Caputo Derivative: Definition & Applications

• Caputo derivative is a fractional operator that generalizes classical differentiation to non-integer orders while ensuring the derivative of a constant is zero.
• It is pivotal in modeling systems with memory and hereditary properties, with applications in anomalous diffusion, viscoelasticity, and fractional variational calculus.
• Numerical schemes like the L1 method and Fast Implicit Double Recurrence (FIDR) effectively approximate the Caputo derivative, ensuring stability and convergence in fractional PDEs.

The Caputo derivative is a fractional calculus operator that extends the concept of integer-order differentiation to non-integer (fractional) orders, while preserving compatibility with classical initial and boundary value problems. It plays a fundamental role in the analysis and modeling of systems exhibiting memory and hereditary properties, particularly within the frameworks of anomalous transport, viscoelasticity, fractional diffusion, and fractional variational calculus.

1. Definition and Fundamental Formulation

For a function $f$ defined on the interval $[a, b]$ and sufficiently smooth (e.g., $f \in C^n([a,b])$), the left Caputo fractional derivative of order $\alpha$ ($0 < \alpha < 1$) is defined as

$${}^C_{a}D_{t}^\alpha f(t) = \frac{1}{\Gamma(1-\alpha)} \int_a^t (t-\tau)^{-\alpha} f'(\tau)\,d\tau,$$

where $\Gamma$ denotes the Gamma function. The right Caputo derivative of order $\beta$ is defined analogously: $${}^C_{t}D_{b}^\beta f(t) = -\frac{1}{\Gamma(1-\beta)} \int_t^b (\tau-t)^{-\beta} f'(\tau)\,d\tau.$$ These definitions generalize to arbitrary order $\alpha > 0$, with

$[a, b]$0

The Caputo derivative differs structurally from the Riemann–Liouville derivative in that the integer derivative appears inside the convolution kernel, yielding compatibility with standard (classical) initial conditions and ensuring that the fractional derivative of a constant vanishes (Almeida et al., 2011, Almeida et al., 2010).

2. Key Analytical Properties

• Linearity: For scalars $[a, b]$1 and appropriately smooth functions $[a, b]$2,

$[a, b]$3

• Vanishing on constants: $[a, b]$4 for constant $[a, b]$5.
• Classical limit and compatibility: As $[a, b]$6,

$[a, b]$7

For integer order $[a, b]$8, the Caputo derivative reduces to the ordinary $[a, b]$9th derivative.

• Relation to Riemann–Liouville derivative:

$f \in C^n([a,b])$0

• Fractional integration by parts: For $f \in C^n([a,b])$1, the Caputo operator satisfies

$f \in C^n([a,b])$2

where $f \in C^n([a,b])$3 is the right-sided Riemann–Liouville fractional integral (Almeida et al., 2010, Almeida et al., 2011, Malinowska et al., 2010).

3. Combined Caputo Derivative and Extensions

Malinowska & Torres introduced the combined Caputo derivative as a convex combination of the left and right Caputo operators: $f \in C^n([a,b])$4 for $f \in C^n([a,b])$5, $f \in C^n([a,b])$6. Special cases recover standard left ($f \in C^n([a,b])$7) and right ($f \in C^n([a,b])$8) Caputo derivatives, and the symmetric (Klimek) derivative ($f \in C^n([a,b])$9, $\alpha$0). This operator is linear in $\alpha$1, admits a suitable integration-by-parts formula, and generalizes variational principles built on the Caputo model (Malinowska et al., 2011, Malinowska et al., 2010).

4. Fractional Variational Calculus and Euler–Lagrange Equations

In the Caputo calculus, the fractional variational principle is formulated for functionals of the form

$\alpha$2

where $\alpha$3 may depend on left and right Caputo derivatives of $\alpha$4. The associated Euler–Lagrange equation is

$\alpha$5

with analogous forms in the combined Caputo framework (Malinowska et al., 2011, Malinowska et al., 2010, Almeida et al., 2011, Lazo et al., 2012).

