Uncertain Logic

Xiang Li
Uncertainty Theory Laboratory Department of Mathematical Sciences Tsinghua University Beijing 100084, China xiang-li04@mails.thu.edu.cn http://orsc.edu.cn/xli

July 29, 2008

Copyright c 2008 by UTLAB

Reference to this book should be made as follows: Li X, Uncertain Logic, http://orsc.edu.cn/process/ulogic.pdf

UTLAB ONLINE BOOKS

Edited by Baoding Liu Uncertainty Theory Laboratory Tsinghua University, Beijing Email: liu@tsinghua.edu.cn Vol.1: Baoding Liu, Uncertainty Theory, 3rd ed., http://orsc.edu.cn/liu/ut.pdf. Vol.2: Baoding Liu, Theory and Practice of Uncertain Programming, 2nd ed., http://orsc.edu.cn/liu/up.pdf. Vol.3: Zhongfeng Qin and Xiang Li, Fuzzy Calculus for Finance, http://orsc.edu.cn/process/fc.pdf. Vol.4: Xiang Li, Uncertain Logic, http://orsc.edu.cn/xli/ulogic.pdf. For up-to-date information, please visit http://orsc.edu.cn/utlab!

Contents
Preface 1 Probabilistic Logic 1.1 Random Proposition 1.2 Fuzzy Formula . . . 1.3 Truth Function . . . 1.4 Truth Value . . . . 1.5 Independence . . . . 1.6 Modus Ponens . . . 2 Credibilistic Logic 2.1 Fuzzy Proposition 2.2 Fuzzy Formula . . 2.3 Truth Function . . 2.4 Truth Value . . . 2.5 Independence . . . 2.6 Modus Ponens . . 2.7 Inference Function 2.8 Inference Rule . . 2.9 Entailment . . . . vii 1 1 1 2 2 5 7 1 1 1 2 2 5 7 7 8 13 15 15 16 16 16 21 21 22 25

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

3 Hybrid Logic 3.1 Hybrid Proposition 3.2 Hybrid Formula . . 3.3 Truth Function . . . 3.4 Truth Value . . . . 3.5 Modus Ponens . . . 3.6 Inference Function . 3.7 Inference Rule . . . 3.8 Entailment . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

vi 4 Uncertain Logic 4.1 Uncertain Proposition 4.2 Uncertain Formula . . 4.3 Truth Function . . . . 4.4 Truth Value . . . . . 4.5 Modus Ponens . . . . 4.6 Inference Function . . 4.7 Inference Rule . . . . 4.8 Entailment . . . . . .

Contents

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

27 27 27 27 27 29 29 29 30 31 32

List of Frequently Used Symbols Index

Preface
Stochastic calculus was well established as one of the most useful tools of probability theory and was applied widely. In order to deal with fuzzy world, we also need such a tool called fuzzy calculus. The main purpose is just to introduce the theory of fuzzy calculus with application to nance. Baoding Liu proposed a fuzzy process, a dierential formula and a fuzzy integral in a seminar during summer holidays of 2007. Later, the community renamed those three footstones Liu process, Liu formula and Liu integral due to their importance and usefulness. Fuzzy calculus originated with Baoding Lius pioneering paper Fuzzy Process, Hybrid Process and Uncertain Process in 2008. After that, fuzzy calculus was developed steadily. Fuzzy dierential equation is a type of dierential equations driven by Liu process. The study of fuzzy dierential equation was started by Baoding Lius pioneering paper Fuzzy Process, Hybrid Process and Uncertain Process in 2008. Numerous research results were then reported in the literature. Fuzzy calculus and fuzzy dierential equations were introduced to nance by Baoding Liu when he addressed the 2007 First Intelligent Computing Conference held on Lushan Mountain and presented an audacious and surprising assumption that stock price follows geometric Liu process, thus producing a fuzzy stock model known as Lius stock model. The rst option pricing formula for this model was given by Qin and Lis paper Option Pricing Formula for Fuzzy Financial Market in 2008. With the development and extension of fuzzy stock models, a new scientic discipline, fuzzy nance, is rapidly growing. Xiang Li Tsinghua University July 29, 2008

Chapter 1

Probabilistic Logic
probabilistic logic was designed by Nilsson as a generalization of classical logic for dealing with random knowledge. Random proposition, Random formula, truth value and probabilistic modus ponens are key concepts in Probabilistic logic.

1.1

Random Proposition

A random proposition is a statement with probability value belonging to [0, 1]. For example, It rains in Beijing with probability 0.8 is a random proposition, where It rains in Beijing is a statement, and its truth value is 0.8 in probability. Generally speaking, we use to express the random proposition and use p to express its probability value. In fact, the random proposition is essentially a random variable = 1, with probability p 0, with probability 1 p

where = 1 means random proposition is true and = 0 means random proposition is false. Please note that we use to express both a random proposition and a random variable.

1.2

Random Formula

In addition to proposition symbols and , negation symbol and disjunction symbol are also the primitive symbols in probabilistic logic. Based on

Chapter 1 - Probabilistic Logic

these symbols, random formula is dened as a member of the minimal set S of nite sequence of primitive symbols satisfying: (a) S for each random proposition ; (b) if X S, then X S; (c) if X1 S and X2 S, then X1 X2 S. In fact, a random formula X is essentially a random variable taking values 0 or 1. First, it is clear that each random proposition is a random variable. Furthermore, if a random formula X is a random variable, then X is also a random variable dened as X = 1 X. Finally, since random formulas X1 and X2 are randm variables, X1 X2 is a random variable dened as X1 X2 = max{X1 , X2 } where X = 1 means random formula X is true and X = 0 means random formula X is false. For any random formulas X1 and X2 , we dene conjunction symbol and implication symbol as X1 X2 = (X1 X2 ), X1 X2 = X1 X2 .

It is clear that X1 X2 and X1 X2 are random formulas.

1.3

Truth Function

Assume X is a random formula containing random propositions 1 , , n . It is well-known that there is a function f : {0, 1}n {0, 1} such that X = 1 if and only if f (1 , , n ) = 1. In probabilistic logic, we will call f the truth function of X. For example, the truth function of random formula ( ) is f (1, 1, 1) = 0, f (1, 1, 0) = 1, f (1, 0, 1) = 0, f (1, 0, 0) = 1, f (0, 1, 1) = 0, f (0, 1, 0) = 1, f (0, 0, 1) = 1, f (0, 0, 0) = 1.

1.4

Truth Value

Denition 1.4.1 Let X be a random formula. Then its truth value is dened as T (X) = Pr{X = 1}. (1.1) For any random proposition , it is easy to prove that Pr{ = 1} = c. That is, the truth value of each random proposition is just its probability value. If T (X) > 0.5, then random formula X is somewhat true, and T (X) = 1 means

Section 1.4 - Truth Value

X is certainly true. If T (X) < 0.5, then X is somewhat false, and T (X) = 0 means X is certainly false. Especially, T (X) = 0.5 means X is totally uncertain. The higher the truth value is, the more true the random formula is. For any random formula X containing random propositions 1 , , n with truth function f , we have T (X) = Pr{f (1 , , n ) = 1}. Theorem 1.4.1 Let X be a random formula containing random propositions 1 , , n whose truth function is f (x1 , , xn ). Then T (X) =
f (x1 , ,xn )=1

m(x1 , , xn )

where m is the joint mass function of random variables 1 , , n . Theorem 1.4.2 (Self-Duality) For any random formula X, we have T (X) = 1 T (X). Proof: It follows from the self-duality of probability measure that T (X) = Pr{X = 1} = Pr{X = 0} = 1 Pr{X = 1} = 1 T (X).

The proof is complete. Theorem 1.4.3 (Monotonicity) For any random formulas X1 and X2 , we have T (X1 X2 ) T (X1 ) T (X2 ). Proof: It follows from the monotonicity of probability measure that T (X1 X2 ) = Pr{X1 X2 = 1} = Cr{{X1 = 1} {X2 = 1}} Pr{X1 = 1} Pr{X2 = 1} = T (X1 ) T (X2 ).

The proof is complete. Theorem 1.4.4 (Subadditivity) For any random formulas X1 and X2 , we have T (X1 X2 ) T (X1 ) + T (X2 ).

