{"id":"e5e321ec-12e6-4bb4-b3bc-2d4a68043818","arxiv_id":"2608.07431","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"GALA codes are a new construction family of rate-1/2 quantum LDPC codes with AOD-compatible moves and certified distances up to 16, including compact instances like [[132,30,12]].","lead":"Researchers introduce GALA, a family of high-rate quantum error-correcting codes built from group products, and report compact instances with certified distances that map onto reconfigurable atom-array hardware. The codes promise lower qubit overhead and millisecond-scale error-correction cycles, potentially bringing fault-tolerant quantum computing closer to current devices.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Exact distance certifications for the two barrier-breaking codes rest on an unshipped custom exhaustive enumerator; independent verification is required before the headline d>w claim is accepted.","rationale":"The reader's weakest assumption is exactly the one I find most load-bearing: the claimed exact distances, and hence the barrier-breaking status of [[1752,880,14]] and [[2232,1120,16]], rest on an unshipped custom exhaustive enumeration. The full text makes the central claim depend on this certification: Section S6.1 says every reported instance is 'certified exactly' by complete exclusion of lower-weight logicals plus a verified witness, and the abstract's 'exactly certified distances greater than check weights' is repeated without qualification. The manuscript gives explicit group-ring generators in Table S3, so the codes themselves are reproducible in principle; what is not reproducible is the exact-distance proof. My proposed check, an independent re-derivation of the certification from the published generators, would settle whether the concern lands. I do not see a stronger or more central concern: the LER extrapolation is explicitly labeled 'extrapolated' in the abstract and is presented with caveats in Section S6.2, so it is secondary to the code-parameter claims; the constructive bounds in Section S3 are proofs, not computational claims. The appropriate verdict remains CONDITIONAL: the construction and reported parameters are plausible and well motivated, but the headline exact-distance barrier-breaking claim should be accepted only once the certification is independently verified or the code and algorithm are released.","tokens_in":46051,"tokens_out":3211,"duration_ms":35266,"concrete_test":"Independently reproduce the exact distance certification for [[2232,1120,16]]: from the Table S3 generators (L=12, S_3 × C_62, with the listed F and G lifts), construct H_X and H_Z; verify ranks give k=1120; then run an independent exhaustive verifier, or the authors' code if released, to confirm that no X-type or Z-type logical operator of weight ≤15 exists and that an explicit weight-16 witness exists in each sector. Repeat for [[1752,880,14]] at threshold d=14. If either enumeration cannot be reproduced, is found to be incomplete, or yields a lower distance, the certified-distance and barrier-breaking claims should be downgraded accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that [[1752,880,14]] and [[2232,1120,16]] are barrier-breaking is exactly certified distance greater than check weight w=12, with no '≤' on any reported GALA parameter. This certification is load-bearing and is not independently checkable from the manuscript. Section S6.1 describes a two-part procedure: exclusion of every logical operator of weight below d, followed by an explicit weight-d witness, using an enumeration described only as 'our own implementation.' No code is shipped, no pruning strategy is specified, and no proof of enumeration completeness is given. For n=2232 and d=16, a naive exhaustive scan over all supports of weight at most 15 is astronomically large (C(2232,15) ≈ 10^35), so the feasibility and correctness of the claimed certification depend on undocumented structure or pruning. The manuscript's own note that a 2160-qubit instance has only its X-sector enumeration complete, while the Z sector 'is not yet complete,' underscores that the enumeration is a substantial custom computation. If that enumeration is incomplete, buggy, or implicitly relies on an unverified property of the lifted generators, the exact distances d=14 and d=16 fail, and with them the 'barrier-breaking' status of the two headline rate-1/2 codes. This is a reproducibility gap, not an accusation of error: the test matrices are present in Table S3, but the certified-distance claim cannot be settled from the paper alone.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces GALA codes, a family of CSS qLDPC codes constructed by lifting a block-circulant proto-matrix over a product or semidirect product of a small non-abelian group (typically S_3 or S_4) and a cyclic group. The factorization gives active orthogonality from the small non-abelian factor and symmetries/AOD move schedules from the abelian factor. The authors prove closed-form bounds on rate, distance, and girth, give explicit logical operators and automorphisms, construct ZX-dual variants with fold-transversal Clifford gates, and report a search producing compact rate-1/2 codes including [[132,30,12]], [[672,336,12]], and the headline barrier-breaking instances [[1752,880,14]] and [[2232,1120,16]], all claimed to have exactly certified distances at check weight 12. The paper also gives syndrome-extraction cycle-time estimates and circuit-level logical error rate simulations.","tokens_in":46361,"tokens_out":12631,"duration_ms":139727,"significance":"If the claims are correct, the paper makes a strong co-design contribution: it provides a transparent algebraic framework that contains previously known AOD-compatible Kasai codes, supplies analytic bounds, and yields concrete rate-1/2 codes with d>w that are short enough for near-term atom-array hardware. The closed-form bounds in Section S3 and the explicit move schedules in Section S5 are genuine strengths, as is the use of an end-to-end case study for [[132,30,12]]. However, the central distance claims rest on an exhaustive enumeration whose code and algorithmic details are not provided, and the 'all smaller than previously known' abstract claim needs qualification. The result is significant if reproducible, but the certification step is currently a black box.","major_comments":[{"comment":"The exact-distance certifications for [[1752,880,14]] and [[2232,1120,16]] rest entirely on an exhaustive enumeration described only as 'our own implementation' in Section S6.1. No code, pseudocode, pruning strategy, complexity analysis, or completeness proof is given. For n=2232 and d=16, a naive search over all supports of weight at most 15 is astronomically large (C(2232,15) is on the order of 10^35), so the feasibility and correctness of the claimed certification depend on undocumented structure or pruning. The manuscript itself states that a 2160-qubit instance has only its X-sector enumeration complete, which underscores that this is a substantial custom computation. Because the 'barrier-breaking' status of the two headline codes is exactly the claim that d>w=12, the central result cannot be verified from the paper alone. I recommend that the authors release the enumerator or provide a detailed algorithm with complexity analysis and validation tests, together with certification logs for each reported instance.","section":"S6.1, Table S3"}],"minor_comments":[{"comment":"The abstract says the two barrier-breaking codes are 'all smaller than previously known hardware compatible rate-1/2 codes,' but the main text acknowledges that the pair-partition code [[1524,766,14]] of Ref. [10] is shorter at d=14. Please clarify the exact comparison class (e.g., AOD-compatible codes only) and align the abstract with that qualification.","section":"Abstract and main text"},{"comment":"The abstract describes [[132,30,12]] as 'almost girth-6,' while the main text and Table S5 report girth 4 (t4=660). Please use consistent terminology and explicitly state that the code is girth 4 despite being called almost girth-6.","section":"Main text, Table S5"},{"comment":"The proof says 'All rows sum up to 1^{1×n},' but the sum of the active rows is the all-ones vector only when the column weight J is odd; for even J (e.g., J=2) the sum is zero. Please state the parity condition and verify the claimed J−1 rank-deficiency count for even J, or restrict the proposition to the odd case.","section":"S3.1, Proposition 7"},{"comment":"The extrapolated logical error rates at p=10^-3 depend on two fit forms with parameters c0,c1,c2 and d_eff. The text already calls this an extrapolation, but the figure caption should note explicitly that the two fit forms can disagree and that the quoted 10^-10 LER for [[672,336,12]] is from the chosen form.","section":"S6.2, Figure S3"},{"comment":"Several certified frontier codes listed in Table S3, such as [[1764,886,14]] and [[2328,1168,16]], are not discussed in the main text. Please add a sentence explaining their status relative to the headline claims and why they are not included in the abstract.","section":"Table S3"}],"recommendation":"major_revision","confidential_remarks":"The construction and analytic bounds are solid enough that I would be happy to see the paper published once the exact-distance certification is made reproducible. The main obstacle is not the mathematical framework but the unshipped enumerator behind the headline d=14 and d=16 claims. I would not reject on this basis alone, because the generator data in Table S3 allows independent verification in principle, but for a serious journal the certification procedure needs to be a documented artifact, not a private implementation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"GALA is a genuinely useful reframing. The product-group factorization — small non-abelian top for active orthogonality, large abelian bottom for AOD moves and automorphisms — turns hardware compatibility into a design input rather than a post-hoc filter, and the paper shows it subsumes the Kasai codes while exposing the parameter tradeoffs in closed form. The bounds in S3 are real derivations, not fits, and the explicit generators in Tables S3–S5 are enough to reconstruct the reported codes. The [[132,30,12]] case study, with its top/bottom logical chains and fold-transversal gates, is concrete and careful.