Add initial orbital guess for DMRG CI/SCF

lcyyork commented 2 months ago

Description

This PR adds DMRG initial orbitals based on localization (active space only) + reordering from Fiedler vector, which is controlled by keyword DMRG_REORDER_INIT_ORBS. This reordering appears effective to pseudo-1D molecules (see test case dmrg-ci-2). For less obvious molecules like N2 or Cr2, this option should be disabled and orbitals should be adjusted based on orbital characters (σ, σ, π, π, δ, δ*, ...). This reordering is also performed only once when entering Forte, which seems to be enough to have a good DMRG-SCF results for Cr2 and acene.

A test case (dmrg-ci-2) on H20 chain is added to test the effect of various orbital localization schemes and Fiedler reordering. I leave dmrg-ci-1 to the test case of PR #413.

Other changes that I considered to be minor includes:

The Cholesky orbitals are added directly inside Localize such that CholeskyLocal is removed.
I modified proc/orbital_helpers.py, although it is subject to reallocation to somewhere else in the future.
The infamous die_if_not_converged is reintroduced to ActiveSpaceSolver and ActiveSpaceMethod to speed up CI of MCSCF.
Improve printing for permanent dipole in ActiveSpaceSolver.

User Notes

[x] Orbital localization and Fiedler reordering for DMRG-CI.
[x] Add Cholesky orbitals inside Localize class.

Checklist

[x] Added/updated tests of new features and included a reference output.ref file
[x] Removed comments in code and input files
[x] Documented source code
[x] Checked for redundant headers
[x] Checked for consistency in the formatting of the output file
[x] Documented new features in the manual
[x] Ready to go!

brianz98 commented 2 months ago

@lcyyork is this PR now ready for review? Happy to take a look if it is!

lcyyork commented 2 months ago

@lcyyork is this PR now ready for review? Happy to take a look if it is!

I think it is. Maybe there will be other tweaks. But I am using it to run acene jobs.

fevangelista commented 1 month ago

I think this can be merged if tests pass. We need to do a follow up PR to move the DMRG-specific functions outside of the psi4 module. This function should go in the driver instead.

evangelistalab / forte