The ABINIT Users Mailing List ( CLOSED )

[paul.boulanger@umontreal.ca]

Hi Mark and Eric,

         I have been investigating the memory leak and I have found a few 
interesting points. First, to verify my intuition that the problem is in 
every SCF cycle, I have made a calculation of the Hydrogen molecule, with one 
dataset. I chose a ridiculously small ECUT and chose a convergence parameter 
such that the numerical error prohibits convergence. I put the nstep variable 
extremely high, a hundred thousand. The job began with less then 15 Mo of 
real memory and 60 of virtual memory. But, after about 20 minutes and a few 
thousands of SCF cycles, the job took 700 Mo of real memory and 853 Mo of 
virtual memory. Clearly, the memory leak is present in this job. We can then 
be sure that the problem is inside each SCF procedure. The files are 
available on demand. 

        With this in mind, I plunged my self in the source codes of PAW in 
Abinit 4.6.4. I found two bizarre statements that can be causing some memory 
leaks. 

•The first is in the file: Src_3paw/pawmkrhoij.F90   line 136

The variable pawrhoijnk_disk is allocated at this line. But is never 
deallocated in the rest of the file. The strange part is that in the 
declarations in the beginning of the file, this variable is defined as local. 
Then we expect that this variable should be deallocated in this file. Maybe 
this is causing some memory leaks.

•The second is in the file: Src_3paw/pawxcm.F90     line 335 , see line: 269 
and 277

The problem here seems to come from the deallocate for the variable d1vxc  
which is inside a if statement: if (option < 3). But if we look earlier in 
the code, we see that it is allocated for all possible values of the variable 
option. This signifies that the variable is not deallocated for values of 
option bigger then 3. Again, in the declaration of the variables in the 
beginning of the file, we see that this variable is local. But this is maybe 
not an error, because is seems line 335 was corrected by Mikami.


That concludes my findings for today. Have a great weekend.

paul