UA MATH563 The central limit theorem of mathematical foundations of probability theory 25 The continuity theorem of characteristic functions of random variables
Continuity Theorem
assumes {μ n}, μ \{\mu_n\},\mu{
μn},μ is the probability measure,{ϕ n}, ϕ \{\phi_n\},\phi{
fn},ϕ is their characteristic function:
- μ n ⇒ μ \ mu_n \ Rightarrow \ mu μn⇒μ, 则ϕ n → ϕ \ phi_n \ to \ phiϕn→ϕ
- ϕ n → ψ \ phi_n \ to \ psi ϕn→ψ andψ \psiψ is continuous at zero, thenψ \psiψ is the characteristic function, ifμ n ⇒ ν \mu_n \Rightarrow \nuis also knownμn⇒ν, 则ψ \ psiψ是ν \ nowCharacteristic function of ν .
Explanation
According to the two conclusions of this theorem, we can get ϕ n → ϕ \phi_n \to \phiϕn→ϕ等价 于μ n ⇒ μ \ mu_n \ Rightarrow \ muμn⇒μ . There is an interesting condition in the second conclusion,ψ \psiψ is continuous at 0. Why is this condition needed? We can use an example to illustrate that if this condition does not hold, the theorem does not hold:
If X n ∼ N (0, n) X_n \sim N(0,n)Xn∼N(0,n ) , thenX n X_nXnThe density function of will become flatter and flatter, consider the characteristic function
ϕ X n (t) = e − n 2 t 2/2 → ψ (t) = {1, t = 0 0, t ≠ 0 \phi_{X_n}( t) = e^{-n^2t^2/2} \to \psi(t) = \begin{cases} 1, t = 0 \\ 0, t \ne 0 \end{cases}ϕXn(t)=e−n2 t2/2→ψ ( t )={
1,t=00,t=0
Obviously ψ \psiψ is not continuous at 0. It is not difficult to verify that
μ n (− ∞, x) = ∫ − ∞ x / n 1 2 π e − s 2/2 ds → 1/2 \mu_n(-\infty,x) = \int_{-\infty)^ {x/\sqrt{n}}\frac{1}{\sqrt{2\pi}}e^{-s^2/2}ds \to 1/2μn(−∞,x]=∫−∞x/n2 π1e−s2/2ds→1/2
So obviously X n X_nXnThe limit distribution of does not exist.
Proving the idea The
first conclusion, if μ n ⇒ μ \mu_n \Rightarrow \muμn⇒μ , according to the monotonic convergence theorem,ϕ n → ϕ \phi_n \to \phiϕn→F ?
The second conclusion: suppose ϕ n → ψ \phi_n \to \psiϕn→ψ andψ \psiψ is at0 00 is continuous, andμ n ⇒ ν \mu_n \Rightarrow \nuμn⇒ν , the proof is divided into the following steps:
- Description {μ n} \{\mu_n\}{ μn} Is tight;
- According to the Bolzano-Weierstrass theorem, {μ n} \{\mu_n\}{ μn} 'S sub-column{ϕ nk} \{\phi_{n_k}\}{ fnk} Weak convergence, according to the first conclusion,ϕ nk → ψ \phi_{n_k} \to \psiϕnk→ψ且ψ \ psiψ is the characteristic function
Here is a proof of Durrett: