96 DYNAMICAL SYSTEMS

CHAPTER 9 SYMBOLIC DYNAMICS 97

CHAPTER. 9

Symbolic Dynamics

—A
A=1 A=4
Fig. 8.11 The orbit diagram for the logistic function
using the initial seed zo = 0.123...,
18. In all of the orbit diagrams, there are some very obvious darker
“curves”
running through the orbit diagram. Explain what these curves are. Why
does the orbit of 0 seem to accumulate on these points? Hints: Compare
the histograms for various c-values to the slice of the orbit diagram for
the
corresponding c's. Compute the orbit diagram using only a few points on
the orbit of 0 without throwing away any of these points.
In this chapter we will introduce one of the most powerful tools for under-
standing the chaotic behavior of dynamical systems, symbolic dynamics. We
will convert the complicated behavior that we have seen for certain quadra-
tic functions to what appears at first to be a completely different dynamical
system. However, we will see that the two systems are really the same and
that, more importantly, we can understand the new system completely.
This chapter is more theoretical than the preceding chapters. We in-
troduce a new level of abstraction involving a *space” of sequences and a
mapping on this space that will later serve as a model for the quadratic
maps. We will see in the next section that this abstraction is totally justi-
fied when we show that iteration of our model mapping can be completely
understood. This is in contrast to the quadratic case which is impossible to
deal with analytically.

9.1 Itinerarijes

In this section we will deal exclusively with a quadratic function Qq(z) =

a? 4 ¢ where c < ~2. Recall that for such a function all of the interesting
dynamics occurred on the interval I = [—p+, p+]. Here py is the fixed point
of Q. given by
p+=%(1+\/1—40).
In this interval there is a subinterval A; that consists of all points that leave
T under one iteration of Q.. The set A of all points z € I whose orbits never
leave I therefore lies in [— Aj. The set I — Az consists of two closed intervals
 98 DYNAMICAL SYSTEMS CHAPTER 9 SYMBOLIC DYNAMICS 99

that we denote by Iy and I; with o to the left of I;. Remember that both I The set £ consists of all possible sequences of 0's and 1's. That is, ele-
and 71 actually depend on ¢, and for each ¢ < —2 these intervals are disjoint ments of T are sequences, not numbers. For example, (0000...), (010101...),
and symmetrically located about 0. (101010.. .), and (1111...) are all distinct elements of D.
Any point in A has an orbit that never leaves I and hence remains for We are going to try to work geometrically with the space D. That means
all iterations in Iy U I. Soifz € A, Q¥(z) € hUT for each n. This allows we have to know what this space looks like. At first, this may sound like an
us to make the following important definition: unattainable goal. After all, the elements of ¥ are sequences, not numbers
that fit conveniently on a line or points that live in the plane or three-
Definition. Let z € A. The itinerary of z is the infinite sequence of 0's and dimensional space. However, we will see that with a little metric space
1's given by theory, we can actually develop a mental image of what I looks like.
Let's begin rather naively. Suppose we try to identify ¥ as some subset
$(z) = (508182...)
of the plane. Who knows what ¥ would look like in the plane? Maybe a

T
where s; = 0if Q1(z) € To and s; = 1 if Q{(z) € H. triangle or a square or even a banana. Whatever our picture of ¥ is, each
“point” in T must be a sequence. That is, abstractly, we should think of
the entire string of digits composing a sequence in £ as corresponding-to one
For example, the fixed point py lies in I for all iterations, so p4 has
point in this picture. Perhaps T is the image sketched in Figure 9.2. If so,
the itinerary S(p4) = (1111...). Similarly, the eventually fixed point —p, then each point in this “space” must be a sequence of 0's and 1's, and two
has itinerary 5(—p+) = (01111...). Note that any periodic point has an distinct points should be two different sequences, as indicated in the figure.
itinerary that is a repeating sequence. A more general itinerary is shown in
Figure 9.1.

Iy A I,
(101010...) (010101...)
Fig. 9.1 The itinerary of zq is $(zo) = (001011...).

