Time Fractional Subdiffusion Equation

• Time fractional subdiffusion equation is a partial differential equation using fractional time derivatives (0 < α < 1) to model anomalously slow transport in complex systems.
• Analytical and numerical approaches, including Caputo and Riemann–Liouville formulations and integral-balance techniques, yield accurate self-similar solutions and computational simulations.
• Extensions involving variable kernels, reactive terms, and time-dependent domains enable modeling of heterogeneous media and enhance approaches to inverse problems.

A time fractional subdiffusion equation is a partial differential equation incorporating a time-fractional derivative—usually of Caputo or Riemann–Liouville type—with order $0 < \alpha < 1$, to model diffusion processes in which the mean waiting time for particle jumps diverges, resulting in sublinear mean square displacement. This framework generalizes classical Fickian diffusion (Brownian motion) to account for anomalous transport mechanisms frequently encountered in disordered and complex systems. The fundamental model is

$$D_t^\alpha\,C(x,t) = D\,\frac{\partial^2 C}{\partial x^2},$$

where $D_t^\alpha$ denotes the fractional time derivative and $D$ is the generalized subdiffusion coefficient. The memory kernel in the fractional derivative encodes system-wide temporal nonlocality.

1. Mathematical Formulation and Fractional Derivatives

The principal operators used in time fractional subdiffusion equations are the Riemann–Liouville and Caputo fractional derivatives. For $0 < \alpha < 1$, these are defined as:

• Riemann–Liouville:

$$D_t^\alpha f(t) = \frac{1}{\Gamma(1-\alpha)}\frac{d}{dt}\int_0^t (t-\tau)^{-\alpha} f(\tau)\,d\tau.$$

• Caputo:

$${}^C D_t^\alpha f(t) = \frac{1}{\Gamma(1-\alpha)}\int_0^t (t-\tau)^{-\alpha} f'(\tau)d\tau.$$

In the context of subdiffusion, both formulations yield nonlocal-in-time, power-law memory, but they differ in initial condition prescription and technical nuances in boundary-value problems (Hristov, 2010, Hristov, 2011).

Boundary and initial conditions are selected to reflect the physical process—fixed values, flux, or mixed types—often adapting classical Dirichlet, Neumann, or Robin forms to the fractional context.

2. Integral-Balance and Self-Similar Solutions

Integral-balance techniques provide explicit, self-similar approximations to fractional subdiffusion problems. The methodology proceeds as follows (Hristov, 2010):

1. Weak Power-Law Profile: Assume the solution within the propagation/penetration layer is

   $$C_a(x, t) = C_\infty + (C_s - C_\infty)\left(1 - \frac{x}{\delta(t)}\right)^n,$$

where $\delta(t)$ is the (unknown) penetration depth and $n$ an exponent determined by error minimization.

1. Integral-Balance Condition: Integrate the governing PDE over $[0, \delta(t)]$ and enforce endpoint and average constraints.
2. Penetration Depth: The resulting ODE for $\delta(t)$ gives the self-similar law

   $\delta(t) \sim t^{\mu/2},$

for order $0 < \mu < 1$ subdiffusion.

1. Profile Optimization: Local minimization of residuals $E(n)$ yields $n \approx 1$ for strong subdiffusion ($\mu \lesssim 0.3$), with larger values (1.2–1.5) preferred for moderate $\mu$.

Numerical comparisons with exact solutions expressed in terms of Wright or Mittag–Leffler functions show that the integral-balance method achieves accuracy within a few percent for most of the penetration layer—making it valuable for rapid parameter estimates in engineering scenarios.

3. Fundamental Solutions and Green Functions

The Green function for the time fractional subdiffusion equation, with initial datum $\delta(x-x_0)$, is typically obtained via Laplace and Fourier transforms: $$\mathcal{L}[C(x,t)](s) = \frac{1}{2\sqrt{D}\,s^{\alpha/2-1}} \exp\left(-\frac{|x-x_0|}{\sqrt{D}\,s^{\alpha/2}}\right).$$ The inverse transform yields representations involving special functions, notably the Wright function: $$C(x,t) = \frac{1}{2\sqrt{D}\;f_{-1+\alpha/2,\,\alpha/2}(t;\,|x-x_0|/\sqrt{D})},$$ where $f_{\nu, \mu}$ is a Mainardi/Wright function (Kosztołowicz et al., 2021, Kosztołowicz et al., 2022).

