本文为 I n t r o d u c t i o n Introduction Introduction t o to to P r o b a b i l i t y Probability Probability 的读书笔记
Cumulative Distribution Functions
- The CDF of a random variable X X X is denoted by F X F_X FX and provides the probability P ( X ≤ x ) P(X \leq x) P(X≤x). In particular, for every x x x we have
In what follows, any ambiguous specification of the probabilities of all events of the form { X ≤ x } \{ X \leq x\} { X≤x}, be it through a PMF, PDF, or CDF. will be referred to as the probability law of the random variable X X X.
Properties of a CDF
- F X ( x ) F_X(x) FX(x) tends to 0 as x → − ∞ x\rightarrow-\infty x→−∞, and to 1 as x → ∞ x\rightarrow\infty x→∞.
- If X X X is discrete, then F X ( x ) F_X(x) FX(x) is a piecewise constant function of x x x.
- If X X X is discrete and takes integer values, the PMF and the CDF can be obtained from each other by summing or differencing:
F X ( k ) = ∑ i = − ∞ k p X ( i ) p X ( k ) = P ( X ≤ k ) − P ( X ≤ k − 1 ) = F X ( k ) − F X ( k − 1 ) F_X(k)=\sum_{i=-\infty}^kp_X(i)\\ p_X(k)=P(X\leq k)-P(X\leq k-1)=F_X(k)-F_X(k-1) FX(k)=i=−∞∑kpX(i)pX(k)=P(X≤k)−P(X≤k−1)=FX(k)−FX(k−1)for all integers k k k.
- If X X X is discrete and takes integer values, the PMF and the CDF can be obtained from each other by summing or differencing:
- If X X X is continuous, then F X ( x ) F_X(x) FX(x) is a continuous function of x x x.
- The PDF and the CDF can be obtained from each other by integration or differentiation:
F X ( x ) = ∫ − ∞ x f X ( t ) d t f X ( x ) = d F X ( x ) d x F_X(x)=\int_{-\infty}^xf_X(t)dt\\ f_X(x)=\frac{dF_X(x)}{dx} FX(x)=∫−∞xfX(t)dtfX(x)=dxdFX(x)
- The PDF and the CDF can be obtained from each other by integration or differentiation:
Sometimes, in order to calculate the PMF or PDF of a discrete or continuous random variable. respectively. it is more convenient to first calculate the CDF.
Example 3.6. The Maximum of Several Random Variables.
You are allowed to take a certain test three times. and your final score will be the maximum of the test scores. Thus,
X = m a x { X 1 , X 2 , X 3 } , X = max\{X_1, X_2, X_3\}, X=max{
X1,X2,X3},where X 1 , X 2 , X 3 X_1, X_2, X_3 X1,X2,X3 are the three test scores and X X X is the final score. Assume that your score in each test takes one of the values from 1 to 10 with equal probability 1 / 10 1/10 1/10, independently of the scores in other tests. What is the PMF p X p_X pX of the final score?
SOLUTION
- We have
F X ( k ) = P ( X ≤ k ) = P ( X 1 ≤ k , X 2 ≤ k , X 3 ≤ k ) = P ( X 1 ≤ k ) P ( X 2 ≤ k ) P ( X 3 ≤ k ) = ( k 10 ) 3 \begin{aligned}F_X(k) &= P(X\leq k) \\&= P(X_1\leq k, X_2\leq k, X_3\leq k) \\&= P(X_1\leq k)P(X_2\leq k) P(X_3\leq k) \\&=(\frac{k}{10})^3\end{aligned} FX(k)=P(X≤k)=P(X1≤k,X2≤k,X3≤k)=P(X1≤k)P(X2≤k)P(X3≤k)=(10k)3where the third equality follows from the independence of the events { X 1 ≤ k } \{ X_1\leq k\} { X1≤k}, { X 2 ≤ k } \{X_2\leq k\} { X2≤k}, { X 3 ≤ k } \{X_3\leq k\} { X3≤k}. Thus, the PMF is given by
p X ( k ) = F X ( k ) − F X ( k − 1 ) = ( k 10 ) 3 − ( k − 1 10 ) 3 , k = 1 , 2 , . . . , 10 p_X(k)=F_X(k)-F_X(k-1)=(\frac{k}{10})^3-(\frac{k-1}{10})^3,\ \ \ \ \ k=1,2,...,10 pX(k)=FX(k)−FX(k−1)=(10k)3−(10k−1)3, k=1,2,...,10 - The preceding line of argument can be generalized to any number of random variables X 1 . . . . . X n X_1 ..... X_n X1.....Xn.
F X ( x ) = F X 1 ( x ) . . . F X n ( x ) F_X(x) = F_{X_1(x)}...F_{X_n(x)} FX(x)=FX1(x)...FXn(x)
Problem 7.
