Skip to content

Sequences, Series, limsup and liminf

CSIR NET real analysis: monotone and Cauchy sequences, Bolzano-Weierstrass, limsup and liminf, standard limits and every convergence test for series.

  • 20 min read
  • Updated
On this page · 14 sections
  1. Prerequisites
  2. The Hook: Steps That Shrink but Never Settle
  3. Sequences and Their Limits
  4. limsup and liminf
  5. Standard Limits
  6. Series
  7. Counter-examples Table
  8. Memory Tricks
  9. Don’t Confuse
  10. Common Mistakes
  11. Practice
  12. Quick Revision Table
  13. Key Takeaways
  14. What’s Next

Prerequisites

The Hook: Steps That Shrink but Never Settle

Sequences and Their Limits

A real sequence \( (a_n) \) converges to \( L \) if for every \( \varepsilon > 0 \) there is \( N \) with \( \lvert a_n - L \rvert < \varepsilon \) for all \( n \ge N \). Limits are unique, convergent sequences are bounded, and limits respect sums, products, quotients (non-zero denominator) and non-strict inequalities.

Sandwich theorem. If \( a_n \le b_n \le c_n \) eventually and \( a_n, c_n \to L \), then \( b_n \to L \).

Monotone convergence theorem

A monotone sequence converges if and only if it is bounded. An increasing bounded sequence converges to \( \sup_n a_n \). This is the workhorse for recursive sequences.

Subsequences and Bolzano–Weierstrass

If \( a_n \to L \), every subsequence converges to \( L \). Conversely, \( a_n \to L \) if every subsequence has a further subsequence converging to \( L \). And if \( a_{2n} \to L \) and \( a_{2n-1} \to L \), then \( a_n \to L \).

Every real sequence has a monotone subsequence (the “peak” argument). Combined with monotone convergence:

Boundedness is essential (\( a_n = n \)), and the theorem fails in infinite-dimensional spaces: in \( \ell^2 \) the unit vectors \( e_n \) are bounded but \( \lVert e_n - e_m \rVert = \sqrt{2} \) for \( n \ne m \). See Metric Spaces and Several Variables.

Cauchy sequences

\( (a_n) \) is Cauchy if for every \( \varepsilon > 0 \) there is \( N \) with \( \lvert a_m - a_n \rvert < \varepsilon \) for all \( m, n \ge N \). In \( \mathbb{R} \), Cauchy \( \iff \) convergent. A sequence is Cauchy if it is contractive: \( \lvert a_{n+2} - a_{n+1} \rvert \le c \lvert a_{n+1} - a_n \rvert \) for a fixed \( c < 1 \).

limsup and liminf

For a real sequence \( (a_n) \) define

$$\limsup_{n\to\infty} a_n = \inf_{n \ge 1} \sup_{k \ge n} a_k = \lim_{n\to\infty} \sup_{k \ge n} a_k, \qquad \liminf_{n\to\infty} a_n = \sup_{n \ge 1} \inf_{k \ge n} a_k.$$

Both always exist in \( [-\infty, \infty] \): \( \sup_{k \ge n} a_k \) decreases in \( n \).

Equivalent description. Let \( E \) be the set of subsequential limits (in the extended reals). Then \( \limsup a_n = \max E \) and \( \liminf a_n = \min E \).

\( \varepsilon \)-description (for finite \( \limsup = \beta \)): for every \( \varepsilon > 0 \), \( a_n > \beta + \varepsilon \) for only finitely many \( n \), and \( a_n > \beta - \varepsilon \) for infinitely many \( n \).

PropertyStatement
Ordering\( \liminf a_n \le \limsup a_n \)
Convergence\( a_n \to L \iff \liminf a_n = \limsup a_n = L \)
Sign flip\( \limsup(-a_n) = -\liminf a_n \)
Sum\( \limsup(a_n + b_n) \le \limsup a_n + \limsup b_n \) (when the right side is defined)
Sum, one convergentequality if \( b_n \to b \)
Product, \( a_n, b_n \ge 0 \)\( \limsup(a_n b_n) \le \limsup a_n \cdot \limsup b_n \)

The sum inequality can be strict: \( a_n = (-1)^n \), \( b_n = (-1)^{n+1} \) give \( \limsup(a_n + b_n) = 0 < 2 \).

