[FPLO] memory allocation error - again

Ormeci, Alim Alim.Ormeci at cpfs.mpg.de
Fri Aug 22 16:20:51 CEST 2008


Hello,

>>>>>>>>>>>>>>>>>>>
Trying to calculate a rather large structure (115 Atoms/cell, see attachment) 
I keep running into "memory allocation errors" like this one:

FATAL ERROR in MODULE SYMMETRY(sym_analysis):
 Could not allocate memory for cc !
>>>>>>>>>>>>>>>>>>>

I noticed that you are using nband = -1, meaning the size of Hamiltonian matrices,
(valence x valence and valence x core blocks) are determined from the total number
of basis functions that you use. Since you have "3d" functions for each B atom, you
are guaranteed to have a very large number for the sizes of these matrices. 

Now, from the output file read out the actual number of valence electrons in the unit
cell (search for MATR and go up a little bit) or you can calculate this number yourself
according to the =.basis file. Divide that number by two, and add let's say 10 or so.
Enter the obtained number as the nband value in the fedit menu. This should save you
enormous amount of memory!

Alim

PS. 2s, 2p and 3d per B atom means 9 basis functions per B atom. If you have, for example,
a calculation for hcp B, let's say, this will give an 18 x 18 valence-valence matrix. But you
have only 6 electrons per unit cell. At some k points only 3 eigenvalues will be sufficient
to occupy this many electrons! But, nband = -1 flag will use the full 18x18 matrix. For the
hcp case, of course this is peanuts. But, it can be a completely different story, if you
have one hundred B atoms in a unit cell !
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://biela.ifw-dresden.de/pipermail/fplo-users/attachments/20080822/20964bff/attachment.html 


More information about the fplo-users mailing list