Altering Default MaxFunEvals

Hi, I'm having trouble altering the default number of evaluations (300) of an fsolve. My .m file looks like this (function omitted):

function solveeqs()

guess=[1.6 0 300];

options = optimset('MaxFunEvals',1e10);

[result, fval, exit, output]=fsolve(@eqns, guess);

result

fval

eqns(guess)

exit

output

end

I thought that the options command might fix it, but the output defines the evaluation limit at 300 anyways. Any tips?

Answer 1

[Sean de Wolski]

You're not passing the options structure to fsolve()

options = optimset('MaxFunEvals',1e10);
[result, fval, exit, output]=fsolve(@eqns, guess,options);

Just make sure options is defined before you call fsolve.

[Alex]

I've defined it as above, and it accepts it, but my function evaluation limit still isn't increasing. It now says 400 evaluations instead of 300, but still defines 400 as the default. I'm at a different computer now though, so maybe that makes sense.

[Sean de Wolski]

The default maxfunevals is 100 times the number of variables. Can you post _exactly_ what you are running?

I would worry a little that there is something more seriously wrong if it has not converged after that nmany iterations.

[Alex]

The equations it's working on are very complex non-linear differential equations, but I'll post the code below, line for line, with the output:

function solveeqs()

guess=[1.6 0 300];

options = optimset('MaxFunEvals',1e10);

[result, fval, exit, output]=fsolve(@eqns, guess, options);

result

fval

eqns(guess)

exit

output

end

function fcns=eqns(z)

n=z(1);

k=z(2);

d=z(3);

fcns(1)=((((1-(n-1i.*k))./(1+(n-1i.*k)))+(((n-1i.*k)-(1.694-1i.*3.666))./((n-1i.*k)+(1.694-1i.*3.666))).*(exp(1).^(-2.*1i.*(2.*pi./250).*(n-1i.*k).*d)))./(1+((1-(n-1i.*k))./(1+(n-1i.*k))).*(((n-1i.*k)-(1.694-1i.*3.666))./((n-1i.*k)+(1.694-1i.*3.666))).*(exp(1).^(-2.*1i.*(2.*pi./250).*(n-1i.*k).*d))))-.922269;

fcns(2)=((((1-(n-1i.*k))./(1+(n-1i.*k)))+(((n-1i.*k)-(1.7046-1i.*3.7066))./((n-1i.*k)+(1.7046-1i.*3.7066))).*(exp(1).^(-2.*1i.*(2.*pi./251).*(n-1i.*k).*d)))./(1+((1-(n-1i.*k))./(1+(n-1i.*k))).*(((n-1i.*k)-(1.7046-1i.*3.7066))./((n-1i.*k)+(1.7046-1i.*3.7066))).*(exp(1).^(-2.*1i.*(2.*pi./251).*(n-1i.*k).*d))))-.92401;

fcns(3)=((((1-(n-1i.*k))./(1+(n-1i.*k)))+(((n-1i.*k)-(1.7152-1i.*3.7472))./((n-1i.*k)+(1.7152-1i.*3.7472))).*(exp(1).^(-2.*1i.*(2.*pi./252).*(n-1i.*k).*d)))./(1+((1-(n-1i.*k))./(1+(n-1i.*k))).*(((n-1i.*k)-(1.7152-1i.*3.7472))./((n-1i.*k)+(1.7152-1i.*3.7472))).*(exp(1).^(-2.*1i.*(2.*pi./252).*(n-1i.*k).*d))))-.925751;

end

Outputs:

run eqnsr

Solver stopped prematurely.

fsolve stopped because it exceeded the iteration limit,

options.MaxIter = 400 (the default value).

result =

1.0e+002 *

1.3445 + 5.5767i 5.5777 - 1.3435i 3.7721 + 3.5028i

fval =

1.0e-003 *

0.0734 - 0.9412i -0.0454 - 0.0117i -0.1829 + 0.9172i

ans =

-1.7380 - 0.1528i -1.7283 - 0.2051i -1.7148 - 0.2567i

exit =

0

output =

iterations: 400

funcCount: 1580

algorithm: 'trust-region dogleg'

firstorderopt: 5.9222e-004

message: [1x127 char]

Since this is a test of it's functionality (I know what values it should be returning (approximately), I'm concerned that it just isn't getting close enough. A result of .0011 isn't as close to 0 as I would like.