Release 2026.5.28

Release 2026.5.28

Release 2026.5.28 introduces multiple new features. We highlight (1) a new step-by-step tutorial on using Machine-Learned Forcefields in the platform to perform structural relaxation, phonon calculations, and similar, highlighting Microsoft’s Mattersim as a specific implementation and NVIDIA GPUs as an accelerator; (2) a new Jupyter notebook demonstrating how to use MACE universal forcefield to do structural relaxation inside a web browser with no additional infrastructure needed, and (3) New jupyter notebooks demonstrating the calculations of surface energy and the equation of state.

Features

• Platform Tutorial: Using Python-based MLFFs (MatterSim) workflows, including GPU acceleration
• [Jupyter Notebooks]: MACE in Jupyter/JupyterLite for pre-relaxing structures
• [Jupyter Notebooks] Add NB for Surface Energy
• [Jupyter Notebooks] Add NB for Equation of State
• [Webinar, Mat3ra-2D] Season 2. Episode 2. Relaxation with MLFF + DFT.

Improvements

• Update Surface Energy workflow to reuse build metadata from mat3ra-made

Fixes

• Fix: CLI Jobs not registering in the web interface
• Fix: Reusing identical Python virtualenvs
• Fix: [MD/WA] Save Combinatorial Set materials

Developers

• Iterate Mat3ra-2D manuscript at https://arxiv.org/abs/2603.27886
• Update https://github.com/Exabyte-io/api-examples/ publish name to mat3ra-notebooks-utils

Highlight 1: Python-based MLFFs, MatterSim, GPU-support

Per the video tutorial below:

Also available at https://docs.mat3ra.com/tutorials/ml/run-mlff-python-workflows-mattersim/.

‍

Highlight 2: pre-relax structures with MLFF in JupyterLite/Pyodide

New notebook highlighting the use of MLFF (MACE)

Highlight 3: Surface Energy Workflow updated.

We continue populating the api-example notebooks with example calculations.

Highlight 4: Mat3ra-2D webinar, S2E2.

Structural relaxation with MLFF and DFT.

‍

Try the new functionality online at https://platform.mat3ra.com/

‍