n2p2

Official Resources

• Homepage: https://compphysvienna.github.io/n2p2/
• Documentation: https://compphysvienna.github.io/n2p2/
• Source Repository: https://github.com/CompPhysVienna/n2p2
• License: GPL v3

Overview

n2p2 is a C++ library and set of tools for training and using neural network potentials (Behler-Parrinello type). It allows for the training of high-dimensional neural network potentials (HDNNP) and provides a highly efficient interface for LAMMPS. It is known for its performance and parallel scaling.

Scientific domain: Machine learning potentials, molecular dynamics
Target user community: HPC users, MD researchers

Capabilities (CRITICAL)

• Training: Kalman filter training, gradient descent.
• Descriptors: Symmetry functions.
• LAMMPS: Native, highly optimized pair style (pair_style hdnnp).
• Pruning: Tools to optimize network structure.
• Standalone: Can be used as a library.

Sources: n2p2 website

Inputs & Outputs

• Input formats: input.data (RuNNer format)
• Output data types: Potential weights

Interfaces & Ecosystem

• LAMMPS: Primary simulation engine.
• RuNNer: Compatible file formats.

Workflow and Usage

1. Generate input.data (structures/energies/forces).
2. Configure input.nn.
3. Train: nnp-train.
4. Run MD in LAMMPS.

Performance Characteristics

• C++ implementation ensures high performance.
• MPI/OpenMP parallelization.

Application Areas

• Water (aqueous solutions)
• Surfaces and interfaces
• Phase transformations

Community and Support

• Developed by CompPhysVienna (University of Vienna)
• Active development

Verification & Sources

Primary sources:

1. Homepage: https://compphysvienna.github.io/n2p2/
2. GitHub: https://github.com/CompPhysVienna/n2p2

Confidence: VERIFIED

Verification status: ✅ VERIFIED

• Website: ACTIVE
• Documentation: COMPREHENSIVE
• Source: OPEN (GitHub)
• Development: ACTIVE
• Applications: HDNNP, LAMMPS