You are on page 1of 9

The Cantor Set And Integration of Opposite Values

Last time, Dr. Wu had discussed how the set of 0-1 sequences represented in a way the “all
possibilities” characteristic of the Infinite. He said that the set of all such sequences corresponded
to a special set of real numbers, the Cantor set. In our last meeting, Dr. Wu gave us a challenge:
to build a set K having an odd combination of properties. Xinyuan and I both suspected that if
we could create such a set K, we would have on our hands the Cantor set, but we had not been
successful so far.1 We were both eager to see a solution.
”Last time,” Dr. Wu began after he arrived, “you will remember that we talked about the
‘all possibilities’ quality of the Infinite. Contacting the Infinite in meditation, for instance, you
may experience being bathed in all possibiities — that creativity and problem-solving abilities are
greatly enhanced by just dipping into this field of all possibilities even for a few minutes.
“Another expression of ‘all possibilities’ that some people notice is a spontaneous ability to
embody opposite ends of certain personality traits — when this happens, it can be quite a surprise!
Details differ for different people. But there are some striking examples that most of us have
encountered at one time or another. For instance, a parent who has the ability to be both strict
and fun-loving with his or her children. Strictness would seem to preclude being fun-loving, and
conversely; yet, you will find parents that somehow embody both qualities with their kids. As
another example, as a mathematician I have known a number of extraordinary researchers who
simultaneously embody extreme precision of speech and a high degree of fluidity and charm. Again,
one usually finds these qualities in different people; when they show up at the same time in a single
individual, it is a kind of inspiration. You can’t help thinking to yourself that here, Life has found
a way to create a kind of wonder by putting these apparently mutually exclusive characteristics
together.
“Of course such people may have nothing to do with meditation. The fact is that the Infinite
is the fundamental level of anybody’s awareness, whether one happens to experience it directly or
not. And one’s life will reflect the qualities of this field to whatever extent the influence of this
field is able to show up. The nice thing about practicing some form a meditation, as we explore the
Infinite, is that you can see for yourself how these qualities really do become enlivened by contact
with the Infinite.
“This trait of the Infinite to unify opposites is expressed in the Tao Teh Ching like this2 :”

Yield and overcome.


1
Actually, I found out later that Xinyuan had come up with a construction but hadn’t finished
verifying that it satisfied the three properties Dr. Wu had mentioned. After class, she found that
her construction did work.

2
Verse 22.

1
Bend and be straight.
Empty and be full.
Wear out and be new.
Have little and gain.
Have much and be confused.

“The verse shows the natural relationship between opposite values. Each depends on the other
and the Tao is the glue that binds them. There is great practical wisdom in this verse. You may
have noticed, for instance, that to overcome an obstacle, there are times one seems to be making
progress, and times when one needs to ease up and temporarily allow things to simply be as they
are; from this perspective of acceptance and stillness, new avenues for further progress appear. One
can try to make use of this wisdom, of the connectedness of overcoming and yielding, but for the
full effect, the text advises the aspirant to dip into the Tao and bathe in its nature; the result is
that the ability to flow between the poles of opposites is thereby greatly enhanced, because this is
the characteristic of the Tao itself. On this point, the text says (Verse 23),”

Open to the Tao, and then allow action to flow.


In this way, everything falls into place.

“This property of integration of opposite values shows up in all kinds of ways in mathematics,
often in contexts that do not appear to have anything to do with infinite sets. Today, I want to
focus on a certain aspect of this quality that is particularly mentioned in the Vedic literature in
rather paradoxical language. They say that wholeness is bigger than the biggest and smaller than
the smallest 3 .
He wrote on the board:

an
. oran
. ı̄yān mahatomahı̄yān
“The significance, as we will discuss further in future lessons, is that the ultimate Infinite is
not only to be found as the veritable container of all that exists (‘bigger than the biggest’), but
also at every point in the universe (‘smaller than the smallest’); again, wholeness is being declared
as both ‘full,’ being vast, and ‘empty,’ having its reality at each point.
“In the context of mathematical infinity, what I want to show you today is how this capacity of
wholeness to be enormous and infinitesimal all at once finds expression in the existence of certain
unusual subsets of R, which simultaneously exhibit properties of being very big and extremely
small. Again, we will see that it is because of the richness of the uncountable infinity represented
by R that such an example could exist at all.
3
Kat.ha Upanishad I.2.20

