{"id":"9b97b0a7-ca2c-430b-a427-d0af35ba9a1b","arxiv_id":"2505.13679","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A pedagogical re-derivation of balanced product quantum LDPC codes using parity-check matrices, with worked examples and proof sketches, containing no new theorem beyond the known literature.","lead":"This paper is a review that rebuilds balanced product quantum error-correcting codes from plain parity-check matrices instead of abstract graph products. It is a useful entry point for researchers who want the construction, a distance proof sketch, and worked examples.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section III's unproved distance-balancing formula \\tilde d_X = d_X d_C carries the O([[N,N^{4/5},N^{3/5}]]) headline; the paper explicitly declines to rule out shorter dressed operators, so the final scaling is not established by the review.","rationale":"The reader's weakest assumption already identifies the distance-balancing step, and I agree that this is the load-bearing gap. The paper is honest about the omission, but the final asymptotic is built on it, so CONDITIONAL is appropriate: either prove the distance lower bound or explicitly label it as a result imported from [20] before the headline is used. I do not see a fatal flaw: the construction is reproducible, small examples check out, and the theorem is known from the cited literature. I also noticed an apparent algebraic slip in the Section II.C X-distance proof: after bounding |v(t0)| <= ((s-1)alpha+1)/s l and |I v(t0)| <= 2 gamma_X l, the chosen gamma_X < beta((s-1)alpha+1)/(2s) does not contradict expansion because expansion only gives |I v(t0)| > beta(1-alpha)/s l; replacing the constant with beta(1-alpha)/(2s) appears to close the gap. This is a repairable proof error rather than a threat to the theorem, but it should be corrected in a revision. On the big question, the unproved distance-balancing formula is the one condition whose failure would change the central claimed scaling.","tokens_in":15288,"tokens_out":29390,"duration_ms":265234,"concrete_test":"Take the [[360,26,5]] code of Example 2 (or the [[12,2,3]] example if resources are tight), balance it with the [7,4,3] Hamming parity-check matrix H_C (3 x 7) via Definition 3, and compute the exact distance of the resulting CSS code with the SAT solver [26]. If an X-type logical operator of weight < 5*3 = 15 (for Example 2) exists, then \\tilde d_X = d_X d_C is false and the final scaling is not established; if the minimum is exactly 15, the concern is resolved for this case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III (Definition 3 and the paragraph after the logical-operator construction) asserts \\tilde d_X = d_X d_C and \\tilde d_Z = d_Z, but after asking \"could there be anything shorter?\" it states \"We will not give the full details here.\" The supplied argument verifies only the logical-qubit count K = k k_C via row dependencies; it never lower-bounds the weight of X-type logical operators in the balanced code. This is load-bearing because Section IV.A's final asymptotic depends on it: the LPS construction gives, before balancing, N = O(q^3), k = O(q^2), d_X = O(q), d_Z = O(q^3); choosing a classical code with d_C = O(q^2) is what produces N = O(q^5), k = O(q^4), d = O(q^3). If a dressed X-logical of weight below d_X d_C exists, the balanced code's distance is no longer O(q^3) and the claimed O([[N,N^{4/5},N^{3/5}]]) scaling is unsupported. Since [20] is cited for this step, the underlying result may be sound, but the review does not provide the proof its own construction needs.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a pedagogical review of the balanced product quantum LDPC code construction of Breuckmann and Eberhardt. It presents the construction directly in terms of parity-check matrices: for an expanding binary matrix I with an order-l permutation symmetry, the CSS code is defined by H_X = [I^T, 1+C] and H_Z = [1+R, I] (Eq. (3)). The paper proves the CSS condition, counts logical qubits, gives a distance proof for the asymmetric code following Panteleev--Kalachev, describes a distance-balancing procedure from Evra--Kaufman--Zémor, and combines these with LPS expander graphs to claim codes with parameters O([[N,N^{4/5},N^{3/5}]]). Worked