Unit 3: α-Cuts, Decomposition Theorems & The Extension Principle

1. Connecting Fuzzy Sets to Classical Subsets

One of the most powerful and fruitful methodologies in fuzzy mathematics is the $\alpha$-cut representation. Rather than treating a fuzzy set as an exotic non-classical entity, $\alpha$-cuts decompose any fuzzy set into an ordered family of standard, crisp subsets.

Definition 3.1 ($\alpha$-Cut and Strong $\alpha$-Cut): Let $A$ be a fuzzy set on universe $X$ with membership function $\mu_A(x)$, and let $\alpha \in [0, 1]$.

1. The $\alpha$-cut (or weak $\alpha$-cut / $\alpha$-level set) of $A$, denoted $A_\alpha$ or $[A]_\alpha$, is the crisp set of all elements whose membership grade is at least $\alpha$:

1. The strong $\alpha$-cut of $A$, denoted $A_{\alpha^+}$ or $(A)_\alpha$, is the crisp set of all elements whose membership grade strictly exceeds $\alpha$:

Boundary Cases:

• For $\alpha = 0$: $A_0 = \{x \in X \mid \mu_A(x) \ge 0\} = X$ (the entire universe).
• The strong 0-cut is the Support: $A_{0^+} = \{x \in X \mid \mu_A(x) > 0\} = \text{Supp}(A)$.
• The 1-cut is the Core: $A_1 = \{x \in X \mid \mu_A(x) = 1\} = \text{Core}(A)$.
• The strong 1-cut is always empty: $A_{1^+} = \{x \in X \mid \mu_A(x) > 1\} = \emptyset$.

2. Level Set of a Fuzzy Set

The set of all distinct values assumed by the membership function $\mu_A(x)$ is called the level set (or spectrum) of $A$:

If $X$ is finite, $\Lambda(A) = \{\alpha_1, \alpha_2, \dots, \alpha_k\}$ is a finite ordered subset of $[0, 1]$.

1. The Monotonicity Property of Cuts

Theorem 3.1 (Cut Monotonicity): For any fuzzy set $A$ and any $\alpha, \beta \in [0, 1]$:

Furthermore, for any $\alpha \in [0, 1)$:

Proof:

Let $x \in A_\beta$. Then $\mu_A(x) \ge \beta$. Since $\beta \ge \alpha$, transitivity yields $\mu_A(x) \ge \alpha$, which means $x \in A_\alpha$. Hence $A_\beta \subseteq A_\alpha$. Similarly, if $x \in A_{\alpha^+}$, then $\mu_A(x) > \alpha \ge \alpha$, so $x \in A_\alpha$. $\blacksquare$

2. Commutativity with Set Theoretic Operations

Theorem 3.2 (Preservation of Intersections and Unions): For any fuzzy sets $A$ and $B$ under standard Zadeh operations, and for any $\alpha \in [0, 1]$:

1. $(A \cap B)_\alpha = A_\alpha \cap B_\alpha$
2. $(A \cup B)_\alpha = A_\alpha \cup B_\alpha$
3. $(A \cap B)_{\alpha^+} = A_{\alpha^+} \cap B_{\alpha^+}$
4. $(A \cup B)_{\alpha^+} = A_{\alpha^+} \cup B_{\alpha^+}$

Proof of (1):

Non-Preservation of Complements:

Does $(A^c)_\alpha = (A_\alpha)^c$? NO! In fact:

Therefore:

Complementation inverts the cut level and swaps weak cuts with strong cuts!

