![Page 1: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/1.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 1
Counting processes, stochastic equations, and
asymptotics for stochastic models
• Poisson processes and Watanabe’s theorem
• Counting processes and intensities
• Poisson random measures
• Stochastic integrals
• Stochastic equations for counting processes
• Embeddings in Poisson random measures
• Example: Model with two time-scales
• Example: Nonlinear Hawkes model
• Age dependent birth and death processes
![Page 2: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/2.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 2
Poisson processes
A Poisson process is a model for a series of random observations occur-ring in time.
x x x x x x x x
t
Let Y (t) denote the number of observations by time t. Note that fort < s, Y (s) − Y (t) is the number of observations in the time interval(t, s]. We assume the following:
1) Observations occur one at a time.
2) Numbers of observations in disjoint time intervals are independentrandom variables, i.e., if t0 < t1 < · · · < tm, then Y (tk)− Y (tk−1),k = 1, . . . ,m are independent random variables.
3) The distribution of Y (t+ a)− Y (t) does not depend on t.
![Page 3: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/3.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 3
Characterization of a Poisson process
Theorem 1 Under assumptions 1), 2), and 3), there is a constantλ > 0 such that, for t < s, Y (s) − Y (t) is Poisson distributed withparameter λ(s− t), that is,
PY (s)− Y (t) = k =(λ(s− t))k
k!e−λ(s−t).
If λ = 1, then Y is a unit (or rate one) Poisson process. If Y is a unitPoisson process and Yλ(t) ≡ Y (λt), then Yλ is a Poisson process withparameter λ.
Note that
PYλ(t+ ∆t)− Yλ(t) ≥ 1|FYλt = 1− e−λ∆t ≈ λ∆t
![Page 4: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/4.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 4
Watanabe’s theorem
N is a counting process if N(0) = 0 and N is constant except for jumpsof +1.
Theorem 2 Suppose that N is a counting process, λ > 0, and that Mdefined by
M(t) = N(t)− λtis a martingale. Then N is a Poisson process with intensity λ.
![Page 5: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/5.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 5
Conditional intensities for counting processes
Assume N is adapted to Ft.
λ ≥ 0 is the Ft-conditional intensity if (intuitively)
PN(t+ ∆t) > N(t)|Ft ≈ λ(t)∆t
or (precisely)
M(t) ≡ N(t)−∫ t
0λ(s)ds
is an Ft-local martingale, that is, if τk is the kth jump time of N ,
E[M((t+ s) ∧ τk)|Ft] = M(t ∧ τk)
for all s, t ≥ 0 and all k.
![Page 6: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/6.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 6
Time-change representation
Lemma 3 If N has Ft-intensity λ, then there exists a unit Poissonprocess (may need to enlarge the sample space) such that
N(t) = Y (
∫ t
0λ(s)ds)
![Page 7: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/7.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 7
Modeling with counting processes
Specify λ(t) = γ(t, N), where γ is nonanticipating in the sense thatγ(t, N) = γ(t, N(· ∧ t)).
Martingale problem. Require
N(t)−∫ t
0γ(s,N)ds (1)
to be a local martingale.
Time-change equation. Require
N(t) = Y (
∫ t
0γ(s,N)ds). (2)
These formulations are equivalent in the sense that the solutions havethe same distribution.
![Page 8: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/8.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 8
Systems of counting processes
Lemma 4 (Meyer (1971), Kurtz (1980) ) Assume N = (N1, . . . , Nm)is a vector of counting processes with no common jumps and λk is theFt-intensity for Nk. Then there exist independent unit Poisson pro-cesses Y1, . . . , Ym (may need to enlarge the sample space) such that
Nk(t) = Yk(
∫ t
0λk(s)ds)
Specifying nonanticipating intensities λk(t) = γk(t, N):
Nk(t) = Yk(
∫ t
0γk(s,N)ds)
![Page 9: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/9.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 9
Representing continuous time Markov chains
IfPX(t+ ∆t)−X(t) = l|FX
t ≈ βl(X(t))∆t, l ∈ Zd.
then we can write
Nl(t) = Yl(
∫ t
0βl(X(s))ds),
where the Yl are independent, unit Poisson processes. Consequently,
X(t) = X(0) +∑l
lNl(t)
= X(0) +∑l
lYl(
∫ t
0βl(X(s))ds).
