Vertex Operator Algebras, Modular Forms and

Moonshine
Geoffrey Mason
University of California, Santa Cruz

Contents
Lecture 1.
1. The Monster simple group
2. J and V \
3. Monstrous Moonshine
4. Vertex algebras
5. Locality and quantum fields
6. CFT axioms
Lecture 2.
7. Lie algebras and local fields
8. Vertex operator algebras
9. Super vertex operator algebras
Lecture 3.
10. Modules over a VOA
11. Lattice theories
12. Partition functions and modular-invariance
Lecture 4.
13. The Lie algebra on V1
14. Automorphisms
15. Twisted sectors
16. The Moonshine module
17. AutV \
References.

1
 Lecture 1
§1 The Monster simple group
Group theorists conceived the Monster sporadic simple group M in the early
1970s, although it was not officially born until 1982. Many features of M
were understood well before that time, however. In particular the complete
character table was already known. Here is a small part of it.

1 2A 2B
χ1 1 1 1
χ2 196883 4371 275
χ3 21296876 91884 − 2324

Let Vi be the M -module that affords the character χi . From the character
table one can compute branching rules Vi ⊗ Vj = ⊕k cijk Vk . In particular,
the tensor square V2⊗2 decomposes into the sum of symmetric and exterior
squares S 2 (V2 ) ⊕ Λ2 (V2 ), and the branching rules show that c222 = 1 with
V2 ⊆ S 2 (V2 ). So there is a canonical M -invariant surjection V2 ⊗ V2 → V2 ,
and it gives rise to a commutative, nonassociatve algebra structure on V1
whose automorphism group contains M . We can formally add an identity
element 1 to obtain a unital, commutative algebra

B = V1 ⊕ V2 (1)

(V1 = C1) with M ⊆ Aut(B).

§2 J and V \
Up to an undetermined constant, there is a unique modular function of weight
0 on the full modular group Γ := SL2 (Z) with a simple pole of residue 1 at
∞. Such functions can be represented as quotients of holomorphic modular
forms of equal weight. For example we have

θE8 (τ )3
J + 744 = = q −1 + 744 + 196884q + 21493760q 2 + . . . (2)
∆(τ )
θΛ (τ )
J + 24 = = q −1 + 24 + 196884q + 21493760q 2 + . . . . (3)
∆(τ )

2
Here,

J = q −1 + 196884q + 21493760q 2 + . . . (4)

is the modular function with constant 0, ∆(τ ) is the discriminant

∞
Y
∆(τ ) = q (1 − q n )24 , (5)
n=1

X
θL (τ ) = q (α,α)/2 (6)
α∈L

is the theta function of an even lattice L, and E8 , Λ denote the E8 root lattice
and Leech lattice respectively.
John McKay noticed that the first few Fourier coefficients in (4) are sim-
ple linear combinations of dimensions of the irreducible M -modules Vi with
nonnegative coefficients. This suggests that we replace the coefficients by
the putative M -modules that correspond to them - a process that sometimes
goes by the abysmal name of ‘categorification’. From (3)-(6), the coefficients
of J are all nonnegative, so at least they correspond to linear spaces. Shifting
the grading by 1 for later convenience, we obtain a Z-graded linear space

V \ := V0\ ⊕ V2\ ⊕ V3\ ⊕ . . . (7)

with

V0\ = V1
V2\ = V1 ⊕ V2 (8)
V3\ = V1 ⊕ V2 ⊕ V3
...

and with dim Vn\ = coefficient of q n−1 in (4). McKay’s observation was pro-
moted to the conjecture that each Vn\ carries a ‘natural’ action of M . Note
that V2\ is identified with the algebra B.

3
§3 Monstrous Moonshine
With the conjectured Z-graded M -module V \ in hand, for each g ∈ M we
can take the graded trace of g and obtain another q-series
∞
X
−1
Zg = Zg (q) := q TrVn\ (g)q n . (9)
n=0

It was John Thompson who first asked what one can say about these
additional q-expansions. (There are 174 of them, one for each conjugacy
class of M .) We have Z1A (1, q) = J by construction, and from the character
table and (8) we see that

Z2A (q) = q −1 + 4372q + 96256q 2 + . . .

Z2B (q) = q −1 + 276q − 2048q 2 + . . .

In a celebrated paper, John Conway and Simon Norton resoundingly an-

swered Thompson’s question. They gave overwhelming evidence for the con-
jecture that each of the trace functions (9) was a hauptmodul for a subgroup
of SL2 (Q) commensurable with SL2 (Z). This means that for each g we have
a subgroup Γg ⊆ SL2 (Q) with |Γg : Γg ∩ Γ|, |Γ : Γg ∩ Γ| < ∞ such that the
following hold:

(i) Each Zg is the q-expansion of a modular function of weight zero

on Γg . (10)
(ii) If H is the complex upper half-plane, the compact Riemann surface
Γg \ H∗ is a Riemann sphere whose function field is C(Zg ).

If g = 1A then of course Z1 = J and Γ1 = Γ. Conway-Norton proposed

formulae for each Zg . For example,

η(τ )24
Z2B (q) = + 24, (11)
η(2τ )24
where η(τ ) is the Dedekind eta-function
∞
Y
1/24
η(τ ) = q (1 − q n ). (12)
n=1

Z2B (q) is a hauptmodul for the index 3 subgroup Γ0 (2) ⊆ Γ.

4
§4 Vertex algebras
The problem is now to define a natural action of the Monster M on V \ (59)
so that the graded traces Zg satisfy the Conway-Norton moonshine conjec-
tures (10). Borcherds’ radical proposed solution involved the idea of a vertex
algebra, which may be defined as follows. It is a pair (V, 1) consisting of a
nonzero C-linear space V and a distinguished vector 1 6= 0. Moreover, V is
equipped with bilinear products
µn : V ⊗ V → V (n ∈ Z),
u ⊗ v 7→ u(n)v (u, v ∈ V ),
satisfying the following axioms for all u, v, w ∈ V :
1) There is n0 = n0 (u, v) ∈ Z such that u(n)v = 0 for n ≥ n0 , (13)
2) v(n)1 = 0 (n ≥ 0) and v(−1)1 = v, (14)
3) For all p, q, r ∈ Z we have
∞  
X p
{u(r + i)v}(p + q − i)w = (15)
i=0
i
∞  
i r
X
(−1) {u(p + r − i)v(q + i)w − (−1)r v(q + r − i)u(p + i)w}.
i=0
i
Thanks to 1), both sums in 3) are finite so that the identity in question is
sensible.
At this point the reader may well be asking, where did these identities
come from, what are they good for, and what do they have to do with Mon-
strous Moonshine? The point of these lectures is to address these questions.
We begin by specializing (15) in various ways. It is convenient to consider
u(n) ∈ End(V ) to be the linear operator such that v 7→ u(n)v (v ∈ V ).
Taking r = 0, the binomial ri vanishes unless i = 0 and (15) reduces to the

operator identity
∞  
X p
[u(p), v(q)] = {u(i)v}(p + q − i), (16)
i=0
i
called the commutator formula. Similarly, taking p = 0 yields the associativ-
ity formula
∞  
i r
X
{u(r)v}(q) = (−1) {u(r − i)v(q + i) − (−1)r v(q + r − i)u(i)}. (17)
i=0
i

5
With n0 (u, v) as in (13), we obtain
∞  
X
i r
(−1) {u(p + r − i)v(q + i) − (−1)r v(q + r − i)u(p + i)} = 0 (18)
i=0
i

whenever r ≥ n0 , which is sometimes referred to as commutivity.

Assuming (13), it is not too hard to show that (15) is a consequence of
the commutator and associativity formulas, and thus is equivalent to them.
There are other equivalent ways to reformulate (15) that are useful. We
explain one of them (cf. (24)) in the next Section.

§5 Locality and Quantum fields

In the succeeding two Sections we will explain how the idea of a vertex algebra
corresponds to the physicist’s conformal field theory.
The important idea of a vertex operator, or quantum field, or simply field,
defined on an arbitrary linear space V is as follows. It is a formal series
X
a(z) := an z −n−1 ∈ End(V )[[z, z −1 ]]
n∈Z

of operators an on V such that if v ∈ V then an v = 0 for all large enough n.

Set

F(V ) = {a(z) ∈ End(V )[[z, z −1 ]] | a(z) is a field}.

F(V ) is a linear subspace of End(V )[[z, z −1 ]].

If (V, 1) is a vertex algebra, we set
X
Y (u, z) := u(n)z −n−1 (u ∈ V ),
n∈Z

where we are using notation introduced in the previous Section. By con-

struction,

{Y (u, z) | u ∈ V } ⊆ F(V ), (19)

and we can think of Y as a linear map

Y : V → F(V ), u 7→ Y (u, z). (20)

6
 We use obvious notation when manipulating fields, eg.,
X
Y (u, z)v := {u(n)v}z −n−1 ∈ V [[z]][z, z −1 ].
n

In this language, (14) reads

X
Y (u, z)1 = u + {u(n)1}z −n−1 .
n≤−2

In particular, it follows that the Y map (20) is injective.

A pair of fields a(z), b(z) ∈ F(V ) are called mutually local if

(z1 − z2 )k [a(z1 ), b(z2 )] = 0 (some integer k ≥ 0). (21)

This means that the (operator) coefficients of each monomial z1p z2q in the
following identity coincide:

(z1 − z2 )k a(z1 )b(z2 ) − (z1 − z2 )k b(z2 )a(z1 ) = 0. (22)

Indeed,

(z1 − z2 )k a(z1 )b(z2 )

k  
i k
X X X
= (−1) z1k−i z2i am z1−m−1 bn z2−n−1
i=0
i m n
(   )
i k
XX X X
= (−1) am bn z1−p−1 z2−q−1
p q k−i−m=−p i−n=−q
i
( k   )
XX X k
= (−1)i ap+k−i bq+i z1−p−1 z2−q−1 .
p q i=0
i

Therefore also

(z1 − z2 )k b(z2 )a(z1 ) = (−1)k (z2 − z1 )k b(z2 )a(z1 )

( k   )
XX X k
= (−1)k (−1)i bq+k−i ap+i z1−p−1 z2−q−1 ,
p q i=0
i

7
whence locality (21), (22) holds if, and only if, for all integers p, q, and some
nonnegative integer k we have
k  
X k 
i
(−1) ap+k−i bq+i − (−1)k bq+k−i ap+i = 0.
i=0
i

The last display is identical with the commutivity formula (18) if we take
r = k. Because (18) holds for all r ≥ n0 , it certainly holds for some positive
integer k in place of r. Combining this with (19), we have established

If (V, 1) is a vertex algebra then any two vertex operators

Y (u, z1 ), Y (v, z2 ) (u, v ∈ V ) are mutually local fields. (23)

In a similar vein, let δ(z) := n∈Z z n be the formal delta-function, and

P
consider the identity
   
−1 z1 − z2 −1 z2 − z1
z0 δ Y (u, z1 )Y (v, z2 ) − z0 δ Y (v, z2 )Y (u, z1 )
z0 −z0
 
−1 z1 − z0
= z2 δ Y (Y (u, z0 )v, z2 ). (24)
z2
Here, the delta-functions are expanded as power series in the second variable
in the numerator, e.g.,
z1 − z2 X
δ( )= z0−n (z1 − z2 )n
z0 n∈Z
∞
!n
X X X
= (z1 /z0 )n (1 − z2 /z1 )n + (z0 /z1 )n (z2 /z1 )i .
n=0 n>0 i≥0

With this convention, identifying the operator coefficients for each monomial
z0p z1q z2r on the lhs and rhs of (24) yields exactly the identity (15).

§6 CFT axioms
(23) is the ‘main’ axiom for (2-dimensional) conformal field theory (CFT). We
now discuss the other axioms. Let (V, 1) be a vertex algebra, and introduce
the endomorphism

D : V → V, u 7→ u(−2)1. (25)

8
Using (14) and associativity (17) with q = 2, we have
∞  
X n i
{u(n)v}(−2)1 = (−1) {u(n − i)v(−2 + i)1}
i=0
i
= u(n)v(−2)1 − nu(n − 1)v(−1)1
= u(n)v(−2)1 − nu(n − 1)v. (26)

Therefore,
X
[D, Y (u, z)]v = {Du(n)v − u(n)Dv}z −n−1
n
X
= {(u(n)v)(−2)1 − u(n)v(−2)1}z −n−1
n
X
= {−nu(n − 1)v}z −n−1
n
X
= {(−n − 1)u(n)v}z −n−2
n
d
= Y (u, z)v,
dz
where d/dz is the formal derivative. Hence, we obtain
d
[D, Y (u, z)] = Y (u, z).
dz
If we take u = v = w = 1 and p = q = r = −1 in (15) we find that
1(−2)1 = 1(−2)1 + 1(−2)1. Thus 1(−2)1 = 0, that is D1 = 0.
We have arrived at the following set-up: a quadruple (V, Y, 1, D) consist-
ing of a linear space V , a distinguished nonzero vector 1 ∈ V , an endomor-
phism D : V → V with D1 = 0, and a linear injection Y : V 7→ F(V ),
satisfying the following for all u, v ∈ V :

Locality: Y (u, z1 ), Y (v, z2 ) are mutually local fields,

Creativity: Y (u, z)1 = u + O(z), (27)
Translation covariance: [D, Y (u, z)] = d/dzY (u, z).

The axioms (27) amount to a mathematical formulation of 2-dimensional

CFT, and we have shown that a vertex algebra (V, 1) naturally defines a

9
CFT. Conversely if (V, Y, 1, D) is a CFT then it can be shown that (V, 1) is
a vertex algebra. Basically, this means that the full strength of (15) can be
recovered (27).
The nomenclature in (27) is fairly standard in the physical literature, and
we use it in what follows. In addition, 1 is the vacuum vector, V is a Fock
space, elements in V are states, Y is the state-field correspondence, u(n) is
the nth mode of Y (u, z). Creativity is interpreted to mean that the state u
is created from the vacuum by the field Y (u, z) corresponding to u.
There are several other useful identities that follow without difficulty from
our axiomatic set-up. Among them we mention the following.
Y (1, z) = IdV , (28)
∞ n
X D u n
Y (u, z)1 = ezD u = z ,
n=0
n!
∞
X (−D)i
u(n)v = (−1)n+1 v(n + i)u. (29)
i=0
i!

(29) is called skew-symmetry.

Lecture 2
§7 Lie algebras and local fields
Certain infinite-dimensional Lie algebras naturally give rise to mutually
local fields. In this Section we discuss some important examples that illus-
trate some of the ideas developed so far.
1. Affine algebras.
Let L be a (complex) Lie algebra with bracket [a, b] (a, b ∈ L), equipped with
a symmetric, invariant, bilinear form h , i : L ⊗ L → C. (Invariant means
that h[a, b], ci = ha, [b, c]i for a, b, c ∈ L.) The associated affine Lie algebra is
b := L ⊗ C[t, t−1 ] ⊕ CK with central element K and bracket
L
[a ⊗ tm , b ⊗ tn ] = [a, b] ⊗ tm+n + mδm+n,0 ha, biK.
There is a triangular decomposition
L b− ⊕ L
b=L b0 ⊕ L
b+

10
with
b− := {a ⊗ tm | m < 0}, L
L b+ := {a ⊗ tm | m > 0}, L
b0 := {a ⊗ t0 } ⊕ CK.

Let W be a (left) L-module. Extend W to a L b+ ⊕ L b0 -module by letting

b+ annihilate W ; K act as a scalar l called the level. The induced module
L

V = V (l, W ) := IndLLb+ ⊕Lb0 (W ) ∼ b− ) ⊗ W

= S(L (30)
b

is a left L-module
b affording the representation π, say. (The linear isomor-
phism in (30) comes from the Poincaré-Birkhoff-Witt theorem.) A typical
vector in V is a sum of vectors that look like
(b1 ⊗ tn1 ) . . . (bk ⊗ tnk ) ⊗ w (bi ∈ L, w ∈ W, n1 ≤ . . . ≤ nk ≤ −1),
and
π(a ⊗ tn ){(b1 ⊗ tn1 ) . . . (bk ⊗ tnk ) ⊗ w} =
(a ⊗ tn )(b1 ⊗ tn1 ) . . . (bk ⊗ tnk ) ⊗ w (31)

where the product on the left is in the universal enveloping algebra of L.

b
Set
X
Y (a, z) := π(a ⊗ tn )z −n−1 (a ∈ L). (32)
n∈Z
P
It is easy to see that if n + i ni > 0 then (31) reduces to 0. In par-
ticular, Y (a, z) ∈ F(V ). The following calculation, showing that the fields
Y (a, z) (a ∈ L) are mutually local of order 2 (i.e. we may take k = 2 in (21)),
gives a first insight into how locality comes into play. Thus
(z1 − z2 )2 [Y (a, z1 ), Y (b, z2 )]
X
= (z1 − z2 )2 [π(a ⊗ tm ), π(b ⊗ tn )]z1−m−1 z2−n−1
m,n∈Z
X
= (z1 − z2 )2 π([(a ⊗ tm ), (b ⊗ tn )])z1−m−1 z2−n−1
m,n∈Z
X
= (z1 − z2 ) 2
{π([a, b] ⊗ tm+n ) + mδm+n,0 ha, biπ(K)}z1−m−1 z2−n−1
m,n∈Z
( )
X X X
= (z1 − z2 )2 π([a, b] ⊗ tp ) z1−m−1 z2m−p−1 + ha, biπ(K) mz1−m−1 z2m−1
p∈Z m∈Z m∈Z

11
 X X
= z2−p−2 (z1 − z2 )2 π([a, b] ⊗ tp ) z1−m−1 z2m+1 +
p∈Z m∈Z
X
z2−2 (z1 2
− z2 ) ha, biπ(K) mz1−m−1 z2m+1
m∈Z
 2 X    2  
z1 z1 z1 z1
= z2−p −1 p
π([a, b] ⊗ t )δ − − 1 ha, biπ(K)δ 0
.
z2 p∈Z
z2 z2 z2

(33)

Here
P δ(z)n−1is as in Section 4 (cf. comments preceding (24)), and δ 0 (z) :=
n∈Z nz . Now check that (z − 1)k δ(z) = 0 for k ≥ 1, (z − 1)k δ 0 (z) = 0
for k ≥ 2. In particular, (33) vanishes and (z1 − z2 )2 [Y (a, z1 ), Y (b, z2 )] = 0,
as asserted.
When W = Cv0 is the trivial 1-dimensional L-module we can go further,
and see the begginnings of a CFT. Here,

V = V (l, Cv0 ) ∼ b−1 ) ⊗ Cv0

= S(L
= S(⊕∞ m=1 L ⊗ t
−m
) ⊗ Cv0 (34)
= C(1 ⊗ v0 ) ⊕ (L ⊗ t−1 ) ⊗ Cv0
⊕ L ⊗ t−2 ⊕ S 2 (L ⊗ t−1 ) ⊗ Cv0 ⊕ . . .


∼
= C1 ⊕ L ⊕ (L ⊕ S 2 (L)) ⊕ . . .
∼
=
where we have used the natural identification L → L ⊗ t−1 , a 7→ a ⊗ t−1 , set
1 ⊗ v0 = 1, and dropped v0 from the notation for convenience.
In this way, the field Y (a, z) (32) is associated with the state a ∈ V .
Y (a, z) is creative (cf. (27)) because
X
Y (a, z)1 = {π(a ⊗ tn )(1 ⊗ v0 )}z −n−1
n∈Z
X∞ −∞
X
−n−1
= n
{1 ⊗ (a ⊗ t )v0 }z + {(a ⊗ tn ) ⊗ v0 }z −n−1
n=0 n=−1
−∞
X
= a ⊗ t−1 ⊗ v0 + {(a ⊗ tn ) ⊗ v0 }z −n−1
n=−2
= a + O(z).

12
(Because Cv0 is the trivial L-module then (a ⊗ tn )v0 = 0 for n ≥ 0.) Y (a, z)
is also translation covariant (loc. cit.): if m ≥ 1 then

[d/dt, Y (a, z)](b ⊗ t−m )

d X X
= {a ⊗ tn .b ⊗ t−m ⊗ v0 }z −n−1 + m {a ⊗ tn .b ⊗ t−m−1 ⊗ v0 }z −n−1
dt n∈Z n∈Z
d X
= {a ⊗ tn .b ⊗ t−m ⊗ v0 }z −n−1 +
dt n<0
d X
{[a, b] ⊗ tn−m ⊗ v0 + nδn,m ha, biK ⊗ v0 }z −n−1 +
dt n≥0
X
m {a ⊗ tn .b ⊗ t−m−1 ⊗ v0 }z −n−1
n∈Z
X
= {na ⊗ tn−1 .b ⊗ t−m ⊗ v0 − ma ⊗ tn .b ⊗ t−m−1 ⊗ v0 }z −n−1 +
n<0
X X
{(n − m)[a, b] ⊗ tn−m−1 ⊗ v0 }z −n−1 + m {a ⊗ tn .b ⊗ t−m−1 ⊗ v0 }z −n−1
n≥0 n∈Z
X X
−m −n−1
= {na ⊗ tn−1
.b ⊗ t ⊗ v0 }z +m {a ⊗ tn .b ⊗ t−m−1 ⊗ v0 }z −n−1 +
n<0 n≥0
X
{(n − m)[a, b] ⊗ tn−m−1 ⊗ v0 }z −n−1
n≥0
X X
= {na ⊗ tn−1 .b ⊗ t−m ⊗ v0 }z −n−1 + m {[a, b] ⊗ tn−m−1 ⊗ v0 }z −n−1 +
n<0 n≥0
X
m(m + 1)ha, bi{K ⊗ v0 }z −m−2 + {(n − m)[a, b] ⊗ tn−m−1 ⊗ v0 }z −n−1
n≥0
X X
−m −n−1
= {na ⊗ tn−1
.b ⊗ t ⊗ v0 }z + n{[a, b] ⊗ tn−m−1 ⊗ v0 }z −n−1 +
n<0 n≥0
−m−2
m(m + 1)ha, bi{K ⊗ v0 }z
(
d X X
= − {a ⊗ tn−1 .b ⊗ t−m ⊗ v0 }z −n + {[a, b] ⊗ tn−m−1 ⊗ v0 }z −n +
dz n<0 n≥0

mha, bi{K ⊗ v0 }z −m−1

13
 ( )
d X
= − {a ⊗ tn−1 .b ⊗ t−m ⊗ v0 }z −n
dz n∈Z
d
= − Y (a, z)b ⊗ t−m .
dz
This shows that [D, Y (a, z)] = d/dzY (a, z) where D = −d/dt, and be-
cause 1 is independent of t then D1 = 0. It should come as no surprise that
in fact (V (l, Cv0 ), Y, 1 ⊗ v0 , −d/dt) is a vertex algebra/CFT. Indeed, based
on what we already know, the result follows from the following general result.

V is a linear space with 0 6= 1 ∈ V, D ∈ End(V ), and mutually local,

translation covariant, creative fields y(u, z) ∈ F(V ) (u ∈ S ⊆ V ).
If V is spanned by states u1 (n1 ) . . . uk (nk )1 (ui ∈ S, ni ∈ Z) then (35)
there is a vertex algebra (V, Y, 1, D) with Y (u, z) := y(u, z) (u ∈ S).

In this situation, we say that S generates V . Thus (V (l, Cv0 ), Y, 1⊗v0 , −d/dt)
is a vertex algebra generated by L = L ⊗ t−1 . We will denote this vertex
algebra by V (L, l).

