Two minor questions on the new version

[xueshanxihua]

Dear developers,

I have some small questions about the newest version (i.e., 4.4 and 4.5) :

1. For ver. 4.5.0, ypp has two optional parameters for exciton postprocessing, i.e., "ypp -e <opt> <opt>", and the second one is the qpoint index. Does it means yambo has done or plans to do a finite-momentum BSE module? Can we already use it at this moment?

2. I remember that, for extremely old versions such as 4.0.x, using more band fragments (p2y -b) helps to slove the memory issues. Does this tip still work for the newest version, because I find that the databases translation using "p2y -b" is now notably slower than the default one without fragment.

Thank you very much,

[Daniele Varsano]

Dear Zeyu,

1) The finite-momentum BSE module has been coded indeed and that why ypp has been modified accordingly. Unfortunately, it has not yet been resealed in the GPL, and as you noticed the ypp procedure is a bit inconsistent. Thanks for reporting, we will correct it soon. Anyway, I'm confident we will release the finite-q BSE very soon as it has been already sufficiently tested.

2) Well, in line of principle, it should work. Probably the I/O is different in the two cases and this can explain the differences but I would not expect bug differences, can you report with more detail (execution time etc, detail of the fragmentation..) what you observed?

Best,
Daniele

[Davide Sangalli]

2. I remember that, for extremely old versions such as 4.0.x, using more band fragments (p2y -b) helps to slove the memory issues. Does this tip still work for the newest version, because I find that the databases translation using "p2y -b" is now notably slower than the default one without fragment.

For fragmentation over the bands

Code: Select all

p2y -b XX

should be used. XX is the number of bands you would like in each fragment.
It should give an error if you just run

Code: Select all

p2y -b

without specifying a number.

However notice that since version 4.2 the I/O of the wave--functions was improved a lot.
Now the variable wf_disk, which previously was allocated with the size of each fragment, is allocated with a smaller size, depending on the number of WFs to be loaded.
Thus, especially for parallel runs, I'd say fragmentation over the bands is not useful anymore.

Best,
D.