![Page 10: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/10.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 10
Poisson random measures
Let ν be a σ-finite measure on (U, dU), a complete, separable metricspace. Let N (U) denote the collection of counting measures on U .
Definition 5 A Poisson random measure with mean measure ν is arandom counting measure ξ (a N (U)-valued random variable) such that
a) For A ∈ B(U), ξ(A) has a Poisson distribution with expectationν(A)
b) ξ(A) and ξ(B) are independent if A ∩B = ∅.
For f ∈M(U), f ≥ 0, define
ψξ(f) = E[exp−∫U
f(u)ξ(du)] = exp−∫
(1− e−f)dν
(Verify the second equality by approximating f by simple functions.)
![Page 11: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/11.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 11
Space-time Poisson random measures
Let ξ be a Poisson random measure on U × [0,∞) with mean measureν × ` (where ` denotes Lebesgue measure).
ξ(A, t) ≡ ξ(A× [0, t]) is a Poisson process with parameter ν(A).
ξ(A, t) ≡ ξ(A× [0, t])− ν(A)t is a martingale.
Definition 6 ξ is Ft-compatible, if for each A ∈ B(U), ξ(A, ·) isFt-adapted and for all t, s ≥ 0, ξ(A× (t, t+ s]) is independent of Ft.
![Page 12: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/12.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 12
Predictable processes
Definition 7 The σ-algebra PU of Ft-predictable sets in U×[0,∞)×Ω is the smallest σ-algebra containing all sets of the form
A× (t, t+ s]×H
where t, s ≥ 0, A ∈ B(U), H ∈ Ft.
A stochastic process Z taking values in M(U) is Ft-predictable if themapping (u, t, ω) ∈ U × [0,∞)× Ω→ Z(u, t, ω) is PU -measurable.
![Page 13: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/13.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 13
Properties of integrals
Lemma 8 If ξ is Ft-compatible and Z is Ft-predictable and satis-fies
E[
∫U×[0,t]
|Z(u, s)|ν(du)ds] <∞, t > 0,
then∫U×[0,t] Z(u, s)ξ(du× ds) exists almost surely,
M(t) =
∫U×[0,t]
Z(u, s)ξ(du× ds)−∫U×[0,t]
Z(u, s)ν(du)ds
is a martingale, and
E[
∫U×[0,t]
Z(u, s)ξ(du× ds)] = E[
∫U×[0,t]
Z(u, s)ν(du)ds]
![Page 14: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/14.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 14
Centered Poisson random measures
If Z is predictable (e.g., left continuous and adapted) and
E[
∫U×[0,t]
|Z(u, s)|2ν(du)ds] <∞, t > 0,
then
M(t) =
∫U×[0,t]
Z(u, s)ξ(du× ds)
is a square integrable martingale satisfying
[M ]t =
∫U×[0,t]
Z(u, s)2ξ(du× ds)
and
E[M(t)2] = E[[M ]t] = E[
∫U×[0,t]
Z(u, s)2ν(du)ds]
![Page 15: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/15.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 15
Obtaining counting processes from space-time Pois-son processes
Let ξ be a space-time Poisson process on [0,∞)× [0,∞) with Lebesguemean measure, that is for Borel A ⊂ [0,∞) × [0,∞), ξ(A) is Poissondistributed with mean |A|. (It follows that if A1 and A2 are disjoint,then ξ(A1) and ξ(A2) are independent.)
Then the solution of
N(t) = ξ((u, s) : u ≤ γ(s−, N), s ≤ t)
=
∫[0,∞)×[0,t]
1[0,γ(s−,N)](u)ξ(du× ds)
has the same distribution as (1) and (2). This representation goes backat least to Kerstan (1964) and Grigelionis (1971).
The multidimensional version also works.
![Page 16: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/16.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 16
Comparison of stochastic equations
γk nonanticipating, Yk independent unit Poisson processes, ξk inde-pendent Poisson random measures with Lebesgue mean measure on[0,∞)× [0,∞).
Nk(t) = Yk(
∫ t
0γk(s,N)ds) Nk(t) =
∫[0,∞)×[0,t]
1[0,γk(s−,N)](u)ξk(du×ds)
Properties
• N and N have the same distribution.
• Equation for N is more intuitive.
• N adapted to
Ft = σ(ξk(A× [0, s]) : s ≤ t, A ∈ B([0,∞), k = 1, 2, . . .regardless of choice of γ.
![Page 17: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/17.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 17
Couplings
Given intensities γ and γ,
E[|N(t)− N(t)|] = E[|Y (
∫ t
0γ(s,N)ds)− Y (
∫ t
0γ(s, N)ds)|]
= E[|∫ t
0γ(s,N)ds−
∫ t
0γ(s, N)ds|]
or
E[|N(t)− N(t)|] = E[|∫
[0,∞)×[0,t](1[0,γ(s−,N)](u)− 1[0,γ(s−,N)](u))ξ(du× ds)|]
≤ E[
∫ t
0|γ(s,N)− γ(s, N)|ds]
![Page 18: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/18.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 18
Alternative coupling
Note that using the time-change representation, we can write
N(t) = Y1(
∫ t
0γ(s,N) ∧ γ(s, N)ds)
+Y2(
∫ t
0(γ(s,N)− γ(s,N) ∧ γ(s, N))ds)
N(t) = Y1(
∫ t
0γ(s,N) ∧ γ(s, N)ds)
+Y3(
∫ t
0(γ(s,N)− γ(s,N) ∧ γ(s, N))ds)
![Page 19: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/19.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 19
Example: Model with two time-scales
Xn1 (t) = Xn
1 (0) + Y1(nλt)− Y2(n
∫ t
0µXn
1 (s)ds)
Xn2 (t) = Xn
2 (0) + n−1Y3(n
∫ t
0αXn
1 (s)ds)− n−1Y4(n
∫ t
0βXn
2 (s))ds)
Then ∫ t
0Xn
1 (s)ds→ λ
µt
and Xn2 converges to the solution of
x2(t) = x2(0) +αλ
µt−
∫ t
0βx2(s)ds
![Page 20: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/20.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 20
Central limit theorem
Setting V n2 (t) =
√n(Xn
2 (t)− x2(t))
V n2 (t) = V n
2 (0) +1√nY3(n
∫ t
0αXn
1 (s)ds)− 1√nY4(n
∫ t
0βXn
2 (s))ds)
+√n
∫ t
0(αXn
1 (s)− αλ
µ)ds−
∫ t
0βV n
2 (s)ds
= V n2 (0) +
1√nY3(n
∫ t
0αXn
1 (s)ds)− 1√nY4(n
∫ t
0βXn
2 (s))ds)
− 1√n
α
µ(Y1(nλt)− Y2(n
∫ t
0µXn
1 (s)ds))−∫ t
0βV n
2 (s)ds
+Xn
1 (0)−Xn1 (t)√
n
![Page 21: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/21.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 21
Limiting process
V2(t) = V2(0) +W3(αλ
µt)−W4(
∫ t
0βx2(s)ds)−
α
µ(W1(λt)−W2(λt))
−∫ t
0βV2(s)ds
![Page 22: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/22.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 22
Example: Nonlinear Hawkes model Bremaud and Mas-soulie (1996)
Let
γ(s,N) = φ(
∫(−∞,t]
h(t− s)N(ds))
Assume |φ(x)− φ(y)| ≤ α|x− y| and
α
∫ ∞
0|h(s)|ds < 1.
