4.4  The Algebra and Order of Limits

So far every limit has been found by hand: we computed the distance from an to the candidate limit and bounded it, as in Example (First Limits) and Exercise B1 of Section 4.3. This section proves the rules that make most such work unnecessary. Sums, multiples, products and quotients of convergent sequences converge to the sum, multiple, product and quotient of the limits; limits respect ≤; and a sequence trapped between two sequences with a common limit has that limit too. Everything holds in an arbitrary ordered field F. No result of this section uses the Archimedean property; only the examples and exercises set in Q do, each time through Example (First Limits) (b) or the non-example after Definition 4.3.4.

Throughout, F is an ordered field and F>0 its set of positive elements, as in Section 4.3.

4.4.1. Two tools

The first tool says that a small factor wins against a factor that merely stays bounded.

Proposition 4.4.1 (Null Times Bounded Is Null)

Let (an) be a null sequence in F, and let (bn) be a sequence with |bn|≤M for every n, where M∈F>0. Then (anbn) is null.

Proof. Let ε∈F>0. Then ε/M∈F>0 (Proposition 2.5.2 (a) and (O4)). Since (an) is null, there is N (depending on ε and M) with |an|<ε/M for every n≥N. For such n, by Proposition 4.1.3 (b), |anbn−0|=|an||bn|≤|an|M<εMM=ε, where the first inequality multiplies |bn|≤M by |an|≥0 (Proposition 4.1.2 (d)), and the second multiplies |an|<ε/M by M>0 (Proposition 2.3.5 (c)).

For instance, ((−1)n/(n+1)) is null in Q: it is 1/(n+1), null by Example (First Limits) (b), times (−1)n, bounded by 1. The bound M is fixed before ε and enters the tolerance ε/M. That is the pattern of the product law below.

The second tool is the fact behind every division by bn: if bn→b≠0, the terms eventually stay away from 0, by a margin we can name.

Lemma 4.4.2 (Terms Near a Non-zero Limit)

Let bn→b in F with b≠0. Then there is N0∈N with |bn|>|b|/2 for every n≥N0. If b>0, then moreover bn>b/2 for every n≥N0.

Proof. Since b≠0, |b|>0 (Proposition 4.1.3 (a)), so ε:=|b|/2∈F>0 (Proposition 4.1.2 (b)). Choose N0 (depending on (bn) and b) with |bn−b|<|b|/2 for every n≥N0. For such n, the reverse triangle inequality (Proposition 4.1.3 (f)), together with x≤|x| and |−x|=|x| (parts (c), (b)), gives |b|−|bn|≤||b|−|bn||≤|b−bn|=|bn−b|<|b|2, so |bn|>|b|−|b|/2=|b|/2, adding |bn|−|b|/2 to both sides (Proposition 2.3.5 (b)). If b>0, then |b|=b, and Proposition 4.1.3 (d) turns |bn−b|<b/2 into b−b/2<bn, that is, bn>b/2.

In particular bn≠0 for every n≥N0, since |bn|>|b|/2>0. The lemma needs b≠0. In Q, bn:=(1+(−1)n)/(n+1) tends to 0 (the warning after Example (First Limits) in Section 4.3: 0≤bn≤2/(n+1), and 1/(n+1)→0 by Example (First Limits) (b), which uses Archimedes), yet bn=0 at every odd n, so for b=0 no N0 has |bn|>|b|/2=0 for every n≥N0. Even 1/(n+1), which tends to 0 (Example (First Limits) (b)) and is never 0, is bounded away from 0 by no positive margin.

4.4.2. The limit laws

Theorem 4.4.3 (Limit Laws)

Let an→a and bn→b in F, and let c∈F. Then:

  • (a) an+bn→a+b;
  • (b) can→ca; in particular −an→−a and an−bn→a−b;
  • (c) anbn→ab;
  • (d) if b≠0 and bn≠0 for every n, then an/bn→a/b.

Idea. Work backwards from each target distance. For (a), |(an+bn)−(a+b)|≤|an−a|+|bn−b|: a sum of two terms, each of which we can make small, so we give each half the tolerance. For (c), write anbn−ab=an(bn−b)+b(an−a). The factors bn−b and an−a can be made small, but an cannot. It is, however, bounded (Proposition 4.3.5), so we replace |an| by a bound M and ask |bn−b| to be smaller than ε/(2M). For (d), it is enough to treat 1/bn and then use (c). Before reading on: why must M be fixed before ε?

