Sei sulla pagina 1di 25

modified:1994-08-30

370 revision:1994-08-30

On the number of non-isomorphic subgraphs

S. Shelah Institute of Mathematics Hebrew University, Jerusalem

L. Soukup Mathematical Institute of the Hungarian Academy of Sciences

October 6, 2003

1

Abstract Let K be the family of graphs on ω 1 without cliques or independent subsets of size ω 1 . We prove that

it is consistent with CH that every G ∈ K has 2 ω 1 many pairwise non-isomorphic subgraphs,

the following proposition holds in L: () there is a G ∈ K such

=

(a)

(b)

(c)

that for each partition (A, B) of ω 1 either G

G[B],

the failure of () is consistent with ZFC.

= G[A]

or

G

Introduction

We assume only basic knowledge of set theory — simple combinatorics for section 2, believing in L |= + defined below for section 3, and finite support iterated forcing for section 4.

The first author was supported by the United States Israel Binational Science Foun- dation, Publication 370 The second author was supported by the Hungarian National Foundation for Scientific Research grant no. l805

1

October 6, 2003

2

modified:1994-08-30

370 revision:1994-08-30

Answering a question of R. Jamison, H. A. Kierstead and P. J. Nyikos [5] proved that if an n-uniform hypergraph G = V, E is isomorphic to each of its induced subgraphs of cardinality |V |, then G must be either empty or complete. They raised several new problems. Some of them will be investi- gated in this paper. To present them we need to introduce some notions.

An infinite graph G = V, E is called non-trivial iff G contains no clique or independent subset of size |V |. Denote the class of all non-trivial graphs on ω 1 by K. Let I(G) be the set of all isomorphism classes of induced subgraphs of G = V, E with size |V |. H. A. Kierstead and P. J. Nyikos proved that |I(G)| ≥ ω for each G ∈ K and asked whether |I(G)| ≥ 2 ω or |I(G)| ≥ 2 ω 1 hold or not. In [3] it was shown that (i) |I(G)| ≥ 2 ω for each G ∈ K, (ii) under + there exists a

G ∈ K with |I(G)| = ω 1 . In section 2 we show that if ZFC is consistent, then

so is ZFC + CH + “|I(G)| = 2 ω 1 for each G ∈ K”. Given any G ∈ K we will investigate its partition tree. Applying the weak principle of Devlin

and Shelah [2] we show that if this partition tree is a special Aronszajn tree, then |I(G)| > ω 1 . This result completes the investigation of problem 2 of [5] for ω 1 . Consider a graph G = V, E . We say that G is almost smooth if it is isomorphic to G[W ] whenever W V with |V \ W | < |V |. The graph

G is called quasi smooth iff it is isomorphic either to G[W ] or to G[V \ W ]

whenever W V . H. A. Kierstead and P. J. Nyikos asked (problem 3) whether an almost smooth, non-trivial graph can exist. In [3] various models of ZFC was constructed which contain such graphs on ω 1 . It was also shown that the existence of a non-trivial, quasi smooth graph on ω 1 is consistent with ZFC. But in that model CH failed. In section 3 we prove that + , and so V=L, too, implies the existence of such a graph. In section 4 we construct a model of ZFC in which there is no quasi- smooth G ∈ K. Our main idea is that given a G ∈ K we try to construct a

partition (A 0 , A 1 ) of ω 1 which is so bad that not only G G[A i ] in the ground

model but certain simple generic extensions can not add such isomorphisms to the ground model. We divide the class K into three subclasses and develop different methods to carry out our plan. The question whether the existence of an almost-smooth G ∈ K can be proved in ZFC is still open. We use the standard set-theoretical notation throughout, cf [4]. Given a

graph G = V, E we write V (G) = V and E(G) = E. If H V (G) we

=

October 6, 2003

3

modified:1994-08-30

370 revision:1994-08-30

define G[H] to be H, E(G) [H] 2 .

