BSE collision-out of memory and wrong SHG

Various technical topics such as parallelism and efficiency, netCDF problems, the Yambo code structure itself, are posted here.

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano, Conor Hogan, Nicola Spallanzani

Post Reply
sunxl
Posts: 4
Joined: Wed Aug 14, 2024 8:05 am

BSE collision-out of memory and wrong SHG

Post by sunxl » Thu Oct 10, 2024 3:11 am

Dear all:
1、When I calculate the GW-BSE collision step, I always encounter a memory overflow issue. Then I used 100 CPUs and set up

Code: Select all

RT_CPU= "10.2.5"                 # [PARALLEL] CPUs for each role
RT_ROLEs= "k.b.q"                # [PARALLEL] CPUs roles (k,b,q,qp)
There was an error as follows:

Code: Select all

 [ERROR] STOP signal received while in[06] Extended collisions
 [ERROR] USER parallel structure does not fit the current run parameters. HARTREE+SEX COLLISIONS
Then I changed the settings of the CPUs,

Code: Select all

RT_CPU= "100.1.1"                 # [PARALLEL] CPUs for each role
RT_ROLEs= "k.b.q"                # [PARALLEL] CPUs roles (k,b,q,qp)
When the task is calculated to [06.01] SEX+HARTREE (e-e correlation) this step, the process will be stuck. Could you kindly advise on how we might allocate CPU resources more effectively in such scenarios? I appreciate any guidance and tutorial on that!


2、While calculating the two-dimensional material, I observed that the second harmonic generation (SHG) output in the z-direction (Xz**) is showing no value. However, based on the material's symmetry, I would expect it to have a non-zero value. Could you please help clarify why this might be the case?Thank you for your assistance!

Best,
sunxl
You do not have the required permissions to view the files attached to this post.

User avatar
claudio
Posts: 473
Joined: Tue Mar 31, 2009 11:33 pm
Location: Marseille
Contact:

Re: BSE collision-out of memory and wrong SHG

Post by claudio » Thu Oct 10, 2024 8:14 am

Dear sunxl

1) regarding the collisions
*) I advise you to try with a small k-points sampling at the beginning and then increase it
*) use yambo_rt to calculate the collisions because it is in single-precision and it requires half of the memory and it is faster, then you can perform non-linear reponse with yambo_nl
*)You can also try this approximated collisions:
https://wiki.yambo-code.eu/wiki/index.p ... r_response
but you have to check if this approximation is good for your system, just do linear response and compare with standard BSE


2) Since the non-linear response code is based on dynamical Berry phase, non periodic directions are not implemented.
So you are right the SHG in z-direction is not zero but you cannot calculate it with Yambo at present

best
Claudio
Claudio Attaccalite
[CNRS/ Aix-Marseille Université/ CINaM laborarory / TSN department
Campus de Luminy – Case 913
13288 MARSEILLE Cedex 09
web site: http://www.attaccalite.com

User avatar
Davide Sangalli
Posts: 623
Joined: Tue May 29, 2012 4:49 pm
Location: Via Salaria Km 29.3, CP 10, 00016, Monterotondo Stazione, Italy
Contact:

Re: BSE collision-out of memory and wrong SHG

Post by Davide Sangalli » Thu Oct 10, 2024 8:44 am

Dear sunxl,

just an addendum to the suggestions from Claudio.

After the interrupted runs of your log files, yambo will leave inside your "03_coll" folder the files "ndb.em1s*"

Those are fine. Just keep them and re-run the code. Yambo will skip the (long) step

Code: Select all

[05] Static Dielectric Matrix
and, instead, it will load the screening from the "ndb.em1s*" files and directly start the computation of the collisions.

Best,
D.

P.S.: please add the signature to your user profile, it is a rele of the forum
Davide Sangalli, PhD
CNR-ISM, Division of Ultrafast Processes in Materials (FLASHit) and MaX Centre
https://sites.google.com/view/davidesangalli
http://www.max-centre.eu/

sunxl
Posts: 4
Joined: Wed Aug 14, 2024 8:05 am

Re: BSE collision-out of memory and wrong SHG

Post by sunxl » Thu Oct 10, 2024 11:30 am

Dear claudio and Davide,

Thank you for your suggestions!

Additionally, I also conducted a test. For the two-dimensional material, I set the k-point in the z-direction to 2, which allowed me to obtain the SHG component Xz**output in the z-direction. Could you please confirm if the results obtained this way are correct?

Best.
sunxl

Post Reply