The two dimensional advection equation#

  1. Consider the 2d advection of field \(F\) under the velocity field \(\vec{V}\):

\[ \frac{\partial F}{\partial t} + \vec{V}\cdot \nabla F = 0, \quad \vec{V}=u\vec{i}+v\vec{j}. \]

Let \((x_m,y_l)=(m\Delta x, l\Delta y)\) and \(t^n=n\Delta t\). Applying central difference formulas to the time and spatial derivatives of \(F\), we obtain:

(63)#\[\begin{split} \frac{F_{m,l}^{n+1}-F_{m,l}^{n-1}}{2\Delta t}=-u\left( \frac{F_{m+1,l}^{n}-F_{m-1,l}^{n}}{2\Delta x}\right) - v\left( \frac{F_{m,l+1}^{n}-F_{m,l-1}^{n}}{2\Delta y}\right) \Leftrightarrow \\ F_{m,l}^{n+1}-F_{m,l}^{n-1}=-u\frac{\Delta t}{\Delta x}\left( F_{m+1,l}^{n}-F_{m-1,l}^{n}\right) - v\frac{\Delta t}{\Delta y}\left( F_{m,l+1}^{n}-F_{m,l-1}^{n}\right). \end{split}\]

We shall apply Von Neumann stability analysis to (63), by substituting a solution of the type:

\[ F_{m,l}^n=B^ne^{k_xm\Delta x+k_yl\Delta y}, \]

where \(k_x\) and \(k_y\) are the wavenumbers in \(x\) and \(y\). After some manipulation, we have

(64)#\[ B^{n+1}=B^{n-1}-2i\Delta t\left( u\frac{\sin k_x\Delta x}{\Delta x}+v\frac{\sin k_y\Delta y}{\Delta y}\right)B^n. \]

Defining \(D^n=B^{n-1}\), we can write (64) as matrix recurrence relationship:

(65)#\[\begin{split} \begin{bmatrix} B^{n+1}\\ D^{n+1} \end{bmatrix} = \begin{bmatrix} -2i\Delta tr & 1\\ 1 & 0 \end{bmatrix} \begin{bmatrix} B^{n}\\ D^{n} \end{bmatrix}, \end{split}\]

where \(r=u\frac{\sin k_x\Delta x}{\Delta x}+v\frac{\sin k_y\Delta y}{\Delta y}\). The eigenvalues of the recurrence matrix of (65) are given by

\[ \lambda_\pm = -i\Delta tr \pm \sqrt{1-(\Delta tr)^2}. \]

For the scheme to be stable, we must have \(|\lambda_\pm| \leq 1\), which is true if \(\Delta tr \leq 1\), or

\[ \Delta t \left| u\frac{\sin k_x\Delta x}{\Delta x}+v\frac{\sin k_y\Delta y}{\Delta y}\right| \leq 1. \]

Let \(u=V\cos\theta\) and \(v=V\sin\theta\) and write

\[ \Delta t V\left| \cos\theta\frac{\sin k_x\Delta x}{\Delta x}+\sin\theta\frac{\sin k_y\Delta y}{\Delta y}\right| \leq 1. \]

For \(\Delta x=\Delta y=d\), we shall have

\[ \frac{\Delta t V}{d}\left| \cos\theta\sin k_x\Delta x+\sin\theta\sin k_y\Delta y\right| \leq 1. \]

When \(\sin k_x\Delta x=1\) and \(\sin k_y\Delta y=1\),

\[ \frac{\Delta t V}{s}\left| \cos\theta+\sin\theta\right| \leq 1, \]

which is maximum when \(\theta=\pi/4\) and \((\cos\theta+\sin\theta)_{\theta=\pi/4}=\sqrt{2}\), resulting in

\[ \sqrt{2}\frac{\Delta t V}{s} \leq 1 \Leftrightarrow \frac{\Delta t V}{s} \leq \frac{1}{\sqrt{2}}=0.707\dots. \]

So we find that for 2d advection there is a reduction in the stability limit, which has to be 30% smaller than the limit in the 1d case.

To understand how this reduction comes about, let us consider a wave moving at a \(\pi/4\) angle to the \(x\) axis (Fig. 10).

../_images/advection-2d-stability.png

Fig. 10 Wave moving at a \(\pi/4\) angle to the \(x\) axis.#

The minimum stability is associated with a wave of wavelength \(2d\) in both the \(x\) and \(y\) directions. The absolute wavenumber of this wave is

\[ k_{2D}=\sqrt{k_x^2+k_y^2}=\sqrt{2\left(\frac{\pi}{d}\right)^2}=\sqrt{2}\frac{\pi}{d}=\sqrt{2}k_{1D}, \]

where \(k_{1D}=\pi/d\) is the wavelength of the shortest resolvable wave in a 1-d grid of size \(d\).

Since \(k_{2D}\) is larger than \(k_{1D}\) by a factor of \(\sqrt{2}\), the effectice wavelength in 2-d will be shorter than the wavelength in 1-d by an equal factor, and therefore the time must be reduced by that factor to fullfill the stability criteria.