\n\nThe soft spot is exactly where the reader put it: the claim that [[1752,880,14]] and [[2232,1120,16]] have exactly certified distances above check weight rests on an exhaustive enumerator that is not shipped and not described in enough detail to assess. For n=2232 and d=16, a naive enumeration is astronomically large, so the feasibility depends on pruning structure that the paper does not explain. The paper's own confession that a 2160-qubit instance has only its X-sector complete underscores that this is a substantial custom computation. I don't doubt the authors ran a real search, and the distances may well be correct; but the headline 'barrier-breaking' status is conditioned on a black box. A referee cannot verify it from the manuscript. The test matrices are there, so a determined reader could re-run the enumeration independently, but that's a heavy ask that the authors should remove by shipping code or writing the pruning argument.\n\nOne smaller point: the abstract's 'below 10^-10 LER' is labeled extrapolated, so the reader's charge of overstatement is too strong. The extrapolation itself is a rough estimate, and the paper says so. The practical-cycle numbers are estimates from a movement model, which is normal for this literature.\n\nNet: this is a significant construction paper with a real reproducibility gap in the part that matters most. The algebra and the small-code results hold up; the exact distances are probably right but unproven as presented. I'd send it to a serious referee, and I'd make code/data release or a complete enumeration description a condition of acceptance.","headline":"A strong co-design construction with proven bounds; the exact-distance certification needs to be reproducible before the barrier-breaking claims are accepted.","tokens_in":46885,"tokens_out":2560,"would_cite":true,"duration_ms":25867,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.Pp"],"model":"deepseek-v4-flash","headline":"This paper claims that GALA codes — built by lifting over a product of a small non-abelian group and a large cyclic group — produce certified rate-1/2 codes with distances 12, 14, and 16 and millisecond QEC cycles on atom arrays.","keywords":["quantum LDPC codes","reconfigurable neutral-atom arrays","active orthogonality","group-action lifts","acousto-optic deflectors","code automorphisms","ZX-duality","fold-transversal Clifford gates"],"falsifier":"Run an independent exclusion search on the published group-ring generators of $[[1752,880,14]]$ and $[[2232,1120,16]]$: enumerate every vector in $\\langle H_Z\\rangle^\\perp\\setminus\\langle H_X\\rangle$ and in $\\langle H_X\\rangle^\\perp\\setminus\\langle H_Z\\rangle$ of weight below the claimed distance (even weights through 12 and 14, since all logicals are even). Finding any such vector refutes the certified distance, and a verifier that reproduces both exclusions plus the weight-$d$ witnesses would confirm it.","tokens_in":45868,"feed_emoji":"⚛️","tokens_out":20567,"duration_ms":153470,"temperature":0.7,"pith_summary":"This paper claims that the conflict between high encoding rate, large distance, and physical implementability in quantum LDPC codes can be resolved by factoring the code construction itself. The GALA construction lifts a block-circulant pair of check matrices over a product group $G=H_k\\times C_m$ whose two factors have separate jobs: a small non-abelian factor (such as $S_3$) enforces active orthogonality, which keeps the stabilizer rows light while turning the discarded latent rows into logical operators whose distance is no longer bounded by the check weight, and a large cyclic factor $C_m$ supplies the symmetries that become explicit parallel row-and-column atom moves and code automorphisms. The concrete payoff is a self-dual $[[132,30,12]]$ code on 132 atoms with a 3.1 ms syndrome-extraction cycle, a girth-6 rate-1/2 $[[672,336,12]]$ code, and rate-1/2 $[[1752,880,14]]$ and $[[2232,1120,16]]$ codes whose certified distances exceed their check weight, all shorter than previously known hardware-compatible rate-1/2 codes. If the claims hold, near-term atom-array processors become a credible path