Re: Memory allocation error in parallel
Posted: Thu Mar 19, 2015 1:52 pm
Dear Juanjo,
finally the two way are almost the same thing.
Your job crashed for memory reason, this was due to the enormous number of Gvectors (that qe needed for the description of the density).
In particular it crashed when trying to load the wavefunctions an by default it loads all the available gvectors.
The number of the stored gvectors components can be reduced.
You can do it at the beginning setting MaxGvecs, in this case the initialization database are done with the assigned number, next all the other calculation can be done using at maximum the number assigned in MaxGvectors in the setup, so you cannot use more than that. In this way it is not necessary to set the FFTGvecs.
In the second way, you consider all the Gvectors in the setup (so without defining MaxGvecs that by default is the density cutoff), and next in your specific calculation you limit the component to be considered when performing the FFT, so essentially the Gvectors that will be loaded for the wavefunctions. Please note that in the HF calculation you have also EXXRLvcs, in your case the values was extremely low and surely out of convergence. In the first case you cannot consider a number bigger than MaxGvecs, anyway the Gvectors of the wfs will be quite surely enough to have a converged results.
Hope it is clear,
Daniele
finally the two way are almost the same thing.
Your job crashed for memory reason, this was due to the enormous number of Gvectors (that qe needed for the description of the density).
In particular it crashed when trying to load the wavefunctions an by default it loads all the available gvectors.
The number of the stored gvectors components can be reduced.
You can do it at the beginning setting MaxGvecs, in this case the initialization database are done with the assigned number, next all the other calculation can be done using at maximum the number assigned in MaxGvectors in the setup, so you cannot use more than that. In this way it is not necessary to set the FFTGvecs.
In the second way, you consider all the Gvectors in the setup (so without defining MaxGvecs that by default is the density cutoff), and next in your specific calculation you limit the component to be considered when performing the FFT, so essentially the Gvectors that will be loaded for the wavefunctions. Please note that in the HF calculation you have also EXXRLvcs, in your case the values was extremely low and surely out of convergence. In the first case you cannot consider a number bigger than MaxGvecs, anyway the Gvectors of the wfs will be quite surely enough to have a converged results.
Hope it is clear,
Daniele