{"id":"7950a16c-4c41-4052-918d-dca5040665f3","arxiv_id":"2607.29359","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Enforcing positive semidefiniteness of the BSE-based self-energy adds particle-hole T-matrix scattering channels and brings computed HOMO energies to within ~0.1 eV of EOM-CCSD accuracy on the Marie-Loos and GW100 benchmarks.","lead":"Quantum-chemistry approximations for electron energies sometimes give impossible negative densities of states. This paper derives a positivity-enforcing ('PSD') version of a widely used BSE+GW self-energy that fixes that flaw and computes molecular ionization energies within about 0.1 eV of coupled-cluster quality at a fraction of the cost.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Near-CCSD claim is reference- and property-selective: PSD-I fails to improve over the parent BSE self-energy on plain HF and for HOMO-1 under the recommended HF+γGW protocol.","rationale":"The reader's conditional verdict already captures much of this concern, but their weakest_assumption emphasizes inheritance of BSE excitation errors and the reference-selection issue. My stress test identifies a more direct and quantitative counterexample within the paper's own figures: the recommended protocol fails to improve over the parent self-energy for HOMO-1, and the improvement over the parent is absent on plain HF for HOMO. This does not invalidate the formal PSD construction or the specific HOMO result, but it does undermine the broader wording of the abstract and discussion. A focused GW100 test with confidence intervals would settle whether the HOMO-1 failure is systematic or an outlier. Because the paper already acknowledges some reference sensitivity and the GW100 HOMO result is a genuine out-of-sample check, I do not recommend rejection; the verdict should remain conditional until the scope of the claim is tightened or the HOMO-1 behavior is explained.","tokens_in":18045,"tokens_out":18602,"duration_ms":183760,"concrete_test":"On GW100, compute Σ_PSD-I and Σ_BSE HOMO and HOMO-1 MADs for at least two reference protocols: plain HF and HF+γGW. Report paired bootstrap 95% confidence intervals for the MAD difference (Σ_PSD-I − Σ_BSE). If the HOMO-1 confidence interval under HF+γGW excludes zero in favor of Σ_PSD-I and the HOMO advantage survives under plain HF, the generalization is supported; otherwise the claims should be narrowed to 'HOMO at the HF+γGW reference.'","verdict_should_be":"UNCHANGED","load_bearing_attack":"The formal PSD construction (Eq. 11) is not the weak point; the weak point is the empirical generalization from 'HOMO at one tuned reference' to 'consistently improves quasiparticle energies' and 'valence ionization energies.' On plain HF, Fig. 6 shows Σ_PSD-I has MAD 0.307 eV, worse than the parent Σ_BSE at 0.280 eV; on RSH+γGW, plain GW (0.105 eV) beats Σ_PSD-I (0.134 eV). On the recommended HF+γGW protocol, the HOMO MAD is indeed 0.095 eV, but the same protocol gives HOMO-1 MAD 0.288 eV—worse than Σ_BSE (0.239 eV) and far from EOM-CCSD (0.164 eV) in Fig. 7. The recommended reference was selected after inspecting Marie–Loos performance; the GW100 out-of-sample test confirms HOMO but does not test HOMO-1 or alternate reference states. Thus the central claim of near-CCSD accuracy for valence ionization energies rests on a single property (HOMO) and a performance-selected reference, while the abstract and discussion state a broader conclusion.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper derives positive-semidefinite (PSD) extensions of the BSE-based self-energy of Ref. 1. Using the NEGF partitioning of Ref. 31, the exchange diagram is split into half-diagrams a and b, and the minimal complete square (a+b)(a+b) yields Eq. (7). Restricting the intermediate excitations to singlets gives PSD-I (Eq. 11), and including triplets gives PSD-II (Eq. 14). The authors implement these self-energies in MOLGW and benchmark them on the Marie–Loos set (23 molecules) and GW100 (100 molecules) for HOMO and HOMO-1 ionization energies. They report that PSD-I on an HF+γGW reference achieves HOMO MADs of 0.095 eV and 0.10 eV on the two sets, close to EOM-CCSD, while restoring spectral-function positivity. The formal derivation is internally consistent, but the empirical claims are narrower than the abstract and discussion suggest: the improvement is HOMO-specific, reference-dependent, and the HOMO-1 results under the recommended protocol are worse than the parent BSE self-energy.","tokens_in":18239,"tokens_out":5529,"duration_ms":54922,"significance":"If the claims were fully validated, this would be a valuable contribution: it provides a principled, parameter-free completion of the BSE-based self-energy