{"id":"b8fadc58-a854-4e2a-b957-e57642769c54","arxiv_id":"2506.15579","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The authors derive and benchmark Langevin-based thermostats, introducing BPCL, a Maxwell-element colored thermostat, and a momentum-conserving bin thermostat that preserves long-wavelength dynamics.","lead":"This paper analyzes stochastic thermostats for molecular dynamics and proposes three refinements: a 'best-possible' conventional Langevin thermostat, a colored thermostat built from Maxwell elements, and a lean momentum-conserving thermostat. The authors argue that Grønbech-Jensen type thermostats allow larger time steps and more accurate thermal sampling than common global velocity-rescaling schemes, which matters for anyone simulating materials or fluids.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"MCL's bulk 3D claim rests only on an 8-atom 1D harmonic chain; random-origin bins and atom exchange are never tested, so the hydrodynamic-preservation claim is unverified.","rationale":"The reader's CONDITIONAL verdict is appropriate, and my concern is the second part of the reader's weakest_assumption stated more sharply. The BPCL optimality issue is also real: the sentence before Eq. (19) asserts that no conventional Langevin thermostat can have a smaller leading error than omega0^2 dt^2/4, but no lower-bound proof is given. I do not make that the primary concern because it mainly affects the 'best-possible' branding; BPCL would still be a serviceable thermostat even if a smaller leading coefficient existed. MCL is different. The central novelty of the paper, and its most consequential claim, is that MCL preserves long-wavelength modes. That claim is inherently about bulk 3D behavior, yet the only evidence is a 1D 8-atom harmonic chain at very low temperature. In 3D, random-origin bins and inter-bin atom exchange change the thermostatted degrees of freedom from step to step, so the fixed-frame BPCL derivation does not transfer automatically. The paper's own text concedes this risk and even drops the GJ version of MCL for this reason. A 3D test of both static averages and small-q dynamics would settle the issue. I also note that the numerical comparisons in Figs. 6-9 lack error bars; this is a secondary weakness that the same 3D run could partially address. Credit is due for the public code repository and for the relatively transparent asymptotic derivations; those are not in question.","tokens_in":22527,"tokens_out":12378,"duration_ms":128794,"concrete_test":"Run MCL exactly as given in the Sect. II E pseudo-code in a 3D Lennard-Jones liquid (N about 4000, density about 0.8 sigma^-3, kBT=2U0/3, bin size about 2 sigma, random origin each step), comparing against GJL and NHC at identical tau and Delta-t. First, check the mean potential energy per atom against the paper's own acceptance band +/- D kBT/400 with D=3. Second, compute the longitudinal current autocorrelation C_l(q_min,t) at q_min=2*pi/L and compare its decay with microcanonical MD and with a lab-frame Langevin thermostat at the same tau. If MCL's energy error exceeds the band, or if C_l is damped as strongly as in the lab-frame case, the MCL claims of correct bulk sampling and preserved hydrodynamic modes fail. A static pass alone would not establish the hydrodynamic claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II E proposes MCL and the conclusions claim it 'preserves long-wavelength vibrational modes and hydrodynamic interactions, avoiding the overdamping artifacts introduced by thermostats acting in a fixed frame of reference.' The only demonstration is Fig. 10: a one-dimensional Lennard-Jones chain with N=8 atoms at kBT=0.001 epsilon_LJ, with the damping time set to twice the largest period so every mode is underdamped. The bulk algorithm, however, uses random-origin spatial bins and atoms that cross bin boundaries, and it is never simulated. This is not a minor gap: the BPCL coefficients in Eqs. (4), (21), and (24)-(26) were derived for a single particle in a fixed inertial frame, whereas MCL thermostats velocities relative to a bin center of mass that is redefined every step, with membership changing as atoms cross boundaries. The paper itself notes that the accuracy of advanced thermostats 'can be decremented' in this setting and that the GJ variant was abandoned for this reason. Nothing in the submitted data shows that BPCL-based MCL retains correct thermal averages or the q-squared damping scaling in three dimensions. The headline MCL claim is therefore unverified, not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops and tests several Langevin-based thermostats. It introduces a generic form for conventional Langevin solvers, derives the lowest-order Langevin (LOL) and 'best-possible conventional Langevin' (BPCL) coefficients, discusses the Grønbech-Jensen (GJ) scheme and an advanced Brownian thermostat, proposes a Maxwell-element version called GJFM, and introduces a momentum-conserving Langevin (MCL) thermostat based on spatial