Topics in Set Theory
Lectured by O. Kolman
Michaelmas Term 2012
Axiomatics. The formal axiomatic system of ordinary set theory (ZFC). Models of set theory.
Absoluteness. Simple independence results. Transnite recursion. Ranks. Reection principles.
Constructibility.
[4]
Innitary combinatorics. Conality. Stationary sets. Fodors lemma. Solovays theorem.
Cardinal exponentiation. Beth and Gimel functions. Generalized Continuum Hypothesis. Singular Cardinals Hypothesis. Prediction principles (diamonds, squares, black boxes). Partial
orders. Aronszajn and Suslin trees. Martins Axiom. Suslins Hypothesis.
[6]
Forcing. Generic extensions. The forcing theorems. Examples. Adding reals; collapsing
cardinals. Introduction to iterated forcing. Internal forcing axioms. Proper forcing.
[4]
Large cardinals. Introduction to large cardinals. Ultrapowers. Scotts theorem.
[2]
Partition relations and possible conality theory. Partition relations. Modeltheoretic
methods. Ramseys theorem; Erd
osRado theorem. Kunens theorem. Walks on ordinals.
Todorcevics theorem. Introduction to pcf theory.
[4]
Applications. Selection from algebra, analysis, geometry, and topology.
[4]
Please let me know of corrections: glt1000@cam.ac.uk
Last updated: Tue 14th Jan, 2014
Prerequisite Mathematics
Logic and Set Theory is essential.
Literature
Basic material
Drake, F. R., Singh, D., Intermediate Set Theory, John Wiley, Chichester, 1996.
Eklof, P. C., Mekler, A. H., Almost Free Modules, rev. ed., NorthHolland, Amsterdam, 2002.
Halbeisen, L., Combinatorial Set Theory With a Gentle Introduction to Forcing, Springer, Berlin,
2012.
Kanamori, A., The Higher Innite, 2nd ed., Springer, Berlin, 2009.
Kunen, K., Set Theory, reprint, Studies in Logic, 34, College Publications, London, 2011.
Advanced topics
Burke, M. R., Magidor, M., Shelahs pcf theory and its applications, Ann. Pure Appl. Logic 50
(1990), 207254.
Kanamori, A., Foreman, M., Handbook of Set Theory, Springer, Berlin, 2012.
Shelah, S., Proper and Improper Forcing, 2nd ed., Springer, Berlin, 1998. Chapters 1 and 2.
Shelah, S. Cardinal Arithmetic, Oxford University Press, New York, 1994.
Todorcevic, S., Combinatorial dichotomies in set theory, Bull. Symbolic Logic 17 (2011), 172.
Lecture 1
Introduction
The Continuum Hypothesis (CH) asserts that if X is any uncountable subset of the real
numbers R, then there is a bijection from X onto R. It is an answer to Cantors Continuum
question: for which ordinal is 20 = ?
This is Hilberts First Problem (1900). However, there are other answers. Indeed, the response to
Cantors question is among the most striking and surprising results of 20th Century mathematics.
Stated loosely, CH is independent of the principles of ordinary mathematics. More precisely, if
ZFC is consistent then ZFC does not prove CH and ZFC does not prove CH (the negation of
CH, namely 20 > 1 ).
One of the aims of this course is to provide an introduction to the ideas and methods developed
since Godel (1938) and Cohen (1963) to handle this and other insoluble problems.
Independence results appear everywhere, not just in set theory. Let us consider some examples
(to correct any naive misconceptions/prejudices, etc).
Example 1. Combinatorics and the order type of the real line.
Let T = hT, 6T i be a tree, i.e.
(1) 6T is a partial order on T ;
(2) {y T : y <T x} is well ordered.
A chain in T is a set C T such that C is totally ordered: x 6 y or y 6 x for all x, y C. An
antichain in T is a set A T such that (x 6= y A)(x 66 y y 66 x).
Suppose is an innite cardinal. A Suslin tree is a tree T = hT, 6T i such that
(1) T  =
(2) every chain and every antichain in T has cardinality less than .
Proposition. There are no 0 Suslin trees.
Proof. Let T be an innite tree with no innite antichains. We show that T has an innite
chain. Note that since T  = 0 , there exists x0 T such that {y T : y > x0 } is innite:
for if each x had only nitely many descendants then we could inductively construct an
innite antichain.
Now continue by induction to dene {xn : n }, as follows: given xn T such that
{y T : y > xn } = 0 , pick xn+1 > xn such that {y T : y > xn+1 } = 0 .
Then {xn : n } is an innite chain in T. So T is not 0 Suslin.
Question. What about 1 Suslin trees?
Could we prove (as in the case of 0 Suslin trees) that there are none?
We shall see that this question cannot be answered on the basis of ZFC. The assertion that there
are no 1 Suslin trees is called Suslins Hypothesis.
Example 2. Commutative Algebra.
L
Denition. An innite abelian group A is free if A
= iI Z for some set I, where Z is the
innite cyclic group under addition.
Denition. An abelian group G is called a Whitehead group (W group) if when we write
G = F/K with F and K free, every homomorphism f : K Z extends to a homomorphism from F to Z.
For example, a free abelian group is a W group. Every countable W group is free (harder).
Question. Is every W group free?
It emerges that the existence of an uncountable nonfree W group is independent of ZFC.
Example 3. Complex Analysis.
Let H(C) be the family of entire holomorphic functions.
Denition. A family F H(C) is called orbitcountable if for all n C, we have {f (w0 ) :
f F } 6 0 .
Every countable family F H(C) is orbitcountable.
Question. If a family F H(C) is orbitcountable, is F countable?
The answer (due to Erd
os) is that the existence of an uncountable orbitcountable family is
equivalent to CH.
Example 4. Euclidean Geometry (Sierpinski; Kuratowski; Davies)
Consider the assertions about the Euclidean plane and space:
(P1) Euclidean 3dimensional space can be decomposed into 3 sets Ei (i = 1, 2, 3) such that
each line parallel to the coordinate axis OXi intersects Ei in only a nite number of points.
(Q1) The Euclidean plane can be decomposed into 3 sets Ei (i = 1, 2, 3) such that for some 3
directions vi in the plane, each line in the direction vi intersects Ei in only a nite number of
points.
Theorem (Sierpinski et al). (P1) (Q1) 20 = 1 .
So, for example, the rudimentary geometric assertion (Q1) about partitioning the Euclidean
plane can neither be proved nor disproved from the principles of ordinary mathematics.
Example 5. Commutative Algebra (Los).
Suppose :
iI
Z Z is a homomorphism of groups.
Question. If LiI Z 0, is 0 ?
Theorem (Specker). If I = N, and LiI Z 0, then 0.
Theorem (Lo
s). The following are equivalent:
(1) There exists a nonzero group homomorphism :
(2) The cardinal I is a measurable cardinal.
iI
Z Z such that LiI Z 0.
Measurable cardinals are examples of large cardinals. They are sets that are so big that their
existence cannot be proved from the principles of ordinary mathematics.
Q Yet, they determine
whether there exist nontrivial homomorphisms of the Cartesian power iI Z fullling (1).
Denition. An uncountable cardinal is measurable i there is a complete nonprincipal
ultralter on , i.e. there is a set U P() such that
(1) U is a lter on (i.e., A, B U implies A B U ; U ; and if A U and
A B , then B U )
(2) for all A , either A U or \ A U
(3) for all < , {}
/U
(4) if < and A U for all < , then
<
A U .
Logical questions. Although these ve examples are not explicitly settheoretic in character,
the solutions are all independent of the principles of ordinary set theory. Why? What is the
source of the complexity that gives rise to settheoretic independence? What sorts of properties
and classes of structures are immune to independence phenomena?
Lecture 2
Chapter 1. Axiomatics
In this lecture, we review the mathematical logic required for independence results in set theory,
specify the vocabulary, the language and the axioms of the rstorder theory ZFC. We then describe the architecture of independence proofs, and introduce some useful models of (subtheories
of) ZFC. As a payo, we shall be able to show:
(1) any proof of the existence of the set of real numbers in rstorder set theory must necessarily use the power set axiom.
(2) the rstorder theory ZFC is not nitely axiomatisable
(3) the existence of a strongly inaccessible cardinal cannot be proved from ZFC
What does (3) mean?
Denition. A cardinal is strongly inaccessible i
1. is an uncountable regular cardinal, and
2. is a strong limit cardinal.
Denition. A cardinal is regular i cannot be written as a union of less than many
sets of size less than .
For example, 0 is regular it cannot be written as a nite union of nite sets. Also, 1
is regular countable
S unions of countable sets are countable, under AC. But is not
regular, since = n< n . (Nonregular innite cardinals are called singular.)
Denition. A cardinal is a strong limit i ( < )(2 < ).
For example, 0 is a strong limit.
We can manufacture strong limits using the i function (bet function). We dene it by
transnite recursion:
i0 = 0
i+1 = 2i
[
i =
i when is a limit ordinal
<
So i is a strong limit cardinal: for if < i then 6 in for some n , and so
2 6 2in = in+1 < i . However, i is not regular.
The question is then natural: can one prove there exist strongly inaccessible cardinals?
Let us turn to the technical denitions from mathematical logic.
Denition. A vocabulary is a family of symbols that includes relation symbols, function
symbols and constant symbols.
The vocabulary of set theory has one binary relation symbol .
If is a vocabulary, then the rstorder language L( ) or L is the family of terms and formulas
formed from using the syntactic rules of rstorder logic.
For deniteness, we specify that the logical vocabulary has the symbols:
and ), (, =, , ,
variables v0 , v1 , v2 , . . .,
But of course we shall use the customary metalinguistic abbreviations , , , and other
conventions too. We write L() or L for the language of set theory.
Suppose is a vocabulary. A theory is a family T of sentences (i.e., sentences in L( )).
For a theory T and a sentence (in L( )), we write T to mean that is deducible (provable)
from T , i.e. there is a formal rstorder deduction of from T .
For families S, T L( ), we write T S to mean T for all S.
A theory T L( ) is consistent i for some sentence , T 6 (i.e., T does not prove );
equivalently, for every L( ), we have T 6 .
A theory T is complete i T is maximal consistent, i.e. if T T and T is consistent then
T = T . Equivalently, T is complete i for every L( ), either T or T .
With this review of syntax over, we can dene the theory ZFC precisely. The axioms are the
transcriptions into L() of the formal principles of naive (or semiaxiomatic) set theory.
The axioms of ZFC are:
(A0) Set Existence. x(x = x)
(A1) Extensionality. x(x a x b) a = b
Notes. 1. We are using the metalanguage to write this axiom.
2. We tacitly adopt the generality interpretation of free variables: that free variables a
and b are understood to be universally quantied.
4
(A2) Null set. xy (y x)
(A3) Pair set. xy(y x y = a y = b)
(A4) Union. xy(y x z(y z z a))
(A5) Power set. xy(y x y a), where y a abbreviates z(z y z a)
(A6) Separation (restricted comprehension, subset). For each in L() in which x does not
appear,
xy(y x y a (y))
(i.e., x = {y a : (y)} is a set)
(A7) Replacement. For each (z, y) in L() in which x does not appear,
zuv (z, u) (z, v) u = v xy y x z(z a (z, y))
(i.e., x = {F (z) : z a} is a set, where F (z) = y (z, y).)
(A8) Foundation. x(x a) x(x a x a = )
(A9) Infinity. w( w x(x w S(x) w)), where S(x) = x {x}
(A10) Choice.
xy x a y a x 6= y x y = x(x a x 6= )
cx(x a u(c x = {u}))
Lecture 3
Denition. The rstorder theories ZF and ZFC are those with axioms A0A9 and A0A10
respectively.
Comments. Why these choices?
1. Why axiomatise?
reduce vagueness of naive set theory and avoid paradoxes
provide rigorous concept of provability (and unprovability)
discover new settheoretic principles
2. Why this firstorder axiomatisation?
the vocabulary {} is economised and suces to express everything pertinent
rstorder logic is sound, complete and compact
innitary logics are not compact (innitary logic allows innite conjunctions and
innite strings of quantiers)
secondorder and higher order logics are generally not compact and the quantiers
range over subsets of domains
3. Why these axioms?
see the Shoeneld handout (or Part II Logic & Set Theory)
The semantics and model theory of set theory
We discuss the concepts from rstorder model theory that we need for the analysis of ZF and
ZFC.
Denition. Let (x) be a formula in L(). We refer to {x : (x)} as a class. (Classes need
not be sets: e.g., V is the class {x : x = x}.)
We may allow parameters x1 , . . ., xn V . E.g., M = {x : (x, x1 , . . ., xn )}.
Denition. Suppose is a vocabulary. A structure is a pair M = (M, M ) where M is a
nonempty class, and M is a family of relations, operations and constants on M which
interpret the symbols of :
if R is an nary relation symbol then RM M n
if f (x1 , . . ., xn ) , then f M : M n M (and we allow dom(f ) M n )
if c is a constant symbol, then cM M
In the case of L(), a structure is simply a pair (M, M ) where M M M .
Denition.
1. Suppose L() is a sentence. We write M = to mean is true in M.
2. If free variables of the formula are included in x1 , . . ., xn and a1 , . . ., an M , we
write M = [a1 , . . ., an ] to mean a1 , . . ., an satisfy (x1 , . . ., xn ) in M (when x1 , . . ., xn
are assigned the values a1 , . . ., an )
3. If T is a theory, we write M = T to mean M = for every T , and say that M
is a model of T .
We write Mod(T ) for the class of models of T .
4. We say a theory T is nitely axiomatisable if there exists a nite family T0 of
sentences such that Mod(T ) = Mod(T0 ).
The fundamental results from rstorder logic we need are the following.
Theorem (Completeness). Suppose T is a theory and is a sentence.
(1) T = (i.e., every model of T is a model of ) i T .
(2) T is consistent i T has a model.
Theorem (Compactness). Suppose T is a theory. Then T has a model i every nite
subfamily of T has a model (i.e., T is nitely satisable).
Let us dene some important examples of structures for L().
Denition. We dene by transnite recursion on Ord the set V as follows:
(1) V0 =
(2) V+1 = P(V ), the power set of V
S
(3) V = < V , when is a limit ordinal
Denition. The class {V : Ord} is called the von Neumann hierarchy.
Exercise. V is a model of ZFC\{Infinity}.
6
Proposition.
1. If < then V V .
2. V is transitive: if x y V then x V (i.e., x V = x V ).
(Hint: use transnite induction on .)
Denition. The class WF of wellfounded sets is dened by the formula WF (x)
( Ord)(x V )
Picture.
V =
n<
Vn
..
.
..
.
..
.
V+1
V2 = PP
V7
V1 = P
V0 =
V1
Denition. Let be an innite cardinal. We dene H() = {x : trcl(x) < }, where trcl(x)
is the transitive closure of x.
H(), also written H , is the family of sets of cardinality hereditarily less than .
Lecture 4
Reflection Principles and Nonfinite Axiomatizability of ZF and ZFC
We motivate the Reection Principle with a simple proposition.
Proposition. ZF x WF (x), i.e. ZF x Ord (x V ).
Proof. Let x V and suppose x
/ WF .
Consider u = {y trcl({x}) : y
/ WF }. Then u 6= (as x U ).
