Barycentric Interpolation
and
Coefficients
Mahbuba Perveen
10/20/2020
Rational Functions
A rational function is a ratio of polynomials p(x)/q(x).
If the numerator p(x) and the denominator q(x) have no roots in common, then the
rational function is in reduced form.
(x2+1)/(x3+3x+1) is in reduced form.
(x-2)/(x2-4) is not in reduced form, because x=2 is a root of both numerator and
denominator.
(x-2)/(x2-4) = (x-2)/(x-2)(x+2) = 1/(x+2)
2
Poles
For a rational function in reduced form, the poles are the values of x where the
denominator is equal to zero.
In other words, the rational function is not defined at its poles.
Example:
The function 1/(x2+8x+7) has poles at x=-1 and x=-7
The function (x-2)/(x2-4) = 1/(x+2) has only one pole, x=-2
The function (x2+1) has no poles
3
Poles (cont.)
4
Rational Interpolation
For the interpolation problem, a rational function is constructed to go through a set
of tabulated functional values.
While constructing a global approximation on the entire table of values using all
the given nodes x0, x1, … xN-1, one potential drawback is that the approximation
can have poles inside the interpolation interval, even if the original function has no
poles there.
5
Rational Interpolation (cont.)
● We can make the degree of both the numerator and the denominator in eqn.
(3.4.1) be N-1
○ There would be no poles anywhere on the real axis
○ Allows the actual order of approximation to be specified to be any integer d<N
● This requires that the p’s and the q’s not be independent, so that eqn. 3.4.2
no longer holds
6
Barycentric Algorithm
Barycentric form of the rational interpolant:
Example:
7
N is the number of nodes, d is the desired order
Barycentric Interpolation
● By a suitable choice of the weights wi, every rational interpolant can be
written in the barycentric form.
○ As a special case, polynomial interpolants as well
● Barycentric rational interpolation competes very favorably with splines
○ It’s error is often smaller
○ The resulting approximation is infinitely smooth (unlike splines)
● If we want our rational interpolant to have approximation order d, i.e., if the
spacing of the points is O(h), the error is O(hd+1) as h -> 0
8
Runge’s example with Barycentric Interpolation
Figure: Interpolating Runge’s example with d = 3 and n = 10, 20, 40, 80. 9
Coefficients of Polynomials
10
Coefficients of the Interpolating Polynomial
● Sometimes we may need the coefficients of a polynomial, rather than the
actual value of the interpolating polynomial
○ For example, to compute simultaneous interpolated values of the function and several of its
derivatives
○ To convolve a segment of the tabulated function with some other function, where the moments
of the other function (i.e., its convolution with powers of x) are known analytically
● Generally the coefficients of the interpolating polynomial can be determined
much less accurately than its value at a desired abscissa
○ Therefore, it is not a good idea to determine the coefficients only for use in calculating
interpolating values
○ Interpolated values calculated this way will not pass exactly through the tabulated points
11
Vandermonde Matrix
Let’s take the tabulated points to be:
If the interpolating polynomial is written as:
Then the ci’s are required to satisfy the linear equation:
This is a Vandermonde matrix.
12
Coefficients (cont.)
● For high degrees of interpolation, precision of coefficients are essential
○ Interpolation error is compounded by inaccuracy of coefficients
● Vandermonde systems are notoriously ill-conditioned
○ In such cases, no numerical method gives a very accurate result
● Only practical for small datasets
○ As N increases, the Vandermonde system becomes more ill-conditioned
● It’s better to compute Vandermonde problems in double precision or higher
13
References
[1]https://ocw.mit.edu/courses/mathematics/18-03sc-differential-equations-fall-201
1/unit-iii-fourier-series-and-laplace-transform/poles-amplitude-response-connectio
n-to-erf/MIT18_03SCF11_s31_1text.pdf
[2]Press, William H., and William T. Vetterling. Numerical Recipes. Cambridge
Univ. Press, 2007.
[3]https://www.semanticscholar.org/paper/Barycentric-rational-interpolation-with-n
o-poles-of-Floater-Hormann/221ed06a9edf2f0f2c96dd062d20994d6eb07abb/figur
e/0
14
