Levin-Wen String-Net Models Overview

• Levin–Wen string-net models are exactly solvable lattice Hamiltonians that use fusion-category data to realize 2+1D topological phases with anyonic excitations.
• They employ trivalent lattices with vertex constraints and plaquette operators based on F-symbol recouplings to enforce consistent topological orders.
• Recent generalizations extend these models to arbitrary multiplicity-free fusion categories, enabling the study of gapped boundaries and nonchiral phase realizability.

Searching arXiv for recent and foundational papers on Levin–Wen string-net models. Levin–Wen string-net models are exactly solvable lattice Hamiltonians for $2+1$-dimensional topological phases built from categorical input data, most commonly a unitary fusion category. In their canonical form, the degrees of freedom live on edges of a trivalent lattice, local vertex constraints enforce admissible fusion, and plaquette operators insert and recouple virtual loops using $F$-symbols. This construction realizes doubled, nonchiral topological orders and provides a microscopic route from fusion-category data to commuting-projector Hamiltonians, topological ground states, and anyonic excitations (Jr, 2011, Lan et al., 2013, Kim et al., 2024). Subsequent work has refined both the scope and the limitations of the framework: it has clarified the relation to Turaev–Viro TQFT (Jr, 2011), exhibited quantum double models as an explicit subclass (0907.2670), generalized the Hamiltonian to arbitrary multiplicity-free unitary fusion categories without tetrahedral symmetry (Hahn et al., 2020), characterized which abelian phases are realizable (Lin et al., 2014), and identified pseudo-unitarity as the sharp boundary between formal topological consistency and ordinary Hilbert-space realizability (Yang, 11 Jan 2026).

1. Categorical definition and lattice Hamiltonian

The standard input of a Levin–Wen model is a fusion category specified by simple objects, duals, fusion multiplicities, quantum dimensions, and associativity data encoded by $F$-symbols. In the common multiplicity-free case, admissibility at a trivalent vertex is encoded by $\delta_{ij}^{k}\in\{0,1\}$, while in the more general formulation one uses fusion spaces such as $\Hom(1,a\otimes b\otimes c)$ (Hahn et al., 2020, Kim et al., 2024). In the string-net presentation, an oriented edge carries a label $a$, and orientation reversal replaces it by its dual, for example

$\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$

in the representation-category formulation (0907.2670).

The Hamiltonian has the characteristic commuting-projector form

$H=-\sum_v Q_v-\sum_p B_p$

or, in closely related notations,

$$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$

(Yang, 11 Jan 2026, Hahn et al., 2020, 0907.2670). The vertex term projects onto locally admissible fusion channels. In one notation,

$$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$

where $F$0 when $F$1, and $F$2 otherwise (Yang, 11 Jan 2026). In the multiplicity-free convention this becomes

$F$3

(Hahn et al., 2020).

The plaquette term inserts a virtual loop and resolves it by $F$4-moves. In canonical weighted form,

$F$5

or equivalently with $F$6 and $F$7 (0907.2670, Hahn et al., 2020, Yang, 11 Jan 2026). The total quantum dimension is thus

$F$8

For a hexagonal plaquette, explicit matrix elements are products of $F$9-symbols around the plaquette boundary (0907.2670). In graphical language, contractible loops evaluate to their quantum dimensions, and local recouplings are governed by the associator (Yang, 11 Jan 2026, Lin et al., 2014).

The exact solvability of the model rests on the pentagon identity and related coherence conditions for the $F$0-symbols. In compressed form, the consistency condition is that different sequences of associativity moves agree,

$F$1

and, in optimization-based formulations,

$F$2

(Yang, 11 Jan 2026). The commuting-projector relations

$F$3

are explicit in the abelian construction and persist in generalized unitary settings (Lin et al., 2014, Hahn et al., 2020).

2. Topological meaning and relation to TQFT

A central interpretation of Levin–Wen models is that they are lattice realizations of doubled topological phases. Kirillov gave a precise equivalence between the string-net space of a spherical fusion category $F$4 and the Turaev–Viro state space: $F$5 for closed oriented surfaces (Jr, 2011). In that formulation, the string-net space is the vector space of colored graphs on a surface modulo local null relations, and puncture projectors $F$6 implemented by weighted loops recover the Turaev–Viro cylinder projector (Jr, 2011).

This TQFT relation is also reflected in the conceptual role of the Drinfeld center. The bulk quasiparticles of a Levin–Wen model are described by $F$7, the Drinfeld center of the input fusion category $F$8, and in boundary formulations one finds

$F$9

so that boundary conditions and bulk excitations are both encoded by center data (Jr, 2011). More broadly, the Levin–Wen construction is widely understood to realize doubled, achiral phases, with the bulk invariant expected to be $\delta_{ij}^{k}\in\{0,1\}$0 (Kim et al., 2024, Kawagoe et al., 2024).

