{"id":"e69da9df-0b43-4057-9ba3-e3d1ae893c72","arxiv_id":"2507.19454","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Random isometries almost never approximate-mask into two subsystems, but almost always into many subsystems, with the number of physical qubits scaling linearly in the number of logical qubits.","lead":"Random ways of hiding quantum information almost never work when the system is split into two parts, but almost always work when split into many parts, and the overhead grows only linearly with the number of hidden qubits. The paper also links approximate masking to approximate quantum error correction, producing random codes with constant rate and exponentially small errors.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The AQECC consequence is proven only for replacement noise; the abstract and Section V.B advertise a general equivalence, so the advertised error-correction claim is broader than the theorems support.","rationale":"The reader's weakest-assumption analysis correctly identifies the replacement-channel restriction in Lemma 1 as the main load-bearing gap. I reviewed the bipartite no-go results (Proposition 2, Theorems 1 and 2) and the multipartite positive results (Theorems 4-7, Corollaries 1 and 2) and found the derivations internally consistent; the concentration arguments via epsilon-nets and the union bound over bipartite cuts are standard, and the lower bound w>1/9 for dC≥2 is supported by Appendix C. The unproved random-matrix approximation in Eq. (31) is non-essential, and the Appendix F typo about the code-rate coefficient does not affect the qualitative conclusion. The one place where the paper overclaims is the AQECC implication: Eq. (52) and Lemma 1 are explicitly restricted to replacement channels, yet the abstract and Section V.B describe the result as a general equivalence between AQIM and AQECC. This does not invalidate the random AQIM theorems, but it does mean the advertised error-correction consequence is broader than the proof supports. The appropriate response is to keep the conditional verdict and require the authors to prominently qualify the noise model in the abstract and in the summary of implications. My concrete test would settle whether the inequality fails for a simple non-replacement noise; if it does, the qualification is mandatory.","tokens_in":38145,"tokens_out":35601,"duration_ms":359590,"concrete_test":"Verify whether Lemma 1 (Eq. 52) remains valid for a non-replacement noise on a k-party subsystem, e.g., the single-qudit depolarizing channel N = id_{S^c} ⊗ D_p with D_p(ρ)=(1-p)ρ + p I/d_S, applied to a code from Theorem 8 with m=6, k=2, dC=4. Numerically compute the QEC inaccuracy eη(E,N) and the subsystem variance η(HC,k); if eη exceeds sqrt(dC)η(HC,k), the equivalence does not extend to general noise and the abstract must be qualified to replacement channels.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Corollary 4 and Theorem 8 rely on Lemma 1 from reference [51], whose inequality (52) bounds the QEC inaccuracy eη(E,R_S) by the subsystem variance η(HC,k) only when the noise R_S is a replacement channel R_S(ψ)=Tr_S(ψ)⊗γ_S. The paper states this restriction in the technical statement ('replacement errors') but the abstract and parts of Section V.B present the result as a general equivalence between AQIM and AQECC, yielding 'approximate quantum error correction codes with constant code rates and exponentially small correction inaccuracies'. For general noise, e.g., dephasing or depolarizing noise on a subsystem, the equivalence between subsystem variance and QEC inaccuracy is not established by the cited lemma. This is not a flaw in the random AQIM theorems themselves, but it is the weakest load-bearing premise of the advertised AQECC implication: without the replacement-channel qualifier, the central claim that AQIM 'naturally gives rise to approximate quantum error correction codes' is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies approximate quantum information masking (AQIM) through random isometries. It introduces maximal and average versions of approximate k-uniform masking and several figures of merit, establishing inequalities among them (Proposition 1). For bipartite systems, it proves that the expected average subsystem variation of a random subspace is bounded below by w > 1/9 (Theorem 1), that the probability of falling below w − α is exponentially small (Theorem 2), and hence that random isometries are almost never approximate maskers (Theorem 3). For multipartite systems with