Canonical transformations, observables, and Mach's principle
par
Salle 1180, bâtiment E2
Salle des séminaires
Canonical transformations (CTs) preserve the form of Hamilton equations but change the Hamiltonian. For example, they can map it to zero. Moreover, they can relate to each other {\it any} two systems of same number of degrees of freedom. For example, a stable system can be canonically mapped to an unstable one. This creates a paradoxal impression that CTs change physics. We argue that the paradox arises only if the system with Hamiltonian $H(p,q,t)$ is considered as totally isolated. The resolution is to put it into contact with an external detector measuring the system coordinates $p,q$. The system+detector can be described by ${\cal H}=H(p,q,t)+\lambda qQ+H_{det}(P,Q)$, which determines the detector response $P(t),Q(t)$. Having chosen ${\cal H}$, the coordinates $p,q$ can be canonically transformed, which changes $H(p,q,t)$, but the detector response does not change. Therefore, the observed physics remains CT independent. It follows that a totally isolated system cannot have dynamics, since its Hamiltonian can be canonically transformed to any value. The dynamics emerges via coupling to the outside world represented by the detector/observer. This viewpoint is reminiscent of Mach's principle, according to which dynamics can be defined only in the presence of ``distant stars''. It follows also that classical mechanics, similarly to quantum theory, needs the external observer for its logical consistency.