The EOM-CCSD code has a few errors that yield poor convergence of the Davidson algorithm. This PR corrects those and improves the convergence from 70 iterations to 12-14 even with an order of magnitude tighter criterion in the excitation energies.
Changes
[X] Correct initial unit-vector guesses to correspond to roughly HOMO-LUMO singles rather than 1s core to LUMO singles.
[X] Correct a mathematical error in the preconditioner in which only a single element of the orbital energy denominators was used rather than the entire vector.
[X] Correct the conversion factor from Hartrees to eV. (An incorrect value of 22.211 was hardcoded; I've converted this to use the qcelemental entry.)
[x] Adjust the size of the guess spaces to force the algorithm to collapse when the maximum allowed dimension is reached. This is only to ensure that the relevant part of the code is tested and demonstrated.
Description
The EOM-CCSD code has a few errors that yield poor convergence of the Davidson algorithm. This PR corrects those and improves the convergence from 70 iterations to 12-14 even with an order of magnitude tighter criterion in the excitation energies.
qcelementalentry.)Status