{"id":"10d50f78-a76a-49e2-b5a7-baa50f7b046c","arxiv_id":"1908.01490","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"RPA with singles corrections gives more accurate adsorption energies for seven molecules in zeolite chabazite than MP2, at roughly one tenth of the computational cost.","lead":"This paper compares two high-accuracy quantum chemistry methods, MP2 and RPA with singles corrections, for computing how strongly small molecules stick to the zeolite chabazite. It finds that RPA with singles is both more accurate (against CCSD(T) references) and about ten times cheaper than MP2, making it a practical choice for adsorption energies in porous materials.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Cluster benchmarks do not support the claimed <2 kJ/mol or 5% error bound; Table III shows RPA+RSE error of 2.6 kJ/mol (8%) for ethylene on the 4T cluster, and error growth with cluster size is unquantified.","rationale":"The strongest claim is that RPA with singles provides near-reference adsorption energies and outperforms MP2 in accuracy and cost. The accuracy part is the load-bearing element: if it fails, the recommendation to use RPA+singles as a reference method collapses. The evidence for accuracy comes only from 2T/4T clusters with CCSD(T) references. My concern sharpens the reader's weakest assumption: the clusters are not merely potentially unrepresentative—their own data already contradict the stated error bound. Table III shows RPA+RSE error for ethylene of 2.6 kJ/mol (8%) on the 4T cluster, which is larger than both '2 kJ/mol' and '5%'. Moreover, the error growth from 2T (0.4 kJ/mol) to 4T (2.6 kJ/mol) for ethylene, and similar growth for alkanes, indicates that the benchmark is not converged with cluster size. Without a 6T/8T or embedded-cluster check, the bulk claim is an extrapolation. I also note that the efficiency claim is weaker than stated because the comparison uses the standard MP2 implementation against a cubic-scaling RPA code; the O(N^4) MP2 of Ref. 59 is acknowledged as more efficient but not used. However, I regard the accuracy/transfer issue as the single most load-bearing because it concerns the scientific conclusion itself. The reader identified the same general area (cluster representativeness), so agreement is partial. The recommended verdict remains CONDITIONAL, matching the reader's verdict; the manuscript needs to add cluster-size convergence data or qualify the error bound.","tokens_in":17022,"tokens_out":9002,"duration_ms":84015,"concrete_test":"Compute RPA+RSE and PNO-CCSD(T)(F12*) adsorption energies for ethylene and methane on a 6T (or larger) cluster cut from the same chabazite geometry used in the paper. If the RPA+RSE error for ethylene remains above 2 kJ/mol (or 5%) or grows beyond the 4T value of 2.6 kJ/mol, the headline error bound and the transfer-to-bulk claim are not supported. A cheaper analytical check: re-analyze Table III directly—ethylene RPA+RSE error is 2.6 kJ/mol (8%) vs the claimed <2 kJ/mol/5% bound; if confirmed, the Discussion statement must be qualified regardless of new calculations.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central accuracy claim—RPA with singles gives adsorption energies with 'errors of less than 2 kJ/mol or 5 %' (Sec. V)—is not supported by the presented benchmarks. The only direct checks against CCSD(T) are on 2T and 4T clusters. On the largest cluster, Table III lists ethylene: RPA+RSE gives -35.3 kJ/mol vs PNO-CCSD(T) -32.7 kJ/mol, an error of 2.6 kJ/mol (8%), exceeding both stated thresholds. The 2T error for the same molecule is 0.4 kJ/mol (Table II), so the error grows by about 2 kJ/mol between 2T and 4T, showing cluster-size non-convergence; no 6T/8T or extrapolated cluster data are provided. Because the bulk system cannot be checked against CCSD(T), the 'reference quality' claim rests entirely on transferring cluster accuracy to the periodic system, and the available cluster data already fail the stated bound for one of seven molecules. The close MP2/RPA agreement in bulk (Table I) only shows the two approximations agree, not that either is within the claimed error. The conclusion that RPA is 'superior to MP2 in accuracy' is also weaker