Issues

In this section known problems with the code can be found

General

  • No checking of input files at all at any stages of the code
  • Basisset object has a bad structure
  • Loops should be replaced with np.einsum() in many cases
  • Documentation of integral code is out-of-date

Performance

  • Alot of recalculations in R, when calculating higher order angular momentum integrals
  • Not sure variables are passed around in integral code the right way with respect to Cython
  • The SCF seem to have trouble converging if BOMD time steps are above 1.0 a.u.
  • All information is stored directly in memory; Integrals scale as N^4 for memory

Need testing

  • DCPT2 not tested, no case in paper, where cartesian integrals was used (all had D orbitals)
  • Numerical Forces are not tested yet

Broken

  • Self overlap integrals is non-one for contracted basisfunctions
  • No check for singularity in DIIS, H2/STO3G breaks the code if used with DIIS
  • Changing Cython directive_defaults seem to make the code break sometimes, not very reproduceable