arXiv subjects

Johannes Kästner

Publications and source records attributed to Johannes Kästner.

At least 19 recordsLinked to original sources

arXiv2026

Curvature-Aware Flow Matching for Molecular Structure Generation

Generative models produce three-dimensional molecular structures with high geometric accuracy, yet they are trained on distributions of atomic coordinates without explicit access to the underlying physics. The potential energy surface (PES) describes how the energy changes around a molecular geometry, and the RMSD between a generated structure and its reference captures this local landscape only indirectly. Supervising the PES during training could close this gap, but conformers and transition states (TSs) are stationary points at which the forces vanish, so first-order quantities carry little information about them. What remains is the local curvature, encoded in the Hessian of the PES. Quantum-chemical Hessians are too expensive to be available at the scale required for training, but recently proposed machine-learned predictors make curvature inexpensive enough to supervise directly. We introduce CurvFM, a flow matching (FM) formulation whose training objective constrains the curvature at the generated structure. A frozen Hessian prediction model is evaluated at the generated endpoint and at the reference structure, which removes the need for quantum-chemical Hessians and keeps the residual cheap enough to impose at every training step. Across TS and conformer generation, CurvFM substantially improves physical fidelity at unchanged global structural accuracy, reducing the median deviation in maximum force by 63% on GEOM-QM9 and raising the fraction of valid TSs by up to 10% across three reaction datasets.

physics.chem-ph↗

arXiv2026

ElemeNet: Multiscale Molecular Machine Learning with Uncertainty Quantification Across the Periodic Table

Advances in deep learning architectures and representations have enabled ML-driven chemical property prediction, but state-of-the-art (SOTA) models have remained largely confined to independent codebases and lack support for diverse chemical species. This work introduces ElemeNet, a unified, general-purpose software package for molecular machine learning. The ElemeNet software package enables the training of advanced ML models for diverse properties and datasets with an enlarged range of elemental compositions. We define molecular representations compatible with elements 1-100, supporting diverse organometallic and biological systems in addition to organic chemistry already well-served by the Chemprop ML toolkit. As well as more common atom-, bond-, and molecule-level predictions, we introduce moiety predictions. We also natively define optional conditioning on charge and spin states. Advanced E(3)-equivariant and transformer architectures are supported, as well as classical 2D models, with all classes including built-in uncertainty quantification through deterministic and statistical measures. We benchmark our protocols for ML model training against representative datasets from organic, inorganic, coordination, and biological chemistry, achieving competitive and SOTA performance relative to literature baselines and favorable scaling to millions of molecules. The entire workflow is exposed through a concise command-line interface, lowering the barrier to entry for non-expert users. We anticipate ElemeNet will empower non-computational researchers to leverage modern deep learning methods across the chemical and physical sciences.

physics.chem-ph↗

arXiv2026

How to Train a Shallow Ensemble

Shallow ensembles provide a convenient strategy for uncertainty quantification in machine learning interatomic potentials, that is computationally efficient because the different ensemble members share a large part of the model weights. In this work, we systematically investigate training strategies for shallow ensembles to balance calibration performance with computational cost. We first demonstrate that explicit optimization of a negative log-likelihood (NLL) loss improves calibration with respect to approaches based on ensembles of randomly initialized models, or on a last-layer Laplace approximation. However, models trained solely on energy objectives yield miscalibrated force estimates. We show that explicitly modeling force uncertainties via an NLL objective is essential for reliable calibration, though it typically incurs a significant computational overhead. To address this, we validate an efficient protocol: full-model fine-tuning of a shallow ensemble originally trained with a probabilistic energy loss, or one sampled from the Laplace posterior. This approach results in negligible reduction in calibration quality compared to training from scratch, while reducing training time by up to 96%. We evaluate this protocol across a diverse range of materials, including amorphous carbon, ionic liquids (BMIM), liquid water (H$_2$O), barium titanate (BaTiO$_3$), and a model tetrapeptide (Ac-Ala3-NHMe), establishing practical guidelines for reliable uncertainty quantification in atomistic machine learning.

physics.chem-ph↗

arXiv2026

Atom Addition Formation of Thionylimide (HNSO) on Interstellar Dust Grains: Chemical routes requiring oxygen and nitrogen atom surface diffusion

