Zhen, The convergence on EF is very important if you perform transport calculations. Otherwise, the important quantity here is the total energy. Now I am a bit confused by your calculations... If you are using such a big supercell, it could be that this huge k-point sampling is simply overkill and maybe your problem is not in the pseudopotential. Where do you get false magnetism (to quote Derek) - in the supercell, or in the unit cell of hcp Ti? One other thing: as Andrei mentioned, incomplete convergence of the SCF cycle could cause you to get false magnetism. What is your tolerance for the SCF convergence? Have you made sure that you have a fully converged SCF cycle, that is, that the SCF calculation ends in a number of SCF steps smaller than, or equal to, the maximum number of SCF steps? If possible, send us also the output of the calculation (the .out file).
One even more basic question. For the unit cell of hcp Ti... does everything com out ok? I mean, structure, lattice parameters, magnetic moment, all come out fine for the unit cell of hcp Ti? If this is not the case, going to huge supercells is simply wasting time - both yours and the CPU time of your cluster... Marcos On Thu, Mar 4, 2010 at 6:20 PM, Zhen Huang <[email protected]> wrote: > Marcos- > Thanks for comment on it. First I took advice from previous note and use > kgrid_cutoff 25-28 Ang in recent calculation instead of k points which gives > me 18 by 18 by 11. I have increased the cutoff to 28 which gives me Ef > (-5.8018 eV) which is less than one percent higher energy than 25 Ang > (-5.8072). > Thanks for correcting me on the way SIESTA's calculation. I used 3*3*3 and > the 5*5*5 is running for two days already which is very slow with kgrid_cut > off 25 Ang. I am hoping the dispersion calculation will be done in couple of > days. As you said it could be the number of supercell can not satisfy the > require. > > Best Regards > ------- > Zhen (Alex) Huang > Ph.D. Student > Nanoscale Transport Research Group > Laboratory for Computational Methods in Emerging Technologies > Cooling Technologies Research Center > School of Mechanical Engineering > Purdue University > Tel: 765 237 9733 > > > 2010/3/4 Marcos Veríssimo Alves <[email protected]> >> >> Zhen, >> >> Again I ask you: what k-point sampling are you using in your >> calculations? The reason I am being pushy on this is because I am very >> familiar with an extreme case, which is graphene. You can get very >> reasonable results when you perform calculations for graphene with a >> low k-sampling, if you enforce symmetries - for example, setting >> >> LatticeConstant xxxx Ang >> >> %block LatticeVectors >> 0.5 0.866025404 0.0 >> -0.5 0.866025404 0.0 >> 0.0 0.0 10.0 >> %endblock LatticeVectors >> >> %block kgrid_monkhorst_pack >> 9 0 0 0.0 >> 0 9 0 0.0 >> 0 0 1 0.0 >> %endblock kgrid_monkhorst_pack >> >> but when you do a variable-cell calculation, in the end, you will find >> a sizeable deviation, in the angle between the in-plane lattice >> vectors, from the expected 60 degrees. It only goes away when you set >> a pretty high value for the MP grid, such as 21x21x1. Being extremely >> pushy, what you should do is convergence studies in your parameters, >> if you still haven't done so. It could well be that some parameters in >> your calculation are not good, which happens often when systematic >> convergence studies are not performed. Please disregard this if you >> have performed said studies, but contact you supervisor if you don't >> know how to perform one - he/she should be able to teach you how to do >> this. If not, let us know. >> >> It could be that, as Derek states, you could need semicore states or >> non-linear core corrections to be included in some cases. It would be >> advisable to check on the literature if this is the case for your >> metallic Ti. >> >> Now, one correction. Phonons in siesta are not calculated by >> frozen-phonon calculations, although siesta (and any other ab initio >> software, for that matter) can perform frozen-phonon calculations. >> What siesta does is a calculation of the force constant matrix in real >> space by finite differences. Then the phonon frequencies are >> calculated by diagonalization of the FC matrix. >> >> Now the following wild guess occurs to me... Your phonon softening >> seems to occur for large values of the phonon wavevector, q, and for >> small values it's pretty well-behaved. Could it be that your supercell >> is not large enough to support such a mode, and then it gives you this >> spurious softening? Phonon experts on the list, am I just guessing it >> wrong? >> >> Marcos >> >> >> On Thu, Mar 4, 2010 at 4:54 PM, Zhen Huang <[email protected]> wrote: >> > Dear Derek and folks- >> > Thanks for you responds. I have tried ABINIT LDA from SIESTA website and >> > it >> > also give very small lattice constants without any treatment to spin >> > (2.67 >> > Ang comparing to 2.95 Ang from textbook). Further, the one of the angles >> > changed from 120 to 108 degree. I could not tell what is the problem is. >> > I >> > have used PWSCF to obtain phonon dispersion of Ti before and it worked >> > out >> > fine. Although, I have never gone into the details of Isaev's paper. >> > The reason I switch to SIESTA, is the phonon calculation is based on >> > frozen >> > phonon and when it comes to large system it is much faster than Quantum >> > espresso. However, seems at this point, I am not able to get the Ti >> > running >> > properly. >> > Best Regards >> > ------- >> > Zhen (Alex) Huang >> > Ph.D. Student >> > Nanoscale Transport Research Group >> > Laboratory for Computational Methods in Emerging Technologies >> > Cooling Technologies Research Center >> > School of Mechanical Engineering >> > Purdue University >> > Tel: 765 237 9733 >> > > >
