{"id":"6237214f-0756-4aee-a5ea-e1ffa1c4c97a","arxiv_id":"2505.23256","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For one-layer QAOA on complete-graph MAXCUT, the ratio of summed entanglement entropy with and without tensor-network truncation collapses onto a universal curve, and an analytic approximation is proposed.","lead":"This paper studies how entanglement, measured with tensor networks, changes in a quantum optimization algorithm called QAOA. It finds a simple curve that describes the truncation error caused by approximating the quantum state, and proposes a mathematical function for that curve.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (27) is derived by replacing the measured first/second and truncation-normalization ratios with the idealized values 1:2 and 1; Tables I–III show those ratios are not those values, so the proposed scaling function is not validated by the paper's own data.","rationale":"The paper has two distinct claims: (i) the empirical data collapse of P Sχ/P S onto a single curve in Fig. 2, and (ii) the analytic scaling function Eq. (27). Claim (i) is plausible and honestly presented with 100-sample averaging, but the paper does not release code, data, or per-instance error bars, so it cannot be independently verified from the text. Claim (ii) is the load-bearing new result, and it is not supported by the paper's own Tables. The derivation in Section IV combines Eq. (18), Eq. (20), and Eq. (22) into Eqs. (23)–(27). Each of these is an empirical assertion taken from the same simulations used to test the formula, and each deviates from the assumed value by an amount that matters: Table I is 2–3σ from 0.5; Table II ranges from 0.436 to 0.550; Table III ranges from 0.844 to 1.031. Since the final function is obtained by substituting exactly 1/2 and 1, the function is essentially an interpolation of processed summary statistics, not a derivation. The paper's own Fig. 5 shows that Eq. (27) misses the numerical curve, and the listed possible reasons include the assumptions themselves. Thus the strongest claim 'Eq. (27) as an analytic approximation' is conditional at best; the safe verdict is to require either an independent derivation or an honest reframing as a heuristic fit. This matches the reader's weakest-assumption analysis, so no change to the reader's CONDITIONAL verdict is needed. A useful next step is to recompute with the actual measured ratios, or to generate an out-of-sample N=14 point, to distinguish a repairable calibration issue from a genuinely invalid analytic form.","tokens_in":13624,"tokens_out":9171,"duration_ms":104747,"concrete_test":"Recompute Eq. (27) without imposing the idealized ratios: use the per-case means and standard errors from Tables I–III directly in Eqs. (25)–(26), propagate the uncertainties, and compare the corrected curve to the raw 100-sample averages underlying Fig. 2 over the full range 0 < 2 log2 χ/N < 1. If the corrected curve does not lie within the scatter of P Sχ/P S, the analytic function is not a valid approximation of the data; if it does, the original Eq. (27) still fails because it uses the wrong constants, and the authors must refit or justify them.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central analytic claim, Eq. (27), rests on three empirical proportionality assumptions. Eq. (18) sets (Σ_space S)_first : (Σ_space S)_second ∼ 1:2, but Table I reports 0.4166±0.0395, 0.4296±0.0262, and 0.4361±0.0223 for N=8,10,12, i.e., 2–3σ away from 0.5. Eq. (20) sets the same ratio in the truncated circuit, but Table II gives 0.436–0.550. Eq. (22) sets the normalized second-half sums equal with and without truncation, but Table III gives 0.844–1.031, with several entries more than 1σ from 1. Because the derivation divides out α and then builds the x-dependence of Eq. (27) precisely from these assumed 1/2 and 1 ratios, any violation changes the functional form. The paper's own Fig. 5 shows a visible mismatch between Eq. (27) and the numerical curve, and the discussion in Section IV attributes this to 'errors in the relationships' and the rough N^2 replacements; but the identified alternatives are themselves part of the central derivation. The empirical collapse of Fig. 2 may still be a real scaling relation, but Eq. (27) is not an independently derived or quantitatively accurate scaling function for the presented data; it is a heuristic curve whose three calibration constants are assumed, not obtained from a fit or from a first-principles argument.