The relation to topological phase classification was sharpened in a state-based setting by a theorem that, assuming exact entanglement-bootstrap axioms and a strong local notion of gappable boundary, a representative state on a disk can be mapped by a geometrically local constant-depth circuit to a canonical Levin–Wen ground state built from a unitary fusion category $\delta_{ij}^{k}\in\{0,1\}$1 (Kim et al., 2024). In that setting, the phase classification is expressed in terms of doubled unitary modular tensor categories, conditional on the standard identification of the bulk anyons with $\delta_{ij}^{k}\in\{0,1\}$2 (Kim et al., 2024).

A more categorical reformulation interprets Levin–Wen models as gauge theories whose gauge symmetry is the tube algebra $\delta_{ij}^{k}\in\{0,1\}$3, rather than an ordinary finite group. In that framework, the Levin–Wen Hamiltonian arises by gauging a trivial $\delta_{ij}^{k}\in\{0,1\}$4-symmetric phase, and the anyon theory again coincides with $\delta_{ij}^{k}\in\{0,1\}$5 through the equivalence

$\delta_{ij}^{k}\in\{0,1\}$6

(Kawagoe et al., 2024). This viewpoint extends ordinary group gauging to non-group examples, including the doubled Fibonacci phase (Kawagoe et al., 2024).

3. Quantum doubles, boundaries, and excitations

A foundational bridge between Levin–Wen models and more concrete lattice gauge theories is the exact mapping from Kitaev’s quantum double models to string-net models based on $\delta_{ij}^{k}\in\{0,1\}$7 (0907.2670). After a local Fourier transform from group-element variables to representation variables,

$\delta_{ij}^{k}\in\{0,1\}$8

the quantum double Hilbert space becomes an enlarged string-net Hilbert space in which the irrep labels $\delta_{ij}^{k}\in\{0,1\}$9 play the role of string types and the matrix indices $\Hom(1,a\otimes b\otimes c)$0 become auxiliary edge-end degrees of freedom (0907.2670). In this representation basis, the vertex projector is exactly the projection onto the trivial isotypic component of the tensor product of incident irreducible representations, and the plaquette term matches the Levin–Wen plaquette operator with

$\Hom(1,a\otimes b\otimes c)$1

(0907.2670).

This mapping shows that quantum double models form a specific subclass of string-net models. It also imports the excitation classification of the Drinfeld double $\Hom(1,a\otimes b\otimes c)$2: magnetic excitations by conjugacy classes, electric excitations by irreducible representations of $\Hom(1,a\otimes b\otimes c)$3, and dyons by pairs $\Hom(1,a\otimes b\otimes c)$4 with $\Hom(1,a\otimes b\otimes c)$5 a conjugacy class and $\Hom(1,a\otimes b\otimes c)$6 an irrep of the corresponding centralizer (0907.2670).

For general input categories, a finite computational method for extracting the quasiparticle data from the string-net wavefunction is the $\Hom(1,a\otimes b\otimes c)$7-algebra approach. The cylinder ground-state space defines an associative algebra, and a finite-dimensional representative $\Hom(1,a\otimes b\otimes c)$8 can be constructed whose simple modules classify simple bulk quasiparticles (Lan et al., 2013). In this formulation, the input UFC determines the $\Hom(1,a\otimes b\otimes c)$9-algebra, and from its irreducible modules one computes quantum dimensions, topological spins, and the modular $a$0-matrix, thereby recovering the bulk UMTC $a$1 (Lan et al., 2013). This provides a concrete boundary-to-bulk map: the anomalous $a$2D edge data described by a UFC determine the $a$3D bulk topological order through the center construction (Lan et al., 2013).

Tube algebras also organize localized excitations in generalized and condensed Levin–Wen systems. In ordinary models, irreducible $a$4-representations of $a$5 are equivalent to simple objects of $a$6, so anyons are exactly the simples of the Drinfeld center (Christian et al., 2023). In lattice models for anyon condensation, one can interpolate between the uncondensed Levin–Wen phase with topological order $a$7 and a condensed phase with topological order $a$8, where $a$9 is a condensable algebra (Christian et al., 2023). In that setting, a quotient tube algebra $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$0 classifies the deconfined excitations of the condensed phase and is Morita equivalent to $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$1 (Christian et al., 2023).

4. Generalizations of the original construction

A major line of development has been the removal of extra assumptions present in the original Levin–Wen formulation. The original graphical derivation relied on tetrahedral symmetry of the $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$2-symbols, but this symmetry is not automatic for arbitrary unitary fusion categories. A generalized construction for arbitrary multiplicity-free unitary fusion categories avoids ambiguous horizontal-line diagrams and instead uses orientation-sensitive $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$3- and $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$4-moves (Hahn et al., 2020). The resulting plaquette matrix elements are given explicitly in terms of $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$5-symbols and their complex conjugates, and the Hamiltonian remains Hermitian, projector-valued, and commuting (Hahn et al., 2020).