9.2 The Sequence Space

(U11..
To set up a model system for the dynamics of Q. on A, we need to have Fig. 9.2 Perhaps the space D is a Smiley Face.
a “space” on which the dynamical system takes place. Unlike all of the
previous systems we have discussed, this model system will not take place We should also be able to recognize certain subsets of D. For example,
on the real line. Rather, it will take place on a much more abstract space the set
called the sequence space.
My={sEL|50
=0}

Definition. The sequence space on two symbols is the set { consists of all sequences whose zeroth entry is 0. Its complement is

2= {(sos152...)|5, = 0 or 1}. Mi={sE£|50

=1).
s
 100 DYNAMICAL SYSTEMS CHAPTER 9 SYMBOLIC DYNAMICS 101
Roughly speaking, My and A divide B into equal halves, perhaps as de- We
picted in Figure 9.3. Similarly, distance between the same two points as displayed on a car's odometer.
choose to put the following metric on .

My ={s EZ|59 =0, 1 =1} .

Definition. Let s = (595152 ...) and t = (tgt1t2 ...) be two points in T. The
Mo ={s ED[0=0,
51 =0} distance between s and t is given by
divide My into equal halves as shown in Figure 9.4,
d =y=, [s; P— ti]
i=0

Li a complicated
Thisi may look k like i ion,
expression, but it is often easy to com-
pute. For example, let s = (000...), t = (111...), and u = (010101...).
Then
(101010....)

M, 1
(11...)
Fig. 9.3 Mo and M, each occupy halfof .

1 1 1
(010101...
dsul=g+mtEt:
1& 1)’_2
=32(0) =3
Note that the series defining dfs, t] always converges. I\ſſde_ed, $i 'und t;
Fig. 9.4 Mog and My, each occupy halfof My. are each cither 0 or 1, 80 |s; —t;| = 0 or 1. Therefore, this series is dommz;ted
by the geometric series 1-32(1/2)%, which converges to 2. Thus thle fa;t =
Of course, given what we know so far (just the definition apart any two points in D may be is 2 units, as in the first example above.
of %), we really
have no earthly way of knowing what this set and
its subsets look like. What
we need is some way of understanding when two points A function d is called a metric on a set X if for any z,y,2 € X
are close together Definition.
and when they are far apart. That is, we need a the following three properties hold: )
way to measure distances
between two elements or points in the set. Mathem
need a metric on the set. Now there are
atically, this means we 1. d[z,y] > 0, and d[z,y] = 0 if and only if z = y.
many ways to put a metric on a 2. d[z,y] = d[y,z].
set, just as in real life where there are many
ways to measure distances, such
as the straight line distance between two points (“as
the crow flies") or the
3. dz,z] < d[z,y] + dly,=]
The pair X,d is called a metric space.
FACULDADE D
 102 DYNAMICAL SYSTEMS CHAPTER 9 SYMBOLIC DYNAMICS 103
This last property, an important one, is called the triengle incquality.
Nohcſie chatſi all three of these properties are familiar properties of the
On the other hand, if 5; # t; for some j < n, then we must have
usual
metric or distance function in Euclidean space.
d&,t] > 551_1
2 00
Proposition. The distance d on & given by Consequently, if dfs, t] < 1/2", then s; = t; for i < n.

- i=0
liztl In words, two sequences are close if their first few entries agree. We can
guarantee that two sequences are within 1/2" of each other if their first n
entries are the same. l
13 @ metric on X. This proposition shows that our naive picture of % in Figures 9.2 to 9.4 is
wrong. For example, any point in Mo must be at least 1 unit away from any
point in M, since the initial entries of these sequences disagree. Similarly,
Proof: Let s = (sgs152...), t = (totit2...
2 ) and u = (u (upujuy...). Clearl any point in Mog must be at least 1/2 unit away from a point in Moy (see
d[s,t] > 0 and dfs,t) = 0 if and only if s = t. Since lsi —ti| = )t, — s ?'Ly
follows that dfs,t] = d{t, s]. Finally, for any three real numbers 5;,t;,u,, we Figure 9.5). Continuing in this fashion, we see that our picture of ¥ cannot
have the usual triangle inequality
contain any planar regions—any such region must eventually be subdivided
' into infinitely many disjoint pieces. This is, of course, reminiscent of the
totally disconnected Cantor set we encountered previously. We will make
lsi = til +[t; = wil 2 |si — wi] this analogy precise later.

from which we deduce that

ds,t] + d{t, u] > d[s, u.

This completes the proof.

- The reason we introduced the metric d was to decide when twi 0 Sequ

1
i‘?:
”
ences (010101...)
in ¥ were close together. The next theorem scttles this issno.

:
The Proximity Theorem. Let s,t € I and suppose s;
0,1,...,nn. Then d[s, t] < 1/2". Conversely, if; d[s,t] < 1/2",; = then
P t; for i
s; = t; Distance = 1

Fig. 9.5 The metric d cuts T into disjoint picces.

Proof. If s; = t; for i < n, then

d[s, t] 9.3 The Shift Map

Now that we have the space that will be the setting for our model dy-
namical system, it is time to introduce the model mapping itself. This is the
shift map on T.
 104 DYNAMICAL SYSTEMS
CHAPTER9 SYMBOLIC DYNAMICS 105
Definition. The sh;; map 0:
shift : — ¥ isi defined by
and the like. Nevertheless, like its more familiar counterparts, o is a continu-
o(s0s182...) = (818253.. ous function. To see why this is true, we cannot use the simplified definition
of continuity often given in calculus courses (“you can draw the graph with-
out lifting your pen”). Rather, we have to return to first principles and
invoke the theoretical definition of continuity.
That is, o simply drops the first entry of any point in D. For exampl
e
”(010101...) = (101010...) Definition. Suppose F: X — X is a function and X is a set equipped with
a metric d. Then F is continuous at z9 € X if, for any € > 0, there is a
o(101010...) = (010101...) § > 0 such that, if d{z,zo) < 6 then d[F(z), F(zg)] < e. We say that F is a
continuous function if F is continuous at all zg € X.
o(0111...)=(1111..,).
Note that it is easy to iterate o—we sim; To prove that a function is continuous at zg, we must therefore be able
ply continue dropping leading entri
as we iterate. That es to accomplish the following: given any € > 0 no matter how small, we must
is,
be able to produce a 6 > 0 such that, whenever z and zq are within 6 units
o7(598152.. ) =(s28384.. ) of each other, their images are closer together than e.
This definition of continuity is one of those extremely complicated def-
0"(999192..) = (3n8n+18n42..). initions that often drives students mad. So before proving that the shift
map is continuous at all points, we'll give a simple example as a warm-up.
This makes it easy to find the periodic
points of o, If s is a repeating Let's begin by showing that o is continuous at the fixed point (0000...).
sequence of the form
A To prove this, we assume that someone gives us an € > 0. Our job is to
find a 6 that *works.” By this we mean that, if s is a sequence in T with
s:(5531,..57._13051.,.5,._1...) d[s, (0000...)] < &, then d{o(s), (0000...)] < e. Now we don't know which
= 3051 - .- 5a21),
sequences in T have images that are within an arbitrary € of (0000...),
then o™(s)
7 = s, 5 Conversely, : an any perioiodic poi of perioi d n for o must be
dic point but we do know those that are within 1/2" of (0000...) for each n. And
repeating sequence. For example,
the only fixed points for o are that will be good enough, as long as we choose n so that 1/2" < e. So
and (1111 ..). The period 2 points are (010101...) and (01;00 °3
of course, ¢ maps one to the other (101010.. ), and we must find 6 that guarantees that d[o(s95152...), (0000...)] < 1/2", But
. There are two 3-cycles for o given that's easy! By the Proximity Theorem, the only way this can be true is
by
if s; = 0 for i = 0,1,2,...,n+1. Therefore we choose 6 = 1/2"+!, If
(00T) — (D10) » (T00) (00T) d{{s98152
...), (0000...)] < 6, then

(T10) - (TOT) - (0TT) — (T0). — d[o(595152 .-.), (0000...)] = d[(s15253...),(0000...)] < 1/2" < e.

. > ay clearly writ,. Thus we have found a 6 that works.

down all cycles of period n for & Just by writing any block of n 0's and ſilſſvs ;:]; The proof in the general case is not much more difficult.

this is just one of the reasons why o is a
The shift map is clearly much different
tered in calculus uch 25 pelymomich m from othe 1
usu al functions encoun-
ctions, exponentials,
Theorem. The function 0: ¥ — T is continuous at all points in T.
— © Proof: Suppose we are given ¢ > 0 and s = (505152...). We will show that
: o is continuous at s.
 106 DYNAMICAL SYSTEMS CHAPTER 9 SYMBOLIC DYNAMICS 107

