Section 6.9

The Lagrange Interpolation Formula

[10/100%]
Show
LaTeX
Problem 6.67

Let the polynomial of nnth degree

f(x)=a0xn+a1xn−1+⋯+an−1x+an f(x) = a_0 x^n + a_1 x^{n-1} + \cdots + a_{n-1}x + a_n

have only distinct zeros x1,x2,…,xnx_1, x_2, \ldots , x_n. Then we have

σk=∑ν=1nxνkf′(xν)={0,if 0≦k≦n−2,a0−1,if k=n−1. \sigma _k = \sum _{\nu =1}^{n} \frac{x_\nu ^k}{f'(x_\nu )} = \begin{cases} 0, & \text{if } 0 \leqq k \leqq n-2, \\ a_0^{-1}, & \text{if } k = n-1. \end{cases}

Furthermore σk\sigma_k is independent of ana_n for k=n,n+1,…,2n−2k = n, n+1, \ldots , 2n-2 and is linear in ana_n with coefficient −a0−2-a_0^{-2} for k=2n−1k = 2n-1.

?
Problem 6.68

(Continuation of problem 67.) We have

∑ν=1nkxνk−1f′(xν)−xνkf′′(xν)[f′(xν)]3={0,if 0≦k≦2n−2,a0−2,if k=2n−1. \sum _{\nu =1}^{n} \frac{kx_\nu ^{k-1}f'(x_\nu ) - x_\nu ^k f''(x_\nu )}{[f'(x_\nu )]^3} = \begin{cases} 0, & \text{if } 0 \leqq k \leqq 2n-2, \\ a_0^{-2}, & \text{if } k = 2n-1. \end{cases}
?
Problem 6.69

(Continuation of problem 67.) Let x1,x2,…,xnx_1, x_2, \ldots , x_n be different from 00 and −1-1. Prove that

∑ν=1nxνnf(xν−1)f′(xν)(1+xν)=(−1)n−1(1−x1x2⋯xn). \sum _{\nu =1}^{n} \frac{x_\nu ^n f(x_\nu ^{-1})}{f'(x_\nu )(1+x_\nu )} = (-1)^{n-1}(1 - x_1 x_2 \cdots x_n).
?
Problem 6.70

Let x0,x1,x2,…,xnx_0, x_1, x_2, \ldots , x_n be arbitrary integers, x0<x1<x2<⋯<xnx_0 < x_1 < x_2 < \cdots < x_n. Every polynomial of nnth degree of the form

xn+a1xn−1+a2xn−2+⋯+an x^n + a_1 x^{n-1} + a_2 x^{n-2} + \cdots + a_n

assumes at the points x0,x1,x2,…,xnx_0, x_1, x_2, \ldots , x_n values, at least one of which is of absolute value ≧n!/2n\geqq n!/2^n.

?
Problem 6.71

If one chooses the zeros of Tn(x)T_n(x) (problem 1), i.e.the numbers xν=cos⁡(2ν−1)π/(2n)x_\nu = \cos (2\nu -1)\pi /(2n), ν=1,2,…,n\nu = 1, 2, \ldots , n, as interpolation points, then the basis polynomials are

fν(x)=(−1)ν−11−xν2nTn(x)x−xν,ν=1,2,…,n. f_\nu (x) = \frac{(-1)^{\nu -1}\sqrt{1-x_\nu ^2}}{n} \frac{T_n(x)}{x-x_\nu }, \qquad \nu = 1, 2, \ldots , n.

Hence every polynomial P(x)P(x) of degree n−1n-1 may be represented as follows:

P(x)=1n∑ν=1n(−1)ν−11−xν2 P(xν)Tn(x)x−xν. P(x) = \frac{1}{n}\sum _{\nu =1}^{n} (-1)^{\nu -1}\sqrt{1-x_\nu ^2}\, P(x_\nu ) \frac{T_n(x)}{x-x_\nu }.

The square roots are to be taken everywhere as positive.

?
Problem 6.72

If one chooses the zeros of Un(x)U_n(x) (problem 1), that is the numbers xν=cos⁡νπ/(n+1)x_\nu = \cos \nu \pi /(n+1), ν=1,2,…,n\nu =1,2,\ldots ,n, as interpolation points, then the basis polynomials are

fν(x)=(−1)ν−11−xν2n+1Un(x)x−xν,ν=1,2,…,n. f_\nu (x) = (-1)^{\nu -1} \frac{1-x_\nu ^2}{n+1} \frac{U_n(x)}{x-x_\nu }, \qquad \nu = 1, 2, \ldots , n.

