Duality Defects in Quantum Field Theories

• Duality defects are topological interfaces that localize duality transformations in spacetime, enabling non-invertible symmetry operations.
• They occur in various settings—from 2D conformal field theory to higher-dimensional gauge and lattice models—where they implement order–disorder transmutation.
• Their fusion typically yields a sum of invertible symmetry defects, revealing rich categorical structures that constrain duality relations and infrared dynamics.

Duality defects are topological defects or interfaces that localize a duality transformation in spacetime. In two-dimensional conformal field theory they appear as line defects that convert order fields into disorder fields and whose fusion with the reverse defect reduces to symmetry lines; in other settings they arise as codimension-two monodromy defects, as codimension-one interfaces obtained by gauging only half of spacetime, or as self-orbifold interfaces in lattice, vertex-operator-algebra, and gauge-theoretic constructions (0909.5013, Gadde et al., 2014, Choi et al., 2021). A recurring feature is non-invertibility: the defect typically does not square to the identity, but to a sum or superposition of invertible symmetry defects. This places duality defects at the intersection of orbifolding, generalized symmetry, categorical symmetry, and symmetry topological field theory.

1. Defects, dualities, and locality data

In the two-dimensional CFT formulation, a defect line is a one-dimensional hypersurface on the world sheet across which fields may have discontinuous or nontrivial gluing behavior. If a defect with condition $X$ is placed on the unit circle, it defines a linear operator $D_X$ on the bulk state space $\mathcal H$. Conformal compatibility is encoded by

$[L_m-\bar L_{-m},D_X]=0,$

while topologicality is strengthened to

$[L_m,D_X]=0=[\bar L_m,D_X].$

A topological defect is therefore transparent to the stress tensor and may be deformed without changing correlators, provided it does not cross field insertions or other defects (0909.5013).

A standard misconception is that a defect is merely a folded boundary condition. The defect-line approach explicitly rejects this: folding misses defect junctions, and these junctions carry essential structural information. In rational CFT, this additional local data controls associativity, duality transformations, and orbifold consistency (0909.5013).

A distinct but closely related usage appears in higher-dimensional QFT, where duality defects are codimension-two objects around which couplings or parameters undergo a duality monodromy. In this formulation a defect is specified by an element $\gamma\in\Gamma$ of the duality group, and the defect is labeled by the conjugacy class of $\gamma$. These objects are naturally interpreted as boundaries of codimension-one duality walls (Gadde et al., 2014). This suggests that the term “duality defect” refers less to a fixed codimension than to a fixed role: the localization of a duality action.

2. Fusion, order–disorder duality, and categorical structure

For topological defects one can define fusion geometrically, and in operator language the fused defect is represented by $D_XD_Y$. The canonical example is the Ising model, with elementary topological defects ${\bf 1},\sigma,\varepsilon$ obeying

$$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$

Here $D_X$0 is group-like and implements the $D_X$1 spin-flip symmetry, whereas $D_X$2 is the duality defect (0909.5013).

The defining physical mechanism is order–disorder transmutation. In the Ising model, moving the $D_X$3-defect past the spin field $D_X$4 produces the disorder field $D_X$5, with an $D_X$6-line ending at the corresponding junction. Both $D_X$7 and $D_X$8 have conformal weights $D_X$9, while the topological junction field has weight $\mathcal H$0. Inserting a closed $\mathcal H$1-loop and sliding it across insertions therefore implements Kramers–Wannier duality directly on correlators (0909.5013).

In rational CFT this mechanism admits an intrinsic algebraic criterion: a defect $\mathcal H$2 is a duality defect if $\mathcal H$3 is a sum of group-like defects only, where $\mathcal H$4 denotes the defect with reversed orientation. By contrast, group-like defects satisfy $\mathcal H$5 and form a symmetry group acting on correlators (0909.5013). In this sense, group-like defects encode ordinary symmetries, whereas duality defects encode order–disorder transformations whose square closes only onto symmetries.

Defect junctions supply the higher coherence data. For group-like defects $\mathcal H$6, rebracketing the two ways of composing three junctions differs by a phase $\mathcal H$7, and the corresponding cohomology class lies in $\mathcal H$8. This 3-cocycle is the associator data of the monoidal category of invertible defects. Orbifolding by a subgroup $\mathcal H$9 requires $[L_m-\bar L_{-m},D_X]=0,$0, and the remaining ambiguity is discrete torsion in $[L_m-\bar L_{-m},D_X]=0,$1 (0909.5013).

