Week 1 Continuity of the inverse function. Infinite series

Version 2025/02/02 Week 1 in PDF All notes in PDF To other weeks

These notes are being developed to reflect the content of the Real Analysis course as taught in the 2024/25 academic year. The first half of the course is lectured by Dr Yuri Bazlov. Questions and comments on these lecture notes should be directed to Yuri.Bazlov@manchester.ac.uk. The second half will be lectured by Dr Mark Coleman.

Pre-requisite: Mathematical Foundations and Analysis

We build upon what was achieved in the Mathematical Foundations and Analysis (MFA) course, taught in Semester 1. Limit of a sequence, continuous function and limit of a function remain key notions in Real Analysis, which develops the β€œanalysis” part of MFA further. Important functions of real variable, used in MFA, will be formally defined, and their properties proved. This includes the power function x𝛼 with arbitrary real 𝛼; exponential function ax and the logarithm; trigonometric functions.

Hence, the Real Analysis course will constantly require the students to call on their knowledge of definitions and results introduced in MFA.

An informal introduction (not covered in lectures)

The first part of the course will introduce infinite series. Recall Euler’s identity

eiπœ‹+1=0,

often brought up as an example of a beautiful mathematical result. But what exactly is there to prove? Let us try to understand this equation and its connection to Real Analysis.

First of all, e is the irrational number discovered by Jacob Bernoulli in 1683 (and initially written as β€œb”) in his study of compound interest. It is defined as

e=limnβ†’βˆž(1+1n)n, equivalently, e=βˆ‘n=0∞1n!=1+11+11β‹…2+11β‹…2β‹…3+….

Both formulae involve an infinite process, which in practice can only approximate e to a given precision. Note how the limit of a sequence, defined in MFA, appears next to something new: a β€œsum of infinitely many numbers”.

What is eiπœ‹, though? Raising e to an imaginary power is done via the rule

ez=1+z1!+z22!+z33!+…

This is where the meaning of Euler’s identity starts to come across. The β€œinfinite sum” for eiπœ‹ breaks down into the real and imaginary parts, leading to two equations,

1βˆ’πœ‹22!+πœ‹44!βˆ’πœ‹66!+β‹―=βˆ’1 and πœ‹βˆ’πœ‹33!+πœ‹55!βˆ’πœ‹77!+β‹―=0.

Adding up ten terms of each β€œinfinite sum”, we obtain approximately βˆ’1.000000004 and βˆ’5.3Γ—10βˆ’10 (check the calculation here!) which suggests that if we keep adding new terms generated by the same rule, in the limit we will indeed get βˆ’1 and 0, respectively. This is mysterious: πœ‹ is transcendental, so no finite sum like this can be a rational number. A proper way to prove that πœ‹ satisfies these equations is to express cos⁑(x) and sin⁑(x), functions given by ratios of sides in a right-angled triangle where x is in radians, as

cos⁑(x)=1βˆ’x22!+x44!βˆ’x66!+…,   sin⁑(x)=xβˆ’x33!+x55!βˆ’x77!+…

These infinite expansions, apparently known already to Madhava (ca. 1400), are enormously important in mathematics and have many compelling applications in the sciences. To prove these formulas is the same as to prove the formula eix=cos⁑(x)+isin⁑(x) (used in MFA without proof). Complete and rigorous theory, leading to these expansions, is part of what is covered in the Real Analysis course. In the first part of MATH11112, we will:

  • β€’ formally define infinite series and make sense of β€œsums of infinitely many numbers”;

  • β€’ learn about ways to tell whether a given series converges, i.e., has a sum;

  • β€’ understand power series which consist of power of x with some coefficients, and see why they define β€œsmooth” continuous functions of x if they converge.

Expansion of functions as power series is intimately connected with differentiation, a formal treatment of which begins the second part of the course. Higher derivatives of a function are key to approximating the function by polynomials, called Taylor polynomials. It is this theory that allows us to write a β€œgood” function as sum of a Taylor series.

The course concludes with the third part devoted to rigorous treatment of integration. A key result is the Fundamental Theorem of Calculus, which demonstrates that integration is truly a reverse operation to differentiation.

Further study of series: a power series is a sum of infinitely many functions of the form axn. In 1807, Joseph Fourier publicised a class of scientific problems which require calculating infinite sums of more sophisticated functions, such as asin⁑(nx). The theory of series that we develop in Real Analysis serves as a foundation for the study of Fourier series and other advanced series in mathematics, science and engineering.

