Issue with lsqnonlin function solve

Hello!

I am working with some algorithms to find optimal solution by changing 3 parameters.

The function I want to minimize to find optimal parameters is: , where a,b,s are fixed values and x_1 to x_5 are my parameters.

First I did example for one case - which works:

calibrator=@(x)(fix_value-f(a,b,c,x(1),x(2),x(3),x(4),x(5)))^2/sigma^2;
startparameters = [0.2 2 0.6 0.5 2 ];

options = optimset('Display','iter');
startparameters = [0.2 2 6 0.25 1 ];
calibrator=@(x)(fix_value-f(a,b,c,x(1),x(2),x(3),x(4),x(5),N))^2/sigma^2;
fminsearch(calibrator,startparameters, options)

Now I want to do the same for set of equations which are non linear. Now inputs are columns (fix_value, sigma) and function f returns column as well. I found function lsqnonlin as a way to solve my issue but id does not work:

calibrator=@(x)(fix_values-fs(a,b,c,x(1),x(2),x(3),x(4),x(5))).^2./sigmas.^2;
startparameters = [0.2 2 0.6 0.25 1];
options = optimoptions('lsqnonlin', 'Display', 'iter');
xopt = lsqnonlin(calibrator,startparameters, ...
    [eps eps -1+eps eps eps  ],
    [Inf Inf 1-eps Inf  Inf  ]);
    
Error using lsqncommon (line 15)
Objective function is returning undefined values at initial point. lsqnonlin cannot continue.

Error in lsqnonlin (line 260)
   lsqncommon(funfcn,xCurrent,lb,ub,options,defaultopt,optimgetFlag,caller,...
 

Any idea how to handle this issue?

Answer 1

[Walter Roberson]

fix_values is a column. You subtract something from it (suppose it is a column as well), getting a column. Then you try to ^2 the column. However ^ is the Matrix Power operator, which can only be used with square matrices. You need to use .^2 instead.

Your sigma is also a column, you said, so you need sigma.^2

You would then have column / column which would be a matrix fitting operation. You need ./ instead of /

[Walter Roberson]

We need your data and your code to test with.

[Walter Roberson]

FT_prob_dens is needed by heston_call

[Walter Roberson]

You deleted the link to your github code and data, which makes it impossible for anyone else to have a look to see if they can see something I missed, or if they can make educated guesses about what the missing FT_prob_dens() function does and so potentially get further without waiting for you to make the code available.