Simultaneous Learning of Static and Dynamic Charges

Philipp Stärk, Henrik Stooß, Marcel F. Langer, Egor Rumiantsev, Alexander Schlaich, Michele Ceriotti, Philip Loche · 2026

Read Simultaneous Learning of Static and Dynamic Charges by Philipp Stärk, Henrik Stooß, Marcel F. Langer, Egor Rumiantsev, Alexander Schlaich, Michele Ceriotti, Philip Loche (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-2601-03656.txt.

Introduction

Philipp Stärk and six co-authors compare ways to teach machine learning models two kinds of atomic charge at once. Static charges drive the electric forces between atoms, while dynamic charges describe how atoms answer an external field. Learning both well matters for accurate simulations of materials and liquids, yet the two are usually handled separately.

Using bulk water and water clusters as test cases, the team compared learning the charges independently with two coupled schemes. Coupling uses the physical link between the charges and must correct for dielectric screening. A single global screening constant worked poorly in uneven systems like clusters, where a local, environment-dependent factor did better. The findings guide builders of atomistic models toward charge schemes that respect real chemistry.

Catalog record

CREATORPhilipp Stärk, Henrik Stooß, Marcel F. Langer, Egor Rumiantsev, Alexander Schlaich, Michele Ceriotti, Philip Loche
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-03656

