AIQM2 Geometry Optimization with ORCA Opt, SCAN, and Sella TS search

I recently uploaded a set of utilities that combine the AIQM2 machine-learning potential with established quantum chemistry tools. The aim is to simplify geometry refinement and transition-state location in complex systems. Oficial implementations are available in MLatom (ASE, geomeTRIC and Gaussian), but as for the date I am writting this, the oficial implementations lacked the possibility of adding positional and distance restraints during TS search and SCAN. Also, I like the TS search algorithm from Sella, and the ability of using the analytical hessian for the TS search. (I tried to implement this on ORCA, but I failed to use the computed hessian for the TS search)

• ORCA Opt and SCAN minimization – a workflow that drives geometry relaxation by coupling AIQM2 calculator with ORCA’s Opt and SCAN algorithms. Implementation on GitHub
• Sella transition state search – harnesses the Sella optimizer for efficient saddle-point searches using AIQM2 energies, gradients and hessian. Implementation on GitHub

SCAN example

Below is a SCAN-driven optimization example. The plot shows the energy evolution over optimization cycles.

Interactive trajectory

Use the slider below to browse individual frames from the SCAN trajectory.