math.wiki
Probability › Foundations › Probability axioms and sample spaces
Sign in

Probability axioms and sample spaces

Whatever the experiment, which rules must every assignment of probabilities obey, and how much of probability follows from those rules alone?

Where you are 4 in Foundations

Why this exists

In Basic probability a probability is a count: when all outcomes are equally likely, it is the number of outcomes in the event divided by the number of outcomes in all. Three ordinary experiments break that definition. A loaded die has outcomes that are not equally likely. Rolling a pair of dice until a double appears can take 1,2,3,\dots rolls with no largest value, so there is no total to divide by. A pointer spun on a circle can stop at uncountably many angles, each with probability 0, while the whole circle has probability 1.

The way out, published by Andrey Kolmogorov in 1933, is to stop defining probability by a formula and instead list the properties every assignment of probabilities must have. Three properties suffice. Any function on events that has them is a probability; which one describes a particular die or pointer is a separate modeling decision. Every rule met in school, the complement rule and the addition rule among them, becomes a theorem, and one theory covers finite, countably infinite and continuous experiments together.

Definition

An experiment is anything with an uncertain result: a roll of two dice, a game of backgammon, a run of rolls continued until something happens.

The sample space \Omega is the set of all possible outcomes of the experiment, and a single outcome is written \omega. School lessons such as Listing a sample space: lists and tables and Sample spaces, outcomes and events call this set S; this course writes \Omega.

An event is a subset A\subseteq\Omega. The event A occurs when the outcome \omega that actually happens lies in A. The operations on Sets then read as statements about events.

SetRead as an event
A\cup BA or B (or both) occurs
A\cap BA and B both occur
A^cA does not occur
A\subseteq Bwhenever A occurs, so does B
A\cap B=\varnothingA and B cannot both occur: they are disjoint, or mutually exclusive
\Omega and \varnothingthe sure event and the impossible event

Events A_1,A_2,A_3,\dots are pairwise disjoint when A_i\cap A_j=\varnothing for every i\ne j.

A probability on \Omega is a function P that assigns a real number P(A) to every event A and satisfies three axioms.

Axiom 1 (nonnegativity). P(A)\ge 0 for every event A.

Axiom 2 (normalization). P(\Omega)=1.

Axiom 3 (countable additivity). If A_1,A_2,A_3,\dots are pairwise disjoint events, then

P\left(\bigcup_{i=1}^{\infty}A_i\right)=\sum_{i=1}^{\infty}P(A_i).

The pair (\Omega,P) is a probability space. When \Omega is finite or countably infinite, every subset of \Omega is an event. When \Omega is an interval or a region, the events form a smaller collection, for a reason given under Boundaries; nothing in this lesson depends on which collection it is, as long as it contains \Omega and is closed under complements and countable unions (and so, by De Morgan’s laws, under countable intersections).

Intuition

Think of probability as one unit of mass spread over \Omega, and of P(A) as the mass sitting on A. Mass is never negative (Axiom 1), there is exactly one unit of it (Axiom 2), and the mass on a region made of separate pieces is the total of the masses on the pieces (Axiom 3). Every result below can be read off a picture with this in mind. In Figure 1 the events A and B cut \Omega into four disjoint pieces, and the addition rule is a way of not counting the mass on the middle piece twice.

\Omega
A
B
A\cap B^c
A\cap B
A^c\cap B
A^c\cap B^c
Figure 1Two events cut \Omega into four disjoint pieces. Every rule about A\cup B is additivity applied to these pieces.

The axioms are also exactly what long-run frequencies obey. If an experiment is repeated n times and f_n(A) is the fraction of repetitions in which A occurs, as in Experimental probability and relative frequency, then f_n(A)\ge 0, f_n(\Omega)=1, and f_n(A\cup B)=f_n(A)+f_n(B) when A and B are disjoint, because each repetition is counted in at most one of them. The axioms keep only what every frequency has. Countable additivity goes one step past what a finite run can show: it is the assumption that makes limits behave, and Result 8 shows what it buys.

The axioms never say what any particular probability is. That two fair dice give each of the 36 ordered rolls probability \tfrac{1}{36} is a model of the dice, chosen as in Probability models: equally likely or not; the axioms only decide which models are consistent.

