[Wien] Problems in convergence due to application of mBj potential
Peter Blaha
pblaha at theochem.tuwien.ac.at
Thu Aug 30 08:53:58 CEST 2012
For sure, forces are NOT ok when you switch on spin-orbit coupling.
But please: think about the physics: InP is not that heavy and is a
semiconductor. You do NOT need SO for any structural relaxation as
it has negligible effects anyway.
(However, you may NEED SO for bandstructure properties, effective masses,....,
but calculate this afterwards with fixed positions, NEVER with MSR1a but with MSR1).
Furthermore, in your mail it sounds as if you mixed up various schemes
with WIEN2k.
When using MSR1 you do NOT optimize positions (only "electrons"),
MSR1a you ALWAYS will optimize positions (in fact, when you do not
have case.inM, it will be generated automatically !)
And for both schemes you use the command run_lapw !
Alternatively, you can use min_lapw (and MSR1, not MSR1a), which will
also optimize positions (and if case.inM does not exist, will generate
it automatically).
Both, MSR1a or min_lapw accept "constraint positions", when you execute
x pairhess -copy (creates case.inM)
edit case.inM and set some values to zero for constraining them.
x pairhess -copy (so that the constrains are honored in the hessian for PORT).
PS: At least for insulators (but usually even for difficult 3d-metallic systems)
I strongly recommend MSR1a over min_lapw, as it is almost always much
faster and more stable. The only problem left is to determine a unique way
when to stop MSR1a (this means: what stopping criteria one should use),
but when you accept some remaining forces (less than 5 mRy/bohr),
even this is not a serious problem and default criteria (or even a bit
larger value than 2.0 in case.inM) are sufficient and reasonable.
PPS: When one tries MSR1a together with mBJ, I'd first converge mBJ (at least
partly) with MSR1 (or PRATT and later on MSR1), and then remove case.in0_grr
(this will fix the value of "c" in mBJ and make things for MSR1a more stable).
Am 29.08.2012 19:03, schrieb Luis Carlos Ogando Dacal:
> Dear Prof. Laurence Marks,
>
> Let me answer you using your previous message.
>
>> First, a reminder. Using MSR1a with mBJ is a computational experiment.
>> It may not give reasonable results, or it might -- I do not know and I
>> am not sure that anyone does, yet. The "standard" method is not to
>> vary the atomic positions with mBJ, but this may not be optimal.
>
> Yes, I know. I just tested MSR1a as you asked the WIEN mailing list
> in a message you sent in August, 22. I will keep on the calculation
> using the standard and reliable methods. My report was just a little
> contribution to your efforts.
>
>> One thing to check, in your case what are the positions like in the
>> substrate? Assuming that you have made this thick enough (e.g. 20 au
>> or more) in the center the positions should be close to those with PBE
>> and/or the bulk. If they are wildly different this implies that the
>> bulk lattice parameters for PBE do not match well those from mBJ. (Of
>> course, if you have only used 2 layers or so of substrate you cannot
>> test this.)
>
> Actually, I did not optimized the structure using MSR1a. I did it
> in a previous calculation with MSEC1 and PBE with case.inM fixing the
> substrate positions. Only after that, I used mBJ to improve the gap.
> When I employed mBJ with PRATT (0.2 followed by 0.4 mixing factors) I
> got the convergence in the SCF cycle and a smooth decrease in ETEST.
> In the MSR1a test, I noticed abrupt changes in ETEST and after the
> last iteration I got the message "energy in SCF NOT CONVERGED".
> As I was not optimizing the structure when using MSR1a, I had not a
> case.inM file and all the atoms changed position during MSR1a
> calculations. Despite the fact that I used a thin substrate (I will
> improve this in a next step), the atomic positions changed only by a
> factor of 10^(-3) when compared to the previously PBE relaxed
> structure (It is important to remember that MSR1a did not converged
> and I used the last generated case.struct to compare with PBE relaxed
> structure)
>
>> When you say "it did not converge with MSR1a", what exactly do you
>> mean? It may be better to send the case.struct and case.scf files to
>> my email directly.
>
> I was talking about the message at the end of the MSR1a calculation
> ("energy in SCF NOT CONVERGED"). I have not calculated any physical
> property of the system.
> I will send you the files in another message.
>
>> If you fix some atoms in case.inM, then the ones which are not fixed
>> will move. While some people argue that this is OK, I have
>> reservations. If you fix all the atoms in case.inM then MSR1a will
>> crash on you -- you have to use MSR1 (or MSEC3).
>
> All the best,
> Luis
>
>
>
>> On Wed, Aug 29, 2012 at 7:50 AM, Luis Carlos Ogando Dacal
>> <ogando at ieav.cta.br> wrote:
>>> Dear Prof. Laurence Marks (and WIEN2k users),
>>>
>>> I would like to report my recent experience in using MSR1a with
>>> mBJ. I have a semiconductor system composed by In and P atoms that I
>>> relaxed using PBE. After that, I tried the mBJ SCF cycle using PRATT
>>> as recommended in section 4.5.9 of the WIEN Users Guide. I got the
>>> convergence after a long cycle (strict convergence criteria), but no
>>> convergence was obtained with MSR1a.
>>> If you want any detail of my system and/or calculation, just send
>>> me an e-mail.
>>> Another point, my system tries to simulate a substrate and a cap
>>> stressed layer. As a consequence, I need to fix the substrate atoms
>>> during the SCF cycles. I would like to know if this can be done with
>>> case.inM when using MSR1a (or any other way). I believe that fixing
>>> atoms leads MSR1a to behave like MSR1. Is this right ?
