About Achieving self-consistency in solving KS equation

[pcthu]

Why cannot simply taking the output density as the new input to the next?
I've read the paper by [P.H.Dederichs and R.Zeller],but more physics picture like answer I wonder to know. Thanks all

 

[kanato]

Usually with self-consistent algorithms like this, the output will have moved in approximately the right direction from the input, but much too far. So if you just take the output as the input for your next iteration, your density will wildly oscillate and never converge. Even if it does converge, it may go extremely slowly, taking thousands of iterations before converging. Implementing some intelligent mixing scheme like conjugate-gradient or Anderson mixing will greatly improve the convergence.

By the way, writing the paper by [author x] is not a proper citation, we need the journal name, volume and page number of the first page (year is good too) to know what paper you are talking about.