equal local dimension d, it shows that a random subspace of dimension d_C is an approximate k-uniform masker with inaccuracy d^{k−m/2}+α except with exponentially small probability (Theorem 6), and gives analogous statements for other figures of merit (Theorem 7, Proposition 4). The paper then derives implications: random multipartite subspaces yield approximate k-uniform states (Corollary 3) and, via Lemma 1 of [51], approximate quantum error-correcting codes (AQECCs) with constant code rate and exponentially small inaccuracy (Corollary 4, Theorem 8). Numerical illustrations and appendices support the main concentration proofs.","tokens_in":38330,"tokens_out":5365,"duration_ms":49352,"significance":"If the central results hold, this paper gives a clean and somewhat counterintuitive separation: random isometries fail to be approximate maskers in bipartite systems even when dimensions are large, yet succeed in multipartite systems with a number of physical qubits that scales only linearly in the number of logical qubits. The proofs use standard tools (Lévy's lemma, epsilon-nets, Lipschitz bounds) and are presented in detail in appendices, which is a strength. The connection to AQECC, if properly scoped, would extend the relevance of masking to quantum error correction and random codes. However, the advertised AQECC equivalence is currently stated more broadly than the theorems support, and one statement of the random-code theorem asserts an equality where the derivation gives only an upper bound. These issues are local and fixable, but they affect the paper's headline claims.","major_comments":[{"comment":"The AQECC consequence is proved only for replacement noise, but the abstract and Section V.B advertise a general equivalence between AQIM and AQECC. Corollary 4 and Theorem 8 rely on Lemma 1 of [51], whose inequality (52) bounds the QEC inaccuracy by the subsystem variance only for replacement channels R_S(ψ)=Tr_S(ψ)⊗γ_S (Section V.B, Eq. (52)). For general noise, e.g., dephasing or depolarizing channels on a subsystem, the equivalence is not established by the cited lemma. This is load-bearing because the advertised claim that 'AQIM naturally gives rise to approximate quantum error correction codes' is unsupported outside the replacement-channel setting. Please revise the abstract, the introduction, Section V.B, and Theorem 8 to state prominently that the AQECC implication holds for replacement errors, or provide a proof for the general case.","section":"Section V.B, Corollary 4, Theorem 8; Abstract"},{"comment":"Theorem 8 states that 'the probability that H_C is an AQECC with inaccuracy eη(E,R_S) = sqrt(d_C)(u+α)' is bounded as in Eq. (55). This asserts an equality, but the preceding results support only an upper bound: Corollary 4 gives eη(E,R_S) ≤ sqrt(d_C) ΛM(H_C,k) and Theorem 7 gives ΛM(H_C,k) ≤ u+α with high probability. The theorem should state eη(E,R_S) ≤ sqrt(d_C)(u+α) (and similarly in the surrounding discussion and Appendix F); as written, it overstates the guarantee.","section":"Theorem 8"},{"comment":"Equation (31) introduces the approximation ΛA_B1(H_C) ≈ (4/(3π)) sqrt(d_1 Tr[(Δψ_B1)^2]) and Eq. (32) derives a corresponding approximate upper bound. This random-matrix approximation is not proved, and it is not used in the subsequent concentration theorems. The text, however, presents it as an analytical result rather than a heuristic. Please either provide a rigorous derivation with the necessary assumptions, or explicitly label this as a numerical/heuristic estimate and clarify that the rigorous results in Theorems 4 and 5 do not depend on it.","section":"Section III.B, Eqs. (31)-(32)"}],"minor_comments":[{"comment":"The statement 'If α = d^{k−m/2} is smaller than d^{k−m/2}' is self-contradictory; it should read 'If α = d^{k−m/2}/c for some constant c>1' or simply 'If α ≤ d^{k−m/2}'.","section":"Corollary 2, case (3)"},{"comment":"Theorem 3 states that the probability of being a δ-approximate masker with δ=1/9 is exponentially small, but the proof is not given explicitly. Since Theorem 2 controls V^A(H_C), the proof should state that V^M(H_C) ≥ V^A(H_C) and that δ=1/9 falls below the lower bound w, to make the inference transparent.","section":"Theorem 3"},{"comment":"In the proof of Lemma 3, after the line showing that terms with i≥2 are non-positive, the next displayed equation writes '2X i=1' (likely a typographical artifact). Please clarify which terms are retained in the summation, so the reader can follow the bound leading to the value 1/3.","section":"Appendix