arXiv subjects

Julia Westermayr

Publications and source records attributed to Julia Westermayr.

At least 19 recordsLinked to original sources

arXiv2026

Hierarchical generative modeling for the design of multi-component systems

The functionality of catalysts, enzymes, and supramolecular assemblies emerges not from individual molecules alone, but from the subtle interplay between multiple components arranged in complex systems. Designing such systems is a grand challenge, the combinatorial explosion of possible chemical compositions and spatial arrangements makes brute-force exploration infeasible, while many current generative approaches remain limited to isolated molecules. In this work, we introduce a hierarchical generative optimization framework that overcomes this barrier by coupling a genetic algorithm for configurational search with a generative model for molecular design. This closed-loop approach enables simultaneous refinement of geometry and composition, efficiently steering discovery toward systems with targeted functionality. As a proof of concept, we design catalytic environments for the Claisen rearrangement of p-tolyl ether by optimizing surrounding components around a fixed reference transition-state geometry. Despite this constraint during the search phase, post-hoc validation via Climbing-Image Nudged Elastic Band calculations confirm a 30% reduction in activation barrier. Beyond this example, our framework provides a general strategy for data-driven discovery of functional multi-component systems, opening the door to automated design of catalysts, enzyme active sites, and advanced materials. Scientific contribution. The study presents a closed loop generative framework that enables joint optimization of molecular components and their spatial organization in multi-component systems. The method moves generative molecular design beyond single molecules toward larger and more complex systems.

physics.comp-ph↗

arXiv2026

Transferable excited-state dynamics enable screening of fluorescent protein chromophores

Transferable excited-state dynamics offer a route to efficient screening of photophysical behavior across molecular systems, but conventional nonadiabatic simulations remain prohibitively expensive. Here we introduce X-MACE, a transferable machine-learning potential for excited-state dynamics that predicts multiple potential energy surfaces, forces and oscillator strengths, and combine it with curvature-driven surface hopping to enable data-efficient screening of photochemical pathways. We apply this framework to fluorescent chromophores as an example application, using green fluorescent protein chromophore variants to demonstrate how subtle structural modifications reshape excited-state relaxation, lifetimes and photoisomerization yields. Fine-tuning a single pretrained model with fewer than 100 reference geometries per derivative yields accurate dynamics across a chemically diverse set of analogues. The screening reveals two governing design principles: steric crowding on the phenolate ring lowers the torsional barrier and accelerates access to twisted conical intersections, whereas conjugation extension stabilizes planar excited-state configurations, suppresses non-radiative decay and prolongs fluorescence. More broadly, this workflow provides a general framework for scalable excited-state screening and interpretable design of photophysical properties.

physics.chem-ph↗

arXiv2025

Machine learning for smell: Ordinal odor strength prediction of molecular perfumery components

Predicting olfactory perception directly from molecular structure is central to fragrance design that plays a role in a wide range of industries, such as perfumery, food and beverage, and health care. Among olfactory attributes, odor strength is a key factor in shaping odor perception, but its modeling has been impeded by scarce and fragmented intensity data. In this work, we introduce an ordinal odor strength data set of over 2,000 molecules by integrating two different public sources, mapping structures to odorless, low, medium, and high categories. Across several molecular encodings and supervised learning algorithms we compared different prediction strategies. Dimensionality reduction and SHAP analysis identifies molecular size, polarity, ring features, and branching as primary drivers, consistent with mass-transport constraints on volatility, sorption, and receptor access. This scalable ordinal framework enables reliable odor-strength estimation for novel molecules and provides a foundation for in silico fragrance design.

physics.chem-ph↗

arXiv2025

Statistics makes a difference: Machine learning adsorption dynamics of functionalized cyclooctine on Si(001) at DFT accuracy

