Source Code
MAPLE is developed openly on GitHub. Contributions, bug reports, and feature requests are welcome.
Changelog
Version 2025.3 — September 12, 2025
Major release with AutoNEB support and improved convergence behavior.
- New: AutoNEB method — Automatic nudged elastic band with adaptive image insertion for transition state searches. Reduces manual setup for complex reaction paths.
- Improved convergence — Refined step-size control for L-BFGS and RFO optimizers, leading to more robust convergence on flat potential energy surfaces.
- Growing String Method enhancements — Better interpolation for initial path generation and improved reparameterization.
- Bug fix: Fixed incorrect thermochemistry corrections when using symmetry-reduced Hessians in frequency calculations.
- Bug fix: Resolved memory leak in multi-structure scan calculations that affected long-running jobs.
- Performance: Reduced GPU memory footprint for MACE models by approximately 15% through optimized graph construction.
Version 2025.2 — June 3, 2025
Expanded model support and new PES scanning capabilities.
- New model: MACE-OMol — Added support for the MACE-OMol potential trained on the OMol25 data set for broad organic molecular coverage.
- New model: EGRET — Added support for the EGRET equivariant graph neural network potential.
- N-dimensional PES scans — Extended scan module to support arbitrary N-dimensional relaxed and rigid scans along multiple internal coordinates simultaneously.
- Dimer method improvements — Better initial rotation algorithm and convergence criteria for dimer-based transition state searches.
- Bug fix: Corrected coordinate transformation for systems with linear molecular fragments during optimization.
- Bug fix: Fixed incorrect energy reporting in scan output when using AIMNet2-NSE with D4 dispersion.
Version 2025.1 — March 15, 2025
Initial public release of MAPLE with core computational capabilities.
- Frequency analysis — Normal mode analysis with mass-weighted Hessian diagonalization, RRHO thermochemistry corrections, and IR spectrum generation.
- ML potential support — ANI-2x, ANI-1x, ANI-1ccx, ANI-1xnr, AIMNet2, AIMNet2-NSE, MACE-OFF23 (s/m/l), and UMA.
- Geometry optimization — L-BFGS, RFO, steepest descent, conjugate gradient, DIIS, and bracketing L-BFGS methods.
- Transition state search — NEB, CI-NEB, P-RFO, Dimer, and Growing String methods.
- IRC — Intrinsic reaction coordinate with GS, LQA, HPC, and EulerPC integrators.
- PES scan — 1D relaxed and rigid scans along bonds, angles, and dihedrals.
- Solvation — GBSA implicit solvation and explicit solvent models.
- Constraints — Fix atoms, bonds, angles, and dihedral angles during optimization.
- GPU acceleration — Full CUDA support for all ML potential evaluations.
- D4 dispersion — Optional DFT-D4 dispersion correction for all calculation types.