§ Canonical Transformations

The content of this post is based on the notes written by Giancarlo Benettin. These are not only exceptionally well written, but are also very rigorous in their approach. I will try to summarize briefly the most important concepts to clarify what canonical transformations are.
Definition I. Let \(\Omega\subseteq \mathbb{R}^{2n}\) and \(\tilde{\Omega}\subseteq \mathbb{R}^{2n}\) be two open subsets. Consider coordinates \((\mathbf{p},\mathbf{q})\) on \(\Omega\) and \((\tilde{\mathbf{p}},\tilde{\mathbf{q}})\) on \(\tilde{\Omega}\). Let $$ \begin{align} w\colon \tilde{\Omega} &\to \Omega\\ (\tilde{\mathbf{p}},\tilde{\mathbf{q}}) &\mapsto w(\tilde{\mathbf{p}},\tilde{\mathbf{q}}):= (u(\tilde{\mathbf{p}},\tilde{\mathbf{q}}), v(\tilde{\mathbf{p}},\tilde{\mathbf{q}})) \end{align} $$ be a diffeomorphism (differentiable map whose inverse is also differentiable). We will say that \(w\) defines a restricted time-independent canonical transformation if, for any \(H\) defining a Hamiltonian system in \(\Omega\), the function \(\tilde{H}:=H\circ w\) defines a hamiltonian system in \(\tilde{\Omega}\), i.e. $$ \dot{\tilde{\mathbf{p}}}= - \dfrac{\partial \tilde{H}}{\partial \tilde{\mathbf{q}}} \quad \quad \dot{\tilde{\mathbf{q}}}= \dfrac{\partial \tilde{H}}{\partial \tilde{\mathbf{p}}} $$
Note that, abusing (a bit) of notation, we can view the coordinates \(p,q\) as functions of \((\tilde{\mathbf{p}},\tilde{\mathbf{q}})\), this meaning that we may call \(u=p\), and \(v=q\).
Example. Let's consider \(\Omega = \tilde{\Omega}=\mathbb{R}^2\) and for \(a,b\in\mathbb{R}\) the diffeomorphism $$ \begin{align} w\colon \tilde{\Omega} &\to \Omega\\ (\tilde{p},\tilde{q}) &\mapsto (u(\tilde{p},\tilde{q}), v(\tilde{p},\tilde{q})):=(\tilde{p}+a,\tilde{q}+b) \end{align} $$ In order to see that \(w\) defines a restricted time-independent canonical transformation, we need to verify that, for any Hamiltonian \(H\) that defines the Hamiltonian system $$ \dot{p} = -\dfrac{\partial H}{\partial q} \quad \dot{q} = \dfrac{\partial H}{\partial p} $$ the fucntion \(\tilde{H}:= H\circ w\) also defines a Hamiltonian system, but in the new variables, namely that $$ \dot{\tilde{p}} = -\dfrac{\partial \tilde{H}}{\partial \tilde{q}} \quad \dot{\tilde{q}} = \dfrac{\partial \tilde{H}}{\partial \tilde{p}} $$ Indeed, recalling the abuse of notation \(p=u\), we have: $$ \begin{align} \dfrac{\partial \tilde{H}}{\partial \tilde{q}} &= \dfrac{\partial}{\partial \tilde{q}}(H\circ w)= \begin{pmatrix} \dfrac{\partial H}{\partial p} & \dfrac{\partial H}{\partial q}\end{pmatrix} \begin{pmatrix} \dfrac{\partial p}{\partial \tilde{q}} \\ \dfrac{\partial q}{\partial \tilde{q}}\end{pmatrix}= \begin{pmatrix} \dot{q} & -\dot{p}\end{pmatrix} \begin{pmatrix} 0 \\ 1\end{pmatrix}=\\ &= -\dot{p} = -\dot{\tilde{p}+a}= -\dot{\tilde{p}} \end{align} $$ Thus, we get the desired result $$ \dot{\tilde{p}} = -\dfrac{\partial \tilde{H}}{\partial \tilde{q}} $$ Similarly one can also check that the result holds for for the other Hamilton equation.
