Repetition Laplace functional Point Processes Repetition Laplace functional General Poisson process Consider a sample space Ω and anonther space E ⊂ Rd for some d ≥ 1 A sequence of random variables Xn : Ω → E for n ∈ N is defined Let E be the Borel σ-algebra on E Given E ⊂ Rd and a measure µ, the point processes N is PRM(µ) (Poisson random measure (µ)) if N(A) is Poisson distributed with mean µ(A) for all A ∈ E For any A ∈ E, let {An }kn=1 disjoint ⇒ {N(An )}kn=1 independent N(A) = N(ω, A) X Xn (A) = n∈N = X 1(X 1 n ∈ A) n∈N Stochastic processes III, lecture 3 Repetition Stochastic processes III, lecture 3 Laplace functional Repetition Laplace functional The Laplace functional Let {Ij ; 1 ≤ j ≤ k} be a (finite) collection of bounded rectangles in E ⊂ Rd The finite P dimensional distributions of the point process N = n∈N Xn are the collection of multivariate mass functions: P(N(Ij ) = nj ; 1 ≤ j ≤ k}, for k, n1 , n2 · · · nk ∈ N. Proposition 4.7.1: The finite dimensional distributions of the point process N uniquely determine the distribution of N Let B+ , be the non-negative and bounded functions on E . For f ∈ B+ , and a point process N define Z X N(f ) := f (x)dN(x) := f (Xn ) x∈E The Laplace functional of N is defined as h i −N(f ) ΨN (f ) := E e The proof is skipped in this course and follows from basic measure theory Stochastic processes III, lecture 3 n∈N Stochastic processes III, lecture 3 Repetition Laplace functional Repetition Laplace functional The Poisson process If {Ij ; 1 ≤ j ≤ k} is a collection of bounded rectangles in k X E ⊂ Rd and f (x) = λj1(x 1 ∈ Ij ) for λi ≥ 0, then Let µ be a measure on E ⊂ Rd , where E has Borel σ-algebra E Recall: a point process is a Poisson process with mean measure µ (PRM) on E if j=1 h Pk i ΨN (f ) = E e − j=1 λj N(Ij ) , 1 For A ∈ E ( which is the joint Laplace transform of (N(Ij ); 1 ≤ j ≤ k) P(N(A) = k) = This implies ΨN (f ) determines the joint distribution of (N(Ij ); 1 ≤ j ≤ k) 2 Propositions 4.7.1 and 4.7.2 give that the Laplace functional uniquely determines the distribution of N Repetition Laplace functional The distribution of PRM(µ) is uniquely determined by For A ∈ E ( P(N(A) = k) = {An }kn=1 disjoint subsets of E ⇒ {N(An )}kn=1 independent We want to show that the properties above determine a unique point process and we want to identify the form of the Laplace functional Repetition Laplace functional e −µ(A) (µ(A))k k! 0 if µ(A) < ∞, if µ(A) = ∞. First we prove that the form of the Laplace functional follows from the properties of the Poisson process If f (x) = λ1 1(x ∈ A), where λ > 0 then N(f ) = λN(A), and N(A) ∼ Poisson(µ(A)) This gives h i h i ΨN (f ) := E e −N(f ) = E e −λN(A) = {An }kn=1 disjoint subsets of E ⇒ {N(An )}kn=1 independent Furthermore, the point process N is PRM(µ) if and only if its Laplace functional is of the form Z ΨN (f ) = exp[− (1 − e −f (x) )µ(dx)], f ∈ B+ E = ∞ X e −µ(A) (µ(A))k k! k=0 e −λk = exp[−(1 − e −λ )µ(A)] = Z (1 − e −λ )1 1(x ∈ A)µ(dx)] = Z Z −λ1 1(x∈A) = exp[− (1−e )µ(dx)] = exp[− (1−e −f (x) )µ(dx)] = exp[− E E Stochastic processes III, lecture 3 if µ(A) < ∞, if µ(A) = ∞. Proof Theorem (Theorem 4.8.2) 2 0 Stochastic processes III, lecture 3 Stochastic