SURREAL

NUMBERS

D. E. KNUTH
 This book is to be returned on or before
the last date stamped below.

27. $/tt

H.HQV1994

62577
512.7 KNU
A/C 063577

30107 000 532 710

 how two ex-students turned on to pure mathematics and found total happiness

a mathematical novelette by D. E. Knuth

SURREAL
NUMBERS

ADDISON-WESLEY PUBLISHING COMPANY

Reading, Massachusetts •
Menlo Park, California London • Amsterdam • Don Mills, Ontario •
Sydney
 CONTENTS
1 The Rock 1

2 Symbols 8

3 Proofs 14

4 Bad Numbers 19

5 Progress 27
This book has been set in Modern Extended jfl with Albert us chapter
headings. The cover and illustrations were designed by Jill C. Knuth. 6 The Third Day 34

f.O. 7 Discovery 41
63577
8 Addition 48
C- ASS
r/a-? km 7 The Answer 55

10 Theorems 63
</. 23 JAM
N i^A. t« isjH {
1 1 The Proposal 71

/ 12 Disaster 76

13 Recovery 82
Copyright © 1974 by Addison -Wesley Publishing Company, Inc.
Philippines copyright 1974 by Addison-Wesley Publishing Company, Inc. 14 The Universe 89

All rights reserved. No part of this publication may be reproduced, stored 15 Infinity 97
in a retrieval system, or ransmitted,
I in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise, wit houl he i

prior written permission of the publisher. Printed in the United States of 16 Multiplication 106
America. Published simultaneously in Canada. Library of Congress Catalog
Card Xo. 74-5998.
v / Postscript 113
ISBN 0-201-03812-9
BCDEFGHIJ-AL-79876S
1
 THE ROCK

•
""""-"•**-*.

W*v^.,

A. Bill, do you think you've found yourself?

B. What?
A. I —
mean here we are on the edge of the Indian Ocean, miles
away from civilization. It's been months since wo ran off to
avoid getting swept up in the system, and "to find ourselves."
I'm just wondering if you think we've done it.
B. Actually, Alice. I 've been thinking about the same thing. These B. Hey, what's that big black rock half-buried in the sand over
past months together have been reallygreat—we're com- there ?

pletely free, we know each other, and we feel like real people A. Search me, I've never seen anything like it before. Look,
again instead of like machines. But lately I'm afraid [*ve bean some kind of
it's got graffiti on the back.
missing some ofthfl things we've "escaped" from. You know.
Can you help me dig it out ? It looks like a museum
a hook to read —any
B. Let's see.
I've got this fantastic craving for I nick.
piece. Unnh! Heavy, too. The carving might be some old
even a textbook, even a math textbook. It sounds crazy, hut
Arabian script. . .no, wait, I think it's maybe Hebrew; let's
I've been lying here wishing I had a crossword puzzle to work
turn it around this way.
on.
A. Hebrew ! Are you sure ?
A. Oh, c'mon, not a crossword puzzle; that's what your parents
B. Well. I learned a lot of Hebrew when I was younger, and I
like to do. But I know what you mean, we need some mental .

can almost read this. . ..

stimulation. It's kinda like the end of summer vacations when
we were kids. In .May every year we couldn't wait to get out A. I much archaeological digging around
heard there hasn't been

of school, and the days simply dragged on until vacation

these parts.Maybe we've found another Rosetta Stone or
something. What does it say, can you make anything out ?
started, but by September we were real glad to be back in the
classroom. B. Wait a minute, gimme a chance Up here at the top right
iswhere it starts, something like "Li the beginning everything
B. Of course, with a loaf of bread, a jug of wine, and thou beside
was void, and. .." .

me, these days aren't exactly "dragging on." But I think

maybe the most important thing I've learned on this hip is
A. Far out! That sounds like the first book of Moses, in the

that the simple, romantic life isn't enough for me. I need Bible. Wasn't he wandering around Arabia for forty years
with his followers before going up to Israel? You don't
something complicated to think about.
suppose ....
A. Well, I'm sorry I'm not complicated enough for you. Why
B. No, no, it goes on much different from the traditional account.
don't we gel up and explore some more of the beach? Maybe
I^et's lug this thing back to our camp, I think I can work
We'll find some pebbles or something that we can use to make
out a translation.
up some kind of a game.
A. Bill, this is wild, just what you needed!
B. (sitting up) Yeah, that's a good idea. But first I think I'll take
a little swim. B. Yeah, I did say I was dying for something to read, didn't I.

Although this wasn't exactly what had mind!

A. (running toward the water) Me, too — bet you can't catch me I

hardly wait to get a good look at it

I in

—some of the things are

I can

kinda strange, and I can't figure out whether it's a story or

what. There's something about numbers, and. . ..

5
A. It seems to be broken off at the bottom ; the stone was origin- and said that it shall be a sign to separate positive num-
ally longer. bers from negative numbers. Conway proved that zero
was less than or equal to zero, and he saw that it was
B. A good thing, or we'd never be able to carry it. Of course
good. And the evening and the morning were the day of
it'll be just our luck to find out that the message is getting
zero. On the next day, two more numbers were created,
interesting, right when we come to the broken place.
one with zero as its left set and one with zero as its right
A. Here we are. I'll go pick some dates and fruit for supper set. And Conway called the former number "one," and
while you work out the translation. Too bad languages aren't the latter he called "minus one." And he proved that
my thing, or I'd try to help you. minus one is less than but not equal to zero and zero is

less than but not equal to one. And the evening . . .

That's where it breaks off.

B. Okay, Alice, I've got it. There are a few doubtful places, a A. Are you sure it reads like that?
couple signs I don't recognize; you know, maybe some obso- B. More or less. I dressed it up a bit.
lete word forms. Overall I think I know what it says, though
A. But "Conway". . .that's not a Hebrew name. You've got to
I don't know what it means. Here's a fairly literal translation:
be kidding.
In the beginning, everything was void, and J. H. W. H. B. No, honest. Of course the old Hebrew writing doesn't show any
Conway began to create numbers. Conway said, "Let vowels, so the real name might be Keenawu or something;
there be two rules which bring forth all numbers large maybe related to the Khans ? I guess not. Since I'm translating
and small. This shall be the first rule: Every number into English, I just used an English name. Look, here are the
corresponds to two sets of previously created numbers, shows up on the stone. The
places where it J. H. W. H.
such that no member of the left set is greater than or might also stand for "Jehovah."
equal to any member of the right set. And the second rule
A. No vowels, eh ? So it's real . . . But what do you think it means ?
shall be this: One number is less than or equal to another
B. Your guess is as good as mine. These two crazy rules for
number if and only if no member of the first number's
numbers. Maybe it's some ancient method of arithmetic that's
left set is greater than or equal to the second number, and
been obsolete since the wheel was invented. It might be fun
no member of the second number's right set is less than
to figure them out, tomorrow; but the sun's going down pretty
or equal to the first number." And Conway examined
soon so we'd better eat and turn in.
these two rules he had made, and behold they were very
!

good. A. Okay, but read it to me once more. I want to think it over,

and the first time I didn't believe yon were serious.
And the first number was created from the void left set
."
and the void right set. Conway called this number " zero," B. (pointing) "In the beginning.. .
SYMBOLS

rffr*

j-

A. I think your Conway Stone makes sense after all, Bill. I was
thinking about it during the night.

B. So was I, but I dozed off before getting anywhere. What's

the secret ?

A. It's not so hard, really; the trouble is that it's all expressed
in words. The same thing can be expressed in symbols and
then you can see what's happening.

9
B. You moan we're actually going to use the New Math to de- A. Good point, but that's the whole beauty of Conway's scheme.
cipher this old stone tablet. Each element of X L and X R must have been created previ-

A. I hate to admit it, but that's what it looks like. Here, the ously, but on the first day of creation there weren't any
first rule says that every number x is really a pair of sets,
previous numbers to work with; so both X, and X R are taken
called the left set x L and the right set xR :
as the empty set!

B. I never thought I'd live to see the day when the empty set
* = (*/.. *«)•
was meaningful. That's really creating something out of

B. Wait a sec, you don't have to draw in the sand, I think we

nothing, eh? But is XL £ XR when A', and X R are both equal
to the empty set? How can you have something unequal
si ill have a pencil and some paper in my backpack. Just a
itself?
minute. . .Here, use this.
Oh yeah, yeah, that's okay since it means no element of the
A. » = (*t, **)• empty than or equal to any element of the empty
set is greater

set — a true statement because there aren't any elements

it's

These xL and x K are not just numbers, they're sets of numbers; in the empty set.
and each number in the set is itself" a pair of sets, and so on.
A. So everything gets started all right, and that's thenumber
B. Hold it, your notation mixes me up. I don't know what's a called zero. Using the symbol to stand for the empty set,
set and what's a number. we can write

A. Okay, I'll use capital letters for sets of numbers and small
= (0, 0).
letters for numbers. Conway's first rule is that

B. Incredible.
x = (X L , XR ), where XL J XR .
(1)
A. Now on the second day, it's possible to use in the left or

This means if xL is any number in XL and HzB is any number right sets, so Conway gets two more numbers
in A'„, they must satisfy xL £ x R And
. that means xL is not
- 1 = (0, {0}) and 1 = ({()}, 0).
greater than or equal to xR .

B. (scratching his head) I'm afraid you're still going too fast B. Let me see, does this check out ? For - 1 to be a number, it

for me. Remember, you've already got this thing psyched out, has to be true that no element of the empty set is greater
but I'm still at the beginning. If a number is a pair of sets than or equal to 0. And for 1, it must be that is not greater
of numbers, each of which is a pair of sets of numbers, and than any element of the empty set. Man, that empty set
so on and so on, how does the whole thing get started in the sure gets around! Someday I think I'll write a book called
first place ? Properties of the. Empty Set.

10 11
 A. You'd never finish.
These are the symbols I couldn't decipher yesterday, and
If X L or X R is empty, the condition XL £ X R is true no matter your notation makes it all crystal clear ! Those double dots
what is in the other set. This means that infinitely many separate the left set from the right set. You must be on the
numbers are going to be created. right track.

B. Okay, but what about Conway's second rule ? A. Wow, equal signs and everything! That stone-age carver must

A. That's what you use to tell whether XL £ XR , when both

have used ^ to stand for — 1; I almost like his notation

sets are nonempty; it's the rule defining less-than-or-cqual. better than mine.

Symbolically, 15. I bet we've underestimated primitive people. They must have
had complex lives and a need for mental gymnastics, just like
x < // means XL £ y and X * )•„. (2) us —at least when they didn't have to fight for food and shel-
ter. We always oversimplify history when we look back.
B. Wait a minute, you're way ahead of me again. Look, X is
L
A. Yes, but otherwise how could we look back?
a set of numbers, and y is a number, which means a pair of
Beta of numbers. What do you mean when you write
X B. I see your point.
L
Z !/? A. Now comes the part of the text I don't understand. On the
A. I mean that every element of X L satisfies x L £ y. In other first day of creation, Conway "proves" that 5 0. Why
words, no element of XL is greater than or equal to y. should he bother to prove that something is less than or equal

B. Oh, I see, and your to itself, since it's obviously equal to itself. And then on the
rule (2) says also that x is not greater than
second day he "proves" that — 1 is not equal to 0; isn't that
or equal to any element of YR . Let me check that with the
text... obvious without proof, since -1 is a different number?

A. The Stone's version a

B. Hmm. I don't know about you, but I'm ready for another
is little different, but x < y must mean
the same thing as y > swim.
x.

Yeah, you're right. Hey, wait a

A. Good idea. That surf looks good, and I'm not used to so
I'..
sec, look here at these carv-
ings off to the side: much concentration. Let's go!

• =<:>

12 13
PROOFS

B. An idea hit me while we were paddling around out there.

.Maybe my translation isn't 0OET60t.

A. What? It must be okay, we've already checked so much of

it out.

