Re: [Bug-glpk] Assertion at glpipm.c; line 571

[Andrew Makhorin]

> I'm using glpk-4.8 (great tool, btw!) and am running into a problem
>  using
> the 
> interior point solver.  The output from the run looks like:
> 
> lpx_interior: original LP problem has 4212 rows and 6739 columns
> lpx_interior: transformed LP problem has 702 rows and 1402 columns
> lpx_interior: A has 2662 non-zeros
> lpx_interior: S has 1682 non-zeros (upper triangle)
> lpx_interior: minimal degree ordering...
> lpx_interior: computing Cholesky factorization...
> lpx_interior: U has 1682 non-zeros
> lpx_interior: guessing initial point...
> Assertion failed: x[j] > 0.0 && z[j] > 0.0; file glpipm.c; line 571
> 
> I've tried this on three different datasets and get the same error. 
> However, 
> I have used it before successfully on other datasets, and I'm also
>  dealing 
> with some new data processing code, so it's quite possible that my 
> application could be stomping on the library.
> 
> However, I am curious to know if anyone else has seen this assertion.
>   If
> not, 
> could characteristics of the data be at fault?
> 
> The simplex solver does find the optimal solution, btw, so if I'm
>  doing any 
> stack stomping it's very (un)lucky stuff.

Thank you for the bug report.

Looks like the bug appears due to excessive round-off errors in the
solution of the normal system, because the problem is badly scaled.

Could you please gzip the problem data causing the bug in mps format
and send it to me?

Andrew Makhorin