Masgent: Materials Simulation Agent

Masgent is a materials simulation AI agent that streamlines DFT workflows and analysis, fast machine-learning-potential (MLP) simulations, and lightweight ML modeling for materials science. With automated tools for structure handling, VASP input generation, workflow preparation & analysis, and rapid property prediction, Masgent simplifies complex simulation tasks and boosts productivity for both researchers and students.

• Basic Usage:
  e.g., prepare POSCAR files and generate SQS structures

  

• AI Agent:
  e.g., prepare POSCAR files and run EOS calculations simply by chatting

  

1. Density Functional Theory (DFT) Simulations

• 1.1 Structure Preparation & Manipulation

  • 1.1.1 Generate POSCAR from chemical formula
  • 1.1.2 Convert POSCAR coordinates (Direct <-> Cartesian)
  • 1.1.3 Convert structure file formats (CIF, POSCAR, XYZ)
  • 1.1.4 Generate structures with defects (Vacancies, Substitutions, Interstitials)
  • 1.1.5 Generate supercells
  • 1.1.6 Generate Special Quasirandom Structures (SQS)
  • 1.1.7 Generate surface slabs
  • 1.1.8 Generate interface structures

• 1.2 VASP Input File Preparation

  • 1.2.1 Prepare full VASP input files (INCAR, KPOINTS, POTCAR, POSCAR)
  • 1.2.2 Generate INCAR templates (relaxation, static, etc.)
  • 1.2.3 Generate KPOINTS with specified accuracy
  • 1.2.4 Generate HPC job submission script

• 1.3 Standard VASP Workflows Preparation

  • 1.3.1 Convergence test (ENCUT, KPOINTS)
  • 1.3.2 Equation of State (EOS)
  • 1.3.3 Elastic constants calculations
  • 1.3.4 Ab-initio Molecular Dynamics (AIMD)
  • 1.3.5 Nudged Elastic Band (NEB) calculations

• 1.4 (Planned) Workflow Output Analysis

  • 1.4.1 (Planned) Convergence test analysis
  • 1.4.2 (Planned) Equation of State (EOS) analysis
  • 1.4.3 (Planned) Elastic constants analysis
  • 1.4.4 (Planned) Ab-initio Molecular Dynamics (AIMD) analysis
  • 1.4.5 (Planned) Nudged Elastic Band (NEB) analysis

2. Fast Simulations Using Machine Learning Potentials (MLPs)

• Supported MLPs:
  • 2.1 SevenNet
  • 2.2 CHGNet
  • 2.3 Orb-v3
  • 2.4 MatSim
• Implemented Simulations for all MLPs:
  • Single Point Energy Calculation
  • Equation of State (EOS) Calculation
  • Elastic Constants Calculation
  • Molecular Dynamics Simulation (NVT)

3. Simple Machine Learning for Materials Science

• 3.1 Data Preparation & Feature Analysis
  • 3.1.1 Feature analysis and visualization
  • 3.1.2 Dimensionality reduction (if too many features)
  • 3.1.3 Data augmentation (if limited data)
• 3.2 Model Design & Hyperparameter Tuning
• 3.3 Model Training & Evaluation

1. Requirements:
   • Python >= 3.11, < 3.14
2. Optional:
   • OpenAI API key for AI agent features: platform.openai.com
   • Materials Project API key for MP structure access: materialsproject.org
3. Install Masgent:

   pip install -U masgent

4. Setup POTCAR path for Pymatgen, see instructions: https://pymatgen.org/installation.html#potcar-setup

• After installation, simply run:

  masgent

• You'll guided by an interactive menu and can invoke the AI agent anytime. Ask anything in AI chat, for example:

  > Generate a POSCAR file for NaCl.
  > Prepare VASP input files for a graphene structure.
  > Add defects to a silicon crystal POSCAR.
  > ...

Found a bug? Have a feature request?
Please open an issue here: https://github.com/aguang5241/masgent/issues

If you use Masgent in your research, please cite the following reference:

@article{
  title={Masgent: A Materials Simulation Agent},
  journal={},
  volume={},
  pages={},
  year={},
  issn={},
  doi={},
}

Masgent builds on the open-source materials ecosystem, including ASE, Pymatgen, Icet, and modern Machine Learning Potentials. We thank the developers of these tools for making advanced materials simulation possible.