Chapter 1 - Probabilistic Logic

Proof: It follows from the subadditivity of probability measure that T (X1 X2 ) = Pr{{X1 = 1} {X2 = 1}} Pr{X1 = 1} + Pr{X2 = 1} = T (X1 ) + T (X2 ).

The proof is complete. Remark 1.1: The probabilistic logic and classical logic are consistent. (a) For any random proposition with probability value p. If is true, we have T () = p = 1, and if is false, we have T () = p = 0. (b) For any random formula X, if T (X) = 1, it follows from the self-duality theorem that T (X) = 1 T (X) = 0, and if T (X) = 0, we have T (X) = 1. (c) For any random formulas X1 and X2 , if T (X1 ) = 1 or T (X2 ) = 1, it follows from the monotonicity theorem that T (X1 X2 ) T (X1 ) T (X2 ) = 1, which implies that T (X1 X2 ) = 1. If T (X1 ) = 0 and T (X2 ) = 0, it follows from subadditivity theorem that T (X1 X2 ) T (X1 ) + T (X2 ) = 0. That is, T (X1 X2 ) = 0. Theorem 1.4.5 (Normality) For any random formula X, we have T (X X) = 1, T (X X) = T (X).

Proof: It follows from the normality of probability measure that T (X X) = Pr{X X = 1} = Pr{X = 1 or 0} = 1. Furthermore, it follows from Denition ?? that T (X X) = Cr{X X = 1} = Cr{X = 1} = T (X). The proof is complete. Theorem 1.4.6 For any random formulas X1 and X2 , we have T (X1 X2 ) T (X1 ) T (X2 ). Proof: It follows from the monotonicity theorem that T (X1 X2 ) = 1 T (X1 X2 ) 1 T (X1 ) T (X2 ) = T (X1 ) T (X2 ).

The proof is complete.

Section 1.5 - Independence

Theorem 1.4.7 For any random formulas X1 and X2 , we have T (X1 X2 ) T (X1 ) + T (X2 ) 1. Proof: It follows from the subadditivity theorem that T (X1 X2 ) = 1 T (X1 X2 ) 1 (T (X1 ) + T (X2 )) = T (X1 ) + T (X2 ) 1.

The proof is complete. Theorem 1.4.8 For any fuzzy formula X, we have T (X X) = 0, T (X X) = T (X).

Proof: It follows from the normality theorem that T (X X) = 1 T (X X) = 1 1 = 0, T (X X) = 1 T (X X) = 1 T (X) = T (X). The proof is complete. Theorem 1.4.9 For any fuzzy formulas X1 and X2 , we have T (X1 X2 ) 1 T (X1 ) + T (X2 ). Proof: It follows immediately from subadditivity theorem and the fact X1 X2 = X1 X2 . Theorem 1.4.10 For any fuzzy formulas X1 and X2 , we have (1 T (X1 )) T (X2 ) T (X1 X2 ) 1 T (X1 ) + T (X2 ). Proof: It follows immediately from the monotonicity theorem and the fact X1 X2 = X1 X2 . Theorem 1.4.11 For any fuzzy formula X, we have T (X X) = 1, T (X X) = 1 T (X).

Proof: It follows from Theorem 3.4.4 that T (X X) = T (X X) = 1, T (X X) = T (X X) = T (X) = 1 T (X). The proof is complete. Theorem 1.4.12 For any fuzzy formulas X1 and X2 , we have T (X2 X1 ) = T (X1 X2 ). Proof: It follows from denition of implication symbol that T (X2 X1 ) = T (X2 X1 ) = T (X1 X2 ) = T (X1 X2 ). The proof is complete.

Chapter 1 - Probabilistic Logic

1.5

Independence

The independence of fuzzy variables were dened by Liu and Gao (2007). That is, fuzzy variables 1 and 2 are independent if Cr{{1 B1 } {2 B2 }} = Cr{1 B1 } Cr{2 B2 } for any sets B1 and B2 of .

Denition 1.5.1 Fuzzy formulas X1 and X2 are called independent if they are independent fuzzy variables. Let X1 = p q, X2 = r and X3 = p r s. Then X1 and X2 are independent. However, X1 and X3 are not independent because if X1 = 0, it is clear that X3 = 1. That is, they are dependent. Theorem 1.5.1 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = T (X1 ) T (X2 ). Proof: Since fuzzy variables X1 and X2 are independent, we have T (X1 X2 ) = Cr{X1 X2 = 1} = Cr{{X1 = 1} {X2 = 1}} = Cr{X1 = 1} Cr{X2 = 1} = T (X1 ) T (X2 ). (1.2)

The proof is complete. Example 1.1: have If fuzzy formulas X1 and X2 are not independent, we may T (X1 X2 ) > T (X1 ) T (X2 ). For example, let p be a fuzzy proposition with credibility value 0.5. It follows from Denition 3.4.1 that T (p) = 0.5, T (p) = 0.5 and T (p p) = Cr{p p = 1} = Cr{p = 1 or 0} = 1. Therefore, we get T (p p) > T (p) T (p). Theorem 1.5.2 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = T (X1 ) T (X2 ). Proof: Since X1 and X2 are independent, we have X1 and X2 are also independent. It follows from Theorem 2.5.1 that T (X1 X2 ) = T ((X1 X2 )) = 1 T (X1 X2 ) = 1 T (X1 ) T (X2 ) = T (X1 ) T (X2 ).

The proof is complete.

Section 1.6 - Modus Ponens

Theorem 1.5.3 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = (1 T (X1 )) T (X2 ). Proof: Since X1 and X2 are independent, we have X1 and X2 are also independent. It follows from Theorem 2.5.1 that T (X1 X2 ) = T (X1 X2 ) = T (X1 ) T (X2 ) = (1 T (X1 )) T (X2 ).

The proof is complete.

1.6

Modus Ponens

As a rule of inference, the modus ponens plays an important role in classical logic and allows us to infer Y from X and X Y . When X and X Y are assumed fuzzy formulas, how do we infer the truth value T (Y ) from T (X) and T (X Y )? We will answer this question by credibilistic modus ponens. Credibilistic Modus Ponens: From T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ). (1.3)

If T (X) = 1 and T (X Y ) = 1, it follows from (2.3) that T (Y ) = 1. In other words, if X and X Y are true, then Y is also true. Hence, the credibilistic modus ponens is consistent with the classical one. For example, let X Y be a fuzzy formula If someone lives in Beijing, then he works in Beijing with credibility 0.9, and X a fuzzy formula Tom lives in Beijing with credibility 0.8. Then it follows from (2.3) that Tom works in Beijing with credibility belonging to [0.7, 0.9]. Remark 1.2: Assume T (X) = 1 and T (X Y ) = 1. It follows from the credibilistic modus ponens (2.3) that T (Y ) = 1. Hence, credibilistic modus ponens (2.3) is also consistent with the classical one. How do we infer the truth value T (X) from T (Y ) and T (X Y )? We will answer this question by credibilistic modus tollens. Credibilistic Modus Tollens: From T (Y ) and T (X Y ) we infer T (Y ) + T (X Y ) 1 T (X) T (X Y ). (1.4)

If T (Y ) = 1 and T (X Y ) = 1, it follows from (2.4) that T (X) = 1. In other words, if Y is false and X Y is true, then X is false. Hence, the credibilistic modus tollens is consistent with the classical one. For example, if we know If someone lives in Beijing, then he works in Beijing with credibility 0.9, and Tom does not work in Beijing with credibility 0.8. Then it follows from (2.4) that Tom does not live in Beijing with credibility belonging to [0.7, 0.9].

Chapter 2

Credibilistic Logic
Credibilistic logic was designed by Li and Liu as a new branch of multi-valued logic for dealing with fuzzy knowledge. Fuzzy proposition, fuzzy formula, truth value and credibilistic modus ponens are key concepts in credibilistic logic. One advantage of credibilistic logic is the well consistency with classical logic.

2.1

Fuzzy Proposition

A fuzzy proposition is a statement with credibility value belonging to [0, 1]. For example, Tom lives in Beijing with credibility 0.8 is a fuzzy proposition, where Tom lives in Beijing is a statement, and its truth value is 0.8 in credibility. In addition, Ellis is a man with credibility 0.1 is a fuzzy proposition, where Ellis is a man is a statement, and its truth value is 0.1 in credibility. Generally speaking, we use p to express the fuzzy proposition and use c to express its credibility value. In fact, the fuzzy proposition p is essentially a fuzzy variable 1, with credibility c p= 0, with credibility 1 c where p = 1 means p is true and p = 0 means p is false.

