rapmat 0.3.2

Rapmat - rapid materials discovery using MLIPs and random search

rapmat

Rapid materials discovery TUI tool, based on random crystal generation and machine learning interatomic potentials.

Installation

An Nvidia GPU is highly recommended. Linux is recommended as well, since all backends are currently supported on Linux systems. Conda may be useful.

Linux

Install pytorch<2.10.0 with CUDA support if you have an NVIDIA GPU, otherwise skip this step:

pip install torch==2.9.1 torchvision --index-url https://download.pytorch.org/whl/cu126

Then install rapmat:

# basic install
pip install rapmat

# mattersim support
pip install rapmat[mattersim]

# all calculators at once
pip install rapmat[all-calculators]

Run its TUI:

rapmat

Windows

On Windows, only the UPET and MatterSim (the latter if built with its prerequisites installed) MLIPs are supported.

Install WSL2

One way to overcome the Windows limitations is WSL2 -- check the Nvidia or Ubuntu guides.

Update the linker paths

NequIP compiles its CUDA kernels and needs the CUDA (stub) libraries on the linker path before installation:

export LIBRARY_PATH=/usr/local/cuda/lib64/stubs:$LIBRARY_PATH

Add it to the ~/.bashrc to make it permanent.

Proceed to the linux install guide

Usage

Basic concepts

A study defines the system (e.g. Al-O) you are working on and the calculation settings like calculator, forces convergence criterion or pressure. A run defines a specific formula x [formula units range]: e.g. Al2O3 x 6..8 constituting the unit cell being calculated.

Each run belongs to exactly one study, while a study may have many runs. Runs in one study may overlap, but you can view and perform actions such as deduplication or thickness filtering for only one run at a time.