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates entanglement entropy (EE) in one-layer QAOA on complete-graph MAXCUT simulated with matrix product states (MPS). The main empirical claim, Eq. (16), is that the ratio of the summed entanglement entropy over all bonds and time steps with truncation to that without truncation, P Sχ/P S, collapses onto a single curve as a function of 2 log2 χ/N. The paper further proposes Eq. (27) as an analytic approximation to this curve, derived from the assumed ratio relations Eqs. (18), (20), and (22) and from the theoretical maxima in Eqs. (17) and (19). Numerical results are shown in Figs. 2 and 5; the latter displays a visible discrepancy between Eq. (27) and the numerical data.","tokens_in":14023,"tokens_out":7673,"duration_ms":78549,"significance":"The empirical scaling collapse, if it persists under a proper statistical test, is a noteworthy extension of the scaling relations found in [19,21], and the use of entanglement entropy as the vertical axis makes the quantity amenable to further theoretical study. The observation in Fig. 7 that the spatial sum of EE grows approximately linearly and then saturates is also interesting. However, the paper's analytic scaling function is not validated: Eq. (27) is calibrated to the same data from which Tables I–III are drawn, and the mismatches shown in Fig. 5 are acknowledged by the authors. The contribution is therefore primarily an empirical scaling law; the theoretical claim needs substantial revision.","major_comments":[{"comment":"Eqs. (18), (20), and (22) are load-bearing inputs to the derivation of Eq. (27), but they are inferred from the same numerical data to which Eq. (27) is compared. Eq. (18) assumes a 1:2 ratio, yet Table I reports 0.417, 0.430, and 0.436 for N=8,10,12, each at least two standard deviations from 1/2; Table II supports Eq. (20) with values 0.436–0.550; and Table III gives 0.844–1.031 instead of 1 for Eq. (22). Because the final x-dependence in Eq. (27) is built from these assumed ratios, the function is not a parameter-free prediction but a curve calibrated to the dataset. Fig. 5 confirms a clear mismatch. The authors should either provide a first-principles derivation of Eq. (27), show that the deviations in Tables I–III do not affect the functional form within statistical error, or reframe Eq. (27) as an empirical fit and test it on independent data.","section":"Section IV"},{"comment":"The claimed scaling collapse is the central empirical result, but no error bars or quantitative collapse metric are provided. With only N=8,10,12 and a small set of χ values, the visual collapse in Fig. 2 cannot be evaluated rigorously. The authors should report uncertainties over the 100 weight samples, provide a quantitative measure such as the maximum deviation from a single curve, and address the discreteness of the horizontal axis for small χ.","section":"Fig. 2 and Eq. (16)"},{"comment":"The 'theoretical basis' of Eqs. (18), (20), and (22) is presented as the observation that the spatial sum of EE increases linearly in the early part of the circuit and saturates afterward. This observation is made from the same numerical data and does not constitute an independent derivation; the three proportionality assumptions remain empirical calibration assumptions. The authors should state this explicitly or derive the ratios from a concrete model of EE growth, otherwise the phrase 'theoretical analysis' overstates what Eqs. (18), (20), and (22) provide.","section":"Appendix C"},{"comment":"The row 'N = 12, χ = 8' in Table II is listed as 0.43605110 ± 0.02234253, which is identical to the N=12 row of Table I. This is almost certainly a transcription error. Because Table II is used to motivate Eq. (20), the authors must correct or explain this entry.","section":"Table II"}],"minor_comments":[{"comment":"The display 'P Sχ P S' should be formatted as a ratio with a slash or fraction bar, and the denominator contains a typo: 'λ2 i,j,,k' has a double comma.","section":"Eq. (16)"},{"comment":"The text says Fig. 5 plots the 'theoretical approximation given by the second line of Eq. (27)', while the caption says Eq. (27); please clarify which