examples, including a 6-cycle repetition code and an LPS graph with p=3, q=5, are checked with a SAT solver whose code and data are deposited in two repositories.","tokens_in":15514,"tokens_out":10550,"duration_ms":93911,"significance":"If the exposition is accurate, the paper serves a useful purpose: it lowers the barrier to understanding a landmark construction in quantum LDPC code theory, provides explicit parity-check matrices rather than abstract chain complexes, and gives independently checkable examples and data. The claimed asymptotic O([[N,N^{4/5},N^{3/5}]]) is already established in the literature, so the paper's contribution is pedagogical rather than novel in results. The main strengths are the concrete matrix formulation, the worked examples with machine-checked distances, and the attempt to give a self-contained distance proof for the asymmetric code. However, the final asymptotic is only as strong as the distance-balancing lower bound, which the manuscript does not prove, and the main distance proof contains a gap in the X-distance part.","major_comments":[{"comment":"The distance-balancing lower bounds \\tilde d_X = d_X d_C and \\tilde d_Z = d_Z are asserted immediately after Definition 3, but the text explicitly says \"We will not give the full details here\" and only constructs logical operators of weight d_X d_C and d_Z. The final claim in Section IV.A that the balanced product with LPS graphs yields O([[N,N^{4/5},N^{3/5}]]) is load-bearing on exactly this missing lower bound: the balanced distance is obtained by multiplying the O(q) X-distance by a classical code with d_C = O(q^2). As written, the manuscript does not rule out shorter dressed operators in the balanced code, so the headline scaling is not established by the review. Please either provide the full proof of the distance-balancing lower bounds or state the result as an imported theorem with a precise citation to [20], and explicitly mark in Section IV that the final scaling depends on this external result.","section":"Section III, Definition 3"},{"comment":"In the X-distance proof, t0 is defined as the smallest integer such that |v(t0)| > (1−α)l/s, and the text derives |v(t0)| ≤ ((s−1)α+1)l/s. It then claims that choosing γ_X < β((s−1)α+1)/(2s) contradicts the expansion property. This does not follow: expansion only gives |Iv(t0)| ≥ β|v(t0)| > β(1−α)l/s, while the proof's upper bound is |Iv(t0)| ≤ 2|u| < 2γ_X l. To obtain a contradiction one needs 2γ_X l < β(1−α)l/s, which is not implied by the displayed choice of γ_X because ((s−1)α+1) > (1−α). As written, the proof does not establish the claimed d_X = γ_X l bound; the argument would need either the smaller choice γ_X < β(1−α)/(2s) or an additional argument showing that |v(t0)| is actually close to its upper bound.","section":"Section II.C, X-distance proof"},{"comment":"The asymptotic claim depends on the existence of a family of expanding matrices I with the exact order-l symmetry of Definition 1, bounded row and column weight s, and kernel dimension scaling as q^2; this is imported from the LPS graphs [22], and the manuscript states without proof that these graphs have \"an appropriate symmetry of order l = q.\" Similarly, the random-expander claim in the first paragraph of Section IV (\"With high probability, we get a code that is O([[nl,nl,l,n]]) before distance balancing\") is asserted without proof. Since this is a review, importing established results is acceptable, but the provenance and precise statements of these existence claims should be made explicit in the text so that the final O([[N,N^{4/5},N^{3/5}]]) is clearly presented as a combination of cited results rather than a derivation completed in this paper.","section":"Section IV"}],"minor_comments":[{"comment":"The remark \"Assuming a large enough code 2βγXl<αn that a token case can be neglected\" is too vague for a proof that claims to give full details; the precise condition on n/l needed for the expansion property to apply to v(t0) should be stated in the main text.","section":"Section II.C, footnote 30"},{"comment":"In the list of orbit elements, \"AH, AH 2, AH 4, AH 4\" appears to contain a typo; it should presumably be \"AH, AH^2, AH^3, AH^4\".","section":"Example 2"},{"comment":"The dimension formula is written as k = n − mA − mB, but the symbols mA and mB are not defined at that point; the later text uses mX and mZ, so the notation should be harmonized.","section":"Section I"},{"comment":"The sentence \"The X and Z distances satisfy dX,dZ ≤ 5\" followed by \"Hence, the code is [[360,26,5]]\" would be clearer if the text explicitly noted that the upper bounds are verified to be tight by the SAT solver.","section":"Example 2"}],"recommendation":"major_revision","confidential_remarks":"This is a review/tutorial manuscript rather than an original research contribution. Its value depends on the correctness and transparency of the reconstructed proofs. The Section II.C gap and the unproved distance-balancing lower bound in Section III are fixable but are load-bearing for the final asymptotic claim, so they should be addressed squarely. The self-cited repositories appear to provide checkable artifacts and no circularity issue is apparent. If the journal publishes review articles, the low novelty is acceptable, but the technical accuracy needs to be brought to the standard the paper itself claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is a genuinely useful review, not a new-result paper. The authors rewrite balanced product codes from [1] in parity-check language, which does demystify the construction. The worked LPS example (p=3, q=5) with SAT-verified distance and explicit logical operators is a nice concrete anchor that the original papers lack. The distance proof for the main theorem is a coherent adaptation of [19] and is mostly self-contained for the left block.\n\nThe main soft spot is Section III. The distance-balancing formula \\tilde d_X = d_X d_C is asserted but not proved. The authors explicitly say \"We will not give the full details here\" and cite [20]. For a review that is acceptable signaling, but the final asymptotic O([[N,N^{4/5},N^{3/5}]]) depends on exactly that formula. A reader who wants to trust the balanced product construction for new work must go to [20]. The paper would be stronger if it either proved the lower bound or marked the formula clearly as an imported result. The stress-test note is fair: the logical-operator construction gives an upper bound on distance, not a lower bound.\n\nOther points are minor. The random expander comment in Section IV is plausible but unproved. Footnote 30 acknowledges a neglected token case. The self-cited repositories are checkable, which is good practice.\n\nOverall, this is an honest, carefully written review. It does not overclaim. The math that is shown is correct, and the attribution is clear. For someone wanting to understand balanced product codes without wading through the original group-theoretic presentation, this is a good entry point. It deserves serious peer review, probably with a request to either prove the balancing bound or label it as a cited result. I'd send it to review.","headline":"Solid review that earns its keep as a teaching tool; just be aware the headline distance scaling rests on an unproved (but cited) balancing lemma.","tokens_in":16067,"tokens_out":2365,"would_cite":false,"duration_ms":21795,"reading_group":"maybe","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":"The paper shows that balanced product quantum codes reduce to a concrete pair of parity-check matrices built from an expanding matrix with a cyclic symmetry, and that these matrices yield low-density CSS codes with distance scaling…","keywords":["balanced product code","quantum LDPC codes","CSS codes","expander graphs","parity-check matrices","distance balancing","storage density","LPS expander graphs"],"falsifier":"Fix the LPS incidence matrix for a growing sequence of primes $q$ (holding $p$ fixed), build $H_X$ and $H_Z$ from Definition 1, and compute the exact minimum weights of dressed logical operators; if the $X$-distance or $Z$-distance stops growing linearly in $l$ or $m$ with fixed positive constants, Theorem 1's distance claim is false. Then apply the same exact search to the distance-balanced matrices and look for any logical operator of weight below $d_X d_C$ or $d_Z$.","tokens_in":15031,"feed_emoji":"⚛️","tokens_out":20206,"duration_ms":144171,"temperature":0.7,"pith_summary":"The paper sets out to make balanced product quantum LDPC codes usable by rewriting the construction directly in terms of parity-check matrices, removing the chain-complex machinery of the original presentation. Its central claim is that when a binary matrix $I$ with an $(\\alpha,\\beta)$-expanding property carries an order-$l$ permutation symmetry ($R I = I C^T$ with $R^l = C^l = 1$), the two matrices $H_X = [I^T, 1+C]$ and $H_Z = [1+R, I]$ automatically define a CSS code: the commutation $H_X H_Z^T = 0$ holds, the code is low density, and its distances are linear in $l$ and $m$ respectively. The paper works through the distance proof, then adds a distance-balancing construction that equalizes the two distances, and applies the whole recipe to LPS expander graphs. If correct, this gives a concrete path from a classical expander graph to a quantum code with parameters $O([[N, N^{4/5}, N^{3/5}]])$, a storage density $kd^2$ on the order of $N^2$, and a distance that grows faster than $\\sqrt{N}$.","feed_headline":"Two matrices build quantum codes that beat the square-root barrier","feed_subtitle":"The parity-check recipe makes dense quantum LDPC codes explicit, reaching a distance beyond the square-root barrier.","key_machinery":"The central object is the balanced product parity-check pair, $H_X = [I^T, 1+C]$ and $H_Z = [1+R, I]$. The matrix $I$ is the incidence matrix of an expanding graph; the permutation matrices $R$ and $C$ implement an order-$l$ symmetry $R I = I C^T$, and the blocks $1+R$ and $1+C$ are parity-check matrices of repetition codes over each orbit. The symmetry does two jobs: it ensures the CSS orthogonality $H_X H_Z^T = 0$ modulo 2, and it is the mechanism that quotients out the orbit structure, removing the tensor-product blow-up in qubit count that limits hypergraph products. The distance argument then runs on expansion: any short dressed logical operator would produce a short vector whose image under $I$ or $I^T$ expands beyond the allowed weight. A secondary mechanism is Definition 3's distance balancing, a block construction combining the quantum code with a good classical LDPC code to equalize $d_X$ and $d_Z$ at the cost of more physical qubits.","core_discovery":"The central claim is Definition 1 plus Theorem 1: for an $(\\alpha,\\beta)$-expanding matrix $I \\in \\{0,1\\}^{m\\times n}$ with row and column weight at most $s$, a cyclic symmetry $R I = I C^T$ whose orbits all have length $l$, and $\\max(|\\ker I|, |\\ker I^T|) = k_0$, the parity-check matrices $H_X = [I^T, 1+C]$ and $H_Z = [1+R, I]$ define a low-density CSS code with parameters $[[n+m, k_0/l, d_X = \\gamma_X l, d_Z = \\gamma_Z m]]$. The distance proof shows that expansion forbids short logical operators: a dressed logical operator localized on the left block would force a vector $u(t_0)$ of weight below $\\alpha m$ whose image under $I^T$ is too heavy, while the symmetric orbit-sum construction gives $X$-logicals of weight $l$. The paper further claims that Definition 3's distance balancing converts such a code into one with equalized distances, and that with the LPS expander family this yields $O([[N,N^{4/5},N^{3/5}]])$ quantum LDPC codes.","pith_inferences":["A natural next search is for families of expanding matrices whose kernels contain vectors with far more symmetry than the $k_0/l$ orbit average guarantees; if such families exist, the lower bound $k \\geq k_0/l$ could be beaten and balanced product codes would exceed, rather than match, the $kd^2 \\sim N^2$ threshold.","Because the paper leaves the part of distance balancing that rules out shorter logical operators to the reader, the balancing claim can be stress-tested on finite examples by building the parity-check matrices explicitly and checking whether any dressed logical operator below $d_X d_C$ or $d_Z$ exists.","The Tanner code variant suggests a design freedom the paper does not develop: choosing local codes with growing distance $d_l$ could multiply the $X$-distance by $d_l$ while adding physical qubits, potentially producing a different trade-off curve between $N$, $k$, and $d$.","The small LPS example with parameters $[[360,26,5]]$ and its subsystem variant offers a concrete benchmark for recomputing exact distances and for probing how the constants $\\gamma_X$ and $\\gamma_Z$ behave as $q$ grows."],"forward_implications":["Given any expanding matrix $I$ with the required cyclic symmetry, the recipe produces a quantum LDPC code immediately from its incidence matrix, with no chain-complex or fiber-bundle abstraction needed.","For LPS expander graphs with fixed $p$ and growing $q$, the balanced product gives a code whose parameters before balancing are $O(q^3)$ physical qubits, $O(q^2)$ logical qubits, and distances $O(q)$ and $O(q^3)$; after distance balancing this becomes $O([[N,N^{4/5},N^{3/5}]])$.","The construction reaches $kd^2 \\sim N^2$, the same storage-density scaling as hypergraph product codes, but with distance scaling $N^{3/5}$ instead of $\\sqrt{N}$, so the code's storage density is no worse while its distance grows strictly faster.","The paper's bound $k \\geq k_0/l$ means the number of logical qubits could in principle be larger than $k_0/l$; in the explored examples the bound is saturated, so exceeding it would push past the $kd^2 \\sim N^2$ threshold.","The distance-balancing step turns any asymmetric $[[n,k,d_X,d_Z]]$ code into an $O([[n\\max(d_X,d_Z)/\\min(d_X,d_Z), k\\max(d_X,d_Z)/\\min(d_X,d_Z), \\max(d_X,d_Z)]])$ code, so the balancing recipe is reusable beyond balanced products."],"supporting_citations":[{"why":"Supplies the original balanced product construction and its main theorem, which this paper re-presents in parity-check language.","marker":"[1]"},{"why":"Provides the expansion-based distance proof technique that Theorem 1's proof follows.","marker":"[19]"},{"why":"Supplies the distance-balancing construction used as Definition 3 to equalize $d_X$ and $d_Z$.","marker":"[20]"},{"why":"Provides the LPS expander graph family with the required symmetry and expansion properties used for the $O([[N,N^{4/5},N^{3/5}]])$ scaling.","marker":"[22]"},{"why":"Gives the hypergraph product construction, the $l=1$ special case of the balanced product and the $kd^2 \\sim N^2$ baseline this paper compares against.","marker":"[17]"}],"fun_headline_variants":["Balanced product codes: dense quantum LDPC with distance beyond sqrt(N)","How balanced product codes beat the square-root distance barrier","New look at balanced product codes: explicit construction for good quantum LDPC","Balanced product codes demystified: parity-check matrices for high-density codes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that a family of expanding matrices with exactly the cyclic symmetry of Definition 1 exists, together with the paper's explicit choice in Section III not to prove that distance balancing creates no dressed logical operators shorter than $d_X d_C$ or $d_Z$.","fun_headline_variants_meta":{"raw":{"variants":["Balanced product codes: dense quantum LDPC with distance beyond sqrt(N)","How balanced product codes beat the square-root distance barrier","New look at balanced product codes: explicit construction for good quantum LDPC","Balanced product codes demystified: parity-check matrices for high-density codes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000766,"raw_usage":{"total_tokens":3363,"prompt_tokens":877,"completion_tokens":2486,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":2408}},"tokens_in":493,"tokens_out":2486,"duration_ms":19334,"temperature":1.0,"reasoning_tokens":2408,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:12:14.202684+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix the LPS incidence matrix for a growing sequence of primes $q$ (holding $p$ fixed), build $H_X$ and $H_Z$ from Definition 1, and compute the exact minimum weights of dressed logical operators; if the $X$-distance or $Z$-distance stops growing linearly in $l$ or $m$ with fixed positive constants, Theorem 1's distance claim is false. Then apply the same exact search to the distance-balanced matrices and look for any logical operator of weight below $d_X d_C$ or $d_Z$.","supporting_citations":[{"cited_title":"Hence, the number of edges is q(q2− 1)(p + 1)/2","cited_arxiv_id":null,"evidence_quote":"Supplies the original balanced product construction and its main theorem, which this paper re-presents in parity-check language."},{"cited_title":"Yamasaki and M","cited_arxiv_id":null,"evidence_quote":"Gives the hypergraph product construction, the $l=1$ special case of the balanced product and the $kd^2 \\sim N^2$ baseline this paper compares against."}],"review_version":1}