Keyword SHIFT

This keyword activates the level-shift procedure.
Options:

	$+<$Real$>$		Fixed level-shift value [in a.u.].
	$-<$Real$>$		Dynamic level-shift value [in a.u.].
	MIN=$<$Real$>$		Minimum value for dynamical level-shift [in a.u.].

Description:
The level-shift procedure enables enlargement of the HOMO-LUMO gap during the SCF iterations [191]. This stabilizes the initial SCF configuration preferentially. Because of the small HOMO-LUMO gap common in DFT calculations, this procedure can be very valuable for improving SCF convergence. Beware, however, that with a fixed level-shift, incorrect convergence to a spurious excited state is possible. When using SHIFT therefore, it is imperative to check the converged orbital occupations with the MOS option of the PRINT keyword (see 4.12.2). The problem can be avoided if the dynamical level-shift procedure [155] is used. In that case, the shift value is adapted to the MinMax SCF error. For large errors, which usually occur at the beginning of the SCF procedure, the full shift value is used. As the error decreases, the shift value is also decreased (see Example on page of the tutorial). A small shift value may remain at the end of the SCF procedure. In fact a minimum shift value can be requested with the MIN option. Small remaining shift values at the end of the SCF are usually of no concern. The dynamical level-shift procedure has proven very valuable for the SCF convergence of transition-metal systems [192]. In the case of ROKS calculations, the level-shift procedure applies twice, once between the doubly and singly occupied orbitals and again between the singly-occupied and the unoccupied orbitals. In both cases, the specified level-shift value is used.

If the SHIFT keyword is used, the level-shift procedure is switched on for all calculation steps (optimization, frequencies, properties, etc.). In the case that the level-shift procedure is used only to prepare a start density, a single-point SCF must be performed with the SHIFT keyword and the resulting density then has to be read in as the restart density (see 4.5.5) in the subsequent calculation without the SHIFT keyword.