Resources overflow in manganese complex simulation

[fabio]

Dear all,
we're trying to reproduce EPR spectra of a Manganese-Oxygen Complex starting from a DFT simulation with ORCA package. So far we obtained a .prop file in attachment here:

system.prop

ORCA output prop file

(4.52 KiB) Downloaded 366 times

using the eprnmr section in the ORCA input

Code: Select all

%eprnmr
  gtensor 1
  nuclei = all Mn {aiso, adip, aorb, fgrad, rho}
end

The lines we're using in matlab are the following:

Code: Select all

Sys=orca2easyspin('system.prop')
Sys.lwpp=1
Exp.Range = [150 520]
Exp.mwFreq = 9.5
[x,y] = pepper(Sys,Exp)

A similiar system with spin 1/2 works smoothly with both pepper and garlic and bring a good results in a few seconds. This system instead takes many hours of computation: all the threads of the machine are automatically involved by matlab and with an increasing ram consumption until the overflow crash the system. The same happened on multiple machines with different Matlab versions. I suppose something is missing in our input but it's not very clear to me, so I would like to ask for suggestions.

Thanks in advance!

[Stefan Stoll]

The system is very large. I am afraid there is no way around this, you need to find a system with more memory. We'll make a note to check for possible memory savings in the the next version.

[fabio]

Thanks for the kind reply! I'm actually wondering if the disproportion between the two system can be explained just by the different size. The 1/2 spin system take a few second and a few MB of ram, while this 5/2 spin system (which has basically the same geometrical structure) ran for about 36 hours on a supercomputer, it took 40 threads (all the avaliable ones) and 130 GB of RAM. Do you think we need to keep going with the simulations or we're missing something else?

[Stefan Stoll]

Have you tried with Opt.Method = 'perturb2'? I would test this with 2 of the 4 nuclei to see if it doesn't introduce error.