Results

Every statement holds in every probability space (\Omega,P), for any events named.

Result 1 (the impossible event). P(\varnothing)=0.

Result 2 (finite additivity). If A_1,\dots,A_n are pairwise disjoint, then P(A_1\cup\dots\cup A_n)=P(A_1)+\dots+P(A_n).

Result 3 (complement rule, as in Complements and probabilities that add to one). P(A^c)=1-P(A).

Result 4 (monotonicity). If A\subseteq B, then P(B\cap A^c)=P(B)-P(A), so P(A)\le P(B). In particular 0\le P(A)\le 1 for every event.

Result 5 (addition rule, as in The general addition rule).

P(A\cup B)=P(A)+P(B)-P(A\cap B).

Result 6 (union bound). For any events A_1,A_2,A_3,\dots, disjoint or not,

P\left(\bigcup_{i=1}^{\infty}A_i\right)\le\sum_{i=1}^{\infty}P(A_i),

and the same holds for finitely many events.

Result 7 (discrete probability spaces). Let \Omega be finite or countably infinite. If p:\Omega\to[0,\infty) satisfies \sum_{\omega\in\Omega}p(\omega)=1, then P(A)=\sum_{\omega\in A}p(\omega) defines a probability on \Omega. Conversely, every probability on \Omega has this form, with p(\omega)=P(\{\omega\}).

Result 8 (continuity). If A_1\subseteq A_2\subseteq A_3\subseteq\cdots, then P\left(\bigcup_n A_n\right)=\lim_{n\to\infty}P(A_n). If A_1\supseteq A_2\supseteq A_3\supseteq\cdots, then P\left(\bigcap_n A_n\right)=\lim_{n\to\infty}P(A_n). In words: along a growing chain of events the probabilities climb to the probability of the union, and along a shrinking chain they fall to the probability of the intersection.

Why it is true

Each proof uses only the axioms and the results before it.

Result 1. Take A_i=\varnothing for every i. These are pairwise disjoint and their union is \varnothing, so Axiom 3 gives P(\varnothing)=\sum_{i=1}^{\infty}P(\varnothing). The left side is a real number c\ge 0, and a series whose terms all equal c diverges unless c=0.

Result 2. Extend the list by A_{n+1}=A_{n+2}=\dots=\varnothing. The longer sequence is still pairwise disjoint with the same union, so Axiom 3 and Result 1 give P(A_1\cup\dots\cup A_n)=\sum_{i=1}^{n}P(A_i)+0+0+\cdots.

Result 3. A and A^c are disjoint with union \Omega, so 1=P(\Omega)=P(A)+P(A^c) by Axiom 2 and Result 2.

Result 4. When A\subseteq B, the set B splits into the disjoint pieces A and B\cap A^c. Result 2 gives P(B)=P(A)+P(B\cap A^c), which rearranges to the stated difference, and P(B\cap A^c)\ge 0 by Axiom 1. Taking B=\Omega gives P(A)\le 1.

Result 5. Figure 1 shows A\cup B as the union of the three disjoint pieces A\cap B^c, A\cap B and A^c\cap B, and shows A and B each made of two of them. Result 2 gives

\begin{aligned}P(A)&=P(A\cap B^c)+P(A\cap B),\\P(B)&=P(A^c\cap B)+P(A\cap B),\\P(A\cup B)&=P(A\cap B^c)+P(A\cap B)+P(A^c\cap B).\end{aligned}

The sum of the first two lines is the third line plus one extra P(A\cap B), which is the addition rule.

Result 6. Make the events disjoint without changing their union: let B_1=A_1 and, for n\ge 2, B_n=A_n\cap(A_1\cup\dots\cup A_{n-1})^c, the outcomes of A_n not in any earlier A_i. An outcome lying in some A_n lies in exactly one B_n, the one for the first such n, so the B_n are pairwise disjoint with the same union as the A_n. Also B_n\subseteq A_n. Axiom 3 and Result 4 give P\left(\bigcup A_n\right)=\sum P(B_n)\le\sum P(A_n). For finitely many events, set the rest equal to \varnothing.

