PAO

References: Schuett2018, Berghold2002

Polarized Atomic Orbital Method [Edit on GitHub]

Subsections

• LINE_SEARCH
• MACHINE_LEARNING
• PRINT

Keywords

• CG_INIT_STEPS

• CG_RESET_LIMIT

• CHECK_GRADIENT_TOL

• CHECK_UNITARY_TOL

• EPS_PAO

• EPS_PGF

• LINPOT_INITGUESS_DELTA

• LINPOT_PRECONDITION_DELTA

• LINPOT_REGULARIZATION_DELTA

• LINPOT_REGULARIZATION_STRENGTH

• MAX_CYCLES

• MAX_PAO

• MIXING

• NUM_GRADIENT_EPS

• NUM_GRADIENT_ORDER

• OPTIMIZER

• PARAMETERIZATION

• PENALTY_DISTANCE

• PENALTY_STRENGTH

• PRECONDITION

• PREOPT_DM_FILE

• REGULARIZATION

• RESTART_FILE

Keyword descriptions

CG_INIT_STEPS: integer = 2

Number of steepest descent steps before starting the conjugate gradients optimization. [Edit on GitHub]

CG_RESET_LIMIT: real = 1.00000000E-001

The CG is reseted if the cosine of the angle between the last search direction and the new gradient is larger that the limit. [Edit on GitHub]

CHECK_GRADIENT_TOL: real = -1.00000000E+000

Tolerance for check of full analytic gradient against the numeric one. Negative values mean don’t check at all. [Edit on GitHub]

CHECK_UNITARY_TOL: real = -1.00000000E+000

Check if rotation matrix is unitary. Negative values mean don’t check at all. [Edit on GitHub]

EPS_PAO: real = 1.00000000E-005

Mentions: ⭐PAO-ML

Convergence criteria for PAO optimization. [Edit on GitHub]

EPS_PGF: real

Sets precision for potential and descriptor matrix elements. Overrides DFT/QS/EPS_DEFAULT value. [Edit on GitHub]

LINPOT_INITGUESS_DELTA: real = 0.00000000E+000

Eigenvalue threshold used for calculating initial guess. [Edit on GitHub]

LINPOT_PRECONDITION_DELTA: real = 0.00000000E+000

Eigenvalue threshold used for preconditioning. [Edit on GitHub]

LINPOT_REGULARIZATION_DELTA: real = 0.00000000E+000

Mentions: ⭐PAO-ML

Eigenvalue threshold used for regularization. [Edit on GitHub]

LINPOT_REGULARIZATION_STRENGTH: real = 0.00000000E+000

Strength of regularization on linpot layer. [Edit on GitHub]

MAX_CYCLES: integer = 1000

Mentions: ⭐PAO-ML

Maximum number of PAO line search cycles for a given hamiltonian. [Edit on GitHub]

MAX_PAO: integer = 1000

Mentions: ⭐PAO-ML

Maximum number of PAO basis optimization steps. [Edit on GitHub]

MIXING: real = 5.00000000E-001

Mentions: ⭐PAO-ML

Mixing fraction of new and old optimizied solutions. [Edit on GitHub]

NUM_GRADIENT_EPS: real = 1.00000000E-008

Step length used for the numeric derivative when checking the gradient. [Edit on GitHub]

NUM_GRADIENT_ORDER: integer = 2

Order of the numeric derivative when checking the gradient. Possible values are 2, 4, and 6. [Edit on GitHub]

OPTIMIZER: enum = CG

Valid values:

• CG Conjugate gradient algorithm

• BFGS Broyden-Fletcher-Goldfarb-Shanno algorithm

Optimizer used to find PAO basis. [Edit on GitHub]

PARAMETERIZATION: enum = ROTINV

Valid values:

• ROTINV Rotational invariant parametrization (machine learnable)

• FOCK Fock matrix parametrization

• GTH Parametrization based on GTH pseudo potentials

• EXP Original matrix exponential parametrization

• EQUIVARIANT Equivariant parametrization

Parametrization of the mapping between the primary and the PAO basis. [Edit on GitHub]

PENALTY_DISTANCE: real = 1.00000000E-001

Distance at which approaching eigenvalues are penalized to prevent degeneration. [Edit on GitHub]

PENALTY_STRENGTH: real = 5.00000000E-003

Strength of the penalty term which prevents degenerate eigenvalues. [Edit on GitHub]

PRECONDITION: logical = F

Lone keyword: T

Apply a preconditioner to the parametrization. [Edit on GitHub]

PREOPT_DM_FILE: string

Read pre-optimized density matrix from given file. [Edit on GitHub]

REGULARIZATION: real = 0.00000000E+000

Strength of regularization term which ensures parameters remain small. [Edit on GitHub]

RESTART_FILE: string

Reads given files as restart for PAO basis [Edit on GitHub]