Skip to content

§7 Limits of Sequences

Why this matters

Derivatives, integrals, infinite sums and continuity are all defined through limits, so this one definition carries the weight of the whole subject. The goal of this section is to replace the vague “sns_n gets close to ss” with a statement precise enough to prove things about.

A sequence is a function whose domain is a set of integers of the form {n∈Z:n≥m}\{n \in \mathbb{Z} : n \ge m\}, where mm is usually 11 or 00. Instead of s(n)s(n) we write sns_n, and we write the whole sequence as (sn)n=m∞(s_n)_{n=m}^\infty, as (sm,sm+1,sm+2,… )(s_m, s_{m+1}, s_{m+2}, \dots), or just (sn)(s_n). In this chapter every sns_n is a real number.

Example 1.
  1. sn=1n2s_n = \dfrac{1}{n^2} gives (1,14,19,116,… )\left(1, \tfrac14, \tfrac19, \tfrac1{16}, \dots\right).
  2. an=(−1)na_n = (-1)^n for n≥0n \ge 0 gives (1,−1,1,−1,… )(1, -1, 1, -1, \dots). Its set of values is just {−1,1}\{-1, 1\}, but the sequence itself has infinitely many terms.
  3. cos⁡ ⁣(nπ3)\cos\!\left(\tfrac{n\pi}{3}\right) cycles through 12,−12,−1,−12,12,1,…\tfrac12, -\tfrac12, -1, -\tfrac12, \tfrac12, 1, \dots
  4. an=n1/na_n = n^{1/n} gives (1,1.4142,1.4422,1.4142,1.3797,… )(1, 1.4142, 1.4422, 1.4142, 1.3797, \dots), with a100≈1.0471a_{100} \approx 1.0471 and a1000≈1.0069a_{1000} \approx 1.0069.
  5. bn=(1+1n)nb_n = \left(1 + \tfrac1n\right)^n gives (2,2.25,2.3704,2.4414,… )(2, 2.25, 2.3704, 2.4414, \dots), with b1000≈2.7169b_{1000} \approx 2.7169.

Keep the notation straight: parentheses ( ⋅ )(\,\cdot\,) denote a sequence and braces { ⋅ }\{\,\cdot\,\} denote a set. Many results depend on which one you are working with.

Informally, ss is the limit of (sn)(s_n) if sns_n is close to ss for all large nn, not just for some of them. That is why 11 is not a limit of (−1)n(-1)^n: the even terms equal 11, but the odd terms keep jumping back to −1-1.

7.1 Definition.

A sequence (sn)(s_n) of real numbers converges to the real number ss if

for each ε>0 there exists a number N such thatn>Nimplies∣sn−s∣<ε.(1)\begin{gathered} \text{for each } \varepsilon > 0 \text{ there exists a number } N \text{ such that} \\ n > N \quad\text{implies}\quad |s_n - s| < \varepsilon. \end{gathered} \tag{1}

We then write lim⁡n→∞sn=s\lim_{n\to\infty} s_n = s or sn→ss_n \to s, and call ss the limit of (sn)(s_n). A sequence that does not converge to any real number diverges.

Used in §8, §9, Theorem 9.1, §10

Some things to notice:

  • By the Archimedean property, we may always take NN to be a positive integer.
  • ε\varepsilon is an ordinary positive real number. The interesting cases are the small ones.
  • Condition (1) is infinitely many statements, one for each ε>0\varepsilon > 0. NN is allowed to depend on ε\varepsilon, and usually it must get larger as ε\varepsilon gets smaller.

Intuition

Read ε\varepsilon as a tolerance and NN as how long you must wait. Convergence says: name any tolerance, however strict, and from some point on every term stays within that tolerance of ss, forever. Equivalently, every interval (s−ε,s+ε)(s - \varepsilon, s + \varepsilon) contains all but finitely many terms. One useful consequence: changing, inserting or deleting finitely many terms never affects whether a sequence converges, or what its limit is. You only need to take NN larger.

Common pitfall“closer and closer” is not the definition

The terms of 1n\frac1n get closer to −1-1 at every step, yet 1n→0\frac1n \to 0, not −1-1. Shrinking distance is not enough: the distance has to drop below every ε>0\varepsilon > 0. Nor does the approach have to be steady. (−1)nn\frac{(-1)^n}{n} hops from one side of 00 to the other and still converges to 00. Finally, NN may depend on ε\varepsilon but never on nn. “Take N=nN = n” is not a valid choice.

The terms of sn=(−1)n/ns_n = (-1)^n / n jump between positive and negative values forever. Does the sequence converge?

