P. 1
Carl D Olds - Continued Fractions

Carl D Olds - Continued Fractions

|Views: 694|Likes:
Published by gsbs58

More info:

Published by: gsbs58 on Oct 28, 2010
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less






published by The Mathematical Association of America

Editorial Comrnittee Basil Gordon, Chairman (1975-76) University of California, L.A, Ivan Niven (1975-77) M. M. Schiffer (1975-77) Anneli Lax, Editor New York University
UniverGty of &egm Stanford UniversZty

The New Mathematical Library (NML) was begun in 1961 by the School Mathematics Study Group to make available to high school students short expository books on various topics not usually covered in the high school syllabus. In a decade the NML matured into a steadily growing series of some twenty titles of interest not only t o the originally intended audience, but to college students and teachers at al1 levels. Previously published by Random House and L. W. Singer, the NML became a publication series of the Mathematical Association of America (MAA) in 1975. Under the awpices of the MAA the NML will continue to grow and will remain dedicated to its original and expanded purposes.


C. D. Olds
8an Jose Stats Uniuersity


Illustrations by Carl Bass

Ninth Printing @ Copyright, 1963, by Yale University
Al1 rights reserved under Intemational and Pan-American Copyright

Conventions. Published in Washington by the Mathematical Association of America Library of Congress Catalog Card Number : 6 1-1 21 85 Manufactured in the United States of America

sections containing thoroughly familiar material may be read very quickly. New York. he will have to make an intellectual effort. even within a single book. some of which may require considerable thought. The reader is urged to acquire the habit of reading with paper and pencil in hand. they vary in difficulty. If the reader has so far encountered rnathematics only in classroorn work. Y. N. NEW YORKUN~VERSITY. in this way mathematics will become increasingly meaningful to him. 10012.cannot be read quickly. Most of the volumes in the New Mathematical úibrary cover topics not usually included in the high school curriculum. Street. New Mathematical Library. Editor. while the reader needs little technical knowledge to understand most of these books. The authors and editorial committee are interested in reactions to the books in this series and hope that readers will write to: Anneli Lax. On the other hand. often an argument will be clarified by a subse quent remark. THECOURANT INSTITUTE OF WTHENATICAL251 Mercer SCIENCES.Note to the Reader his book is one of a series written by professional rnathematicians in order to malte some important mathematical ideas interesting and understandable to a large audience of high school students and laymen. The best way to learn mathematics is to do mathematics. some parts require a greater degree of concentration than others. and. Thus. Nor must he expect to understand al1 parts of the book on first reading. He should feel free to skip complicated parts and return to them later. and each book includes problems. he should keep in mind that a book on mathemati. The Editors T .

Magnus . Based on the EGtvGs 1 1 12 Competitions 1894-1905 and 1906-1928. E. Sal ind 18 FIRST CQNCEPT'S OF TOPOLOGY by W. M. Coxeter and S L Greitzer . Compiled and with solutions by Charles T. J. D. Salkind and J. Yaglom. Steenrod 19 GEOMETRY REVlSlTED by H. S M. Chinn and N. l 'i Other titles I n preparatlon . 2 0 INVITATION TO NUMBER THEORY by Oystein Ore 21 GEOMETRIC TRANSFORMATIONS ll by l M. Aaboe 14 GROUPS AND THEIR GRAPHS by l Grossman and W. S a ~ y e f 3 AN INTRODUCTION TO INEQUALITIES by E F. Bellman . Rapaport 13 EPISODES FROM THE EARLY HISTORY OF MATHEMATICS by A. 25 THE CONTEST PROBLEM BOOK 11 Annual Hlgh School Contests 1966-1972 1 Complled and with solutions by C T. Sheni I . Beckenbach and R. Shields . Davls 7 USES OF INFlNlTY by Leo Zippin 8 GEOMETRIC TRANSFORMATIONS by l M. O. Friedrichs 17 THE CONTEST PROBLEM BOOK I Annual Hi h School Contests 1961-1965. . 2 WHAT IS CALCULUS ABOUT? by W. translated by E. translated by A. 9 CONTINUED FRACTIONS by Carl D. W. Yaglom. translated by A. Kazarinoff 5 THE CONTEST PROBLEM BOOK 1 Annual High School Contests 1950-1960. translated by A. I Compiled and with solutions by Charles T. 4 GEOMETRIC INEQUALITIES by N. Sinkov 23 INGENUITY IN MATHEMATICS by Ross Honsberger 24 GEOMETRIC TRANSFORMATIONS I Iby 1 M.NEW MATHEMATICAL LIBRARY 1 NUMBERS: RATIONAL AND IRRATIONAL by lvan Niven. Salkind 6 THE LORE OF LARGE NUMBERS by P. Olds 10 GRAPHS AND THEIR USES by Oystein Ore 1 HUNGARIAN PROBLEM BOOKS Iand 1. Shlelds . Yaglom. Earl . G. 22 ELEMENTARY CRYPTANALYSIS-A Mathematical Approach by A. 15 THE MATHEMATICS OF CHOICE by luan Niven 16 FROM PYTHAGORAS TO ElNSTElN by K.

2 2.5 2.5 3.4 2.4 Expansion of Rational Fractions Introduction Definitions and Notation Expansion of Rational Fractions Expansion of Rational I'ractions (General Discussion) 1.by = f_ 1 The General Solution of ax . and Monkeys 31 31 32 36 1 42 1 44 46 48 Chapter 2 2.7 Some Historical Co mments Diophantine Equations Introduction The Method Used Extensively by Euler The Indeterminate Equation ax . b ) The General Solution of Ax z: By = +_ C Sailors.3 2.9 + = = Expansion of Irrational Numbers 51 Introduction 51 Preliminary Examples 52 Convergents 58 Additional Theorems on Convergents 61 63 Some Xotions of a Limit Infinite Continued Fractions 66 Approximation Theorems 70 Geometrical Interpre tation of Continued Fractions 77 80 Solution of the Equation x2 = ax 1 81 3.3 3.2 3.Contents Pref ace Chapter 1 1.6 3.7 Chapter 3 3.3 1.8 3.1 1.4 3.6 Diff erences of Convergents 1.2 1. b) The General Solution of a x by = c.1 2.bu = c. Coconuts.5 Convergents and Their Properties 1.1 t A Method for CaJculating Logarithms + . (a.1 3. (a.10 Fibonacci Numbers 84 3.6 2.7 3.

4 Periodic Continued Fractions Introduction Purely Periodic Continued Fractions Quadratic Irrationals Reduced Quadratic Irrationals Converse of Theorem 4.1 Lagrange's Theorem The Continued Fraction for 4 Ñ Pell's Equation.5 4.6 4. x2 . Appendix 11.7 4.3 5.1 Has No Integral Solutions Appendix 1.Ny2 = 4 1 How to Obtain Other Solutions of Pell's Equation Epilogue Introduction Statement of the Problem Hurwitz' Theorem Conclusion Proof That x2 .4 4.1 4.viii CONTENTS Chapter 4 4.1 5. Some Misceiianeous Expansions Solutions to Problems References Index .3 4.3y2 = .9 Chapter 5 5.2 5.2 4.8 4.



The plan in this book is to present an easygoing discussion of simple continued fractions that can be understood by anyone who has a minimum of mathematical training. and then. Continued fractions were studied by the great mathematicians of the seventeenth and eighteenth centuries and are a subject of active investigation tc?day. and this attitude is reflected in the pages that follow. This chapter should be easy to read. how rational fractions can be expanded into continued fractions. but these accounts are condensed and rather difficult for the beginner. particularly into the nature of numbers. called "continued fractions". that fractions of this form. Nearly al1 books on the theory of numbers include a chapter on continued fractions. Mathematicians often think of their subject as a creative art rather than a s a science. . Chapter 1 shows how continued fractions might be discovered accidentally.Preface At first glance nothing seems simpler or Iess significant than writing a number. however. more detailed tha n necessary . if anything. In Chapter 2 these results are applied to the solution of linear Diophantine equations. for example 3. in the form It turns out. by means of examples. Gradually more general nohtion is introduced and prelimimry theorems are stated and proved. it is. provide much insight into many mathematical problems.

I a m also grateful to my wife who typed the original manuscript. 1901. he should plan to return to it later and tackle it once again until it is mastered. It goes without saying that one should not "read" a mathematics book. I n addition he should test his grasp of the subject by working the problems a t the end of the sections. refers to item 2 listed in the references. I n the text "Crystal [2]". Anneli Lax. who prepared the final ty pescript. Numbers: Rational and írrational. C. Vinally. California. and to Mrs. and Appendix 1 is a collcction of mis1 cellaneous expansions designed to show how the subjcct has developed. Ruth Murray.D. Particular thanks are due to Dr. and to the Editorial Panel for suggestions which have irnproved the book. A student of mathematics should wrestle with every step of a proof .1 has no solution in integers. These are mostly of an elementary nature. there is a short list of refercnces. It is better to get out pencil and paper and rewrite the book. Here the famous theorem of Hurwitz is discussed. and should not present any difficulties. Chapter 5 is designed to give the reader a look into the future. and other theorems closely related to it are mentioned. The first of the two appendices &ves a proof that x2 . Olds Los Altos. but also for her critica1 reading of the text. many of these expansions are dificult to obtain. 1 wish t express my thanks to the School Mathematics Study o Group for including this book in the New Mathematical Library series. for example. . Here one sees how continued fractions can be used to give better and better rational approximations to irrational numbers. The reader will find this chapter more challenging than the others. if he does not understand it in the first round. so freely given. These and later results are closely connected with and supplement similar ideas developed in Niven's book. The periodic properties of continued fractions are discussed in Chapter 4. Their answers appear a t the end of the book. and to suggest further study of the subject. closely related to the text. The main part of the chapter develops a proof of Lagrange's theorem that the continued fraction expansion of every quadratic irrational is periodic after a cei-tain stage. but the end results are rewarding.3y2 = . not only for technical advice.4 PREFACE Chapter 3 deals with the expansion of irrational numbers into infinite continued fractions. and includes an introductory discussion of the idea of limits. this fact is then used as the key to the solution of Pell's equation.

CHAPTER ONE Expansion of Rational Fractions 1. This gives + Repeating this replacement of x by 3 obtains the expression + l/x several more times he . narnely 3 l/x.1 Introduction Imagine that an algebra student atternpts to solve the quadratic equation as follows: He first divides through by x and writes the equation in the forrn The unknown quantity x is still found on the right-hand side of this equation and hence can be replaced by its equal.

obhined by stopping at consecutive stages. is in agreernent to three decimal places with the last result above.303. what is meant when we say that the infinito dec&l 0. Then will the expression on the right of (1. First. if we calculate more and more convergents (1. he does not seem to be getting any closer to the solution of the equation (1.2) the nontermina ting expression + where the three dots stand for the words "and so on" and indicate that the successive fractione are continued without end. suppose we consider the process used to get (1.333 * is + .1).2).6 CONTINUED FRACTIONS Since x continues to appear on the right-hand side of this "multipledecked" fraction.3)) will we continue to get better and better approximations to x = f (3 di3)? Second. when rounded to 3. These preliminary calculations suggest some interesting questions. The quadratic formula ehows that this root ie actually equal to - which. We see that it contains a succession of fractions. give in turn the nubere It then comes as a very pleasant surpriee t o discover that these numbers (or convergents as we shall cal1 them later) give better and better approximations to the positive root of the given quadratic equation (1.4) actually be equal to +(3 fi) ? This reminds us of an infinite decimal. For example. These numbers.2) as being continued indefinitely.1). so that we have in place of (1. But let us look more closely at the right side of equation (1. when converted into fractions and then into decimals.

A study of these fractions and their many properties and applicatians forms one of the most intriguing chapters in mathematics. In this monograph. until we come to Chapter 3.RATIONAL FRACTIONS 7 equal to ? These and many other questions will eventually be discussed and answered. are positive integers. The first of these is the introduction of basic definitions. * 9 bi. we shall restrict our discussion to simple continued fractions. we shall further restrict the discussion to Jinite simple continued fractions. and the number of terms may be finite or infinite. 1. bz.2 Desnitions and Notation + An expression of the form is mlled a continued fraction. In general. Multiple-decked fractions like (1.) a4. ba. These have the form .a.2) and (1.a3. We must start with simpler things. and where the terms a2.. however. may be any real or complex numbers. the numbers al.4) are called continued fractions. however. In fact. These have the form where the first term al is usually a positive or negative integer (but could be zero).a2.

a. continued fraction. From now on.7) is signs after the first one are lowered to remind us of the where the "stepdown" process in forming a continued fraction. For exarnple. a. It is also convenient to denote the continued fraction (1. so that [al. .3 Expansion of Rational Fractions A rational number is a fraction of the form p/q where p and q are integers with q # O. . . unless the contrary is stated.a .8 CONTINUED FRACTIONS with only a finite number of terms al. is How did we get this result? First we divided 67 by 29 to obtain the quotient 2 and the remainder 9. so that 9 Note that on the right we have replaced gg by the reciprocal of 9.az. can be expressed as a Jinite simple continued fraction..8) by the symbol . Such a fraction is called a terminating continued fraction. or rational number. We shall prove in the next section that everg rational fraction. A much more convenient way of writing (1.as. az.. un.]. the words continued fraction will imply that we are dealing with a Jinite simple continued jrmtion. are called the partial quotients of the 1.. + The terms al. the continued fraction for 8. .


Next we divided 29 by 9 to obtain

Finally, we divided 9 by 2 to obtain

at which stage the process terminates. Now substitute (1.12) intr (1.11), and then substitute (1.11) into (1.10) to get

We should notice that in equation (1.10) the number 2 29 is the largest multiple of 29 that is less than 67, and consequently the remainder (in this case the nurnber 9) is necessarily a number 2 0 but definitely <29. t Next consider equation (1.11). Here 3 9 is the largest rnultiple of 9 that is less than 29. The remainder, 2, is necessarily a number 2 0 but <9. In (1.12) the number 4 2 is the largest multiple of 2 that is less than 9 and the remainder is 1, a nurnber 2 0 but <2. Finally, we cannot go beyond equation (1.12), for if we write

then 2 1 is the largest multiple of 1 that divides 2 and we simply end up with

so the calculation terminates. t If a number a is l a s than a number b we write a < b. If a is less than or
equal to b we write q 5 b. Likewise, if a is greater than b, or if a is greater than or eqiial to b, we write, respectiveIy, a > b, a b. For a detailed discumioii of inequaIities, see E. Beckenbach and R. Bellman (11.




The process for finding the continued fraction expansion for can be arranged as follows: 20)67 (2 = al


Divide 67 by 20. 2 29 = 58; subtract 58 from 67. 58 9)29(3 = a2 Divide 29 by 9. 27 3 9 = 27 ; subtract 27 from 29. 2)0(4 = a3 Divide0 by2. 4 2 = 8 ; subtract 8 from 0. 8 1)2(2 = a4 Divide 2 by 1. 2 2 1 = 2; subtract 2 from 2. O Procesa terminates.


We observe, in this exrtmple, that in the successive divisions the remainders 0, 2, 1 are exactly determined non-negative numbers each smaller than the corresponding divisor. Thus the remainder 9 is less than the divisor 29, the remainder 2 is less than the divisor 9, and so on. The remainder in each division becomes the divisor in the next division, so that the successive remainders become smaller and smaller non-negative integers. Thus the remainder zero must be reached eventually, and the process must end. Each remainder obtained in this process is a unique non-negative number. For example, can you divide 67 by 29, obtain the largest quotient 2, and end up with a remainder other than 0? This means that, for the given fraction +$, our process yields exactly one sequence of remaindew. As a second example, let us find the continued fraction expansion for gig. We obtain



Notice that in this example al = O. To check our results, al1 we have t do is simplify the continued fraction o

A comparison of the expansion = [2,3,4, 2 with the expan1 sion of the reciprocal # = [O, 2, 3,4, 21 suggests the result that, if p is greater than q and


The reader is asked to state a similar result for p

< q.

The following examples will help to answer some questions which may have occurred to the attentive student. Pirst , is the expansion

f the only expansion of +&as a simple finite continued fraction? I we go back and study the method by which the expansion was obtained, the answer would seem to be "yee". And this would be true except that a slight change can always be made in the last term, or last partial quotient, a4. Since a = 2, we can write 4

-1= - 1 - . =
a 4 Hence it is also true tbrtt


1 l+i

6. 1. aa are positive. O Thus .3. For example. but as. 6.3.p / q . proceed as follows: (Search for a negative quotient which. a4. 3. aa]. f or (1. and then expand the resulting fraction. The third question is this: I we multiply the numerator and f denominator of by sume number. FVe shall see in the more general discussion which follows that this is the only way we can get a "different" expansíon. when multiplied by 44 and subtracted from . say 3. g$+. 21. 21 44 = [-1. leaves the srnallest posz'title remainder.-l.1. az. ar. 6 - a ? .4.1] can be changed back to its original form [2.[-1. let us consider how t obtain the expansion of a negative o rational number . 1 = [al.12 CONTINUED FRACTIONB Clearly the expansion [2.14) 87)2oi(2 174 27)87 (3 81 6) 27 (4 24 3)6(2 .) Thus a . t o find the continued fraction expansion of . This requires only a slight variation of the process already explained. as. 1 Notice that al is negative. Next. a4.37. will the continued fraction for ZgiF be the same as that for We shall see that the expansions are identical.4.g .3.

al. - 1. (b) $9. 4. Show that. Find p / q if 3. 21. ++.unJ. . and conversely.54 = 100 . 5.e.a2. 1..4 Expansion of Rational Fractions (General Discussion) So far we have introduced the terminology peculiar t o the study of continued fractions and have worked with particular examples. . 2. . We always obtain a rational fraction p / q in its lowest terms. a.7.14159 2. Find p / q if p / q = [O. Problem Set 1 1. not to S#. Convert p / q to a decimal and compare with the value of u. a2.. 23 (e) -23 = 100 355 354 (c) 3. Find p / q if p / q = [3.]. a. (g) 3. a fraction for which p and q have no factors greater than 1 i n common. Working with symbols instesd of with actual numbers frees . If we calculated we would get back to $g.15. al. 4. Can you discover a t this stage a reason for this? Later an explanation will be given. az.. Convert each of the folIowing into finite simple continued fractions.a. if q / p = [O. then q / p = [O. 11. Find the simple continued fraction expansions of (a) compare these with the expansions in Problem 1 (a). i. But to make real progress in our study we must discuss more general resulte. if p > q and p / q = [al. . (b). 6.RATIONAL FRACTIONS 13 This illustrates an interesting property of continued fractions. a*.].] then p / q = [al. .

dividing q by rl to obtain Notice now that q/rl is a positive fraction. be any rational fraction. If rl f O. Thus.14 CONTINUED FRACTIONS the mind and allows us to think abstractly. We divide p by q to obtain where al is the unique integer so chosen as to make the remainder rl greater than or equal to O and less t h n q. f the process krminaks and the continued fraction expansion for P / Q is [all. I r2 = 0 . liero. let p/q. with the exceptions to be noted below. first sentence in this theorem is quite clear from what The we have explained in our worked examples. I rl = 0 . PROOF.15) to obtain P 9 = al 1 + .= [al. a number bet'ween O and rl.16) into (1. any rational number p/q can be represented as a jinite simple continued fraction. THEOREM Any jinite simple continued fraction represents a rational number. once this has been accomplished a host of other ideas quickly follows. Conversely. the process stops and we substitute f q/rl = a2 from (1. we write and repeat the division process. for if any expansion terminates we can always "back track" and build the expansion into a rational fraction. 1. or positive. . az] a2 as the continued fraction expansion for p / q . so a2 is the unique largest positive integer that b k e s the remainder r. q > O. while our first theorem merely expresses in general terms what we did in the worked examples. al can be negative.1. t h representation. or expansion. i s unique. As we saw in the worked examples. To prove the converse.

1s it possible never to arrive a t an rn which is zero.16) in the form and repeat the division process using rl/r2. which is equal to zero. with the equation in which the reminder rn is equal to zero. Hence. we shall be in the ridiculous position of having discovered an infinite number of distinct positive integers al1 less than a finite positive integer q. after a certain finite number of divisions. From the first two equations in (1. for the remainders rl. r2. sequence of non-negative integers q > rl > r2 > r3 > . by successive divisions we obtain a sequence of equations: terminating. we write (1.RATIONAL FRACTIONS If r2 # O. TB.. and unless we come eventually to a remainder r. so that the division process continues indefinitely? This is clearly form a decreasing impossible.18) we have . We observe that the calculations stop when we come to a remainder rn = O. It is now easy to represent p / q as a finite simple continued fraction.

1) +1 1 - so that (1.16 CONTINUED FRACTIONS Using the third equation in (1. a. . This statement must be accompanied. as we choose. so that the nurnber of terms in the expansion is either euen or odd.19) can be replaced by On the other hand. by the remark that once the expansion has been obtained we can always'modify the tast terrn a.18) we replace r1/r2 by and so on. however. 1 .2.]. To see this. +a. if a.19) becomes Hence we have the following theorem: THEOREM Any rational nunzber p / q can be expressed as a 1. then so that (1. . = 1. It is interesting to notice that the equations (1.az.19) Q =al+z 1 1 +G + . . The uniqueness of the expansion (1. Jinite sinzple continued fraction in which the East termican be mod$ed so as to make the nulnber of terms in the expansion eather even or odd.18) are precisely the equations used in a procedure known as Euclisl's algorithm for . is greater than 1 we can write a. notice that if a. (a.= fal. a3) . until finally we obtain the expansion (1.19) follows from the manner in which the si's are calculated.

. .) of any two integers p and q is the largest integer which divides both p and q. any integer d which divides both b and c will also divide a.. similarly for the other equations. then b = dbi. The number d is the g. In the theory of numbers the g. ( p . . Likewise. theh any integer d which divides both a and b must divide c. of p and q.d. . We need only one more observation: I a.d. Since a . since (a) d = 3 . of the integere p and q is denoted by the symbol ( p . .18) by multiplying both sides by the denorninator q.c...-1 is the g. b. 5 divides both p and q. we see that so that d divides c.d.13. and c are integers f such that a = b+c..RATIONAL FRACTIONS 17 finding the greakst common divisor of the integers p and q. we first state the two conditions that the g. We shall prove that the last nonvanishing remainder r.18) in the form: . .c... + t The greatest commn divisor (g.. m . and if d divides b.. For example. .. let p = 3 5 11 and let q = 32 5 . p = alq rl..b = c.C.c... . J. In order to do this. and (b) the common divisors 3 and 5 of p and q divide d . however it is known to be of earlier origin. [This procedure occurs in the seventh book of Euclid's Elemnts (about 300 B. The first equation. of two integers must satisfy. bl an integer.] To find the greatest common divisor of p and q by means of Euclid's algorithm.c. .. q). of p and q is d = 3 5. we write the equations (1.). is obtained from the first equation in (1. of two integers p and q if (a) d divides both integers p and q. For if d divides a. and (b) any common divisor c of p and q divides d. Then the g. ..d..d. . thus. q) = d means that d is the largest integral factor common to both p and q. then a = d& where al is an integer.c.d.c..

