@ Michael: Just one thing... The preservation of symmetry in siesta calculations may depend largely, and more strongly, on the meshcutoff you use, or on a good use of gridcellsampling, than on the basis set and most likely only very weakly on the basis set. Remember that the forces, in siesta, always have an eggbox effect component, which will tend to spoil the preservation of symmetry after some CG/Broyden steps. The larger the mesh cutoff, the finer your real-space grid, and the smaller the eggbox effect. So you should probably use a good grid cell sampling for the removal of the eggbox effect and a mesh cutoff as high as possible. It could also be good to symmetrize the coordinates (Google FINDSYM to find out a nice, free-of-charge coordinate symmetrizer - with a web interface!) after a while and restart your calculation from there, after relaxation is reached for the first time. Ideally, you should re-symmetrize and relax as many times as necessary until you are satisfied with the degree of symmetry of the final, relaxed coordinates.
@Yefei: I think that if you set, in the geometry constraints block, stress 1 2 4 5 6 you'd have the c axis (which I am assuming to be along z) fixed because the stress on the x and y axes would be fixed. Therefore, you'd be doing a calculation in which a and b would have fixed lengths, c would vary, and the crystallographic angles would all remain the same. Please do a test on a small system to check if this is so before taking my word on it, because I think have done this before but I am not sure right now. @Bin: The first time I optimized a basis set, I did it with a very rough and primitive implementation of the simplex method, which was not made available at the time in the siesta distribution. Now you are advised to use the very flexible and powerful utility the developers have distributed with siesta. You have quite a bit of control and it is automated after you set the initial parameters. Marcos On Mon, Aug 23, 2010 at 3:18 PM, Michael R. C. Williams < [email protected]> wrote: > Dear Yefei, > > In that example I was using one of the VdW functionals, which calculated > slightly smaller unit cells than experiment. For the same system, using a > GGA functional like PBE does result in calculated dimensions that are larger > than experiment, as you expect. > > In fact, I did not include anything to ensure vector 'a' and vector 'b' > would stay the same. I wanted to use the preservation or loss of that > symmetry as another indicator of the accuracy of the calculation. Hope that > is helpful, > > Mike > > On Aug 23, 2010, at 8:48 AM, yefare li wrote: > > > Dear Michael R. C. Williams: > > Your advise for basis is interesting. I really want to known the > > unit cell vectors calculated by plane waves methods with the same xc > > functional. As my experience, the calculated unit cell vectors is > > often slightly larger than experiments. However, your calculated unit > > cell vectors is all smaller than experiments. I feel it is somehow > > strange. Besides, would you plz tell me how do you fix the conditions > > of a=b and the cell vectors' angles? Did siesta-3.0 have the new > > function? Thanks. > > Yefei > > > > On Mon, Aug 23, 2010 at 1:28 PM, Bin Shao <[email protected]> wrote: > >> Dear Marcos, > >> > >> Thank you for your advice of optimizing the basis. > >> > >> I wonder to optimizing the basis of NiO with a simple cubic structure. > The > >> following is my basis set: > >> > >> %block PAO.Basis # Define Basis set > >> Ni 2 # Species label, number of l-shells > >> n=4 0 2 P # n, l, Nzeta > >> 5.5 5.5 # rc > >> n=3 2 2 P > >> 5.0 5.0 > >> O 2 > >> n=2 0 2 P > >> 6.0 6.0 > >> 1 2 > >> 5.5 5.5 > >> %endblock PAO.Basis > >> > >> To get this, I optimize it manually. First, I fix the rc's of Ni-3d, > O-2s > >> and O-2p and calculate the total energy of the system with the rc of the > >> Ni-4d from 4.0 to 8.0 increased by 0.5. Then, use the value of rc with > the > >> lowest energy. Next, I get the rc's of Ni-3d, O-2s and O-2p with the > same > >> steps. So in this process, I need to do the calculation of the system > with > >> many times. Moreever, I used the BasisSize of DZP, there are two rc's of > the > >> same l and if I want to optimize all of them, I should do the optimize > one > >> more time and it's complicated. > >> > >> So, I want to know how do you optimize the basis, manually, or by the > >> program like Simplex? > >> > >> Thank you in advance. > >> > >> Best, > >> > >> 2010/8/17 Marcos Veríssimo Alves <[email protected]> > >>> > >>> Robert, > >>> Try optimizing the basis set for a smaller system and using it in your > >>> larger system. For example, if you have LaTiO3 in your system, optimize > the > >>> basis set of La, Ti and O for bulk BiTiO3, then use the optimized basis > set > >>> for O in LaTiO3 with fixed rc's and optimize the other basis sets for > the > >>> bulk compound involving Bi. The idea is to use the chemical environment > of > >>> each atom, which should not change much, and take advantage of it to > create > >>> a basis set that will have enough flexibility to perform well in an > >>> environment that will be a bit different, but not radically different > from, > >>> the "original" one. Worked well for my compound, where the chemical > >>> environment doesn't change much from the original ones - I have been > >>> checking that against all-electron calculations and things come out > very > >>> much fine. > >>> Cheers, > >>> Marcos > >>> > >>> On Tue, Aug 17, 2010 at 4:13 PM, Koch, Robert J <[email protected]> > wrote: > >>>> > >>>> Hello all, > >>>> > >>>> First let me warn you that this is a bit of an involved question, but > I > >>>> have spent some time on it and have been unsuccesfull in finding a > solution. > >>>> Thanks in advance for any help you can provide. This list has been a > >>>> critical factor in my success in this past summer internship. > >>>> > >>>> I am in the process of optimizing a basis for a complex inorganic unit > >>>> cell (Orthorhombic cell with a, b and c on the order of 5, 5 and 36 > >>>> Angstroms respectively). It contains a rare-earth and transitional > metal > >>>> perovskite phase between bismuth oxide layers. I eventually hope to > study > >>>> the band structure. > >>>> > >>>> As you might imagine, getting the SCF calculation to finish with a > simple > >>>> "confinement-energy" basis was a process in itself. I started with an > >>>> electronic temperature of 1000K and a mixing weight of 0.0001. > Periodic > >>>> restarts were necessary to shorten the steps between kicks. Once the > SCF > >>>> converged, I'd lower the electronic temperature and start again, > untill I > >>>> got to 0 K. > >>>> > >>>> Now I am running into serious problems with optimizing a basis for the > >>>> system. First, I would run the initial calculation on 6 nodes. As > far as I > >>>> can tell, the Simplex utility included with Siesta cannot be made to > drive a > >>>> program in parallel (adding an mpi execute command to the run_script > crashes > >>>> the optimization routine). So currently I am stuck running each > indivdual > >>>> job to calculate the basis enthalpy on one node, which looks to make > the job > >>>> about three weeks long. > >>>> > >>>> Also, I've found that I cannot use the above iterative restart > process > >>>> for completing the SCF calculation when optimizing my basis. That is, > >>>> restarts of jobs are not permitted using simplex. > >>>> > >>>> Has anybody encountered issues such as these? I can't imagine that > >>>> everyone optimizes their basis using one node per job. > >>>> > >>>> Also, how necessary is an optimized basis for what I am interested in? > >>>> Particularly, I want to be able to report the primary character of > the top > >>>> of the valence band and the bottom of the conduction band. > >>>> > >>>> Thanks so much for all of your help so far! > >>>> Rob Koch > >>>> > >>>> B.S. Candidate, Ceramic Enginnering > >>>> Inamori School of Engineering, Alfred University > >>>> > >>>> > >>> > >> > >> > >> > >> -- > >> Bin Shao, Ph.D. Candidate > >> College of Information Technical Science, Nankai University > >> 94 Weijin Rd. Nankai Dist. Tianjin 300071, China > >> Email: [email protected] > >> > >
