Remote deathwath and terminated messages

I have an Akka Cluster (2.2) set up with multiple front end nodes and a
single back end node.

Certain actors on the front end node register themselves with a singleton
actor on the back end node. Each of them puts the other on a deathwatch.
Do my assumptions hold?

1) If a deathwatch triggers on the front end actor, it will trigger on the
backend actor too.
2) The deathwatch triggers on both actors on a network partition.
3) Once the network partition heals and the front end node tries to
re-register with the back-end node, the actor-ref for the front end node
actor will not change.

Thanks,
Rajiv