Next we must show that if c is any common divisor of both p and q. shows that rn-1 divides.22) shows that c divides rt. of p = 6381 and q = 5163. I c divides both p f and q. then c divides 1 .-~ divides both p and q. let us use Euclid's algorithm to determine the g.d.2 .s and rn.a. and hence divides rl. and we conclude that r. and condition (a) is satisfied. the first equation in (1. and finally.22). of P q. we find that rn-1 divides ra and r2. shows that rn-1 divides rn-3. Hence.22) and work our way down. Dividing r2 and rl. it divides p.-1 is the g.c. the second equation in (1. and hence divides rn-l. Thus condition (b) is satisfied. This time we start with the first equation in (1.18 C O N T I N U E D FRACTIONS We now return to the equations (1. We find that . In the same way. it dividea q .c. or is a factor of.-a and r. since it divides both rn. we arrive at the next to in the last equation. namely The equation . r.-2. Working up from the bottom in thii fashion. since it divides rn-1 and r n 4 . from the equation we see that rn-1 divides rn-r. dividing both rl and q. in which c divides r. As an example.22) shows that c divides rl. The last equation there.d. But if c divides both q and rl. Contin~iing this manner. directly above it.

where 709 is a prime number.c. of 6381 and 5163. third.d. in succesaion. but before we can put them to effective use we must study some of their properties in greater detail. and hence is the g.1449 (b) 1517. Expand the following rational fractions into finite simple continued fractions with an even number of terms and also with an odd number of terms: 2.2015 ( e ) 2299.) Thus 3 is the o d y factor common to these two numbrs.3800 (d) 3528. and 5163 = 3 1721 where 1721 is alm a prime number.d. a. are positive integers. .c. From'these we can form the fractions obfained.1 of s the following pairs of numbers: (a) 1380. by cutting off the expansion procem after the firat. az. Actually.6 Convergents and Their Properties Continued fractions are of great service in solving many interesting problems. These fractions are called the m .RATIONAL FRACTIONS 19 hence 3 is the g. From now on we will cal1 the a2. 6381 = 32.c. and where .d. the partid quotients or quotients of the numbers al.7455 1. aa. a. steps. In Section 1. Problem Set 2 1. U e Euclid's algorithm to find the greatest common divisor (g.709.4 we saw that any rational fraction p/q could be expanded into a finite simple contiilued fraction where al is a positive or negative integer. aecond. (A prime number is a number with precisely two positive integral divisors: 1 and the number itself. or zero. continued fraction.

Next we write where p2 = aia2 +1 and q2 = a2. then and so on.23). The nth convergent. atl. ql = 1. It is important to develop a systematic way of computing ihese convergents. convergents. We write where pl = al.]. it equal to the continued fraction itself.a. We notice that so that Again. cn = al + -r + a 1 .20 C O N T I N U E D FRACTIONS first. second. Now let us take a closer Iook at the convergent c3.-+a. of the continued fraction (1. respectively. 1 . . third. that so that . by factoring.= [al. from c4 we observe.

.RATIONAL FRACTIONS From (1. n. of course. . where That the equations (1.25) we might guess that if then and that in general. .] satisfy the equations with the initial vaEues PROOF. although convinced of their correctness. but it is a genuine example of inductive thinking.28) we get + . 5.24) and (1.27) are true for i = 3 . .3. U. . This. . 4 .n. a2. We guess the formulas from the first few calculations. ith convergent ci of the continued jraction [ a l . have seen already that cl = pl/ql = a l / l and that We c2 = p2/q2 = (azul l)/a2.. would not give us proof that the equations (1. I we substitute i = 3 in equations f (1. .4. Thus we state and then prove by induction the followirig theorem : THEOREM The nurnerators pi and the denorninators qí of the 1. then. for i = 3. 5 .26)are correct can be confirmed by a direct calculation. we must still supply a forma1 proof. .

3 necessarily holds for the next 1. In equation (1. pk-1. that is. up to some integer k.30) with j replaced by k.4. To see this. first replace j by k .5.30). that is. simply compare + with This suggests that we should be able to calculate c k + l from the formula for c k obtained from (1. that for the integers 3. qk-1 did not change their values when we tamper with ak.3 is true.1.2. Let us assume that Theorem 1. On the basis of this assumption.30) to help us supply a integer k proof that + The next few steps will require concentration. from This we could certainly do if we were sure that the numbers pk-2.22 CONTINUED FRACTIONS again in agreement with the direct calculation of Ca. k.1. qk-2. we wish to prove that Theorem 1.4. We obtain in succession : and . let us look at the manner in which they are calcillated. Notice first that ck+l differs from c k only in having (ak l/ak+l) in place of ak. 9 k .To do this we use equations (1. and then by k . or has been verified by direct calculation. To see that they do not. m for j = 3.5.

6. that if the expression for the convergent Cj.1 quotients al. a2. 5.30). This. .In (1. then it also holds for the next convergent c k + ~ = pk+l/qk+l.4. given by (1. Le.30) holds for j = k = 3. as we have explained. qk-1 depend only upon the number ak-1 and the numbers pk-2.32) replace a by k (ak l/ak+l) to obtain. But we actualiy know by a direct calculation that (1. we obtain and rearranging the terms. . . We are now ready to calculate ck+~. holds for the values j = 3. Ic. and q's.. thrtt akpk-1 pk-2 = pk. then. ak-1 and hence are independent of ak. Thus the numbers pk-2. This means that they wili not change when ak is replaced by (ak l/ak+l). 7. respectively. Qk-1 depend only upon the first k . multiplying numerator and denominator by ak+l. pk-1. . qk-2.30) hold for j = Ic. qk-3. + Hence the terms in parentheses in the numerator and denominator of our last expression for ck+l can be replaced.RATIONAL FRACTIONS 23 We notice that the numbers pk-1. + In studying this proof. al1 of which in turn depend upon preceding a's. we obtain We have proved.proves Theorem 1. Hence it is true for the next integer Ic 1 = 4. . t n.3. . qk-2. p's. we get At this point we use the assumption that formulas (1. Thus. notice that nowhere h v e we used the fact that the quoticnts ai are integers. + + Now. by p k and q k . pk-3. and likewise for Ic = 5. Although each ai is an integer.

28)) and using (1.28) we get the undeJined terms po. and the first two values.28) could also reproduce the first two convergents given by (1. n.33))we get for i = 2 we get Hence.Setting i equal to 1 in (1. I we put i = 1 . the assigned values (1.28) will hold for i = 1 . . Nevertheless its substitution for ak in the proof causes no breakdown of the argument. 2 . However. The calculation of successive convergents can now be systematized. 2. The continued fraction expansion for '.33) enable us to dispense with equations (1. qo. n.24 CONTINUED FRACTIONS the number ak l/ak need not be one. q-1. will reproduce equations ( 1 -29). . 2 . p-1.29).1 . if we asxign the values + to these undefined terms. i = 1 .n . But notice that p-l/q-l and po/qo are not convergents. . I t would be convenient if the equations (1. 2 in f (1. 3.Ts" is We form the following table : TABLE 1 .29) and to use instead equations (1.28)) with i = 1 . then equations (1. An example will make this clear.

The special values p-1 = O. Thus. p = 49. First. for example. Then we calculate the pi's. . 0. q-l = 1.28). from equations (1. Express each of the following continued frsctions in an equivslent form but with s n odd number of psrtisl quotients. Cr = H. under i Problem Set 3 = 4. o under i = . 2t-5 . i = 0. Thus. For i = 3. ci have been listed. under the values of i to which they correspond. qo = O are entered a t the left. (a) (b) % o (c) M (d) 2. using i = 1. p = 1. pi. 2.) 2 under i = 1 in the third row. Espsnd the following rationsI numbers into simple continued frsctions snd calculate the successive convergents ci for esch number. qr = 20. Note: Starred problems are more difficult snd could be omitted the first time over. respectively. For i = 2. and so on.'s we follow the same scheme. entering the values we obtain in the row labeled qi. 4 We form our table in this way: We write the values a in the seci ond row. under i = 4 we find a4 = 2. p3 a3p2 pi = 4 5 2 = 22.1. we 2 which is recorded under i = 2 in the same row. 1. qi. we get i = - (Follow the first system of arrows We record pl obtain = O+-1 # ' . . To calculate the q.RATIONAL FRACTIONS 25 Explanation of tabb: The entries In the first row of the table are the values of i: i 1. Under each value of i the corresponding values of ai. + + so 20 is recorded in the fourth row. 1.

26 C O N T I N U E D FRACTIONS 3.) + + + + + 5.3.t. + .n .~ q . 51. 1. hneralize Problem 4. .4.l+ p+2. *8.~ . (See Problem 7.p.4771 (d) 0. take pn/qs = pa/q6.2. (&e Problem 8 below. 1. 3. (a) 3.) 6. Do the same with qa/qr. and add the multing expressions. . For each continued fraction in Problem 2. 4. al]. Note that a.6] and show that p6 = 5p6 5p4 4pa 3p2 2pl 2. .a. for example. show that .n]. hence and so on.2. the last convergent. 4. Calculate the convergents of the continued fraction [1. let i be equal to 1.-lp. -.718 (T) (e) Pn .a. .2. For 13. We know that p.. and Rints. Calculate the successive convergents to the following approximations to the numbers in parentheses. of [ l . . let n be the numbe* oi par tia1 quotients and calculate pnqne1 . In 2 (a). -. p2/q2. calculate p5 and p4. If pllql.[as. hence We also know that p-1 = u.3.. then calculate the corresponding quantity after these fractions have been expressed with an odd number of partial quotients. 4.5.a + p. Pn-I (c) 0. Then convert p6/p4 into a simple continued fraction and compare it with the original fraction. = n.-~. G-2. ps/qn are the convergents Hint: In the relation pi = ipi-1 pi-2.3010 (Ioglo 3) (10glo 2) S S S .14159 (b) 2..

34).poql = al O -1 1 = (-1)l. + + .3. i = k i equations (1. we see that But this is the statement of the theorem for i = Ic 1. so we have proved t h t the theorern holda for i = k 1 if z't holds for i = h.28)] we know that for i = i 1.3 [see holds for the next integer. .pk-lqk) * We assume t h a t the theorem holds for i = k. When i when i = 1 . + + hence we can write pk+lqk (1.1)') where i > 0. Plqo = 0. PROOF: Direct calculations show that the theorem is true for i = 0. THEOREM If pi = aipi-1 1.4. This is a corollary to the following fundamental theorem.pkqk+l = = + pk-1)qk ~k(ak+lqk+ qk-1) ak+lpkqk + pk-lqk . when i = 2. then it also 1.6 DBerences of Convergents Those who worked the preceding exercises will already have guessed that the convergents to a finite simple continued fraction are always in their lowest terms.34) . &$ned as in Theorern 1.pkqk-1 (ak+lpk " = (- 1 ) (pkqk-1 . 1 . We shall prove that i f the theorem holds for i = Ic.RATIONAL FRACTIONS 1. that Substituting this result into the last line in (1.pi-iqi = (. From Theorem 1.ak+lpkqk . 2 . that is. then piqi-1 + pi-2 and qi = aiqi-i + qi-a are .

.pi-2qi-1 is the same as piqi-1 .c. and therefore for i r= 1 1 = 2. we obtain the final result.he symbol d = (a. . Hence this reduction. 1. p.pi-iqi but with i replaced by i 1.2. 2. PROOF.piq2.8 . can be repeated.d. p2q1 . and so on for al1 values of i = 0. 1 . 51 by calculating in turn poq-1 .pi-tqi-i = (-1) ( p i .pi-tqi-1).1. Since + piqi-i . 2 . n .z q i . performed in succession. continued fraction i s in its lowest terrns. of a simple 1.1 are the only common divisors of pi and qi.1)" But the only divisors of ( .1. The expression pi-lqi-r .pi-tqi-2) After i reductions of the sarne sort. p 2 / q t r rational fraction in its lowest terrns. q i ) = 1. of a and b. The earliest traces of the idea of a continued fraction are somewhat confused. we used t.l)i are 1 and . and qi have no cornrnon divisors other than 1 or . + + Problem Set 4 1.* . i 2 1. but there was no systematic development of the subject. that i s . Notice that = (-1) (pi-iqi-2 . etc. yielding pi-iqi-2 . it follows that any number which divides both p i and q i must be a divisor of ( . ~ 6 / q 6 is Also verify that each convergent p l / q l .pi-iqi = (-l)i.I ~ O . for many ancient arithmetical results are suggestive of thesg fractions. . I n oiir discussion of Euclid's algorithm. hence the numbers 1 and .4 using the continued fraction [3.b) to indicate that d was the g. plqo .4 using the following hints. 1. since 1 is the largest number that divides both pi and qi. we can now state that ( p i .5.p . hence i t holds for i = O+ 1 = 1. 2. + COROLLARY Everg convergent ci = p i / q i .p0q1. or "stepdown" from i to i 1. Give another proof of Theorern 1. Check Theorern 1.28 CONTXNUED FRACTIONS We know the theorem holds for i =O. - - We end this chapter with a few brief remarks concerning the history of the theory of continued fractions.

A referente to continued fractions is found in the works of the Indian mathematician Aryabhata. of 177 and 233. he expressed in the form 0 This he modified.c. Further traces of the general concept of a continued fraction are fouiid occasionally in Arab and Greek writings. essentially. In our modern symbolism he showed. +#. who died around 550 A. Bis work contains one of the earliest attempts a t the general solution of a linear indeterminate equation (see next chapter) by the use of continued fractions. This is perhaps the earliest (c. Germany. into the form which is substantially the modern form A third early writer who deserves mention is Daniel Schwenter (1585-1 636)) who was a t vario us times professor of Hebrew. of two numbers is essentially that of converting a fraction into a continued fraction.D. 1530)' a native of Bologna. also a native of Bologna. +. and mathematics a t the University of Altdorf. +.d. and +. for example. that This indicates that he knew. Oriental languages. His treatise on algebra (1572) contains a chapter on square roots.c. and from these calculations he determined the convergents *. 300 B. In his book Geornetrica Practica he found approximations to 8 8 by finding the g. for convenience in printing.d. . Most authorities agree that the modern theory of continued fractions began with the writings of Rafael Bombelli (born c.) important step in the development of the concept of a continued fraction.C. In a treatise on the theory of roots (1613). that The next writer to consider these fract'ions was Pietro Antonio Cataldi (1548-1626).RATIONAL FRACTIONS 29 We have already seen that Euclid's method for finding the g.

hgrange (1736-1813). New York: McGraw-Hill Book Company. laid the foundation for the modern theory. Wdlis stated a good many of the elementary properties of the convergents to general continued fractions. continued fractions are used to give approximations to various complicated functions. 1956 (Chapter 9). From this beginning great mathematicians such as Euler (17071783). Hildebrsnd. This is described in his treatise Descriptio Autornati Planetarii. t See F. They constitute a most important tool for new discoveries in the theory of numbers and in the field of Diophantine approximatioiis. an extensive area for present and future research. Euler's great memoir. and once coded for the electronic machines. Christiaan Huygens (1629-1695) used continued fractions for the purpose of approximating the correct design for the toothed wheels of a planetarium (1698).t. astronomer. In the computes field. De Fractionibus Continius (1737). published posthumously in 1698. Continued fractions play an important role in present d a y mathematics. The great Dutch mathematician. He also used for the first time the name "continued fraction". and physicist. In particular. Lambert (1728-1777).30 CONTINUED FRACTIONS The next writer of prominence to use continued fractions was Lord Brouncker (1620-1684). Introductwn io Numerical Anula(t?is. into the continyed fraction but made no further use of these fractions. . He transformed the interesting infinite product discovered by Che English mathematician John Wallis (1655). mechanician. give rapid numerica1 results valuable to scientists and to those working in applied mathematical fields. In the discussion of Brouncker's fraction in his book Arithrnetica Infinitorum. including the rule for their formation. and many others developed the theory as we kiiow it today. B. There is the important generalization of coiitinued fractions called the analytic theory of continued fractions. published in 1655. the first President of the Roya1 Society.

we can f give x any value. riddles. we restrict the values of x and y to be integers. and a number of pigs at $50 each. we have the equation which is equivalent to I nothing limits the values of x and y in equation (2. and trick questions lead to mathematical equations whose solutions must be integers. If. say x = 9. this sense.21. (2. How many cows and how many pigs did he buy? If x is the number of cows and y the number of pigs. as the farmer is likely to do (since he is probably not interested in half a cow). His bill was $810.CHAPTER TWO Diophantine Equations 2. which means that we can always find some value of y corresponding to any value we choose for x. getting y = 9.2) is a n indeterminate equation. and then solve the resulting equation In for y. Here is a typical example: A farmer bought a number of cows a t $80 each. however.1 Introduction A great many puzzles. then our example belongs to an extensive class of problems .

The second method will show how the theory of continued fractions can be applied to solve such equations. it should be noted.1) can be solved in many ways. . we solve the equation for y.82 a non-negative multiple of 5. 3. if we write equation (2. Indeterminate equations ta be solved in integers (and sometimes in rat'ional numbers) are ofterl cdled Diophantine equations in honor of Diophantus.32 CONTINUED FRACTIONS requiring that we search for integral solutions x and y of indeterminate equations. take on the values 0. I n f m t there is no harrn in solving such equations b y trial and error or b y rnakiq intelligent guesses. who wrote a book about such equations. So the farmer could buy 2 cows and 13 pigs.D. The calculations are hence the two solutions to our problem are (3. We shall give two additional methods. getting t For additional examples. see 0.5). Equation (2. 13) and (x..2) in the form 81 8 2 = 5y. in turn. There are other ways of solving Diophantine equations. . . Our problem. [lo]. - we need only search for positive integral values of x such that 81 . a Greek mathematician of about the third century A.2) and hence equation (2. Ore . y) = (2. Icor example. we find that x = 2 and x = 7 are the only non-negative values which make 81 .2 The Method Used Extensively by Eulert Let us consider again the equation Since y has the smaller coefficient. 2. 10.8x is a mitltiple of 5 . y) = (7. The first of these was used extensively by Euler in his popular text Algebra. 1. published in 1770. Letting x.2. or 7 cows and 5 pigs. has the further restridion that both x and y must not only be integers but must be positive.


Both 81 and 8 contain multiples of 5, that is, 81=5.16+1 therefore, from (2.4), we have and 8=5*1+3;


Since x and y must be integers, we conclude from equation (2.5) that t must be a n integer. Our .task, therefore, is to find integers x and t satisfying equation (2.6). This is the essential idea in Euler's method, i.e., to show that integral solutions of the given equation are in turn connected with integral solutions of similar equations with smaller coefficients. We now reduce this last equation to a simpler one exactly as we reduced (2.3) to (2.6). Solving (2.6) for x, the term with the smaller coefficient, we get


Again, since x and t must be integers, u must also be an integer.



Conversely, if u is an integer, equation (2.8) shows that is an integer; x also is an iriteger since, from (2.7)) Substituting x 2


- 5u and

t = 3u


1 in (2.5) gives

so that y is an integer. This shoivs that the general integral solution of (2.3) is

where u is any integer, positive, negative, or zero, Le., u = O , f l , f 2 ) f3, . . .


A direct substitution into (2.3) shows indeed that
Consequently (2.3) has an infinite number of solutions, one for each integral value of u. A few solutions are listed below:

I the problern is such that we are limited to positive values of x f and y, then two inequalities must be solved. For example, if in (2.9) both x and y are to be positive, we must solve the two inequalities

- 5u > 0,
2 u<-? 5


+ 8u > O,
u >

for u. These inequalities require that u be an integer such that and

13 -8


and a glance at Figure I shows that the only two possible integral values of u are O and -1. Substituting, in turn, u = O and u = 1 in (2.9) gives (x,y) = (2,13) and (x,y) = (7,5), the original answers to the farmer's problem.




Figure 1 Going back over the solution of equation (2.3) we can raise certain questions. For example, why should we solve for y, rather than for x, simply because y has the smaller coefficient? If we had solved first for x, could we have arrived a t a shorter solution? In the second line below equation (2.4) we replaced 8 by 5 I 3. Why not replace 8 by 5 - 2 - 2 ' In solving equation (2.3) the writer did not ! have in rnind the presentation of the shortest solution. We leave i t to the reader to experiment and try to obtain general solutions in the least number of steps.


Problem Set 5
1. Use Euler's method to solve the following linear Diophantine equations.

In each case list the positive integral solutions, if any, (a) 15x 478 = 2 (b) 3 1 x + 7 y = 1



(c) 152 (d) 13s

+ 47y = 4 + 21y = 295

2. Does the indeterminate equation 62 15y = 17 have integral solutions? Note that the left side of the equation is divisible by 3. What about the right-hand side? What happens if we go ahead and use Euler's method anyway? 3. Return to equation (2.9)and fill out the following table for the values of u indicated.


On ordinary graph paper plot the points (x,y) and join them by a straight line. Use this graph to pick out the positive solutions of the equation 8x 5y = 81.


4. A man buys horses and cows for a total amount of $2370. If one horse costs $37 and one cow $22, how many horses and cows does he buy?

5. Show that the equation 172 integral solutions.

- 15y = 5

has infinitely many positive



6. Find integers u and u such that u tl = 84 and auch that u ia divisible by 9 and u is divisible by 13. Hint: Let u = Ox, u 13y.



7. Find a number N which leaves a remainder 2 when divided by 20 and a remainder 12 when divided by 30.Hint: Find integers x and y so that the required number N = 20x 2 = 30y 12. Hence solve the equation 20x - 30y = 10.



2.3 The Indeteminate Equation ax

- by



We are now ready to show how continued fractions can be used to by = c where a, b, solve the linear indeterminate equation ax and c are given integers, and where x and y are the unknown integers. Our approach to this will be a step-by-step process, through easy stages, culminating in the final mastery of the solution of any solvabk equation of the form ax by = c. We start with the restrictions that the coefficients of x and y are of different signs and that they ha,ve no common divisor but 1. Thus we first learn to solve the equation



where a and b are positive integers. [The equation - a x by = 1 , ( a , b) = 1, is of the same form with the roles of x and y interchanged.] The integers a and b can have no ditrisors greater than 1 in common; for, if an integer d divides both a and b, it also divides the integer 1 on the right-hand side of the equation and hence can have only the value d = 1. In other words, a and b must be relatively prime, or d = ( a , b) = 1. We shall now state and prove


THEOREM The equation ax - by = 1, where a and b are 2.1. relatively prime positive integers, has un injinite number of integral aolutions (x,y).
We first convert a / b into a finite simple continued fraction
a - = [al, a2, b

- , a,-1,


and calculate the convergents


,cn-1, cn.

The last two


con vergent S,

are the key to the solution, for they satisfy the relation stated in Theorem 1.4, namely that and since p,


a, q, = b, this gives

If n is even, that is if we have a n even number of partial quotients al, a2, , a,, then (- l)n = 1 and (2.12) berornes

Comparing this with the given equation we see that a solution to this equation is

This, however, is a particular solution and ilot the general solution We indicate particular solutions by the ilotation (xa,ya). On the other hand, if n is odd so that (-1)" = -1, we can modify the continued fraction expai~sion(2.11) by replacing

1 un or by replacing



(a. - 1)


1 1

if a,

> 1,

Thus, if (2.11) has an odd humber of partial quotients, i t may be transformed iiito [al, a2, or into [a1,a2,...,a,-~+l], if a n = l ;

, a - 1, 1

if an > 1,



in both cases the number of partial quotients is even. Using these continued fractions, one case or the other, we re-calculate p,- Jqn-r and p,/q, = a/b, and equation (2.13) is satisfied once more. Once a particular solution (xo, yo) of equation (2.10) has been found, i t is an easy matter to find the general solution. To this end, let (x, y) be any other solution of (2.10). Then and and a subtraction @ves (2.14) a(z

- xo)



- yo).

This shows that b divides the left side of the equation. But b cannot divide a since a and b are relatively prime; hence b must divide x - xo, that is, x - xo is an integral multiple of b, and we may write x-xo=tb (taninteger),




+ tb.

But if this is true, (2.14) shows that a(tb) so that y form

b(y - yo),

- yo = at.
= 1

It follo\vs that any other solution (x,y) of ax - by

has the

Conversely, if (xo, yo) is any particular solution of ax - by = 1, and if we set up the equations (2.15) with t any integer whatever, then the values (x, y) will mtisfy the given equation, because (axo - byo)


+ tab - tab

We cal1 the values of x and given by equations (2.15) the general solution of the indeterminate equation ax - by = 1.

Here n = 6 . 21 has an odd SOLUTION. We convert a/b into a finite simple continued fraction with a n odd number of convergents.8. Here the integers 205 = 5 . 4 1 and 93 = 3 31 are relatively prime.by = 205x .DIOPHANTINE EQUATIONS 39 EXAMPLE Find integral solutions of the indeterminate equation 1.12) becomes . = [2. I n this case equation (2. q .4. As a general check we have since the terms involving t cancel.93y = 1 is As a check. but it can be replaced by s5 the equivalent expansion with an even number of quotients.29109 = 1. let t = 1 . y = 313 and 205(142) .15)) the general solution of the equation ax . . 1. The convergents are then computed.10). so the equation has solutions. andhence. The continued fraction number of partial quotients. p n _ ~ = p 6 = 1 0 8 = y o .93(313) = 29110 .93y = 1. then x = 142. The method for solving the equation is quite similar to t h a t used to solve (2.l = q 6 = 4 9 = x o i by (2. 205x .

numbers 205 and 93 are relatively prime. hence the given The equation has integral solutions. . SOLUTION. f2. Comparing this equation with ax we see that XO . f l . The continued fraction espansion for is and has an odd number of partial quotients. . The general solution. take t = . so ( .1. then (x.40 CONTINUED FRACTIONS since n is odd. therefore.y) = ( -49. as before. EXAMPLE Find integral solutions of the equation 2.by = -1.1) = -1 a s Our calculations show that en-r = pn-i/qn-i = p4/44 %.* . x = $0 + tb y = yo + ta t = O . heme a particular solution of the given equation is xo = q4 = 44 and yo = p4 = 97. -3. = Qn-1) Y O 'p n . . and . the general solution being.1 is a particular solution of the @ven equation.108). To find the convergents we set up the table = (.1)" required. is As a check.

& 2 . 1 3 ) we know that aq. = (q.p. which a e with the aolution given for Example 2. = (49. 1 6 ) will be yl=a-yo=afor then axl Pn-1. --• 9 EXAMPLE Show that we can solve Example 2 if we have already solved 3.bpnWl= 1 .qn-1) .1 will then be (2. = a ( b .1 .18). 108) is a particular solution of 205x . P . Hence the general solution.DIOPHANTINE EQUATIONS 41 It is interesting to notice t h a t once we have calculated the yo) particular solution ($0. That is.93y = +l. of y the equation The particular solution of ( 2 . f is according to (2.18) x = xl+tb 2/=y1+h and this can be checked by a direct substitution. k 3 . l ) .b(a .93y = . ms . Using equations (2.-1. knowing that (xo. The general solution of the equation ax . cal1 it (si. yo) SOLUTION.-1 . .~ ) of the equation we can immediately obtain a particular solution. f l .17)we find that is a particuIar solution o 205x . . Example 1.by.-1) since from ( 2 .by = .93y = . t = O.1 . solve the equation 205x .

For example.yo) is any particular solution of (2. then .1 7 y = . (a) 132 .1 (c) 652 .1 (e) 56x . Check f each answer. Problem Set 6 1. Since (xo. .93y = .) and the general solution becomes Notice that equations (2.20) for t = 3. = (-49. Find the general integral solutions o the following equations.20) reproduce the same values of x and y but not for the same values of t. SOLUTION.17y = 1 (b) 1 3 2 .19) gives (x.56y = 1 (d) 6 5 2 .1.93y = . we know that + If we multiply through by .5 6 y = = .21).507).21) m-by= where a and b are two relatively prime positive integers.y) = (230. (a. the same values obtained from (2. = c.4 The General Solution o ax f .1 we see that hence (xi. = (49. provided we know a particular solution of Example 1. EXAMPLE Give a third solution of the equation 2052 4.42 C O N T I N U E D FRACTIONS There is still another way to solve Example 2.108) is a particular solution of the equayo) tion 205x . .1.19)and (2.by 1. suppose that (xo.108) is a particuIar solution of 2052 . This is illustrated in the following example. For.65y = 1 =1 2. y.933 = 1. t = 2 in (2.b) Once we have learned t o solve the indeterminate equation (2. i t is a simple matter to soIve the equation where c is any integer.

93(745) = 69290 . Multiplying both sides by 5 we get so that (5x0. The general solution. yo) = (49. cyo) is a particular solution of (2. 108) is a particular solution of theequation 205x .93y = 1. -540) is a particular solution of the given . Example 1 of this section we recalled that 205(49) .93(108) = 1.22). we know that (xo.22). T h u s the general solution of equation (2. take t = 1 . In SOLUTXON. 205(49) . EXAMPLE Solve the equation 2. then (x. we obtain so that (cxo.. I 2 .22) w i l be This can easily be verified by a direct substitution into (2. according to (2. that is. Section 2. EXAMPLE Solve the equation 1.69285 = 5 .DIOPHANTINE EQUATIONS and multiplying both sides b y c. Ir.745) and 205(338) .3. m S m 9 As a check. 540) is a particular solution of the given equation. SOLUTION. From Example 1 . 5yo) = (245. y = 540 + 205t t = o. will be . Multiplying through by -5 we get so that (xOryo) = (-245.93(108) = 1. y) = (338.23).

Still assuming t h a t a a n d b are positive integers. The general solution.56y = 7 (c) 562 .-.z.93(-130) -12095 + 12090 = -5.44 CONTINUED FRACTIONS equation.-. take t 205(-59) = 2.17y = 5 (b) 652 .. to that of the equation ax . and q.6 The General Solution of ax + by = c. then (x.b) = 1.3 to obtain the general integral aolutions of the following equations. .x) = b(y + cpn-1). -130). Prohlem Set 7 1. we first find a particular solution of the equation (a. but (4 b) = 1. b) = 1 = -3 2. T o do this. so b cannot divide a. ax + b y = 1. Then as before. Use particular solutions obtained from the yroblems at the eild of Section 2.6% (a.23). Therefore b divides cqn-1 .26) x = cq.24) a(cqn-1 + by = c . T h e discussion of this equation is similar. expand a/b as a simple continued fraction with a n even numher of partial quotients. and . Check each answer. The trick now is to write the given equation ax i n the form Rearrange terms to obtain (2.by = c.-1 . From the table of convergents read off p. except for some minor changes. is then To check this. This shows that b divides the left side of the equation. y) = = (-59. (a) 132 . so t h a t there is a n integer t such that (2.tb. according to equation (2.

tb) + b(at . so that 17 divides x Fkplacing x . and so the given equation may be written in the form 13s 17y = 300(13 4 .27) y = at .1200 -13t .25) into (2.17y = 1. ~ + This shows that (2.1200. or 1 3 . yo) = (4. SOLUTION. -.-1) ~ = acqn_.cpn-1.DIOPHANTINE EQUATIONS Suhstitute (2. or by 171 in (2.26) and (2. 3) is a particular solution of the We equation 13%.17 3).29) 1 3 ( ~ 1200) = -17(y + 900).17(3 . Thus the general solution + by = c is EXAMPLE Solve the indeterminate equation 1.300). or that 13(4) .cp. Hence the general solution of the given equation is x = 1200 + 171 .17(3) = 1.24)to g e t a n d solve for y to ohtain (2.bcpn-i so the equation ax of the equation ax + by = c is satisfied.27) info ax by gives + ax + by = a ( ~ q .900. find that ($0. Conversely. a direct substitution of (2.13(4 300) = -17y . for any integer t.29) gives y = . .tab + tab .

and the second two can be replaced. = -300.13t. where A and B are positive integers. For example. or y = and equation (2.30) can be solved in . To see this. SOLUTION. 13% + 171. Hence the general solution of the given equation is 2.30) A x + By = fC. if d does not divi& C. let d be the greatest common divisor of A and B. It follows that 17 divides x x = -1200 and replacing x + 1200 by 171 gives 900 .46 CONTINUED FRACTIONS EXAMPLE Solve the indeterrninate equation 2. any equation of the form f A x k By= C can be reduced to one or the other of the forms (2. Not al1 equations of the form (2. + 17t. respectively.By = fC. neither of the equations (2. Then. of the four equations the first two are already in the required form.7y = -10. A x .29%) + 17y = -300(13 + 1200.30) have solutions. second equation in the solution of Example 1 now The becomes 132 (2.6 The General Solution of Ax & By = 1 C By multiplying through by -1.900).17 3).29) is replaced by 13(x + 1200) = -17(y .? 3s + 7y = -10 and 3s . by . 4 .

then we can divide both sides of the eqiiations (2. On the other hand. namely where a and b are relatively prime.c. The general solution of 205s . divide A.186y we find that The main results obtained from our study of the linear Diophantine equation can be summarized as follows: Summary. Since d = 2 divides 10. the g. and where c is a positive or negative integer. for the left side of each would be divisible by d while the right side is not.30) by d. EXAMPLE Solve the equation l.the equation can be solved. This is the equation solved in Example 1 of Section 2.4. Divide the given equation by 2 to obtain where now 205 and 93 are relatively prime. Since 410 = 2 5 41.31) will aufomatically @ve solutions of equations (2. Any equation of the form Ax f By = f C has integral solutions x. SOLUTION. The next step is t expand a/b as a simple continued fraction with an even number n . and of which we know the solutions. and C by d = ( A . Conversely. if d does divide C. reducing them respectively to equations of the form we have just discussed.93y = 5 found there was x = 245 93t. any solution of equations (2. reducing the given equation to either the form or the form where in both equations a and b are relatively prime positive integem. 186 = 2 3 31.d. y. B.D I O P H A N T I N E EQUATIONS 47 integers x. In this case. of 410 and 186 is d = 2. B).y only if the greatest common divisor of A and B divides C.30). - + tlnd substituting it into 410% .

He divided the remainder . Express f i as the sum of two fractions whose denominators are 7 and 11.=x +u 77 7 11 - 3 The sum of two positive integers a and b is 100. A little later a second sailor awoke and had the same idea as the first.48 CONTINUED FRACTIONS n of partial quotients.-~ . He then hid his share and went back to sleep. and the general solution of (i) is Likewise the general solution of (ii) is The solutions (iii) and (iv) represent. b = 9y 5 and use the fact that a b = 100.174y = 9 (c) 772 $ 6 3 ~= 40 + + + m Hint: Find integers x and y such that . Find the general solution in integers of the others. + + + 4. He divided the nuts into five equal piles and discovered that one nut was left over. Five sailors were cast away on an island. and Monkeys The following problem is of considerable age and. and if b is divided by 9 the remainder is also 5. (d) 342 . respectively. Problem Set 8 1. Find a and b. y) of 13s + 17y = 300. so he threw this extra one to the monkeys. During the night one of the sailors awoke and decided to take his share of the coconuts. To provide food.49y = 5 (e) 34x 49y = 5 (f) 562 208 = 11 2. (hey collected al1 the coconuts they could find. and from the table of convergents read off pn-i and qn-1. Coconuts. Find positive integral solutions (x. continues to appear from time to time.bpn-l = 1. If a is divided by 7 . (a) 1832 174y = 9 (b) 183s . Then aq. the remainder is 5.7 Sailors. Two of these six equations do not have integral solutions. 2. Hint: Let a = 7% 5. in one form or another. for the cases (i) and (ii) the general solution of Ax f By = f C .

discovered also that one was left over. The problem is to Jind the smallest number of nuts in the original pile. and threw it to the monkeys. The next morning the sailors. we find that the third. divided the remaining nuts into five equal piles. We first seek a particular solution (xl. Then he hid his share. Now the number of nuts in the last pile must be a multiple of 5 since it was divided evenly into five piles with no nuts left over. no nuts being left over this time. al1 looking as innocent as possible. nuts. The first sailor took +(x . aild fifth sailors left. Similarly the second sailor took coconuts and left four times this number. or 16x .36 Similarly. hence these numbers are relatively prime and the equation (2. each throwing a coconut to the monkeys.yl) of the equation To this end. Hence where y is some integer. fourth. respective1y.1). the convergents of the continued fraction .32) has integral solutions.DIOPHANTINE EQUATIONS 49 of the nuts into five equal piles. Multiplying both sides by 3126 we obtain the indeterminate equation Factoring into primes we find 1024 = 2l0 and 15625 = 56. let x be the original nurnher of coconuts.1) coconuts and left $(x . I n order to solve this problem. In their turn the other three sailors did the same thing.

50 CONTINUED FRACTIONS are calculated: The convergent c9 yields the particular solution x l q g = 10849. we search for the vaiue of t which gives the smallest positive value of x and which at the same time makes y positive. From (2. = yo = 8404~1 . . 5975244 will be a particular solution of equation = (2. Hence xo = 8 4 0 4 ~ ~91174996.32). published by the National Council of Teachers of Mathematics. For an interesting discussion of this and related proble& see the article entitled "Mathematical Games" by Martin Gardner in ScientiJic American. A Guide to the Literature.33). April. by William L. we finally obtain which means that the original number of coconuts was 3121 and each sailor received 204 in the final distribution. y1 = p9 = 711 of equatlon (2.34) we find that t must be an integer satisfying the two inequalities Hence the required vaiue is t = . Recreationat Mathematics. The genera1 solution is - Since both x and y must be positive. Schaaf.34). 1938.5833. One should also keep in mind the excellent collection of references. Introducing this value of t into equations (2.

and where D is a positive integer not a perfect square. an'd we shall see that these fractions do not terminate but go on forever.CHAPTER THREE Expansion of Irrational Numbers 3. and. . is irrational. We proved that a rational number can be expanded into a finite simple continued fraction. & are integers. The numbers are al1 irrational.1 Introduction So far our discussion has been limited to the expansion of rational numbers. Any number of the form where P. A number of this form is called a quudratic irrational or quadratic surd since it is the root of the quadratic equation Our discussion will be limited fo the expansion of quadratic irrationals. An irrational number is one which cannot be represented as the ratio of two integers. every finite simple continued fraction represents a rational number. D. conversely. This chapter will deal with the simple continued fraction expansion of irrational numbers.

and express 2 in the form where the number is irrational.2 Prellminary Examples The procedure for expanding an irrational number is fundamentally the same as that used for rational numbers.14159 . for.e. Calculate al. namely algebraic irrational numbers and transcendental numbers. . . using decimal approximations to these numbers. The irrational number 4%the solution of the algebraic equation x2 . . 3. an equation of the form are integers. and to study the deeper properties of each should read the first monograph in the NML (New Mathematical Library) series: Numbers: Rational and Irrational. the largest integer less than x2.2 = 0. Let x be the given irrational number. To continue. by Ivail Niven. al. but the methods of obtaining the expansions of x and e given in Appendix 1 are beyond the scope of this monograph. such as x = 3. An algebraic number is a number x whch satisfies an algebraic equation. if an integer is subtracted from sil irrational number. algebraic is called a transcendental number. 1 Those who wish to learn &out the two classes of irrational numbers. and e = 2. is one example. the greatest integer less than x. It is quite difñcult to expand transcendental numhers into continued f ractions . calculate a2. we can calculate a few of the first terms of their continued fractioii expansions. It can be proved that x is transcendental.. not al1 zero. and t See 1.71828 . the result and the reciprocal of the result are irrational.t The number e is also transcendental. Niveri [8]. The irrational number x = 3.14159 . but this not easy to d0.52 CONTINUED FRACTIONS There are irrational numbers which are not quadratic surds. i. is and is therefore called an "algebraic number". A number which is not where a*.

az.. then XJ from the third into this result.. which is impossible since each successive z is i irrational. for x. x2.1) into the first equation. a. producing in succession the equations where al. to be equal to x.inued fractlion .. produces the required infinite simple cont. This calculation may be repeated indefinitely.x3. . . . 2 4 . and so on. This process cannot terminate. the number is irrational.IRRATIONAL NUMBERS express x2 in the form where. again. the only way this could happen would be for sorne integer a. are al1 integers and where the numbers are al1 irrational. . Substituting z2 from the second equation in (3.

e + The bar over the 2 on the right indicates that the number 2 ie repeated over and over. so The z EXAMPLEExpand 1. d into an infinite simple continued fraction. . Thus al1 the subsequent partial quotients will be equal to 2 and the infinite expansion of 4 will be 2 + + x4. the calculations of . Solving this equation for xz. an example or two should be worked to make sure the expansion procedure is understood. we get Consequently.414 . will al1 produce the same result.414 . is aa = 2.54 CONTINUED FRACTIONS where the three dots indicate that the process is continued indefinitely. namely -\/S 1. . The largest integer < xs =2 < + 1 = 2. = . z So~unorr.\/S 1. is al = 1. x6. largest integer < d = 1. Before discussing some of the more "theoretical" aspects of injtnite simple continued fractiom. so where At this stage we know that Since xa = 2 < 1 is the same as x. .

S] actually represents the irrational number 4 5 ? Certainly there is more to this than is evident a t first glance. and it will be one of the more difficult questions to be discussed in this chapter. we write hence or 1 = 1. roughly speaking. that we go through certain manipulations. With this understanding. We can. which tells us nothing about x. we can write from which we see that Thus .IRRATIONAL NUMBERS 55 Imrnediately some questions are raised. give a formal answer to this question. is it possible to prove that the infinite continued fraction [l. For example. 2. however. A fomnal answer means. ] = [l. but no claim is made that every move is necessarily justified. using the same idea. 2. However. .

6 3 -fi-3 2 X-1 3+4533-453 > 1. Then where x*=-. SOLUTION. This theorem will be proved in Chapter 4. These examples are illustrations of a theorem first proved by Lagrange in 1770 to the effect that the cmttinued fraction expansion of any quadratic irrational ia perz'odic after a certain stage. EXAMPLE Find the infinite continued fraction expansion for 2. so where Thus x4 = 23. proceed exactly as in Example 1. Since fiis between We 7 and 8. .CONTINUED FRACTIONS Some additional examples of a similar sort are: I n each of these examples the numbers under the bar form the periodic part of the expansion. the number 4% having quite a long period. and so the last caloulation will repeat over and over again. The largest integer < xl is a9 = 2. the largest integer < x is al = 1. so where x3re-= 1 X4-2 2 = 453-7 aa = dE+7 2 The largest integer < xa is 7.1 22 3 .

we find 7+dG. the required expansion is so that finally we obtain = [l. . Now let us reverse the process. l . let us start with the infinite expansion and try to get back to the original value of x.we obtain which ia the o r i ~ i n a value of x. I t is convenient to replace where Then y satisfies the equation Solving for y (by the quadratic formula) a.2. 2 Hence Simplifying the right-hand side.nd noting that y that 0 = > O. 71.IRRATIONAL NUMBERS Hence.

. we define p-1 = O. The convergent cn = pn/qn is calculated by the same formulas = 1. and q.C.. . = 7" = 3. but greater than 3 8 = 3. this is quite a remarkable result considering . where. ) stated that the ratio of the circumference of any circle to its s diameter i less than 31. less accurate than the above Egyptian value. Calculate the first five convergents. Archimedes (c. The computational scheme is the a m e .l are calculated in exactly the same way as before. The out as follows: 'A = 'A = 3. was used by the Babylonians.1604.7. Translated into our decimal notation. 2 -.14159 starts [3. . The approximation -U = 3. as before.292. 15. 1. SOLUTION. E~AMPLE infinite continued fraction for 1. for al1 n 2 1. = O. . 1.14084 . 9-1 = 1. 225 B . .l .C O N T I N U E D FRACTIONS 3.3 Convergents The convergents to the infinite continued fract ion x=at+a2 1 +G + 1 = [al. table of convergents is as follows: The In this connection it is interesting t o note that the earliest approximation to a is to be found in the Rhind Papyrus preserved in the British Museum and dated about 1700 B. the value of u stated there is 3. 1. These convergents give successively better approximations t o 'A. . C . a 9 as.14285 .

