{"id":"0c821016-ab6b-4b89-bafb-be477e1c82a0","arxiv_id":"2412.17204","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An automated multilevel embedding framework delivers CCSD(T)-quality adsorption enthalpies for 19 adsorbate-ionic-surface systems at near-DFT cost.","lead":"Scientists built an automated, open-source framework that applies gold-standard coupled cluster calculations to molecules on ionic crystal surfaces at a fraction of the usual cost. It reproduces experimentally measured adsorption strengths for 19 molecule-surface combinations and settles long-running disputes about how several molecules bind.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Open-shell NO monomer energies rely on CCSD(T) corrections of 290–660 meV despite the paper's own caveat that CCSD(T) performs poorly for open-shell radicals; a multireference cross-check is needed before accepting the >80 meV monomer/dimer gap.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing concern: the reliability of DLPNO-CCSD(T) with an unrestricted Hartree-Fock reference for open-shell NO monomers, given the anomalously large ΔCC corrections. My independent reading of the main text, Methods S5, Table S8 and Table S12 confirms that this is the single point on which the strongest specific claim—the >80 meV stabilization of the NO dimer over all monomer configurations—depends. The paper acknowledges in the Discussion that CCSD(T) performs poorly for open-shell radicals, yet uses it for all monomer configurations without reporting multireference diagnostics or an alternative benchmark. The inconsistency is reinforced by the paper's own use of MRMP2 rather than CCSD(T) for the NO dimer cohesive energy (Sec. S7.4). This is not a disagreement with consensus; it is a correctness risk internal to the argument, because the magnitude of ΔCC indicates the method is operating outside its trusted regime. The reader's CONDITIONAL verdict is appropriate: the broad 19-system framework and the closed-shell results are credible and well supported, but the NO monomer conclusion requires an additional check. Since my concern matches the reader's, I recommend no verdict change and mark agreement as full.","tokens_in":61982,"tokens_out":2434,"duration_ms":28264,"concrete_test":"Recompute the SKZCAM-cluster interaction energies for the Bent-Bridge and Bent-O NO monomer configurations with a multireference method—for example, MRMP2 or CASPT2 with an active space spanning the NO π/π* orbitals and relevant surface states, or a DMC benchmark on the same embedded cluster—and compare the resulting Eint to the DLPNO-CCSD(T) values used in Table S12. If the multireference monomer Eint differs by more than ~80 meV, or if the monomer–dimer Hads gap falls below 80 meV, the central NO-dimer claim is not established and the paper's verdict should be reconsidered.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest specific claim is that NO on MgO(001) binds as the cis-(NO)2 dimer, with all monomer configurations at least 80 meV less stable. This conclusion depends on the autoSKZCAM interaction energies for five open-shell NO monomer configurations, which were computed with DLPNO-CCSD(T) from an unrestricted Hartree-Fock reference (Methods S5). The paper's Discussion explicitly states that CCSD(T) 'performs poorly for open-shell molecules of radical character,' and the magnitude of the ΔCC corrections in Table S8 confirms the concern: Bent-Bridge NO has ΔCC = +626 meV, Bent-O NO has +659 meV, and Vertical-Hollow NO has +289 meV, whereas nearly all closed-shell systems have ΔCC below 35 meV. These corrections are so large that the final monomer Eint values (e.g., Bent-Bridge: MP2 -661 meV plus ΔCC +626 meV plus -26 meV corrections gives about -62 meV) are almost entirely determined by an unreliable correction. If the true ΔCC for these open-shell monomers is smaller, or of opposite sign, the monomer energies could shift by hundreds of meV and the claimed >80 meV separation from the dimer would collapse. The paper itself uses MRMP2 rather than CCSD(T) for the NO dimer cohesive energy because CCSD(T) underbinds that quantity; a parallel treatment for the monomer interaction energies is missing. No multireference diagnostics (T1, D1, %T1) are reported for these open-shell cases. This is the load-bearing soft spot: the headline agreement for NO, and the resolution of the NO adsorption debate, rests on CCSD(T) being reliable precisely in the regime where the authors state it is not.