Some theoretical results on fractional-order continuous information measures (Q6541968)

Shannon defined differential entropy of a continuous random variable \(X\) with probability density function \({f_X}(x)\) and support \({S_X} \subseteq \mathcal{R}\) as \(h(X): = \int_{{S_X}} {{f_X}} (x)\log {f_X}(x)dx\). The main idea of this paper is to replace Shannon's formula with \(h(X): = \mathop {\lim }\limits_{t \to - 1} \frac{d}{{dt}}\int_{{S_X}} {f_X^{ - 1}(x)dx} \). Assuming that the function \(f\) is integrable and differentiable and the integral and the limit exist the ordinary differential operator \(\frac{d}{{dt}}\) is replaced with the Rieman-Liouville (RL) fractional derivative \(({}^{RL}D_{{a^ + }}^\alpha f)(t) = {(\frac{d}{{dt}})^n}({}^{RL}I_{{a^ + }}^{n - \alpha }f)(t)\), \(a \in \mathcal{R}\), \(t > a\), \(\alpha > 0\), which reduces to \(\frac{d}{{dt}}\) when \(\alpha \to 1\); the RL fractional integral is defined as \(({}^{RL}I_{{a^ + }}^\alpha f)(t) = \frac{1}{{\Gamma (\alpha )}}\int_a^t {{{(t - x)}^{\alpha - 1}}f(x)dx} \), where \(\Gamma ( \cdot )\) is the Euler's gamma function. Some examples of the fractional entropy of some common continuous probability distributions are given and some properties of the fractional joint and conditional differential entropy and the fractional divergence and mutual information are proved; these may find application, e.g. in computer vision and information theory.