K. 1 know; but now that I think of it, I wasn't quite sure of that
word I translated "equal to." Maybe it has a weaker meaning,

15
 "similar to" or "like." Then Conway's second rule becomes A. Yes, but so far we've been using the empty set in almost every
"One number is less than or like another number if and only argument, so the full implications of the rules aren't clear
if. ." And later on, he proves that Have you
.
zero is less than or like yet. noticed that almost everything we've proved
zero, minus one is less than but not like zero, and so forth. so far can be put into a framework like this: "If X and Y
A. Oh, right, that must be it, he's using the word
are any sets of numbers, then x = (0, X) and y = ( Y, 0) are
in an abstract
technical sense that must be defined by the numbers, and X ^ y."
rules. So of course
he wants to prove that is less than or like 0, in order to B. It's neat the way you've just proved infinitely many things,
see that his definition makes a number "like" itself. by looking at the pattern I used in only a couple of cases. I
B. So does his proof go through By guess that's what they call abstraction, or generalization, or
must show that ? rule (2), he
no element of the empty set is greater than or like 0, and something. But can you also prove that your X is strictly less

that is not greater than or like any element of than y ? This was true in all the simple cases and I bet it's true
the empty
set. .
.Okay, it works, the empty set triumphs again. in general.

A. More interesting is how he could prove that -

A. Uh huh. . .Well no. not when X and Y are both empty, since
1 is not like 0.
The only way I can think of is that he proved that
that would mean % 0. But otherwise it looks very interest-
is not
- ing. Let's look at the case when X is the empty set, and Y
less-than-or-like 1. I mean, we have rule (2) to tell whether
one number is less than or is not empty; is it true that is less than (
}', ())?
like another; and if a; is not less-
than-or-like y, it isn't less than y and it isn't like y. B. If so, then I'd call (}', 0) a "positive" number. That must be
what Conway meant by zero separating the positive and neg-
B. I we want to show that
see, < -1 is false. This is rule (2)
ative numbers.
with x = and )'„ = {0}, so < -1 and only
if if J 0.
But is > 0, we know that, so $ - 1. He was
A. Yes, but look. According to rule (2), we will have (Y, 0) ^
right.
if and only if no member of > is greater than or like 0. So if,
A. I wonder if Conway also tested - 1 against I; I suppose he for example, Y is the set {-1}, then (Y, 0) ^ 0. Do you want
did, although the rock doesn't say anything about it. If the positive numbers to be < ?
rules are any good, it should be possible to prove that - 1 is
less than 1.
Too bad 1 didn't take you up on that bet.

B. Well, let's see:- 1 is (0, {0}) and is ({0}, 0), so once again
B. Hmm. You mean (
}', 0) is going to be positive only when )
1

the empty set makes - 1 < 1 by rule (2). On the other hand,
contains some number that is zero or more. I suppose you're
right. But at least we now understand everything that's on
1 < - 1 is the same as saying that £ - and £ 0, ac- 1 1
the stone.
cording to rule (2), but we know that both of these are
false.
Therefore 1 $ - 1, and it must be that - 1 < 1. Conway's A. Everything up to where it's broken off.

rules seem to be working. B. You mean . . . ?

16 17
A. I wonder what happened on the third day.

B. Yes, we should beable to figure that out, now that we know

the rules. It might be fun to work out the third day, after
lunch.

A. You'd better go catch some fish; our supply of dried meat is

getting kinda low. I'll go try and find some coconuts.

18
BAD NUMBERS

'tn,^^

B. I've been working on that Third Day problem, and I'm

afraid it's When more and more num-
going to be pretty hard.
bers have been created, the number of possible sets goes up
fast. I bet that by the seventh day, Conway was ready for a
rest.

A. Right. I've been working on it too and I get seventeen num-

bers on the third day.

21
B. Really? I found nineteen; you must have missed two. Here's than or like y, but it shouldn't be necessary to make so many
my list:
tests. If any element of XL
> y, then the largest element of
is

X L ought to be > y. Similarly, we need only test a: against the

<:><-:><•:>< :> <-•:><-•:> <• :> smallest element of YR .

<-••:> 0-> <:•> <:•> <:-•><:-!> B. Yeah, that oughta be right ...
1}, 0) just like I proved it was
I can prove that
less than
1 is less than
the extra
({(), ({1}, 0);

<:••><:-•»><-••><•:•><-•:•><-:••> "0" in X L didn't seem to make any difference.

A. If what I said is true, it will save us a lot of work, because

A. I see you're using the Stone's notation. But why did you each number (X L , XR ) will behave in all < relations exactly
include \Z/ ? That was created already on the first day. as if XL were replaced by its largest element and XR by its

smallest. We won't have to consider any numbers in which

B. Well, we have to test the new numbers against the old, in
order to see how they fit in.
XL or XR have two or more elements; ten of those twenty
numbers in the list will be eliminated!
A. But I only considered new numbers in my list of seventeen,
B. I'm not sure I follow you, but how on earth can we prove
so there must actually be twenty different at the end of the
such a thing?
third day. Look, you forgot to include:
A. What we seem to need is something like this:

<-:•> if x < y and y < z, then x < z. (Tl)

in your list. I don't see that this follows immediately, although it is con-
B. (blinking) So I did. Hinin ... 20 by 20, that's 400 different sistent with everything we know.
cases we'll have to consider in rule (2). A lot of work, and not B. At any rate, it ought to be true, if Conway's numbers arc; to
much fun either. But there's nothing else to do, and 1 know be at all decent. We could go ahead and assume it. but it

it'll bug me until I know the answer. would be neat to show once and for all that it was true, just

A. Maybe we'll think of some way to simplify the job once we by using Conway's rules.

get started. A. Yes, and we'd be able to solve the Third Day puzzle without
B. Yeah, that would be nice . . .
much more work. Let's see, how can it be proved ....
B. Blast these flies! Just when I'm trying to concentrate. Alice,
Well, I've got one result, 1 is less than ({1}, 0). First I had to
prove that £ ({I}, 0).
can you —no, I guess I'll go for a little walk.

A. I've been trying out a different approach. Rule (2) says we

have to test every element of X L to see that it isn't greater Any progress?

22 23
A. No, I seem to be going in circles, and the £ versus < is A. Bill ! You've got it.

confusing. Everything is stated negatively and things get

B. Havel?
incredibly tangled up.
A. If (x, y, z) are three bad numbers, there are two possible cases.
B. .Maybe (Tl) isn't true.
Case i, some x L > z: Then (y, z, xL ) are three more bad num-
A. But it has to be true. Wait, that's it! We'll try to disprove it.
bers.
And when we fail, the cause of our failure will be a proof!
Case. 2, some zH < x: Then (z R , x, y) are three more bad num-
B. Sounds good— it's always easier to prove something wrong bers.
than to prove it right.
B. But aren't you still going in circles? There's more and more
A. Suppose we've got three numbers x. y, and z for which bad numbers all over the place.

x < y, and y <. z, and x $ z. A. No, in each case the new bad numbers are simpler than the
original ones; one of them was created earlier. We can't go
What does rule (2) tell us about "bad numbers" like this?
on and on finding earlier and earlier sets of bad numbers, so
B. It says that
there can't be any bad sets at all

Xi * y, B. (brightening) Oho! What you're saying is this: Each number

and x £ )'„, a: was created on some day d[x). If there are three; bad num-
and YL i z, bers (x, y, z), for which the sum of their creation days is

and i, £ Z„> d(x) + d(y) + d(z) = n, then one of your two cases applies
and gives three bad numbers whose day-sum is less than n.
and then also x $ z, which means w hat ! Those, in turn, will produce a set whose day -sum is still less

A. It means one of the two conditions fails. Either there is a and so on; but that's impossible since there are no three num-
bers whose day-sum is less than 3.
number x L in X L for which x L > z. M there is a number zR

in Z„ for which x > z

R With all . these facts about x, y, and A. Right, the sum of the creation days is a nice way to express
:. ire ought to be able to prove something. the proof. If there are no three bad numbers (x, y, z) whose
B, Well, since x L is in A',, it can't lie greater than or equal to y.
day-sum is less than n, the two cases show that there are
Say it's less than y.But y < z, so xL must be no, . . . sorry, none whose day-sum equals n. I guess it's a proof by induction
I can't use facts about numbers we haven't proved. on the day -sum.

Going the other way. we know that y < < x L and

z and z B. You and your fancy words. It's the idea that counts.
y £ x L so this gives us three
; more bad numbers, and we can A. True; but we need a name for the idea, so we can apply it
get more facte again. But that looks hopelessly complicated. more easily next time.

24
25
B. Yes, I suppose there will be a next time . . .

Okay, I guess there's no reason for me to be uptight any more

about the New Math jargon. You know it and I know it;
we've just proved the transitive law.

A. (sigh) Not bad for two amateur mathematicians!

B. It was really your doing. I hereby proclaim that the transi-

tive law (Tl) shall be known henceforth as Alice's Theorem.

A. C'mon. I'm sure Conway discovered it long ago.

B. But does that make your efforts any less creative ? 1 bet every

great mathematician started by rediscovering a bunch of " well-

known " results.

A. Gosh, let's not get earned away dreaming about greatness!

Let's just have fun with this.

26
PROGRESS

B. Ijust thought of something. Could there possibly be two

numbers that aren't related to each other at all? I mean

x £ y and y £ x,

like one of them is out of sight or in another dimension or

something. It shouldn't happen, but how would we prove it?

29
A. I suppose we could try the same technique that worked before.
A. But I . . . Okay, you're right, excuse me for going off on such
x and y are bad numbers in
If this sense, then either some a silly tangent. We have x < xLL and xLL < xL so the transi- ,

xL > y or x > some yR .

tive law tells us that x < x L . This probably gets around the
B. Hmm. Suppose y < xL Then.
xL <
we would have y < x
if x, need for extra subscripts.
by our transitive law, and we have assumed that
y $ x. So B. Aha, that docs it. We can't have a: < xL , because that would
xL % x. In the other case, y R < x, the same kind of figuring
mean XL £ xL which
, (a impossible since xL is one of the
would show that y $ yR .
elements of XL .

A. Hey, that's very shrewd! All we have to do now to show

that A. Good point, but how do you know that xL < xL .

such a thing can't happen is prove something I've suspected

for a long time. Every number x must lie between all B. What You mean we've come this far and
? haven't even proved
the
that a number is like itself? Incredible there must be an
elements of its sets X L and X R . I mean,
. . .

easy proof.
XL <x and x < Xa . (T2) A. Maybe you can see it, but I don't think it's obvious. At any
rate, let's try to prove
B. That shouldn't be hard to prove. What does xL % x say?
A. Either there is a number xLL in X LL , with xLL > x, or else there
x < x. (T3)
isa number xB in X R with x L > x R But the second case can't
happen, by rule (1).
.

This means that XL £ x and x £ XR .

B. I knew we were going to use B. It's curiously like (T2). But uh-oh, here we are in the same
rule (1) sooner or later. But what
can we do with xLL l I don't spot again, trying to show that x < x, is impossible.
like double subscripts.
A. Well, xLL is an element of the left set of a; Since x was A. This time it's all right, Bill. Your argument shows that x ^ x L
t L .

created earlier than x, we can at least assume that x implies xL £ x L which , is impossible by induction.
LL < x L ,

by induction. B. Beautiful! That means (T3) is true, so everything falls into

B. I^ead on. place. We've got the "A',. < x" half of (T2) proved, and the
other half must follow by the same argument, interchanging
A. Let's sec, xLL <, xL says that xLLL £ x L and . . .
left and right everywhere.
B. (interrupting) don't want to look at this—your subscripts
I
A. And like we said before, (T2) is enough to prove that all num-
are getting worse.
bers are related; in other words
A. You're a big help.
if a; $ .// then y < x. (T4)
B. Look, I am helping, I'm telling you to keep away from those
hairy subscripts! B. Right. Look, now we don't have to bother saying things so

30
31
 I

indirectly any more, since "x £ y" is exactly the same as

A. You know, when you look over everything we've proved, it's
"x is less than y."
really very pretty. It's amazing that so much flows out of
A. I see, it's the same as "z is less than or like y but not like y." Conway's two rules.
We can now write
B. Alice, I'm seeing a new side of you today. You really put to
x < y rest the myth that women can't do mathematics.

A. Why, thank you, gallant knight!

in place of x £ y, and the original rules (1) and (2) look miicli B. I know it sounds crazy, but working on this creative stuff
nicer. I wonder why Conway didn't define things that way? with you makes me feel like a stallion! You'd think so much
Maybe because a third ride would be needed to define what
it's

"less than" means, and he probably wanted to keep

brainwork would turn off any physical desires, but really —
down haven't felt quite like this for a long time.
the number of rules.
A. To tell the truth, neither have I.
B. I wonder if it's possible to have two different numbers which
B. Look at that sunset, just like in the poster we bought once.
are like each other. I mean, can we have both x < y and
And look at that water.
x Z y when X L is not the same set as }',. I

A. (ninning) Let's go!

A. Sure, we saw something like that before lunch. Don't yon
remember, we found that < y and y < when y = ({- 1},
0). And I think ({<>, 1}, 0) will turn out to be like ({1}. 0).

B. You're right. When X < y and x > y, I guess x and y are

effectively equal for all practical purposes, because
the tran-
sitive law tells us that x < z if and only if// <
z. They're
interchangeable.

A. Another thing, we've also got two more transitive laws, I

mean

if x < y and 2/ < 2 then x < z; (T5)

d x < y and y < z then x < z. (TO)

B. Very nice—in fan. these both follow immediately from (Tl),

il we consider "a- < y" equivalent
to "x y". There's no £
need to use (T2), (T3), or (T4) in the proofs of (T5) and (T6).

32
33
 THE THIRD DAY

;*&
•i

B. Boy, I never slept so well.

A. Me too. It's so great to wake up and be really awake, not just

"coffee-awake."

B. Where were we yesterday, before we lost our heads .and forgot

all about mathematics ?
A. (smiling) I think we had just proved that Conway's numbers

35
 behave like all little
numbers should; they can be arranged
in a from smallest to largest, with every number
line,
x = (X L , XR ) is a number, and we take any other sets of
being numbers Y L and Y R where ,
greater than those to its left and less
than all those on its
right.
YL <x< YR .

B. Did we really prove that?

Then I think it's true that x is like z, where
A. Yes. anyway at least the unlike numbers keep in
line, because
of (T4). Every new number created
must fall into place among z = (Y L KJX L ,X R U Y r ).
the others.

B. Now it should be
pretty easy for aa to figure out what
hap-
In other words, enlarging the sets XL and XR , by adding
pened cm the Third Day: those 20 x 20
calculations must be
numbers on the appropriate sides, doesn't really change x.

reduced way down. Our theorems

(T2) and (T.i) show that B. Let's see, that sounds plausible. At any rate, z is a number,
according to rule (1); it will be created sooner or later.

A. In order to show that z < x, we have to prove that

so seven of the numbers are placed

already and just a
matter of fitting the other ones
it's
YL U XL < x and z < XR .

in.

You know, now that it's getting

than a crossword puzzle.
easier, this is t.uu-l, more fun But that's easy, now, since we know that Y L < x, XL < x,

and 2 < XR U YR , by (T3).

A. We also know, for example, that And same argument, interchanging
B. the left and right, shows
that x < z. You're right, it's true:
<-:•>
if YL < x < Y l:
,

lies somewhere between and | x = (YL U

. Let's check it against then A' t , A' s U YR ). (T7)
the middle element 0.

B. Hmm. it's both < and > 0, so it must be (I'm going to write "x m z," meaning x is like z, I mean
according to
like 0,
rule (3). As I said yesterday, it's effectively equal to 0, so we
x < 2 and 2 ^ a;.)

might as well forget it. That's eight down and welve A. That proves just what we want. For example,
I to go.
A. Let's try to get rid of thus,, ton msm where X, or XR have
more than one element, like I tried to do
yesterday morning.
<-•:> = <•:.><: -•> = <:->
I had an idea during the
night which might work.
Suppose and so on.

36
37
B. So we're left with only two cases: ^«» *\ and S\ |\ . B. (shaking his head) Incredible, Holmes!

A. Actually, (T~) applies to both of these, too, with X = 0! A. Elementary, my dear Watson. One simply uses deduction.
B. Cle-ver. So the Third Day is now completely analyzed: only B. Your subscripts aren't very nice, but I'll ignore it this time.
those seven numbers we listed before are essentially different. What would you do with the number ({a^-j}, fo + i}) if i < }1
A. I wonder if the same thing won't work for the following days, A. (ahrngging her shoulders) I was afraid you'd ask that. I don't
too. Suppose the different numbers at the end of n days are know.

B. Your same argument would work beautifully if there was a

x t < x2 < < xm .

number x where each element of X L is < x, _ : and each element

Then maybe the only new numbers created on the (n + I)st

of X H is >x j + 1 .

day will be A. Yes, you're right, I hadn't noticed that. But all those ele-
xt
(#,{*»», <W»to». • «**-x). W). ({*«). «)•
B.
ments x,,xt+1

I suppose so . . .
in

No, I've got

between might

it!
interfere.

Let x be the one of x,, x l+1 ,

B. Alice, you're wonderful! If we can prove this, it will solve

. . ., Xj which was created first. Then X L and X n can't involve

infinitely many days in one swoop! You'll get ahead of the any of the others! So ({a^-j}, {«/ + i}) = x.

Creator himself. A. Allow me to give you a kiss for that.

A. But maybe we can't prove it.

B. Anyway let's try some special eases. Like, what if we had the
number ({a-'i-i}. {*i+i})i 't would have to be equal to one of
the others. B. (smiling) The problem isn't completely solved, yet; we have
A. Sure, it equals x„ because of (T7). Look, each element of to consider numbers like (0,
{% +1}) and ({*,_,}, 0). But in the
X IL is <#(_!, and each clement of X lR is >.r, + 1
. Therefore, lirst case, we get the first -created number of Xlt x2 ., x,, , . .

by (T7), we have and in the second case it's the first-created number of r,,

x(+1 . . . ., xm .

*« = (fa-i) UX iL ,
X IR u {.r
( + 1 }). A. What number wasn't unique? I mean, what
if the first-created
ifmore than one of the x„ ...', Z; were created on that earliest
And again by (T7),
day?

= (Xil VJ fa-i}, U A' 1H B. Whoops. . . No, it's okay, that can't happen, because the proof
({*i-i}. fo+i}) fa +1 } ).

is still valid and it would show that the two numbers are both
By the transitive law. x t
= ({a^-j}, {x, + l }).
like each other, which is impossible.

38 39
A. Neato! You've solved the problem of all the days at once.
B. With your help Lei 'a see, on the fourth day there will be 8
new numbers, then on the fifth day there arc 16 more, and so
on.

A. Yes, after the »th day, exactly 2" - 1 different numbers will
have been created.

B. You know, I don't think that guy Conway was so smart after

all. Imean, he could have just given much simpler rules.

with the same effect. There's no need to talk about sets of
numbers, and all that nonsense: he simply would have to say
that the new numbers are created between existing.adjaeent
ones, or at the ends.

C. Rubbish. Wait until you get to infinite sets.

A. What was that? Did you hear something? It sounded like

thunder.

B. I'm afraid we'll be getting into the monsoon season pretty

soon.

40
DISCOVERY

...''

,.- •-"^W

A. Well, we've solved everything on that rock, hut I can't help

feeling there's still a lot missing.

B. What do you mean ?

A. I mean, we know what happened on the third day, four
like

new numbers were created. But we don't know what Conway

called them.

43
 B. Well, one of the numbers was bigger than 1. so I suppose
he A. Look, there's a cave over by that cliff. Let's wait out the storm
called it "2." And another was between and 1, so maybe Hey, really pouring!
" |." in there. it's
he called it

A. That's not really the point; what

really bothers nu- is. why
are they numbers 1 I mean, in
order to be numbers you have to
B. Sure is dark in here. Ouch! I stubbed my toe on something.
be added, subtracted, and that sort
of thing.
Of all the . . .

B. (frowning) I see. You think Conway gave some more rales in

A. Bill ! You've found it! You stubbed your toe on the other part
the broken-oil part of the rock, which made
the number* of the Conway Stone!
numerical. All we have is a bunch of
objects ordered neatly in
aline, but ire haven't got anything B. (wincing) Migosh, it look's like you're right. Talk about fate!
to do with them!
But my toe isn't as pleased about it as the rest of me is.
A. I don't think I'm clairvoyant enough to
guess what he did— A. Can you read it, Bill ? Is it really the piece we want, or is it
if he did do something.
something else entirely ?

B. That means we're stuck, unless we

can find the missing part
B. It's too dark in here to see much. Help me drag it out inThe
of that rock. And don't
I remember where we found the first rain, the water will wash the dust off and . . .

part.
Yup, I can make out the words "Conway" and "number,"
A. Oh, I remember was careful to note exactly where
that, I
it so it must be what we're looking for.
was in case we ever wanted to go back.
A. Oh, good, we'll have plenty to work on. We're saved!
B. What would I do Without you? Come on, let's go! But I'm going back
B. The info we need is hero all right. in the
A. Hey wait, don't you think we
should have a little lunch first? cave, it can't keep raining this hard for very long.

B. Right, I got so wrapped up in this A. (following) Right, we're getting drenched.

J forgot all about food.
Okay, let's grab a quick bite and then start digging.

B. I wonder why this mathematics is so exciting now, when it

A. (digging) Oh, I'm afraid this isn't going to work.

Bill,
was so dull in school. Do you remember old 1'rofessor Lan-
The
dirt under the sand is so hard, we need dau's lectures? I used to really hate that class: Theorem,
special tools.
proof, lemma, remark, theorem, proof, what a total drag.
B. Yeah, just scraping away with this knife isn't getting us very
A. Yes, I remember having a tough time staying awake. But
far. Uh oh, here comes the rain, too. Should
we dash back
- camp?
fco
look —wouldn't our beautiful discoveries be just about the
same?

44
45
 B. True. I've got this mad urge to get up before a class and
A. Exactly. But teachers are so conservative, they'd be afraid
present our results: Theorem, proof, lemma,
remark. I'd make of searing off the "grind" type of students who obediently
it so slick, nobody would be
able to guess how we did it, and
and mechanically do all the homework. Besides, they wouldn't
everyone would be so impressed.
like the extra work of grading the answers to nondirected
A. Or bored.
questions.
B. Yes, there's that. I guess the way to put off all creative aspects until
excitement and the beauty comes The traditional is

in the discovery, not the

hearing. the last part of graduate school. For seventeen or more years,
A. But it is beautiful. And a student is taught examsmanship. then suddenly after pass-
enjoyed hearing your discoveries
I

almost as much as making my own. ing enough exams in graduate school he's told to do something
So what's the real differ-
ence? original.

B. I guess you're right, at thai.

was able to really appreciate
I
B. Right. I doubt if many of the really original students have
what you did. because I had already been struggling with the stuck around that long.
same problem myself. A. Oh, I don't know, maybe they're original enough to find a
A. It was dull before, because we
weren't involved at way to enjoy the system. Like putting themselves into the
all; we
were just being told to absorb what
somebody subject, as we were saying. That would make the traditional
else did. and
for all we knew there was nothing special about it.
college courses tolerable, maybe even fun.

15. From now on whenever I read a math book, I'm going B. You always were an optimist. I'm afraid you're painting too
to
try to Ggure out by myself how everything was done, before rosy a picture. But look, the rain has stopped, let's lug this
looking at the solution. Even if [ rock back to camp and see what it says.
don't figure it out, I think
I'll be able to see the beauty
of a proof then.
A. And think we should also try to guess
I
what theorems are
coming up; or at least, to figure out how
and why anybody
would try to prove such theorems in the first
place. We should
imagine ourselves in the discoverer's
place. The creative part
is really more
interesting than the deductive
part. Instead of
concentrating just on finding good answers
to questions, it's
more important to learn bow to find good
questions!
B. You've got something 1hen-. wish our teachers would give us
I

problem* like, -Find something interesting about x."

instead
of "Prove*."

46
47
ft

A. The two pieces fit pretty well, it looks like we've got almost
the whole message. What does it say I

B. This part is a little harder to figure out, there are some ob-
scure words, but I think it goes like this:

. . . day. And Conway said, "Let the numbers be added to

each other in this wise : The left set of the sum of two
numbers shall be the sums of all left parts of each number

49
 with the other; and in like manner the right set shall be A. What a weird ending. And what do you mean "aleph day"?
from the right parts, each according to his kind." Conway
B. Well, aleph is a Hebrew letter and it's just standing there by
proved that every number plus zero is unchanged, and
itself, look: X. It seems to mean infinity. Let's face it, it's
he saw that addition was good. And the evening and the
heavy stuff' and it's not going to be easy to figure out what
morning were the third day.
iliis means.
And Conway "Let the negative of a number have
said,
A. Can you write it all down while I fix supper? It's too much
as its sets the negatives of the number's opposite sets;
for me to keep in my head, and I can't read it.
and let subtraction be addition of the negative." And it
was Conway proved that subtraction was the inverse
so. B. Okay, that'll help me get it clearer in my own mind too.

of addition, and this was very good. And the evening

and
the morning were the fourth day.

