Possibility versus probability: falling shadows versus falling integrals Probability – falling integrals In probability theory, the dependency between two random variables X, Y with probability density functions fX(x) and fY(y) respectively, can be characterized through their joint probability density function fX,Y(x,y), where ∫ R f X ,Y ( x, t )dt = f X ( x), ∫ R f X ,Y (t , y )dt = fY ( y ) for all x, y 0 R. fX(x) and fY(y) are called the marginal probability density functions of joint random variable (X,Y). X and Y are said to be independent if fX,Y(x,y) = fX(x) fY(y). Probability – expected value Let a, b 0 R c {–∞, ∞} with a # b, then the probability that X takes its value from [a, b] is computed by b P( X ∈ [a, b]) = ∫ f X ( x)dx. a The expected value of random variable X is defined by E( X ) = ∫ xf X ( x)dx. R If g is a function of X then the expected value of g(X) is E( g ( X )) = ∫ g ( x) f X ( x)dx. R Probability – measure of dependency If h is a function of X and Y then the expected value of h(X,Y) can be computed by E(h( X , Y )) = ∫ 2 h( x, y ) f X ,Y ( x, y )dxdy. R The covariance of two random variables X and Y is defined by Cov( X , Y ) = E(( X − E ( X ))(Y − E (Y ))) = ∫ 2 xyf X ,Y ( x, y )dxdy − ∫ xf X ( x)dx ∫ yfY ( y )dy. R R R Probability – measure of variability The variance of random variable X is defined as the covariance of X and itself, that is Var ( X ) = E(( X − E( X )) ) = ∫ x f X ( x)dx − 2 2 R (∫ xf R ) 2 X ( x)dx . For any random variables X and Y and real numbers λ and µ the following relationship holds Var (λX + µY ) = λ2 Var ( X ) + µ 2 Var (Y ) + 2λµCov( X , Y ). Probability – correlation coefficient Let X and Y be random variables with non-zero variances. Then, the correlation coefficient of X and Y is defined by Cov( X , Y ) ρ ( X ,Y ) = , Var ( X )Var (Y ) where –1 # ρ(X, Y) # 1. If X and Y are independent random variables then Cov( X , Y ) = ρ ( X , Y ) = 0. Possibility – concept of fuzzy number A fuzzy set A in R is a fuzzy number if it is normal, fuzzy convex and has an upper semi-continuous membership function of bounded support. The family of all fuzzy numbers are denoted by ö. A γ-level set of a fuzzy set A in Rm is defined by [ A]γ = cl{x ∈ R m : A( x) > γ }. If A 0 ö is a fuzzy number then [A]γ is a convex and compact subset of R for all γ 0 [0,1]. Possibility – concept of joint distribution Let a, b 0 R c {–∞, ∞} with a # b, then the possibility that A 0 ö takes its value from [a, b] is defined by Pos( A ∈ [a, b]) = max A( x). x∈[ a ,b ] A fuzzy set B in Rm is said to be a joint possibility distribution of fuzzy numbers Ai 0 ö, i = 1, ..., m, if max B( x1 , K, xm ) = Ai ( xi ) x j ∈R , j ≠ i for all xi 0 R, i = 1, ..., m. Ai is called the i-th marginal possibility distribution of B. Possibility – falling shadows The joint possibility distribution always uniquely defines its marginal distributions – the shadow of B on the i-th axis is exactly Ai –, but not vice versa. If Ai 0 ö, i = 1, ..., m, and B is their joint possibility distribution then B( x1 , K , xm ) ≤ min{ A1 ( x1 ),K , Am ( xm )}, [ B]γ ⊆ [ A1 ]γ × L × [ Am ]γ hold for all x1, ..., xm 0 R and γ 0 [0,1]. B 1 γ A G B 1 γ A H Possibility – concept of non-interactivity Definition 1. Fuzzy numbers Ai 0 ö, i = 1, ..., m, are said to be non-interactive if their joint possibility