Fitting Cu2+ and 4 Nitrogen with Chili and ESFit

[olback]

Heyho,

I am new to this forum, so at first I'd like to thank the developers for such a great software it is really helpful.

So let's start with my problem. I think it should not be a problem of coding, because my script works under certain conditions. I'm trying to simulate and fit a Cu2+ solution spectrum with Chili and ESFit. The Cu2+ is coordinated by 4 nitrogens, so far nothing complicated. If I try to start the simulation it works as long as I just try to simulate Cu2+ and 1 nitrogen, but if I go for more than 1 nitrogen it seems that the calculation takes a huge amount of ressources, cause it stops always at the initial step of fitting. For example, if I use "simplex fcn" and "perturb" it says "initial simplex ..." even after hours nothing more happens. This happens with all other methods as well. So I thought, I need more ressources and went on to a cluster and used 32 cores and 64 GB memory, but still the same problem nothing happens except of the initial calculation.

I tried using various fitting methods (simplex, levmar, etc.), I tried "pertrub" and "matrix" and now I don't know what to do anymore. Like I said with just one nitrogen it works fine, it takes a bit longer than with just Cu2+ but it works.
Is there a way to get this working or is it really a ressource problem and I need to go for something like 128 cores and a huge amount of memory?

Here's the output of the matlab console when I start the script:

>> TestChili
-- esfit ------------------------------------------------
Simulation function: chili
Problem size: 1 spectra, 1 components, 14 parameters
Minimization method: Nelder/Mead simplex
Residuals computed from: data as is
Scaling mode: lsq0
---------------------------------------------------------
initial simplex...


I should add that if I use garlic or pepper it works perfectly, but due to being in a slow motion regime the fit is not really great.

I am thankful for any kind of help.

Have a nice day,

Olli

[Stefan Stoll]

Olli, Cu+4xN is a really large spin space for slow-motion simulations using the Stochastic Liouville equation (SLE). No wonder that you run out of memory. 648 spin states times 100s of rotational basis functions gives 100'000x100'000 matrices. So, although in principle possible, there is no chance to run this full system using the SLE on a typical computer or cluster unless you have plenty of memory.

Now there is another way: The nitrogen splittings in your complex are most likely small and therefore are in the fast limit for typical rotational correlation times. You can then treat the nitrogens as perturbation and use the SLE only for the electron spin + Cu. So in the first step, simulate your Cu slow-motion spectrum, then convolve it with the nitrogen splitting pattern, which you can calculate separately using garlic.

For fitting, you can write your own fitting function. See the ES documentation here.

[olback]

Hi Stefan, thank you for your answer.

Yep, that seems to be a legit conclusion, that there's no way to run it on a typical computer

Ok, then I'm going to try to write a fitting function by myself, thank you for your help.

Best regards,

Olli