Non-Linear Curve fit: Error using svd

I am conducting a non linear fit with multiple parameters, the data set is pretty big with 622 elements. The code I've used is shown below.

G = 6.67*(10^-11);
H = 73.8;
fun = @(b,X) ((100.*G.*(H^2)).*((b(1).^3)./X).*((log(abs(1+(X./b(2))))-((X./b(2))./((X./b(2))+1)))./(log(abs(1+(b(1)./b(2))))-((b(1)./b(2))./((b(1)./b(2))+1))))) + ((4.*pi.*G).*b(3).*b(4).*((X./(2.*b(4))).^2).*((besseli(0,(X./(2.*b(4)))).*besselk(0,(X./(2.*b(4)))))-(besseli(1,(X./(2.*b(4)))).*besselk(0,(X./(2.*b(4))))))) + (G.*22.665.*(b(5).^2).*b(6));

opts = statset('nlinfit');
%opts.RobustWgtFun = 'bisquare';

b0 = [5000; 3; 340; 90; 300; 600];
mdl = fitnlm(x,y,fun,b0,'Options',opts);

I get the following error:

Error using svd

Input to SVD must not contain NaN or Inf.

Error in internal.stats.isEstimable (line 108)

[V,Sx,U] = svd(X,0);

Error in NonLinearModel/fitter (line 1146)

internal.stats.isEstimable(eye(numel(model.Coefs)),'DesignMatrix',J_out,'TolSVD',TolSVD);

Error in classreg.regr.FitObject/doFit (line 94)

model = fitter(model);

Error in NonLinearModel.fit (line 1446)

model = doFit(model);

Error in fitnlm (line 99)

model = NonLinearModel.fit(X,varargin{:});

Error in Curve_Fit_Take_2 (line 9)

mdl = fitnlm(x,y,fun,b0,'Options',opts);

I believe the error is with some NaNs during processing, how do I go about this as I'm very sure there are no NaN values in my data set. Let me know if you also have a better way to find a non-linear fit.

Thanks in advance !!

Answer 1

[Matt J]

There are regions of the b(i) space where your model function will evaluate to a complex number, or to a non-finite value. You need to constrain the range of the search so that those b(i) are not tested. Possibly, you intend for all the b(i) to be positive for example. For that, you will probably need an alternative solver that supports bounds, e.g. lsqcurvefit.

[Adhitya Shreyas]

Thank you Matt

[yongguang zhou]

Thank you for helping me tackle similar problem as well.