Fuzzy Sets: Foundations and Applications

• Fuzzy sets are collections defined by membership functions mapping elements to [0,1], offering a graded notion of belonging that generalizes crisp sets.
• They employ algebraic operations—such as max/min union, intersection, and alternative summation—and extend to various generalizations including interval and type-2 fuzzy sets.
• Fuzzy sets underpin diverse applications across possibility theory, topology, machine learning, and quantum formulations, providing a flexible tool for uncertainty modeling.

Fuzzy sets are set-theoretic objects in which membership is graded rather than binary. For a nonempty universe $X$, a fuzzy set $A$ is determined by a membership function $\mu_A:X\to[0,1]$, equivalently $A=\{(x,\mu_A(x))\mid x\in X\}$; crisp sets are recovered when $\mu_A(x)\in\{0,1\}$ for all $x$ (Fujita et al., 12 Mar 2026). In the literature represented here, fuzzy sets are described both as the cornerstone of a non-additive uncertainty theory, namely possibility theory, and as a basis for developments in category theory, topology, algebra, analysis, machine learning, and uncertainty-oriented generalizations (Kirişci, 2017, Jardine, 2019).

1. Foundational definition and algebraic operations

The standard formal setting identifies the collection of fuzzy sets on $X$ with

$FS(X)=\mathrm{Hom}(X,[0,1]).$

On this space, the pointwise order is

$A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$

and the standard Zadeh operators are defined pointwise by

$$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$

With these operations, $A$0 is a complete distributive lattice (Lobillo et al., 2017, Fujita et al., 12 Mar 2026).

This pointwise formulation is the canonical type-1 model. It encodes partial belonging by a single scalar degree $A$1 for each element, and it is the reference object from which many later constructions depart. In the comparative hierarchy summarized in the survey literature, fuzzy sets are the simplest “one-coordinate” case; intuitionistic fuzzy sets add non-membership, neutrosophic sets add explicit indeterminacy, and plithogenic sets index appurtenance by attribute values together with contradiction information (Fujita et al., 12 Mar 2026).

The same foundational model also supports alternative algebraic operators. One cited example is the algebraic sum,

$A$2

together with bounded sum and algebraic product, although the basic fuzzy-set preliminaries emphasize the max/min/complement triple as the standard presentation (Fujita et al., 12 Mar 2026).

2. Membership functions, $A$3-cuts, and fuzzy numbers

A substantial part of fuzzy-set practice is organized around explicit membership-function families. The survey literature records piecewise-linear examples such as a “comfortable” fuzzy set on $A$4 that is $A$5 outside $A$6, rises linearly on $A$7, equals $A$8 on $A$9, and decreases linearly on $\mu_A:X\to[0,1]$0, as well as a trapezoidal “premium customer” fuzzy set with breakpoints $\mu_A:X\to[0,1]$1 (Fujita et al., 12 Mar 2026). These constructions exemplify the standard use of piecewise-linear membership maps in decision-making, control, and linguistic modeling.

A particularly important numerical specialization is the triangular fuzzy number. It is represented by

$\mu_A:X\to[0,1]$2

with membership function

$\mu_A:X\to[0,1]$3

For triangular fuzzy numbers, addition and scalar multiplication are defined componentwise, and with the metric

$\mu_A:X\to[0,1]$4

the resulting space is a complete metric space (Kirişci, 2017).

Quantitative comparison of fuzzy sets is frequently expressed through $\mu_A:X\to[0,1]$5-cuts. For fuzzy sets $\mu_A:X\to[0,1]$6 and $\mu_A:X\to[0,1]$7 on a real universe,

$\mu_A:X\to[0,1]$8

When these cuts are compact intervals $\mu_A:X\to[0,1]$9 and $A=\{(x,\mu_A(x))\mid x\in X\}$0, a signed Hausdorff distance can be defined by

$A=\{(x,\mu_A(x))\mid x\in X\}$1

This yields directional variants of the Ralescu–Ralescu and Chaudhuri–Rosenfeld distances, with the sign indicating whether $A=\{(x,\mu_A(x))\mid x\in X\}$2 lies to the right or to the left of $A=\{(x,\mu_A(x))\mid x\in X\}$3 “on average” (McCulloch et al., 2013).

The distance framework extends beyond normal convex fuzzy sets. The cited construction includes non-normal sets, where some $A=\{(x,\mu_A(x))\mid x\in X\}$4-cuts may be empty, and non-convex sets, where an $A=\{(x,\mu_A(x))\mid x\in X\}$5-cut may decompose into disjoint intervals whose intervalwise directional distances are averaged (McCulloch et al., 2013). This makes explicit that comparison in fuzzy-set theory is not exhausted by unsigned similarity or overlap; directionality can itself be part of the formal data.

