Normalization

Logical Design of Relational Schemas

Erős Levente, 2024.

What is the problem with this
relation?
productID supplierName supplierAddr price
1 Erik és Fia Bt. Budapest, Fő utca 1 1000 HUF
2 Erik és Fia Bt. Budapest, Fő utca 1 3000 HUF
3 NASA 300 E. Street SW 90000000 HUF
WDC
4 Erik és Fia Bt. Budapest, Fő utca 1 2500 HUF
5 Erik és Fia Bt. Budapest, Fő utca 1 5000 HUF
6 NASA 300 E. Street SW 180000000 HUF
WDC
7 NASA 300 E. Street SW 70000000 HUF
WDC
8 NASA 300 E. Street SW 60000000 HUF
WDC
9 NASA 300 E. Street SW 7770000000 HUF
WDC
Introduction
• another way for creating relational schemas
• which was the 1st? ER  relational mapping
• there might be unnecessary data
• When is data unnecessary –
is supplier address (supplierAddr) unnecessary?
• Product(productID, supplierName, supplierAddr, price)
• supplierAddr is REDUNDANT (it can be figured out)
• Product(productID, supplierName, price)
Supplier(supplierName, supplierAddr)
• supplierAddr is NOT REDUNDANT
• check out next slide for redundancy
Examples
productID supplierName supplierAddr price
1 Erik és Fia Bt. Budapest, Fő utca 1 1000 HUF
2 Erik és Fia Bt. Budapest, Fő utca 1 3000 HUF

productID supplierName price

1 Erik és Fia Bt. 1000 HUF
2 Erik és Fia Bt. 3000 HUF

supplierName supplierAddr
Erik és Fia Bt. Budapest, Fő utca 1
Why is redundancy a problem?
• It causes anomalies
• Insertion anomaly
• can’t insert a piece of data without other pieces of data
• can’t insert supplier (name+address) without at least one product
• Deletion anomaly
• by deleting some data, other data is deleted too
• if you delete the last product of the supplier, its address disappears too
• Update anomaly
• changing the value of an attribute in one row means changing the
same attribute in other rows
• supplier changes address  all its supplierAddr values shall be changed
• resolved by decomposition (not discussed)
Functional dependencies (FDs)
• Example:
• supplier name defines supplier address ==
• supplier address is functionally dependent on supplier name
==
• if supplier name is repeated in a relation, supplier address is
repeated too
• Definition:
• Given is relational schema R that includes attribute sets X and
Y.
Y is functionally dependent on X, i.e. XY
• if in the case of any relation r(R)
• if an arbitrary pair of elements t and t’ of r
• if the values of X in t equal the values of X in t’, i.e. t[X]==t’[X]
Functional dependencies (FDs)
• Remarks

• XY can stand if X has no repeating values.

• occassional dependencies (each Bob in the table has brown

hair) are not real dependencies (each dentist has a university
diploma)

• FD XY only causes redundancy if X can be repeated,

otherwise not
Some definitions
• If X,Y ⊆ R and X → Y, but ∄X ′ ⊂ X so that X ′ → Y, then X
is the determinant of Y
• minimal (not minimum) set of attributes that define Y
• If X,Y ⊆ R and X → Y, but ∄X ′ ⊂ X so that X ′ → Y, then Y
is fully dependent on X (full dependency)
• the left side of the FD cannot be further reduced
• If X,Y ⊆ R and X → Y, and ∃X ′ ⊂ X so that X ′ → Y, then Y
is partially dependent on X (partial dependency)
• the left side of the FD can be reduced and will still define the
right side
Keys
• X is a superkey iff X defines the schema, i.e. XR
• X is a key if it is a superkey and none of its real subsets are
superkeys
• X is a key if the schema is fully dependent on it
• one key is primary key
• rest of the keys are candidate keys
• this distinction has no mathematical relevance

• A is a primary attribute if it is included by at least one

of the keys
• A is a secondary attribute if it is included by none of
the keys
True and deducible FDs
• The functional dependency set F includes all the
dependencies of a schema R
• A FD XY is true over a functional dependency set F if it
stands on all those relations r(R) on which the
dependencies of F stand. Notation: F ⊨ X → Y
• A FD XY is deducible from a functional dependency
set F if it can be obtained by repeatedly applying
Armstrong’s axioms on the FDs in F. Notation: F ⊢ W →
Z
• Theorem: Iff a FD is true over F then it is deducible from
F.
Armstrong’s axioms
• X ⊆ Y ⊨ Y → X, trivial dependency
• X → Y ∧ Y → Z ⊨ X → Z, transitivity
• this has (almost) nothing to do with
transitive dependency
• X → Y ⊨ XZ → YZ, expandability
Normal forms
• Goal: reducing redundancy
• Normal forms are defined
• each excludes certain types of redundancy
• If a relational schema R is in a given normal form,
then an arbitrary relation r(R) is guaranteed to exclude
the given types of redundancy
 1NF
First normal form, 1NF 2NF
3NF
0NF
• Definition: Each attribute is atomic
BCNF
• 0NF includes non-atomic attributes
• Question: When is an attribute atomic?
• if we don’t need parts of attribute values
• depends on application, e.g. date (YYYY-MM-DD)
• is it atomic or not?