Standard Limits

LimitValue
\( a^{1/n} \), \( a > 0 \)1
\( n^{1/n} \)1
\( x^n \), \( \lvert x \rvert < 1 \)0
\( n^p x^n \), \( \lvert x \rvert < 1 \), any \( p \)0
\( \frac{x^n}{n!} \)0
\( \left(1 + \frac{x}{n}\right)^n \)\( e^x \)
\( \frac{(\log n)^p}{n^q} \), \( q > 0 \)0
\( \frac{(n!)^{1/n}}{n} \)\( \frac{1}{e} \)
\( 1 + \frac{1}{2} + \dots + \frac{1}{n} - \log n \)Euler’s constant \( \gamma \approx 0.5772 \)

Growth order. \( \log n \ll n^q \ll a^n \ll n! \ll n^n \) for \( q > 0 \), \( a > 1 \).

Cesàro means. If \( a_n \to L \) then \( \frac{a_1 + \dots + a_n}{n} \to L \). The converse is false (\( a_n = (-1)^n \) has means tending to 0). The multiplicative version: if \( a_n > 0 \) and \( a_n \to L \), then \( (a_1 \cdots a_n)^{1/n} \to L \).

Stolz–Cesàro. If \( (b_n) \) is strictly increasing and unbounded and \( \frac{a_{n+1} - a_n}{b_{n+1} - b_n} \to L \), then \( \frac{a_n}{b_n} \to L \). Example: \( \frac{1 + \frac{1}{2} + \dots + \frac{1}{n}}{\log n} \to 1 \).

Ratio and root. For \( a_n > 0 \),

$$\liminf \frac{a_{n+1}}{a_n} \le \liminf a_n^{1/n} \le \limsup a_n^{1/n} \le \limsup \frac{a_{n+1}}{a_n}.$$

So if \( \frac{a_{n+1}}{a_n} \to L \), then \( a_n^{1/n} \to L \). Applied to \( a_n = \frac{n!}{n^n} \): the ratio is \( \left(\frac{n}{n+1}\right)^n \to \frac{1}{e} \), which gives the \( (n!)^{1/n}/n \) entry in the table.

Series

\( \sum a_n \) converges if its partial sums \( s_n = a_1 + \dots + a_n \) converge. By the Cauchy criterion, this happens if and only if for every \( \varepsilon > 0 \) there is \( N \) with \( \lvert a_{n+1} + \dots + a_m \rvert < \varepsilon \) for \( m > n \ge N \). In particular:

$$\boxed{\sum a_n \text{ converges} \Rightarrow a_n \to 0}$$

The converse fails (harmonic series). A series of non-negative terms converges if and only if its partial sums are bounded.

Benchmark series

SeriesConverges iff
\( \sum x^n \)\( \lvert x \rvert < 1 \) (sum \( \frac{1}{1-x} \) from \( n = 0 \))
\( \sum \frac{1}{n^p} \)\( p > 1 \)
\( \sum \frac{1}{n (\log n)^p} \)\( p > 1 \)
\( \sum \frac{(-1)^{n}}{n^p} \)\( p > 0 \) (absolutely iff \( p > 1 \))
\( \sum \frac{1}{n^{1 + 1/n}} \)never: diverges (limit comparison with \( \frac{1}{n} \), since \( n^{1/n} \to 1 \))

Tests for non-negative terms

TestStatementBest for
Comparison\( 0 \le a_n \le b_n \): \( \sum b_n \) conv \( \Rightarrow \sum a_n \) convbounding by a p-series
Limit comparison\( \frac{a_n}{b_n} \to c \in (0, \infty) \): both converge or both divergerational functions, \( \sin\frac{1}{n} \)
Ratio (d’Alembert)\( \limsup \frac{a_{n+1}}{a_n} < 1 \): conv; \( \frac{a_{n+1}}{a_n} \ge 1 \) eventually: divfactorials, exponentials
Root (Cauchy)\( \alpha = \limsup a_n^{1/n} \): \( \alpha < 1 \) conv, \( \alpha > 1 \) div\( n \)-th powers
Condensation\( a_n \) decreasing: \( \sum a_n \) and \( \sum 2^k a_{2^k} \) converge togetherlogs
Integral\( f \ge 0 \) decreasing: \( \sum f(n) \) and \( \int_1^\infty f \) converge togetherlogs, explicit antiderivatives
Raabe\( n\left(\frac{a_n}{a_{n+1}} - 1\right) \to L \): \( L > 1 \) conv, \( L < 1 \) divratio limit equal to 1

The root test is strictly stronger than the ratio test (from the inequality chain above), and both are inconclusive at 1: \( \sum \frac{1}{n} \) and \( \sum \frac{1}{n^2} \) both give limit 1.

