AMS driver’s tasks and properties

MLPotential is an engine used by the AMS driver. While the specific options for the MLPotential engine are described in this manual, the definition of the system, the selection of the task and certain (potential-energy-surface-related) properties are documented in the AMS driver’s manual.

On this page, you will find useful links to the relevant sections of the AMS driver’s Manual.

Geometry, System definition

The definition of the system, i.e. the atom types and atomic coordinates (and optionally, the lattice vectors and atomic masses for isotopes) are part of the AMS driver input. See the System definition section of the AMS manual.

Note

MLPotential models that support 3D-periodic systems can also be used for 1D-periodic chains and 2D-periodic slabs.

Tasks: exploring the PES

The job of the AMS driver is to handle all changes in the simulated system’s geometry, e.g. during a geometry optimization or molecular dynamics calculation, using energy and forces calculated by the engine.

These are the tasks available in the AMS driver:

• GCMC (Grand Canonical Monte Carlo)

• Geometry Optimization

• IRC (Intrinsic Reaction Coordinate)

• Molecular Dynamics

• NEB (Nudged Elastic Band)

• PESScan (Potential Energy Surface Scan, including linear transit)

• Single Point

• Transition State Search

• Vibrational Analysis

Properties in the AMS driver

The following properties can be requested from the MLPotential engine in the AMS driver’s input:

• Elastic tensor

• Hessian

• Nuclear gradients (forces)

• Normal modes

• PES point character

• Phonons

• Stress tensor

• Thermodynamic properties

In addition to the above properties, for supported machine learning models, the MLPotential engine will also automatically calculate the condensed Fukui functions for aperiodic systems. In order to do this, the model must accept the total system charge and predict atomic charges. The results include \(f^+\), \(f^-\), \(f^0\), the dual descriptor, the vertical electron affinity, and the vertical ionization energy (see Fukui Function for definitions). The Python example Predicting Reactivity via Condensed Fukui Functions demonstrates how to calculate and analyze these reactivity descriptors with AIMNet2.