Lifting Lifted Product Codes

AI summary

Overview

Research area: Quantum error correction — specifically quantum low-density parity-check (qLDPC) codes and the lifted product (LP) code family, with side discussions of fault-tolerant logical operations and decoherence-induced phase transitions.

Technical level: Advanced. The paper is written in the language of group algebras, group extensions, chain complexes, balanced products, and chain/cochain maps; the physics applications (code surgery, coherent information) sit on top of that algebraic machinery.

Scope (one sentence): The paper introduces a general algebraic procedure for "lifting" any lifted product code over an arbitrary group into a larger code with the same local Tanner-graph structure, and then uses that procedure for code search, for transferring logical-operation gadgets, and for probing thermodynamic (scaling) families of qLDPC codes.

What This Paper Is About

Lifted product codes are built as lifts of hypergraph product codes, but there has been no general recipe for lifting a general LP code further while keeping the code an LP code. The authors supply one: a group extension turns a base LP code over one group into a larger LP code over an extended group, whose Tanner graph is a finite lift of the base Tanner graph. They then show what this buys you — better code parameters, cheaper code-surgery gadgets, and a candidate way to define a thermodynamic family for algebraically defined qLDPC codes that have no underlying Euclidean lattice.

Key Contributions

1. A general lifting construction for LP codes over arbitrary groups. Starting from an LP code over the group algebra 𝔽₂[G] built from two classical complexes, the authors lift those complexes along a short exact sequence 1 → K → H → G → 1, then take the balanced product over 𝔽₂[H]. They prove (Proposition 1) that the resulting Tanner graph is a |K|-lift of the original Tanner graph.

2. A matching projection and transfer machinery. Applying the quotient homomorphism π entrywise to the boundary maps produces a "projected code" over 𝔽₂[G], and this projection induces chain and cochain maps between the lifted and projected codes (Proposition 2). These maps are the mechanism for relating code parameters, logical operators, and gadgets across a lifted family.

3. Applications to code search. The framework recovers known constructions as special cases (hypergraph product codes as the trivial-group case; sequences of BB codes from [76] as a special case of a ℤ_u × ℤ_t extension of ℤ_l × ℤ_m), and, applied more broadly, yields LP codes with better parameters than previously reported, including improved lifted BB codes and enhanced clustered cyclic (CC) codes.

4. Gadget transfer and thermodynamic families. Code-surgery gadgets are shown to transfer across finite lifts through chain maps, sometimes with lower space overhead; parallel product surgery is developed for lifted clustered cyclic codes; and coherent information under bit-flip noise is computed for lifted families, exhibiting finite-size crossings.

Main Findings

• Lifting preserves locality while growing the code. The base and lifted codes share the same local incidence structure at every vertex, even though their global structure differs. The illustrative example: a cyclic code over ℤ₆ with boundary maps H_A = (x + x²) and H_B = (x + x⁻¹) has parameters ⟦12, 2, 3⟧, and its ℤ₁₈ version has parameters ⟦36, 2, 3⟧.

• Better parameters than previously reported. The authors state that their lifting protocol gives a systematic route to code search and produces LP codes with improved parameters relative to earlier results, including lifted BB codes that fall outside the BB family, and lifted CC codes with enhanced parameters. (The specific parameter values for these improved instances are given in the paper's tables, which are not visible in the excerpt analyzed here.)

• Parameter bounds from the base code. In several settings the parameters of the lift are lower bounded by those of the base code, so lifting a good base code is a useful ansatz that shrinks the search space.

• Surgery gadgets transfer through chain maps, sometimes more cheaply. Code-surgery gadgets defined on the projected code can be carried to the lift via the induced chain maps, and the lifted structure lets one build a surgery ancilla code with smaller space overhead in several cases.

• Logical addressability survives lifting for CC codes. The authors identify conditions under which lifted CC codes retain the same logical addressability under parallel product surgery as their base CC codes, and show the lifted CC codes reported in the paper preserve the relevant logical-operation capabilities.

• Finite-size crossings in coherent information. Under bit-flip noise, coherent information computed for several base codes and selected lifts shows finite-size crossings, consistent with a common asymptotic behavior across the lifted family. A worked non-Abelian example uses G = S₃ × S₃ with a base code of parameters ⟦72, 4, 6⟧ and lifts with K = ℤ₂, ℤ₃, ℤ₄ (Table 6, studied in Sec. V).

• Local structure alone is not enough. The authors explicitly flag caveats: the crossings are suggestive, but additional conditions beyond preservation of local structure are needed to single out a unique thermodynamic family.

• Position relative to prior lifting work. Earlier approaches either lift arbitrary CSS codes abstractly via finite coverings of a two-dimensional Tanner cone-complex (classified by finite-index subgroups of the fundamental group, which is computationally demanding and does not necessarily yield an LP code), or are restricted to specific subfamilies (BB codes, and more recently two-block group algebra codes, and quantum Tanner codes via coverings of the underlying square complex). This work applies to any LP code over an arbitrary group and stays inside the LP class.

