Maksim Zhukovskii – Zero-one k-laws for G(n,n−α)
description
Transcript of Maksim Zhukovskii – Zero-one k-laws for G(n,n−α)
Zero-one k-laws and extended zero-one
k-laws for random distance graphs
Popova Svetlana
Lomonosov Moscow State University
Workshop on Extremal Graph Theory
Moscow, June 6, 2014
1/22
Denitions: Erdos-Renyi random graph G(n, p) and random graph G(Gn, p)
Denition. Erdos-Renyi random graph G(n, p) is a random element
with values in Ωn and distribution Pn,p on Fn, where
Ωn = (V = 1, ..., n, E), Fn = 2Ωn ,
Pn,p(G) = p|E|(1− p)C2n−|E|.
Denition. Let Gn be a sequence of graphs Gn = (Vn, En).
Random graph G(Gn, p) is a random element with values in ΩGnand distribution PGn,p on FGn , where
ΩGn = G = (V,E) : V = Vn, E ⊆ En,
FGn = 2ΩGn , PGn,p(G) = p|E|(1− p)|En|−|E|.
2/22
Denitions: Erdos-Renyi random graph G(n, p) and random graph G(Gn, p)
Denition. Erdos-Renyi random graph G(n, p) is a random element
with values in Ωn and distribution Pn,p on Fn, where
Ωn = (V = 1, ..., n, E), Fn = 2Ωn ,
Pn,p(G) = p|E|(1− p)C2n−|E|.
Denition. Let Gn be a sequence of graphs Gn = (Vn, En).
Random graph G(Gn, p) is a random element with values in ΩGnand distribution PGn,p on FGn , where
ΩGn = G = (V,E) : V = Vn, E ⊆ En,
FGn = 2ΩGn , PGn,p(G) = p|E|(1− p)|En|−|E|.
2/22
Denitions: rst-order properties and zero-one law
Denition. First-order properties of graphs are dened by
rst-order formulae, which are built of
predicate symbols ∼,=logical connectivities ¬,⇒,⇔,∨,∧variables x, y, . . .
quantiers ∀,∃
Denition. The random graph G(n, p) is said to follow zero-one law
if for any rst-order property L either
limn→∞
Pn,p(L) = 0
orlimn→∞
Pn,p(L) = 1.
3/22
Denitions: rst-order properties and zero-one law
Denition. First-order properties of graphs are dened by
rst-order formulae, which are built of
predicate symbols ∼,=logical connectivities ¬,⇒,⇔,∨,∧variables x, y, . . .
quantiers ∀,∃
Denition. The random graph G(n, p) is said to follow zero-one law
if for any rst-order property L either
limn→∞
Pn,p(L) = 0
orlimn→∞
Pn,p(L) = 1.
3/22
Denitions: zero-one k-law
Denition. The random graph G(n, p) is said to follow zero-one
k-law if for any property L dened by a rst-order formula with
quantier depth at most k either
limn→∞
Pn,p(L) = 0
orlimn→∞
Pn,p(L) = 1.
4/22
Zero-one law for Erdos-Renyi random graph G(n, p)
Theorem(Glebski et al., 1969; Fagin, 1976)
Let a function p = p(n) satisfy the property
∀β > 0 min(p, 1− p)nβ →∞ when n→∞.
Then the random graph G(n, p) follows the zero-one law.
Theorem(Shelah, Spencer, 1988)
Let p(n) = n−β and β be an irrational number, 0 < β < 1.
Then the random graph G(n, p) follows the zero-one law.
5/22
Zero-one law for Erdos-Renyi random graph G(n, p)
Theorem(Glebski et al., 1969; Fagin, 1976)
Let a function p = p(n) satisfy the property
∀β > 0 min(p, 1− p)nβ →∞ when n→∞.
Then the random graph G(n, p) follows the zero-one law.
Theorem(Shelah, Spencer, 1988)
Let p(n) = n−β and β be an irrational number, 0 < β < 1.
Then the random graph G(n, p) follows the zero-one law.
