{"id":"c4ebb923-3366-4aa9-a56f-8156ce5cb9cd","arxiv_id":"2412.18395","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Frozen natural spinors and Cholesky decomposition are combined in an X2CAMF-based two-component CCSD/CCSD(T) implementation that matches four-component accuracy for tested molecules at much lower cost.","lead":"This paper combines two established approximations, Cholesky decomposition and frozen natural spinors, in a two-component relativistic coupled cluster method to lower storage and compute time. The authors benchmark accuracy against four-component calculations and show the method can handle a medium-sized uranium complex on a single workstation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The uranium flag-ship run is at the LOOSE FNS threshold with no canonical or tighter reference, so the 'similar accuracy to 4c' claim is untested in the large-system regime, and the abstract's 'over 1000 virtual spinors' contradicts the 562 actually correlated.","rationale":"The reader's weakest assumption identifies MP2-based FNS ranking as load-bearing; I agree it is central, but I think the more falsifiable gap is the absence of any validation at the exact setting used for the paper's headline demonstration. The small-molecule benchmarks (Table 1) and HX comparisons to 4c (Tables 2-3) provide real support; they are not the problem. The problem is extrapolating from those benchmarks to a LOOSE-threshold, 562-virtual uranium complex with no reference energy. The abstract's 'over 1000 virtual spinors' is an internal inconsistency: the full virtual space is 1392, but only 562 are correlated after truncation, so the demonstration as written cannot support the 'routine >1000' efficiency claim. I would keep the reader's CONDITIONAL verdict; the method is plausible and has genuine benchmark support, but the flagship demonstration needs a tighter-threshold/full-virtual validation and a corrected description of what was correlated. This is a request for evidence, not an accusation: the promoted large-system capability is simply not yet demonstrated by the reported numbers.","tokens_in":16479,"tokens_out":12493,"duration_ms":118013,"concrete_test":"Rerun the Section 4.4 [UO2(NO3)3]- CCSD calculation with NORMALFNS and TIGHTFNS thresholds, and if feasible also with the full canonical virtual space in the same CD-X2CAMF implementation, recording the retained virtual count and total CCSD energy at each setting. If the LOOSE-to-TIGHT energy difference exceeds the claimed accuracy (about 1 kcal/mol) or is not captured by the Delta-EMP2 correction of Eqs. (35)-(37), then the LOOSE-threshold flagship calculation does not establish the method's accuracy in the large-system regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that FNS-CD-X2CAMF-CC gives 'similar accuracy as canonical four-component ... at a fraction of the cost.' This hinges on the MP2-based FNS truncation and its Delta-EMP2 correction (Eqs. 35-37) keeping truncation error small. In the flagship application (Section 4.4, [UO2(NO3)3]-), the calculation is run only at the LOOSEFNS threshold (FNS 1e-4, CD 1e-3), which reduces the virtual space from 1392 to 562 spinors, and no comparison is made to the canonical full-virtual CD-X2CAMF result or to a tighter FNS threshold. Table 1 shows LOOSE settings already have absolute maximum errors of 0.89 kcal/mol (CCSD) and 0.57 kcal/mol (CCSD(T)) in small benchmark complexes, so the threshold is not self-evidently negligible. The abstract's statement that the demonstration involves 'the correlation of over 1000 virtual spinors' is contradicted by the 562 correlated virtuals reported in Section 4.4. Thus the advertised routine >1000-spinor capability and the accuracy of the large-system demonstration are both unsupported; the central claim should be treated as conditional until a tighter-threshold or full-virtual reference validates this regime.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript describes an implementation of frozen natural spinors (FNS) combined with Cholesky decomposition (CD) in an exact two-component atomic mean-field (X2CAMF) based coupled cluster framework, covering CCSD and CCSD(T). The virtual space is truncated using MP2-based natural spinors, a perturbative correction is applied for the truncation error, and three- and four-virtual integrals are generated on the fly from Cholesky vectors. The paper benchmarks the method with dissociation enthalpies for 18 coinage-metal cation complexes, compares bond lengths and harmonic frequencies of the hydrogen halides against canonical four-component and FNS four-component references, reports storage and timing comparisons for HI, and presents a CCSD calculation on [UO2(NO3)3]−.","tokens_in":16736,"tokens_out":8420,"duration_ms":80473,"significance":"If the central claim is accepted, the method offers a practical route to relativistic CCSD/CCSD(T) for medium-sized heavy-element systems with substantially reduced storage and floating-point costs. The manuscript has clear strengths: the FNS and CD thresholds are predefined convergence parameters rather than fitted values; the 18-complex dissociation enthalpy set provides statistical