help-glpk
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Help-glpk] Error message from glpk : Assertion failed: ap[q] != 0.0


From: Andrew Makhorin
Subject: Re: [Help-glpk] Error message from glpk : Assertion failed: ap[q] != 0.0; file glpspx1.c; line 1525
Date: Thu, 23 Jun 2005 01:00:38 +0400

> I am using the glpk C API to solve real linear programming
> problems with 
> the simplex algorithm and I am encountering a problem potentially 
> originating from the glpk library. The program crashes telling:
> Assertion failed: ap[q] != 0.0; file glpspx1.c; line 1525
> Writing the problem in .mps file (attached to the email) and
> using 
> glpsol reproduces the error   ( glpsol prob.mps --max ) .
> I haven't investigated that much the linear problem, it may be 
> infeasible, but previous infeasible problems had been correctly
> detected 
> by glpk.

I run your instance with glpsol 4.8 using various options, however,
could not catch the error. The most stable results were obtained using
options --max --nopresol --noscale:

lpx_read_mps: reading problem data from `prob.mps'...
lpx_read_mps: problem name not specified
lpx_read_mps: 1162 rows, 713 columns, 4538 non-zeros
lpx_read_mps: 3637 cards were read
lpx_adv_basis: size of triangular part = 1124
      0:   objval =   0.000000000e+00   infeas =   1.000000000e+00 (31)
    200:   objval =   0.000000000e+00   infeas =   4.201680672e-03 (20)
    237:   objval =   0.000000000e+00   infeas =   1.387626018e-13 (15)
*   237:   objval =   0.000000000e+00   infeas =   3.302549922e-11 (15)
*   250:   objval =   1.519754903e+04   infeas =   1.196406822e-09 (15)
PROBLEM HAS UNBOUNDED SOLUTION
Time used:   1.0 secs
Memory used: 1.0M (1056492 bytes)

I think the error appears due to the same reason as explained in two
messages you found on the list, i.e. because glpk simplex solver is not
robust and sometimes numerical difficulties cause its crashing. Try to
change some options (e.g. --noscale, --nopresol, --std, etc.).

Andrew Makhorin






reply via email to

[Prev in Thread] Current Thread [Next in Thread]