TWO INFINITE QUANTITIES AND THEIR SURPRISING RELATIONSHIP

DÁNIEL T. SOUKUP

As early as the 17th century, Galileo Galilei

wondered how to compare the sizes of infinite sets.
Fast forward almost four hundred years, and in the
summer of 2017, at the 6th European Set Theory
arXiv:1803.04331v1 [math.HO] 12 Mar 2018

Conference, a young model theorist, Maryanthe

Malliaris, and the well-known polymath, Saharon
Shelah, received the Hausdorff Medal for the most
influential work in set theory published in the last
five years. Malliaris and Shelah made significant
breakthroughs both regarding a model theoretic
classification problem (that is, sorting certain ob-
jects into types), and proved that two well-studied
infinite quantities, p and t, are in fact the same.
This latter result is the focus of our expository S. Shelah and M. Malliaris in the center
paper. (Photo by Joan Bagaria)

Galilei considered the set of natural numbers N and the set of perfect squares {1, 4, 9, 16 . . . }. The
argument, that these two sets have the same size, goes as follows: since any perfect square has exactly
one positive root, and any positive natural number is the root of some perfect square, there should be
the same amount of perfect squares and positive natural numbers. On the other hand, there are many
natural numbers which are not perfect squares; indeed, if one looks at the ratio of perfect squares to all
natural numbers in larger and larger intervals, this quantity tends to zero rather fast. Galilei thought of
these observations as a paradox, which prevents us from distinguishing between the sizes of infinite sets.
In the 1870s, Georg Cantor came forward with the following definition, now accepted as standard:
two sets are equinumerous or have the same cardinality exactly if there is a one-to-one correspondence
between the elements of the two sets. So, Galilei’s first argument proves that N and the set of perfect
squares have the same cardinality. Those sets which are equinumerous with N are called countably infinite
and we use ℵ0 (in plain words ’aleph zero’) to denote their size; the ℵ0 notation refers to the fact that
this is the smallest possible infinite size, or cardinality in other words.1
One of Cantor’s great contributions to logic was that he did not consider Galileo’s argument as an
irresolvable paradox, but instead, he started to develop a rich theory of infinities.
Quite surprisingly, he proved that even the set
of rational numbers Q is countably infinite. Now,
how about the set of all real number R? Can-
tor claims, that no matter how we produce a list
x1 , x2 , x3 , . . . of real numbers, there is always a
real number y which is missing from our enumer-
ation. Indeed, if y differs from x1 at the first dec-
imal place, and y and x2 differ at the second dec-
imal place, and so on, then y cannot possibly ap-
pear on the list. Hence, there is no one-to-one
correspondence between the natural numbers and

1We say that a set X has cardinality at most the cardinality of Y , if X and some subset of Y are equinumerous.

1
2 D. T. SOUKUP

the set R, and so R must be uncountable. We use the notation 2ℵ0 (the ’continuum’) for the cardinality
of R, and in turn, we proved the following inequality:
the cardinality of N = ℵ0 < 2ℵ0 = the cardinality of R.
Based on Cantor’s theory, one can compare the cardinalities of any two sets, and in any non empty
collection of sets, there is a smallest in size. So, it makes sense to define the first infinite cardinality which
is uncountable, and we use ℵ1 to denote this. The next larger cardinality will be denoted by ℵ2 , then ℵ3 ,
and so on.2

At this point, we defined a strictly increasing

sequence ℵ0 < ℵ1 < ℵ2 < . . . of infinite cardinali- Is there a largest infinite cardinality? Not
ties. Where does 2ℵ0 sit in this list? Interestingly, according to Cantor’s definition. Any set X has
this question is undecidable using the generally ac- more subsets than elements, in other words,
cepted ZFC axioms of mathematics.3 Put it differ- κ < 2κ for any cardinality κ.
ently, in some models of mathematics the equation
2ℵ0 = ℵ1 holds i.e., R has the smallest possible uncountable cardinality, and in this case, we say that the
Continuum Hypothesis holds. In many other interesting and important models, the Continuum Hypoth-
esis fails and the gap between ℵ0 and 2ℵ0 can be arbitrary large: some models satisfy ad hoc equations,
such as 2ℵ0 = ℵ16 .
A simple analogy might shed more light on this undecidability business: R and Q, as algebraic struc-
tures, both satisfy certain axioms of addition and multiplication,4 however the equation x2 = 2 has two
roots in R, but none in Q. In other words, whether the statement ’x2 = 2 has a solution’ is true or
false, is not decided by the axioms of addition and multiplication. Similarly, while the ZFC axiom system
decides that ’the inner angles of a plane triangle sum to 180 degrees’, the axioms do not decide whether
2ℵ0 = ℵ1 or 2ℵ0 > ℵ1 holds; one of these statements holds in any model, but it depends on the particular
model which. Another quite similar story was unfolding with the discovery of non-standard geometries
in the 19th century and the independence of the parallel postulate from other geometric axioms.
In the last fifty years, people constructed var-
How does one make new models? Kurt ious different models of mathematics, which have
Gödel, in 1938, discovered the so-called con- plenty of cardinalities between ℵ0 and 2ℵ0 , and
structible universe, and proved that the Contin- a significant portion of modern set theory focuses
uum Hypothesis holds in this model of mathemat- on the analysis of these models. Now, two models
ics. Then, in the 1960s, Paul Cohen genuinely sur- of the ZFC axioms which both satisfy 2ℵ0 = ℵ16
prised the set theoretic community: he was the can behave very differently, even if one only con-
first to show that, in some other models of ZFC, siders theorems from algebra, measure theory or
the Continuum Hypothesis fails i.e., ℵ1 < 2ℵ0 . Co- topology. To the surprise of many specialists,
hen’s technique is called forcing, and its underlying the Whitehead-problem in group theory or the
idea is fairly simple: given a set theoretic model existence of outer automorphisms on the Calkin-
M , we can construct a larger model N by adding algebra are both undecidable using the usual ax-
a generic object G to M . We can use the generic ioms. However, in light of present techniques, it is
G to increase the value of 2ℵ0 , which leads to the hard to see how say the Riemann-hypothesis or the
failure of the Continuum Hypothesis. Cohen re- existence of general solutions for the Navier-Stokes
ceived a Fields Medal for his work in set theory in equations could be undecidable.5
1966. On the bright side, the study of the real line and
its fundamental topological and measure theoretic
properties is possible through looking at combinatorial properties of families of subsets of the natural

2The list of ℵ’s does not stop here: the ℵ cardinalities indexed by the natural number have a supremum, denoted by
n
ℵω , the next strictly larger cardinality is ℵω+1 , then ℵω+2 ...
3ZFC stands for the Zermelo-Fraenkel axiom system with the ’Axiom of Choice’.
4Think about commutativity, distributivity or the field axioms in general.
5Anyone who solves one of these problems, will be awarded the Millenium Prize and its one million dollar prize purse.
 TWO INFINITE QUANTITIES AND THEIR SURPRISING RELATIONSHIP 3

numbers; indeed, there is a one-to-one correspondence between the real numbers and subsets of N.6
Studying the so-called cardinal characteristics of the continuum deals with exactly such matters.

What attracts many people to this area of research is that one can understand a breakthrough result
such as Malliaris and Shelah’s with minimal background. From now on, we will only talk about subsets
of the natural numbers, and our concern is the following relation: if A and B are sets of natural numbers,
then we write A ⊆∗ B (in plain words, ’A is almost contained in B’) if all but finitely many elements
of A are elements of B as well. For example, the set A = {1, 3, 5, 7, . . . } is almost contained in B =
{4, 5, 6, 7, . . . }, since each element of A, apart from 1 and 3, is an element of B too.
What is the advantage of working with such a weak relation instead of the real containment? Lets
take the sets An of positive natural numbers which are divisible by n: so A1 is the set of all positive
natural numbers, A2 collects the even natural numbers, A3 = {3, 6, 9 . . . }, and so on.
It is easy to see that if we take finitely many sets
A tower A1 , A2 . . . An until a fixed n, then these sets have infinite
intersection. Indeed, the multiples of n! = 1 · 2 · . . . ·
n are contained in each of the sets A1 , A2 . . . An . We
usually say in this case that the system of sets {An } has
the finite intersection property, a rather counterintuitive
name for sets with infinite intersection...
Of course we cannot find any positive natural number
which is divisible by all the numbers, that is, there is
no real intersection to the whole family {An }. On the
The
nite intersection property other hand, we can easily find infinite sets B so that
B ⊆∗ An for all n; it suffices to make sure that the nth
element of B is selected from A1 ∩ A2 ∩ · · · ∩ An , so for example B = {n!}n=1,2... works. Since B behaves
somewhat like an intersection to the family {An }, we call B a pseudo-intersection of {An }. It is also easily
checked, that one can add B to the family {An }, and the finite intersection property is still preserved.
Now, given an arbitrary family of sets A with the finite intersection property, we can extend A to a
maximal family Amax with the finite intersection property.7 This maximal family Amax however cannot
have a pseudo-intersection anymore.8 This leads to our first main definition:

The pseudo-intersection number p is the cardinal-

ity of the smallest family of sets A with the finite
intersection property, such that A does not have a
pseudo-intersection.

