Dear Steven and all MPB users,

I have some problem when using mpb-mpi for 3d simulations (resolution 
(32,32,32) and computation region about (1, 10, 4):
1, It seems that the targeted-frequency method is not accurate enough, even if 
the error is set to be 1e-8 or 1e-9. In 2D, smooth curves can be obtained. 
While in 3D, the curves are ugly, even for a small freqency region (fcen ~0.28, 
df ~0.005). Good results can be obtained by using the normal method.
2, For a small-size (num of k-points and band-nums are small) computation, 
mpb-mpi will work, while for a large-size computation it gives errors after the 
first k-point. The errors info is: CHECK failure on line 243 of blasglue.c: non 
positive-definite matrix in potrf .....MPI_ABORT was invoked on rank 19 in 
communicator MPI_COMM_WORLD with errorcode 1 ..... 

What's the reason of the above problems? Is there any limitation of the 
computation size by using mpb/mpb-mpi? The MPB-MPI is run on openmpi. Only a 
small cluster based on NFS is used.

Thanks in advance.

Best,
Jiangjun

_______________________________________________
mpb-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/mpb-discuss

Reply via email to