dipdip causes hang in thm9 or dos calculation
Posted: Wed Feb 15, 2012 2:36 pm
For systems where large numbers of q points are needed for good DOS or thermodynamical quantities (typically metals) adding dipdip 1 (dipole dipole interaction) to the calculation is usually not needed, unless you are in an intermediate case e.g. with a small gap system.
If dipdip is turned on, each q interpolation calls ewald9 to calculate the dipole contribution, and in this there is a call to the erfc function, which basically hangs my calculations when many q-points are involved (thmflag 1 or prtdos > 0).
Could someone more knowledgeable about phonons in insulators tell me what to do? Should we drop the dipdip (in principle the long range IFC will converge quite slowly) or can we use dipdip selectively in the interpolation to avoid an exploding numerical cost. Another possibility would be to approximate the erfc(), with a spline or just evaluating for arguments within a small range.
thanks
Matthieu
Machine details:
abinit 6.10.2
jade cines (running on front end) in france
x86_64_linux
intel 11.1 compiler
If dipdip is turned on, each q interpolation calls ewald9 to calculate the dipole contribution, and in this there is a call to the erfc function, which basically hangs my calculations when many q-points are involved (thmflag 1 or prtdos > 0).
Could someone more knowledgeable about phonons in insulators tell me what to do? Should we drop the dipdip (in principle the long range IFC will converge quite slowly) or can we use dipdip selectively in the interpolation to avoid an exploding numerical cost. Another possibility would be to approximate the erfc(), with a spline or just evaluating for arguments within a small range.
thanks
Matthieu
Machine details:
abinit 6.10.2
jade cines (running on front end) in france
x86_64_linux
intel 11.1 compiler