C.1.b, Eq. (C25)"},{"comment":"In the table row for E_H_C[ (Π^{(B_1)}_C − e1_{B_1})^2 ], the notation contains a comma inside the norm, '∥Π^{(B_1)}_C , e1_{B_1}∥', which should be a minus sign: '∥Π^{(B_1)}_C − e1_{B_1}∥'.","section":"Appendix A, table"}],"recommendation":"major_revision","confidential_remarks":"The core random-AQIM theorems appear technically sound and are well documented in the appendices. The paper is publishable once the AQECC claims are scoped to replacement noise, the equality in Theorem 8 is corrected to an upper bound, and Eq. (31) is flagged as a heuristic. The abstract overstates the current result; this should be fixed before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper's real new result is the bipartite no-go theorem: for a random subspace of a bipartite Hilbert space, the average subsystem variation is bounded below by a universal constant w > 1/9, and the probability of falling below w - alpha is exponentially small. That is a clean and surprising statement, and it genuinely generalizes the no-masking intuition to the random approximate setting. The proof is analytic, uses standard Levy/epsilon-net tools, and looks correct.\n\nThe multipartite positive result is also good, though less novel than the abstract suggests. The proof is a union-bound over bipartite cuts of the known concentration results from Hayden-Leung-Winter. The linear scaling of physical qubits with logical qubits follows by simple counting from the asymptotic inequalities. Still, it is a useful and clearly presented construction, and the figures of merit are defined carefully with the relationships among them worked out in the appendices.\n\nThe stress-test concern about the AQECC implication is fair. The technical statements are careful: Corollary 4 and Theorem 8 explicitly mention replacement errors, because the lemma imported from [51] bounds QEC inaccuracy by subsystem variance only for replacement channels R_S(psi)=Tr_S(psi) ⊗ gamma_S. But the abstract and portions of Section V.B describe the consequence as yielding approximate quantum error-correcting codes with constant rates, without prominently flagging the replacement-channel restriction. A reader could take the equivalence to be general, which it is not. That is a presentation flaw, not a flaw in the masking theorems themselves.\n\nOther soft spots are minor. Theorem 8 states the QEC inaccuracy as equal to sqrt(dC)(u+alpha), while Corollary 4 gives this only as an upper bound. The random-matrix approximation in Eq. (31) is introduced without proof, but it is not load-bearing for the main results. There is a typo in Appendix F. None of these undermine the central bipartite and multipartite claims.\n\nOverall, the paper is worth serious refereeing. It gives a genuine no-go theorem for random approximate masking in bipartite systems, a clean positive result in the multipartite case, and a useful set of figures of merit. The main revision should be to make the replacement-channel limitation visible in the abstract and Section V.B, and to fix the equality/upper-bound inconsistency in Theorem 8. If you work on masking, random codes, or AQECC, this is worth citing after those fixes.","headline":"The bipartite no-random-AQIM bound is a solid new result; the multipartite construction is nice but mostly a union-bound extension, and the advertised AQECC implication is broader than the theorems support since the equivalence holds only for replacement noise.","tokens_in":38845,"tokens_out":1510,"would_cite":true,"duration_ms":15948,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes a no-random approximate masking theorem for bipartite systems and a random approximate masking theorem for multipartite systems, and connects approximate masking to approximate quantum error correction.","keywords":["approximate quantum information masking","random isometries","no-masking theorem","multipartite entanglement","k-uniform states","approximate quantum error correction","concentration of measure","random subspaces"],"falsifier":"Sample Haar-random isometries from a $d_C$-dimensional logical space into a bipartite system and estimate the average subsystem variation $V^A(H_C)$; Theorem 1 predicts it stays above $w>1/9$ (approaching $1/6$ for large dimensions), and Theorem 2 predicts deviations below $w-\\alpha$ appear with probability at most $\\exp(-d_{12}\\alpha^2/16)$. Many trials with average variation clearly below $w$ would refute the no-random-AQIM claim.","tokens_in":1726,"feed_emoji":"🎲","tokens_out":5581,"duration_ms":106190,"temperature":0.7,"pith_summary":"This