Hi Sebastião,

On 12/13/05, Sebastião Alves <[EMAIL PROTECTED]> wrote:
> Hi,
>
>  I'm a professor in the University of the State of Rio Grande do Norte,
> Brazil. I'm an entusiast of OpenOffice.org and i work in the operation
> research and combinatorial optimization area. I would like to know if you
> need some help to do the solver componnent. Is there someone with it? What's
> the algorithm to be implemented? Will it cover just linear programming or
> non-linear programming too?

I'm currently working on creating a solver add-on.  Although the hours
I can spend vary from week to week, I try to spend at least 3-5 hours
a week to keep the effort moving, but in some weeks I spend more,
especially when ideas keep coming in.

For linear programming, I'm currently developing the revised simplex
algorithm, which is my current focus and I pay a lot of attention to
it to make it more robust.  I also have the upper-and-lower-bounded
revised simplex, but this one needs more debugging and testing. 
Thinking long-term, I would also like to implement the interior point
method for its efficiency and scalability, but no effort has been done
for its implementation.  Luckily I have a friend who is into the
interior point method, so he may be able to contribute some, though I
don't want to give him any pressure into doing so (only he is
interested). ;-)

For non-linear programming, not much has been done so far, but the
initial focus will be to develop a robust unconstrained non-linear
algorithm.  I have written code for Quasi-Newton BFGS method, though
this still needs a lot of testing.  Once that goes somewhere, we will
move on to developing a constrained non-linear based on it.  There are
two possible algorithms: penalty and barrier methods.  We willl look
into both methods to see which one we should use.

Again, thinking long-term, I would like to develop genetic algorithm
to solve constrained non-linear, but I think we should set priority to
developing either penalty and barrier method first before coming to
this route.

What would help tremendouly is for you to take a look at my revised
simplex implementation, and test it against constrained linear models
you may have.  If my solver fails to solve it, suggest why it failed,
and how we can modify the existing algorithm to fix it.  That's also
what I'm actively doing at the moment, and I don't want to move on to
developing non-linear unless the linear algorithm is robust enough to
solve the majority of common scenarios.

What would also help is to suggest what algorithms other commercial or
non-commercial solvers that you know of use for their implementation,
especially for non-linear, any pitfalls and difficulties, and ways to
overcome them.  I'm studying as I go along, so any professional
suggestions will help me tremendously, and help the project as a
whole.

Best regards,
Kohei




--
Kohei Yoshida
http://kohei.us/

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to