ξ a Poisson random measure on [0,∞)× (−∞,∞) with Lebesgue meanmeasure.
N(r, t] =
∫[0,∞)×(r,t]
1[0,γ(s−,N)](u)ξ(du× ds)
![Page 23: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/23.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 23
Recursion
Let N (0)(r, t] =∫
[0,∞)×(r,t] 1[0,λ](u)ξ(du× ds) and
N (k+1)(r, t] =
∫[0,∞)×(r,t]
1[0,γ(s−,N (k))](u)ξ(du× ds)
Then N (k) has stationary increments.
N (k+1)4N (k)(r, t] =
∫[0,∞)×(r,t]
|1[0,γ(s−,N (k))](u)−1[0,γ(s−,N (k−1))](u)|ξ(du×ds)
and setting E[N (k+1)4N (k)(r, t]] = ρ(k)(t− r)
ρ(k) = E[
∫ 1
0|γ(r,N (k))− γ(r,N (k−1))|dr]
≤ αE[|∫ 0
−∞h(−s)N (k)(ds)−
∫ 0
−∞h(−s)N (k−1)(ds)|]
≤ αρ(k−1)∫ ∞
0|h(s)|ds
![Page 24: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/24.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 24
Birth and death processes
Let ν ∈ P([0,∞)), and let ξ be a space-time Poisson random measureon [0,∞)× [0,∞) with mean measure ν × `. Let X be given by
X(t) = ξ(B(t))− ξ(D(t))
where
B(t) = (y, u) : u ≤∫ t
0λ(X(s))ds
D(t) = (y, u) : y ≤ t, u ≤∫ t−y
0λ(X(s))ds
![Page 25: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/25.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 25
Law of large numbers for Poisson random measures
Lemma 9 For n = 1, 2, . . ., let ξn be a Poisson random measure withmean measure nν×`. Then for A ∈ B([0,∞)× [0,∞)) with ν×`(A) <∞,
n−1ξn(A)→ ν × `(A)
Definition 10 A is a lower layer if (y1, u1) ∈ A and y2 ≤ y1, u2 ≤ u1
implies (y2, u2) ∈ A.
For α > 0, let Lα be the collection of lower layers contained in [0,∞)×[0, α].
Theorem 11 Stute (1976) For each α > 0,
supA∈Lα
|n−1ξn(A)− ν × `(A)| → 0.
![Page 26: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/26.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 26
A fluid limit
LetXn(t) = n−1ξn(Bn(t))− n−1ξn(Dn(t))
where
Bn(t) = (y, u) : u ≤∫ t
0λ(Xn(s))ds
Dn(t) = (y, u) : y ≤ t, u ≤∫ t−y
0λ(Xn(s))ds
so
Xn(t) =
∫[0,∞)×[0,∞)
(1Bn(t)(y, u)− 1Dn(t)(y, u))ξn(dy × du)
For each t, Bn(t) and Dn(t) are predictable.
ξn(Bn(t)) is a counting process with intensity nλ(Xn(s)) and
ν × `(Bn(t)−Dn(t)) =
∫ t
0λ(Xn(s))ν(t− s,∞)ds
![Page 27: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/27.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 27
Theorem 12 Let ταn = inft :∫ t
0 λ(Xn(s))ds ≥ α. Then
supt≤ταn|Xn(t)−
∫ t
0λ(Xn(s))ν(t− s,∞)ds| → 0
and if λ is Lipschitz, for each T > 0, supt≤T |Xn(t) − x(t)| → 0 wherex is the unique solution of
x(t) =
∫ t
0λ(x(s))ν(t− s,∞)ds
Wang (1975, 1977); Kurtz (1983); Foley (1986); Solomon (1987); Frickerand Jaıbi (2007); Decreusefond and Moyal (2008)
![Page 28: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/28.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 28
Gaussian approximation
For each A ∈ B([0,∞)× [0,∞)),
ξn(A)− nν × `(A)√n
⇒ W (A),
where W is space-time Gaussian white noise with E[W (A1)W (A2)] =ν × `(A1 × A2), but the convergence is not uniform over lower layers.
