Subfactors, Jones Index, and Q-Systems

A finite-index inclusion of local factors is encoded by a Q-system: a positive special Frobenius algebra object built from the canonical endomorphism. The Q-system reconstructs the extension, while braided commutativity is the extra condition that makes the reconstructed net local. Jones index, algebra-object dimension, and locality therefore answer different questions and must not be conflated.

Required background. Strong Additivity, Haag Duality, and the μ-Index supplies finite-index local inclusions; DHR Sectors and Modular Tensor Categories of Nets supplies the endomorphism category; and Operator Algebras and Positive Functionals: a Bridge supplies factors and commutants. Helpful background. Operator-Algebra Quantum Error Correction gives another use of inclusions, while Constructions from Gauging, Duality, and Condensation gives the categorical interpretation of condensable algebras.

From a subfactor to its canonical endomorphism

Let $\iota:N\hookrightarrow M$ be an inclusion of type-III factors with finite Jones index. A conjugate homomorphism $\bar\iota:M\to N$ exists, together with solutions of the conjugate equations. The dual canonical and canonical endomorphisms are

\theta=\bar\iota\iota\in\operatorname{End}(N), \qquad \gamma=\iota\bar\iota\in\operatorname{End}(M).

Their dimensions obey

d(\theta)=d(\iota)^2=[M:N].

This identity is an immediate normalization check: $d(\iota)$ is the square root of the subfactor index, whereas $d(\theta)$ is the index itself. Confusing the two introduces a factor-of-two error in logarithmic index formulas and a square-root error in categorical dimensions.

A Q-system in $\operatorname{End}(N)$ is a triple $(\theta,w,x)$ with

w\in\operatorname{Hom}(\operatorname{id}_N,\theta), \qquad x\in\operatorname{Hom}(\theta,\theta^2),

satisfying unit, associativity, Frobenius, specialness, and adjoint-compatibility relations. Diagrammatically, $w$ creates the algebra unit and $x$ is the adjoint of multiplication after the standard C*-normalization. The conjugate equations for $\iota,\bar\iota$ produce these maps. Conversely, a Q-system reconstructs a finite-index extension $N\subset M$ , unique up to the appropriate unitary equivalence. This reconstruction, including the relative-locality condition for nets, is Theorem 4.9 of Longo and Rehren 1995, pp. 590–592.

Locality is a braided condition

Suppose $N=\mathcal A(I)$ and $\theta$ lies in the braided DHR category of a conformal net. The reconstructed extension $\mathcal B$ is local precisely when the multiplication is commutative with respect to the DHR braiding:

\varepsilon(\theta,\theta)x=x.

Without this equation the Q-system still gives a finite-index inclusion and often a relatively local or graded-local extension, but spacelike separated charged generators can braid nontrivially. Without the C*-positivity and specialness relations, a formal Frobenius algebra need not act on a positive Hilbert space. Thus “algebra object,” “Q-system,” and “commutative Q-system” have progressively stronger content.

The proof mechanism adjoins to $N$ an isometry representing the charged generator, uses $w$ and $x$ to reduce products to $N$ , and obtains associativity from the Q-system identities. Transportability glues the interval inclusions into a net. Exchanging generators in disjoint intervals produces $\varepsilon(\theta,\theta)$ , so the displayed commutativity equation is exactly the step that upgrades relative locality to locality.

The finite-index Ising construction

Let $\mathcal C$ be the Ising DHR category with sectors $\mathbf1,\sigma,\varepsilon$ . In $\mathcal C\boxtimes\mathcal C^{\mathrm{rev}}$ , the Longo–Rehren or full-center object is

\Theta=(\mathbf1\boxtimes\mathbf1) \oplus(\sigma\boxtimes\sigma) \oplus(\varepsilon\boxtimes\varepsilon).

Its dimension is

d(\Theta)=1+(\sqrt2)^2+1=4.

The canonical Q-system therefore gives an index-four extension of the left–right chiral product. The reverse braiding in the second factor cancels the first-factor monodromy, so the full-center multiplication is commutative and the extension is local. Proposition 4.10 of Longo and Rehren 1995, pp. 592–595 gives the canonical sector sum and its global index. This provides the concrete finite-index inclusion behind the region-algebra discussion in Regions, Causal Complements, and Nets of Observables: the larger dual algebra contains charged intertwiners absent from the componentwise algebra.

An independent check is multiplicativity. Since $d(\sigma\boxtimes\sigma)=d(\sigma)^2=2$ , the three summands have dimensions $1,2,1$ and add to $4$ , matching the Ising $\mu$ -index. In contrast, the chiral object $\mathbf1\oplus\varepsilon$ has dimension $2$ but $\theta_\varepsilon=-1$ ; its multiplication is not bosonically commutative. It can describe a graded or nonlocal extension, not an ordinary local conformal net.

Equivalence and reconstruction checks

Two unitarily equivalent Q-systems reconstruct isomorphic inclusions that fix the subnet, but equality of their underlying endomorphism $\theta$ is weaker: distinct multiplication intertwiners can encode inequivalent extensions. At the net level one must also transport the construction coherently between intervals and intertwine covariance. A useful check is to recover the canonical conditional expectation $E:M\to N$ from the Q-system and verify that its minimal index equals $d(\theta)$ . If that number disagrees with the sector-dimension sum, either the normalization of $x$ or the proposed canonical endomorphism is wrong.

Adversarial failure

Start with an associative positive Q-system but drop $\varepsilon(\theta,\theta)x=x$ . Reconstruction still yields a subfactor, yet the exchange of two charged generators leaves a nontrivial braid operator. The licensed conclusion is finite-index relative locality, not locality. More severely, a fusion-ring sum such as $\mathbf1\oplus X$ need not admit any associative positive multiplication at all. Its integer dimension or appealing sector decomposition is not a converse existence theorem.

Exercises

For the Ising full-center object $\Theta$ , compute the Jones index and explain why replacing the second Ising category by one with the same braiding, rather than reverse braiding, jeopardizes locality.

Solution

The index is $d(\Theta)=1+2+1=4$ . With reverse braiding, the monodromy phases of corresponding left and right sectors cancel. With the same braiding they multiply, so $\varepsilon(\Theta,\Theta)x=x$ need not hold; the sector sum alone does not prove a local extension.