Limit theorems for conditioned non-generic Galton-Watson trees
Igor Kortchemski (Université Paris-Sud, Orsay) PIMS 2012
Igor Kortchemski (Université Paris-Sud, Orsay) Condensation in Galton-Watson trees
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
Two approaches: - Scaling limits
- Local limits
What happens when µis not critical?
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches: - Scaling limits
What happens when µis not critical?
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches: - Scaling limits
What happens when µis not critical?
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches: - Scaling limits
What happens when µis not critical?
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches:
- Scaling limits
State of the art Non-generic trees Limit theorems Extensions
What is this about?
Goal: understand the structure of largeconditionedGalton-Watsontrees.
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches:
- Scaling limits - Local limits
Typical framework:
- the offspring distribution µis critical (P
i>0iµ(i)=1).
- µhas finite variance.
- one studies GWµtreesconditionedon having a fixed (large) number of vertices (or edges).
Two approaches:
- Scaling limits - Local limits
What happens when µis not critical?
I. State of the art (critical case) II. Non-generic trees
III. Limit theorems for non-generic trees IV. One conjecture and one problem
I. State of the art
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root. 2. for everyj>1withρ(j)>0, underPρ(·|k∅=j), the number of children
of thejchildren ofthe rootare independent with distributionρ. Here,k∅=2.
Here,ζ(τ) =5.
Letζ(τ)denote the total number of vertices ofτ.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1.
Here,k∅=2. Here,ζ(τ) =5.
Letζ(τ)denote the total number of vertices ofτ.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
Here,k∅=2. Here,ζ(τ) =5.
Letζ(τ)denote the total number of vertices ofτ.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root.
Letζ(τ)denote the total number of vertices ofτ.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root.
2. for everyj>1withρ(j)>0, underPρ(·|k∅=j), the number of children of thejchildren ofthe rootare independent with distributionρ.
Letζ(τ)denote the total number of vertices ofτ.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root.
2. for everyj>1withρ(j)>0, underPρ(·|k∅=j), the number of children of thejchildren ofthe rootare independent with distributionρ.
Here,k∅=2.
State of the art Non-generic trees Limit theorems Extensions
Recap on Galton-Watson trees
Trees will be planar and rooted.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root.
2. for everyj>1withρ(j)>0, underPρ(·|k∅=j), the number of children of thejchildren ofthe rootare independent with distributionρ.
Here,k∅=2.
Letζ(τ)denote the total number of vertices ofτ.
Letρ be a probability measure onN={0, 1, 2, . . .}withP
iiρ(i)61and ρ(1)<1. AGalton-Watson tree with offspring distributionρis a random tree such that:
1. k∅ has distributionρ, wherek∅ is the number of children ofthe root.
2. for everyj>1withρ(j)>0, underPρ(·|k∅=j), the number of children of thejchildren ofthe rootare independent with distributionρ.
Here,k∅=2.
Here,ζ(τ) =5.
Letζ(τ)denote the total number of vertices ofτ.
Scaling limits
Order the vertices in thethe lexicographical order: k∅=u(0)< u(1)<· · ·< u(ζ(τ) −1).
Letku be the number of children of the vertexu.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
20 21
22 23
24 25
Order the vertices in thethe lexicographical order: k∅=u(0)< u(1)<· · ·< u(ζ(τ) −1).
Letku be the number of children of the vertexu.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
Order the vertices in thethe lexicographical order: k∅=u(0)< u(1)<· · ·< u(ζ(τ) −1).
Letku be the number of children of the vertexu.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4 5 6
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12
13 20
21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4 5 6
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12
13 20
21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4 5 6
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4 5 6
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4
Proposition
The Lukasiewicz path of a GWµtree has the same distribution as a random walkwith jump distributionν(k) =µ(k+1),k>−1, started from0, stopped when it hits−1.
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
Definition (of the contour function)
A capybara explores the tree at unit speed. For 06t62(ζ(τ) −1),Ct(τ)is the distance between the beast at timetand the root.
Definition (of the contour function)
A capybara explores the tree at unit speed. For 06t62(ζ(τ) −1),Ct(τ)is the distance between the beast at timetand the root.
Definition (of the contour function)
A capybara explores the tree at unit speed. For 06t62(ζ(τ) −1),Ct(τ)is the distance between the beast at timetand the root.
0 10 20 30 40 50 0
1 2 3 4 5 6
Definition (of the contour function)
A capybara explores the tree at unit speed. For 06t62(ζ(τ) −1),Ct(τ)is the distance between the beast at timetand the root.
0 5 10 15 20 25 30 0
1 2 3 4 5 6
0 10 20 30 40 50
0 1 2 3 4 5 6 7
Figure: The Lukasiewicz path and the contour function.
I The Lukasiewicz path behaves like arandom walk.
State of the art Non-generic trees Limit theorems Extensions
Scaling limits
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
State of the art Non-generic trees Limit theorems Extensions
Scaling limits
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
State of the art Non-generic trees Limit theorems Extensions
Scaling limits
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
Theorem (Aldous ’93, Duquesne ’04)
Letσ2 be the variance ofµ. Then :
1
√nW[nt](tn), 1 2√
nC2nt(tn)
06t61
−→(d) n→∞
σ·e(t),1 σe(t)
06t61
,
where eis the normalized Brownian excursion.
State of the art Non-generic trees Limit theorems Extensions
Scaling limits
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
Theorem (Aldous ’93, Duquesne ’04)
Letσ2 be the variance ofµ. Then :
1
√nW[nt](tn), 1 2√
nC2nt(tn)
06t61
−→(d) n→∞
σ·e(t),1 σe(t)
06t61
,
where eis the normalized Brownian excursion.
0.2 0.4 0.6 0.8 1
0.
State of the art Non-generic trees Limit theorems Extensions
Scaling limits
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
Theorem (Aldous ’93, Duquesne ’04)
Letσ2 be the variance ofµ. Then :
1
√nW[nt](tn), 1 2√
nC2nt(tn)
06t61
−→(d) n→∞
σ·e(t),1 σe(t)
06t61
,
where eis the normalized Brownian excursion.
Remark:
I Duquesne ’04: extension to the case whereµ is in the domain of attraction of a stable law.
Letµ be a critical offspring distribution with finite variance. Lettn be a Pµ[·|ζ(τ) =n]tree. What doestn look like fornlarge ?
Theorem (Aldous ’93, Duquesne ’04)
Letσ2 be the variance ofµ. Then :
1
√nW[nt](tn), 1 2√
nC2nt(tn)
06t61
−→(d) n→∞
σ·e(t),1 σe(t)
06t61
,
where eis the normalized Brownian excursion.
Consequences:
- limit theorem for the height oftn,
Theorem (Aldous ’93, Duquesne ’04)
Letσ2 be the variance ofµ. Then :
1
√nW[nt](tn), 1 2√
nC2nt(tn)
06t61
−→(d) n→∞
σ·e(t),1 σe(t)
06t61
,
where eis the normalized Brownian excursion.
Consequences:
- limit theorem for the height oftn,
- convergence in the Gromov-Hausdorff sense oftn, suitably rescaled, towards the Brownian CRT.
II. Non-generic trees
II. 1) Exponential families
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Letµ be an offspring distribution with0<µ(0)<1.
Lemma (Kennedy ’75)
Letλ >0 be such that
Zλ=X
i>0
µ(i)λi<∞.
Then a GWµ treeconditionedon having nvertices has the same distribution as a GWµ(λ) treeconditionedon havingnvertices.
Consequence:
I if there existsλ >0 such thatZλ<∞andµ(λ)is critical, then we are back to the critical case.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Letµ be an offspring distribution with0<µ(0)<1.
Lemma (Kennedy ’75)
Letλ >0 be such that
Zλ=X
i>0
µ(i)λi<∞.
Consequence:
I if there existsλ >0 such thatZλ<∞andµ(λ) is critical, then we are back to the critical case.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Letµ be an offspring distribution with0<µ(0)<1.
Lemma (Kennedy ’75)
Letλ >0 be such that
Zλ=X
i>0
µ(i)λi<∞. Set
µ(λ)(i)= 1
Zλµ(i)λi, i>0.
Consequence:
I if there existsλ >0 such thatZλ<∞andµ(λ) is critical, then we are back to the critical case.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Letµ be an offspring distribution with0<µ(0)<1.
