Re: In an RDBMS, what does "Data" mean?

From: Chris Hoess <choess_at_stwing.upenn.edu>
Date: Wed, 19 May 2004 06:41:35 +0000 (UTC)
Message-ID: <slrncam0cu.9q6.choess_at_force.stwing.upenn.edu>


In article <PAu0b6GWsqpAFwSz_at_thewolery.demon.co.uk>, Anthony W. Youngman wrote:
>
> It's just that I find Newtonian mechanics an excellent analogy. To
> express it in computerese, both Newtonian Mechanics and Relational
> Theory are instances of the class Mathematical_Theory. BOTH are
> mathematically perfect (well, I know Newtonian Mechanics is).

But you're missing an important point, namely, Newtownian mechanics incorporates into it distinct physical concepts such as mass, distance, and time. Relational theory does not. This is why we can't set up some experiment to test "relational theory" as such against the real world and see what happens: only by creating a specific schema which links together machine-readable definitions of relations and constraints and the semantic import of those relations can we try and test relational theory, or any other general theory of data modelling, against the real world.

To put it another way, relational theory is analogous to the equation for a Gaussian distribution, f(x) = ae^(-bx^2). Were I to assert that Gaussian distributions are useful in describing scientific phenomena, you might ask me for a test; and what are f, a, b, and x? And when I tell you that it depends on the phenomenon we are trying to describe, and that f, a, b, and x can be many different things, you might mistake it to be of no practical value, as it makes no verifiable predictions. But if I were to substitute for f C, the concentration, for a C0/sqrt(4piDt), for b 1/4Dt, and proclaimed x to be distance, I would have made use of a Gaussian distribution to describe the process of diffusion, and it could be checked experimentally and the predictions of the equation (Fick's Second Law) verified. Only by giving a physical interpretation to the variables of the Gaussian distribution does it become a scientifically verifiable theory; and only by creating a schema which we associate with semantics are we able to test the application of the relational model to our problems.

Having established that the relational model is an underlying mathematical framework bound to reality by the "glue" of the schemas we create, we're on better grounds to discuss the applicability of the model without premature calls for "experiment". We know that data in the relational model is formulated as logical propositions whose validity is evaluated by first-order logic. Hence my tenative suggestion in a post here about a month ago for examining alternatives to the relational model: are logical propositions the best way to formulate data, and do we need more power than first-order logic can bring us (and what trade-offs does that present)?

(Incidentally, can we agree that while consistency is not sufficient to prove the correctness of a data model, it is necessary?)

-- 
Chris Hoess
Received on Wed May 19 2004 - 08:41:35 CEST

Original text of this message