GW calculations stops at dipoles

[burkzdemir]

Dear developers,

I ran spin-orbit calculation and then I tried to perform GW calculation but got the following error. My LOG and setup files are attached.

yambo:26243 terminated with signal 11 at PC=7fa728 SP=7ffc316ab3a8. Backtrace:
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x7fa728]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x6f2fba]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x57324e]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x501105]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x4f1cc7]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x40adbc]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x75d8a5]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x4099b5]
/lib64/libc.so.6(__libc_start_main+0xf5)[0x7f00dc655555]
/home/LTEN/ozdemir_b/yambo-5.0.4/bin/yambo[0x4097e9]

[Daniele Varsano]

Dear Burak,
from the log/report it is not easy to understand the reason of the failure.
It could be a memory issue, in this case I suggest you to optimise the parallel strategy to optimise the memory distribution of dipoles:
1 CPU on k
5 CPU on v
8 CPU in c

Best,

Daniele

[burkzdemir]

Dear Daniele,

I tried the parallelization but I got the same error. What else should I check/do? Could it be related to the fact that I am using intel environment?

[Daniele Varsano]

Dear Burak,

intel environment should not give any problem provided that the compilation is performed correctly and consistent.
What you can try to do is:
1. reduce the number of bands in order to see if it could be a memory problem
2. Recompile the code using the flag --enable-int-linalg in the configure in order to exclude you have miscomputation of the intel linear algebra.

Best,
Daniele

[burkzdemir]

Dear Daniele,

I tried recompiling with --enable-int-linalg. And now it seems to be working but this time it stuck at the end of dipole calculation when ran parallel. Serial run has no problem. I appreciated any help. Here is the output,

<02s> P40-nazare070: Dipoles: P, V and iR (T) | | [000%] --(E) --(X)
<07s> P40-nazare070: Dipoles: P, V and iR (T) |######### | [023%] 05s(E) 20s(X)
<12s> P40-nazare070: Dipoles: P, V and iR (T) |################### | [049%] 10s(E) 20s(X)
<17s> P40-nazare070: Dipoles: P, V and iR (T) |############################## | [075%] 15s(E) 19s(X)
<22s> P40-nazare070: Dipoles: P, V and iR (T) |########################################| [100%] 19s(E) 19s(X)


<36s> P40-nazare070: [DIP] Checking dipoles header
P40-nazare070: [ERROR] STOP signal received while in[04] Dipoles
P40-nazare070: [ERROR] DIPOLES database not correct or missing

Best,
Burak

[Daniele Varsano]

Dear Burak,
it is not easy to help if you do not post all the relevant information (report/logs/input) and eventually config.log

Daniele

[burkzdemir]

Ok, here is the link for the files.

https://drive.google.com/drive/folders/ ... sp=sharing

Burak

[Daniele Varsano]

Dear Burak,

it is not very clear what is happening here.

can you try to include in the input file the parallel structure I have suggested before?
Remove first the ndb.dipoles* file

Code: Select all

DIP_CPU= "1 8 5"                      # [PARALLEL] CPUs for each role
DIP_ROLEs= "k c v"                    # [PARALLEL] CPUs roles (k,c,v) ,

and also for the X:

Code: Select all

X_and_IO_CPU= "1 1 8 5"                 # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "k q c v"   

Daniele

[burkzdemir]

Dear Daniele,

I tried this, but again same result. I realized the error is only in the LOG files for CPU's belonging to the second node, so I ran the calculation with only 1 node and now it works without a problem. Why it stops with more than 1 node? There is some problem with inter-node communication? Could you help me with this problem?

Best,
Burak

[Daniele Varsano]

Dear Burak,

if there is a problem in node communication we cannot help much, you should maybe enquire your system administrator.
Please note that according to your input the size of your calculation is rather small so you should be able to afford it using one node.
Important, pleas note that using,

Code: Select all

NGsBlkXp= 1 RL

you are considering the response function as a scalar and this for sure provides you highly not converged calculation. This is a parameter that you need to converge, you can have a look to one of the tutorial in the yambo wiki e.g.:
http://www.yambo-code.org/wiki/index.ph ... 21_version

Best,
Daniele