Math.

450, Spring 2013

Existence of real numbers, decimal representation of real numbers Bolzano-Weierstrass Theorem, Cauchy sequences, and Cauchy Completeness Theorem
Jacek Polewczak Existence of real numbers
Denition 1. An ordered set is a set S = with relation < satisfying two properties: (a) If x S and y S then one and only one of the statements x < y, is true (b) If x, y, z S , if x < y and y < z , then x < z . For example Q, the set of rational numbers, is an ordered set if r < s, for r, s Q, is dened to mean that s r is a positive rational number. Remark 1. The statement x < y may be read as x is less than y or x is smaller than y or x precedes y . The notation x y indicates that x < y or x = y , without specifying which of these two is to hold. Equivalently, x y is the negation of x > y . Denition 2. Let E S be a subset of an ordered set S . If there exists a S such that x for every x E , we say that E is bounded above, and call an upper bound of E . Remark 2. Lower bounds are dened similarly, with in place of . Denition 3. Let E S be a bounded above subset of an ordered set S . If there exists an S satisfying the following properties: (i) is an upper bound of E . (ii) If < then is not an upper bound of E . Then is called the least upper bound of E or the supremum of E and we write = sup E . The least upper bound is unique. Remark 3. When E R (R denotes here the set of real numbers), then is the least upper bound of E if and only if (i) is an upper bound of E , and (ii) given any > 0, there is a number x( ) E for which x( ) > . Remark 4. Q is an ordered set that does not have the least-upper-bound property. Indeed, the set E = {r Q : r > 0 and r2 < 2} contains no largest number in Q. Notice that E as a subset of R has the least upper bound equal to 2. 1 + xn As another (complementary) example, consider set E = {xn }, where x1 = 1 and xn+1 = 2 , for n 2. 2 + xn Then, the sequence {xn } is an increasing sequence of rational numbers (i.e., xn Q with xn < xn+1 , for every 1 + xn n 1) such that xn = 2 2, for n 1. (Can you check it? ) Next, one shows (Can you do it? ) that 2 + xn lim xn = 2 Q. Thus, E does not posses the least upper bound in Q. 1 x = y, y<x

Denition 4. A eld is a set F with two operations, addition and multiplication, which satisfy so called eld axioms (A), (M), and (D). (A) Axioms for addition (A1) If x F and y F , then their sum x + y is in F . (A2) Addition is commutative: x + y = y + x for all x, y F . (A3) Addition is associative: (x+y)+z=x+(y+z) for all x, y, z F . (A4) F contains an element 0 such that 0 + x = x for every x F . (A5) To every x F corresponds and element x F such that x + (x) = 0. (M) Axioms for multiplication (M1) If x F and y F , then their product xy is in F . (M2) Multiplication is commutative: xy = yx for all x, y F . (M3) Multiplication is associative: (xy)+z=x(yz) for all x, y, z F . (M4) F contains an element 1 = 0 such that 1x = x for every x F . (M5) To every x F corresponds and element 1/x F such that x(1/x) = 1. (D) The distributive law x(y + z ) = xy + xz holds for all x, y, z F . Denition 5. An ordered eld is a eld F which is also and ordered set, such that (i) x + y < x + z if x, y, z F and y < z , (ii) xy > 0 if x, y F , x > 0 and y > 0. If x > 0, we call x positive ; if x < 0, x is negative. The set of rational numbers, Q, is an ordered eld. Theorem 1. There exists an ordered eld R which has the least-upper-bound property. R is unique in the following sense. If R1 and R2 are such elds, there exists an isomorphism of ordered elds between them, i.e., there exists a bijective mapping : R1 R2 which preserves the structure: for any x, y R1 , we have (x + y ) = (x) + (y ) and (xy ) = (x)(y ); for any x, y R1 with x < y , we have (x) < (y ). Moreover, R contains Q as a subeld. Proof. For details of the proof see, for example, Principles of Mathematical Analysis by Walter Rudin, or Mathematical Analysis An Introduction by Andrew Browder. The idea behind the proof is to construct members of R as certain subsets of Q, called Dedekind cuts. A (Dedekind) cut is, by denition, any set Q with the following three properties: (i) = and = Q, (ii) If p , q Q, and q < p, then q , (iii) If p , then p < s for some s (i.e., has no largest member). Moreover, every subset Q that satises (i)-(iii) has the form {r Q : r < a} for some a R; in fact, a = sup . 2

