Intuitionistic Fuzzy Sets Overview

• Intuitionistic fuzzy sets are defined by a membership function, a nonmembership function, and a hesitation degree, extending classical and ordinary fuzzy set theory.
• They support componentwise operations, ordered algebraic structures, and have been applied to semigroups, Banach algebras, and topological studies.
• Their versatile framework underpins practical applications from decision-making and pattern recognition to uncertainty modeling in machine learning.

Intuitionistic fuzzy sets (IFS), introduced by Atanassov as a generalization of fuzzy sets, assign to each element both a membership degree and a non-membership degree, subject to a consistency constraint, and thereby retain an explicit residual term for hesitation or indeterminacy. In this form, IFS distinguish partial support, partial opposition, and unresolved uncertainty in a way that ordinary fuzzy sets do not. The framework has become a foundational model in algebraic, topological, soft-set, metric, and decision-theoretic developments, and it also serves as a reference point for later constructions such as q-rung orthopair, neutrosophic, and plithogenic models (Sardar et al., 2011, Wu et al., 2021, Fujita et al., 12 Mar 2026).

1. Formal definition and core semantics

An intuitionistic fuzzy set $A$ on a universe $X$ is specified by two functions

$\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$

where $\mu_A(x)$ is the membership degree and $\nu_A(x)$ is the non-membership degree, with

$$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$

The residual quantity

$\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$

is the hesitation degree, also called indeterminacy. Thus each element is represented by the triplet $(\mu_A(x),\nu_A(x),\pi_A(x))$ with total mass $1$ (Çağman et al., 2013).

This definition contains two important special cases. Classical crisp sets are recovered when $\mu_A(x)\in\{0,1\}$ and $X$0, so $X$1. Ordinary fuzzy sets are recovered by imposing $X$2 for all $X$3, again forcing zero hesitation. IFS therefore generalize fuzzy sets by permitting $X$4, which records explicit uncertainty rather than encoding everything through a single membership grade (Çağman et al., 2013).

Standard set-theoretic operations are adapted componentwise. For IFS $X$5 and $X$6 on the same universe,

$X$7

$X$8

$X$9

$\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$0

The induced order reverses the non-membership coordinate, reflecting the fact that larger sets have stronger membership but weaker non-membership (Çağman et al., 2013).

At the level of individual values, an intuitionistic fuzzy value (IFV) is a pair

$\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$1

Two scalar summaries recur throughout the literature: $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$2 called the score and accuracy, respectively. These functions support ranking, order construction, and metric design on the IFV space (Wu et al., 2021).

2. Ordered value spaces, algebraic structure, and topology

The space of all intuitionistic fuzzy values is

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

A major structural line of work studies $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$4 not merely as a set of admissible pairs but as an ordered algebraic object. One standard linear order is the Xu–Yager order $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$5, defined lexicographically by score and then accuracy: lower score implies lower rank, and ties in score are broken by accuracy. A second linear order, $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$6, is built from a similarity quantity

$\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$7

again refined by accuracy. These two orders are order-isomorphic via an explicit bijection $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$8 (Wu et al., 2021).

Under these linear orders, $\mu_A : X \to [0,1], \qquad \nu_A : X \to [0,1],$9 is a complete lattice. Moreover, a piecewise-defined negation on IFVs is shown to be a strong negation, and with the induced meet and join the resulting structure is a Kleene algebra. This result is significant because it places the basic truth-value space of IFS within a mature order-algebraic setting rather than treating it only as a numerical constraint domain (Wu et al., 2021).

The same work establishes nontrivial topological properties. With the order topology induced by either $\mu_A(x)$0 or $\mu_A(x)$1, the IFV space is compact and connected, but not separable and not metrizable. It is therefore not homeomorphic to $\mu_A(x)$2 with the usual topology, despite the superficial similarity of both spaces as bounded continua (Wu et al., 2021).

A related ordering problem arises for intuitionistic fuzzy numbers, where each number is a pair of fuzzy numbers satisfying $\mu_A(x)$3. A total order on the class of all such intuitionistic fuzzy numbers has been constructed using a double upper dense sequence in $\mu_A(x)$4, cut-based decompositions, and a lexicographic sequence of real-valued functionals $\mu_A(x)$5. This extends Wang and Wang’s total ordering for fuzzy numbers to the intuitionistic setting (Nayagam et al., 2016).

The IFV space also connects directly to q-rung orthopair fuzzy values. The mapping

$\mu_A(x)$6

is an order-isomorphism between suitable q-rung orthopair orders and $\mu_A(x)$7, allowing lattice, negation, and topological results to transfer across the two frameworks (Wu et al., 2021).

3. Algebraic realizations in semigroups and Banach algebras

One substantial algebraic development studies IFS over semigroups. If $\mu_A(x)$8 is a semigroup and $\mu_A(x)$9 is an intuitionistic fuzzy subset of $\nu_A(x)$0, then $\nu_A(x)$1 is an intuitionistic fuzzy subsemigroup when

$\nu_A(x)$2

Analogous definitions exist for intuitionistic fuzzy bi-ideals, $\nu_A(x)$3-ideals, left ideals, right ideals, ideals, and semiprime ideals (Sardar et al., 2011).

