So recently I've been doing some self-study on canonical transformations and relating together different Hamiltonian systems. I've found this paper (PDF) with a remarkable result showing that any two Hamiltonian systems $H(q,p)$ and $K(Q,P)$ with the same degrees freedom, are locally equivalent and connected via specific canonical transformation derived from the solution to the Hamiltonian-Jacobi equations. The construction essentially shows (to my understanding) that the action functional can be looked at as a generating function mapping, which transforms the Hamiltonian to systems where nothing changes. Applying this to both systems $H$ and $K$ and requiring that the constants match links the two systems. However, this construction goes through the HJ canonical transformation, which is time-dependent as it essentially contains information on the dynamics of the original system. Nevertheless, the canonical transformations that connect both systems together in their examples always end up being time-independent.

So my question is for something I did not currently see a clear idea how to prove or disprove - given two Hamiltonian systems $H(q,p)$ and $K(Q,P)$ with the same degrees of freedom is it true that there always exists a time-independent canonical transformation $f: (q,p) \to (Q,P)$ such that the dynamics of $K(f(q,p))$ in terms of $(q,p)$ are the same as the dynamics of $H(q,p)$ (at least locally since the transformation might not be one-to-one)?


1 Answer 1


TL;DR: The answer is No, since a time-independent canonical transformation (CT) cannot turn a non-zero Hamiltonian into a vanishing Hamiltonian. However OP's conjecture is true for 2 non-zero sufficiently well-behaved Hamiltonians.

More details:

  1. In the same way as time-dependent Hamilton-Jacobi (HJ) theory, is governed by Hamilton's principal function $S(q,P,t)$, the corresponding time-independent problem is governed by Hamilton's characteristic function $W(q,P)$, cf. e.g. Ref. 1.

  2. The underlying mathematical problem comes down to guaranteeing the local existence of a solution to the time-independent HJ equation, which is a first-order non-linear PDE. To ensure this, OP's 2 Hamiltonians $H$ & $\widetilde{H}$ would have to be sufficiently well-behaved.

  3. One new feature (in the time-independent case) is that the 2 Kamiltonians $K$ & $\widetilde{K}$ (produced by time-independent HJ theory) are not necessarily zero (as in the time-dependent case), so as an extra step one should make CTs to ensure that $K=P_1$ and $\widetilde{K}=\widetilde{P}_1$. (This last step could fail if precisely one of the Kamiltonians vanishes.) $\Box$


  1. H. Goldstein, Classical Mechanics; Section 10.3.
  • 2
    $\begingroup$ can you expand a bit on the following obvious example: turn (or at least relate by CT) $H=p^2/2+ x^2/2$ to $\tilde H=p^2/2-x^2/2$? $\endgroup$ Commented Aug 15, 2020 at 14:03

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.