The interpretation of experiments on reactive semiconductor surfaces requires statistically significant sampling of molecular dynamics, but conventional ab initio methods are limited due to prohibitive computational costs. Machine-learning interatomic potentials provide a promising solution, bridging the gap between the chemical accuracy of short ab initio molecular dynamics (AIMD) and the extensive sampling required to simulate experiment. Using ethinyl-functionalized cyclooctyne adsorption on Si(001) as a model system, we demonstrate that conventional AIMD undersamples the configurational space, resulting in discrepancies with scanning tunnelling microscopy and X-ray photoelectron spectroscopy data. To resolve these inconsistencies, we employ pre-trained equivariant message-passing neural networks, fine-tuned on only a few thousand AIMD snapshots, and integrate them into a "molecular-gun" workflow. This approach generates 10,000 independent trajectories more than 1,000 times faster than AIMD. These simulations recover rare intermediates, clarify the competition between adsorption motifs, and reproduce the experimentally dominant on-top [2+2] cycloaddition geometry. Our results show that fine-tuning of pre-trained foundational models enables statistically converged, chemically accurate simulations of bond-forming and bond-breaking events on complex surfaces, providing a scalable route to reconcile atomistic theory with experimental ensemble measurements in semiconductor functionalization.

cond-mat.mtrl-sci↗

arXiv2025

A practical guide to machine learning interatomic potentials -- Status and future

The rapid development and large body of literature on machine learning interatomic potentials (MLIPs) can make it difficult to know how to proceed for researchers who are not experts but wish to use these tools. The spirit of this review is to help such researchers by serving as a practical, accessible guide to the state-of-the-art in MLIPs. This review paper covers a broad range of topics related to MLIPs, including (i) central aspects of how and why MLIPs are enablers of many exciting advancements in molecular modeling, (ii) the main underpinnings of different types of MLIPs, including their basic structure and formalism, (iii) the potentially transformative impact of universal MLIPs for both organic and inorganic systems, including an overview of the most recent advances, capabilities, downsides, and potential applications of this nascent class of MLIPs, (iv) a practical guide for estimating and understanding the execution speed of MLIPs, including guidance for users based on hardware availability, type of MLIP used, and prospective simulation size and time, (v) a manual for what MLIP a user should choose for a given application by considering hardware resources, speed requirements, energy and force accuracy requirements, as well as guidance for choosing pre-trained potentials or fitting a new potential from scratch, (vi) discussion around MLIP infrastructure, including sources of training data, pre-trained potentials, and hardware resources for training, (vii) summary of some key limitations of present MLIPs and current approaches to mitigate such limitations, including methods of including long-range interactions, handling magnetic systems, and treatment of excited states, and finally (viii) we finish with some more speculative thoughts on what the future holds for the development and application of MLIPs over the next 3-10+ years.

cond-mat.mtrl-sci↗

arXiv2025

Incorporating Long-Range Interactions via the Multipole Expansion into Ground and Excited-State Molecular Simulations

Simulating long-range interactions remains a significant challenge for molecular machine learning potentials due to the need to accurately capture interactions over large spatial regions. In this work, we introduce FieldMACE, an extension of the message-passing atomic cluster expansion (MACE) architecture that integrates the multipole expansion to model long-range interactions more efficiently. By incorporating the multipole expansion, FieldMACE effectively captures environmental and long-range effects in both ground and excited states. Benchmark evaluations demonstrate its superior performance in predictions and computational efficiency compared to previous architectures, as well as its ability to accurately simulate nonadiabatic excited-state dynamics. Furthermore, transfer learning from foundational models enhances data efficiency, making FieldMACE a scalable, robust, and transferable framework for large-scale molecular simulations.

physics.comp-ph↗

arXiv2025

Transferable Machine Learning Potential X-MACE for Excited States using Integrated DeepSets

Conical intersections serve as critical gateways in photochemical reactions, enabling rapid nonradiative transitions between potential energy surfaces that underpin fundamental processes such as photosynthesis or vision. Their calculation with quantum chemistry is, however, extremely computationally intensive and their modeling with machine learning poses a significant challenge due to their inherently non-smooth and complex nature. To address this challenge, we introduce a deep learning architecture designed to precisely model excited states and improve their accuracy around these critical, non-smooth regions. Our model integrates Deep Sets into the Message Passing Atomic Cluster Expansion (MACE) framework resulting in a smooth representation of the non-smooth excited-state potential energy surfaces. We validate our method using numerous molecules, showcasing a significant improvement in accurately modeling the energy landscape around conical intersections compared to conventional excited-state models. Additionally, we apply ground-state foundational machine learning models as a basis for excited states. By doing so, we showcase that the developed model is capable of transferring not only from the ground state to excited states, but also within chemical space to molecular systems beyond those included in the training dataset. This advancement not only enhances the fidelity of excited-state modeling, but also lays the foundations for the investigation of more complex molecular systems.

