bsenjean / openfermion-dirac Goto Github PK

Multiplicity can be set in the molecular object but does actually not appear anywhere in the DIRAC input file. How to tell DIRAC which spin-state one is interested in ?
This should be fixed.

Traceback (most recent call last):
  File "/disk-samsung/freebsd-ports/science/py-OpenFermion-Dirac/work-py39/Openfermion-Dirac-cc08a66/examples/H4.py", line 3, in <module>
    from openfermion.utils import eigenspectrum
ImportError: cannot import name 'eigenspectrum' from 'openfermion.utils' (/usr/local/lib/python3.9/site-packages/openfermion/utils/__init__.py)

OpenFermion-1.5.1
Python-3.9
FreeBSD 13.1

CASCI calculations are now wrong in some cases, like the LiH molecule. Before, we had:

Hartree-Fock energy of -7.8309055884430281 Hartree.
The spin-orbitals and their associated energy:{1: -2.3611878858249176, 3: -0.25010666181875163, 5: 0.07327903918143776, 7: 0.16210566454657324, 9: 0.1621056645465733, 11: 0.4326451477796373, 2: -2.3611878858249176, 4: -0.25010666181875163, 6: 0.07327903918143776, 8: 0.16210566454657324, 10: 0.1621056645465733, 12: 0.4326451477796373}
CASCI energies: [-7.86108778 -7.78953175 -7.78953175 ... 0.79376582 1.28714689
1.61687514]

And now:
Hartree-Fock energy of -7.8309055884430441 Hartree.
The spin-orbitals and their associated energy:b'{1: -2.361187885824919, 2: -0.25010666181875174, 3: 0.07327903918143933, 4: 0.16210566454657374, 5: 0.16210566454657382, 6: 0.4326451477796374, 7: -2.361187885824919, 8: -0.25010666181875174, 9: 0.07327903918143933, 10: 0.16210566454657374, 11: 0.16210566454657382, 12: 0.4326451477796374}'
CASCI energies: [-11.18701071 -11.08526707 -11.08526707 ... -0.20254019 -0.20254019
0.79376582]

This should be fixed easily, but I have to understand what happened. To be continued...

CC amplitudes are not yet fully working. I only managed to make them work for relativistic calculation, but the ordering does not match the one of OpenFermion-Psi4 or OpenFermion-PySCF. Some additional checks and fixes have to be made.

Now that we can use DIRAC21, I noticed some bugs that need to be fixed. For instance, the CCSD energy of the H2 molecule in the minimal basis is now wrong. Before using DIRAC21, one had the following energies (as shown in the tutorial on github):

Without NUCMOD:
Hartree-Fock energy of -1.0661086545096314 Hartree.
MP2 energy of -1.086665369123029 Hartree.
CCSD energy of -1.101150334318182 Hartree.

With NUCMOD:
Hartree-Fock energy of -1.0661086554855703 Hartree.
MP2 energy of -1.086665370107966 Hartree.
CCSD energy of -1.100909326556841 Hartree (not entirely converged with 30 iterations)

Now, we have:
Without NUCMOD:
Hartree-Fock energy of -1.0661086545096314 Hartree.
MP2 energy of -1.086665369123031 Hartree.
CCSD energy of -1.084147034280662 Hartree. (not converged)

With NUCMOD:
Hartree-Fock energy of -1.0661086554855708 Hartree.
MP2 energy of -1.086665370107968 Hartree.
CCSD energy of -1.210059543226092 Hartree.

The different is huge, and I am confident that the previous result is correct as it was extremely close to the FCI one.

I attach the 2 output files with DIRAC21 and with and without NUCMOD.
I also enclose the files where I used DIRAC20 to compare.

H2_STO-3G_R1.0_ccsd_NUCMOD.txt
H2_STO-3G_R1.0_ccsd.txt
H2_STO-3G_R1.0_ccsd_NUCMOD_DIRAC20.txt
H2_STO-3G_R1.0_ccsd_DIRAC20.txt