2. PROBABILITY SPACES 2.1 Probability functions Whenever probabilities cannot reasonably be assigned to all subsets of the sample space, it is important to distinguish between events and other subsets of the sample space. The probability function P:F[0,1] assigns probabilities to certain subsets (called events) of the sample space . Not all subsets of need to be events. If A is not in the domain F of the probability function P, then P(A) is undefined and A is not an event. Experiment: A die is tossed twice. Possible outcomes: (1,1),(1,2),…,(1,6), (2,1),(2,2),…,(2,6), (3,1),(3,2),…,(3,6), (4,1),(4,2),…,(4,6), (5,1),(5,2),…,(5,6), (6,1),(6,2),…,(6,6) Sample space: ={(1,1),(1,2),…,(6,5),(6,6)} Simple events: {(1,1)},{(1,2)},…,{(6,5)},{(6,6)} Simple event probabilities: P({(1,1)})=…=P({(6,6})= 361 In this example, probabilities can easily be assigned to all subsets of the sample space just by summing simple event probabilities, hence all subsets are events. For example, P({(4,5),(4,6),(5,6)})=P({(4,5)})+P({(4,6)})+P({(5,6)}) = 361 + 361 + 361 = 363 = 121 . 2.1 2.2 Uncountably infinite sample spaces An uncountably infinite sample space (e.g., an interval) has not only simple subsets (e.g., subintervals) but also more complicated ones, many of which are so odd that it is absolutely impossible to assign probabilities to them in a reasonable manner (see Appendix A). Example of a probability function P that is defined on all semiclosed subintervals of =(0,1]: P((a,b])=b-a The probability of an interval (a,b] is given by its length. The domain of P is F={(a,b]:0ab1}. Clearly, we should not call any function P a probability function unless it meets some obvious requirements. In our example, there seems to be no problem in that respect. Indeed, and P()=P((0,1])=1-0=1, P()=P((a,a])=a-a=0, P((0,b])+P((0,b]c)=(b-0)+P((b,1])=b+(1-b)=1. Moreover, if a1a2b1b2, then P((a1,b1](a2,b2])=P((a1,b2]) =b2-a1 =(b1-a1)+(b2-a2)-(b1-a2) =P((a1,b1])+P((a2,b2])-P((a2,b1]) =P((a1,b1])+P((a2,b2])-P((a1,b1](a2,b2])). 2.2 2.3 Extending the domain of a probability function Let P be the probability function defined by P((a,b])=b-a for all semiclosed subintervals of =(0,1]. The domain of P is given by F={(a,b]:0ab1}. If AF and BF, then ABF, but Ac, Bc, and AB need not necessarily be elements of F. A=(0.1,0.3], B=(0.5,0.7]) AB=F, Ac=(0,0.1](0.3,1]F, Bc=(0,0.5](0.7,1]F, AB=(0.1,0.3](0.5,0.7]F A=(0.1,0.6], B=(0.5,1]) AB=(0.5,0.6]F, Ac=(0,0.1](0.6,1]F, Bc=(0,0.5]F, AB=(0.1,1]F Despite the fact that some complements and unions of elements of F are not in F, we could easily assign probabilities to them by putting P(Ac)=1-P(A) and P(AB)=P(A)P(B)-P(AB), respectively. It seems therefore reasonable to increase the domain of P by those subsets of to which we can assign probabilities in a consistent manner. 2.3 2.4 -algebras Henceforth we will only consider probability functions whose domains are closed under the operations of taking complements, countable intersections and countable unions. A set F of subsets of a nonempty set is called a -algebra (or -field) if it satisfies ,F, AF AcF, and A1,A2,…F A1A2…F A1A2…F. Exercise 2.a: Let ={1,2,3}. Examine whether or not and F1={,{1},{2},{3},{1,2},{1,3},{2,3},}, F2={,}, F3={,{1},{2,3},}, F4={,{1},{2},{2,3},{1,3},} are -algebras. F1, F2, and F3 are -algebras, but F4 is not a -algebra, because {1}{2}={1,2}F4 and {1,3}{2,3}={3} F4. The smallest -algebra of subsets of the real line =(-,) that contains all intervals is called the Borel -algebra and its elements are called Borel sets. 2.4 2.5 Probability distributions A function P:F[0,1] that is defined on a -algebra F is called a probability distribution (or probability measure) if it satisfies (i) P()=0, P()=1, (ii) P(Ac)=1-P(A) if AF, and (iii) P(A1A2...)=P(A1)+P(A2)+… if A1,A2,… are pairwise disjoint elements of F. The triple (,F,P) is called a probability space. Exercise 2.b: Let ={1,2,3} and F={,{1},{2,3},}. Examine whether or not the functions given by and P1()=0, P1({1})=0, P1({2,3})=1, P1()=1, P2()=0, P2({1})= 87 , P2({2,3})=- 17 , P2()=1, P3()=0, P3({1})= 15 , P3({2,3})= 52 , P3()=1, P4()=0, P4({1})= 23 , P4({2,3})= 13 , P4()=1, P5()=0, P5({1})=0, P5({2,3})=0, P5()=1, respectively, are probability distributions. Obviously, only P1 and P4 are probability distributions. 2.5 2.6 Further exercises Exercise 2.c: Let P be the probability function defined by P((a,b])=b-a for all semiclosed subintervals of =(0,1]. Define both P((a,b)) and P([a,a]) in a consistent manner. 1 1 1 P((a,b))=P((a,b- 101 ](b- 101 ,b- 100 ](b- 100 ,b- 1000 ]…) 1 1 1 =P((a,b- 101 ])+P((b- 101 ,b- 100 ])+P((b- 100 ,b- 1000 ])+… 1 1 1 =((b- 101 )-a)+((b- 100 )-(b- 101 ))+((b- 1000 )-(b- 100 ))+… 1 1 1 =b-a- 101 +( 101 - 100 )+( 100 - 1000 )+… =b-a P([a,a])=1-P([a,a]c) =1-P((0,a)(a,1]) =1-(P((0,a))+P((a,1])) =1-((a-0)+(1-a)) =0 Exercise 2.d: Let ={0,1,2,3}. Examine whether or not F={,{0},{3},{1,2},{0,1,2},{1,2,3},{0,3},} is a -algebra. Exercise 2.e: Find the smallest -algebra of subsets of ={4,6,8,10,12,…} that contains {4,6} and {4,6,8}. 2.6