We investigate the formation of the recently detected HNSO molecule using quantum chemical calculations on ices and astrochemical models in tandem. Our results indicate that HNSO is efficiently produced on grain surfaces through reactions involving atomic oxygen and nitrogen atoms with the radicals NS and SO, forming NSO as a key intermediate. Subsequent hydrogenation of NSO leads to HNSO, with a clear preference for the lowest energy cis conformer, while the trans form is metastable and may be short-lived under typical interstellar conditions. The models predict that solid HNSO can reach abundances comparable to icy OCS, placing it among the major sulfur-bearing species in interstellar ices. Gas-phase abundances, in contrast, remain lower than those of OCS. The implementation of a multibinding scheme in our models clarifies the role of diffusive chemistry in the production of HNSO at early times, improving agreement with observations. These findings suggest that reactions involving diffusing O and N atoms on icy grains contribute significantly to sulfur chemistry and beyond in dense clouds and motivate further searches for molecules containing simultaneously H, N, O and S in other astronomical environments.

astro-ph.GA↗

arXiv2026

Beyond the Training Domain: Robust Generative Transition State Models for Unseen Chemistry

Transition states (TSs) govern the rates and outcomes of chemical reactions, making their accurate prediction a central challenge in computational chemistry. Although recent machine-learning models achieve near chemical accuracy in the prediction of TS structures and the associated reaction barriers for small organic reactions, their ability to generalize beyond the training domain remains largely unexplored. Here, we introduce targeted benchmarks to probe chemical and structural novelty in generative TS prediction. Building on Transition1x, a large-scale dataset of reactions involving small organic molecules, we construct curated extensions incorporating controlled elemental substitutions and diverse transition-metal complexes (TMC). These benchmarks reveal fundamental limitations of generative models in the generalization to previously unseen elements. As a result, they produce unphysical geometries and large energetic errors, even for reactions structurally similar to well-predicted organic systems. To address this challenge, we introduce a self-supervised pretraining strategy based on equilibrium conformers that exposes generative TS models to novel chemical environments prior to targeted fine-tuning. Across the newly proposed benchmarks, self-supervised pretraining substantially improves TS prediction for previously unseen systems, lowering the median RMSD of TS geometries on T1x-TMC reactions from 0.39 to 0.19 $\mathring{A}$ and reducing fine-tuning data requirements by up to 75%, enabling reliable performance even in low-data regimes. Overall, the integration of generative TS models with self-supervised pseudo-reaction pretraining provides an efficient, scalable, and chemically robust framework for elucidating TSs well beyond the small organic molecule domain, establishing a foundation for investigating complex and catalytically relevant reaction landscapes.

physics.chem-ph↗

arXiv2026

Enhanced Representation-Based Sampling for the Efficient Generation of Datasets for Machine-Learned Interatomic Potentials

In this work, we present Enhanced Representation-Based Sampling (ERBS), a novel enhanced sampling method designed to generate structurally diverse training datasets for machine-learned interatomic potentials. ERBS automatically identifies collective variables by dimensionality reduction of atomic descriptors and applies a bias potential inspired by the On-the-Fly Probability Enhanced Sampling framework. We highlight the ability of Gaussian moment descriptors to capture collective molecular motions and explore the impact of biasing parameters using alanine dipeptide as a benchmark system. We show that free energy surfaces can be reconstructed with high fidelity using only short biased trajectories as training data. Further, we apply the method to the iterative construction of a liquid water dataset and compare the quality of simulated self-diffusion coefficients for models trained with molecular dynamics and ERBS data. Further, we active-learn models for liquid water with and without enhanced sampling and compare the quality of simulated self-diffusion coefficients. The self-diffusion coefficients closely match those simulated with a reference model at a significantly reduced dataset size. Finally, we compare the sampling behaviour of enhanced sampling methods by benchmarking the mean squared displacements of \ce{BMIM+BF4-} trajectories simulated with uncertainty-driven dynamics and ERBS and find that the latter significantly increases the exploration of configurational space.

physics.chem-ph↗

arXiv2025

CO Adsorption Sites on Interstellar Water Ices Explored with Machine Learning Potentials. Binding energy distributions and snowline

