Chapter 4 · Real Analysis

4.2 Sequences and Series of Functions, Ascoli–Arzelà and Weierstrass

Pointwise convergence and why it is not enough; uniform convergence as convergence in the sup metric; the M-test, Dirichlet, Abel and Dini; when limits, integrals and derivatives may be interchanged; power series and their radius of convergence; and the two great theorems about $C[a,b]$ — Ascoli–Arzelà and Weierstrass approximation.

Many of the most useful functions in mathematics are born as limits. The exponential is the sum of its power series; the solution of an initial value problem is the limit of its Picard iterates (§4.1, §6.1); a vibrating string is described by a Fourier series; a continuous function on $[0,1]$ is the limit of its Bernstein polynomials. In each case we want to know what survives the limit. If every $f_n$ is continuous, is the limit continuous? Is the integral of the limit the limit of the integrals? May a series be differentiated term by term? The naive answer "yes" is wrong for the most natural notion of limit — convergence at each point — and right for a stronger one, uniform convergence, which is nothing but convergence in the sup metric $d_\infty$ of §4.1.

GATE MA examines this material through short, sharp questions: does this sequence converge uniformly on this set; what is $\lim\int_0^1f_n$; which of these series converge uniformly; what is the radius of convergence of this power series and what is its sum at a point; which of these families of functions has a uniformly convergent subsequence; can this function be approximated uniformly by polynomials. This section gives the theorems that answer them — with proofs of the central ones — and the counterexamples that make the MSQ options false.

Pointwise and Uniform Convergence

Throughout, $D$ is a nonempty set (usually an interval) and $f_n,f:D\to\mathbb{R}$.

📘 Definition 4.9 — Pointwise convergence; series of functions
$(f_n)$ converges pointwise to $f$ on $D$ if $f_n(x)\to f(x)$ for every $x\in D$: for each $x$ and each $\varepsilon\gt0$ there is $N=N(\varepsilon,x)$ with $|f_n(x)-f(x)|\lt\varepsilon$ for $n\ge N$. A series of functions $\sum u_n$ converges pointwise if its partial sums $S_N=\sum_{n\le N}u_n$ do; the limit is its sum.

Pointwise convergence loses almost everything. (1) Continuity: $x^n\to0$ for $0\le x\lt1$ and $\to1$ at $x=1$, so continuous functions converge to a discontinuous one; equivalently $\lim_n\lim_{x\to1^-}x^n=1\ne0=\lim_{x\to1^-}\lim_nx^n$. (2) Integrals: the tent $g_n$ of height $n$ on $[0,\frac2n]$ (zero elsewhere) tends to $0$ at every point of $[0,1]$, yet $\int_0^1g_n=1$ for every $n$. (3) Derivatives: $h_n(x)=\frac{\sin nx}{\sqrt n}\to0$ (even uniformly), but $h_n'(x)=\sqrt n\cos nx$ does not converge at all (at $x=0$ it is $\sqrt n$). The fix is to make $N$ independent of $x$.

📘 Definition 4.10 — Uniform convergence, the sup norm, uniformly Cauchy

$(f_n)$ converges uniformly to $f$ on $D$, written $f_n\rightrightarrows f$, if for every $\varepsilon\gt0$ there is $N$ (depending only on $\varepsilon$) with $|f_n(x)-f(x)|\lt\varepsilon$ for all $x\in D$ and all $n\ge N$. Geometrically: from $N$ on, the whole graph of $f_n$ lies in the $\varepsilon$-tube around the graph of $f$.

The sup norm of $g:D\to\mathbb{R}$ is $\|g\|_D=\sup_{x\in D}|g(x)|$ (written $\|g\|_\infty$ when $D$ is clear); for continuous functions on $[a,b]$ it is the sup metric $d_\infty(f,g)=\|f-g\|_\infty$ of §4.1. $(f_n)$ is uniformly Cauchy on $D$ if for every $\varepsilon\gt0$ there is $N$ with $\|f_n-f_m\|_D\lt\varepsilon$ for $m,n\ge N$. A series converges uniformly if its partial sums do.

📐 Theorem 4.13 — Testing for uniform convergence
  1. Sup-norm criterion. $f_n\rightrightarrows f$ on $D$ if and only if $M_n=\|f_n-f\|_D\to0$.
  2. Cauchy criterion. $(f_n)$ converges uniformly on $D$ (to some $f$) if and only if it is uniformly Cauchy on $D$.
  3. Negative tests. (a) If there are points $x_n\in D$ with $|f_n(x_n)-f(x_n)|\not\to0$, the convergence is not uniform. (b) If $\sum u_n$ converges uniformly on $D$, then $\|u_n\|_D\to0$. (c) If each $f_n$ is continuous on $[a,b]$ and $(f_n)$ converges uniformly on $(a,b)$, it converges uniformly on $[a,b]$ — so if $(f_n(a))$ diverges, the convergence on $(a,b)$ is not uniform.

