Polynomial Interpolation

Polynomial Interpolation

Numerical Methods 401-0654 6 Polynomial Interpolation Throughout this chapter n N is an element from the set N if not otherwise specified. ∈ 0 0 6.1 Polynomials D-ITET, ✬ ✩D-MATL Definition 6.1.1. We denote by n n 1 n := R t αnt + αn 1t − + ... + α1t + α0 R: α0, α1,...,αn R . P ∋ 7→ − ∈ ∈ the R-vectorn space of polynomials with degree n. o ≤ ✫ ✪ ✬ ✩ Theorem 6.1.2 (Dimension of space of polynomials). It holds that dim n = n +1 and n P P ⊂ 6.1 C∞(R). p. 192 ✫ ✪ n n 1 ➙ Remark 6.1.1 (Polynomials in Matlab). MATLAB: αnt + αn 1t − + ... + α0 Vector Numerical − Methods (αn, αn 1,...,α0) (ordered!). 401-0654 − △ Remark 6.1.2 (Horner scheme). Evaluation of a polynomial in monomial representation: Horner scheme p(t)= t t t t αnt + αn 1 + αn 2 + + α1 + α0 . ··· − − ··· Code 6.1: Horner scheme, polynomial in MATLAB format 1 function y = polyval (p,x) 2 y = p(1); for i =2: length (p), y = x y+p( i ); end ∗ D-ITET, D-MATL Asymptotic complexity: O(n) Use: MATLAB “built-in”-function polyval(p,x); △ 6.2 p. 193 6.2 Polynomial Interpolation: Theory Numerical Methods 401-0654 Goal: (re-)constructionofapolynomial(function)frompairs of values (fit). ✬ Lagrange polynomial interpolation problem ✩ Given the nodes < t < t < < tn < and the values y ,...,yn R compute −∞ 0 1 ··· ∞ 0 ∈ p n such that ∈P j 0, 1,...,n : p(t )= y . ∀ ∈ { } j j ✫ ✪ D-ITET, D-MATL 6.2 p. 194 6.2.1 Lagrange polynomials Numerical Methods 401-0654 ✬ ✩ Definition 6.2.1. Consider nodes <t <t < <tn < ( Lagrange interpolation). −∞ 0 1 ··· ∞ → Then define the Lagrange polynomials L ,L ,...,Ln n associated to (t ,...,tn) through 0 1 ∈P 0 n t tj L (t) := − i t t j=0 i − j jY=i 6 for all t R and all i 0, 1,...,n . ∈ ∈ { } ✫ ✪D-ITET, D-MATL 1 : i = j Note that L (t )= δ = . i j i,j 0 : i = j ( 6 6.2 p. 195 8 L 0 Numerical L 2 Methods L 5 401-0654 6 Example 6.2.1. Consider n N and Lagrange ∈ polynomials for uniformly spaced nodes 4 2j t = 1+ ,j 0, 1,...,n . 2 j − n ∈ { } Plot ➙ 0 n = 10, j =0, 2, 5 Lagrange Polynomials ✸ -2 -4 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 t Fig. 36 D-ITET, D-MATL ✬ ✩ Definition 6.2.2 (Lagrange interpolation operator). Consider <t0 <t1 <...<tn < . n+1 −∞ ∞ By I : R n we denote the linear mapping defined by t0,...,tn →P n n+1 R (y ,...,yn) I y ,...,yn := y L n . ∋ 0 7→ t0,...,tn 0 i i ∈P i=0 X ✫ ✪ 6.2 p. 196 ✬ ✩Numerical Theorem 6.2.3 (Existence & uniqueness of Lagrange interpolation polynomial). Methods 401-0654 The general polynomial interpolation problem (6.2) admits a unique solution p n and it holds ∈P that n p(t)= y L (t)= I (y ,...,yn) (t) i · i t0,...,tn 0 Xi=0 for all t R. ∈ ✫ ✪ 6.2.2 Conditioning of polynomial interpolation D-ITET, D-MATL Necessary for studying the conditioning: norms on vector space of continuous functions C(I, R) where I R is a non-empty bounded and closed interval: ⊂ supremum norm f ,I := sup f(t) : t I , (6.2.4) k k∞ {| | ∈ } L2-norm f := f(t) 2 dt , (6.2.5) k k2,I | | sZI 6.2 L1-norm f := f(t) dt . (6.2.6) k k1,I | | p. 197 ZI Numerical Methods 401-0654 ✬ ✩ Lemma 6.2.6 (Absolute conditioning of polynomial interpolation). Consider interval I R, ⊂ nodes < t <...<tn < and associated Lagrange polynomials L ,...,Ln. Then −∞ 0 ∞ 0 the Lagrange interpolation operator satisfy I t0,...,tn(y) ,I n It ,...,t = sup k k∞ = Li , (6.2.7) 0 n L( , ,I) y i=0 k·k∞ k·k∞ y Rn+1 0 | | ,I ∈ \{ } k k∞ X ∞ I 1 t0,...,tn(y) 2,I n 2 2 It ,...,t = sup k k L . (6.2.8) 0 n L( 2, 2,I) i=0 i 2,I k·k k·k y Rn+1 0 y 2 ≤ k k ∈ \{ } k k X ✫ ✪D-ITET, D-MATL n Terminology: Lebesgue constant of t0,...,tn and I R: λt ,...,t (I) := Li . ⊂ 0 n i=0 | | ,I ∞ X Example 6.2.4 (Estimation of the Lebesgue constant). Consider n N, I = [ 1, 1] and t = ∈ − k 6.2 p. 198 2k 1+ n , k 0, 1,...,n (uniformly spaced nodes). Then Numerical − ∈ { } Methods 401-0654 1 1 3 ... n 3 n+1 2n 1 λ (I) L (1 1) = n · n · n · · −n · n ···· n− t0,...,tn ≥| n/2 − n | 2 2 4 n 2 1 n · n ····· −n · 1 3 5 ... (n 3) (n + 1) (n + 2) ... (2n 1) = · · · · − · · · · − [2 4 6 ... (n 2) n]2 · · · · − · n (n + 1) (n + 3) ... (2n 1) √2 · · · − C 1.4n ≥ 2 4 ... (n 4) (n 2)2 n2 ≥ (n 2) n2 ≥ · · · · − · − · − for all n 6, 8, 10,... and some constant C (0, ). ∈ { } ∈ ∞ Theory[1]: for uniformly spaced nodes n/2 for all N and some constant λt0,...,tn(I) Ce n D-ITET, ≥ ∈ D-MATL C (0, ). ∈ ∞ ✸ Example 6.2.5 (Oscillating interpolation polynomial: Runge’s counterexample). Between the nodes the interpolation polynomial can oscillate excessively and overestimate the changes in the values: bad approximation of functions! 6.2 p. 199 2 Interpol. polynomial Numerical Methods 401-0654 1.5 Let n N. Interpolation polynomial with uniformly ∈ spaced nodes: 1 10j 1 tj := 5+ n , yj = 2 − 1+(tj) 0.5 for all j 0, 1,...,n . Plot n = 10 ➙ Interpolation polynomial ∈ { } See Example 6.4.1. 0 -0.5 -5 -4 -3 -2 -1 0 1 2 3 4 5 t D-ITET, D-MATL Attention: strong oscillations of the interpolation polynomials of high degree on uniformly spaced nodes! ✸ 6.3 p. 200 6.3 Polynomial Interpolation: Algorithms Numerical Methods 401-0654 6.3.1 Newton basis polynomials and divided differences Drawback of the Lagrange basis: adding another data point affects all basis polynomials! Alternative, “update friendly” method: Newton basis of n. P ✬ ✩ Definition 6.3.1. Consider t ,...,tn R with t ,...,tn = n +1. Then define Newton 0 ∈ |{ 0 }| basis polynomials N0,N1,...,Nn n associated to (t0,...,tn) through D-ITET, ∈P D-MATL n 1 − N (t):=1 , N (t):=(t t ) , ... , Nn(t) := (t t ) (6.3.1) 0 1 − 0 − i iY=0 for all t R. ∈ ✫ ✪ ➣ LSE for polynomial interpolation problem in Newton basis: R R Given (yj)j 0,1,...,n , find (aj)j 0,1,...,n such that ∈{ } ⊂ ∈{ } ⊂ 6.3 j 0, 1,...,n : a N (t )+ a N (t )+ + anNn(t )= y . p. 201 ∀ ∈ { } 0 0 j 1 1 j ··· j j triangular linear system Numerical ⇔ Methods 401-0654 a0 y0 N0(t0) N1(t0) N2(t0) Nn(t0) ··· a1 y1 N (t ) N (t ) N (t ) Nn(t ) . . 0 1 1 1 2 1 ··· 1 . N (t ) N (t ) N (t ) Nn(t ) . = . , 0 2 1 2 2 2 ··· 2 . . . an 1 yn 1 N0(tn) N1(tn) N2(tn) Nn(tn) − − ··· an yn i.e., 10 0 0 a0 y0 ···.. 1 (t1 t0) 0 . a1 y1 − . . . 1 (t2 t0) (t2 t0)(t2 t1) . . −. − .. − .. . = . 0 . . n 1 D-ITET, − an 1 yn 1 D-MATL 1 (tn t0) (tn t0)(tn t1) (tn ti) − − − − − ··· i=0 − an yn Q 6.3 p. 202 Solution of the system with forward substitution: Numerical Methods 401-0654 a0 = y0 , y a y y a = 1 − 0 = 1 − 0 , 1 t t t t 1 − 0 1 − 0 y1 y0 y2 y0 y1 y0 y2 a0 (t2 t0)a1 y2 y0 (t2 t0) t −t t −t t −t a = − − − = − − − 1− 0 = 2− 0 − 1− 0 , 2 (t t )(t t ) (t t )(t t ) t t 2 − 0 2 − 1 2 − 0 2 − 1 2 − 1 y3 y0 y1 y0 y3 a0 (t3 t0) a1 (t3 t0)(t3 t1) a2 t −t t −t (t3 t1) a2 a = − − − − − − = 3− 0 − 1− 0 − − 3 (t t )(t t )(t t ) (t t )(t t ) 3 − 0 3 − 1 3 − 2 3 − 1 3 − 2 y y y y y y y y 3− 0 1− 0 2− 0 1− 0 t3 t0 − t1 t0 t2 t0 − t1 t0 − t t − − t t − = 3− 1 − 2− 1 (t t ) D-ITET, 3 − 2 D-MATL . Simpler and more efficient algorithm using divided differences: y[ti] = yi y[ti+1,...,ti+k] y[ti,...,ti+k 1] y[t ,t ,...,t ] = − − (recursion) (6.3.2) i i+1 i+k t t 6.3 i+k − i p. 203 Recursive calculation by divided differences scheme: Numerical Methods 401-0654 t0 y[t0] > y[t0,t1] t1 y[t1] > y[t0,t1,t2] > y[t1,t2] > y[t0,t1,t2,t3], t2 y[t2] > y[t1,t2,t3] > y[t2,t3] t3 y[t3] Then n n t R: y L (t)= I (y ,...,yn) (t)= y[t ,...,t ] N (t) ∀ ∈ i · i t0,...,tn 0 0 i · i i=0 i=0 X X D-ITET, D-MATL and y[t0,t1,...,yi]= y[tπ(0),tπ(1),...,tπ(i)] for all permutations π : 0, 1,...,i 0, 1,...,i and all i 0, 1,...,n (see, e.g., Satz 2.2 { } → { } ∈ { } and Korollar 2.1 in [44, Rannacher, Einführung in die Numerische Mathematik, Vorlesungsskriptum, 2006]).

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    51 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us