Section OPTIMIZE_BASIS

• describes a basis optimization job, in which an ADMM like approach is used to find the best exponents and/or coefficients to match a given training set.

• Section path: CP2K_INPUT / OPTIMIZE_BASIS

• This section cannot be repeated.

• Subsections

  • FIT_KIND
  • OPTIMIZATION
  • TRAINING_FILES

• Keywords

  • BASIS_COMBINATIONS
  • BASIS_OUTPUT_FILE
  • BASIS_TEMPLATE_FILE
  • BASIS_WORK_FILE
  • CONDITION_WEIGHT
  • GROUP_PARTITION
  • RESIDUUM_WEIGHT
  • USE_CONDITION_NUMBER
  • WRITE_FREQUENCY

• Keyword descriptions

  BASIS_COMBINATIONS
  	BASIS_COMBINATIONS {Integer} ...
  	If multiple atomic kinds are fitted at the same time, this keyword allows to specify which basis sets should be used together in optimization (underived set ID=0). If skipped all combinations are used. The order is taken as the kinds and sets are specified in the input
  	This keyword can be repeated and it expects a list of integers.

  BASIS_OUTPUT_FILE
  	BASIS_OUTPUT_FILE {Word} ...
  	Name of the basis set file containing the optimized basis
  	This keyword cannot be repeated and it expects a list of words.
  	Default value: BASIS_OUTPUT_FILE

  BASIS_TEMPLATE_FILE
  	BASIS_TEMPLATE_FILE {Word} ...
  	Name of the basis set file, containing the structure of the new basis set
  	This keyword cannot be repeated and it expects a list of words.
  	Default value: BASIS_SET

  BASIS_WORK_FILE
  	BASIS_WORK_FILE {Word} ...
  	Name of the basis set file which is created to be read as initial guess
  	This keyword cannot be repeated and it expects a list of words.
  	Default value: BASIS_WORK_FILE

  CONDITION_WEIGHT
  	CONDITION_WEIGHT {Real}
  	This keyword allows to give different weight factors to the condition number of different basis combinations (LOG(cond) is used). The first entry corresponds to the original basis sets. Every further value is assigned to the combinations in the order given for BASIS_COMBINATIONS.
  	This keyword can be repeated and it expects precisely one real.
  	Default value: 1.00000000E+000

  GROUP_PARTITION
  	GROUP_PARTITION {Integer} ...
  	Allows the specification of the group mpi group sizes in parallel runs. If less Groups than tasks are speciefied, consecutive calculations Will be assigned to one group (derived basis sets and then training sets) If keyword is skipped, equal group sizes will be generated trying to fit all calculations.
  	This keyword can be repeated and it expects a list of integers.

  RESIDUUM_WEIGHT
  	RESIDUUM_WEIGHT {Real}
  	This keyword allows to give different weight factors to the residuum of the different basis combinations. The first entry corresponds to the original basis sets. Every further value is assigned to the combinations in the order given for BASIS_COMBINATIONS.
  	This keyword can be repeated and it expects precisely one real.
  	Default value: 1.00000000E+000

  USE_CONDITION_NUMBER
  	USE_CONDITION_NUMBER {Logical}
  	Determines whether condition number should be part of optimization or not
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  WRITE_FREQUENCY
  	WRITE_FREQUENCY {Integer}
  	Frequency at which the intermediate results should be written
  	This keyword cannot be repeated and it expects precisely one integer.
  	Default value: 5000