Hi,
For a Bluegene, I would not bother with a SIESTA 1.3f1 version, which is
an old legacy version. SIESTA 2.0 has a couple of bugs removed and is far
superior when it comes to scalability, which is key for a BlueGene system.
While I have successfully build SIESTA 2.0 and 2.0.1 using xlf9.1 for IBM
Power5 machines, I have never tried building SIESTA 2.0 on our Bluegene.
Please keep us posted how it goes.
Best wishes
Joachim
Dear all,
We are recently trying to compile siesta1.3f1p at the Bluegene in Juelich
(IBM BlueGene/L, Jon von Neumann Institute fur Computing Froschungszentrum
Julich), Germany. The compiler is IBM XL Fortran V10.1. But
the results (the total energies and Fermi energies) of our test
calculations (inputs coming with the Siesta package) show to be way off. I
attach the results of this bluegene
and our local machine (rama) for comparison.
The apparent difference as I can see is as follows:
1) exchange and correlation energy
2) harris energy
3) SCF falls into an energy close to E_harris (harris electron density)
for the bluegene.
