{"id":"aba2e286-4dfe-46ee-b677-686101733589","arxiv_id":"2607.24574","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"Near the AT/Δ boundary the AdResS thermodynamic force equals the missing average force through the Δ/TR interface, given by a one-dimensional integral from the open-system Liouville hierarchy.","lead":"A large part of the corrective force used in adaptive-resolution molecular simulations is the average force missing across the artificial interface, derived from open-system Liouville theory. That link lets the force be computed in a much smaller box and checked with interface-only tests.","discovery_kind":"extension","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The empirical support for F_th = F_av near AT/Δ is partly seeded: F_av was the iteration's initial guess, and the agreement region is where F_av decays to zero; the decisive zero-initial-guess control is asserted in one sentence without data.","rationale":"I agree with the reader that the soft spot is Eq. (30) — whether ˆF fades before the AT/Δ boundary — but I locate the weakness one level deeper: the evidence offered for that fade is partially circular. Because F_av seeds the iteration and the density-flatness convergence criterion means the correction is minimal where the density is already flat, agreement between converged F_th and F_av near AT/Δ is close to a property of the initialization unless the zero-guess control is demonstrated. The paper asserts that control in a single sentence; that assertion, if documented, would largely settle the matter, which is why this is CONDITIONAL rather than REJECT. Mitigating factors: the derivation itself is sound (I verified the angular integration in Eq. 22 and the potential/force consistency of Eq. 27); the full-AT validation of the average force through the AT/∆ boundary (Fig. 7, third row) independently confirms that interface artifacts do not reach the boundary statistics, which supports the spirit of Eq. (30) even if not the pointwise identification; and the comparison across three thermodynamic states and three box geometries would be unlikely to agree under seeding artifacts alone. The parameter-independence claims for L_y,z and L_AT are directly tested and convincing. What is missing is quantification: an error norm for F_th−F_av in the fade region and a shown, zero-seeded convergence run. This is a one-afternoon computational check with the authors' existing public MRMD setup. Until it is shown, the central identification rests on an assertion rather than demonstrated evidence, so I would move ACCEPT to CONDITIONAL, with the condition being exactly the concrete test above.","tokens_in":24362,"tokens_out":5508,"duration_ms":92974,"concrete_test":"Re-run the 7-iteration procedure for all three states with F^0_th ≡ 0 and with F^0_th = ½F_av, keeping all else fixed. Plot each converged F_th against πρ(r_∂²I_0−I_2) over x_AT/Δ ≤ x ≤ x_Δ/TR−1σ and report max- and L2-norm of F_th−F_av relative to peak |F_th| and to the statistical noise floor of the binned force. If the zero-guess result converges to F_av within noise, Eq. (30) is independently supported; if it deviates by >~10% of peak in the fade region, the Fig. 6 agreement was seeded and the identification weakens.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The analytical derivation (Eqs. 7–27) checks out: for a bulk homogeneous simple fluid at a fictitious planar boundary, ρ^(2)=ρ²g(r) is exact, the angular integration in Eq. (22) is correct, and Eq. (27) is consistent with F=+dV/dr_∂. The load-bearing step is therefore empirical: Eq. (30), F_th(x)=F^x_av(x_Δ/TR−x) for x_AT/Δ≤x≤x_fade, rests on the observed agreement between the converged iterated force and the a priori integral in Fig. 6. Two features weaken that observation as evidence. (1) Seeding: F_av (via Goldman's g(r)) was used as F^0_th in the iteration of Eq. (3). Near AT/Δ the density produced by that guess is already flat, so the correction −c∇ρ is small there and the iteration has little mechanism to move F_th away from F_av in exactly the region where agreement is claimed. The paper states that iterations without the initial guess \"converged to the same result,\" but this is one sentence in §IVC with no figure, no quantification, and no statement of the x-range or tolerance. (2) Comparison band: the agreed region is where F_av is decaying to zero (F_av(x_AT/Δ)=0 because L_Δ=r_cut; from Fig. 5 the force there is ≲0.5ϵ/σ versus a ~2ϵ/σ peak near Δ/TR). Two small, smooth, monotonically decaying functions can agree \"very well\" while hiding a residual ˆF of comparable local magnitude; no error norm is reported. If the converged F_th depends measurably on the initial guess, or if the agreement tolerance is loose relative to local force magnitude, the claim that ˆF fades before x_fade — the reader's flagged assumption — loses its only empirical support, and the independence claims (γ, r_cap, L_AT, L_y,z) inherit the gap.","agreement_with_reader":"partial"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript reinterprets the thermodynamic (coupling) force of adaptive resolution simulations (AdResS) through the Liouville-type hierarchy for open systems. Starting from the marginalized Liouville equation, the authors identify the average force through the open system's boundary (Eq. 7), and for a simple homogeneous fluid with a planar boundary reduce it to a one-dimensional integral, F^x_av(r_∂) = πρ(r_∂²I_0 − I_2) (Eqs. 23–25), with a corresponding potential (Eq. 27). They then argue that in AdResS the thermodynamic force must reintroduce precisely this missing average force through the Δ/TR interface, and hypothesize (Eq. 30) that F_th(x) = F^x_av(x_{Δ/TR} − x) on [x_{AT/Δ}, x_fade], up to a residual F̂(x;γ,r_cap) assumed to fade within Δ. This yields parameter-dependence predictions (Tab. II), a proposed 'minimal' AdResS setup without an AT region for computing F_th, and new interface-local validation criteria (boundary density, Maxwell-Boltzmann momenta, particle-crossing statistics, average force through AT/Δ). Simulations of supercritical LJ fluids at three states with three box setups show that converged F_th from minimal/no-AT/complete boxes collapse, agree with the a priori F_av near AT/Δ, and pass the proposed validations against a full-AT reference. The analytical derivation is transparent and, in my reading, correct: ρ⁽²⁾=ρ²g(r) is exact in the bulk, the angular integration in Eq. (22) is right, and Eq. (27) is consistent with F=+dV/dr_∂. The load-bearing,","tokens_in":24802,"tokens_out":3942,"duration_ms":131651,"significance":"If the central identification holds, the paper provides (i) a closed one-dimensional integral (Eq. 25) for the missing average force, derived without free parameters from g(r) and V, which can serve as an a priori initial guess for the thermodynamic-force iteration; (ii) testable parameter-independence predictions (Tab. II) with practical value — notably the demonstration that a 'minimal' AdResS box with no AT region suffices to compute the thermodynamic force and reproduce correct boundary statistics; and (iii) new validation criteria tied to the reservoir-coupling terms of the hierarchy rather than bulk AT-region observables. The work also supplies reproducible code: simulation scripts in a public MRMD fork with pinned commit, plus the GhostFace and MRMDAnalysis packages with pinned commits. The conceptual link it draws — the thermodynamic force as the mean-field reintroduction of the average force through the Δ/TR interface — is a useful reframing that clarifies which artifacts the force must and must not compensate. These are solid contributions to the AdResS/open-systems literature provided the empirical identification is properly controlled.","major_comments":[{"comment":"Eq. (30), §IIIC and §IVC/Fig. 6: the central empirical claim F_th(x)=F_av(x_{Δ/TR}−x) near AT/Δ is supported by agreement between the converged iterated force and the a priori integral — but F_av (via Goldman's g(r)) was the iteration's initial guess F^0_th in Eq. (3). Near AT/Δ that guess already produces a flat density, so the correction −c∇ρ is small precisely where agreement is claimed, giving the iteration little mechanism to move away from F_av there. The decisive control — iterations started without the initial guess — is asserted in one sentence in §IVC ('converged to the same result') with no figure, no x-range, and no tolerance. This control exists (the runs were performed) and should be shown. Note that Fig. 6 does show the iteration moving substantially away from the initial guess within 1σ of Δ/TR, so the iteration is demonstrably not inert; a quantitative comparison of seed","section":"§IIIC, Eq. (30); §IVC, Fig. 6"},{"comment":"Related to the above: no error norm is reported for the agreement in the region x_AT/Δ ≤ x ≤ x_fade. From Fig. 5, F_av at the AT/Δ boundary is ≲0.5 ϵ/σ versus a ~2 ϵ/σ peak near Δ/TR (L_Δ = r_cut forces F_av(x_AT/Δ)=0), so the agreement band lies where both functions are small and smoothly decaying; two such functions can agree 'very well' while hiding a residual F̂ comparable to the local force magnitude. Please report a relative residual |F_th−F_av|/|F_th| (or an L2 norm) over the claimed interval, and use it to give a quantitative estimate of x_fade, which currently is never determined from data.","section":"§IVC, Figs. 5–6"},{"comment":"§IVD (and the Conclusions): the statement that 'close to the AT/∆ boundary, the only artifact left to compensate is the missing average force' is stronger than what the validation shows. The three criteria in Fig. 7 (momentum distribution, rescaled crossing statistics, measured average force through AT/Δ against the full-AT reference) establish that the *combined* action of thermostat, capping, and F_th reproduces the correct boundary statistics, i.e. Eq. (28)/(11). They do not, by themselves, establish the decomposition Eq. (29) with F̂→0. This is genuinely independent, unseeded evidence for boundary consistency and should be credited as such — but the text should distinguish it from evidence for the identification F_th=F_av itself, which rests only on the seeded comparison in Fig. 6.","section":"§IVD, Fig. 7"},{"comment":"§IIID and Table II: the parameter-independence predictions are a principal selling point, but coverage is limited to L_y,z and L_AT. Independence of γ is directly testable with the existing setup and is arguably the most consequential untested entry: the theory predicts it because F_av is momentum-independent, yet the simulations use a *global* Langevin thermostat with the rather strong γ=20τ⁻¹, and §IVD itself flags that the global thermostat may affect crossing statistics. One or two additional iteration runs at different γ (and/or r_cap) would substantially strengthen Tab. II and the claim that F̂(x;γ,r_cap) is what fades rather than something the thermostat is masking.","section":"§IIID, Table II"}],"minor_comments":[{"comment":"§IVA: the three thermodynamic states are described as '(1) low density and low temperature, (2) high density and high temperature and (3) high density and high temperature'; state (2) should presumably read 'high density and low temperature' per Table IV.","section":"§IVA"},{"comment":"Table IV and Appendix A: densities are printed as '0.296σ³'/'0.370σ³'; the exponent should be −3 (σ⁻³) as elsewhere in the manuscript.","section":"Table IV, Appendix A"},{"comment":"Table VI lists 'niter 7/10' and Appendix A states n_iter=10 with convergence at iteration 7; §IVC says the procedure was 'stopped once... sufficiently flat' at the 7th iteration. Please clarify whether iterations 8–10 were run and discarded, and state the actual numerical flatness tolerance used as the stopping criterion ('sufficiently flat' and 'prescribed tolerance' in §IIA are never quantified).","section":"Table VI, §IVC, Appendix A"},{"comment":"§IIIA, first paragraph: 'the conditional probability density in Eq. 7' should read 'Eq. (7)'; 'Morsali et. al' in Appendix A should be 'Morsali et al.'","section":"§IIIA, Appendix A"},{"comment":"Fig. 5: the x-axis is shifted by x_AT/Δ while the caption text discusses distances 'greater than 1σ away from the Δ/TR interface'; since x_{Δ/TR}−x_{AT/Δ}=2.5σ, readers must mentally re-shift. Marking the 1σ-from-Δ/TR position or stating the offset explicitly would help.","section":"Fig. 5"},{"comment":"Fig. 7, second row: the rescaling of the crossing distributions for the minimal box uses the factor L1/L2=20/30 from Eq. (B19); this is stated only implicitly. Please state the applied factor and the number of snapshots/windows entering each distribution so the reader can assess statistical uncertainty of the rescaled curves.","section":"Fig. 7, Appendix B"},{"comment":"Appendix B is a careful but lengthy derivation of the standard CLT rescaling of Gaussian distributions; it could be condensed substantially (the result Eq. (B19) is the only part used), improving readability without loss of rigor.","section":"Appendix B"},{"comment":"§IIID, item on r_cap independence: the argument that g(r)=0 in the capped repulsive region renders the integral insensitive to r_cap is plausible but is an assumption about g(r) at the simulated states; a one-line reference to a measured g(r) inside r_cap (e.g., from the equilibration runs) would close it.","section":"§IIID"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful core is straightforward: they reduce the Liouville-hierarchy boundary force for a simple planar open fluid to a single integral F_av = πρ(r_∂²I_0 − I_2), then argue that near the AT/Δ edge the AdResS thermodynamic force is essentially that missing average force through the Δ/TR cut. That gives a real a-priori initial guess, a justification for iterating the force in a no-AT (even minimal Ly,z) box, and interface-native checks (density, Maxwellian momenta, crossing fluctuations, measured F_av) instead of only bulk AT observables.\n\nThe math is standard and transparent once you accept bulk g(r) and planar geometry. The LJ runs at three supercritical states are done carefully: minimal / no-AT / complete boxes collapse on F_th near AT/Δ, match the Goldman-based guess there, and pass the full-AT interface statistics. Circularity is low—F_th is still the usual density-flattening iteration; F_av is comparison and seed, not definition. Code and protocols are public. That is enough to make the practical claims (cheaper force setup, interface validation) credible for simple fluids.\n\nSoft spots, in proportion. Eq. (30) is explicitly an assumption that residual artifacts ˆF fade inside Δ; the paper treats agreement in Fig. 6 as support, not proof. The stress-test point lands partially: they seeded the iteration with F_av, the agreement band is where F_av is already small, and the zero-guess control is one sentence with no figure or tolerance. That weakens how hard you can lean on “ˆF always fades” and on the unscanned independences (γ, r_cap, L_Δ). Density wiggles near Δ/TR never go away; only LJ; global thermostat on both AdResS and reference. None of that breaks the LJ evidence or the derivation, but generalization and the load-bearing fade assumption stay empirical and scoped.\n\nWho it is for: people who run or extend AdResS / open-boundary MD and particle–continuum coupling. Not a foundational rewrite of open-system statistical mechanics, but a solid bridge between the hierarchy papers and day-to-day force construction. I would send it to referees; the central claim is supported as stated for the systems they simulated. Worth engaging if you touch this toolkit.","headline":"Clean derivation of the open-boundary average force plus a practical minimal-box recipe for AdResS thermodynamic force; the F_th≈F_av claim near AT/Δ is useful but only partly stress-tested.","tokens_in":25327,"tokens_out":607,"would_cite":true,"duration_ms":20865,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A large part of the AdResS thermodynamic force is the missing average force from the open-system Liouville hierarchy, given by a one-dimensional integral.","keywords":["Adaptive resolution simulation","thermodynamic force","Liouville-type hierarchy","open molecular systems","average boundary force","Lennard-Jones fluid","interface artifacts"],"falsifier":"Compute the iterated thermodynamic force and the integral F_av on the same grid for a supercritical Lennard-Jones fluid; if they disagree throughout the half of the Δ region nearest the AT/Δ boundary, or if particle-crossing fluctuations and the measured average force through that boundary deviate from a full-atomistic reference, the claim is false.","tokens_in":25018,"feed_emoji":"⚗️","tokens_out":895,"duration_ms":17665,"temperature":0.7,"pith_summary":"Adaptive resolution simulations keep a fully interacting atomistic region open by surrounding it with transition and tracer zones, but the abrupt switching of interactions creates interface artifacts that must be cancelled by an external thermodynamic force. This paper shows that, near the atomistic boundary, a substantial part of that force is exactly the average force that particles would feel through the outer interface if interactions had not been turned off. Starting from the source terms of the Liouville-type hierarchy for open systems, the authors derive a one-dimensional integral for that average force in a simple fluid. The integral immediately implies which simulation parameters the force can and cannot depend on, which in turn licenses cheaper box