The same perspective leads to generalized orbifolds. A topological defect $[L_m-\bar L_{-m},D_X]=0,$2 need not be a sum of group-like defects; it suffices that $[L_m-\bar L_{-m},D_X]=0,$3 admit junction fields obeying the same local move invariance. In the rational case this is controlled by special symmetric Frobenius algebras in the category of $[L_m-\bar L_{-m},D_X]=0,$4-modules, and the correlator construction via Frobenius algebras agrees with the defect-network construction. One major consequence is that every rational CFT with identical left and right chiral symmetry $[L_m-\bar L_{-m},D_X]=0,$5, well-defined on closed oriented surfaces of genus $[L_m-\bar L_{-m},D_X]=0,$6 and $[L_m-\bar L_{-m},D_X]=0,$7, with a unique vacuum and non-degenerate two-point function, is a generalized orbifold of the Cardy case for $[L_m-\bar L_{-m},D_X]=0,$8 (0909.5013).

A later structural refinement asks when duality defects are group-theoretical. For $[L_m-\bar L_{-m},D_X]=0,$9 in $[L_m,D_X]=0=[\bar L_m,D_X].$0d and $[L_m,D_X]=0=[\bar L_m,D_X].$1 in $[L_m,D_X]=0=[\bar L_m,D_X].$2d, a duality defect is group-theoretical if and only if its Symmetry TFT is a Dijkgraaf–Witten theory; equivalently, the relevant gauge theory admits an EM-stable Lagrangian boundary condition. In $[L_m,D_X]=0=[\bar L_m,D_X].$3d this occurs if and only if $[L_m,D_X]=0=[\bar L_m,D_X].$4 is a perfect square, while in $[L_m,D_X]=0=[\bar L_m,D_X].$5d on spin manifolds it occurs if and only if $[L_m,D_X]=0=[\bar L_m,D_X].$6 and $[L_m,D_X]=0=[\bar L_m,D_X].$7 is a quadratic residue mod $[L_m,D_X]=0=[\bar L_m,D_X].$8 (Sun et al., 2023).

Higher-categorical versions of the same fusion pattern also occur. In $[L_m,D_X]=0=[\bar L_m,D_X].$9d, duality defects obtained by half-spacetime gauging of $\gamma\in\Gamma$0 obey a Tambara–Yamagami-type fusion 2-category, with

$\gamma\in\Gamma$1

In $\gamma\in\Gamma$2d, self-duality defects for finite abelian 1-form symmetry $\gamma\in\Gamma$3 are organized by generalized Tambara–Yamagami fusion 3-categories, or $\gamma\in\Gamma$4 categories, described via Brauer–Picard and Picard 4-groupoids (Cui et al., 2024, Bhardwaj et al., 2024).

3. Two-dimensional realizations in CFT and VOA theory

The Ising model remains the archetype, but two-dimensional realizations extend far beyond minimal models. Lin and Shao identified a genuinely non-invertible topological defect line $\gamma\in\Gamma$5 in the Monster CFT. The Monster theory is self-dual under the $\gamma\in\Gamma$6 orbifold, and the defect obeys

$\gamma\in\Gamma$7

so the defect category contains an Ising subcategory $\gamma\in\Gamma$8. The same work shows that fermionizing the Monster CFT with respect to $\gamma\in\Gamma$9 gives Baby Monster $\gamma$0 Majorana–Weyl fermion, and introduces a defect McKay–Thompson series twisted by $\gamma$1, whose coefficients decompose into dimensions of irreducible representations of $\gamma$2 (Lin et al., 2019).

In the holomorphic $\gamma$3 lattice VOA, duality defects coming from cyclic $\gamma$4 symmetries are classified by a two-step criterion: one chooses a non-anomalous cyclic symmetry and then an order-2 automorphism of the fixed-point VOA $\gamma$5 that fully swaps the electric and magnetic axes of

$\gamma$6

This produces a $\gamma$7 Tambara–Yamagami action and hence a non-invertible Kramers–Wannier duality defect. The construction is worked out explicitly for $\gamma$8, and the resulting defect partition functions are checked against fermionization and Potts-model constructions (Burbano et al., 2021).

A parallel program exists for $\gamma$9 meromorphic theories associated with Niemeier lattices. For $D_XD_Y$0-type Niemeier lattice CFTs, non-anomalous $D_XD_Y$1 orbifolds can map one meromorphic theory to another, and in self-dual cases one can extract duality-defect partition functions. Besides standard diagram automorphisms, exchange automorphisms between identical current-algebra factors generate additional defect partition functions; the resulting defects fall into either the Ising or Ising $D_XD_Y$2 Tambara–Yamagami class (Grover et al., 2023).

In $D_XD_Y$3 compact boson CFTs with discrete shift orbifolds, duality defects are classified into four types according to whether the T-duality identification uses mirror symmetry, spacetime inversion, both, or neither. The condition that a toroidal branch theory be self-dual under a shift-orbifold can be reformulated as quadratic equations in integers, and for “almost all” theories the solutions can be enumerated explicitly. The analysis also provides evidence for duality defects along parameter families such as $D_XD_Y$4 with $D_XD_Y$5 (Furuta, 2024).

4. Higher-dimensional generalizations and half-space gauging

A major modern development constructs duality defects by gauging generalized symmetries only on half of spacetime. For a $D_XD_Y$6-dimensional theory $D_XD_Y$7 with finite higher-form symmetry $D_XD_Y$8, if $D_XD_Y$9 and ${\bf 1},\sigma,\varepsilon$0, then gauging on one side of a codimension-one interface and imposing Dirichlet boundary conditions on the emergent gauge field produces a topological defect ${\bf 1},\sigma,\varepsilon$1. In ${\bf 1},\sigma,\varepsilon$2 dimensions this reproduces the Kramers–Wannier line; in ${\bf 1},\sigma,\varepsilon$3 dimensions, for ${\bf 1},\sigma,\varepsilon$4, one finds

${\bf 1},\sigma,\varepsilon$5

so the defect is non-invertible precisely because its self-fusion is a sum of symmetry surfaces rather than the identity (Choi et al., 2021).

A related construction starts not from self-duality under gauging alone, but from a mixed ’t Hooft anomaly between a 0-form symmetry and a discrete higher-form symmetry. After gauging the higher-form symmetry, the anomalous symmetry defect must be dressed by a lower-dimensional TQFT, and the resulting object becomes a Kramers–Wannier-like non-invertible defect. In ${\bf 1},\sigma,\varepsilon$6 dimensions this yields examples in ${\bf 1},\sigma,\varepsilon$7 Yang–Mills at ${\bf 1},\sigma,\varepsilon$8, ${\bf 1},\sigma,\varepsilon$9 $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$0 super Yang–Mills, and $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$1 $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$2 super Yang–Mills at $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$3 (Kaidi et al., 2021).

Maxwell theory on non-spin manifolds supplies a complementary continuum realization. Because line-operator statistics matter, one obtains four non-spin Maxwell theories distinguished by whether fundamental Wilson and ’t Hooft lines are bosonic or fermionic. Three are non-anomalous and related by electric or magnetic $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$4 one-form gauging, while the all-fermion theory is anomalous. Topological interfaces for $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$5 and $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$6, together with gauging interfaces built by half-gauging BF theory, can be stacked into composite self-maps at special couplings. When gauging is involved, the resulting duality defects are generally non-invertible and act as condensation defects projecting out odd Wilson or ’t Hooft lines (Kan et al., 2024).

In $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$7 dimensions, simultaneous half-spacetime gauging of $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$8 with trivial mixed anomaly produces non-invertible topological surface defects. If the theory is self-dual under the combined gauging operation, the defect $$D_\varepsilon D_\varepsilon=D_{\bf 1},\qquad D_\varepsilon D_\sigma=D_\sigma D_\varepsilon=D_\sigma,\qquad D_\sigma D_\sigma=D_{\bf 1}+D_\varepsilon .$$9 satisfies

$D_X$00

and the full structure is organized by a Tambara–Yamagami-type fusion 2-category (Cui et al., 2024).

Subsystem versions exist as well. In $D_X$01d lattice models with subsystem $D_X$02 symmetry, gauging produces a subsystem Kramers–Wannier transformation. Gauging only a half-space yields subsystem duality defects whose self-fusion gives a grid operator built from many subsystem symmetry lines rather than a single global symmetry element. Unlike defects of invertible subsystem symmetries, subsystem Kramers–Wannier defects are mobile in both spatial directions by local unitaries, although the corresponding duality symmetry is argued to be anomalous (Cao et al., 2023).

One dynamical issue is whether duality defects obstruct trivial gapping. For $D_X$03d $D_X$04 self-duality defects, direct analysis of one-form SPT phases yields strong arithmetic constraints: for bosonic odd $D_X$05, a trivially gapped infrared phase is possible only if each prime factor of $D_X$06 is $D_X$07; for bosonic even $D_X$08, no trivially gapped phase is possible. A crucial subtlety is that the mere existence of a duality defect does not automatically imply a nontrivial infrared phase, as shown by non-abelian examples related to $D_X$09 triality (Choi et al., 2021).

5. Geometric and string-theoretic realizations

A geometrically distinct viewpoint treats duality defects as codimension-two monodromy loci in physical spacetime. In four-dimensional $D_X$10 theories, couplings are regarded as background fields varying over a transverse plane $D_X$11, and supersymmetry requires a holomorphic map $D_X$12. For pure $D_X$13 Seiberg–Witten theory, the simplest map is $D_X$14, which identifies the physical $D_X$15-plane with the Seiberg–Witten $D_X$16-plane. The monopole and dyon singularities then become actual spacetime locations of codimension-two duality defects, with monodromies

$D_X$17

This picture generalizes to holomorphic Lefschetz fibrations, where singular fibers correspond to duality defects and duality twists become Dehn twists in the mapping class group (Gadde et al., 2014).

The same theme admits a six-dimensional origin. Compactifying the $D_X$18d $D_X$19 theory on an elliptically fibered Kähler three-fold $D_X$20 yields $D_X$21d $D_X$22 super Yang–Mills with spacetime-dependent complex coupling $D_X$23. Singular fibers produce 3d duality walls along branch cuts of $D_X$24 and 2d surface defects around which the 4d theory undergoes $D_X$25 duality. Localized chiral fields on the defects arise from modes of the six-dimensional two-form $D_X$26, and intersections of surface defects enhance flavor symmetry, producing a nested 4d–3d–2d–0d “Matroshka” defect structure (Assel et al., 2016).

A top-down type-IIB construction places D3-branes at the tip of a Calabi–Yau cone $D_X$27. Reduction of the IIB topological term on $D_X$28 yields a 5d symmetry TFT encoding generalized symmetry data of the 4d theory. Wrapped $D_X$29 7-branes on $D_X$30 generate codimension-one topological duality and triality defects via their $D_X$31 monodromies. Different field-theoretic realizations of duality defects, including half-space gauging and Kramers–Wannier-like constructions, correspond in this framework to different placements of the 7-brane branch cuts in the 5d bulk (Heckman et al., 2022).

Related codimension-two defects in the $D_X$32d $D_X$33 theory $D_X$34 are not, in this formulation, themselves identified as duality defects, but they furnish a crucial organizing language for four-dimensional duality. Regular defects are described uniformly by Hitchin singularities, Nahm boundary conditions, and Toda primaries, with the correspondence mediated by nilpotent orbits, Springer theory, and order-reversing duality maps. This dictionary explains how puncture data encode S-duality frames in class $D_X$35 theories (Balasubramanian, 2014).

6. Lattice dynamics, transmission, and current directions

On the lattice, duality defects can be made fully operational. In one-dimensional quantum spin systems, a duality or symmetry can be represented by a matrix product operator $D_X$36, and applying $D_X$37 only on half the chain produces a Hamiltonian

$D_X$38

with a localized impurity Hilbert space emerging from the virtual bond space of the MPO. When the two halves are related by a duality, the resulting interface is topological: it can be moved by a unitary rearrangement, and an incoming wavepacket is transmitted with unit transmittance rather than reflected. The outgoing excitation, however, is generally re-encoded as a nonlocal string-like excitation. In the Ising example, a spin-flip wavepacket passes through the Kramers–Wannier interface with 100% transmission and emerges as a domain wall or string excitation; analogous behavior occurs in the Kennedy–Tasaki duality of the spin-1 chain and in a $D_X$39 example (Ueda et al., 30 Oct 2025).

A weaker but structurally related dynamical theme concerns long-lived defect-bound observables. In a deformed transverse-field Ising chain with a Kramers–Wannier duality defect, the defect Hamiltonian can be constructed in three ways: by a half-chain Kramers–Wannier transformation, from the Ising fusion category, or by a defect-modified weak integrability-breaking deformation. The defect remains topological in the sense that its position can be shifted by a local unitary. Defect-modified higher charges become quasi-conserved under the weak deformation, displaying slower decay, and the corresponding defect-twisted Floquet model contains an isolated zero mode in the integrable limit whose decay under deformation is again parametrically slow (Yan et al., 2024).

These lattice results sharpen the physical meaning of duality defects. They show that non-invertibility is not merely a formal feature of fusion algebra: it can control transmission, locality, dressed excitations, quasi-conservation, and the fate of defect zero modes. This suggests a broad operational interpretation in which a duality defect is a transparent interface for energy transport but not for operator locality.

Across the subject, several points remain structurally important. A defect is not merely a boundary condition in disguise (0909.5013). Not every duality defect is group-theoretical (Sun et al., 2023). The existence of a duality defect can constrain infrared behavior, but it does not do so uniformly in all theories (Choi et al., 2021). What is uniform is the central mechanism: a duality defect packages the passage between equivalent descriptions of a theory into a topological object whose fusion, junctions, and twisted sectors record more information than an ordinary symmetry line or wall can carry.