DHR Sectors and Modular Tensor Categories of Nets

The superselection theory of a completely rational conformal net is a unitary modular tensor category because localization on the circle produces a braiding, finite Jones index produces conjugates and finite dimensions, and the two-interval theorem removes transparent nontrivial sectors. Each clause is needed: a braided category is not automatically modular, and abstract modular data do not reconstruct a net.

Required background. Split Property and Complete Rationality supplies the finiteness theorem; Superselection Sectors and DHR Reconstruction supplies localized endomorphisms; and Fusion, Junctions, and Endpoints supplies the physical fusion interpretation. Helpful background. What Is a Topological Field Theory? explains why the resulting category resembles anyon data without identifying a chiral net with a topological theory.

Localized endomorphisms and braiding

Let $\mathfrak A$ be the quasilocal C*-algebra obtained by norm-closing the union of all interval algebras in the vacuum representation. A DHR sector may be represented by a transportable endomorphism $\rho\in\operatorname{End}(\mathfrak A)$ localized in an interval $I$ :

\rho(A)=A\qquad\text{for every }A\in\mathcal A(I').

Transportability means that for every interval $J$ there is an equivalent endomorphism localized in $J$ . Intertwiners are bounded elements $t$ satisfying $t\rho(A)=\sigma(A)t$ for all $A\in\mathfrak A$ . Composition is fusion, $\rho\otimes\sigma=\rho\circ\sigma$ , and the vacuum sector is the identity endomorphism.

On $S^1$ , two localized charges can be transported past one another clockwise or counterclockwise. Locality makes the resulting intertwiner depend only on the ordering, yielding unitary braiding operators

\varepsilon(\rho,\sigma)\in \operatorname{Hom}(\rho\sigma,\sigma\rho).

The two components of the complement retain an orientation, so the braiding need not square to the identity. This is the decisive difference from DHR theory in at least $2+1$ spatial dimensions, where exchange is symmetric. Finite-index sectors have conjugates $\bar\rho$ and statistical dimension

d(\rho)=\sqrt{[\mathcal A(I):\rho(\mathcal A(I))]}.

Standard solutions of the conjugate equations supply evaluation and coevaluation maps. Positivity of the Hilbert-space adjoint makes the category a C*-tensor category; it is not enough to know only a fusion ring.

Why complete rationality gives modularity

For a completely rational net, there are finitely many isomorphism classes of irreducible DHR sectors, all have finite statistical dimension, every sector decomposes as a finite direct sum, and the braiding is nondegenerate. Equivalently, the Müger center contains only the vacuum: if

\varepsilon(\rho,\sigma)\varepsilon(\sigma,\rho)=\mathbf 1 \quad\text{for every }\sigma,

then $\rho$ is a sum of vacuum sectors. These results are Corollaries 37–39 of Kawahigashi, Longo, and Müger 2001, pp. 657–658. The proof identifies the canonical endomorphism of the two-interval inclusion with the Longo–Rehren endomorphism $\bigoplus_i\rho_i\boxtimes\bar\rho_i$ . Finite $\mu$ bounds the total squared dimension; equality $\mu=\sum_i d_i^2$ exhausts the canonical endomorphism; a transparent sector would produce extra central data and contradict this exhaustion.

The resulting finite semisimple rigid braided C*-category, with nondegenerate braiding and compatible twist, is a unitary modular tensor category. This is a theorem about the DHR category of the net. Its converse fails: a unitary modular tensor category can have no known conformal-net realization, and even a realizable category need not identify a unique net.

The Ising net

The completely rational Ising net has three irreducible sectors $\mathbf1,\sigma,\varepsilon$ with fusion

\sigma\otimes\sigma\cong\mathbf1\oplus\varepsilon, \qquad \sigma\otimes\varepsilon\cong\sigma, \qquad \varepsilon\otimes\varepsilon\cong\mathbf1.

Taking statistical dimensions gives $d_{\mathbf1}=d_\varepsilon=1$ and $d_\sigma=\sqrt2$ . The twists are $1,e^{\pi i/8},-1$ in the order $(\mathbf1,\sigma,\varepsilon)$ . In the same order,

S=\frac12 \begin{pmatrix} 1&\sqrt2&1\\ \sqrt2&0&-\sqrt2\\ 1&-\sqrt2&1 \end{pmatrix}.

Direct multiplication gives $S^*S=\mathbf1$ and $\det S=-1$ , so the braiding is nondegenerate. The global-dimension check is $1^2+(\sqrt2)^2+1^2=4=\mu$ . This supplies the exact sector, fusion, braiding, dimension, and twist calculation for Minimal Models and Fusion Rules. The Ising data and the role of complete rationality are summarized with operator-algebraic hypotheses in Kawahigashi 2015, §§3.4–3.5, pp. 24–30.

Which categorical data have actually been recovered

Fusion coefficients alone record only dimensions of spaces of intertwiners. The DHR construction also supplies the concrete C*-categories of intertwiners, their associative composition, unitary braiding, conjugate solutions, and positive categorical trace. From these one obtains twists and modular matrices; the order cannot be reversed from a numerical pair $(S,T)$ without a realization theorem. Even the full unitary modular tensor category describes superselection data rather than the interval net itself. It forgets the vacuum Hilbert space, covariance representation, local type-III factors, embeddings for nested intervals, and stress tensor. Thus categorical equivalence is a powerful invariant of a completely rational net, but it is not net isomorphism.

There is also a localization boundary. DHR sectors are transportable and localizable in bounded intervals. Solitonic representations localized only on half-lines, boundary sectors, or sectors of a nonrational net can form different tensor categories and need not enter the finite modular theorem. Before applying a Verlinde formula, one must therefore check both completeness of the DHR list and nondegeneracy of its braiding.

Adversarial test and independent check

Drop finite $\mu$ and suppose a nontrivial sector $z$ has trivial monodromy with every sector. The corresponding row of the unnormalized Hopf-link matrix is proportional to the dimension row, so $S$ has dependent rows and cannot be inverted. Thus braided does not imply modular. Independently, the Ising fusion equation gives $d_\sigma^2=d_{\mathbf1}+d_\varepsilon=2$ , fixing $d_\sigma=\sqrt2$ by positivity before the $S$ -matrix is used.

Exercises

Use the displayed Ising $S$ matrix to verify the Verlinde coefficient $N_{\sigma\sigma}^{\varepsilon}=1$ .

Solution

All entries are real, and $S_{0x}=(1,\sqrt2,1)/2$ . Hence $N_{\sigma\sigma}^{\varepsilon}=\sum_x S_{\sigma x}^2S_{\varepsilon x}/S_{0x}$ . The three contributions are $1/2$ , $0$ , and $1/2$ , giving $1$ . This checks the fusion rule without assuming it separately.