{x, y} ∈ E}. If G and H are graphs we write G = H to mean that G and H are isomorphic. If f : V (G) V (H) is a function we denote by f : G H the fact that f is an isomorphism between G and H. Given a set X let Bij p (X) be the set of all bijections between subsets of X. If G = V, E is a graph take

Given x V

take G(x) = {y

V

=

:

Iso p (G) = {f Bij p (V ) : f : G[dom(f )] G[ran(f )]} .

=

We denote by Fin(X, Y ) the set of all functions mapping a finite subset of

X to Y .

Given a poset P and p, q P we write p P q to mean that p and q are compatible in P . The axiom + claims that there is a sequence S α : α < ω 1 of contable

sets such that for each X ω 1 we have a closed unbounded C ω 1 satisfying

X ν S ν and C ν S ν for each ν C.

We denote by TC(x) the transitive closure of a set x. If κ is a cardinal take H κ = {x : |TC(x)| < κ} and H κ = H κ , . Let us denote by D ω 1 the club filter on ω 1 .

2 I(G) can be always large

Theorem 2.1 Asume that GCH holds and every Aronszajn-tree is special. Then |I(G)| = 2 ω 1 for each G ∈ K.

Remark: S.Shelah proved, [7, chapter V. §6,7], that the assumption of theorem 2.1 is consistent with ZFC. During the proof we will apply the following definitions and lemmas.

Lemma 2.2

Assume that G ∈ K, A [ω 1 ] ω 1 and |{G(x) A : x ω 1 | = ω 1 .

Then |I(G)| = 2 ω 1 .

Proof: See [3, theorem 2.1 and lemma 2.13].

2 ω 1 . Proof: See [3, theorem 2.1 and lemma 2.13]. Definition 2.3 Consider a

Definition 2.3 Consider a graph G = ω 1 , E .

October 6, 2003

4

modified:1994-08-30

370 revision:1994-08-30

1. For each ν

ξ

take

ω 1

let us define the ordinal γ ν ω 1 and the sequence

= 0 and if ξ α : α < γ is defined, then

γ ν : γ γ ν as follows: put ξ

ν

0

ν

ξ γ ν = min {ξ : α < γ

ξ > ξ α and ({ξ α , ξ} ∈ E

ν

ν

iff {ξ α , ν} ∈ E)} .

ν

If ξ

ν = ν, then we put γ ν = γ.

γ

2. Given ν, µ ω 1 write ν G µ iff ξ

γ ν = ξ

γ µ for each γ γ ν .

3. Take T G = ω 1 , G . T G is called the partition tree of G.

Lemma 2.4 If G = ω 1 , E ∈ K with |I(G)| < 2 ω 1 , then T G is an Aronszajn tree.

Proof: By the construction of T G , if ν, µ ω 1 , ν < µ and G(ν) ν = G(µ) ν, then ν G µ. So the levels of T G are countable by lemma 2.2. On the other hand, T G does not contain ω 1 -branches, because the branches are prehomogeneous subsets and G is non-trivial.

branches are prehomogeneous subsets and G is non-trivial. Definition 2.5 ⊂ ω 1 . function g

Definition 2.5

ω 1 .

function g : ω 1 2 is an A-diamond for F

{α A : F (h α) = g(α)} is a stationary subset of ω 1 .

1.

Let F

:

(2 ω ) <ω 1

2

and A

We say that a

(2 ω ) ω 1 ,

iff,

for any h

2. A ω 1 is called a small subset of ω 1 iff for some F function is an A-diamond for F .

3. J = {A ω 1 : A is a small subset of ω 1 }.

: (2 ω ) <ω 1 2 no

In [2] the following was proved:

Theorem 2.6 If 2 ω < 2 ω 1 , then J is a countably complete, proper, normal ideal on ω 1 .

After this preparation we are ready to prove theorem 2.1. Proof: Assume that G = ω 1 , E ∈ K.

|I(G)| < 2 ω 1 and a contradiction will be derived. Since 2 ω 1 = ω 2 , we can fix a sequence {G ν : ν < ω 1 } of graphs on ω 1

Write

such that for each Y G ν = ω 1 , E ν .