Proof.

  • (a) Let ε∈F>0. Then ε/2∈F>0 (Proposition 4.1.2 (b)). Since an→a, choose N1 (depending on ε) with |an−a|<ε/2 for every n≥N1; since bn→b, choose N2 (depending on ε) with |bn−b|<ε/2 for every n≥N2. Let N:=max{N1,N2}. For n≥N, both estimates hold, and the triangle inequality (Proposition 4.1.3 (e)) gives |(an+bn)−(a+b)|=|(an−a)+(bn−b)|≤|an−a|+|bn−b|<ε2+ε2=ε, where the strict inequality is Proposition 2.3.5 (b) and the last equality Proposition 4.1.2 (b).

  • (b) If c=0, then can=0=ca for every n (Proposition 2.3.2 (b)), a constant sequence, which converges to 0 (Example (First Limits) (a)). Let c≠0, so |c|>0 (Proposition 4.1.3 (a)), and let ε∈F>0. Then ε/|c|∈F>0 (Proposition 2.5.2 (a) and (O4)). Choose N (depending on ε and c) with |an−a|<ε/|c| for every n≥N. For such n, by distributivity and Proposition 4.1.3 (b), |can−ca|=|c(an−a)|=|c||an−a|<|c|ε|c|=ε, multiplying by |c|>0 (Proposition 2.3.5 (c)). With c=−1, and (−1)x=−x (Proposition 2.3.2 (c)), this gives −an→−a. Then (a), applied to (an) and (−bn), gives an−bn=an+(−bn)→a+(−b)=a−b.

  • (c) By Proposition 4.3.5 there is M∈F>0 with |an|≤M for every n∈N. We fix this M now; it depends on the sequence (an) only. Now let ε∈F>0. Since |b|≥0, the element |b|+1 is ≥1>0 (Proposition 2.3.5 (b), (e)). Since 2>0 (Proposition 4.1.2 (b)), 2M>0 and 2(|b|+1)>0 by (O4), so Proposition 2.5.2 (a) and (O4) give ε2M∈F>0andε2(|b|+1)∈F>0. Since bn→b, choose N1 with |bn−b|<ε/(2M) for every n≥N1. Since an→a, choose N2 with |an−a|<ε/(2(|b|+1)) for every n≥N2. Both depend only on ε, M and b. Let N:=max{N1,N2}, and let n≥N. Using the identity of the Idea, the triangle inequality and the product rule (Proposition 4.1.3 (e), (b)), |anbn−ab|≤|an||bn−b|+|b||an−a|≤M|bn−b|+(|b|+1)|an−a|<Mε2M+(|b|+1)ε2(|b|+1)=ε2+ε2=ε. The second line uses Proposition 4.1.2 (d) twice, with 0≤|an|≤M and 0≤|b|≤|b|+1, and adds the two estimates (Proposition 2.3.5 (b)). The third multiplies the two estimates by the positive elements M and |b|+1 (Proposition 2.3.5 (c)) and adds them (Proposition 2.3.5 (b)); the last equality is Proposition 4.1.2 (b). So anbn→ab.

  • (d) First we show 1/bn→1/b. By Lemma 4.4.2 there is N0 (depending on (bn) and b) with |bn|>|b|/2 for every n≥N0. For such n, Lemma 4.1.1 (c) and Proposition 4.1.3 (b) give |1bn−1b|=|b−bnbnb|=|bn−b||bn||b|. Multiplying |bn|>|b|/2 by |b|>0 gives |bn||b|>|b|2/2>0, so 1/(|bn||b|)<2/|b|2 (Proposition 2.5.2 (b)). Multiplying by |bn−b|≥0, |1bn−1b|≤2|b|2|bn−b|(n≥N0). Now let ε∈F>0. Then ε|b|2/2∈F>0 ((O4), Proposition 4.1.2 (b)); choose N1 (depending on ε and b) with |bn−b|<ε|b|2/2 for every n≥N1, and let N:=max{N0,N1}. For n≥N, the last display and multiplication of |bn−b|<ε|b|2/2 by 2/|b|2>0 (Proposition 2.5.2 (a), (O4); Proposition 2.3.5 (c)) give |1/bn−1/b|<(2/|b|2)(ε|b|2/2)=ε. So 1/bn→1/b. Finally an/bn=an⋅(1/bn), and (c) gives an/bn→a⋅(1/b)=a/b. This proves the theorem.

