{"id":"663dace6-7fd8-4e13-a400-f216f0acc635","arxiv_id":"2512.06551","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The DPS entanglement hierarchy can be block diagonalized for CLDUI/LDUI/LDOI states, and the dual symmetric hierarchy is characterized by r-sos polynomials and linked to copositive programming.","lead":"This paper shows how to make the standard hierarchy for detecting quantum entanglement cheaper for a large family of diagonally invariant bipartite states, by block-diagonalizing its semidefinite programs. It also characterizes the dual hierarchy for Bose-symmetric states in terms of sums of squares and tests the speedups numerically.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.4 rests on a false polynomial identity as printed; the gDPS–K^{(t)} bridge is not proved until the Hadamard-square correction is verified.","rationale":"I read the paper as making two main contributions: block-diagonalizing the DPS hierarchy for CLDUI/LDUI/LDOI states, and connecting the symmetric dual hierarchy to the copositive hierarchy. The first contribution is supported by detailed structural lemmas, explicit clique tables, and numerical experiments; I did not find a comparably serious flaw there. The second contribution, however, hinges on Theorem 4.4, whose proof contains a concrete algebraic error: the displayed real-variable identity omits the Hadamard square in the quadratic form. The reader identified exactly this step, and my independent check agrees. The error is almost certainly typographical — the surrounding argument and the definition of K^{(t)} make the intended corrected identity clear — but the theorem as printed does not follow. Because the central claims of Section 4 are directly downstream of Theorem 4.4, the paper should not be accepted without the correction being verified. I do not see grounds to reject: the main Section 3 machinery appears internally consistent, the overlap with GNP25 is acknowledged, and the flaw is localized and repairable. Thus the appropriate posture remains conditional, unchanged from the reader's verdict.","tokens_in":53736,"tokens_out":20659,"duration_ms":172634,"concrete_test":"Independently re-derive the polynomial identity in Theorem 4.4 for generic A and, say, n=2, t=2. Expand \\|x\\|^{2(t-1)}\\sum A_{hk}|x_h|^2|x_k|^2 in x=x^{\\mathrm{Re}}+i x^{\\mathrm{Im}} and verify that it equals \\|w\\|^{2(t-1)}(w^{\\circ 2})^T(J_2\\otimes A)w^{\\circ 2}, not \\|w\\|^{2(t-1)}w^T(J_2\\otimes A)w. Then re-run the argument with the corrected identity and check that the r-sos condition is equivalent to J_2\\otimes A\\in K_{2n}^{(t-1)} and hence, via [GL07, Lemma 15], to A\\in K_n^{(t-1)}. If the corrected identity restores the proof, Theorem 4.4 is salvageable; if the equivalence fails, Theorem 4.5 is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is in §4.2, proof of Theorem 4.4. After defining p(x,\\bar x)=\\|x\\|^{2(t-1)}\\langle M^{T_B}_{S,T},xx^*\\otimes xx^*\\rangle, the proof passes to real variables w=(x^{\\mathrm{Re}},x^{\\mathrm{Im}}) and asserts\np_{\\mathrm{Re}}(w)=\\|w\\|^{2(t-1)} w^T (J_2\\otimes A) w.\nThe left-hand side is actually\n\\|x\\|^{2(t-1)}\\sum_{h,k} A_{hk}|x_h|^2|x_k|^2,\nwhich in real variables equals\n\\|w\\|^{2(t-1)} (w^{\\circ 2})^T (J_2\\otimes A) w^{\\circ 2},\nnot the printed quadratic form. The printed expression has degree 2t in w, whereas the actual polynomial has degree 2t+2; already for n=1, t=2 the two disagree. Because this identity is what links r-sos of p to the definition of K_{2n}^{(t-1)}, and because the reduction [GL07, Lemma 15] is applied to that (incorrect) quadratic form, the claimed equivalence M^{T_B}_{S,T}\\in(\\widetilde{\\mathrm{DPS}}^{(t)}_n)^* \\Leftrightarrow A\\in K_n^{(t-1)} is not proved as written. The error is repairable — the Hadamard square w^{\\circ 2} should appear — but until the corrected identity is checked and the [GL07] reduction re-run, Theorems 4.4–4.5 and the claimed agreement with GNP25 are unsupported. Secondary omissions (proofs of Theorem 3.9 and Lemma 3.10) increase verification burden but are not the crux.