Lemma (Kennedy ’75)
Letλ >0 be such that
Zλ=X
i>0
µ(i)λi<∞. Set
µ(λ)(i)= 1
Zλµ(i)λi, i>0.
Then a GWµtreeconditionedon having nvertices has the same distribution as a GWµ(λ) treeconditionedon havingnvertices.
Letµ be an offspring distribution with0<µ(0)<1.
Lemma (Kennedy ’75)
Letλ >0 be such that
Zλ=X
i>0
µ(i)λi<∞. Set
µ(λ)(i)= 1
Zλµ(i)λi, i>0.
Then a GWµtreeconditionedon having nvertices has the same distribution as a GWµ(λ) treeconditionedon havingnvertices.
Consequence:
I if there existsλ >0 such thatZλ<∞andµ(λ) is critical, then we are back to the critical case.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Definition
We say that µisnon-genericif there exist noλ >0such that Zλ<∞and µ(λ)is critical.
Example:
Example: µ(i)∼c/iβ withc >0andβ >2.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Definition
We say that µisnon-genericif there exist noλ >0such that Zλ<∞and µ(λ)is critical.
Example:
- µis subcritical (P
iiµ(i)<1)
Example: µ(i)∼c/iβ withc >0andβ >2.
State of the art Non-generic trees Limit theorems Extensions
Exponential families
Definition
We say that µisnon-genericif there exist noλ >0such that Zλ<∞and µ(λ)is critical.
Example:
- µis subcritical (P
iiµ(i)<1) - and the radius of convergence ofX
i>0
µ(i)zi is1.
Definition
We say that µisnon-genericif there exist noλ >0such that Zλ<∞and µ(λ)is critical.
Example:
- µis subcritical (P
iiµ(i)<1) - and the radius of convergence ofX
i>0
µ(i)zi is1.
Example: µ(i)∼c/iβ withc >0andβ >2.
II. 2) Large non-generic trees
State of the art Non-generic trees Limit theorems Extensions
Large non-generic trees
Fixµnon-generic. What does aPµ[·|ζ(τ) =n]tree look like for nlarge (Jonsson & Stefánsson 11’)?
Condensationphenomenon
State of the art Non-generic trees Limit theorems Extensions
Large non-generic trees
Fixµnon-generic. What does aPµ[·|ζ(τ) =n]tree look like for nlarge (Jonsson & Stefánsson 11’)?
Condensationphenomenon
State of the art Non-generic trees Limit theorems Extensions
Large non-generic trees
Fixµnon-generic. What does aPµ[·|ζ(τ) =n]tree look like for nlarge (Jonsson & Stefánsson 11’)?
Condensationphenomenon
Condensationphenomenon (which also appears in the zero-range process !).
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2.
Lettn be aPµ[·|ζ(τ) =n]tree and mbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) Then:
tn
−→(d) n→∞ Tb
We havetn
−→(d)
n→∞ Tbwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) Then:
tn
−→(d) n→∞ Tb
where the convergence holds for the local convergence
We havetn
−→(d)
n→∞ bTwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) Then:
tn
−→(d) n→∞ Tb
where the convergence holds for the local convergence and the treebThas the following form:
We havetn
−→(d)
n→∞ bTwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) Then:
tn
−→(d) n→∞ Tb
where the convergence holds for the local convergence and the treebThas the following form:
The spine has a finite random lengthS, where:
P[S=i] = (1−m)mifori>0
We havetn
−→(d)
n→∞ bTwhere bTis:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Remarks:
I In the critical case the spine is infinite (Kesten ’86).
I Janson ’12: Assertion 1) holds for every non-genericµ.
I A GWµ tree has in expectation1+m+m2+· · ·=1/(1−m)vertices. Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
- What are the fluctuations of the maximal degree? - Where is located the vertex of maximal degree ? - What is the height oftn?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Remarks:
I In the critical case the spine is infinite (Kesten ’86).
I Janson ’12: Assertion 1) holds for every non-genericµ.
I A GWµ tree has in expectation1+m+m2+· · ·=1/(1−m)vertices. Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
- What is the height oftn?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Remarks:
I In the critical case the spine is infinite (Kesten ’86).
I Janson ’12: Assertion 1) holds for every non-genericµ.
