{"id":"47bbed00-5ee1-4cff-bf64-f01538a81322","arxiv_id":"2501.11688","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Two remote parties can efficiently estimate the overlap of any pair of real Clifford-transformed W-states using Bell sampling and Pauli measurements, even though these states resist classical MPS and low-magic learning.","lead":"This paper introduces a family of highly entangled quantum states, built by applying real Clifford circuits to the W state, and proves that two distant labs can estimate their overlap using only local measurements and classical communication. The result matters because it shows efficient cross-device verification is possible for states that are too complex for classical computers to describe.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 1's proof fails for odd n: Eq. (C6) claims F=0 for ε<4/n², but W-state Pauli expectations reach 1/n², so F(3/n²)>0.","rationale":"The reader's weakest_assumption identifies the realness requirement as the main concern. While realness is an acknowledged practical limitation, the most load-bearing technical flaw in the central claim is the incorrect CDF bound for odd n. This flaw directly undermines the proof of Proposition 1, the theorem that rDIPE is efficient on all CW sequences, and it also propagates to the robustness argument. It is an internal inconsistency, not a matter of consensus or physical applicability. The issue is concrete and checkable: for n=3 the claimed F=0 region is violated. It is repairable by choosing a smaller ε₂ for odd n, at the cost of modified constants, so the verdict remains conditional rather than reject. Other concerns, such as the unproven mixed-state extension of Eq. (4), affect robustness but do not touch the ideal-state efficiency claim; the failure of Proposition 3 for n=1,2 is a minor edge-case error. Thus the reader's verdict of CONDITIONAL is unchanged, but for a different and more fundamental reason.","tokens_in":12740,"tokens_out":34780,"duration_ms":322691,"concrete_test":"For n=3, compute F_{W_3}(1/3) exactly from the Pauli expectation values given in Appendix C: the six Paulis with z=1 or z=2 have squared expectation 1/9, so F(1/3)=6·(1/9)/8=1/12>0, contradicting Eq. (C6). Then determine whether replacing the threshold 4/n² by 1/n² for odd n and setting ε₂=1/n² still yields a valid polynomial bound for Proposition 1 with adjusted constants.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Proposition 1, the central efficiency theorem, relies on the claim in Appendix C that the Pauli CDF of every CW state vanishes for ε < 4/n² (Eq. C6). This is false for odd n. For the n-qubit W state, Pauli expectation values with z = (n±1)/2 Z-factors equal ±1/n, giving squared expectation 1/n². For n=3, for instance, there are 6 Paulis with squared expectation 1/9, so F_{W_3}(1/3) = 6·(1/9)/8 = 1/12 > 0, even though 1/3 < 4/9. Since the CDF is Clifford-invariant, this affects all states in CW(3). The proof of Proposition 1 sets ε₂ = min{(ε/8)², 3/n²} and asserts F(ε₂)=0 because 3/n² < 4/n²; for odd n this is wrong, so the inequality 4ε₁+4√ε₂+2F(ε₂) ≤ ε can fail and the sample-complexity bound is not established for such n. The same erroneous bound is used in the robustness proof (Appendix D, via Eq. C7), where ε₂ = 1/n² requires F(2/n²)=0, again false for odd n. The error is repairable — using ε₂ = 1/n² for odd n and adjusting constants, or handling finitely many small n by direct tomography — but as written the main theorem is unproven for all odd n. This is an internal inconsistency in the proof of the central claim, distinct from the acknowledged realness limitation.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the class CW(n) of states obtained by applying real Clifford unitaries to the n-qubit W state, and claims that for any two sequences in CW the rDIPE protocol of Ref. [18] estimates the normalized inner product tr(ρσ)/sqrt(trρ²trσ²) with polynomially many copies. The authors argue that CW contains states with linear half-system second Rényi entropy and states that need more than n/4 T gates to approximate, so the protocol goes beyond the usual learn-and-share regime. They also claim robustness to trace-distance noise and propose an experimental procedure to pre-estimate the required number of samples.","tokens_in":13145,"tokens_out":21846,"duration_ms":251713,"significance":"If the main claims are correct, the paper is a valuable conceptual advance: it gives an explicit family of states that are highly entangled and highly non-Clifford, yet for which distributed inner product estimation is efficient via Pauli sampling, and it provides a concrete near-term measurement schedule. The class CW is simple, the Clifford-invariance observation is clean, and the paper correctly identifies that the key quantity is the Pauli CDF gap. The direct calculations of Pauli expectations for W-like states and the average entanglement computation are useful. However, the central efficiency theorem has a parity-dependent gap in its CDF argument, and the mixed-state concentration bound needed for the robustness statement is imported without proof; both issues must be fixed before the claims