← 2. Sequences

Exercises

Problems to check your understanding of this chapter. The solutions are folded away — try them yourself first.

Easy

1Easy

Let an=n+sin⁡nn2+1a_n = \frac{n + \sin n}{n^2 + 1}. Prove directly from the ε\varepsilon-NN definition that lim⁡n→∞an=0\lim_{n \to \infty} a_n = 0, and give an explicit formula for a usable N(ε)N(\varepsilon) in terms of ε>0\varepsilon > 0.

∣n+sin⁡nn2+1∣≤2n\left|\frac{n + \sin n}{n^2 + 1}\right| \le \frac{2}{n}
N(ε)=⌈2ε⌉N(\varepsilon) = \left\lceil \frac{2}{\varepsilon} \right\rceil
ε=0.1\varepsilon = 0.1: the moment the tail fits entirely inside the ε\varepsilon-band is exactly the NN we found.
Show solution
  1. First trap the oscillating term in the numerator. Since ∣sin⁡t∣≤1|\sin t| \le 1 for every real tt, ∣n+sin⁡n∣≤n+1|n + \sin n| \le n + 1. For n≥1n \ge 1 we have n+1≤2nn + 1 \le 2n and the denominator satisfies n2+1≥n2n^2 + 1 \ge n^2, so
    ∣an−0∣=∣n+sin⁡n∣n2+1≤2nn2=2n.|a_n - 0| = \frac{|n + \sin n|}{n^2 + 1} \le \frac{2n}{n^2} = \frac{2}{n}.
  2. Given ε>0\varepsilon > 0, put N(ε)=⌈2/ε⌉N(\varepsilon) = \lceil 2/\varepsilon \rceil. If n>Nn > N then n>2/εn > 2/\varepsilon, i.e. 2/n<ε2/n < \varepsilon. Combined with the estimate above, ∣an−0∣<ε|a_n - 0| < \varepsilon, so an→0a_n \to 0 by the ε\varepsilon-NN definition.
2Easy

For n≥1n \ge 1 let an=⌊n2⌋na_n = \frac{\lfloor n\sqrt{2} \rfloor}{n}, where ⌊x⌋\lfloor x \rfloor is the greatest integer ≤x\le x. (a) Show each ana_n is rational. (b) Prove an→2a_n \to \sqrt{2} from the ε\varepsilon-NN definition. (c) Prove 2\sqrt{2} is irrational. Conclude that a sequence of rationals can converge to an irrational limit.

0≤n2−⌊n2⌋<10 \le n\sqrt{2} - \lfloor n\sqrt{2} \rfloor < 1
∣an−2∣<1n\left|a_n - \sqrt{2}\right| < \frac{1}{n}
Show solution
  1. (a) ⌊n2⌋\lfloor n\sqrt{2} \rfloor is an integer, so ana_n is an integer divided by nn: rational. (b) From the definition of the floor function, 0≤n2−⌊n2⌋<10 \le n\sqrt{2} - \lfloor n\sqrt{2} \rfloor < 1. Dividing by nn gives ∣an−2∣<1/n|a_n - \sqrt{2}| < 1/n. For ε>0\varepsilon > 0 put N=⌈1/ε⌉N = \lceil 1/\varepsilon \rceil; if n>Nn > N then ∣an−2∣<1/n<ε|a_n - \sqrt{2}| < 1/n < \varepsilon, so an→2a_n \to \sqrt{2}.
  2. (c) Suppose 2=p/q\sqrt{2} = p/q with coprime positive integers p,qp, q. Then p2=2q2p^2 = 2q^2, so p2p^2 is even, hence pp is even (an even square has an even root): p=2rp = 2r. Substituting gives 4r2=2q24r^2 = 2q^2, i.e. q2=2r2q^2 = 2r^2, so by the same argument qq is even. This contradicts coprimality. Hence 2\sqrt{2} is irrational.
  3. The rational sequence (an)(a_n) converges to the irrational 2\sqrt{2}, so the expectation that a limit of rationals stays rational is false. A sequence limit can "escape" the set its terms live in.
3Easy

Let an=(2+(−1)n)(1−1n+1)a_n = (2 + (-1)^n)\left(1 - \frac{1}{n+1}\right). Find lim sup⁡n→∞an\limsup_{n \to \infty} a_n and lim inf⁡n→∞an\liminf_{n \to \infty} a_n, with a rigorous proof using subsequence arguments.

a2k=3(1−12k+1),a2k+1=1−12k+2a_{2k} = 3\left(1 - \frac{1}{2k+1}\right), \qquad a_{2k+1} = 1 - \frac{1}{2k+2}
Show solution
  1. Separate even and odd terms. For n=2kn = 2k, (−1)n=1(-1)^n = 1, so a2k=3(1−1/(2k+1))→3a_{2k} = 3(1 - 1/(2k+1)) \to 3. For n=2k+1n = 2k+1, (−1)n=−1(-1)^n = -1, so a2k+1=1−1/(2k+2)→1a_{2k+1} = 1 - 1/(2k+2) \to 1. Hence lim sup⁡an≥3\limsup a_n \ge 3 and lim inf⁡an≤1\liminf a_n \le 1.
  2. For the reverse inequalities, trap every term. Since 1≤2+(−1)n≤31 \le 2 + (-1)^n \le 3 and 0<1−1/(n+1)<10 < 1 - 1/(n+1) < 1, we have an<3a_n < 3 for all nn. Hence lim sup⁡an≤3\limsup a_n \le 3, and with the even subsequence, lim sup⁡an=3\limsup a_n = 3. Likewise an≥1⋅(1−1/(n+1))a_n \ge 1 \cdot (1 - 1/(n+1)) and 1−1/(n+1)→11 - 1/(n+1) \to 1, so lim inf⁡an≥1\liminf a_n \ge 1. With the odd subsequence, lim inf⁡an=1\liminf a_n = 1.

Medium

4Medium

Let an>0a_n > 0 with an→L>0a_n \to L > 0. Prove the geometric mean Gn=(a1a2⋯an)1/nG_n = (a_1 a_2 \cdots a_n)^{1/n} converges to LL. (Hint: taking logarithms turns this into a Cesaro mean.)

log⁡Gn=log⁡a1+⋯+log⁡ann\log G_n = \frac{\log a_1 + \cdots + \log a_n}{n}
Gn=exp⁡(1n∑k=1nlog⁡ak)→exp⁡(log⁡L)=LG_n = \exp\left(\frac{1}{n}\sum_{k=1}^{n} \log a_k\right) \to \exp(\log L) = L
Show solution
  1. Put bn=log⁡anb_n = \log a_n. Since log⁡\log is continuous at L>0L > 0, bn→log⁡Lb_n \to \log L. By Cesaro's theorem (if a sequence converges, so do its arithmetic means, to the same limit),
    b1+⋯+bnn⟶log⁡L(n→∞).\frac{b_1 + \cdots + b_n}{n} \longrightarrow \log L \qquad (n \to \infty).
  2. But log⁡Gn=1n∑k=1nlog⁡ak\log G_n = \frac{1}{n}\sum_{k=1}^n \log a_k, so by continuity of the exponential, Gn=exp⁡(log⁡Gn)→exp⁡(log⁡L)=LG_n = \exp(\log G_n) \to \exp(\log L) = L. The hypothesis L>0L > 0 is essential for taking logarithms.

Hard

5Hard

For a sequence with an>0a_n > 0, prove lim inf⁡n→∞an+1an≤lim inf⁡n→∞an1/n≤lim sup⁡n→∞an1/n≤lim sup⁡n→∞an+1an\liminf_{n \to \infty} \frac{a_{n+1}}{a_n} \le \liminf_{n \to \infty} a_n^{1/n} \le \limsup_{n \to \infty} a_n^{1/n} \le \limsup_{n \to \infty} \frac{a_{n+1}}{a_n}. Additionally, if an+1an→L>0\frac{a_{n+1}}{a_n} \to L > 0, take logarithms and use the Stolz-Cesaro theorem to show an1/n→La_n^{1/n} \to L.

lim inf⁡an+1an≤lim inf⁡an1/n≤lim sup⁡an1/n≤lim sup⁡an+1an\liminf \frac{a_{n+1}}{a_n} \le \liminf a_n^{1/n} \le \limsup a_n^{1/n} \le \limsup \frac{a_{n+1}}{a_n}
an+1an→L>0⟹an1/n→L\frac{a_{n+1}}{a_n} \to L > 0 \quad \Longrightarrow \quad a_n^{1/n} \to L
Show solution
  1. The middle inequality lim inf⁡≤lim sup⁡\liminf \le \limsup is trivial. We prove the last one. Let R=lim sup⁡an+1/anR = \limsup a_{n+1}/a_n; if R=∞R = \infty there is nothing to prove, so assume R<∞R < \infty. Take β\beta with R<β<∞R < \beta < \infty. By definition, an+1/an<βa_{n+1}/a_n < \beta for n≥Nn \ge N. Hence aN+k<aNβka_{N+k} < a_N \beta^k for k≥1k \ge 1, and taking (N+k)(N+k)-th roots,
    aN+k1/(N+k)<aN1/(N+k)⋅βk/(N+k)⟶1⋅β=β.a_{N+k}^{1/(N+k)} < a_N^{1/(N+k)} \cdot \beta^{k/(N+k)} \longrightarrow 1 \cdot \beta = \beta.
  2. (aNa_N is fixed and the exponent tends to 00, so aN1/(N+k)→1a_N^{1/(N+k)} \to 1.) Hence lim sup⁡an1/n≤β\limsup a_n^{1/n} \le \beta; letting β↓R\beta \downarrow R gives lim sup⁡an1/n≤R\limsup a_n^{1/n} \le R. The first inequality is symmetric: let r=lim inf⁡an+1/anr = \liminf a_{n+1}/a_n. If r=0r = 0 there is nothing to prove; assume r>0r > 0. Take 0<α<r0 < \alpha < r; then an+1/an>αa_{n+1}/a_n > \alpha for n≥Nn \ge N, and from aN+k>aNαka_{N+k} > a_N \alpha^k we get aN+k1/(N+k)>aN1/(N+k)αk/(N+k)→αa_{N+k}^{1/(N+k)} > a_N^{1/(N+k)} \alpha^{k/(N+k)} \to \alpha. So lim inf⁡an1/n≥α\liminf a_n^{1/n} \ge \alpha, and α↑r\alpha \uparrow r gives lim inf⁡an1/n≥r\liminf a_n^{1/n} \ge r.
  3. Finally the limit form: if an+1an→L>0\frac{a_{n+1}}{a_n} \to L > 0 then log⁡an+1−log⁡an=log⁡an+1an→log⁡L\log a_{n+1} - \log a_n = \log\frac{a_{n+1}}{a_n} \to \log L. Applying Stolz-Cesaro to numerator log⁡an\log a_n and denominator nn gives log⁡ann→log⁡L\frac{\log a_n}{n} \to \log L, and exponentiating, an1/n=exp⁡(log⁡ann)→exp⁡(log⁡L)=La_n^{1/n} = \exp(\frac{\log a_n}{n}) \to \exp(\log L) = L. This is why the ratio test implies the root test.