WorksPhysics and Astronomy
Local gluing classification and bulk-locked clock rings for models on a Klein bottle
Abstract
We classify local reflection gluings of the plaquette model on lattice Klein bottles and solve the obstructed literal seam. Within the monomial far-corner ansatz, all commuting gluings form a family labelled by a unit of . They exist for every at even seam length and, at odd length, exactly when is a square modulo . We determine the anyon transport, including exchange seams with for charge conjugation , and the relation phases that distinguish an unfrustrated code from a commuting but frustrated Hamiltonian. For the literal gluing, the noncommuting plaquettes form self-dual clock rings whose sectors are locked by the bulk. This gives an exact reduction of the spectrum to clock-ring spectra, closed-form ring ground energies for , and a closed-form ground-state degeneracy. We further determine twisted-coefficient homology on the stated surface domains and compare it with logical modules on registered lattice grids; composite dimensions need not give squares of cyclic groups. Dressed-isometry gates and monodromy are reported with their finite-grid scope. Finally, the qudit Gottesman–Knill theorem bounds the computational interpretation: Clifford deformations and Weyl measurements with stabilizer inputs and no external magic states cannot supply a non-stabilizer resource. Charge-conjugation measurements that do supply such states lie outside this operation model. The new results concern explicit seam classification, phase-sensitive consistency and bulk-locked spectra, not a new general no-go theorem or a universal computing architecture.
Introduction
A reflection gluing can preserve a lattice check algebra, frustrate its vacuum sector through a relation phase, or produce a noncommuting seam Hamiltonian. These possibilities are distinct even when the bulk topological order is unchanged. We study them explicitly for the plaquette model on a Klein bottle. Our main results are a classification of commuting local gluings within a stated ansatz and an exact reduction of the obstructed literal seam to bulk-locked clock rings.
In [1] we studied the toric code on a lattice Klein bottle. In the edge register the reflection gluing preserves the checks; in the site register (Wen’s plaquette model [2]) a Hadamard dressing restores commutation, while the literal gluing produces locked Majorana rings. The extension to adds structures absent at : a unit-labelled -family, an arithmetic obstruction at odd circumference, order-four exchange transport with , and logical modules with composite- torsion. Here is charge conjugation on anyon labels. None of these label relations alone specifies a measurement or a non-Clifford resource.
Contributions and results map. Theorems 1 and 2 classify all commuting gluings within the local monomial far-corner ansatz, determine their seam actions and code dimensions, and track the relation phases needed for an unfrustrated code. Theorems 3 and 4 reduce the non-commuting literal seam to bulk-locked clock rings with bulk-constrained sectors. These are the two central microscopic results. Theorem 5, on its stated domain, and the separate projective-plane calculation in Example 1 describe twisted homology. Propositions 2, 3 and 4 compare logical modules, dressed isometries and monodromy with their stated proof, argument and grid support. Finally, Corollary 1 and Remark 2 delimit the computation allowed by the operation model below; Proposition 5 identifies the non-stabilizer measurements in the Wigner examples.
Relation to known results. Non-orientable qudit homological codes and parity-invariant-anyon counts are known [3, 4], as are reflection-enriched phases [5] and cross-cap Clifford gates at [6]. Recent quotient-code methods connect twisted compactification to logical operators and gates [7]. We do not claim these general principles, the mere possibility of order-four reflection transport, or stabilizer closure as new. Our contribution is their resolution in the specified plaquette realization: the complete local -gluing classification, its phase-sensitive consistency conditions, and the locked clock-ring spectrum of the literal obstruction. The composite- and gate results retain the evidence levels of Table 1. § 7 compares the constructions and assumptions in detail. The results of Ref. [1] are re-derived rather than re-claimed.
| result | content | support | provenance |
| Thm. 1 | commuting gluings = -family; odd iff | proof; enumeration; grid | pred. E-1 H1, H2 |
| Thm. 2(i, iii) | ; frustration iff odd and ; Lemma 1 | proof; grid | pred. E-1 H3, H4; Lemma 1 post hoc |
| Thm. 2(ii, iii) | seam action ; ; ; phases | proof; sealed property checks | cross (not pre-registered) |
| Thm. 3, 4 | literal gluing: clock rings, locking rule, spectrum, , GSD | proof; grid; numerics | pred. E-2 H1–H4 (reduced coverage) |
| Prop. 1 | row-exchanging reflection ; Kramers–Wannier action | proof; grid | pred. E-2 H5 |
| Prop. 2 | logical module = twisted homology (edge / site register) | proof / grid, argument | pred. E-3 H1 |
| Sec. 5.1 | logical operators; distances (at the listed sizes) | grid | pred. E-3 H2, H3 |
| Prop. 3 | admissible dressed isometries, logical actions, gate groups | argument; grid | pred. E-3 H4, H5 |
| Thm. 5 | : consistency, , block quotient, direct sums | proof; grid | pred. E-4 H1; other structures: reading |
| Remark 1 | -defect fusion count consistent with ( odd) | arithmetic consistency | pred. E-4 H2 |
| Prop. 4 | double-winding closure counts; monodromy multiplier | grid | pred. E-4 H3; : reading |
| Cor. 1 | no non-stabilizer resource from cross-caps within | proof (Gottesman–Knill) | not computed |
| Prop. 5 | Wigner minima; outcome probabilities of three measurements | proof; grid | pred. E-4 H4 |
| Sec. 6.2 | no protection from non-orientability itself | hand assessment | not computed |
Table 1. Results and their evidence (labels as in Sec. 1; grids in Appendix A.1). “pred.”: pre-registered prediction (stage, hypothesis); “cross”: sealed cross-check. Check counts and details: Appendix A.3.
Operation model and computational boundary. For the computational discussion we specify : static codes generated by Weyl stabilizers; code deformations implemented by Clifford unitaries and Weyl measurements; classical feed-forward; stabilizer inputs and no external non-stabilizer states. Lattice isometries with single-site Clifford dressings and cross-cap moves belong to only when implemented by these operations. The qudit Gottesman–Knill theorem then excludes production of a non-stabilizer state, independently of the seam classification. In proposals using topological order, additional resources enter through a symmetry-constrained charge measurement [8] or coherent interferometry with charge-conjugation defects [9]. Extra symmetry and fractionalization data, cross-cap-state preparation, coherent superpositions of mobile-defect paths, dynamical non-Pauli measurements and non-Clifford pumps are not supplied by our static gluing and lie outside .
Scope. The classification is restricted to local monomial far-corner gluings of the fixed bulk plaquette model. We construct lattice codes only on the alternating Klein bottle; elsewhere we compute homology without asserting a lattice realization. In the exchange branch the double-winding external monodromy is , but a single passage is , and at the exchange is not . Fusion counting is a consistency check, not an implementation of a defect theory. The no-go applies to , not to arbitrary physical manipulations of cross-caps. No protected non-Clifford implementation, noise threshold or universal architecture is established.
Method and evidence. The work was done in five stages: E-0, an engine calibrated against the results of Ref. [1] (no pre-registration), and E-1 to E-4, whose predictions and grids were frozen before the stage’s runs; results known beforehand are listed in each pre-registration and not counted as predictions. Each final run was sealed by sha256 after development or rehearsal runs (Appendix § 10). Table 1 labels each result by its support—proof, argument (a sketch whose general validity is not established), grid (exhaustive on the pre-registered grid), numerics—and its provenance: pre-registered prediction, cross-check (sealed, not pre-registered), reading (sealed, without registered expectation) or post hoc (after sealing, unsealed). Results of Ref. [1] that are the case of ours are re-derived, not re-claimed.
Model and conventions
Weyl operators. With , and , so . Weyl strings , (up to phases, which all computations track exactly), satisfy with .
