arXiv subjects

Jyri Kimari

Publications and source records attributed to Jyri Kimari.

5 recordsLinked to original sources

arXiv2026

A Self-Evolving Machine-Learning-Based Kinetic Monte Carlo Method for Modelling Thin-Film Growth

We present a kinetic Monte Carlo (KMC) simulation framework parameterized by automatically sampling machine-learning (ML) for modeling thin-film growth atom by atom. Given an interatomic potential energy function, the KMC algorithm builds an ML-based regression model for rate parameters on runtime, being trained on the local atomic environments encountered during the system evolution. New environments are continuously added to the training set in a self-evolving manner at points where the ML model estimates high uncertainty. As the simulation progresses, the ML model gains confidence, and the quick estimation of rates increasingly overtakes the relatively-expensive nudged elastic band calculations, promoting computational efficiency while retaining high fidelity description of the atomic diffusion kinetics. As a test case, we simulate the sub-monolayer growth of Ag on Ag {111}, where we demonstrate adatom islands forming in shapes and densities in accordance with the underlying atomistic interaction model, the theoretical framework, and available experimental results related to thin-film nucleation and growth.

cond-mat.mtrl-sci↗

arXiv2022

Biased self-diffusion on Cu surface due to electric field gradients

Under strong electric fields, an arc of strong current flowing through plasma can link two metal surfaces even in ultra high vacuum. Despite decades of research, the chain of events leading to vacuum arc breakdowns is hitherto unknown. Previously we showed that a tall and sharp Cu nanotip exposed to strong electric fields heats up by field emission currents and eventually melts, evaporating neutral atoms that can contribute to plasma buildup. In this work, we investigate by means of molecular dynamics simulations whether surface diffusion biased by the presence of an electric field gradient can provide sufficient mass transport of atoms toward the top of the nanotip to maintain supply of neutrals for feeding plasma. To reach the necessary timescales and to add electric field in MD, we utilized a novel combination of collective variable~-driven hyperdynamics acceleration and coupling to a finite element mesh. In our simulations, we observed biased self-diffusion on Cu surfaces, that can contribute to the continuous replenishment of particle-emitting nanotips. This mechanism implies a need to reduce the rate of surface diffusion in devices that are susceptible to vacuum arcs. Finding suitable alloys or surface treatments that hinder the observed biased diffusion could guide the design of future devices, and greatly improve their efficiency.

cond-mat.mtrl-sci↗

arXiv2020

Gradient-Based Training and Pruning of Radial Basis Function Networks with an Application in Materials Physics

Many applications, especially in physics and other sciences, call for easily interpretable and robust machine learning techniques. We propose a fully gradient-based technique for training radial basis function networks with an efficient and scalable open-source implementation. We derive novel closed-form optimization criteria for pruning the models for continuous as well as binary data which arise in a challenging real-world material physics problem. The pruned models are optimized to provide compact and interpretable versions of larger models based on informed assumptions about the data distribution. Visualizations of the pruned models provide insight into the atomic configurations that determine atom-level migration processes in solid matter; these results may inform future research on designing more suitable descriptors for use with machine learning algorithms.

cs.LG↗

arXiv2020

Data sets and trained neural networks for Cu migration barriers

Kinetic Monte Carlo (KMC) is an efficient method for studying diffusion. A limiting factor to the accuracy of KMC is the number of different migration events allowed in the simulation. Each event requires its own migration energy barrier. The calculation of these barriers may be unfeasibly expensive. In this article we present a data set of migration barriers on for nearest-neighbour jumps on the Cu surfaces, calculated with the nudged elastic band (NEB) method and the tethering force approach. We used the data to train artifcial neural networks (ANN) in order to predict the migration barriers for arbitrary nearest-neighbour Cu jumps. The trained ANNs are also included in the article. The data is hosted by the CSC IDA storage service.

physics.comp-ph↗

arXiv2020

Application of artificial neural networks for rigid lattice kinetic Monte Carlo studies of Cu surface diffusion

Kinetic Monte Carlo (KMC) is a powerful method for simulation of diffusion processes in various systems. The accuracy of the method, however, relies on the extent of details used for the parameterization of the model. Migration barriers are often used to describe diffusion on atomic scale, but the full set of these barriers may become easily unmanageable in materials with increased chemical complexity or a large number of defects. This work is a feasibility study for applying a machine learning approach for Cu surface diffusion. We train an artificial neural network on a subset of the large set of $2^{26}$ barriers needed to correctly describe the surface diffusion in Cu. Our KMC simulations using the obtained barrier predictor show sufficient accuracy in modelling processes on the low-index surfaces and display the correct thermodynamical stability of these surfaces.

physics.comp-ph↗