2. Virasoro algebra. (Several aspects of this case are similar to the previous
one, so we give less detail.)
The Virasoro algebra is the Lie algebra with underlying linear space
V ir := ⊕n∈Z CLn ⊕ CK with central element K and bracket

m3 − m
[Lm , Ln ] = (m − n)Lm+n + δm+n,0 K. (36)
12
(The denominator 12 is conventional here; it can be removed by rescaling.)
There is a triangular decomposition

V ir = V ir+ ⊕ V ir0 ⊕ V ir−

with

V ir+ := ⊕n>0 CLn , V ir0 := CL0 ⊕ CK, V ir− = ⊕n<0 CLn .

Let W = Cv0 be the 1-dimensional V ir0 -module such that L0 v0 = hv0 , Kv0 =
cv0 , extend to a V ir+ ⊕ V ir0 -module by letting V ir+ annihilate v0 , and form

14
the induced module

V = V (c, h) = IndVV ir
ir+ ⊕V ir0 W
∼
= S(V ir− ) ⊗ Cv0
= S(⊕n<0 CLn ) ⊗ Cv0
∼
= C1 ⊕ CL−1 ⊕ . . .

where 1 := 1 ⊗ v0 . h and c are called the conformal weight and central charge
respectively. Introduce
X
Y (ω, z) := Ln z −n−2 . (37)
n∈Z

One sees easily that Y (ω, z) ∈ F(V ). Note the slight change in convention
regarding powers of z in (37), which is standard. The reader may enjoy
proving that Y (ω, z) is a (self-) local field. Indeed, we have

(z1 − z2 )4 [Y (ω, z1 ), Y (ω, z2 )] = 0. (38)

Note that
X
Y (ω, z)1 = {Ln 1}z −n−2 = h1z −2 + L−1 1z −1 + L−2 1 + . . . . (39)
n∈Z

So there is no chance that Y (ω, z) is creative, because L−1 1 is nonzero by

construction. Furthermore, as it stands ω is just an abstract symbol, not a
state in V . We do not deal systematically with these issues here, but move
on to the definition of vertex operator algebra, where in some sense they get
resolved.

§8 Vertex operator algebras

A vertex operator algebra (VOA) is a vertex algebra with additional struc-
ture that arises from a special Virasoro field of the type discussed in Section
7. Specifically, a VOA is a vertex algebra/CFT (V, Y, 1, D) together with
a distinguished state ω ∈ V (called the conformal or Virasoro vector) such

15
that the following hold:
X
1) Y (ω, z) = L(n)z −n−2 and the modes L(n) generate an action of
n∈Z
the Virasoro algebra V ir (36) in which K acts on V as a scalar c,
called the central charge of V .
2) L(0) is a semisimple operator on V . Its eigenvalues lie in Z, are
bounded below, and have finite-dimensional eigenspaces.
3) D = L(−1).

This definition requires some discussion. Because (V, Y, 1, D) is a vertex

algebra, the fields Y (u, z) (u ∈ V ) are required to be mutually local and
creative. In particular, Y (ω, z) is necessarily self-local - a condition that can
be independently verified (38). Furthermore, comparison with (39) shows
that in the present situation we must have L(0)1 = L(−1)1 = 0 (otherwise
Y (ω, z) is not creative) and ω = L(−2)1 (because ω is created from the
vacuum by the field which corresponds to it). Note that L(n) = ω(n + 1).
The associativity formula (17) yields

(L(−1)u)(n) = (ω(0)u)(n) = ω(0)u(n) − u(n)ω(0) = [L(−1), u(n)].

Thanks to 3) and the last display, translation covariance may then be written

d/dzY (u, z) = [L(−1), Y (u, z)] = Y (L(−1)u, z).

In particular, Du = L(−1)u = u(−2)1, and 3) is consistent with (25).

For n ∈ Z we let Vn be the L(0)-eigenspace with eigenvalue n. Ac-
cording to 2), we have the fundamental spectral decomposition (into finite-
dimensional graded pieces)
∞
M
V = Vn (40)
n=n0

where n0 is the smallest eigenvalue of L(0). Because L(0)1 = 0 then 1 ∈ V0 .

We usually denote a VOA by the quadruple (V, Y, 1, ω). It is a model for
the creation and annihilation of bosons (particles of integer spin).

16
 The vertex algebra V (L, l) can sometimes be given the structure of a VOA
- we just have to find the right conformal vector. We describe two important
cases where this can be achieved.

1. Heisenberg algebra, or free bosonic theories.

Here, the Lie algebra L is abelian (i.e. [a, b] = 0 (a, b ∈ L)) of dimension
l, equipped with the (unique) nondegenerate symmetric bilinear form h , i
(which is automatically invariant). The levelP(the scalar by which K acts)
is also l. The conformal vector is ω := 1/2 li=1 vi (−1)vi where {vi } is an
orthonormal basis of L, and it transpires that the central charge is c = l. The
grading by L(0)-eigenvalues (40) coincides with the natural tensor product
grading in which L⊗t−m has degree m (cf. (34)). This is the rank l Heisenberg
VOA. It models l free (noninteracting) bosons. The special case when l = 24
underlies the bosonic string.

2. Kac-Moody theories, or WZW models.

In this case, L is a finite-dimensional simple Lie algebra, and h , i is the
Killing form (which is unique up to an overall
Pdimscalar). The conformal vector
L
is similar to the last case, namely ω = 1/2 i=1 vi (−1)vi for an orthonormal
basis {vi } of L. The central charge is c = l dim L/(l + h∨ ), and we obtain a
VOA as long as l + h∨ 6= 0 (h∨ is the dual Coxeter number of L).

§9 Super vertex algebras

Physically realistic theories incorporate both bosons and fermions. Ax-
iomatically, this corresponds to super vertex (operator) algebras (SV(O)A).
We limit ourselves here to the basic definitions.
The Fock space for a SVA is a linear super space, i.e. a linear space V
equipped with a Z2 -grading V = V 0 ⊕ V 1 , and a nonzero vacuum vector
1 ∈ V 0 . Here and below, superscripts will always lie in {0, 1} regarded as
the two elements of Z/2Z. We write |u| = p if u ∈ V p . V 0 and V 1 are called
the even and odd parts of V respectively.
There is a correspondence u 7→ YP(u, z) between states u ∈ V and mutu-
ally local, creative fields Y (u, z) := n∈Z u(n)z −n−1 , and we have

u(n) : V p → V p+|u| .

17
Finally, we require the super version of the basic identity (15), namely
∞  
X p
{u(r + i)v}(p + q − i)w =
i=0
i
∞  
X r
i
(−1) {u(p + r − i)v(q + i)w − (−1)r+|u||v| v(q + r − i)u(p + i)w}.
i=0
i

The delta-function version of this (cf. (24)) reads

   
−1 z1 − z2 |u||v| −1 z2 − z1
z0 δ Y (u, z1 )Y (v, z2 ) − (−1) z0 δ Y (v, z2 )Y (u, z1 )
z0 −z0
 
−1 z1 − z0
= z2 δ Y (Y (u, z0 )v, z2 ). (41)
z2

Note that the substructure (V 0 , Y, 1) is a vertex algebra. As in the case

of vertex algebras, these axioms are equivalent to a SCFT for which super
locality, the super analog of (22), is as follows:

(z1 − z2 )k [Y (u, z1 ), Y (v, z2 )] = (−1)|u||v| (z1 − z2 )k [Y (v, z2 ), Y (u, z1 )]

A SVOA is a quadruple (V, Y, 1, ω) such that analogs of 1)-3) of Section 8

hold. The only change is that eigenvalues of L(0) are allowed to lie in 1/2Z.
L(0) leaves V 0 invariant, and on this subspace the eigenvalues lie in Z. Thus
(V 0 , Y, 1, ω) is a VOA.
There are further variations on this theme, where it is assumed that
additional special states and fields exist. These lead to so-called N = 1
SCFT, N = 2 SCFT, etc. They play a rôle in certain geometric and physical
applications, although we will not discuss them here.

