Dear all,
I’m having a problem with a simple calculation (see attached
files). I have 3 inputs, c0, c10 and c20, which are defined by type 14. I run
the following calculation:
c-pred= teller / noemer, with teller
= (c10^2 - c0*c20) and noemer = (2*c10 - (c0+c20)).
When I output these values, I get strange differences
compared to values calculated in XL (see attached xl-file). At the first 48
timesteps everything goes fine with relative errors on the order of 10^-6, but
from the 49th step on, errors on the order of 10^-4 occur. Strangly these occur
only at the uneven timesteps. The errors occur only in teller and noemer but
are, I suppose, levelled out after division.
Can anyone help with this? I suspect it has something to do
with rounding-off errors but I can not find the exact cause. The only change I
see in the input values at step 49 is that c10 changes from below 10^3 to above
10^3.
I tried decreasing the tollerances and also using Powell
instead of Succesive, to no avail. When I change the timebase from 100h to
1000h (also for the type 14), the errors occur at different timesteps and
different rates…
Thanks in advance,
Stefan de Kool
Cauberg-Huygen Raadgevende Ingenieurs B.V.
Bezoekadres: Boterdiep 48, 3077 AW
ROTTERDAM
Postadres: Postbus 9222, 3007 AE
ROTTERDAM
Telefoon: +31 (0)10-2443638
Fax: +31 (0)10-425 44 43
http://www.chri.nl
