---
title: Finite Index Conformal Embeddings
url: https://www.emergentmind.com/topics/conformal-embeddings-of-finite-index
type: topic
---

# Finite Index Conformal Embeddings

Conformal embeddings of finite index are embeddings of chiral conformal structures in which the smaller theory and the larger theory share the same stress tensor, while the larger object restricts to only finitely many irreducible sectors over the smaller one. In affine vertex operator algebras this means equality of the relevant Sugawara Virasoro vectors together with a finite decomposition of the ambient affine VOA into modules for the embedded affine subalgebra; in tensor-categorical and conformal-net formulations it is encoded by étale algebras or Q-systems and by finite Jones index. The subject therefore sits at the intersection of affine Kac–Moody theory, branching rules, fusion rules, simple-current extensions, orbifold methods, and subfactor theory [1101.4484] [1210.6602] [2112.15396] [2503.13641].

## 1. Definition and conformality criteria

For a simple finite-dimensional Lie algebra $\mathfrak g$ with dual Coxeter number $h^\vee$ and level $k \neq -h^\vee$, the universal affine VOA $V_{\mathfrak g}(k)=N_{\mathfrak g}(k,0)$ carries the Sugawara Virasoro vector with central charge
\[
c_{\mathfrak g}(k)=\frac{k\,\dim\mathfrak g}{k+h^\vee}.
\]
If $\mathfrak g' \subset \mathfrak g$ is reductive with simple summands $\mathfrak g'_1,\dots,\mathfrak g'_r$, and if $k'_i=a_i k$ is the induced level on each simple summand, then
\[
c_{\mathfrak g'}(k')=\sum_{i=1}^r \frac{k'_i\,\dim\mathfrak g'_i}{k'_i+h_i^\vee},
\]
with zero contribution from abelian factors. Adamović–Perše give a general criterion: the embedding
\[
V_{\mathfrak g'}(k') \hookrightarrow V_{\mathfrak g}(k)
\]
is conformal if and only if two conditions hold, namely $c_{\mathfrak g}(k)=c_{\mathfrak g'}(k')$ and, writing $\mathfrak g=\mathfrak g' \oplus W$ as a $\mathfrak g'$-module, every weight space of $W$ has conformal weight $1$ under the $V_{\mathfrak g'}(k')$ Sugawara action. Equivalently,
\[
L^{\mathfrak g'}(0)\bigl(x(-1)\mathbf1\bigr)=x(-1)\mathbf1 \qquad \text{for all } x\in W
\]
[1101.4484].

A more computational form of the same criterion appears in the Adamović–Perše condition for reductive $\mathfrak k \subset \mathfrak g$. If the complement $\mathfrak p=\mathfrak g \ominus \mathfrak k$ decomposes as $\mathfrak p=\bigoplus_i M_i$, and if $\gamma_{ij}$ is the $\mathfrak k_j$-Casimir eigenvalue on $M_i$, then conformality is equivalent to
\[
\sum_j \frac{\gamma_{ij}}{2\,(k+h_j^\vee)}=1
\]
for every $i$. In the equivalent highest-weight form,
\[
\sum_j \frac{(\mu_{ij},\mu_{ij}+2\rho_j)}{2\,(k+h_j^\vee)}=1
\]
for every $i$ [1702.06089].

In several important classification problems this general criterion collapses to central-charge matching alone. For maximal equal-rank reductive subalgebras $\mathfrak g^0 \subset \mathfrak g$, one has
\[
V_{k'}(\mathfrak g^0)\hookrightarrow V_k(\mathfrak g)\ \text{conformal}\iff c_k(\mathfrak g)=c_{k'}(\mathfrak g^0),
\]
and in the classical maximal setting the finite-decomposition property is likewise equivalent to the same central-charge equality [1509.06512] [1210.6602].

The superalgebra analogue is formally similar. For a basic classical simple Lie superalgebra $\mathfrak g$ with even part $\mathfrak g_{\bar 0}$,
\[
c(\mathfrak g,k)=\frac{k\cdot \operatorname{sdim}\mathfrak g}{k+h^\vee},
\]
and conformality of $V_k(\mathfrak g_{\bar 0}) \subset V_k(\mathfrak g)$ is characterized by $c(\mathfrak g,k)=c(\mathfrak g_{\bar 0},k)$; solving this equation yields a complete list of conformal levels for the basic classical families [1903.03794].

