Surface Code Twist Defects

• Surface code twist defects are extrinsic endpoints in toric codes that exchange electric and magnetic anyons, implementing an anyon automorphism.
• They introduce mixed Pauli stabilizers, often with a Y component, which modify logical operator topology and enable novel braiding and lattice surgery protocols.
• Twist defects facilitate enhanced code rates and fault-tolerant Clifford operations by supporting Majorana-like fusion and code deformations.

Surface code twist defects are extrinsic endpoints of domain walls in two-dimensional surface or toric codes that exchange electric and magnetic anyons, $e \leftrightarrow m$. In stabilizer language they mark locations where the usual CSS separation into purely $X$-type and $Z$-type checks fails and mixed-Pauli stabilizers appear, often with a single $Y$ component. Because strings crossing the wall change Pauli type, twists modify the topology of logical operators, support Ising-like or Majorana-like fusion behavior, and enable code deformations, braiding protocols, and lattice-surgery measurements that are not available with ordinary rough and smooth boundaries alone (Kesselring et al., 2018, Zheng et al., 2015).

1. Topological meaning and anyon-theoretic interpretation

The anyon content relevant to the surface code is the toric-code set

$A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$

with fusion rules

$$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$

and mutual braiding $M_{e,m}=-1$. A transparent domain wall implements an anyon automorphism preserving fusion, topological spin, and braiding. The canonical surface-code wall is the electric-magnetic duality

$$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$

A twist defect is the endpoint of such a wall; transporting an anyon once around the twist applies the automorphism. In the toric code this gives the familiar projective non-Abelian fusion rule

$\tau\times\tau = 1+\psi,$

which is the sense in which twists behave as Ising-like or Majorana-like defects (Kesselring et al., 2018).

This picture extends the conventional distinction between rough and smooth boundaries. Rough boundaries condense $e$, smooth boundaries condense $X$0, and a corner where the boundary type changes is therefore an endpoint of an $X$1 wall. One immediate consequence is that the ordinary planar code already contains twist defects at its corners, even before any interior dislocation or branch cut is added (Brown et al., 2016).

Within the broader symmetry-defect framework, the surface-code $X$2-$X$3 twist is the $X$4 case of a genon or twist defect associated with an anyon-permuting symmetry. For the toric/surface-code limit, the $X$5 reduction reproduces Majorana-like Ising braiding up to an overall phase, while more general host phases can realize projective braiding that is not equivalent to ordinary Ising braiding (Barkeshli et al., 2012).

2. Stabilizer realizations and graph-based formalisms

Several equivalent lattice realizations are used in the literature. In the standard toric-code presentation with qubits on edges, the stabilizers are

$X$6

In the Wen/Bombín vertex-qubit formulation, bulk plaquettes can be written as

$X$7

while a twist plaquette becomes a pentagon operator

$X$8

The appearance of $X$9 is the local stabilizer signature of the branch cut that exchanges $Z$0 and $Z$1 across it (Zheng et al., 2015).

A more general formalism associates a surface code to any graph $Z$2 embedded on any $Z$3-dimensional manifold such that qubits are associated to the vertices of $Z$4, stabilizers are associated to faces, and twist defects are associated to odd-degree vertices. In that setting, the central combinatorial notion is checkerboardability, defined by the condition that the dual of $Z$5 is bipartite. When the dual is bipartite, the construction reduces to Kitaev’s CSS surface-code layer with a standard $Z$6 partition. When the dual is not bipartite, the usual CSS separation fails, and face stabilizers require mixed Pauli assignments. The generic form is

$Z$7

with $Z$8 selected by local rotation-system data and parity information (Sarkar et al., 2021).

This graph-based perspective puts twist defects on the same footing as topological graph properties. It reproduces known surface-code families with and without twists, produces new examples, and relates code parameters to graph invariants such as genus, systole, and face-width. It also places the usual locality tradeoff in the form

$Z$9

with $Y$0 the number of physical qubits, $Y$1 the number of logical qubits, and $Y$2 the distance (Sarkar et al., 2021).

Concrete mixed stabilizers near domain walls and twist endpoints appear in several constructions. One explicit example uses a modified plaquette

$Y$3

along the wall and a twist-endpoint stabilizer

$Y$4

showing directly that twists are not merely reduced boundary checks but genuinely mixed $Y$5 objects (Jia et al., 2018).

