+ All Categories
Home > Documents > Asymptotic behaviour of exponential functionals of Lévy...

Asymptotic behaviour of exponential functionals of Lévy...

Date post: 25-May-2020
Category:
Upload: others
View: 1 times
Download: 0 times
Share this document with a friend
40
Transcript
Page 1: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Asymptoti behaviour of exponential fun tionals of

Lévy pro esses with appli ations to random

pro esses in random environment

Charline Smadi, Irstea (with Sandra Palau and Juan Carlos

Pardo, CIMAT)

May 8th, 2017

1 / 40

Page 2: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Motivations

Aim

Study the asymptoti behaviour of E[F (It

(ξ))] as t goes to in�nity,

where F is non in reasing at in�nity and with at least a polynomial

de ay at in�nity, ξ is a Lévy pro ess, and

I

t

(ξ) =

∫t

0

exp(−ξs

)ds.

Motivations

Appli ation to the long time behaviour of bran hing pro esses in

random environment

2 / 40

Page 3: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

GW pro esses in random environment

Lévy pro esses and CSBP

CSBP in random environment

Main result

3 / 40

Page 4: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Galton-Watson pro esses

◮Models for population dynami s in dis rete time and spa e.

◮No intera tion between individuals

Z

n

=

Z

n−1∑

i=1

ξ(n)i

where (ξ(k)j

, (j , k) ∈ N2) are iid with law Q.

4 / 40

Page 5: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

◮Galton-Watson pro esses in iid random environment:

reprodu tion law hanges at ea h generation

5 / 40

Page 6: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Galton-Watson pro esses in random environment

Z

n

=

Z

n−1∑

i=1

ξ(n)i

, where (ξ(n)j

, j ∈ N) are iid with law Qn

.

◮De�ne the random walk (S

n

, n ∈ N) by

X

n

= logm(Qn

) S

n

=

n∑

k=1

X

i

where

m(Qn

) =

∞∑

y=0

yQn

({y})

◮Then Z

n

e

−S

n

is a martingale =⇒ Properties of (Zn

, n ∈ N)determined by its asso iated random walk (S

n

, n ∈ N)

6 / 40

Page 7: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Asymptoti behaviour of Galton-Watson pro esses in random

environment has been well studied

◮First work: Smith and Wilkinson 69, Athreya and Karlin 71

◮Reviews: Birkner, Geiger and Kersting 05, Dyakonova, Vatutin

and Sagitov 11

7 / 40

Page 8: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

GW pro esses in random environment

Lévy pro esses and CSBP

CSBP in random environment

Main result

8 / 40

Page 9: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Lévy pro esses

De�nition

Sto hasti pro ess issued from the origin with iid in rements and

almost sure right ontinuous paths

= CONTINUOUS VERSION OF RANDOM WALKS

9 / 40

Page 10: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Continuous State Bran hing Pro esses (CSBP)

De�nition

◮Non-negative strong Markov pro ess (Y

t

, t ≥ 0) where 0 and

∞ are two absorbing states

◮Bran hing property: if Y

(x)is the pro ess with initial state x ,

Y

(x+y) L= Y

(x) + Y

(y),

where Y

(x)and Y

(y)are independent

(Lamperti 1967)

CSBP ⇔ SCALING LIMITS OF GALTON-WATSON PROCESSES

10 / 40

Page 11: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Properties

The law of Y is ompletely hara terized by its Lapla e transform

Ex

e

−λYt = e

−xu

t

(λ), ∀x > 0, t ≥ 0,

where u is a di�erentiable fun tion in t satisfying

∂ut

(λ)

∂t= −Φ(u

t

(λ)), u

0

(λ) = λ.

Φ: bran hing me hanism. Given by the Lévy-Khin thine formula

Φ(λ) = −aλ+ γ2λ2 +

(0,∞)

(e

−λx − 1+ λx)µ(dx),

where (a, γ) ∈ R2

, µ σ-�nite measure s.t.

∫ (x ∧ x

2

)µ(dx) <∞.

