Thermodynamic modeling of complex solid solutions in the Lu-H-N system via graph neural network accelerated Monte Carlo simulations

Pinwen Guan,Catalin Spataru,Vitalie Stavila,Reese Jones,Peter Sharma,Matthew Witman
DOI: https://doi.org/10.26434/chemrxiv-2024-6g37p
2024-08-19
Abstract:Metal hydrides are important across diverse applications such as hydrogen storage, batteries, gas sensors, nuclear reactions and high-temperature superconductivity. Previous computational studies of metal hydrides under extreme pressures, e.g., O(10^2) GPa, usually treat them as stoichiometric compounds without considering interstitial lattice disorder. As pressures become more moderate in the O(10^0) GPa and below range, hydrogen disorder at interstitial lattice sites becomes prominent, e.g. in the N-doped Lu hydride that was recently claimed superconducting near 1 GPa. Further adding compositional complexity from alloying and/or multi-element interstitial occupation makes elucidating pressure- and temperature-dependent observables intractable by first-principles calculations alone. We therefore propose a lattice graph neural network surrogate modeling approach to predict configuration- and pressure-dependent equation-of-state properties. Their efficiency permits Monte Carlo simulations to calculate Gibbs energies and pressure-dependent phase diagrams, thereby revealing insights into the synthesis conditions required for achieving desired phase equilibria. We demonstrate this concept for the compositionally complex cubic Lu(H, N,Va)3 system where three constituents (hydrogen, nitrogen and vacancy) have disordered multi-element interstitial occupancies and insights into pressure-dependent phase equilibria are critically needed, e.g., N-doping levels can significantly lower dehydrogenation temperatures and provide a new strategy to optimize hydrogen-storage alloys. This work can improve the thermodynamic understanding of the Lu-H-N system and help rational synthesis of N-doped Lu hydrides, but more generally demonstrates an efficient approach to model pressure-dependent thermodynamics of multi-component solid solutions.
Chemistry
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve This paper aims to address the thermodynamic modeling of complex solid solutions under high-pressure conditions (specifically at or below 100 GPa). Specifically, the study focuses on the cubic Lu(H, N, Va)₃ solid solution in the Lu-H-N system. The paper mainly focuses on the following aspects: 1. **Disorder of Hydrogen and Nitrogen at Interstitial Sites**: - At moderate pressures (at or below 100 GPa), the disorder of hydrogen at interstitial sites becomes significant, especially in the recently reported N-doped Lu hydrides, where this disorder has an important impact on superconducting properties. - When alloying or multiple elements occupy interstitial sites, this disorder makes it very difficult to predict pressure and temperature-dependent observables through first-principles calculations. 2. **Prediction of Thermodynamic Behavior of Multi-Component Solid Solutions**: - Traditional first-principles calculation methods are inefficient in handling the complexity of multi-component solid solutions, especially under high-pressure conditions. - The paper proposes a surrogate model method based on Graph Neural Networks (GNN), combined with Monte Carlo simulations to predict configuration and pressure-dependent equation of state properties, thereby calculating Gibbs free energy and pressure-dependent phase diagrams. 3. **Optimization of Hydrogen Storage Alloy Design**: - Through thermodynamic modeling, a new doping strategy is proposed to optimize the performance of hydrogen storage alloys. Specifically, N doping can significantly reduce the dehydrogenation temperature, providing a new approach to optimizing hydrogen storage materials. ### Solutions 1. **Graph Neural Network (GNN) Surrogate Model**: - Using GNN to directly predict high computational cost first-principles (DFT) relaxed properties from idealized lattice structure inputs, such as zero-pressure formation energy (Ef,0) and specific volume (ν0). - By training the GNN model, sufficiently accurate predictions can be obtained with a reasonable amount of data (about 1000 DFT relaxations), enabling efficient phase space sampling. 2. **Monte Carlo Simulations**: - Utilizing the GNN surrogate model to perform large-scale Monte Carlo simulations to estimate free energy dependent on composition, pressure, and temperature, thereby calculating different types of phase equilibria and thermodynamic properties. 3. **Application Example**: - Using the cubic Lu(H, N, Va)₃ system as an example, the effectiveness of this method in predicting the PCT (pressure-composition-temperature) behavior of high-pressure, multi-element interstitial hydrides is demonstrated. - By calculating Gibbs free energy, phase diagrams under different pressures and temperatures are predicted, and the impact of N doping levels on the phase diagram is analyzed. ### Conclusion This study provides a general method for efficiently modeling the thermodynamic behavior of multi-component solid solutions, particularly suitable for complex materials such as high-entropy alloys, high-entropy ceramics, and mixed anion compounds. Additionally, this method can be combined with semi-empirical thermodynamic modeling frameworks (such as CALPHAD) and their deep learning counterparts, further extending to applications in low-dimensional systems (such as surfaces), such as high-entropy catalysts. By introducing other machine learning techniques (such as active learning) to label the data required to train the GNN model, this method can be further accelerated.