This is a formula I have seen stated in two different places in a relativistic context, but I do not know the proof and seem unable to derive it on my own. $E$ is the energy of a free particle, $v$ is its speed and $p$ its momentum.


The fundamental energy-momentum relation in special relativity is $E^2=p^2+m^2$ (let’s ignore factors of $c$, since they will factor out in the end — that is, let’s work in natural units).

Taking the derivative with respect to $p$ on both sides gives

$$2E\frac{\partial E}{\partial p}=2p,$$

since $m$ is a constant. Thus,

$$\frac{\partial E}{\partial p}=\frac{p}{E}.$$

Now, noting that $E=m\gamma$ and $p=mv\gamma$, where $\gamma=1/\sqrt{1-v^2}$, the result is proven.

In the last step, we could have instead used the fact that the four-momentum is given by


So that $E=m\,\mathrm{d}t/\mathrm{d}\tau$ and $p=m|\mathrm{d}\vec{x}/\mathrm{d}\tau|$ (this only works in the massive case of course, but proving your identity is trivial in the massless case, since $E=p$). Thus,

$$\frac{\partial E}{\partial p}=\frac{p}{E}=\frac{|\mathrm{d}\vec{x}/\mathrm{d}\tau|}{\mathrm{d}t/\mathrm{d}\tau}=\left|\frac{\mathrm{d}\vec{x}}{\mathrm{d}t}\right|,$$

which is the definition of the speed. Note that in the second step we used the chain rule.

I hope this helps!

  • $\begingroup$ Doing these manipulations using $\gamma$ and normalized velocity vectors is needlessly complicated and, as you note, forces you to handle the massless case separately. It's both simpler and more general if you just prove $v=p/E$ from the fact that the energy-momentum vector has to be parallel to the velocity vector (which can be unnormalized). (If it weren't parallel, then we would have a violation of rotational symmetry.) $\endgroup$ – user4552 Oct 14 '19 at 23:05

We have $E=\gamma mc^2$ (relativistic total energy) and $p=\gamma mv$, where $\gamma=1/\sqrt{1-v^2/c^2}$. Therefore, we can write:

$$E=\frac{pc^2}{v} \space,$$

differentiating both sides WRT $p$ implies:

$$\frac{dE}{dp}=c^2\left(\frac{v-p\frac{dv}{dp}}{v^2}\right)=\frac{c^2}{v^2}\left(v-\frac{p}{\frac{dp}{dv}}\right) \space. \tag{*}$$

Moreover, we have:

$$\frac{dp}{dv}=\frac{d(\gamma mv)}{dv}=m\frac{d}{dv}\left(\frac{v}{\sqrt{1-\frac{v^2}{c^2}}}\right)=\frac{m}{\left(1-\frac{v^2}{c^2}\right)^\frac{3}{2}}={\gamma^3}m \space. \tag{**}$$

Substituting both $p=\gamma mv$ and its derivative [Eq. $(**)$] into Eq. $(*)$ infers:

$$\frac{dE}{dp}=\frac{c^2}{v^2}\left(v-\frac{\gamma mv}{\gamma^3 m}\right)=\frac{c^2}{v}\left(1-\frac{1}{\gamma^2}\right)=\frac{c^2}{v}\left(1-1+\frac{v^2}{c^2}\right)=v \space.$$

  • $\begingroup$ This is much more complicated than necessary, and only works in the case where $m>0$. $\endgroup$ – user4552 Oct 14 '19 at 23:09

I use the signature convention $(+,-)$.

Any spacetime-displacement vector can be written as $$\tilde T=a\hat t+ \tilde s,$$ where $\hat t$ is unit and future-timelike and $\tilde s\cdot\hat t=0$ (that is, $\tilde s$ is purely spacelike to the observer with 4-velocity $\hat t$). Assume $a \gt 0$. Since $a$ is the "temporal displacement" and $\tilde s$ is the "spatial displacement", then we can define the "spatial velocity according to the $\hat t$-observer" as $$\tilde v\equiv \displaystyle\frac{\tilde s}{a}.$$ So, $\tilde T=a(\ \hat t+\tilde v\ )$. For non-spacelike vectors ($\tilde T\cdot\tilde T \geq 0$), we have $1=\left\|\hat t\right\|^2=\hat t\cdot \hat t\ \ \geq\ -\tilde v \cdot \tilde v=\left\|\vphantom{\hat t}\tilde v\right\|^2=v^2$.

Similarly, any [future-pointing non-spacelike] energy-momentum vector can similarly be decomposed into $$\tilde P=E\hat t+ \tilde p,$$ where $E$ is the relativistic-energy and $\tilde p$ is the relativistic momentum. When $\tilde T$ is chosen to be parallel to $\tilde P$ (so the particle with this constant momentum $\tilde P$ will traverse this displacement $\tilde T$), we have $$\tilde v =\displaystyle\frac{\tilde p}{E},$$ as suggested by @Ben Crowell .

Now, as @Bob Knighton does, \begin{align} m^2 \equiv \tilde P\cdot \tilde P &=E^2(\ \hat t\cdot \hat t\ )+(\ \tilde p\cdot \tilde p\ )\\ &=E^2 - p^2 \end{align} and, by differentiation, \begin{align} 0 &= 2 E \frac{dE}{dp} - 2p, \end{align} so that \begin{align} \frac{dE}{dp} = \frac{p}{E} =v, \quad \mbox{where the last equality uses the relation above}. \end{align} This should be in line with @Ben Crowell , who hints that a good proof should be able to cover both the timelike ($m^2>0$) and lightlike ($m^2=0$) cases.

  • $\begingroup$ Sweet coordinate-independent presentation. At the point where you say $\tilde{v}=\tilde{p}/E$, it seems to me that you ought to invoke the fact that the energy-momentum is parallel to the displacement. Otherwise it's not obvious why this $\tilde{v}$ is the same as the earlier $\tilde{v}$. $\endgroup$ – user4552 Oct 15 '19 at 14:41
  • $\begingroup$ You start off with a displacement that can be in any direction regardless of light-cone, then make some remarks about norms in the non-spacelike case, then make an argument about the $\tilde{P}$ vector that's restricted to future non-spacelike $\tilde{P}$. It's not clear to me where you ever used the facts about the norms (maybe you didn't?), or whether the restriction to future non-spacelike $\tilde{P}$ was ever necessary. Is there something in the argument that fails for tachyons? $\endgroup$ – user4552 Oct 15 '19 at 14:49
  • $\begingroup$ I agree that it should be specified that $\tilde P$ is parallel to $\tilde T$. Thanks... I'll make the change. I restricted the displacement's temporal component to avoid the $a=0$ case (infinite velocity in $\hat t$'s frame) and to keep it pointing forward in time (in $\hat t$'s frame). I also wanted $\tilde P$ to be future-pointing causal. $\endgroup$ – robphy Oct 15 '19 at 16:56

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.