Error: Failure in initial user-supplied objective function evaluation when using fminunc ?

Hi, I plan to calculate a minimum distance estimator using fminunc function.

[b,fun,flag]    = fminunc(@cov,b0);

However, during execution, Matlab send me an error message:

Error: Failure in initial user-supplied objective function evaluation

Following some suggestion from previous thread, I command

dbstop if caught error

And then run my code again. Matlab located the error at

Caught-error breakpoint was hit in optimget>optimgetfast at line 102. The error was:

Reference to non-existent field 'DiffMinChange'.

102     value = options.(name);

the cov function is defined as

%******************************
%  Minimum Distance Procedure
%******************************
function f=cov(b):
fmm = fcn(b); %parameterized moment
peso= diag(diag(var));
f   = (m-fmm)'*inv(peso)*(m-fmm); %m is vector of actual covariances - fmm is implied covariances
end

Answer 1

[Walter Roberson]

It appears to me that your installation somehow combines multiple versions of the fmincon support routines, as if someone had copied some of them instead of doing a complete installation.

At the very least you should

rehash toolboxcache

and if the problem continues then you will need to check your MATLAB path and possibly reinstall the Optimization Toolbox

[Lu zhang]

After I rehash the toolbox, the error message remains

Error using feval
Error: File: cov.m Line: 4 Column: 18
Unexpected MATLAB operator.

Error in fminunc (line 250)
        f = feval(funfcn{3},x,varargin{:});

Caused by:
    Failure in initial user-supplied objective function evaluation. FMINUNC cannot
    continue.

and shift+F5 gives me

Reference to non-existent field 'DiffMinChange'.

Error in optimget>optimgetfast (line 102)
    value = options.(name);

Error in optimget (line 25)
    o = optimgetfast(options,name,default);

Error in fminunc (line 201)
options.DiffMinChange = optimget(options,'DiffMinChange',defaultopt,'fast');

How to restall the optimization toolbox?

[Walter Roberson]

I just noticed that you are using dbstop if caught error. The error you are detecting is not necessarily a problem, since it is caught and potentially dealt with smoothly. Sometimes Mathworks writes "try it and see" code instead of checking more directly for errors.

Where is your cov getting m from? Is your cov a nested function grabbing from a shared variable?