generate two commuting copies
of the Odake algebra [Oda89].

4 G2 holonomy manifolds
In this section we review some properties of the manifolds with G2 holonomy that
will be used extensively in the computations. The references for this section are
[Joy07, Sal89].
The Lie group G2 can be defined as the group of linear automorphism of R7
that preserves the cross product, where we identify R7 = Im(O), and the cross
product operation is induced from the octonion multiplication u × v = im(v̄.u),
u, v ∈ O. The cross product operation is encoded by a 3-form ϕ0 ∈ ∧3 (R7 )∗
defined as

ϕ0 (u, v, w) =< u × v, w > .

If we choose coordinates (x1 , . . . , x7 ) on R7 and denote by dxij...k the exterior

form dxi ∧ dxj ∧ · · · ∧ dxk on R7 , ϕ0 can be written as:

16
 ϕ0 = dx123 + dx145 + dx167 + dx246 − dx257 − dx347 − dx356 .

In the other way, the identity

6 < u, v > dvol7 = (uyϕ0 ) ∧ (vyϕ0 ) ∧ ϕ0 , (4.1)

expresses the inner product of R7 in terms of the volume form of R7 and the 3-form
ϕ0 . Observe that the Lie algebra g2 of the Lie group G2 can be described as:

g2 = {(aij ) ∈ so(7)|aij ϕ0 ijk = 0 ∀ k}. (4.2)

Definition 4.1. Let M be a smooth 7-dimensional manifold, and F the frame bun-
dle of M . We say that M has a G2 -structure if there is a principal subbundle P of
F with fibre G2 .

Equivalently M has a G2 -structure if there exists ϕ ∈ Ω3 (M ) that is pointwise

modelled in ϕ0 , i.e., for each x ∈ M , ϕ0 is the image of ϕ(x) under a linear
oriented isomorphism Tx M → R7 . Thus, we have a 1-1 correspondence between
3-forms ϕ pointwise modelled in ϕ0 and G2 -structures P on M .
Given a G2 -structure in M this determines in particular an orientation and
a metric g in M because any G2 -structure is an instance of an SO(7)-structure.
Furthermore, let ϕ be the 3-form that corresponds to this G2 -structure. Then, the
identity (4.1) determines the metric in terms of ϕ up to a conformal factor. Let
Hol(g) denote the holonomy group of M , then Hol(g) ⊆ G2 if and only if ϕ is
parallel with respect to the Levi-Civita connection of g, i.e., ∇ϕ = 0.

Lemma 4.2. [Sal89] The holonomy group of the Riemannian metric g induced by
ϕ is contained in G2 if and only if dϕ = 0 and d ∗ ϕ = 0, where ∗ denotes the
Hodge star for g.

We say that a manifold with a G2 -structure is a G2 -manifold if Hol(g) ⊆ G2 .

If we write ϕ and ∗ϕ in local coordinates: ϕ = ϕijk dxijk , ∗ϕ = ψijkl dxijkl
then dϕ = 0 and d ∗ ϕ = 0 are equivalent to:

∂
0= ϕijk − Γsmi ϕsjk − Γsmj ϕisk − Γsmk ϕijs , (4.3)
∂xm
∂
0= ψijkl − Γsmi ψsjkl − Γsmj ψiskl − Γsmk ψijsl − Γsml ψijks , (4.4)
∂xm
for all m = 1, . . . , 7.

Lemma 4.3. [Sal89] Let (M, g) be a Riemannian 7-manifold. If Hol(g) ⊆ G2 ,

then g is Ricci-flat.

17
 Let M be a 7-dimensional manifold with G2 –structure given by the 3-form ϕ,
denote by g and ψ = ∗ϕ the associated metric and 4-form respectively. Then the
following identities are satisfied, they are collected from [Kar09]:
Contractions of ϕ and ψ

ϕijk ϕabc gia g jb gkc =42, (4.5)

jb kc
ϕijk ϕabc g g =6gia , (4.6)
ϕijk ϕabc gkc =gia gjb − gib gja − ψijab , (4.7)

ϕijk ψabcd gib gjc gkd =0, (4.8)

jc kd
ϕijk ψabcd g g = − 4ϕiab , (4.9)
ϕijk ψabcd gkd =gia ϕjbc + gib ϕajc + gic ϕabj
− gaj ϕibc − gbj ϕaic − gcj ϕabi , (4.10)

ψijkl ψabcd gld = − ϕajk ϕibc − ϕiak ϕjbc − ϕija ϕkbc

+ gia gjb gkc + gib gjc gka + gic gja gkb − gia gjc gkb
− gib gja gkc − gic gjb gka − gia ψjkbc − gja ψkibc
− gka ψijbc + gab ψijbc − gac ψijkb . (4.11)

Another useful identity is:

 
ϕ ∧ (ω ϕ) =(−2) ψ ∧ w♭ , or in coordinates
ϕ[ijk ϕmn]l =(−2)gl[m ϕnijk] (4.12)

where w is a vector field in M and w♭ be the 1–form dual to w; [ ] denotes the

anti–symmetrization of the indices.
Lemma 4.4. (Lemma 4.9 in [Kar09]) Let M be a manifold with a G2 -structure and
let Rijkl be the Riemann curvature tensor of M . Then we have Rijkl ψ ijkm = 0.
The next lemma follows easily from the fact that if Hol(g) ⊆ G2 then Rijkl ∈
Sym2 (g2 ), and the above definition (4.2) of g2
Lemma 4.5. Let M be a G2 -manifold and let Rijkl be the Riemann curvature
tensor of M . Then we have Rijkl ϕijm = 0.

5 The Chiral de Rham Complex over a G2 -manifold

In this section we prove the main results of this paper. We begin remembering the
conjecture that is the leitmotiv of this paper.

18
 Let (M, g) be a G2 -manifold, let ϕ be the corresponding defining 3-form. Let
Φ± be the global sections of CDR that correspond to ϕ by Theorem 3.2, and let
K± := G(0) (Φ± ). In [EHKZ13] the following conjecture was stated:

Conjecture 5.1. [EHKZ13, Conjecture 7.3] The sections pairs {Φ+ , K+ } and
{Φ− , K− } generate two commuting copies of the Shatashvili-Vafa G2 supercon-
formal algebra given in subsection 2.1 (see Remark 2.13).

5.1 G2 holonomy manifolds are superconformal

Before stating our main result we define some sections:
1  1   1 
X± := Φ± (0) Φ± , G± := (− ) Φ± (1) K± , L± := G± (0) G± .
6 3 2
Note that these are well defined global sections because they are defined in terms
of Φ± and K± .

Theorem 5.2. The pairs of sections {G± , L± } and {Φ± , X± } generate two N = 1
superconformal algebras of central charge 21 7
2 and 10 respectively. Furthermore the
plus signed sections commute with the minus signed sections.

Proof. The proof is based on explicit computations and some abstract manipula-
tions. To make the proof clear, every time that an explicit computation should be
made we will indicate the appropriate subsection of the section 6 where the com-
putation is performed. Note that to compute a λ-bracket we can choose a local
coordinate chart and perform the computation in this chart because we are working
with well defined global sections.
We begin proving that {Φ± , X± } generate two N = 1 superconformal algebras of
the desired central charges respectively. We work the plus case (the minus case is
similar).
We have by (3.4)

Φ+ = 16 ϕijk ei+ ej+ ek+ + 21 ϕijk Γimn gjm ∂γ n ek+ . (5.1)

In 6.1 it is shown that

[Φ+ λ Φ+ ] = (− 72 )λ2 + 6X+ , (5.2)

where X+ is given by:

1
X+ = − 24 ψijkl ei+ ej+ ek+ el+ − 14 ψijkl Γimn gjm ∂γ n ek+ el+
− 81 ψijkl Γim1 n1 gjm1 ∂γ n1 Γkm2 n2 g lm2 ∂γ n2 − 12 gij ∂(ei+ )ej+
− 12 gij Γjkl ∂γ k el+ ei+ − 14 Γijk Γkil ∂γ j ∂γ l . (5.3)

In 6.2 we see that:

19
 X+ (2) Φ+ = 0, X+ (1) Φ+ = − 15
2 Φ+ . (5.4)

Then, using skewsymmetry (2.6), we get Φ+ (2) X+ = 0, Φ+ (1) X+ = − 15

2 Φ+ . To
compute Φ+ (0) X+ we observe that:

[Φ+ λ [Φ+ µ Φ+ ]] = 6[Φ+ λ X+ ] (5.5)

by (5.2). Using the Jacobi identity (2.7) on the left side of (5.5) we have:

[Φ+ λ [Φ+ µ Φ+ ]] = (−1)[Φ+ µ [Φ+ λ Φ+ ]] + [[Φ+ λ Φ+ ]λ+µ Φ+ ]

= (−1)[Φ+ µ 6X+ ] + [6X+ λ+µ Φ+ ]
 
= (−6) Φ+ (0) X+ + (− 152 )Φ + µ + (−1)6[Φ+ −λ−µ−∂ X+ ]
 
= (−6) Φ+ (0) X+ + (− 152 )Φ+ µ (5.6)
 
+(−6) Φ+ (0) X+ + 15 2 (λ + µ + ∂)Φ + .

Equating (5.5) and (5.6) we obtain a polynomial equality in λ and µ. Taking

the coefficient of λ0 µ we get Φ+ (0) X+ = − 52 ∂Φ+ . Then

[Φ+ λ X+ ] = − 15 5
2 Φ+ λ − 2 ∂Φ+ . (5.7)
To compute [X+ λ X+ ] observe that, using (5.2) and the Jacobi identity, we
have:

6[X+ λ X+ ] = [X+ λ [Φ+ µ Φ+ ]] = [Φ+ µ [X+ λ Φ+ ]] + [[X+ λ Φ+ ]λ+µ Φ+ ].

Then

10 15
6[X+ λ X+ ] = [Φ+ µ − 2 ∂Φ+ − 2 Φ+ λ] + [[X+ λ Φ+ ]λ+µ Φ+ ]. (5.8)

This is a polynomial identity in λ and µ. Putting µ = 0 we get:

6[X+ λ X+ ] = −30∂X+ − 45X+ λ + [[X+ λ Φ+ ]λ Φ+ ]

= −30∂X+ − 45X+ λ + [ 10
2 ∂Φ+ −
15
2 Φ+ λ λ Φ+ ]
= −30∂X+ − 45X+ λ + [− 2 ∂Φ+ λ Φ+ ] + [− 15
10
2 Φ+ λ Φ+ ]λ,

Using sesquilinearity (2.5) it follows that:

35 3
[X+ λ X+ ] = 24 λ − 10X+ λ − 5∂X+ . (5.9)

20
We have proved that {Φ± , X± } satisfy the λ-brackets (5.2), (5.7) and (5.9) and
therefore, by Remark 2.12 we conclude that {Φ+ , X+ } and {Φ− , X− } generate
7
two N = 1 superconformal algebras of central charge 10 respectively. Further-
i i
more, as the e+ ’s commute with the e− ’s (3.2), we have that Φ+ commutes with
Φ− and, consequently, the algebra generated by {Φ+ , X+ } commutes with the
algebra generated by {Φ− , X− }.
Now we are going to prove that {G± , L± } generate two N = 1 superconformal
1  
algebras of central charge 21
2 . We compute explicitly G ± := (− ) Φ ± (1) K±
3
(see 6.3) and find that:

1 i i 1 ij 1 ij l k
G+ = 1
2 c βi + 2 ∂γ bi + 2 g bi βj + 2 g Γik c bj bl + 12 g ij Γkij ∂bk
+ 21 gij ∂γ i cj + gij Γkil Γljm∂γ m bk ,

1 i i 1 ij 1 ij l k
G− = 1
2 c βi + 2 ∂γ bi + (− 2 )g bi βj + (− 2 )g Γik c bj bl + (− 12 )gij Γkij ∂bk
+(− 12 )gij ∂γ i cj + (−1)g ij Γkil Γljm ∂γ m bk .

Note that if we define G := G+ + G− , we have G = ci βi + ∂γ i bi , and

1
{G, L := G(0) G} generate an N = 1 superconformal algebra of central charge
2
c = 21, and with respect to L the conformal weight of the generators is as follows
(see example 2.10 and the definition of G(0) in section 3):

1 1
∆bi = , ∆ci = , ∆γ i = 0, ∆βi = 1, i ∈ {1, . . . , 7}. (5.10)
2 2
This implies that the conformal weights of the local sections in the coordinate chart
are always positive. We will use this observation below.
1
In 6.4, we prove that [G+ λ G− ] = 0. By definition, L+ = G+ (0) G+ and
2
1
L− = G− (0) G− . Then we have that {G+ , L+ } commute with {G− , L− }, and
2
that L = L+ + L− . We will show that {G+ , L+ } is an N = 1 superconformal
algebra (the same proof works for {G− , L− }). First we take care of [L+ λ L+ ]. As
∆L+ = 2, the observation above about conformal weights implies:

[L+ λ L+ ] = [Lλ L+ ] = ∂L+ + 2L+ λ + Aλ2 + Bλ3 ,

where A and B are two fields with ∆A = 1 and ∆B = 0.

Consequently:
[Lλ A] = ∂A + Aλ + Cλ2 , [Lλ B] = ∂B,
where C is a field with ∆C = 0, i.e., [Lλ C] = ∂C. Using the Jacobi identity (2.7),
we obtain

[L+ λ [Lµ L]] = [L+ µ [L+ λ L+ ]] + [[L+ λ L+ ]λ+µ L+ ]. (5.11)

21
Also:

[L+ λ [Lµ L]] = [L+ λ ∂L+ + 2L+ µ]. (5.12)

Working the right side of (5.12) :

 
[L+ λ ∂L+ + 2L+ µ] = (∂ + λ) ∂L+ + 2L+ λ + Aλ2 + Bλ3 (5.13)
 
+ 2µ ∂L+ + 2L+ λ + Aλ2 + Bλ3 .

Working separately the summands on the right side of (5.11) and putting µ = 0:

[L+ µ [L+ λ L+ ]]|µ=0 = [Lµ ∂L+ + 2L+ λ + Aλ2 + Bλ3 ]|µ=0

= ∂ 2 L+ + 2∂L+ λ + ∂Aλ2 + ∂Bλ3 ,

[[L+ λ L+ ]λ L+ ] = [∂L+ + 2L+ λ + Aλ2 + Bλ3 λ L]

 
= (−λ) ∂L+ + 2L+ λ + Aλ2 + Bλ3
 
+2λ ∂L+ + 2L+ λ + Aλ2 + Bλ3 + Aλ3
+(−C)λ4 + (−2)∂Cλ3 + (−1)∂ 2 Cλ2 + (−1)∂Bλ3 .

