Originally posted by: DavidGravot

Hi

I was trying to help convergence on large LP file that sometimes ends up in  apparent infeasibility, but i suspect more numerical illness at some point since switching the algorithm sometimes leads to a solution, or launching in another computer as well 

I set feasibilty tolerance (CPX_PARAM_EPRHS) . Increasing this tolerance from 1e-6 (default) to 1e-2 helped

Checking the documentation, it says 'degree to which values of the basic variables calculated by the simplex method may violate their bounds'

is there any other parameter that affects feasibility of a constraint  ? Such as 3x+2y == 4 would be considered true even when right hand side and left hand side only differ by some tolerance ?

Thanks

David

Hi,

have you had a look at CPX_PARAM_EPINT

?

regards

Originally posted by: DavidGravot

Sorry, I forgot to mention my problem is purely continuous 

Actually, EpRhs does what you want. A small violation in an equality constraint is basically a small violation in the bounds of a slack/surplus/artificial variable.

You might want look at the numerical emphasis switch.