End of the informal introduction.

The Inverse Function Theorem

We begin with a result which could have been proved in Mathematical Foundations and Analysis (MFA), given more time. The Inverse Function Theorem will be used to define a function x↦xr where r is rational. First, a definition.

Definition.

Let AβŠ†R. A real-valued function f on A is

.
increasing, if βˆ€x,xβ€²βˆˆA, x<xβ€² ⟹ f(x)≀f(xβ€²);
strictly increasing, if βˆ€x,xβ€²βˆˆA, x<xβ€² ⟹ f(x)<f(xβ€²);
decreasing, if βˆ€x,xβ€²βˆˆA, x<xβ€² ⟹ f(x)β‰₯f(xβ€²);
strictly decreasing, if βˆ€x,xβ€²βˆˆA, x<xβ€² ⟹ f(x)>f(xβ€²).

A function satisfying one of the above conditions is called (strictly) monotone.

The above applies to sequences which are functions on A=N.

We can use monotone sequences to calculate limits of functions. The following is an MFA-style result:

Lemma 1.1: limit of f from the left via strictly increasing sequences.

For a function f, defined on (a,b), the following are equivalent:

  • (i) limxβ†’bβˆ’f(x)=β„“.

  • (ii) For all strictly increasing sequences (xn) such that xnβ†’b as nβ†’βˆž, one has limnβ†’βˆžf(xn)=β„“.

Remark: the Lemma can be expressed in words as follows:

The limit of f at b from the left is the common limit of all sequences (f(xn))nβ‰₯1, where a sequence (xn)nβ‰₯1 is strictly increasing and converges to b.

  • Proof of the Lemma (not given in class). (i) β‡’ (ii): let πœ€>0 be arbitrary. First, we use the definition of limxβ†’bβˆ’f(x)=β„“ to generate 𝛿>0 such that |f(x)βˆ’β„“|<πœ€ for all x∈(bβˆ’π›Ώ,b).

    Now we let (xn)nβ‰₯1 be a strictly increasing sequence, and use the above 𝛿>0 in the definition of β€œxnβ†’b as nβ†’βˆžβ€ to generate N∈N such that nβ‰₯N implies |xnβˆ’b|<𝛿. That is, xN,xN+1,β€¦βˆˆ(bβˆ’π›Ώ,b+𝛿). Since the sequence is strictly increasing with limit b, no term can exceed b, so in fact xN,xN+1,β€¦βˆˆ(bβˆ’π›Ώ,b).

    But then, by the choice of 𝛿, |f(xn)βˆ’β„“|<πœ€ for all nβ‰₯N. We have shown that β„“ satisfies the definition of limit for the sequence (f(xn))nβ‰₯1, and so (ii) is proved.

    (ii) β‡’ (i): to prove the contrapositive of this implication, we assume that the statement β€œlimxβ†’bβˆ’f(x)=ℓ” is false. This means that there exists some πœ€0>0 such that for all 𝛿>0, the interval (bβˆ’π›Ώ,b) contains a point, say x(𝛿), with |f(x(𝛿))βˆ’β„“|β‰₯πœ€0.

    Choose 𝛿1=1 and construct x(𝛿1)∈(bβˆ’π›Ώ1,b). We have |f(x(𝛿1))βˆ’β„“|β‰₯πœ€0.

    Then, for each nβ‰₯2, choose 𝛿n=min(1n,bβˆ’x(𝛿nβˆ’1)) and construct x(𝛿n)∈(bβˆ’π›Ώn,b).

    Since bβˆ’1n<x(𝛿n)<b, by the Sandwich Rule x(𝛿n)β†’b as nβ†’βˆž. Also, since x(𝛿n)>bβˆ’(bβˆ’x(𝛿nβˆ’1))=x(𝛿nβˆ’1), the sequence x(𝛿1),x(𝛿2),… is strictly increasing.

    We still have, by construction, that |f(x(𝛿n))βˆ’β„“|β‰₯πœ€0 for all n. Therefore, β„“ fails to satisfy the definition of the limit of the sequence (f(x(𝛿n)))nβ‰₯1, i.e. (ii) is false. β–‘

(dots on a graph of a function converging to a limit)

Figure 1.1: Lemma 1.1 says that the limit of f at b from the left is the same as the common limit of all sequences f(x1),f(x2),… where x1,x2,… strictly increase and converge to b

The Lemma is illustrated by Figure 1.1. The next result mirrors the Lemma to deal with a limit from the right:

Corollary: limit of f from the right via strictly decreasing sequences.

