Graph Planar Algebra Embedding Theorem

• Graph Planar Algebra Embedding Theorem is a realization theorem in subfactor theory that embeds finite depth subfactor planar algebras into bipartite graph planar algebras, preserving grading, *-structure, positivity, and the tracial state.
• The theorem transforms abstract standard invariants into concrete loop-based models, enabling explicit computations, combinatorial constructions, and obstruction arguments in classification problems.
• It generalizes the classical principal graph embedding to include fusion graph targets, unifying planar algebra, tensor-categorical, and operator-algebraic approaches within a combinatorial framework.

Searching arXiv for the cited papers and closely related work to ground the article. The Graph Planar Algebra Embedding Theorem is a realization theorem in subfactor theory stating that a finite depth subfactor planar algebra can be embedded, as a shaded planar $*$-algebra, into the bipartite graph planar algebra of an associated graph. In its original form, the target graph is the principal graph of the subfactor planar algebra; in its later generalization, the target may be the fusion graph of any cyclic pivotal $C^*$-module over the corresponding projection category. The theorem converts an abstract standard invariant into a concrete loop-based planar algebra model, preserving grading, the $*$-structure, positivity, and the tracial state, and thereby provides a combinatorial framework for construction, computation, and obstruction arguments in finite depth subfactor theory (Jones et al., 2010, Coles et al., 2018).

1. Statement of the theorem and its variants

In the finite depth setting, let $P$ be a finite depth subfactor planar algebra of modulus $\delta>0$, assumed to be a shaded spherical $C^*$-planar algebra. Let $\Gamma$ be the principal graph of $P$. Then there exists an injective planar algebra homomorphism

$\phi:P\to G(\Gamma),$

into the bipartite graph planar algebra of $\Gamma$, making $C^*$0 a planar subalgebra. Equivalently, for each $C^*$1 and shading $C^*$2, $C^*$3 restricts to an injective $C^*$4-homomorphism of finite-dimensional $C^*$5-algebras

$C^*$6

intertwining all planar tangle operations. The map preserves grading, the $C^*$7-structure, positivity, and the tracial state. The modulus is $C^*$8, and the index satisfies $C^*$9 (Jones et al., 2010).

A later reformulation replaces the principal graph target by a more general fusion graph target. Let $*$0 be a finite depth subfactor planar algebra, and let $*$1 be an indecomposable finitely semisimple pivotal right module $*$2 category over the projection category of $*$3, with simple basepoint $*$4. Then $*$5 embeds into the bipartite graph planar algebra of the fusion graph of the cyclic module $*$6. In this form, the principal graph embedding is recovered by choosing the cyclic module given by the planar module consisting of the $*$7 tower itself; choosing the dual tower recovers the embedding into the dual principal graph (Coles et al., 2018).

The generalized statement is strictly broader than the principal-graph version. It covers embeddings into graph planar algebras associated to graphs that are not the principal or dual principal graphs, provided they arise as fusion graphs of cyclic pivotal modules. This explains why constructions associated with graphs used in Haagerup–Izumi or extended Haagerup settings fit naturally into the graph planar algebra framework (Coles et al., 2018).

2. Strongly Markov inclusions and the canonical relative commutant planar algebra

The operator-algebraic input is a strongly Markov inclusion

$*$8

of finite von Neumann algebras. Two conditions characterize this setting in the formulation used for the embedding theorem: the canonical semifinite trace $*$9 on the basic construction $P$0 is finite and normalized so that $P$1, and there exists a Pimsner–Popa basis $P$2 for $P$3 over $P$4, equivalently

$P$5

Such a basis satisfies

$P$6

for all $P$7, and the Watatani index is

$P$8

independent of the choice of basis (Jones et al., 2010).

The associated Jones tower

$P$9

is built inductively by $\delta>0$0, with

$\delta>0$1

The projections $\delta>0$2 satisfy the Temperley–Lieb–Jones relations

$\delta>0$3

Equivalently, with $\delta>0$4,

$\delta>0$5

The Markov normalization is therefore identical to the planar algebra loop parameter (Jones et al., 2010).

From a strongly Markov inclusion one defines a canonical shaded planar $\delta>0$6-algebra by relative commutants:

$\delta>0$7

where $\delta>0$8 is the identification

$\delta>0$9

with $C^*$0. Multiplication is induced by product in $C^*$1, the $C^*$2-structure is reversal of tensors, inclusions add vertical strings, and capping operations are implemented by conditional expectations. In particular, the commutant expectation has the explicit formula

$C^*$3

independent of the chosen Pimsner–Popa basis (Jones et al., 2010).

This relative-commutant planar algebra is the algebraic core of the embedding theorem. It supplies an intrinsic planar algebra attached to the Jones tower before any graph model is introduced.

