1. Newton Form of the Interpolating Polynomial

Definition

The same unique polynomial (see 3.1), written in nested-product form:

⚠️ The last product stops at , so there are factors for degree . (The slide writes ; that’s a typo, since it would give degree .)

Finding the coefficients

Impose one node at a time; each new node determines one new coefficient:

  • and so on: is the -th divided difference.

2. Divided Differences

Definition





Then

Property – Symmetry

Divided differences are invariant under permutation of their arguments, e.g. .

Divided-difference table

Order 0Order 1Order 2Order 3

The coefficients are the diagonal entries (in bold).

flowchart LR
    f0["f[x₀]"] --> d01["f[x₀,x₁]"]
    f1["f[x₁]"] --> d01
    f1 --> d12["f[x₁,x₂]"]
    f2["f[x₂]"] --> d12
    f2 --> d23["f[x₂,x₃]"]
    f3["f[x₃]"] --> d23
    d01 --> d012["f[x₀,x₁,x₂]"]
    d12 --> d012
    d12 --> d123["f[x₁,x₂,x₃]"]
    d23 --> d123
    d012 --> d0123["f[x₀,x₁,x₂,x₃]"]
    d123 --> d0123

Each entry is (lower-left − upper-left) / (difference of the two outermost ‘s).

Example (extra) – rocket velocity at with nodes 10, 15, 20, 22.5

Order 1Order 2Order 3
10227.04
15362.78
20517.35
22.5602.97

At : , ,

Stopping after the order-2 term gives m/s, the same values as the direct method in 3.1 (the interpolating polynomial is unique).

⚠️ Newton’s form is incremental: going from quadratic to cubic just adds one term, . Nothing already computed changes.


3. Choosing the Degree

Rule

When the order isn’t specified, keep computing columns until the divided differences become . With points, if the divided differences of order are all zero, a polynomial of degree interpolates the data exactly.

Example (extra)

Data from :

Order 1Order 2Order 3
00
111
2431
39510

Order 3 is zero → degree :

⚠️ Constant order- differences mean the data come from a degree- polynomial. Real measured data are rarely this clean, so the differences will just get small.


4. Newton vs Lagrange (summary)

SituationBetter choiceWhy
Add a new data pointNewtonAdd one row and one column to the table; the previous are kept
Same , new valuesLagrangeThe bases depend only on the , so they’re reused; Newton would need the whole table again