GENERAL

ARTICLE

Transfinite Numbers
What is Infinity?
S M Srivastava
In a series of revolutionary articles written during the
last quarter of the nineteenth century, the great German mathematician Georg Cantor removed the age-old
mistrust of infinity and created an exceptionally beautiful and useful theory of transfinite numbers. This is
an introductory article on this topic. 1
S M Srivastava is with the
Indian Statistical
Institute, Calcutta. He
received his PhD from the
Indian Statistical Institute
in 1980. His research
interests are in
Descriptive Set Theory.

The author would like to

gratefully acknowledge the
help received from 5 C Bagchi.
T Krishnan. B V Rao and H
Sarbadhikari while preparing
this article.

Introduction
How many points are there on a line? Which is more in numberpoints on a line or lines in a plane? These are some natural
questions that have occurred to us sometime or the other. It is
interesting to note the difference between the two questions. Do
we have to know how many points and lines there are to answer
the second question? Even primitive man could see whether he
had more cows than his neighbour without knowing the actual
number of cows they had-he would just pair off his cows with
his neighbour's. So, to compare the size of two sets, we can start
pairing off their elements. If one of them exhausts first we say
that that particular set has fewer elements than the other one.
Otherwise we say that the two sets have the same number of elements. If we are considering finite collections only, everything
seems to be alright. Now imagine yourself walking into a hotel
with infinitely many rooms which are all occupied. The receptionist shifts the guest in Room 1 to Room 2, the one in Room
2 to Room 3 and so on. Now Room 1 falls vacant and she gives
it to you. As we can pair off the two sets of guests, question
arises: are there more guests in the hotel now? Thus we see
that the idea of pairing off elements leads to counterintuitive
results for infinite sets, e.g., an infinite set can have as many
elements as a part of it. Georg Cantor (1845-1918) showed that
this is a characteristic difference between finite and infinite sets
and created an immensely useful branch of mathematics based
on this idea which had a great impact on the whole of mathematics. For example, the question of what is a number (finite or
infinite) is almost a philosophical one. However Cantor's work
turned it into a precise mathematical concept.

--------~~------

GENERAL

ARTICLE

Countable Sets

How many points

The simplest sets are those whose elements can be counted as

the first, the second, the third etc. with the possibility that the
counting may not stop. Such sets are called countable. Mathematically, a set A is countable if there is a one-one map from
the set N of natural numbers 0,1,2, ... , or from {O, 1"" n -I}
(n being a natural number) onto A. (For n = 0 we take the set
{O, 1, n -I} to be the empty set.) If a set is not countable we
call it uncountable. You can easily see that a set is countable if
and only if its elements can be enumerated as aO, aI, a2, .. " may
be by repeating some of its elements. It follows that a subset of
a countable set is countable.

are there on a
line? Which is
more in numberpoints on a line or
lines in a plane?

Examples
We can enumerate the set N x N of ordered pairs of natural numbers by the diagonal method as shown in the following
diagram

(1,2)

(2,1)

(2,2)

That is, we enumerate the elements of NxN as (0,0), (1, 0), (0, 1),
(2,0), (1,1), (0,2), .... By induction we see that the set of all
k-tuples of natural numbers is also countable.
An argument similar to the one above shows that if Ao, AI, A2, ...
is a sequence of countable sets then their union A = UO An is
also countable.
The set Q of all rational numbers is countable. (Exercise: Try
to deduce this from the first example.)
The most natural question that arises now is: are there uncountable sets? The answer is yes. Here is an example.
If a < b are real numbers then the interval [a, b] is uncountable.

Proof. If possible, let [a, b] be countable and let {ao, al,"'} be

an enumeration of [a, b]. Define an increasing sequence {b n} and

--------~--------

GENERAL I ARTICLE

Say that sets A

and Bare
equinumerous or
are of the same
cardinality, written
A == B, if there
exists a one--one
map f from A onto
B.

a decreasing sequence {en} of real numbers in [a, b] inductively

as follows: Put bo = a and CO = b. For some n EN, let
bo

< bl < ... <

bn

< C n < ... < CI < co

be defined. Let in be the first integer i such that bn < ai < en

and jn be the first integer j such that ain < aj < Cn. Put
bn+l = a'n and Cn+l = ajn. Now, the number x ~ sup{bn } is in
[a, b] but is different from each of an. This is a contradiction.
Combining the above observations we easily see that every interval with more than one point contains uncountably many
numbers which are not algebraic. 2 Such numbers are caJled tmnscendentaL This simple argument of Cantor generated a lot of
interest (as well as scepticism for not giving an example of
a transcendental number but only showing their existence) in
set theory at that time. Later Hermite and Lindemann showed
respectively that 11" and e are transcendental.