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Doherty-Parrilo-Spedalieri (DPS) hierarchy for separable bipartite quantum states under two structural assumptions: diagonal-unitary invariance (CLDUI/LDUI/LDOI states) and Bose symmetry. In the diagonal-unitary case it proves that the sparsity/block structure of the state is inherited by DPS certificates at every level, giving block-diagonal SDP formulations; it also develops a moment-formulation reduction and reports numerical experiments on PPT2-type benchmark instances. In the Bose-symmetric case it introduces a symmetry-adapted hierarchy gDPS(t), proves an r-sos characterization of its dual that mirrors Fang-Fawzi's theorem, and relates that dual to the copositive hierarchy K(t). The relation to the concurrent GNP25 work is explicitly acknowledged.","tokens_in":1396,"tokens_out":8258,"duration_ms":163353,"significance":"If completed, the paper would be a solid contribution: it offers substantial SDP-size reductions for classes of structured entangled states, an independent and more compact route to the gDPS-K(t) correspondence, and concrete numerical evidence. Strengths include the detailed proofs of the main structural results (Theorems 3.1 and 4.1), explicit block-size tables, and a public implementation. The overlap with GNP25 is handled transparently, and the authors are honest about the unresolved strictness question for gDPS(t). The central obstacle is one incorrect identity in the proof of Theorem 4.4; it is local and likely repairable, but it currently leaves Theorems 4.4-4.5 and the claimed equivalence with GNP25 unsupported.","major_comments":[{"comment":"The displayed real-variable identity in the proof of Theorem 4.4 is false as written. From (64), the real part of p is equal to the norm of w squared to the power (t-1) times the sum over h,k of A_hk times (w_h^2 + w_{n+h}^2)(w_k^2 + w_{n+k}^2). This equals the norm squared to the power (t-1) times the Hadamard square of w transposed times (J2 tensor A) times the Hadamard square of w, not the printed quadratic form using w itself. The left-hand side has degree 2t+2 in w, whereas the printed quadratic form has degree 2t; already for n=1, t=2 the two disagree. This identity is the step converting the r-sos of p into J2 tensor A in K_{2n}^{(t-1)} and then, via [GL07, Lemma 15], into A in K_n^{(t-1)}. Consequently Theorems 4.4 and 4.5 are not proved as printed. The correction is local and seems repairable, but the Hadamard-square identity and the subsequent reduction must be re-verified.","section":"Section 4.2, Eq. (64) and the following display"},{"comment":"The LDUI/LDOI transport theorem and the clique-size bound are stated without proofs ('The proof is similar, thus omitted'; 'We omit the proof'). These results underlie the main efficiency claims for LDUI and LDOI states, including Corollary 3.11 and Tables 5-6. The later Lemma 5.4 is proved in the moment framework and is not explicitly shown to imply Lemma 3.10. Please supply the missing proofs or give a fully explicit reduction to the CLDUI case; as written the reader cannot verify the LDUI/LDOI block decompositions and size bounds from the text.","section":"Section 3.3, Theorem 3.9 and Lemma 3.10"}],"minor_comments":[{"comment":"Spedaglieri appears to be a typo for Spedalieri (the author of [DPS02, DPS04]).","section":"Abstract and Introduction"},{"comment":"In the summand 'B + sum_{s=0}^{tau} W,' the last term appears to be missing its subscript: it should be W_s.","section":"Section 4.1, Theorem 4.1(iii)"},{"comment":"The external reduction [GL07, Lemma 15] is invoked without stating the lemma. Since it is load-bearing for the corrected proof, a statement would improve self-containedness.","section":"Section 4.2"},{"comment":"The notation m_k (k blocks of size m) is explained in the text but not in the table captions; please add it to the captions.","section":"Tables 4-6"},{"comment":"The proof is labelled a sketch and essentially follows from [GLS21, Section 5]. Since this lemma drives the moment-formulation implementation and the block-size tables, please provide a precise derivation or exact references.","section":"Section 5.1, Lemma 5.1"}],"recommendation":"major_revision","confidential_remarks":"I found no circularity or attribution problem: the overlap with GNP25 is disclosed and the proof routes are independent. The main issue is the false identity in the proof of Theorem 4.4. It is clearly repairable, but until the corrected identity and the [GL07] reduction are checked, the central gDPS-K(t) equivalence and Theorem 4.5 remain unproved. The omitted proofs for Theorem 3.9 and Lemma 3.10 should also be supplied, since they support the efficiency claims of Section 3."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a serious, solid paper with one genuine flaw in the printed proof. The block diagonalization of the DPS hierarchy for CLDUI/LDUI/LDOI states is new and practically useful; the r-sos characterization of the symmetric dual is a real extension of the known results. But the connection to the copositive hierarchy rests on an identity that is false as written.