A vertex operator algebra replaces one fixed representation matrix by an entire field of operators depending on a formal variable. The coefficients of those fields form infinitely many modes; their commutation and locality relations encode conformal symmetry, and their modules can carry fusion rules whose characters transform under the modular group.

The subject sits at a meeting point of infinite-dimensional representation theory, algebraic combinatorics, complex analysis and two-dimensional conformal field theory. Yet its basic mechanism can be approached in stages. A state a produces a formal field Y(a,z). Expanding that field gives operators a_n. A special state ω produces the Virasoro operators L_n. Modules carry compatible fields, characters package graded dimensions, and intertwining operators describe how modules fuse.

This article connects the BTT guides on Lie Algebra Representations, Fusion Categories, Quantum Groups and Categorification. The emphasis is mathematical. Physical conformal-field interpretations require additional analytic and model-specific assumptions.

State-field map · Virasoro structure · Heisenberg example · Modules · Characters · Fusion · Ising model · Modularity · Practice

Why ordinary algebraic multiplication is not enough

In an associative algebra, two elements multiply to one element. In a Lie algebra, two elements produce a bracket. In a vertex algebra, a state a does not merely multiply b once. It produces a formal Laurent series of possible operations on b.

Write

Y(a,z)=Σ_n∈Za_nz⁻ⁿ⁻¹.

Each coefficient a_n is a linear operator. The integer n labels a mode. Acting on another state b gives a formal series whose coefficients are the states a_nb.

The extra formal variable stores an infinite family of products in one object. Locality relations then control how two such fields can be reordered.

The state-field correspondence

A vertex algebra contains a vector space V, a distinguished vacuum vector 1, a translation operator T and a state-field map

Y:V→End(V)[[z,z⁻¹]].

For each a∈V, the field Y(a,z) is lower-truncated on every b: a_nb=0 for sufficiently large n. This makes each coefficient extraction meaningful.

The vacuum satisfies Y(1,z)=id_V, while Y(a,z)1 begins with a and contains no negative powers after the standard expansion. Translation satisfies

[T,Y(a,z)]=dY(a,z)/dz

and T1=0.

These axioms say that states, fields and infinitesimal translation are compatible rather than independent data.

Locality replaces ordinary commutativity

Two vertex fields need not commute term by term. Instead, for every a,b∈V there exists a sufficiently large integer N such that

(z−w)^N[Y(a,z),Y(b,w)]=0.

This is formal locality. Away from the possible singularity z=w, the fields behave as though they commute. The finite order of the pole controls the failure of naive commutativity.

The Jacobi identity used in many definitions is an equivalent compact formulation of locality, associativity and the mode relations under the standard vertex-algebra axioms.

Thus vertex algebras do not abandon algebraic control. They replace one multiplication law with a structured operator-product law.

Operator product expansions are compressed mode relations

In physics-inspired notation, the singular part of the product of fields is written as an operator product expansion, or OPE.

If the products a_nb vanish for n≥N, then the singular terms of Y(a,z)Y(b,w) involve only finitely many powers (z−w)^−r. Their coefficients are fields corresponding to states constructed from a and b.

The OPE therefore packages infinitely many commutator relations into a finite singular expansion when locality has bounded order.

A formal OPE is not automatically an analytic convergence statement. Vertex-algebra identities are first identities of formal series. Convergence belongs to an additional analytic layer used in conformal field theory and tensor-category constructions.

A conformal vector produces the Virasoro algebra

A vertex operator algebra includes a distinguished conformal vector ω. Expand its field as

Y(ω,z)=Σ_n∈ZL_nz⁻ⁿ⁻².

The modes L_n satisfy the Virasoro commutation relations

[L_m,L_n]=(m−n)L_m+n+[c/12](m³−m)δ_m+n,0I,

where c is the central charge.

The operator L₀ provides the conformal grading, and L₋₁ is the translation operator T. The central charge is therefore part of the representation-theoretic data of the conformal symmetry.

Grading turns one vector space into infinitely many finite layers

In a standard VOA,

V=⊕_n∈ZV_n, L₀v=nv for v∈V_n,

with suitable lower-boundedness and finite-dimensional homogeneous pieces.

The vacuum has weight 0 and the conformal vector has weight 2. If a has weight h, its mode a_n shifts conformal weight by h−n−1.

This degree rule is a strong bookkeeping device. Before calculating any mode action, one can often determine which graded subspace it must land in.

Derive the mode weight shift

The conformal covariance relation gives

[L₀,Y(a,z)]=Y(L₀a,z)+z dY(a,z)/dz.

If L₀a=ha and Y(a,z)=Σa_nz⁻ⁿ⁻¹, compare coefficients of z⁻ⁿ⁻¹ to obtain

[L₀,a_n]=(h−n−1)a_n.

Thus a_n sends a weight-r vector to weight r+h−n−1 when nonzero.

This is the infinite-dimensional analogue of using weights in a Lie-algebra representation to predict how raising and lowering operators move vectors.

The Heisenberg VOA: oscillator modes

The rank-one Heisenberg algebra has modes a_n satisfying

[a_m,a_n]=mδ_m+n,0I.

Positive modes annihilate the vacuum. Negative modes create states:

a_−n₁a_−n₂···a_−nᵣ1, nᵢ>0.

The conformal weight is n₁+···+nᵣ. Therefore the dimension of the weight-N subspace is the partition number p(N): each partition of N records the multiset of oscillator creation modes.

This gives a direct bridge from integer partitions to a graded infinite-dimensional representation.

First Heisenberg graded pieces

At weight 0 there is the vacuum 1. At weight 1 there is a₋₁1. At weight 2 there are two states:

a₋₂1;
a₋₁²1.

At weight 3 there are three partition patterns: 3, 2+1 and 1+1+1. The basis states are a₋₃1, a₋₂a₋₁1 and a₋₁³1.

The generating function for these dimensions is

∏_n≥1(1−qⁿ)⁻¹.

The algebraic oscillator construction has therefore produced the same partition generating function that appears throughout combinatorics.

Modules carry compatible vertex operators

A V-module M has its own state-field map

Y_M(a,z)=Σa_n^Mz⁻ⁿ⁻¹

satisfying module versions of the vacuum, locality/Jacobi and conformal relations.

For an ordinary grading-restricted module, L₀ acts with finite-dimensional eigenspaces and weights bounded below. The lowest conformal weight h need not be an integer.

This is how the Ising model later produces lowest weights 0, 1/2 and 1/16 even though the VOA itself is integer graded.

Heisenberg Fock modules carry continuous charge

For the rank-one Heisenberg VOA, one can construct a Fock module M_α with a highest vector |α⟩ satisfying

a₀|α⟩=α|α⟩, a_n|α⟩=0 for n>0.

Negative modes create oscillator descendants exactly as in the vacuum module. In the standard normalisation, the lowest conformal weight is α²/2.

The parameter α can vary continuously. Therefore this representation category is not a finite fusion category with only finitely many simple labels. It is a useful warning that “VOA module category” does not automatically mean “modular tensor category”.

Characters package graded dimensions

For a module M with central charge c and L₀ grading, the standard shifted character is

χ_M(τ)=Tr_Mq^L₀−c/24, q=e^2πiτ.

If M=⊕M_h+n, then

χ_M(τ)=q^h−c/24Σ_n≥0(dim M_h+n)qⁿ.

The leading exponent therefore reveals h−c/24, while the coefficients count graded dimensions. A character is more informative than total dimension because the modules are usually infinite-dimensional.

Heisenberg character and the Dedekind eta function

The rank-one Heisenberg VOA has central charge c=1. Its Fock module M_α has one partition basis over the highest vector. Therefore

χ_α(τ)=q^{α²/2−1/24}∏_n≥1(1−qⁿ)⁻¹.

Using η(τ)=q^1/24∏_n≥1(1−qⁿ), this becomes

χ_α(τ)=q^α²/2/η(τ).

The eta function is not a decorative rewrite. Its modular transformation properties help organise how characters behave under τ↦−1/τ.

Character equality does not automatically prove module equivalence

A character remembers L₀-graded dimensions and, when refined, traces of additional commuting operators. It does not generally remember every intertwining map or extension.

In semisimple rational settings, the vector of irreducible characters is an extremely strong invariant and transforms in a finite-dimensional modular representation. In non-semisimple logarithmic settings, ordinary characters can fail to distinguish modules and may need pseudotraces or other refined functions.

The same lesson appeared in category O and categorification: decategorified traces can be powerful without becoming complete classification data.

Fusion is controlled by intertwining operators

Given V-modules W₁,W₂,W₃, an intertwining operator of type

(W₃; W₁ W₂)

is a field-like map that combines a state from W₁ with a state from W₂ and produces a formal series with values in W₃, satisfying compatibility with the VOA action.

