GaussianProcessNEBParameters

class GaussianProcessNEBParameters(kernel=None, length_scale=None, magnitude=None, noise_variance=None, max_displacement=None, convergence_factor=None, max_iterations=None, training_interval=None, trajectory_filename=None, history=None, min_bond_fuzz_factor=None)

Class that stores all the parameters associated with optimizing a NudgedElasticBand using a Gaussian process regression surrogate model.

Parameters:

• kernel (InverseAtomDistance | RadialBasisFunction | Matern52) – The kernel to use in the Gaussian process regression.
  Default: Matern52.

• length_scale (PhysicalQuantity of type length) – The kernel length scale hyperparameter.
  Default: 1.0 Angstrom.

• magnitude (PhysicalQuantity of type energy) – The covariance function’s magnitude hyperparameter.
  Default: 1.0 eV.

• noise_variance (float) – The Gaussian process model’s noise variance. Usually a small value to avoid numerical problems.
  Default: 1e-10.

• max_displacement (PhysicalQuantity of type length) – The maximum displacement of the surrogate path from the last known path before the next reference points are calculated. If set to Automatic, the value is calculated as half of the total length of the last known path.
  Default: Automatic.

• convergence_factor (float) – The convergence factor used to determine if the surrogate optimization has converged. This is relative to the max_forces criterion of the true path. It is recommended to choose this value to be smaller than 1.0.
  Default: 0.5.

• max_iterations (int) – The maximum number of iterations of the surrogate optimization.
  Default: 20.

• training_interval (int) – The number of iterations between optimizing the Gaussian Process model hyperparameters. A value of 0 means that the hyperparameters are never optimized.
  Default: 0.

• trajectory_filename (str) – The filename to save the trajectory of the surrogate optimization. None means that the trajectory is not saved.
  Default: None.

• history (non-negative int) – The number of most recent configurations to use in training and fitting the model. If 0, all configurations are used. The smaller the number, the faster the training and fitting, but the less accurate the model.
  Default: 0

• min_bond_fuzz_factor (float) – The factor by which the covalent radii are multiplied to determine if two atoms are too close to each other. This is used to determine if a configuration is unphysical. A higher value corresponds to a more strict criterion and more configurations classified as unphysical, while a lower value corresponds to a more relaxed criterion. If the factor is 0, the check for unphysical structures is not done.
  Default: 0.7

nlinfo()

Returns:

The nlinfo.

Return type:

dict

uniqueString()

Return a unique string representing the state of the object.