Two-Parameter Mittag-Leffler Function

• The two-parameter Mittag-Leffler function is a generalization of the exponential function defined by an infinite power series with parameters that control its monotonic or oscillatory behavior.
• It plays a central role in fractional calculus by modeling memory effects and fractional differential equations, thereby enabling advanced analytical and numerical solutions.
• Advanced approximation techniques, including rational approximants and contour integrals, efficiently evaluate the function even for complex and matrix arguments.

The two-parameter Mittag-Leffler function, typically denoted as $E_{\alpha,\beta}(z)$, is a prominent special function fundamental to fractional calculus, the analytic theory of fractional differential and integral equations, and models with memory effects. Entire in $z \in \mathbb{C}$ for all $\Re \alpha > 0$, $\beta \in \mathbb{C}$, it generalizes the exponential and trigonometric functions and exhibits a spectrum of behaviors dictated by its parameters, ranging from complete monotonicity to oscillatory, sign-changing dynamics. Its computation, asymptotics, parameter sensitivities, inequalities, and matrix analogues are active research foci due to both theoretical interest and the demands of time-fractional numerical simulation.

1. Series Definition, Special Cases, and Entireness

The canonical definition is the entire power series

$$E_{\alpha,\beta}(z) = \sum_{k=0}^\infty \frac{z^k}{\Gamma(\alpha k + \beta)},$$

where $\alpha > 0$ and $\beta \in \mathbb{C}$, and $\Gamma(\cdot)$ is Euler's Gamma function. The radius of convergence is infinite for all such parameter values, as assured by

$$\Gamma(\alpha k + \beta) \sim (2\pi)^{1/2} (\alpha k)^{\alpha k + \beta - 1/2} e^{-\alpha k} \quad (k \to \infty),$$

hence the ratio test yields $R = \infty$ (Honain et al., 2023, Mieghem, 2020, Rogosin et al., 2024).

Distinguished special cases include:

• $E_{1,1}(z) = e^z$, the exponential,
• $E_{2,1}(-z^2) = \cos z$, $z E_{2,2}(-z^2) = \sin z$,
• For $\beta=1$, $E_{\alpha,1}(z) \equiv E_\alpha(z)$ is the classical (one-parameter) Mittag-Leffler function.

The function is of order $1/\alpha$ and type $1$ as an entire function. As a consequence, the zero set exhibits a density and distribution determined by $\alpha$; for $0 < \alpha < 1$ and $\beta \ge 1$, $E_{\alpha, \beta}(-x)$ is non-oscillatory and completely monotone for $x>0$, while for $\alpha>1$ and suitable $\beta$, oscillations and sign changes appear (Honain et al., 2023, Garrappa et al., 2024).

2. Integral, Asymptotic, and Functional Representations

Multiple transformations provide alternative forms critical for analysis and computation:

• Hankel- or Bromwich-type representations: For suitable parameters, the function admits

$$E_{\alpha,\beta}(z) = \frac{1}{2\pi i}\int_\gamma \frac{w^{\alpha-\beta} e^w}{w^\alpha - z} dw,$$

where $\gamma$ is a suitable contour encircling the singularities (Cardoso, 2023, Mieghem, 2020).

• Real-variable integral representations: For $\rho>1/2$, explicit real-kernel forms exist (often denoted Representation “A” on a ray plus arc, or “B” as a single integral) (Saenko, 2020, Saenko, 2020). For example,

$$E_{\rho,\mu}(z) = \int_{1+\varepsilon}^\infty K_{\rho,\mu}(r,...) dr + \int_{...}^{...} P_{\rho,\mu}(1+\varepsilon,\phi,...) d\phi,$$

with kernels $K$, $P$ explicitly constructed to facilitate robust quadrature.

• Mellin-Barnes integrals and Laplace transforms:

$$E_{\alpha,\beta}(z) = \frac{1}{2\pi i}\int_C \frac{\Gamma(s)\Gamma(1-s)}{\Gamma(\beta - \alpha s)}(-z)^{-s} ds,$$

supporting sectorial asymptotic analysis and parameter differentiation (Rogosin et al., 2024, Paris, 2019).

• Asymptotic Expansions: For large $|z|$ ($0<\alpha<2$), the main regime is

$$E_{\alpha,\beta}(z) \sim \frac{1}{\alpha} z^{\frac{1-\beta}{\alpha}} \exp(z^{1/\alpha}) - \sum_{j=1}^N \frac{z^{-j}}{\Gamma(\beta - \alpha j)} + O(|z|^{-N-1}),$$

with algebraic and exponential contributions depending strongly on angular sectors in $z$ (Honain et al., 2023, Mieghem, 2020, Paris, 2019).

3. Parameter Dependence, Monotonicity, and Inequalities

The parameter regime $(\alpha,\beta)$ crucially dictates analytic, monotonic, and oscillatory properties:

• Completely monotone regime: If $0 < \alpha \leq 1$, $\beta \geq \alpha$, then $E_{\alpha, \beta}(-t)$ is completely monotone for $t \geq 0$; log-convexity and corresponding inequalities $|E_{\alpha,\beta}(z)| \leq E_{\alpha,\beta}(\Re z)$ hold globally (Garrappa et al., 2024).
• Reverse inequalities: For $1 \leq \alpha < 2$, $\beta \in [\alpha-1, \alpha]$, $1/E_{\alpha,\beta}(x)$ is completely monotone, and in specified regimes, $|E_{\alpha,\beta}(z)| \geq E_{\alpha,\beta}(\Re z)$ holds (Garrappa et al., 2024).
• Zero Distribution and Oscillatory Transitions: For $\alpha \in (1,2)$, as $\alpha \to 2$, the number of real zeros of $E_{\alpha, \beta}(-t)$ grows, and complete monotonicity is lost. Phase diagrams with curves $\varphi(\alpha), \psi(\alpha)$ separate monotone from oscillatory regions, with transitions where the zeros move to the derivative or function remains sign-definite but oscillatory (Honain et al., 2023).
• Log-convexity/concavity: $E_{\alpha,\beta}(x)$ is log-convex for $\alpha\le1, \beta \ge h(\alpha)$; log-concave for $\alpha \ge 1, \beta\le h(\alpha)$ with $h(\alpha)$ an explicit function derived from Gamma identities (Garrappa et al., 2024).

