[Bonmin] another unexpected infeasible problem !
Pierre Bonami
pierre.bonami at gmail.com
Wed Sep 12 10:59:57 EDT 2007
Dear Nicolas,
You should refer to the documentation of Bonmin. This software is not
currently designed at giving global optima of non-convex problems (problems
which have a non-convex continuous relaxation) and the default algorithm is
particularly not designed at non-convex problem.
The least you should do if you want to try to solve a non-convex problem is
to set the option
bonmin.algorithm B-BB
(If I do that on your problem, bonmin finds a feasible solution).
For more options, you could look at the documentation, there are a few extra
tricks we can play to try to be lucky with finding good solution of
non-convex problems. For instance you can find it here:
http://www.coin-or.org/Bonmin/options_set/index.html#sec:opt_nonconv
I am very happy to learn that RealPlayer (isn't that name already taken by
somebody?) solves gear4.
I have a question. I see that you are using a modeling language different
than ampl. Does your code come with a modeling language for Non-linear
problems? Is it able to compute derivative? Is it free?
If it does the three it might be of interest to us sometime.
Cheers,
Pierre
On 9/12/07, Nicolas BERGER <Nicolas.Berger at univ-nantes.fr> wrote:
>
> Bonjour à nouveau,
>
>
> I am trying to solve the following problem (it is an optimisation problem
> --"gear4"-- from MINLPlib which I have changed into a simple constraint
> satisfaction problem) :
>
> ----------------------------------------------------------------
> param best_value := 1.643428;
> param eps := 1; #= max(1, 1% x best_value)
>
> var i1 integer >= 10 <= 25 ; #solu=19.0000000000,
> var i2 integer >= 0 <= 20 ; #solu=16.0000000000,
> var i3 integer >= 37 <= 50 ; #solu=43.0000000000,
> var i4 integer >= 45 <= 55 ; #solu=49.0000000000,
> var x6 >= 0 <= 100 ; #solu=,
> var x7 >= 0 <= 100 ; #solu=1.64342800000;
>
>
> subject to
> cons1:-1.0e6*i1*i2/(i3*i4)-x6+x7=-144279.32477276;
> obj1:-x6-x7-(0) >= - best_value - eps;
> obj2:-x6-x7-(0) <= - best_value + eps;
>
> option solver bonmin ;
>
> solve ;
> display i1,i2,i3,i4,x6,x7;
> ----------------------------------------------------------------
>
>
>
>
> Here is what I get from Ampl :
>
> -----------------------------------------------------------------
> WARNING: Tried to set option "expect_infeasible_problem" to a value of
> "yes",
> but the previous value is set to disallow clobbering.
> The setting will remain as: "expect_infeasible_problem no"
> WARNING: Tried to set option "mu_strategy" to a value of "adaptive",
> but the previous value is set to disallow clobbering.
> The setting will remain as: "mu_strategy monotone"
> WARNING: Tried to set option "mu_oracle" to a value of "probing",
> but the previous value is set to disallow clobbering.
> The setting will remain as: "mu_oracle quality-function"
> bonmin:
>
>
> ******************************************************************************
> This program contains Ipopt, a library for large-scale nonlinear
> optimization.
> Ipopt is released as open source code under the Common Public License
> (CPL).
> For more information visit http://projects.coin-or.org/Ipopt
>
> ******************************************************************************
>
> Error in an AMPL evaluation. Run with "halt_on_ampl_error yes" to see
> details.
> IpOp0008I
> Num Status Obj It time
> IpOp0009I 1 OPT 0 17 0.012001
> Cbc0013I At root node, 0 cuts changed objective from 0 to 0 in 1 passes
> Cbc0014I Cut generator 0 (Outer Approximation Supporting Hyperplanes for
> NLP optimum) - 0 row cuts (0 active), 0 column cuts
> Cbc0014I Cut generator 1 (GMI) - 1 row cuts (0 active), 0 column cuts
> Cbc0014I Cut generator 2 (Probing) - 0 row cuts (0 active), 1 column cuts
> Cbc0014I Cut generator 3 (covers) - 0 row cuts (0 active), 0 column cuts
> Cbc0014I Cut generator 4 (MIR) - 0 row cuts (0 active), 0 column cuts
> Cbc0014I Cut generator 5 (Outer Approximation local enumerator) - 0 row
> cuts (0 active), 0 column cuts
> Cbc0014I Cut generator 6 (Outer Approximation feasibility checker) - 0 row
> cuts (0 active), 0 column cuts
> Cbc0001I Search completed - best objective 1e+50, took 1 iterations and 0
> nodes (0.42 seconds)
>
> "Finished"
> Finished
>
> bonmin: Infeasible problem
> i1 = 0
> i2 = 0
> i3 = 0
> i4 = 0
> x6 = 0
> x7 = 0
> ----------------------------------------------------------------
>
>
>
>
> This problem is said to be infeasible, but it *DOES* have a solution when
> I use my constraint satisfaction tool (RealPaver) :
>
> -------------------------------------------------------------------
> problem pb_gear4 /*MINLPlib*/
> num
> best_value := 1.643428,
> eps := 1;
>
> var
> i1 : int ~ [10, 25] /*19.0000000000*/,
> i2 : int ~ [0, 20] /*16.0000000000*/,
> i3 : int ~ [37, 50] /*43.0000000000*/,
> i4 : int ~ [45, 55] /*49.0000000000*/,
> x6 : real ~ [0, 100] /**/,
> x7 : real ~ [0, 100] /*1.64342800000*/;
>
>
>
> ctr
> -1.0e6*i1*i2/(i3*i4)-x6+x7=-144279.32477276,
> -x6-x7-(0) >= - best_value - eps,
> -x6-x7-(0) <= - best_value + eps;
>
>
>
> solve * ;
> end ;
> -------------------------------------------------------------------
>
>
>
> Solving this CSP by RealPaver gives me the following output :
>
> -------------------------------------------------------------------
> Box 0 [11340 ms] [width 8.3528e-09] [vol 0]
> i1 = 16
> i2 = 19
> i3 = 43
> i4 = 49
> x6 ~ [0..8.352799341082573e-09]
> x7 ~ [1.643428473966196..1.643428482318996]
>
> Solving in 11340ms
> 1 solution(s)
> 379903 split(s)
> 29 box(es) created in memory
> -------------------------------------------------------------------
>
>
>
> I need an explanation... :(
>
>
> Nicolas
>
>
> _______________________________________________
> Bonmin mailing list
> Bonmin at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/bonmin
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.coin-or.org/pipermail/bonmin/attachments/20070912/26291031/attachment.html
More information about the Bonmin
mailing list