And Conway said to the numbers, "Be fruitful and multi- A. It's curious that the four numbers created on the third day
Let part of one number be multiplied by another and
ply.
aren't mentioned. I still wonder what Conway called them.
added to the product of the first number by part of the
other, and let the product of the parts be subtracted. This
B. Maybe if we try the rules for addition and subtraction we
shall be done in could figure out what the numbers are.
all possible ways, yielding a number in
the left set when the parts are of the same
of the product A. Yeah, if we can figure out those rules for addition and sub-
kind, but in the right set when they are of opposite traction. Let's see if we can put the addition rule into symbolic
kinds."
Conway proved that every number times one is unchanged. form, in order to see what it means ... I suppose "its own

And the evening and the morning were the fifth day. kind" must signify that left goes with left, and right with

And behold! When the numbers had been created for

right. What do you think of this:
infinitely many days, the universe itself appeared. And
the evening and the morning were X day. * + V = UXl +9)V(7l + x), (Y R + x) u (X R + y)).

(3)
And Conway looked over all the rules he had made for
numbers, and saw that they were very, very good. And B. Looks horrible. What does your rule mean?
he commanded them to be for signs, and series, and
quo- A. To get the left set of a: + you take all numbers of the form
y,
tients, and roots.
xL + y, where x L is in XL , and also all numbers y, + x where
Then there sprang up an infinite number less than in- y L is in Y L The right
. set is from the right parts, "in like
finity. And infinities of days brought forth multiple orders manner."
of infinities.
B. I see, a "left part" of # is an clement of XL . Your symbolic
That's the whole bit.
definition certainly seems consistent with the prose one.

50
51
A. And it makes sense too, because each xL + y and x + yL A. Congratulations on coming up with the world's longest proof
ought to lie less than x + y. that 1 + 1 is 2.

B. Okay, I'm willing to try it and see how it works. I see you've B. Have you ever seen a shorter proof?
called it rule (3).
A. Not really. Look, your calculations help me too. I get 1+2
A. Now after the third day, we know that there are seven num- = ({2}, 0), a number that isn't created until the fourth day.
bers, which we might call 0, 1, - 1, a, b, c, and d. B. I suggest we call it "3."
B. No, I have an idea that we can use left-right symmetry and A. Bravo. So rule (3) is working; let's check if b is \ by calculating
call them b + b ...

-a < -1 < -6 < < 6 < 1 < a, B. Hmm, that's odd, it comes out to ({6}, {b + 1}), which hasn't
been created yet.
where
A. And ft + 1 is ({6, 1}, {2}), which is like ({1}, {2}), which is

created on the fourth day. So 6 + b appears on the fifth day.

-a = = a B. Don't tell me b + b is going to be equal to another number we
-1 = don't know the name of.
= |=1
A. Are we stuck?
-6 =
<-:•> <•:> = 6
B. We worked out a theory that tells us how to calculate all
numbers that are created, .so we should be able to do this.
<:> - «
Let's make a table for the first four days.

A. Brilliant! You must be right, because Conway's next rule

A. Oh, Bill, that's too much work.
is B. No, it's a simple pattern really. Look:

-x = (-X R , -A",).
(4)

Day I

B. So it is! Okay— now we can start adding these numbers. Like,

what's 1 + 1, according to rule (3)? l).l\ 'J -1

A. You work on that, and I'll work on 1 + a. Day :t -2

B. Okay, 1 get ({0 + 1,0+ 1}, 0). And + 1 is ({0 + 0}, 0),
Day i -3 -(/,.|) d 6+1
+ is (0, 0) = 0. Everything fits together, making + 1 1

= ({!}. 0) = a. Just as we thought, a must be 2

52 53
A. Oh + b is (b, b + 1), which is formed from rum-
I see, so 6
adjacenl numbers And our theory says it is the earliest-
. . .

created number between them.

B. (beaming) And that's 1, because 1 makes the scene before c.

A. So b is 4 after all, although its numerical value wasn't estab-

lished until two.days later. It's amazing what can be proved
from those few rules -they all hang together so tightly, it
boggles the mind.

B. I'll bet d is $ and c is %.

A. But the sun is going down. Let's sleep on it, Bill; we've got
lots of time and I'm really drained.

B. (muttering) d + c = . . . Oh, all right. G'night.

54
THE ANSWER

{ftfa

A. Are you awake already?

B. What a miserable night! I kept tossing and turning, and my
mind was racing in circles. I dreamed I was proving things
and making logical deductions, but when I woke up they were
all foolishness.

A. Maybe this mathematics isn't good for us after all. We were

so happy yesterday, but

57
B. (interrupting) Yeah, yesterday we were high on math, but each XL is < some yL by , (T4). Thus XL < y, and similarly, y <
today it's turning sour. I can't get it out of my system, we've XB Soy^zby
. (T7).
got to get more results before I can rest. Where's that pencil now we have
?
It's pretty easy to work out the proof that all

A. Bill, you need some breakfast. There are some apricots and this ammunition to work with.
figs over then*.
A. The nice thing about (T8) is that itmakes the calculation
B. Okay, but I've gotta get right to work. we did last night much easier. Like when we wete calculating
A. Actually I'm curious to see what happens too, but promise b + b = ({&}, {b + 1}), we could have seen immediately that

me one thing. 1 is the first number created between {6} and {b + 1}.

B. What? B. Hey, let me try that on c + c: It's the first number created
A. We'll only work on addition and subtraction today; not multi- between b + c and + c. Well, it must be 6+1,1 mean 1|,
1

so c is f
plication. Wo won't even look at that other part of the tablet

until later. That's a surprise, I thought it would be §.

B. Agreed. I'm almost willing to postpone the multiplication A. And d is

J.

indefinitely, since it looks awfully complicated. B. Right.

A. (kissing him) Okay, now relax. A. I think the general pattern is becoming clear now: After four
B. (stretching) You're so good to me, Alice. days the numbers > are

A. That's better. Now I was thinking last night about how you 3
0, I h i, I, #, 2,
solved the problem about all the numbers yesterday morning.
I think it's an important principle that we. ought to write and after five days they will probably be
down as a theorem. I mean:
B. (interrupting)
Given any number y. if x is the first number
created with the property that YL < x and 0,i,i,|, h it. i. l.i U.2,f,3, 4.
(T8)
a: < Y R then , a; = y.
A. Exactly. Can you prove it?

B. Hmm, 1 what we proved. Let's see if we can

guess that is
B
reconstruct the proof,
in this new symbolism. As I recall we
Yes, but not so easily as I thought. For example, to figure out
constructed the number z = )', u A',.. X u )'„), and then = which turned out to be
R ( '

the value of/ ({£ }, {2}), J, I calcu-

we had x = z by (T7). On the other hand, no element x of lated/ + This is the number created between 3 and
first
L /.
X L satisfies Y L < xL , since a-,, was created before .r; therefore 4, and I had to "look ahead" to see that it was I. I'm con-

58 59
 vinccd we have the right general pattern, but it would be
B. Sounds logical, since adding y should "shift" things over by
nice tohave a proof.
y units. Whoops, no, take x = 1 ; that would say y + 1 is
A. On the fourth day we calculated § by knowing that it was ({y}< 0)> which fails when y is $.

1 + £, not by trying 2 + 2. Maybe adding 1 will do the trick. A. Sorry. In fact, your rule for 1 + x doesn't work when x =
B. Let's see . . . According to the definition, rule either.
(3),

B. Right, I proved it only when x is positive.

I + •->((! +A' t )u{4,I + XR ),
A. I think we ought to look at rule (3), the addition rule, more
closely and see what can be proved in general from it. All
assuming that + x = x. In fact, isn't it true that . . . sure,
numbers we can always choose X so that we've got are names for the numbers. These names must be
for positive
L 1 + XL correct if Conway's numbers behave like actual numbers, but
has an element >x, so it simplifies to
we don't know that Conway's rules are really the same. Be-
i +*- (i + xu i + x„) sides, I think it's fun to derive a whole bunch of things from
just a few basic rules.

in this case.
B. Let's see. In the first place, addition is obviously what we
A. That's it, Bill! Look at the last eight
numbers on the fifth might call commutative, I mean
day, they are just one greater than the eight numbers on the
fourth day.
x + y = y + x. (T9)

B. A perfect fit. Now all we have to do is prove the pattern for A. True. Now let's prove what Conway claimed, that
the numbers x between and 1 ... but that can always be
done by looking at x + x + = x. (T10)
x, which will be less than 2!

A. Yes, now I'm sure we've got the right pattern. B. The rule says that

B. What a load off my mind. I don't even feel the need to form- x + = (X L + 0, XR + 0).
alize the proof now; 1 know it's right.

A. I wonder if our rule for + x

So all we do is a "day of creation" induction argument,
1 isn't a special case of a more
general rule. Like, isn't again; we can assume that X L + is the same as X L and ,

XR + is XR ,
since all those numbers were created before x.

'J + * = (V + XL , y + XR ) ?
Q.E.I).

A. Haven't we proved that x + = x, not =a;?

That would bo much simpler than Conway's complicated rule. B. You're a nit-picker, you are. I'll change (T10) if you want me

60
61
 to, since it really won't make any difference. But actually
doesn't the proof actually show that x + is identically the 10
same pair of sets as x ?
A. Excuse me again. You're right.

B. That's ten theorems. Shall we try for more while we're hot ?

G2
10 THEOREMS

in*

A. How about the associative law,

(x + y) + z = x + (y + z). (Til)

B. Oh, I doubt if we'll need that; it didn't come up in the calcula-

tions. But I suppose it won't hurt to try it, since my math
teachers always used to think it was such a great thing.

65
 .

One associative law, coming right up. Can you work out the A. In fact, we might as well prove a stronger statement,
definition ?

if X < y, then x + z < y + z, (T13)

A. (x + y) + z = (((X L y) + z) u (()',. + x) + z)
+
«-»
&t (* + V)), ((X R + y) + z)
+ because this will immediately prove (T12).
u ((}' R + x) + z) u (Z
K + (x + y))) B. I see, because x = y if and only if x < y and y < x. Also
a: + (y + z) = {(X L + (y + z )) u ((Y L + z) + x) (T13) looks like it will be useful. Shouldn't we also prove
V ((Z L + y) + x), (A" B + (y + z)) more, I mean
*-» ((*"* + 3) + *) U ((Z„ + y) + «)).
and w <
if x <, y z,

B. You're really
then x + w < y + z\
good at these hairy formulas. But how can such
monstrous things be proved equal ?
A. Oh, that follows from (T13), since x + w<y + w = iv+y
A. It's not hard, just using a day-sum argument on (x, y, z) as <z + y = y + z.

we did before. See, (X L + y) + z = XL + (y + z) because H. Okay, that's good, because (T13) is simpler. Well, you're the
(#l- ?/- 2) has a smaller day-sum than (a;, y, z), and we can expert on formulas, what is (T13) equivalent to?
The same for the other
induct on that. five sets, using the
commutative law in some cases. A. Given that XL < y and x < YR , we must prove that X, + z

< y + z, ZL + x < y + z, x + z < YB + z, and x + z < ZB

B. Congratulations! Another Q.E.D., and another proof of = + y-
instead of =
B. Another day -sum induction, eh? Really, these are getting too
A. That worries me a little. Bill. We showed that we could easy.
substitute like elements for like elements, with respect
to
A. Not quite so easy, this time. I'm afraid the induction will
< and < but don't we have to verify this also for addition
I mean,
, '.

only give us XL + z < y + z, and so on; it's conceivable that

xL < y but x L + z = y + z.
if x = y. then x + z = y + z. (T12)
B. Oh yeah. That's interesting. What we need is the converse,

if x + z < y + z then x < y. (T14)

B. I suppose so, otherwise we wouldn't strictly be allowed to
make the simplifications we've been making in our names
A. Brilliant! The converse is equivalent to this: Given that
for the numbers. As long as we're proving things, we might
XL + z < y + z, ZL + x < y + z, x + z < YK + z, and
as well do it right.
x + z < ZH + y, prove that XL < y and x < YR .

