[Couenne] Couenne error with final solution

[Pietro Belotti]

Hi Akshay,

> I am running Couenne within AMPL since I have to solve some other models 
> first and hence not using a .nl file as input to Couenne. I do not think 
> it is writing a .sol solution file in this case. Is there another way to 
> check ?

Yes, by printing the value of the objective function and the violation of 
all constraints using the variables you defined in AMPL. After the 
"solve;" command, the variables should be set at a feasible solution if 
one was found.

> This error was reproduced on a number of instances I tried. Is this a 
> bug or is there some disconnect (w.r.t. tolerances etc.) between Cbc and 
> Couenne that I am not aware of ?

Recently Couenne started writing solutions it deems feasible regardless of 
whether Cbc think they aren't (it was implemented by François Margot). 
This is in both stable 0.4 and in trunk. The disconnect is due to Couenne 
printing Cbc's output, but what should be really looked at is Couenne's 
version of the facts, i.e., if there is a finite upper bound, then there's 
a feasible solution.

Pietro

--
Pietro Belotti
Dept. of Mathematical Sciences
Clemson University
email: pbelott at clemson.edu
phone: 864-656-6765
web:   http://myweb.clemson.edu/~pbelott

On Wed, 4 Apr 2012, Akshay Gupte wrote:

> Hi Pietro,
>
> I am running Couenne within AMPL since I have to solve some other models 
> first and hence not using a .nl file as input to Couenne. I do not think 
> it is writing a .sol solution file in this case. Is there another way to 
> check ?
>
> This error was reproduced on a number of instances I tried. Is this a 
> bug or is there some disconnect (w.r.t. tolerances etc.) between Cbc and 
> Couenne that I am not aware of ?
>
> Regards,
> Akshay
>
> --
> Akshay Gupte
> Ph.D. student, Operations Research
> School of Industrial&  Systems Engineering
> Georgia Institute of Technology
> Email: akshayg at gatech.edu
>
>
> On 4/4/2012 5:43 PM, Pietro Belotti wrote:
>> Akshay,
>> 
>> Couenne sometimes finds feasible solutions that Cbc deems infeasible, and 
>> hence keeps printing 1e50 as the best upper bound (the lines that begin 
>> with "After 100 nodes," are printed by Cbc). You may want to check if the 
>> reported upper bound of -3.562291e+04 corresponds to a feasible solution, 
>> i.e., is there a .sol file?
>> 
>> Cheers,
>> Pietro
>> 
>> -- 
>> Pietro Belotti
>> Dept. of Mathematical Sciences
>> Clemson University
>> email: pbelott at clemson.edu
>> phone: 864-656-6765
>> web:   http://myweb.clemson.edu/~pbelott
>> 
>> On Wed, 4 Apr 2012, Akshay Gupte wrote:
>> 
>>> Hi,
>>> 
>>> I am attaching a sample output below:
>>> 
>>> 
>>> Cbc0013I At root node, 1 cuts changed objective from -37772.749 to 
>>> -37772.688 in 2 passes
>>> Cbc0014I Cut generator 0 (Couenne convexifier cuts) - 1 row cuts average 
>>> 2.0 elements, 320 column cuts (321 active)
>>> Cbc0010I After 0 nodes, 0 on tree, 1e+50 best solution, best possible 
>>> -37772.688 (777.18 seconds)
>>> Cbc0010I After 100 nodes, 44 on tree, 1e+50 best solution, best possible 
>>> -37739.148 (819.40 seconds)
>>> Cbc0010I After 200 nodes, 93 on tree, 1e+50 best solution, best possible 
>>> -37724.941 (835.35 seconds)
>>> Cbc0020I Exiting on maximum time
>>> Cbc0005I Partial search - best objective 1e+50 (best possible -37724.941), 
>>> took 23515 iterations and 255 nodes (838.10 seconds)
>>> Cbc0035I Maximum depth 80, 0 variables fixed on reduced cost
>>>         "Not finished"
>>> Stats: /tmp/at2730.nl   233 [var]    0 [int]  592 [con]  823 [aux]   1595 
>>> [root]     1595 [tot] 0.008999 [sep]  838.775 [time]  838.782 [bb] 
>>> -3.772494e+04 [lower]
>>> -3.562291e+04 [upper]     255 [nodes]
>>> 
>>> couenne: Optimization interupted on limit.
>>> 
>>> In the "Stats" line, a finite upper bound is being reported. However, just 
>>> before exiting on maximum time, the "best solution" is +Inf.
>>> The reported lower bounds are consistent in the two lines.
>>> 
>>> What is going on here ?
>>> 
>>> -- 
>>> Regards,
>>> Akshay
>>> -- 
>>> Akshay Gupte
>>> Ph.D. student, Operations Research
>>> School of Industrial & Systems Engineering
>>> Georgia Institute of Technology
>>> Email: akshayg at gatech.edu
>>> 
>>> 
>