3. Bipartite graph planar algebras and Perron–Frobenius normalization

For a connected unital inclusion of finite-dimensional $C^*$4-algebras $C^*$5 with the Markov trace, the Bratteli diagram $C^*$6 is a finite connected bipartite multigraph whose even and odd vertices index simple summands of $C^*$7 and $C^*$8. If $C^*$9 is the bipartite adjacency matrix, the Markov trace vectors $\Gamma$0 and $\Gamma$1 satisfy

$\Gamma$2

with $\Gamma$3. Equivalently, if

$\Gamma$4

then the Perron–Frobenius eigenvalue of $\Gamma$5 is $\Gamma$6, and one uses a positive Perron–Frobenius eigenvector $\Gamma$7 with $\Gamma$8; in the normalization of the theorem one writes

$\Gamma$9

as the positive eigenvector encoding the trace weights (Jones et al., 2010).

The bipartite graph planar algebra $P$0, also denoted $P$1, is a shaded planar $P$2-algebra whose box spaces are spanned by based loops of length $P$3 in $P$4. For the even shading,

$P$5

where the loop starts at an even vertex; $P$6 is defined analogously from odd vertices. Multiplication is path concatenation with matching conditions at the seam, and the adjoint is loop reversal:

$P$7

The action of tangles is computed by summing over states and multiplying by local correction factors

$P$8

so that contractible closed loops evaluate to $P$9 (Jones et al., 2010).

The Temperley–Lieb structure is built directly into the graph model. Jones projections are realized by cap-cup elements, and in algebraic form satisfy

$\phi:P\to G(\Gamma),$0

The trace on loop basis elements is diagonal with Perron–Frobenius weights. For loops $\phi:P\to G(\Gamma),$1,

$\phi:P\to G(\Gamma),$2

and the loop basis carries the Markov trace after the appropriate Perron–Frobenius normalization (Jones et al., 2010).

A central identification theorem states that the canonical relative-commutant planar algebra of a strongly Markov inclusion of finite-dimensional $\phi:P\to G(\Gamma),$3-algebras is isomorphic to the bipartite graph planar algebra of the Bratteli diagram of the inclusion. In the principal-graph embedding theorem, that Bratteli diagram is the principal graph of the finite depth subfactor planar algebra; in the module embedding theorem, it becomes the fusion graph of a cyclic module (Jones et al., 2010, Coles et al., 2018).

4. Construction of the embedding

The original proof proceeds in three steps. First, one chooses a finite shift $\phi:P\to G(\Gamma),$4 large enough so that the shifted inclusion

$\phi:P\to G(\Gamma),$5

is standard; finite depth guarantees the existence of such a shift. Setting

$\phi:P\to G(\Gamma),$6

one forms the canonical relative-commutant planar algebra $\phi:P\to G(\Gamma),$7 attached to the strongly Markov inclusion $\phi:P\to G(\Gamma),$8 (Jones et al., 2010).

Second, one defines the embedding map by adding strings on the left. In the principal-graph form, the map

$\phi:P\to G(\Gamma),$9

is given by adding $\Gamma$0 strings on the left for $\Gamma$1 and $\Gamma$2 strings on the left for $\Gamma$3. In the generalized categorical form, if $\Gamma$4 is a cyclic pivotal module category over the projection category of $\Gamma$5, one constructs the tower

$\Gamma$6

chooses $\Gamma$7 such that $\Gamma$8 is strongly Markov, and defines

$\Gamma$9

Diagrammatically, this “adds $C^*$00 alternating strings on the left” together with the module strand corresponding to $C^*$01 (Coles et al., 2018).

Third, one verifies that $C^*$02 intertwines the planar algebra structure. In the finite depth principal-graph case, $C^*$03 preserves multiplication, the $C^*$04-structure, Jones projections, inclusions, and both ordinary and commutant conditional expectations. Injectivity follows from trace-preservation and positivity on finite-dimensional box spaces: if $C^*$05, then

$C^*$06

hence $C^*$07 (Jones et al., 2010).

In the module form, the verification uses the tower description together with the fact that left capping is averaging over a Pimsner–Popa basis:

$C^*$08

The key identity is that the map $C^*$09 commutes with left capping after choosing Pimsner–Popa bases adapted to the compression and shift. The resulting planar $C^*$10-algebra morphism lands in the canonical relative-commutant planar algebra $C^*$11, and composing with a non-canonical planar $C^*$12-isomorphism

$C^*$13

produces the desired embedding into the graph planar algebra (Coles et al., 2018).

Different choices entering the construction are controlled. Shifting by two strings produces isomorphic canonical planar algebras; compression by projections with central support $C^*$14 produces isomorphic canonical planar algebras; and different choices of Pimsner–Popa bases change the identification with the graph planar algebra by a planar $C^*$15-automorphism. Thus the resulting embeddings are equivalent up to planar $C^*$16-automorphisms of the target (Coles et al., 2018).

