Sequences, series and uniform convergence
Introduced| English | Meaning |
|---|---|
| pointwise/ˈpɔɪntwaɪz/ | The convergence of function values separately at each fixed point, allowing the required index to depend on that point. |
| uniform convergence/ˈjuːnɪfɔːm kənˈvɜːdʒəns/ | A form of function convergence in which one index threshold controls the error at every point of the domain. |
| conditionally/kənˈdɪʃənəli/ | The convergence of a series whose absolute-value series diverges. |
A brief peak can still affect an average
- When estimating an average signal, settling at every fixed time does not control a narrowing spike. Pointwise 逐点 convergence checks each fixed point; uniform convergence 一致收敛 controls the worst error over the whole domain.
- For example, each $x^n$ is continuous on $[0,1]$, yet its limit is $0$ for $x<1$ and $1$ at $x=1$. We will distinguish sequences, infinite sums and function limits before interchanging operations.
Cauchy depends on the ambient space
- A real sequence is Cauchy when, for every $\varepsilon>0$, all sufficiently late pairs of terms satisfy $|a_m-a_n|<\varepsilon$. In $\mathbb R$, Cauchy and convergent are equivalent; this is completeness.
- The rational truncations $1,1.4,1.41,1.414,\ldots$ approach $\sqrt2$ in $\mathbb R$. Their tails become arbitrarily close, so the sequence is Cauchy. It has no limit in $\mathbb Q$, since a limit there would also be its unique real limit, which is irrational.
Every Cauchy sequence of rational numbers converges to a rational number.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Rational decimal truncations of √2 are Cauchy, but their unique real limit is irrational. The equivalence between Cauchy and convergent depends on completeness of the ambient space.
Boundedness gives less than convergence
- A bounded monotone real sequence converges: an increasing one approaches its supremum, and a decreasing one its infimum. Boundedness alone is insufficient; $a_n=(-1)^n$ alternates forever.
- Bolzano–Weierstrass says that every bounded real sequence has a convergent subsequence. Here the even-indexed terms converge to $1$ and the odd-indexed terms to $-1$. Since a convergent sequence makes every subsequence share its limit, these two limits prove that the full sequence diverges.
A series is a limit of partial sums
- The series $\sum_{n=1}^{\infty}a_n$ converges when its partial sums $S_N=\sum_{n=1}^N a_n$ converge to a finite number. The condition $a_n\to0$ is necessary but not sufficient: the harmonic series has terms tending to zero and still diverges.
- Absolute convergence means $\sum|a_n|$ converges and implies convergence of the original series. The term conditionally 条件地 refers to the convergence of a series whose absolute-value series diverges. For example, $\sum(-1)^n/n$ converges by the alternating-series test, since $1/n$ decreases to zero. Rearrangements may change a conditionally convergent real series’s sum, but preserve an absolutely convergent one.
$\sum (-1)^n/n$ converges, but not absolutely: it converges ____.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Its absolute series is harmonic.
If $a_n\to0$, then $\sum_{n=1}^{\infty}a_n$ must converge.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The terms 1/n tend to zero, but the harmonic series diverges. Tending to zero is necessary, not sufficient.
Apply the absolute ratio test
- Suppose $a_n\ne0$ eventually and $L=\lim|a_{n+1}/a_n|$ exists, possibly as infinity. If $L<1$, the series converges absolutely. If $L>1$, including infinity, it diverges because its terms cannot approach zero.
- If $L=1$, the test gives no conclusion: $\sum1/n$ diverges while $\sum1/n^2$ converges. For $a_n=n/2^n$, the ratio is $(n+1)/(2n)\to1/2$, so the series converges absolutely.
For $\sum n/2^n$, the ratio test gives
Difficulty 1/5 · 100 coins + a small bonus on first completion today
$\frac{n+1}{2n} \to \frac12$.
Quantifiers distinguish function limits
- Pointwise refers to the convergence of function values separately at each fixed point, allowing the required index to depend on that point. For each $x\in E$ and each $\varepsilon>0$, there is an $N$ such that $|f_n(x)-f(x)|<\varepsilon$ whenever $n\geq N$.
