From NWChem
Viewed 1128 times, With a total of 3 Posts

Clicked A Few Times
Threads 5
Posts 19


2:58:26 PM PST  Tue, Feb 2nd 2016 

Hello,
I'm encountering an issue with using NWChem 6.6 (fully patched, built for LINUX64 with Intel MKL) to perform CDFT calculations. Specifically, whenever I try to apply a charge constraint using the Becke population scheme, the calculation crashes during the first iteration with the error message "CDFT: multipliers go over limit."
Searching online for this error brings up an old thread from the mailing list where this error message is attributed to the target diabatic state being exceptionally unphysical. However, I do not believe this to be the case here; this error has been reproducible for a variety of systems, including for systems built from the appropriate fragment guesses at large separation.
These systems converge just fine when Lowdin populations are used instead.
The minimal system for which I can reproduce this error is a He dimer with a net charge of +1. This will converge just fine when Lowdin populations are used, but fails for Becke populations. Another example is the CDFT test included in the NWChem QA suite (QA/tests/cdft_jarv1/cdft_jarv1.nw). This is another case where convergence is fine when the default Lowdin populations are used, but the "multipliers go over limit" error occurs in the first iteration if Becke populations are used.
The minimal He dimer input file:
# He2+ dimer, 8 A separation, fragments then CDFT calculation
geometry left
symmetry c1
He 0.0000 0.0000 4.0000
end
geometry right
symmetry c1
He 0.0000 0.0000 4.0000
end
geometry full
symmetry c1
He 0.0000 0.0000 4.0000
He 0.0000 0.0000 4.0000
end
basis spherical
He library "augccpvtz"
end
# left He, neutral
set geometry left
charge 0
dft
xc m06hf
grid xfine
odft
mult 1
vectors input atomic output left.vec
end
title "left He, neutral"
task dft energy
# right He, charge +1
set geometry right
charge 1
dft
xc m06hf
grid xfine
odft
mult 2
vectors input atomic output right.vec
end
title "right He, charge +1"
task dft energy
# dimer with constraints
set geometry full
charge 1
dft
xc m06hf
grid xfine
odft
mult 2
vectors input fragment left.vec right.vec output full.vec
cdft 2 2 charge 1 pop becke
end
title "dimer with constraints"
task dft energy
This system converges fine with the Lowdin population  in fact, the multiplier optimizes to 0, indicating that the He atoms are sufficiently separated such that the system naturally localizes the charge without the need for a constraint. Therefore, it doesn't make sense for the CDFT multipliers to be over the limit.
Ordinarily I would simply proceed with the Lowdin populations and a spherical basis set. However, since I am trying to calculate offdiagonal coupling elements in a diabatic basis, Becke populations are required in order to correctly reproduce the convergence of the coupling elements in the complete basis set limit.
Any insight into why the Becke populations are failing would be much appreciated.

Edited On 3:02:15 PM PST  Tue, Feb 2nd 2016 by Buchwj




Clicked A Few Times
Threads 2
Posts 8


5:28:50 AM PST  Tue, Feb 23rd 2016 

I also have this problem with becke populations on my NWCHEM 6.6 that was compiled and installed a few days ago.
0:0:CDFT: multipliers go over limit:: 1
(rank:0 hostname:ca096 pid:22555):ARMCI DASSERT fail. ../../ga54/armci/src/common/armci.c:ARMCI_Error():208 cond:0
CDFT: multipliers go over limit 0

Edited On 5:50:34 AM PST  Tue, Feb 23rd 2016 by Brose




Clicked A Few Times
Threads 5
Posts 19


8:54:03 PM PST  Tue, Feb 23rd 2016 

Brose,
I've been trying to troubleshoot this for a few weeks now, and one thing that I've noticed is that you have to explicitly tell NWChem to partition the electron density using the Becke scheme. The CDFT code reads the partition weights directly from the grid files, and if you use the default grid settings (erf1), the "Becke populations" won't actually be Becke populations. Try this in your DFT section:
dft
grid fine becke
end
In my experience, this fixes the multiplier limit issue in about 1/3 of the cases I tried, though it's by no means a complete fix  just a possible step in the right direction.




Clicked A Few Times
Threads 1
Posts 7


8:42:34 PM PST  Fri, Feb 26th 2016 

Becke populations

Buchwj ,
You may use the Cartesian option or an alternative exchangecorrelation potential to get your system with Becke populations converged. In both cases NWChem runs fine and without failure report.
Regards
Manfred



AWC's:
2.5.10 MediaWiki  Stand Alone Forum Extension
Forum theme style by: AWC