Alan King Suggestion on Solver and Modeling Languge Interfaces

[Kipp Martin <kmartin@gsb.uchicago.edu>]

Hi:

I agree totally that the TC should address standards for interfaces 
between modeling languages and solvers.  A standard for representing a 
problem instance should be tied to a standard for solver APIs. For 
example, if a solver API wants constraints expressed with upper and 
lower bounds, it makes little sense to represent a problem instance 
constraint with a single right hand side and then a <=, >=, or = 
depending on the type of constraint. Obviously the conversion can be 
made in a software library, but it is extra work. Indeed, when my 
colleagues (Bob Fourer and Leo Lopes) and I designed our schema for 
linear programming we looked at the IBM/COIN OSI and LINDO API so that 
our data structures would be reasonably compatible. There is chicken and 
egg problem here of which comes first.  Specifying an API for solvers 
and modeling languages will have a huge impact on the best way to 
represent problem instances. The converse is also true. I think the COIN 
OSI is a good place to start for a solver API.

The data issue is an important one I would like to address in a 
different thread.

The more I think about this, the more I think the TC should just focus 
on LP and IP at first.
-- 
Kipp Martin
Professor of Operations Research
    and Computing Technology
University of Chicago
Graduate School of Business
1101 East 58 Street
Chicago IL 60637
773-702-7456