MP smearing - comparison between metallic systems

[william_watson]

Hi,

I am examining the properties of several actinide compounds (uranium and plutonium based) which have a mix of bonding but display metallic characteristics with 5f states at the Fermi level. I have relaxed most systems using the Gaussian smearing method initially but have subsequently done some convergence testing on SIGMA for Methfessel-Paxton smearing on a subset of the actinide compounds, as well as bulk U and Pu metal. However, doing so for all the systems I wish to examine would take a lot of time, and I am unsure about whether consistent values of SIGMA should be applied to pure metals and their respective compounds (e.g. Pu and PuC). I am also using DFT+U and occupation matrix control and have already seen in some of my systems that the ground state can vary depending on whether ISMEAR=0 or 1. In addition, I also wish to generate phonon dispersion plots for structures which are believed to exist experimentally but have not been crystalographically identified therefore I am conscious that I should use Methfessel-Paxton smearing for this application to get accurate forces. Given the methodology and objectives outlined and for the number of systems involved, would it be appropriate to use the same SIGMA value identified for bulk U metal and Pu metal for a range of uranium and plutonium (ceramic) compounds?

Thanks,

William

[manuel_engel1]

Hi William,

Thanks for reaching out. The Methfessel-Paxton method can be very sensitive to the choice of SIGMA. Therefore, we recommend to always check your SIGMA value for each metallic compound that you are computing. If you are unsure, then the Gaussian smearing method is a good starting point, as you have already done.

For ceramics, avoid using ISMEAR>0, since this can lead to large errors. Instead, use the tetrahedron method (if you have many k-points), or the Gaussian smearing method with a small SIGMA.

In the end, I think it's too difficult to say whether a particular ISMEAR/SIGMA choice works for materials without testing. It might be possible to transfer the SIGMA value from one material to a closely related material, but this is not something that is very systematic, so we still recommend to be cautious and check your results.

[william_watson]

Thank you Manuel for your quick response.

Given that most of the systems that I am looking at are ceramics/semi-metals, or I do not know the electron characterisation of the material as it may not exist under equilibrium conditions (e.g. possible metastable state), would the safest approach to be apply Gaussian smearing generally? Or perhaps just for the unknown structures (where there is no experimental data for comparison e.g. lattice parameters) but systematically assess range of SIGMA values under MP scheme for structures where comparisons can be made with experimental structural data?

[manuel_engel1]

I'm happy to help.

Some good general recommendations about the use of ISMEAR/SIGMA is given on the ISMEAR wiki page. We always recommend to start with Gaussian smearing and a small SIGMA if the material is unknown.