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces autoSKZCAM, an automated embedded-cluster workflow that computes CCSD(T)-level interaction energies for adsorbates on ionic surfaces through the SKZCAM protocol, mechanical embedding, and local correlation approximations, with DFT ensembles used for relaxation and vibrational contributions. The framework is applied to 19 adsorbate–surface systems on MgO(001), TiO2 rutile(110), and anatase(101), with additional validation on LiH(001) and NaCl(001), and is used to benchmark density functional approximations and to address literature debates on adsorption geometries, most prominently the claim that NO on MgO(001) binds as a cis-(NO)2 dimer with all monomer configurations at least 80 meV less stable.","tokens_in":62359,"tokens_out":8599,"duration_ms":79275,"significance":"If the central results hold, this is a substantial methodological advance: it brings correlated wave-function-level adsorption energetics to a cost competitive with periodic hybrid DFT, automates a previously manual protocol, and provides a reusable open-source tool plus a CCSD(T)-level benchmark dataset for DFA assessment. The authors also deserve credit for transparent error propagation, for using experimental data purely as a benchmark rather than as fitting targets, and for validating the protocol against independent DMC and CCSD(T) results for LiH and NaCl. However, the strongest specific claim—the resolution of the NO adsorption debate—rests on DLPNO-CCSD(T) corrections of several hundred meV for open-shell radical monomers, a regime the paper itself identifies as problematic for CCSD(T); until a multireference cross-check is supplied, that particular conclusion should be treated as provisional.","major_comments":[{"comment":"The ΔCC corrections for the five NO monomer configurations are +289, -68, +626, -48, and +659 meV, whereas the closed-shell systems have corrections below 35 meV in magnitude. These corrections dominate or reverse the MP2 interaction energies: for Bent-Bridge, the MP2 Eint of -661 meV becomes -62 meV after the ΔCC and other corrections (Table S12). The paper's Discussion explicitly states that CCSD(T) 'performs poorly for open-shell molecules of radical character,' and for the NO dimer cohesive energy the authors deliberately switch to MRMP2 because CCSD(T) underbinds it (SI §S7.4, Table S19). No T1, D1, or %T1 diagnostics, and no alternative multireference or DMC calculation, are reported for the NO monomer configurations. Because the >80 meV monomer–dimer separation and the resolution of the NO adsorption debate (main-text Fig. 3, SI Table S1) rest on these energies, this is a load-bearing gap that needs to be addressed with a multireference cross-check for at least the most stable monomer configurations.","section":"SI §S6.3, Table S8; Methods §S5; Discussion"},{"comment":"The blanket statement that all 19 systems 'reproduce' experimental Hads is weakened by the size of the reported uncertainties for several systems: C6H6 has a total uncertainty of ±100 meV, Vertical-Hollow NO has Hads = 68 ± 91 meV, and H2O on rutile(110) has Hads = -1007 ± 57 meV. With uncertainties of this magnitude, agreement within error bars is a much weaker test than for systems with ±20 meV errors. The authors should qualify 'reproduced' with a precision-sensitive statement or separately identify which systems achieve chemical accuracy (43 meV), since the screening window of ~150 meV cited in the Introduction makes this distinction important.","section":"Main text Fig. 2; SI Table S30"}],"minor_comments":[{"comment":"In the sentence 'A similar confidence interval has been calculated for the the individual terms', the duplicated article 'the the' should be corrected.","section":"Main text, Results"},{"comment":"The phrase 'computational costs approaching DFT' is supported for the interaction-energy step, but the full Hads workflow also requires periodic DFT geometry optimizations and vibrational frequency calculations; the sentence should specify this scope.","section":"Abstract"},{"comment":"The definition of ϵgeom as twice the RMSE over the DFA ensemble assumes that the error distribution is symmetric around the revPBE-D4 result; this assumption should be stated explicitly because the final error bars in Table S30 rest on it.","section":"SI §S8.3"},{"comment":"The GitHub link is useful, but for archival reproducibility the authors should deposit a versioned release with a DOI (e.g., Zenodo) and cite it.","section":"Code Availability"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong methodological contribution and well within the scope of a broad physical chemistry journal. My main concern is the open-shell NO monomer treatment: the ΔCC corrections are so large that the monomer–dimer separation is not yet established. A targeted multireference or DMC calculation for the NO monomer configurations, plus a more cautious wording of the 19-system reproduction claim, would be sufficient; I would not require recalculating all systems. The open-source code, transparent error analysis, and independent validation on LiH/NaCl are clear strengths."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real advance, and the benchmark dataset is worth having. The autoSKZCAM framework automates the SKZCAM protocol, adds Δbasis/Δcore ONIOM layers, and scales to 19 adsorbate–surface systems including clusters and monolayers at roughly hybrid-DFT cost. The error analysis is unusually careful — 95% intervals everywhere, no parameters fitted to the target adsorption enthalpies, and cross-checks against DMC and independent CCSD(T) for CO/MgO, H2O/MgO, LiH and NaCl. The open-source code and deposited data make the work reproducible. That part deserves credit.\n\nThe soft spot is exactly where the reader and stress-test put it. The dimer-vs-monomer conclusion for NO on MgO rests on DLPNO-CCSD(T) corrections of +289 to +659 meV for five open-shell monomer configurations, where closed-shell systems sit below 35 meV. Those corrections flip MP2 binding energies of −650 to −660 meV to near-zero final values, so the monomer energies are almost entirely defined by a method the paper's own Discussion says 'performs poorly for open-shell molecules of radical character.' No T1/D1 diagnostics are reported, and for the NO dimer cohesive energy the authors themselves abandon CCSD(T) for MRMP2 because CCSD(T) underbinds. I think the physical conclusion is probably right — FTIR and EPR independently say dimer, and the dimer Hads matches experiment — but the quantitative '>80 meV' gap is not established by these numbers. A multireference cross-check on 2–3 of the monomers would settle it.\n\nTwo smaller things. First, several headline 'reproductions' carry ±90–100 meV error bars (C6H6, vertical-hollow NO), which is honest but weakens the reproduction claim for exactly those systems. Second, the CO2/MgO resolution leans on re-analyzing the Meixner physisorption experiment at a higher desorption temperature so it lands on the chemisorbed value; that's a post hoc reinterpretation, plausible but convenient. The Campbell–Sellers re-analysis of experimental ν also moves the target (RMSD drops from 102 to 58 meV), though it's transparently documented.\n\nWho gets value: anyone doing oxide surface chemistry, and especially DFA developers — the Eint benchmark table is the most immediately usable output. It deserves serious refereeing; my recommendation is to engage, and to ask for multireference diagnostics on the NO monomers before the debate is declared closed.","headline":"A genuinely useful, open-source embedding framework and benchmark set with an honest error budget — but the NO-on-MgO monomer energies ride on 290–660 meV CCSD(T) corrections in a regime the authors themselves flag as unreliable.","tokens_in":62947,"tokens_out":4709,"would_cite":true,"duration_ms":43120,"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":"Adsorption enthalpies on ionic surfaces can be computed with coupled-cluster accuracy at roughly the cost of hybrid density functional theory, as demonstrated on 19 experimental measurements.","keywords":["adsorption enthalpy","coupled cluster theory","embedded cluster method","surface chemistry","ionic materials","density functional benchmarking","NO dimer on MgO(001)","SKZCAM protocol"],"falsifier":"Run a multireference method (for example CASPT2/NEVPT2 or diffusion Monte Carlo) on the same SKZCAM cluster series for the five NO monomer configurations; if any monomer interaction energy shifts by more than roughly 80 meV relative to the reported DLPNO-CCSD(T) values, the claim that all monomer configurations are at least 80 meV less stable than the cis-(NO)2 dimer is overturned.","tokens_in":61798,"feed_emoji":"🧪","tokens_out":11913,"duration_ms":100608,"temperature":0.7,"pith_summary":"This paper claims that adsorption enthalpies for molecules on ionic-material surfaces can be computed with gold-standard coupled-cluster accuracy for roughly the cost of a hybrid density functional calculation. The central proposal is an automated multilevel embedding workflow that splits the adsorption enthalpy into an interaction energy treated with