Result 7. The sum \sum_{\omega\in A}p(\omega) is a series of nonnegative terms, and by a theorem of calculus such a series has the same sum, finite or +\infty, in every order and under every grouping of its terms. So P(A) is well defined and at most \sum_{\omega\in\Omega}p(\omega)=1. Axioms 1 and 2 are immediate. For Axiom 3, when A_1,A_2,\dots are pairwise disjoint, each outcome \omega of \bigcup A_i lies in exactly one A_i; grouping the terms p(\omega) of P\left(\bigcup A_i\right) by that A_i turns the sum into \sum_i P(A_i). For the converse, an event A is the union of the countably many disjoint events \{\omega\}, \omega\in A, so Axiom 3 (or Result 2 when A is finite) gives P(A)=\sum_{\omega\in A}P(\{\omega\}).

Result 8. For a growing chain, let B_1=A_1 and B_n=A_n\cap A_{n-1}^c for n\ge 2, the outcomes that first appear at stage n. The B_n are pairwise disjoint, B_1\cup\dots\cup B_n=A_n, and \bigcup_n B_n=\bigcup_n A_n. By Axiom 3, the definition of the sum of a series, and Result 2,

\begin{aligned}P\left(\bigcup_n A_n\right)&=\sum_{k=1}^{\infty}P(B_k)\\&=\lim_{n\to\infty}\sum_{k=1}^{n}P(B_k)\\&=\lim_{n\to\infty}P(A_n).\end{aligned}

For a shrinking chain the complements A_n^c grow, and \left(\bigcap_n A_n\right)^c=\bigcup_n A_n^c by De Morgan’s law. The first part and Result 3 give

\begin{aligned}P\left(\bigcap_n A_n\right)&=1-\lim_{n\to\infty}\left(1-P(A_n)\right)\\&=\lim_{n\to\infty}P(A_n).\end{aligned}

Worked examples

Example 1

Two records of the same roll

Mei and Kofi play backgammon, where each turn begins with a roll of two fair dice. Take as the model the 36 ordered rolls (i,j), first die then second, each with probability \tfrac{1}{36}. Kofi keeps a log that records only the unordered roll, such as \{3,5\} or the double \{4,4\}. Find the probability of each outcome in Kofi’s sample space, check that these values form a probability, and find the probability that a roll is a double and that it contains a 6.

Solution

Kofi’s sample space has 6 doubles \{i,i\} and \binom{6}{2}=15 pairs \{i,j\} with i<j, so 21 outcomes. Each of them is an event in the ordered model, so its probability is the number of ordered rolls inside it over 36 (Result 7). A double \{i,i\} contains only (i,i), so p(\{i,i\})=\tfrac{1}{36}. A non-double \{i,j\} contains (i,j) and (j,i), so p(\{i,j\})=\tfrac{2}{36}=\tfrac{1}{18}.

These values are nonnegative and

6\cdot\tfrac{1}{36}+15\cdot\tfrac{1}{18}=\tfrac{6}{36}+\tfrac{30}{36}=1,

so by Result 7 they define a probability on Kofi’s 21 outcomes. It is not uniform, and nothing requires it to be: the axioms ask only for nonnegative values that total 1.

The event “a double” is the six outcomes \{i,i\}, with probability 6\cdot\tfrac{1}{36}=\tfrac{1}{6}. The event “contains a 6” is \{6,6\} together with \{i,6\} for i=1,\dots,5, with probability \tfrac{1}{36}+5\cdot\tfrac{2}{36}=\tfrac{11}{36}. Both agree with counting ordered rolls, as they must, since the two sample spaces describe the same dice. The uniform assignment \tfrac{1}{21} on Kofi’s outcomes also satisfies the axioms, but it describes different dice, ones on which a double has probability \tfrac{6}{21}=\tfrac{2}{7}.

Example 2

Entering from the bar

A checker that has been hit waits on the bar and must re-enter in the opponent’s home board, whose points are numbered 1 to 6. It can enter on point k when a die shows k and that point is open. Kofi holds four of his six home points closed. Find the probability that Mei enters on her next roll, and how much it falls if Kofi closes a fifth point.

