(

7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE
BIPARTITE GRAPH
SH706
SAHARON SHELAH
Abstract. For cardinals , , we consider the class of graphs of cardinality
which has no subgraph which is (, )-complete bipartite graph. The question
is whether in such a class there is a universal one under (weak) embedding. We
solve this problem completely under GCH. Under various assumptions mostly
related to cardinal arithmetic we prove non-existence of universals for this
problem. We also look at combinatorial properties useful for those problems
concerning -dense families.
Date: May 5, 2010.
Key words and phrases. Set theory, graph theory, universal, bipartite graphs, black boxes.
The author thanks Alice Leonhardt for the beautiful typing. This research was supported by
the United States-Israel Binational Science Foundation. Done 1,2 - 5/99; 3 in ?/99
First Typed - 12/16/99.
1
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


2 SAHARON SHELAH
Anotated Content
0 Introduction
1 Some no we-universal
[We dene Pr(, ) and using it gives sucient conditions for the non-
existence of we-universal in H
,,
mainly when =
+
. Also we give
sucient conditions for no ste-universal when =

, 2

.]
2 No we-universal by Pr(, ) and its relatives
[We give ner sucient conditions and deal/analyze the combinatorial prop-
erties we use; Pr(, ) says that there are partial functions f

from to
for < , which are dense, = otp(Dom(f

)) and > otp(Dom(f

)
Dom(f

)) for ,= .]
3 Complete characterization under G.C.H.
[Two theorems cover G.C.H. - one assume is strong limit and the other
assume =
+
= 2

+2
<
= . Toward this we prove mainly some results
on existence of universe.]
4 More accurate properties
0. Introduction
On the problem of among graphs with nodes and no complete subgraph with
nodes, is there a universal one (i.e. under weak embedding) is to a large extent
solved in Komjath-Shelah [KoSh:492], see more there. E.g. give a complete solution
under the assumption of GCH.
Now there are some variants, mainly for graph theorists embedding, i.e. a one
to one function mapping an edge to an edge, called here weak or we-embedding;
for model theorist an embedding also maps a non-edge to a non-edge, call strong or
ste-embedding. We have the corresponding we-universal and ste-universal. We deal
here with the problem among the graphs with nodes and no complete (, )-
bipartite sub-graph, is there a universal one?, see below on earlier results. We
call the family of such graphs H
,,
, and consider both the weak embedding (as
most graph theorists use) and the strong embedding. Our neatest result appears
in section 3 (see 3.5, 3.15).
{0.A}
Theorem 0.1. Assume
0
.
1) If is strong limit then: there is a member of H
,,
which is we-universal
(= universal under weak embedding) i there is a member of H
,,
which is ste-
universal (= universal under strong embeddings) i cf() cf() and ( <
cf() < cf()).
2) If = 2

=
+
and = 2
<
, then
(a) there is no ste-universal in H
,,
(b) there is we-universal in H
,,
i =

and = .
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH706 3
We give many sucient conditions for the non-existence of universals (mainly we-
universal) and some for the existence, for this dealing with some set-theoretic prop-
erties. Mostly when we get no G H
,,
is we/ste-universal we, moreover, get
no G H
,,
is we/ste-universal among the bipartite ones. Hence we get also
results on families of bi-partite graphs. We do not look at the case <
0
here.
Rado has proved that: if is regular >
0
and 2
<
= , then H
,,1
has a ste-
universal member (a sucient condition for G

being ste-universal for H

,,1
is: for
any connected graph G with < nodes, of the components of G are isomorphic
to G). Note that G H
,,1
i G has nodes and the valency of every node is < .
Erd os and Rado (see [EH74], in Problem 74) ask what occurs, under GCH to say

. By [Sh:26, 3.1] if is strong limit singular then there is a ste-universal graph

in H
,,1
.
Komjath and Pach [KP84] prove that
1
no universal in H
1,1,0
, this holds
also for H

+
,
+
,
when
S

holds; subsequently the author showed that 2

=
+
suce (Theorem 1 there). Then Shar (see [Sha01, Th.1]) presents this and proves
the following:
([Sha01, Th.2]): if = cf(),
S

and there is a MAD family on []

of cardinality
, then H

+
,,
has no we-universal.
([Sha01, Th.3]): if 2

and there is / []

of cardinality such that no

B []

is included in of them, then H

,,
has no ste-universal member.
([Sha01, Th.4]): if 2

and
S

then H
,,
has no ste-universal members.
Here we characterize H
,,
has universal under GCH (for weak and for strong
embeddings). We also in 1.1 prove
S

no we-universal in H

+
,
+
,
(compared
to [Sha01, Th.2], we omit his additional assumption no MAD / []

, [/[ = );
in 1.2 we prove more. Also (1.5) =

no universal under strong

embedding in H
,,
(compared to [Sha01, Th.3] we omit an assumption).
Lately some results for which we originally used [Sh:460] now instead use [Sh:829],
[Sh:922] which gives stronger results.

Notation 0.2.
We use , , , , for cardinals (innite if not said otherwise)
We use , , , , , , i, j for ordinals, for limit ordinals
For = cf() < , S

= < : cf() =
[A]

= B A : [B[ = . We use G for graphs and for bipartite graphs;

see below Denition 0.3(1), 0.4(1), it will always be clear from the context
which case we intend.
{0.1}
Denition 0.3. 1) A graph G is a pair (V, R) = (V
G
, R
G
), V a non-empty set,
R a symmetric irreexive 2-place relation on it. We call V the set of nodes of G
and [V [ is the cardinality of G, denoted by |G|, and may write G instead of
V
G
.
Let E
G
= , : R
G
, so we may consider G as (V
G
, E
G
).
2) We say f is a strong embedding of G
1
into G
2
(graphs) if :
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


4 SAHARON SHELAH
(a) f is a one-to-one function from G
1
into G
2
; pedantically from V
G1
into
V
G2
(b)
st
for , G
1
we have
R
G1
f()R
G2
f().
3) we say f is a weak embedding of G
1
into G
2
if
(a) above and
(b)
we
for , G
1
we have
R
G1
f()R
G2
f().
4) The -complete graph K

is the graph (, R) were R ,= or any graph

isomorphic to it.
{0.2}
Denition 0.4. 1) G is a bipartite graph means G = (U, V, R) = (U
G
, V
G
, R
G
)
where U, V are disjoint non-empty sets, R U V . For a bipartite graph G, we
would like sometimes to treat as a usual graph (not bipartite), so let G as a graph,
G
[gr]
, be (U
G
V
G
, (, ) : , V U and R
G
R
G
). The cardinality
of G is ([U
G
[, [V
G
[) or [U
G
[ +[V
G
[.
2) We say f is a strong embedding of the bipartite graph G
1
into the bipartite
graph G
2
if :
(a) f is a one-to-one function from U
G1
V
G1
into U
G2
V
G2
mapping U
G1
into U
G2
and mapping V
G1
into V
G2
(b) for (, ) U
G1
V
G1
we have
R
G1
f()R
G2
f().
3) We say f is a weak embedding of the bipartite graph G
1
into the bipartite graph
G
2
if
(a) f is a one-to-one function from U
G1
V
G1
into U
G2
V
G2
, f mapping U
G1
into U
G2
and mapping V
G1
into V
G2
(b) for (, ) U
G1
V
G1
we have
R
G1
f()R
G2
f().
4) In parts (2), (3) above, if G
1
is a bipartite graph and G
2
is a graph then we
mean G
[gr]
1
, G
2
.
5) The (, )-complete bipartite graph K
,
is (U, V, R) with U = i : i < , V =
+i : i < , R = (i, +j) : i < , j < , or any graph isomorphic to it.
{0.3}
Denition 0.5. 1) For a family H of graphs (or of bipartite graphs) we say G is
ste-universal [or we-universal] for H i every G

H can be strongly embedded [or

weakly embedded] into G.
2) We say H has a ste-universal (or we-universal) if some G H is ste-universal (or
we-universal) for H.
{0.4}
Denition 0.6. 1) Let H
,,
= H
gr
,,
be the family of graphs G of cardinality
(i.e. with nodes) such that the complete (, )-bipartite graph cannot be weakly
embedded into it; gr stands for graph.
2) Let H
bp

,,
be the family of bipartite graphs G of cardinality

such that the
complete (, )-bipartite graph cannot be weakly embedded into it. If

= (, )
we may write (similarly in (3)); bp stands for bipartite.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH706 5
3) Let H
sbp

,{,}
= H
sbp

,,
be the family of bipartite graphs G of cardinality

such
that K
,
(the (, )-complete bipartite graph) and K
,
(the (, )-complete bi-
partite graph) cannot be weakly embedded into it; sbp stands for symmetrically
bipartite.
4) H

= H
gr

is the family of graphs of cardinality and H

bp

is the family of bipartite

graphs of cardinality

.
{0.5}
Observation 0.7. 1) The following are equivalent:
(a) in H
sbp
,,
there is a we-universal
(b) in G
[gr]
: G H
sbp
,,
there is a we-universal.
2) Similarly for ste-universal.
3) If G is ste-universal for H then it is we-universal for H (in all versions).
4) Assume that for every G H
,,
there is a bipartite graph from H
,,
not x-
embeddable into it, then in H
bp
,,
and in H
bp
,,
and in H
sbp
,,
there is no x-universal
member; for x we, ste.
Proof. (1) (a) (b): Trivially.
(b) (a): Assume G is we-universal in G
[gr]
: G H
sbp
,,
and let A
i
: i < i

)
be its connectivity components. Let A
i
be the disjoint union of A
i,0
, A
i,1
with
no G-edge inside A
i,0
and no G-edge inside A
i,1
(exists as G = G
[gr]

for some
G

H
sbp
,,
, note that A
i,0
, A
i,1
is unique as G A
i
is connected). Let A
m,
i,
for i < i

, < 2, m < 2, < be pairwise disjoint sets with [A

m,
i,
[ = [A
i,k
[ when
m+ = k mod 2. Let G

be the following member of H

bp
,,
: let U
G

be the disjoint
union of A
,
i,
for i < i

, < 2, < and V

G

be the disjoint union of A

1,
i,
for
i < i

, < 2, < and R

G

= R

i,
: i < i

, < 2 where R

i,
are chosen such
that (A
0,
i,0
, A
1,
i,1
, R

i,0
)

= (A
i,0
, A
i,1
, R
G
A
i,0
A
i,1
)

= (A
0,
i,1
, A
1,
i,0
, R

i,1
). Easily
G

H
bp
,,
is we-universal.
2) The same proof.
3), 4) Easy.
0.7
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


6 SAHARON SHELAH
1. Some no we-universal
We show that if =

then in H
,,
there is no ste-universal
graph (in 1.5); for we-universal there is a similar theorem if =
+
, Pr(, ), see
1.4 + Denition 1.2, (this holds when =

= cf(),
S

).
{1.1}
Claim 1.1. Assume is regular and
S

(see Denition 1.2 below). Then there

is no we-universal in H

+
,
+
,
.
{1.2}
Denition 1.2. 1) For regular < let S

= < : cf() = cf().