that guarantees a nonnegative spectral function and adds a physical T-matrix channel, rather than a damped ad hoc correction. The formal construction (Eqs. 7, 11) is elegant and machine-checkable in the sense of following from explicit half-diagram algebra, and the GW100 out-of-sample HOMO test is a useful sanity check. However, the significance of the numerical claim is currently limited by reference selection and by the absence of evidence for deeper valence states. The paper's central conclusion—'consistently improves quasiparticle energies' and 'accuracy comparable to coupled-cluster'—is supported only for HOMO at one performance-selected reference, which weakens the claim from a general methodological statement to a benchmark-specific observation.","major_comments":[{"comment":"The Abstract and Discussion claim 'consistently improves quasiparticle energies' and 'accuracy comparable to coupled-cluster reference calculations.' But under the recommended HF+γGW protocol, Fig. 7 gives PSD-I HOMO-1 MAD = 0.288 eV, which is worse than the parent Σ_BSE (0.239 eV) and far from EOM-CCSD (0.164 eV). The Discussion sentence correctly says 'HOMO energies,' yet the broader wording in the Abstract is not supported for valence ionization energies generally. Please restrict the claims to HOMO or provide evidence that the improvement carries over to deeper valence states.","section":"Abstract; Discussion; Figs. 6–7"},{"comment":"The recommended scheme (HF+γGW with PSD-I) was selected after inspecting Marie–Loos performance across four references and several synthetic mixtures. On plain HF, PSD-I (MAD 0.307 eV) does not improve over the parent Σ_BSE (0.280 eV); on RSH+γGW, PSD-I (0.134 eV) is worse than plain GW (0.105 eV). The GW100 out-of-sample test uses only the selected protocol and only HOMO, so it does not validate the selection procedure or the general claim. Unless an a priori criterion for choosing the reference is provided, or a nested training/test split is used, the '0.1 eV accuracy' should be presented as one favorable protocol rather than a robust property of PSD-I.","section":"Numerical Results; Fig. 6; 'Synthetic approaches'"},{"comment":"The PSD construction only rearranges half-diagrams; it cannot repair systematic errors in the input singlet BSE excitation manifold. The triplet-channel failure of PSD-II is correctly attributed to underestimated triplet Ων, but no analogous validation of singlet excitation energies or amplitudes against reference is provided. Since PSD-I inherits all singlet BSE errors, the near-CCSD claim for HOMO is only as good as the static-kernel BSE singlet spectrum. Please either benchmark the singlet intermediates or explicitly state this inherited error as a limitation of the method.","section":"Eqs. (3), (11), (14); Numerical Results (PSD-II paragraph)"}],"minor_comments":[{"comment":"Captions state errors are with respect to 'a reference CCSD(T)ΔSCF evaluation,' while the text says the Marie–Loos reference is CIPSI near-FCI and the GW100 reference is ΔCCSD(T). Please align the captions with the actual reference definitions.","section":"Figs. 6–8 captions"},{"comment":"The acronym PSD is used in the Abstract without expansion; spell out 'positive semidefinite' at first use or define it there.","section":"Abstract"},{"comment":"The basis set is written 'aug-cc-pvQZ'; the standard notation is aug-cc-pVQZ. Also 'LibXClibrary' should be 'libxc.'","section":"Computational Details"},{"comment":"The factor of 3 in the triplet contribution is introduced without explanation. A brief comment that the three degenerate triplet components (Eq. 23) give this multiplicity would help the reader.","section":"Eq. (14) and text after Eq. (11)"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper supplies a genuinely clean fix for a real flaw in the BSE-based self-energy of Ref. 1 — the negative spectral weight — and shows that on one protocol (HF+γGW) the fixed self-energy gets HOMO MADs near 0.1 eV on both Marie-Loos and GW100, close to EOM-CCSD. The other thing to know: the abstract's \"consistently improves quasiparticle energies\" is not supported by the paper's own numbers. The improvement is real for HOMO at the recommended reference, and much weaker elsewhere.\n\nThe genuinely new content is the derivation of the PSD completions Σ_PSD-I and Σ_PSD-II, plus the working equations (13) and (15). The PSD framework is prior work (Stefanucci and Pavlyukh), but applying it to this self-energy is new, and the construction is a formal identity — a complete square, Eq. (7) — with no fitted constants. The spin analysis is rigorous: only singlets enter Σ_d and Σ_2x, so the minimal completion Σ_PSD-I is the natural choice, and the triplet channel degrades results exactly as expected given the static-kernel BSE's known triplet bias. Fig. 5 shows the negative dips in the rate function becoming small positive satellites. The GW100 test is a genuine out-of-sample check for HOMO, and the error distribution is narrow and centered at zero.