In a curved $(3+1)$ dimensional spacetime with metric components $g_{\mu \nu}$, the covariant derivative of a $4$ vector $\mathbf V = (V^0, \vec V)$ is given by $$\nabla_\mu~ V^\mu = \frac{1}{\sqrt{-g}}\partial_\mu (\sqrt{-g}~V^\mu).$$

I expect that this relation can also be used to derive the expression for divergence of the $3$ vector $\vec V$ in a flat spatial hypersurface in a curvilinear coordinate system, eg. the cylindrical polar coordinates $(r,\phi,z)$. We will then need to replace the $\sqrt{-g}~$ by $\sqrt g~$ as the metric of the spatial hypersurface has a positive determinant. This will then give $$\vec \nabla \cdot \vec V = \nabla_i V^i = \frac{1}{r}\partial_r(r~V^r) + \partial_\phi V^\phi + \partial_zV^z.$$

However, the actual expression for the divergence of a $3$ vector in cylindrical polar coordinates is $$\vec \nabla \cdot \vec V = \frac{1}{r}\partial_r(r~V^r) + \frac{1}{r}\partial_\phi V^\phi + \partial_zV^z.$$ Can you please point out and explain where I am going wrong?


1 Answer 1


Both identities are correct. The point is that they refer to different bases. The former uses the decomposition, $$\vec{V} = V^r \partial_r + V^\theta \partial_\theta + V^z \partial_z\:,$$ the latter uses the decomposition, $$\vec{V} = V'^r \vec{e}_r + V'^\theta \vec{e}_\theta + V'^z \vec{e}_z\:,$$ where $\vec{e}_i = \frac{1}{\sqrt{g_{ii}}} \partial_{x^i}$ are unit vectors, so that $V'^i = \sqrt{g_{ii}} V^i$ (there is no sum over the repeated index $i$).

  • 3
    $\begingroup$ For the OP, this is a common hangup when graduating from vector calculus to differential geometry (as it is used in general relativity). Vector calculus almost always is taught using unit vectors for a basis, but the natural basis vectors corresponding to a set of coordinates are typically not unit. $\endgroup$
    – Muphrid
    Apr 28, 2014 at 14:19
  • 1
    $\begingroup$ Vector calculus is a silly thing to teach, since you have to unlearn most of it... $\endgroup$ Apr 28, 2014 at 15:18
  • $\begingroup$ Thank you for clarifying this. So, in the first case, $\partial_{x_i}$ are the coordinate basis vectors. But I do not realise how one shall know that the unit vectors in the second basis will be given by $\vec e_i = \frac{1}{\sqrt{g_{ii}}} \partial_{x_i}$. $\endgroup$
    – damaihati
    Apr 29, 2014 at 10:31
  • 1
    $\begingroup$ One has $g\left(\frac{1}{\sqrt{g_{ii}}}\partial_i, \frac{1}{\sqrt{g_{ii}}}\partial_i\right)= \frac{1}{g_{ii}}g(\partial_i,\partial_i)= \frac{1}{g_{ii}} g_{ii}=1$. Therefore $\frac{1}{\sqrt{g_{ii}}}\partial_i$ is a unit vector, tangent to the coordinate $x^i$. Consequently, if $i=r$ it coincides with $\hat{r}= \vec{e}_r$, if $i=\theta$ it coincides with $\hat{\theta}= \vec{e}_\theta$, if $i=z$ it coincides with $\hat{\theta}= \vec{e}_z$. $\endgroup$ Apr 29, 2014 at 11:06
  • $\begingroup$ Sorry, I used spherical coordinates but your referred to cylindrical ones, in my answer $\phi$ should be replaced for $z$. I corrected it. $\endgroup$ Apr 29, 2014 at 11:11

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.