Re: [math-optimize-discuss] Scope

[Alan King <kingaj@us.ibm.com>]

Hi Leo,

I think Kipp had it right before, so I am just expanding on what he wrote.
The issue is that for some purposes solvers would like to be able to access
more than just the model coefficients.

Concretely, let us suppose for discussion a portfolio optimization model
with a representation of the returns R in terms of underlying factors Y:

Model {
sets                { I;J;}
data               { F[I,J]; D[I]; R[I]; Y[J]; LAMBDA; TARGET;}
variables     { x[I];z;RISK;RETURN; }
equations    {R := sum[J] F*Y + D;        // factor model
                         RETURN := sum[I] R*x;
                         RISK := max (TARGET- RETURN,0);
                         z:= RETURN - LAMBDA*RISK; }
}

There are many methods that could be employed to propose a solution for
this problem: mean-lower partial mean, mean-semivariance, stochastic
recourse programming, robust optimization, etc, etc.  Each one has
requirements for access to the information in the data, sets, and equation
stanzas.  Obviously, the most important is to be able to access the factor
model.

My point is that a standard that supports the exposure of the relationship
between the information in the model stanzas and the final model
coefficients will help promote a faster pace of development and deployment
of novel solution methods --- of course in stochastic programming, but also
possibly in other fields like nonlinear.

Alan


Alan King
Mathematical Sciences, IBM Research
http://www.research.ibm.com/
http://www.research.ibm.com/people/k/kingaj/

Campus Relationship Manager: University of Washington
http://www.washington.edu
http://www.ibm.com/university


                                                                           
             Leo Lopes                                                     
             <leo@sie.arizona.                                             
             edu>                                                       To 
                                       math-optimize-discuss@lists.oasis-o 
             08/10/2004 01:15          pen.org                             
             AM                                                         cc 
                                                                           
                                                                   Subject 
             Please respond to         [math-optimize-discuss] Scope       
                    leo                                                    
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           




Dear All,

This message is in response to Kipp's and Alan's discussion. Personally,
I'd be really happy if we got the persistent representation problem
taken care of. Just to make sure: Alan, could you expand a little on
your description of "nterfaces between modeling languages, solvers, and
data sources"? Did you mean something like OSI, something like FML, or
something else?

Thanks,
Leo.

--
========================================================================
Leonardo B. Lopes                                    leo@sie.arizona.edu
Assistant Professor                                        (520)626-1780
SIE - University of Arizona  http://www.sie.arizona.edu/faculty/leolopes