CCSD 2emet 9 returns error
Gets Around
5:13:55 AM PST - Mon, Dec 5th 2016
Hi NWCHEM developers.

I found an issue when using 2emet=9 in CCSD calculation.

It is reproduced in ubuntu 16.04 and NWCHEM from apt repo or compiled from source by standard procedure.

Input file
title "H2 CCSD/cc-pVDZ energy"

permanent_dir /var/tmp
scratch_dir /var/tmp

memory stack 100 mb heap 100 mb global 2000 mb

geometry
 symmetry C1
 H       0.0      0.0      0.6
 H       0.0      0.0     -0.6
end

scf
 direct
end

basis spherical
 H library cc-pVDZ
end

tce
 ccsd
 2eorb
 2emet 9
 freeze atomic
end

driver
 xyz
end

task tce energy


Tail of the output
 Global array virtual files algorithm will be used
 
 Parallel file system coherency ......... OK
 
 Fock matrix recomputed
 1-e file size   =              100
 1-e file name   = /var/tmp/H2.f1......
 Cpu & wall time / sec            0.0            0.0
 4-electron integrals stored in orbital form
 
 v2    file size   =             7462
 4-index algorithm nr.   9 is used
 imaxsize =       30
 imaxsize ichop =        0

 tce_mo2e_disk_2eorb: fast2e=9
 2_EL_BATCH AND COEFF MATRIX
 size_ic                20001
 size_icc                   90
 tile_dim                    9
 max_na                   10
 n_4a_files                    1
 n_3a1m_files                    1
 n_2a2m_files                    2
 n_1a3m_files                    2
 atpart                    1
 noa+nva                    2
 noa                    1
 nva                    1
 before sf_create
 n_4a_files                    1
 sizes:
size_4a_m(i)     1       10000
 -----------------
 after sf_create
 4A integrals cpu,wall    1.9999999999999990E-002   9.6671581268310547E-003
 STEP2 4index
 3A1M integrals cpu wall   0.0000000000000000        7.1525573730468750E-006
 STEP3 4index
 2A2M integrals cpu wall   0.0000000000000000        2.7894973754882812E-005
 STEP4 4index
 ------------------------------------------------------------------------
 zones put: sf problem21-b                   1
 ------------------------------------------------------------------------
 ------------------------------------------------------------------------
  current input line :.
    33: task tce energy
 ------------------------------------------------------------------------
 ------------------------------------------------------------------------
 An error occured while trying to read or write to disk space
 ------------------------------------------------------------------------
 For more information see the NWChem manual at http://nwchemgit.github.io/index.php/NWChem_Documentation


Best Vladimir.

Forum Vet
11:02:00 AM PST - Mon, Dec 5th 2016
emet 15
Vladimir
I would recommend use of emet=15 since it is much faster

Gets Around
6:30:03 PM PST - Mon, Dec 5th 2016
Quote:Edoapra Dec 6th 1:02 am
Vladimir
I would recommend use of emet=15 since it is much faster

Thanx Edo.

Clicked A Few Times
9:57:24 AM PST - Tue, Dec 6th 2016
Hi Vladimir,
2emet 9 is disk based procedure using shared files.
I recommend using GA based procedures:
2eorb
2emet 13
(or mentioned by Edo 2emet 15, which is very fast but more memory thirsty).
You may also use "2emet 14" option followed by "split" keyword - works the same as "2emet 13"
but builds atomic 2-electron integrals in batches (which saves a lot of memory).

Best,
Karol

Gets Around
1:10:54 PM PST - Thu, Feb 22nd 2018
I created 2emet=9 in 2008 specifically to run big jobs on my laptop. It worked then but has not been tested much and may have bugs. I recall it only works on one node. 2emet=14 is a much better option that has a much higher probability of working.

Forum Vet
8:14:13 PM PST - Thu, Feb 22nd 2018
This calculation includes a ROHF final molecular orbital analysis in NWCHEM6.8.

With NWCHEM6.8 on Ubuntu17.10 and openmpi-dev on my HP workstation using four threads,
2emet 9 fails with
...
 Parallel file system coherency ......... OK



Fock matrix recomputed

1-e file size   =              100

1-e file name   = /var/tmp/h2ccsd.f1

Cpu & wall time / sec            0.0            0.0

4-electron integrals stored in orbital form

mpirun noticed that process rank 0 with PID 0 on node hp-HP-Z2-Mini-G3-Workstation exited on signal 8 (Floating point exception).
, but 2emet 13, 14 , and 15 all can finish with with the same results
...
Total times cpu: 0.2s wall: 0.2s (2emet 15)
This calculation is very robust and fast on this computer.

Employing macOS High Sierra 10.13.3 (with NWCHEM6.8 compiled on 10.13.2) and mpich 3.2_2 on my MAC with three cores,
2emet 9 fails with
...
Parallel file system coherency ......... OK

Fock matrix recomputed

1-e file size   =              100

1-e file name   = /var/tmp/h2ccsd.f1

Cpu & wall time / sec            0.0            0.0

4-electron integrals stored in orbital form

=======================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= PID 671 RUNNING AT appledeMac-mini.local
= EXIT CODE: 8
= CLEANING UP REMAINING PROCESSES
= YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
=======================================================================
YOUR APPLICATION TERMINATED WITH THE EXIT STRING: Floating point exception: 8 (signal 8)
This typically refers to a problem with your application.
Please see the FAQ page for debugging suggestions
...,
but 2emet 13, 14 and 15 finishes with results almost indentical with those from Ubuntu
...
Total times  cpu:        1.2s     wall:        1.2s(2emet 15)


On both platform, it takes the time with surely negligible differences for each succesfull 2emet.

If the github topic is created, I will put the log files there.


Forum >> NWChem's corner >> Running NWChem