processes III, lecture 3 1 e −µ(A) (µ(A))k k! Stochastic processes III, lecture 3 E Repetition Laplace functional Repetition P Let f (x) = kn=1 λn1(x 1 ∈ An ), where λn ≥ 0 and {An }kn=1 disjoint subsets of E This gives h i h Pk i ΨN (f ) := E e −N(f ) = E e − n=1 λn N(An ) = Z k k h i Y Y −λn N(An ) −λn1(x∈A 1 n) = E e = exp − (1 − e )µ(dx) = n=1 = exp[− For f ∈ B+ , let n n2 X i −1 i i −1 1(f 1 (x) ∈ [ n , n )) + n1 1(f (x) ≥ n) fn (x) = n 2 2 2 i=1 E n=1 Z X k fn (x) % f (x), thus by monotone convergence N(fn ) % N(f ) By 0 < e −f ≤ 1 and dominated convergence we have i h h i lim ΨN (fn ) := lim E e −N(fn ) = E lim e −N(fn ) = ΨN (f ) 1 n) (1 − e −λn1(x∈A )µ(dx) = E n=1 Z = exp[− (1 − e − Pk n=1 Laplace functional λn1(x∈A 1 n) n→∞ )µ(dx)] = n→∞ n→∞ E Z = exp[− (1 − e −f (x) )µ(dx)] E Stochastic processes III, lecture 3 Stochastic processes III, lecture 3 Repetition Laplace functional Laplace functional We still have to prove that Z −f (x) )µ(dx) , ΨN (f ) = exp − (1 − e The previous step yields f ∈ B+ E Z −fn (x) ΨN (fn ) = exp − (1 − e )µ(dx) E By 1 − e −fn % 1 − e −f and monotone convergence, we have Z −f (x) ΨN (f ) = lim ΨN (fn ) = exp − (1 − e )µ(dx) n→∞ Repetition E determines a Poisson process uniquely We know that the Laplace transform uniquely determines a point process (Theorem 4.7.2) f (x) = λ1 1(x ∈ A) ⇒ Z h i −λ1 1(x∈A) ΨN (f ) = exp − (1 − e )µ(dx) = exp −(1 − e −λ )µ(A) , E which is the Laplace transform of a Poisson(µ(A)) distributed random variable Stochastic processes III, lecture 3 Stochastic processes III, lecture 3 Repetition Laplace functional Pk ∈ An ), with {An }kn=1 disjoint, then Z h Pk i P − n=1 λn N(An ) 1 − kn=1 λn1(x∈A n) E e = exp − (1 − e )µ(dx) E " Z k # X −λn1(x∈A 1 ) n = exp − (1 − e )µ(dx) f (x) = 1 n=1 λn1(x Repetition Construction of general Poisson process Theorem Let S be a countable, possibly infinite set. Assume {Ei , i ∈ S} are disjoint elements of E E = ∪i∈S Ei E n=1 = = k Y n=1 k Y Laplace functional µ(Ei ) < ∞ for i ∈ S h i exp −(1 − e −λn )µ(An ) h i E e −λn N(An ) n=1 So, the joint Laplace transform of (N(An ); 1 ≤ n ≤ k) factors into a product of Laplace transforms, which shows independence Consider an independent sequence of random variables (Yi ; i ∈ S), where Yi ∼ Poisson(µ(Ei )). Define a probability measure on Ei : Fi (dx) = µ(dx)/µ(Ei ). Let {Xi,n ; 1 ≤ n ≤ Yi } be i.i.d. random variables on Ei with distribution Fi . {Xi,n ; 1 ≤ n ≤ Yi , i ∈ S} are independent. Yi XX Then, N = Xi,n is PRM(µ). i∈S n=1 Stochastic processes III, lecture 3 Stochastic processes III, lecture 3 Repetition Laplace functional Repetition Laplace functional Proof For E = ∪i∈S Ei we have Step 1: Check it for Ei ΨN (f ) = E[e − P Yi j=1 f (Xi,j ) ] = = = = = ∞ X h E e k=0 ∞ X − Pk j=1 f (Xi,j ) i (µ(E ))k i e −µ(Ei ) k! ik (µ(E ))k i E e e −µ(Ei ) k! k=0 h i exp −µ(Ei )(1 − E[e −f (Xi,1 ) ]) Z −f (x) µ(dx) exp −µ(Ei ) 1 − [e ] µ(Ei ) Ei Z exp − (1 − e −f (x) )µ(dx) h ΨN (f ) = Z −f (x) exp − (1 − e )µ(dx) Ei i∈S " −f (Xi,1 ) = exp − XZ i∈S # (1 − e −f (x) )µ(dx) Ei Z −f (x) = exp − (1 − e )µ(dx) E This completes the construction of a Poisson process on E Ei Stochastic processes III, lecture 3 Y Stochastic processes III, lecture 3