dly wrote:
[Interesting history deleted]

> In general linear programming models the results can assume any
> values in the feasible region and are allowed to be continuous.
> Altho for some problems such as transportation assignment or network
> flow, integer solutions can be obtained automatically, the extreme
> points of a general linear programming problem is not ordinarily
> restricted to having only integer-valued components.  Moreover one
> cannot merely round the fractional values of an optimal solution and
> expect it to produce the optimal value solution.  In a mixed integer-
> continuous variable problem only some of the variables are required
> to be integers.
>

Donna:

Thanks for this information.  I guess I misunderstood (or am
misunderstanding now): you first raised the linear programming/vector
processor issue as an example as to why 111r333 contained more information
than 1r3.  I completely agree that approximating a rational number by a
floating point number may be inappropriate in some circumstances, however
if we are talking about rational numbers (which are defined as equivalence
classes), I still don't see the difference between representatives.

The rational type in J can be useful, but it is easily added to any
language.  In fact a rational number class is a common exercise in OOP
courses, especially if the language has operator overloading.

By the way, floating point numbers are rational, too: in fact 64 bit
floating point numbers represent more rational numbers than <32 bit
integers>r<32 bit integers>.  So you also have the reverse problem:
approximating a floating point number by a rational.  More generally, the
fact that finitely many rational numbers tell us anything about
uncountably many real numbers is remarkable in itself.

Best wishes,

John



----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to