Closed vanderhe closed 4 months ago
@bhourahine If we are willing to accept the less tight regression criteria I think this PR is ready for review. My testing indicates that the modern Broyden converges considerably faster in some situations. I'm not happy with the current layout of the source tree, e.g. just dumping all the files in a single folder, but this I deal with in a separate PR.
Substitutes the existing Simple- and Broyden-mixer by a more recent implementation taken from DFTB+.
For the regression tests to pass I had to soften the tolerances of the regression infrastructure (abs. tol.: 1e-10 $\rightarrow$ 1e-08, rel. tol.: 1e-09 $\rightarrow$ 1e-07), which demands for further investigations/discussions. Actually, the current settings were simply far too tight...
To do: