Re: [QE-users] Convergence with interstitials
Thank you. I will check those. Thank you so much for your time, and comments. Regards Tejas From: users On Behalf Of Mostafa Youssef via users Sent: Tuesday, February 2, 2021 02:12 PM To: Quantum ESPRESSO users Forum Subject: Re: [QE-users] Convergence with interstitials Dear Tejas, There are many tests to check and test here. But one thing that comes to my mind is the pseudopotentials you used. you could try to use another set of pseudopotentials (especially for N). Try the recommended set here: https://www.materialscloud.org/discover/sssp/table/efficiency. Also one needs to make sure that W lattice constant is close to experiment. Lastly, I would reduce the starting magnetization on W to 0.1 and on N to 0.3 or so. No rationale here, just intuition. Regards, Mostafa AUC ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Convergence with interstitials
Dear Tejas, There are many tests to check and test here. But one thing that comes to my mind is the pseudopotentials you used. you could try to use another set of pseudopotentials (especially for N). Try the recommended set here: https://www.materialscloud.org/discover/sssp/table/efficiency. Also one needs to make sure that W lattice constant is close to experiment. Lastly, I would reduce the starting magnetization on W to 0.1 and on N to 0.3 or so. No rationale here, just intuition. Regards, Mostafa AUC ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Convergence with interstitials
Thank you for your quick attention. I added starting_magnetization to both atoms. It behaved similarly (attached). I didn’t remove from W and add to N. Should that factor in? I had to reduce k-points to 444111 as at 888111 memory requirement was crossing 250 GBs. Regards Tejas From: users On Behalf Of Mostafa Youssef via users Sent: Monday, February 1, 2021 09:13 PM To: Quantum ESPRESSO users Forum Subject: Re: [QE-users] Convergence with interstitials Dear Tejas, Did you test giving an initial magnetic moment to N instead of W? For example: starting_magnetization(2)= 0.5 Regards, Mostafa AUC WN-M.relax.in Description: WN-M.relax.in WN-M.relax.out Description: WN-M.relax.out ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Convergence with interstitials
Dear Tejas, Did you test giving an initial magnetic moment to N instead of W? For example: starting_magnetization(2)= 0.5 Regards, Mostafa AUC ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Convergence with interstitials
Thanks Mostafa, I tried both options, but continue to have convergence issues. I repeated the check with up to 8 x 8 x 8 k-points to some sense (but not to the threshold) of convergence (0.0x Ry at about 100 iterations), but larger than that the memory requirements overshoot > 256 GB). CG diagonalization also doesn’t help. Ends in same large scf error divergence quickly. I have attached one of the runs’ “in and out” here. I am trying a finer k-point mesh 12x12x12 and a bit larger smearing with 3x3x3 cell now to see if as discussed in some other threads to avoid issues with occupation states. I have been perusing literature and see some where even 2 x 2 x 2 cells work for some (I do note that not many are for metallic systems), but there are studies on Fe, NI and W where even 2x2x2 or 3x3x3 super cell seem to work. Not sure what I am missing. Any thoughts? Regards, Tejas From: users On Behalf Of Mostafa Youssef via users Sent: Saturday, January 9, 2021 03:48 PM To: users@lists.quantum-espresso.org Subject: [QE-users] Convergence with interstitials Dear Tejas, I see two issues: 1) Your supercell size is too small. This corresponds to a high concentration of N. I would go for 4x4x4 conventional cells of W. That is 128 atoms. 2) A magnetic moment on N is needed since the odd number of electrons is because of N. Regards, Mostafa AUC WN.relax.in Description: WN.relax.in WN.relax.out Description: WN.relax.out ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Convergence with interstitials
Dear Mustafa, Thank you for the comment. I will check both options. I imagine with tungsten maybe large cells are needed. My main concern was since the same converges easily if I use Fe in place of W. Regards Tejas From: users On Behalf Of Mostafa Youssef via users Sent: Saturday, January 9, 2021 03:48 PM To: users@lists.quantum-espresso.org Subject: [QE-users] Convergence with interstitials Dear Tejas, I see two issues: 1) Your supercell size is too small. This corresponds to a high concentration of N. I would go for 4x4x4 conventional cells of W. That is 128 atoms. 2) A magnetic moment on N is needed since the odd number of electrons is because of N. Regards, Mostafa AUC ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
[QE-users] Convergence with interstitials
Dear Tejas, I see two issues: 1) Your supercell size is too small. This corresponds to a high concentration of N. I would go for 4x4x4 conventional cells of W. That is 128 atoms. 2) A magnetic moment on N is needed since the odd number of electrons is because of N. Regards, Mostafa AUC ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
[QE-users] Convergence with interstitials
Hello everyone, To study diffusion barrier using NEB, to prepare the initial and final states, I am performing "relax" calculations with interstitials inserted into a vc-relaxed pure bcc W cell structure. I have used k-points, cut-offs and cells size after verifying the convergence, but with interstitial added, I am having issues. I have tried "mv" and "mp" smearing options, and I am USPP pseuodopotentials. With pure cell, convergence is usually between 16 to 40 iterations. I tried the possibility of W being magnetic, and spin polarization option as well. With an interstitial atom added, it doesn't converge well. I have retried with finer k-point grids as well, but no luck. With the interstitial setup, I realize that the cell has odd number of valence electrons. Does the odd number of valence electrons in a cell matter? I have searched through the mailing list archives, but can't seem to find any related posts on treating interstitials. I have attached the input and output files, if someone could see what I am missing. Regards, Tejas R PhD Student WN.relax.in Description: WN.relax.in WN.relax.out Description: WN.relax.out ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users