vibinabraham / fermicluster Goto Github PK

Implement the code using a sparse sigma vector form to avoid full matrix formation. Will be helpful for systems requiring larger config space.

The CI and PT energies are alright but the number of determinants are not exactly same.

Many of the test jobs are redundant and slow - we should go through these carefully to minimize the time it takes to run the tests, while also maintaining code coverage.

I keep seeing a memory leak on large systems where I do a PT2 correction, and then immediately go do another TPSCI calculation. I am able to do the corresponding calculation if I run the steps separately, but running sequentially in the same file crashes the system (I even filled up a 500gb node)

tools.py is huge - we need to separate these functions into files call:

• cmf.py
• matvec.py
• hamiltonian_build.py
• pt2.py
• maybe others too

and we also should rename all the bc_cipsi stuff to better names like tpsci and tucker_tpsci etc

Currently this gets crazy slow as the size of the vector gets large. Need to improve the code and parallelize it.

We have no documentation system, which we need to start

do test simplification first

Compute the 2rdm to get the gradient and avoid doing matvec. Will make the oo much faser and also can be used for a CAS-SCF extension.