H2 @ direct CCSD(T)/6-31G* fails
Clicked A Few Times
12:43:18 AM PDT - Fri, Jul 17th 2015
Unique to diatomic hydrogen.

Fails with direct CCSD(T), reports Out Of Memory error irrespective of generous allocation.
Succeeds using TCE.
Succeeds using 6-31G** or larger Pople basis set.

Test script reproduces failure:

start H2

#memory 10 gb

print all



title "H2"

charge 0



geometry units angstroms print xyz

 symmetry C1

 H    -0.372  0.0  0.0

 H     0.372  0.0  0.0

end



basis

 H library 6-31G**

end



#succeeds

task ccsd(t) energy



unset "basis:*"



basis

 H library 6-31G*

end



# FAILS

task ccsd(t) energy

-drh

Forum Vet
11:21:24 AM PDT - Fri, Jul 17th 2015
Thanks for the bug report.

It might be a memory leak.

Forum Vet
1:14:22 PM PDT - Fri, Jul 17th 2015
fix checked into the svn repository

Clicked A Few Times
8:38:58 PM PDT - Fri, Jul 17th 2015
Quote:Edoapra Jul 17th 1:14 pm
fix checked into the svn repository
(*blink!*)
That was quite fast.
Thank you very much.

-drh

Gets Around
11:45:17 PM PST - Sat, Nov 28th 2015
CCSD is an exact wavefunction for a two-electron system. The (T) contribution should be zero. Why are you trying to run this?

Anyways, I'm glad to see that Edo fixed the issue already.

Clicked A Few Times
11:30:52 AM PST - Sun, Nov 29th 2015
Quote:Jhammond Nov 28th 10:45 pm
CCSD is an exact wavefunction for a two-electron system. The (T) contribution should be zero. Why are you trying to run this?

Simple malice.

(*cough!*)

The H and H2 energies are first test cases in the Gaussian-N composite thermochemical methods, even before water, ammonia, and the G2, G3 test suites.

In the G3(MP2) and G4(MP2) methods we got working in NWChem, the post-HF and task statements are general and under python control. The H2 run aborted with an unrelated failure message instead of a modest warning like "not enough electrons for non-iterative triples, (T) correlation = 0.0", i.e., a graceful failure.

The G3(MP2) and G4(MP2) pythonic Nwchem code is currently languishing on Github at
[https://github.com/mattbernst/composite-thermochemistry-nwchem]

It's still manually installed with all that implies about finish level.

-drh


Forum >> NWChem's corner >> Feedback