I A GWµ tree has in expectation1+m+m2+· · ·=1/(1−m)vertices.
Hence a forest ofcntrees GWµhas in expectationcn/(1−m)vertices. Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Remarks:
I In the critical case the spine is infinite (Kesten ’86).
I Janson ’12: Assertion 1) holds for every non-genericµ.
I A GWµ tree has in expectation1+m+m2+· · ·=1/(1−m)vertices.
Hence a forest ofcntrees GWµhas in expectationcn/(1−m)vertices.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
- What are the fluctuations of the maximal degree?
State of the art Non-generic trees Limit theorems Extensions
Theorem (Jonsson & Stefánsson ’11)
Letµ be a subcritical offspring distribution such thatµ(i)∼c/iβ withc >0, β >2. Lettn be aPµ[·|ζ(τ) =n]tree andmbe the mean ofµ.
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
- What are the fluctuations of the maximal degree?
- Where is located the vertex of maximal degree ?
1) We havetn −→(d)
n→∞ bTwhereTb is:
2) The maximal degree oftn, divided byn, converges in probability towards 1−m.
Questions:
- Do the Lukasiewicz path and contour function oftn, properly rescaled, converge?
- What are the fluctuations of the maximal degree?
- Where is located the vertex of maximal degree ? - What is the height oftn?
III. Limit theorems for non-generic trees
State of the art Non-generic trees Limit theorems Extensions
Assumptions
We consider an offspring distribution µsuch that:
- µis subcritical (0<P
iiµ(i)<1)
Lettn be aPµ[·|ζ(τ) =n]tree.
State of the art Non-generic trees Limit theorems Extensions
Assumptions
We consider an offspring distribution µsuch that:
- µis subcritical (0<P
iiµ(i)<1)
- There exists a slowly varying functionLsuch that µ(n)= L(n)
n1+θ, n>1 with fixedθ >1.
State of the art Non-generic trees Limit theorems Extensions
Assumptions
We consider an offspring distribution µsuch that:
- µis subcritical (0<P
iiµ(i)<1)
- There exists a slowly varying functionLsuch that µ(n)= L(n)
n1+θ, n>1 with fixedθ >1.
(Lis slowly varying ifL(tx)/L(x)→1whenx→∞, ∀t >0.)
We consider an offspring distribution µsuch that:
- µis subcritical (0<P
iiµ(i)<1)
- There exists a slowly varying functionLsuch that µ(n)= L(n)
n1+θ, n>1 with fixedθ >1.
(Lis slowly varying ifL(tx)/L(x)→1whenx→∞, ∀t >0.) Lettn be aPµ[·|ζ(τ) =n]tree.
III. 1) Convergence of the Lukasiewicz path
0 1
2 3
4
5 6
7
8 9 10
11 12 13
14 15
16
17 18
19
20 21
22 23
24 25
0 5 10 15 20 25 30
0 1 2 3 4 5 6
Letku be the number of children of the vertexu.
Definition
The Lukasiewicz pathW(τ) = (Wn(τ), 06n6ζ(τ))of a treeτis defined by : W0(τ) =0, Wn+1(τ) =Wn(τ) +ku(n)(τ) −1.
State of the art Non-generic trees Limit theorems Extensions
Convergence of the Lukasiewicz path
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
State of the art Non-generic trees Limit theorems Extensions
Convergence of the Lukasiewicz path
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
66 n
(iii)
W[nt]∨(U(tn)+1)(tn)
n , 06t61
(d)
n→−→∞ ((1−m)(1−t))06t61.
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
State of the art Non-generic trees Limit theorems Extensions
Convergence of the Lukasiewicz path
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
State of the art Non-generic trees Limit theorems Extensions
Convergence of the Lukasiewicz path
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
(i) U(tn)/nconverges in probability towards0 asn→∞.
(iii) W[nt]∨(U(tn)+1) n
n , 06t61 −→(d)
n→∞ ((1−m)(1−t))06t61.
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
State of the art Non-generic trees Limit theorems Extensions
Convergence of the Lukasiewicz path
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
(i) U(tn)/nconverges in probability towards0 asn→∞. (ii) sup
06i6U(tn)
Wi(tn) n
(P) n→−→∞ 0.