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Simultaneous Learning of Static and Dynamic Charges Philipp Stärk, Henrik Stooß, Marcel F. Langer, Egor Rumiantsev, Alexander Schlaich, Michele Ceriotti, Philip Loche Abstract Long-range interactions and electric response are essential for accurate modeling of condensed-phase systems, but capturing them efficiently remains a challenge for atomistic machine learning. Traditionally, these two phenomena can be represented by static charges, that participate in Coulomb interactions between atoms, and dynamic charges such as atomic polar tensors—aka Born effective charges—describing the response to an external electric field. We critically compare different approaches to learn both types of charges, taking bulk water and water clusters as paradigmatic examples: (1) Learning them independently; (2) Coupling static and dynamic charges based on their physical relationship with a single global coupling constant to account for dielectric screening; (3) Coupled learning with a local, environment-dependent screening factor. In the coupled case, correcting for dielectric screening is essential, yet the common assumption of homogeneous, isotropic screening breaks down in heterogeneous systems such as water clusters. A learned, environment-dependent screening restores high accuracy for the dynamical charges. However, the accuracy gain over independent dynamic predictions is negligible, while the computational cost increases compared to using separate models for static and dynamical charges. This suggests that, despite the formal connection between the two charge types, modeling them independently is the more practical choice for both condensed-phase and isolated cluster systems. I Introduction Electric fields influence the structure, dynamics, and reactivity of molecular and condensed-phase systems. Processes such as catalysis, charge transport, or characterization techniques such as infrared (IR) spectroscopy depend sensitively on how atoms respond to internal and external electric fields Pan et al. 2022 ; Ciampi et al. 2018 ; Salles et al. 2020 . First-principles electronic-structure methods capture these responses with high accuracy, but their computational cost limits their use for large, heterogeneous systems or long simulation times. Machine-learned interatomic potentials (MLIPs) have begun to address this challenge by providing near-quantum-chemical accuracy for the structure and dynamics of molecules and condensed phases at drastically reduced cost Unke et al. 2021 . Modern architectures—often graph neural networks with equivariant message-passing—excel at learning many-body interactions Schütt et al. 2017 ; Thomas et al. 2018 ; Batzner et al. 2022 ; Batatia et al. 2022 . However, most MLIPs remain restricted to field-free simulations: they do not natively incorporate the effects of external fields or predict electrical response properties, limiting their applicability in technologically relevant scenarios. Several strategies have been developed to endow MLIPs with electric-field awareness. One class of methods learns dipoles or local dipole contributions Falletta et al. 2024 ; Stocco et al. 2025 ; Veit et al. 2020 , but these approaches face fundamental challenges as described within the modern theory of polarization. For ionic (i.e. systems with free charges) and periodic systems in general, dipoles are defined only modulo a polarization quantum Resta and Vanderbilt 2007 ; Spaldin 2012 ; Stocco et al. 2025 , possibly complicating the training. Other approaches bypass these issues by learning scalar effective charges Dufils et al. 2023 ; Li et al. 2025 or tensorial quantities such as Born effective charges (BECs), also called atomic polar tensors (APTs) Schienbein 2023 ; Joll et al. 2024 ; Bergmann et al. 2025 . These models couple external fields to learned charges, generating auxiliary field-dependent forces and thus avoid the conceptual ambiguities of polarization-based methods. BEC-based models perform well in capturing the linear electric response. However, these models typically neglect an important physical ingredient: the external electric-field response of atoms is intimately linked to the electrostatic interaction between atoms. Effective charges used to model long-range Coulomb forces reflect internally screened interactions, whereas BECs describe the unscreened response to an external field. Importantly, this distinction becomes pronounced in heterogeneous systems such as water clusters, interfaces, or molecular mixtures. Besides electric-field response, long-range electrostatics have already been incorporated in MLIPs using fixed charges Bartók et al. 2010 ; Deng et al. 2019 , ML-fitted surrogate charges such as Hirshfeld charges Artrith et al. 2011 ; Morawietz et al. 2012 , dipole-matching schemes Yao et al. 2018 , or global charge-equilibration networks Ko et al. 2021 ; Dufils et al. 2023 ; Li et al. 2025 ; Gao et al. 2024 . Recently, differentiable Ewald summation frameworks Cheng 2025 ; Loche et al. 2025 have enabled end-to-end charge learning, removing ambiguities in charge assignment Gross et al. 2002 ; Reed et al. 1985 ; Hirshfeld 1977 . A few recent studies attempt to combine long-range electrostatics with electric-field response, for example by deriving BECs from end-to-end learned internal charges Zhong et al. 2025 ; Kim et al. 2025 . However, these approaches rely on the assumption of homogeneous isotropic screening in the model connecting learned charges with BECs, which breaks down outside bulk systems as we will show in this work. This breakdown raises a central open question: should static (effective) and dynamic (BEC) charges be learned together, constrained by their physical relationship, or treated as independent quantities? Here, we address this question by systematically comparing models that learn static and dynamic charges either independently or in coupled forms with isotropic, homogeneous screening, or environment-dependent screening factors. Using high-quality finite-field reference data for water clusters and bulk water, we critically compare different models in terms of accuracy and computational costs. II Theory and Models II.1 Static and Dynamic Charges A key concept for describing the response of atomistic systems to externally applied fields is the BEC (or APT) of an atom, which is defined via Schienbein 2023 ; Joll et al. 2024 Z^{i}_{\alpha\beta}=V\partialderivative{\mathcal{P}_\alpha}{r^i_{\beta}}=\partialderivative{F^i_{\beta}}{E_{\alpha, \text{ext}}}, (1) where \mathcal{P}_{\alpha} is the system polarization density, r^{i}_{\beta} the position of atom i , V is the system volume, F_{\beta}^{i} the force on atom i and E_{\alpha,\text{ext}} an externally applied field. The variables \alpha,\beta\in\{x,y,z\} specify Cartesian components of the tensorial quantities. The right-hand side of eq. 1 motivates the use of these charges for describing electrostatic interactions with external fields, as it represents the corresponding linear-response coefficient—the change in force induced by an applied external field. We consider for the moment non-periodic systems, where we define partial charges q_{\text{IR}}^{i} that reproduce the polarization \bm{P}=\mathcal{P}V \bm{P}=\sum_{j}q_{\text{IR}}^{j}\bm{r}^{j}. (2) This allows us to relate BECs Z^{i} to static charges q^{i}_{\text{IR}} through Milani and Castiglioni 2010 Z^{i}_{\alpha\beta}=q^{i}_{\text{IR}}\delta_{\alpha\beta}+\sum_{j}\partialderivative{q _{\text{IR}}^j}{r^i_\beta}r_{\alpha}^{j}, (3) where the first term mimics the “static charges” q^{i}_{\text{IR}} , which are sometimes also referred to as the “IR charge” (as it reproduces the system’s dipole and thus yields the infrared spectrum, as we will discuss below). The second term in eq. 3 describes the charge flux: the change in the dipole moment due to a collective alteration of the charge distribution as the i -th atom is displaced by an infinitesimal amount. Because Z^{i} encodes these charge redistribution effects, BECs are also called “dynamic charges” . II.2 Long-Ranged Machine Learned Potentials Via Learned Static Charges As briefly explained in the introduction, a wide range of methods have been proposed for assigning partial “static” charges to nuclei from the continuous electron density. Popular schemes include Mulliken charges Mulliken 1955 , Hirshfeld charges Hirshfeld 1977 and the aforementioned IR charges Milani and Castiglioni 2010 . Each of these methods corresponds to a different definition of the partial charges, and there is no

Read the complete text