Alloy nanoparticle stability prediction

# Nanoparticle-stability-prediction<br>The training dataset, prediction dataset and code to reproduce the results of nanoparticle stability prediction<br><br>1. Training procedure for nanoparticle stability prediction was followed the instruction that described in detail in the paper.<br> "The MLIP package: moment tensor potentials with MPI and active learning". DOI 10.1088/2632-2153/abc9fe.<br> After training process, trained MTP_stability can be used to predict energy for nanoparticle as described in detail in the MLIP package. <br><br>2. The generation of bulk structure was performed based on the cluster approach to statistical mechanics (CASM) package.<br> "CASM — A softwCASM — A software package for first-principles based study of multicomponent crystalline solidsare package for first-principles based study of multicomponent crystalline solids". DOI 10.1016/j.commatsci.2022.111897<br><br>3. The command directory include the code to generate nanoparticle (NP) with diverve morphology and diameter.<br> - Gen_Solid_Solution_NP: generate the ordered solid solution nanoparticle with desired diameter from bulk structure. <br> - Gen_Core_Shell_NP: generate the core-shell nanoparticle by keeping the same the number of composition and atoms in solid solution nanoparticle<br> - Gen_Disordered_NP: generate the disordered nanoparticle.<br> - Convert_cif_cfg: converting the cif to cfg for further prediction using MLIP package.<br> - Convexhull_plot_NP: ploting convexhull based on the total energy prediction from MLIP package.<br> - Gen_surface_train: generate surface structure to construct necessary training dataset for MLIP package. <br><br>4. The validation data directory include the data to reproduce the parity plot for both force and energy.<br> - Before DFT optimization: include .cif files of structures of both core shell and solid solution NP in validation dataset and INCAR file for VASP full optimization. <br> - After DFT optimization: include .cif files of full optimization structures.<br> - Prediction results: include .cfg files with energy and force predicted from MLIP package of structures in Before DFT optimization directory. Trained stability potential (MTP_stability) include the trained potential that can be used directly to predict energy of unrelaxed NP.<br> - Energy_Force_validation: CS_F and SS_F directories include the force for each NP in validation dataset. E_valid.csv lists the energy/ atom (eV/atom) predicted from MLIP package and DFT full optimization of each NP in validation dataset. F_valid.csv lists the force (eV/Angstrong) predicted from MLIP package and DFT full optimization of each NP in validation dataset.<br><br>5. Hyperparameter optimization directory include the data to reproduce the hyperparameter optimization selection.<br> - lev_max (potential compolexity): the .mtp is the potential used to predict energy and force of corresponding lev_max. The prediction results include the .cfg file (comprising energy and force predicted by MLIP package) and corresponding force (.csv file).<br> - r cutoff (cutoff radius): the .mtp is the potential used to predict energy and force of corresponding r_cutoff. The prediction results include the .cfg file (comprising energy and force predicted by MLIP package) and corresponding force (.csv file).<br> - weight_ratio: the .mtp is the potential used to predict energy and force of corresponding energy-to-force-weight. The prediction results include the .cfg file (comprising energy and force predicted by MLIP package) and corresponding force (.csv file).<br><br>6. The prediction directory include the prediction data (test dataset) for nanoparticle stability prediction. The predictions were performed for PtNi, PtAu, PdAu, PtCu (FCC-FCC), PtCo (FCC-HCP) and PtFe (FCC-BCC). The generation of training dataset and prediction dataset for the nanopaticle in same class is similar. Therefore, the directory show the representative nanoparticle for each class upto nanoparticle diameter of 3nm. The training dataset, prediction dataset for the other nanoparticle with different nanoparticle diameter can be generated using the code provied in command directory. <br> - PtNi: include Prediction dataset, Prediction results, Training dataset and Trained potential.<br> - Prediction dataset: .cif file of nanoparticle that need to predict energy.<br> - Prediction results: .cfg file of nanoparticle that predicted by MLIP package.<br> - Training dataset: bulk and surface structure used to trained MLIP.<br> - PtCo, PtFe: the directory structure is the same as PtNi.<br><br>