For simplicity, I only calculated half of the commutator. I didn't leave everything in components because I'm uncomfortable considering (I previously messed up the indices. The following is the corrected version by Sean) \begin{equation} \begin{split} \nabla_{V^i e_i}(\nabla_{W^je_j}X^k e_k)=\nabla_{V^i e_i}(W^j (\partial_j X^k+\Gamma_{j\rho}^k X^\rho)e_k)=V^{i}{[\partial_i(W^j(\partial_j X^k+\Gamma^k_{j\rho}X^\rho)+\Gamma^k_{i\sigma}W^j(\partial_jX^\sigma+\Gamma^\sigma_{j\rho}X^\rho)]e_k}=V^i[(\partial_i W^j)(\partial_j X^k+\Gamma^{k}_{j\rho}X^\rho)+W^j\partial_i(\partial_j W^k+\Gamma^k_{j\rho}X^\rho)+\Gamma^k_{i\sigma}W^j(\partial_jX^\sigma+\Gamma^\sigma_{j\rho}X^\rho)]e_k \end{split} \end{equation} (The first term in the last line should be zero, if the vector fields in nabla are simply elements of the coordinate basis $\partial_i$). But this doesn't give me the torsion term, i.e. there's no minus term that arises when you consider $\nabla_\mu X^\nu$ as a (1,1) tensor. Like here,Sean Carroll

we have $-\Gamma^\lambda_{\mu\nu}\partial_\lambda V^\rho-\Gamma_{\mu\nu}^\lambda \Gamma_{\lambda\sigma}^{\rho}V^{\sigma}$ in the third line. Why would they give different result? What's wrong with my derivation.

  • 1
    $\begingroup$ Please use MathJax, not images of math. $\endgroup$
    – Ghoster
    Commented Feb 18 at 21:48
  • $\begingroup$ Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. $\endgroup$
    – Community Bot
    Commented Feb 18 at 22:04

2 Answers 2


I prefer the notation $\nabla_X$ to the apparently more explicit $X^\mu\nabla_\mu$ because in the presence of torsion there is a dangerous ambiguity lurking in the notation $\nabla_\mu$. To see this recall that the curvature definition $$ [\nabla_X,\nabla_Y]Z- \nabla_{[X,Y]}Z= {\bf R}(X,Y) Z $$ holds even when the connection has torsion. Now take $X\to \partial_\mu$ and $Y\to \partial_\nu$, i.e. $X^\lambda =\delta^\lambda_\mu$, $Y^\lambda =\delta^\lambda_\nu$ so that we are tempted to write $$ \nabla_{\partial_\mu}\stackrel{}{=}\delta^\lambda_\mu \nabla_\lambda\stackrel{?}{=} \nabla_\mu, \quad \nabla_{\partial_\nu}=\delta^\lambda_\nu \nabla_\lambda \stackrel{?}{=}\nabla_\nu. $$ Now, remembering that $[\partial_\mu,\partial_\nu]=0$, our curvature definition gives us
$$ ([\nabla_\mu,\nabla_\nu] Z)^\lambda \stackrel{?}{=} Z^\alpha{R^\lambda}_{\alpha \mu\nu}. $$ If, however, we include a Christoffel symbol ${\Gamma^\alpha}_{\nu\mu}$ when $\nabla_\mu$ acts from the left on the index $\nu$ on $\nabla_\nu$ and similarly a ${\Gamma^\alpha}_{\mu\nu}$ when $\nabla_\nu$ is on the left and acting on the index $\mu$, we end up with $$ ( [\nabla_\mu,\nabla_\nu] Z)^\lambda \stackrel{?}{=} Z^\alpha{R^\lambda}_{\alpha \mu\nu}- {T^\sigma}_{\mu\nu} \nabla_\sigma Z^\lambda. $$ Which is equation is correct? With the last version, remembering that
$$ \nabla_X Y-\nabla_Y X= [X,Y] + T(X,Y), $$ we can use Liebnitz rule to find
$$ ( [X^\mu\nabla_\mu,Y^\nu \nabla_\nu]Z)^\lambda = (\nabla_X Y-\nabla_YX)^\sigma \nabla_\sigma Z^\lambda +X^\mu Y^\nu (Z^\alpha{R^\lambda}_{\alpha \mu\nu}- {T^\sigma}_{\mu\nu} \nabla_\sigma Z^\lambda)\nonumber\\ = Z^\alpha{R^\lambda}_{\alpha \mu\nu}X^\mu Y^\nu + [X,Y]^\sigma \nabla_\sigma Z^\lambda, \nonumber $$ which reproduces the curvature definition for general vector fields $X$, $Y$. Both formulae are correct in their own ways. The difference lies in whether we regard the $\mu$ on $\nabla_\mu$ as a shorthand label specifying the co-ordinate basis vector ${\partial}_\mu$, or as an index on the numerical component of a covector.