Think of (1) as a game. Your opponent picks a band width ε\varepsilon. You must answer with an NN so that every point after NN lies inside the band (s−ε, s+ε)(s - \varepsilon,\, s + \varepsilon). The sequence converges to ss exactly when you can win no matter which ε\varepsilon is chosen.

n ≤ N (ignored)n > N, inside the bandn > N, outside the band

✗ N = 2 fails: n = 3 gives |sn − 3/7| = 0.160 ≥ ε. The smallest N that works is 4.

Things to try:

  • For sn=3n+17n−4s_n = \frac{3n+1}{7n-4}, make ε\varepsilon smaller and watch the smallest working NN grow, roughly like 1949ε\frac{19}{49\varepsilon}.
  • For (−1)n(-1)^n with candidate limit 11, no NN ever works once ε≤2\varepsilon \le 2: there is always a later odd nn outside the band.
  • Compare how fast n1/n→1n^{1/n} \to 1 and (1+1/n)n→e(1 + 1/n)^n \to e by looking at how large NN must be for the same ε\varepsilon.

The widget above lets you drag ε yourself. Here the roles are fixed, as in the definition: the page is an opponent that announces an ε\varepsilon, you commit to an NN, and then it searches the whole tail n>Nn > N for a point outside the band. Beat five ε in a row, down to 0.0020.002. (The same game for functions, with δ\delta in place of NN, is in §17.)

Then switch Who moves first? and play the swapped order, where you must name NN before seeing ε\varepsilon. The opponent now wins on every sequence except the constant one. This is exactly the difference between

∀ε>0 ∃N ∀n>N: ∣sn−s∣<εand∃N ∀ε>0 ∀n>N: ∣sn−s∣<ε.\forall \varepsilon > 0\ \exists N\ \forall n > N:\ |s_n - s| < \varepsilon \qquad\text{and}\qquad \exists N\ \forall \varepsilon > 0\ \forall n > N:\ |s_n - s| < \varepsilon.

The second statement says ∣sn−s∣<ε|s_n - s| < \varepsilon for every ε\varepsilon, i.e. sn=ss_n = s for all n>Nn > N. Also try the “spiky” claim: five rounds can pass while the claim is false, because the opponent only needs a single far-out spike.

Round 1 of 5. The opponent announces ε = 0.500. Choose N so every n > N has |sn − 3/7| < ε.

n ≤ N (ignored)n > N, inside the bandn > N, outside the band
Example 2.

Let sn=3n+17n−4s_n = \dfrac{3n+1}{7n-4}. Writing sn=3+1/n7−4/ns_n = \dfrac{3 + 1/n}{7 - 4/n} suggests that lim⁡sn=37\lim s_n = \tfrac37. By Definition 7.1, this claim means: for each ε>0\varepsilon > 0 there is an NN such that

n>N  ⟹  ∣3n+17n−4−37∣<ε.n > N \implies \left| \frac{3n+1}{7n-4} - \frac37 \right| < \varepsilon.

In §8 we will show that N=1949ε+47N = \dfrac{19}{49\varepsilon} + \dfrac47 works. For ε=1,0.1,0.01,0.001\varepsilon = 1, 0.1, 0.01, 0.001 that gives N≈0.96, 4.45, 39.35, 388.33N \approx 0.96,\ 4.45,\ 39.35,\ 388.33.

nsn|sn − 3/7|
20.70000.2714
30.58820.1597
40.54170.1131
50.51610.0876
60.50000.0714
400.43840.0098
4000.42950.0010

A table of numbers can only check finitely many of these statements, so we will need proofs.

Example 3.

Back to Example 1:

  1. lim⁡1n2=0\lim \frac{1}{n^2} = 0 (proved in §8, Example 1).
  2. (−1)n(-1)^n diverges, so “lim⁡an\lim a_n” is meaningless here (§8, Example 4).
  3. cos⁡ ⁣(nπ3)\cos\!\left(\frac{n\pi}{3}\right) diverges (Exercise 8.7).
  4. lim⁡n1/n=1\lim n^{1/n} = 1 (Theorem 9.7(c)).
  5. lim⁡(1+1n)n=e≈2.7182818\lim \left(1 + \frac1n\right)^n = e \approx 2.7182818 (see §16 and §37).
Theorem(Uniqueness of limits).

If lim⁡sn=s\lim s_n = s and lim⁡sn=t\lim s_n = t, then s=ts = t.

Proof

Let ε>0\varepsilon > 0. Since sn→ss_n \to s, there is N1N_1 such that n>N1n > N_1 implies ∣sn−s∣<ε2|s_n - s| < \frac{\varepsilon}{2}. Since sn→ts_n \to t, there is N2N_2 such that n>N2n > N_2 implies ∣sn−t∣<ε2|s_n - t| < \frac{\varepsilon}{2}. Take any n>max⁡{N1,N2}n > \max\{N_1, N_2\}. By the Triangle Inequality,

∣s−t∣=∣(s−sn)+(sn−t)∣≤∣s−sn∣+∣sn−t∣<ε2+ε2=ε.|s - t| = |(s - s_n) + (s_n - t)| \le |s - s_n| + |s_n - t| < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \varepsilon.

