3
$\begingroup$

[I know this is a topic in mathematical physics, but I feel like my question is more on the mathematics than the physics. At any rate, feel free to migrate it to Physics.SE if necessary.]

In my notes, one of the implications of Noether's theorem (in the Lagrangian formalism) is stated briefly as such:

Theorem. If the Euler-Lagrange equations hold, and an infinitesimal transformation is invariant, then there exists a conserved quantity.

An infinitesimal transformation is one that takes the form $Q = q + \delta q$ and, for simplicity, we assume that it does not modify the time coordinate. An invariant transformation is one such that the Euler-Lagrange equations take the same functional form, i.e.: if the Lagrangian function is $L_q(q,\dot q,t)$ in the old coordinates and $L_Q(Q,\dot Q ,t)$ in the new, then the latter also lies in the kernel of the Euler-Lagrange operator $$\frac{\mathfrak D}{\mathfrak D q^k} = \frac{\partial}{\partial q^k} - \frac d {dt} \frac{\partial}{\partial \dot q^k}$$ expressed in the old coordinates. We have seen a lemma that states that such a transformation must be such that the new and old Lagrangians are connected by the formula $$L_Q(Q,\dot Q,t) = L_q(Q,\dot Q,t) + \frac{d\Lambda}{dt}(Q,t) \tag1 $$ for some arbitrary regular function $\Lambda$ (actually, we have only proven that if it satisfies $(1)$, the transformation is invariant, but I trust that the other implication holds).

The demonstration goes like this: consider the infinitesimal transformation $Q = q + \delta q$ and suppose that it is invariant, so that it satisfies $(1)$. The Lagrangian is uniquely defined (its value is not dependent on the chosen chart), so $L_q(q,\dot q,t) = L_Q(Q,\dot Q,t)$ and we may revert to the old Lagrangian on the RHS of $(1)$. Substituting the formula for $Q$, we get $$L_q(q,\dot q,t) = L_q(q+\delta q, \dot q + \delta \dot q,t) + \frac{d\Lambda}{dt}(q+\delta q,t) $$ We assume that the Lagrangian and the $\Lambda$ are sufficiently regular, so that we may expand both to first order and obtain $$\frac{\partial L_q}{\partial q^k}\delta q^k + \frac{\partial L_q}{\partial \dot q^k}\delta \dot q^k + \frac d {dt} \left(\Lambda(q,t) + \frac{\partial \Lambda}{\partial q^k}\delta q^k \right) = 0 $$ and, by a reverse Leibniz rule, $$\frac{\partial L_q}{\partial q^k}\delta q^k + \frac{d}{dt}\left(\frac{\partial L_q}{\partial \dot q^k}\delta q^k\right) - \frac d{dt}\left( \frac{\partial L_q}{\partial \dot q^k}\right) \delta q^k + \frac d {dt} \left(\Lambda(q,t) + \frac{\partial \Lambda}{\partial q^k}\delta q^k \right) = 0. $$ Since by hypothesis the E-L equations hold in the old variables $\forall k$, the first and third terms disappear and we are left with the total time derivative $$\frac {d}{dt} \left(\frac{\partial L_q}{\partial \dot q^k}\delta q^k + \Lambda + \frac{\partial \Lambda}{\partial q^k}\delta q^k \right) = 0$$ which implies that the quantity in parenthesis is conserved. However, in my notes, somewhere along the way the $\Lambda$ term disappears, without justification! And indeed the correct conserved quantity is $$\frac{\partial L_q}{\partial \dot q^k}\delta q^k + \delta \Lambda = \text{const}. $$

Why does this happen? Is $\Lambda (q,t)$ itself zero for some reason? Is its total time derivative zero? And how can either option be justified thoroughly?

$\endgroup$
1
$\begingroup$

I don't think the last term $\frac{d\Lambda}{dt}(Q,t)$ of (1) is correct. It needs to vanish when $\delta q=0$, but in this circumstance, it gives $\frac{d\Lambda}{dt}(q,t)$.

Let us consider an example, and denote $L[q]=L(q, \dot q,t)$. Let $L=\dot q^2/2$, and $Q=q+vt$ (Galilean boost). Then $L[Q]=\dot q^2+2v\dot q+v^2=L[q]+\frac{d}{dt}(2vq+v^2t)$. Up to a constant, we have $\Lambda=2vq+v^2t$ (I will add a minus sign to $\Lambda$ here). Since $\delta q=vt$, we can identify $v$ as the small parameter. Observe then that $\Lambda=0$ when $v=0$.

I think a more correct way to write (1) is like this: $$ L[Q]=L[q]+\frac{d}{dt}\left(\delta\Lambda[q]+O(\delta^2)\right). $$ In the above example, we would have $\delta\Lambda[q]+O(\delta^2)=2vq+v^2t$, or $\delta\Lambda[q]=2vq$. Expanding this to first order then yields: $$ \frac{\partial L}{\partial q}\delta q+\frac{\partial L}{\partial \dot q}\delta\dot q=\frac{d}{dt}\delta\Lambda[q]. $$ And from here we can finish Noether: $$ \delta q\left(\frac{\partial L}{\partial q}-\frac{d}{dt}\frac{\partial L}{\partial \dot q}\right)=\frac{d}{dt}\left(-\delta q\frac{\partial L}{\partial \dot q}+\delta\Lambda[q]\right). $$

$\endgroup$
  • $\begingroup$ So would it be correct to say that $\Lambda$ must be chosen so that it vanishes for the identity transformation? Is this also true for general invariant transformation or just specific to Noether’s theorem? $\endgroup$ – giobrach Jul 25 '18 at 7:42
  • $\begingroup$ I believe yes to both questions, because otherwise the $O(1)$ term does not vanish when you expand to first order in $\delta$. So I would say this is true for invariant continuous transformations. $\endgroup$ – user254433 Jul 25 '18 at 8:07
  • $\begingroup$ So this is a huge stretch, but: would it be correct to say that if we are given a family of invariant transformations that depend continuously on a parameter $\alpha$ and forms a group with respect to said parameter, then the set of “gauges” $d\Lambda/dt$ corresponding to those transformations also forms a group that is isomorphic to the former through the map $\text{transf}(\alpha) \mapsto \text{gauge}(\alpha)$, so that the identity must be preserved? $\endgroup$ – giobrach Jul 25 '18 at 8:35
  • $\begingroup$ Well, there are examples where gauge$(\alpha)\equiv 0$, like $\delta q=1$ in the above example, so we would hope for a homomorphism. But I don't think the target space has a group law. If $X=a(t,q)\partial/\partial q$ is the infinitesimal generator, then we have gauge$(\alpha)=[\exp(\alpha X)-1]L[q]$. One try is $g(\alpha)+g(\beta)=g(\alpha+\beta)$, but the right hand side gives $(e_\alpha+e_\beta-2)L$. Another try is $\exp(\beta X)g(\alpha)=g(\alpha+\beta)$, but the RHS is $[\exp(\alpha+\beta)X-\exp(\beta)]L=g(\alpha+\beta)-g(\beta)$. $\endgroup$ – user254433 Jul 25 '18 at 9:42
  • $\begingroup$ I see. Thank you! $\endgroup$ – giobrach Jul 25 '18 at 10:23

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.