Geometry Optimization
The Geopt class allows to perform a geometry optimization to
relax the forces on a given structure, using a machine
learning model.
- class mlcalcdriver.workflows.geopt.Geopt(posinp, calculator, forcemax=0.01, step_size=0.002, max_iter=500)[source]
Bases:
objectThis class allows to relax the input geometry of a given system in order to find the structure that minimizes the forces. The final result obtained depends on the trained machine learning model used.
- Parameters
posinp (mybigdft.Posinp) – Starting configuration to relax
calculator (Calculator) – mlcalcdriver.Calculator instance that will be used in the created Job to evaluate properties.
forcemax (float) – Stopping criterion on the forces (in eV/Angstrom). Default is 0.01.
step_size (float) – Step size for each relaxation step. Default is 0.003 Angstrom<sup>2</sup>/eV.
max_iter (int) – Maximum number of iterations. Default is 500.
- property calculator
- Returns
The Calculator object to use for the Jobs necessary to perform the geometry optimisation.
- Return type
- property final_posinp
- Returns
Final posinp of the geometry optimization or None if the the optimization has not been completed
- Return type
Posinp or None
- property forcemax
- Returns
Stopping criterion on the forces (in eV/Angstrom)
- Return type
float
- property step_size
- Returns
Step size for each relaxation step
- Return type
float
- property max_iter
- Returns
Maximum number of iterations
- Return type
int
- run(batch_size=128, recenter=False, verbose=0)[source]
- Parameters
batch_size (int) – Size of the mini-batches used in predictions. Default is 128.
recenter (bool) – If True, the structure is recentered on its centroid after the relaxation. Default is False.
verbose (int) – Controls the verbosity of the output. If 0 (Default), no written output. If 1, a message will indicate if the optimization was succesful or not and the remaining forces. If 2 or more, each iteration will provide an output.