Fix NOCI routine - Githubissues

GQCG / GQCP

The Ghent Quantum Chemistry Package for electronic structure calculations

https://gqcg.github.io/GQCP/

GNU Lesser General Public License v3.0

32 stars 10 forks source link

Fix NOCI routine #1033

Closed xdvriend closed 1 year ago

xdvriend commented 1 year ago

Short description

This PR will ensure the NOCI routines in the code are verified correct. The python bindings for the Biorthogonal basis will also be enabled.

Related issues

Some NOCI results dove below FCI. This PR will go through the code and verify everything to ensure the correctness of the code.

To do

[x] Verify RLowdinPairingBasis
[x] Verify ULowdinPairingBasis
[x] Verify GLowdinPairingBasis
[x] Verify evaluation of matrix elements in RNonOrthogonalStateBasis
[x] Verify evaluation of matrix elements in UNonOrthogonalStateBasis
[x] Verify evaluation of matrix elements in GNonOrthogonalStateBasis
[x] Add python bindings for R/U/GLowdinPairingBasis

xdvriend commented 1 year ago

All classes of the LowdinPairingBasis are now checked and verified against the non-orthogonal matrix element library by H. Burton.

The next step is to verify all routines used to calculate the matrix elements.

Tagging @guacke to update on progress.

xdvriend commented 1 year ago

Unrestricted matrix evaluation verified.
Generalised matrix evaluation checked with unrestricted examples.

I am working on a way to use the reference code with generalised basis states (this was no issue for the lowdown pairing, however it is to check the matrix evaluation). Once I can generate reference data I will:

Verify additional real and complex generalised matrix evaluation.
Verify the restricted algorithms. (This has a lower priority, optionally I can temporarily disable the restricted case and fix this at a later stage.)

Tagging @guacke to update on progress.

EDIT: Multiple basis states and complex basis states tested for matrix evaluation in G basis and verified.

xdvriend commented 1 year ago

All NonOrthogonalStateBasis classes have been checked with the reference code.
I am still looking for a test that calls the two zero algorithm in the restricted basis. If I find one, I will add it to the testing suite.
The NOCI testing suite fails, as it is not yet updated with the corrected code. I will perform all diagonalizations separately in order to provide reference data for these tests.
After the NOCI tests have been corrected, it has to be checked whether or not the bindings for the biorthogonal basis work.

If all of the above is done, this PR should be ready and we can merge it. Tagging @guacke to provide update on progress.

xdvriend commented 1 year ago

All NOCI code is now conform to the reference code and each of the test suites has been updated and verified with the updated code. As such, I expect the final build to succeed. (EDIT: If, at least, I don't make any typo's before pushing...)

I will now check the biorthogonal basis python bindings. If these work, this PR should be ready to be merged.

Tagging @guacke for progress update.

xdvriend commented 1 year ago

Bindings should now be enabled and an example notebook is provided. If the build succeeds, this PR is ready to be merged.

Tagging @guacke for final update/approval for merging.