[TRAC] cbc.exe -- Linear relaxation infeasible

[qtothec]

Moved from TRAC as part of COIN fORgery: https://projects.coin-or.org/Clp/ticket/58

Hello,

I'm using cbc.exe (built from CoinAll?-1.6.0) to solve a linear problem. The presolve is set to off.

If I use DualSimplex? Cbc doesn't find the solution and says :

Coin:dualS
Perturbing problem by 0.001 % of 15328.673 - largest nonzero change 0.014293713
(% 0.13843725) - largest zero change 0.014850191
0  Obj -1.0011806e+012 Primal inf 1.228312e+015 (120)
4  Obj 3.7294764e+019 Primal inf 1.0874013e+024 (114)
7  Obj 1.7568362e+023 Primal inf 5.0281442e+024 (113)
9  Obj 2.1954681e+028 Primal inf 5.6899824e+029 (121)
10  Obj 2.1954682e+028 Primal inf 5.6899957e+029 (124)
11  Obj 2.1954682e+028 Primal inf 5.6900909e+029 (122)
13  Obj 2.1956923e+028 Primal inf 5.6901586e+029 (123)
15  Obj 2.1956936e+028 Primal inf 5.6913518e+029 (126)
16  Obj 2.1956936e+028 Primal inf 5.6913518e+029 (130)
18  Obj 2.1956936e+028 Primal inf 5.6913518e+029 (131)
20  Obj 2.1956936e+028 Primal inf 5.6913518e+029 (133)
21  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (133)
23  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (140)
24  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (142)
25  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (143)
26  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (144)
28  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (145)
30  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (144)
32  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (144)
75  Obj 6.5876509e+028 Primal inf 1.7075533e+030 (109)
76  Obj 6.5888849e+028 Primal inf 1.7075533e+030 (112)
77  Obj 6.5888849e+028 Primal inf 1.7075533e+030 (111)
79  Obj 6.5890839e+028 Primal inf 1.7075533e+030 (111)
80  Obj 6.5890839e+028 Primal inf 1.7075533e+030 (113)
81  Obj 6.5890839e+028 Primal inf 1.7075533e+030 (114)
82  Obj 6.5891369e+028 Primal inf 1.7075533e+030 (114)
83  Obj 6.5891439e+028 Primal inf 1.7075533e+030 (122) Dual inf 0.00034051428 (1
)
84  Obj 6.5891439e+028 Primal inf 1.7075533e+030 (124)
85  Obj 6.5894741e+028 Primal inf 1.7075533e+030 (140)
97  Obj 6.5894741e+028 Primal inf 1.7075533e+030 (114)
98  Obj 2.6350363e+029 Primal inf 1.7075735e+030 (115)
98  Obj 2.6350363e+029 Primal inf 1.7075735e+030 (115)
Primal infeasible - objective value 2.0628922e+029
PrimalInfeasible objective 2.062892195e+029 - 98 iterations time 0.042

Result - Linear relaxation infeasible

Enumerated nodes:           0
Total iterations:           0
Time (CPU seconds):         5271.35
Time (Wallclock Seconds):   5271.35

But if I directly make another call to DualSimplex? Cbc finds the solution ?! :

Coin:dualS
Perturbing problem by 0.001 % of 15328.673 - largest nonzero change 0.013690577
(% 5.2125114) - largest zero change 0.014466123
0  Obj 1.2372034e+011 Primal inf 2.5945057e+009 (37) Dual inf 2110055.5 (7)
6  Obj 4.8619578e+012 Primal inf 6.528013e+008 (39)
10  Obj 6.5771001e+012 Primal inf 3.6421095e+008 (30)
15  Obj 7.1419465e+012 Primal inf 2.7443405e+008 (30)
17  Obj 7.1419509e+012 Primal inf 2.5510148e+008 (28)
32  Obj 7.1461345e+012 Primal inf 2.5386987e+008 (10)
34  Obj 7.1461345e+012 Primal inf 3.3829262e+008 (16)
35  Obj 7.1461345e+012 Primal inf 3.3819038e+008 (17)
36  Obj 7.1461345e+012 Primal inf 3.4818665e+008 (16)
36  Obj 7.1461345e+012 Primal inf 3.4818665e+008 (16)
36  Obj 7.1461345e+012 Primal inf 3.4818665e+008 (16)
54  Obj -1.2740085e+013 Primal inf 7.3715614e+012 (41)
55  Obj -1.2740085e+013 Primal inf 7.3713401e+012 (43)
57  Obj -1.0186659e+013 Primal inf 5.4809212e+012 (42)
60  Obj 5.0106785e+011 Primal inf 31121055 (15)
64  Obj 5.2745316e+011 Primal inf 88664.554 (7)
69  Obj 5.2816431e+011 Primal inf 4398607.9 (15)
76  Obj 5.2848012e+011 Primal inf 109.27733 (2)
82  Obj 5.2860406e+011
82  Obj -1.1280161e+011 Primal inf 1.1778729e+015 (25) Dual inf 7.0426814e+014
7)
94  Obj 3.8456318e+011 Primal inf 9.7273275e+008 (10) Dual inf 61020822 (1)
107  Obj 2.2089191e+011 Dual inf 0.022728333 (2)
108  Obj 2.2089191e+011
Optimal - objective value 2.2089191e+011
Optimal objective 2.208919053e+011 - 108 iterations time 0.042

Clp and Cplex both find the solution. I know it would be more convenient to use Clp but for some reason I have to go with Cbc.

Is there a way to get rid of the infeasible linear relaxation problem ? (if it's the origin of the problem) What should I do ?

Thanks in advance.