Series with signs

Absolute convergence. If \( \sum \lvert a_n \rvert \) converges, so does \( \sum a_n \). A convergent series that is not absolutely convergent is conditionally convergent.

Leibniz (alternating series) test. If \( b_n \) decreases to 0, then \( \sum (-1)^n b_n \) converges, and the error after \( n \) terms is at most \( b_{n+1} \).

Dirichlet test. If the partial sums of \( \sum a_n \) are bounded and \( b_n \) decreases to 0, then \( \sum a_n b_n \) converges. Since \( \sum_{k=1}^n \sin kx \) and \( \sum_{k=1}^n \cos kx \) are bounded for \( x \notin 2\pi\mathbb{Z} \), the series \( \sum \frac{\sin nx}{n} \) and \( \sum \frac{\cos nx}{n} \) converge for such \( x \).

Abel test. If \( \sum a_n \) converges and \( b_n \) is monotone and bounded, then \( \sum a_n b_n \) converges.

Rearrangements and products

  • Absolutely convergent series can be rearranged freely: every rearrangement converges to the same sum.
  • Riemann rearrangement theorem. If \( \sum a_n \) converges conditionally, then for any \( \alpha \le \beta \) in \( [-\infty, \infty] \) there is a rearrangement whose partial sums have liminf \( \alpha \) and limsup \( \beta \). In particular, rearrangements can converge to any real number or diverge.
  • Cauchy product \( c_n = \sum_{k=0}^n a_k b_{n-k} \). Mertens’ theorem: if \( \sum a_n = A \) converges absolutely and \( \sum b_n = B \) converges, then \( \sum c_n = AB \). Without absolute convergence it can fail: the Cauchy product of \( \sum_{n \ge 0} \frac{(-1)^n}{\sqrt{n+1}} \) with itself has \( \lvert c_n \rvert \ge \frac{2(n+1)}{n+2} \), which does not tend to 0, so it diverges.

Counter-examples Table

False claimCounter-example
\( a_n \to 0 \Rightarrow \sum a_n \) converges\( \sum \frac{1}{n} \)
\( a_{n+1} - a_n \to 0 \Rightarrow (a_n) \) Cauchy\( \sqrt{n} \), \( \log n \)
Bounded \( \Rightarrow \) convergent\( (-1)^n \)
Cesàro means converge \( \Rightarrow \) sequence converges\( (-1)^n \)
\( \sum a_n \) conv \( \Rightarrow \sum a_n^2 \) conv\( a_n = \frac{(-1)^n}{\sqrt{n}} \)
\( \sum a_n \) conv, \( a_n > 0 \Rightarrow \sum \sqrt{a_n} \) conv\( a_n = \frac{1}{n^2} \)
\( \sum a_n \) conv \( \Rightarrow \sum \lvert a_n \rvert \) conv\( \sum \frac{(-1)^n}{n} \)
\( \sum a_n, \sum b_n \) conv \( \Rightarrow \sum a_n b_n \) conv\( a_n = b_n = \frac{(-1)^n}{\sqrt{n}} \)
Ratio test decides whenever root test does\( \frac{1}{2} + \frac{1}{3} + \frac{1}{2^2} + \frac{1}{3^2} + \dots \) (Practice Q7)
\( \limsup (a_n + b_n) = \limsup a_n + \limsup b_n \)\( a_n = (-1)^n \), \( b_n = -(-1)^n \)
Alternating with terms \( \to 0 \Rightarrow \) converges\( b_n = \frac{1}{n} \) for even \( n \), \( \frac{1}{n^2} \) for odd \( n \) (not monotone; \( \sum (-1)^n b_n \) diverges)
Every rearrangement of a convergent series has the same sum\( 1 - \frac{1}{2} + \frac{1}{3} - \dots \)
A sequence with exactly one real subsequential limit converges\( 1, 2, 1, 3, 1, 4, \dots \) (unbounded; \( +\infty \) is a second limit in the extended reals)

Memory Tricks

Don’t Confuse

ThisvsThat
\( \sup_n a_n \) (largest value)\( \limsup a_n \) (largest subsequential limit)
Cauchy: all pairs \( m, n \ge N \) closeconsecutive terms close (much weaker)
\( a_n \to 0 \) (necessary for series)\( \sum a_n \) converges (not implied)
absolute convergence (order-proof)conditional (order-sensitive)
ratio test limit 1 (inconclusive)ratio \( \ge 1 \) eventually (divergence)

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.

