Re: fsolve regression

[Jaroslav Hajek]

On Wed, Feb 25, 2009 at 7:16 AM, John W. Eaton <address@hidden> wrote:
> On 24-Feb-2009, Jaroslav Hajek wrote:
>
> | Sorry for the delay. It was a trivial problem after all caused by not
> | updating iteration count after unsuccesful iteration, which caused
> | trust-region radius to be reinitialized over and over.
> | I'm not sure, however, if this is compatible with Matlab - does it
> | count unsuccesful iterations? I hope it does not matter to anyone. If
> | Matlab uses a different enough algorithm, it may not even make sense.
> |
> | With the new patch, your script gives:
> | z0 =
> |
> |    0.133000000000000
> |    0.325115654084158
> |
> | z =
> |
> |    0.134806289085259
> |    0.323387260675138
> |
> | You can get even closer by setting lower TolX and TolFun (by default
> | they're at sqrt(eps), ~eps requests maximum precision).
>
> OK, now I'm able to run a fairly large set of examples that I have
> that use fsolve.  I haven't looked at all the results in detail
> (requires comparing plots), but the ones I have looked at seem fine.
>
> Thanks,
>
> jwe
>

You're welcome. If you happen to have further issues, please let me
know. I'm very interested in making the new fsolve a more robust
solver. Maybe some of your examples could be converted to test and/or
demos?

cheers

-- 
RNDr. Jaroslav Hajek
computing expert
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz