Property Calculation failed

[Peter Walton]

Hi there,

I have been attempting a series of property calculations on various compounds (all less than 100 heavy atoms) and have been struggling to get the calculation to complete. I have tried giving the calculations more memory (so far up to 12000mb per core and 8 cores) and I still get the same error:

      ---------------------------------------------------
      Calculating Fermi-Contact and Spin-Dipole Responses
      ---------------------------------------------------

                                NWChem CPHF Module
                                ------------------


  scftype          =     UHF
  nalpha           =      123
  nbeta            =      123
  variables        =   105288
  # of vectors     =      767
  tolerance        = 0.10D-03
  level shift      = 0.00D+00
  max iterations   =       50
  max subspace     =     7670

 SCF residual:   3.311812999091719E-004


Iterative solution of linear equations
  No. of variables   105288
  No. of equations      767
  Maximum subspace     7670
        Iterations       50
       Convergence  1.0D-04
        Start time   1581.2


   iter   nsub   residual    time
   ----  ------  --------  ---------
0:CreateSharedRegion:kr_malloc failed KB=: -555836

My input file is as follows:

echo
START
memory 12000 mb
geometry noautosym noautoz
C  0.000000  0.000000  0.000000
O  1.422300  0.000000  0.000000
C  1.964700  0.000000  1.332700
C  1.674300  -1.244200  2.113000
C  0.887500  -1.395700  3.317000
C  0.138600  -0.336700  4.083700
C  -1.075900  -0.561300  4.697700
C  -1.604100  0.591200  5.367100
C  -2.798600  0.750100  6.114100
C  -3.131700  1.982300  6.697300
C  -4.370100  2.109900  7.533700
C  -2.266000  3.062200  6.544000
C  -1.098900  2.941100  5.782400
O  -0.274900  4.044700  5.682300
C  -0.608100  4.784200  4.507800
C  -0.760400  1.693900  5.215800
S  0.634400  1.276700  4.294200
C  0.891300  -2.745300  3.647100
C  0.472200  -3.581000  4.701300
N  0.371600  -3.154100  6.037000
N  0.300400  -4.917000  4.519800
C  0.823300  -5.433900  3.451700
N  1.499000  -4.826100  2.480700
N  1.502700  -3.396100  2.615600
C  2.071900  -2.529300  1.724300
C  2.868500  -2.975900  0.534400
N  4.262500  -3.335700  0.854100
C  5.025400  -2.231400  1.437900
C  6.401900  -2.753400  1.858300
N  7.106900  -3.398900  0.771800
C  6.426300  -4.140600  -0.163200
O  7.054300  -4.909200  -0.893700
C  4.929800  -3.853600  -0.358500
H  -0.389500  0.833500  0.592700
H  -0.394300  -0.949300  0.373200
H  -0.338400  0.123400  -1.032600
H  1.634000  0.913400  1.824100
H  3.050800  0.098400  1.224700
H  -1.585400  -1.517800  4.659300
H  -3.454200  -0.109300  6.241000
H  -5.160600  1.448200  7.164200
H  -4.761200  3.132300  7.499700
H  -4.148000  1.850600  8.573400
H  -2.489400  4.016200  7.015200
H  -1.625700  5.185000  4.572600
H  -0.495800  4.179900  3.601500
H  0.083400  5.629200  4.437700
H  0.216700  -2.171500  6.177600
H  -0.197000  -3.788900  6.574600
H  0.731100  -6.516200  3.270800
H  2.363200  -3.837800  0.079000
H  2.857100  -2.189700  -0.230700
H  5.134600  -1.408500  0.719900
H  4.543900  -1.832600  2.336600
H  6.303500  -3.490300  2.664100
H  7.031300  -1.934700  2.220700
H  8.062200  -3.696300  0.910400
H  4.848700  -3.142200  -1.189900
H  4.457700  -4.798000  -0.652900
end
basis
 * library 6-31G*
end
dft
 direct
 xc b3lyp
 fukui
 print "Fukui information"
 odft
end
property
 all
end
task dft property

Is there anything I'm missing that can help these calculations finish properly?

Kind Regards

Peter

[Edoardo Aprà]

Peter

Could you provide details about the NWChem installation (if NWChem was installed by compiling the source code, please provide the settings used)and about the hardware you are using (for example, amount of memory available)

[Edoardo Aprà]

Could you try to add the following line just before the "task dft property" line

set fock:mirrmat f

In other words, the last lines of the input file should be

...

end

set fock:mirrmat f

task dft property

[Edoardo Aprà]

The root cause of your memory problem is likely to be due to the fact the in the calculation of spin-spin coupling all the atom pairs of your molecule are considered

https://nwchemgit.github.io/Properties.html#nmr-and-epr

Since you have 59 atoms, this results in 59*58=3422 atom pairs to be considered. Because of that, NWChem ends up allocating around a few thousand distributed matrices.

The fix for this input file is to specify the number of pairs needed.

For example, "spinspin 1 1 2" will calculate the coupling for one pair, and the coupling will be between atoms 1 and 2).

[Peter Walton]

Dear Edoardo,

Many thanks for your help on this. I'll set off the calculation with the changes listed above. The installation of NWChem in on an HPC cluster so I'll ask the admin for information about the system. I'll let you know how these changes work. 

Kind Regards,

Peter