Pith. sign in

REVIEW 2 major objections 5 minor 35 references

A single contraction graph turns EOM-CC relaxation into reverse-mode differentiation, so excited-state forces and nonadiabatic couplings fit and run on an 8 GB consumer GPU.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 04:46 UTC pith:DOMKCUV5

load-bearing objection Solid systems paper: real consumer-GPU EOM-CCSD gradients/NACMEs with strong small-scale validation and honest framing of the chromophore demo. the 2 major comments →

arxiv 2607.05622 v1 pith:DOMKCUV5 submitted 2026-07-06 physics.chem-ph cs.DCphysics.comp-ph

EOM-CC Excited-State Gradients and Nonadiabatic Couplings on a Consumer GPU from a Contraction-DAG with Laplace-Transform J/K Kernels

Rub\'en Dar\'io Guerrero This is my paper
classification physics.chem-ph cs.DCphysics.comp-ph
keywords EOM-CCexcited-state gradientsnonadiabatic couplingscontraction DAGLaplace-transform J/Kconsumer GPUAO-directreverse-mode differentiation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Photochemistry at conical intersections needs both excited-state forces and the nonadiabatic couplings that let neighbouring electronic states exchange population. With equation-of-motion coupled cluster those quantities have traditionally required hand-derived response equations per state and hardware that is out of reach for most groups. This paper shows that both quantities fall out of one contraction directed acyclic graph: its reverse-mode transpose is the EOM-CC relaxation, and an atomic-orbital-direct Laplace-transform Coulomb/exchange kernel, made non-symmetric by the transition densities, resolves every energy denominator without ever forming a four-index molecular-orbital tensor. The resulting pipeline is validated end-to-end at small scale against finite differences, full configuration interaction, and an independent code, and it executes a complete excited-state gradient and Q–B coupling of the chlorophyll-core chromophore Mg-porphine inside 8 GB. The claim is that the combination of reverse-mode graph transpose and memory-bounded non-symmetric kernels puts correlated excited-state forces and couplings onto hardware a student already owns.

Core claim

The complete non-Hermitian EOM-CCSD excited-state gradient and interstate nonadiabatic coupling can be obtained from a single contraction DAG whose reverse-mode transpose is the relaxation, fed through an AO-direct Laplace J/K kernel that never materializes a four-index MO tensor, and that entire pipeline validates end-to-end at small scale and executes a chromophore-scale calculation of Mg-porphine within the 8 GB of a consumer GPU.

What carries the argument

The contraction DAG whose reverse-mode transpose is the EOM-CC relaxation: every multilinear density contraction has a single adjoint rule, so one graph emits the forward densities, the amplitude-response (zeta) equations, the interstate Z-vector, and the chunked non-symmetric J/K schedule that keeps the build inside 8 GB.

Load-bearing premise

That small-scale end-to-end correctness of the same kernels, plus piece-wise checks and machine-zero translational invariance, is enough to treat the chromophore-scale Mg-porphine result as a trustworthy capability demonstration even though frozen-natural-virtual truncation leaves an uncertainty of roughly 0.01 hartree per bohr and no external large-scale oracle exists.

What would settle it

Recompute the Mg-porphine bright-Q gradient and Q–B NACME at substantially higher virtual retention (or without FNO compression on larger memory) and check whether the per-component values move by more than the stated ~0.01 Eh/a0 bound; an independent CFOUR-class cross-code comparison on a mid-sized molecule would also settle the transport of small-scale correctness.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The manuscript presents a memory-bounded GPU implementation of EOM-CCSD excited-state analytic gradients and interstate NACMEs that fits a single 8 GB consumer GPU. The central construction is a single contraction DAG whose reverse-mode transpose supplies the non-Hermitian amplitude response (ζ / interstate Z-vector), combined with an AO-direct Laplace-transform J/K kernel made ordered and non-symmetric by transition densities (J^x(A,B)≠J^x(B,A)), so that no four-index MO tensor is formed. Small-scale end-to-end validation is extensive: finite differences across four spin multiplicities, an exact FCI identity for two electrons to <10^{-12}, spin-adapted open-shell CSF gates, a ζ=0 control, and independent Psi4 agreement on excitation energies and excited-state gradients from H2O through benzene. Kernels and the ground-state CD-RCCSD solve reach ~730 AO; with FNO virtual compression the authors execute a complete Q-state gradient and Q–B NACME of Mg-porphine (def2-SVP, 439 AO) on the card, framed explicitly as a capability demonstration bounded by ~10^{-2} Eh/a0 FNO truncation uncertainty rather than a spectroscopic result.

