Modeling Short-Range and Three-Membered Ring Structures in Lithium Borosilicate Glasses Using a Machine-Learning Potential

Abstract

Lithium borosilicate (LBS) glass is a prototypical lithium-ion conducting oxide glass available for an all-solid-state buttery. Nevertheless, the atomistic modeling of LBS glass using ab initio (AIMD) and classical molecular dynamics (CMD) simulations has critical limitations due to computational cost and inaccuracy in reproducing the glass microstructures, respectively. To overcome these difficulties, a machine-learning potential (MLP) was examined in this work for modeling LBS glasses using DeepMD. The glass structures obtained by this MLP possessed 4-fold coordinated boron (4B) confirmed well with the experimental data and abundance of three-membered rings. The models were energetically more stable compared with those constructed with a functional force field even though both the models included reasonable 4B. The results confirmed MLP to be superior to model the boron-containing glasses and address the inherent shortcomings of the AIMD and CMD. This study also discusses some limitations of MLP for modeling glasses.

Accurate and Transferable Machine Learning Potential for Molecular Dynamics Simulation of Sodium Silicate Glasses

“…The predicted atomic forces also give satisfactory results with an MAE of 0.154 eV/Å. These values are comparable with MAEs obtained in other works on MLP trained on specific glass compositions. ,− No particular trend was observed either with the composition changes or with different elements present in the systems, suggesting that the proposed MLP describes with comparable accuracy glasses within the full compositional range of glass formability. Albeit the virial was not included in the training, its prediction is accurate with MAE = 3.254 eV.…”

“…Machine learning-based potentials (MLP) have attracted more and more attention in recent years as an alternative approach to simulate atomic interactions in glasses. − They have been already applied in other fields such as battery materials, catalysis, and drug design . The basic idea is to first train the chosen ML algorithm using accurate ab initio data, typically from DFT, and then to reproduce the potential energy surface at a much lower computational cost, making larger systems and longer timescales accessible without losing accuracy.…”

Boron coordination and three‐membered ring formation in sodium borate glasses: a machine‐learning molecular dynamics study

“…Therefore, the less amount of 3‐rings generated by the MLP‐MD simulations compared to the KMG model at low Na 2 O content is not due to the modeling procedures. A previously developed MLP for modeling lithium borate glasses 26 also underestimated boroxol rings in borate glass. One possible explanation for this is the significant challenge in obtaining borate glass models with an adequate number of boroxol rings using any molecular modeling technique, such as CMD and AIMD simulations, which hinders to obtain appropriate training data for constructing MLP.…”

“…The other challenge facing CMD simulations is the formation of the 3‐rings. Indeed, it has been demonstrated that the force fields reproduce few 3‐rings, even if boron atoms appropriately bind to oxygen atoms 18,26 . Takada et al.…”

“…Recently, machine‐learning potentials (MLPs), which estimate the energy and force of each atom by machine‐learning algorithms using structural descriptors, have garnered significant attention as a novel method for obtaining accurate and efficient force fields. For instance, MLPs have already been applied in modeling and investigating various materials, such as inorganic electrolyte, 29 copper surface, 30 reaction between H + CH 3 OH, 31 alanine dipeptide, 32 and oxide glasses 26,33–35 . It has been demonstrated that MLPs exhibit comparable accuracy with AIMD simulations when sufficient DFT datasets were appropriately utilized for training these MLPs.…”