Corollary 1. Suppose that R is an ordered eld with the least-upper-bound property. (a) (b) Let = B R and B is bounded below. Let L be the set of all lower bounds of B . Then = sup L exists in R, and = inf B . is the greatest lower bound of B and B has the greatest-lower-bound property. If x, y R and x > 0, then there exists n N (N denotes here the set of natural numbers) such that nx > y. (c) If x, y R and x < y , then there exists a r Q such that x < r < y .

proof of (a). Since B is bounded below, L = . Since L consists of exactly those y R which satisfy the inequality y x for every x B , we see that every x B is an upper bound of L. Thus L is bounded above. By our assumption about R, L has a supremum in R; call it . If < then (see Denition 3) is not an upper bound of L, hence, B . It follows that x for every x B . Thus, L. If < then L, since is an upper bound of L. Thus we have shown that L but L if > . In other words, is lower bound of B and, but is not if > . This means that = inf B . proof of (b). Let A = {nx : n N}. If (b) were false, then y would be an upper bound of A. Then A has a least upper bound in R, = sup A. Next, x > 0 implies x < and x is not an upper bound of A. Thus, x < mx for some m N. But then < (m + 1)x, which is impossible, since is an upper bound of A. proof of (c). Combining x < y (i.e., y x > 0) with (b) yields n N such that n(y x) > 1. Applying (b) again (with x = 1), we obtain m1 , m2 N such that m1 > nx and m2 > nx, or equivalently m2 < nx < m1 . Hence there exists an integer m (with m2 m m1 ) such that m 1 nx < m. (Do you know why? ) Combining these inequalities we obtain nx < m 1 + nx < ny. However, n > 0, thus x< This proves (c) with r = m/n. Remark 5. Part (a) shows that an ordered eld with the least-upper-bound property has also the greatest-lowerbound property. Part (b) is referred to as the Archimedean property of R, while part (c) states that Q is dense in R: between any two real numbers there is a rational number. Theorem 2. For every real x > 0 and every integer n > 0 there is one and only one positive real y such that y n = x. This number y is written as n x or x1/n . Proof. For a proof see, for example, Principles of Mathematical Analysis by Walter Rudin. m < y. n

Decimal representation of real numbers

Let x > 0 be real and let n0 be the largest integer such that n0 x. We denote it by [x] and call it the integer part of x and x [x] the fractional part of x. Observe that existence of such n0 is guaranteed by the Archimedean property of R (see part (b) of Corollary 1). We will represent x[x] in the form a1 a2 a3 . . ., where each ai is a digit, i.e., ai is from the set {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}. The process of selecting ai is done inductively as follows. Divide the interval [0, 1), which contains x [x], into ten open equal parts 0, 1 2 2 3 9 1 , , , , , ..., ,1 10 10 10 10 10 10

and take as a1 the digit (numerator of left endpoint) corresponding to the subinterval that contains x [x]. Next, divide [a1 /10, (a1 + 1)/10) into ten equal parts 1 a1 1 a1 2 a1 9 a1 + 1 a1 a1 , + 2 , + 2, + 2 , ..., + 2, 10 10 10 10 10 10 10 10 10 10 and choose as a2 the digit (multiplier of 1/102 for left endpoint) corresponding to the subinterval that contains x [x]. Continue, to get a1 a2 a3 . . . an hence the decimal of x: x = n0 .a1 a2 a3 . . . . If x happens to be an endpoint of one of the subintervals, the digits for x are all 0 from some stage on, and conversely. In this case x is just the nite sum x = n0 + a1 a2 ak + 2 + + k 10 10 10

and the decimal can be viewed as an abbreviation for this sum. If x is not an endpoint, the decimal has innitely many nonzero digits and a resulting sum requires a further consideration. To this end, let E be the set {n0 + a1 a2 ak + 2 + + k : ak {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}, k = 0.1, 2, 3, . . . .} 10 10 10

Since set E is bounded above by n0 + 1 (Do you know why? ), sup E exists (see Theorem 1) and x = sup E = n0 .a1 a2 a3 . . . Proposition 1. Distinct numbers have distinct decimal representations. Proof. Suppose x < y . Since the set {10k : k = 1, 0, 1, 2, . . . } is unbounded, the Archimedean property implies that there is k such that 1 0< < 10k . yx Hence, clearly, the choices of ak for x and y are dierent.

Remark 6. (a) The endpoints of the intervals of subdivision are assigned decimals that we call terminating, i.e., they conclude with an innite repetition of the digit 0. For example, 1 125 = = 0.125000 . . . 8 1000 If we settled on left-open, right-closed intervals for the subdivisions, the only dierences in the decimal representations would be at the endpoints, where a decimal is of the form xxx999 . . . would result, such as 1 = 0.124999 . . . 8 Another alternative is to consider closed subintervals and accept an ambiguity in the decimal representation of some numbers; each endpoint would belong to two subinetervals at some stage, and two choices ak 000 . . . and (ak 1)999 . . . would result, amounting to 0.125000 . . . = 0.124999 . . . . (b) Instead of dividing intervals in ten equal parts and assigning digits from the set {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}, one can make division into b equal parts (2 b N) and use the digits 0, 1, . . . b 1. The result is the base b expansion of numbers. We observe that assuming the closed interval method in the subdivision process, the successive digits designate a succession of closed intervals In (in which the numbers is to be sought) with the following property: I1 I2 I3 . . . Such a sequence of closed intervals In is called a nest.

and length In 0

Note that for any nest of closed intervals in an ordered eld R with Archimedean property, the intersection
n=1

In

is either empty or a single point. The property of R that guarantees the existence of a number in R for every base b expansion is

Nested Intervals Property: For every nest of closed intervals in R the intersection
n=1

In is nonempty.

Bolzano-Weierstrass Theorem
Denition 6. For a given a set S = (the set S does NOT have to be a subset of real numbers), a sequence is a function whose domain is a set of the form {n Z : n m} and its range is S . m is often 1 or 0. It is customary to denote a sequence by a letter such as x and to denote its value at n as xn rather than x(n). Denition 7. An innite sequence of real numbers {xn }, n 1 converges to x as n if and only if for each > 0 there is N N such that |xn x| < for all n > N . We will use the notation
n

lim xn = x or xn x as n .

Denition 8. Given as sequence of real numbers {xn }, consider a sequence {nk } of positive integers such that n1 < n2 < n3 < . The sequence {xnk } is called a subsequence of {xn }. If {xnk } converges as k , its limit is called a subsequential limit of {xn }. Remark 7. Note that {xn } converges to x if and only if every subsequence of {xn } converges to x. (Prove it ! ) 5

Denition 9. A sequence of real numbers {xn }, n 1 is increasing if xn xn+1 for all n N and strictly increasing if xn < xn+1 for all n N. A sequence of real numbers {xn }, n 1 is decreasing if xn xn+1 for all n N and strictly decreasing if xn > xn+1 for all n N. All four kinds of these special sequences are called monotone. Lemma 1. Every innite sequence of real numbers has a monotone subsequence. Proof. It is enough to consider sequences that have no increasing subsequence and show that they each have necessarily a decreasing subsequence. Suppose {xn } has no increasing subsequence. We argue that there is a largest value among {xn : n N}, in other words, max xn exists. If not, then x1 is not the largest, so there is a rst index n1 with x1 < xn1 . But xn1 is not the largest, so there is a rst index n2 giving xn1 < xn2 and n2 > n1 . Continuing, we nd an increasing subsequence contrary to assumption. Now, choose as xm1 the rst occurrence of the largest value. Consider {xn : n > m1 }, which has no increasing subsequence, and take as xm2 the rst occurrence of its largest value. We have xm1 xm2 . Next, we continue by induction to obtain a decreasing subsequence. Denition 10. A sequence {xn } of real numbers is bounded if there exists 0 < M < such that |xn | M for all n N. Remark 8. Every convergent sequence of real numbers is bounded. (Can you prove it? ) Theorem 3. A bounded and monotone innite sequence of real numbers converges. Proof. It is enough to consider a bounded increasing sequence. The case of a bounded from below and decreasing sequence follows a very similar line of arguments as below, if one applies part (a) of Corollary 1. Let {xn } be a bounded increasing sequence. If E = {xn : n N}, then Theorem 1 implies that x = sup E exists and x R. We will show that lim xn = x. Let > 0. Since x is not an upper bound of E (see (ii) of
n

Denition 3), there exists N N such that xN > x . Since {xn } is increasing, we have xN xn for all n > N . This together with xn x for all n N yields x xn x < x + , The above inequality implies that |xn x| < for all n > N . Equivalently, lim xn = x.
n

Theorem 4 (Bolzano-Weierstrass Theorem). Any bounded innite sequence {xn } of real numbers contains a convergent subsequence. Proof. Lemma 1 implies that {xn } contains a bounded and monotone subsequence, while Theorem 3 shows that such a subsequence is convergent.

