Menu
▾
▴
Re: [Opium-talk] about the opium package
|
From: Eric J. W. <ew...@pa...> - 2004-07-29 15:54:48
|
I think that is not a hard and fast rule for transition metals. How do other psps do for the systems you are trying? Eric On Thu, Jul 29, 2004 at 11:34:29PM +0800, Wu Rongqin wrote: > Thanks. > I am testing some of the pps in the teste package and find that the produce larger lattice constant than experimental values using lda. General speaking lda tends to underestimate the lattice by 2-3%. So what can cause such a problem? This is also true in the prb publication introducing opium. There the copper is used but the lattice constant is a bit larger. > Rongqin > > -----Original Message----- > From: Eric J. Walter [mailto:ew...@pa...] > Sent: 2004??7??29?? 23:17 > To: Wu Rongqin > Cc: ej...@wm... > Subject: Re: [Opium-talk] about the opium package > > > Dear Rongqin, > > Well, this is where we start to get into gray areas. Personally, I > like to get less than 10 or for the total error, if possible. > Sometimes it is easy to get this done, sometimes it is not. > > The way that I usually progress is: > > 1) First I pick the reference state based on the target problem. > > 2) Then , I pick the largest rc's I can live with based on the problem > I am going to do. I.e. what is the largest rc I can pick such that > there is no overlap of pseudopotentials in the target calculation (the > solid). > > 3) Once I have done this, I pick the smallest qc's I can get away with > such that the convergence error (the error in the PS report) is ok. > > 4) Then I test for ghosts and transferability. If ok I am done, if > not ok, I can pull in rc's to make the transferability better. If > I have a ghost for all choices of local potential, I may have to > try a different reference state. > > Hope this helps, > > Eric > > > > On Thu, Jul 29, 2004 at 10:58:38PM +0800, Wu Rongqin wrote: > > Dear Dr Eric, > > In the test of excited state, what an difference between ae orbital eigenvalues and nl orbital eigenvalues is acceptable? This is a test of Al.param > > ### TC report ######################################## > > > > AE:Orbital Filling Eigenvalues[Ry] Norm > > ---------------------------------------------------------- > > 100 2.000 -110.9387148261 > > 200 2.000 -8.4832751122 > > 210 6.000 -5.7425682806 > > 300 2.000 -1.0948994753 0.8204921422 > > 310 0.000 -0.6738512277 0.9000537643 > > > > E_tot = -482.1875622936 Ry > > > > NL:Orbital Filling Eigenvalues[Ry] Norm Ghost > > ------------------------------------------------------------------ > > 100 2.000 -1.0936103881 0.8203602771 no > > 210 0.000 -0.6732005072 0.8981291718 no > > > > E_tot = -3.4576095285 Ry > > > > AE-NL:Orbital Filling Eigenvalues[mRy] Norm[1e-3] > > AE-NL- -------------------------------------------------------------- > > AE-NL- 100 2.000 -1.2890872401 0.1318651965 > > AE-NL- 210 0.000 -0.6507204399 1.9245925469 > > AE-NL- total error = 1.9398076801 2.0564577434 > > > > > > Thanks > > > > Rongqin > > > > > > -----Original Message----- > > From: Eric J. Walter [mailto:ew...@pa...] > > Sent: 2004??7??29?? 21:29 > > To: Wu Rongqin > > Subject: Re: [Opium-talk] about the opium package > > > > > > > > http://opium.sourceforge.net/ contains all of the documentation > > that is completed. > > > > This includes the installation and user's guide links. The examples, FAQ > > and references are still be written. > > > > Do you see it now? > > > > Eric > > > > > > > > > > On Thu, Jul 29, 2004 at 09:22:08PM +0800, Wu Rongqin wrote: > > > Thanks. > > > I cannot find the documentation. Can you send me a copy? > > > Rongqin > > > > > > -----Original Message----- > > > From: Eric J. Walter [mailto:ew...@pa...] > > > Sent: 2004??7??29?? 