Fiddling with the mysterious inner workings of how to best represent atomic orbitals so they appear to behave just like in experimental observations...
It's like the simmering of a secret sauce where each ingredient must be mixed in the right amount and added at the right time. Otherwise dramatic explosions break out in the form of error messages like "Jacobian has collapsed", "5s is unbound", "x1 must be monotonically increasing", or "failed to bracket eigenvalues".
Now I know why Snape the potion master is perennially frustrated. At least my research group appreciated the Shakespeare reference and the clip-art of a bubbly cauldron.
