{"id":"32782b6f-0bfd-4807-90ea-2e84ce591eba","arxiv_id":"2608.11529","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A new four-component multireference second-order perturbation theory with Dirac-Coulomb-Breit Hamiltonians is developed and benchmarked, finding a Breit correlation contribution of roughly 4% for xenon.","lead":"4C-MRPT2, a four-component relativistic multireference perturbation theory, is formulated for Dirac-Coulomb-Breit Hamiltonians and implemented within the small tensor product distributed active space framework. Benchmarks on noble-gas and group-13 atoms show the Breit contribution to the correlation energy grows with atomic number, reaching about 4% for xenon.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"4C-MRPT2's >1 eV overcorrelation vs MRCISD (Table 4) leaves the 2.03 eV / 4% Breit share for Xe (Table 2) without a variational benchmark; the NVPA issue is real but secondary.","rationale":"The reader identified the no-virtual-pair approximation as the weakest assumption, but the paper's own Section 3.2 already treats NVPA as a deliberate, standard approximation and explicitly attributes negative-energy dynamic correlation to QED. In contrast, Section 3.3 contains an unaddressed internal warning: 4C-MRPT2 overcorrelates relative to 4C-MRCISD by more than 1 eV in the shared correlation space, with the error increasing for heavier atoms. The central advertising result, the Breit contribution reaching about 4% of the total correlation energy in Xe, is obtained from full-space MRPT2 energies in which that overcorrelation is most likely largest. The shared-space comparison further shows that MRPT2 and MRCISD disagree on the Breit difference itself (0.15 eV versus 0.28 eV for Xe), so the issue is not merely a uniform shift in total energies. The method itself is coherent: the STP-DAS formulation is plausible, the fine-structure splittings for Ga and In agree with experiment to within a few percent once dynamic correlation is added, and the frozen-core and frozen-virtual tests are internally consistent. Those results support the methodology's qualitative value but do not validate the quantitative Breit share, since the fine-structure splittings are dominated by valence correlation and do not test the deep-core contributions that dominate the 2 eV Breit difference. The requested check, a variational extrapolation of the DCB minus DC difference over growing correlation spaces, would directly test the 4% claim without requiring a full 406-spinor MRCISD calculation. Pending that test, conditional acceptance with a required revision is the appropriate outcome, preserving the likely novelty of the implementation while withholding approval of the headline physical conclusion.","tokens_in":16572,"tokens_out":7476,"duration_ms":78869,"concrete_test":"For Xe, recompute the DCB minus DC Breit correlation-energy difference with 4C-MRCISD in a sequence of systematically enlarged correlation spaces (for example, 18 electrons in 58, 100, 150, and 200 positive-energy spinors) using the same Dyall-cv3z basis, then extrapolate to the full 406-spinor external space; compare the extrapolated variational Breit difference with the full-space 4C-MRPT2 value of about 2.03 eV. If the extrapolated MRCISD difference is substantially smaller, or if the trend with correlation-space size is non-monotonic, the 4% Breit share in Table 2 is not established.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative claim in Section 3.3 is that the Breit contribution to the all-electron correlation energy of Xe is about 2.03 eV, roughly 4% of the total, obtained as the DCB minus DC difference of full-space 4C-MRPT2 energies in Table 2. The only internal accuracy check of 4C-MRPT2 against a variational method is Table 4, where, in an identical (18e,58o) correlation space, 4C-MRPT2 overcorrelates relative to 4C-MRCISD by 1.47 eV (DC), 1.59 eV (DCG), and 1.59 eV (DCB) for Xe, and the overcorrelation grows with atomic number. The full-space calculation extends precisely into the high-energy core and virtual excitations where second-order Epstein–Nesbet perturbation theory is least reliable, so the full-space total energies, and their DCB minus DC difference, have no demonstrated accuracy. In the shared Table 4 space the MRPT2 Breit correlation difference for Xe is 0.15 eV, while the variational MRCISD value is 0.28 eV, showing that the PT2 approximation changes the Breit difference substantially even in a space only about one-seventh of the full correlation space. The bias is therefore not simply that PT2 inflates the Breit share; it is that full-space PT2 results are unbenchmarked and the known overcorrelation is Hamiltonian-dependent enough to shift the Breit difference. Without a variational or higher-order benchmark over a systematically enlarged space, the 2.03 eV and 4% numbers are unsupported. The NVPA limitation discussed in Section 3.2 is separate: it concerns negative-energy correlation, whereas this concern persists even if NVPA is accepted.