![Page 29: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/29.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 29
Convergence of stochastic integrals
For ϕ ∈ L2(ν), define
Wn(ϕ, r) =1√n
∫[0,∞)×[0,r]
ϕ(y)ξn(dy × du)
and observe that for ϕi ⊂ L2(ν)
(Wn(ϕ1, ·), . . . ,Wn(ϕm, ·))⇒ (W (ϕ1, ·), . . . ,W (ϕm, ·)).Assume that ξn is Fn
t -compatible.
Theorem 13 Suppose Zn is a cadlag, L2(ν)-valued, Fnt -adapted pro-
cess and that for each ϕk ⊂ L2(ν),
(Zn,Wn(ϕ1, ·), . . . ,Wn(ϕm, ·))⇒ (Z,W (ϕ1, ·), . . . ,W (ϕm, ·))in DL2(ν)×Rm[0,∞). Then∫
U×[0,·]Zn(y, u−)Wn(dy × du)⇒
∫U×[0,·]
Z(y, u−)W (dy × du).
![Page 30: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/30.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 30
Proof. See Kurtz and Protter (1996).
The result also holds for Rd-valued Zn, and under additional conditions,for function-valued Zn.
![Page 31: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/31.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 31
Central limit theorem for Xn
Taking Zn(y, u−) = 1Bn(t)(y, u)− 1Dn(t)(y, u) and u0 >>∫ t
0 λ(x(s))ds
Wn(Bn(t))−Wn(Dn(t)) =
∫[0,∞)×[0,u0]
Zn(y, u−)Wn(dy × du)
converges to∫[0,∞)×[0,u0]
(1B(t)(y, u)− 1D(t)(y, u))W (dy × du) = W (B(t))−W (D(t))
and more generally, the finite dimensional distributions converge.
![Page 32: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/32.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 32
Limiting Gaussian process
Verifying tightness and setting Vn(t) =√n(Xn(t)− x(t)),
Vn(t) = Wn(Bn(t))−Wn(Dn(t))
+√n
∫ t
0(λ(Xn(s))− λ(x(s)))ν(t− s,∞)ds
converges to the solution of
V (t) = W (B(t))−W (A(t)) +
∫ t
0λ′(x(s)))V (s)ν(t− s,∞)ds
![Page 33: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/33.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 33
Some related work
Kurtz (1980) treats infinite systems of time-change equations.
Kurtz and Protter (1996) treats infinite systems of Poisson embeddingequations.
Stochastic equations for spatial birth and birth and death processes arediscussed in Massoulie (1998); Garcia and Kurtz (2006, 2008).
![Page 34: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/34.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 34
References
Pierre Bremaud and Laurent Massoulie. Stability of nonlinear Hawkes processes. Ann. Probab., 24(3):1563–1588, 1996. ISSN 0091-1798. doi: 10.1214/aop/1065725193. URL http://dx.doi.org.ezproxy.
library.wisc.edu/10.1214/aop/1065725193.
Laurent Decreusefond and Pascal Moyal. A functional central limit theorem for the M/GI/∞ queue.Ann. Appl. Probab., 18(6):2156–2178, 2008. ISSN 1050-5164. doi: 10.1214/08-AAP518. URL http:
//dx.doi.org.ezproxy.library.wisc.edu/10.1214/08-AAP518.
Robert D. Foley. Stationary Poisson departure processes from nonstationary queues. J. Appl. Probab., 23(1):256–260, 1986. ISSN 0021-9002.
Christine Fricker and M. Raouf Jaıbi. On the fluid limit of the M/G/∞ queue. Queueing Syst., 56(3-4):255–265, 2007. ISSN 0257-0130. doi: 10.1007/s11134-007-9041-x. URL http://dx.doi.org.ezproxy.
library.wisc.edu/10.1007/s11134-007-9041-x.
