Talk:DIIS

This is a notable mathematical technique. A case for a merge could be made (with what?). Please do not delete. Doug (talk) 19:06, 28 February 2012 (UTC)

How could this possibly not be notable? DIIS is arguably one of the most powerful and versatile fixed point convergence accelerators known to mankind. Amongst countless uses electronic structure theory, it also lies at the very heart of computational chemistry itself; because this (and nothing else) is what made actually /converging/ the Hartree-Fock and Kohn-Sham equations possible in the first place. Without this, there would be no density functional theory, no post-Hartree-Fock methods, no routine calculation of molecular properties. Additionally, DIIS predates similar techniques like GMRES invented by mathematicans, which aim to perform a similar task. No one is arguing that those techniques are "not notable".128.112.122.172 (talk) 15:38, 20 April 2012 (UTC)

I agree. Added a source (which was easy enough to find; many more could be added, I just used the first one that came up in a search of JCP). Since there is now at least one secondary source, I've removed the "not notable" tag David.S.Hollman (talk) 00:45, 20 December 2012 (UTC)

Umm Arnoldi iteration was invented 1950, Pulay in 1980. In function spaces, overcomplete just wins, and in discrete spaces, basis-free methods like Krylov or Arnoldi iteration are just more easily generalized. The problem with DIIS is that it relies too heavily on there being a smooth error function that is easy to compute. For numerical PDE except for global norms, there might not be any guide available in an information slice. 129.93.161.205 (talk) 01:22, 18 July 2025 (UTC)