[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [Help-glpk] having problem with solver (numerical instability ?)
From: |
Sayan Mitra |
Subject: |
RE: [Help-glpk] having problem with solver (numerical instability ?) |
Date: |
Tue, 2 Mar 2004 11:04:49 -0500 (EST) |
Thanks for your response!
However, it is not clear to me how I can scale this problem. The large
number M arises from the "Big M" method for linearizing products of
binary and real variables. As I have mentioned before, if I reduce M,
then the solver complaints with :
lpx_simplex: numerical instability (primal simplex, phase II)
Best regards,
sayan
On Tue, 2 Mar 2004, Robert Horvath wrote:
>
>
> For me it seems like your problem is badly scaled (meaning the coefficient
> matrix A has elements very small in absolute value and very large as well).
> 1. Try to scale your problem first.
> 2. If 1. did not work, try to decrease the number M to as small a value as
> possible for your problem.
>
>
> Regards
> Robert
>
>
>
> -----Original Message-----
> From: address@hidden
> [mailto:address@hidden On Behalf Of
> Sayan Mitra
> Sent: Tuesday, March 02, 2004 1:55 AM
> To: address@hidden
> Subject: [Help-glpk] having problem with solver (numerical instability ?)
>
> Hello,
>
> I am having the following difficulty with the solver; can anyone offer
> fixes or a way to get around this? In the following constraints, aa[i]
> is
> an array of booleans, y[i] is real positive, e is a small number (set to
> 0), and M is a large number (set to 10.E+8),
>
> s.t. effect_1{i in J}: y[i] >= aa[i] * e ;
>
> s.t. effect_2{i in J}: y[i] <= aa[i] * M ;
>
> therefore, when aa[i] is 0, then y[i] should be 0, but in the solution I
> am getting positive values. I am given to believe that this is an issue
> with the numerical instability of the solver, because if I reduce the
> value of M, I get:
>
> lpx_simplex: numerical instability (primal simplex, phase II)
>
> Will look forward to your responses,
>
> Thanks,
>
> sayan
>
>
>
> _______________________________________________
> Help-glpk mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/help-glpk
>
>
>