[om-list] Re: Unified Modeling of Arbitrary Complexity

[Mark Butler]

Tom wrote:

>     Isn't there a problem with the fact that if you give objects
> point-position values, position will become ambiguous: two objects can have
> the same centroid.  (I anticipate "yes" in some applications, and "no" in
> others.)

Strictly speaking, if two objects have the same attributes in all real
dimensions, they are either indistinguishable or are actually the *same*
object. Position is just one of many dimensions that distinguish an object.

For macroscopic objects, the standard way to store position distribution
information is to break it down into its subcomponents and store the positions
of each of those.

>     Personally, I think we should find a fully general way of modelling any
> arbitrary level of complexity, but one that is simple enough that you don't
> have too much wasted overhead if you only model simplicity.  (I call it
> mathesis, even before I know what mathesis looks like completely.)

The difference is that Mathesis is a theory and we are currently trying to
build a prototype of a physically realizable system. Everything can be
accomodated in theory.

>     If that makes it run to slowly, we'll have the option to increase the
> speed and the storage.  I think OM should be about options: the user has the
> option to use a powerful feature if he doesn't mind additional costs
> elsewhere.

I agree in principle, but in order to accomodate that desire in version one,
the first thing we would have to do is give up on using any existing database
technology and adopt a language like LISP.

The big difference is that our current proposed model is quite capable of
modeling the majority of what people actually know - i.e. what they can
actually fit inside their heads and what they can actually use to reason
with.  ITs major strength is scalabity - we can easily store information about
millions of objects.

The type of information you are talking about is very useful for a wide class
of scientific and engineering problems, but is so dense that you are lucky to
store information about a few thousand objects.

The impedance mismatch between these two sets of requirements means that using
current technology you need to split into two different versions that are
optimized appropriately:

1. DBMS based, disk oriented, very large numbers of objects, simple attributes
2. File based, memory oriented, smaller number of objects, 
   very complex attributes.

The only way to support both well is to develop a superior database
technology, which is roughly the same complexity as developing a new operating
system.
 
Short of that, you can certainly implement the same model using technology
appropriate for your application.

- Mark