5

A fairly well-known puzzle asks for the resistance between two adjacent nodes of an infinite square resistor grid, like the points A and B in the drawing. If the resistors are all $1\Omega$, the answer is $\frac12\Omega$, which can be seen by superposition of the two situations where you inject a current in node A, or an opposite current in node B, and use symmetry to argue that in each situation $1/4$th of the current flows through each of the directly adjacent resistors.

Question: Is there also a simple way to see the impedance between diagonally adjacent nodes A and D? (And maybe even for other cases, like A to C, or the knight-jump A to E, etc.?)

grid

Those resistance values can be calculated with higher mathematics but that method is not convincing to many people (for instance some will insist that points A and D should give a resistance of $\frac12\sqrt2$). For completeness, nevertheless, here we'll derive the values in that way. We denote nodes on the grid by vectors ${\bf x}=(n,m)$ with integer coordinates. From Kirchhoff's and Ohm's laws we then have for the voltage $V({\bf x})$ on each node: $$ I({\bf x}) = 4V({\bf x})-V({\bf x}-{\bf e}_1)-V({\bf x}+{\bf e}_1) -V({\bf x}-{\bf e}_2)-V({\bf x}+{\bf e}_2) \tag{Eq1} $$ with ${\bf e}_1$ and ${\bf e}_2$ the basis vectors $(1,0)$ and $(0,1)$, and $I({\bf x})$ an external current injected in node ${\bf x}$. Since this is essentially the discrete Laplace operator, we can transform to the Fourier domain and write:

$$ V({\bf x}) = \int_{-\pi}^{\pi} \frac{dk}{2\pi} \ \int_{-\pi}^{\pi} \frac{dq}{2\pi} \ e^{i{\bf p}\cdot{\bf x}}\ \hat V({\bf p}) $$ where the transformed $\hat V$ in Fourier space now depends on ${\bf p}=(k,q)$. Doing the same for $I({\bf x})$ we can invert the discrete Laplace, and write [Eq1] as: $$ \hat V({\bf p}) = \frac1{4-2\cos k-2\cos q} \ \hat I({\bf p}) $$ $$ \Rightarrow \quad V({\bf x}) = \int_{-\pi}^{\pi} \frac{dk}{2\pi} \ \int_{-\pi}^{\pi} \frac{dq}{2\pi} \ \frac{e^{i{\bf p}\cdot{\bf x}}}{4-2\cos k-2\cos q} \ \hat I({\bf p}) \tag{Eq2} $$ If we now inject two opposite unit currents at points A and B, a trivial Fourier transform from ${\bf x}$ space to ${\bf p}$ space will give us: $$ \hat I({\bf p}) = e^{-i {\bf p}\cdot{\bf x_B}} - e^{-i {\bf p}\cdot{\bf x_A}} $$ Using [Eq2] we then transform back and find the potential difference between any two other points C and D, as: $$ V({\bf x}_D)-V({\bf x}_C) = \int_{-\pi}^{\pi} \frac{dk}{2\pi} \ \int_{-\pi}^{\pi} \frac{dq}{2\pi} \ \frac{e^{i\mathbf{p}\cdot\mathbf{x}_D} -e^{i\mathbf{p}\cdot\mathbf{x}_C} } {4-2\cos k-2\cos q} \ (e^{-i {\bf p}\cdot{\bf x}_B} - e^{-i {\bf p}\cdot{\bf x}_A}). $$

For the resistor puzzle, points C and D are the same as A and B, and we can choose one of them to be the origin, which leaves us with one vector ${\bf x} = {\bf x}_B $. The numerator in the integral then becomes:

$$ (1-e^{-i {\bf p}\cdot{\bf x}}) \ (1-e^{i {\bf p}\cdot{\bf x}}) = 2-2\cos({\bf p}\cdot{\bf x}) = 2-2\cos(n k + m q) = $$ $$ = 2-2\cos(n k)\cos(m q)+2\sin(n k)\sin(m q). $$ Since the denominator is totally symmetric in $q$ and $k$ we can keep only the symmetric part of this last expression. Also we'll subsequently use four times the integral in only one quadrant. If we then denote $\vec B-\vec A={\bf x}=(n,m)$ we get:

$$ R(n,m)= 8 \int_0^\pi \frac{dk}{2\pi} \ \int_0^\pi \frac{dq}{2\pi} \ \ \frac{1-\cos(n k)\cos(m q) } {4-2\cos k-2\cos q}. $$

For the case $m=0$, doing the $q$ integral leaves us with:

$$ R(n,0)= \int_0^\pi \frac{dk}{\pi} \ \frac{1 - \cos(k n)} {\sqrt{3-4\cos k +\cos^2 k}}. $$

Doing also the $k$ integral then shows us $R(1,0)=1/2$, which we already knew, just like $R(0,0)=0$, but we also see that for instance $R(2,0)= 2-4/\pi$, which is new! We can of course redo the $q$ integral for $m=1$. This gives:

$$ R(n,1)= \int_0^\pi \frac{dk}{\pi} \Big(\cos(k n) + \frac{(\cos k - 2) \cos(k n) + 1} {\sqrt{3-4\cos k +\cos^2 k}} \Big). $$ and this finally gives us the "diagonal" resistance: $R(1,1)=2/\pi$, which is clearly 10% smaller than the $\frac12\sqrt2\ $ some might have expected, and the knight jump: $R(2,1)=4/\pi-1/2$. In summary:

$$ R(0,1) = \frac12 $$ $$ \qquad R(0,2) = 2-\frac4\pi $$ $$ \quad\qquad R(0,3) = \frac{17}2-\frac{24}\pi $$ $$ R(1,1) = \frac2\pi $$ $$ \qquad R(1,2) = \frac4\pi-\frac12 $$ $$ \qquad R(1,3) = \frac{46}{3\pi}-4 $$

Of course the procedure can nicely be generalized, e.g. for the anisotropic case where the horizontal resistors have a different value than the vertical ones, or for more than $2$ dimensions. And the approach is of course standard practice for particle physicists, but it is still an interesting question whether at least some of these values, like the diagonal $R(1,1)$, have a simpler derivation...

Qmechanic
  • 201,751
  • 3
    https://xkcd.com/356/ – PM 2Ring Mar 09 '24 at 15:11
  • Possible duplicates: https://physics.stackexchange.com/q/2072/2451 , https://physics.stackexchange.com/q/10308/2451 and links therein. – Qmechanic Mar 09 '24 at 15:28
  • 1
    The $\pi$ (a transcendental number) in the expression suggests there should be integrals or summations of infinitely long series involved. So, depending on how "simple" you are looking for, the answer is probably no. The simplest you could hope for still gets an infinite series like $(1-1/3+1/5-1/7+...)$ somewhere along the proof. – Ali Mar 09 '24 at 16:06
  • Sorry for missing the duplicates! Stackexchange did not show them when I submitted this. I should have done a more active search, of course... (Anyhow they do not give an answer to finding a simpler derivation, in fact the derivations I see look more like unnecessary long.) – Jos Bergervoet Mar 09 '24 at 16:37

0 Answers0