Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

A reduced cost equation of motion coupled cluster method for excited states based on state-specific natural orbitals: Theory, Implementation, Benchmark

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read By giving each excited state its own compressed virtual space and adding a low-cost correction, the paper reproduces canonical EOM-CCSD excitation energies to within about 0.02 eV for valence, Rydberg, and charge-transfer states.

desk verdict A credible reduced-cost EOM-CCSD variant whose sub-0.02 eV MAE claim holds on standard benchmarks, but the perturbative correction is empirical and worsens the worst Rydberg case; worth a serious referee. read the letter →

arxiv 2506.16894 v2 pith:YVPDEML3 submitted 2025-06-20 physics.chem-ph

Amrita Manna , Achintya Kumar Dutta This is my paper · ORCID
classification physics.chem-ph
keywords equation-of-motioncoupledclusterEOM-CCSDstate-specificfrozennaturalorbitalsauxiliaryfunctionsperturbativecorrectionADC(2)Rydbergstatescharge-transfer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Equation-of-motion coupled cluster with singles and doubles (EOM-CCSD) is among the most reliable ways to compute molecular excitation energies, but its steep cost limits it to small molecules. This paper presents a way to run it in a compressed orbital space: for each excited state, a separate set of frozen natural orbitals is built from the low-cost ADC(2) or CIS(D) one-particle density, and the auxiliary basis is compressed with natural auxiliary functions. A perturbative correction estimated with the same ADC(2) method compensates for most of the truncation error. Tested on valence, Rydberg, and charge-transfer benchmarks, the method reproduces full EOM-CCSD excitation energies with mean absolute deviations typically below 0.02 eV when ADC(2) natural orbitals and the correction are used. With default thresholds, roughly half the virtual and auxiliary functions are retained, and the overhead of building the state-specific orbitals is small.

What carries the argument

The machinery has three pieces. First, state-specific frozen natural orbitals: for the ground state, MP2 natural orbitals; for each excited state, the virtual-virtual block of the state's correlated one-particle density matrix is diagonalized, and orbitals with occupation below a threshold are discarded. Second, natural auxiliary functions: a singular value decomposition of the three-center two-electron integral matrix compresses the density-fitting auxiliary basis, dropping functions with small singular values. Third, a perturbative correction: the difference between canonical and truncated ADC(2) excitation energies is added to the truncated EOM-CCSD value, using the cheap method's truncation error as a surrogate for the expensive method's. The two thresholds are the only user-set parameters, making the approach effectively black-box.

What would settle it

Take a set of valence, Rydberg, and charge-transfer states, compute the truncated EOM-CCSD excitation energy both with and without the ADC(2) correction at several thresholds (for example $10^{-3}$, $10^{-4}$, and $10^{-5}$), and test whether the corrected value sits closer to the canonical EOM-CCSD value at every threshold; if for some state the correction moves the result in the wrong direction, or the gap between canonical and truncated ADC(2) does not track the gap for EOM-CCSD, the key assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that excitation energies from EE-EOM-CCSD can be computed in a drastically smaller orbital basis without losing the accuracy of the full method, provided the truncation is tailored to each excited state. The paper proposes truncating the virtual space separately for each target state using natural orbitals obtained from the correlated one-particle density matrix of that state at the ADC(2) or CIS(D) level, and truncating the density-fitting auxiliary space using natural auxiliary functions. A perturbative correction, taken as the difference in excitation energy between canonical and truncated ADC(2), is added to the truncated EOM-CCSD value. With a $10^{-4}$ FNO threshold and a $10^{-2}$ NAF threshold, the method keeps roughly half the virtual and auxiliary functions and yet achieves mean absolute deviations below 0.02 eV relative to canonical EE-EOM-CCSD across valence, Rydberg, and charge-transfer states; the worst-case deviations in the presented statistics are around 0.02 to 0.08 eV depending on the test set.

Load-bearing premise

The correction assumes that truncating the virtual and auxiliary spaces changes the cheap ADC(2) excitation energy by about the same amount as it changes the full EOM-CCSD excitation energy; if the two methods respond differently to the same truncation, especially for diffuse Rydberg states, the correction could shift the answer away from the canonical value.

Editorial extensions

If this is right

  • With default thresholds, the method retains roughly half the virtual and auxiliary functions, so EOM-CCSD-level excitation energies become practical for molecules beyond the usual ten-heavy-atom limit.
  • The two-threshold protocol behaves consistently across valence, Rydberg, and charge-transfer states, so users can trust the same defaults without state-type-specific tuning.
  • The perturbative correction recovers accuracy at looser thresholds, giving a systematic accuracy-versus-cost dial rather than a single fixed approximation.
  • Building the state-specific natural orbitals adds negligible time compared with the CCSD and EOM-CCSD steps.
  • The SS-FNO framework is general and can be attached to other excited-state methods such as ADC(3) or CC3, as the authors state in the conclusions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test core-excited or double-excitation states, so whether the same thresholds and correction transfer to those regimes remains open.
  • Because the perturbative correction relies on an ADC(2) calculation in the full canonical space, that calculation is a hidden cost floor; an on-the-fly or itself-truncated ADC(2) would be a natural extension to push the method to larger systems.
  • The state-specific loop means cost grows with the number of requested roots; for dense spectra, a state-averaged FNO space shared across roots could save time at the price of some per-state optimality, a trade-off the paper does not quantify.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper presents a reduced-cost equation-of-motion coupled-cluster singles-and-doubles (EE-EOM-CCSD) method based on state-specific frozen natural orbitals (SS-FNOs). Two sets of natural orbitals are generated from CIS(D) or ADC(2) densities, one for the ground state and one for each target excited state, and the auxiliary space is also truncated via natural auxiliary functions (NAFs). A perturbative correction is defined in Eq. (31) as the difference between canonical and truncated ADC(2) excitation energies added to the truncated EOM-CCSD result. The method is benchmarked against canonical EE-EOM-CCSD on valence (Thiel), charge-transfer (Kozma), and Rydberg (Waterloo) test sets, with statistical error tables and timing comparisons on water clusters and a 31-atom molecule.

Significance. If the reported accuracy is robust, the SS-FNO approach is a practically useful way to extend EOM-CCSD-level excited-state calculations to larger molecules, and the state-specific treatment of the virtual space is a genuine improvement over ground-state-only FNO schemes. The benchmark protocol is well designed: three standard test sets, canonical EOM-CCSD references, and clear error statistics. The timing analysis on water clusters and N-methyl-2,3-benzcarbazole supports the cost-reduction claim. The paper also ships numerical data in the Supporting Information for all test sets, which is a strength for reproducibility.