## 2. Meanings of finite index

In affine VOA theory, finite index is first of all a statement about restriction. If $V_{\mathfrak g}(k)$ is completely reducible as a $V_{\mathfrak g'}(k')$-module, then
\[
V_{\mathfrak g}(k)=\bigoplus_{\lambda\in P^+_{k'}} M_\lambda\,L_{\mathfrak g'}(k',\lambda),
\]
where $P^+_{k'}$ is the finite set of level-$k'$ dominant weights and $M_\lambda \in \mathbb N$ is the multiplicity. The index is then defined as
\[
[\,V_{\mathfrak g}(k):V_{\mathfrak g'}(k')\,]=\sum_{\lambda\in P^+_{k'}} M_\lambda<\infty.
\]
Equivalently,
\[
[\,V_{\mathfrak g}(k):V_{\mathfrak g'}(k')\,]
=
\sum_{\lambda\in P^+_{k'}} M_\lambda \bigl(\mathrm{qdim}\,L_{\mathfrak g'}(k',\lambda)\bigr)^2
=
\frac{\mathrm{Dim}_q\,V_{\mathfrak g}(k)}{\mathrm{Dim}_q\,V_{\mathfrak g'}(k')}.
\]
In many finite-index conformal embeddings all $M_\lambda$ are $0$ or $1$ and all quantum dimensions are $1$, so the index reduces to the number of summands [1101.4484].

A related tensor-categorical formulation uses the étale algebra object attached to a conformal embedding. For
\[
\mathcal V(\mathfrak g_1,k_1)\subset \mathcal V(\mathfrak g_2,k_2),
\]
Huang–Kirillov–Lepowsky or Xu theory produces an étale algebra $A$ in $\operatorname{Rep}(\mathcal V(\mathfrak g_1,k_1))$, and the extension size is measured by
\[
\operatorname{Index}\bigl(\mathcal V(\mathfrak g_1,k_1)\subset \mathcal V(\mathfrak g_2,k_2)\bigr)
=
\dim(A)
=
\sum_{X\subset A}(\dim_q X)^2.
\]
The category of $A$-modules is equivalent to $\operatorname{Rep}(\mathcal V(\mathfrak g_2,k_2))$ [2503.13641].

In conformal nets, finite index means finite Jones index of the local inclusion $A(I)\subset B(I)$, equivalently the existence of a faithful normal conditional expectation satisfying the Pimsner–Popa bound. Gui’s Connes-fusion approach identifies finite-index extensions with Q-systems in $\operatorname{Rep}(A)$, and the numerical index is
\[
[\,B_Q(\mathbb I):A(\mathbb I)\,]=d(H_a)^2,
\]
where $d(H_a)$ is the quantum dimension of the underlying object of the Q-system [2112.15396].

A distinct convention also appears in the affine literature: one source writes
\[
I(\varphi)=\frac{c_k(\mathfrak g)}{c_k(\mathfrak k)}
\]
or gives a level-ratio expression for $I(\varphi)$; for a conformal embedding this equals $1$. This use of “index” measures conformality rather than the size of a branching decomposition. The coexistence of these conventions is a persistent feature of the subject [1702.06089].

| Framework | Finite-index datum | Formula |
|---|---|---|
| Affine VOA restriction | Number of summands / quantum-dimension sum | $\sum_\lambda M_\lambda$ or $\sum_\lambda M_\lambda(\mathrm{qdim}\,L_\lambda)^2$ |
| Étale algebra in a modular tensor category | Algebra dimension | $\dim(A)=\sum_{X\subset A}(\dim_q X)^2$ |
| Conformal net extension | Jones index | $[B_Q(\mathbb I):A(\mathbb I)]=d(H_a)^2$ |

## 3. Affine Lie algebra cases: classification and explicit finite decompositions

For maximal conformal embeddings of reductive $\mathfrak l$ into semisimple $\mathfrak g$ of classical type,
\[
V_k(\mathfrak l)\hookrightarrow V_1(\mathfrak g),
\]
the finite-index condition is equivalent to the central-charge identity
\[
c_{\mathfrak l}(k)=c_{\mathfrak g}(1).
\]
Kac–Möseneder–Papi–Xu describe the classical cases and compute the corresponding simple-current groups that measure the index. Examples include diagonal and adjoint-type families, as well as $\mathfrak{so}(m)\times\mathfrak{so}(n)\subset\mathfrak{so}(mn)$ and $\mathfrak{sl}(p)\oplus\mathfrak{sl}(q)\oplus\mathbb C\subset\mathfrak{sl}(p+q)$, where the order of the simple-current subgroup can be a $\gcd$-quantity such as $\gcd(p+1,q+1)$ [1210.6602].

