# Turn vector equation into matrix multiplication

I have physical equations of motion that describe the dependence of one vector field on the components of another vector field. Without getting too much into detail, my differential equations involve a double curl that mixes the vector components on one side of the equation. Furthermore, I would like to discretize the space on which these equations are defined. So far I was able to set up the discretized equations:

$$\frac{d}{dt}\ \begin{pmatrix} \vdots\\ P_x(i)\\ P_y(i)\\ P_z(i)\\ \vdots \end{pmatrix} = \begin{pmatrix} \vdots\\ A_y(i)-A_y(i+1)\\ -A_x(i)+A_x(i+1)-a*A_z(i)\\ b*A_y(i)\\ \vdots \end{pmatrix}\tag{1}$$ where a and b are some arbitrary constants and $$\vec{A}$$ and $$\vec{P}$$ are the two vector fields whose relation I want to determine. The lattice coordinates in one dimension are written as i (e.g. $$\frac{d}{dt} P_x(7) = A_y(7)-A_y(8)$$).

Is there any way to automatically turn this into a matrix multiplication? I would want the result to look something like

$$\frac{d}{dt}\ \begin{pmatrix} \vdots\\ P_x(i)\\ P_y(i)\\ P_z(i)\\ P_x(i+1)\\ P_y(i+1)\\ P_z(i+1)\\ \vdots \end{pmatrix} = \begin{pmatrix} \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots \\ \ddots & 0 & 1 & 0 & 0 & -1 & 0 & \ddots\\ \ddots & -1 & 0 & -a & 1 & 0 & 0 & \ddots\\ \ddots & 0 & b & 0 & 0 & 0 & 0 & \ddots\\ \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots\\ \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots\\ \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots\\ \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots & \ddots\\ \end{pmatrix} \begin{pmatrix} \vdots\\ A_x(i)\\ A_y(i)\\ A_z(i)\\ A_x(i+1)\\ A_y(i+1)\\ A_z(i+1)\\ \vdots \end{pmatrix} \tag{2}$$

You can already see from this example that there is some regularity as P(i+1), P(i+2), etc. will have the same entries in the matrix as the ones for P(i), only that the entries within the matrix will be at different positions.

So the question is: If I gave you the vector on the right-hand side of equation (1), could you give me the matrix on the right-hand side of equation (2)?

• SparseArray[{ {i_, j_} /; Mod[i, 3] == 1 && j == i + 1 :> 1, {i_, j_} /; Mod[i, 3] == 1 && j == i + 4 :> -1, {i_, j_} /; Mod[i, 3] == 2 && j == i + 1 :> 1, {i_, j_} /; Mod[i, 3] == 2 && j == i + 3 :> -a, {i_, j_} /; Mod[i, 3] == 2 && j == i + 4 :> -1, {i_, j_} /; Mod[i, 3] == 0 && j == i + 1 :> b, ...}, {n, n}]? – Michael E2 Sep 26 '20 at 21:36
• @MichaelE2, I tried your suggestion and must be doing something wrong because when I copy-paste what you wrote, into v12.1 on a Mac, it gives an error "Syntax::sntxf" saying SparseArray cannot be followed by what is following it. I cleared variables so I don't think that is my problem and unfortunately, it doesn't work for me. – Mark R Sep 26 '20 at 21:54
• @MarkR The ellipsis is an ellipsis, not a RepeatedNull. The code is about as complete as the post and is merely a hint. – Michael E2 Sep 26 '20 at 22:44
• @MichaelE2 - thanks for letting me know. I should have realized it was just showing the general idea. I should have looked more closely before trying to copy-paste. – Mark R Sep 27 '20 at 19:48

You could use CoefficientArrays. For example:

m = CoefficientArrays[
{
Ay[i]-Ay[i+1],
-Ax[i]+Ax[i+1]-a Az[i],
b Ay[i]
},
{Ax[i],Ay[i],Az[i],Ax[i+1],Ay[i+1],Az[i+1]}
][[2]];
m //Normal //TeXForm


$$\left( \begin{array}{cccccc} 0 & 1 & 0 & 0 & -1 & 0 \\ -1 & 0 & -a & 1 & 0 & 0 \\ 0 & b & 0 & 0 & 0 & 0 \\ \end{array} \right)$$

• Very neat solution! – xabdax Sep 26 '20 at 23:08