Site register. A qudit sits on every site , , . The plaquette , with corners SW , SE , NE , NW , carries
the plaquette model [10]; for it is Wen’s model, [1, 2]. Plaquettes and sites have colour A ( even) or B (odd). A site receives its -type factors from the two plaquettes of its own colour and its -type factors from the other two. Conjugating the odd sites by the Fourier transform (, ) makes the A plaquettes -type and the B plaquettes -type: on a torus with even sides Eq. (1) is the toric code of the lattice whose vertices, edges and faces are the A plaquettes, the sites and the B plaquettes. A plaquette with eigenvalue carries charge ; the charges of the two colours are the anyons and , with labels , braiding phase and spin ; a one-site translation exchanges them [2].
Edge register (control). Qudits sit on the oriented edges and ; carries on the edges leaving and on those entering it, carries along the counterclockwise boundary ( along the arrow). This is the toric code [11], the homological code of the cellularization.
Mirror Klein bottle. Row is identified with row 0 by (offset ). In the edge register the vertical edge leaving enters the vertex , and the top edge of the plaquette , traversed counterclockwise, is the row-0 edge traversed along its arrow, so the seam-row carries there instead of . The code is again a homological code of a cellulation of the Klein bottle , and the seam reverses the sign of fluxes: on . In the site register the seam plaquettes have their far corners on row 0, NE at column and NW at column (Figure 1a); are the row-0 plaquettes, all others are bulk plaquettes. There are plaquettes and qudits. The reflection reverses the pattern , so the far corners must be chosen:
Figure 1. (a) Mirror Klein bottle of the site register (, , ; colours A/B). The virtual row is row 0 read at column (top labels). A seam plaquette keeps , on its near corners; its far corners land on row 0 at columns and , where the -family places and (Theorem 1). (b) Non-commutation graph of the literal gluing (): (squares) and (circles), labelled by column; meets with phase and with , giving two rings of length (even ) or one ring of length (odd ) (Theorem 3). For these are the graphs of Ref. [1], Fig. 2(a,b).
Definition 1 (local monomial far-corner gluing). Assign to every column of row 0. In the NE far corner (column ) carries and the NW far corner (column ) carries ; everything else is as in Eq. (1). The literal gluing has , : the far corners keep and .
Phases on the monomials affect only the scalars of relations Theorem 2(iii). By Theorem 1 the commuting gluings are the -gluings: the NE and NW far corners on column carry and , with for even and for odd ; is the Fourier dressing (, alternating by column), and , is the Hadamard-dressed gluing of Ref. [1]. For even there are two branches. In the diagonal branch ( odd) every site still receives its -type factors from plaquettes of its own colour: the colouring continues across the seam, which maps each anyon type to itself. In the exchange branch ( even) the colours, and with them and , are exchanged across the seam. For odd the colouring is inconsistent around and there is one charge type. “Grid” statements hold on the pre-registered grids (Appendix A.1) and are not extrapolated.
Local gluings: the -family
Theorem 1 (local gluings). Let , , . A local monomial far-corner gluing makes all plaquettes commute if and only if, for every column ,
Equivalently, the NE and NW far corners on column carry and with and , so for even and for odd . For even every unit gives a commuting gluing ( of them); for odd the chain closes iff . Hence commuting gluings with odd exist iff is a square mod , i.e. iff and every odd prime factor of is .
Proof. shares far-corner sites only with (; distinct as ) and with , whose far corners sit on columns ; all other overlaps are those of the torus and cancel ( keeps the row-0 plaquettes away from the seam row). Writing , , , , and (one near and one far corner in common). Setting them to zero gives (2); then is a unit and . The recursion has period two, so it closes on even cycles for every and on odd ones iff . The last statement is the classical criterion for . □
For even every unit works, and for odd only the closing roots ; locally every unit is allowed because the -gluing is the quotient by the glide dressed with , ( or by sublattice), which maps every plaquette to some (derivation; checked post hoc): a symmetry of the stabilizer group for every , of only for . The arithmetic condition is the integer set of the time-reversal criterion for Chern–Simons theory [12]; specific here is its role as a criterion for local lattice gluings.
Theorem 2 (seam action and code dimension). For a -gluing with , let be the relation module and .
(i) in the diagonal branch, and in the exchange branch and for odd .
(ii) A charge carried upward through the seam has its label mapped, in the diagonal and the exchange branch respectively, by
and for odd (one charge type) by with . Both matrices invert the braiding phase and the spin, , and , the number of -invariant labels, in every case.
(iii) With phase-free monomials all relation scalars are , except for odd and , where and no state satisfies all plaquettes. Phases change only the scalars: multiplying one seam plaquette by leaves unchanged and, for even in the exchange branch or with odd , where is generated by , flips its scalar. The code dimension is whenever all scalars are 1; only is an invariant of the gluing.
Lemma 1 (sign of the dressed product). For and every -gluing, .
Proof. and satisfy and ( odd), the qubit Pauli relations. , and as every is odd the far corners become (NE) and (NW): the product is that of the Hadamard-dressed gluing. Order its plaquettes by and multiply site by site. A site with receives its four factors in the order () or (): product . Row-0 site receives from , reversed at , followed by from the far corners of , reversed at ; two pairs in the same order give or , in opposite orders . So the sign is at the sites if , and at all sites if ; either way .
For this sign was only observed, at the sizes computed, in Ref. [1]; the proof above was supplied by the internal review of draft v0.1 and checked post hoc.
Proof of Theorem 2. (i) iff the exponents vanish on every site. A site with is SW and NE corner of and and SE and NW corner of and , so is constant on colour classes, and ( for odd ). Row-0 site gives and , and have the same colour iff is odd. Hence (diagonal), and (exchange), (odd ), with no division by non-units. (ii) (Derivation.) A at the last site of a string along is continued through the NW far corner of by on row 0; an string is continued through the NE far corner by . Each string carries one colour, so a label continued on colour A is multiplied by and one continued on colour B by , colours being preserved (diagonal) or exchanged (exchange branch): Eq. (3), up to a sign convention for labels. The entries of multiply to , which inverts and , and the equations of (i) are the fixed-point equations of . (iii) A relation supported on one colour has only -type or only -type factors on each site and scalar 1; in the diagonal branch these generate . Otherwise, for even , is generated by , whose scalar is 1 for ( and commute) and for (Lemma 1). Multiplying by multiplies that scalar by .
The diagonal branch with is an order-two parity action, with as for the edge register; the exchange branch realizes an order-four pair. (), and encodes nothing for odd . For odd the case is the sign-frustrated dressed gluing of Ref. [1]; on the grid the frustration recurs for , and the codes are one-dimensional for .
The obstructed seam: locked clock rings
Theorem 3 (obstruction of the literal gluing). Let , , , any . In the literal gluing the non-commuting pairs are exactly and , with and . They form two cycles of length for even and one of length for odd (Figure 1b). Along consecutive operators satisfy with no operator inverted, so each cycle is a self-dual clock ring of (even ) or (odd ) sites, with the seam plaquettes as field terms and the row-0 plaquettes as bond terms.
Proof. With , carries on column and on of row 0, and carries on , on . For the common site contributes , for it contributes , and for both common sites carry commuting factors; other overlaps cancel as on the torus (also for ). A step along the cycle advances the column by two, preserving its parity for even and visiting all columns for odd . , represents the ring.
Let and be the products of the seam and of the row-0 plaquettes of ring (ring charge and ring twist); let on sites mod , restricted to , with lowest eigenvalue at , and let if and are both even, otherwise.
