Still trying to teach myself some basic differential geometry in relation to general relativity. I've read that, in relation to basis vectors $e_{\nu}=\partial_{\nu}$ and basis one-forms $\omega^{\mu}=dx^{\mu}$, then $$e_{\nu}\omega^{\mu}=\partial_{\nu}dx^{\mu}=\frac{\partial x^{\mu}}{\partial x^{\nu}}=\delta_{\nu}^{\mu}.$$ Fair enough, but why does the partial derivative operator $\partial_{\nu}$ (which I thought meant $\frac{\partial}{\partial x^{\nu}}$ , where the slot is for an arbitrary function) when applied to $dx^{\mu}$ give $\frac{\partial x^{\mu}}{\partial x^{\nu}}$ and not $\frac{\partial\left(dx^{\mu}\right)}{\partial x^{\nu}}$? Or does $\frac{\partial x^{\mu}}{\partial x^{\nu}}=\frac{\partial\left(dx^{\mu}\right)}{\partial x^{\nu}}$? But if that's the case what does $\partial_{\nu}x^{\mu}$ give? It can't also be correct that

$$\partial_{\nu}x^{\mu}=\frac{\partial x^{\mu}}{\partial x^{\nu}}=\delta_{\nu}^{\mu},$$ but I can't see why not. I'm confused. Thanks.


I like Amitai Yuval's answer, but let me try to say something a little more basic. It's necessarily a bit vague, but maybe it will give you the idea of what's going on.

The coordinates $x^1, \dots, x^n$ define functions on your manifold. In a physics context, $x^1$, $x^2$ and $x^3$ may represent position with respect to a particular set of axes whilst $x^4$ labels time. Tangent vectors are little arrows at points in your space, and we have a nice basis at each point given by the arrows pointing along each of the coordinate directions. In our example these would correspond to arrows in the directions of the $x^1$-, $x^2$- and $x^3$-axes, and a fourth arrow which points forwards in time but has no spatial component. We write these vectors as $\partial/\partial x^a$ or $\partial_a$, but this is just notation for now.

A one-form is 'a thing you plug vectors into': you feed it a vector and it spits out a number which depends linearly on the input. For each $a$ we can define a one-form $\mathrm{d}x^a$ to be the one-form which gives $1$ when you input $\partial_a$ and $0$ when you input $\partial_b$ for $b \neq a$. Again this is just notation. In other words, the statement $\mathrm{d}x^a(\partial_b)=\delta^a_b$ is basically a tautology. Back to our example, if you plug a vector $v$ into $\mathrm{d}x^4$ then what you get out is the time component of $v$.

Now, given a vector $v$ at a point $p$ and a function $f$ defined near $p$ you can take the directional derivative of $f$ in the direction $v$, which you could write as $v(f)$ (although personally I don't like this notation). If you take $v$ to be $\partial_a$ then this directional derivative really just is the partial derivative $\partial f/\partial x^a$. This is why the notation $\partial_a$ or $\partial/\partial x^a$ is sensible. So in our example $f$ could be something like density and then $\partial_4(f)$ represents the time-derivative of the density.

Thinking of the vector $v$ in the expression $v(f)$ as variable, we get a thing which takes vectors and spits out numbers (given the vector $v$ it outputs $v(f)$). In other words it defines a one-form: the differential (or derivative) of $f$, which we write as $\mathrm{d}f$. Happily, the one-form we defined to be $\mathrm{d}x^a$ purely as a piece of notation really just is the differential of the function $x^a$.

So in terms of directional derivatives it is true that $\partial_a(x^b)=\delta^b_a$. Whilst in terms of feeding vectors into one-forms it is true that $\mathrm{d}x^a(\partial_b)=\delta^a_b$. By the above comment about differentials, relating directional derivatives to one-forms, these statements are really saying the same thing.

The expression you write as $\partial_\nu \mathrm{d}x^\mu$ is a bit ambiguous. It definitely doesn't mean 'differentiate $\mathrm{d}x^\mu$'. The sensible interpretation (which makes it equal to $\delta^\mu_\nu$) is that you're feeding the one-form into the vector using double-duality. But this is probably a little too abstract at present.

  • $\begingroup$ Thanks. In your para 2, could an example of these $x^{1}$ etc coordinates on the manifold be Cartesian coordinates $x,y,z$ (assuming we're in Euclidean space). If these coordinates were defined in terms of spherical coordinates $r,\theta,\phi$, we could say $z=r\cos\phi$ and thus $\frac{dz}{dr}=\cos\phi$ and $dz=\cos\phi dr$. Is $dz$ a one-form? Also, in para 4 you say vector $v=\frac{\partial}{\partial x^{a}}$. In para 3 you say if we plug $v$ into $dx^{4}$ we get the time component of $v$. But surely $\frac{\partial}{\partial x^{a}}dx^{4}=\frac{\partial}{\partial x^{4}}dx^{4}=1$? $\endgroup$ – Peter4075 Oct 17 '15 at 13:00
  • $\begingroup$ Yes, if you're in 3d Euclidean space then you could take $x^1$, $x^2$ and $x^3$ to be the Cartesian coordinates. You're right that $\mathrm{d}z$ is a one-form, but when you pass to spherical coordinates you should get $\mathrm{d}z=\cos \phi \mathrm{d}r-r \sin \phi \mathrm{d}\phi$. $\endgroup$ – Jez Oct 17 '15 at 15:33
  • 1
    $\begingroup$ As I mentioned at the end of my answer, it's a bit ambiguous to write a vector followed by a one-form. The correct statement is $\mathrm{d}x^4(\partial_a)=\delta^4_a$. Why do you think you should replace the index $a$ by $4$ in the last line of your comment? $\endgroup$ – Jez Oct 17 '15 at 15:37
  • $\begingroup$ $\mathrm{d}z/\mathrm{d}r$ is not really a thing, since $z$ is a function of three variables: $r$, $\theta$ and $\phi$ (although it's actually independent of $\theta$). $\endgroup$ – Jez Oct 17 '15 at 15:39
  • $\begingroup$ Sorry, I meant to use the total differential of $z$ $(\mathrm{d}z=\cos\phi\mathrm{d}r-r\sin\phi\mathrm{d}\phi)$. But at least now that I'm using a simple example, the language of coordinate functions is becoming a little easier to visualise. I used $a=4$ in $\mathrm{d}x^{4}(\partial_{a})=\delta_{a}^{4}$ because any other value would give zero. Does that mean time component of $v$ in this case equals 1? $\endgroup$ – Peter4075 Oct 17 '15 at 17:14

It is only a matter of notation. There are a few different ways to look at things, the following is one of those ways.

Take a coordinate chart $\varphi:U\to\mathbb{R}^n$. Then $\varphi$ consists of $n$ real valued functions which we write as $x^1,\ldots,x^n$. The derivative of the $i$th function is $dx^i$, and this is how we get the local frame of the cotangent bundle $dx^1,\ldots,dx^n$.

On the other hand, the chart $\varphi$ also induces a local frame of the tangent bundle, which we write as $\frac{\partial}{\partial x^1},\ldots,\frac{\partial}{\partial x^n}.$

Now, there are two standard notations for the derivative of $f$ in the direction $X$ - one of them is $X(f)$ and the other is $df(X)$. Using the former, we can write$$\frac{\partial}{\partial x^i}(x^j)=\delta_i^j,$$whereas the latter yields$$dx^j\left(\frac{\partial}{\partial x^i}\right)=\delta_i^j.$$Both equations mean the same.

  • $\begingroup$ I'm afraid that's above my head. Any chance of a more high school level type answer? I'm OK with contravariant vector components as tangent vectors, and one-form components as gradients and that these things live at a point on a manifold (4d for spacetime). I understand that $$e_{\nu}\omega^{\mu}=\delta_{\nu}^{\mu}.$$ But "charts", "bundles" and "$n$ real valued functions" lose me. $\endgroup$ – Peter4075 Oct 14 '15 at 14:32
  • $\begingroup$ Thanks to Amitai Yuval, but the second paragraph loses me. What are these $n$ real valued functions? Are they coordinate transformations from one chart to another? The reason for my question was trying to understand how the product of a one-form and a contravariant vector gives a scalar. Hence my trying to see how the product of the basis vectors $e_{\nu}\omega^{\mu}$ gives $\delta_{\nu}^{\mu}$. As is painfully obvious, my knowledge of manifolds is via a very basic introduction to the physics of spacetime. $\endgroup$ – Peter4075 Oct 15 '15 at 17:53
  • $\begingroup$ @Peter4075 I see what you mean. Unfortunately, I don't know how I can help at the moment. It seems to me that you should pick an introductory book on manifolds and find in it the basic terminology. It may take a few days, but if you really want to understand geometry, it is definitely worth the effort. $\endgroup$ – Amitai Yuval Oct 15 '15 at 18:02

I found this question because I was having the same problem. I had to figure it out and I believe i did. Way too late for the original questioner, but I respond now for people with the same question. I think this is the solution:

If you were reading somebody's book who uses the same notation as Carroll's book on GR, you probably saw three different d's. Carroll uses $d$ for differentials, $\partial{}$ for partial derivatives and the letter d for a one-form. Note the italics $d$ versus regular d. Schutz uses the ~ with the d and writes $\tilde{\text{d}}$ for a one form, and I like that better.

Now, what I think you saw was d$x_i (\frac{\partial{}}{dx^i}) =\delta^j_i\;$ and what it meant was $\;\tilde{\text{d}}x_i (\frac{\partial{}}{dx^i}) =\delta^j_i$ and not $\;dx_i (\frac{\partial{}}{dx^i}) =\delta^j_i$. That is, what was being said is that a one-form basis takes the vector basis and produces $\delta$.

The question now becomes, what is the one form basis? So to be complete, let me say this: The basis denoted $\partial{}_i$ is generated in this way: A one form and a vector produce a invariant scalar. This is bedrock and we can write the invariant scalar as: $d\phi/dS.$ Where $\phi$ is a scalar field and S is a path. The description of the scalar field $\phi$ will change with a change of coordinates as will the description of a path S. But the invariant scalar $d\phi/dS$ will not change with a change of coordinates.

Now, our invariant scalar, the contraction of a vector and a one form, $d\phi/dS$ is: $$\frac{d\phi}{dS} = \frac{\partial{\phi}}{\partial{x}}\frac{dx}{dS} + \frac{\partial{\phi}}{\partial{y}}\frac{dy}{dS}...\tag{EQ 1}$$

In "the modern way" we see a vector as a tangent vector to some curve through some (unnamed) scalar function. That is, in the above equation, from the point of view of a vector, $\phi$ could be any scalar function, so we have: $$a\;vector = \frac{d}{dS} = \frac{\partial{}}{\partial{x}}\frac{dx}{dS} + \frac{\partial{}}{\partial{y}}\frac{dy}{dS}...$$ The right side of the above equation takes the form of a vector and we can see that the vector will have components $\frac{dx^i}{dS}$ and basis $\frac{\partial{}}{\partial{x_i}}$ or $\partial{}_i$ This leaves the definition of a one-form, I always thought of it as: $$d\phi = \frac{\partial{\phi}}{\partial{x}}dx + \frac{\partial{\phi}}{\partial{y}}dy... $$ But that leads back to the same problem stated in the question. If the one-form was written as above, the contraction with the vector written above as $d/dS$ would yield: $$\frac{d\phi}{dS} = \frac{\partial{\phi}}{\partial{x}}\frac{dx}{dS}\frac{\partial{}}{\partial{x}}dx + \frac{d\phi}{dy}\frac{dy}{dS}\frac{\partial{}}{\partial{y}}dy...\;\;(wrong)$$

This cannot be correct. But it was exactly the same result as was found when mixing up the symbols d and $d$ as described in the question. When I didn't know the solution to the first error, this reinforcement of error nearly drove me mad.

So, I believe the correct way to look at the one-form is the same as the modern way of looking at a vector. That is, we see a one-form as the rate of change of a scalar field along some (unnamed) path $$\frac{d\phi}{d} = \frac{\partial{\phi}}{\partial{x}}\frac{dx}{d} + \frac{\partial{\phi}}{\partial{y}}\frac{dy}{d}...$$

The contraction now looks proper as: $$\frac{d\phi}{dS} = \frac{\partial{\phi}}{\partial{x}}\frac{dx}{dS}\frac{\partial{}}{\partial{x}}\frac{dx}{d} + \frac{d\phi}{dy}\frac{dy}{dS}\frac{\partial{}}{\partial{y}}\frac{dy}{d}...$$

and above equation reduces to our definition (EQ 1). Lastly, we can write d$x$ =$dx/d$ or more clearly $\tilde{\text{d}}x$ =$dx/d$

I do still have a problem with Carroll's equation 2.25 which states: $$\text{d}x^u(\partial{}_v)=\frac{\partial{x^u}}{\partial{x^v}} $$ Where we now get $$\text{d}x^u(\partial{}_v)=\frac{d{x^u}}{\partial{x_v}} $$ But this is close enough for me.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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