Discrete Fractional Calculus: Theory & Applications

• Discrete fractional calculus is a framework defining non-integer order difference operators on discrete grids and time scales, uniting discrete and continuous analysis.
• It employs operators such as Riemann–Liouville and Caputo forms, using discrete convolutions with Gamma functions to capture nonlocal memory effects.
• The theory underpins computational schemes, variational principles, and practical applications in signal processing, anomalous diffusion, and optimization.

Discrete fractional calculus is the field studying discrete analogues of fractional integration and differentiation operators—specifically, operators of non-integer order defined on difference spaces such as the integer lattice $\mathbb{Z}$, uniform grids $h\mathbb{Z}$, or arbitrary time scales. This theory unifies nonlocal memory effects with discrete-time modeling, serving both as a theoretical framework and as a numerical bridge between classical discrete differences and continuous fractional calculus.

1. Foundational Operators and Algebraic Structure

The central objects of discrete fractional calculus are the discrete fractional sum (integral) and difference operators of Riemann–Liouville and Caputo type, defined in delta (forward) and nabla (backward) variants. On the integer lattice $\mathbb{N}_a = \{a, a+1, \dots\}$:

• Left (forward) delta Riemann–Liouville fractional sum of order $\alpha > 0$:

$$(_a\Delta^{-\alpha} f)(t) = \frac{1}{\Gamma(\alpha)} \sum_{s=a}^{t-\alpha} (t-\sigma(s))^{(\alpha-1)} f(s)$$

with $\sigma(s) = s+1$, and $(x)^{(y)} = \Gamma(x+1)/\Gamma(x+1-y)$.

• Left (backward) nabla Riemann–Liouville fractional sum:

$$(_a\nabla^{-\alpha} f)(t) = \frac{1}{\Gamma(\alpha)} \sum_{s=a}^{t} (t-\rho(s))^{\overline{\alpha-1}} f(s)$$

with $\rho(s) = s-1$, $(x)^{\overline{y}} = \Gamma(x + y)/\Gamma(x)$.

Fractional differences are obtained via compositions with (forward/backward) difference operators: $h\mathbb{Z}$0 and analogously for nabla.

Key properties include linearity, the semigroup (composition) law for sums and differences, and inversion formulas connecting sums and differences. Representation in terms of discrete convolutions with kernel sequences (falling/rising factorials weighted by Gamma functions) underlies the translation to analytic and generating-function techniques (Ferreira, 2021, Ferreira, 2020, Abdeljawad et al., 2012).

2. Equivalence of Approaches and Duality

Distinct definitions of discrete fractional operators—Riemann–Liouville, Caputo, Grünwald–Letnikov, and binomial-coefficient-based differences—are equivalent under mild regularity and correct domain shifting. For instance, the binomial (Grünwald–Letnikov) form

$h\mathbb{Z}$1

can be proved to coincide term-by-term with the Riemann–Liouville discrete sum/difference (Abdeljawad et al., 2012).

Dual identities, such as relations between left and right fractional operators and between delta and nabla forms, provide transformation tools that are essential for boundary value problems and the formulation of initial-boundary conditions in discrete fractional equations.

3. Discrete Fractional Calculus on Uniform Grids and Time Scales

Generalization to uniform grids $h\mathbb{Z}$2 introduces the $h\mathbb{Z}$3-factorial: $h\mathbb{Z}$4 enabling the definition of left/right fractional $h\mathbb{Z}$5-sums and $h\mathbb{Z}$6-differences: $h\mathbb{Z}$7 and

$h\mathbb{Z}$8

with analogous right-hand versions (Bastos et al., 2010, Ferreira et al., 2011, Ferreira, 2010, Bastos, 2012). As $h\mathbb{Z}$9, these operators limit to continuous Riemann–Liouville integrals and derivatives. The general time scales theory further extends discrete fractional calculus to arbitrary closed subsets of $\mathbb{N}_a = \{a, a+1, \dots\}$0, allowing for hybrid dynamical systems and applications in nonuniform temporal discretization (Benkhettou et al., 2015, Bastos, 2012).

On arbitrary time scales, nonsymmetric (nabla/delta) and symmetric fractional derivatives are defined pointwise through limiting difference quotients with exponent $\mathbb{N}_a = \{a, a+1, \dots\}$1 (Benkhettou et al., 2015), unifying the discrete and continuous case. The symmetric operator interpolates between the forward and backward settings: $\mathbb{N}_a = \{a, a+1, \dots\}$2 with explicit weights derived from time scale graininess.

