Материал: Крючков Фундаменталс оф Нуцлеар Материалс Пхысицал Протецтион 2011
D(x1 + x2... + xn ) = D(x1) + D(x2 ) + ... + D(xn ) = nσ 2 , |
(4.8) |
||||||||
x1 |
+ x2... + xn |
= |
1 |
|
+ x2... + xn ) = |
σ 2 |
(4.9) |
||
D |
|
|
|
|
D(x1 |
. |
|||
|
n |
|
|||||||
|
|
|
n |
|
n |
|
|||
Distribution function
A random quantity can be considered to be fully characterized if for any value х the distribution function F(x) is known:
F(x) = P(X<x). |
(4.10) |
Unlike the distribution law that unambiguously relates the values р and х, the distribution function gives a random quantity a somewhat different and more general characterization. The probability that a random quantity will assume any value satisfying the inequality x1 ≤ x < x2 , is equal to the increment of its distribution function on this interval:
P(x1 ≤ x < x2 ) = F (x2 ) − F (x1) .
Continuous random quantity
We shall introduce the concept of continuous random quantity.
Random quantity is called continuous if its distribution function is everywhere continuous with its derivative (to the exclusion of the finite number of points on a finite interval).
Probability density of continuous random quantity |
|
Probability density ϕ(x) for a continuous random quantity |
is its |
distribution function derivative: |
|
ϕ(x) = F '(x) . |
(4.11) |
The probability of a continuous random quantity Х to assume any value in the interval (а,b) is equal to a certain integral of its probability density in the limits of а, b:
161
b
P(a < X < b) = ∫ϕ (x)dx = F (b) − F (a),
a
+∞
∫ϕ (x)dx = 1.
−∞
The probability density ϕ (x) for a random quantity Х distribution function F(x) mutually determine each other. Actually,
x
F (x) = P(−∞ < X < x) = ∫ϕ (x)dx .
−∞
(4.12)
and its
(4.13)
The geometrical interpretation of the formulas at hand is fairly obvious. The distribution curve for a continuous random quantity is called its probability density plot.
The mathematical expectation М(X) for a continuous random quantity:
+∞ |
|
M ( X ) = ∫ xϕ(x)dx . |
(4.14) |
−∞ |
|
The variance D(X) of a continuous random quantity: |
|
+∞ |
|
D( X ) = ∫(x − a)2 ϕ(x)dx, где a = M ( X ). |
(4.15) |
−∞
All earlier found mathematical expectation and variance properties are also just in the event of a continuous random quantity. One example is the
property (4.5) for a continuous random quantity:
+∞
D( X ) = ∫ x2ϕ (x)dx − [M (X )]2 .
−∞
Apart from mathematical expectation and variance, other numerical measures are used that reflect various distribution peculiarities. Consider the following major measures of distribution.
Moments
Special numerical measures of a random quantity are moments around zero and moments around mean. Mathematical expectation and variance are known to be moments, and more precisely, mathematical expectation is the first-order moment around zero, and variance is the second-order moment around mean. All basic distributions relate to the moments of the
162
first two orders (these are parameters of these distributions). If, however, one has to give a detailed characterization to a random quantity, moments of higher orders can be used.
By the moment around zero of the order k we shall mean:
ν k = M ( X k ), |
(4.16) |
and by the moment around mean of the order k we shall mean
μk = M [(X − M (X ))k ]. |
(4.17) |
Median
This is used to indicate to the center for the grouping of a random quantity’s values alongside with mathematical expectation. For a continuous random quantity, median is the boundary left and right of which the random quantity values are with a probability 0.5. For a normal distribution: Ме(Х) = М(Х) = а.
Mode
For a continuous random quantity, mode is the point of the probability density function local maximum. For a normal distribution, Мо(Х) = М(Х)
= а.
Quartiles
Quartiles are boundaries that divide the whole of the distribution into four equally probable parts. There are three quartiles in all. It is obvious that the central quartile is the median.
Quantiles
By the quantile of the level q (or q– quantile) they mean such value xq
of a random quantity with which the distribution function thereof assumes the value equal to q:
F (xq ) = P( X < xq ) = q . |
(4.18) |
163
Some quantiles have special names. It is obvious that the above introduced median is the quantile of the level 0.5, that is, Me( X ) = x0,5 .
The concept of quantile has a close link to the concept of percentage
point. |
|
A 100q% point is understood as quantile |
x1−q , i.e. as such value of the |
random quantity Х that makes the following condition just: |
|
P( X ³ x1−q ) = q . |
(4.19) |
Binomial distribution law
The distribution law for the occurrence х of the event А in n independent tests, in each of which this can occur with the constant probability р, obeys to binomial distribution law, the expression of which is Bernoulli’s formula:
P(x = m) = Cnm pmqn−m , m = 0,1, 2,... , n . |
(4.20) |
The mathematical expectation of a binomially distributed random quantity is equal to np, and the variance is equal to npq:
M(X) = np; D(X) = npq.
Typically, binomial distribution is interpreted as:
·the probability of an event to occur exactly т times in п independent tests (Bernoulli case) provided the probability is constant and equal to р;
·(with a test regarded to be sampling) the probability that a replicate random sample of the size п contains exactly т elements of the given type, given the universe of the size N contains pN elements of the given type.
Hypergeometrical distribution law
A generalization of binomial law is hypergeometrical distribution law:
N1 |
N - N1 |
|
|
m n−m |
|
|
|
|
|
|
|
|
|
P( X = m) = m |
n - m |
|
= |
CN1CN − N1 |
, |
(4.21) |
N |
|
n |
||||
|
|
|
|
CN |
|
|
|
n |
|
|
|
|
|
164
where N ³ n , |
N ³ N1 = pN ; |
|
|
|
|
|
|
|
|
|
||
|
nN1 |
|
nN1 |
(N − N1) |
|
n −1 |
|
|
n −1 |
|||
M ( X ) = |
|
= pn; D(X) = |
|
|
1 |
− |
|
|
= np(1 − p) 1 |
− |
|
. |
N 2 |
|
N 2 |
|
|
||||||||
|
|
|
|
|
N −1 |
|
|
N −1 |
||||
A typical interpretation of hypergeometrical distribution law: Р(Х = т) in (4.21) is the probability that a non-replicate random sample of the size п contains т elements of the given type, if this sampling is done from a universe of N elements in which N1 = pN elements of the given type.
The following passage to the limit is just: it is obvious that if |
N → ∞ |
while п and p = N1 / N remain fixed, the geometrical law |
tends to |
binomial. In fact, a non-replicate sample differs little from a replicate sample if the ratio п / N is small. The given approximation is just if n / N < 0,1.
Note that the mathematical expectation for the relative frequency of the event А in п independent tests, in each of which it may occur with the probability р, is equal to this probability and the variance is equal to pq/n.
Thus, M(X/n) = p, D(X/n) = pq/n. So: |
|
||||
|
|
|
|
|
|
σ = |
pq |
. |
(4.22) |
||
|
|||||
|
|
n |
|
||
Therefore, an increase in the number of tests makes the relative frequency of an event decreasingly less dispersed about this probability.
Poisson’s law
If a random quantity Х is capable of assuming only integer nonnegative
values т = 0, 1, 2, 3,… with the probability: |
|
|
P( X = m) = λme−λ |
, |
(4.23) |
m! |
|
|
where λ is the parameter, it is said to be distributed by Poisson law.
This is the law of rare events, the probability of which (р) is small and the number п is large, say, severe accidents, birth of triplets and so on.
165