It is big challenge and time-consuming procedure to converge the SCF calculations for the large systems. For a similar case, It seems that the g09 (with defaults) reaches the convergence in smaller steps in comparison with GAMESS_US.
I have checked FDIFF, DIIS, SHIFT, DAMP.... but I am not able to find a standard method to do large systems calculations (about 200 atoms) with GAMESS.
It would be appreciable if anyone suggest a efficient programme to overcome.
I have faced to such problem too. The gaussian algorithms in scf calculations have been extensively improved compare to GAMESS.If you use DFT , I suggest ADF.
In my experience GAMESS for some reason needs finer DFT grid to converge, then other programs. Default grid is frequently not enough.
In gaussian 09 the default is SCF=TIGHT which means SCF(CONVER=8) which translates to 1.00D-08 in convergence on RMS density matrix and 1.00D-06 in convergence in energy change.
For Gamess the default is (or used to be) 1.00D-06 in energy. So in theory Gamess jobs should converge faster. If it's not the case it means (as Masoud suggested) that gaussian algorithms are simply better - in the sense that they converge in fewer steps.
As I have proposed there is not significant difference as CPU time between Gaussian and Gamess. Because of Gamess uses DIIS, while the equivalen in the Gaussian is GDIIS.
Please find below the CPU times for an example C2H4 computed at HF/STO-G level of theory.
Most probably the problem consists of not suitable preliminary optimization of the geometry at a lower level of theory. Even two subsequent optimizations at lower levels before
a final precise one would be suitable for the requested case, due to the relatively large size of the atomic data set of 200 atoms.
Gamess:
POINT 1 X Y Z (BOHR) CHARGE
0.000009 0.000006 0.000527 -0.00 (A.U.)
DX DY DZ /D/ (DEBYE)
-0.000003 -0.000001 -0.000000 0.000003
...... END OF PROPERTY EVALUATION ......
STEP CPU TIME = 0.02 TOTAL CPU TIME = 0.9 ( 0.0 MIN)
TOTAL WALL CLOCK TIME= 1.2 SECONDS, CPU UTILIZATION IS 72.52%
$VIB
IVIB= 0 IATOM= 0 ICOORD= 0 E= -77.0739547733
2.209097295E-05 9.327679542E-06 9.528467057E-07-1.569538173E-05-7.787761423E-07
-8.702235840E-07 1.061018695E-06 3.272413748E-06-6.153105356E-06-7.183160800E-06
-2.805177427E-06 5.625168995E-06-5.977571470E-06-3.619930194E-06-5.661771938E-06
5.704122386E-06-5.396209540E-06 6.107085174E-06
-2.950169122E-06-1.070583300E-06-3.905447555E-07
......END OF GEOMETRY SEARCH......
STEP CPU TIME = 0.00 TOTAL CPU TIME = 0.9 ( 0.0 MIN)
TOTAL WALL CLOCK TIME= 1.3 SECONDS, CPU UTILIZATION IS 71.65%
580000 WORDS OF DYNAMIC MEMORY USED
EXECUTION OF GAMESS TERMINATED NORMALLY Tue Sep 29 12:11:52 2015
DDI: 263224 bytes (0.3 MB / 0 MWords) used by master data server.
----------------------------------------
CPU timing information for all processes
========================================
0: 0.982 + 0.62 = 1.44
1: 0.00 + 0.00 = 0.00
----------------------------------------
ddikick.x: exited gracefully.
----- accounting info -----
Gaussian:
Dipole moment (Debye):
X= 0.0000 Y= 0.0000 Z= 0.0006 Tot= 0.0006
Quadrupole moment (Debye-Ang):
XX= -11.9477 YY= -11.8422 ZZ= -13.4623
XY= 0.0000 XZ= 0.0000 YZ= 0.0000
Octapole moment (Debye-Ang**2):
XXX= 0.0000 YYY= 0.0000 ZZZ= 0.0003 XYY= 0.0000
XXY= 0.0000 XXZ= 0.0015 XZZ= 0.0000 YZZ= 0.0000
YYZ= 0.0010 XYZ= 0.0000
Hexadecapole moment (Debye-Ang**3):
XXXX= -62.3649 YYYY= -23.8167 ZZZZ= -10.8725 XXXY= 0.0001
XXXZ= 0.0000 YYYX= 0.0000 YYYZ= 0.0000 ZZZX= 0.0000
ZZZY= 0.0001 XXYY= -12.2206 XXZZ= -12.6833 YYZZ= -6.2760
XXYZ= 0.0000 YYXZ= 0.0000 ZZXY= 0.0000
N-N= 3.374246449815D+01 E-N=-2.468296898586D+02 KE= 7.662601763857D+01
Test job not archived.
1|1|UNPC-UNK|FOpt|RHF|STO-3G|C2H4|PCUSER|29-Sep-2015|0||# OPT=GDIIS RH
F/STO-3G POP=(MINIMAL) TEST||Untitled-2||0,1|C,-0.6529350734,-0.002164
6669,0.0001014927|C,0.6529350714,0.002164665,0.0001014937|H,-1.2261497
733,-0.9202704471,-0.0003117394|H,-1.232233423,0.9121165498,-0.0002972
.............................................................
Job cpu time: 0 days 0 hours 0 minutes 11.0 seconds.
File lengths (MBytes): RWF= 6 Int= 0 D2E= 0 Chk= 15 Scr= 1
Dear Professor Ivanova,
At first thanks for your complete answer and time,
I have optimized my system using other level of theory,
but it should be mentioned that we are talking about same geometry as a input for both of codes, and I expect same behavior,
Furthermore, I faced this problem in the case of large number of atoms (~more than 100), and there is no difference for smaller systems
Thanks again
Did you try converging the system with GAUSSIAN first and THEN with GAMESS? It is like helping GAMESS with better input structure
Apart from the scf code, Gaussian consider polar coordinates while GAMESS use Cartesian code. If these things are same and runs in a single node, DIIS with same initial guess i mean huckel then it should take same time for both software. If it is not then prof Gordon would only the man can answer you.
- Badges
- Science topic
- Similar topics
- Engineering
- Chemical Engineering