Equating the right sides of (5.11) and (5.12), putting µ = 0, and looking at the
coefficient of λ4 and λ3 we get that C = 0 and ∂B = A respectively.
Working separately the summands on the right side of (5.11) and putting λ = 0:
 
[L+ µ [L+ λ L+ ]]|λ=0 = [L+ µ ∂L+ ] = (∂ + µ) ∂L+ + 2L+ µ + Aµ2 + Bµ3 ,
 
[[L+ λ L+ ]λ+µ L+ ]|λ=0 = [∂L+ µ L+ ] = (−1)µ ∂L+ + 2L+ µ + Aµ2 + Bµ3 .
Equating the right sides of (5.11) and (5.12), putting λ = 0, and looking at the
coefficient of µ3 we get that ∂B = 0. But this implies that A = 0. Also, as
∆B = 0, we know that B is a function, therefore ∂B = 0 implies that B is a
constant.
Hence [L+ λ L+ ] = (∂ + 2λ) L+ + c12+ λ3 , where c+ is a constant.
Now we compute [L+ λ G+ ]. Proceeding as above we get that ∆G+ = 23 and,
hence:
3
[L+ λ G+ ] = [Lλ G+ ] = ∂G+ + G+ λ + Cλ2 ,
2
1
where C is a field with ∆C = , and
2
1
[Lλ C] = ∂C + Cλ.
2
Using the Jacobi identity (2.7), we obtain

[G+ λ [L+ µ L+ ]] = [Lµ [G+ λ L+ ]] + [[G+ λ L+ ]λ+µ L]. (5.14)

22
We also have:
[G+ λ [L+ µ L+ ]] = [G+ λ (∂ + 2µ)L+ ]. (5.15)
Working the right side of (5.15):
 
3
[G+ λ (∂ + 2µ)L+ ] = (−1)(λ + ∂) ∂G+ + (−λ − ∂)G+ + (λ + ∂)2 C
2
 
3
+2µ(−1) ∂G+ + (−λ − ∂)G+ + (λ + ∂)2 C .
2

Working separately the summands on the right side of (5.14) and putting µ = 0:
1 3
[Lµ [G+ λ L+ ]]|µ=0 = [Lµ ∂G+ + G+ λ + (−C)λ2 ]|µ=0
2 2
1 2 3
= (∂ G+ ) + ∂G+ λ + (−1)∂Cλ2 ,
2 2

1 3
[[G+ λ L+ ]λ+µ L]|µ=0 = [ ∂G+ + G+ λ + (−C)λ2 λ L]
2  2 
1 1 3 2
= (− )λ ∂G+ + G+ λ + (−C)λ
2 2 2
 
3 1 3 1
+ λ ∂G+ + G+ λ + (−C)λ + (− )Cλ3 .
2
2 2 2 2

Equating the right sides of (5.14) and (5.15) putting µ = 0, and looking at the
3
coefficient of λ3 we get that C = 0 which implies [L+ λ G+ ] = ∂G+ + G+ λ.
2
Now we compute [G+ λ G+ ]. By definition of L+ and conformal weights positivity
we have:
[G+ λ G+ ] = [Gλ G+ ] = 2L+ + Dλ + Eλ2 ,
with D and E two fields of conformal weights 1 and 0 respectively.
Using the Jacobi identity (2.7) we obtain

[L+ λ [Gµ G]] = [G+ µ [L+ λ G+ ]] + [[L+ λ G+ ]λ+µ G+ ]. (5.16)

We also have:
c+ 3
[L+ λ [Gµ G]] = [L+ λ 2L] = [L+ λ 2L+ ] = 2(∂ + 2λ)L+ + 2 λ . (5.17)
12
Working separately the summands on the right side of (5.16):
3
[G+ µ [L+ λ G+ ]] =[G+ µ ∂G+ + G+ λ]
2
  3  
=(∂ + µ) 2L+ + Dµ + Eµ2 + λ 2L+ + Dµ + Eµ2 ,
2

23
 3
[[L+ λ G+ ]λ+µ G+ ] =[∂G+ + G+ λ G+ ]
2 λ+µ
 
=(−1)(λ + µ) 2L+ + D(λ + µ) + E(λ + µ)2
3  
+ λ 2L+ + D(λ + µ) + E(λ + µ)2 .
2
Equating both the right sides of (5.16) and (5.17) and looking at the coefficient of
λ2 with µ = 0 we get that D = 0 while looking at the coefficient of λ3 with µ = 0
c+ c+ 2
we get that E = . Therefore [G+ λ G+ ] = 2L+ + λ .
3 3
We have proved that {G+ , L+ } and {G− , L− } are two commuting N = 1 super-
conformal algebras of central charge c+ and c− respectively, such that c+ + c− =
21.
We can calculate the central charge by computing explicitly the lambda bracket
[G+ λ G+ ] = [Gλ G+ ] and looking at the coefficient of λ2 .
Computing the coefficient of λ2 :
!
1 ∂ 2   7 1   1  lm n  1 7
G+ λ G+ = + (− ) g,llm + g Γlm + g lm Γnla Γamn +
2 ∂λ 4 4 ,m 4 ,n 2 4
7 1 1  lm n 
= + g lm Γnla Γamn + g Γlm
2 2 2 ,n
 
7 1 1
= + − g lm Γnla Γamn + g lm (Γnlm ),n
2 2 2
7 1
= + g lm Rlm
2 2
7
= .
2
The last equality follows because G2 -manifolds are Ricci flat, Lemma 4.3. It is
interesting to note that, as in other places, the scalar curvature of the manifold, i.e.,
g ij Rij , appears here explicitly. Then, the central charge of the N = 1 superconfor-
mal algebra {G+ , L+ } is c+ = 21 2 and consequently as c+ + c− = 21, the central
charge of {G− , L− } is c− = 21 2 .
To conclude the proof of the theorem we only need to check that G+ (resp. G− )
commutes with Φ− (resp. Φ+ ). This is accomplished by performing an explicit
computation in 6.5.

Remark 5.3. Note that X+ (5.3) is not exactly the section produced by Theorem
3.2 using the 4-form, because, besides the correction (3.5), we need to add other
terms. That this would happen was already observed in [EHKZ13] while working
the flat case. For this reason, Conjecture 5.1 was formulated using only the 3-form.
The expression (5.3) that we have obtained here works in any coordinate system.
Remark 5.4. It should be noticed that the complexity of the explicit computations
performed is greater than in the other holonomy cases. This is due to the lack of
special coordinate systems simplifying the corrections to the fields. This in turn
is a reflection of our lack of knowledge about the geometry of manifolds with G2
holonomy.

24
5.2 Proof of the Conjecture
In this subsection, we prove Conjecture 5.1.

Theorem 5.5. The sections pairs {Φ+ , K+ } and {Φ− , K− } generate two commut-
ing copies of the Shatashvili-Vafa G2 superconformal algebra.

In the above subsection 5.1, besides the global sections Φ± and K± = G(0) Φ± ,
we introduced the global sections:
1  1   1 
X± := Φ± (0) Φ± , G± := (− ) Φ± (1) K± , L± := G± (0) G± .
6 3 2
Now we introduce M± := G(0) X± = G± (0) X± . Note that, again, these sec-
tions are globally well defined because they are defined in terms of X± using the
supersymmetric generator G(0) .
Remark 5.6. The reader should note that Theorem 5.5 not only proves that we have
two pair of commuting copies of the N = 1 superconformal algebra. Namely it
is implicit in the proof that {Φ± , X± } satisfy the commutation rules of the same
named fields of the Shatashvili-Vafa G2 superconformal algebra (subsection 2.1),
i.e.,:
7 15 5
[Φ± λ Φ± ] = (− )λ2 + 6X± , [Φ± λ X± ] = − Φ± λ − ∂Φ± ,
2 2 2
35 3
[X± λ X± ] = λ − 10X± λ − 5∂X± .
24
It follows from the way we define the global sections (see also Remark 2.13) that
{L+ , X+ , K+ , M+ } (resp. {L− , X− , K− , M− }) can be expressed in terms of Φ+
and G+ (resp. Φ− and G− ). Remember also that in Theorem 5.5 we proved that
{Φ+ , G+ } commute with {Φ− , G− }. Therefore {L+ , G+ , Φ+ , X+ , K+ , M+ }
commute with {L− , G− Φ− , X− , K− , M− }.
By the remark above, to prove the conjecture we only need to check the
λ–brackets between the fields {L± , G± , Φ± , X± , K± , M± } satisfy the same λ-
brackets of the Shatashvili-Vafa G2 algebra (ruling out the ones have already been
checked in the way of proving Theorem 5.5). We work the plus case, the minus
case is similar.
We prove first the linear λ-brackets (subsection 5.2.1); and then the non-linear λ-
brackets (5.2.2) under the assumption that we have proved the relation (2.13):

0 = 4 : G± X± : −2 : Φ± K± : −4∂M± − ∂ 2 G± . (5.18)

Finally (subsection 5.2.3) we prove the above relation.

Remark 5.7. We need the assumption (5.18) only to prove the non-linear λ-
brackets, the linear ones follow without the assumption. In fact, the relation (5.18)
is used only to prove the λ-bracket [Φ+ λ M+ ], the others non-linear λ-brackets are
deduced from this one. Similarly we could have computed first another non-linear

25
λ-bracket using the relation, and then deduce the others ones, that is, there is noth-
ing special in [Φ+ λ M+ ]. Even more, if we are able to check a non-linear λ-bracket
without the assumption, then we can prove the relation (5.18) using the Jacobi iden-
tity (2.7), remenber that the space of global sections of the CDR is a vertex algebra.
The moral here is that we need to verify at least one non-linear identity among the
fields. We opted here to check the relation (5.18).
To simplify the notation we denote the sypersymmetric generator G(0) by D.
Remember that D is an odd derivation of all the n-products; and that [D, ∂] = 0
because D 2 = ∂.

5.2.1 Linear λ-brackets

1) [Φ+ λ K+ ]
By definition of G+ we have that Φ+ (1) K+ = (−3)G+ . We are left with
Φ+ (2) K+ and Φ+ (0) K+ , because Φ+ (n) K+ = 0 for n ≥ 3 by the positivity
of the conformal weight.
Computing Φ+ (2) K+ :
By skewsymmetry (2.6), we have [Φ+ λ K+ ] = (−1)[K+ −∂−λ Φ+ ]. Then,
2
Φ+ (2) K+ λ2 + Φ+ (1) K+ λ + Φ+ (0) K+
2
= − K+ (2) Φ+ (−∂−λ)
2 − K+ (1) Φ+ (−∂ − λ) − K+ (0) Φ+ .
Expanding the right side of the last equality and looking at the coefficient of
λ2 we get that Φ+ (2) K+ = −K+ (2) Φ+ . On the other hand, as D is an odd
derivation of the n-products, we have
0 = D(Φ+ (2) Φ+ ) = DΦ+ (2) Φ+ + (−1)Φ+ (2) DΦ+ = K+ (2) Φ+ − Φ+ (2) K+ .
This implies Φ+ (2) K+ = 0.
Computing Φ+ (0) K+ :
Applying Borcherds identity (2.2) with a = Φ+ , b = K+ , c = |0i, n =
−1, m = 2, k = −2, we obtain
   
2  2 
Φ+ (0) K+ |0i + Φ+ (1) K+ |0i
1 (−1) 2 (−2)
   
−1   −1  
= Φ+ (1) K+ (−2) |0i + (−1) Φ+ (0) K+ (−1) |0i .
0 1
Then
Φ+ (0) K+ = 3∂G+ + Φ+ (1) ∂K+ . (5.19)
As we have already computed [Φ+ λ Φ+ ], using sesquilinearity (2.5), we have
that Φ+ (1) ∂Φ+ = 6X+ . Using that D is an odd derivation we get 6M+ =
D(Φ+ (1) ∂Φ+ ) = K+ (1) ∂Φ+ − Φ+ (1) ∂K+ . Then,

Φ+ (1) ∂K+ = K+ (1) ∂Φ+ − 6M+ . (5.20)

26
  
We also have ∂ Φ+ (1) K+ = ∂Φ+ (1) K+ + Φ+ (1) ∂K+ . Which implies

(−3)∂G = ∂Φ+ (1) K+ + Φ+ (1) ∂K+ . (5.21)

As Φ+ (n) K+ = 0 for n ≥ 2, by skewsymmetry (2.6) we have

[∂Φ+ λ K+ ] = (−1)[K+ −λ−∂ ∂Φ+ ]

 
2
= (−1) K+ (2) ∂Φ+ (−λ − ∂) + K+ (1) ∂Φ+ (−λ − ∂) + K+ (0) ∂Φ+ .

Then
 
∂Φ+ (1) K+ = K+ (1) ∂Φ+ − ∂ K+ (2) ∂Φ+ . (5.22)

Applying Borcherds identity (2.2) with a = Φ+ , b = K+ , c = |0i, n =

1, m = −2, k = 1 we obtain
 
−2     
Φ+ (1) K+ |0i =K+ (2) Φ+ (−2) |0i + (−1)K+ (1) Φ+ (−1) |0i
0 (−1)

Φ+ (1) K+ =K+ (2) ∂Φ+ + (−1)K+ (1) Φ+ .

Then using skewsymmetry (2.6) we get:

K+ (2) ∂Φ+ = 2Φ+ (1) K+ = (−6)G+ . (5.23)

Finally substituting the equations (5.23), (5.22), (5.21), (5.20) and (5.19), we
get: Φ+ (0) K+ = (−3)M+ + (− 32 )∂G+ . Then we have proved that
 
1
[Φ+ λ K+ ] = −3G+ λ − 3 M+ + ∂G+ .
2

2) [K+ λ K+ ]  
As D is an odd derivation of the n-products we have D Φ+ (n) K+ =
K+ (n) K+ − Φ+ (n) ∂Φ+ for all n ∈ Z. Then
 
K+ (n) K+ = D Φ+ (n) K+ + Φ+ (n) ∂Φ+ . (5.24)

We already know [Φ+ λ K+ ] and [Φ+ λ Φ+ ], specialising equation (5.24) for

n = 3, 2, 1, 0 we conclude that:
21 3
[K+ λ K+ ] = − λ + 6 (X+ − L+ ) λ + 3∂ (X+ − L+ ) .
6

3) [G+ λ Φ+ ]
By positivity of the conformal weight and the definition of K+ :=
G(0) Φ+ = G+ (0) Φ+ we have that

λ2
[G+ λ Φ+ ] = A + Bλ + K+ ,
2

27
 where A and B are two fields. To find A and B we perform an explicit
computation 6.6, and found that A = B = 0. Then we conclude:

[G+ λ Φ+ ] = K+ .

4) [L+ λ Φ+ ]
3
As ∆Φ+ = 2 we have

3
[L+ λ Φ+ ] = [Lλ Φ+ ] = Aλ2 + (∂ + λ)Φ+ ,
2
where A is a field. To find A we perform and explicit computation 6.7 and
found that A=0. Then we conclude:
3
[L+ λ Φ+ ] = (∂ + λ)Φ+ .
2

6) [L+ λ X+ ]
Applying the Jacobi identity (2.7) we have
1 
[L+ λ X+ ] = [L+ λ 61 [Φ+ µ Φ+ ]] = [Φ+ µ [L+ λ Φ+ ]] + [[L+ λ Φ+ ]λ+µ Φ+ ] .
6
From where
7 3
[L+ λ X+ ] = − λ + 2X+ λ + ∂X+ .
24

7) [L+ λ K+ ]
Applying the Jacobi identity (2.7) we have

[L+ λ K+ ] = [L+ λ [G+ µ Φ+ ]] = [G+ µ [L+ λ Φ+ ]] + [[L+ λ G+ ]λ+µ Φ+ ].

We conclude that:

[L+ λ K+ ] = (∂ + 2λ)K+ .

8) [G+ λ X+ ]
Computing in much the same way as [L+ λ X+ ], we obtain:

1
[G+ λ X+ ] = − G+ λ + M+ .
2

9) [G+ λ K+ ]
In the same way as [L+ λ K+ ], we obtain:

[G+ λ K+ ] = 3Φ+ λ + ∂Φ+ .

