[Coin-lpsolver] CLP Presolve

[Matthew Galati]

> I have fixed on main branch (and on development branch where it is 
> CoinPresolve....).  An array was not long enough for this odd shaped 
> problem - 5833 rows and only 23 columns and 437 elements.


Thanks John!


> As to your questions:
> 
> 1) CoinPresolve does not produce a proof of infeasibility - it just 
> declares the problem infeasible.  Presumably it could produce a proof. 
>  I suspect that the problem could be proved to be infeasible by presolve 
> only in first few iterations.


So, if presolve determines the problem infeasible, and then I call 
resolve( ), it won't speed anything up? The problems I am looking at are 
feasibility problems with an exponential number of variables - solved 
with column generation. So, if presolve could return a proof of 
infeasibility without ever going into simplex - that would be great. Is 
that possible?



> 2)  Which codes provide multiple rays?  If others do and people think it 
> is a good idea then I can look into it.



I meant that there may exist multiple rays - there is no code I know of 
to extract them. Once upon a time, Ted Ralphs and I wrote a routine 
using CPX to do this by searching through the basis inverse. I can dig 
around to find this, clean it up to make it OsiCpx (and perhaps OsiClp) 
friendly and contribute. Hopefully, we can then make it work for other 
solvers.


Matt



-- 
Matthew Galati
ISE Lehigh University
IBM Service Parts Solutions
610.758.4042 (Office)
610.882.0779 (Home)
magh at lehigh.edu, magal11 at us.ibm.com
http://sagan.ie.lehigh.edu/mgalati/