The answer to the prompt: the tolerance ε/(2M) contains M, so M must exist before that tolerance is written down. And M must not depend on n: it has to bound |an| for every n≥N at once, including indices not yet chosen. That is why it comes from Proposition 4.3.5, a single bound for the whole sequence, and not from “|an| is some element of F”. The +1 in |b|+1 is there because b may be 0, and ε/(2|b|) would then mean nothing.

Part (a) is the move we call split ε: to make a sum of k small terms smaller than ε, give each term the tolerance ε/k, and combine the finitely many thresholds by a maximum. Part (c) is bound, then shrink: replace the factor you cannot control by a bound fixed in advance, then make the other factor small.

Alternatively, (c) follows from the earlier tools without new estimates. By Proposition 4.3.3, (bn−b) and (an−a) are null; an(bn−b) is null by Proposition 4.4.1, with the bound M; b(an−a) is null by (b); so their sum anbn−ab is null by (a), and Proposition 4.3.3 again gives anbn→ab. The bound M is still fixed first: it is the hypothesis of Proposition 4.4.1.

In (d), the hypothesis “bn≠0 for every n” is there only so that an/bn is defined for every n. By Lemma 4.4.2, bn≠0 holds eventually anyway once b≠0. If some early bn vanish, the quotient is a sequence from some N0 on, and Proposition 4.3.2 (b) applies the law to the shifted sequences (an+N0) and (bn+N0).

Warning. Part (d) needs b≠0, even when every bn≠0. In Q, an=1/(n+1) and bn=1/(n+1)2 are both null (the first by Example (First Limits) (b), the second by (c), since bn=anan), with every bn≠0, and an/bn=n+1. Since n+1>n, the sequence (n+1) is not bounded (the non-example after Definition 4.3.4), and so has no limit (Proposition 4.3.5). The “law” lim(an/bn)=0/0 names nothing.

Quick check. Find lim(3n2−n)/(n2+5) in Q, citing a law, or another result, for each step.

Solution. The limit is 3. Let u0:=0 and un:=1/n for n≥1. The shifted sequence (un+1)n=(1/(n+1))n is null (Example (First Limits) (b); this is where Archimedes enters), so (un) is null by Proposition 4.3.2 (b). By Theorem 4.4.3: 3−un→3−0=3 (the constant 3 converges to 3 by Example (First Limits) (a); then (b) and (a)); un2→0 ((c)); 1+5un2→1 ((b), (a)). Every 1+5un2≥1>0: un2≥0 (Proposition 2.3.5 (e) if un≠0, and u02=0), so 5un2≥0 (Proposition 4.1.2 (d)), and 1+5un2≥1 by Proposition 2.3.5 (b). Also the limit 1≠0, so (d) gives cn:=3−un1+5un2→31=3. For n≥1, multiplying numerator and denominator by n2 shows cn=(3n2−n)/(n2+5). The two sequences agree from N0=1 on (at n=0 they differ: 3 versus 0), so Proposition 4.3.2 (a) gives the limit 3.

4.4.3. Limits and order

Inequalities between terms pass to the limits, provided we accept that a strict inequality may become an equality.

Proposition 4.4.4 (Limits Preserve Weak Inequalities)

Let an→a and bn→b in F, and let N0∈N.

  • (a) If an≤bn for every n≥N0, then a≤b.
  • (b) In particular, for c∈F: if an≤c for every n≥N0, then a≤c; and if an≥c for every n≥N0, then a≥c.
  • (c) (Squeeze.) Let (cn) be a sequence in F. If a=b and an≤cn≤bn for every n≥N0, then cn→a.

Idea. For (a), suppose instead that b<a. The midpoint (a+b)/2 separates them, and from some index on an lies above it and bn below it, contradicting an≤bn. The tolerance that achieves this is half the gap, (a−b)/2. For (c), from some index on an and bn both lie in the window (a−ε,a+ε), and cn is trapped between them.

Proof.

  • (a) Suppose, for a contradiction, that b<a. Let ε:=a−b2, which lies in F>0 by Proposition 2.3.5 (a) and Proposition 4.1.2 (b). Choose N1 (depending on ε) with |an−a|<ε for every n≥N1, and N2 (depending on ε) with |bn−b|<ε for every n≥N2. Let n:=max{N0,N1,N2}. By Proposition 4.1.3 (d), at this n, an>a−ε=a+b2=b+ε>bn. This contradicts an≤bn, which holds because n≥N0. So b<a is false, and a≤b by trichotomy (O1).

  • (b) A constant sequence c converges to c (Example (First Limits) (a)). Apply (a) to (an) and the constant sequence c for the first claim, and to the constant sequence c and (an) for the second.

  • (c) Let ε∈F>0. Choose N1 (depending on ε) with |an−a|<ε for n≥N1 and N2 (depending on ε) with |bn−a|<ε for n≥N2 (recall b=a), and let N:=max{N0,N1,N2}. For n≥N, Proposition 4.1.3 (d) gives a−ε<an≤cn≤bn<a+ε, so |cn−a|<ε by the same part (d). Hence cn→a.

The two midpoint equalities in (a) are field algebra: a−(a−b)/2=(2a−a+b)/2=(a+b)/2, and likewise b+(a−b)/2=(a+b)/2. Chapter 5 uses the squeeze (c) repeatedly, mostly with F=Q to show that a sequence is null. In the squeeze (c), the middle sequence is not assumed to converge: its convergence is the conclusion.

Warning. Strict inequalities do not pass to the limit. In Q, 1/(n+1)>0 for every n, and the limit is 0, not positive (Example (First Limits) (b)). From an<bn for every n, (a) gives only a≤b.

4.4.4. Exercises

A. Check your understanding

Exercise (A1)

These questions are about the proof of the product law, Theorem 4.4.3 (c).

  • (a) Why is M fixed before ε, and what does it depend on?
  • (b) Which hypothesis supplies N1, which supplies N2, and which supplies M?
  • (c) Why does the proof use |b|+1 rather than |b|?
  • (d) State the idea of the proof in two sentences.
  • (e) Run the proof in Q on an=1+1/(n+1), bn=2 and ε=1/10: give a valid M, the least N1, the least N2, and N.

Solution.

  • (a) The tolerance ε/(2M) is built from M, so M must be available first. It depends only on the sequence (an): it bounds |an| for every n, independently of ε and of n. A “bound” depending on n would be useless, since the tolerance for |bn−b| must be named before n is.

  • (b) N1 comes from bn→b, applied with the tolerance ε/(2M); N2 from an→a, with the tolerance ε/(2(|b|+1)); M also from an→a, through Proposition 4.3.5.

  • (c) b may be 0, and then ε/(2|b|) is undefined. Since |b|+1≥1>0, the tolerance ε/(2(|b|+1)) always exists, and |b|≤|b|+1 keeps the estimate valid.

  • (d) Write anbn−ab=an(bn−b)+b(an−a), so that each term has a factor tending to 0. The other factor of the first term is not small, but it is bounded by a constant M fixed in advance, so making |bn−b|<ε/(2M) and |an−a|<ε/(2(|b|+1)) makes each term smaller than ε/2.

  • (e) Here a=1 and b=2. Since 0<1/(n+1)≤1, |an|≤2 for every n, so M=2 is valid. Then ε/(2M)=1/40, and |bn−b|=0<1/40 for every n: the least N1 is 0. Next ε/(2(|b|+1))=(1/10)/6=1/60, and |an−a|=1/(n+1)<1/60 if and only if n+1>60, that is, n≥60: the least N2 is 60. So N=max{0,60}=60. Check at n=60: a60b60−ab=2(1+1/61)−2=2/61<1/10.

B. Practice

Exercise (B1)

Find the following limits in Q by the limit laws, naming the result used at each step.

  • (a) n+23n+1.
  • (b) n2+1(n+1)2.
  • (c) (1n+1−2)(3+(−1)nn+1).

Hint: write each term through un:=1/(n+1).

Solution. Let un:=1/(n+1), so un→0 (Example (First Limits) (b)), and constant sequences converge to their value (part (a) of that example). All laws cited are parts of Theorem 4.4.3.

  • (a) With m:=n+1≥1: (n+2)/(3n+1)=(m+1)/(3m−2)=(1+un)/(3−2un), dividing numerator and denominator by m. Now 1+un→1 by (a), and 3−2un→3 by (b) and (a). Since 0<un≤1, 3−2un≥1>0 for every n, and the limit 3≠0, so (d) gives the limit 1/3.

  • (b) With m=n+1: (n2+1)/(n+1)2=((m−1)2+1)/m2=1−2un+2un2. By (c), un2→0; by (b) and (a), 1−2un+2un2→1−0+0=1. No quotient law is needed.