Level 1: Foundation

  1. Find limsup and liminf of \( a_n = (-1)^n \left(1 + \frac{1}{n}\right) \).
  2. Does \( \sum_{n \ge 2} \frac{1}{n \log n} \) converge?
  3. State the Bolzano–Weierstrass theorem for sequences.
Answers
  1. limsup 1, liminf \( -1 \). Even terms tend to 1, odd terms to \( -1 \). (The sup is \( \frac{3}{2} \), at \( n = 2 \).)
  2. No. Condensation gives \( \sum 2^k \cdot \frac{1}{2^k k \log 2} = \frac{1}{\log 2} \sum \frac{1}{k} \), divergent. Or \( \int_2^\infty \frac{dx}{x \log x} = [\log \log x]_2^\infty = \infty \).
  3. Every bounded real sequence has a convergent subsequence.

Level 2: Part B Style

Practice Q1. \( \lim_{n\to\infty} \frac{(n!)^{1/n}}{n} \) equals:

(a) 0 (b) 1 (c) \( \frac{1}{e} \) (d) \( e \)

Answer: (c)

Let \( a_n = \frac{n!}{n^n} \). Then \( \frac{a_{n+1}}{a_n} = \left(\frac{n}{n+1}\right)^n = \left(1 + \frac{1}{n}\right)^{-n} \to \frac{1}{e} \), so \( a_n^{1/n} = \frac{(n!)^{1/n}}{n} \to \frac{1}{e} \). (Alternatively, \( \frac{1}{n}\sum \log\frac{k}{n} \to \int_0^1 \log x\, dx = -1 \).)

Practice Q2. Which of the following series converges?

(a) \( \sum \frac{1}{n^{1 + 1/n}} \) (b) \( \sum \sin\frac{1}{n} \) (c) \( \sum \left(1 - \cos\frac{1}{n}\right) \) (d) \( \sum_{n \ge 2} \frac{1}{(\log n)^2} \)

Answer: (c)

\( 1 - \cos\frac{1}{n} = 2\sin^2\frac{1}{2n} \sim \frac{1}{2n^2} \): converges by limit comparison with \( \sum \frac{1}{n^2} \). (a) \( \frac{1/n^{1+1/n}}{1/n} = n^{-1/n} \to 1 \): diverges. (b) \( \sin\frac{1}{n} \sim \frac{1}{n} \): diverges. (d) \( (\log n)^2 < n \) for large \( n \), so the terms exceed \( \frac{1}{n} \): diverges.

Practice Q3. Let \( x_1 = 1 \) and \( x_{n+1} = \sqrt{6 + x_n} \). Then \( (x_n) \):

(a) converges to 2 (b) converges to 3 (c) converges to \( \sqrt{6} \) (d) diverges

Answer: (b)

If \( x_n < 3 \) then \( x_{n+1} < \sqrt{9} = 3 \). Also \( x_{n+1} > x_n \iff x_n^2 - x_n - 6 < 0 \iff (x_n - 3)(x_n + 2) < 0 \), true for \( 0 < x_n < 3 \). Increasing and bounded, so it converges; \( L^2 = 6 + L \) gives \( L = 3 \) (reject \( -2 \)).

Practice Q4. The series \( \sum_{n \ge 1} \frac{(-1)^n}{n^p} \) converges conditionally (that is, converges but not absolutely) exactly when:

(a) \( p > 1 \) (b) \( 0 < p \le 1 \) (c) \( p > 0 \) (d) \( p \ge 1 \)

Answer: (b)

For \( p > 0 \), \( \frac{1}{n^p} \) decreases to 0, so Leibniz gives convergence. Absolute convergence needs \( \sum \frac{1}{n^p} < \infty \), that is \( p > 1 \). For \( p \le 0 \) the terms do not tend to 0. So conditional convergence is \( 0 < p \le 1 \).

Level 3: Part C Style

Practice Q5. Let \( (a_n) \) be a real sequence. Which statements are true?

(a) If \( a_n \to 0 \), then \( \frac{a_1 + \dots + a_n}{n} \to 0 \). (b) If \( \lvert a_{n+1} - a_n \rvert \to 0 \), then \( (a_n) \) is Cauchy. (c) If every subsequence of \( (a_n) \) has a further subsequence converging to \( L \), then \( a_n \to L \). (d) If \( (a_n) \) is bounded and \( \limsup a_n = \liminf a_n \), then \( (a_n) \) converges.