For maximal equal-rank reductive subalgebras $\mathfrak g^0\subset\mathfrak g$, the complete classification shows a sharp dichotomy. At the distinguished level $k=1$ with $\mathfrak g^0$ semisimple, the decomposition of $V_1(\mathfrak g)$ as a $V_1(\mathfrak g^0)$-module is always finite, multiplicity-free, and a simple-current extension. Outside $k=1$, the only additional conformal levels with finite decomposition are:
\[
C_n \text{ with } C_h\oplus C_{n-h}\text{ at }k=-\tfrac12,\qquad
B_n \text{ with } B_h\oplus B_{n-h}\text{ at }k=-n+\tfrac32,
\]
\[
G_2 \text{ with } A_2\text{ at }k=-\tfrac53,\qquad
F_4 \text{ with } B_4\text{ at }k=-\tfrac52.
\]
All other reductive but non-semisimple equal-rank cases have infinite decomposition [1509.06512].

The non-equal-rank non-integrable classification adds further negative-rational and negative-integer levels. The finite-index list includes
\[
\mathfrak{sl}(n)\times\mathfrak{sl}(m)\hookrightarrow\mathfrak{sl}(nm)\ \text{at }k=-1,\qquad
\mathfrak{sp}(2n)\times\mathfrak{so}(m)\hookrightarrow\mathfrak{sp}(2nm)\ \text{at }k=-\tfrac12,
\]
\[
\mathfrak{so}(n)\times\mathfrak{so}(m)\hookrightarrow\mathfrak{so}(nm)\ \text{at }k=-1,
\]
together with the spin cases $B_3\hookrightarrow D_4$ and $G_2\hookrightarrow B_3$ at $k=-2$, and exceptional embeddings at Deligne-series levels such as $k=-6,-4,-3,-\tfrac52,-\tfrac53$ [1702.06089].

Representative finite decompositions illustrate the range of possibilities. Adamović–Perše exhibit both unitary and non-unitary examples:
\[
V_{D_{n+1}}(1)=V_{B_n}(1)\oplus L_{B_n}(1,\omega_1),
\]
so the index is $2$;
\[
V_{A_2}(1)\cong V_{A_1}(1)\oplus L_{A_1}(1,4\omega_1),
\]
again of index $2$;
\[
V_{E_6}(-3)=V_{F_4}(-3)\oplus L_{F_4}(-3,\omega_4),
\]
also of index $2$; and for $A_2\supset A_1$ at admissible level $-\tfrac53$, and for $G_2\supset A_2$ at the same level, the index is $3$ [1101.4484].

At level $1$, the decomposition formulas can be highly structured. Two standard examples are
\[
V_1(G_2)\cong V_1(A_2,0)\oplus V_1(A_2,\omega_1)\oplus V_1(A_2,\omega_2)
\]
and
\[
V_1(E_8)\cong\bigoplus_{i=0}^4 V_1(A_4,\omega_i)\otimes V_1(A_4,\omega_{4-i}),
\]
both multiplicity-free simple-current extensions [1509.06512].

| Embedding | Decomposition | Index |
|---|---|---|
| $V_{B_n}(1)\subset V_{D_{n+1}}(1)$ | $V_{D_{n+1}}(1)=V_{B_n}(1)\oplus L_{B_n}(1,\omega_1)$ | $2$ |
| $V_{A_1}(1)\subset V_{A_2}(1)$ | $V_{A_2}(1)\cong V_{A_1}(1)\oplus L_{A_1}(1,4\omega_1)$ | $2$ |
| $V_{F_4}(-3)\subset V_{E_6}(-3)$ | $V_{E_6}(-3)=V_{F_4}(-3)\oplus L_{F_4}(-3,\omega_4)$ | $2$ |
| $V_1(A_2)\subset V_1(G_2)$ | $V_1(G_2)\cong V_1(A_2,0)\oplus V_1(A_2,\omega_1)\oplus V_1(A_2,\omega_2)$ | $3$ |

A recurrent misconception is that central-charge equality by itself always guarantees a finite decomposition. The classification results show that this is true in specific settings, notably maximal equal-rank embeddings and the classical maximal level-$1$ families, but not as an unrestricted general principle [1509.06512] [1210.6602].

## 4. Vertex superalgebras and finite non-simple-current extensions

For affine vertex superalgebras, the first step is the classification of conformal levels for the embedding of the even part:
\[
V_k(\mathfrak g_{\bar 0})\subset V_k(\mathfrak g).
\]
The complete list includes the families $\mathfrak{sl}(m|n)$, $\mathfrak{psl}(m|m)$, $B(m,n)=\operatorname{osp}(2m+1|2n)$, $D(m,n)=\operatorname{osp}(2m|2n)$, $C(n+1)=\operatorname{osp}(2|2n)$, $F(4)$, $G(3)$, and $D(2,1;a)$, with explicit conformal levels such as $k=\pm1$ in several cases and distinguished non-unitary values like $k=-2$ for $F(4)$ and $k=-3$ for $G(3)$ [1903.03794].

