Technical item 2: SP -- was: Re: [Coin-standards] An argument for public domain modeling infrastructure

[Alan King]

Hi Leo,

I haven't seen the proposal you presented in Miami.  I looked for it on
your webpage, but I'm a little slow I guess  --- is it clear where to find
it?

In answer to your questions...

Sparse matrix inheritance: The main SMPS goal was to allow users to easily
create stochastic programs from existing MPS files.  The main design
principle was to require no changes to the original MPS file, which became
the "core" file.  The "time" and "stochastic" information is developed
assuming the existence of this core file.

Multiple Core files:  In principle there should be no difficulty in stating
that the SMPS standard should allow more than one core file.  So far no
practical solution has been offered.  (If there is one, please send it to
me.)

A,T,W matrices: As a fellow developer I do sympathize with your plea.  It
is possible to infer the relationships from the time information, though.
Notation-wise it seems that the A_{st} style dominates the literature.

I am flattered (of course!) to be called an expert.  In actual fact I'm
just a survivor who has managed to convince an employer to have patience
with me while I try to do some interesting and I hope useful projects.
Along those lines let me share one of the more successful operating
principles I learned at IBM: "keep it simple".  (This is in a sense what
you are hearing from Irv.  Why go beyond MPS if no-one is asking for it?)
With "simple" as a by-word, you can see why SMPS evolved along the lines
that it did: one core file, scenarios format, library-style interfaces.
Anything else requires so much more effort.  Every choice given the user
requires documentation, support, hassles when things change or break, etc,
etc.

Alan

Alan King
http://www.research.ibm.com/people/k/kingaj/
Mathematical Sciences Department
IBM Thomas J. Watson Research Center
(914) 945-1236, 8-862-1236


                                                                                                                                       
                      "Leonardo B.                                                                                                     
                      Lopes"                   To:       Alan King/Watson/IBM at IBMUS                                                    
                      <leo at iems.nwu.edu        cc:       coin-standards at www-124.southbury.usf.ibm.com                                  
                      >                        Subject:  Technical item 2: SP -- was: Re: [Coin-standards] An argument for public      
                                                domain modeling infrastructure                                                         
                      04/11/2002 03:42                                                                                                 
                      PM                                                                                                               
                                                                                                                                       
                                                                                                                                       



Once again I'm going to try to take the opportunity of having some really
experienced people around to throw some technical issues out there...

Dear Dr. King,

             Have you had a chance to look at the stage separation aspects
of
the proposal I presented in Miami? I've since changed them a little,
moving the T definition to the child stage, which makes much more sense. I
can explain it to you if something doesn't make sense. I feel there are
two aspects there which are fundamental in the SP context:

             1) Sparse matrix inheritance. How did the SMPS authors arrive
at
the definitions they chose? I had some trouble working with that mechanism
in my previous paper, so in my proposal I have a more sophisticated
inheritance scheme, equivalent to having possible more than one "core". As
far as the solver is concerned, of course, this should be transparent.

             2) I find it a lot easier to work with SP objects if the T, W,
and
A matrices are explicitly denoted. The A matrix in my notation is the part
of the the W martix whose rows refer only to columns in the current stage.
Yet other than my own papers, I have only found that notation in a couple
of other places. I'd like to speculate that the reason for that is that a
lot of our notation was set up by solving two-period problems, in which
case the distinction is not very important. But maybe there is something
broken about doing things that way that I haven't seen yet. Does that seem
like a reasonable idea to you?

             Basically what I am asking, not only to Dr. King, but to any
of
the other SPers who may be on this list, is to try to break the SP part of
the proposal that is on the web page, or come up with a completely
different initial design. I should warn that there is something currently
wrong with it (the stage of the T matrix), but I'll fix it if I find
someone who is willing to take a look at it and try and break it.

Thanks,
Leo.

========================================================================
Leonardo B. Lopes                                       leo at iems.nwu.edu
Ph.D. Student                                              (847)491-8470
IEMS - Northwestern University              http://www.iems.nwu.edu/~leo