hi,
I am running a job involving a surface (metallic) with an adsorbate atom with ISMEAR = 0. The number of electrons is 237 and I am doing spin polarized calculation.
When I run the optimization job with sigma of 0.01, the energy (energy(sigma->0) ) is -311.846998 and the magnetization is 2.8993584.
When I run it with sigma of 0.1, the energy (energy(sigma->0)) is -311.851548 and the magnetization is 2.3561691.
Lastly, when doing a single point calculation with ISMEAR=-5 and sigma=0.2, the energy (energy(sigma->0) ) is -311.838427 and the magnetization is 1.651752.
Which energy and, more importantly, magnetization values are most accurate? Which method (smearing) should I use in my further calculations?
Can someone please comment on it.
Thanks.
sigma
Moderators: Global Moderator, Moderator
-
tlchan
sigma
You can refer to the vasp manual about which smearing method to use. In principle, a calculation is the most accurate for sigma = 0. However, an appropriate choice of sigma can help improve convergence.
For metallic systems that involve more than the gamma point for k-point sampling, a large number of k-points may be needed for convergence. For this case, you can plot what you want to calculate (for example adsorption energy, magnetization) as a function of sigma (for each sigma, you have to test the convergence with respect to k-points). From the trend, you decide on a maximum possible sigma that gives a reasonable result with the smallest number of k-points.
For gamma-only calculations, smearing serves as an electronic temperature which helps to improve electronic iterations. You can tune sigma to reduce the number of self-consistent electronic iterations while maintaining the accuracy of your results.
The determination of the optimal sigma might be tedious. However, you only need to do it once, and use the same sigma throughout the whole project.
For metallic systems that involve more than the gamma point for k-point sampling, a large number of k-points may be needed for convergence. For this case, you can plot what you want to calculate (for example adsorption energy, magnetization) as a function of sigma (for each sigma, you have to test the convergence with respect to k-points). From the trend, you decide on a maximum possible sigma that gives a reasonable result with the smallest number of k-points.
For gamma-only calculations, smearing serves as an electronic temperature which helps to improve electronic iterations. You can tune sigma to reduce the number of self-consistent electronic iterations while maintaining the accuracy of your results.
The determination of the optimal sigma might be tedious. However, you only need to do it once, and use the same sigma throughout the whole project.
Last edited by tlchan on Sat Jun 25, 2011 3:01 pm, edited 1 time in total.