# About the group of canonical transformations and the matrices representing them

Recently I have come to know that for a system with $$2n$$ dimensional phase space, the set of all canonical transformations form a group $${\rm Sp(2n, R)}$$. But in contrast to other Lie groups e.g. $${\rm SO(3)}, {\rm SU(2)}$$ etc, I find this group to be quite abstract. Let me explain.

Any matrix $$M\in{\rm Sp(2n, R)}$$ act on the column vector $$\underline{z}=(q_1,...,q_n,p_1,...,p_n)^T$$ and preserve the Poisson bracket structure. But unlike rotations, CTs $$(Q,P)\to(q,p)$$ are most often not linear. Therefore, in contrast to rotation matrices $$R_{\hat{n}}(\vec{\theta})\in {\rm SO(3)}$$ where the group elements $$R_{\hat{n}}(\vec{\theta})$$ depend only on the group parameters $$\vec{\theta}$$, the matrices $$M$$ seems to depend on $$(q,p)$$ variables too. For instance, for the hamiltonian $$H=\frac{1}{2}(p^2+q^2)$$, a CT like $$q=\sqrt{2P}\cos Q, ~ p=\sqrt{2P}\sin Q$$ cannot be written as $$(Q, P)^T=M ~(q,p)^T$$ with an $$M$$ that is independent of $$(q,p)$$.

• This does not feel correct to me. I am accustomed to seeing linear transformations forming a group, not nonlinear transformations.

####################################################################

Next, the CTs can be widely different in their appearance. To state more clearly, for systems with $${\rm 2D}$$ dimensional phase space, all the transformations below

$$Q=p~ \& P=-q$$,

$$Q=\log \frac{1}{q}\sin p~ \&~ P=q\cot p$$,

$$Q=\tan^{-1}(q/p) ~ \& ~ P=\frac{1}{2}(p^2+q^2)$$

... etc, are examples of CTs (in the sense that all of them preserve canonical PB structure). Therefore, if I am not terribly wrong, all these transformations are elements of $${\rm Sp(2,R)}$$. These transformations are wildly different from each other; in no way they one is obtainable from the other by continuous variation of group parameters. This is also quite bizarre.

• Again, I surely hold serious misunderstanding but can't find it. Can someone point out what's wrong with my line of thought? Thanks!

## 1 Answer

1. The full groupoid of all (possibly non-linear) canonical transformations (CTs) is infinite-dimensional. Infinitesimal CTs are Hamiltonian vector fields $$\delta z^I ~=~{\{F(z,t),z^I\}}, \qquad I~\in~\{1,\ldots,2n\}.\tag{1}$$

2. The symplectic group $$Sp(2n, \mathbb{R})$$ of dimension $$n(2n+1)$$ is the group of all linear time-independent CTs. The corresponding symplectic Lie algebra $$sp(2n,\mathbb{R})$$ is the set of all linear time-independent infinitesimal CTs, which have time-independent quadratic generating functions $$F(z)~=~\frac{1}{2}\sum_{I,J=1}^{2n}a_{IJ}z^Iz^J.\tag{2}$$

• Okay. Therefore, for the 2D phase space all CTs of the form $Q=\alpha q+\beta p, P=\gamma q+\delta p$ ($\alpha,\beta,\delta,\gamma$ are independent of $q,p$) belong to the group ${\rm Sp(2,R)}$. My 2nd and 3rd examples of CTs, which involve logarithmic, trigonometric functions do not belong to ${\rm Sp(2,R)}$. Am I correct? – mithusengupta123 Jun 30 '20 at 15:58
• $\uparrow$ Right. – Qmechanic Jun 30 '20 at 16:10
• Isn't then the Wikipedia notation $M(z,t)\in Sp(2n,R)$ misleading? en.wikipedia.org/wiki/Symplectic_group#Classical_mechanics – mithusengupta123 Jun 30 '20 at 16:43
• $\uparrow$ Yes. – Qmechanic Jun 30 '20 at 17:08