11 / 40

Page 12: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Can be de�ned as the unique strong solution (Fu and Li 10) of the

SDE

Y

t

= Y

0

+

∫t

0

aY

s

dz+

∫t

0

√2γ2Y

s

dB

s

+

∫t

0

∫ ∞

0

∫Y

s

0

zN(ds, dz , du),

where B is a standard Brownian motion, N(ds, dz , du) is a Poisson

random measure with intensity dsµ(dz)du independent of B , and

N is the ompensated measure of N.

12 / 40

Page 13: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Self similar CSBP: a parti ular lass

Φ(λ) = βλβ+1

, λ ≥ 0,

for some β ∈ (−1, 0) ∪ (0, 1] and β is su h that

{ β < 0 if β ∈ (−1, 0), β > 0 if β ∈ (0, 1].

Y is the unique non-negative strong solution of (Fu and Li 2010)

Y

t

= Y

0

+1β=1

∫t

0

√2 βYs

dB

s

+1β 6=1

∫t

0

∫ ∞

0

∫Y

s−

0

zN(ds, dz , du)

B Brownian motion, N Poisson random measure independent of B

with intensity ββ(β + 1)dsdzdu/(Γ(1 − β)z2+β), and N is the

ompensated Poisson measure.

13 / 40

Page 14: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

GW pro esses in random environment

Lévy pro esses and CSBP

CSBP in random environment

Main result

14 / 40

Page 15: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Self similar CSBP in a Lévy environment

◮CSBP in iid random environment? Catastrophes, ontinuous

variations,...?

◮Boeinghof and Hutzenthaler 12, Palau and Pardo 15:

Brownian motion, Bansaye, Pardo and S. 13: ompound

Poissons pro esses

◮A Lévy pro ess somewhere...

◮See Bansaye and Simatos 2015 for general results on s aling

limits of GW pro esses in random environment

◮See Palau and Pardo 2015 and He, Li and Xu 2016 for

existen e and uni ity of strong solutions

15 / 40

Page 16: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Self similar CSBP in a Lévy environment

Unique non-negative strong solution of (Palau and Pardo 2015, He,

Li and Xu 2016)

Z

t

=Z0

+ 1β=1

∫t

0

√2 βZsdBs

+ 1β 6=1

∫t

0

∫ ∞

0

∫Z

s−

0

zN(ds, dz , du) +

∫t

0

Z

s−dSs ,

where S independent Lévy pro ess

S

t

= αt+σWt

+

∫t

0

(0,1)

(ev−1)M(ds, dv)+

∫t

0

R\(0,1)

(ev−1)M(ds, dv),

α ∈ R, σ ≥ 0,W Brownian motion, M Poisson random measure in

R+ × R independent of W with intensity dsπ(dy), and π σ-�nitemeasure su h that∫

R

(1 ∧ v

2)π(dv) <∞ and

(−1,0)|ev − 1|π(dv) <∞.

16 / 40

Page 17: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Random environment

S

t

= αt + σWt

+

∫t

0

(0,1)(ev − 1)M(ds, dv)

+

∫t

0

R\(0,1)(ev − 1)M(ds, dv)

If we de�ne the auxiliary pro ess,

K

t

=

(α− σ2

2

−∫

(0,1)(ev − 1− v)π(dv)

)t

+σWt

+

∫t

0

(0,1)vM(ds, dv) +

∫t

0

R\(0,1)vM(ds, dv),

then Z

t

e

−K

t

martingale =⇒ Will give the long term behaviour of

the pro ess.

17 / 40

Page 18: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Proposition (Bansaye, Pardo and S. 2013, Palau and Pardo

2015, He, Li and Xu 16)

For all z , λ > 0 and t ≥ 0, we have

Ez

[exp

{− λZ

t

e

−K

t

}∣∣∣K]

= exp

{−z(λ−β + β β

∫t

0

e

−βKu

du

)−1/β}.

18 / 40

Page 19: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Proof.

We introdu e M

t

= Z

t

e

−K

t

and take G (s, x) = exp

−xv

t

(s,λ,K).