can be regarded as established.","major_comments":[{"comment":"The claimed CDF gap F(ε)=0 for ε<4/n² is false for odd n. For the n-qubit W state with n odd, Pauli strings consisting only of Z factors with weight (n±1)/2 have expectation ±1/n and hence squared expectation 1/n², so F(1/n²)>0. Consequently the proof of Proposition 1, which sets ε₂=min{(ε/8)², 3/n²} and uses '3/n² < 4/n²' to conclude F(ε₂)=0, is invalid for odd n: the inequality 4ε₁+4√ε₂+2F(ε₂)≤ε is not established. The same erroneous gap is used in Eq. (C7), where Lemma 3 is applied with 2ε₂<4/n², and in Appendix D, where ε₂=1/n² requires F(2/n²)=0. As written, the main efficiency claim and the robustness theorem are unproven for all odd n. The error appears repairable by choosing ε₂<1/n² for odd n (or by treating finitely many small n separately), but the proof must be corrected.","section":"Appendix C, Eq. (C6); Proposition 1"},{"comment":"The paper relies on a mixed-state version of the rDIPE concentration bound, Eq. (4), but does not prove it. The sentence 'The proof of this statement applies with minor modifications also to our case of mixed states' is not a derivation, and the robustness theorem is stated precisely for mixed states with trρ²,trσ²>1/2. Since Eq. (9) is obtained by substituting into Eq. (4), the mixed-state extension is load-bearing. In addition, Algorithm 1 and Eq. (4) assume that the purities A and B are known exactly; the text in Appendix A acknowledges this but does not incorporate the estimation error into the concentration bound. Please provide a precise statement and proof of the mixed-state bound, with the purity-estimation error included or explicitly accounted for in the theorem hypotheses.","section":"Section II, Eq. (4); Appendix D"},{"comment":"The step 'Therefore, there exists at least one Pauli operator P with |⟨P⟩_ρ−⟨P⟩_ρ′|≥1/4' does not follow from the two displayed cardinality inequalities. The high-expectation set for ρ has size <2^{3n/4} and the set of ±1-expectation Paulis for ρ′ has size ≥2^{3n/4}, but this only shows that a Pauli with |⟨P⟩_ρ|>3/4 either lies in the ±1 set of ρ′ (which would give the desired gap) or lies outside it; in the latter case no bound on |⟨P⟩_ρ′| is provided. Additional structural information about t-doped states is needed to rule out the possibility that all large-expectation Paulis of ρ have intermediate expectation in ρ′. This proof gap affects the 'highly-doped' part of the paper's central claim and should be closed.","section":"Proposition 3 proof"}],"minor_comments":[{"comment":"The input specification reads 'computed by using usingN1 copies' and should read 'using N1 copies'.","section":"Algorithm 1, input line"},{"comment":"The sample pre-estimation procedure explicitly assumes that the prepared state is real, which is consistent with the paper's stated realness limitation, but this assumption should be restated in the theorem or proposition that summarizes Section V so that it is not read as applying to arbitrary noisy inputs.","section":"Section V"},{"comment":"Even after the parity fix, Eq. (C6) should state the gap separately for even and odd n: for even n the gap is 4/n², while for odd n it is 1/n². The current unified statement is misleading.","section":"Eq. (C6)"},{"comment":"The bound TV(q_{ρ′},q_ρ)≤∥ρ−ρ′∥_tr is quoted from the data-processing inequality; the derivation would be clearer if the factor 1/2 in the trace-distance convention were stated explicitly, since the final constant k=29 depends on the convention.","section":"Appendix D, Eq. (D1)"}],"recommendation":"major_revision","confidential_remarks":"The parity-dependent error in Appendix C is the most serious issue and, as written, invalidates Proposition 1 and the robustness theorem for all odd n. The authors should be given the opportunity to repair it, as the fix appears local. The mixed-state version of Eq. (4) also needs a self-contained proof or a precise reduction to Ref. [18]. If these are addressed, the paper could be suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper gives a new family of real states, CW(n) = rCl(n)|W_n>, and shows three things: rDIPE is efficient on them, they contain linearly entangled states, and they are hard to approximate by t-doped states for t ≤ n/4. The construction is clean and the hardness arguments are solid, especially the counting bound in Proposition 3. The average entanglement calculation via the real Clifford 2-design is also neat. This is a genuine advance in the 'verify without learning' direction.\n\nThe soft spot is in the CDF estimate that underpins Proposition 1. Appendix C claims F(ε)=0 for ε < 4/n² for all CW states. That is false for odd n: the W state has Pauli expectations ±1/n (from strings with z=(n±1)/2 Z factors), so the squared expectations reach 1/n², and F(3/n²) > 0 for n=3. The proof of Proposition 1 sets ϵ₂ = min{(ε/8)², 3/n²} and uses F(ϵ₂)=0. For odd n and moderately large ε this fails, so the stated sample-complexity bound is not proven as written. This is repairable: choose ϵ₂ = 1/(2n²) or handle the odd-n case directly, and the efficiency claim should survive. But the current version's main theorem is unproven for all odd n.