1

$A^\alpha$ is a given contravariant vector (when $\alpha\in {0,1,2,3}$ a $4$-vector in Minkowski space)

I need to show that the derivative $\frac{\partial A^\alpha}{\partial x^\beta}$ is a mixed tensor of rank 2, so:

$\frac{\partial A^\alpha}{\partial x^\beta}=T^\gamma_\delta$

(I'm not sure if $\gamma=\alpha$ and $\delta=\beta$)

The subject is Special Relativity so the space taking diffrential 2-form.

Andrews
  • 3,953
  • Non-contracted indices on either side should match. So it should be $\frac{\partial A^\alpha}{\partial x^\beta} = T^\alpha_\beta$ (indices under the "fraction line" in a derivative are moved, so the $\beta$ in $\frac{}{\partial x^\beta}$ is a lower index). – Arthur Mar 26 '19 at 11:10

1 Answers1

2

The most straight-forward, brute-force way to show whether something is a tensor is to perform a change of coordinates and see how the components change.

Given concrete coordinates $x^\beta$ and an arbitrary point $p$ in our space, $A^\alpha$ is a $4$-tuple of numbers (some might call it a vector, or even a column vector), and $T^\alpha_\beta$ is a $4\times 4$ table of numbers describing how the $4$-tuple of $A^\alpha$ change when we move from $p$ to a point with slightly changed $i$th coordinate.

Now, let's say we have a new set of coordinates, $x^{\beta'}$. Since $A$ is a tensor, the $4$-tuple $A^{\alpha'}$ of numbers it gives at point $p$ with the new coordinates is given by $$ A^{\alpha'} = A^\alpha\frac{\partial x^{\alpha'}}{\partial x^\alpha} $$ (I have barely touched tensors in two years, and I still remember which coordinate functions to differentiate with respect to which. Such is the power of index notation and knowing how indices must match up.)

If we now differentiate both sides with respect to the new coordinate systems, we get $$ T^{\alpha'}_{\beta'} = \frac{\partial A^{\alpha'}}{\partial x^{\beta'}} = \frac{\partial}{\partial x^{\beta'}} \left(A^\alpha\frac{\partial x^{\alpha'}}{\partial x^\alpha}\right)\\ = \frac{\partial A^{\alpha}}{\partial x^{\beta'}} \frac{\partial x^{\alpha'}}{\partial x^\alpha} + A^{\alpha}\frac{\partial}{\partial x^{\beta'}}\frac{\partial x^{\alpha'}}{\partial x^\alpha} $$ By the chain rule, the first term here is $$ \frac{\partial A^{\alpha}}{\partial x^{\beta'}} \frac{\partial x^{\alpha'}}{\partial x^\alpha} = \frac{\partial A^{\alpha}}{\partial x^{\beta}}\frac{\partial x^{\beta}}{\partial x^{\beta'}} \frac{\partial x^{\alpha'}}{\partial x^\alpha} = T^\alpha_\beta\frac{\partial x^{\beta}}{\partial x^{\beta'}} \frac{\partial x^{\alpha'}}{\partial x^\alpha} $$ which is the correct change-of-coordinates expression for a $(1, 1)$ tensor like $T$. However, there is a second term: $$ A^{\alpha}\frac{\partial}{\partial x^{\beta'}}\frac{\partial x^{\alpha'}}{\partial x^\alpha} $$ And we sadly cannot expect this to be non-zero in general. We can see this, for instance, in $2$ dimensions with Cartesian versus polar coordinates, where $$ \frac{\partial}{\partial\theta}\frac{\partial r}{\partial x} $$ clearly cannot be $0$, as $\frac{\partial r}{\partial x}$ is $0$ for a point on the $y$ axis and $1$ for a point on the (positive) $x$-axis, so as a function of $\theta$ it cannot be constant.

Which is to say $T^\alpha_\beta$ is not a tensor. However, there is a way to mend this. The mend comes in the form of the so-called Christoffel symbols and the covariant derivative. Very briefly, the covariant derivative consists of taking the partial derivative, then for each index throw in a correction term using the Christoffel symbols (which have three indices: $\Gamma^\gamma_{\mu\nu}$, and is thus a $4\times 4\times 4$ table of numbers at each point) to ensure that what you get out in the end is indeed a tensor. The covariant derivative of $A^\alpha$ is $$ \nabla_\beta A^\alpha = \frac{\partial A^\alpha}{\partial x^\beta} + \Gamma^\alpha_{\beta\eta}A^\eta $$

Arthur
  • 199,419
  • I'm a bit confused by the term $\frac{\partial}{\partial x^{\beta'}}\frac{\partial x^{\alpha'}}{\partial x^\alpha}$. It equals to $\frac{\partial^2 x^{\alpha'}}{\partial x^{\beta'}\partial x^\alpha}$, exchange the order of two partial derivatives, it equals to $\frac{\partial^2 x^{\alpha'}}{\partial x^\alpha\partial x^{\beta'}} = \frac{\partial}{\partial x^\alpha}\delta^{\alpha'}_{\beta'} = 0$. Did I make a mistake somewhere? – Guangliang Sep 17 '20 at 16:45
  • 1
    @Guangliang You cannot exchange the order of two partial derivatives from two different coordinate systems. Just check for yourself with, say, Cartesian and polar coordinates in the Euclidean plane, on any of the four coordinate functions. For instance, take the $x$-function, and differentiate with respect to $x$ and $r$. First $x$ then $r$ will give $0$, while first $r$ then $x$ will give $\dfrac{y^2}{\sqrt{x^2 + y^2}^3} = \dfrac{\sin\theta}{r}$. – Arthur Sep 17 '20 at 18:41
  • The reason that it doesn't work is that the direction of increasing $r$ (the direction whose directional derivative is the polar coordinate $\frac{\partial}{\partial r}$) also changes $x$ (and vice versa). This doesn't happen when the partial derivatives come from the same coordinate system. – Arthur Sep 17 '20 at 18:42
  • Man, God bless you, I had been stuck on this problem for two days and it was from your answer that I understood. I really can't put in words how grateful I am for this answer – tryst with freedom May 09 '21 at 11:24
  • I have a few questions: 1. Why did you decide to differentiate with the new system? pls see my post here , 2. For the sake of illustration, could you show how to evaluate the second term in the sum (see this question] – tryst with freedom May 09 '21 at 11:27
  • @Buraian As Grinfeld likes to say, at least in his video lectures on this topic, most theorems in tensor calculus comes about from setting up the right identity and then differentiate.

    You have $$ T^{i'} = T^{i} J_{i}^{i'} $$ We want $\frac{\partial T^{i'}}{\partial Z^{j'}}$, so it seems very natural to differentiate with respect to the primed coordinates $Z^{j'}$. I don't know that much can be sone with the second term in general. You just have to keep it as it is, and watch it cancel against the Christoffel term in the covariant derivative.

    – Arthur May 09 '21 at 12:17
  • It's been a while since I did this, though. I may have forgotten a trick. – Arthur May 09 '21 at 12:22
  • Wait, so you learned from Pavel too? xD – tryst with freedom May 09 '21 at 13:26
  • @Buraian His videos were very good supplemental material when I tried to get through general relativity half a decade back. – Arthur May 09 '21 at 13:40