The dimension of the space of such intertwining operators is the fusion-rule coefficient

N_W₁,W₂^W₃.

In a suitable tensor-category setting one packages these universal intertwining maps into a tensor product W₁⊠W₂.

This tensor product is not generally the ordinary vector-space tensor product with componentwise modes. Its universal property is built from intertwining operators and the analytic geometry of insertion points.

Why the insertion point matters

In the Huang–Lepowsky theory, one first constructs P(z)-tensor products depending on a nonzero complex number z. Different insertion points are related by parallel transport, and associativity compares products formed in different convergence regions.

This is one reason VOA tensor categories require more analytic control than a finite fusion table. Formal series must be related to convergent multivalued analytic functions in appropriate domains.

A fusion coefficient can be finite and correct while a full braided tensor-category structure still requires substantial convergence, extension and associativity theorems.

Heisenberg fusion adds charge

In the standard Heisenberg Fock setting, the charge labels add:

M_α⊠M_β≈M_α+β

under the usual category and normalisation assumptions.

This resembles weight addition under tensor products of abelian group representations. The difference is that the modules are infinite-dimensional conformal objects with oscillator descendants above each charge.

Because α ranges continuously, this example again lies outside the finite semisimple fusion-category situation used by the Verlinde formula below.

The Ising VOA gives a finite worked fusion system

The Virasoro minimal model at central charge c=1/2 has three irreducible modules, with lowest conformal weights

1: h=0;
ε: h=1/2;
σ: h=1/16.

Their leading character powers are therefore

χ₁ begins q^−1/48(1+···);
χ_ε begins q^23/48(1+···);
χ_σ begins q^1/24(1+···).

Each exponent is h−c/24. For σ, 1/16−1/48=1/24, an immediate arithmetic check.

The Ising fusion rules

The simple-module fusion rules are

ε⊠ε≈1, ε⊠σ≈σ, σ⊠σ≈1⊕ε.

These are exactly the fusion rules analysed abstractly in the earlier Fusion Categories guide.

The VOA adds a source for that category. The objects are not merely labels satisfying a multiplication table; they are modules carrying vertex operators, conformal gradings and characters.

Thus the same fusion ring now has a concrete infinite-dimensional representation-theoretic realisation.

Quantum dimensions from fusion

Let d(1)=1. The equation ε²=1 and positivity give d(ε)=1. Then σ²=1+ε gives

d(σ)²=2,

so d(σ)=√2 in the unitary modular category.

This √2 is not the vector-space dimension of the VOA module σ, which is infinite-dimensional. It is the categorical dimension controlling tensor growth.

The distinction is especially important here because the same object simultaneously has infinite ordinary dimension, finite-dimensional graded pieces and irrational quantum dimension.

Characters transform under the modular group under strong hypotheses

For suitable rational and C₂-cofinite vertex operator algebras, the finite-dimensional span of irreducible characters is preserved by modular transformations such as

S:τ↦−1/τ, T:τ↦τ+1.

Zhu established foundational modular-invariance results under finiteness conditions, and Huang proved the Verlinde conjecture and modular tensor-category results under precise rationality, cofiniteness, self-duality and positivity hypotheses. [1–3]

The hypotheses belong to the theorem. An arbitrary vertex algebra can have infinitely many irreducible modules, non-semisimple representation theory or characters requiring more elaborate modular objects.

The Ising modular S-matrix

Order the simple modules as (1,ε,σ). A standard normalized modular S-matrix is

S = (1/2) [ 1    1    √2 ]
          [ 1    1   −√2 ]
          [ √2  −√2   0 ].

Direct multiplication gives S²=I because all three Ising simples are self-dual. The first row recovers quantum dimensions through dᵢ=S₀ᵢ/S₀₀, giving 1,1,√2.

The same matrix appeared in the abstract Fusion Categories article. Here it is also the matrix describing τ↦−1/τ on the vector of irreducible VOA characters, with compatible normalization.

Use Verlinde to recover σ fusion

For a normalized unitary modular S-matrix, the Verlinde formula is

N_ij^k=Σ_rS_irS_jroverline(S_kr)/S_0r.

Take i=j=σ. For output 1, the unit row cancels the denominator and gives (√2/2)²+(−√2/2)²=1. For output ε, the result is again 1. For output σ, the two nonzero terms cancel and give 0.

Therefore σ⊠σ=1⊕ε. A modular transformation of character functions has recovered an algebraic tensor-product multiplicity.

graded character functions → modular S-matrix → Verlinde formula → fusion rules.

