Introduction to random walks in random and non-random environments
Nadine Guillotin-Plantard
Institut Camille Jordan - University Lyon I
Grenoble – November 2012
Outline
1 Simple Random Walks in Zd Definition
Recurrence - Transience
Asymptotic distribution forn large Asymmetric random walk
2 Random Walks in Random Environments Definition
Recurrence-Transience
Valleys (or traps) - Slowing down Asymptotic distributions forn large
3 Random Walk in Random Scenery
Simple Random Walks inZd Definition
At time 0, a walker starts from the site 0, tosses a coin. If he gets ”Head”, then he goes to the site +1, otherwise to the site -1. (”Tail”)
Simple Random Walks inZd Definition
Simple Random Walks inZd Definition
Natural questions
Does the walker come back to the origin ? Notion of Recurrence - Transience.
Mean position of the walker, fluctuations around this position, large deviations...
Probability that the walker be at site x at timen (Local limit theorem)
Number of distinct sites visited by the walker up to timen (Range) Maximal (or minimal) position of the walker before time n
Number of visits to a fixed sitex. (Local time ) The last time the random walker visits 0 before time n
The number of positive values of the random walk before time n Number of self-intersections up to time n.
Favorite sites of the walker and so on...
Simple Random Walks inZd Definition
Let (Xi)i≥1 be i.i.d. random variables taking values +1 or−1 with equal probability.
{Xi = +1}={ The walker gets ”Head” at timei}.
The position of the walker at time n is given by : S0:= 0
and for any n≥1,
Sn:=
n
X
i=1
Xi (Sn)n≥0 is called simple random walk onZ. From this writing, we can compute
E(Sn) = 0 and Var(Sn) =E(Sn2) =n. Therefore,Sn∼√
n
Simple Random Walks inZd Recurrence - Transience
For n integer,
P(S2n= 0) = Number of paths of length 2n from 0 to 0 Number of paths of length 2n
= C2nn 22n
= (2n)!
4n(n!)2
∼ 1
√πn for n large using Stirling’s formula
n!∼n e
n√ 2πn.
Simple Random Walks inZd Recurrence - Transience
A random walk is said recurrent iff Ph
lim sup
n
{Sn= 0}i
=Ph
Sn= 0 i.o.i
= 1
Otherwise, it is called transient. SinceSn is a Markov chain, we have this useful criterion :
Theorem
(Sn)n is recurrent iff
+∞
X
n=0
P(Sn= 0) = +∞
Since P(Sn= 0)∼C/√
n, the simple random walk onZ isrecurrent.
Simple Random Walks inZd Recurrence - Transience
Simple random walk in Z
2Simple random walk in Z2
Simple Random Walks inZd Recurrence - Transience
Georges P´olya (1887 – 1985)
Simple Random Walks inZd Recurrence - Transience
In higher dimension
Theorem (P´olya (1921) )
There exists some constant C =C(d) s.t. for n large enough P(Sn= 0)∼C n−d/2.
Main tool: Fourier Inversion Formula P(Sn= 0) = 1
(2π)d Z
[−π,π]d
E(eiΘ·Sn)dΘ Use that Sn is a sum of i.i.d. random vectors and for||Θ|| small,
E(eiΘ·X1) = 1−||Θ||2
2d +o(||Θ||2)
Simple Random Walks inZd Recurrence - Transience
Theorem
A simple random walk in Zd is recurrent for d= 1 or 2, but is transient for d ≥3.
Another way to say that :
”All roads lead to Rome except the cosmic paths ! ”
Simple Random Walks inZd Asymptotic distribution fornlarge
Local limit theorem
For n andx integers s.t. n+x is even,
P(Sn=x) = Number of paths of length n from 0 to x Number of paths of length n
= Cn(n+x)/2
2n
∼ r2
π.e−x
2
√2n
n for n large and |x|=o(n2/3) Therefore, for any x∈Rs.t. n+ [x√
n] is even,
P(Sn= [x√ n]) ∼
r2 π.e−x
2
√2
n for n large
Simple Random Walks inZd Asymptotic distribution fornlarge
Let a,b ∈Rwith a<b, P
S2n∈[a√ 2n,b√
2n]
= X
k∈[a√ 2n,b√
2n]
P(S2n=k)
∼ r2
n
X
m∈[a,b]∩√2Z
2n
√1 2πe−m
2 2
→ 1
√2π Z b
a
e−x
2
2 dx =P(X ∈[a,b]) where X is distributed as the Normal distribution N(0,1).
Notation: As n tends to infinity, Sn
√n
−→ NL (0,1).
Simple Random Walks inZd Asymptotic distribution fornlarge
Maximum of the path at time n
Define
Mn := max
k=0..nSk
= max
t∈[0,1]S[nt]
For anyt >0, as n large, S√[nt]n ∼ N(0,t) and Mn
√n = max
t∈[0,1]
S[nt]
√n
Functional of the path from 0 to time n, a convergence in distribution on the space of the c`ad-l`ag paths (φ(t))t∈[0,1] is needed.
Simple Random Walks inZd Asymptotic distribution fornlarge
Simple Random Walks inZd Asymptotic distribution fornlarge
Functional limit theorem
The sequence S
√[nt]
n
t≥0 converges in law to the real Brownian motion (Bt)t≥0, that is a stochastic process satisfying :
B0 := 0
Stationarity of the increments : Bt−Bs ∼Bt−s for s <t Independence of the increments : Bt−Bs independent from Bs
Bt ∼ N(0,t)
The law of the maximum of the Brownian motion is well-known :
t∈[0,1]max Bt∼B1∼ N(0,1)
Simple Random Walks inZd Asymptotic distribution fornlarge
Arcsine distributions
With the same method, we can compute the asymptotic distributions of many functionals of the random walk :
Nn= max{k = 1. . .n ;Sk = 0} the last time the random walker visits 0 before time n
Vn= #{k = 1. . .n ;Sk >0}the number of positive values of the random walk before time n
We have for any x∈(0,1), asn is large, P(Nn≤xn)∼ 2
π arcsin(√ x) and
P(Vn≤xn)∼ 2
πarcsin(√ x).
Simple Random Walks inZd Asymptotic distribution fornlarge
Simple Random Walks inZd Asymmetric random walk
The random walker moves to the right with probability p and to the left with probability q = 1−p.
Same questions as before: Recurrence, Transience, Asymptotic distribution,....
Simple Random Walks inZd Asymmetric random walk
Let (Xi)i≥1 be i.i.d. random variables taking values +1 or−1 with probability p andq = 1−p respectively. The position of the walker at time n is given by :
S0:= 0 and for any n≥1,
Sn:=
n
X
i=1
Xi From this writing, we can compute
E(X1) =p−q 6= 0 The strong law of large numbers gives : as n→+∞,
Sn
n = 1 n
n
X
i=1
Xi →E(X1) =p−q a.s.
The random walk (Sn)n is transient, tends to +∞ (resp. −∞) when p >q (resp. p<q).
Simple Random Walks inZd Asymmetric random walk
Asymptotic distribution for n large
As n→+∞,
Sn−n (p−q)
√n
−→ NL (0, σ2) whereσ2 = 4p(1−p).
Indeed, forn integer andx∈Zs.t. n+x is even, P(Sn=x) = Cn(n+x)/2p(n+x)/2q(n−x)/2
∼ ....
Random Walks in Random Environments Definition
Random Environment : Letωx,x∈Z, be i.i.d. random variables with values in [0,1], uniformly bounded away from 0 and 1.
For a given realization of the environment, we consider the Markov chain (Sn)n which jumps to the sitex+ 1, with probabilityωx and to x−1 with probability 1−ωx, given it is located at x.
They were introduced by A.A. Chernov in 1967 in order to model the replication of DNA.
Random Walks in Random Environments Definition
Quenched law: Denote by Pω the law of the walk (starting from 0) in the environmentω.
Annealed law: IfPdenotes the law of the environment, P =P×Pω
defined as
P(.) = Z
Pω(.)dP(ω) Fundamental remark :
Under P, the random walk isnota Markov chain.
(Under Pω, the random walk is a Markov chain (inhomogeneous in space))
Random Walks in Random Environments Recurrence-Transience
Solomon I
The ratio
ρx := 1−ωx ωx
plays an important role in the study of RWRE.
Theorem (Solomon (1975))
IfE(lnρ0)<0 (resp. >0) then the random walk istransientand
n→+∞lim Sn= +∞ (resp− ∞) P −a.s.
