{"id":"0034dfa6-747e-4ebe-abf8-195ea760c5d1","arxiv_id":"2505.02930","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Entanglement entropy analysis shows that omitting the second d-shell orbitals overestimates 4s-sigma* correlation in late 3d oxides and mischaracterizes static and dynamic correlation in transition metal diatomics.","lead":"This paper studies why simulating transition metal molecules, from catalysts to molecular qubits, is so hard by zooming in on a known issue called the double d-shell effect. Using quantum information measures, the authors show exactly when a smaller approximation overestimates electron correlation and misleads the calculation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Orbital-basis dependence of mutual information may undermine the \"overestimation\" claim: the 0.25 vs 0.07 comparison is not proven rotation-invariant.","rationale":"The reader's CONDITIONAL verdict is warranted, but I do not think the weakest assumption is primarily the transferability of the entropy-magnitude thresholds. Those thresholds affect only the static/dynamic labels, not the raw numerical observation. The more load-bearing assumption is that orbital-resolved mutual information can be compared across two active spaces whose CASSCF orbital bases are different and, to a large extent, arbitrarily rotated within the active space. Since the entire diagnostic claim rests on this comparison, the orbital-basis dependence should be tested explicitly. The proposed localization scan is decisive and inexpensive because the active spaces are small enough for exact diagonalization, so the wavefunction can be exactly re-expressed in any rotated basis. If the large-active-space 4s-sigma* mutual information is rotation-sensitive, the conclusion that valence-only active spaces overcorrelate the 4s orbital would need to be rephrased in an orbital-invariant or localization-consistent way. I therefore keep the verdict at CONDITIONAL, with this additional robustness condition.","tokens_in":16129,"tokens_out":9407,"duration_ms":112406,"concrete_test":"Using the NiO large-active-space CASSCF wavefunction, apply a unitary rotation among the active orbitals of A1 symmetry (e.g., projection-onto-4s localization or Pipek-Mezey localization) while freezing the wavefunction, then recompute I(4s, sigma*) from Eqs. (1)-(3) in the rotated orbital basis. If the value remains near 0.07 and well below the small-space 0.25, the overestimation claim is robust; if rotations can drive it toward 0.25, the claim is an artifact of the CASSCF orbital gauge. Repeat the same rotation scan on the small-active-space wavefunction so the two active spaces are compared in like-for-like orbital bases.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative result, the reduction of I(4s, sigma*) in NiO from 0.25 to 0.07 upon adding the d' orbitals (Sec. III, Fig. 3a-b), is computed from one- and two-orbital RDMs in the converged CASSCF orbital basis (Sec. II, Eqs. 1-3). Orbital entanglement measures are not invariant under unitary rotations among active orbitals, a point the paper itself acknowledges in the Discussion (refs. 75, 97-101). The small [14,9] and large [14,14] active spaces produce different optimized orbitals, so the functions labeled 4s and sigma* are not the same one-electron orbitals in the two calculations. The drop in mutual information therefore conflates the physical effect of correlating the d' shell with a change in the active-orbital basis, including arbitrary active-active rotations fixed only by CASSCF convergence. The statement that the small active space \"overestimates\" the 4s-sigma* entanglement presupposes this particular orbital basis; without a rotation-invariant check, the magnitude and even the direction of the claimed second-d-shell distortion are not uniquely defined.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the double d-shell (second d-shell) effect in a series of 3d and 4d transition metal hydrides and oxides by comparing active spaces with and without the d' orbitals. Using CASSCF and CASPT2 for geometries and frequencies, and DMRG with M=1000 for near-exact active-space wavefunctions, the authors compute one-orbital entropies and mutual information. They report that adding d' orbitals changes vibrational frequencies and Mulliken charges, that one-orbital entropies of the 3d orbitals are underestimated in the smaller active spaces, and that for late 3d oxides such as NiO the small active space strongly overestimates the 4s-sigma* mutual information (0.245 vs 0.073). They also analyze d-d' interactions in CrH and MoH and conclude that the second d-shell modulates both static and dynamic correlation in a system-dependent way.","tokens_in":16412,"tokens_out":6171,"duration_ms":72600,"significance":"If the central claim holds, the paper provides a practical quantum-information diagnostic: orbital entanglement can flag when a valence-only active space is unbalanced, which would be useful for active-space selection in transition metal chemistry. The calculations are performed with standard methods, the active spaces are small enough that the stated DMRG energy errors (about 1e-6 to 1e-4 a.u.) indicate numerical reliability, and the comparison across a periodic 