paper asks whether random isometries can approximately mask quantum information — hiding the identity of an input state in the correlations of a composite system. In a bipartite system the answer is negative: the average subsystem variation of a random subspace is bounded below by $w > 1/9$ (Theorem 1), and the probability of doing better than $w - \\alpha$ decays as $\\exp(-d_{12}\\alpha^2/16)$ (Theorem 2). In a multipartite system with $m$ equal-dimensional parties the answer reverses: a random subspace of dimension $d_C$ is an approximate $k$-uniform masker with inaccuracy $d^{k-m/2} + \\alpha$ except with exponentially small probability (Theorem 6). Because approximate masking is tied to approximate quantum error correction, the paper concludes that random subspaces can serve as approximate quantum error-correcting codes with constant code rate and exponentially small inaccuracy.","feed_headline":"Random isometries fail to mask in two parts, work in many","feed_subtitle":"Almost all random encodings fail at two-party masking yet succeed with many parties, giving constant-rate quantum codes.","key_machinery":"The workhorse is the average and maximum subsystem variation of a random subspace, together with concentration of measure on the Grassmannian. Proposition 2 decomposes $\\mathbb{E}_{H_C}[V^A_X(H_C)]$ as a dimension-dependent prefactor times the corresponding variation of the full bipartite space, and Lemma 3 shows the two one-party variations of a Haar-random bipartite state add to at least $1/3$; this carries the lower bound in Theorem 1. On the multipartite side, the key mechanism is the union bound over $\\binom{m}{k}$ bipartite cuts combined with concentration results (Theorems 4 and 5) stating that a random subspace's reduced states are close to the maximally mixed state or to the marginal of the subspace projector. The bridge to approximate quantum error correction is Lemma 1, which equates the maximum subsystem inaccuracy with the subsystem variance that controls the QEC inaccuracy for replacement channels.","core_discovery":"The central discovery is a pair of contrasting theorems about Haar-random isometries. In a bipartite target system, the average trace distance between the reduced states of two random image states cannot be pushed below a universal constant: $\\mathbb{E}_{H_C}[V^A(H_C)] \\ge w > 1/9$, with $w = \\frac{1}{6}\\frac{(2d_C-2)(2d_{12}-1)}{(2d_C-1)(2d_{12}-2)}$, and a deviation below $w-\\alpha$ occurs with probability at most $\\exp(-d_{12}\\alpha^2/16)$. Thus almost all random isometries fail to be even approximate maskers, which the authors cast as a no-random-AQIM theorem extending the original no-masking theorem. In multipartite systems with equal local dimension $d$, however, a random isometry is an approximate $k$-uniform masker almost surely: the probability that the maximum subsystem inaccuracy exceeds $d^{k-m/2}+\\alpha$ is exponentially small. The paper further shows that, since the maximum subsystem inaccuracy coincides with the subsystem variance used in approximate quantum error correction for replacement channels, random approximate maskers yield approximate quantum error-correcting codes whose code rate is constant and whose inaccuracy falls exponentially.","pith_inferences":["The bipartite/multipartite contrast suggests that the obstruction to random masking is fundamentally a two-party tradeoff: any encoding that hides information from every single party must spread it over at least three shares, because no random two-share split can keep both marginals flat.","Because the error-correction conclusion is proven only for replacement channels, a natural next test is whether the equivalence survives for general noise; if not, the AQECC claim would need a different recovery argument.","The concentration bounds used here are strong enough that unitary $k$-designs, rather than Haar-random unitaries, may suffice for the multipartite masking construction; comparing the two would give a practically implementable version of the result.","The connection between approximate masking, approximate $k$-uniform states, and approximate error correction suggests that masking is a generic property of highly entangled random subspaces, with consequences for thermalization and code capacity that the paper leaves open."],"forward_implications":["Random isometries cannot serve as approximate maskers in any bipartite system, no matter how large the local dimensions; this generalizes the no-masking theorem to