International Journal of Engineering and Computational Applications  |  ISSN (Online): 3107-6580  |  Double-Blind Peer Review  |  Open Access  |  CC BY 4.0

Current Issues
     2026:2/5

International Journal of Engineering and Computational Applications

ISSN: (Print) | 3107-6580 (Online) | Open Access

Coefficient-Guided Chebyshev–PCHIP Approximation for Mixed-Regularity Functions

Full Text (PDF)

Open Access - Free to Download

Download Full Article (PDF)

Abstract

In this paper we propose and evaluate a coefficient-decay-triggered hybrid approximation for one-dimensional functions with changing regularity in their domain. If the coefficient-tail indication and independent validation residual are small, local Chebyshev interpolation is appropriate; otherwise the interval is recursively bisected. Shape-preserving cubic piecewise Hermite interpolation (PCHIP) reconstructs consistently uneven end intervals. A jump classifier activates one-sided PCHIP pieces to prevent misleading continuous transitions. validation guided, practical acceptance test, not an a posteriori limit, the off-grid residual. The five deterministic benchmarks are analytic, cusp, algebraic-singularity, jump and mixed regularity. The proposed method is benchmarked against global Chebyshev interpolation, global PCHIP, cubic splines, and an adaptive Chebyshev-only baseline. On the mixed benchmark it achieves RMSE 1.41×10⁻⁵ versus adaptive Chebyshev (5.24×10⁻³), global Chebyshev (2.28×10⁻²), and global PCHIP (2.13×10 At a construction budget of 1,317 function evaluations, global Chebyshev, PCHIP, and cubic splines hold RMSE levels of 1.07×10⁻², 6.00×10⁻³, and 6.56×10⁻³, respectively, whereas the hybrid representation holds only 344 degrees of freedom Global Chebyshev still dominates the analytic benchmark, showing that switching is generally advantageous when regularity is spatially nonuniform. Sensitivity studies demonstrate that the baseline mixed-function partition is insensitive to tail thresholds of 3 × 10⁻⁵ to 10⁻² and three to eight tail coefficients The jump classifier identifies 20 piecewise-constant leaps across five locations and four amplitudes after automatic bracketing, whereas selected non-jump controls are not misclassified for thresholds 4 to 24. Instead of substituting the spectral approximation, we propose a regularity dependent validation guided switching approach.

How to Cite This Article

Hameed Hasan Obaid Yasari (2025). Coefficient-Guided Chebyshev–PCHIP Approximation for Mixed-Regularity Functions . International Journal of Engineering and Computational Applications (IJECA), 1(6), 70-82. DOI: https://doi.org/10.54660/.IJECA.2025.1.6.70-82

Export Citation:

BibTeX RIS EndNote

Share This Article: