Multi-Parameter Curve Fitting --- How to use lsqcurvefit ?

Hello!

I am major in chemistry and inexperience with Matlab.

Currently, I'm trying to fit a nonlinear curve with my model, so I use lsqcurvefit to get the parameters of my function. After finishing the code, I obtain this output:

"Error using lsqfcnchk (line 80)

FUN must be a function, a valid character vector expression, or an inline function object.

Error in lsqnsetup (line 46)

funfcn = lsqfcnchk(FUN,caller,lengthVarargin,funValCheck,flags.grad);"

--------------------------------------------------------------------------------------------------------------------------------------------------------------------

My curve fitting problem can be described as follows:

In order to display my code in a simple way, I set N=2, and I make a fake Objective function J_aim for testing. My code is as follows:

global Dim_N

Dim_N = 2;

%--------Objective Function (Curve)---------------

x = linspace(1.4,1.9,1000);

J_aim = 0.01./((x-1.6).^2+0.01^2)+0.05./((x-1.7).^2+0.05^2);

plot(x,J_aim);

hold on

%-----It is just a test, not the real curve I want to fit-----------

%-------Fitting Parameters----------------

gT = [0.5,0.2];

g = transpose(gT);

w = [1.6,0 ; 0, 1.7];

kappa = [0.01,0.05];

%-------Reshape these Parameters----------

para = [gT, kappa, reshape(w,[1,Dim_N^2])];

%-------Initial Gauss---------------------

g0 = [0.3,0.3]; w0 = [1,0; 0, 1]; kappa0 = [0.1, 0.1];

para0 = [gT, kappa, reshape(w,[1,Dim_N^2])];

para_opt = lsqcurvefit(FunModel(para0,x),para0,x,J_aim); % lsqcurvefit fitting method

%-------Fitting Model---------------------

function [J_mod] = FunModel(para,x)

global Dim_N

gT = para(1:Dim_N);

kappa = para(Dim_N+1:2*Dim_N);

w = zeros(Dim_N);

%-------Reconstruct the w matrix----------

point = 0;

for index = 2*Dim_N+1: length(para)

point = point + 1;

i = fix( (point-1)/Dim_N ) + 1;

j = point - (i-1)*Dim_N;

w(i,j) = para(index);

end

%-------My fitting model "J_mod"----------

J_mod = zeros(1,length(x));

g = transpose(gT);

Ht = w - 0.5i*diag(kappa);

for i=1:length(x)

temp = Ht - x(i)*eye(Dim_N);

J_mod(i) = imag( gT*inv(temp)*g ); % J_mod = imag( g^T.(1/(Ht-x)).gT )

end

end

--------------------------------------------------------------------------------------------------------------------------------------------------------------------

So my question is that why is Matlab telling me that the code is wrong? How can I fix my code and return the fitting parameters I want?

Any help will be appreciated. Thanks!

Answer 1

[Vaibhav]

Hi Siwei

The error is due to lsqcurvefit expecting a function handle as its first input.

Here is a way to modify the code:

1. Define FunModel as a separate function or as an anonymous function that accepts the parameters and the independent variable (x) as inputs and returns the dependent variable (J_mod).
2. Pass a handle to this function as the first argument to lsqcurvefit.

Here is a code snippet to help you get started:

global Dim_N
Dim_N = 2;
%--------Objective Function (Curve)---------------
x = linspace(1.4,1.9,1000);
J_aim = 0.01./((x-1.6).^2+0.01^2)+0.05./((x-1.7).^2+0.05^2);
plot(x,J_aim);
hold on

%-------Fitting Parameters----------------
gT = [0.5,0.2];
g = transpose(gT);
w = [1.6,0 ; 0, 1.7];
kappa = [0.01,0.05];
%-------Reshape these Parameters----------
para = [gT, kappa, reshape(w,[1,Dim_N^2])];
%-------Initial Gauss---------------------
g0 = [0.3,0.3]; w0 = [1,0; 0, 1]; kappa0 = [0.1, 0.1];
para0 = [g0, kappa0, reshape(w0,[1,Dim_N^2])];  % Fixed to use initial guesses

% Define the model function as an anonymous function
FunModel = @(para,x)modelFunction(para, x);

% Use lsqcurvefit with the function handle
para_opt = lsqcurvefit(FunModel, para0, x, J_aim);

Solver stopped prematurely.

lsqcurvefit stopped because it exceeded the function evaluation limit,
options.MaxFunctionEvaluations = 8.000000e+02.

%-------Fitting Model---------------------
function J_mod = modelFunction(para, x)
    global Dim_N
    gT = para(1:Dim_N);
    kappa = para(Dim_N+1:2*Dim_N);
    w = zeros(Dim_N);
    %-------Reconstruct the w matrix----------
    point = 0;
    for index = 2*Dim_N+1: length(para)
        point = point + 1;
        i = fix( (point-1)/Dim_N ) + 1;
        j = point - (i-1)*Dim_N;
        w(i,j) = para(index);
    end
    %-------My fitting model "J_mod"----------
    J_mod = zeros(1,length(x));
    g = transpose(gT);
    Ht = w - 0.5i*diag(kappa);
    for i=1:length(x)
        temp = Ht - x(i)*eye(Dim_N);
        J_mod(i) = imag( gT*(temp\g) );   % Corrected to use matrix left division
    end
end