28
 10) [L+ λ M+ ]

[L+ λ [G+ µ X+ ]] =[L+ λ (− 12 )G+ µ + M+ ] (5.25)

=(− 21 )[L+ λ G+ ]µ + [L+ λ M+ ]. (5.26)

On the other hand using the Jacobi identity (2.7) we have:

[L+ λ [G+ µ X+ ]] = [G+ µ [L+ λ X+ ]] + [[L+ λ G+ ]λ+µ X+ ]. (5.27)

Equating the right sides of the equations (5.25) and (5.27), expanding using
the already known λ-brackets and then looking at the coefficients of the λ’s
we get:
1 5
[L+ λ M+ ] = − G+ λ2 + M+ λ + ∂M+ .
4 2
11) [G+ λ M+ ]
Computing in much the same way as [L+ λ M+ ], we obtain:
7 3
[G+ λ M+ ] = − λ + (L+ + 4X+ ) λ + ∂X+ ,
12

5.2.2 Non linear λ-brackets

1) [Φ+ λ M+ ]
Computing Φ+ (3) M+ :
 
Φ+ (3) M+ = Φ+ (3) G+ (0) X+ .

Using the Borcherds identity (2.2) with m = 3, k = 0, n = 0, a = G+ ,

b = X+ , and c = Φ+ , we obtain:
   
3  3 
G+ (0) X+ Φ+ + G+ (1) X+ Φ+
0 (3) 1 (2)
   
= G+ (3) X+ (0) Φ+ + (−1)X+ (0) G+ (3) Φ+ .

implying that M+ (3) Φ+ = 0 and, by skewsymmetry (2.6), that Φ+ (3) M+ =

0.
Computing Φ+ (2) M+ :
 
Φ+ (2) M+ = Φ+ (2) G+ (0) X+ .

Using the Borcherds identity (2.2) with m = 2, k = 0, n = 0, a = G+ ,

b = X+ , and c = Φ+ , we obtain:
   
2  2 
G+ (0) X+ Φ+ + G+ (1) X+ Φ+
0 (2) 1 (1)
   
= G+ (2) X+ (0) Φ+ + (−1)X+ (0) G+ (2) Φ+ .

29
Then M+ (2) Φ+ = 0 and by skewsymmetry (2.6) Φ+ (2) M+ = 0.
Computing Φ+ (1) M+ :
 
Φ+ (1) M+ = Φ+ (1) G+ (0) X+ .

Using the Borcherds identity (2.2) with m = 1, k = 0, n = 0, a = G+ ,

b = X+ , and c = Φ+ , we obtain:
   
G+ (0) X+ Φ+ + (− 12 )K+ = G+ (1) (− 10
2 )∂Φ + .
(1)

Then M+ (1) Φ+ = − 29 K+ and by skewsymmetry (2.6) Φ+ (1) M+ = 29 K+ .

Computing Φ+ (0) M+ :
By the assumption (5.18) we have
0 =4 : G+ X+ : −2 : Φ+ K+ : −4∂M+ − ∂ 2 G+ .
Using quasi-commutativity (2.8) we have
Z 0
: G+ X+ : − : X+ G+ := [G+ λ X+ ]dλ = 41 ∂ 2 G+ + ∂M+ ,
−∂

and
Z 0
: Φ+ K+ : − : K+ Φ+ := [Φ+ λ K+ ]dλ = (−3)∂M+ .
−∂

Thus,
0 =4 : X+ G+ : −2 : K+ Φ+ : +6∂M+ .

It follows that
[Φ+ λ 4 : X+ G+ : −2 : K+ Φ+ : +6∂M+ ] = 0.
The non-commutative Wick formula (2.10) implies:
[Φ+ λ 4 : X+ G+ :] =4 : [Φ+ λ X+ ]G+ : +4 : X+ [Φ+ λ G+ ] :
Z λ
 
+4 [Φ+ λ X+ ]µ G+ dµ
0
=(−25)K+ λ2 + (−30) : Φ+ G+ : λ + (−10) : ∂Φ+ G+ :
+ 4 : X+ K + : .

We already know that [Φ+ λ M+ ] = 92 K+ λ + Φ+ (0) M+ . The non-

commutative Wick formula (2.10) then gives:
[Φ+ λ (−2) : K+ Φ+ :] =(−2) : [Φ+ λ K+ ]Φ+ : + : K+ [Φ+ λ Φ+ ] :
Z λ !
 
+ [Φ+ λ K+ ]µ Φ+ dµ
0
 
=(−2)K+ λ2 + 6Φ+ (0) M+ + 6 : G+ Φ+ : +(−27)∂K+ λ
+ 6 : M+ Φ+ : +3 : ∂G+ Φ+ : +(−12) : K+ X+ : .

30
 By sesquilinearity (2.5), we have
   
[Φ+ λ 6∂M+ ] = 27K+ λ2 + 6Φ+ (0) M+ + 27∂K+ λ + 6∂ Φ+ (0) M+ .

Looking at the coefficient of λ in [Φ+ λ 4 : X+ G+ : −2 : K+ Φ+ :

+6∂M+ ] = 0 and using that : Φ+ G+ : + : G+ Φ+ := ∂K+ (by quasi-
commutativity (2.8)), we obtain

Φ+ (0) M+ = (−3) : G+ Φ+ : + 52 ∂K+ ,

 
from where we conclude that [Φ+ λ M+ ] = 29 K+ λ− 3 : G+ Φ+ : − 25 ∂K+ .

2) [K+ λ M+ ]
Using that D is an odd derivation we get:
 
D Φ+ (n) M+ = K+ (n) M+ − Φ+ (n) ∂X+ .
 
In others words, K+ (n) M+ = D Φ+ (n) M+ + Φ+ (n) ∂X+ and, we can
compute [K+ λ M+ ] in terms of already known data. We conclude that:

15 11
[K+ λ M+ ] = − Φ+ λ2 − ∂Φ+ λ + 3 (: G+ K+ : −2 : L+ Φ+ :) .
2 2

3) [X+ λ K+ ]
Using that D is an odd derivation we have:
 
D K+ (n) M+ = ∂Φ+ (n) M+ + K+ (n) ∂X+ .
 
Then K+ (n) ∂X+ = D K+ (n) M+ − ∂Φ+ (n) M+ , that is, we can compute
[K+ λ X+ ] in terms of already known data. We conclude that:

[X+ λ K+ ] = −3K+ λ + 3 (: G+ Φ+ : −∂K+ ) .

4) [X+ λ M+ ]
We have [M+ λ X+ ] = [M+ λ 16 [Φ+ µ Φ+ ]]. Using the Jacobi identity (2.7) we
can express

[M+ λ [Φ+ µ Φ+ ]] = (−1)[Φ+ µ [M+ λ Φ+ ]] + [[M+ λ Φ+ ]λ+µ Φ+ ].

Putting µ = 0 in the above equation we can compute [M+ λ X+ ] in terms of

already known data. We obtain:
   
9 2 9 7 3 2
[X+ λ M+ ] = − G+ λ − 5M+ + ∂G+ λ + 4 : G+ X+ : − ∂M+ − ∂ G+ .
4 4 2 4

31
 5) [M+ λ M+ ]
Using that D is an odd derivation we have:
 
D X+ (n) M+ = M+ (n) M+ + X+ (n) ∂X+ .
 
Then M+ (n) M+ = D X+ (n) M+ − X+ (n) ∂X+ , that is, we can compute
[M+ λ M+ ] in terms of already known data. We conclude that:
 
35 4 1 2 9
[M+ λ M+ ] = − λ + (20X+ − 9L+ ) λ + 10∂X+ − ∂L+ λ
24 2 2
 
3 2 3 2
+ ∂ X+ − ∂ L+ − 4 : G+ M+ : +8 : L+ X+ : .
2 2

5.2.3 Checking the relation

To conlude the proof of the Conjecture 5.1 we need to prove the relation (5.18)
0 = 4 : G+ X+ : −2 : Φ+ K+ : −4∂M+ − ∂ 2 G+ .
used in the computation of the non-linear λ-brackets [Φ+ λ M+ ]. We check this
relation among the fields performing again an explicit computation. This is a
long calculation, even longer than the ones already performed in the proof of
Theorem 5.5. Besides the Mathematica package OPEdefs [Thi91], we have
used the computer algebra system Cadabra [Pee07], this last software was
proved very useful for simplifying tensorial expressions with many terms. Despite
its length this is a straightforward computation.
For these reasons we do not present here the computations, they are show up
online at the URL:
http://www.ime.unicamp.br/~lazarord/checking_the_
relation.pdf
Remark 5.8. As in Theorem 5.5 we make extensive use of: the Ricci flatness, the
contractions on page 18, the fact that ϕ, ψ and g are parallel and the symmetries
of the Riemann curvature tensor. Unlike the computations in Theorem 5.5, we
should point out that we need to use the identity (4.12), in fact this one appears
many times and is a key identity. It is also interesting to note the appearance of
the first Pontryagin class p1 (M ) = 8π1 2 Tr (R ∧ R) of the manifold, it appears as
one of the coefficients of the term ∂γ i ∂γ j ∂γ k cl , i.e, 8π1 2 Rijmn Rklmn ∂γ i ∂γ j ∂γ k cl ,
nevertheless this expression is identically zero because R is antisymmetric in i and
j while ∂γ i commutes with ∂γ j .

6 Computations
To perform all the computations below we are assuming that we are working in a
local coordinate chart where the volume form is constant, then we can assume that:
p
i ∂ log |g|
Γij = = 0, (6.1)
∂xj

32
where|g| denotes the absolute value of the determinant of the metric tensor g.
For convenience of the reader we recall the expression of the Riemann curvature
tensor Rlijk and the Ricci tensor Rij in terms of the Christoffel symbols. We also
recall some of its symmetries.

Rlijk = (Γlik ),j − (Γlij ),k + Γljs Γsik − Γlks Γsij , (6.2)

Riklm = Rlmik , Riklm = −Rkilm = −Rikml , (6.3)

Riklm + Rimkl + Rilmk = 0 (first Bianchi identity). (6.4)

Rij = Rlilk = glm Riljm = (Γlij ),l − (Γlil ),j + Γlij Γm m l

lm − Γil Γjm . (6.5)

The Ricci tensor is symmetric.

6.1 [Φ+ λ Φ+ ]
We have
Φ+ = 16 ϕijk ei+ ej+ ek+ + 21 ϕijk Γimn gjm ∂γ n ek+ .
To simplify the notation we drop the plus subscript.

a) Computing [ 16 ϕijk ei ej ek λ 61 ϕmnl em en el ]

For this we first compute [em en el λ ei ej ek ] and we can assume anti–
symmetrization in the indices {i, j, k} and {m, n, l} due to the future con-
traction with the three form ϕ.
 
[em en el λ ei ej ek ] =(−3)g im g jn g kl λ2 + 18g im g jn ek el − 6∂(g im g jn g kl ) λ
+ 9g im ej ek en el − 18g im g jn ∂(el )ek + 36∂(g im )g jn ek el
− 18∂(g im )∂(g jn )g kl .

Then using the non-commutative Wick formula (2.10):

1 1 1
[ ϕijk ei ej ek λ ϕmnl em en el ] = : ϕmnl [em en el −λ−∂ ϕijk ei ej ek ] :
6 6 36
1  
= : ϕmnl : ϕijk [em en el µ ei ej ek ] : :.
36 |−λ−∂

Collecting terms first by the order of λ and then by the number of factors e’s:
λ2

1 1 7
(−3)ϕmnl ϕijk gim g jn gkl = (−3)42 = − .
36 36 2

33
Here was used the identity (4.5).

λ
1 
−6ϕmnl ∂(ϕijk gim gjn g kl ) − 18ϕijk ϕmnl gim g jn ek el
36 
+ 6ϕmnl ϕijk ∂(gim gjn g kl )
1 1
= − ϕmnl ∂(ϕijk )gim gjn g kl − ϕijk ϕmnl gim gjn ek el
6 2
1 ijk k l
= − ϕ ∂(ϕijk ) − 3gkl e e
6
= 0.
Here was used the identity (4.6) and that gij ei ej = −gij ej ei + gij ∂(gij ) by
quasi-commutativity (2.8), then

1 1  
gij ei ej = gij ∂(gij ) = gij −Γiab gaj − Γjab gia ∂γ b = −Γiib ∂γ b . (6.6)
2 2
We also have by (4.3) and (4.6) that

 
ϕijk ∂(ϕijk ) = ϕijk Γaib ϕajk + Γajb ϕiak + Γakb ϕija ∂γ b = 18Γiib ∂γ b .
(6.7)
without λ
ei ej ek el
9 1
ϕijk ϕmnl g im ej ek en el = −ψjknl + gjn gkl − gjl gkn ej ek en el


36 4
1 1
= − ψijkl ei ej ek el + gij gkl ∂(g ik )∂(g jl )
4 8
1 ij kl
− gij gkl ∂(g )∂(g )
8
1 1
= − ψijkl ei ej ek el + Γcib Γicd ∂γ b ∂γ d
4 4
1 ic m j 1
+ gjm g Γib Γcd ∂γ ∂γ d − gij gkl ∂(g ij )∂(g kl ).
b
4 8
Here was used the identity (4.7), quasi-associativity (2.9), the equality
gij ei ej = 12 gij ∂(gij ) proved above and that metric g is covariantly constant.
∂(ei )ej
1 
−18ϕmnl ϕijk gim g jn ∂(ek )el − 18ϕmnl ϕijk g im gjn ek ∂(el )
36 
−18ϕmnl ϕijk gim g jn ∂(el )ek
3
= −3glk ∂(el )ek − glk ∂ 2 glk .
2

34
Here we used the identity (4.6) and quasi-commutativity (2.8):
Z 0
i j j i 1
e ∂(e ) = −∂(e )e + [ei λ ∂ej ]dλ = −∂(ej )ei + ∂ 2 (gij ).
−∂ 2

ei ej
1
− ϕmnl ∂(ϕijk g im g jn )ek el + ϕmnl ϕijk ∂(g im )g jn ek el
2
1
= − ϕmnl ∂(ϕijk )g im g jn ek el
2
1 1
= − ϕmnl ϕajk Γaib g im g jn ∂γ b ek el − ϕmnl ϕiak Γajb g im g jn ∂γ b ek el
2 2
1
− ϕmnl ϕija Γakb g im g jn ∂γ b ek el
2
1
= −ϕmnl ϕajk Γaib g im g jn ∂γ b ek el − ϕmnl ϕija Γakb g im g jn ∂γ b ek el
2
1
= (−1) (gam glk − gmk gla − ψmlak ) Γaib g im ∂γ b ek el + (− )6gla Γakb ∂γ b ek el
2
= ψmlak Γib g ∂γ e e + (−1)glk Γib ∂γ e e + (−2)gla Γakb ∂γ b ek el ,
a im b k l i b k l

1
= ψmlak Γaib g im ∂γ b ek el + (−2)gla Γakb ∂γ b ek el + (− )glk ∂(g lk )Γiib ∂γ b ,
2
1
= ψmlak Γib g ∂γ e e + (−2)gla Γkb ∂γ e e + gij ∂(g ij )glk ∂(g lk ).
a im b k l a b k l
4

Here was used that dϕ = 0, i.e., (4.3), the identities (4.7) and (4.6), and
(6.6).
without e’s
1   1
− ϕmnl ∂ 2 ϕijk g im g jn g kl − ϕmnl ϕijk ∂(g im )∂(g jn )g kl
12  2
1  
im jn kl
+ ϕmnl ∂ ϕijk ∂ g g g
6
1 3
= − ϕmnl ∂ 2 (ϕijk )g im g jn g kl + glk ∂ 2 (g lk )
12 2
1 2 3 2 lk
= − ϕijk ∂ (ϕijk ) + glk ∂ (g )
12 2
9   1 3
= ∂ gij ∂(g ) + ∂(ϕijk )∂(ϕijk ) + glk ∂ 2 (g lk )
ij
12 12 2
3 1 9
= ∂(gij )∂(g ) + ∂(ϕ )∂(ϕijk ) + glk ∂ 2 (g lk )
ij ijk
4 12 4
3 c i 3 ic a j
= − Γbi Γcd ∂γ ∂γ − gaj g Γbi Γcd ∂γ b ∂γ d − Γcbi Γicd ∂γ b ∂γ d
b d
2 2
1 i j 1 9
− Γbi Γjd ∂γ ∂γ − ψamjn Γabi g mi Γicd g nc ∂γ b ∂γ d + glk ∂ 2 (g lk )
b d
2 2 4
5 c i 3 1
= − Γbi Γcd ∂γ ∂γ − gaj g Γbi Γcd ∂γ ∂γ − ψamjn Γabi g mi Γjcd g nc ∂γ b ∂γ d
b d ic a j b d
2 2 2
1 ij lk 9 2 lk
= − gij ∂(g )glk ∂(g ) + glk ∂ g .
8 4

35
 In this chain of equalities was used that
 
∂ ϕijk ∂(ϕijk ) = ∂(ϕijk )∂(ϕijk ) + ϕijk ∂ 2 (ϕijk )

and ϕijk ∂(ϕijk ) = 9gij ∂(gij ); this last identity is proved combining (6.6)
and (6.7). We also used that dϕ = 0 i.e., (4.3), that g is covariantly constant
and the identity (4.7).

b) Computing
[ 16 ϕijk ei ej ek λ 21 ϕmnl Γm na b l 1 i ja b k 1 m n l
ab g ∂γ e ] + [ 2 ϕijk Γab g ∂γ e λ 6 ϕmnl e e e ].