  • (c) un−2→−2 by (b) and (a). The sequence (−1)nun is null by Proposition 4.4.1, since un is null and |(−1)n|=1 (Example (First Limits) (d)); so 3+(−1)nun→3 by (a). By (c), the product tends to (−2)⋅3=−6.

In each part Archimedes enters once, through un→0; every other step is a law valid in any ordered field.

Exercise (B2: Put the proof in order)

Here are lines of the proof of Proposition 4.4.4 (a), shuffled, together with two lines that do not belong. Recall the hypotheses: an→a, bn→b, and an≤bn for every n≥N0. Write the seven lines of a correct proof in a correct order, say which other orders are also admissible, and say why each of the other two lines fails.

  • (i) Let n:=max{N0,N1,N2}.
  • (ii) Let ε:=a−b, which lies in F>0.
  • (iii) Choose N2 with |bn−b|<ε for every n≥N2.
  • (iv) Suppose, for a contradiction, that b<a.
  • (v) This contradicts an≤bn; so b<a is false, and a≤b by trichotomy.
  • (vi) Let ε:=(a−b)/2, which lies in F>0 by Proposition 4.1.2.
  • (vii) Let n:=N1.
  • (viii) Then an>a−ε=(a+b)/2=b+ε>bn.
  • (ix) Choose N1 with |an−a|<ε for every n≥N1.

Solution. An admissible order is (iv), (vi), (ix), (iii), (i), (viii), (v). The only other admissible order exchanges (ix) and (iii), which use only ε and one hypothesis each. Nothing else moves: (vi) uses the assumption b<a of (iv) to make ε positive; (ix) and (iii) use ε; (i) uses N1 and N2; (viii) uses the two estimates at this n, and needs n≥N1 and n≥N2; (v) uses (viii) and n≥N0, and closes the assumption of (iv).

Line (ii) is an ε of the wrong size. It is positive, but with it (viii) becomes an>a−(a−b)=b and bn<b+(a−b)=a, and “an>b and bn<a” is no contradiction with an≤bn: for instance an=bn=(a+b)/2 satisfies all three at once. The two windows of half-width a−b around a and b overlap; half-width (a−b)/2 makes them touch only at the midpoint, which neither open window contains.

Line (vii) chooses an index that depends on the wrong thing. With n:=N1, the estimate for an holds, but nothing makes n≥N2 or n≥N0: the estimate for bn and the hypothesis an≤bn may both fail at this n. The index must lie beyond all three thresholds, which the maximum in (i) ensures.

Exercise (B3: Close to a convergent sequence)

Let an→a in F, and let (bn) be a sequence in F such that (bn−an) is null. Prove that bn→a. Here is the start of a proof.

Let ε∈F>0. Then ε/2∈F>0. Since an→a, choose N1 with |an−a|<ε/2 for every n≥N1. Since (bn−an) is null, choose N2 with |bn−an|<ε/2 for every n≥N2. Let N:=max{N1,N2}.

…

Name the move, then complete it.

Solution. The move is split ε, with two terms: each gets ε/2, and the two thresholds are combined by a maximum.

Completion. Let n≥N. Then n≥N1 and n≥N2, and the three-point triangle inequality (Proposition 4.1.3 (e)) gives |bn−a|≤|bn−an|+|an−a|<ε2+ε2=ε. Since ε∈F>0 was arbitrary, bn→a. (The same result also follows from the laws: bn=(bn−an)+an→0+a by Theorem 4.4.3 (a). The exercise is the law’s proof, run once more.)

C. Going deeper

Exercise (C1: Exit check: the product law)

Prove, from Definition 4.3.2, that if an→a and bn→b in an ordered field F, then anbn→ab. You may use that convergent sequences are bounded (Proposition 4.3.5), the rules for the absolute value (Proposition 4.1.3), Proposition 4.3.3 and Proposition 4.4.1, but not Theorem 4.4.3.

Solution. By Proposition 4.3.5 there is M∈F>0 with |an|≤M for every n∈N. It depends only on (an), and we fix it before any tolerance is named.

Let ε∈F>0. Since |b|+1≥1>0, both ε/(2M) and ε/(2(|b|+1)) lie in F>0. By bn→b there is N1 with |bn−b|<ε/(2M) for every n≥N1; by an→a there is N2 with |an−a|<ε/(2(|b|+1)) for every n≥N2. These depend only on ε, M and b. Let N:=max{N1,N2}.