bins. The numerical sections compare mean potential energy errors for a harmonic oscillator, a sinusoidal potential, a one-dimensional Lennard-Jones chain, bulk liquid Lennard-Jonesium, and liquid copper, and show a single one-dimensional chain demonstration for MCL. The central claims are that BPCL is the optimal conventional Langevin scheme, that GJ reproduces exact harmonic-reference averages and remains accurate for anharmonic systems, and that MCL preserves long-wavelength vibrational modes and hydrodynamic interactions.","tokens_in":22697,"tokens_out":5809,"duration_ms":57781,"significance":"The paper has genuine strengths: the asymptotic derivations in Section II are compact and mostly checkable, the code and data are made publicly available, and the paper usefully contrasts global kinetic-energy controls with local Langevin thermostats. If the BPCL optimality claim and the GJ harmonic-exactness claim were fully proven, and if MCL were validated in bulk systems, the work would provide practitioners with a practical comparison of thermostat algorithms and with new thermostat options. However, as it stands, the two most distinctive claims—the 'best-possible' property of BPCL and the hydrodynamic-preservation claim of MCL—rest on an unproved lower bound and on a single one-dimensional 8-atom simulation, respectively. The significance of the paper is therefore conditional on closing those gaps.","major_comments":[{"comment":"The statement that no conventional Langevin thermostat can have a systematically smaller relative leading error in the mean potential energy of a harmonic oscillator than omega_0^2 dt^2/4 is asserted without proof. Equation (14) expresses <x^2> in terms of the four propagation coefficients, but no optimization over the allowed c_vv, c_vf, c_vg, c_xv parameter space is performed. Since this lower bound is what justifies the name 'best-possible conventional Langevin' and is quoted in the conclusions, it needs a rigorous derivation, or the claim should be weakened to a statement about the specific scheme proposed.","section":"Section II.B.1, before Eq. (19)"},{"comment":"The paper acknowledges that the coefficient choices make Eq. (41) a 'self-fulfilling prophecy,' and then states that exactness can be confirmed by evaluating all possible covariances, but that evaluation is not shown. The abstract and conclusions nonetheless claim that GJ is exact for harmonic references. As written this is a circular validation; a complete closed-form verification of the second-moment relations, or a reference to a full proof, is required before the exactness claim can be considered established.","section":"Section II.C.2, Eqs. (41)-(45) and the passage after Eq. (45)"},{"comment":"The MCL algorithm is introduced for spatial bins with random origins and atoms crossing bin boundaries, but the only simulation presented is a one-dimensional 8-atom Lennard-Jones chain at k_B T = 0.001 epsilon_LJ with damping set to twice the largest period. The conclusions claim that MCL 'preserves long-wavelength vibrational modes and hydrodynamic interactions' in general, but the bulk three-dimensional algorithm is never simulated. The paper itself notes that 'the accuracy of advanced thermostats can be decremented' in this setting and that the GJ variant was abandoned for this reason; therefore the central MCL claim is currently unverified.","section":"Section II.E and Fig. 10"},{"comment":"The numerical comparisons of mean potential energy are presented without statistical error bars or confidence intervals, even though Section II.H mentions multiple replicas for on-the-fly error estimation. Without such information, quantitative claims such as the factor-of-80 reduction in leading-order error in Fig. 5 and the factor-of-9 time-step advantage cannot be distinguished from statistical noise.","section":"Figures 5-9"}],"minor_comments":[{"comment":"The relation c_vf = c_xv/m appears twice, once as Eq. (44) and again in the line after Eq. (45); this duplication obscures the derivation of c_xv and should be cleaned up.","section":"Section II.C.2, Eqs. (44)-(45)"},{"comment":"The caption contains the misspelling 'momentem-conserving'; it should read 'momentum-conserving.'","section":"Figure 10 caption"},{"comment":"The symbol T is used both for the oscillation period and for temperature; although the text warns the reader about this convention, using a different symbol for one of the two quantities would avoid confusion.","section":"Section III.A"},{"comment":"The potential is written as U(x) = -U_0 cos(qx), but the text immediately afterward uses kappa = q^2 |V_0|; the notation for the potential amplitude should be harmonized.","section":"Section III.C, Eq. (65)"},{"comment":"The pseudo-code line 'g = g - mean(g)' is ambiguous because it is not specified whether the mean is subtracted per Cartesian component; this should be stated explicitly.","section":"Section II.E, pseudo-code"},{"comment":"The parenthetical remark about sex-based differences in voice