to low-overhead fault-tolerant quantum memory.","feed_headline":"Rate-1/2 quantum codes certified to distance 16","feed_subtitle":"GALA codes bring rate-1/2 error correction to millisecond cycles on today's reconfigurable atom arrays","key_machinery":"The load-bearing object is the product-group lift with active orthogonality. Generators $F=\\sum_i z_i\\otimes F_i\\otimes f_i$ and $G=\\sum_i z_i\\otimes G_i\\otimes g_i$ live in the group algebra $\\mathbb{F}_2[\\mathbb{Z}_{L/2}\\times H_k\\times C_m]$; their commutators satisfy $[F_i,G_j]=0$ iff the small non-abelian factors commute, so the choice of the active set $\\Gamma$ — the $J\\times J$ band of block rows that will remain stabilizers — is a finite pattern inside $H_k$ (e.g. $S_3$), enumerable independently of $m$. The abelian factor then does the hardware work: the diagonal copy $I_{kL}\\times C_m$ is a group of code automorphisms that organizes the logical space into hypercubes, and the syndrome-extraction schedule reads off as rigid row and column translations, one pair of AOD sweeps per round. Closed-form bounds tie design parameters to outcomes: rate $\\ge 1-2J/L$; for monomial lifts the barrier relation $d\\le w=L$ holds whenever $J>L/4$ or $L<12$, which is why the compact instances use polynomial lifts (sums of group elements per entry) to decouple the weight from $L$; and quotient inflation gives $d\\le m\\,d_{Q_H}$ and $d\\le k\\,d_{Q_C}$. ZX-dual variants impose a fold symmetry $F_i=G_{r(i)}^T t$ on the generators, yielding fold-transversal Clifford gates $H$ and $S$ at the cost of one parallel rearrangement plus one gate layer.","core_discovery":"The paper's central claim is that the augmentation barrier — the rule that deleting stabilizer rows to raise the encoding rate caps the distance at the stabilizer weight — can be broken by construction rather than by post-hoc search. The mechanism is active orthogonality: in a two-block CSS code lifted over the group ring $\\mathbb{F}_2[H_k\\times C_m]$, the condition $H_XH_Z^T=0$ is enforced only on the $J$ active block rows that are kept as stabilizers, while the remaining latent rows are deliberately made non-orthogonal and removed; those removed rows become logical degrees of freedom whose weight is not limited by the bounded check weight. Because commutativity of the lifted generators $F_i,G_j$ is decided entirely inside the small factor $H_k$, the orthogonality pattern is a finite, enumerable combinatorial condition independent of the large factor $C_m$, which is then free to supply automorphisms and AOD schedules. The paper reports that exhaustive enumeration certifies every reported distance exactly — no logical operator below weight 12, 14, or 16 exists for the flagship instances — and that circuit-level simulation extrapolates logical error rates below $10^{-10}$ per cycle at $10^{-3}$ physical error for the rate-1/2 $[[672,336,12]]$.","pith_inferences":["The paper's reported trade — about one-and-a-half decades worse logical error rate for an eightfold reduction in qubit overhead — suggests a scaling law worth probing: the abelian factor controls cycle time and automorphisms while the non-abelian top sets the distance ceiling, so scanning tops beyond $S_3$ and $S_4$ at fixed $L,J$ is the natural next search dimension.","Because the certified distances rest on an unshipped exhaustive enumerator, the decisive reproducibility test is an independent re-implementation of the exclusion search on the published generators; the paper enables this by tabulating explicit group-ring generators for every reported code.","The constructive quotient logicals can be up to $k$ times heavier than the minimum weight, so the move-based gates are not yet a fault-tolerant logical library; composing the shift automorphisms with the homomorphic CNOT gadgets along quotient chains is the implied route to a complete Clifford set, with fault tolerance still to be proven."],"forward_implications":["Rate-1/2 codes with check weight 12 and certified distances up to 16 are realizable in the few-hundred-to-two-thousand-qubit regime of present-day atom-array devices, with syndrome-extraction cycles of roughly 3–9 ms depending on the number of crossed AODs.","The claimed $[[1752,880,14]]$ and $[[2232,1120,16]]$ codes are shorter than every previously known hardware-compatible rate-1/2 code at those distances, breaking the $d\\le w$ barrier with exact certification.","Hardware compatibility and logical operations become inputs rather than post-hoc checks: every GALA code inherits an explicit AOD move schedule and shift automorphisms, and ZX-dual members carry fold-transversal Clifford gates costing one rearrangement and one gate layer — the same primitives as one QEC round.","The GALA family reproduces the previously known rate-1/2 affine-permutation codes as special cases, so the new parameter