The Poissonian formulation of mechanics tells us that for a generating function $g(q,p,t)$, the Poisson bracket of some function/variable $f(q,p,t)$ with the generating function corresponds with an infinitesimal change in $f$ along the transformation or "motion" generated by $g$.

$$\delta f = \epsilon \left\{f,g \right\}$$

An example of this is momentum conservation due to invariance under infinitesimal translations. To show this, take $f$ to be the Hamiltonian and $g$ to be $\mathbf{p}\cdot\hat{n}$, where $\mathbf{p}$ is the momentum $p_x \hat{x}+p_y\hat{y}+p_z\hat{z}$ and $\hat{n}$ is an arbitrary unit vector. The canonical transformation generated by $\mathbf{p}\cdot \hat{n}$ is an infinitesimal translation along the $\hat{n}$ direction of the system variables with which the Hamiltonian is evaluated.

$$\begin{align*} \epsilon\left\{H,\mathbf{p}\cdot\hat{n}\right\}&=\epsilon\left(\sum_i \frac{\partial H}{\partial q_i}\frac{\partial\,(\mathbf{p}\cdot\hat{n})}{\partial p_i}-\frac{\partial H}{\partial p_i}\frac{\partial\,(\mathbf{p}\cdot\hat{n})}{\partial q_i}\right)\\ &=\epsilon\left(\sum_i \frac{\partial H}{\partial q_i}(\hat{n})_i\right)\\ &=\epsilon (\nabla_q H)\cdot \hat{n}\\ &\\ &\implies \left\{H,\mathbf{p}\cdot\hat{n}\right\}=(\nabla_q H)\cdot \hat{n} \end{align*}$$

Now, if we were to take an polar angle $\theta$ about some axis $\hat{n}$ to be a coordinate, the above procedure with $\mathbf{l}$, the angular momentum, in place of $\mathbf{p}$ would then translate as an infinitesimal "translation" of the $\theta$ variable - i.e. a rotation about the $\hat{n}$ axis. An example of this is given in Landau & Lifshitz, Goldstein, and many other mechanics textbooks - the rotation of a constant vector $\mathbf{c}$ about a specified axis.


In terms of the interpretation of the Poisson brackets through generating functions (which I just gave), I can see why this would be true. The vector $\mathbf{c}$ changes by an amount $d\theta(\hat{n}\times\mathbf{c})$ when rotated by an infinitesimal angle $d\theta$ about an axis $\hat{n}$, and that result can be reached by simple analytical geometry. However, by direct evaluation of the Poisson bracket, I can't see why this isn't zero (as $\mathbf{c}$ is a constant). The angular momentum operator (vector-valued function in terms of phase space variables) is given by

$$\begin{align*} \mathbf{l}&=\mathbf{r}\times\mathbf{p}\\ &=(yp_z-zp_y)\hat{x}+(zp_x-xp_z)\hat{y}+(xp_y-yp_x)\hat{z} \end{align*}$$

Note that this, assuming a typical classical Hamiltonian, entirely in terms of phase space variables. Now, the Poisson bracket of this with a constant vector is

$$\begin{align*} \left\{\mathbf{c},\mathbf{l}\cdot\hat{n}\right\}&=\sum_i\left(\frac{\partial \mathbf{c}}{\partial q_i}\frac{\partial (\mathbf{l}\cdot\hat{n})}{\partial p_i}-\frac{\partial \mathbf{c}}{\partial p_i}\frac{\partial (\mathbf{l}\cdot\hat{n})}{\partial q_i}\right)\\ &=0\,\,\,\,(\mathbf{c}\textrm{ doesn't depend on phase space variables)} \end{align*}$$

Please, could you tell me how to resolve this paradox?

P.s: I originally wrote this question extremely briefly because I thought somebody would certainly know what I'm talking about.

  • $\begingroup$ In the definition of the Poisson bracket derivatives wrt. the canonical coordinates and momenta are taken. The derivatives of $c$ vanish and hence the Poisson bracket as well. $\endgroup$
    – Urgje
    Commented Feb 22, 2016 at 9:42
  • 2
    $\begingroup$ @Urgje $\vec c$ is a vector, that implies a certain transformation behaviour of the components on changes of the coordinate system, therefore a constant vector $\vec c$ does not necessarily have constant components (consider, for example spherical coordinates where the vector basis changes from point to point). $\endgroup$ Commented Feb 22, 2016 at 14:22
  • 1
    $\begingroup$ @SebastianRiese That $c$ is a vector does not imply that it depends on the canonical variables. $\endgroup$
    – Urgje
    Commented Feb 22, 2016 at 22:38
  • 3
    $\begingroup$ @ArturodonJuan (3 comments up) for the record, when a question is closed (technically, put on hold), that's not necessarily such a bad thing. The whole point of putting on hold is to keep answers away until you can edit the question to be good. $\endgroup$
    – David Z
    Commented Feb 23, 2016 at 3:46
  • $\begingroup$ I have a strong feeling this has to do with the fact that $\mathbf{l}\cdot\hat{n}$ is the momentum conjugate to the polar angle about the $\hat{n}$ axis, and so the angular momentum operator/function that I wrote down in terms of the standard $x$, $y$, and $z$ doesn't apply - i.e. I need to express \mathbf{c} in terms of coordinates of which at least one needs to be the angle $\theta$ about the \hat{n} axis. $\endgroup$ Commented Feb 23, 2016 at 16:02