Why modularity is a representation-theoretic return path

A character begins as a trace over one infinite-dimensional module. Modular invariance mixes the characters of different irreducible modules. The S-matrix describing that mixing simultaneously diagonalises the fusion algebra.

Thus three layers agree:

operator layer: modules carry vertex modes;
analytic layer: graded traces transform under SL₂(Z);
tensor layer: irreducible modules fuse with integer multiplicities.

The agreement is a theorem under the stated hypotheses, not a general rule for any collection of q-series with an invertible matrix.

Affine Lie algebras produce major VOA families

Start with a finite-dimensional simple Lie algebra 𝔤 and form its affine Kac–Moody algebra. At a chosen level k, a vacuum representation can be given a vertex-algebra structure; away from the critical level, a Sugawara construction supplies a conformal vector under standard assumptions.

At positive integral levels, integrable highest-weight modules produce finite families whose fusion rules are truncated versions of ordinary tensor-product rules.

This is a major bridge from the earlier highest-weight representation article to VOA tensor categories. The affine extension supplies infinitely many modes, while the level bounds which highest weights remain integrable.

Lattice VOAs turn discrete geometry into conformal algebra

Let L be a positive-definite even lattice. One can combine Heisenberg oscillators with group-algebra data from L to construct a lattice VOA V_L.

For suitable positive-definite even L, irreducible modules are indexed by cosets in the finite discriminant group L*/L. Fusion adds cosets.

The lattice inner product controls conformal weights and braiding phases. Thus Euclidean lattice geometry, finite abelian group data and conformal representation theory are linked in one construction.

This is another example where an apparently simple fusion law does not exhaust the theory: the same finite group of labels also carries quadratic-form and modular information.

The Monster example shows how large finite symmetry can sit inside a VOA

The moonshine vertex operator algebra V^♮ has the Monster finite simple group as its automorphism group. Its graded character is the modular function J(τ)=j(τ)−744, beginning

q⁻¹+196884q+···.

The coefficient 196884 decomposes into 1+196883, dimensions of the trivial and smallest nontrivial Monster representations. Monstrous moonshine connects this graded representation theory to modular functions.

The example is a landmark application, not a statement that every modular function comes from a VOA or that every finite group admits an analogous moonshine construction.

Twisted modules describe symmetry sectors

If g is an automorphism of a VOA, a g-twisted module allows vertex operators with fractional mode powers determined by the eigenvalues of g.

Twisted modules are central to orbifold theory, where one studies the fixed-point subalgebra V^G under a finite automorphism group and the sectors needed to reconstruct its representation theory.

The fractional modes are controlled by the automorphism; they are not arbitrary branch choices. Different twisted sectors encode different boundary or monodromy conditions.

Logarithmic modules show the boundary of semisimple theory

In logarithmic conformal field theory, L₀ can act non-semisimply. Generalized weight spaces contain Jordan blocks, and correlation functions can acquire logarithmic terms.

The module category is then non-semisimple. Fusion products can involve indecomposables and extensions, ordinary quantum dimensions may behave differently, and the classical semisimple Verlinde formula requires modification.

This mirrors Modular Representation Theory: once semisimplicity fails, composition factors are no longer the whole object.

A verification workflow

1. State the VOA convention. Record the mode indexing in Y(a,z).
2. Identify the vacuum and conformal vector. Verify their weights and translation relation.
3. Check one locality or mode commutator. Do not infer an OPE from notation alone.
4. Use L₀ to control degrees. Confirm where each mode can send a homogeneous vector.
5. For a module, record the lowest conformal weight. It determines the leading character exponent together with c.
6. Separate ordinary tensor product from VOA fusion. Fusion is built through intertwining operators.
7. State rationality and cofiniteness hypotheses before using a finite modular S-matrix.
8. Check S normalization. Verlinde denominators depend on it.
9. Recompute a fusion coefficient from S. This returns analytic character data to algebra.
10. Distinguish semisimple from logarithmic settings. Characters and fusion can require different tools.

Common mistakes

Calling Y(a,z) one operator: it is a generating field whose coefficients are infinitely many modes.
Treating formal locality as an ordinary numerical limit: the initial identities live in formal series.
Ignoring the c/24 shift: modular characters use q^L₀−c/24.
Calling fusion an ordinary vector-space tensor product: VOA fusion uses intertwining operators and tensor-category constructions.
Assuming every VOA has finitely many simples: the Heisenberg example already has continuously labelled Fock modules.
Using Verlinde without hypotheses: rationality, cofiniteness and related conditions matter.
Confusing quantum dimension with ordinary module dimension: an Ising σ module is infinite-dimensional as a vector space but has quantum dimension √2.
Assuming every non-semisimple theory is covered by the classical modular tensor category theorem: logarithmic theories require modified structures.