Discussion Problern. 4 . 41 = d 6 (b) [5. The following is one of the classical straight-edge and compass problems. 5.4s in the second half of Example 2. it possible t o construct the length n we could then construct l/?r 113' the following means: Let AB = r . Construct. . 2. . Then x = BD = prove this use the similar triangles ABU and CBD. 1)raw BU perpendicular to AC. 1. 2 . SO a square with the same area would have a side Were equal to 6. square equal in area to a circle of radius 1. see To Figure 2 . A circle of radius 1 has an area A = r r 2 = A.IRRATIONAL NUMBERS the very limited means at his disposal. 101 = 6 3.2. - (a) [2. a. 2 . Figure 2 .V 1 5 17 = [1. Problem Set 9 1. G] (d) 24 . 1. Section 3. 4 . using only a straightedge and cornpass. More information on the use of continued fractions to give rational approximations t o irrational niimbers will be taken u p in C h a p t e r 5. 4 . The approximation is correct to six decimal places. verify that the following continued fractions represent the irrational numbers written on the right. Verify the following expansions and calculate the first five convergents: 7 (a) d = [ 2 .6 ] = [2. BC = 1 and draw a semicircle with center a t O and passing through A and C . 1. fi] 2.

and also verify that the con- both numerators and denominators being formed from the sequence of Fibonacci numbers Each of these numbers is the sum of the preceding two.4F = 2.141592 . Then prove that . 1. + + Figure 3 4. However. FG parallel to E 0 and FH ~)arallelto DG. there are many interesting approximate constructions. 2. Show that vergents are : d+ 1) = [l. and i t only remains to construct a line equal in length to 3 AH.3. During the second. can be 5. and so on. The Fibonacci numbers Fi = 1. F3 = 2. Skctcli such a trec after a five-year growing . Let O D = i. A discussion of these interesting numbers will be given in Section 3. Let . then branches again. discussed a t the end of Section 3. 1 .CONTINUED FRACTIONS It can be proved that a length equal to ñ cannot be constructed with straightedge and compass. then "rests" for ayear. F = 3. 4 into this formula.10. F 2 = 1. 3. Since e • the approximation to a can easily be constructed as follows: Let O be the center of ti circle with radius OE = 1. Imagine that each branch of a certain tree has the following pattern of growth. I t produces no new branches during its first year of growth. 6. (i . it puts forth one braiich. Verify this by substituting n = 1.1. For example. 3. d iii the general formula reproduced by substituting n = 1. Jakob de Gelder in 1849 gave the following construction using the convergent +#= 3.4H = 42/(72 82).iB be a diameter perand . see Figure 3. 2. Draw pendicular to OR. 1.

if we regard the trunk and its extensions a s branches..4 Additional Theorems on Convergents The nurnerators p.3. in the second year two branches. satisfy the f u n d a m e n t a l recurrence relation (3. he \vil1 win the game. m 7. and in general the number of branches will reproduce the Fibonacci numbers 1. So unless A makes a mistake. 19 (1953). a n d denominators qn of the convergents c. 8. 8. construct a point G on a line segment AB such that (AG) = dGB). Using only a straightedge and compass. S. of the infinite simple continued fraction [al.4). Use tlie results of Problem 8 to show hoiv to construct a rcgiilar pentagon using only a straightedgc and compass. then he must remove a n equal number of counters from each. = p.IRRATIONAL NUMBERS 61 period and show that. +(di + + 9. M. t h e proof given there being independent of . If he wishes to take counters from both heaps. 3. vol. az. .l)". then in the first year of the tree's growth it has one branch (the trunk). * .4. = -1 (. Scripta Mathematica. Alternately two players A and B remove counters from two heaps according t o the follo~ving rules: At his turn a player may take any number of counters from the first or froin the second heap. For more details about this game and related subjects see H. Wythoff). Phyllotaxis. 2. I t can be proved that the nth pair of numbers forming a safe combination is given by where í = 1) and where { z ] stands for the greatest integer less than or equal to x. 2. pp./q. A. and TVythofs Garne. 3. where T = i(1 4).2) pnqn-i . Coxeter: The Golden Section. and A can always convert this back into a safe combination (safe for . The player who takes the Iast counter from the table wins. In order for player A to win he should. m . proved in Thcorern 1. IYythoJ's game (invented in 1907 by W. a.Pn-lQn n 2 0. leave one of the folIowing safe combinations (safe for -4): Then no matter what B does in the next move he \vil1 leave a n unsafe combinatio~i(unsafe for B). after his move. 135-143. 5. Verify this statement for n = 1.

equation (3. upon dividing both sides by qnqn-i.2. . we see that respectively.un(- 1).2. This proves Theorem 3. These theorems give us important information as to how the convergents cn change as n increases.~ n . we find that Since cn = pn/qn. These inequalities show that (3-4) CI < c2 and that c3 < cq. and recall that the qn's are positive.1.3) can be stated as Similarly we can prove THEOREM 3.62 CONTINUED FRACTIONS whether the continued fraction was finite or infinite.-l Qnqn-2 t n 2 3. Cn . substitute obtaining where the last equality follows from equation (3. PROOFClearly .2 .1.2) with n replaced by n . In the numerator on the right. From this equation. If we set 7% = 2 and then n = 3 in Theorem 3.

in succession.1. Eies between the two preceding cmvergents. f o r m a decrea* sequence.2. using n = 3. setting n = 3 in Theorem 3.4) proves that Similarly. q ~ a3 are al1 positive numbers. Moreouer. n 2 3. continued fraction fornz a n increasing sequence. Hence c~ < cg.2 shows that . Problem Set 10 1. as we have seen. The conversion of an irrational number x into an infinite cont inued fraction gave. we see that C3 < C 4 < C2.IRRATIONAL NUMBERS 63 On the other hand. and every odd convergent i s lesa than a n y even convergent.+l of u n inJinzte simple 3. . followed by n = 4 in Theorem 3. Give a numerical verification of Theorem 3. and the even convergents cz. since qa.. and combining this result with those in (3.3 using the convergents to d . each convergent c. we obtain the inequalities Combining these inequalities we obtain the fundamental result We state it as a theorem: THEOREM T h e odd convergents c2. i 36 Some Notions of a L M . then n = 4 in Theorem 3. Proceeding step by step in this fashion..3.

The implication is that we can somehow carry out an infinite number of operations and thereby arrive a t a certain number which is asserted to be x. if we add the numbers 1. To make this clear. is irrational. We shall see. Realizing this. that the only way to attach a rnathe~naticalvzeaning to such an infinite process i s to introduce the notion of a linzit. . Which of the follolving infinite surns have meaning? Clearly. so we say the sum A becornes infinite as the number of terms added increases indefinitely. one is tempted to write (as we did) which irnplies that the infinite continued fraction on the right actually represents the irrational number x. and such a result is not of much use to us. I t is advisable to reflect on the fneaning of such a statement.1)st calculation we had where x. On the other hand. +! 1 1 T ?8.CON'I'INU'ED F R A C T I O N S so that. let us first go back to ordinary addition. we get i succession the partial sutns n . the given irrational number. we can make the "sum" as large as we please. a t the end of the (n . and we saw that the process could be continued indefinitely. however. if we add 1 to itself over and over.

that is.a.IRRATIONAL NUMBERS which can be represented graphically as shown in Figure 4. as n . 1 In order to prove that they continua!ly approsch this upper limit 2. (+)"-l . . or approaches 2 as a limit. they are al1 bounded above by the constant 2. where so that the partial sums continua2ly increase. we w ~ t e so that Subtracting the second line from the first. But each partial sum sn is less than 2 . approaches zero. that is. we obtain which implies that As n increases indefinitely. and so S. geb closer and closer to 2.

increasing sequence of numbers al1 bounded above by the convergent c 2 = U. c3. then they hable a linzit l ~ where EL 2 L. the limit Ev must be a number less than al1 the even convergents. since al1 odd convergents are less than al1 t.. We say that converges to the value 2 as n -+ e .c2.4. and i. On the other hand. 53. a t For a discussion of limits of sequences. form a decreasing sequence of numbers al1 bounded below by the convergent cl = L. 2s Zess than U. S?. hence they will converge to a limit Eu 5 U.66 CONTINUED FRACTIONS S. Zippin [15]. Our task is to attach a meaning to the infinite continued fraction form an Theorem 3. . . that is. S Z . sa.3 states that the odd convergents cl. It also illustrates a fundamental theorem of analysis whích we state but do not attempt to pr0ve. We then assign this limit 2 as the valw of the infinite sum in question and we write This illustrates. see L.heeven convergents. . S .t THEOREM I f a sequence o j numbers si. cb. cs. which alao treats this fundamental theorem of analysis (Theorem 3. . the mathematical notion of a limit needed to attach meaning to an infinite continued fraction. continually decrease but are all .f jor each n. or in symbols. s3. continually increases. Moreover.haue a limit Eu. 3. 36 Infinite Continued Fractions . that is.c4. then the numbers SI. where U is some $sed number. 8 2 . in an admittedly rough fashion.4). greater than L. lim n+ S. where 1~ 5 U. = 2. We return to the discussion of infinite simple continued fractions. the even convergents c2. If the nunzbers SI.

that the q. . If lu # EL we would be in trouble. (.5. the fraction 1/ ~ ~ approaches~xero as k approaches infinity. Biit then from ~ q ~ . > Figure 5 To this end.1 by 2k . where ZL is a number greater than every odd convergent.c2k-1 approaches zero as h.IRRATIONAL NUMBERS 67 so that even convergents approach the limit EL L.~ equlttion (3.e) the denominator q2kq2k-1 of the fraction in (3. however.6) increases ivithout bound as k increases. Looking a t the convergents graphically (see Figure S ) .n 2 1) is a positive integer.'s increase without bound as n iricreases.he a m e limiting value 2 = lu = ZL. we see that all we have proved so far is that the even convergents have a lirnit ZI. We have proved: 3. How far have we progressed? 1s this limit E the same niimbcr x which gave rise to the continued fraction in the first place? Actually it is. Every infinite simple continued fraction converges THEOREM lo a limit E which is greater than any odd convergent and Eess than any even convergent. that is. retuin to Theorem 3. Weget The numbers q.6) we conclude that the difference C2k . but this must be proved. are calculated by means of the recurrence relation therefore it follows. and the only way this can happen is for both C28 and c2k-~ to have t. 13enc. and the odd convergents havc a fmit Eu. that ZU = EL.1.1 and replace n by 26 and 3 n . since each a. We can prove. (n 2 2) and each q . approaches infinity.

51. where s. that is. is the "rest" of the fraction. (3.9 it follows that xn+ 1 anf 1 so.8) shows that n Again.+i. Similarly. To this end. according to the second line i (3. we see that The next step in the proof is to show that x lies between cn and c. and return to the expansion (3. we compare the three expressions: . =an+- I 9 Zn+i where again The second line in (3.68 CONTINUED FRACTIONS To do so.7). combining these results. 1 1 and since . let z he the givcn irrational niimber.< -.7) shows that since Xn+l is positive.

On the other hand. we see = al 1 / x 2 . and since that c i < x.9) cl = + Thus .9) gives al < x l . hence al and x l = x. narnely 1 -r an 1 -r Xn and 1 a. (3.9) we know that and we can conclude frorn (3. where b y (3. either A direct calculation shows that for.IRRATIONAL NUMBERS We first observe that these expressions have the term in common so that it is necessary only to compare the terms in which they differ. x a2 < 2 2 or l / x z < l / a 2 . 1 +-an+i But by (3.10) that x will always lie between two consecutive convergents cn and cn+1. that is.

between C J and c4> between cr and Ca. Therefore it is permissible to write and we have proved THEQREM If un irrational number x i s expanded i n t o un 3.. . a. .6. a2. approach x from the left. This is true.7 Approximation Theorems Our experience with continued fractions and in particular our study of Theorern 3. Before stating such a result as a theorem we make some preliminary remarks.6 have supplied ample evidence that each convergent in the continued fraction expansion of an irrational numher x is nearer to the value of x than is the preceding cohvergent. equations (3. and so on. then the tirnit to which the conuergmts ci. Since al1 odd convergents are less than al1 even convergents. in expanded forrn. c4. that Thus we see that the convergents cl. ca. . . 3. and c2.of the fraction [al.9 m m Cn. hence x and E must be one and the %me. But we know that as k increases indefinitely. a2.. U n . infinik simple continued fraction [ai.70 CONTINUED FRACTIONS Similarly. and the reader will find it impossible to expand. as explained. ' . any given irrational numbers in two different ways. c2. This theorem should be followed by an additional theorem stating that the expansion of any irrational number into an infinite simple continued fraction is unique.10) show that x lies between c2 and cs. a ] according to the rules described. Let the expansion of the irrational number x be . approach x fronl the right. . ive are forced to the conclusion that or. . the odd convergents ctk-1 and the even convergents C2k approach a lirnit 1. - ] converge i s the number x which gaue rise to the j r a c t i m in the first place.

. equation (3. an+2. pn. a?. however. the given irrational number. . if we calculate convergents in the usual manner. a. we write (3. as before. by analogy with our study of finite continued fractions.-1. . Then. and consequently we have no right to treat it as though it were a legitimate partial quotient. when n = 0. are al1 positive numbers. this should be equal to x. yn-l depend only upon the integers al. p. In particular. qn. (3. the last "convergent" (in Theorem 1. Thus it seems reasonable to write where.I+Ias a legitimate partial quotient. Suppose.+1 contains an infinite number of integral partial quotients un+*. When n = I. . also note that xl = x.12) gives and by definition. While x. .IRRATIONAL NUMBERS where We assume that ~ 2 ~ x 3 . it should be stressed. take i = n 1 and an+1 = xn+l) woiild be + @ and.12) gives . it need not itself be an integer.3.11) in the f o m of a "finite" continued fraction aild treat X. .

according to (3. 11 = 7. . 1-71 7 - if a < 0.t hence We know that for n '_ 2. then la( = lb1 tcl.12) holds for al1 n can be proved in exactly the same way as we proved Theorem 1.12)) and from this we obtain or.+~> 1. rneans a ifa > O . hence > q n . r e d "absoiute value of tal c a". the expansion of the given irrational number x be Let where Then.tinuedfraction than is the preceding convergent.72 C O N T I N U E D FRACTIONS That (3.a For example. and 1-al = lal. we obtain Now if a = b . simple con. t The syrnbol la[. x. we have for n 2 2 Dividing through by xn+iqn. the successive steps being nearly identical.3.~> O . la1 = . We are now ready to state the main theorem of this section: THEOREM Each convergeni is nearer to the value of un infinite 3. 7 .c . rearranging. PROOF.7. and that q.

nqn+t < 1/~2.7 that x is closer t.IRRATIONAL NUMBERS and so Thus (3. we know already from Theorem 3.13) shows that or.+i than it is to cn. . and c. This shows that cn is closer to x than is c.+1. that + Taking the absolute value of both sides. or estanate.o c. we know from Theorem 3. so that c. of just how closely c. this tells us that Figiire G Moreover. and it follows that the absolute value of the difference between x and cn will always be greater than one-half the absolute value of the difference between c.+i > q n J Hence we can state qnqn+i > 4R and SO I/q. is to the left of c. In fact. with n repIaced by n 1. It would be interesting to have some measure. and the theorem is proved..1. Clearly AB < AC < AD.-1. This becomes clear if the situation is studied graphícally. what is the same thing. approximates x.+I. or Since q. Figure 6 shows the case when n i s odd.

such that This is the beginning of the theory of rational approximation t o irrational numbers. ( p . actiially approaches a limit can be suggested by numerical evidence: The number e ia taken as the base of the system of natural logarz'thms.74 C O N T I N U E D FRACTIONS If x is irrational. satisfying Theorem 3. Thus we have the following theorem : T W E ~ R E M I f x i s irrationul. Commenta. a decimal fraction which approximates e correctly to six decimal places.just as 10 is used as the base for cmrnon logarithms.718282 * - give better and better approximations to this number. EXAMPLEShow that the first few convergents to the number 1.. . there exists an infinite number of convergents pn/q.9. q > O. q ) = 1. The continued fraction expansion of e is the proof is quite difficult. These convergents should be calculated by finding the first few convergents to 2.718282. a subject we shall discuss briefly in Chapter 5. there exists un injinite number of 3. t! = 2.8. rational fractions p / q . The irrational niimber e arises quite naturally in the atudy of calculus and is defined as That the sequence of numbers (1 + (1 +) (1+ !)') .

--e The inequality o Theorem 3. Convert convergen&. S ] . 1. . find a fraction with a smaller EXAMPLE Given the fraction +&#ea6-. rnight be considerably improved.IRRATIONAL NUMBERS 75 SOLUTION. 2. that in succession these give better and better approximations to e . The tsblc givcs the numerical resulta: w- . and this sugvalue of e gests that Theorem 3. I n the next f example we s h l l approximate a rational number. 1. notice that p7/q7 = hence it should be true that w. We shall see in Chapter 5 that this is indeed the case. As a check on Theorem 3. or being content with the approximation we find that e = [2.00065746. The corresponding convergents are and a conversion to decimals shows. 2. 4 . We observe that the and this is certainly less than 1/39" is approximately one-half that of 1/39%. numerator and a smaller denominator whose value approximates that of the given fraction correctly to t hree decimal places. into a continued fraction and calculate the SOL~ION.9.9. 1. Assurning the above expansion for e. . A numerical calculation shows that 0.8 is true for rational or irrational x. 1 .. indeed. regarded as an approximation theorem.

.+.P5 --197 q5 will suffice.q.CONTINUED FRACTIONS Referring now to Theorem 3. Expand fiinto an infinite simple eontinued fraction and find a fraction which will approximate 6 9 mith accuracy to four decimal places. 7 . 3.+i > 1 / e . qn Problem Set 11 1./q. = pn/qn and c+i = pn+l/q. Given the fraction find a fraction with a smaller numerator and a smaller denominator whose value approsimates that of the given fraction correctly to three decimal places.Note that if we had worked with the fraction l / q : instead of l/qnq. Use Theorem 3. 1 . < e .q.8. M+.y 87 38 p4 q4 451 . that is. 1 . 1 . witl-i an error less thau half a unit in the fourth decimal place. we wish to approximttte Z#'$J.+i our answer would have been the next convergent +S.= .8 to investigate how closely the first four convcrgcnts approximate A. A little esperimentation soon shows that . 15.+l which will make That is. with an error of less than 5 units in the fourth place. where e is any given number. a]. 292. folloning this by an use a table of squares and first check that additional check to see if q. such that l/q. for Hence the required fraction is 3%. The continued fraction expansion of n is [3.0005.q. In order to find values of q. 2. we could > 1 / e . 1 .by p. since 1/38' is not less than 0. we search for two convergents e.+.

1907. called lattice points.he line.IRRATIONAL NUMBERS 3. and a = y/x would be a rational number. to t. I we move the thread f away from the line in the other direction. some of whose works are available today in reprint form. .]. See Figure 7. . toward the left. At these points. t F.he thread in our hand. a?. respectively. . if it did there would be a point (x? y) with integral coordinates satisfying the equation y = ax. 17-25. We pul1 the thread taut so that the end in oiir hand is at the origin. we move our hand away from the origin. . the thread will catch on certain pegs above t.he odd convergents. . The pegs contacted by the thread on the lower side are situated at tlhe lattice points with coordinates and correspond. 9 Cn. and that we hold the other end of t.8 Geometrical Interpretation of Continued Fractions A striking geometrical interpretation of the manner in which the convergents ci. . pp. . Felix Klein was not only a prominent mathematician but a most popular mathematical expositor. imagine that pegs or pins are inserted. a. assume a positive. This is impossible since a! is irrational. Klcin : A ~trycrodhlleKapitel def. Keeping the thrcad taitt. .. Zahledheorie. Let a be an irrational niimber whose expansion is [al. and whose convergents are 1:or simplicity. Kext plot the line p = ax. and on graph paper mark with dots al1 points (x. This line does not pass through any of the lattice points. c2. it wilI catch on certain other pegs. of a continued fraction for an irrational number converge to the value of the given number was given by FeIix Kleint in 1897. Now imagine that a piece of thin blactk thread is tied to an infinitely remote point on the line y = ax. . for.. y) whose coordinates x and ' 1 ~ are positive integers. Teubner.

The pegs contacted above the line are sitilated a t the lattice points .78 CONTINUED FRACTIONS O T 2 3 4 5 6 Figure 7 which are al1 less than a.

(5. Those points corresponding to the even convergents are (1. 1 Y3!4 (Chapter 8).IRRATIONAL NUMBERS corresponding t o the even convergents. EXAMPLE. p4) = (3. . 8 R. New York: The MaoiniIIan Compaiiy. for example. or 5/3 > g/3 = a. 13). I n fact. y) marked in Figure 7. 5). 21. t See . 1. and are al1 below the line. (8. 3).lost of the elementary properties of continued fractions have geometrical interpretations. and are above the line. Draw a Klein diagram for the continued fraction espansion of 1 1 + d 5 a= = [1. Hancock. 2 - -1. that the point (qr. l). the theory of simple continued fractions can bc developed geometrically. Let us show. (di - 2. (2. Development of the Minkowski Geomelry of Numhers. or a = 4/3. we see that y = a 3. Since it is on the line y = ax. see Figure 7. Consider the point (3. convergents are The The points or pegs corresponding to the odd convergents are (1. The point (3. Construct a Klcin diagram for the continued fraction expansion of 1)/2.? Problem Set 12 1. 8). (3. 1. 1. hence the corivergent 5/3 > a. al1 of which are greater than a. . Construet a Klein diagrarn for the eontinued fraction expansion of 2/3. 5) corresponds to the even convergent p4/q4 =. Each of the two positions of the string forms a polygonal path which approaches the line y = ax more and more closely the farther out we go. . H. 5) is above the line so 5 > 4. which is greater than a. SOLUTION. 5/3.

1. 1. getting so that For example. when a = 1.3. (a) x2 . provided.]. better and better approximations to the actual solution i ( 1 See also Problem 4 of Section 3. we have only to divide both sides of (3. We shall now examine the quadratic polynomial equation the positive root of any quadratic equation of the form (3. . that it has such a root. . of course.14) has the continued fraction expansion If a > 0.5x -1=0 .1 = O (1)) x2 . and the successive convergei~ts this continued fraction will give to 45). + Problem Set 13 1. A more detailed discussion of this particular number foIlows in the next section. Use the quadratic fo~mmulato find the positive roots o the following f equations and compare the exact solutions with the approximate solutions obtained by computing the first few convergents to the continiied fraction expansioiis of these positive roots. the equation x2=x+1 has a positlive root x = [1. To see this.3x .CONTINUED FRACTIONS 3.14) by x.9 Solution of the Equation x2 = ax +1 Continued fractions can be used to approximate the positive root of any polynomial equation. 1.

p n / ~ n . by giving the positive integers a and b particular values. and has the va. Suppose that and that b is a multiple of a.10 Fibonacci Wumbers The simplest of al1 infinite simple continued fractions is T = [ l .2.2/q. that is b Show that then x satisfies the equation 22-bx-c-O = ac (where c is an integer).liie 3. p n + 2 / q n + 2. and by selecting particular convergents pn.IRRATIONAL NUMBERS 2. that if 3. where T satisfies the equation which has the positive root The convergents to T are both numerators and denominators being formed from t h e sequence hof integers ... l ?1 . Verify.

