BFGS

Provides parameters to tune the BFGS optimization [Edit on GitHub]

Subsections

• RESTART

Keywords

• METHOD

• RESTART_FILE_NAME

• RESTART_HESSIAN

• TRUST_RADIUS

• USE_MODEL_HESSIAN

• USE_RAT_FUN_OPT

Keyword descriptions

METHOD

Type: enum
Default: DIRECT
Usage: METHOD INVERSE

Description: Matrix updated by full-memory BFGS. DIRECT updates the Hessian; INVERSE updates its inverse without a full diagonalization at each step. [Edit on GitHub]

Valid values:

• DIRECT

• INVERSE

RESTART_FILE_NAME

Type: string
Usage: RESTART_FILE_NAME {filename}

Description: Specifies the name of the file used to read the initial Hessian. [Edit on GitHub]

RESTART_HESSIAN

Type: logical
Default: F
Lone keyword: T
Usage: RESTART_HESSIAN

Description: Reads the initial BFGS matrix from file. Hessian and inverse-Hessian restarts are converted to the selected METHOD when necessary. [Edit on GitHub]

TRUST_RADIUS

Type: real
Default: 2.50000000E-001 [angstrom]
Usage: TRUST_RADIUS {real}

Description: Trust radius used in BFGS. Previously set to 0.1. Large values can lead to instabilities [Edit on GitHub]

USE_MODEL_HESSIAN

Type: logical
Default: F
Lone keyword: T
Usage: USE_MODEL_HESSIAN

Description: Uses a model Hessian as initial guess instead of a unit matrix. Should lead in general to improved convergence might be switched off for exotic cases [Edit on GitHub]

References: Lindh1995

USE_RAT_FUN_OPT

Type: logical
Default: F
Lone keyword: T
Usage: USE_RAT_FUN_OPT

Description: Includes a rational function optimization to determine the step. Previously default but did not improve convergence in many cases. Not supported by inverse-Hessian update method. [Edit on GitHub]