Methodology in Plain English

The authors work entirely algebraically, at the level of the matrices ("boundary maps") that define a code rather than at the level of pictures of Tanner graphs.

1. Pick a base LP code defined over a group algebra 𝔽₂[G], that is, a code whose boundary-map entries are sums of group elements.

2. Choose a group extension 1 → K → H → G → 1. The kernel K is what sets the lift index: the new Tanner graph covers the old one exactly |K| times.

3. Lift the boundary maps. Each group element g appearing in a boundary-map entry is replaced by a chosen preimage h in H, written h = χ(γ)·σ(g) using a section σ and a kernel element γ. The ranks of the modules stay the same; only the group acting on them changes.

4. Take the balanced product of the two lifted classical complexes over 𝔽₂[H] to get the new LP code, and verify the covering property of the Tanner graphs.

5. Go backwards with the quotient map. Applying π to every group-algebra coefficient collapses the lifted code onto a smaller code and induces chain and cochain maps, which are then used as the transfer channel for logical operators and gadgets.

6. Test on instances. The framework is applied to BB codes, to clustered cyclic codes, and to a non-Abelian S₃ × S₃ example; parameters are computed, surgery constructions are built and compared for space overhead, and coherent information under bit-flip noise is evaluated numerically to look for finite-size crossings.

Why This Matters

Impact on research. Lifted product codes include quasi-cyclic codes, bivariate bicycle codes, and other finite-size instances with good performance, and the LP family is one route to asymptotically good qLDPC codes. Having a general, purely algebraic lifting operation — rather than one restricted to a subfamily — turns "guess a good code" into "systematically grow a good code," and simultaneously gives a principled way to talk about scaling families for codes that have no Euclidean lattice behind them.

Potential applications:

• Designing qLDPC codes with better parameters for quantum memory at fixed qubit counts.
• Reducing the space overhead of fault-tolerant logical operations (code surgery) by exploiting the lifted structure for ancilla preparation.
• Building families of codes of increasing size from a single validated base code, useful for benchmarking decoders and for studying error-correction thresholds and decoherence transitions.
• Providing candidate code families for architectures where long-range qubit connectivity is available, since the lift changes global structure while keeping check weights local.

Industry relevance. The results speak directly to groups building qLDPC-based fault-tolerant architectures: cheaper surgery ancillas translate into fewer physical qubits per logical operation, and a systematic search procedure over lifted families is more practical than exhaustive search over arbitrary codes.

Future Directions

• Pin down what makes a thermodynamic family unique. The observed finite-size crossings are necessary-looking but not sufficient; the authors state that additional conditions beyond local structure are needed to determine a unique family, leaving the correct criterion open.
• Extend the search over inequivalent lifts. For a fixed group extension, inequivalent choices of the lifted elements h_ij(g) can produce distinct LP codes; systematically exploring these choices (and non-equivalent π for given (K, H, G)) is a natural next step for code discovery.
• Broaden the gadget-transfer program. Demonstrating transfer of further fault-tolerant logical-operation protocols beyond code surgery and parallel product surgery, and characterizing exactly when the reduced-overhead ancilla constructions apply.
• Connect to decoding and thresholds. The paper analyzes coherent information under bit-flip noise; extending the family analysis to more realistic noise models and to concrete decoders would test whether the lifted families behave as hoped at scale.

Target Audience

Researchers in quantum information theory and quantum error correction who are comfortable with group theory and chain complexes — particularly those working on qLDPC code construction, code surgery and fault-tolerant logical operations, and statistical-mechanics-style studies of decoding and decoherence transitions. Readers wanting only the practical code parameters can read the introduction, the examples in Sec. III, and the tables; the algebraic core in Sec. II is aimed at specialists.

Authors’ abstract

Lifted product (LP) codes form an important class of quantum error correcting codes with favorable code parameters. We introduce a systematic construction of LP code families based on group extensions and graph lifts. The construction increases the code size while preserving the local structure of the Tanner graph, and relates code parameters, logical operators, and fault-tolerant logical-operation gadgets within the families through chain and cochain maps. As a first application, we obtain LP codes with better code parameters than previously reported ones. We then demonstrate that code-surgery gadgets can be transferred across the selected finite lifts through chain maps and, in several cases, implemented with lower space overhead. We also develop parallel product surgery for lifted clustered cyclic codes. Finally, we propose lifting as a systematic first step toward defining thermodynamic families for algebraically defined qLDPC codes without an underlying Euclidean lattice. For several base codes and selected lifts, coherent information exhibits finite-size crossings, while our results also indicate that additional conditions are needed to determine a unique family.