physics.chem-ph↗

arXiv2025

Artificial intelligence for objective assessment of acrobatic movements: How to apply machine learning for identifying tumbling elements in cheer sports

Over the past four decades, cheerleading has evolved from a sideline activity at major sporting events into a professional, competitive sport with growing global popularity. Evaluating tumbling elements in cheerleading relies on both objective measures and subjective judgments, such as difficulty and execution quality. However, the complexity of tumbling - encompassing team synchronicity, ground interactions, choreography, and artistic expression - makes objective assessment challenging. Artificial intelligence (AI) has revolutionized various scientific fields and industries through precise data-driven analyses, yet their application in acrobatic sports remains limited despite significant potential for enhancing performance evaluation and coaching. This study investigates the feasibility of using an AI-based approach with data from a single inertial measurement unit to accurately identify and objectively assess tumbling elements in standard cheerleading routines. A sample of 16 participants (13 females, 3 males) from a Division I collegiate cheerleading team wore a single inertial measurement unit at the dorsal pelvis. Over a 4-week seasonal preparation period, 1102 tumbling elements were recorded during regular practice sessions. Using triaxial accelerations and rotational speeds, various ML algorithms were employed to classify and evaluate the execution of tumbling manoeuvres. Results indicate that certain machine learning models can effectively identify different tumbling elements despite inter-individual variability and data noise, achieving high accuracy. These findings demonstrate the significant potential for integrating AI-driven assessments into cheerleading and other acrobatic sports, providing objective metrics that complement traditional judging methods.

cs.CY↗

arXiv2025

Excited-state nonadiabatic dynamics in explicit solvent using machine learned interatomic potentials

Excited-state nonadiabatic simulations with quantum mechanics/molecular mechanics (QM/MM) are essential to understand photoinduced processes in explicit environments. However, the high computational cost of the underlying quantum chemical calculations limits its application in combination with trajectory surface hopping methods. Here, we use FieldSchNet, a machine-learned interatomic potential capable of incorporating electric field effects into the electronic states, to replace traditional QM/MM electrostatic embedding with its ML/MM counterpart for nonadiabatic excited state trajectories. The developed method is applied to furan in water, including five coupled singlet states. Our results demonstrate that with sufficiently curated training data, the ML/MM model reproduces the electronic kinetics and structural rearrangements of QM/MM surface hopping reference simulations. Furthermore, we identify performance metrics that provide robust and interpretable validation of model accuracy.

physics.chem-ph↗

arXiv2024

Parameterizing Intersecting Surfaces via Invariants

We introduce and analyze numerical companion matrix methods for the reconstruction of hypersurfaces with crossings from smooth interpolants given unordered or, without loss of generality, value-sorted data. The problem is motivated by the desire to machine learn potential energy surfaces arising in molecular excited state computational chemistry applications. We present simplified models which reproduce the analytically predicted convergence and stability behaviors as well as two application-oriented numerical experiments: the electronic excited states of Graphene featuring Dirac conical cusps and energy surfaces corresponding to a sulfur dioxide ($SO_2$) molecule in different configurations.

math.NA↗

arXiv2023

Machine learning interatomic potentials for reactive hydrogen dynamics at metal surfaces based on iterative refinement of reaction probabilities

Reactive chemistry of molecular hydrogen at surfaces, notably dissociative sticking and hydrogen evolution, plays a crucial role in energy storage and fuel cells. Theoretical studies can help to decipher underlying mechanisms and reaction design, but studying dynamics at surfaces is computationally challenging due to the complex electronic structure at interfaces and the high sensitivity of dynamics to reaction barriers. In addition, ab initio molecular dynamics, based on density functional theory, is too computationally demanding to accurately predict reactive sticking or desorption probabilities as it requires averaging over tens of thousands of initial conditions. High-dimensional machine learning-based interatomic potentials are starting to be more commonly used in gas-surface dynamics, yet, robust approaches to generate reliable training data and to assess how model uncertainty affects the prediction of dynamic observables are not well established. Here, we employ ensemble learning to adaptively generate training data while assessing model performance with full uncertainty quantification for reaction probabilities of hydrogen scattering on different copper facets. We use this approach to investigate the performance of two message-passing neural networks, SchNet and PaiNN. Ensemble-based uncertainty quantification and iterative refinement allow us to expose the shortcomings of the invariant pairwise-distance-based feature representation in the SchNet model for gas-surface dynamics.

physics.chem-ph↗

arXiv2023

Reinforcement learning for traversing chemical structure space: Optimizing transition states and minimum energy paths of molecules

In recent years, deep learning has made remarkable strides, surpassing human capabilities in tasks like strategy games, and it has found applications in complex domains, including protein folding. In the realm of quantum chemistry, machine learning methods have primarily served as predictive tools or design aids using generative models, while reinforcement learning remains in its early stages of exploration. This work introduces an actor-critic reinforcement learning framework suitable for diverse optimization tasks, such as searching for molecular structures with specific properties within conformational spaces. As an example, we show an implementation of this scheme for calculating minimum energy pathways of a Claisen rearrangement reaction and a number of SN2 reactions. Our results show that the algorithm is able to accurately predict minimum energy pathways and thus, transition states, therefore providing the first steps in using actor-critic methods to study chemical reactions.

physics.chem-ph↗

arXiv2023

Decoding Energy Decomposition Analysis: Machine-Learned Insights on the Impact of the Density Functional on the Bonding Analysis

The concept of chemical bonding is a crucial aspect of chemistry that aids in understanding the complexity and reactivity of molecules and materials. However, the interpretation of chemical bonds can be hindered by the choice of the theoretical approach and the specific method utilized. This study aims to investigate the effect of choosing different density functionals on the interpretation of bonding achieved through energy decomposition analysis (EDA). To achieve this goal, a data set was created, representing four bonding groups and various combinations of functionals and dispersion correction schemes. The calculations showed significant variation among the different functionals for the EDA terms, with the dispersion correction terms exhibiting the highest variability. More information was extracted by using unsupervised learning in combination with dimensionality reduction on the data set. Results indicate that, despite the differences in the EDA terms obtained from different functionals, the functional has the least significant impact, suggesting minimal influence on the bonding interpretation.

physics.comp-ph↗

arXiv2023

Machine learning for accelerated bandgap prediction in strain-engineered quaternary III-V semiconductors

Quaternary III-V semiconductors are one of the major promising material classes in optoelectronics. The bandgap and its character, direct or indirect, are the most important fundamental properties determining the performance and characteristics of optoelectronic devices. Experimental approaches screening a large range of possible combinations of III- and V-elements with variations in composition and strain are impractical for every target application. We present a combination of accurate first-principles calculations and machine learning based approaches to predict the properties of the bandgap for quaternary III-V semiconductors. By learning bandgap magnitudes and their nature at density functional theory accuracy based solely on the composition and strain features of the materials as an input, we develop a computationally efficient yet highly accurate machine learning approach that can be applied to a large number of compositions and strain values. This allows for a computationally efficient prediction of a vast range of materials under different strains, offering the possibility for virtual screening of multinary III-V materials for optoelectronic applications.

cond-mat.mtrl-sci↗

arXiv2023

High-throughput property-driven generative design of functional organic molecules