4. Advanced Approximation and Computational Schemes

Due to its entire nature and nontrivial growth, direct evaluation via the power series is computationally expensive, especially for $z$ large or for matrix arguments. The following schemes are established:

• Rational Approximants and “Derooting”: For $\alpha \in (1,2)$, the oscillatory nature of $E_{\alpha,\beta}(-t)$ is captured by the derooted decomposition:

$$E_{\alpha,\beta}(-t) = (-t)^r E_{\alpha,\beta+\alpha r}(-t) + P_{\alpha,\beta}^{(r-1)}(-t),$$

where $P_{\alpha,\beta}^{(r-1)}$ is an explicit polynomial. Rational Padé-type approximants $R^{m,n}_{\alpha,\beta}(t)$ are constructed for $E_{\alpha, \beta + \alpha r}(-t)$; the full approximant $R_{\alpha,\beta}^{m,n,r}(t)$ then matches zeros up to $r$ and achieves high uniform accuracy across oscillatory and monotone regimes (Honain et al., 2023). For instance, $R_{\alpha,\beta}^{13,4,5}(t)$ tracks five real zeros with max error $\sim10^{-3}$.

• Matrix Argument Evaluation: For $E_{\alpha,\beta}(A)$ ($A \in \mathbb{C}^{N\times N}$), rational approximants are implemented by evaluating $$R^{m,n}(A) := [\Gamma(\beta - \alpha)]^{-1} Q_n(A)^{-1} P_m(A)$$ and its derooted variants, using direct inversion, Sylvester-type linear solves, partial fraction expansion, or diagonalization. This yields speedups of up to $40\times$ relative to standard routines, with relative errors as low as $10^{-6}$ for $N=100$ matrices (Honain et al., 2023).
• Laplace Transform and Contour Integral Numerical Inversion: The Laplace transform formula

$$\mathcal{L}\{ t^{\beta-1} E_{\alpha,\beta}(\lambda t^\alpha) \} = s^{\alpha - \beta}/(s^\alpha - \lambda)$$

is inverted along a parabolic or Hankel-type optimal contour, allowing evaluation with rigorously controlled error via trapezoidal summation and residue subtraction (Garrappa, 2015). This facilitates double-precision or higher-precision implementations across a range of parameter values.

5. Analytical Properties, Parameter Differentiation, and Higher Generalizations

• Parameter Differentiation and Uniform Convergence: Term-by-term differentiation with respect to $\alpha$ and $\beta$ is justified by uniform convergence. For example,

$$\frac{\partial}{\partial \alpha} E_{\alpha, \beta}(z) = -\sum_{k=1}^\infty \frac{k \psi(\alpha k + \beta)}{\Gamma(\alpha k + \beta)} z^k,$$

where $\psi$ is the digamma function. Mellin-Barnes integral representations support differentiation under the integral sign, reinforcing the analytic structure of parameter dependence (Rogosin et al., 2024).

• Generalizations to Multi-parameter Functions: Methods extend to the Prabhakar function, Le Roy and Wright types, and higher-order parameter families via explicit series and Mellin-Barnes integrals, maintaining analytic and computational tractability (Rogosin et al., 2024).

6. Applications in Fractional Calculus and Mathematical Physics

The two-parameter Mittag-Leffler function is central in closed-form solutions to fractional relaxation/oscillation, fractional diffusion-wave equations, fractional plasma oscillations, and convolution-quadrature schemes in fractional integral evolution. Models with memory, anomalous transport, and viscoelasticity exploit $E_{\alpha, \beta}$ as the analytic backbone, and the efficient evaluation of $E_{\alpha, \beta}(-tA)$ in PDE solvers enables precise, scalable simulation in high-dimensional or matrix-variable settings (Honain et al., 2023).

7. Summary Table: Key Regimes of $E_{\alpha, \beta}(-t)$ Behavior

$\alpha$ Range	$\beta$ Range	Monotonicity / Oscillation	Numerical Regime
$0 < \alpha \le 1$	$\beta \ge \alpha$	Completely monotone	Standard rational approximants
$1 < \alpha < 2$	$\beta \ge 1$	Oscillatory, real zeros	Derooted rational approximants
$\alpha \ge 2$	$\beta$ below threshold	Monotone, all zeros real	Rational/Padé, efficient solvers

The operational regime—monotone, oscillatory, or mixed—determines both the analytic and computational strategy. The modern approximation schemes exploit decompositions, parameter transformations, and contour analysis to ensure accuracy and stability across domains (Honain et al., 2023, Sarumi et al., 2019).

---

For a rigorous survey of analytic theory, rational approximation, and computational practice for $E_{\alpha, \beta}$ in oscillatory and non-oscillatory regimes, see (Honain et al., 2023). For parameter dependence, monotonicity, and inequality theory, consult (Garrappa et al., 2024). For real-variable integral representations and robust numerical quadrature, see (Saenko, 2020, Saenko, 2020). For matrix argument computation, (Cardoso, 2023) and (Honain et al., 2023) provide detailed methodologies.