[Wien] Regarding mBJ plus U
Gavin Abo
gsabo at crimson.ua.edu
Wed Mar 9 14:23:54 CET 2016
If you ran as you describe:
runsp_lapw -p -orb -ec 0.0001 -cc 0.0001 -i 60
runsp_lapw -p -orb -ec 0.0001 -cc 0.0001 -i 60
The first runsp_lapw should have ran the calculation from cycle 1 up to
cycle 60. The second runsp_lapw should have then removed the broyd
files and repeated the same calculation again from cycle 1 up to cycle 60.
Now, if you had did:
runsp_lapw -p -orb -ec 0.0001 -cc 0.0001 -i 60
runsp_lapw -p -orb -ec 0.0001 -cc 0.0001 -i 60 -NI
The first runsp_lapw should have ran the calculation from cycle 1 up to
cycle 60. The second runsp_lapw should have then kept the broyd files
(because of the -NI) and continued the calculation from cycle 61 up to
cycle 120.
However, your new description that the energy and charge convergence are
constant in every cycle is much clearer. So, I now don't think that
increasing the maximum number of iterations will solve your problem.
Of the few mBJ calculations that I have ran, I don't remember ever
having any problem just doing:
run[sp]_lapw ... -i 80 (with the default MSR1 in case.inm)
If I were to encounter a convergence problem, then I would restore my
pre-mBJ calculation and try starting with a few iterations of PRATT and
switch back to MSR1 [
https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg07403.html
]:
restore_lapw
run[sp]_lapw ... -i 15 (with PRATT in case.inm)
run[sp]_lapw ... -i 80 -NI (with MSR1 in case.inm)
If the mBJ instructions were written today, maybe PRATT would not be
mentioned in the UG mBJ procedure.
I believe the PRATT part of the UG mBJ procedure was written back in the
day when MSEC1 was the default (before MSR1 existed) [
http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg03783.html
]. I suspect that mBJ calculations starting with MSEC1 failed
frequently, such that PRATT for a few iterations was the workaround to
the MSEC1 problem.
There was one user in the past that reported that they had the same
problem with a mBJ calculation, where the energy and charge convergence
became constant in every cycle:
https://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg07150.html
It looks like it might have been because they used too large of a mixing
greed for PRATT. Since the thread of posts seem to stop, the user
likely solved it by using a smaller (but not too small) PRATT mixing greed.
Unfortunately, you did not run the exact command that Prof. Marks sent
and provide the output as it looks like it would have included
information about the mixing parameters that were used during the
calculation.
On 3/8/2016 9:37 AM, sikander Azam wrote:
> Dear Laurence Marks
> <https://plus.google.com/u/0/115978139881852385108?prsrc=4>
> Thanks for your reply, Sir, actually we did mBJ+U calculations for
> parent compound. We got the convergence. Then we produce the vacancies
> in the parent compound, for that when we run the mBJ+U we get this
> problem i.e. the energy is not converged.
> So this is the only worry that why our calculations is not converging.
> Firstly we give 60 cycles, then 80, but this also didn't solve our
> problem i.e. we didnt got the convergence.
>
>
> *Note:* By *grep :DIS *.scf* we get 0.21815 0.21815 0.21815
> ..........20 times
> and by *grep :ENE *.scf* we get 0.5014 0.5014
> .............................20 times
>
> with best regard
> Azam
>
>
>
>
> On Tue, Mar 8, 2016 at 6:31 AM, Laurence Marks
> <L-marks at northwestern.edu <mailto:L-marks at northwestern.edu>> wrote:
>
> Please run the command below and post the result; it will provide
> some information about what is going on (although it may not
> answer everything). You can either run it in a linux shell (under
> bash) or paste it to a file (e.g. ~/bin/Check, $WIENROOT/Check or
> $WIENROOT/Check_lapw), do a "chmod a+x file" then just do "Check"
> (or whatever name you used).
>
> tail -n 100000 *.scf $1 | grep -e :ADIST -e :DIR -e :MV -e GREED
> -e :FRMS -e :ENE -e :CHARG -e PRATT \
> -e :DIS -e "MIXING SC" -e ":RANK" -e PLANE | \
> grep -v -e "with 1.0" -e scheme -e CONTRIBUTION | \
> tail -n 50
>
>
> On Tue, Mar 8, 2016 at 8:17 AM, sikander Azam
> <sikander.physicst at gmail.com <mailto:sikander.physicst at gmail.com>>
> wrote:
>
> Dear Gavin abin
> Thanks Sir for the reply, but I ran it for 60 cycle two times
> with commands
> runsp-lapw -p -orb -ec 0.0001 -cc 0.0001 -i 60
> Regards
> Sikander
>
> On 8 Mar 2016 14:36, "Gavin Abo" <gsabo at crimson.ua.edu
> <mailto:gsabo at crimson.ua.edu>> wrote:
>
> How many cycles did the calculation run for and what did
> you set for the maximum number of iterations (i.e., the
> number that you set for the -i option)?
>
> If you did not use the -i option, just
>
> run_lapw or runsp_lapw
>
> The program will run up to 40 cycles by default.
>
> It is common for a mBJ calculation to not be able to reach
> converge within 40 cycles [1-3], which is why it suggests
> during the running of init_mbj that you increase the
> maximum number of interactions. The example the init_mbj
> script gives is to set -i to 80:
>
> run_lapw -i 80 (or runsp_lapw -i 80)
>
> [1]
> http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg06119.html
> [2]
> http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg09879.html
> [3]
> http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg02833.html
>
> On 3/8/2016 3:51 AM, sikander Azam wrote:
>
>
> Dear all
> I am doing calculations using mBJ plus U, but when the
> calculations complete, I get that energy is not
> converged. But when I check the SCF file I get a
> constant value of 0.52341000 for energy convergence
> and 0.345680000 values for charge convergence.
> So I don't understand on the logic. please help me
> that what should I do. Should I continue with these
> calculations and find the properties or some one will
> give me some suggestions.
> Thanks in advance.
> Regards
> Azam
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20160309/4fe43137/attachment.html>
More information about the Wien
mailing list