CCSDTLR Not Compiled Error in Example ccsdt polar small.nw


Jump to page 1Prev 16
Forum Vet
Thanks a lot. I have already removed '2eorb' according to the manual , or the program would stop earlier.

Forum Vet
Thanks very much for your advice. The compilation is successful.
Now I encounter " failed ga_create size/nproc bytes:Received an Error in Communication" during ccsdt iterations.
when I set the memory to 1200mw, or change io ga to 2emet=3, the problem still appears.
How to set memory size in this case?

Forum Vet
Quote:Xiongyan21 Oct 8th 10:51 pm
Thanks very much for your advice. The compilation is successful.
Now I encounter " failed ga_create size/nproc bytes:Received an Error in Communication" during ccsdt iterations.
when I set the memory to 1200mw, or change io ga to 2emet=3, the problem still appears.
How to set memory size in this case?

Please show the last 100 lines of your output files and your complete input file
Thanks

Forum Vet
Input:
echo
start ccsdt_polar_small
  1. permanent_dir ./perm
  2. scratch_dir ./scratch

    1. memory 12000 mb

  1. R = 2.068 bohr

geometry units au
symmetry d2h
N 0 0 -1.034
N 0 0 1.034
end

basis spherical
* library aug-cc-pVDZ
end

scf
 singlet
rhf
thresh 1e-8
end

tce
 scf
ccsdt
2emet 3
end
......

Forum Vet
Output
......Z axis ( b1u symmetry)

---------------------------------------
Iterations converged

CCSD-LR dipole moments / hartree & Debye

Forum Vet


X       0.0000000      0.0000000
Y 0.0000000 0.0000000
Z -0.0000000 -0.0000000
--------------------------------

Warning: you have not solved the response equations for all axes.  Please analyze the results carefully as the average and anisotropic polarizabilities are surely wrong.


CCSD Linear Response polarizability / au 
Frequency = 0.1518778 / au
Wavelength = 300.0000066 / nm


 available GA memory               7101856  bytes
------------------------------------------------------------------------
createfile: failed ga_create size/nproc bytes 9248672
------------------------------------------------------------------------
------------------------------------------------------------------------
current input line :
46: task tce energy
------------------------------------------------------------------------
------------------------------------------------------------------------
------------------------------------------------------------------------
For more information see the NWChem manual at http://nwchemgit.github.io/index.php/NWChem_Documentation


For further details see manual section:                                                                                                                                                                                                                                                                
0:createfile: failed ga_create size/nproc bytes:Received an Error in Communication
application called MPI_Abort(comm=0x84000000, 9248672) - process 0

Gets Around
You are trying to allocate 12 GB of memory per processor. Do you have that much RAM? This job does not need much memory. This worked fine for me:

echo

start tce_polar_ccsdt_small

memory 100 mb

geometry units au
 symmetry d2h
 Ne  0  0  0
end

basis spherical
 * library aug-cc-pVDZ
end

tce
  freeze core
  ccsdt
  io ga
end

set tce:lineresp T
set tce:respaxis F F T
set tce:afreq 0.0 0.1

task tce energy


It also worked fine with my default memory settings, no memory directive in the input file itself, but I wanted to see if I could shrink the default memory allocation since top didn't show much resident memory per process. Try using less memory.

Forum Vet
Test tce_ccsdt_polar_small can pass and has the similar result with the default memory allocation.

Forum Vet
Cannot Increase GA Needed for Example ccsdt_polar_small
The input is
echo

start ccsdt_polar_small


    1. memory 1000 mb
  1. R = 2.068 bohr

geometry units au
symmetry d2h
N 0 0 -1.034
N 0 0 1.034
end

basis spherical
* library aug-cc-pVDZ
scf
 singlet
rhf
thresh 1e-8
end

tce
 freeze core
ccsdt
io ga
end# This turns on linear response

set tce:lineresp T

  1. Calculate T(1) w.r.t. X and Z only

set tce:respaxis T F T

  1. Dynamic polarizability frequencies in atomic units

set tce:afreq 0.00000000 0.08855851 0.104551063 0.12977315 0.15187784 # INF, 514.5, 435.8, 351.1, and 300 nm, respectively

task tce energy

This input also cannot work.

Forum Vet
I have tried memory=200mb, 100mb, 1000mb, 2000mb,3500mb but still got the same errors.

