[Ipopt] changing linear solver to ma86 which is not an the option list

[Greg Horn]

The number if variables and constraints is not enough information to
determine how many iterations your problem should take to solve. What sort
of problem are you working with?

On Sun, Dec 20, 2015, 19:26  <1252685 at tongji.edu.cn> wrote:

> Thank you very much for your reply,and I don't have to try other linear
> solvers.
>
> But I still can't solve my initial problem.How can I reduce the iteration
> count?
>
> I'm solving an nlp,which has 2 equality constraints,5 inequality
> constraints and  19 variables with lower and upper bounds.It takes 358
> iterations to find the final solution. It's quite unacceptable for such a
> small question.
>
> And in the program output,I found that the overall_error,constraint_viol
> and complementary_inf were all smaller than the corresponding tolerance in
> the 17th iteration except the dual_inf which has the value of 40.00341,and
> the program took another 341 steps only to reduce the dual_inf.Is there
> anyway to reduce the iteration count except increasing the tolerance of
> dual_inf?
>
> Thanks in advance.
>
> Dongming Jin
>
> > -----原始邮件-----
> > 发件人: jonathan.hogg at stfc.ac.uk
> > 发送时间: 2015-12-20 20:53:08 (星期日)
> > 收件人: 1252685 at tongji.edu.cn, ipopt at list.coin-or.org
> > 抄送:
> > 主题: RE: [Ipopt] changing linear solver to ma86 which is not an the
> option     list
> >
> > This isn't terribly surprising - they should all give the same answer
> for solving the linear system!
> >
> > The only time changing the linear solver might help reduce iteration
> count is if there is some form of breakdown (eg singular systems), where
> different solvers might offer different scaling techniques or similar that
> will determine if they can successfully extract meaningful information from
> the system.
> >
> > Changing the linear system may, however, decrease time per iteration.
> >
> > Jonathan.
> > ________________________________________
> > From: Ipopt [ipopt-bounces at coin-or.org] on behalf of
> 1252685 at tongji.edu.cn [1252685 at tongji.edu.cn]
> > Sent: 20 December 2015 11:18
> > To: ipopt at list.coin-or.org
> > Subject: [Ipopt] changing linear solver to ma86 which is not an the
> option      list
> >
> > Hello.
> > I'm solving some nolinear problem with ipopt,and trying to change the
> linear solver to improve its efficiency.
> >
> > And I have tried mumps,ma27 and ma57 so far.But the outcome isn't really
> satisfying.The solving process with these three linear solvers have almost
> the same number of iteration steps.I wondering if I have change the linear
> solver successfully,but the program output says that "this is ipopt version
> 3.7 stable,runing with linear solver ma57",Can this guarantee that i have
> changed the linear solver from default mumps to ma57 successfully?if
> so,maybe i should change some other linear solvers like HSL_MA86? But the
> ipopt only have linear solver options like mumps,ma27,ma57.The ma86,ma77
> are not in the option list.how can I change the linear solvers to another
> one like ma86?
> >
> > Can you give me some instructions,or if there is any website or post
> discussing about it?I'm changing linear solver by dynamically loading some
> separate DLL. Working on windows,using msvs and intel fortran compiler.
> >
> > Sorry for my poor English,and thanks in advance!
> > Best regards,
> >
> > Dongming Jin
> > _______________________________________________
> > Ipopt mailing list
> > Ipopt at list.coin-or.org
> > http://list.coin-or.org/mailman/listinfo/ipopt
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20151221/b4f6287f/attachment.html>