Error in anisotropic calculation

[mdogan]

Hello,

I am trying to run an anisotropic Eliashberg calculation on a hydrogen phase. Although I'm able to run an isotropic calculation without an issue (epw3.in and epw3.out), the anisotropic calculation fails with the BAD TERMINATION error (epw4.in and epw4.out). The full error message is unfortunately not very descriptive either:

Code: Select all

munmap_chunk(): invalid pointer
srun: error: r220: task 8: Broken pipe
srun: error: r263: task 11: Broken pipe
[mpiexec@r133.ib.bridges2.psc.edu] wait_proxies_to_terminate (../../../../../src/pm/i_hydra/mpiexec/intel/i_mpiexec.c:527): downstream from host r133 exited with status 141

I successfully ran anisotropic calculations in other crystal phases of this material with analogous input files, and changing the number of nodes hasn't made a difference, so this is a total mystery to me.

All the files are available here: [https://www.dropbox.com/sh/a12q7198eiq7 ... b06Ha?dl=0]

I appreciate any advice in advance.

Best,
Mehmet Dogan

[hpaudya1]

Hi Mehmet,

This looks like a memory issue to me. Did you notice;

"Size of allocated memory per pool: ~= 1.8548 Gb"

This is even before reading *.ephmat files. I recommend using larger memory nodes if available or try running in more cores.

HappyEPWing,
Hari Paudyal

[hlee]

Dear Mehmet Dogan:

Could you show me make.inc you are using when building the code?

Sincerely,

H. Lee

[mdogan]

Dear Hari and H. Lee,

I'm not sure I understand the memory allocation. I'm using 32 nodes with 128 GB memory each. Given that I specify 128 pools, that should mean that each pool has 32 GB of memory available. So, shouldn't that be plenty?

I've also added the make.inc file to the Dropbox folder.

Thank you both very much!

Best,
Mehmet

[hlee]

Dear Mehmet Dogan:

I would suggest you the followings:
(1) Use of gcc environments instead of intel.
(2) Include "--enable-debug" in the options of configure like "./configure --enable-debug ..."
(3) Run calculations again.

Sincerely,

H. Lee

[mdogan]

Dear H. Lee,

Thank you for your suggestion! I recompiled and reran as you suggested, and the resulting error message was:

Code: Select all

At line 522 of file supercond.f90
Fortran runtime error: Index '-332' of dimension 1 of array 'lambda_pairs' below lower bound of 1

Error termination. Backtrace:
#0  0x4f4035 in __supercond_MOD_evaluate_a2f_lambda
        at /jet/home/mdogan/qe-6.7-v2/EPW/src/supercond.f90:522
#1  0x425178 in eliashberg_eqs_
        at /jet/home/mdogan/qe-6.7-v2/EPW/src/eliashberg.f90:61
#2  0x406fdf in epw
        at /jet/home/mdogan/qe-6.7-v2/EPW/src/epw.f90:163
#3  0x406afc in main
        at /jet/home/mdogan/qe-6.7-v2/EPW/src/epw.f90:20
srun: error: r171: task 2: Broken pipe
[mpiexec@r120.ib.bridges2.psc.edu] wait_proxies_to_terminate (../../../../../src/pm/i_hydra/mpiexec/intel/i_mpiexec.c:527): downstream from host r120 exited with status 141

I also put the new output file and the error log in the shared folder. I guess the error is connected with not finding but I'm not sure how. Please let me know what you think.

Best,
Mehmet

[mdogan]

* I meant "I guess the error is connected with not finding lambda pairs but I'm not sure how."

[hlee]

Dear Mehmet:

At line 522 of file supercond.f90
Fortran runtime error: Index '-332' of dimension 1 of array 'lambda_pairs' below lower bound of 1

The error message above tells us that there is a out-of-bound error; the first dimension of the array 'lambda_pairs' is smaller than one; this should not happen.

Sincerely,

H. Lee

[mdogan]

Dear H. Lee,

Thank you! Do you have any idea what may be causing this anomalous situation? As far as I can see, lambda_pairs is allocated in line 497 of supercond.f90 but I don't see where its entries get populated.

Best,
Mehmet

[hlee]

Dear Mehmet:

Please look at this line (https://gitlab.com/QEF/q-e/-/blob/devel ... d.f90#L520).
In your case, ibin has a negative value of -332.

Usually, dosef, wf, and g2 have non-negative values and the issue above doesn't happen (I assume that you don't set shortrange to .true. and eps_acustic>0).
You can find its origin by looking at the lines of supercond.f90 around the line of 520 and following relevant values which affect the ibin.

Sincerely,

H. Lee