Proof: Interpolation Error Theorem
We prove the interpolation error formula, establishing the quantitative relationship between approximation error and function smoothness.
Given: , interpolates at distinct nodes .
To Prove: For any , there exists with:
Proof:
If for some , both sides equal zero (by interpolation), so assume for all .
Define the error function:
Observe that has the following properties:
- for (since and the numerator vanishes)
- (by construction)
Thus has at least zeros in .
By Rolle's theorem, has at least zeros in . Applying Rolle's theorem repeatedly:
- has zeros
- has zeros
- has zero, call it
Now compute :
Since has degree , we have .
The product is a monic polynomial of degree , so:
Therefore:
Setting where :
Solving for :
This completes the proof.
Key Insights:
- The proof technique parallels Taylor's theorem with Lagrange remainder, using Rolle's theorem to locate an intermediate point
- The error depends on evaluated at an unknown point (depending on ), not at a fixed point
- The product term depends only on node locations, motivating optimal node selection
The formula immediately implies that if is a polynomial of degree , then , so (exact interpolation).
This error analysis extends to derivatives: if interpolates , then approximates with error where is the maximum node spacing, forming the basis for finite difference approximations to derivatives.