major comments (2)
  1. [§2.E, Eq. (31)] The perturbative correction is introduced as an empirical additivity assumption without a derivation or a perturbative analysis. The shift is the difference between canonical and truncated ADC(2) excitation energies, which is then added to the truncated EOM-CCSD value. This implicitly assumes that the virtual-space truncation error of ADC(2) is a good proxy for that of EOM-CCSD in the same SS-FNO space. That assumption is load-bearing for the central accuracy claim, but Table 3 shows that for the Rydberg set the corrected ADC(2) MAD is 0.077 eV, larger than the uncorrected value of 0.069 eV; the worst-case Rydberg state is moved away from the canonical value. A formal justification (e.g., a second-order analysis of the truncation error in both methods) or a separate validation study focused on diffuse states is needed before the correction can be claimed as generally beneficial.
  2. [§3.1 (threshold selection)] The default thresholds (excited-state FNO 10^-4, ground-state FNO 10^-4, NAF 10^-2) are selected from convergence curves for a single valence state of pyrrole (Figures 2 and S1) and then applied to all charge-transfer and Rydberg states without further convergence checks in those classes. The behavior of the perturbative correction differs qualitatively by state type (compare Tables 2 and 3), so a single calibration point does not establish that the thresholds are transferable. The 'nearly black-box' claim would be considerably strengthened by showing at least one convergence curve with respect to the FNO and NAF thresholds for a representative Rydberg state and a representative charge-transfer state.
minor comments (5)
  1. [Abstract and §2.D] The abstract and Section 2 state that the method requires 'two adjustable thresholds,' but the actual method uses three: a ground-state FNO threshold, an excited-state FNO threshold, and a NAF threshold. Section 3.1 sets the FNO thresholds to 10^-4 and the NAF threshold to 10^-2. Please correct the count.
  2. [Table 3 and §3.2.3] The text says 'The use of perturbative corrections leads to a slight improvement in the magnitude and spread of the error' for the Rydberg set, but the maximum absolute deviation degrades from 0.069 eV (uncorrected ADC(2)) to 0.077 eV (corrected). This should be acknowledged explicitly, as it is directly relevant to the reliability of the correction for diffuse states.
  3. [Figure 3 caption] The caption reads 'Theil test set' but the correct name is the 'Thiel test set' (as used in the text). Please correct the typo.
  4. [§3.3] The molecule name appears as 'N-methyl-2,3-benzcarbaxole' in the text and Figure 7; the correct chemical name is likely N-methyl-2,3-benzcarbazole. Also, the timing details (2 days 16 hours versus 1 day 13 hours with THC) could be presented more clearly in a single table.
  5. [§2.B, Eq. (10)] The notation in Eq. (10) is not fully defined: the indices c in the sum and the relationship of D_ab to the MP2 amplitudes should be stated explicitly, and the equation as printed appears to have an ambiguous sign pattern. Clarifying this will help readers reproduce the density matrix construction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the SS-FNO-EE-EOM-CCSD accuracy claim is benchmarked against held-out canonical EOM-CCSD references, and the ADC(2)-based perturbative shift is an independent lower-level correction, not a refit of the target quantity.

full rationale

The paper's central claim is that truncated SS-FNO-EE-EOM-CCSD reproduces canonical EE-EOM-CCSD excitation energies. This is established by comparison with canonical EOM-CCSD values on three external benchmark sets (Thiel valence, Kozma charge-transfer, Waterloo Rydberg); these references are not used as fitting inputs. The FNO and NAF thresholds are selected on one pyrrole state and then held fixed, and the reported statistics cover many distinct states, so the benchmarks are not forced by construction. Equation (31) applies a shift computed from the ADC(2) method used to build the natural orbitals: omega_corrected = omega_uncorrected + (omega_canonical_ADC2 - omega_truncated_ADC2). Although this additivity assumption is heuristic and not derived in the paper, it is not circular: the canonical ADC(2) energy is an independent approximate excited-state energy, not the EOM-CCSD target, and the shift is not fitted to EOM-CCSD data. The paper's own Rydberg table even shows the correction increases the ADC(2)-density MAD from 0.069 eV to 0.077 eV, which is a validity concern for the transferability of the assumption, but it does not indicate that the results reduce to their inputs. Self-citations (refs. 23, 26, 33, 40, 54) describe prior implementations, related methods, or the BAGH code; none is invoked as an unverified uniqueness theorem or as the sole justification for the central accuracy claim. No step in the derivation chain is self-definitional, no fitted parameter is renamed as a prediction, and the central result is not equivalent to its inputs by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central method leans on standard coupled-cluster background, on the empirical quality of MP2/ADC(2)-based natural orbitals, on the transferability of thresholds chosen from one molecule, and on an additivity assumption for the perturbative correction. No genuinely new physical entities are introduced.