\n\nThe soft spots, in order. First, property selectivity: on HOMO-1 under the recommended HF+γGW protocol, PSD-I gives 0.288 eV MAD against 0.239 for the parent Σ_BSE and 0.164 for EOM-CCSD. The text admits the HOMO-1 advantage is \"less systematic,\" but the abstract and discussion generalize anyway. The construction also inherits the input manifold's errors; the triplet failure shows how sensitive it is. Second, reference selection: HF+γGW was picked after seeing Marie-Loos performance, and GW100 only validates HOMO. Legitimate, though not fatal — the GW100 confirmation is real. Third, on plain HF, PSD-I (0.307 eV) doesn't beat Σ_BSE (0.280 eV), and on RSH+γGW, plain GW (0.105 eV) beats PSD-I (0.134 eV); the RSH comparison is muddied by per-method optimal tuning. None of this sinks the recommended protocol, but it does sink the broad framing. Fourth, no uncertainty estimates, so the sub-0.02 eV differences among PSD variants are not meaningful. Fifth, minor: the Fig. 6/7 captions call the Marie-Loos reference CCSD(T)ΔSCF when it is CIPSI, say \"23 molecules\" when two are omitted, and the code is only \"available upon request.\"\n\nWho gets value: anyone working on vertex corrections, spectral functions, or molecular GW/BSE. It deserves a serious referee — the math holds up and the HOMO result is interesting. I'd send it out, asking for a scoped abstract, confidence intervals, corrected captions, and the code.","headline":"Clean formal fix for the BSE self-energy's negative spectral weight, with real ~0.1 eV HOMO gains on a selected reference — but the 'consistently improves' framing overreaches once you look at HOMO-1 or other starting points.","tokens_in":18900,"tokens_out":10026,"would_cite":true,"duration_ms":77827,"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":"Completing the square of a diagrammatic self-energy restores spectral positivity and reaches near coupled-cluster accuracy for ionization energies.","keywords":["Bethe-Salpeter equation","self-energy","positive semidefinite spectral function","GW approximation","nonequilibrium Green's functions","quasiparticle ionization energies","coupled-cluster benchmark"],"falsifier":"Find a molecule where the static-kernel BSE singlet excitation energies carry a bias comparable to the known triplet bias (for example, a strong charge-transfer or double-excitation case). If Σ_PSD-I's HOMO MAD there degrades into the 0.2–0.5 eV range typical of Σ_PSD-II, the claimed accuracy would be shown to be inherited from the input manifold rather than produced by the PSD construction. A direct spectral-function calculation that exhibits residual negative weight on any molecule would also falsify the positivity guarantee.","tokens_in":17759,"feed_emoji":"⚛️","tokens_out":8760,"duration_ms":91928,"temperature":0.7,"pith_summary":"The paper claims that the negative spectral weight produced by a Bethe-Salpeter-equation-based self-energy can be eliminated by a positive semidefinite (PSD) completion: recombining the diagram's two half-diagrams into a complete square. The minimal completion, Σ_PSD-I, doubles the exchange term and adds a particle-hole T-matrix self-energy built from singlet BSE excitations, turning negative dips into real satellite peaks. Benchmarked on 23 small molecules and on the 100-molecule GW100 set, the corrected self-energy gives HOMO ionization energies with a mean absolute deviation of about 0.1 eV from near-full-CI references, close to the intrinsic accuracy of EOM-CCSD, at essentially the cost of the parent BSE-based scheme. The point is that a formal positivity constraint doubles as an accuracy improvement, not just a cure for an artifact.","feed_headline":"Positivity fix lands BSE self-energy at near-coupled-cluster accuracy","feed_subtitle":"Minimal PSD completion removes negative spectral weight and matches EOM-CCSD on 123 molecules.","key_machinery":"The load-bearing device is the PSD completion of half-diagrams on the Keldysh contour. Cutting the exchange diagram at the boundary between the minus and plus branches gives two half-diagrams a^ν_kσ and b^ν_kσ (Eqs. 4) expressed through BSE transition amplitudes; a self-energy built as the complete square (a+b)(a+b) = a·a + 2a·b + b·b has rates in Fermi golden rule form, hence a nonnegative spectral function. The new b·b term is a particle-hole T-matrix self-energy Σ_Tph(S) from singlet BSE excitation energies Ω_ν and amplitudes Z^ν, while the doubled a·b term restores the original exchange contribution. Because the statically screened interaction W0 is kept distinct from the bare Coulomb v,","core_discovery":"The authors derive the minimal positive semidefinite extension of the BSE-based self-energy. Using the nonequilibrium Green's function formalism, they cut the exchange-type self-energy diagram along its Keldysh-contour boundary into two half-diagrams a and b, built from Coulomb and screened-Coulomb vertices with BSE transition amplitudes, and recombine them as the complete square (a+b)(a+b). This yields Σ_PSD-I = Σ_d + 2Σ_2x + Σ_Tph(S), where the new particle-hole T-matrix term Σ_Tph(S) uses singlet BSE excited states. The completed square guarantees a nonnegative rate function and a nonnegative spectral function, converts the parent theory's small negative spectral weight into positive sate","pith_inferences":["Since the PSD completion only rearranges existing diagrams, its accuracy is bounded by the quality of the input singlet BSE excitation manifold; improving the triplet kernel (e.g., with dynamical screening) should directly unlock Σ_PSD-II's potential.","The same half-diagram square-completion logic could apply to other vertex corrections or cumulant-type satellite constructions, where positivity of the spectral function is also a requirement.","The strong reference sensitivity (on plain HF, Σ_PSD-I does not beat the parent scheme) indicates that starting-point optimization remains a controlling factor; a self-consistent evaluation could remove this dependence.","The demonstrated accuracy on light molecules invites testing on extended systems and on spectra beyond valence ionization (electron affinities, core levels), where positive-definiteness of the spectral function is equally mandatory."],"forward_implications":["The BSE-based self-energy can be made spectral-function-positive without sacrificing efficiency: Σ_PSD-I costs essentially the same as the parent BSE-based self-energy.","On the 23-molecule set, Σ_PSD-I with a density-matrix-corrected HF reference gives a HOMO MAD of 0.095 eV, versus 0.082 eV for EOM-CCSD and 0.43 eV for one-shot GW.","On the GW100 set, Σ_PSD-I yields a HOMO MAD of 0.10 eV, roughly halving the GW error and improving on the parent Σ_BSE (0.15 eV).","Admitting triplet intermediate states (Σ_PSD-II) degrades results (MADs 0.2–0.5 eV) because the static-kernel BSE underestimates triplet excitation energies; the singlet-only minimal completion is therefore the recommended scheme.","The parent theory's negative spectral weight reappears as small positive satellite peaks at nearly the same energies, enriching the satellite spectrum."],"fun_headline_variants":["PSD fix lands BSE self-energy at CC accuracy","Minimal PSD extension gives BSE near-CC precision","Positive semidefinite BSE self-energy matches coupled-cluster","BSE self-energy: positivity fix yields CC-level accuracy","From negative spectral weight to near-CC: BSE fixed by PSD"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The strongest result depends on two premises: the static-kernel BSE singlet excitation manifold is accurate enough for the recombined self-energy to inherit its quality, and the non-self-consistent HF+γGW reference protocol is a valid starting point—on plain HF, Σ_PSD-I does not improve over the parent theory.","fun_headline_variants_meta":{"raw":{"variants":["PSD fix lands BSE self-energy at CC accuracy","Minimal PSD extension gives BSE near-CC precision","Positive semidefinite BSE self-energy matches coupled-cluster","BSE self-energy: positivity fix yields CC-level accuracy","From negative spectral weight to near-CC: BSE fixed by PSD"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000265,"raw_usage":{"total_tokens":1463,"prompt_tokens":781,"completion_tokens":682,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":525,"completion_tokens_details":{"reasoning_tokens":609}},"tokens_in":525,"tokens_out":682,"duration_ms":7090,"temperature":1.0,"reasoning_tokens":609,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T08:35:54.324451+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a molecule where the static-kernel BSE singlet excitation energies carry a bias comparable to the known triplet bias (for example, a strong charge-transfer or double-excitation case). If Σ_PSD-I's HOMO MAD there degrades into the 0.2–0.5 eV range typical of Σ_PSD-II, the claimed accuracy would be shown to be inherited from the input manifold rather than produced by the PSD construction. A direct spectral-function calculation that exhibits residual negative weight on any molecule would also falsify the positivity guarantee.","supporting_citations":[],"review_version":1}