Example: Let’s compare how well a 3rd order polynomial interpolation does when the data points are chosen to be at the Chebyshev nodes, compared to when they are equally spaced as previous. f (x) = sin x, x ∈ [0, π]. Using n = 3 gives 4 equally-spaced points x0 = 0, x1 = π/3, x2 = 2π/3 and x4 = π, with √ f (0) = f (π) = 0, f (π/3) = f (2π/3) = 3/2. We have previously found p3 (x) that passes through these four points: p3 (x) : 0 :0 = L0 (x)f (x0 ) + L1 (x)f (x1 ) + L2 (x)f (x2 ) + L3 (x)f (x3 ) √ √ 3 3 x(x − π/3)(x − π) x(x − 2π/3)(x − π) · + · = (π/3)(−π/3)(−2π/3) 2 (2π/3)(π/3)(−π/3) 2 We would like to compare this with the approximation p∗3 (x) based on the “Chebyshev zeros” of Tn+1 (x) = T4 (x). We can find the zeros immediately, as they are given by property 2 of Chebyshev polynomials: xCk = cos (2k + 1)π 2(n + 1) k = 0, 1, 2, 3. It’s clear that Tn+1 (xCk ) = 0, since by definition Tn+1 (x) = cos((n + 1) cos−1 x). Evaluating the zeros: xC0 = cos xC2 = cos π ∼ 0.9239, 8 3π ∼ 0.3827 8 7π = cos ∼ −0.9239 8 xC1 = cos 5π ∼ −0.3827, 8 xC3 These points lie between [−1, 1]. To map them onto the range [a, b] = [0, π], we use the formula: xk = 1 1 π π (a + b) + (b − a)xCk = + xCk . 2 2 2 2 Hence, the Chebyshev nodes for our particular problem are x0 = x1 = x2 = x3 = π 2 π 2 π 2 π 2 π 2 π + 2 π + 2 π + 2 + · 0.9239 ∼ 3.0220 ∼ 0.9619π · 0.3827 ∼ 2.1719 ∼ 0.6913π · −0.3827 ∼ 0.9697 ∼ 0.3087π · −0.9239 ∼ 0.1196 ∼ 0.0381π Using these nodes imply we should use the data points (x0 , f (x0 )) = (0.961π, 0.1192), (x1 , f (x1 )) = (0.6913π, 0.8247) (x2 , f (x2 )) = (0.3087π, 0.8247), (x3 , f (x3 )) = (0.0381π, 0.1192) Substituting these values into the Lagrange interpolation formula gives p∗3 (x) = −1.3663 × 10−8 x3 − 0.3948x2 + 1.2405x − 0.0000. 1 Let’s see how well these do compared to each other: 1.0 0.02 0.8 0.00 f(x) f(x) - p(x) 0.6 0.4 -0.02 0.2 -0.04 0.0 0 pi/3 2pi/3 pi 0 x pi/3 2pi/3 pi x Figure 3.1: (a) f (x) = sin x between [0, π] (thick solid line) ; 3rd order interpolation polynomial (p3 (x)) with equally spaced interpolation points (thin solid line, interpolation points shown as crosses); 3rd order polynomial interpolation with Chebyshev nodes (p∗3 (x)) (dash-dotted line, Chebyshev nodes shown as diamonds). (b) Error in the interpolation: f (x) − p3 (x) (solid line) and f (x) − p∗3 (x). We see that overall, using the Chebyshev nodes achieves a more accurate fit, even though in places, the other interpolation polynomial does better. 2
© Copyright 2026 Paperzz