The mean square displacement (MSD) grows as $\langle x^2(t) \rangle \sim t^\alpha$, demarcating subdiffusive behavior [$\alpha < 1$ implies slower diffusion than Fickian].

4. Extensions: Variable Kernels, Transient Regimes, and Heterogeneous Media

The basic equation is extended in several directions:

• $g$-Subdiffusion: Replaces the fractional time derivative kernel $(t-\tau)^{-\alpha}$ by $[g(t)-g(\tau)]^{-\alpha}$, capturing evolving or multi-regime anomalous transport. With $g(t) \sim t$ at early times and $g(t) \sim t^{\beta/\alpha}$ at late times, one obtains crossovers from $\alpha$- to $\beta$-type subdiffusion (Kosztołowicz et al., 2022, Kosztołowicz, 2022, Kosztołowicz et al., 2021).
• Spatially Evolving and Growing Domains: Incorporation of growth functions to describe subdiffusion on time-dependent domains introduces comoving derivatives and new history kernels, constructed via pullbacks along Lagrangian trajectories (Angstmann et al., 2017).
• Immobilization and Trapping: Including a probability of permanent immobilization at each site modifies the kernel and leads, at long times, to a stationary exponential spatial profile (Kosztołowicz, 2023).
• Reactive and Nonlocal/Nonlinear Terms: Problems with nonlinear nonlocal initial data, reaction terms, or fractional Fokker–Planck operators admit analysis via Green-function methods and Banach fixed-point contraction techniques, under appropriate Lipschitz and kernel bounds (Ashurov et al., 24 Jun 2025, Kay et al., 2023).

5. Stochastic Representations and Numerical Algorithms

Time fractional subdiffusion admits a natural stochastic representation:

• CTRW/Inverse Subordinator: Solutions are distributions of Markov processes time-changed by the inverse of a (possibly stable or infinitely divisible) subordinator with Laplace exponent dictating the memory kernel. This captures the nonlocality in time as arising from random operational time (Magdziarz et al., 2015, Lawley, 2020, Nichols et al., 2017).
• Monte Carlo Simulation: Discrete-time random walks with Sibuya power-law waiting times provide efficient, pathwise algorithms for approximating fractional subdiffusion processes. The key advantage is the expected computational cost per sample path grows as $n^\alpha$, significantly faster than standard finite-difference methods for small $\alpha$ (Nichols et al., 2017).
• Galerkin and Spectral Methods: Fractional Grönwall inequalities undergird stability and convergence of spectral discretization for both linear and nonlinear subdiffusion (Yang et al., 2019).

6. Analytical Results, Decay Rates, and Critical Phenomena

In $\mathbb{R}^d$, decay rates for the Cauchy problem are governed by the order $\alpha$ and spatial dimension $d$:

• The fundamental solution decays as $t^{-\alpha d/2}$ in $L^r$ norms, with critical dimension phenomena: above a threshold, the decay saturates at $O(t^{-\alpha})$ (Kemppainen et al., 2014).
• Pointwise kernel estimates distinguish ‘inner’ ($|x|^2 \le t^\alpha$) and ‘outer’ ($|x|^2 \gg t^\alpha$) regimes, with transition regions governed by subordination from a Gaussian kernel.
• Energy methods and Fourier-multiplier approaches yield optimal and robust decay estimates, respectively, with physical consequences for long-time relaxation and asymptotic profile sharpness.

7. Inverse Problems, Uniqueness, and Well-Posedness

Identification of the fractional order or coefficients in subdiffusion is possible using overposed data (e.g., observation of certain functionals at fixed time). Representing the solution in eigenfunction-Mittag–Leffler series, one can prove uniqueness and develop efficient numerical schemes for order recovery via nonlinear equation inversion (Ashurov et al., 2020, Ashurov et al., 7 Nov 2025).

Backward (in time) problems for the subdiffusion equation are severely ill-posed in all classical norms, but conditional stability results can be derived in strong (Sobolev-type) spaces for sufficiently regular target and source data (Alimov et al., 2021).

---

The time fractional subdiffusion equation, in both its classical and generalized forms, provides a rigorous and flexible mathematical structure for modeling a broad spectrum of anomalously slow transport phenomena, connecting deterministic and stochastic viewpoints, and supporting both analytical and computational inquiry across physics, engineering, and applied mathematics (Hristov, 2010, Hristov, 2011, Kosztołowicz et al., 2022, Nichols et al., 2017, Kemppainen et al., 2014).