Studying the Spontaneous Crystallization of Metals under Extreme Conditions Using Classical and Deep Learning Interatomic Potentials for Molecular Dynamics

The spontaneous crystallization of metals in highly unusual conditions is a remarkable occurrence that takes place when systems are far from equilibrium. This intriguing phenomenon holds the potential to pave the way for the advancement of groundbreaking and transformative metastable metals that possess exceptional and unconventional properties. The unique characteristics of spontaneous crystallization present a fascinating opportunity to explore the generation of metastable metals from amorphous metallic phases. This research's underlying idea is that by subjecting these materials to various external factors like temperature and pressure, it is possible to create extreme yet controlled conditions for crystallization. These conditions, unlike other methods, can lead to the development of metastable metals. The research within this dissertation involves an exploration of the spontaneous crystallization process in amorphous metal bulk structures and nanoparticles, along with an examination of how metastable phases form during this crystallization. This investigation employs non-equilibrium molecular dynamics (MD) simulations, utilizing both classical and deep learning-based interatomic potentials. In the case of Nickel (Ni) and Cobalt (Co) bulk structures, the process of creating amorphous metals involves melting and rapid cooling, followed by the introduction of external seeds representing various crystalline phases. This is done under different temperature and pressure conditions using the NPT ensemble, leading to the discovery of self-propagating crystallization waves and the formation of metastable phases. For amorphous Co (a-Co) bulk structure with a face-centered cubic (fcc) seed at T=800 K and P=1 MPa, the crystallization process results in an impressive 97.4% fcc metastable phase, the highest achievement in these simulations. However, in the case of a-Ni bulk structure, the variation of hexagonal close-packed (hcp) metastable phase during crystallization under different temperature and pressure conditions is not consistent, making it challenging to determine the maximum percentage of metastable phase using MD simulations. To address this, Bayesian Optimization (BO) with Gaussian Processes (GPs) regression is utilized as a surrogate model to maximize the hcp-Ni phase fraction and find the values of control variables that lead to this fraction. This BO-guided active learning approach achieves maximum hcp-Ni fractions of 43.38% and 58.25% in the final crystallized phase when fcc and hcp crystallites are used, respectively, as seeds for crystallization from the amorphous phase. In the context of crystallizing nanoparticles with both external seeds (matching the bulk structure) and internal seeds (spherical shape), a-Ni and a-Co nanoparticles yield maximum metastable phase fractions of 20% and 50%, respectively. Furthermore, the crystallization of a-Ni and a-Co nanoparticles takes place under NVT and NVE ensembles without periodic boundary conditions. Deep potential-smooth edition (DeepPot-SE) is employed within the framework of DeepMD-kit to develop deep learning interatomic potentials (DLIPs) for the MD simulations of amorphous ß-tungsten (a-ß-W) and amorphous silicon carbide (a-SiC) bulk structures. The developed DLIP for ß-W predicts a melting temperature of 3560 K, in good agreement with experimental data, demonstrating the proper accuracy of this deep learning potential/force field. Subsequent MD simulations include tasks like melting, rapid cooling, and crystallization, which, when adding a ß-W seed to a-W, result in approximately 95% metastable ß-W phase. Additionally, computations with different temperature and pressure values for ß-W crystallization reveal that independent from pressure, the crystallization wave propagates for temperatures above 1500 K, with increasing temperature causing the self-propagating crystallization wave to accelerate. At temperatures below 1500 K, the crystallization wave progresses very slowly and eventually ceases to propagate. For bulk c-SiC, DLIP is also developed to gain insights into the structure's behavior under extreme conditions. During the melting phase of the amorphization process, heating bulk c-SiC to temperatures exceeding 4200 K leads to dissociation/decomposition, accompanied by Carbon atoms clustering. For temperature equal to 4200 K, thermalization under the NPT ensemble at P=1 bar results in a composition of 58% amorphous and 42% c-SiC phases.