ProofComplete

Proof: Interpolation Error Theorem

We prove the interpolation error formula, establishing the quantitative relationship between approximation error and function smoothness.

ProofPolynomial Interpolation Error

Given: fCn+1[a,b]f \in C^{n+1}[a,b], pnp_n interpolates ff at distinct nodes x0,,xnx_0, \ldots, x_n.

To Prove: For any x[a,b]x \in [a,b], there exists ξx(a,b)\xi_x \in (a,b) with: f(x)pn(x)=f(n+1)(ξx)(n+1)!i=0n(xxi)f(x) - p_n(x) = \frac{f^{(n+1)}(\xi_x)}{(n+1)!} \prod_{i=0}^n (x - x_i)

Proof:

If x=xix = x_i for some ii, both sides equal zero (by interpolation), so assume xxix \neq x_i for all ii.

Define the error function: ϕ(t)=f(t)pn(t)[f(x)pn(x)]i=0n(txi)i=0n(xxi)\phi(t) = f(t) - p_n(t) - [f(x) - p_n(x)] \frac{\prod_{i=0}^n (t - x_i)}{\prod_{i=0}^n (x - x_i)}

Observe that ϕ\phi has the following properties:

  1. ϕ(xi)=0\phi(x_i) = 0 for i=0,,ni = 0, \ldots, n (since f(xi)=pn(xi)f(x_i) = p_n(x_i) and the numerator vanishes)
  2. ϕ(x)=0\phi(x) = 0 (by construction)

Thus ϕ\phi has at least n+2n+2 zeros in [a,b][a,b].

By Rolle's theorem, ϕ\phi' has at least n+1n+1 zeros in (a,b)(a,b). Applying Rolle's theorem repeatedly:

  • ϕ\phi' has n+1\geq n+1 zeros
  • ϕ\phi'' has n\geq n zeros
  • \vdots
  • ϕ(n+1)\phi^{(n+1)} has 1\geq 1 zero, call it ξx(a,b)\xi_x \in (a,b)

Now compute ϕ(n+1)(t)\phi^{(n+1)}(t): ϕ(n+1)(t)=f(n+1)(t)pn(n+1)(t)[f(x)pn(x)]dn+1dtn+1[i=0n(txi)i=0n(xxi)]\phi^{(n+1)}(t) = f^{(n+1)}(t) - p_n^{(n+1)}(t) - [f(x) - p_n(x)] \frac{d^{n+1}}{dt^{n+1}}\left[\frac{\prod_{i=0}^n (t - x_i)}{\prod_{i=0}^n (x - x_i)}\right]

Since pnp_n has degree n\leq n, we have pn(n+1)(t)=0p_n^{(n+1)}(t) = 0.

The product i=0n(txi)\prod_{i=0}^n (t - x_i) is a monic polynomial of degree n+1n+1, so: dn+1dtn+1[i=0n(txi)]=(n+1)!\frac{d^{n+1}}{dt^{n+1}}\left[\prod_{i=0}^n (t - x_i)\right] = (n+1)!

Therefore: ϕ(n+1)(t)=f(n+1)(t)[f(x)pn(x)](n+1)!i=0n(xxi)\phi^{(n+1)}(t) = f^{(n+1)}(t) - [f(x) - p_n(x)] \frac{(n+1)!}{\prod_{i=0}^n (x - x_i)}

Setting t=ξxt = \xi_x where ϕ(n+1)(ξx)=0\phi^{(n+1)}(\xi_x) = 0: 0=f(n+1)(ξx)[f(x)pn(x)](n+1)!i=0n(xxi)0 = f^{(n+1)}(\xi_x) - [f(x) - p_n(x)] \frac{(n+1)!}{\prod_{i=0}^n (x - x_i)}

Solving for f(x)pn(x)f(x) - p_n(x): f(x)pn(x)=f(n+1)(ξx)(n+1)!i=0n(xxi)f(x) - p_n(x) = \frac{f^{(n+1)}(\xi_x)}{(n+1)!} \prod_{i=0}^n (x - x_i)

This completes the proof.

Remark

Key Insights:

  1. The proof technique parallels Taylor's theorem with Lagrange remainder, using Rolle's theorem to locate an intermediate point
  2. The error depends on f(n+1)f^{(n+1)} evaluated at an unknown point ξx\xi_x (depending on xx), not at a fixed point
  3. The product term i=0n(xxi)\prod_{i=0}^n (x - x_i) depends only on node locations, motivating optimal node selection

The formula immediately implies that if ff is a polynomial of degree n\leq n, then f(n+1)0f^{(n+1)} \equiv 0, so f=pnf = p_n (exact interpolation).

This error analysis extends to derivatives: if pnp_n interpolates ff, then pnp_n' approximates ff' with error O(hn)O(h^n) where hh is the maximum node spacing, forming the basis for finite difference approximations to derivatives.