Fuzzy Arithmetic and Membership Functions: Alpha-Cuts, Worked Examples and Exam Traps

Ordinary sets feel simple until an element belongs only partly. Triangular fuzzy-number shortcuts are easy to memorise without knowing when they fail. A membership grade lies in [0,1], but it is not a probability. Alpha-cuts turn those grades into nested intervals; addition, subtraction, multiplication and division then follow from interval rules, including sign and zero checks.

Fuzzy sets and membership functions: what a membership grade means

A fuzzy set A on universe X is defined by mu_A: X -> [0,1]. A crisp characteristic function permits only 0 and 1; fuzzy membership also permits intermediate grades. Fuzzy Sets in AI: Membership Functions and Worked Examples develops the broader foundation of shapes, standard operations, relations and max-min composition. Exact alpha-cut sums, differences, products, quotients and extension-principle memberships instead require interval arithmetic.

For X={1,2,3,4,5}, let A={0.2/1, 0.7/2, 1.0/3, 0.6/4, 0.1/5}. Here 0.7/2 means mu_A(2)=0.7. The slash is notation, not division, and this does not mean a 70 per cent chance.

• Support: points with mu_A(x)>0, so support(A)={1,2,3,4,5}.

• Core: points with mu_A(x)=1, so core(A)={3}.

• Height: sup mu_A(x)=1, making A normal.

• Crossover point: a point with grade 0.5. None of these five sampled points is one.

Membership-function shapes and alpha-cuts: compute a triangle exactly

A triangular fuzzy number T=(a,m,b) has membership 0 outside [a,b], (x-a)/(m-a) for a <= x <= m, and (b-x)/(b-m) for m <= x <= b. A trapezoid (a,b,c,d) instead has core [b,c]; a singleton has one non-zero point.

For Warm=(20,25,30):

• mu_Warm(20)=0

• mu_Warm(23)=(23-20)/(25-20)=3/5=0.6

• mu_Warm(25)=1

• mu_Warm(28)=(30-28)/(30-25)=2/5=0.4

• mu_Warm(30)=0

Thus support(Warm)=(20,30), core(Warm)={25}, and height(Warm)=1.

For 0 < alpha <= 1, the ordinary alpha-cut is T_alpha={x: mu_T(x) >= alpha}. Solving each slope gives [a+alpha(m-a), b-alpha(b-m)]. Therefore Warm_0.6=[20+0.6(5), 30-0.6(5)]=[23,27]. Under the standard fuzzy-number convention, T_0=closure(support(T))=[a,b]; the literal condition mu >= 0 over the whole universe would include every point. A strong alpha-cut uses T_alpha^+={x: mu_T(x) > alpha}. For this triangle, its endpoints are excluded, so the strong 0.6-cut is (23,27).

Fuzzy union, intersection and complement: do not confuse them with arithmetic

Keep A=[0.2,0.7,1.0,0.6,0.1] and take B=[0.1,0.5,0.8,1.0,0.4]. With the standard Zadeh operators, union is pointwise max, intersection is pointwise min, and complement is 1-mu.

The output vectors are [0.2,0.7,1.0,1.0,0.4], [0.1,0.5,0.8,0.6,0.1], and [0.8,0.3,0,0.4,0.9]. At x=4, these operations give max(0.6,1.0)=1.0, min(0.6,1.0)=0.6, and 1-0.6=0.4. Min and max extend the 0/1 AND/OR boundary discussed in Propositional and Predicate Logic: Truth Tables, Quantifiers. Adding the grades gives 1.6, which is neither standard union nor a valid membership grade.

Fuzzy arithmetic with alpha-cuts: one complete worked example

Let positive triangular fuzzy numbers be P=(1,2,3) and Q=(2,4,6). Their cuts are P_alpha=[1+alpha,3-alpha] and Q_alpha=[2+2alpha,6-2alpha].

For intervals P=[pL,pU] and Q=[qL,qU], sum is [pL+qL,pU+qU]; difference is [pL-qU,pU-qL]. Product and quotient use the minimum and maximum of all four endpoint results, with division allowed only when Q excludes zero. Here both intervals are positive, so product becomes [pL*qL,pU*qU] and quotient [pL/qU,pU/qL].

The alpha=0 row uses the closed-support convention. Also, 5/6=0.8333..., not 0.8.

The full sum cut is [3+3alpha,9-3alpha], hence P+Q=(3,6,9). The difference is [-5+3alpha,1-3alpha], hence P-Q=(-5,-2,1). Its lower endpoint must be lower P minus upper Q, not lower minus lower.

Multiplication and division remain the exact cut families [(1+alpha)(2+2alpha),(3-alpha)(6-2alpha)] and [(1+alpha)/(6-2alpha),(3-alpha)/(2+2alpha)]. Their boundaries are generally curved, so triangular inputs do not produce exact triangular products or quotients. Division is undefined through any alpha-cut of Q containing 0.

Extension principle: why alpha-cut arithmetic works

For addition, the extension principle is mu_(P+Q)(z)=sup_(x+y=z) min(mu_P(x),mu_Q(y)). It examines every input pair producing z. Alpha-cuts turn that search into interval arithmetic when the operation and domain conditions are valid.

At z=6, (x,y)=(2,4) gives min(1,1)=1. At z=4.5, (1.5,3) gives min(0.5,0.5)=0.5; correspondingly, the alpha=0.5 sum cut runs from 1.5+3=4.5 to 2.5+5=7.5.

This agrees with P+Q=(3,6,9): mu_(P+Q)(5)=(5-3)/(6-3)=2/3, mu_(P+Q)(6)=1, and mu_(P+Q)(7.5)=(9-7.5)/(9-6)=0.5.

Fuzzy arithmetic mistakes and a numerical self-check

For [-2,3]*[-4,5], the candidates are 8,-10,-12,15, so the interval is [-12,15], not [8,15]. For positive cuts, P_0.5*Q_0.5=[4.5,12.5].

Use this four-step self-check: write the membership or cut definition, state the interval rule, calculate every candidate endpoint, then confirm that each higher-alpha result cut nests inside its lower-alpha cut.

How questions test membership functions and fuzzy arithmetic

Rapid checks: mu_Warm(28)=0.4; Warm_0.6=[23,27]; at x=4, mu_(A intersection B)=0.6; the modal value of P+Q is 6; P_0.5-Q_0.5=[-3.5,-0.5]; and [-2,3]*[-4,5]=[-12,15].

For mixed practice beyond this worked example, use the GATE CS Exam Preparation category.

Fuzzy arithmetic and membership functions: the short version and next step

Recall the chain: interpret mu, select the function branch, form the correct alpha-cut, apply the valid interval rule, inspect signs and zero, then verify nested cuts. The anchors are Warm_0.6=[23,27], P+Q=(3,6,9), and P-Q=(-5,-2,1).

Try this transfer exercise: for R=(0,2,4) and S=(1,3,5), find their alpha=0.5 cuts, sum, and difference.

Answer: R_0.5=[1,3], S_0.5=[2,4], R_0.5+S_0.5=[3,7], and R_0.5-S_0.5=[-3,1].

For a structured preparation route, use GATE Guidance by Sanchit Sir. For broader applied-AI preparation, AI & ML for Placements develops placement-oriented AI and machine-learning skills.