Assessing the impact of nodal surface optimization in fixed-node diffusion Monte Carlo on non-covalent interactions
Kousuke Nakano, Benjamin X. Shi, Dario Alfè, Andrea Zen · 2026
Read Assessing the impact of nodal surface optimization in fixed-node diffusion Monte Carlo on non-covalent interactions by Kousuke Nakano, Benjamin X. Shi, Dario Alfè, Andrea Zen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-04329.txt.
Introduction
Nakano and colleagues test whether optimizing the nodal surface improves diffusion Monte Carlo predictions for non-covalent interactions. Diffusion Monte Carlo and coupled cluster are standard benchmarks here, yet recent work finds gaps between them. In diffusion Monte Carlo the error is expected to come from the fixed-node approximation, which takes its nodal surface from one determinant. Using a new antisymmetrized geminal power ansatz with natural orbitals across 12 compounds, optimized nodes improve agreement with coupled cluster for hydrogen-bonded systems but barely matter for dispersion ones.
Catalog record
| CREATOR | Kousuke Nakano, Benjamin X. Shi, Dario Alfè, Andrea Zen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26474 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.04329
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Assessing the impact of nodal surface optimization in fixed-node diffusion Monte Carlo on non-covalent interactions Kousuke Nakano, Benjamin X. Shi, Dario Alfè, Andrea Zen Abstract Diffusion quantum Monte Carlo (DMC) and coupled cluster theory [CCSD(T)] are widely-employed benchmark methods for noncovalent interactions (NCIs). However, recent studies have reported notable discrepancies across several hydrogen-bonded and dispersion-dominated systems, raising questions on the accuracy of the approximations underlying each approach. In DMC, the dominant error is expected to stem from the fixed-node approximation, where the nodal surface is typically taken from a single Slater determinant derived from a density functional theory or Hartree–Fock calculation. In this work, we assess the impact of nodal surface optimization on DMC predictions for 12 compounds spanning diverse NCIs, using a recently proposed antisymmetrized geminal power ansatz with natural orbitals. We find improved agreement with CCSD(T) for hydrogen-bonded systems, while having negligible effect for dispersion-dominated systems. These results provide a practical and computationally efficient route to resolving discrepancies in hydrogen-bonded interactions, while offering insight into the remaining differences in dispersion-dominated systems. † † email: [email protected] † † affiliation: Center for Basic Research on Materials (CBRM), National Institute for Materials Science (NIMS), 1-2-1 Sengen, Tsukuba, Ibaraki 305-0047, Japan † † affiliation: Initiative for Computational Catalysis, Flatiron Institute, 160 5th Avenue, New York, NY 10010, USA † † affiliation: Dipartimento di Fisica Ettore Pancini, Università di Napoli Federico II, Monte S. Angelo, I-80126 Napoli, Italy † † affiliation: Department of Earth Sciences, University College London, Gower Street, London WC1E 6BT, United Kingdom † † affiliation: Thomas Young Centre and London Centre for Nanotechnology, 17-19 Gordon Street, London WC1H 0AH, United Kingdom † † email: [email protected] † † affiliation: Dipartimento di Fisica Ettore Pancini, Università di Napoli Federico II, Monte S. Angelo, I-80126 Napoli, Italy † † affiliation: Department of Earth Sciences, University College London, Gower Street, London WC1E 6BT, United Kingdom The central task of ab initio electronic structure calculations is to solve the many-body Schrödinger equation. Several methods have been developed to solve this equation, including wavefunction methods (quantum chemistry calculations), density functional theory (DFT), and quantum Monte Carlo (QMC). On the one hand, coupled cluster theory with single, double, and perturbative triple excitations [CCSD(T)] 37 has been regarded as a reference method within the wavefunction methods. On the other hand, the Fixed-node Diffusion quantum Monte Carlo with Jastrow Slater Determinant ansatz (FN-SD-DMC) 15 has also been regarded as a reference method, especially in materials science and condensed matter physics. For small molecular systems, FN-SD-DMC and CCSD(T) typically yield consistent values, e.g., for intermolecular binding energies. However, in 2021, Al-Hamdani et al. reported significant discrepancies between CCSD(T) and FN-SD-DMC calculations when evaluating the binding energies of large molecules bound by van der Waals forces, such as C 60 buckyball inside a [6]-cycloparaphenyleneacetylene ring (C 60 +5PPA) 1 . Fishman et al. reported