3. Sheaf-theoretic and categorical formulations

A more structural formulation replaces the codomain $A=\{(x,\mu_A(x))\mid x\in X\}$6 by a locale $A=\{(x,\mu_A(x))\mid x\in X\}$7. In that setting, a fuzzy set over $A=\{(x,\mu_A(x))\mid x\in X\}$8 is simply a function

$A=\{(x,\mu_A(x))\mid x\in X\}$9

The category $\mu_A(x)\in\{0,1\}$0 has these functions as objects, and morphisms $\mu_A(x)\in\{0,1\}$1 consist of a map $\mu_A(x)\in\{0,1\}$2 such that

$\mu_A(x)\in\{0,1\}$3

Barr’s key insight, as developed in the presheaf-theoretic treatment, is that such a $\mu_A(x)\in\{0,1\}$4 can be encoded as a sheaf of injections over the augmented locale $\mu_A(x)\in\{0,1\}$5, leading to an equivalence

$\mu_A(x)\in\{0,1\}$6

where $\mu_A(x)\in\{0,1\}$7 denotes the sheaves of monomorphisms (Jardine, 2019).

For dense total orders $\mu_A(x)\in\{0,1\}$8, the presheaf approach makes limits and colimits explicit. If $\mu_A(x)\in\{0,1\}$9 in $x$0, then for $x$1 its membership grade is

$x$2

If $x$3, then for $x$4 its grade is

$x$5

Accordingly, $x$6 is complete and cocomplete, and the meet/join formulas above give explicit grades for limits and colimits (Jardine, 2019).

Boolean localization further yields a stalk theory when $x$7 approximates the structure of a closed interval in the real line. For a sheaf $x$8 on $x$9 and $X$0, the stalk is

$X$1

A morphism of sheaves of monomorphisms is an isomorphism if and only if it induces bijections on all stalks (Jardine, 2019). This suggests that fuzzy membership can be treated as local data varying over an ordered base, not only as a pointwise scalar annotation.

The Vietoris–Rips construction supplies an explicit bridge to TDA. For a finite data cloud $X$2, the simplicial grading

$X$3

defines a simplicial fuzzy set in $X$4, and under Barr’s equivalence the associated sheaf is exactly the Vietoris–Rips system $X$5 (Jardine, 2019). In that formulation, the generic fibre is the full simplex $X$6, and the paper further shows that an inclusion of data clouds $X$7 induces an isomorphism of simplicial sheaves $X$8 if and only if $X$9.

4. Generalizations and lattice embeddings

The basic model $FS(X)=\mathrm{Hom}(X,[0,1]).$0 has been generalized in several directions, often by enlarging the codomain of the membership map while retaining pointwise lattice structure (Lobillo et al., 2017, Seth et al., 2020).

Model	Representation	Structural note
FS	$FS(X)=\mathrm{Hom}(X,[0,1]).$1	Complete distributive lattice
IVFS	$FS(X)=\mathrm{Hom}(X,[0,1]).$2	Lattice under pointwise interval operations
SVFS	$FS(X)=\mathrm{Hom}(X,[0,1]).$3	Complete lattice only if $FS(X)=\mathrm{Hom}(X,[0,1]).$4-values are allowed
CVFS	$FS(X)=\mathrm{Hom}(X,[0,1]).$5	Distributive lattice extending standard union and intersection
T2FS	$FS(X)=\mathrm{Hom}(X,[0,1]).$6	Complete lattice under levelwise Zadeh operators

A central issue in these extensions is that naive singleton embeddings do not always preserve the desired order or operations. In particular, seeing fuzzy sets as set-valued or type-2 fuzzy sets whose membership degrees are singletons is not order-preserving, and the naive singleton map from $FS(X)=\mathrm{Hom}(X,[0,1]).$7 to $FS(X)=\mathrm{Hom}(X,[0,1]).$8 does not preserve intersection (Lobillo et al., 2017). The cited remedy is a family of explicit lattice embeddings, including

$FS(X)=\mathrm{Hom}(X,[0,1]).$9

for $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$0, together with embeddings $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$1, $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$2, $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$3, and $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$4, yielding a commutative chain of lattice embeddings (Lobillo et al., 2017).

Closed-valued fuzzy sets $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$5 were introduced precisely to repair the hesitant-fuzzy setting and to allow mixed membership degrees such as closed intervals and finite sets. The order $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$6 on $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$7 and the operations $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$8 make $A\le_F B \iff \forall x\in X:\;A(x)\le B(x),$9 a distributive lattice, and pointwise lifting gives a $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$0-lattice that extends standard union and intersection on ordinary fuzzy sets (Lobillo et al., 2017).

