I need to prove the following identity. $\Omega \subset \mathbb{R}^d$ is open and $g$ is a metric field on $\Omega$. Further $\Gamma_{\,kl}^j$ denotes the Christoffel symbol of second kind. $g^{ij}$ is the inverse matrix to $g_{ij}$ and $g$ denotes the determinant.

$$ \Gamma_{\,jl}^j = \frac{\partial}{\partial x^l} \log \sqrt{\vert g \vert} $$

Here is what I have so far: $$ \Gamma^j_{\;jl} = \frac{1}{2} g^{jk} \left( \frac{\partial }{ \partial x^l} g_{kj} + \frac{\partial}{ \partial x^j} g_{kl} - \frac{\partial}{ \partial x^k} g_{jl} \right) = \frac{1}{2} g^{jk} \frac{\partial }{ \partial x^l} g_{kj} $$

The second and the third term cancel each other out. Now as stated here (it's about the sixth equation, after the line starting with "The contracting relations..."

$$ \frac{1}{2} g^{jk} \frac{\partial }{ \partial x^l} g_{kj} = \frac{1}{2g} \frac{\partial}{ \partial x^l} g = \frac{\partial}{ \partial x^l} \log \sqrt{\vert g \vert} $$

I don't get the second last step. I suspect it has got something to do with the Laplace formula and the Cramer Rule. However i seem not to be able to figure it out, the derivation is bugging me.

Any hints are appreciated!


1 Answer 1


First the slick method: write $\bar g_{ik}$ for the value of $g_{ik}$ at the specific $x^l = \bar x^l$ (so $\bar g_{ik}$ is constant). Let $x^l$ be a function of $\epsilon$ that is a perturbation of $\bar x^l$, so that $x^l = \bar x^l$ at $\epsilon=0$.

Let $$h_{ij} = \bar g^{ik} g_{kj} . \tag1$$ Then $h_{ij}$ is a perturbation of the identity: $$h_{ij} = \delta_{ij} + \epsilon \frac{\partial h_{ij}}{\partial \epsilon} + o(\epsilon).$$ Then it is easily computed that $$ \frac{\partial h}{\partial \epsilon} = \frac{\partial h_{ii}}{\partial \epsilon} + o(\epsilon)$$ Differentiate $(1)$ with respect to $\epsilon$ $$ \frac{\partial h_{ii}}{\partial\epsilon} = \bar g^{ik} \frac{\partial g_{ki}}{\partial \epsilon} $$ Also $h = \bar g^{-1} g$. Hence $$ \bar g^{-1} \frac{\partial g}{\partial \epsilon} = \bar g^{ik} \frac{\partial g_{ki}}{\partial \epsilon} + o(\epsilon). $$ In particular, at $\epsilon = 0$ we can remove the overlines, and we get: $$ g^{-1} \frac{\partial g}{\partial \epsilon} = g^{ik} \frac{\partial g_{ki}}{\partial \epsilon}. $$

Next the direct way: so $$g = \sum_\pi\sigma(\pi) \prod_i g_{i\pi(i)}.$$ Here the sum is over permutations $\pi$ of $\{1,\dots,n\}$. Then $$\frac{\partial g}{\partial x^l} = \sum_\pi \sigma(\pi) \sum_i \frac{\partial g_{i\pi(i)}}{\partial x^l} \prod_{j \ne i} g_{j\pi(j)} \\ =\sum_{ik} \frac{\partial g_{ik}}{\partial x^l}\sum_{\pi:\pi(i)=k} \prod_{j\ne i}\sigma(\pi) g_{j\pi(j)} \\ =\sum_{ik} \frac{\partial g_{ik}}{\partial x^l} g \ g^{ki}. $$

  • $\begingroup$ Hello Stephen, thanks for your answer! I'm still not sure whether I completely understood your direct solution. In order for it to work I would need to move the $g g^{ki}$ in the last equation to the left. But it seems they depend on $x^l$ so i would not be allowed to do that. In other words: what is the derivation acting on? $\endgroup$
    – stebu92
    Dec 28, 2013 at 13:28
  • $\begingroup$ When I write $\frac{\partial A}{\partial b} c$, I mean $\left(\frac{\partial A}{\partial b}\right) c$. $\endgroup$ Dec 28, 2013 at 15:21
  • $\begingroup$ $\sum_{\pi:\pi(i)=k} \prod_{j\ne i}\sigma(\pi) g_{j\pi(j)}$ is the formula for the transpose of the adjoint matrix - en.wikipedia.org/wiki/Adjugate_matrix - and hence it is $g g^{ki}$. $\endgroup$ Dec 28, 2013 at 15:25
  • $\begingroup$ I rewrote the slick proof using a more easy to understand notation here: math.stackexchange.com/questions/620771/… $\endgroup$ Dec 28, 2013 at 21:35
  • $\begingroup$ I have one more question: Is it trivial to state that this equation is invariant under diffeomorphisms i.e. $$ \Gamma_{\,jl}^j[\phi] = \frac{\partial}{\partial x^l} \log \sqrt{\vert g[\phi] \vert} $$ where $\phi \in$ Diff$(\Omega)$? If not, how would I go about proving that? My suspicion is that since when I write down the def. of the CS in terms of $g_ij$ two terms cancel out due to the symmetry and I'm left with one, so i can just state it holds for diffeomorphisms. But I'm not sure. $\endgroup$
    – stebu92
    Jan 19, 2014 at 7:38

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.