1. Scaling a Crisp Set by a Scalar $\alpha$

To reconstruct a fuzzy set from its crisp cuts, we define the product of a scalar $\alpha \in [0, 1]$ and a crisp set $C \subseteq X$:

Definition 3.2 (Special Fuzzy Set $\alpha C$): For $\alpha \in [0, 1]$ and $C \subseteq X$, the fuzzy set $\alpha C$ (or $\alpha \cdot C$) is defined by the membership function:

2. The First Decomposition Theorem

Theorem 3.3 (First Decomposition Theorem): For any fuzzy set $A$ on universe $X$:

where $\bigcup$ denotes the standard fuzzy union ($\sup$ operator). That is, for all $x \in X$:

Proof:

For any chosen point $x \in X$, let $\mu_A(x) = a \in [0, 1]$. Evaluate the right-hand supremum:

Recall that $x \in A_\alpha \iff \mu_A(x) \ge \alpha \iff a \ge \alpha$. Therefore:

Substituting into the product:

Taking the supremum over all $\alpha \in [0, 1]$:

3. The Second Decomposition Theorem

Theorem 3.4 (Second Decomposition Theorem): For any fuzzy set $A$ on universe $X$:

where $\Lambda(A)$ is the level set of $A$.

This proves that ANY fuzzy set is completely and losslessly determined by its crisp $\alpha$-cuts! Every theorem in fuzzy set theory can be proved by proving it for its crisp cuts.

1. Motivation of the Extension Principle

Suppose we have a crisp mathematical function $f: X \to Y$ (for example, $f(x) = x^2$ or $f(x) = \sin x$), and we wish to evaluate $f$ when the input is not a crisp number, but a fuzzy set $A$ on $X$. How do we compute the induced fuzzy set $B = f(A)$ on $Y$? In 1975, Lotfi A. Zadeh formulated the Extension Principle, which is considered the single most important operational engine in fuzzy mathematics.

2. Formal Definition of the Extension Principle

Definition 3.3 (Zadeh's Extension Principle): Let $f: X \to Y$ be a crisp mapping from universe $X$ to universe $Y$. Let $A$ be a fuzzy set on $X$ with membership function $\mu_A(x)$. The mapping $f$ induces a fuzzy set $B = f(A)$ on $Y$ whose membership function $\mu_B(y)$ is defined for all $y \in Y$ by:

where $f^{-1}(y) = \{x \in X \mid f(x) = y\}$ is the preimage of $y$.

Special Cases:

1. One-to-One (Invertible) Mapping: If $f$ is injective, each $y \in f(X)$ has a unique preimage $x = f^{-1}(y)$, so:

2. Many-to-One Mapping: If multiple distinct points $x_1, x_2, \dots$ map to the same $y$, the principle assigns $y$ the maximum (or supremum) of their membership grades:

(Optimistic principle: an outcome $y$ is as possible as the most possible input that can produce it!)

1. Extension Principle for Multi-Argument Functions

Let $f: X_1 \times X_2 \times \dots \times X_n \to Y$ be a crisp multi-variable function (e.g. addition $f(x_1, x_2) = x_1 + x_2$, or multiplication $f(x_1, x_2) = x_1 \cdot x_2$). Let $A_1, A_2, \dots, A_n$ be fuzzy sets defined on universes $X_1, X_2, \dots, X_n$.

Definition 3.4 (Multi-Variable Extension Principle): The image fuzzy set $B = f(A_1, \dots, A_n)$ on $Y$ has membership function:

where $f^{-1}(y) = \{(x_1, \dots, x_n) \mid f(x_1, \dots, x_n) = y\}$.

2. Cut-Level Preservation of the Extension Principle

Theorem 3.5 (Nguyen's Theorem / Cut Preservation): If $X$ and $Y$ are Euclidean spaces and $f: X \to Y$ is continuous, and if $A$ has compact $\alpha$-cuts, then the $\alpha$-cut of the extended fuzzy set $f(A)$ is identically the crisp image of the $\alpha$-cut of $A$:

Similarly, for multi-argument functions:

Profound Consequence:

This theorem guarantees that evaluating complicated non-linear functions of fuzzy sets via the extension principle is completely mathematically equivalent to evaluating standard interval analysis on each $\alpha$-cut level! This is the foundational cornerstone of all fuzzy arithmetic.