Documentation

Overview

SimpleAtom is a free, browser-based interface to the MACE family of machine-learning interatomic potentials.

What is SimpleAtom?

SimpleAtom lets you run quantum-accurate atomistic calculations — single-point energies and forces, geometry optimizations, and molecular dynamics — directly in your web browser. You upload or sketch a structure, choose a pre-trained MACE model, and read back energies, per-atom forces, and trajectories with interactive 3D visualization and publication-quality charts.

There is nothing to install: no Python environment, no command line, and no high-performance-computing (HPC) allocation. Calculations are executed server-side with ASE (the Atomic Simulation Environment) and mace-torch, and returned to the browser as structured results.

What are machine-learning interatomic potentials?

Interatomic potentials predict a system’s potential energy (and its gradient, the forces on each atom) as a function of atomic positions. Classical force fields are fast but approximate; density functional theory (DFT) is accurate but expensive. Machine-learning interatomic potentials (MLIPs) are trained on DFT (or higher-level) reference data to reproduce that accuracy while running orders of magnitude faster.

MACE (Multi-Atomic Cluster Expansion) is a higher-order equivariant message-passing neural network introduced by Batatia et al. (NeurIPS 2022). Its foundation models are pre-trained on large DFT datasets and generalize across chemistries out of the box, which is what makes a zero-configuration web interface possible.

The mission: accessible computational chemistry

MLIPs have reached DFT-level accuracy, yet using them still typically demands Python scripting, command-line fluency, and cluster access — barriers that exclude experimentalists, students, researchers in under-resourced labs, and scientists with accessibility needs. SimpleAtom exists to remove that first barrier and put a good idea one click away from a result.

Why this matters

The interface is built with accessibility as a first principle: keyboard navigation, ARIA semantics, a colorblind-safe data palette (Paul Tol), and a calm, low-contrast visual design.

How it works

The workflow mirrors how computational chemists actually work:

  1. Provide a structure — upload a file (XYZ, CIF, POSCAR, PDB), paste a SMILES string, sketch a molecule, or pick from a benchmark catalog.
  2. Choose a model & calculation — a MACE foundation model (or your own fine-tuned checkpoint), then a single-point, geometry-optimization, or molecular-dynamics run.
  3. Run — the structure is evaluated server-side with ASE + mace-torch.
  4. Explore & share — inspect results in 3D and in charts, export (CSV, JSON, PDF), or publish a permanent shareable link.

Where to next