Re: [ojAlgo-user] v25.1

[Anders P. <an...@op...>]

On 24 nov 2008, at 13.33, Oddvar Grønning wrote:

> Yes, this makes sense.
>
> The last few days I have been checking my system for poor condition  
> setup, but have not found any structural issues.
> My system is sort of a sequential QP, the equality constraints are  
> nonlinear, so I linearize and iterate till I find the solution.
>
> I guess some of my sub-systems are badly scaled (in lack of a  
> better term), and should be fixed before calling the QP-solver.
>
> Do you have any hints/links that can put me on the right track to  
> pre-solve the system myself?

Don't feel I know the best way to fix your problem.

The AE matrix contains some very small elements like for instance  
-3.4914813388431334E-15. This is a constraint parameter?! Having  
parameters with very different magnitude could be a problem, but in  
this case the number itself is just too small.

> Or do you have any plans for developing a pre-solver?

Yes...

/Anders


> Thanks for your help!
>
> - Oddvar
>
> -----Original Message-----
>> From: Anders Peterson [mailto:an...@op...]
> Sent: 23. november 2008 21:10
> To: Anders Peterson
> Subject: Re: [ojAlgo-user] v25.1
>
> On 21 nov 2008, at 12.40, Anders Peterson wrote:
>
>> On 21 nov 2008, at 10.07, Oddvar Grønning wrote:
>>
>>> I have a dynamic optimization problem to solve that mostly is
>>> overdetermined, so the optimization problem has a set of solutions
>>> instead of one global solution. I guess this is the reason of the
>>> numerical difficulties of my systems.
>>
>> Yes, I think the problem is that you have several (near) redundant
>> constraints.
>
> Don't think so anymore...
>
> I've long felt that the QP solvers needed a pre-solver that analysed
> and fixed the model/problem for better numerical performance. I think
> that the Matlab solver has one, and that's the difference. By
> experimenting a bit I managed to improve the quality of the solution
> to your latest model/problem:
>
> {{0.0},
> {0.0175},
> {-0.0175},
> {1.4639},
> {5.0},
> {4.8768},
> {4.4581},
> {-6.7724},
> {0.2257}}
>
> This solution is VERY close to the what you got from Matlab!
>
> Until ojAlgo actually has such a pre-solver you have to think about
> how you construct your matrices.
>
> /Anders
> ---------------------------------------------------------------------- 
> ---
> This SF.Net email is sponsored by the Moblin Your Move Developer's  
> challenge
> Build the coolest Linux based applications with Moblin SDK & win  
> great prizes
> Grand prize is a trip for two to an Open Source event anywhere in  
> the world
> http://moblin-contest.org/redirect.php?banner_id=100&url=/
> _______________________________________________
> ojAlgo-user mailing list
> ojA...@li...
> https://lists.sourceforge.net/lists/listinfo/ojalgo-user
>
>
>
> ---------------------------------------------------------------------- 
> ---
> This SF.Net email is sponsored by the Moblin Your Move Developer's  
> challenge
> Build the coolest Linux based applications with Moblin SDK & win  
> great prizes
> Grand prize is a trip for two to an Open Source event anywhere in  
> the world
> http://moblin-contest.org/redirect.php?banner_id=100&url=/
> _______________________________________________
> ojAlgo-user mailing list
> ojA...@li...
> https://lists.sourceforge.net/lists/listinfo/ojalgo-user
>
>