In Sean Carroll's spacetime and geometry chapter 5 Carroll states the following

In addition we always have another constant of the motion for geodesics: the geodesic equation (together with metric compatibility) implies that the quantity$$\epsilon=-g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^\nu}{d\lambda}\tag{5.55}$$ is constant along the path. (For any trajectory we can choose the parameter $\lambda$ such that $\epsilon$ is a constnat; we are simply noting that this is compatible with affine parameterization along a geodesic.)

I feel like I understand this now but my understanding of formal tensors is still a bit shaky so could you tell if my reasoning is valid and correct any misunderstandings?

  1. From the metric components $g_{\mu\nu}$ we can construct the metric tensor given by $g(x)=g_{\mu\nu}(x)dx^\mu\otimes dx^\nu$. Since it is a tensor it is reparametrization invariant but can still vary over space.
  2. Metric compatibility tells us that actually it doesn't vary over space. Since $\nabla_\alpha g_{\mu\nu}=0$ the metric can be parallel transported to any point in space so it is constant.
  3. If $x(\lambda)$ is a geodesic then from the geodesic equation we have $\nabla_{\dot x}\dot x=0$. This means the direction of the tangent vector $\frac{d}{d\lambda}$ is conserved. By a suitable (reparametrization) of $\lambda$ we get that $\frac{d}{d\lambda}$ is conserved.
  4. Since $g_{\mu\nu}dx^\mu\otimes dx^\nu$ and $\frac{d}{d\lambda}$ are both conserved we have that $$g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^\nu}{d\lambda}=g\left(\frac{d}{d\lambda}\otimes \frac{d}{d\lambda}\right)=\left(g_{\mu\nu}dx^\mu\otimes dx^\nu\right)\left(\frac{d}{d\lambda}\otimes \frac{d}{d\lambda}\right)$$ is also conserved.
  • 1
    $\begingroup$ I think you're a bit confused with what a tensor is. The metric g takes two vectors as arguments, not a (2,0)-tensor. $\endgroup$
    – Krup'a
    Dec 13, 2020 at 14:42
  • $\begingroup$ @Krup'a So would you write 4. as $g\left(\frac{d}{d\lambda},\frac{d}{d\lambda}\right)=\left(g_{\mu\nu}dx^\mu\otimes dx^\nu\right)\left(\frac{d}{d\lambda},\frac{d}{d\lambda}\right)$? So still with a tensor product between the differentials? $\endgroup$ Dec 13, 2020 at 14:46
  • $\begingroup$ As a note, the metric tensor $\textbf{g}$ should be written as $\textbf{g}=g_{\mu\nu}(\textbf{e}^\mu \otimes \textbf{e}^\nu)$. $\endgroup$
    – TaeNyFan
    Dec 13, 2020 at 15:51

2 Answers 2


You don't need any knowledge of tensor calculus to understand this.

What Does $g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^ \nu}{d\lambda}$ Represent?

For a curve $x^\mu(\lambda)$ in a manifold, the quantity $$g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^ \nu}{d\lambda}$$ is the square of the length of the tangent vector $\textbf{t}$ to the curve $x^\mu(\lambda)$ at any point $P$.

To see this, note that at any point $P$ on the curve, the tangent vector $\textbf{t}$ is defined as $$\textbf{t}=\frac{d\textbf{s}}{d\lambda},$$ where $d\textbf{s}$ is the infinitesimal separation vector between point $P$ and a nearby point $Q$ on the curve corresponding to the parameter value $\lambda+d\lambda$.

In a given coordinate system with basis vectors $\textbf{e}_\mu$, we can write $d\textbf{s}=\textbf{e}_\mu dx^\mu$ so that the tangent vector is now $$\textbf{t}=\frac{dx^\mu}{d\lambda}\textbf{e}_\mu.$$ (Note: More explicitly, $dx^\mu \equiv dx^\mu(\lambda)$.)

The square of the length of the tangent vector $\textbf{t}$ is then $$|\textbf{t}|^2=g_{\mu\nu}t^\mu t^\nu=g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^\nu}{d\lambda}.$$

So the quantity $g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^ \nu}{d\lambda}$ being constant throughout the curve $x^\mu(\lambda)$ means that the tangent vector $\textbf{t}$ has a constant length throughout the curve.

Parameterising A Curve

For any curve $x^\mu(\lambda)$, we can paramaterise the curve such that the length of the tangent vector is constant. Note that if we choose the parameter $\lambda$ to be $$\lambda=as+b,$$ where $s$ is the distance measured along the curve and $a$, $b$ are constants, the length of the tangent vector will be constant. This can be shown through $$|\textbf{t}|=\frac{d|\textbf{s}|}{d\lambda}=\frac{ds}{d\lambda}={1\over a},$$ where $1\over a$ is a constant.

Summary: For any curve $x^\mu(\lambda)$, we can always parameterise it such that the length of the tangent vector is constant throughout the curve. The length of the tangent vector is given by $|\textbf{t}|=(g_{\mu\nu}\frac{dx^\mu}{d\lambda}\frac{dx^ \nu}{d\lambda})^{1\over2}$.


  1. Hobson, Efstathiou & Lasenby General Relativity: An Introduction for Physicists pg. 75

This is simply a choice of parametrization of the geodesic $x^\mu(\lambda)$. If we were in a Euclidean-signature manifold, $\lambda$ would be proportional to the arc length along the curve. Here, for a timelike geodesic, it would be proportional to the proper time along the curve.

  • $\begingroup$ So would you say 3. is correct? Is the direction of $d/d\lambda$ conservered for a general geodesic? $\endgroup$ Dec 13, 2020 at 14:06
  • $\begingroup$ It depends on what you mean by $d/d\lambda$. Do you mean $(d x^\mu/d \lambda) \partial_\mu$ with $\partial_\mu$ considered as a tangent space basis vector? Certainly geodesics are autoparallels. I don't like saying that "direction is conserved" as directions at different points are not comparable. $\endgroup$
    – mike stone
    Dec 13, 2020 at 14:14
  • $\begingroup$ To me $d/d\lambda$ and $(dx^\mu/\lambda)\partial_\mu$ are two expressions of the same thing. I don't know what autoparallels are. $\endgroup$ Dec 13, 2020 at 14:17
  • $\begingroup$ But I have to agree that saying directions are conserved is a bit of a vague statement. $\endgroup$ Dec 13, 2020 at 14:19
  • $\begingroup$ I mean that parallel trasnporting the tangent along the geodesic takes you to the tangent at the new point. If you parallel transport to the same point along a different curve you will get a different vector. So directions at different points are not comparable. The attempted comparison depends on how you try to do it. $\endgroup$
    – mike stone
    Dec 13, 2020 at 14:19

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.