Significance. If the claims hold, this is a genuine systems contribution to correlated nonadiabatic dynamics: it unifies reverse-mode response for non-Hermitian EOM-CC with an AO-direct non-symmetric J/K path under a consumer-GPU memory envelope, a regime that currently requires cluster allocations. Strengths that should be credited include the multi-oracle validation stack (FD, determinant/FCI oracle, reverse-mode AD check of ξ, independent Psi4 cross-code through aromatic benzene), the explicit validated-versus-demonstrated framing, the ζ=0 load-bearing control, spin-adapted open-shell gates, and the detailed device algorithm (DAG chunking, fused ladder, arena allocator, screening-collapse fix). The combination of reverse-mode DAG transpose specialized to non-Hermitian EOM with ordered transition-density J/K on commodity hardware is a useful specialization even though the individual ingredients (AD response, Laplace AO correlation, EOM gradients/NACMEs) have mature literatures.

major comments (2)
  1. [Sect. 4.3, Table 6; SI S4] Sect. 4.3 and Table 6: the integrated device EOM-CCSD gradient/NACME validation uses the CD non-symmetric J/K path with exact orbital-energy denominators; the Laplace-grid resolution of Sect. 2.2 is explicitly not exercised and is only checked as an LT-MP2 energy proxy (SI S4). The title, abstract, and contribution (1) present the Laplace-transform J/K kernel as co-equal with the DAG for the memory-bounded claim. Please either (i) add a small-scale end-to-end FD (or dense-reference) gate of the excited-state gradient/NACME with the LT-AO path active, or (ii) re-scope the abstract/title/contribution language so that Laplace is clearly the memory enabler validated by proxy, not part of the end-to-end EOM gradient gate. As written, the strongest validated claim is the non-symmetric CD device path plus DAG transpose, not the full LT-AO EOM pipeline.
  2. [Abstract; Sect. 4.3; SI S10] Sect. 4.3 / SI S10: the Mg-porphine gradient and Q–B NACME are correctly labeled a capability demonstration with ~10^{-2} Eh/a0 FNO uncertainty, but the Abstract and Conclusions still lead with that chromophore run as the systems headline. Given that the FNO proxy study finds non-smooth gradient convergence and that NACME truncation is not separately convergence-tested, please keep the Abstract/Conclusions emphasis aligned with the paper’s own distinction: validated small-scale pipeline and memory-bounded kernel architecture first; chromophore execution second, with the FNO bound stated in the same sentence as the claim. This is a framing fix, not a request to drop the demo.
minor comments (5)
  1. [Sect. 2.1–2.4] Notation: A is used both for the EOM Jacobian (Eq. 5) and as a density argument in J^x(A,B) (Eq. 18). The text flags this once; a consistent typographic distinction (e.g. script A for the Jacobian) would reduce reader load in Sects. 2.2–2.4.
  2. [Table 1; Sect. 4.2] Table 1 quartet NACME rows are best-effort and not gated; the caption is clear, but a one-line pointer in the main text to the spin-adapted Table 3 resolution would help readers who stop at Table 1.
  3. [Sect. 4.4, Fig. 14] Fig. 14 is a surrogate-density assembly timing, not an end-to-end EOM gradient; the caption says so, but the main-text sentence that introduces it should restate that limit in one clause to avoid mis-citation.
  4. [Conclusions] The claim that the DAG template extends to CASPT2/DMRG etc. is carefully caveated as structural only (Conclusions). Consider moving that paragraph fully to SI or shortening it further so the validated EOM-CC scope remains the focus.
  5. [Data and code availability] Data availability: depositing the validation harnesses and geometries is excellent. Please state explicitly which tables/figures can be regenerated from the public stack alone versus which require the proprietary CUDA kernels (roofline, chromophore wall times).