2.2

Fuzzy Formula

In addition to proposition symbols p and q, negation symbol and disjunction symbol are also the primitive symbols in credibilistic logic. Based on these

Chapter 2 - Credibilistic Logic

symbols, fuzzy formula is dened as a member of the minimal set S of nite sequence of primitive symbols satisfying: (a) p S for each fuzzy proposition p; (b) if X S, then X S; (c) if X1 S and X2 S, then X1 X2 S. In fact, a fuzzy formula X is essentially a fuzzy variable taking values 0 or 1. First, it is clear that each fuzzy proposition is a fuzzy variable. Furthermore, if a fuzzy formula X is a fuzzy variable, then X is also a fuzzy variable dened as X = 1 X. Finally, since fuzzy formulas X1 and X2 are fuzzy variables, X1 X2 is a fuzzy variable dened as X1 X2 = max{X1 , X2 } where X = 1 means X is true and X = 0 means X is false. For any fuzzy formulas X1 and X2 , we dene conjunction symbol and implication symbol as X1 X2 = (X1 X2 ), X1 X2 = X1 X2 .

It is clear that X1 X2 and X1 X2 are fuzzy formulas.

2.3

Truth Function

Assume X is a fuzzy formula containing fuzzy propositions p1 , , pn . It is well-known that there is a function f : {0, 1}n {0, 1} such that X = 1 if and only if f (p1 , , pn ) = 1. In credibilistic logic, we will call f the truth function of X. For example, the truth function of fuzzy formula (p q) r is f (1, 1, 1) = 0, f (1, 0, 1) = 0, f (0, 1, 1) = 0, f (0, 0, 1) = 1, f (1, 1, 0) = 1, f (1, 0, 0) = 1, f (0, 1, 0) = 1, f (0, 0, 0) = 1.

2.4

Truth Value

Denition 2.4.1 Let X be a fuzzy formula. Then its truth value is dened as T (X) = Cr{X = 1}. (2.1) For any fuzzy proposition p, it is easy to prove that Cr{p = 1} = c. That is, the truth value of fuzzy proposition is just its credibility value. If T (X) > 0.5, then fuzzy formula X is somewhat true, and T (X) = 1 means X is certainly true. If T (X) < 0.5, then X is somewhat false, and T (X) = 0 means X is certainly false. Especially, T (X) = 0.5 means X is

Section 2.4 - Truth Value

totally uncertain. The higher the truth value is, the more true the fuzzy formula is. For any fuzzy formula X containing fuzzy propositions p1 , , pn with truth function f , we have T (X) = Cr{f (p1 , , pn ) = 1}. Theorem 2.4.1 Let X be a fuzzy formula containing fuzzy propositions p1 , , pn whose truth function is f (x1 , , xn ). Then T (X) = 1 2 max
f (x1 , ,xn )=1

(x1 , , xn ) + 1

max
f (x1 , ,xn )=0

(x1 , , xn )

where is the joint membership function of fuzzy variables p1 , , pn . Proof: The theorem follows immediately from the credibility inversion theorem (??). Theorem 2.4.2 (Self-Duality) For any fuzzy formula X, we have T (X) = 1 T (X). Proof: It follows from the self-duality of credibility measure (??) that T (X) = Cr{X = 1} = Cr{X = 0} = 1 Cr{X = 1} = 1 T (X).

The proof is complete. Theorem 2.4.3 (Monotonicity and Subadditivity) For any fuzzy formulas X1 and X2 , we have T (X1 ) T (X2 ) T (X1 X2 ) T (X1 ) + T (X2 ) Proof: It follows from the monotonicity of credibility measure (??) that T (X1 X2 ) = Cr{X1 X2 = 1} = Cr{{X1 = 1} {X2 = 1}} Cr{X1 = 1} Cr{X2 = 1} = T (X1 ) T (X2 ).

Furthermore, it follows from the subadditivity of credibility measure (??) that T (X1 X2 ) = Cr{{X1 = 1} {X2 = 1}} Cr{X1 = 1} + Cr{X2 = 1} = T (X1 ) + T (X2 ).

4 The proof is complete.

Chapter 2 - Credibilistic Logic

Remark 2.1: The credibilistic logic and classical logic are consistent. (a) For any fuzzy proposition p with credibility value c. If p is true, we have T (p) = c = 1, and if p is false, we have T (p) = c = 0. (b) For any fuzzy formula X, if T (X) = 1, it follows from Theorem 4.4.1 that T (X) = 1 T (X) = 0, and if T (X) = 0, we have T (X) = 1. (c) For any fuzzy formulas X1 and X2 , if T (X1 ) = 1 or T (X2 ) = 1, it follows from Theorem 4.4.2 that T (X1 X2 ) T (X1 ) T (X2 ) = 1, which implies that T (X1 X2 ) = 1. If T (X1 ) = 0 and T (X2 ) = 0, it follows from Theorem 4.4.2 that T (X1 X2 ) T (X1 ) + T (X2 ) = 0. That is, T (X1 X2 ) = 0. Theorem 2.4.4 For any fuzzy formula X, we have T (X X) = 1, T (X X) = T (X).

Proof: It follows from the normality of credibility measure that T (X X) = Cr{X X = 1} = Cr{X = 1 or 0} = 1. Furthermore, it follows from Denition 3.4.1 that T (X X) = Cr{X X = 1} = Cr{X = 1} = T (X). The proof is complete. Theorem 2.4.5 For any fuzzy formulas X1 and X2 , we have T (X1 ) + T (X2 ) 1 T (X1 X2 ) T (X1 ) T (X2 ). Proof: It follows from Theorem 4.4.2 that T (X1 X2 ) = 1 T (X1 X2 ) 1 T (X1 ) T (X2 ) = T (X1 ) T (X2 ), = 1 T (X1 X2 ) 1 (T (X1 ) + T (X2 )) = T (X1 ) + T (X2 ) 1.

T (X1 X2 )

The proof is complete.

Section 2.5 - Independence

Theorem 2.4.6 For any fuzzy formula X, we have T (X X) = 0, T (X X) = T (X).

Proof: It follows from Theorem 3.4.4 that T (X X) = 1 T (X X) = 1 1 = 0, T (X X) = 1 T (X X) = 1 T (X) = T (X). The proof is complete. Theorem 2.4.7 For any fuzzy formulas X1 and X2 , we have (1 T (X1 )) T (X2 ) T (X1 X2 ) 1 T (X1 ) + T (X2 ). Proof: It follows immediately from Theorem 4.4.2 and the fact X1 X2 = X1 X2 . Theorem 2.4.8 For any fuzzy formula X, we have T (X X) = 1, T (X X) = 1 T (X).

Proof: It follows from Theorem 3.4.4 that T (X X) = T (X X) = 1, T (X X) = T (X X) = T (X) = 1 T (X). The proof is complete. Theorem 2.4.9 For any fuzzy formulas X1 and X2 , we have T (X2 X1 ) = T (X1 X2 ). Proof: It follows from denition of implication symbol that T (X2 X1 ) = T (X2 X1 ) = T (X1 X2 ) = T (X1 X2 ). The proof is complete.

2.5

Independence

The independence of fuzzy variables were dened by Liu and Gao (2007). That is, fuzzy variables 1 and 2 are independent if Cr{{1 B1 } {2 B2 }} = Cr{1 B1 } Cr{2 B2 } for any sets B1 and B2 of .

Chapter 2 - Credibilistic Logic

Denition 2.5.1 Fuzzy formulas X1 and X2 are called independent if they are independent fuzzy variables. Let X1 = p q, X2 = r and X3 = p r s. Then X1 and X2 are independent. However, X1 and X3 are not independent because if X1 = 0, it is clear that X3 = 1. That is, they are dependent. Theorem 2.5.1 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = T (X1 ) T (X2 ). Proof: Since fuzzy variables X1 and X2 are independent, we have T (X1 X2 ) = Cr{X1 X2 = 1} = Cr{{X1 = 1} {X2 = 1}} = Cr{X1 = 1} Cr{X2 = 1} = T (X1 ) T (X2 ). (2.2)

The proof is complete. Example 2.1: have If fuzzy formulas X1 and X2 are not independent, we may T (X1 X2 ) > T (X1 ) T (X2 ). For example, let p be a fuzzy proposition with credibility value 0.5. It follows from Denition 3.4.1 that T (p) = 0.5, T (p) = 0.5 and T (p p) = Cr{p p = 1} = Cr{p = 1 or 0} = 1. Therefore, we get T (p p) > T (p) T (p). Theorem 2.5.2 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = T (X1 ) T (X2 ). Proof: Since X1 and X2 are independent, we have X1 and X2 are also independent. It follows from Theorem 2.5.1 that T (X1 X2 ) = T ((X1 X2 )) = 1 T (X1 X2 ) = 1 T (X1 ) T (X2 ) = T (X1 ) T (X2 ).

The proof is complete. Theorem 2.5.3 If fuzzy formulas X1 and X2 are independent, then T (X1 X2 ) = (1 T (X1 )) T (X2 ). Proof: Since X1 and X2 are independent, we have X1 and X2 are also independent. It follows from Theorem 2.5.1 that T (X1 X2 ) The proof is complete. = T (X1 X2 ) = T (X1 ) T (X2 ) = (1 T (X1 )) T (X2 ).

Section 2.7 - Inference Function

2.6

Modus Ponens

As a rule of inference, the modus ponens plays an important role in classical logic and allows us to infer Y from X and X Y . When X and X Y are assumed fuzzy formulas, how do we infer the truth value T (Y ) from T (X) and T (X Y )? We will answer this question by credibilistic modus ponens. Credibilistic Modus Ponens: From T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ). (2.3)

If T (X) = 1 and T (X Y ) = 1, it follows from (2.3) that T (Y ) = 1. In other words, if X and X Y are true, then Y is also true. Hence, the credibilistic modus ponens is consistent with the classical one. For example, let X Y be a fuzzy formula If someone lives in Beijing, then he works in Beijing with credibility 0.9, and X a fuzzy formula Tom lives in Beijing with credibility 0.8. Then it follows from (2.3) that Tom works in Beijing with credibility belonging to [0.7, 0.9]. Remark 2.2: Assume T (X) = 1 and T (X Y ) = 1. It follows from the credibilistic modus ponens (2.3) that T (Y ) = 1. Hence, credibilistic modus ponens (2.3) is also consistent with the classical one. How do we infer the truth value T (X) from T (Y ) and T (X Y )? We will answer this question by credibilistic modus tollens. Credibilistic Modus Tollens: From T (Y ) and T (X Y ) we infer T (Y ) + T (X Y ) 1 T (X) T (X Y ). (2.4)

If T (Y ) = 1 and T (X Y ) = 1, it follows from (2.4) that T (X) = 1. In other words, if Y is false and X Y is true, then X is false. Hence, the credibilistic modus tollens is consistent with the classical one. For example, if we know If someone lives in Beijing, then he works in Beijing with credibility 0.9, and Tom does not work in Beijing with credibility 0.8. Then it follows from (2.4) that Tom does not live in Beijing with credibility belonging to [0.7, 0.9].

2.7

Inference Function

Let X, Y and X Y be fuzzy formulas. It follows from Table 1 that if X is false and X Y is true, then Y must be true; if X is true and X Y is true, then Y may be true or false; if X and X Y are all false, then Y must be false; nally, X is true and X Y is false is illogical because X is true implies that X Y is true. Suppose that X1 , , Xn and Y are fuzzy formulas. An assignment X1 = x1 , , Xn = xn may be illogical. Otherwise, there are three possible outcomes about Y . That is, Y = 1 (true), Y = 0 (false) and Y = 1 or 0 (undetermined).

8 X 1 1 0 0 Y 1 0 1 0

Chapter 2 - Credibilistic Logic

X Y 1 1 1 0

Table 2.1: The truth table of disjunction Denition 2.7.1 Let X1 , , Xn and Y be fuzzy formulas. Then the inference function g : {0, 1}n {0, 0.5, 1} of Y on X1 , , Xn is dened as follows: (i) g(x1 , , xn ) = 1 if Y = 1 with X1 = x1 , , Xn = xn ; (ii) g(x1 , , xn ) = 0.5 if Y = 1 or 0 with X1 = x1 , , Xn = xn ; (iii) g(x1 , , xn ) = 0 if X1 = x1 , , Xn = xn are illogical or Y = 0 with X1 = x1 , , Xn = xn . For example, the inference function of Y on X and X Y is g(0, 1) = 1, g(1, 1) = 0.5, g(0, 0) = 0, g(1, 0) = 0.

Let X be a fuzzy formula containing fuzzy propositions 1 , , n . It is clear that any assignment 1 = x1 , , n = xn is logical. In addition, for each assignment, there are only two possible outcomes about X. That is, X = 0 and X = 1. Hence, it follows from Denition 2.7.1 that the inference function of X on 1 , , n degenerates to its truth function.

2.8

Inference Rule

Suppose that X1 , , Xn and Y are fuzzy formulas, and g is the inference function of Y on X1 , , Xn . It follows from the monotonicity axiom of credibility measure that Cr {g(X1 , , Xn ) = 1} Cr {g(X1 , , Xn ) > 0} . Inference Rule: Suppose that X1 , , Xn and Y are fuzzy formulas, and g is the inference function of Y on X1 , , Xn . Then we infer Cr {g(X1 , , Xn ) = 1} T (Y ) Cr {g(X1 , , Xn ) > 0} . Let X be a fuzzy formula containing fuzzy propositions 1 , , n . In this case, since the inference function of X on 1 , , n degenerates to its truth function f , we have Cr {g(1 , , n ) = 1} = Cr {f (1 , , n ) = 1} ,

Section 2.8 - Inference Rule

Cr {g(1 , , n ) > 0} = Cr {f (1 , , n ) > 0} = Cr {f (1 , , n ) = 1} . It follows from the inference rule that T (X) = Cr {f (1 , , n ) = 1} = Cr{X = 1}. Therefore, the inference rule and the denition of truth value are consistent. Remark 2.3: Since the inference function of Y on X and X Y is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0.5, g(0, 0) = 0,

it follows from inference rule that T (Y ) Cr{g(X, X Y ) > 0} = Cr{X Y = 1} = T (X Y ), and T (Y ) Cr{g(X, X Y ) = 1} = Cr{X = 1, X Y = 1} Cr{X = 1} + Cr{X Y = 1} 1 = T (X) + T (X Y ) 1. That is, T (X)+T (X Y )1 T (Y ) T (X Y ). Hence, the credibilistic modus ponens is a special case of inference rule. Credibilistic modus ponens essentially provides a lower bound estimation and an upper bound estimation for T (Y ). For example, from T (X) = 0.9 and T (X Y ) = 0.8 we infer T (Y ) 0.7 and T (Y ) 0.8. However, these estimations may be trivial if we apply credibilistic modus ponens continuously. Example 2.2: For example, if T (X) = 0.7, T (X Y ) = 0.6 and T (Y Z) = 0.7, then we have T (Z) T (Y ) + T (Y Z) 1 T (X) + T (X Y ) + T (Y Z) 2 = 0. In order to solve this problem, we will prove a modied credibilistic modus ponens under certain condition. Theorem 2.8.1 (Modied Credibilistic Modus Ponens) Let X and X Y be fuzzy formulas with T (X) + T (X Y ) > 1. Then we have T (X) T (X Y ) T (Y ) T (X Y ). Proof: Since Cr{X = 1} + Cr{X Y = 1} = T (X) + T (X Y ) > 1, it follows from (??) that T (Y ) Cr{X = 1, X Y = 1} = Cr{X = 1} Cr{X Y = 1} = T (X) T (X Y ).

10

Chapter 2 - Credibilistic Logic

On the other hand, it follows from the credibilistic modus ponens that T (Y ) T (X Y ). The proof is complete. If T (X) T (X Y ) > 0.5, it is clear that T (X) + T (X Y ) > 1, and it follows from the modied credibilistic modus ponens that T (Y ) = T (X Y ). Example 2.3: Let us reconsider Example 2.8. Since T (X)+T (X Y ) > 1, it follows from the modied credibilistic modus ponens that T (Y ) T (X) T (X Y ) = 0.7 0.6 = 0.6. Furthermore, since T (Y ) + T (Y Z) 0.6 + 0.7 > 1, we get T (Z) T (Y ) T (Y Z) 0.6 0.7 = 0.6.

Remark 2.4: Since the inference function of X on Y and X Y is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0.5, g(0, 0) = 0,

it follows from the inference rule that T (X) Cr{X Y = 1} = T (X Y ) and T (X) Cr{Y = 1, X Y = 1} Cr{Y = 1} + Cr{X Y = 1} 1 = T (Y ) + T (X Y ) 1. That is, T (Y ) + T (X Y ) 1 T (X) T (X Y ). Hence, the credibilistic modus tollens is a special case of inference rule. Theorem 2.8.2 (Modied Credibilistic Modus Tollens) Let Y and X Y be fuzzy formulas with T (Y ) + T (X Y ) > 1. Then we have T (Y ) T (X Y ) T (X) T (X Y ). Proof: The theorem follows immediately from the modied credibilistic modus ponens and the fact T (X Y ) = T (Y X). If T (Y ) T (X Y ) > 0.5, it is clear that T (Y ) + T (X Y ) > 1, and it follows from the modied credibilistic modus ponens that T (Y ) = T (X Y ). Theorem 2.8.3 Let X and X Y be fuzzy formulas with T (X) + T (X Y ) > 1. Then we have T (X) T (X Y ) T (Y ) T (X Y ).

Section 2.8 - Inference Rule

11

Proof: Since T (X Y ) = T (X Y ), we have T (X) + T (X Y ) > 1. It follows from the modied credibilistic modus ponens that T (Y ) T (X Y ) = T (X Y ) and T (Y ) T (X) T (X Y ) = T (X) T (X Y ). The proof is complete. Theorem 2.8.4 Let X Y and Z X be fuzzy formulas with T (X Y ) + T (Z X) > 1. Then we have T (Y Z) T (Z X) T (X Y ). Proof: Since (X Y ) (Z X Y Z) is a tautology in classical logic, we have T ((X Y ) (Z X Y Z)) = 1. According to the modied credibilistic modus ponens, we get T (Z X Y Z) T (X Y ). In addition, it follows from T (Z X) + T (Z X Y Z) T (Z X) + T (X Y ) > 1 that T (Y Z) T (Z X) T (Z X Y Z) T (Z X) T (X Y ). The proof is complete. Theorem 2.8.5 Let X Z and Y Z be fuzzy formulas with T (X Z) + T (Y Z) > 1. Then T (X Y Z) = T (X Z) T (Y Z). Proof: It is easy to prove that the inference function of X Y Z on X Z and Y Z is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0, g(0, 0) = 0.

It follows from the inference rule that T (X Y Z) = Cr{X Z = 1, Y Z = 1} = Cr{X Z = 1} Cr{Y Z = 1} = T (X Z) T (Y Z). The proof is complete. Corollary 2.1: Let X Z and X Z be fuzzy formulas with T (X Z) + T (X Z) > 1. Then T (Z) = T (X Z) T (X Z). Proof: The proof follows immediately from Theorem 3.7.3 and the fact T (Z) = Cr{Z = 1} = Cr{X X Z = 1} = T (X X Z).

12

Chapter 2 - Credibilistic Logic

Theorem 2.8.6 Let X Z, Y Z and X Y be fuzzy formulas with T (X Y ) + T (X Z) > 1, T (X Y ) + T (Y Z) > 1 and T (X Z) + T (Y Z) > 1. Then we have T (X Y ) T (X Z) T (Y Z) T (Z) T (X Z) T (Y Z). Proof: Since T (X Z) + T (Y Z) > 1, it follows from Theorem 3.7.3 that T (X Y Z) = T (X Z) T (Y Z). It is easy to prove that T (XY )+T (XY Z) = (T (XY )+T (X Z))(T (XY )+T (Y Z)) > 1. Hence, it follows from the modied credibilistic modus ponens that T (X Y ) T (X Z) T (Y Z) T (Z) T (X Z) T (Y Z). The proof is complete. Theorem 2.8.7 Let X Y and X (Y Z) be fuzzy formulas with T (X Y ) + T (X (Y Z)) > 1. Then we have T (X Y ) T (X (Y Z)) T (X Z) T (X (Y Z)). Proof: Since T (X (Y Z)) = T (X Y Z) = T ((X Y ) (X Z)), we have T (X Y ) + T ((X Y ) (X Z)) > 1. It follows from the modied credibilistic modus ponens that T (X Z) T ((X Y ) (X Z)) = T (X (Y Z)), T (X Z) T (X Y )T ((X Y ) (X Z)) = T (X Y )T (X (Y Z)). The proof is complete. Theorem 2.8.8 Let X Y and Y Z be fuzzy formulas. If T (X Y ) + T (Y Z) > 1, then we have T (X Z) T (X Y ) T (Y Z), if T (X Y ) + T (Y Z) < 1, then we have T (X Z) T (X Y ) T (Y Z).

Section 2.9 - Entailment

13

Proof: It is easy to prove that the inference function of X Z on X Y and Y Z is g(1, 1) = 1, g(1, 0) = 0.5, g(0, 1) = 0.5, g(0, 0) = 0.

If T (X Y ) + T (Y Z) > 1, then it follows from (??) that T (X Z) Cr{X Y = 1, Y Z = 1} = Cr{X Y = 1} Cr{Y Z = 1} = T (X Y ) T (Y Z). If T (X Y ) + T (Y Z) < 1, then it follows from (??) that T (X Z) Cr{X Y = 1 or Y Z = 1} Cr{X Y = 1} Cr{Y Z = 1} = T (X Y ) T (Y Z). The proof is complete.

2.9

Entailment

In this section, we will introduce the concepts of entailment and quasientailment in credibilistic logic. Denition 2.9.1 Let X1 , , Xn and Y be fuzzy formulas. Then we say X1 , , Xn entails Y if T (Xi ) = 1 for all 1 i n implies T (Y ) = 1. We will use a standard notation for entailment: where X1 , , Xn and Y are fuzzy formulas, X1 , , Xn |= Y means X1 , , Xn entails Y . Since entailment is within a system, we will use unsubscripted |= to indicate entailment in classical logic, and |=C to indicate entailment in credibilistic logic. Theorem 2.9.1 X1 , , Xn |= Y if and only if X1 , , Xn |=C Y . Proof: Assume X1 , , Xn |= Y . Then X1 Xn Y is a tautology in classical logic, which implies that T (X1 Xn Y ) = 1. Furthermore, if T (Xi ) = 1 for all 1 i n, it follows from Theorem ?? that T (X1 Xn ) = T (X1 ) T (Xn ) = 1. Finally, it follows from the modied credibilistic modus ponens that T (Y ) = 1. Hence, we have X1 , , Xn |=C Y . In addition, the converse follows immediately from the consistency of credibilistic logic and classical logic. The proof is complete. Let X be a fuzzy formula. If T (X) > 0.5, then X is somewhat true because it follows from the self-duality that T (X) < 0.5, which implies that T (X) < T (X). Denition 2.9.2 Let X1 , , Xn and Y be fuzzy formulas. Then we say X1 , , Xn quasi-entails Y if T (Xi ) > 0.5 for all 1 i n implies T (Y ) > 0.5.

14

Chapter 2 - Credibilistic Logic

We will use a notation X1 , , Xn |=Q Y to express that X1 , , Xn quasi-entails Y . Theorem 2.9.2 If X1 , , Xn |= Y , then X1 , , Xn |=Q Y . Proof: Assume X1 , , Xn |= Y . Then we have T (X1 Xn Y ) = 1. If T (Xi ) > 0.5 for all 1 i n, we have T (Xi ) + T (Xj ) > 1 for each i = j and 1 i, j n. According to Theorem ??, we get T (X1 Xn ) = T (X1 ) T (Xn ) > 0.5. It follows from the modied credibilistic modus ponens that T (Y ) T (X1 Xn )T (X1 Xn Y ) = T (X1 ) T (Xn ) > 0.5. Hence, we have X1 , , Xn |=Q Y . The proof is complete. Since X, X Y entails Y in classical logic, we have X, X Y |=Q Y . That is, if X and X Y are all somewhat true, then Y is somewhat true. Similarly, it is easy to prove that Y, X Y |=Q X, X Y, Y Z |=Q X Z and X Y, X Z, Y Z |=Q Z.

Chapter 3

Hybrid Logic
Probabilistic logic and credibilistic logic are used to deal with random knowledge and fuzzy knowledge, respectively. In this section, a hybrid logic will be introduced within the framework of chance theory for dealing with random knowledge and fuzzy knowledge simultaneously. A brief introduction about chance theory may be found in Appendix 2.

3.1

Hybrid Proposition

Denition 3.1.1 A hybrid formula is dened as a member of the minimal set S of nite sequence of primitive symbols satisfying: (a) S for each fuzzy proposition ; (b) S for each random proposition ; (c) if X S, then X S; (d) if X S and Y S, then X Y, X Y, X Y S.

Example 3.1: Let be a fuzzy proposition Tom lives in Beijing with credibility , a random proposition It rains in Beijing with probability and a random proposition Tom stays home with probability . Then the sentence If Tom lives in Beijing and it rains in Beijing, then Tom stays home is a hybrid formula, denoted by . Assume X is a hybrid formula containing fuzzy propositions 1 , , n and random propositions 1 , , m . Then X is essentially a hybrid variable dened by its truth function f as X = f (1 , , n , 1 , , m ). In this equation, the symbols 1 , , n and 1 , , m are considered as fuzzy variables and random variables, respectively.

16

Chapter 3 - Hybrid Logic

3.2 3.3 3.4

Hybrid Formula Truth Function Truth Value

Denition 3.4.1 Let X be a hybrid formula. Then its truth value is dened as T (X) = Ch{X = 1}. (3.1) In addition, if X = f (1 , , n , 1 , , m ), then it follows from Denition 3.4.1 that T (X) = Ch{f (1 , , n , 1 , , m ) = 1}. Remark 3.1: If the hybrid formula X degenerates to a random formula, then it follows from the chance composition theorem (??) that T (X) = Ch{X = 1} = Pr{X = 1}. On the other hand, if X degenerates to a fuzzy formula, then we have T (X) = Ch{X = 1} = Cr{X = 1}. Hence hybrid logic, probabilistic logic and credibilistic logic are consistent. Example 3.2: Let be a fuzzy proposition with credibility , and let be a random proposition with probability . Then (i) is a hybrid formula with truth value T () = Ch{ = 1} = Ch{ = 1, = 1} = Cr{ = 1}Pr{ = 1} = ; (ii) is a hybrid formula with truth value T () = Ch{ = 1} = Ch{ = 1 or = 1} = Cr{ = 1}Pr{ = 1} = ; (iii) is a hybrid formula with truth value T ( ) = Ch{ = 1} = 1 Ch{ = 1, = 0} = 1 Cr{ = 1} (1 Pr{ = 1}) = 1 (1 ) = (1 ) .

Example 3.3: Let be a fuzzy proposition with credibility , and let and be random propositions with probabilities and , respectively. Then

Section 3.4 - Truth Value

17

is a hybrid formula with truth value T ( ) = 1 Ch{ = 1, = 1, = 0} = 1 Cr{ = 1} Pr{ = 1, = 0} = 1 Cr{ = 1} (Pr{ = 1} Pr{ = 0}) = 1 ( (1 )).

Example 3.4: Suppose that and are fuzzy propositions with credibilities and , respectively, and is a random proposition with probability . Then is a hybrid formula with truth value T ( ) = 1 Ch{ = 1, = 1, = 0} = 1 Cr{ = 1, = 0} Pr{ = 1} = 1 Cr{ = 1} Pr{ = 0} Pr{ = 1} = 1 (1 ).

Theorem 3.4.1 Let X be a hybrid formula containing fuzzy propositions 1 , , n and random propositions 1 , , m . If its truth function is f (x1 , , xn , y1 , , ym ), then its truth value is (x) (x) max (y) , if max (y) < 0.5, x x 2 2 f (x,y)=1 f (x,y)=1 T (X) = (x) 1 max (x) (y) , if max (y) 0.5, x x 2 2
f (x,y)=0 f (x,y)=1

where x = (x1 , , xn ), y = (y1 , , ym ), is the joint membership function of (1 , , n ) and is the joint probability mass function of (1 , , n ). Proof: Since X = f (1 , , n , 1 , , m ), we have T (X) = Ch{X = 1} = Ch{f (1 , , n , 1 , , m ) = 1}. The theorem follows immediately from the denition of chance measure. Remark 3.2: Let x be an n-dimensional vector (x1 , , xn ) with xi {0, 1} for all 1 i n, and y an m-dimensional vector (y1 , , ym ) with yj {0, 1} for all 1 j m. Since fuzzy variable i is dened as i = its membership function is i (xi ) = 2ci 1, if xi = 1 2(1 ci ) 1, if xi = 0, 1, with credibility ci 0, with credibility 1 ci ,

18

Chapter 3 - Hybrid Logic

and the joint membership function of fuzzy vector (1 , , n ) is (x) = min i (xi ).
1in

In addition, since random variable j is dened as j = 1, with probability pj 0, with probability 1 pj ,

its probability mass function is j (yj ) = pj , if yj = 1 1 pj , if yj = 0,

and the joint probability mass function of random vector (1 , , m ) is (y) =

1jm

j (yj ).

Theorem 3.4.2 (Self-Duality) For any hybrid formula X, we have T (X) = 1 T (X). Proof: It follows from the self-duality of chance measure (??) that T (X) = Ch{X = 1} = Ch{X = 0} = 1 Ch{X = 1} = 1 T (X). The proof is complete. Generally speaking, the higher the truth value is, the more true the hybrid formula is. For any hybrid formula X, T (X) = 1 means X is certainly true and T (X) = 0 means X is certainly false. Theorem 3.4.3 (Monotonicity and Subadditivity) For any hybrid formulas X and Y , we have T (X) T (Y ) T (X Y ) T (X) + T (Y ) Proof: It follows from the monotonicity of chance measure (??) that T (X Y ) = Ch{X Y = 1} = Ch{X = 1 or Y = 1} Ch{X = 1} Ch{Y = 1} = T (X) T (Y ). Furthermore, it follows from the subadditivity of chance measure (??) that T (X Y ) = Ch{X = 1 or Y = 1} Ch{X = 1} + Ch{Y = 1} = T (X) + T (Y ).

Section 3.4 - Truth Value

19

The proof is complete. Remark 3.3: The hybrid logic and classical logic are consistent. First, for any hybrid formula X, if T (X) = 1, it follows from Theorem 3.4.2 that T (X) = 1 T (X) = 0, and if T (X) = 0, we have T (X) = 1. Furthermore, for any hybrid formulas X and Y , if T (X) = 1 or T (Y ) = 1, it follows from Theorem 3.4.3 that T (X Y ) T (X) T (Y ) = 1, which implies that T (X Y ) = 1. If T (X) = 0 and T (Y ) = 0, it follows from Theorem 3.4.3 that T (X Y ) T (X) + T (Y ) = 0. That is, T (X Y ) = 0. Theorem 3.4.4 For any hybrid formula X, we have T (X X) = 1, T (X X) = T (X).

Proof: It follows from the normality of chance measure that T (X X) = Ch{X X = 1} = Ch{X = 1 or X = 0} = 1. In addition, it follows from Denition 3.4.1 that T (X X) = Ch{X X = 1} = Ch{X = 1} = T (X). The proof is complete. Theorem 3.4.5 For any hybrid formulas X and Y , we have T (X) + T (Y ) 1 T (X Y ) T (X) T (Y ). Proof: It follows from Theorem 3.4.3 that T (X Y ) = 1 T (X Y ) 1 T (X) T (Y ) = T (X) T (Y ), T (X Y ) = 1 T (X Y ) 1 (T (X) + T (Y )) = T (X) + T (Y ) 1. The proof is complete. Theorem 3.4.6 For any hybrid formula X, we have T (X X) = 0, T (X X) = T (X).