1 1 1
[ ϕijk ei ej ek λ ϕmnl Γm na b l il na m b j k
ab g ∂γ e ] = ϕijk ϕmnl g g Γab ∂γ e e .
6 2 4
Using skew-symmetry (2.6) and arranging the indices we have:
1 1 1 1
[ ϕijk Γiab g ja ∂γ b ek λ ϕmnl em en el ] = [ ϕijk ei ej ek λ ϕmnl Γm na b l
ab g ∂γ e ],
2 6 6 2
then
1 1 1 b k 1
[ ϕijk ei ej ek λ ϕmnl Γm na b l i ja m n l
ab g ∂γ e ] + [ ϕijk Γab g ∂γ e λ ϕmnl e e e ]
6 2 2 6
1
= ϕijk ϕmnl gil gna Γm b j k
ab ∂γ e e
2
1
= gjm gkn − gjn gkm − ψjkmn gna Γm b j k


ab ∂γ e e
2
1 1 1
= gjm Γm b j k m b j k
kb ∂γ e e + (− )gkm Γjb ∂γ e e + (− )ψjkmn g Γab ∂γ e e
na m b j k
2 2 2
1 1
=(− )ψjkmn gna Γm b j k m b j k jk m
ab ∂γ e e + (−1)gkm Γjb ∂γ e e + gjm ∂(g )Γkb ∂γ
b
2 2
1
=(− )ψjkmn gna Γm b j k m
ab ∂γ e e + (−1)gkm Γjb ∂γ e e
b j k
2
1 1 k c
+ (− )gjm gck Γjcd Γm b d
kd ∂γ ∂γ + (− )Γcd Γkb ∂γ ∂γ .
b d
2 2
Here were used the identity (4.7), (6.6) and that g is covariantly constant.

c) Computing [ 21 ϕijk Γirs gjr ∂γ s ek λ 21 ϕmnl Γm na b l

ab g ∂γ e ].

1 1
[ ϕijk Γirs gjr ∂γ s ek λ ϕmnl Γm na b l
ab g ∂γ e ]
2 2
1
= ϕmnl ϕijk g lk Γm na b i jr
ab g ∂γ Γrs g ∂γ
s
4
1
gmi gnj − gmj gni − ψmnij Γm na b i jr s


= ab g ∂γ Γrs g ∂γ
4
1 1
= − ψmnij Γm na b i jr s na m i
ab g ∂γ Γrs g ∂γ + gmi g Γab Γns ∂γ ∂γ
b s
4 4
1
− Γm Γi ∂γ b ∂γ s .
4 ib ms

36
Finally we combining the result in a), b) and c) to obtain

[Φ+ λ Φ+ ] = (− 27 )λ2 + 6X+ , (6.8)

where X+ is:
1
X+ = − 24 ψijkl ei+ ej+ ek+ el+ − 14 ψijkl Γimn gjm ∂γ n ek+ el+
− 81 ψijkl Γim1 n1 gjm1 ∂γ n1 Γkm2 n2 g lm2 ∂γ n2 − 12 gij ∂(ei+ )ej+
− 12 gij Γjkl ∂γ k el+ ei+ − 14 Γijk Γkil ∂γ j ∂γ l ,

1
X− = − 24 ψijkl ei− ej− ek− el− + 14 ψijkl Γimn gjm ∂γ n ek− el−
− 81 ψijkl Γim1 n1 gjm1 ∂γ n1 Γkm2 n2 g lm2 ∂γ n2 + 12 gij ∂(ei− )ej−
+ 12 gij Γjkl ∂γ k el− ei− − 14 Γijk Γkil ∂γ j ∂γ l .

To get this precise form of X+ we only need to manipulate a little more the terms
without e’s that come from a), b) and c):

− 43 ψamjn Γabi g mi Γjcd g nc ∂γ b ∂γ d − 3Γcbi Γicd ∂γ b ∂γ d − 32 gaj g ic Γabi Γjcd ∂γ b ∂γ d

+ 43 glk ∂ 2 g lk
3 3  
= − 43 ψamjn Γabi g mi Γjcd g nc ∂γ b ∂γ d − Γcbi Γicd ∂γ b ∂γ d + ∂ gij ∂(g ij )
2 4
3 a mi j nc b d 3 c i b d
= − 4 ψamjn Γbi g Γcd g ∂γ ∂γ − Γbi Γcd ∂γ ∂γ .
2
The first equality is proved using:

∂(gij ∂(gij )) =∂(gij )∂(gij ) + gij ∂ 2 (gij )

=(−2)Γcbi Γicd ∂γ b ∂γ d + (−2)gaj g ic Γabi Γjcd ∂γ b ∂γ d + gij ∂ 2 (gij ),

which itself is proved using that g is covariantly constant. The second equality
follows from (6.6) and (6.1).

6.2 X+ (2) Φ+ and X+ (1) Φ+

To simplify the notation we drop the “+” subscript.
Below we only compute the λ-bracket between the sumands of X and Φ contribut-
ing to X(2) Φ and X(1) Φ.

37
 1
a) Computing [− 24 ψabcd ea eb ec ed λ 16 ϕijk ei ej ek ].

1 1
[− ψabcd ea eb ec ed λ ϕijk ei ej ek ]
 24 6  
1 1
= − ψabcd ϕijk g g g e λ + − ψabcd ϕijk gbi gdj ea ec ek
ai bj ck d 2
6 4

3 bi cj dk a 1 bi cj dk a
− ψabcd ϕijk g g ∂(g )e + ϕijk ∂(ψabcd g g g e ) λ
8 6
+ terms without λ
 
a c k 1 bi cj dk a
= −ϕack e e e + ϕijk ∂(ψabcd )g g g e λ + terms without λ
6
 
= −ϕijk e e e + (−2)ϕijk Γilm glj ∂γ m ek λ + terms without λ.
i j k

Here were used the identities (4.8) and (4.9), and that d ∗ ϕ = 0, i.e., (4.4).

b) Computing [− 41 ψabcd Γamn g bm ∂γ n ec ed λ 16 ϕijk ei ej ek ].

1 1
[− ψabcd Γamn gbm ∂γ n ec ed λ ϕijk ei ej ek ]
 4 6 
1 a bm n ci dj k
= ϕijk ψabcd Γmn g ∂γ g g e λ + terms without λ
4
 
= −ϕabk Γamn gbm ∂γ n ek + terms without λ.

Here was used the identity (4.9).

c) Computing [− 21 glb ∂(eb )el λ 16 ϕijk ei ej ek ].
1 1
[− glb ∂(eb )el λ ϕijk ei ej ek ]
 2 6 
1 2 1 1
= (− )ϕijk g e λ + (− )ϕijk ei ej ek + (− )ϕijk glb ∂(g bi g kl ej )
ij k
2 4 2

1
+ ϕijk ∂(g ik ej ) λ + terms without λ
2
 
1 i j k
= (− )ϕijk e e e λ + terms without λ.
4

d) Computing [− 21 glb ∂(eb )el λ 12 ϕijk Γimn g jm ∂γ n ek ].

1 1
[− glb ∂(eb )el λ ϕijk Γimn g jm ∂γ n ek ]
 2 2 
1 i jm n k
= − ϕijk Γmn g ∂γ e λ + terms without λ.
4

38
 e) Computing [− 21 gam Γanb ∂γ b en em λ 16 ϕijk ei ej ek ].

1 1
[− gam Γanb ∂γ b en em λ ϕijk ei ej ek ]
 2 6
1
= − ϕijk Γinb gnj ∂γ b ek λ + terms without λ.
2

Combining a), b), c), d) and e) we get that X+ (2) Φ+ = 0 and X+ (1) Φ+ =
− 15
2 Φ+ .

6.3 [Φ± λ K± ]
To perform this computation we express Φ± explicitly in terms of the bc − βγ
system.

Φ+ = 121√2 ϕijk ci cj ck + 1
√ ϕ gil cj ck bl + 4√
4 2 ijk
1
ϕ g il gjm ck bl bm
2 ijk
+ 1√
ϕ gil g jm gkn bl bm bn + 2√
12 2 ijk
1
ϕ Γi g jm ∂γ n gkl bl
2 ijk mn
+ 1
√ ϕ Γi gjm ∂γ n ck ,
2 2 ijk mn

i
Φ− = 12√ ϕ ci cj ck −
2 ijk
i
√ ϕ gil cj ck bl + 4√i 2 ϕijk g il gjm ck bl bm
4 2 ijk
− i
√ ϕ gil g jm gkn bl bm bn + 2√i 2 ϕijk Γimn g jm ∂γ n gkl bl
12 2 ijk
− i
√ ϕ Γi gjm ∂γ n ck .
2 2 ijk mn

As by definition K± := G(0) (Φ± ) we need to apply G(0) to each of the sum-

mands of Φ± . To simplify the notation we denote G(0) simply by D.

D(ϕijk ci cj ck ) = (ϕijk,l cl )ci cj ck + 3ϕijk ∂γ i cj ck ,

D(ϕijk gil cj ck bl ) =D(ϕljk cj ck bl )

=(ϕljk,mcm )cj ck bl + 2ϕljk ∂γ j ck bl + ϕljk cj ck βl ,

D(ϕijk gil g jm ck bl bm ) =D(ϕlmk ck bl bm )

=(ϕlmk,n cn )ck bl bm + ϕlmk ∂γ k bl bm + 2ϕlmk ck bl βm ,

D(ϕijk g il gjm gkn bl bm bn ) =D(ϕlmn bl bm bn )

=(ϕlmn,a ca )bl bm bn + 3ϕlmn bl bm βn ,

39
 D(ϕijk Γimn gjm ∂γ n gkl bl ) =D(ϕijk Γimn gjm gkl ∂γ n bl )
l
=(Fn,a ca )∂γ n bl + Fnl ∂cn bl + Fnl ∂γ n βl ,
where Fnl := ϕijk Γimn gjm gkl ,
D(ϕijk Γimn gjm ∂γ n ck ) = (Fkn,a ca )∂γ n ck + Fkn ∂cn ck + Fkn ∂γ n ∂γ k ,
where Fkn := ϕijk Γimn gjm .
Now we collect the non-zero λ-bracket between the summands of Φ± and the
summands of K± that contains terms with λ. To compute this λ-brackets we used
the Mathematica package [Thi91]:
[ϕijk ci cj ck λ D(ϕijk gil gjm ck bl bm )]

= 6ϕijk ϕjkn,s ci cn cs + 6ϕijk ϕkj n ∂γ n ci

+6ϕkmn ϕijk,m ci cj cn λ + terms without λ,

[ϕijk ci cj ck λ D(ϕijk gil g jm g kn bl bm bn )]

  λ2
= 6ϕijk ϕijk,s cs + 18ϕjkn ϕijk,n ci
 2
+ (−18)ϕijk ϕ ,s c c bl + (−18)ϕijk ∂ϕsjk,s ci
ljk i s

+6∂ϕijk ϕijk,s cs + (−18)ϕijk ϕjkn ci βn


+18ϕlkn ϕijk,n ci cj bl + 18ϕjkn ϕijk,n ∂ci λ
+ terms without λ,

[ϕijk g il cj ck bl λ D(ϕijk gil cj ck bl )]

 
= 4ϕlmi ϕijk,l cj ck cm + 4ϕlmi ϕijl,scj cm cs + (−4)ϕijk ϕkin ∂γ j cn λ
+ terms without λ,

[ϕijk g il cj ck bl λ D(ϕijk gil gjm ck bl bm )]

  λ2
= 2ϕlji ϕij k,l ck + (−2)ϕlji ϕij l,s cs + (−4)ϕij k ϕkmi,j cm
 2
l ij k m k ij m s
+ (−4)ϕ mj ϕ k,l c c bi + (−4)ϕ mj ϕ k,s c c bi
+(−4)ϕk mj ∂(ϕij k,i )cm + (−2)ϕlji ϕij k,s ck cs bl
+2∂(ϕl ji )ϕij k,l ck + (−2)∂(ϕk ji )ϕij k,s cs + 2ϕij k ϕlji ∂γ k bl
+4ϕij k ϕkmi cm βj + (−4)ϕij k ϕlmi,j ck cm bl

+(−2)ϕij k ϕkmn,j cm cn bi + (−4)ϕij k ϕkmi,j ∂cm λ
+ terms without λ,

40
 [ϕijk g il cj ck bl λ D(ϕijk gil gjm gkn bl bm bn )]

= 6ϕlmk ϕijk,l cm bi bj + 6ϕlkj ϕijk,s cs bi bl + 6ϕlkj ∂(ϕijk,i )bl

+6∂(ϕlkj )ϕijk,l bi + (−6)ϕijk ϕlij bl βk


+12ϕijk ϕlmj,k cm bi bl + 6ϕijk ϕlij,k ∂bl λ
+ terms without λ,

[ϕijk g il cj ck bl λ D(ϕijk Γimn gjm ∂γ n g kl bl )]

  λ2 
= (−2)Fji ϕj in cn + (−2)ϕk in Fj,k i
∂γ j cn
2 
+Fji ϕj mn cm cn bi + (−2)Fji ϕj in ∂cn + (−2)Fji ∂ϕj in cn λ
+ terms without λ,

[ϕijk g il cj ck bl λ D(ϕijk Γimn gjm ∂γ n ck )]

 
= Fji ϕimn cj cm cn λ + terms without λ,

[ϕijk gil gjm ck bl bm λ D(ϕijk ci cj ck )]

 
= −6ϕij k ϕlmj,i ck cl cm − 6ϕij k ϕlji,s ck cl cs + 6ϕij k ϕlji ∂γ l ck λ
+ terms without λ,

[ϕijk gil g jm ck bl bm λ D(ϕijk g il cj ck bl )]

  λ2
= (−4)ϕij k ϕkmj,i cm + 2ϕij k ϕkij,s cs + 2ϕlij ϕij k,l ck
 2
ij l k m ij l k s
+ 4ϕ k ϕ mj,i c c bl + 2ϕ k ϕ ji,s c c bl
+(−2)ϕij k ϕkmn,j cm cn bi + 4ϕij k ϕkmj,s cm cs bi
+2ϕij k ∂(ϕlji,l )ck + 2∂(ϕij k )ϕk ij,scs − 2∂(ϕij k )ϕk mj,i cm
+2∂(ϕij k )ϕkjn,i cn + (−4)ϕij k ϕkmj ∂γ m bi

+2ϕij k ϕlji ck βl + (−4)ϕlmj ϕij k,l ck cm bi + 2ϕlij ϕij k,l ∂ck λ
+ terms without λ,

[ϕijk gil gjm ck bl bm λ D(ϕijk gil g jm ck bl bm )]


= 4ϕkmn ϕij k,mcn bi bj + 8ϕij k ϕlkn,j cn bi bl
−4ϕij k ϕlkj,s cs bi bl + 4ϕij k ϕkmi bj βm
+4ϕij k ∂(ϕlk j,l )bi + 4∂(ϕij k )ϕlkj,i bl

+4ϕkmj ϕij k,m∂bi λ + terms without λ,

41
 [ϕijk g il gjm ck bl bm λ D(ϕijk gil gjm gkn bl bm bn )]
 
= 12ϕij k ϕlmk,i bj bl bm λ + terms without λ,

[ϕijk gil g jm ck bl bm λ D(ϕijk Γimn gjm ∂γ n g kl bl )]

  λ2 
= 2ϕij k Fik bj + 2ϕij k Fm,ik
∂γ m bj
2 
+2ϕij k Fjl ck bi bl + 2ϕij k Fik ∂bj + 2∂(ϕij k )Fik bj λ
+ terms without λ,

[ϕijk g il gjm ck bl bm λ D(ϕijk Γimn gjm ∂γ n ck )]

  λ2 
= 2ϕij k Fij ck + 2ϕij k Fil,j ∂γ l ck + 2ϕij k Fmi ck cm bj
2 
+2ϕij k Fij ∂ck + 2∂(ϕij k )Fij ck λ + terms without λ,

[ϕijk gil gjm gkn bl bm bn λ D(ϕijk ci cj ck )]

  λ2
= 18ϕijk ϕlkj,i cl + 6ϕijk ϕijk,s cs
 2
+ 18ϕ ϕlmj,k c c bi + 18ϕ ϕljk,s cl cs bi
ijk l m ijk

+18∂(ϕijk )ϕlkj,i cl + 6∂(ϕijk )ϕijk,s cs


+18ϕijk ϕlkj ∂γ l bi λ + terms without λ,

[ϕijk gil g jm g kn bl bm bn λ D(ϕijk gil cj ck bl )]


= 12ϕijk ϕlmj,k cm bi bl + 6ϕijk ϕljk,scs bi bl
+6ϕijk ∂(ϕlkj,l )bi + 6∂(ϕijk )ϕlkj,i bl + 6ϕijk ϕlkj bi βl

+6ϕlmk ϕijk,l cm bi bj + 6ϕljk ϕijk,l ∂bi λ
+ terms without λ,

[ϕijk g il gjm gkn bl bm bn λ D(ϕijk gil gjm ck bl bm )]

 
= 12ϕijk ϕlmi,j bk bl bm λ + terms without λ,

[ϕijk gil gjm gkn bl bm bn λ D(ϕijk Γimn gjm ∂γ n gkl bl )]

 
= 3ϕijk Fil bj bk bl λ + terms without λ,

42
 [ϕijk gil gjm gkn bl bm bn λ D(ϕijk Γimn gjm ∂γ n ck )]
  λ2 
= 6ϕijk Fij bk + 6ϕijk Fil,j ∂γ l bk
2 
+3ϕijk Fmi cm bj bk + 6ϕijk Fij ∂bk + 6∂(ϕijk )Fij bk λ
+ terms without λ,

l
[Fm ∂γ m bl λ D(ϕijk gil cj ck bl )]
  λ2 
= 2Fij ϕijk ck + Fil ϕijk cj ck bl
2 
+2ϕijk Fm,i
k
∂γ m cj + 2∂(Fij )ϕijk ck λ + terms without λ,

l
[Fm ∂γ m bl λ D(ϕijk gil gjm ck bl bm )]
  λ2 
= 2Fjk ϕij k bi + 2Fjl ϕij k ck bi bl + (−2)ϕij k Fm,j
k
∂γ m bi
2
+2∂(Fjk )ϕij k bi λ + terms without λ,

 
l
[Fm ∂γ m bl λ D(ϕijk gil g jm g kn bl bm bn )] = 3Fkl ϕijk bi bj bl λ + terms without λ,

 
l
[Fm ∂γ m bl λ D(ϕijk Γimn gjm ∂γ n gkl bl )] = 2Fm
j i
Fj ∂γ m bi λ + terms without λ,

 
l
[Fm ∂γ m bl λ D(ϕijk Γimn gjm ∂γ n ck )] = Fm
i
Fji ∂γ m cj λ + terms without λ,

 
[Fji ∂γ i cj λ D(ϕijk gil cj ck bl )] = Fji ϕimn cj cm cn λ + terms without λ,

[Fji ∂γ i cj λ D(ϕijk gil gjm ck bl bm )]

  λ2 
= 2Fji ϕij n cn + 2Fji ϕlin cj cn bl
2 
+2ϕjmn Fji,m ∂γ i cn + 2∂(Fji )ϕij n cn λ + terms without λ,

[Fji ∂γ i cj λ D(ϕijk gil gjm g kn bl bm bn )]

  λ2 
= 6Fji ϕijm bm + 3Fji ϕlmi cj bl bm
2 
+6ϕljn Fji,n ∂γ i bl + 6∂(Fji )ϕijl bl λ + terms without λ,

 
[Fji ∂γ i cj λ D(ϕijk Γimn gjm ∂γ n gkl bl )] = Fji Fm
i
∂γ m cj λ + terms without λ.

43
6.3.1 (Φ+ (1) K+ ) + (Φ− (1) K− )
Now we want to compute (Φ+ (1) K+ ) + (Φ− (1) K− ), it should be notice that the
λ-brackets used to compute Φ+ (1) K+ and Φ− (1) K− are the same modulo a sign.
Then to compute (Φ+ (1) K+ ) + (Φ− (1) K− ) we only need to take into account the
λ-brackets that have the same sign an consider each one twice.
We compute (Φ+ (1) K+ ) + (Φ− (1) K− ) analyzing the coefficient of each type of
term that appears. All the coefficients were obtained after a long but straightfor-
ward computation using identities (4.6) and (4.7), and the fact that dϕ = 0 and
∇g = 0, except the coefficient of ∂γ i cj that is more involved and is detailed below.

coefficient of bi bj bk : 0,
coefficient of ci cj bk : 0,
coefficient of ∂ci : 0,
coefficient of ∂γ i bi : (−3)∂γ i bi ,
coefficient of ci βi : (−3)ci βi ,
coefficient of ∂γ i cj : 0.
Computations to obtain the coefficient of ∂γ i cj :
Denote by A1 the terms of type ∂γ i cj that appear in the computations of the ci βi
coefficient due to quasi-associativity (2.9):
1 1 1
A1 = − ∂ϕijk ϕjkn,n ci + (− )∂ϕjkn ϕijk,nci + ∂ϕij k ϕkmi,j cm
8 8 4
1 k ij m 1 ij l k 1 l ij k
+ ∂ϕ mi ϕ k,j c + ∂ϕ k ϕ ji,l c + ∂ϕ ji ϕ k,l c ,
4 8 8
Collecting the other terms that contain ∂γ i cj : denote by A2 the sum of the terms
that does not contain derivatives of the Christoffel symbols, denote by A3 the sum
of the terms containing derivatives of the Christoffel symbols.
We have:
1 ml n 1
A3 = ϕ i ϕ ls (Γimn ),r ∂γ r cs + ϕsrl ϕmli (Γimn ),s ∂γ n cr .
2 2
Using the identity (4.7) we get:
1 1 1
A3 = gis gmn (Γimn ),r ∂γ r cs + (− )gri gsm (Γimn ),s ∂γ n cr + (Γsmn ),s ∂γ n cm
2 2 2
1 sm i
+ ψ r i (Γmn ),s ∂γ n cr .
2
Let R denote the Riemann curvature, using the identity (6.2) we can work the first

44
two summands of A3 :
1
gis gmn (Γimn ),r ∂γ r cs
2
1 1
= gis gmn Rinrm ∂γ r cs + gis g mn (Γrin ),m ∂γ r cs
2 2
1 1
+ (− )gis gmn Γria Γman ∂γ r cs + gis g mn Γmia Γran ∂γ r cs
2 2
1 1 1
= Rsr ∂γ c + gis g (Γr n ),m ∂γ r cs + (− )Γias Γj ij ∂γ s ca
r s mn i
2 2 2
1
+ Γiaj Γij s ∂γ s ca ,
2

1
gri gsm (Γimn ),s ∂γ n cr
2
1 1
=(− )gri g sm Rinsm ∂γ n cr + (− )gri gsm (Γsin ),m ∂γ n cr
2 2
1 1
+ gri g sm Γsia Γman ∂γ n cr + (− )gri gsm Γmia Γsan ∂γ n cr
2 2
1 sm i n r
=(− )gri g (Γs n ),m ∂γ c
2
1
= − gis gmn (Γrin ),m ∂γ r cs ,
2
then
1 1 1 1
A3 = Rsr ∂γ r cs + (− )Γias Γj ij ∂γ s ca + Γiaj Γij s ∂γ s ca + (Γsmn ),s ∂γ n cm
2 2 2 2
1 sm i
+ ψ r i (Γmn ),s ∂γ n cr .
2
We also have
1 1 1
A1 + A2 =(− )Γaij Γij s ∂γ s ca + Γias Γj ij ∂γ s ca + (− )Γiaj Γij s ∂γ s ca
2 2 2
1 j i k l s a
+ Γi k Γl s ψ a j ∂γ c ,
2

45
then

A1 + A2 + A3
1 1 1
=(− )Γaij Γij s ∂γ s ca + Γij k Γlis ψ ka lj ∂γ s ca + Rsr ∂γ r cs
2 2 2
1 s n m 1 sm i n r
+ (Γmn ),s ∂γ c + ψ r i (Γmn ),s ∂γ c
2 2
1 s 1 1
=( (Γmn ),s ∂γ n cm + (− )Γaij Γij s ∂γ s ca ) + ( ψ sr mi (Γimn ),s ∂γ n cr
2 2 2
1 sm i a n r 1 r s
+ ψ r i Γa s Γm n ∂γ c ) + Rsr ∂γ c
2 2
1 1 sm i 1
= Rmn ∂γ c + ψ r i R msn ∂γ n cr + Rsr ∂γ r cs
n m
2 2 2
1
=Rmn ∂γ n cm + Rabcd ψ cdlm g am ∂γ b cl ,
2
=0.

To conclude the last equality is zero we use the Lemma 4.3 and Lemma 4.4.
Finally we have proved that

(Φ+ (1) K+ ) + (Φ− (1) K− ) = (−3)∂γ i bi + (−3)ci βi

6.3.2 Φ± (1) K±
 
Now we want to compute Φ+ (1) K+ Φ− (1) K− , as was noted in 6.3.1 the
λ-brackets used to compute Φ+ (1) K+ and Φ− (1) K− are exactly the same mod-
 
ulo a sign. Then to compute (Φ+ (1) K+ ) Φ− (1) K− we only need to con-
sider the λ-brackets that change sign and remember  to add one half of the sum
(Φ+ (1) K+ ) + (Φ− (1) K− ). We compute Φ+ (1) K+ Φ− (1) K− analyzing the co-
efficient of each type of term that appears. All the coefficients were obtained after
a long but straightforward computation using identities (4.6) and (4.7), and the fact
that dϕ = 0 and ∇g = 0, except the coefficient of ∂γ i bj that is more involved and
is detailed below.
coefficient of bi βj : (− 32 )gij bi βj ,
coefficient of ∂bi : (− 32 )gij Γkij ∂bk ,
coefficient of ci cj ck : 0,
coefficient of ci b j bk : (− 32 )gij Γlik ck bj bl ,
coefficient of ∂γ i cj : (− 32 )gij ∂γ i cj ,
coefficient of ∂γ i bj : (−3)gij Γkil Γljm ∂γ m bk .

46
 Computations to obtain the coefficient of ∂γ i bj .

Denote by A1 the terms of type ∂γ i bj that appear in the computations of the

bi βj coefficient due to quasi-associativity (2.9):

1 1 1
A1 =(− )∂(ϕijk )ϕlij,k bl + (− )∂(ϕlij )ϕijk,k bl + ∂(ϕijk )ϕlkj,l bi
16 16 16
1 1 1
+ ∂(ϕlkj )ϕijk,l bi + ∂(ϕij k )ϕkmi,m bj + ∂(ϕkmi )ϕij k,mbj .
16 8 8
Collecting the other terms that contain ∂γ i bj : denote by A2 the sum of the terms
that does not contain derivatives of the Christoffel symbols, denote by A3 the sum
of the terms containing derivatives of the Christoffel symbols.
We have:
1 1 7
A1 + A2 =(− )Γij s Γaij ∂γ s ba + Γias Γj ij ∂γ s ba + (− )Γiaj Γij s ∂γ s ba
2 2 2
1 a kl j i s
+ (− )ψ j Γi k Γl s ∂γ ba ,
2

1 1 1
A3 = glk (Γkil ),s ∂γ s bi + (− )grm (Γmis ),r ∂γ s bi + gam (Γmis ),i ∂γ s ba
2 2 2
1 ramn
+ ψ gin (Γmis ),r ∂γ s ba .
2
Let R denote the Riemann curvature, using the identity (6.2) we can write:
1 lk
g (Γkil ),s ∂γ s bi
2
1 1 1
= gkl Rilsk ∂γ s bi + gkl (Γsil ),k ∂γ s bi + (− )gkl Γsia Γkal ∂γ s bi
2 2 2
1 kl i a s
+ g Γk a Γs l ∂γ bi ,
2
and using identity (6.5) we have
1 am 1 1
g (Γmis ),i ∂γ s ba = gam Rms ∂γ s ba + gam Γmj l Γslj ∂γ s ba .
2 2 2
Then
1 1 1
A3 = Ris ∂γ s bi + (− )gkl Γsia Γkal ∂γ s bi + + gkl Γkia Γsal ∂γ s bi
2 2 2
1 a s 1 am j l s 1 ramn
+ R s ∂γ ba + g Γm l Γs j ∂γ ba + ψ gin (Γmis ),r ∂γ s ba .
2 2 2

47
Using again the identity (6.2) we can write:
1 1
(− )ψ aj kl Γij k Γlis ∂γ s ba + ψ ramn gin (Γmis ),r ∂γ s ba
2 2
1 a kl j i 1
=(− )ψ j Γk i Γl s ∂γ ba + (− )ψ aj kl (Γlj s ),k ∂γ s ba
s
2 2
1 a kl j
=(− )ψ j R lks ∂γ s ba
2
1
= Rksjl ψ jli k gai ∂γ s ba ,
2
then
1
A1 + A2 + A3 =(−3)gkl Γkia Γsal ∂γ s bi + Rij ∂γ j bi + Rksjl ψ jli k gai ∂γ s ba
2
=(−3)gij Γkil Γljm∂γ m bk .

To establish the last equality we use the Lemma 4.3 and Lemma 4.4.
Finally we have proved that:

Φ+ (1) K+ = − 23 ci βi − 23 ∂γ i bi − 32 g ij bi βj − 23 gij Γlik ck bj bl − 23 gij Γkij ∂bk

− 32 gij ∂γ i cj − 3gij Γkil Γljm ∂γ m bk ,

Φ− (1) K− = − 23 ci βi − 23 ∂γ i bi + ( 32 )gij bi βj + ( 32 )gij Γlik ck bj bl + ( 32 )gij Γkij ∂bk

+ ( 32 )gij ∂γ i cj + 3gij Γkil Γljm∂γ m bk .

