[Coin-lpsolver] Warm Starts with Dual Algorithm

[rickob at internode.on.net]

Hi,

We have been running some relatively large problems (10's of
thousands of variables) which are repeated with a warm start
after a relatively small change to the problem.  We have
found that after some time, perhaps 40 or 50 warm starts,
the dual algorithm does not produce correct results. 
However, the primal algorithm with warm starts seems to be
fine. l suspect that at some stage the solution produced
from a previous solve was not feasible once the problem had
been modified in the next solve period. 

My question is what tests, etc, should l run before
performing a warm start? Should l check to see if the
previous solution was proven feasible or optimal, or should
the warm start be able to work from any previous solution? 

Thanks for any help,
Rick