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a four-component multireference second-order perturbation theory (4C-MRPT2) within the STP-DAS framework for the Dirac-Coulomb (DC), Dirac-Coulomb-Gaunt (DCG), and Dirac-Coulomb-Breit (DCB) Hamiltonians. The formalism uses an Epstein-Nesbet partitioning with a multiconfigurational reference, a positive-energy (no-virtual-pair) Hilbert space, and STP-DAS tensor structure for the external space. Benchmarks cover computational timings for HBr, positive-negative-energy orbital-rotation effects for He-like ions, all-electron noble-gas correlation energies from Ne to Xe, frozen-core and frozen-virtual approximations for Ga and In, and fine-structure splittings for Ga and In. The paper's headline physical claim is that the Breit contribution to all-electron correlation energy increases rapidly with atomic number and is about 2.03 eV (approximately 4% of the total correlation energy) for Xe; it also reports that the method recovers fine-structure splittings to within a few percent of experiment.","tokens_in":16830,"tokens_out":8288,"duration_ms":83243,"significance":"The work is significant for relativistic quantum chemistry if the claims hold. It appears to be the first four-component MRPT2 formulation that treats DCB consistently at the reference and perturbation levels, and the STP-DAS implementation enables external spaces of about 2.17×10^10 determinants. The fine-structure splittings are falsifiable predictions and agree with experiment to within 0.3-4.3%, providing a strong internal check on the method. The formal equations are standard, and no fitted parameters enter the physical claims. However, the headline Breit correlation share is not yet supported by a variational or higher-order benchmark in the full-space regime where it is claimed, and the no-virtual-pair approximation is an acknowledged but unquantified limitation for that specific number.","major_comments":[{"comment":"The central quantitative claim, that the Breit contribution to the all-electron correlation energy of Xe is about 2.03 eV (~4%), is the difference between full-space 4C-MRPT2 DCB and DC energies in Table 2. The only variational check in the paper, Table 4, uses an identical (18e,58o) space for Xe and shows that 4C-MRPT2 overcorrelates by 1.47 eV (DC) and 1.59 eV (DCB) relative to 4C-MRCISD. In that shared space the Breit correlation difference is +0.15 eV for MRPT2 versus +0.28 eV for MRCISD, whereas the full-space Table 2 value is -2.13 eV. The PT2 error is therefore not a uniform shift: it changes the Breit difference by roughly a factor of two in a space about one-seventh the size of the full space (58 of 406 correlated spinors), and the difference changes sign between the shared and full spaces within MRPT2. Because the full-space calculation extends into high-energy core and virtual excitations, where second-order Epstein-Nesbet perturbation theory is least reliable, the 2.03 eV/4% figure has no demonstrated accuracy. The authors should provide a systematic enlargement benchmark (e.g., MRCI, CC, or higher-order PT in progressively larger spaces) or explicitly reframe the number as an exploratory NVPA estimate that is not yet quantitatively benchmarked.","section":"Section 3.3 (Tables 2 and 4)"},{"comment":"The NVPA is a stated limitation: the positive-energy-only Hilbert space excludes negative-energy dynamic correlation, and the paper acknowledges that the Breit interaction directly couples positive- and negative-energy states and that negative-energy dynamic correlation can only be captured through a QED treatment. This means the reported Breit correlation contribution is, by construction, incomplete. The paper's own Section 3.2 data show that positive-negative orbital-rotation effects grow strongly with Z and are largest for Breit/Gaunt; a similar growth is plausible for the neglected negative-energy dynamic correlation. The manuscript should either provide a quantitative estimate or bound for this omitted contribution (e.g., via two-component or QED-oriented benchmark studies of heavy atoms or ions) or consistently present the 4% claim as an NVPA-limited value.","section":"Section 3.2 (No-virtual-pair approximation)"},{"comment":"The text in Section 3.3 states that the MRCI correlation energy decreases to -2.35 eV for Xe and that the DC-DCB MRPT2 difference reaches 2.03 eV for Xe. Table 2 lists -2.439 eV for the former; for the latter, the CASSCF+ MRPT2 row gives DCB -55.473 eV and DC -53.343 eV, a difference of 2.130 eV, while the CASCI MRPT2 