Equinumerosity-Order of Infinity
An algebraic number is a
complex number which is a root
ofa polynomial having rational
numbers for coefficients. For
eg: Ii is an algebraic number;
it satisfies XZ-2=O.
2

Say that sets A and B are equinumerous or are of the same

cardinality, written A == B, if there exists a one-one map f
from A onto B.

Exercise Show that if I and J are two intervals having more

than one point (which may be bounded or unbounded, mayor
may not include their end points) they have the same cardinality.
In the last section we saw that the idea of equinumerosity, besides being a natural one, is useful too. So let us study it.
Let X and Y be sets. The collection of all subsets of a set X is
itself a set called the power set of X denoted by P(X). Similarly
the collection of all functions from Y to X forms a set which
we denote by X Y If A is a subset of X then its chamcteristic
function is the map XA : X ---+ {O, I} where
XA (x) = {

You can easily verify that A

P(X) onto {O,l}x.

I ifXEA
0 otherwise.
---+ XA

defines a one-one map from

So far we have seen only two different 'orders of infinity': those

equinumerous either with N or with the set of real numbers ~.
Are there any more? Cantor's answer was simple but breathtaking.

--------~--------

GENERAL

ARTICLE

Say tha.t the ca.rdinality of a set A is less than or equal to the

cardinality of set B, written A ::5c B, if there is a one-one function f from A to B. If A ::5c B but A 1= B then we say that the
cardinality of A is less than the cardinality of B and symbolically
write A <c B.

Is there an
uncountable set of
real numbers of
cardinality less
than that of 1{?

Notice that N <c!R.

Theorem(Cantor) For any set X, X <c P(X).
(Note that the power set of the empty set is not empty-it
contains the empty set.)
Proof. The map x ---+ {x} from X to "'(X) is one-one. Therefore, X ::;;c P(X). Now take any map f : X ----+ P(X). We show
that f cannot be mapped onto P(X). To show this consider the
set
A = {x E xix f(x)}.
(A may not have any point, thus it may be an empty set-yet
a subset of X).

H A = I(xo) for some xo E X then note that

xO E

A <==> xo A.

This contradiction proves our claim.

Now we see that

N <c peN) <c P(P(N <c ....

Let T be the union of all the sets N, peN), P(P(N,
Then T is of cardinality larger than each of the sets described
above. We can now similarly proceed with T and get a never
ending class of sets of higher cardinalities! Cantor at this point
raises a very interesting problem. Is there an infinite set whose
cardinality is different /rom the cardinalities of each of the sets
so obtained? In particular, is there an uncounta.ble set of real
numbers of cardinality less than that of!R? These turned out
to be among the most fundamental problems not only in set
theory but in the whole of mathematics. Cantor could neither
solve these problems nor did he live long enough to see their
surprising solution. We shall discuss these later.
The following result is very useful in proving equinumerosity of
two sets.
Schroder-Bernstein Theorem. For any two sets X and Y,
X $c Y & Y $c X ===> X

== Y.

-RE-S-O-N-A-N-C-E--I-M-a-rc-h-1-9-97--------------~-------------------------------~

GENERAL

ARTICLE

Proof. (Dedekind) Fix one-one maps / : X

and 9 : Y
defined by

--+

--+ Y

X. Consider the map 1t : 'P(X)

1t(A) = X \ g(Y \ /(A,

--+

P(X)

X.

Then
i) A C B eX==? 1t(A) c 1t(B), and
ii) 1t(Un An) = Un 1t(An).
Now, let
Ao =

0, and

= 1t(An), n = 0,1,2, ....

= Un An. Then, 1t(E) = E.

An+l
Let E

Define h: X ~ Y by

h(x) = { lex)
if x E E
9 -1 (x) otherwise.
The map h : X

--+

Y is easily seen to be one-one and onto.

Here are some applications of Schroder-Bernstein theorem .

The function A --+ LneA 3n\1 from 'P (N) to ~ is one-one.
Therefore, 'P(N) ~c~. On the other hand the map x ---+
{r E Qlr < x} from ~ to 'P(Q) is one-one and so, ~ ~c P(Q).
But Q == N. Therefore, ~ ~c peN). By Schroder-Bernstein
theorem, ~ == peN) (== {O, l}N) .
Fix a one-one map x --+ (xo, Xl! X2, .. ) from ~ onto {O, l}N,
the set of sequences ofO's and l's. Then the function (x,y) --+
(xo, YO, Xl, Y1, ... ) of the pairs of real variables to {O, l}N is oneone and onto. So, ~2 ==~. Thus there are precisely as many
points on the X -axis as in the entire plane! By induction on
positive integers k you can now show that ~Ic and ~ are equinumerous. Similarly, using N x N == N, you can show that ~ and
~N, the set of sequences of real numbers are equinumerous .
Show that the sets of points on a line and lines in a plane are
equinumerous.
The Axiom of Choice
Are the sizes of any two sets necessarily comparable? That is,
if X, Yare sets, is it true that at least one of the relations
X ~c Y or Y ~c X holds? To answer this question, we need a
hypothesis on sets known as the axiom of choice.

_ _ _ _ _ _ _ _LAAAAAA_ _ _ _ _ __ _
62
V VVV
RESONANCE I March 1997
v

GENERAL I ARTICLE

The Axiom of Choice (AC): If {Aihel is a family of nonempty sets then there is a function I : 1 --+ U Ai such that
I{i) E Ai for every i E 1.
Such a function I is called a choice function. Note that if 1
is finite then by induction on the number of elements in 1 we
can show that a choice function exists. If 1 is infinite then we
do not know how to prove the existence of such a map. The
problem can be explained by the following example of Russell.
Let Ao, Al, A2,'" be a sequence of pairs of shoes. Let I{n)
be the left shoe in the n-th pair An and so the choice function
in this case certainly exists. Instead, let Ao, Al, A2, . .. be a
sequence of pairs of socks. Now we are unable to give a rule to
define a choice function for the sequence of socks Ao, All A2, .. .!
AC asserts the existence of a function without giving any rule
or any construction for defining such a function. Because of its
non-constructive nature, AC met with serious criticism. For an
excellent account of AC see Gregory Moore's book mentioned in
Suggested Readings. We only remark that AC is indispensable
not only for the theory of cardinal numbers but for the whole
of mathematics. Note that while proving that the union of a
sequence of countable sets Ao, A 1, ... is countable we have used
AC. For each n, we chose an enumeration of An. But there are
infinitely many such enumerations and we did not specify any
rule to choose them. The next result shows that every infinite
set has a proper equinumerous subset. In its proof we use AC.
Theorem If X is infinite and A C X finite then X \ A and X
have the same cardinality.
Proof. By AC, fix a choice function J : P(X) \ {0} --+ X such
that f (E) E E for every non-empty subset E of X. Let A =
{ao, al, .. , an}. Inductively define an+l, a n +2, ... such that
an+k+l

N ow define h : X

--+

= f(X

\ {ao, al,, an+k}).

X \ A by

h{x) = {xan+k +1 if x = ale

otherwise.

Clearly, h : X

--+

X \ A is one-one and onto.

There are many equivalent forms of AC. We shall use Zorn's

lemma--one of its equivalent forms-to prove further results.
We explain Zorn's lemma now.
A partial order on a set P is a binary relation R such that for
x,y,zEP,

-E-S-O-N-A-N-C-E--I-M-a-rc-h-1-9-97----------------~~-----------------------------63-

GENERAL I ARTICLE

i) x Rx (reflexive),
ii) xRy and yRz implies xRz (transitive), and
iii) xRy and yRx implies x = y (anti-symmetric).
A set P with a partial order is called a partially ordered set or
simply a poset. A linear order on a set X is a partial order R
on X such that any two elements of X are comparable, i.e., for
every x, y EX, at least one of xRy or yRx holds. H X is a
set with more than one element then the inclusion relation c
on 'P(X) is a partial order which is not a linear order. Another
example of a partial order which is not a linear order is obtained
as follows. Let X and Y be sets with Y having more than one
element. Let l' be the set of all partial functions from X to Y
(functions with domain a subset of X and range contained in
Y). Order l' by 'extension' (if I, 9 E l' then 9 extends I if the
domain of f is contained in the domain of 9 and I(x) = g(x)
for every x in the domain of f).
Fix a poset (P, R). A chain in P is a subset C of P such that
R restricted to C is a linear order, i.e., for any two elements x,
y of C either xRy or yRx must be satisfied. HAc P then an
upper bound for A is an x E P such that y Rx for all yEA. An
x E P is called a maximal element of P if for no yEP different
from x, xRy holds. In the second example above, any partial
function I with domain X or range Y is a maximal element.
So, there may be many maximal elements in a poset which is
not linearly ordered.