Hence every polynomial of degree n−1n-1 may be represented as follows:

P(x)=1n+1∑ν=1n(−1)ν−1(1−xν2)P(xν)Un(x)x−xν. P(x) = \frac{1}{n+1}\sum _{\nu =1}^{n} (-1)^{\nu -1}(1-x_\nu ^2)P(x_\nu ) \frac{U_n(x)}{x-x_\nu }.
?
Problem 6.73

If one chooses the zeros of Un−1(x)(x2−1)U_{n-1}(x)(x^2-1), that is the numbers xν=cos⁡νπ/nx_\nu = \cos \nu \pi /n, ν=1,2,…,n−1\nu =1,2,\ldots ,n-1, and in addition x0=1x_0=1, xn=−1x_n=-1 as interpolation points, then the basis polynomials f0(x),f1(x),f2(x),…,fn(x)f_0(x), f_1(x), f_2(x), \ldots , f_n(x) are the following:

fν(x)=(−1)νnUn−1(x)(x2−1)x−xν,ν=1,2,…,n−1, f_\nu (x) = \frac{(-1)^\nu }{n} \frac{U_{n-1}(x)(x^2-1)}{x-x_\nu }, \qquad \nu = 1, 2, \ldots , n-1, f0(x)=12nUn−1(x)(x+1),fn(x)=(−1)n2nUn−1(x)(x−1). f_0(x) = \frac{1}{2n} U_{n-1}(x)(x+1), \qquad f_n(x) = \frac{(-1)^n}{2n} U_{n-1}(x)(x-1).

Hence every polynomial P(x)P(x) of degree nn may be represented in the following form:

P(x)=12nUn−1(x)[P(1)(x+1)+(−1)nP(−1)(x−1)]+1n∑ν=1n−1(−1)νP(xν)Un−1(x)(x2−1)x−xν. P(x) = \frac{1}{2n} U_{n-1}(x)\left[P(1)(x+1) + (-1)^n P(-1)(x-1)\right] + \frac{1}{n}\sum _{\nu =1}^{n-1} (-1)^\nu P(x_\nu ) \frac{U_{n-1}(x)(x^2-1)}{x-x_\nu }.
?
Remark.
?

Cf.I. Schur: Math. Z. 4, pp. 273--274 (1919).

Problem 6.74

If one chooses the nnth roots of unity εν=e2πiν/n\varepsilon_\nu = e^{2\pi i \nu /n}, ν=1,2,…,n\nu = 1, 2, \ldots , n, as interpolation points, then the basis polynomials are

fν(x)=ενnxn−1x−εν,ν=1,2,…,n. f_\nu (x) = \frac{\varepsilon _\nu }{n} \frac{x^n-1}{x-\varepsilon _\nu }, \qquad \nu = 1, 2, \ldots , n.

Hence for an arbitrary polynomial P(x)P(x) of degree n−1n-1 we have the representation

P(x)=1n∑ν=1nενP(εν)xn−1x−εν. P(x) = \frac{1}{n} \sum _{\nu =1}^{n} \varepsilon _\nu P(\varepsilon _\nu ) \frac{x^n-1}{x-\varepsilon _\nu }.
?
Problem 6.75

Let the polynomials P(x)P(x) and Q(x)Q(x) be both of degree nn, and let x0,x1,…,xnx_0, x_1, \ldots , x_n be any n+1n+1 distinct or in part coinciding real or complex numbers. From the n+1n+1 equations

P(x0)=Q(x0),P′(x1)=Q′(x1),P′′(x2)=Q′′(x2),…,P(n)(xn)=Q(n)(xn) P(x_0) = Q(x_0), \qquad P'(x_1) = Q'(x_1), \qquad P''(x_2) = Q''(x_2), \qquad \ldots , \qquad P^{(n)}(x_n) = Q^{(n)}(x_n)

it follows that we have identically

P(x)=Q(x). P(x) = Q(x).
?
Problem 6.76

It is possible, and in fact in a unique manner, to determine, corresponding to n+1n+1 arbitrary given numbers c0,c1,c2,…,cnc_0, c_1, c_2, \ldots , c_n, a polynomial P(x)P(x) of degree ≦n\leqq n such that

P(0)=c0,P′(1)=c1,P′′(2)=c2,…,P(n)(n)=cn P(0) = c_0, \qquad P'(1) = c_1, \qquad P''(2) = c_2, \qquad \ldots , \qquad P^{(n)}(n) = c_n

(cf.problem 75). This polynomial P(x)P(x) may be represented in the form

P(x)=P(0)A0(x)+P′(1)A1(x)+P′′(2)A2(x)+⋯+P(n)(n)An(x), P(x) = P(0)A_0(x) + P'(1)A_1(x) + P''(2)A_2(x) + \cdots + P^{(n)}(n)A_n(x),

where A0(x),A1(x),…,An(x)A_0(x), A_1(x), \ldots , A_n(x) are arithmetically determined polynomials, independent of the arbitrarily assigned numbers c0=P(0)c_0 = P(0), c1=P′(1)c_1 = P'(1), , cn=P(n)(n)c_n = P^{(n)}(n). (They may be thought of as the ``basis polynomials'' of this interpolation, which is essentially different from the Lagrange interpolation.)

?