Database Reference
In-Depth Information
Formally, an annotation for a mapping
M
is a function
α
that maps every position of a
variable in a target atom into cl or op . We refer to the pair (
) as an annotated mapping .
To define the notion of a canonical solution under an annotated mapping, we need to
consider annotated instances: these are instances in which every constant or null is labeled
with cl or op . The semantics of such an instance T is a set Rep ( T ) on instances over C ONST
obtained as follows: after applying a valuation
M
,
α
to T , any tuple ( ..., a op ,... ) in
ν
ν
( T )
can be replicated arbitrarily many times with ( ..., b ,... ),for b
C ONST . For example,
Rep ( { ( a cl ,⊥
op ) } ) contains all binary relations whose projection on the first attribute is
( a cl ,
cl )
{ a }
,and Rep (
{
}
) contains all one-tuple relations
{
( a , b )
}
with b
C ONST .
).These
generate the annotated canonical universal solution C AN S OL ( M, α) ( S ) in the same way
as the usual mapping will generate the canonical universal solution except that now we
have an annotated instance. For example, if we have a tuple (#1,'Data exchange') in re-
lation Papers , the first st-tgd in the mapping will produce a tuple (#1 cl ,
Now suppose we have a source instance S and an annotated mapping (
M
,
α
op ) in relation
Assignment .
The first observation is that the two extreme cases of annotated mappings correspond
precisely to the open-world and the closed-world semantics we have seen earlier. Let
(
in which every position is anno-
tated op (respectively, cl ). Then, for each source instance S we have
M
, op ) (respectively, (
M
, cl )) stand for the mapping
M
S OL M ( S )= Rep (C AN S OL ( M, op ) ( S ));
S OL CWA
M
( S )= Rep (C AN S OL ( M, cl ) ( S )).
Annotated solutions can be defined in a way similar to the CWA solutions, using the
notion of justification. Since we consider settings without target constraints, we can obtain
their algebraic characterization: annotated solutions T for S under (
) are homomor-
phic images of C AN S OL ( M, α) ( S ) so that there is a homomorphism from T into an expan-
sion of C AN S OL ( M, α) ( S ). An expansion of the annotated instance C AN S OL ( M, α) ( S ) is
obtained by expanding op -annotated elements; that is, every tuple in the expansion must
agree with some tuple in the instance on all cl -annotated attributes.
With this, we have some expected properties of annotated solutions:
M
,
α
Rep (C AN S OL ( M, α) ( S )) is precisely the set of all annotated solutions under (
M
,
α
);
α is obtained from
if
α
by changing some cl annotations to op , then each annotated
α ).
solution under (
M
,
α
) is an annotated solution under (
M
,
Given an annotated mapping (
M
,
α
) and a query Q over the target schema, we define
certain ( M, α) ( Q , S )= Q ( D )
)-solution .
|
D
Rep ( T ) where T is an (
M
,
α
For unions of conjunctive queries, this notion coincides with those we have seen.
Proposition 8.34
If
M
is a mapping, Q is a union of conjunctive queries, S is a source
instance, and
α
is an arbitrary annotation, then
certain M ( Q , S )= certain CWA
( Q , S )= certain ( M, α) ( Q , S ) .
M
Search WWH ::




Custom Search