Skip to content

MACE Force Fields — Web Interface

Run DFT-accuracy atomistic simulations entirely in your browser. No installation. No command line. No barriers.

Live Demo: mace-lake.vercel.app


What is this?

A browser-based interface for MACE (Multi-Atomic Cluster Expansion) machine learning interatomic potentials. Upload a crystal or molecular structure, pick a model, and get publication-quality energies, forces, and trajectories in seconds.

Why it exists and the story behind it

I am a freshman college student majoring in computer science. I was lucky to be assigned a research project by Oak Ridge National Laboratory, focusing on the use of MACE.

MACE is a powerful tool for quantum chemistry, but I initially had no idea how to use it. The detailed documentation and Google Colab are not user-friendly, at least for me.

So I decided to act.

I am not an expert in quantum chemistry or machine-learning interatomic potentials, but I'm a front-end developer. I can visualize complex, abstract tools and simplify the command-line interface into a modern software tool accessible to researchers and even students in my research team.

Rowan Scientific, a molecular design and simulation platform for scientists, truly inspires me. They share my vision of making complex scientific tools more accessible to researchers. I would like to include their quote:

"We're starting Rowan because we think that scientific software shouldn't be hard to use."

It is inspiring when I realize they had the exact same vision as I did!

Especially with the age of AI, modern software using AI agents becomes increasingly easy to build, and the quality is amazing. Even someone without a PhD in quantum chemistry can build a research-grade web interface for others to use.

Machine learning interatomic potentials like MACE have reached DFT-level accuracy while running orders of magnitude faster. But using them still requires Python scripting, command-line fluency, and environment setup that shuts out many researchers — especially those with accessibility needs, those in under-resourced labs, or students encountering computational chemistry for the first time.

This project removes that barrier.

Key capabilities

Feature Details
Structure input Drag-and-drop .xyz, .cif, .poscar, .pdb or pick from a built-in catalog of 14 benchmark structures
Foundation models MACE-MP-0 (89 elements, materials) and MACE-OFF (organic molecules) in small/medium/large
Custom models Upload your own .model file and compare against foundation models
Calculations Single-point energy & forces, geometry optimization (BFGS), molecular dynamics (NVE/NVT/NPT)
Visualization Dual 3D viewers (3Dmol.js + WEAS), interactive Plotly charts, MD trajectory player
Sharing Every result becomes a permanent shareable URL via MACE Link
Benchmark Batch-evaluate multiple models across multiple structures with leaderboard and export
Accessibility Keyboard navigation, ARIA labels, colorblind-safe palette (Paul Tol)

Built by Zicheng Zhao · Northeastern University