3

4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


On Taylors problem
P. Komjath and S. Shelah
By extending nite theorems Erd os and Rado proved that for every innite cardinal
there is a -chromatic triangle-free graph [3]. In later work they were able to add
the condition that the graph itself be of cardinal [4]. The next stage, eliminating 4-
circuits, turned out to be dierent, as it was shown by Erd os and Hajnal [1] that every
uncountably chromatic graph contains a 4-circuit. In fact, every nite bipartite graph
must be contained, but odd circuits can be omitted up to a certain length. This solved the
problem which nite graphs must be contained in every -chromatic graph for every >
. The next result was given by Erd os, Hajnal, and Shelah [2], namely, every uncountably
chromatic graph contains all odd circuits from some length onward. They, as well as
Taylor, asked the following problem. If , are uncountable cardinals and X is a -
chromatic graph, is there a -chromatic graph Y such that every nite subgraph of Y
appears as a subgraph of X. In [2] the following much stronger conjecture was posed. If X
is uncountably chromatic, then for some n it contains all nite subgraphs of the so-called
n-shift graph. This conjecture was, however, disproved in [5].
Here we give some results on Taylors conjecture when the additional hypotheses
|X| = , |Y | = are imposed.
We describe some (countably many) classes K
n,e
of nite graphs and prove that if

0
= then every
+
-chromatic graph of cardinal
+
contains, for some n, e, all members
of K
n,e
as subgraphs. On the other hand, it is consistent for every regular innite cardinal
that there is a
+
-chromatic graph on
+
that contains nite subgraphs only from K
n,e
.
We get, therefore, some models of set theory, where the nite subraphs of graphs with
|X| = Chr(X) =
+
for regular uncountable cardinals are described.
We notice that in [6] all countable graphs are described which appear in every graph
with uncountable coloring number.
Notation. x will denote a nite string of ordinals. x < y means that max(x) < min(y).
Denition. Assume that 1 n < , e : {1, 2, . . . , 2n} {0, 1} is a function with
|f
1
(0)| = n. We are going to dene the structures in K
n,e
as follows. They will be of the
form H = (V, <, U, X, h
1
, . . . , h
n
) where (V, <) is a nite linearly ordered set, U V , X
is a graph on U, h
i
: U V satisfy h
1
(x) < < h
n
(x) = x for x U. The elements
in K
n,e
0
are those isomorphic to (V, <, U, X, h
1
, . . . , h
n
) where V = {1, 2, . . . , n}, < is the
natural ordering, U = {n}, X = , h
i
(n) = i (1 i n).
If H = (V, <, U, X, h
1
, . . . , h
n
) is a structure of the above form, and x V , we form
the edgeless amalgamation H

= H +
x
H as follows. Put H

= H +
x
H = (V

, <

, U

, X

, h

1
, . . . , h

n
) where (V

, <

) has the <

-ordered decomposition V

= W V
0
V
1
, if
we put V

i
= W V
i
for i < 2 then the structures
_
V

i
, <

|V

i
, U

i
, h

1
|V

i
, . . . , h

n
|V

i
_
The rst author acknowledges the support of the Hungarian OTKA grant 2117.
Publication number No. 346 on the second authors list. His research was partially
covered by the Israel Academy Basic Research Fund.
1
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


are both isomorphic to H for i < 2 and min(V
i
) correspond to x under the isomorphisms.
If H = (V, <, U, X, h
1
, . . . , h
n
) is a structure of the above form, and x U, we also
form the one-edge amalgamation H

= H
x
H as follows. Enumerate in increasing order
e
1
(0) as {a
1
, . . . , a
n
} and e
1
(1) as {b
1
, . . . , b
n
}. Put H

= (V

, <

, U

, X

, h

1
, . . . , h

n
)
where (V

, <

) has the ordered decomposition V

= V
0
V
1
V
2n
; H

|(V
0

{V
i
:
e(i) = }) are isomorphic to H ( = 0, 1) if x
0
, x
1
are the points corresponding to x, then
h

i
(x
0
) = min(V
a
i
), h

i
(x
1
) = min(V
b
i
), and the only extra edge in X

is {x
0
, x
1
}.
We then put
K
n,e
t+1
=
_
H +
x
H : H = (V, <, U, . . .) K
n,e
t
, x V
_

_
H
y
H : H = (V, <, U, . . .) K
n,e
t
, y U
_
,
and nally K
n,e
=

{K
n,e
t
: t < }.
Theorem 1. If |G| = Chr(G) =
+
,

0
= , then, for some n, e, G contains every graph
in K
n,e
as subgraph.
We start with some technical observations.
Lemma 1. If t
n
:
+

+
are functions (n < ), then there is a -coloring F :
+

such that for F() = F(), i, j < , < t
i
() < t
j
() may not hold.
Proof. As

0
= , it suces to show this for two functions t
0
(), t
1
(), with t
1
() > .
We prove the stronger statement that there is a function F :
+
[]

such that if
< t
0
() < t
1
() then F() F() = . Let N

: <
+
be a continuous, increasing
sequence of elementary submodels of
+
; <, t
0
, t
1
, . . . with

= N

+
<
+
. C = {

:
<
+
} is closed, unbounded. We dene F|

by transnite recursion on . If F|

is
given, and has t
0
() <

<
+1
, by elementarity = sup{t
1
() : < t
0
()} <

,
and there is a

with t
0
(

) = t
0
(), <

<

. Put H() = F(

), otherwise, i.e., when

t
0
(), put H() = . To get F|[

,
+1
), we disjointize {H() :

<
+1
}, i.e.,
nd F() H() of cardinal such that F(
0
) F(
1
) = for
0
=
1
. We show that
this F works. Assume that F(), F() are not disjoint. By induction we can assume that
either or is between

and
+1
. By the disjointization process some of them must be
smaller than

. If <

<
+1
then t
0
() <

as N

is an elementary submodel,
so t
0
() < . Assume now that <

<
+1
. Our construction then selected a

with t
0
(

) = t
0
() and F() H() = F(

) which is, by the inductive hypothesis,

disjoint from F().
Lemma 2. If C = {

: <
+
} is a club then there is a function K : [
+
]

0
such
that if K(A) = K(B) and A [

,
+1
) = and B [

,
+1
) = for some <
+
then
A
+1
= B
+1
= and so A B is an initial segment both in A and B.
Proof. Fix for every <
+
an into function F

: such that for

0
<
1
<
2
,
F

1
(
0
) = F

2
(
1
) holds. This can be done by a straightforward inductive construction.
If A [
+
]

0
put X(A) = { : A [

,
+1
) = }. Let tp(X(A)) = . Enumerate
X(A) as {
A

: < }. Let K(A) be a function with domain , at < , if

A

= , let
K(A)() =

{F

(
A

) :

< }, {F

+1
(y) : y A
+1
}
_
.
2
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


Assume now that K(A) = K(B), X(A) X(B). If =
A

=
B

then =

by the
properties of F above. The second part of the denition of K(A) gives that A
+1
=
B
+1
.
Proof of Theorem 1. We rst show that one can assume that G is
+
-chromatic on
every closed unbounded set.
Lemma 3. There is a function f :
+

+
such that if C
+
is a closed unbounded
set then
_
[, f()] : C
_
is
+
-chromatic.
Proof. Assume that the statement of the Lemma fails. Put f
0
() = , for n < let
C
n
witness that f
n
:
+

+
is not good and f
n+1
() = min(C
n
( + 1)). As, by
assumption,

{[, f
n
()] : C
n
, n < } is -chromatic, there is a
/
_
_
_
, f
n
()

: C
n
, n <
_
, > min
_

_
C
n
: n <
__
.
Clearly, / C
n
(n < ), and if now
n
= max( C
n
), then
n
< , and
n+1
<
n
(n < ), a contradiction.
By slightly re-ordering
+
we can state Lemma 3 as follows. If C
+
is a closed
unbounded set, then S(C) =

{[, ( + 1)) : C} is
+
-chromatic. Put, for < ,
C
+
a club set, S

(C) =

{ + : C}. If, for every < there is some closed
unbounded C

that S

(C

) is -chromatic, then for C =

{C

: < }, S(C) is the union

of at most graphs, each -chromatic, a contradiction.
There is, therefore, a < such that S

(C) is
+
-chromatic whenever C is a closed
unbounded set. Mapping + to we get a graph on
+
, order-isomorphic to a subgraph
of the original graph which is
+
-chromatic on every closed unbounded set. From now on
we assume that our original graph G has this property.
We are going to build a model M =
+
; <, , G, . . . by adding countably many new
functions.
For n, e as in the Denition, a rst order formula, let G
n,e

be the following graph.