Circularity Check

0 steps flagged

No significant circularity: reverse-mode transpose is the method, not a forced prediction; numerical claims are gated against FD, FCI, and independent Psi4.

full rationale

The load-bearing derivation is that EOM-CC amplitude response (and the interstate Z-vector) is obtained as the reverse-mode transpose of one forward density-contraction DAG, with non-symmetric AO-direct Laplace J/K builds and a two-sided Davidson. That identity is the standard reverse-mode AD Jacobian-vector product specialized to the non-Hermitian density functional F_AB; it is the implementation mechanism, not a numerical prediction that reduces to fitted inputs. Gradients and NACMEs are validated against central finite differences of the same energy, an exact full-CI determinant oracle (including bit-exact H2), a spin-adapted open-shell CSF path, and an independent Psi4 cross-code check of the excited-state gradient from H2O to benzene. The Mg-porphine chromophore run is explicitly framed as a capability demonstration with a stated FNO-truncation bound (~10^{-2} E_h/a0), piece-wise anchors, and machine-zero translational invariance—not as a spectroscopically forced result. Self-citation of the author's RECURSUM recurrence codegen is peripheral to these gates. No step reduces a claimed prediction to a fit or to an unverified self-citation chain.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 2 invented entities

The central claim rests on standard EOM-CC and Laplace-transform identities plus engineering assumptions about memory layout and FNO truncation, not on fitted physical constants. Free parameters are numerical controls (Laplace grid size, FNO retention, screening thresholds, FP32/FP64 boundary). Invented entities are software abstractions (contraction DAG as relaxation, ordered non-symmetric kernel), not new physical objects.

free parameters (4)
  • Laplace quadrature size n_τ and minimax grid
    Chosen for target accuracy over [Dmin,Dmax]; accuracy depends on n_τ and Dmax/Dmin. Not fitted to the gradient claim but controls denominator error.
  • FNO virtual retention n'_v (e.g. 90 of 353 for Mg-porphine)
    Hand-chosen compression that enables the chromophore eigensolve; direct proxy study shows ~10^{-2} Eh/a0 gradient error at ~25% retention—load-bearing for the capability demo.
  • Cholesky/screen thresholds (δ, screen_tol=0.01δ)
    Engineering knobs to keep PSD and fit 8 GB; validated against direct ERIs but chosen, not derived.
  • FP32/FP64 mixed-precision boundary
    Which kernels run FP32 vs FP64 is a design choice validated by machine-ε and ≤10^{-7} gates, not a free physical parameter, but it affects numerical floors.
axioms (6)
  • domain assumption EOM-CC similarity-transformed non-Hermitian eigenproblem with biorthonormal left/right eigenvectors yields correct excitation energies and transition densities.
    Standard EOM-CC theory (Stanton/Bartlett et al.); invoked throughout Sect. 2.1.
  • standard math Laplace identity 1/D ≈ Σ w_α exp(-D τ_α) with factorization into one-index orbital factors allows AO-direct builds without four-index MO tensors.
    Almlöf/Häser Laplace transform; Sect. 2.2 Eqs. (11)–(12).
  • standard math Reverse-mode differentiation of the multilinear density-contraction DAG yields the amplitude-response equation A^T ζ = -ξ (relaxation = transpose).
    Standard reverse-mode AD / backpropagation identity specialized to EOM; Sect. 2.3 Eq. (22).
  • domain assumption Shifted Laplace reweighting 1/(D_k - ω) is valid for ω < min D_k (HOMO–LUMO-gap regime).
    Stated limitation in Sect. 2.2 Eq. (21) and Conclusions; higher roots need contour shift or IP-EOM.
  • domain assumption Cumulant decomposition of the two-particle transition density reduces the separable part to ordered one-particle J^x/K^x builds.
    Sect. 2.4 Eqs. (23)–(24); active-only cumulant remainder handled separately.
  • ad hoc to paper Kernel identity: chromophore-scale kernels are the same objects validated end-to-end at small scale, so small-scale FD/Psi4 agreement transports.
    Explicit validation strategy at scale in Sect. 4.3; necessary because no large-scale external oracle exists.
invented entities (2)
  • Single contraction DAG as the EOM-CC relaxation (transpose = ζ / interstate Z-vector) independent evidence
    purpose: Unify gradient and NACME response across EE/IP/EA without per-state hand derivation
    Software/method abstraction specializing known AD and TCE ideas; not a new physical entity. Independent evidence is the small-scale FD/AD agreement on ξ and gradients.
  • Ordered non-symmetric AO-LT J^x(A,B)≠J^x(B,A) GPU kernel for transition densities independent evidence
    purpose: Evaluate non-Hermitian EOM densities without symmetrizing L=R
    Engineering realization of known non-Hermitian density structure; validated by FD and Psi4. Not a new particle or force.