than stated: on the 4T cluster, MP2 is closer to the reference for the three alkanes (errors 0.5-0.8 kJ/mol vs 1.4-1.7 kJ/mol for RPA+RSE).","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript benchmarks adsorption energies of seven small molecules (methane, ethane, ethylene, acetylene, propane, CO2, H2O) in zeolite chabazite, using periodic VASP calculations at the MP2 and RPA-with-singles-corrections (RSE and GWSE) levels, and on 2T and 4T finite clusters with CCSD(T) references. It reports close agreement between MP2 and RPA+singles for the bulk material and finds on clusters that RPA+RSE is generally closer to CCSD(T) than MP2 for unsaturated molecules. The paper concludes that RPA with singles is superior to MP2 both in accuracy and in computational cost, and that it can provide reference-quality adsorption energies at 0 K with errors below 2 kJ/mol or 5%.","tokens_in":17338,"tokens_out":5548,"duration_ms":53663,"significance":"If the central claims held, the paper would establish RPA with singles corrections as a practical reference method for adsorption in porous materials, which is a useful and important result. The benchmark design has real strengths: an external CCSD(T) reference on clusters, systematic basis-set extrapolation, use of hard PAW potentials, finite-size corrections, and parameter-free many-body methods with no fitting to the target adsorption energies. However, the data support a more limited conclusion: RPA with singles is accurate to roughly 1-2 kJ/mol for most systems and is clearly better than MP2 for ethylene and acetylene, but the blanket 'less than 2 kJ/mol or 5%' claim and the order-of-magnitude cost advantage are not established by the presented evidence.","major_comments":[{"comment":"The central accuracy claim is not supported as stated. Section V states that RPA with singles provides energies 'with errors of less than 2 kJ/mol or 5 %.' Against the largest available cluster (4T, Table III), RPA+RSE gives -35.3 kJ/mol for ethylene versus PNO-CCSD(T) -32.7 kJ/mol, an error of 2.6 kJ/mol (8%), exceeding both thresholds. Moreover, on the same 4T cluster MP2 is closer to the reference for methane (0.6 vs 1.5 kJ/mol), ethane (0.5 vs 1.4), and propane (0.8 vs 1.7), so the statement that RPA is 'superior to MP2 in accuracy' is too broad. Please revise the accuracy claims to be consistent with the largest cluster results, or provide additional cluster-size data that resolve the discrepancy.","section":"Section V, Tables II and III"},{"comment":"The claim that RPA has a computational cost 'about an order of magnitude smaller' than MP2 is asserted without supporting timing data. No CPU times, wall-clock times, or hardware details are given anywhere in the manuscript. Since the abstract and conclusions present this cost advantage as a headline result, the authors should either include comparative timings or explicitly temper the cost claim.","section":"Section IV.A and Abstract"},{"comment":"Cluster-size convergence is not established, which matters because the accuracy of the periodic adsorption energies is inferred by transfer from the cluster benchmarks. Only 2T and 4T clusters are reported, and the RPA+RSE error for ethylene grows from 0.4 kJ/mol on the 2T cluster (Table II) to 2.6 kJ/mol on the 4T cluster (Table III). No 6T/8T cluster or extrapolated cluster-size limit is provided, so the convergence behavior is unknown. Without this, the claim that RPA with singles provides 'reference quality' energies in the periodic system, where no CCSD(T) reference exists, is not quantitatively supported.","section":"Section IV.B"}],"minor_comments":[{"comment":"The geometries are optimized with optB88-vdW and not relaxed with the correlated methods; this is acknowledged but the possible sensitivity of the method ranking to this choice is not discussed.","section":"Section II"},{"comment":"The figure caption mentions HF but the legend does not include an HF curve; please either add it to the legend or update the caption.","section":"Figure 2"},{"comment":"The table heading says 'PNO-CCSD(T)' while the text describes 'PNO-CCSD(T0)(F12*)' with a T/T0 scaling correction; please make the notation consistent.","section":"Table III"},{"comment":"The phrase 'errors of less than 2 kJ/mol or 5 %' is ambiguous: it should state whether this is a mean absolute error or a maximum error, and whether both criteria must be met or only one.","section":"Section V"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a useful benchmark dataset and a careful convergence protocol, but the headline claims exceed what the data show. The most important fix is to revise the accuracy and cost statements, or to supply the missing cluster-size and timing evidence. With those changes the paper could be a solid contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real content here is the systematic cluster benchmark: MP2 overbinds ethylene and acetylene by 10–17% relative to CCSD(T), while RPA+RSE stays within a few kJ/mol for all seven molecules. That is a new and chemically sensible result for zeolite adsorption, and the convergence testing—basis set, cell size, PAW potentials, CCSD(T)(F12*) references—is careful. The 2T and 4T cluster construction is sound, and the finite-size corrections are done properly.\n\nThe soft spots are in the claims, not the data. The abstract and Section V say RPA with singles gives errors of less than 2 kJ/mol or 5%. Table III does not support that. On the 4T cluster, RPA+RSE errors are: methane 1.5 kJ/mol (10%), ethane 1.4 (8%), ethylene 2.6 (8%), propane 1.7 (9%), CO2 1.2 (4%), H2O 1.0 (1%). So the 5% threshold fails for four of seven molecules and the 2 kJ/mol threshold fails for ethylene. The accuracy comparison with MP2 is also not a clean win: on the 4T cluster, MP2 is closer to the reference for the three alkanes (0.5–0.8 kJ/mol versus 1.4–1.7 for RPA+RSE). The superiority of RPA is real for the unsaturates but not general, as the authors acknowledge only indirectly.\n\nThe “order of magnitude smaller” cost claim appears without any timing data, so it is an assertion, not a result. And because the supplementary material is not attached to the arXiv version, the structures and exact input parameters are not independently checkable from the preprint. That is a reproducibility issue for the version in front of us.\n\nOne more thing worth flagging: the cluster-size trend. RPA+RSE alkane errors grow from about 0.3 kJ/mol on 2T to 1.5 kJ/mol on 4T. That does not invalidate the method, but it means the claim of bulk “reference quality” is an extrapolation. A 6T or 8T point for at least one or two molecules would have made the transfer argument much stronger.\n\nWho is this for? Anyone choosing reference methods for adsorption in porous materials, and anyone benchmarking dispersion-corrected DFT against correlated wavefunction methods. The MP2 overbinding result is the lasting takeaway and is worth citing on its own.\n\nRecommendation: send it to referees. The benchmark data are real and the main physical finding is well supported. The paper needs revision to recalibrate the accuracy claim to what Tables II and III actually show, add timing data or drop the cost claim, and make the SI available. That is normal referee work, not a fatal flaw.","headline":"A useful seven-molecule zeolite benchmark shows MP2 overbinds ethylene and acetylene, but the advertised '<2 kJ/mol or 5%' error bound for RPA+RSE is not met on the 4T cluster, and the cost claim is unquantified.","tokens_in":17823,"tokens_out":3965,"would_cite":true,"duration_ms":37247,"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 paper claims that RPA with singles corrections gives near-reference adsorption energies in zeolites, outperforming MP2 at lower cost.","keywords":["adsorption energy","random phase approximation","singles corrections","MP2","zeolite","chabazite","dispersion interactions","coupled cluster benchmark"],"falsifier":"Compute adsorption energies for the same seven molecules on a larger cluster model, for example a full double-six-ring or 6T site, with CCSD(T)-level theory and compare with RPA+RSE and MP2; if RPA+RSE errors exceed 2 kJ/mol or MP2 matches or beats it, the central claim fails. A complementary test is to compare RPA+GWSE 0 K energies to experimental adsorption enthalpies after including zero-point and thermal corrections, since a systematic drift beyond 2 kJ/mol would also