\n\nThe other concerns are minor by comparison: the robustness section imports a mixed-state version of Eq. (4) without proof, and the estimator assumes exact knowledge of the purities A and B. The realness condition is honestly flagged as a limitation. No code or data is attached, which is fine for a theory paper.\n\nOverall, this is a worthwhile paper with a fixable proof gap. It deserves a serious referee, but the referee should require a corrected CDF bound and a proof of the mixed-state extension before acceptance.","headline":"Nice family of states and clean hardness proofs, but the CDF vanishing claim is wrong for odd n, leaving the main efficiency theorem unproven as written (though likely repairable).","tokens_in":13614,"tokens_out":4714,"would_cite":true,"duration_ms":46783,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P68"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"This paper introduces a class of highly entangled real quantum states and proves that two parties can efficiently estimate the inner product of any two such states via Bell sampling, without sharing classical descriptions.","keywords":["cross-device verification","distributed inner product estimation","Bell sampling","Pauli sampling","Clifford-rotated W-states","real Clifford group","t-doped stabilizer states","Rényi entanglement entropy"],"falsifier":"Randomly sample real Clifford unitaries C on n qubits, compute the half-system second Renyi entropy of C|W_n>, and average: if the average does not grow linearly in n, Proposition 2 is false. Likewise, an explicit search for an (n/4)-doped stabilizer state within trace distance 1/4 of a CW state would refute Proposition 3.","tokens_in":12568,"feed_emoji":"⚛️","tokens_out":9203,"duration_ms":90632,"temperature":0.7,"pith_summary":"The paper introduces a family of real quantum states—real Clifford rotations of the W-state—and proves that two parties holding copies of any states in this family can efficiently estimate their overlap using Bell sampling and classical communication, a task called distributed inner product estimation. The estimates require only polynomially many two-copy Bell measurements and single-copy Pauli measurements, and the sample count can be determined experimentally in advance. This matters because previous efficient schemes relied on learning a classical description of each state and sharing it, an approach that fails for states with high entanglement or many non-Clifford gates. The paper shows the W-state class contains states with linearly growing half-system entanglement and states that cannot be approximated by circuits with few T-gates, yet their overlap is still efficiently verifiable. The protocol is also robust to preparation noise up to a small trace-distance error.","feed_headline":"Bell sampling verifies entangled states classical tools cannot learn","feed_subtitle":"Remote parties estimate the overlap of Clifford-rotated W-states with only polynomial samples.","key_machinery":"The key object is the state class CW(n) = {C|W_n> : C ∈ rCl(n)}, where rCl(n) is the real Clifford group and |W_n> is the n-qubit W-state. The protocol is rDIPE, a Bell-sampling estimator of the normalized overlap c(rho,$\\sigma$) that uses N1 two-copy Bell measurements to sample from the mixture of Bell distributions and N1 N2 single-copy Pauli measurements to estimate the Pauli expectations at the sampled strings. The proof that rDIPE is efficient on CW rests on two facts: for real pure states the Pauli distribution p_rho equals the Bell distribution q_rho, making the total-variation term $\\Delta$ vanish; and the CDF of the Pauli expectation values of a W-state satisfies F_rho(eps)=0 for eps < 4/$n^{2}$, since its only nonzero expectations are 0, ±2/n, and 1-2z/n. These bounds turn the protocol's error estimate into a polynomially decaying exponential. The entanglement and non-approximability results use the real Clifford group being an orthogonal 2-design and a counting argument over Pauli strings with large expectation.","core_discovery":"The paper's central claim is that for every sequence of states taken from CW(n) = {C|W_n> : C ∈ rCl(n)}, the rDIPE protocol—the real-state version of the Pauli-sampling inner-product estimator—is efficient: for any accuracy epsilon and failure probability $e^{{-delta}}$, polynomially many samples suffice to estimate c(rho,$\\sigma$)=tr(rho $\\sigma$)/$\\sqrt$(tr $rho^{2}$ tr $sigma^{2}$) within epsilon. Efficiency relies on two features of CW. Because the states are real and pure, the Pauli distribution equals the Bell distribution, so the distribution-mismatch term $\\Delta$ vanishes exactly. And because the underlying W-state has Pauli expectation values only 0, ±2/n, or 1-2z/n, its CDF vanishes below 4/$n^{2}$, making the error bound polynomial. The same class contains, by an orthogonal-2-design averaging argument, states whose second Renyi entropy across a half-system cut grows linearly in n, and, by a stabilizer-counting argument, states that any t-doped stabilizer approximation must miss by trace distance at least 1/4 unless t > n/4.","pith_inferences":["If the realness condition can be relaxed by randomized