Theorem 4 (locking rule and exact reduction). For the literal gluing with , , : (a) the relations form , generated by the product of all plaquettes and, for , the product of the colour-A plaquettes, which commute pairwise; both scalars are 1, the product of all plaquettes taken in the order , bulk (the index order gives the same); (b) the centre is generated by the bulk plaquettes , the ring charges and the ring twists ; with eigenvalues , , , the realized joint eigenvalues are exactly those with
and, for , , where is the ring whose seam plaquettes have colour A and the other; each realized joint eigenspace has dimension (even ) or (odd ), in which each ring carries one irreducible representation of dimension with multiplicity (even ) or (odd ), independent of except through the parity of ; (c) every eigenvalue of is , over the allowed labels, with multiplicity times the ring multiplicities; (d) the ground states have all and every ring in :
For this is the result of [1]; for a ring is two Ising chains and as for , so every lattice has the same for as for . For and we give no closed form for and use the numerical ground energy of the ring.
| size | states | dev. | GSD | ||
| 2 | 4 096 | d | |||
| 3 | 6 561 | d | |||
| 4 | 4 096 | d | |||
| 3 | 19 683 | d, L | |||
| 4 | 531 441 | L | |||
| 5 | 1 953 125 | L | |||
| 3 | 13 458 907 | g | |||
| 4 | 16 777 216 | g | |||
| 6 | 10 077 696 | g |
Table 2. Ground energies of the literal gluing, Eq. (5), against exact diagonalization on 9 of the 17 lattices of E-2 (dev. = largest deviation over ). GSD (): counted from complete spectra (d: dense; b: block-resolved by bulk labels), or of Theorem 4 in italics where Lanczos does not resolve degeneracies (L: with low levels; g: ground energy only).
Proof. (a) Each site carries one , , , , so relation coefficients are constant on colour classes; the seam identifies with , which crosses colours unless is even; nothing is divided, so composite adds no torsion. In the stated order every site contributes the scalar 1 except, for , two row-0 sites contributing and ; the cancellation is global, not site by site. For the non-commuting pairs join opposite colours, and each site carries and , or and , from the colour-A plaquettes. (b), (c): the cycle algebra has a unique irreducible representation, of dimension , for each central character; count dimensions. For fixed , on all states has non-positive off-diagonal elements in the basis and is irreducible, so by Perron–Frobenius its ground state is unique, positive, and has . The ring translation exchanges and , so and for . For , a Gray-code map writes the ring as two critical Ising chains.
:::
Proposition 1 (row-exchanging reflection). The site permutation , for , implemented by the unitary , , is a symmetry of , and exchanges the seam row and row 0 ( up to inverses), hence the field and bond terms of the rings: for odd it acts on the ring as , a Kramers–Wannier duality followed by a translation; for even it exchanges the two rings. A uniform detuning of the seam row against row 0 breaks .
Proof. Lift to the double cover, where is ; bulk plaquettes map as .
For reweighted plaquette terms, invariance under and the half translation forces both ring masses to vanish for ; for a staggered @page 5/15 starts_mid=1 ends_mid=1 and , with and . They form two cycles of length for even and one of length for odd (Figure 1b). Along consecutive operators satisfy with no operator inverted, so each cycle is a self-dual clock ring of (even ) or (odd ) sites, with the seam plaquettes as field terms and the row-0 plaquettes as bond terms.
Proof. With , carries on column and on of row 0, and carries on , on . For the common site contributes , for it contributes , and for both common sites carry commuting factors; other overlaps cancel as on the torus (also for ). A step along the cycle advances the column by two, preserving its parity for even and visiting all columns for odd . , represents the ring.
Let and be the products of the seam and of the row-0 plaquettes of ring (ring charge and ring twist); let on sites mod , restricted to , with lowest eigenvalue at , and let if and are both even, otherwise.
Theorem 4 (locking rule and exact reduction). For the literal gluing with , , $L_y\geq 2: (a) the relations form , generated by the product of all plaquettes and, for , the product of the colour-A plaquettes, which commute pairwise; both scalars are 1, the product of all plaquettes taken in the order , bulk (the index order gives the same); (b) the centre is generated by the bulk plaquettes , the ring charges and the ring twists ; with eigenvalues , , , the realized joint eigenvalues are exactly those with
and, for , , where is the ring whose seam plaquettes have colour A and the other; each realized joint eigenspace has dimension (even ) or (odd ), in which each ring carries one irreducible representation of dimension with multiplicity (even ) or (odd ), independent of except through the parity of ; (c) every eigenvalue of is , over the allowed labels, with multiplicity times the ring multiplicities; (d) the ground states have all and every ring in :
For this is the result of [1]; for a ring is two Ising chains and as for , so every lattice has the same for as for . For and we give no closed form for and use the numerical ground energy of the ring.
| size | states | dev. | GSD | ||
| 2 | 4 096 | d | |||
| 3 | 6 561 | d | |||
| 4 | 4 096 | d | |||
| 3 | 19 683 | d, L | |||
| 4 | 531 441 | L | |||
| 5 | 1 953 125 | L | |||
| 3 | 13 458 907 | g | |||
| 4 | 16 777 216 | g | |||
| 6 | 10 077 696 | g |
Table 2 (PDF p. 5). Ground energies of the literal gluing, Eq. (5), against exact diagonalization on 9 of the 17 lattices of E-2 (dev. = largest deviation over ). GSD (): counted from complete spectra (d: dense; b: block-resolved by bulk labels), or of Theorem 4 in italics where Lanczos does not resolve degeneracies (L: with low levels; g: ground energy only).
Proof. (a) Each site carries one , , , , so relation coefficients are constant on colour classes; the seam identifies with , which crosses colours unless is even; nothing is divided, so composite adds no torsion. In the stated order every site contributes the scalar 1 except, for , two row-0 sites contributing and ; the cancellation is global, not site by site. For the non-commuting pairs join opposite colours, and each site carries and , or and , from the colour-A plaquettes. (b), (c): the cycle algebra has a unique irreducible representation, of dimension , for each central character; count dimensions. For fixed , on all states has non-positive off-diagonal elements in the basis and is irreducible, so by Perron–Frobenius its ground state is unique, positive, and has . The ring translation exchanges and , so and for . For , a Gray-code map writes the ring as two critical Ising chains.
:::
Proposition 1 (row-exchanging reflection). The site permutation , for , implemented by the unitary , , is a symmetry of , and exchanges the seam row and row 0 ( up to inverses), hence the field and bond terms of the rings: for odd it acts on the ring as , a Kramers–Wannier duality followed by a translation; for even it exchanges the two rings. A uniform detuning of the seam row against row 0 breaks .
Proof. Lift to the double cover, where is ; bulk plaquettes map as .
For reweighted plaquette terms, invariance under and the half translation forces both ring masses to vanish for ; for a staggered reweighting invariant under , translation and gives the rings opposite non-zero masses.
Self-dual clock rings are known: three-state Potts criticality for , two Ising chains for , and a critical phase containing the self-dual point for [13, 14]. As a pre-registered anchor, the ring gives the ratio of the lowest charged-sector gap () to the first neutral gap, at ( extrapolated in ; Potts ratio of scaling dimensions ), and central charge from the registered fit, with the velocity inferred from assuming ; both are within the registered , and the exact velocity gives (reading). As for , the ring degeneracy belongs to the frustrated Hamiltonian; we make no protection claim.
Arithmetic and codes
Klein-bottle codes
For a stabilizer group , denotes the Weyl operators commuting with (modulo phases) and the logical module; the code dimension is .
Proposition 2 (logical module). For the -gluing with even , on the registered grid,
with trivial along rows and across the seam: in the diagonal branch and in the exchange branch. The edge register is the diagonal case , , and for every size.
Proof for the edge register; argument for the site register. In the edge register consists of -type cocycles and -type cycles and of coboundaries and boundaries, so . In the site register, string operators modulo stabilizers are 1-chains whose label coefficients are twisted by across the seam; for , , , Fox calculus gives , , hence Eq. (8). In the diagonal branch, conjugating the odd sites by gives a CSS code with the A and B plaquettes as checks and unit coefficients, i.e. of a cellulation of with unit-twisted incidence numbers; that this complex (or its exchange-branch analogue) is the twisted complex of Eq. (8) is not proved here.
