{"id":"682274d8-d5a0-4d67-9c22-6cca619fc5a7","arxiv_id":"2509.05214","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For a class of graph states with commuting controlled-rotation gates, the Fubini-Study entanglement distance depends only on the vertex degree sequence, and this paper works out explicit formulas for four example graph families.","lead":"This paper computes a geometric entanglement measure for quantum states built from directed graphs, showing that only the number of edges per node matters, not their direction. It derives explicit formulas for four example network families, including layered, tree, and bridged-cycle shapes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (13)/(A5) fail for directed graphs with 2-cycles: the two-vertex bidirectional graph gives E=1−cos²2θ, not 1−cos⁴θ, because Γ→ and Γ← overlap.","rationale":"The reader's weakest assumption correctly identified the reliance on identical product states and commuting diagonal controlled rotations, but did not identify the more specific and more damaging hidden condition that Γ→(i) and Γ←(i) be disjoint. The Appendix A derivation assumes independent outgoing and incoming neighbor sets; when a 2-cycle is present, the same qubit is traced over twice and the binomial factorization fails. A direct calculation for the minimal 2-cycle graph contradicts Eq. (13). This is a genuine correctness risk for the stated scope, since the paper never excludes 2-cycles and the abstract claims full independence of edge orientation. The fix is straightforward: either restrict the theorem to oriented graphs (no 2-cycles) or derive a separate formula for overlapping neighbor sets. Because the applications all appear to use acyclic/oriented layered graphs, the paper's worked examples may still be correct, but the central claim as stated is not. This preserves the conditional recommendation with a new, concrete condition.","tokens_in":10059,"tokens_out":41681,"duration_ms":415310,"concrete_test":"Compute the per-qubit ED of the two-qubit state U12U21|+>|+> with Uab of Eq. (5) using Eq. (4), and compare with Eq. (13). If E=1−cos²2θ rather than 1−cos⁴θ, the theorem requires an explicit no-2-cycle hypothesis. Repeat for a 3-vertex graph with edges (1,2), (2,1), (2,3) to confirm the discrepancy persists beyond the minimal case.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central degree-only formula (13) and its general-α extension (A5) are derived in Appendix A by treating the d→ outgoing neighbors and d← incoming neighbors as independent sets of qubits. This factorization requires Γ→(i) ∩ Γ←(i) = ∅ for every vertex i. The paper never states this condition: L is defined as any set of ordered pairs, and standard simple-direct-graph terminology permits both (a,b) and (b,a). When the same neighbor appears in both sets, the binomial partial-trace argument double-counts that neighbor. Explicit counterexample: the two-vertex graph with both arcs. For p=1/2, ψ=0, U12U21|+>|+> = 1/2(|00>+e^{iθ}|01>+e^{iθ}|10>+e^{-2iθ}|11>). The reduced single-qubit state has off-diagonal ρ01 = (1/2)cos2θ e^{iθ}, so ||⟨σ⟩||² = cos²2θ and E = 1−cos²2θ. Eq. (13) with d(1)=d(2)=2 gives E = 1−cos⁴θ. At θ=π/4 the two values are 1 and 0.75. For general p the A5 prediction 4p(1−p)r⁴ disagrees with the exact 4p(1−p)|(1−p)e^{−iθ}+p e^{3iθ}|². Thus the orientation-independence claim is false for 2-cycles unless the graph is explicitly restricted to an oriented (2-cycle-free) digraph.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies entanglement in pure multipartite states associated with directed graphs. The states are generated by products of commuting controlled rotations U_ab = Pi_0^a I^b + Pi_1^a Ubar^b with diagonal Ubar, acting on identical single-qubit product inputs. The central claim is that the Entanglement Distance per qubit (Eq. 4) depends only on the total degree sequence: E(theta;{d(i)}) = 1 - (1/M) sum_i [cos theta]^(2 d(i)) (Eq. 13), independent of edge orientation. Appendix A extends this to arbitrary identical single-qubit input states, obtaining E^(i) = 1 - (1-2p)^2 - 4p(1-p) r^(2d(i)) with r = sqrt(cos^2 theta + sin^2 theta (1-2p)^2) (Eq. A5). The paper applies these formulas to four graph families: a Young-Fibonacci-like graph, deep feed-forward networks, full binary trees, and linear bridged cycle graphs.","tokens_in":10481,"tokens_out":11637,"duration_ms":117901,"significance":"If the degree-only characterization were valid for general directed graph states, it would be a strong and useful result: multipartite entanglement in this class would be fully determined by the degree distribution, and the explicit closed forms for nontrivial graph families would be valuable for quantum-network design. The appendix derivation is algebraically transparent and, under the correct graph restriction, reproduces Eq. (13); the application formulas are consistent