5/22
Random distance graph
Random distance graph G(Gdistn , p)
Gdistn = (V distn , Edistn )
a = a(n), c = c(n)
V distn =
v = (v1, . . . , vn) : vi ∈ 0, 1,
n∑i=1
vi = a
Edistn = u,v ∈ V distn × V dist
n : (u,v) = c
6/22
Zero-one law for random distance graph
Let a function p = p(n) satisfy the property
∀β > 0 min(p, 1− p)|V distn |β →∞ when n→∞.
Theorem
Let a(n) = αn, c(n) = α2n, α ∈ Q, 0 < α < 1. Then
the random graph G(Gdistn , p) doesn't follow the zero-one law, but
there exists a subsequence G(Gdistni, p) following the zero-one law.
7/22
Zero-one law for random distance graph
Let a function p = p(n) satisfy the property
∀β > 0 min(p, 1− p)|V distn |β →∞ when n→∞.
Theorem
Let a(n) = αn, c(n) = α2n, α ∈ Q, 0 < α < 1. Then
the random graph G(Gdistn , p) doesn't follow the zero-one law, but
there exists a subsequence G(Gdistni, p) following the zero-one law.
7/22
Questions
When does a given subsequence G(Gdistni, p) follow zero-one
law?
Does there exist a rst-order property L and a subsequenceG(Gdistni
, p) such that
limi→∞
PGdistni,p(L) ∈ (0, 1)
What limiting probabilities PGdistni,p(L) can we get?
8/22
Questions
When does a given subsequence G(Gdistni, p) follow zero-one
law?
Does there exist a rst-order property L and a subsequenceG(Gdistni
, p) such that
limi→∞
PGdistni,p(L) ∈ (0, 1)
What limiting probabilities PGdistni,p(L) can we get?
8/22
Questions
When does a given subsequence G(Gdistni, p) follow zero-one
law?
Does there exist a rst-order property L and a subsequenceG(Gdistni
, p) such that
limi→∞
PGdistni,p(L) ∈ (0, 1)
What limiting probabilities PGdistni,p(L) can we get?
8/22
Extended zero-one k-law
Denition. The random graph G(Gn, p) is said to follow extended
zero-one k-law if for every property L dened by a rst-order
formula with quantier depth at most k any partial limit of the
sequence PGn,p(L) equals either 0 or 1.
Goal. Find conditions on the sequence G(Gdistni, p) under which one
of the following takes place:
zero-one k-law holds
zero-one k-law doesn't hold, but extended zero-one k-law holds
extended zero-one k-law doesn't hold
9/22
Extended zero-one k-law
Denition. The random graph G(Gn, p) is said to follow extended
zero-one k-law if for every property L dened by a rst-order
formula with quantier depth at most k any partial limit of the
sequence PGn,p(L) equals either 0 or 1.
Goal. Find conditions on the sequence G(Gdistni, p) under which one
of the following takes place:
zero-one k-law holds
zero-one k-law doesn't hold, but extended zero-one k-law holds
extended zero-one k-law doesn't hold
9/22
Ehrenfeucht game EHR(G,H, k)
EHR(G,H, k)
Graphs G,H, number of rounds kTwo players Spoiler and Duplicator
i-th round:
Spoiler chooses a vertex either from G or from H
Duplicator chooses a vertex of the other graph
Let x1, . . . , xk, y1, . . . , yk be vertices chosen from graphs G andH respectively.Duplicator wins if and only if G|x1,...,xk ∼= H|y1,...,yk.
Theorem
The random graph G(Gn, p) follows zero-one k-law if and only if
P(Duplicator wins the game EHR(G(Gn, p), G(Gm, p), k))→ 1
as n,m→∞.
10/22
Ehrenfeucht game EHR(G,H, k)
EHR(G,H, k)
Graphs G,H, number of rounds kTwo players Spoiler and Duplicator
i-th round:
Spoiler chooses a vertex either from G or from H
Duplicator chooses a vertex of the other graph
Let x1, . . . , xk, y1, . . . , yk be vertices chosen from graphs G andH respectively.Duplicator wins if and only if G|x1,...,xk ∼= H|y1,...,yk.
Theorem
The random graph G(Gn, p) follows zero-one k-law if and only if
P(Duplicator wins the game EHR(G(Gn, p), G(Gm, p), k))→ 1
as n,m→∞.
10/22
Full level extension property
Denition. The graph G = (V,E) is said to satisfy full level t
extension property if for any vertices v1, . . . ,vl, u1, . . . ,ur(l + r ≤ t) there exists a vertex v adjacent to v1, . . . ,vl and
non-adjacent to u1, . . . ,ur.
Proposition
Let G(Gn, p) satisfy full level (k − 1) extension property
asymptotically almost surely. Then the random graph G(Gn, p)follows zero-one k-law.
Corollary
Let G(Gn, p) satisfy full level t extension property a.a.s for every
t ∈ N. Then the random graph G(Gn, p) follows the zero-one law.
11/22
Full level extension property
Denition. The graph G = (V,E) is said to satisfy full level t
extension property if for any vertices v1, . . . ,vl, u1, . . . ,ur(l + r ≤ t) there exists a vertex v adjacent to v1, . . . ,vl and
non-adjacent to u1, . . . ,ur.
Proposition
Let G(Gn, p) satisfy full level (k − 1) extension property
asymptotically almost surely. Then the random graph G(Gn, p)follows zero-one k-law.
Corollary
Let G(Gn, p) satisfy full level t extension property a.a.s for every
t ∈ N. Then the random graph G(Gn, p) follows the zero-one law.
11/22
Full level extension property
Denition. The graph G = (V,E) is said to satisfy full level t
extension property if for any vertices v1, . . . ,vl, u1, . . . ,ur(l + r ≤ t) there exists a vertex v adjacent to v1, . . . ,vl and
non-adjacent to u1, . . . ,ur.
Proposition
Let G(Gn, p) satisfy full level (k − 1) extension property
asymptotically almost surely. Then the random graph G(Gn, p)follows zero-one k-law.
Corollary
Let G(Gn, p) satisfy full level t extension property a.a.s for every
t ∈ N. Then the random graph G(Gn, p) follows the zero-one law.
11/22
Full level extension property for random distance graph
Proposition
Let a(n) = αn, α ∈ Q, 0 < α < 1. Then G(Gdistni, p) satises
full level t extension property a.a.s for every t ∈ N if and only if
c = α2n and ∀m ∈ N m|ni for suciently large i.
Proposition
Let a(n) = αn, c = α2n, α ∈ Q, 0 < α < 1, t ≤ 5. Then
G(Gdistni, p) satises full level t extension property a.a.s if and only if
Dt|a(ni)− c(ni) for suciently large i, where
D2 = 1, D3 = 2, D4 = 6, D5 = 60.
12/22
Full level extension property for random distance graph
Proposition
Let a(n) = αn, α ∈ Q, 0 < α < 1. Then G(Gdistni, p) satises
full level t extension property a.a.s for every t ∈ N if and only if
c = α2n and ∀m ∈ N m|ni for suciently large i.
Proposition
Let a(n) = αn, c = α2n, α ∈ Q, 0 < α < 1, t ≤ 5. Then
G(Gdistni, p) satises full level t extension property a.a.s if and only if
Dt|a(ni)− c(ni) for suciently large i, where
D2 = 1, D3 = 2, D4 = 6, D5 = 60.
12/22
Zero-one k-laws for random distance graph
Notation. a = αn, c = α2n, α = s/q, (s, q) = 1.
Theorem (zero-one 4-law)
The random graph G(Gdistn , p) follows extended zero-one 4-law.
The sequence G(Gdistni, p) follows zero-one 4-law if and only if
∃i0 such that all the numbers a(ni)− c(ni) for i > i0 have thesame parity.
Theorem (zero-one 5-law)
Let a sequence ni be such that a(ni)− c(ni) are even for
suciently large i. Then
G(Gdistni, p) follows extended zero-one 5-law,
G(Gdistni, p) follows zero-one 5-law if and only if ∃i0 such that
either ∀i > i0 3|a(ni)− c(ni) or ∀i > i0 3 - a(ni)− c(ni).
13/22
Zero-one k-laws for random distance graph
Notation. a = αn, c = α2n, α = s/q, (s, q) = 1.
Theorem (zero-one 4-law)
The random graph G(Gdistn , p) follows extended zero-one 4-law.
The sequence G(Gdistni, p) follows zero-one 4-law if and only if
∃i0 such that all the numbers a(ni)− c(ni) for i > i0 have thesame parity.
