User Tools

Site Tools


gpw

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision
Previous revision
gpw [2014/02/10 11:54] – [ot / diag] oschuettgpw [2020/08/21 10:15] (current) – external edit 127.0.0.1
Line 11: Line 11:
 $n$ and $\tilde n$ are not equal, and this introduces an error in the calculation. $\tilde n$ converges toward $n$ when the cutoff (that controls the grid spacing) goes to infinity (and gridspacing to 0). Which cutoff is sufficient to represent a density depends on how sharp is the gaussian basis set (or that of the potential, but it is always broader). $n$ and $\tilde n$ are not equal, and this introduces an error in the calculation. $\tilde n$ converges toward $n$ when the cutoff (that controls the grid spacing) goes to infinity (and gridspacing to 0). Which cutoff is sufficient to represent a density depends on how sharp is the gaussian basis set (or that of the potential, but it is always broader).
  
-For historical reasons the density of the grid is given as the energy (in Ry) of the highest reciprocal vector that can be represented on the on the grid. This can be roughly given as $0.5(\pi/dr)^2$ where $dr$ is the gridspacing in Bohr. The characteristic length of a gaussian with exponent A is given by $1/\sqrt{a}$ (up to a factor 2 depending on the convention used). This means that the cutoff to represent in the same way a gaussian depends linearly on the exponent. Thus one can get a first guess for an acceptable guess can be take from the knowledge that for water with $\alpha_H=47.7$ a good cutoff for doing MD is 280 Ry.+For historical reasons the density of the grid is given as the energy (in Ry) of the highest reciprocal vector that can be represented on the grid. This can be roughly given as $0.5(\pi/dr)^2$ where $dr$ is the gridspacing in Bohr. The characteristic length of a gaussian with exponent A is given by $1/\sqrt{a}$ (up to a factor 2 depending on the convention used). This means that the cutoff to represent in the same way a gaussian depends linearly on the exponent. Thus one can get a first guess for an acceptable guess can be take from the knowledge that for water with $\alpha_H=47.7$ a good cutoff for doing MD is 280 Ry.
  
 It turns out that if one wants to put the whole density on the grid, the core electrons of even the simplest atoms cannot be represented, thus one has to remove to core electrons and use pseudopotentials for the atoms. In Cp2k we use the [[GTH_pseudopotentials|Godeker-Theta-Hutter pseudopotentials]], these are harder than other pseudopotentials, but also more accurate. It turns out that if one wants to put the whole density on the grid, the core electrons of even the simplest atoms cannot be represented, thus one has to remove to core electrons and use pseudopotentials for the atoms. In Cp2k we use the [[GTH_pseudopotentials|Godeker-Theta-Hutter pseudopotentials]], these are harder than other pseudopotentials, but also more accurate.
Line 17: Line 17:
 ===== Smoothing ===== ===== Smoothing =====
 $\tilde n$ is optimized for the electrostatic part, but is used also to calculate the exchange and correlation potential. Because of this, and because the [[GTH pseudopotential]] goes almost to 0 close to the core of the atom, the xc potential, especially for gradient corrected functionals, converges badly. Instead of using very high cutoffs one can perform a smoothing of the density, and calculate the derivatives on the grid with other methods than the G-space based derivatives. $\tilde n$ is optimized for the electrostatic part, but is used also to calculate the exchange and correlation potential. Because of this, and because the [[GTH pseudopotential]] goes almost to 0 close to the core of the atom, the xc potential, especially for gradient corrected functionals, converges badly. Instead of using very high cutoffs one can perform a smoothing of the density, and calculate the derivatives on the grid with other methods than the G-space based derivatives.
-For MD of water using a cutoff of 280 Ry ''XC_SMOOTH_RHO NN10'' and ''XC_DERIV SPLINE2_SMOOTH'' (in the ''FORCE_EVAL%DFT%SC%XC_GRID'' section) give good results, please note that these options renormalize the total energy, and the amount of renormalization is dependent on the cutoff. Thus energies with different cutoffs cannot be easily compared, only interaction energies or forces can be calculated.+For MD of water using a cutoff of 280 Ry ''XC_SMOOTH_RHO NN10'' and ''XC_DERIV SPLINE2_SMOOTH'' (in the ''FORCE_EVAL%DFT%XC%XC_GRID'' section) give good results, please note that these options renormalize the total energy, and the amount of renormalization is dependent on the cutoff. Thus energies with different cutoffs cannot be easily compared, only interaction energies or forces can be calculated.
  
 Methods that do not redefine the total energy are ''XC_SMOOTH_RHO NONE'' and ''XC_DERIV'' equal to either ''PW, SPLINE3'' or ''SPLINE2''. These are listed from the one that assumes more regularity (''PW'' the the one that assumes less regularity ''SPLINE2''. Normally ''SPLINE2'' is a good choice, but for high cutoffs (600 Ry for water) ''SPLINE3'' is better. The default (''PW'') is not bad, but generally inferior to the others. Methods that do not redefine the total energy are ''XC_SMOOTH_RHO NONE'' and ''XC_DERIV'' equal to either ''PW, SPLINE3'' or ''SPLINE2''. These are listed from the one that assumes more regularity (''PW'' the the one that assumes less regularity ''SPLINE2''. Normally ''SPLINE2'' is a good choice, but for high cutoffs (600 Ry for water) ''SPLINE3'' is better. The default (''PW'') is not bad, but generally inferior to the others.
Line 41: Line 41:
  
 <code> <code>
-&XC_FUNCTIONAL BLIP+&XC_FUNCTIONAL BLYP
 &END XC_FUNCTIONAL &END XC_FUNCTIONAL
 </code> </code>
gpw.1392033295.txt.gz · Last modified: 2020/08/21 10:15 (external edit)