How is mpw computed - paral_kgb = 1
Posted: Mon Apr 06, 2015 8:31 pm
Hi,
I am using the version 7.10.2 of Abinit with parallelization over Kpt/bands/fft and I am encountering a warning I don't really know how to deal with. I am using the code under somewhat extreme conditions and in the log file I have a warning saying that my nband is higher than mpw and therefore I shouldn't expect any good results. I do understand that indeed, the hamiltonian matrix would have a lower rank than the expected number of eigenstates and I should have at best a degeneracy or a crash. But that's not what I get and the results seem pretty reasonable.
Furthermore, I don't understand how the mpw number is computed because, when I compute the number of G vectors based on the Ecut I use I have a gigantic number compared to the mpw written in the log file (15720 versus 31). But it seems commensurate if I multiply mpw by the number of processors used. Last, I have a number of bands of 256. So my questions are:
- how is mpw computed? is it different in the parallelized case?
- shall I worry about my results? or is it actually not mpw that is meaningful in determining the rank of the hamiltonian matrix?
Thank you very much in advance for your help.
Best.
I am using the version 7.10.2 of Abinit with parallelization over Kpt/bands/fft and I am encountering a warning I don't really know how to deal with. I am using the code under somewhat extreme conditions and in the log file I have a warning saying that my nband is higher than mpw and therefore I shouldn't expect any good results. I do understand that indeed, the hamiltonian matrix would have a lower rank than the expected number of eigenstates and I should have at best a degeneracy or a crash. But that's not what I get and the results seem pretty reasonable.
Furthermore, I don't understand how the mpw number is computed because, when I compute the number of G vectors based on the Ecut I use I have a gigantic number compared to the mpw written in the log file (15720 versus 31). But it seems commensurate if I multiply mpw by the number of processors used. Last, I have a number of bands of 256. So my questions are:
- how is mpw computed? is it different in the parallelized case?
- shall I worry about my results? or is it actually not mpw that is meaningful in determining the rank of the hamiltonian matrix?
Thank you very much in advance for your help.
Best.