1

I'm trying to understand how tensor components transform under a change of coordinate system, but something doesn't look right. Below is the derivation that I wrote out:


We start with a general example of the tensor transformation equation for a tensor $\mathbf{T}$ of rank greater than 1, in this case 3:

$$\begin{align} T^{\mu^\prime}{}_{\nu^\prime \lambda^\prime} = \frac{\partial x^{\mu^\prime}}{\partial x^{\mu}}\frac{\partial x^{\nu}}{\partial x^{\nu^\prime}}\frac{\partial x^{\lambda}}{\partial x^{\lambda^\prime}} T^{\mu}{}_{\nu\lambda} \end{align}$$

To test out this equation, we'll simplify it to the case of a rank-1 tensor, which can be thought of as a vector, and work in 2 dimensions, where an index value of $0$ represents the first coordinate and an index value of $1$ represents the second coordinate: $$T^{\mu^\prime} = \frac{\partial x^{\mu^\prime}}{\partial x^{\mu}}T^{\mu} = \frac{\partial x^{\mu^\prime}}{\partial x^{0}}T^{0} + \frac{\partial x^{\mu^\prime}}{\partial x^{1}}T^{1}$$

Let's use the familiar example of 2-dimensional polar coordinates, mapped from the Cartesian $x$ and $y$ as follows:

$$\begin{align} x^{0^\prime} = r = \sqrt{x^2 + y^2} \\\\[4pt] x^{1^\prime} = \theta = \arctan\left(\frac{y}{x}\right) \end{align}$$

Here's what I expect the results to be based on standard linear algebra: $$\begin{bmatrix}T^r \\\ T^\theta\end{bmatrix} = \begin{bmatrix}\cos\theta & \sin\theta \\\ -\sin\theta & \cos\theta\end{bmatrix}\begin{bmatrix}T^x \\\ T^y\end{bmatrix}$$ Let's try to verify this in tensor form:

Using the above expression for $T^{\mu^\prime}$, we can write out the components $T^{0^\prime}$ and $T^{1^\prime}$ explicitly by setting $\mu^\prime = 0$ and then $\mu^\prime = 1$. These are the expected components of our rank-1 tensor (vector) when transforming to the new coordinate system:

$$\begin{align} T^{0^\prime} = \frac{\partial x^{0^\prime}}{\partial x^{\mu}}T^{\mu} = \frac{\partial x^{0^\prime}}{\partial x^{0}}T^{0} + \frac{\partial x^{0^\prime}}{\partial x^{1}}T^{1} \\\\[6pt] T^{1^\prime} = \frac{\partial x^{1^\prime}}{\partial x^{\mu}}T^{\mu} = \frac{\partial x^{1^\prime}}{\partial x^{0}}T^{0} + \frac{\partial x^{1^\prime}}{\partial x^{1}}T^{1} \end{align}$$

It looks like we have some partial differentiation to do, so let's take all 4 of these partial derivatives $\partial x^{\mu^\prime} / \partial x^\mu$:

$$\begin{align} \frac{\partial x^{0^\prime}}{\partial x^0} & = \frac{\partial r}{\partial x} = \frac{\partial}{\partial x}\sqrt{x^2 + y^2} = \frac{x}{\sqrt{x^2 + y^2}} \\\\[6pt] \frac{\partial x^{0^\prime}}{\partial x^1} & = \frac{\partial r}{\partial y} = \frac{\partial}{\partial y}\sqrt{x^2 + y^2} = \frac{y}{\sqrt{x^2 + y^2}} \\\\[6pt] \frac{\partial x^{1^\prime}}{\partial x^0} & = \frac{\partial \theta}{\partial x} = \frac{\partial}{\partial x}\arctan\left(\frac{y}{x}\right) = -\frac{y}{x^2 + y^2} \\\\[6pt] \frac{\partial x^{1^\prime}}{\partial x^1} & = \frac{\partial \theta}{\partial y} = \frac{\partial}{\partial y}\arctan\left(\frac{y}{x}\right) = \frac{x}{x^2 + y^2} \end{align}$$

