ROHF_GRAD_DESCENT

• WORK IN PROGRESS

This modules contains the basics to perform a fixed step (or fixed by part) gradient descent algorithm for minimizing the ROHF energy.

The details of the algorithms are given in the report.

---

There are three main files :

1. :file:`initialization.irp.f` contains the providers for : * the overlap matrice, its square root and inverse square root. * the routine |init_guess| providing initial densities. By defaut : Core Guess * the function |test_projs| that tests whether the projectors are in the right space. ie if : Tr(Pd) = Nd, Tr(Ps) = Ns, Pd^2 = Pd, Ps^2 = Ps, Pd*Ps = 0
2. :file:`energy_and_gradients.irp.f` contains the providers and subroutine to compute the energy and the projected gradient at each loop. It also contains the subroutine |retraction|.
3. :file:`ROHF_GRAD_DESCENT.irp.f` is the main file.

---

There are two other files, needed to obtain canonical MOs from final densities.

The procedure is such :

1. call $qp run extract_data to obtain the |H_core| matrix, |overlap matrix|, and the number of total, d and s orbitals. The last part of the file generates the |four-index tensor| but is commented since it is not needed for this particular task.

   These files are writen in a /data_dir directory.

2. Copy your file :file:`Pd.dat` and :file:`Ps.dat` in /data_dir, containing the converged density matrices.

   [ a b c; a b c

   The format is for M = d e f; M.dat = d e f

   g h i] g h i

   !!! By default, the routine ROHF_GRAD_DESCENT generate those files when it converges. !!!

3. run the julia code :file:`extract_mos.jl` (outside /data dir). It should generate the file :file:`new_orbitals.dat` in /data_dir

4. call $ qp run read_and_canonicalize_mos. Orbitals should be canonical.