Does the TCE officially support QCISD(T)?
Gets Around
1:33:45 AM PDT - Sun, May 18th 2014
In src/tce/tce_input.F I see that there are handlers for "qcisd(t)" and "cr-qcisd(t)". I took QA/tests/tce_h2o/mp4sdq_h2o.nw and replaced

tce
mbpt4(sdq)
end


with

tce
qcisd(t)
end


and the modified input file appeared to run successfully, i.e. it produced a number that looked reasonable and didn't crash during execution.

But I don't know if it is really ready for use since the manual doesn't mention QCISD(T):

http://nwchemgit.github.io/index.php/TCE

I am interested in implementing the Gaussian-n composite thermochemical methods. I want to be able to treat open shell molecules so the implementations in GAMESS and Psi4 are currently inadequate; both require RHF reference functions for their QCISD(T) functionality. If there's something weird about the QCISD(T) functionality implemented through the TCE I don't have a second program to double-check calculations on open shell molecules using it.

Gets Around
8:58:00 AM PST - Wed, Dec 10th 2014
Quote:Mernst May 18th 8:33 am
In src/tce/tce_input.F I see that there are handlers for "qcisd(t)" and "cr-qcisd(t)". I took QA/tests/tce_h2o/mp4sdq_h2o.nw and replaced

tce
mbpt4(sdq)
end


with

tce
qcisd(t)
end


and the modified input file appeared to run successfully, i.e. it produced a number that looked reasonable and didn't crash during execution.

But I don't know if it is really ready for use since the manual doesn't mention QCISD(T):

http://nwchemgit.github.io/index.php/TCE

I am interested in implementing the Gaussian-n composite thermochemical methods. I want to be able to treat open shell molecules so the implementations in GAMESS and Psi4 are currently inadequate; both require RHF reference functions for their QCISD(T) functionality. If there's something weird about the QCISD(T) functionality implemented through the TCE I don't have a second program to double-check calculations on open shell molecules using it.


That code might be correct. I recall that (T) is the same for QCISD and CCSD. It should be valid to run the (T) code with QCISD amplitudes.

I am surprised that QCISD(T) for open-shell systems cannot be verified via CFOUR. Can you verify this?

In any case, QCISD(T) only has a quadratic singles term and thus I think CCSD(T), which has higher-order singles terms, should be much more accurate for open-source systems. QCISD(T) for open-shell and multi-reference systems is a known problem in the literature.

In the event you find an error in NWChem, please contact me and I will try to debug the issue. My email address is found on http://jeffhammond.github.io/.

Best,

Jeff

Gets Around
1:23:56 PM PST - Wed, Dec 10th 2014
CFOUR may well offer this functionality for comparison. In the past I signed dozens of academic software license agreements, nowadays I have license agreement fatigue. Perhaps I should to endure it one more time.

QCISD(T) surely does not offer the best accuracy for its cost among current methods, but it is needed to reproduce historical work. Calculations that required long jobs on expensive hardware 20 years ago are now easily completed on a laptop, if the software support is there. My ultimate motivation is that I don't think people should have to pay $$$$ to G******* Inc just to reproduce literature results from the 1980s and 1990s. In any case I only get to work on this problem in dribs and drabs of spare time, so progress is slow.

Thank you for the response. I have learned a lot from your mailing list posts and web pages in the past.

Clicked A Few Times
11:14:41 AM PDT - Tue, Jul 14th 2015
QCISD(T) problem for larger molecules
echo
title "QCISD"
permanent_dir ......
start QCISD
memory stack 2320 mb heap 180 mb global 2000 mb noverify

geometry units angstrom
45 atoms
end

basis
O library 6-311g
H library 6-311g
C library 6-311g
end

scf
thresh 1.0e-8
tol2e 1.0e-8
singlet
rhf
maxiter 200
end

tce
QCISD(t)
maxiter 60
diis 3
thresh 1.e-6
2eorb
2emet 3
attilesize 40
tilesize 8
freeze core atomic
end

task tce energy


After running for 8 hours in 15 nodes I got a following error.

t2 file size = 372112063
t2 file name   = ./QCISD.t2          

t2 file handle =       -996

CCSD(T)

Using plain CCSD(T) code
41: error ival=11
75: error ival=11
(rank:41 hostname:g432 pid:19888):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)
35: error ival=11
(rank:35 hostname:g431 pid:15507):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)
49: error ival=11
45: error ival=11
(rank:49 hostname:g433 pid:5427):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)

I appreciate your help to resolve this issue.

Thank you.
Sujala

Gets Around
11:27:18 AM PDT - Tue, Jul 14th 2015
After running for 8 hours in 15 nodes I got a following error. 

t2 file size   =        372112063

t2 file name   = ./QCISD.t2          

t2 file handle =       -996

CCSD(T)

Using plain CCSD(T) code

41: error ival=11

75: error ival=11

(rank:41 hostname:g432 pid:19888):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)

35: error ival=11

(rank:35 hostname:g431 pid:15507):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)

49: error ival=11

45: error ival=11

(rank:49 hostname:g433 pid:5427):ARMCI DASSERT fail. ../../ga-5-3/armci/src/devices/openib/openib.c:armci_send_complete():459 cond:(pdscr->status==IBV_WC_SUCCESS)

This is usually associated with out-of-memory errors, but could also be due to the design of ARMCI for InfiniBand. Errors like this are why I use ARMCI-MPI (https://wiki.mpich.org/armci-mpi/index.php/Main_Page) instead. You should report this issue to the GA team (hpctools@emsl.pnl.gov) and if they fail to address the issue, you may consider trying ARMCI-MPI (which I co-wrote, hence I am biased about its utility).

Clicked A Few Times
12:16:23 PM PDT - Tue, Jul 14th 2015
QCISD(T) reference values for smaller compounds and basis sets are at CCCBDB

http://cccbdb.nist.gov/energy1.asp


Forum >> NWChem's corner >> NWChem functionality