Top Page > Browsing
Problems with MD restart and rmm-diisv
Date: 2021/08/02 21:55
Name: Pavel Ondracka   <pavel.ondracka@email.cz>

So I've recently started to use the rmm-diisv as it had superior performance for the core-hole calculations, now I tried it also for some MD runs. The strange thing is that the restart doesn't seem right. For start take the Gly_NH.dat example.

The first MD run with the default mixing (rmm-diis) looks like this:

SCF history at MD= 1
***********************************************************
***********************************************************

SCF= 1 NormRD= 1.000000000000 Uele= -16.244018526070
SCF= 2 NormRD= 0.450834596730 Uele= -15.486785282075
SCF= 3 NormRD= 2.778607595326 Uele= -11.211298777219
SCF= 4 NormRD= 3.308972202005 Uele= -11.152915776263
SCF= 5 NormRD= 3.287297214339 Uele= -11.151626196263
SCF= 6 NormRD= 3.266024307052 Uele= -11.150401386299
SCF= 7 NormRD= 1.350871560421 Uele= -12.549472975026
SCF= 8 NormRD= 0.719802536162 Uele= -12.286418713388
SCF= 9 NormRD= 0.571992127632 Uele= -12.153080261961
SCF= 10 NormRD= 0.127216017684 Uele= -11.919407359026
SCF= 11 NormRD= 0.073395503939 Uele= -11.974299589815
SCF= 12 NormRD= 0.029456207071 Uele= -11.951442418928
SCF= 13 NormRD= 0.013447725776 Uele= -11.938594265490
SCF= 14 NormRD= 0.005152167076 Uele= -11.938275316123
SCF= 15 NormRD= 0.003433464033 Uele= -11.932939108394
SCF= 16 NormRD= 0.000826664781 Uele= -11.935546571542
SCF= 17 NormRD= 0.000445565863 Uele= -11.936186286280
SCF= 18 NormRD= 0.000067854121 Uele= -11.936043550950
SCF= 19 NormRD= 0.000020282520 Uele= -11.935932346385
SCF= 20 NormRD= 0.000001975507 Uele= -11.935958390459
SCF= 21 NormRD= 0.000000410793 Uele= -11.935958227260
SCF= 22 NormRD= 0.000000053027 Uele= -11.935958152839

and the second one needs less steps due to the restart.

SCF history at MD= 2
***********************************************************
***********************************************************

SCF= 1 NormRD= 1.000000000000 Uele= -11.938292769246
SCF= 2 NormRD= 0.009244276365 Uele= -11.928476167512
SCF= 3 NormRD= 0.005345767428 Uele= -11.930039201708
SCF= 4 NormRD= 0.003509413811 Uele= -11.930819285040
SCF= 5 NormRD= 0.002491356661 Uele= -11.931227292448
SCF= 6 NormRD= 0.000603818813 Uele= -11.931545975308
SCF= 7 NormRD= 0.000050031100 Uele= -11.931734389221
SCF= 8 NormRD= 0.000021962205 Uele= -11.931733774154
SCF= 9 NormRD= 0.000002424051 Uele= -11.931736447354
SCF= 10 NormRD= 0.000000846808 Uele= -11.931736452813

If I change to the rmm-diisv, the convergence is faster in the first MD step

SCF history at MD= 1
***********************************************************
***********************************************************

SCF= 1 NormRD= 1.000000000000 Uele= -16.244018526070
SCF= 2 NormRD= 0.170603684126 Uele= -15.685583500909
SCF= 3 NormRD= 0.152037342941 Uele= -15.141971328172
SCF= 4 NormRD= 0.133528231755 Uele= -14.549840414055
SCF= 5 NormRD= 0.113677727344 Uele= -11.923450504783
SCF= 6 NormRD= 0.031098401332 Uele= -11.950802823416
SCF= 7 NormRD= 0.024475717194 Uele= -11.928488196414
SCF= 8 NormRD= 0.005911415820 Uele= -11.922206631451
SCF= 9 NormRD= 0.001470568167 Uele= -11.934507549455
SCF= 10 NormRD= 0.000512551580 Uele= -11.935212726971
SCF= 11 NormRD= 0.000148343565 Uele= -11.935826524159
SCF= 12 NormRD= 0.000051799286 Uele= -11.936154857675
SCF= 13 NormRD= 0.000011888924 Uele= -11.935899626492
SCF= 14 NormRD= 0.000006850317 Uele= -11.935955321126
SCF= 15 NormRD= 0.000000694657 Uele= -11.935959432507
SCF= 16 NormRD= 0.000000242967 Uele= -11.935958243581
SCF= 17 NormRD= 0.000000013538 Uele= -11.935958176646

however the second step does not seem to start from the old charge density and so takes much longer?

SCF history at MD= 2
***********************************************************
***********************************************************

SCF= 1 NormRD= 1.000000000000 Uele= 0.882981571544
SCF= 2 NormRD= 0.312320826996 Uele= -1.492964360096
SCF= 3 NormRD= 0.249057221546 Uele= -3.384871494301
SCF= 4 NormRD= 0.201911383171 Uele= -4.893599105098
SCF= 5 NormRD= 0.166248458600 Uele= -11.993198856904
SCF= 6 NormRD= 0.011873076588 Uele= -12.034887234258
SCF= 7 NormRD= 0.004213080176 Uele= -11.940504127606
SCF= 8 NormRD= 0.003687804244 Uele= -11.938487312303
SCF= 9 NormRD= 0.000531638193 Uele= -11.933694586283
SCF= 10 NormRD= 0.000085168118 Uele= -11.932515466583
SCF= 11 NormRD= 0.000045752572 Uele= -11.931749986210
SCF= 12 NormRD= 0.000011149088 Uele= -11.931734764035
SCF= 13 NormRD= 0.000002145140 Uele= -11.931726627568
SCF= 14 NormRD= 0.000000844652 Uele= -11.931736373442
SCF= 15 NormRD= 0.000000086764 Uele= -11.931736484789
SCF= 16 NormRD= 0.000000018390 Uele= -11.931736295446
SCF= 17 NormRD= 0.000000000939 Uele= -11.931736292237

Is this expected? BTW rmm-diisk seem to work OK
メンテ
Page: [1]

Re: Problems with MD restart and rmm-diisv ( No.1 )
Date: 2021/08/04 21:40
Name: T. Ozaki

Hi,

Thank you very much for your report.
In fact, this is a bug which will be fixed in an upcoming patch.

Temporary, one can correct the bug by changing the line 48 of Mixing_V.c:

if (MD_iter==1 && SCF_iter==1){

to

if (SCF_iter==1){

I have confirmed that it works for the example of Gly_NH.dat.

Regards,

TO
メンテ
Re: Problems with MD restart and rmm-diisv ( No.2 )
Date: 2021/08/04 21:40
Name: Pavel Ondracka  <pavel.ondracka@email.cz>

Just to be 100% sure, this should not have any negative consequences when I'm using rmm-diisv and with the core-hole final-state calculations (when restarting from the initial state), right? The MD_iter==1 condition should be fulfilled IMO.
メンテ
Re: Problems with MD restart and rmm-diisv ( No.3 )
Date: 2021/08/04 21:44
Name: T. Ozaki

Hi,

There is no negative effect by the modification for the line 48 of Mixing_V.c.

Regards,

TO
メンテ

Page: [1]

Thread Title (must) Move the thread to the top
Your Name (must)
E-Mail (must)
URL
Password (used in modification of the submitted text)
Comment (must)

   Save Cookie