Practice questions

1. What does the coefficient a_n mean in Y(a,z)? 2. State the Virasoro commutator. 3. If a has weight h, by how much does a_n shift conformal weight? 4. How many rank-one Heisenberg vacuum states have weight 4?

5. What is the leading q-exponent for an Ising module of weight 1/16? 6. What does a fusion coefficient count? 7. Why is Heisenberg fusion not a finite fusion category? 8. Compute d(σ) from σ²=1+ε.

9. Use the Ising S-matrix to find N_σσ^σ. 10. What is the role of C₂-cofiniteness/rationality assumptions? 11. Why can two modules with the same ordinary character still require further analysis in a non-semisimple setting? 12. What extra data does a VOA provide beyond an abstract fusion ring?

Worked answers

1–4. Modes and grading

1. a_n is the linear operator multiplying z⁻ⁿ⁻¹ in the field Y(a,z). Acting on b gives the state a_nb.

2. [L_m,L_n]=(m−n)L_m+n+[c/12](m³−m)δ_m+n,0I.

3. By h−n−1. This follows from [L₀,a_n]=(h−n−1)a_n.

4. p(4)=5, corresponding to partitions 4, 3+1, 2+2, 2+1+1 and 1+1+1+1.

5–8. Characters and fusion

5. c=1/2, so c/24=1/48. Therefore 1/16−1/48=1/24.

6. It is the dimension of the relevant space of intertwining operators, equivalently the multiplicity of the output simple in a semisimple fusion product.

7. The Fock modules M_α are parametrised by a continuous charge α, so there are infinitely many simple labels rather than a finite set.

8. d(ε)=1, so d(σ)²=1+1=2. Positivity gives d(σ)=√2.

9–12. Modularity and retained structure

9. The two nonzero contributions cancel, giving 0. Therefore σ does not appear in σ⊠σ.

10. They provide finiteness and semisimplicity conditions needed for a finite-dimensional modular character representation and, with the additional hypotheses in Huang’s theorem, a modular tensor category of modules.

11. A character records graded traces but can forget extension or Jordan-block information. Non-semisimple categories may require pseudotraces or other refined invariants.

12. The VOA provides actual infinite-dimensional modules, mode operators, a conformal grading, intertwining operators, character functions and analytic compatibility. A fusion ring retains only the integer tensor multiplicities after much of that structure has been forgotten.

How to teach vertex operator algebras through returns

Begin with the Heisenberg oscillator because its basis is visible. Ask the learner to list weight-3 and weight-4 creation-mode states and recognise partition numbers. This establishes why an infinite-dimensional module can still have finite graded pieces.

Then introduce Y(a,z) as a machine that stores all modes. Use L₀ to predict degree shifts before calculating. Only after the grading is secure should the Virasoro central term and formal locality be added.

Finish with the Ising model. Compute the three leading character exponents, write the S-matrix and recover σ²=1+ε with Verlinde. The learner then sees a full cycle: operator algebra → modules → characters → modular transformation → fusion.

Sources and further study

[1] Yi-Zhi Huang, Lecture Notes on Vertex Operator Algebras and Tensor Categories, for modules, intertwining operators, tensor products and the modular-category framework. [2] Yi-Zhi Huang, Vertex Operator Algebras and the Verlinde Conjecture, for the Verlinde theorem under precise hypotheses. [3] Huang, Vertex Operator Algebras, the Verlinde Conjecture and Modular Tensor Categories, for rigidity and nondegeneracy consequences. [4] David Ben-Zvi and Edward Frenkel, Geometric Realization of the Segal–Sugawara Construction, for affine and conformal vertex-algebra context. The Ising S-matrix and fusion calculations above are checked explicitly in the normalization stated.

Representation Mathematics — Batch 07

Study canonical Hecke-basis multiplicities in Kazhdan–Lusztig Theory, nilpotent flag geometry in Springer Correspondence, and dual-group reconstruction in Geometric Satake. Return to the BTT Mathematics Learning Hub.

Small Group Tutorials

Vertex Operator Algebras | Modules, Characters, Fusion and Conformal Symmetry