# Evaluating Seebeck coefficient using DFT

I am working with semiconductor materials using the SIESTA DFT package. I have tried the BoltzTraP2 software interface with SIESTA, however the results (Seebeck coefficient of MgO and Si) did not match with those published in literature. Are there any other options available to do this with reduced computational cost?

• The issue could be that using DFT and some pseudopotential/functionals fail to predict the band structure (band gaps) for semiconductors. So, moving to other DFT codes. will give the same results.
– Camps
Sep 29 at 9:35
• I don't know much about this program, but I think it will help if you include the settings you used for the calculation, your results, and the literature results for these systems.
– Tyberius
Sep 29 at 13:06
• Hi, Thanks for the insightful comments. I used the SIESTA DFT package with DZP Basis (with GGA exchange correlation functional). Also I verified the bandgap and the band diagram of bulk Si before using that with Boltztrap2. However, the data for MgO were directly taken from those provided by the Boltztrap2 wiki (where they provide all the output files from a SIESTA calculation).
– PBH
Sep 29 at 14:54
• Well, if you use all the data from the program site, I thong you should be in contact with the developers through the user group.
– Camps
Sep 29 at 15:32
• Yes I am working on that. But in the meanwhile, I am trying to figure out the other options which are available to do this.
– PBH
Sep 30 at 1:59

I'm posting this here as an answer so that if anyone else runs into the same problem, they can get it clarified.

After searching endlessly (because the SIESTA to BoltzTraP2 interface provided by the developers failed), I figured it out by myself. What happened was a problem of the k-point output file from SIESTA not being what BoltzTraP2 expects.

If you are working to link the two packages, follow the instructions provided in their wiki with one exception. At the place where the SYSTEM_NAME.KP file is read by BoltzTraP2, make sure to multiply the array of k-points by 2 before being used by BoltzTraP2.

Also, the BoltzTraP2 package works with a gamma centered, uniform k-grid. To achieve this, the best way would be to use odd number of k-points (to have equal numbers of k-points on each side of the gamma point) and to set 0. to the dk value (to obtain a gamma centered grid). However, I cannot verify this statement for every unit cell implementation. So in such a case, plot the k-points generated by the Monkhorst-Pack scheme and see if the resultant grid is uniform or not.

Additionally, the current version of BoltzTraP2 does not support non-periodic boundary conditions. To analyze non-periodic systems, I currently set 3 k-points in the non-periodic direction (which includes a vacuum and so no interactions would be present between adjacent unit cell images) and ignore the BoltzTraP2 results for that direction. I am also working on a modification to the code to enable support for 2D and 1D systems, and shall update this thread if I succeed.