Theorem (zero-one 5-law)
Let a sequence ni be such that a(ni)− c(ni) are even for
suciently large i. Then
G(Gdistni, p) follows extended zero-one 5-law,
G(Gdistni, p) follows zero-one 5-law if and only if ∃i0 such that
either ∀i > i0 3|a(ni)− c(ni) or ∀i > i0 3 - a(ni)− c(ni).
13/22
Zero-one k-laws for random distance graph
Notation. a = αn, c = α2n, α = s/q, (s, q) = 1.
Theorem (zero-one 4-law)
The random graph G(Gdistn , p) follows extended zero-one 4-law.
The sequence G(Gdistni, p) follows zero-one 4-law if and only if
∃i0 such that all the numbers a(ni)− c(ni) for i > i0 have thesame parity.
Theorem (zero-one 5-law)
Let a sequence ni be such that a(ni)− c(ni) are even for
suciently large i. Then
G(Gdistni, p) follows extended zero-one 5-law,
G(Gdistni, p) follows zero-one 5-law if and only if ∃i0 such that
either ∀i > i0 3|a(ni)− c(ni) or ∀i > i0 3 - a(ni)− c(ni).
13/22
Zero-one k-laws for random distance graph
Theorem (zero-one 6-law)
Let q 6= 5 and a sequence ni be such that a(ni)− c(ni) are
divisible by 12 for suciently large i. Then
G(Gdistni, p) follows extended zero-one 6-law,
G(Gdistni, p) follows zero-one 6-law if and only if ∃i0 such that
either ∀i > i0 5|a(ni)− c(ni) or ∀i > i0 5 - a(ni)− c(ni).
14/22
Disproof of extended zero-one laws for random distance graph
∀β > 0 min(p, 1− p)|V distn |β →∞ as n→∞. (∗)
Theorem (disproof of extended zero-one 6-law)
Let one of the following two cases take place:
q = 5 and a sequence ni is such that a(ni)− c(ni) are not
divisible by 5 for suciently large i,
α = 12 and a sequence ni is such that a(ni)− c(ni) are not
divisible by 4 for suciently large i.
Then there exists a function p(n) satisfying (∗) such thatG(Gdistni
, p) doesn't follow extended zero-one 6-law.
15/22
Disproof of extended zero-one laws for random distance graph
Theorem (disproof of extended zero-one law)
Let q be even, α ∈ (14 ,
34) and a sequence ni be such that
a(ni)− c(ni) are not divisible by 4 for suciently large i. Then
there exists a function p(n) satisfying (∗) such that G(Gdistni, p)
doesn't follow extended zero-one law.
16/22
Special sets of vertices
Denition. Vertices v1, . . . ,vt of a graph G = (V,E) are said to
form a special t-set if there doesn't exist a vertex v ∈ V adjacent
to all of the vertices v1, . . . ,vt.
Let Rt be a property of spanning subgraphs of Gn:for any vertices v1, . . . ,vt not forming a special t-set in Gn and
for any subset U ⊆ v1, . . . ,vt there exists a vertex v
adjacent to all vertices from U and non-adjacent to all vertices
from v1, . . . ,vt \ U .
Proposition
For every t ∈ N the random graph G(Gdistn , p) satisfyes Rt a.a.s.
17/22
Special sets of vertices
Denition. Vertices v1, . . . ,vt of a graph G = (V,E) are said to
form a special t-set if there doesn't exist a vertex v ∈ V adjacent
to all of the vertices v1, . . . ,vt.
Let Rt be a property of spanning subgraphs of Gn:for any vertices v1, . . . ,vt not forming a special t-set in Gn and
for any subset U ⊆ v1, . . . ,vt there exists a vertex v
adjacent to all vertices from U and non-adjacent to all vertices
from v1, . . . ,vt \ U .
Proposition
For every t ∈ N the random graph G(Gdistn , p) satisfyes Rt a.a.s.
17/22
Special sets of vertices
Denition. Vertices v1, . . . ,vt of a graph G = (V,E) are said to
form a special t-set if there doesn't exist a vertex v ∈ V adjacent
to all of the vertices v1, . . . ,vt.
Let Rt be a property of spanning subgraphs of Gn:for any vertices v1, . . . ,vt not forming a special t-set in Gn and
for any subset U ⊆ v1, . . . ,vt there exists a vertex v
adjacent to all vertices from U and non-adjacent to all vertices
from v1, . . . ,vt \ U .
