ENCUTGWSOFT

ENCUTGWSOFT = [real]

Default: ENCUTGWSOFT	= ENCUTGW $\times 0.8$	for ALGO=ACFDT
	= ENCUTGW $\times 0.8$	as of VASP.6.3
	= ENCUTGW	else

Important: For vasp.6.3 and later releases ENCUTGWSOFT always defaults to ENCUTGW

\times 0.8

.

Descprition: The flag ENCUTGWSOFT sets the energy cutoff for response function, such that it allows to truncate the Coulomb kernel slowly between the energy specified by ENCUTGWSOFT and ENCUTGW using a cosine window function.

This usually leads to much smoother energy-volume curves in ACFDT calculations and MP2 calculations. The modified Coulomb kernel is in this case: ${\displaystyle v_{G} = \frac{4 \pi e^2} {G^2} \frac{1}{2} \left( 1 + \cos \left( \pi \, \frac{ \frac{\hbar^{2} G^2 }{2 m_e} - \mathrm{ ENCUTGWSOFT} }{ \mathrm{ENCUTGW} - \mathrm{ENCUTGWSOFT}} \right) \right) \qquad \mbox{for} \quad \frac{\hbar^2 G^2 }{2 m_e} > \mathrm{ENCUTGWSOFT}}$

If LSCK is set to .TRUE., the squeezed Coulomb kernel is used instead of the cosine window:^[1]

${\displaystyle v_{G} = 4 \pi e^2 \frac{ (G_{max}-G_{min})(G_{max}-G) }{ (G_{min}^2 - G(2G_{min}-G_{max}))^2 } \qquad \mbox{for} \quad \mathrm{ENCUTGWSOFT}=\frac{\hbar^2G_{min}^2}{2m_e}<\frac{\hbar^2 G^2}{2m_e}<\frac{\hbar^2G_{max}^2}{2m_e}=\mathrm{ENCUTGW}}$

This kernel squeezes contributions from large wave vectors $G>G_{max}$ into the window given by ENCUTGWSOFT. Effectively, this extrapolates the RPA correlation energy to the ENCUTGW $\to \infty$ limit, assuming that the basis set incompleteness error falls off as $1/\mathrm{ENCUTGW}^{3/2}$ .