2
The Cantor Set

“The set I have in mind is Cantor’s middle-thirds set, which I mentioned briefly at the end of
our last lesson. Were either of you able to build your own Cantor set with the hints I have last
time?”
We both shook our heads.
“Don’t worry. It’s a tricky construction. C will embody the principle of integration of opposite
values in a striking way—in a way that we don’t see at the level of countable infinities, but which
seems to emerge when we start looking at R and its uncountable subsets.
“The construction goes like this. Start with the closed interval [0, 1]. We are going to start
removing certain subsets from [0, 1]. The set C will be what is left over after we are done removing
the subsets we have in mind. Step 1 consists of removing the middle third open interval from [0, 1].
In other words, remove the open interval (1/3, 2/3). You are left with two closed intervals [0, 1/3]
and [2/3, 1], as in the diagram.”

“Step 2 consists of removing the middle third interval from each of the two remaining closed
intervals. From the first interval, we will remove (1/9, 2/9) and from the second interval, we will
remove (7/9, 8/9). The result is that we are left with four closed intervals: [0, 1/9], [2/9, 1/3],
[2/3, 7/9], [8/9, 1].

“Now, Step 3, as you would expect, will have us remove the middle third intervals from each
of these four remaining closed intervals. This step will leave us with eight closed intervals.
“In general, Step n + 1 will involve removing middle third intervals from each of the 2n closed
intervals that remain after Step n; the result will be 2n+1 closed intervals.
“The set C is what is left over after carrying out Step n for every natural number n; that is,
C is what is left over after repeating this process infinitely many times. Another way to say this
is that C consists of all points p for which, for every Step n, p lies in one of the remaining closed
intervals after middle-third intervals have been removed.

3
“A reasonable question at this point is, ‘Is there anything left at all?’ Is C just the empty
set?”
I could see that, even though an awfully big chunk of [0, 1] was being removed in all these
Steps, still certain points certainly survived through it all, like the endpoints 0 and 1. Likewise, the
points 1/3 and 2/3 didn’t seem to lie in any of the middle-third segments being removed. In fact,
as I thought about it some more, it looked like the endpoints of the middle-third open intervals
that were being removed were never themselves taken out. I told Dr. Wu that it looked like C had
to be infinite.
“Yes, that’s right, the endpoints of the middle-third intervals all end up in C,” he said, “and
there certainly are infinitely many of them—in fact, it is not hard to arrange these endpoints in a
sequence...”
“So there is only a countable infinity of them,” I interrupted.
“Right. And this leads to the next question: Is C countably infinite or uncountable? If it’s
uncountable, then where are all those extra points to be found, the ones that are not endpoints of
middle-third intervals?”
From what Dr. Wu had said earlier, I knew that C was supposed to be uncountable, but I
certainly didn’t see why.
“Let me heighten the suspense a little by emphasizing how ‘small’ C has to be. I want to
give you some idea of how much ‘substance’ of [0, 1] was removed in the creation of C. Do you
remember what the length of the interval [0, 1] is?”
“Sure,” I said, “it has length 1.”
“Right. Now what I’d like to do is have you guess what the sum of the lengths of all the
middle-thirds intervals might be.”
I had no idea. The middle-third interval at Step 1 was just (1/3, 2/3), so it has length 1/3.
At Step 2, there are two middle thirds intervals, each of length 1/9, so the total length removed at
Step 2 is 2/9. I was starting to see a pattern. Each middle third interval at Step 3 had length 1/27

4
and there were 4 of them, so the sum of the lengths of the middle-third intervals in Step 3 had to
be 4/27. I told Dr. Wu my observation.

The “Smallness” Of The Cantor Set

“Good, you have seen a pattern,” Dr. Wu began. “I would like to invoke a little formula that
can be used for adding up an infinite series like the one you have discovered. I am going to do some
maneuvers that may seem a bit like magic, but I will also give you an exercise that will clarify what
I’m doing and help you to understand the steps better.
“The formula I want to invoke is this: Suppose r is a real number between 0 and 1, and we
wish to figure out the sum of these numbers:

r + r2 + r3 + r4 + . . . + rn + . . .

“For example, r might be the number 2/3; we will be able to use our formula, therefore, to compute
the sum
2 2 2 2 3 2 n
+ + +...+ +...
3 3 3 3
“Now, I’ll state the formula as a theorem.”