pith-pipeline@v1.1.0-grok45 · 46693 in / 4157 out tokens · 33497 ms · 2026-07-11T04:46:52.870983+00:00 · methodology

0 comments
read the original abstract

We present a unified, memory-bounded GPU realization of equation-of-motion coupled-cluster (EOM-CC) excited-state gradients and interstate nonadiabatic couplings (NACMEs) on a single 8\,GB consumer GPU. Both are built from one contraction directed acyclic graph: the EOM-CC relaxation is the reverse-mode transpose of the forward density build rather than a per-state re-derivation, and an atomic-orbital-direct Laplace-transform $J/K$ kernel, made non-symmetric ($J^x(A,B)\neq J^x(B,A)$) by the transition densities, resolves every energy denominator with no four-index molecular-orbital tensor; a two-sided Davidson returns both eigenvectors from one device-resident, spin-pure solve. The pipeline is \emph{validated end to end at small scale}: gradients and NACMEs match finite differences across four spin multiplicities and full configuration interaction to $<\!10^{-12}$ for two electrons, and the excited-state gradient matches the independent \textsc{Psi4} code to $\le\!4.6\times10^{-7}~E_h/a_0$ from \ce{H2O} to aromatic benzene. The kernels and the ground-state solve reach chromophores ($\le\!730$ AO) in 8\,GB, and a frozen-natural-virtual compression lets the eigensolver \emph{execute} a complete excited-state gradient and $Q$--$B$ NACME of the chlorophyll-core chromophore \ce{Mg}-porphine (def2-SVP, $439$ AO) on the card. We present that run as a \emph{capability demonstration} -- executed and translationally invariant to machine zero, but anchored only piece-wise and bounded by a direct convergence study at ${\sim}10^{-2}~E_h/a_0$ -- not a converged spectroscopic result. The validated small-scale capability and the memory-bounded implementation are the contribution.

Figures

Figures reproduced from arXiv: 2607.05622 by Rub\'en Dar\'io Guerrero.

