{"id":"2c46ace7-65b2-4601-8090-0559e4a83b94","arxiv_id":"2501.11085","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In the double-scaling limit, the singular-value moments of products of truncated Haar unitary matrices equal a closed expression involving the Erlang delay function, from which the entropy reduction is computed.","lead":"This paper derives exact formulas for the singular value moments and von Neumann entropy of products of truncated random unitary matrices in a large-N, large-L limit. The result connects the problem to the Erlang delay function from queueing theory and gives a universal description of purification in monitored quantum circuits.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"All-p Erlang identification is assumed without proof, and the density and entropy results depend on an unverified analytic continuation.","rationale":"I read the paper in good faith. The recursion (3.3) is clearly stated, the first six polynomials match the Erlang formula, and the numerical comparisons are encouraging. The load-bearing weakness is exactly what the reader identified: the all-p identification with the Erlang delay function is assumed, not proven, and the density and entropy results rely on an analytic continuation that is not independently justified. The paper's own sentence in Sec. IV.A admits this gap. I do not see a reason to accuse the argument of an internal inconsistency; the issue is proof completeness and control of the large-p regime. The numerical tests are suggestive but limited, covering only small p and moderate N. A generating-function derivation from the recursion would settle the integer-p identity and provide the foundation for the analytic continuation. Therefore the conditional verdict is appropriate: the result is plausible and potentially correct, but the central technical claim requires a proof or a substantially stronger test before acceptance.","tokens_in":7472,"tokens_out":6549,"duration_ms":67184,"concrete_test":"Derive the bivariate generating function C(z,w)=sum_{p>=1} sum_{k=0}^{p-1} c_{k,p} z^p w^k from the continuum limit of recursion (3.3), writing S_p(tau)=e^{-p tau} sum_{k=0}^{p-1} c_{k,p} tau^k. Compare this generating function directly with the generating function of the Erlang polynomials (4.3). If they agree, the all-p identity is proved and the analytic continuation used in Secs. V and VI is justified. If they disagree, the central result (4.1) fails for some integer p and the entropy formula (6.8) is called into question.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central technical result (4.1) is obtained from the recursion (3.3), but the step from the first six matching polynomials (4.2) to the Erlang delay function (4.3) for all p is explicitly assumed: 'We assume the correspondence between Eqs. (4.2) and (4.3) holds for all p, although we have no insight why a delay function from queueing theory would appear in the present context' (Sec. IV.A). This is not a cosmetic gap. The singular-value density is reconstructed by inverse Laplace transform (5.2), which requires S_p(L) as an analytic function of complex p, and the von Neumann entropy (6.8) is obtained from the alpha-to-1 limit of the Renyi entropy, which requires G_alpha(tau)/Gamma(alpha) for alpha near 1. The recursion determines only integer p, and infinitely many analytic continuations agree at the integers. Unless the Erlang continuation is the correct one, Eqs. (5.3) and (6.8) lack foundation. Moreover, the Gaussian approximation leading to (3.3) is derived at fixed p; its error may grow with p, so the large-p moments used for the Laplace inversion tail are the least controlled. The numerical evidence (N=300, 20 realizations, p up to 6) does not constrain the large-p or complex-p behavior. A proof or a much stronger test of the all-p identity is needed.