handyfloss

According to an e-mail by Txema, if you have a Gaussian job whose SCF does not converge, you can try the following:

scf(noincfock,conver=6,maxcyc=2025) iop(5/22=20)

Explanation:

• NoIncFock: Prevents use of Incremental Fock formation
• Conver=6: Request a mere 10^-6 convergence criterion (whereas the default is 10^-9, I believe).
• iop(5/22=20): Use DIIS extrapolation, with energy-based mixing. OK, OK, I have no idea what this means, and I use it out of superstition… yeah, sue me.

Other keywords with (in)direct impact on convergence: QC, VShift, FullLinear.