measures beyond a single error; the HX section compares directly with canonical four-component results; and the HI section gives concrete storage and timing comparisons. The main gaps are that the headline accuracy claim rests on only four closed-shell HX molecules, and the large-system uranium demonstration lacks a canonical or tighter-threshold reference; these issues need to be addressed before the claim can be taken at face value.","major_comments":[{"comment":"The abstract and Introduction state that the uranium demonstration involves \"the correlation of over 1000 virtual spinors,\" but §4.4 reports that after LOOSEFNS truncation the [UO2(NO3)3]− calculation has 562 correlated virtual spinors out of 1392 canonical virtuals. This internal inconsistency should be corrected. More importantly, the uranium calculation is not benchmarked against the canonical full-virtual CD-X2CAMF result or against a tighter FNS threshold. Since Table 1 shows that LOOSEFNS already gives absolute maximum errors of 0.89 kcal/mol (CCSD) and 0.57 kcal/mol (CCSD(T)) in the small benchmark set, the large-system accuracy component of the central claim is unsupported. A tighter-threshold or full-virtual reference, or at least an explicit error estimate for the LOOSEFNS truncation in this system, is needed.","section":"Abstract/Introduction and §4.4"},{"comment":"The conclusion that the FNS-CD-X2CAMF approach gives \"similar accuracy as that of the canonical four-component relativistic coupled cluster method\" is based on only four closed-shell hydrogen halides. The agreement at the TIGHT and NORMAL thresholds is good, but this set is too narrow to establish the broad claim for heavy-element, open-shell, and actinide systems such as the uranium complex in §4.4. Either extend the four-component comparison to more representative systems or explicitly qualify the claim as demonstrated only for the HX series.","section":"§4.2, Tables 2–3"},{"comment":"The ΔEMP2 correction for the frozen-virtual truncation error is a key assumption, but the manuscript reports only the final corrected errors. Since this assumption is what makes the FNS approach reliable, please provide the uncorrected FNS errors in Table 1 or the Supporting Information to demonstrate that the MP2 correction is actually improving the CCSD/CCSD(T) results, and state explicitly whether Table 1 includes the correction. Without this decomposition, the reader cannot assess whether the method is robust across the threshold settings.","section":"§2.4, Eqs. (35)–(37)"}],"minor_comments":[{"comment":"The heading contains a typo: \"Comparision\" should be \"Comparison.\"","section":"§4.2 heading"},{"comment":"Please clarify the caption and column labels in Tables 2 and 3: the \"4c\" column contains the reference value, while the remaining numeric columns are errors relative to that reference. It would also help to state that the FNS-4c error values are taken from Ref. 40.","section":"Tables 2–3"},{"comment":"The notation ΔECCSD/CCSD(T) = ECanonical − EFNS is ambiguous; please state the sign convention and specify whether all reported errors are FNS − canonical or canonical − FNS, so that the direction of the correction is clear.","section":"§2.4, Eq. (35)"},{"comment":"In Table 4, the \"—\" entries for VVVV and OVVV in the FNS-CD-X2CAMF columns should be explained in the caption, because these integrals are not stored or formed in the present implementation; without that explanation the entries look like missing data.","section":"Table 4"},{"comment":"The concluding paragraph says the method \"can be routinely used for accurate relativistic calculations of small molecules,\" but the preceding demonstration is for a medium-sized uranium complex; please revise to \"small and medium-sized molecules\" or otherwise reconcile the wording.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reports a useful implementation, and I am not recommending rejection. However, the abstract's \"over 1000 virtual spinors\" statement is factually inconsistent with Section 4.4, and the flagship uranium calculation currently does not validate accuracy in the large-system regime. The authors should correct the claim and either add a tighter-threshold or full-virtual reference for the uranium system or clearly limit the accuracy claim to the benchmarked systems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real implementation advance, not a repackaging. The direct formation of LVV Cholesky vectors in the FNS basis is the key move; it avoids storing the full virtual block and yields a 38x speedup over canonical 4c CCSD on HI. The 18-complex benchmark with three thresholds is carefully done, and the NORMALFNS errors (MAE 0.13 kcal/mol CCSD(T)) are reasonable. The HX tables show FNS-CD-X2CAMF matching FNS-4c errors at similar thresholds, which supports the 'similar accuracy' claim as far as it goes. The MP2-based FNS ranking is a standard approach, properly cited, so I don't see a novel weakness there.