Lecture 3

§10 Modules over a VOA

Suppose that (V, Y, 1) is a VA. A module over this structure, i.e. a V -
module, is a linear space W and a linear map YW : V → F(W ), v → 7

18
YW (v, z) = n∈Z vW (n)z −n−1 such that YW (1, z) = IdW and the analog of
P
(15) holds, i.e. for all u, v ∈ V, w ∈ W we have
∞  
X p
{u(r + i)v}W (p + q − i)w = (42)
i=0
i
∞  
i r
X
(−1) {u(p + r − i)W v(q + i)W w − (−1)r v(q + r − i)W u(p + i)W w}.
i=0
i

As before, there are a number of auxiliary consequences of this identity.

We mention only that the fields YW (u, z) (u ∈ V ) are mutually local. Gener-
ally W has no analog of the vacuum vector, so creativity has no meaning for
the fields YW (u, z).
Suppose that (V, Y, 1, ω) is a VOA. A module over this structure is a
V -module (in the previous sense) such that LW (0) is a semisimple operator
(on W ) with finite-dimensional eigenspaces. The eigenvalues of LW (0) are
truncated below in the following sense: given an eigenvalue λ, there are only
finitely many eigenvalues of the form λ − n (n ∈ N). In particular, we have
a spectral decomposition of W analogous to (40).
W is called irreducible, or simple, if the only subspaces invariant
P under all
modes uW (q) (u ∈ V, q ∈ Z) are 0 and W . It is easy to see that n∈Z Wλ+n
is always an invariant subspace. Hence, if W is a simple V -module then the
spectral decomposition takes the form
∞
M
W = Wh+n (43)
n=0

for some uniquely determined h = hW ∈ C called the conformal weight of

W.
We give a few examples of V -modules.
1. If (V, Y, 1, ω) is a VOA then V is itself a V -module, called the adjoint
module.
2. Suppose that V is a VOA and W ⊆ V satisfies u(n)w ∈ W (u ∈ V, w ∈ W ).
Then W is a V -module. W is called an ideal of V , because we also have
w(n)u ∈ W (use skew-symmetry (29)). It follows that V /W is the Fock
space of a VOA (the quotient VOA of V ) in which the mode (u + W )(q) of
Y (u + W, z) is the operator induced on V /W by u(q). We call V simple if

19
the only ideals are the trivial ones V and 0. For example, any Heisenberg
VOA V (l, Cv0 ) is simple.
3. If (V, Y, 1, ω) is a SVOA (cf. Section 9) the odd part V 1 is a module over
the even part V 0 . In this case the conformal weight of V 1 is necessarily
a half-integer. (A VOA may have no modules with half-integral conformal
weight except for (direct sums of) the adjoint module. Thus the VOAs that
can occur as the even part of a SVOA are severely restricted.)
4. Recall the rank l Heisenberg VOA V (l, Cv0 ) (cf. Section 8) generated by a
rank l abelian Lie algebra L. For an L-module W we constructed (Section 7)
a space V (l, W ) and mutually local fields Y (a, z) ∈ F(V (l, W )) (a ∈ L). It is
not hard to see that V (l, W ) is a V (l, Cv0 )-module, and it is simple whenever
dim W = 1.

§11 Lattice theories

An integral lattice L is a finitely generated free abelian group equipped with
a nondegenerate, symmetric, Z-valued bilinear form h , i : L × L → Z. Let
l := rkL. Set H := C ⊗Z L and let h , i also denote the linear extension to H.
We regard H as an abelian Lie algebra equipped with a symmetric invariant
bilinear form. As such, there is the associated Heisenberg VOA V (l, Cv0 ) (cf.
Sections 7 and 8).
Fix β ∈ L. Let Ceβ be the 1-dimensional linear space spanned by eβ ,
regarded as an H-module through the action

α.eβ := hα, βieβ (α ∈ H). (44)

Associated to this L-module is the simple V (l, Cv0 )-module V (l, Ceβ ). Note
that Ce0 is the trivial L-module, so that it can be identified with Cv0 . Also,
we have a linear isomorphism V (l, Ceβ ) ∼= S(H b − ) ⊗ Ceβ (cf. (30)). We form
the Fock space
M
VL := V (l, Ceβ )
β∈L

∼
M
b −) ⊗
= S(H Ceβ (45)
β∈L
b−
= S(H ) ⊗ C[L].

20
(It is convenient to identify the group algebra C[L] of L with ⊕β Ceβ .) We
discuss the following result:
VL carries the structure of a SVOA; if L is an even
lattice (i.e. hβ, βi ∈ 2Z for β ∈ L), then VL is a VOA.
S(H b − ) is naturally identified with the Heisenberg VOA itself, and in par-
ticular it is generated (cf. (35)) by the fields Y (α, z) (α ∈ H). Because each
V (l, Ceβ ) is a Heisenberg module, the Y (α, z) naturally extend to (mutually
local) fields on VL . To get a generating set of fields for VL (loc. cit.) we would
need to extend the set of Y (α, z) to a larger set of mutually (super) local
fields by defining fields Y (1 ⊗ eβ , z) (β ∈ L) directly. We will skip the details
here. RecallP (cf. Section 8) that the conformal vector for the Heisenberg VOA
is ω := 1/2 li=1 vi (−1)vi for an orthonormal basis {vi } of H. This state is
also taken as the conformal vector of VL . In Pparticular, the central charge of
−n−2
VL is the rank l of L. The field Y (ω, z) = n L(n)z determined by ω is
defined in the natural way, i.e. on V (l, Ceβ ) it acts as YV (l,Ceβ ) (ω, z). Since
each summand in (45) is a Heisenberg module, L(0) acts semisimply on each
of them, and therefore on VL . We consider the eigenvalues and eigenspaces
of L(0) in the next Section. Finally, we note that VL is a simple VOA if L is
even.
§12 Partition functions
Suppose that (V, Y, 1, ω) is a VOA of central charge c (cf. Section 8, axiom
1)), and spectral decomposition (40) into L(0)-eigenspaces. The partition
function of V is the formal q-series
∞
X
−c/24
Z(q) = ZV (q) := q dim Vn q n . (46)
n=n0

(This is the first place that c has played a rôle in the proceedings.) More
generally, for a simple V -module W with spectral decomposition (43), the
corresponding partition function is
∞
X
Z(q) = ZW (q) := q h−c/24 dim Vn q n . (47)
n=0

These expressions make sense because L(0)-eigenspaces in both cases are

finite-dimensional. Indeed, it will be convenient to define the partition func-
tion for any graded space in the same way, as long as it too makes sense. One

21
can often check the VOA axioms regarding the conformal vector (Section 8,
axiom 2)) by directly computing the corresponding partition function. We
will carry this out in the case of the Fock spaces for the Heisenberg VOA
and the lattice theory VL .
For the rank l Heisenberg theory V = V (l, Cv0 ) we saw (34) that V
has a tensor decomposition S(⊕∞ m=1 L ⊗ t
−m
) ⊗ Cv0 (L is the abelian Lie
algebra of rank l). It is not hard to see that the L(0)-grading respects this
decomposition, and that L ⊗ t−m is an eigenspace with eigenvalue m. Since
symmetric powers are multiplicative over direct sums, we obtain
∞
Y
−l/24
partition function of S(L ⊗ t−m )


ZV (l,Cv0 ) (q) = q
m=1
Y∞
= q −l/24 (1 + q m + q 2m + . . .)l
m=1
Y∞
= q −l/24 (1 − q m )−l = η(q)−l ,
m=1

η(q) being the eta function (12).