20 Proof: It follows from Theorem 3.4.4 that

Chapter 3 - Hybrid Logic

T (X X) = 1 T (X X) = 1 1 = 0, T (X X) = 1 T (X X) = 1 T (X) = T (X). The proof is complete. Theorem 3.4.7 For any hybrid formulas X and Y , we have (1 T (X)) T (Y ) T (X Y ) 1 T (X) + T (Y ). Proof: It follows from Theorem 3.4.3 and the fact X Y = X Y that T (X Y ) = T (X Y ) T (X) T (Y ) = (1 T (X)) T (Y ), T (X Y ) = T (X Y ) T (X) + T (Y ) = 1 T (X) + T (Y ). The proof is complete. Theorem 3.4.8 For any hybrid formula X, we have T (X X) = 1, T (X X) = 1 T (X).

Proof: It follows from Theorem 3.4.4 that T (X X) = T (X X) = 1, T (X X) = T (X X) = T (X) = 1 T (X). The proof is complete. Theorem 3.4.9 For any hybrid formulas X and Y , we have T (Y X) = T (X Y ). Proof: Since X Y = X Y , we have T (Y X) = T (Y X) = T (X Y ) = T (X Y ). The proof is complete. Theorem 3.4.10 For any fuzzy formula X and random formula Y , we have (a) T (X Y ) = T (X) T (Y ); (b) T (X Y ) = T (X) T (Y ); (c) T (X Y ) = (1 T (X)) T (Y ) where T (X) represents the truth value in credibilistic logic and T (Y ) represents the truth value in probabilistic logic.

Section 3.6 - Inference Function

21

Proof: Part (a). It follows from chance composition theorem (??) that T (XY ) = Ch{XY = 1} = Ch{X = 1, Y = 1} = Cr{X = 1}Pr{Y = 1} = T (X)T (Y ). Part (b). It follows from the self-duality theorem and part (a) that T (X Y ) = T ((X Y )) = 1 T (X) T (Y ) = T (X) T (Y ). Part (c). It follows from the self-duality theorem and part (b) that T (X Y ) = T (X Y ) = T (X) T (Y ) = (1 T (X)) T (Y ). The proof is complete.

3.5 3.6

Modus Ponens Inference Function

Let X, X Y and Y be hybrid formulas. It follows from Table 1 that if X is false and X Y is true, then Y must be true; if X is true and X Y is true, then Y may be true or false; if X and X Y are all false, then Y must be false; X is true and X Y is false is illogical because X is true implies that X Y is true. X 1 1 0 0 Y 1 0 1 0 X Y 1 1 1 0

Table 3.1: The truth table of disjunction Generally, suppose that X1 , , Xn and Y are hybrid formulas. An assignment X1 = x1 , , Xn = xn may be illogical. Otherwise, there are three possible outcomes about Y . That is, Y = 1 (true), Y = 0 (false) and Y = 1 or 0 (undetermined). For any hybrid formulas X1 , , Xn and Y , dene a function g : {0, 1}n {0, 0.5, 1} as (i) g(x1 , , xn ) = 1 if X1 = x1 , , Xn = xn implies Y = 1; (ii) g(x1 , , xn ) = 0 if X1 = x1 , , Xn = xn implies Y = 0; (iii) g(x1 , , xn ) = 0.5 if X1 = x1 , , Xn = xn implies Y = 0 or Y = 1; (iv) g(x1 , , xn ) = 0 if X1 = x1 , , Xn = xn is illogical. Such a function g is called inference function of Y on X1 , , Xn . At times it causes no confusion, the function g will be called the inference function of Y simply.

22

Chapter 3 - Hybrid Logic

For example, the inference function of Y on X, X Y is g(0, 1) = 1, g(1, 1) = 0.5, g(0, 0) = 0, g(1, 0) = 0.

In addition, it follows from Table 2 that the inference function of Y on X, X Y is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0.5, g(0, 0) = 0.

X 1 1 0 0

Y 1 0 1 0

XY 1 0 1 1

Table 3.2: The truth table of implication

3.7

Inference Rule

Inference Rule: Suppose that X1 , , Xn and Y are hybrid formulas, and g is the inference function of Y on X1 , , Xn . Then we infer Ch {g(X1 , , Xn ) = 1} T (Y ) Ch {g(X1 , , Xn ) > 0} . Let X be a hybrid formula containing fuzzy propositions 1 , , n and random propositions 1 , , m . It is clear that any assignment 1 = x1 , , n = xn , 1 = y1 , , m = ym is logical. In addition, for each assignment, there are only two possible outcomes about X. That is, X = 0 and X = 1. Hence, the inference function of X degenerates to its truth function. In this case, we have Ch {g(1 , , n , 1 , , m ) > 0} = Ch {g(1 , , n , 1 , , m ) = 1} = Ch {f (1 , , n , 1 , , m ) = 1} , which implies that the inference rule and the denition of truth value are consistent. Theorem 3.7.1 (Hybrid Modus Ponens) From T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ).

Section 3.7 - Inference Rule

23

Proof: Since the inference function of Y on X, X Y is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0.5, g(0, 0) = 0,

we have g(X, X Y ) > 0 if and only if X Y = 1 and g(X, X Y ) = 1 if and only if X = 1 and X Y = 1. It follows from the inference rule that T (Y ) Ch{g(X, X Y ) > 0} = Ch{X Y = 1} = T (X Y ), and T (Y ) Ch{g(X, X Y ) = 1} = Ch{X = 1, X Y = 1} Ch{X = 1} + Ch{X Y = 1} 1 = T (X) + T (X Y ) 1. The proof is complete. For example, let be a hybrid formula If Tom lives in Beijing and it rains in Beijing, then Tom stays home with chance value , and a fuzzy proposition Tom lives in Beijing with credibility and a random propositionIt rains in Beijing with probability . Then it follows from the hybrid modus ponens that Tom stays home with truth value belonging to [ + 1, ]. Corollary 3.1: (Hybrid Modus Tollens) From T (Y ) and T (X Y ) we infer T (Y ) + T (X Y ) 1 T (X) T (X Y ). Proof: Since T (X Y ) = T (Y X), it follows from the hybrid modus ponens that T (X) T (Y X) = T (X Y ), T (X) T (Y ) + T (Y X) 1 = T (Y ) + T (X Y ) 1. The proof is complete. Corollary 3.2: (Hybrid Left Disjoint) From T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ). Proof: Since T (X Y ) = T (X Y ), it follows from the hybrid modus ponens that T (Y ) T (X Y ) = T (X Y ), T (Y ) T (X) + T (X Y ) 1 = T (X) + T (X Y ) 1. The proof is complete. Theorem 3.7.2 From T (X Y ) and T (Z X) we infer T (X Y ) + T (Z X) 1 T (Y Z) 1.

24

Chapter 3 - Hybrid Logic

Proof: Since the inference function of Y Z on X Y, Z X is g(1, 1) = 1, g(1, 0) = 0.5, g(0, 1) = 0.5, g(0, 0) = 0.5,

we have g(X Y, Z X) = 1 if and only if X Y = 1 and Z X = 1. It follows from the inference rule that T (Y Z) Ch{g(X Y, ZX) > 0} = 1 and T (Y Z) Ch{g(X Y, Z X) = 1} = Ch{X Y = 1, Z X = 1} Ch{X Y = 1} + Ch{Z X = 1} 1 = T (X Y ) + T (Z X) 1. The proof is complete. Theorem 3.7.3 From T (X Z) and T (Y Z) we infer T (X Z) + T (Y Z) 1 T (X Y Z) T (X Z) T (Y Z). Proof: It is easy to prove that the inference function of X Y Z on X Z, Y Z is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0, g(0, 0) = 0.

It follows from the inference rule that T (X Y Z) = Ch{X Y = 1, Y Z = 1}. According to the monotonicity of chance measure (??) that T (X Y Z) Ch{X Z} Ch{Y Z} = T (X Z) T (Y Z), and it follows from the subadditivity of chance measure (??) that T (XY Z) Ch{X Z}+Ch{Y Z}1 = T (X Z)+T (Y Z)1. The proof is complete. Corollary 3.3: From T (X Z) and T (X Z) we infer T (X Z) + T (X Z) 1 T (Z) T (X Z) T (X Z). Proof: It follows from Denition 3.4.1 that T (Z) = Ch{Z = 1} = Ch{X X Z = 1} = T (X X Z). The corollary follows immediately from Theorem 3.7.3. Theorem 3.7.4 From T (X Y ) and T (X (Y Z)) we infer T (X Y ) + T (X (Y Z)) 1 T (X Z) T (X (Y Z)).