2) For regular and stationary subset S of let
S
means that for some

A = A

:
S, limit) we have
(a) A

is an unbounded subset of
(b) if A is an unbounded subset of then for some (equivalently stationarily
many) S we have A

A.
2A) For < let

,
mean that for some family / []

of cardinality we have
(B []

)(A /)(A B).

3) Pr(, ) for cardinals > means that some T exemplies it, which means
(a) T is a family of functions
(b) every f T is a partial function from to
(c) if f T then = otp(Dom(f)) and f strictly increasing
(d) f ,= g T > [Dom(f) Dom(g)[
(e) if g is a partial (strictly) increasing function from to such that Dom(g)
has cardinality , then g extends some f T.
4) Pr

(, ) is dened similarly for a limit ordinal but clauses (c) + (d) are replaced
by
(c)

if f T then = otp(Dom(f)) and f is one to one

(d)

if f ,= g T then Dom(f) Dom(g) is a bounded subset of Dom(f) and

of Dom(g).
{1.2A}
Observation 1.3. 1) We have
(i) = cf() and
S

Pr(, )
(ii) Pr

(, ) Pr(, )

,
(iii) for any cardinal we have Pr(, ) Pr

(, ).
2) If we weaken clause (c) of 1.2(3) to
(c)

f T [Rang(f)[ = = [Dom(f)[
we get equivalent statement (can combine with 1.3(3)).
3) The one to one in Denition 1.2(4), clause (c)

is not a serious demand, that

is, omitting it we get an equivalent denition.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH706 7
Proof. Easy.
1) E.g., clause (iii) holds because for any one to one f : Ord, for some A []

the function f A is strictly increasing (note it rst to regular ).

2) Left to the reader.
3) Why? Let pr: be 1-to-1 onto and pr
1
,pr
2
: be such that
= pr(pr
1
(), pr
2
()).
Let T be as in the Denition 1.2(4) old version. Then pr
1
f : f T will
exemplify the new version, i.e. without the 1-to-1.
For the other direction, just take f T : f is one to one.
1.3
Proof. Proof of 1.1 It follows from 1.4 proved below as
S

easily implies Pr(

+
, )
for regular (see 1.3(1)).
1.1
{1.3}
Claim 1.4. If Pr(, ), so > then in H
,
+
,
there is no we-universal.
Moreover, for every G

H
,
+
,
there is a bipartite G H
,
+
,
of cardinality
not we-embeddable into it.
Proof. Let G

be a given graph from H

,
+
,
; without loss of generality V
G

= .
For any A let
()
0
(a) Y
0
A
=: < : is G

-connected with every A

(b) Y
2
A
=: < : is G

-connected with members of A

(c) Y
1
A
=: < : is G

-connected with every A except possibly

< of them.
Clearly
(d) A B Y
0
A
Y
0
B
and Y
2
A
Y
2
B
and Y
1
A
Y
1
B
(e) [A[ Y
0
A
Y
1
A
Y
2
A
.
We now note
()
1
if A []

then [Y
0
A
[ .
[Why? Otherwise we can nd a weak embedding of the (,
+
)-complete bipartite
graph into G

]
()
2
if A []

then [Y
1
A
[ .
[Why? If not choose pairwise disjoint subsets A
i
of A for i < each of cardinality
, now easily Y
1
A
[i < : / Y
0
Ai
[ < so Y
1
A

i<
Y
0
Ai
hence if [Y
1
A
[ >
then for some i < , Y
0
Ai
has cardinality > , contradiction by ()
1
.]
Let T = f

: < exemplify Pr(, ). Now we start to choose the bipartite

graph G:

0
U
G
= , V
G
= , R
G
=

<
R
G

and R
G

(, (, )) : < and
Dom(f

) and < U
G
V
G
where R
G

is chosen below; we let

G

= (U
G
, V
G
, R
G

).
Now

1
G is a bipartite graph of cardinality

2
the (
+
, )-complete bipartite graph ( H
bp
(
+
,)
) cannot be weakly embed-
ded into G.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


8 SAHARON SHELAH
[Why? As for any (, ) V
G
the set < : R
G
(, ) is equal to Dom(f

)
which has cardinality which is <
+
; note that we are speaking of weak embedding
as bipartite graph, side preserving]

3
the (,
+
)-complete bipartite graph cannot be weakly embedded into G
provided that for each < ,

K
,
+ cannot be weakly embedded into (U
G
, V
G
, R
G

).
[Why? Let U
1
U
G
, V
1
V
G
have cardinality ,
+
respectively and let V

1
=
: (, ) V
1
for some . If [V

1
[ 2 choose 9
1
,
1
), (
2
,
2
) V
1
such that

1
,=
2
, so < : (, (

)) R
G
for = 1, 2 include U
1
hence by the
denition of R
G
we have [Dom(f
1
) Dom(f
2
)[ [U
1
[ = , but
1
,=
2

[Dom(f
1
) Dom(f
2
)[ < by clause (d) of Denition 1.2(3) - think.]

4
G cannot be weakly embedded into G

provided that for each < :

there is no weak embedding f of (U

G
, V
G
, R
G

) into G

extending f

.
[Why? Assume toward contradiction that f is a one-to-one mapping from U
G
V
G
into V
G

= mapping edges of G to edges of G

. So f U
G
is a one-to-one
mapping from to hence by the choice of T = f

: < to witness Pr(, )

see clause (e) of Denition 1.2(3) there is such that f

f U
G
. So clearly
R
G
(, ) and Dom(f

) implies f(), f(, ) = f

(), f((, )) E
G

,
hence (, (, )) R
G

f(), f((, )) E
G

. This clearly contradicts

4

which we are assuming.]

So we are left with, for each < , choosing R

(, (, ) : Dom(f

), <
to satisfy
3

+
4

. The proof splits to cases, xing .

Let us denote B

= Rang(f

), A

= Dom(f

) for = 0, 1 we let A

=:
A

: otp(A

) = mod 2 and B

=: f

() : A

.
Case 1: Y
2
B
has cardinality for some B [B

.
Choose such B = B

and let A

= A

: f

() B

. There is a sequence

C = C

: <
+
), C

[]

such that < [C

[ < .
Let R

= (, (, )) : <
+
, A

and otp( A

) C

. Now
4

holds
because if f is a counter-example, then necessarily by the pigeon-hole principle
for some <
+
we have f((, )) / Y
2
B

, but clearly (, ) is G

-connected
to members of A

hence f((, )) is G

-connected to members of B

hence
f((, )) Y
2
B

and we get a contradiction. Also

3

holds as < <

+

[C

[ < .
So we may assume, for the rest of the proof, that

5
[Y
2
B
[ > for every B [B

.
Case 2: For some < 2, [Y
2
B

[ > and for some Z:

(i) Z Y
2
B

Y
1
B

(ii) [Z[
(iii) for every
0
(Y
2
B

Y
1
B

)Z there is
1
Z such that > [ B

:
is G

-connected to
0
but is not G

-connected to
1
[.
So we choose such = () < 2, Z = Z

and then we choose a sequence B

,
:
Z

) such that:
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH706 9

6
(a) B
,
is a subset of B

(b) [B
,
[ =
(c)
1
,=
2
Z B
,1
B
,2
=
(d) is not G

-connected to any B
,
(this is possible as Z

/ Y
1
B

).
Now we can nd a sequence C
,
: <
+
) satisfying

7
() C
,
B

() [C
,
[ = moreover Z

[C
,
B
,
[ =
() for < we have [C
,
C
,
[ <
(e.g. if = cf() >
0
by renaming B

= , each B
,
is stationary, choose
nonstationary C
,
inductively on ; if > cf() reduce it to construction on
regulars, if =
0
like = cf() >
0
).
Lastly we choose R

= (, (, )) : A

, <
+
and f

() C
,
.
Now
3

is proved as in the rst case, as for

4

, if f is a counter-example then
clearly for <
+
, f((, )) Y
2
B
, so as [Y
2
B
[ > by
5
and [Z

[ and
[Y
1
B

[ (by ()
2
) necessarily for some <
+
,
0
=: f((, )) Y
2
B

Y
1
B

.
Let
1
Z

be as guaranteed in clause (iii) in the present case. Now

0
is G

-
connected to every member of C
,
as
0
= f((, )). Hence
0
is G

-connected
to members of B
,1
(see clause () above and the choice of R

); but
1
is not
G

-connected to any member of B

,1
(see clause (d) above). Reading clause (iii),
we get contradiction.
Case 3: Neither Case 1 nor Case 2.
Recall that is xed. For 0, 1 we choose Z

,
by induction on <
+
,
such that

8
(a) Z

,
a subset of Y
2
B

of cardinality
(b) Z

,
is increasing continuous in
(c) Y
1
B

,0
(d) if = + 1 then there is

,
Z

,
Z

,
Y
1
B

such that
for every

,
Y
1
B

we have
= [ B

: is G

-connected to

,
but not to

[
(e) if = + 1 and Z

,
hence = [ B

: is connected to [
(e.g. =

,
), then Y
1
{B

: is G

-connected to }
is included in
Z

,
(f) Z
0
,
(Y
2
B
0

Y
2
B
1

) = Z
1
,
(Y
2
B
0

Y
2
B
1

).
Why possible? For clause (c) we have [Y
1
B

[ by ()
2
, for clause (d) note that
not Case 2 trying Z

,
Y
1
B

as Z, and for clause (e) note again [Y

1
{B

: is G

-connected to

,
}
[
by ()
2
.
Having chosen Z

,
: <
+
, < 2), we let
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


10 SAHARON SHELAH
R

= (, (, )) : for some < 2 we have:

A

and f

() is G

-connected to =

,2+
so <
+
.
Now why
3

holds? Otherwise, we can nd A A

, [A[ = and B
+
, [B[ =
+
such that A and B R

(,
()
,
) where = () mod 2, so for some
< 2 we have [A A

[ = , and let
A

= f

() : A A

.
Easily [B[ =
+
and [A

[ = and A

and B R
G

,
, contradiction to
K
,
+ is not weakly embeddable into G

.
Lastly, why
4

holds? Otherwise, letting f be a counterexample, let <

+
and < 2. Clearly f((, )) is G

-connected to every

which is G

-
connected to

,2+
hence f((, )) cannot belong to Z

,2+
Y
1
B

(by the demand

in clause (d) of
8
), but it has to belong to Z

,2++1
(by clause (e) of
8
), so
f((, )) (Z

,2++1
Z

,2+
) Y
1
B

. Putting together = 0, 1 we get f((, ))

((Z
0
,2+1
Z
0
,2
) Y
1
B
0

) ((Z
1
,2+2
Z
1
,2+1
) Y
1
B
1

) hence f((, )) Y
1
B
0

Y
1
B
1

,
but [Y
1
B

[ <
+
; as this holds for every <
+
this is a contradiction to f is one
to one.
1.4
{1.4}
Claim 1.5. 1) Assume 2

and =

(e.g. = 2

).
Then in H
,,
there is no ste-universal (moreover, the counterexamples are
bipartite).
2) Assume Pr(, ), , 2

. Then the conclusion of (1) holds.

Proof. 1) By the simple black box ([Sh:300, Ch.III,4]) or [Sh:e, Ch.VI,1], i.e.
[Sh:309])
there is

f = f

:

), f

a function from i : i < into such

that for every f :
>
for some

we have f

f.
Let G

H
,,
and we shall show that it is not ste-universal in H
,,
, without
loss of generality V
G

= . For this we dene the following bipartite graph G:

1
(i) U
G
=
>
and V
G
=

(ii) R
G
= R
G

:

and f

is a one-to-one function where

R
G

( i, ) : i u

where u

is dened as follows

2
for

we choose u

such that if possible

()
,u
for no < do we have (i < )[f

( i)R
G

i u

].
If for every

for which f

is one to one for some u we have ()

,u
holds,
then clearly by we are done.
Otherwise, for this

, f

is one to one and: there is

u
< satisfying
(i < )(f

( i)R
G

u
i u) for every u . But then A

=: f

( 2i) : i <
and B

=:
u
: u and (i < )2i u form a complete (, 2

)-bipartite
subgraph of G

, contradiction.
2) The same proof.
1.5
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70611
2. No we-universal by Pr(, ) and its relatives
We dene here some relatives of Pr. Here Ps is like Pr but we are approximating
f : , and Pr
3
(, , , ) is a weak version of ( + )
||
(Denition 2.5);
we give sucient conditions by cardinal arithmetic (Claim 2.6, 2.8). We prove
more cases of no we-universal: the case limit (and Pr(, )) in 2.2, a case of
Pr

(,
+
) in 2.4. We also note that we can replace Pr by Ps in 2.9, and
strong limit singular of conality > cf() in 2.10.
{2.1}
Convention 2.1.
0
.
{2.2}
Claim 2.2. If is a limit cardinal and Pr(, ), then there is no we-universal
graph in H
,,
even for the class of bipartite members.
Proof. Like the proof of 1.4, except that we replace cases 1-3 by:
for every < we let R

= (, (, )) : Dom(f

) and < [Y
0
Dom(f)
[
+
.
Now
3

holds as [Y
0
Dom(f)
[ < (by ()
1
there) hence [Y
0
Dom(f)
[
+
< as is a
limit cardinal. Lastly
4

holds as for some we have f

f hence the function

f maps (, ) : < [Y
0
Dom(f)
[
+
into Y
0
Dom(f)
but f is a one to one mapping,
contradiction.
2.2
Recall
{pr.22}
Denition 2.3. For a cardinal and a limit ordinal , Pr

(, ) holds when for

some T:
(a) T a family of functions
(b) every f T is a partial function from to
(c) f T otp(Dom f) = , and f is one to one
(d) f, g T, f ,= g (Dom f) (Dom g) is a bounded subset of Dom(f) and
of Dom(g)
(e) if g : is a partial function, one to one, and [Dom g[ = , then g
extends some f T.
{2.4}
Claim 2.4. Assume
(a) Pr

(,

),

= , ordinal product
1
(b) =
+
.
Then there is no we-universal in H
gr
,,
even for the class of bipartite members.
Proof. Let G

H
,,
and we shall prove it is not we-universal; let without loss of
generality V
G

= .
Let T be a family exemplifying Pr

(,

), let T = f

: < let A

=
Dom(f

) and let it be
,,i
: i < , < such that [
,(1),i(1)
<
,(2),i(2)

(1) < (2) ((1) = (2) and i(1) < i(2))] and for i < let A
,i
=
,,i
: < ,
so clearly
()
1
A
,i
[]

and (
1
, i
1
) ,= (
2
, i
2
) [A
1,i1
A
2,i2
[ < .
1
this is preserved by decreasing
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


12 SAHARON SHELAH
[Why the second assertion? As
,,i
: < is an unbounded subset of A

(of
order type ).]
For (, i) let f
,i
= f

A
,i
let B
,i
= Rang(f
,i
) so [A
,i
[ = [B
,i
[ =
and let Y
0
,i
= < : is G

-connected to every member of B

,i
, so as
G

H
,,
clearly [Y
0
,i
[ < . As =
+
> , clearly for each < for some

< we have
2
: X

:= i < : [Y
0
,i
[

has cardinality . As

< also

:=
+

is <
+
= , so
+

= [X

[. We choose by induction on <

+

an
ordinal i

,
X

such that:
()
2
i

,
/ i

,
: < .
Recall that

is a successor, hence a regular cardinal. So if S

= <
+

:
cf() =

recalling

=
+

> [Y
0
,i

,
[ there is < such that (Y
0
,i

,
Y
0
,
:
< Y
0
,
: < ). Let g() be the rst ordinal having this property, so g
is a well dened function with domain S

. Clearly, g is a regressive funciton.

By Fodors lemma for some stationary S

, and for some B

of cardinality

we have S

Y
0
,i

,
Y
0
,j
: j < i

,
, in fact: B

= Y
0
,i

,
:
<

where g S

is constantly

is O.K., we can decrease B

but immaterial
here
Now
()
3
for ,= from S

there is no < such that:

is G

-connected to every B
,i

,
is G

-connected to every B
,i

,
is not in B

.
Let (, j) : j <
+

) list S

in an increasing order. Let G be the bipartite graph

U
G
=
V
G
=
R
G
=

(, (, )) : < and <

+

and A
,i

,(,2)
A
,i

,(,2+1)

1
G is a bipartite graph of cardinality

2
the (, )-complete bipartite graph ( H
bp
(,)
) cannot be weakly embedded
into G.
[Why
2
holds? So let ((1), (1)) ,= ((2), (2)) belongs to V
G
, the set set
U
G
: connected to ((1), (2)) and to ((2), (2)) is included in

(1),(2){0,1}
(A
(1),i

((1),2(1)+(1)

A
(2),i

((2),2(2)+(2)
) which is the union of four sets each of cardinal < (by ()
1
)
hence has cardinality < .]

3
the (, )-complete bipartite graph cannot be weakly embedded into G.
2
in fact by 2.2 without loss of generality is a successor cardinal, so without loss of generality

=
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70613
[Why? Toward contradiction assume U
1
U
G
, V
1
V
G
have cardinality ,
respectively and U
1
(, ) V
1
R(, ).
Let (, ) V
1
, clearly U
1
R
G
(, ) A
,i

,(,2)
A
,i

,(,2+1)
.
So U
1
A
i

,(,2)
A
i

,(,2+1)
.
Now if (
1
,
1
), (
2
,
2
) V
1
and
1
,=
2
then i, j < [A
1,i
A
2,j
[ <
by the choice of T, so necessarily for some

< we have V
1

. But if
(,
1
) ,= (,
2
) V
1
then (,
1
), (,
2
) has no common neighbour, contradiction.

4
there is no weak embedding f of G into G

.
[Why? Toward contradiction assume that f is such a weak embedding. By the
choice of T and f

: < ) we can choose < such that f

fU
G
. As
f is a weak embedding < < R
G
(, ) f()R
G

f(, ), hence

i
A
,i
= Dom(f

) < R
G
(, ) f

()R
G

f(, ). Hence if <

then f(, ) is G

-connected to every B
,(,2)
B
,(,2)

,
hence
f(, ) Y
0
,(,2)
Y
0
,(,2+1)
which implies that f(, ) B

. So the function
f maps the set (, ) : < into B

. But f is a one-to-one function and B

has cardinality < , contradiction.

2.4
{2.4A}
Denition 2.5. 1) For < and < we dene Ps(, ) and Ps

(, ) similarly
to the denition of Pr(, ), Pr

(, ) in Denition 1.2(3),(4) except that we replace

clause (e) by
(e)

if g is a one to one function from to , then g extends some f T

(so the dierence is that Dom(g) is required to be ).
2) Let Pr
3
(, , , ) means that for some T:
(a) T a family of partial functions from to
(b) [T[
(c) f T otp(Dom(f)) =
(d) if g

(f T)(f g).
{2.5}
Claim 2.6. 1) Assume is strong limit, > and cf() > cf().
Then Pr

(,

) holds if

< has conality cf().

2) If =
+
= 2

, cf(

) ,= cf(),

< then Pr

(,

) holds.
3) If < is a limit ordinal and =
||
then Ps

(, ).
4) If = cf(), < < , =

and Pr
3
(, , , ) for every < , then
Ps

(, ).
5) Pr(, ) Ps(, ), Pr

(, ) Ps

(, ) and similarly with instead of .