5. Module categories, Markov towers, and the generalized embedding theorem

A major conceptual advance of the towers-of-algebras approach is the equivalence between several formulations of “module over a subfactor planar algebra.” The paper establishes the following chain of equivalences: subfactor planar algebras are equivalent to unitary $C^*$17 multitensor categories with generator; modules over planar algebras are equivalent to pivotal module $C^*$18-categories over the corresponding multitensor category; and pivotal $C^*$19-module $C^*$20-categories are equivalent to connected Markov towers of tracial finite-dimensional von Neumann algebras and to pointed bipartite graphs with Frobenius–Perron vertex weights (Coles et al., 2018).

A Markov tower

$C^*$21

consists of an increasing sequence of finite-dimensional von Neumann algebras, faithful normal tracial states with $C^*$22, and Jones projections satisfying the Temperley–Lieb–Jones relations with modulus $C^*$23. The canonical trace-preserving conditional expectation $C^*$24 is implemented by $C^*$25:

$C^*$26

and the Markov property is

$C^*$27

The tower also has a pull-down property $C^*$28, and a decomposition of each level into “old stuff” and “new stuff,” with the principal graph defined from the simple summands in the new stuff (Coles et al., 2018).

The converse construction starts from a cyclic pivotal right $C^*$29-module $C^*$30-category $C^*$31. One sets

$C^*$32

takes $C^*$33 to be the normalized trace induced by $C^*$34, and defines Jones projections from the coevaluation and evaluation morphisms of $C^*$35 and their daggers. These satisfy the Temperley–Lieb–Jones relations with modulus $C^*$36, and the principal graph of the resulting tower is exactly the fusion graph of the cyclic module $C^*$37 with respect to the generator $C^*$38 (Coles et al., 2018).

This equivalence yields a classification theorem for cyclic pivotal right $C^*$39-module $C^*$40-categories with simple basepoint. Their equivalence classes are classified by a connected bipartite graph $C^*$41 with distinguished base vertex $C^*$42 and a positive function $C^*$43 satisfying the Frobenius–Perron condition

$C^*$44

equivalently

$C^*$45

with normalization $C^*$46. The data determine, and are determined by, the Markov tower and its principal graph, hence by the $C^*$47-module (Coles et al., 2018).

In this formulation, the generalized embedding theorem becomes a structural statement: every finite depth subfactor planar algebra embeds into the graph planar algebra of the fusion graph of any of its cyclic modules. The original principal graph embedding is the special case obtained from the canonical cyclic module defined by the planar module tower itself (Coles et al., 2018).

6. Applications, obstructions, and significance for classification

One immediate consequence of the embedding theorem is that any finite depth subfactor planar algebra may be studied concretely inside a loop algebra model. This permits explicit computations with generators, Jones projections, traces, and annular actions using paths on a graph rather than only abstract planar operad relations. The theorem is therefore a central tool in classification problems, especially at small index, where low-box-space relations can be checked inside graph planar algebras (Jones et al., 2010).

A notable application is the obstruction of principal graphs developed from the embedding theorem for a family of $C^*$48-supertransitive subfactors. In that setting, if the principal graph begins as a Temperley–Lieb chain through depth $C^*$49, bifurcates at depth $C^*$50 to vertices $C^*$51 and $C^*$52, and continues to $C^*$53 and $C^*$54 at depth $C^*$55, then one studies the image of a $C^*$56-box lowest weight rotational eigenvector

$C^*$57

inside $C^*$58. The vector $C^*$59 satisfies the quadratic relation

$C^*$60

where $C^*$61 is the ratio $C^*$62 or its reciprocal if $C^*$63, and $C^*$64 is the $C^*$65-strand Jones–Wenzl idempotent (Morrison, 2013).

The graph planar algebra model turns annular lowest-weight conditions into explicit linear equations on coefficients of loops. For example, the relations

$C^*$66

and analogous formulas at deeper vertices reduce the calculation to finitely many collapsed loops. In the family considered, there are $C^*$67 collapsed loops and $C^*$68 rotation orbits; after imposing lowest-weight and rotation constraints, the solution space has dimension $C^*$69 for $C^*$70 and dimension $C^*$71 for $C^*$72 (Morrison, 2013).

The decisive consequence is an “early cycle” obstruction. If there is no vertex at depth $C^*$73 adjacent to both $C^*$74 and $C^*$75, then the quadratic and annular relations force either the Haagerup polynomial

$C^*$76

or a contradiction. For $C^*$77, this yields the index

$C^*$78

which is the Haagerup index. Hence, in this family, absence of a cycle by depth $C^*$79 implies that the standard invariant is that of the Haagerup subfactor; otherwise a cycle must appear by depth $C^*$80 (Morrison, 2013).

The broader significance is twofold. First, the embedding theorem transforms structural questions about subfactor planar algebras into concrete graph-theoretic calculations with Perron–Frobenius weights and loop concatenations. Second, the module embedding theorem shows that non-principal graph targets are not exceptional phenomena but are organized by cyclic pivotal module categories and their fusion graphs. This unifies planar algebra, tensor-categorical, and operator-algebraic approaches to subfactors within a single graph-planar-algebra framework (Coles et al., 2018).