OptimizationManager

class optking.optimize.OptimizationManager(molsys: Molsys, history_object: History, params: OptParams, computer: ComputeWrapper)

Bases: OptimizationInterface

Recommended use of Optking’s Optimization Algorithms is to create this class and then loop over take_step. OptimizationFactory will either return the appropriate OptimizationAlgorithm or return itself if management of multiple algorithms is required (linesearching). Currently only 1 linesearch method is implemented so changing linesearch_method to anything that does not evaluate to None will turn linesearch on.

This class’ primary purpose is to abstract the interface for an OptimizationAlgorithm, Linesearch, or IRC so no special handling is needed for an IRC optimization as opposed to a NR optimization.

# TODO add dynamic_level management here

Methods Summary

alg_error_handler(error)	consumes an AlgError.
check_maxiter()	Check iterations < geom_maxiter.
clear()	Reset history (inculding all steps) and molecule
converged(E, fq, dq[, step_number, str_mode])	Test whether the optimization has finished.
finish([error])
from_dict(d, molsys, history, params, computer)	Reload attributes from the provided dictionary.
get_hessian_protocol()	Determine action to take for how to compute a hessian.
opt_error_handler(error)	OptError indicates an unrecoverable error.
start_step(H)	Initialize coordinates and perform any coordinate transformations of gradients and hessians.
take_step([fq, H, energy, return_str])	Take whatever step (normal, linesearch, IRC, constrained IRC) is next.
to_dict()	Convert attributes to serializable form.
unknown_error_handler(error)	Unknown errors are not recoverable error.
update_requirements()	Get the current requirements for the next step.

Methods Documentation

alg_error_handler(error)

consumes an AlgError. Takes appropriate action

check_maxiter()

Check iterations < geom_maxiter. For IRC’s check total_steps_taken.

clear()

Reset history (inculding all steps) and molecule

converged(E, fq, dq, step_number=None, str_mode=None)

Test whether the optimization has finished. An optimization can only be declared converged If a gradient has been provided (linesearching cannot terminate an optimization)

finish(error=None)

classmethod from_dict(d, molsys, history, params, computer)

Reload attributes from the provided dictionary. Create all necessary classes.

To prevent duplication, OptHelper handles converting the molsys, history, params, and computer to/from dict

get_hessian_protocol()

Determine action to take for how to compute a hessian. Handles alternate IRC behavior

Returns

str

Return type

one of (‘compute’, ‘update’, ‘guess’, ‘unneeded’)

opt_error_handler(error)

OptError indicates an unrecoverable error. Print information and trigger cleanup.

start_step(H: ndarray)

Initialize coordinates and perform any coordinate transformations of gradients and hessians.

Returns

• H (np.ndarray) – 2D. Hessian in appropriate coordinates

• f_q (np.ndarray) – 1D. forces in appropriate coordinates

• E (float) – energy

take_step(fq=None, H=None, energy=None, return_str=False, **kwargs)

Take whatever step (normal, linesearch, IRC, constrained IRC) is next.

fq: Union[np.ndarray, None]

forces

H: Union[np.ndarray, None]

hessian

energy: Union[np.ndarray, None] return_str: bool

if True return string with information about step (information is logged regardless)

to_dict()

Convert attributes to serializable form.

unknown_error_handler(error)

Unknown errors are not recoverable error. Print information and trigger cleanup.

update_requirements()

Get the current requirements for the next step.

Notes

If linesearching requirements can change. Always safe to provide a gradient regardless.