Crisp Sets, Fuzzy Sets & Membership Foundations

1. Classical Set Theory and the Law of Excluded Middle

In classical (Cantorian) set theory, an element $x$ from a universe of discourse $X$ either strictly belongs to a subset $A$ or does not. This rigid binarism is grounded in Aristotle's classical laws of thought:

1. The Law of Contradiction: $A \cap A^c = \emptyset$ (nothing can be both $A$ and not-$A$).
2. The Law of Excluded Middle (Tertium Non Datur): $A \cup A^c = X$ (everything must be either $A$ or not-$A$; there is no middle ground).

2. The Characteristic Function

Formally, any classical ("crisp") set $A \subseteq X$ is uniquely identified by its characteristic function $\chi_A: X \to \{0, 1\}$:

The set of all subsets of $X$ is the power set $\mathcal{P}(X) = 2^X$. The classical set operations are completely isomorphic to Boolean operations on characteristic functions:

• Intersection: $\chi_{A \cap B}(x) = \min(\chi_A(x), \chi_B(x)) = \chi_A(x) \cdot \chi_B(x)$
• Union: $\chi_{A \cup B}(x) = \max(\chi_A(x), \chi_B(x)) = \chi_A(x) + \chi_B(x) - \chi_A(x)\chi_B(x)$
• Complement: $\chi_{A^c}(x) = 1 - \chi_A(x)$

3. Limitations in Modeling Real-World Vagueness

While crisp sets are ideal for exact mathematics and discrete computing, human cognition and natural language deal inherently with imprecision, vagueness, and continuous gradation. Concepts such as "high temperature", "young person", "fast vehicle", or "safe following distance" do not possess abrupt, crisp boundaries:

• Is a person aged 29 years and 364 days "young", but instantaneously "not young" at midnight of their 30th birthday?
• This classical dilemma is known as the Sorites Paradox (the paradox of the heap).

To model such gradual transitions rigorously, Lotfi A. Zadeh (1965) generalized the characteristic function from the discrete binary codomain $\{0, 1\}$ to the real continuous unit interval $[0, 1]$.

1. Definition of a Fuzzy Set

In his seminal 1965 paper "Fuzzy Sets", Lotfi A. Zadeh introduced the fundamental concept:

Definition 1.1 (Fuzzy Set): Let $X$ be a non-empty classical set called the universe of discourse. A fuzzy set $A$ in $X$ is characterized by a membership function:

where the real number $\mu_A(x)$ represents the degree of membership (or grade of belonging) of element $x$ in the fuzzy set $A$.

• $\mu_A(x) = 1$ denotes full, unequivocal membership.

• $\mu_A(x) = 0$ denotes absolute non-membership.

• $0 < \mu_A(x) < 1$ represents intermediate, gradual degrees of membership.

Representation Notations:

1. Discrete Universe: If $X = \{x_1, x_2, \dots, x_n\}$ is discrete:

(Note: The summation $\sum$ and slash $/$ denote collection and pairing, NOT arithmetic addition or division!)

1. Continuous Universe: If $X$ is a continuous real domain (e.g. $\mathbb{R}$):

2. Fundamental Geometric Regions of a Fuzzy Set

Definition 1.2 (Support, Core, Height, and Boundary): Let $A$ be a fuzzy set in universe $X$ with membership function $\mu_A(x)$.

• Support of $A$: The crisp subset of $X$ containing all elements with strictly positive membership:

• Core (Kernel) of $A$: The crisp subset of $X$ containing all elements with full membership:

• Height of $A$: The supremum of the membership grades over the entire universe:

• Boundary of $A$: The region of intermediate, uncertain membership:

Normal vs Subnormal Fuzzy Sets:

• A fuzzy set $A$ is called normal if its height is exactly 1:

• Otherwise, if $h(A) < 1$, the fuzzy set is called subnormal. Any subnormal set with $h(A) > 0$ can be normalized via:

1. From Boolean Logic to Infinite-Valued Logic

Classical Boolean logic is restricted to truth values $T = \{0, 1\}$. In 1920, Jan Łukasiewicz introduced 3-valued logic ($T_3 = \{0, \frac{1}{2}, 1\}$) and subsequently generalized it to the continuum of truth values $T_\infty = [0, 1]$. Fuzzy logic is an infinite-valued logic in which truth values are degrees of truth in the closed interval $[0, 1]$.

2. Major Fuzzy Logic Valuations

Let $v(P) \in [0, 1]$ and $v(Q) \in [0, 1]$ denote the truth values of propositions $P$ and $Q$.

1. Standard Zadeh Logic:

• Negation: $v(\neg P) = 1 - v(P)$
• Conjunction ($\wedge$): $v(P \wedge Q) = \min(v(P), v(Q))$
• Disjunction ($\vee$): $v(P \vee Q) = \max(v(P), v(Q))$
• Implication (Kleene-Dienes): $v(P \implies Q) = \max(1 - v(P), v(Q))$

2. Łukasiewicz Logic:

• Conjunction (Bounded Product): $v(P \wedge_L Q) = \max(0, v(P) + v(Q) - 1)$
• Disjunction (Bounded Sum): $v(P \vee_L Q) = \min(1, v(P) + v(Q))$
• Implication (Łukasiewicz Implication):

3. Gödel-Dummett Logic:

• Implication (Gödel Residuated Implication):

4. Product (Goguen) Logic:

• Conjunction: $v(P \cdot Q) = v(P) \cdot v(Q)$
• Implication (Goguen Implication):

3. Failure of Classical Tautologies in Fuzzy Logic

In classical logic, $P \vee \neg P \equiv 1$ (Law of Excluded Middle) and $P \wedge \neg P \equiv 0$ (Law of Contradiction) are universal tautologies. In Zadeh fuzzy logic with $v(P) = 0.5$:

Thus, fuzzy sets do NOT generally satisfy the Law of Excluded Middle or the Law of Contradiction! This non-trivial property reflects the presence of inherent fuzziness.

