A linear mass-lumped finite element method for the Landau-Lifshitz-Gilbert equation: unconditional energy dissipation and length preservation

• The LLG equation models magnetization dynamics in ferromagnetic materials as a time-dependent nonlinear equation.Source: §1 · p. 1
• Standard finite element methods typically fail to preserve the exact non-convex unit-length constraint |m| = 1.Source: §1 · p. 2
• Renormalization enforces unit length by normalizing intermediate magnetization, but rigorous error analysis remains challenging.Source: §1 · p. 2
• Existing normalization methods are not known to be unconditionally energy stable, while some alternatives require modified energy laws or new finite element spaces at each time step.Source: §1 · p. 2
• The paper proposes a fully discrete linear finite element scheme combining interpolation, mass lumping, and renormalization to enforce |m| = 1 and unconditional original energy dissipation.Source: §1 · p. 3
• Its hybrid framework combines finite element structure with finite-difference nodewise properties for scheme construction and error analysis.Source: §1 · pp. 2–3

Notation and mesh assumptions

• The paper fixes Sobolev-space notation, time-step notation, quasi-uniform triangular partitions, and primarily periodic or homogeneous Neumann boundary conditions.The mesh consists mainly of adjacent triangle pairs forming approximate parallelograms.Source: §2 · pp. 3–4

Mass-lumped finite elements

• The mass-lumped finite element setting uses piecewise linear functions, vertex-based Lagrange interpolation, and a discrete inner product defined through nodal values.The section also records an estimate for the discrepancy between lumped and standard inner products.Source: §2 · p. 4

Grid construction

• On rectangular domains, right-triangular meshes generated from quasi-uniform grids support the finite-difference operators, with general triangulations handled under corresponding mesh conditions.The extension to general domains is described as natural under the stated conditions.Source: §2 · p. 4

Grid operators

• The computational grid, grid-function space, discrete gradient, and componentwise matrix operations are introduced for scalar and vector grid functions.The discrete gradient inner product combines corresponding matrix entries through multiplication and summation.Source: §2 · pp. 4–5

Analytical tools

• Interpolation and inverse inequalities, matched central interpolation identities, and a discrete Gronwall lemma provide analytical tools for subsequent estimates.The stated inequalities apply under dimension- and exponent-dependent conditions, while the constants in the inverse estimate depend only on the domain.Source: §2 · p. 5

Stability and preservation

• The constructed scheme preserves the node-wise unit-length constraint and the original energy dissipation law.The analysis introduces a weak Laplacian and derives the energy law from the scheme identities and a node-wise estimate.Source: §3 · pp. 5–7

Scheme construction

• The scheme combines time projection with mass-lumped finite element spatial discretization to construct a fully discrete method.The projection includes a node-wise normalization operation that is simple to implement.Source: §3 · pp. 5–6
• The method is connected to the original LLG equation through mass-lumped integration by parts and vector identities at mesh nodes.The resulting reformulation uses the approximation mt = −βm × ∆m −γm × (m × ∆m).Source: §3 · p. 6

Stability and preservation

• The scheme admits unconditional stability, including a uniform H1 bound for the numerical solution.The stability result also yields existence and uniqueness of the scheme solution.Source: §3 · p. 7

Mesh considerations

• The stability analysis extends to general triangular mesh partitions under a Delaunay-type condition.Under this condition, the stated inequality keeps the analysis valid on general unstructured triangular meshes.Source: §3 · p. 8

4. Theoretical analysis.

• The equivalent reformulation avoids the complicated nonlinear damping term and enables optimal-rate error analysis through linearized stability estimates.Summation by parts is used to derive the reformulation before the convergence argument.Source: §4 · p. 8
• Mass-lumping and interpolation operators express finite element inner products and nonlinear gradients through finite-difference approximations on triangular cells.The construction uses node values and constant cellwise gradients to obtain the required expansions.Source: §4.1 · pp. 8–9
• Nodewise identities and interpolation values are used to transform nonlinear inner-product terms into forms needed for the convergence proof.These identities support the discrete expansions and consistency estimates used later in the analysis.Source: §4.1 · p. 10
• The consistency analysis combines interpolation and nodewise finite-difference estimates, with super-convergence and auxiliary interpolation results supplying precise truncation-error bounds.The argument assumes a smooth exact solution and uses estimates for the Lagrange interpolation operator.Source: §4.2 · pp. 11–12
• Under the theorem’s regularity, smallness, and scaling assumptions, the fully discrete scheme satisfies an optimal error estimate with L2 error bounded by C0(h2+∆t).The proof closes through discrete Gronwall and comparison estimates between discrete and continuous norms.Source: §4.3 · pp. 13, 19

5.1. Convergence test.

• Second-order spatial and first-order temporal convergence are observed in the discrete L2_h norm, matching Theorem 4.5.The tests use Δt = h^2 for spatial convergence and Δt = h for temporal convergence.Source: §5.1 · p. 20
• The convergence tables on quasi-uniform grids satisfy the mesh condition hmax/hmin ≤ 3.This supports testing the method beyond uniform meshes.Source: §5.1 · p. 21, Table 4.4 · p. 21

5.2. Energy dissipation.

• The original energy decreases monotonically across different spatial mesh sizes, confirming dissipative behavior.The simulation uses time step Δt = 0.01.Source: §5.2 · p. 21

5.3. Phenomenon of blowup.

• In the blowup simulation, the magnetization preserves (0, 0, 1)^T at the origin while turning toward (0, 0, −1)^T nearby over time.Snapshots are taken at t = 0, 0.06, 0.15, 0.30, 0.32, and 0.35.Source: §5.3 · p. 22, Figure 4.3 · p. 22, Figure 4.4 · p. 23

5.2. Energy dissipation.

• Centrally refined quasi-uniform 128 × 128 meshes produce monotonic energy decay and match the uniform 256 × 256 reference considerably well.The comparison includes a standard uniform 128 × 128 mesh and the reference solution from [26].Source: §5.3 · pp. 22–23, Figure 4.2 · p. 22

6. Conclusion.

• The scheme enforces |m_h| = 1 at every finite element node and preserves unconditional original energy dissipation.The method is presented as a classical finite element scheme distinct from the tangent-plane finite element method.Source: §6 · p. 23
• Interpolation operators combine finite element and finite difference analysis to control the nonlinear term and derive optimal-order error estimates.The analysis uses an equivalent weak formulation and triangular meshes.Source: §6 · p. 23
• Developing higher-order finite element methods that retain both unit-length preservation and unconditional original energy dissipation remains a significant challenge.Source: §6 · p. 23