Since € > 0, we may pick n such that 1/2" < e, We then choose § = We will conclude this chapter with a proof of this important fact, but
1/2”, Iftis a point in £ and d[t,s] < 6, then by the Proximity Theorem we before this we turn to another property of S.
must have s; = t; for i = 0,1,...n+1. That is, t = (5 ... $n+1tn+2tn+3 .. ).
Now o(t) = (51 ... 8n+1tn+2tn+3 - .) has entries that agree with those of SoQc(z)=00 S(z).
Theorem. Ifz € A, then
o(s) in the first n + 1 spots*. Thus, again by the Proximity Theorem,

This result is clear: if z € A has itinerary (sos152...), then by definition

e
d[o(s), o(t)] <1/2” < e.
z €Iy
Therefore, o is continuous at s. Since s was arbitrary, o is a continuous

S
function. This completes the proof. Qi(z) E I, m
and so forth. Here, Iy; is either,Tq or I, depending upon the digit s;. Thus
Qc(z) €I,
9.4 Conjugacy
Qi(e) € Iy
Qe(s) € Ly, |

S A S
Now that we have our model system, the shift map o on the sequence
50 that S(Qc(z)) = (515253 ...) which is o(S(z)) as claimed.
space L, it is time to relate this dynamical system to Q. on A. We already
have a function that relates the two, namely the itinerary function S. Recall We may describe this result pictorially as a commutative diagram:
that, given a point z € A, the itinerary of z is a sequence S(z) in E. Hence

M
we have a mapping 5: A — L. a &®
Our first observation is that A and I, though defined in very differ-
ent manners, are essentially identical spaces. The way to show that two 5| |s

i
spaces are identical is to exhibit a homeomorphism between them. A home-
omorphism is a one-to-one, onto, and continuous function with continuous 7
inverse. If we have a homeomorphism between two sets, these scts are said
to be homeomorphic. From a qualitative point of view, two sets that are The theorem tells us that if we start with any z € A. in the upper l?ft-hf\nd
homeomorphic are essentially the same. Not only may we put the points
of
corner of this diagram and then follow the arrows in either possible direction,
one set into one-to-one correspondence with those of the other set, but, we always end up at the same point in T in the lower right-hand corner.
more
importantly, we may do this in a continuous fashion, taking nearby
points Since S is a homeomorphism, we also have
in one set to nearby points in the other.
Homeomorphism allows for sets to be distorted, as, for example, two Q.05 =8loSogeos”!
finite
closed intervals are homeomorphic, even though their lengths differ. But
an
=SlogooSos”
open interval and a closed interval are never homcomorphic. Rather than =SToo.
interrupt the discussion of symbolic dynamics with a lengthy description
of
the properties of homeomorphisms at this point, we refer you to Appendix Thus, this diagram commutes:
A.1 for a more complete discussion of this topic.
a E A
T.
Theorem. (A and T are the same!) Suppose ¢ < —(5 + 2V/5)/4. Then
S:A — T is o homeomorphism.
* Technically, these sequences agree in the zeroth through nth spots = L

aj
 108 DYNAMICAL SYSTEMS CHAPTER 9 SYMBOLIC DYNAMICS 109

Stephen Smale z
ition. — X and G:Y
Let F:X q — Y be twoc functions. We say that
Be:nſ:il(; are conjugate if there is a homeomorphism h: X — Y such that
Aſter winning the Fields Medal in 1966 for resolving the a conjugacy.
Poincaré Conjecture, Stephen Smale (1930- ) turned his hoF =Goh. The map h is called
attention to dynamical systems. In his famous *horse- 3
shoe” example, Smale showed that very simple dynamical The previous theorems therefore give:
systems could behave quite chaotically. Moreover, he in-
troduced the use of symbolic dynamics to help understand The Conjugacy Theorem. The shift map on ¥ is conjugate to Qc on A
this chaotic behavior. The quadratic map in this chapter
is a simplified version of the horseshoe example. Smale
when ¢ < —(5+2v3)/4.
is currently Professor of Mathematics at the University of < ¢ < —2, but the
California, Berkeley. We remark that this result is valid for —(5+ 2V/5)/4
)
in thisthi range are more di ifficult.
ents needed in
|
. To prcctw_e1 this
argul\?;w let’s finish the proof that A and ¥ are homeomorphic
S and 57 are
We also have we need to show that S is one-to-one and onto and that both
SoQt=0050Q7t continuous.
=o 0 So?q7=? Assume,. hcſi-wever, that
-to- : Suppose that z,y € A with = # y.
cl}:zes;v-:l:
?{3 0 05?;) This means that Q!(s) and Qe(y) alwa)}'ls lfſſ:hſiſieseSider
1\? 'ls
=0"085. subinterval Iy or I;. We know that Q. is one-to-one on eac]
Now Foſi;ſioſi the
This is an important fact. and that |Q{(z)] > p > 1 forall z € LUl and some .
It says that S converts orbits of Q. to orbits of in one-tcz-c* \\-;e1 as vion onto
o. Pictorially, we have interval [z,y]. For each n, Q7 takes this interval
[Q*(z), Q3(y)]. But, as we saw in Chapter 7, the Mean Value
a % , %%, impli at
s length [Q*(z), Q3(y) 2 " length [z,y].
15
Since y" — 0o, we have a contradiction unless # = y.
Sj ls

T + % 5 o5 T Onto: We first introduce the following notation. Let J C I be a closed

Similarly, we have interval. Let
QtoS'=5"00" Q7%(J)={zETIQ!(2€ )T}
80 51 converts o-orbits to Qc-orbits. Therefore, Qc-orbits are in one-to-one In i
particula -
r, Q7*(J) i
denotes the preimag e off J. The ? main observation
correspondence with o-orbits. In particular, if s is a periodic point for o, 7 is a closed interval, then Q7!() consists of two clozeſid
then S-1(s) is a periodic point for Q. with the same period. If we have subintevrvals, one in Iy and one in T1. (see Figure 9.6; also consult Append
ix
an eventually periodic point for o, then 5”! gives us an analogous point A3).
for Qc. Indeed, S-? converts any dynamical behavior we observe for o to To find z € A with S(z) = s, we define
corresponding behavior for Qc. Thus the dynamics of & on T and Qcon
A
are essentially the same. As we will see in the next chapter, the dynamical Lipspousn = (2 €115 € Lip, Qc(@) € Iy, - Q2 (2) € L}
behavior of & is chaotic. Therefore the same is true for Qc on A. depending on
Homeomorphisms like S play an important role in the study of dynamical Since 5; = 0 or 1 for each j, the set I, is really one of Iy or I;
5ystems, since they show that two apparently different systems may the digit s;. Using the notation above we may write
actually
be dynamically equivalent. These maps are called conjugacies. Tgsroan = D N Q7 D) N N QT (L)
 110 DYNAMICAL SYSTEMS
CHAPTER 9 SYMBOLIC DYNAMICS 111

Incidentally, note that Nayolsgs,..s, consists of a unique point. This

follows immediately from the fact that S is one-to-one. In particular, we
have that diam T,gg1...z — 0 as n — oco.

Continuity: To prove that S is continuous, we again invoke the theoretical

definition of continuity given above. Let z € A and suppose that S(z) =
sos152... - We will show that $ is continuous at z. Let e > 0. Then pick
n 50 that 1/2" < e. Consider the closed subintervals Iy, ., defined above
for all possible combinations tot1 ...tz. These subintervals are all disjoint,
and A is contained in their union. There are 2**! such subintervals, and
Iygsy...sn is one of them. Hence we may choose 6 such that |z — y| < § and
y € A implies that y € I,,,..s,. To do this we simply choose 6 so small that
Fig. 9.6 The preimage of a closed interval J is a the interval of length 26 centered at z overlaps none of the Iy,. 4, with the
pair of closed intervals, one in I and one in 11 exception of Iys,..s,. Therefore, S(y) agrees with S(z) in the first n + 1
For any pair of intervals A and B, we have terms. Hence, by the Proximity Theorem, we have

QAN D) = Q7' (4)nQ;\(B). dS(z),S(y) < 55 < e

i
B
Thus we may also write
This proves the continuity of S. It is easy to check that 5~ is also continuous.
Thus, S is a homeomorphism.
Tenwmn = Lo N 07 (TN N Q7 (~1)(1,,)