that such discrepancy is also observed in \pi -stacking complexes such as acene and alkadiene dimers 14 . Furthermore, in 2025, more binding energies were benchmarked on the S66 dataset 46 of medium-sized dimers. It was reported that discrepancies between CCSD(T) and FN-SD-DMC were observed not only for large molecules bound by van der Waals forces but also for some molecules bound by hydrogen bonds. Since neither CCSD(T) nor FN-SD-DMC provides exact solutions, but rather each employs approximations, this is why no definitive conclusion has been reached regarding which is more accurate. Figure 1: The binding energies computed by FN-SD-DMC, FN-AGPn-DMC, FN-BF-DMC, CCSD(T), CCSDT(Q), and CCSD(cT)-fit for (a) hydrogen bonds and (b) Dispersion-dominated bonds. The CCSD(T), CCSDT(Q), CCSD(cT)-fit (except for HCOOH–HCOOH, Ethene–Ethyne, and Ethyne–Ethyne, which were computed in this work using the same setting in Ref. 46 ), and FN-BF-DMC values are taken from Refs. 46 , 22 , 46 , and 21 , respectively. E_{b} in the x labels refer to the binding energies (kcal/mol) obtained by FN-SD-DMC in this work. The most uncontrollable approximation in FN-SD-DMC is the fixed-node approximation with a mean-field nodal surface, arising from the so-called Fermion sign problem 15 . That is, FN-SD-DMC results are determined by the nodal surface provided by a preceding mean-field calculation, such as HF or DFT calculations. In this situation, what can be done from the FN-SD-DMC side is to investigate whether the binding energy obtained by using a better wavefunction beyond a mean-field nodal surface remains unchanged, or approaches the CCSD(T) value. This allows discussion of whether the discrepancy between CCSD(T) and FN-SD-DMC stems from the mean-field nodal surfaces. Many variational ansatzes beyond the single Slater-determinant have been proposed. Examples are multideterminant ansatz 13 ; 53 ; 52 , antisymmetrized geminal power (AGP) ansatz 8 , backflow 12 ; 24 ; 4 , Pfaffian 3 ; 4 ; 16 , and more recently, neural-network-based ansatzes 18 ; 35 . They have been applied to binding energy calculations of various molecules 17 ; 47 ; 2 ; 30 ; 16 ; 38 ; 21 . A pioneering work in the context of the present study is the FN-DMC calculation with the backflow ansatz (denoted as FN-BF-DMC in this study) for AcOH–AcOH and water-peptide by Lambie et al. 21 The study revealed that the consistency between CCSD(T) and FN-BF-DMC is much better than that between CCSD(T) and FN-SD-DMC; thus, the reported discrepancy between CCSD(T) and FN-SD-DMC is attributed to the mean-field nodal surface used in FN-SD-DMC. However, in any ansatz, the drawback is the inevitable trade-off between flexibility and computational cost. More flexible ansatzes yield lower variational energies but it requires higher computational cost. This prevents one from studying diverse compounds at a reasonable computational cost. In fact, Lambie et al. 21 stated that their FN-BF-DMC calculations used 2.9 million and 1.5 million core-hours for AcOH–AcOH and water-peptide on an AMD EPYC cluster machine, respectively, while their FN-SD-DMC counterparts cost 0.42 and 0.27 million core-hours, respectively. This can make it challenging to go towards larger compounds. We recently developed a DMC workflow along this line based on AGP ansatz with natural orbitals (denoted as FN-AGPn-DMC in this study) 32 , where only a relatively small number of pairing coefficients of the geminal pairing function are optimized. This is obtained by writing the pairing function in terms of natural orbitals constructed from a beyond HF/DFT calculation and keeping them frozen in the optimization while optimizing their weights. Since AGPn is a generalization of a single Slater determinant, FN-AGPn-DMC is variationally better than FN-SD-DMC. Moreover, the computational cost of FN-AGPn-DMC has the same scaling as that of FN-SD-DMC 32 . This enables one to study large systems with better nodal surfaces than mean-field ones at an affordable computational cost. This Letter reports DMC binding energy calculations from the FN-AGPn-DMC approach for diverse non-covalent interaction (NCI) compounds, and a comparison with FN-SD-DMC and CCSD(T) evaluations. We selected 12 NCI compounds where discrepancies between FN-SD-DMC and CCSD(T) have been reported. The 12 compounds are taken from the S22 19 , S66 42 , and A24 39 datasets, and are divided into two primary categories: hydrogen bonds and dispersion-dominated interactions ( \pi -stacking, pure London dispersion, and mixed-interactions). Our calculation reveals that, for hydrogen bonds, the FN-AGPn-DMC yields smaller binding energies than the FN-SD-DMC does; thus, provides bonding energies very close to