form is plotted, since the final line of Eq. (27) uses the additional approximation ⌊log2 χ⌋ ≈ log2 χ.","section":"Section IV and Fig. 5"},{"comment":"The approximations N(N−1)/2 ≈ N²/2 and N(N−1) ≈ N² introduce errors of roughly 8–14% for N=8–12, which are comparable to the discrepancies in Fig. 5; the discussion should quantify the contribution of these approximations to the observed mismatch.","section":"Section IV, Eqs. (25)-(26)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of a quantum-information journal, but the central analytic formula is not supported by the presented evidence. The empirical scaling law in Fig. 2 could be a publishable contribution after a careful revision that either treats Eq. (27) as an empirical fit or replaces it with an independent derivation, adds error bars, and corrects the apparent transcription error in Table II."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know: the empirical collapse in Fig. 2 looks real and useful, but the analytic scaling function in Eq. (27) is not what the paper claims it is—it is calibrated to assumptions the authors' own data contradict.\n\nThe solid part is the observation that, for one-layer QAOA on complete-graph MAXCUT, the ratio of summed entanglement entropy (over all bonds and all CNOT steps) with and without MPS truncation collapses as a function of 2 log2 χ/N. That extends the energy-based scaling of Dupont et al. and Nakhl et al., and the paper is clear that the summation over all bonds is essential because single-bond ratios do not collapse. The authors also deserve credit for showing the scaling data in Fig. 2 without hiding the spread, and for explicitly discussing the mismatch in Fig. 5.\n\nThe broken part is the derivation of Eq. (27). It rests on three assumed ratios: Eqs. (18), (20), and (22) set the first/second half entropy ratio to 1:2 and the truncation-invariance ratio to 1. The paper's own Tables I–III show these assumptions are violated: the first/second ratios are 0.42–0.44 rather than 0.5, and the truncation ratios range from 0.84 to 1.03. The theoretical curve then divides out the α that would absorb these discrepancies and builds its functional form from exactly the idealized numbers. It is no surprise that Eq. (27) does not match Fig. 5; the paper's suggested explanations (rough N^2 replacements) are secondary. Eq. (27) is a heuristic fit with hand-picked constants, not a derived scaling law.\n\nThe missing error bars in Fig. 2 are a minor but fixable issue. More important is the absence of code and data; without them, a referee cannot check whether the collapse is robust across graph realizations. The paper should be revised to reframe Eq. (27) as an empirical fit, provide the data, and either derive the assumed ratios from first principles or abandon them.\n\nThis is a paper for people interested in classical simulability of QAOA and tensor network methods. The empirical scaling relation, if it survives scrutiny, is a modest contribution. The analytic claim needs major rework. I would still send it to peer review: the empirical observation is worth checking, and a good referee can help the authors separate the wheat from the chaff. I would not cite it in my own work until the analytic part is fixed. Serious thinker: yes.","headline":"Fig. 2's empirical entropy-sum scaling collapse is plausible; Eq. (27) is a calibrated heuristic that the paper's own data contradict.","tokens_in":14499,"tokens_out":3350,"would_cite":false,"duration_ms":37114,"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":"For one-layer QAOA on complete-graph MAXCUT, the ratio of summed entanglement entropy with versus without MPS truncation is a universal function of $2\\log_2\\chi/N$, independent of the number of qubits, and the paper proposes an analytic…","keywords":["QAOA","entanglement entropy","matrix product states","tensor network simulation","scaling relation","MAXCUT","truncation error"],"falsifier":"Run the same one-layer complete-graph MAXCUT simulation at fixed $2\\log_2\\chi/N$ for system sizes beyond the fitted range (say $N=14,16,20$) with fresh random edge weights, and check whether the measured $PS_\\chi/PS$ points land on the curve defined by the second line of Eq. (27) within the scatter seen in Fig. 5; if they systematically drift away as $N$ grows, the claimed $N$-independence and