# Why does isotropy principle require existence of inertial transformation when axes are reversed?

Assuming one spatial and one termporal dimension, let's assume an intertial transformation $A(v)$ as follows: $$\begin{pmatrix} t' \\ x' \\ \end{pmatrix} = A(v) \begin{pmatrix} t \\ x \\ \end{pmatrix}$$ where $$A(v) = e^{\sigma v} \begin{pmatrix} 1 & 0 \\ -v & 1 \\ \end{pmatrix}$$ for some constant $\sigma \ge 0$ (we get the Galileo transformation for $\sigma=0$ as a special case). It is simply a Galileo transformation combined with a velocity-dependent (and not reflection invariant, as shown below) global dilatation.

It can be easily shown, that $A(0) = 1$, $A^{-1}(v) = A(\bar v)$ with $\bar v=-v$ and $A(u)A(v)=A(w)$ with $w=u+v$, so $A(v)$ forms a group. The transformation is linear, as required by space homogeneity.

Following [1], if the space is isotropic, then $(t, -x)$ and $(t, -x')$ qualify as well for equivalent coordinates. Introducing the space reflection (parity) by the matrix $T=diag(1, -1)$, we can see, that $T A(v) T$ does not belong into the group (i.e. there is no $\bar v$ so that $T A(v) T = A(\bar v)$) unless $\sigma=0$ (then $\bar v=-v$), because $$T A(v) T = e^{\sigma v} \begin{pmatrix} 1 & 0 \\ +v & 1 \\ \end{pmatrix}$$

In other words, if an event has coordiantes $(t, x)$ and $(t', x')$ in the two inertial frames, then for $\sigma > 0$ there is no inertial transformation (of the above form), that would connect $(t, -x)$ and $(t, -x')$, so the transformation $A(v)$ does not respect isotropy.

What is the physical meaning of the above $A(v)$ transformation for $\sigma > 0$? Does it mean that it depends on the axes orientation? How could an experiment be constructed to show that the space is not isotropic? This transformation must be preferring a space direction, but I would like to see this explicitly.

[1] Levy-Leblond, J.-M. (1976). One more derivation of the Lorentz transformation. American Journal of Physics, 44(3), 271–277.

## 2 Answers

Adding the factor of $\exp(\sigma v)$ to the Galilean transformation matrix simply means that every infinitesimal Galilean transformation is "decorated" by an extra uniform infinitesimal rescaling of $t,x$ by the same factor. Such a "velocity-dependent scaling of time (and space)" has of course never been considered in classical mechanics where time was universal and absolute (including the normalization).

But you may consider it. When you do, the very form of the factor $\exp(\sigma\cdot v)$ makes it obvious that $\sigma$ is really a spatial vector (and not a scalar), too. If there's one spatial dimension only, spatial vectors and scalars may get confused but it's easy to unconfuse oneself, especially if you try to construct higher-dimensional generalizations of this construction (higher than 1+1).

Because $\sigma$ is a spatial vector, it's obvious that under the spatial reflection $x\to -x$, $\sigma$ goes to $-\sigma$, too. That's why you can't construct a parity-symmetric theory with a fixed nonzero $\sigma$: the parity transformation simply wants to revert the sign of $\sigma$ but it isn't allowed because the theory wants a fixed $\sigma$. I find the ordering of the description of the problems of this theory you offered kind of upside down because you are trying to "impose" the parity symmetry and claim that the Galilean transformations won't exist to relate two events in spacetime. It seems more logical to me to assume/demand the more nontrivial symmetry, the Galilean one, and then you may simply see that the parity symmetry is broken for $\sigma\neq 0$.

• Thanks for the answer and good point about the spatial vector. What you wrote is correct, but I still don't understand why we want to reject such transformations (=how exactly would the anisotropy manifest in physical experiments based on this particular transformation?). This is just a counterexample used in [1] to show that if isotropy of space is not imposed (but you still want linearity and a group), we get this as a possible transformation (there are more such examples, this is just one of them). Commented Jan 16, 2013 at 17:01
• Dear @OndřejČertík , a theory invariant under the transformations with a nonzero $\sigma$ isn't isotropic simply because $\sigma$ is a spatial vector that breaks the symmetry between different directions of space. If you don't impose the condition that the theory yields isotropic phenomena (as required by the cosmological principle, for example) you may think that these transformations are indeed OK - from a purely group-theoretical viewpoint. But if you do require isotropy etc., these transformations are no good. You won't find interesting theories invariant under this group. Commented Jan 17, 2013 at 8:17
• I don't know what to do with questions like "How exactly would the anisotropy manifest in physical experiments?" You want to know "something" exactly except that you describe what this "something" is totally inexactly and fuzzily. Which theories? Which experiments? I can't answer "exactly" what will happen with an experiment according to a theory if you don't tell me what is either. But by pure maths, group theory, I know and I have shown that it's impossible that phenomena in such a theory will preserve isotropy (or parity). Commented Jan 17, 2013 at 8:19
• Hi Luboš, I apologize for my late reply (my son was just born...) and also that I was not clear before, but I finally got back to this. I have posted my own answer, which shows such an experiment. I hope I didn't make a mistake in the derivation, but I am now happy with my own answer, assuming it is correct. Commented Jan 30, 2013 at 6:44

I'll try to answer my own question. Here is the experiment which shows the space anisotropy. The above transformation implies the following time dilatation: $$\Delta t' = e^{\sigma v} \Delta t$$ which depends on the direction of $v$. So we consider two synchronized clocks A and B at x=0, then move A to $x=1$ and B to $x=-1$ with velocities $\pm v_1$. Then we exchange A, B with velocities $\pm v_2$. Finally we move A, B back to $x=0$ with velocities $\pm v_1$. The idea is to use $v_1 \ll v_2$, so that $v_1$ virtually doesn't cause any dilatation, while $v_2$ causes almost all the dilatation. As such, the clock B must show more time.

Now in equations (I hope I didn't make a mistake below). Let's start with t=t'=0. For the clock A, by moving from $x=0$ to $x=1$ it will show the time $t'=t_1'$: $$\begin{pmatrix} t_1' \\ 0\\ \end{pmatrix} = A(v_1) \begin{pmatrix} t_1 \\ 1 \\ \end{pmatrix}\,,$$ from which $t_1 = \frac{1}{v_1}$ and $t_1' = e^{\sigma v_1} t_1 = e^{\sigma v_1}\frac{1}{v_1}$. Now we move to $x=-1$, the clock A will show time $t'=t_1'+t_2'$, where $t_2' = e^{-\sigma v_2}\frac{2}{v_2}$. Finally, by moving back to $x=0$, the clock will show the time $t'=t_1' + t_2' + t_3'$ where $t_3' = e^{\sigma v_1}\frac{1}{v_1}$. All together then: $$t' = t_1' + t_2' + t_3' = e^{\sigma v_1}\frac{1}{v_1} + e^{-\sigma v_2}\frac{2}{v_2} + e^{\sigma v_1}\frac{1}{v_1} = e^{\sigma v_1}\frac{2}{v_1} + e^{-\sigma v_2}\frac{2}{v_2}$$ In the same way, the clock B will show: $$t' = e^{-\sigma v_1}\frac{2}{v_1} + e^{\sigma v_2}\frac{2}{v_2}$$ Using for example $v_1=1$ and $v_2=10$ and $\sigma=0.1$ we obtain for clock A: $$t' = 2.2839...$$ and clock B $$t' = 2.3533...$$ as can be calculated by this simple Python script:

In [1]: from math import exp
In [2]: v1 = 1.
In [3]: v2 = 10.
In [4]: sigma = 0.1
In [5]: exp(sigma*v1) * 2 / v1 + exp(-sigma*v2) * 2/v2
Out[5]: 2.2839177243855837
In [6]: exp(-sigma*v1) * 2 / v1 + exp(sigma*v2) * 2/v2
Out[6]: 2.3533312017637282


This experiment is symmetrical, but we still got different times for A, B. This means that we can use this experiment to measure if time goes faster when moving to the left or to the right.