6) If 2

then = U

() Pr(, ).
7) If Ps

(, ) then Ps(, ).
Remark 2.7. Recall U

() = U
J
bd

(), and for an ideal J on , U

J
() = Min[T[ :
T []

is such that for every f

for some A T we have i < : f(i)
A ,= mod J.
Proof. 1) Let
i
: i < cf()) be increasing continuous with limit such that

<
0
and 2
i
<
i+1
, hence for limit ,

is strong limit cardinal of conality

cf(). For S
cf()
cf()
= < cf() : cf() = cf(), let f
,
: < 2

) list the
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


14 SAHARON SHELAH
partial one-to-one functions from

to

with domain of cardinality

. We choose
by induction on < 2

a subset A
,
of Dom(f
,
) of order type

unbound in

such that < sup(A

,
A
,
) <

; possible as we have a tree with cf()

levels and 2

cf()-branches, each giving a possible A

,
Dom(f
,
) and each
A
,
( < ) disqualies

+[[ of them.
Now T = f
,
A
,
: S
cf()
cf()
and < 2

is as required because if f is
a partial function from to such that [Dom(f)[ = and f is one to one then
< cf() : (

i <

)(i Dom(f) f(i) <

) contains a club of cf().

2) This holds as
S
for every stationary S < : cf() ,= cf(), see [Sh:108],
and without any extra assumption by [Sh:922].
3) By the simple black box (see proof of 1.5, well it was phrased for but the same
proof, and we have to rename ,
<
as ; see [Sh:e, Ch.IV], i.e. [Sh:309]).
4) We combine the proof of the simple black box and the denition of Pr
3
. Let

i
: i ) be increasing continuous,
0
= 0,

= . By Pr
3
(, , , ) with
=

i+1

i
, for each i < , we can nd T
i
such that
()
1
T
i
g : g a partial function from to , otp(Dom(g)) =

i+1

()
2
[T
i
[
()
3
for every g

there is g g

from T
i
By easy manipulation
()
+
3
if g

and < then there is g T
i
such that g g

and Dom(g)
[, ).
Clearly T
i
exists by the assumption Pr
3
(, , , ) for < so let T
i
= g
i,
:
< . Now for every

let f
0

be the following partial function from (

>
)
to :
()
4
if i < , < and Dom(g
i,(i)
) then f
0

(( i, )) = g
i,(i)
().
Let h be a one to one function from (
>
) onto such that (if cf() then
also in ()
5
(b) we can replace ,= by <)
()
5
(a)
>
< < h((, )) < h(, )
(b)
>
< < Dom(g
g(),(g())
) h((, )) ,=
h((, )).
Let f

be the following partial function from to satisfying f

() = f
0

(h
1
())
so it suces to prove that T = f

:

and f

is one-to-one exemplies
Ps

(, ).
First, clearly each f

is a partial function from to . Also for each i < and

< the function g
i,
has domain of order type

i+1

i
, hence by ()
5
(a) also


i < Dom(f

h(i, ) : < has order type

i+1

i
. By ()
5
(b)
also Dom(f

) has order type =

i
(

i+1

i
).
Now if f T then f

is one-to-one by the choice of T. Second, let f :

be a one-to-one function and we shall prove that for some , f

T f

f.
We choose
i

i
by induction on i such that j < i
j
=
i
j and

i


f

h(
j
, ) : j < i, Dom(g
j,i(j)
) f.
For i = 0 and i limit this is obvious and for i = j + 1 use ()
+
3
. So

and
f

f hence is one-to-one hence f

T so we are done.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70615
5) Easy (recalling 1.3(3)).
6) Easy.
7) Easy because if f : is one-to-one then for some u of order type , fu
is increasing (trivial if is regular, easy if is singular).
2.6
{2.5A}
Claim 2.8. 1) Each of the following is a sucient condition to Pr
3
(, , , ),
recalling Denition 2.5(2):
(a)
||
= = >
(b) = > [[ and (
1
< )(
||
1
< )
(c) =

([[).
2) If
1

2
,
1

2
,
1

2
,
1

2
then Pr
3
(
1
,
1
,
1
,
1
) implies Pr
3
(
2
,
2
,
2
,
2
).
Proof. 1) If clause (a) holds, this is trivial, just use T = f : f a partial function
from to with = otp(Dom(f)). If clause (b) holds, note that for every
f

, for some i
1
, i
2
< we have otp(j < i
1
: j < and f(j) < i
2
) and
let T = f : f a partial function from to with bounded range and bounded
domain if = such that = otp(Dom(f)). If clause (c) holds, use [Sh:460].
2) Trivial.
2.8
{2.5B}
Claim 2.9. 1) In 1.4, 1.5(2) and in 2.2 we can weaken the assumption Pr(, ) to
Ps(, ).
2) In 2.4 we can weaken the assumption Pr

(,

) to Ps

(,

).
Proof. The same proofs.
We can get another answer on the existence of universals.
2.9
{2.5C}
Claim 2.10. If is strong limit, cf() > cf() and > ( ), then in H
,,
there is no we-universal member even for the class of bipartite members.
Proof. Let :=
+
(recalling is a limit cardinal) by 2.6(1) we have Pr

(, )
hence by 2.4 we are done.
2.10
Note that Ps may fail.
Claim 2.11. Assume < , cf() cf() and < [[
cf()
< .
Then Ps(, ) fail (hence also Pr(, ), Pr

(, ), Ps

(, ).
Proof. Toward contradiction let T witness Pr(, ) so T is of cardinality . Let
f

: < ) list T; choose an increasing sequence

i
: i < cf()) such that

i
= (
i
)
cf()
and =
i
: i < cf(). We choose |
i
by induction on i such that:
()
1
i
(a) |
i

(b)
i
|
i
(c) [|
i
[ =
i
(d) if f T and Dom(f) |
i
is unbounded in Dom(f), then Dom(f)
Rang(f) .
For clause (d) note that if Dom(f) |
i
is unbounded in Dom(f) then there is
u Dom(f) |
i
unbounded in Dom(f) of order-type cf() and such u determines
f in T uniquely
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


16 SAHARON SHELAH
Now choose f

: such that f

maps [

j<i()
,
i()
)(|
i
|
j
: j < i) into
[
i(1)
,
i(j)+1
) when i(2) i(j) < cf() and be increasing f

contradict the choice

of T.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70617
3. Complete characterization under GCH
We rst resolve the case is strong limit and get a complete answer in 3.5 by
dividing to cases (in 3.1, 3.2, 3.4 and 2.10), in 3.4 we deal also with other cardinals.
This includes cases in which there are universals (3.1, 3.2) and the existence of
we-universal and of ste-universal are equivalent. In fact in 3.4 we deal also (in part
(2)) with another case: = cf(), =

<,<
[[
||
(and then there is no
we-universal).
Next we prepare the ground for resolving the successor case under GCH (or
weaken conditions using also 3.4(2)). If =
+
= , =

there is a we-universal
in H
,,
(3.7), if =
+
= 2

, < (in 3.8, 3.12) we give a sucient condition for

existence. In 3.15 we sum up. We end with stating the conclusion for the classes
of bipartite graphs (3.16, 3.17).
{2.6}
Claim 3.1. Assume is strong limit, cf() cf(), < and < cf() <
cf(); hence cf() < < so is singular.
Then in H
,,
there is a ste-universal member.
Proof. Denote = cf() and let
i
: i < ) be increasing continuous with limit
such that
0
> and (
i+1
)
i
=
i+1
. For any graph G H
,,
we can nd
V
G
i
: i < ) such that: V
G
i
V
G
, V
G
i
: i < ) is increasing continuous with i
with union V
G
such that [V
G
i
[ =
i
and
()
1
if x V
G
V
G
i+1
then [y V
G
i+1
: y is G-connected to x[ < .
As = cf() cf() it follows
()
2
if i < is a limit ordinal and x V
G
V
G
i
then (cf(i) < cf() hence)
[y V
G
i
: y is G-connected to x[ < .
For i let T
i
=

j<i
2
j+1
, T
s
= T
i+1
: i < , T =

i<
T
i
.
Let

A = A

: T
s
) be a sequence of pairwise disjoint sets such that
T
i+1
[A

[ =
i
. For T T

let B

= A
j
: j < g(), j a successor
ordinal. Now we choose by induction on i , for each T
i
a graph G

such
that:
(a) V
G
= B

(so for =<> this is the graph with the empty set of
nodes) and so [V
G
[ =
j
: j < g()
successor
(b) if then G

is an induced subgraph of G

, moreover
(x V
G
V
Gv
)(x is G

-connected to < nodes in V

G
)
(c) if i < , T
i
, G a graph such that [V
G
[ =
i+1
and G H
i+1,,
and G

is an induced subgraph of G and (x V

G
V
G
)(x is
G-connected to < members of V
G
), then for some < 2
i+1
there is an isomorphism from G onto G

which is the
identity on B

(d) G

H
|G|,,
.
[Can we carry the induction? For i = 0 this is trivial. For i = j + 1 this is easy,
the demand in clause (c) poses no threat to the others. For i limit for T
i
, the
graph G

is well dened satisfying clauses (a), (b) (and (c) is irrelevant), but why
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


18 SAHARON SHELAH
G

H
|G|,,
? Toward contradiction assume A
0
, A
1
B

, A
0
A
1
R
G
and
[A
0
[, [A
1
[ = , . If < 2 and cf([A

[) ,= cf(i) then for some j < i, [A

B
j
[ =
[A

[ so without loss of generality A

B
j
, but then by clause (b) no x B

B
j
is G

-connected to members of B
j
and [A

[ Min, = , hence A
1

B
j
, so we get contradiction to the induction hypothesis. So the remaining case
is cf([A
0
[) = cf(i) = cf([A
1
[) hence cf() = cf() = cf(i); so as we are assuming
cf() cf() cf(i), we necessarily get cf(i) = cf() = cf() = cf(). By the
last assumption of the claim (i.e. < cf() > cf()) we get that < and
without loss of generality [A
0
[ = , [A
1
[ = , so for some j < cf() we have j < i
and [A
1
B
j
[ , so as above A
0
B
j
, hence again as above A
1
B
j
and
we are done.]
We let G

= G

: T. Now we shall check that G

is as required. First
assume toward contradiction that A, B V
G

and A B R
G

and [A[, [B[ =

, . A set C V
G

will be called

A-at if it is included in some B

, T T

.
Easily above if B is not

A-at then A is

A-at. So without loss of generality for
some T

we have A B

but then x G

(for some i(x) < and

T
i(x)
we have x B

) [y B

: x is G

-connected to y[ < , so as
we get B B

, hence A, B V
G
and we get contradiction to clause (d) of
.
So K
,
does not weakly embed into G

; also [V
G

[ = so G

H
,,
. Lastly, the
ste-universality follows from the choice of V
G
i
: i < ) for any G H
,,
. That is,
we can choose by induction on i,
i
T
i
and an isomorphism f
i
from G V
G
i
onto
G
i
if i > 0, with f
i
increasing continuous (and
i
increasing continuous) using for
successor i clause (c) of .
3.1
In the previous claim we dealt with the case of , < . In the following claim
we cover the case of = :
{2.7}
Claim 3.2. Assume is strong limit, cf() cf() < = ; hence is
singular.
Then in H
,,
there is a ste-universal member.
Proof. Similar to the previous proof and [Sh:26, Th.3.2, p.268]. Let = cf()
and

=
i
: i < ), T
i
: i ), T
s
, T be as in the proof of 3.1. For any graph
G H
,,
let h
G
:

(V
G
) be dened by: if [x

: < [ = then h
G
( x) is the
rst i < such that
i
[y V
G
: y is G-connected to every x
i
, i < = g( x)[,
otherwise h
G
( x) is not dened. Now we choose V
G
i
: i < ) as an increasing
continuous sequence of subsets of V
G
with union V
G
such that if i < then
[V
G
i
[
i
and x

(V
G
i+1
) [Rang( x)[ = h
G
( x) i + 1 (y V
G
)(y is
G-connected to x
i
for every i < y V
G
i+1
).
Then when (as in the proof of 3.1) we construct G

: T
s
) we also construct
h

: T
s
), h

:

(B

) with the natural demands. In the end we have to

check that K
,
is not strongly embeddable into G

; if cf() = we need to look

at slightly more (as in the end of the proof of 3.1).
3.2
Remark 3.3. More generally see [Sh:829].
{2.8}
Claim 3.4. 1) Assume is strong limit, = , cf() = cf(), then in H
,,
there is no we-universal graph, even for the members from H
sbp
,,
.
2) Assume
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70619
(a) =
(b) ( < )( < )[[[
||
], (recall = )
(c) cf() cf().
Then in H
,,
there is no we-universal graph, even for the H
sbp
,,
.
Proof. 1) By part (2).
2) Let = cf(). Let
i
: i < ) be (strictly) increasing with limit = .
Without loss of generality
[
i+1

i
[ is (nite or) a cardinal (< ) with conality ,= cf().
[Why? If is a limit cardinal, trivial as if a cardinal < fails its successor is O.K.;
if is a successor cardinal,
i
= i is O.K. and also
i
= i or
i
=
1
i is O.K.]
Given a graph G

in H
,,
without loss of generality V
G

= .
For i < let T
i
= f : f is a partial one-to-one mapping from
i
into = V
G

with bounded range such that j < i 0, 1 [

j+1

j
[/2 = 1,
j

j+1
and mod 2[ and 2, 2+1 Dom(f) f(2)R
G

f(2+1). Let T =

i<
T
i
,
so T is a tree with nodes and levels; for T let i() < be the unique
i < such that T
i
. Let T
+
= : for some , g() = + 1, T is
-maximal in T and () = 0.
Note
()
1
if i < is a limit ordinal, f
j
: j < i) is -increasing, f
j
T
j
then

j<i
f
j
T
i
[in other words if f is a function from
i
to such that j < i
f
j
T
j
then f T
i
].
[Why? The least obvious demand is sup Rang(f) < which holds as cf(i) i <
= cf() cf().]
()
2
there is no

f = f
i
: i < ) increasing such that i < f
i
T
i
.
[Why? As then

i<
f
i
weakly embed a complete (, )-bipartite graph into G

.]
We dene a bipartite graph G
U
G
= : T, i() is even
(, ) : T
+
, < is even
V
G
= : T, i() is odd
(, ) : T
+
and < is odd
R
G
= R
G
1
R
G
2
where R
G
1
=

, (, ) : T
+
, i() is a succcessor ordinal and U
G
T
(and
i()1
<
i()
) and
and ,= g() mod 2

F
G
2
=

(,
1
), (,
2
) : T
+
, i() is a successor ordinal,

1
,
2
[i() 1, i() and
1
,=
2
mod 2

.
Now
(a) [T[ by clause (b) of the assumption, [
i()1
,
i()
] which is a weak form
of is strong limit
(b) [T
+
[ .
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


20 SAHARON SHELAH
If (|
1
(T
+
)) ,= ,= (V
1
(T
+
)) then we can choose (
1
,
1
) |
,
(
2
,
2
)
V
1
. [Why? As T
+
sup Rang() < , see ()
2
recalling i = g() < ,
Rang()
i
< .]
(c) [U
G
[ [T
1
[ = and [V
G
[ [T
2
[ =
hence
(d) [U
G
[ = = [V
G
[ so
(e) G H
sbp
,,
.
[Why? Being bipartite is obvious; so toward contradiction assume (U
1
U
G

V
1
V
G
) (U
1
V
G
V
1
U
G
) have cardinality (recall that = ) and
U
1
V
1
R
G
. If (

) V
1
and i < cf() = be such that

<
i
, then U
1

: (, ) U
G
V
G
: =

and <
i
which clearly has cardinality
< , contradiction. Hence V
1
(T
+
) = and by symmetry |
1
(T
+
) = ,
hence V
1
T, U
1
T hence U
1
V
1
(set of unordered pairs) is disjoint to R,
contradiction.]
(f) G is not weakly embeddable into G

.
[Why? If f is such an embedding, we try to choose by induction on i < , a member

i
of T
i
, increasing continuous with i such that ( Dom(
i
))(j < i)[
j
<

j+1

i
() = f((
i

j
, ))]. If we succeed we get a contradiction to G

H
,,
by ()
2
, so we cannot carry the induction for every i < . For i = 0 and i limit
there are no problems (see ()
1
), so for some i = j +1 < , f
j
is well dened but we
cannot choose f
i
. But if j < consider f
i
= f
j
(, f((
j
, )) : [
j
,
i
). This
gives a contradiction except possibly when = sup Rang(
i
), but then necessarily
by , [
i+1

i
[ has conality ,= cf(), so for < 2, for some

< the set

:
j
<
j+1
and f(
j
, ) <

and = mod 2 has cardinality [

i

j
[/2
which has conality ,= cf(), and then f
i
:= f
j
(, f(
j
, )) :
j
<
i
and
f(
j
, ) <
max{0,1}
is O.K., contradiction.]
3.4
{2.9}
Theorem 3.5. Assume
0
and is a strong limit cardinal. There is
a we-universal in H
,,
i cf() cf() and ( < cf() < cf()) i there is a
ste-universal in H
,,
.
{2.9d}
Remark 3.6. Similarly for the universal for G
[gr]
: g H
sbp
,,
.
Proof. We use freely 0.7(3) and below in each case the middle condition in 3.5
clearly holds or clearly fails and the other conditions hold or fail by the claim
quoted in the case.
If < and cf() > cf() by 2.10 the family H
,,
has no we-universal.
If = and cf() > cf() then Pr

(, ) holds by 2.6(1) (and recall that Pr(, )

is equivalent here to Pr

(, ), since is a cardinal, see 1.3(1)(iii)) hence by 2.2 the

family H
,,
has no we-universal member.
If cf() < cf() and < by 3.1 the family H
,,
has a ste-universal member;
the second statement in Theorem 3.5 holds as: if < easy, if then =
hence cf() < cf() = cf().
If cf() < cf() (hence ,= so < ) and = (so < ) by 3.2 the family
H
,,
has a ste-universal member.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70621
So the remaining case is cf() = cf(). If < < by 3.1 in H
,,
there is an
ste-universal; if = by 3.4(1) in H
,,
there is no we-universal member. If
< = by 3.2 in H
,,
there is a ste-universal member. We have checked all
possibilities hence we are done.
3.5
We turn to successor cardinal. In the following case, possibly the existence
of we-universal and ste-universal are not equivalent, see 3.12(2) + 3.12(4) and
Theorem 3.15.
{2.10}
Claim 3.7. Assume (
0
and)
(a) =
+
(b) < and =
(c) =

.
Then in H
,,
there is a we-universal member.
Proof. If G H
,,
(and without loss of generality V
G
= ) and < , then
()

< : is G-connected to elements < is bounded in say by

< .
Hence there is a club C = C
G
of such that:
(i) cf() ,= cf(), C, [, ) > otp < : is G-connected to

(ii) cf() = cf(), C, [, ) otp < : is G-connected to

(iii) if C then / and > sup(C ) cf() ,= cf().

We shall dene G

with V
G

= below. For each < divisible by let a

i
: i <
) list T

= a : a , and [a[ < or otp(a) = and = sup(a), each appearing

times, possible as [[ = =

, and let
R
G

, +i : < is divisible by , < , i < , a

+i, +j : i ,= j < and < is divisible by

.
Now clearly we have + < > [ < : is G

-connected to [
hence K
,
(which is K
,
by the assumptions) cannot be weakly embedded into
G

. On the other hand if G H

,,
without loss of generality V
G
= and let
C
G
be as above, and let

: < ) list in increasing order C

G
0, and we can
choose by induction on , a weak embedding f

of G

into G

( ). So G

is as required.
3.7
{2.11}
Claim 3.8. Assume (
0
and)
(a) = 2

=
+
, is a singular cardinal
(b) < and
3
<
3
in fact, = is O.K., but already covered by 3.4(2)
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


22 SAHARON SHELAH
(c) for every T []

of cardinality for some

4
B []

, for sets A T
we have B A
(d) cf() = cf().
Then in H
,,
there is no we-universal member (even for the family of bipartite
graphs).
Proof. Let G

H
,,
, without loss of generality V
G

= and we shall construct

a G H
sbp
,,,
not weakly embeddable into it. Now we choose U
G
= , V
G
= .
Notice that

= (by (a)), so let (f

, B

) : < ) list the pairs (f, B)

such that f : is one to one, B []

and f B is increasing such that each

pair appears times. Let
B
= sup + 1 : < is G

-connected to members
of B for B []

(and we shall use it for B []

, i.e. for subsets of V

G
). Now

B
is < by clause (c) of the assumption. We shall now choose inductively C

for
[, ) such that
(i) C

is unbounded of order type

(ii) no
Rang(f)
is G

-connected to every f

(), C

(iii) < [C

[ < .
In stage choose B

of order type such that ()[ < sup(B

) < ), that is B

is bounded in equivalently C

equivalently in B

for
< ; this is possible by diagonalization, just remember cf() = cf() and >
and clause (i).
Now there is C satisfying
()

C
C B

is unbounded of order type such that no

Rang(f)
is
G

-connected to every f

() for C.
[Why? Otherwise for every such C there is a counterexample
C
and we can easily
choose C
,i
by induction on i < such that:
(i) C
,i
B

(ii) sup(C
,i
) = sup(B

) =
(iii) otp(C
,i
) =
(iv) (j < i)[ > [C
,i
C
,j
[]
(iv)
+
moreover, if j < i then > [C
,i
< : f

() is G

-connected to

C,0C,j
[).
This is easy: for clause (iv)
+
note that for C = C
,j
C
,0
by the choice of
C
we
have
C

Rang(f)
hence by the choice of
Rang(f)
clearly D
C
=: i < : f

(i)
is well dened and G

-connected to
C
has cardinality < , so we can really carry
the induction on i < , that is any C B

unbounded in of order type such

that j < i [C D
C,jC,0
[ < will do.
Let A
0
= C
,0
, A
1
=
C,0C,1+i
: i < they form a complete bipartite subgraph
of G

by the denition of
C,0C,i
and [C
,0
[ = = [A
0
[ (by (iii) of ) and
[A
1
[ = (the last: by (iv)
+
), contradiction. So there is C such that ()

C
.]
Choose C

as any such C such that ()

C
. Lastly dene G
4
note that if () this clause always holds; and if 2

it is hard to fail it, not clear if

its negation is consistent
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70623
U
G
=
V
G
=
R
G
= (, ) : V
G
, C

.
Clearly G H
sbp
,,
recalling < and
1
,=
2
[C
1
C
2
[ < . Suppose
toward contradiction that f : is a weak embedding of G into G

, hence the
set Y = < : and f

= f is unbounded in and without loss of

generality Y
Rang(f)
=

, i.e. is constant. As f is one to one for every

Y large enough, f() (

, ) and we get easy contradiction to clause (ii)

for and we are done. (Note that we can add nodes to U
G
).
3.8
For the next claim, we need another pair of denitions:
{2.12m}
Denition 3.9. 1) H

is the class of G = (V
G
, R
G
, P
G
i
)
i<
where (V
G
, R
G
) is a
graph, [V
G
[ = and P
G
i
: i < ) is a partition of V
G
.
2) We say f is a strong embedding of G
1
H

into G
2
H

when it strongly
embeds (V
G1
, R
G1
) into (V
G2
, R
G2
) mapping P
G1
i
into P
G2
i
for i < .
3) G H

is ste-universal is dened naturally.

{2.12u}
Denition 3.10. For , U

() = min[P[ : P []

and (A []

)(B
P)([A B[ = ).
{2.12uu}
Remark 3.11. If is a strong limit cardinal and cf() < cf() < , then
U

() = .
{2.12}
Claim 3.12. 1) Assume (
0
and)
(a) =
+
= 2

(b) < and cf() ,= cf()

(c) 2

and U

() =
(d) (i) > cf() or
(ii) < or
(iii) < cf() and there are C

of order type for <

such that u []

otp[C

: u] > .
Then in H
,,
there is no we-universal even for the bipartite graphs in H
,,
.
2) In part (1) if we replace clause (d) from the assumption by (d)
1
or (d)
2
where
(d)
1

(d)
2
= and among the graphs or cardinal there is no ste-universal
(d)
3
= and in H

there is no ste-universal, then still there is no ste-universal,

even for the bipartite graphs in H
,,
.
3) If (a) + (b) of part (1) and 2
<
= =

= then there is a ste-universal.

4) If (a),(b) of part (1) and (c),(d) below, then there is a ste-universal in H
,,
(c) =

(d) in H

there is a universal, see 3.9(1).

(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


24 SAHARON SHELAH
{2.12d}
Remark 3.13. Note that part (2) is not empty: if is strong limit singular, 2

=
=
+
, =
<
< and P is the forcing of adding Cohen subsets to , then in
V
P
clause (d)
2
holds.
Proof. 1) Let G

H
,,
and without loss of generality V
G

= . As in the proof
of 3.7 using assumption (c) there is a club C of such that
(i) C, < otp < : is G

-connected to
(ii) C, cf() ,= cf(), < > otp < : is G

-connected to

(iii)
2
divides for every C.
Let S =: C : cf() = cf(); as we have
S
, see [Sh:922], so let

f = f

:
S), f

be a one-to-one function such that (f

)(
stat
S)[f is one-to-one
f

= f ]. For S let

= Min(C( + 1)), and for i [,

) we let
a
,i
= < : f

() is G

-connected to i, so otp(a
,i
) by the choice of C, and
let B

= i : i <

and [a
,i
[ .
Now for S we choose a

unbounded of order type such that (i

B

)(a

a
,i
).
[Why? First assume (d)(i), i.e. > cf() let [,

) =

< cf()
A
,
, [A
,
[ <
, A
,
is -increasing continuous with and let
,
: < ) be increasing con-
tinuous with limit satisfying [
,
(remember S
2
[). Now choose

,
[
,
,
,+1
) a
,i
: for some < cf(), = mod cf() and i A
,
for
< and let a

,
: < , it is as required.
Second assume case (ii) of clause (d) of the assumption, so < hence .
For S we choose a sequence

C

= C
,i
: i < ) of pairwise disjoint sets, C
,i
an
unbounded subset of S of order type , always exist as
2
[ (we could have asked
moreover that f

C
,i
is increasing with limit . Now if f : is one-to-one
then S : f

= f and for f

we can choose

C

is a stationary subset of
so this is O.K. but not necessary). If for some i < the set C
,0
C
,1+i
is as
required on a

, ne, otherwise for every i < there is

i
< which is G

-connected
to every y Rang(f (C
,0
C
,1+i
)). As any
i
is G

-connected to ordinals
< and C
,i
: i < ) are pairwise disjoint, clearly [j :
j
=
i
[ hence we
can nd Y such that
i
: i Y ) is with no repetitions and [Y [ = . So
A
0
= Rang(f C
,0
), A
1
=
i
: i Y exemplify that a complete (, )-bipartite
graph can be weakly embedded into G

, contradiction.
Lastly, the case clause (iii) of clause (d) holds. let
,
: < ) be an increasing
limit with such that [
,
; let C

i
: i < ) be as in clause (iii) of (d) of the
assumption and let C
,i
:= + 1: for some < we have
,
<
,
+ and

,
C

i
and otp(C

i
(
,
)) < .
Lastly, repeat the proof of Second....]
Lastly dene the bipartite graph G by V
G
= , R
G
= (, ) : S, a

.
Easily G H
sbp
,,
and is not weakly embeddable into G

by the choice of

f.
2) Let G

H
,,
, V
G

= . We choose the club C, the set S and the sequence

f = f

: S) as in the proof of part (1). We shall choose a

,i
: S, i < ) and
dene G by V
G
= , E
G
=

S
E

, E

= , +i : a

i
, i < i

+i, +j :
(i, j) R

. Naturally a

i
is an unbounded subset of of order-type .
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70625
Now it is sucient to nd for S an unbounded subset C = C

of of order
type such that for no =
C
< do we have ( < )( C f

()R
G

), in
this case i

= 1. If this fails then such

C
is well dened for any unbounded C
of order type ; C
,i
unbounded of order type , pairwise distinct for i <
and C
,1+i
C
,0
= ; then A
0
=: f

() : C
,0
, A
1
=:
C
,0
C
,1+i
: i <
exemplies that the complete (, )-bipartite graph can be weakly embedded into
G

, contradiction.
Clearly (d)
2
(d)
3
so without loss of generality (d)
3
holds. For S let
C
,j
: j ) be a sequence of distinct subsets of which include < :
f

()R
G

( + i) : i < i

and let M

be the model expanding G

[, + i

] by
P
M
j
= + i : i < i

and ( < )[f

()R
G

( + i) C
,j
]. As M

is not
universal in H

, so let N

witness this; without loss of generality the universe

of N in [, +), and let a
,j
= f

() : C
,j
and R

= R
N

.
Now check.
3) It suces to prove that the assumptions of part (4) holds, the non-trivial part
is clause (d) there, i.e. H

has a universal member. But 2

= so either is
regular so =
<
or is strong limit singular and in both cases this holds by
Jonsson or see [Sh:88r].
4) We choose G

for by induciton on such that

(a) G

is a graph with set of nodes (1 +)

(b) if < then G

is an induced subgraph of G
(c) if = + 1 and G is a graph with nodes and id
G

is a strong
embedding of G

into G such that x V

G
V
G

(x is connected to nodes of G

) then there is a
strong embedding of G into G

which extends id
G

.
The construction is possible by clause (d) of the assumption. Now as in the proof
of 3.8 G

H
,,
is ste-universal.
3.12
{2.12A}
Remark 3.14. 1) In the choice of

f (in the proof of 3.12) we can require that for
every f

the set S : f

= f and Rang(f) = Rang(f

) is stationary
and so deal with copies of the complete (, )-bipartite graph with the part after
the part.
2) Probably we can somewhat weaken assumption (c).
{2.14}
Theorem 3.15. Assume
0
and = 2

=
+
and 2
<
= .
1) In H
,,
there is a we-universal member i

= = i there is no
G

H
,,
we-universal for G
[gr]
: G H
sbp
,,
.
2) In H
,,
there is ste-universal, i

= = i there is no G

H
,,
ste-universal for G
[gr]
: G H
sbp
,,
.
Proof. 1) The second i we ignore as in each case the same claims cited give it too
or use 3.17 below. We rst prove that there is a we-universal except possibly when

= = .
Proving this claim, whenever we point out a case is resolved we assume that it
does not occur. We avoid using 2
<
= when we can.
If =

+
then by 2.6(3) we have Ps

(,
+
) so by 2.4 + 2.9(2) there is no
we-universal; hence we can assume that <

+
so (as =
<
) clearly
+
hence = =
+
that is = = .
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


26 SAHARON SHELAH
If = then by 3.4(2) there is no we-universal, so we can assume that ,=
hence < , hence =

so by 2.6(3) we have Ps

(, ) hence by 2.6(7) we
have Ps(, ). So if =
+
then by 1.4 more exactly, 2.9(1) there is no we-universal
so without loss of generality
+
< hence < . If cf() = cf() then by 3.8 we
are done except if
()
1
clause (c) of 3.8 fails, (and cf() = cf(), < )
is impossible as 2
<
= .
But (c) of 3.8 so we can assume
()
2
cf() ,= cf(),
so as 2
<
= , < we get
()
3
U

() = and 2

.
Now we try to apply 3.12(1), so we can assume that we cannot; but clauses (a)-(c)
there hold hence clause (d) there fails. So cf() (recalling sub-clauses
(i),(ii) of 3.12(1)(d)) as cf() ,= cf() by ()
2
and we have < cf() and
= . As 2
<
= this implies

= and = =
+
as promised. All this gives
the implication ; the other direction by 3.7 gives there is a we-universal.
2) By part (1) and 0.7(3), the only open case is

= and = =
+
then Claim
3.12(3),(4) applies (clause (c) there follows from =

).
3.15
Recalling Denition 0.6
{2.15}
Claim 3.16. The results in 3.15 hold for H
sbp
,,
and for H
bp
,,
.
Proof. The no universal clearly holds by 3.15, so we need the positive results,
and we are done by 3.17 below.
3.16
{2.16}
Claim 3.17. The results of 3.1, 3.2, 3.7 and 3.12(3),(4) hold for H
sbp
,,
and H
bp
,,
.
Proof. In all the cases the isomorphism and embeddings preserve x U
G
, y
V
G
.
For H
sbp
,,
, in 3.7 we redene G

as a bipartite graph (recalling a

i
: i < ) lists
a : otp(a) and if equality holds then = sup(a) for < divisible by
)
U
G

= 2 : <
V
G

= 2 + 1 : <
R
G

= (2, 2 + 1) : for some < divisible by we have 2, 2 + 1 [, +]

(2, + 2i + 1) : < divisible by , i < , 2 < , a

( + 2i, 2 + 1) : < divisible by , 2 + 1 < , i < , a

The proof is similar. For H

bp
,,
, 3.7 we redene G

U
G

= 2 : <
V
G

= 2 + 1 : <
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70627
R
G

= (2, 2 + 1) : for some < divisible by , 2, 2 + 1 [, +]

(2, + 2i + 1) : < divisible by , i < , 2 < and a

(2 + 1, 2) : 2 + 1 < 2.
The proof of 3.1, 3.2 for H
sbp
,,
is similar to that of 3.1, 3.2. The G

is from
H
sbp
,g()
so the isomorphism preserve the x U
G
, y V
G
. For H
bp
,,
without loss
of generality ,= hence < (otherwise this falls under the previous case).
We repeat the proof of the previous case carefully; making the following changes,
say for 3.1, V
G
i
: i < ) is increasing continuous with union V
G
, U
G
i
: i < )
increasing continuous with union U
G
.
()

1
if x V
G
V
G
i+1
then > [y U
G
i
: y is G-connected to x[.
We leave 3.12(3),(4) to the reader.
3.17
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


28 SAHARON SHELAH
4. More accurate properties
{4.1}
Denition 4.1. Let Q(, , , ) mean: there are A
i
[]

for i < such that for

every B []

there are < ordinals i such that B A

i
.
{4.2}
Denition 4.2. 1) For let set(, ) = A : A is a subset of of
cardinality such that i < > [A ( i)[ and let set(, ) = []

for
< .
2) Assume , . Let Qr
w
(, , , ) mean that some

A exemplies it,
which means
(a)

A = A
i
: i < )
(b) A
i
set(, ) for i <
(c)

A is (, )-free which means (A set(, ))(
<
i < )(A A
i
)
(d)
(e) if

A

= A

i
: i <

) satises clauses (a),(b),(c),(d), then for some one to

one function from

i<

i
into

i<
A
i
and one-to-one function from

to (or to ) we have i <

(A

i
) A
(i)
.
3) Qr
st
(, , , ) is dened similarly except that we change clause (e) to (e)
+
demanding (A

i
) = A
(i)
. Let Qr
pr
(, , , ) be dened similarly omitting clause
(e).
4) Assume , and x w, st. Let NQr
x
(, , , ) mean that
Qr
x
(, , , ) fails.
{4.3}
Claim 4.3. 1) Assume NQr
w
(, , , ) and =
+
and ( >

) ( < ).
Then
(a) in H
,,
there is no we-universal member
(b) moreover, for every G

H
,,
there is a member of H
sbp
,,
not weakly
embeddable into it.
2) Assume NQr
st
(, , , ) and =
+
and ( >

) ( ). Then
(a) in H
,,
there is no ste-universal member
(b) moreover, for every G

H
,,
there is a member of H
sbp
,,
not strongly
embeddable into it.
3) In parts (1), (2) we can weaken the assumption =
+
to
=

> and there is T f : f a partial one to one function from

to , [Dom(f)[ = of cardinality such that for every f

there
5
is
f T, f f

.
Proof. 1), 2) Let x = w for part (1) and x = st for part (2).
Now suppose that G

H
,,
and without loss of generality V
G

= and we
shall construct G H
sbp
,,
not x-embeddable into G

(so part (b) will be proved,

and part (a) follows).
Case 1: < so set(, ) = []

.
5
we can add there are functions f F, f f

, with pairwise disjoint domains, and

possibly increasing F we get it
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70629
Similar to the proof of 1.5. Let

f = f

:

) be a simple black box for one
to one functions. It means that each f

is a one to one function from i : i <

into , such that for every f :
>
for some

we have f

f. We dene
the bipartite graph G as follows:
() (a) U
G
=
>
and V
G
= (

)
(b) R
G
= R
G

:

where R
G

( , (, i)) : < , i < .

Now for each

, we choose
,i
: i <

) listing without repetitions the

set < : is G

-connected to members of Rang(f

) and without loss of

generality

= [

[ and A
,i
= < :
,i
, f

( ) are G

-connected.
As G

H
,,
clearly A []

[i <

: A A
,i
[ < hence

= [

[
2

+ but (see Denition 4.2(2)) we have assumed and (in 4.3) =

so 2

hence

; next let

A

= A
,i
: i <

) and as

A

cannot be a
witness for Qr
x
(, , , ) but clauses (a), (b), (c), (d) of Denition 4.2(1) hold,
hence clause (e) fails so there is

A

= A

,i
: i <

) exemplies the failure of

clause (e) of Denition 4.2 with

A

,

A

here standing for

A,

A

there.
Let
R
G

= ( , (, i)) : < , i <

and A

,i
.
The proof that G cannot be x-embedded into G

is as in the proof of 1.5.

Case 2:

< (and =

, < ).
First note that by the assumptions of the case

1
there is

f = f

:

) such that
(a) f

is a function from

<
( ) into
(b) if f is a function from (
>
) to then we can nd

:

)
such that
(i)

g()

(ii)
1

2

1

2
(iii) if < < and
>
then

,=

(iv) f

f for

.
We commit ourselves to

2
(a) U
G
= (
>
) and V
G
= (, i) :

, i <
(b) R
G
= R
G

:

where
(c) R
G

(( , j), (, i)) : j < , i < , < .

We say

is G

-reasonable if f

is one to one and for every < and y V

G

the set ( , j) : < , j < and f

(( , j)) is G

-connected to y has
cardinality < . We decide

3
(a) if is not G

-reasonable then R
G

=
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


30 SAHARON SHELAH
(b) if is G

-reasonable let
,i
: i <

) list without repetitions the

set < : is G

-connected to at least members of

Rang(f

); and let A
,i
= (, j) : < , j < and f

(( , j))
is G

-connected to
,i
; clearly A
,i
set(, ) and
let

A

= A
,i
: i <

)
(c) as

A

cannot guarantee Qr
x
(, , , ) necessarily there is

= A

,i
: i <

) exemplifying this so

and let
R
G

= (( , j), (, i)) : i <

, and (, j) A

,i
.
The rest should be clear; for every f :
>
letting

:

) be as in
above, for some

,

is G

-reasonable.
Case 3: = .
Left to the reader (as after Case 1,2 it should be clear).
3) As in the proof of 2.6(4), it follows that there is

f as needed.
4.3
{4.4}
Claim 4.4. 1) NQr
st
(2

, , 2

, ).
2) If = = 2

then in H
,,
there is no ste-universal even for members of H
sbp
,,
.
Proof. 1) Think.
2) By part (1) and 4.3.
4.4
{4.5}
Claim4.5. 1) Assume < and Qr
x
(, 1, , ) and H
sbp
(,),,
,= , then H
sbp
(,),,
=
H
bp
(,),,
has a x-universal member.
Proof. Read the denitions.
4.5
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70631
5. Independence results on existence of large almost disjoint
families
This deals with a question of Shar
{3.1}
Denition 5.1. 1) Let Pr
2
(, , , ) mean: there is / []

such that
(a) [/[ =
(b) if A
i
/ for i < and i ,= j A
i
,= A
j
then [

i<
A
i
[ < .
If we omit we mean , if we omit we mean 2

.
{3.3}
Claim 5.2. Pr
2
(, , , ) has obvious monotonicity properties.
{3.2}
Claim 5.3. Assume
() =
<
< =

