Re: Core function SQP

[Juan Pablo Carbajal]

Dear Olaf,

Yes, the problem I posted isn't feasible.
There isn't a vector with \ell^2 norm equal to 10 and each component
lower than 1. This was just to show that sqp is not handling info == 6
of qp
This is fixed by adding (as minimal solution) the lines

    if info == 6
        warning("sqp: QP: The problem is infeasible.");
    end

at line 445 of sqp.m
In this way the user is forced to check whether their equalities and
inequalities are compatible. That is, the user is not forced to check
why sqp is trying to multiply a matrix with an empty vector.

On the other hand I bumped into this with a nonlinear problem which
linearization, as far as I can see, makes it not feasible. Can it be
possible? so far, I haven't found an answer in Bertsekas books.

At the moment I am trying to write the derivatives of constraints and
cost function, to see whether the discrete approximation is the issue.


Thanks

JPi


On Thu, Aug 4, 2011 at 10:09 AM, Olaf Till <address@hidden> wrote:
> On Wed, Aug 03, 2011 at 12:36:17PM +0200, Juan Pablo Carbajal wrote:
>> ...
>> Now, I am not an expert on SQP, but I guess the non-feasibility of the
>> problem is rooted (sometimes at least)in the linearization passed to
>> qp.
>>
>> Is there a way to perturb the linearized system passed to qp and
>> try again? Maybe in this way it gets unstuck, maybe not.
>>
>> I am taking a look at the book of Bertsekas to see whether the problem
>> is mentioned there.
>
> This would be nice anyway, but also remember that your posted example
> with the added 5 inequality constraints isn't necessarily feasible, so
> there might be no way for sqp to recover in this special case.
>
> Olaf
>



-- 
M. Sc. Juan Pablo Carbajal
-----
PhD Student
University of Zürich
www.ailab.ch/carbajal