Thus a line segment is said to be divided according to the golden mean if one part is T times the other.. after the first two. In geometry. especially in the printing and advertising crafts. which has many forms. although he was not the first to use them. The Greeks claimed that the creations of nature and art owed their beauty to certain underlying mathematical patterns. it arises from what some cal1 the "most pleasing" division of a line segment AR by a point C .16) are known as the Fibonacci nu~nbers. The figures and drawings were made by Leonardo da Vinci. Luca Pacioli pubIished a book. The numbers (3. + + If we now let fc = b / a . the ratio 3 to 5 is approximately eqiial to the ratio 1 to T. . In this book Pacioli described thirteen interesting properties of T. in the proportions of the human body. and so on. l'or example. whose sides are in the approximate ratio 1 to T.82 CONTINUED FRACTIONS Each of these numbers. Witness the popularit$ of the 3 X 5 index card. One of these was the law of the golden nzean. Divina Proportione. aesthetically. The golden mean appears a t many unexpected turns: in the pentagonal symmetry of certain flowers and marine animals. is equal to the sum of the preceding two. devoted to a study of the number T . 3 = 2 1. or golden section. thus 2 = 1 1. 1170-1250). This is said to be attained by selecting a point C such that the ratio of the parts a to b (see I'igure 8) is the same as the ratio of b to the whole segment a + b. + I n 1509. Le. we have so that z = b/a = f (1 1/5) = T. or b = Tu. named after the great thirteenth century mathernatician Leonardo Fibonacci (c. Man has employed the golden mean in the creative arts and in various aspects of conternporary design. the majority of people considers that rectangle to be most pleasing. and so on.

q ( p = q6) as the dirnensions of our square. q6 as the dimensions of our rectangle. while that of the rectangle with what seem to be the same component parts is 5 13 = 65. The number T occurs in connection with many mathematical games. as shown in Figure 9a.D do not lie on a straight line but are the vertices of a parallelogram ABCD (for a n exaggerated picture of the situation see Figure 9b) whose area is exactly equaI to the "extra" unit of area. o Figure 9a This puzzle is based on the fbllowing facts: The convergents (3. . the obtuse a n d e s ADC and ABC differ from straight angles by less than f +O. and p6. which. B. The most familiar. and the above rela6 ~ tion tells us that the areas of these figures differ by only one unit.8 = 64. for n = 6. C.15) have the property that the denominator of each is the numerator of the previous one. is the one involving a square 8 units by 8. Ir1 case of the rectangle of Figure 9a. I n particular.IRRATIONAL NUMBERS 83 In geometry. The area ~f the square is 8 . can seemingly be broken up and fitted together again to form a rectangle 5 by 13. and the convergents to T also occur in connection with certain geometrical deceptions. the points A. perhaps. so that somehow the area has been increased by 1 unit. Now consider the relation which in this case. the golden mean is the key to the construction of the regular pentagon. becomes We have chosen p6. Actually.

I Ftn is large (say F 2 n = 144. and Fk = Fk-2 + Fk-1 for k > 2. pp. . if the Fibonacci numbers are defined by the relations F x = 1. then it can be shown that when the parts are reassembled to form a rectangle. a hole in the shape of a parallelogram ABCD of unit area will appear and the altitude of this parallelogram is 1 / 1 / ~ : . t This section is rather technical and may be omitted without loes of continuity. No. 8. . then the hole is so f narrow that it is difficult indeed to detect it. + Fzn Figure 9c 3. fMathematica2 Tab2es and Other Aids to Computation. 45. describes a method for calculating logarithms which is worth recording because of its adaptability to high-speed computing machines. 60-641. April 1954. FZn-2 = 55). in a journal devoted to numerical comput~tions. F 2 = 1. Vol.CONTINUED FRACTIONS Figure 9b More generally.~. . F . (with even siibscript) is divided into parts as shown in Figure 9c.11 A Method for Calcuiating Logaritbmst 1 Daniel Shanks. and if a square with a side equal to a Fibonacci number F2.

Calculate . and the sequence of positive integers where the numbers n l . be.17) where l / x l bo = b1 ni+. nz. we then calculate bz = - bo bf' and determine a n integer nz for which b$Qbl<b2 ntf 1 . bz.l. then The procedure is now continued. If n2 is such an integer.+l. This shows that (3. we first find an integer n l such that o. the relations are determined by means of Thus.IRRATIONAL NUMBERS 85 To calculate the logarithm loa.' < bo < b. 1 ) < 1. bl to the base bo of a number bi (where 1 < bl < bo) we compute two sequences: bz) b3. ba.

18) we have On the other hand. Solving equation (3. bl notice that from equations (3. from (3. To see that we are actually calcuhting lo&. and hence we can write Similarly we can show that and so on.17) and (3.86 CONTINUED FRACTIONS and find an integer n3 such that b i a < b o < b 3na+l . . by the deíinition of a logarithm. whence and so on.19).17) for bi and using these results we have and so.

we find that so that ni = 3 and bg = 10/28 = 1. The paper by Shanks gives the following resulta: This shows that log 2 = 3+3+9+2+2+*** 1 1 1 1 1 = [O.024.the value of log 2 to 11 places is 0. 2. . 3.25)Q = 1. b~ = 2. SOLUTXON. Next we calculate the convergents: The convergent cs gives the approximation 0. . 3.30103093 . we see that Thus n2 = 3 and ba = 2/(1. With bo = 10.IRRATIONAL NUMBERS EXAMPLE. -1. 2. Subsequent calculations become more difficult but can easily be done with the aid of a desk calculator. 9. in general.30102999566. . each convergent approxirnafes log 2 to one more correct decimal place than does the previous convergent.25. I t can be shown that. Using a table of powers. Calculate loglo 2.

and that irrational numbers have non-terminating. or quadratic surds. It is not hard to show that . i.1 Introduction Our study so far has shown that rational numbers have finite continued fraction expansions. In aH the examples considered. + where. For example . the bar over the partial quotients indicates those numbers which are repeated indefinitely.CHAPTER FOUR Periodic Continued Fractions 4. as before. expansions. with irrational numbers of the forrn ~ k v 5 where P. Q. or they were periodic from some point onward. In Chapter 3 we dealt kainly with the expansion of qvadratic irrationals.e. Iike the expansion of f (1 d%)below. the expansions of such numbers were either purely periodic. or infinite.. D are integers and where D is positive and not a perfect square.

1 W ) . I n 1657 Fermat stated that equation (4.1) was given by Lagrange about 1766. every periodic continued fraction represents a quadratic irrational.1) is known as Pe2E7sequation. The most interesting proposed by Fermat as a challenge to English mathematicians of the time. The first complete discussion of (4. 2. at the age of thirteen. 1 1 i u wtw c*lccteda fellow of the Roya1 Society in 16&3. p. fe11 had mastered eight languages before he wsil twcnty.1) has infinitely many solutions. 3411.4 supplies a deeper study of reduced quadratic irrationals.5. equation (4. .1652).1 Many authors refer to the equation a s Fermat's equation.PERIODIC CONTINUED FRACTIONS 89 any purely periodic continuedfraction. This will be accomplished in several stages. References to indeterminate equations of the Pell type occur throughout the history of mathematics. t Lord Brouncker in the same year gave a systematic method for solving the equation. The aim of this chapter is the presentation of the proofs of these theorems. a t Rreda ( 1040. Carnbridge. and Section 4. and. was first proved by Lagrange in 1770. This is followed by the proof of Lagrange's theorem which states that the continued fraction expansion of any quadratic irrational is periodic from some point on. and where N is a given integer not a perfect square.2 and is followed by the proof for the general case. For a complete history of the subject see Dickson [4. and he was Crornwell's representative in Switzerland (10454. The chapter will end with a brief discussion of the indeterminate equation where x and y are unknown integers. an example is presented a t the beginning of Section 4. t Actually 1John Pell(1611-1685) was a great teacher and scholar. called a reduced quadratic irrational. The more difficult theorem. These sections contain the tools necessary for proving. conversely. that any quadratic irrational has a continued fraction expansion which i s periodic after a certain stage. vol. represents a quadratic irrational. but this is unjustified since Pell did not make any independent contribution to the subject. He waa profeasor of mathemat ica a t Amsterdam (1643lt)46). but he did not supply the proof. or any fraction which i s perz'odic frum so?12e point onward. Commonly. First it will be shown that a purely periodic continued fraction represents a quadratic irrational of a special kind.3 furnishes a more detailed discussion of quadratic irrationals. Section 4. that any reduced quadratic irrational has a purely periodic continued fraction expansion. Admitted to Trinity College. in Section 4.

See Heath f6.f The solution of this problem contains eight unknowns (each representing the nurcber of cattle of various kinds) which satisfy certain equations and conditions. 4. p. Eumbers represented by purely periodic continued fractions are quadratic irrationals of a particular kind. I n fact some historians doubt that the problem had any connection with Archimedes. Others. The smallest solution of the cattle problem corresponding to these values of x and y consists of numbers with hundreds of thousands of digits. New York: Sirnon and Rchuater. 3421. ( a ) A nu~nerical exa?nple. . Newrnan. 1211.90 CONTINUED FRACTIONS example arises i connection with the so-called "cattle problem" of n Archimedes. The problem can be reduced to the equation whose smallest solution involves numbers x and y with 45 and 41 digits respectively.2 Purely Periodic Continued Fractions Certain continued fractions. There is no evidence that the ancients came anywhere near to the solution of the problem. 187-188. like are periodic only after a certain stage. like are periodic from the beginning on and are called purely periodic continued fractions. vol. pp. Consider some purely periodic continued fraction. 1956. while others are convinced that it was propounded by Archimedes to Eratosthenes. 2. Dickson f4. p. such as 8 Vde can write t For a statement of the cattle problem see The World of Mdhemalics by James R. and we shall now investigate how these numbers can be distinguished from other quadratic irrationals.

There we showed that if where then where p .~ and pn/qn are the convergents corresponding. respectively. We form the table al = 3.. + a2 a3 1 ~ ~ ~ 1 . + a. t o the partial quotients anWland a.6) to the special case (4. as = 2. (4. a . I n the case of a purely periodic continued fraction a = [al.5) shows that a can be calculated from the equation We now apply (4. and that in calculating CY we can regard as though it were a legitimate partial quotient. using a* = 1. .+ .3) as though it were a finite continued fraction.~ / q . . I n effect.5) shows that we can treat (4.2). ] = al 1 1 1 1 + . a = 13.P E R I O D I C C O N T I N U E D FRACTIONS 91 It is now necessary to recal a result studied in Section 3.1 we see that and hence equation (4. 1.7. .a2) = . 21.+ .

we obtain This leads to the quadratic equation which is the same equation we would have obtained had we worked with equation (4. We now consider the number /3 obtained from a by reversing the period. 0 > 1.1l).92 C O N T I N U E D FRACTXONS Hence. where . and so a! and . Moreover. and so . has the positive root a and the negative root a = . a = 5+dZ 3 and a = ' 5 . This shows that the quadratic equation (4.9) can be written in the form Comparing (4. or (4.7) has two roots.7).1 < . that is. These roots cannot be equal are positive. the number Applying (4.1/@ have opposince both a and site signs.1/@.6) to 0.f i 3 .1 < a < 0.2).l/@< O. The quadratic formula shows that (4. ' ' It is easy to check these results numerically.7) and (4.10) we see that the quadratic equation has solutions x = a and x = -1/@. we get this leads to the quadratic equation Equation (4.

1. respectively.. so that .l/qi represent. . a2.].-¡.all. a 11 is the continued fraction for a with the period reversed. ~ ~ . a' lies between .1/@ = a is the ' second. 4. and. a = 3. Moreover.9) is and hence ' which shows that .-.694 > 1. PROOF.1)st conver~ents the continued fraction [a.if 8 = [a.361. a. the purely [al.l / @is equnl to a. a.1...[a. namely that if . and a = -0. are posz'tive integers. the nth and (n .-1. or conjugate root. al] = Q n 7 J and Where p:/qL and pk . . to three decimal places. a2. . then . y an] is greakr ¿han 1 and as ¿he positiue root of a quadratic equation with integral coeficients. (b) The general case. a..1 < a < 0. periodic continued fraction = .. require t ~ v oresiilts stated in Problem 7 of Set 3. equally inzportant. a. then Pn -- pn-I .a27 . of the quadratic equataon satisfied b y a.1 and 0. a2. P.PERIODIC CONTINUED FRACTIONS 93 The positive root @ of (4. We shall now prove THEOREM I f al. We page 26. of . . ' ' The purely periodic continued fraction a is indeed a quadratic irrational. a. kioreover.= [al. rln Pn . a2. ..

9.15) we can replace (4. a2./qk and p .1 are. we obtain and again.-l/q. equivalent to the quadratic equation Revetsing the period in a. . respectively. according to (4.-1 = O. it follows that Since a is purely periodic we can write it in the form and. Equation (4.6). Using the results stated in (4. . in the form where p. .]. respectively.-l are defined. and p.-1.-1) (. .p.16) is and (n . .19) qn (- ./qk .6). as the nth y a. P a.].94 CONTINUED FRACTIONS Since convergents are in their lowest terrns.. according to (4.f) .1)st convergents of [al.(p.l)st : convergents to [a./q..18) by so that fl satisfies the equation which is equivalent to the equation (4. the nth and (n . a. we see that where p.

though the result was implicit in the earlier work of Lagrange. bz. < ' the root a = . such as a = [al.NOW. a.1//3 lies between 1 and O. while in caee (iii) the second root is necessarily greater than O. and if the second root &' of this quadratic equation lies between . (ii) Those with an acyclic part consisting of a single quotient al. 0 = [al. ba. .-1.1/@. thus we have p > 1. and the root $2 = .]. . then the continued fraction expansion of a is purely periodic.1 and O. (iii) Those with an acyclic part containing at least two quotients. whose second root a lies between . such as a == [al. a. This remarkable fact was first proved by Galois in 1828. siich as . and hence satisfies a quadratic equation with integral coefficients.1 is also true (and will be proved in Section 4. bz. . In cases (ii) and (iii) it can ' also be proved that a is a quadratic irrational satisfying a quadratic equation with integral coefficients.17) and (4. .-1. bl. al are al1 positive integers.. . . . Simple recurring continued fractions may be grouped as follows: (i) Fractions which have no acyclic (or non-repeating) part. ' What is to be emphasized is that these few conditions on a and a conzpletely characterixe the nui~zberswhich have purely periodic continued fraction expansions. as. O < l/@ 1. where a. al].PERIODIC CONTINUED FRACTIONS 95 Comparing equations (4.].. We will not prove theae last two results. a. This means that if a > 1 is a quadratic irrational number. a2..19). In other words. . - m S We proved. bi. .5). that a is a quadratic irrational which satisfies a quadratic equation with integral coefficients.1 and O. as. for fractions of type (i). . This completes the proof.P stands for the purely periodic continued fraction [a.. a - The converse of Theorem 4. and so -1 < -1/@ < O. bn]. we conclude that the quadratic equation has two roots: The root XI = a.. b. but in case (ii) the second root a ' of this quadratic equation is either less than -1 or greater than O.

1 and 0. are irrational. then = Bt. (b) that y = 11. not a perfect square. (b) find the equation of which a is a root. and that a therefore lies between -1 and 0. TOprove this.3 Quadratic irrationals I n this section we shall be concerned mainly with nurnbers of the form where A and R are arbitrary rational numbers. does not lie between . aside from trivial variations such as + + In other words. for an arbitrary but fixed positive integer D. so that . If a = [2.A2 = O or A l = AS. 4. 2. write the above . First we observe that.Ri) 45. 31 satisfies an equation whose other root. Hence the assumption that B # Rp leads to a contradiction and we must conclude that i B . a'. Verify numerically (a) that a == [l. and therefore. if and only if Al = A 2 and BI equality in the form Ai if B2 f R1.A2 = (B2 . 1 (a) verify nurnerically that a > 1 and f? > 1.CONTINUED FRACTIONS Problem Set 14 l. 21. there is only one way of writing the number A B ds. of this equation satisfies the relation a' = -1/@. and where D is a fixed positive integer not a perfect square. 31 satisfies an equation whose other root. . a'. is positive. A l .\/E. = B2. and hence also A B d D . ' - - 2. b would be rational. (c) show that the other root. y'. 6 and f? = [6. 2. contrary to assumption.

a >O . if a # O. When we speak of quadratic irrationals.4ac. the result is again of this form. we clairn that when numbers of this form are combined by any of the elementary operations of arithmetic (addition. but cal1 attention to the fact that in this connection. i. subtraction. We leave the proofs of these properties to the reader (see Problem 1 of Set 15). division). Clearly if a = O. multiplication.PERIODIC CONTINUED FRACTIONS 97 Next. i s the root of a quadratic equation ax2 bx c = O. + + + + In order to prove the statement in italics we recall that any quadratic equation ax2 bx c = 0. + We prove next that every number x = A B 45. has roots + + where D = b2 . and consequently Hence. "numbers of the form A B " include those for which R = O. we shall assurne B # O. c are integers and where b2 .e.4ac > O. however. not a perfect square. since otherwise the number under consideration would be rational. where the coe& cients a > O. we can verify by direct substitution that . b. where A and R # O are rational and D i s a positive integer. ordinary rational numbers. x = -c/b would be rational and hence could not represent the irrational number A R 1/Ti. we can replace a s 2 + bx + c = O by c . Conversely.

the common denominator of the rational numbers 2A and A 2 .4ac is not a perfect square. a > 0.4ac of this last equation is positive . satisjies one and only one B quadratic equution as2 bx c = O where a. since D was assumed to be positive.B2D) = A2 k ZAR 45 + B2D .B 0 need not have integral coefñcients. for. c have nofactors in were a root of comrnon. Finally. For. # O.2A2 T 2AB Z/D + A 2 . but if we multiply through by a. Observe also that b2 . provided B tr O. The above discussion leads us to a precise definition of a quadratic irrationai. b = -2aA.ZA(A I B fi)+ (A2 . The numbers A B <D we have been dealing with are therefore al1 quadratic surds according to this definition. we obtain the quadratic equation where the three coefñcients .98 CONTINUED FRACTIONS = (and x (A A-B a) this last equation: satisfies A B 0 and A . The equation x2 . + B f i 1 2 . and c = a(A2 . or quudratic surd.B2D) = O satisfied by + + A quadrcatic surd A R 0. the discriminant b2 . if x = A R and also of g2(x) = + + + + 4s a2x2 + b2x + c2 = 0.R2D) are integers.aib2 # O. b a then it would also be a root of the equation Now if a2bl . b. then this would imply that .2Ax + (A2 .R2D = 0. it is a number which satisfies a quadratic equation whose coeficients are integers and whose discriminant is positive but not a perfect square.B2D.

leaving the rest as problems. difference. then 2. difference. The conjugate of the conjugate of a quadratic irrational number a is a. a' also satisfies this equation. or from consequence l. b2 = kbl. or quotient of their conjugates. or quotient of two quadratic surds al and a2 is equal. contrary to the assumption that x is irrational. The conjugate of the sum. and hence that that a2 = kal. (Why?) + bx + c = O. if a2bi .. In symbols. SO Every quadratic irrational has a conjugate at=A-Rdii formed by merely changing the sign of the coefñcient B of < . On the other hand. This follows directly from the definition of a conjugate. one being merely a constant multiple of the other. c2 = kcl and the two quadratic equations gl(x) = O and g2(z) = O are actually equivalent. - Al + Bi fi and a 2 = A2 +B 2 6 . . because a quadratic equation has only two roots. this means that (al az)' = al I &. We prove the first assertion. D This definition has a number of useful consequences: 1 If a satisfies the quadratic equation ax2 . 3.alc2 = 0.a1b2 = O then the equation + implies that a2c1 . Henoe in this case x = A R 1/D could not satisfy both equations. Thus if a. product. product.PERIODIC C O N T I N U E D F R A C T I O N S 99 is rational. respectively. to the sum.

+ .1112. B. Show that.0 2 . l l h ) + A2) + (Bl + B ~ ). then 01 ffa. a2 as in Problem 1. B1. N are positive integers. if al = Al Bl al = A2 B2 (where Al.B 2 G ) and cornparing the two results we see that Problern Set 15 1. Using the same representation of a.\/liÍ/.a:. has roots .\/E + + I 2. (al a2)' = a. shoiv that + (al . the surn of the conjugates is = (Al -B + (A2 . B. ai . prove that A = B = C = 0.\/E]' On the other hand.100 C O N T I N U E D FRACTIONS then the conjugate of the surn is (al + a2)' = [(Al = (Al + A21 . + 6. al/aa (a2 # O). 01 . not perfect squarea.4 Reduced Quadratic Irrationals The quadratic equation where a. If A B fi C fi = O and if A.B2 are rational and D is a positive integer. c are integers. snch that . and 3..az)' = al . + + 4. caneach beexpressed in the form A B fi with rational A. not a perfect square). A2. and denoting the coiijugate of a by a'. a:. b. C are rational and M .\/F is not rational.(Ri + Be) h.

or that O < P < fi. We have shown that if a is a reduced quadratic irrational of the f o m (4. given ' by (4.P < Q < V% and + P < 2 . . fi .1 Finally we that a > 1 < O. that u = a. >1 and and a ' -1 <a'= P . D will be as defined by (4. Under these assumptions the quadratic irrational a given by (4. i.1 and 0.\/»> shows that observe that -Q.20).23) O < P < fi. Q. I we assume that D > O is not a perfect square. Suppose. The implies that P + 0 > Q . Q. then the integers P. that the value of a given by (4. Throughout the rest of this chapter. D satisfy the conditions (4.21).1 irnply that a ! + a' > O.20) i s sazd to be reduced if a i s greater than 1and if its conjugate a. P + f i Q a ! > 1.f i - Q < 0. P. then the roots a and u are quadratic surds of the form A f B ' where A = P/Q and B = 1/Q are rational. we conclude that P > O. and since Q > O. from and it follows inequality a'> . It is important in what follows to find out more about the form and properties of reduced quadratic irrationals.20) is a reduced quadratic irrational..e. or The conditions > . Also.\/D. then.PERIODIC CONTINUED FRACTIONS 101 and where f are integers. lies between .22). and the inequality PP-. or fl-P<Q.

for.4ac > O not a perfect square. However. there is only a finite number of positive integers P and Q such that P<1/Dand Q < 2 0 . This follows directly from the inequalities (4. for any given D > 1. a > O. To establish this result. namely + where X is the largest integer lessthan With this determination of A. once D is fixed. l/al satisfies the + + + + . there exists always at least one reduced quadratic surd assoez'ated wz'th it. This idea. and is intimately related t o the theory of reduced quadratic forms. Q = 2a.20). and where al is again a reduced quadratic surd. The quadratic equation satisfied by a and a' is + 6.23). however.22). i t muy be expressed in the form where al i s t h largest integer less than a. the importance of the idea depends upon the fact that for a n y given D there i s only a Jinite number of reduced quadratic surds of the form (4.102 CONTINUED FRACTIONS The reason for introducing the notion of reduced quadratic surds has not been explained. where al is the greatest integer less than a.0 obeys . c are integers. It is necessary to have the following result: I f a i s a reduced quadratic surd. not a perfect square. Write cw in the form a = al l/ai.1 < a' < O. Clearly a = a. where a. is a well-established concept in the theory of numbers. For our purpose. b. X 0 = a is clearly greater than 1. P = -b. see (4. Could it happen that there are no reduced quadratic surds of the form ( P 1/D )/Q associated with a given D? If so we might be talking about an empty set of reduced surds. and its conjugate a' = X . let the reduced quadratic surd a be the root of the equation ax2 bx c = O. and D = b2 .