Gets Around
This runs successfully for me. It takes several hours though.

echo

start ccsdt_polar_small

memory 1000 mb

#
# R = 2.068 bohr
#
geometry units au
 symmetry d2h
 N  0  0 -1.034
 N  0  0  1.034
end

basis spherical
 * library aug-cc-pVDZ
end

scf
  singlet
  rhf
  thresh 1e-8
end

tce
  scf
  ccsdt
  io ga
end

#
# This turns on linear response
#
set tce:lineresp T
#
# Calculate T(1) w.r.t. X and Z only
#
set tce:respaxis T F T
#
# Dynamic polarizability frequencies in atomic units
#
set tce:afreq 0.00000000 0.08855851 0.104551063 0.12977315 0.15187784 # INF, 514.5, 435.8, 351.1, and 300 nm, respectively

task tce energy


The final outputs are slightly different from the reference outputs given in the input file comments. The formatting and the actual values have both changed. I've put the results on pastebin because the forum is doing that charming thing where it can't handle some text:

http://pastebin.com/Bt48fMgY

Forum Vet
Although the memory is increased or "set tce: xmem 250" is used, GA cannot be changed, so the calculation fails.
My question is how to increase GA?
Thanks for your advice.

Forum Vet
Although the memory size is increased or "set tce: xmem 250" used, GA cannot be changed, so the calculation fails.
My question is how to increase GA, e.g., throuhg shmmax setting?

Gets Around
If you need to increase shmmax, here is how to do it on Linux: http://www.linuxforums.org/forum/red-hat-fedora-linux/17025-how-can-i-change-shmmax.html

Changing shmmax on OS X: http://www-01.ibm.com/support/knowledgecenter/SSGU8G_11.70.0/com.ibm.relnotes.doc/notes/if...

Forum Vet
Thanks a lot.
I have successfully increased shmmax on MAC OS X, but still got the same error message.

Forum Vet
Thanks, the total memory ,GA, etc., all can be changed.

Forum Vet
In order to save time, only 0.0885585 / au was chosen for the test.
Parallel calculations were unable to finish the last step trivial calculation after tens of hours, following
the converged CCSDT-LR iterations, and a sequential one gave
...

isotropic 8.4607437 1.2537531
anisotropic 13.3830280 1.9831606

Gets Around
Quote:Xiongyan21 Oct 9th 1:04 am
Thanks a lot. I have already removed '2eorb' according to the manual , or the program would stop earlier.


Yeah, I think this is a regression from when I originally wrote this code, but the fix you found is the best one. Fixing it in the source would take an enormous amount of effort and would achieve the same thing as changing that one line of the input file.

Sorry for the trouble.

Jeff

Gets Around
Quote:Xiongyan21 Oct 11th 3:03 pm
Although the memory size is increased or "set tce: xmem 250" used, GA cannot be changed, so the calculation fails.
My question is how to increase GA, e.g., throuhg shmmax setting?


That input option is not relevant to CCSDT-LR.

Gets Around
Quote:Xiongyan21 Oct 12th 3:19 am
Thanks a lot.
I have successfully increased shmmax on MAC OS X, but still got the same error message.


This test might be too larger for a Mac laptop. I ran it on a workstation when I created it.

On Mac, I use ARMCI-MPI to avoid various GA issues.

Jeff

Forum Vet
Quote:Jhammond Dec 10th 7:47 am
Quote:Xiongyan21 Oct 12th 3:19 am
Thanks a lot.
I have successfully increased shmmax on MAC OS X, but still got the same error message.


This test might be too larger for a Mac laptop. I ran it on a workstation when I created it.

On Mac, I use ARMCI-MPI to avoid various GA issues.

Jeff


Thanks.
This is caused by the sign "#" not deleted in the input line, a funny mistake.

Forum Vet
Thanks a lot.
ARMCI-MPI is wonderful if the various GA issues can be avoided when it is compiled on MAC, which makes the TCE calculations of larger molecules for me feasible.

Gets Around
@Mernst

Can you send me the two log files via email so I can vimdiff them locally? I think you know my email already.

Thanks,

Jeff

Gets Around
According to QA/tests/tce_polar_ccsdt_small/tce_polar_ccsdt_small.out, that test took 20 minutes on 2 nodes 6 years ago. However, that was using the code that still allowed the use of 2eorb with CCSDT.


Forum >> NWChem's corner >> Compiling NWChem
Jump to page 1Prev 16