Solution

Mei fails exactly when both dice show numbers of closed points. Call this event F. Each die has 4 such numbers, so F contains 4\cdot 4=16 of the 36 rolls and P(F)=\tfrac{16}{36}=\tfrac{4}{9}. Entering is F^c, so by the complement rule

P(F^c)=1-\tfrac{4}{9}=\tfrac{5}{9}.

The complement avoids sorting the entering rolls into those with one good die and those with two.

With a fifth closed point the failure event becomes F', with 5\cdot 5=25 rolls, and P(F'^c)=1-\tfrac{25}{36}=\tfrac{11}{36}. Every roll that failed before still fails, so F\subseteq F', and Result 4 gives the drop as the probability of the rolls that fail only now:

P(F'\cap F^c)=P(F')-P(F)=\tfrac{25}{36}-\tfrac{16}{36}=\tfrac{1}{4}.

Check: \tfrac{5}{9}-\tfrac{11}{36}=\tfrac{20}{36}-\tfrac{11}{36}=\tfrac{9}{36}=\tfrac{1}{4}.

Example 3

A six or a large total

In one position Kofi wants a roll that shows at least one 6 or totals 10 or more. Find the probability.

Solution

Let A be “at least one die shows 6” and B be “the total is at least 10”. The complement of A is “neither die shows 6”, with 5\cdot 5=25 rolls, so P(A)=1-\tfrac{25}{36}=\tfrac{11}{36}. The event B is

B=\{(4,6),(6,4),(5,5),(5,6),(6,5),(6,6)\},

so P(B)=\tfrac{6}{36}. Every roll in B except (5,5) contains a 6, so P(A\cap B)=\tfrac{5}{36}. The addition rule gives

P(A\cup B)=\tfrac{11}{36}+\tfrac{6}{36}-\tfrac{5}{36}=\tfrac{12}{36}=\tfrac{1}{3}.

Check: A\cup B is A together with the one roll (5,5), so it has 11+1=12 rolls.

Example 4

Probabilities with nothing to count

Over a long season, let W be the event that Mei wins a game and D the event that the doubling cube is turned at least once in it. Suppose P(W)=0.55, P(D)=0.40 and P(W\cup D)=0.70. Find the probability that Mei wins with the cube turned, that she wins without it, that neither happens, and that exactly one of the two happens. Could a report of P(W\cap D)=0.45 fit these figures?

Solution

No equally likely outcomes are in sight; the axioms and their consequences are all there is. The addition rule rearranges to

\begin{aligned}P(W\cap D)&=P(W)+P(D)-P(W\cup D)\\&=0.55+0.40-0.70=0.25.\end{aligned}

W is the disjoint union of W\cap D and W\cap D^c, so P(W\cap D^c)=0.55-0.25=0.30. “Neither” is W^c\cap D^c=(W\cup D)^c by De Morgan’s law, with probability 1-0.70=0.30. “Exactly one” is W\cup D with W\cap D removed, and W\cap D\subseteq W\cup D, so Result 4 gives 0.70-0.25=0.45. Check: the four pieces of Figure 1 carry 0.30, 0.25, 0.15 and 0.30, which total 1.

A value P(W\cap D)=0.45 is impossible whatever else is true: W\cap D\subseteq D, so Result 4 forces P(W\cap D)\le P(D)=0.40.

Example 5

Rolling until a double

Kofi rolls the pair of dice again and again and stops at the first double. The outcome is the number n of the roll on which it appears, so \Omega=\{1,2,3,\dots\}. Take as the model p(n)=\left(\tfrac{5}{6}\right)^{n-1}\tfrac{1}{6}: the first n-1 rolls are not doubles and the nth is, and rolls do not influence each other, as in Independent events and the multiplication rule. Show that this is a probability, and find the probability that more than 3 rolls are needed and that the number of rolls is even.

Solution

Each p(n)>0, and by the geometric series

\sum_{n=1}^{\infty}\left(\tfrac{5}{6}\right)^{n-1}\tfrac{1}{6}=\tfrac{1}{6}\cdot\frac{1}{1-\frac{5}{6}}=1,

so Result 7 makes p a probability on this countably infinite \Omega (Figure 2).

