2
$\begingroup$

First off, I did look through some other questions:

Covariant Derivative of Metric Tensor

Why is the covariant derivative of the metric tensor zero?

https://math.stackexchange.com/q/2174588/

But they either give the covariant derivative as:

$\nabla_{\rho} g_{\mu\nu}=\frac{\partial g_{\mu\nu}}{\partial x^{\rho}} - g_{\mu\sigma}\Gamma^{\sigma}_{\nu\rho}-g_{\tau\nu}\Gamma^{\tau}_{\mu\rho}$

By basis of the connection being chosen so that the covariant derivative of the metric is zero.

However, is this relation not derived from the tensor product rule of the covariant derivative, such that one can find:

$\nabla_{\rho}( g_{\mu\nu} \vec{e}^{\mu} \otimes \vec{e}^{\nu})=(\frac{\partial g_{\mu\nu}}{\partial x^{\rho}} - g_{\mu\sigma}\Gamma^{\sigma}_{\nu\rho}-g_{\tau\nu}\Gamma^{\tau}_{\mu\rho})\vec{e}^{\mu} \otimes \vec{e}^{\nu}$?

It seems to relate to the third resource I've linked, since I don't think it makes sense to say the covariant derivative of a tensor, that is, the tensor components and the tensor product of basis vectors/covectors, equals the covariant derivative of the COMPONENTS times the tensor product of basis stuff, since the covariant derivative is supposed to describe curved space by affecting the basis stuff, which it does, so we can't factor it out as that viewpoint would imply.

In the third linked source, it said to view the first formula as a component of the covariant derivative, not the covariant derivative of the component, which I can get behind.

So, in summa, why is it said:

$\nabla_{\rho} g_{\mu\nu}=\frac{\partial g_{\mu\nu}}{\partial x^{\rho}} - g_{\mu\sigma}\Gamma^{\sigma}_{\nu\rho}-g_{\tau\nu}\Gamma^{\tau}_{\mu\rho}$

And not:

$\nabla_{\rho}( g_{\mu\nu} \vec{e}^{\mu} \otimes \vec{e}^{\nu})=(\frac{\partial g_{\mu\nu}}{\partial x^{\rho}} - g_{\mu\sigma}\Gamma^{\sigma}_{\nu\rho}-g_{\tau\nu}\Gamma^{\tau}_{\mu\rho})\vec{e}^{\mu} \otimes \vec{e}^{\nu}$?

And if it is said as the latter, then we say:

$\nabla_{\rho}( g_{\mu\nu} \vec{e}^{\mu} \otimes \vec{e}^{\nu})=0$

And not:

$\nabla_{\rho}g_{\mu\nu}=0$

Since it might be that instead:

$\nabla_{\rho}g_{\mu\nu}=\frac{\partial g_{\mu\nu}}{\partial x^{\rho}}$?

I suppose part of my motivation for this question is in the Einstein-Hilbert Action, where we use the virtue of covariant derivative of the metric being zero to factor it into a covariant derivative so that we can prove one equation is zero, so that the EFEs pop out.

Addendum Auctoris: Okay, I made a fe- a lot of index oopsies, but that should be taken care of now. Thanks for pointing that out.

Addendum II: It seems that I'm bad at LaTeX, or I forget things easily.

$\endgroup$
  • $\begingroup$ I just realized the in summa took up half the post. I guess that was unnecessary then. $\endgroup$ – M. V. Jun 29 at 14:52
  • 1
    $\begingroup$ you need to check your question, for example how you expand $g$ in components and basis, your expression is not covariant and this matters. $\endgroup$ – Nelson Vanegas A. Jun 29 at 15:10
  • $\begingroup$ Oops, switched subscripts and superscripts. Hold on. $\endgroup$ – M. V. Jun 29 at 15:16
  • $\begingroup$ So taking $g$ as $g_{\mu\nu} \vec{e}^{\mu} \otimes \vec{e}^{\nu}$, did you mean the expression, the right side of the second equation prior to my edit, does not transform covariantly? I hope I fixed that. $\endgroup$ – M. V. Jun 29 at 15:23
  • $\begingroup$ Forgot to make partial derivatives as well. $\endgroup$ – M. V. Jun 30 at 19:25
2
$\begingroup$