Conformality does not imply finite index in the super setting either. For $g=\mathfrak{sl}(m|n)$ at $k=1$ with $m\neq n,n-2$, the decomposition of $V_1(\mathfrak{sl}(m|n))$ over
\[
\mathbb C\cdot \omega \oplus \mathfrak{sl}(m)\oplus \mathfrak{sl}(n)
\]
is an infinite direct sum indexed by $q\in\mathbb Z$. Likewise, for $g=C(n+1)=\operatorname{osp}(2|2n)$ at $k=1$, the decomposition over $\mathbb C\cdot\omega\oplus \mathfrak{sp}(2n)$ is again infinite [1903.03794].

Finite-index super examples divide into simple-current and non-simple-current types. For $g=\mathfrak{psl}(m|m)$ at $k=1$,
\[
V_1(\mathfrak{psl}(m|m))
=
\bigoplus_{j=0}^{m-1}
L_{\mathfrak{sl}(m)}(1,j\omega_1)\otimes
L_{\mathfrak{sl}(m)}(-1,(m-j)\omega_1),
\]
and the index is $m$. For $B(m,n)$ and $D(m,n)$ at $k=1$ or at the other simple-current level, one obtains two-term decompositions [1903.03794].

The most distinctive superalgebra phenomena occur in finite non-simple-current extensions. The paper isolates four cases:
\[
g=\operatorname{spo}(2|3),\ k=-\tfrac34,\ \text{index }4;
\qquad
g=F(4),\ k=1,\ \text{index }3;
\]
\[
g=G(3),\ k=1,\ \text{index }4;
\qquad
g=\operatorname{osp}(2n+8|2n),\ k=-2,\ \text{index }n+1.
\]
For the last family,
\[
V_{-2}(\operatorname{osp}(2n+8|2n))
\cong
\bigoplus_{i=0}^{n}
L_{so(2n+8),-2}(i\omega_1)\otimes L_{sp(2n),1}(\omega_i),
\]
so the extension is finite but not a simple-current extension [1903.03794].

This superalgebraic sector is important because it demonstrates that finite-index conformal embeddings are not exhausted by multiplicity-free simple-current behavior. The $\operatorname{osp}(2n+8|2n)_{-2}$ decomposition is explicitly described as a finite, non simple current extension [1903.03794].

## 5. Fusion rules, orbifolds, semisimplicity, and intermediate extensions

A central technical problem is not merely to identify conformal levels, but to prove that the corresponding restriction is semisimple and finite. The standard strategy in the affine VOA setting is modular. One first checks that the complementary summands in $\mathfrak g\ominus\mathfrak g'$ all have conformal weight $1$ under the smaller Sugawara action. One then realizes $\mathfrak g'$ as the fixed-point subalgebra of a finite-order automorphism of $\mathfrak g$, extends this automorphism to $V_{\mathfrak g}(k)$, and decomposes the latter into orbifold sectors. Fusion rules for $V_{\mathfrak g'}(k')$, together with tensor-product decompositions of the finite-dimensional $\mathfrak g'$-modules appearing in $\mathfrak g\ominus\mathfrak g'$, are then used to show closure on a finite list of highest-weight modules. A standard “no-new-singular-vector” argument eliminates further constituents, and simplicity together with orbifold-theory results such as Dong–Mason ’97 yields semisimplicity of the decomposition [1101.4484].

In the classical maximal setting, this analysis becomes especially rigid. If
\[
V_k(\mathfrak l)\subset W\subset V_1(\mathfrak g)
\]
is an intermediate simple VOA with the same Virasoro vector, then every such $W$ is either the full $V_1(\mathfrak g)$ or a simple-current extension of $V_k(\mathfrak l)$. Equivalently,
\[
W=\bigoplus_{x\in H} W_x
\]
for a finite subgroup $H$ of the simple-current subgroup
\[
S_R\bigl(V_1(\mathfrak g),V_k(\mathfrak l)\bigr),
\]
and the set of intermediate simple VOAs is in bijection with the set of subgroups $H\subset S_R$ [1210.6602].

The decomposition itself can often be organized by classical invariant theory. In the symmetric-pair realization used for classical type, the irreducible $V_k(\mathfrak l)$-summands correspond to Borel-stable abelian subspaces $\mathfrak I\subset \mathfrak p$, and one obtains an explicit formula
\[
V_1(\mathfrak g)\cong \bigoplus_{\mathfrak I\in\Sigma} L_{\mathfrak l}\bigl(\Lambda(\mathfrak I)\bigr)
\]
[1210.6602].

The contrast with infinite-decomposition phenomena is equally important. In the study of $A_1\times D_4\hookrightarrow C_8$ at $k=-\tfrac12$, subsingular vectors appear in the relevant Weyl-algebra realization. The resulting decomposition of the free-field ambient module is not semisimple and exhibits an “infinite-index”-type behavior. This shows that free-field realizations can detect obstructions not visible from central charges alone [1702.06089].

## 6. Conformal nets, Q-systems, and diagrammatic interpolation

In the conformal-net formulation, an irreducible local Möbius covariant net $A$ on $S^1$ assigns a type $\mathrm{III}_1$ factor $A(I)$ to each interval and carries isotony, locality, Möbius covariance, and a vacuum vector. An extension $A\subset B$ is of finite Jones index when each inclusion $A(I)\subset B(I)$ has finite index, equivalently when there exists a faithful normal conditional expectation satisfying the Pimsner–Popa bound. Connes fusion turns $\operatorname{Rep}(A)$ into a braided rigid $C^*$-tensor category, and Gui proves that finite-index extensions are classified by Q-systems
\[
Q=(H_a,\mu,\eta)
\]
in $\operatorname{Rep}(A)$ [2112.15396].

The basic structural theorems are exact. Given a Q-system, one constructs an extension $A\subset B_Q$ of finite index, with local algebras generated by left-fusion operators and commutants generated by right-fusion operators. Conversely, for any extension $A\subset B$, the following are equivalent: the underlying object $H_a$ is dualizable in $\operatorname{Rep}(A)$; each $A(\mathbb I)\subset B(\mathbb I)$ has finite Jones index; and there exists a unique Q-system with underlying object $H_a$ such that $B=B_Q$. Local Möbius or conformal covariance of the extension is characterized by commutativity of the Q-system and trivial twist,
\[
\mu\circ \varepsilon_{a,a}=\mu,
\qquad
\theta_{H_a}=\mathrm{id},
\]
and then
\[
[B_Q(\mathbb I):A(\mathbb I)]=d(H_a)^2
\]
[2112.15396].

A canonical example is the Longo–Rehren Q-system built from a dualizable object $H_i$:
\[
Q=(H_i\boxtimes H_{\bar i},\ \mu=1\boxtimes ev\boxtimes 1,\ \eta=\mathrm{coev}).
\]
It is commutative and irreducible, and the resulting local Möbius extension has index
\[
[B_Q(\mathbb I):A(\mathbb I)]=d(H_i)^4
\]
[2112.15396].

Recent work also places finite-index conformal embeddings into a diagrammatic and interpolative framework. For the family
\[
\mathcal V(\mathfrak{sl}_N,N)\subset \mathcal V(\mathfrak{so}_{N^2-1},1),
\]
the category of local modules is described by a two-color skein theory with a black strand for the defining representation $X$, a red invertible strand $g$, and a trivalent vertex
\[
X\otimes X^*\to g.
\]
The defining coefficients are rational functions in
\[
q=e^{2\pi i/(4N)},
\]
or more generally in the ring
\[
R=\mathbb C[q,q^{-1},(q-q^{-1})^{-1}],
\]
which permits specialization to a continuous family of categories $\mathcal E_t$ for $t\in\mathbb C\setminus\{-1,0,1\}$. At the special roots $t=e^{2\pi i/(4N)}$, semisimplification reproduces the module category of the conformal embedding. In this case the index is computed combinatorially as
\[
\operatorname{Index}\bigl(\mathfrak{sl}_N{}_N\subset \mathfrak{so}_{N^2-1}{}_1\bigr)=2^{N-1},
\]
and the Jones–Kosaki index of the corresponding subfactor is likewise $2^{N-1}$ [2503.13641].

The same source lists four infinite discrete series of finite-index conformal embeddings, including the $\mathfrak{sl}_N,N\pm2$ and $\mathfrak{sl}_N,N\subset \mathfrak{so}_{N^2-1},1$ families, and states the expectation that analogous Deligne-interpolation methods should apply to the remaining classical $A,B,C,D$ series. This suggests that finite-index conformal embeddings admit a common skein-theoretic and tensor-categorical organization beyond the individual branching calculations [2503.13641].

Source: https://www.emergentmind.com/topics/conformal-embeddings-of-finite-index