[ω 1 ] ω 1

there is a ν

< ω 1

with G[Y ]

=

G ν .

October 6, 2003

5

modified:1994-08-30

370 revision:1994-08-30

. Since every Aronszajn-tree

is special and I is a countably complete ideal on ω 1 , there is an antichain S

in T G with S / J . Take

Consider the Aronszajn-tree T G =

ω 1 , G

A = α ω 1 : σ S(α G σ) .

Now property () below holds:

()

σ S ρ

(S A) \ σ + 1

α A σ ({σ, α} ∈ E iff {ρ, α} ∈/ E).

Indeed, if for each α

A σ

we had {σ, α} ∈ E

iff {ρ, α} ∈ E, then

σ G ρ would hold by the construction of T G .

S,

embedding. Define F (ν, σ, T, f ) 2 as follows:

Let ν

ω 1 ,

σ

T

S σ

and f

:

G[(A σ) T ]

G ν

be an

F (ν, σ, T, f ) = 1 iff x G ν (α

A σ)({x, f (α)} ∈ E ν iff {σ, α} ∈ E).

In case ωσ = σ, under suitable encoding, F can be viewed as a function from (2 ω ) <ω 1 to 2.

S

Since S / J , there is a g 2 ω 1 and f : G[A T ] G ν , the set

such that for every ν ω 1 = 2 ω , T

=

S T = {σ S : g(σ) = F(ν, σ, T σ, f σ)}

is stationary. Take T = {σ S : g(σ) = 0}. Choose an ordinal ν < ω 1 and a

function f with f : G[A T ] G ν .

by (), that

For each σ < ω 1 with σ = ωσ it follows,

=

σ T

iff x ω 1 α S σ ({x, f (α)} ∈ E ν iff {σ, α} ∈ E).

Thus g(σ) = 0 iff F (ν, σ, T σ, f σ) = 1, for each σ S, that is, S T which is a contradiction.

σ ∈ S , that is, S T which is a contradiction. = ∅ , 3

= ,

3 A quasi-smooth graph under +

Theorem 3.1 If + holds, then there exists a non-trivial, quasi-smooth graph on ω 1 .

October 6, 2003

6

modified:1994-08-30

370 revision:1994-08-30

Proof: Given a set X, A⊂P(X) and F ⊂Bij p (X) take

Cl(A, F ) = {B : B ⊃

A and B 0 , B 1 ∈ B ∀f

∈ F ∀Y

[X] <ω

{B 0 B 1 , f B 0 , B 0 Y }⊂B} .

We say that A is F -closed if A = Cl(A, F ). Given A, D⊂P (X), we say that

D is uncovered

by A if |D\A| = ω for each A ∈ A and D ∈ D.

Lemma 3.2 Assume that F ⊂Bij p (X) is a countable set, A 0 , A 1 P(X) are countable, F -closed families. If D⊂P (X) is a countable family which is uncovered by A 0 ∪ A 1 , then there is a partition (B 0 , B 1 ) of X such that D is uncovered by Cl(A i ∪ {B i } , F ) for i < 2.

Proof: We can assume that F is closed under composition. Fix an enumer- ation { D n , k n , F n , i n , A n : n ω} of D × ω × F <ω × { i, A : i 2, A ∈ A i }. By induction on n, we will pick points x n X and will define finite sets, B and B n , such that B n B n = and B n B

., f k1 . Take

n

0

1

0

1

i

i

n+1 .

Assume that we have done it for n 1. Write F n = f 0 ,

B n1 = B

0

n1 B

1

n1 and

B

n

= B n1 f

j

B n1 : j < k .

Pick an arbitrary point x n D n \(A n B

n

). Put

B

i

n

n

= B

i

n1

n

and

B

1i n

n

= B

1i n

n1

{x n } ∪ f

1

j

(x n ) : j < k .

with B i ⊃ ∪{B n : n < ω} for i < 2.

