If I want to prove that the derivative of a vector $V$, $\partial_\mu V^\nu$, is not a tensor, I can apply the tensor transformation rule to $V^\nu$:

$$\partial_\mu V^\nu \rightarrow \partial_{\mu\prime} V'^{\nu\prime} = \partial_{\mu\prime}\left( \frac{\partial x'^{\nu'}}{\partial x^\nu}V^\nu\right)$$

Now the next correct step would be to apply the chain rule to $\partial_{\mu'}=\frac{\partial x^\mu}{\partial x'^{\mu'}}\frac{\partial}{\partial x^\mu}$ and then expand. Everything from this point is clear to me.

But my course notes mention that it would not be correct to differentiate the expression in the brackets with $\partial_{\mu\prime}$ right away, but I do not understand why.

Why would that step be illegal? Why did we have to apply the chain rule first?

  • $\begingroup$ and what if it is a tensor, indeed? $\endgroup$
    – basics
    Nov 8, 2022 at 14:31
  • $\begingroup$ apologies, i should have said that V is a vector, I will edit it in $\endgroup$
    – NX37B
    Nov 8, 2022 at 14:32
  • $\begingroup$ this is not the way you transform tensors. An answer will follow soon $\endgroup$
    – basics
    Nov 8, 2022 at 14:35

2 Answers 2


The shorthand notation is tripping you up. Let's say that in a particular coordinate chart $x$, your vector components take the form $V^\mu (x)$. You now decide to differentiate these components, to obtain an expression $$\frac{\partial}{\partial x^\nu}\big[ V^\mu(x)\big]$$

Now you wish to change to a different coordinate chart $y$. The components of $V$ transform as $$V^\mu(x) \mapsto J^\mu_{\ \ \alpha}(y) \hat V^\alpha(y)$$ where $J^\mu_{\ \ \alpha}(y)\equiv \frac{\partial x^\mu (y)}{\partial y^\alpha}$ is the Jacobian of the transformation. Plugging that in yields

$$\frac{\partial}{\partial x^\nu}\left[J^\mu_{\alpha}(y)\hat V^\alpha(y)\right]$$

The problem now is that the quantity inside the brackets is a function of the coordinates $y$, so you cannot differentiate them with respect to the coordinates $x$. Instead, you need to transform your derivative operator: $$\frac{\partial}{\partial x^\nu} \mapsto \big(J^{-1}(y)\big)^\beta_{\ \ \nu} \frac{\partial}{\partial y^\beta}$$

Now you can expand everything out using the product rule. To recover the shorthand notation, you can drop the arguments $(x)$ and $(y)$ from all of the functions, write the Jacobian as $\partial x/\partial y$, and replace $y$ with $x'$. From there, replace $\frac{\partial}{\partial x^\mu}$ with $\partial_\mu$ and $\frac{\partial}{\partial x'^\mu}$ with $\partial_{\mu'}$. (Can you see how this gets confusing for beginners?)

As an example, we can transform the vector field $\mathbf V = \frac{x}{\sqrt{x^2+y^2}}\frac{\partial}{\partial x}+\frac{y}{\sqrt{x^2+y^2}}\frac{\partial}{\partial y}$ from Cartesian coordinates $(x,y)$ to polar coordinates $(r,\theta)$. Clearly this is simply equal to $\frac{\partial}{\partial r}$, so this is what we should expect to find at the end.

Because $x(r,\theta)=r\cos(\theta)$ and $y(r,\theta) = r\sin(\theta)$, the Jacobian and its inverse take the form $$J(r,\theta) = \pmatrix{\cos(\theta) & -r\sin(\theta)\\ \sin(\theta) & r\cos(\theta)} \qquad J^{-1}(r,\theta) = \pmatrix{\cos(\theta) & \sin(\theta)\\ -\frac{1}{r}\sin(\theta) & \frac{1}{r}\cos(\theta)}$$

The initial Cartesian components are $V^x(x,y) = \frac{x}{\sqrt{x^2+y^2}}$ and $V^y(x,y) = \frac{y}{\sqrt{x^2+y^2}}$. Expressed in terms of $r$ and $\theta$, we have that $V^x = \cos(\theta)$ and $V^y = \sin(\theta)$. According to the transformation rule,

$$\pmatrix{V^x\\V^y} = J(r,\theta) \pmatrix{\hat V^r\\ \hat V^\theta} \implies \pmatrix{\hat V^r\\ \hat V^\theta} = J^{-1}(r,\theta) \pmatrix{\cos(\theta)\\\sin(\theta)} = \pmatrix{1\\0}$$ And so $\hat V^r(r,\theta) = 1$ and $\hat V^\theta(r,\theta) = 0$, as anticipated.

