---
title: Crank-Nicolson Approach in Fractional PDEs
url: https://www.emergentmind.com/topics/crank-nicolson-approach
type: topic
---

# Crank-Nicolson Approach in Fractional PDEs

The Crank-Nicolson Approach is a family of implicit, time-centered numerical schemes renowned for their stability and accuracy in solving diverse classes of differential and integro-differential equations, especially those involving fractional derivatives, dispersive and advective terms, and stiff or degenerate parabolic dynamics. Originating as a second-order method for parabolic PDEs, the Crank-Nicolson methodology now encapsulates a wide spectrum of extensions—including multi-step, high-order, and fractional-derivative variants—engineered for demanding applications in transport, anomalous diffusion, and memory-dominated physical models. Its appeal arises from unconditional stability, controlled accuracy, and flexibility in spatial and temporal discretization, particularly in regimes where explicit schemes are hampered by stringent CFL-type restrictions or inefficient low-order convergence.

## 1. Theoretical Foundation and Discrete Formulation

At its core, the Crank-Nicolson approach centers on a trapezoidal, time-centered discretization of differential equations of the form
\[
u_t = \mathcal{L}u + f,
\]
producing the canonical time-stepping stencil
\[
\frac{U^{n+1} - U^n}{k} = \frac12 \left( \mathcal{L}U^{n+1} + \mathcal{L}U^n \right) + \frac12\left(f^{n+1} + f^n\right),
\]
where \( U^n \) approximates \( u(t_n) \) and \( k \) is the time step. In the classical setting, this yields second-order accuracy in time and, with proper spatial discretization, high-order accuracy in space.

For equations involving memory or non-local fractional time derivatives, such as the time-variable fractional mobile-immobile advection-dispersion equation
\[
u_t(x,t) = -c\,D_t^\beta u + u_{xx} - u_x + f(x,t), \quad 0 < \beta(x,t) < 1,
\]
the Crank-Nicolson framework is nontrivial. Here, repeated convolution-type sums and shifted quadrature rules are invoked to approximate nonlocal terms, with precise weights and remainder bounds required for stability and consistency. In the approach of [2205.05077], the temporal grid is enriched by half-steps and predictor-corrector sublevels to align the discrete scheme with the internal structure of the Caputo derivative and to facilitate optimal error propagation.

## 2. Fourth-Order Two-Step Explicit Euler/Crank-Nicolson Hybrid Scheme

The scheme analyzed in [2205.05077] achieves higher spatial accuracy via fourth-order finite-difference stencils:
\[
u_x(x_j) \approx \delta_x^4 u_j = \frac{-u_{j+2} + 8u_{j+1} - 8u_{j-1} + u_{j-2}}{12h},
\]
\[
u_{xx}(x_j) \approx \delta_{2x}^4 u_j = \frac{-u_{j+2} + 16u_{j+1} - 30u_j + 16u_{j-1} - u_{j-2}}{12h^2},
\]
paired with a two-step, time-shifted Euler/Crank-Nicolson hybridization:
- **Predictor**: an explicit-Euler step employs L1-type convolution quadrature for the Caputo derivative and spatial fourth-order discretization, solving a pentadiagonal linear system at the fractional time level \( t_{n+\frac12+\alpha} \).
- **Corrector**: a Crank-Nicolson-type step at \( t_{n+1+\alpha} \) further balances temporal errors and refines the solution by including additional weighted fractional-derivative terms and quadrature remainders.

Notationally, primary unknowns \( U_j^n \approx u(x_j,t_n) \) propagate on grids \( x_j = L_0 + jh \), \( t_n = nk \), and the Caputo derivative is discretized at fractional time levels using precomputed, decaying, strictly positive weights \( a_{n+\frac12,\ell+\frac12}^{\alpha,\beta} \).

