I am optimizing a small organic molecule in a triplet second excited state using TD-DFT in ORCA 5.0. It is giving error termination in SCF gradient. How to get rid of it? Check the error below. Please provide solution.
ORCA finished by error termination in SCF gradient
Calling Command: mpirun -np 16 -machinefile molecule_b3lyp_6311plusgdp_trip_r4_tdopt.nodes /home/app/orca_5_0/orca_scfgrad_mpi molecule_b3lyp_6311plusgdp_trip_r4_tdopt.scfgrad.inp molecule_b3lyp_6311plusgdp_trip_r4_tdopt
Check for SCF gradient-logfiles
[file orca_main/maingopt.cpp, line 1235]: ORCA finished with an error in the gradient calculation
Here's a quote from the manual, maybe this is just your case:
----
The standard SCF implementation in ORCA uses the DIIS algorithm [142,143] for initial and an approximate second-order converger for final convergence [144, 145]. This approach converges quickly for most chemical systems. However, there are many interesting systems with a more complicated electronic structure for which the standard SCF protocol converges either slowly (“creeping”), converges to an excited state, or diverges. In those cases, a newly developed trust-region augmented Hessian (TRAH) SCF approach [146–149] should be used. The TRAH-SCF method always converges to a local minimum and converges quadratically near the solution.
You can run TRAH from the beginning by adding
! TRAH
to the simple input line if you expect convergence difficulties. Open-shell molecules notoriously have SCF convergence issues, in particular, if they are composed of many open-shell atoms. The errors of the electronic gradient or residual vector converge almost steadily below the default TRAH ccuracy
of 10−6.
Alternatively, TRAH is launched automatically if standard SCF (DIIS/SOSCF) fails shows converge problems (default) what is called AutoTRAH.
%scf
AutoTRAH true
end
Alexey N. Masliy Prasanta Bandyopadhyay
I read in the Orca forum that it is a type of memory issue. Yet, as a beginner in ORCA, I did not understand which input should work. Here is my input. Please assist.
! B3LYP 6-311+G(d,p) opt NumFreq nososcf
%pal
nprocs 16
end
%scf
MaxIter 1500
AutoTRAH true
end
%geom
MaxIter 1500
end
%tddft
nroots 5
Iroot 4
Irootmult 3
maxdim 5
end
*xyz 0 1
...
*
I do not think you need that much of SCF iteration or GEOM optimization cycle. SCF iteration within 150 is good. Look at the geometry optimization cycle; if it is converging then you may want to decrease it. Now, coming to your main point of memory management,
Try this,
%maxcore 3000
For details, look at this page,
https://sites.google.com/site/orcainputlibrary/orca-common-errors-and-problems
BTW, why are you using Numfreq? Is Anfreq (analytical frequency) not available?
Prasanta Bandyopadhyay isnt numfreq faster? Also, if I use spectra, numfreq may be necessary. I have not had a clear idea in orca yet.
Are you suggesting that there is numerical noise in the system? Will RIJCOSX or COSX type grid work? Actually, I thought orca 5 does not have this issue, the site says grids are unclear there. No idea.
Enlighten.
Also, what is GTOInt?
Anfreq is faster, how faster, just do a simple calculation with small molecule. It is worth checking.
You can use RIJCOSx and COSx, these are cheaper approximation.
Why don't you share your input here? Do not give the coordinates.
There will be people who can assist you better in this case.
- Badges
- Science topic