Proof. (1) "$|f_n(x)-f(x)|\lt\varepsilon$ for all $x$" says $\|f_n-f\|_D\le\varepsilon$; this is the definition with $\varepsilon$ and $\le$ (harmless). (2) If $f_n\rightrightarrows f$ then $\|f_n-f_m\|\le\|f_n-f\|+\|f-f_m\|$. Conversely, if $(f_n)$ is uniformly Cauchy, each $(f_n(x))$ is Cauchy in $\mathbb{R}$, with limit $f(x)$. Choose $N$ with $|f_n(x)-f_m(x)|\lt\varepsilon$ for all $x$ and $m,n\ge N$, and let $m\to\infty$: $|f_n(x)-f(x)|\le\varepsilon$ for all $x$ and $n\ge N$. (3)(a) $M_n\ge|f_n(x_n)-f(x_n)|$. (b) $u_n=S_n-S_{n-1}$ and $\|S_n-S_{n-1}\|\to0$ by (2). (c) For $m,n$ fixed, $|f_n(a)-f_m(a)|$ equals $\lim_{x\to a^+}|f_n(x)-f_m(x)|$, which is at most $\|f_n-f_m\|_{(a,b)}$; the same holds at $b$; so the sequence is uniformly Cauchy on $[a,b]$. ∎

🧭 Deciding uniform convergence in GATE time
  1. Find the pointwise limit $f$ (treat endpoints and special points such as $x=0$ separately). If every $f_n$ is continuous and $f$ is not, stop: not uniform near the discontinuity (Theorem 4.16).
  2. Compute $M_n=\sup|f_n-f|$ — usually by calculus: differentiate $f_n-f$, locate the maximum (it typically sits at $x\approx\frac1n$ or $\frac1{\sqrt n}$), and evaluate. Or bound $|f_n-f|$ by something independent of $x$ that tends to $0$.
  3. Or exhibit a bad sequence $x_n$ (try $\frac1n$, $\frac1{\sqrt n}$, $1-\frac1n$, $n$) with $|f_n(x_n)-f(x_n)|$ bounded away from $0$.
  4. The domain matters. Restricting to $[a,1]$ with $a\gt0$, or to a bounded interval, often makes the convergence uniform; check whether the bad points escape the new domain.
  5. For series: M-test first; if $\sum\|u_n\|$ diverges, try Dirichlet/Abel or the alternating-series bound; if $\|u_n\|\not\to0$, the series is not uniformly convergent.
✏️ Example 1 — Four sup-norm computations
Find the pointwise limit and decide whether the convergence is uniform. (a) $f_n(x)=x^n$ on $[0,1)$ and on $[0,a]$, $0\lt a\lt1$. (b) $f_n(x)=\dfrac{nx}{1+n^2x^2}$ on $[0,1]$ and on $[a,1]$, $0\lt a\lt1$. (c) $f_n(x)=\dfrac{x}{1+nx^2}$ on $\mathbb{R}$. (d) $f_n(x)=x^n(1-x)$ on $[0,1]$.
(a) The limit is $0$ on $[0,1)$. There, $M_n=\sup_{[0,1)}x^n=1$ for every $n$ (approach $x\to1$): not uniform, although the limit $0$ is continuous on $[0,1)$. On $[0,a]$, $M_n=a^n\to0$: uniform.
(b) $f_n(0)=0$ and for $x\gt0$, $f_n(x)\le\frac{nx}{n^2x^2}=\frac1{nx}\to0$; the limit is $0$. With $t=nx$, $f_n=\frac t{1+t^2}$, maximal at $t=1$: $M_n=f_n(\frac1n)=\frac12$ for every $n$. Not uniform on $[0,1]$ — the bump of height $\frac12$ slides into $0$. On $[a,1]$ the maximum point $\frac1n$ leaves the interval once $n\gt\frac1a$, $f_n$ is decreasing there, and $M_n=f_n(a)=\frac{na}{1+n^2a^2}\to0$: uniform.
(c) The limit is $0$ ($f_n(0)=0$; $|f_n(x)|\le\frac1{n|x|}$ otherwise). $f_n'(x)=\frac{1-nx^2}{(1+nx^2)^2}=0$ at $x=\pm\frac1{\sqrt n}$, so $M_n=f_n(\frac1{\sqrt n})=\frac{1/\sqrt n}{2}=\frac1{2\sqrt n}\to0$: uniform on all of $\mathbb{R}$.
(d) The limit is $0$ (at $x=1$ the factor $1-x$ vanishes). $f_n'=x^{n-1}(n-(n+1)x)=0$ at $x=\frac n{n+1}$, so $M_n=\left(\frac n{n+1}\right)^n\frac1{n+1}\le\frac1{n+1}\to0$: uniform. The factor $(1-x)$ kills exactly the part of $x^n$ that did not converge uniformly in (a).

Series of Functions: the M-test, Dirichlet, Abel and Dini

📐 Theorem 4.14 — The Weierstrass M-test; Dirichlet's and Abel's tests
  1. M-test. If $|u_n(x)|\le M_n$ for all $x\in D$ and $\sum M_n\lt\infty$, then $\sum u_n$ converges absolutely and uniformly on $D$.
  2. Dirichlet's test. If the partial sums $A_n(x)=\sum_{k=1}^na_k(x)$ are uniformly bounded on $D$, $|A_n(x)|\le A$, and for each $x$ the sequence $b_k(x)$ is decreasing in $k$ with $b_k\rightrightarrows0$ on $D$, then $\sum a_kb_k$ converges uniformly on $D$. In particular (with $a_k=(-1)^{k+1}$): if $b_k(x)\downarrow0$ uniformly, $\sum(-1)^{k+1}b_k(x)$ converges uniformly, with tail bound $\left|\sum_{k\gt n}(-1)^{k+1}b_k(x)\right|\le b_{n+1}(x)$.
  3. Abel's test. If $\sum a_k(x)$ converges uniformly on $D$ and $b_k(x)$ is monotone in $k$ for each $x$ and uniformly bounded, $|b_k(x)|\le B$, then $\sum a_kb_k$ converges uniformly on $D$.

