Graph neural network framework for energy mapping of hybrid monte-carlo molecular dynamics simulations of Medium Entropy Alloys (2411.13670v1)

Abstract: Machine learning (ML) methods have drawn significant interest in material design and discovery. Graph neural networks (GNNs), in particular, have demonstrated strong potential for predicting material properties. The present study proposes a graph-based representation for modeling medium-entropy alloys (MEAs). Hybrid Monte-Carlo molecular dynamics (MC/MD) simulations are employed to achieve thermally stable structures across various annealing temperatures in an MEA. These simulations generate dump files and potential energy labels, which are used to construct graph representations of the atomic configurations. Edges are created between each atom and its 12 nearest neighbors without incorporating explicit edge features. These graphs then serve as input for a Graph Convolutional Neural Network (GCNN) based ML model to predict the system's potential energy. The GCNN architecture effectively captures the local environment and chemical ordering within the MEA structure. The GCNN-based ML model demonstrates strong performance in predicting potential energy at different steps, showing satisfactory results on both the training data and unseen configurations. Our approach presents a graph-based modeling framework for MEAs and high-entropy alloys (HEAs), which effectively captures the local chemical order (LCO) within the alloy structure. This allows us to predict key material properties influenced by LCO in both MEAs and HEAs, providing deeper insights into how atomic-scale arrangements affect the properties of these alloys.

• The paper introduces a novel GCNN framework that accurately predicts potential energy in MEAs using hybrid MC/MD simulations.
• It employs graph-based representations of NiCoCr alloys, achieving R² values between 0.93 and 0.98 with MAPE as low as 0.0184.
• The study demonstrates strong cross-temperature generalization while highlighting challenges in high-temperature extrapolation due to sparse data.

Overview of Graph Neural Network Framework for Energy Mapping in Medium Entropy Alloys

The paper presents a paper on the application of graph neural networks (GNNs) in energy mapping of medium-entropy alloys (MEAs) using hybrid Monte Carlo/Molecular Dynamics (MC/MD) simulations. It highlights the use of ML methods, specifically graph convolutional neural networks (GCNNs), for predicting potential energy in MEAs, which are of interest due to their unique mechanical properties.

Methodological Framework

The research introduces a computational framework wherein hybrid MC/MD simulations are employed to derive thermally stable structures of a NiCoCr MEA. These simulations yield atomic configurations at various annealing temperatures, translated into graph representations. Each atom forms a node connected to its 12 nearest neighbors without explicit edge feature encoding. The nodes include feature matrices encompassing atom types and velocities.

The constructed graphs serve as inputs to a GCNN architecture. This architecture facilitates robust local chemical ordering (LCO) recognition within MEAs and high-entropy alloys (HEAs), permitting the prediction of potential energy across different configurations.

Numerical Analysis and Findings

The performance of the GCNN model is evaluated through three case studies:

1. Temperature-Specific Model Training: Separate models were trained on datasets from specific annealing temperatures (450K, 650K, 950K). The results demonstrated high prediction accuracy with R² values ranging from 0.93 to 0.98 and Mean Absolute Percentage Error (MAPE) as low as 0.0184.
2. Cross-Temperature Generalization: A single model was trained on combined datasets of multiple temperatures (450K, 650K, 850K) and exhibited R² values close to 0.97, underscoring its ability to generalize across varying temperature conditions.
3. Extrapolation to Unseen Configurations: The model, trained on select temperatures (350K, 650K, 850K, 1150K), was evaluated for its predictive capacity on unseen annealing setups. While strong performance was noted for lower temperatures, challenges emerged in high-temperature scenarios due to sparse data points at higher energy levels.

Implications and Future Directions

The paper underscores the potential of GNNs to enhance prediction efficacy in complex material systems like MEAs by capturing atomic interactions and LCO. The presented model accurately reflects potential energy variations linked to LCO, providing a pathway for adopting graph-based methodologies in material property prediction.

Future advancements might include the integration of more sophisticated node and edge features, enabling finer-resolution predictions. Additionally, incorporating diverse alloy systems and further optimizing model architectures could enhance predictive performance over broader temperature ranges and configurations.

In conclusion, the research provides vital insights into the role of GNNs in materials science, particularly within the context of MEAs and HEAs, setting the stage for further exploration of graph-centric approaches in the field.