66
67
 B. Hmm. The converse would go through by induction— except A. The same statement also follows immediately from rule (2);
that we might have a case with, say, xL + z < y + z but I mean, y < if and only if Y L < and < y if and only
*,. y. Such cases would be ruled out by (T13), but . . . if < }' B So now what we have to prove is
.

A. But we need (T13) to prove (T14), and (T14) to prove (T13).

And (T13) to prove (T12). xl + (
— *) < 0i and (-x R + x <
) 0,

and.r R + {-x) > 0, and (-xL ) + x > 0,

B. We're going around in circles again.

A. Ah, but there's a way X L and

out, we'll prove them both together 1 for all xL in all xR in XR .

We can prove the combined statement "(T13) and (T14)"

by induction on the day-sum of
B. Hmm. Aren't we allowed to assume that xL + (
— xL = ) and
(z, y, z) !

Xr + (-**) = 0?
B. (glowing) Alice, you're a genius! An absolutely gorgeous, tan-
A. Yes, since we can be proving (T15) by induction.
talizing genius!

A. Not so B. Then I've got it! If x L + (-x) were >0, then (-A')„ + xL
fast, we've still got work to do. We had better show-
that would be >0, by definition. But (-.X) R is -(X L ), which con-
tarns -xL , and (-x,) + x L is not >0. Therefore xL + (-x)
x —x= 0. (T15) must be < 0, and the same technique works for the other

cases too.
B. What's that minus sign We
? never wrote down Conway's rule
for subtraction.
A. Bravo! That settles (T15).

B. What next?
A. x ~ y = x + (-y). (5)
A. How about this I

B. 1 notice you put the m in (CIS); okay, it's clear that % + ( -*)

won't be identically equal to -(-*) = *• (T16)

0, I mean with empty left and
right sets, unless x is 0.
B. Sssss. That's trivial. Next?
A. Rules (3), (4), and (5) say that (T15) is equivalent to this:
A. All I can think of is Conway's theorem,
((.V, + (~x))KJ((-X R + x), )

&a + (-*»U((-JC£) + a!))« o.

(x + y)-y = x. (T17)

B. Uh oh, it looks hard. How do we show something = any- B. What's that equivalent to ?

way ? ... By (T8), y m if and only if Y L < and < >'„, A. It's a real mess . . . Can't we prove things without going back
since was the first created number of all. to the definitions each time ?

68
 B. Aha! Yes, it almost falls out by itself:

11
(X + y) -y = (x + y) + (_ y) by (5)
= X+ (y + (-y)) by (Til)
= x + (y - y) by (5)
=x + by (T12) and (T15)
= X. by (T10)

We've built up quite a pile of useful results—even the associ-

ative law has come in handy. Thanks for suggesting it against
my better judgment.

A. Okay, we've probably exhausted the possibilities

of addition,
negation, and subtraction. There are some
more things we
could probably prove, like

-(x + y) = (-x) + (-y), (T18)

if x < y, then -y ^ -x, (T19)

but I don't think they involve any new ideas; so there's little
point in proving them unless we need 'em.
B. Nineteen theorems, from just a few primitive rules.

A. Now you must remember your promise: This afternoon we

take a vacation from mathematics, without looking
at the
rest of the stone again. I don't want that
horrible multiplica-
tion jazz to rob you of any more sleep.

B. We've done a good day's work, anyhow—all the problems

are
resolved. Look, the tide's just right again.
Okay— the last one
into the water has to cook supper!

70
11 THE PROPOSAL

war AX

A. That sure was a good supper you cooked.

B. (lying down beside her) Mostly because of the fresh fish you
caught.

What are you thinking about now ;

A. (blushing) Well, actually I was wondering what would happen

if I got pregnant.
 B. You mean, here we are, near the Fertile Crescent, and ... ? B. And it was kind of an aplirodisiac, besides.
A. Very funny. .And after all our work to prove that 1 + 1 = 2 A. (gazing at the stars) One nice thing about pure mathematics
we'll discover that 1 + 1=3. the things we proved today will never be good for anything,
B. Okay, you win. no more jokes. But so nobody will be able to use them to make bombs or stuff
come to think of it,
Conway's rules for numbers are like copulation. I like that.
mean the
left set meeting the right set, But we
. . .
B. Right. can't be in an ivory tower all the time, cither.
A. You've got just one thing—no, two things— There are lots of problems in the world, and the right kind of
on your mind.
But seriously, what would we do if I really were pregnant
math might help to solve them. You know, we've been away
?

B. Well, I've been thinking we'd better

from newspapers for so long, we've forgotten all the problems.
go back home pretty
soon anyway; our money's running out, A. Yeah, sometimes I feel guilty about that ....
and the weather is
going to get bad. Maybe the right kind of mathematics would help solve some
Actually, I really want to marry you in any of these problems, but I'm worried that it could also be
case, whether
you're pregnant or not. If you'll have me, of misused.
course.
A. That's just what I feel too. This trip has proved that we're B. That's the paradox, and the dilemma. Nothing can be done
ready for a permanent relationship. without tools, but tools can be used for bad things as well as

wonder good. If we stop creating things, because they might be harm-

I
. . . When our children grow up, will we teach them
ful in the wrong hands, then we also stop doing useful things.
OUT theory of numbers |

A. Okay, I grant you that pure mathematics isn't the answer

B. No, it would be more fun for them to discover it for themselves.
to everything. But are you going to abolish it entirely just
A. But people can't discover everything for
themselves, there has because it doesn't solve the world's problems ?

to be some balance.
B. Oh no, don't misunderstand me. These past few days have
B. Well, isn't learning really a process of self-discovery
all
? Don't
shown me that pure mathematics is beautiful it's an art form —
the best teachers help their students to think like poetry or painting or music, and it turns us on. Our natural
on their own ?
A. In a way, yes. Whew, we're getting philosophical.
curiosity has to be satisfied. It would destroy us if we couldn't
have some fun, even in the midst of adversity.
B. I still can't get over how great I fed when I'm doing this crazy
mathematics; A. Bill, it's good to talk with you like this.
it really turns me on right now, bin 1 used to
hate it.
B. I'm enjoying it too. It makes me feel closer to you, and sort

A. Yes, I've been high on it, too. I think it's of peaceful.

a lot better than
drugs; I mean, the brain can stimulate il self
naturally.

74
75
12 DISASTER

«;-~

B. Are you awake already ?

A. About an hour ago I woke up and realized that there's a big,

gaping hole in what we thought we proved yesterday.

B. No!

A. Yes, I'm afraid so. We forgot to prove that z + y is a number.

B. You're kidding. Of course it's a number, it's the sum of two

 numbers! Oh, wait, I see we have to check that rule (1)
is satisfied.
. . .

A. Yes, and i © J
= |. But then I tried (-i) © i, and I was
stopped cold.
A. Yes, the definition of addition isn't legitimate unless we can © ©
prove that L + y < Xr X + y, and X L + y < YR +
B. You mean . . . ? I see, ( - i) © ft
= ({(- 1) 0}, {0 1}),
or, and
which is ({0}, {0}).
> L +x < XR + y, and YL + x < YR + x.
B. These would follow from (T13)
A. And that's not a number. It breaks rule (1).
and (TI4), but ... I see your
point, we proved (TI3) and (T14) assuming that the sum of
B. So your definition of © wasn't legit.

two numbers is a number. How did you ever think of this A. And I realized that you can't just go making arbitrary defi-
problem ? nitions: they have to be proved consistent with the other rules
A. Well, that's kind of interesting. I was wondering what would too. Another problem with © was, for example, that ({ — 1},
happen if we defined addition like this: 0) s Obut({-l},0)©l ft 0©1.

*®y«<x£ e B. Okay, © is out, but I suppose we can fix up the real definition
>'/, A'„ © r fi ).
of +.
I called this © because it wasn't obviously going to come out A. I don't know; what I've just told you is as far as I got. Except
thesame as + But it was pretty easy to see that I thought about pseudo-numbern.
.

commutative and associative operation, so I

that © was a
wanted to see B. Pseudo-numbers?
what it turned out to be.
A. Suppose we form (A* t , XR ) when XL is not necessarily <X R .

B. I see; the sum of a: and y lies between X L + YL and X R + YR ,

Then rule (2) can still be used to define the < relation between
so this definition might turn out to be simpler than Conway's.'
such pseudo-numbers.
A. But my hopes were soon dashed, when I discovered that B. I see . . . like ({1}, {0}) turns out to be less than 2.

O©* = A. Right. And I just noticed that our proof of the transitive law
(Tl) didn't use the £ part of rule (1), so that law holds for
for all x.
pseudo-numbers too.
B. Ouch! Maybe © means multiplication? B. Yes, I remember saying that the full rule (1) wasn't used
A. Then I proved that
© x = for x >
1 1 all 0, and 2 ©x = 2
until (T2). That seems like a long time ago.
for all x > 1, and 3 © x = 3 for x > all 2, and A. Now get ready for a shock. The pseudo-number
. . . . ({1}, {0}) is
B. For <
I see. all and u, m
positive integers m
» is the minimum © neither nor 2: !

of and m n. That's commutative and associative,

all right. B. Far out!
So your © operation did turn out to be interesting.
A. Yes, think I can prove that < a number y
1 ({1}, {0}) is if

78
79
 and only if y > l, and it is> a number x if and only if x < 0. A. Good idea . . . Hey, the same proof goes through for all
It's not related at all to any numbers between and 1.
pseudo-numbers: x is always like x.
B. Where's the pencil? 1 want to check that out ... I think
B. This is great but I'm afraid it's taking us away from the main
you're .right. This is fun, we're proving things about quantities problem, whether or not + is well-defined.
that don't even exist.
A. Well, our proofs that x y = y + x, x + = x, and even+
A. Well, do pseudo-numbers exist
any less than Conway's num- the associative law, work for pseudo-numbers as well as num-
bers ? What you mean is, we're
proving things about quantities bers. If the inequality theorems (T13) and (T14) also go
that are purely conceptual,
without real-world counterparts through for pseudo-numbers, then + will be well-defined.
as aids to understanding . . . Remember that V=l was once
B. I see, that's beautiful! So far we've established (Tl), (T3),
thought of as an imaginary number,
and VI wasn't even
('IT)), (TO), (T9), (T10), (Til) for all pseudo-numbers. Let's
thought to be "rational."
look at (T13) again.
B. Conway's rule for adding normal
numbers also gives us a
way to add pseudo-numbers. A. But I'm afraid . . . uh, oh. Bill! We were too gullible yesterday
I wonder what this leads to«
If* = in our acceptance of that day-sum proof for (T13) and (T14);
({!}, {0}), then 1 +x is . . .
({2}, {1}).
it was too good to be true.
A. And x + x is ({1 + *•}, {*}), a secon d-order pseudo-number.
B. What do you mean ?
B. Pure mathematics is a real mind-expander.
A. We were proving that ZL + x < y + z by induction, right?
But did you notice that
({1}, {0}) isn't even < itself? Well, to get this it takes two steps, first ZL + x <, ZL + y
A. I.H s see. * < , ,„ean.s that A', < * < A'*, so this could only and then Z L + y < z + y. Induction gives us the first part
be true if XL < XR .
all right, but the second part involves (2,. z, y), which might
No, wait, we aren't allowed to use "<" have a larger day-sum than (x, y, z).
i„ place of "%*> for
pseudo-numbers, since (T4) isn't true in
general. We have to B. So we really blew it. Conway would be ashamed of us.
go back to the original rule
(2), which says that x < x if and
A. Good thing we didn't see this yesterday, or would have
only if X L i x and x > X So it
R ({1}, {0}) is «s itself after all.
.

spoiled our day.

B. Touch*] I'm glad I was
wrong, since every x ought to be like B. I guess it's back to the drawing boards . . . but hey, we've
itsell, even when it's a pseudo-number.
gotta eat some breakfast.
A. Maybe there issome more complicated pseudo-number
that
isn't< itself. It's hard to visualize, because
the sets XL and
X„ might include pseudo-numbers too.
B. Let's look back at our proof
of (T3) and see if it breaks down.