We turn to the lattice theory VL . The partition function for VL is the
product of those for the two factors S(H b − ) and C[L] in (45). Moreover, the
first of these is just the partition function for the Heisenberg theory that
we just computed. As for the second factor, using the module version of
associativity (17) we have
l
X
L(0).1 ⊗ eβ = 1/2 (vi (−1)vi )(1).1 ⊗ eβ
i=1
Xl X∞
= 1/2 {(vi (−1 − j)vi )(1 + j) + vi (−j)vi (j)}1 ⊗ eβ
i=1 j=0
l
X
= 1/2 {vi (0)vi (0)}1 ⊗ eβ
i=1
Xl
= 1/2 hvi , βi2 1 ⊗ eβ = 1/2hβ, βi1 ⊗ eβ .
i=1

(Here, we used that vi (j) moves across the tensor sign if j ≥ 0 and annihilates
eβ if j ≥ 1, as well as (44). The last equality holds because {vi } is an

22
orthonormal basis of H.) The upshot is that 1 ⊗ eβ is an eigenvector for L(0)
with eigenvalue 1/2hβ, βi. We therefore see that
X
partition function of C[L] = q 1/2hβ,βi = θL (q)
β∈L

is the theta function of L (6). Altogether then, we obtain

θL (q)
ZVL (q) = , (48)
η(q)l
and in particular the L(0)-eigenspaces are indeed finite-dimensional.
Let L0 ⊆ L consist of those β ∈ L such that hβ, βi ∈ 2Z. Because L is
an integral lattice, L0 is a sublattice of L with |L : L0 | ≤ 2. If L = L0 then
L0 is an even lattice and VL is a VOA. If |L : L0 | = 2, choose γ ∈ L \ L0 .
Then, with an obvious notation, there is a decomposition
b − ) ⊗ C[L0 ] ⊕ S(H
VL = S(H b − ) ⊗ C[L0 + γ],

where S(H b − ) ⊗ C[L0 ], S(H

b − ) ⊗ C[L0 + γ] are the parts of VL graded by Z
and 1/2 + Z respectively. In this case, VL is a SVOA and S(H b − ) ⊗ C[L0 ] and
b − ) ⊗ C[L0 + γ] are the even and odd parts.
S(H
With this discussion, we have at last made contact with the ideas of
Section 2. For if we take L to be the Leech lattice Λ (a (self-dual) even
lattice of rank 24), then according to (48) we have
θΛ (q)
ZVΛ (q) = , (49)
∆(q)
and (using (5)) this is the partition function (3). Similar comments apply to
(2), which is now seen to be the partition function for V3E8 .
Thanks to (48) and known transformation properties of θ- and η-functions,
it follows that the partition function ZVL (q) of a lattice theory is a modular
function of weight zero on a congruence subgroup of the modular group. We
derived this result only after explicitly computing the partition function, but
in fact there is a large class of VOAs for which a priori results about the
partition function and its transformation properties can be proved without
explicitly knowing what the partition function is. This is the class of regular
VOAs.

23
 One point that we will not pursue but that deserves mention is this:
the partition function of a VOA is a formal q-expansion, with no a priori
convergence properties. On the other hand, at least for a regular VOA, the
partition function turns out to be holomorphic in the complex upper half-
plane H when we think of it as a function ZV (τ ) with q = e2πiτ , τ ∈ H. For
this reason, we now write partition functions as functions of τ rather than q.
Although there will be no time to develop the general theory of regular
VOAs in these lectures, we can illustrate some of the ideas using the lattice
theory VL . If V is an arbitrary VOA, the set of modules over V are the objects
of a category V -Mod. A morphism f : W1 → W2 between two V -modules
W1 , W2 is a linear map such that

f (u(n)w) = u(n)(f (w)) (u ∈ V, n ∈ Z, w ∈ W1 ).

In terms of fields, this reads f YW1 (u, z) = YW2 (u, z)f . Roughly speaking, V
is called rational if V -Mod is semisimple, i.e. every V -module is a direct sum
of simple V -modules. (In fact, one has to include additional types of modules
that we did not discuss in Section 10.) It can be shown that a rational VOA
has only finitely many (isomorphism classes of) simple V -modules. A VOA
is regular if it is both rational in the above sense and satisfies an additional
condition that we will not discuss here.
If L is an even lattice as before then VL is indeed a regular VOA. It
therefore has only finitely many inequivalent simple modules, and in fact
they are enumerated by the quotient group L0 /L where L0 is the dual lattice
of L. If we set E := R ⊗Z L then the dual lattice is

L0 := {α ∈ E | hα, βi ∈ Z (β ∈ L)}.

Because L is integral and positive-definite then L ⊆ L0 is a subgroup of finite

index. The simple VL -modules have a structure that is parallel to VL itself.
The Fock spaces are
M
VL+γ := V (l, Ceβ )
β∈L+γ

∼
M
b −) ⊗
= S(H Ceβ (50)
β∈L+γ
b − ) ⊗ C[L + γ],
= S(H

24
(compare with (45)), where L + γ ∈ L0 /L. The partition function is

θL+γ (τ )
ZVL+γ (τ ) = ,
η(τ )l

which is once again a modular function of weight zero on a congruence sub-

group of the modular group. Indeed, one knows that the linear space

P := hθL+γ (τ )/η(τ )l | L + γ ∈ L0 /Li

spanned by these partition functions furnishes a representation of the mod-

ular group (through the usual action τ 7→ aτ +b
cτ +d
). This set-up is conjectured
to hold for all rational VOAs V ; that is, if P is the span of the partition
functions of the (finitely many) simple V -modules then P affords a repre-
sentation of the modular group that factors through a principal congruence
subgroup. This phenomenon is often called modular-invariance of rational
VOAs.
An important special case of these ideas arises when the VOA V is not
only rational, but has (up to isomorphism) a unique simple module, namely
the adjoint module V . We call such a V holomorphic. Then our discussion
of modular-invariance shows that the partition function ZV (τ ) of a holomor-
phc VOA is a modular function on the full modular group (perhaps with a
character). For example, since the simple VL -modules are indexed by the
cosets of L in L0 , it follows that VL is holomorphic if, and only if, L = L0
is self-dual. The Leech lattice Λ and orthogonal sums of the E8 root lattice
are examples of self-dual lattices, and indeed their partition functions (2),
(3) are modular functions on the full modular group.

Lecture 4

§13 The Lie algebra on V1

We have seen that a regular VOA that is holomorphic (i.e. has a unique
simple module) has a partition function that is a modular function of weight
0 on the full modular group. A case in point is the Leech lattice theory VΛ ,
which has central charge 24 (= rkΛ) and partition function ZVΛ (τ ) = J +24 =
q −1 + 24 + 196884q + . . .. Our goal now is to construct a holomorphic VOA

25
V \ , also of central charge 24, whose partition function is J (4), which has
constant term 0. This is the1 Moonshine module.
Although the VOAs VΛ and V \ have partition functions differing only in
their constant term, many of their algebraic properties are quite different.
Indeed, these properties are to a large extent governed by the constant term.
For this reason, we begin with a general discussion of this point. We restrict
attention to VOAs of CFT-type, which means that in the spectral decom-
position (40) the pieces Vn vanish for n < 0 and V0 = C1. (Recall that we
always have 1 ∈ V0 .) There are many interesting VOAs that are not of CFT-
type, nevertheless CFT-type theories are natural from a physical standpoint
because they arise from ’unitarity’ assumptions. Be that as it may, our basic
assumption here is that the spectral decomposition of V has the shape

V = C1 ⊕ V1 ⊕ V2 ⊕ . . .

For states u, v ∈ V1 we define a bracket [ ] by setting [uv] := u(0)v. Now

L(0)u(0)v = [L(0), u(0)]v + u(0)L(0)v = u(0)v.

([L(0), u(0)] = 0 by translation covariance and L(0)v = v because v ∈ V1 ).

