>On second thought, the definition I had for 4NF didn't make any difference between keys; I think it said that a relation was in 4NF if whenever there was a multivalued non-trivial dependency X->>Y, X was a candidate key for R. And I'm pretty sure that the BCNF definition didn't make any difference either between PK and CK's.
>
>Anyway, I would say that NF definitions aren't that important; the problems they apply to are what matters, and adding a surrogate key doesn't make them disappear.

Frederico,

BCNF certainy does address CK's, it says that an entity is in BCNF if it is in 3rd NF for any possible candidate for the primary key. 4th and 5th specifically discuss the primary key and don't address candidate keys at all.

Quoted from Felmming and von Halle;

4th Normal Form: "Fourth normal form addresses recognition and separation of independently multivalued attributes constituting a composite primary key." The rule is stated as, "Reduce BCNF entities to 4th NF by removing any independently multivalued components of the primary key to two new parent entities. Retain the original (now child) entity only if it contains other, non-key attributes."

5th Normal Form: "Fifth normal form addresses cyclic dependencies that warrant decomposition of an entity into three or more equivalent entities (i.e. representing the same information) with less redundancy."
The rule is stated as: "Reduce 4th NF entities to 5th NF by removing pairwise cyclic dependencies (appearing within composite primary keys with three or more component attributes) to three or more new parent entities."

You can see that both 4th and 5th NF specifically refer to the primary key.