Now all that's left is to plug in our partial derivatives $\partial x^{\mu^\prime} / \partial x^\mu$ into our above expression for $T^{0^\prime}$ and $T^{1^\prime}$ to obtain the components of $\mathbf{T}$ when transforming to polar coordinates: $$\begin{align} T^{0^\prime} = T^r & = \frac{x}{\sqrt{x^2 + y^2}}T^0 + \frac{y}{\sqrt{x^2 + y^2}}T^1 \\\\[6pt] & = \frac{x}{r}T^0 + \frac{y}{r}T^1 \\\\[6pt] & = T^0\cos\theta + T^1\sin\theta \end{align}$$

This looks great. Likewise, for the angular component, $$\begin{align} T^{1^\prime} = T^\theta & = -\frac{y}{x^2 + y^2}T^0 + \frac{x}{x^2 + y^2}T^1 \\\\[6pt] & = \frac{1}{\sqrt{x^2 + y^2}}\left[-\frac{y}{\sqrt{x^2 + y^2}}T^0 + \frac{x}{\sqrt{x^2 + y^2}}T^1\right] \\\\[6pt] & = \frac{1}{r}\left[-\frac{y}{r}T^0 + \frac{x}{r}T^1\right] \\\\[6pt] & = \frac{1}{r}\left[T^1\cos\theta - T^0\sin\theta \right] \\\\[6pt] \end{align}$$ Now, the bracketed term is what I expect, but I'm obtaining a $1/r$ term in front, which seems strange to me. I'm not sure whether my mistake is simple algebra or if I'm fundamentally misunderstanding something here. Any ideas as to where I went wrong?

Thanks!

JBL
  • 13
  • Are you sure of your very first equality? I thing you have mixed up some lower and upper indexes – Didier Oct 02 '22 at 16:06
  • 1
    There is no mistake in your calculation. That is the correct formula. Why does the result seem strange to you? Why do you only expect the bracketed term? – Jackozee Hakkiuz Oct 02 '22 at 16:11
  • Based on the transformation matrix of sines and cosines above, it doesn't look like there should be an $1/r$ term in the result. I'm not sure where the $1/r$ term comes from. – JBL Oct 02 '22 at 16:36
  • @JackozeeHakkiuz ^ – JBL Oct 02 '22 at 16:50

1 Answers1

3

To every coordinate system $(x^1,\dots,x^n)$, there is a naturally associated (I will call it canonical) basis $\partial_{x^1},\dots,\partial_{x^n}$. Tensor equations such as $$T^{\mu'} = \frac{\partial x^{\mu'}}{\partial x^\mu} T^\mu$$ assume you are using these canonical bases everywhere, with respect to the both the old and the new coordinates: $$ T = T^\mu \partial_{x^\mu} = T^{\mu'}\partial_{x^{\mu'}}. $$ Your calculations are correct, so the matrix you obtained $$ \begin{bmatrix} \frac{\partial x^{0^\prime}}{\partial x^0} & \frac{\partial x^{0^\prime}}{\partial x^1} \\ \frac{\partial x^{1^\prime}}{\partial x^0} & \frac{\partial x^{1^\prime}}{\partial x^1} \end{bmatrix} = \begin{bmatrix} \frac{x}{\sqrt{x^2 + y^2}} & \frac{y}{\sqrt{x^2 + y^2}} \\ -\frac{y}{x^2 + y^2} & \frac{x}{x^2 + y^2} \end{bmatrix} = \begin{bmatrix} \cos\theta & \sin\theta \\ -\frac{1}{r}\sin\theta & \frac{1}{r}\cos\theta \end{bmatrix} $$ correctly transforms components with respect to the canonical basis of cartesian coordinates: $\partial_x,\partial_y$ into components with respect to the canonical basis of polar coordinates: $\partial_r,\partial_\theta$: $$\begin{align} T &= T^x \partial_x + T^y\partial_y \\ &= T^{r}\partial_r + T^{\theta}\partial_\theta \end{align}$$ where $$ \begin{bmatrix} T^{r} \\ T^{\theta} \end{bmatrix} = \begin{bmatrix} \cos\theta & \sin\theta \\ -\frac{1}{r}\sin\theta & \frac{1}{r}\cos\theta \end{bmatrix} \begin{bmatrix} T^x \\ T^y \end{bmatrix} $$ are the components of $T$ with respect to the basis $\partial_x,\partial_y$ and the basis $\partial_r,\partial_\theta$.

The mistake is that the equation

$$ \begin{bmatrix} T^{r} \\ T^{\theta} \end{bmatrix} \overset{?}{=} \begin{bmatrix} \cos\theta & \sin\theta \\ -\sin\theta & \cos\theta \end{bmatrix} \begin{bmatrix} T^{x} \\ T^{y} \end{bmatrix} \tag 1 $$ assumes a different basis. This last equation $(1)$ supposes you are using the canonical basis $\partial_x,\partial_y$ in Cartesian coordinates, but you are using the orthonormal basis $e_r,e_\theta$ in polar coordinates. Actually, let me denote the components of $T$ with respect to the orthonormal basis $e_r,e_\theta$ as $\hat T^r,\hat T^\theta$. Now the correct version of $(1)$ is

$$ \begin{bmatrix} \hat T^{r} \\ \hat T^{\theta} \end{bmatrix} = \begin{bmatrix} \cos\theta & \sin\theta \\ -\sin\theta & \cos\theta \end{bmatrix} \begin{bmatrix} T^{x} \\ T^{y} \end{bmatrix} \tag 2 $$

The relation between the bases $\partial_r, \partial_\theta$ and $e_r,e_\theta$ is \begin{align} \partial_r &= e_r \tag 3\\ \partial_\theta &= r e_\theta. \tag 4 \end{align}

That is: the only difference between both bases is that the canonical angular basis vector $\partial_r$ has length $r$ but the orthonormal angular basis vector $e_\theta$ has length $1$.

To get the expression you want (the one in orthonormal coordinates) all you have to do is to take the (correct) expression you got (the one in canonical coordinates): \begin{align} T &= T^r\partial_r + T^\theta \partial_\theta \\ &= [T^x\cos\theta + T^y\sin\theta]\partial_r +\frac{1}{r} [-T^x\sin\theta + T^y\cos\theta]\partial_\theta. \end{align} and substitute with equations $(3)$ and $(4)$; that is $\partial_r=e_r$ and $\partial_\theta = r e_\theta$.

Personally I wouldn't reccomend it. If you wanted to further change coordinates from polar to, say, some slanted affine coordinate system $x^{\mu''}$, you would need to undo this in order to apply the tensor formula $$T^{\mu''} = \frac{\partial x^{\mu''}}{\partial x^{\mu'}} T^{\mu'}.$$

Jackozee Hakkiuz
  • 6,119
  • 1
  • 17
  • 38
  • This is amazing -- thank you! Do you know of a general formula to convert between canonical bases and orthonormal bases? – JBL Oct 02 '22 at 19:13
  • @JBL In the case of polar coordinates, you have an associated orthonormal basis because the canonical basis is already orthogonal, so all you have to do is normalize everything. The same happens in cylindrical or spherical coordinates, but it doesn't happen in every coordinate system. In general there are many orthonormal bases from which you have to choose some. The general method to switch between orthonormal bases and canonical ones is called the method of orthonormal tetrads, the "tetrad formalism", or also vielbein, or also moving Cartan frames. – Jackozee Hakkiuz Oct 02 '22 at 19:31
  • 1
    @JBL By the way, my terminology of "canonical basis" is not standard. What I called canonical basis is usually called holonomic basis or coordinate basis in the literature. – Jackozee Hakkiuz Oct 02 '22 at 19:31
  • Very helpful explanation thank you very much – JBL Oct 02 '22 at 21:27
  • @JBL my pleasure. I'm glad my answer was helpful :) – Jackozee Hakkiuz Oct 03 '22 at 00:59