Toric codes all the way down: decoupling translation-invariant topological CSS codes in higher dimensions

AI summary

Overview

Research area: Quantum error correction, specifically the classification and simplification of translation-invariant topological Calderbank–Shor–Steane (CSS) codes in higher spatial dimensions using homological algebra and module theory.

Technical level: Advanced. The paper assumes familiarity with CSS stabilizer codes, finite-depth local Clifford circuits, chain complexes, free resolutions, and Ext modules over Laurent polynomial rings.

Scope: The paper establishes when a translation-invariant topological p-ary CSS code in D ≥ 3 dimensions can be mapped, by a finite-depth local CSS-preserving Clifford circuit commuting with superlattice translations, to a stack of higher-form toric codes, and gives an explicit construction of that circuit.

What This Paper Is About

Topological codes in three or more dimensions can offer transversal non-Clifford gates and single-shot decoding, but it is often unclear whether these features are genuinely new or inherited from simpler toric codes. The central question is when a translation-invariant topological CSS code is “locally equivalent” — decoupled — to a stack of higher-form toric codes, allowing finite-depth local CSS-preserving Clifford circuits, ancillas, and coarse-graining to a superlattice. In two dimensions the known criterion is that translations must act trivially on every anyon type; in higher dimensions excitations are organized into superselection layers that can remain “glued” together, and this paper shows that gluing is the only remaining obstruction and can always be removed by further coarse-graining.

Key Contributions

1. A complete decoupling criterion (Theorem 1). A translation-invariant topological p-ary CSS code can be decoupled on a superlattice exactly when, after coarse-graining to that superlattice, translations act trivially on every superselection layer and the inter-layer gluing vanishes.

2. Proof that gluing is the only remaining obstruction and that coarse-graining removes it. For a code with finitely many excitation types in each layer and m nonzero layers, once translations act trivially on every layer, further coarse-graining by a factor of p^{m−1} in every direction eliminates the gluing. For codes with at most two nonzero layers — as in every three-dimensional code — the paper also determines the minimal coarse-graining.

3. Explicit algorithms and circuit construction. The first algorithm finds a sufficient superlattice; the second constructs the decoupling circuit on it. Both run in polynomial time for fixed spatial dimension once the superselection profile has been computed (Proposition 1), and the circuit is built from SUM (qudit CNOT) and SWAP gates.

4. A finite-torus criterion and a worked obstruction. The same criterion determines when a code on a finite torus encodes the maximal number of logical qudits: exactly when the code can be decoupled on its period lattice. The paper also analyzes a specific glued three-dimensional code from Ref. [15] and explicitly removes its gluing.

Main Findings

• Gluing is the sole additional obstruction. In higher dimensions, excitations are sorted by dimension into superselection layers, and even when translations act trivially on each layer, the layers can remain glued together. The paper proves that this gluing is the only remaining obstruction to decoupling beyond the two-dimensional anyon-preserving condition.

• Coarse-graining by p^{m−1} suffices. For a code with finitely many excitation types in each layer and m nonzero layers, once translations act trivially on every layer, coarse-graining by a factor of p^{m−1} in every direction removes the gluing. Each further coarse-graining by a factor of p in every direction detaches one more layer from those below it.

• Minimal coarse-graining for at most two layers. When at most two layers are nonzero — which the paper identifies as the situation for every three-dimensional code — the minimal coarse-graining is determined, so the general p^{m−1} bound can be more than necessary.

• Decoupling criterion. A code is decouplable on a superlattice exactly when, after coarse-graining to it, its superselection profile is that of a toric-code stack: translations act trivially on every layer and the gluing vanishes.

• Profile is invariant under the allowed transformations. The superselection profile is unchanged by translation-invariant CSS-preserving Clifford circuits and by adjoining ancillas (Lemma 7, Appendix A.4). Since the j-form toric code TC_j^(D)(F_p) has a single nonzero layer E_X^j ≅ F_p, a toric-code stack has profile ⊕_j F_p[−j]^{⊕ t_j}; a decoupling therefore requires the code’s profile to have the same form.

• Only degrees 1 through D−1 occur. The superselection layers are the modules E_X^j ≅ Ext_R^j(M_X-bar, R) for j ≥ 1, and only 1 ≤ j ≤ D−1 can occur (Lemma 3, Appendix A.2), so a topological code in D dimensions has excitations of dimension at most D−2.

• Fracton models are outside the scope. The requirement that every layer have finitely many excitation types excludes fracton models such as Haah’s cubic code and the X-cube model.

• Finite-torus result. A torus encodes the maximal number of logical qudits exactly when the code can be decoupled on its period lattice.

Methodology in Plain English

The authors turn the physics problem into algebra. Translation invariance lets each code be represented by matrices whose entries are Laurent polynomials in variables x_1, …, x_D, with coefficients in F_p, where p is prime and p = 2 corresponds to qubits. A code is described by a short complex of free modules, with the qudits in degree zero, the X-type checks as one map, and the Z-type checks as the other. The topological condition is expressed as ker H_X = im H_Z† and ker H_Z = im H_X†.