1 < a < O. QI.23) are automatically inherited by P1. and it only remains to prove that . It will now be shown that al is a reduced quadratic surd. for. as required. QI = 2(aa: + bu1 + c). so al > 1.1/ar is also a reduced quadratic irrational. .(2aa1 + b). therefore O < l/al < 1. we obtain where Pi = and . we prove that if a is a reduced quadratic irrational. To this end. and hence the inequalities (4.PERIODIC CONTINUED FRACTIONS 103 quadratic equation Solving for the positive root al. It is also clear that Pi. . we recall that al is the greatest integer less than a .1 < 4 < O. by hxpotheses. then its asaociatc /3 = . and DI = D are integers.1 < a: < O. we obtain + Theref ore ' since al 2 1 and. or . Solving the equation a = al (I/al) for al and taking the conjugate of the result (see page 99).and DI = D. Finally. QI. Thus al is a reduced quadratric surd. These expressions give us the expficit form of al. and has the same irrational part t/D as a has. If follows that O < -a: < 1.

+ the fornl 4.23) imply that 1 < a and . then we show that this expansion is necessarily purely periodic . Determine al1 the reduced quadratic irrationals of (P f i 3 ) / & . i f + 2. This we established in Section 4.1 < a' < O imply that /3 that 0' = . prove that conditions (4. then al is a reduced quadratic irrational.4. . and Problem Set 16 1. PROOF. and where is again a reduced quadratic irrational associated with D.1 quadratic irrational. + ( l l a l ) . In other words. . so that a > 1 i s the root of a quadratic equation with integral coejiaents whose conjugate root U!' lies between .1 We are now ready to prove THEOREM (CONVERSE THEOREM ) . > 1. then the continued fracticm for a i s purely periodic. If a i. where that.104 CONTINUED FRACTIONS the inequalities a > 1.1 and 0. .s a reduced 4. The first step is to express the reduced quadratic irrational U! in the form - where a l is the largest integer less than a.6 Converse of Theorem 4. Show a = al a = t ( 5 4 1 is expressed in the form 3) al is the largest integer le^ than a . Shoiv that the conditions (4.23) are necessary and sufficient conditions for a [defined by equation (4.1 and 0. < af < o.20)]to be a reduced quadratic irrational.2 OF 4.l / a lies between . 3 .l . first investigate the actual expansion of a into a We continued fraction.

PERIODIC CONTINUED FRACTIONS 105 Step (4.24) is the first step in converting U! into a continued fraction. and where Since u is irrational this procesa never comes to an end. Repeating the procesa on al. and is a reduced quadratic irrational. and Continuing the process. al. At this stage we have where a. al. are al1 reduced quadratic irrationals associated with D.we obtain where a2 is the largest integer less than al. a2. and hence we seemingly are gerierating an infinite number of reduced surds . a2 are reduced. we generate step-by-step a string of equations where a0 = a.

For this purpose. from which it follows that To prove (ii) we shall show that e = az+3. It then follows that the recipro~ l cals of aa+l and a ~ are equal and hence that a k + l = QZ+I. we obtain and and hence ... O<k<l. is repeated.. we have and taking conjugates. we use the conjugates of the equal complete quotients a k and al. al1 associated with D. al. . and. . we must arrive eventually at a n reduced surd which has occurred before. also yields a[k+2 = a1+2. in other words. a k = a~ implies a k + l = al+'^.. and that al is the first one whose value has occurred before.. l1 .4 that there can only be a finite number of reduced a l a associated with a given D . = a. in other words.. so that al = a k . a = ao.1 = a . al-I are different. -- . . Thk argument.106 ao. therefore. (ii) The very first complete quotient. a 2 . Suppose. a. ai. we merely recall that are the greatest integers less than a k = a*. ak-2 = al-%. ak+3 for O < k < 1 implies a k . al. To prove (i). . al1 subsequent completa quotients are repeated. obtaining oi. then.. QO = al-k. It is then possibIe to prove that: (i) Once a complete quotient is repeated. that i the sequence a11 the complete quotients ao. the sequence a = ao. when repeated. a k = al Now if k # O. a k + 2 = QZ+~. since ak+i and we may conclude tbat a k + = al+l.. . e C O N T I N U E D FRACTIONS But we proved in Section 4. is purely periodic.

ak-1 I and . also.t?k = 01. Pk-1 1 and o < -al-l t =P1-1 1 < 1. it follows that a k = al and hence.27) are the largest integers less than Pkj 01. NOW 7C . if a k ifJ not the very first complete quotient.27) Since so that 1 B k = a k + Pk-1 ak-1.. This shows that the a k . and since . in expanding the reduced quadratic irrational a into a continued fraction we generate the string of equations .7 = al .28) is and the right side is al-l.. and 61-1 al-1 are reduced. a 2 6 =al+1 1 1 which is the same as saying (4..he first a. if i. until we arrive a t t. ak-3 QE-3. we have and -1 -l<ak-i<O < a[-l < O o < -ak-1 t = -<1. al in (4. etc.ak .1 # O.~ = al-l.al-1.e.PERIODIC CONTINUED FRACTIONS 107 t 1 .7 - ffk . we may repeat this argument 7c times to prove that ak-2 = al-2. that Since the left side of (4. and obtain Thus. we have shown that a k = al implies a k . respectively.

. x" ia 1x1. . and then G to F(a.). will also repeat : Therefore. we have adoptcd thc t The traditional notation for "largest integer less than bracee hcre. a. al. a.] of a purely periodic continued fraction. This completes the proof of Theorem 4. a2. .+l. into l/an+r and G maps I/a.. that is. the reciprocal of the next a . a 2 . such that F maps a. and from this point on the a's repeat. observe that where ak+ is the largest integer less than a k .. To define the funct'ion F. to every ak. but since this conflicts with oiir notat.-1 are al1 different. and where a. F(x) and G(x). Since for every a k > 1. = a. . Before extending the proof to al1 quadratic irrationals (reduced and not reduced).. we shall obtain a. we present a graphical illustration of the periodic character of the complete quotients al. By first applying F to some a.+l into its reciprocal. it is clear that the sequence al. in the expressions We shall define two functions. S S .. a. there exists exactly one biggest integer ak less than a k . t'he continued fraction for a has the form - a = [al.108 CONTINUED FRACTIONS where a. a. a ~ .+l. Let the symbol { x ] denote the largest integer less than x. .ion for continued fract.2. t then we may write and we define the function F accordingly : 4 We now have a function which assigns. a2. .ions.

The projection of C 0nt. consists of one branch of the equilateral hyperbola y = l/x.PERIODIC CONTINUED FRACTIONS 109 h'ow. the appropriate definition of G is simply so that In other words. plot the functions F(x) = x . We then find the point on the graph of G(x) which has the same ordinate as the point B. since the reciprocal of the reciprocal of a number is the number itself. The graph of F(x) consists of the parallel line segments and the graph of G(x). for positive x. G[F(ak)l = W+I. Let a be the given quadratic irrational. namely l/crl.{xj and G(x) = l / x on the same graph paper.. Figure 10 In order to apply this scheme graphically. we cal1 this point C. hecause .e. see Figure 10.o the x-axis represents the value of al. to the point F(a) = B]. We locate it on the horizontal axis (poii~tA ) and find F(a) = l/ai by rneasuring the vertical distance from A so the graph of F ( x ) [i.

a double arrow from a to a 2 . eventually a reduced complete quotient a + is reached. . has a periodic continued fraction expansion.2. are periodic. then there will be a repetition and the a are i periodic. the abscissa of C' represenb the value of a . If. in the course of our path. Observe that the partial quotients al. Any quadratic irrational number EM a has a continued fraction expansion which is periodic frorn some point onward. going from A' to B' to C ' . al. then the path will eventually repeat itself . Problem Set 17 1. . we are led t o a l point on the hyperbola which was already covered by the earlier part of the path. if the a.5) we know that .110 CONTINUED FRACTIONS Starting with al. PROOF.we now repeat this procesa. etc. Shoiv that a 2.4]. 4. a single arrow leading from a to al. and . respectively. by equation (4. Show that = 1 d is reduced and verify that its expansion is 2 the purely periodic continued fraction [$l.3. central idea of the proof is to show that when any The quadratic irrational number a is developed into a continued fraction. [2. a2. Let the expansion of a be # Then. although not purely periodic. Conversely. Use tl~e graphical method explained at the end of the last section in order to show that d5.6 Lagrange's Theorem T H E ~ R 4. can be determined by recording which segment of F(x) is hit by the part of the path issuing from a. 3. + di is not reduced and that ita eontinued fraction expansioii is not purely periodic.l from then on the fraction will be periodic by Theorem 4. s The arrows in the figure indicate the paths that lead from each a to the next.

but they will be alternately slightly less than 1 and slightly greater than 1. . < 1. tend to the limit a.1 and O.29) will of necessity líe between .PERIODIC CONTINUED FRACTIONS 111 where a and a.30) slightly less than 1. We know also that the convergents c. solving for %+l. the values of the fraction (4. are convergents to a. = pn/q. once we have found a value of n which makes the fraction (4. by Theorem 4. and hence eventually. and consequently a' . But from our study of convergents in Chapter 3 we know that as n increases indefinitely. we get f a.2 the continued fraction for a will be periodic from there on. Factoring the numerator and the denominator.. as n increases. We notice also that in (4.30) will not only get closer and closer to 1.29) the numbers and are both positive integers and (see page 67) that O < qn-1 < q. both en-1 and c. are alternately less than a and greater than a. Thus.+l are quadratic irrationals and conjugates of both sides o this equation.-Jq. so that qp. Thus Lagrange's "theorem has been proved.cR-1 a' cn - tends to a ' 01' -a -a = 1 as n approaches infinity. This proves that a is redueed . Taking or. the wlue o f given by (4.+l > 1. this gives where c = n l / l and c.

+. which is reduced. First ilotice that 4 Ñ is greater than 1. and hence its conjugate . < a l . Show that a = i(8 + dE) not reduced. Adding al to both sides of (4. the continued fraction for 4 Ñ has ail interesting form.CONTINUED FRACTIONS Problem Set 18 1. and its expansion cannot be purely periodic. if is we eventually come to an a. .N -\/Ñ al is redoced. so N < is no¿ reduced.31) we get + + + aild since this expansion is purely periodic it must have the forrn Consequently. 4.4 cannot lie % between . since al is the largest integer less than d í the number 4% al is greater than 1. does lie hetween -1 and 0. h. . and verify that the expansion is periodic from then on. so and its conjugate.7 The Continued Fraction for -\r~ If N > O is an int'eger which is not a perfect square.1 and O. For example. On the other hand. the expansion for 1 is 6 where the period starts after the first term and ends mrith the term 2al. but that.

Wc assume that N is not a . that continued fraction expansions are uniqiie. al.32). ar.dr al is the conjugate of a = -\TN al. a2. al.PERIODIC CONTINUED FRACTIONS 113 Notice that. - .35). x2 . for addit'ional examples. subtracting al frorn both sides of this equation yields and the reciprocal of this expression is We know.33) . 4. however. recall from Section 4. hence. Hence.. To investigate the symmetrical part.2 that if a' = . un. page 116. See Tahle 2. and where x and y are unknown integerri whou* vriliic*~wcB are seeking.8 Peii's Equation. we obtain + + On the other haiid. except for the term 2a1.34) and (4. 2~11. an. but with the period reversed. then the expansion of -l/a' is the same as that of a. we conclude that It follows that the continued fraction for f orm di necessarily has the d~ = [al. the periodic part is symnzetrvical. ñ e can obtaiii the expansion for (fial)-L quite easily from (4. The symmetrical part may or may not have a central term.Ny2= -t 1 At the beginning of this chapter we mentioned that the cattle problem of Archimedes reduced t o the solution of the equation In this section we shall discuss the solutions in integers x and y of the equation wheru! N > O in s given integer. reversing the period in (4. comparing (4.

3). Hence the last equation requires that and alqn + + + qn-I 'Pn* . 1)2 . c.l t cn = pn/qn which come immediately before the term 2a1 in (4. otherwise the equation is of little interest. p. multiplying both sides by the denominator. d are integers and ..114 CONTLNUED FRACTIONS perject square.l / ~ n . since the difference of two perfect squaresis never equal to 1 except in the special cases (+.-. Replacing a. (Why?) The continued fraction expansion for -\TN supplies al1 the equipment we need to solve Pell's equation x2 . b.\/Ñ is irrational. we get which is equivalent to Now this is an equation of the form a bN < =c dd Ñ . qn-l. We know that - -al+--+ - - + + an+í an - ? where We again use the fact that where p.37).Ny2 = 1. and this implies that a = c and b = d (see Section 4.+l in (4. where a. or x2 .39) by the right side of (4.38) yields then. provided solutions exist..02. and q. are calculated from the two convergents cn-l = p n .N1~2= 1.

4 we know that and. Notice that so that the term a.40).Ny2 = 1. If n is odd and we still desire a solution the equation x2 . But from Theorem 1.41) becomes and hence a particular solution of Pell's equation x2 .-1 and 9.. then 2 2 fin .1)2n= 1.-1 and qn-I from (4.PERIODIC CONTINUED FRACTIONS 115 Solving these equations for p. is actually the term azn. where it occurs for the second that time. when it occurs again. we move ahead t o the second period in the expansion of TN. with the values of p. . is.1 . Y1 = Q2nl gives us again a particiitar uolution of the equation x2 . . this equation has the form If n is euen.Ny2 = . out to the term a. Y1 = qn gives a particular solutioii of the equation x2 . equation (4.Ny2 = 1. then and X1 = Pn.-1 we find that in terms of pn and qn.Nqzn = (.Ny2 = 1 is If n is odd.and so 5 1 = P2n.

81 [4. 11 1. 1."u. 1. 1.1 [2. 2. 3. 81 14. 2. m 4 17 170 9 55 197 24 5 5 26 127 70 11 1520 17 23 35 6 6 37 25 19 1 4 39 2 12 42 5 1 1 5 24 13 2 273 3 4 6 1 1 6 4 3 -1 +l f l +l +l +1 +l +1 -1 +l +l -1 +l +l +l +l 15. -1 +l 3 5 6 7 8 10 11 12 13 14 15 17 18 19 20 21 22 23 24 26 27 28 29 30 31 32 33 34 35 37 38 39 40 [1121 [l. +1 +1 -1 +l +l +l [6. .116 CONTINUED FRACTIONS 2 N Continued Fraction ~ O F d f xi 1 2 u1 1 1 x : .41 [2. 3 3 1 [4. 8 1 14. 1. 2. 3. 1. 1. 0 [S.i.l [3. 27 4. 2.2. S.] [4.B1 15. 1. 2 5 8 3 3 10 7 18 15 4 1 2 3 1 1 3 2 5 4 1 -1 +l +l -ti -1 +1 +1 -1 +1 +l m 6 1 [4.1. 11 61 [3. 1. 1. 1. 3. 3.101 2 [S. 101 [5. 61 [3. 4. 1. 1. 81 I4. 1 [2. 1 1. 1. 2. f i ~ [S. - 10 11 . m31 16. 41 I2. 1. 11 3. 1. 1 [31 61 [313.l [3.]. 1. 2 8 1 1 14. 1. 101 1 ~ 5 ~ 1 . 101 5 1 -1 15. 1 1 1. 101 I5. (6J3-1 m 1 . 5. 1.c 0 2 1 1 2 1. WI 15. 11 1.

Here we shall confine our examples to eqriations that have solutions.1 and not + l . give x2 . yi = q6 = 13.3 y 2 = .N y 2 = .21 y2 = 1 . x2 . 81 = [al. A calculation hence XI = 55. 2. The next period gives the convergen& and so.1 can be solved. Here N = 21. Hence. expansion of The 4% is so that n = 5. EXAMPLI.1 has no integral solutions. it can be proved (see Appendix I a t the end of this book) thilt the equation x 2 . For example. and the continued fraction expansion given in Table 2 is fi= [4.1. The first five convergents are But xl = p6 = 70. if we take XI Z = 9801. 1.96059600 = 1. an even number.13* = . so that and = as. = as.29y2 = 1 . h'ot al1 equations of the form x 2 . : . yl = 12 is a particular solution of the given equation. an odd number. 6. SOLUTION. a*. cvhich shows that a.: Fiiid a particular solution of the equation 1. we get X: . as. 1 . 1. SOLTJTION. we must move on to the next period.2 9 ~ = 96059601 . 2.29y2 the value 702 . as.PERIODIC CONTINUED FRACTIONS 117 The above analysis shows that we can always find particular solutions of the equation and sometimes particular solutions of the equation x 2 . 2a11. so that n shows that c6 = 35. y1 = 1820. 1 .N y2 = .29 . EXAMPLK Find a particular solution of the equation x2 .

Ny:' = 1 or x: : = -1. y. in turn.9 How to Obtain Other Solutions of Pell's Equation We have seen that Pell's equation x 2 . which indicates that \ve have to move to the next period to obtain a solution of the equation x2 .1. - THEOREM 4. we can systematically generate al1 the other positive solutions... y1 = 5 is a solution of the equation x2 . Show that xl = 8. In this Table.29y2 = + l . but that not al1 equations of the form x 2 . for the Table shows that - and gives a solution $1 = 70.1.118 CONTINUED FRACTIONS The solutions arrived at in Example 1 can be checked against Table 2. that is.1.1.8]..13y2 = . We shall state the main theorems involved and illustrate them by examples. Problem Set 19 l. then al1 the other positive solutions (x. ?ji > O such that x .Ng2 = . These statements will not be proved.2. Once the least positive solution has been obtained. However.13g2 = 1. Likewise we can check Example 2. y1 = 12 of the equation x2 21y2 = 1. = 3 is a solution of the equation z2 . as indicated in Table 2 . = 13 of the equation x2 .4. 4.1 have solutions. and further to the right we find listed a solution xl = 55.8 will always produce the least positive (minimal) solution. The values of x. N a positive integer not a perfect square. . and proceed to the next period to find a solution of the equation x2 . 3. 2. If ( 2 1 . if either of these equations has solutions.N y 2 = 1. yl) i s the least posz'tive solution of x2 . it will always produce the two smallest integers X I > 0.42) by expanding the term (xl y 1 d Ñ ) n by the binomial theorem and equating the + . Show that xl = 18. are obtained from (4.7 y 2 = 1. y. 2.N y 2 = I . n = 1. then the method outlined in Section 4. opposite N = 21 we find the expansion o f dE=~=[4.) can be obtuined jrom the equation by setting. and y . can always be solved.1.29y2 = -1 .

N y 2 = 1. and use (4.42). y1 = 12 is a solution (minimal) of the equation x2 . are solutions of the equation x2 .PERIODIC CONTINUED FRACTIONS 119 rational parts and purely irrational parts of the resulting equation. Using these values.8 we found that X I = 55 and 1. y.43) y. if (xl.N y 2 = 1. A second solution ( 2 2 . N < = ( X I .y2) can be found by taking n = 2 in (4.42). x. this gives .43): - r'. Since the conjugate of a ~ r o d i i c t t. and y. are calculated b y equation (4.Ny.42).y*) is a solution of x2 . 1/2) can be obtained by setting n = 2 in (4.he product of the conjugates.y1 fl)'. For example. Now we factor xn . EXAMPLE In Example 1 of Section 4. yl) is the least positive solution of x2 . from (4.N y . where there are n factors in tbe expression on the right-hand side..21y2 = 1. a since by assumption ( x l . We have.N?/: = (xn + yn1/Ñ)(xn - l/a fi) Thus x.42). . if x. and y2 direct calculation shows that + = 2xlyl.42) and (4.N y 2 = 1. This gives so that $ 2 = x. then x: . then the solution (x2. It is easy to show that. = 1. is this @ves (4.

A second solution is obtained from (4. y l ) be the least positive solution.) of x 2 .. .5.120 CONTINUED FRACTIONS which implies that xz = üO49. = 1 is the minimal solution .\/Ñ . -1. 3 (26)2. THEOREM Assuming that x 2 .1~1.\/Ñln. x 2 + y * v ' i = (2 = + 1 da2= 7 + 4 di.N y 2 = . A third solution (x3.1. y1 = 1 is a solution of the equation x2 ..3y2 = 1. 7 . = so that x3 = 7 . this is a solution since (117)2. We have EXAMPLE Table 2 shows that 3. (XI.44) by setting 21 = +1 = 117 + 37 dio. since In general.45) r. y. of x2 .1 can be calculated from the equation bu setting n = 1.21y2 = 1. y3 = 4. fiforeover. = (XI + . = 37.10(37)2= 13689 If we take n = 2 in (4.10y2 = n = 3. so that x3 = 117. These values satisfy the equation xa . the solutions of Pell's equation become large very fast. using the sume values x1. and 7 2 . . 4. A second solution (x2. y3 = 15. + y.ritive solutions of x2 . This is true since .45).13690 = .N112 = 1 are given by (4. x3 + y3 6 = ( 3 0 3. al2 po. y2 = 1320. 2. with n = 2. .1 i s soluable.3 42 given by the equation x3 1. Then all positive solutions (x. 2 .y3) is + y3 6 = ( 2 + 1 d ) = 26 + 15 4 3 .675 = 1.N y 2 = . The procedure can be continued.y2) is given by the equation XI EXAMPLE Tabfe 2 shows that 2. 3. let 4. 6. ..3(15)2 = 676 so that x3 = 26. we get . 5.

x = & l . C .10y2 = 1. equations of the form Ax2 + Bxy + C y 2 + D X+ E y + F = 0. see Figure 1 1 . Table 2 shomrs that xi = 18. Then. In concluding this section. Now the problem is reduced to finding integral solutions of the equation Solve t h h equation and list the first four solutions of x2 * such that y . Use Theorem 4.n2) = (m2 n2)2. y -x = m 2 .1 . then the values + y2 = z2. This involves an extensive study. 2. y2 = 6.N y 2 = 1 is preliminary to the study of the most general equation of second degree in two unknowns. and where x and y are the unknown integers.2n2 = -+l. find two solutions of the equation x2 . and l g a . where A .n 2 .13y2 = 1. B . = 17. + + + y2 = z2 . By means of certain substitutions for the variables x and y.5 to find the next solution. y1 = 4 is the minimal solution of the equation x2 . ure remark that the study of the equation x2 . E and F are integers. + + We now propose the problem of finding right triangles with legs of lengths x and y. - 3.PERIODIC CONTINUED FRACTIONS 121 This gives $ 2 = 19. 62 = 1 so t h a t these values are solutions of x 2 . $he solutions of this equation (if they exist) can be made to depend upon the corresponding solutions of an equation of the form x2 . Consider the Pythagorean equation x2 integers. Problem Set 20 1. Table 2 indicates that x. Use Theorem4.4 to find the next two solutions. Also. D . yl 5 is the minimal solution of the equation x2 . Let m . so that x and y are consecutive integers.Ny2 = dd.2mn = (nz . so that m = u n = u v.10 .18y2 = 1. + if ?n and n are will always give integral solutions of x2 y2 = z 2 because of the identity (2mn) (m2 .n = u.n ) 2 .13y2 = . and so we must be content with this introduction. n = u.

y. the angle e between x a n d z approaches 600. Find sets of integers (x. as these integers increase. Figure 11 .122 CONTINUED FRACTIONS 4. 2) for the sides of the right triangle of Figure 11 such that.

however.2 Statement of the Problem Throughout this chapter let a! be a given irrational number. and would lead to the general solution. Proofs of the theorems stated here and of many related theorems can be found in the books by Niven [8]. in integers. and let p/q be a rational fraction. however small.Ny2 = M could be undertaken. if E s is any given number. with positive q. in other words. q such that .1 Introduction In this chapter we shall preview some results t h a t can be studied once the first four chapters of this book have been mastered. on theorems related to the approximation of an irrational number by a rational fraction. 6. where p and q have no factors in common. We have already indicated that a complete study of Pell's equation x 2 .CHAPTER FIVE Epilogue 6. of the equation Ax2 + Bxy + Cy2 + Dx + Ey + F = 0. It is clear that we can always find a rational fraction p/q. as close a we please to a. and Hardy and Wright [5]. we can always find relatively prime integers p. We shall concentrate now.

with q 2 1 . of the form p/q.3 Hurwitz's Theorem * Inequality (5. . 5.3) has this interesting feature: I f a is anu irrational number. to the conAny of the convergents p l / q l .1. how small can we make e ? We have already accomplished something along these lines.9..124 CONTINUED FRACTIONS But this is not the interesting point. is there a number k > 2 such that the inequality has infinitely many solutions p/q ? If so. then how large can k be? . p2/4?./q. the inequality (5.2). Theorem 3. It is possible to sharpen the inequality (5. such that then it can be proved that p/q is necessarily one of ¿he convergents of the simple continued fraction expansion of a. p.+l/q. . tinued fraction expansion of a can serve as the fraction p/q in (5. given a and q. q > O. stated here without proof.2) as shown by the following theorem. THEOREM Of any tlwo consecutive convergents p.how large must q be? Or. p. there exists an infinite number of rational fractions in lowest terms. We proved in Chapter 3. such that . that if a is irrational.3) immediately suggests the following question concerning still better approximations. Given an irrational niimber a ./q. at least one (call it p/q) satisfies the inequalit y Moreover. and if p/q is a rational fraction i n lowest terms.+l to the continued fraction expansion of a.l).. and 5. What we should like to know is : Given a and e in (5. .

