Section DIAGONALIZATION

• Set up type and parameters for Kohn-Sham matrix diagonalization.

• Section path: CP2K_INPUT / FORCE_EVAL / DFT / SCF / DIAGONALIZATION

• This section cannot be repeated and can be optional.

• Subsections

  • DIAG_SUB_SCF
  • OT

• Keywords

  • SECTION_PARAMETERS
  • ALGORITHM
  • CHECK_MOS_CONV
  • EPS_ADAPT
  • EPS_ITER
  • EPS_JACOBI
  • EPS_KRYLOV
  • EPS_STD_DIAG
  • JACOBI_THRESHOLD
  • MAX_ITER
  • NBLOCK_KRYLOV
  • NKRYLOV

• Keyword descriptions

  SECTION_PARAMETERS
  	&DIAGONALIZATION T
  	controls the activation of the diagonalization method
  	This optional keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  ALGORITHM
  	ALGORITHM STANDARD
  	Algorithm to be used for diagonalization
  	This optional keyword cannot be repeated and it expects precisely one keyword.
  	Default value: STANDARD
  	List of valid keywords: LANCZOS Block Krylov-space approach to self-consistent diagonalisation OT Iterative diagonalization using OT method STANDARD Standard diagonalization: LAPACK methods or Jacobi.

  CHECK_MOS_CONV
  	CHECK_MOS_CONV T
  	This requires to check the convergence of MOS also when standard diagonalization steps are performed, if the block krylov approach is active.
  	This optional keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  EPS_ADAPT
  	EPS_ADAPT 0.01
  	Required accuracy in iterative diagonalization as compared to current SCF convergence
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: 0.00000000E+00

  EPS_ITER
  	EPS_ITER 1.E-8
  	Required accuracy in iterative diagonalization
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-08

  EPS_JACOBI
  	EPS_JACOBI 1.0E-5
  	Below this threshold value for the SCF convergence the pseudo-diagonalization method using Jacobi rotations is activated. This method is much faster than a real diagonalization and it is even speeding up while achieving full convergence.However, it needs a pre-converged wavefunction obtained by at least one real diagonalization which is further optimized while keeping the original eigenvalue spectrum. The MO eigenvalues are NOT updated. The method might be useful to speed up calculations for large systems e.g. using a semi-empirical method.
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: 0.00000000E+00
  	This keyword cites the following reference: [Stewart1982]

  EPS_KRYLOV
  	EPS_KRYLOV 0.00001
  	Convergence criterion for the MOs
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-07

  EPS_STD_DIAG
  	EPS_STD_DIAG 0.001
  	Level of convergence to be reached before starting the Lanczos procedure. Above this threshold a standard diagonalization method is used. If negative Lanczos is started at the first iteration
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: -1.00000000E+00

  JACOBI_THRESHOLD
  	JACOBI_THRESHOLD 1.0E-6
  	Controls the accuracy of the pseudo-diagonalization method using Jacobi rotations
  	This optional keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-07
  	This keyword cites the following reference: [Stewart1982]

  MAX_ITER
  	MAX_ITER 20
  	Maximum number of iterations in iterative diagonalization
  	This optional keyword cannot be repeated and it expects precisely one integer.
  	Default value: 2

  NBLOCK_KRYLOV
  	NBLOCK_KRYLOV 1
  	Size of the block of vectors refined simultaneously by the Lanczos procedure
  	This optional keyword cannot be repeated and it expects precisely one integer.
  	Default value: 32

  NKRYLOV
  	NKRYLOV 20
  	Dimension of the Krylov space used for the Lanczos refinement
  	This optional keyword cannot be repeated and it expects precisely one integer.
  	Default value: 4