Zorn's Lemma: H P is a non-empty partially ordered set such

that every chain in P has an upper bound in P then P has a
maximal element.
Some Consequences of the Axiom of Choice
Theorem For any two sets X and Y at least one of the relations
X ~c Y or Y ~c X holds.
Proof. Without loss of generality we can assume that both
X and Y are non-empty. We need to show that either there
exists a one-one map I : X --+ Y or there exists a one-one
map 9 : Y --+ X. Towards this let P be the set of all one-one
partial functions from X to Y. The set P is clearly non-empty.
Order it by 'extension'. Let C = {Ji : i E I} be a chain in P.
Then C is a consistent family of one-one partial functions. Let
D = UiEldomain(Ji). Define I : D --+ Y by
I(x) = Ji(x) if x E domain(Ji).

_ _ _ _ _ _ _ _LAAAAAA_ _- - - - - 64
v V VVV v
RESONANCE I March 1997

GENERAL

ARTICLE

Since Ii's are consistent and one-one, I is well defined and oneone. Thus, I is an upper bound of C. By Zorn's lemma, P
has a maximal element, say 10. We must have domain(/o) =
X or range(/o) = Y. Otherwise, choose a E X.\ domain(/o)
and b E Y\ range(fo). We get a one-one extension of 10 to
domain(/o) U{ a} by assigning b to a. This contradicts the
maxima1ity of 10. H domain(/o) = X then X ~c Y and if
range(/o) = Y then Y ~c X.

A hotel with
infinitely many
rooms even when

full can
accommodate as
many new guests
as already there!

The next result shows that a hotel with infinitely many rooms
even when full can accommodate 88 many new guests as are
already there!
Theorem For every infinite set X, X x {O, I}

== X.

Proof. Let
P

{(A, f) : A

c X and

f: A x {O, I}

--+

a bijection}.

Since X is infinite it c9ntains a countably infinite set, say D.

Since, D x {O,l} == D, P is non-empty. Partially order P by
'extension'. As before, by Zorn's lemma, get a maximal element
(A, f) of P. The result will follow if we show that A == X. Since
X is infinite, by the theorem proved in the last section, it will
be sufficient to show that X \ A is finite. Suppose not. Then
by AC there is aBc X \ A such that B == N. SO, there is
a one-one map 9 from B x {O, I} onto B. Combining f and 9
we get a bijection h : (A UB) x {O, I} --+ A UB which extends
f. This contradicts the maximality of (A, f). Hence, X \ A is
finite. Therefore, A == X. The proof is complete.
Corollary Every infinite set can be written as the union of kmany pairwise disjoint equinumerous sets where k is any positive
integer.
Theorem H X is an infinite set then X x X

== X.

Proof. Let
P

{(A, f) : A C X

and f: A x A

--+

A a bijection}.

Partially order P by 'extension'. By Zorn's lemma take a maximal element (A, f) of P. Note that A must be infinite. To
complete the proof, we shall show that A == X. Suppose not.
Then A <c X.
We first show that X \ A == X. If possible, suppose X \ A <c X.
Either A ~c X \ A or X \ A ~c A. Assume first X \ A ~c A.
Take two disjoint sets Al, A2 of the same cardinality as A. Now,

---------~---------

GENERAL

ARTICLE

This is a contradiction. Similarly using the other inequality we

arrive at the same contradiction. Thus X \ A == X. Now get
B C X \ A such that B == A. Using the last'result, write B as
the union of three disjoint sets say Bt, B2 and B3 each of the
same cardinality as A. Since there is a bijection from A x A to
A there exist bijections 11 : B x A --+ Bl, 12 : B x B --+ B2
and /3 : A x B --+ B3. Let C = AU B. Combining these four
bijections we get a bijection 9 : C x C --+ C which extends I,
This contradicts the maximality of I.

Exercise Show that an infinite set X is equinumerous to the

set of all its finite subsets.
The Continuum Hypothesis (CH):
We now get back to the two famous problems of Cantor mentioned earlier. He conjectured the following. (We are assuming
AC.)

