Answers are grounded in this paper and cite their sources.
An entropy computed from an interpolated record depends on both reconstruction accuracy and the amplitude quantizer. We bound the difference between the entropies of the computed reconstruction and the reference interpolant under a common sampling rule and quantizer. For nonlinear fractal interpolation operators, including those whose optimal Lipschitz constant equals one, ordinate perturbations and inexact iteration yield an explicit uniform reconstruction budget. Hard-bin Shannon entropy has no uniform continuity modulus in ∥f−g∥∞ alone; a computable conditional bound uses the weight of samples near bin boundaries. Piecewise-linear soft assignments instead give a distributional error controlled by E/h, where E bounds the amplitude error and h is the grid spacing. When E≤h, an adjacent-label coupling yields a sharp Shannon bound involving at most three labels, independently of the total alphabet size when it is at least three. Its small-error order is (E/h)log(h/E). Sharpness concerns quantization alone. The nonlinear data response can dominate the reconstruction budget even before quantization. We also give uniform and data-dependent bounds for Rényi orders greater than one. Combining these estimates gives stopping criteria and sufficient conditions for reliable entropy comparisons. A reproducible example distinguishes observed differences from the guarantees supplied by the reconstruction budget and quantizer resolution.
No takes yet. Share an insight, caveat, or question.
Attia et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: