{"id":"ec59fb3c-3d25-445c-a234-d4609b4cbd08","arxiv_id":"2505.14327","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"CSS code lifting can be performed with the Freedman-Hastings handlebody realization, equivalent to Tanner cone-complex lifting, classifying hypergraph-product code lifts by subgroups of π1(T1)×π1(T2).","lead":"Quantum CSS codes can be lifted by taking covering spaces of their handlebody realization, a path shown here to be equivalent to the earlier Tanner cone-complex lifting. This gives a topological classification of all lifts of hypergraph-product codes and clarifies when the two lifting frameworks coincide.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.10's classification may fail because Lemma 4.14 assumes all cellular realizations share a 1-skeleton, while Z-type 1-cells depend on pairing choices and can vary.","rationale":"The reader's central concern is the dependence of the cellular-lift/Tanner-lift equivalence on support-preserving Z-lifts (§4.3, Appendix A). That is a real caveat, but Theorem 4.10 is about HPCs, which do admit support-preserving Z-lifts, so that assumption is satisfied for the main classification claim. The more load-bearing issue for Theorem 4.10 is internal: the proof that every L ∈ L(C) has the required fundamental group is incomplete. Lemma 4.14 invokes Lemma 4.7's 'identical 1-skeletons' hypothesis without establishing it, and the construction in §4.1 allows Z-type 1-cells to vary with the edge pairing used to form Gz. If the 1-skeleton can vary, π1(L) can vary, and the classification by subgroups of π1(T1)×π1(T2) overcounts or undercounts lifts. Additionally, Lemma 4.12's retraction construction assumes matchings that are not guaranteed for arbitrary Tanner graphs. These are fixable if the classification is restricted to a fixed pairing/1-skeleton or if a different proof of π1-invariance is supplied, but as written they are gaps in the central theorem. I therefore retain the reader's CONDITIONAL verdict, with a different, more specific reason. The concrete test would settle whether the gap is real: constructing two cellular realizations with different pairing choices and computing their fundamental groups.","tokens_in":22585,"tokens_out":14362,"duration_ms":123202,"concrete_test":"Take the HPC with T1 any connected bipartite graph (e.g., a 3-cycle with doubled edges to make it bipartite) and T2 the Tanner graph of the length-2 repetition code, H2 = [1 1]. Using the paper's Z-lift, construct two cellular realizations L1, L2 ∈ L(C) from the same Z-lift but with different edge pairings at an X-vertex of degree ≥4 in some eTz, one pairing keeping Gz connected and one creating two components. Compute π1(L1) and π1(L2) directly (e.g., via a presentation from the 2-skeleton). If they differ, Lemma 4.14 is false as stated and Theorem 4.10's classification by subgroups of π1(T1)×π1(T2) does not cover all cellular lifts. Separately, check whether the required edge sets in Lemma 4.12 exist for this T2; if not, the lemma's proof fails for a concrete case.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 4.10 claims that cellular-lifts of a hypergraph-product code C are classified by subgroups of π1(T1)×π1(T2). The proof chain is: Lemma 4.12 constructs one L ∈ L(C) with L^2 ≃ T1×T2; Lemma 4.14 asserts every L ∈ L(C) satisfies π1(L) ≅ π1(T1)×π1(T2); then Theorem 4.6 (invariance under 2-cell choices) plus the Galois correspondence gives the classification.\n\nThe gap is in Lemma 4.14. Its proof says: 'There exists a cellular realization homotopy equivalent to T1×T2 ... By Theorem 4.7, any other cellular realization has an isomorphic fundamental group.' Lemma 4.7 (which the text also calls Theorem 4.7) requires the two cellular realizations to have identical 1-skeletons. But L(C) as defined in §4.1 does not fix the 1-skeleton: the Z-type 1-cells are introduced to connect the components of each graph Gz, and the number of components of Gz depends on the pairing of oppositely signed edges at X-vertices in §3.4. That pairing is not determined by the Z-lift. Even with the support-preserving HPC Z-lift, a connected subgraph eTz can be split by the pairing (e.g., a star graph with center x becomes several degree-2 vertices, producing multiple components), so different L ∈ L(C) can have different numbers of Z-type 1-cells and hence different fundamental groups. Lemma 4.14 never verifies that all L have identical 1-skeletons; without that, the isomorphism π1(L) ≅ π1(T1)×π1(T2) is unproved, and the classification is incomplete.\n\nIndependently, Lemma 4.12's retraction requires selecting edge sets E^f_1 ⊂ E1 and E^f_2 ⊂ E2 with no shared endpoints, covering every check of T1 and every bit of T2 respectively. Such matchings need not exist for arbitrary bipartite Tanner graphs; e.g., when T2 is the length-2 repetition code (|V2b|=2, |V2c|=1), no matching covering all bits exists. Thus the lemma's construction is not generally applicable, and the paper provides no alternative proof of the homotopy equivalence L^2 ≃ T1×T2.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a topological code-lifting procedure based on the Freedman-Hastings handlebody realization of a CSS code. It defines a 5-dimensional cellular realization L(C) of a code, introduces the notion of a cellular-lift via finite covers of L(C), and gives explicit formulas for the lifted boundary maps. It then argues that, when the code admits a support-preserving Z-lift, the cellular-lift is equivalent to the author's earlier Tanner cone-complex lift, and it uses this equivalence to classify cellular-lifts of hypergraph-product codes: for C = C1 ⊗ C2*, the lifts are classified by subgroups of π1(T1) × π1(T2), where T1 and T2 are the Tanner graphs of the two classical codes. The paper also contains an appendix exhibiting a CSS code with no support-preserving Z-lift and a corollary asserting that asymptotically good quantum LDPC codes can be obtained by this cellular-lifting procedure.","tokens_in":22988,"tokens_out":8169,"duration_ms":71340,"significance":"If the main theorem and its supporting lemmas are correct, the paper provides a useful handlebody-based perspective on code lifting and an explicit topological counterpart to the author's Tanner cone-complex machinery. The explicit lifted boundary-map formulas in Eqs. (1) and (3), together with the clean proof of the chain-complex condition in Theorem 4.4, are concrete assets, and the support-preserving Z-lift for hypergraph-product codes is given explicitly. The classification statement, subgroups of π1(T1) × π1(T2), is a clean and potentially useful reformulation of known results. However, the paper is transparent that the classification is equivalent to a prior result from [Gue25], and the proof of this equivalence via cellular realizations is where the main technical gaps occur; the significance of the paper therefore depends on whether those gaps can be closed.","major_comments":[{"comment":"The proof of Lemma 4.14 is not valid as written. The lemma claims that every L ∈ L(C) has π1(L) ≅ π1(T1) × π1(T2), by combining Lemma 4.12's existence of one realization homotopy equivalent to T1 × T2 with 'Theorem 4.7'. However, Theorem 4.6/4.7 applies only to cellular realizations with identical 1-skeletons, and L(C) as defined in §4.1 does not fix the 1-skeleton. In §3.4, the graph Gz is constructed using a pairing of oppositely signed edges at each X-vertex, and the paper itself notes that Gz can be disconnected 'due to the choice of edge pairing'; in §4.1, the Z-type 1-cells are introduced precisely to connect the components of Gz. Different pairings can therefore produce different numbers of Z-type 1-cells and, potentially, different fundamental groups. The proof of Lemma 4.14 never establishes that all L have the same 1-skeleton, so the isomorphism π1(L) ≅ π1(T1) × π1(T2) is not proved for arbitrary L, and Theorem 4.10 is not established by this argument.","section":"§4.4, Lemma 4.14"},{"comment":"Lemma 4.12 contains an unproved, and as stated questionable, graph-theoretic assertion. The proof requires that for each check vertex of T1 and each bit vertex of T2 one can choose edge sets E1^f and E2^f such that, for every z-vertex (v1,v2) ∈ V1^b × V2^c, there exists e ∈ E1^f such that for all v ∈ N(v2) ∪ (v1,v2) the horizontal edge (e,v) lies in the star of (v1,v2), with a symmetric vertical condition. No argument is given that such marked edges exist for arbitrary bipartite Tanner graphs, and the displayed condition is not well formed as written: it mixes a vertex (v1,v2) with the neighbor set N(v2), and e ∈ E1^f need not be incident to v1 merely because it was selected for some check vertex of T1. The proof also asserts that after several local retractions a global deformation retraction can be obtained, but the required spanning-forest property is not proved. Since Lemma 4.12 is the only place where the homotopy equivalence L^2 ≃ T1 × T2 is established, this is a load-bearing gap in the proof of Theorem 4.10.","section":"§4.5, Lemma 4.12"},{"comment":"The claimed classification depends essentially on the existence of a support-preserving Z-lift, and the paper is not fully self-contained on this point. Section 4.3 states that the cellular-lift and the Tanner-lift are equivalent only when a support-preserving Z-lift exists, and Appendix A gives a CSS code with no such lift. For codes without this property, the paper itself notes that the fundamental group of the cellular realization can differ from that of the Tanner cone-complex. Consequently, Theorem 4.10 is not a classification of all cellular-lifts of all HPCs beyond the support-preserving case, and the proof does not provide an independent derivation of the Tanner-lift classification. The paper should either prove the equivalence in the support-preserving case with full detail or explicitly state Theorem 4.10 as a reformulation of the known classification from [Gue25] rather than as a new classification theorem.","section":"§4.3 and Appendix A"}],"minor_comments":[{"comment":"The phrase 'π1(L) and T(C) (the Tanner cone-complex)' conflates the Tanner graph T(C) with the Tanner cone-complex K(C); the proof should refer to the fundamental group of the cone-complex K(C), not of the graph T(C).","section":"§4.3, Lemma 4.5 proof"},{"comment":"There are several typographical errors: 'bundary' in Section 3.1, 'indix' in Section 3.1, 'classication' in Section 4.4, and 'repeatidly' in the proof of Lemma 4.7.","section":"§3.1 and §4.4"},{"comment":"The reference [hes] is given as a MathOverflow URL with the author name in the citation key; it should be formatted as a standard bibliographic entry with author, title, URL, and access date.","section":"References"},{"comment":"The proof of Theorem 4.6 states that replacing the 2-cells 'does not change the incidence' between lifts of a Z-check and its neighboring qubits; this step is asserted without justification and should be expanded, since it is part of the claimed invariance of the resulting code.","section":"§4.2, Theorem 4.6 proof"},{"comment":"In Example 4.13, 'the simplest spanning forest has on edge out of two' should read 'has one edge out of two'.","section":"§4.5, Example 4.13"}],"recommendation":"major_revision","confidential_remarks":"The paper is clearly related to the author's earlier work [Gue25], and the classification result is presented as an equivalent formulation of an already published result. This is acceptable if the equivalence proof is solid, but the current gaps in Lemmas 4.12 and 4.14 concern exactly that equivalence. I would ask for a complete proof of the spanning-forest property in Lemma 4.12 and a rigorous argument in Lemma 4.14 showing that the relevant cellular realizations share a 1-skeleton, or a reformulation of Theorem 4.10 that avoids claiming invariance across all of L(C). The explicit lifted boundary maps and Theorem 4.4 are useful and should be preserved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this paper makes the Freedman–Hastings handlebody into a working code-lifting device and shows it is essentially the same as the Tanner cone-complex lifting from the author's earlier paper [Gue25]. The construction is explicit, the lifted boundary formulas in Section 4.2 are new, and Theorem 4.4 gives a clean proof that the lifted maps compose to zero. That part is solid.\n\nWhat's genuinely new: the cellular realization of a code, the explicit covering construction on the 1-skeleton and then lifting 2-cells, and the invariance statements (Lemma 4.7, Theorem 4.6) for realizations with identical 1-skeletons. The paper is also honest about the conditionality: the equivalence with Tanner-lifts requires a support-preserving Z-lift, and Appendix A gives a Fano-plane code that admits none. That limitation is stated clearly, not buried.\n\nThe soft spots are in the self-contained classification of HPC lifts (Theorem 4.10). Lemma 4.12's proof requires choosing edge sets E1^f and E2^f that cover all checks/bits with no shared endpoints. That's a spanning-forest assumption that can fail for simple Tanner graphs—e.g., a length-2 repetition code on the C2 side has one check adjacent to both bits, so no such selection exists. The proof doesn't address this. And Lemma 4.14 invokes Lemma 4.7 to claim all cellular realizations have the same pi_1, but Lemma 4.7 only applies when the 1-skeletons are identical. The number of Z-type 1-cells can vary with the pairing choices at X-vertices, so that precondition isn't verified. So the proof of Theorem 4.10 as written has a real gap.\n\nThat said, the theorem itself is likely true: the paper explicitly says it is an equivalent version of a result already used in [Gue25], and the equivalence with the Tanner-lift, when the support-preserving Z-lift exists, covers HPCs. So the gap is in the self-contained proof, not necessarily in the mathematical claim. Still, a referee should ask the author to either fix the spanning-forest argument, restrict the statement, or rely explicitly on the prior classification rather than claiming a self-contained proof.\n\nWho is this for? People working on quantum LDPC constructions and topological codes. It gives a different geometric lens and some reusable explicit formulas. It deserves a serious referee; the conditional claims are honest and the construction is genuinely useful. I'd recommend sending it to review, with the expectation that the HPC classification proof needs revision.","headline":"Useful geometric translation of code lifting with honest caveats; the self-contained HPC classification proof has gaps that need patching.","tokens_in":23624,"tokens_out":3988,"would_cite":true,"duration_ms":37154,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-07T15:36:57.028261+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}