compiling (which the paper explicitly does not rely on), the same Bell-sampling identity would apply to a wider class of near-real states; we would predict that the O(tau) robustness bound is tight in the imaginary-error direction, i.e., that small coherent Y-type errors are the dominant deviation.","The counting argument suggests a general trade-off: any family of states with a limited number of large Pauli expectation values will be hard for t-doped stabilizer approximation but easy for rDIPE, so searching for other such families (e.g., states from Haar-random circuits with real amplitudes) could extend the result.","Since the protocol reveals only sampled Pauli expectations, it provides a route to confidential verification—parties can certify they hold the same state without communicating enough information to reconstruct it; this is an implicit connection to quantum cryptography that the paper leaves as a future direction.","A testable extension: run the numerical simulation from Fig. 2 on the CW class itself for n up to ~30 and compare the empirical CDF with the analytical 4/n^2 threshold; a mismatch would indicate either an implementation error or a subtlety in the real-Clifford twirl assumption."],"forward_implications":["Cross-device verification no longer requires the states being compared to have efficient classical descriptions; highly entangled, highly doped states can be verified with polynomial local measurements.","The protocol is implementable in the near term, since each party only needs to prepare two copies of its state in a Bell measurement and perform single-copy Pauli measurements; the total sample complexity scales polynomially in n and 1/epsilon.","The required number of samples can be computed in advance from the parties' own measurement data using the empirical CDF and the Dvoretzky–Kiefer–Wolfowitz inequality, so the protocol is self-calibrating.","The protocol remains useful under realistic noise: states prepared within trace distance tau of a real CW state yield overlap estimates accurate up to O(tau), with polynomial resources.","The construction generalizes to Dicke states with constant excitation number and to superpositions of poly(n) computational basis states under real Cliffords, so the class of verifiable states is broader than the W-state itself."],"supporting_citations":[{"why":"Supplies the rDIPE/Bell-sampling estimator and the error bound (Eq. 4) that the paper adapts to mixed states.","marker":"[18]"},{"why":"Establishes the equality p_rho = q_rho for real pure states, which sets Delta = 0 in the protocol analysis.","marker":"[19]"},{"why":"Proves the real Clifford group is an orthogonal 2-design, used to average the entanglement entropy in Proposition 2.","marker":"[21]"},{"why":"Gives the entropy-scaling criterion that linear Rényi entropy prevents approximation by polynomial-bond-dimension MPS.","marker":"[5]"},{"why":"Provides the efficient learning algorithm for t-doped stabilizer states that motivates the comparison class in Proposition 3.","marker":"[6]"},{"why":"The Dvoretzky–Kiefer–Wolfowitz bound used in Section V to estimate required sample numbers from empirical CDFs.","marker":"[28]"}],"fun_headline_variants":["Bell sampling verifies states classical tools can't learn","Polynomial Bell sampling verifies entangled states beyond classical learning","Cross-device verification for states requiring many non-Clifford gates","Efficient Bell sampling wins where classical learning fails"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire efficiency argument assumes the states have real amplitudes, so that p_rho equals q_rho and the distribution-mismatch term $\\Delta$ vanishes.","fun_headline_variants_meta":{"raw":{"variants":["Bell sampling verifies states classical tools can't learn","Polynomial Bell sampling verifies entangled states beyond classical learning","Cross-device verification for states requiring many non-Clifford gates","Efficient Bell sampling wins where classical learning fails"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000301,"raw_usage":{"total_tokens":1769,"prompt_tokens":1015,"completion_tokens":754,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":631,"completion_tokens_details":{"reasoning_tokens":688}},"tokens_in":631,"tokens_out":754,"duration_ms":8668,"temperature":1.0,"reasoning_tokens":688,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:58:51.053989+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Randomly sample real Clifford unitaries C on n qubits, compute the half-system second Renyi entropy of C|W_n>, and average: if the average does not grow linearly in n, Proposition 2 is false. Likewise, an explicit search for an (n/4)-doped stabilizer state within trace distance 1/4 of a CW state would refute Proposition 3.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves the real Clifford group is an orthogonal 2-design, used to average the entanglement entropy in Proposition 2."},{"cited_title":"Chia, C.-Y","cited_arxiv_id":null,"evidence_quote":"Provides the efficient learning algorithm for t-doped stabilizer states that motivates the comparison class in Proposition 3."}],"review_version":1}