composition of exciton by ypp

[caowd]

Dear Daniele Varsano

The problem has been solved by reducing the BSK Dimension according to previous discussions in the forum.

Best Wishes.

[Daniele Varsano]

Dear Wen-Dong,
looking at you report you are try to diagonalize an enormous matrix more than 200Kx 200K. Most probably there are problem of memory allocation.

Code: Select all

|Dimension               : 256000
|Bands                   :  9  24

This is impossible with traditional method.
You can either:
1) Diagonalize it with iterative methods by setting in the input file : BSSmod= "h"
But note that in this way you will have access to the spectrum but not to the eigenvector so that you cannot see the composition of the excitation.

2) Try to reduce your matrix by reducing the BSE bands. Note you have 4000 points in the entire BZ, so the dimension of the matrix become very large when adding bands.
The dimension is Nk x Nv x Nc, and consider that diagonalize a matrix let's say larger than 20k x 20k become very demanding. Try to see if you can converge it with very few bands, or may be you can use a smaller k sampling.

Best,

Daniele


Addendum: I answered your post before looking your last post. Great to know you solved.

[haseebphysics1]

Dear Daniele,

I am also facing the issue of disk space storage of large BS matrix, so as was suggested in this thread that we can use the following to avoid disk-IO:

DBsIOoff= "BS"
But in my case BS matrix is still being written in a large file (ndb.BS_PAR_Q1) (more than 290 GB) and I am unable to proceed due to disk storage quota!

I am using yambo-4.5.1, compiled with HDF5 support.

I added manually the following lines in the input:

Code: Select all

StdoHash=  40                  # [IO] Live-timing Hashes
DBsIOoff= "BS"               # [IO] Space-separated list of DB with NO I/O. DB 
 (DIP,X,HF,COLLs,J,GF,CARRIERs,OBS,W,SC,BS,ALL)
DBsFRAGpm= "none"              # [IO] Space-separated list of +DB to FRAG and -DB to NOT FRAG. DB=(DIP,X,W,HF,COLLS,K,BS,QINDX,RT,ELP
#WFbuffIO                      # [IO] Wave-functions buffered I/O

Thanks,

[Daniele Varsano]

Dear Haseeb,
this is happening because this feature was not added for the hdf5 support.
Applying the patch attached here (thanks Davide!) the DBsIOoff= "BS" should work also for the hdf5 support.

Best,
Daniele

[haseebphysics1]

Thank you, developers!


Regards,

[haseebphysics1]

.