CCL: GAMESS:SCF-Problem with DFT
- From: Gustavo Mercier <gamercier _ yahoo.com>
- Subject: CCL: GAMESS:SCF-Problem with DFT
- Date: Mon, 21 Dec 2009 08:47:37 -0800 (PST)
Sent to CCL by: Gustavo Mercier [gamercier::yahoo.com]
Hi!
Several points:
1. Normally if you go with DIIS then Damping and Vshifting is turned off. Look
at the REF.doc in the Gamess distribution where they give hints
for helping in convergence. They also make a suggestion in
the INPUT.doc under the SCF namelist where they suggest using ETHRESH
and
SWDIIS. I do not recommend the SWDIIS option because what it does is switch
from DIIS to the second order method (SOSCF), and this has not worked out as
well for me.
2. Set ETHRESH (option under SCF) to a large value like 2E+0. This
will make DIIS kick in early, and should stay on pretty much all the time.
3. Another thing is the number of points used for computing the functional
integrals. This is done numerically and is controled by the DFT namelist.
The program uses a low number of points in the early iterations, and then
switches to the full grid. I find that I have to start the full grid
earlier. I've also had to use larger than default values and go up to 434 and
590. I do work on Mn coordination complexes and they seem to be fairly soft with
a fairly flat potential energy curve. Read under REF.doc details about the
grids and expected accuracy in the energy gradients.
I would also recommend doing the geometry optimization with delocalized
coordinates:
$ZMAT DELOC=.T. AUTO=.T. $END
you also need to specify the number of variables in the $CONTRL namelist
using (I believe) NZVAR=3*N-6 where N is the number of atoms.
This will reduce the number of iterations and I find that it makes for a smarter
optimization.
On another note... if you intend to do thermochemistry, I've found that the
'approximate' hessian kicked out at the end of the geometry optimization is not
bad. For DFT the second derivatives are computed using numerical methods from
(mostly) analytic gradients. Seems tha the augmented hessian methods that are
used for default in geometry optimizations do a good job at getting the hessian
ok, or at least as good as the default one-point numerical differentiation
formula.
Hope this helps!
--
Gustavo A. Mercier, Jr., MD, PhD
gamercier,+,yahoo.com
Gustavo.Mercier,+,bmc.org
gumercie,+,bu.edu
617-414-6440
----- Original Message ----
> From: Rudolf Herrmann rudolf.herrmann__physik.uni-augsburg.de
<owner-chemistry,+,ccl.net>
> To: "Mercier, Gustavo, " <gamercier,+,yahoo.com>
> Sent: Mon, December 21, 2009 6:42:44 AM
> Subject: CCL: GAMESS:SCF-Problem with DFT
>
>
> Sent to CCL by: "Rudolf Herrmann"
[rudolf.herrmann:+:physik.uni-augsburg.de]
> Hello all,
> I'm using GAMESS(US), rev. R3 from Jan 12, 2009. I encountered a problem
> in obtaining SCF with DFT. Part of the input:
>
> $CONTRL SCFTYP=UHF MULT=2 RUNTYP=Optimize COORD=unique INTTYP=HONDO $END
> $CONTRL DFTTYP=PBE0 MAXIT=200 ICHARG=0 ECP=SBKJC $END
> $SYSTEM TIMLIM=80000 MEMORY=80000000 $END
> $SCF DIRSCF=.T. DAMP=.T. SHIFT=.T. DIIS=.T. SOSCF=.F. $END
> $BASIS GBASIS=SBKJC $END
> $STATPT OPTTOL=0.0001 NSTEP=200 $END
> $GUESS GUESS=HUCKEL $END
> $DATA
> 4*Fuenfring - Phthalo cctt als Cu-Komplex UHF
> C1
> C 6.0 -5.216897 -0.910406 -10.919197
> ....... (many atoms)
>
> And that is the result of the first SCF calculation:
>
>
>
NONZERO
>
>
>
>
> BLOCKS
> ITER EX TOTAL ENERGY E
CHANGE DENSITY CHANGE DIIS ERROR
> VIR. SHIFT DAMPING
INTEGRALS SKIPPED
> 1 0 -1036.5378101022 -1036.5378101022
1.230919053 0.821519137
> 0.000000000 1.000000000 6231081460
193615138
> * * * INITIATING DIIS
PROCEDURE * * *
> 2 1 -1041.6461370603
-5.1083269581 11.464555047 0.402615779
> 0.000000000 1.000000000 6236128017
193099683
> 3 2 -1002.7358262561
38.9103108042 11.474994560 1.302218004
> 0.000000000 1.000000000 6241249413
191679251
> 4 3 -1041.8568160646 -39.1209898086
0.195994971 0.402579497
> 0.000000000 1.000000000 6237469221
191939968
> 5 4 -1042.3054537170
-0.4486376523 0.226115870 0.343346341
> 0.000000000 1.000000000 6057043846
201451242
> 6 5 -1042.6171017767
-0.3116480597 0.279322397 0.266341119
> 0.000000000 1.000000000 6061097833
201455155
> 7 6 -1042.9966679031
-0.3795661264 0.106460306 0.081821934
> 0.000000000 1.000000000 6077977863
201147673
> 8 7 -1043.0591297051
-0.0624618020 0.046857045 0.011477052
> 0.000000000 1.000000000 5989046608
202974307
> 9 8 -1043.0669450821
-0.0078153770 0.011381853 0.005948448
> 0.000000000 1.000000000 5736168713
207005748
> 10 9 -1043.0697884590
-0.0028433770 0.013485505 0.004417363
> 0.000000000 1.000000000 5475629752
210127430
> 11 10 -1043.0723408038 -0.0025523448
0.017067266 0.003869563
> 0.000000000 1.000000000 5458702402
210313186
> 12 11 -1043.0751463423 -0.0028055385
0.014434645 0.003675831
> 0.000000000 1.000000000 5453595333
210460303
> 13 12 -1043.0777030610 -0.0025567187
0.019791837 0.003849315
> 0.000000000 1.000000000 5397573182
211084337
> 14 13 -1043.0813673308 -0.0036642698
0.004903916 0.004281832
> 0.000000000 1.000000000 5491928111
210122347
> CONVERGED TO SWOFF, SO DFT CALCULATION IS NOW SWITCHED ON.
> * * * INITIATING DIIS
PROCEDURE * * *
> 15 14 -1058.1568375748 -15.0754702440
1.625183076 0.052558260
> 6225076316 194161508
> 16 15 -1056.2538409178 1.9029966571
0.866868728 0.796456193
> 6203667374 195873448
> 17 16 -1058.5031727428 -2.2493318250
0.157468336 0.035593489
> 6097769151 200991564
> 18 17 -1058.2872778511 0.2158948917
0.070399380 0.134520819
> 6058543947 201342138
> 19 18 -1058.5442781186 -0.2570002675
0.113766853 0.013742123
> 5927217989 204237218
> 20 19 -1058.5522842356 -0.0080061170
0.023549798 0.007512045
> 5840741652 205391057
> 21 20 -1058.5530282008 -0.0007439652
0.017420717 0.010226036
> 5670929315 208075365
> 22 21 -1058.5559993807 -0.0029711799
0.047523393 0.009589402
> 5564367770 209481859
> 23 22 -1058.5585808178 -0.0025814371
0.029346400 0.024600824
> 5707543406 207871809
> 24 23 -1058.5575623133 0.0010185045
0.013103014 0.017673475
> 5514532287 210353312
> NEXT ENERGY RISES: SWITCHING DIIS OFF FOR 5 ITERS, NEW ETHRSH=
1.59E-02
> 25 24 -1058.5566503140 0.0009119993
0.065560927 0.026774000
> 5383825524 211393480
> 26 25 -1058.5245291667 0.0321211473
0.955346110 0.097075114
> 5824682851 206204169
> 27 26 -1045.4354021475 13.0891270192
19.549841437 1.414811739
> 6176465140 198256544
> 28 27 -820.5152084492 224.9201936983
183.173769694 2.570654172
> 6246989834 190654722
> 29 28 615.6159645678
1436.13117301703000.606626971 2.010533226
> 6254865291 185798793
> 30 29 1752.6067873608
1136.99082279303001.172518613 10.378186893
> 6256392203 177999716
> 31 30 1765.6511795332
13.04439217243095.179109875 2.823919783
> 6256414149 176971112
> 32 31 2820.6124946084
1054.96131507523095.984312651 2.819419434
> 6256313973 177970697
> 33 32 2012.2921032636
-808.32039134473958.491129374 3.341335928
> 6256340795 177974590
> 34 33 3111.0312907781
1098.73918751453958.190267766 2.732484925
> 6256322037 177808512
> 35 34 2059.9131221528
-1051.11816862534124.537686577 3.295129177
> 6256330123 177699772
> ....... and so on (oscillates).
>
> My question is: Everything seems reasonable until the program's decision
> to turn off DIIS for five iterations. How can I switch off this behaviour?
> I did not find any hint in the manual (maybe I have simply overlooked it).
>
> Thanks a lot for suggestions.
>
> Merry Christmas and lots of successful calculations in the New Year!
>
> R. Herrmann.
>
> --
> Dr. Rudolf Herrmann
> Institut für Physik
> Universität Augsburg
> Universitätsstr. 1
> D-86159 Augsburg
>
>
>
> -= This is automatically added to each message by the mailing script =-
> To recover the email address of the author of the message, please change
> the strange characters on the top line to the ,+, sign. You can also
> look up the X-Original-From: line in the mail header.
>
> E-mail to subscribers: CHEMISTRY,+,ccl.net or use:
> http://www.ccl.net/cgi-bin/ccl/send_ccl_message
>
>
>
> Subscribe/Unsubscribe:
> http://www.ccl.net/chemistry/sub_unsub.shtml
>
>
> Conferences: http://server.ccl.net/chemistry/announcements/conferences/
>
> Search Messages: http://www.ccl.net/chemistry/searchccl/index.shtml
>
> If your mail bounces from CCL with 5.7.1 error, check:
> http://www.ccl.net/spammers.txt
>