For a function f, defined on (a,b), the following are equivalent:

  • (i) limxβ†’a+f(x)=β„“.

  • (ii) For all strictly decreasing sequences (xn) such that xnβ†’a as nβ†’βˆž, one has limnβ†’βˆžf(xn)=β„“. β–‘

In other words,

the limit of f at a from the right is the common limit of all sequences (f(xn))nβ‰₯1, where a sequence (xn)nβ‰₯1 is strictly decreasing and converges to a.

We are now ready to prove the first theorem of the course.

Theorem 1.2: the Inverse Function Theorem for strictly increasing functions.

A strictly increasing continuous function f:[a,b]β†’[f(a),f(b)] has an inverse g:[f(a),f(b)]β†’[a,b] which is strictly increasing and continuous.

  • Proof. f:[a,b]β†’[f(a),f(b)] is surjective, because for every d∈[f(a),f(b)] the Intermediate Value Theorem (and its corollary in MFA) gives c in [a,b] such that f(c)=d.

    A strictly increasing f is injective: indeed, if x1β‰ x2, then either x1<x2 and so f(x1)<f(x2), or x1>x2 and so f(x1)>f(x2). In either case f(x1)β‰ f(x2).

    We have shown that f is bijective, hence it has an inverse g=fβˆ’1:[f(a),f(b)]β†’[a,b].

    We prove that g is strictly increasing by contradiction. Assume not, then there exist y1, y2 such that y1<y2 and g(y1)β‰₯g(y2). Since f is increasing, f(g(y1))β‰₯f(g(y2)). Since f=gβˆ’1, this reads y1β‰₯y2, but at the same time y1<y2, a contradiction.

    We prove that g is continuous at an arbitrary point d of its domain by verifying the criterion of continuity, seen in MFA:

    limy→dg(y)=g(d).

    By results from MFA, this is equivalent to

    limyβ†’dβˆ’g(y)=limyβ†’d+g(y)=g(d).

    We first show that limyβ†’dβˆ’g(y)=g(d). We would like to use Lemma 1.1 for this, so we let (yn) be a strictly increasing sequence in [f(a),f(b)] which converges to d. Since g is a strictly increasing function, (g(yn)) is a strictly increasing sequence; it is also bounded (lies in [a,b]), hence by a result from MFA, (g(yn)) has a limit, say c, in [a,b].

    Then by Lemma 1.1, limnβ†’βˆžf(g(yn))=limxβ†’cβˆ’f(x) which is f(c) as f is continuous. Since f=gβˆ’1, this says that limnβ†’βˆžyn=f(c). Thus, d=f(c), hence g(d)=c.

    We have proved that the common limit of all sequences (g(yn)), where yn strictly increases and converges to d, is g(d). By Lemma 1.1, this means that limyβ†’dβˆ’g(y)=g(d).

    The proof that limyβ†’d+g(y)=g(d) is completely similar, based on the Corollary to Lemma 1.1, and we omit it. Continuity of g at d is proved. β–‘

Example: the pth root function.

Let p∈N. Show that the pth power function [0,+∞)β†’[0,+∞), x↦xp, has a continuous inverse (denoted y↦yp and called the pth root function).

Solution: define f:[0,+∞)β†’[0,+∞) by f(x)=xp. Then f is strictly increasing on [0,+∞). Apply Inverse Function Theorem 1.2 to the restriction [0,b]β†’f[0,bp] to get a continuous inverse  p:[0,bp]β†’[0,b]. Since b>0 can be made arbitrarily large, this defines the continuous function  p on all of [0,+∞).

We compose continuous, strictly increasing functions to define a rational power function:

Example: raising to rational power pq where p∈Z, q∈N.

Define xpq=(xq)p. This is a continuous function of x where x∈(0,+∞).

Remark: one can deduce from the definition of a rational power that

x>1, r,s∈Q, r<s⟹xr<xs.

This allows us to formally define arbitrary real powers of x:

Definition: x𝛼 where x>0 and π›ΌβˆˆR.

If xβ‰₯1, define x𝛼=sup{xr:r∈Q,r≀𝛼}.

If 0<x<1, define x𝛼=(1/x)βˆ’π›Ό.

A disadvantage of this definition is that proving the expected properties of powers such as x𝛼x𝛽=x𝛼+𝛽 requires work. We will soon obtain a more useful expression for powers via the exponential function.

