TCE error ival=4 problem

Dear NWChem users and Developers!

As discussed in another topic I try to run CCSD(T) calculations in the TCE module to obtain electric dipole moment and polarizability. Now I finished the test calculations and turned to the actual problem. For these systems I want to do the following

tce

freeze atomic or core

ccsd(t)

io ga

2eorb

2emet 3

tilesize 15

end

set tce:lineresp T
set tce:afreq 0.000
set tce:respaxis T T T

task tce energy

for which I use a 6-311g** basis set. If I do so for higher symmetry structures with C2v, D2h symmetry the calculation converges normally. With the same number of atoms but lower symmetry (of course by symmetry more integrals have to be calculated explicitly) , C2 or Cs, the calculations produce the error

0: error ival=4

This error always occurs in one of the first steps of CCSD Lambda iterations so after the CCSD routine is finished.
I am aware of the fact that these calculations are computationally more expensive but I also doubled the amount of processors used in the calculation but still the error keeps appearing. So far I have used
memory stack 800 mb heap 100 mb global 1000 mb
and increased it to
memory stack 2320 mb heap 180 mb global 2000 mb
but without solving the problem.

I would be pleased for any suggestion how I could solve this issue.

Thanks in advance and all the best

Sven

Hi Bert!

No I just mentioned core or atomic since I was not sure if there is a difference. I tried both options but only one at a time. Right now I am using 80 processors with, as far as I know, 4.5 GB per Processor but if that is an important parameter I could ask the admin of the cluster. The full input is

start Si11_II_CC

memory stack 2320 mb heap 180 mb global 2000 mb

geometry units angstroms

  Si  -3.13917621     1.04569897     0.00000000

  Si  -0.48608866    -2.05930346     0.00000000

  Si  -1.44054750    -0.17461559    -1.23279420

  Si  -0.91486947     2.10257468     0.00000000

  Si   0.81574414     1.05614863    -1.29483629

  Si   2.74618502     1.01873330     0.00000000

  Si   0.55884417    -1.29069123    -2.07231568

  Si   1.92586771    -1.28938713     0.00000000

  symmetry cs

end

basis spherical

 si library 6-311g**

end

tce

freeze atomic

ccsd(t)

io ga

2eorb

2emet 3

tilesize 15

end

set tce:lineresp T
set tce:afreq 0.000
set tce:respaxis T T T

task tce energy

All the best

Sven

As I was not sure with the old settings on the cluster I reran the calculation with

memory stack 800 mb heap 100 mb global 1000 mb

with 20 nodes with 4 cores on each and using 8GB per node. Still the error keeps
appearing.

Sven

Mhh, this is strange it is maybe a compilation problem. Unfortunately, I do not know how the program is compiled and what libraries are used.
The rest of the error message is

CCSD Lambda iterations

---------------------------------------------

Iter          Residuum            Cpu    Wall

---------------------------------------------

   1         65.6301924841714   407.4   417.5

0: error ival=4
(rank:0 hostname:u4n075 pid:24296):ARMCI DASSERT fail. openib.c:armci_call_data_server():2010 cond:(pdscr->status==IBV_WC_SUCCESS)
4: error ival=4
(rank:4 hostname:u4n053 pid:29430):ARMCI DASSERT fail. openib.c:armci_call_data_server():2010 cond:(pdscr->status==IBV_WC_SUCCESS)

Please let me know if you are able to run the calculation with io ga. If so I think I have to contact the admin of the cluster.

Sven

Sven,

I was able to run it successfully with your input deck.

What kind of hardware are you running on?

Do you set ARMCI_DEFAULT_SHMMAX in your job script or environment?

Do you set MA_USE_ARMCI_MEM in your job script or environment?

Bert

Thanks for the moment!

I'll give it a try. About the speed: I know there were some issues compiling NWChem v6.0 and it took the admin quite a while.
Maybe it is a general compilation problem of the program on the cluster. Would it be ok if our admin contacts you if the error
keeps appearing?

Thanks

Sven