Artem, For a moment I thought I was completely hallucinating about my previous experience with graphene :) but then I remembered that if you don't set a variable explicitly in the input, the default values are used. I have just run a variable-cell calculation for graphene using your input and, indeed, if you don't set the target pressure, Siesta's default - 1 GPa, which is huge!!! - is too high, since the final pressure is 0.4 GPa, which is still too high. I set MD.TargetPressure and MD.TargetStressTol to 0.01 GPa and the lattice parameter comes out extremely bad - check it out. Actually, what I see is that, at some point, the c-axis length starts to diminish enormously (surely a spurious effect) and I didn't even bother letting the calculation finish. I think now this should prove my points.
Marcos On Thu, Apr 29, 2010 at 11:49 PM, Artem Baskin <abas...@uic.edu> wrote: > Dear Marcos, > > Thank you once again. > I recalculated the same jobs (systems) with MD relaxation parameters: > MD.TypeOfRun CG > MD.NumCGsteps 50 > to compare the results. The results are the same no matter with or without > relaxation steps (no changing in structural parameters at the end of > calculations compared to the initial ones). As to forces and stresses, > they were (and are) much below their respective tolerances. So the > convergence is reached very fast (within 17 - 22 SCF iterations). > > I was looking for answers to my questions in the SIESTA archive but still > I am stuck at the same. I still need some help. Could you push me to the > right direction? > > Thanks a lot, > > Artem Baskin, > PhD student, > University of Illinois at Chicago > > > On Wed, April 28, 2010 2:25 am, Marcos Veríssimo Alves wrote: > > Artem, > > > > Before proceeding to any other discussion: if this is indeed the complete > > input, then the results will be the same, since you don't specify the > > number > > of relaxation steps; Siesta uses the default value, which is zero. So > > nothing is moved, and all results are identical in what regards the > > positions of the atoms and that of the Fermi level. However, how do the > > forces and cell stresses look in each of the cases? Can you confirm that > > Siesta is indeed performing a non-zero number of CG steps and that, by > the > > end of your calculation, all forces and stresses are below their > > respective > > tolerances? > > > > Marcos > > > > On Wed, Apr 28, 2010 at 3:46 AM, Artem Baskin <abas...@uic.edu> wrote: > > > >> Dear Marcos, > >> > >> Thank you for your elucidative response. I learned a lot. I also > >> apologize > >> for my not well defined questions and for that I did not provide my > >> input > >> file for graphene. > >> Here it is. (that one that used initially) > >> > >> NumberOfAtoms 2 > >> NumberOfSpecies 1 > >> > >> %block ChemicalSpeciesLabel > >> 1 6 C # Species index, atomic number, species label > >> %endblock ChemicalSpeciesLabel > >> Meshcutoff 800.00 Ry > >> LongOutput T > >> WriteDenchar T > >> AtomicCoordinatesFormat Ang > >> %block AtomicCoordinatesAndAtomicSpecies > >> 0.000 0.000 0.000 1 > >> 1.42 0.00 0.000 1 > >> > >> %endblock AtomicCoordinatesAndAtomicSpecies > >> > >> %block PAO.BasisSizes > >> C DZP > >> %endblock PAO.BasisSizes > >> > >> LatticeConstant 2.459512147 Ang > >> %block LatticeVectors > >> 0.866025403 0.5 0.0 > >> 0.866025403 -0.5 0.0 > >> 0.0 0.0 60.0 > >> > >> %endblock LatticeVectors > >> > >> %block BandLines > >> 1 0.0 -1.33333333 0.0 > >> 30 0.0 0.0 0.0 > >> 30 0.577350269 1.0 0.0 > >> 20 0.0 1.33333333 0.0 > >> %endblock BandLines > >> > >> I did what you suggested. I used the the content of STRUCT_NEXT_ITER > >> files > >> to see the difference in structural parameters and here they are : > >> ---------------------------------------------------------------- > >> 1) MD.VariableCell .false.; no k-point sampling; Meshcutoff=800 Ry > >> > >> 2.129999998 1.229756074 0.000000000 > >> 2.129999998 -1.229756074 0.000000000 > >> 0.000000000 0.000000000 147.570728820 > >> 2 > >> 1 6 0.000000000 -0.000000001 0.000000000 > >> 1 6 0.333333333 0.333333334 0.000000000 > >> > >> FE=2.916573 > >> ----------------------------------------------------------------- > >> 2) MD.VariableCell .true.; no k-point sampling; Meshcutoff=800 Ry > >> > >> 2.129999998 1.229756074 0.000000000 > >> 2.129999998 -1.229756074 