esfit parameterization workflow

[wfagan]

Hello EasySpin community,

I am a graduate student performing a multicomponent simulation of an aqueous system that includes hydroxyl, superoxide, and sulfate radicals. Radicals were trapped using DMPO. My goal of this analysis is to determine the percent composition of these 3 radicals. I am using garlic and esfit to accomplish this.

My question is regarding esfit parameterization workflow. Seeing that the component weights and spin system parameters (hfccs, Gaussian/Lorentzian linewidths, isotropic g tensors) can be varied and knowing that we should limit the number of varying parameters for a given fit, what is a good approach to deduce spin system parameters to accurately calculate radical-adduct weights? The reason I ask this is because I have varied different combinations of spin system parameters (i.e. vary g, and keep lwpp constant, and vice versa) that have good fits (correlation coefficient > 0.99) but the determined weights from these fits can vary plus/minus 10%. Is there an order of operations for adjusting certain spin parameters to reproducibly attain component weights? Perhaps a different optimization algorithm or target function would be more appropriate for my system.

Thank you for your help!

-William Fagan

Attached: sample code, a picture of the esfit GUI, data file.

Code: Select all

clear;
clc;
close all;

[Mag,RawData,params]=eprload('USPS40_25.DSC');

Mag=Mag/10; %Convert G to mT

Exp.Range=[Mag(1) Mag(end)]; % 345.5 to 357.514 mT
Exp.mwFreq=9.86; 
Exp.nPoints=1024; 
Exp.ModAmp=0.1; 

%DMPO-OH 
Sys1.g =[2.00 2.00]; 
Sys1.Nucs = '14N,1H';
Sys1.n=[1 1];
Sys1.A_N = mt2mhz([15.023]/10); % G --> MHz 
Sys1.A_H = mt2mhz([14.702]/10); 
Sys1.A = [Sys1.A_N, Sys1.A_H]; 
Sys1.lwpp=[0.1 0.0135];
Sys1.weight=1;

%DMPO-OOH
Sys2.g=[2.00 2.00];
Sys2.Nucs='14N,1H,1H';
Sys2.n=[1,1,1];
Sys2.A_N=mt2mhz([15.27]/10);
Sys2.A_H_beta=mt2mhz([10.579]/10);
Sys2.A_H_gamma=mt2mhz([1.166]/10);
Sys2.A=[Sys2.A_N,Sys2.A_H_beta,Sys2.A_H_gamma];
Sys2.lwpp=[0.1 0.1];
Sys2.weight=0.25;

%DMPO-SO4
Sys3.g=[2.00 2.00];
Sys3.Nucs='14N,1H,1H,1H';
Sys3.n=[1,1,1,1];
Sys3.A_N=mt2mhz([13.67]/10);
Sys3.A_H_beta=mt2mhz([10.24]/10);
Sys3.A_H1_gamma=mt2mhz([1.52]/10);
Sys3.A_H2_gamma=mt2mhz([0.79]/10);
Sys3.A=[Sys3.A_N,Sys3.A_H_beta,Sys3.A_H1_gamma,Sys3.A_H2_gamma];
Sys3.lwpp=[0.09 0.09];
Sys3.weight=0.25;

Vary1.g=[0.005 0.005];
Vary1.lwpp=[0.05 0.05];

Vary2.g=[0.005 0.005];
Vary2.lwpp=[0.05 0.05];
Vary2.weight=0.25;

Vary3.g=[0.005 0.005];
Vary3.lwpp=[0.09 0.09];
Vary3.weight=0.25;

FitOpt.Method='simplex fcn';
FitOpt.Scaling='lsq0'; 

esfit('garlic',RawData,{Sys1,Sys2,Sys3},{Vary1,Vary2,Vary3},Exp,[],FitOpt);

esfitGUI.PNG (42.42 KiB) Viewed 2960 times

[Stefan Stoll]

It looks like your simulation is quite over-parameterized given the simplicity of the spectrum. You need to find ways to determine some of the parameters independently (literature?). There is no optimization algorithm that can fix this problem. If your model has too many parameters, given the data, then you will end up with multiple minima and significant uncertainties.

[katarkon]

William, as it was noted by Stefan Stoll, there are too much parameters should to be varied in Your simulation.
Firstly, what the reason to use anisotropic g-factor for isotropic spectra?
Secondly, all of EPR parameters of the spin system like g-factor, HFC constants, etc should be defined once and should not to varied anymore. Only linewidth and weight of the spectrum should be varied if You want to obtain reproducible result of the spin quantation.
Thirdly, why gaussian component in the linewidth is used, moreover, its contribution proposed to be equal to lorentzian? It's wrong for my opinion. Lorentzian contribution should be predominant for organic radicals. I think that gaussian component should be either zero either at order less then lotetzian one and should not be varied anyway.
At last, the spectrum is qute noisy, and the noise level is comparable to the minor signals one. So, some drift of found weightts of these signals is expectable.
Good luck.
UPD. The values of g=2.0 and MWFreq=9.86 are extremely rough. They should be defined more closely.

[wfagan]

Stefan and katarkon,

Thank you for your inputs! I have since reduced the parameterization of the model, and I am getting more reproducible fits (and weightings).

katarkon, in regards to your questions
-The anisotropic g-factor was a mistake. I have since fixed this and found a g-value that works nicely. I have also defined the g-factor and MWFreq more specifically.
-Thanks for pointing out the discrepancy regarding the gaussian and lorentzian line broadening. I will ensure that the lorenztian contribution predominates, and refer to the literature to see how other researchers have accounted for the linewidths of these adducts.
-The spectrum is indeed noisy. These radicals are formed by high-intensity ultrasound, a system that favors OH radicals compared to the sulfate and superoxide radicals. I'll have to account for this uncertainty.

Thank you again for your help
-William Fagan

[siyavarma]

Stefan and katarkon,
Thank you so much for this great post.and regular updates.good work