Convergence problems with Au cluster


Jump to page 12Next 16Last
Clicked A Few Times
8:28:56 AM PST - Sat, Jan 28th 2017
Dear NWChem users and developers,
I have some problems in reaching convergence with a pretty simple calculation on an Au20 cluster.
The geometry should reproduce that of the bulk gold, so I would expect to achieve fast convergence in an energy dft run. But the calculation does not converge.
I am using lanl2dz_ecp basis set (common in literature) with effective core potential.
I tried to vary the damping factor and I increased it up to 99, but still it does not converge, even after thousands of iterations.

Here is the input file (a part from the geometry)

basis spherical
  Au library lanl2dz_ecp
end
ecp
  Au library lanl2dz_ecp
end

dft
  iterations 990

  convergence damp 99

  convergence ncydp 990

  grid fine

  direct

  noio
end
task dft energy

Any suggestion on how to overcome this problem? Does anyone ever face the same problem?
Thanks a lot

Alessandro

Forum Vet
8:39:10 AM PST - Tue, Jan 31st 2017
Could you provide the complete input file?

Clicked A Few Times
12:44:14 PM PST - Tue, Jan 31st 2017
Yes, the full input file is below.
I should also comment that with an all-electron basis set (vdz) the calculation converges. I have tried several other ECP, but I achieved convergence only with CRENBL.

start Au10
Title "Au10 cluster"

memory total 2000 mb

charge 0

geometry units angstroms print xyz noautoz
Au 0.0000000e+00 0.0000000e+00 -5.0000000e+00
Au -1.4421450e+00 -8.3262280e-01 -7.3550129e+00
Au 1.4421450e+00 -8.3262280e-01 -7.3550129e+00
Au 0.0000000e+00 1.6652456e+00 -7.3550129e+00
Au -2.8842900e+00 -1.6652456e+00 -9.7100258e+00
Au 0.0000000e+00 -1.6652456e+00 -9.7100258e+00
Au 2.8842900e+00 -1.6652456e+00 -9.7100258e+00
Au -1.4421450e+00 8.3262280e-01 -9.7100258e+00
Au 1.4421450e+00 8.3262280e-01 -9.7100258e+00
Au 0.0000000e+00 3.3304912e+00 -9.7100258e+00
end

basis spherical
 Au library lanl2dz_ecp
end
ecp spherical
 Au library lanl2dz_ecp
end
dft 
 iterations 990

 convergence damp 99

 convergence ncydp 990

 direct

 noio
end

task dft energy

Forum Vet
1:21:38 PM PST - Tue, Jan 31st 2017
Try this input file
start Au10
Title "Au10 cluster"

memory total 2000 mb n

charge 0

geometry units angstroms print xyz noautoz
Au 0.0000000e+00 0.0000000e+00 -5.0000000e+00
Au -1.4421450e+00 -8.3262280e-01 -7.3550129e+00
Au 1.4421450e+00 -8.3262280e-01 -7.3550129e+00
Au 0.0000000e+00 1.6652456e+00 -7.3550129e+00
Au -2.8842900e+00 -1.6652456e+00 -9.7100258e+00
Au 0.0000000e+00 -1.6652456e+00 -9.7100258e+00
Au 2.8842900e+00 -1.6652456e+00 -9.7100258e+00
Au -1.4421450e+00 8.3262280e-01 -9.7100258e+00
Au 1.4421450e+00 8.3262280e-01 -9.7100258e+00
Au 0.0000000e+00 3.3304912e+00 -9.7100258e+00
end

basis spherical
 Au library lanl2dz_ecp
end
ecp spherical
 Au library lanl2dz_ecp
end
dft
 iterations 990
 convergence damp 40 ncydp 0 dampon 1d99 dampoff 1d-3
 direct
 smear
 noio

end
set quickguess t
task dft energy

Clicked A Few Times
2:06:14 AM PST - Wed, Feb 1st 2017
I have tried, but it still does not converge...

Forum Vet
6:43:43 AM PST - Wed, Feb 1st 2017
Could you post the output file to a public website? I don't see any problem running it with NWChem 6.6

Forum Vet
6:48:30 AM PST - Wed, Feb 1st 2017
Could you post the output file to a public website? I don't see any problem running it with NWChem 6.6

http://pastebin.com/Z9DhkwNs

Clicked A Few Times
2:21:25 PM PST - Wed, Feb 1st 2017
Here is my output, I don't know why it does not converge...

http://pastebin.com/6EEbbLCC

Forum Vet
9:41:54 PM PST - Wed, Feb 1st 2017
Alessandro
Your NWChem installation fails to produce correct eigenvalues for the overlap matrix (you can spot this by comparing the two output files). Could you upload the following file
$NWCHEM_TOP/src/tools/build/config.log
What compilers and versions are you using?

Clicked A Few Times
1:23:31 PM PST - Thu, Feb 2nd 2017
I uploaded both the config and build_nwchem files here

http://pastebin.com/dREhnUnY

Thank you for your help

Alessandro

Forum Vet
5:24:54 PM PST - Thu, Feb 2nd 2017
Please do NOT use the following settings that show up in your log
export FOPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops -unroll-aggressive"
export COPTIMIZE="-O3 -xSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 -no-prec-div -funroll-loops"
There are lost of of other safer and more effective performance optimizations that can be used in your installation,
e.g. definition of BLAS and SCALAPACK.
To sum it up, please do NOT set neither FOPTIMIZE nor COPTIMIZE since the options used in the NWChem have been validated using the NWChem QA suite, while any other value of those two values might have unexpected consequences.

Another suggestion is to stick to gcc (default) for the choice of C compiler.

Forum Vet
10:20:34 PM PST - Thu, Feb 2nd 2017
Dear Edoapra
  I employed your input without n in the memory group,
carried out the calculation with three cores, and the calculation finishes at the 14th step.
...



convergence    iter        energy       DeltaE   RMS-Dens  Diis-err    time

---------------- ----- ----------------- --------- --------- ---------  ------
d= 0,ls=0.0,diis 14 ... -1.96D-07 7.74D-06 1.95D-07 57.0
...

This is the result from NWCHEM6.6 obtained by your input.

Pastebin.com is not accessible to me.


Very Best Regards!

Clicked A Few Times
8:22:59 AM PST - Sat, Feb 4th 2017
Ok, Edoapra, it seems I need your help with the compilation of NWChem.
My current build file is already here

http://pastebin.com/dREhnUnY

and the machine in which I want to install and run the code is the following

http://www.fz-juelich.de/ias/jsc/EN/Expertise/Supercomputers/JURECA/Configuration/Configur...

I already tried to comment the optimisation options and to switch to gcc, but still I get the wrong results in my test calculation on Au10. Maybe something went wrong with the MPI settings/location... Sorry but I am not expert, can you give me any suggestion?

Thanks a lot

Alessandro

Forum Vet
1:56:07 PM PST - Mon, Feb 6th 2017
Alessandro
I have tried to reproduce your problem, but to no avail so far.

Since you are using way too many processors for your small problem (1024!), could you please try to use just 4 processes and see if the problem is still there? I would curious to see a run just using a single process.
Thanks, Edo

Clicked A Few Times
7:10:39 AM PST - Wed, Feb 8th 2017
Dear Edo,
sorry for my late reply but I was very busy. I tried to run the code on 1 single node, but I still have the same problem as in a parallel running...

Thanks for your help
Alessandro

Forum Vet
7:53:25 PM PST - Thu, Feb 9th 2017
Fix available
Alessandro,
I managed to reproduce your failure with ifort 2017
Here is a patch that should fix this problem

http://nwchemgit.github.io/download.php?f=Hfmke.patch.gz

How to apply this patch?

cd $NWCHEM_TOP
wget http://nwchemgit.github.io/images/Hfmke.patch.gz
gzip -d Hfmke.patch.gz
patch -p0 < Hfmke.patch


Forum >> NWChem's corner >> Running NWChem
Jump to page 12Next 16Last