Context. Carbon monoxide (CO) is arguably the most important molecule for interstellar organic chemistry. Its binding to amorphous solid water (ASW) ice regulates both diffusion and desorption processes. Accurately characterizing the CO binding energy (BE) is essential for realistic astrochemical modeling. Aims. We aim to derive a statistically robust and physically accurate distribution of CO BEs on ASW surfaces, and to evaluate its implications for laboratory temperature-programmed desorption experiments and interstellar chemistry, with a focus on protoplanetary disks. Methods. We trained a machine-learned potential (MLP) on 8321 density functional theory (DFT) energies and gradients of CO interacting with differently-sized water clusters (22-60 water molecules). The DFT method was selected after extensive benchmark. With this potential we built realistic non-porous and porous ASW surfaces, and computed a BE distribution. We used symmetry-adapted perturbation theory to rationalize the interaction of CO on the different binding sites. Results. We find that both ASW morphologies yield similar Gaussian-like BE distributions with mean values near 900 K. However, the nature of the binding interactions is rather different and critically depends on surface roughness and dangling-OH bonds. Simulated TPD curves reproduce experimental trends across several coverage regimes. From an astrochemical point of view, the application of the full BE distribution has a dramatic influence on the CO distribution in protoplanetary disks, leading to a broader CO snowline region, improving predictions of CO gas-ice partitioning, and suggesting an equally broader distribution of organics in these objects.

astro-ph.GA↗

arXiv2025

Adaptive Transition State Refinement with Learned Equilibrium Flows

Identifying transition states (TSs), the high-energy configurations that molecules pass through during chemical reactions, is essential for understanding and designing chemical processes. However, accurately and efficiently identifying these states remains one of the most challenging problems in computational chemistry. In this work, we introduce a new generative AI approach that improves the quality of initial guesses for TS structures. Our method can be combined with a variety of existing techniques, including both machine learning models and fast, approximate quantum methods, to refine their predictions and bring them closer to chemically accurate results. Applied to TS guesses from a state-of-the-art machine learning model, our approach reduces the median structural error to just 0.088 $\unicode{x212B}$ and lowers the median absolute error in reaction barrier heights to 0.79 kcal mol$^{-1}$. When starting from a widely used tight-binding approximation, it increases the success rate of locating valid TSs by 41\% and speeds up high-level quantum optimization by a factor of three. By making TS searches more accurate, robust, and efficient, this method could accelerate reaction mechanism discovery and support the development of new materials, catalysts, and pharmaceuticals.

physics.chem-ph↗

arXiv2025

Apax: A Flexible and Performant Framework For The Development of Machine-Learned Interatomic Potentials

We introduce Atomistic learned potentials in JAX (apax), a flexible and efficient open source software package for training and inference of machine-learned interatomic potentials. Built on the JAX framework, apax supports GPU acceleration and implements flexible model abstractions for fast development. With features such as kernel-based data selection, well-calibrated uncertainty estimation, and enhanced sampling, it is tailored to active learning applications and ease of use. The features and design decisions made in apax are discussed before demonstrating some of its capabilities. First, a data set for the room-temperature ionic liquid EMIM+BF4- is created using active learning. It is highlighted how continuously learning models between iterations can reduce training times up to 85 % with only a minor reduction of the models' accuracy. Second, we show good scalability in a data-parallel training setting. We report that a Gaussian Moment Neural Network model, as implemented in apax, achieves higher accuracy and up to 10 times faster inference times than a performance-optimized Allegro model. A recently published Li3PO4 dataset, reported with comparable accuracy and inference performance metrics, is used as a point of comparison. Moreover, the inference speeds of the available simulation engines are compared. Finally, to highlight the modularity of apax, an equivariant message-passing model is trained as a shallow ensemble and used to perform uncertainty-driven dynamics.

physics.chem-ph↗

arXiv2025

Diffusive vs. non-diffusive paths to interstellar hydrogen peroxide. A machine learning-based molecular dynamics study

Context. Radical chemical reactions on cosmic dust grains play a crucial role in forming various chemical species. Among different radicals, the hydroxyl (OH) is one of the most important ones, with a rather specific chemistry. Aims. The goal of this work is to simulate the recombination dynamics of hydroxyl radicals and the subsequent formation of hydrogen peroxide (H$_{2}$O$_{2}$). Mthods. We employed neural network potentials trained on ONIOM(QM/QM) data, combining multi-reference (CASPT2) and density functional theory (DFT) calculations. This approach allows us to model the recombination of hydroxyl radicals on ice surfaces with high computational efficiency and accuracy. Results. Our simulations reveal that the initial position of the radicals plays a decisive role in determining recombination probability. We found that the formation of hydrogen-bond between radicals, competes with the formation of hydrogen peroxide, reducing the recombination efficiency, contrary to expectancy. This competition reduces the recombination probability for radicals that are initially formed approximately 3 Å apart. Recombination probabilities also depend on the kinetic energy of the added radicals, with values around 0.33 for thermal radicals and a wide range of values between 0.33 and 1.00 for suprathermal OH radicals. Conclusions. Based on our calculations we provide recommendations for introducing OH radical recombination into kinetic astrochemical models, differentiating between thermal and suprathermal radicals. The recombination behavior varies significantly between these two cases: while thermal radicals are sometimes trapped in hydrogen-bonded minima, the case of suprathermal radicals varies with the added energy. Our most important conclusion is that OH radical recombination probability cannot be assumed 1.0 for a wide variety of cases.