A further direction uses interval type-2 fuzzy sets. An interval type-2 fuzzy set $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$1 is specified by secondary membership intervals $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$2, and its footprint of uncertainty is

$$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$3

Within hesitant fuzzy linguistic term sets, each linguistic term can be assigned an interval type-2 trapezoidal membership function, producing type-2 fuzzy sets based hesitant fuzzy linguistic term sets for linguistic decision making (Seth et al., 2020). In that framework, the lower and upper membership functions encode intra-uncertainty, while hesitant collections of terms encode inter-uncertainty.

5. Construction rules, cardinality, and computability

One recent line of work studies objective construction rules for building new fuzzy sets from known fuzzy or classical sets on a finite universe’s superstructure. If $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$4 is a fuzzy set on a finite $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$5, and $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$6 is constructed on $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$7, the cited rules include $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$8 and, for nonempty $$\mu_{A\cup B}(x)=\max(\mu_A(x),\mu_B(x)),\qquad \mu_{A\cap B}(x)=\min(\mu_A(x),\mu_B(x)),\qquad \mu_{\overline A}(x)=1-\mu_A(x).$$9,

$A$00

With scalar cardinality defined by

$A$01

the fuzzy power-set cardinality theorem states

$A$02

This extends the classical identity $A$03 to the fuzzy setting (Zhou, 2024).

The same paper presents a construction for realizing an arbitrary membership value $A$04 by nested singletons $A$05. If $A$06, then the recursive relations are

$A$07

The corresponding theorem states that every $A$08 is associated with a unique binary sequence satisfying $A$09, and the paper gives a greedy algorithm for constructing that sequence (Zhou, 2024).

A different computational perspective is supplied by the Fuzzy Ershov Hierarchy. Here a fuzzy subset of $A$10 is a map $A$11, usually with $A$12. A fuzzy set is $A$13-c.e. if it admits a computable approximation $A$14 converging to $A$15, starting at $A$16, and changing direction of monotonicity at most $A$17 times for each $A$18 (Bazhenov et al., 2021). The hierarchy does not collapse; each fuzzy $A$19-c.e. set can be represented as a Boolean combination of fuzzy c.e. sets; and, contrary to the classical case, the Fuzzy Ershov Hierarchy does not exhaust the class of all $A$20 fuzzy sets (Bazhenov et al., 2021). This marks an important limit phenomenon: finite mind-change behavior is only one part of fuzzy computability.

6. Analytical, learning-theoretic, and contemporary applications

Another strand of work re-engineers membership functions directly. The $A$21-function,

$A$22

deforms the crisp box-function into a smooth fuzzy membership profile controlled by a fuzziness parameter $A$23. It satisfies

$A$24

and provides a $A$25 family of memberships and aggregation operators (García-Morales, 2017). The applications developed in that framework include a nonlinear switching function, wallpaper and frieze patterns, symmetry-breaking controlled by $A$26, and a theorem for shaping prescribed limit cycles in smooth deterministic nonlinear dynamical systems (García-Morales, 2017).

In machine learning, kernels on fuzzy sets turn membership functions into inputs for RKHS-based methods. Four kernel classes are identified: cross-product, intersection, non-singleton, and distance-based kernels. Their purpose is to define positive-definite similarity measures on $A$27-valued functions so that standard kernel algorithms such as SVMs, kernel PCA, kernel k-means, kernel ridge regression, and two-sample tests can be applied directly to fuzzy-set inputs (Guevara et al., 2019). This is one of the clearest examples of fuzzy sets being reinterpreted as objects in a modern statistical learning pipeline.

Noise-robust approximation is another active direction. In fuzzy rough sets, the standard lower and upper approximations use $A$28 and $A$29, and the cited study emphasizes that this creates sensitivity to outliers. Fuzzy quantifier-based fuzzy rough sets replace rigid universal/existential semantics by general unary and binary quantification models, and the paper highlights Yager’s Weighted Implication-based binary quantification model as both a significant improvement on VQFRS and a worthy competitor to OWAFRS (Theerens et al., 2022). Empirically, the study uses 24 real datasets, balanced accuracy, 5-fold stratified CV, and $A$30 label noise (Theerens et al., 2022).

A recent extension moves from scalar degrees to quantum states. In that framework, the unit interval is embedded into the Bloch ball via density matrices,

$A$31

and a quantum fuzzy set is a map from $A$32 into density matrices, with the classical limit characterized by simultaneous diagonalizability (Mannucci, 22 Mar 2026). The same work introduces a category $A$33, a global Q-Matrix whose local sections are obtained by partial trace, and an obstruction to a fully internal Frobenius-algebra treatment caused by the no-cloning constraint (Mannucci, 22 Mar 2026). This suggests that the contemporary theory of fuzzy sets is no longer confined to graded subsets of a universe, but is also being reformulated in categorical and quantum-informational terms.