Re: DMFTwDFT Not work for many atoms?

[Uthpala Herath]

Hello Franck, 

What is the KPAR value you have in your INCAR?

I think I'm figuring out what might be going on. 

Best,

Uthpala

On Mon, Aug 23, 2021 at 11:15 AM Uthpala Herath <ukh...@mix.wvu.edu> wrote:

Hello Franck, 

I was checking on the calculation again and this time I also received **** for the Fermi energy. 

I also noticed that the forces on the structure are huge!

Is this what it should be?

Best,

Uthpala

On Tue, Aug 10, 2021 at 4:43 PM Singh, Vijay Ramdin <vsin...@uic.edu> wrote:

No problem, Uthpala. That would be great, thank you!

---

From: Uthpala Herath <ukh...@mix.wvu.edu>
Sent: Tuesday, August 10, 2021 3:20 PM
To: Singh, Vijay Ramdin <vsin...@uic.edu>
Cc: Franck <franck...@gmail.com>; Park, Hyowon <hyo...@uic.edu>
Subject: Re: DMFTwDFT Not work for many atoms?

 

Hello Vijay,

I’m currently in Maryland helping my brother move for grad school so I am not sure if I can. But I’ll be back home by the end of the week so we can reschedule it then.

Have a great day! 

Thank you,

Best,

Uthpala 

On Mon, Aug 9, 2021 at 12:45 PM Singh, Vijay Ramdin <vsin...@uic.edu> wrote:

Dear Uthpala, 

Good morning.

Do you have some time after our meeting tomorrow to discuss Franck's molecules?

Let me know so accordingly I shall create a zoom link. 

Thank you and have a great week ahead.

Sincerely,

Vijay

---

From: Franck <franck...@gmail.com>
Sent: Monday, August 9, 2021 10:20 AM
To: Singh, Vijay Ramdin <vsin...@uic.edu>
Cc: Uthpala Herath Mudiyanselage <ukh...@mix.wvu.edu>; Park, Hyowon <hyo...@uic.edu>

Subject: Re: DMFTwDFT Not work for many atoms?

 

Dear Vijay,

thank you for you message;

Do you have any ideas?

Franck O.

On Tue, Aug 3, 2021 at 4:41 PM Singh, Vijay Ramdin <vsin...@uic.edu> wrote:

Dear Franck and Uthpala,

Thank you for the message. 

I can also try if time permits and we can sit together and see if we can solve the problem.

Let us hope for the best.

Thank you.

Sincerely,

Vijay

---

From: Franck <franck...@gmail.com>
Sent: Tuesday, August 3, 2021 7:20 AM
To: Uthpala Herath Mudiyanselage <ukh...@mix.wvu.edu>
Cc: Singh, Vijay Ramdin <vsin...@uic.edu>; Park, Hyowon <hyo...@uic.edu>
Subject: Re: DMFTwDFT Not work for many atoms?

 

Hi,

Yes indeed it's the same message that I had and I thought that it's maybe a detail of the code that didn't take into account the energies of a certain order. or that simply didn't calculate well the fermi energy. because for a simple DFT calculation the fermi energy is well taken into account.

I tried both versions DMFTwDFT and DMFTwDFT_eb and nothing works.

Bests,

Franck O.

On Tue, Aug 3, 2021 at 4:36 AM Uthpala Herath <ukh...@mix.wvu.edu> wrote:

Hello Vijay and Franck, 

I was running it and in the self-consistent VASP step I get a garbage Fermi energy:

E-fermi : ********

I'm not sure what is going on here so I have to investigate it further. 

Did you get the same issue Franck when you run the non DMFTwDFT_eb version? 

The DMFTwDFT is the latest version on github. 

Best,

Uthpala

On Mon, Aug 2, 2021 at 10:43 AM Singh, Vijay Ramdin <vsin...@uic.edu> wrote:

Dear Franck,

Good morning. 

