I was trying to prove, that for a transformation to be Canonical, one must have a relationship:

$$ \left\{ Q_a,P_i \right\} = \delta_{ai} $$

Where $Q_a = Q_a(p_i,q_i)$ and $P_a = P_a(p_i,q_i)$.

Now to do the proof I started with $\dot{Q_a}$:

  1. Chain rule and Hamilton's equation for initial coordinates $q_i,p_i$ $$ \dot{Q_a} = \frac{\partial Q_a}{\partial q_j} \dot{q_j} + \frac{\partial Q_a}{\partial p_j} \dot{p_j} = \frac{\partial Q_a}{\partial q_j} \frac{\partial H_a}{\partial p_j} - \frac{\partial Q_a}{\partial p_j} \frac{\partial H_a}{\partial q_j} $$

  2. Then I apply chain rule for the Hamiltonian derivatives: $$ \dot{Q_a} = \frac{\partial Q_a}{\partial q_j} \left( \frac{\partial H}{\partial Q_i} \frac{\partial Q_i}{\partial p_j} + \frac{\partial H}{\partial P_i} \frac{\partial P_i}{\partial p_j} \right) - \frac{\partial Q_a}{\partial p_j} \left( \frac{\partial H}{\partial Q_i} \frac{\partial Q_i}{\partial q_j} + \frac{\partial H}{\partial P_i} \frac{\partial P_i}{\partial q_j} \right) $$

  3. Now reordering the terms yields us: $$ \dot{Q_a} = \frac{\partial H}{\partial Q_i} \left\{ Q_a,Q_i \right\} + \frac{\partial H}{\partial P_i} \left\{ Q_a,P_i \right\} $$

Now here the problem, for the transformation from a coordinate system $(q_i,p_i)$ to a coordinate system $(Q_a(q_i,p_i), P_a(q_i,p_i))$ to be canonical we require:

$$\left\{ Q_a,P_i \right\} = \delta_{ai}$$

But why we have as well the following requirement, or is it just too obvious or true because of some property of any coordinate transformation?

$$\left\{ Q_a,Q_i \right\} = 0$$

The problem I am having is as follows. I agree, that the following two are true (if I used the covariant notation correctly):

$$ \left\{ q^i,q_j \right\}_{q,p} = \frac{\partial q^i}{\partial q^k} \frac{\partial q_j}{\partial p_k} - \frac{\partial q^i}{\partial p^k} \frac{\partial q_j}{\partial q_k} = 0 $$

$$ \left\{ Q^i,Q_j \right\}_{Q,P} = 0 $$

But why its the case that the following is also true?

$$ \left\{ Q^i,Q_j \right\}_{q,p} = 0 $$


1 Answer 1


A canonical transformation $(q^i,p_j) \to (Q^i,P_j)$ preserves the form of Hamilton's equations.

Similarly, a symplectic transformation$^1$ $(q^i,p_j) \to (Q^i,P_j)$ preserves the Poisson structure, aka. as a symplectomorphism. In other words, all the fundamental Poisson brackets (PB)

$$ \{ q^i,p_j \} ~=~ \delta^i_j, \qquad \{q^i,q^j \}~=~0, \qquad \{ p_i,p_j \} ~=~ 0,\qquad i,j \in\{1, \ldots, n\},$$

have the same form in the new coordinates

$$ \{ Q^i,P_j \} ~=~ \delta^i_j, \qquad \{Q^i,Q^j \}~=~0, \qquad \{ P_i,P_j \} ~=~ 0,\qquad i,j \in\{1, \ldots, n\}. $$

In particular, to answer OP's question(v2), the relations $\{Q^i,Q^j \}=0$ and $\{P_i,P_j\} = 0$ are only trivial if $n=1$, because of skewsymmetry of PB.

As is well-known, canonical and symplectic transformations are the same. For a proof [at least in the case of restricted transformations, i.e. transformations without explicit time dependence], see e.g. Ref.1, which uses so-called symplectic notation. An important point is that the Jacobian matrix of a symplectic transformation must be a symplectic matrix.


  1. H. Goldstein, Classical Mechanics, Section 9.4 in eds. 3 or Section 9.3 in eds. 2.


$^1$ In this answer we will for simplicity only discuss non-degenerate Poisson brackets in finite dimensions using globally defined coordinates.


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.