The Continuum Hypothesis (CH) Every uncountable subset of a:l is of the same cardinality as ~.
The Generalised Continuum Hypothesis (GCH) If X is
an infinite set then there is no set A such that X <c A <c P(X).
CH says that a:l is the least numerous uncountable set. If GCH
were true then every infinite set is equinumerous to one of the
sets obtained by iterating the power set operation on N. Are
these statements true? These turned out to be very hard problems and their solutions were startling. In 1938 Kurt Godel
obtained deep results on models of set theory and showed that
based on certain axioms for sets (the so calledZermelo-Praenkel
axioms), CH and GCH cannot be disproved by producing a
'model' of set theory satisfying the Zermelo-Fraenkel axioms
where CH and GCH are satisfied.This was the first time metamathematics entered in a non-trivial way to answer a problem
in mathematics.Godel's result does not say that CH and GCH
can be proved.In 1963 Paul Cohen developed a very powerful
technique, known as forcing, to build models of set theory and

--------~--------

GENERAL I ARTICLE

showed that CH and GCH cannot be proved either. A similar

and equally famous result was proved in the nineteenth century.
It was shown that the axiom of parallels-through a point outside a given line passes a unique line not intersecting the giveJ.1
line---can neither be proved nor can be disproved. (See K Paranjape's series of articles on Geometry in this journal for details.)
Godel and Cohen's results are generally considered to be among
the greatest in twentieth century mathematics. Cohen's method
is particularly useful in proving such consistency results. A further discussion on this is beyond the scope of this article.

Arithmetic of Cardinal Numbers

For sets X, Y and Z we can easily check the following.
i) X == X,

== Y then
iii) if X == Y and
ii) if X

Y
Y

== X, and
== Z then X == z.

So, to each set X we can assign a symbol, say

cardinal number such that

== Y

<::=:}

IXI

and

IYI

lXI,

called its

are the same.

The symbols IXI are called cardinal numbers or simply cardinals.

(Cardinal numbers are denoted by Greek letters K, A, J-L with or
without suffixes.) The cardinal number of a finite set with n
elements is denoted by n itself. We put INI = No and I~I = c.
As in the case of natural numbers, we can add, multiply and
compare cardinal numbers. Towards this, let A and J-L be two
cardinal numbers and X and Y be two sets such that IX I = A
and IYI = J-L.

Definition

a) A + J-L = I(X
b) A./L = IX X
c) )..11- = IXYI

{o}) U(Y

{1})I

YI

d) A ~ J-L if X ~c Y

e) ).. <

J-L

if X < c Y

--------~--------

GENERAL

ARTICLE

The above definitions are easily seen to be independent of the

choices of X and Y Further, these extend the corresponding
notions for natural numbers. We can extend the notion to that
of sum and product for infinite collection of cardinals too. Let
{Ai} be a set of cardinal numbers. Fix a family {Xi} of sets of
cardinality Ai. We define IliAi = I Xi Xii To define Ei Ai we
take Xi's to be pairwise disjoint and put Ei Ai = I Ui Xii In
terms of these notations note that ~o.~o = ~o, 2l-to = " No < "
,l-to = ,etc.
Cardinal arithmetic is very similar to the arithmetic of natural
numbers and follows more or less the same laws. Whatever we
have proved about equinumerosity of sets, i.e., the results concerning union, product, $;c etc., translate into corresponding
results about cardinal numbers. Thus we obtain a very beautiful
and useful cardinal arithmetic, remarkably similar to the arithmetic of natural numbers. For instance, the Schroder-Bernstein
theorem translates as follows: A $; I-' & I-' ~ A ==} A == 1-'; the
result on comparability of cardinals becomes: for.
cardinals A and I-' at least one of A ~ I-' or J. ~ A holds. We
get some notable differences for infinite cardinals: if A is infinite
then A == A+ A = A.A.
Here is how we do cardinal arithmetic.
2' $;

=
=

Address for Correspondence

S M Srivastava
Stat-Math Unit
Indian Statistical Institute
203 B T Road
Calcutta 700 035
e-mail: [email protected]

(since 2 ~ c)
(2 o)' (since c = 2l-tO)
2l-to.'
(since, for non-empty sets
X, Y, Z, (XY) Z == XYXZ)

"

2'"
(since No < c)
2'
(since C.c = c).
So, by the Schroder-Bernstein theorem 2' = ,'. From this we
conclude that there are as many 0-1 valued maps of real numbers
as real valued ones!
$;

Suggested Reading
K Kuratowski and A Mostowski. Set theory. North-Holland publishing company. Amsterdam, New York,
Oxford,1976.
Gregory Moore. Zermelo's axiom of choice -its origin, dewlopment and influertCS. Springer.. Verlag. New
York., 1982
Georg Cantor. Contributiom to thefo;unding of the theory of transfinite numbers. Dover Publications Inc.
New York, 1992.

Nicolas Bourbaki. Elements of the history of mathematics. Springer-Verlag. Berlin, Heidelberg, New
York,1994.
Yiannis N Moschovakis. Notes on set theory. UTM. Springer-Verlag. New York, Berlin, Heidelberg, 1994

--------~--------