Figure 1
Figure 1. Figure 1: EOM-CCSDTQ excitation manifolds. The right (and left) eigenvectors of [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The universal gradient/NACME pipeline. Stages S1–S4 (orange) are method [PITH_FULL_IMAGE:figures/full_fig_p022_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The contraction-DAG leverage. The per-method term list compiles into one for [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The relaxation is automatic differentiation applied to a density build: reading [PITH_FULL_IMAGE:figures/full_fig_p026_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Computation graph (DAG) for the EOM-CCSDTQ excited-state gradient and [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The HGP-OS recurrence kernel: one CUDA thread evaluates one quartet’s full [PITH_FULL_IMAGE:figures/full_fig_p031_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Coulomb (J) and exchange (K) energy build as cuBLAS contraction-DAGs over the compact Cholesky factor B, with the N4 ERI tensor never formed. J is two bandwidth￾bound FP64 GEMVs over B; K streams the auxiliary index in ∼256 MB blocks as an FP32 GEMM cascade (unpack → half-transform → permute → K += B⊤ hBh). The transition￾density (EOM, L ̸= R) variant uses two distinct half-transforms BU = UB, BV = V B ⇒ K… view at source ↗
Figure 8
Figure 8. Figure 8: Gradient execution. The relaxed-density assembly is a forward ( [PITH_FULL_IMAGE:figures/full_fig_p039_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Kernel roofline on the RTX 4060 (AD107, sm 89): achieved throughput vs. arithmetic intensity for the production J/K and gradient kernels, against the FP32 (15.1 TFLOP/s) and FP64 (0.236 TFLOP/s) compute peaks and the 272 GB/s DRAM roof (FP32 ridge at AI = 56). Circles are the symmetric/shared (ground-state-density) builds; tri￾angles are the non-symmetric transition-density builds—the GPU realization of th… view at source ↗
Figure 10
Figure 10. Figure 10: Cost decomposition of one Born–Oppenheimer molecular-dynamics step for the [PITH_FULL_IMAGE:figures/full_fig_p059_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Per-SCF-iteration energy-term breakdown across the eleven benchmark systems, [PITH_FULL_IMAGE:figures/full_fig_p060_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Scaling on a single RTX 4060 over the gradient systems (31–137 atoms). (a) per [PITH_FULL_IMAGE:figures/full_fig_p061_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Speedup of the GPU backend (OSX-GPU, CD, FP32-mixed) over single-thread [PITH_FULL_IMAGE:figures/full_fig_p062_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: In-method transition-density response gradient on the RTX 4060 (the production [PITH_FULL_IMAGE:figures/full_fig_p063_14.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    and Bartlett, Rodney J

    Stanton, John F. and Bartlett, Rodney J. , title =. J. Chem. Phys. , year =

  2. [2]

    , title =

    Stanton, John F. , title =. J. Chem. Phys. , year =

  3. [3]

    Quasidiabatic states described by coupled-cluster theory , journal =

    Ichino, Takatoshi and Gauss, J. Quasidiabatic states described by coupled-cluster theory , journal =. 2009 , volume =

    2009

  4. [4]

    Laplace transform techniques in

    H. Laplace transform techniques in. J. Chem. Phys. , year =

  5. [5]

    and Webb, Simon P

    Aikens, Christine M. and Webb, Simon P. and Bell, Robert L. and Fletcher, Graham D. and Schmidt, Michael W. and Gordon, Mark S. , title =. Theor. Chem. Acc. , year =

  6. [6]

    and Bouduban, Marine E

    Hohenstein, Edward G. and Bouduban, Marine E. F. and Song, Chenchen and Luehr, Nathan and Ufimtsev, Ivan S. and Mart. Analytic first derivatives of complete active space self-consistent field energies on graphical processing units , journal =. 2015 , volume =

    2015

  7. [7]

    and Parrish, Robert M

    Hohenstein, Edward G. and Parrish, Robert M. and Mart. Tensor hypercontraction density fitting. J. Chem. Phys. , year =

  8. [8]

    Implementation of analytic gradients for

    Feng, Xintian and Epifanovsky, Evgeny and Gauss, J. Implementation of analytic gradients for. J. Chem. Phys. , year =

  9. [9]

    , title =

    Shavitt, Isaiah and Bartlett, Rodney J. , title =

  10. [10]

    and DePrince, III, A

    Rubin, Nicholas C. and DePrince, III, A. Eugene , title =. Mol. Phys. , year =

  11. [11]

    and Yuwono, Stephen H

    Liebenthal, Marcus D. and Yuwono, Stephen H. and Koulias, Lauren N. and Li, Run R. and Rubin, Nicholas C. and DePrince, III, A. Eugene , title =. J. Phys. Chem. A , year =

  12. [12]

    1979 , doi =

    Pauncz, Ruben , title =. 1979 , doi =

    1979

  13. [13]

    , title =

    Warren, Jr., Henry S. , title =

  14. [14]

    and Lam, Monica S

    Aho, Alfred V. and Lam, Monica S. and Sethi, Ravi and Ullman, Jeffrey D. , title =

  15. [15]

    Automated Code Engine for Graphical Processing Units: Application to the Effective Core Potential Integrals and Gradients , journal =

    Song, Chenchen and Wang, Lee-Ping and Mart\'. Automated Code Engine for Graphical Processing Units: Application to the Effective Core Potential Integrals and Gradients , journal =. 2016 , volume =

