Machine learning interatomic potentials (MLIPs) have become emerging tools in molecular modeling and computational chemistry. By learning high-dimensional potential energy surfaces from quantum chemical data, MLIPs enable accurate and efficient predictions of structural, thermodynamic, and dynamical properties. However, such models have limitations in predictions of electronic properties and the effects of static electron correlation due to their lack of electronic structure information. This work presents OrbGNN, an electronic structure graph architecture analogous to molecular graph and MLIP frameworks, where pair-orbital interactions constitute the graph representation, while orbital entanglement encodes the connectivity between them. By embedding information derived from orbital correlation metrics directly into the graph topology, OrbGNN provides a compact representation of a molecule s orbital landscape and electron correlation patterns. Analysis of the behavior of the feature space in an orbital graph are shown to demonstrate model robustness. The model is evaluated for the dissociation of nitrogen and for a larger dataset of diatomic molecules. Finally, the OrbGNN model is applied to a set of octahedral iron(II) complexes to predict spin-state energy gaps.
Brody Quebedeaux, Shahzad Akram, Markus Reiher et al.· 0 citations
Transition state (TS) search is a crucial step in understanding chemical reactivity and mechanisms, yet conventional algorithms remain computationally intensive and heavily reliant on initial guesses, user s expertise, and chemical intuition. While recent machine learning approaches have shown promise, they demand either large training datasets or geometric interpolation between known endpoints, limiting their generality. In this work, we introduce a TS search model based on the soft actor-critic model, an advanced reinforcement learning algorithm in which an agent learns to navigate potential energy surfaces directly from local energetic and curvature information starting from a given reactant and its corresponding product. By formulating the search as a sequential decision-making process in internal coordinates, the agent adaptively proposes chemically meaningful structural updates through a reward function designed to promote movement towards saddle point regions. Without labelled trajectories or prescribed reaction pathways, the method successfully identifies TS geometries for standard benchmark reactions, operating directly on realistic molecular potential energy surfaces. These results highlight the potential of RL as a general strategy for reducing dependence on initial guesses and enabling scalable, automated reaction discovery across diverse chemical systems.
Utham Suresh, Konstantinos D. Vogiatzis· 0 citations