This is something we are working on for EclipseLink 1.1. In EclipseLink
1.0 you can map the relationship using a DescriptorCustomizer and use
either a OneToManyMapping or AggregateCollectionMapping.
For the OneToManyMapping, you will need to map the foreign key in the
target object some how, either Basic or ManyToOne (of coarse then you
could just use JPA as normal).
For AggregateCollectionMapping the target object needs to be treated like
an Embeddable and marked as an aggregateCollectionDescriptor.
In general the best way to model a OneToMany is to either have a ManyToOne
back, or use a join table.