row gives 1.992 eV. The quoted numbers do not match any row of Table 2, so the discussion should clarify which reference method (CASCI vs CASSCF+) and which row is used for the headline claim, and the numbers must be made internally consistent.","section":"Section 3.3 (Table 2 and surrounding text)"}],"minor_comments":[{"comment":"The statement that the σ-build and PT2 steps are 'independent of the underlying Hamiltonian' should be clarified as independence of computational cost; the Hamiltonian enters through the transformed two-electron integrals, which are different for DC, DCG, and DCB.","section":"Section 3.1 (Table 1)"},{"comment":"There is a typo in 'with he Dyall-cv3z basis'; this should read 'with the Dyall-cv3z basis'.","section":"Section 3.2"},{"comment":"The imaginary level-shift machinery is described but not used in the reported calculations; a sentence explaining why the chosen systems are free of intruder states would help the reader understand the practical need for Section 2.6.","section":"Section 2.6"},{"comment":"The notation in Eqs. (15)-(16), especially the barred Kronecker deltas and the index domains, is dense and would benefit from a worked example or more explicit definitions; the current presentation makes the implementation difficult for readers to verify.","section":"Section 2.4 (Eqs. 14-16)"},{"comment":"The statement that the DC result for In arises from fortuitous cancellation is helpful; it would be useful to quantify the cancellation (e.g., by separating correlation and two-electron spin-orbit contributions) so that readers can assess how robust the agreement is.","section":"Section 3.4 (Table 5)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope, and the fine-structure benchmarks and NVPA analysis are valuable contributions. The stress-test concern about the unbenchmarked full-space Breit share lands: the headline 4% number needs either stronger benchmarks or a more cautious framing. The issues are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper from the Li group is genuinely new: a four-component multireference second-order perturbation theory that includes the Dirac-Coulomb-Breit Hamiltonian at both the reference and perturbation levels, and does it inside the STP-DAS framework so the external space can be large. The fine-structure splittings for Ga and In come out within a few percent of experiment, and the frozen-core/frozen-virtual analysis gives practical guidance. The paper is also honest about its main weakness: it reports more than 1 eV of overcorrelation relative to MRCISD in a shared correlation space, and it flags the no-virtual-pair limitation in Section 3.2.\n\nThe central claim I would push back on is the 2.03 eV / roughly 4% Breit share of the total correlation energy for Xe. That number is a difference of full-space 4C-MRPT2 energies, and there is no variational or higher-order benchmark in that full space. The only head-to-head check, Table 4, shows two things: PT2 overcorrelates by about 1.5 eV, and the Breit difference in that shared space is 0.15 eV at MRPT2 versus 0.28 eV at MRCISD. So the PT2 bias is not Hamiltonian-independent; it materially changes the very quantity the paper highlights. The full-space calculation extends into high-energy core and virtual excitations, where second-order Epstein-Nesbet perturbation theory is least reliable. I do not think this makes the method wrong, but it makes the 4% number unsupported as stated. The NVPA issue is real but secondary: Section 3.2 acknowledges it, and this concern survives even if you accept the NVPA.\n\nWhat is missing is reproducibility. No code, no input files, no tabulated outputs are provided. For a method paper that is a legitimate referee request, not a fatal flaw. I would send this to peer review, with the request that the authors benchmark full-space PT2 against a systematically enlarged MRCISD or a higher-order PT treatment, and that they provide the Xe data so the Breit difference can be checked. The method itself and the fine-structure results deserve referee time; the quantitative Breit-correlation claim needs support before it becomes a citation-level fact.","headline":"Useful first implementation of four-component MRPT2 with Breit, but the headline 4% Breit share for Xe rests on full-space PT2 energies that have no variational benchmark.","tokens_in":17513,"tokens_out":2009,"would_cite":true,"duration_ms":21747,"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":"4C-MRPT2 gives a working four-component multireference second-order perturbation theory for the Dirac-Coulomb-Breit Hamiltonian, and the benchmark calculations show the Breit contribution reaches about 4% of the total correlation energy…","keywords":["four-component relativistic quantum chemistry","multireference