A central operator here is intuitionistic fuzzy magnified translation. For $\nu_A(x)$4 and

$\nu_A(x)$5

it is defined by

$\nu_A(x)$6

This simultaneously scales and shifts the membership and non-membership functions while preserving the IFS constraint. The construction generalizes earlier fuzzy translation and fuzzy multiplication, and it preserves the standard intuitionistic fuzzy substructures: the magnified translation of an intuitionistic fuzzy subsemigroup, bi-ideal, $\nu_A(x)$7-ideal, ideal, or semiprime ideal has the same property if and only if the original set does (Sardar et al., 2011).

The semigroup-theoretic payoff is a collection of characterization theorems. Intra-regular semigroups are characterized by the condition that the magnified translation of every intuitionistic fuzzy ideal is semiprime; left regular and right regular semigroups admit analogous formulations using left and right ideals; regular semigroups are characterized through the equality

$\nu_A(x)$8

for every intuitionistic fuzzy right ideal $\nu_A(x)$9 and left ideal $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$0; and in archimedean semigroups the magnified translation of an intuitionistic fuzzy semiprime ideal is constant (Sardar et al., 2011).

Functional analysis provides another algebraic realization. An intuitionistic fuzzy normed algebra $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$1 is an algebra equipped with membership and non-membership functions on $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$2 satisfying intuitionistic fuzzy norm axioms together with multiplicative compatibility conditions. A complete intuitionistic fuzzy normed algebra is called an intuitionistic fuzzy Banach algebra (Dinda et al., 2010).

In that setting, the set $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$3 of invertible elements is open, the complement $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$4 of non-invertible elements is closed, and the inversion map $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$5 is strongly intuitionistic fuzzy continuous. The paper also defines topological divisors of zero and proves that every topological divisor of zero belongs to the closed set of non-invertible elements (Dinda et al., 2010). These results are direct intuitionistic fuzzy analogues of classical Banach algebra theorems, but expressed through the two-sided information carried by $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$6 and $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$7.

4. Soft, parameterized, and relational extensions

A major extension of IFS combines them with soft sets. If $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$8 is a universe of objects, $$0 \le \mu_A(x) + \nu_A(x) \le 1 \qquad \forall x \in X.$$9 a parameter set, and $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$0, an intuitionistic fuzzy soft set may be written as

$\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$1

where each parameter $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$2 is assigned an intuitionistic fuzzy set on $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$3, and $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$4 is the intuitionistic fuzzy empty set for $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$5. This introduces a two-level uncertainty structure: parameters organize the model, and each parameter carries its own membership/non-membership profile over the object universe (Çağman et al., 2013).

A closely related construction is the intuitionistic fuzzy parametrized soft set. For an intuitionistic fuzzy set $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$6 on the parameter space $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$7, the associated intuitionistic FP-soft set over $\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$8 is

$\pi_A(x) = 1 - \mu_A(x) - \nu_A(x)$9

where $(\mu_A(x),\nu_A(x),\pi_A(x))$0 and $(\mu_A(x),\nu_A(x),\pi_A(x))$1 if $(\mu_A(x),\nu_A(x),\pi_A(x))$2 and $(\mu_A(x),\nu_A(x),\pi_A(x))$3. Here $(\mu_A(x),\nu_A(x),\pi_A(x))$4 and $(\mu_A(x),\nu_A(x),\pi_A(x))$5 represent the importance and unimportance of parameter $(\mu_A(x),\nu_A(x),\pi_A(x))$6, while $(\mu_A(x),\nu_A(x),\pi_A(x))$7 records which objects satisfy it (Deli et al., 2013).

These parameterized structures inherit the usual lattice-style operations. Inclusion is defined by

$(\mu_A(x),\nu_A(x),\pi_A(x))$8

complement swaps $(\mu_A(x),\nu_A(x),\pi_A(x))$9 and $1$0 and replaces $1$1 by $1$2, and union and intersection use the ordinary IFS max–min and min–max patterns on the parameter side together with set-theoretic union and intersection on the approximation side. Additional OR/AND sums and products extend algebraic sum and product operations from IFS to the parameterized soft setting (Deli et al., 2013).

Relations on intuitionistic fuzzy soft sets generalize binary relations between parameterized families of IFS. For IF soft sets $1$3 and $1$4, the cartesian product $1$5 is indexed by $1$6, and an intuitionistic fuzzy soft relation from $1$7 to $1$8 is an intuitionistic fuzzy soft subset of this product. The theory includes inverse relations, composition, and the properties of being symmetric, transitive, and reflexive, thereby extending ordinary relation theory into a parameterized intuitionistic fuzzy environment (Dinda et al., 2012).

These soft and relational variants are significant because they decouple uncertainty about objects from uncertainty about parameters. This makes them especially suitable when the criteria themselves are graded, disputed, or only partially relevant.

