{"id":"54900d4d-9003-40a6-a00a-49c3ebe8761e","arxiv_id":"2412.16405","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Diffusion Monte Carlo interaction energies for the full S66 dataset reveal systematic deviations from CCSD(T) that correlate with the ratio of electrostatic to dispersion contributions.","lead":"Researchers used diffusion Monte Carlo to compute interaction energies for all 66 complexes in the S66 benchmark set and compared them with coupled cluster references. They found a systematic pattern: DMC binds hydrogen-bonded systems more strongly and dispersion-bound systems more weakly than CCSD(T), with the size of the discrepancy tracking the electrostatic-to-dispersion ratio.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"CCSD(T) deformation energies enter every DMC interaction energy; unpropagated and validated only at finite time step, so the central trend is conditional.","rationale":"The reader's weakest assumption identifies the use of CCSD(T) deformation energies in Eq. (2) as the main unquantified systematic risk. I agree. The paper provides a valuable DMC benchmark for S66 and a plausible correlation with ELST/DISP, and the validation in Table S3 demonstrates that DMC and CCSD(T) deformation energies agree within about 0.12 kcal/mol for 43 cases, including the monomers with the largest deformations. However, this uncertainty is not propagated into the reported DMC error bars nor into the regression in Fig. 4, and the validation is performed at a single time step (0.01 au) rather than the extrapolated zero time-step limit used for the final interaction energies. Since the deformation energies for uracil, acetic acid, and other monomers reach 0.3-1.3 kcal/mol, a systematic offset in the unvalidated or finite-time-step cases could shift individual points by more than their stated error and alter the slope or R^2 of the central correlation. The proposed Monte Carlo test uses the existing validation data to quantify whether the observed deviations are sufficient to change the conclusion. I considered other potential concerns, such as the DLA localization approximation and fixed-node errors, but those are not as directly tied to a specific overlooked approximation in the workflow. The reader's conditional verdict is appropriate, and no change is needed.","tokens_in":40182,"tokens_out":12457,"duration_ms":108802,"concrete_test":"Perform a Monte Carlo sensitivity analysis: for each of the 43 validated monomer cases, treat the DMC-CCSD(T) deformation-energy deviation (Table S3) as a random draw from a distribution (mean and sigma from the 43 values); resample the deformation energies for all 132 monomer instances, recompute all 66 DMC interaction energies via Eq. (2), and refit the Fig. 4 linear correlation 10,000 times. If the R^2 drops below 0.6 or the slope changes sign in more than 5% of resamples, the deformation-energy approximation is load-bearing; otherwise the trend is robust to this uncertainty.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The DMC interaction energies in Eq. (1) are not computed with DMC monomer energies at the in-dimer geometries. Instead, Eq. (2) of Sec. II A uses DMC total energies at a reference geometry plus CCSD(T) deformation energies. This is an essential approximation: for several monomers (uracil, acetic acid, acetamide, peptide) the deformation energies are 0.3-1.3 kcal/mol (Table S2), comparable to the reported MAD of 0.21 kcal/mol and to the deviations used to establish the trend. The authors validate DMC vs CCSD(T) deformation energies for 43 monomer cases in Table S3, finding deviations within 0.12 kcal/mol. However, this validation (i) does not cover all monomer/dimer combinations; (ii) uses DMC deformation energies at a single time step (0.01 au) rather than the zero time-step limit used for the final DMC estimates; and (iii) does not propagate the observed deviations into the final error bars or into the correlation analysis of Fig. 4. If the deformation-energy error is systematically positive or negative for monomers that dominate one interaction class (e.g., uracil in dispersion-bound dimers vs H-bonded dimers), the slope and R^2 of the log(ELST/DISP) correlation could shift. Because the central claim is a systematic trend, this unquantified systematic risk is load-bearing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports fixed-node diffusion Monte Carlo (DMC) interaction energies for the full S66 dataset, using eCEPP pseudopotentials, determinant locality approximation, LDA trial wave functions, and extrapolation to the zero time-step limit. The DMC values are compared to an average of three literature CCSD(T) composite estimates. The authors find a mean absolute deviation of 0.21 kcal/mol overall, with a systematic pattern: DMC binds hydrogen-bonded (electrostatic-dominated) systems more strongly than CCSD(T) and dispersion-dominated systems more weakly. The relative deviation is reported to correlate linearly with the natural logarithm of the SAPT electrostatic-to-dispersion ratio (R²=0.78). The paper also applies an empirically fitted CCSD(cT) correction to dispersion-dominated complexes, reducing the mean absolute deviation to 0.09 kcal/mol, and identifies the acetic acid dimer and uracil–cyclopentane dimer as compact model systems with pronounced discrepancies.","tokens_in":40475,"tokens_out":3496,"duration_ms":32543,"significance":"If the reported trends hold, this is a valuable contribution to the benchmarking literature for non-covalent interactions: it provides the first systematic DMC reference set for the widely used S66 database, subject to careful convergence testing (time-step extrapolation, localization scheme, trial wavefunction, pseudopotential for the acetic acid dimer). The correlation with a simple SAPT ratio is a potentially useful, falsifiable descriptor for anticipating DMC-CCSD(T) disagreement. The authors also make all data and analysis scripts publicly available, which aids reproducibility and further analysis. The main limitation is that the central trend is conditional on an unquantified approximation in how monomer deformation energies are combined with DMC total energies, as detailed in the major comments.","major_comments":[{"comment":"The DMC interaction energies are not obtained from DMC monomer energies at the in-dimer geometries; instead, Eq. (2) of the SI combines DMC total energies at a reference geometry with CCSD(T) deformation energies. The authors validate this procedure for 43 monomer cases (Table S3) and find deviations within 0.12 kcal/mol, but this validation is performed at a single time step (0.01 au) rather than at the zero time-step limit used for the final DMC estimates, and it does not cover all monomer/dimer combinations. For several monomers (e.g., uracil, acetic acid, acetamide) the deformation energies are 0.3–1.3 kcal/mol, comparable to the reported MAD of 0.21 kcal/mol and to the deviations that define the trend in Fig. 4. The observed validation deviations are not propagated into the reported error bars nor into the correlation analysis. A systematic bias in the unvalidated cases, especially for monomers that dominate one interaction class, could shift the slope and R² of the log(ELST/DISP) correlation. I request that the authors either compute DMC deformation energies for all relevant monomer/dimer pairs at the zero time-step limit, or explicitly add the measured 0.12 kcal/mol (or a conservative version of it) as a systematic uncertainty in the interaction energies and re-evaluate the trend and its significance under that perturbation.","section":"Sec. II A / SI Sec. S2.1, Eq. (2)"}],"minor_comments":[{"comment":"In Eq. (2) of the main text, the variables MP2 corr. and CCSD corr. are not defined in the main text; they are only defined in the SI. Please add a brief definition or a pointer to the SI.","section":"Sec. II B, Eq. (2)"},{"comment":"The statement that “The stronger binding of DMC over CCSD(T) has not been (systematically) reported before” is rather strong; several recent works (e.g., Refs. 49, 50) discuss related discrepancies. Please qualify this claim to avoid overstatement.","section":"Sec. III, Figs. 1–3"},{"comment":"In Table S4, the entries “σcubic fit” and “Δ linear fit/cubic fit” appear as line breaks in the table; this is a formatting issue that may confuse readers. Please format the error-type entries unambiguously.","section":"SI Sec. S5, Table S4"},{"comment":"The main text reports R²=0.78 for the correlation in Fig. 4 but does not report the slope, intercept, or a measure of statistical significance. Adding these values (and ideally the 95% confidence interval) would strengthen the claim.","section":"Sec. III / SI Sec. S9"},{"comment":"The SI correctly states that the CCSD(cT)-fit has only been parametrized for dispersion-dominated complexes and its transferability is unconfirmed; this caveat is not mentioned in the main text when discussing CCSD(cT) results. Please include this limitation in the