\n\nSoft spots: the abstract says the uranium application involves 'correlation of over 1000 virtual spinors,' but Section 4.4 reports 562 correlated virtuals after truncation. That is a concrete mismatch and must be fixed. On a related note, the uranium run uses only the LOOSE threshold with no canonical or tighter reference, so the accuracy in that regime is unverified—Table 1 shows LOOSE errors up to 0.89 kcal/mol in smaller systems. The four-component accuracy baseline covers only four hydrogen halides, which is thin for a general statement, though the results there are consistent. No code or data are released, which is common for quantum chemistry implementation papers and not a dealbreaker.\n\nThese are fixable issues, not fatal ones. The method itself, the timing numbers, and the benchmark design are credible. This paper is for people working on relativistic CC for heavy-element systems, and they will want to see it. It deserves a serious referee. My recommendation: send it to review, but require a corrected abstract and either a tighter-threshold or full-virtual reference for the uranium complex.","headline":"Genuine new implementation combining FNS with CD in X2CAMF-CC; solid benchmarks, but fix the overstated 1000-spinor claim and validate the loose-threshold uranium run before trusting the central claims.","tokens_in":17280,"tokens_out":2148,"would_cite":true,"duration_ms":21055,"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 the FNS-CD-X2CAMF-CCSD/CCSD(T) implementation reproduces four-component relativistic coupled cluster accuracy for heavy-element molecules while cutting storage and wall time by one to two orders of magnitude.","keywords":["frozen natural spinors","Cholesky decomposition","exact two-component","atomic mean-field spin-orbit","relativistic coupled cluster","CCSD(T)","heavy-element chemistry","occupation number truncation"],"falsifier":"Compute the CCSD(T) atomization energy of a heavy-element hydride such as AtH with the NORMALFNS threshold and compare it to the canonical four-component CCSD(T) value; if the FNS truncation plus MP2 correction error exceeds about 1 kcal/mol, the MP2 occupation ranking is not reliable for spin-orbit-coupled virtual spinors.","tokens_in":16290,"feed_emoji":"⚛️","tokens_out":5646,"duration_ms":50558,"temperature":0.7,"pith_summary":"The paper presents an implementation of relativistic coupled cluster (CCSD and CCSD(T)) for heavy-element molecules that combines three existing ideas: the exact two-component atomic mean-field Hamiltonian (X2CAMF), Cholesky decomposition of two-electron integrals, and frozen natural spinors (FNS) built from an MP2 density matrix. The claim is that this combination reproduces the accuracy of canonical four-component relativistic coupled cluster at a small fraction of the cost, while removing the storage bottleneck that kept such calculations to small molecules. The central evidence is a series of benchmarks: 18 coinage-metal cation complexes for ligand dissociation enthalpies, spectroscopic constants for the hydrogen halides, timing comparisons on HI, and a CCSD calculation on the uranyl nitrate complex with 562 correlated virtual spinors. A reader who cares about heavy-element chemistry would care because the method offers a practical path to quantum-chemical accuracy on systems where four-component CC was previously infeasible on a single computing node.","feed_headline":"Relativistic CC matches 4-component accuracy at 38x speed","feed_subtitle":"Frozen natural spinors and Cholesky decomposition cut storage and cost for heavy-element molecules.","key_machinery":"Frozen natural spinors are the eigenvectors of the correlated one-body density matrix; here the virtual-virtual block is built from MP2 amplitudes, diagonalized, and truncated by occupation-number threshold, then re-canonicalized. Cholesky decomposition writes the two-electron integrals as sums of products of Cholesky vectors, so three- and four-virtual integrals are reconstructed on the fly instead of stored. The X2CAMF Hamiltonian puts spin-orbit coupling into an effective one-electron atomic mean-field operator, leaving the two-electron part nonrelativistic. Together these ingredients let the implementation transform only LOO- and LOV-type Cholesky vectors into the canonical basis, form LVV vectors directly in the FNS basis, and never construct four- or three-virtual integrals.","core_discovery":"The central discovery is that MP2-derived natural spinor occupation numbers, combined with a Cholesky-decomposed two-component Hamiltonian whose spin-orbit coupling enters through an atomic mean-field potential, yield a compact virtual space whose CCSD/CCSD(T) energies differ from canonical four-component results by only a few tenths of a kcal/mol for thermochemical benchmarks and by under a milliangstrom and a few wavenumbers for the hydrogen halide series, once an MP2 correction for the truncation error is added. The cost saving is not incremental: for HI, the FNS-CD-X2CAMF-CCSD calculation is about 38 times faster than canonical four-component CCSD and 11 times faster than the four-component FNS-CCSD, with the formerly dominant VVVV and OVVV integrals no longer stored at all.","pith_inferences":["The paper benchmarks only ground-state energies and dissociation enthalpies; nothing here says the FNS truncation is safe for excited states, response properties, or spin-orbit-induced near-degeneracies, and those would need separate tests.","Because the Cholesky and FNS thresholds enter independently, a two-parameter error surface could be mapped to find the cheapest threshold pairing for a target accuracy; the paper only scans three discrete settings.","The same MP2-based FNS ranking could be reused to accelerate equation-of-motion CC or analytic gradients, but the perturbative MP2 correction would have to be revalidated for energy differences.","If a system is substantially multi-reference, MP2 occupation numbers can mis-order spinors; a cheap diagnostic would be comparing MP2 and CCSD natural spinor occupations for a difficult heavy atom before trusting the truncation."],"forward_implications":["At the TIGHTFNS threshold (FNS 10^-5, CD 10^-5), errors for bond lengths and harmonic frequencies of HX (X=F, Cl, Br, I) stay within about 0.0003 Å and 3 cm^-1 of canonical four-component results, so the method can replace canonical four-component CC for spectroscopic constants.","With the NORMALFNS threshold, mean absolute errors against canonical CD-X2CAMF-CC remain below 0.14 kcal/mol for CCSD and CCSD(T) over 18 metal-ligand dissociation enthalpies, so one threshold setting can be recommended for thermochemistry.","The method removes the need to store four-virtual (VVVV) and three-virtual (OVVV) integrals; only OOVV and OOOV type integrals are stored, and the particle-particle ladder contraction becomes a dominant cost.","For the uranyl nitrate complex, a CCSD calculation with 562 correlated virtual spinors and 2441 Cholesky vectors completed in about two days on one node, indicating that mid-sized heavy-element systems are accessible without a multi-node cluster."],"supporting_citations":[{"why":"Supplies the four-component frozen natural spinor framework whose virtual-space reduction and perturbative correction the present method adapts.","marker":"[40]"},{"why":"Provides the Cholesky-decomposed two-component X2CAMF-CC framework that the present work extends with natural spinors.","marker":"[36]"},{"why":"Establishes that MP2-generated frozen natural orbitals work in relativistic two-component correlated calculations, justifying the MP2 route used here.","marker":"[43]"},{"why":"Introduces frozen natural orbitals with a perturbative correction for coupled cluster truncation, the direct nonrelativistic ancestor of the FNS+MP2 correction used here.","marker":"[55]"},{"why":"Shows that occupation-based natural orbital truncation plus a perturbative correction is an accurate cost-reduction strategy for coupled cluster, supporting the choice of occupation-number thresholds.","marker":"[57]"},{"why":"Supplies the atomic mean-field spin-orbit approximation that defines the X2CAMF effective one-electron Hamiltonian.","marker":"[27]"}],"fun_headline_variants":["Frozen spinors and Cholesky cut cost of relativistic CC","38x speedup for relativistic CC with 4-component accuracy","Frozen natural spinors: cheaper 2-component relativistic CC","MP2 natural spinors speed up relativistic CC by 38x","Cholesky + frozen spinors: heavy-element CC at low cost"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole approach rests on the premise that MP2 occupation numbers rank the most important virtual spinors for CCSD and CCSD(T), and that the MP2 correction for the truncated virtual space is close to the CCSD/CCSD(T) correction; if that ranking is wrong for spin-orbit-coupled heavy elements, the reported agreement with canonical four-component results would not hold.","fun_headline_variants_meta":{"raw":{"variants":["Frozen spinors and Cholesky cut cost of relativistic CC","38x speedup for relativistic CC with 4-component accuracy","Frozen natural spinors: cheaper 2-component relativistic CC","MP2 natural spinors speed up relativistic CC by 38x","Cholesky + frozen spinors: heavy-element CC at low cost"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000759,"raw_usage":{"total_tokens":3313,"prompt_tokens":826,"completion_tokens":2487,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":442,"completion_tokens_details":{"reasoning_tokens":2397}},"tokens_in":442,"tokens_out":2487,"duration_ms":15737,"temperature":1.0,"reasoning_tokens":2397,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T04:42:56.559498+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the CCSD(T) atomization energy of a heavy-element hydride such as AtH with the NORMALFNS threshold and compare it to the canonical four-component CCSD(T) value; if the FNS truncation plus MP2 correction error exceeds about 1 kcal/mol, the MP2 occupation ranking is not reliable for spin-orbit-coupled virtual spinors.","supporting_citations":[{"cited_title":"G.; Bartlett, R","cited_arxiv_id":null,"evidence_quote":"Introduces frozen natural orbitals with a perturbative correction for coupled cluster truncation, the direct nonrelativistic ancestor of the FNS+MP2 correction used here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows that occupation-based natural orbital truncation plus a perturbative correction is an accurate cost-reduction strategy for coupled cluster, supporting the choice of occupation-number thresholds."}],"review_version":1}