For composite , is in general not the square of a cyclic group (Table 3): at , give and give ; at all four units give , split differently between the colours. (At , gives and .)
| dim | |||||||
| 2 | 1 | 2 | 2 | 4 | 4 | yes | |
| 3 | 1 | 3 | 3 | 3 | 2 | no | |
| 4 | 1 | 4 | 2 | 8 | 4 | yes | |
| 5 | 2 | 1 | 1 | 1 | 1 | no | |
| 6 | 1 | 6 | 2 | 12 | 4 | yes | |
| 8 | 1 | 8 | 2 | 16 | 8 | yes | |
| 8 | 3 | 2 | 4 | 8 | 2 | no | |
| 10 | 3 | 2 | 2 | 4 | 4 | yes | |
| 12 | 5 | 4 | 6 | 24 | 8 | yes | |
| 15 | 4 | 3 | 5 | 15 | — | — | |
| 16 | 1 | 16 | 2 | 32 | 16 | yes | |
| 16 | 7 | 2 | 8 | 16 | 4 | no |
Table 3. Site codes at , (diagonal branch; E-3): logical module , dimension , order of the group generated by the logical actions of all dressed isometries modulo logical Weyl operators ( is not on the isometry grid), and whether the Dehn twist acts (). Edge register, , : for .
Logical operators (grid). The row loop places on the even and on the odd sites of row ; its label is on colour and on colour . In the diagonal branch it commutes with , lies in iff its label has , , and mod ; and have orders , . The axis diagonals , , and , , close iff , resp. , and are then non-trivial modulo and the row loops. Each colour has a one-site-per-row vertical representative, with two turns where the straight diagonals all carry the other colour (, even; 74 codes). In the exchange branch with even, iff is even, and for even the colour sums , , are non-trivial and anticommute with . With the four edge-register loop types, these and generate on every grid code.
Distances (grid only; site register with , , and , in the diagonal branch, and even in the exchange branch). Diagonal branch (dimension ): ; exchange branch: ; edge register, even: , as for ; odd: and at the computed sizes. The last fails off the grid: an -type logical made of two dual cross-cap cores and a seam-row string has weight 13 at , , , and 14 at , (; readings).
Lattice isometries and gates
The isometries of the lattice Klein bottle are the maps , , that normalize the deck group generated by and : for , $c \in for , ; maps in all [1]. The twist iff is the half-period choice (the half translation is not isotopic to the identity and acts on homology as a Dehn twist [1]); the colour parity is the change of plaquette colour under the map. A dressed automorphism is the induced site permutation followed by single-site Cliffords mapping each plaquette to a unit power of the plaquette at the image position; such dressings are monomial. The multiplier () acts as , on even and with on odd sites. In the diagonal branch we use the basis of : the row loops above and the classes of one-site-per-row vertical loops of colour A and B whose labels generate the two components of (orders ); the edge register uses its and loops in the same roles. Modulo logical Weyl operators acts as . Let be the natural map, sending a vertical loop to the row loop with the same label (, ; ), and .
Proposition 3 (gates from lattice isometries). On the isometry grid, within the single-site dressing class: (a) A dressed automorphism exists iff, in the diagonal branch, or ; in the exchange branch, iff is even or is a square mod ; always in the edge register. The dressings of the identity are the multipliers (; in the exchange branch), and each admissible isometry has as many. (b) Modulo logical Weyl operators, colour-preserving dressed automorphisms in the diagonal branch and all those of the edge register act as ; on codes of dimension a non-trivial colour exchange occurs in the diagonal branch only for (on the grid: , and with ) and acts outside ; in the exchange branch ( even) every dressed automorphism, including the geometric colour exchanges allowed by (a), is a logical Weyl operator. (c) The generated group has order , (diagonal), and (edge), and iff , i.e. iff is even and , being the largest power of dividing .
Necessity in (a) is an argument valid beyond the grid: the induced (anti-)automorphism of the labels must satisfy ; sufficiency, (b) and (c) are exhaustive on the grid. At the isometries give [1]; for the ingredients separate. The Dehn twist acts as , a CNOT-type map on the order-two parts; for odd the half translation is a multiplier up to logical Weyl operators. A non-trivial colour exchange (, dimension ) forces and dimension for . The multipliers exist equally on the torus, where every isometry, colour exchanges included, has a dressing (controls at , , ); within this class the scarcity of Fourier-type gates comes from the non-orientable seam. Locality-preserving gates are constrained by anyon data and compatibility with mapping-class-group actions [15]. The Clifford restriction of Bravyi and Koenig applies to their two-dimensional topological qubit stabilizer-code setting [16]; it is not used here as a theorem for composite . Proposition 3 gives the dressed-isometry criteria within its stated class and grid.
Twisted homology of surfaces with cross-caps
Let (; ) be the connected sum of tori and cross-caps, , , coefficients , , with and the uniform pattern or the alternating pattern (k even). The Klein bottle of § 5.1 is alternating with (, ). Let .
Theorem 5 (twisted homology). Let , and .1 Parts (b)–(e) assume the consistency condition in (a).
(a) The local system is consistent iff : for the uniform pattern iff (exchange: even or ; diagonal: ); the alternating pattern always is.
(b) For the uniform pattern,
, ; for the alternating pattern the -th blocks carry and prefactors .
(c) For both patterns
that is, in the exchange branch and in the diagonal branch.
(d) for the alternating pattern, and for the uniform one whenever or is prime; in the exchange branch with even , . These conditions are sufficient, not necessary.
(e) . For , the auxiliary term on the right is given by Example 1.
Proof. For the one-vertex complex, and . The Fox derivatives of the commutators vanish under , and , giving (b), (e) and , hence (a). As is invertible, and , giving Eq. (10); for is computed without dividing by non-units. For (d), coordinates with () put the -component of in the form , with (alternating) or (uniform); when it vanishes the quotient splits, and for prime the order fixes the structure.
Example 1 (projective plane). [derivation; not on the preregistered grid] For , set and . The relation requires . In the same cellular convention as above the complex is
so direct calculation from this complex gives
In the diagonal branch consistency is , including for every . With , a positive integer for these consistent cases,
In particular, gives . In the exchange branch , so consistency requires ; then and . These are homology statements, not a construction of a lattice code or new sealed grid data.
Proof. The one-cell presentation has attaching word , whose Fox derivative is ; hence the two differentials above. Their product is , so consistency puts inside . The order follows by taking the quotient and using . For with , one has . In its two cyclic components the kernels of have orders and , and the images of have orders and , respectively. Each quotient is cyclic of order ; this also proves its integrality. For , . For at , swaps the two components and both and are the diagonal subgroup, so their quotient is trivial.
Scope. Lattice codes exist here only for the alternating local system on the Klein bottle, (§ 5.1). For every other surface and local system—handles, or , and the uniform pattern on the Klein bottle itself, whose holonomy along the rows would be —
| branch | pattern | dim | ||||
| 2 | 1 | exch. | uniform | 8 | ||
| 3 | 1 | exch. | altern. | 1 | ||
| 3 | 1 | exch. | uniform | inconsistent | — | |
| 3 | 1 | exch. | uniform | 9 | ||
| 3 | 1 | exch. | uniform | 81 | ||
| 3 | 1 | exch. | altern. | 9 | ||
| 4 | 1 | exch. | uniform | 32 | ||
| 5 | 2 | diag. | uniform | inconsistent | — | |
| 7 | 2 | diag. | uniform | 7 | ||
| 8 | 3 | diag. | uniform | 64 |
Table 4. Twisted homology of (E-4). , the dimension a code would have if its logical module were ; a lattice code exists only for the alternating row (the Klein bottle of § 5.1). as in Remark 1. *Reading: not direct-sum form.
Theorem 5 and the separate projective-plane calculation in Example 1 count homology only. A local cross-cap can be realized as a slit whose two lips are glued by a reflection; its ends merge into a vertex of cone angle , where the compatible gluings would have to be classified anew. If such a code exists and its logical module is the twisted homology, its dimension is , e.g. 9 (two qutrits) for , in the exchange branch (Table 4); we do not claim that every lattice realization would have this logical module.
