Extensions, Orbifolds, Cosets, and Alpha-Induction

Finite-index extensions, finite-group fixed points, and cosets are constructions of nets; alpha-induction is a method for transporting chiral sectors through an inclusion and extracting a coupling matrix. Under finite-index, braided, nondegeneracy, and chiral-locality hypotheses that matrix is modular invariant. The reverse inference is false: a matrix commuting with $S$ and $T$ need not supply a positive commutative Q-system.

Required background. DHR Sectors and Modular Tensor Categories of Nets supplies the braided sector system, Subfactors, Jones Index, and Q-Systems supplies extension data, and Minimal Models and Fusion Rules supplies rational examples. Helpful background. Cosets and Orbifolds develops the physical constructions and twisted sectors.

Three constructions and their hypotheses

A finite-index extension is an isotone family $\mathcal A(I)\subset\mathcal B(I)$ described intervalwise by one transportable Q-system. It is local only when the Q-system is braided-commutative. If a finite group $G$ acts properly on $\mathcal B$ , preserves the vacuum, and commutes with covariance, the fixed-point assignment

\mathcal A(I)=\mathcal B(I)^G

is the orbifold subnet. Its representation theory includes twisted information not visible in the naive invariant part of the vacuum Hilbert space. A coset begins with a covariant subnet $\mathcal A\subset\mathcal B$ and defines

\mathcal C(I)=\mathcal A(I)'\cap\mathcal B(I).

For $\mathcal C$ to be a conformal net with the expected stress tensor, one must check isotony, locality, covariance, vacuum cyclicity in the coset Hilbert space, and suitable normality or cofinite-index conditions. A formal central-charge subtraction is not a construction of these relative commutants.

Alpha-induction and the coupling matrix

Let $\iota:N\hookrightarrow M$ have finite index and dual canonical endomorphism $\theta=\bar\iota\iota$ . For a braided finite system $\lambda\in\operatorname{End}(N)$ containing the irreducible summands of $\theta$ , alpha-induction defines endomorphisms $\alpha_\lambda^\pm$ of $M$ . On the range of $\bar\iota$ , the defining formula is

\alpha_\lambda^\pm =\bar\iota^{-1}\circ \operatorname{Ad}\!\bigl(\varepsilon^\pm(\lambda,\theta)\bigr) \circ\lambda\circ\bar\iota,

with the inverse understood through the canonical extension construction. Both signs restrict through $\iota$ to $\iota\lambda$ ; they differ in whether over- or under-braiding is used with $\theta$ . This formula and its finite-index domain are developed in Böckenhauer, Evans, and Kawahigashi 1999, §3.3, pp. 457–459.

The coupling matrix is

Z_{\lambda\mu}= \dim\operatorname{Hom}_M(\alpha_\lambda^+,\alpha_\mu^-).

Its entries are nonnegative integers and $Z_{00}=1$ for an irreducible extension. If the braiding is nondegenerate and the canonical endomorphism satisfies chiral locality, then $Z$ commutes with the modular $S$ and $T$ matrices. The intertwining-braiding-fusion relations move a braiding through the induced morphisms; summing the resulting identities produces $ZS=SZ$ and $ZT=TZ$ . The exact theorem, including the distinction between nondegenerate braiding and chiral locality, is Böckenhauer, Evans, and Kawahigashi 1999, Definition 5.5, Theorem 5.7, and Corollary 5.8, pp. 472–476.

A local simple-current extension

For the $SU(2)_4$ net, label sectors by $j=0,1,2,3,4$ . The simple current $J=4$ has order two and conformal weight

h_J=\frac{J(J+2)}{4(k+2)}=1,

so its twist is $e^{2\pi i h_J}=1$ . The object $\theta=0\oplus4$ admits the bosonic commutative Q-system. Alpha-induction gives the $D_4$ coupling

Z=\lvert\chi_0+\chi_4\rvert^2+2\lvert\chi_2\rvert^2.

The equal $T$ phases of sectors $0$ and $4$ check $TZ=ZT$ , while the multiplicity two at the fixed point $j=2$ records fixed-point resolution. The Q-system’s braided commutativity supplies locality; modular invariance is then a conclusion rather than an assumption. This is the concrete simple-current calculation returned to Cosets and Orbifolds, alongside the warning that the orbifold must include twisted sectors and the coset must be defined by relative commutants.

An independent dimension check is $d_0+d_4=2$ , hence the extension index is $2$ . Since $J$ is invertible, $J\otimes J=0$ and the algebra closes. At level $k=2$ , by contrast, the analogous terminal simple current has half-integer conformal spin; the same sector sum is not a bosonic local extension.

Induced endomorphisms versus genuine sectors

The notation $\alpha^\pm_\lambda$ can tempt one to treat every induced endomorphism as a DHR sector of the extension. In general it is only a solitonic endomorphism: it may fail to be localizable in a bounded interval because moving it across the extension remembers the chosen over- or under-braiding. The ambichiral endomorphisms appearing in both the plus and minus systems are the candidates that recover genuine local sectors. Consequently $Z_{\lambda\mu}$ measures an intertwiner space between two induced systems; it is not by itself the fusion matrix of the extension.

Orbifolds illustrate the same missing-data issue. Taking fixed vectors in the vacuum representation constructs the invariant subnet, but its complete representation theory requires twisted sectors. A dimension or character calculation that retains only untwisted invariants can satisfy elementary modular checks while failing sector completeness. The operator-algebraic construction keeps the fixed-point inclusion, its index, and its canonical endomorphism visible, which is precisely the information needed to detect the omitted sectors.

Adversarial failure and nonconverse

Start only with a nonnegative integer matrix $Z$ that commutes with $S$ and $T$ . It contains no multiplication $x\in\operatorname{Hom}(\theta,\theta^2)$ , no associativity witnesses, no C*-positivity, and no proof of $\varepsilon(\theta,\theta)x=x$ . It may therefore be a numerical modular invariant without being realized by a local extension of the chosen net. Likewise, alpha-induction without chiral locality can still produce a coupling matrix, but the stronger type-I extension interpretation fails. Modular commutation is a necessary shadow of a local Q-system in the theorem’s regime, not a converse construction theorem.

Exercises

Verify directly that the $SU(2)_4$ extension pairs only sectors with equal $T$ phase in the displayed expression.

Solution

For $SU(2)_4$ , $h_j=j(j+2)/24$ . Thus $h_0=0$ , $h_4=1$ , and the corresponding $T$ phases agree after the common $-c/24$ shift. Sector $2$ appears only with itself, so its phase matches trivially. Hence every nonzero entry of $Z$ connects equal $T$ eigenvalues.