Hi there,
I am currently trying to optimise a set of non-linear equations iteratively – I am trying to solve equation (4) in this paper (https://aip.scitation.org/doi/pdf/10.1063/1.3416910?casa_token=i26EckyYkb8AAAAA:UQ1axvJNde0OswFL9jgt24izc_sVX6XhurCYzC0M1BfNObq-K6OOOjuykKQuwt37mR6zALj-QrhD) where the author uses a non-linear least square optimisation approach to solve for 4 variables with 3 non-linear equations. Simply put, the 3 equations can be represented by the following:
1: y1 = sqrt(((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4)));
2: y2 = sqrt(((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4)));
3: y3 = sqrt(((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4)));
where y1, y2, y3, are the respective y values and xdata the x values that I know. Essentially, I have 3 sets of data points that I want to fit with the 4 unknown variables (x(1) to x(4)) through optimisation. I have tried the following:
x = optimvar('x',4);eq1 = y1 == sqrt(((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4)));eq2 = y2 == sqrt(((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4)));eq3 = y3 == sqrt(((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4)));prob = eqnproblem;prob.Equations.eq1 = eq1;prob.Equations.eq2 = eq2;prob.Equations.eq3 = eq3;show(prob)x1.x = [0.5e-3 0.5e-3 0.5e-3 0];[sol,fval,exitflag] = solve(prob,x1);disp(sol.x)
But these are the comments generated:
Solving problem using fsolve.Warning: Trust-region-dogleg algorithm of FSOLVE cannot handle non-square systems; using Levenberg-Marquardt algorithminstead. > In fsolve (line 323) In optim.problemdef.EquationProblem/callSolver In optim.internal.problemdef.ProblemImpl/solveImpl In optim.problemdef.EquationProblem/solve No solution found.fsolve stopped because the relative size of the current step is less than thevalue of the step size tolerance, but the vector of function valuesis not near zero as measured by the value of the function tolerance.
I was able to fit a single equation using the lsqcurvefit function by defining the following:
fun = @(x,xdata) sqrt(((1+4.*xdata.^2.*x(1)^2).^(-2) + x(3))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(3)));x0 = [0.1e-3 0.1e-3 0];options = optimoptions('lsqcurvefit','Algorithm','levenberg-marquardt');x = lsqcurvefit(fun,x0,xdata,y1,[],[],options)
However, when I tried defining the the 3 equations with a single function handle with the following codes:
fun = @(x,xdata) [sqrt(((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4))); sqrt(((1+4.*xdata.^2.*x(1)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4))); sqrt(((1+4.*xdata.^2.*x(2)^2).^(-2) + x(4))./((1+4.*xdata.^2.*x(3)^2).^(-2) + x(4)))];x0 = [0.5e-3 0.5e-3 0.5e-3 0.1e-5];options = optimoptions('lsqcurvefit','Algorithm','levenberg-marquardt');x = lsqcurvefit(fun,x0,xdata,y1,y2,y3,[],[],options)
It shows the following:
Warning: Length of lower bounds is > length(x); ignoring extra bounds. > In checkbounds (line 27) In lsqnsetup (line 77) In lsqcurvefit (line 210) Warning: Length of upper bounds is > length(x); ignoring extra bounds. > In checkbounds (line 41) In lsqnsetup (line 77) In lsqcurvefit (line 210) Exiting due to infeasibility: 4 lower bounds exceed the corresponding upper bounds.
Unfortunately, I am not too familiar with the optimisation procedures with MATLAB, so am wondering if anyone has any insights or suggestions on what else I should try. Thank you for your time.
Best Regards,
Wayne
Best Answer