In this language, the question in the OP relates to expressions of the form

$$\frac{\partial}{\partial x} \pmatrix{V^x(x,y)\\ V^y(x,y)} = \frac{\partial}{\partial x} \left[ J(r,\theta) \pmatrix{\hat V^r(r,\theta)\\\hat V^\theta(r,\theta)}\right]$$

and the fact that it is inappropriate to differentiate the RHS before transforming the derivative operator.

  • $\begingroup$ "The problem now is that the quantity inside the brackets is a function of the coordinates y, so you cannot differentiate them with respect to the coordinates x." Ah I think I see. I missed that the Jacobian is actually a function of x, not x'. $\endgroup$
    – NX37B
    Nov 8, 2022 at 15:43

What you write $A_{\mu}^{\nu} = \partial_{\mu} V^{\nu}$ is the component representation of the tensor

$\mathbb{A} = \mathbf{b}^{\mu}\otimes\mathbf{b}_{\nu} A_{\mu}^{\nu} = \mathbf{b}^{\mu}\otimes\mathbf{b}_{\nu} \partial_{\mu} V^{\nu}$,

assuming $\mathbf{b}_{\nu}$ as the base vectors you're referring to, and $\mathbf{b}^{\mu}$ the vectors of the reciprocal basis, s.t. $\mathbf{b}_{\nu} \cdot \mathbf{b}^{\mu} = \delta^{\mu}_{\nu}$.

We can write the components of your tensor in a different basis, whose vectors $\mathbf{b}'_{\alpha}$ can be as a linear combination of the vectors of the original basis, using the

$\mathbf{b}'_{\alpha} = T_{a}^{\nu} \mathbf{b}_{\nu}$$\qquad , \qquad$$\mathbf{b}'^{\beta} = \tilde{T}^{\beta}_{\mu} \mathbf{b}^{\mu}$
$\mathbf{b}_{\nu} = \tilde{T}^{a}_{\nu} \mathbf{b}'_{\alpha}$$\qquad , \qquad$$\mathbf{b}^{\mu} = {T}_{\beta}^{\mu} \mathbf{b}'^{\beta}$ ,

being $\tilde{T}^{\beta}_{\mu}$ the inverse matrix $ T_{a}^{\nu}$. Replacing the expression of the vectors of the basis in the expression of the tensor $\mathbb{A}$ we get

$\mathbb{A} = \mathbf{b}^{\mu}\otimes\mathbf{b}_{\nu} A_{\mu}^{\nu} = \mathbf{b}'^{\beta}\otimes\mathbf{b}'_{\alpha} \underbrace{ {T}_{\beta}^{\mu} \tilde{T}^{a}_{\nu} A_{\mu}^{\nu}}_{{A'_{\beta}}^{\alpha}} = \mathbf{b}'^{\beta}\otimes\mathbf{b}'_{\alpha} {A'_{\beta}}^{\alpha}$

  • $\begingroup$ I am sorry I do not understand. My course notes and everywhere else I looked at say that the partial derivative of a vector is not a tensor. I am only familiar with the component representation of tensors, and I've never seen the 'cross in a circle' operation that you used here, what is it? $\endgroup$
    – NX37B
    Nov 8, 2022 at 15:02
  • $\begingroup$ @NX37B The cross in a circle is a tensor product. $\endgroup$
    – Ghoster
    Nov 8, 2022 at 18:12
  • $\begingroup$ @NX37B. Technically any linear combination of the basis vectors, or tensor products of the basis vectors, is a tensor. However the differential of a vector field is not a “good” tensor in that $\partial_j v^i \pmb{e}_i \otimes \pmb{e}^j \neq \partial’_j v’^i \pmb{e}’_i \otimes \pmb{e}’^j$. $\endgroup$
    – J Peterson
    Nov 17, 2022 at 4:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.