A New Fractional Step Structure Preserving Method for The Landau-Lifshitz-Gilbert Equation

1. Problem Statement

The paper addresses the numerical simulation of the Landau-Lifshitz-Gilbert (LLG) equation, which governs the dynamics of magnetization in ferromagnetic materials. The equation combines two physical phenomena:

• Gyromagnetic precession: Conserves energy and magnetization magnitude.
• Dissipative relaxation: Models energy dissipation into the lattice.

Key Challenges:

• Nonlinearity and Norm Constraint: The magnetization vector $\mathbf{m}$ must satisfy the pointwise constraint $|\mathbf{m}| = 1$ at all times.
• Stiffness and Stability: Standard implicit methods (like Crank-Nicolson) offer excellent stability and energy conservation but result in nonlinear systems that are computationally expensive and difficult to solve.
• Explicit Limitations: Explicit methods are linear but suffer from strict CFL (Courant-Friedrichs-Lewy) stability conditions, requiring very small time steps.
• Goal: Develop a numerical scheme that is linear, unconditionally stable, structure-preserving (specifically preserving the magnetization norm), and accurate.

2. Methodology

The authors propose a Structure-Preserving Fractional Step Method based on a Crank-Nicolson type formulation combined with an implicit Gauss-Seidel fractional iteration.

A. Governing Equations

The dimensionless LLG equation is given by:
$$\mathbf{m}_t = -\mathbf{m} \times \Delta \mathbf{m} - \alpha \mathbf{m} \times (\mathbf{m} \times \Delta \mathbf{m})$$
subject to homogeneous Neumann boundary conditions and the constraint $|\mathbf{m}| = 1$.

B. The Proposed Scheme

The method avoids solving a fully nonlinear system by decoupling the components of the magnetization vector through a three-step fractional iteration (Gauss-Seidel type).

1. Semi-Implicit Formulation: The scheme uses an auxiliary variable $\mathbf{g}_h = (I - \Delta t \Delta_h)^{-1} \mathbf{m}_h$ to handle the Laplacian term implicitly while maintaining linearity.
2. Fractional Steps: For the undamped case ($\alpha=0$), the update from time $n$ to $n+1$ is split into three linear sub-steps:
   • Step 1: Solve for the first component update using known values from step $n$.
   • Step 2: Solve for the second component, utilizing the updated first component from Step 1.
   • Step 3: Solve for the third component, utilizing the updated first and second components.
3. Damping Term: The method is extended to include the damping term ($\alpha > 0$) by modifying the right-hand side of the linear systems to include cross-product terms involving the damping coefficient, while maintaining the fractional step structure.

C. Theoretical Properties

• Linearity: Unlike the standard implicit Crank-Nicolson method which requires solving a nonlinear system, the proposed method solves a sequence of linear systems.
• Norm Preservation: The scheme is proven to preserve the magnitude of the magnetization vector ($\|\mathbf{m}^{n+1}\|_2 = \|\mathbf{m}^n\|_2$) exactly, without requiring a projection step.
• Stability: Spectral analysis of the coefficient matrices in the three steps demonstrates that the eigenvalues of the inverse matrices have magnitudes $\le 1$. This implies the method is unconditionally stable.
• Accuracy:
  • Time: First-order accuracy ($O(\Delta t)$).
  • Space: Second-order accuracy ($O(h^2)$).

3. Key Contributions

1. Linearization of Implicit Schemes: The paper successfully transforms the typically nonlinear implicit Crank-Nicolson scheme for LLG into a linear algorithm using a fractional step Gauss-Seidel approach.
2. Structure Preservation: The method inherently preserves the geometric constraint $|\mathbf{m}|=1$ (length preservation) and energy conservation properties (for the undamped case) without post-processing projections.
3. Unconditional Stability: Theoretical analysis proves the scheme is unconditionally stable, overcoming the restrictive time-step limitations of explicit methods.
4. Efficiency: By avoiding nonlinear solvers (like Newton-Raphson) and projection steps, the method reduces computational complexity per time step while maintaining high stability.

4. Numerical Results

The authors validated the method through extensive numerical experiments in 1D and 3D settings.

• Accuracy Verification:
  • 1D Tests: Using an exact analytical solution, the method demonstrated first-order convergence in time and second-order convergence in space.
  • 3D Tests: Similar convergence rates were observed, confirming the method's scalability to higher dimensions.
• Norm Preservation:
  • Tests showed that the deviation from the unit norm ($\|\|\mathbf{m}_h\|_2 - 1\|_\infty$) remained at the level of machine precision ($\approx 10^{-15}$), confirming exact norm preservation.
• Robustness: The method was tested with various initial conditions (smooth and complex profiles) and damping coefficients ($\alpha = 0.01$), consistently showing stable and accurate numerical profiles.
• Stability: The method remained stable even with large time steps where explicit methods would fail, verifying the unconditional stability claim.

5. Significance and Future Work

• Theoretical Value: This work bridges the gap between the stability of implicit methods and the computational efficiency of linear schemes for micromagnetic simulations.
• Practical Application: The method offers a robust tool for simulating complex magnetic dynamics where long-time integration and strict adherence to physical constraints (norm preservation) are critical.
• Future Directions:
  • Extending the analysis to Finite Element Methods (FEM) for spatial discretization.
  • Applying the method to more complex micromagnetic problems.
  • Further rigorous proofs of convergence and stability for the damped case with the normalizing step.

In conclusion, Changjian Xie presents a novel, efficient, and theoretically sound numerical method that overcomes the nonlinearity bottleneck of traditional implicit schemes for the LLG equation while strictly preserving the physical structure of the magnetization vector.