the random setting.","In multipartite systems with enough parties, almost all random isometries are approximate $k$-uniform maskers with exponentially small failure probability.","Masking a logical space of $l$ qubits requires only a linear number $m^* \\propto l$ of physical qubits (or qudits), not exponential.","Random approximate maskers are simultaneously approximate quantum error-correcting codes with constant code rate and exponentially small correction inaccuracy, for replacement-channel noise on any set of at most $k$ parties.","The same concentration results imply that Haar-random multipartite pure states are approximate $k$-uniform states and have generalized Meyer-Wallach entanglement near 1 with high probability."],"supporting_citations":[{"why":"Supplies the no-masking theorem, the exact no-go result that approximate masking seeks to bypass.","marker":"[16]"},{"why":"Introduced probabilistic and approximate masking and gave a necessary condition; the starting point for AQIM.","marker":"[30]"},{"why":"Established the connection between k-uniform quantum information masking and quantum error-correcting codes with distance k+1.","marker":"[21]"},{"why":"Defined the subsystem variance and proved the inequality that bounds QEC inaccuracy for replacement channels; load-bearing for Corollary 4 and Theorem 8.","marker":"[51]"},{"why":"Supplies the generic entanglement and concentration results for random subspaces used in Theorems 4-6.","marker":"[40]"},{"why":"Provides the epsilon-net discretization technique and randomizing-state methods used in the concentration proofs.","marker":"[32]"},{"why":"Supplies concentration of measure for Lipschitz functions on Grassmannians and special unitary groups, used in Theorems 2 and 5.","marker":"[59]"}],"fun_headline_variants":["Random masking: no for pairs, yes for many","Random isometries fail bipartite, win multipartite","No random masking bipartite, but multipartite works","Random maskers: two fail, many succeed, constant-rate QEC","Random approximate masking: dual results, constant-rate codes"],"cache_read_input_tokens":41088,"weakest_assumption_plain":"The step that turns approximate masking into approximate quantum error correction relies on the noise being a replacement channel — one that discards a subsystem and replaces it with a fixed state — and the paper does not establish the masker-error-correction equivalence for general noise.","fun_headline_variants_meta":{"raw":{"variants":["Random masking: no for pairs, yes for many","Random isometries fail bipartite, win multipartite","No random masking bipartite, but multipartite works","Random maskers: two fail, many succeed, constant-rate QEC","Random approximate masking: dual results, constant-rate codes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000583,"raw_usage":{"total_tokens":2835,"prompt_tokens":1131,"completion_tokens":1704,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":747,"completion_tokens_details":{"reasoning_tokens":1621}},"tokens_in":747,"tokens_out":1704,"duration_ms":12591,"temperature":1.0,"reasoning_tokens":1621,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:52:33.554280+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Sample Haar-random isometries from a $d_C$-dimensional logical space into a bipartite system and estimate the average subsystem variation $V^A(H_C)$; Theorem 1 predicts it stays above $w>1/9$ (approaching $1/6$ for large dimensions), and Theorem 2 predicts deviations below $w-\\alpha$ appear with probability at most $\\exp(-d_{12}\\alpha^2/16)$. Many trials with average variation clearly below $w$ would refute the no-random-AQIM claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the no-masking theorem, the exact no-go result that approximate masking seeks to bypass."},{"cited_title":"Li, S.-h","cited_arxiv_id":null,"evidence_quote":"Introduced probabilistic and approximate masking and gave a necessary condition; the starting point for AQIM."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defined the subsystem variance and proved the inequality that bounds QEC inaccuracy for replacement channels; load-bearing for Corollary 4 and Theorem 8."},{"cited_title":"Nelson, G","cited_arxiv_id":null,"evidence_quote":"Supplies the generic entanglement and concentration results for random subspaces used in Theorems 4-6."},{"cited_title":"Harrow, P","cited_arxiv_id":null,"evidence_quote":"Provides the epsilon-net discretization technique and randomizing-state methods used in the concentration proofs."}],"review_version":2}