pitch is not connected to the technical content of the paper and should be removed or placed in a clearly separate discussion.","section":"Introduction"}],"recommendation":"major_revision","confidential_remarks":"The paper is a methods-development and benchmarking study written in good faith, and it contains useful derivations and public code. The main obstacles are the gap between the advertised 'best-possible' and 'preserves hydrodynamic interactions' claims and the demonstrated results. I would not recommend rejection: with a rigorous treatment of the BPCL optimality claim, a complete proof or citation for the GJ harmonic-exactness property, and at least one bulk three-dimensional MCL test with error bars, the paper could become a solid contribution. One additional concern is that the proliferation of new acronyms (BPCL, MCL, GJFM) may overstate the novelty relative to existing known integrators; the authors should position these as parameterizations of previously proposed schemes where appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read arXiv:2506.15579. Short version: worth taking seriously. The new thermostat variants are real, the derivations are more transparent than the usual propagator treatments, and the code is public. But the 'best-possible' label is not actually proven, and the momentum-conserving thermostat's flagship claim about hydrodynamic modes is unverified outside a single 8-atom 1D test.\n\nWhat's genuinely new: BPCL, a four-coefficient conventional Langevin scheme with no memory of random numbers; GJFM, which attaches GJF Brownian thermostats to Maxwell elements; and MCL, a bin-based momentum-conserving thermostat built on BPCL. The paper also gives compact asymptotic derivations of the known GJ and GJF schemes that many readers will find easier to follow than the standard operator-splitting versions. The numerical comparisons against CSVR, NHC, and LOL on the LJ chain, 3D LJ liquid, and liquid copper are useful; the claim that GJ allows 5-10 fs steps in copper at 1400 K is plausible and practically valuable. The public GitHub repository with codes and LAMMPS inputs counts for a lot.\n\nNow the soft spots.\n\nFirst, the 'best-possible' claim. The assertion just before Eq. (19) - no conventional Langevin thermostat can have systematically smaller leading relative error in the mean potential energy of a harmonic oscillator than omega0^2 dt^2/4 - is given without proof or reference. That is load-bearing. If a competitor scheme had a smaller leading coefficient, the paper's central branding collapses. The authors should supply the lower-bound argument or rename the scheme. The GJ derivation, in contrast, is properly honest: they state that the moment requirements are a self-fulfilling prophecy and then check consistency. Fine.\n\nSecond, the MCL bulk claim. The conclusions say MCL 'preserves long-wavelength vibrational modes and hydrodynamic interactions, avoiding the overdamping artifacts introduced by thermostats acting in a fixed frame of reference.' The only demonstration is Fig. 10: an 8-atom 1D LJ chain at kBT=0.001 epsilon, with damping set to twice the largest period. The algorithm as described uses random-origin spatial bins redefined every step, with atoms crossing bin boundaries. None of that is tested in a bulk 3D system. The paper itself notes that the accuracy of advanced thermostats 'can be decremented' here, and says GJ was abandoned for this reason. So the MCL headline claim is unverified, not established. That is the biggest gap, and it is fixable: run MCL on a 3D liquid, report center-of-mass conservation and the q-dependence of damping.\n\nThird, Figs. 5-9 have no error bars. The text mentions parallel replicas for error estimation, but the figures show single-point estimates. For the 3D LJ results, where the GJ advantage shrinks, that matters.\n\nAlso, the statement that GJFM is 'exact for harmonic references' is asserted without a full covariance check of the coupled system; that needs a few lines of derivation.\n\nBottom line: a solid, carefully written methods paper with real contributions and honest self-assessment. It deserves a serious referee. I'd require the BPCL proof-or-rename, a 3D MCL demo, and error bars before acceptance. I'd cite BPCL and GJFM, not the MCL claim.","headline":"Useful thermostat variants with transparent derivations and public code; the 'best-possible' branding is unproved and MCL's hydrodynamic claim rests on a single 8-atom 1D chain.","tokens_in":23308,"tokens_out":4864,"would_cite":true,"duration_ms":38465,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82C31","82C80","65C30"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that a recycled-random-number Langevin integrator gives exact harmonic thermal averages at two-to-ten-fold larger time steps, while a bin-based momentum-conserving Langevin scheme preserves long-wavelength modes.","keywords":["Langevin thermostat","molecular dynamics","thermal averages","harmonic oscillator","momentum-conserving