CCSD(T) and a set of smaller relaxation, zero-point, and thermal terms treated with a six-functional DFT ensemble. On 19 adsorbate–surface systems spanning almost 1.5 eV of binding strength, the computed enthalpies fall within experimental error bars. The same machinery is used to settle configuration debates, most notably predicting that NO adsorbs on MgO(001) as a covalently bonded cis-(NO)2 dimer, with all monomer configurations at least 80 meV less stable. If this holds, correlated wave-function methods become a routine, not heroic, tool for surface chemistry screening and DFT benchmarking.","feed_headline":"19 adsorption enthalpies reproduced at near-DFT cost","feed_subtitle":"An automated embedded-cluster workflow reproduces 19 adsorption enthalpies and settles disputed binding geometries.","key_machinery":"The load-bearing machinery is the SKZCAM protocol, a set of rubrics that automatically generates a converging series of embedded quantum clusters for the adsorbate–surface system, surrounded by formal point charges and effective core potentials that reproduce the ionic environment. Interaction energies at the MP2 level on moderately sized clusters are extrapolated to the bulk limit, then a small ΔCC correction raises the result to CCSD(T) using local natural-orbital or domain-based local pair-natural-orbital approximations on smaller clusters; additional Δbasis and Δcore corrections recover basis-set and semi-core correlation effects. This mechanical embedding is layered in an ONIOM-style scheme, and the automation adds further intermediate layers that cut the cost of a CO-on-MgO interaction energy to about 600 CPU-hours, two orders of magnitude below the previous manual implementation. The remaining relaxation, zero-point, and thermal terms come from a six-functional DFT ensemble, with a quasi-rigid-rotor harmonic-oscillator treatment of low-frequency modes, giving conservative error bars on the final Hads.","core_discovery":"The paper's central claim is that the autoSKZCAM framework delivers CCSD(T)-quality adsorption enthalpies Hads for ionic surfaces at a cost comparable to periodic hybrid DFT. The framework partitions Hads = Eint + Erlx + EZPV + ET − RT, computing the dominant interaction energy Eint with local coupled-cluster theory embedded in an electrostatic point-charge environment and extrapolated to the bulk limit via the SKZCAM protocol, while obtaining geometric relaxation, zero-point, and thermal contributions from an ensemble of six density functional approximations. Against experiment, the framework reproduces the adsorption enthalpy for 19 adsorbate–surface systems, including monolayers and molecular clusters, and identifies the most stable adsorption configuration for systems where DFT studies disagree. In the specific case of NO on MgO(001), it finds the cis-(NO)2 dimer configuration most stable with Hads consistent with experiment, while all five monomer configurations are more than 80 meV less stable. It also predicts chemisorbed carbonate for CO2 on MgO(001), a tilted geometry for CO2 on rutile(110), a parallel geometry for N2O on MgO(001), and partially dissociated hydrogen-bonded tetramers for H2O and CH3OH on MgO(001).","pith_inferences":["The paper's own error analysis identifies the DFT-geometry error ϵgeom as the largest contribution to the Hads error bars; this suggests that obtaining reference geometries from a more accurate level of theory would tighten predictions most for strongly chemisorbed systems such as CO2 on MgO(001), where ϵgeom reaches 188 meV.","The NO monomer conclusion rests on an unrestricted-Hartree-Fock DLPNO-CCSD(T) treatment with ΔCC corrections as large as 659 meV, inside a regime the paper itself flags as questionable for CCSD(T); this suggests a multireference or quantum Monte Carlo check on the monomer clusters as the natural next test.","The point-charge-plus-effective-core-potential environment limits the framework to insulating ionic surfaces; extending the same partition-of-Hads idea to metals or covalent materials would require replacing that environment with a quantum embedding that couples the cluster to its surroundings through the electron density or Green's function.","The interaction-energy benchmark table may be the most durable scientific output, since it gives DFT developers reference values on surfaces, a class of systems the paper notes is underrepresented in existing benchmark sets."],"forward_implications":["Correlated wave-function methods become applicable routinely, not just for one or two showcase systems, so