6.4 [G+ λ G− ]
Now we compute [G+ λ G− ],
1 i i 1 ij 1 ij l k
G+ = 1
2 c βi + 2 ∂γ bi + 2 g bi βj + 2 g Γik c bj bl + 12 g ij Γkij ∂bk
+ 21 gij ∂γ i cj + gij Γkil Γljm∂γ m bk ,

1 i i 1 ij 1 ij l k
G− = 1
2 c βi + 2 ∂γ bi + (− 2 )g bi βj + (− 2 )g Γik c bj bl + (− 12 )gij Γkij ∂bk
+(− 12 )gij ∂γ i cj + (−1)g ij Γkil Γljm ∂γ m bk .

We list the non-zero λ-brackets between the summands of G+ and the summands
of G− , to compute these we used the Mathematica package [Thi91].

[ci βi λ ∂γ j bj ] = 27 λ2 + (ci bi )λ + ∂ci bi + ∂γ j βj ,

  λ2  
[ci βi λ g lm bl βm ] = (−1)(g,llm ),m + g,llm βm + (−1)(g,ilm ),m ci bl λ
2
+ g βl βm + g,i c bl βm + (−1)(g,ilm ),m ∂ci bl + ∂(g,llm )βm ,
lm lm i

48
 [ci βi λ glm Γnla ca bm bn ]
 
= (glm Γnla ),n ca bm + (−1)(g lm Γnla ),m ca bn λ
+ (glm Γnla )ca bm βn + (−1)(g lm Γnla )ca bn βm
+ (−1)(glm Γnla ),i ca ci bm bn + ∂[(glm Γnla ),n ]ca bm
+ (−1)∂[(glm Γnla ),m ]ca bn ,

  λ2  
[ci βi λ glm Γnlm ∂bn ] = (glm Γnlm ),n + (glm Γnlm )βn + ∂[(glm Γnlm ),n ] λ
2
1
(g Γlm )∂βn + (glm Γnlm ),i ci ∂bn + ∂ 2 [(glm Γnlm ),n ],
lm n
2

[ci βi λ glm ∂γ l cm ] =glm ∂cl cm + (glm ),i ∂γ l ci cm ,

[ci βi λ glm Γnla Γamr ∂γ r bn ]

  λ2  
= glm Γnla Γamn + glm Γnla Γami ci bn + (g lm Γnla Γamr ),n ∂γ r λ
2
+ g lm Γnla Γami ∂ci bn + (g lm Γnla Γamr )∂γ r βn
+ (g lm Γnla Γamr ),i ∂γ r ci bn + ∂[(glm Γnla Γamr ),n ]∂γ r ,

 
[∂γ i bi λ cj βj ] = 27 λ2 + (−1)ci bi λ + (−1)ci ∂bi + ∂γ i βi ,

[∂γ i bi λ glm bl βm ] =glm ∂bm bl ,

[∂γ i bi λ glm Γnla ca bm bn ] =glm Γnli ∂γ i bm bn ,

[∂γ i bi λ glm ∂γ l cm ] =glm ∂γ l ∂γ m ,

  λ2 
[g lm bl βm λ ci βi ] = (−1)(g,llm ),m + (−1)(g,llm )βm + (g,ilm ),m ci bl
2
+(−1)∂[(g,llm ),m ] λ + glm βl βm + g,ilm ci bl βm
+ (−1)g,llm ∂βm + (g,ilm ),m ci ∂bl + ∂[(g,ilm ),m ]ci bl
1
+ (− )∂ 2 [(g,llm ),m ],
2

[glm bl βm λ ∂γ i bi ] =(−1)g lm bm ∂bl ,

49
 [glm bl βm λ gij bi βj ]
 
lm ij
= (g,j )(g,m )bi bl λ + (−1)glm g,m
ij
bi bl βj + glm (g,m
ij
),j bi ∂bl
+ g ij (g,j
lm
)bi bl βm + g ij (g,j
lm
),m bi ∂bl + g ij ∂[(g,j
lm
),m ]bi bl
lm ij lm ij
+ g,j g,m bi ∂bl + g,j ∂[(g,m )]bi bl + ∂(glm )(g,m
ij
),j bi bl ,

[glm bl βm λ gij Γaik ck bj ba ]

 
= (−1)glm (g ij Γail ),m ba bj λ + (−1)glm (gij Γail )ba bj βm
+ glm (gij Γaik ),m ck ba bj bl + (−1)g lm ∂[(gij Γail ),m ]ba bj
+ (−1)∂(glm )(g ij Γail ),m ba bj ,

[g lm bl βm λ gij Γkij ∂bk ] =(−1)g lm (gij Γkij ),m ∂bk bl ,

[glm bl βm λ gij ∂γ i cj ]
1  
= λ2 + (−1)cj bj + g lm (gil ),m ∂γ i + glm ∂(glm ) λ
2
+ (−1)cj ∂bj + ∂γ i βi + (−1)g lm (gij ),m ∂γ i cj bl
+ glm ∂[(gil ),m ]∂γ i + (−1)∂(glm )gmj cj bl + 2∂(glm )(gil ),m ∂γ i
1
+ glm ∂ 2 (glm ) + ∂(gil )(glm ),m ∂γ i ,
2

[g lm bl βm λ gij Γkin Γnja∂γ a bk ]

 
= (−1)g lm gij Γkin Γnjm bk bl λ + (−1)glm gij Γkin Γnjm bk ∂bl
+ (−1)g lm (gij Γkin Γnja ),m ∂γ a bk bl + (−1)∂(g lm )gij Γkin Γnjmbk bl ,

[gij Γlik ck bj bl λ cm βm ]
 
= (−1)(gij Γm ik ),m ck
b j + (g im l
Γ ik ),m ck
b l λ
+ (gij Γm k im l k ij l k m
ik )c bj βm + (−1)(g Γik )c bl βm + (−1)(g Γik ),m c c bj bl
(−1)(gij Γm k im l k ij m k
ik ),m c ∂bj + (g Γik ),m c ∂bl + (−1)(g Γik ),m ∂c bj
+ (gim Γlik ),m ∂ck bl ,

[gij Γlik ck bj bl λ ∂γ m bm ] =gij Γlim ∂γ m bj bl ,

50
 [g ij Γlik ck bj bl λ gmn bm βn ]
 
= (−1)g mn (gij Γlim ),n bj bl λ + (gij Γlim )gmn bj bl βn
+ (−1)g mn (gij Γlim ),n bj ∂bl + (−1)g mn (gij Γlik ),n ck bj bl bm
+ (−1)g mn (gij Γlim ),n ∂bj bl ,

[gij Γlik ck bj bl λ gmn Γrma ca bn br ]


= gij Γlir gmn Γrml bj bn + (−1)gij Γlin gmn Γrml bj br

+(−1)g ij Γlir gmn Γrmj bl bn + g ij Γlin g mn Γrmj bl br λ
+ gij Γlik gmn Γkma ca bj bl bn + (−1)g ij Γlik gmk Γrma ca bj bl br
+ gij Γlik gmn Γrml ck bj bn br + (−1)gij Γlik g mn Γrmj ck bl bn br
+ gij Γlik gmn Γkml ∂bj bn + (−1)gij Γlik gmk Γrml ∂bj br
+ (−1)g ij Γlik gmn Γkmj ∂bl bn + gij Γlik gmk Γrmj ∂bl br
+ ∂(gij Γlik )gmn Γkml bj bn + (−1)∂(gij Γlik )gmk Γrml bj br
+ (−1)∂(gij Γlik )g mn Γkmj bl bn + ∂(gij Γlik )gmk Γrmj bl br ,

 
[gij Γlik ck bj bl λ gmn Γamn ∂ba ] = gij Γlik gmn Γkmn bj bl λ + gij Γlik gmn Γkmn bj ∂bl
+ g ij Γlik gmn Γkmn ∂bj bl + ∂(gij Γlik )gmn Γkmn bj bl ,

[gij Γlik ck bj bl λ gmn ∂γ m cn ] =gij Γlik glm ∂γ m ck bj + (−1)Γlik ∂γ i ck bl ,

[g ij Γlik ck bj bl λ gmn Γams Γsnr ∂γ r ba ] =gij Γlik g mn Γkms Γsnr ∂γ r bj bl ,

  λ2  
[gij Γkij ∂bk λ cm βm ] = (gij Γkij ),k + (−1)(gij Γkij )βk λ
2
+ (g Γij ),m cm ∂bk + (−1)∂(gij Γkij )βk ,
ij k

[g ij Γkij ∂bk λ g lm bl βm ] =(−1)g lm (gij Γkij ),m ∂bk bl ,

[gij Γkij ∂bk λ glm Γnla ca bm bn ]

 
= (−1)g ij Γkij g lm Γnlk bm bn λ + (−1)∂(gij Γkij )glm Γnlk bm bn ,

 
[gij Γkij ∂bk λ glm ∂γ l cm ] = (−1)gij Γkij glk ∂γ l λ + (−1)∂(gij Γkij )glk ∂γ l ,

51
 [gmn ∂γ m cn λ ci βi ] =(−1)gmn cm ∂cn + (gmn ),i ∂γ m ci cn ,

[gmn ∂γ m cn λ ∂γ i bi ] =gmn ∂γ m ∂γ n ,

[gmn ∂γ m cn λ gij bi βj ]
λ2  
= + ci bi + (−1)g ij (gmi ),j ∂γ m + gij ∂(gij ) λ
2
+ ∂ci bi + ∂γ j βj + (−1)gij (gmn ),j ∂γ m cn bi + (−1)gij (gmi ),j ∂ 2 γ m
1
+ ∂(gmn )gim cn bi + ∂ 2 (gij )gij + ∂(gij )(gmi ),j ∂γ m
2
ij m
+ ∂(gmi )(g ),j ∂γ ,

[gmn ∂γ m cn λ gij Γlik ck bj bl ] =glm gij Γlik ∂γ m ck bj + (−1)Γlik ∂γ i ck bl ,

[gmn ∂γ m cn λ gij Γkij ∂bk ]

 
= gmk gij Γkij ∂γ m λ + gmk gij Γkij ∂ 2 γ m + ∂(gmk )gij Γkij ∂γ m ,

[gmn ∂γ m cn λ gij Γkil Γlja ∂γ a bk ] =gmn gij Γnil Γlja ∂γ a ∂γ m ,

[gij Γkil Γlja ∂γ a bk λ cm βm ]

  λ2 
= g ij Γkil Γljk + (−1)gij Γkil Γljm cm bk
2 
+(−1)(gij Γkil Γlja ),k ∂γ a + ∂(gij Γkil Γljk ) λ + (−1)gij Γkil Γljm cm ∂bk
+ (gij Γkil Γlja )∂γ a βk + (g ij Γkil Γlja ),m ∂γ a cm bk + (−1)(gij Γkil Γlja ),k ∂ 2 γ a
+ (−1)∂(gij Γkil Γljm)cm bk + 12 ∂ 2 (gij Γkil Γljk ),

[gij Γkil Γlja∂γ a bk λ gmn bm βn ]

 
= gij Γkil Γlja gma bk bm λ + gij Γkil Γlja gma ∂bk bm
(−1)(gij Γkil Γlja),n gmn ∂γ a bk bm + ∂(gij Γkil Γlja)gma bk bm ,

[gij Γkil Γlja ∂γ a bk λ gmn Γrms cs bn br ] =gij Γkil Γljagmn Γrmk ∂γ a bn br ,

[gij Γkil Γlja ∂γ a bk λ g mn ∂γ m cn ] =g ij Γkil Γlja gmk ∂γ a ∂γ m .

52
6.4.1 Coefficient of λ2
We prove that it is zero after an easy computation using only that the metric is
covariantly constant.

6.4.2 Coefficient of λ
We compute the coefficient of each type of term that appears:

• coefficient ci bj : 0,

(−1)(glm Γnla ),n ca bm + (−1)g lm Γnla Γami ci bn

=gsm Γlsn Γnla ca bm + (−1)glm (Γnla ),n ca bm
=gsm (Γnsa ),n ca bm + (−1)g sm Rsa ca bm + (−1)glm (Γnla ),n ca bm
=(−1)gsm Rsa ca bm
=0,
by Lemma (4.3).

• coefficient ∂γ i : 0,
1
(−1)(gij Γkil Γlja ),k ∂γ a + ∂(gij Γkil Γljk ).
2
Expanding each summand separately:
(−1)(gij Γkil Γlja ),k ∂γ a
ij k l
=(−1)g,k Γil Γja ∂γ a + (−1)gij (Γkil ),k Γlja∂γ a + (−1)g ij Γkil (Γlja ),k ∂γ a
=(gsj Γisk + gis Γjsk )Γkil Γlja ∂γ a + (−1)gij (Γm k l a ij l
ik Γlm )Γja ∂γ + (−1)g Ril Γja ∂γ
a
h i
+ (−1)g ij Γkil Rjkal
+ (Γljk ),a − Γlks Γsja + Γlas Γsjk ∂γ a
=(−1)gij Ril Γlja ∂γ a + (−1)g ij Γkil (Γljk ),a ∂γ a + gij Γkil Γlks Γsja ∂γ a .
Above, to cancel out the summand involving the curvature tensor we used
that the Riemann’s curvature tensor is antisymmetric in the first two indices
(6.3).
1
∂(gij Γkil Γljk )
2
1 ij k l 1 1
= g,a Γil Γjk ∂γ a + gij (Γkil ),a Γljk ∂γ a + gij Γkil (Γljk ),a ∂γ a
2 2 2
1  sj i 
= −g Γsa − g Γsa Γil Γjk ∂γ + g Γil (Γljk ),a ∂γ a
is j k l a ij k
2
=(−1)gsj Γisa Γkil Γljk ∂γ a + gij Γkil (Γljk ),a ∂γ a .
Combining the two expansions and using Lemma 4.3 we get the desired
result.

53
6.4.3 Coefficient of λ0
We compute the coefficient of each type of term that appears:
• coefficient ∂ci bj : 0,
1 lm 1 1
(g ),m ∂ci bl + (− )glm Γnla Γami ∂ci bn + (− )(gij Γm k
ik ),m ∂c bj
4 ,i 2 4
1
+ (gim Γlik ),m ∂ck bl
4
1 lm n a 1
=(− )g Γla Γmi ∂ci bn + (− )(gij Γm k
ik ),m ∂c bj
2 2
1 1 ij m
= g aj Γima Γm k
ik ∂c bj + (− )g (Γik ),m ∂c bj
k
2 2
1
=(− )gsj Rsk ∂ck bj
2
=0,
by Lemma 4.3.

• coefficient ∂γ i βj : 0.

• coefficient ci bj βk : 0.

• coefficient ci ∂bj : 0,
1 1
(− )glm Γnla Γami ci ∂bn + (− )(gij Γm k
ik ),m c ∂bj
2 2
1 1 ij m
= gaj Γima Γm k k
ik c ∂bj + (− )(g Γik ),m c ∂bj
2 2
1
=(− )gsj Rsk ck ∂bj
2
=0,
by Lemma 4.3.

• coefficient ci cj bm bn : 0.

• coefficient ∂γ i cj bk : 0,
using that the metric is covariantly constant we get:
1 h i 1
(− )∂ (glk Γnlm ),n cm bk + (− )∂(gij Γkil Γljm )cm bk
2 2
1 h lk n i
=(− )∂ (g Γlm ),n + gij Γkil Γljm cm bk
2
1
=(− )∂ [gsm Rsa ] ca bm
2
=0,

54
 by Lemma 4.3.

• coefficient ∂βm : 0.

• coefficient ∂ci cj : 0.

• coefficient ∂γ i cj ck : 0.