This generalization is substantial because it includes categories such as the Haagerup $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$6 fusion category, which fail tetrahedral symmetry and were not covered by the original construction (Hahn et al., 2020). The general philosophy is that unitarity and sphericality are sufficient for the lattice Hamiltonian once the graphical calculus is formulated correctly, whereas tetrahedral symmetry is an unnecessary restriction tied to a particular presentation (Hahn et al., 2020).

Another extension concerns parity and time-reversal breaking phases. A generalized string-net formalism introduces extra local data $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$7 and $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$8, related to $\lvert a\rangle_{e^\ast}=\lvert a^\ast\rangle_e$9 and $H=-\sum_v Q_v-\sum_p B_p$0 Frobenius–Schur indicators, and drops the parity-invariance assumption of the original model (Lin et al., 2014). In the abelian case, the local data are

$H=-\sum_v Q_v-\sum_p B_p$1

with the consistency conditions

$H=-\sum_v Q_v-\sum_p B_p$2

$H=-\sum_v Q_v-\sum_p B_p$3

and the unitarity condition

$H=-\sum_v Q_v-\sum_p B_p$4

(Lin et al., 2014). These generalized models realize phases inaccessible to the original construction, including examples that break parity and time reversal (Lin et al., 2014).

A further extension replaces a single flavor of strings by multiple flavors labeled by groups $H=-\sum_v Q_v-\sum_p B_p$5, with same-flavor branching but inter-flavor crossing. The resulting intersecting string-net models are exactly solvable and realize the same set of abelian phases as ordinary string-net models with

$H=-\sum_v Q_v-\sum_p B_p$6

while making the Künneth decomposition of $H=-\sum_v Q_v-\sum_p B_p$7 concrete in lattice terms (Lin, 2016). This construction gives, for example, a $H=-\sum_v Q_v-\sum_p B_p$8 string-net model realizing a non-abelian topological phase by intersecting three toric-code layers (Lin, 2016).

5. Realizability, gapped edges, and gauge-theoretic subclasses

The scope of string-net realizability is sharpest in the abelian case. A complete criterion states that an abelian topological phase is realizable by a string-net model if and only if it has vanishing thermal Hall conductance and at least one Lagrangian subgroup; equivalently, if and only if it supports a fully gapped edge to the vacuum (Lin et al., 2014). In $H=-\sum_v Q_v-\sum_p B_p$9-matrix language, realizability is equivalent to the existence of $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$0 linearly independent integer null vectors $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$1 for an even-dimensional $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$2-matrix: $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$3 The associated string-net $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$4-matrices have the form

$$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$5

(Lin et al., 2014).

This result has several implications. First, commuting-projector string-net Hamiltonians necessarily have zero chiral central charge and therefore zero thermal Hall conductance (Lin et al., 2014). Second, nonchiral abelian phases without a Lagrangian subgroup are nevertheless excluded, so vanishing thermal Hall conductance alone is not sufficient (Lin et al., 2014). Third, the realizability criterion is controlled by boundary physics: string-net models realize exactly the abelian topological orders with a fully gapped boundary (Lin et al., 2014).

The relation between string-net models and discrete gauge theories is especially explicit for representation categories $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$6, where quantum double models are realized as string-net models after Fourier transform (0907.2670). In this subclass the string types are irreducible representations, the $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$7-symbols are group-theoretic $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$8-symbols built from intertwiners, and the anyon theory is $$H=-\sum_{\mathbf v} Q_{\mathbf v}-\sum_{\mathbf p} B_{\mathbf p}, \qquad H^{\mathrm{SN}}=-\sum_v A_v^{\mathrm{SN}}-\sum_p B_p^{\mathrm{SN}}$$9 (0907.2670). Closely related work on $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$0 fusion algebras shows that solutions of the local constraints correspond to $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$1 gauge theory and doubled Chern–Simons theories with quantum groups, and that the string-net model is exactly dual to a triangular-lattice spin model coupled to a $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$2 gauge field (Hung et al., 2012).

A cautionary point concerns the relation to Turaev–Viro. One critique identifies pointed-category counterexamples showing that the slogan “Levin–Wen equals microscopic Hamiltonian formulation of Turaev–Viro” is too naive unless additional categorical structure, specifically unimodality, is imposed in the relevant group-category setting (Stirling, 2010). A later mathematical treatment using spherical fusion categories gives the precise correspondence between string-net spaces and Turaev–Viro state spaces, including surfaces with boundary (Jr, 2011). This suggests that the issue is not the existence of a relation, but the exact categorical hypotheses under which it holds (Stirling, 2010, Jr, 2011).

