How Atoms Interact Within Molecules
Adil Kabylda, Malte Esders, Matteo Gori, Stefan Chmiela, Klaus-Robert Müller, Alexandre Tkatchenko · 2026
Read How Atoms Interact Within Molecules by Adil Kabylda, Malte Esders, Matteo Gori, Stefan Chmiela, Klaus-Robert Müller, Alexandre Tkatchenko (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-2605-28960.txt.
Introduction
Adil Kabylda and colleagues compute interatomic forces in molecules of hundreds of atoms using quantum field theory for long-range electron correlation and machine-learned force fields. Average interaction strength falls off polynomially with separation, but the scatter of forces stays robust and strongly directional. Larger molecules amplify that scatter and direction dependence further, a pattern missed by textbook empirical models. The results shift attention from single atoms to interacting hotspots that may steer how biopolymers fold, and they set new benchmarks for force models.
Catalog record
| CREATOR | Adil Kabylda, Malte Esders, Matteo Gori, Stefan Chmiela, Klaus-Robert Müller, Alexandre Tkatchenko |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50763 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.28960
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
How Atoms Interact Within Molecules Adil Kabylda, Malte Esders, Matteo Gori, Stefan Chmiela, Klaus-Robert Müller, Alexandre Tkatchenko Abstract Fundamental understanding of interatomic forces in molecules must emerge from quantum mechanics, yet widely used empirical force fields rely on simplified mechanistic approximations that often fail to capture the complexity of many-body systems. Here we employ recent developments in quantum field theory (QFT) for long-range electron correlation and machine learning force fields (MLFFs) to directly compute the depth and scatter of interatomic forces for molecular systems containing hundreds of atoms. We find that while the average interaction strength decays polynomially with interatomic separation, the interaction scatter remains robust and exhibits substantial anisotropy. Both QFT and MLFFs demonstrate that increasing the molecular size further amplifies this scatter and anisotropy – a phenomenon not considered in traditional textbook empirical models. These results provide new benchmarks for force models, shift the focus from interacting atoms to interacting “hotspots” that might determine the folding pathways of (bio)polymers, and rationalize why MLFFs are uniquely successful in capturing the nuances of complex molecular systems. Our findings offer a roadmap for the development of more accurate and quantum-aware molecular force fields. I Introduction Understanding the nature of interatomic interactions is fundamental to molecular and materials science, as it directly impacts our ability to model and predict their behavior. Although all interactions are inherently quantum mechanical, their complexity has historically necessitated the use of simplified models 13 ; 9 ; 31 . The characteristics of interatomic forces, particularly their range, anisotropy (angular deviation from the vector connecting two atoms), and functional dependence, remain an area of active investigation 45 ; 19 ; 43 . The central questions are: What is the effective distance range of interatomic interactions needed for accurate simulations, how does it scale with molecular size, and how anisotropic are interactions at various interatomic distances? The motivation for this study goes back to the pioneering work of Langmuir and Gurvich more than a century ago, who debated the characteristic distances of interatomic and intermolecular interactions, proposing ranges from roughly 1 to 300 Å 16 ; 28 . Today, advances in computational methods and experimental techniques have refined these scales, suggesting that critical bonded interactions occur below 5 Å distances (Fig. 1 A). Although short-range anisotropy in interatomic potentials has been well-established (for example, due to non-spherical valence electron distributions such as lone pairs and \pi -bonds) 44 , long-range anisotropy has received comparatively little attention. Traditional models tend to assume isotropy at longer ranges, relying on pairwise additive schemes for non-covalent electrostatic and dispersion forces that decay as R^{-2} or R^{-7} , respectively 23 ; 30 ; 48 ; 15 . However, these assumptions neglect the multiscale nature of interactions that potentially leads to anisotropic contributions that may persist or amplify at greater distances between atoms or larger molecular fragments. Indeed, evidence from multiple domains of molecular science points to a more complex picture. Second derivatives of the ab initio energy surface 10 ; 18 ; 17 , orientation-dependent fits from high-level quantum theory 50 , and symmetry-adapted perturbation theory (SAPT) studies