2. -1.-1 = pn and I t can be shown that. ] = [O. a. . if the continued fraction expansion of a is [al. a.. the "simplest" numbers are the worst in the following sense: The "simplest" irrational num ber is f. . where each ai has the smallest possible value. ./q. approxima¿ions p/q which satisf y the inequality The number d is ¿he best possible number. then the rational approximations p. /5 These remarks suggest the truth of the following theorem.EPILOGUE 125 I t can be shown that. On the other hand if there are small numbers in the sequence al. . no matter how far out we go. a. and if p. Iarger number were substituted f o t 4. a2. . 2 m - . 5 . then m hence we can get very good approximations to U! if the numbers al./qn cannot be too good for small a. From the point of view of stpproximation. the expression gets closer and closer to 1 d qi..T]. The convergents to are the fractions SO that q.. the theorem would become i false if an. get large very fast. a2.1 = [O) 1. . 1... is the nth convergent. for n very large. . THEOREM Any irrational nulnber U! has un in$nity of rational 5. ... . = 2/5 . first proved by Hurwitz in 1891. . .as) .

not an infinite number. the irrational nunzber O < fl < 1 lies between two consecutive fractions p / q . + . Substituting for /3 in the above inequality. we let 8 = a ..6).n where n is the greatest integer less than a. the one important for this study is: If for any n.2 depends upon the fact that. the sequence F . Hunvitz did not use continued fractions. r / s can be used for x / l j in the inequalit y + + In order to make such an inequality valid also for an irrational number a > 1.\/5 is the best possible number in this sense. is the set of rational numbers a/b with O 5 a 5 b 5 n. The first four sequences are: 4 5 These sequences have many useful properties. b) = 1. in the continued fraction expansion of a.2. (a. instead he based his proof on properties of certain fractions known as the Farey seqzdences. r / s of the sequence F. This is the central idea in Hurwitz'e proof. One proof (by means of continued fractions) of Theorem 5. then there exists only afinite number of such rational approximations p/q to a.126 CONTINUED FRACTLONS By "false" we mean here that if were replaced by any number k > fi. For ' complete details see LeVeque [7]. For any positive integer n. Niven [83 gives an elementary proof that . ( p r ) / ( q S ) . arranged in increasing order of magnitude. I n his original proof of Theorem 5. then at least one of the three ratios p / q . at least one of every three consecutive convergents beyond the first satisfies the inequality (5. we obtaiii where x = ny x .

that a a. if a real number a! has the contjnued fraction expansion a = ! [al. (i) refle. then y x. Now.2. d satisfying the and siich that . Such a statement always seems to stimiilate further research. An equivalence relation divides the set of a11 numbers into equivalente classes in such a way that each number belongs tJoone and only one equivalence class. i.e. I t is equivalence just defined has al1 the properties lente relation. I a certain class of irrationals f were ruled out.q. could this constant perhaps be replaced by a larger number? Indeed this can be done. (5. 6 2 1. then x N Z. it follows from p.6). and ud . N N - x). x condition N y) if there are integers a. ffn+~]. b = 3.p. Hence if a! and /3 are any two real numbere with coiitiriiied fraction expansions and from - .7) and (5. c. m . i.\/2: and x = (2 d because x = (a b)/(c fi d) with d = 1. = ( .+~ [cf. b. (iii) transitive.EPILOGUE 127 Mathematicians-are never content mith a "best possible result". if y = .bc = 2 . and are therefore just as hard to approximate.1) which forced us to accept .-~q.\/5 as the "best possible" 6 constant iTi the inequality (5. if x y and y Z.e.1)" (see Theorem 1 .e. The class of irratjonals to be excliided consists of al1 numbers eqztivalent to the critica1 number l = 4 ( . a n .. We shal show that al1 numbers equivalent t o l have the same periodic part a t the end of their continued fraction expansions as 6 has.3 = -1.2: can be expressed in terms of y by the fraction s Forexample. if x y..-1 . x a = 2 .8)]. DEFINITION a number x is said to be equivalent to a number : Here y (in symbols.. i. namely that it be + + + 3)/(* + l). 4 . such as the constant & in Theorem 5. c = easy tso see that the reqiiired of an equivaN - y 1. every x is equivalent to itself (x (ii) symmetric.cive.

There are infinitely many irrational numbers equivalent to f = *(& . C O . then a an+l 0. - 11.. and since 1 1. C I . . -1. b. +(a . SO a N P. and hence al1 these equivalent irrat'ionals play essentially the same role in Hurwitz's theorem as the number E = + ( 4 5 . x y.1/51 has an infinity of rational approximations p/q which satisfu the inequality There is a chain of theorems similar to this one. that is. each of these is expanded into a simple continued fraction.ain place on. by Theorem 5. 13 . Then..3. Any irrational number /3 not equivalent ¿o [ = +(1 . and only 2f 0 . each of these expansions will contain the same sequence of quotients. from a cert.+i b. I n particular any two rational numbers x and y are equivalent.1 ) does. . It seems reasonable t o guess that if we rule out the number f and al1 irrationals equivalent to it.bn. let us suppose that. THEOREM 5. b. c2. i f and only i f the sequence of quotients in a afier the mth i s the sume as ¿he sequence ~ T LP after the nth. The question as to when one irrational number is equivalent t o another is answered by the following theorem. stated here without proof . co. .4. .a2. then the constant 4 5 i1i Hurwitz's theorem could be replaced by a larger number.9) can be replaced b y any number less than or equal lo 4 2 2 1 / 5 .ci.in (5. [al. . .128 CONTINUED FRACTIONS and if an+l = Pm+l. In fact the following theorem caii be proved. then the number . THEOREM Two irrational nurnbers a and 0 are equivalent if 5. = [bl.. For example. Kow let us return t o Hurwitz's theorem. .bz. . a. .. for their expansions can always be written in the form 2 = Y N .1).3.1) or 4 2 . ca. = [bi. if 0 is no¿ equivalent to either . .

No. and a very simple proof using Farey sequences was recently given by Ivan Niven. 9. vol. notice that the lower bound is not just the negative of the upper bound. this is Hurwitz's Theorem. but the study of continued fractions is. and probaibly w i l remain. THEOREM For any real nulnber r 2 0. of course. 121-123. and the expression is unsymmetrical. 6. One coiild. hl. the inequality has infinitely many ~olut~ions.? 5. 2. Using continued fractions. Robinson (1947) gave a proof of Segre's theorem. go into the subject of continued fractions more deeply by reading such books as Perron [ll]. For example. In recent years severa1 new methods for solving problems in this field have been invented. this monograph can serve as the point of departure for further study of a variety of topics. the basic stepping stone for those wishing to explore this subject. Segre in 1946.side of Hurwitz's ineiuality can be strength eiied without essentially weakening the other. The Michigan Math. the following theorem was proved by B.5. 11i02. . and also proved that given e > 0. R. The subject has a long history. The field of Diophantine approximations by no means exhausts the avenues of exploration open to the interested student.4 Conclusion Hurwitz's theorem is an example of a whole class of related theorems and problems studied under the general title of Diophantine approximations. pp.EPILOGUE 129 Recently interest has been showii in "lop-sided" or unsymmetrical approximations to irrational numbers. t On Aayrnmelrtr Diophantiae Approximations. result is interesting since it shows This that one . Journal. a n irrational number a! can be approzimated by infinitelg m a n y rational fractions p/q in szich a way that When r = 1. For r # 1. yet there are still many challenging prohlems Ieft t o be solved.

If x = f ( l ~ ) show that y = ( . and (iii) transitive.130 CONTINUED FRACTIONS Alternatively. Problem Set 21 l. r/cr of the Farey sequence Fp on page 126 and verify that a t least one of the numbers p/q. a subject initiated by Stieltjes and others. . a t least one satisfies Hurwitz's inequality (5. 24. see Hardy and Wright [53. . + 2. + + 4. there is t h e extension to analytic continued fractions (see Wall [14]). Calculate the first six convergents to a = f (1 4%) show that and of every three of these consecutive convergents beyond the first.6). Expand both x and y into simple continued fractions and use these to give a numerical check on Theorem 5.6). Chapters 3.10s 7)/(72. founded by Minkowski. (ii) symmetric. (p r)/(q S). Locate a = +(di6 2) betseen two successive elements p/q. 3. Fg. Calculate the next row. a n d there is the beautiful and closely related subject of the geometry of numbers. Prove that the equivalence relation defined on page 127 is (i) reflexive. r/s satisfies the inequality (5.5) is equiva.of the Farey sequences given on page 126. + + 5.3. lent to s. For an introduction t o t h e geometry of numbers.

if it did. if x2 . y = 2y are both even integers. Then - + and since yl and y1 1 are consecutive integers one of them must be even.3y2 = -1 is to have integral solutions. then we cannot have - - But an integer of the form 41 1 41 = -2. we notice first that x and y cannot be both even or both odd. hence 4y1(y1 1) is divisible by 8. y odd. in the second case.1 is not solvable in integers x. cannot have the value . or x odd.1. Suppose that x is even and y is odd.6n 1 is an integer. y 2y1 1. and therefore 1 = i t to the reader to show that if x odd and y evan. y. Hence. So yl(y1 1) is divisible by 2. then we must have x even. and. from (1)) we conclude that y2 has the form 8n . then + is also even (twice an integer) and again could not equal -1. We leave xa Bya -1.3y2 = . would not be an integer.1 Has No Integral Solutions To show that the equation $2 . Similarly. y even. For. where n is an integer.1. Then + + + where 2 = xi . if x = 2xl 1. if x = 2xi.3y2 = .APPENDIX 1 Proof That x2 . so that x = 2x1. in the first case. then is even and so could not be equal to .1. y = 2y1+ 1 are both odd integers. + -+ .

at. . and others. 2al]. This last equation is closely connected with a famous theorem stated by Fermat in 1640 and proved by Euler in 1754: + THEOREM: Every prime p of the form 4c 1 can be expressed as the sum 7 of two squares. in Section 4. then the equation x2 .1 is soluble. if N = p is a prime number of the form 4 c 1. Legendre's method depends upon the fact that the periodic part of the continued fraction for has a symmetrical part a2. . if x2 . a*. ar. al. whenever N is such that N . that if the symmetrical part has no central term (n odd).a2 followed by 2a1.+I by reversing its period.1 always has solutions. .py2 = .1 has no solutions. is equal to &.8.132 CONTINUED FRACTIONS Hence there do not exist integral solutions z. Q such that p = P2 Q2. as.. We proved.y of the equation In fact.. . 1 6 This we write in the equivalent form where.. the equation x2 . + + Once this theorem became known it was natural for mathematicians to search for ways to calculate the numbers P and Q in terma of the given prime p..+l. there exists one and only one pair oj integers P.py2 = .3 is an integral multiple of 4. as. and this representation is unique. The converse is also true.1) Moreover. Gauss (1825). then the equation 7 x2 . the period in the expansion of a... a3. a3. Now CY. hence the continued fraction for dc has the form = [al. Without entering the details of the proof. obtained from CY. beginning at the middle of the symmetrical part. . however. Serret (1848). a. On the other hand. . namely. a. we shall present the essential idea of Legendre's construction.py2 = -1 is soluble then there is no central term in the symmetrical part of the period. Constructions were given by Legendre (1808). That is.Ny2 = .~ is a purely periodic continued fraction and hence has the form (see Theorem 4.+~ is symmetrical and hence the number 0.

2.1.3 + 1. Problem Set 22 1 . ~ --1 P = that d or Q Q As an illustration.APPENDIX 1 133 But according to Theorem 4. take p = 13 so that Hence all we have to do is calculate a . i f one soldier is added or taken away from one of the squarca. c Show also that. Q = 3. the conjugate by of am+l is related to /3 so that This means that ~ + 1 / i . . . giving p = 13 = 22 + 3a. Expanding fiwe obtain Thus so that P = 2. 3 p = P2 + Qa. thc two cl(btrrclirncntscan sometimos be coinbined into a square. 3. Show that it is impossible to combine the two squnres into a single square of soIdiers. Ex~iressp = 29 as the sum of two squnres. There are two equal detachments of soldiers arranged in two squares. = 4 . each containing b rows of b soldiers. Espress p = 433 as the sum of two squares.

Bombelli.A P P E N D I X 11 Some Miscellaneous Expansions The following is a small collection of miscellaneous continued fractions. 1.? The list is not restricted to simple continued fractions. Cataldi. He expmssed the continued fraction expansi*n of in the form fi and also in the form t See D. E. mainly of historical interest. In modern notation he knew essentially that 2. . 1572.Smith [13]. 1613.

He found the following expansions involving the number the base of the natural logarithms. about 1658. . . EuIer.A P P E N D I X 11 3. 1737. m 4.14159 . both discoveries were important steps in the history of a = 3. Lord Brouncker. This expansion is closely connected historically with the infinite product given by Wallis in 1655.

the 7th convergent to (e . For example. Lambert used these expansions t o conclude that a) If x is a rational number. This number differs from the value of e by one unit in the 12th decimal place. since tan (n/4) = 1. 5.CONTINUED FRACTIONS This Iast expansion affords a quick approsimation t o e . neither x/4 nor x can be rational.then ez cannot be rationaI. b) If z is a rational number. not 0. not 0. 1770. - x '. - 6.1)/2 is 342762/398959. tan x = 1 - 1 3 - 1 x x - 1 5 - x - 1 7 -. Lambert. 1766. approximately. Some weaknesses in Lambert's proof were remedied by Legendre in hia Éléments de gdometrie (1794). then tan x cannot be rational. so that. . Lambert. Thus.

. both numerators and denominators The convergents are f. 5. being formed from the sequence of Fibonacci numbers 1. 10. . 13. 1. 8. 1833. 2. +. 3. The convergents t o ?r are m the fraction approximates n with an error of at most 3 units in the 7th decimal place. . 9. a .A P P E N D I X 11 137 Unlike the espansion of e.1415926536 does not seem to have any regularity. the simple continued fraction expansion of R = 3. Stern. g. S.

1776. 1776. Lambert.CONTINUED FRACTIONS sin x = 1 x + x2 ( 2 .x2)+ 4 5x2 (6 7 . 1812. Lambert. Gauss. tanh x = 14.x2) + (4 5 m 2 3x2 .3 . tan x = x x2 13. Iog (1 Lagrange. 1770. .x2) +- 12. 1770. arctan Lambert.Lagrange.

1776. 18. . Lagrange.16.= 1-2 - 1 . x2 9 17. Lagrange. 1813. This is the probability integra1 used in the theory of probability and in statistics. 1805 . Laplace. log . Legendre. 1826.

and > q > O. .1415926536 .Solutions Set 1. anI- . . is an integer > O. +f. is 1 +an 1 where a . 3. . al. . then Q = n = 3.1415929204 . . 4. 6. If p Q P > 1 and - [a. fg = 3. a*] = al + + 1 as +-. The reciprocal of P Q = [O... . al. page 13 2. . 4. an.

1. p d p . 11 l . 15) (c) 10. 11. 1. 4. 3. 2.SOLUTIONB T O PROBLEMS Conversely. 8 . (d) -1. Even nurnber of quotients: (a) p d q s = #*.p4q5 = 51 0 . y. 21. page 19 Set 3. - . 1. (a) [5. e. 3. 1 0 + 3 * 3 + 2 .4. 0 1 7 2 convergents +) Y) 1 1 2 3 13s) s8~ 2 8.460 (b) -1. 4. compare with original fraction. 1. 1. T 9 (b) [3. 81. 1 e. 4 3 + 4 . hence psq4 . 5 ) 1. 51 = [5. 1. 1. 3 .23 20 = 459 . 1. then o P is of the form and its reciprocal is Set 2. 4. 1. convergents f . page 25 121 1. p 6 / ~ 5 = *+1 p4/q4 * 9. 1.psq6 = 51 11 . 1. 1. 2. 111. 1393 = 5. = = Odd number of quotients: (a) [2. 1.28 20 = 561 . -gal -f%. 51. z+.2]. -2-3 -z+-. 1 + 2 . 1./qc Bf = 15. (d) [5. g. &A0. (c) -1. 1.$. .f-. p5/q5 = hence psq5 .560 (b) 1. convergents . = J# = [5. (c) 1 (4 1. * 5. if p < p. 1.225 + 5 . 1 Similarly. 1. convergents $.

8.1.2 In this relation let n have in turn the value . n. 3 . . From pn = c ~ p . 1..-3 Similarly The required result is then obtained from these equations by successive substitutions. n .2. we used the fact that p.~ ~ ~ .i is proved in a like rnanner. The resuft for qn/q.142 CONTINUED FRACTIONB 7. 2. This gives the following equatibns: + pn-r = (n . = n p . I n constructing our table of convergents.. .1)pn-a + Pn-a .+Pn-2 + 1 pn-2 P. .i .~ pn-2 + we see that and from the fact that pnel = an-~pn12 pn-3 we see that Pn. n .=a.

pzq1 . Set 5. +. pn-2. there are no integral solutions with both x and y positive. poq-1-p-iqo=1-1 .SOLUTIONS T O P R O B L E M S Adding the Ieft and right sides of these equations. .p2. . u must be an integer such that u < & and u > 97. we obtain the required expression. Hence Required solution is Forboth x and y to be positive integers.p o q i = 3 ~ 0 . (a) Showthat x = . no such integer exists. namely m m Set 4. no integer u can be simultaneously lem than and greatcr than . . f 2 .3 1 = (-1)2) etc.p1q2 = 4 . @)z--2+7u. k l . y 15 = 1 .7t .21 '1 Thus.u. on the left and subtracting the terms pn-1. . page 35 1. p~ from both sides of the equation. where u = t / 2 or t = 2u. g=Q-31u) u = O .1 . hence. Note that some solutions rnight have a different parametric forrn but still reproduce the same values of x and y. for.o m o = l = (-1)Q) p i q o . . Clearly. l'here are no poaitive integral solutions . page 28 1.3 y + t when t = -S 2 . we obtain Leaving p. The second part of the problem is accomplished by simple calcuIations.1 ~ 1=(-l)'.

15u. xo = q3 = 4. The general solution is x = 22t 4.37t. y) = (2. = 36. 41 1 . 13) and (x. >5 6. yo = p3 = 3. 5). Why? . x = xo tb = 4 = + + n + (c) (d (e) g$ = [1. + m ) ++ n = 4. 31 + 56t. 6. y) = (7. u = 13(3 . f2.1.2 and ~ Hence t = 0. 1. 64). 17t. 2.9t. & 2 .144 CONTINUED PRACTIONS (c) x = -6+47u. w = 0 . page 42 (In al1 cases t 1. Then 372 22y = 2370.65t.18. y = 3 . . hence y = x = 25 1 . 1. y = 101 . x = + 56t. (48. 27). For positive solutions t must be an integer between . y = yo ta = 3 13t.4. y = 17u . The straight line if carefully graphed should pass through the two points (x. 4. when u = -1. = = 4. y) = (4.1. Set 6. Positive solutions require that u and u > A. For example. + 65t. . But no integral value of y will make u an integer. -1. y = 31 56t. 6. Hence w = 1. No positive solutions. The solution of the equation 2s . hence 98 = [O. For positive solutions w must be an integer less than $4 and greater than &. 1. 21. '1. Hence N = 20x+ 2 = 60u . The given equation has no integral solutions. n = 29 + 65t. + 3. Let x = number of horses and let y = number of cows. 36 4. hence u = 1. . 5.2 1 w . Hence u = 9(5 13t). hence [O. (d) x = 3 4 . . 31. y = 2u . 6). y = 2 .3y . 1. xo = qt. By Euler's method you would arrive a t other equations which cannot be solved in integers. and the positive solutions are (x. yo = p3 = 29. + + + yo = p4 y = 36 = 36. The solution of the equation 9x 13y = u u = 84 is x = 5 13t. + 5. .5. u = 0 .u where u = i ( 1 3y). 41. (a) = 0. 2. and the only positive solution is (x. + + w. xo = qs = 25. where u is any integer. xo = q4 = 31. + 1. 11. x= n = 6. 4. y=13w-7. 3. f 3. y) = (13.3y = 1 is x = 3u . + 2. 101)) (26. where t is sny integer. x = 15u . yo = p. hcnce + . 2. For example x = 3 .