80
81
 —

13 RECOVERY

i
.

mt

if^" 1

'*•-
HET, •
*-*.:^*J0^^Kw _
L _ .. ;
., /

A. We've missed lunch, Bill.

B. (pacing the ground) Have we? This stupid problem is driving

me up the wall.

A. Just staring at this paper isn't helping us any, either. We

need a break; maybe if we ate something
B. What we really need is a new idea. Gimme an idea, Alice.

83
 A. (beginning to eat) Well, when m ware going around in circles I(* t ,y), I(*,Y L ), l(XR ,y), I(x,Y B ),
like this before, how did we break out? The main thing was III(X R> X L , y),
to use induction, I mean to show that the
proof in one case IU(x,Xu y), U(y,YR ,x),
depended on the truth in & previous case, which
on a still previous case, and so on, where
depended 111(2/, }',, x), U{x, XR , y),
the chain must Ill(Y R ,YL ,x).
eventually terminate.

B. Like our day-sum argument. For example, we have to prove among other things that
Xi. + y < YR + x. In other words, for and y R all xL in XL
A. Right. The other way we broke
the circle was by proving more in Y R we should have previously established that x L + y
than we first thought we needed. I mean, in order to
keep the < y R + x. Now III(x, x L y) and (T3) show that x +
,
L y
induction going, we had to keep proving several things simul- < x + and U(y, y R x) shows that y + x < y R + x. Right
y, , ?
taneously.
A. It looks good; except I don't see why you included those first
B. Like when you combined (T13) and (T14). Okay, Alice, four, 1(X L y) through I(a;, Y R ). I mean, even if xL + y wasn't
right ,

after lunch I'm going to sit down

and write out the total a number, that wouldn't matter; all we really need to know
picture, everything we need to prove,
and perhaps even more. is that xL and y themselves are numbers. After all, < and
And I'm going to try and prove everything < are defined for pseudo-nunibers,
simultaneously and the transitive laws
by induction. The old battering-ram approach. work too.
If that doesn't
work, nothing will.
B. No, rule (1) says that elements of the left part like xL + y
A. That sounds hard but probably the best way. Here, have
it's have to be numbers. Anyway it doesn't really matter, because
some oat cakes. if we're proving I(x, y) we can assume I(x
t y) and so on for ,

free; induction takes care of them,

A. It's complicated, but keep going, this looks promisini:

B. This approach has to work or we're sunk. Okay, the proof

B. Okay, hen we go. We want to prove three things about num-
bers, and they all seem to depend on one another. of U.(x, y, z), namely (T13), will follow if we have previously
proved
I. x + y is a number.
I[. if x < y, then x + z y + z
<, .
iiKy, xb z),

in. x + < y + U(x,y,Z L ), UI(z,Z L ,y),

if z z , then x < y.
UI{Y R ,x,z),
Now II(x, y,Z R III(Z R ,z,x).
I'm not mistaken, the proof of
if
l(x, y) will follow if
),
we
have previously proved
That's curious — this one really doesn't require l(x, y). How

84
85
 come we thought we'd have to prove that tho sum of two
I think it was a good idea to introduce this new notation,
numbers is a number, before proving (T13)?
like I(x, y) and so on, because it makes the patterns become
A. That was before we knew much about pseudo-numbers. clearer. Now all we have to do is find some way to rig up an
It's
strange how a fixed idea will remain as a mental induction hypothesis that goes from these six things to
block!
Remember ? This was the first reason we said it was going to \\(x,y,z).
be hard to prove x + y is a number, because we thought A. But uh-oh. it doesn't work. Look, lV(x, y, z) depends on
(T13)
depended on After learning that pseudo-numbers satisfy
this. IV(2, z L , y), which depends on IV (y R , y, z), which depends on
the transitive laws, we forgot to reconsider the original IV(z, z L y) again; we're in a loop. It's the same stupid prob-
,

source of trouble. lem I noticed before, and now we know it's critical.

B. So at least this big picture method is getting us B. (pounding the dirt) Oh no! Well, there's one more thing
somewhere, . . .

if only because it helps organize our

thoughts. I'll try before giving up. Let's go all tho way and prove a more
Now two down and one
it's to go. The proof of lift*,
general version of (T13):
y, z)
depends on knowing < and
V. if .x- x' y < y',

then x + y <, x' + y'.

U(X L ,y,z),
H(», YR , z). This is what we really are using in our proofs, instead of
doing two steps with (T13). And it's symmetrical; that might
A. Again. [(.<:. //) wasn't required. So we
can simply prove (T13) help.
and (T14) without worrying whether or not x
+ y is a number. A. We'll also need a converse, generalizing (T14).
B. I see— then later, z + y will turn out to be a number, because B. I think what we need is
of (TI3) and (T14). Great!
VI. if x + y > x' + y' and y < y\
A. Now and III depend on each other, so wo can combine
II
then x > x'.
them into a single statement like we did before.
A. Your notation, primes and all. looks very professional.
B. Good point. Let's sec, if I write IV(a;, y, z) to stand for the
combined statement II(.c, y, z) and III my B. (concentrating) Thank you. Now the proof of V(x, x', y, y')
(x, y, z), lists show
that lV(x, y, z) depends on depends on

VI(X t *', y, y% ,

IV(//.A- ;., Z ), IV(s, y, ZL ), IV(z, ZL , y), Vl(yt ,y', *,*'),

iv(y„,*, Z ), IV(*, y, ZR ), IV(2 z, x).
fi , V\(x,X'R ,y,y'),
mx L ,y,z), IV(x,YR ,z).
V%, Y'R , x, x').

86
87
 !

Hey, this is actually easier than the other one. the symmetry
is helping. 14
Finally, to prove VI (*, x', we need
y, y'), . . . the suspense is
killing me, I can't think . . .

V(x, XI y. y'), V (A' fi

, *', y, y').

A. (jumping up) Ixiok, a day-sum argument,

applied to the com-
bination of V and VI, now finishes the induction
!

B. (hugging her) We've won

A. Bill, I can hardly believe it, but our proof of these tun state-
ments actually goes through for all pseudo-numbers x, x' , y,
and y'.
B. Alice, this has been a lot of work,
but it's the most beautiful
thing I ever saw.

A. Yes, we spent plenty of energy on what we

both took for
granted yesterday.

I Conway himself had a simpler way to prove those

wonder if

laws? Maybe he did, but even so I like ours

because it taught
us a lot about proof techniques.
B. Today was going to be the day we studied multiplication.
A. We'd better not start it now, it might ruin our sleep •
again.
Let's just spend the rest of the afternoon
working out a proof
that -a: is a number, whenever x is.

B. Good idea, that should be easy now. And I

wonder if we can
prove something about the way negation acts
on pseudo-
numbers ?

88
I 14 THE UNIVERSE

.•

k
m
„,.w.

B. (stretching) Good morning, love; did you think of any more

mistakes in our math, during the night?

A. No, how about you ?

B. You lemma I never look for mistakes. But a thought did strike
me: Here we're supposed to have rules for creating all the

numbers, but actually \ never appears. Remember, 1 was

91
 !

expecting to see it on the "fourth day." hut the number B. After infinitely many days. I guess (.'on way looked out over
turned out to ho kind of thought, well.
j. |
J a little slow
is all those binary numbers he had created, and . . . Omigosh!
>" arriving, but it will get here sooner or later. Just now it I bet he didn't stop.
struck me that we've analyzed all the numbers, but still
,]
A. You're right I never though* of it before, but the stone does
has never showed. I

seem to say he went right on. And . . . sure, he gets more

A. All the numbers that are created
have a finite representation numbers, too, because for the first time he can choose X L and
in the binary number
system. I mean like 3| = 11.101 in X R to be infinite sets!
binary. And on
the other hand, every number with
a finite B. Perhaps time doesn't How at a constant rate. I mean, to us the
binary representation does get
created, sooner or later. Like, days seem like they're of equal length; but from Conway's
3| was created on the eighth day. . . .
point of view, as he peers into our universe, they might be
B. Binary numbers are used on computers. Maybe
Conway was going faster and faster in some absolute extra-celestial time
creating a computerized world.
scale. Like, the first earth day lasts one heavenly day, but
What's the binary representation of the second earth day lasts only half a heavenly day, and the
J anyway?
A. I don't know, but it must next is one fourth, and so on. Then, after a total of two
have one.
heavenly days, zap! Infinitely many earth days have gone
B. Oh, I remember, you sort of
do long division but with base by, and we're ready to go on.
2 instead of 10. Let's see ... I get
A. I never thought of that, but it makes sense. In a way, we're
\ = .0101010101 ... now exactly in Conway's position after infinitely many earth-
days went by. Because we really know everything that tran-
and so on ad infinitum. It doesn't terminate, that's spired, up until X day!
why it
wasn't created.
B. (gesticulating) Another plus for mathematics: Our finite minds
A. "Ad infinitum." That reminds
me of the last part of the can comprehend the infinite.
inscription. What do you suppose the rock means A. At least the countably
about X infinite.
day and all that ?
B. But the real numbers are uncountable, and we can even com-
B. It sounds like some metaphysial or rc igious praise of the
| prehend them.
number system to me. Typical of ancient
writings. A. I suppose so, since every real number is just an infinite decimal
On the other hand, it's sort of strange that expansion.
Conway was still
around and talking, after infinitely
many days. "Till the end B. Or binary expansion.
of time," but time still hadn't
ended.
A. You're in great voice today.
A. Hoy !
I know now what happened on X day —the real numbers
were all created

92
93
 — !

B. (eyes popping) Migoah, I believe you're

right. A. (not listening) Bill, there's another number also created on
A. Sure, we get £ by taking X L to be, say, X day, a number that's not in the real number system. Take
XR to be empty, and
{.01, .0101, .010101, .01010101, . . .
}
.Y, ={1,2,3,4,5,...}.
in binary notation, and XR would be numbers that, get closer
Tl lis number is larger than all the others.
and closer to £ from above, like
B. Infinity ! Outa sight
{.1, .011, .01011, .0101011, .010101011
}. A. I think I'll denote it by the Greek letter o> since I always
liked that letter. Also —to was created, I mean minus infinity.
B. And number like n gets created in roughly the same way.
a
B. X day was a busy, busy day.
I don't know the binary representation ofTr,
but let's Bay Jt'a
A. Now the next day
77 = 11.00100100001111 . . .
; B. You mean X wasn't the end!

A. Oh no, why should Conway stop then ? I have a hunch he

we get FI t by stopping at every "1,"
was only barely getting started. The process never stops, be-
ll L = {11.001, 11.001001. 11.00100100001,
cause you can always take XR empty and X L to be the set
. . .
}
of all numbers created so far.

and II fl by stopping at every "0" and adding

1,
B. But there isn't much else to do on the day after X, since the
real numbers fit together so densely. The noninfinite part of
UR = {11.1, 11.01, 11.0011, 11.00101, . . .
}.
the universe is done now, since there's no room to put any
more numbers in lidween two "adjacent" real numbers.
A. There are lots of other sets that could be used for U L and n fi , A. Xo, Bill; that's what / thought too, until you said it just now.
infinitely many in fact. But they all produce numbers equiva-
lent to this one,
I guess it just proves I like to argue with you. How about
because is the first number created that
it
taking A',. =
and {0} XR = {I . !.. ',. ,!,... }. It's a number
is greater than n t and less than Il fi .
j

greater than zero and less than all positive real numbers! We
H. (hugging her again) So might
what the Conway Stone means
that's call it e.

When it says the universe was created on

X day: the real B. (fainting) Ulp . . . That's okay, I'm all right. But this is almost
numbers are the universe.
too much; I mean, there's gotta be a limit.
Have you ever hoard of the "big bang" theory
the cosmolo- What surprises me most is that your number e was actually
gists talk about? This is it, X day: Bang! created on X day, not the day after, because you could have

94
95
 taken X„ = {1, |, J, |, ^ . , .
J.
Also, there are lots of other
crazy numbers in there, like
15
(OMH.li, li.lA,...})

which is just a hair bigger than 1.

And I suppose there's a number like this right next to all

numbers, like n . . . no, that can't be . . .

