lsqnonlin projectBox error when using OutputFcn

Hello,

I am trying to check the number of iterations and residuals after each iteration of lsqnonlin. I have more unknowns than equations and have set it to use the levenberg-marquardt algorithm. I also made the following output function:

function [stop]=residual_check_fun(x,optimValues,state)

keyboard
iteration=optimValues.iteration;
residual=optimValues.residual;
if iteration>5 && residual>10000
    stop=true;
else
    stop=false;
end
end

This function works when I have one equation and one unknown, but does not work for more unknowns than equations. The error I am getting is the following:

Unable to perform assignment because the left and right sides have a different number of elements.

Error in projectBox (line 39)

dx(activeLB) = lb(activeLB) - x(activeLB);

Error in levenbergMarquardt>@(x,dx)projectBox(x,dx,lb,ub) (line 125)

project = @(x,dx) projectBox(x,dx,lb,ub);

Error in levenbergMarquardt>computeFirstOrderOpt (line 385)

a = norm(project(XOUT,-gradF),Inf);

Error in levenbergMarquardt>callOutputAndPlotFcns (line 410)

optimValues.firstorderopt = computeFirstOrderOpt(project,xOutputfcn,gradF);

Error in levenbergMarquardt (line 147)

[optimValues, stop] = callOutputAndPlotFcns(outputfcn,plotfcns,caller,reshape(XOUT,xShape),...

Error in lsqncommon (line 186)

levenbergMarquardt(funfcn,xC,lb,ub,flags.verbosity,options,defaultopt,initVals.F,initVals.J, ...

Error in lsqnonlin (line 260)

lsqncommon(funfcn,xCurrent,lb,ub,options,defaultopt,optimgetFlag,caller,...

Answer 1

[Alan Weiss]

The error is because optimValues.residual is a vector, not a scalar. Possibly you want to write norm(optimValues.residual).

Alan Weiss

MATLAB mathematical toolbox documentation

[Dennis Nikitaev]

But that line is after the keyboard line. The error occurs before the body of the function. It seems that the error is in between the lsqnonlin and the output function.

Answer 2

[Bianca Romanski]

tl;dr Try updating your Matlab version.

Hello Dennis,

I got exactly the same error as you did. To understand the error better, I reproduced it with some code as simple as possible:

% Call optimization
x0 = [0 1];
options = optimoptions(@lsqnonlin, 'Algorithm','levenberg-marquardt','OutputFcn',@outfun,... 
    'Display','iter');
[xsol,fval] = lsqnonlin(@objfun,x0,[-1 -1],[1 1],options);
% [xsol,fval] = lsqnonlin(@objfun,x0,[],[],options);
 
 function stop = outfun(x,optimValues,state)
     stop = false;        
 end
 
 function f = objfun(x)
     f = exp(x(1))*(4*x(1)^2 + 2*x(2)^2 + 4*x(1)*x(2) +... 
                    2*x(2) + 1);
 end

I found out, that the projectBox error occurs only when constraints are used. Without constraints the script worked perfectly fine. As I couldn't think about any reason why output functions only work for unconstraint problems, I was wondering whether this might be a bug in the Matlab function. As bugs are fixed sometimes, I updated my Matlab and ran the above code in Matlab 2022a (instead of 2020b) and it worked! Maybe updating your Matlab version will also solve your issue. :)

[Bruno Luong]

@Bianca Romanski Could you try to run with lo and hi reshaped as column vector p, 2020b?

[Bianca Romanski]

I changed the 5th line, so that upper and lower bounds are column vectors.

[xsol,fval] = lsqnonlin(@objfun,x0,[-1; -1],[1; 1],options);

That didn't change anything, the same error occurs.

[Bruno Luong]

Thanks