• Further normal forms will assume that the 1NF condition

is met.
Second normal form, 2NF
• Definition: 1NF and no secondary attribute depends on
a real subset of a key
• Equivalent definition: Each secondary attribute is fully
dependent on each key
• Counter example:
R(A,B,C)
F={BC}
key: ?
• Why is this redundant?
Finding keys
• Classify attributes based FDs in the FD set:
• appear only on left hand side – included in each key (nobody defines
them)
• appear only on right hand side – included in none of the keys
(defined by others)
• apper on neither side – included in each key (are only defined by
themselves)
• appear on both sides – left right both none
the magic happens here 
observations&brute force find keys B C - A
• Example: R(A,B,C), F={BC}
• A and B in all keys. Let’s check what AB defines: ABABC (itself plus
C)
Second normal form, 2NF
• Definition: 1NF and no secondary attribute depends on
a real subset of a key
• Equivalent definition: Each secondary attribute is fully
dependent on each key
• Counter example:
R(A,B,C)
F={BC}
key: AB
• Why is this redundant?
Second normal form, 2NF
• Definition: 1NF and no secondary attribute depends on
a real subset of a key
• Equivalent definition: Each secondary attribute is fully
dependent on each key
A B C
• Counter example:
R(A,B,C) Jake Bp U
F={BC}
key: AB Alice Bp U
• Why is this redundant? Bob Szl R
Transitive dependencies
• NOT a single dependency but a dependency system
• NOT to be confused with transitivity
• the cause of redundancy (as far as we’re concerned in
this course)
• A is transitively dependent on X if there is a Y for which
• XY
• Y⇸X (Y doesn’t define X)
• YA
• A∉Y
• Homework: Think of how this causes redundancy
Third normal form, 3NF
• Definitions
• 1NF and for each XA X is a superkey or A is prime attribute
• 1NF and no secondary attribute depends on any key transitively
• Counter example: A B C
• R(A,B,C) 1 y a
• F={AB, BC} 2 y A
• key: A, all other attributes: secondary, 3 x B
and C transitively depends on A
• Question: What’s the highest normal formS of R? C Z
• Example for 3NF: Józsefhegyi Budapest 102
• R(S(treet), C(ity), Z(IP code)) = R(S,C,Z) utca 5
• F={SCZ, ZC} Józsefhegyi Törökbáli 204
utca nt 5
• Question: Leftover reundancy?
Törökvész út Budapest 102
Boyce-Codd normal form, BCNF
• Definitions
• 1NF and for each XA X is a superkey S Z
Józsefhegyi
• 1NF and no attribute depends on any key 102
utca 5
transitively
Józsefhegyi 204
• Example for 3NF: utca 5
• R1(S,Z); R2(Z,C) Törökvész út 102
C 5
Z
• F1={}; F2{ZC}
Budapest 102
• Question: Problems? 5
• Btw. any 2-attribute schema is BCNF Törökbáli 204
nt 5
Closures
• The closure of attribute set X is the largest attribute set
W for which XW stands. Formally: X+(F) = {A|A ∈ R and F
⊨ X → A}
• calculated in linear time:
• X+=X initially
• in each step
• find a FD the left hand side of which is in X+, while its right hand side is not
• add the right hand side to X+

• The closure of a functional dependency set F is the

largest set of those FDs which can be deduced from F using
Armstrong’s axioms. Formally: F+ = {X → Y |F ⊨ X → Y}
• calculated in exponential time
Closures, examples
• F={AB, BC}
• F+={AB, BC, AA, BB, CC, ABAB, ACAC, BCBC,
ABCABC, ABA, ABB, ACA, ACC, BCB, BCC, ABCA,
ABCB, ABCC, AC, ABC, ACB, ABAC,...}
• A+(F)=?
• A+(F)0=A
• A+(F)1=AB (based on AB)
• A+(F)2=ABC (based on BC)
• A+(F)3= A+(F)2, no further expansion is possible, we’re done.
• Question: How to determine whether AC is in F+ or not?
• Solution I: Find F+ and see – takes a long time
• Solution II: Calculate A+(F) and see whether C is in it – efficient