# Re: exact and approximate

From: <petsa@us.ibm.com>
Date: Thu, 20 May 1999 14:46:04 -0400

Message-ID: <85256777.00671795.00@D51MTA03.pok.ibm.com>
```Marina:

You raise 3 points:
1. Yes, h must >= 0.  We will fix this.
2. For some specific values x = v.
3. Since values in C are approximations for values in M,
many values in M map to a single value in C.  Perhaps
this needs to be made more specific.

Regards, Ashok

(Embedded
image moved to Marina_Nikitina@infoimage.com
file:          05/18/99 02:22 PM
pic03455.pcx)

cc:    (bcc: Ashok Malhotra/Watson/IBM)
Subject:  exact and approximate

In 2.5.1.4 you are defining approximate datatypes as follows:

Then for every value v' in C, there is a corresponding value v in M and a real
value h such that P(x) = v' for all X in M such that |v - x| < h.

First of all, because of absolute value,  h should be >0 (or >=0).
Then x in M must  be different from v in M. So, you better say , for example,
... and a real value h >0 such that P(x) = v' for all x != v in M such that |v -
x| < h

Next:  Furthermore, for at least one value v' in C, there is more than one value
y in M such that P(y) = v'
Question: is it possible for  v'' from C and v'' !=v' that P(y) =v' and
P(y)=v''?
I am just trying to understand your concept of "approximate datatype".

Thanks

```
Received on Thursday, 20 May 1999 14:46:18 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Sunday, 6 December 2009 18:12:46 GMT