| Author |
 Topic  |
|
|
biotecDD
Germany
Posts |
 Posted - 05/28/2004 : 10:23:48 AM
|
Dear sirs,
I need to fit my data with the following function:
(1-erf(sqrt(1+x)))*exp(1+x)/sqrt(1+x)
According to the analysis, this should be a continuous positive decreasing function of positive argument. However, for arguments x<~30, Origin computes reasonable values, whereas for x>30 the values computed are exactly 0.
Obviously, this is related to the limited precision of computer arithmetic.
In principle it is possible to approximate this funtion for x>~10 by an asymptotic series, which will not converge well for small x (say, for x~1).
Is it possible to combine in one fitting function the analytical expression for x<10 and series expansion for x>10?
Or could you please suggest another solution for this problem? |
|
|
easwar
USA
1964 Posts |
Posted - 05/28/2004 : 10:40:21 AM
|
quote:
Is it possible to combine in one fitting function the analytical expression for x<10 and series expansion for x>10?
Or could you please suggest another solution for this problem?
Hi,
You can certainly define the fitting function to be different above and below a certain x, such as:
if ( x < 10 ) y = <expression1>;
else y = <expression2>;
This appears to be the only way since erf() converges to 1 at low x.
Easwar
OriginLab
Edited by - easwar on 05/28/2004 10:41:41 AM |
 |
|
| |
Topic |
|
|
|
Problems with computer precision?