refute the claimed reference quality.","tokens_in":16857,"feed_emoji":"🧪","tokens_out":7826,"duration_ms":73540,"temperature":0.7,"pith_summary":"This paper targets a practical question: which affordable correlated electronic-structure method should be trusted to compute adsorption energies in porous materials such as zeolites? The authors calculate adsorption energies of seven small molecules—methane, ethane, ethylene, acetylene, propane, carbon dioxide, and water—in the zeolite chabazite, comparing MP2 with the random phase approximation (RPA) augmented by singles corrections (RSE or GWSE). Their central claim is that RPA with singles reproduces coupled-cluster reference values to within about 2 kJ/mol or 5 percent, while MP2 is less reliable exactly where electron screening matters, notably for ethylene and acetylene, and is about an order of magnitude more expensive. If the claim holds, RPA with singles becomes the recommended method for generating reference-quality adsorption energies in zeolites, and the paper's data also expose systematic biases in dispersion-corrected density functional approximations.","feed_headline":"RPA with singles beats MP2 for zeolite adsorption","feed_subtitle":"Seven molecules in chabazite: near-reference accuracy at a tenth of MP2's cost.","key_machinery":"The central object is the RPA correlation energy augmented by a singles correction, either RSE (renormalized single excitations) or GWSE (screened single excitations), evaluated on exact exchange built from a semilocal DFT starting point. The singles correction rotates the one-particle density matrix toward the Hartree-Fock one, which repairs the too-weak binding of bare RPA. What makes the comparison possible is a set of small cluster models (2T and 4T tetrahedral-site clusters capped with hydrogen) small enough for CCSD(T) reference calculations while still representing the acidic chabazite adsorption site; on these clusters the accuracy of MP2 and RPA+RSE is ranked directly against the coupled-cluster reference.","core_discovery":"The paper's core discovery is that the inclusion of singles corrections fixes the main deficiency of RPA for adsorption: bare RPA underestimates binding by about 10–20 percent, but adding either the renormalized singles excitation (RSE) or the GW-screened singles excitation (GWSE) brings adsorption energies to near CCSD(T) quality, with the two variants differing by less than 1 kJ/mol in the bulk. MP2, by contrast, overbinds ethylene and acetylene by about 2–3 kJ/mol relative to RPA+GWSE because it lacks the infinite-order screening that RPA includes; this pattern is confirmed on 2T and 4T cluster models where CCSD(T) and PNO-CCSD(T)(F12*) references exist. On those clusters RPA+RSE stays within about 3 percent of the reference for all seven molecules, whereas MP2 errors reach 10 percent for the unsaturated molecules. The paper therefore concludes that RPA with singles surpasses MP2 both in accuracy and in cost for adsorption in zeolites, and that it can serve as a parameter-free source of near-reference 0 K adsorption energies.","pith_inferences":["If the accuracy ranking transfers to the full pore, the same test on a larger-pore or higher-silica zeolite should reproduce the pattern that MP2's overbinding grows with the extent of the molecule's delocalized electrons; a molecule like benzene adsorbed in a large cage would sharpen this prediction.","The observed sign correlation between MP2 minus RPA differences and the second-order exchange contribution suggests that adding a screened second-order exchange term to RPA should shrink the remaining error; that is a testable calculation.","The observed 0.5–1.5 kJ/mol sensitivity to pseudopotential hardness implies that older zeolite adsorption benchmarks using softer potentials may carry a systematic bias; re-evaluating those systems with harder potentials is a concrete next step."],"forward_implications":["For saturated hydrocarbons and water, MP2 and RPA+GWSE agree to within about 1–2 kJ/mol, so either method could serve as a reference; for ethylene and acetylene, MP2 is the outlier and overbinds because it lacks screening.","Bare RPA underestimates adsorption energies by roughly 10 