This shows that u(0)v ∈ V1 , so that we have a bilinear product [ ] : V1 ×V1 →
V1 . One can check that this makes V1 into a Lie algebra. (Use (29) for skew-
symmetry [uv] = −[vu] and the associativity formula (17) for the Jacobi
identity [[uv]w] + [[wu]v] + [[vw]u] = 0.)
For a V OA V of CFT-type and central charge c = 24, the partition
function has the general shape ZV (τ ) = q −1 + dim V1 + . . .. So for such
theories, the constant term is the dimension of the Lie algebra on V1 .
If L is an even lattice of rank l, the nature of the partition function (48)
of the lattice theory shows that

ZVL (τ ) = q −l/24 (1 + (l + |L2 |)q + . . .), (51)

where L2 = {α ∈ L | hα, αi = 2} are the roots of L. In particular, VL

is of CFT-type. The Lie algebra on (VL )1 is reductive, being a direct sum
a ⊕ g where a is abelian and g is semisimple with root system L2 . (The
set of roots in an even lattice always carries the structure of a semisimple
1
It is expected that there is a unique VOA with partition function J, but this remains
open.

26
root system embedded in the ambient Euclidean space E = R ⊗ L.) For
example, if L = 3E8 then the Lie algebra on (VL )1 is semisimple, being the
sum of three copies of the E8 Lie algebra. (Note that dim E8 = 248, so that
dim(VL )1 = 744, in agreement with (2).) Similarly, the Leech lattice Λ has
no roots, whence (VΛ )1 is abelian of rank l = 24.
Because J has no constant term, a VOA V \ with partition function J
and central charge c = 24 necessarily has no corresponding Lie algebra. In
particular, V \ cannot be a lattice theory, because the weight one piece never
vanishes for a lattice theory (cf. (51)).
§14 Automorphisms
Let V be a (S)VOA. An automorphism of V is an invertible linear map
g : V → V such that g(ω) = ω and gv(q)g −1 = g(v)(q) for all v, q, i.e.
gY (v, z)g −1 = Y (g(v), z) (v ∈ V ). (52)
We give some basic examples of automorphisms.
1. One checks (use induction and (16) or (17)) that for n ≥ 0,
n  
i n
X
n
(u(0) v)(q) = (−1) u(0)n−i v(q)u(0)i (u, v ∈ V, q ∈ Z).
i=0
i
Therefore,
∞
u(0)
X 1
(u(0)n v)(q)


e .v (q) =
n=0
n!
∞ X
n
X (−1)i
= u(0)n−i v(q)u(0)i
n=0 i=0
i!(n − i)!
= e v(q)e−u(0) ,
u(0)

showing that (52) holds with g = eu(0) . Furthermore, if V is of CFT-type

(cf. Section 12) and u ∈ V1 then
u(0)ω = u(0)L(−2)1
= [u(0), L(−2)]1 + L(−2)u(0)1
= [u(0), ω(−1)]1
= −[ω(−1), u(0)]1
X∞
= − (−1)i (ω(i)u)(−1 − i)1
i=0

27
 = − {(L(−1)u)(−1) − (L(0)u)(−2) + (L(1)u)(−3)} 1
= 0.

(For the last two equalities use translation covariance, L(0)u = u (because
u ∈ V1 ), L(1)u ∈ V0 = C1, L(n)u ∈ V1−n = 0 for n ≥ 2, and 1(q) = δq+1,0 IdV
(cf. (28).)
It follows from this calculation that if V is a VOA of CFT-type then
u(0)
{e | u ∈ V1 } is a set of automorphisms of V . In the previous Section we
learned that V1 carries the structure of a Lie algebra with bracket [uv] =
u(0)v. Now we see that the usual action of the associated Lie group G
generated by exponentials eadu extends to an action of G as automorphisms
of V .
2. Suppose that V is a SVOA. Then there is a canonical involutorial au-
tomorphism which acts as +1 on the even part of V and −1 on the odd
part.
3. A related example (and the one we will need later) is an involutorial
automorphism t of a lattice VOA VL , defined to be a lifting of the −1 au-
tomorphism of the lattice L. t also acts as −1 on the abelian Lie algebra
C ⊗ L and then acts as naturally on the associated Heisenberg VOA (cf. (34)
- where L is the Lie algebra, not the lattice!) and on VL , where

t(u ⊗ eβ ) = t(u) ⊗ e−β b −)

(u ∈ S(H (53)

(cf. (45)).

If g is an automorphism of V then gY (ω, z)g −1 = Y (g(ω), z) = Y (ω, z),

in particular gL(0)g −1 = L(0). Therefore g acts on the eigenspaces of V ,
i.e. the homogeneous pieces Vn . We may therefore define additional partition
functions
∞
X
−c/24
ZV (g, τ ) := q (TrVn g)q n .
n=n0

Let’s compute this trace function for the automorphism t of VL . It is clear

from (53) that the only contributions to the trace arise from states u ⊗ e0 , i.e.

28
 b − ). Therefore by (34),
from states in the Heisenberg VOA Fock space S(H

ZVL (t, τ ) = Trace t on q −l/24 S(⊕m>0 H ⊗ t−m )

O
= Trace t on q −l/24 S(Cu ⊗ t−m )l
m>0
Y
−l/24
= q (1 − q + q 2m − . . .)l
m

m>0
Y
−l/24
= q (1 + q m )−l
m>0
 l
η(τ )
= . (54)
η(2τ )

This is a modular function of weight 0. If l = 24 it is almost equal to (11)!

§15 Twisted sectors

Let (V, Y ) be a VOA and g an automorphism of V of finite order R. A
g-twisted V -module, or g-twisted sector, is a generalization of V -module (to
which it reduces if g = 1). Precisely, it is a pair (Wg , Yg ) consisting of a Fock
space Wg and a Y -map Yg : V → F(Wg ), u 7→ Yg (u, z) where
X
Yg (u, z) := u(n)z −n−1 ∈ End(W )[[z 1/R , z −1/R ]]
n∈r/R+Z

whenever g(u) = e−2πir/R (r ∈ Z), and Yg (1, z) = IdWg . The twisted vertex
operators Yg (u, z) are required to satisfy twisted analogs of the basic identity
(15). In the delta-function formulation (cf. (24)) this reads
   
−1 z1 − z2 −1 z2 − z1
z0 δ Yg (u, z1 )Yg (v, z2 ) − z0 δ Yg (v, z2 )Yg (u, z1 )
z0 −z0
 −r/R  
−1 z1 − z0 z1 − z2
= z2 δ Yg (Y (u, z0 )v, z2 ). (55)
z2 z0

Finally, the operator Lg (0) (the zero mode of Yg (ω, z)) is required to be
semisimple with finite-dimensional eigenspaces. The eigenvalues satisfy a
truncation condition analogous to that for V -modules (cf. the discussion in
Section 10 preceding display (43)). There is an obvious notion of irreducible

29
(or simple) g-twisted module, and as in the untwisted case (cf. (43)) the
spectral decomposition of a simple g-twisted module takes the form
∞
M
Wg = (Wg )hg +n/R (56)
n=0

for a scalar hg (the conformal weight). Needless to say, the twisted sector
has an associated partition function
∞
X
−c/24+hg
ZWg (τ ) := q dim(Wg )n q n/R .
n=0

Let us specialize to the case of the (even, self-dual) Leech lattice Λ with
its associated VOA VΛ and canonical involution t (cf. Section 14). In this
case there is (up to isomorphism) a unique simple t-twisted module, denoted
by VΛ (t, τ ). The following transformation law can be proved:

ZVΛ (t, −1/τ ) = ZVΛ (t) (τ ).

Using (54), the partition function of the t-twisted sector must be

 24
η(−1/τ )
ZVΛ (t) (τ ) =
η(−2/τ )
 24
12 η(τ )
= 2
η(τ /2)
∞
Y
12 1/2
= 2 q (1 + q n/2 )24 , (57)
n=1
√
(using the transformation law η(−1/τ ) = ( τ /i)η(τ )). Because the central
charge is c = 24, it follows that the conformal weight of VΛ (t, τ ) is 3/2.
Similarly to the rank 24 Heisenberg VOA, the product term in (57) is the
partition function of a symmetric algebra S(⊕n>0 H ⊗ t−n/2 ) (cf. (34)). This
suggests how one might try to construct the twisted sector, though we must
skip the details here. (The curious factor 212 turns out to correspond to a
Clifford algebra. Cf. Section 17 for further comment.)