IfE(lnρ0) = 0, then the random walk is recurrentand lim sup
n→+∞
Sn= +∞ and lim inf
n→+∞Sn=−∞ P −a.s.
Random Walks in Random Environments Recurrence-Transience
Solomon II
Theorem (Solomon (1975)) P-almost surely,
n→+∞lim Sn
n =v where
v =
1−E(ρ0)
1+E(ρ0) if E(ρ0)<1
E(1/ρ0)−1
E(1/ρ0)+1 if 1<1/E(ρ−10 )
0 if 1/E(ρ−10 )≤1≤E(ρ0) Comparison with the random walk in Z:
1-|v|<|E(S1)| −→ some slowdown already occurs.
2- The random walk can be transient with zero speed !
Random Walks in Random Environments Valleys (or traps) - Slowing down
Potential:
V(x) =
x
X
k=1
log(ρk) if x≥1
0 if x= 0
−
0
X
k=x+1
log(ρk) if x ≤ −1
(V(x))x is a realrandom walk with mean E[logρ0] and variance E[(logρ0)2].
Remark also that
ωx = e−V(x)
e−V(x−1)+e−V(x) > 1 2 if and only if
V(x−1)>V(x).
When the potential decreases (resp. increases), the random walker tends to go to the right (resp. left).
Random Walks in Random Environments Valleys (or traps) - Slowing down
Random Walks in Random Environments Valleys (or traps) - Slowing down
Random Walks in Random Environments Asymptotic distributions fornlarge
Recurrent case – E [log ρ
0] = 0
Theorem (Sinai (1982), Kesten (1986), Golosov (1986)) Denote
σ2 =E(logρ0)2∈ ]0,+∞[
Then,
σ2
(logn)2Sn−→L b∞
where b∞ is a symmetric random variable with Laplace transform E(e−λ|b∞|) = cosh(√
2λ)−1 λcosh(√
2λ) , λ >0.
Random Walks in Random Environments Asymptotic distributions fornlarge
Transient case – E [log ρ
0] < 0
Under both assumptions :
1- There exists κ >0 s.t. E(ρκ0) = 1 andE(ρκ0(logρ0)+)<∞.
2- The distribution of log(ρ0) is non-lattice.
Theorem (Kesten-Kozlov-Spitzer (1975)) Whenκ <1, (v = 0)
n→+∞lim PSn nκ ≤x
= 1−Lκ,b(x−1/κ) Whenκ∈(1,2),
n→+∞lim P
Sn−nv v1+1/κn1/κ ≤x
= 1−Lκ,b(−x).
Random Walks in Random Environments Asymptotic distributions fornlarge
Transient case – E [log ρ
0] < 0
Lκ,b is a stable distribution with characteristic function ˆLκ,b(t) = exp
−b|t|κ
1−i t
|t|tan(πκ/2)
The value of b for κ∈ (0,2) was determined by Enriquez, Sabot and Zindy (’09)
Theorem (Kesten-Kozlov-Spitzer (1975)) Whenκ >2,
Sn−nv
√n
−→ NL (0, σ2) where σ2>0.
Random Walk in Random Scenery
Riddle
Is this random walk recurrent or transient ?
Random Walk in Random Scenery
Theorem (Campanino – P´etritis (2003))
The random walk (Sn)n is transient for almost every realization of the orientations.
A local limit theorem can even be proved.
Theorem (Castell – Guillotin-Plantard – P`ene – Schapira (AOP, 2011)) For n large,
P[Sn= 0]∼ C n5/4.
Random Walk in Random Scenery
(Sn)n has the same distribution as (Xn,Yn)n where Yn is the ”blue” random walk on Z.
Xn is the random walk (Yn) in random scenery (”H”,”T”) : ξi = 1 (resp. −1) if ”Tail” (resp. ”Head”) at site i ∈Z,
Xn =
n−1
X
k=0
ξYk
Random Walk in Random Scenery
We have
P[Sn= 0] = P[Xn= 0;Yn= 0]
∼ P[Xn= 0|Yn= 0]P[Yn= 0]
We know that
P[Yn= 0]∼ C n1/2 and (not easy !)
P[Xn= 0|Yn= 0]∼ C n3/4