Parallel methods

As for direct methods, the entire Hartree-Fock and Kohn-Sham DFT parts of the DALTON program has been parallelized using MPI as message passing interface. The use of the parallel code requires, however, that the code has been installed as a parallel code, which is being determined during the building of the program as described in Section 3.3.

All that is needed to do of changes in the DALTON.INP file is to add the keyword .PARALLEL in the general input section, as demonstrated for a calculation of vibrational frequencies:

**DALTON INPUT
.RUN PROPERTIES
.PARALLEL
**WAVE FUNCTIONS
.HF
**PROPERTIES
.VIBANA
**END OF DALTON INPUT

The number of nodes to be used in the calculation is requested to the dalton run script after the -N option (see Section 5.4), or as stated in local documentation. Note that the master/slave paradigm employed by DALTON will leave the master mainly doing sequential parts of the calculation and distribution of tasks, thus very little computation compared to the N-1 slaves, see Ref. [90].

By default the two-electron integrals will be screened [89], using an increasingly tighter integral screening threshold during the SCF iterations. Thus, direct calculations calculated using integral screening will, when converged, be almost as exact as the results obtained with integral screening turned off. The thresholds can be changed with the keywords .IFTHRS and .ICEDIF. To turn of integral screening in direct or parallel calculations altogether, .IFTHRS should be set to 20 or larger.