Proposition
For every t ∈ N the random graph G(Gdistn , p) satisfyes Rt a.a.s.
17/22
Proof of zero-one k-laws: special sets of vertices without edges
Suppose
(1) Gn = (Vn, En) doesn't have special (t− 1)-sets
(2) G(Gn, p) satisfyes Rt a.a.s.
Proposition
Let a sequence Gn = (Vn, En) satisfy (1), (2) and the followingconditions:
Gn has special t-sets,
for every special t-set any two of its vertices are non-adjacent.
Then the random graph G(Gn, p) follows zero-one (t+ 1)-law.
18/22
Proof of zero-one k-laws: special sets of vertices without edges
Suppose
(1) Gn = (Vn, En) doesn't have special (t− 1)-sets
(2) G(Gn, p) satisfyes Rt a.a.s.
Proposition
Let a sequence Gn = (Vn, En) satisfy (1), (2) and the followingconditions:
Gn has special t-sets,
for every special t-set any two of its vertices are non-adjacent.
Then the random graph G(Gn, p) follows zero-one (t+ 1)-law.
18/22
Proof of zero-one k-laws: special sets of vertices with edges
Proposition
Suppose Gn = (Vn, En) satises (1), (2) and for any vertices
v1, . . . ,vi where i < t one of the following holds:
for any vertex vi+1 such that v1, . . . ,vi+1 can be extended toa special t-set there exist Ω(|Vn|β) dierent vertices each of
which can be mapped onto vi+1 by an automorphism of Gnxing v1, . . . ,vi (where β is a positive constant),
|(vi+1, . . . ,vt) : v1, . . . ,vt is a special t-set| = O(1).
Then the random graph G(Gn, p) follows extended zero-one(t+ 1)-law.
19/22
Disproof of extended zero-one k-laws
Let L be a property of subgraphs G ⊆ Gn:for any (v1, . . . ,vi) that can be extended to a special t-set with
edges in Gn there exist vi+1, . . . ,vt extending (v1, . . . ,vi) to a
special t-set with edges in G.
Let K(v1, . . . ,vi) be the number of (vi+1, . . . ,vt) extending
(v1, . . . ,vi) to a special t-set with edges in Gn.
If there exists (v1, . . . ,vi) with
K(v1, . . . ,vi)→∞, K(v1, . . . ,vi) = |Vn|o(1),
then PGn,p(L) can approach any number from (0, 1).
20/22
Disproof of extended zero-one k-laws
Let L be a property of subgraphs G ⊆ Gn:for any (v1, . . . ,vi) that can be extended to a special t-set with
edges in Gn there exist vi+1, . . . ,vt extending (v1, . . . ,vi) to a
special t-set with edges in G.
Let K(v1, . . . ,vi) be the number of (vi+1, . . . ,vt) extending
(v1, . . . ,vi) to a special t-set with edges in Gn.
If there exists (v1, . . . ,vi) with
K(v1, . . . ,vi)→∞, K(v1, . . . ,vi) = |Vn|o(1),
then PGn,p(L) can approach any number from (0, 1).
20/22
Disproof of extended zero-one k-laws
Let L be a property of subgraphs G ⊆ Gn:for any (v1, . . . ,vi) that can be extended to a special t-set with
edges in Gn there exist vi+1, . . . ,vt extending (v1, . . . ,vi) to a
special t-set with edges in G.
Let K(v1, . . . ,vi) be the number of (vi+1, . . . ,vt) extending
(v1, . . . ,vi) to a special t-set with edges in Gn.
If there exists (v1, . . . ,vi) with
K(v1, . . . ,vi)→∞, K(v1, . . . ,vi) = |Vn|o(1),
then PGn,p(L) can approach any number from (0, 1).
20/22
Disproof of extended zero-one k-laws
Replace L by a rst-order property L:
L = ∀v1 . . . ∀vi ∃vi+1 . . . ∃vt Q(v1, . . . ,vt),
where Q approximately says that either (v1, . . . ,vi) can't be
extended to a special t-set with edges in Gn or (v1, . . . ,vt) forms a
special t-set with edges in G(Gn, p).
21/22
Thanks
Thank you for your attention!
22/22