Definition II. Let \(\Omega\subseteq \mathbb{R}^{2n}\) and \(\tilde{\Omega}\subseteq \mathbb{R}^{2n}\) be two open subsets. Consider coordinates \((\mathbf{p},\mathbf{q})\) on \(\Omega\) and \((\tilde{\mathbf{p}},\tilde{\mathbf{q}})\) on \(\tilde{\Omega}\). Let $$ \begin{align} w\colon \tilde{\Omega} &\to \Omega\\ (\tilde{\mathbf{p}},\tilde{\mathbf{q}}) &\mapsto w(\tilde{\mathbf{p}},\tilde{\mathbf{q}}):= (u(\tilde{\mathbf{p}},\tilde{\mathbf{q}}), v(\tilde{\mathbf{p}},\tilde{\mathbf{q}})) \end{align} $$ be a diffeomorphism (differentiable map whose inverse is also differentiable). We will say that \(w\) defines a general time-independent canonical transformation if, for any \(H\) defining a Hamiltonian system in \(\Omega\), there exists a function \(\tilde{H}\colon \tilde{\Omega}\to\mathbb{R}\) that defines a hamiltonian system in \(\tilde{\Omega}\), i.e. $$ \dot{\tilde{\mathbf{p}}}= - \dfrac{\partial \tilde{H}}{\partial \tilde{\mathbf{q}}} \quad \quad \dot{\tilde{\mathbf{q}}}= \dfrac{\partial \tilde{H}}{\partial \tilde{\mathbf{p}}} $$
Note that this differs from Definition I only by the fact that here, no special definition of \(\tilde{H}\) is given. This makes Definition II more general than Definition I. In particular, this means that there might exist transformations that are general time-independent canonical that are not restricted time-independent canonical. For instance, in the case \(\Omega = \tilde{\Omega}=\mathbb{R}^2\) and \(a,b\in\mathbb{R}\) verifying \(a\cdot b= c^{-1}\) $$ \begin{align} w\colon \tilde{\Omega} &\to \Omega\\ (\tilde{p},\tilde{q}) &\mapsto (u(\tilde{p},\tilde{q}), v(\tilde{p},\tilde{q})):= (a\tilde{p}, b\tilde{q}) \end{align} $$ is general time-independent canonical when \(\tilde{H}=cH\circ w\), which is (generally) different from \(H\circ w\).
It is also common to define canonical transformations as those who preserve the Poisson brackets. We now make clear what we mean by this.
Definition III. Let \(\Omega\subseteq \mathbb{R}^{2n}\) and \(\tilde{\Omega}\subseteq \mathbb{R}^{2n}\) be two open subsets. Consider coordinates \((\mathbf{p},\mathbf{q})\) on \(\Omega\) and \((\tilde{\mathbf{p}},\tilde{\mathbf{q}})\) on \(\tilde{\Omega}\). Let $$ \begin{align} w\colon \tilde{\Omega} &\to \Omega\\ (\tilde{\mathbf{p}},\tilde{\mathbf{q}}) &\mapsto w(\tilde{\mathbf{p}},\tilde{\mathbf{q}}):= (u(\tilde{\mathbf{p}},\tilde{\mathbf{q}}), v(\tilde{\mathbf{p}},\tilde{\mathbf{q}})) \end{align} $$ be a diffeomorphism (differentiable map whose inverse is also differentiable). Furthermore, let $$ f,g\colon \Omega \to \mathbb{R} $$ be two differentiable functions. We will say that \(w\) preserves the Poisson bracket \(\{f,g\}\) if $$ \{f\circ w, g\circ w\}_{(\tilde{\mathbf{p}}, \tilde{\mathbf{q}})} = \{f, g\}_{(\mathbf{p}, \mathbf{q})}\circ w $$
Proposition IV. \(w\) preserves the Poisson bracket if and only if it is a restricted time-independent canonical transformation.
Proposition IV. \(w\) preserves the Poisson bracket if and only if $$ \{u_i, u_j\}_{(\tilde{\mathbf{p}}, \tilde{\mathbf{q}})} = 0 = \{v_i, v_j\}_{(\tilde{\mathbf{p}}, \tilde{\mathbf{q}})} \quad \{v_i, u_j\}_{(\tilde{\mathbf{p}}, \tilde{\mathbf{q}})}= \delta_{ij} \quad \forall i,j = 1,\ldots, n $$
For generalized time-independent canonical transformations, the preservation of the Poisson brackets takes a different form.