[Coin-discuss] OSI clarifications

[Erling D. Andersen]

Hi,


>The biggest grey area I see here is with isProvenPrimalInfeasible()
>and isProvenDualInfeasible().  Technically, it is possible for a
>problem to be both primal and dual infeasible, though this situation
>is often ignored.  To be fair, I think that it is more of a
>"pathological" case that rarely (if ever) occurs in real problems.
>When most people say "dual infeasible", I think they really mean
>"primal unbounded", since this is more of a concern.  CPLEX, for
>example, tells you whether the problem is primal infeasible and primal
>unbounded, but it does not easily reveal the case of primal and dual
>infeasibility.

I think the reverse is the case. Usually people say their problem is 
unbounded meaning it is dual infeasible.

Note some interior-point methods actually detect dual infeasibility 
but not unboundedness. This is the case for MOSEK (not linked to
COIN yet).

Erling

_______________________________________________
>Coin-discuss mailing list
>Coin-discuss at www-124.ibm.com
>http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-discuss

*************************************************************************
EKA Consulting ApS         
Fruebjergvej 3, Boks 16
DK-2100 Copenhagen O
Denmark

Phone (work): +45 3917 9907
Mobile-phone: +45 2362 9520
Fax:               +45 3917 9823
Email to phone: 23629520 at sms.tdm.dk  
Email: e.d.andersen at mosek.com
Homepage: http://www.mosek.com/homepages/e.d.andersen/

*************************************************************************