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
(i) U(tn)/nconverges in probability towards0 asn→∞. (ii) sup
06i6U(tn)
Wi(tn) n
(P) n→−→∞ 0.
(iii)
W[nt]∨(U(tn)+1)(tn)
n , 06t61
(d)
n→−→∞ ((1−m)(1−t))06t61.
0 0.2 0.4 0.6 0.8 1
−0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
We have:
(i) U(tn)/nconverges in probability towards0 asn→∞. (ii) sup
06i6U(tn)
Wi(tn) n
(P) n→−→∞ 0.
(iii)
W[nt]∨(U(tn)+1)(tn)
n , 06t61
(d)
n→−→∞ ((1−m)(1−t))06t61. Remarks:
I The limit is deterministic and depends only onm(the mean ofµ).
LetU(tn)be the index of the first vertex with maximal degree oftn.
Theorem (K. 12’)
We have:
(i) U(tn)/nconverges in probability towards0 asn→∞. (ii) sup
06i6U(tn)
Wi(tn) n
(P) n→−→∞ 0.
(iii)
W[nt]∨(U(tn)+1)(tn)
n , 06t61
(d)
n→−→∞ ((1−m)(1−t))06t61.
Remarks:
I The limit is deterministic and depends only onm(the mean ofµ).
I With high probability, there is one vertex with degree roughly(1−m)nand the others have degree o(n).
State of the art Non-generic trees Limit theorems Extensions
Idea of the proof
I We know thatW(tn)has the law of arandom walk(Wn)n>0 with jump distributionν(k) =µ(k+1),k>−1
all the other jumps are asymptotically independent (the distribution ofW1
is(0, 1]–subexponential).
State of the art Non-generic trees Limit theorems Extensions
Idea of the proof
I We know thatW(tn)has the law of arandom walk(Wn)n>0 with jump distributionν(k) =µ(k+1),k>−1, conditioned on
W1>0,W2>0, . . . ,Wn−1>0 andWn = −1.
all the other jumps are asymptotically independent (the distribution ofW1
is(0, 1]–subexponential).
State of the art Non-generic trees Limit theorems Extensions
Idea of the proof
I We know thatW(tn)has the law of arandom walk(Wn)n>0 with jump distributionν(k) =µ(k+1),k>−1, conditioned on
W1>0,W2>0, . . . ,Wn−1>0 andWn = −1.
I ButE[W1] =m−1<0.
State of the art Non-generic trees Limit theorems Extensions
Idea of the proof
I We know thatW(tn)has the law of arandom walk(Wn)n>0 with jump distributionν(k) =µ(k+1),k>−1, conditioned on
W1>0,W2>0, . . . ,Wn−1>0 andWn = −1.
I ButE[W1] =m−1<0.
I By the “one big jump principle”,W(tn)makes one macroscopic jump, and all the other jumps are asymptotically independent
I We know thatW(tn)has the law of arandom walk(Wn)n>0 with jump distributionν(k) =µ(k+1),k>−1, conditioned on
W1>0,W2>0, . . . ,Wn−1>0 andWn = −1.
I ButE[W1] =m−1<0.
I By the “one big jump principle”,W(tn)makes one macroscopic jump, and all the other jumps are asymptotically independent (the distribution ofW1 is(0, 1]–subexponential).
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree
Recall the local con- vergence oftn to
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree,∆(tn)its degree
Recall the local con- vergence oftn to
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree,∆(tn)its degree and|u?(tn)|its height.
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree,∆(tn)its degree and|u?(tn)|its height.
I The fluctuations of∆(tn)around(1−m)nare of ordern2∧θ.
Recall the local con- vergence oftn to
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree,∆(tn)its degree and|u?(tn)|its height.
I The fluctuations of∆(tn)around(1−m)nare of ordern2∧θ.
I Fori>0,P[|u?(tn)|=i] −→
n→∞ (1−m)mi.
Recall the local con- vergence oftn to
State of the art Non-generic trees Limit theorems Extensions
Applications
Letu?(tn)be the vertex of maximal degree,∆(tn)its degree and|u?(tn)|its height.
I The fluctuations of∆(tn)around(1−m)nare of ordern2∧θ.
I Fori>0,P[|u?(tn)|=i] −→
n→∞ (1−m)mi.
Recall the local con- vergence oftn to