Remark 1 (charge-conjugation defects: consistency only). For odd , fixes only the vacuum, so the counting of symmetry defects [17] gives one -defect type , with and quantum dimension (for defects of other anyon symmetries of models see Refs. [10, 18]); such defects ( even) have vacuum channels, and in the exchange branch with even , by Eq. (10) (432 grid cases). This is one count written twice: consistent with, not a proof of, cross-caps acting as charge-conjugation defects. For even , fixes the four labels , so the same counting gives four -defect types, related by fusion with abelian anyons and hence of equal dimension, with , i.e. (derivation, not computed).
Around a cross-cap: the double-winding monodromy
A -type path from column of row 0 runs through , (a on fails to commute only with the plaquettes and , of colour ), enters the seam plane at what appears its last site and continues, as in the proof of Theorem 2(ii), from its NW far corner at column ; a second pass returns it to (Fig. 3a): twice through the seam, once around a cross-cap. -type paths run along . The closure count is the number of strings on the path sites, with only (or only ) powers, that commute with all stabilizers; it equals for the multiplier picked up once around.
Figure 3. (a) -type double-winding path on , (diagonal branch), sites in order. The first pass (filled) runs from ; from its last site the string enters the seam plate through the NW far corner, i.e. row 0 at column (top labels), where the second pass (open) starts; it re-enters at column 0 . Here . (b) Closure counts of all 1800 registered paths (sealed E-4 results) by and branch, diagonal-branch paths split by whether : each group has one value, (diagonal) or (exchange). Grey line: .
Proposition 4 (double-winding monodromy). For all 540 codes with , all units, , , and all 1800 registered paths, the closure count is in the diagonal and in the exchange branch. The multiplier read generator by generator is on all 720 diagonal-branch paths and on all 1080 exchange-branch paths; for , occurs in the diagonal branch exactly for (64 paths: , ; , ).
The closure counts were pre-registered; is a reading, equal on all paths to : the landing columns sum to and have equal parity in the diagonal branch () and opposite parity in the exchange branch ().
These readings concern the external monodromy of the defined strings: on the registered lattices (derivation and reading), a charge carried once around a cross-cap is acted on by , the sign depending on the direction— in the exchange branch, the multiplier in the diagonal branch ( if ). They do not determine the defect type, the cross-cap states, or and data (Remark 1).
Cross-caps and magic: a conditional boundary
A stabilizer state of qudits is, up to phase, the unique common eigenvector of an abelian group of Weyl operators, i.e. of a Lagrangian subgroup () with phases; a logical stabilizer state is a code state fixed by a maximal abelian group of logical Weyl operators.
Corollary 1 (conditional no-go). In the operation model every operation maps stabilizer states to stabilizer states, and the outcome statistics of any protocol in applied to a stabilizer input can be computed efficiently classically. Hence, within , cross-cap geometry—whatever the seam action , the number of cross-caps and the
monodromy—produces no non-stabilizer state from stabilizer inputs, and no protocol in maps a logical stabilizer state of a code of this type to a non-stabilizer state.
Proof. This is the qudit Gottesman–Knill theorem [19, 20, 21, 22] restricted to ; we check the steps, including composite . Clifford unitaries, among them qudit permutations and single-site dressings, map Weyl operators to Weyl operators up to phases, hence Lagrangian groups to Lagrangian groups. Measuring with outcome on the state of gives, if has non-zero probability, a state stabilized by the elements of that commute with and by , i.e. by . is isotropic, and since for subgroups of and , the modular law gives is Lagrangian. Feed-forward corrections are Weyl or Clifford operations, and tracking with its phases is efficient [22]. A logical stabilizer state is a stabilizer state: the preimage in of a Lagrangian subgroup of is isotropic of order .
Remark 2 (Boundary of the operation model). Corollary 1 applies only to protocols in . Coherent path superpositions of mobile charge-conjugation defects and their interferometric measurements, as used in a different construction [9], are outside the present scope and are not implemented here. Gauging a separately specified charge-conjugation symmetry, when consistent symmetry and defect data permit it [17], changes the theory and is outside the present scope. Reflection-enriched data, fractionalization and cross-cap states [5] are outside the present scope and are not fixed by seam monodromy alone. In particular, the additional symmetry structure supporting the charge measurement considered by Cong, Cheng and Wang [8] is not constructed from our gluing. External magic states [23] are excluded by assumption: protocols supplied with them are outside the present claim. Moving a defect by the Clifford code deformations already allowed in does not by itself supply coherent path control. The label relation alone supplies none of these additional structures. These exclusions neither establish impossibility outside nor demonstrate any such resource in the lattice codes studied here.
The closure argument is the standard stabilizer argument; the contribution here is its geometric setting and boundary. Theorem 1 classifies commuting gluings within the local monomial far-corner ansatz, and Proposition 3 identifies the dressed isometries on the registered grid within its single-site dressing class. For every realization and protocol satisfying , the obstruction is uniform in the seam action, the number of cross-caps and the monodromy, because these parameters do not alter stabilizer closure under the permitted operations. This does not assert that arbitrary physical motions or measurements of cross-caps admit an implementation in . The argument needs no classification of locality-preserving gates [16] (we neither use nor extend that result to composite or non-orientable codes).
What the post-measurement states show
For odd we use the Wigner function of Gross [24], defined by and , so that . Let () and ; for prime and of order , let be the eigenstates of on an orbit .
Proposition 5 (Wigner negativity). (a) The minimum of is for , at , and for . (b) Measuring on gives with probability each; measuring on gives with probability and with probability , . (c) For prime , , and measuring on , , gives each eigenvalue with probability .
Proof. (a), (b) by direct computation: vanishes off , , . (c) A subset of ( prime) with at least two elements that is closed under midpoints is all of ; as , some have their midpoint . In that column , with , and , so the at most negative entries give .
A pure state of odd dimension has a non-negative Wigner function iff it is a stabilizer state [24], and negativity is necessary for magic-state distillation to help [25]; at is the strange state, [23]. Proposition 5 therefore certifies only that one non-stabilizer measurement—of , or of a charge that does not distinguish a label from its conjugate, such as —yields non-stabilizer states with fixed probabilities. That measurement lies outside . Nothing here shows distillability, fault tolerance, or a universal gate set.
Symmetry protection, by hand
Could the orientation reversal itself protect a measurement that does not distinguish a charge from its conjugate? Take a tunnelling term . The invariance of the Hamiltonian can be maintained in three ways. (i) A unitary with requires . (ii) An antiunitary also conjugates scalars: if , is invariant for every complex ; only if is -even, , does invariance give . This is the mechanism of Cong, Cheng and Wang: in their fractional quantum spin Hall realization “the time-reversal symmetry exchanges the two layers”, the charge is time-reversal invariant, and so “the tunneling amplitude of has to be real”; “the requirement that is real is beyond topological protection” [8]. (iii) A unitary spatial reflection that maps the coupling path to itself and to pins exactly as does: on the literal gluing (Proposition 1), so a -invariant term on a -fixed plaquette and , e.g. at , has real.
The Klein-bottle gluing, by contrast, is a gluing rule, not a symmetry of the Hamiltonian, and by itself constrains no local amplitude. The three pinning symmetries are microscopic symmetries that must be imposed; they are equally available on orientable lattices (a mirror through the coupling on a plane or torus) and fragile to disorder. If a Klein-bottle lattice carries such a reflection, its protection is symmetry protection, not topological, and does not come from non-orientability. Within the answer is negative; Corollary 1 and are unchanged. This is a hand assessment; the exclusions of apply, among them reflection-enriched data [5] and path superpositions of mobile defects [9].