Nancy L. Garcia and Thomas G. Kurtz. Spatial birth and death processes as solutions of stochastic equations.ALEA Lat. Am. J. Probab. Math. Stat., 1:281–303, 2006. ISSN 1980-0436.
Nancy L. Garcia and Thomas G. Kurtz. Spatial point processes and the projection method. In Inand out of equilibrium. 2, volume 60 of Progr. Probab., pages 271–298. Birkhauser, Basel, 2008.doi: 10.1007/978-3-7643-8786-0 13. URL http://dx.doi.org.ezproxy.library.wisc.edu/10.1007/
978-3-7643-8786-0_13.
B. Grigelionis. The representation of integer-valued random measures as stochastic integrals over the Poissonmeasure. Litovsk. Mat. Sb., 11:93–108, 1971. ISSN 0132-2818.
![Page 35: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/35.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 35
Johannes Kerstan. Teilprozesse Poissonscher Prozesse. In Trans. Third Prague Conf. Information Theory,Statist. Decision Functions, Random Processes (Liblice, 1962), pages 377–403. Publ. House Czech. Acad.Sci., Prague, 1964.
Thomas G. Kurtz. Representations of Markov processes as multiparameter time changes. Ann. Probab., 8(4):682–715, 1980. ISSN 0091-1798.
Thomas G. Kurtz. Gaussian approximations for Markov chains and counting processes. In Proceedings of the44th session of the International Statistical Institute, Vol. 1 (Madrid, 1983), volume 50, pages 361–376,1983. With a discussion in Vol. 3, pp. 237–248.
Thomas G. Kurtz and Philip E. Protter. Weak convergence of stochastic integrals and differential equations.II. Infinite-dimensional case. In Probabilistic models for nonlinear partial differential equations (Monteca-tini Terme, 1995), volume 1627 of Lecture Notes in Math., pages 197–285. Springer, Berlin, 1996.
Laurent Massoulie. Stability results for a general class of interacting point processes dynamics, and applica-tions. Stochastic Process. Appl., 75(1):1–30, 1998. ISSN 0304-4149. doi: 10.1016/S0304-4149(98)00006-4.URL http://dx.doi.org.ezproxy.library.wisc.edu/10.1016/S0304-4149(98)00006-4.
P. A. Meyer. Demonstration simplifiee d’un theoreme de Knight. In Seminaire de Probabilites, V (Univ.Strasbourg, annee universitaire 1969–1970), pages 191–195. Lecture Notes in Math., Vol. 191. Springer,Berlin, 1971.
Wiremu Solomon. Representation and approximation of large population age distributions using Poissonrandom measures. Stochastic Process. Appl., 26(2):237–255, 1987. ISSN 0304-4149.
Winfried Stute. On a generalization of the Glivenko-Cantelli theorem. Z. Wahrscheinlichkeitstheorie undVerw. Gebiete, 35(2):167–175, 1976.
![Page 36: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/36.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 36
F. J. S. Wang. Limit theorems for age and density dependent stochastic population models. J. Math. Biol.,2(4):373–400, 1975. ISSN 0303-6812.
Frank J. S. Wang. A central limit theorem for age- and density-dependent population processes. StochasticProcesses Appl., 5(2):173–193, 1977. ISSN 0304-4149.
![Page 37: Counting processes, stochastic equations, and asymptotics ...kurtz/Lectures/edin10.pdfFirst Prev Next Go To Go Back Full Screen Close Quit 1 Counting processes, stochastic equations,](https://reader035.vdocuments.net/reader035/viewer/2022071516/6138f200a4cdb41a985b62c9/html5/thumbnails/37.jpg)
•First •Prev •Next •Go To •Go Back •Full Screen •Close •Quit 37
AbstractCounting processes, stochastic equations, and asymptotics for stochastic models
Many stochastic models can be represented as solutions of stochastic equations for which Poisson processesor Poisson random measures are the primary inputs. A variety of examples will be given illustrating theseequations and how they can be used to study limit theorems for stochastic models.