fminsearch for complex input for curve fitting

Dear all, I would like to fit experimental data with custom equations. I aim for the best fit of the theoretical curve to the experimental data by minimizing the residuals and use fminsearch to find minimal error.

The attached code works well for the real inputs

[R, fval] = fminsearch(err, 2.11)% finds the minimum of err But it fails for the

[R, fval] = fminsearch(err, 2.0-0.064i)

Help for fminsearch suggests to input split into real, imaginary parts and work to obtain the best fit. I have a little idea of doing this.

Could somebody help me with this problem? Thanks all.

[VISWANATH]

I used

% Absolute error err = @ (n1) sum((abs(Rtot_s(n1)).^2 - S).^2 + (abs(Rtot_p(n1)).^2 - P).^2) % Minimum search for the error value x0 = complex(2.1,-0.07) [R, fval, exitflag] = lsqnonlin(@(x)err(x), real(x0),imag(x0) )

O/p R =

    2.1084

fval =

   28.0003

exitflag =

    5.2915

Answer 1

[Matt J]

Assuming your err function is real-valued,

    [R, fval] = fminsearch(@(x) err(  complex( x(1),x(2) )  ),   [2.0,0.064])

[VISWANATH]

tried you suggestion, seems not working Matt.

[Matt J]

Means little. You haven't shown us what you've actually done. You haven't shown what you've actually observed.

[VISWANATH]

When i used

[R, fval, exitflag] = lsqnonlin(@(x)err(x),complex(2.11,-0.064) )

gave the best values for fit but solution did not converge!!! Where do you think i am going wrong?

R =

   2.1151 - 0.0640i

fval =

   1.6100e-04

exitflag =

    0.0127

[Matt J]

The objective function must be a mapping from reals to reals. The initial guess x0 must also be real. So, complex(2.11,-0.064) is not legal as an initial point.

Answer 2

[VISWANATH]

Hi Matt, It looks like the following example may help me to solve my problem. If you have any suggestions i would be glad to hear from you. Thanks a lot. Fit a Model to Complex-Valued Data