geometries for computing the force and supplies interfacial validation checks that do not require a large atomistic region. Molecular-dynamics tests on the supercritical Lennard-Jones fluid at three thermodynamic states confirm that the iterated force collapses onto the integral expression near the boundary and that the proposed minimal setups reproduce the correct one-body density, Maxwellian momenta, particle-crossing statistics and average force.","feed_headline":"AdResS coupling force is mostly a missing average boundary force","feed_subtitle":"A one-dimensional integral from open-system theory simplifies force calculation and validation","key_machinery":"The average boundary force F^x_av(r_∂)=πρ(r_∂^{2}I_0-I_2) derived from the hierarchy source term Ψ_n; it supplies both the physical identification of the thermodynamic force and the parameter-independence statements used for simplification and validation.","core_discovery":"Near the AT/Δ boundary the thermodynamic force of AdResS is the missing average force through the Δ/TR interface, F_th(x)=F^x_av(x_Δ/TR-x), where F^x_av is the explicit one-dimensional integral πρ(r_∂^{2}I_0-I_2) obtained by marginalising the open-system Liouville hierarchy for a simple atomistic fluid.","pith_inferences":["The same integral construction should extend, with only technical changes, to multi-site molecules or multi-component mixtures once the appropriate pair correlations are supplied.","Because the force is independent of thermostat strength and capping radius near the boundary, one can systematically weaken or localize the thermostat without re-deriving the thermodynamic force.","The derived potential of mean force offers a natural starting point for time-dependent or non-equilibrium AdResS reservoirs whose target density and temperature vary slowly."],"forward_implications":["The thermodynamic force near the open-system boundary is completely fixed by density, temperature, pair potential and distance to the outer interface.","Force iteration can be performed in a minimal box with no atomistic region and reduced transverse dimensions, then transferred unchanged to a production run.","Validation reduces to interfacial observables (one-body density, Maxwellian momenta, net crossing statistics, average force) that do not require a large AT domain.","The same integral supplies an a-priori initial guess that already matches the converged force near the boundary."],"fun_headline_variants":["AdResS thermodynamic force equals missing average boundary force","Open-system Liouville hierarchy reframes AdResS coupling force","Thermodynamic force is average force missing at Δ/TR interface","One-D integral from Liouville hierarchy yields AdResS force","AdResS force near AT/Δ matches open-system boundary term"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The residual artifact force from excluded-volume overlap, thermostat residuals and thermodynamic mismatch must fade inside the transition region before it reaches the atomistic boundary; if it does not, the force-hierarchy identification fails.","fun_headline_variants_meta":{"raw":{"variants":["AdResS thermodynamic force equals missing average boundary force","Open-system Liouville hierarchy reframes AdResS coupling force","Thermodynamic force is average force missing at Δ/TR interface","One-D integral from Liouville hierarchy yields AdResS force","AdResS force near AT/Δ matches open-system boundary term"]},"model":"grok-4.5","effort":"low","cost_usd":0.003678,"raw_usage":{"total_tokens":1126,"prompt_tokens":715,"num_sources_used":0,"completion_tokens":73,"cost_in_usd_ticks":36784000,"prompt_tokens_details":{"text_tokens":715,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":338,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":715,"tokens_out":73,"duration_ms":4493,"temperature":1.0,"reasoning_tokens":338,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T11:21:35.343711+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Compute the iterated thermodynamic force and the integral F_av on the same grid for a supercritical Lennard-Jones fluid; if they disagree throughout the half of the Δ region nearest the AT/Δ boundary, or if particle-crossing fluctuations and the measured average force through that boundary deviate from a full-atomistic reference, the claim is false.","supporting_citations":[],"review_version":1}