CC2 Energy: wrong MP2 values leading to failure

[Jean Q.]

Hi,

I'm trying to run 'simple' CC2 energy calculations on 'smallish' systems (~100 electrons), starting with a small basis set (cc-pVDZ) but the direct coupled cluster energy program's initial iteration gives a incorrect MP2 value: -63693.1 a.u. vs -630.1 a.u. obtained with a separate MP2 calculation with the same basis set. The following CC2 iterations energies are out of format and the calculation stops after the following message: "ZERO PIVOT - MATRIX SINGULAR".

I'm using Dalton2016.1 compiled with intel 17.0.1, 64-bit integers, and MKL lib. I have the same issue with intel 14.0.2.

Thanks in advance,

Jean.

[Jean Q.]

More information:
if I use the Choleski decomposition, the MP2 energy is correct and the calculation finishes normally.

[sonco]

Hello,
I can try and have a look by the end of next week.
Sonia

[Jean Q.]

Hello,
Out of curiosity I tried to compute excitation energies with "RUN PROPERTIES" and surprisingly (to me at least), the CC2 energy converges properly and the calculation finishes normally.
Jean.

[kennethruud]

Hmmm, somewhat surprised myself. Could you try a Hartree-Fock calculation and see if the end result is the same. I am somewhat concerned it continues without an error/warning, using a Hartree-Fock wave function and density. If this is the case, then we need to insert a stop so that this does not happen.

Thanks and best regards,

Kenneth

[Jean Q.]

Looks like you're right, the excitation energies (and oscillator strengths) are the same for the CC2 calculation and HF calculation:
exc HF (eV) CC2 (eV)
1 4.709882 4.709882
2 5.334553 5.334552
3 5.701641 5.701640
4 5.799008 5.799008
5 6.130453 6.130453

(sorry for the bad formatting, I couldn't figure out how to make a table or tabulations)

[Jean Q.]

Since there seems to be an issue with the properties module, I also tried the response one. As expected, the calculation failed at the CC2 energy step, just like for the energy calculation (wrong MP2 then failure to converge).

I also have an excitation energies calculation with CC2 and Choleski running (input taken from the 2016 manual, p. 178). I'm not sure to fully understand that input, but it looks like it uses a specific module for CC linear response.

[Jean Q.]

CC2 Choleski excitation energies are different from the HF ones:
exc HF (eV) CC2/Chol (eV)
1 4.709882 4.05682
2 5.334553 4.55965
3 5.701641 5.05287
4 5.799008 5.17112
5 6.130453 5.89983

[Jean Q.]

Any news?

[Jean Q.]

A colleague has the same issue but with CCSD.

[Jean Q.]

The issue is still present in Dalton 2017.alpha