3d/4d series is a genuine strength. The analysis is parameter-free in the sense that no observables are fitted; the entropy classification thresholds are imported from prior literature. The main reservations concern the orbital-basis dependence of the key mutual-information numbers and the unvalidated transfer of entropy-magnitude classifications to these systems.","major_comments":[{"comment":"The central quantitative result, the reduction of I(4s, sigma*) in NiO from 0.245 to 0.073 upon adding the d' shell, is computed in the converged CASSCF orbital bases of the [14,9] and [14,14] active spaces. These are different optimized orbital sets, so the labels '4s' and 'sigma*' do not denote the same one-electron orbitals in the two calculations; mutual information is not invariant under unitary rotations among active orbitals, as the Discussion itself acknowledges. The statement that the small active space 'overestimates' the 4s-sigma* entanglement is therefore not uniquely defined unless the comparison is made in a common orbital representation. The authors should recompute the entropies for both active spaces in a common localized or natural orbital basis, or provide a rotation-invariant comparison, before drawing the overestimation conclusion.","section":"Sec. III, Fig. 3; Sec. IV, refs. 97-101"},{"comment":"The classification of one-orbital entropy values into non-dynamic (>0.5), static (0.1-0.5), and dynamic (small) correlation regimes is taken from ref. 29 and applied to these 3d/4d hydrides and oxides without re-validation in the present systems. The qualitative conclusions, including the claim that d-d' interactions in CrH are 'likely not nondynamic', depend on these absolute thresholds. The authors should support the threshold transfer with system-specific evidence (e.g., natural occupation numbers, configuration weights, or comparison with a known dissociation curve) or explicitly soften the classification language to avoid over-interpreting small differences in entropy magnitudes.","section":"Sec. II, Eq. (1)-(3); Sec. III, Tables II-III"},{"comment":"The conclusion that 3d-3d' interactions in CrH are not nondynamic is based on a DMRG-FCI calculation with a smaller basis, whose reduced mutual information is then used to interpret the larger-basis [7,12] active-space results. Because the FCI calculation uses a different one-electron basis, its entropy values do not automatically transfer to the larger-basis calculation. The authors should either perform the FCI check in the same basis or demonstrate explicitly that the basis change does not affect the reported entropy comparison.","section":"Sec. III, CrH discussion and SI Sec. S4"}],"minor_comments":[{"comment":"The Methods section states that the large active space for the oxides contains 14 orbitals, but Fig. 3(d) and Table I use [14,15] for PdO. Please clarify which active space was used and whether the extra virtual orbital affects the comparability of the PdO results with the other oxides.","section":"Sec. II and Fig. 3(d)"},{"comment":"The manuscript repeatedly refers to Supporting Information Tables S1-S7 and Sections S4-S6, but the SI was not included in the submitted version available for review. Please include the SI so that the active-space compositions and the DMRG-FCI details can be checked.","section":"General"},{"comment":"The notation is inconsistent: the title and most of the text use 'double d-shell' or 'second d-shell', while the abstract and some phrases use 'doubled-shell'. Please unify the terminology.","section":"Title and Introduction"},{"comment":"There are several typographical errors in the Discussion, including 'demonstracted', 'transiton metal', 'correlateld electronic structure', and 'availabe' in Sec. V. These should be corrected in revision.","section":"Sec. IV"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope and the computational methodology is sound. The main technical concern is that the key mutual-information comparison is orbital-basis dependent, and the Discussion explicitly acknowledges that orbital localization affects such measures; this needs to be addressed with a common or localized basis before the overestimation claim can be considered robust. The missing SI should also be supplied. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a serious look. It applies orbital entanglement entropy, established metrics from Boguslawski, Legeza, and Reiher, to a periodic series of transition metal hydrides and oxides, comparing small valence-only active spaces with ones containing the second d-shell. That systematic comparison is genuinely new, and the calculations are solid: standard CASSCF/CASPT2 and DMRG with small active spaces that are nearly exact, with energy errors stated at 1e-6 to 1e-4 a.u. The qualitative finding, that omitting d' distorts the orbital entanglement picture, especially in late 3d oxides, comes through consistently in one-orbital entropies, Mulliken charges, and vibrational frequency shifts. Credit where it's due: the paper doesn't oversell the frequencies, notes the CrH case is sensitive to CASPT2 parameters, and acknowledges the orbital-localization issue in the discussion. For a computational chemist choosing active spaces, this is a practical and useful diagnostic.