Chapter IV · Limits of functions of a positive integral variable
Chapter IV problems
Eleven of Hardy’s Miscellaneous Examples on Chapter IV, with graded hints and full solutions.
How to use this page
These problems are chosen from the Miscellaneous Examples at the end of Hardy’s Chapter IV, for variety rather than difficulty: monotone sequences, recursions, averages, functions defined by limits, series and complex numbers. Each is restated briefly; Hardy’s wording is in the book. Try each problem for a while before opening the first hint, and open the hints one at a time: the first only points in a direction, the last is nearly a solution. Dots show difficulty (one: routine; three: needs a real idea).
Sequences and their limits
The function takes the values for . Express it by a formula without trigonometric functions.
Hint
Hint
Solution
Check the four residues: gives ; gives ; gives ; gives . Since has period , this covers all .
The idea generalises. If is a -th root of unity other than , then . Applying this to , where , shows that is when divides and otherwise.
Let and for . Show that increases, decreases, and ; deduce that both tend to limits. Then show that the two limits are equal.
Hint
Hint
Hint
Solution
increases. . By Bernoulli, , so the ratio exceeds .
decreases. , so . By Bernoulli, , so the ratio exceeds .
. .
Both converge. , so is increasing and bounded above; , so is decreasing and bounded below. By §69 both have limits, and .
The limits are equal. For and : (Bernoulli again), so . With this gives . Hence , and . The common limit is (Chapter IX).
Let be the arithmetic mean of the products over all ways of writing as a sum of two distinct positive integers (order ignored). Show that .
Hint
Hint
Solution
Using and ,
so .
If is odd: . If is even: . (Check: has splittings , , mean product .) In both cases : and .
The answer does not depend on the reading of “distinct pairs”: over ordered pairs the mean is , with the same limit. The limit is in disguise.
Recursions
Let , and . Let be the positive root of . Show that increases if , decreases if , and tends to in either case.
Hint
Hint
Hint
Solution
The side of is preserved. , and , so has the same sign as . By induction every is on the same side of as (or all equal , if ).
Monotone. with . So , and since this has the sign of . If , all and the sequence increases; if , it decreases.
The limit. In the first case increases and is bounded by ; in the second it decreases and is bounded below by . Either way it has a limit (§69). Letting in gives , so or ; as , .
Try this. This is the recursion with . Change and find how large must be for . For the limit is .
- band l ± δ
- l
- n₀ (drag)
- n ≥ n₀, good
- n ≥ n₀, bad
What to notice
The terms climb steadily and never overshoot , as the proof says. The gap shrinks by roughly the factor at each step, because .
Let , , and . Show that
and deduce that .
Hint
Hint
Solution
Write , so that for all . Then
so , where (all are positive, so ). By induction , which is Hardy’s identity.
Since , , so and . Solving for gives .
The exponent explains Heron’s speed: once is about , the next step makes it about . The number of correct digits roughly doubles at every step.
Try this. Set to and then to , and find the smallest each time. Change .
- band l ± δ
- l
- n₀ (drag)
- n ≥ n₀, good
- n ≥ n₀, bad
What to notice
Making a thousand times smaller costs only about one more step: the error is squared at every step.
Let , and for . Express in terms of the ’s, and show that if then .
Hint
Hint
Solution
Formula. , so by induction
Limit. Put with . Then
The first part tends to since . For : the are bounded, say . Let and choose with for . For ,
The first term tends to as , so it is below for large , and then . Hence and . The old terms are not small, but they are damped by high powers of ; the recent terms are not damped much, but they are small.
Averages
Show that if then . Show that the converse is false by taking .
Hint
Hint
Hint
Solution
Since with , it is enough to show that the average of a null sequence is null.
The are bounded (a convergent sequence is bounded), say . Let and choose with for . For ,
With fixed, as soon as . So the average is less than in absolute value for all large , and tends to .
This is Hardy’s “split into two groups” argument, which he urges the reader to study with great care: the first group has terms that need not be small, but their number is fixed while grows; the second group has many terms, but each is small. (Hardy lets the split point grow slowly with ; a fixed chosen from works equally well.)
Converse false. For the sum is the number of odd numbers up to , that is , so the average lies between and and tends to , although oscillates. (Compare the series , whose partial sums these are: averaging assigns it the value .)
Corollary (Misc. Examples IV, 17). If , then : apply the result to (with ), whose average is .
Try this. These are the averages for . Find for .
- band l ± δ
- l
- n₀ (drag)
- n ≥ n₀, good
- n ≥ n₀, bad
What to notice
The averages settle on , approaching it from above at the odd places and sitting on it at the even ones, although the sequence being averaged never settles at all.
Functions defined by limits
Show that is when is an integer and otherwise. Deduce that equals except at the integers, where it equals .
Hint
Solution
If is an integer, and every term is . Otherwise , so and .
For the second expression: at an integer every term is . Otherwise divide top and bottom by :
So a formula built from smooth pieces can change a function’s values on the integers alone.
Series
Show that and, for , .
Hint
Hint
Solution
Put ( factors). By the second hint,
So the partial sum telescopes:
As the last fraction tends to , so the sum is . For this is .
Show that if then .
Hint
Hint
Solution
A harder one
Show that if , then and is a rational number whose denominator is a power of .
Hint
Hint
Solution
. From , the sequence , and hence , tends to . Letting in gives , so or . Letting in gives . If , then , so , but then , a contradiction. So , and gives .
is dyadic. Write with an integer and . Then , and since is close to on only near , . Choose with for all . For such , with , so (by uniqueness of the representation) . By induction for all . The left side has absolute value less than while the right side is unbounded unless . Hence , so and , a rational number with a power of as denominator.
Conversely, if then is an integer for , so from then on. Hardy adds that for other bases the limit need not be : for every , because .