    2016

  16. [16]

    Elimination of energy denominators in

    Alml. Elimination of energy denominators in. Chem. Phys. Lett. , year =

  17. [17]

    Principles for a direct

    Alml. Principles for a direct. J. Comput. Chem. , year =

  18. [18]

    Takatsuka, Akio and Ten-no, Seiichiro and Hackbusch, Wolfgang , title =. J. Chem. Phys. , year =

  19. [19]

    Hackbusch, Wolfgang , title =. Comput. Vis. Sci. , year =

  20. [20]

    Grace and Fajen, O

    Wang, Yuanheng and Hait, Diptarka and Johnson, K. Grace and Fajen, O. Jonathan and Zhang, Juncheng Harry and Guerrero, Rub\'en D. and Mart\'. Extending. J. Chem. Phys. , year =

  21. [21]

    RECURSUM: Automated Code Generation for Recurrence Relations Exceeds Expert Optimization via LayeredCodegen

    Guerrero, Rub\'en D. , title =. arXiv:2604.18585 [cs.PL] , year =. 2604.18585 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  22. [22]

    and Gauss, J

    Stanton, John F. and Gauss, J. Analytic energy derivatives for the equation-of-motion coupled-cluster method: Algebraic expressions, implementation, and application to the. Theor. Chim. Acta , year =

  23. [23]

    and Wang, Tao and Krylov, Anna I

    Levchenko, Sergey V. and Wang, Tao and Krylov, Anna I. , title =. J. Chem. Phys. , year =

  24. [24]

    Analytic evaluation of the nonadiabatic coupling vector between excited states using equation-of-motion coupled-cluster theory , journal =

    Tajti, Attila and Szalay, P. Analytic evaluation of the nonadiabatic coupling vector between excited states using equation-of-motion coupled-cluster theory , journal =. 2009 , volume =

    2009

  25. [25]

    , title =

    Faraji, Shirin and Matsika, Spiridoula and Krylov, Anna I. , title =. J. Chem. Phys. , year =

  26. [26]

    Communication: Non-adiabatic derivative coupling elements for the coupled cluster singles and doubles model , journal =

    Kj. Communication: Non-adiabatic derivative coupling elements for the coupled cluster singles and doubles model , journal =. 2023 , volume =

    2023

  27. [27]

    Automatic differentiation in quantum chemistry with applications to fully variational

    Tamayo-Mendoza, Teresa and Kreisbeck, Christoph and Lindh, Roland and Aspuru-Guzik, Al. Automatic differentiation in quantum chemistry with applications to fully variational. ACS Cent. Sci. , year =

  28. [28]

    Zhang, Xing and Chan, Garnet Kin-Lic , title =. J. Chem. Phys. , year =

  29. [29]

    and Chan, Garnet Kin-Lic , title =

    Zhang, Xing and Li, Chenghan and Ye, Hong-Zhou and Berkelbach, Timothy C. and Chan, Garnet Kin-Lic , title =. J. Chem. Phys. , year =

  30. [30]

    Hirata, So , title =. J. Phys. Chem. A , year =

  31. [31]

    Schweizer, Sabine and Doser, Bernd and Ochsenfeld, Christian , title =. J. Chem. Phys. , year =

  32. [32]

    Passing the one-billion limit in full configuration-interaction (

    Olsen, Jeppe and J. Passing the one-billion limit in full configuration-interaction (. Chem. Phys. Lett. , year =

  33. [33]

    Theoretical study of the electronic spectrum of magnesium-porphyrin , journal =

    Rubio, Mercedes and Roos, Bj. Theoretical study of the electronic spectrum of magnesium-porphyrin , journal =. 1999 , doi =

    1999

  34. [34]

    Gouterman, Martin , title =. J. Mol. Spectrosc. , volume =. 1961 , doi =

    1961

  35. [35]

    Smith, Daniel G. A. and Burns, Lori A. and Simmonett, Andrew C. and Parrish, Robert M. and others , title =. J. Chem. Phys. , volume =. 2020 , doi =

    2020