Infinite series: definition

Definition: infinite series, convergent series, sum.

For real numbers an, an infinite series is an expression of the form βˆ‘n=1∞an (also written as a1+a2+β‹―+an+…, βˆ‘nβ‰₯1an or just βˆ‘an).

The nth partial sum of this series is the finite sum of terms up to and including an: sn=a1+β‹―+an=βˆ‘i=1nai.

If the sequence of partial sums converges: limnβ†’βˆžsn=s, we say that the series βˆ‘n=1∞an is convergent with sum s, and write βˆ‘n=1∞an=s.

Remarks on the definition: (i) βˆ‘n=1∞an=s is a actually a shorthand which means β€œthe series βˆ‘n=1∞an is convergent with sum s”.

(ii) Any series that is not convergent is said to be a divergent series.

(iii) A series can start from n=N (any integer) instead of n=1: aN+aN+1+β‹―=βˆ‘n=N∞an. For example, it is common to start from n=0. The nth partial sum will still be the sum which ends with an: e.g., for the series a0+a1+a2+a3+…,

  • β€’ a0 is the 0th partial sum,

  • β€’ a0+a1 is the 1st partial sum, a0+a1+a2 is the 2nd partial sum, and so on.

Basic examples of convergent/divergent series are discussed in week 1 supervision classes.

Alert: a strict definition of convergence.

The definition of convergence of the series βˆ‘n=1∞an, used in Real Analysis, is very strict: if the number sequence a1, a1+a2, a1+a2+a3,… does not have a limit, then the series has no sum.

Weaker definitions can assign a β€œsum” to some particular types of series which we consider divergent: CesΓ ro sum, Abel sum etc. They are used in specialist applications which are beyond this course.

The geometric series

The next example is simple yet important: we will see that more complicated series can be studied by comparing them to a geometric series. We revisit a result seen in MFA.

Proposition 1.3: convergence and sum of geometric series.

Let a,r∈R. The geometric series with initial term a and ratio r,

a+ar+ar2+ar3+β‹―=βˆ‘n=0∞arn,

is convergent if |r|<1, with sum a1βˆ’r.

  • Proof. The nth partial sum of the series is sn=a(1+r+r2+β‹―+rn). The calculation

    (1+r+r2+β‹―+rn)(1βˆ’r)=1βˆ’r+rβˆ’r2+r2βˆ’r3+β‹―+rnβˆ’rn+1=1βˆ’rn+1 where the intermediate terms cancel, gives us the formula

    sn=a1βˆ’rn+11βˆ’r.

    If |r|<1, we recall from MFA that rn+1 tends to 0 as nβ†’βˆž, so by Algebra of Limits of convergent sequences,

    s=limnβ†’βˆžsn=a1βˆ’01βˆ’r=a1βˆ’r.

    The sum of the series, is, by definition, the limit of partial sums if it exists. Hence the sum of the geometric series is a/(1βˆ’r) as claimed. β–‘

Convergence of series with non-negative terms

Unlike the geometric series, usually there is no nice formula for the nth partial sum sn. We still want to decide if a series is convergent, so we prove theorems known as β€œconvergence tests”. Our first few tests work for series where all terms are non-negative.

Theorem 1.4: boundedness test for non-negative series.

Let a series a1+a2+… have anβ‰₯0 for all n. The following are equivalent:

  • (i) the partial sums s1,s2,… are bounded above;

  • (ii) the series is convergent.

If (i) and (ii) hold, the sum of the non-negative series is the least upper bound, sup{sn:nβ‰₯1}, of its partial sums.

  • Proof. The partial sums of a non-negative series form an increasing sequence, because sn+1=sn+an+1β‰₯sn for all n. We know from MFA that an increasing sequence (sn)nβ‰₯1 of real numbers has a limit iff it is bounded above, and then the limit is the supremum of the terms of the sequence. β–‘

Corollary: only two convergence types for non-negative series.

A non-negative series a1+a2+… is either

  • β€’ convergent with a non-negative finite sum: βˆ‘n=1∞an=s, 0≀s<+∞, or

  • β€’ divergent if sup{sn:nβ‰₯1}=+∞.

In the latter case we use the symbolic notation β€œβˆ‘n=1∞an=+βˆžβ€ and say that the non-negative series diverges to +∞.

Alert.

