The definition that DiSSCo (Distributed System of Scientific Collections) uses is: “a persistent identifier is a string (functioning as a symbol/name) that identifies a digital object. The identifier can be persistently and reliably resolved to digitally actionable meaningful information about the identified digital object.”
We don’t say that it is globally unique, but it otherwise reliable resolution won’t occur. Also, we don’t mention metadata because that is a characteristic of the object, not a characteristic of its identifier. As Eugene @0000-0002-5119-2271 said, Handles (which are the PIDs DiSSCo will use) are just pointers but only once they have a PID Record associated with them. Before that, they’re just names, like my name “Alex”. It doesn’t tell you anything about me or how/where to find me.
Metadata is often needed of course, to tell you something about the identified thing and to make connections to other things, as has been mentioned already.
It’s also not true to say that PIDs exist in perpetuity. They exist only for as long as they are needed, which can be a very long time (>100 years in the case of DiSSCo). How long, is a policy decision related to the purpose for which they are being used. There are use cases, for example where workflows can create huge numbers of PIDs for intermediate results during multiple parameters sweeps and data sweep executions that don’t need to be retained beyond the workflow runs.
However, while PIDs do exist they must persistently identify the correct thing and persistently resolve reliably and stably. It is the identifier that is persistent, not necessarily the thing to which it resolves, although in many use cases that is also the case.