Singular matrix (or other numeric) error by fitting an implicit function with ODR-Package
gr4nd
I'm new to Igor and this Forum, but by now i've read a lot of the Igor's user help files, since i got a big problem with the ODR-Fitting Package.
I got a function to fit to my measured data and tried this with the fitfunc /ODR=3 command, since the function is implicit and i want to know the coefficients.
But I always get a "Singular matrix or other numeric error". I checked all the hints that gave the "explain error" button, but nothing helped. But maybe i misunderstood something.
By the way, sorry for my bad english... :(
So here is what i did.
This is my function, where j and U are the variables (measured data). You will see, that u cant get it into U=f(j) or j=f(U):
j=Isc-j0*(exp((U+j*Rs)/(n*k*T))-1)-(U+j*Rs)/Rp
Thats why i subtracted j to get it into the /ODR=3 required form.
0= Isc - j0*..... - (U+j*Rs)/Rp - j
The "coefficients" i want to know are the following (starting values in brackets).:
j0 (5.6E-9), Rp(450), Isc(0.0188)
All other values (Rs,n,k,T) are constants by now. Once i got this to work, i will try to get Rs,n and T as coefficients too, but i think it's hard enough by now.
The starting value's should be fine, because if i plot the function with these values (and of cause the constants i found by trying a long time ^^), it nearly completely matches the measured data (by eye).
This is what my procedure window is like.
#pragma rtGlobals=1 // Use modern global access method.
constant Rs=2.8
constant n=1.5
constant T=314
Function I_U_Kennlinie(w,U,j) : FitFunc
Wave w
Variable U
Variable j
//CurveFitDialog/
//CurveFitDialog/ Coefficients 3
//CurveFitDialog/ w[0] = j0
//CurveFitDialog/ w[1] = Rp
//CurveFitDialog/ w[2] = Isc
return w[2]-w[0]*(exp((U+j*Rs)/(n*8.617343E-5*T))-1)-(U+j*Rs)/w[1]-j
end
constant Rs=2.8
constant n=1.5
constant T=314
Function I_U_Kennlinie(w,U,j) : FitFunc
Wave w
Variable U
Variable j
//CurveFitDialog/
//CurveFitDialog/ Coefficients 3
//CurveFitDialog/ w[0] = j0
//CurveFitDialog/ w[1] = Rp
//CurveFitDialog/ w[2] = Isc
return w[2]-w[0]*(exp((U+j*Rs)/(n*8.617343E-5*T))-1)-(U+j*Rs)/w[1]-j
end
And now the command to get the fit starting:
FuncFit/ODR=3 I_U_Kennlinie, w /X={U_Werte, I_Werte}/XD={U_WerteFit,I_WerteFit}
My only explanation for the error is, that j0 is simply to small and has only little of influence on the function (Great changes are needed to modify the function's graph with a change to j0) for getting a change in the derivatives. The constraints range of j0 is about 1e-12 to 1e-6.
But ODR=3 does'nt support an epsilon wave to be used.
Does anyone have a solution for my problem? I would really appreciate any help.
Thanks a lot
Martin
Making implicit fits work can be tricky, and it's hard to give advice since this is such a new feature of Igor. Please send me a copy of your experiment file with the data and fitting function so that I can look at it. Send it to support@wavemetrics.com.
John Weeks
WaveMetrics, Inc.
support@wavemetrics.com
October 7, 2008 at 11:34 am - Permalink