## 3. Stability and Error Analysis

The main theoretical result is unconditional stability and optimal error bounds in the strong \( L^\infty(0,T;L^2) \) norm, even for non-smooth fractional solutions. The crucial analytic components are:
- **Quadrature Control**: The remainder in the fractional-derivative quadrature is provably \( O(k^{2-\beta}) \) (see Lemma 1 of [2205.05077]), with the weights \( a_{n+\frac12,\ell+\frac12}^{\alpha,\beta} \) shown to be positive and strongly decaying (Lemma 2).
- **Discrete Fractional-Energy Argument**: By employing telescoping in time and "fractional-energy" inequalities,
  \[
  (u^{n+s}, \Delta_t^{\beta} u^{n+s+\alpha}) \geq \frac12 \Delta_t^\beta ( (u^{n+s+\alpha})^2 ),
  \]
  all history terms due to the Caputo derivative are controlled.
- **Operator Dissipativity**: The fourth-order spatial operator
  \[
  L_h = \delta_{2x}^4 - \delta_x^4
  \]
  is dissipative, with
  \[
  (L_h u, u) \leq 0, \qquad (-L_h u, u)\geq c_1 \|\delta_x u\|_2^2,
  \]
  enforcing energy decay at the discrete level.
- **Convergence**: For \( 0 < \beta < 2/3 \), \( 0 < \alpha < 1/2 \), one obtains
  \[
  \max_{0\leq n\leq N-1} \|U^{n+\frac12}\|_2, \ \max_{0\leq n\leq N} \|U^n\|_2 \leq \|u\|_{L^\infty(0,T;L^2)} + C(k + k^{2-\beta} + k^2 + h^4)
  \]
  and—using discrete Grönwall arguments—
  \[
  \max_n \|e^n\|_2 + \max_n \|e^{n+\frac12}\|_2 \leq C\,(k + h^4)
  \]
  where \( e^n = U^n - u(x_j, t_n) \). Thus, the method achieves first-order accuracy in time and fourth-order in space, with unconditional stability ensured by the dissipativity and energy telescoping.

## 4. Computational Implementation and Practical Considerations

Algorithmic features include:
- **Linear Algebra**: All implicit solves are for pentadiagonal, nonsymmetric systems arising from the high-order spatial discretization. These are efficiently handled by direct band solvers or Krylov subspace methods (e.g., GMRES) with band-circulant preconditioners.
- **Fractional Weights**: The a priori computation of all required weights \( a_{n+\frac12,\ell+\frac12}^{\alpha,\beta} \) is recommended; these can be reused across time steps due to the time-invariance for constant \( \beta \).
- **No CFL Constraint**: The unconditional stability of the method obviates any explicit time-step restriction; for optimal accuracy, it is natural to choose \( k \sim h^4 \).
- **Boundary Conditions**: Dirichlet conditions are enforced directly; ghost values may be set to prescribed boundary data without affecting the solvability or stability of the pentadiagonal systems.

The step-by-step procedure involves initialization, computation of a shifted start value \( U^\alpha \), predictor-corrector sweeps at each time level, and direct enforcement of boundary values for all temporal sub-levels.

## 5. Comparison with Classical Crank-Nicolson and Other Schemes

Relative to standard integer-order Crank-Nicolson (CN) methods:
- **Classical CN**: Second-order in time and usually second-order in space unless augmented with compact higher-order stencils. Unconditional stability is standard but proven only for linear, constant-coefficient problems.
- **Modified Two-Step CN**: The present approach features a hybrid explicit Euler (L1-type) predictor and a CN-type corrector, shifted by an adjustable parameter \( \alpha \). This adaptation is necessary for fractional time derivatives, which induce nonlocal temporal memory, and achieves first-order temporal accuracy (optimal for fractional dynamics with limited time regularity) and fourth-order spatial accuracy, with unconditional stability and pentadiagonal algebraic structure. It avoids step-size constraints that often accompany explicit/singly-implicit handling of the fractional term.

A comparative summary:

| Feature             | Classical CN (PDE)      | Modified Euler/CN (Fractional)      |
| ------------------- | ---------------------- | ----------------------------------- |
| Time Accuracy       | Second-order           | First-order (optimal for fractional)|
| Space Accuracy      | Second-order           | Fourth-order                        |
| Stability           | Unconditional (linear) | Unconditional (fractional, any k)   |
| Solve Structure     | Tridiagonal            | Pentadiagonal                       |
| Fractional Support  | Not directly           | Yes (Caputo, variable β)            |

## 6. Extensions and Generalizations

The explicit–implicit predictor-corrector and the energy-based unconditional-stability framework are extensible to broader classes of nonlocal, subdiffusive, or multi-term fractional PDEs. The technical machinery—discrete fractional-convolution quadrature, telescopic energy inequalities, high-order spatial discretization, and two-step start-up—carries over to systems with spatially and temporally variable fractional order, nonlinear memory, and stiff multi-physics couplings. For further details regarding the development and application of these high-order fractional schemes, and precise stability arguments, refer to the full theoretical and numerical discussion in [2205.05077].

Source: https://www.emergentmind.com/topics/crank-nicolson-approach