Notation βˆ‘n=1∞an=+∞ (to mean that the series is divergent) and βˆ‘n=1∞an<+∞ (to mean that the series is convergent) is used only for series with non-negative terms!

The next test is used very often.

Theorem 1.5: the comparison test for non-negative series.

Assume that 0≀an≀bn for all n. If the series b1+b2+… is convergent (with sum T), then the series a1+a2+… is convergent (with sum at most T).

If a1+a2+… is divergent, then b1+b2+… is also divergent.

  • Proof. Write sn=a1+β‹―+an and tn=b1+β‹―+bn. As a1≀b1, a2≀b2 etc, we have sn≀tn, where tn≀T by Theorem 1.4. Hence s1,s2,… have an upper bound T, so by boundedness test, Theorem 1.4, the series a1+a2+… is convergent.

    The sum βˆ‘n=1∞an is the least upper bound of (sn)nβ‰₯1, and T is an upper bound. Hence βˆ‘n=1∞an≀T.

    Now β€œa1+a2+… is divergent β‡’ b1+b2+… is divergent” follows by contrapositive. β–‘

To use the Comparison Test, we need to compare with some easy series βˆ‘bn, yet the Test does not tell us how to find it. Hence we develop further convergence tests.

Theorem 1.6: the Ratio Test for positive series.

For a positive series βˆ‘nβ‰₯1an, suppose that the limit β„“=limnβ†’βˆžan+1an exists. Then if 0≀ℓ<1, the series is convergent, and if β„“>1, the series is divergent.

  • Proof. The case 0≀ℓ<1. Choose a positive πœ€ such that β„“+πœ€<1. For example, πœ€=(1βˆ’β„“)/2 works.

    Since β„“=limnβ†’βˆžan+1an, there exists N such that an+1an<β„“+πœ€ for all nβ‰₯N. Write this as an+1<ran, where r=β„“+πœ€. Then aN+2<raN+1<r2aN, and, repeating this, we obtain aN+k<rkaN. We have

    a1+a2+β‹―+aN+k≀a1+β‹―+aNβˆ’1+aN+raN+β‹―+rkaN≀(a1+β‹―+aNβˆ’1)+aN1βˆ’r. The upper bound that we have obtained is a finite constant which does not depend on k. Thus, partial sums of the series a1+a2+… are bounded, so by Theorem 1.4, the series is convergent.

    The case β„“>1. Put πœ€=β„“βˆ’1. There is N such that β„“βˆ’πœ€<an+1an for nβ‰₯N. But β„“βˆ’πœ€=1, so 1<an+1an, equivalently an<an+1, for nβ‰₯N. In particular, all an for n>N are greater than the positive constant aN. Hence snβ‰₯(nβˆ’N)aN which is unbounded. β–‘

Alert: the Ratio test may be inconclusive.

If β„“=1 or the limit does not exist, this test does not tell us anything: the harmonic series βˆ‘nβ‰₯11n is divergent (see the next Chapter), and the series of inverse squares βˆ‘nβ‰₯11n2 is convergent (see the first exercise sheet). Both have β„“=1.

In the following test (not taught in lectures, not examinable), we use the nth root function x2n, defined earlier.

Theorem 1.7: The nth Root Test for non-negative series.

For a non-negative series βˆ‘nβ‰₯1an, suppose that the limit β„“=limnβ†’βˆžann exists. Then if 0≀ℓ<1, the series is convergent, and if β„“>1, the series is divergent.

Remark: Again, if β„“=1 or the limit does not exist, this test does not tell us anything.

  • Proof. (not given in class: very similar to the proof of the Ratio Test; not examinable.) The case 0≀ℓ<1. Choose a positive πœ€ so that r=β„“+πœ€ is still less than 1; for example, πœ€=(1βˆ’β„“)/2 works.

    By definition of limit, there is N such that ann<β„“+πœ€=r for all nβ‰₯N. Then an<rn for nβ‰₯N, so partial sums of the series are bounded by a1+β‹―+aN+11βˆ’r, implying convergence.

    The case β„“>1. Put πœ€=β„“βˆ’1. Since β„“ is the limit of ann, there is N such that β„“βˆ’πœ€<ann for nβ‰₯N. But β„“βˆ’πœ€=1, so 1<ann, equivalently 1<an, for nβ‰₯N. We therefore have snβ‰₯nβˆ’N which is unbounded. β–‘

Version 2025/02/02 Week 1 in PDF All notes in PDF To other weeks