By Foundation, u has an minimal element y.
For all z y, we have z WF (by minimality of y), so there
exists z minimal such that z Vz +1 .
{z : z y} is a set, by Replacement. So sup{z + 1 : z y}
Ord, and for all z y, we have z V . So y V , and hence
y V+1 = P(V ), contradicting y
/ WF .
So x must belong to WF . So we have shown that V = WF .
xr
V
rz
Vz +1
Thus our picture (under Foundation) simplies to:
V = WF
Comments.
1. Note, we do not attempt to show ZF xWF (x) by constructing a formal derivation
from the axioms. We argue informally in V (a class), and then observe that the
argument could be recast as a formal deduction.
2. We can restate the Proposition as follows. Let ZF0 be ZF without Foundation.
By the Deduction Theorem of rstorder logic, the Proposition is exactly ZF0
Foundation (V = WF ).
Exercise. Show that ZF0 (V = WF ) Foundation.
Returning to the picture of V , it is natural to ask: how similar are V and V ? We consider
which rstorder properties of V are reected in V , for large .
There are two aspects to this problem.
1. For an axiom of ZFC or rstorder property in L(), and Ord, is V a model of ?
2. For which axioms of ZFC can one prove there exist (arbitrarily large) Ord such
that V = .
There is a technical diculty to overcome: what does it mean to write V = , or A = when
A is a proper class?
Recall by Tarskis Theorem on indenability of truth that we cannot dene (in ZFC) V = .
We deal with this by relativising formulas to a class.
Denition. Suppose A = {x : A(x)} is a class where A(x) is a formula in L() with free
variable x.
For each formula in L() we dene A , the relativisation of to A, as follows:
if is vi vj or vi = vj , then A is vi vj or vi = vj
A
A
if is 1 2 or 3 , then A is A
1 2 or (3 )
if is y(y), then A is y(y A A (y)) more formally, A is y(A(y) A (y))
Exercise. Write out A for axioms (A0)(A5).
Intuitively, ZF A means A = , i.e. A is a model of when the bound quantiers are
relativised to the class A.
Denition. Let A be a class. We say {A : Ord} is a cumulative hierarchy for A if
(1) A is a set
(2) < implies A A
S
(3) A = < A if is a limit ordinal
S
(4) A = Ord A
8
Theorem (Reection Principle). Suppose A =
pose (y1 , . . ., yn ) is a formula in L().
Ord
A is a cumulative hierarchy. Sup
Then there exist unboundedly many Ord such that:
x1 , . . ., xn A , (x1 , . . ., xn )A (x1 , . . ., xn )A
Proof. The idea is to close under Skolemlike functions.
Write y for (y1 , . . ., yn ). List all of the subformulas of (
y ) as 1 , . . ., k . Wlog all existential subformulas (i.e. of the form u(. . .)) appear at the beginning of the list, and if (
y)
is x(x, y) then 1 is (
y ). Let ji + 1 be the number of free variables in i .
The Skolemlike functions. For y1 , . . ., yji A, dene fi : Aji Ord by
the least ordinal such that
if i (x, y1 , . . ., yji )A for some x A
then i (x, y1 , . . ., yji )A for some x A
fi (y1 , . . ., yji ) =
0 otherwise
For any , fi Aji is a set by Replacement, so there is an ordinal i () greater than all
j
ordinals in fi Ai .
Let () = max{i () : i 6 k}.
Now, given Ord, dene hn : n < i by
0 = ,
n+1 = (n ),
and = sup{n : n }
Claim. is as required. I.e., for all y1 , . . ., yn A , we have
(
y )A (
y )A
This we prove by induction on complexity of .
Lecture 5
The claim is trivial if (
y ) is atomic. If (
y ) is 1 (
y ) 2 (
y ) or 3 (
y ), then the claim
follows by induction hypothesis on 1 , 2 , 3 .
The interesting case is when the formula (
y ) is xi (x, y).
If (
y )A then there is x A such that i (x, y)A . So by induction hypothesis applied
to i (x, y) we obtain i (x, y)A , i.e. xi (x, y)A , i.e. (
y )A .
Conversely, if (
y )A and y Aji , then there is an m < with y1 , . . ., yji Am . What
do we know about = fi (
y ) ? Well, 6 m+1 .
We are given that there is x A such that i (x, y)A , so by construction there is x Afi (y)
such that i (x, y)A . But Afi (y) A , so by induction hypothesis we obtain i (x, y)A ,
or in other words (
y )A , as required.
2
Corollary 1. Suppose (
x) is a formula in L().
y ).
Then ZF > y1 , . . ., yn V (
y )V (
Proof. By the Proposition, ZF V = WF , so {V : Ord} is a cumulative hierarchy for V .
Now apply the Reection Principle with A = V and A = V .
2
Corollary 2. The theories ZF and ZFC are not nitely axiomatisable.
Proof. By contradiction and G
odels Second Incompleteness Theorem.
Suppose T0 is a nite set of sentences in L() and Mod(ZF C) = Mod(T0 ).
V
Let be T0 , the conjunction of T0 . Now, by the Reection Principle applied to using
{V : Ord}, there exists Ord such that V .
ZFC , so ZF C V , so V = , and so V = for every ZFC.
In other words, we have proved in ZFC that there exists a set (namely V ) which is a model
of ZFC. That is to say, we have proved the consistency of ZFC just using the axioms of
ZFC. This contradicts G
odels Second Incompleteness Theorem.
2
We might prefer a proof that does not appeal to Godel. Here is the idea.
For the V in the above proof, V = ZFC ()
Let be minimal with this property ().
V = ZFC
V = V V
If VV : < were a cumulative hierarchy for V , then in V we could apply the Reection
V
V
Principle to = T0 to get a new ordinal < such that V V .
V
But V , so V
. Since V = ZFC, VV must be V .
So < and V reects , contradicting the minimality of .
This is a vaguely expressed argument, but it can be made rigorous. The issue concerns how or
whether dened terms change their meanings as one moves from one model to another. This
will lead us to study the concept of absoluteness.
However, some further comment on the strength of the Reection Principle is necessary.
Theorem. Let LM (standing for LevyMontague) be the axioms of ZFC without the Replacement axiom scheme, but with all instances of the Reection Principle (one for each
formula (
y )).
Then, for every formula (
y ), LM Replacement(y) .
Proof. The idea is to use the axiom of Separation combined with RP(y) to nd an appropriate
witness for Replacement.
Details will be in an exercise.
10
Cumulative hierarchies are very useful as a source of new axioms of set theory.
Denition. Suppose A is a set. Then the collection Def(A) is the family of sets y A such
that for some formula (x, y1 , . . ., yn ) in L() and a1 , . . ., an A, we have y = {x :
(x, a1 , . . ., an )A }.
Def(A) is the collection of denable subsets of A.
We dene the cumulative hierarchy of constructible sets as follows.
1. L0 =
2. L+1 = Def(L )
S
3. L = < L when is a limit ordinal.
S
The class L = Ord L is the universe of constructible sets. The Axiom of Constructibility
is the assertion that V = L, i.e. x Ord(x L ).
The Axiom of Constructibility is independent of ZFC. It is a very powerful statement about the
regularity of the universe V . Although the study of the class L is not the goal of this course, we
note that L is a transitive class model of ZFC+GCH. It is the smallest transitive model of ZFC
containing all of the ordinals.
Lecture 6
The intuition behind the denition of the class L is that instead of the maximally fat power
set operation, one puts into the power set of a set x only those subsets of x that necessarily
must be there, i.e. the ones that have names (in the sense of denability).
It is instructive to begin the study of L by comparing the basic properties of V and L .
Theorem (Basic properties of V ).
(1) 6 V V
(2) < V V
(3) V is transitive: x V x V
(4) V = {x WF : rank(x) < }, where rank(x) = min{ Ord : x V+1 }
(5) y x rank(y) < rank(x)
(6) rank() = for all Ord
(7) Ord V =
= { : < }
(8) n, Vn  < 0 , and V  = 0 , and, with (AC), V+  = i for all Ord
Proof. All are done by transnite induction on Ord. (Exercise.)
Theorem (Basic properties of L ).
(0) L is a set and L V
(1) L is transitive
(2) Ln  < 0 , Ln = Vn for n < , and L = V
(3) L  =  for >
(4) 6 L L
(5) < L L (use the formula x = x), and L ( L
(6) L+1 and
/ L
11
Proof. In most cases, these are like the proofs for V using transnite induction. (Exercise.) 2
Comment. For most Ord, we have L ( V , by (8). Thus in L, we have greater control
over the power set operation.
These basic properties of V and L will be useful when we come to establish relative consistency
results.
We dene next the concepts of relative consistency and inner models.
Denition. Let be a sentence in L() and T a theory. We say that is consistent relative
to T if the consistency of T implies the consistency of T {}.
Analogously, we can dene the concept of S being consistent relative to T for another theory S.
There is a very useful criterion for relative consistency.
Theorem. Let T be a theory in L() (or some expansion of L()), and be a sentence.
Suppose A(x) is a formula in L() with exactly the free variable x. If
(1) T xA(x)
(2) for every axiom in T , T A
(3) T A
then is consistent relative to T .
Proof. Let us prove the contrapositive. So T {} is inconsistent. Then there is a nite subset
{1 , . . ., n } T such that S0 = {1 , . . ., n , } is inconsistent. Thus S0 has no model.
Hence, = 1 . . . n .
Recall from Logic, if is any sentence, then = implies = xA(x) A ()
So by (), = xA(x) (1 . . . n )A .
That is, = xA(x) (1A . . . nA A ).
Now by (1) and (2), T A . But by (3), T A . Thus T A A , so T is
inconsistent.
2
Denition. A class A is called an inner model of a theory T (with respect to a theory S) if
(1) A is transitive
(2) for every in T , we have S A
The previous theorem helps us nd inner models. We now have several useful classes (V , L ,
H , WF , L) for the application of the criterion to obtain relative consistency results.
An example is the following theorem of Godel: L is an inner model of ZFC+V = L+GCH.
12
Lecture 7
Chapter 2. Infinitary Combinatorics
Denition. A cardinal is a successor cardinal if = + for some cardinal .
A cardinal is a limit cardinal if is not a successor cardinal.
Remark. Every innite cardinal is a limit ordinal.
Denition. Suppose that P = (P, 6) is a partial order. A set A P is a cover of P if
(p P )(q A)(p 6 q)
The conality of P is the least cardinal for which there exists a cover of P of size :
cf(P) = min{A : A P is a cover of P}
Notes.
(1) cf(P) 6 P
(2) cf() 6 for every ordinal
(3) cf( + 1) = 1 (since {} is a cover)
(4) cf(P(X), ) = 1
Lemma 1. Suppose Ord. Then there exists a function f : cf() such that
(1) range(f ) is a cover of
(2) f is strictly increasing: < f () < f ()
Proof. Let A be a cover of of size cf(), say A = { : < cf()}. Dene f : cf() by
f () = max , sup{f () + 1 : < }
Why is f welldened for all cf()? By minimality of A, f is welldened. Then (1)
and (2) are immediate.
2
Denition. We say that f : is a conal map (or that f maps into conally) if
range(f ) is unbounded in .
Proposition 2. The following are equivalent:
(1) cf() =
(2) is the least ordinal such that there is a conal map f :
Proof.
(1) (2). Let A be a cover of cardinality cf() = . So is a cardinal. By Lemma 1,
there is an increasing function f : such that range(f ) is a cover of . So
range(f ) is unbounded in and is therefore conal.
is the least ordinal with this property. For if < and g : , then range(g) 6
 6 < = cf(), so range(g) is not a cover and hence is not conal.
Thus is the least cardinal such that there is f : conally.
(2) (1). Let f : be conal, and is minimal, so  = . Then A = range(f )
is a cover of , so cf() 6 A 6  = cf().
2
13
Denition. A cardinal is regular if there is no family {A : < } such that =
with A  < and < .
<
A ,
If is not regular, we say that is singular.
S
Examples. 0 is regular, as is 1 . But is singular, since =
singular.
n<
n . And i is also
Proposition 3. A cardinal is singular i cf() < .
Proof.
() Suppose = cf(), and < and A is a cover of size , say A = {A : < }.
Let A = { < : < + 1} = + 1.
Then A  =  + 1 =   < , and < , and =
then A for some since A is a conver of .)
<
A . (Take any 6 ,
So is singular.
() Suppose is singular, and let be minimal such that =
<
A with A  < .
Case 1: some A is a cover of . Then cf() 6 A  < , and we are done.
Case 2: otherwise. So no A covers . Thus for all < , there is such that
A , < . Let A = { : < }. Then A is a cover of , and A 6 < , so
cf() < .
2
Corollary 4. cf(+ ) = + for every innite cardinal .
Proof. Let = cf(+ ). Write =
<
A , with A  < + .
S
Then + = < A = sup{A  : < } = max , sup{A  : < } .
But A  6 , so + = max{, } = = cf(+ ).
Lecture 8
Lemma 5. Suppose is a limit ordinal, and f : is strictly increasing and conal. Then
cf() = cf().
Proof. First, cf() 6 cf(). Let g : cf() be as in Lemma 1.
Now, range(f g) is a cover of . So cf() 6 cf().
Second, cf() 6 cf(). Let h : cf() be conal, and let
k() be the least < such that f () > h().
Claim: k : cf() is conal.
Why? If i < then f (i) < . There exists with f (i) 6 h(),
and k() = min{ : f () > h()}, and so i < k() = .
(If < i then f () < f (i) and we get a contradiction.)
So cf() 6 cf().
14
cf()
f ()
h()
Corollary 6. cf(cf()) = cf().
Proof. By Lemma 1, there is a strictly increasing conal map f : = cf() . By Lemma
5, since cf() is a limit ordinal (since cf() is a cardinal), it follows that cf() = cf(), i.e.
cf(cf()) = cf().
2
Thus cf() is always a regular cardinal.
Corollary 7. 0 is regular.
Proof. Any cover is innite.
Corollary 8. If is a limit ordinal, then cf( ) = cf().
Proof. By Lemma 1, there is a strictly increasing conal map f : cf() . Dene g : cf()
by g() = f () .
So cf( ) = cf(), by Lemma 5, because g is strictly increasing and conal.
Exercise. Prove that for all Ord, we have 6 .
Corollary 9. If is an uncountable regular limit ordinal, then = .
Proof. We have
=
=
6
6
cf( )
cf()
by regularity
by Corollary 8
by the exercise
So = .
Observation. The converse is false. I.e., there exists a singular with = .
For example, let 0 = and n+1 = n , and = supn< n .
Then = , but cf( ) = .
Conality enables us to improve Cantors Theorem, that 2 > .
Denition. Suppose that {Xi : i I} are pairwise disjoint sets, with Xi  = i . We dene the
(cardinal) sum
X
[
i =
Xi
iI
iI
and we dene the (cardinal) product
Y
S
i = f : f is a function from I into X, with f (i) Xi i I
iI
Theorem 10 (K
onig). Suppose < , for all < . Then
<
Proof. Clearly,
<
<
. (Exercise: use the natural inclusion map.)
<
<
Note
<
S
Q
S
.
{}
{} . Suppose that G :
=
<
<
15
<
We show that G is not a surjection.
