Load of old vectors failed
Clicked A Few Times
8:56:55 AM PDT - Mon, Sep 17th 2012
Hi,

I'm facing a new problem when trying to load old vectors in a geometry optimization. I simply resubmitted a job that hadn't converged by specifying the input vectors (which I have done many times without any problem), but this time, I get a new warning about a basis mismatch before nwchem fails to load the movecs file :

!! nbf/nmo/basis-name mismatch

          nbf=1976 nbf_file=1976

          nmo=1962nmo_file=1963

          basis="ao basis" basis_file="ao basis"



 Either an incorrect movecs file was

specified, or linear dependence has changed,

 or the basis name was changed.



Loading old vectors from job with title :



 Load of old vectors failed.   Forcing atomic density guess

The thing is, I haven't changed anything in the input file (same basis, same atoms, same everything). I tried resubmitting the job with another set of older vectors (again from the same geometry) and got the same error. It doesn't seem like the vector file is corrupted as I have already loaded this same file previously without any problem. In fact, it looks like it always fails at the same step of the geometry optimization even though there is nothing special about that step. Has anyone ever seen this?

Chloe

Forum Vet
11:56:41 AM PDT - Mon, Sep 17th 2012
Please post the input file
Chloe
Could you please post the input file?
Thanks, Edo

Clicked A Few Times
5:57:02 AM PDT - Tue, Sep 18th 2012
Here's the input :

echo
restart
memory 1500 mb

geometry full units angstrom noautoz nocenter noprint
[...]
symmetry c1
end

basis
 C library 3-21g*  file ~/libraries/

 H library 3-21g*  file ~/libraries/

 Pd library lanl2dz_ecp  file ~/libraries/
end

ecp
 Pd library lanl2dz_ecp file ~/libraries/
end

set tolguess 1e-7

set geometry:actlist 1:108

driver
loose

maxiter 200
end

dft
direct

xc xpbe96 cpbe96

vectors input Pdarmchair7x11opt.movecs output Pdarmchair7x11opt.movecs

iterations 700

convergence ncyds 700
end
task dft optimize

Forum Vet
2:23:22 PM PDT - Tue, Sep 18th 2012
All that was needed for a restart is:

echo
restart
memory 1500 mb

task dft optimize


The rest will be taken from the runtime database.

Which geometry did you specify in your restart input?

What you are dealing with is a change in the number of linear dependent basis functions as a result of a reasonably large change in geometry during the optimization. One way to get around this is to set the number of linear dependent functions that can be discarded to set using "set lindep: n_dep 0" or to adjust the linear dependence tolerance using "set lindep:tol 1.0d-5" or some other value.

Bert
Quote:Chloe Sep 18th 12:57 pm
Here's the input :

echo
restart
memory 1500 mb

geometry full units angstrom noautoz nocenter noprint
[...]
symmetry c1
end

basis
 C library 3-21g*  file ~/libraries/

 H library 3-21g*  file ~/libraries/

 Pd library lanl2dz_ecp  file ~/libraries/
end

ecp
 Pd library lanl2dz_ecp file ~/libraries/
end

set tolguess 1e-7

set geometry:actlist 1:108

driver
loose

maxiter 200
end

dft
direct

xc xpbe96 cpbe96

vectors input Pdarmchair7x11opt.movecs output Pdarmchair7x11opt.movecs

iterations 700

convergence ncyds 700
end
task dft optimize

Clicked A Few Times
6:43:06 AM PDT - Wed, Sep 19th 2012
Thanks Bert, but "set lindep" doesn't work. I have a feeling it has more to do with the warning about the basis mismatch than the linear dependence.

By the way, I was using the geometry from the last step as the input geometry.

Chloe

Forum Vet
3:48:11 PM PDT - Mon, Sep 24th 2012
Chloe,

At this point in your caluclation lindep should not help you, only if you start a new simulation. I can only suggest starting with the latest geometry and without the old vectors.

Bert


Quote:Chloe Sep 19th 1:43 pm
Thanks Bert, but "set lindep" doesn't work. I have a feeling it has more to do with the warning about the basis mismatch than the linear dependence.

By the way, I was using the geometry from the last step as the input geometry.

Chloe

Clicked A Few Times
8:21:05 AM PDT - Wed, Oct 3rd 2012
If "lindep:tol 1.0d-5" doesn't do it, should I set it to a larger or smaller value?

Chloe

Forum Vet
11:16:34 AM PDT - Wed, Oct 3rd 2012
If you know how many linear dependent functions you should have you can use

 set lindep:n_dep <# of linear dependent functions it should remove>

Bert

Quote:Chloe Oct 3rd 3:21 pm
If "lindep:tol 1.0d-5" doesn't do it, should I set it to a larger or smaller value?

Chloe

Clicked A Few Times
6:24:27 AM PDT - Thu, Oct 18th 2012
I've tried everything, restarted the calculation from a previous geometry and building a new guess for the vectors. It always ends up converging to the same vectors that fail on the next step. I can't get passed this point. I know I'm supposed to have 13 linear dependencies and that the calculation fails when there are 14, but setting n_dep to 1 doesn't change anything, nor does changing the tolerance. Whatever I do I always get 14 linear dependencies and this message about the the basis mismatch. I also tried to change the geometry a little manually so that the calculation doesn't follow the exact same optimization steps, but it didn't help. This is really bothering me. Is there any workaround?

Forum Regular
7:02:31 PM PDT - Thu, Oct 18th 2012
Hi Chloe,

Can you send us your complete input file ? If you don't want to post it just email it to us.

Best,
-Niri

niri.govind@pnnl.gov


Forum >> NWChem's corner >> Running NWChem