\n\nNow the soft spots. The stress-test concern about rotation invariance is real and lands directly on the paper's most quoted number: the NiO 4s-sigma* mutual information dropping from 0.25 to 0.07 when d' is added. Mutual information is computed in the converged CASSCF orbital basis, and the small and large active spaces produce different optimized orbitals. So the claim that the small active space \"overestimates\" the entanglement presupposes that particular orbital basis. The paper itself cites refs. 75, 97-101 on localization effects but never tests whether the NiO reduction survives an orbital-invariant quantity or a different localization. This doesn't sink the paper, because the qualitative trend across the series is supported by other metrics, but the quantitative framing should be softened or backed by a rotation-invariant check. Also worth noting: data is only available upon request, which is annoying for a study whose value is partly as a benchmark; vibrational frequencies have no error bars; and the CrH DMRG-FCI verification lives in the SI with assumed transferability to the larger basis. The entropy classification thresholds from prior literature are adopted without re-validation for 3d/4d systems, reasonable as a starting point but the static/dynamic attributions hang on them.\n\nNone of this is fatal. The paper is honest, clear, and the core numerical results are reproducible in principle. It deserves peer review, with the request that the authors address the basis-invariance issue and make the data available. I'd bring it to our group meeting; it's a good example of how quantum information metrics can inform practical active space selection, even if the quantitative claims need more care.","headline":"A systematic and useful entanglement-entropy study of the double d-shell effect in 3d/4d diatomics, but the headline NiO numbers are more basis-dependent than the paper lets on.","tokens_in":713,"tokens_out":1853,"would_cite":true,"duration_ms":41596,"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":"Omitting the second d-shell overestimates 4s–σ* mutual information in NiO by more than a factor of three, and entanglement entropy diagnoses the imbalance.","keywords":["double d-shell effect","orbital entanglement entropy","mutual information","transition metal hydrides","transition metal oxides","active space selection","static and dynamic correlation","DMRG wavefunctions"],"falsifier":"A near-exact FCI-quality calculation of NiO in a basis large enough to describe both the 4s and d′ radial functions would settle it: if the 4s–σ* mutual information remains near 0.25 rather than falling to about 0.07 when the d′ shell is present, the central diagnostic claim fails; a larger-basis CrH calculation would test the transferability assumption.","tokens_in":15931,"feed_emoji":"⚛️","tokens_out":10893,"duration_ms":100415,"temperature":0.7,"pith_summary":"This paper tries to characterize the double $d$-shell effect—the need to correlate a second, radially compact set of $d$ orbitals (the $d'$ orbitals) alongside the valence $d$ set—by looking at the entanglement entropy of correlated wavefunctions. It compares small valence-only active spaces with large active spaces that add the $d'$ orbitals for a periodic series of $3d$ and $4d$ transition-metal hydrides and oxides. The central finding is that omitting the $d'$ orbitals systematically distorts the entanglement picture: in late $3d$ oxides like NiO, the small active space overestimates the $4s$–$\\sigma^*$ mutual information by roughly a factor of three, mislabeling the balance of static and dynamic correlation. Including the $d'$ orbitals drops that mutual information from 0.25 to 0.07 and shifts vibrational frequencies and Mulliken charges toward experiment. The paper argues that entanglement entropy can serve as a practical diagnostic for when a correlated calculation's active space is unbalanced.","feed_headline":"Omitting the second d-shell overstates 4s–σ* entanglement in NiO","feed_subtitle":"Adding d′ orbitals drops the 4s–σ* mutual information from 0.25 to 0.07, rebalancing correlation.","key_machinery":"The load-bearing object is the orbital entanglement analysis built from one-orbital entropies $s_i$ and pairwise mutual information $I_{i,j}$ computed from the correlated wavefunction. The paper computes these quantities with DMRG for small active spaces (valence $d$, $\\sigma$, $\\sigma^*$, and $s$ orbitals only) and large active spaces that also contain the second $d$-shell orbitals $d'$, and reads the change between the two as the signature of the double $d$-shell effect. Magnitude thresholds from earlier work—one-orbital entropy above about 0.5 with large mutual information indicating non-dynamic correlation, 0.1–0.5 with moderate mutual information indicating static correlation, and small values indicating dynamic correlation—turn the raw entropies into statements about correlation type.","core_discovery":"The central claim is that the second $d$-shell effect has a characteristic information-entropy signature: a valence-only active space overestimates the quantum mutual information between the $4s$ and $\\sigma^*$ orbitals of late $3d$ oxides, and including the $d'$ orbitals