I. EVALUACIÓN DE LAS POLÍTICAS PÚBLICAS CUESTIONES
I.3. ARTICULACIÓN DE UN PROCESO DE EVALUACIÓN
I.3.2. Criterios de Evaluación
I.3.2.5. La sostenibilidad
4.5.4 A different Perspective for MVDs
M VDs have been defined as expressions
X -
Y whereX, Y E Sub(N).
Alternatively, we can view MVDs as expressionsX -
Y whereX, Y E
eN and (eN, �'U,
n, -'-eN , JN) is the Brouwerian algebra of closed subsets of the PO-space on the join-irreducible elements J Nof
Sub(N).
A setr
�dom(N) satisfies
the MVDX -
y
on eN, denoted byFr X -
Y, if and only if there isa t E r
with1r�(t)
=1r�(t1)
for allB E XUY
and1rg(t)
=1rg(t2)
for allC E X U
(JN-'-cNY) whenever7r1(tl) = 7r1(t2)
for allA E X
holds for anyt1 , t2 E r.
This view can again be justified in the following sense. Lemma 3.9 shows that for allr
�dom(N)
we have
Fr X
- y forX, y E
eN if and only ifFr u X - u
y in terms of Definition 4. 1 . The minimal axiomatisation from Theorem 4.44 reads then as follows. The following inference rulesx - Y, x - z x - Y u z
are minimal, sound and complete for the implication of MVDs in the presence of records and lists.
4.6 Related and Future Work
MVDs have been studied very well in relational databases. The next goal is the proposal of a nested list normal form for nested attributes with respect to the class of MVDs and the class of FDs and MVDs, to semantically justify this proposal and generalise the decomposition approach. Research papers that may be used as guidelines are [103, 133, 289, 290] . The proposal of such a normal form can be found in Section 6.2 of this thesis. It is desirable to improve the running time of Algorithm 4.4. 1 for deciding the implication of FDs and MVDs. Substantial research on that subject has again been done for relational databases and the papers [98, 1 18, 135, 152, 173, 223, 239, 277] may give some more information. The paper [2 1 6] proposes algorithms how to obtain
reduced MVDs
andminimal covers
of sets of MVDs for relational databases. The concept of apure set of FDs and MVDs
was introduced in [154] . An MVDX
----» Y of a setE
of FDs and MVDs on a relation schema R is called pure iff it is non-trivial and neitherX
-t Y norX
-t (R - Y) are inE+.
A related definition aimed at factoring out MVDs which cannot be derived from FDs appears in the concept of anenvelope set
due to [301, 302] in a work on desirable 4NF decompositions. So-calledconflict-free MVDs
are introduced in [247] . MVDs of this class have the property that they allow a unique 4NF dependency preserving database schema. Moreover it is stated that non conflict-free sets of dependencies are inadequately specified. It is interesting to study these different notions in the context of complex object types.Multi-valued dependencies have been the subject of
data mining.
In [242] two algo rithms for the discovery of multi-valued dependencies from relations are presented. The top-down algorithm enumerates the hypotheses from the most general to more specific hy potheses which are checked on the input relation. The bottom-up algorithm first computes the invalid multi-valued dependencies. Starting with the most general dependencies, the algorithm iteratively refines the set of dependencies to conform with each particular invalid dependency. The implementation of the algorithms is analysed and some empirical results are presented. A different approach is proposed in [300] .Recent papers that study multi-valued dependencies in the context of XML are [286, 287] . The work in [286] introduces MVDs in XML (XMVDs) and justifies the definition by showing that for a general class of mappings from relations to XML, a relation satisfies an MVD if and only if the corresponding XML document satisfies the corresponding XMVD. As this justification of XMVDs already suggests, XMVDs provide semantics for XML documents that are exported or imported from relational databases. Therefore, XMVDs do not cover multi-valued dependencies among complex objects such as lists. The definition of XMVDs is again based on the notion of a path. The work in [287] proposes an extension of the well-known fourth normal form ( 4NF) from relational data bases to XML in order to syntactically describe semantically well-designed XML documents with respect to XMVDs as studied in [286] .
A conceptual treatment of MVDs is introduced in [266] . It is proposed that entity relationship modelling techniques enable a more natural and intuitive way of handling MVDs. Based on the concept of
competing MVDs
it is proven in which case a unique entity-relationship schema representation exists. If MVDs are competing, then either one4.6. RELATED AND FUTURE WORK Sebastian Link of the competing schemata is chosen or an approximation which combines the competing schemata can be used.
For more comments on future work see Section 6.2. Let us finally look at a further example of MVDs among complex objects. Suppose we store nucleotide sequences together with certain genes that occur in it, i.e. sequences of amino acids, and together with a certain base and the sequence of positions in which that base appears in the original nucleotide sequence. We may use the nested attribute
Genes(Sequence[N ucleotide] , Gene[ Amino-Acid] , Occurs(Base, Position[N umber] ) ) . There might b e several genes encoded within the nucleotide sequence, and there are dif ferent bases together with a certain sequence of positions in which they occur. The set of genes, however, is independent from the set of bases and the corresponding sequence of occurrences. We therefore have the following MVDs
Genes(Sequence[Nucleotide] ) ----* Genes(Gene[Amino-Acid] ) and Genes(Sequence[Nucleotide]) ----* Genes(Occurs(Base, Position[Number]) ) . Moreover there are the FDs
Genes(Sequence[Nucleotide] ,Occurs(Base)) -+ Genes(Occurs(Position[Number] ) ) and
Genes (Sequence[N ucleotide J, Occurs (Position [Number] ) ) -+ Genes( Occurs (Base) ) . It appears that the chance of MVDs occurring among complex objects is as good as the chance of MVDs occurring among fiat data. The techniques provided in this chapter may therefore help to cover more application domains.