- Uniform convergence is a form of function convergence in which one index threshold controls the error at every point of the domain. For each $\varepsilon>0$, choose one $N$ that works for every $x\in E$. Equivalently, the supremum error $\sup_{x\in E}|f_n(x)-f(x)|$ tends to zero.
The endpoint changes the supremum
- For $f_n(x)=x^n$ on $[0,1]$, the pointwise limit is zero on $[0,1)$ and one at $1$. The error at $1$ is therefore zero, but its supremum over $[0,1)$ is $1$ for every $n$. This supremum is not attained; it still prevents uniform convergence.
- A moving witness makes the failure explicit: $x_n=2^{-1/n}<1$ gives error $1/2$ for every $n$. On the smaller domain $[0,1/2]$, however, the limit is zero and the maximum error is $2^{-n}\to0$. Changing the domain changes the answer.
On $[0, \tfrac12]$, what is $\sup |x^3 - 0|$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
x³ increases on [0, ½], so the supremum of |x³| is at x = ½: (½)³ = 1/8 = 0.125.
A bound must hold on the whole domain
- On $[0,1]$, $|x^n/n|\leq1/n$, so $x^n/n$ converges uniformly to zero. Similarly, $|\sin(nx)/n|\leq1/n$ proves uniform convergence to zero on all of $\mathbb R$.
- In contrast, $x/n\to0$ at every fixed real $x$, but its error is unbounded on $\mathbb R$ for each $n$. Restricting to a fixed bounded interval gives a bound of the form $C/n$ and restores uniform convergence.
The worst point can move
- On $[0,\infty)$, $f_n(x)=nx e^{-nx}$ tends to zero at every fixed $x$, including $x=0$. Its derivative has the sign of $1-nx$, so the maximum occurs at $x=1/n$ and equals $1/e$. The peak moves, but does not shrink.
- On $[0,1]$, $g_n(x)=nx(1-x)^n$ has a similar peak at $x=1/(n+1)$. Its maximum is $(n/(n+1))^{n+1}\to1/e$, although each fixed value tends to zero. Checking a few fixed points cannot establish a supremum bound.
Pointwise only, or uniform?
Decide uniform convergence from the supremum error.
Which sequence converges uniformly on $[0, 1]$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The sequence $x^n/n$ has supremum error $1/n$, so converges uniformly to zero. For $x^n$ the error supremum is $1$. For $nx(1-x)^n$ the maximum is $(n/(n+1))^{n+1}$, tending to $1/e$ rather than zero. For $x^{1/n}$, the pointwise limit is $0$ at $x=0$ and $1$ for $x>0$, so the discontinuous limit rules out uniform convergence of these continuous functions.
At which $x\geq0$ does $5xe^{-5x}$ attain its maximum?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The derivative is 5e^(−5x)(1 − 5x), positive before x = 1/5 and negative after it. The maximum occurs at x = 1/5 = 0.2.
Compare $x^n$ on $[0,1]$, $x^n$ on $[0,1/2]$, and $x^n/n$ on $[0,1]$. Give each pointwise limit and supremum error; use a moving witness where uniformity fails.
For $x^n$ on [0,1], the limit is 0 for x < 1 and 1 at x = 1. The error at 1 is 0, but values from below approach 1, so the supremum error is 1 for every n. At $x_n=2^{-1/n}$, the error is always 1/2, proving failure of uniform convergence. On [0,1/2], the limit is zero and the maximum error is $2^{-n}$, tending to zero. For $x^n/n$ on [0,1], the limit is zero and the maximum error is 1/n. Thus the last two sequences converge uniformly.
Uniform limits preserve continuity
- If every $f_n$ is continuous on the same domain $E$ and $f_n\to f$ uniformly there, then $f$ is continuous on $E$. The domain need not be compact for this conclusion.
