[Date Index] [Thread Index] [Author Index]

Numeric Accuracy

• To: mathgroup@smc.vnet.net
• Subject: [mg11945] Numeric Accuracy
• From: hello@there.com (Rod Pinna)
• Date: Thu, 9 Apr 1998 00:33:35 -0400
• Organization: UWA

Hello all,

I have a problem which is confusing me a bit. I have a rather  long
calculation, the results of which I am comparing to a finite  element
solution. Now, in defining the problem, I have a set of  physical
dimensions.

For the sake of discussion, they are a=10, t=0.05, L=16.1886,
W=207x10^9, n=0.3.

Leaving these as is, my analytic solution gives two solutions, 395325
and 128095. There is also a "wave number" (an integer)  associated with
these. The second of these solutions is correct  within 3%, and
predicts the same wave number as the FE solution.

However, when enter the dimensions as follows:
a=10,t=0.05`1000,L=16.1886`1000,W=207x10^9,n=0.3`1000

The solution now produces two solutions which are almost exactly  the
same, and have a value about 310551.7, which is quite a long  way from
the correct solution.

So, my question is, should I trust the machine precision  solution, on
the arbitrary precision solution?


Rod Pinna
(rpinnaX@XcivilX.uwa.edu.au  Remove the X for email)