2 Answers 2


The starting point is that the $3$-vector $\vec{\bf c}$ transforms in the $3$-dimensional irreducible vector representation of the rotation group $SO(3)$,

$$\tag{1} \{ \vec{\bf c}, \vec{\bf L}\cdot \hat{\bf n} \}_{PB}~=~ \hat{\bf n}\times \vec{\bf c},$$

where $\hat{\bf n}$ is an arbitrary unit vector, whose Poisson bracket (PB) with anything vanishes

$$\tag{2} \{ \hat{\bf n}, \cdot \}_{PB}~=~0.$$

We assume that $\vec{\bf c}$ is not identically zero. Since the PB with $\vec{\bf c}$ does not vanish, the $3$-vector $\vec{\bf c}$ cannot be a constant. It must be a function of the phase space variables $\vec{\bf r}$ and $\vec{\bf p}$. It can be thought of as being of the form

$$\tag{3} \vec{\bf c}~=~\vec{\bf r}f+ \vec{\bf p}g+ \vec{\bf L}h,$$


$$\tag{4} f~=~f(r^2,p^2,\vec{\bf r}\cdot\vec{\bf p},L^2), \quad g~=~g(r^2,p^2,\vec{\bf r}\cdot\vec{\bf p},L^2), \quad\text{and}\quad h~=~h(r^2,p^2,\vec{\bf r}\cdot\vec{\bf p},L^2), $$ are three suitable functions of the phase space $SO(3)$ scalars

$$\tag{5} r^2,\quad p^2,\quad \vec{\bf r}\cdot\vec{\bf p},\quad\text{and}\quad L^2.$$


  1. H. Goldstein, Classical Mechanics; Section 9-6 in 2nd edition or Section 9.7 in 3rd edition.
  • $\begingroup$ The question specifies $\mathbf c$ as a constant, though. Are you saying this is impossible? $\endgroup$ Commented Feb 23, 2016 at 18:30
  • $\begingroup$ $\uparrow$ Yes. I updated the answer. $\endgroup$
    – Qmechanic
    Commented Feb 23, 2016 at 19:23
  • 1
    $\begingroup$ That makes very little sense to me. $f:\mathrm{PS}\to\mathbb R$, $f\equiv 1$ is a perfectly valid constant function on phase space. Why is $f:\mathrm{PS}\to\mathbb R^3$, $f\equiv (1,0,0)$ suddenly impossible? $\endgroup$ Commented Feb 23, 2016 at 19:42
  • $\begingroup$ Because then $\{ f, \cdot \}_{PB}=0$. $\endgroup$
    – Qmechanic
    Commented Feb 23, 2016 at 20:03
  • 1
    $\begingroup$ Yeah, that's sort of what the OP is asking - why is that not allowed? Heck, you could even rephrase the PB as $$\left\{\hat{e}\cdot \mathbf{c},\mathbf{l}\cdot\hat{n}\right\}_\mathrm{PB}=\hat{e} \cdot (\hat{n}\times\mathbf{c}),$$ and then choosing $\mathbf c=\hat n=(1,0,0)$ you reduce to the real-valued case $f\equiv 1$. Is $f\equiv 1$ allowed but $f\equiv \hat e \cdot \hat e$ forbidden? There are obviously issues with rotational invariance and active vs passive transformations, but it doesn't help to not talk about them. $\endgroup$ Commented Feb 23, 2016 at 20:19

Let's just do it for a simple example. By $\vec{c}$ I imagine you mean the location of the particle relative to some origin, so $\vec{c}=\vec{r}$. Later on for simplicity we'll suppose further the particle is located on the x-axis (but it is important to do this only after differentiating as we will see).

We'll also suppose we are rotating around the z axis so that $\hat{n}=\hat{z}$.

Then we have \begin{equation} \{\vec{c} , \vec{l}\cdot\vec{n}\}=\{\vec{r},xp_y - y p_x\} =\{x\hat{x}+y\hat{y}+z\hat{z},xp_y - y p_x\}= -y \hat{x}+x\hat{y}. \end{equation}

Now that we have differentiated (meaning, evaluated the brackets) we can set $y=0$ and $x=R$ (that is, we can suppose our particle started on the $x$-axis at the position $R$). Then \begin{equation} \{R \hat{x},\vec{l}\cdot\hat{z}\}=R \hat{y}=\hat{z}\times(R\hat{x}) \end{equation} which is consistent with your formula.

Incidentally, you might be worried that I started off by setting $\vec{c}=\vec{r}$. I think in the framework you are working in--particle mechanics--the vectors should all start from the same origin. If you want to start taking poisson brackets of vectors with different origins, I think you really need to generalize this discussion to field theory (which will complicate the story a bit because in addition to rotating the direction of the vector you need to rotate the origin, so you will end up with an additional term). So I think that may be what you have in mind but that is a more complicated story.

  • $\begingroup$ I'm not sure field theory really needs to be introduced to handle multiple origins. Field theory may be where that comes in withing most physics curricula, but logically speaking I don't see the need. $\endgroup$
    – DanielSank
    Commented Feb 23, 2016 at 18:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.