1. 21 Mar, 2022 1 commit
  2. 20 Mar, 2022 2 commits
  3. 18 Mar, 2022 1 commit
  4. 17 Mar, 2022 2 commits
  5. 15 Mar, 2022 1 commit
  6. 14 Mar, 2022 2 commits
  7. 07 Mar, 2022 4 commits
  8. 06 Mar, 2022 5 commits
  9. 05 Mar, 2022 1 commit
  10. 04 Mar, 2022 2 commits
  11. 03 Mar, 2022 2 commits
  12. 02 Mar, 2022 2 commits
  13. 01 Mar, 2022 1 commit
  14. 28 Feb, 2022 2 commits
  15. 25 Feb, 2022 2 commits
  16. 24 Feb, 2022 4 commits
  17. 22 Feb, 2022 3 commits
  18. 19 Feb, 2022 2 commits
    • agrofe's avatar
    • agrofe's avatar
      SCF Refactor/Abstraction complete. Works with SingleSlater, and NEOSingleSlater · b0dbe94d
      agrofe authored and agrofe's avatar agrofe committed
      Cleaned up the input file options for CEDIIS to make it more intuitive
      
      Added unit tests for CEDIIS. All tests pass except for GHGGA_RESIDUE, which fails for QZ decomposition
      
      Added checks on CEDIIS input arguments, Added testing for ROHF, Allowed ROHF damping to be performed.
      
      Implemented the SCF abstraction for SingleSlater and NEOSingleSlater. The code compiles on my local machine and passes all of the tests.
      
      MPI Bug Fix for Newton-Raphson SCF
      
      Changed to use reference_wrapper for MOs/Debugged Conventional SCF for RHF,UHF,GHF
      
      Added Natural Orbitals to SS/Implemented SCF for NEO and RO/Most tests pass
      
      Added Unit tests for QuasiNewton SCF/Separated Orthogonalization into its own library/Finished debugging rebase
      
      Fixed MPI compilation bugs
      
      Fixed the MPI bug for SCF. The unit tests for most of the SCF_MPI pass.
      
      The bug was caused by the orthogonalization objects only being computed
      by the Rank 0  MPI process.  Thus, all of the orthogonalization
      functions(ao2orthoDen,ao2orthoFock,etc) need to be protected by ROOT_ONLY.
      
      Added MPI unit tests for QN_SCF. Separated 4C SCF tests.
      
      Changed CEDIIS switch to 0.05 from 0.1 to yield better convergence for
      O2 unit tests. Before the O2 results for UHF could be chaotic because of
      small changes in the d-orbitals.  This resulted in two possible
      solutions when only changing the number of threads. Now, it is more
      deterministic.
      
      Cleaned up the computation of the maximum of orbital gradient in
      Newton-Raphson SCF.
      
      Added formDensity in SAD guess to make the density N-representable
      before the Fock matrix build. Here, the natural orbitals of the density
      are calculated first.
      
      Updated the guess functions so that new orbitals are not generated to
      preserve the guess=readDen.  Cleaned up NewtonRaphsonSCF object, and
      separated the diagonal hessian approximation from the gradient.
      Added OrthoAOMO functionality to orthogonalization.
      
      Removed obsolete modifyorbitals/neoscf and added computation of 4c natural orbitals
      
      Cleaned up NewtonRaphsonSCF header.
      
      Modified SCF objects to satisfy comments from code reviewers.
      
      1. The const attribute was added to all getter functions and printing
         functions
      2. A more general interface to NewtonRaphsonSCF was implemented which
         got rid of the need to pass in nOcc/nC. Instead the clients tell
         the algorithm which orbitals are being rotated using a new data
         structure.
      3. Two functions were moved out of the NewtonRaphsonSCF object and
         added to the CQLinAlg library. These include the SVDInverse used
         for computing L-SR1, and the MatExp function to compute the
         matrix exponential used to compute the unitary rotation from
         an anti-Hermitian matrix.
      4. An alias template was defined in ModifyOrbitals for the major data
         structures that are passed between the SCF and the client (e.g.
         vector<shared_ptr<SquareMatrix<MatsT>>> -> VecShrdPtrMat). This
         makes the code much easier to read and hopefully code.
      5. The MO's and eigenvalues are now passed by reference instead of
         by value.
      b0dbe94d
  19. 08 Feb, 2022 1 commit