\n\nWhat's good: Theorem 3.1 plus the clique analysis (Lemmas 3.5–3.6) gives a clean, transferable sparsity reduction. The moment formulation in Section 5 makes the computational savings concrete, and the n=3 example in Section 3.4 is a nice analytic result. Theorem 4.1's characterization in terms of r-sos mirrors Fang–Fawzi and is the right way to see the symmetric dual. The authors are honest about the overlap with GNP25 and their proof route is independent.\n\nThe weak spot is exactly where the stress-test puts it. In the proof of Theorem 4.4, after writing x = x_Re + i x_Im, the displayed identity for p_Re is wrong. The left side is ||x||^{2(t-1)} times a sum over A_{hk}|x_h|^2|x_k|^2, which in real variables is ||w||^{2(t-1)} (w^{∘2})^T (J_2⊗A) w^{∘2}. The printed quadratic form w^T(J_2⊗A)w has the wrong degree: already for n=1, t=2 the two sides disagree. This identity is the bridge that identifies r-sos of p with A∈K_{2n}^{(t-1)}, so Theorem 4.4 and its corollaries are not proved as printed. It looks like a missing Hadamard-square, and I expect it's repairable, but it is load-bearing; a referee should ask for the correction and the [GL07] reduction to be re-run. Minor additions: Theorem 3.9 and Lemma 3.10 have proofs deferred or omitted, which raises the verification burden but is not central.\n\nWho gets value: anyone working on symmetric quantum states, semidefinite hierarchies, or copositive programming. It's a within-subfield advance, not a paradigm shift.\n\nRecommendation: send it to peer review. The main structural work deserves referee time, but the version currently posted should not be accepted without the Theorem 4.4 fix. I'd cite the block-diagonalization part; I'd wait on the copositive bridge until the corrected proof confirms.","headline":"Block diagonalization of DPS for diagonal unitary invariant states is new and valuable, but the proof of Theorem 4.4 has a false displayed identity that must be fixed before the copositive hierarchy connection is credible.","tokens_in":54644,"tokens_out":2673,"would_cite":true,"duration_ms":24560,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C22","81P40","15B48"],"pacs":[],"model":"deepseek-v4-flash","headline":"For bipartite states invariant under diagonal unitaries, every level of the DPS hierarchy can be block-diagonalized, making high-level entanglement tests tractable; in the Bose-symmetric case the dual hierarchy is exactly characterized by r","keywords":["quantum entanglement","DPS hierarchy","separable states","diagonal unitary invariance","Bose symmetry","copositive cone","sums of squares","block diagonalization"],"falsifier":"Evaluate the displayed identity in the proof of Theorem 4.4 at t=2, n=2, A=I_2, x=(1+i,1): the left side has degree 6 in x,xbar while the right side as printed has degree 2 in xRe,xIm, so the line cannot be true; replacing xRe,xIm by their Hadamard squares gives the intended degree-6 identity. Then check the equivalence with K^{(t-1)} using Theorem 4.1(ii) for a matrix A known to lie in K^{(1)}_5 setminus K^{(0)}_5.","tokens_in":53577,"feed_emoji":"⚛️","tokens_out":8904,"duration_ms":88418,"temperature":0.7,"pith_summary":"This paper shows that structural symmetry of a bipartite quantum state can be pushed into every level of the DPS hierarchy used to distinguish separable from entangled states. For states with diagonal-unitary invariance (the CLDUI, LDUI, and LDOI classes), the matrices in the semidefinite relaxations split into small independent blocks, so higher relaxation levels become computable in seconds instead of minutes. For Bose-symmetric states, the paper gives a dual characterization of the symmetry-adapted hierarchy: the associated complex polynomial must be a real sum of squares, exactly parallel to the known generic case. It then proves an equivalence between this symmetric hierarchy and the standard sum-of-squares approximations of the copositive cone, so entanglement detection for these states is tied to a well-studied optimization object. A concrete family of CLDUI states is shown to be separable exactly when it passes level two of the hierarchy, illustrating that the reductions do not lose detecting power.","feed_headline":"Symmetry shrinks entanglement checks to tiny blocks","feed_subtitle":"Exploiting symmetry in quantum states lets separability tests reach higher levels in seconds instead of minutes.","key_machinery":"The main engine is the sparsity pattern of CLDUI/LDOI states: their nonzero entries are indexed by equality of multisets or parity conditions, and the same patterns define the support of extended DPS certificates. These support graphs are disjoint unions of cliques, so positive semidefiniteness and partial-transpose conditions factor into independent blocks. The moment reformulation indexes certificates by monomial