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes products C_L = tilde U_L ... tilde U_1 of L independently Haar-distributed unitary matrices of dimension N, each truncated by zeroing the first delta N rows and columns, in the double-scaling limit L, N -> infinity with fixed tau = L delta N / N. The equivalent random-projector product (Sec. II.B) permits a large-N Gaussian average over the projector vectors, which yields a closed system of first-order recursion relations (Eq. 3.3) for the singular-value moments S_p(n) = (1/N) E Tr(Q_n Q_n^dagger)^p. The paper's central claim is that in the double-scaling limit S_p(L) = e^{-p tau} G_p(tau) / Gamma(p), with G_p the Erlang delay function of queueing theory (Eqs. 4.1-4.3); the match is verified numerically for p = 1..6. From the claimed analytic continuation in p, the author reconstructs the singular-value density by inverse Laplace transformation (Sec. V), derives the large-p tail (4.4) and the square-root edge lambda_min = tau - 1 - ln tau of the density (5.3), and obtains the von Neumann entropy reduction S_L - ln N = -ln tau + e^tau (tau - 1) Gamma(0, tau) - gamma_Euler (Eq. 6.8) by analytically continuing the Rényi entropy. The entropy asymptotics agree with three independent weak-measurement models (Refs. 15-17). The paper states explicitly that the all-p identification with the Erlang function, and hence the analytic continuation, is assumed rather than proved.","tokens_in":7677,"tokens_out":22611,"duration_ms":227801,"significance":"If Eq. (4.1) holds for all p, this is a clean, parameter-free characterization of random contraction products in the double-scaling limit, with new falsifiable predictions: the large-p decay (4.4), the square-root density edge (5.3), and the full Rényi entropy curve (6.7). The paper earns credit for deriving the moment recursion from elementary Wick contractions, for checking both unitary and orthogonal ensembles numerically (N = 300, 20 realizations), for reporting the explicit match to a known special function, and for confirming the entropy asymptotics against independent models (Refs. 15-17). The crossover from linear to logarithmic entropy reduction in tau is robust and clearly presented. The central limitation is that the technical basis of the density and entropy results, the all-p Erlang identification and its analytic continuation to non-integer p, is explicitly conjectural; the present evidence does not determine the quantities in Eqs. (5.3) and (6.8).","major_comments":[{"comment":"The transition from the recursion (3.3) to the closed form (4.1) is not demonstrated (the text says 'We find'), and the matching of the six polynomials in (4.2) to the Erlang delay function (4.3) is explicitly stated as an assumption: 'We assume the correspondence between Eqs. (4.2) and (4.3) holds for all p, although we have no insight why a delay function from queueing theory would appear in the present context.' This assumption is load-bearing for Eq. (4.4), for the density near the edge (5.3), and for the Rényi and von Neumann entropies (6.7)-(6.8). The numerical check (N = 300, 20 realizations, p up to 6) does not constrain the large-p moments that determine those quantities. Because the recursion (3.3) is a closed deterministic system for integer p within the leading-order Gaussian theory, the identification can be checked without further random-matrix sampling to moderately large p (for example, p up to about 30) by iterating the recursion symbolically, and ideally proved by induction if a closed-form solution exists. I ask that the author either supply the derivation of (4.1) for general p, or at minimum add the high-p recursion check, and state explicitly which numbered results are theorems and which are conjectural.","section":"Sec. IV.A, Eqs. (4.1)-(4.3)"},{"comment":"The analytic continuation to non-integer p is used but not justified. The recursion (3.3) defines S_p(n) only for positive integer p, yet the inverse Laplace transform (5.2) evaluates S_p(L) on a vertical contour in the complex p-plane, and the Rényi entropy expression (6.7) requires S_alpha for real alpha, including the limit alpha -> 1 that gives the von Neumann entropy (6.8). Infinitely many entire functions interpolate the integer moments, so the Erlang continuation is one of many possibilities unless the physical moment sequence is shown to be its restriction. For (6.8) it would suffice to prove correctness of the continuation in a neighborhood of alpha = 1; for the density (5.3), a direct verification that the claimed density reproduces all integer moments, or a moment-problem determinacy argument applied to the full integer moment sequence of the claimed density, would close the gap.","section":"Secs. V.A and VI, Eqs. (5.1)-(5.2), (6.4)-(6.8)"},{"comment":"The Gaussian approximation is not controlled uniformly in p. Equation (3.3) is quoted to leading order in 1/N with O(1/N^2) remainders, and the number of recursion steps is L = tau N / delta N, so the accumulated error is O(tau / (N delta N)) only if the remainders are uniform in n and do not grow too rapidly with p. The Wick-contraction count in the average over the projector vectors grows factorially with p, so the large-p tail (4.4), which generates the square-root edge (5.3), is the part of the result least protected by the N -> infinity limit. A brief statement on the p-dependence of the discarded remainders, or a numerical comparison of the tail against the recursion solution at moderately large p, would address the point.","section":"Sec. II.C and Eq. (3.3)"}],"minor_comments":[{"comment":"There is a typesetting error: 'TrBp n]' should read 'Tr(B_n^p)', and 'B n' should read 'B_n' in the same line.","section":"Eq. (3.1)"},{"comment":"The definition of d_i as the smallest integer delta such that x_{i+delta mod p} = 1 is difficult to parse; a worked example for p = 4 would help the reader verify the recursion.","section":"Eq. (3.3)"},{"comment":"The correspondence between the length-L truncated product C_L and the (L+1)-term projector product, which underlies the shifted comparison tau = (L+1) delta N / N in Fig. 1 and footnote 30, should be stated explicitly in the main text.","section":"Sec. II.B"},{"comment":"References [24] and [29] cite the same book by R. B. Cooper with different publishers; they should be merged into a single entry.","section":"Refs. [24] and [29]"},{"comment":"The numerical Laplace inversion of (5.2) is not described (contour choice, discretization); one sentence on the method would aid reproducibility.","section":"Sec. V.A"},{"comment":"The notation 'N = 2N' uses the same symbol N for the Hilbert-space dimension and the number of spin-1/2 degrees of freedom; distinct symbols would avoid confusion.","section":"Footnote 32"},{"comment":"The replacement of the random traces by their expectation values in Eq. (6.5) presumes self-averaging of Tr(C_L C_L^dagger)^alpha; a brief variance estimate would make this step explicit.","section":"Sec. VI, Eq. (6.5)"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the paper is a cleanly written letter whose central technical result is, by the author's own admission, an assumption ('We assume the correspondence between Eqs. (4.2) and (4.3) holds for all p'). I do not see circularity: the Erlang identification is inferred from the recursion-generated polynomials and is externally supported by the agreement of the entropy asymptotics with Refs. 15-17. The gap is, however, load-bearing and well delimited: the recursion (3.3) determines every integer moment within the leading-order Gaussian theory, so the conjecture is verifiable, and possibly provable, without new random-matrix simulation. My request for major revision is therefore not a request to change scope but to supply the missing proof or high-p verification and to mark each numbered result as theorem or conjecture. The paper fits the journal's scope; I would also note the duplicated reference [24]/[29]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nYou should know two things: this paper gives a compact formula for singular-value moments of products of truncated Haar random matrices in the double-scaling limit L,N→∞ at fixed τ=LδN/N, and it uses that formula to get a closed expression for the von Neumann entropy of the associated density matrix. The moment formula (4.1) with the Erlang delay function, the reconstructed density, and the entropy formula (6.8) are new. The numerics in Figures 1, 2, and 4 back them up at the level tested.\n\nWhat the paper does well: it is genuinely simple. The Gaussian approximation gives a first-order linear recursion, solvable sequentially, and the first six polynomials match the Erlang delay function. The entropy crossover from −τ to 1−lnτ−γ agrees with existing weak-measurement models, an independent check. The paper is transparent: the author states openly that the all-p Erlang correspondence is assumed and that there is no insight into why a queueing-theory delay function appears. That honesty is rare.\n\nThe soft spots are exactly where the author says they are. The step from six polynomials to the all-p Erlang function is a conjecture. It is not circular—it is pattern matching on polynomials, not fitting to the target entropy—but it is load-bearing. The density reconstruction and the entropy formula both require analytic continuation to complex p and to α near 1, and infinitely many analytic continuations agree with the moment sequence at the integers. Unless the Erlang continuation is the right one, the density and entropy formulas lack justification. The recursion is derived at fixed p, so the large-p moments used for the Laplace-inversion tail are the least controlled, and the numerics (N=300, 20 realizations, p up to 6) do not constrain the large-p or complex-p behavior.\n\nThe conjecture is crisp and falsifiable. A referee can ask for a proof of the all-p correspondence or for a much stronger numerical test (larger N, larger p, error bars). This is the kind of paper that deserves peer review: written by an expert, honest about its own gap, new and likely to be used. Even if the referee demands heavy revision, it deserves referee time.\n\nWho it is for: random matrix products, monitored circuits, purification dynamics. I would bring it to a reading group with an RMT session. I would cite the integer-p moment formula now; I would wait on the entropy formula until the analytic continuation is proven or much better tested.\n\nRecommendation: send it to peer review, conditional on the author either proving the all-p Erlang identification or providing a more demanding numerical check. The author should also flesh out the derivation of (4.1); 'We find' is thin for a central result.","headline":"Clean double-scaling moment formula and entropy for truncated unitary products, honest about a load-bearing unproven Erlang conjecture; deserves peer review.","tokens_in":8258,"tokens_out":3015,"would_cite":true,"duration_ms":27902,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","15B52"],"pacs":[],"model":"deepseek-v4-flash","headline":"The singular-value moments of products of truncated Haar unitary matrices are given by the Erlang delay function in the double-scaling limit, yielding an explicit von Neumann entropy reduction formula.","keywords":["truncated unitary matrices","singular value moments","Erlang delay function","von Neumann entropy","monitored quantum circuits","random contraction process","double-scaling limit","weak measurements"],"falsifier":"Take $N$ large with fixed truncation depth $\\delta N$, form many independent products of $L=\\tau N/\\delta N$ truncated Haar unitary matrices, and numerically histogram $\\lambda=-\\ln\\sigma^2$ or compute the R\\'enyi entropy $E_\\alpha$ at a non-integer index such as $\\alpha=1/2$ or $3/2$ directly from the singular values. Compare with the inverse Laplace transform of (4.1) and with (6.7): the integer-moment checks in the paper cannot distinguish the assumed all-$p$ Erlang continuation, but a non-integer R\\'enyi entropy or a high-resolution density histogram can, so a systematic disagreement would disprove the continuation while leaving the integer moments intact.","tokens_in":7196,"feed_emoji":"🎲","tokens_out":7972,"duration_ms":81900,"temperature":0.7,"pith_summary":"Products of independently Haar-distributed unitary matrices, each truncated to an $(N-\\delta N)$-dimensional subspace, model measurement-induced purification in monitored quantum circuits. This paper claims that in the double-scaling limit $L,N\\to\\infty$ at fixed ratio $\\tau=L\\,\\delta N/N$, every integer moment of the squared singular values of the product equals $\\langle\\sigma^{2p}\\rangle = e^{-p\\tau}G_p(\\tau)/\\Gamma(p)$, where $G_p(\\tau)$ is the Erlang delay function from queueing theory. Because the Erlang expression can be continued to complex $p$, the full singular-value density follows by Laplace inversion, and the von Neumann entropy of the normalized product obeys an explicit formula in terms of $\\tau$. The upshot is a single-parameter description of purification: entropy falls linearly for $\\tau\\ll 1$ and logarithmically for $\\tau\\gg 1$, with a crossover at $\\tau=1$ where the singular-value distribution is close to an exponentiated chi-square. If correct, the result unifies the projective contraction picture with earlier weak-measurement