Section OUTER_SCF

• parameters controlling the outer SCF loop

• Section path: CP2K_INPUT / FORCE_EVAL / DFT / XAS / SCF / OUTER_SCF

• This section cannot be repeated.

• Subsections

  • none

• Keywords

  • SECTION_PARAMETERS
  • BISECT_TRUST_COUNT
  • BROYDEN_TYPE
  • CONTINUE_LS
  • DIIS_BUFFER_LENGTH
  • EPS_SCF
  • EXTRAPOLATION_ORDER
  • FACTOR_LS
  • JACOBIAN_FREQ
  • JACOBIAN_RESTART
  • JACOBIAN_STEP
  • JACOBIAN_TYPE
  • JACOBIAN_VECTOR
  • MAX_LS
  • MAX_SCF
  • OPTIMIZER
  • STEP_SIZE
  • TYPE

• Keyword descriptions

  SECTION_PARAMETERS
  	&OUTER_SCF {Logical}
  	controls the activation of the outer SCF loop
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  BISECT_TRUST_COUNT
  	BISECT_TRUST_COUNT {Integer}
  	Maximum number of times the same point will be used in bisection, a small number guards against the effect of wrongly converged states.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 10

  BROYDEN_TYPE
  	BROYDEN_TYPE {Keyword}
  	Specifies the Broyden optimizer variant to use.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: BT1
  	List of valid keywords: BT1 Broyden's first method, also known as the good method. The initial Jacobian is built from MD history if available. Otherwise switches to SD for one SCF iteration until a Jacobian can be built from the SCF history. BT1_EXPLICIT Same as BT1, but computes the explicit Jacobian with finite differences. Requires a CDFT SCF procedure to be active. BT1_EXPLICIT_LS Same as BT1_EXPLICIT, but uses backtracking line search for optimizing the step size. BT1_LS Same as BT1, but uses backtracking line search for optimizing the step size (see optimizer NEWTON_LS). BT2 Same as BT1, but uses Broyden's second method, also known as the bad method. BT2_EXPLICIT Same as BT1_EXPLICIT, but using Broyden's second method. BT2_EXPLICIT_LS Same as BT2_EXPLICIT, but uses backtracking line search for optimizing the step size. BT2_LS Same as BT2, but uses backtracking line search for optimizing the step size.

  CONTINUE_LS
  	CONTINUE_LS {Logical}
  	Continue backtracking line search until MAX_LS steps are reached or the norm of the CDFT gradient no longer decreases. Default (false) behavior exits the line search procedure on the first step that the gradient decreases.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  DIIS_BUFFER_LENGTH
  	DIIS_BUFFER_LENGTH {Integer}
  	Maximum number of DIIS vectors used
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 3

  EPS_SCF
  	EPS_SCF {Real}
  	The target gradient of the outer SCF variables. Notice that the EPS_SCF of the inner loop also determines the value that can be reached in the outer loop, typically EPS_SCF of the outer loop must be smaller than EPS_SCF of the inner loop.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-005

  EXTRAPOLATION_ORDER
  	EXTRAPOLATION_ORDER {Integer}
  	Number of past states used in the extrapolation of the variables during e.g. MD
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 3

  FACTOR_LS
  	FACTOR_LS {Real}
  	Control parameter for backtracking line search. The step size is reduced by this factor on every line search iteration. Value must be between 0 and 1 (exclusive).
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 5.00000000E-001

  JACOBIAN_FREQ
  	JACOBIAN_FREQ {Integer} {Integer}
  	Defines parameters that control how often the explicit Jacobian is built, which is needed by some optimizers. Expects two values. The first value determines how many consecutive CDFT SCF iterations should skip a rebuild, whereas the latter how many MD steps. The values can be zero (meaning never rebuild) or positive. Both values cannot be zero.
  	This keyword cannot be repeated and it expects precisely 2 integers.
  	Default values: 1 1

  JACOBIAN_RESTART
  	JACOBIAN_RESTART {Logical}
  	Restart the inverse Jacobian using the vector defined with keyword JACOBIAN_VECTOR.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  JACOBIAN_STEP
  	JACOBIAN_STEP {Real}
  	Step size to use in the calculation of the inverse Jacobian with finite differences.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-004

  JACOBIAN_TYPE
  	JACOBIAN_TYPE {Keyword}
  	Finite difference method used to calculate the inverse Jacobian needed by some optimizers. Compatible only with CDFT constraints.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: FD1
  	List of valid keywords: FD1 First order forward difference (one extra energy evaluation per constraint). FD1_BACKWARD First order backward difference (one extra energy evaluation per constraint). FD1_CENTRAL First order central difference (two extra energy evaluations per constraint). FD2 Second order forward difference (two extra energy evaluations per constraint). FD2_BACKWARD Second order backward difference (two extra energy evaluations per constraint).

  JACOBIAN_VECTOR
  	JACOBIAN_VECTOR {Real} ...
  	Defines the inverse Jacobian matrix. Useful for restarting calculations. Expects n^2 values where n is the total number of constraints. The matrix should be given in row major order.
  	This keyword cannot be repeated and it expects a list of reals.

  MAX_LS
  	MAX_LS {Integer}
  	The maximum number of backtracking line search steps to perform.
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 5

  MAX_SCF
  	MAX_SCF {Integer}
  	The maximum number of outer loops
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 50

  OPTIMIZER
  	OPTIMIZER {Keyword}
  	Method used to bring the outer loop to a stationary point
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: NONE
  	List of valid keywords: BISECT Bisection of the gradient, useful for difficult one dimensional cases BROYDEN Broyden's method. Variant defined in BROYDEN_TYPE. DIIS Uses a Direct Inversion in the Iterative Subspace method NEWTON Newton's method. Only compatible with CDFT constraints. NEWTON_LS Newton's method with backtracking line search to find the optimal step size. Only compatible with CDFT constraints. Starts from the regular Newton solution and successively reduces the step size until the L2 norm of the CDFT gradient decreases or MAX_LS steps is reached. Potentially very expensive because each iteration performs a full SCF calculation. NONE Do nothing, useful only with the none type SD Takes steps in the direction of the gradient, multiplied by step_size SECANT Secant method. Only for one dimensional cases. See Broyden for multidimensional cases.

  STEP_SIZE
  	STEP_SIZE {Real}
  	The initial step_size used in the optimizer (currently steepest descent). Note that in cases where a sadle point is sought for (constrained DFT), this can be negative. For Newton and Broyden optimizers, use a value less than the default 1.0 (in absolute value, the sign is not significant) to active a damped optimizer.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 5.00000000E-001

  TYPE
  	TYPE {Keyword}
  	Specifies which kind of outer SCF should be employed
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: NONE
  	List of valid keywords: BASIS_CENTER_OPT Optimize positions of basis functions, if atom types FLOATING_BASIS_CENTER are defined BECKE_CONSTRAINT Enforce a constraint on the Becke weight population, requires the corresponding section CDFT_CONSTRAINT Enforce a constraint on a generic CDFT weight population, requires the corresponding section which determines the type of weight used (currently only Gaussian Hirshfeld supported) DDAPC_CONSTRAINT Enforce a constraint on the DDAPC, requires the corresponding section NONE Do nothing in the outer loop, useful for resetting the inner loop, S2_CONSTRAINT Enforce a constraint on the S2, requires the corresponding section