lsqnonlin abort due matrix multiplication error

When using lsqnonlin I get an error message saying:

" Incorrect dimensions for matrix multiplication. Check that the number of columns in the first matrix matches the number of rows in

the second matrix. To perform elementwise multiplication, use '.*'.

Error in atamult (line 23)

V = A'*Y;

Error in snls (line 396)

newgrad = feval(mtxmpy,newA,newfvec(:,1),-1,varargin{:});

Error in lsqncommon (line 164)

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

Error in lsqnonlin (line 253)

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

Error in lsqnonlinTest14 (line 71)

[x,resnorm,residual,exitflag,output] = lsqnonlin(problem);"

myFunction does the following:

• read an input vector from lsqnonlin
• create an input file for external finite element software
• start external solver, reads results, calculates error compared to experimental results
• returns error as vector (err in the .mat file attached)
• my function only contains matrix multiplication

lsqnonlin is implemented with these options:

set.scale=[1000 100];

problem.objective = 'myFunction';
problem.x0 = [20 250].*set.scale; 
problem.lb = [5 220].*set.scale;    
problem.ub = [40 320].*set.scale;   
problem.solver = 'lsqnonlin';
problem.options = optimoptions(@lsqnonlin,'UseParallel',true);
problem.options.Algorithm = 'trust-region-reflective';
problem.options.PlotFcn = 'optimplotstepsize';
problem.options.Display = 'iter-detailed';
problem.options.StepTolerance = 1e-1;
problem.options.Diagnostics='on';
problem.options.OptimalityTolerance = 1e-3;

[x,resnorm,residual,exitflag,output] = lsqnonlin(problem);

The error appears randomly, sometimes after 6, 20...n iterations of lsqnonlin.

From the command line I can reconstruct the following:

                                    Norm of      First-order 
Iteration  Func-count     f(x)      step          optimality
1          6     2.83613e+11        8.95184       8.57e+11    
2          9     1.93657e+11        5.54999       6.05e+11 
3         12     1.08155e+11        7.76516       4.04e+11     
4         15       7.431e+10         3.7301       4.13e+11    
5         18     4.42101e+10        4.20724       2.14e+11          
6         21     2.90596e+10        2.85327        1.4e+11  

So finally my questions are:

• are my options wrong?
• Is this a known issue?
• is my scaling correct?

Thank you!

[Walter Roberson]

What does your function do if the external solver returns an unexpected size ?

[Fabian Teichmann]

Do you mean size of the error vector or absolute value of the error values? Both have not happend so far. What should the function do in that case?

[Matt J]

It's not something that can be investigated here without being provided the means to run the optimization.

[Walter Roberson]

The evidence suggests that a size error is occurring, unless the multiplication error is before you call the external function?

[Fabian Teichmann]

Right. The error occurrs exactly after one call of myFunction endet sucessfully and the next call starts. I have checked this by writing textfiles at discrete points of the function...

[Walter Roberson]

I recommend using

dbstop if caught error

to track down the problem.