[Octopus-users] Does the 'cg poisson solver' work?
nsai at physics.ucsd.edu
Thu Dec 23 19:36:06 WET 2004
Thanks for your reply.
I am really asking about the different methods for solving the Poisson
equations. I would like to use the "cg" instead of the "fft" method for a
special reason. The "fft" method works fine for a smaller system, but I
can not use it for a large system because the double FFT box is so huge
and it requires too much memory.
Anybody has an idea if the "cg" method works and why doesn't it work in
the example calculation I provided below?
BTW there is something that seems inconsistent in the manual and the code:
In Poisson.F90, you can see that one has to use "PoissonSolver = 5" for
the "cg" method and "0" for the fft method with spherical curoff. The "3"
option is for the fft with no cutoff. However, in the manual, it says
PoissonSolver = cg or 0
fft or 3.
I wonder if this is a mistake that should be corrected in the manual.
> Try to play with the convergence criteria (MaximumIter,
> ConvAbsDens.....) to see if it works
>>Does anybody know if the "cg" method of the PoissonSolver work in any
situation? The eigenvalues start to look crazy at the second scf step
already. I tried a simple example as below, and it still does not work.
>>there any parameter that I should pay attention to in order to use this
> I think you mixed up "cg" for solving the Possion equation with "cg" for
solving the eigenvalue equation here. By the way does PoissonSolver = 5
> Good luck
>>verbose = 30
>>DebugLevel = 15
>>SystemName = "3D"
>>CalculationMode = gs_start
>>Units = "eVA"
>> "Na" | 0 | 0 | 0 | no
>>BoxShape = minimum
>>Radius = 4.6
>>Spacing = 0.3
>>PoissonSolver = 5
> Octopus-users mailing list
> Octopus-users at tddft.org
More information about the Octopus-users