We introduce in $\mathbb{R}^3$ with the usual Euclidean metric $\langle \cdot, \cdot \rangle$, the connection defined in Cartesian ccoordinates $(x^1, x^2, x^3)$ by $$\Gamma_{jk}^i = \omega \varepsilon_{ijk}$$ where $\omega: \mathbb{R}^3 \rightarrow \mathbb{R}$ is a smooth function and $$\varepsilon_{ijk} = \begin{cases} +1 \quad \text{if }(i,j,k) \text{ is an even permutation of }(1,2,3)\\ -1 \quad \text{se }(i,j,k) \text{ is an odd permutation of }(1,2,3)\\ 0 \quad \text{ otherwise} \end{cases}$$ Show that $\nabla$ is compatible with $\langle \cdot, \cdot \rangle$;

I've taken a local coordinate system $(x_1, x_2, \dots, x_n)$ and used the

We know that a connection $\nabla$ is compatible with a metric $\langle \cdot, \cdot \rangle$ if and only if for every differentiable curve $c: I \rightarrow M$ and every pair $V, \,W$ of vector fields differentiable along $c$ we have $$\frac{d}{dt} \langle V, W \rangle = \left\langle \frac{DV}{dt}, W \right\rangle +\left\langle V, \frac{DW}{dt} \right\rangle, \quad t \in I$$

Let $(x_1(t), x_2(t), x_3(t))$ be the parametrization of $c(t)$ in the local coordinate system, and let $X_i = \frac{\partial}{\partial x_i}$, and write $V$ as $V =\sum_{j=1}^3 v^j X_j$ (and similarly for $W$), where $v_j = v_j(t)$. We can express the covariant derivative as

$$ \begin{align}\frac{DV}{dt} &= \sum_{k=1}^3 \left[ \frac{dv^k}{dt} + \sum_{i,j} v^j \frac{dx_i}{dt} \Gamma_{ij}^k\right] X_k\\ &= \sum_{k=1}^3 \left[ \frac{dv^k}{dt} + \sum_{i,j} v^j \frac{dx_i}{dt} \omega \varepsilon_{ijk}\right] X_k \end{align}$$

I thought about calculating $\frac{d}{dt} \langle V, W \rangle$ and $ \left\langle \frac{DV}{dt}, W \right\rangle +\left\langle V, \frac{DW}{dt} \right\rangle$ in these terms and showing they are equal, but it would be too messy. I think there must be a smart solution that doesn't require too many calculations.

  • 1
    $\begingroup$ You're using Godinho & Natário, right? Great book. $\endgroup$
    – Ivo Terek
    Feb 14 '21 at 22:48
  • $\begingroup$ Yes, I'm using that book. $\endgroup$
    – José
    Feb 15 '21 at 0:06

Since $\nabla\langle \cdot,\cdot\rangle$ is a tensor, it suffices to check that $(\nabla_{\partial_k}\langle\cdot,\cdot\rangle)(\partial_i,\partial_j) = 0$ for all $i,j,k \in \{1,2,3\}$. In other words, we must show that $$\partial_k \langle \partial_i,\partial_j\rangle = \langle \nabla_{\partial_k}\partial_i,\partial_j\rangle + \langle \partial_i, \nabla_{\partial_k}\partial_j\rangle$$for all $i,j,k \in \{1,2,3\}$. The left side is obviously zero, so the situation doesn't look so bad. By the definition of $\nabla$, we have that $$\begin{align} \langle \nabla_{\partial_k}\partial_i,\partial_j\rangle + \langle \partial_i, \nabla_{\partial_k}\partial_j\rangle &= \langle \omega \varepsilon^r_{~ki}\partial_r,\partial_j\rangle + \langle \partial_i, \omega \varepsilon^r_{~kj}\partial_r\rangle \\ &= \omega \varepsilon^r_{~ki} \delta_{rj} + \omega \varepsilon^r_{~kj}\delta_{ir} \\ &= \omega(\varepsilon^j_{~ki} + \varepsilon^i_{~kj}) \\ &= 0,\end{align}$$because since $(j,k,i) \mapsto (i,k,j)$ is an odd permutation, we have that $\varepsilon^j_{~ki} =- \varepsilon^i_{~kj}$.


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.