astro-ph.GA↗

arXiv2024

Uncertainty-biased molecular dynamics for learning uniformly accurate interatomic potentials

Efficiently creating a concise but comprehensive data set for training machine-learned interatomic potentials (MLIPs) is an under-explored problem. Active learning, which uses biased or unbiased molecular dynamics (MD) to generate candidate pools, aims to address this objective. Existing biased and unbiased MD-simulation methods, however, are prone to miss either rare events or extrapolative regions -- areas of the configurational space where unreliable predictions are made. This work demonstrates that MD, when biased by the MLIP's energy uncertainty, simultaneously captures extrapolative regions and rare events, which is crucial for developing uniformly accurate MLIPs. Furthermore, exploiting automatic differentiation, we enhance bias-forces-driven MD with the concept of bias stress. We employ calibrated gradient-based uncertainties to yield MLIPs with similar or, sometimes, better accuracy than ensemble-based methods at a lower computational cost. Finally, we apply uncertainty-biased MD to alanine dipeptide and MIL-53(Al), generating MLIPs that represent both configurational spaces more accurately than models trained with conventional MD.

physics.comp-ph↗

arXiv2024

Formation of the interstellar sugar precursor, (Z)-1,2-ethenediol, through radical reactions on dust grains

In recent years, the continued detection of complex organic molecules of prebiotic interest has refueled the interest on a panspermic origin of life. The prebiotic molecule glyceraldehyde is proposed to be formed from (Z)-1,2-ethenediol, a molecule recently detected towards the G+0.693-0.027 molecular cloud at the galactic center. In this work, we computationally simulate the formation of (Z)-1,2-ethenediol from vinyl alcohol on the surface of amorphous solid water in a two-step synthesis involving a OH addition and a H abstraction reaction. In total, we considered all reaction possibilities of the 1,1 and 1,2-OH addition to vinyl alcohol followed by H-abstraction or H-addition reactions on the resulting radicals. The combination of these reactions is capable of explaining the formation of (Z)-1,2-ethenediol provided a suprathermal diffusion of OH. We also conclude that our proposed formation pathway is not selective and also yields other abstraction and addition products. Key in our findings is the connection between the adsorption modes of the reactants and intermediates and the stereoselectivity of the reactions.

astro-ph.GA↗

arXiv2024

In-Depth Exploration of Catalytic Sites on Amorphous Solid Water: I. The Astrosynthesis of Aminomethanol

Chemical processes taking place on ice-grain mantles are pivotal to the complex chemistry of interstellar environments. In this study, we conducted a comprehensive analysis of the catalytic effects of an amorphous solid water (ASW) surface on the reaction between ammonia (NH$_3$) and formaldehyde (H$_2$CO) to form aminomethanol (NH$_2$CH$_2$OH) using density functional theory. We identified potential catalytic sites based on the binding energy distribution of NH$_3$ and H$_2$CO reactants, on a set-of-clusters surface model composed of 22 water molecules and found a total of 14 reaction paths. Our results indicate that the catalytic sites can be categorized into four groups, depending on the interactions of the carbonyl oxygen and the amino group with the ice surface in the reactant complex. A detailed analysis of the reaction mechanism using Intrinsic Reaction Coordinate and reaction force analysis revealed three distinct chemical events for this reaction: formation of the C--N bond, breaking of the N--H bond, and formation of the O--H hydroxyl bond. Depending on the type of catalytic site, these events can occur within a single, concerted, albeit asynchronous, step, or can be isolated in a step-wise mechanism, with the lowest overall transition state energy observed at 1.3 kcal mol$^{-1}$. A key requirement for the low-energy mechanism is the presence of a pair of dangling OH bonds on the surface, found at 5\% of the potential catalytic sites on an ASW porous surface.