Proof. (1) For $m\gt n$ and every $x$, $\left|\sum_{k=n+1}^mu_k(x)\right|\le\sum_{k=n+1}^mM_k$, which is small for large $n$ because $\sum M_k$ converges. So the partial sums are uniformly Cauchy, and Theorem 4.13(2) applies (absolute convergence follows the same way).

(2) Summation by parts: with $A_0=0$ and $a_k=A_k-A_{k-1}$,

$$\sum_{k=m}^na_kb_k=A_nb_n-A_{m-1}b_m+\sum_{k=m}^{n-1}A_k(b_k-b_{k+1}).$$

Since $b_k-b_{k+1}\ge0$ and $b_k\ge0$, the absolute value is at most $A\,b_n+A\,b_m+A(b_m-b_n)$, that is, at most $2A\,b_m(x)\le2A\|b_m\|_D$, which tends to $0$: the series is uniformly Cauchy. For the alternating case $|A_n|\le1$, and the sharper tail bound is the usual alternating-series estimate applied at each $x$. (3) is proved in the same way, applying summation by parts to the tails $\sum_{k\gt n}a_k$, which tend to $0$ uniformly. ∎

📐 Theorem 4.15 — Dini's theorem
Let $K$ be a compact metric space (e.g. $[a,b]$), $f_n:K\to\mathbb{R}$ continuous, $f_n\to f$ pointwise with $f$ continuous, and suppose the convergence is monotone: $f_n(x)\le f_{n+1}(x)$ for all $n,x$ (or $\ge$ for all $n,x$). Then $f_n\rightrightarrows f$ on $K$. In particular, a series of non-negative continuous functions with a continuous sum converges uniformly on a compact set.

Proof. Put $g_n=f-f_n$ (for increasing $f_n$): continuous, $g_n\ge g_{n+1}\ge0$, $g_n\to0$ pointwise. Fix $\varepsilon\gt0$ and let $K_n=\{x:g_n(x)\ge\varepsilon\}$, a closed (Theorem 4.2) hence compact subset of $K$. The sets decrease, $K_1\supseteq K_2\supseteq\cdots$, and their intersection is empty because $g_n(x)\to0$ for each $x$. The open sets $K\setminus K_n$ therefore cover $K$, so finitely many do, and since they increase, $K\setminus K_N=K$ for one $N$: $K_N=\varnothing$. Then $0\le g_n\le g_N\lt\varepsilon$ on $K$ for all $n\ge N$. ∎

Each hypothesis is needed: $x^n$ on $[0,1]$ (limit not continuous), $x^n$ on $[0,1)$ (domain not compact), and the sliding bumps $\frac{nx}{1+n^2x^2}$ of Example 1(b) (not monotone in $n$) all converge pointwise but not uniformly.

✏️ Example 2 — Which series converge uniformly?
(a) $\displaystyle\sum_{n\ge1}\frac{\sin nx}{n^2}$ on $\mathbb{R}$. (b) $\displaystyle\sum_{n\ge0}x^n$ on $(-1,1)$ and on $[-a,a]$, $0\lt a\lt1$. (c) $\displaystyle\sum_{n\ge1}\frac{(-1)^n}{n+x^2}$ on $\mathbb{R}$. (d) $\displaystyle\sum_{n\ge1}\frac{\sin nx}{n}$ on $[\delta,2\pi-\delta]$, $0\lt\delta\lt\pi$, and on $(0,2\pi)$. (e) $\displaystyle\sum_{n\ge0}\frac{x^2}{(1+x^2)^n}$ on $\mathbb{R}$.
(a) Yes M-test with $M_n=\frac1{n^2}$. The sum is continuous on $\mathbb{R}$ (Theorem 4.16).
(b) On $(-1,1)$: $\sup_{(-1,1)}|x^n|=1\not\to0$, so not uniform (Theorem 4.13(3b)). On $[-a,a]$: M-test with $M_n=a^n$, uniform.
(c) Yes, not absolutely $b_n(x)=\frac1{n+x^2}$ decreases in $n$ and $\|b_n\|_{\mathbb{R}}=\frac1n\to0$, so the alternating form of Dirichlet's test applies: $|\text{tail after }n|\le\frac1{n+1}$ for every $x$. The M-test cannot work, since $\sum\|u_n\|=\sum\frac1n$ diverges — and indeed the series does not converge absolutely at any $x$.
(d) On $[\delta,2\pi-\delta]$, $\left|\sum_{k=1}^n\sin kx\right|=\left|\frac{\sin(nx/2)\sin((n+1)x/2)}{\sin(x/2)}\right|\le\frac1{\sin(\delta/2)}$, and $\frac1n\downarrow0$: uniform by Dirichlet. On $(0,2\pi)$ the sum is $\frac{\pi-x}2$, which tends to $\frac\pi2$ as $x\to0^+$, while every partial sum vanishes at $x=0$. If the convergence were uniform on $(0,2\pi)$, it would be uniform on $[0,2\pi]$ by Theorem 4.13(3c), and the sum would be continuous at $0$ with value $0$ — but its right-hand limit is $\frac\pi2$. Not uniform.
(e) No For $x\ne0$ this is a geometric series with ratio $\frac1{1+x^2}\lt1$: the sum is $\frac{x^2}{1-\frac1{1+x^2}}=1+x^2$. At $x=0$ every term is $0$. The terms are continuous but the sum jumps at $0$, so the convergence is not uniform on any interval containing $0$. It is uniform on $\{|x|\ge a\}$ for each $a\gt0$: the tail after the term $n=N$ is $\sum_{n\gt N}\frac{x^2}{(1+x^2)^n}=(1+x^2)^{-N}\le(1+a^2)^{-N}$.