• coefficient bi bj β k : 0,
using that the metric is covariantly constant we get:
1  
(− ) glm gia Γjam bi bl βj = 0,
2
because the factor between parentheses is symmetric in i and l.
Due to the quasi-associativity (2.9) we obtain some terms of type ∂γ i bj bk :
1   1 lm  aj i 
Q1 =(− )∂(glm ) gaj Γiam bi bl + (− )g,j ∂ g Γam bi bl
2 ,j 2
1 lm

ia j
 1 lm  ia j 
+ (− )∂(g ) g Γam bi bl + (− )g,j ∂ g Γam bi bl
2 ,j 2
1   1 lm  ij a 
+ ∂(glm ) gij Γail ba bj + g,m ∂ g Γil ba bj . (6.9)
2 ,m 2

• coefficient γ i γ j : 0,
using that the metric is covariantly constant we get:
1 1 1
(− )∂[(glm Γnla Γamr ),n ]∂γ r + ∂ 2 (gij Γkil Γljk ) + (− )gij ∂ 2 (gij )
2 4 8
1 2 1 rm 1 m r
+ (− )∂ (gij )g + g gal Γrs Γim ∂γ ∂γ + Γrs Γmi ∂γ s ∂γ i
ij l a s i
8 2 2
=0.

Here we used the following three identities

1 1 1 1
0 = (− )∂ 2 (gij gij ) = (− )∂ 2 gij gij + (− )∂gij ∂gij + (− )gij ∂ 2 gij ,
8 8 4 8
1 1 1
∂gij ∂gij = (− )Γlis Γilr ∂γ s ∂γ r + (− )gaj g il Γais Γjlr ∂γ s ∂γ r ,
4 2 2
1 1 1
0 = (− )∂[(glm Γnla Γamr ),n ]∂γ r +(− )(g lm Γnla Γamr ),n ∂ 2 γ r + ∂ 2 (gij Γkil Γljk ).
2 2 4
The last identity follows taking the derivative of the equality:
1
0 = (−1)(gij Γkil Γlja ),k ∂γ a + ∂(gij Γkil Γljk ) (this is exactly the coefficient
2
of ∂γ i in the terms with λ in the subsection 6.4.2 above).

55
 Due to the quasi-associativity (2.9) we obtain some terms of type ∂ 2 γ i :
1
Q2 = (gij Γkil Γlja ),k ∂ 2 γ a . (6.10)
2

• coefficient ∂ 2 γ i : 0,
just using that the metric is covariantly constant and taking into account the
term (6.10) that comes from the coefficient of γ i γ j .

• coefficient ci bj bk bl : 0,

1 1
(− )glm gij (Γaik ),m ck ba bj bl + (− )gij gmn Γlik Γkma ca bj bl bn
2 2
1 lm sj i a k
+ g g Γsm Γik c ba bj bl
2
1 lm ij a 1
= g g R ikm ck ba bj bl + (− )g lm gij (Γaim ),k ck ba bj bl
2 2
1 lm ij a x k 1
+ (− )g g Γkx Γim c ba bj bl + glm gij Γamx Γxik ck ba bj bl
2 2
1 ij mn l k a 1
+ (− )g g Γik Γma c bj bl bn + g lm g sj Γism Γaik ck ba bj bl
2 2
1 ajl k
= − R k c ba bj bl
2
1 [ajl] k
=− R k c ba bj bl
2
=0.

Here [ ] denotes the anti-symmetrization of the indices, the last equality

[ajl]
follows from the Bianchi’s first identity (6.4) which implies that R k = 0.

• coefficient ∂bi bj : 0,
denote by A1 terms containing derivatives of the Christoffel symbols are:
1 lm ia j 1 1
A1 = g g (Γam ),j bi ∂bl + (− )g lm g ij (Γkij ),m bl ∂bk + g ij g am (Γlaj ),m bi ∂bl ,
2 2 2
denote by A2 the sum of terms that doesn’t contain derivatives of the
Christoffel symbols:
1 1 1
A2 = (− )Γiji Γj la ba ∂bl + (− )Γijl Γaij ba ∂bl + ( )Γija Γilj ba ∂bl .
2 2 2
Replacing the derivatives of the Christoffel symbol in A1 using (6.2) and

56
 (6.5) we get:
1 1
A1 = glm gia Ram bi ∂bl + glm gia Γxay Γymx bi ∂bl
2 2
1 lm ij k 1
+ g g R ijm bl ∂bk + (− )g lm g ij (Γkim ),j bl ∂bk
2 2
1 lm ij k s 1
+ (− )g g Γjs Γim bl ∂bk + glm gij Γkms Γsij bl ∂bk
2 2
1 ij am l 1 ij am l
+ g g R jam bi ∂bl + g g (Γjm ),a bi ∂bl
2 2
1 ij am l s 1
+ g g Γas Γjmbi ∂bl + (− )gij gam Γlms Γsja bi ∂bl
2 2
1 lm ia 1 lm ia x y
= g g Ram bi ∂bl + g g Γay Γmx bi ∂bl
2 2
1 lm ij k 1
+ g g R ijm bl ∂bk + glm gij Γkms Γsij bl ∂bk
2 2
1 ij am l 1
+ g g R jam bi ∂bl + (− )gij g am Γlms Γsja bi ∂bl ,
2 2
then
1 1 1
A1 + A2 = g lm g ia Ram bi ∂bl + glm gij Rkijm bl ∂bk + gij g am Rljam bi ∂bl
2 2 2
1 ij am l
= g g R jam bi ∂bl
2
=0,

using the symmetries (6.3).

• coefficient ∂γ i bj bk : 0,
due to the quasi-associativity in the computations of the bi bj β k coefficient
we need to take into account (6.9).
Collecting the terms that contain ∂γ i bj bk : denote by A1 the sum of the
terms that does not contain derivative of the Christoffel symbols, denote by
A2 the sum of the terms containing derivative of the Christoffel symbols. We
have:
1 lm aj i
A1 = g,j g,s Γam ∂γ s bi bl + glm g,m
ij k n
Γin Γja ∂γ a bk bl
2
+ (−1)g ij g mn Γlik Γkms Γsnr ∂γ r bj bl
1 j 1 lm xj i
= g lx gak Γm i s k n
xj Γks Γam ∂γ bi bl + (− )g g Γxm Γin Γja ∂γ bk bl
a
2 2
+ (−1)g lm gix Γjxm Γkin Γnja ∂γ a bk bl + (−1)g ij gmn Γlik Γkms Γsnr ∂γ r bj bl ,

57
 1
A2 =grm g ia Γlrj (Γjam ),s ∂γ s bi bl + (− )grm gij Γlrs (Γail ),m ∂γ s ba bj
2
1 lr ij m a
+ (− )g g Γrs (Γil ),m ∂γ ba bj + g lm g ij Γkin (Γnja ),m ∂γ a bk bl
s
2
+ g g Γja (Γkin ),m ∂γ a bk bl .
lm ij n

Substituting the derivative of the Christoffel symbols in the second, third and
fourth summand of A2 by the Riemann curvature tensor (6.2) we get:
1
A2 =g rm gia Γlrj (Γjam ),s ∂γ s bi bl + g rm gij Γlrs Ralim ∂γ s ba bj
2
1 rm ij l 1
+ (− )g g Γrs (Γalm ),i ∂γ s ba bj + (− )grm gij Γlrs Γaix Γxlm ∂γ s ba bj
2 2
1 rm ij l a x s 1 lr ij m a
+ g g Γrs Γmx Γli ∂γ ba bj + g g Γrs R lim ∂γ s ba bj
2 2
1 lr ij m a 1
+ (− )g g Γrs (Γlm ),i ∂γ s ba bj + (− )glr g ij Γm a x s
rs Γix Γlm ∂γ ba bj
2 2
1
+ glr gij Γm a x s lm ij k n
rs Γmx Γli ∂γ ba bj + (−1)g g Γin R jam ∂γ bk bl
a
2
+ g lm gij Γkin (Γnjm ),a ∂γ a bk bl + glm gij Γkin Γnax Γxjm ∂γ a bk bl
+ (−1)g lm gij Γkin Γnmx Γxja ∂γ a bk bl + glm gij Γnja (Γkin ),m ∂γ a bk bl
1
=(−1)g rm gij Γlrs Γaix Γxlm ∂γ s ba bj + (− )glm g ij Γkin Γnmx Γxja ∂γ a bk bl
2
1 lr ij m a x s
+ g g Γrs Γmx Γli ∂γ ba bj + g g Γin Γax Γjm ∂γ a bk bl .
lm ij k n x
2
Finally we check that A1 + A2 = 0.

We conclude that [G+ λ G− ] = 0.

6.5 [G+ λ Φ− ] and [G− λ Φ+ ]

Now we compute [G+ λ Φ− ], [G− λ Φ+ ] is computed similarly.

1 1 1 1 1
G+ = ci βi + ∂γ i bi + gij bi βj + g ij Γlik ck bj bl + g ij Γkij ∂bk
2 2 2 2 2
1 i j ij k l m
+ gij ∂γ c + g Γil Γjm ∂γ bk ,
2

i −i i
Φ− = √ ϕijk ci cj ck + √ ϕijk g il cj ck bl + √ ϕijk gil gjm ck bl bm
12 2 4 2 4 2
−i i
+ √ ϕijk g g g bl bm bn + √ ϕijk Γimn gjm ∂γ n gkl bl
il jm kn
12 2 2 2
i
− √ ϕijk Γimn gjm ∂γ n ck .
2 2

58
We list the non-zero λ-brackets between the summands of G+ and the summands
of Φ− , to compute these we used the Mathematica package [Thi91].

[ci βi λ ϕlmn cl cm cn ] =ϕlmn,i ci cl cm cn

= ϕamn Γail + ϕlan Γaim + ϕlma Γain ci cl cm cn

=0,

 
[ci βi λ ϕlmn cm cn bl ] = ϕlmn,l cm cn λ + ϕlmn cm cn βl
 
+ ϕlmn,i ci cm cn bl + ∂ ϕlmn,l cm cn ,

 
[ci βi λ ϕlmn cn bl bm ] =2 ϕlmn,m cn bl λ + 2ϕlmn cn bl βm
 
+ ϕlmn,i ci cn bl bm + 2∂ ϕlmn,m cn bl ,

 
[ci βi λ ϕlmn bl bm bn ] =3 ϕlmn,n bl bm λ + 3ϕlmn bl bm βn
 
+ ϕlmn,i ci bl bm bn + 3∂ ϕlmn,n bl bm ,

   
i
[c βi λ ϕlmn Γlms ∂γ s bn ] = ϕlmn Γlms cs bn mn l
+ ϕl Γms s
∂γ λ
,n
 
+ ϕlmn Γlms ∂cs bn + ϕlmn Γlms ∂γ s βn
   
+ ϕlmn Γlms ∂γ s ci bn + ∂[ ϕlmn Γlms ]∂γ s ,
,i ,n

 
[ci βi λ ϕlmn Γlms ∂γ s cn ] = ϕlmn Γlms cs cn λ + ϕlmn Γlms ∂cs cn
 
+ ϕlmn Γlms ∂γ s ci cn ,
,i

[∂γ i bi λ ϕlmn cl cm cn ] =3ϕlmn ∂γ n cl cm ,

[∂γ i bi λ ϕlmn cm cn bl ] =2ϕlmn ∂γ m cn bl ,

[∂γ i bi λ ϕlmn cn bl bm ] =ϕlmn ∂γ n bl bm ,

[∂γ i bi λ ϕlmn Γlms ∂γ s cn ] =ϕlmn Γlms ∂γ n ∂γ s ,

59
    
[g ij bi βj λ ϕlmn cl cm cn ] =3 gni ϕlmn,i cl cm λ + 3 gni ϕlmn cl cm βi
+ (−1)gij ϕlmn,j cl cm cn bi + 3g ni ∂[ϕlmn,i ]cl cm
+ 6∂(gni )ϕlmn,i cl cm + 3∂ ϕlmn g,ini cl cm ,

   
[g ij bi βj λ ϕlmn cm cn bl ] =(−2) gni ϕlmn,i cm bl λ + 2 g mi ϕlmn cn bl βi
+ gij ϕlmn,j cm cn bi bl + 2gmi ∂[ϕlmn,i ]cn bl
 
+ 4∂(gmi )ϕlmn,i cn bl + 2∂ ϕlmn g,imi cn bl ,

   
[g ij bi βj λ ϕlmn cn bl bm ] = gni ϕlmn,i bl bm λ + gni ϕlmn bl bm βi
 
+ (−1)gij ϕlmn,j cn bi bl bm + gni ∂ ϕlmn,i bl bm
   
+ 2∂ gni ϕlmn,i bl bm + ∂ ϕlmn g,ini bl bm ,

[gij bi βj λ ϕlmn bl bm bn ] =g ij ϕlmn,j bi bl bm bn ,

 
[gij bi βj λ ϕlmn Γlms ∂γ s bn ] = gij ϕlmn Γlmj bi bn λ + gij ϕlmn Γlmj ∂bi bn
 
+ gij ϕlmn Γlms ∂γ s bi bn + ∂(gij )ϕlmn Γlmj bi bn ,
,j

[gij bi βj λ ϕlmn Γlms ∂γ s cn ]


= (−1)gij ϕlmn Γlmj cn bi + gnj (ϕlmn Γlms ),j ∂γ s
  
+∂(g )ϕl n Γms λ + (−1)g ϕl n Γms c ∂bi + g ϕl n Γms ∂γ s βj
ns m l is m l n nj m l

   
+ (−1)g ij ϕlmn Γlms ∂γ s cn bi + gnj ∂[ ϕlmn Γlms ]∂γ s
,j ,j
 
+ (−1)∂(gis )ϕlmn Γlms cn bi + 2∂(gnj ) ϕlmn Γlms ∂γ s
,j
1 2 ns m l nj
 
+ ∂ (g )ϕl n Γms + g,j ∂ ϕlmn Γlms ∂γ s ,
2

 
[g ij Γrik ck bj br λ ϕlmn cl cm cn ] = 6g in Γm ϕ
ik lmn ck l
c λ + 3gij Γnik ϕlmn ck cl cm bj
+ 3g im Γrik ϕlmn ck cl cn br + 6g in Γm k l
ik ϕlmn ∂c c
 
+ 6∂ g in Γm k l
ik ϕlmn c c ,

60
 [gij Γrik ck bj br λ ϕlmn cm cn bl ]

= 2gin Γm l k ij n l
ik ϕ mn c bl + 2g Γil ϕ mn c bj
m

  
+ 2g Γil ϕ mn c br + 2∂ g Γil ϕ mn λ + 2gij Γnik ϕlmn ck cm bj bl
im r l n in m l

+ 2g in Γrik ϕlmn ck cm bl br + gij Γril ϕlmn cm cn bj br + 2gij Γnil ϕlmn cm ∂bj

 
+ 2g im Γril ϕlmn cn ∂br + 2gin Γm ϕ
ik mn
l
∂c k
bl + 2∂ g in m
Γ l
ik ϕ mn c bl
k

     
+ 2∂ gij Γnil ϕlmn cm bj + 2∂ gim Γril ϕlmn cn br + ∂ 2 gin Γm l
il ϕ mn ,

[gij Γrik ck bj br λ ϕlmn cn bl bm ]

 
= 2gij Γnim ϕlmn bj bl + 2gin Γrim ϕlmn bl br λ + 2gin Γrim ϕlmn bl ∂br
+ gij Γnik ϕlmn ck bj bl bm + (−1)gin Γrik ϕlmn ck bl bm br
 