For this derivation, we first need to calculate the partial derivative of the covarinat metric tensor (which can be expressed, as the dot product of two covariant basis vectors). \begin{align} \partial_{\omega}{g_{{\mu}{\nu}}} =\partial_{\omega}\langle{\boldsymbol{\varphi}_{\mu},\boldsymbol{\varphi}_{\nu}}\rangle= \langle{\partial_{\omega}\boldsymbol{\varphi}_{\mu},\boldsymbol{\varphi}_{\nu}}\rangle+ \langle{\boldsymbol{\varphi}_{\mu},\partial_{\omega}\boldsymbol{\varphi}_{\nu}}\rangle \end{align} By the definition of the covariant derivative, acting on a vector field: \begin{align} \nabla_{\omega}\mathbf{F}=\boldsymbol{\varphi}_{\mu}\nabla_{\omega}F^{\mu} \end{align} According to the product rule and the fact that any vector can be expressed as $\mathbf{F}=\boldsymbol{\varphi}_{\mu}F^{\mu}$ this implies that, the covariant derivative of the basis vectors $\nabla_{\omega}\boldsymbol{\varphi}_{\mu}$ is zero. Now, we can compute the partial derivative of the covariant basis vector. \begin{align} 0=\nabla_{\omega}\boldsymbol{\varphi}_{\mu} =\partial_{\omega}\boldsymbol{\varphi}_{\mu}- \Gamma^{\alpha}_{{\mu}{\omega}}{\,}\boldsymbol{\varphi}_{\alpha} \Longrightarrow \partial_{\omega}\boldsymbol{\varphi}_{\mu} =\Gamma^{\alpha}_{{\mu}{\omega}}{\,}\boldsymbol{\varphi}_{\alpha} \end{align} We can now write the partial derivative of the metric tensor as follows: \begin{align} \partial_{\omega}{g_{{\mu}{\nu}}} =\Gamma^{\alpha}_{{\mu}{\omega}}\langle{\boldsymbol{\varphi}_{\alpha},\boldsymbol{\varphi}_{\nu}}\rangle +\Gamma^{\alpha}_{{\omega}{\nu}}\langle{\boldsymbol{\varphi}_{\mu},\boldsymbol{\varphi}_{\alpha}}\rangle =g_{{\alpha}{\nu}}\Gamma^{\alpha}_{{\mu}{\omega}}+g_{{\mu}{\alpha}}\Gamma^{\alpha}_{{\omega}{\nu}} \end{align} But the covariant derivative of the metric tensor was: \begin{align} \nabla_{\omega}g_{{\mu}{\nu}}=\partial_{\omega}{g_{{\mu}{\nu}}}-g_{{\alpha}{\nu}}\Gamma^{\alpha}_{{\mu}{\omega}}-g_{{\mu}{\alpha}}\Gamma^{\alpha}_{{\omega}{\nu}}=0 \end{align} And therefore, it has to be zero. Q.E.D.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ I'm not familiar with the definition on the second line. I learned the covariant derivative as a regular total/partial derivative of a vector field's components and basis and using product rule. From that I was taught that $\nabla_{\omega} \phi_{\mu} = \Gamma^{\alpha}_{\omega \mu} \phi_{\alpha}$, where $\phi_{\mu}$ is a basis vector, I'm afraid I was not taught that $\partial_{\omega} \phi_{\mu} - \Gamma^{\alpha}_{\omega \mu} \phi_{\alpha}$, although I've seen the covariant derivative being calculated as this form online. Does the third line use that definition of the covariant derivative? $\endgroup$ – M. V. Aug 21 at 18:19
  • 1
    $\begingroup$ It uses the rule common rule of the covariant derivative $\nabla_{\omega}F_{\mu}=\partial_{\omega}F_{\mu}-\Gamma^{\alpha}_{{\mu}{\omega}}F_{\alpha}$ which can be seen here en.wikipedia.org/wiki/Covariant_derivative for example. $\endgroup$ – Great Stokes Aug 21 at 21:25
  • 1
    $\begingroup$ Ah, I see the definition, however in that definition, isn’t the partial taking the component of the vector field as an argument? If we take phi to be a basis vector, then wouldn’t the partial be of the vector’s components in the phi basis (which I guess is constant so partial of zero?), not the vector itself? $\endgroup$ – M. V. Aug 21 at 21:34
  • 1
    $\begingroup$ The covariant derivative (like the partial derivative) doesn't need to only act on the components of vectors or tensors itself for this rule to apply. It only needs to have one (or multiple) co- or contravariant indices. It can act on any tensor or vector as long as these conditions are satisfied. You may remember that in the euclidian coordinate space, the partial derivative of the basis vectors $\mathbf{e}_{x}$ are zero because they are of constant value. The co- and contravariant basis vectors in the generalized coordinate system however, are not. $\endgroup$ – Great Stokes Aug 21 at 22:03
  • 1
    $\begingroup$ Ah yes, I agree that the covariant derivative would be applied as long as at least one co- or contravariant index is present in the argument. However, I think the issue lies with how to apply the covariant derivative. Sorry if it seems like I'm just restating my second comment, but the definition of the covariant derivative in your first comment seems to only be if the argument of the covariant derivative is a vector field (which I admit a basis vector to be) with components (as measured against the basis of said field) that change in space. Yes, I agree that in the generalized system, the... $\endgroup$ – M. V. Aug 21 at 22:45
4
$\begingroup$