3. Boundaries, corners, logical operators, and Majorana structure

The effect of a twist is most visible at the level of strings and logical operators. In the checkerboardable CSS case, logicals are ordinary homologically nontrivial $Y$6- and $Y$7-strings. In the presence of a twist, a string crossing the defect line changes type: an $Y$8-string becomes a $Y$9-string, and a $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$0-string becomes an $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$1-string. Accordingly, a single continuous curve can realize a mixed logical operator, and strings can terminate at twist endpoints where both $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$2 and $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$3 may condense (Kesselring et al., 2018, Jia et al., 2018).

This is why corners of the planar code are operationally equivalent to twists. If a smooth boundary segment is covered by an $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$4 defect line drawn just inside the edge, the places where rough and smooth boundaries meet are exactly the twist endpoints. In this description the standard planar-code logicals are recovered unchanged, but their support is reinterpreted as strings attached to four boundary twists (Brown et al., 2016).

Pairs of twists encode fermionic parity in the same way that pairs of Majorana zero modes do. In the explicit Jordan-Wigner analysis of surface-code twists, each twist binds an unpaired Majorana zero mode, and the parity of a pair is

$A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$5

This parity operator commutes with all stabilizers but is not itself a product of local plaquettes, so it acts as a logical operator. The resulting fusion structure is the same $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$6 ambiguity familiar from Ising anyons: two twists fuse either to vacuum or to the fermion $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$7, distinguished by their parity (Zheng et al., 2015).

A common misconception is that twist defects are interchangeable with holes or with ordinary boundary terminations. Holes modify local checks around a two-dimensional puncture but preserve pure $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$8-type or pure $A=\{1,e,m,\psi\}, \qquad \psi=e\times m,$9-type structure on their boundaries. Twists instead require a one-dimensional wall implementing $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$0, mixed-type checks along that wall, and an endpoint stabilizer that condenses both $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$1 and $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$2 (Jia et al., 2018).

4. Braiding, code deformation, and lattice surgery

Because exchanging twists acts on the encoded Majorana parity, braiding and code deformation implement Clifford transformations. For the planar code viewed as four boundary twists, exchanging the two bottom twists maps

$$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$3

while exchanging the two left twists maps

$$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$4

These operations generate the single-qubit Clifford group up to Pauli-frame updates, and they can be realized by moving corners into the bulk through sequences of local single-qubit measurements and stabilizer updates (Brown et al., 2016).

A complementary approach is to track single-qubit Clifford gates classically and use twist-based lattice surgery only for the nontrivial logical parity measurements. In edge tracking, $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$5 and $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$6 are absorbed into a reassignment of which edge realizes $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$7, $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$8, or $$e\times e = m\times m = \psi\times \psi = 1,\qquad e\times m=\psi,$$9. The resulting surgery layer must then measure mixed parities such as $M_{e,m}=-1$0 or $M_{e,m}=-1$1, which is exactly what dislocations and twists permit. In this formulation, twists appear as the endpoints of dislocation lines on the surgery interface, and a weight-five twist check with a central $M_{e,m}=-1$2 realizes the required mixed observable (Litinski et al., 2017).

At the circuit level, explicit twist-based lattice surgery protocols require elongated checks and weight-five twist stabilizers, together with modified ancilla connectivity and scheduling. A protocol using a four time-step schedule for two-qubit gates showed that the timelike logical-failure threshold remains near $M_{e,m}=-1$3, that the threshold is slightly reduced relative to twist-free surgery, and that comfortably below threshold, with CNOT infidelities below $M_{e,m}=-1$4, the degradation is mild and preferable over proposed twist-free alternatives for $M_{e,m}=-1$5-containing measurements (Chamberland et al., 2022).

Twists also give a direct route to $M_{e,m}=-1$6-basis access. Diagonal fusion of corner twists across a single $M_{e,m}=-1$7 patch turns $M_{e,m}=-1$8 into a known product of post-transition stabilizers in a degenerate $M_{e,m}=-1$9 boundary configuration. The resulting in-place logical $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$0-basis measurement or initialization uses

$$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$1

syndrome rounds, never leaves the original patch bounding box, and preserves the full code distance $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$2 throughout (Gidney, 2023).