free parameters (3)
  • Ground-state FNO truncation threshold = 1e-4
    Chosen from convergence of the pyrrole 1^1B2 excitation energy against canonical EOM-CCSD in Section 3.1, then applied to all benchmark molecules.
  • Excited-state FNO truncation threshold = 1e-4
    Chosen from the same pyrrole convergence study, then fixed for all subsequent calculations.
  • NAF truncation threshold = 1e-2
    Chosen from pyrrole NAF convergence (Figure S1), then used as the default for ground and excited states.
assumptions (5)
  • standard math EOM-CCSD and coupled-cluster machinery are correct and standard.
    Used throughout Section 2.A without proof; this is accepted background in quantum chemistry.
  • domain assumption Natural orbitals from MP2 (ground state) and CIS(D)/ADC(2) (excited state) densities provide a systematically truncatable virtual space.
    Sections 2.B and 2.C assume these densities rank orbital importance for correlation and excitation energy; demonstrated only empirically via benchmarks.
  • ad hoc to paper ADC(2) truncation error approximates EOM-CCSD truncation error in Equation (31).
    The perturbative correction assumes transferability of truncation error between two different levels of theory; stated in Section 2.E without derivation.
  • ad hoc to paper Thresholds selected on pyrrole transfer to all other molecules and state types.
    Section 3.1 fixes 1e-4, 1e-4, and 1e-2 after a single calibration molecule; no per-molecule adaptation is performed.
  • domain assumption Density fitting and natural auxiliary function truncation introduce negligible errors.
    The implementation relies on DF and NAF approximations throughout Section 2.D, assuming the resulting integral errors are small for the reported quantities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A reduced cost equation of motion coupled cluster method for excited states based on state-specific natural orbitals: Theory, Implementation, Benchmark." pith.science (2026). https://pith.science/paper/YVPDEML3

@misc{pith2026250616894,
  author       = {Pith},
  title        = {Pith review of: A reduced cost equation of motion coupled cluster method for excited states based on state-specific natural orbitals: Theory, Implementation, Benchmark},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YVPDEML3}},
  note         = {Machine review of arXiv:2506.16894}
}
read the original abstract

We present a reduced-cost equation-of-motion coupled-cluster method for excited states, built on a new state-specific frozen natural orbital (SS-FNO) framework. This approach enables systematic and controllable truncation of the virtual spaces, significantly reducing computational demands while maintaining high accuracy. The method allows black-box application via two adjustable thresholds and includes a perturbative correction that compensates for truncation errors. We have tested the performance of both CIS(D) and ADC(2) methods in generating appropriate natural orbitals for excited states. Benchmarking on valence, Rydberg, and charge-transfer excited states demonstrates excellent agreement with canonical EE-EOM-CCSD results, with mean absolute deviations typically below 0.02 eV when ADC(2) natural orbitals with perturbative corrections are applied.

Figures

Figures reproduced from arXiv: 2506.16894 by the authors.

Figure 1
Figure 1. The convergence error in SS-FNO-EE-EOM-CCSD and canonical EE-EOM￾CCSD method with respect to percentage of active virtual 11B2 state of pyrrole molecule in aug-cc-pVTZ basis set. The canonical EE-EOM-CCSD result with 100% active virtual has been used as the reference [PITH_FULL_IMAGE:figures/full_fig_p024_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Fast and Accurate Charge Transfer Excitations via Nested Aufbau Suppressed Coupled Cluster

    physics.chem-ph 2025-05 conditional novelty 7.0 of 10

    Nesting a small coupled cluster treatment inside a new Aufbau suppressed second-order perturbation theory reproduces about 0.1 eV charge transfer excitation accuracy at non-iterative N^5 plus iterative N^3 cost.

Reference graph

Works this paper leans on

12 extracted references · 10 canonical work pages · cited by 1 Pith paper

  1. [1]

    Introduction: Simulation of excited states is one of the most active branches of theoretical chemistry, which has implications in spectroscopy and photochemistry, as well as in emerging fields like energy conversion and storage. A variety of theoretical methods for the quantum chemical calculation of molecular excited states, starting from semi -empirical...

    work page

  2. [2]

    Equation of motion coupled cluster (EOM-CC) theory: We will represent the following convention for classifying different types of indices

    Theory: 2.A. Equation of motion coupled cluster (EOM-CC) theory: We will represent the following convention for classifying different types of indices. ● i, j, k, l: Indexes of occupied molecular orbitals in Hartree-Fock determinant. ● a, b, c, d: Indexes of unoccupied (virtual) molecular orbitals. ● p, q, r, s: General indexes of any molecular orbital. ●...

    work page

  3. [3]

    Solve the HF equations

    work page

  4. [4]

    Generate the three-centre two-electron integrals using DF approximations

    work page

  5. [5]

    Save the amplitudes to disk

    Solve the MP2, CIS and EE-ADC(2)/CIS(D) equations for all the excited state in canonical basis. Save the amplitudes to disk

    work page

  6. [6]

    Generate the ground state frozen natural orbitals by the MP2 density matrix and transform the molecular orbital space to frozen natural orbital basis by using the transformation matrix GRU . 4a. Transform the virtual MO indices J to the ground state FNO basis ()J . 4b. Generate the W from J , and transform the auxiliary index of J to NAF basis. 4c. Solve ...

    work page

  7. [7]

    Form the excited state density matrix from equation (20)

    Loop over excited states obtained from EE-ADC(2)/CIS(D) method 5a. Form the excited state density matrix from equation (20). 5b. Generate the excited state FNO and molecular orbital to the excited state frozen natural orbital transformation matrix () EEU . 5c. Transform the virtual MO indices of J to the excited state FNO basis ()J . 5d. Generate the W fr...

    work page

  8. [8]

    Retrieve the canonical MP2 amplitudes from the disk, transform them to the excited state FNO basis, solve EE -ADC(2)/CIS(D) equations, and calculate the perturbative correction to SS-FNO-EE-EOM-CCSD values. In the EE-EOM-CCSD method, the total computational cost is dominated by both ground state CCSD and excited state EOM calculation , in which the most t...

    work page

Show all 12 references
  1. [9]

    Convergence with respect to the threshold: The excited states of atoms and molecules can be broadly classified into three cat egories: valence, Rydberg, and charge transfer

    Result & Discussion: 3.1. Convergence with respect to the threshold: The excited states of atoms and molecules can be broadly classified into three cat egories: valence, Rydberg, and charge transfer. In the ideal case, any approximations to the EE-EOM- CCSD method should give ...

    2018 arXiv

  2. [363]

    (18) Neese, F

    https://doi.org/10.1016/0009-2614(93)87156-W. (18) Neese, F. An Improvement of the Resolution of the Identity. J. Comput. Chem. 2003, 24 (14), 1740–1747. (19) Jung, Y.; Sodt, A.; Gill, P. M. W.; Head-Gordon, M. Auxiliary Basis Expansions for Large- Scale Electronic Structure C...

    2003 doi

  3. [622]

    (30) Helmich, B.; Hättig, C

    https://doi.org/10.1016/S0009-2614(02)01639-1. (30) Helmich, B.; Hättig, C. A Pair Natural Orbital Implementation of the Coupled Cluster Model CC2 for Excitation Energies. J. Chem. Phys. 2013, 139 (8), 084114. https://doi.org/10.1063/1.4819071. (31) Frank, M. S.; Hättig, C. A ...

    2013 doi

  4. [6697]

    (20) Beebe, N

    https://doi.org/10.1073/pnas.0408475102. (20) Beebe, N. H. F.; Linderberg, J. Simplifications in the Generation and Transformation of Two‐ electron Integrals in Molecular Calculations. Int. J. Quantum Chem. 1977, 12 (4), 683–705. https://doi.org/10.1002/qua.560120408. (21) Aqu...

    1977 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.