{"id":"64f9f705-6c45-4b10-8dda-2bee3e732469","arxiv_id":"2509.22164","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"In isotropic vector resonant relaxation, the one-loop Martin-Siggia-Rose closure improves two- and three-point correlation predictions over the bare direct-interaction approximation, matching N-body simulations to within 4 to 20 percent.","lead":"This paper tests a mathematical framework for predicting how chaotic, interacting systems evolve statistically, using a model of stars orbiting a supermassive black hole as an example. The authors show that adding one-loop corrections makes the framework's predictions for correlation functions match computer simulations noticeably better.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The one-loop fixed-point solution is not shown to be independent of the Gaussian seed; because the paper itself states uniqueness is not guaranteed, the central predictions may be branch-dependent until a seed-independence test is performed.","rationale":"The reader's weakest assumption identifies exactly the same load-bearing risk: the fixed-point scheme is not tested for independence from the Gaussian initial propagator, and the paper itself concedes that uniqueness is not guaranteed. My stress-test agrees that this is the central weakness. In particular, Section IV A reports that starting from the bare propagator gives slow and marginally stable convergence, while the Gaussian seed converges quickly; no evidence is offered that both routes terminate at the same [G,Γ]. Since the Gaussian seed is constructed from the same coherence time T_c used to normalize the N-body comparisons, seed-dependence would directly contaminate the apparent agreement in the two-point and three-point correlation functions. This is not a fatal flaw, but it is a correctable gap: a seed-independence test would either resolve the concern or reveal that the one-loop closure requires a branch-selection rule. The paper has substantial independent support through its public code, convergence checks in Appendix F, and the nontrivial success of the one-loop skewness predictions for triangles where the bare order vanishes, so a conditional acceptance with revision is the appropriate disposition. No additional rejection-level concern emerged from my reading.","tokens_in":34691,"tokens_out":5661,"duration_ms":52644,"concrete_test":"Run the fixed-point iteration of Eq. (26) from at least three initial propagators: (i) G(0)=g (bare), (ii) G(0)=G_G with T_c scaled by 0.5 and by 2, and (iii) a one-parameter family G(0)(λ)=(1−λ)g+λG_G for λ=0.25,0.5,0.75, keeping Γ(0)=γ as in the paper. Use a reduced but still converged grid (e.g., LMAX=5, TMAX=T_c, DT=T_c/40, ITER=15 with under-relaxation if needed). Compare the terminal C_ℓ(t) for ℓ=2,3,4 and the maximum of Γ_L for triangle (d). If the terminal states spread by more than a few percent, the central predictions are seed-dependent and the one-loop closure is not yet uniquely defined; if they collapse to the same state, the seed-dependence concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is not the size of the one-loop improvement but whether \"the\" one-loop fixed point is a well-defined object. The numerical scheme (Eq. 26) solves the coupled Dyson and vertex equations (Eqs. 22 and 24) by fixed-point iteration, and Section IV A explicitly states that a unique, well-behaved solution is not guaranteed. For the one-loop closure, the authors initialize G(0) with a Gaussian propagator G_G built from the coherence time T_c, and report that initializing from the bare propagator g gives slow, marginally stable convergence. They do not test whether the final [G,Γ] is independent of this seed. If multiple fixed points exist, the one-loop closure has no unique prediction unless a branch-selection rule is supplied; the chosen Gaussian seed is informed by the same T_c that sets the N-body timescale, so the excellent agreement in Figs. 2, 4, and 5 could partly reflect the seed rather than the closure. This is acknowledged as an open point in the text, but it is load-bearing because every quantitative claim in Section V is produced from one seed only.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a numerical implementation of the one-loop Martin-Siggia-Rose (MSR) closure for the isotropic Vector Resonant Relaxation (VRR) model, a purely quadratically nonlinear long-range interacting system with no linear term. The authors solve the coupled Dyson equation (Eq. 22) and the self-consistent vertex equation (Eq. 24) by fixed-point iteration, exploiting isotropy, time stationarity, the fluctuation-dissipation theorem, and angular-momentum contraction rules to manage the numerical cost. They compare the resulting two-point correlation function C_ell (Fig. 2) and three-point skewness S^L (Figs. 4-5) with direct N-body simulations, reporting 4-20% agreement. The one-loop closure is shown to improve upon the bare DIA prediction, in particular by producing nonzero skewness for odd triangles where the bare vertex exactly vanishes. The authors also show that a naive one-loop expansion in powers of the bare vertex diverges (Fig. 9), whereas the self-consistent one-loop closure is stable. Numerical convergence with respect to ITER, LMAX, DT, and TMAX is documented in Appendix F, and the code is publicly available.","tokens_in":34930,"tokens_out":6055,"duration_ms":55804,"significance":"If the central claim holds, this is a substantial methodological advance: it provides a parameter-free, self-consistent one-loop MSR calculation in a fully nonlinear, non-perturbative setting, with falsifiable predictions for both two- and three-point statistics. The manuscript is careful in several respects: it documents convergence with respect to all numerical truncation parameters (Appendix F), reproduces the bare and naive benchmarks from the companion paper, uses N-body measurements with explicit bootstrap error bars, and makes the code publicly available. The reported improvement over bare DIA is concrete and quantified, and the demonstration that a non-self-consistent naive expansion diverges is an important sanity check. The main concerns are not with the derivation but with the uniqueness of the fixed point and with the strength of the validation claims: the quantitative results rest on a single fixed-point seed, and the abstract overstates the validation of the renormalised vertex, which is not directly measured. These issues are addressable and do not, on the present evidence, invalidate the derivation.","major_comments":[{"comment":"The one-loop predictions are obtained from a fixed-point iteration whose uniqueness is explicitly not guaranteed: Section IV A states that \"a unique, well-behaved fixed-point solution for [G,Gamma] in the MSR scheme is not guaranteed,\" and the scheme is initialized with G(0)=G_G, a Gaussian propagator built from the coherence time T_c. The paper reports that starting from the bare propagator g gives \"slow and marginally stable convergence,\" but it does not test whether the final [G,Gamma] is independent of the seed. Since every quantitative claim in Section V (Figs. 2, 4, and 5) is generated from this single seed, the excellent agreement with N-body data could in principle reflect a branch selected by the Gaussian initial condition rather than the intrinsic content of the one-loop closure. Please add a seed-independence study: for example, continue iterations from the bare-propagator seed to large ITER, and also run a perturbed Gaussian seed, then compare the resulting C_ell and S^L within the quoted few-percent accuracy. If such a test is computationally infeasible, the conclusions and abstract must be softened to state that the predictions are conditional on the chosen fixed-point branch.","section":"Section IV A and Section V"},{"comment":"The abstract lists among the validated predictions \"(ii) the renormalised three-point interaction vertex,\" but Section V B explicitly states that Gamma is challenging to measure directly in N-body simulations and that doing so \"will be the topic of a future work.\" The only quantitative N-body comparisons involving Gamma are made indirectly through the skewness S^L in Section V C, which is a convolution of G and Gamma (Eq. E4), not a direct measurement of the vertex itself. Thus the renormalised vertex is predicted and examined structurally (Fig. 3), but it is not quantitatively validated against N-body data. Please rephrase the abstract and any summary statements to say that the formalism predicts Gamma and that the skewness comparisons provide an indirect consistency test, or add a direct measurement of Gamma.","section":"Abstract and Section V B"},{"comment":"The convergence tests in Appendix F (Figs. 10-13) monitor the two-point correlation function C_ell only; no equivalent convergence diagnostic is reported for the renormalised vertex Gamma^L or for the residual of the vertex equation (Eq. 26a). Since Gamma^L is the central new object and the one-loop diagram is the computational bottleneck, it would be valuable to report the relative change in Gamma^L between successive iterations (or the norm of the residual of Eq. 26a) alongside the C_ell convergence. This would strengthen the claim that the fixed point is actually resolved for the vertex, not only for the two-point function.","section":"Appendix F and Section IV C"}],"minor_comments":[{"comment":"There is a typo in the phrase \"methods from stastical closure theory\"; it should read \"statistical closure theory.