with the stated degree distributions (apart from a typesetting issue in Eq. (21)). However, the central claim as stated is not valid for arbitrary directed graphs, because the derivation double-counts vertices that are both incoming and outgoing neighbors. This is a load-bearing defect, but it can be repaired by explicitly restricting to oriented graphs, and the paper's own examples already satisfy that restriction.","major_comments":[{"comment":"The degree-only formula requires Gamma->(i) ∩ Gamma<-(i) = ∅ for every vertex i, but this disjointness is never stated. The definition of L in §II permits both (a,b) and (b,a), and condition i) does not exclude this because U_ab and U_ba commute. When a neighbor is both incoming and outgoing, the binomial partial-trace argument in Appendix A counts the same qubit twice. Concrete counterexample: M=2, L={(1,2),(2,1)}, p=1/2, ψ=0. Then U_12 U_21 |++> = (1/2)(|00> + e^{iθ}|01> + e^{iθ}|10> + e^{-2iθ}|11>). The reduced single-qubit state has off-diagonal element ρ_01 = (1/2)e^{iθ} cos 2θ, so ||<σ>||^2 = cos^2 2θ and E = 1 − cos^2 2θ. Equation (13) with d(1)=d(2)=2 gives E = 1 − cos^4 θ; at θ=π/4 the two values are 1 and 3/4. Thus the orientation-independence and degree-only claims are false for directed graphs containing 2-cycles. Please restrict the theorem to oriented graphs (no opposite ar","section":"§II, §III.A, Appendix A (Eqs. A1–A5)"}],"minor_comments":[{"comment":"The full binary tree expression appears to have lost superscripts: the factors written as 2N−1 should be 2^{N−1}, and the last coefficient should be 2^{N−1}−2. As typeset, the formula is inconsistent with the degree distribution N(d)=2^{N−1}δ_{d,1}+δ_{d,2}+2(2^{N−2}−1)δ_{d,3}.","section":"§IV.C, Eq. (21)"},{"comment":"The sentence 'The ED equals M if |G> is maximally entangled' conflicts with the per-qubit definition in Eq. (4), whose maximum is 1. Either define a total ED that sums over qubits or change the sentence to refer to the per-qubit value.","section":"§III.A, Eq. (4)"},{"comment":"Please clarify the graph terminology: exclude loops explicitly and state whether opposite arcs are allowed. In the standard 'simple directed graph' convention both (a,b) and (b,a) may coexist; the theorem requires the stricter 'oriented graph' convention.","section":"§II"},{"comment":"The caption says 'recurrent neural network' but the section describes a deep feed-forward neural network; the caption should match the text.","section":"Figure 5 caption"},{"comment":"Minor language issues: 'commutate' should be 'commute'; 'direct graph' in a few places should be 'directed graph'.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The main result is essentially inherited from the authors' own Ref. [18], with the new contributions being the general-α extension in Appendix A and the application examples. The 2-cycle counterexample is not a matter of interpretation: Eq. (13) is plainly false under the stated definition of L unless the graph is oriented. Since the applications are already oriented, the authors can likely fix this with a clear restriction and a corrected statement; this is why I recommend major revision rather than rejection. Please ensure the revised version redefines the class of graphs consistently."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know upfront: the central formula, Eq. (13), is not valid for all directed simple graphs as claimed. The derivation in Appendix A implicitly assumes that no vertex appears in both the outgoing and incoming neighbor sets of another vertex—i.e., no directed 2-cycles. The paper defines a directed simple graph as any set of ordered pairs, which allows both (a,b) and (b,a). For the two-vertex graph with both arcs, a direct calculation gives E = 1 − cos²2θ, while Eq. (13) with the paper's own degree definition (number of distinct neighbors, d=1) gives 1 − cos²θ. So the orientation-independence claim fails. The stress-test note gets the counterexample right, though it misstates the degree as 2; the paper defines degree as the union of in- and out-neighbors, so d=1. Either way, the formula is wrong for 2-cycles.\n\nThe good news is that the paper has real content once you add the missing hypothesis. For oriented graphs (no antiparallel edges), the Appendix A derivation is algebraically sound and gives a clean extension to arbitrary identical single-qubit input states. The four example families—Young-Fibonacci, feed-forward neural network, full binary tree, and linear bridged cycle—are all oriented, so the closed-form ED expressions are correct apart from a likely typo in Eq. (21) (the coefficient of cos⁶θ should be 2^{N-1}−2, not 2^N−3). The examples are worked carefully, and the degree-distribution method is clearly explained. The paper is a straightforward extension of the authors' own prior work, but it does add the general-α formula and the four specific families, which are not in the cited references.