5. Similarity, distance, aggregation, and ranking

Similarity and distance are among the most active technical components of IFS theory. For intuitionistic fuzzy soft sets over a finite universe $1$9 and parameter set $\mu_A(x)\in\{0,1\}$0, several metric-type distances are standard: $\mu_A(x)\in\{0,1\}$1 corresponding to Hamming, normalized Hamming, Euclidean, and normalized Euclidean distances. They compare both membership and non-membership values across all parameter–object pairs, and they induce similarity measures by decreasing transformations such as

$\mu_A(x)\in\{0,1\}$2

A distinct similarity measure based on the net score $\mu_A(x)\in\{0,1\}$3 is also defined for intuitionistic fuzzy soft sets and used in a medical diagnosis procedure (Çağman et al., 2013).

More recent work refines the axiomatic side of distance design. The notions of strict intuitionistic fuzzy distance measure (SIFDisM) and strict intuitionistic fuzzy similarity measure (SIFSimM) were introduced to require strict order-consistency along Atanassov chains and to force maximal distance only for the opposite endpoints $\mu_A(x)\in\{0,1\}$4 and $\mu_A(x)\in\{0,1\}$5. Within that framework, an improved Jensen–Shannon-divergence-based distance $\mu_A(x)\in\{0,1\}$6 is proved to be a SIFDisM, its dual similarity is a SIFSimM, and the induced entropy is an intuitionistic fuzzy entropy (Wu et al., 2022).

Aggregation theory has likewise been generalized beyond the product t-norm. For any continuous t-norm $\mu_A(x)\in\{0,1\}$7 with dual t-conorm $\mu_A(x)\in\{0,1\}$8, four basic operations on intuitionistic fuzzy values are defined: $\mu_A(x)\in\{0,1\}$9

$X$00

$X$01

$X$02

On this basis, the IF weighted average (IFWA), IF weighted geometric (IFWG), and IF mean weighted average and geometric (IFMWAG) operators are defined. The same work proves that a continuous t-norm is power stable if and only if it is the minimum t-norm, or it is strict, or it is an ordinal sum of strict t-norms (Wu et al., 2022). In the corresponding decision-making application, IFMWAG is designed to remove what the paper calls the hindrance of indiscernibility on the boundaries (Wu et al., 2022).

Ranking by distance has also become a point of methodological criticism. A hypervolume-based study proves that, for nonlinear distance functions, the assumption that the shortest distance to the positive ideal solution must coincide with the greatest distance from the negative ideal solution is not valid. As an alternative, it proposes the HyperVolume based ASsessment method (HVAS), with

$X$03

and an extension

$X$04

when hesitancy is incorporated through a perception factor $X$05 (Deveci et al., 2022). This line of work treats ranking by dominance geometry rather than by nonlinear point-to-point distance.

6. Generalizations, applications, and ongoing debates

IFS occupy a central position in a broader hierarchy of uncertainty models. One line of work embeds them into single-valued neutrosophic sets by identifying

$X$06

so that the neutrosophic triplet satisfies $X$07. In that perspective, IFS are the special case of neutrosophic sets with dependent components and fixed total sum, whereas neutrosophic sets allow

$X$08

with independent components (Smarandache, 2019). The same paper argues that intuitionistic fuzzy aggregation operators treat indeterminacy as a residual quantity, while neutrosophic operators process it explicitly (Smarandache, 2019). A broader survey places IFS as a foundational model and a stepping stone toward neutrosophic and plithogenic theories, and it treats fuzzy sets as the zero-hesitation special case of IFS (Fujita et al., 12 Mar 2026).

The application range is correspondingly broad. The survey literature lists intuitionistic fuzzy relations, graphs, hypergraphs, algebras, topological spaces, metric spaces, probability spaces, Markov chains, dynamical systems, differential equations, decision methods, aggregation operators, entropy measures, and similarity measures (Fujita et al., 12 Mar 2026). Concrete implementations include medical diagnosis via intuitionistic fuzzy soft similarity (Çağman et al., 2013) and pattern recognition for building materials using admissible similarity measures compatible with the linear orders on IFVs (Wu et al., 2021).

A recent application brings IFS directly into LLM alignment. In side-by-side preference annotation, each response $X$09 is assigned a support degree, an opposition degree, and a hesitation degree, with

$X$10

The framework defines IFS-based inter-annotator agreement, dynamic annotator weights, and dataset quality scores, and it reports average agreement $X$11 versus $X$12 for binary annotation and $X$13 for Likert annotation, together with a $X$14 reduction in annotation time and a $X$15 improvement in win-rate against baseline models in downstream RLHF experiments (Du, 30 May 2025). This suggests that the original intuitionistic fuzzy distinction between support, opposition, and hesitation remains operationally useful even in contemporary machine-learning pipelines.

The enduring conceptual issue is therefore not whether IFS can represent uncertainty, but how their three-part semantics should be ranked, aggregated, and extended. Subsequent theories frequently present themselves as generalizations of IFS, but the persistent technical work on value-space orderings, strict similarity axioms, t-norm-based aggregation, soft relations, and algebraic characterization shows that IFS remain a mathematically active framework in their own right (Wu et al., 2021, Wu et al., 2022, Fujita et al., 12 Mar 2026).