1. Parametric Membership Function Families

In applications to engineering, control theory, and expert systems, membership functions are parameterized mathematically to enable efficient computation and gradient-based tuning.

2. Piecewise Linear Membership Functions

1. Triangular Membership Function $\text{trimf}(x; a, b, c)$:

Defined by three parameters $a < b < c$ where $b$ is the peak (core) and $[a, c]$ is the support:

2. Trapezoidal Membership Function $\text{trapmf}(x; a, b, c, d)$:

Defined by four parameters $a < b \le c < d$, where $[b, c]$ is the core and $[a, d]$ is the support:

3. Smooth Differentiable Membership Functions

3. Gaussian Membership Function $\text{gaussmf}(x; c, \sigma)$:

Characterized by center $c$ and standard deviation $\sigma > 0$:

• Core: $\{c\}$
• Support: $\mathbb{R}$ (infinitely supported, but practically localized within $c \pm 3\sigma$)
• Strictly smooth ($C^\infty$) and non-zero everywhere.

4. Generalized Bell Membership Function $\text{gbellmf}(x; a, b, c)$:

Characterized by half-width $a$, slope parameter $b > 0$, and center $c$:

5. Sigmoidal Membership Function $\text{sigmf}(x; a, c)$:

Used for representing open-ended linguistic concepts such as "large" or "high":

where $a$ governs the steepness of the transition at inflection point $c$.

1. Fuzzy Singletons

Definition 1.3 (Fuzzy Singleton): A fuzzy set $A$ on universe $X$ whose support is a single point $x_0 \in X$ with membership grade $\mu_A(x_0) = \alpha \in (0, 1]$ is called a fuzzy singleton:

If $\alpha = 1$, it represents the crisp point $x_0$ embedded into fuzzy set theory.

2. Convex Fuzzy Sets

In classical geometry, a set $S \subseteq \mathbb{R}^n$ is convex if for any two points $x_1, x_2 \in S$, the entire line segment connecting them lies in $S$: $\lambda x_1 + (1 - \lambda) x_2 \in S$ for all $\lambda \in [0, 1]$. In fuzzy set theory on $\mathbb{R}^n$, convexity is generalized through membership grades:

Definition 1.4 (Convex Fuzzy Set): A fuzzy set $A$ in $\mathbb{R}^n$ is called fuzzy convex if and only if for all $x_1, x_2 \in \mathbb{R}^n$ and all $\lambda \in [0, 1]$:

Critical Distinction:

A fuzzy convex membership function $\mu_A(x)$ is quasiconcave in the terminology of real analysis! Its graph does NOT need to be a convex curve; rather, its upper contour sets (level sets) must be convex crisp sets.

• If $X = \mathbb{R}$, a fuzzy set is convex if and only if its membership function is monotonically non-decreasing up to the core, and monotonically non-increasing thereafter (i.e. single-peaked or plateau-topped).

3. The Convexity Theorem for $\alpha$-Cuts

Theorem 1.1 (Equivalence with Convex $\alpha$-Cuts): A fuzzy set $A$ on $\mathbb{R}^n$ is fuzzy convex if and only if every crisp $\alpha$-cut:

is a classical convex set for all $\alpha \in (0, 1]$.

Proof:

1. Necessity ($\implies$): Assume $A$ is fuzzy convex. Let $\alpha \in (0, 1]$ and choose $x_1, x_2 \in A_\alpha$.

Then $\mu_A(x_1) \ge \alpha$ and $\mu_A(x_2) \ge \alpha$. By fuzzy convexity, for any $\lambda \in [0, 1]$:

Therefore, $\lambda x_1 + (1 - \lambda) x_2 \in A_\alpha$, proving $A_\alpha$ is a convex set.

1. Sufficiency ($\impliedby$): Assume $A_\alpha$ is convex for all $\alpha \in (0, 1]$.

For any $x_1, x_2 \in \mathbb{R}^n$, set $\alpha_0 = \min(\mu_A(x_1), \mu_A(x_2))$. If $\alpha_0 = 0$, then $\mu_A(\lambda x_1 + (1 - \lambda) x_2) \ge 0 = \alpha_0$ trivially. If $\alpha_0 > 0$, then $x_1 \in A_{\alpha_0}$ and $x_2 \in A_{\alpha_0}$. Since $A_{\alpha_0}$ is convex, $\lambda x_1 + (1 - \lambda) x_2 \in A_{\alpha_0}$ for all $\lambda \in [0, 1]$. This implies $\mu_A(\lambda x_1 + (1 - \lambda) x_2) \ge \alpha_0 = \min(\mu_A(x_1), \mu_A(x_2))$. $\blacksquare$