perturbation theory","Dirac-Coulomb-Breit Hamiltonian","no-virtual-pair approximation","STP-DAS","electron correlation","fine-structure splittings","heavy-element atoms"],"falsifier":"Run the xenon $^1S_0$ 4C-MRPT2 calculation in progressively larger basis sets (for example Dyall cv3z, cv4z, and beyond) and track the DCB-minus-DC correlation gap of 2.03 eV: if the gap grows by more than a few tenths of an electronvolt as the virtual space expands, the 4% Breit share is not converged. A second check is to compare with a treatment that includes negative-energy pair contributions, since the present method restricts the correlated space to positive-energy spinors.","tokens_in":16257,"feed_emoji":"⚛️","tokens_out":9942,"duration_ms":90749,"temperature":0.7,"pith_summary":"This paper introduces a four-component multireference second-order perturbation theory, 4C-MRPT2, built on the small tensor product distributed active space (STP-DAS) framework, so that the perturbative correction can run over very large external determinant spaces without the memory bottleneck of conventional four-component configuration interaction. The working equations are formulated for the Dirac-Coulomb (DC), Dirac-Coulomb-Gaunt (DCG), and Dirac-Coulomb-Breit (DCB) Hamiltonians, which means the Breit interaction enters consistently at both the reference and the perturbation levels. Benchmarks on Ne through Xe recover all-electron dynamic correlation from the full positive-energy orbital space and show that the Breit contribution to the total correlation energy grows steeply with atomic number, reaching approximately 4% for xenon. For the gallium and indium $^2P$ ground states, adding the second-order correction moves fine-structure splittings from 12-15% error to within a few percent of experiment. A sympathetic reader would care because a practical four-component MRPT2 with explicit Breit interactions has been an open gap between two-component relativistic methods and full four-component CI.","feed_headline":"Breit correlation reaches 4% of xenon's total correlation energy","feed_subtitle":"New four-component second-order method also brings gallium and indium splittings close to experiment.","key_machinery":"The load-bearing machinery is the STP-DAS decomposition of the external configuration space, combined with the no-virtual-pair positive-energy spinor basis and the relativistic integral transformation for the DC, DCG, and DCB Hamiltonians. In STP-DAS, each external determinant is represented as a tensor product of sub-determinants from distributed active subspaces, and the Hamiltonian couplings between reference and external spaces are built as a sequence of tensor loops (the $\\sigma$-build), so the full external expansion never has to be stored globally. The perturbative step uses an Epstein-Nesbet partitioning in which the zeroth-order Hamiltonian is block-diagonal in the reference space and diagonal on external determinants, giving $E^{(2)}_m = \\sum_{K^+ \\in Q^+} |\\sigma_{m,K^+}|^2 / (E^{(0)}_m - E_{K^+K^+})$. The reference wavefunctions come from four-component CASCI or CASSCF with the positive-energy projection, and the AO-to-MO two-electron integral transformation is the main computational bottleneck once the Hamiltonian is upgraded from DC to DCB.","core_discovery":"The central claim is that a scalable four-component multireference perturbation theory at second order can be built for the Dirac-Coulomb-Breit Hamiltonian, and that including the Breit interaction changes the quantitative picture of relativistic correlation in heavy atoms. Within the no-virtual-pair approximation, the second-order correction is evaluated as a sum over all single and double excitations from the active space into the positive-energy external spinor space, with the Epstein-Nesbet denominator $\\Delta E = E^{(0)}_m - E_{K^+K^+}$. The paper reports that the DCB-minus-DC correlation energy difference grows from 0.05 eV for Ne to 2.03 eV for Xe, about 4% of the total correlation energy at this level, and that the Gaunt approximation overestimates this Breit contribution because it omits the partial cancellation from the gauge term. It further finds that the second-order correlation energy depends only weakly on whether the reference orbitals come from 4C-CASCI, 4C-CASSCF+, or 4C-CASSCF$\\pm$, and that freezing high-lying virtual spinors is a cheap approximation while freezing the entire 3d core in gallium removes about a third of the correlation.","pith_inferences":["Editorial inference: the observed weak dependence on reference orbitals suggests 4C-MRPT2 may be robust for excited states and bond-breaking regions, where state-specific orbital optimization is difficult; the paper demonstrates only ground states, so this remains to be tested.","Editorial