Strange excitation energies from ADC(3)

Dear users and developers. I have just started my adventure with the ADC(3) method, and I have done some preliminary calculations with 2,2’-bithiophene, for which I have tested a number of correlated methods and have quite a good knowledge of what I should expect of a good theoretical model for excited states. And the ADC(3) model works fine both for the optically allowed Bu states and for the totally symmetric, dark Ag states. The contribution of the triply excited configurations is of particular importance for the latter, as at least some of them are spanned by the doubly occupied electronic configurations.

Yet my primary target at the moment is the so called TIPS-tetracene, which is a tetracene molecule substituted with 2 acetyl-triisopropyl silyl groups. For the sake of simplicity I have replaced the isopropyl moieties with hydrogens, which does not noticeably affect the chromophore properties. I have had high hopes for the ADC(3) method for its favourable scaling (N^6) and accuracy comparable to that of CC3 (which scales as N^7). Having overcome the technical issues related to the memory bottleneck the calculations went smoothly and fast, indeed. For such a big system and a reasonable basis set (not as large as one might want, but one cannot have everything :wink: - ccpVDZ - optimizations of 4 excited states took less than 20 hours on 12 cores.

To my surprise, however, the excitation energy to the lowest A1 excited singlet state (in the C2v symmetry group) went down the energy extremely far, dropping from 3.409eV (CC2) or 3.245eV (CC3) to 1.45 eV (ADC(3)), with the doubly occupied configurations constituting over 80% of the wavefunction. The red shift of the doubly occupied states can be expected when going from CC2 to ADC(3), but 1. I have never seen one so large, 2. the CC3 calculations did not reveal such strong changes (for the CC3 method - I have not been able to converge the calculations yet due to time/memory issues, yet the energies are already so stable that I believe that can be taken into considerations). CC3 shows that it is not the 2A1 state (the CC2 orderiing) that shifts down in energy, but the 3A1, but then it goes down only by some 0.7 eV, which is a very reasonable number. Nearly 2eV - this is quite a lot and I fear that this may indicate the failure of this method fir such a large, unsaturated system with a rather small HOMO-LUMO gap.

Another strange issue concerns the lowest triplet state (of B1 symmetry). Its energy obtained in both CC2 and DFT (not time-dependent, but regular DFT calculations) is about 1eV (B3LYP - 0.93eV, CC2 - 1.1 eV). The ADC(3) calculations, however, yielded the excitation energy of 0.389 eV. This is, again, surprisingly low a value, obtained for a state that is, according to qchem - a regular, singly occupied one, for which already CC2 and DFT are likely to produce rather good values of excitation energies.

So, finally (sorry, if it was overly long) I would like to ask, whether such a behaviour is known for the ADC(3) model, or has any of you come across similar problems. Perhaps it can be, somehow, avoided or circumvented? Any assistance will be most welcome.
Yours sincerely!
Marcin Andrzejak