Answer: (a), (c) and (d)
  • (a) True. Cesàro: given \( \varepsilon \), split the average at \( N \) where \( \lvert a_n \rvert < \varepsilon \); the first \( N \) terms contribute \( O(\frac{1}{n}) \).
  • (b) False. \( a_n = \sqrt{n} \): \( a_{n+1} - a_n = \frac{1}{\sqrt{n+1} + \sqrt{n}} \to 0 \), but \( a_n \to \infty \).
  • (c) True. If \( a_n \not\to L \), there is \( \varepsilon > 0 \) and a subsequence with \( \lvert a_{n_k} - L \rvert \ge \varepsilon \); no further subsequence of it can converge to \( L \).
  • (d) True. Both equal some finite \( L \) (bounded), and \( \liminf = \limsup = L \) forces \( a_n \to L \) by the \( \varepsilon \)-description.

Practice Q6. Which of the following series converge?

(a) \( \sum \frac{(-1)^n}{\sqrt{n}} \) (b) \( \sum \frac{\sin n}{n} \) (c) \( \sum \frac{n!}{n^n} \) (d) \( \sum \frac{1}{n^{1 + 1/n}} \)

Answer: (a), (b) and (c)
  • (a) Converges by Leibniz (conditionally).
  • (b) Converges by Dirichlet: \( \sum_{k \le n} \sin k \) is bounded and \( \frac{1}{n} \) decreases to 0.
  • (c) Converges by the ratio test: \( \frac{a_{n+1}}{a_n} = \left(\frac{n}{n+1}\right)^n \to \frac{1}{e} < 1 \).
  • (d) Diverges. Limit comparison with \( \frac{1}{n} \): the ratio is \( n^{-1/n} \to 1 \). The exponent \( 1 + \frac{1}{n} \) exceeds 1 for each \( n \), but not by a fixed amount; that is the trap.

Practice Q7. Define \( a_{2k-1} = 2^{-k} \) and \( a_{2k} = 3^{-k} \) for \( k \ge 1 \), so the series is \( \frac{1}{2} + \frac{1}{3} + \frac{1}{2^2} + \frac{1}{3^2} + \dots \). Which statements are true?

(a) \( \limsup \frac{a_{n+1}}{a_n} = \infty \). (b) The ratio test is inconclusive for \( \sum a_n \). (c) \( \limsup a_n^{1/n} = \frac{1}{\sqrt{2}} \), so the root test proves convergence. (d) \( \sum a_n = \frac{3}{2} \).

Answer: all four, (a), (b), (c) and (d)
  • (a) True. \( \frac{a_{2k+1}}{a_{2k}} = \frac{2^{-(k+1)}}{3^{-k}} = \frac{1}{2}\left(\frac{3}{2}\right)^k \to \infty \).
  • (b) True. Also \( \frac{a_{2k}}{a_{2k-1}} = \left(\frac{2}{3}\right)^k \to 0 \). The ratio is neither eventually \( \ge 1 \) nor has limsup \( < 1 \).
  • (c) True. \( a_{2k-1}^{1/(2k-1)} = 2^{-k/(2k-1)} \to 2^{-1/2} \) and \( a_{2k}^{1/(2k)} = 3^{-1/2} \). The larger is \( \frac{1}{\sqrt{2}} < 1 \).
  • (d) True. The series is absolutely convergent, so regroup: \( \sum_{k \ge 1} 2^{-k} + \sum_{k \ge 1} 3^{-k} = 1 + \frac{1}{2} = \frac{3}{2} \).

Part C can have all options correct. Do not leave one unmarked because “four can’t all be right”.

Quick Revision Table

IdeaFact
Monotone convergencemonotone + bounded \( \iff \) convergent
Bolzano–Weierstrassbounded \( \Rightarrow \) convergent subsequence
Cauchy in \( \mathbb{R} \)\( \iff \) convergent
limsup\( \inf_n \sup_{k \ge n} a_k \) = largest subsequential limit
Ratio–root chain\( \liminf \frac{a_{n+1}}{a_n} \le \liminf a_n^{1/n} \le \limsup a_n^{1/n} \le \limsup \frac{a_{n+1}}{a_n} \)
\( \sum \frac{1}{n^p} \), \( \sum \frac{1}{n (\log n)^p} \)converge iff \( p > 1 \)
Condensation\( \sum a_n \sim \sum 2^k a_{2^k} \) for decreasing \( a_n \ge 0 \)
Raabe\( n(\frac{a_n}{a_{n+1}} - 1) \to L \): \( L > 1 \) conv
Dirichletbounded partial sums \( \times \) decreasing to 0
Rearrangementabsolute: sum fixed; conditional: any sum
Mertensone absolutely convergent \( \Rightarrow \) Cauchy product converges to \( AB \)

Key Takeaways

What’s Next