+ 2gij Γril ϕlmn cn bj bm br + 2gij Γnim ϕlmn ∂bj bl + 2∂ gij Γnim ϕlmn bj bl
 
+ 2∂ gin Γrim ϕlmn bl br ,

[gij Γrik ck bj br λ ϕlmn bl bm bn ] =3g ij Γrin ϕlmn bj bl bm br ,

[gij Γrik ck bj br λ ϕlmn Γlms ∂γ s bn ] =gij Γrin Γlms ϕlmn ∂γ s bj br ,

[g ij Γrik ck bj br λ ϕlmn Γlms ∂γ s cn ]

=g ij Γnik Γlms ϕlmn ∂γ s ck bj + (−1)gin Γrik Γlms ϕlmn ∂γ s ck br ,

   
[gij Γkij ∂bk λ ϕlmn cl cm cn ] = 3g ij Γm ϕ
ij lmn cl n
c λ + 3∂ g ij m
Γ l n
ij ϕlmn c c ,

   
[gij Γkij ∂bk λ ϕlmn cm cn bl ] = 2gij Γnij ϕlmn cm bl λ + 2∂ g ij Γnij ϕlmn cm bl ,

[gij Γkij ∂bk λ ϕlmn cn bl bm ]

 
= (−1)g ij Γnij ϕlmn bl bm λ + (−1)∂(gij Γnij )ϕlmn bl bm ,

[gij Γkij ∂bk λ ϕlmn Γlms ∂γ s cn ]

   
= (−1)g ij Γnij ϕlmn Γlms ∂γ s λ + (−1)∂ gij Γnij ϕlmn Γlms ∂γ s ,

61
 [gij ∂γ i cj λ ϕlmn cm cn bl ] =gil ϕlmn ∂γ i cm cn ,

[gij ∂γ i cj λ ϕlmn cn bl bm ] =2gim ϕlmn ∂γ i cn bl ,

[gij ∂γ i cj λ ϕlmn bl bm bn ] =3gin ϕlmn ∂γ i bl bm ,

[gij ∂γ i cj λ ϕlmn Γlms ∂γ s bn ] =gin ϕlmn Γlms ∂γ i ∂γ s ,

[gij Γkia Γajr ∂γ r bk λ ϕlmn cl cm cn ] =3gij Γnia Γajr ϕlmn ∂γ r cl cm ,

[gij Γkia Γajr ∂γ r bk λ ϕlmn cm cn bl ] =(−2)gij Γnia Γajr ϕlmn ∂γ r cm bl ,

[g ij Γkia Γajr ∂γ r bk λ ϕlmn cn bl bm ] =g ij Γnia Γajr ϕlmn ∂γ r bl bm ,

[gij Γkia Γajr ∂γ r bk λ ϕlmn Γlms ∂γ s cn ] =g ij Γnia Γajr ϕlmn Γlms ∂γ r ∂γ s .

6.5.1 Coefficient of λ
We prove that it is zero after an easy computation using only that the metric is
covariantly constant and that dϕ = 0, (4.3).

6.5.2 Coefficient of λ0
We compute the coefficient of each type of term that appears:

• Coefficient of ci cj βk : 0.

• Coefficient of ci cj ck bl : 0, we use that dϕ = 0.

• Coefficient of ∂γ s ci cj :

After some simplifications using that dϕ = 0 and ∇g = 0 we arrived at the

following expression:
i 
(− √ ) gij Γklm Γljs ϕkin ∂γ s cm cn + (−1)g ij Γksl Γljmϕkin ∂γ s cm cn
4 2
    
ij k s m n ij k s m n
+g Γjs ϕkin ∂γ c c + (−1)g Γjm ϕkin ∂γ c c
,m ,s
i
=(− √ )gij ϕkin Rkjms ∂γ s cm cn
4 2
=0,
the last equality is zero by Lemma 4.5.

62
• Coefficient of ci bj βk : 0.

• Coefficient of ci cj bl bm : 0, we use that dϕ = 0.

• Coefficient of ∂γ s ci bj :
After some simplifications using that dϕ = 0 and ∇g = 0 we arrived at the
following expression:

i    
√ ϕijk Γijs ∂γ s cl bk + (−1)ϕijk Γijl ∂γ s cl bk
4 2 ,l ,s

+ϕijk Γila Γajs ∂γ s cl bk + (−1)ϕijk Γisa Γajl ∂γ s cl bk

i    
+ √ g ij ϕlmn Γlms ∂γ s cn bi + (−1)g ij ϕlmn Γlmj ∂γ s cn bi
4 2 ,j ,s

ij m l a s n ij m l a s n
+g ϕl n ΓjaΓms ∂γ c bi + (−1)g ϕl n Γsa Γmj ∂γ c bi
i i
= √ ϕijk Rijls ∂γ s cl bk + √ gij ϕlmn Rlmjs ∂γ s cn bi
4 2 4 2
=0,

the last equality is zero by Lemma 4.5.

• Coefficient of bi bj βk : 0.

• Coefficient of cl bi bj bk : 0, we use that dϕ = 0.

• Coefficient of ∂γ s bi bj :
After some simplifications using that dϕ = 0 and ∇g = 0 we arrived at the
following expression:

i    
√ gij ϕlmn Γlms ∂γ s bi bn + (−1)g ij ϕlmn Γlmj ∂γ s bi bn
2 2 ,j ,s

ij mn l a s ij mn l a s
+g ϕl Γja Γms ∂γ bi bn + (−1)g ϕl Γsa Γmj ∂γ bi bn
i
= √ gij ϕlmn Rlmjs ∂γ s bi bn
2 2
=0,

the last equality is zero by Lemma 4.5.

• Coefficient of ci ∂bj : 0.

• Coefficient of ∂ci bj : 0.

63
 • Coefficient of ∂ci cj : 0.

• Coefficient of bi bj bk bl : 0, we use that dϕ = 0.

• Coefficient of ∂bi bj : 0.

• Coefficient of ∂γ s βi : 0.

• Coefficient of ∂γ s ∂γ r :
After some simplifications using that dϕ = 0 and ∇g = 0 we arrived at the
following expression:

(−i)    
√ gij Γkir Γlkm ϕmjl ∂γ s ∂γ r + (−1)gij Γkir Γlms ϕmjl ∂γ s ∂γ r
2 2 ,s ,k
    
ij k l a m s r ij k l a m s r
+g Γir Γsa Γmk ϕ jl ∂γ ∂γ + (−1)g Γir Γka Γms ϕ jl ∂γ ∂γ

(−i) 
+ √ gij Γljs (Γm n r s ij l m n
in ),r ϕ lm ∂γ ∂γ + (−1)g Γjs (Γnr ),i ϕ lm ∂γ ∂γ
r s
2 2

+gij Γljs (Γm Γ a
ra ni ) ϕn
lm ∂γ r
∂γ s
+ (−1)g ij l
Γ js (Γ m a
Γ
ia nr ) ϕn
lm ∂γ r
∂γ s

(−i) (−i)
= √ gij Γkir ϕmjl Rlmsk ∂γ s ∂γ r + √ g ij Γljs ϕnlm Rmnri ∂γ r ∂γ s
2 2 2 2
=0,

the last equality is zero by Lemma 4.5.

Then we conclude that [G+ λ Φ− ] = 0, similarly we obtain [G− λ Φ+ ] = 0.

6.6 [G+ λ Φ+ ]
G = ci βi + ∂γ i bi ,
1 1 1
Φ+ = √ ϕijk ci cj ck + √ ϕijk g il cj ck bl + √ ϕijk gil gjm ck bl bm
12 2 4 2 4 2
1 1
+ √ ϕijk g il gjm gkn bl bm bn + √ ϕijk Γimn gjm ∂γ n gkl bl
12 2 2 2
1
+ √ ϕijk Γimn gjm ∂γ n ck .
2 2
We can use the λ-brackets performed in section 6.5.

6.6.1 Coefficient of λ2
We realize inmediatly that the coefficient of λ2 is zero because there are not λ2
terms.

64
6.6.2 Coefficient of λ
• Coefficient of ci cj : 0, we use that ∇ϕ = 0.

• Coefficient of ci bj : 0, we use that ∇ϕ = 0.

• Coefficient of bi bj :
1
√ ϕlmn,n bl bm = 0.
4 2
As ∇ϕ = 0 (4.3) we have,

ϕlmn,n = −ϕamn Γlan − ϕlan Γm

an − ϕ
lma n
Γan = 0,

to conclude that the last equality is zero we used the symmetries of ϕ and
(6.1).

• Coefficient of γ s :
1  
√ ϕlmn Γlms ∂γ s = 0.
2 2 ,n
As ∇ϕ = 0 (4.3) we have
   
ϕlmn Γlms =ϕlmn,n Γlms + ϕlmn Γlms
,n ,n
 
ϕamn Γaln − ϕlma Γnan Γlms + ϕlmn Γlms
ϕlan Γm


= − an
,n
 
=ϕamn Γaln Γlms + ϕlmn Γlms
,n
   
mn l mn
=ϕl Γms + (−1)ϕl Γlmn + ϕlmn Γlna Γams
,n ,s
+ (−1)ϕl Γsa Γamn
mn l

=ϕlmn Rlmns
=0.

That the last equality is zero follows by Lemma 4.5.

Then [G+ λ φ+ ] = K+ and similarly [G− λ φ− ] = K− .

6.7 [L+ λ Φ+ ]

1 1
L =∂γ i βi − ci ∂bi + ∂ci bi ,
2 2

65
 1 1
Φ+ = √ ϕijk ci cj ck + √ ϕijk g il cj ck bl
12 2 4 2
1 1
+ √ ϕijk gil g jm ck bl bm + √ ϕijk g il gjm gkn bl bm bn
4 2 12 2
1 1
+ √ ϕijk Γimn g jm ∂γ n gkl bl + √ ϕijk Γimn gjm ∂γ n ck .
2 2 2 2
The only brackets that can produce λ2 are:
 
[ci ∂biλ ϕlmn cm cn bl ] = 2ϕiin cn λ2 + terms linear in λ,

 
[ci ∂biλ ϕlmn cn bl bm ] = 2ϕlii bl λ2 + terms linear in λ,

 
[∂ci biλ ϕlmn cm cn bl ] = 2ϕiin cn λ2 + terms linear in λ,

 
[∂ci biλ ϕlmn cn bl bm ] = 2ϕlii bl λ2 + terms linear in λ.

Then the coefficient of λ2 is zero. We conclude that:

[L+ λ φ+ ] = (∂ + 23 λ)φ+ , similarly we obtain [L− λ φ− ] = (∂ + 32 λ)φ− .

References
[Blu92] R Blumenhagen, Covariant construction of N = 1 super W-algebras, Nuclear Physics
B 381 (1992), no. 3, 641–669. ↑4, 11
[Bor86] Richard E. Borcherds, Vertex algebras, Kac-Moody algebras, and the Monster, Proc. Nat.
Acad. Sci. U.S.A. 83 (1986), no. 10, 3068–3071. ↑7
[BZHS08] David Ben-Zvi, Reimundo Heluani, and Matthew Szczesny, Supersymmetry of the chiral
de Rham complex, Compositio Mathematica 144 (2008), 503–521. ↑3, 5, 14
[CHNP15] Alessio Corti, Mark Haskins, Johannes Nordström, and Tommaso Pacini, G2 -manifolds
and associative submanifolds via semi-Fano 3-folds, Duke Math. J. 164 (2015), no. 10,
1971–2092. MR3369307 ↑5
[dBNS08] Jan de Boer, Asad Naqvi, and Assaf Shomer, The topological G2 string, Adv. Theor.
Math. Phys. 12 (2008), no. 2, 243–318. MR2399312 ↑5
[DK98] A. D’Andrea and V. G. Kac, Structure theory of finite conformal algebras, Selecta Math-
ematica 4 (1998), no. 3, 377–418. ↑4
[DSK05] Alberto De Sole and Victor G. Kac, Freely generated vertex algebras and non-linear Lie
conformal algebras, Comm. Math. Phys. 254 (2005), no. 3, 659–694. ↑9
[DSK06] , Finite vs affine W-algebras, Japanese Journal of Mathematics 1 (2006), no. 1,
137–261. ↑6
[EHKZ13] Joel Ekstrand, Reimundo Heluani, Johan Källén, and Maxim Zabzine, Chiral de Rham
Complex on Riemannian Manifolds and Special Holonomy, Communications in Mathe-
matical Physics 318 (2013), no. 3, 575–613. ↑3, 4, 5, 13, 14, 15, 16, 19, 24

66
 [FO97] José M. Figueroa-O’Farrill, A note on the extended superconformal algebras associated
with manifolds of exceptional holonomy, Phys. Lett. B392 (1997), 77–84. ↑12, 13
[FOS91] José M. Figueroa-O’Farrill and Stany Schrans, Extended superconformal algebras,
Phys.Lett. B257 (1991), 69–73. ↑13
[FOS92] , The Conformal bootstrap and super W algebras, Int.J.Mod.Phys. A7 (1992),
591–618. ↑13
[Hel09] Reimundo Heluani, Supersymmetry of the chiral de Rham complex 2: Commuting sec-
tors, International Mathematics Research Notices 2009 (2009), 953–987. ↑3, 5
[HP91] P.S. Howe and G. Papadopoulos, A note on holonomy groups and sigma models, Physics
Letters B 263 (1991), no. 2, 230 –232. ↑3
[HP93] P. S. Howe and G. Papadopoulos, Holonomy groups and W -symmetries, Comm. Math.
Phys. 151 (1993), no. 3, 467–479. MR1207260 ↑3, 11, 13
[HRD15] Reimundo Heluani and Lázaro O. Rodríguez Díaz, The Shatashvili-Vafa G2 supercon-
formal algebra as a quantum Hamiltonian reduction of D(2, 1; α), Bull. Braz. Math.
Soc. (N.S.) 46 (2015), no. 3, 331–351. ↑12
[Joy07] Dominic D Joyce, Riemannian holonomy groups and calibrated geometry, Oxford Grad-
uate Texts in Mathematics, Oxford Univ. Press, UK, 2007. ↑16
[Kac98] Victor Kac, Vertex operator algebras for beginners, University Lecture Series 10 (1998).
↑6, 10
[Kar09] Spiro Karigiannis, Flows of G2 Structures, I, The Quarterly Journal of Mathematics 60
(2009), no. 4, 487–522. ↑18
[KW04] Victor G. Kac and Minoru Wakimoto, Quantum reduction and representation theory of
superconformal algebras, Advances in Mathematics 185 (2004), no. 2, 400–458. ↑10
[MSV99] Fyodor Malikov, Vadim Schechtman, and Arkady Vaintrob, Chiral de Rham Complex,
Communications in Mathematical Physics 204 (1999), no. 2, 439–473. ↑3, 5, 13, 14
[Oda89] Satoru Odake, Extension of N = 2 Superconformal Algebra and Calabi-Yau Compacti-
fication, Mod.Phys.Lett. A4 (1989), 557. ↑3, 16
[Pee07] Kasper Peeters, Cadabra: a field-theory motivated symbolic computer algebra system.,
Comput. Phys. Commun. 176 (2007), no. 8, 550–558 (English). ↑4, 32
[Sal89] Simon Salamon, Riemannian geometry and holonomy groups, Pitman research notes in
mathematics series, Longman Scientific & Technical New York, 1989. ↑16, 17
[SV95] S.L. Shatashvili and C. Vafa, Superstrings and manifolds of exceptional holonomy, Se-
lecta Mathematica 1 (1995), no. 2, 347–381. ↑3, 5, 11, 12, 13
[Thi91] K. Thielemans, A Mathematica package for computing operator product expansions, Int.
J. Mod. Phys. C2 (1991), 787–798. ↑4, 32, 40, 48, 59

67