faster NMR coupling calculation

[Attila Balázs]

Dear list, 

We would like to calculate a big couple of molecules H spin-spin couplings. But the test run looks wery slow. This running 48 hours. Here is the input file:

echo

memory total 16 gb

restart haloperidol_nmr_couplings

title "haloperidol NMR spin-spin coupling test"

#https://www.chem.wisc.edu/areas/reich/nmr/h-data/hdata.htm

charge 0

geometry units angstroms print xyz autosym

Cl        -2.95350        9.35840        0.50610

F          5.30580        5.24650       -2.05990

O         -3.63330        2.82170       -1.27210

O          4.38050        0.33690        1.93490

N         -0.76010        1.17340        0.43840

C         -2.66220        3.35770       -0.38640

C         -1.28960        3.07280       -1.05900

C         -2.77230        2.53150        0.93210

C         -0.75320        1.63380       -0.95570

C         -2.14350        1.12500        0.92040

C         -0.10960       -0.13570        0.60060

C         -2.76890        4.87820       -0.12060

C          1.42680       -0.04930        0.48170

C         -3.57090        5.68530       -0.94280

C         -1.98500        5.49770        0.86720

C          2.09010        0.83950        1.56450

C         -3.63450        7.06250       -0.74500

C         -2.04770        6.87490        1.06470

C          3.58770        1.06890        1.33590

C         -2.87670        7.65200        0.26190

C          4.01940        2.18770        0.42980

C          3.09070        2.96810       -0.28010

C          5.38910        2.45780        0.28050

C          3.52710        3.99430       -1.11470

C          5.81800        3.48580       -0.55570

C          4.88670        4.24970       -1.24910

H         -1.34130        3.36070       -2.11000

H         -0.52020        3.75240       -0.69100

H         -2.37950        3.06570        1.79470

H         -3.82270        2.44590        1.21280

H          0.26630        1.63960       -1.34210

H         -1.31070        0.94840       -1.59910

H         -2.74150        0.41440        0.34400

H         -2.14590        0.75450        1.94830

H         -0.33560       -0.52580        1.59590

H         -0.50990       -0.86280       -0.11020

H          1.82260       -1.06240        0.57580

H          1.71300        0.25150       -0.52680

H         -4.42240        2.62680       -0.75110

H         -4.15580        5.24920       -1.74090

H         -1.30780        4.92190        1.48230

H          1.59580        1.80730        1.63330

H          1.95550        0.39070        2.54830

H         -4.26590        7.67140       -1.37660

H         -1.44610        7.33960        1.83330

H          2.02520        2.79900       -0.20540

H          6.13500        1.87690        0.80900

H          2.81390        4.59460       -1.66060

H          6.87200        3.69370       -0.66950

end

basis

 * library 6-31+G**

end

scf

  direct filesize 0 memsize 500000000 

end

dft

  mult 1

  xc b3lyp

  grid fine

  direct

  noio

  odft

  convergence fast

  iterations 200 

end

#task dft optimize

task dft energy

task dft freq

property

  spinspin 5  28 29 31 32 33 34 35 36 37 38

end

task dft property

[Attila Balázs]

I accidentally sent it before I finished the full letter, here are some comments:

• I don't want to use smaller basis.  We would like put the calculations in to an article.
• We have ~130 molecules for calculating, 
• System is an Ubuntu 18.04.5, officially .deb package from the github. I do not compile it for myself, but if it's neccesearly I can do it.
• This run 8 processor core, 16Gbyte ram, minimised the disk usage.
• When I did not setup the disk usage, the temporarly files size was 600-700Gbyte. And continously read and write it.

So, any idea how can the NWchem do faster this calculation?

[Edoardo Aprà]

Could you change the basis set from cartesian (default) to spherical? This is not going to have only a modest improvement in time to solution, but it could improve numerical stability (therefore requiring fewer SCF cycles to converge and/or CPHF iterations)

https://nwchemgit.github.io/Basis.html#-spherical-or-cartesian

basis spherical

 * library 6-31+G**

end

Are you using 8 cores in your runs? Is your computer free of any other heavy process running?

If you have 16Gb total and use 8 processes, then you should ask only 16/8=2GB in your input line

[Edoardo Aprà]

Why did you set

odft

mult 1

?

If it is a singlet, you make the calculation 2x more expensive by choosing odft.

[Attila Balázs]

Hi Edoardo, 

The computer what I am using is for calculations, so now I've got some source to calculate. (24core, 50Gbyte ram, what I can use now)

The answer your second question, is here: https://nwchemgit.github.io/Special_AWCforum/st/id2011/Indirect_Spin-Spin_Coupling_Tens....html

The calculation run in to an error, without the odft flag. I tryed it now. I have the same error message:

Indirect spin-spin coupling not calculated for RHF

      wave functions: needs UHF wave function

I don't know this is a bug or a feature?

[Edoardo Aprà]

You are right, I forgot about this.

The only other suggestion I have is to check how NWChem was installed ... did you compile it yourself?

What value of ARMCI_NETWORK did you choose?

Do you plan to perform just single node runs or multi-node runs, too?

[Attila Balázs]

This is the standard release from github:  nwchem_7.0.2-1_amd64.ubuntu_bionic.deb, nothing special. I don't want to compile. 

But if necessarly I can compile it.

And we  perform a single node run.