next up previous contents
Next: 4.8 Numerical Simulations of Up: 4. Geometric integration of Previous: 4.6 Accuracy tests for   Contents

4.7 Finite differences spatial discretization of LLG equation

Up to this point the considerations we made about the properties and the implementation of mid-point rule were rather independent from the spatial discretization used. In the following, to test the method we have chosen a specific technique based on finite-differences method. The magnetic body is subdivided into a collection of rectangular prisms with edges parallel to the coordinate axes. The edge lengths are $ d_x,d_y,d_z$. In this framework, it is convenient to identify each cell by three indices $ i,j,k$ instead of using the index $ l$ as we did before. The magnetization $ \textbf{{m}}_{i,j,k}$ is assumed to be uniform within the generic $ (i,j,k)$ cell. With this notation, the effective field in the generic $ (i,j,k)$ cell can be expressed as

$\displaystyle \textbf{h}_$eff;i,j,k$\displaystyle =\textbf{h}_$ex;i,j,k$\displaystyle +\textbf{h}_$m;i,j,k$\displaystyle +\textbf{h}_$an;i,j,k$\displaystyle +\textbf{h}_$a;i,j,k (4.57)

The exchange field (4.8) is computed by means of a 7-point laplacian discretization, which is second order accurate in space. In the generic ``internal'' cell $ (i,j,k)$, the exchange field can be expressed as follows:

$\displaystyle \textbf{h}_$ex;i,j,k$\displaystyle =\frac{2A}{\mu_0
 M_s^2}\bigg[\frac{\textbf{{m}}_{i+1,j,k}+\textb...
...-1,j,k}}{d_y^2}+\frac{\textbf{{m}}_{i,j+1,k}
 +\textbf{{m}}_{i,j-1,k}}{d_x^2} +$    
$\displaystyle +\frac{\textbf{{m}}_{i,j,k+1}+\textbf{{m}}_{i,j,k-1}}{d_z^2}-\lef...
...ac{2}{d_y^2}+\frac{2}{d_x^2}+
 \frac{2}{d_z^2}\right)\textbf{{m}}_{i,j,k}\bigg]$ (4.58)

A similar expression holds for the boundary cells where the Neumann boundary condition (4.6) has to be taken into account. Since the exchange interaction is a first-neighbors interaction, one can easily observe that the matrix $ {\underline{\textrm{C}}}_$ex is a block-diagonal matrix. As seen in section 3.1, the magnetostatic field (4.9) can be expressed as discrete convolution:

$\displaystyle \textbf{h}_$m;i,j,k$\displaystyle =\sum_{i'\neq i}\sum_{j'\neq j}\sum_{k'\neq k}
 N_{i-i',j-j',k-k'}\cdot\textbf{{m}}_{i',j',k'}  d_x d_y d_z$ (4.59)

where $ N_{i-i',j-j',k-k'}$ is the $ 3\times3$ block of the magnetostatic interaction matrix $ {\underline{\textrm{C}}}_$m which describes the magnetostatic interaction between the cells $ i,j,k$ and $ i',j',k'$. The discrete convolution (4.59) is computed by means of 3D Fast Fourier Transform (FFT) using the zero-padding algorithm [64]. The kernel of the convolution is obtained in exact analytical form by means of a slight generalization of the one proposed in Ref. [93] for cubic cells. As far as anisotropy is concerned, we assume that the body has uniaxial anisotropy defined by the anisotropy constant $ K_1$. Therefore, the anisotropy field is

$\displaystyle \textbf{h}_$an;i,j,k$\displaystyle =\frac{2K_1}{\mu_0M_s^2}
  (\textbf{{m}}_{i,j,k}\cdot\mathbf{e}_x)\mathbf{e}_x \quad.$ (4.60)

and the matrix $ {\underline{\textrm{C}}}_$an is a diagonal matrix.
Figure: (Left) Thin-film geometry for $ \mu$-mag standard problem n. 4. (Right) Initial equilibrium S-state.
\begin{figure}
\begin{center}
\epsfig{figure=mumag4_geom.eps,width=6.6cm}
\epsfig{figure=s_state.eps,width=6.6cm}
\end{center}
\end{figure}

next up previous contents
Next: 4.8 Numerical Simulations of Up: 4. Geometric integration of Previous: 4.6 Accuracy tests for   Contents
Massimiliano d'Aquino 2005-11-26