\"","section":"Section II"},{"comment":"The sentence \"the one-loop prediction correctly recovers the oscillation of the correlation function that predicts a negative correlation for t/T_c by 0.6\" is awkward; consider rephrasing as \"correctly recovers the oscillation, predicting a negative correlation for t/T_c greater than about 0.6.\"","section":"Section V A"},{"comment":"The bottom panel of Fig. 2 shows only the one-loop absolute error; including the bare-order error in the same panel would make the claimed 80% improvement directly visible to the reader.","section":"Figure 2"},{"comment":"The norm introduced in Eq. (48) is not named; please state explicitly that it is the L2 norm over the square domain [0,T_c]^2.","section":"Equation (48)"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about seed dependence is valid and is the main reason for the major-revision recommendation. The abstract's claim of validating the renormalised vertex is also too strong, since only the skewness is compared with N-body data. These are fixable with additional numerical tests or by appropriately softening the claims; I do not see a fatal flaw in the derivation or in the reported N-body comparisons."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a serious look. What is actually new: this paper takes the MSR one-loop closure for VRR from formal equations to a numerically converged fixed point, and shows it beats bare DIA in two respects: the two-point correlation error drops from about 0.15 to 0.03 for ell=4, and the three-point skewness becomes nonzero for triangles where bare order predicts exactly zero, matching N-body within 4-20 percent. That is a concrete, reproducible result: the code is public, convergence tests for LMAX, DT, TMAX, ITER are in the appendix, and the N-scaling analysis (even vs odd triangles) is a nice piece of work in itself.\n\nThe soft spots are two, and neither sinks the paper. First, the abstract says they \"quantitatively validate\" the renormalized vertex Gamma, but in Section V B they explicitly say Gamma is not directly measured; what is compared is the three-point correlation built from Gamma. The abstract overclaims. Second, and more substantive: the fixed-point solver is seeded with a Gaussian propagator built from the coherence time T_c, and the paper states that a unique fixed point is not guaranteed and that starting from the bare propagator converges slowly and only marginally stably. That means the final [G, Gamma] might depend on the seed. The concern is real but not fatal: T_c is derived from the Hamiltonian, not fitted to the N-body data, so the seed is physically motivated, and the bare-order case was checked to be seed-independent. Still, the one-loop predictions are all produced from a single seed; a revision should either test independence from the initial G(0) or argue, with evidence, that the branch selected is the physical one. Without that, the claim to have \"the\" one-loop prediction is provisional.\n\nThe math looks careful. The derivation follows MSR and FF25, and the contraction rules for the Elsasser coefficients are standard. The citation pattern is appropriate: self-citations to FF25 serve as a benchmark, not as a substitute for evidence.\n\nWho is this for? People working on closures for long-range interacting systems, kinetic turbulence, and plasma physics. It deserves a serious referee; the request should be for a revision that tempers the abstract and addresses seed dependence.","headline":"A concrete one-loop MSR closure for VRR that improves on bare DIA and produces nonzero skewness where bare order vanishes; the main caveat is the un-tested seed independence of the fixed point.","tokens_in":35429,"tokens_out":3137,"would_cite":true,"duration_ms":30416,"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":"The one-loop MSR closure predicts isotropic VRR two- and three-point correlations, cutting the two-point error by 80 percent and matching N-body skewness within 4 to 20 percent.","keywords":["vector resonant relaxation","statistical closure theory","MSR formalism","one-loop renormalisation","Direct Interaction Approximation","two-point correlation","three-point skewness","fixed-point iteration"],"falsifier":"Run the same one-loop fixed-point iteration starting from several distinct initial propagators—the bare propagator, the Gaussian propagator with coherence time $T_c$, and a perturbed Gaussian—and compare the converged $[G,\\Gamma]$ and predicted $C_\\ell$; if the final correlation functions differ by more than numerical tolerance, the claimed one-loop prediction is not unique. Alternatively, measure the renormalised vertex directly in N-body simulations by inverting the measured two- and three-point propagators and compare its structure to the predicted $\\Gamma$.","tokens_in":34470,"feed_emoji":"🌀","tokens_out":11741,"duration_ms":88827,"temperature":0.7,"pith_summary":"This