Relation to neighbouring work
Homological codes. Lei shows that a qudit homological code on any 2-complex, for any qudit dimension , has dimension [3]; our edge-register codes are instances. Chan, Teo and Ryu relate the degeneracy on non-orientable surfaces to parity-invariant anyons [4]; Theorem 2 establishes this count, , for the seam actions of local lattice gluings, including the order-four ones. Bombin and Martin-Delgado state that “for nonorientable surfaces it is impossible to construct homological codes based on qudits of dimension ” [26], tracing this to “the torsion subgroup” of through their requirement that the stabilizer subspace have “a basis which is a linearly independent set” in , so that the code is an integer number of -dimensional qudits (their example: on the projective plane). For odd that torsion is absent (). Their no-go concerns their construction, a free linear subspace of . Codes whose stabilizer is a module, which allows torsion, exist on non-orientable surfaces for every : Lei’s theorem gives them in general, and explicitly the edge register on , the toric code of a cellulation, has commuting stabilizers and, for odd , dimension . What fails is therefore the broad reading that no non-orientable code with exists; we do not claim that the two constructions coincide. For even the obstruction concerns only the integer-number-of-qudits requirement, which the torsion violates; our edge-register codes (dimension ) do not impose it and are the twisted homology of Proposition 2 with . The site-register -codes are not cellular codes in the usual sense (their checks change type and carry unit multipliers across the seam), but on the grid their logical modules are twisted homology groups as well.
Cross-cap gates and reflection-enriched order. Cross-cap defects implement a fault-tolerant logical Hadamard in the surface code [6]; Proposition 3 is a comparison for lattice isometries. Barkeshli, Bonderson, Cheng, Jian and Walker develop reflection- and time-reversal-enriched phases on non-orientable manifolds, with cross-cap states and fractionalization data [5]. Their Sec. II acknowledges order-four and higher reflection actions before adopting an order-two convention for the subsequent analysis. Our exchange-branch is an explicit local-system monodromy of the stated lattice gluing; we do not claim order-four reflection itself as new or identify our cross-caps with their defect classes.
Quotient codes and twisted compactification. Hsin [7] formulates monomial stabilizer-code quotients through twisted compactification, with a transfer theorem for logical operators and distance, a Clifford restriction for descended monomial symmetries, and a separate discussion of composite dimensions. Our focus is the explicit local far-corner seam ansatz of the plaquette model: its commuting -family, phase-sensitive consistency, and the noncommuting literal seam with bulk-locked clock spectra. These are model-resolved questions, rather than a general quotient transfer theorem. The non-Clifford constructions of Ref. [7] use additional cup-product circuits or fermionic pumps outside ; they do not contradict the stabilizer-closure statement of Corollary 1.
Clifford limits and routes to magic. Webster and Bartlett prove non-universality for combinations of defect braiding and locality-preserving logical operators, including Abelian quantum-double models; their stronger Clifford restriction applies to a specified standard defect encoding [27]. Corollary 1 instead follows directly from the explicitly chosen stabilizer operation model. In the charge-4e superconductor of Shi, Han, Raghu and Vishwanath the fluxes are charge-conjugation defects of a chiral order: braiding gives the many-qutrit Clifford group, and a single-probe interferometric measurement prepares magic states [9]. Our exchange-branch cross-caps also have double-winding monodromy , but within a cross-cap moves only by code deformation, and no coherent superposition of paths arises.
Other neighbours. Lattice twists on a Klein bottle must represent its fundamental group [28]. Takahashi, Li and Yoshida construct an order-two locality-preserving Clifford realization of electric–magnetic duality for odd and exclude it for even on periodic toric-code lattices sufficiently large compared with the locality range; their obstruction locally emulates a cross-cap [29]; Proposition 3 asks instead which isometries of the Klein-bottle codes admit single-site dressings; its condition for a non-trivial colour exchange, , comes from compatibility with the seam action. Self-dual chains appear in parafermion chains [13], twist-defect chains [30] and edges of Wen’s model [31]; lattice defects implement Kramers–Wannier duality in the Ising model [32], and lattice reflections implement it in one dimension [33]; specific here is its realization by the Klein-bottle reflection exchanging the two plaquette rows of an obstructed gluing.
Discussion
The classification covers local, monomial, far-corner gluings only; non-local gluings, multi-site far corners, modified bulk plaquettes and the glide gluings of Ref. [1] at were not studied. Grid statements are not extrapolated (the odd- edge distance fails beyond the grid); the site-register logical module is verified on the grid and only argued in general; the gate results hold within the single-site dressing class; lattice codes exist only for the alternating Klein bottle, and elsewhere we only count homology; the ring degeneracy is not claimed to be protected. The no-go holds within , and the symmetry assessment is by hand. Producing a non-stabilizer resource from stabilizer inputs requires an operation outside ; the extensions listed in Remark 2 are outside the present scope, and their availability or sufficiency in these lattice codes is not established here. Thresholds, decoders and noise were not considered.
Acknowledgments
The research program was designed and directed by the author. Derivations, pre-registered predictions, code, figures and drafting were produced with AI assistance (Anthropic Claude). Independent AI instances, including OpenAI Codex, assisted with cross-checks; these were AI-assisted checks, not independent human verification or journal peer review. The reports and dispositions are archived with the accompanying material (Appendix A). The author takes full responsibility for the content. This work received no external funding.
Evidence, provenance and reproducibility
Registered grids
E-1: , , , , all units. E-2: , , , ; 920 cases. E-3: site register , all , , , , plus , (1586 codes); edge register , , , (495 codes); isometries on 1378 site ( and 16) and 264 edge codes (); distances on 135 site and 30 edge codes. E-4: Sec. 5.3, Propositions 4, 5.
Pre-registration and runs
E-0 is a calibration without pre-registration. For E-1 to E-4 the predictions, anchors, grids and falsification criteria were frozen in prereg/eK_prereg.md before the stage’s runs in this package (sha256 and time in logs/eK_prereg_stamp.txt, Table 5; records, not cryptographic proofs of priority) and were not changed afterwards. Each lists what was known beforehand, which is not counted as prediction: for E-1, E-0 readings on the Fourier-dressed code (code dimensions at , , , part of H3; the odd- sign, part of H4); for E-2, the literal-gluing anatomy and the results; for E-3, the E-1 results and the codes; for E-4, the seam action , planning-stage checks by another AI instance outside the package ( for identical exchange-branch cross-caps on the sphere, , even ; the Wigner minima , ) and a hand computation of the uniform , , case, which overlap H1(b) and H4(a). The seam action , and the phase statement of Theorem 2(iii) were not pre-registered (E-1 registered and the phase-free frustration rule); first appears in a pre-registration as the stated basis of E-3 H1.
| stage | prereg frozen | seal | sealed |
| E-0 | (calibration) | 0901726bc670b9df | 10-01 10:11:53 |
| E-1 | 09-30 15:20:37 | 55136031bc669cbb | 10-01 09:52:44 |
| E-2 | 10-01 14:49:29 | 599bdc643a1cdd7d | 10-02 07:41:10 |
| E-3 | 10-02 09:04:58 | cf526247f4931553 | 10-02 11:10:21 |
| E-4 | 10-02 17:45:44 | 97811bc292074cfe | 10-02 18:35:22 |
| checks: | E-1 21 108/10 027/2 707; E-2 15 956/13/1 410; | ||
| E-3 25 656/2 408/6 698; E-4 8 798/2 121/7 330 |
Table 5. Pre-registration freeze, seal (first 16 hex digits; full values in SEALS.txt) and sealing time of each stage (UTC; all dates in 2026); checks are prediction/anchor/cross-check items of the sealed run.
Each sealed run is one execution of code/run_eK.sh that started with no result files of its stage (the manifest lists the files present at the start). It was preceded by full development or rehearsal runs, whose results were inspected and in part reviewed, and code, gates and check counts changed between them. For example, E-0 ran twice against the v0.4 targets of Ref. [1] on 2026-09-30 and again on 2026-10-01; E-1 ran on 2026-09-30 at 16:37 UTC (57 018 checks; reviewed) and 19:21 (59 042); E-2 was sealed, reviewed and rerun with a changed gate and identical checks (superseded seal 6d0c3e1f…); E-3 had two rehearsals (the reviewed one with 36 329 checks; sealed run 34 762); after E-4’s reviewed rehearsal, 1800 cross-checks of were added.