The boundary conditions—so-called transversality conditions—require careful handling, as the presence of fractional derivatives naturally introduces nonlocal contributions, realized through fractional integrals of the Lagrangian derivatives at the interval endpoints (Malinowska et al., 2010, Malinowska et al., 2011).

For isoperimetric problems, additional constraints are incorporated via augmented Lagrangians, with Lagrange multipliers and complementary slackness conditions as in the classical theory (Malinowska et al., 2011, Malinowska et al., 2010, Almeida et al., 2010).

5. Numerical Discretization and Computational Algorithms

The Caputo derivative's nonlocal structure leads to quasi-memory effects in numerical implementations. The L1 scheme, the shifted Grünwald–Letnikov method, and sum-of-exponentials approximations are standard approaches to discretization. For a uniform mesh $\alpha$6, the L1 approximation reads

$\alpha$7

Recent results quantify truncation errors and stability for both fixed and small fractional orders $\alpha$8, and establish unconditional convergence for implicit schemes applied to fractional PDEs (Płociniczak, 2021, Zhang et al., 2021, Dimitrov et al., 2018, Li et al., 2019).

In time-fractional PDE solvers, efficient history evaluation, such as the Fast Implicit Double Recurrence (FIDR) algorithm, is essential to mitigate computational cost, especially in the small $\alpha$9 regime. Discrete convolution coefficients must be carefully analyzed for positivity and monotonicity to guarantee stability and equipartition of error (Zhang et al., 2021, Li et al., 2019).

6. Generalizations: Variable and Distributed Order Caputo Derivatives

Extensions to variable-order and distributed-order settings have been developed to address systems where the fractional order itself may vary in space, time, or be governed by a probability distribution. The variable order Caputo derivatives take the form

$0 < \alpha < 1$0

with specific expansions and error control for numerical schemes (Tavares et al., 2015).

The distributed-order Caputo derivative is formalized as

$0 < \alpha < 1$1

where $0 < \alpha < 1$2 is a nonnegative integrable weight, with existence and regularity results for corresponding fractional diffusion equations derived under minimal regularity of data and coefficients (Kubica et al., 2017).

7. Analytical and Applied Implications

The Caputo derivative provides an operator-theoretic and functional-analytic framework for nonlocal models, with spectral characterizations linked to fractional powers of Volterra and elliptic operators in appropriate Sobolev spaces (Gorenflo et al., 2014). It is suitable both for modeling physical processes exhibiting nonlocality or memory and for rigorous mathematical analysis, due to its compatibility with classical data and well-posedness structures.

Monotonicity results for Caputo derivatives extend classical criteria: monotonicity of a function on an interval is characterized by the sign of all its Caputo derivatives of order $0 < \alpha < 1$3 on that interval (Diethelm, 2015). The Caputo calculus also enables generalizations of the fundamental lemma of the calculus of variations and derivation of Euler–Lagrange equations involving only Caputo derivatives, thus eliminating the need for Riemann–Liouville terms in boundary conditions (Lazo et al., 2012).

In computational contexts, the Caputo fractional derivative informs the design of stable, efficient iterative optimization algorithms, such as Caputo-fractional gradient descent, which exhibit smoothing properties and close links to Tikhonov regularization in quadratic cases (Shin et al., 2021).

The Caputo derivative's analytical tractability, compatibility with physical initial and boundary data, and adaptability to numerical schemes ensure its centrality in the theory and applications of fractional differential equations and variational problems across mathematical physics, engineering, and applied analysis (Malinowska et al., 2011, Almeida et al., 2010, Malinowska et al., 2010, Almeida et al., 2011, Płociniczak, 2021, Gorenflo et al., 2014, Zhang et al., 2021, Diethelm, 2015, Kubica et al., 2017, Tavares et al., 2015, Lazo et al., 2012, Dimitrov et al., 2018, Li et al., 2019, Shchedrin et al., 2017).