main text.","section":"Sec. IV / SI Sec. S6"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is well-suited to the journal's scope and the DMC calculations are, in my assessment, carefully conducted for the systems studied. The central trend is plausible and interesting, but the load-bearing approximation in the deformation-energy assembly needs to be addressed before publication. I am not requesting new large-scale DMC calculations for all 132 monomers if the authors can demonstrate, via a sensitivity analysis or a targeted subset, that the trend is robust to the observed deformation-energy discrepancies. Please ensure the authors address the major comment directly rather than only clarifying the text."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline is that this paper gives the first DMC interaction energies for the full S66 set, and the systematic pattern—DMC binds H-bonded systems more strongly than CCSD(T) and dispersion-dominated ones more weakly—looks real. The correlation with log(ELST/DISP) is a new, practical diagnostic, and it survives a second SAPT level in the SI. Credit where due: the DMC calculations are carefully converged and tested (time-step extrapolation, pseudopotential and trial-wavefunction checks for acetic acid), and the data and analysis notebooks are on GitHub. That is the standard the field needs.\n\nThe soft spot is the deformation-energy protocol. DMC monomer energies are computed at a reference geometry, and CCSD(T) deformation energies are added to get the in-dimer geometry. The SI validates this against DMC for 43 monomer cases and finds agreement within 0.12 kcal/mol, which is reassuring. But the validation uses a single time step (0.01 au) rather than the zero-time-step limit used in the final numbers, and the uncertainty is not propagated. For uracil and acetic acid, deformation energies of 0.3–1.3 kcal/mol are comparable to the reported MAD of 0.21 kcal/mol, so this is not a trivial formal point. If there were a systematic offset in the unvalidated monomer geometries—especially one that splits by interaction class—the slope and R² of the descriptor correlation could shift. I don't think it would erase the trend: the validated cases include the largest deformations, the deviations scatter around zero, and the spread in log(ELST/DISP) is large enough to absorb random 0.1 kcal/mol noise. But the error bars on the DMC points are arguably optimistic, and a referee should ask for either DMC deformation energies for the remaining monomer cases or a sensitivity analysis.\n\nThe CCSD(cT)-fit analysis is clearly presented as empirical and secondary; it weakens binding for all dispersion systems but does not resolve the uracil–cyclopentane outlier. That is handled honestly. The comparison to three CCSD(T) references from the literature is fair, and the averaging procedure is transparent.\n\nWho is this for? Anyone benchmarking DMC or coupled cluster on non-covalent interactions, and developers of energy decomposition descriptors. It deserves a serious referee. I would send it out.","headline":"First DMC benchmark for the full S66 set shows a real systematic trend against CCSD(T), and the deformation-energy shortcut is a legitimate but not fatal soft spot.","tokens_in":40992,"tokens_out":2667,"would_cite":true,"duration_ms":23222,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Reference methods split by interaction type across 66 dimers.","keywords":["diffusion Monte Carlo","CCSD(T)","non-covalent interactions","S66 dataset","energy decomposition analysis","SAPT","interaction energy benchmarks","electrostatics vs dispersion"],"falsifier":"Compute DMC deformation energies directly for the monomers with the largest deformation corrections, especially uracil and acetic acid; if those DMC values deviate from the CCSD(T) values beyond the stated ~0.12 kcal/mol validation margin, the trend line in Fig. 4 shifts or breaks. Alternatively, run a backflow or multideterminant fixed-node DMC on the acetic acid dimer (ID 20); if the 0.78 kcal/mol overbinding against CCSD(T) disappears, the hydrogen-bond side of the discrepancy is a nodal-surface artifact.","tokens_in":40003,"feed_emoji":"🧪","tokens_out":7016,"duration_ms":56181,"temperature":0.7,"pith_summary":"The paper tries to establish that two of the most trusted quantum-chemistry methods, diffusion Monte Carlo (DMC) and coupled-cluster CCSD(T), disagree in a systematic rather than random way on medium-sized molecular complexes. Across the full S66 