Top Page > Browsing

scf.restart option Date: 2007/04/10 12:11 Name: Peter Dear Dr. Ozaki, I have noticed that using 'scf.restart on' doesn't help to achieve convergence quickly. In order to check it, I performed one calculation (Nomd) and achieve convergence at some level (1.0d-6) using 55 SCF steps. Then I 'restarted' and expected to see that convergence will be achieved very quickly. However, I see in *.DFTSCF file: SCF= 1 NormRD= 1.000000000000 Uele= -537.004048407343 SCF= 2 NormRD= 52.510284076431 Uele= -485.834372406457 SCF= 3 NormRD= 58.741730517761 Uele= -485.448945817214 SCF= 4 NormRD= 60.387830595028 Uele= -485.443743507890 SCF= 5 NormRD= 60.384375104995 Uele= -485.491538702194 SCF= 6 NormRD= 45.806870093147 Uele= -489.750088977022 SCF= 7 NormRD= 30.144613952982 Uele= -494.648133057033 SCF= 8 NormRD= 22.158101095890 Uele= -497.486055856183 SCF= 9 NormRD= 14.186001896701 Uele= -498.881583202137 SCF= 10 NormRD= 11.696943840861 Uele= -499.544913834972 SCF= 11 NormRD= 10.571498994659 Uele= -499.854091654455 SCF= 12 NormRD= 9.513790693403 Uele= -500.048040175641 ... It looks like 'from scratch' calculations. In the output file the line : ******************* MD= 1 SCF= 1 ******************* <Restart> Found restart files <Poisson> Poisson's equation using FFT... <Band> Eigenvalue problem... ... Am I missing something? I also found that using restart files with mixing method 'kerker' and mixing method 'rmm-diisk' leads to very unpredictable results (for example, magnetization on atoms disappears). So, it seems, it is better do not use restart files, which is opposite what manual suggests. Best wishes, Peter Inishon

Page: [1]

Re: scf.restart option ( No.1 ) Date: 2007/04/10 15:02 Name: Vasilii Artyukhov That seems strange, since everything was fine for me when I used the restart option. What are your mixing parameters? I think it might be that the mixing at the first iteration destroys your SCF. Re: scf.restart option ( No.2 ) Date: 2007/04/10 20:25 Name: T.Ozaki Hi, I wonder that you used a different 'System.Name' in the second calculation from that in the first calculation. Regards, TO Re: scf.restart option ( No.3 ) Date: 2007/04/10 22:26 Name: Peter Dear Vasilii and Dr. Ozaki, Thank you for your suggestions. I use the following settings scf.Mixing.Type Rmm-Diisk # Simple|Rmm-Diis|Gr-Pulay|Kerker|Rmm-Diisk scf.Init.Mixing.Weight 0.1 # default=0.30 scf.Min.Mixing.Weight 0.001 # default=0.001 scf.Max.Mixing.Weight 0.10 # default=0.40 scf.Mixing.History 7 # default=5 scf.Mixing.StartPulay 30 # default=6 scf.criterion 1.0e-7 # default=1.0e-6 (Hartree) scf.Mixing.EveryPulay 5 # default = 5 Also, I use LDA+U settings. I didn't change 'System.Name' in the second calculation. As a matter of fact, I only turn on 'scf.restart' option (you can see in my first post, that code reports 'found restart files'). Best wishes, Peter Inishon Re: scf.restart option ( No.4 ) Date: 2007/04/23 15:11 Name: T.Ozaki Hi, Thank you for showing a more details. I have noticed that this is a bug in DFT.c, and fixed it. Please replace the routines by http://www.openmx-square.org/bugfixed/07Apr13/patch3.2.2.tar.gz Thank you for your cooperation in advance. Regards, TO

Page: [1]