Yes, Uthapala attempted to run those calculations on his cluster, but he was facing some issues. 

Uthpala, if you have time, please let us know the situation of those calculations. 

Thank you and have a great week ahead.

Sincerely,

Vijay

---

From: Franck <franck...@gmail.com>
Sent: Monday, August 2, 2021 3:17 AM
To: Singh, Vijay Ramdin <vsin...@uic.edu>
Cc: Park, Hyowon <hyo...@uic.edu>; Uthpala Herath Mudiyanselage <ukh...@mix.wvu.edu>
Subject: Re: DMFTwDFT Not work for many atoms?

 

Dear Vijay,

I am writing to you to know if you were able to do the calculation as expected?

Best,

Franck O.

On Fri, May 21, 2021 at 3:44 PM Singh, Vijay Ramdin <vsin...@uic.edu> wrote:

Dear Frank,

Good morning.

Thank you for the message and the files. I shall try to run it and update you soon.

Have a fruitful day.

Sincerely,

Vijay

---

From: Franck <franck...@gmail.com>
Sent: Thursday, May 20, 2021 10:38 PM
To: Singh, Vijay Ramdin <vsin...@uic.edu>
Cc: Park, Hyowon <hyo...@uic.edu>; Uthpala Herath Mudiyanselage <ukh...@mix.wvu.edu>
Subject: Re: DMFTwDFT Not work for many atoms?

 

Dear Vijay,

thank you for your message. I did the NCSC calculation (attached below) and it worked really well. increasing the DMFT iteration from 1 to 10 does not change anything, the calculation turns but does not converge and when it does the DMFT + DFT calculation it crashes at the second iteration as when I had the DMFT iteration equal to 1.

It was suggested to me by sir Uthpala to take into account including the ibridation with the nitrogen atoms N closer (figure attached), I have done it and the only thing that has changed is the weather calculation longer and at the end I have the same error (I changed "n_tot": 5.4 ,; "atomnames": ["Mn"] ,; "orbs": ["d"] ,; "L_rot ": [0], to --->" n_tot ": 51 ,;" atomnames ": [" Mn "," N "],;" orbs ": [" d "," p "],;" L_rot ": [0, 0],)

Franck O

POSCAR.eps

(7,067 Ko)

 NCSC_MnPc_For_Vijay.tar.gz

Le mer. 19 mai 2021 à 01:06, Singh, Vijay Ramdin <vsin...@uic.edu> a écrit :

Dear Frank,

Sorry for the late reply.

Yes, I received it. I see that the eigenvalue is not correct for the charge SCF run. 

We need to run it on our platform and get back to you. Meanwhile, you can increase the DMFT iteration from 1 to 10 and see if it helps.

Have you done it without charge for self-consistency?

Have a nice evening.

Sincerely,

Vijay

---

From: OSIRISI <franck...@gmail.com>
Sent: Saturday, May 15, 2021 1:30 PM
To: Singh, Vijay Ramdin <vsin...@uic.edu>
Cc: Park, Hyowon <hyo...@uic.edu>
Subject: DMFTwDFT Not work for many atoms?

 

Dear vijay,

I attach here the complete file of my calculation. given that I have a molecule I used "q": [1, 1, 1], and I unitilized "Ndft": 50 it seems to me that after the 50 iteration, the calculation does not converge, and therefore the closed energy is also erroneous and since the energy window written in wannier90.in is a function of the closed energy which is itself false, the calculation stops there!

I remain at your disposal to bring other elements. you can answer me here or at f.ng...@yahoo.fr

Franck O.