multiple adsorption configurations can be compared at CCSD(T) quality with costs approaching those of hybrid DFT.","The 19-system dataset provides adsorption interaction-energy benchmarks for density functional approximations: PBE-MBD/FI and rev-vdW-DF2 come closest to the CCSD(T) values (mean absolute deviations of 26 and 25 meV), RPA underbinds the MgO(001) subset (58 meV), and PBE-D3, SCAN-rVV10 and r2SCAN-D4 overbind.","The configuration rulings resolve long-standing debates: NO on MgO(001) is a cis-(NO)2 dimer, CO2 on MgO(001) chemisorbs as a carbonate, CO2 on rutile(110) prefers a tilted geometry, N2O on MgO(001) binds parallel, and H2O and CH3OH adsorb as partially dissociated clusters on MgO(001).","Because the workflow is open source, it can operate as a screening tool in catalyst discovery and generate reference interaction energies for fitting machine-learned density functionals."],"supporting_citations":[{"why":"Introduces and validates the SKZCAM embedded-cluster protocol whose cluster series and bulk extrapolation carry the interaction-energy calculation.","marker":"48,49,51"},{"why":"Supplies the local natural-orbital CCSD(T) approximation that makes the cluster calculations affordable.","marker":"52,53"},{"why":"Supplies the domain-based local pair-natural-orbital CCSD(T) used for the open-shell NO monomer configurations.","marker":"54–57"},{"why":"Provides the compiled single-crystal temperature-programmed desorption measurements that the 19 computed adsorption enthalpies are compared against.","marker":"59"},{"why":"Provides the system-specific adsorbate entropies used to re-analyze experimental pre-exponential factors and set the experimental error bars.","marker":"60"},{"why":"Introduces the quasi-rigid-rotor harmonic-oscillator treatment used to convert static adsorption energies into adsorption enthalpies.","marker":"61"},{"why":"Experimental infrared evidence that NO forms dimers on MgO(001), corroborating the predicted most-stable configuration.","marker":"62"},{"why":"Experimental electron-paramagnetic-resonance evidence that NO forms dimers on MgO(001), corroborating the predicted most-stable configuration.","marker":"63"},{"why":"The temperature-programmed desorption experiment reporting a weak physisorbed CO2 enthalpy on MgO(001) that the framework argues is inconsistent.","marker":"67"},{"why":"The temperature-programmed desorption experiment reporting a strong chemisorbed CO2 enthalpy on MgO(001) that the framework reproduces.","marker":"68"}],"fun_headline_variants":["Automated framework: CCSD(T) accuracy at near-DFT cost for ionic surfaces","Near-DFT cost CCSD(T) for surface adsorption: 19 systems match experiments","Embedded-cluster workflow reproduces 19 adsorption energies cheaply","Open-source framework settles adsorption disputes at DFT-like cost","Ionic surface chemistry: correlated accuracy without the cost"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central numerical claim assumes that DLPNO-CCSD(T) with an unrestricted Hartree-Fock reference reliably captures the interaction energy of the open-shell NO monomer on MgO(001), even though the paper itself notes CCSD(T) performs poorly for open-shell radicals and reports ΔCC corrections up to 659 meV for these monomers.","fun_headline_variants_meta":{"raw":{"variants":["Automated framework: CCSD(T) accuracy at near-DFT cost for ionic surfaces","Near-DFT cost CCSD(T) for surface adsorption: 19 systems match experiments","Embedded-cluster workflow reproduces 19 adsorption energies cheaply","Open-source framework settles adsorption disputes at DFT-like cost","Ionic surface chemistry: correlated accuracy without the cost"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001124,"raw_usage":{"total_tokens":4700,"prompt_tokens":992,"completion_tokens":3708,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":608,"completion_tokens_details":{"reasoning_tokens":3613}},"tokens_in":608,"tokens_out":3708,"duration_ms":24576,"temperature":1.0,"reasoning_tokens":3613,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T05:42:20.642083+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a multireference method (for example CASPT2/NEVPT2 or diffusion Monte Carlo) on the same SKZCAM cluster series for the five NO monomer configurations; if any monomer interaction energy shifts by more than roughly 80 meV relative to the reported DLPNO-CCSD(T) values, the claim that all monomer configurations are at least 80 meV less stable than the cis-(NO)2 dimer is overturned.","supporting_citations":[],"review_version":1}