Rank-reduced equation-of-motion coupled cluster formalism with full inclusion of triple excitations
Pith reviewed 2026-05-18 21:04 UTC · model grok-4.3
The pith
A rank-reduced EOM-CCSDT method using Tucker decomposition on triple amplitudes achieves N^6 scaling while introducing errors several times smaller than the parent method's inherent error.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The rank-reduced formalism introduces an error several times smaller than the inherent error of the parent theory with the proposed default settings for a wide range of problems, while recovering the canonical method by increasing a single parameter.
What carries the argument
Tucker decomposition of the ground- and excited-state triply-excited amplitudes tensors, whose linear scaling with system size N enables the overall cost reduction.
If this is right
- Routine EOM-CCSDT-quality calculations become feasible for molecules larger than those treatable by the canonical method.
- Potential energy curves and spectroscopic parameters for excited states of mixed single, double, and higher character can be obtained at reduced cost.
- Accuracy can be increased systematically toward the full canonical result by raising one parameter without changing the code structure.
- The same decomposition strategy can be applied to ground-state CCSDT or to other properties that depend on triple amplitudes.
Where Pith is reading between the lines
- The approach may extend naturally to other tensor-network or low-rank approximations inside high-order coupled-cluster methods.
- It could enable excited-state studies of medium-sized systems where triple excitations matter but full canonical scaling remains prohibitive.
- Further validation on transition-metal compounds or larger charge-transfer complexes would test whether the error ratio holds outside the current test set.
Load-bearing premise
The Tucker decomposition retains enough information in the triple amplitudes that the added error remains smaller than the parent method's own error for the tested molecules and excitation characters.
What would settle it
A calculation on any molecular system or excited state where the rank-reduced error exceeds the parent EOM-CCSDT error by more than a small factor under default settings would challenge the central practical claim.
read the original abstract
In this work we describe the rank-reduced variant of the equation-of-motion coupled cluster theory with complete inclusion of single, double, and triple excitations. The advantage of the proposed formalism in comparison with the canonical theory stems from the application of Tucker decomposition format to the ground- and excited-states triply-excited amplitudes tensors. By exploiting the linear scaling of the dimension of the decomposed amplitudes with respect to the system size $N$, one can reduce the computational cost of the method to the level of $N^6$ and storage requirements to $N^4$. While in practice the proposed rank-reduced formalism introduces an error, we show that it is several times smaller than the inherent error of the parent theory with the proposed default settings for a wide range of problems. Higher level of accuracy can be achieved by increasing the value of a single parameter present in this formulation, recovering the canonical method in an appropriate limit. We illustrate the accuracy and performance of the proposed method by calculations for a group of molecules with excited states of different character -- from dominated by single excitations with respect to the reference determinant to states with moderate and large contribution of double and higher excitations. We report calculations of potential energy curves and related spectroscopic parameters for the first four singlet excited states of magnesium dimer, as well as the potential energy curve for the excited state of charge-transfer character in the NH$_3$-F$_2$ complex as a function of intermolecular separation.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper presents a rank-reduced formulation of equation-of-motion coupled-cluster theory with full inclusion of singles, doubles, and triples (EOM-CCSDT) in which Tucker decomposition is applied to the ground- and excited-state triply-excited amplitudes tensors. By exploiting the linear scaling of the decomposed tensor dimension with system size N, the method reduces computational cost to N^6 and storage to N^4. The authors state that, with proposed default settings for the decomposition rank, the additional approximation error remains several times smaller than the inherent error of the parent EOM-CCSDT method across tested molecules with varying excitation characters, while the canonical method is recovered by increasing a single rank parameter. Numerical illustrations include potential energy curves and spectroscopic constants for the first four singlet excited states of Mg2 and the charge-transfer excited state of the NH3-F2 complex.
Significance. If the error control and rank scaling hold beyond the small-molecule tests, the approach would make high-accuracy excited-state calculations accessible for systems where full EOM-CCSDT remains prohibitive, while retaining a tunable accuracy parameter. The explicit demonstration on both single-excitation-dominated and charge-transfer states is a positive feature.
major comments (2)
- [Numerical results and discussion of accuracy] The central claim that default Tucker ranks keep the added error several times below the parent method's inherent error while preserving N^6 scaling rests on the assumption that ranks remain bounded or grow slowly with N. The reported calculations are limited to small systems (Mg2, NH3-F2) with localized or moderate excitations; no data or analysis is provided on rank growth for larger molecules or delocalized/charge-transfer states where tensor ranks typically increase, which would simultaneously degrade both the effective scaling and the error-control guarantee.