- At a point $x_0$, choose one $f_N$ whose error from $f$ is below $\varepsilon/3$ everywhere. Continuity of $f_N$ controls its change near $x_0$ by another $\varepsilon/3$. Adding the two approximation errors and this change proves that $f$ is continuous.
A uniform limit of continuous functions is continuous.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
True. If continuous fₙ → f uniformly, an ε/3 argument shows f is continuous. Pointwise limits need not be: xⁿ on [0, 1] has a discontinuous limit.
The M-test concerns a series of functions
- Suppose $|u_n(x)|\leq M_n$ for every $x\in E$, where the nonnegative constants $M_n$ satisfy $\sum M_n<\infty$. The Weierstrass M-test gives absolute and uniform convergence of the series $\sum u_n(x)$; it concerns its partial sums, not just individual terms tending to zero.
- On $[-1,1]$, the terms $u_n(x)=x^n/n^2$ satisfy $|u_n(x)|\leq1/n^2$. Therefore $\sum x^n/n^2$ converges uniformly. For $N\geq1$, its uniform tail is at most $\sum_{n>N}1/n^2\leq1/N$.
Which series is uniformly convergent on $[-1,1]$ by the M-test with a summable bound?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For the first series, |xⁿ/n²| ≤ 1/n² and ∑1/n² converges. Each other series diverges at x = 1, so cannot converge uniformly on the whole interval.
Interchange integrals on a finite interval
- Let every $f_n$ be Riemann integrable on a fixed finite interval $[a,b]$. If $f_n\to f$ uniformly, then $f$ is Riemann integrable there and the integrals converge to the integral of $f$.
- Put $E_n=\sup_{[a,b]}|f_n-f|$. The estimate $|\int_a^b(f_n-f)|\leq(b-a)E_n$ controls the integral error and tends to zero. Thus a uniformly convergent series of continuous functions can be integrated term by term on this interval; pointwise convergence alone does not provide this bound tending to zero.
Justify termwise integration of $\sum_{n=1}^{\infty}x^n/n^2$ on $[0,1]$. Express its integral as a numerical series, and give a uniform tail bound after $N$ terms.
Each term is continuous and $|x^n/n^2|\leq1/n^2$ on [0,1]. Since the numerical bound series converges, the M-test gives uniform convergence of the function series. For $N\geq1$, the uniform tail is at most $\sum_{n>N}1/n^2\leq1/N$ by integral comparison. The finite-interval integration theorem therefore permits termwise integration, giving the numerical series $\sum_{n=1}^{\infty}1/(n^2(n+1))$. A closed-form value is not needed to justify the interchange.
A shrinking triangle keeps its area
- For $n\geq2$, let $h_n$ be the piecewise linear triangle with vertices $(0,0)$, $(1/n,n)$ and $(2/n,0)$, and zero for $x>2/n$ in $[0,1]$. Every fixed $x>0$ eventually lies outside its support, and $h_n(0)=0$, so the pointwise limit is zero.
- Each triangle has base $2/n$ and height $n$, giving area $1$. Hence $\int_0^1h_n(x)\,dx=1$ while the integral of the pointwise limit is zero. Each $h_n$ is continuous, but the supremum is $n$; the convergence is not uniform.
A triangular $h_n$ has base $2/n$ and height $n$ on $[0,1]$, for $n\geq2$. What is $\int_0^1h_n(x)\,dx$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The area is half the base times the height: (1/2)(2/n)n = 1 for every n. Pointwise convergence to zero does not make these integrals tend to zero.
Compare the shrinking triangles $h_n$ on $[0,1]$ with $g_n(x)=1/n$ for $0\leq x\leq n$ and $g_n(x)=0$ for $x>n$ on $[0,\infty)$. Both integrals equal 1. Which convergence and domain hypotheses explain the two failures to pass the limit through the integral?