Alvin throws darts at a circular target of radius r r r and is equally likely to hit any point in the target. Let X X X be the distance of Alvin’s hit from the center. The target has an inner circle of radius t t t. If X ≤ t X\leq t X≤t, Alvin gets a score of S = 1 / X S = 1/ X S=1/X. Otherwise his score is S = 0 S = 0 S=0. Find the CDF of S S S. Is S S S a continuous random variable?
SOLUTION
- For s < 0 s<0 s<0, F S ( s ) = 0 F_S(s)=0 FS(s)=0
- For 0 ≤ s < 1 / t 0 \leq s < 1/t 0≤s<1/t, we have
F S ( s ) = P ( S ≤ s ) = 1 − P ( X ≤ t ) = 1 − t 2 r 2 F_S(s) = P(S \leq s) = 1-P(X\leq t) = 1-\frac{t^2}{r^2} FS(s)=P(S≤s)=1−P(X≤t)=1−r2t2 - For 1 / t < s 1/t < s 1/t<s, the CDF of S S S is given by
F S ( s ) = P ( S ≤ s ) = P ( X ≤ t ) P ( S ≤ s ∣ X ≤ t ) + P ( X > t ) P ( S ≤ s ∣ X > t ) = t 2 r 2 P ( 1 / X ≤ s ∣ X ≤ t ) + ( 1 − t 2 r 2 ) ⋅ 1 = t 2 r 2 P ( 1 / s ≤ X ≤ t ) P ( X ≤ t ) + ( 1 − t 2 r 2 ) = t 2 r 2 ( 1 − 1 s 2 t 2 ) + ( 1 − t 2 r 2 ) = 1 − 1 s 2 r 2 \begin{aligned}F_S(s) &= P(S \leq s) = P(X \leq t)P(S \leq s |X \leq t) + P(X > t)P(S \leq s |X > t) \\&=\frac{t^2}{r^2}P(1/X\leq s|X\leq t)+(1-\frac{t^2}{r^2})\cdot1 \\&=\frac{t^2}{r^2}\frac{P(1/s\leq X\leq t)}{P(X\leq t)}+(1-\frac{t^2}{r^2}) \\&=\frac{t^2}{r^2}(1-\frac{1}{s^2t^2})+(1-\frac{t^2}{r^2}) \\&=1-\frac{1}{s^2r^2} \end{aligned} FS(s)=P(S≤s)=P(X≤t)P(S≤s∣X≤t)+P(X>t)P(S≤s∣X>t)=r2t2P(1/X≤s∣X≤t)+(1−r2t2)⋅1=r2t2P(X≤t)P(1/s≤X≤t)+(1−r2t2)=r2t2(1−s2t21)+(1−r2t2)=1−s2r21
Problem 8.
Consider two continuous random variables Y Y Y and Z Z Z, and a random variable X X X that is equal to Y Y Y with probability p p p and to Z Z Z with probability 1 − p 1 - p 1−p.
- (a) Show that the PDF of X X X is given by
f X ( x ) = p f Y ( x ) + ( 1 − p ) f Z ( x ) f_X(x) = pf_Y(x) + (1 - p)f_Z(x) fX(x)=pfY(x)+(1−p)fZ(x)
Note that E [ X ] = p E [ Y ] + ( 1 − p ) E [ Z ] E[X]=pE[Y]+(1-p)E[Z] E[X]=pE[Y]+(1−p)E[Z]
- (b) Calculate the CDF of the two-sided exponential random variable that has PDF given by
where λ > 0 \lambda> 0 λ>0 and 0 < p < 1 0 < p < 1 0<p<1
SOLUTION
- (a) By the total probability theorem, we have
F X ( x ) = P ( X ≤ x ) = p P ( Y ≤ x ) + ( 1 − p ) P ( Z ≤ x ) = p F Y ( x ) + ( 1 − p ) F Z ( x ) F_X(x) = P(X \leq x) = pP(Y \leq x) + (1 - p)P(Z \leq x) = pF_Y (x) + (1 - p)F_Z(x) FX(x)=P(X≤x)=pP(Y≤x)+(1−p)P(Z≤x)=pFY(x)+(1−p)FZ(x)By differentiating, we obtain
f X ( x ) = p f Y ( x ) + ( 1 − p ) f Z ( x ) f_X(x) = pf_Y(x) + (1 - p)f_Z(x) fX(x)=pfY(x)+(1−p)fZ(x) - (b) Consider the random variable Y Y Y that has PDF
and the random variable Z Z Z that has PDF
We note that the random variables − Y -Y −Y and Z Z Z are exponential. Using the CDF of the exponential random variable, we see that the CDFs of Y Y Y and Z Z Z are given by
We have f X ( x ) = p f Y ( x ) + ( 1 − p ) f Z ( x ) f_X(x) = pf_Y (x) + (1 - p)f_Z(x) fX(x)=pfY(x)+(1−p)fZ(x), and consequently F X ( x ) = p F Y ( x ) + ( 1 − p ) F Z ( x ) F_X(x) = pF_Y (x) + (1 -p)F_Z(x) FX(x)=pFY(x)+(1−p)FZ(x). It follows that
Problem 10. Simulating a continuous random variable.