the analytic function are falsified.","tokens_in":13427,"feed_emoji":"⚛️","tokens_out":10284,"duration_ms":91205,"temperature":0.7,"pith_summary":"The paper tries to establish that, for one-layer QAOA solving MAXCUT on a complete graph, the error introduced by truncating the matrix-product-state bond dimension is governed by a single dimensionless variable, $2\\log_2\\chi/N$, rather than by system size $N$ and bond dimension $\\chi$ separately. It reports that the ratio of summed entanglement entropies with and without truncation collapses onto one universal curve, and it proposes an explicit analytic function, Eq. (27), approximating that curve. If correct, this would be the first theoretical account of the scaling relations previously observed for QAOA energy expectation values, and it would let a simulator predict truncation error from system size and bond dimension alone. The derivation rests on three empirical ratio assumptions extracted from the same simulations used to test the formula, so the proposal is an extension and partial calibration of earlier scaling results rather than a standalone derivation.","feed_headline":"One curve predicts QAOA truncation error at any qubit count","feed_subtitle":"For one-layer MAXCUT, entanglement loss depends on one ratio, not system size.","key_machinery":"The central object is the space-time summed entanglement entropy, $\\sum_{\\text{space\\&time}} S$, obtained by summing the bipartite entanglement entropy over every bond of every matrix product state immediately after each CNOT gate in the one-layer QAOA circuit. The carrying mechanism is the chain of proportionality $(\\sum_{\\text{space}} S)_{\\text{first}} : (\\sum_{\\text{space}} S)_{\\text{second}} : (\\sum_{\\text{space}} S)_{\\text{max}} \\sim \\alpha/2 : \\alpha : 1$ and its truncated analogue, which lets the ratio in Eq. (16) be replaced by ratios of closed-form maxima, $N^2/4$ (Eq. (17)) and Eq. (19). Dividing the resulting expressions for the truncated and untruncated sums yields the proposed function $H(2\\log_2\\chi/N)$ in Eq. (27).","core_discovery":"On its own terms, the paper's discovery is that the effect of MPS bond-dimension truncation on one-layer QAOA for complete-graph MAXCUT is governed by a universal curve. With $PS_\\chi/PS$ defined as the ratio of the summed entanglement entropy over all bonds and all CNOT steps with and without truncation, the data collapse according to $PS_\\chi/PS = G(2\\log_2\\chi/N)$ (Eq. (16)), with the same function $G$ for every system size $N$. The paper then proposes an explicit analytic approximant, the second line of Eq. (27), built from the theoretical maxima of the summed entropy in Eqs. (17) and (19) and from three empirical ratio assumptions, Eqs. (18), (20), and (22). The claim is that this gives the first theoretical analysis of the previously noted scaling relations, and that the universality traces to an $N$-independent pattern in how entanglement entropy grows along the QAOA circuit.","pith_inferences":["The authors do not test whether the collapse persists for more than one QAOA layer; a natural extension is to check whether the scaling variable becomes $p(2\\log_2\\chi/N)$ or $(2\\log_2\\chi/N)/p$, which would turn the curve into a resource-estimation tool for deeper circuits.","Because the assumed ratios in Tables I–III deviate from the ideal values (first/second around 0.42–0.44 rather than 0.5, and Eq. (22) ratios between 0.84 and 1.03 rather than 1), the proposed function is best read as an interpolation anchored to the numerics rather than a fully predictive derivation.","If the universal curve is confirmed at larger $N$, a classical simulator could choose the bond dimension $\\chi$ in advance from a target entanglement fidelity by inverting $H(2\\log_2\\chi/N)$, making the scaling relation a practical preconditioning step for QAOA simulations."],"forward_implications":["At fixed $2\\log_2\\chi/N$, doubling the number of qubits requires doubling the bond dimension to keep the same entanglement-entropy ratio, so the MPS cost of a fixed-fidelity one-layer QAOA simulation grows only linearly in $N$.","The analytic approximant Eq. (27) makes a concrete numerical prediction for truncation error, so deviations between simulation and Eq. (27) can be used to test whether the 1:2 and truncation-invariance assumptions are the correct mechanism.","Because the ratio saturates near 1 as $2\\log_2\\chi/N \\to 1$, the curve identifies a threshold bond dimension $\\chi \\sim 2^{N/2}$ beyond which truncation causes almost no entanglement loss.","The paper argues that any quantum algorithm whose summed entanglement entropy grows linearly and then saturates with circuit depth would show an analogous scaling relation, making the phenomenon a candidate fingerprint of QAOA-like optimized circuits."],"supporting_citations":[{"why":"Established the original scaling relation between MPS truncation and QAOA energy expectation values that this work extends.","marker":"[19]"},{"why":"Extended the scaling relation to more layer counts and graph types and reported an analogous relation for entanglement entropy.","marker":"[21]"},{"why":"Provides the matrix product state formalism used to represent the QAOA state and to define bond dimensions and truncation.","marker":"[20]"},{"why":"Supplies the singular value decomposition used for truncation and for computing entanglement entropies from singular values.","marker":"[16]"},{"why":"Provides the QAOA circuit simulation backend used for parameter optimization in the numerical experiments.","marker":"[22]"},{"why":"Provides the MPS simulation engine used to generate the truncated and untruncated entanglement data.","marker":"[23]"}],"fun_headline_variants":["QAOA truncation error collapses to one curve","Single ratio predicts QAOA simulation error","QAOA error scales with one ratio, not size","Entanglement ratio determines QAOA truncation error","New analytic formula for QAOA error scaling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivation depends on treating the numerical ratios in Tables I–III as exact equalities — first-half to second-half summed entropy of 1:2 and truncation-invariance of the normalized second-half sum — even though the same tables show deviations (about 0.42–0.44 and 0.84–1.03) that exceed the stated error bars in some rows.","fun_headline_variants_meta":{"raw":{"variants":["QAOA truncation error collapses to one curve","Single ratio predicts QAOA simulation error","QAOA error scales with one ratio, not size","Entanglement ratio determines QAOA truncation error","New analytic formula for QAOA error scaling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001075,"raw_usage":{"total_tokens":4516,"prompt_tokens":974,"completion_tokens":3542,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":3473}},"tokens_in":590,"tokens_out":3542,"duration_ms":21903,"temperature":1.0,"reasoning_tokens":3473,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:49:05.144760+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same one-layer complete-graph MAXCUT simulation at fixed $2\\log_2\\chi/N$ for system sizes beyond the fitted range (say $N=14,16,20$) with fresh random edge weights, and check whether the measured $PS_\\chi/PS$ points land on the curve defined by the second line of Eq. (27) within the scatter seen in Fig. 5; if they systematically drift away as $N$ grows, the claimed $N$-independence and the analytic function are falsified.","supporting_citations":[{"cited_title":"scaling relation","cited_arxiv_id":null,"evidence_quote":"Established the original scaling relation between MPS truncation and QAOA energy expectation values that this work extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extended the scaling relation to more layer counts and graph types and reported an analogous relation for entanglement entropy."},{"cited_title":"Bisgard, Analysis and Linear Algebra: The Singular V alue Decomposition and Applications","cited_arxiv_id":null,"evidence_quote":"Provides the matrix product state formalism used to represent the QAOA state and to define bond dimensions and truncation."},{"cited_title":"Okunishi, T","cited_arxiv_id":null,"evidence_quote":"Supplies the singular value decomposition used for truncation and for computing entanglement entropies from singular values."},{"cited_title":"Dupont , N","cited_arxiv_id":null,"evidence_quote":"Provides the QAOA circuit simulation backend used for parameter optimization in the numerical experiments."},{"cited_title":"Perez-Garcia, F","cited_arxiv_id":null,"evidence_quote":"Provides the MPS simulation engine used to generate the truncated and untruncated entanglement data."}],"review_version":1}