1. Definition of Interpolation
Motivation
Often a function isn’t known explicitly, only its values at a set of points (a table). Interpolation estimates the function between tabulated values.
Definition – Interpolant
Given data , , find a (polynomial) function such that
- is the interpolant (interpolation function) of the points
- the are the nodes
- goal: estimate at an that isn’t in the table
⚠️ Interpolation passes exactly through every data point. Regression (Chapter 4) doesn’t; it only fits the trend.
2. Existence and Uniqueness
Theorem
With data points and an interpolating polynomial of degree (which has coefficients):
| Case | Result |
|---|---|
| interpolation does not exist (in general: too many conditions) | |
| exists but is not unique (too few conditions) | |
| exists and is unique |
Statement: for distinct nodes there is a unique polynomial of degree with , .
Example
2 points → unique line (degree 1); 3 points → unique parabola (degree ≤ 2); 4 points → unique cubic (degree ≤ 3).
3. Linear Interpolation
Formula
Line through and :
(You can equally anchor at .) Error: .
Example (slides) – rocket velocity at s
| (s) | 0 | 10 | 15 | 20 | 22.5 | 30 |
|---|---|---|---|---|---|---|
| (m/s) | 0 | 227.04 | 362.78 | 517.35 | 602.97 | 901.67 |
The two nodes bracketing are and :
⚠️ Choose the nodes closest to the target and bracketing it (). Using far-away nodes such as and is still interpolation, but less accurate.
4. Polynomial Interpolation – Direct Method
Method
Polynomials are the most common interpolants because they’re easy to evaluate, differentiate and integrate. For points, write
and impose . This gives an linear system (solve with Chapter 1’s methods):
Example (slides) – by linear, quadratic and cubic interpolation
Linear (nodes 15, 20):
Quadratic (nodes 10, 15, 20):
Cubic (nodes 10, 15, 20, 22.5):
| Order | Nodes used | (m/s) | Change from previous order |
|---|---|---|---|
| 1 | 15, 20 | 393.70 | – |
| 2 | 10, 15, 20 | 392.19 | 1.51 |
| 3 | 10, 15, 20, 22.5 | 392.06 | 0.13 |
⚠️ The values settle as the order increases. The change between successive orders is a practical estimate of the interpolation error.
⚠️ The direct (Vandermonde) system becomes ill-conditioned for high degree or large values (columns like quickly dominate), so round carefully. Newton and Lagrange (3.2, 3.3) avoid solving this system.