Re: fsolve

[Jaroslav Hajek]

On Sun, May 25, 2008 at 8:57 PM, yakup murat mert <address@hidden> wrote:
> function result=Uval(Us,Ud,Sbt,Uf,Vg)
> result=(0.0256*Us.+0.0256*Sbt.*(exp(Uf).*(exp(-Us).+Us.-1)+exp(-Uf).*(exp(Us.-Ud).-Us.-exp(-Ud))).^0.5).-Vg;
> ...
>
> ...
>
> Us=fsolve(@(Us) Uval(Us,Ud,Sbt,Uf,Vg) , init);
>
> ....
>
> ...
>
> Code above works on matlab. The equation is parametric. And depends on "Ud"
> "Sbt" "Uf" "Vg"  parameters.  "Us"  is unknown (like x) "Ud" is 1*100 matrix
> and rest are constants.  In matlab  after this code "Us" becomes 1*100
> matrix and has been solved 100 times according to "Ud" .
>
> This code hasnt been accepted by octave. I checked everything but coulndt
> find equivalent...
>
> any suggestion....?
>


Currently, fsolve is implemented to use column vectors exclusively -
you can switch to that as a quick remedy. Matlab is probably more
smart here; it always shapes the
input to user function according to the shape of user's initial guess.
It even allows things like
fsolve (@fcn, ones(3,3,3)) (returns a 3x3x3-array)

The attached two changesets solve this. I noticed that
octave_value::vector_value with force_vector_conversion = true didn't
work on N-d arrays, and generally the XXX_vector_value methods were
sloppy, making unnecessary copies. Hence the first changeset.

The second changeset modifies the fsolve function to behave like
Matlab's. It can be applied without the first one, but then N-d arrays
still won't work.

make check seems to go without problems.

regards,

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

ov-xxx-value.diff
Description: Text Data

fsolve_xdims.diff
Description: Text Data