- [Formulation of the rank-reduced EOM-CCSDT equations] The derivation of reduced scaling from the linear dimension of the decomposed amplitudes (abstract and formulation sections) is formally correct only if the chosen decomposition rank is independent of system size. The manuscript does not supply explicit bounds, scaling plots of rank versus N, or tests on systems larger than those shown, leaving the practical N^6 claim unverified for the regime where the method would be most useful.
minor comments (2)
- [Method description] Specify the exact default value of the decomposition rank parameter and its relation to the Tucker core tensor dimensions in the main text rather than only in the abstract.
- [Results for Mg2 and NH3-F2] Include a table or figure that directly compares the rank-reduced error to the parent EOM-CCSDT error (e.g., difference from a higher-level reference) for each state examined.
Simulated Author's Rebuttal
We thank the referee for the careful reading of our manuscript and the constructive comments. We appreciate the positive assessment of the method's potential significance. Below we respond to each major comment, providing clarifications on the scaling assumptions and error control. We have revised the manuscript to address the points raised where possible.
read point-by-point responses
-
Referee: The central claim that default Tucker ranks keep the added error several times below the parent method's inherent error while preserving N^6 scaling rests on the assumption that ranks remain bounded or grow slowly with N. The reported calculations are limited to small systems (Mg2, NH3-F2) with localized or moderate excitations; no data or analysis is provided on rank growth for larger molecules or delocalized/charge-transfer states where tensor ranks typically increase, which would simultaneously degrade both the effective scaling and the error-control guarantee.
Authors: We agree that the numerical tests are confined to small systems and that explicit data on rank growth versus N for larger molecules are not provided. In the revised manuscript we have added a paragraph in the discussion section analyzing the Tucker ranks required to achieve the reported accuracy thresholds for the Mg2 and NH3-F2 cases, including the charge-transfer excitation. These ranks remain modest and do not grow rapidly within the tested range. The formulation treats the rank as a single, user-adjustable parameter that can be increased to recover the canonical EOM-CCSDT result; this provides a practical mechanism for error control independent of any assumed bound on rank growth. We acknowledge that benchmarks on substantially larger systems would be needed to confirm the scaling behavior in the regime of greatest practical interest and have noted this explicitly as a direction for future work. revision: partial
-
Referee: The derivation of reduced scaling from the linear dimension of the decomposed amplitudes (abstract and formulation sections) is formally correct only if the chosen decomposition rank is independent of system size. The manuscript does not supply explicit bounds, scaling plots of rank versus N, or tests on systems larger than those shown, leaving the practical N^6 claim unverified for the regime where the method would be most useful.
Authors: The scaling analysis in the formulation section is derived under the condition that the Tucker rank is held fixed for a chosen accuracy target, which reduces the effective dimension of the decomposed triple amplitudes to scale linearly with N and thereby yields N^6 computational cost. We have revised the relevant sections to state this assumption more explicitly and to clarify that the rank parameter serves as the convergence control: increasing it systematically recovers the parent method without altering the underlying contraction structure. No mathematical bounds on rank growth are supplied because the rank is selected numerically to meet a prescribed error tolerance rather than being derived from a priori estimates. We have not added new scaling plots because the present work focuses on establishing the method and demonstrating accuracy on representative small-molecule cases with varying excitation character; larger-system verification remains outside the current scope. revision: yes
- Explicit numerical verification of rank growth and effective scaling on molecular systems significantly larger than Mg2 and NH3-F2.
Circularity Check
No significant circularity; derivation self-contained via explicit rank parameter and numerical validation
full rationale
The paper presents a rank-reduced EOM-CCSDT method by applying Tucker decomposition to the triply-excited amplitudes tensors. Reduced scaling to N^6 follows directly from the stated linear dimension of the decomposed tensors with system size N, which is a property of the chosen decomposition format rather than a fitted or self-referential result. The central accuracy claim (error several times smaller than parent method's inherent error with default settings) is asserted via explicit numerical tests on molecules with varied excitation characters, not by construction or reduction to inputs. A single tunable rank parameter is introduced to control accuracy and recover the canonical limit, with no evidence of self-definitional loops, fitted quantities renamed as predictions, or load-bearing self-citations. The formulation is independent of external benchmarks and does not invoke uniqueness theorems or prior ansatze from the same authors in a circular manner.
Axiom & Free-Parameter Ledger
free parameters (1)
- decomposition rank parameter
axioms (2)
- domain assumption Tucker decomposition can be applied to the triply-excited amplitudes tensors while preserving the essential physics of the excitation.
- standard math Standard EOM-CCSDT framework remains valid when the triple amplitudes are replaced by their decomposed form.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.