thermostat","GJ Langevin scheme","stochastic integrators","hydrodynamic modes"],"falsifier":"Scan the full four-parameter space of the conventional update (or solve its covariance equations symbolically) for a harmonic oscillator and check whether any memory-free coefficient set yields a mean potential-energy error below $\\omega_0^2\\Delta t^2/4$; a counterexample refutes BPCL's optimality. For MCL, run the scheme in a three-dimensional Lennard-Jones liquid with random bin origins and measure the damping of the longest-wavelength mode; if it follows the $q$-dependence of laboratory-frame damping rather than the momentum-conserving prediction, the central MCL claim fails.","tokens_in":22246,"feed_emoji":"⚛️","tokens_out":10498,"duration_ms":99746,"temperature":0.7,"pith_summary":"The paper argues that thermostats which rescale every velocity by a common factor—global kinetic-energy controls—cannot be true canonical samplers, because degenerate, weakly coupled vibrational modes stay phase-locked forever, corrupting quantities such as elastic response. Its counter-proposal is a family of Langevin-equation solvers whose update coefficients are fixed by matching exact limits: a memory-free “best-possible conventional Langevin” (BPCL) scheme, and the GJ scheme, which recycles a Gaussian random number from the previous step and thereby reproduces harmonic-oscillator thermal averages exactly. The paper also introduces a lean momentum-conserving Langevin (MCL) thermostat, which damps only motion relative to a small local center of mass, leaving long-wavelength, hydrodynamic modes underdamped. If these claims are right, standard molecular-dynamics practice can shift from global rescaling toward these Langevin solvers, gaining time steps two to ten times larger while keeping mean potential-energy errors within 0.0025 $k_B T$ per degree of freedom.","feed_headline":"Langevin schemes keep MD accurate at 10x time steps","feed_subtitle":"Global rescaling freezes weakly coupled modes; the paper's GJ and MCL schemes keep dynamics physical.","key_machinery":"The machinery is the four-coefficient leapfrog update $v_{n+1/2}=c_{vv}v_{n-1/2}+c_{vf}f_n+c_{vg}g_n$, $x_{n+1}=x_n+c_{xv}v_{n+1/2}$, with coefficients fixed by exact limits. BPCL chooses $c_{vv}=\\exp(-\\Delta t/\\tau)$ and enforces exact constant-force drift, exact equipartition, and exact kinetic energy for a force-free particle. The GJ variant changes the noise term from one fresh Gaussian per step to a fresh plus a recycled Gaussian, which closes every harmonic covariance identity and makes the oscillator distribution exact; BPCL is the best possible version without that memory. MCL applies BPCL to coordinates measured relative to each bin's center of mass and advances the center of mass symplectically, so only relative motion is damped and noisy. The job of this machinery is to push the leading error in the harmonic mean potential energy down to $O(\\Delta t^2)$ and, for GJ, to remove it entirely.","core_discovery":"The central claim is that the GJ scheme is the most accurate Langevin-type thermostat for harmonic and near-harmonic systems, yielding correct thermal averages across a broad range of damping constants and time steps, and that the new MCL thermostat preserves long-wavelength vibrational modes and hydrodynamic interactions, avoiding the overdamping artifacts of thermostats acting in a fixed laboratory frame. The derivation proceeds by writing the integrator as a four-coefficient update and fixing those coefficients through asymptotic requirements: correct free-particle decay, correct drift under a constant force, equipartition, and exact kinetic energy for a force-free particle. Adding a second, recycled random number closes all covariance equations for a harmonic reference, which is why GJ reproduces the exact second and fourth moments of the oscillator while BPCL and simpler schemes carry a leading error of $\\omega_0^2\\Delta t^2/4$ in mean potential energy. Across single-sinusoidal potentials, one- and three-dimensional Lennard-Jones liquids, and liquid copper, the paper reports energy errors below $k_B T/400$ per degree of freedom at time steps two to ten times larger than conventional stability limits.","pith_inferences":["A decisive bulk test not yet performed is MCL in a three-dimensional Lennard-Jones liquid with random bin origins; if the long-wavelength damping follows the derived $q$-dependence, MCL becomes a cheap stand-in for dissipative particle dynamics.","Because MCL conserves linear but not angular momentum, it will add some viscosity; a natural next measurement is the excess viscosity as a function of bin size and damping, with bin masses tuned to offset it.","The asserted lower bound on conventional Langevin schemes is stated without proof; a direct search over the four propagation coefficients for a harmonic oscillator would settle whether the bound is true.","The same coefficient-fixing strategy could be applied to auxiliary variables such as cell-shape degrees of freedom, where