benchmark of 66 dimers, DMC binds hydrogen-bonded (electrostatically dominated) systems more strongly than CCSD(T), while dispersion-dominated systems bind more weakly. The relative size of the discrepancy tracks the natural logarithm of the electrostatic-to-dispersion ratio taken from energy decomposition, with R² = 0.78, so interaction character predicts where the methods will diverge. The authors also single out two small complexes, acetic acid dimer and uracil-cyclopentane dimer, where the disagreement is large enough to serve as focused benchmarks. If the pattern holds, future method development has clear targets and a cheap descriptor for anticipating disagreements between reference methods.","feed_headline":"DMC and CCSD(T) split predictably across 66 dimers","feed_subtitle":"Diffusion Monte Carlo overbinds hydrogen-bonded dimers and underbinds dispersion-bound ones; a simple ratio predicts the gap.","key_machinery":"The central organizing object is the ratio of electrostatic to dispersion contributions, ELST/DISP, obtained from symmetry-adapted perturbation theory (SAPT) energy decomposition. Its natural logarithm serves as one descriptor that sorts the 66 dimers along a line: positive values for hydrogen-bonded systems, negative values for dispersion-dominated ones, and a linear relation with the relative DMC-versus-CCSD(T) difference. The computational protocol that makes the dataset tractable is a two-part scheme: DMC total energies are computed only for 14 monomers at reference geometries, and CCSD(T) deformation energies move those monomers to their in-dimer geometries, with the deformation correction checked against DMC on a subset of 43 monomer cases. This protocol, combined with cubic time-step extrapolation of the interaction energy itself, is what allows the whole 66-dimer comparison to be carried out at sub-0.12 kcal/mol statistical accuracy.","core_discovery":"On the authors' terms, fixed-node DMC, extrapolated to the zero-time-step limit and with monomer deformation energies supplied by CCSD(T) (validated on a subset), yields interaction energies for all 66 S66 dimers with statistical errors mostly below 0.10 kcal/mol. Compared with the average of three recent near-basis-set-limit CCSD(T) references, DMC shows a mean absolute deviation of 0.21 kcal/mol, but the deviations are not noise: DMC overbinds nearly every hydrogen-bonded dimer, up to 0.78 kcal/mol for acetic acid (ID 20), and underbinds most dispersion-dominated dimers, up to 0.47 kcal/mol for uracil-cyclopentane (ID 42). Plotting the relative difference against the natural logarithm of the SAPT electrostatic-to-dispersion ratio collapses the dataset onto a single linear trend with R² = 0.78. For dispersion-dominated systems, a recently proposed empirical CCSD(cT)-fit correction moves CCSD(T) closer to DMC, reducing that subset's MAD to about 0.09 kcal/mol, though a notable gap remains for uracil-cyclopentane. The paper stops short of identifying the physical origin of the split, leaving fixed-node error or missing higher-order coupled-cluster terms as open possibilities.","pith_inferences":["If this descriptor generalizes beyond S66, other dispersion-heavy benchmarks, including host-guest and π-stacked complexes, should show DMC/CCSD(T) gaps that grow with increasingly negative log(ELST/DISP); that prediction could be tested on existing datasets of larger non-covalent complexes.","The DMC-overbinding side of the trend, including the A24 water-ammonia and HCN cases, suggests a nodal-surface bias specific to hydrogen bonds rather than a missing-triples problem; a direct test would be a backflow or multideterminant fixed-node calculation on the acetic acid dimer.","Because monomer deformation energies are as large as 1.3 kcal/mol and DMC checks cover only a subset, the cleanest way to harden the trend is to compute DMC deformation energies for all 66 dimers; if the unvalidated cases deviate, the R² = 0.78 line would shift.","An empirical (cT) correction fitted only to dispersion-bound systems may worsen the H-bonded side if applied there, so a single correction is unlikely to close both ends of the correlation."],"forward_implications":["If DMC is the more accurate reference, CCSD(T)-based benchmark values for hydrogen-bonded complexes inherit a systematic weak-binding bias of up to about 0.8 kcal/mol, which would propagate into density-functional and machine-learning potentials trained on those references.","The log(ELST/DISP) correlation can be used