FUZZY SYSTEMS
Tolerance and Equivalence Relations
Relations exhibit certain properties as discussed below. Consider the
following figure.
Fig. 1 Three vertex graphs for the properties (a) Reflexivity, (b) Symmetry and (c) Transitivity
This figure describes a universe of three elements, which are labeled as the vertices of
this graph, 1, 2, and 3, or in set notation, x 1, x2, x3 . The properties discussed here are
reflexivity, symmetry, and transitivity.
Reflexivity: When a relation is reflexive every vertex in the graph originates a single loop, as
shown in Figure (a).
Symmetry: If a relation is symmetric, then in the graph for every edge pointing (the arrows on
the edge lines in Figure b) from vertex i to vertex j (i, j 1, 2, 3), there is an edge pointing in
the opposite direction, that is, from vertex j to vertex i .
Transitivity: When a relation is transitive, then for every pair of edges in the graph, one
pointing from vertex i to vertex j and the other from vertex j to vertex k (i, j, k 1, 2, 3),
there is an edge pointing from vertex i directly to vertex k , as seen in Figure c (e.g., an arrow
from vertex 1 to vertex 2, an arrow from vertex 2 to vertex 3, and an arrow from vertex 1
to vertex 3).
Crisp Equivalence Relation
A relation R on a universe X can also be thought of as a relation from X to
X. The relation R is an equivalence relation if it has the following three
properties: (1) reflexivity, (2) symmetry, and (3) transitivity.
For example, for a matrix relation, the following properties will hold:
Reflexivity (xi , xi ) ∈ R or χR(xi , xi ) = 1. Eq.
(a)
Symmetry (xi , xj ) ∈ R −→ (xj , xi ) ∈ R Eq.
(b)
or χR(xi , xj ) = χR(xj , xi ).
Transitivity (xi , xj ) ∈ R and (xj , xk ) ∈ R −→ (xi , x k ) Eq.
∈R (c)
or χR(xi , xj ) and χR(xj , xk ) = 1 −→ χR(xi , xk )
= 1.
Crisp Tolerance Relation
A tolerance relation R (also called a proximity relation) on a universe X is
a relation that exhibits only the properties of reflexivity and symmetry. A
tolerance relation, R, can be reformed into an equivalence relation by at most (n
1) compositions with itself, where n is the cardinal number of the set defining
R, in this case X, that is
Rn−1 = R1 ◦ R1 ◦· · · ◦ R1 = R
Example:
1) Suppose in an airline transportation system we have a universe composed of
five elements: the cities Omaha, Chicago, Rome, London, and Detroit. The
airline is studying locations of potential hubs in various countries and must
consider air mileage between cities and takeoff and landing policies in the
various countries. These cities can be enumerated as the elements of a set,
that is,
X = {x1, x2, x3, x4, x 5 }= {Omaha, Chicago, Rome, London, Detroit}.
Further, suppose we have a tolerance relation, R1, that expresses
relationships among these cities:
This relation is reflexive and symmetric. The graph for this tolerance
relation would involve five vertices (five elements in the relation), as shown
in Figure 2.
Fig. 2 Five-vertex graph of tolerance relation (reflexive and symmetric)
The property of reflexivity (diagonal elements equal unity) simply
indicates that a city is totally related to itself. The property of symmetry
might represent proximity: Omaha and Chicago (x1 and x2) are close (in a
binary sense) geographically, and Chicago and Detroit (x2 and x5) are close
geographically.
This relation, R1, does not have properties of transitivity, for example,
(x1 , x2 ) ∈ R1 (x2 , x5 ) ∈ R1 but (x1 , x5 ) ∈/ R1 .
Using Eq. Rn−1 = R1 ◦ R1 ◦· · · ◦ R1 = R
Now, we see in this matrix that transitivity holds, that is, (x1, x5) R1, and R is an
equivalence relation.
Fig. 3 Five-vertex graph of equivalence relation (reflexive, symmetric, and transitive)
FUZZY TOLERANCE AND EQUIVALENCE RELATIONS
A fuzzy relation, R, on a single universe X is also a relation from X to X. It is a fuzzy
equivalence relation if all three of the following properties for matrix relations define it:
Reflexivity μR(x=i , xi )= 1. Eq. (d)
∼
Symmetry μR(xi , xj ) = μR(xj , xi ). Eq. (e).
∼ ∼
Transitivity μR(xi , xj ) = λ1 and μR(xj , xk) = λ2 −→ μR(xi , xk) = λ, Eq. (f)
∼ ∼ ∼
where λ ≥ min[λ1, λ2].
It can be shown that any fuzzy tolerance relation, R1, that has properties of
∼
reflexivity and symmetry can be reformed into a fuzzy equiva−lence
relation by at most (n-1) compositions, just as a crisp tolerance relation can
be reformed into a crisp equivalence relation.
That is,
Rn−1 = R1 ◦ R1 ◦· · · ◦ R1 = R
Example 2: Suppose, in a biotechnology experiment, five potentially new strains of
bacteria have been detected in the area around an anaerobic corrosion pit on a new aluminum–
lithium alloy used in the fuel tanks of a new experimental aircraft. In order to propose
methods to eliminate the biocorrosion caused by these bacteria, the five strains must first be
categorized. One way to categorize them is to compare them to one another. In a pairwise
comparison, the following “similarity” relation, R1, is developed.
For example, the first strain (column 1) has a strength of similarity to
the second strain of 0.8, to the third strain a strength of 0 (i.e., no relation),
to the fourth strain a strength of 0.1, and so on. Because the relation is for
pairwise similarity it will be reflexive and symmetric.
Hence,
The system with relation R1 is reflexive and symmetric. But not transitive.
For example
Thus applying the fuzzy composition on relation R1.
By Max-Min composition,
Thus doing one more fuzzy composition between R12 and R1
R= R13 = R12 o R1
Here
Eq. (g)
The relation R is reflexive, symmetric and transitive and satisfies the conditions of transitivity
as clear from Eq. (g).
Thus the fuzzy relation R formed by composition is an equivalence relation.
OTHER FORMS OF THE COMPOSITION OPERATION
Max– min and max– product (also referred to as max–dot ) methods of
composition of fuzzy relations are the two most commonly used techniques. Many
other techniques are mentioned in the literature. Each method of composition of
fuzzy relations reflects a special inference machine and has its own significance
and applications. The max– min method is the one used by Zadeh in his original
paper on approximate reasoning using natural language if– then rules. Many have
claimed, since Zadeh’s introduction, that this method of composition effectively
expresses the approximate and interpolative reasoning used by humans when they
employ linguistic propositions for deductive reasoning (Ross, 1995).
The following additional methods are among those proposed in the literature for
= ◦ B A R, where A is the input, or antecedent defined on
the composition operation
∼ ∼ ∼ ∼
the universe X, B is the output, or co nse quent defi ned on universe Y, and R is
∼
a fuzzy relation characterizin g the relationship between specific inputs (x ) and
∼
specifi c outputs (y ):
Eg (h)
·
where f ( .) is a logistic function (such as a sigmoid or a step function) that limits
the value of the function within the interval [0, 1]. This composition method is
commonly used in applications of artificial neural networks for mapping
between parallel layers in a multilayer network.
Exercise
1. Determine the values of the composition B = A o R using the min-max, max-max,
min-min, max-average, sum-product fuzzy compositions.