Cauchy sequences
Denition 11. A sequence of real numbers {xn }, n 1 is called a Cauchy sequence if and only if for each > 0 there is N N such that |xn xm | < for all n > N and all m > N . Lemma 2. Every convergent sequence is a Cauchy sequence. 6

1 2 Let xm be any element of {xn } with m > N . We have |xn x| <

Proof. Suppose lim xn = x and let n N N such that

> 0. Then from the denition of convergence (see Denition 7) there is for all n > N .

|xn xm | = |xn x + x xm | |xn x| + |xm x| < Thus {xn } is a Cauchy sequence. Lemma 3. Every Cauchy sequence is bounded.

1 1 + = . 2 2

Proof. From the denition of Cauchy sequence (see Denition 11) with = 1, there is N0 N such that |xn xm | < 1 for all n > N0 and m > N0 . In particular, |xn xN0 +1 | < 1 if n > N0 . Also, |xn | = |xn xN0 +1 + xN0 +1 | |xn xN0 +1 | + |xN0 +1 | < 1 + |xN0 +1 |, for all n > N0 . Now, if M = sup{|x1 |, |x2 |, . . . , |xN0 +1 , 1 + |xN0 +1 |}, then |xn | M for all n N. Theorem 5 (Cauchy criterion for convergence). A necessary and sucient condition for convergence of a sequence of real numbers {xn } is that it be a Cauchy sequence. Proof. Lemma 2 shows that a convergent sequence is a Cauchy sequence. Now, assume that {xn } is a Cauchy sequence. Lemma 2 implies that {xn } is a bounded sequence, while Theorem 4 yields that there is a subsequence {xnk } of {xn } which converges to the limit x. We shall show that the sequence {xn } itself converges to x. Let > 0. Since {xn } is Cauchy sequence there is N1 N such that |xn xm | <
k

1 2

for all n > N1 and m > N1 .

Next, since lim xnk = x, there is N2 N such that |xnk x| < 1 2 for all k > N2 .

Now, x k N2 such that nk N1 . (Observe that such a choice is always possible since n1 < n2 < n3 < .) Then for all n N1 we have |xn x| = |xn xnk + xnk | |xn xnk | + |xnk x| < Thus lim xn = x.
n

1 1 + = . 2 2

Theorem 5 is very useful because it is often easier to show that a sequence is Cauchy than to show that it converges. The following example shows that we can prove that a sequence is Cauchy even when we have no idea what its limit is. 7

Example 1. Any sequence of real numbers satisfying |xn xn+1 | is convergent. Proof. For m > n we have |xn xm | = |xn xn+1 + xn+1 xn+2 + xn+2 + + xm1 xm | |xn xn+1 | + |xn+1 xn+2 | + + |xm1 xm | 1 1 n + m1 2 2 mn 1 1 1 1 1 = n1 = 2 2k 2n1 2mn k=1 1 2n1 . 1 , 2n n N,

For a given > 0, choose N N such that n N implies 1/2n1 < . Thus, we have proved that {xn } is Cauchy. By Theorem 5, therefore, it converges to some real number. Remark 9. A sequence of real numbers satisfying lim (xn+1 xn ) = 0 is not necessarily Cauchy. Indeed, xn = log n n has the property n+1 xn+1 xn = log(n + 1) log n = log log 1 = 0 n as n . However, the sequence {log n} does not converge, in fact, it diverges to as n .

Cauchy Completeness Theorem

One proves the following result: Theorem 6 (Cauchy Completeness Theorem). Assume that R is an ordered eld with the Archimedean property (see part (b) of Corollary 1) and consider the following statements: (a) (b) (c) R has the least-upper-bound property, R has the Nested Intervals Property, Every Cauchy sequence in R converges.

Then (a) (b) (c). Theorem 6 provides equivalent characterizations of any ordered eld with the Archimedean property, while Theorem 1 exhibits existence of a unique (up to isomorphism) such eld, which we call the real numbers and denote by R. Furthermore, the rational numbers Q R are dense in R (see part (c) of Corollary 1) and R \ Q are called the irrational real numbers. Can you prove the Cauchy Completeness Theorem ?