degrees rather than register sequences, eliminating the remaining duplicate rows. On the dual side, the bridge is the identity expressing the real part of the polynomial ||x||^{2(t-1)} <M, xx*⊗xx*> as a weight times a quadratic form in the squared real and imaginary parts, which i","core_discovery":"The central claim is that the DPS relaxation at any order t can be reduced without loss for diagonal-unitary-invariant states. The support of an extended certificate in CLDUI(t), LDUI(t), or LDOI(t) is contained in a graph that is a disjoint union of cliques, and this holds also after partial transposes; therefore the certificate is block diagonal, with blocks of size at most t! n^{ceil(t/2)} in the tensor formulation. Rephrasing the hierarchy in moment form removes the remaining replicated rows, shrinking the largest block further; the paper reports a ratio bound of at most t! / n^{ceil(t/4)} relative to the generic case. For Bose-symmetric states, the dual of the symmetric hierarchy gDPS^{","pith_inferences":["The reported runtimes suggest the block-size advantage grows with the hierarchy level, so the highest relaxation levels benefit most from the sparsity reduction; the paper's tables show this but it is not stated as a theorem.","The Bose-symmetric equivalence implies a transfer of hardness: matrices that escape the sum-of-squares approximations of the copositive cone should yield Bose-symmetric entangled states that escape the symmetric DPS hierarchy for many levels.","The cliqued-support mechanism is not tied to unitary invariance specifically; analogous block-diagonal hierarchies should arise for other diagonal symmetry groups, such as orthogonal sign symmetries, following the same projection-and-clique argument."],"forward_implications":["For any CLDUI, LDUI, or LDOI bipartite state, testing membership in DPS level t reduces to semidefinite programs whose largest block is at most t! n^{ceil(t/2)}; in the moment formulation the largest block is at most t! n^{ceil(t/4)} times as large as in the generic case.","The block-diagonal implementation makes levels 4–7 practical for small n where the generic formulation stalls at level 2–3, so known separability criteria become testable at much higher strength.","For Bose-symmetric states rho^TB_{(X,X)}, the symmetry-adapted hierarchy is equivalent to the copositive approximation hierarchy (K^{(t)})^*; consequently separability of these states can be certified by checking X against (K^{(t-1)})^*.","For the family rho_{a,a'} defined in the paper, DPS level two is exact: the state is separable exactly when a, a' are at least 1, even though level one only certifies aa' >= 1.","The symmetric hierarchy converges to the Bose-symmetric separable cone, and any strict inclusion in the copositive hierarchy yields explicit states that escape the symmetric hierarchy."],"fun_headline_variants":["Symmetry shrinks quantum separability tests","Block diagonalization speeds up entanglement checks","Symmetry reduces semidefinite hierarchy size","Quantum state tests made faster by symmetry"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"Everything rests on the claim that projecting an arbitrary DPS certificate onto the CLDUI/LDOI/LDUI subspace preserves the certificate conditions (positive semidefiniteness after every partial transpose and correct partial trace); for the copositive equivalence, Theorem 4.4 additionally depends on a polynomial identity whose printed form omits Hadamard squaring of the real and imaginary parts and is false as written, so the corrected identity is load-bearing.","fun_headline_variants_meta":{"raw":{"variants":["Symmetry shrinks quantum separability tests","Block diagonalization speeds up entanglement checks","Symmetry reduces semidefinite hierarchy size","Quantum state tests made faster by symmetry"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000155,"raw_usage":{"total_tokens":1090,"prompt_tokens":824,"completion_tokens":266,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":213}},"tokens_in":568,"tokens_out":266,"duration_ms":3472,"temperature":1.0,"reasoning_tokens":213,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:08:02.362815+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the displayed identity in the proof of Theorem 4.4 at t=2, n=2, A=I_2, x=(1+i,1): the left side has degree 6 in x,xbar while the right side as printed has degree 2 in xRe,xIm, so the line cannot be true; replacing xRe,xIm by their Hadamard squares gives the intended degree-6 identity. Then check the equivalence with K^{(t-1)} using Theorem 4.1(ii) for a matrix A known to lie in K^{(1)}_5 setminus K^{(0)}_5.","supporting_citations":[],"review_version":1}