Section 3.8 - Entailment

25

Proof: Since the inference function of X Z on X Y, X (Y Z) is g(1, 1) = 1, g(1, 0) = 0, g(0, 1) = 0.5, g(0, 0) = 0,

it follows from the inference rule that T (X Z) Ch {g(X Y, X (Y Z)) > 0} = Ch{X (Y Z) = 1} = T (X (Y Z)), and T (X Z) Ch {g(X Y, X (Y Z)) = 1} = Ch{X Y = 1, X (Y Z) = 1} Ch{X Y = 1} + Ch{X (Y Z) = 1} 1 = T (X Y ) + T (X (Y Z)) 1. The proof is complete. Theorem 3.7.5 From T (X Y ), T (X Z) and T (Y Z) we infer T (X Y ) + T (X Z) + T (Y Z) 2 T (Z) T (X Z) T (Y Z). Proof: Since the inference function of Z on X Y , X Z, Y Z is g(1, 1, 1) = 1, g(0, 1, 1) = 0.5, g(1, 1, 0) = 0, g(0, 1, 0) = 0, g(1, 0, 1) = 0, g(0, 0, 1) = 0, g(1, 0, 0) = 0, g(0, 0, 0) = 0,

it follows from the inference rule that T (Z) Ch{X Z = 1, Y Z = 1} Ch{X Z = 1} Ch{Y Z = 1} = T (X Z) T (Y Z), according to the subadditivity of chance measure (??) we get T (Z) Ch{X Y = 1, X Z = 1, Y Z = 1} Ch{X Y = 1} + Ch{X Z = 1} + Ch{Y Z = 1} 2 = T (X Y ) + T (X Z) + T (Y Z) 2. The proof is complete.

3.8

Entailment

Chapter 4

Uncertain Logic
In this section, we introduce an uncertain logic within the framework of uncertainty theory as a generalization of probabilistic logic, credibilistic logic and hybrid logic. A brief knowledge about uncertainty theory may be found in Appendix 3.

4.1 4.2 4.3 4.4

Uncertain Proposition Uncertain Formula Truth Function Truth Value

In uncertain logic, we dene the truth value for each proposition as uncertain measure value. Hence, we use the terms uncertain proposition and uncertain formula instead of proposition and formula, respectively. Generally speaking, we use to express an uncertain proposition and use u to express its uncertain measure value. If we use = 1 to express is true, and use = 0 to express is false, then is essentially an uncertain variable dened as = 1, with uncertain measure u 0, with uncertain measure 1 u.

Let X be an uncertain formula containing uncertain propositions 1 , , n . It is clear that X is essentially an uncertain variable taking values 0 or 1 dened by its truth function f as X = f (1 , , n ). In this equation, the symbols 1 , , n are considered as uncertain variables.

28

Chapter 4 - Uncertain Logic

Denition 4.4.1 For each uncertain formula X, its truth value is dened as T (X) = M{X = 1}. For any uncertain proposition , it is easy to prove that M{ = 1} = u. That is, the truth value of each uncertain proposition is just its uncertain measure value. In addition, if X = f (1 , , n ), then T (X) = M{f (1 , , n ) = 1}. Example 4.1: If is an uncertain proposition with uncertain measure value , then is an uncertain formula and its truth value is T ( ) = M{ = 1} = 1 M{ = 1} = 1 T (X) = 1 .

Example 4.2: Suppose that and are two uncertain propositions with uncertain measure values and , respectively. Then we have (i) is an uncertain formula with truth value belonging to [ , ( + ) 1] because T ( ) = M{ = 1 or = 1} M{ = 1} M{ = 1} = , T ( ) = M{ = 1 or = 1} M{ = 1} + M{ = 1} = + ; (ii) is an uncertain formula with truth value belonging to [( + 1) 0, ] because T ( ) = M{ = 1, = 1} M{ = 1} + M{ = 1} 1 = + 1, T ( ) = M{ = 1, = 1} M{ = 1} M{ = 1} = ; (iii) is an uncertain formula with truth value belonging to [(1 ) , (1 + ) 1] because T ( ) = T ( ) (1 ) , T ( ) = T ( ) 1 + .

Theorem 4.4.1 (Self-Duality) For any uncertain formula X, we have T (X) = 1 T (X). Proof: It follows from the self-duality of uncertain measure that T (X) = M{X = 1} = 1 M{X = 1} = 1 T (X). The proof is complete.

Section 4.7 - Inference Rule

29

Theorem 4.4.2 (Monotonicity and Subadditivity) For any uncertain formulas X and Y , we have T (X) T (Y ) T (X Y ) T (X) + T (Y ) Proof: It follows from the monotonicity of uncertain measure that T (X Y ) = M{X Y = 1} M{X = 1} M{Y = 1} = T (X) T (Y ). On the other hand, it follows from the subadditivity of uncertain measure that T (X Y ) = M{X Y = 1} M{X = 1} + M{Y = 1} = T (X) + T (Y ). The proof is complete. Remark 4.1: It follows from the self-duality, monotonicity and subadditivity of uncertain truth value that uncertain logic and classical logic are consistent. Theorem 4.4.3 For any uncertain formulas X and Y , we have (a) T (X X) = 1, T (X X) = T (X); (b) T (X X) = 0, T (X X) = T (X); (c) T (X) + T (Y ) 1 T (X Y ) T (X) T (Y ); (d) T (Y X) = T (X Y ); (e) T (X X) = 1, T (X X) = 1 T (X); (f) (1 T (X)) T (Y ) T (X Y ) 1 T (X) + T (Y ). Proof: Part (a). It follows from the normality of uncertain measure that T (X X) = M{X X = 1} = M{X = 1 or X = 0} = 1. Furthermore, it follows from Denition 4.4.1 that T (X X) = M{X X = 1} = M{X = 1} = T (X). The other parts may be proved by a similar way of hybrid case.

4.5 4.6 4.7

Modus Ponens Inference Function Inference Rule

In uncertain logic, the inference function of uncertain formula Y on X1 , , Xn may be dened by a similar way of hybrid case.

30

Chapter 4 - Uncertain Logic

Inference Rule: Suppose that X1 , , Xn and Y are uncertain formulas, and g is the inference function of Y on X1 , , Xn . Then we infer

M {g(X1 , , Xn ) = 1} T (Y ) M {g(X1 , , Xn ) > 0} .

As special cases of this rule, we introduce some inference theorems including the uncertain modus ponens, uncertain modus tollens and uncertain left disjoint, which may be proved by a similar way of hybrid case. (a) Uncertain Modus Ponens: from T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ). (b) Uncertain Modus Tollens: from T (Y ) and T (X Y ) we infer T (Y ) + T (X Y ) 1 T (X) T (X Y ). (c) Uncertain Left Disjoint: from T (X) and T (X Y ) we infer T (X) + T (X Y ) 1 T (Y ) T (X Y ).

4.8

Entailment

List of Frequently Used Symbols

, , , , Xt , Yt , Zt Nt Bt Ct , Cr Pr E (, A, Pr) (, P, Cr)
n

fuzzy variables fuzzy vectors fuzzy process, stochastic process fuzzy renewal process Brownian motion Liu process membership functions empty set credibility measure probability measure expected value probability space credibility space set of real numbers set of n-dimensional real vectors maximum operator minimum operator

Index
American option, ?? Asian option, ?? Brownian motion, ?? credibility inversion theorem, ?? credibility measure, ?? credibility space, ?? European option, ??, ?? exotic option, ?? expected value, ?? rst passage time, ?? fuzzy variable, ?? fuzzy dierential equation, ?? fuzzy process, ?? fuzzy renewal process, ?? fractional Liu process, ?? Gaos stock model, ?? Ito formula, ?? Ito integral, ?? Liu formula, ?? Liu integral, ?? Liu process, ?? Lius stock model, ?? membership function, ?? Pengs stock model, ?? Qin-Li option pricing formula, ?? reected principle, ?? stochastic dierential equation, ??