predictively: a cheap SAPT decomposition of any new dimer indicates whether DMC and CCSD(T) will diverge and roughly by how much.","For dispersion-dominated systems, replacing the perturbative triples (T) with the empirical (cT) correction cuts the mean absolute deviation against DMC from about 0.24 kcal/mol to about 0.09 kcal/mol, pointing to the triples treatment as a major contributor on that side of the trend.","The acetic acid dimer (64 electrons) and uracil-cyclopentane dimer (98 electrons) provide compact, cost-effective benchmarks, nearly ten times smaller in electron count than the buckyball-ring system that originally exposed the discrepancy.","The relative differences are small for H-bonded systems (mean 2.45%) but much larger for dispersion-dominated ones (mean 8.21%), so the practical impact of the discrepancy depends strongly on the interaction type."],"supporting_citations":[{"why":"Defines the S66 dataset of 66 dimers with geometries and interaction categories that the whole comparison rests on.","marker":"[64]"},{"why":"One of the three CCSD(T) estimates averaged as the reference; supplies revised interaction energies for the dataset.","marker":"[75]"},{"why":"The 'SILVER' CCSD(T) reference, the second of the three averaged estimates.","marker":"[89]"},{"why":"The '14k-GOLD' CCSD(T) reference, the third of the three averaged estimates.","marker":"[92]"},{"why":"Provides the SAPT electrostatic and dispersion decomposition used to build the log(ELST/DISP) descriptor.","marker":"[94]"},{"why":"Introduces the empirical CCSD(cT)-fit relation that the paper applies to reduce the dispersion-side discrepancy.","marker":"[50]"},{"why":"Documents the large DMC/CCSD(T) disagreement for big dispersion-bound molecules that motivates the systematic search.","marker":"[49]"},{"why":"Supplies DMC results on the A24 set showing similar H-bonded overbinding, corroborating the trend.","marker":"[96]"}],"fun_headline_variants":["DMC and CCSD(T) split by interaction type","Ratio of electrostatics to dispersion predicts method gap","Systematic DMC-CCSD(T) discrepancies across S66","DMC overbinds hydrogen bonds, underbinds dispersion","Simple ratio explains DMC-CCSD(T) differences"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported DMC interaction energies rely on CCSD(T) deformation energies standing in for DMC deformation energies for every monomer in the dataset, even though direct DMC checks cover only a subset of the monomer cases.","fun_headline_variants_meta":{"raw":{"variants":["DMC and CCSD(T) split by interaction type","Ratio of electrostatics to dispersion predicts method gap","Systematic DMC-CCSD(T) discrepancies across S66","DMC overbinds hydrogen bonds, underbinds dispersion","Simple ratio explains DMC-CCSD(T) differences"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000374,"raw_usage":{"total_tokens":2087,"prompt_tokens":1128,"completion_tokens":959,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":744,"completion_tokens_details":{"reasoning_tokens":876}},"tokens_in":744,"tokens_out":959,"duration_ms":7992,"temperature":1.0,"reasoning_tokens":876,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:36:21.233249+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute DMC deformation energies directly for the monomers with the largest deformation corrections, especially uracil and acetic acid; if those DMC values deviate from the CCSD(T) values beyond the stated ~0.12 kcal/mol validation margin, the trend line in Fig. 4 shifts or breaks. Alternatively, run a backflow or multideterminant fixed-node DMC on the acetic acid dimer (ID 20); if the 0.78 kcal/mol overbinding against CCSD(T) disappears, the hydrogen-bond side of the discrepancy is a nodal-surface artifact.","supporting_citations":[{"cited_title":"R ez \\'a c , author K","cited_arxiv_id":null,"evidence_quote":"One of the three CCSD(T) estimates averaged as the reference; supplies revised interaction energies for the dataset."},{"cited_title":"a fer , author A. Irmler , author A. Gallo , \\ and\\ author A. Gr \\","cited_arxiv_id":null,"evidence_quote":"Introduces the empirical CCSD(cT)-fit relation that the paper applies to reduce the dispersion-side discrepancy."},{"cited_title":"Basis set incompleteness errors in fixed-node diffusion Monte Carlo calculations on non-covalent interactions","cited_arxiv_id":"2412.00368","evidence_quote":"Supplies DMC results on the A24 set showing similar H-bonded overbinding, corroborating the trend."}],"review_version":1}