It's just notation. We take $\nabla_\rho g_{\mu\nu}$ to mean $(\nabla g)_{\rho\mu\nu}$, that is, the component of the covariant derivative of the metric tensor, because it's convenient. After all, we already have the notation $\partial_\rho g_{\mu\nu}$ for the derivative of the components.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

So, $g$ is a tensor field which in a chart with the appropriate base could be written as $$ g = g_{\mu \nu} \, \mathrm{d}x^{\mu} \otimes \mathrm{d}x^{\nu}.$$

Now, the covariant derivative has a Leibniz rule (it is indeed a derivative), when applied to $g$ acts as $$\nabla_{\rho}g = \nabla_{\rho}g_{\mu \nu} \, \mathrm{d}x^{\mu} \otimes \mathrm{d}x^{\nu} + g_{\mu \nu}\, (\nabla_{\rho}\mathrm{d}x^{\mu}) \otimes \mathrm{d}x^{\nu} + g_{\mu \nu}\, \mathrm{d}x^{\mu} \otimes (\nabla_{\rho} \mathrm{d}x^{\nu}).$$

Since by components $g_{\mu \nu} = g_{\mu \nu}(x)$ is a scalar function and $(\nabla_{\rho}\mathrm{d}x^{\mu}) = - \Gamma^{\mu}_{\,\rho \sigma} \mathrm{d}x^{\sigma},$ then you recover the expression you want. Have in mind that by definition the affine connection takes your tensor and a vector and maps them onto another tensor:

$$\nabla_{\mathrm{_X}} g = \mathrm{X}^{\rho} \, \nabla_{\rho} g,$$ (where you plug in the expression you found). I hope you can see now why your second equation is wrong.

Finally, for different reasons one demands that the metric is covariantly constant $$\nabla_{\mathrm{_X}} g = 0,$$ mainly to obtain a metric compatible connection (and a torsion-free metric), something that is very useful in general relativity.

Note that $\nabla_{\mathrm{_X}} g = 0,$ would lead to $$\nabla_{\rho}\,[ g(Y,Z)] = 0$$ when you parallel transport the vectors the along a geodesic with a metric compatible (Levi-Civita) connection.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Sorry, it seems that I've made far too many mistakes in my LaTeX formatting for my equations. Now it should line up with your second equation. Did I need to indicate field components? Because it seems that I can just take the covariant derivative along the basis of the vector field $X=X^{\rho} \frac{\partial}{\partial x^{\rho}}$, or was my equation wrong solely because of the errors I kept making in formatting? $\endgroup$ – M. V. Jun 30 at 19:39
  • 1
    $\begingroup$ My take is that the second equation kind of emphasizes that there is second rank tensor only involved and in a way forgets that there has to be vector field in whose direction we consider the covariant derivative, and that the $\rho$ index gives it a richer structure. One usually drops the $X^{\rho}$ for sure... but it does not mean that formally is there. $\endgroup$ – Nelson Vanegas A. Jun 30 at 21:57
  • $\begingroup$ Ah, so basically my equation can be easily misconstrued so that one may neglect fundamental properties of the covariant derivative on this manifold? $\endgroup$ – M. V. Jun 30 at 22:03
  • 1
    $\begingroup$ Yes, I believe so... others might disagree. $\endgroup$ – Nelson Vanegas A. Jul 1 at 2:27

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.