 MnPc_for_Vijay.tar.gz

--

Uthpala Herath

PhD Candidate

Department of Physics and Astronomy

West Virginia University

Morgantown, WV 26505

Tel. (304) 216-2535

Email: ukh...@mix.wvu.edu

Website: https://uthpalaherath.github.io/ 

--

Uthpala Herath

PhD Candidate

Department of Physics and Astronomy

West Virginia University

Morgantown, WV 26505

Tel. (304) 216-2535

Email: ukh...@mix.wvu.edu

Website: https://uthpalaherath.github.io/ 

--

Uthpala Herath

PhD Candidate

Department of Physics and Astronomy

West Virginia University

Morgantown, WV 26505

Tel. (304) 216-2535

Email: ukh...@mix.wvu.edu

Website: https://uthpalaherath.github.io/ 

--

Uthpala Herath

PhD Candidate

Department of Physics and Astronomy

West Virginia University

Morgantown, WV 26505

Tel. (304) 216-2535

Email: ukh...@mix.wvu.edu

Website: https://uthpalaherath.github.io/ 

[Franck]

Hello Uthpala,

since I have a molecule I use a single point k and therefore the value and of KPAR is the default one so KPAR=1. the only value related to the parallelization in my INCAR is NCORE=1.

Franck O.

[Uthpala Herath]

Hello Franck, 

The reason I brought this up is cause I am seeing some discrepancies with parallelization based on the KPAR parameter. The charge conservation seems to be off when we use the KPAR != num. of processors. 

I know it's absurd for a large system like yours, but have you ever tested it on a single core?

Best,

Uthpala

[Franck]

I thought that setting NCORE=1 would solve this kind of problem. I will try now with one heart. even if this I think it will be extremely slow.

[Franck]

I did the calculation with only one core but it turns but I still have E-fermi : ******** .

I don't know if this is related to the way the electron.F file is modified etc... normally I have a negative fermi energy.

[Uthpala Herath]

Hello Franck, 

I see.

So when I'm running the calculation with KPAR != <number of cores> the number of electrons is not right. This can be seen in vasp.out.

DAV:  12    -0.208198652688E+05   -0.50294E+00   -0.72718E-03   408   0.523E-01BRMIX: very serious problems
 the old and the new charge density differ
 old charge density:    60.00178 new   60.00058

However, the NELECT for this case is 197 so there's something happening. But I checked the total number of electrons in the Wannier subspace as calculated with the DMFT occupancy matrix and that was around 8 (dmft-nkij_eig_vasp.dat dmft-nkij_eig_amn_vasp.dat) which is the correct value. 

In other materials (SrVO3) I noticed that using the KPAR value as the number of cores fixes the issue with the charge density. Here we probably can't do that given it is 1x1x1. 

Did you see a similar issue when not using the correct KPAR?

Best,

Uthpala

[Uthpala Herath]

So with the serial calculation does the issue with the *** Fermi energy disappear for you?

Yes, there seems to be something going on with the parallelization. 

Best,

Uthpala

On Wed, Sep 1, 2021 at 8:39 AM Franck <franck...@gmail.com> wrote:

Then if n_tot represents the number of electrons of the corrected layer (here the d-layer of manganese), then it is between 5 and 5.5 as it is in the OUTCAR file there is already a publication on the question of the filling of the d-layer of my molecule at the time of the DFT+DMFT calculation and they find effectively a value of 5.4

this is the paper attached

""this mBRMIX: very serious problems the old and the new charge density differ" this message is not an error.

I made several tests

1-) I restarted the calculation on a core it has been running for several 12 hours and produced only 4 itteration DFT+DMFT("Niter":80, "Ndft":100, "Nit":2,) except that at this speed the computer will spend several days with 1 core. the problem is that on our computer center when you make the request of a single core on a node that has 28, it is the totality that is counted in your consumption hours.

2-) when the launches it more than one knots it does not work.

I don't know how to manage it but it's certain that the problem is in the parallel calculation.

[Franck]

Yes the calculation runs normally. (I don't know if during the calculation it displays **** for fermi energy but still continues but one thing is for sure it runs very well on 1 core) this problem of parrallelisation I raised last December.

there is no way to solve this problem in the script DMFT.py or RUNDMFT.py

nobody else has an idea?

Franck O.