A decoupling circuit is then encoded as a commuting square of invertible polynomial matrices U and V: the circuit sends a Z-type operator Z(v) to Z(Uv) and an X-type operator X(u) to X((U†)^{−1}u). Finding a decoupling becomes finding such a square between the code’s complex and the complex of a toric-code stack.

To capture all excitations, including looplike and higher-dimensional ones, the authors use the superselection profile: a positive-degree truncation of a derived Hom complex built from a free resolution of the module of X-type Pauli operators modulo X-type stabilizers. Its cohomology modules are the superselection layers, which record fusion and translation action on excitation types. Toric codes are analyzed through the Koszul complex of the cubic lattice, whose differential sends a k-cell to its oriented boundary. Coarse-graining to a superlattice corresponds to restricting the Laurent polynomial ring to a subring, which is how the paper shows that gluing can be removed. The constructive part turns the algebraic factorization into a circuit of SUM and SWAP gates.

Why This Matters

This work gives a systematic classification of higher-dimensional translation-invariant topological CSS codes. It separates features that are genuinely tied to a nontrivial topological phase from features inherited from toric-code stacks, and it shows that many apparent obstructions disappear after coarse-graining. That matters for decoding, because a code that decouples to toric codes can be decoded using toric-code decoders, and for logical gates, because properties such as transversal non-Clifford gates must be interpreted relative to the decoupled stack. The paper’s motivating example is the three-dimensional color code: it has a transversal non-Clifford gate but is locally equivalent to three copies of the three-dimensional toric code [5].

Potential real-world applications include:

• Fault-tolerant quantum computing: identifying which higher-dimensional codes offer genuine advantages for logical gate synthesis rather than advantages already present in toric-code stacks.
• Decoder design: reducing decoding of a complicated CSS code to decoding of higher-form toric codes after an explicit local circuit.
• Code resource estimation: using the finite-torus criterion to determine when a code encodes the maximal number of logical qudits on a given period lattice.
• Code construction and benchmarking: using the decoupling algorithms to compare candidate translation-invariant codes, including bivariate bicycle codes, against toric-code stacks.

Industry relevance: The results are relevant to groups designing fault-tolerant quantum architectures with higher-dimensional or translation-invariant codes, and to teams that need practical decoding and logical-gate protocols derived from a code’s topological phase rather than from its microscopic presentation.

Future Directions

• Efficient computation of the superselection profile in higher dimensions. The decoupling algorithms are polynomial time for fixed spatial dimension once the profile is known, so improving profile computation is a direct next step.

• Minimal coarse-graining beyond two layers. The paper determines the minimal coarse-graining when at most two layers are nonzero; extending this to codes with three or more layers is left open.

• Extending beyond translation-invariant p-ary CSS codes. The framework assumes translation invariance and CSS structure. Generalizing to non-translation-invariant codes, non-CSS codes, or non-abelian anyon models is a natural direction.

• Applying decoupling to decoding and logical-gate protocols. Section VI discusses applications to decoding and logical gates, and the explicit SUM/SWAP circuit construction provides a starting point for turning the classification into concrete fault-tolerant protocols.

Target Audience

This paper is aimed at quantum information theorists, mathematical physicists, and quantum error-correction researchers with a background in stabilizer codes and homological algebra. Graduate students working on topological codes, higher-form symmetries, or translation-invariant quantum codes will benefit, as will researchers designing decoding and logical-gate schemes for higher-dimensional fault-tolerant quantum computers. Readers seeking an accessible experimental introduction to quantum error correction will find the paper’s algebraic machinery demanding.

Authors’ abstract

Topological Calderbank-Shor-Steane (CSS) codes in more than two spatial dimensions offer features useful for fault-tolerant quantum computation, such as transversal non-Clifford gates and single-shot decoding, but a systematic understanding of them is still lacking. One approach is to decouple a code: map it to a stack of higher-form toric codes by a finite-depth local CSS-preserving Clifford circuit, possibly after adding ancillas. For translation-invariant $p$-ary CSS codes, where $p$ is prime ($p=2$ for qubits), we require the circuit to commute with the translations of a superlattice. In two dimensions, such a decoupling exists once these translations act trivially on every anyon type. In higher dimensions, excitations are sorted by their dimension into superselection layers, which can remain glued together even when translations act trivially on each layer. We show that this gluing is the only remaining obstruction: a code can be decoupled on a superlattice exactly when, after coarse-graining to it, translations act trivially on every layer and the gluing vanishes. Moreover, for a code with finitely many excitation types in each layer and $m$ nonzero layers, once translations act trivially on every layer, further coarse-graining by a factor of $p^{m-1}$ in every direction removes the aforementioned gluing. Physically, the same criterion determines when the code encodes the maximal number of logical qudits on a finite torus. Finally, we give an explicit algorithm that constructs the decoupling as a circuit of SUM (qudit CNOT) and SWAP gates.