A Newton Method for Shape-Preserving Spline Interpolation
Summary (1 min read)
If the integral over [a, b] of the piecewise linear function (
- Furthermore, in this case quadratic convergence of the Newton method would follow directly from [13] .
- The following example of dimension 2 shows that such an argument does not work.
- Hence the function above is not piecewise smooth.
- Then the positive semidefiniteness of the elements of ∂F − (λ) follows from the fact that any matrix in the generalized Jacobian of the gradient of a convex function must be symmetric and positive semidefinite.
- By combining the above lemmas and applying Theorem 1.1, the authors obtain the main result of this paper which settles the question posed in [11] .
4. Numerical results.
- In Figures 1-5 , the dashed line is for the resulting shape-preserving cubic spline (using the data obtained with the starting point λ 0 = sign(d)); the solid line is for the natural spline (using the MATLAB SPLINE function), and "o" stands for the original given data.
- In Table 1 for results of the numerical experiments the authors use the following notation: 1 , they observe that Algorithm 3.2 converges rapidly to the solution from both starting points for all problems except Example 4.4 (y 9 = 4), to which the algorithm within 30 iterations failed to produce an approximate solution meeting the required accuracy.
- These observations indicate that how far away from zero each divided difference is may make a big difference in the numerical performance of the algorithm.
- The problem is, however, nonsmooth, and here the authors are entering a new territory.
Did you find this useful? Give us your feedback
Citations
Cites background from "A Newton Method for Shape-Preservin..."
...Dontchev, Qi, Qi and Yin [ 11 ] proved the nonsmooth equation reformulation is strongly semismooth and established quadratic convergence of the generalized Newton method for solving this problem....
[...]
References
9,498 citations
"A Newton Method for Shape-Preservin..." refers background in this paper
...As a side result, from Lemma 2.3 and the Clarke inverse function theorem [2, Theorem 7.1.1], we obtain that the solution of the problem (3) is a Lipschitz continuous function of the interpolation values yi....
[...]
...For a locally Lipschitz continuous function G : Rn → Rn, the generalized Jacobian ∂G(x) of G at x in the sense of Clarke [2] is the convex hull of all limits obtained along sequences on which G is differentiable: ∂G(x) = co { lim xj→x ∇G(xj) | G is differentiable at xj ∈ Rn } ....
[...]
...For a locally Lipschitz continuous function G : R → R, the generalized Jacobian ∂G(x) of G at x in the sense of Clarke [2] is the convex hull of all limits obtained along sequences on which G is differentiable:...
[...]
1,464 citations
296 citations
136 citations
"A Newton Method for Shape-Preservin..." refers methods in this paper
...Furthermore, in this case quadratic convergence of the Newton method would follow directly from [13]....
[...]
130 citations
"A Newton Method for Shape-Preservin..." refers methods in this paper
...For more discussion of the inexact Newton method, we refer to [3, 7, 14]....
[...]