Decomposition example into 3NF Remember it is always possible to decompose a relation with functional dependencies F into a decomposed set

of tables which is 3NF, dependency preserving, and loss-less join. The notes describe algorthims which are quite formal and perhaps better suited to computer implementation where the problems are complex. Below is a more intuitive approach which is ussually sufficient for simple problems where you can easily draw the dependencies and inspect them for properties. That is, the approach below embodies the algorthms but the steps are carried out by simple inspection. Consider the following example. R=A,B,C,D,E,F,G F= { AB->CD, C->EF, G->A, G->F, CE->F } Step One, Find the attributes that could be keys for the relation since no primary key is identified (i.e. underlined). G must be in any key since it is not determined by anything (does not appear in the right side of any dependency) B must be in any key for the same reason. Since GB+ contains all the others (i.e. GB determine everything else by transitivity) GB is a key. Since any key must contain GB, GB is in fact the only minimal key.

Step Two, Clean up the set of dependencies by removing any that are implied by others (duplicates), or have unecessary attributes in their left hand sides. You can also combine dependencies with the same left hand sides. (In effect we are finding a minimal cover, but by inspection) So F will reduce as follows. combine G->F and G->A into G->AF combine C->EF and CE->F into C->EF only (E is not necessary in CE->F, because C->F alone) So we have F'c = { AB->CD, C->EF, G->AF } Step Three. Make a table for each of these dependencies with the left hand of the dependency being a key. R1=ABCD, R2=CEF, R3=GAF. If none of these tables contains the original key (GB) make a table for the key. So we have as our decompostion: R1=ABCD, R2=CEF, R3=GAF, R4=GB. Claim: R1,R2,R3,R4 forms a lossless join, 3nf, dependency preserving decompostion. Proof: Each of the minimal (necessary) dependencies appears with one of the tables so they are all there. No other dependencies are introduced that were not in our minimal set so we did not add new constraints. None of the tables will have any depencency which violates 3nf (i.e. has a left side that is neither a superkey nor a proper subset of key)

The tables can be re-joined in such a way that whenever two tables are joined, the attributes they have in common is a key, or superkey of one of them. Intuitively, if XY is the original key. Then any dependency is of the form XY->..., X->..., XW->..., or W->,... In the first two cases the tables can be joined with our table that has the key, in case 3 W must be dependent directly or transitively on XY or X or Y otherwise W would have to be in the key as well. Finally in case W->..., W must be transitively dependent on XY or part of XY otherwise W would have to be part of the key. In any other these cases an appropriate join order can be found. For example: Lets join the 4 relations to reform the original: R34=R3*R4=GABF, i.e. GB * GAF have G in common which is a superkey of GAF. R134=R1*R34=GABCDF, i.e. ABCD * GABF have AB in common which is a superkey of ABCD R1234 = R2 * R134=GABCDEF, i.e. CEF * GABCDF have CF in common which is a superkey of CEF

Example 2 Example: Consider R(ABCDEF) subject to the following FDs F = { A BCE, B DE, ABD CF, AC DE }

/Give a lossless, dependeny preserving 3NF decomposition also / Solution: Find a canonical cover for ,F. 1. After Left Reduction: Rule ABD CF becomes A CF Rule AC DE becomes A DE F= { A BCE, B DE, A CF, A DE }
2.

After Right Reduction: Rule A BCE becomes A B Rule A DE becomes A E F= { A B, B DE, A CF, A E }

3. After Removing Redundant Rule(s): Rule A E is redundant (Trans. on A B, B E ) F= { A B, B DE, A CF } same as F= { A BCF, B DE }

Final Partition = ( ABCF, BDE )