The design of molecules and materials with tailored properties is challenging, as candidate molecules must satisfy multiple competing requirements that are often difficult to measure or compute. While molecular structures, produced through generative deep learning, will satisfy those patterns, they often only possess specific target properties by chance and not by design, which makes molecular discovery via this route inefficient. In this work, we predict molecules with (pareto)-optimal properties by combining a generative deep learning model that predicts three dimensional conformations of molecules with a supervised deep learning model that takes these as inputs and predicts their electronic structure. Optimization of (multiple) molecular properties is achieved by screening newly generated molecules for desirable electronic properties and reusing hit molecules to retrain the generative model with a bias. The approach is demonstrated to find optimal molecules for organic electronics applications. Our method is generally applicable and eliminates the need for quantum chemical calculations during predictions, making it suitable for high-throughput screening in materials and catalyst design.

physics.chem-ph↗

arXiv2022

BuRNN: Buffer Region Neural Network Approach for Polarizable-Embedding Neural Network / Molecular Mechanics Simulations

Hybrid quantum mechanics/molecular mechanics (QM/MM) simulations have advanced the field of computational chemistry tremendously. However, they require the partitioning of a system into two different regions that are treated at different levels of theory, which can cause artefacts at the interface. Furthermore, they are still limited by high computational costs of quantum chemical calculations. In this work, we develop BuRNN, an alternative approach to existing QM/MM schemes, which introduces a buffer region that experiences full electronic polarization by the inner QM region to minimize artefacts. The interactions between the QM and the buffer region are described by deep neural networks (NNs), which leads to high computational efficiency of this hybrid NN/MM scheme while retaining quantum chemical accuracy. We demonstrate the BuRNN approach by performing NN/MM simulations of the hexa-aqua iron complex.

physics.chem-ph↗

arXiv2022

NQCDynamics.jl: A Julia Package for Nonadiabatic Quantum Classical Molecular Dynamics in the Condensed Phase

Accurate and efficient methods to simulate nonadiabatic and quantum nuclear effects in high-dimensional and dissipative systems are crucial for the prediction of chemical dynamics in condensed phase. To facilitate effective development, code sharing and uptake of newly developed dynamics methods, it is important that software implementations can be easily accessed and built upon. Using the Julia programming language, we have developed the NQCDynamics.jl package which provides a framework for established and emerging methods for performing semiclassical and mixed quantum-classical dynamics in condensed phase. The code provides several interfaces to existing atomistic simulation frameworks, electronic structure codes, and machine learning representations. In addition to the existing methods, the package provides infrastructure for developing and deploying new dynamics methods which we hope will benefit reproducibility and code sharing in the field of condensed phase quantum dynamics. Herein, we present our code design choices and the specific Julia programming features from which they benefit. We further demonstrate the capabilities of the package on two examples of chemical dynamics in condensed phase: the population dynamics of the spin-boson model as described by a wide variety of semi-classical and mixed quantum-classical nonadiabatic methods and the reactive scattering of H2 on Ag(111) using the Molecular Dynamics with Electronic Friction method. Together, they exemplify the broad scope of the package to study effective model Hamiltonians and realistic atomistic systems.

physics.chem-ph↗

arXiv2022

Long-range dispersion-inclusive machine learning potentials for structure search and optimization of hybrid organic-inorganic interfaces

The computational prediction of the structure and stability of hybrid organic-inorganic interfaces provides important insights into the measurable properties of electronic thin film devices, coatings, and catalyst surfaces and plays an important role in their rational design. However, the rich diversity of molecular configurations and the important role of long-range interactions in such systems make it difficult to use machine learning (ML) potentials to facilitate structure exploration that otherwise require computationally expensive electronic structure calculations. We present an ML approach that enables fast, yet accurate, structure optimizations by combining two different types of deep neural networks trained on high-level electronic structure data. The first model is a short-ranged interatomic ML potential trained on local energies and forces, while the second is an ML model of effective atomic volumes derived from atoms-in-molecules partitioning. The latter can be used to connect short-range potentials to well-established density-dependent long-range dispersion correction methods. For two systems, specifically gold nanoclusters on diamond (110) surfaces and organic $π$-conjugated molecules on silver (111) surfaces, we train models on sparse structure relaxation data from density functional theory and show the ability of the models to deliver highly efficient structure optimizations and semi-quantitative energy predictions of adsorption structures.

cond-mat.mtrl-sci↗