Dear Andrei: I don't see how your system can be a perovskite, nor how CH3N3SnI3 sums up to 12 atoms. I'll appreciate if you give a hint, just for my education. However, this is beyond the point. In principle, a mere passing to a supercell should not result in an increase of the band gap. The mapping of k-points may be not exact, i.e., Gamma of the 3*3*3 supercell maps onto 0, 1/3, 2/3 [*(2*pi/a)] divisions along axes of the reciprocal cell, none of which points includes R=(1/2 1/2 1/2). You can check what the band gap (in the primitive cell) is at (1/3 1/3 1/3), or at (3/8 3/8 3/8), the closest of your k-points at the 8x8x8 mesh. [However, you say that you have an automatic k sampling, then the chances are that your mesh is shifted... and does not contain the R point..?] Apart from these tests, I think that some simple error in structure (as simple as a wrong lattice constant in a supercell, sorry) is a very good candidate for this kind of problem. To check it, it might be a good idea to have a view at both structures (the primitive one and the supercell) as stored in corresponding XV files, and to compare their corresponding DOS.
Again apart from all this, if your system contains methyl azide, I think it could be tricky to initialize the different charge states at different N atoms correctly... If you just let it go as it wish the calculation may converge to something wrong... Even to differently wrong in two different supercells. Sorry for too many suggestions. Best regards Andrei Postnikov > Dear Siesta users, > > I'm having very bizarre problem. > I'm simulating the primitive cubic unit cell(12 atoms) of the CH3N3SnI3 > perovksite with GGA and PBE and gives me band gap of Eg=0.9 eV at R point > with 8x8x8 Moshkrof automatic k sampling(i know it's wrong in DFT, but > still) and then i use supercell approach built in with 3x3x3 unit > cell(432 atoms) with Gamma point only and i get a bangap of 1.6 eV. I've > checked against plane cutoff convergence seems ok. I thought maybe > dispersion is too big so one should sample it more finely, but it turns > out meff=0.13me, which should be fine. Any ideas ? > > > With Best Regards, Andrei Buin. >