5. Twist-derived code families and resource trade-offs

Twist defects are not only computational primitives; they also define alternative patch geometries and higher-rate code families. A prominent example is the triangle code, a patch-based encoding built from a modified twist. In this construction all plaquette checks remain weight four, boundary loops have weight two, and only the central mixed-type plaquette at the origin contains a single $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$3. For odd distance $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$4, the code uses

$$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$5

data qubits per logical qubit, compared with $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$6 for a rotated surface-code patch, giving $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$7 fewer physical qubits at the same distance. The same work reports ideal-syndrome thresholds of about $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$8 and noisy-syndrome thresholds near $$W_{em}: e\mapsto m,\quad m\mapsto e,\quad \psi\mapsto \psi.$$9 for bit-phase-flip noise under MWPM decoding (Yoder et al., 2016).

Twists also appear in higher-symmetry patch families. Stellated surface codes place a twist at the center connected to the boundary by a domain wall and increase rotational symmetry $\tau\times\tau = 1+\psi,$0. For odd $\tau\times\tau = 1+\psi,$1, the encoding-rate constant is $\tau\times\tau = 1+\psi,$2; for even $\tau\times\tau = 1+\psi,$3, $\tau\times\tau = 1+\psi,$4; and $\tau\times\tau = 1+\psi,$5 as $\tau\times\tau = 1+\psi,$6 grows. These constructions generalize the “surface code with a twist” at $\tau\times\tau = 1+\psi,$7 while keeping local checks (Kesselring et al., 2018).

More aggressive packing is possible when twist pairs are yoked into dense planar architectures. On a regular two-dimensional hex grid, one proposal uses dense packing of twist defects together with new stabilizer measurement cycles using four layers of nearest-neighbor two-qubit gates. Under a one- and two-qubit $\tau\times\tau = 1+\psi,$8 uniform depolarizing model, the estimated encoding rate reaches up to $\tau\times\tau = 1+\psi,$9 that of a rotated surface-code patch. The same work gives padding-free lattice-surgery protocols in an optimal bounding box of $e$0 data and measurement qubits per patch and asymptotic dense-pack density approaching one logical qubit per $e$1 physical qubits (Low et al., 28 May 2026).

These constructions illustrate a recurrent trade-off. Twists improve code rate, support direct $e$2-access and Clifford operations, and enable denser patch layouts, but they replace the uniform CSS bulk with local mixed-type checks and more complicated decoding graphs. The practical question is therefore not whether twists change the code, but where the additional syndrome complexity buys enough reduction in area or spacetime volume to justify their use.

6. Experiments, synthetic twists, and current directions

The non-Abelian interpretation of surface-code twists has been demonstrated experimentally at small scale. A five-qubit truncation of a matching-code realization implemented the simplest exchange operation of two surface-code Majoranas on the IBM 5Q processor. In that experiment the expected rotation into the logical $e$3 basis was observed, and the measured correlation

$e$4

from $e$5 shots agreed with the predicted exchange behavior in the presence of device noise (Wootton, 2016).

A larger patch-based realization was reported on a $e$6-qubit superconducting quantum processor. There, merge and split, patch expansion and shrinkage, and domain-wall or twist-defect deformations were composed into logical routing, CNOT, Hadamard, and phase operations on distance-three rotated surface-code patches with multi-round syndrome extraction, neural-network decoding, and no post-selection. In that implementation, $e$7 was realized by twist-defect motion together with a final transversal Hadamard domain wall, and $e$8 was realized by teleportation using a $e$9 ancilla prepared in place by diagonal twist-defect motion. The reported average logical gate fidelities were $X$00 for $X$01 and $X$02 for $X$03 (Lin et al., 1 Jul 2026).

Recent work has also shifted attention from static lattice twists to synthetic twists generated by local perturbations of an otherwise standard surface-code Hamiltonian. In a Wen-plaquette realization with perturbation

$X$04

the active degrees of freedom along the perturbed cut map to a Kitaev Majorana chain. The topological regime corresponds to

$X$05

where unpaired Majorana zero modes localize at the cut endpoints and realize synthetic twist defects. Numerical spectra show a gap closing near the transition and exponential splitting of the endpoint-mode doublet with cut length, giving a controlled finite-size route to creating and moving twists without geometrically modifying the lattice (Kairys et al., 11 May 2026).

Related generalizations appear in Majorana surface codes, where twist defects can be bosonic or fermionic and can encode both logical qubits and logical Majorana zero modes. In that setting all Clifford gates on logical qubits can be implemented by braiding twist defects, and lattice-surgery methods can produce the effect of Clifford gates with zero time overhead (McLauchlan et al., 2022). This suggests that the notion of a twist defect is best understood not as a peculiar corner case of the qubit surface code, but as a general defect mechanism linking anyon permutation, mixed stabilizers, and fault-tolerant logical control across both bosonic and fermionic code architectures.