He wrote:

Power Series Formula Suppose r is a real number between 0 and 1, exclusive. Then the sum

r + r2 + r3 + r4 + . . . + rn + . . .

is equal to
1
− 1.
1−r
“I realize that at this stage, this formula may raise a lot of questions. How could it be, after all,
that we can add up infinitely many numbers together and expect to get a number at all? As I recall,
before we ever began this Directed Reading, I gave examples in class of infinite series that clearly
did not add up to anything. For instance, the series 1 − 1 + 1 − 1 + 1 . . ., as you might remember,
produces a paradoxical answer when you try to add up the terms — if you insert parentheses like
this: (1 − 1) + (1 − 1) + (1 − 1) + . . ., you get a sum of 0, but if you insert parentheses like this
1 + (−1 + 1) + (−1 + 1) + (−1 + 1) + . . ., you get a sum of 1. Another example I gave was the series
1 + 2 + 3 + 4 + 5 + . . ., which couldn’t add up to any number since, for any natural number n that
you might think would be the sum, n itself is actually one of the terms of the series.
“I also mentioned during this lecture, however, that certain infinite series do have a legitimate
sum, and there are some well-established criteria for deciding whether or not this will happen. As
I say, I will give you a few exercises to acquaint you with some of these ideas. One consequence of
the theorems in this field is that ‘power series’ in the form I have given them — r + r2 + r3 + r4 +

5
. . . + rn + . . . , 0 < r < 1 — always do have a sum, and the Power Series Formula tells you exactly
what the sum is.
“Now try out the formula on the series”

2 2 2 2 3 2 n
+ + +...+ +...
3 3 3 3

I plugged 2/3 in for r in the formula and got

1
− 1 = 2.
1 − 2/3

“Good,” he said after I told him my answer. “Now let’s try adding up the lengths of the middle
third segments that were removed in the construction of C. To see the pattern, let’s write out our
series involving 2/3 in more detail:

2 4 8 16
+ + + + ...
3 9 27 81

“Compare that with the sum we get when we add the lengths of the intervals removed at Step 1,
then Step 2, then Step 3, etc. in the construction of C. As you observed, we get:”

1 2 4 8
+ + + + ...
3 9 27 81

“Notice how this series is closely related to the series that begins with 2/3?”
I did notice. In fact, if you multiply every term in the second series by 2, you get this first
series that begins with 2/3.
“Yes, you do,” he said. “Or, another way of saying it is that if you divide each term in the
first series by 2, you get the second series. This means that if you divide the final sum of the first
series by 2, you will get the sum of the second series. In other words, since the sum of the series
beginning with 2/3 is 2, the sum of the second series is 1:

1 2 4 8
+ + + + . . . = 1.
3 9 27 81

“Now let’s think about what this computation tells us. It says that if we add up the lengths of all
the intervals that were removed in the construction of C, we get 1, which is the length of [0, 1] itself!
In other words, we have completely removed ‘all the length’ of [0, 1] in creating C. In mathematical
language, one says that C has measure zero—which means that there is nothing left in terms of
length.4

The “Bigness” of the Cantor Set


4
See the Appendix for a brief introduction to the concept of measure zero sets.

6
“However, as you say, we are about to establish that C is actually uncountable, and in fact,
structurally very much like [0, 1] itself. Anyone who first hears this finds this at least a little bit
amazing. In terms of length, C would seem to be a closer relative to the empty set than to [0, 1].
But in terms of size and structure, C is very much like [0, 1]. C is therefore a striking blend of
‘opposite’ values—tiny in terms of measure but big in terms of cardinality.
“Well, let’s verify that C really is uncountable. We will show that there is a very natural 1-1
correspondence between the set 2N of 0-1 sequences and C.
“Given a 0-1 sequence s1 , s2 , s3 , . . ., I need to tell you a point in C that corresponds to it.
There is a compelling visual way to do this. To give you the idea of how it works, let’s start with
a specific example of a 0-1 sequence and figure out which number in C it corresponds to. To make
things easy, let’s use the following sequence:

0011111 . . ..