A computer has a subroutine that can generate values of a random variable U U U that is uniformly distributed in the interval [ 0 , 1 ] [0, 1] [0,1]. Such a subroutine can be used to generate values of a continuous random variable with given CDF F ( x ) F(x) F(x) as follows. If U U U takes a value u u u, we let the value of X X X be a number x x x that satisfies F ( x ) = u F(x) = u F(x)=u. For simplicity, we assume that the given CDF is strictly increasing over the range S S S of values of interest. where S = { x ∣ 0 < F ( x ) < 1 } S = \{ x|0 <F(x) < 1\} S={
x∣0<F(x)<1}. This condition guarantees that for any u ∈ ( 0 , 1 ) u \in(0, 1) u∈(0,1), there is a unique x x x that satisfies F ( x ) = u F(x) = u F(x)=u.
- (a) Show that the CDF of the random variable X X X thus generated is indeed equal to the given CDF.
(b) How can this procedure be generalized to simulate a discrete integer-valued random variable?
SOLUTION
- (a) By definition, the random variables X X X and U U U satisfy the relation F ( X ) = U F(X) =U F(X)=U. Since F F F is strictly increasing, we have for every x x x, X ≤ x X \leq x X≤x if and only if F ( X ) ≤ F ( x ) F(X)\leq F(x) F(X)≤F(x). Therefore,
P ( X ≤ x ) = P ( F ( X ) ≤ F ( x ) ) = P ( U ≤ F ( x ) ) = F ( x ) P(X\leq x)=P(F(X)\leq F(x))=P(U\leq F(x))=F(x) P(X≤x)=P(F(X)≤F(x))=P(U≤F(x))=F(x) - (b) Let F F F be the desired CDF. To any u ∈ ( 0 , 1 ) u \in(0, 1) u∈(0,1), there corresponds a unique integer x u x_u xu such that F ( x u − 1 ) < u ≤ F ( x u ) F(x_u - 1) < u\leq F(x_u) F(xu−1)<u≤F(xu). This correspondence defines a random variable X X X as a function of the random variable U U U. We then have, for every integer k k k
P ( X = k ) = P ( F ( k − 1 ) < U ≤ F ( k ) ) = F ( k ) − F ( k − 1 ) P(X=k)=P(F(k-1)<U\leq F(k))=F(k)-F(k-1) P(X=k)=P(F(k−1)<U≤F(k))=F(k)−F(k−1)Therefore, the CDF of X X X is equal to F F F, as desired.
The Geometric and Exponential CDFs
- Because the CDF is defined for any type of random variable, it provides a convenient means for exploring the relations between continuous and discrete random variables. A particularly interesting case in point is the relation between geometric and exponential random variables.
- Let X X X be a geometric random variable with parameter p p p; Thus, for k = 1 , 2 , . . . , k =1, 2, ... , k=1,2,..., we have P ( X = k ) = p ( 1 − p ) k − 1 P(X = k) = p(1 -p)^{k-1} P(X=k)=p(1−p)k−1 and the CDF is given by
F g e o ( n ) = ∑ k = 1 n p ( 1 − p ) k − 1 = 1 − ( 1 − p ) n , f o r n = 1 , 2 , . . . F_{geo}(n)=\sum_{k=1}^np(1-p)^{k-1}=1-(1-p)^n,\ \ \ \ \ for\ n=1,2,... Fgeo(n)=k=1∑np(1−p)k−1=1−(1−p)n, for n=1,2,... - Suppose now that X X X is an exponential random variable with parameter λ > 0 \lambda> 0 λ>0. Its CDF is given by
F e x p ( x ) = P ( X ≤ x ) = 0 , f o r x ≤ 0 F_{exp}(x)=P(X\leq x)=0,\ \ \ \ \ for\ x\leq0 Fexp(x)=P(X≤x)=0, for x≤0and
F e x p ( x ) = ∫ 0 x λ e − λ t d t = 1 − e − λ x , f o r x > 0 F_{exp}(x)=\int_0^x\lambda e^{-\lambda t}dt=1-e^{-\lambda x},\ \ \ \ \ for\ x>0 Fexp(x)=∫0xλe−λtdt=1−e−λx, for x>0 - To compare the two CDFs above, let us define δ = − l n ( 1 − p ) / λ \delta = - ln(1 - p)/\lambda δ=−ln(1−p)/λ,so that
e − λ δ = 1 − p e^{-\lambda\delta}=1-p e−λδ=1−pThen, we see that the values of the exponential and the geometric CDFs are equal whenever x = n δ x = n\delta x=nδ, with n = 1 , 2 , . . . . n= 1, 2,.... n=1,2,....
F e x p ( n δ ) = F g e o ( n ) , n = 1 , 2 , . . . F_{exp}(n\delta)=F_{geo}(n),\ \ \ \ \ \ n=1,2,... Fexp(nδ)=Fgeo(n), n=1,2,... and are close to each other for other values of x x x.