4. Calculus of Variations and Discrete Fractional Euler–Lagrange Theory

Discrete fractional calculus furnishes the variational framework for extremal problems involving nonlocal memory on discrete grids. Given a functional

$\mathbb{N}_a = \{a, a+1, \dots\}$3

direct variational arguments yield discrete fractional Euler–Lagrange equations: $\mathbb{N}_a = \{a, a+1, \dots\}$4 for appropriate discrete fractional difference operator $\mathbb{N}_a = \{a, a+1, \dots\}$5 (delta or nabla) (Bastos et al., 2010, Bastos et al., 2010, Pooseh et al., 2012, Abdeljawad, 2017). Second-order Legendre-type conditions have been established to eliminate false extremals (Bastos et al., 2010, Bastos, 2012), involving quadratic forms in second derivatives of $\mathbb{N}_a = \{a, a+1, \dots\}$6 with explicit Gamma-function-weighted corrections.

The unification extends to multi-term functionals, higher-order and isoperimetric problems, and systems on general time scales (Ferreira, 2010, Bastos, 2012). Special summation-by-parts formulas, involving shifted summation indices and endpoint Gamma-factor terms, are necessary to transfer fractional differences and establish natural fractional boundary conditions.

5. Computational Schemes and Finite Difference Approximations

Numerical realization of discrete fractional operators relies on finite difference schemes, including Grünwald–Letnikov weights and binomial–coefficient expansions. For mesh nodes $\mathbb{N}_a = \{a, a+1, \dots\}$7, the approximation

$\mathbb{N}_a = \{a, a+1, \dots\}$8

enables the construction of discrete fractional variational problems as static nonlinear systems (Pooseh et al., 2012). The error is typically $\mathbb{N}_a = \{a, a+1, \dots\}$9, with first-order convergence under suitable smoothness (Pooseh et al., 2012). Matrix-based approaches further generalize this by defining two-band or triangular strip matrices to represent fractional difference operators; Balakrishnan’s semigroup theory provides the matrix fractional power, with explicit formulas for arbitrary real $\alpha > 0$0 (Kolokoltsov et al., 11 Dec 2025, 0811.1355). This is essential for large-scale simulations in fractional PDEs, offering favorable computational complexity and spectral stability properties compared to classical marching-in-time schemes.

Table: Discretization of Fractional Operators

Operator Type	Discrete Definition	Example Reference
Delta left sum	$\alpha > 0$1	(Bastos et al., 2010)
Nabla left sum	$\alpha > 0$2	(Abdeljawad, 2017)
GL finite difference	$\alpha > 0$3	(Pooseh et al., 2012)

6. Extensions: Generalized Discrete Fractional Operators and Exterior Calculus

Recent advancements include the abstraction of discrete fractional operators via convolution pairs $\alpha > 0$4, enabling the unification of standard, Caputo, and Riemann–Liouville forms and their semigroup properties in a single framework (Ferreira, 2021). Duality relations and the generalized fundamental theorem of calculus are proved for wide classes of convolution kernel pairs.

The integration of discrete exterior calculus (DEC) and fractional calculus has led to the definition of Caputo-like and structure-preserving discrete exterior derivatives, applicable to nonlocal and anomalously diffusive PDEs on arbitrary meshes. The key construction uses weighted matrices depending on simplex distances, providing first- or second-order accurate approximations to continuous fractional derivatives, and preserving algebraic (exactness) relations such as $\alpha > 0$5 (Jacobson et al., 2022, Crum et al., 2019).

Fractional “diamond” operators—convex combinations of delta and nabla sums—provide parametric families interpolating between forward and backward fractional calculus, with full inheritance of semigroup, linearity, and product rules, useful for applications requiring flexible memory modeling (Bastos et al., 2010).

7. Applications, Examples, and Theoretical Implications

Discrete fractional calculus is employed in modeling long-memory discrete-time systems, variational optimization on lattices, nonlocal discrete PDEs, signal and image processing, anomalous transport, and physical models such as fractional radial Schrödinger equations (Ozturk, 2018). Worked examples confirm that as the fractional order becomes integer, discrete fractional solutions reduce to their classical discrete-time analogues, and as the grid step $\alpha > 0$6, discretizations converge to the continuous-time fractional calculus (Bastos et al., 2010, Bastos et al., 2010, Bastos, 2012). The explicit algebraic and convolution structures support the derivation of hypergeometric identities, such as Saalschutz-type summations, and form a discrete basis for applications in $\alpha > 0$7-calculus and special function theory (Ferreira, 2020).

A plausible implication is that continuing advances in the algebraic and computational apparatus—matrix representations, general convolution frameworks, and structure-preserving discretizations—will further expand the scope of discrete fractional calculus in both theoretical analysis and numerical simulation of nonlocal, memory-dependent discrete dynamical systems.