TURBOMOLE Users Forum
TURBOMOLE Modules => Ridft, Rdgrad, Dscf, Grad => Topic started by: antti_karttunen on May 22, 2007, 09:01:29 PM
-
Hello,
I have encountered a cumbersome problem while running dscf. The details of the system: 80 atoms, S6-symmetry, 1520 basis functions (def-TZVP). Starting from the first SCF iteration, dscf prints out the following warnings at each iteration:
Symm. of e-repr. not preserved in 630 cases
Max deviation in the symm. blocks: 9.134290479551055E-008
WARNING
Antisymm. of e-repr. not preserved in 670 cases
Max deviation in the antis. blocks: 7.036113913727604E-008
WARNING
Symm. of e-repr. not preserved in 1517 cases
Max deviation in the symm. blocks: 9.134290479551055E-008
WARNING
Antisymm. of e-repr. not preserved in 854 cases
Max deviation in the antis. blocks: 8.136587343118647E-008
I have encountered such warnings previously with rimp2, but the results have still been OK. However, this time the calculation ends up in a disaster. After the energy and density converge, dscf calculates one more iteration, yields a reasonable (converged) energy, and prints the following message:
Deviation from symm. for 213 eu :
1.056785083619616E-005 > 1.000000000000000E-005
fatal error in redrep.
dscf ended abnormally
I guess the warnings and the error are related to the complex reducible E-representations present in the S6 point group. I have really learned to abhor them to date :). So, any tricks or tips how this error could be avoided? As a first measure, I tried to tighten $scftol from about 0.2E-12 to 0.1E-14, but it did not help. Of course lowering the symmetry would be a simple solution, but I tested C3-symmetry and encountered the same problem. I would not like to lower the symmetry to C1, because the optimization might end up into wrong minimum (this is a quite tricky system with many very closely lying minima).
With def-SVP basis the final deviation is 4.895235694579725E-008 and dscf ends normally (no warnings are given during the iterations in this case).
Architechture em64t, TM version 5.9.1
-
Hi,
interesting error - in former Turbomole times, the molecular orbitals have been orthogonalized with Schmidt, which sometimes was not accurate enough for e representations. Nowadays, Cholesky is used and that solved the errors with redrep - but obviously it survived in at least one case...
Could you please send the input to the Turbomole Support?
Regards,
Uwe