distribution B is given by B( x1 , K , xm ) = min{ A1 ( x1 ),K, Am ( xm )} or, equivalently, [ B]γ = [ A1 ]γ × L× [ Am ]γ for all x1, ..., xm 0 R and γ 0 [0,1]. B 1 γ A AHB Non-interactive possibility distributions Possibility – central value Definition 2. Let A 0 ö be a fuzzy number with [A]γ = [a1(γ), a2(γ)], then the central value of [A]γ is defined by γ C ([ A] ) = 1 ∫ [ A ]γ ∫ dx [ A ]γ a1 (γ ) + a2 (γ ) . xdx = 2 Definition 6. Let A1, ..., An 0 ö be fuzzy numbers, let B be their joint possibility distribution, and let g: Rn 6 R be a continuous function. The central value of g on [B]γ is 1 C[ B ]γ ( g ) = g ( x)dx. γ ∫ [ B] dx ∫ γ [ B] Possibility – central value The central value operator is linear for any fixed joint possibility distribution B and for any γ 0 [0,1]. Let the projection functions on R2 be denoted by πx and πy, that is, πx (u,v) = u and πy (u,v) = v for u,v 0 R. Let A, B 0 ö be non-interactive fuzzy numbers, then C[ A× B ]γ (π x + π y ) = C[ A×B ]γ (π x ) + C[ A× B ]γ (π y ) = C[ A]γ (id ) + C[ B ]γ (id ) = C ([ A]γ ) + C ([ B ]γ ), C[ A×B ]γ (π xπ y ) = C[ A]γ (id )C[ B ]γ (id ) = C ([ A]γ )C ([ B ]γ ). Possibility – mean value Definition 3. A function f : [0,1] 6 R is said to be a weighting function if f is non-negative, monotone increasing, and satisfies 1 ∫ 0 f (γ )dγ = 1. Definition 4. The expected (or mean) value of A with respect to a weighting function f is defined by a1 (γ ) + a2 (γ ) E f ( A) = ∫ C ([ A] ) f (γ )dγ = ∫ f (γ )dγ . 0 2 0 1 γ 1 Possibility – mean value Definition 6. Let A1, ..., An 0 ö be fuzzy numbers, let B be their joint possibility distribution, and let g: Rn 6 R be a continuous function. The expected (or mean) value of g on B with respect to a weighting function f is defined by 1 E f ( g ; B) = ∫ C[ B ]γ ( g ) f (γ )dγ 0 1 =∫ 0 1 ∫ [ B ]γ ∫ dx γ [B] g ( x)dx f (γ )dγ . Possibility – measure of interactivity Definition 8. Let A, B 0 ö be fuzzy numbers, and let C be their joint possibility distribution. The interactivity relation of [A]γ and [B]γ is defined by R[C ]γ (π x , π y ) = C[C ]γ ((π x − C[C ]γ (π x ))(π y − C[C ]γ (π y ))) = C[C ]γ (π xπ y ) − C[C ]γ (π x )C[C ]γ (π y ). The covariance of A and B with respect to a weighting function f is defined by 1 Cov f ( A, B) = ∫ R[C ]γ (π x , π y ) f (γ )dγ . 0 Possibility – measure of variability The variance of A 0 ö is defined by 1 Var f ( A) = Cov f ( A, A) = ∫ R[ A]γ (id, id ) f (γ )dγ . 0 Theorem 3. Let A, B 0 ö be fuzzy numbers, let C be their joint possibility distribution, and let λ, µ 0 R. Then, R[C ]γ (λπ x + µπ y , λπ x + µπ y ) = λ2 R[C ]γ (π x , π x ) + µ 2 R[C ]γ (π y , π y ) + 2λµR[C ]γ (π x , π y ), Var f (λA + µB) = λ2 Var f ( A) + µ 2 Var f ( B) + 2λµCov f ( A, B). Possibility – correlation coefficient Theorem 4. Let A, B 0 ö be fuzzy numbers with nonzero variances, and let C be their joint possibility distribution. Then, the correlation coefficient of A and B ρ f ( A, B) = Cov f ( A, B) Var f ( A)Var f ( B) satisfies the property –1 # ρ f (A, B) # 1 for any weighting function f. If A, B 0 ö are non-interactive fuzzy numbers then Cov f ( A, B ) = ρ f ( A, B ) = 0. B 1 γ A C The case of ρf(A, B) = 1 B 1 γ A D The case of ρf(A, B) = –1 B 1 A γ E The case of ρf(A, B) = 1/3 B 1 A γ F The case of ρf(A, B) = –1/3 B 1 γ A I The case of interactivity when ρf(A, B) = 0