Re: PROV-ISSUE-204 (end-of-entity): What marks the end of an entity? [prov-dm]

On Wed, Jan 11, 2012 at 13:30, Luc Moreau <L.Moreau@ecs.soton.ac.uk> wrote:

> You expressed some views about entity deletion, a while back.
> Could you explain them again?

In some cases we can know for sure that an entity has stopped
existing. It might be useful for the asserter to include this in the
account, as to make a closure on entities, which we then know are not
involved with anything beyond that end time.


This is the case for the LucInBoston-kind of entity - the moment you
are no longer in Boston the LucInBoston seizes to exist.


So just from the nature of the characterisation we will have both a
beginning and end of the life of the entity. We have used activities
to denote how an entity was generated, even if this generation is just
a simple state-change in a thing in the world, rather than assembling
a whole new thing.

I believe that similarly we can model the end of an entity's
characterisation interval by a 'Destruction' event. (we might want to
have a more neutral term). A Destruction is a kind of Usage, but which
makes the entity unavailable for any further interactions after its
end. Mirroring generation there can only be one Destruction per
entity.

We don't need to specify what kind of destruction we are talking about
- so I would not call it 'deletion' which is a bit more specific then
'destruction'.

It could be the end of the characterisation, like LucInBoston - or it
could be the end of existence, like when a dog dies, or when flour and
water stop being flour and water, and start being bread. It could be a
file being overwritten, deleted or moved.


Thus an Entity will also have both a start (wasGeneratedAt) and can
have an end (wasDestroyedAt) - a duration. This comes with similar
constraints as for generation - for an entity to be used by an
activity, both their durations must overlap, even if unspecified. For
an entity to be generated by an activity, its beginning must be coved
by the activity's duration, and to be destroyed the entity end must be
covered by the activity duration.


If we add this to PROV-DM it will also make all the explanations of
characterisations and partial ordering more consistent, like the
timelines in figure 2 would be actually expressible in PROV.


I am planning a separate, more complete proposals on how to model the
events and partial ordering within PROV-DM, but as a stepping stone I
would support this issue.

-- 
Stian Soiland-Reyes, myGrid team
School of Computer Science
The University of Manchester

Received on Thursday, 12 January 2012 15:00:38 UTC