TCE Frequencies calculation
Just Got Here
12:39:54 AM PST - Fri, Feb 28th 2014
Dear All,

One question has been raised when doing the "TCE Frequencies" calculation:

" ------------------------------------------------------------------------
stpr_gen_hess_at: reference point energy failed        7

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

This type of error is most commonly associated with calculations not reaching

convergence criteria

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

For more information see the NWChem manual at

http://nwchemgit.github.io/index.php/NWChem_Documentation"


=
Input file:

START H2

MEMORY stack 1900 mb heap 200 mb global 3800 mb

GEOMETRY
symmetry group d2h
H 0.0 0.0 -0.39
H 0.0 0.0 0.39
END

BASIS
* library 6-31G*
END

SCF
SINGLET

RHF

thresh 1e-15

tol2e 1e-15
END

TCE
CCSD

FREEZE CORE ATOMIC

NROOTS 2

symmetry

targetsym b1u

target 1

thresh 1.0d-8
END

set tce:thresheom 1.0d-6
set tce:threshl 1.0d-5

set gradient:usegroups T
set subgroups_number 2


task tce optimize numerical
task tce frequencies numerical

=

I am trying to figure out what is the requirement of the " convergence criteria".
Since already very high thresholds already been used in both SCF and TCE blocks.

Would anyone here give some hints about the criteria of which parameters are needed to be improved.

Thanks very much!

Hansly

Forum Regular
11:51:44 PM PDT - Tue, Apr 8th 2014
Hi Hansly,

The error message:

 This type of error is most commonly associated with calculations not reaching

 convergence criteria

is in this case a complete red herring. The reason you get this message is because the energy calculation failed and this fact is passed up to the gradient or hessian driver. That routine only knows the energy evaluation failed but not really why. It then pronounces this wild guess that most likely something did not converge.

In actual fact in this case there is an entirely different problem. The problem is that the finite difference hessian code cannot use symmetry (the finite difference gradient code actually can). So when you hit the hessian driver it turns all symmetry off. However, you used symmetry to specify the target state of interest. With all symmetry gone TCE all of a sudden can no longer identify the state you are after and instead it now reports this a failure of the energy code back to the gradient and hessian drivers (which then confuse it for a convergence problem).

So the fix for this problem is to set the whole calculation up without symmetry (e.g. specify symmetry c1 in the geometry block).

I hope this helps,

Huub


Forum >> NWChem's corner >> Running NWChem