30
§16 The Moonshine Module
Retaining the notation of the previous Section, consider
VΛ ⊕ VΛ (t). (58)
The involution t acts naturally on the twisted sector: in the ‘usual way’
on S(⊕n>0 H ⊗ t−n/2 ) and as −1 on the 212 constant part. The Moonshine
Module is then defined to be the space of t-invariants
V \ := VΛ+ ⊕ VΛ (t)+ . (59)
Now every state u ⊗ eβ ∈ VΛ (β 6= 0) produces a t-invariant u ⊗ eβ +
t(u) ⊗ e−β . On the other hand, the partition function of the Heisenberg VOA
(consisting of states u⊗e0 ) is 1/∆(τ ) and the graded trace of t is ∆(τ )/∆(2τ )
(the case l = 24 of (54)). It follows that
ZV + (τ )
Λ

= (ZVΛ (τ ) − 1/∆(τ ))/2 + (1/∆(τ ) + ∆(τ )/∆(2τ ))/2

= (ZVΛ (τ ) + ∆(τ )/∆(2τ ))/2
= ((q −1 + 24 + 196884q + . . .) + q −1 (1 − 24q + 276q 2 + . . .))/2
= q −1 + 98580q + . . .
On the other hand, a similar calculation using (54) and the nature of the
twisted sector as a symmetric algebra shows that
ZVΛ (t)+ (τ ) = 212 (∆(τ )/∆(τ /2) − q∆(τ /2)/∆(τ ))/2
∞ ∞
!
Y Y
= 211 q 1/2 (1 + q n/2 )24 − (1 − q n−1/2 )24
n=1 n=1
= 98304q + . . .
Altogether, we have
ZV \ (τ ) = ZV + (τ ) + ZVΛ (t)+ (τ )
Λ

= (q −1 + 98580q + . . .) + (98304q + . . .)
= q −1 + 196884q + . . . (60)
It is clear from the above that ZV \ (τ ) is a modular function of weight 0 and
level at most 2, and it is easy to check that in fact it is invariant under the
full modular group. Thus from the q-expansion we arrive at the identity
ZV \ (τ ) = J.

31
 The space (58) has the structure of an abelian intertwining algebra, a
generalization of VOA and SVOA. The main missing ingredient, which we
cannot go into here, is the definition of fields YVΛ ⊕VΛ (t) (u, z) for states u in
the twisted sector satisfying an appropriate variation of the basic identity
(15), (24). Once this is done, t is seen to be an automorphism of this larger
structure. Then it is easy to see that the t-invariant subspace V \ , together
with the restriction of the fields to this subspace, defines the structure of a
VOA on V \ with central charge 24. Furthermore, VΛ+ ⊕ VΛ (t)− is a SVOA
with even part VΛ+ . (Indeed, it is an N = 1 superconformal field theory, a
term we alluded to but did not define in Section 9.)
Consider a VOA V of CFT-type (cf. Section 13) with trivial Lie algebra
V1 :

V = V0 ⊕ V2 ⊕ . . .

(V \ satisfies these conditions, as follows from (60).) If u, v ∈ V2 , define

u.v := u(1)v. It is easy to check that u(1)v ∈ V2 , so that we have a
(nonassociative) bilinear product on V2 . By skew-symmetry (29), v(1)u =
u(1)v − L(−1)u(2)v + L(−1)2 u(3)v/2 − . . . But u(2)v ∈ V1 = 0, u(3)v ∈ C1
and L(−1)1 = 0, and all other u(q)v (q ≥ 4) lie in Vn with n < 0 and hence
also vanish. The upshot is that u(1)v = v(1)u, so that V2 has the structure
of a commutative, nonassociative algebra. In the case of V \ , this is precisely
the algebra B that we discussed in Section 1.

§17 AutV \
Consider the CFT

VΛ = C1 ⊕ (VΛ )1 ⊕ . . .

where Λ is, as before, the Leech lattice. Because Λ has no roots, it follows
from (51) that dim(VΛ )1 = 24, and the Lie algebra on (VΛ )1 is abelian. So the
automorphisms eu(0) (u ∈ (VΛ )1 ) generate a 24-dimensional complex torus
T . Additional automorphisms of VΛ arise from the automorphism group
Co0 := Aut(Λ) of the Leech lattice, and there is a (nonsplit) short exact
sequence

1 → T → AutVΛ → Co0 → 1.

32
The automorphism t of Λ (or of VΛ ) is a central involution of Co0 , and the
quotient Co1 := Co0 /hti is the largest sporadic (simple) Conway group of
order 221 . . ..
Because t acts as −1 on T , its only fixed elements are those of order at
most 2. So the centralizer C(t) of t in AutVΛ (the elements that commute
with t) is described by another short exact sequence (also nonsplit)

1 → 224 → C(t) → Co0 → 1.

(224 is a direct product of 24 copies of Z2 , the 2-torsion of T .) Note that

|C(t)| = 246 . . .
As regards the Monster, the relevance of C(t) is that it preserves the
decomposition (59). This is a bit subtle: t acts trivially by definition, but
the action of C(t)/hti is projective on VΛ (t)+ . When it is linearized, we obtain
a third group Cb occuring as the middle term of a short exact sequence

1 → 21+24 → C
b → Co1 → 1,

where now 21+24 is the (nonabelian) linearization of the projective action of

the 2-torsion of T . 21+24 is a so-called extra-special group. It is familiar in
physics (24 × 24 Pauli matrices) and the theory of theta-functions. It has a
unique faithful irreducible representation, realizable on the 212 -dimensional
b = 246 . . . and
Clifford algebra that we identified at the end of Section 15. |C|

b ⊆ AutV \
C

It turns out that the decomposition (59) breaks the symmetry of V \ in

the sense that there are further automorphisms that do not preserve (59)
and hence do not lie in C.
b The Monster M is the full automorphism group
\
of V and also of the algebra B, and

|M | = 246 . . . 47.59.71

These results are not easily obtained, and we say no more about them here.
The graded traces ZV \ (g, τ ) for g ∈ M turn out to be hauptmoduln as
described in Section 1. This result is also difficult. We end these Notes with
the computation for a single automorphism g of order 2 that acts trivially

33
on VΛ+ and as −1 on VΛ (t)+ . A previous calculation shows that its graded
trace is a modular function of weight 0 and level 2. Specifically,
ZV \ (g, τ ) = ZV + (τ ) − ZVΛ (t)+ (τ )
Λ
−1
= (q + 98556q + . . .) − (98304q + . . .)
= q −1 + 276q + . . .
is the hauptmodul for the Monster element 2B (11).

References
The following textbooks and monographs cover the material in these Notes
and much more.
J. Conway et al, ATLAS of Finite Groups, Clarendon Press, OUP, 1985.
C. Dong and J. Lepowsky, Generalized Vertex Algebras and Relative Vertex
Operators, Birkhäuser, 1993.
I. Frenkel, J. Lepowsky and A. Meurman, Vertex Operator Algebras and the
Monster, Academic Press, 1988.
T. Gannon, Moonshine Beyond the Monster: The Bridge Connecting Alge-
bra, Modular Forms, and Physics, CUP, 2006.
V. Kac, Vertex Algebras for Beginners, 2nd ed., Univ. Lect. Ser. Vol. 10,
AMS, 1998.
J. Lepowsky and H. Li, Introduction to Vertex Operator Algebras and Their
Representations, Birkäuser, 2004.
A. Matsuo and K. Nagatomo, Axioms for a Vertex Algebra and the Locality
of Quantum Fields, Math. Soc. of Japan Memoirs, Vol. 4, 1999.
G. Mason and M. Tuite, Vertex operators and modular forms, in A Window
into Zeta and Modular Physics, MSRI Publ. No. 57, CUP, 2010.
Moonshine, The First Quarter Century and Beyond, J. Lepowsky, J. McKay
and M. Tuite eds., LMS Lect. Note Series No. 372, CUP, 2010.
Moonshine, the Monster, and Related Topics, C. Dong and G. Mason eds.,
Contemp. Math. No. 193, AMS, 1996.

34