1 2 3 4 5 6 7 8 9 10 11 12 0.05 0.10 0.15
p(n)
n
Figure 2The probabilities p(n)=\left(\tfrac{5}{6}\right)^{n-1}\tfrac{1}{6} for the roll on which the first double appears. Infinitely many bars, total height 1.

“More than 3 rolls” is the complement of \{1,2,3\}, so

P(n>3)=1-\tfrac{1}{6}\left(1+\tfrac{5}{6}+\tfrac{25}{36}\right)=1-\tfrac{91}{216}=\tfrac{125}{216},

which is \left(\tfrac{5}{6}\right)^3, the probability that the first three rolls all miss. The even outcomes \{2,4,6,\dots\} form an infinite event, and only countable additivity lets its probability be the sum of infinitely many p(n):

\begin{aligned}P(\text{even})&=\sum_{k=1}^{\infty}\left(\tfrac{5}{6}\right)^{2k-1}\tfrac{1}{6}=\tfrac{5}{36}\sum_{j=0}^{\infty}\left(\tfrac{25}{36}\right)^{j}\\&=\tfrac{5}{36}\cdot\tfrac{36}{11}=\tfrac{5}{11}.\end{aligned}

This \Omega has no outcome for “no double ever appears”, and Result 8 shows that nothing is lost. Let B_m be the event that no double appears in the first m rolls. These events shrink as m grows, “never” is their intersection, and P(B_m)=\left(\tfrac{5}{6}\right)^m. So any model of the endless run of rolls that agrees with these values gives “never” the probability \lim_{m\to\infty}\left(\tfrac{5}{6}\right)^m=0.

Example 6

A lower bound for both

Prove that P(A\cap B)\ge P(A)+P(B)-1 for any events A and B. Then: on a given evening Mei is free to play with probability 0.95 and Kofi with probability 0.9. With nothing known about how their evenings are connected, what can be said about the probability that both are free?

Solution

By Result 4, P(A\cup B)\le 1. Substituting the addition rule gives P(A)+P(B)-P(A\cap B)\le 1, which rearranges to P(A\cap B)\ge P(A)+P(B)-1.

For the evening, the bound gives P(\text{both free})\ge 0.95+0.9-1=0.85. Monotonicity bounds it from above: “both free” is contained in “Kofi free”, so P(\text{both free})\le 0.9. The two probabilities alone place P(\text{both free}) in [0.85,0.9] and no closer. One extra assumption would pin it down: if the two evenings are independent, as in Independence of events and variables, it is 0.95\cdot 0.9=0.855.

Boundaries

Additivity is countable, not uncountable. Let a pointer stop at a uniformly random point of [0,1), the model with P([a,b))=b-a; that such a P exists on intervals and the sets built from them is proved in measure theory. Each single point has probability 0 by Result 4, since \{x\}\subseteq[x,x+\varepsilon) for every small \varepsilon>0. The interval is the union of its points, yet 1\ne 0+0+\cdots. There is no contradiction, because the points of an interval are uncountably many and Axiom 3 speaks only about sequences. The same example shows that probability 0 does not mean impossible, and probability 1 does not mean certain.

No uniform model on a countably infinite set. “Pick a positive integer, each equally likely” describes no probability at all (Exercise 17). A finite set carries a uniform model and an interval carries one by length, but \{1,2,3,\dots\} does not.

Not every subset of an interval can be an event. There is no way to give every subset of [0,1) a probability that satisfies the axioms, agrees with length on intervals, and is unchanged when a set is slid along the interval, wrapping around past 1. The counterexample, the Vitali set, uses the axiom of choice and belongs to measure theory; it is stated here without proof. That is why, for intervals and regions, events are a collection closed under complements and countable unions, which contains every set this course meets.

The axioms do not choose the model. They rule out inconsistent figures, such as P(W\cap D)=0.45 beside P(D)=0.40 in Example 4, but they never say whether dice are fair. That comes from symmetry, from data, or from assumptions such as rolls not influencing one another, and a conclusion is only as good as the model it came from.

