GW calculation stoping with Dipoles: P, V and iR (T) |###########| [100%]

[Daniele Varsano]

Dear Luo,

it seems that code stops after calculating the dipole so starting to calculate the response function.
Here is where it enters the distribution of the X (response function).

Code: Select all

 X_and_IO_CPU= "1 1 1 12 4" 
 X_and_IO_ROLEs= "q g k c v"

Here it means that you are using a parallelisation strategy aimed to distribute on bands (c and v) and not on the other variables. Parallelization over bands allow to distribute memory.
At this stage you can completely neglect the SE_CPU specification as the self energy convolution is done after calculating the response function.

Best,
Daniele

[Luo Yan]

Dear Daniele
Thanks. From your suggestions, I will try "DIP_CPU= "4 12 4" # [PARALLEL] CPUs for each role; DIP_ROLEs= "k c v" # [PARALLEL] CPUs roles (k,c,v)" and "DIP_CPU= "4 12 4" # [PARALLEL] CPUs for each role; DIP_ROLEs= "k c v" # [PARALLEL] CPUs roles (k,c,v) " together in my inouts. However, I can not understand well with bands "c,v" for actual parameters in my calculations. Are the bands the %QPkrange
1|61|29|39|, c "29" v "39" ???

[Daniele Varsano]

Dear Luo,
DIP_ROLE and DIP_CPU governs how the dipole matrix elements are distributed with the CPUs in calculating the dipoles <c|r|v>.
The product of the number you assign has to be equal to the CPU used in your calculations.

Anyway it seems to me that the dipoles are calculated and code stops when starting to calculate the response function.

%QPkrange this is related with SE_CPU (q,qp,b)
q and b distribute q points and bands in the GW summations, qp instead are the diagonal matrix elements you want to calculate independently.
qp cannot be larger than the total number of QP energies you want to calculated indicated in QPkrange. Parallelizing over b allow you to distribute memory.

I suggest you to look if you have any error message in one of the log files in the LOG directory.
Then try to parallelise as much as you can in "c" and "v" in X_and_IO_CPU

Here you can have a look to a lecture about yambo parallel structure:
https://www.youtube.com/watch?v=i3uVgytO4qc&t=1s

starting from around min.13

Best,
Daniele

[Luo Yan]

Dear
Thanks for your detailed suggestions. i solve this problem by another compiled version (4.5.3). Maybe there are some incomplete operations when conplied yambo-5.0 version. Could you direct me to install it for normal compilation ?

[Daniele Varsano]

Dear Luo,

I do not know how your compilation was done, anyway if the compilation was successful it should be ok.
If you want, you can post the report file for the successful (4.5) and unsuccessful (5.0) run, possibly with the same number of CPUs, in order to see if some difference appears.

Best,
Daniele

[Luo Yan]

Dear Daniele,
Thanks a lot. I will try.