“The idea is that after the first two 0’s, the sequence consists of all 1’s. I am going to tell you
which number in C this sequence corresponds to. I want to treat the sequence as a code for picking
a certain closed interval at each stage of the construction.
“The digit that occurs first in the sequence—in this example, 0— will tell me how to pick a
closed interval that was obtained at the end of Step 1. At the end of Step 1, there are two closed
intervals. The one on the left, [0, 1/3], I will think of as being specified by the digit 0, and the one
on the right, [2/3, 1], by the digit 1. In this example, the first digit of my sequence is 0, so this tells
me to pick the left interval, [0, 1/3].
“Now notice that in Step 2, each of the closed intervals in Step 1 was, in effect, divided in
two. For our example, we will be interested in how the left interval [0, 1/3] was divided in two:
After removing the middle third, we are left with [0, 1/9] and [2/9, 1/3]. I again wish to pick one
of these closed intervals, and I will decide which one by looking at the second digit s2 of my given
sequence. In this example, the second digit is again 0, so we again must choose the left interval,
namely [0, 1/9].
“The third term of my sequence in this example is 1; this tells me to pick the right interval
after [0, 1/9] has been divided in two in Step 3. The right interval in this case is [2/27, 1/9].
“We keep obtaining narrower and narrower closed intervals; the nth term of the given sequence
will tell me which closed interval to pick after my current interval has been divided in two in Step
n.

7
“By the First Completeness Property Theorem, my selected closed intervals telescope down to
a single point p. This is necessarily a point of C, and will be the point in C that corresponds to
the sequence s1 , s2 , s3 , . . .. In this particular example, if you think a little, you will see that p must
be the number 1/9.”
I thought I saw why this was true. The sequence we are using in the example consists of all
1’s after the first two 0’s. This tells me that I will always pick the interval on the right as we
further subdivide [2/27, 1/9]. This has the effect of scooting the left endpoint of the intervals that
we create farther to the right, but never moving the endpoint 1/9 at all. The end result then is
that p = 1/9.
“We therefore have established a natural 1-1 correspondence between 2N and C, obtained by
thinking of each 0-1 sequence as coding a sequence of telescoping closed intervals, one from each
Step in the construction, and in this way specifying a particular point in C. We have proven the
following:”

He wrote:

Size Of Cantor Set Theorem. |C| = |2N |.

“One can actually prove more—that C and 2N are actually homeomorphic as spaces; this
means that they are structurally identical, the only difference being in the notational differences in
our way of representing the elements of these spaces.
“Now, if we combine the Size Of Cantor Set Theorem with the Cardinality Of 2N Theorem,
we conclude that the Cantor set C has the same size as R. But, as I mentioned earlier, one can
also prove that C is structurally very similar to R and [0, 1], though it is not quite true that they
are homeomorphic.

8
“We have established that C exhibits these strikingly contrasting qualities of ‘being large’
(having the same size and structure as R) and ‘being small’ (having measure 0), at the same time.
It is, as I have said, the combinatorial richness of R, made possible by its enormous size, that allows
such a combination of properties to reside in a single set. This is the example of the integration
of opposite values—and in particular, of the property of being ‘bigger than the biggest and smaller
than the smallest’—that I was wanting to show you.

Other Sets Of Reals That Are Both “Small” And “Large”

“Researchers in this century have had, understandably, a certain fascination with sets of reals,
like C, that are simultaneously large and small, with respect to different measures of size. By
now, there are quite a number of other ways to measure the bigness or smalleness of a set of reals,
besides the concept of cardinality and the concept of measure. Without going into the definitions
of these terms, let me just mention two other ways, among the many that are known, that a set
of reals can be considered ‘small’: A set could be considered ‘small’ if it is first category or has
Marczewski measure zero. Again, I don’t want to define these terms, but let me just tell you one
result. It turns out that the Cantor set C is also first category, but does not have Marczewski
measure zero. But it is known that there is a subset of C which also has the same cardinality as
R, but which, simultaneously, is first category and has measure zero and Marczewski measure zero.
This particular set illustrates the property of being ‘big and small at the same time’ perhaps even
more than the Cantor set.”

Appendix: A Brief Introduction to Measure Zero Sets.

A set A of reals has measure zero if for any positive real number r (no matter how small), one
can find a sequence of open intervals I1 , I2, I3 , . . . such that each point of A is in one of the intervals
and the sum of the lengths of the intervals is less than r.

Exercise. Which of the following sets has measure zero? Prove your answers.
(1) Any finite set of reals (like {1, 2, 8}, for example).
(2) N
(3) Q (Hint: List the rationals q1 , q2 , q3, . . .. For a given r > 0, let I1 be an open interval of length
< r/2 that contains q1 ; let I2 be an open interval of length < r/4 that contains q2 ; and so
forth.)
(4) [0, 1]
(5) C, the Cantor set.

You might also like