Error in TD-DFT (setman/rpa_dia.C): Unable to diagonalize A-B

Currently I am using IP-tuned wB97XD3 with and without PCM on aggregates of perylene derivatives. For one of the systems the calculations crashes after the TDA calculation with the error:

 Direct TDDFT calculation will be performed
 Exchange:     0.1957 Hartree-Fock + 1.0000 wB97X-D3 + LR-HF
 Correlation:  1.0000 wB97X-D3
 Using SG-2 standard quadrature grid
 Dispersion:   Grimme D3
 Iter    Rts Conv    Rts Left    Ttl Dev     Max Dev

 Q-Chem fatal error occurred in module setman/rpa_dia.C, line 59:

 Unable to diagonalize A-B

The input options read as follows:

CIS_N_ROOTS      8
CIS_TRIPLETS     false
RPA              true
state_analysis    true
GUI = 2
MEM_TOTAL = 7000

Theory      CPCM

Dielectric 3
OpticalDielectric 3

Is it possible to use a different algorithm for the digonalization of A-B matrices?
Thank you

This could probably do with a better error message, rather than the default one that is thrown by BLAS, but when you see that the problem is almost always a triplet instability in the reference determinant. This manifests as a negative eigenvalue in the orbital Hessian (A-B), so the calculation is fundamentally ill-posed. The solution is to use the Tamm-Dancoff approximation (RPA=0, which is the default), as the B matrix is not needed and A is positive definite. Alternatively you can try a different density functional, cross fingers that you get a stable ground-state solution. Hartree-Fock theory is generally more susceptible to triplet instabilities (especially at stretched bond lengths) and functionals with larger fractions of exact exchange tend to inherent this sensitivity.

Thank you for the quick answer and sorry for the late reply. It was indeed a triplet instability. It just took me quite some more investigations to believe it, since I did not at all expect a triplet ground state for this systems. As it happens, I messed up my structure input.

Thanks again and sorry for occupying your time with my rookie mistakes.

No need to apologize, that error is rather cryptic if you’re not thinking about instabilities.