astro-ph.GA↗

arXiv2024

ZnTrack -- Data as Code

The past decade has seen tremendous breakthroughs in computation and there is no indication that this will slow any time soon. Machine learning, large-scale computing resources, and increased industry focus have resulted in rising investments in computer-driven solutions for data management, simulations, and model generation. However, with this growth in computation has come an even larger expansion of data and with it, complexity in data storage, sharing, and tracking. In this work, we introduce ZnTrack, a Python-driven data versioning tool. ZnTrack builds upon established version control systems to provide a user-friendly and easy-to-use interface for tracking parameters in experiments, designing workflows, and storing and sharing data. From this ability to reduce large datasets to a simple Python script emerges the concept of Data as Code, a core component of the work presented here and an undoubtedly important concept as the age of computation continues to evolve. ZnTrack offers an open-source, FAIR data compatible Python package to enable users to harness these concepts of the future.

cs.SE↗

arXiv2023

Thermally Averaged Magnetic Anisotropy Tensors via Machine Learning Based on Gaussian Moments

We propose a machine learning method to model molecular tensorial quantities, namely the magnetic anisotropy tensor, based on the Gaussian-moment neural-network approach. We demonstrate that the proposed methodology can achieve an accuracy of 0.3--0.4 cm$^{-1}$ and has excellent generalization capability for out-of-sample configurations. Moreover, in combination with machine-learned interatomic potential energies based on Gaussian moments, our approach can be applied to study the dynamic behavior of magnetic anisotropy tensors and provide a unique insight into spin-phonon relaxation.

physics.comp-ph↗

arXiv2023

Predicting Properties of Periodic Systems from Cluster Data: A Case Study of Liquid Water

The accuracy of the training data limits the accuracy of bulk properties from machine-learned potentials. For example, hybrid functionals or wave-function-based quantum chemical methods are readily available for cluster data but effectively out-of-scope for periodic structures. We show that local, atom-centred descriptors for machine-learned potentials enable the prediction of bulk properties from cluster model training data, agreeing reasonably well with predictions from bulk training data. We demonstrate such transferability by studying structural and dynamical properties of bulk liquid water with density functional theory and have found an excellent agreement with experimental as well as theoretical counterparts.

physics.comp-ph↗

arXiv2023

A Framework and Benchmark for Deep Batch Active Learning for Regression

The acquisition of labels for supervised learning can be expensive. To improve the sample efficiency of neural network regression, we study active learning methods that adaptively select batches of unlabeled data for labeling. We present a framework for constructing such methods out of (network-dependent) base kernels, kernel transformations, and selection methods. Our framework encompasses many existing Bayesian methods based on Gaussian process approximations of neural networks as well as non-Bayesian methods. Additionally, we propose to replace the commonly used last-layer features with sketched finite-width neural tangent kernels and to combine them with a novel clustering method. To evaluate different methods, we introduce an open-source benchmark consisting of 15 large tabular regression data sets. Our proposed method outperforms the state-of-the-art on our benchmark, scales to large data sets, and works out-of-the-box without adjusting the network architecture or training code. We provide open-source code that includes efficient implementations of all kernels, kernel transformations, and selection methods, and can be used for reproducing our results.

stat.ML↗

arXiv2023

Processing of hydroxylamine, NH2OH, an important prebiotic precursor, on interstellar ices

Hydroxylamine, NH2OH, is one of the already detected interstellar molecules with the highest prebiotic potential. Yet, the abundance of this molecule found by astronomical observations is rather low for a relatively simple molecule, $\sim$ 10$^{-10}$ relative to H2. This seemingly low abundance can be rationalized by destruction routes operating on interstellar dust grains. In this work, we tested the viability of this hypothesis under several prisms, finding that the origin of a lower abundance of \ce{NH2OH} can be explained by two chemical processes, one operating at low temperature (10 K) and the other at intermediate temperature (20 K). At low temperatures, enabling the hydrogen abstraction reaction HNO + H -> NO + H2, even in small amounts, partially inhibits the formation of NH2OH through successive hydrogenation of NO, and reduces its abundance on the grains. We found that enabling a 15--30 % of binding sites for this reaction results in reductions of \ce{NH2OH} abundance of $\sim$ 1-2 orders of magnitude. At warmer temperatures (20 K, in our study), the reaction NH2OH + H -> HNOH + H2, which was found to be fast (k$\sim$10$^{6}$ s$^{-1}$) in this work, followed by further abstractions by adsorbates that are immobile at 10 K (O, N) are the main route of \ce{NH2OH} destruction. Our results shed light on the abundance of hydroxylamine in space and pave the way to constraining the subsequent chemistry experienced by this molecule and its derivatives in the interstellar prebiotic chemistry canvas.

astro-ph.GA↗