paper tries to establish that the MSR statistical closure scheme, taken one order beyond the bare Direct Interaction Approximation, accurately predicts correlation functions in a fully non-linear long-range system: vector resonant relaxation, the reorientation of stellar orbital planes around a supermassive black hole. The authors implement the one-loop MSR equations with a fixed-point iteration and compare the predicted two-point two-time correlation, renormalised three-point vertex, and three-point skewness against direct N-body simulations. Their central finding is that the one-loop prediction improves the two-point correlation by about 80 percent over the bare DIA, and produces non-zero skewness within 4 to 20 percent of simulations for triangles where the bare order predicts exactly zero. A sympathetic reader would care because this system has no linear term and no external driving, so it puts closure theory to the test in the strongly non-linear regime where standard perturbation theory fails.","feed_headline":"One-loop closure cuts correlation error by 80 percent","feed_subtitle":"Self-consistent vertex renormalisation matches N-body skewness within 4 to 20 percent.","key_machinery":"The load-bearing object is the renormalised three-point interaction vertex $\\Gamma_{123}$, defined by $\\Gamma_{123}=G^{-1}_{11'}G^{-1}_{22'}G^{-1}_{33'}G_{1'2'3'}$; it is the target of the closure and a proxy for the skewness. The closure is supplied by the one-loop relation $\\Gamma_{123}=\\gamma_{123}+(\\delta\\Sigma_{12}/\\delta G_{45})G_{44'}G_{55'}\\Gamma_{34'5'}$ together with the Dyson equation $G^{-1}=g^{-1}-\\Sigma$, with self-energy $\\Sigma_{12}=\\tfrac12\\gamma_{134}G_{33'}G_{44'}\\Gamma_{23'4'}$. The numerical scheme iterates these coupled equations for the pair $[G,\\Gamma]$ until convergence, exploiting isotropy, time stationarity, the fluctuation-dissipation theorem, and angular-momentum contraction rules to keep the cost manageable; the auxiliary tensor $\\Lambda_{123}=\\Gamma_{123'}G_{3'3}$ reduces the dominant one-loop evaluation from $O(N_{\\rm STEPS}^8L_{\\rm MAX}^6)$ to $O(N_{\\rm STEPS}^5L_{\\rm MAX}^6)$ operations.","core_discovery":"The paper's central claim is that the self-consistent one-loop MSR closure, in which the renormalised three-point vertex $\\Gamma$ and the dressed two-point propagator $G$ are solved together, gives quantitatively accurate predictions for isotropic vector resonant relaxation. Concretely, for harmonic $\\ell=4$ the maximum absolute error of the two-point correlation drops from about 0.15 at bare DIA order to about 0.03 at one-loop order, and the one-loop prediction recovers the late-time exponential decay and the sign change of the correlation. For triangles $(\\ell_1,\\ell_2,\\ell_3)$ whose bare vertex, and hence bare skewness, vanishes, the one-loop renormalisation generates a non-zero skewness that matches N-body measurements within roughly 4 to 20 percent, with the $(1,3,3)$ triangle underestimated by about 20 percent. The same scheme shows that expanding the vertex in powers of the bare coupling diverges at late times, while the self-consistent closure in terms of $\\Gamma$ converges.","pith_inferences":["Because the governing equation $\\partial_t\\varphi=\\tfrac12\\gamma\\varphi\\varphi$ appears in many settings, including plasmas, decaying turbulence, and cosmological structure formation, the validated one-loop vertex closure suggests the same self-consistent treatment could improve correlation predictions wherever no linear timescale exists.","A direct test of internal consistency would be to measure the dressed vertex $\\Gamma$ in simulations via its definition in terms of the two- and three-point propagators; the paper notes this is difficult, but such a measurement would distinguish genuine one-loop dressing from an effective fit.","The documented sensitivity to the initial propagator seed implies a uniqueness test: exploring additional seeds, including a delayed Gaussian or a vertex initialised away from the bare value, could expose fixed-point branching not visible in the ITER=10 runs reported here.","The cost scaling suggests that solving the MSR equations in Fourier time, or replacing the vertex expansion with a regulator-based flow, are natural next steps; if either works, two-loop predictions could become feasible and would test whether the residual 20 percent underestimate in the $(1,3,3)$ triangle shrinks."],"forward_implications":["The one-loop MSR closure reduces the maximum absolute error of the two-point correlation for $\\ell=4$ from about 0.15 to about 0.03, an 80 percent improvement over the bare DIA.","For odd triangles whose bare skewness vanishes, the one-loop scheme predicts a non-zero skewness that agrees with N-body