A. The one just greater than n doesn't

come until the day after
X. Only terminating binary numbers
get an infinitely close
neighbor on X day.
B. On the day after X we're also going to get a number between
and e. And you think Conway was just getting started.
A. The neatest thing, we not only have the real
Bill, is that
numbers and infinity and all the in-betweens ...
we also have
rules for telling which of two numbers is larger, and for adding
and subtracting them.

B. That's right. We proved all these rules, thinking we already

feu id what we were proving— it was
just a game, to derive
all the old standard laws of arithmetic from
Conway's few
rules. But now we find that our
proofs apply also to infinitely
many unheard-of cases! The numbers are limited
only by our
imagination, and our consciousness is
expanding, and . .

A. You know, all this is sort of like a religious

experience for me;
I'm beginning to get a better appreciation
of God. Like He's
everywhere . . .

B. Even between the real numbers.

A. C'mon, I'm serious.

96
15 INFINITY

,..
••

m,
&
^mm

B. I've been doing a few calculations with infinity. Like, rule

(4) tells us immediately that

to + 1 - ({a,, 2, 3, 4, 5, . . .
}, 0),

which simplifies to

to + 1 m ({to}, 0).

99
A. That was created on the day after X day.
I guess we can call this 2co, even though we don't have multi-
B. Right, and plication yet, because we'll certainly prove later on that

(z + y)z = xz + yz. That means 2z m (1 + 1)2 z + z.

co + 2 = ({co + 1}, 0)
A. Right, and
on the day after. Also,
3co = ({2co + 1, 2<o + 2, 2co + 3, 2co + 4, . . . }, 0)

« + I - ({«}, {« + !})•
will be created on (3X) day, and so on.

A. What about co - 1? B. We still don't know about multiplication, but I'm willing to
bet that co times will turn out to be
B. co — 1! I never thought of subtracting from infinity, because co

a number less than infinity is supposed to be finite. But, let's co

2
= ({<u, 2co, 3co, 4tu, . . . 0).
},
grind it out by the rules and see what happens . . . Look
at that, A. Created on X2 day. Just imagine what Conway must be doing
to all the smaller numbers during this time.
co - 1 ({1, 2, 3, 4, . . .
}, {co}).
B. You know, Alice, this reminds me of a contest we used to
Of course — it's the first number created which is larger than have on our block when I was a kid. Every once in a while

all integers, yet less than co. we'd start shouting about who knows the largest number.
1'ivtty soon one of the kids found out from his dad that
A. So that's what the Stone meant about an infinite number less
infinity was the largest number. But I went him one better
than infinity.
by calling out "infinity plus one." Well, the next day we got
Okay, I've got another one for you, what's co + w? up to infinity plus infinity, and soon it was infinity times
B. Easy: infinity.

CO + IT = (CO + I!/, CO + Tl R ).
A. Then what happened?
B. Well, after reaching "infinityfinityfinityfinity . .
." repeated as
This was created on . . . (2X) day ! And so were co + e and long as possible without taking a breath, we sort of gave up
co — e. the contest.

A. Oho! Then there must also be a number 2co. I mean, co + co. A. But there are still a lot more numbers left. Like
B. Yup, "»" = ({«. "2 , CO
3
, CO*,... }, 0).

co + co = ({co + 1, co + 2, co + 3, co + 4, . . .}, 0).

And still we're only at the beginning.

100
101
B. You mean, there's a/ ", to°> °",
and the limit of this, and so on. in XL . If this assumption is false, then xLL + hasn't been

Why didn't I think of that when I was a kid? proved equal to x lL for some x LL ; or, I guess, some x LR might
be the culprit. Any counterexample would imply an infinite
A. It's a whole new vista . . . But I'm afraid our proofs aren't
sequence of counterexamples.
correct any more, Bill.
A. All we have to do now is show that there is no infinite ancestral
B. What ? Not again. We already fixed them.
sequence of numbers.
Oh-oh, I think I see what you're getting at. The day-sums.
-
Xi, .x 2, x$, a;
A. Right. We can't argue by induction on the day-sums because 4, . . .

they might be infinite.

such that x, + is in X U X, R .
j IL

B. Maybe our theorems don't even work for the infinite cases. B. That's a nice way to put it.

It sure would be nice if they did, of course.

mean, what a I
A. Also, it's true, because every number (in fact, every pseudo-
feeling of power to be proving things about all these numbers
number) is created out of previously created ones. Whenever
we haven't even dreamed of yet.
we create a new number x, we could prove simultaneously
A. We didn't have any apparent trouble with our trial calcula- that there is no infinite ancestral sequence starting with
tions on infinite numbers. Let me think about this for awhile. xx = x, because we have previously proved that there's no
infinite sequence that proceeds from any of the possible
choices of x 2 in X L or X R .

B. That's logical, and beautiful . . . But it almost sounds like

It's okay, I think we're okay, we don't need "day-sums." you're proving the validity of induction, by using induction.

B. How do you manage it ?

A. I suppose you're right. This must actually be an axiom of
A. Well, remember how we first thought of induction in terms some sort, it formalizes the intuitive notion of "previously

of "bad numbers." What we had to show was that if a theo- created" which we glossed over in rule (1). Yes, that's it,

rem fails for x, say, then it also fails for some element xL in rule (1) will be on a rigorous footing if we formulate it in this

X L and then it also fails for some xu in X LL and so on. But

, ,
way.
if every such sequence is eventually finite; I mean if eventually B. What you've said covers only the one-variable case. Our day-
we must reach a case with X IL L empty, then the theorem sum argument has been used for two, three, even four vari-
can't have failed for .t in the first place. ables, where the induction for [x, y, z) relies on things like

B. (whistling) I see. For example, in our proof that x + = x, (y, z, x L ) and so on.
we have a: + = (X L + 0, A' R + 0). We want to assume by A. Exactly. But in every case, the induction went back to some
induction that xL + has been proved equal to xL for all xL permutation, of the variables, with at least one of them getting

102 103
 an additional L or Ii subscript. Fortunately, this means that B. Let's sec, n + it = (n + Y\ L , n + II R ), which . . . Okay, I see,

there can't be any infinite chain such as there arc infinitely many branches of the calculations but they
all terminate after finitely many steps.
(x, y, z) -> (y, z, x L ) -* (z R . y, x L ) -+ . . .
,
A. The neat thing about the kind of induction we've been using

and so on; if there were, at least one of the variables would is that we never have to prove the "initial case" separately.

have an infinite ancestral

The way I learned induction, we always had to prove P(l)
chain all bv itself, contrary to rule
first, or something like that. Somehow we've gotten around
(I).
this.
B. (hugging her once again) Alice, I love you, in infinitely many
ways. B. You know. 1 think I understand the real meaning of induction
for the first time. And I can hardly get over the fact that all
A. (giggling) "How do I love thee I Let me count the ways." and
our theory is really valid, for the infinite infinitesimal
1, to, co
2
, co
u a/ '"'"
,
numbers as well as the finite binary ones.
B. It seems that we have gotten around this infinite con-
still
A. Except possibly (T8), which talks about the "first number
struction in a sneaky and possibly suspicious way. Although
created" with a certain property. We'd have to fix up a defi-
I can't see anything wrong with your argument, I'm still nition of what that means ... I suppose we could assign a
leery of it.
number to each day, like say the largest number created on
A. As I see it, the difference is between proof and calculation. that day, and order the days that way . . .

There was no essential difference

in the finite case, when we
B. I sort of follow you. I've noticed that a number seems to be
were just talking about numbers created before day X. But
the largest created on its day when XR is empty and X, is
now there is a definite distinction between proof and the
all the previously created numbers.
ability to calculate. There are no infinite ancestral sequences,
but they can be arbitrarily long, even when they start with A. Maybe that explains why there was X day and (X + 1) day,

the same number. For example, <o, », n - but no (X - 1) day.

1, is a se- 1

quence of ancestors of to, for all n. B. Yeah, I guess, but this is all too deep for me. I'm ready to

B. Right. I've just been thinking about the ancestral sequences tackle multiplication now, aren't you ?
2
of co . They're all finite, of course; but they can be so long, the
liniteness isn't even obvious.
A. This unbounded liniteness means that we can make valid
proofs, for example, that 2 X it m n + ir, but we can't neces-
sarily calculate u + «• in a finite number of steps. Only God
can finish the calculations, but we can finish the proofs.

104 105
MULTIPLICATION

A. Let's see that paper where you wrote down Conway's rule
for multiplication. There must be a way to put it in symbols
. . . Hmm, wc already know what he means by "part of the

same kind."

B. Alice, this is too hard. Let's try to invent our own rule for

multiplication instead of deciphering that message.

107
 Why don't we just do like he did for addition. I mean, xy
A. That's it, the product of positive numbers must be positive!
should lie between X,j/ \JxYt and X Ry
Uil'„. At least, it
and
ought to do this
The other three conditions for xy to lie between its left
when negative numbers are excluded.
right sets are essentially saying that
A. But that definition would be identical to addition, so the
product would turn out to be the same as the sum. (*r ~ x)[y R - y) > 0,

(* - xL )(yR - y) > 0,
B. Whoops, so it would ... All right, I'm ready to appreciate
{xR - x)(y - yL ) > 0.
Conway's solution, let's look at that paper.

A. Don't feel bad about it, you've got exactly the right attitude. Okay, the definition looks sensible, although we haven't
Remember what we said about always trying things first ? proved anything.

B. Hah, 1 guess that's one lesson we've learned. B. Before we get carried away trying to prove the main laws

A. The best can make out about multiplication, I want to check out a few simple cases
I is that Conway chooses the left set
of xy to be all numbers of the form just to make sure. Let's see . . .

xy = yx; (T20)
Xi2) + xyL - *dfc or x^ + xyR - x RyR ,

0y = 0; (T21)
iy = y. (T22)
and the right set contains all numbers of the form

Those were all easy.

xtfl + xy R ~ XlUr or x R y + xyL - XRy L .

A. Good, zero times infinity is zero. Another easy result is

You see, the left set gets the "same kinds" and the right set
gets the "opposite kinds" of parts. Does this definition make -(xy) = (-*),/. (T23)

any sense ?
B. Right on. Look, here's a fun one:
B. I.enimc see, it looks weird. Well, xy is supposed to be greater
than its left part, so do we have \x = (l_X L U(Z- U' fi ),
(x - JZL U ) JA' fi ). (T24)

xy > xtf + xyL - x,y L ! A. Hey, I've always wondered what half of infinity was.

B. Half of infinity! . . . Coming right up.

This is like . . . yeah, it's equivalent to

- \u>m ({1,2,3,4,...},
(* xL )(y - yL ) > 0.
{(o - 1, cu - 2, w - 3, to - i, ... }).

108
109
 ! — !

It's interesting to prove that |tu + Jw = o> . . Wow, here's A. Hmm . . . What about division ? ... I bet if x is between
another neat one:
and 1, it'll be possible to prove that

eta a 1,
1
1 -
1 + x
Our infinitesimal number turns out to be the reciprocal of
infinity ({x, X - X2 + X3 X
,
- X2 + x3 - X* + X5 },

{x - X2 X - X2 +
, X3 - X*, . . . }).
A. While you were working that out, I was looking at multipli-
cation in general. It looks a freaky for pseudo-numbers
little
At least, this is got J, for x = i. Perhaps we'll be able
how we
I found a pseudo-number p for which ({1}, Q)p is not like
to show that every nonzero number has a reciprocal, using
({0, 1}, Q)p, even though ({1}, 0) and ({0, 1}, 0) are both equal
some such method.
to 2. In spite of this difficulty, I applied your Big Picture
method and B. Alice! Feast your eyes on this!
I think it is possible to prove

x{y + z) = xy + xz, (T25) V^ S ({l,2,3,4,...},| ,-,-,

x(yz) = (xy)z (T26)
T ? ...}).
for arbitrary pseudo-numbers, and V^({ e ,2 e ,:k,4 6
,...},{j,H
-J,...}).

if x > x' and y > y'

A. (falling into his arms) Bill! Every discovery leads to more,
then (x - x')(y - y') > (T27)
and more

for arbitrary numbers. It will follow that xy is a number B. (glancing at the sunset) There are infinitely many things yet

whenever x and y are. to do and only a finite amount of time

. . .
. . .

B. Theorem (T27) can be used to show that

if x = y then xzmyz (T28)

for all numbers. So all of these calculations we've been making

arc perfectly rigorous.