For arbitrary vectors $V,W$ and $X$, we have \begin{equation} \begin{split} \nabla_V \nabla_W X &= \nabla_{V^\lambda \partial_\lambda}(\nabla_{W^\tau \partial_\tau}X^k e_k) \\ &=\nabla_{V^\lambda \partial_\lambda}(W^\tau (\partial_\tau X^k+\Gamma_{\tau j}^k X^j)e_k) \\ &=V^\lambda[\partial_\lambda (W^\tau(\partial_\tau X^k+\Gamma^k_{\tau j}X^j))+\Gamma^k_{\lambda i}W^\tau(\partial_\tau X^i+\Gamma^i_{\tau j}X^j)]e_k \\ &=V^\lambda[ (\partial_\lambda W^\tau)(\partial_\tau X^k+\Gamma^k_{\tau j}X^j) + W^\tau \partial_\lambda (\partial_\tau X^k+\Gamma^k_{\tau j}X^j) \\ &\qquad +\Gamma^k_{\lambda i}W^\tau(\partial_\tau X^i+\Gamma^i_{\tau j}X^j)]e_k \end{split} \end{equation} Now if $V=\partial_\mu$, $W=\partial_\nu$, we have $V^\lambda=\delta^\lambda_\mu$, $W^\tau=\delta^\tau_\nu$. Substituting $V^\lambda$ and $W^\tau$ into the above equation, we get \begin{equation} \begin{split} \nabla_{\partial_\mu} \nabla_{\partial_\nu} X &\equiv \nabla_{\delta^\lambda_\mu \partial_\lambda}(\nabla_{\delta^\tau_\nu \partial_\tau}X^k e_k) \\ &= [ \partial_\mu \partial_\nu X^k + \partial_\mu \Gamma^k_{\nu j}X^j + \Gamma^k_{\nu j}\partial_\mu X^j + \Gamma^k_{\mu i}\partial_\nu X^i + \Gamma^k_{\mu i}\Gamma^i_{\nu j}X^j]e_k. \end{split} \end{equation} Then \begin{equation} \begin{split} \nabla_{\partial_\mu} \nabla_{\partial_\nu} X - \nabla_{\partial_\nu} \nabla_{\partial_\mu} X &= [\partial_\mu \partial_\nu X^k + \partial_\mu \Gamma^k_{\nu j}X^j + \Gamma^k_{\nu j}\partial_\mu X^j + \Gamma^k_{\mu i}\partial_\nu X^i + \Gamma^k_{\mu i}\Gamma^i_{\nu j}X^j]e_k \\ &\quad - [ \partial_\nu \partial_\mu X^k + \partial_\nu \Gamma^k_{\mu j}X^j + \Gamma^k_{\mu j}\partial_\nu X^j + \Gamma^k_{\nu i}\partial_\mu X^i + \Gamma^k_{\nu i}\Gamma^i_{\mu j}X^j]e_k \\ &= ( \partial_\mu \Gamma^k_{\nu j} - \partial_\nu \Gamma^k_{\mu j} + \Gamma^k_{\mu i}\Gamma^i_{\nu j} - \Gamma^k_{\nu i}\Gamma^i_{\mu j} ) X^j e_k \\ &= {R^k}_{j\mu\nu} X^{j} e_k. \end{split} \end{equation}

$\nabla_\mu \nabla_\nu X^k$ is defined as the component of the $(1,2)$-tensor: \begin{equation} \nabla\nabla X = (\nabla_\mu \nabla_\nu X^k) \mathrm{d}x^\mu \otimes \mathrm{d}x^\nu \otimes \partial_k. \end{equation} We have \begin{equation} \nabla_\mu \nabla_\nu X^k - \nabla_\nu \nabla_\mu X^k = {R^k}_{j\mu\nu} X^j - 2 \Gamma^\lambda_{[\mu\nu]} \nabla_\lambda X^k. \end{equation} But $\nabla_{\partial_\mu} \nabla_{\partial_\nu} X$ is a vector. So $(\nabla_{\partial_\mu} \nabla_{\partial_\nu} X)^k \neq \nabla_\mu \nabla_\nu X^k$. And \begin{equation} [\nabla_\mu, \nabla_\nu] X^k \neq ([\nabla_{\partial_\mu}, \nabla_{\partial_\nu}] X)^k. \end{equation}

  • $\begingroup$ Thank you! But that still wouldn't give the minus terms right? $\endgroup$ Commented Feb 19 at 4:42
  • $\begingroup$ I edited my reply. $\endgroup$
    – Sean
    Commented Feb 19 at 13:48
  • $\begingroup$ Thank you so much! I now see my problem is with the interpretation of \nabla_mu X^k is DEFINED to be a (1,1) tensor which is not at all the same thing as (\nabla_mu X)^k. $\endgroup$ Commented Feb 19 at 18:39
  • $\begingroup$ A follow up question. How does the (1,1) tensor (\nabla_mu X^nu) dx^mu \otimes \partial_nu acts on V\timesV*? $\endgroup$ Commented Feb 19 at 18:59
  • $\begingroup$ $(\nabla_\mu X^\nu) dx^\mu \otimes \partial_\nu (V,\omega) = (\nabla_\mu X^\nu) dx^\mu(V) \omega(\partial_\nu) = (\nabla_\mu X^\nu) V^\mu \omega_\nu$. $\endgroup$
    – Sean
    Commented Feb 20 at 1:58

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.