measurements within 4 to 20 percent, so the closure captures weak non-Gaussianities generated purely by renormalisation.","The self-consistent one-loop vertex expansion converges at late times, while expanding the same diagram in powers of the bare vertex diverges; convergence requires dressing the vertex in terms of itself.","The renormalised vertex stays close to the bare vertex, with dressing amplitudes of order $10^{-4}$ relative to the bare maximum, indicating that the closure corrections are small perturbative improvements rather than large resummations.","The same fixed-point scheme formally extends to two-loop order, but the estimated cost, about $10^{10}$ hours on 128 cores for the chosen parameters, makes two-loop predictions impractical without new numerical methods."],"supporting_citations":[{"why":"It supplies the MSR functional equations, including the one-loop vertex relation and the Dyson equation, that the paper solves.","marker":"[30]"},{"why":"It defines the VRR model and the bare DIA prediction that the one-loop result improves upon, including the earlier naive one-loop equation.","marker":"[31]"},{"why":"It provides the Direct Interaction Approximation, the bare-order closure used as the comparison baseline.","marker":"[25]"},{"why":"It gives the VRR Hamiltonian, the coherence time $T_c$, and the fluctuation-dissipation relation used for initial conditions and the Gaussian seed.","marker":"[51]"},{"why":"It provides the angular-momentum contraction rules that preserve isotropy and make the diagram evaluations tractable.","marker":"[76]"},{"why":"It gives the direct-substitution inversion for triangular Toeplitz matrices used to recover the response function from the Dyson equation.","marker":"[77]"}],"fun_headline_variants":["One-loop closure cuts correlation error 80% in stellar dynamics","Self-consistent vertex renormalisation matches N-body skewness","Closure theory: 80% lower error for black-hole orbital alignment","Fivefold error reduction with one-loop closure in stellar dynamics","N-body skewness predicted within 20% by one-loop closure"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The fixed-point iteration is assumed to converge to a unique physical solution, but the paper only demonstrates convergence from one Gaussian-seeded initial condition; if multiple fixed points exist, the one-loop prediction depends on the starting guess.","fun_headline_variants_meta":{"raw":{"variants":["One-loop closure cuts correlation error 80% in stellar dynamics","Self-consistent vertex renormalisation matches N-body skewness","Closure theory: 80% lower error for black-hole orbital alignment","Fivefold error reduction with one-loop closure in stellar dynamics","N-body skewness predicted within 20% by one-loop closure"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000916,"raw_usage":{"total_tokens":3928,"prompt_tokens":935,"completion_tokens":2993,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":551,"completion_tokens_details":{"reasoning_tokens":2905}},"tokens_in":551,"tokens_out":2993,"duration_ms":19665,"temperature":1.0,"reasoning_tokens":2905,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:44:29.415255+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same one-loop fixed-point iteration starting from several distinct initial propagators—the bare propagator, the Gaussian propagator with coherence time $T_c$, and a perturbed Gaussian—and compare the converged $[G,\\Gamma]$ and predicted $C_\\ell$; if the final correlation functions differ by more than numerical tolerance, the claimed one-loop prediction is not unique. Alternatively, measure the renormalised vertex directly in N-body simulations by inverting the measured two- and three-point propagators and compare its structure to the predicted $\\Gamma$.","supporting_citations":[{"cited_title":"Yokoyama and M","cited_arxiv_id":null,"evidence_quote":"It supplies the MSR functional equations, including the one-loop vertex relation and the Dyson equation, that the paper solves."},{"cited_title":"Tarpin, L","cited_arxiv_id":null,"evidence_quote":"It defines the VRR model and the bare DIA prediction that the one-loop result improves upon, including the earlier naive one-loop equation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the Direct Interaction Approximation, the bare-order closure used as the comparison baseline."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the VRR Hamiltonian, the coherence time $T_c$, and the fluctuation-dissipation relation used for initial conditions and the Gaussian seed."},{"cited_title":"Goldreich and S","cited_arxiv_id":null,"evidence_quote":"It gives the direct-substitution inversion for triangular Toeplitz matrices used to recover the response function from the Dyson equation."}],"review_version":2}