= cf() and ( < )([[

<
< ) and 2

=
+
< (so

++
).
Then for some forcing notion P
(a) [P[ =

(b) P satises the

++
-c.c.
(c) P is -complete
(d) P neither collapses cardinals nor changes conalities
(e) in V
P
we have 2

, 2

(f) in V
P
we have Pr
2
(, ,
+
, ) but Pr
2
(
++
, , , ) for < recalling
5.2.
Proof. The forcing is as in a special case of the Q one in [Sh:918, 2], see history
there. Let E be the following equivalence relation on
E + = +.
We dene the partial order P = (P, ) by
P = f : f is a partial function from to 0, 1
with domain of cardinality such that
( < )([Dom(f) (/E)[ < )
f
1
f
2
i f
1
, f
2
P, f
1
f
2
and
> [ Dom(f
1
) : f
1
(/E) ,= f
2
(/E)[.
We dene two additional partial orders on P:
f
1

pr
f
2
i f
1
f
2
and f
1
, f
2
P and
( Dom(f
1
))[f
1
(/E) = f
2
(/E)].
f
1

apr
f
2
i f
1
, f
2
P, f
1
f
2
and Dom(f
2
) /E : Dom(f
1
).
We know (see there)
()
0
P is
++
-c.c., [P[ =

(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


32 SAHARON SHELAH
()
1
P is -complete and (P,
pr
) is
+
-complete, in both cases the union of an
increasing sequence forms an upper bound
()
2
for each p, P q : p
apr
q is
+
-c.c. of cardinality

= and -complete
()
3
if p r then for some q, q

we have p
apr
q
pr
r and p
pr
q

apr
r;
moreover q is unique we denote it by inter(p, r)
()
4
if p

Ord then for some q we have
(a) p
pr
q
(b) if < and q r and r
P

() = then inter(q, r)
P

() =
.
This gives that clauses (a),(b),(c),(d) of the conclusion hold. As for clause (e),
2

follows from ()
5
+[P[ =

and 2

, too.
Dene:
()
5
(a) f

:= p : p G

(b)
P
f

= G

P
is a function from to 0, 1
(c) for < let A

= < : f

( +) = 1.
Also easily
()
6
if p

then for some u []

and -Borel funtion B :

u
2

2
and q we have
p
pr
q
q

= B(f

u)
()
7
A

moreover < [, +) A

,= and [, +) A

and
,= A

,= A

.
[Why? By density argument.]
()
8
/ := A

: < exemplies Pr
2
(, ,
+
, ).
Why? By ()
7
+()
5
(c), / []

, [/[ = so we are left with proving clause (b) of

5.1; its proof will take awhile. So toward contradiction assume that for some p P
and

: <
+
) we have

1
p
P

< ,

,=

for < <

+
and [

<
+
A

[ .
By induction on
+
we choose p

P such that:

2
() p
0
= p
() p

is
pr
-increasing continuous
() there is r

such that p
+1

apr
r

and r

() Dom(p
+2
) (

/E) ,= .
No problem because (P,
pr
) is
+
-complete and
+
<
+
and ()
3
.
Let
6
q = p

+.
We can nd r

for <
+
such that q
apr
r

and r

. Let
u

= Dom(r

) Dom(q), so [u

[ < by the denition of

apr
. By the -system
6
if we dene P such that it is only -complete, we rst choose y
2
, u
3
, u

, v

, v

and then
q = p
()
for () = min{ <
+
: | y| = }
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70633
lemma, recalling =
<
there is Y
+
, [Y [ =
+
and u

such that for <

from Y we have u

= u

. Without loss of generality Y r

= r

.
Let v

= < :

+ Dom(r
+2
) so v

[]
<
.
Possibly further shrinking Y without loss of generality
,= from Y v

= v

.
So v

[]
<
(in fact follows).
Let () = Min(Y ).
We claim
r
()

<
+
A

.
As p r
()
this suces.
Toward contradiction assume that r, are such that

3
r
()
r P and r

<
+
A

.
Recall clause () of
2
(and <
+
p

pr
p

+ = q), we know <

+

Dom(q) (

/E) ,= and, of course, q r

()
r so by the denition of in P,
for every <
+
large enough

4
(a) (

/E) Dom(r

)Dom(r
()
) = hence
(b) r, r

are compatible functions (hence conditions)

(c) / v

.
Let r
+
= r r

+, 0).
So easily

5
(a) r r
+
P
(b) r

r
+
hence r
+

(c) r
+
/ A

.
So we have gotten a contradiction thus proving ()
8
.
()
9
Pr
2
(
++
, , , ) if < .
Why? So toward contradiction suppose p

= B

: <
++
[]

exemplies Pr
2
(
++
, , , ).
For each <
++
we can nd p

such that p, r and then S, q:

1
(a) p = p
,i
: i ) is
pr
-increasing continuous (in P)
(b) p
0
= p

(c) p
,i
r
,i
and r
,i

P

,i
B

i
(d) p
,i

pr
p
,i+1

ap
r
,i
(e) S

is stationary
(f) v
,i
:= Dom(r
,i
) Dom(p
,i+1
) : i S) is a -system with heart
v

, so [v

[ <
(g) r
,i
v

: i s) is constantly r

(h) p

= p
,
r

P so p

q.
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


34 SAHARON SHELAH
Let u

= /E : Dom(p

) so u

[]

and Dom(r
,i
) u

for i < .
For some Y [
++
]

++
and

<
+
and stationary S and =
i
: i S)
we have

2
if Y then otp(u

) =

and S

= S and
,i
: i S) = .
Let g
,
be the order preserving function from u

onto u

.
Again as 2

=
+
without loss of generality

3
For , Y
(a) p

= p

g
,
and p
,i
= p
,i
g
,
and r
,i
= r
,i
g
,
for i <
(b) u

= u

for < <

++
(c) g
,
is the identity on u

and

4
if p

apr
r

, p

apr
r

, r

= r

g
,
and < then
(a) r

(b) r

/ B

/ B

.
Choose

: <
++
) an increasing sequence of ordinals from Y .
Let p

<
p

and () =

.
So:

5
(a) P [= p

< q

for <
(b) if p

q then for each < for every large enough i S, the

conditions q, r
,i
are compatible hence
(c) if q

q then for every large enough i S the universal q

+
=
q r
,i
: < is a well dened function, belongs to P
and is common upper bound of r
,i
: < q
hence force
i
B

for < .
As
i
i clearly

6
q

<
B

is unbounded in hence has cardinality .

5.3
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


UNIVERSALITY AMONG GRAPHS OMITTING A COMPLETE BIPARTITE GRAPH SH70635
6. Private Appendix
Continued from pg.36:
(b) Q = P q : p

apr
q satises the
+
-c.c.
(c) if p

r and r

<
B

then < inter(p

, r) B

but
inter(p

, r) inter(p

, r)
hence inter(p

, r)

<
B

for <
(d) u = < : p

<
B

is a set of cardinality < (as it is forced
to be a set of cardinality < , and cf() > and (b) + (c))
(e) there are p
+
, p
()

apr
p
+
and u such that p
+
B

()
.
[Why? As p
()
B

()
[]

, for some r, we have u, p

()
r and
r B
()
. Now inter(p
()
, p
+
) is as required.]
Let p
+

= p
+
g
(),
and let
p
+
=

<
p
+

.
So p

apr
p
+
and
p
+

<
B

;
a contradiction.
5.2
(
7
0
6
)


r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
5
-
0
4







m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
5
-
0
5


36 SAHARON SHELAH
7. Private Appendix 2
{2.3}
Claim 7.1. Assume
(a) = cf() > , 2

> (and )
(b) / []

, [/[ = , and (X []

)(A /)(A X)
(c) for every < we have > [A / : A [
(d) cf(

, <

) > .
Then there is no we-universal in H
,,
.
Proof. Saharon, recall.
References
[EH74] Paul Erd os and Andras Hajnal, Solved and Unsolved Problems in set theory, Proc. of the
Symp. in honor of Tarksis seventieth birthday in Berkeley 1971 (Leon Henkin, ed.), Proc. Symp
in Pure Math., vol. XXV, 1974, pp. 269287.
[KP84] Peter Komj ath and Janos Pach, Universal graphs without large bipartite subgraphs, Math-
ematika 31 (1984), 282290.
[Sha01] Ofer Shar, Two cases when d and d

are equal, Fundamenta Mathematicae 167 (2001),

1721.
[Sh:e] Saharon Shelah, Nonstructure theory, vol. accepted, Oxford University Press.
[Sh:26] , Notes on combinatorial set theory, Israel Journal of Mathematics 14 (1973),
262277.
[Sh:88r] , Abstract elementary classes near
1
, Chapter I. 0705.4137. 0705.4137.
[Sh:108] , On successors of singular cardinals, Logic Colloquium 78 (Mons, 1978), Stud.
Logic Foundations Math, vol. 97, North-Holland, Amsterdam-New York, 1979, pp. 357380.
[Sh:300] , Universal classes, Classication theory (Chicago, IL, 1985), Lecture Notes in
Mathematics, vol. 1292, Springer, Berlin, 1987, Proceedings of the USAIsrael Conference on
Classication Theory, Chicago, December 1985; ed. Baldwin, J.T., pp. 264418.
[Sh:309] , Black Boxes, , 0812.0656. 0812.0656. 0812.0656.
[Sh:460] , The Generalized Continuum Hypothesis revisited, Israel Journal of Mathematics
116 (2000), 285321, math.LO/9809200.
[KoSh:492] Peter Komjath and Saharon Shelah, Universal graphs without large cliques, Journal
of Combinatorial Theory. Ser. B 63 (1995), 125135.
[Sh:829] Saharon Shelah, More on the Revised GCH and the Black Box, Annals of Pure and
Applied Logic 140 (2006), 133160, math.LO/0406482.
[Sh:918] , Many partition relations below density, Israel Journal of Mathematics accepted,
0902.0440.
[Sh:922] , Diamonds, Proceedings of the American Mathematical Society 138 (2010),
21512161, 0711.3030.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The He-
brew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathe-
matics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110
Frelinghuysen Road, Piscataway, NJ 08854-8019 USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at