0.000000000 > >> 0.000000000 0.000000000 147.570728820 > >> 2 > >> 1 6 0.000000000 -0.000000001 0.000000000 > >> 1 6 0.333333333 0.333333334 0.000000000 > >> FE=2.916573 > >> ------------------------------------------------------------------- > >> 3) MD.VariableCell .true.;k-point sampling: > >> %block kgrid_Monkhorst_Pack > >> 21 0 0 0.0 > >> 0 21 0 0.0 > >> 0 0 1 0.0 > >> %endblock kgrid_Monkhorst_Pack ; Meshcutoff=800 Ry > >> 2.129999998 1.229756074 0.000000000 > >> 2.129999998 -1.229756074 0.000000000 > >> 0.000000000 0.000000000 147.570728820 > >> 2 > >> 1 6 0.000000001 -0.000000001 0.000000000 > >> 1 6 0.333333334 0.333333335 0.000000000 > >> FE=-3.788733 > >> ------------------------------------------------------------------- > >> 4) MD.VariableCell .true.;k-point sampling: > >> %block kgrid_Monkhorst_Pack > >> 6 0 0 0.0 > >> 0 6 0 0.0 > >> 0 0 1 0.0 > >> %endblock kgrid_Monkhorst_Pack ; Meshcutoff=800 Ry > >> 2.129999998 1.229756074 0.000000000 > >> 2.129999998 -1.229756074 0.000000000 > >> 0.000000000 0.000000000 147.570728820 > >> 2 > >> 1 6 0.000000001 0.000000000 0.000000000 > >> 1 6 0.333333334 0.333333335 0.000000000 > >> FE=-3.785969 > >> > >> Is it that info that you asked for? So, if it is that, you may easily > >> see > >> that there is no difference in cell vectors and atomic positions > >> regardless of the presence of the options you indicated. The only > >> significant difference is the Fermi Level (as it is expected). It is > >> also > >> interesting that the band structure looks very similarly in all these > >> cases but the crossing points (at the K-points) are shifted and do not > >> tally with the calculated level of FE (in case without k-point > >> sampling). > >> The fact that there is no difference in structure parameters may be the > >> simple consequence of that I used the initial geometry very closed to > >> equilibrium one. Is it so? > >> > >> Now I have more questions. > >> 1) Why do you want to compare these results? Are they indicative for > >> what? > >> 2) How to specify (what reasons or ideas to follow) the k-point sampling > >> in case of GNRs? > >> 3) Why I got the more or less correct band structure (merely shifted > >> with > >> respect to correct FE) for Zigzag NR without sampling and can not get > >> the > >> same WITH k-point sampling along that ribbon? > >> 4) Why I could not get the more or less reasoning BS for Armchair NR > >> until > >> I used > >> %block kgrid_Monkhorst_Pack > >> 45 0 0 0.0 > >> 0 1 0 0.0 > >> 0 0 1 0.0 > >> %endblock kgrid_Monkhorst_Pack ? > >> 5) How does k-point sampling influence the convergence of SCF loop? > >> 6) Why the doubling of unit cell for Armchair NR (with the same or > >> larger > >> k-point sampling along the ribbon) does not lead to convergence and any > >> reasonable results? > >> 7) What should I do now to circumvent all these hurdles? > >> > >> I'm looking forward for you instructive (enlightening) advice. > >> > >> Artem Baskin, > >> PhD student, > >> University of Illinois at Chicago > >> > >> > >> > >> > >> > >> > >> > >> > >> On Thu, April 22, 2010 7:35 pm, Marcos Veríssimo Alves wrote: > >> > Artem, > >> > > >> > Let's go step by step. The fact that you get "good" results with > >> graphene > >> > only at the Gamma point seems extremely strange (weird might not sound > >> so > >> > good, I acknowledge that) for the following. > >> > > >> > Graphene has an extremely small Fermi surface - actually, a *single* > >> > k-point, > >> > which is located at the high-symmetry point K of the Brillouin zone. > >> To > >> > accurately sum over the bands of graphene, you need a specific k-point > >> > sampling, which includes the coordinates of K in reciprocal space, as > >> well > >> > as a pretty dense k-point mesh in order to get the (linear, in the > >> > vicinity > >> > of K) dispersion of the bands in this region. > >> > > >> > Why is this important? Well, due to symmetry reasons, the Fermi level > >> lies > >> > * > >> > exactly* at this point. Thus, it is important to accurately sample the > >> > region of the BZ close to K, both including this particular point in > >> your > >> > sampling, *and* including enough k-points in order to accurately sum > >> over > >> > the occupied states and get a good total energy, which is the > >> variational > >> > quantity in DFT. Andf, as you may easily realize, Gamma-only sampling > >> is > >> > not > >> > enough for that. > >> > > >> > You do not provide enough details on your graphene calculation, > >> therefore > >> > I > >> > cannot know what you have done (this is the reason we always ask for > >> the > >> > input files. Repetitive? Yes, but... an image is worth a thousand > >> words, > >> > if > >> > you catch my drift). So, I propose the following test for you, in your > >> > graphene unit cell with two atoms. Start from ideal positions ((0,0,0) > >> ; > >> > (0.3333333333,0.3333333333,0.3333333333) in fractional coordinates) > >> for > >> > the > >> > C atoms, and the following cell: > >> > > >> > LatticeConstant 2.44 Ang > >> > %block LatticeParameters > >> > 1.00 1.00 10.0 90.0 90.0 60.0 > >> > %endblock LatticeParameters > >> > > >> > (check the convention for the angles between the lattice vectors in > >> > crystallographic format in the manual, but I think this would be the > >> > correct > >> > setting). Now, do two calculations: > >> > > >> > 1) one with fixed cell, and only atomic coordinate relaxation > >> > (MD.VariableCell .false.), and > >> > 2) one with MD.VariableCell .true. . > >> > > >> > For those, check the cell parameters and angle between the cell > >> vectors > >> at > >> > the end of the second calculation. > >> > > >> > After this, do a third calculation, starting from the same ideal > >> > coordinates > >> > and with MD. VariableCell set to .true., but now using > >> > > >> > %block kgrid_Monkhorst_Pack > >> > 21 0 0 0.0 > >> > 0 21 0 0.0 > >> > 0 0 1 0.0 > >> > %endblock kgrid_Monkhorst_Pack > >> > > >> > This is a sampling that, for the graphene cell described, includes the > >> > high-symmetry point K in the sampling, and has a rather dense k-point > >> > grid. > >> > As a final calculation, lower the Monkhorst-Pack grid to > >> > > >> > %block kgrid_Monkhorst_Pack > >> > 6 0 0 0.0 > >> > 0 6 0 0.0 > >> > 0 0 1 0.0 > >> > %endblock kgrid_Monkhorst_Pack > >> > > >> > which still includes the high-symmetry point K, but has a much less > >> dense > >> > grid. Do it still with MD.VariableCell .true., and look at the > >> > structural parameters. Use a 800 Ry cutoff in all of them, since this > >> will > >> > help avoid the infamous egg-box effect. Please tell me if you see any > >> > difference in the structural parameters from each of the calculations. > >> You > >> > can use, if you wish, an automatically generated DZP basis set with an > >> > EnergyShift 0.2 eV. An anticipation on the results: if there is no > >> > difference between these results, there is something wrong :) > >> > > >> > The fact that you use Stephan Roche's results as a benchmark is good - > >> his > >> > calculations are very reliable. Nevertheless, I would say that you are > >> > somehow inadvertently forcing symmetry in your system, which is the > >> cause > >> > of > >> > the - much likely fortuitous - agreement of whatever result you are > >> > obtaining in your calculations, with those of S. Roche. > >> > > >> > Some years ago, there were several posts on graphene on the list, with > >> > some > >> > very enlightening paricipations by Stephanie Reich. Search the list > >> > archives > >> > for these posts - you will learn a lot from them. I did, when I was > >> still > >> > working with graphene. > >> > > >> > Later on we can deal with different beasts like GNRs. > >> > > >> > Cheers, > >> > > >> > Marcos > >> > > >> > On Thu, Apr 22, 2010 at 11:32 PM, Artem Baskin <abas...@uic.edu> > >> wrote: > >> > > >> >> Dear Marcos, > >> >> Thanks for your help, but the situation looks really WEIRD or I don't > >> >> understand anything at all (that may be also true). > >> >> When I was calculating the BS for graphene and zigzag NR I did not > >> use > >> >> any > >> >> specific supercell (I did not use the supercell block, the naive > >> >> supercell > >> >> factors were 6*6*1 for graphene (initially my elementary unit cell > >> >> consisted of 2 atoms) and 6*1*1 for Zigzag NR (with 22 atoms - 5 > >> rings > >> >> width)). Of course, the SCF procedure converged in these two cases. > >> >> As to criterion to determine whether my calculations are good or not, > >> I > >> >> used the comparison both with tight-bonding method results and the > >> DFT > >> >> calculations that were already done (using Siesta) by Pr. Stephan > >> Roche > >> >> and co-workers. My results tallied with the those ones perfectly. > >> >> > >> >> As to convergence in case with the zigzag NR (with k-point grid), > >> double > >> >> unit cell of armchair NR and its flakes, the procedure did not > >> >> converged, > >> >> and I don't know how to have it converged. Moreover, form one step > >> of > >> >> the > >> >> iteration to another one the charges on the edge H-atoms are > >> fluctuating > >> >> getting more and more asymmetrical. This is another manifestation of > >> the > >> >> non-zero dipole moment. > >> >> > >> >> Best, > >> >> Artem Baskin, > >> >> PhD student, > >> >> University of Illinois at Chicago > >> >> > >> >> > >> >> On Thu, April 22, 2010 1:16 pm, Marcos Veríssimo Alves wrote: > >> >> > Artem, > >> >> > > >> >> > Good results for graphene without any k-point sampling (only Gamma > >> >> point)? > >> >> > And also for a zigzag NR? Now THAT's weird... unless you are using > >> a > >> >> huge > >> >> > supercell for each of them. What is your criterion for saying that > >> >> your > >> >> > calculation is good? For a zigzag NR you might get away with a > >> smaller > >> >> > supercell, if it's a semiconductor (I guess it is, I don't remember > >> it > >> >> > right > >> >> > now). > >> >> > > >> >> > What is your procedure for calculating the band structure? Are you > >> >> making > >> >> > sure you re-use the converged DM file? > >> >> > > >> >> > Marcos > >> >> > > >> >> > > >> >> > On Thu, Apr 22, 2010 at 8:06 PM, Artem Baskin <abas...@uic.edu> > >> wrote: > >> >> > > >> >> >> Dear all users, > >> >> >> I am involved in band structure calculations for armchair and > >> zigzag > >> >> >> nanoribbons and I have problems with the appropriate k-point > >> >> sampling. > >> >> >> My > >> >> >> problem is like this: > >> >> >> Initially, I calculated the band structure for graphene and zigzag > >> NR > >> >> >> without specifying any k-points. The results (with Meshcutoff > >> level > >> >> 800 > >> >> >> Ry > >> >> >> and 400 Ry respectively) were quite good. But when I tried to > >> >> calculate > >> >> >> the band structure for armchair (5 rings width) NR I got > >> absolutely > >> >> >> wrong > >> >> >> results, moreover, in spite of symmetrical configuration I > >> obtained > >> >> the > >> >> >> non-zero significant dipole moment perpendicular to the ribbon > >> (even > >> >> >> with 400 Ry MCutoff). > >> >> >> Then I generated a k-point grid using > >> >> >> %block kgrid_Monkhorst_Pack > >> >> >> 45 0 0 0.5 > >> >> >> 0 1 0 0.5 > >> >> >> 0 0 1 0.5 > >> >> >> %endblock kgrid_Monkhorst_Pack > >> >> >> and I got correct results (no dipole moment and reasonable FE and > >> >> band > >> >> >> structure). > >> >> >> > >> >> >> FIRST problem that I encountered was that when I tried to > >> recalculate > >> >> BS > >> >> >> for zigzag NR with the SAME k-point grid I got an enormous dipole > >> >> moment > >> >> >> (perpendicular to the ribbon) and wrong BS and FE level. Why is it > >> >> so? > >> >> >> > >> >> >> SECOND problem is that when I doubled the elementary cell for > >> 5-rings > >> >> >> width armchair ribbon (using the SAME k-point grid and 400 Ry > >> MCoff > >> >> >> level) > >> >> >> just to make sure that I will get the same good results I got > >> again > >> >> an > >> >> >> enormous Dip moment and incorrect Band Structure. > >> >> >> > >> >> >> And THIRD (mysterious) problem is that when I tried to calculate > >> Band > >> >> >> Structure for an isolated flake of armchair NR (3*5 rings size) I > >> got > >> >> >> non > >> >> >> zero Dip moment (in this case for x- and y- directions) that is > >> >> >> nonsense. > >> >> >> > >> >> >> I guess, the problem is how to define correctly k-grid in these > >> three > >> >> >> cases, but I don't know how. I performed the convergence test > >> >> increasing > >> >> >> the number of k-points but neither dipole moment nor FE level > >> >> >> demonstrate > >> >> >> the monotonic behaviour, so I could not reach the correct results. > >> >> >> > >> >> >> Any help will be appreciated. > >> >> >> > >> >> >> Artem Baskin, > >> >> >> PhD student, > >> >> >> University of Illinois at Chicago > >> >> >> > >> >> >> > >> >> >> > >> >> >> > >> >> > > >> >> > > >> >> > >> >> > >> >> > >> > > >> > >> > >> > > > > >