We claim that it works. Indeed, a typical element of Cl(A i ∪ {B i }, F ) has

the form

Next choose a partition (B 0 , B 1 ) of X

i

C

= A f

j

B i : j < k ,

where A ∈ A, k < ω and f 0 ,

, f k1 ∈ F . So, if D ∈ D, then

D\C ⊃ {x n : D n = D, A n

= A, i n = i and F n = f 0 ,

1

because x n / A and f

j

(x n ) B 1i by the constuction.

, − 1 because x n ∈ / A and f j ( x n )

, f k1 }

October 6, 2003

7

modified:1994-08-30

370 revision:1994-08-30

, f n1 . Given a family F ⊂Bij p (X) we

say that F each i < n.

assume that the empty term denotes the identity function on X. If l n

take (l) F = f 0 ,

∈ F , for We will

Consider a sequence F = f 0 ,

is an F -term provided f i = f

or f i = f 1 for some f

We denote the function f 0 ◦ ··· ◦ f n1 by F as well.

, f l1 and F (l) = f l ,

,

f n1 . Let

Sub(F ) = f i 0 ,

, f i l1 : l n, i 0

<

< i l1 < n .

Given f ∈ F and x, y X with x / dom(f ) and y / ran(f ) let F f,x,y be the term that we obtain replacing each occurrence of f and of f 1 in F with

f ∪ { x, y } and with f 1 ∪ { y, x }, respectively.

Lemma 3.3 Assume that F ⊂Bij p (X), A⊂P (X) is F -closed, F 0 ,

, F n1

are F -terms,

z 0 ,

, z n1 X, A 0 ,

, A n1 ∈ A such that for each i < n

()

z i / {F A i :

F Sub(F i )} .

 

If f

∈ F , x

X\dom(f ), Y

[X\ran(f )] ω

with

|A Y |

<

ω

for each

A ∈ A, then there are infinitely many y Y such that () remains true when replacing f with f ∪ { x, y }, that is,

(∗∗)

z i / F A i : F Sub(F

i

f,x,y

)

for each i < n.

Proof: It is enough to prove it for n = 1.

z = z 0 . Take

Write F = f 0

Y F,A = {y Y

: (∗∗) holds for y} .

, f k1 , A = A 0 ,

Now we prove the lemma by induction on k. If k = 0, then Y F,A = Y \A. Suppose we know the lemma for k 1. Using the induction hypothesis we can assume that () below holds:

()

Y = Y G,F (l) A : l n, G Sub( (l) F f,x,y ), G

=

F f,x,y .

Assume that |Y F,A | < ω and a contradiction will be derived. First let us remark that either f k1 = f or f k1 = f 1 by (). Case 1:f k1 = f 1 . Then Y F,A Y \A by (), so we are done.

October 6, 2003

8

modified:1994-08-30

370 revision:1994-08-30

Case 2:f k1 = f. In this case x

F f,x,y (x). Then for each y, y Y take

A and for all but finitely many y

Y

we have z

=

l(y, y ) = max l n : i < l F

(i)

f,x,y

(x) = F

(i)

f,x,y

(x) .

By Ramsey’s theorem, we can assume that l(y, y ) = l whenever y, y Y .

Clearly l

Thus z = (l1) F f,x,y (x) for each

Y , which contradicts () because x A. The lemma is proved. We are ready to construct our desired graph. First fix a sequence M α : α < ω 1 of countable, elementary submodels of some H λ with M γ : γ < α M α for each α < ω 1 , where λ is a large enough regular cardinal. Then choose a -sequence S α : α < ω 1 M 0 for the uncountable sub- sets of ω 1 , that is , {α < ω 1 : X α = S α } ∈/ NS(ω 1 ) whenever X [ω 1 ] ω 1 . We can also assume that S α is cofinal in α for each limit α. We will define, by induction on α,

y

f l = f 1 and F

(x), so

Then F

(l)

f,x,y

(l1)

f,x,y

(x)

f,x,y

F

(l)

(x)

but F

f,x,y

(l1)

(x)

f,x,y

F

(l1)

<

n.

=

=

(x) = x for each y Y .

( l − 1) < n . = = ( x ) = x for each

1. graphs G α = ωα, E α with G β = G α [ωβ] for β < α,

2. countable sets F α Iso p (G α ),

satisfying the induction hypotheses (I)–(II) below:

(I) {S ωγ : γ α} is uncovered by I α J α where

and

I α = Cl({G(ν) ν : ν

ωα} ,

βα

F β )

J α = Cl({ν\G(ν) : ν ωα} ,

βα

F β ).

To formulate (II) we need the following definition.

Definition 3.4 Assume that α = β + 1 and Y ωα. We say that Y is large if n ω, f i , x i : i < n , h if

October 6, 2003

9

modified:1994-08-30

370 revision:1994-08-30

1. i < n α i < β f i ∈ F α i ,

2. i < n ωα i x i < ωβ,

3. i < n ran(f i )Y ,

4. i

= j < n ran(f i ) ran(f j ) =

5. h Fin(Y ωβ, 2) and dom(h) {ran(f i ) : i < n} = ,

then y Y [ωβ, ωα) such that

6. i < n x dom(f i ) ({y, f i (x)} ∈ E α iff

7. z dom(h)

{y, z} ∈ E α iff h(z) = 1.

Take

(II)

If α = β + 1, then ωα is large.

{x i , x} ∈ E α ),

The construction will be carried out in such a way that G β : β α M α

and F β : β < α M α .

To start with take G 0 = , and F = {∅}. Assume that the construc-

tion is done for β < α. Case 1:α is limit. We must take G α = ∪{G β : β < α}. We will define sets F

and will take F α = F Let

0

α

α

∪ F α .

0

1

1

, F α Iso p (G α )

0 = {f Iso p (G α ) M α : α n : n < ω α

F

α

sup {α n : n < ω} = α,

f ωα n ∈ F α n and f ωα n : G α n

= G α n [ran(f )] for each n ω} .

Take F =

with F M α+1 , so M α+1 |= “|F | = ω”. Obviously both I F -closed and S = {S ωβ : β α} is uncovered by them.

β<α

F β ∪ F

0

α

, I

α

=

β<α I β and J

α

=

β<α J β . Clearly F M α

and J α are

α

1

¿From now on we work in M α+1 to construct F α . For W ωα write

L W = {ν Take

< α : W (ων + ω) is large}.

W α = { W, f (P(ωα) ∩ M α ) × (β<α F β ) : L W

is cofinal in α

October 6, 2003

10

modified:1994-08-30

370 revision:1994-08-30

and f : G γ f

= G γ f [W ωγ f ] for some γ f < α .

We want to find functions g W,f f for W, f ∈ W α such that

(A)

(B)

g W,f : G α G α [W]

=

taking F α = g W,f : W, f ∈ W α the induction hypothesis (I) remains true.

1

First we prove a lemma:

Lemma 3.5 If W,

f ∈ W α , g

Iso p (G α , G α [W ]), g

(i) for each x W \dom(f ) the set

f , |g\f | < ω, then

{y W : g ∪ { x, y } ∈ Iso p (G α , G α [W ])}

is cofinal in ωα.

(ii) for each y W \ran(f ) the set

{x W : g ∪ { x, y } ∈ Iso p (G α , G α [W ])}

is cofinal in ωα.

Proof: (i): Define the function h : ran(g)\ran(f ) 2 with h(g(z)) = 1

iff {z, x}

Since

W (ωβ + ω) is large, we have a y W [ωβ, ωβ + ω) such that

E α .

Choose β L W with ran(h)ωβ and γ f

β.

1. {y, f (z)} ∈ E α iff {x, z} ∈ E α for each z dom(f )

2. {y, g(z)} ∈ E α iff h(g(z)) = i for each z dom(g)\dom(f ).

But this means that g ∪ { x, y } ∈ Iso p (G α , G α [W ]). (ii) The same proof works using that ωβ + ω is large for each β < α.

using that ωβ + ω is large for each β < α . By induction on

By induction on n, we will pick points z n

W,f

n

: n < ω will work.

W,f

n

ωα and will construct

=

families of partial automorphisms, g

g

: W, f ∈ W α such

that g W,f

During the inductive construction we will speak about F α -terms and

about functions which are represented by them in the n th step.