6. Representative models, phase transitions, and non-unitary obstructions

Several benchmark categories recur throughout the literature. The abelian $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$3 model has trivial dimensions

$$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$4

and scalar $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$5-cocycle $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$6-symbols $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$7, with unitarity $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$8 (Yang, 11 Jan 2026). The rank-5 Tambara–Yamagami category $$Q_v |i,j,k;\mu\rangle=\delta_{ijk}|i,j,k;\mu\rangle,$$9 has simple objects $F$00 and $F$01, with

$F$02

illustrating that non-integral positive quantum dimensions do not obstruct Hermitian string-net Hamiltonians (Yang, 11 Jan 2026). The non-Abelian category $F$03 provides a concrete unitary benchmark with

$F$04

and an explicitly unitary real-gauge $F$05-matrix on the $F$06 channel (Yang, 11 Jan 2026).

The Fibonacci input category is central in both topological and dynamical studies. Its nontrivial object obeys

$F$07

(Schulz et al., 2012, Ritz-Zwilling et al., 2020). The corresponding golden string-net model on the honeycomb lattice exhibits a doubled Fibonacci topological phase near the exactly solvable Levin–Wen point (Schulz et al., 2012). In one study combining high-order series expansions and exact diagonalization, the doubled Fibonacci phase occupies

$F$08

with second-order transitions into two distinct trivial phases and a first-order transition between those trivial phases at $F$09 (Schulz et al., 2012). A later tensor-network variational study, using PEPS ansätze adapted to the string-net condensate, found instead first-order transitions for Fibonacci near

$F$10

while reproducing the known second-order behavior in the $F$11 case (Schotte et al., 2019). This divergence indicates that the precise critical behavior of string-tension-driven breakdown of non-Abelian string-net order remains subtle (Schulz et al., 2012, Schotte et al., 2019).

Loop observables provide another dynamical probe. In the presence of string tension, the perturbed Hamiltonian

$F$12

interpolates between a topological deconfined phase and a trivial confined phase (Ritz-Zwilling et al., 2020, Schotte et al., 2019). For arbitrary modular input categories in the charge-free sector, Wegner–Wilson loop operators $F$13 obey a perimeter law in the topological phase and a modified area law in the trivial phase, with leading coefficients controlled by quantum dimensions $F$14 and total quantum dimension

$F$15

(Ritz-Zwilling et al., 2020). For Abelian $F$16 with $F$17, one has the exact identity

$F$18

so these excitations remain completely deconfined (Ritz-Zwilling et al., 2020).

The sharpest modern limitation of the Levin–Wen construction concerns non-unitary input. A recent analysis separates topological consistency from physical realizability and proves that a physical realization on a Hilbert space exists if and only if the spherical fusion category is pseudo-unitary (Yang, 11 Jan 2026). The benchmark non-unitary case is the Yang–Lee category $F$19, with simple objects $F$20, fusion rule

$F$21

and negative quantum dimension

$F$22

(Yang, 11 Jan 2026). Numerically, one can satisfy the pentagon equations while failing the unitarity condition: $F$23 The best-fit $F$24-matrix is pseudo-unitary with respect to

$F$25

but not unitary in the ordinary sense (Yang, 11 Jan 2026). The resulting state space is a Krein space rather than a Hilbert space, and the Hamiltonian is pseudo-Hermitian rather than Hermitian: $F$26 The theorem proved there states: $F$27 This refines the traditional presentation of Levin–Wen models by showing that sphericality alone guarantees topological consistency, but positive-definite quantum mechanics requires pseudo-unitarity (Yang, 11 Jan 2026).

Ground-state and excited-state degeneracies also reveal where universal topological data end and model-dependent structure begins. In generalized string-net models restricted to the charge-free fluxon sector, the ground-state degeneracy on a closed surface depends only on the Drinfeld center: $F$28 but excited-state degeneracies include additional factors $F$29 from the tube algebra decomposition (Ritz-Zwilling et al., 2023). The general puncture-space dimension becomes

$F$30

showing that Morita-equivalent categories with the same center can nevertheless differ in their excited-level degeneracies (Ritz-Zwilling et al., 2023).

Taken together, these developments define the modern understanding of Levin–Wen string-net models. They are a categorical construction of exactly solvable, generally nonchiral topological lattice phases; they realize Drinfeld centers of suitable fusion-category input; they encompass quantum doubles and admit multiple Hamiltonian generalizations; and they are now understood to be bounded on the physical side by pseudo-unitarity and on the phase-classification side by gappability criteria (0907.2670, Hahn et al., 2020, Lin et al., 2014, Kim et al., 2024, Yang, 11 Jan 2026).