This thread is locked.Only browsing is available.
Top Page > Browsing
queation about DIIS ,H matrix.
Date: 2007/09/21 20:21
Name: Liger Chen   <>

Hi :

The Newton method for structure opimize, be :

x_new = x_old + H^(-1)*g
Where x_new be the new atomic potitions waiting been calculated, x_old be the last output potitions, H be the Hessian, and g be the moving distance.

the well known DIIS algorithm , which introducts the alpha for acceleration of convergence and discuss last m steps :

x_new = x_old_m + (alpha)*g .

Obviously, The algorithm nelgects the H^(-1) term. So in some case the DIIS could not reach structure optimizing criterion. But in vasp , The DIIS performs so well.
May I ask why ?
thank you


Page: [1]

Re: queation about DIIS ,H matrix. ( No.1 )
Date: 2007/09/25 17:23
Name: T.Ozaki


I am not sure the implementation in the VASP code.
But, theoretically the DIIS method can fail to converge.
It is found that some examples are given in the VASP HP at

Best regards,


Page: [1]