Re: Strange result with exponents...

["Andy Holmer" <aholmer@xxxxxxxxxx>]

Not wanting to be the only member that didn't post on this one so here
goes....

> These two types of problems differ in your ability to know quantities 
> with absolute exactness.  It is easy to know how many fingers I have 
> (an exact
> entity) but it is very hard to know precisely how much they weigh (an
> inexact entity).  Decimal expressions are used for problems that match
the
> first type (most iSeries software falls into this category).  Floating
> point expressions are used for problems of the second type.  Do not
use
> the wrong type of expressions for the wrong problem.

There are two metrics of a measurement to consider when solving a
problem.  Accuracy and precision.  Accuracy is how close my answer is to
the correct answer.  If I added 1 + 1 and got 1.6, 2.4, 1.7, 2.3, 2.0
and 1.9, my answer is relatively accurate--they all average to 2, but
they lack precision.  If I get 1.5002, 1.4997, 1.5001, 1.5000 my answers
are very precise, but they lack accuracy.  To get a true correct answer
I need both accuracy and precision.  The problem lies in the fact that
nobody wants to spend $1000 to get a ruler accurate to 10 decimal
places, so we sacrifice one or the other.

The place that nature and business applications really cross is in the
quality control department.  Consider the engineer that specifies a
piece of aluminum that is one inch square to hang 1000 pounds off of.
The aluminum holds 1100 pounds per square inch.  If I hang 1000 pounds
from it does it break?  No - at least not on the drawing board.

Consider reality.  My aluminum was made by a tired factory worker on a
Monday morning.  It is really only .99 x .98 (which is within my quality
control specification).  Consider that 1100 pound figure, can it really
hold 1100 pounds per square inch or just somewhere between 1050 and
1150, depending on the batch of aluminum?  Lastly, does that weight
really weigh 1000 pounds or did they think that somewhere between 980
and 1020 was good enough?  So now if I hang the weight, does it break?
It broke  1 time out of 25 --- whoops.

So, if I wrote a procedure in RPG that would do all that statistical
math and you just passed it parameters, I would have two choices,
declare all my variables really big (19,9P in RPG III) and pray for no
overflow or use a float data type.  In areas where absolute precision is
required (aka General Ledger), float data types are not a very good
answer, but within other areas they are exactly what the doctor ordered.
IMHO one of the tricks to programming is knowing which time is which.

Andy