ALMO_OPTIMIZER_PCG
Controls the PCG optimization of block-diagonal ALMOs. [Edit on GitHub]
Keywords
Keyword descriptions
- CONJUGATOR
Type: enum
Default: HAGER_ZHANG
Usage: CONJUGATOR POLAK_RIBIEREDescription: Various methods to compute step directions in the PCG optimization
Valid values:
ZEROSteepest descentPOLAK_RIBIEREPolak and RibiereFLETCHER_REEVESFletcher and ReevesHESTENES_STIEFELHestenes and StiefelFLETCHERFletcher (Conjugate descent)LIU_STOREYLiu and StoreyDAI_YUANDai and YuanHAGER_ZHANGHager and Zhang
- EPS_ERROR
Type: real
Default: 1.00000000E-005
Usage: EPS_ERROR 1.E-6Description: Target value of the MAX norm of the error
- EPS_ERROR_EARLY
Type: real
Default: -1.00000000E+000
Usage: EPS_ERROR_EARLY 1.E-2Description: Target value of the MAX norm of the error for truncated SCF (e.g. Langevin-corrected MD). Negative values mean that this keyword is not used.
- LIN_SEARCH_EPS_ERROR
Type: real
Default: 1.00000000E-003
Usage: LIN_SEARCH_EPS_ERROR 1.E-2Description: Target value of the gradient norm during the linear search
- LIN_SEARCH_STEP_SIZE_GUESS
Type: real
Default: 1.00000000E+000
Usage: LIN_SEARCH_STEP_SIZE_GUESS 0.1Description: The size of the first step in the linear search
- MAX_ITER
Type: integer
Default: 20
Usage: MAX_ITER 100Description: Maximum number of iterations
- MAX_ITER_EARLY
Type: integer
Default: -1
Usage: MAX_ITER_EARLY 5Description: Maximum number of iterations for truncated SCF (e.g. Langevin-corrected MD). Negative values mean that this keyword is not used.
- MAX_ITER_OUTER_LOOP
Type: integer
Default: 0
Usage: MAX_ITER_OUTER_LOOP 10Description: Maximum number of iterations in the outer loop. Use the outer loop to update the preconditioner and reset the conjugator. This can speed up convergence significantly.
- PRECONDITIONER
Type: enum
Default: DEFAULT
Usage: PRECONDITIONER DOMAINDescription: Select a preconditioner for the conjugate gradient optimization
Valid values:
NONEDo not use preconditionerDEFAULTSame as DOMAIN preconditionerDOMAINInvert preconditioner domain-by-domain. The main component of the linear scaling algorithmFULLSolve linear equations step=-H.grad on the entire space
- PRECOND_FILTER_THRESHOLD
Type: real
Default: -1.00000000E+000
Usage: PRECOND_FILTER_THRESHOLD 0.1Description: Select eigenvalues of the preconditioner that are smaller than the threshold and project out the corresponding eigenvectors from the gradient. No matter how large the threshold is the maximum number of projected eienvectors for a fragment equals to the number of occupied orbitals of fragment’s neighbors.