Evidence by result
The proofs of draft v0.1 were re-derived by an independent AI reviewer; those added in v0.2 (Corollary 1, the edge-register part of Proposition 2) were checked by the external AI review of draft v0.2 but have not been re-derived independently by a human. Sealed checks per hypothesis (correlated within a case): E-1 H1–H4 4182, 261, 4536, 9252; E-2 H1–H5 3105, 7600, 196, 399, 6040; E-3 H1–H5 7531, 11 566, 390, 4926, 4602; E-4 H1–H4 5720, 864, 5400, 414. Theorem 1: complete enumeration of all far-corner assignments in 16 small cases (644 782 854 gluings: , ; , ; , 5, , , ) finds exactly the -family; a pruned exact search returns the predicted sets on all 1152 grid cases; for () odd- gluings exist exactly for . Theorem 2: 7272 stabilizer-group orders agree between three algorithms (Howell form [34], integer determinant, Smith form); state-vector ranks equal the algebraic dimension in 21 cases up to 16 777 216 states. Part (ii) rests on 17 208 sealed property checks of the seam map (colour exchange iff even; entries multiplying to ; square exactly in the exchange branch and for ; fixed labels); the equality of the 239 stored matrices with Eq. (3) was checked post hoc; the overall sign, a convention, is not tested. The phase statement of (iii) is a cross-check added after the internal review (846 cases). Theorems 3, 4, Proposition 1: 920 grid cases, including the dimensions of 186 712 joint eigenspaces (54 613 allowed); complete spectra on nine dense lattices (up to 6561 states) and block-resolved at , ; Lanczos ground energies and low levels on five lattices (up to 1 953 125 states), ground energies only on three (up to 16 777 216); all 34 ground energies agree with Eq. (5) to , which also reproduces the six energies and twenty degeneracies of Ref. [1]. The ring identity was checked on 68 ring sizes (). Two registered tests were not run (low levels of the three largest lattices; complete ring spectra beyond ), so (c) is confirmed with reduced coverage. Propositions 2, 3: invariant factors of all 1586 site and 495 edge codes by two routes (Smith form of the commutation Gram matrix; Howell counting), with the
square of the dimension of Theorem 2; gates exhaustive on 1378 site and 264 edge codes. For odd , (b) and (c) of Proposition 3 follow from (a), so its informative tests are at even ; at the groups of Ref. [1] are reproduced in its 30 calculators. Theorem 5: 2074 cases, 1570 consistent, invariant factors by two routes; (d) holds on its 1282 cases (550 with prime, implied by the order). The 288 consistent uniform cases outside its conditions (all diagonal, composite ) are ranges: 252 of direct-sum form, 36 not (: 10; : 26), e.g. at , , , not ; a post-hoc recomputation from Eq. (9) reproduces all 288. The alternating case reproduces the modules of all E-3 codes; H2 follows from H1(b). Propositions 4, 5: the closure counts separate the branches on 1120 paths; , added after the internal review because closure counts cannot tell from when , is a reading; torus control: 28 checks. Wigner: the least negative orbit state is , ( against the bound ); stabilizer states have . Post hoc (paper/verify_v02.py, exact arithmetic, unsealed): Lemma 1 (, , , all ; for ), the dressed glide and the exponents of Theorem 1, Theorem 4(a) in both orders, the CSS form used for Proposition 2, and the -fixed plaquettes of § 6.2.
Engine, seals and gates
A NumPy-only engine represents operators as Weyl strings with exact phases and uses the Howell form [34] for orders, relations and membership, and dense diagonalization and Lanczos for spectra. E-0 calibrated it at against the six result files of Ref. [1] (v0.7, unchanged in v0.8): 8005 comparisons (606 stored as truth values), 1934 definitional items, 453 internal cross-checks, 12 self-tests; errors specific to are addressed only by each stage’s self-tests and anchors.
Each sealed run writes the sha256 of every code file (gate and seal.py included), input and result to results/eK_files.sha256, whose own sha256 is the seal recorded in SEALS.txt after the gate passed (Table 5); both can be checked with sha256sum alone. The gate code/eK-gate.py re-evaluates every stored comparison, fixes the number and kinds of checks and, for E-2 to E-4, the coverage case by case; the E-3 and E-4 gates recompute 5196 and 13 787 expected values from the preregistered formulas. The gates guard against accidents, not against a writer who rewrites results and seal list consistently; mutation tests (unsealed record) behaved as expected. The gate summaries results/eK_gate.json are not sealed, and running a gate rewrites them. The only complete check is a rerun.
Deviations. Besides the two E-2 tests not run (Appendix § 10.§ 3), the E-2 ring range was fixed in code, not in the pre-registration, and 110 of the 7988 off-grid E-1 cross-checks duplicate a repeated random draw (no verdict affected). Grid-only statements, readings and post-hoc checks are labelled where they occur.
Reviews. Before sealing, an AI instance without the authoring context reviewed each of E-1 to E-4 adversarially, re-implementing the central computations and attacking the gates; its findings (gate defects, two rated P0 in E-1; coverage gaps; an off-grid distance failure; uninformative checks; the blindness of closure counts to ) were fixed before sealing or recorded. OpenAI Codex reviewed every stage. Its first E-0 review found two P0 problems—the gate trusted cached result files, and the odd- dressed gluing of Ref. [1] is sign-frustrated rather than a code with GSD = 2—which led to the seal mechanism, the reruns of E-0 and E-1, and the erratum v0.7 of Ref. [1]; on the current seals it found no P0. The external review of draft v0.2 identified a P1 error in the projective-plane domain of Theorem 5; the v0.3 disposition incorrectly summarized that review as having no P1. Draft v0.4 restores and separates Example 1, without changing the sealed results. The correction and the preserved v0.3 files are documented in ERRATA v0.4.md; the earlier wording fixes remain in place. For E-2 its point on case-by-case gate coverage led to a changed gate and a full rerun; its other P1 points were adopted. Reports and dispositions are under reviews/, E_campaign/review/ and, for this paper, paper/review_v0.1/ and paper/review_codex_v0.2/.
Data availability. The accompanying ancillary archive, paperE_reproducibility_v0.6.zip, contains the sealed E-0–E-4 code, inputs, results and supporting records in their campaign-relative layout, together with the number-tracing and post-hoc verification scripts. Historical reports retain their original statements; the scientific scope and corrections in this manuscript take precedence. Version DOI for the full reproducibility archive: https://doi.org/10.5281/zenodo.24185575.
Reproduction. bash code/run_eK.sh reruns a stage (sealed wall times 66, 61, 59, 10 min and 7 s for E-0 to E-4 on a two-core machine; Python 3.11.15, NumPy 2.4.4; about 4 GB for E-1, E-2) and gives a new seal; rerunning E-3 invalidates the E-4 seal. paper/claim_numbers.md traces every number, paper/check_numbers.py prints them from the sealed files after verifying the seals (marking unsealed inputs), paper/make_figs.py draws Figure 2–Figure 4, and paper/verify_v0.2.py holds the post-hoc checks. No second human has checked derivations or computations.
Figure 2. Structure of , uniform pattern, diagonal branch, composite (columns: units grouped by ; rows: ; sealed E-4 results). Blue: , direct sum confirmed. Green: the condition fails, the direct sum holds (readings). Orange, : not a direct sum; the extension of Eq. (9) (readings; post-hoc check). Grey: inconsistent, . The alternating pattern, the exchange branch and prime give direct sums throughout the grid. None of these local systems is realized by a lattice code here.
Figure 4. Wigner negativity of post-measurement states (sealed E-4 results; odd). (a) of for the 28 registered , against and ; the triangles below the orange line are the four cases with (two coincide at ). (b) Multiplier eigenstates for prime (116 states) against the bound : all above the line, by a factor .