Let n≥N. Since anbn−ab=an(bn−b)+b(an−a), the triangle inequality and |xy|=|x||y| give |anbn−ab|≤|an||bn−b|+|b||an−a|≤M|bn−b|+(|b|+1)|an−a|<ε2+ε2=ε, where the second line uses 0≤|an|≤M and 0≤|b|≤|b|+1 (Proposition 4.1.2 (d)), and the third multiplies the two estimates by M>0 and |b|+1>0. So for every ε∈F>0 there is N∈N with |anbn−ab|<ε for every n≥N: anbn→ab.

Marking scheme. (10 points) - 2: fixes a bound M∈F>0 for (|an|), justified (convergent sequences are bounded), before ε. - 2: writes anbn−ab as a sum of two terms, each with a factor tending to 0. - 3: chooses N1, N2 depending only on ε, M and b, with the case b=0 handled, and N as their maximum. - 2: carries out the estimate with the triangle inequality and the product rule for |⋅|. - 1: concludes with the quantifiers in the order ∀ε∃N∀n≥N. Fatal errors (0 for the step): a threshold or tolerance that depends on n; the bound “|an|≤|a|+1 for every n” without the finite part for n<N; dividing by |b| without b≠0. A route through null sequences ((bn−b) and (an−a) are null by Proposition 4.3.3; an(bn−b) and b(an−a) are null by Proposition 4.4.1, with bounds M and |b|+1; the sum of two null sequences is null, proved from Definition 4.3.2 by splitting ε) earns the same points: decomposition 2, the bound 2, the two null products and the proved sum 5, conclusion 1. Citing Theorem 4.4.3 for the sum or for b(an−a) gets 0 for that step. A three-term route (anbn−ab=(an−a)(bn−b)+a(bn−b)+b(an−a), with a tolerance such as δ=min{1,ε/(1+|a|+|b|)} chosen from ε, a and b only) earns the same points: decomposition 2; the choice of δ, independent of n and with δ≤1 controlling the product term (in place of the bound M), 2; a threshold N for both sequences at tolerance δ (the case a=0 or b=0 covered by the 1 in the denominator) 3; the estimate 2; conclusion 1.

Exercise (C2: Non-negative terms and reciprocals)

Let an→a in F, with an≥0 for every n.

  • (a) Prove that a≥0.
  • (b) Suppose moreover a>0. Prove that there is N0∈N with an>0 for every n≥N0, and that the sequence (1/an+N0)n∈N converges to 1/a.
  • (c) Give an example in Q where an>0 for every n and (1/an) does not converge.

Solution (Hint 1). For (a), compare with a constant sequence. For (b), which result keeps terms away from a non-zero limit?

Solution (Hint 2). For (b), start: “By Lemma 4.4.2, since a>0, there is N0 with an>a/2 for every n≥N0.”

Solution.

  • (a) Since an≥0 for every n≥0, Proposition 4.4.4 (b) with c=0 gives a≥0.

  • (b) Since a>0, Lemma 4.4.2 gives N0 with an>a/2>0 for every n≥N0. The shifted sequence (an+N0)n converges to a by Proposition 4.3.2 (b), and each of its terms is non-zero. By Theorem 4.4.3 (d), applied to the constant sequence 1 and to (an+N0)n, whose limit a is non-zero, 1/an+N0→1/a.

  • (c) an=1/(n+1): every term is positive, and 1/an=n+1 is not bounded in Q: for any M∈Q>0, the non-example after Definition 4.3.4 gives n with M<n<n+1=|1/an|. So it does not converge (Proposition 4.3.5). The hypothesis a>0 of (b) fails here: a=0.

Marking scheme. (10 points) - 3: (a), by comparison with the constant sequence 0 (or by contradiction with ε=−a/2). - 3: (b), an index N0 beyond which an is bounded away from 0, from a>0. - 2: (b), 1/an+N0→1/a, either by the quotient law (Theorem 4.4.3 (d)) applied to the shift, with its convergence to a and a≠0 checked, or directly from |1/an+N0−1/a|≤(2/a2)|an+N0−a|, which uses an+N0>a/2. - 2: (c), an example with the failure of convergence justified. Fatal errors (0 for the step): concluding a>0 in (a) from an>0; applying the quotient law without checking that the limit is non-zero.