Hello everyone! I have been using MRCC to try and find excitation energies with the LR-UCCSDT/aug-cc-pVTZ scheme and I am looking to overcome issues with incorrect or unconverged excitation energies. I appreciate if anyone is able to review the details of the errors and my example input file to help me locate the problem.
The precise issue with the values is not constant. For two-atom systems, one of the following errors tends to occur depending on the choice of convergence thresholds, choice to use PGS and choice to use automatic restart rest=2:
- Excitation energies always converge to a value in the 10-20 eV range, usually around 14 eV, and come out in a random order (i.e. the third excitation energy may be weaker than the second and/or first).
- State energies are NaN's. This has only happened with rest=2.
- Excitation energies converge to meaningful, but very inaccurate values (differences of a few eV or more).
- Excitation energies do not converge even when 10000 CC iterations are allowed.
These issues have happened when I have used the following MRCC builds:
- Pre-compiled binaries of the 2023-08-28 software release with all patches up to 2023-11-01.
- Source-code release of MRCC 2023-08-28 with patches up to 2023-11-01 compiled with build-mrcc Intel -pOMP, with both Intel 2021.3.0 and Intel 2022.0.1 classic compilers and MKL.
- As above, but with maxit increased to 10000 and ndiis increased to 100 in MRCCCOMMON.
The calculations I am attempting fail even for builds which pass the test suite. I have attached an example of a calculation which fails to converge, using the software build with more iterations enabled and Intel 2022.0.1 compilers and MKL. I am happy to attach additional files for other error types if they would be helpful for finding the issue. Thank you very much!