So ∣s−t∣<ε|s - t| < \varepsilon for every ε>0\varepsilon > 0, which forces ∣s−t∣=0|s - t| = 0, i.e. s=ts = t.

∎
Lean 4 proof machine-checked
theorem lim_unique {s : ℕ → ℝ} {a b : ℝ} (ha : Lim s a) (hb : Lim s b) : a = b := by
by_contra hne
have hpos : 0 < |a - b| := abs_pos.mpr (sub_ne_zero.mpr hne)
obtain ⟨N₁, h₁⟩ := ha (|a - b| / 2) (by linarith)
obtain ⟨N₂, h₂⟩ := hb (|a - b| / 2) (by linarith)
have h1 := h₁ (max N₁ N₂ + 1) (by omega)
have h2 := h₂ (max N₁ N₂ + 1) (by omega)
have key : |a - b| ≤ |s (max N₁ N₂ + 1) - a| + |s (max N₁ N₂ + 1) - b| := by
calc |a - b| = |(s (max N₁ N₂ + 1) - b) - (s (max N₁ N₂ + 1) - a)| := by ring_nf
_ ≤ |s (max N₁ N₂ + 1) - b| + |s (max N₁ N₂ + 1) - a| := abs_sub _ _
_ = _ := add_comm _ _
linarith

Checked against Mathlib in CI. Definitions: lean/Ross/Defs.lean.

Reading alongside Abbott

Abbott’s Section 2.2 uses the same definition. It also restates it in terms of neighbourhoods: (sn)→s(s_n) \to s when every interval centred at ss contains all but finitely many terms. That version is worth keeping in mind; it is the one that generalizes to metric spaces (§13).

The scratch work is done; the formal proof runs forwards. The gaps are the algebra, the choice of NN, and the inequality that connects them.

Fill in the gaps: lim (3n+1)/(n+2) = 3

This is the skeleton of the proof. In each gap choose the option that makes the argument work, then check.

  1. Let ε>0\varepsilon > 0 and let sn=3n+1n+2s_n = \frac{3n+1}{n+2}. First, ∣sn−3∣=∣3n+1−3(n+2)n+2∣=|s_n - 3| = \left|\frac{3n + 1 - 3(n+2)}{n+2}\right| = .
  2. Choose N=N = . This depends only on ε\varepsilon, not on nn.
  3. Now let n>Nn > N. Then ∣sn−3∣=5n+2|s_n - 3| = \frac{5}{n+2} <ε< \varepsilon, since n>5εn > \frac5\varepsilon.
  4. So ∣sn−3∣<ε|s_n - 3| < \varepsilon for every n>Nn > N, which is the definition of lim⁡sn=3\lim s_n = 3.

    Take the quiz (3 questions)

    When does (sn)(s_n) fail to converge to ss?

    In Definition 7.1, what does NN depend on?

    Which sequence converges?

    Exercise 7.1

    Write out the first five terms of the following sequences.

    1. sn=13n+1s_n = \dfrac{1}{3n+1}
    2. bn=3n+14n−1b_n = \dfrac{3n+1}{4n-1}
    3. cn=n3nc_n = \dfrac{n}{3^n}
    4. sin⁡ ⁣(nπ4)\sin\!\left(\dfrac{n\pi}{4}\right)
    Solution
    1. 14,17,110,113,116\tfrac14, \tfrac17, \tfrac1{10}, \tfrac1{13}, \tfrac1{16}
    2. 43,1,1011,1315,1619\tfrac43, 1, \tfrac{10}{11}, \tfrac{13}{15}, \tfrac{16}{19}
    3. 13,29,19,481,5243\tfrac13, \tfrac29, \tfrac19, \tfrac4{81}, \tfrac5{243}
    4. 22,1,22,0,−22\tfrac{\sqrt2}{2}, 1, \tfrac{\sqrt2}{2}, 0, -\tfrac{\sqrt2}{2}

    Tags: Warm-upComputation

    Used in Exercise 7.2

    Exercise 7.2

    For each sequence in Exercise 7.1, determine whether it converges. If it converges, give its limit. No proofs are required.

    Solution
    1. converges to 00
    2. converges to 34\tfrac34
    3. converges to 00
    4. diverges

    For (d), the sequence cycles through 22,1,22,0,−22,−1,−22,0,…\tfrac{\sqrt2}{2}, 1, \tfrac{\sqrt2}{2}, 0, -\tfrac{\sqrt2}{2}, -1, -\tfrac{\sqrt2}{2}, 0, \dots

    Tags: Warm-upComputation

    Exercise 7.3

    For each sequence below, determine whether it converges and, if it converges, give its limit. No proofs are required.

