1
$\begingroup$

I am wondering how can I calculate the error of an interpolation (Lagrange, Newton, spline) or approximation (trigonometric) for a function whose mathematical expression I don't know, but I have some points from its representation that I will use to calculate the form of the interpolation / approximation.

I was thinking of an error given by the difference between the:

->mean of the function's values (y's) in the given points that are used to calculate the others

and

->mean of the function's values (y's) in the calculated points

But this seems to be far from giving any relevant results.

$\endgroup$
1
$\begingroup$

The usual way of doing this involves assuming a function space, say $C^{k}$, and then using this to prove that the error is bounded by a constant times the spacing between sample points. Unfortunately this constant is unknown unless an explicit form of the function is known.

I would just interpolate half the points, then calculate the error from the interpolant to the actual points. If you truly know nothing about your function, that's as good as you can do. If you can assume a function class then you can probably demonstrate that a particular interpolant has $O(h^{k})$ accuracy, for some $k$. Then, assume that the spacing between samples is halved, you are safe in assuming the error is $1/2^k$ of the error between the coarse interpolant.

$\endgroup$
  • $\begingroup$ Thank you for your answer! I believe that your first suggestion is what I need. I think of using the Lagrange interpolation to find the approximation of the values and then I'll bound the error and get the maximum one to find the "real" values in those points. Unfortunately, there will probably be a lot of errors caused by the approximated (n+1) th derivatives. $\endgroup$ – Alex M. May 16 '17 at 18:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.