nlinfit gives no result

Hello!

When trying to fit data with nlinfit, I get the following error message:

Warning: Rank deficient, rank = 0, tol = 0.000000e+00. > In nlinfit>LMfit (line 574) In nlinfit (line 276) In TRMC_GuiFit_v2_tmax_SD_nlfit>FITANDPLOT_Callback (line 491) In gui_mainfcn (line 95) In TRMC_GuiFit_v2_tmax_SD_nlfit (line 42) In matlab.graphics.internal.figfile.FigFile/read>@(hObject,eventdata)TRMC_GuiFit_v2_tmax_SD_nlfit('FITANDPLOT_Callback',hObject,eventdata,guidata(hObject)) Warning: Some columns of the Jacobian are effectively zero at the solution, indicating that the model is insensitive to some of its parameters. That may be because those parameters are not present in the model, or otherwise do not affect the predicted values. It may also be due to numerical underflow in the model function, which can sometimes be avoided by choosing better initial parameter values, or by rescaling or recentering. Parameter estimates may be unreliable. > In nlinfit (line 373) In TRMC_GuiFit_v2_tmax_SD_nlfit>FITANDPLOT_Callback (line 491) In gui_mainfcn (line 95) In TRMC_GuiFit_v2_tmax_SD_nlfit (line 42) In matlab.graphics.internal.figfile.FigFile/read>@(hObject,eventdata)TRMC_GuiFit_v2_tmax_SD_nlfit('FITANDPLOT_Callback',hObject,eventdata,guidata(hObject)) Warning: Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND = 1.999277e-56. > In nlparci (line 104) In TRMC_GuiFit_v2_tmax_SD_nlfit>FITANDPLOT_Callback (line 492) In gui_mainfcn (line 95) In TRMC_GuiFit_v2_tmax_SD_nlfit (line 42) In matlab.graphics.internal.figfile.FigFile/read>@(hObject,eventdata)TRMC_GuiFit_v2_tmax_SD_nlfit('FITANDPLOT_Callback',hObject,eventdata,guidata(hObject)) Warning: Imaginary parts of complex X and/or Y arguments ignored > In TRMC_GuiFit_v2_tmax_SD_nlfit>FITANDPLOT_Callback (line 657) In gui_mainfcn (line 95) In TRMC_GuiFit_v2_tmax_SD_nlfit (line 42) In matlab.graphics.internal.figfile.FigFile/read>@(hObject,eventdata)TRMC_GuiFit_v2_tmax_SD_nlfit('FITANDPLOT_Callback',hObject,eventdata,guidata(hObject)) >>

Posting the code is a bit tricky, but I will do my best.

      [xfit,resnorm, Jacob, CovB, MSE] = nlinfit(  handles.timecorrcrop,handles.datacorrcrop',@funexp2conv410, handles.x0 ); 

handles.timecorrcrop = -1e-7:250e-12:1e-6;
handles.a1_GUESS_VALUE = 0.6;
handles.G1_GUESS_VALUE = 2.5e-8;
handles.a2_GUESS_VALUE = 0.4;
handles.G2_GUESS_VALUE = 3.5e-7;
handles.ETAMU_0_GUESS_VALUE = 0.04; 

handles.x0 = [ handles.a1_GUESS_VALUE; 1/handles.G1_GUESS_VALUE; handles.a2_GUESS_VALUE; 1/handles.G2_GUESS_VALUE; handles.ETAMU_0_GUESS_VALUE ];

With the function being:

   function f = funexp2conv410( param, t )

      RC = 18e-9; % Time constant of cavity cell
      a1 = param(1); % Parameters for bi-exponential decay
      D1 = param(2);
      a2 = param(3);
      D2 = param(4);
      etamu1 = param(5); %starting value of conductivity in cm2/Vs

      assignin('base', 'D1', D1);
      assignin('base', 'D2', D2);

       filename1 = 'pulse410.txt';
       pathname1 = 'D:\PulseShapes\';
       pulse410 = importfile1([pathname1, filename1]);

      [P,I] = max(pulse410(:,1)); %find the coordinates of the maximum value of t

       t_in = find(t==P); 
       t_x = t(1:t_in); 
       pulse410s=spline(pulse410(:,1), pulse410(:,2), t_x); % make the pulse and the experimental data have the same time scale

      f1x = conv(a1 * exp(-t*D1) + a2 * exp(-t*D2), pulse410s./75.5856) 
      f1(1:4000) = etamu1*f1x(1:4000);
      assignin('base','f1',f1);

      f2x = conv(exp(-t/RC),f1);
      f(1:4000) = f2x(1:4000)/72.501;

I am not sure what the problem is. Are there too many convolutions in the fit? Thanks for your help!

[Eric]

In your function funexp2conv410, your line

 t_in = find(t==P);

runs the risk of being empty, thereby causing output f to always be zeros, which is likely why it is complaining. Consider replacing those lines with

 t_x = t(t<=P);

or whatever makes sense for your situation to avoid problems. Not sure if that's the only issue.

[Silke]

Hi Eric,

thanks for this comment. It did not solve the problem totally, but at least the fit is now starting and I actually get different values for the fit (even though they make no sense and the curves are not overlapping at all). It seems that at least the fit tries to do something, which I did not see before. Anyway, the error message in the end is still the same, so if someone has other ideas, please come up with them.

[Silke]

Indeed, there were quite some more bugs in the fitting related to the time. I think I solved them. Thanks for your help!

[Eric]

Is the problem solved? If so, I will re-post my comment as an answer.

[Silke]

Yes, the problem is solved. Your answer gave the right indication on where to find the problem, and partly solved it. Thank you!

Answer 1

[Eric]

In your function funexp2conv410, your line

 t_in = find(t==P);

runs the risk of being empty, thereby causing output f to always be zeros, which is likely why it is complaining. Consider replacing those lines with

 t_x = t(t<=P);

or whatever makes sense for your situation to avoid problems. Not sure if that's the only issue.

EDIT: Not a direct solve, but led OP to the answer. See comments to question.