I guess that takes care of everything

it says on the tablet.

Kxcept the vague reference to "series, and quotients, and

roots."

110 111
 POSTSCRIPT
Tho reader may have guessed that this is not a true story. The late Hungarian mathematician Alfred Renyi composed three
However, "J.BL W. H. Conway" does exist -he is I'rol'rssor stimulating "Dialogues on Mathematics," which were published by
John Horton Conway of Cambridge University. The real Holdeii-Day of San Francisco in 1!)67. His first dialogue, set in ancient
Conway has established many remarkable results about t bese
Greece about 440 B.C., features Socrates and gives a beautiful descrip-
rxtraordinal" numh.rs, beaidee wlml has been mentioned tion of the nature of mathemat ics. The second, which supposedly takes
here. For example, every polynomial of odd degree, with place ill 212 u.c, contains Archimedes' equally beautiful discussion of
arbitrarynumbers as coefficients, has a root. Also, every the applications of mathematics, Renyi's third dialogue is about math-
pseudo-number p corresponds to a position in a two-person ematics and science, and we hear Galileo speaking to us from about
game between players Left and Wight, where the four A.D. 1600.
relations have prepared Surreal Numbers as a mathematical dialogue of the
I
1970's, emphasizing the nature of creative mathematical explorations.
p > 0, P < 0, Of course, I wrote this mostly for fun, and I hope that it will transmit
P = 0, P °II
some pleasure to its readers, but I must also admit that I also had a
correspond respectively to the four conditions serious purpose in the back of my mind. Namely, I wanted to provide
some material which would help to overcome one of the most serious
I.'fi wins. Right wins.
shortcomings our present educational system, the lack of training
in
Second player wins, First player wins.
for research u oikhere is comparatively little opportunity for students
: t

starting at position p. The i heory is still very much in its

to experience how new mathematics is invented, until they reach
may wish to play
infancy, ami the reader with some of the graduate school.
many unexplored topics: What can be I decided that <i rat ivity can't be taught using a textbook, but that
said about log-
arithms! continuity! multiplicative properties of pseudo- an "anti-text" such as this novel might be useful. I therefore tried to
numbers? generalized diophantine equal ions* write the exact opposite of Landau's Grundlayen der Mathemalik; my
otc.
aim was to show how mathematics can be "taken out of the classroom
and into life, and to urge the reader to try his or her own hand at
exploring abstract mathematical ideas.
The best way to communicate the techniques of mathematical
research probably to present a detailed case study. Conway's recent
is

approach to numbers struck me as the perfect medium for illustrating

the important aspects of mathematical explorations, because it is a
rich theory that is almost self-contained, yet with close ties to both
algebra and analysis, and because it is still largely unexplored.
In other words, primary aim is not really to teach Conway's
my
theory, it is to teach how one might go about developing such a theory.
Therefore, as the two characters in this book gradually explore and
build up Conway's number system, I have recorded their false starts
and frustrations as well as their good ideas. I wanted to give a reason-

112 113
 ^
ably faithful portrayal of the important principles, techniques, joys,
4. After Chapter 6. When we are developing the theory of Conway's
passions, and philosophy of mathematics, so I wrote the story as I was
numbers, from these few axioms, is it legitimate to be using the
actually doing the research myself (using no outside sources except a
properties we already "know" about numbers, in the proofs?
vague memory of a lunchtime conversation I had had with John
(For example, the use of subscripts like i — 1 and j + 1, and so
Conway almost a year earlier).
I have intended this book primarily for college mathematics
on.) [Warning: This may lead to a discussion about mctamathe-
matics for which the instructor may have to be prepared.]
students at about the sophomore or junior level. Within a traditional
math curriculum it can probably be used best either (a) as supplemen- 5. After Chapter 9. Find a complete formal proof of the general pat-
tary reading material for an "Introduction to Abstract Mathematics" tern after n days. [This makes an instructive exercise in design of
course or a "Mathematical Tx)gic" course; or (b) as the principal text notations. There are many and the students should
possibilities,

in an undergraduate seminar intended to develop the students' strive to find a notation that makes a rigorous proof most under-
abili-
ties for doing independent work. standable, in that it matches Alice and Bill's intuitive informal
Books which arc used in classrooms usually are enhanced by exer- proof.]

cises; so at
the risk of destroying the purity of this "novel" approach, 6. After Chapter 9. fa there a simple formula telling the day on which
I have compiled a few suggestions for supplementary problems. When a given binary number is created ?

used with seminars, such exercises should preferably be brought up

7. After Chapter 10. Prove that x = y implies — x = — y.
early in each class hour, for spontaneous class discussions, instead of
being assigned as homework. 8. After Chapter 12. Establish the value of x © y for as many x, y as
you can.
1. After Chapter 3. What is "abstraction,*' and what is "generaliza-
tion"? 9. After Chapter 12. Change rules (1) and (2), replacing > by < in

all three places; and add a new rule:

2. After Chapter 5. Assume that g is a function from numbers to num-
bers such that x < y implies g(x) < g(y). Define x < y if and only if x < y and .'/ $ x-

m=(f(XL)V{g(x)}J(X R )).
Now develop the theory of Conway's numbers from scratch, using
Prove that/(.r) < J(y) if and only if a: < y. Then in the special case these definitions. [This leads to a good review of the material in
that g(x) is identically 0, evaluate /(a:) for as many numbers as you the first chapters; the arguments have t<> be changed in several
can. [Note: After Chapter 12, this exercise
makes sense also when places. The major hurdle is to prove x < x for all numbers; there
"numbers" are replaced by "pseudo-numbers."] is a fairly short proof, not easy to discover, whieh I prefer not to
3. After Chapter Let x,y be numbers whose
5. left and right parts are reveal here. The students should be encouraged to discover that
"like" but not identical. Formally, let the new < relation is not identical to Conway's, with respect to
pseudo-numbers (although of course it is the same for all numbers).
fi.'- Xl-*Yu Sr- Xk
In the new case, x < x does not always hold; and if
Ul-Y l gR :YB -+Xa
be functions such that/jzj xL ,fR (x R = xR> gL(yL ) m yL gR(yR )
)
* = ({({0}, {0})}, »),
,

m y R Prove that x m y. [Alice

. and Bill did not realize that this we have x = in Conway's system but x s 1 in the new one!
lemma was important in some of their investigations, they assumed
Conway's definition has nicer properties but the new relation is
it without proof. The lemma holds also for pseudo-numbers.]
instructive.]

114 115
 Show how 17. After Chapter Hi. Call x a real number if —n < x < n for some
10. After Chapter 13. to avoid Alice and Bill's circularity
problem another way, by eliminating 111(2, Z L y) and l\l(Z R z, x)
(nongeneralized) integer n, and if
, ,

from the requirements needed to prove II(:e, y, z). In other words, x=({z- l,x- lx- I ...},{x +l,x+ l,x+ {, .. .}).
prove directly that we can't have z + y < zL + y for any 2,,.
Prove that the real numbers are closed under addition, subtrac-
11. After Chapter 14. Determine the "immediate neighborhood" of tion,and multiplication, and that they are isomorphic to real
each real number during the first few days after X day. numbers defined in more traditional ways. [This exercise and those
12. After Chapter 15. Construct the largest infinite numbers you can which follow were suggested by John Conway.]
think of, and also the smallest positive infinitesimals. 18. After Chapter 16. Change rule (1), allowing {X L , XR ) to be a number
13. After Chapter 15. Does it suffice to restrict A',, and X„ to countable only when X £ X R and the following
L condition is satisfied:

sets? [This is difficult but it may lead to an interesting discussion. XL has a greatest element if and only if XH has a
The instructor can prepare himself by boning up on ordinal num- least element.
bers.]
Show thai precisely the real numbers (no more, no less) are created
14. Almost anywhere. What are the properties of the operation de- in these circumstances.
fined by
19. After Chapter 16. Find a pseudo-number p such that p +p = ({0},
xoy = (X L nY L ,X R vY R )1 {0}). [This is surprisingly difficult and it leads to interesting sub-
problems.]
[The class should discover that this is not min(a:, y)\ Many other
20. After Chapter 15 or The pseudo-number ({0}, {({0}, {0})}) is >0
16.
operations are interesting to explore, e.g., when x ° y is defined to
and <x numbers x. It's really infinitesimal! But
for all positive
be
({0}, {({0}, { — 1})}) is smaller yet. And any pseudo-number p >

(A, Y»X R uYR )

is > ({0}, {({0}, {—*))}) for some suitably large number x.

or (X L i

y uxoY L ,X K uYR ) 21. After Chapter 16. For any number x define

and so on.] wx = ({0} U {ttort \xL eX L , n = 1,2,3, . .

.},

15. After Chapter 16. If X is the set of all numbers, show that (X, 0)
is not equivalent to any number. [There are paradoxes in set (p«* I «*sXjM n = l,2,3,...n.
theory unless care is taken. Strictly speaking, the class of all

numbers isn't a set. Cf. the "set of all sets"' paradoxes.] Prove that m'u" = mz * v .

Hi. After Chapter 16. Call x a generalized integer if 22. After Chapter Explore the properties of the symmetric pseudo-
16.

numbers S such that

X E3 ({X - 1}, {* + 1}).

(
PL P R )eS
, if and only if P L = PR £ 5.
Show that generalized integers are closed under addition, subtrac-
In other words, the elements of S have identical left and right sets,
tion, and multiplication. They include the usual integers », as
as numbers like <u + n, £u>, etc. [This exercise is due
and so do the elements of their left and right parts. Show that S
well to Simon
is closed under addition, subtraction, multiplication. Explore fur-
Norton.]

116 117
 ther properties of S (e.g., how many unlike elements of S are as well as on the mathematical content; say 50-50. They must be told
created on each day, and is their arithmetic interesting in any that a math term paper should not read like a typical homework paper.
way?). [This open-ended problem is perhaps the best on this list, The latter is generally a collection of facts in tabular form, without
because there is an extremely rich theory lurking here.] motivation, etc., and the grader is supposed to recognize it as a proof;

I will send hints to the solutions of exercises 9, 19, and 22 to any the former is in prose style like in math textbooks. Another way to

bona fide teachers who request them by writing to me at Stanford provide experience in writing is to have the students take turns pre-

University. paring resumes of what transpires in class; then all the other students

Now I would like to close this postscript with some suggestions willbe able to have a record of the discussions without being distracted
addressed specifically to teachers who will be leading a seminar based by taking notes themselves. In my opinion the two weaknesses in
on this book. (All other people, please stop reading, and close the book our present mat hematics education are the lack of training in creative
at once.) thinking and the lack of practice in technical writing. I hope that

Dear Teacher: Many topics for class discussion are implicit in the the use of t his little book can help make up for both of these deficiencies.

story. The first few chapters will not take much time, but before long Stanford, California D.E.K.
you may well be covering less than one chapter per class hour. It may May 1974
be a good idea for everyone to skim the whole book very quickly at
first, because the developments at the end are what really make he- I

beginning interesting. One thing to stress continually is to ask the

students to "distill off" the important general principles, the modus
operandi, of the characters. Why do they approach the problem as
they do, and what is good or bad about their approaches? How does
Alice's "wisdom"
differ from Bill's? (Their personalities air distinctly

different.) Another ground rule for the students is that they should
check over the mathematical details which arc often only hinted at;
this is the only way they can really learn what is going on in the book.
Preferably they should tackle a problem first themselves before reading
on. When an ellipsis, i.e., " . .
.

" appears, this oftenmeans the charac-

ters were thinking (or writing), and the reader should do the same.
When holding class discussions of such exercises as these, I have
found it a good rale t<> limit t he number of times each person is allowed
to speak up. This keeps the loquacious people from hogging the floor
and ruining the discussion; everybody gets to participate.
Another recommendation is that the course end with a three or
four- week assignment, to write a term paper that explores some topic
not explicitly worked out in the book. For example, the open-ended
exercises in the above list illustrate several possible topics. Perhaps
t he si udents can do their research groups of two. The students should
in

also be told that they will be graded on their English expository style

118 119
 \

512.

Dm

ADDISON-WESLEY PUBLISHING COMPANY

Reading, Massachusetts • Menlo Park, Cai
<lon • Amster Ontario •
Sydney