270. y = l i t . For positive values of a and b it is sufficient to require that x _> O. + + 2. and since 3 divides 9 the equation is solvable. The gcni*rul noliitioii iu x = 1200 . x = cxo bt = 175 56t.c.65t. (a) x o = 4.204. The given equation has no integral solutions. When t = 39. (f) The g.d. from ~ which the solution of the given equation according t o (2. y = 13t . The only solution with both x and y positive is x = 3. hence c = 3. of 183(= 3 . (a) The g.-i = 39. y = 10 and a 5. .360/9 and > 270/7.61) and 174(= 2 3 29) is 3. * ! 1 Hence the solutioil of the given equation. x = O. We first solve the equation 612: .45. 3. x a 10.9t. '1. of 56(= 23 . y 2 O.d.1 . is a particular solution of 562 . solution.58y = 1 for which the espansion $ = [l. b = 32. (d) Since 34(= 2 17) and 49(= 72) are relatively prime. y = 7t . 11 and 63 = 32 7 so the g.49t.23) is (c) An unsolvable equation since 77 = 7 . y = 3 and a = 68. The value t 70 lairtln to tlic* orily ponitivc. page 44 (In al1 cases t = 0. we have only to solve the given equation by the rnethods of Section 2. c = 5. 19. according to eqiration (2. - + + + + + + + + 7= + + + + Set 8. page 4 8 1. F'rom the hint we find that 7s 9y = 90. x = 9. When t = 40. Thus we can take t = 39 and t = 40.5 8 = 3. Divide both sides of the given equation by 3 and solve the resulting equation 612: 58y = 3.28).65y = . yo = 25. or that t be an integer < . - - .7t. yo = 29. m ) 17t. l.900. x = c x o + bt = 20 y = c y ~ at = 15 13t.SOLUTIONS S O PROBLERlS Set 7.17t. The required solution is x = 65 49t.4.7) and 20(= 22 5) is 4 and does not divide 1l. y = 10. x = cxo bt = 87 65t. 2. y = 45 34t. 1 S ~ O W S that xo 9.c. yo = 3. of 77 and 63 is 7 and does not divide 40(= Z3 5). yo = pn-i = 41. is + (b) In this case we must solve the equation 611: . b = 95. f2. y = cyo at = 75 56t. y = 34t . y c y o at = 203 -t.c.d. y = 5 given by t = 19. (e) x = 65 . Check: 13(20 17t) .17(15 1st) 5. The general soiution of this equation is x = 360 . (b) xo = 25. + - + 4. The solution of the equation l l x 7y 68 is x = 136 . c = 7. (c) xo = 29.

The first five convergents are: 2 S I .1 . (a) Let = 1 x=[2.4] = 2 + .2 and z = 2 + ( 4 6 . Then 1 1 1 1 1 (b) Let x = ~ + .2. -. Hence +4 2 6- 1 4 6 . Hence.C O N T I N U E D FRACTIONS Set 9. Faj consists of the number of branches O that are a t least one year old and the number of branches . note that in triangle AOD. In the nth year.218 5 22 49 6 7 13 46 59 (a) 9 20 89 ( c ) i ' -1y e ?2 .7 ' 9 2 3 5 8 3 7 (b) i ' i ) S ' 3 ' 5 2.1 ? l 2 y=2+ 1 1 q+y. 6 42 72 3. But we know already that AF/ A D = AG/l. by division. From the similar triangles AGF and AOD r e see that and. page 59 1. the total number of branches. 6. Expansions are given in the problems. To show that AH = + 8* see Figure 3. (AD)2 = 72 +" s2 .5 ) = 6. = 42 72 + 8* ' On the other hand from the similar triangles AHF and AGD we see that AF/ AD = AH/ AG. since AF = 4.2) = 4 0 .J 1 / = 1 + i + i + s + ~ ?/ Hence z = 5 + ( . .

With E as center and radius EB describe the arc BF. = 0. Construct the point E such that AE = E D and draw EB = 4 4 5 2 .+I = F.-I for n = 2. F.S O L U T I O N S TO P R O B L E M S Figure 12 Y .3. the next year. 8. Thus F. and FI = 1 (because only the trunk was present during the first year) yield the recursion formula for these F'ibonacci numbers. I n symbols. Figure 13a . First solutwn: Construct the square ABCD of aide z = A B . that are less than one p a r old. During branches.. + F. see Figure 13a. there are + Y. On = Fn-i. Since the number of at-least-one-year-old branches constitutes fhe total number of branches of the previous year.

With A as a center and radius AF draw the where T = arc FG. and that B H is parallel to F I . then I D = x . Hence BG = C D . x 9. and G D parallel to BE.= . Second solution: Construct the right triangle B A C such that A B = x . Then A C C D = T X .= -D GB DE TX -7. and if we let A D = x. With C as center and radius B C = d5 x construct point D. But C D = BG = A l . . we see that A D / A I = C D / H I . . Now B C = 1 = A I . hence A D / A I = A I / I D . and x ( x . so that x = = +(1 + 4) 5. A C = 2 x . Using similar triangles. Sirnilarly. and so BF = H I .1. Then + + + AG A .4B = s.1 = O.CONTINUED FRACTIONS Then 1). or ' X . see Figure 13b.x . Draw BE. where T = +(1 d5). and AG=T(GB). For the regular pentagon A B C D E whose sides have length 1. Construct point E such that D E = . or ( A D ) ( I D ) = ( A I ) 2 . and H I = BF = I D . Clearly AG = X / T and so : d+ (i Consequently. first prove that A D is parallel to B C and that BE is parallel to C D . prove that H I is parallel to B E .1 ) = 1.

and al1 are less 2 .. . . 8. 4 .. than 4 2 = 1.00134 d 9 in place of r ia at most 0. 1..9 2 ~ ) . 1. . The first five convergents to ?r are in turn. and with C and D as centers and radius AC = C D = T.00005.t.0005. + Set 11. $8. draw line C D = 1. + $E. . fi= 4. hence the required approximation is J+ In this problem it would have sufficed to use l/qE in place of l/q5qs since (&)2 < 0. 5 . Hence to construct a regular pentagon. The convergent c-i = "$A1gives 1/3262 < 0. hence the error in using - For example. 170 3. is the required approximation. 1 .7 7 106 2.i] me +. . 7 106 1 -= 0. . The even convergents are +.. 1 .00134 - . ~ = [ 2 . Points B and E can then be constructed since AB = BC = AE = DE 1. - Set 10.g. S]. construct point A . A calculation shows that C4 = f3 g6 r &. The convergents are 1421 +. page 76 1 . etc. -y. 1 1 -> . 2 and c i = f l C Z =1-1~ -c. 2. . 1 113 33102 . 5 . = [4. 2. The odd convergents to 4 =. = 8 6 6 . C~aIculate. Moreover < 5 < 8.358899 l/q. 4% = .i Z + f r ] c5 =: +$j+.SOLUTIONS T O PROBLEMS Figure 14 A line segment of length T can be constructed using the results of Problem 4.414 . hence c .9 1. +8. 3. 2 . 9+#. page 63 1. and are al1 greater than 4 5 . [l. . sqT+2-. 2.

1. Try . 1. ( 1 . (a) x = f ( 3 + fi) 3. 51. a = = 0.62. page 79 1. then x = [O. . 7 ) ) ( 1 1 . i / ) = (qn.30277 = . a = +(fi [O. 3.4(2) + O = O. ~ l othe points t Also carefully plot the line y = a where a x value of 3 4 5 . 2.1. 26). 21. The first few convergents are Q.c = 0. Let a. 6 = 1. page 80 1.. b = 2. and x also satisfies the equivalent equation xe . +. - ( x .2. +. 1. .ac = O since b = ac. .1 = [O.. + 2)pn + pn-a = O gives 8 . the apgroximate & = [ l .. $. 1.19258 vergents are : + . 191. The values of the first few con- 2. For example. (3. . - then pn+a . let a = 1.2].1). 1. (15. Set 13.abx . Write then ax2 . l ) .1) = .732 .CONTLNUED FRACTIONS Set 12.l. 1.bx . (4. The values of the iirst few convergents are: (b) z = &(5 d 2 9 ) = 5. Plot the points ( 1 . pn) Also plot the line y = 1/3 x. 21.(ab other cases. 2.


Set 14, page 90
i.(a)a=+(.\/IS+3)>1, /3=fi+3>1. (b) 3ac2 - 6a - 1 = 0. (c) a' = f ( 3 -0.154 , -1/p = -1/(d12 3) = 3 3 - 6 2 ) .



2. (a) 2a2 2a - 7 = O, a = p(l/i5 - 1) > 1, a' = ;(-1 - 1/15)< -2. (b) 3y2 - 57 1 =O, s> y = +(5 4 ) 1, > 0. y' = i ( 5 - 4 s ) = 0.232




Set 15, page 100
1. al f a 2 = (Al f A2) (BI f B 3 a 1 . a ~ (Ai 4- Bi 4 5 ) ( ~ 1 34- B2 = = AiA2 BiBzD (AiB2 AzBi)





a) D + 4

for if A:

- BeD = O ,

then D would be a perfect square.

(ai a2)'


= =


a 2

(AiAs BIBzD) - (AiBe BlAo) 4 5 ' I (Al - Bi *)(A2 - B2 d D ) = al az; (A1& - B ~ B ~ D ) ( A z l - & B ~ ) A: - B;D - B;D



on the other hand

= C2N - A 2 - B2M. 3. A B 6 f = -C I Ñ ; therefore 2AB If AB # 0, the left side of this equation is irrational, the right side f rational; this is impossible. I AB = O, then A = O or B O. If B< M C 4 Ñ = O we see that A O, B # O, then from A d % / d Ñ = -C/B, contrary to hypothesis. Hence if A = O, then B = 0, and hence C = O. If B = 0, -4 C 9= O, hence Amo, C-0.










Set 16, page 104
l. The largest integer l e s ~ than f (5


is 3. If
-1 =


5 + f i 3

= a + - ?1


- 4 + I E , 3


is On the other hand, a: = +(4 - fi) approximately -1 < a: < O. Hence a l is reduced.
2. From O



< P < v'D


I D - P < Q < dD + P ,

see (4.23),

it follows that

Sinee Q


and P

- d\/D O, <


de - P

< Q implies that ((.\/E- P ) / Q <
a = '

1 so that

a-fi Q



3. The totality of expressions of the form

P + G P and Q are where

integers satisfying condition (4.23) are obtained as follows: If 43 - 1 < Q < 4 1, i.e. 6 Q 5 7; thisyields




If P


2, 5



If P

= 3,


< 9;

this yields


By the same procedure, for P = 4, 5, 6, we find

í/43 wieh

k = 3, 4, . . . , 10;


+ 6 with 1 = 2, 3, 1

, 11;


+ 1/43

with m


1, 2, . .

, 12.

Set 17, page 110
1. a = 1

+ d > 1, i

a =1 '

- fi = 1 - 1.414
a =1 l

lies between -1


e ]


1 + d = 2 + -, i a

+ d = a, 2


a = [2,2, 2, . .


2. a = 4 > 1, a = 8 ' = (2, 1 4 . ,]

4 s

4 8

does not lie between

and O.

Set 18, page 112

a = a2, where s

a 2

is a reduced quadratic irrational. Hence

8 -

+ d 3 7 = [1, 1, 1, 3, 21 .

Notice that a and a are not reduced, but that l

hence as is reduced and the continued fraction is periodic from then on. Set 19, page 118
1. 4 = [2, 1, 1, 1, 41. The convergents are 2 7 SO pc 21 8, q 4 - yi 3, and



5/2, 8/3 = p Jq4

xi = 4 . fi= [3. = 1. 4 / 1 . g.18ya = 1 = 2. According to Theorem 4. 3. 1. According to Theorem 4.1.) ~f the equation x2 . mí 2 .18yi = (577)2 .5. and since xl = 17. Proceeding to the tenth convergent we find plo/qlo = 649/180.18(136)2= 1. y1 = 5 is the minimal solution which determines the next solution $3 = x f 39xls/:. V I = ni = 1.-1. and X: . - Set 20. ya).154 C O N T I N U E D FRACTIONS 2. Since A B fi = C E d D if and only if A = C and B = E. the next solution of x2 obtained from . 61. 1. 1. 2. vi = 1 is fhe minimal aolution of u . The ñrst five convergents are 3/1.13y2 -1 is xl = 18. 7 / 2 . yi = 4. F'rom fhe relation for x3.13y2 = 1 are obtained from + + $2 + y2 fi = ($1 + yl dGl2and + y4 fi= ( x l + y1 fi)i $4 The computations are Ieft fo the diligent reader. Thus x2 = 649.2v2 . y2 = 180 is a solution of x2 . 1. page 121 1.4.13y2 = 1. we have + + + + + y2 = 2x1~1 2 17 4 = 136. qs = y1 = 5 gives a solution of x2 13y2 = . ya) and ( 2 4 . This yields u. SO ps = $1 = 18. y3 = 32:yi 13y:. Solutions ($2. we have so that I f 17 is substituted for may be verified. . 1 1 3 18/5 = p ~ / q s . XI and 4 for yl. yz) and (xs. the next two solutions. Table 2 indicates that ul'= 1. the relation x: . are obtained from 2 2 y2 d 8 = (xl y1 d18)2 1 and + + The first relation yields x2 1/18 = x: 18y: 2xly1 6 8 . ($2.

n2/m2. then 8/2 would approach 3Q0 and 8 would approach 60". x4 = 696. (m2 . 4. ms = 12.but much cIoser to 60". approach 1 / d . we convert & info the continued fraction and let mi and ni be the numerator and denominator. y4 = 697. . . and u1 = 7. 5822. . z3 = 169. . . . Set 20. respectively. : 2 = Thus. ua+ = (1 vs = ns = 5. xs = 120. and mi. z2 = 29. the lengths of the sides may be written where m and n are positive integers. xi + 6 + + y: = u2 = 3. 6722) and its angle 8 is between 60" and 61°. ~herefore sin 8 y12 1 cos 8 1 (x/z) 1 + + + 2mn/(ma n2) = -n. u4 For k = 4. As explained in the statement of Problem 3. and = = n4 = 12. u4 u 4 lh = (1 12)4 17 12 1/S. 34). 119. of the convergent c. (16. We find and corresl~onding triangles with sides (3. such that the ratios nl/ml. 02 = nz = 2.The sixth triangle ha9 sides (3360. for k = 2. m4 = 29. 30.. 4. page 121. s/s + = 7 7 5 6 . 5). 841 For k = 3 .. z4 = 985. n2.n2)/(m2 n2) m + + If we could find sequences of integers ni. x2 = 20.. y2 = 21. = 17. u2 vz =3 2 m2 = 5. m 0 tan 2 = > n. m2. + + + U. To find these sequences.SOLUTIONS T O PROBLEMS Other solutions of u'* + 2v2 = k 1 are obtained from A. and =.

O = 1 .(7)(7) = 1.cz + D ' . respectively. Z.BC = ad .bc = + 1. d = 1. and the convergent 3 satisfies the inequalities (5.b'c' = _+l. -7 lies between 9 and k. [ . Of the numbers and o 1 O +. we can write and y = cfz x=cy +d - a'z + b' + dt ' where. 1.bc = f1 (aa' (ea' and a'd' . c = O. then Y where A D (iii) Since x N = -dx+b cx-a . i ] = x = [i] and y = -169 .2 .. The other two come close but do not make it.Ax+B = C x f D .6). b = O. = + + (ii) If x = ay + b -and CY +d ad . +#. 4 . Hence x . 4. Hence y x. a = .= 1 -l 1+2 3 - 1 9 2 the first satisfies (5.bc = 1 . x N y because (-lo)(-5) .6). + 4s) + 3.387 .CONTINUED FRACTIONS Set 21.3874 1. ad . In F2. Shen + bct)x + ab' + bd' A Z + B + del)z + cbf + ddt . ). +S. page 130 = 1.4 5 118 (i) Since x ax b -7 with a = 1. The first sis convergents to a = f (1 are +. we have cx d ad .bc y and y N fl . note that also satisfies (5. 4) 3.6). 1 .x .

and 29 a11: Z2 + 52. hence P 17 = 12. 1 . . Q = 17 and 433 = 1Z2 +1p.b'c') Set 22.S O L U T I O N S TO PROBLEMS and AD ..a'bcd' . 2. 5 2.2q2 = f1 or pz & 1 = 2qz = q2 + q2.BC = = (aa' + bc')(cb' + dd') . 101.bc) (ad . . S Q we always have p2 .3.17 41 99 . + .bfc'(ad . Hence the second part of the problem can be solved by values of p and q such that p2 1 = 242.bc)(a'd' . On the other hand and the convergents to this continued fraction are - 1 1' - 3 2' 7 . Q = 5. or p2 . page 1.(ab' + bdf)(ca' + dc') + bb'cc' .I = 2q2.ab'c'd aa'dd' = = a'dt(ad . 6= 15. so we must calculate a1 " 1 12 + d 4 3 3 . . Since that d /SS irrational it is impossible to find two intcgers i d =2 a b 7 a and b such or such that a2 = 2bZ = be + b2.3 1. .5 ' 12' 2 9 ' 7 0 ' e . hence P = so \ve must calculate a ~ a : 3 = = z + 6 . 3. 1.bc) . 2.


The only extended account of the subject in English is that given in Vol. Hardy and E. Davenport. Edwin Beckenbach and Richard Bellman. for he gets to the heart of the matter quickly and with very little fussing. Wa~hington Carnegie Institiite of Washington. 1 of Chrystal's 1 Algebra. vol. The book by Daveilport is very good reading. 11. G . an old-fashioned yet still valuable text. 4th ed. 1919 . An Introduction to the Theory o Numbere. : New York : Chelsea. The Higker Arithmetic. 2. New York: Random House. London : Hutchinson's University Library. G . 1 1 1. H. vols. 1959. or deal with subject matter that has been referred to in the text. Edinburgh: Adam and Black. 1. 1952. The standard treatise on contiked fractions is Perron's Kettenbrüche. Algebra. 1950. New York : Chelsea. f . 1. 11. 1961.. E.References The books listed below either contain chapters on continued fractions. but this book is for the specialist. New Mathematical Library 3. 5. An Introduction to Inqualities. L. M. H. 1889 . Oxford : Clarendon Press. Inc. 3. Chrystal. reprinted. Wright. 1960. reprinted. No attempt has been made to compile a complete bibliography. Historly-ojthe Theory of Numbers.. 4. Dickson.

Xe~vRlathematical Library 1. Reading. 14. Inc. LeVeque. Number Theory and Its History7 Xew York: RicGra~vHill. 1929. E. 1956. Soc. . Xew York: llover Publications. 1962. 12. Xew York: Rai~dorn House Inc. Oskar Perroil. 1956. D. 9. J. 8. 2nd ed. A Studg in the History of Greelc Algebra. Robinson. Bullet'iiz American R4at. Ivan Kiveil. S. 1948. 351-361. 10. Van Xostrai~d Company. 7. Leipzig a i ~ d Berlin: Teiibner. 15. 1958. Cambridge: Cambridge University Press.160 CONTINUED FRACTIONS 6. vols. Iilc. Topics i n Number Theory.. Smit'h.Historg o j nlathematics.. 53 (1947) pp. 1 and 11. vol. Diophantus of Alexandria. Kew York: D. W. Die Lehre von den ICettenbrüchen. Mass. Inc. vols.. Ore. Numbers: Rational and Irrational. Unsymmetrical Approximations of Irrational Numbers. 1 and 11.. Irrational Numbers.. 11.. Kew I'ork : Random Hoiise. : Addison-Wesley. Kew RIat'hematical Library 7. 0. Uses of Injinity. Carus Rlonograph 11. 13. Wall. H. Leo Zippii~. Reprint. 1910. Analytic Theory of Continued Fractions. 1949. Heath. Xew York: John Wiley aild Sons. Ivan Niven. Raphael M. Sir Thomas L. 1961.h.

127 equivalent. 89. 82 e. 112ff symmetrical periodic pa* Of. 118 particular solution of. 77 periodic. 88ff periodic part of. 82-83 greatest common divisor. 103 Babylonians.19 differences of. 7 for d Ñ . 37 minimal solution of. 135 Cataldi. 72 acyclic. 60. 90 Hildebrand. 124ff. H. 89 discriminant. 7 fiilite simple. 32. 16 Euclid's Elements. 134 coconuts and monkeys. 90. 61 Coxeter. 123. 56 purely periodic. de Gelder. Jakob. F. i0ff . 135-136 equivalente classes.. 66 interpretation of.. 48ff conjugate. 132. 61. 7 geometrical interpretation of. 60 Dickson. 128 Huygens. 11 . 30 113 terminating. . 84. 35. 126 Fermat. 30. 95 Gardner. 53-54. 27. 30 Hurwitz's theorern. 138 geometric constructions. 127ff Eratosthenes. 50 Gauss. 81-82.Index absolute value. 98 Divina Proportione. 89. 113 Aryabhata. 99 continued fraction. B.lrchimedes. 130 Heath. 137. . A l . Rafael. 90 Euclid's algorithm. 11 ponvergents. 74-75. 132 Fibonacci numbers. 58 Beckenbach. d . 58 . 95 approximation theorems. 132."cattle l~roblem"of. 17 Hardy.. 90ff simple. H. 134 Brouncker. Martin.. 58. 9 Bellnlan. 77 infinite simple. 29 associate. 89-90 Diophantine equat ions. 29. 59 golden section (mean). Pietro Antonio. Lord. 29. 52. 30. 17 Euler. E. Cliristiaan. 6. 3 1ff general. R. 135 Farey sequences. 9 Bombelli. 37 solutions of. 147 Galois. 7 uniqueness.

74 method for calculating. 123. 51 reduced.52 Stern. 52 qTall. 8. 16 Klein. 51. b) = 1. 130 Newman. 31 m-by=ff. 58 Robinson. 135 Wright. 89. 132. Daniel.by = c. 108 irrational numbers. 77 Lagrange. 29 Segre. 129 number algebraic. 89 having no integral solutions. 52. 113ff) 118ff. 51ff + partial quotients. John. 127 transitive. 123. 90 Niven. B. 121 quadratic irrational. 138 Laplace. 130 Wythoff. 58. 30. M. xa . Ivan. 123 quadratic equation. 44. 95. 13 reflexive. W. x2 .. 61 pi ( T ) . 127 Rhind Papyrus. 61 game. (a. 6 4 6 5 Pell.Ny2 = k 1. 19 partial sums. 136 Le Veque. 61 Zippin. 52 irrational. 123 insoluble. R. 44 Ax f By = 4. 88 reduced. 46 integral part of x. 131ff minimal solution of. 137 Stieltjes. L. 51ff approximation of. E. x2 .Ny2 = f1. 136-137 prime number.. Luca. 77 Legendre. D. 19 Pythagorean equation. (a. 89. 84ff natural. 136.. 113ff. 121 m + b y = c . 76. b) = 1. James R. 129 Schwenter. 52. 51ff prime. 36 ax . 89. 96-98. 10Off rational fractions expansion of. 134 square root of 2 (G).. 31ff. 131ff PelI's equation. 8. John. 74 Minkowski. 5. A. 126. 139 Lagrange's t heorem. 89 PeIl'e equation. 127 uniqueness. 126 limit.3ya = -1.. 89. 84 Smith... 66 Pacioli. Daniel. 139 lattice points. 129 Serret. 123ff expansion of. 82 .162 CONTINUED FRACTIONS indeterminate equations Diophantine. 129 phyllotsxis. lOOff quadratic surd. F. 60. 56. 130 \Vallis. C. [x]. 19 transcendental. 11Off Lambert. 30. 132 Shanks. 63 logarithms conlnton. 30. 96-98. 42 ux by = c . 130 symmetric. 118ff. 118 Perron.

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->