R
=1y N Q7 (Iy..00)- Exercises
We claim that the I, 5051...99 Are closed intervals that are nested.
Clearly I,,
is a closed interval. By induction we assume that 7 51...3, 18 @ closed 1. List all cycles of prime period 4 for the shift map.
interval.
Then, by the observation above, QP(I,,...s,) consists of
a pair of intervals, Compute d[s, t] where:
T C
one in o and one in 11. In either event, L,nQ7Y( Lotoon) = Trgn.c.am is 2
single closed interval. 2. s = (100), t = (001).
These intervals are nested because
ety
A S
3. s = (100), t = (010).
Tog.wmm = Trowanms N QGE(Ln) C Lugo,, - 4. s = (1011), t = (0101).
Therefore we conclude that 5. Find all points in T whose distance from (000...) is exactly 1/2.
M Loosion
n>0 6. Give an-example of a sequence midway between (000...) and (111...).
is nonempty since a nested intersection o f closed intervals is always nonemp Give a second such example. Are there any other such points? Why or why
See Appendix A.3 for a discussion of this fact. Note that if z
ty. not?
then z € I, Qe(z) € L,,, and so forth. Hence S(z) € Nn>ologetic.ons
= (5051 . ..). This proves 7. Let Moy = {s € S|so = 0,8, = 1} and My, = (s € T|so = 1,8, =
that S is onto.
0,52 = 1}. What is the minimum distance between a point in My, and a
 112 DYNAMICAL SYSTEMS
CHAPTER 9 SYMBOLIC DYNAMICS 113
point in M1q1? Give an example of two sequences that are
other.
this close to each t i
was discussed i the exercises
in i at the end of Chapter 7. . There e we proved
4
:Z:t? ={z € [0,1]|7*(z) € [0,1] for all n} was the Cantor middle-thirds
8. What is the maximum distance between a point set. Now define an itinerary function S:T' — £, where I is the space of
in My, and a point in
Mia? Give an example of two sequences that are
this far apart.
sequences of 0's and 1's. Prove that S is a homeomorphism.
Remarks:
9. Let S be the itinerary map as defined in this
chapter. Prove that ifs € &
is periodic under the shift map, then S=(s) 1. The above exercise shows that the set A for the quadratic map and
is a periodic point for Q. in A
with the same period. What happens if s is eventu the Cantor middle-thirds set are actually homeomorphic, for they are both
ally periodic?

i Gl
homeomorphic to T.
The N-Shift:
2. We now have two infinite sequences attached to each poinſit z in the Cantor
The following seven exercises deal with the analog of the shift map and set. One is the ternary expansion of z and the other is the ltmerary'of z. It
sequence space for sequences that have more
than two possible entries, the is tempting to seek a relationship between these two sequences. While there

s
space of sequences of N symbols.
is a relationship between them, it is by no means obvious. For example, the
10. Let Ty denote the space of sequences point 1 has ternary expansion 0.222... but §(1) = (1000...).
whose entries are the
positive
integers 0,1,...,N — 1, and let oy be the shift
map on Ly. For s,t € Sy,
let 18. Each of the following defines a function on the space of sequences L. In
oo
Isi —ti| each case, decide if the given function is continuous. If so, prove it. If not,
dyfs,t] = ) = explain why.
= N a. F(sos1s2...) = (Osps1s2...)
Prove that dy is a metric on Zy.
b. G(505182 -..) = (0590510s2 ...)
11. What is the maximal distance between two
sequences in Ly ? c. H(5g5192- (515053528554...) ) l )
12. How many fixed points does on have?
How many 2-cycles?
d. J(sos1s2...) = (303132...) where §; = 1if 5; = 0 and 3; = 0 if
cycles of prime period 27 How many s;=1
13. How many points in Dy e. K(sos1s3...) = ((1=s0)(1 —51)(1 = s2)...)
are fixed by o}ſi{,?
f. L(sosys2 (s0s25456...) )
14. Prove that oy: £y — Sy is continuous. g M(sos1s2...) = (5151051005100
- --
15. Now define —_— h. N(sos1sz...) = (totitz...) where t; = 89 +51 +- - +s; mod 2. That
s,t) is, t5 = 0if 50 +-+-+5 is even, and & = 1if 80 +-- +3; is odd.
&[s,t] = 3 A2 i. P(sgs153...) = (totits...) where t; = lima—o 5n if this limit exists.
= N*
where 6;(s,t)= 0if 54 = #; and 8k(s,t) Otherwise, t; = s;.
= 1if s # t. Prove that is also
a metric on Ty. 19. Define a different distance function d on 3 by [s,t] = 1/(k + 1) where
16. What is the maximum distance between
k is the least index for which s # ¢ and ds,s] = 0. Is d a metric?
two points in Ly when the
metric dg is used?
17. Recall the function

T(z) {
3z ifa<1/2
3-3z ifa>1/2