The vertex set is V

= {x
0
, . . . , x
n
: x
0
< < x
n
, M |= (x
0
, . . . , x
n
)} and x
0
, . . . , x
n
,
y
0
, . . . , y
n
are joined, if x
0
= y
0
, {x
1
, . . . , x
n
}, {y
1
, . . . , y
n
} interlace by e, and nally
{min(x
n
), min(y
n
)} G. We introduce a new quantier Q
n,e
with Q
n,e
meaning that
the above graph, G
n,e

is
+
-chromatic. If, however, Chr(G
n,e

) , we add a good -
coloring to M. We also assume that M is endowed with Skolem functions.
Lemma 4. There exist n, e and
1
< <
n
<
+
such that t(
i
) <
i+1
if t :
+

+
is a function in M and if x
0

1
, x
i
[
i
,
i+1
) (1 i < n), x
n
[
n
,
+
), min(x
i
) =
i
,
and is a formula, M |= (x
0
, . . . , x
n
), then M |= Q
n,e
.
Proof. Assume that the statement of the Lemma does not hold, i.e., for every n, e,

1
, . . . ,
n
there exist x
0
, . . . , x
n
contradicting it.
Let, for <
+
, B

+
be a countable set such that B

, and if n, e,

1
, . . . ,
n
B

are given , then a counter-example as above is found with x

0
, . . . , x
n

B

. We require that B

be Skolem-closed. Let B
+

be the ordinal closure of B

, B
+

=
3
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


{(, ) :

} be the increasing enumeration, = (,

). Let {M

: <
+
} be a
continuous, increasing chain of elementary submodels of M such that

= M

+
<
+
.
Clearly, C = {

: <
+
} is a closed, unbounded set. We take a coloring of the sets
{B
+

: <
+
} by colors that satises Lemma 2, if , get the same color then
the structures (B
+

; B

, M) and (B
+

; B

, M) are isomorphic and we also require that if

x
0
, . . . , x
n
B

and y
0
, . . . , y
n
B

are in the same positions, i.e., are mapped onto

each other by the order isomorphism between B

and B

and (x
0
, . . . , x
n
) is colored by
the -coloring of G
n,e

, then (y
0
, . . . , y
n
) is also colored and gets the same color. All this
is possible, as

0
= . We also assume that our coloring satises Lemma 1 with some
functions {t
n
: n < } that B
+

= {t
n
() : n < }.
As G is
+
-chromatic on C, there are < , both in C, joined in G, getting the
same color. By our conditions, B
+

B
+

is initial segment in both, and beyond that they

do not even intersect into the same complementary interval of C. As our structures are
isomorphic, this holds for B

, B

, as well.
We now let B
+

=

{B
+

(i) : i < i

}, B
+

=

{B
+

(i) : i < i

} be the ordered
decompositions given by the following equivalence relations. For x, y B
+

, x y, x y
if either [x, y] B
+

= or [x, y] B
+

[x, y] B
+

. Similarly for B
+

. By Lemma 2,
B
+

B
+

= B
+

(0) = B
+

(0).
Lemma 5. i

, i

are nite.
Proof. Otherwise, as B
+

, B
+

are ordinal closed, = min

_
B
+

()
_
= min
_
B
+

()
_
is in
B
+

B
+

, so B
+

(0), a contradiction.
Enumerate
_

: there is a 0 < i < such that either (, ) = min

_
B
+

(i)
_
or (, ) = min
_
B
+

(i)
_
_

_
as
1
<
2
< <
n
. By Lemma 1, if < , = min(B
+

(i

1)), = min(B
+

(i

1)),
B
+

(i

1) < B
+

(i

1). So
n
=

. We let
i
= (,
i
),
i
= (,
i
). If
i
=
min(B
+

(j)) then
i
B

and, by isomorphism,
i
B

. We show that for every i < n,

t M, t(
i
) <
i+1
and t(
i
) <
i+1
. As (B
+

; B

, M) and (B
+

; B

, M) are isomorphic,
for every i it suces to show this either for
i
or for
i
. For i = n 1 this follows from
the fact that (as well as ) is from C. If i < n then either
i1
and
i
are separated by
an element of B
+

or vice versa. Assume the former. Then, by Lemma 2,

i1
and
i
are
in dierent intervals of C so necessarily t(
i1
) <
i
holds.
Let e be the interlacing type of {
i
: 1 i n}, {
i
: 1 i n}. By our indirect
assumption, there are a formula , x
i
, y
i
(0 i n) in the same position in B

, B

such that x
i
[
i
,
i+1
), y
i
[
i
,
i+1
) etc, and M |= (x
0
, . . . , x
n
) (y
0
, . . . , y
n
)
and (x
0
, . . . , x
n
), (y
0
, . . . , y
n
) are joined in G
n,e

, but they get the same color in the good

coloring of G
n,e

, a contradiction which proves Lemma 4.

4
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


Now x n, e, and
1
< <
n
<
+
as in Lemma 4. We call a formula dense
if there exist x
0

1
, x
i
[
i
,
i+1
) (1 i < n), x
n
[
n
,
+
), min(x
i
) =
i
such
that M |= (x
0
, . . . , x
n
). If H = (V, <, U, X, h
1
, . . . , h
n
) K
n,e
, V = {0, 1, . . . , s}, a
-rich copy of H is some string (y
0
, . . . , y
s
) such that y
0
< < y
s
, if {i, j} X then
{min(y
i
), min(y
j
)} G and for every v U, M |= (y
0
, y
h
1
(v)
, . . . , y
h
n
(v)
).
Lemma 5. For every H K
n,e
if is dense there is a -rich copy of H in G.
Lemma 6. For every H = (V, <, U, X, h
1
, . . . , h
n
) K
n,e
, q U, if is dense, there is a
-rich copy (y
0
, . . . , y
s
) of H such that min(y
h
i
(q)
) =
i
for 1 i n.
We notice that Lemma 5 obviously concludes the proof of Theorem 1 and Lemma
6 clearly implies Lemma 5. Also, they trivially hold for H K
n,e
0
. We prove these two
Lemmas simultaneously.
Claim 1. If Lemma 5 holds for some H then Lemma 6 holds for H, as well.
Proof. Assume that Lemma 5 holds for H = (V, <, U, X, h
1
, . . . , h
n
) K
n,e
and for
any dense but Lemma 6 fails for a certain q U and a dense . This statement
can be written as a formula (
1
, . . . ,
n
). As is dense, M |= (x
0
, . . . , x
n
) for some
appropriate strings, so also M |= (x
0
, . . . , x
n
) where = (min(x
1
), . . . , min(x
n
)).
As is dense, by Lemma 5 there is a -rich copy (y
0
, . . . , y
s
) of H but then M |=
(min(y
h
1
(q)
), . . . , min(y
h
n
(q)
)), a contradiction.
Claim 2. If Lemma 6 holds for H = (V, <, U, X, h
1
, . . . , h
n
) and x V then Lemma 5
holds for H

= H +
x
H.
Proof. Select q U such that x = h
i
(q) for some 1 i n. By Lemma 6, there is a
-rich copy (y
0
, . . . , y
s
) of H such that min(y
h
i
(q)
) =
i
for 1 i n. As
i
> t(
i1
)
holds for every function t in the skolemized structure M there are -rich copies of H which
agree with this below x but their x elements are arbitrarily high. We can, therefore, get a
-rich copy of H

.
Claim 3. If Lemma 6 holds for H = (V, <, U, X, h
1
, . . . , h
n
) and y U then Lemma 5
holds for H

= H
y
H.
Proof. Let (y
0
, . . . , y
s
) be a -rich copy of H such that min(y
h
i
(q)
) =
i
for 1 i
n. The elements in the (y
0
, . . . , y
s
) string can be redistributed as (x
0
, . . . , x
n
) such that
min(x
i
) =
i
and then the fact that they form a -rich copy of H can be written as
M |= (x
0
, . . . , x
n
) for some formula . As is dense, by Lemma 4, M |= Q
n,e
holds,
so there are two strings, (x
0
, . . . , x
n
) and (x

0
, . . . , x

n
) both satisfying , interlacing by e,
and {min(x
n
), min(x

n
)} G. This, however, gives a -rich copy of H

.
Theorem 2. If n, e are as in the Denition, is an innite cardinal,
<
= , then there
exists a
+
-c.c., < -closed poset Q = Q
n,e,
which adds a
+
-chromatic graph of cardinal

+
all whose nite subgraphs are subgraphs of some element of K
n,e
.
Proof. Put q = (V, U, X, h
1
, . . . , h
n
) Q if V [
+
]
<
, U V , X [V ]
2
, every
h
i
is a function U V with h
1
(x) < < h
n
(x) = x for x U and every nite
5
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


substructure of (q, <) is a substructure of some element of K
n,e
. Order Q as follows. q

=
(V

, U

, X

, h

1
, . . . , h

n
) q = (V, U, X, h
1
, . . . , h
n
) i V

V , U = U

V , X = X

[V ]
2
,
h

i
h
i
(1 i n). Clearly, (Q, ) is < -closed.
Lemma 7. (Q, ) is
+
-c.c.
Proof. By the usual -system arguments it suces to show that if the conditions q
i
=
(V V
i
, U
i
, X
i
, h
i
1
, . . . , h
i
n
) are order isomorphic (i < 2), V < V
0
< V
1
then they are
compatible. A nite subset of V V
0
V
1
can be included into some s s
0
s
1
where s
0
and s
1
are mapped onto each other by the isomorphism between q
0
and q
1
. By condition,
q|ss
0
is a substructure of some structure H K
n,e
. But then q|ss
0
s
1
is a substructure
of an edgeless amalgamation of H.
If G Q is generic then Y =

{X : (V, U, X, . . .) G} is a graph on a subset of
+
all whose nite subgraphs are subgraphs of some member of K
n,e
. The following Lemma
clearly concludes the proof of the Theorem.
Lemma 8. Chr(Y ) =
+
.
Proof. Assume, toward a contradiction, that 1 forces that f :
+
is a good coloring
of Y . Let M
1
M
2
M
n
be elementary submodels of (H((2

)
+
); Q, f, . . . , ) with
M
0
, [M
i
]
<
M
i
. Put
i
= M
i

+
<
+
. Notice that cf(
i
) = . Let p

=
(V

, U

, X

, h

1
, . . . , h

n
) where V = {
1
, . . . ,
n
}, U = {
n
}, X = , h
i
(
n
) =
i
. Choose
p = (V, U, X, h
1
, . . . , h
n
) p

forcing f(
n
) = for some < . Let
n
(, x
1
, . . . , x
n
) be
the following formula. is an order isomorphism V
+
, (
i
) = x
i
and (p) forces
that f(x
n
) = . Let
n+1
= . For 0 i < n dene
i
(, x
1
, . . . , x
i
) meaning that
: V
i+1

+
is order preserving and there are arbitrarily large x
i+1
<
+
and

such that
i+1
(

, x
1
, . . . , x
i+1
) holds.
Claim 4.
i
(id|V
i+1
,
1
, . . . ,
i
) for 0 i n.
Proof. This is obvious for i = n. If
i
(id|V
i+1
,
1
, . . . ,
i
) fails, then, by denition,
there would be a bound for the possible x
i+1
values for which
i+1
(

,
1
, . . . ,
i
, x
i+1
)
holds for some

id|V
i+1
. But then this bound is smaller than
i+1
so
i+1
fails,
too.
Returning to the proof of Lemma 8, we dene the following function t. Let {a
1
, . . . , a
n
},
{b
1
, . . . , b
n
} be as in the denition of the one-edge amalgamation. Put, for 1 i n,
t(i) = j i b
j1
< a
i
< b
j
where b
0
= 0, b
n+1
= 2n +1. Set
0
= id|V
1
. We know that

0
(
0
) holds. By induction on 1 i n select
i
in such a way that
i+1

i
, if we let

i
(
i
) =

i
then
i
(
i
,

1
, . . . ,

i
) holds and sup(V
t(i)
) <

i
and Ran(
i
) <
t(i)
. This is
possible as M
1
, . . . , M
n
are elementary submodels. Finally,
n
(p) is a condition interlacing
with p by e and it forces that f(

n
) = . Now if we take the union of them plus the edge
{
n
,

n
} then an argument as in Lemma 7 shows that we get a condition which forces a
contradiction.
Theorem 3. If GCH holds there is a cardinal, conality, and GCH preserving (class)
notion of forcing in which for every n, e, and regular there is a
+
-chromatic graph
on
+
all whose nite subgraphs are subgraphs of some elements of K
n,e
.
6
3
4
6


r
e
v
i
s
i
o
n
:
1
9
9
4
-
0
1
-
3
1







m
o
d
i
f
i
e
d
:
1
9
9
4
-
0
2
-
0
3


Proof. For regular let Q

be the product of Q
n,e,
of Theorem 2 with nite supports
if = , and complete supports otherwise. Notice that Q

is a
+
-c.c. notion of forcing
of cardinal
+
. For singular let Q

be the trivial forcing.

Our notion of forcing is the Easton-support limit of the Q

s, i.e., the direct limit of

P

where P
+1
= P

with Q

dened in the ground model. For limit, p P

i
p() Q

for all < , and |Dom(p) | < for regular.

Given n, e, and as in the statement of the Theorem, the extended model can be
thought as the generic extension of some model rst with Q
n,e,
then with P

which is of
cardinal so it cannot change the chromatic number of a graph from
+
to .
Assume that the conality of some ordinal collapses to a regular . P splits as
P

R where R is -closed, |P

| and Q

is
+
-c.c., so in fact the
+
-c.c. P
+1
changes the conality of which is impossible. This also implies that no cardinals are
collapsed.
If is regular, all subsets of are added by the
+
-c.c. P
+1
of cardinal
+
so 2

remains
+
. If is singular we must bound
cf()
. The sets of size cf() are added by
P
cf()+1
so we can bound the new value of
cf()
by
cf()
+
=
+
.
References
[1] P.Erd os, A.Hajnal: On chromatic number of graphs and set-systems, Acta Math.
Acad. Sci. Hung. 17 (1966), 6199.
[2] P.Erd os, A.Hajnal, S.Shelah: On some general properties of chromatic number, in:
Topics in Topology, Keszthely (Hungary), 1972, Coll. Math. Soc. J. Bolyai 8, 243
255.
[3] P.Erd os, R.Rado: Partition relations connected with the chromatic number of graphs,
Journal of London Math. Soc. 34 (1959), 6372.
[4] P.Erd os, R.Rado: A construction of graphs without triangles having pre-assigned
order and chromatic number, Journal of London Math. Soc. 35 (1960), 445448.
[5] A.Hajnal, P.Komj ath: What must and what need not be contained in a graph of
uncountable chromatic number ?, Combinatorica 4 (1984), 4752.
[6] P.Komj ath: The colouring number, Proc. London Math. Soc. 54 (1987), 1-14.
[7] W.Taylor: Atomic compactness and elementary equivalence, Fund. Math. 71 (1971),
103112.
[8] W.Taylor: Problem 42, Comb. Structures and their applications, Proc. of the Calgary
International Conference, 1969.
Peter Komj ath
Department of Computer Science
E otv os University
Budapest, M uzeum krt. 68
1088, Hungary
e-mail: kope@cs.elte.hu
Saharon Shelah
Institute of Mathematics
Hebrew University
Givat Ram
Jerusalem, Israel
e-mail: shelah@math.huji.ac.il
7