The triangles converge pointwise to zero, but their supremum is n, so convergence is not uniform. Their integrals stay 1 on the fixed finite interval, whereas the integral of their pointwise limit is 0. For the wide blocks, the supremum error is 1/n, so convergence to zero is uniform on [0,∞). However, this interval is unbounded: width n times height 1/n still gives integral 1. The finite-interval theorem does not apply, and its factor b − a cannot be replaced by a fixed finite number here. Uniform convergence alone on an unbounded domain does not justify this improper-integral interchange.
Uniform function limits do not control derivatives
- The functions $f_n(x)=\sin(nx)/n$ converge uniformly to zero, yet $f_n'(x)=\cos(nx)$. At $x=0$, every derivative equals $1$, whereas the derivative of the limit is $0$. Differentiating after taking the limit gives a different result.
- Convergent derivatives alone also leave constants uncontrolled: $f_n(x)=n$ on $[0,1]$ has derivatives identically zero, converging uniformly, but the functions do not converge even at one point. A sufficient derivative theorem must anchor the function values too.
Uniform convergence of differentiable functions guarantees that their derivatives converge to the derivative of the limit.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For fₙ(x) = sin(nx)/n, the function error is bounded by 1/n. Yet fₙ′(0) = 1 for every n, while the derivative of the zero limit is 0.
A sufficient theorem for derivatives
- Suppose the $f_n$ are continuously differentiable on a finite interval $[a,b]$, their derivatives converge uniformly to $g$, and $f_n(x_0)\to c$ at one fixed $x_0\in[a,b]$. Then $f_n$ converge uniformly to $f$, where $f(x)=c+\int_{x_0}^x g(t)\,dt$. Its derivative is $g$ in $(a,b)$.
- Put $A_n=|f_n(x_0)-c|$ and $D_n=\sup_{[a,b]}|f_n'-g|$. The fundamental theorem of calculus bounds the uniform function error by $A_n+(b-a)D_n$, which tends to zero. The uniform limit $g$ of the continuous derivatives is continuous, so the formula for $f$ also gives $f'=g$.
The $f_n$ are continuously differentiable on $[0,1]$ and $f_n\prime$ converge uniformly. Which additional assumption guarantees uniform convergence of the $f_n$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
A convergent function value at one fixed point controls the otherwise arbitrary constants. The example fₙ(x) = n has uniformly convergent zero derivatives and finite continuous function values, but no function limit.
Check both parts of the derivative theorem
- For $f_n(x)=x^2+x/n$ on $[0,1]$, the derivatives are $2x+1/n$, converging uniformly to $g(x)=2x$. Also $f_n(0)=0$ for every $n$, providing the convergent value at one fixed point.
- The theorem gives $f(x)=\int_0^x2t\,dt=x^2$ and $f'(x)=2x$. Directly, the function error is at most $1/n$ and the derivative error is $1/n$. Both estimates hold throughout the interval, rather than only at sampled points.
Is the interchange justified?
Check convergence type, the domain and the hypotheses of the relevant theorem.
On $[0,1]$, let $f_n(x)=c_n+x^3+x/n$. Its derivatives converge uniformly. Compare $c_n=(-1)^n$ with $c_n=1/n$: decide whether the functions converge and whether the derivative theorem applies.
The derivatives are $3x^2+1/n$, converging uniformly to $3x^2$ regardless of cₙ. If $c_n=(-1)^n$, the values $f_n(0)=c_n$ do not converge. Neither the theorem nor convergence of the functions follows; in fact, at every fixed x the alternating constant prevents convergence. If $c_n=1/n$, the values at 0 tend to zero, so all hypotheses of the sufficient theorem hold. The functions converge uniformly to $x^3$, with function error at most 2/n and derivative error 1/n. The derivative of the limit is $3x^2$ in the interior.
Keep the object and hypotheses clear
- A sequence limit concerns its terms; a series limit concerns partial sums. Pointwise convergence allows a threshold for each point, while uniform convergence needs one threshold for the whole domain; use a bound or a moving witness.
- Uniform limits preserve continuity. Integral interchange requires the stated integrability and finite-interval hypotheses; derivative interchange needs stronger control, such as uniform derivative convergence and a convergent function value at one point.