>>> All the best,
>>> Luis Ogando
>>>
>>>
>>>
>>> 2012/8/22 Laurence Marks <L-marks at northwestern.edu>:
>>>> I am sure Peter had a typo - I think he meant MSR1 not MSR1a. Whether MSR1a
>>>> is a good idea with mBJ is currently unclear; there was a recent discussion
>>>> of this, look in the email archives.
>>>>
>>>> A good topic where readers of this list could contribute is testing whether
>>>> MSR1a with mBJ is physically reasonable and reporting back.
>>>>
>>>> On Aug 22, 2012 8:47 AM, "Madhav Ghimire" <ghimire.mpg at gmail.com> wrote:
>>>>>
>>>>> Dear Prof. Marks,
>>>>> Thank you very much for replying immediately. I was just replying to
>>>>> your post.
>>>>> As in userguide of wien2k, it is suggested to edit the case.inm and change
>>>>> MSR1a with PRATT as MSR1a leads to convergence problems in mBj. Hence, I
>>>>> performed the calculations by changing MSR1a with PRATT.
>>>>> I will follow to what you suggest right now and report within a day or
>>>>> two.
>>>>> Please let me know more if I have to be cautious somewhere in the
>>>>> calculations.
>>>>> Thanks.
>>>>> Madhav
>>>>>
>>>>>
>>>>>
>>>>> On Wed, Aug 22, 2012 at 8:21 PM, Laurence Marks <L-marks at northwestern.edu>
>>>>> wrote:
>>>>>>
>>>>>> Can you send the case.scf file to me directly? I am curious why MSR1 does
>>>>>> not converge well for some mBJ and there are some things printed in
>>>>>> case.scfm which may explain.
>>>>>>
>>>>>> ---------------------------
>>>>>> Professor Laurence Marks
>>>>>> Department of Materials Science and Engineering
>>>>>> Northwestern University
>>>>>> www.numis.northwestern.edu 1-847-491-3996
>>>>>> "Research is to see what everybody else has seen, and to think what
>>>>>> nobody else has thought"
>>>>>> Albert Szent-Gyorgi
>>>>>>
>>>>>> On Aug 21, 2012 9:50 PM, "Madhav Ghimire" <ghimire.mpg at gmail.com> wrote:
>>>>>>>
>>>>>>> Dear wien users and developers,
>>>>>>> I am working on some 3d TM oxides. With a normal scf cycle with or
>>>>>>> without inclusion of U value, I got good convergence in energy and charge.
>>>>>>> This oxide material is reported to have a bandgap of approx. 0.3 eV. In GGA,
>>>>>>> I do not observe any gap. In the meantime even with very high value of U,
>>>>>>> the bandgap do not open up. Because of this, I tried to implement mBj
>>>>>>> potential (in order to find the bandgap) both with and without inclusion of
>>>>>>> U, but the energy and charge do not converge.
>>>>>>> Rather even for a large number of iteration (199), the energy and charge
>>>>>>> remains constant without convergence (shown below).
>>>>>>>
>>>>>>> For GGA without mBj the scf cycle smoothly converges as below:
>>>>>>> in cycle 22 ETEST: .0000236850000000 CTEST: .0038743
>>>>>>> in cycle 23 ETEST: .0000184300000000 CTEST: .0012996
>>>>>>> in cycle 24 ETEST: .0000174650000000 CTEST: .0006011
>>>>>>> in cycle 25 ETEST: .0000037600000000 CTEST: .0007451
>>>>>>> in cycle 26 ETEST: .0000016050000000 CTEST: .0001163
>>>>>>>
>>>>>>>> stop
>>>>>>>
>>>>>>> while with mBj+GGA, energy and charge convergence remains constant
>>>>>>> above cycle 103 and could not converge as below:
>>>>>>> in cycle 193 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 194 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 195 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 196 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 197 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 198 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>> in cycle 199 ETEST: .2112103950000000 CTEST: 2.0591251
>>>>>>>
>>>>>>>> energy in SCF NOT CONVERGED
>>>>>>>
>>>>>>> Does anyone have experienced this type of problems. If so, please let me
>>>>>>> know how it can be converged. I followed all the steps as described in
>>>>>>> previous wien mail and userguid but could not solve.
>>>>>>> Your help to solve this issue will be higly appreciated.
>>>>>>> Thanks in advance
>>>>>>>
>>>>>>> Madhav Ghimire
>>>>>>>
>>>>>>> --
>>>>>>> MANA, National Institute for Materials Science (NIMS)
>>>>>>> 1-1 Namiki, Tsukuba, Ibaraki, Japan
>>>>>>> Phone: +81-29-851-3354 (ex.4115)
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Wien mailing list
>>>>>>> Wien at zeus.theochem.tuwien.ac.at
>>>>>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> MANA, National Institute for Materials Science (NIMS)
>>>>>>> 1-1 Namiki, Tsukuba, Ibaraki, Japan
>>>>>>> Phone: +81-29-851-3354 (ex.4115)
>>>>>>>
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Wien mailing list
>>>>>> Wien at zeus.theochem.tuwien.ac.at
>>>>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> MANA, National Institute for Materials Science (NIMS)
>>>>> 1-1 Namiki, Tsukuba, Ibaraki, Japan
>>>>> Phone: +81-29-851-3354 (ex.4115)
>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> Wien mailing list
>>>> Wien at zeus.theochem.tuwien.ac.at
>>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>>
>>> _______________________________________________
>>> Wien mailing list
>>> Wien at zeus.theochem.tuwien.ac.at
>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>>
>>
>> --
>> Professor Laurence Marks
>> Department of Materials Science and Engineering
>> Northwestern University
>> www.numis.northwestern.edu 1-847-491-3996
>> "Research is to see what everybody else has seen, and to think what
>> nobody else has thought"
>> Albert Szent-Gyorgi
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
--
P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300 FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------
More information about the Wien
mailing list