quantifying anisotropy in exchange and dispersion 46 ; 27 , all show that the standard “sum-of-spheres” approximation fails unless non-spherical features such as virtual sites or angular terms are explicitly included. Experimentally, phenomena such as rapid specific protein association 38 and the anisotropic inter-layer van der Waals potentials of 2D materials point to long-range forces acting as directional “steering” mechanisms rather than isotropic attractions 44 ; 34 . Given that such persistent anisotropy is evident even in small simple molecules, the assumption that it averages out or vanishes in larger, more complex systems (in particular, for atoms interacting within molecules) is not well founded. Beyond the covalent distance regime, there are several kinds of contributions to interatomic interactions. They can arise from permanent, induced, and fluctuating multipoles. Electrostatic and polarization interactions depend on permanent and induced dipoles and are usually screened in extended systems 21 ; 35 , either because multipoles do not scale with molecular size or due to orientational polarization and multipole cancellation 45 ; 52 . Dispersion interactions involve both fluctuating and induced multipoles and depend on high-frequency (not static) dielectric properties of matter. Dispersion interactions are usually attractive between ground-state fragments, scale with system size, and arise from correlated zero-point plasmonic modes in the optical frequency range that are wave-like and delocalized throughout the molecular framework 1 ; 42 ; 14 . Consequently, dispersion forces can converge slower with distance than electrostatic forces (see Fig. S1 for molecules studied in this work). Figure 1: Long-range electron correlation interactions exhibit non-uniform decay and growing anisotropy in a protein. (A) Framework for analyzing force contributions F_{ij} between atoms i and j , where \sum_{j}F_{ij}=F_{i} gives the total force on atom i . The angle \theta_{ij} measures the alignment between the interatomic vector d_{ij} and the force contribution F_{ij} . Short-range interactions ( <5 Å ) are known to exhibit anisotropy, while long-range interactions are often assumed isotropic in force fields. (B) Interaction heatmap for Chignolin ( F_{ij} ), obtained through QFT and MLFF decompositions averaged over 100 snapshots from an ab initio molecular dynamics trajectory spanning folded, semi-folded, and unfolded conformations (see Methods for details). (C) Interaction depth, i.e. the decay of the interaction strength |F_{ij}| with interatomic distance. The dashed R^{-4} to R^{-7} reference curves are anchored at R=5 Å to a global fit of |F_{ij}|=c\,R^{-7} over the 5 – 10 Å range. The 2D histogram shows substantial scatter at each distance with non-uniform decay. (D) Polar representation of the angular \theta_{ij} distribution vs. interatomic distance. The heatmap demonstrates that directional dependence (deviation from \theta_{ij}=180^{\circ} , which corresponds to pairwise attractive interactions; dashed sector) grows systematically at longer range. The magnitude and directional character of long-range forces exhibit complex, environment- and distance-dependent behavior not captured by pairwise potentials. This study examines how atoms interact within molecules by leveraging new developments in Quantum Field Theory (QFT) for long-range electron correlation 25 and Machine Learning Force Fields (MLFF) 49 . These methods allow us to probe how interaction depth and directional anisotropy evolve across length scales and system sizes. Across systems ranging from small biomolecular units to a 35-residue protein, we find that the interaction depth exhibits broad scatter that defies simple power-law decay, and that the anisotropy of interatomic forces persists at long-range and amplifies with system size. II Results The central challenge in characterizing interatomic interactions is disentangling the contributions of individual atom pairs from the collective quantum mechanical behavior of the full many-body molecular system. To address this, we develop and apply a framework for extracting pairwise force contributions F_{ij} and their angular alignments \theta_{ij} from three distinct computational approaches: Second-Quantized Many-Body Dispersion (SQ-MBD) theory 47 ; 11 ; 14 (with and without mean-field DFT contributions), the PaiNN machine learning force field 41 , and a Mechanistic Empirical Force Field (MEFF). These methods span a wide range of physical approximations, from the fully quantum, non-local description of SQ-MBD to the atom-type-based MEFF, enabling a systematic assessment of which physical ingredients are necessary to reproduce the true force landscape. We apply this framework to five molecular systems of diverse structural complexity: the tetrapeptide