Could a Legendre transform be equivalent to evaluation in a retarded time?

178 Views Asked by At

We have the following function:

$$U = U(r(t), p(t)) = (α/r)·\sqrt{1 - p^2/c^2}$$

being $α$ and $c$ positive constants, $t$ a real variable $0$ or positive, and $p = dr/dt$.

I would like to show that if we define

$t'= t – r(t)/c$

it is true that

$U' = U - p·(∂U/∂p)$

where the prime ' stands for evaluated at $t'$.

The origin of the problem is electrodynamics. I try to show that the Legendre transform of the electrodynamic potential is equivalent to a retarded potential evaluated at instant $t'$.

I have tried to prove it from:

$\Delta U=\left( \frac{\partial U}{\partial r} \right)\cdot \Delta r+\left( \frac{\partial U}{\partial p} \right)\cdot \Delta p$

being

$\Delta U={U}'-U$, $\Delta r={r}'-r$, $\Delta p={p}'-p$ and $p={dr}/{dt}\;$

Then

${U}'=U+\left( \frac{\partial U}{\partial r} \right)\cdot \Delta r+\left( \frac{\partial U}{\partial p} \right)\cdot {p}'-\left( \frac{\partial U}{\partial p} \right)\cdot p$

In order to prove my equation, then

$0=\left( \frac{\partial U}{\partial r} \right)\cdot \Delta r+\left( \frac{\partial U}{\partial p} \right)\cdot {p}'$

We can assume that (for $\Delta t\ll 0$, and taken into account that ${t}'=t-r(t)/c$

$\Delta r=p\cdot \Delta t=p\cdot \left( {-r}/{c}\; \right)$

And $p’$ should be

${p}'=c\cdot \left( 1-\frac{{{p}^{2}}}{{{c}^{2}}} \right)$

But I can't think of how to prove the latter. I think the approach I have followed is inappropriate. That is why I am asking for help.

1

There are 1 best solutions below

3
On

As far as I understand the comments you look at the time transformation $$t'=t-r(t)/c$$ and evaluate $U(r(t),p(t))$ at $t=t'$ which gives $$\tag{A} U\Big(r(t'),p(t')\Big)=U\Big(r\big(t-r(t)/c\big),p\big(t-r(t)/c\big)\Big)\,. $$ (I do not think we have to denote this by $U'\,.$) Then you ask if (A) is of the form $$\tag{B} U\Big(r(t),p(t)\Big)-p(t)\frac{\partial U}{\partial p}\Big(r(t),p(t)\Big)\,. $$ I find this very unlikely. Taking the $t$ derivative of (A) and (B) leads to a nonlinear system of first order ODEs for $r(t)$ and $p(t)\,.$ Only when these functions satisfy that system we have equality of (A) and (B).

Some background on Legendre transforms:

For simplicity I assume $r>0$ and $p=\dot r>0\,.$ From $$ \frac{\partial U}{\partial p}=\frac{\alpha}{rc^2}\frac{-p}{\sqrt{1-\frac{p^2}{c^2}}}=\frac{-\alpha}{rc^2\sqrt{\frac{1}{p^2}-\frac{1}{c^2}}}\,,\quad \frac{\partial^2U}{\partial p^2}=\frac{\alpha}{2rc^2}\frac{-2p^{-3}}{\Big(\frac{1}{p^2}-\frac{1}{c^2}\Big)^{3/2}} $$ it follows that $U$ is strictly concave in $p\,.$ This allows to express the Legendre transform of $U$ as a nice function of $\partial U/\partial p\,.$ The convex conjugate of $p\mapsto U(r,p)$ is the function $$\tag{1} U^*(r,q)=\sup_{p>0}\{qp-U(r,p)\} $$ which is also called Fenchel-Legendre transform. Since $p\mapsto U(r,p)$ is strictly concave the supremum in (1) is attained for the unique $p$ that satisfies $$\tag{2} q=\frac{\partial U}{\partial p}(r,p)\,. $$ This allows to write $$\tag{3} U^*(r,q)=qp-U(r,p)\,. $$ Note that

  • $p$ in (3) is a function of $q\,.$

  • Therefore, $U^*$ is not a function of $p$ but instead of the conjugate variable $q=\frac{\partial U}{\partial p}\,.$

  • In the physics literature a Legendre transform (3) is often written quite sloppily as $$\tag{4} U^*=p\frac{\partial U}{\partial p}-U\,. $$ This ignores the fact that $U^*$ is not a function of $p\,$. For further details, see Zia et.al..

  • Up to a minus sign, your $U'$ (my (B)) is the sloppy $U^*$ from (4) but this would be a Legendre transform and not a time transform.