Linker IT Software
Google
Web www.oraxcel.com
menubar-top-links menubar-top-rechts
Home Help Search Login
Welcome, Guest. Please Login.
SQL*XL: Database to Excel bridge litLIB: Excel power functions pack ExcelLock: Locking and securing your valuable Excel spreadsheets encOffice: Protect your Excel file easy and safe encOffice: Protect your Excel file easy and safe
Pages: 1
MOLCAS SCF: Convergence problems (Read 8009 times)
Gerrit-Jan Linker
YaBB Administrator
*****




Posts: 75
MOLCAS SCF: Convergence problems
21.03.08 at 23:20:25
 
MOLCAS SCF: Convergence problems
 
Using the thresholds keyword you can specify in SCF lower thresholds for convergence. This allows you to play with the convergence thresholds. The manual says:
 
Specifies convergence thresholds. Four individual thresholds are specified as arguments, which have to be fulfilled simultaneously to reach convergence: EThr, DThr and FThr specify the maximum permissible diverence in energy, density matrix elements and Fock matrix elements, respectively, in the last two iterations. The DltNTh finally specifies the norm of the orbital displacement vector used for the orbital rotations in the second-order/C2-DIIS procedure. The corresponding
values are read in the order given above. The default values are 1.0d-9, 1.0d-5, 0.5d-6, and 0.2d-4, respectively. Note that these thresholds automatically define the threshold used in the direct Fock matrix construction to estimate individual contributions to the Fock matrix such that the computed energy will have an accuracy that is better than the convergence threshold.
Back to top
 
« Last Edit: 01.05.08 at 14:21:15 by Gerrit-Jan Linker »  

Gerrit-Jan Linker
Linker IT Software
Email WWW Gerrit-Jan Linker   IP Logged
Gerrit-Jan Linker
YaBB Administrator
*****




Posts: 75
Re: MOLCAS SCF: Convergence problems
Reply #1 - 01.05.08 at 14:23:15
 
Convergence problems can be overcome using a good set of initial orbitals. See GUESORB.
 
http://www.teokem.lu.se/molcas/wsh/lectures/lecture9.pdf
Back to top
 
 

Gerrit-Jan Linker
Linker IT Software
Email WWW Gerrit-Jan Linker   IP Logged
Gerrit-Jan Linker
YaBB Administrator
*****




Posts: 75
Using start orbitals in MOLCAS SCF
Reply #2 - 11.05.08 at 22:45:32
 
Using start orbitals in MOLCAS SCF
 
To use staring orbitals from a previous SCF calculation follow the procedure below.
 
Edit the job file to contain a statement similar to the one below before the molcas program is started:
 
!ln -fs $Project.ScfOrb.autosaved INPORB
 
Add the LUMORB keyword to the SCF input.
 
Another way to do it is not by making a symbolic link but to copy the file:
 
!cp myproject.ScfOrb.autosaved INPORB
Back to top
 
« Last Edit: 15.11.08 at 08:11:45 by Gerrit-Jan Linker »  

Gerrit-Jan Linker
Linker IT Software
Email WWW Gerrit-Jan Linker   IP Logged
Gerrit-Jan Linker
YaBB Administrator
*****




Posts: 75
Re: MOLCAS SCF: Convergence problems
Reply #3 - 09.06.08 at 10:40:10
 
UHF SCF calculations in Molcas seem to converge slowly. I have calculations where the maximum number of 200 iterations is not sufficient to lead to convergence. The SCF run can be restarted however it is required that all the files from the previous run are supplied.
 
A lazy approach to give extra iterations is to add an extra SCF step in the calculation. Only the restart input is suffient:
 
 &SCF &End
RESTART
End of Input
 
Note that you will also need to set the MOLCAS_TRAP environment variable to OFF so that MOLCAS will continue the job even after encountering errors.  
MOLCAS_TRAP=OFF; export MOLCAS_TRAP
Back to top
 
« Last Edit: 29.04.09 at 13:30:14 by Gerrit-Jan Linker »  

Gerrit-Jan Linker
Linker IT Software
Email WWW Gerrit-Jan Linker   IP Logged
Gerrit-Jan Linker
YaBB Administrator
*****




Posts: 75
Re: MOLCAS SCF: Convergence problems
Reply #4 - 10.06.08 at 23:11:46
 
Another strategy when trying to get convergence is to first try lower conversion thresholds. Once convergence is made at lower thresholds the obtained orbitals can be used as input in a run with lower thresholds eventually returning to the default values.
 
Part of SCF input:
 
Threshold
 1.0D-7 1.0D-3 1.0D-4 1.0D-3
End of input

 
The 4 numbers indicate:
EThr, DThr and FThr specify the maximum permissible
difference in energy, density matrix elements and Fock matrix
elements, respectively, in the last two iterations. The DltNTh finally
specifies the norm of the orbital displacement vector used for the
orbital rotations in the second-order/C2-DIIS procedure
 
Default values: 1.0d-9, 1.0d-5, 0.5d-6, and 0.2d-4
Back to top
 
« Last Edit: 10.06.08 at 23:19:38 by Gerrit-Jan Linker »  

Gerrit-Jan Linker
Linker IT Software
Email WWW Gerrit-Jan Linker   IP Logged
Pages: 1