Stochastic process
大部分来自wiki
虽然想着英文好理解一些, 但是自己写还是会有好多用错词 的啊(
就当latex练习好了
a mathematical object defined as a family of random variables.
Definitions
Stochastic process
- a collection of random variables indexed by some set.
- numerical values of some system randomly changing over time.
Random Function
a stochastic process can also be interpreted as a random element in a function space.
Random Field
If the random variables are indexed by the Cartesian plane or some higher-dimensional Euclidean space, then the collection of random variables is usually called a random field instead.
Discrete-time & Continuous-time Stochastic Processes
(When interpreted as time, ) The index set has a finite or countable number of elements or not.
State Space
where each random variable takes values from.
Discrete/Integer-valued SP
- state space: integers or natural numbers
Real-valued SP
- state space: real line
N-dimensional Vector Process
- state space: n-d Euclidean space
Notation
probability space
( Ω , F , P ) (\Omega,F,P) (Ω,F,P)
where Ω \Omega Ω is a sample space, F F F is a σ \sigma σ-algebra, P P P is a probability measure.
measurable space
( S , Σ ) (S,\Sigma) (S,Σ)
while S S S is the state space.
stochastic process
{ X ( t ) : t ∈ T } \{X(t):t\in{T}\} {X(t):t∈T}
while X ( t ) X(t) X(t) refer to the random variable with the index t t t
T T T is called the index set or parameter set.
distribution function F t 1 , t 2 , ⋯   , t i ( x 1 , x 2 , ⋯   , x i ) F_{t_1, t_2, \cdots, t_i}(x_1,x_2,\cdots,x_i) Ft1,t2,⋯,ti(x1,x2,⋯,xi)
F t 1 , t 2 , ⋯   , t n ( x 1 , x 2 , ⋯   , x n ) = P { X ( t 1 ) ≤ x 1 , X ( t 2 ) ≤ x 2 , ⋯   , X ( t n ) ≤ x n } F_{t_1,t_2,\cdots,t_n}(x_1,x_2,\cdots,x_n) = P\{X(t_1)\leq x_1,X(t_2)\leq x_2,\cdots,X(t_n)\leq x_n\} Ft1,t2,⋯,tn(x1,x2,⋯,xn)=P{X(t1)≤x1,X(t2)≤x2,⋯,X(tn)≤xn}
If the distribution is independent,
P
{
X
(
t
1
)
≤
x
1
,
X
(
t
2
)
≤
x
2
}
=
P
{
X
(
t
1
)
≤
x
1
}
P
{
X
(
t
2
)
≤
x
2
}
P\{X(t_1)\leq x_1,X(t_2)\leq x_2\}=P\{X(t_1)\leq x_1\}P\{X(t_2)\leq x_2\}
P{X(t1)≤x1,X(t2)≤x2}=P{X(t1)≤x1}P{X(t2)≤x2}
mean function m X ( t ) m_X(t) mX(t)
m X ( t ) = E X ( t ) , t ∈ T m_X(t)=EX(t), t\in T mX(t)=EX(t),t∈T
covariance function B X ( s , t ) B_X(s,t) BX(s,t)
B X ( s , t ) = E [ ( X ( s ) − m X ( s ) ) ( X ( t ) − m X ( t ) ) ] B_X(s,t) = E[(X(s)-m_X(s))(X(t)-m_X(t))] BX(s,t)=E[(X(s)−mX(s))(X(t)−mX(t))]
variance function D X ( t ) = B X ( t , t ) D_X(t)=B_X(t,t) DX(t)=BX(t,t)
D X ( t ) = σ X 2 ( t ) = E [ ( X ( t ) − m X ( t ) ) 2 ] = E X 2 ( t ) − m X ( t ) 2 = E X 2 ( t ) − ( E X ( t ) ) 2 D_X(t)=\sigma^2_X(t) = E[(X(t)-m_X(t))^2] = EX^2(t)-m_X(t)^2 = EX^2(t)-(EX(t))^2 DX(t)=σX2(t)=E[(X(t)−mX(t))2]=EX2(t)−mX(t)2=EX2(t)−(EX(t))2
Correlation coefficient R X ( s , t ) R_X(s, t) RX(s,t)
R X ( s , t ) = E [ X ( s ) X ( t ) ] R_X(s,t)=E[X(s)X(t)] RX(s,t)=E[X(s)X(t)]
while m X ( t ) m_X(t) mX(t) is the mean value of X ( t ) X(t) X(t), D X ( t ) D_X(t) DX(t) is the offset of X ( t ) X(t) X(t) to mean value at time t t t ,
B X ( s , t ) B_X(s,t ) BX(s,t)& R X ( s , t ) R_X(s,t) RX(s,t) represents the relevance of SP { X ( t ) , t ∈ T } \{X(t), t\in T\} {X(t),t∈T} from different time s , t s,t s,t .
variance D ( x ) D(x) D(x)
D X = E X 2 − ( E X ) 2 DX=EX^2-(EX)^2 DX=EX2−(EX)2
integral representation of E [ f ( X ) ] , X ∼ U ( 0 , T ) E[f(X)], X\sim U(0, T) E[f(X)],X∼U(0,T)
if
X
∼
U
(
0
,
T
)
X\sim U(0, T)
X∼U(0,T):
E
[
f
(
X
)
]
=
1
T
∫
0
T
f
(
x
)
d
x
E[f(X)]=\frac{1}{T}\int_{0}^{T}f(x)dx
E[f(X)]=T1∫0Tf(x)dx
representing the mean value of every possible X in (0, T)
Process with Orthogonal Increments
stochastic process { X ( t ) , t ∈ T } \{X(t), t\in T \} {X(t),t∈T} ,
if E X ( t ) = 0 EX(t) =0 EX(t)=0, and t 1 < t 2 ≤ t 3 < t 4 ∈ T : E [ ( X ( t 2 ) − X ( t 1 ) ) ( X ( t 4 ) − X ( t 3 ) ‾ ) ] = 0 t_1 \lt t_2 \leq t_3 \lt t_4 \in T : E[(X(t_2)-X(t_1))\overline{(X(t_4)-X(t_3)})]=0 t1<t2≤t3<t4∈T:E[(X(t2)−X(t1))(X(t4)−X(t3))]=0,
X ( t ) , t ∈ T {X(t), t\in T} X(t),t∈T is a process with orthogonal increments.
Specially, if
T
=
[
a
,
∞
)
T=[a, \infty)
T=[a,∞) and
X
(
a
)
=
0
X(a)=0
X(a)=0,
B
X
(
s
,
t
)
=
R
X
(
s
,
t
)
=
σ
X
2
(
m
i
n
(
s
,
t
)
)
B_X(s,t)=R_X(s,t)=\sigma_X^2(min(s,t))
BX(s,t)=RX(s,t)=σX2(min(s,t))
Normal distribution
N ( μ , σ 2 ) E X = μ , D X = σ 2 N(\mu,\sigma^2) \\EX=\mu, DX=\sigma^2 N(μ,σ2)EX=μ,DX=σ2
Specially, in standard Normal distribution,
μ
=
0
,
σ
2
=
1
\mu=0, \sigma^2=1
μ=0,σ2=1
Poisson process
It can be defined as a counting process, which represents the random number of events up to some time.
P
{
X
(
t
+
s
)
−
X
(
s
)
=
n
}
=
e
−
λ
t
(
λ
t
)
n
n
!
P\{X(t+s)-X(s)=n\}=e^{-\lambda t} \frac{(\lambda t)^n}{n!}
P{X(t+s)−X(s)=n}=e−λtn!(λt)n
- has the natural numbers as its state space and the non-negative numbers as its index set.
let
X
(
t
)
,
t
≥
0
{X(t), t \geq 0}
X(t),t≥0 be a Poisson process, for $t,s \in [0,\infty) $ and
s
≤
t
s \le t
s≤t ,
E
[
X
(
t
)
−
X
(
s
)
]
=
D
[
X
(
t
)
−
X
(
s
)
]
=
λ
(
t
−
s
)
E[X(t)-X(s)]=D[X(t)-X(s)]=\lambda(t-s)
E[X(t)−X(s)]=D[X(t)−X(s)]=λ(t−s)
since
X
(
0
)
=
0
X(0)=0
X(0)=0,
m
X
(
t
)
=
λ
t
σ
x
2
(
t
)
=
λ
t
B
X
(
s
,
t
)
=
λ
s
m_X(t) = \lambda t \\ \sigma^2_x(t)=\lambda t \\ B_X(s,t)=\lambda s
mX(t)=λtσx2(t)=λtBX(s,t)=λs
normally,
B
X
(
s
,
t
)
=
λ
min
(
s
,
t
)
B_X(s,t)=\lambda\min(s,t)
BX(s,t)=λmin(s,t)
Poisson distribution
P ( λ ) P ( X = k ) = λ k k ! e − λ E X = D X = λ P(\lambda) \\ P(X=k)=\frac{\lambda^k}{k!}e^{-\lambda} \\ EX = DX = \lambda P(λ)P(X=k)=k!λke−λEX=DX=λ
Compound Poisson process
if { N ( t ) , t ≥ 0 } \{N(t), t \geq 0 \} {N(t),t≥0} is a Poisson process of λ \lambda λ,
{ Y k , k = 1 , 2 , ⋯   } \{Y_k, k=1,2,\cdots\} {Yk,k=1,2,⋯} is a set of independent and identically distributed random variables,
and is independent to
{
N
(
t
)
,
t
≥
0
}
\{N(t), t \geq 0\}
{N(t),t≥0},
X
(
t
)
=
∑
k
=
1
N
(
t
)
Y
k
,
t
≥
0
,
X(t)=\sum_{k=1}^{N(t)}Y_k,\ \ t\geq0,
X(t)=k=1∑N(t)Yk, t≥0,
{
X
(
t
)
,
t
≥
0
}
\{X(t),t\geq0\}
{X(t),t≥0} is a Compound Poisson process.