percent, and the singles corrections are necessary to reach near-reference accuracy.","The claimed <2 kJ/mol or <5 percent error at 0 K means RPA with singles can replace CCSD(T) as the reference generator for zeolites too large for coupled cluster.","The cost advantage, about one order of magnitude less than MP2, makes periodic RPA with singles practical for unit cells up to roughly 3000 cubic angstroms, covering zeolites such as ferrierite, mordenite, and LTA.","The 0 K energies can be combined with resampling or correction schemes and experimental heats of adsorption to yield finite-temperature adsorption enthalpies."],"supporting_citations":[{"why":"Supplies earlier periodic RPA and MP2 adsorption data for chabazite that this study compares against and partly corrects.","marker":"[23]"},{"why":"Introduces the renormalized singles excitation (RSE) correction used to define RPA with singles.","marker":"[38]"},{"why":"Introduces the GW singles correction (GWSE) and the low-scaling implementation used for the periodic calculations.","marker":"[39]"},{"why":"Provides earlier benchmark data showing RPA with singles reaches near-reference accuracy for water adsorption on boron nitride.","marker":"[13]"},{"why":"Shows singles-corrected RPA reproduces molecular-crystal lattice energies, supporting transferability to porous solids.","marker":"[40]"},{"why":"Applies RPA with singles to CO adsorption in a zeolite and demonstrates the route from 0 K energies to measured heats.","marker":"[47]"},{"why":"Provides MP2-based reference adsorption thermodynamics for alkanes in chabazite, used here to validate the periodic MP2 values.","marker":"[76]"},{"why":"Documents the expectation that MP2 is close to CCSD(T) for finite zeolite clusters, the premise this paper tests and refines.","marker":"[46]"}],"fun_headline_variants":["RPA+singles outdoes MP2 in zeolite adsorption","Zeolite adsorption: RPA singles beat MP2","Singles-fixed RPA tops MP2 for adsorption accuracy","RPA+RSE surpasses MP2 in zeolite binding","Near-CCSD(T) adsorption from RPA singles"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the 2T and 4T finite cluster models faithfully represent the periodic chabazite adsorption site, so the accuracy ranking measured against CCSD(T) on those clusters transfers to the bulk material where no CCSD(T) reference exists.","fun_headline_variants_meta":{"raw":{"variants":["RPA+singles outdoes MP2 in zeolite adsorption","Zeolite adsorption: RPA singles beat MP2","Singles-fixed RPA tops MP2 for adsorption accuracy","RPA+RSE surpasses MP2 in zeolite binding","Near-CCSD(T) adsorption from RPA singles"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1324,"prompt_tokens":890,"completion_tokens":434,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":506,"completion_tokens_details":{"reasoning_tokens":350}},"tokens_in":506,"tokens_out":434,"duration_ms":4917,"temperature":1.0,"reasoning_tokens":350,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:10:37.192321+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute adsorption energies for the same seven molecules on a larger cluster model, for example a full double-six-ring or 6T site, with CCSD(T)-level theory and compare with RPA+RSE and MP2; if RPA+RSE errors exceed 2 kJ/mol or MP2 matches or beats it, the central claim fails. A complementary test is to compare RPA+GWSE 0 K energies to experimental adsorption enthalpies after including zero-point and thermal corrections, since a systematic drift beyond 2 kJ/mol would also refute the claimed reference quality.","supporting_citations":[{"cited_title":"Rube s , author M","cited_arxiv_id":null,"evidence_quote":"Applies RPA with singles to CO adsorption in a zeolite and demonstrates the route from 0 K energies to measured heats."},{"cited_title":"Piccini , author M","cited_arxiv_id":null,"evidence_quote":"Provides MP2-based reference adsorption thermodynamics for alkanes in chabazite, used here to validate the periodic MP2 values."},{"cited_title":"Tuma \\ and\\ author J","cited_arxiv_id":null,"evidence_quote":"Documents the expectation that MP2 is close to CCSD(T) for finite zeolite clusters, the premise this paper tests and refines."}],"review_version":1}