From: Yu Li on
Hi,

I am running into problem with fsolve this time. The exit code is mostly 2 with a few exceptions, which gives an exit code of 3. I tried the trust-region-reflective algorithm and it didn’t help. I also tried the levenberg-marquardt algorithm and about half of the time the root converges but the other half got an exitflag of -3. Below is the option I used for fsolve:
options = optimset('algorithm','levenberg-marquardt','Display','iter','Jacobian','on','TolFun',...
1.0e-32,'TolX',2*eps(x(1)));
Any ideas or suggestions? Thank you very much.

From: Alan Weiss on
On 7/14/2010 1:07 PM, Yu Li wrote:
> Hi,
>
> I am running into problem with fsolve this time. The exit code is mostly
> 2 with a few exceptions, which gives an exit code of 3. I tried the
> trust-region-reflective algorithm and it didn’t help. I also tried
> the levenberg-marquardt algorithm and about half of the time the root
> converges but the other half got an exitflag of -3. Below is the option
> I used for fsolve: options =
> optimset('algorithm','levenberg-marquardt','Display','iter','Jacobian','on','TolFun',...
>
> 1.0e-32,'TolX',2*eps(x(1)));
> Any ideas or suggestions? Thank you very much.

You should never set TolFun to be less than eps.
You probably shouldn't set it to be less than 1e-14.
Same for TolX.

For more suggestions, see
http://www.mathworks.com/access/helpdesk/help/toolbox/optim/ug/br44iv5.html

Alan Weiss
MATLAB mathematical toolbox documentation
From: Yu Li on
Hi,

Thank you very much for your suggestion. I have to set the tolerance very small because there are small numbers involved, something in the order of 1e-13. I will read the link you gave me. Thanks again.

Best
Yu

Alan Weiss <aweiss(a)mathworks.com> wrote in message <i1kvmu$5o2$1(a)fred.mathworks.com>...
> On 7/14/2010 1:07 PM, Yu Li wrote:
> > Hi,
> >

>
> You should never set TolFun to be less than eps.
> You probably shouldn't set it to be less than 1e-14.
> Same for TolX.
>
> For more suggestions, see
> http://www.mathworks.com/access/helpdesk/help/toolbox/optim/ug/br44iv5.html
>
> Alan Weiss
> MATLAB mathematical toolbox documentation