Top Page > Browsing

Energy landscape discontinuities Date: 2013/11/08 17:35 Name: Jure Varlec   <jure.varlec@ki.si> Dear OpenMX developers and users, First, let me express my thanks for a nice software package, and especially its documentation. That said, I hit a wall trying to optimize unit cell. As can be seen in the image[1], there are discontinuities in the energy landscape. These discontinuities throw off optimization algorithms. As you can see in the input file, the scf.Ngrid and scf.fixed.grid were specified, just as the manual recommends. I also noticed that the DC solver creates a much smoother landscape that the faster Krylov solver. I am willing to sacrifice precision in order to compute faster, but the discontinuities are a show stopper. Is there any way to avoid them? The input file was fixed for all points in the energy landscape, only Atoms.UnitVectors were changed. Each point represnets length and angle of a rhombohedral cell. Best regards Jure Varlec [1] http://lana.hamradio.si/~s56ex/omx_energy/

Page: [1]

Re: Energy landscape discontinuities ( No.1 ) Date: 2013/12/25 04:03 Name: T. Ozaki Hi, Although you mentioned that " the scf.Ngrid and scf.fixed.grid were specified", I did not see those keywords in your input file. If the discontinuity appear even though you really specified those parameters, I guess that it appears due to use of the order-N methods which may produce such a discontinuity due to sudden change of the number of atoms in truncated clusters. I am wondering why you don't use the conventional diagonalization method for such a small system. In that case, you will not see the discontinuity if you properly specify scf.Ngrid and scf.fixed.grid. Regards, TO

Page: [1]