Spectral Collocation with Variable Transformation for Riemann–Liouville Fractional Integral Equations with Non-Smooth Solutions: Stability and Conditioning Analysis
الملخص
Weakly singular Volterra integral equations of Riemann–Liouville type with rational order have non-smooth solutions, so polynomial collocation converges slowly. This paper analyses spectral collocation after a power-law change of the independent variable that makes the solution smooth. The scheme is the Chebyshev–Gauss case of the Müntz collocation method of D. Hou, Y. Lin, M. Azaiez and C. Xu (J. Sci. Comput. 81 (2019) 2162–2187). The main result is an explicit bound on the condition number that grows like the squared logarithm of the polynomial degree. Its constant uses a new resolvent estimate based on the complete monotonicity of Mittag-Leffler functions. For dissipative problems this removes the exponential growth of the classical Gronwall-type constant: the bound grows quadratically in the coefficient, as observed, and is 35 to 53 times the measured value. For kernels bounded below by a positive constant, a matching lower bound shows that the exponential growth is genuine. Stability for every rational order, error bounds with one logarithmic factor and analyticity of the transformed solution are also proved under Hölder continuity of the kernel; all results are asymptotic. On five test problems, including a fractional relaxation model, the scheme reaches errors of 10⁻¹³ to 10⁻¹⁵ with at most 64 unknowns in 0.4 to 1.8 ms. The method of T. Pu and M. Fasondini (Adv. Comput. Math. 49 (2023) 7), with high-precision matrices, is up to two orders of magnitude more accurate, whereas collocation runs in double precision, accepts non-separable kernels and was several hundred times faster.