Section 5.

7 Hermite Interpolation

Key terms

• Hermite data
• Hermite interpolation
• Method of undetermined coefficients
• Modify the Lagrange basis
• Modify the divided difference basis
• Error in Hermite interpolation
• Hermite Cubic interpolation
• Error in Hermite Cubic interpolation
Hermite Interpolation

Hermite interpolation is an extension of basic polynomial interpolation that not only

matches discrete information at a set of points, but also matches the slope (or rate
of change) at those points.

Hermite Interpolation Theorem

Let S be a discrete data set of distinct points defined as follows:

Then there exists a unique polynomial of degree 2n + 1 or less that

interpolates the data by matching the position of each point and the slope at each
at each point; that is,

is called the Hermite polynomial interpolant to the data set S.

Construction of the Hermite Interpolating Polynomial

There are three ways to construct the Hermite interpolant that are similar to the
procedures for basic polynomial interpolation.

1. Method of undetermined coefficients; requires the solution of a linear system.

2. A modification of the Lagrange basis; the required modification is cumbersome

and not very easy to use computationally.

3. A modification to the Divided Difference basis; this procedure is quite

reasonable when a "modified" divided difference table construction is included
with the process.
Example: Use the method of undetermined coefficients to construct the Hermite
interpolant to set

Here we have data point (1,2) where the slope is to be m = 2, point (3,1) where the
slope is to be m = -1, and point (4,2) where the slope is to be m = 0.

For set S the Hermite interpolant is to be of degree 5, so let

Now use the interpolation requirements listed next to construct a linear system
of 6 equations in the 6 unknown coefficients a5, a4, …, a0.
We obtain the following linear system to solve for the coefficients a5, a4, …, a0. Here
we show the augmented matrix of the system.

The modified Lagrange interpolation scheme is just too algebraic cumbersome for
our purposes.
The divided difference basis for standard polynomial interpolation is

The modification of this basis for

Hermite interpolation is shown
next.
Note the repeats of the linear
factors and that each member is
of a different degree.
Also note that in the expression
for the last data point's x-
coordinate only appears to the
first power.
Hermite interpolants are of odd
degree since there are an even
number, 2n, requirements to be
fulfilled.
We need only determine the corresponding coefficient of these 6 expressions.
To get the coefficients we modify the standard divided difference table construct
process. We use the slopes at the data points to fill-in some entries in the 1st
DD column. The modified table for the case of 3 Hermite data points is
displayed below. Note that we repeat the (x, y) data in the x and 0th DD
columns respectively.
 We use the slopes at the data points to
fill-in some entries in the 1st DD
column.

Note that we repeat the (x, y) data

in the x and 0th DD columns
respectively.

The coefficients for the Hermite divided

difference basis are the entries of the table
from the top diagonal just as in the
standard polynomial interpolation case
when the divided difference basis is used.
We illustrate this in the next example.
The corresponding divided difference table is

-1/2
The higher order
divided differences
are computed in the
usual manner.

The Hermite interpolant is

The Error In Hermite Interpolation
Example: Let f(x) = sin(x) . Construct the Hermite data set for x = 0, π/2 , the
Hermite interpolant to the data, and find a least upper bound on the error.

The Hermite interpolant is

 x 2*(x-pi/2)2
0.4

0.35

0.3

0.25

Comment: It is possible to modify the divided 0.2

difference procedure to handle mixed data; that 0.15

is, a data set, which may contain points where 0.1

0.05
some have derivative information and others do
0

not. 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

PIECEWISE HERMITE CUBIC INTERPOLATION

Computing the piecewise Hermite cubic interpolant

The computation procedure is designed to obtain the Hermite cubic interpolant over
successive subintervals [x0, x1], [x1, x2],…,[xn-1, xn] by solving for the individual
Hermite interpolant from left to right across the data set. We show that this requires
the solution of a 2 by 2 linear system for each of the n subintervals.

We develop the appropriate model for a generic subinterval, obtain a formula for the
exact solution, and then construct a program using the exact solution, thus avoiding
the repeated solution of the linear system.
Using the values determined for d and c, we need only solve the two equations
shown next for a and b.

In matrix form the system is

In the general case, the computation proceeds from interval to interval by computing

for i = 0, 1,…, n-1.

The Hermite interpolant over [xi, xi+1] is then

Note that the data need not be equispaced since h is recomputed for each
subinterval, but recall that the data must be ordered and distinct.
 Piecewise Hermite Cubic Error

Let’s compare Hermite Cubic interpolation to Piecewise Cubic interpolation. The

Hermite cubic interpolant is both continuous and differentiable, while the piecewise
cubic interpolant is only guaranteed to be continuous.

The error bounds are similar with the Hermite cubic interpolant being smaller.