In lecture 3 (about 97 min into the lecture) of Leonard Susskind's general relativity course, he suggests finding the Riemann curvature tensor in terms of the Christoffel symbols as an exercise.

I started from $$D_r D_s V_n - D_s D_r V_n = R_{rs}^t{_n} V_t$$ as a definition of the Riemann curvature tensor.

The covariant derivative is defined by $$D_sV_n = \partial_sV_n - \Gamma_s^u{_n}V_u$$ and the Christoffel symbols are defined in terms of the metric by $$\Gamma_s^u{_n} = \dfrac12 g^{ut}\left( \partial_sg_{tn} + \partial_n g_{st} - \partial_tg_{sn}\right).$$

I'm supposed to find $$R_{rs}^t{_n} = -\partial_r\Gamma_s^t{_n} + \partial_s\Gamma_r^t{_n} + \Gamma_r^w{_n}\Gamma_s^t{_w} - \Gamma_s^w{_n}\Gamma_r^t{_w}.$$

I get that this works, but only if $$\Gamma_r^w{_n}\partial_sV_w = \Gamma_s^w{_n}\partial_rV_w.$$

I'm stuck showing this. I tried expanding the Christoffel symbols out into the metric, but I wind up trying to show things like $$\partial_r g_{tn}\partial_s = \partial_s g_{tn}\partial_r,$$ which I don't see how to justify.

(It may also be that I'm making algebra errors because I'm unfamiliar with tensor calculus, so any advice you have would be helpful here.)

  • 1
    $\begingroup$ The identity is false. Try e.g. acting with it on a function that only depends on one of the coordinates, say, $s$, with $s\neq r$. The r.h.s. vanishes, while the l.h.s. does not. $\endgroup$ Jul 20, 2019 at 15:57
  • $\begingroup$ So is the setup of the calculation right? $\endgroup$ Jul 20, 2019 at 16:04
  • $\begingroup$ I think so, but I didn't double check all the indices. The derivation is pretty standard though, so you should be able to find it online. $\endgroup$ Jul 20, 2019 at 16:09
  • $\begingroup$ Okay, I found this page: einsteinrelativelyeasy.com/index.php/general-relativity/… But it seems to say that the identity I asked about is true. (near the bottom, just below the first blue box) $\endgroup$ Jul 20, 2019 at 16:53
  • 1
    $\begingroup$ I'm not sure I follow; shouldn't I be able to choose $v^i$ arbitrarily to show that if the sum is zero, each individual operator is also zero? I don't know which identity you mean by "the quoted identity". Are you saying that the page I linked above is wrong? $\endgroup$ Jul 20, 2019 at 18:04

1 Answer 1


I'm over a year late to answer this question, but here goes. I think you must have misapplied some identities or got lost in the weeds during your derivation. This has to do with the torsion-freeness of the connection. We have, using the formulas for the covariant derivative of a rank 2 tensor:

\begin{align*} D_rD_sV_n=&D_r(\partial_s V_n-\Gamma^u_{sn}V_u)\\ &=\partial_r \partial_s V_n-(\partial_r \Gamma^u_{sn})V_u-\Gamma^u_{sn}\partial_r V_u\\ &-\Gamma^a_{rs}(\partial_a V_n-\Gamma^u_{an}V_u)\\ &-\Gamma^a_{rn}(\partial_s V_a-\Gamma^u_{sa}V_u) \end{align*}

We want to antisymmetrize with respect to r and s, and I want to use bracket notation $V_{[a} U_{b]}\equiv \frac{1}{2}(V_a U_b-V_b U_a)$ for brevity. To do this without ambiguity, we should try to rearrange all our indices to put r and s next to each other. Using the symmetry of the Christoffel symbol in the lower two indices (a consequence of torsion freeness)...

\begin{align*} D_rD_sV_n=&-(\partial_r \Gamma^u_{sn})V_u-\Gamma^u_{ns}\partial_r V_u\\ &-\Gamma^a_{nr}\partial_s V_a+\Gamma^a_{nr}\Gamma^u_{sa}V_u\\ &+\text{symmetric stuff} \end{align*} Then we have: $$ D_{[r}D_{s]}V_n=-(\partial_{[r} \Gamma^u_{s]n})V_u-\Gamma^u_{n[s}\partial_{r]} V_u-\Gamma^a_{n[r}\partial_{s]} V_a+\Gamma^a_{n[r}\Gamma^u_{s]a}V_u$$

The first and last terms give the Riemann tensor up to factors of two. The middle two terms add up to zero because by definition, $\Gamma^u_{n[s}\partial_{r]}=-\Gamma^u_{n[r}\partial_{s]}$.

What did you do wrong?

You might have forgot to properly distribute the partial derivative across $\partial_r(\Gamma^u_{sn}V_u)$. If you work with the convention that $\partial_r$ acts on everything to the right of it, you have to be very careful about it. If you find yourself making mistakes or overusing parentheses, you can use the symbol $\overrightarrow\partial$ to mean "acts on everything to the right" and the symbol $\partial$ to mean "acts on the thing immediately to the right". So as an operator equation $\overrightarrow\partial_x f=\partial_xf+f\overrightarrow\partial_x$.

You also might have misapplied the covariant derivative of a 2-tensor. We have:

$$D_a (V_{bc})=\partial_a V_{bc}-\Gamma^u_{ab}V_{uc}-\Gamma^u_{ac}V_{bu}$$

There are also plenty of other bookkeeping / index mistakes that could be made here.

Other References

pg. 82 / equation 3.66 of Sean Carroll's lecture notes covers this, making no assumptions about torsion.


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.