bounds, logical bases, and ZX-dual variants apply to that whole family."],"supporting_citations":[{"why":"the affine-permutation lifting construction that first demonstrated active orthogonality; GALA generalizes it and reproduces its rate-1/2 codes as instances.","marker":"[1]"},{"why":"the co-designed baseline codes (such as $[[1152,580,12]]$) that the new instances must beat on length and cycle time; its AOD-compatibility constraint motivates factoring the lift into a product group.","marker":"[2]"},{"why":"bivariate bicycle codes, the rate $\\le 1/12$ baseline used for logical-error-rate and qubit-overhead comparisons.","marker":"[3]"},{"why":"the pair-partition cyclic-lift construction whose $[[1524,766,14]]$ is the comparison point at distance 14 and whose certification standard the exact enumeration follows.","marker":"[10]"},{"why":"the fold-transversal Clifford gate construction used to turn GALA ZX-dualities into logical $H$ and $S$ gates.","marker":"[16]"},{"why":"distance estimation used in the adaptive search cascade before exact certification.","marker":"[17]"},{"why":"atom-movement physics and AOD timing parameters underlying the reported millisecond QEC cycles.","marker":"[22]"},{"why":"the stabilizer simulator used to sample circuit-level logical error rates.","marker":"[24]"}],"fun_headline_variants":["GALA codes break the distance barrier at rate 1/2","Rate-1/2 QEC to distance 16 on atom arrays","Designer codes: rate half, distance 16, millisecond cycles","Active orthogonality unlocks rate-1/2 codes with certified distances","Compact self-dual QEC: rate 1/2, distance 12, built for AOD"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported distances 14 and 16 rest on a custom exhaustive enumeration that the paper does not ship and describes only in outline; if that enumeration is incomplete or contains a bug, the barrier-breaking parameter claims collapse.","fun_headline_variants_meta":{"raw":{"variants":["GALA codes break the distance barrier at rate 1/2","Rate-1/2 QEC to distance 16 on atom arrays","Designer codes: rate half, distance 16, millisecond cycles","Active orthogonality unlocks rate-1/2 codes with certified distances","Compact self-dual QEC: rate 1/2, distance 12, built for AOD"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000291,"raw_usage":{"total_tokens":1856,"prompt_tokens":1255,"completion_tokens":601,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":871,"completion_tokens_details":{"reasoning_tokens":499}},"tokens_in":871,"tokens_out":601,"duration_ms":5924,"temperature":1.0,"reasoning_tokens":499,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T04:42:48.961356+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an independent exclusion search on the published group-ring generators of $[[1752,880,14]]$ and $[[2232,1120,16]]$: enumerate every vector in $\\langle H_Z\\rangle^\\perp\\setminus\\langle H_X\\rangle$ and in $\\langle H_X\\rangle^\\perp\\setminus\\langle H_Z\\rangle$ of weight below the claimed distance (even weights through 12 and 14, since all logicals are even). Finding any such vector refutes the certified distance, and a verifier that reproduces both exclusions plus the weight-$d$ witnesses would confirm it.","supporting_citations":[{"cited_title":"2026 , eprint=","cited_arxiv_id":null,"evidence_quote":"the affine-permutation lifting construction that first demonstrated active orthogonality; GALA generalizes it and reproduces its rate-1/2 codes as instances."},{"cited_title":"2026 , eprint=","cited_arxiv_id":null,"evidence_quote":"the co-designed baseline codes (such as $[[1152,580,12]]$) that the new instances must beat on length and cycle time; its AOD-compatibility constraint motivates factoring the lift into a product group."},{"cited_title":"2025 , eprint=","cited_arxiv_id":null,"evidence_quote":"bivariate bicycle codes, the rate $\\le 1/12$ baseline used for logical-error-rate and qubit-overhead comparisons."},{"cited_title":"Nature Physics , volume=","cited_arxiv_id":null,"evidence_quote":"the pair-partition cyclic-lift construction whose $[[1524,766,14]]$ is the comparison point at distance 14 and whose certification standard the exact enumeration follows."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"the fold-transversal Clifford gate construction used to turn GALA ZX-dualities into logical $H$ and $S$ gates."},{"cited_title":"Degenerate quantum","cited_arxiv_id":null,"evidence_quote":"distance estimation used in the adaptive search cascade before exact certification."},{"cited_title":"2026 , eprint=","cited_arxiv_id":null,"evidence_quote":"atom-movement physics and AOD timing parameters underlying the reported millisecond QEC cycles."},{"cited_title":"2024 , eprint=","cited_arxiv_id":null,"evidence_quote":"the stabilizer simulator used to sample circuit-level logical error rates."}],"review_version":1}