Self-consistent vertex corrected $GW$ with static and dynamic screening using tensor hypercontraction: assessment of molecular ionization potentials

Chia-Nan Yeh; Dominika Zgid; Miguel A. Morales; Ming Wen; Munkhorgil Wang; Pavel Pokhilko

arxiv: 2604.25581 · v1 · submitted 2026-04-28 · ❄️ cond-mat.str-el · physics.chem-ph

Self-consistent vertex corrected GW with static and dynamic screening using tensor hypercontraction: assessment of molecular ionization potentials

Munkhorgil Wang , Ming Wen , Pavel Pokhilko , Chia-Nan Yeh , Miguel A. Morales , Dominika Zgid This is my paper

Pith reviewed 2026-05-07 15:13 UTC · model grok-4.3

classification ❄️ cond-mat.str-el physics.chem-ph

keywords tensor hypercontractionself-consistent GWvertex correctionsionization potentialsmolecular benchmarkselectronic structure theory

0 comments

The pith

Tensor hypercontraction accelerates self-consistent GW and vertex-corrected GW for molecular ionization potentials with negligible error.

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

The paper benchmarks tensor hypercontraction versions of fully self-consistent GW and vertex-corrected GW against standard molecular ionization potential data sets. It establishes that the THC approximation adds almost no deviation from the exact self-consistent results. Vertex corrections produce largely uniform shifts in the predicted IPs rather than better agreement with measured values. This combination shows that the accelerated methods remain faithful to the parent theories while becoming cheaper to run.

Core claim

The THC decomposition introduces negligible errors into self-consistent GW ionization potentials, indicating that the acceleration preserves the underlying fully self-consistent results. Across both the G0W0Γ29 and GW100 benchmark sets, vertex-corrected scGWΓ methods primarily produce systematic shifts in the IPs relative to scGW rather than consistent accuracy improvements. These results identify THC as a reliable route to lower-cost scGW and scGWΓ calculations.

What carries the argument

Tensor hypercontraction (THC) decomposition applied to the integrals that enter the self-energy in the self-consistent GW and GWΓ equations.

Load-bearing premise

The G0W0Γ29 and GW100 molecular sets are representative enough to conclude that vertex corrections yield only systematic shifts without consistent accuracy gains.

What would settle it

A THC-scGW calculation on a molecule outside the GW100 set that produces an ionization potential differing from the non-THC reference by more than the negligible threshold reported in the paper.

Figures

Figures reproduced from arXiv: 2604.25581 by Chia-Nan Yeh, Dominika Zgid, Miguel A. Morales, Ming Wen, Munkhorgil Wang, Pavel Pokhilko.

**Figure 1.** Figure 1: Hedin’s Pentagon. The interacting one-body Green’s function view at source ↗

**Figure 2.** Figure 2: Skeleton Feynman diagrams for the self-energy approximations considered in this view at source ↗

**Figure 3.** Figure 3: Basis set convergence of Etot, E stat 1b and E dyn 2b using the logarithmic scale, with respect to αIpts for selected molecules. The energy contributions are calculated at the THCscGW/cc-pVQZ level. For clarity, energy differences smaller than 10−12 a.u. are set to zero. All 29-molecules are reported in Supporting Information Figs. S1 and S2 interactions40 and magnetic interactions. 60 In view at source ↗

**Figure 4.** Figure 4: The convergence behavior of the trace of the imaginary part of the dynamic self view at source ↗

**Figure 5.** Figure 5: The trace of the imaginary part of the dynamic self-energy ImTr[Σ( view at source ↗

**Figure 6.** Figure 6: The trace of the real part of the dynamic self-energy ReTr(Σ( view at source ↗

**Figure 7.** Figure 7: Upper panel: The first IP differences in eV for the view at source ↗

**Figure 8.** Figure 8: The convergence of the first IP value for water produced for all six THC-sc view at source ↗

**Figure 9.** Figure 9: The first IP values calculated by THC-sc view at source ↗

**Figure 10.** Figure 10: Imaginary part of the trace self-energy, ImTr[Σ( view at source ↗

**Figure 11.** Figure 11: Real part of the trace self-energy, ReTr[Σ( view at source ↗

read the original abstract

In this work, we benchmark tensor hypercontraction (THC)-accelerated fully self-consistent $GW$ (sc$GW$) and vertex-corrected self-consistent $GW$ (sc$GW\Gamma$) methods for predicting molecular first ionization potentials (IPs). The vertex function, $\Gamma$, is inserted into the self-energy in a fully self-consistent manner, and representative sc$GW$ and sc$GW\Gamma$ variants are assessed across the $G_0W_0\Gamma29$ and $GW100$ data sets. We find that the THC decomposition introduces negligible errors into self-consistent $GW$ ionization potentials, indicating that the acceleration preserves the underlying fully self-consistent results. Across both benchmark sets, vertex-corrected sc$GW\Gamma$ methods primarily produce systematic shifts in the IPs relative to sc$GW$ rather than consistent accuracy improvements. These results identify THC as a reliable route to lower-cost sc$GW$ and sc$GW\Gamma$ calculations

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

THC keeps scGW IPs close to the non-accelerated versions on these sets, while vertex corrections mainly shift values without clear accuracy gains.

read the letter

The central result is that tensor hypercontraction introduces negligible errors into the self-consistent GW ionization potentials, so the acceleration largely preserves the original numbers on the molecules they checked. Vertex-corrected scGWΓ versions produce systematic shifts relative to plain scGW rather than consistent improvements against experiment on the G0W0Γ29 and GW100 sets. That combination is the practical takeaway: THC looks usable for making these calculations cheaper without wrecking the IPs, but adding the vertex does not deliver the accuracy boost one might hope for on these benchmarks. The work applies an existing THC technique to fully self-consistent GW and GWΓ, then runs the comparison across the two standard molecular IP datasets. It reports that the decomposition holds up well enough to keep the underlying scGW results intact, which is the kind of concrete check that matters for people who actually want to run these methods on larger systems. The paper is straightforward about the outcomes and sticks to the data it has. The soft spot is that the abstract and the stress-test note leave open whether the THC error checks were performed on the full sets or only on smaller subsets where the overhead is lower. If the validation skipped the larger molecules or did not report size-dependent error statistics, the claim that the acceleration preserves results everywhere rests on an extrapolation that needs explicit support in the text. The finding on vertex corrections is also tied to these particular implementations and datasets, so it may not hold more broadly. This is for readers who work on Green's function methods in quantum chemistry and want to know which accelerations are safe to use. The benchmarks are relevant enough that a serious editor should send it to referees for a detailed check on the error tables and implementation choices.

Referee Report

1 major / 1 minor

Summary. The manuscript benchmarks tensor hypercontraction (THC)-accelerated fully self-consistent GW (scGW) and vertex-corrected scGWΓ methods for predicting molecular first ionization potentials (IPs) on the G0W0Γ29 and GW100 datasets. It asserts that the THC decomposition introduces negligible errors relative to conventional scGW, thereby preserving the underlying fully self-consistent results, and that vertex corrections primarily induce systematic shifts in IPs rather than consistent accuracy gains.

Significance. If the THC error assessment holds across the full benchmark sets, the work establishes a computationally efficient route to self-consistent GW and vertex-corrected GW calculations for molecular IPs, which could enable studies of larger systems. The systematic comparison of scGW and scGWΓ variants on standard datasets also clarifies the practical impact of vertex corrections in this context.

major comments (1)

[Results section (assessment of G0W0Γ29 and GW100)] Results section (assessment of G0W0Γ29 and GW100): the central claim that 'the THC decomposition introduces negligible errors' and 'preserves the underlying fully self-consistent results' requires explicit quantitative metrics (e.g., mean absolute deviation, maximum deviation, and any size dependence) for the complete sets of molecules. If these statistics are reported only for subsets or smaller systems, the extrapolation to the full benchmarks is not load-bearing and must be strengthened with full-set error tables.

minor comments (1)

[Abstract] Abstract: the statement of 'negligible errors' and 'systematic shifts' would benefit from inline numerical values (e.g., average IP shift or MAE) to make the claims immediately quantifiable.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thorough review and constructive feedback on our manuscript. We address the major comment below and have revised the manuscript to strengthen the presentation of our results.

read point-by-point responses

Referee: Results section (assessment of G0W0Γ29 and GW100): the central claim that 'the THC decomposition introduces negligible errors' and 'preserves the underlying fully self-consistent results' requires explicit quantitative metrics (e.g., mean absolute deviation, maximum deviation, and any size dependence) for the complete sets of molecules. If these statistics are reported only for subsets or smaller systems, the extrapolation to the full benchmarks is not load-bearing and must be strengthened with full-set error tables.

Authors: We appreciate the referee highlighting this point. While the original manuscript included error assessments on representative molecules and smaller subsets to illustrate the THC accuracy, we agree that explicit quantitative metrics for the complete G0W0Γ29 and GW100 sets are required to fully support the central claim. In the revised manuscript, we have added comprehensive full-set error tables (new Tables S3 and S4 in the supplementary information, with corresponding discussion in the main text Results section). These tables report the mean absolute deviation, maximum deviation, root-mean-square deviation, and an assessment of size dependence between THC-accelerated and conventional scGW ionization potentials across all molecules in both benchmark sets. The revisions confirm that the THC errors remain negligible for the full datasets. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or validation chain

full rationale

The paper reports numerical benchmarks of THC-accelerated scGW and scGWΓ ionization potentials against conventional (non-THC) implementations on the G0W0Γ29 and GW100 molecular datasets. The central assertion that THC decomposition introduces negligible errors is an empirical observation drawn from direct error statistics, not a mathematical derivation that reduces to its own inputs by definition, a fitted parameter relabeled as a prediction, or a load-bearing self-citation chain. No equations are presented whose outputs are forced by construction from the same quantities used as inputs; the assessment relies on external benchmark sets and reported mean/max errors rather than self-referential logic.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

No new free parameters or invented entities; work relies on established GW theory and THC approximation from prior literature.

axioms (1)

domain assumption GW approximation and its self-consistent and vertex-corrected extensions provide valid starting points for molecular electronic structure.
Invoked when benchmarking scGW and scGWΓ variants on ionization potentials.

pith-pipeline@v0.9.0 · 9469 in / 1032 out tokens · 101197 ms · 2026-05-07T15:13:09.453193+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

(1) Maggio, E.; Kresse, G.GWVertex Corrected Calculations for Molecular Systems.J. Chem. Theory Comput.2017,13, 4765–4778. (2) Wen, M.; Abraham, V.; Harsha, G.; Shee, A.; Whaley, K. B.; Zgid, D. Comparing Self-ConsistentGWand Vertex-CorrectedG 0W0 (G0W0Γ) Accuracy for Molecular Ion- ization Potentials.Journal of Chemical Theory and Computation2024,20, 310...

work page 2017
[2]

Measure- ments of lithium cluster ionization potentials.Chemical Physics Letters1992,197, 433–437

(4) Dugourd, P.; Rayane, D.; Labastie, P.; Vezin, B.; Chevaleyre, J.; Broyer, M. Measure- ments of lithium cluster ionization potentials.Chemical Physics Letters1992,197, 433–437. (5) Trickl, T.; Cromwell, E. F.; Lee, Y. T.; Kung, A. H. State-selective ionization of nitrogen in theX 2Σ+ g v+ = 0 andv + = 1 states by two-color (1+1) photon excitation near ...

work page 2025
[3]

forbidden

(16) Kreile, J.; Schweig, A.; Theil, W. Experimental and theoretical investigation of the photoionization of hydrogen cyanide.Chemical Physics Letters1982,87, 473–476. (17) Vovna, V. I.; Vilesov, F. I.; Lopatin, S. N. Photoelectron Spectra of Hydrazine and Some Alkyl Derivatives.Optics and Spectroscopy1975,38, 143–144. 15 (18) Vorob’ev, A. S.; Furlei, I. ...

work page 2021

[1] [1]

(1) Maggio, E.; Kresse, G.GWVertex Corrected Calculations for Molecular Systems.J. Chem. Theory Comput.2017,13, 4765–4778. (2) Wen, M.; Abraham, V.; Harsha, G.; Shee, A.; Whaley, K. B.; Zgid, D. Comparing Self-ConsistentGWand Vertex-CorrectedG 0W0 (G0W0Γ) Accuracy for Molecular Ion- ization Potentials.Journal of Chemical Theory and Computation2024,20, 310...

work page 2017

[2] [2]

Measure- ments of lithium cluster ionization potentials.Chemical Physics Letters1992,197, 433–437

(4) Dugourd, P.; Rayane, D.; Labastie, P.; Vezin, B.; Chevaleyre, J.; Broyer, M. Measure- ments of lithium cluster ionization potentials.Chemical Physics Letters1992,197, 433–437. (5) Trickl, T.; Cromwell, E. F.; Lee, Y. T.; Kung, A. H. State-selective ionization of nitrogen in theX 2Σ+ g v+ = 0 andv + = 1 states by two-color (1+1) photon excitation near ...

work page 2025

[3] [3]

forbidden

(16) Kreile, J.; Schweig, A.; Theil, W. Experimental and theoretical investigation of the photoionization of hydrogen cyanide.Chemical Physics Letters1982,87, 473–476. (17) Vovna, V. I.; Vilesov, F. I.; Lopatin, S. N. Photoelectron Spectra of Hydrazine and Some Alkyl Derivatives.Optics and Spectroscopy1975,38, 143–144. 15 (18) Vorob’ev, A. S.; Furlei, I. ...

work page 2021