October 6, 2003

11

modified:1994-08-30

370 revision:1994-08-30

If F

where

= h 0 ,

h k1 is an F α -term and n ω take F [n] = j 0 ◦ ··· ◦ j k1

j i =

 

W,f

g

(g

h i

n

W,f

n

) 1

if h i = g W,f ,

if h i = (g W,f ) 1 ,

otherwise.

First fix an enumeration { W n , f n , u n , i n : 1 n < ω} of W α × ωα × 2 and an enumeration F n,i : i < l n , j n , A n,i : i < l n , D n n < ω of the quadru-

F k1 are F α -

or j = 1 and

., A k1 J α . During the inductive construction conditions (i)–(v) below will be satis-

A 0 ,

terms, j 2, D ∈ S

ples F 0 ,

, F k1 , j, A 0 ,

, A k1 , D where k < ω, F 0 ,

, A k1 I

α

,

and either j = 0 and A 0 ,

fied:

g

(ii) g

(i)

(iii)

g

W,f

0

W,f

n

W,f

n

= f

Iso p (G α , G α [W ])

g

W,f

n1 , |g

W,f

n

\f| < ω

(iv) z k /

F [n] A k,i : F Sub(F k,i ) for each i < l k and k < n

(v) if i n = 1, then u n dom(g

W n ,f n

n

),

if i n = 0, then either u n / W n or u n ran(g

W n ,f n

n

).

If n = 0, then take g

W,f

0

= f.

If n > 0, then let g

n

W n , f n . Assume that

). Then, by lemma 3.5, the

= y W : g n1 ∪ { u n , y } ∈ Iso p (G α , G α [W ]) is unbounded in ωα.

are bounded in ωα, we can apply lemma 3.3

∪ { u n , y } condition

W,f

g n1 whenever W, f

=

W,f

=

i n = 0, set Y

Since the members of I

W, f = W n , f n and u n / dom(g

W,f

J

α

α

W n ,f n

n1

to pick a point y Y such that taking g

(iv) holds. If i n = 1 and W, f = W n , f n , then the same argument works. Finally pick a point

W n ,f n

n

W n ,f n

n1

= g

z n

/ D n \

F [n] A n,i : F Sub(F n,i ), i < l n .

The inductive construction is done.

October 6, 2003

12

modified:1994-08-30

370 revision:1994-08-30

Take

g W,f = g

W,f

n

: n < ω . By (v),

g

W,f

n

=

: G ωα

G ωα [W].

By (iv), we have

z k D k \ F

k,i A k,i : i < l k

and so it follows that {S ωβ : β α} is uncovered by I α J α . Case 2:α = β + 1.

To start with we fix an enumeration f pairs f i , x i : i < n , h satisfying 3.4.1–5. If k ω take

ν, x

and

k , x

i

i k : i < n k , h k : k ω of

k

i

) and

k

i

/ E β

B

k

= h

k

0

1

{0} ∪ f

k

i

(ν) : i < n k , ν dom(f

1

B k = h

1

k

{1} ∪ f

k

i

k

(ν) : i < n k , ν dom(f

i

) and

ν, x

k

i

E β .

Applying lemma 3.2 ω-many times we can find partitions (C of ωβ such that taking

+

I

β

= Cl((I β C

1

k

:

k ω ,

γβ

F γ )

0

k

, C

k

1

), k < ω,

and

+

J

β

= Cl((I β C

0

k

:

k ω ,

γβ

F γ )

+

the set {S ωγ : γ β} is uncovered by I

β

+

J β .

1

B k

We can assume that B k C k for i < 2 and k < ω because B

I β . Take

i

i

E α = E β {ν, ωβ + n} : ν < ωβ, n ω

and ν B

1

n

0

k

J β and

and

F α = .

By the construction of G α = ωα, E α , it follows that ωα is large, so (II) holds. On the other hand

I α = X Y

+

: X I

β

,

Y [ωα] <ω

October 6, 2003

13

modified:19