Where this goes next

Most of probability is choosing \Omega and P in a particular way and then using the results above. When \Omega is finite and every outcome gets the same p(\omega), Result 7 becomes P(A)=|A|/|\Omega|, and Counting models develops the counting that formula needs, including inclusion-exclusion for many events, of which Result 5 is the first case. Fixing an event B with P(B)>0 and setting Q(A)=P(A\cap B)/P(B) gives a function that again satisfies all three axioms; that is conditioning, in Conditioning and Bayes' rule, rigorously (school version Finding a conditional probability). Independence, in Independence of events and variables, is the extra assumption under which probabilities multiply. A random variable is a function X on \Omega, and the function p of Result 7, carried over to the values of X, becomes the probability mass function of Discrete distributions: binomial, geometric, Poisson. Result 8 is what later makes a distribution function F_X(x)=P(X\le x) climb to 1 and be continuous from the right.

Exercises

Sample spaces and models

  1. In the opening roll-off of backgammon, Mei and Kofi each throw one fair die. Write the sample space as ordered pairs (Mei’s number, Kofi’s number). How many outcomes does it have? Find the probabilities of T, “the numbers tie”, M, “Mei’s number is higher”, and K, “Kofi’s number is higher”, and check that they total 1.
  2. A game between Mei and Kofi ends in one of four ways: Mei wins a single game, Mei wins a gammon, Kofi wins a single game, Kofi wins a gammon. Which of these assignments, in that order, are probabilities? (a) 0.35,0.15,0.40,0.10 (b) 0.50,0.20,0.40,-0.10 (c) 0.30,0.10,0.30,0.20. For each valid one, find the probability that Mei wins and the probability of a gammon.
  3. Kofi’s single die is loaded so that P(\{k\})=ck for k=1,\dots,6. Find c, the probability of an even number, and the probability of a 5 or a 6.
  4. At the end of a game the doubling cube shows V\in\{1,2,4,8,16,32,64\}. A model sets P(V=2^k)=c\cdot 2^{-k} for k=0,1,\dots,6. Find c, P(V=1) and P(V\ge 8).

Consequences of the axioms

  1. Events A and B have P(A)=0.6, P(B)=0.5 and P(A\cap B)=0.35. Find P(A\cup B), P(A^c\cap B^c), P(A\cap B^c) and the probability that exactly one of A and B occurs.
  2. For one roll of two fair dice, let A be “the total is 7” and B be “at least one die shows 2”. Find P(A), P(B), P(A\cap B) and P(A\cup B).
  3. Mei has a checker on the bar. Find her probability of entering when Kofi holds two home points closed and when he holds three, and the probability of the rolls that enter against two closed points but not against three.
  4. With no points blocked in between, Kofi hits a checker 3 pips away if a die shows 3, if the two dice total 3, or if he rolls double 1s (four moves of one pip each). Write the hitting event as a union of three disjoint events and find its probability.
  5. Events A and B have P(A)=0.7 and P(B)=0.6, and nothing else is known. Find the smallest and largest possible values of P(A\cap B) and of P(A\cup B), and show that each is attained.

Countably infinite sample spaces

  1. In the roll-off of Exercise 1 a tie is thrown again, so the number N of throws has \Omega=\{1,2,3,\dots\} and p(n)=\left(\tfrac{1}{6}\right)^{n-1}\tfrac{5}{6}. Show that p is a probability, and find P(N\ge 3) and the probability that N is odd.
  2. Let p(n)=\dfrac{c}{n(n+1)} for n=1,2,3,\dots. Find c, P(N\ge 10) and P(N\le 3).
  3. Let p(n)=\dfrac{c}{n^2} for n=1,2,3,\dots. Using \displaystyle \sum_{n=1}^{\infty}\frac{1}{n^2}=\frac{\pi^2}{6}, stated here without proof, find c, P(N=1) and the probability that N is even.

Proofs

  1. Prove inclusion-exclusion for any events A, B, C: the probability P(A\cup B\cup C) equals P(A)+P(B)+P(C), minus P(A\cap B)+P(A\cap C)+P(B\cap C), plus P(A\cap B\cap C).
  2. The symmetric difference A\mathbin{\triangle}B=(A\cap B^c)\cup(A^c\cap B) is the event that exactly one of A and B occurs. Prove that P(A\mathbin{\triangle}B)=P(A)+P(B)-2P(A\cap B).
  3. Prove that if P(A_n)=0 for every n, then P\left(\bigcup_n A_n\right)=0, and that if P(A_n)=1 for every n, then P\left(\bigcap_n A_n\right)=1.
  4. Prove that P(A_1\cap\dots\cap A_n)\ge P(A_1)+\dots+P(A_n)-(n-1) for any events A_1,\dots,A_n.
  5. Prove that no probability on \Omega=\{1,2,3,\dots\} gives every outcome the same probability.
  6. Refute with a counterexample: “If P(A)=0, then A=\varnothing.”
  7. Prove that for any events A_1,A_2,A_3,\dots, the probability of A_1\cup\dots\cup A_n tends to P\left(\bigcup_{i=1}^{\infty}A_i\right) as n grows.
Answers
  1. 36 outcomes. P(T)=\tfrac{6}{36}=\tfrac{1}{6} and P(M)=P(K)=\tfrac{15}{36}=\tfrac{5}{12}: swapping the two numbers matches the 30 untied pairs of M with those of K. The three events are disjoint with union \Omega, and \tfrac{1}{6}+\tfrac{5}{12}+\tfrac{5}{12}=1.
  2. Only (a). Its values are nonnegative and total 1; (b) totals 1 but has a negative value, breaking Axiom 1; (c) totals 0.90, breaking Axiom 2. In (a), P(\text{Mei wins})=0.35+0.15=0.50 and P(\text{gammon})=0.15+0.10=0.25.
  3. c=\tfrac{1}{21}, since c(1+2+\dots+6)=21c=1. P(\text{even})=\tfrac{2+4+6}{21}=\tfrac{4}{7} and P(\{5,6\})=\tfrac{11}{21}.
  4. c=\tfrac{64}{127}, since \sum_{k=0}^{6}2^{-k}=\tfrac{127}{64}. P(V=1)=\tfrac{64}{127} and P(V\ge 8)=\tfrac{64}{127}\left(\tfrac{1}{8}+\tfrac{1}{16}+\tfrac{1}{32}+\tfrac{1}{64}\right)=\tfrac{15}{127}.
  5. P(A\cup B)=0.6+0.5-0.35=0.75; P(A^c\cap B^c)=1-0.75=0.25 by De Morgan’s law and the complement rule; P(A\cap B^c)=0.6-0.35=0.25; exactly one: 0.75-0.35=0.40.
  6. P(A)=\tfrac{6}{36}; P(B)=1-\tfrac{25}{36}=\tfrac{11}{36}; A\cap B=\{(2,5),(5,2)\}, so P(A\cap B)=\tfrac{2}{36}; P(A\cup B)=\tfrac{6+11-2}{36}=\tfrac{5}{12}.
  7. Two closed: 1-\tfrac{4}{36}=\tfrac{8}{9}. Three closed: 1-\tfrac{9}{36}=\tfrac{3}{4}. The failing rolls against two closed points lie inside those against three, so by Result 4 the rolls that enter only against two have probability \tfrac{9}{36}-\tfrac{4}{36}=\tfrac{5}{36}, which equals \tfrac{8}{9}-\tfrac{3}{4}.
  8. \tfrac{14}{36}=\tfrac{7}{18}. The pieces are “a die shows 3” (11 rolls), \{(1,2),(2,1)\} and \{(1,1)\}. None of the last three rolls contains a 3, and (1,1) totals 2, so the pieces are disjoint and Result 2 adds them: \tfrac{11+2+1}{36}.
  9. 0.3\le P(A\cap B)\le 0.6 and 0.7\le P(A\cup B)\le 1. Example 6 gives P(A\cap B)\ge 0.7+0.6-1=0.3 and Result 4 gives P(A\cap B)\le P(B)=0.6; then P(A\cup B)=1.3-P(A\cap B). On \Omega=\{1,\dots,10\} with each outcome \tfrac{1}{10}, A=\{1,\dots,7\} and B=\{5,\dots,10\} give 0.3 and 1, while A=\{1,\dots,7\} and B=\{1,\dots,6\} give 0.6 and 0.7.
  10. Every term is positive and \sum_{n\ge 1}\left(\tfrac{1}{6}\right)^{n-1}\tfrac{5}{6}=\tfrac{5}{6}\cdot\tfrac{1}{1-1/6}=1. P(N\ge 3)=1-\tfrac{5}{6}-\tfrac{5}{36}=\tfrac{1}{36}. P(N\text{ odd})=\tfrac{5}{6}\sum_{j\ge 0}\left(\tfrac{1}{36}\right)^j=\tfrac{5}{6}\cdot\tfrac{36}{35}=\tfrac{6}{7}.
  11. c=1: \tfrac{1}{n(n+1)}=\tfrac{1}{n}-\tfrac{1}{n+1}, so the partial sums telescope to 1-\tfrac{1}{m+1}, which tends to 1. In the same way P(N\ge 10)=\sum_{n\ge 10}\left(\tfrac{1}{n}-\tfrac{1}{n+1}\right)=\tfrac{1}{10}, and P(N\le 3)=1-\tfrac{1}{4}=\tfrac{3}{4}.
  12. c=\tfrac{6}{\pi^2}, so P(N=1)=\tfrac{6}{\pi^2}\approx 0.608. P(N\text{ even})=\tfrac{6}{\pi^2}\sum_{k\ge 1}\tfrac{1}{4k^2}=\tfrac{6}{\pi^2}\cdot\tfrac{\pi^2}{24}=\tfrac{1}{4}.
  13. Result 5 applied to the events A\cup B and C writes P(A\cup B\cup C) as P(A\cup B)+P(C) minus P((A\cup B)\cap C). By the distributive law (A\cup B)\cap C=(A\cap C)\cup(B\cap C), and these two events intersect in A\cap B\cap C, so Result 5 again writes P((A\cup B)\cap C) as P(A\cap C)+P(B\cap C) minus P(A\cap B\cap C). Substituting this and P(A\cup B)=P(A)+P(B)-P(A\cap B) gives the formula.
  14. A\cap B^c and A^c\cap B are disjoint, so P(A\mathbin{\triangle}B)=P(A\cap B^c)+P(A^c\cap B) by Result 2. Since A\cap B\subseteq A and A\cap(A\cap B)^c=A\cap B^c, Result 4 gives P(A\cap B^c)=P(A)-P(A\cap B), and in the same way P(A^c\cap B)=P(B)-P(A\cap B). Adding the two gives the result.
  15. By Axiom 1 and the union bound, 0\le P\left(\bigcup_n A_n\right)\le\sum_n P(A_n)=0. For the second statement, P(A_n^c)=0 for every n by the complement rule, so the first statement gives P\left(\bigcup_n A_n^c\right)=0. By De Morgan’s law this union is \left(\bigcap_n A_n\right)^c, so the complement rule gives P\left(\bigcap_n A_n\right)=1.
  16. Write E=A_1\cap\dots\cap A_n. By De Morgan’s law E^c=A_1^c\cup\dots\cup A_n^c, so the complement rule and the union bound give P(E)=1-P(E^c)\ge 1-\sum_{i=1}^{n}P(A_i^c). Since P(A_i^c)=1-P(A_i) by the complement rule, the right side is \sum_{i=1}^{n}P(A_i)-(n-1). For n=2 this is Example 6.
  17. Suppose P(\{n\})=c for every n; Axiom 1 gives c\ge 0. The events \{n\} are pairwise disjoint with union \Omega, so Axioms 2 and 3 give 1=\sum_{n=1}^{\infty}c. That series sums to 0 when c=0 and diverges when c>0, so no value of c works.
  18. False. On \Omega=\{1,2,3\} let p(1)=p(2)=\tfrac{1}{2} and p(3)=0, a probability by Result 7. The event A=\{3\} has P(A)=0 but is not empty. The pointer under Boundaries gives another: every single point of [0,1) has probability 0.
  19. Let C_n=A_1\cup\dots\cup A_n. Then C_1\subseteq C_2\subseteq\cdots, and \bigcup_n C_n=\bigcup_i A_i, because an outcome in A_i lies in C_i and every C_n lies inside \bigcup_i A_i. Continuity along a growing chain (Result 8) gives P(C_n)\to P\left(\bigcup_i A_i\right).
math.wiki · CC-BY-SA-4.0