MIXING
Define type and parameters for mixing procedures to be applied to the density matrix. Normally, only one type of mixing method should be accepted. The mixing procedures activated by this section are only active for diagonalization methods and linear scaling SCF, i.e. not with minimization methods based on OT. [Edit on GitHub]
Keywords
Keyword descriptions
- SECTION_PARAMETERS
Type: logical
Default: T
Lone keyword: T
Usage: &MIXING ONDescription: Controls the activation of the mixing procedure
- ALPHA
Type: real
Default: 4.00000000E-001
Usage: ALPHA 0.2Description: Fraction of new density to be included
- ALPHA_MAG
Type: real
Default: -1.00000000E+000
Usage: ALPHA_MAG 0.8Description: Fraction of new magnetization density to be included (for spin-polarized calculations, ispin=2 channel after rho_total/m transform). A negative value (default) means: use the same value as ALPHA. For magnetic transition-metal systems, a larger value (e.g. 0.8-1.6) than ALPHA often improves convergence.
- BETA
Type: real
Default: 5.00000000E-001 [bohr^-1]
Usage: BETA 1.5Description: Denominator parameter in Kerker damping introduced to suppress charge sloshing: rho_mix(g) = rho_in(g) + alphag^2/(g^2 + beta^2)(rho_out(g)-rho_in(g))
- BETA_MAG
Type: real
Default: -1.00000000E+000 [bohr^-1]
Usage: BETA_MAG 0.0Description: Kerker damping parameter for the magnetization channel (for spin-polarized calculations). A negative value (default) means: use the same value as BETA. Set to 0.0 to disable Kerker screening on the magnetization density, which avoids suppression of long-range magnetic order formation in transition-metal systems.
- BROY_W0
Type: real
Default: 1.00000000E-002
Usage: BROY_W0 0.03Description: Regularization weight used in Broyden mixing. For the original BROYDEN_MIXING method this is the constant diagonal regularization of the small Broyden system. For MODIFIED_BROYDEN_MIXING it is the corresponding diagonal regularization of the dynamically weighted Broyden system. The default follows tblite.
- BROY_WMAX
Type: real
Default: 1.00000000E+005
Usage: BROY_WMAX 100000.0Description: Upper bound for the dynamic residual weight. This keyword is only used by MODIFIED_BROYDEN_MIXING; the original BROYDEN_MIXING path is unchanged. The lower bound is fixed to 1.0. The default follows tblite.
- BROY_WREF
Type: real
Default: 1.00000000E-002
Usage: BROY_WREF 0.01Description: Reference factor for the dynamic residual weight. This keyword is only used by MODIFIED_BROYDEN_MIXING; the original BROYDEN_MIXING path is unchanged. The effective history weight is proportional to BROY_WREF divided by the residual norm, clipped to the interval [1, BROY_WMAX]. The default follows tblite.
- GMIX_P
Type: logical
Default: F
Lone keyword: T
Usage: GMIX_PDescription: Activate the mixing of the density matrix, using the same mixing coefficient applied for the g-space mixing.
- MAX_GVEC_EXP
Type: real
Default: -1.00000000E+000
Usage: MAX_GVEC_EXP 3.Description: Restricts the G-space mixing to lower part of G-vector spectrum, up to a G0, by assigning the exponent of the Gaussian that can be represented by vectors smaller than G0 within a certain accuracy.
- MAX_STEP
Type: real
Default: 1.00000000E-001
Usage: MAX_STEP .2Description: Upper bound for the magnitude of the unpredicted step size in the update by the multisecant mixing scheme
- METHOD
Type: enum
Default: DIRECT_P_MIXING
Usage: METHOD KERKER_MIXINGDescription: Mixing method to be applied
Valid values:
NONENo mixing is appliedDIRECT_P_MIXINGDirect mixing of new and old density matricesKERKER_MIXINGMixing of the potential in reciprocal space using the Kerker dampingPULAY_MIXINGPulay mixingBROYDEN_MIXINGOriginal CP2K Broyden mixing with a constant BROY_W0 regularizationMODIFIED_BROYDEN_MIXINGModified Broyden mixing with dynamic residual weights controlled by BROY_W0, BROY_WREF, and BROY_WMAXMULTISECANT_MIXINGMultisecant scheme for mixingNEW_PULAY_MIXINGNew Pulay mixing using Sundararaman et al.’s metric and preconditioner, with improved convergence behavior and suitable for grand canonical SCF
- NBUFFER
Type: integer
Default: 4
Aliases: NPULAY, NBROYDEN, NMULTISECANT
Usage: NBUFFER 2Description: Number of previous steps stored for the actual mixing scheme
- NMIXING
Type: integer
Default: 2
Usage: NMIXING 1Description: Minimal number of density mixing (should be greater than 0), before starting DIIS
- NSKIP
Type: integer
Default: 0
Aliases: NSKIP_MIXING
Usage: NSKIP 10Description: Number of initial iteration for which the mixing is skipped
- N_SIMPLE_MIX
Type: integer
Default: 0
Aliases: NSIMPLEMIX
Usage: NSIMPLEMIXDescription: Number of kerker damping iterations before starting other mixing procedures
- PULAY_ALPHA
Type: real
Default: 0.00000000E+000
Usage: PULAY_ALPHA 0.2Description: Fraction of new density to be added to the Pulay expansion
- PULAY_BETA
Type: real
Default: 1.00000000E+000
Usage: PULAY_BETA 0.2Description: Fraction of residual contribution to be added to Pulay expansion
- QK
Type: real
Default: 3.00000000E+000 [bohr^-1]
Usage: QK 3.0Description: The control parameter in the denominator of the Kerker preconditioner used in the new Pulay mixing, introduced to suppress charge sloshing: Kerker preconditioner: K(g) = alpha * (g^2 + qkapa^2)/(g^2 + qk^2 + qkapa^2)
- QKAPPA
Type: real
Default: 2.50000000E-001 [bohr^-1]
Usage: QKAPPA 0.25Description: The control parameter in the numerator and the denominator of the Pulay metric and the Kerker preconditioner used in the new Pulay mixing, introduced to ensure a finite and well-defined Pulay metric at g=0 and a non-zero and well-defined Kerker preconditioner at g=0
- QM
Type: real
Default: 7.50000000E-001 [bohr^-1]
Usage: QM 0.75Description: The control parameter in the numerator of the Pulay metric used in the new Pulay mixing, introduced to suppress charge sloshing: Pulay metric: M(g) = (g^2 + qm^2 + qkapa^2)/(g^2 + qkapa^2)
- REGULARIZATION
Type: real
Default: 1.00000000E-005
Usage: REGULARIZATION 0.000001Description: Regularization parameter to stabilize the inversion of the residual matrix {Yn^t Yn} in the multisecant mixing scheme (noise)
- R_FACTOR
Type: real
Default: 5.00000000E-002
Usage: R_FACTOR .12Description: Control factor for the magnitude of the unpredicted step size in the update by the multisecant mixing scheme