21:02 > > > To: Wu Rongqin > > > Subject: Re: [Opium-talk] about the opium package > > > > > > Dear Rongqin, > > > > > > I uploaded the first round of the opium documentation last night. I > > > announced this to the mailing-list. Perhaps you should join? > > > > > > The Opium potentialws may be more rough than the fhi98pp, this > > > does happen. What is important is that your convergence error: > > > > > > > > > ### PS report ######################################## > > > > > > Orbital Conv. error: [mRy/e] [mRy] [meV] Ghost > > > -------------------------------------------------------------------------- > > > 200 0.0166087183 0.0332174366 0.4519497983 no > > > 210 0.0362648745 0.0725297490 0.9868252589 no > > > > > > Tot. error = 0.1057471856 1.4387750572 > > > > > > > > > is small. This basically is saying that there will be about 1.5 meV > > > basis set error when this pseudopotential is used at the ecut it is > > > designed for (the largest qc^2). > > > > > > The optimized pseudopotential method allows one to pick an ecut (via > > > the qc's) and minimized this convergence error below this ecut. For > > > instance, I can pick all my qc's to be 7.07, this then makes the ecut > > > 50 Ry (7.07^2). If I have reasonable convergence error (like a few > > > meV), I know that this potential is well converged in plane waves at > > > 50 Ry. In TM potentials, there is no such control. In general, TM > > > potentials are harder because of this. > > > > > > I believe that the *.ini is all you need to generate the potential in > > > the fhi98pp. At this point, the potentials are just represented in > > > semi-local form, so no local needs to be chosen yet. Once you > > > start testing the potential, the ghost analysis requires the non-local > > > form, therefore a local potential needs to be selected. > > > > > > You shouldn't worry about messages like this: > > > > > > > note: for the local potential no bound excited d state > > > > is found, the ghost state analysis will assume > > > > a zero 1st excited state energy (variable eloc1) > > > > > > There are just some combinations of local and non-local potentials > > > that do not yield a bound state. Opium can have this problem as well. > > > It isn't a problem, it just means that the ghost testing couldn't > > > determine if your choice of local potential was ok. > > > > > > Hope this helps, > > > > > > Eric > > > > > > > > > > > > On Thu, Jul 29, 2004 at 12:02:15PM +0800, Wu Rongqin wrote: > > > > Dear Dr. Eric > > > > Thanks you very much for your help. When will the opium documentation be > > > > ready? > > > > 1 I found that opium generated potential within the core region seems > > > > not so regular and smooth as that from fhi98pp package. Does this > > > > matter? > > > > 2 what is the advantages of opium pps over fhi98pp pps? > > > > 3 would you please give me some time on fhi98pp? > > > > a cu.ini > > > > 29.00 5 3 8 0.00 : z nc nv iexc rnlc > > > > 1 0 2.00 : n l f > > > > 2 0 2.00 > > > > 2 1 6.00 > > > > 3 0 2.00 > > > > 3 1 6.00 > > > > 3 2 10.00 > > > > 4 0 1.00 > > > > 4 1 0.00 > > > > 2 t : lmax s_pp_def > > > > 2 1.5 0 t : lt rct et s_pp_type > > > > 0 2.0 0 t : lt rct et s_pp_type > > > > 1 2.0 0 t : lt rct et s_pp_type > > > > is the attacthed *.ini the only required input file for generate a > > > > pseudopotential? > > > > Are these ( z, nc, nv iexc rnlc n l f lmax s_pp_def lt rct et > > > > s_pp_type) all parameters that are required in *.ini file? > > > > Then I can type psgen -o * *.ini to produce the pseudopotential and > > > > other files. After this action the pseudopotential is generated and has > > > > nothing to do with ' which component is taken as local potential', > > > > right? > > > > So next just try pswatch -i * *.ini -l [0,1,2] to check which > > > > component is suitable for local potential? ( surely sometimes increasing > > > > r_cutoff necessary for remove a ghost state) and then write the > > > > component ( suitable for local potential) to the *.cpi file? > > > > > > > > I also encountered a problem. For example for the upper cu.ini. > > > > lmax=2. when I tried pswatch -cu cu.ini -l 2 it works ok. But -l 1 and > > > > -l 0 the cu.test file indicate something like this: > > > > & dftseq - no bound state found (ctp): e=>0 n l 3 2 > > > > note: for the local potential no bound d state is > > > > found, the ghost state analysis will assume > > > > a zero groundstate energy (variable eloc0) > > > > > > > > & dftseq - no bound state found (ctp): e=>0 n l 4 2 > > > > note: for the local potential no bound excited d state > > > > is found, the ghost state analysis will assume > > > > a zero 1st excited state energy (variable eloc1) > > > > can you tell me how when -l n (n/= lmax)? > > > > Sorry to trouble you so much. But I desperately need your help! > > > > > > > > Regards > > > > Rongqin |