    1. an=nn+1a_n = \dfrac{n}{n+1}
    2. bn=n2+3n2−3b_n = \dfrac{n^2+3}{n^2-3}
    3. cn=2−nc_n = 2^{-n}
    4. tn=1+2nt_n = 1 + \dfrac2n
    5. xn=73+(−1)nx_n = 73 + (-1)^n
    6. sn=21/ns_n = 2^{1/n}
    7. yn=n!y_n = n!
    8. dn=(−1)nnd_n = (-1)^n n
    9. (−1)nn\dfrac{(-1)^n}{n}
    10. 7n3+8n2n3−3\dfrac{7n^3+8n}{2n^3-3}
    11. 9n2−186n+18\dfrac{9n^2-18}{6n+18}
    12. sin⁡nπ2\sin\dfrac{n\pi}{2}
    13. sin⁡(nπ)\sin(n\pi)
    14. sin⁡2nπ3\sin\dfrac{2n\pi}{3}
    15. 1nsin⁡n\dfrac1n \sin n
    16. 2n+1+52n−7\dfrac{2^{n+1}+5}{2^n-7}
    17. 3nn!\dfrac{3^n}{n!}
    18. (1+1n)2\left(1+\dfrac1n\right)^2
    19. 4n2+33n2−2\dfrac{4n^2+3}{3n^2-2}
    20. 6n+49n2+7\dfrac{6n+4}{9n^2+7}
    Hint

    For rational functions of nn, divide the numerator and denominator by the highest power of nn in the denominator. For (q), compare consecutive terms: 3n+1/(n+1)!3n/n!=3n+1\frac{3^{n+1}/(n+1)!}{3^n/n!} = \frac{3}{n+1}.

    Solution

    Converge: (a) 11, (b) 11, (c) 00, (d) 11, (f) 11, (i) 00, (j) 72\tfrac72, (m) 00, (o) 00, (p) 22, (q) 00, (r) 11, (s) 43\tfrac43, (t) 00.

    Diverge: (e), (g), (h), (k), (l), (n).

    Tags: Warm-upComputation

    Exercise 7.4

    Give examples of

    1. a sequence (xn)(x_n) of irrational numbers whose limit is a rational number;
    2. a sequence (rn)(r_n) of rational numbers whose limit is an irrational number.
    Solution

    (a) xn=2n→0x_n = \frac{\sqrt2}{n} \to 0.

    (b) Let rnr_n be the decimal expansion of 2\sqrt2 cut off after nn digits: 1,1.4,1.41,1.414,⋯→21, 1.4, 1.41, 1.414, \dots \to \sqrt2.

    Tags: Warm-upExamples & counterexamples

    Exercise 7.5

    Determine the following limits. No proofs are required, but show any relevant algebra.

    1. lim⁡sn\lim s_n where sn=n2+1−ns_n = \sqrt{n^2+1} - n
    2. lim⁡(n2+n−n)\lim\left(\sqrt{n^2+n} - n\right)
    3. lim⁡(4n2+n−2n)\lim\left(\sqrt{4n^2+n} - 2n\right)
    Hint

    For (a), first show sn=1n2+1+ns_n = \dfrac{1}{\sqrt{n^2+1}+n} by multiplying and dividing by the conjugate n2+1+n\sqrt{n^2+1}+n.

    Solution

    (a) sn=1n2+1+n→0s_n = \dfrac{1}{\sqrt{n^2+1}+n} \to 0.

    (b) n2+n−n=nn2+n+n=11+1/n+1→12\sqrt{n^2+n}-n = \dfrac{n}{\sqrt{n^2+n}+n} = \dfrac{1}{\sqrt{1+1/n}+1} \to \dfrac12.

    (c) 4n2+n−2n=n4n2+n+2n=14+1/n+2→14\sqrt{4n^2+n}-2n = \dfrac{n}{\sqrt{4n^2+n}+2n} = \dfrac{1}{\sqrt{4+1/n}+2} \to \dfrac14.

    Tags: StandardComputation

    Used in Exercise 8.8

    Keyboard shortcuts

    Single-key shortcuts
    KeyAction
    nNext section
    pPrevious section
    jNext exercise
    kPrevious exercise
    hShow or hide all hints
    sShow or hide all solutions
    /Search
    ?Show this list
    Ctrl K or ⌘ KSearch (built in)
    EscClose a dialog

    Shortcuts never fire while you type in a text field or hold Ctrl, Alt or ⌘. Screen reader users in browse mode can press the keys after switching to focus mode, or turn them off here.