inference: since the AO-to-MO integral transformation dominates the DCB wall time (roughly 90%), the practical next step is to accelerate that transformation, for example by Cholesky decomposition or local fitting, which would extend the method to molecules containing several heavy centers.","Editorial inference: extrapolating the reported trend, superheavy elements beyond xenon may show Breit correlation shares of several percent, large enough to affect bond lengths and spin-orbit splittings; this follows from the Z-trend but is not computed in the paper."],"forward_implications":["All-electron four-component correlation calculations for heavy elements become feasible, because the STP-DAS traversal keeps the external-space memory footprint small even when the perturbation runs over hundreds of spinors.","The Breit interaction should be retained in four-component correlation treatments of heavy atoms: its contribution to the total correlation energy grows with nuclear charge and is roughly 4% for xenon at this level.","The Gaunt-only approximation overestimates the Breit correlation because it omits the gauge term that cancels part of the Gaunt contribution, so DCB results are the safer benchmark.","Frozen-virtual and frozen-core approximations are practical: discarding the top 20% of virtual spinors changes the correlation energy by under 1% for Ga and In, although freezing the full 3d core in Ga removes about a third of the correlation.","Because the PT2 energy depends only weakly on the reference orbital choice, cheaper references such as canonical HF or 4C-CASCI can be used without losing much of the recovered dynamic correlation."],"supporting_citations":[{"why":"Provides the small tensor product distributed active space decomposition and the massively parallel determinant-addressing scheme that the MRPT2 external-space traversal inherits.","marker":"[31,32]"},{"why":"Supplies the four-component DCB multiconfigurational SCF references and documents the positive-negative-energy rotation sensitivity that the perturbation analysis builds on.","marker":"[30]"},{"why":"Gives the efficient Pauli-spinor evaluation of the Dirac-Coulomb-Gaunt and Dirac-Coulomb-Breit two-electron integrals used to build the Hamiltonian matrix elements.","marker":"[27,28]"},{"why":"The quaternion Dirac-Coulomb-Breit AO-to-MO integral transformation is the cost-determining step for the DCB-Hamiltonian calculations.","marker":"[43]"},{"why":"Establishes the four-component spinor formalism and restricted kinetic balance that define the positive-energy molecular spinor basis.","marker":"[2]"},{"why":"Justifies the no-virtual-pair approximation that restricts the correlated Hilbert space to positive-energy spinors, the boundary condition on which the Breit-correlation analysis rests.","marker":"[4]"}],"fun_headline_variants":["Breit correlation hits 4% for xenon in new 4C-MRPT2","Scalable relativistic method captures Breit correlation accurately","Gaunt approximation overestimates Breit correlation in heavy atoms","Frozen cores cut cost of 4C-MRPT2 with minimal accuracy loss","Xenon's Breit correlation energy reaches 4%, new method shows"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the no-virtual-pair approximation, namely that excluding negative-energy molecular spinors from both the reference and the perturbation leaves the reported correlation energies and the 4% Breit share essentially unchanged.","fun_headline_variants_meta":{"raw":{"variants":["Breit correlation hits 4% for xenon in new 4C-MRPT2","Scalable relativistic method captures Breit correlation accurately","Gaunt approximation overestimates Breit correlation in heavy atoms","Frozen cores cut cost of 4C-MRPT2 with minimal accuracy loss","Xenon's Breit correlation energy reaches 4%, new method shows"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000651,"raw_usage":{"total_tokens":3010,"prompt_tokens":991,"completion_tokens":2019,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":607,"completion_tokens_details":{"reasoning_tokens":1921}},"tokens_in":607,"tokens_out":2019,"duration_ms":19423,"temperature":1.0,"reasoning_tokens":1921,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:36:13.083061+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the xenon $^1S_0$ 4C-MRPT2 calculation in progressively larger basis sets (for example Dyall cv3z, cv4z, and beyond) and track the DCB-minus-DC correlation gap of 2.03 eV: if the gap grows by more than a few tenths of an electronvolt as the virtual space expands, the 4% Breit share is not converged. A second check is to compare with a treatment that includes negative-energy pair contributions, since the present method restricts the correlated space to positive-energy spinors.","supporting_citations":[],"review_version":1}