Notes - Numerical Analysis HT24, Lagrange interpolation


Flashcards

Give a formula for a polynomial $p _ n \in \Pi _ n$ such that

\[p_n(x_i) = f_i\]

for $n+1$ points $(x _ i, f _ i)$.


\[p_n(x) = \sum^n_{k = 0} f_k L_{n,k} (x) \in \Pi_n\]

where

\[L_{n,k}(x) = \prod^n_{i = 0, i \ne k} \frac{(x -x_i)}{(x_k - x_i)}\]

Quickly prove that the Lagrange interpolating polynomial for $n+1$ distinct points is unique.


Suppose $p _ n, q _ n$ are Lagrange polynomials of degree $n$. $p _ n - q _ n$ has $n + 1$ distinct roots, so since these polynomials are only of degree $n$, they must be identically $0$.

State the theorem that gives you a bound on the error for a Lagrange interpolating polynomial $p _ n$ approximating a function $f$ on the interval $[x _ 0, x _ n]$.


For every $x \in [x _ 0, x _ n]$, there exists $\xi _ x \in (x _ 0, x _ n)$ such that

\[e(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!} \prod^n_{k = 0} (x-x_k)\]

where $e(x) = f(x) - p _ {2n+1}(x)$.

Quickly prove that if:

  • $p _ n$ is the lagrange interpolating polynomial for points $x _ 0, \cdots, x _ n$
  • $f$ is the actual function
  • $e(x) = f(x) - p _ n(x)$

Then:

  • For every $x \in [x _ 0, x _ n]$, there exists $\xi _ x \in (x _ 0, x _ n)$ such that
\[e(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!} \prod^n_{k = 0} (x-x_k)\]

If $x = x _ k$ for any $k$, both sides are $0$ and so the equality follows immediately. So suppose $x \ne x _ k$. Define

\[\phi(t) := e(t) - \frac{e(x)}{\pi(x)} \pi(t)\]

where

\[\pi(t) := \prod_{k=0}^n (x -x_k)\]

Then note that $\phi(t)$ vanishes at $n + 2$ points: $t = x _ k$ for each $k$, and $t = x$. Then by repeated applications of Rolle’s theorem, $\phi^{(k)}(t)$ vanishes at $n + 2 - k$ points in $(x _ 0, x _ n)$ so $\phi^{(n+1)}(t)$ vanishes at an unknown point $\xi \in (x _ 0, x _ n)$.

Then

\[\begin{aligned} \phi^{(n+1)}(\xi) &= e^{(n+1)}(\xi) - \frac{e(x)}{\pi(x)} \pi^{(n+1)}(\xi) \\\\ &= f^{(n+1)} (\xi) - \frac{e(x)}{\pi(x)} (n+1)! \end{aligned}\]

So

\[e(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!} \pi(x)\]

as required.

Quickly prove that if

  • $\hat f$ is the $n$-th degree Taylor approximation at $x _ 0$
  • $f$ is the actual function
  • $e(x) = f(x) - \hat f(x)$ is the error

Then for all $x$, there exists $\xi \in (x _ 0, x)$ such that

\[e(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!} (x - x_0)^{n+1}\]

Define the function

\[\phi(t) := e(t) - \frac{e(x)}{\pi(x)}\pi(t)\]

where

\[\pi(t) = (t - x_0)^{n+1}\]

Since $\phi(x) = 0$ and for each $k$ we have $\phi^{(k)}(x _ 0) = 0$, we can repeatedly apply Rolle’s theorem to determine that $\exists \xi \in (x _ 0, x)$ such that

\[\phi^{(n+1)}(\xi) = 0\]

Then

\[e^{(n+1)}(\xi) - \frac{e(x)}{\pi(x)} \pi^{(n+1)}(\xi) = 0\]

which implies, after rearranging

\[e(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!} \pi(x)\]

Suppose $p \in \Pi _ n$ that interpolates some function for points $x _ 0 < x _ 1 < \cdots < x _ {n-1}$. Can you quickly write down every polynomial $q \in \Pi _ {n+1}$ that also interpolates these points?


\[q(x) = p(x) + \alpha (x-x_0)\cdots(x-x_{n-1})\]



Related posts