From It�'s formula, we observe that (G (s,Ms

), s ≤ t) onditionedon K is a martingale i� v

t

(s, λ,K ) satis�es

∂sv

t

(s, λ,K ) = βvβ+1

t

(s, λ,K )e−βKs , v

t

(t, λ,K ) = λ.

v

t

(s, λ,K ) =

(λ−β + β β

∫t

s

e

−βKu

du

)−1/β

As (G (s,Ms

), s ≤ t) onditioned on K is a martingale,

Ez

[exp{−λMt

}|K ] = Ez

[exp{−M0

v

t

(0, λ,K )}|K ] = e

−zv

t

(0,λ,K).

19 / 40

Page 20: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Ez

[exp

{− λZ

t

e

−K

t

}]

λ→ ∞ ⇒ P(Zt

e

−K

t = 0) = P(Zt

= 0)

λ→ 0 ⇒ P(Zt

e

−K

t <∞) = P(Zt

<∞)

20 / 40

Page 21: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Consequen es

Probability of non explosion

Pz

(Z

t

<∞)= 1{β∈(0,1]}+

1{β∈(−1,0)}E

[exp

{−z(β β

∫t

0

e

−βKu

du

)−1/β}]

Probability of non extin tion

Pz

(Z

t

> 0

)= 1{β∈(−1,0)}+

1{β∈(0,1]}E

[1− exp

{−z(β β

∫t

0

e

−βKu

du

)−1/β}]

21 / 40

Page 22: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Logisti population model in a Lévy random environment

Unique strong solution of (Palau and Pardo 2015)

U

t

= U

0

+

∫t

0

U

s

(µ− kU

s

)ds +

∫t

0

U

s−dSs

where µ > 0 is the drift and k > 0 is the ompetition. The pro ess

U satis�es the Markov property and we have

U

t

=U

0

e

K

t

1+ kU

0

∫t

0

e

K

s

ds

, t ≥ 0.

22 / 40

Page 23: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Di�usion in a Lévy random environment

Informal solution to

dX (t) = dβ(t)− 1

2

V

′(X (t))dt, X (0) = 0,

V two-side Lévy pro ess and β Brownian independent of V . More

rigorously, X di�usion whose onditional generator given V is

1

2

e

V (x) d

dx

(e

−V (x) d

dx

).

(Kawazu and Tanaka 1993, Carmona and al. 1997)

P

(max

t≥0

X (t) > x

)= E

[A

A+ B

x

]

where

A =

∫0

−∞e

V (t)dt independent of B

x

=

∫x

0

e

V (t)dt.

23 / 40

Page 24: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Con lusion

◮ P(Zt

<∞) = E

[exp

{−z(β β

∫t

0

e

−βKu

du

)−1/β}]

, β ∈ (−1, 0)

◮ P(Zt

> 0) = E

[1− exp

{−z(β β

∫t

0

e

−βKu

du

)−1/β}]

, β ∈ (0, 1]

◮ E[Ut

] = E

[U

0

e

K

t/

(1+ kU

0

∫t

0

e

K

s

ds

)]

◮ P (max

x≥0

X (x) > t) = E

[A/(A+

∫t

0

e

V (s)ds

)]

= E

[F

[∫t

0

e

−ξs

ds

]],

where F is non in reasing and goes to 0 at in�nity (at least

polynomially)

24 / 40

Page 25: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

GW pro esses in random environment

Lévy pro esses and CSBP

CSBP in random environment

Main result

25 / 40

Page 26: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Lapla e exponent of a Lévy pro ess

◮ ψ(λ) = logE[eλξ1 ]

◮ E[eλξt ] = e

tψ(λ), t ≥ 0

◮Convex fun tion, ψ(0) = 0

◮ ψ′(0) = E[ξ1

] when it exists

Then (see (Kyprianou 2006) for instan e)

◮ ψ′(0) > 0 =⇒ ξt

→ ∞◮ ψ′(0) = 0 =⇒ lim sup ξ

t

= − lim inf ξt

= ∞◮ ψ′(0) < 0 =⇒ ξ

t

→ −∞

26 / 40

Page 27: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

◮Assumption: de�ned on (θ−, θ+), with θ− < 0 < θ+

◮De�nition: τ in (0, θ+) su h that ψ′(τ) = 0

PSfrag repla ements

0 τ

ψ

27 / 40

Page 28: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Conditions on F

Case (A1)

F is non in reasing at in�nity and satis�es

F (x) = A(x + 1)−p

[1+ (1+ x)−ζh(x)

], 0 < p ≤ τ,

where h is a Lips hitz fun tion whi h is bounded and ζ ≥ 1.