Limits, Integrals and Derivatives under Uniform Convergence

📐 Theorem 4.16 — Continuity, double limits and integrals
  1. Continuity. If each $f_n$ is continuous at $x_0\in D$ and $f_n\rightrightarrows f$ on $D$, then $f$ is continuous at $x_0$. A uniform limit of continuous functions is continuous; the sum of a uniformly convergent series of continuous functions is continuous.
  2. Interchanging limits. If $f_n\rightrightarrows f$ on $D$ and $L_n=\lim_{x\to x_0}f_n(x)$ exists for each $n$ ($x_0$ a limit point of $D$), then $(L_n)$ converges and $\lim_{x\to x_0}f(x)=\lim_nL_n$.
  3. Integration. If each $f_n$ is Riemann integrable on $[a,b]$ and $f_n\rightrightarrows f$ on $[a,b]$, then $f$ is Riemann integrable and
$$\int_a^bf_n\to\int_a^bf;\quad\text{indeed}\quad \int_a^xf_n\rightrightarrows\int_a^xf\ \text{on }[a,b].$$

For series: $\int_a^b\sum u_n=\sum\int_a^bu_n$ whenever $\sum u_n$ converges uniformly on $[a,b]$.

Proof. (1) Given $\varepsilon$, choose $n$ with $\|f_n-f\|_D\lt\frac\varepsilon3$, then $\delta$ with $|f_n(x)-f_n(x_0)|\lt\frac\varepsilon3$ for $|x-x_0|\lt\delta$. For such $x$,

$$|f(x)-f(x_0)|\le|f(x)-f_n(x)|+|f_n(x)-f_n(x_0)|+|f_n(x_0)-f(x_0)|\lt\varepsilon.$$

(2) Letting $x\to x_0$ in $|f_n(x)-f_m(x)|\le\|f_n-f_m\|_D$ gives $|L_n-L_m|\le\|f_n-f_m\|_D$, so $(L_n)$ is Cauchy, with limit $L$. Then the same $\frac\varepsilon3$ argument with $L_n$ in place of $f_n(x_0)$ gives $|f(x)-L|\lt\varepsilon$ near $x_0$.

(3) Let $M_n=\|f_n-f\|$. For any partition, the upper and lower sums of $f$ differ from those of $f_n$ by at most $M_n(b-a)$ each, so $U(f)-L(f)\le U(f_n)-L(f_n)+2M_n(b-a)$, which can be made small: $f$ is integrable. Then $\left|\int_a^xf_n-\int_a^xf\right|\le\int_a^x|f_n-f|\le(b-a)M_n\to0$ uniformly in $x$. ∎

📐 Theorem 4.17 — Differentiation term by term
Let $f_n$ be differentiable on $[a,b]$. Suppose $(f_n')$ converges uniformly on $[a,b]$ to some $g$, and $(f_n(x_0))$ converges for at least one $x_0\in[a,b]$. Then $(f_n)$ converges uniformly on $[a,b]$ to a differentiable function $f$, and $f'=g$, that is, $\left(\lim f_n\right)'=\lim f_n'$. For series: if $\sum u_n'$ converges uniformly and $\sum u_n(x_0)$ converges, then $\left(\sum u_n\right)'=\sum u_n'$.

Proof (when the $f_n'$ are continuous, the case met in practice). By the fundamental theorem of calculus $f_n(x)=f_n(x_0)+\int_{x_0}^xf_n'$. The first term converges to some $c$, and the integral converges uniformly to $\int_{x_0}^xg$ by Theorem 4.16(3). So $f_n\rightrightarrows f$ with $f(x)=c+\int_{x_0}^xg$; $g$ is continuous by Theorem 4.16(1), so $f'=g$ by the fundamental theorem of calculus. (Without continuity of $f_n'$ the mean value theorem applied to $f_n-f_m$ shows that $(f_n)$ is uniformly Cauchy, and Theorem 4.16(2) applied to the difference quotients gives $f'=g$; see 📚 Go deeper.) ∎

Uniform convergence of $(f_n)$ itself is not enough: $\sqrt{x^2+\frac1n}\rightrightarrows|x|$ on $[-1,1]$ (the difference is at most $\frac1{\sqrt n}$), each function is differentiable, and the limit is not differentiable at $0$; and $\frac{\sin nx}{\sqrt n}\rightrightarrows0$ while the derivatives diverge.

✏️ Example 3 — Limits of integrals; term-by-term integration and differentiation
(a) Find $\displaystyle\lim_{n\to\infty}\int_0^1\frac{nx}{1+n^2x^2}\,dx$ and $\displaystyle\lim_{n\to\infty}\int_0^12nx\,e^{-nx^2}\,dx$, and compare each with the integral of the pointwise limit. (b) Evaluate $\displaystyle\int_0^\pi\sum_{n\ge1}\frac{\sin nx}{n^2}\,dx$ as a series. (c) Show that $F(x)=\sum_{n\ge1}\frac{\sin nx}{n^3}$ is differentiable on $\mathbb{R}$ with $F'(x)=\sum_{n\ge1}\frac{\cos nx}{n^2}$.
(a) First $\int_0^1\frac{nx}{1+n^2x^2}dx=\frac{\log(1+n^2)}{2n}\to0$, which equals $\int_0^10$. The interchange holds although the convergence is not uniform (Example 1(b)): uniform convergence is sufficient, not necessary.
(a) Second The pointwise limit is $0$ (for $x\gt0$, $e^{-nx^2}$ beats $n$). But $\int_0^12nxe^{-nx^2}dx=\big[-e^{-nx^2}\big]_0^1=1-e^{-n}\to1\ne0$. The convergence cannot be uniform: $M_n=f_n(1/\sqrt{2n})=\sqrt{2n}\,e^{-1/2}\to\infty$.
(b) M-test with $\frac1{n^2}$ gives uniform convergence on $[0,\pi]$, so integrate term by term: $\int_0^\pi\sin nx\,dx=\frac{1-(-1)^n}n$, and the integral is $\sum_n\frac{1-(-1)^n}{n^3}=\sum_{n\text{ odd}}\frac2{n^3}=\frac74\zeta(3)\approx2.1036$.
(c) The differentiated series $\sum\frac{\cos nx}{n^2}$ converges uniformly on $\mathbb{R}$ (M-test with $\frac1{n^2}$) and $\sum\frac{\sin n\cdot0}{n^3}=0$ converges, so Theorem 4.17 applies on every $[a,b]$. Differentiating once more is not justified this way: $\sum\frac{\sin nx}{n}$ converges, but not uniformly near $0$ (Example 2(d)).
✏️ Example 4 — Prove or disprove
(a) If $f_n\rightrightarrows f$ on $\mathbb{R}$ and each $f_n$ is bounded, then $f$ is bounded. (b) If $f_n\rightrightarrows f$ and $g_n\rightrightarrows g$ on $\mathbb{R}$, then $f_ng_n\rightrightarrows fg$. (c) If $f_n\rightrightarrows f$ on $(0,1)$ and each $f_n$ is uniformly continuous, then $f$ is uniformly continuous. (d) If $f_n\to f$ pointwise on $[0,1]$, all $f_n$ and $f$ are continuous, then the convergence is uniform. (e) If $\sum u_n$ converges uniformly on $D$, then $\sum|u_n|$ converges uniformly on $D$.
(a) True Take $N$ with $\|f_N-f\|\lt1$; then $|f|\le|f_N|+1$ is bounded. (The sequence is then even uniformly bounded: $\|f_n\|\le\|f\|+1$ for $n\ge N$.)
(b) False $f_n(x)=x+\frac1n\rightrightarrows x$ and $g_n(x)=x$; then $f_ng_n-x^2=\frac xn$, whose sup over $\mathbb{R}$ is infinite. It becomes true when all the functions are bounded: $\|f_ng_n-fg\|\le\|f_n\|\|g_n-g\|+\|g\|\|f_n-f\|$.
(c) True Given $\varepsilon$, fix $n$ with $\|f_n-f\|\lt\frac\varepsilon3$ and take the $\delta$ of uniform continuity of $f_n$ for $\frac\varepsilon3$; the $\frac\varepsilon3$ argument of Theorem 4.16(1) works with the same $\delta$ at every point.
(d) False $f_n(x)=\frac{nx}{1+n^2x^2}$ (Example 1(b)): continuous functions, continuous limit $0$, $M_n=\frac12$. Dini's theorem would need monotonicity in $n$, which fails.
(e) False $\sum\frac{(-1)^n}{n+x^2}$ converges uniformly on $\mathbb{R}$ (Example 2(c)) but $\sum\frac1{n+x^2}$ diverges at every $x$.

Power Series

📘 Definition 4.11 — Power series, radius of convergence
A power series about $c$ is $\sum_{n\ge0}a_n(x-c)^n$. Its radius of convergence is $$R=\frac1{\limsup_{n\to\infty}|a_n|^{1/n}}\in[0,\infty],$$ with $\frac10=\infty$ and $\frac1\infty=0$. The set of $x$ where the series converges is its interval of convergence.
📐 Theorem 4.18 — Cauchy–Hadamard, term-by-term calculus, Abel's theorem
  1. Cauchy–Hadamard. The series converges absolutely for $|x-c|\lt R$ and diverges for $|x-c|\gt R$; it converges uniformly on every closed interval $[c-r,c+r]$ with $r\lt R$ (but not necessarily on $(c-R,c+R)$).
  2. Ratio formula. If $\left|\frac{a_{n+1}}{a_n}\right|\to L\in[0,\infty]$, then $R=\frac1L$. (If the ratio has no limit, use the root formula.)
  3. Term-by-term calculus. The series $\sum na_n(x-c)^{n-1}$ and $\sum\frac{a_n}{n+1}(x-c)^{n+1}$ have the same radius $R$. On $(c-R,c+R)$ the sum $f$ is infinitely differentiable, may be differentiated and integrated term by term, and $a_n=\frac{f^{(n)}(c)}{n!}$; in particular the coefficients are determined by $f$.
  4. Abel's theorem. If $\sum a_nR^n$ converges ($0\lt R\lt\infty$, $c=0$), then the series converges uniformly on $[0,R]$, and $\lim_{x\to R^-}\sum a_nx^n=\sum a_nR^n$. (Similarly at $-R$.)

Proof. (1) Take $c=0$. If $r\lt\rho\lt R$, then $\limsup|a_n|^{1/n}\lt\frac1\rho$, so $|a_n|\lt\rho^{-n}$ for $n\ge N$ and $|a_nx^n|\le(r/\rho)^n$ on $[-r,r]$: the M-test gives absolute and uniform convergence there, and every $|x|\lt R$ lies in such an interval. If $|x|\gt R$, then $|a_n|^{1/n}\gt\frac1{|x|}$ for infinitely many $n$, i.e. $|a_nx^n|\gt1$ infinitely often, and the terms do not tend to $0$.

(2) The standard inequalities

$$\liminf\left|\tfrac{a_{n+1}}{a_n}\right|\le\liminf|a_n|^{1/n},$$ $$\limsup|a_n|^{1/n}\le\limsup\left|\tfrac{a_{n+1}}{a_n}\right|$$

force $|a_n|^{1/n}\to L$ when the ratios converge to $L$.

(3) $n^{1/n}\to1$, so $\limsup|na_n|^{1/n}=\limsup|a_n|^{1/n}$, and similarly for $\frac{a_n}{n+1}$. On $[c-r,c+r]$, $r\lt R$, the differentiated series converges uniformly by (1), so Theorem 4.17 gives $f'=\sum na_n(x-c)^{n-1}$; repeat. Setting $x=c$ in the $n$-th derivative gives $f^{(n)}(c)=n!\,a_n$. Integration term by term is Theorem 4.16(3).

(4) (Sketch.) Scale to $R=1$. Then $\sum a_nx^n$ converges uniformly on $[0,1]$, by Abel's test (Theorem 4.14(3)) applied to the constant functions $a_n$ (whose series converges, trivially uniformly) and $b_n(x)=x^n$, which is decreasing in $n$ and bounded by $1$. The sum is then continuous on $[0,1]$ (Theorem 4.16(1)), which is the claim. ∎

The radius does not decide the endpoints: $\sum x^n$ converges at neither endpoint, $\sum\frac{x^n}n$ only at $x=-1$, $\sum\frac{x^n}{n^2}$ at both. Term-by-term differentiation keeps $R$ but can lose endpoints, integration can gain them. And the converse of Abel's theorem is false: $\sum(-1)^nx^n=\frac1{1+x}\to\frac12$ as $x\to1^-$, but $\sum(-1)^n$ diverges. Finally, a smooth function's Taylor series can have a small radius for a reason invisible on $\mathbb{R}$: $\frac1{1+x^2}=\sum(-1)^nx^{2n}$ has $R=1$ because of the complex singularities $\pm i$ (Chapter 5).

✏️ Example 5 — Five radii of convergence
Find the radius of convergence of (a) $\displaystyle\sum\frac{n^2}{3^n}x^n$; (b) $\displaystyle\sum\frac{x^{2n}}{4^n}$; (c) $\displaystyle\sum\frac{n!}{n^n}x^n$; (d) $\displaystyle\sum\big(2+(-1)^n\big)^nx^n$; (e) $\displaystyle\sum_{n\ge1}x^{n!}$.
(a) $3$ $\left(\frac{n^2}{3^n}\right)^{1/n}=\frac{(n^{1/n})^2}3\to\frac13$. (Polynomial factors never change the radius.)
(b) $2$ The coefficients are $a_{2n}=4^{-n}$ and $a_{2n+1}=0$, so $|a_{2n}|^{1/(2n)}=\frac12$ and $\limsup|a_k|^{1/k}=\frac12$. Faster: as a series in $y=x^2$ it converges iff $|y|\lt4$, i.e. $|x|\lt2$. The tempting answers $4$ and $\frac14$ are wrong.
(c) $e$ $\frac{a_{n+1}}{a_n}=\frac{(n+1)!}{(n+1)^{n+1}}\cdot\frac{n^n}{n!}=\left(\frac n{n+1}\right)^n\to\frac1e$, so $R=e$.
(d) $\frac13$ $|a_n|^{1/n}=2+(-1)^n$ alternates between $1$ and $3$: $\limsup=3$, $R=\frac13$. The ratio $\frac{a_{n+1}}{a_n}$ has no limit (it alternates between $3^{n+1}$ and $3^{-n}$), so the ratio formula is useless here.
(e) $1$ $a_k=1$ if $k$ is a factorial and $0$ otherwise, so $\limsup|a_k|^{1/k}=1$. (At $x=\pm1$ the terms do not tend to $0$.)
✏️ Example 6 — Summing series with power series
Find (a) $\displaystyle\sum_{n\ge1}\frac{n^2}{2^n}$; (b) $\displaystyle\sum_{n\ge1}\frac1{n2^n}$; (c) $\displaystyle\sum_{n\ge0}\frac{(-1)^n}{2n+1}$.
(a) $6$ For $|x|\lt1$: $\sum x^n=\frac1{1-x}$; differentiate and multiply by $x$: $\sum nx^n=\frac x{(1-x)^2}$; again: $\sum n^2x^n=x\frac{d}{dx}\frac x{(1-x)^2}=\frac{x(1+x)}{(1-x)^3}$. At $x=\frac12$: $\frac{\frac12\cdot\frac32}{\frac18}=6$.
(b) $\log2$ Integrate $\sum_{n\ge0}t^n=\frac1{1-t}$ from $0$ to $x$ (Theorem 4.18(3)): $\sum_{n\ge1}\frac{x^n}n=-\log(1-x)$ for $|x|\lt1$. At $x=\frac12$: $-\log\frac12=\log2\approx0.6931$.
(c) $\frac\pi4$ Integrating $\frac1{1+t^2}=\sum(-1)^nt^{2n}$ gives $\arctan x=\sum_{n\ge0}\frac{(-1)^nx^{2n+1}}{2n+1}$ for $|x|\lt1$. At $x=1$ the series converges (alternating test), so by Abel's theorem its sum is $\lim_{x\to1^-}\arctan x=\frac\pi4$ (Leibniz's series). Without Abel's theorem we would only know the formula for $|x|\lt1$.