References
- [1]J. Wang. An exactly solvable obstructed seam on a lattice Klein bottle: bulk-locked Majorana rings and register-dependent toric codes, 2026. URL https://arxiv.org/abs/2610.35931. Version 1.
- [2]X.-G. Wen. Quantum orders in an exact soluble model. Phys. Rev. Lett., 90:016803, 2003. doi: 10.1103/PhysRevLett.90.016803.DOI
- [3]Z. Lei. Qudit surface codes and hypermap codes. Quantum Inf. Process., 22:297, 2023. doi: 10.1007/s11128-023-04006-8.arxiv.org/abs/2112.01752
- [4]A. P. O. Chan, J. C. Y. Teo, and S. Ryu. Topological phases on non-orientable surfaces: twisting by parity symmetry. New J. Phys., 18:035005, 2016. doi: 10.1088/1367-2630/18/3/035005.DOI
- [5]M. Barkeshli, P. Bonderson, M. Cheng, C.-M. Jian, and K. Walker. Reflection and time reversal symmetry enriched topological phases of matter: path integrals, non-orientable manifolds, and anomalies. Commun. Math. Phys., 374:1021–1124, 2020. doi: 10.1007/s00220-019-03475-8.DOI
- [6]R. Kobayashi and G. Zhu. Cross-cap defects and fault-tolerant logical gates in the surface code and the honey-comb Floquet code. PRX Quantum, 5:020360, 2024. doi: 10.1103/PRXQuantum.5.020360.DOI
- [7]P.-S. Hsin. Twisted compactification, quotient quantum codes and logical gates, 2026. URL https://arxiv.org/abs/2611.09538. Version 1.
- [8]I. Cong, M. Cheng, and Z. Wang. Universal quantum computation with gapped boundaries. Phys. Rev. Lett., 119:170504, 2017. doi: 10.1103/PhysRevLett.119.170504.DOI
- [9]Z. D. Shi, Z. Han, S. Raghu, and A. Vishwanath. Charge-4e superconductors with parafermionic vortices: A path to universal topological quantum computation, 2026. URL https://arxiv.org/abs/2602.09663v2. Version 2.
- [10]Y.-Z. You and X.-G. Wen. Projective non-Abelian statistics of dislocation defects in a Z_N rotor model. Phys. Rev. B, 86:161107(R), 2012. doi: 10.1103/PhysRevB.86.161107.DOI
- [11]A. Y. Kitaev. Fault-tolerant quantum computation by anyons. Ann. Phys., 303:2, 2003. doi: 10.1016/S0003-4916(02)00018-0.DOI
- [12]D. Delmastro and J. Gomis. Symmetries of Abelian Chern–Simons theories and arithmetic. JHEP, 03:006, 2021. doi: 10.1007/JHEP03(2021)006.DOI
- [13]W. Li, S. Yang, H.-H. Tu, and M. Cheng. Criticality in translation-invariant parafermion chains. Phys. Rev. B, 91:115133, 2015. doi: 10.1103/PhysRevB.91.115133.DOI
- [14]G. Sun, T. Vekua, E. Cobanera, and G. Ortiz. Phase transitions in the Z_p and U(1) clock models. Phys. Rev. B, 100:094428, 2019. doi: 10.1103/PhysRevB.100.094428.DOI
- [15]M. E. Beverland, O. Buerschaper, R. Koenig, F. Pastawski, J. Preskill, and S. Sijher. Protected gates for topological quantum field theories. J. Math. Phys., 57:022201, 2016. doi: 10.1063/1.4939783.DOI
- [16]S. Bravyi and R. König. Classification of topologically protected gates for local stabilizer codes. Phys. Rev. Lett., 110:170503, 2013. doi: 10.1103/PhysRevLett.110.170503.DOI
- [17]M. Barkeshli, P. Bonderson, M. Cheng, and Z. Wang. Symmetry fractionalization, defects, and gauging of topological phases. Phys. Rev. B, 100:115147, 2019. doi: 10.1103/PhysRevB.100.115147.DOI
- [18]M. Barkeshli, C.-M. Jian, and X.-L. Qi. Twist defects and projective non-Abelian braiding statistics. Phys. Rev. B, 87:045130, 2013. doi: 10.1103/PhysRevB.87.045130.DOI
- [19]D. Gottesman. The Heisenberg representation of quantum computers. arXiv:quant-ph/9807006, 1998.arxiv.org/abs/quant-ph/9807006
- [20]D. Gottesman. Fault-tolerant quantum computation with higher-dimensional systems. Chaos Solitons Fractals, 10:1749, 1999. doi: 10.1016/S0960-0779(98)00218-5.DOI
- [21]E. Hostens, J. Dehaene, and B. De Moor. Stabilizer states and Clifford operations for systems of arbitrary dimensions and modular arithmetic. Phys. Rev. A, 71:042315, 2005. doi: 10.1103/PhysRevA.71.042315.DOI
- [22]N. de Beaudrap. A linearized stabilizer formalism for systems of finite dimension. Quantum Inf. Comput., 13:73, 2013. doi: 10.26421/QIC13.1-2-6.DOI
- [23]V. Veitch, S. A. H. Mousavian, D. Gottesman, and J. Emerson. The resource theory of stabilizer quantum computation. New J. Phys., 16:013009, 2014. doi: 10.1088/1367-2630/16/1/013009.DOI
- [24]D. Gross. Hudson’s theorem for finite-dimensional quantum systems. J. Math. Phys., 47:122107, 2006. doi: 10.1063/1.2393152.DOI
- [25]V. Veitch, C. Ferrie, D. Gross, and J. Emerson. Negative quasi-probability as a resource for quantum computation. New J. Phys., 14:113011, 2012. doi: 10.1088/1367-2630/14/11/113011.DOI
- [26]H. Bombin and M. A. Martin-Delgado. Homological error correction: classical and quantum codes. J. Math. Phys., 48:052105, 2007. doi: 10.1063/1.2731356.DOI
- [27]P. Webster and S. D. Bartlett. Fault-tolerant quantum gates with defects in topological stabilizer codes. Phys. Rev. A, 102:022403, 2020. doi: 10.1103/PhysRevA.102.022403.DOI
- [28]N. Seiberg and W. Zhang. Tori, Klein bottles, and modulo 8 parity/time-reversal anomalies of 2+1d staggered fermions. JHEP, 05:264, 2026. doi: 10.1007/JHEP05(2026)264.
- [29]S. Takahashi, Z. Li, and B. Yoshida. A no-go theorem for order-two Clifford electric-magnetic duality, 2026. URL https://arxiv.org/abs/2610.02097v2. Version 2.
- [30]X. Yu, X. Chen, A. Roy, and J. C. Y. Teo. Twofold twist defect chain at criticality. Phys. Rev. B, 96:205435, 2017. doi: 10.1103/PhysRevB.96.205435.DOI
- [31]W. W. Ho, L. Cincio, H. Moradi, D. Gaiotto, and G. Vidal. Edge-entanglement spectrum correspondence in a nonchiral topological phase and Kramers–Wannier duality. Phys. Rev. B, 91:125119, 2015.DOI
- [32]D. Aasen, R. S. K. Mong, and P. Fendley. Topological defects on the lattice: I. The Ising model. J. Phys. A, 49:354001, 2016. doi: 10.1088/1751-8113/49/35/354001.DOI
- [33]S. D. Pace, G. Delfino, H. T. Lam, and Ö. M. Aksoy. Gauging modulated symmetries: Kramers–Wannier dualities and non-invertible reflections. SciPost Phys., 18:021, 2025. doi: 10.21468/SciPostPhys.18.1.021.DOI
- [34]J. A. Howell. Spans in the module (Z_m)^s. Linear Multilinear Algebra, 19(1):67–77, 1986. doi: 10.1080/03081088808817705.



