Hi Jeff,
On 4 Nov 2006, at 20:37, Jeff Hammond wrote:
Dear ALPS users,
I'm trying to get some benchmark numbers for the 2D Hubbard model. I'm figuring out the input from the tutorial and by reading the XML files and source code, but I wonder if there a more complete manual somewhere so I can optimize the input for conserved quantum numbers, etc.? I read the arXiv article and two Powerpoint presentations but haven't found what I'm looking for. For example, can I use SU(2) - | S^2 S_z > - configurations, or better yet, SO (4) - | S^2 S_z J^2 J_z > - configurations?
We have two XML definitions for the Hubbard model in the models.xml file.
One uses the quantum numbers Nup and Nodwn, the other the quantum numbers N, S and Sz on each site.
It was mentioned somewhere that 1.3 would have spatial symmetry built in the diagonalization code but I could not run the 4x4 Hubbard on a workstation, which should have been possible if the translation and C4v groups were utilized. What is the status of that development?
Given the current functionality, what should I need for the half- filled 4x4 Hubbard using Lanczos? I should have almost unlimited resources at NERSC but installation there is not going well and I'd rather use a 4 Xeon-core, 4 GB workstation if possible.
Translation symmetry is built into the 1.3 release, but not the point group symmetries such as C4v which only help for some momenta. The dimension of the Hilbert space using particle number, Sz and momentum conservation is about 10 million elements, which will require about 640 MB memory. However, the ALPS codes also store the matrix in memory, and that will require about 7 GB of memory - hence a 4 GB workstation will not be enough.
For the Hubbard model I thus recommend that you contact Andreas Läuchli at laeuchli@comp-phys.org He has an efficient Hubbard model code that will work well on your workstation.
Finally, we would like to help you solve the problems on the NERSC machine. Please let us know where you are stuck and we will try to help you solve it and can then make ALPS 1.3 install "out of the box" on the NERSC machine.
Matthias