Main   News   Research   Members   Courses   Notes   Publications   Opportunities   Resources   Q-Chem   PSI   Group Area

Search the Sherrill Group

This page contains some workarounds for problems that may occasionally arise for Q-Chem. This information has not been thoroughly checked and should be used at your own risk. Send comments to sherrill@chemistry.gatech.edu.

Too-low SCF energy

Problem Description: The SCF energy crashes to a much lower (unrealistic) value, such as:

    1    -109.3648026460      3.38E-02  000000
    2    -109.4619036371      1.33E-03  000000
    3    -186.3415905859      1.93E-01  000000

Solution: Turn off the incremental Fock builds, i.e., set INCFOCK 0. It may also help to turn off variable threshholding, VARTHRESH 0.

Errors with large basis sets

Problem Description: For large basis sets with near linear dependencies, the default integral cutoffs of 8-9 may be too loose. One symptom of this problem would be negative eigenvalues of the overlap matrix.

Solution: Try changing THRESH to 11 or 12.

Solution: Turn off the incremental Fock builds, i.e., set INCFOCK 0. It may also help to turn off variable threshholding, VARTHRESH 0.

Memory errors

Problem Description: Error messages such as:

Q-Chem memory limit of 1497366528 bytes exceeded on allocation of 28992 bytes of double storage ...

Cause: I think there is a bug in libmdc/NewQAlloc.C which uses the number of words in REM_MEMORY_TOTAL and multiplies it by 1024 for some inconsistent reason. This causes an overflow because the default size of REM_MEMORY_TOTAL is 1E9.

Solution: This should have been fixed by CDS sometime in spring 2001; the code was checked in. If not, try reducing MEMORY_TOTAL by a factor of 1000, to something like 1000000.

Small imaginary frequencies

Problem Description: After an optimization, you get very small imaginary frequencies like 40i.

Cause: This is due to loose convergence criteria which the program uses by default.

Solution: A number of convergence options may be tightened, including SCF_CONVERGENCE, THRESH, and GEOM_OPT_TOL_GRADIENT. The convergence on the Z-vector solver for CIS or MP2 may be tightened by setting SET_CONV to a value greater than 6 but no greater than 8.