E
[
X
(
t
)
]
=
λ
t
E
(
Y
1
)
D
[
X
(
t
)
]
=
λ
t
E
(
Y
1
)
2
E[X(t)]=\lambda t E(Y_1) \\ D[X(t)]=\lambda t E(Y_1)^2
E[X(t)]=λtE(Y1)D[X(t)]=λtE(Y1)2
Markov Chain
probability transition matrix as
P
=
[
p
i
j
]
P = [p_{ij}]
P=[pij]
Two-step transition probability matrix as
P
(
2
)
=
P
P
P^{(2)}=PP
P(2)=PP
State classification of Markov chain
assume state space I = { 1 , 2 , ⋯   , 9 } I=\{1,2,\cdots ,9\} I={1,2,⋯,9} ,
for state 1, the step T T T is the steps it takes to go back from state 1
for set
{
n
:
n
≥
1
,
p
i
i
(
n
)
>
0
}
\{n:n\geq1,p_{ii}^{(n)}\gt0\}
{n:n≥1,pii(n)>0},
d
=
d
(
i
)
=
G
.
C
.
D
{
n
:
p
i
i
(
n
)
>
0
}
d = d(i)=G.C.D\{n:p_{ii}^{(n)}>0\}
d=d(i)=G.C.D{n:pii(n)>0}
d
d
d is the cycle of state
i
i
i,
if d > 1 d\gt1 d>1, state i i i is periodic,
if
d
=
1
d=1
d=1, state
i
i
i is aperiodic.
f
i
j
=
∑
n
=
1
∞
f
i
j
n
f_{ij}=\sum_{n=1}^{\infty}f_{ij}^{n}
fij=n=1∑∞fijn
f
i
j
f_{ij}
fij is the the probability that i can finally reach j,
when
f
i
i
=
1
f_{ii}=1
fii=1 , the state
i
i
i is recurrent. The necessary and sufficient condition is
∑
n
=
0
∞
p
i
i
(
n
)
=
∞
\sum_{n=0}^{\infty}p_{ii}^{(n)}=\infty
n=0∑∞pii(n)=∞
Specially, state
i
i
i is ergodic state if it is aperiodic & recurrent.
stationary distribution
{ π j = ∑ i ∈ I π i p i j , ∑ j ∈ I π j = 1 , π j ≥ 0 , \begin{cases} \pi_j=\sum_{i \in I}\pi_i p_{ij} ,\\\\ \sum_{j \in I}\pi_j =1, \pi_j \geq 0, \end{cases} ⎩⎪⎨⎪⎧πj=∑i∈Iπipij,∑j∈Iπj=1,πj≥0,
the expected time
μ
i
=
1
π
i
\mu_i=\frac{1}{\pi_i}
μi=πi1
The Birth Death process
{
X
(
t
)
,
t
≥
0
}
\{X(t),t\geq 0\}
{X(t),t≥0} is a Birth Death process when
{
p
i
,
i
+
1
(
h
)
=
λ
i
h
+
o
(
h
)
,
λ
i
>
0
,
p
i
,
i
−
1
(
h
)
=
μ
i
h
+
o
(
h
)
,
μ
i
>
0
,
μ
0
=
0
,
p
i
i
(
h
)
=
1
−
λ
i
h
−
μ
i
h
+
o
(
h
)
,
p
i
j
(
h
)
=
o
(
h
)
,
∣
i
−
j
∣
≥
2
,
\begin{cases} p_{i,i+1}(h)=\lambda_ih+o(h),&\lambda_i>0,\\ p_{i,i-1}(h)=\mu_ih+o(h),&\mu_i>0,\mu_0=0,\\ p_{ii}(h)=1-\lambda_ih-\mu_ih+o(h),\\ p_{ij}(h)=o(h), &|i-j|\geq2, \end{cases}
⎩⎪⎪⎪⎨⎪⎪⎪⎧pi,i+1(h)=λih+o(h),pi,i−1(h)=μih+o(h),pii(h)=1−λih−μih+o(h),pij(h)=o(h),λi>0,μi>0,μ0=0,∣i−j∣≥2,
Kolmogorov forward equation
p i j ′ ( t ) = λ j − 1 p i , j − 1 ( t ) − ( λ j + μ j ) p i j ( t ) + μ j + 1 p i , j + 1 ( t ) , i , j ∈ I p'_{ij}(t)=\lambda_{j-1}p_{i,j-1}(t)-(\lambda_j+\mu_j)p_{ij}(t)+\mu_{j+1}p_{i,j+1}(t),\ \ \ \ i,j \in I pij′(t)=λj−1pi,j−1(t)−(λj+μj)pij(t)+μj+1pi,j+1(t), i,j∈I
希望不会挂科吧。。