Hi Shao,

On 7 Jun 2009, at 14:17, Junping Shao wrote:

Hi,

I'm interested in the internal workings of the sparse and full diagonalization programs, because I want to parallelize and adapt it to my studies.

How do you want to parallelize it? Do you want to do the diagonalization of a single model in parallel or parallelize it by farming out various parameters to different nodes?

Could someone point out to me the code section in which the Lanczos is performed,

The Lanczos algorithm is in the file ietl/lanczos.h and ietl/tmatrix.h of the ALPS libraries. This is a C++ adaption of the Cullum & Willoughby Fortran code on netlib.

especially the parts where a)preconditioning according to lattice symmetries, and

There is no preconditioning but we build the matrix directly for the subspace corresponding to fixed momenta.

b)reorthonormalization

Reorthonormalization is not needed in Lanczos.

is done? (I'm a newbie at C++)

Also, I can't find the source code which corresponds to the fulldiag_mpi binary in the fulldiag tutorial directory. Would someone show me where it is, or kindly link me to it, if possible?

The fulldiag code is wherever you installed the ALPS applications to. The default location is ~/ALPS/bin

Don't hesitate to ask more questions

Matthias