rebalances it. For NiO the $4s$–$\\sigma^*$ mutual information falls from 0.25 to 0.07 with the $d'$ shell added, and similar but weaker rebalancing appears in FeO and CoO; the $4d$ analogues RuO, RhO, and PdO show much smaller changes. In early hydrides such as CrH, the $d$–$d'$ interactions are moderate and appear to mediate both static and dynamic correlation, although a smaller-basis DMRG-FCI reference calculation makes the authors cautious about classifying the $3d$–$3d'$ interaction as nondynamic. The paper concludes that simple rules for adding $d'$ orbitals will not work universally, and that orbital entanglement entropy can help decide case-by-case whether an active space is balanced.","pith_inferences":["The same entropy-based comparison could be applied to excited states or spin-crossover systems, where the d′ shell is suspected to matter but no entanglement data exists.","Automated active-space selection methods that score orbitals by entanglement would likely need to include d′ orbitals explicitly, or they may systematically under-rank them for late 3d oxides.","The NiO drop from 0.25 to 0.07 suggests a rough empirical rule of thumb: a valence-only late-3d-oxide calculation showing 4s–σ* mutual information near 0.25 should trigger suspicion that d′ is required.","If the CrH DMRG-FCI result generalizes, d–d′ interactions are not purely nondynamic, which would complicate multireference perturbation corrections that assume a single correlation regime."],"forward_implications":["Valence-only active spaces for late 3d oxides systematically overestimate 4s–σ* entanglement, so property calculations built on them carry an unbalanced static/dynamic mix.","Adding the d′ orbitals moves CASPT2 vibrational frequencies toward experimental values for the oxides, as seen for FeO, CoO, and NiO.","The d′ effect is not uniform: early hydrides like CrH show moderate, mixed static-and-dynamic d–d′ interactions rather than a simple nondynamic channel.","Entanglement entropy can serve as a practical diagnostic for active-space balance, supporting case-by-case decisions on whether the d′ shell is needed."],"supporting_citations":[{"why":"Defines the double d-shell effect for 3d metals in CASSCF treatments.","marker":"[23]"},{"why":"Identifies the short-ranged 3d basis as a source of the effect.","marker":"[24]"},{"why":"Supplies the entropy and mutual-information thresholds used to classify correlation type.","marker":"[29]"},{"why":"Proposes an alternative multi-d-occupancy definition and frames the ongoing debate on the effect's cause.","marker":"[30]"},{"why":"Establishes occupancy sensitivity of 3d orbitals as the origin of second d-shell nondynamic correlation.","marker":"[59]"},{"why":"Provides the DMRG methodology for extracting one- and two-orbital entropies from the wavefunction.","marker":"[75]"}],"fun_headline_variants":["Adding d' orbitals rebalances 4s–σ* entanglement in NiO","Omitting d' shell inflates 4s–σ* mutual information in NiO","Double d-shell effect distorts orbital entanglement in NiO","NiO's 4s–σ* entanglement drops when d' orbitals are added"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The qualitative classification of one-orbital entropy and mutual-information magnitudes into static, dynamic, and non-dynamic correlation regimes is assumed to transfer from earlier literature to these 3d/4d hydrides and oxides, and the CrH conclusion about 3d–3d' interactions relies on a smaller-basis DMRG-FCI calculation whose transferability to the larger-basis results is assumed.","fun_headline_variants_meta":{"raw":{"variants":["Adding d' orbitals rebalances 4s–σ* entanglement in NiO","Omitting d' shell inflates 4s–σ* mutual information in NiO","Double d-shell effect distorts orbital entanglement in NiO","NiO's 4s–σ* entanglement drops when d' orbitals are added"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000867,"raw_usage":{"total_tokens":3761,"prompt_tokens":952,"completion_tokens":2809,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":2724}},"tokens_in":568,"tokens_out":2809,"duration_ms":24018,"temperature":1.0,"reasoning_tokens":2724,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:38:43.383808+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A near-exact FCI-quality calculation of NiO in a basis large enough to describe both the 4s and d′ radial functions would settle it: if the 4s–σ* mutual information remains near 0.25 rather than falling to about 0.07 when the d′ shell is present, the central diagnostic claim fails; a larger-basis CrH calculation would test the transferability assumption.","supporting_citations":[{"cited_title":"Pierloot , author B","cited_arxiv_id":null,"evidence_quote":"Defines the double d-shell effect for 3d metals in CASSCF treatments."},{"cited_title":"Dunning , Thom","cited_arxiv_id":null,"evidence_quote":"Identifies the short-ranged 3d basis as a source of the effect."},{"cited_title":"Boguslawski , author P","cited_arxiv_id":null,"evidence_quote":"Supplies the entropy and mutual-information thresholds used to classify correlation type."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proposes an alternative multi-d-occupancy definition and frames the ongoing debate on the effect's cause."}],"review_version":1}