In the above example, we essentially showed that any family indexed by the natural numbers, which
has the finite intersection property, also has a pseudo-intersection, and in turn, p must be uncountable.
Hence, the following inequality holds:
ℵ0 < p ≤ 2ℵ0 .
There are numerous models of mathematics, where p = 2ℵ0 and this common value can be essentially
any ℵ. On the other hand, ℵ1 = p < 2ℵ0 = ℵ2 can also be true in other models. So the usual axioms do
not decide where the pseudo-intersection number p sits on the list ℵ1 , ℵ2 , . . . , nor if p = 2ℵ0 or p < 2ℵ0
holds.

We need another definition. A typical family with the finite intersection property is not ordered in
any sense: in our original example about divisibility, if one only considers the sets Ap for primes p, no
two of these sets are in ⊆∗ relation. Now, call a family of sets T a tower if for any two sets X, Y from
T , either X ⊆∗ Y or Y ⊆∗ X holds. In other words, the relation ⊆∗ linearly orders T .
6In other words, the cardinality of all subsets of N is 2ℵ0 .
7This is a standard application of Zorn’s lemma.
8Otherwise, we could add this pseudo-intersection to A
max while preserving the finite intersection property, and so
maximality would be violated.
4 D. T. SOUKUP

The so called tower number t is the cardinality of the

smallest tower T without pseudo-intersection.

Since any tower has the finite intersection prop-

erty, any witness for the invariant p must have at Large towers. Surprisingly, even from the subsets
most the size of a tower witnessing t. This shows of N, one can construct towers of size 2ℵ0 . First
the next equation: of all, list all the rational numbers as q1 , q2 , q3 , . . . .
ℵ0 < p ≤ t ≤ 2ℵ0 . Then, for each real number r, we define Xr to be
The value of the invariant t can be manipulated the set of all those indexes n so that qn < r. So Xr
similarly to p. Actually, since the 1940s, more than essentially collects the rational numbers smaller than
a dozen cardinal invariants between ℵ0 and 2ℵ0 r. Now, if r < t are two real numbers then Xr is fully
have been studied. Moreover, we know that apart contained in Xt , moreover, Xt has infinitely many
from certain simple inequalities that were known extra elements. Can we find a pseudo-intersection
from the mid 20th century, there is no provable for this family?
relationship between the invariants. That is, with
various versions and combinations of forcing techniques, one can not only set the value of 2ℵ0 but also
the values of the invariants to any fixed alephs, which do not violate the known inequalities.9
Indeed, the most advanced techniques can indepen-
dently manipulate the values of more than five invari-
ants at the same time.

Despite all these investigations in the last sixty years,

no one constructed a model, where p and t are not equal.
The generally accepted conjecture was that p < t is
possible, however, we knew that a proof could not be too
easy: it was proved a long time ago that if p = ℵ1 , then
t = ℵ1 holds as well. So, p < t can only hold in models
where 2ℵ0 is bigger than ℵ2 , however fine tuning such
models is significantly harder than controlling models
with 2ℵ0 ≤ ℵ2 .
Malliaris and Shelah’s new, unexpected result states
A few cardinal invariants and their provable that
relationships p = t,
no matter which model we look at. What is it exactly that the authors prove? Since p ≤ t was known,
they needed to show t ≤ p: given an arbitrary family A with the finite intersection property but no
pseudo-intersection, no matter how random or involved the overlays between the elements of A are, one
can construct a tower T of size at most the size of A which has no pseudo-intersection. The naive
approach to construct the tower from the elements of A fails quickly, since A might have no two elements
which are related by ⊆∗ , but T must be totally ordered by ⊆∗ .
One can compare Malliaris and Shelah’s result to showing that two algebraic equations have the same
solutions, but without actually determining the value of these solutions. The ZFC axioms do not decide
whether p = ℵ1 or p = ℵ2 or p = ℵ3 , and similarly the value of t is undecidable. However, Malliaris and
Shelah could prove that no matter what the value of p is, it has to be the same as t.

We should mention that not only the authors solved the sixty year old mystery surrounding p and t, but
they uncovered a novel connection between a model theoretic complexity hierarchy, the Keisler-order, and
the theory of cardinal characteristics [6, 7, 8]; unfortunately, it is not in the scope of our paper to sketch
these results. While the original proof of p = t employs serious tools from model theory and modern set
theory, there are now new versions which only require basic knowledge and some perseverance [9]10
9We recommend A. Blass’ classical text on cardinal invariants for an excellent overview.
10The Fields Medalist, Timothy Gowers’ blog also looks at this problem, link here.
 TWO INFINITE QUANTITIES AND THEIR SURPRISING RELATIONSHIP 5

An open problem. We say that a family R The results of Malliaris and Shelah are far from
is unsplit if there is no Y which splits all el- the last of cardinal characteristics, and will more
ements of R at the same time, i.e., no Y so likely spark a renewed interest in the field. So
that the intersection X ∩ Y and the difference what problems does a regular set theorist work
X − Y are both infinite for any X from R. Let on? On one hand, the relationship of some classi-
r denote the cardinality of the smallest unsplit cal invariants are still unknown, and we mention
family. Moreover, let rσ denote the size of the a problem of this sort in the side note [3]. On the
smallest family which cannot be split by even other hand, people are defining new, interesting in-
countably many sets Y0 , Y1 , . . . . It is easy to variants to this day, and it is often a hard task to
see that ℵ0 < r ≤ rσ ≤ 2ℵ0 , however, it remains determine the position of these new invariants rel-
unknown if r < rσ is possible in some model. ative to the classical ones [2]. Finally, a rich theory
The conjecture is that r = rσ holds, and this is is growing out of the study of cardinal characteris-
certainly true in all known models [3]. tics which are defined using families of uncountable
sets, rather than the subsets of N, showing striking
differences with the classical studies [4].

We close by a few words about the awardees:

Maryanthe Malliaris graduated from Berkeley in 2009, currently a professor at the University of Chicago,
she is the recipient of multiple, prestigious awards, and is invited to present at the International Congress
of Mathematicians in 2018.
The name Saharon Shelah might ring a bell for a lot of the readers: the 72 years old mathematician is
the author of 1023 published papers (!), on groundbreaking results from combinatorics and model theory,
to logic and group theory. He still works 6 days a week, splitting the year between Rutgers and the
Hebrew University of Jerusalem.

The author was supported in part by the FWF Grant I1921. A Hungarian version of the current survey was prepared for
the journal Matematikai Lapok. We thank Emese Bottyán, Lajos Soukup, Zoltán Vidnyánszky, and Zita Zádorvölgyi for
their careful reading, while the English version significantly improved thanks to the help of Neil Barton and Vera Fischer.

References
[1] A. Blass. Combinatorial cardinal characteristics of the continuum. Handbook of set theory, pages 395–489, 2010.
[2] A. Blass, J. Brendle, W. Brian, J. D. Hamkins, M. Hardy, and P. B. Larson. The rearrangement number. arXiv
preprint:1612.07830, 2016.
[3] J. Brendle. Around splitting and reaping. Commentationes Mathematicae Universitatis Carolinae, 39(2):269–279, 1998.
[4] J. Brendle, A. Brooke-Taylor, S.-D. Friedman, and D. Montoya. Cichon’s diagram for uncountable cardinals. to appear
in Israel Journal of Mathematics, arXiv:1611.08140, 2016.
[5] G. Galilei. Dialogue concerning the two chief world systems, Ptolemaic & Copernican. Univ of California Press, 1967.
[6] M. Malliaris and S. Shelah. General topology meets model theory, on p and t. Proceedings of the National Academy of
Sciences, 110(33):13300–13305, 2013.
[7] M. Malliaris and S. Shelah. Cofinality spectrum theorems in model theory, set theory, and general topology. Journal of
the American Mathematical Society, 29(1):237–297, 2016.
[8] J. T. Moore. Model theory and the cardinal numbers p and t. Proceedings of the National Academy of Sciences,
110(33):13238–13239, 2013.
[9] G. M. Roccasalvo. Ultraproducts of finite partial orders and some of their applications in model theory and set theory.
Master’s thesis, University of Torino, 2014.

(D.T. Soukup) Universität Wien, Kurt Gödel Research Center for Mathematical Logic, Austria
E-mail address, Corresponding author: [email protected]
URL: http://www.logic.univie.ac.at/∼soukupd73/