Dene h
as follows. For < ,
<
h() = min : \ { G(, ) : < }

{z
}
has cardinality 6
where is the projection of
onto .
<
h is welldened, since < . Now it is evident that h
/ range(G) by construction,
h() 6= G(, ) for all < .
2
Corollary 11. If > cf(), then > .
Proof. Write =
i , with i < . Then by Theorem 10,
i<cf()
i <
= cf() 6
i<cf()
i<cf()
2
Corollary 12. If > 0 , then cf(2 ) > .
Proof. Let = cf(2 ). By Corollary 11, (2 ) > 2 .
But (2 ) = 2 = 2max{,} . So 2max{,} > 2 .
So max{, } > , and so = cf(2 ) > .
Lecture 9
We know from Corollaries 11 and 12 that 0 6 6 cf() and > .
Three natural questions occur:
(1) What is the least such ?
(2) For the least above,
(a) how large is relative to ?
(b) how large is cf() relative to ?
(3) How are 2 and cf() related (if at all)?
Key examples to think about:
(1) What is 0 , given 2n , n < ?
(2) What is 11 , given 2 , < 1 ?
Answers were given by Shelah and Silver.
In this lecture, we examine some hypotheses which bear on these questions.
16
Denition.
(1) The Generalised Continuum Hypothesis (GCH) is the statement:
( > 0 )(2 = + )
(2) The Continuum Hypothesis (CH) is the statement:
20 = 1
(3) The weak Generalised Continuum Hypothesis (wGCH) is the statement:
+
( > 0 )(2 < 2 )
(4) The weak Continuum Hypothesis (wCH) is the statement:
20 < 21
Remark. GCH wGCH, and CH wCH.
We look at how GCH impacts the operation of cardinal exponentiation and innite products of
cardinals.
Notation. We write
Y = {f : f is a function from X into Y }.
Proposition 13. Let > 0 . Then is regular i () ( < ) =
<
Proof.
() Suppose < = cf(). Note key observation: if f , then range(f ) is bounded
in , and so there is < such that f .
So
and clearly
<
. Thus () holds.
<
() Suppose () holds. If < and f : then there is < with f by ().
So f is not conal in . Thus < cf() for all < .
So 6 cf() 6 . I.e., is regular.
Lemma 14. For all > 0 , if < cf() and ( < )(2 6 ), then = .
Proof. If < cf(), then
[
  =
<
()
X
X
X
=
=

<
<

<
<

<
= = 6
<
(Note that () uses the key observation from the proof of Proposition 13.)
Thus GCH answer all of the opening question about , cf() and 2 . GCH implies that
= cf(), and = cf() = + .
17
Denition. For cardinals , , we dene the weak power of to to be
X
< =
<
Lemma 15. For all > 0 ,
(1) if = cf() and ( < )(2 6 ), then < =
(2) Assume GCH. Then = cf() i < = .
Proof.
(1) By Lemma 14, for all < = cf(), we have = .
P
So < =
= = .
<
<
(2) Note GCH implies that ( < ) 2 = + 6 .
() This is part (1).
() If = cf() < , then by Konigs Theorem, > , and so < < .
Lemma 16. For all > 0 ,
(1) > < > 2< >
(2) If = + , then < = 2< = 2
Proof.
(1) This is trivial.
(2) Note 2< = 2 and < = . Hence 2 = 2< 6 < = 6 (2 ) = 2 .
Corollary 17. GCH holds i ( > 0 )(2< = ).
Lemma 18. If is a limit cardinal, then 2 = (2< )cf() .
Proof. Write =
i , with i < . Then
i<cf()
2 = 2
i<cf()
2i 6
2< = (2< )cf() 6 (2 )cf() = 2
i<cf()
2
Lecture 10
Theorem 19 (Bukovsk
yHechler). Let > cf() be such that
0 0 6 < 2 = 20
()
Then 2 = 20 .
Proof. Since cf() < , by () and wlog, cf() 6 0 .
Now, by Lemma 18, 2 = (2< )cf() = (20 )cf() = 20 .
Denition. The Gimel function () is dened on the class of innite cardinals by
() = cf() .
18
Why is () important in current research in cardinal exponentiation?
Note. If , > 0 , then cf( ) > . (Exercise: use Konig again.) In particular, () > .
Denition. The Gimel Hypothesis (Solovay, 1974) states: for every singular cardinal ,
() = max 2cf() , +
Remarks.
(1) GCH implies the Gimel Hypothesis. (Exercise.)
(2) We proved in ZFC that () > 2cf() and () > . The Gimel Hypothesis asserts
that () is the least ZFCpermitted value (relative to the other values of () for
regular ).
For example, under CH, the Gimel Hypothesis says 0 = +1 .
(3) Why only for singular cardinals? We know the for regular cardinals , we have
() = 2 , and it follows from work of Easton (1964, 1970) that for any reasonablydened cardinalvalued function F dened on regular cardinals such that
(i) < F () 6 F ()
(ii) cf(F ()) >
it is consistent with ZFC and the Gimel Hypothesis that for all regular cardinals
we have 2 = F ().
So, in particular, the Gimel Hypothesis does not imply GCH.
Failures of the Gimel Hypothesis are an active subject of research. An equivalent statement is
the following.
Denition. The Singular Cardinals Hypothesis (SCH) states: for every singular cardinal ,
if 2cf() < , then () = +
Exercise. Show that the SCH is equivalent to:
for all regular cardinals , , if 2 < , then =
Corollary 20 (Bukovsk
y). The continuum function 7 2 can be dened in terms of the
Gimel function. In other words, if one knows , then one knows 2 for all .
Proof.
(a) If = cf(), then 2 = ().
(b) If is a limit cardinal and the continuum function is eventually constant below (as
in Theorem 19), then (check)
2 = 2< ()
(c) If is a limit cardinal and the continuum function is not eventually constant below
, then
2 = (2< )
because cf(2 ) = cf(). (Check.)
19
Thus given () we can compute 2 .
To close this section on cardinal exponentiation, we note that the following theorems of Silver
and Shelah illustrate how dierent exponentiation of singular cardinals is.
Theorem.
(1) (Silver, 1974) If 2 = +1 for all < 1 , then 21 = 1 +1
(2) (Shelah, 61994) pp( ) < 4
In simplistic terms, 0 < 4 + (20 )+ .
So, for example, if 20 < , then 0 < 4 .
Note: pp() is dened in terms of conalities of products of regular cardinals.
We leave exponentiation and turn to the study of prediction principles and families of large
subsets of limit ordinals.
Denition. Suppose that is a (limit) ordinal. We say that C is a club of (or in ) if
(1) C is closed in i.e., if C has sup then sup C
(2) C is unbounded in i.e., < C < < .
(Club is from closed unbounded.)
Lecture 11
Remarks.
(1) If is a successor ordinal, then clubs are not very interesting, since e.g. if = + 1,
then {} is a club.
(2) C is closed i C contains all of its limit points: is a limit point if ( <
) ( C) ( < < ) (where is a limit ordinal)
Examples.
(1) is closed in .
(2) If cf() > 0 , then lim() = { < : is a limit ordinal} is club in .
(3) If C is unbounded in , then the set C of limit points of C is club in .
(4) If > 0 and cf() = 0 , then any conal sequence in is club in .
(5) If is a limit cardinal, then { < : is a cardinal} is club in .
(6) If is a limit cardinal, then {+ : hi} is not club in (it is not closed)
Remark. The concept of a club of is most interesting when is an ordinal of uncountable
conality.
Remark. If Ci is closed in , then
Ci is also closed in .
iI
However, arbitrary intersections of clubs will not in general be club. Suppose that = cf()
and f : is conal in .TConsider C = { < : > f ()} = [f (), ). Then C is
club in for all < , but
C = .
<
20
Proposition 23. Suppose cf() > 0 , and < cf().
Then
C is club in whenever C is club in for all < .
<
Proof. Let C =
C . For later, wlog, < C C .
<
Clearly, C is closed. We show that C is unbounded.
C
Let < be given. Let f () = min C \ ( + 1) .
(I.e., remove and its predecessors.)
g(g()) r
r
r
Let g() = sup{f () : < } < so g : . (Recall < cf().)
g()
r
r
r
Dene g0 () = , and gn+1 () = g(gn ()), and g () = sup gn ().
n<
It follows that < g () < , and g () C.
Corollary 24. Suppose = cf() > 0 and < .
If C is club in for all < then
C is club in .
<
Denition. Suppose is a limit ordinal. A set S is stationary in if S C 6= for every
club C of .
Examples. Let = cf() < = cf(), with > 0 .
Let S = { : cf() = }. Then S is stationary in .
Proof. Recall f is continuous if f () = sup{f () : < }.
Given a club C, dene f : C, strictly increasing and continuous. Note f () C and
f () S , so S C 6= .
2
Example. If = 1 and = 0 , then S01 is actually a club in 1 , since S01 = lim(1 ).
However, in general, this is false. S does not contain a club of since any club must contain
elements of every conality < .
Proposition 25.
(1) If C is club and S is stationary in then C S is stationary in .
S
(2) If < cf() and {X : < } is stationary in , then some X is stationary in .
Proof.
1. By Proposition 23.
2. If C X = for all < , then C =
S
C {X : < } = , contradiction.
21
C is club by Proposition 23, but
<
Denition. Suppose = cf() > 0 and X for < . We dene the diagonal intersection to be the set
X = < : ( < )( X )
<
Proposition 26. Suppose that = cf() > 0 and each C is club in . Then C = C is
<
club in .
Lecture 12
Proof. Clearly C is closed. To see that C is unbounded in , dene g() for < as follows:
\
g() = min
C \ ( + 1)
<
 {z }
club by Prop 23
By Proposition 23, g is welldened, and < g() < .
Let g0 () = , and gn+1 = g(gn ()), and g () = sup gn ().
n<
Claim. g () C.
Why? Suppose < g (). Then n (m > n) ( < gn () C ), and we know that C is
closed, so g () C .
Thus C is club in .
Theorem 27 (Pressing Down Lemma / Fodors Lemma). Suppose = cf() > 0 and
S is stationary in . If f : S is regressive, i.e. S f () < , then there
is some stationary S S such that f S is constant. I.e., < such that f 1 ({}) is
stationary.
Proof. Otherwise, there exists a club C such that C f 1 ({}) = . Then, by Proposition
26, C = C is club.
<
In particular, it follows that C S 6= . Pick C S.
If < , then C . So f () 6= . So f () > (1).
But S, and so f () < (2), contradicting (1).
Theorem 28 (Ulam). Suppose = cf() > 0 . Then there exists a family of disjoint
stationary subsets of .
Proof. There are two cases.
Case 1. is a limit cardinal.
Consider {S : = cf() < }. Recall: S = { < : cf() = }
This is as required.
Case 2. = + .
Let {A : < , < } be an Ulam (, )matrix on , i.e.
22
(1) A A = if < <
S
A 6 .
(2) \
<
We can nd A with these properties as in Exercise Sheet 1:
[, + ), let f : be a surjection, and A = { < : f () = }.
Note, by (2),
A is stationary (by Proposition 25).
<
By Proposition 25 again, for some < , A is stationary in .
So there exists such that E = { : = } is stationary.
Thus {A : E} is the required family.
Corollary 29. Let = cf() > 0 . There are 2 stationary subsets of .
Proof. For the family
{B : < } of stationary subsets in Theorem 28 and nonempty X ,
S
B .
2
take F =
X
Prediction Principles
Intuitively, a settheoretic prediction principle is a list of guesses (or approximations) of subsets
of a cardinal. In this sense, CH is a prediction principle:
CH hX : < 1 i Y < 1 Y = X
AC is also a prediction principle:
AC hX : < 2 i Y < 2 Y = X
The syntactic form of these statements leads one to consider stronger statements.
Denition. Let S be a stationary subset of = cf() > 0 .
The diamond on S, denoted 3S , is the statement: hA : Si such that
(1) A
(2) for all X , the set { S : X = A } is stationary in
If S = , we write 3 for 3S , and we write 3 for 31 .
Lecture 13
Remarks.
(1) S S implies 3S 3S . (Recall: for stationary sets.)
(2) 3+ 2 = + , so 3 CH.
(3) 3S is equivalent to: hfa : Si such that
(i) f :
(ii) for all f : , the set { S : f  = f } is stationary.
23
Proof.
(1) is trivial.
(2) We show P() {A : + }.
If X , then there is < + with X . The set E = { < + : X = A } is
stationary. Pick E with > . Now X = X = A .
(3) will be on Exercise Sheet 3.
Remark. V = L implies 3S , but we will show the relative consistency of 3 by forcing.
One of the most striking and important applications of 3 is to prove the existence of a Suslin
tree.
Theorem 30. 3 implies that there exists a Suslin tree.
It will be convenient rst to recall some denitions relating to trees.
Denitions.
(1) A tree is a partial order T = hT, 6T i such that for all x T , the set x
b = {y T :
y <T x} is wellordered.
(2) The ordertype of x
b under <T is called the height of x in T, denoted htT (x).
(3) If Ord, the th level of T is the set T = Lev (T) = {x T : htT (x) = }.
S
T  .
We write T for the partial order T restricted to the set T  , where T  =
<
(4) A branch of T is a linearly ordered subset b such that x b y <T x y b.
We say that b is an branch if b has ordertype .
(5) A branch b is maximal if b is not properly contained in any other branch of T. AC
implies that every branch can be extended to a maximal branch.
Remark. x
b is a branch. If x has no successors in T, then x
b {x} is maximal.
(6) An antichain in T is a subset A T such that no two elements of A are comparable
in <T . An antichain A is maximal if it is not properly contained in any other antichain
of T.
AC implies that every antichain can be extended to a maximal antichain.
Remark. If T 6= , then T is a maximal antichain.
(7) Let be an ordinal and a cardinal. A tree T is a (, )tree if
(i) < , T 6=
(ii) T =
(iii) < , T  <
So a (, )tree has height and width less than .
(8) T has unique limits if whenever is a limit ordinal, if we have x, y T with x
b = yb,
then x = y.
24
(9) A (, )tree T is normal if
(i) T has unique limits
(ii) T0  = 1
(iii) If , + 1 < , x T , then there are distinct y1 , y2 T+1 such that x <T y1
and x <T y2
(iv) If < < and x T , then there is y T such that x <T y
(10) Let be an innite cardinal. A tree is a normal (, )tree.
(11) A Suslin tree is an 1 tree with no uncountable antichains.
Lecture 14
We now prove Theorem 30, that 3 (Suslins Hypothesis).
Proof. Assume 3 and let hA : < 1 i be a 3sequence. By transnite recursion, we construct
a Suslin tree T = hT, <T i, with T = 1 , such that
S
T
(1) T =
<1
(2) T is a (normal) (, 1 )tree
The elements of T are the nite ordinals, and for innite , the elements of T will be
ordinals from the set { : 6 < + }.
How do we go?
(0) T0 = {0}.
(1) If n and Tn+1 is dened, dene Tn+2 by taking the elements of Tn in turn and,
for each element, picking the next two unused nite ordinals to be its successors in
Tn+1 .
(2) If > and T+1 is dened, dene T+2 by using the ordinals in { : 6 <
+ } to provide each element in T with two successors in T+1 .
This is possible, since T  6 0 .
(3) If > and is a limit ordinal, and T is dened, for each x T  , pick an
branch bx with x bx , subject to the condition: if A is a maximal antichain in
T , then bx A 6= . (Remember that A is from the 3sequence one of the
guesses.)
Why is this possible?
There exists a A such that x and a are comparable: x 6T a or a <T x. If x 6T a
then pick bx to be some branch extending b
a {a}, and if a <T x then pick bx to
be an branch extending x
b {x}.
If A is not a maximal antichain in T , pick bx to be any branch containing x.
For each T  , add a 1point
branch bx . This is possible since
Let T =
from { : 6 < + } to the
extension
T  6 0 .
T . Then T is an 1 tree. This is clear.
<1
25
It remains to show that T is Suslin, i.e. T has no uncountable antichains. Its enough to
show that every maximal antichain of T is countable.
Suppose that X 1 , and X is a maximal antichain in T. Consider
C = { 1 : = and X is a maximal antichain in T }
Claim. C is a club.
Proof of claim. Its easy to see that C1 = { < 1 : = } is a club, and check that
C2 = { lim(1 ) : X is a maximal antichain in T } is a club.
Why is C2 club? Easily C2 is closed: n C2 n .
C2 is unbounded. Let 0 < 1 . We nd C2 with > 0 . Given n , dene n+1
to be the least limit ordinal > n such that every element of Tn is comparable to
some element X . Let = sup n . So C2 , and so C = C1 C2 is a club.
n<
So now by 3, the set S = { < 1 : X = A } is stationary in 1 , and so C S 6= .
Pick C S. Then X = A , and so A is a maximal antichain in T .
By construction of T (and every T , > ), every element of T lies above some element
of A = X (i.e., is comparable with some element in X ).
So X is a maximal antichain in T.
So X = X, and so X = X  6  6 0 , since < 1 .
Thus X is countable.
It is natural to ask (and Istvan Juh
asz did) whether this is optimal. Could one replace 3 by a
weaker principle? In this connection, we mention the club principle, .
Denition. asserts that there exists hC : lim(1 )i such that
(1) sup C =
(2) for every uncountable X 1 , the set { lim(1 ) : C X} is stationary in 1 .
Remarks.
(1) 3
(2) Juh
aszs Question: does imply the existence of a Suslin tree?
Lecture 15
Remark. (Regarding 3 6=SH.)
Some model theory allows one a less ad hoc proof that the set C in the proof is a club.
Recall that X is a maximal antichain in T. Consider the structure TX = hT, 6T , Xi.
The set { lim(1 ) : TX  TX } is club by question 5(iv) on Exercise Sheet 3. This is
the set C in the proof.
26
We shall return to 3 later. We now introduce a new axiom MA (Martins Axiom) and use it
to show that there are no Suslin trees.
Denition. Suppose P = hP, 6P i is a partial order. Wlog, P has a least element OP .
(1) Two elements p, q P are compatible if (r P )(p 6 r q 6 r). Otherwise p, q
are incompatible.
(2) A subset G P is directed if (p G)(q G)(r G)(p 6 r q 6 r).
(3) A subset D P is dense in P if (p P )(d D)(p 6 d).
(4) Let be an innite cardinal. We say that P satises the chain condition if every
antichain in P has cardinality less than .
(5) If = 1 , then the 1 chain condition is called the countable chain condition
(CCC).
Caveat. Some sources reverse the order in the above (and subsequent) denitions. E.g., they
say p, q are compatible if (r P )(r 6 p r 6 q).
Denition (cont.)
(6) Suppose D is a family of dense sets in P, and
V G P is directed. We say that G is
G D 6=
Dgeneric (or generic relative to D) if
DD
Proposition 31. Suppose D = {Dn : n < } is a countable family of dense sets in the partial
order P. Then there exists a Dgeneric set G P .
Proof. By induction, we dene hdn : n < i.
D0 6= , so pick d0 D0 . Given dn Dn , since Dn+1 is dense in P, there exists dn+1
Dn+1 with dn 6 dn+1 .
Now G = {dn : n < } is a Dgeneric set, as required.
In general, we cannot improve Proposition 31 to uncountable families of dense sets. Uncountable antichains can wreak havoc. However, if we require that P has the CCC then there is a
relatively consistent statement.
Denition. Let be an innite cardinal. The statement MA asserts: for every CCC partial
order P and every family D = {D : a < } of dense subsets in P, there exists a Dgeneric
set G P in P.
^
MA .
Martins Axiom is the statement:
0 6<20
Remark. CH MA.
So for this reason, we generally tacitly assume CH when we apply MA.
Proposition 32.
(1) ZFC MA0
(2) MA 20 >
(3) ZFC MA20 .
27
Proof.
(1) is immediate from Proposition 31.
(2) We must nd a poset P and some dense sets.
Let hf 2 : < i. We aim to nd g 2 with
g 6= f .
<
We pick P to approximate g. Let P be the set of nite functions from into {0, 1} :
P = f : dom(f ) , f : dom(f ) {0, 1}, dom(f ) < 0
and dene f 6P h i hdom(f ) = f , i.e. h extends f .
P = (P, 6P ) is CCC since P  = 0 .
Let En = {p P : n dom(p)}. Easily, En is dense in P.
For < , let
D = p P : n dom(p), f (n) 6= p(n) .
Then D is dense in P.
Now let D = {D , En : < , n < }. By MA , there exists a Dgeneric set G.
S
Let g = G. Note: g is a function since G is directed.
We have dom(g) = since G En 6= , and range(g) {0, 1}.
And, for all < , we have g 6= f since G D 6= .
Thus g 2 and g 6= f for all < . So 20 > .
(3) is immediate from (2).
Lecture 16
The axioms MA1 and MA say that the universe contains generic sets (for families of dense sets
in CCC posets, provided the families are of size < 20 ). We now prove:
Theorem 33. MA1 implies that there are no Suslin trees.
Lemma. If T is a Suslin tree, then every chain in T is countable.
Proof. By contradiction. Suppose C is an uncountable chain in T. Wlog, C is maximal. So
C T 6= for all < 1 . So there exists c C T , and by normality (iii) of a Suslin
tree, there exists a+1 T+1 \ C such that {a+1 , c+1 } is an antichain in T+1 .
Consider A = {a+1 : < 1 }. Clearly, A is an antichain, and A = 1 , contradiction. 2
Why can one not simply dene inductively an uncountable chain in a Suslin tree by using
normality to grab a chain member at each next level?
This will not work, since at limit levels , there is no uniform single choice for an element c to
continue the chain at the limit level above all the lower elements c , < . By normality, one
can certainly ascend onebyone, but not above a boatload.
28
What is the conality cf(T) of a Suslin tree T?
S
If C is a cover of a Suslin tree T, then T = {b
c : c C} C; so if C were countable, then some
b
c would have cardinality 1 and be an uncountable chain in T. This is impossible since Suslin
trees have no uncountable chains by the Lemma. Hence there is no countable cover of T. Since
T has cardinality 1 , any cover has cardinality at least, and hence exactly, 1 . It follows that
cf(T) = 1 .
Back to Theorem 33.
Proof. Towards a contradiction, suppose that T is a Suslin tree. Then T is a CCC partial order.
Let D = {y T : htT (y) > } for < 1 . Clearly, D 6= and D is dense in T.
By MA1 , there is a Dgeneric set G for D = {D : < 1 }. Clearly, G is a chain in T
since G is directed, and G = 1 , contradicting the lemma.
2
3 and MA have many applications across ordinary mathematics. Here is a more recent short
application of MA to abelian group theory.
Example (sketch). Let A be an innite abelian group. We write A = Hom(A, Z) for the dual
of A. We say that A is free if A
= iI Z for some set I.
Let be an innite cardinal. We say that A is free if every subgroup of A of cardinality
< is free.
For example, Z is 1 free.
The Trivial Dual Conjecture for states: there exists a free group A with A = 0.
This is denoted TDU .
Remarks.
(1) TDU0 is true: e.g., Q = (Q, +).
(2) What about TDU1 ?
TDU1 is true, but its a decent theorem due to Eda (1989).
(3) What about TDU2 ?
Theorem. MA implies that if A is 2 free and A < 20 , then A is separable.
Denition. An abelian group A is separable if every nite subset of A is contained in a
free direct summand of A.
Equivalently, any pure cyclic subgroup of A is a free direct summand of A.
B 6 A is pure in A if: b B, n , if A = x (nx = b), then B = x (nx = b)
Proof. Let h A. Suppose that hhi is pure cyclic. It is enough to show that hhi is a
free abelian summand. Let (h) = 1. We use MA to extend to a homomorphism
Hom(A, Z).
Let P = (P, 6P ) be the set
29
: Hom(A , Z), (h) = 1, A is a pure nitelygenerated subgroup of A
Then 6 if extends .
Some facts which require proof:
P is a CCC partial order
For a A, let Da = { P : a A }
Then D is dense in P.
By MA, there exists a Dgeneric set G since D < 20 = A.
S
Let =
G. Then Hom(A, Z) and by the First Isomorphism Theorem,
A/ ker
= Z. So ker is a direct summand of A. This completes the proof (modulo
some algebra).
2
Open question. What is the status of TDU ?
Lecture 17
Chapter 3. Forcing
In this chapter, we present forcing, a method for constructing extensions of models of (subtheories of) ZFC. We then use the technique to prove the independence of the Continuum Hypothesis,
3, , and some other combinatorial principles, but also V = L.
There are several approaches to the presentation of forcing. We shall follow the presentation of
Shelah in Proper and Improper Forcing, chapter 1.
We start from the assumption that ZFC is consistent, and has a countable model M. Wlog, we
shall also assume:
(1) M is a standard model: if M = (M, E M ) then E M = MM . In other words, the
membership predicate of M is real membership.
(2) The universe of M is a transitive set: x y M x M .
These additional assumptions come for free and they simplify the presentation. (We can justify
(1) and (2) either using the fact that L, the universe of constructible sets, is a standard model
and use the Downward L
owenheimSkolem theorem to obtain a countable elementary submodel;
alternatively, use the Reection Principle to cut down to a V for an arbitrary large nite
fragment of ZFC, and take a countable elementary submodel of V ; nally, use the Mostowski
Collapse to obtain an isomorphic transitive model.)
So throughout this presentation, we shall assume that M is a countable transitive model satisfying (1) and (2). We call M a CTM.
We wish to extend M.
Denition. A forcing is a partial order P = (P, 6P ). We shall, wlog, assume that forcings
have minimal elements. It is also not necessary to assume that P is antisymmetric. In
some sources, this is called a prepartial order or quasiorder.
The elements of a forcing are called (forcing) conditions.
30
The extensions we construct are dened using forcings and generic lters in V . These forcings
and generic sets control the truth values of sentences in the extensions.
Denition. Suppose A is a model of ZFC (or a fragment of ZFC), and P A (so P and 6P
belong to A). We say that G P is a lter in P if
(1) G is directed: (x, y G) (z G) (x 6 z y 6 z)
(2) G is downwardclosed: x 6 y G x G.
Note. The denition does not require G V to belong to A.
G A
q
V
Denition. We say a subset D P is dense open in P if
(1) D is dense in P: p P (d D)(p 6 d)
(2) D is open (upwardclosed): (p, q P )(q 6 p q D p D)
We say that a lter G in P is generic (over A) if G D 6= for every dense open set
D P in A.
Our aim is to build an extension M[G] for a given CTM M and a generic lter G over M with
the properties:
(1) M M[G], i.e. M is a substructure of M[G]
(2) M Ord = M[G] Ord
(3) G M[G]
(4) M[G] is the smallest CTM such that (1), (2), (3) hold.
How will it work?
In intuitive terms, if one thinks of the conditions p P as potential information about G, then
{p : p G} should be contradictionfree, and this amounts to saying that G should be directed.
If q 6 p G, this amounts to saying that stronger information (p) implies weaker information
(q) about G, so G should be downwardclosed.
How about genericity?
The genericity condition on G amounts to saying that G is generic in the sense of nonspecic
or random, i.e. G contains consistent general information.
Are there any generic sets over a CTM? And where?
Proposition 1. Suppose M is a CTM, and P M is a forcing. Then there exists a generic lter
G ( V ) on P over M. In fact, for any p P , there is a generic G over M such that p G.
Lecture 18
Proof. (See Proposition 31 in Chapter 2.)
In V , M is countable, and so there is a list {Dn : n N} of all the dense open sets in P
31
that belong to M. Now, inductively, pick pn+1 > pn with pn+1 Dn+1 (using denseness
of Dn+1 ) and p1 = p. Let G = {q P : n : q 6 pn }.
G is a generic lter over M, as required.
Note. This argument happens in V , not in M.
In general, G
/ M.
Proposition 2. Suppose P is separative, i.e. p P , q, r P , with p 6 q, p 6 r and qr.
(Every condition has two incompatible extensions.)
Then G
/ M , for any generic lter G in P over M.
Proof. By contradiction. Suppose G M. Then the set D = P \ G belongs to M. But D =
6
and D is dense open, so G D 6= . Contradiction.
2
Now let us try to motivate the construction of the forcing extension.
First, note that we do need to consider a new sort of modeltheoretic extension. Why? Two
general reasons:
(1) ZFC and its subtheories are not modelcomplete.
A theory is modelcomplete if whenever A B (substructure) then A B (elementary
substructure), i.e. its submodels are elementary submodels. For example, V V1 , but
V 6 V1 .
Recall. A substructure A B is an elementary substructure of B if for every formula
(x1 , . . ., xn ) and a1 , . . ., an A, we have A = [a1 , . . ., an ] B = [a1 , . . ., an ].
(2) The CTM M might (for all we know) be a model also of V = L. Then if a submodel
M of M contains all of the ordinals of M then M = M, because everything in M is
constructible.
The other way to extend M is to nd an elementary extension M+ . But this does not help to
prove independence, because T h(M) = T h(M+ ) if M M+ , so e.g. if M = CH then M+ = CH.
Recall. If A is a structure, the diagram of A is the collection of atomic sentences and
negations of atomic sentences in the language A ( with constants a for each a A)
which are true in A.
The elementary diagram of A is { : A = , is a sentence in the vocabulary of A }.
In modeltheoretic terms, we are looking for a concept intermediate between the diagram of M
and the elementary diagram of M. Where should we go?
One way to construct an extension of a model A is to add new constants
a (for a A) and a witness constant g, and observe that the elementary
diagram with the sentences a6= g for each a A is nitely satisable, hence
+
by the Compactness Theorem has a model A+ , and g A
/ A.
q g A+ A+
q a A
Of course, this is too much, since A+ is already an elementary extension of A, but the idea of
using names or new constants is a useful strategy. We therefore design names that are more
32
complicated sets in M, dened by transnite recursion for elements of M and P, and whose
interpretations will depend upon the generic set G. In general, the referents of these names
cannot be computed or identied in M, since G
/ M (usually).
Denition. Suppose P M is a forcing. We dene by induction on Ord the Pnames (or
names) of rank 6 as follows.
A set is a Pname of rank 6 if = {(pi , i ) : i < i0 }, where pi P , i is a Pname of
rank 6 i < , and i0 Ord.
A Pname is a Pname of rank 6 for some Ord.
We write MP for the collection of Pnames that are elements of M.
We let the (name) rank of , rkn ( ) = if is a Pname of rank 6 , but not of rank
6 for any < .
: p P, b a}.
We dene for a M , by induction on rk(a), a Pname a = {(p, b)
Note that a is a Pname. (Check.)
Examples of Pnames.
(1) rank 6 0:
(2) rank 6 1: , {(pi , ) : i < i0 }.
Lecture 19
Denition (cont.) We dene the Pname = {(p, p)
: p P}.
We dene the revised (name) rank rkr ( ) =
0Sif = a for some a M
{rkr () + 1 : (p, ) for some p P}
Notation. V P is the class of all Pnames. We use f , , a for Pnames that are not of the form
e
a for a M.
e e
When no confusion arises, we lapse into using a instead of a.
This completes the denitions of the names.
We turn next to dene the values or interpretations of the names. This denition by transnite
recursion is given in V (not in M), since in general G
/ M.
Denition. Suppose P M is a forcing and G P is a generic lter over M. We dene by
induction on Ord.
(1) If is a Pname of rank 6 , say = {(pi , i ) : pi P, i < i0 } where i is a Pname
of rank 6 i , i < , then the interpretation (or value) of relative to G is
[G] = {i [G] : pi G, i < i0 }
(2) M[G] = { [G] : M and is a Pname}.
Note: [G] is also written G , valG ( ), val(, G) and other variants.
Write M[G] =M[G]M[G] .
33
In this course, M is called the ground model, and M[G] the generic extension of M relative
to G (by G).
Finally, we dene the forcing relation
P (or
) which connects truth in generic extensions
with denable sets in M and the combinatorial properties of P.
Denition. Suppose P M is a forcing, 1 , . . ., n are Pnames, and (x1 , . . ., xn ) is a formula
in the vocabulary of set theory (possibly with a unary predicate symbol M (x) for M).
We say that a condition p P forces (1 , . . ., n ), denoted p
P (1 , . . ., n ), if
h
i
M[G] 1 [G], . . ., n [G]
for every generic lter G in P over M such that p G.
Remark. The denition of
P is given in V , apparently. However, we shall prove that the
binary predicate
P is definable in M.
This is a critical fact about forcing.
We can now state precisely the Forcing Theorem.
Theorem 3. Suppose M is a CTM, and P M is a forcing. Then for every generic lter G in
P over M, there exists a CTM M[G] (dened as above) such that
(1) M[G] is an extension of M. M M[G], G M[G].
(2) Ord M = Ord M[G], i.e. M and M[G] have the same ordinals.
(3) For every sentence in the vocabulary of ZFC (possibly with a unary predicate
symbol M (x)),
M[G] i (p G)(p
P )
(4) The predicate
P is denable in M.
(5) M[G] is minimal with the above properties: if M+ is any transitive model of ZFC
with M M+ , G M+ , then M[G] M+ .
It will take some preparatory work to prove M[G] has properties (1)(5).
Let us consider some examples to see how this theorem helps us to establish independence
results.
Example 1. Adding a Cohen real.
o
Let P = f : f : dom(f ) {0, 1}, dom(f ) < 0 , 6P is extension of functions, i.e.
f 6 g gdom(f ) = f .
So we have P M. In M, the sets Dn = {f P : n dom(f )} are dense open and
denable in M, so belong to M. So G Dn 6= .
What do we know about g =
Lecture 20
G? g is a function from to {0, 1}.
S
If G is generic, then g = G is a real, i.e. g : {0, 1}. And since P is separative,
G
/ M and so g
/ M. So g is a real which is new.
34
If we could add enough new reals, then CH would fail in the generic extension M[G].
Example 2. Let P = f : f is a nite function from into {0, 1} , where M, and
f 6 g gdom(f )=f .
If G is generic over M, dene g : {0, 1} by g (n) =
is a sequence of new reals.
G(, n). Then hg : < M i
However, this would not be quite enough to blow up the continuum to size in M[G],
because maybe M could be small in M[G]. We would need to ensure that cardinals do
not collapse.
Example 3. P3 = {
a = hAi : i < i : < , Ai i < }, with a
6 a
a
is an initial
segment of a
.
S
A 3sequence in M[G] will be A = {
a:a
G}, where G is a generic lter over M.
Example 4. P = {f : f is a function from a countable (in M) ordinal into P()}, with f 6
g f g.
A generic lter G gives rise to a map g =
G : 1M P()M .
M[G]
To make sure that M[G] is a model of CH, we must be sure that 1M = 1
P()M[G] .
and P()M =
M[G]
To check that M[G] is an extension of M and that, e.g. in the examples above, 1M = 1
and P()M = P()M[G] , we need to show that for transitive models, the simplest settheoretic
concepts are invariant under extension. The concept of invariance of a property or a term is
important in its own right. We study it briey in a more general setting.
Absoluteness
Denition. Suppose (x1 , . . ., xn ) is a formula in the vocabulary of ZFC (or some expansion),
and A B are classes. We say:
(1) (x1 , . . ., xn ) is absolute between A and B if
x1 , . . ., xn A (x1 , . . ., xn )A (x1 , . . ., xn )B
(2) A term t is absolute between A and B if the formula x = t is absolute between A
and B.
(3) If or t is absolute between A and V , then or t is absolute for A.
(4) If or t is absolute for any transitive class A, then or t is absolute.
Why is absoluteness useful? What concepts and properties are absolute?
Example. Suppose A is a class, x A, and V = (y)(y x). A priori, y need not belong to A.
A
However, if A is transitive, then y A. So for transitive A, y(y x) y(y x) .
This example suggests that atomic formulas and formulas all of whose quantiers are restricted
A
will have the property (y x) (y x) .
35
I.e., properties expressed by formulas with restricted quantiers are absolute, leading to the
following denition of 0 formulas.
Denition. We dene that class of 0 = 0 = 0 formulas as follows:
(1) If is atomic, then is 0 .
(2) If 1 , 2 , 3 are 0 , then 1 2 and 3 are 0 . (And 1 2 , 1 2 , 1 2
are also 0 by conventional abbreviations.)
(3) If is 0 , then x(x y ) is 0 . (We abbreviate x(x y ) as x y .)
Also, x(x y ) is 0 (abbreviated as (x y)).
(See handout for a list of 0 formulas.)
Denition. Suppose T is a theory. We say that is T0 (T provably 0 ) if for some 0 formula
, we have T .
If T is ZF or stronger, we omit T .
Lemma. Suppose (x1 , . . ., xn ) is 0 (or provably equivalent to a 0 formula), and A is a
transitive class.
Then x1 , . . ., xn A (x1 , . . ., xn ) (x1 , . . ., xn )A .
I.e., (x1 , . . ., xn ) is absolute.
Proof. Straightforward induction on the complexity of (x1 , . . ., xn ).
With these matters claried, we turn to the proof of the Forcing Theorem.
Lemma 4. Let G be a generic lter in P over M. Then
(1) For all a M, we have a[G]
= a, and [G] = G.
So M M[G], and G M[G].
(2) M[G] is transitive.
(3) If is a Pname in M, then rkr ( ) 6 rkn ( ), and rank( [G]) 6 rkn ( ).
So Ord M = Ord M[G].
(4) M[G] is minimal: if M+ is a transitive model of ZFC with M M+ and G M+ ,
then M[G] M+ .
Lecture 21
First, a remark on absoluteness of transnite recursion.
(1) If s(y, z) is absolute and ZFC t() = s(t , ), then y = t() is absolute.
(2) If s0 and s1 (y) are absolute terms and ZFC t(0) = s0 t(n + 1) = s(t(n)), then y = t(n)
is absolute.
Proof of Lemma 4.
(1) Induction on Pname rank 6 .
36
a.
x a[G]
x = b[G]
for some b a with (p, b)
By induction, this x = b for some b a. So x a and a[G]
= a.
Then [G] = {(p, p)
: p P }[G] = {p[G]
: p G} = {p : p G} = G.
(2) x [G] M[G] x = i [G] for some Pname i , etc, which x M[G].
(3) The claims about rank are proved by induction (exercise).
We show that Ord M = Ord M[G]. Clearly, Ord M Ord M[G] by (1).
If Ord M[G], say = [G] for some Pname MP .
e
Then = rank() = rank( [G]) 6 rkn ( ) Ord M, since ranks are dened by
e
transnite recursion, so are absolute.
So 6 Ord M, so M, since M is transitive and Ord.
Thus Ord M = Ord M[G].
(4) For MP , we see that M+ , G M+ .
+
So [G] = [G]M M+ , so M[G] M+ .
We start proving M[G] satises the axioms of ZFC. We get some very easily.
Proposition 5. With the usual assumptions on M, P, G, we have that M[G] satises the following.
(0) Set existence
(1) Extensionality
(2) Null set
(3) Foundation
(4) Pair set
(5) Union
(6) Innity
Proof.
(0) 6= M M[G].
(1) M[G] is transitive
(2), (3), (6) absoluteness of terms involved in the axioms.
(4) Let x = [G], y = [G] M[G], with , MP .
Dene a Pname upair(, ) MP , by upair(, ) = {h0P , i, h0P , i}.
Its easy now to check upair(, )[G] = {[G], [G]} = {x, y}.
(5) Suppose that x = [G] M[G]. Two ways to check union.
S
(i) manufacture a Pname u MP such that u[G] = x
S
(ii) on credit. Find a Pname MP such that x [G] and then appeal to
Separation. (But this assumes that we have checked that Separation holds in
M[G].)
37
S
We indicate (1). We want ( x)[G] = z[G] : u[G] x[G], z[G] u[G] .
S
Reecting suggests the following Pname as a candidate to give x in M[G].
u = (r, ) : p, q P, MP , hp, i hq, i p 6 r q 6 r
S
S
This is a Pname in M, and it is easy enough to check that u[G] = x = ( [G]).
S
If one does (2), take = range( ). We have range( ) = { : p P, hp, i }.
S
is a Pname in M, and x [G]. Now apply Separation.
2
It remains to verify Power set, Replacement, Separation and Choice in M[G]. These axioms will
require us to produce more complicated Pnames in M. Hence we must treat the denability of
in M.
Lecture 22
To verify these remaining axioms, we shall assume proven the following two clauses of the Forcing
Theorem.
(3) M[G] = 1 [G], . . ., n [G] i (p G) p
(1 , . . ., n ) .
(4) The relation
P is denable in M; more precisely, given (x1 , . . ., xn ), there is a formula
(x1 , . . ., xn , x, y) which is absolute in M such that for all a1 , . . ., an M,
p
(a 1 , . . ., a n ) i M = (a1 , . . ., an , p, P)
Notes.
(3) is sometimes called the Truth Lemma.
(4) is sometimes called the Denability Lemma.
We shall prove (3) and (4) in an appendix (they are proved by induction).
Proposition 5 12 The axioms of Separation, Power set, Replacement and Choice are all satised
in M[G].
Proof.
(1) Separation.
Well write G , G , . . . instead of [G], [G], . . ., when more convenient.
Suppose that , 1 , . . ., n MP and (x, y, x1 , . . ., xn ) is a formula. We wish to show
that
y = a G : M[G] = [a, G , 1G , . . ., nG ]
belongs to M[G].
We dene an appropriate name MP .
n
o
= hp, i : hp, i P range(), p
(, , 1 , . . ., n )
We have V P . In fact, MP by Denability Lemma (4).
We just check that [G] = y.
First, [G] y. Suppose a [G]. Then a = [G], where hp, i for some p G.
By denition of ,
p
(, , 1 , . . ., n )
38
Since p G,
M[G] = G G [G , G , 1G , . . ., nG ]
(by denition of
).
So a = [G] y.
Second, y [G]. Suppose a y. Then a [G] and (a, G , 1G , . . ., nG )M[G] .
So a = [G] for some range(). So M[G] = G G [G , G , 1G , . . ., nG ].
Thus (p G)p
(, , 1 , . . ., n ) (by Truth Lemma (3)).
So hp, i and G G . So y G .
(2) Power set.
Suppose x M[G], x = [G] for some MP .
Let Z = hq, pi : (p 6 q)(hp, i ) .
Let Z = {u : u Z }M . Let = P Z .
MP . It is an exercise to check that (P x)M[G] = G .
(3) Replacement.
Suppose [G] M[G] and we wish to nd a set [G] M[G] such that (in M[G])
z z [G] (y) y [G] (y, z)
i.e.
o
n
[G] = z : (y [G]) (y, z)M[G]
We seek a suitable name.
Let Z = hq, yi : (r P ) hr, yi r 6 q
Then Z MP .
For each hq, yi Z , consider all Pnames z such that q
(y,
z).
There are too many Pnames z for this to be a set in M. We cut the collection down
to a set in M by using the rank function in M.
Let (q, y) be the least rank of a set z M for which q
(y,
z).
(If no such z exists,
then (q, y) = 0.)
Now, let
n
oM
= hw, zi : (hq, yi Z ) q
(y,
z)
rank(z) 6 (q, y)
Check that G is as required. Note that MP because rank( ) is bounded in M,
and
is denable in M.
(4) Choice.
Suppose x = G M[G]. We show that there is a function f M[G] mapping an
ordinal onto a set containing x as a subset.
Let h : < i be an enumeration of range( ).
39
Since M = AC, we may assume
belongs to M. Let opair(u, v)
that this enumeration
be the ordered pair hu, vi = {u, u}, {u, v} .
)i : p P, < MP
Let f = hp, opair(,
Then f [G] = h, [G]i : < M[G], and x range(f ), dom(f ) = .
So x is wellordered in M[G].
Corollary 6. M[G] = ZFC.
Lecture 23
Proposition 7. Let P M be a forcing. If p
(x)(x (x, 1 , . . ., n ) then there exist
q > p and range() such that q
(, 1 , . . ., n ).
Proof. Suppose p G for some generic lter G.
M[G]
(x) x G (x, 1G , . . ., nG ) , so for some
MP and q G, we have
(q , ) and M[G] = G G (G , 1G , . . ., nG ) ,
By the Truth Lemma, there is q G such that q
(, 1 , . . ., G ).
Wlog, q > p, q .
We turn to the proof of the relative consistency of CH.
Denition. Let P M. We say
(1) P preserves cardinals if for every generic lter G in P over M, for all Ord M,
( is a cardinal)M ( is a cardinal)M[G]
()
If a cardinal in M ceases to be a cardinal in M[G], we say that P collapses (or
that is collapsed by P).
(2) Analogously, we say that P preserves cardinals 6 M if () holds for all 6 ,
Ord M.
Note. The nite ordinals and are absolute, so cardinal preservation is an issue only for > 0 .
Note also that if is a cardinal in M[G], then is a cardinal in M since M M[G].
Denition. Let P M. We say
(1) P preserves conalities if for every limit ordinal M,
cf()M = cf()M[G] ()
(2) Analogously, P preserves conalities 6 M if () holds for 6 .
Lemma 8. If P preserves conalities, then P preserves cardinals.
Proof. Every innite cardinal is regular or a limit.
Case 1. is regular in M. Then cf()M[G] = cf()M = , so is a regular cardinal in
M[G].
Case 2. is a limit cardinal in M. Then the regular (even the successor) cardinals M,
< , are unbounded in . By Case 1, these remain regular cardinals in M[G] and
are still unbounded in , and so is a limit cardinal in M[G].
2
40
Denition.
1. Suppose > 0 . A forcing P is closed (or complete) if for any V
< , every
6P increasing sequence hpi : i < i P has an upper bound p P . I.e.,
pi 6P p.
i<
2. P is countably complete if P is 0 complete.
Lemma 9. Suppose P M, M, and (P is closed)M . Let < , B M, and suppose G is
generic in P over M.
Then (B)M = (B)M[G] .
In M[G], there are no new sequences, i.e. if f : B, f M[G], then f M.
Proof. M[G] = f is a function from into B. Say f = G for some MP .
Suppose f
/ (B)M = B M (absolute) = K M.
By the Truth Lemma, there exists p G such that p
( is a function from into B and
/ K).
Now we work entirely in M and dene by transnite recursion a 6P increasing sequence
hp : 6 i and hb B : < i as follows:
(1) p0 = p
(2) 6 p 6P p
= b for some b B.
(3) p+1
()
Why is this possible to do?
If 6 is a limit ordinal, since P is closed and < , there is p P with
p 6P p.
<
So (2) is satised.
(p = p0 6 p )
If p has been dened, p
is a function from into B.
= x).
So p
(x)(x B (B)
= b .
By Proposition 7, there exists b B and p+1 > p such that p+1
()
This completes the denition.
In M, consider the function g : B given by g() = b . Then g K. In particular,
g M.
Let H be a generic lter in P over M such that p H. Then for all < , p H.
V
M[H]
H () = g(). So in M[H], H = g K
().
<
so M[H] = H
But, p = p0 6 p H, so p H. And p
/ K,
/ K.
So H
/ K, which contradicts (). Thus f M.
41
Corollary 10. Suppose ( is a cardinal and P is complete)M . Then P preserves conalities
6 and cardinals 6 .
Proof. If for some cardinal 6 we have cf()M[G] < cf()M , then in M[G], there exist <
and f M[G] such that f : is strictly increasing and conal in .
By Lemma 9, f M, so cf()M 6 < .
Theorem 11. Con(ZFC) Con(ZFC+(20 = 1 )).
I.e., there is a model of ZFC+(20 = 1 ).
Proof. Let M be a CTM. We construct M[G] such that M[G] = (20 = 1 ).
We wish M[G] to possess a surjection from 1 onto P() in M[G].
In M, let P = f : dom(f ) = < 1 , range(f ) P() , with f 6P g if gdom(f ) = f .
Note:
M[G]
(1) P is 1 complete. So 1M = 1
M
M[G]
(2) (P(w)) = (P())
Let g =
, by Corollary 10.
, since ( 2)M = (2)M[G] , by Lemma 9.
G, where G is a generic lter in P over M.
Then Ea = f : a range(f ) , for (in M), is a dense open set in M.
So G Ea 6= .
M[G]
Hence in M[G], g is a surjection from 1
Lecture 24
onto (P())M[G] .
Lemma 12. Suppose P M and (P is a CCC)M . Suppose f M[G] is a function from to ,
with , Ord and > .
Then there exists y M with range(f ) y and (y 6 )M .
In other words, f M[G] can be approximated by a set in M.
Proof. f = G for some MP . M[G] = (G : is a function).
By the Truth Lemma, there is p G with p
( : is a function).
And p G, so M[G] = ( < )(f () < ).
So p
( < )(
() < ).
= .
Say = f (). Thus there is q G, p 6 q, with q
()
= )
.
For each < , dene y = < (r)(p 6 r)(r
()
Note: y M by the Denability Lemma, and y where = f ().
In M, y is countable. Why? Choose for each y a condition q such that p 6 q and
= .
q
()
Then {q : y } is an antichain in P. This is easy to check.
42
So {q : y } is countable since P has CCC in M, and hence (y is countable)M .
Let y =
y . Then y M since , y M:
<
y =
<a
M
y  = 0  =  .
And range(f ) y since < , we have f () y .
Lemma 13. Suppose (P is CCC)M . Then P preserves conalities and cardinals.
M[G]
Proof. If not, then since M
0 = 0
= cf()M[G] < in M[G].
, the witness to failure = cf() > 0 in M has conality
So in M[G], there is f : , strictly increasing and conal in . Now, by Lemma 12,
there is y M such that range(f ) y and (y = a)M .
So y is conal in in M. Thus cf() 6 y =  <
M
, so cf()M < in M, contradiction.
P must therefore preserve conalities and hence also cardinals.
We shall use a useful result to check CCC.
Lemma 14 (the system Lemma). Suppose A is an uncountable family of nite subsets
of a set X.
Then there exist an uncountable subfamily B A and a set r such that for all a 6= b B,
we have a b = r.
Proof. Wlog, for all a A, a = n for a xed n < . We prove by induction on n.
If n = 1, this is the Pigeonhole Principle. So suppose n > 1.
Case 1. If there
exists x such that x a for uncountably many a A, then evict x and
consider a \ {x} : a as above}. Apply the induction hypothesis to get B and add x
back to each b B.
Case 2. Not case 1. Then every x leaves the as in A after at most countably many as.
So one can easily dene a sequence ha : < 1 i with a A with a a = for
6= .
2
Denition. Fn(A, B, ) = f : f is a function from A to B and f  < .
Corollary 15. Fn( , 2, 0 ) has the CCC.
Proof. If A Fn( , 2, 0) is uncountable, apply Lemma 14 to nd B A with B = 1 ,
and f g = r for all f 6= g B.
So h = f g Fn( , 2, 0) and f 6 h g 6 h.
So A is not an antichain.
Theorem 16. Con(ZFC) Con(ZFC+(20 > 1 )).
I.e., there is a model of ZFC+(20 > 1 ).
43
Proof. Let M be a CTM. In M, let P = Fn( , 2, 0 ) where > 2 in M.
By Corollary 15, P has the CCC, so P preserves conalities and cardinals, and hence
M[G]
= M
1
M[G]
= M
2 6
The following sets are dense open and belong to M:
D,n = {p P : (, n) dom(p)}
So G D,n 6= for all < , n .
In M[G],
G : 2. Let g (n) =
G(, n). Then g : 2, g M[G].
Claim. In M[G], < g 6= g .
Why? Otherwise M[G] = g = g , so there exist , MP , and p G such that
p
= .
Pick n < such that (, n), (, n)
/ dom(p).
Extend p to q, so qdom(p) = p, with q(, n) = 0 and q(, n) = 1.
Let H be generic in P over M, q H.
Then M[H] = H (n) = 0, and M[H] = H = 1, but M[H] = H (n) = H , because
p H (as p 6 q).
Contradiction. Thus in M[G], g 6= g , and so M[G] = (20 > > 2 ).
Corollary 17. The Continuum Hypothesis is independent of ZFC (if ZFC is consistent).
Proof. Theorems 11 and 16.
44
TOPICS IN SET THEORY: Exercise Sheet 1
Department of Pure Mathematics and Mathematical Sciences,
University of Cambridge
Michaelmas 20122013
Dr Oren Kolman
(i) Assume the axiom of choice.
Suppose
P I is a nonempty set and for each i I , i is an infinite cardinal.
Show
iI i  I  supiI i , where  I  is the cardinalitySof I . [Hint:
enumerate i and think up a surjection from  I  onto iI i where
= supiI i .]
(ii) Suppose is an infinite cardinal. Prove that the (cardinal) successor + of
is a regular cardinal. [Hint: part (i).]
(i) Suppose f H(C) , i.e. f is an entire holomorphic function of the single
complex variable z . Let Z(f ) = {z C : f (z) = 0} be the zero set of f .
If f 6= 0 , what is the cardinality of Z(f ) ? [Hint: the zeros of a holomorphic
function have a noteworthy topological property.]
(ii) S
Suppose F H(C) has cardinality 1 . What is the cardinality of the set
f 6=gF Z(f g) ? [Hint: Q1(i).]
(iii) Suppose 20 > 1 and F H(C) has cardinality 1 . By judicious selection
of a w0 C , show that F is not orbit countable.
(iv) Deduce the observation of Erdos that if 20 > 1 , then orbit countability is
equivalent to countability.
(i) Let ZF C be the firstorder theory whose axioms are obtained from ZF C by
omitting the axiom of infinity. Show that (*) (V , (V V )) is a model
of ZF C .
(ii) Show that every element of V is finite. Deduce that the axiom of infinity
cannot be proved from the other axioms of ZF C .
(iii) Can the assertion (*) be proved from ZF C ? Explain.
(i) Consider the assertion
DODGY: there exists a family {A,n : < 1 , n < }
S
such that (i) 1 \ n< A,n is finite for every < 1 , and (ii) if 6= , then
A,n A,n = for all n < . Decide whether DODGY is provable or not
from ZF C . [Hint: it may be easier to try part (ii) first.]
1
S
(ii) Prove there exists a family {A,n : < 1 , n < } such that (i) 1 \ n< A,n
is countable for every < 1 , and (ii) if 6= , then A,n A,n = for all
n < . [Hint: For each ordinal < 1 , choose a surjection f from onto
(why is this possible?), and consider the set A,n = { : f (n) = } .]
(iii) Can you generalize the result of part (ii) to cardinals greater than 1 ? How
about ?
5 Assume that 1 0 = 1 . Prove n 0 = n for all n < . [Hint: induction.]
TOPICS IN SET THEORY: Exercise Sheet 2
Department of Pure Mathematics and Mathematical Statistics,
University of Cambridge
Michaelmas 20122013
Dr Oren Kolman
(i) Suppose that x and y belong to the class W F of wellfounded sets. Find
bounds for the ranks of the following sets in terms of the ranks of x and y :
x , P (x) , {x} , x y , x y , x y , {x, y} , hx, yi , and y x .
(ii) Calculate the ranks of the sets N , Z , Q , R , and C .
(i) Suppose that M is a transitive model of ZF (or a large enough finite fragment
of ZF including the Power Set Axiom) and let x M . Prove that P (x)M =
P (x) M . Deduce that the Power Set Axiom holds in M if and only if
x My M(P (x) M y) .
(ii) Suppose that V reflects (a large enough finite fragment T of) ZF C , let
< . Prove that VV = V . Hence complete the second proof that neither
ZF nor ZF C is finitely axiomatizable.
3 Prove the basic properties of the hierarchy {V : Ord} .
(i) V V .
(ii) < V V .
(iii) V is a transitive set: x V x V .
(iv) V = {x W F : rank(x) < } , where rank(x) = min{ : x V + 1} .
(v) y x rank(y) < rank(x) .
(vi) rank() = .
(vii) Ord V = .
(viii) For every n ,  Vn < 0 ;  V = 0 ;
(AC) for all Ord,  V+ = i .
4 Prove the basic properties of the constructible hierarchy {L : Ord} .
(i) L is a set and L V .
(ii) L is transitive.
1
(iii) For every n ,  Ln < 0 , and Ln = Vn ; L = V .
(iv) For every ,  L =  .
(v) L L .
(vi) < L L and L
L .
(vii) L+1 and
/ L .
5
(i) Suppose that M is a transitive class. Show that ZF C ` Extensionality M
and ZF C ` Regularity M .
(ii) For a L() formula (x, y) with free variables x, y, y1 , . . . , yn , and a variable
z not free in (x, y) , let ()(x,y) be the assertion (x a)(!y)(x, y)
zy(y z x(x a (x, y))) . The notation !w abbreviates the
formula zw( w = z) , where z is the first variable different from w and
not free in . Show that the schema ()(x,y) is an equivalent form of the
schema of the Axiom of Replacement.
(iii) Prove that the Reflection Principle for {V : Ord} implies the Axioms of
Replacement. [Hint: Suppose a V ; referring to part (i), reflect to a V ,
and use the Axioms of Separation to find the right candidate for the image of
a .] Comment: sometimes it may prove less onerous to check the Reflection
Principle instead of Replacement, e.g., when showing that the class L is an
inner model of ZF C , it is equivalent to check that L is an inner model of
LevyMontague set theory LM .
6 **
(i) Consider the assertion that for every formula (x1 , . . . , xn ) in the language of
set theory, for all ordinals < < , for all a1 , . . . , an V ,
(a1 , . . . , an )V (a1 , . . . , an )V .
What is your opinion? Can the assertion be proved in ZF C ? Would your
view change if V and V were replaced by L and L ? Would your views
alter if the assertion were modified to the sharpened form: for all uncountable
cardinals < , for every formula (x1 , . . . , xn ) in the language of set theory,
for all a1 , . . . , an V ,
(a1 , . . . , an )L (a1 , . . . , an )L .
(ii) Suppose there exists an ordinal such that V is a model of ZF C . Show
that the least such ordinal has cofinality .
(iii) Suppose is a strongly inaccessible cardinal. Prove that V is a model of
ZF C . 1
1
Unrelated general gossip: apparently, if the existence of inaccessible cardinals were inconsistent with
ZF C , marvellous phenomena would appear  one could prove that there are no uncountable Grothendieck
universes and the axiom of universes in category theory is false. (An uncountable Grothendieck universe
is exactly H for an inaccessible cardinal , and the axiom of universes asserts that every set is in such
a universe.)
(iv) Deduce that the converse of part (iii) is false.
7
(i) Prove that for every infinite cardinal , 2 = .
(ii) Let , i (i I) be infinite cardinals. Prove:
P
Q
(a) iI i = iI i ;
Q
Q
(b) ( iI i ) = iI i .
TOPICS IN SET THEORY: Exercise Sheet 3
Department of Pure Mathematics and Mathematical Statistics,
University of Cambridge
Michaelmas 20122013
Dr Oren Kolman
1 Suppose and are ordinals. Prove:
(i) +1 < 22 ; implies = 2 .
Q
Q
0 Q
0
; <+ = + 0 .
(ii)
0<n< n = 2 ;
n< n =
(iii) < implies +1 = +1 .
< cf ( );
(iv) GCH implies = +1 cf ( ) ;
+1
.
(v) 1 = 0 21 .
(vi) < 2 implies 2 = 1 22 .
2
(i) Prove that an infinite cardinal is a strong limit cardinal if and only if = i
for some limit ordinal .
(ii) (Tarskis recursion formula) Let be a limit cardinal and > 0 . Let be
a limit ordinal such that < cf () . Suppose that { P
< : < } is a
strictly increasing sequence of cardinals such that = < . Show that
P
= < .
(iii) Prove that if > 0 is a limit ordinal, then cf (i ) = cf () .
P
3 (i) Suppose is a limit cardinal and < cf () . Prove = <   .
(ii) Suppose is a limit cardinal and cf () . Prove = (sup<   )cf () .
(iii) Suppose > cf () is not a strong limit cardinal. Prove < = 2< > .
(iv) Suppose > cf () is a strong limit cardinal.
Prove 2< = and < = cf () .
(v) 20 > implies 0 = 20 .
(vi) 21 = 2 and 0 > 1 implies 1 1 = 0 .
(vii) 20 1 implies ( ) = 20 and (1 ) = 21 .
(viii) 21 = 2 implies 0 6= 1 . [HINT: Cofinalities.]
4 Prove there exists a rigid dense subset D of the linear order (R, ) , i.e. D has no
nontrivial orderautomorphisms. [HINT: Predictively list the orderautomorphisms
from R into R (using the observation that these are uniquely determined by their
restrictions to Q ) and construct two disjoint sets D Q and B by transfinite
recursion so that every potential orderautomorphism of D maps some element of
D into B .]
1
(i) Let = cf () > 0 .
Show U = { < : = } and V = { < : = } are club in .
(ii) Suppose B = cf () > 0 . Show that B is a club of if and only if B
is the range of a continuous strictly increasing function f : .
(iii) Prove that if > cf () 0 , then there is a club C of such that no
member of C is a regular cardinal. [HINT: Try the range of a continuous
function f : cf () { < : > cf ()} .]
(iv) Suppose = cf () > 0 and f : . Show that the set D = { < : is
closed under f } is club in . Deduce that if F is a family of functions from
to and  F  < , then the set E = { < : is closed under F } is club
in . Hence or otherwise, show that for a structure M with universe ,
where   < , the set { < : M  is an elementary submodel of M } is
a club in . [HINT: Skolem functions and the TarskiVaught Criterion.]
6 Let 1 have the order topology and suppose that f : 1 R is a (topologically)
continuous function. Show there exists < 1 such that > , f () = f () .
[HINT: for each real > 0 and limit ordinal > 0 use the inverse image of
(f () , f () + ) to define a regressive function g () , apply Fodors Theorem
and first countability of R .]
7 Let A be a set of cardinality = cf () > 0 . A filtration of A is an indexed
sequence {A : < } such that for all , <
(i)  A  < ;
(ii) < implies A A ;
(iii) lim() implies A = {A : < } ;
(iv) A = {A : < } .
(i) Suppose {A : < } and {B : < } are filtrations of A . Show the
set { : A = B } is a club of .
(ii) Let {A : < = cf ()} be a filtration of A . Prove there exists a club
C of such that for all C ,  A+ \ A = + \  where + is the
successor of in C , i.e. + = inf { C : > } .
(iii) Suppose = cf () > 0 and {A : < } is a filtration of a set A of
cardinality . Prove Fodors Lemma: if S is a stationary subset of and
f : S A is a function such that for all S , f () A , then there exists
a stationary S 0 S such that f  S 0 is constant
8 Prove that implies there exists {Z : < 21 } such that
(i) < 21 Z is a stationary subset of 1 ;
(ii) < < 21 Z Z is countable.
9 Suppose that S is a stationary subset of = cf () > 0 . Prove that S is
equivalent to the statement: there exists {f : S} such that
2
(i) S, f : ;
(ii) f : , { S : f  = f } is a stationary subset of .
10
(i) Show that implies .
(ii) * Prove Devlins Theorem: + CH implies . Deduce that is equivalent
to + CH .
0
11 Let S denote the following statement: there exists {E : S} such that
(i) S, E is a countable set of subsets of ;
(ii) X , { S : X E } is a stationary subset of .
0
Prove that S and S are equivalent in ZF C .
12
(i) Suppose is a bijection from + onto + . Show there exists a club C
of + such that for all C , the restriction map  is a bijection from
onto .
(ii) For a cardinal and a set W , let [W ] = {Y W :  Y  } . If 2 = + ,
let {X : < + } be an enumeration of [+ ] and suppose Z + . Show
that for some club C of + , for all C there are arbitrarily large <
such that for some < , Z = X .
(iii) Suppose cf () = > 0 and h is a function from dom(h) into . Prove
that the following are equivalent:
(a) h is onetoone on some club C of ;
(b) h is strictly increasing on some club D of ;
(c) range(h  S) is unbounded in for every stationary subset S .
13 Open Research Problems.
(i) Juhaszs Problem. Does imply SH , i.e. does imply there exists a
Suslin tree?
(ii) Assume that cf () < and = + = 2 . Determine whether Scf
is a
()
theorem of ZF C (or even ZF C + GCH ).
(iii) Assume that = < = 2 is a regular limit cardinal. Determine whether
is a theorem of ZF C .
TOPICS IN SET THEORY: Exercise Sheet 4
Department of Pure Mathematics and Mathematical Statistics,
University of Cambridge
Michaelmas 20122013
Dr Oren Kolman
This final set of exercises involves, alongside basic drill, some questions that require significant extensions of the material covered in lectures. Attempt a representative selection.
Several questions are optional (they presuppose some elementary nonsettheoretic information from algebra, complex analysis, model theory, or topology); some problems
are difficult (marked ) or open, and they have been included, along with guidance, to
illustrate the richness and flexibility of forcing.
1 Absoluteness results
All formulas and terms are in the vocabulary of ZF C unless otherwise indicated.
Suppose that , 1 and 2 are absolute.
(i) Suppose that is a formula with the same free variables as such that
ZF ` . Show that is absolute.
(ii) Suppose that y1 , z2 and have the same free variables; suppose
ZF ` y1 and ZF ` z2 . Show that is absolute.
(iii) Suppose that (y) and the term t are absolute. Show that (t) is absolute.
(iv) Suppose that t is absolute. Show that x t and t x are absolute.
(v) Suppose that (y) and t are absolute. Show that {y t : (y)} is absolute.
(vi) Prove that the following terms and predicates are absolute (in each instance,
it suffices to produce a ZF provably equivalent absolute or 0 formula):
S
y x ; z = {x, y} ; z = {x} ; z = hx, yi ; z = x ; z = x y ; z = x y ;
z = x \ y; z = x y;
f is a function; y = dom(f ) ; y = range(f ) ;
y = ; y = s(x) ; y = 1 ; y = 2 ;
y = f 00 x (where f 00 x means the application of f to x ); y = f  x ;
x is transitive; x Ord (remember that Foundation is an axiom of ZFC);
x is a limit ordinal; x = .
(vii) Suppose that the term s(y, z) is absolute, and ZF C ` ()t() = s(t  , ) .
Show that the formula y = t() is absolute. [For a formula (x) , ()
abbreviates ( Ord ()) .]
(viii) Prove that the following are absolute:
+ ; ; ; rank(x) ; y is the transitive closure of x .
1
(ix) Determine which of the following are (a) absolute, (b) absolute for V , when
is a strongly inaccessible cardinal, (c) absolute between the ground model
and its generic extensions (brief explanations suffice):
Z , (Q, ) , (R, ) , x is countable, y = P (z) , is a cardinal, 1 , S is
a stationary set, C is a club, P is a forcing, the partial order P is a c.c.c.
forcing, F n(A, B, 0 ) , F n(A, B, 1 ) , T is a Suslin tree ( ), y = ,
= , z = i , x L , ()(x L ) , ZF C ` ;
Optional
() ( ), M is an R module over the commutative ring R , P = N P , the
real Hilbert space `2 , X is a complex Banach space, Y is an inseparable
topological space, the Singular Cardinals Hypothesis ( ), the Riemann
Hypothesis ( ).
2 Generic filters and classical theorems
(i) Prove Cantors theorem on the 0 categoricity of unbounded dense linear orders: if A and B are countable unbounded dense linear orders, then A and
B are isomorphic. [HINT. Let P be the partial order of finite partial isomorphisms between A and B under extension. Show that for all a A and
b B , the sets Da = {p : a dom(p)} and Rb = {q : b range(q)} are
dense open in P . Consider what properties a filter G generic relative to the
family {Da , Rb : a A, b B} might possess.]
(ii) For a set X P
and a cardinal , let [X]< = {Y X :  Y P
< } . Let
1
I = {I : nI n < } . Let Q be the partial order {I : nI n1 < 1}
where I Q J iff I J . For a cardinal , let HS< (I) abbreviate the
statement:
(H [I]< )(I I)(I H)(I I )
where for x, y P (), x y ( x\y < ) . Prove the assertion HS<1 (I) .
What can be said about HS<20 (I) ? [HINT. Show DJ = {q Q : J q} is
dense in I for J I .]
(iii) Consider the following (rash) statement: if the forcing P has the 2 chain
condition and D is a family of less than 21 dense open sets in P , then there
exists a D generic filter G in P . Is this statement provable, refutable, or
independent of ZF C ? (Generalization of Martins Axiom to higher cardinals and discovery of the strongest or optimal form of the generic principle it
expresses are elusive; Martins Maximum ( M M ; see below) and the Proper
Forcing Axiom ( P F A ) are the bestknown candidates.)
3 Concerning forcings, antichains and generic sets
(i) Prove that a filter G is generic in P over M if and only if for every maximal
antichain A M of P  G A  = 1 . [HINT. One direction might use AC.]
(ii) A subset D P is:
(1) predense above p P if (q P)(q p (d D)(d and q are compatible)); D is predense if D is predense above 0P .
2
(2) dense above p P if (q P)(q p (d D)(d q)) . (So D is dense
in P if D is dense above 0P .)
Suppose that E is predense in P and G is generic in P over M . Show that
G E 6= .
Suppose that E is predense above q P and G is generic in P over M .
Show that if q G , then G E 6= .
(iii) Deduce that the following are equivalent for a filter G P M where M is
a transitive model of ZF C .
G is generic in P over M ;
G D 6= for every dense open subset D of P ;
G C 6= for every dense subset C of P ;
G B 6= for every predense subset B of P ;
G A 6= for every maximal antichain A of P .
(iv) Suppose M is a CTM, P M, E P, E M , and G is generic in P over M .
Prove that either G E 6= or (q G)(r E)(r and q are incompatible).
[HINT. Consider {p P : (r E)(r p)} {q P : (r E)(r and q are
incompatible) } M .]
(v) Suppose M is a CTM and P M is a separative forcing. Prove that there
are 20 generic filters in P over M .
4 The forcing relation
P
Suppose that P is a nontrivial forcing, p, q P , and is a formula in the
vocabulary of ZF C which may contain P names. Show:
(i) if p
P and p P q , then q
P ;
(ii) if q
P and p P q p 6= q , then p
P ;
(iii) if (@r)(p P r r
P ) , then p
P ;
(iv) (r)(p P r)(r decides ) , i.e. either r
P or r
P ;
V
(v) if p does not decide , then i=1,2 (ri )(p P ri )(r1
P ) (r2
P ) .
5 Names
Suppose G is generic in P over M .
(i) Suppose , MP . Show G G = ( )G .
(ii) Suppose MP and range( ) {n : n } .
Let = {hp, ni
: (q P)(hq, ni
p and q are incompatible) } . Show
G = \ G . [HINT. Show the set {r P : (q r)(hq, ni
hq, ni
}
is dense.]
(iii) Suppose , MP . Show G G = ( )G .
3
(iv) Suppose A is an antichain in P and for each a A, a is a P name. Show
there exists a P name such that for every a A , if a G , then [G] =
a [G] , and [G] = if G A = . [HINT. Suppose a = {(qa,j , a,j ) : j < ia } .
Consider the P name = {(r, a,j ) : a A, j < ia , r qa,j , and r a} and
refer to Question 3 (i).]
6 Nice Names and Bounds for the Continuum
Suppose G is generic
in P over M . A name MP is a nice name for a subset
S
x of MP if = {A {} : range()} , where each A is an antichain
in P .
(i) Prove that if P M , then for all , MP there exists a nice name such
that
P ( = ) . [HINT. For range() , let A be maximal
relative to the properties (1) (p A )(p
) and (2) A is an antichain
in P ; now use Question 3 (iv) to check as defined above works.]
(ii) Suppose (P is a c.c.c. forcing and is a cardinal) M . Let = ( P  )M . Then
(2 )M[G] . [HINT. For P ()M[G] , count the number of possible nice names
for its members, remembering that P has the countable chain condition.]
(iii) Deduce that if ( is a cardinal and 0 = )M , then there is a generic extension
M[H] such that (20 = )M[H] .
7 Adding Cohen reals and Suslin trees
(i) A tree T is everbranching if for every s T , the set {t T : s T t} is not
linear ordered. Let M be a CTM such that (T is an everbranching Suslin
tree) M . Suppose (P = F n( , 2, 0 ) 0 )M . Prove that for any filter
G generic in P over M , M[G] = (T is a Suslin tree).
(ii) Deduce that there is a model of ZF C in which there is a Suslin tree but CH
fails. Remark: So the existence of Suslin trees does not imply CH (nor a
fortiori ). It is a theorem of Shelah that adding a Cohen real adds a Suslin
tree.
8 Diamonds and Clubs
(i) Prove that the theory ZF C + is relatively consistent. [HINT. It may be
slightly easier to verify in its function form (see Exercise Sheet 3, Question
9): let I = {h, i : < < 1 M } and consider the forcing Q = F n(I, 2, 1 ) .
Show that Q is countably complete, and that
S if G is generic in Q over M ,
then in M[G] , a sequence is given by h( G) : < 1 i . For this, noticing
Q adds no new sequences and 1 M = 1 M[G] , define a sequence of ordinals
and conditions forcing an arbitrary given club to intersect the family of guesses
for a function f : 1 1 . (Refer to K. Kunen, Set Theory, chapter VII, or
S. Shelah, Proper and Improper Forcing, chapter 1, if the abyss looms.)]
(ii) Deduce that is independent of ZF C .
4
(iii) Show if ( is a cardinal and 0 = and )M , then there is a generic
extension M[H] such that (20 = and there is a Suslin tree) M[H] .
(iv) Suppose that (P is a c.c.c. forcing and  P  1 and )M . Show for every G
generic in P over M , M[G] = . [HINT. In M , use to guess nice names
for subsets of 1 .]
(v) Suppose that (P is a c.c.c. forcing) M and M[G] = . Show M = . [HINT.
Remember the equivalent characterization of from Exercise Sheet 3 and the
lemma about approximating functions in c.c.c. generic extensions.]
(vi) Optional Prove that is independent of ZF C .
9 The Generalized System Lemma
V
(i) Suppose < = cf () and < < . Prove that if  A  = and
x A implies  x  < , then there exists B A such that  B  = and
(r)(x
S B)(y B)(x 6= y x y = r) . [HINT. This is a standard result.
WLOG A Vand some < is the order type of every x = hx() : <
i A . Using < < and = cf () , let 0 be the minimal such
that {x() : x A} is cofinal in ; let = sup{x() + 1 : x A < 0 } ,
so (*) x 0 < ; now define by induction {x : < } such that
x (0 ) > max{, sup{x () : < < }} . Use (*) and < to refine
{x : < } and extract a root r .]
(ii) Find a family of finite sets such that no subfamily of size has a root.
10 The Levy collapse and its Basic Properties
Suppose S Ord and is a cardinal. Define Col(, S) = {p : p is a function
and  p  < dom(p) S ((, ) dom(p))(p(, ) = 0 p(, ) )} ,
and p Col(,S) q if and only if p q . This forcing is called Levy forcing. It adds
surjections from onto every S . For example, if > , then Col(, {})
collapses   to . In particular, Col(, ) makes a cardinal into the cardinal
successor of in a generic extension (so the idiom is slightly misleading, since
remains a cardinal and only those cardinals strictly between and are collapsed).
Levy forcing is frequently used when is an inaccessible cardinal, in order to obtain
desirable properties on successor (and other) cardinals in generic extensions. But its
applications are fundamental and much more farreaching in large cardinal theory.
In the following, is a regular cardinal.
(i) Col(, S) is closed.
(ii) If = cf () > and either is inaccessible, or = , then Col(, ) has
the chain condition. [HINT. Apply the Generalized System Lemma, as
in the case of the c.c.c. for F n(A, B, 0 ) .]
(iii) If Col(, ) has the chain condition, then forcing with Col(, ) preserves
cardinals and .
5
(iv) Suppose = cf () , Col(, ) has the chain condition, and G is Col(, ) generic. Then for any f M[G] such that f : Ord , where < , there
exists < such that f M[G Col(, )] . [HINT. Let f = G and for <
, let A be a maximal
antichain such that (p A )()(p
()
= ) .
S
Deduce ()(p < A dom(p) ) . To see f M[G Col(, )] ,
observe f () = if and only if p
()
= where p is the unique member
of G Col(, ) (see Question 3).]
11 Optional: Moderately large cardinals do not decide CH
Recall the arrow notation from the partition calculus: () is the statement:
for every function f : [X] B , where  X  and  B  = , there exists
Y X such that  Y  and f [Y ] b for some b B . In this notation,
Ramseys infinite theorem is written (n )(k )(0 (0 )nk ) . Analogously,
()<
is the statement: for every function f : [X]< B , where  X 
and  B  = , there exists YS X such that  Y  and f [Y ]< b for
some b B , where [X]< = < [X] . In these cases, the function f is called a
colouring of the element subsets of X with colours (or a B colouring of X );
Y is called a homogeneous set for f . A cardinal is Ramsey if ()<
2 .
(i) Prove that the set 2 with the lexicographic order lex contains no increasing
or decreasing sequences of length + . [HINT. Otherwise, suppose H is e.g.
lex increasing, of size + ; WLOG, H = {h : < + } and for some least
, g, h H, g 6= h . Now find < such that {h : < + }
has cardinality + .]
(ii) Prove that 2 9 (+ )22 . [HINT. Otherwise, consider a homogeneous set Y of
size + for the 2 colouring F of [ 2]2 given by F ({f, g}) = 0 if and only if
f lex g . This is due to Sierpi
nski and Kurepa independently.]
(iii) Prove that if is a Ramsey cardinal, then is strongly inaccessible. [HINT.
Regularity S
is easy: use the colouring c({, }) = 0 ()({, } X )
where = < X ,  X  < ; use part (ii) for strong inaccessibility.]
(iv) Prove that is a Ramsey cardinal if and only if for all < , ()<
.
<
[HINT. For the hard direction, if f : [] , define a 2 colouring g as
follows: g({1 , . . . m }) = 0 n = 2m f ({1 , . . . m }) = f ({m+1 , . . . 2m }) .
Notice if H is homogeneous for g and of cardinality , then g [H]n 0
and so H is also homogeneous for f .]
(v) Suppose that M = ( is a Ramsey cardinal and  P < ) . Let G be generic
in P over M . Prove that M[G] = ( is a Ramsey cardinal). [HINT. If
M[G] = (G is a colouring of [)]< ), and this is forced by some condition
p G , consider the colouring g : []< P (P 2) defined by g(a) = {hq, xi :
p P q q
( (a)
= x)}
. Note g M and P (P 2) has cardinality
for some < (by strong inaccessibility). Use part (iv) in M to obtain a
homogeneous set Y M ; show p
(Y is homogeneous for ).]
6
(vi) Deduce that CH is independent of ZF C + ()( is a Ramsey cardinal).
Remark: This type of result in very general form is due to Levy and Solovay
(1967). See A. Kanamori, The Higher Infinite, Springer, 2009. The power of large
cardinals to decide a statement is thus circumscribed by the existence of forcings
of relatively small size if such forcings can used to prove the independence of .
(In the optional 25th lecture, part (v) was proved for measurable cardinals.)
12 Optional: GCH and Diamonds
(i) Let be a regular cardinal and E be a stationary subset of . Let (E)
denote the statement: there exists a family {S [P ()]< : E} such that
X there exists a club C such that X S for all C E .
The sequence {S : E} is called a (E) sequence.
(a) Prove (E) implies (F ) for every F such that E F is stationary
in . (Note for a stationary subset S of , (S) is S in the notation
of the lecture notes, where the additional distinguishing detail was not
required.)
(ii) Suppose = 2 = + and = cf () < . Let E = { < : cf () = } .
(a) Show there is an enumeration {X : < } of all the bounded subsets
of .
(b) Assume = . For E , let S = {Y : Y {X : < }
and  Y  } . Prove {S : E} is a (E) sequence. [HINT. For
all W , define a club D as follows: let 0 D ; if D , let s ,
the successor of in D , be the least ordinal > so that for some
< , X = W . Now consider the set C of limit points of D and
check C is as required.]
(c) Assume is singular, cf () = 6= , and for every < , < .
For E , let {i : i < } be strictly increasing and cofinal in .
Fix a sequence {Uj : j < } of sets such that = j< Uj and for all
j,  Uj  < . Let S = {Y : Y {X : Uj } for some j and
 Y  } . Prove {S : E} is a (E) sequence.[HINT. Consider W
and D as previously defined. If E is a limit point
Sof D , then there
exists {i < : i < } such that if I []< , then iI Xi = W .
Since 6= , there exists j < such that  Uj {i < : i < }  = .]
(iii) Assume GCH . Suppose is a cardinal and cf () = .
(a) Suppose = . Prove + ({ < + : cf () 6= }) holds. [HINT. Use
(ii)(b).]
(b) Suppose 6= . Prove + ({ < + : cf () 6= }) holds. [HINT. Use
(ii)(c).]
(iv) Deduce GCH implies + for all = cf () 1 .
Remark: This exercise sketches an early proof of + from GCH due to Gregory
(1976) and Shelah (1981); in the course lecture notes, there is a short, much simplified, unpublished proof discovered by Peter Komjath, inspired by a new proof of
Shelah: S. Shelah, Diamonds, Proc. Amer. Math. Soc. 138 (2010), 21512161.
13 Optional
(i) Consider the ordinal 1 + 1 with the order topology and let X = (1 + 1) be
the topological space with the product topology. Let A be the set of successor
ordinals in 1 . Show that for A , the set G = {f X : range(f )}
is dense and open in X . Show that the family {H : A} , where H =
{h X : h(0) = } , is a collection of nonempty open pairwise disjoint sets.
(ii) Prove the Continuum Hypothesis is equivalent to the statement: in every
compact Hausdorff space, the intersection of less than 20 many dense open
sets is nonempty. [HINT. Forwards, the Baire Category Theorem; backwards,
part (i) if CH fails.]
14 Optional
(i) Prove that nonisomorphism (of groups) is not absolute. [HINT. Free groups
of different infinite cardinalities; collapse. Now state a generalisation.]
(ii) An infinite abelian group A is called almost free if every subgroup B of
cardinality less than  A  is free. Show the statement that the group Z is
almost free is independent of ZF C . [HINT. You may take it as proven that
Z is 1 free, but is not 2 free.]
15 Optional: Forcing and Partial Isomorphisms
Suppose A and B are structures in a vocabulary . Say A and B are partially
isomorphic, denoted A 'p B , if some nonempty family F Part(A, B) of the
partial isomorphisms from A to B is a backandforth set for A and B :
(f F)(a A)(g F)(f g a dom(g)) and
(f F)(b B)(g F)(f g b range(g)) .
(i) Prove if , A and B are countable, then A 'p B . [Use the idea of your proof
of Cantors theorem from a previous exercise.]
(ii) Show the converse of (i) fails. [HINT. Consider the linear orders Q and R .]
(iii) Show that if two structures A and B are partially isomorphic, then there is
a forcing extension in which A and B are isomorphic.
Remark: Partial isomorphism yields a characterization of elementary equivalence
in the infinitary language L . For a recent introduction to these ideas, see J.
Vaananen, Models and Games, Cambridge University Press, 2011.
16 Optional
(i) Prove Erdoss Theorem: if the Continuum Hypothesis holds then there is
an uncountable orbitcountable family of entire holomorphic functions on C .
[HINT. This is an ingenious elementary argument; see the original paper: P.
Erdos, An interpolation problem associated with the Continuum Hypothesis,
Mich. Math. J. 11, 910 (1964), or pages 335336 in P. Komjath, V. Totik,
Problems and Theorems in Classical Set Theory, Springer, 2006.]
(ii) Deduce that the Continuum Hypothesis is equivalent to the assertion that
some orbitcountable family is uncountable.
(iii) Deduce that the statement (*) every orbitcountable family is countable is
independent of ZF C .
17 Let ZF C be the theory ZF C Power Set Axiom.
(i) Suppose = cf () 1 . Prove that H is a model of ZF C .
(ii) Deduce that no proof of the existence of R avoids some nontrivial use of the
Power Set Axiom.
18 Optional
Let ZF C be the theory ZF C Power Set Axiom. Suppose = cf () 1 .
Let N be a countable elementary submodel of H .
(i) Suppose (x) is a formula in the vocabulary of ZF C (possibly with parameters from N ) such that ZF C ` (!x)(x) . Show if a H and H = [a] ,
then a N .
V
(ii) Suppose f H , dom(f ) = {a1 , . . . , an } N, 1in f (ai ) N . Prove
f N.
(iii) H ; 1 H ; H .
(iv) If {a, A, B, f } N , a A , and f : A B is a function (in V ), then
f (a) N .
(v) If X N and X is countable, then X N .
(vi) For every ordinal 1 {1 } , N is an ordinal.
(vii) If X = {X : 1 } N , then X N for every 1 N .
19 Optional
For a forcing P , a cardinal is large enough (for P ) if = cf () > 1 and the set
of dense subsets of P is an element of H (so, in particular, P , the conditions in
P and every dense subset of P all belong to H ). For a set N , a condition p P
is called N generic if for every D N which is a dense subset of P , D N is
predense above p .
Suppose is large enough for P . Prove the following are equivalent:
9
(1) P has the countable chain condition;
(2) for every countable elementary submodel N of H , 0P is N generic;
(3) every countable subset X of H is contained in a countable elementary
submodel N of H such that 0P is N generic.
[HINT. For (1) (2) , consider an A N maximal relative to the property of
being an antichain contained in D . For (3) (1) , show if A N is a maximal
antichain, then A = {p P : (q A)(q P p)} N is dense.]
20 Optional: Martins Maximum
A forcing P is called stationarypreserving if P does not destroy stationary subsets
of 1 : if M = (S is a stationary subset of 1 ), then M[G] = (S is a stationary
subset of 1 ), whenever G is generic in P over M . Martins Maximum is the
statement M M : for every stationarypreserving forcing P , if ( < 1 )(D is
dense open in P) , then there exists a {D : < 1 } generic filter G in P .
(i) Show that if P is c.c.c, then P is stationarypreserving.
(ii) Give an example of a stationarypreserving forcing that has an uncountable
antichain.
(iii) M M implies M A1 .
Remark: The relative consistency strength of M M is far stronger than that of
M A which is equiconsistent with ZF C ; M M requires a large cardinal axiom for
its consistency.
21 Optional: Normal Functions and Mahlo Cardinals
A (class) function G : Ord Ord is called normal if G is S
increasing ( <
G() < G() ) and continuous (for all limit Ord, G() = < G()) .
(i) (a) Prove in ZF C that every normal function G has a fixed point: there
exists Ord such that G() = .
(b) Call the statement every normal function has a regular fixed point the
regular fixed point axiom RF P A . Show that RF P A is not provable in
ZFC.
(ii) A strongly inaccessible cardinal is called Mahlo if { < : is a regular
cardinal } is stationary in .
(a) Suppose is Mahlo. Show is a regular limit cardinal, and is the
limit of inaccessible cardinals.
(b) Prove that if is Mahlo, then V = RF P A .
22 Open Question: Banach Spaces and Cofinality
The cofinality of a Banach space E is the least ordinal such that there exists
an increasing chain hE : < i of proper closed subspaces of E whose union
is dense in E . Does every infinitedimensional Banach space have cofinality ?
10
Remark: This is an equivalent reformulation of the Separable Quotient Problem:
does every infinitedimensional Banach space X have a separable infinitedimensional
quotient X/Y ? See S. Todorcevic, Combinatorial dichotomies in set theory, Bull.
Symbolic Logic, 17 (2011), 172.
23 Open Question: Abelian groups and the replicating cofinality
Let us call the replicating cofinality, rcf (A) , of an infinite abelian group A , the
least ordinal , if it exists, such that there exists an increasing
chain hA : < i
S
of proper subgroups of A such that A = A and A = < A . [More generally,
the replicating cofinality,
S rcf (B) , of a structure B , is the least ordinal , if it
exists, such that B = < B for some increasing chain hB : < i of proper
submodels of B , where B
= B .]
L
(i) What is the replicating cofinality of the free abelian group
< Z of cardinality , if 1 ?
(ii) Open Question What is the replicating cofinality of the group Z ?
Remark: It is known that Z is not the union of a countable chain of proper
subgroups each isomorphic to Z ; see J. Irwin, A. Blass, Baer meets Baire:
Applications of category arguments and descriptive set theory to Z , in: Arnold,
D. M. et al. (eds.), Abelian Groups and Modules, New York, NY, Marcel
Dekker, Lect. Notes Pure Appl. Math. 182, 1996, pp. 193202.
24 Open Question: Galvins Conjecture
Let D be the least cardinal , if it exists, such that for every partial order P , if
every suborder of P of size less than can be decomposed into countably many
chains, then P can also be decomposed into countably many chains. Galvins Conjecture states that 2 is a possible value for D . See S. Todorcevic, Combinatorial
dichotomies in set theory, Bull. Symbolic Logic, 17 (2011), 172.
Remark. After the classic papers of Godel and Cohen, the following are accessible and
list many further suggestions for reading and research:
Woodin, W. H., The Continuum Hypothesis, part I, Notices Amer. Math. Soc.
48(2001), 567576.
Woodin, W. H., The Continuum Hypothesis, part II, Notices Amer. Math. Soc.
48(2001), 681690.
Dehornoy, P., Recent progress on the Continuum Hypothesis (after Woodin);
http://www.math.unicaen.fr/~dehornoy/Surveys/DgtUS.pdf;
http://www.math.unicaen.fr/~dehornoy/Surveys/Dgt.pdf.
Koellner, P., The Continuum Hypothesis, Stanford Encyclopaedia of Philosophy,
September 2011;
http://www.logic.harvard.edu/EFI_CH.pdf.
11
Steprans, J., History of the Continuum in the Twentieth Century, to appear in:
Vol. 6, History of Logic;
http://www.math.yorku.ca/~steprans/Research/PDFSOfArticles/hoc2INDEXED.pdf
Remark. For research problems in set theory, there are some treasure houses to visit:
Shelah, S., On what I do not understand (and have something to say): Part I, Fund.
Math. 166(2000), 182;
http://matwbn.icm.edu.pl/ksiazki/fm/fm166/fm16612.pdf.
And for the modeltheoretic pendant:
Shelah, S., On what I do not understand (and have something to say ), model theory,
Math Japonica 51 (2000), 329377;
http://shelah.logic.at/files/702.pdf.
Fremlin, D.H., Problems;
http://www.essex.ac.uk/maths/people/fremlin/problems.pdf.
Miller, A.W., Some interesting problems;
http://www.math.wisc.edu/~miller/res/problem.pdf.
S. Todorcevic, Combinatorial dichotomies in set theory, Bull. Symbolic Logic, 17
(2011), 172.
12