models and supplies analytic control over random contraction processes.","feed_headline":"Truncated unitary products: moments match Erlang delay function","feed_subtitle":"One ratio τ = LδN/N controls singular-value density and von Neumann entropy in monitored circuits.","key_machinery":"The mechanism is a recursion for the moments $S_p(n)=N^{-1}\\mathrm{Tr}\\,(Q_nQ_n^\\dagger)^p$ in the large-$N$ Gaussian approximation. Adding one projector $P_{n+1}$ produces a first-order linear recursion in $S_p$ whose coefficients are sums over binary strings $x_i\\in\\{0,1\\}$ of products of lower moments $S_{d_i}(n)^{x_i}$, with $d_i$ determined by cyclic runs of ones. Solving these recursions sequentially and taking $L,N\\to\\infty$ at fixed $\\tau=L\\delta N/N$ gives the polynomial family (4.2), which is then identified with the Erlang delay function $G_p(\\tau)$, a function whose usual home is queueing theory. The decisive feature is that the identification yields an analytic function of $p$, so singular-value density and R\\'enyi and von Neumann entropies can be extracted without a separate moment-generating-function calculation.","core_discovery":"The central discovery is formula (4.1): in the double-scaling limit, the moments $S_p(L)$ of the squared singular values of $C_L=\\tilde U_L\\cdots \\tilde U_1$ are $S_p(L)=e^{-p\\tau}G_p(\\tau)/\\Gamma(p)$, with $G_p(\\tau)$ the Erlang delay function $G_p(\\tau)=(1-\\tau)(p-1)!\\sum_{i=0}^{p-1}(p\\tau)^i/i!+\\tau^p p^{p-1}$. The first six polynomial cases match a known series pattern that identifies $G_p$ as the Erlang delay function, and the paper assumes this identification for all $p$. This closed form allows analytic continuation in $p$, so the moment sequence determines the density of $\\lambda=-\\ln\\sigma^2$ by inverse Laplace transform, producing a delta-function peak at unit singular value for $\\tau<1$, a square-root edge at $\\lambda_{\\min}=\\tau-1-\\ln\\tau$ for $\\tau>1$, and near $\\tau=1$ an exponentiated chi-square distribution. The same continuation yields the von Neumann entropy reduction $S_L-\\ln N = -\\ln\\tau+e^\\tau(\\tau-1)\\Gamma(0,\\tau)-\\gamma_{\\rm Euler}$, whose small- and large-$\\tau$ asymptotics match existing weak-measurement models.","pith_inferences":["Because the Gaussian approximation treats real and complex projection vectors alike, the same Erlang moment formula should hold for products of truncated real orthogonal matrices; a direct finite-$N$ check at non-integer powers of the singular values would test the all-$p$ continuation.","The appearance of the Erlang delay function hints that the log singular values in the double-scaling limit may follow a Poisson or queueing-type process, which would give a mechanistic explanation the paper does not supply.","The $\\tau=1$ exponentiated chi-square distribution could provide a quantitative convergence criterion for Kaczmarz-type algorithms, going beyond the heuristic connection the paper closes with.","The entropy formula suggests that finite-$N$ corrections to the double-scaling limit depend only on $\\delta N/N$; computing those corrections would show whether the Erlang continuation is exact or only leading order."],"forward_implications":["All singular-value statistics in the double-scaling limit depend only on the single parameter $\\tau=L\\delta N/N$, with integer moments $e^{-p\\tau}G_p(\\tau)/\\Gamma(p)$.","The singular-value density separates into regimes: for $\\tau<1$ a fraction $1-\\tau$ of singular values sit at unity plus a continuous part with a square-root edge; for $\\tau>1$ the continuous part starts at $\\lambda_{\\min}=\\tau-1-\\ln\\tau$; at $\\tau=1$ the density is close to an exponentiated chi-square with one degree of freedom.","The von Neumann entropy of the normalized product is $S_L=\\ln N -\\ln\\tau + e^\\tau(\\tau-1)\\Gamma(0,\\tau)-\\gamma_{\\rm Euler}$, giving linear decay in $\\tau$ for small $\\tau$ and logarithmic decay $\\ln(N/\\tau)$ for $1\\ll\\tau\\ll N$.","The $\\tau=1$ case gives a closed-form moment formula for products of $N$ rank-$(N-1)$ random projections, relevant to Kaczmarz-type projection algorithms.","The entropy asymptotics agree with non-projective weak-measurement models, supporting universality of the purification transition in monitored quantum circuits."],"supporting_citations":[{"why":"Establishes the projector-product formulation and the monitored-circuit context in which the truncation model is used.","marker":"[15]"},{"why":"Provides prior microscopic correlation results for products of truncated unitary matrices that the double-scaling moment formula extends.","marker":"[6]"},{"why":"Situates the double-scaling limit within the known exact and asymptotic theory for products of independent random matrices.","marker":"[7]"},{"why":"Supplies the queueing-theory definition and context of the Erlang delay function used to identify $G_p(\\tau)$.","marker":"[24]"},{"why":"Identifies the polynomial family (4.2) with the Erlang delay function through the OEIS series.","marker":"[28]"},{"why":"Provides the large-argument incomplete-Gamma asymptotics that yield the three regimes in (4.4).","marker":"[31]"},{"why":"One of the weak-measurement models whose entropy asymptotics are matched to establish universality of the logarithmic decay.","marker":"[16]"}],"fun_headline_variants":["Erlang delay function governs product unitary moments","Truncated unitaries: singular-value moments match queueing theory","Crossover at τ=1 in entropy of monitored-circuit products","Random matrix product moments: Erlang function from queueing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the polynomial family (4.2), checked for $p=1,\\dots,6$, coincides with the Erlang delay function for every real $p$, including the non-integer values needed for the density and entropy; the author states this is assumed without insight, and if it fails the derived density and entropy formulas would be wrong even though the integer moments are correct.","fun_headline_variants_meta":{"raw":{"variants":["Erlang delay function governs product unitary moments","Truncated unitaries: singular-value moments match queueing theory","Crossover at τ=1 in entropy of monitored-circuit products","Random matrix product moments: Erlang function from queueing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000192,"raw_usage":{"total_tokens":1346,"prompt_tokens":947,"completion_tokens":399,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":332}},"tokens_in":563,"tokens_out":399,"duration_ms":4957,"temperature":1.0,"reasoning_tokens":332,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:38:54.164429+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $N$ large with fixed truncation depth $\\delta N$, form many independent products of $L=\\tau N/\\delta N$ truncated Haar unitary matrices, and numerically histogram $\\lambda=-\\ln\\sigma^2$ or compute the R\\'enyi entropy $E_\\alpha$ at a non-integer index such as $\\alpha=1/2$ or $3/2$ directly from the singular values. Compare with the inverse Laplace transform of (4.1) and with (6.7): the integer-moment checks in the paper cannot distinguish the assumed all-$p$ Erlang continuation, but a non-integer R\\'enyi entropy or a high-resolution density histogram can, so a systematic disagreement would disprove the continuation while leaving the integer moments intact.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the projector-product formulation and the monitored-circuit context in which the truncation model is used."},{"cited_title":"Akemann, Z","cited_arxiv_id":null,"evidence_quote":"Provides prior microscopic correlation results for products of truncated unitary matrices that the double-scaling moment formula extends."},{"cited_title":"Akemann and J","cited_arxiv_id":null,"evidence_quote":"Situates the double-scaling limit within the known exact and asymptotic theory for products of independent random matrices."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the queueing-theory definition and context of the Erlang delay function used to identify $G_p(\\tau)$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Identifies the polynomial family (4.2) with the Erlang delay function through the OEIS series."},{"cited_title":"Ferreira, J","cited_arxiv_id":null,"evidence_quote":"Provides the large-argument incomplete-Gamma asymptotics that yield the three regimes in (4.4)."},{"cited_title":"Gerbino, P","cited_arxiv_id":null,"evidence_quote":"One of the weak-measurement models whose entropy asymptotics are matched to establish universality of the logarithmic decay."}],"review_version":1}