Case (A2)

F is an Hölder fun tion with index α > 0, non in reasing at in�nity,

and satis�es

F (x) ≤ A(x + 1)−p , p > τ.

28 / 40

Page 29: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Theorem (Palau, Pardo and S. 2016)

We have the following �ve regimes for the asymptoti behaviour of

EF

(t) = E

[F

(∫t

0

e

−ξs

ds

)]for large t. (moment ondition

θ+ > p)

i) If ψ′(0+) > 0 and F is a positive and ontinuous fun tion

whi h is bounded, then

lim

t→∞EF

(t) = EF

(∞) > 0.

ii) If ψ′(0+) = 0, θ− < 0, F satis�es (A2),then ∃ 1

su h that

lim

t→∞

√tE

F

(t) =

1

> 0.

29 / 40

Page 30: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

PSfrag repla ements

0 p τ

ψ

PSfrag repla ements

0 p = τ

ψ

PSfrag repla ements

0 pτ

ψ

30 / 40

Page 31: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Theorem (Palau, Pardo and S. 2016)

iii) Suppose that ψ′(0+) < 0:

a) If F satis�es (A1) with ψ′(p) < 0, then ∃ 2

su h that,

lim

t→∞e

−tψ(p)EF

(t) =

2

> 0.

b) If F satis�es (A1), ψ′(p) = 0, ψ′′(p) <∞, then ∃ 3

su h that

lim

t→∞

√te

−tψ(p)EF

(t) =

3

> 0.

) If F satis�es (A2), ψ′(p) > 0 and τ + p < θ+, then ∃ 4

su h

that

lim

t→∞t

3/2e

−tψ(τ )EF

(t) =

4

> 0.

31 / 40

Page 32: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Remark

Reminis ent of asymptoti behaviour of GWRE

◮Strong sub riti al:

P(Zn

> 0) ∼

1

(E[Z1

])n

◮Intermediate sub riti al:

P(Zn

> 0) ∼

2

1√n

(E[Z1

])n

◮Weak sub riti al:

P(Zn

> 0) ∼

3

1

n

3/2(E[E[Z

1

|f0

]τ ])n ,

where

E[E[Z1

|f0

]τ ] = inf

0≤θ≤1

E[E[Z1

|f0

]θ].

32 / 40

Page 33: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Remark

The same week as our paper, preprint of Li and Xu on the same

subje t. Their approa h:

◮Flu tuation theory for Lévy pro esses

◮Lévy pro esses onditioned to stay positive

Our approa h

◮Dis retization of the exponential fun tionals

◮Results on asymptoti behaviour of fun tionals of semi-dire t

produ ts of random variables

33 / 40

Page 34: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Proof of the strong sub riti al ase (p < τ)

Change of measure

dP(β)

dP

∣∣∣∣F

t

= e

βξt

−ψ(β)tfor β ∈ (θ−, θ+),

I

t

(ξ) =

∫t

0

e

−ξt−s

ds = e

−ξt

∫t

0

e

ξt

−ξt−s

ds

(d)= e

−ξt

∫t

0

e

ξs

ds = e

−ξt

I

t

(−ξ)

Hen e using the hange of measure with β = p, we have

E[I

t

(ξ)−p

]= E

[e

pξt

I

t

(−ξ)−p

]= e

tψ(p)E(p)[I

t

(−ξ)−p

].

E(p)[−ξ

1

] = −ψ′(p) > 0 ⇒ E(p)[I

t

(−ξ)−p

]→ E

(p)[I∞(−ξ)−p

]> 0

34 / 40

Page 35: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

∣∣F (I

t

(ξ)) − AI

t

(ξ)−p

∣∣ ≤ MI

t

(ξ)−(1+ε)p .