\n\nMinor issues: the text says \"The ED equals M\" for maximally entangled states, but Eq. (4) is per qubit and maxes at 1; that's a typo. The paper also leans heavily on Ref. [18] without crisply stating what is new here, though the appendix mitigates that.\n\nBottom line: the paper deserves a serious referee, but the referee should require the oriented-graph condition to be stated explicitly and the theorem re-proven under that condition. The counterexample is not fatal to the oriented version, which is still useful for the graph families considered. I would not cite it in its current form, but I'd recommend sending it to peer review with the expectation of a revision.","headline":"The paper's degree-only entanglement formula is only valid for oriented graphs (no 2-cycles), and the paper never says that; the two-vertex bidirectional graph is a clean counterexample.","tokens_in":10917,"tokens_out":9884,"would_cite":false,"duration_ms":101986,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.Mn"],"model":"deepseek-v4-flash","headline":"For a large class of directed graph states, entanglement per qubit is fixed by the degree sequence alone.","keywords":["entanglement distance","directed graph states","degree distribution","Fubini–Study metric","multipartite entanglement","quantum networks","graph topology","controlled-rotation gates"],"falsifier":"Take two three-vertex directed graphs with the same total degree sequence (2,2,2) but different orientations: a directed cycle 1→2→3→1 and a directed acyclic graph 1→2, 1→3, 2→3. Prepare both graph states with |φ⟩ = (|0⟩+|1⟩)/√2 and θ = π/3, measure the per-qubit expectation values ||⟨σ⁽ⁱ⁾⟩||, and compute the ED. Equation (13) predicts identical ED for both; any observed difference falsifies the degree-only claim.","tokens_in":9980,"feed_emoji":"🔗","tokens_out":5806,"duration_ms":57837,"temperature":0.7,"pith_summary":"Graph states are many-qubit quantum states built by applying a fixed entangling gate along each edge of a graph. This paper studies directed graphs, where edges have a chosen orientation, and asks how much multipartite entanglement the resulting state carries, as measured by the Entanglement Distance derived from the Fubini–Study metric. The central finding is that the per-qubit entanglement depends only on the total number of edges touching each vertex, the degree d(i), through E(θ; {d(i)}) = 1 − (1/M) Σᵢ [cos θ]^(2d(i)). Edge orientation and vertex numbering drop out entirely, so the measure is a topological invariant of the graph. A derivation in Appendix A extends this to arbitrary identical single-qubit input states, again yielding a degree-only formula. This matters because it turns a complex combinatorial entanglement calculation into a function of one local statistic, offering a practical handle for designing quantum networks whose entanglement is controlled by topology.","feed_headline":"One formula ties graph-state entanglement to link counts","feed_subtitle":"Edge direction does not matter: a directed graph's entanglement reduces to its degree sequence.","key_machinery":"The carrier of the argument is the Entanglement Distance (ED), E = 1 − (1/M) Σᵢ ||⟨G|σ⁽ⁱ⁾|G⟩||², a Fubini–Study-derived local-unitary invariant for pure multipartite states, applied to graph states built from a single commuting entangling gate U_ab = Π₀ᵃIᵇ + Π₁ᵃŪᵇ with Ū diagonal (Eq. 5). The calculation hinges on phase cancellations in the expectation values of σₓ and σᵧ at each vertex; because the gate is diagonal and identical across edges, the net effect of all incident links collapses to a power cos^(2d(i)) θ, erasing any distinction between incoming and outgoing edges.","core_discovery":"The paper's claim, stated on its own terms: for directed graph states generated by a commuting controlled-rotation gate (Eq. 2 with Ū diagonal, Eq. 5) from an identical pure product input, the Entanglement Distance per qubit is exactly E(θ; {d(i)}) = 1 − (1/M) Σᵢ∈V [cos θ]^(2d(i)) (Eq. 13). Each vertex contributes a power of cos²θ equal to its total degree; incoming and outgoing links contribute identically, and vertex relabeling changes nothing. Appendix A proves the same degree-only dependence for a generic identical input state |φ⟩ = α₀|0⟩ + α₁|1⟩, giving E⁽ⁱ⁾ = 1 − (1−2p)² − 4p(1−p) r^(2d(i)) with r = √(cos²θ + sin²θ (1−2p)²). The maximal-entanglement choice p = 1/2, θ = π/2 recovers Eq.","pith_inferences":["If the degree-only formula holds across this gate class, then ED cannot distinguish two graph states that share a degree sequence but differ in connectivity; a natural next test is whether other entanglement measures also collapse to degree statistics on this gate class.","A direct experimental falsifier: prepare a three-vertex directed cycle and a three-vertex directed acyclic graph with the same total degree sequence (2,2,2) but different edge orientations, using |φ⟩ = (|0⟩+|1⟩)/√2 and θ = π/3; Eq. (13) predicts identical ED, and any measured difference would refute the claim.","The dependence on identical input states suggests a design principle for quantum networks: tune the global input state to maximize the r^(2d(i)) response, concentrating entanglement on high-degree hubs; the authors hint at this for dynamic topology learning in quantum machine learning.","Weighted or non-commuting gates would likely bring back orientation dependence, so the degree-only collapse should be understood as a property of this specific commuting controlled-rotation gate, not of directed graph states in general."],"forward_implications":["The ED of any state in this gate class can be computed from the degree sequence alone, bypassing the full adjacency matrix.","Graphs with the same degree distribution have identical per-qubit entanglement, even if their edge orientations or vertex labels differ.","For layered graphs (variant of the Young–Fibonacci graph and the full binary tree), the infinite-size limit yields closed asymptotic bounds such as 1 − cos⁸θ and 1 − (cos²θ/2)(1 + cos⁴θ).","The linear bridged cycle formula E(θ; M, N) = 1 − (cos⁴θ/M)(M − 2(N−1) sin²θ) shows explicitly how entanglement changes with the number of bridges and cycle sizes.","Any identical single-qubit input state preserves the degree-only character; the input state only rescales the radial factor r in the per-vertex contribution."],"supporting_citations":[{"why":"Companion paper (arXiv:2505.10716) where the degree-only ED formula (Eq. 13) is derived for the |+⟩ input state; the present work extends that result.","marker":"[18]"},{"why":"Introduces the Entanglement Distance measure used throughout the paper.","marker":"[3]"},{"why":"Extends ED to multipartite mixed states, supporting the measure's status as a general entanglement quantifier.","marker":"[4]"},{"why":"Provides the Fubini–Study geometric interpretation of the entanglement metric underlying the ED.","marker":"[9]"},{"why":"Defines the stabilizer graph-state framework and multiparty entanglement in graph states, the formal setting for this class.","marker":"[11]"}],"fun_headline_variants":["Graph entanglement: edge direction irrelevant, degree counts all","Entanglement in directed graphs depends only on vertex degrees","For graph states, entanglement ignores edge direction, uses degrees","Directed graph entanglement: total degree per vertex is the key","No direction needed: graph entanglement tied to degree sequence"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The result assumes every edge applies the same commuting diagonal controlled-rotation gate and every qubit starts in the identical pure product state; if either condition fails, the phase cancellations that erase edge orientation no longer occur, and entanglement can depend on edge direction.","fun_headline_variants_meta":{"raw":{"variants":["Graph entanglement: edge direction irrelevant, degree counts all","Entanglement in directed graphs depends only on vertex degrees","For graph states, entanglement ignores edge direction, uses degrees","Directed graph entanglement: total degree per vertex is the key","No direction needed: graph entanglement tied to degree sequence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000167,"raw_usage":{"total_tokens":1099,"prompt_tokens":757,"completion_tokens":342,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":501,"completion_tokens_details":{"reasoning_tokens":264}},"tokens_in":501,"tokens_out":342,"duration_ms":4701,"temperature":1.0,"reasoning_tokens":264,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:30:53.764848+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two three-vertex directed graphs with the same total degree sequence (2,2,2) but different orientations: a directed cycle 1→2→3→1 and a directed acyclic graph 1→2, 1→3, 2→3. Prepare both graph states with |φ⟩ = (|0⟩+|1⟩)/√2 and θ = π/3, measure the per-qubit expectation values ||⟨σ⁽ⁱ⁾⟩||, and compute the ED. Equation (13) predicts identical ED for both; any observed difference falsifies the degree-only claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Companion paper (arXiv:2505.10716) where the degree-only ED formula (Eq. 13) is derived for the |+⟩ input state; the present work extends that result."},{"cited_title":"G¨ uhne and G","cited_arxiv_id":null,"evidence_quote":"Introduces the Entanglement Distance measure used throughout the paper."},{"cited_title":"Vesperini, G","cited_arxiv_id":null,"evidence_quote":"Provides the Fubini–Study geometric interpretation of the entanglement metric underlying the ED."},{"cited_title":"Vesperini, G","cited_arxiv_id":null,"evidence_quote":"Defines the stabilizer graph-state framework and multiparty entanglement in graph states, the formal setting for this class."}],"review_version":1}