a Brownian-driven Maxwell element may give smoother equilibration than direct Langevin damping."],"forward_implications":["Molecular dynamics of stiff materials can run at time steps of 5–10 fs for copper at 1400 K with the GJ scheme, compared with the usual 1–2 fs, while keeping mean potential-energy errors below $0.0025 k_B T$ per atom.","Global velocity-rescaling thermostats should not be used for response functions built from weakly coupled quasi-harmonic variables, such as box-shape fluctuations used to extract elastic constants.","Applying the paper's additive correction $U_{\\mathrm{BPCL}^*}=U_{\\mathrm{BPCL}}+k_BT/2-\\langle T_{\\mathrm{kin}}\\rangle$ doubles the usable time step of BPCL for a fixed accuracy target.","In near-harmonic systems, a GJ run that stays numerically stable can be trusted as accurate, because the range of time steps where the scheme is stable but inaccurate is extremely narrow or absent.","The MCL thermostat lets long-wavelength modes in large homogeneous systems keep oscillating, which makes it a candidate for equilibrium and nonequilibrium runs that need hydrodynamic interactions intact."],"supporting_citations":[{"why":"Supplies the GJ scheme, the recycled-random-number Langevin solver whose exact harmonic sampling is the paper's main benchmark and extension.","marker":"[9]"},{"why":"Provides the advanced Brownian integrator that is reused to thermostat Maxwell elements in the proposed GJFM scheme.","marker":"[31]"},{"why":"Introduces colored-noise thermostats with memory, which the paper extends by exact harmonic averaging.","marker":"[13]"},{"why":"Defines dissipative particle dynamics, the momentum-conserving scheme the paper's MCL thermostat is designed to beat on cost.","marker":"[10]"},{"why":"Is the stochastic velocity-rescaling scheme treated as the leading global kinetic-energy control whose canonical claims the paper disputes.","marker":"[16]"}],"fun_headline_variants":["Exact moments, 10x steps: GJ thermostat","MCL thermostat preserves long-wavelength modes","Langevin thermostats beat global rescaling at 10x steps","Recycled noise gives exact covariances for harmonic systems","Lean momentum-conserving thermostat avoids overdamping"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that no memory-free Langevin solver can beat the leading error $\\omega_0^2\\Delta t^2/4$ in the harmonic-oscillator mean potential energy; if a competitor did, the claim that BPCL is the best conventional Langevin thermostat collapses, and the MCL branch separately assumes that a one-dimensional eight-atom chain at very low temperature is a reliable guide to bulk three-dimensional bins.","fun_headline_variants_meta":{"raw":{"variants":["Exact moments, 10x steps: GJ thermostat","MCL thermostat preserves long-wavelength modes","Langevin thermostats beat global rescaling at 10x steps","Recycled noise gives exact covariances for harmonic systems","Lean momentum-conserving thermostat avoids overdamping"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001047,"raw_usage":{"total_tokens":4361,"prompt_tokens":869,"completion_tokens":3492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":3411}},"tokens_in":485,"tokens_out":3492,"duration_ms":24656,"temperature":1.0,"reasoning_tokens":3411,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:32:27.301171+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Scan the full four-parameter space of the conventional update (or solve its covariance equations symbolically) for a harmonic oscillator and check whether any memory-free coefficient set yields a mean potential-energy error below $\\omega_0^2\\Delta t^2/4$; a counterexample refutes BPCL's optimality. For MCL, run the scheme in a three-dimensional Lennard-Jones liquid with random bin origins and measure the damping of the longest-wavelength mode; if it follows the $q$-dependence of laboratory-frame damping rather than the momentum-conserving prediction, the central MCL claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the GJ scheme, the recycled-random-number Langevin solver whose exact harmonic sampling is the paper's main benchmark and extension."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the advanced Brownian integrator that is reused to thermostat Maxwell elements in the proposed GJFM scheme."},{"cited_title":"Grønbech-Jensen, Complete set of stochastic Verlet- type thermostats for correct Langevin simulations, Molecular Physics118, e1662506 (2019)","cited_arxiv_id":null,"evidence_quote":"Introduces colored-noise thermostats with memory, which the paper extends by exact harmonic averaging."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines dissipative particle dynamics, the momentum-conserving scheme the paper's MCL thermostat is designed to beat on cost."},{"cited_title":"Soddemann, B","cited_arxiv_id":null,"evidence_quote":"Is the stochastic velocity-rescaling scheme treated as the leading global kinetic-energy control whose canonical claims the paper disputes."}],"review_version":2}