Top Page > Browsing

completness of basis set Date: 2007/06/28 00:02 Name: JessK Dear Dr. Ozaki, I am testing basis set for strontium and want to be sure that the basis set is not overcomplete. I use several 'rc' values - from 6.0 to 11.0 ans different number of basis functions. I tested it on Sr bulk (fcc). I found that s2p2d1 basis is good for almost all rc values (6.0 gave me bad results). For rc=7.0, 8.0, 9.0, 10.0 the lattice constant error is ~6.25%, 4.6%, 3% and 3%, respectively. From this, I thought I have to choose rc=9.0 or rc=10.0. Cohesive energy using those rc values is -2.07 eV and -1.90 eV ; reference data for cohesive energy is -1.73. So, it looks like, that rc=10.0 is probably a good choice. But, I have a doubt that 10.0 may be too high. What do you think? Regards, Jess

Page: [1]

Re: completness of basis set ( No.1 ) Date: 2007/06/28 20:03 Name: T.Ozaki Hi, Though the optimum choice of cutoff radius depends on the systems, 9.0 ang seems to be Okay according to your calculations for Sr bulk (fcc). I wonder that you used the counterpoise method to reduce basis set super position error (BSSE) in the calculation of the cohesive energy. I guess that the 9.0 ang basis gives a closer result to the reference value when the counterpoise method is used. (how to do the counterpoise method: In the single atom calculation the counterpoise method can be applied by putting empty atoms to atomic positions of Sr in the FCC except for one Sr atom in a supercell, where the empty atom should have the same basis functions as PAOs of Sr (to do that you have to replace the numerical tables for only the basis functions in the PAO file of empty atom by the numerical tables for only the basis functions in the PAO file of Sr). The position of empty atoms can be chosen as those of the neighbouring atoms in Sr fcc bulk. Probably, it would be enough to put the empy atoms to the positions of the first neighbouring atoms. Then, it is assured that the quality of basis functions used in the single atom calculation is almost same as the bulk calculation.) If you are going to calculate oxides including Sr, the optimium cutoff becomes shorter since the delocalized s-state can be empty. For such a case it would be better to do test calculations for oxides. Regards, TO

Page: [1]