So, it is enough to show

E

[I

t

(ξ)−(1+ε)p]= o(etψ(p)), as t → ∞.

Again, from the hange of measure with β = (1+ ε)p, we dedu e

E

[I

t

(ξ)−(1+ε)p]= E

[e

p(1+ε)ξs

I

t

(−ξ)−(1+ε)p]

= e

tψ(p(1+ε))E((1+ε)p)

[I

t

(−ξ)−(1+ε)p].

For ε small enough

E(p(1+ε))[−ξ

1

] = −ψ′(p(1 + ε)) > 0

⇒ E(p(1+ε))

[I

t

(−ξ)−p(1+ε)]→ E

(p(1+ε))[I∞(−ξ)−p(1+ε)

]> 0

35 / 40

Page 36: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Key idea

Theorem (Le Page and Peigné 1997, Guivar 'h and Liu 2010)

Let (an

, bn

)n≥0

be a R2

+-valued sequen e of i.i.d. r.v. su h that

E[ln a0

] = 0. Assume that b

0

/(1− a

0

) not onstant a.s. and de�ne

A

0

:= 1, A

n

:=

n−1∏

k=0

a

k

and B

n

:=

n−1∑

k=0

A

k

b

k

, for n ≥ 1.

Let η, κ, ϑ > 0 su h that κ < ϑ, and φ and ψ be two positive

ontinuous fun tions on R+ su h that they do not vanish and for a

onstant C > 0 and for every a > 0, b ≥ 0, b

′ ≥ 0, we have

φ(a) ≤ Ca

κ, ψ(b) ≤ C

(1+ b)ϑ, and |ψ(b)−ψ(b′)| ≤ C |b−b′|η.

36 / 40

Page 37: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Key idea

Theorem (Guivar 'h and Liu 2010)

Moreover, assume that

E[a

κ0

]<∞, E

[a

−η0

]<∞, E

[b

η0

]<∞ and E

[a

−η0

b

−ϑ0

]<∞.

Then there exist two positive onstants (φ, ψ) and (ψ) su h that

lim

n→∞n

3/2E

[φ(A

n

)ψ(Bn

)]= (φ, ψ) and lim

n→∞n

1/2E

[ψ(B

n

)]= (ψ).

37 / 40

Page 38: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Dis retization

(q, n) ∈ N2

. For k ≥ 0, we also de�ne

a

k

= e

−(ξ(k+1)/q−ξk/q)and b

k

=

∫ (k+1)/q−k/q

0

e

−(ξu+k/q−ξk/q)

du.

∫ (i+1)/q

i/qe

−ξu

du = e

−ξi/qb

i

=

i−1∏

k=0

a

k

b

i

:= A

i

b

i

,

whi h implies

∫n/q

0

e

−ξs

ds =

n−1∑

i=0

∫ (i+1)/q

i/qe

−ξu

du =

n−1∑

i=0

A

i

b

i

:= B

n

.

38 / 40

Page 39: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

Example: riti al ase (ψ′(0) = 0)

An appli ation of Guivar 'h and Liu gives

√nE

F

(n/q) ∼ (q), as n → ∞.

Let t > 0. Sin e the mapping s 7→ EF

(s) is non in reasing for large

s, we get

√tE

F

(t) ≤√tE

F

(⌊qt⌋/q) =√

t

⌊qt⌋√⌊qt⌋E

F

(⌊qt⌋/q).

√tE

F

(t) ≥√tE

F

((⌊qt⌋+1)/q) =

√t

⌊qt⌋+ 1

√⌊qt⌋+ 1E

F

((⌊qt⌋+1)/q).

Both ∼ (q)/√q for large t.

39 / 40

Page 40: Asymptotic behaviour of exponential functionals of Lévy …math0.bnu.edu.cn/probab/Workshop2017-05/Talks/SmadiC.pdf · GW ro pcesses in random environment Lévy and CSBP Main result

GW pro esses in random environment Lévy pro esses and CSBP CSBP in random environment Main result

THANK YOU FOR YOUR ATTENTION

40 / 40


Recommended