Equicontinuity, Ascoli–Arzelà and Weierstrass Approximation

In $\mathbb{R}^n$ every bounded sequence has a convergent subsequence. In $(C[0,1],d_\infty)$ this fails — $x^n$ and $\sin nx$ are bounded with no uniformly convergent subsequence (§4.1). The Ascoli–Arzelà theorem identifies the missing ingredient: the functions must be continuous in the same way.

📘 Definition 4.12 — Equicontinuous and uniformly bounded families
Let $K$ be a metric space and $\mathcal{F}$ a family of functions $K\to\mathbb{R}$. $\mathcal{F}$ is equicontinuous if for every $\varepsilon\gt0$ there is $\delta\gt0$ such that $|f(x)-f(y)|\lt\varepsilon$ for all $f\in\mathcal{F}$ and all $x,y$ with $d(x,y)\lt\delta$ — one $\delta$ for all members of the family (and, here, for all points). $\mathcal{F}$ is uniformly bounded if $\sup_{f\in\mathcal{F}}\|f\|_\infty\lt\infty$, and pointwise bounded if $\sup_{f\in\mathcal{F}}|f(x)|\lt\infty$ for each $x$.
📐 Theorem 4.19 — Arzelà–Ascoli theorem

Let $K$ be a compact metric space (e.g. $[a,b]$) and $\mathcal{F}\subseteq C(K)$ with the sup metric. Then $\overline{\mathcal{F}}$ is compact if and only if $\mathcal{F}$ is uniformly bounded and equicontinuous (pointwise bounded and equicontinuous is enough). Equivalently: every sequence in a uniformly bounded, equicontinuous family has a uniformly convergent subsequence.

Working tests for equicontinuity: a common Lipschitz constant, $|f(x)-f(y)|\le L|x-y|$ for all $f\in\mathcal{F}$ — for instance $|f'|\le L$ on an interval; a common Hölder bound $|f(x)-f(y)|\le L|x-y|^\alpha$; or $\mathcal{F}=\{\int_a^bk(x,t)g(t)\,dt:\|g\|_\infty\le1\}$ with $k$ continuous on $[a,b]^2$.

Proof (sketch; full proof in 📚 Go deeper). Sufficiency. (i) $K$ has a countable dense set $\{q_1,q_2,\dots\}$: the union of finite $\frac1m$-nets, $m=1,2,\dots$ (Theorem 4.8). (ii) Given a sequence $(f_n)$ in $\mathcal{F}$, the numbers $f_n(q_1)$ are bounded, so a subsequence converges at $q_1$ (Bolzano–Weierstrass); a further subsequence converges at $q_2$; and so on. The diagonal subsequence $(g_k)$ converges at every $q_j$. (iii) Given $\varepsilon$, take $\delta$ from equicontinuity and a finite set $q_{j_1},\dots,q_{j_r}$ that is a $\delta$-net of $K$; choose $N$ with $|g_k(q_{j_i})-g_l(q_{j_i})|\lt\varepsilon$ for $k,l\ge N$ and $i\le r$. For any $x$, pick $q=q_{j_i}$ with $d(x,q)\lt\delta$:

$$|g_k(x)-g_l(x)|\le|g_k(x)-g_k(q)|+|g_k(q)-g_l(q)|+|g_l(q)-g_l(x)|\lt3\varepsilon.$$

So $(g_k)$ is uniformly Cauchy and converges in the complete space $C(K)$ (§4.1). Necessity. A compact set is bounded and totally bounded; if $f_1,\dots,f_r$ is a finite $\frac\varepsilon3$-net of $\mathcal{F}$, each $f_i$ is uniformly continuous (Heine), so a single $\delta$ works for all $f_i$ with $\frac\varepsilon3$, and the $\frac\varepsilon3$ argument transfers it to every $f\in\mathcal{F}$. ∎

Ascoli–Arzelà is the compactness theorem of analysis: it proves Peano's existence theorem for $y'=F(x,y)$ with $F$ merely continuous, shows that integral operators with continuous kernels are compact operators (§8.4), and gives the existence of minimisers in the calculus of variations.

📘 Definition 4.13 — Bernstein polynomials
For $f:[0,1]\to\mathbb{R}$ and $n\ge1$, the $n$-th Bernstein polynomial of $f$ is $$B_nf(x)=\sum_{k=0}^nf\!\left(\tfrac kn\right)\binom nk x^k(1-x)^{n-k}.$$ The weights $p_{n,k}(x)=\binom nkx^k(1-x)^{n-k}$ are the probabilities of $k$ successes in $n$ independent trials with success probability $x$.
📐 Theorem 4.20 — Weierstrass approximation theorem; Stone–Weierstrass
  1. Weierstrass (Bernstein's proof). For every $f\in C[0,1]$, $B_nf\rightrightarrows f$ on $[0,1]$. Hence for every $f\in C[a,b]$ and $\varepsilon\gt0$ there is a polynomial $p$ with $\|f-p\|_\infty\lt\varepsilon$: the polynomials are dense in $(C[a,b],d_\infty)$.
  2. Stone–Weierstrass (statement). If $K$ is a compact metric space and $\mathcal{A}\subseteq C(K)$ is a subalgebra (closed under sums, products and scalar multiples) that contains the constants and separates points (for $x\ne y$ some $g\in\mathcal{A}$ has $g(x)\ne g(y)$), then $\mathcal{A}$ is dense in $C(K)$. Consequences: polynomials in $x_1,\dots,x_n$ are dense in $C(K)$ for compact $K\subseteq\mathbb{R}^n$; trigonometric polynomials $a_0+\sum_{k=1}^N(a_k\cos kx+b_k\sin kx)$ are dense (for $d_\infty$) in the continuous $2\pi$-periodic functions.
  3. Moments. If $f\in C[a,b]$ and $\int_a^bf(x)x^n\,dx=0$ for all $n\ge0$, then $f=0$.

Proof. (1) Moment identities. From the binomial theorem, for every $x$:

$$\sum_kp_{n,k}=1,\quad \sum_k\tfrac kn\,p_{n,k}=x,$$ $$\sum_k\left(\tfrac kn-x\right)^2p_{n,k}=\frac{x(1-x)}n.$$

(The first is $(x+1-x)^n=1$; the others are the mean $nx$ and variance $nx(1-x)$ of the binomial distribution, or follow by differentiating $(x+y)^n$ with respect to $x$ once and twice and putting $y=1-x$.) In particular $B_n1=1$, $B_nx=x$ and $B_n(x^2)=x^2+\frac{x(1-x)}n$.

The estimate. Let $\varepsilon\gt0$ and $M=\|f\|_\infty$. By Heine's theorem $f$ is uniformly continuous: there is $\delta$ with $|f(x)-f(y)|\lt\varepsilon$ when $|x-y|\lt\delta$. Using $\sum p_{n,k}=1$,

$$|f(x)-B_nf(x)|\le\sum_k\left|f(x)-f(\tfrac kn)\right|p_{n,k}(x).$$

Split the sum. Terms with $|\frac kn-x|\lt\delta$ contribute less than $\varepsilon\sum p_{n,k}=\varepsilon$. For the others $|f(x)-f(\frac kn)|\le2M\le2M\frac{(k/n-x)^2}{\delta^2}$, so they contribute at most

$$\frac{2M}{\delta^2}\cdot\frac{x(1-x)}n\le\frac{M}{2n\delta^2}.$$

Hence $\|f-B_nf\|_\infty\le\varepsilon+\frac M{2n\delta^2}\lt2\varepsilon$ for $n$ large. For $[a,b]$ apply this to $f(a+(b-a)t)$ and substitute back.

(3) Choose polynomials $p_k\rightrightarrows f$. Then $\int_a^bfp_k=0$ for each $k$, and $fp_k\rightrightarrows f^2$ (since $f$ is bounded), so $\int_a^bf^2=\lim\int_a^bfp_k=0$; $f^2$ is continuous and non-negative, so $f=0$. Part (2) is proved in 📚 Go deeper. ∎

Weierstrass's theorem is about compact intervals. On $\mathbb{R}$ a uniform limit of polynomials is itself a polynomial (Exercise IV.7), so $e^x$ and $\sin x$ are not uniform limits of polynomials on $\mathbb{R}$; on $(0,1]$ a uniform limit of polynomials is bounded, so $\frac1x$ is not one. And approximation is not expansion: the Weierstrass polynomials for $|x|$ are not partial sums of any power series, and $e^{-1/x^2}$ (with value $0$ at $0$) is uniformly approximable on $[-1,1]$ although its Taylor series at $0$ is identically zero.

✏️ Example 7 — Compact families and polynomial approximation
(a) Which of these subsets of $(C[0,1],d_\infty)$ have compact closure: $\mathcal{F}_1=\{x^n:n\ge1\}$; $\mathcal{F}_2=\{\sin(x+n):n\ge1\}$; $\mathcal{F}_3=\{f\in C^1[0,1]:|f(0)|\le1,\ |f'|\le1\}$; $\mathcal{F}_4=\{x+n:n\ge1\}$? (b) Compute $B_n(x^2)$ and $\|B_n(x^2)-x^2\|_\infty$; how large must $n$ be for an error below $0.01$? (c) Is the set of even polynomials dense in $C[0,1]$? In $C[-1,1]$?
(a) $\mathcal{F}_1$: no Uniformly bounded by $1$, but not equicontinuous at $1$: for any $\delta$, $x=1-\frac\delta2$ gives $|1^n-x^n|\to1$. Directly, a uniformly convergent subsequence would have the discontinuous pointwise limit.
$\mathcal{F}_2$: yes Bounded by $1$, and $|\frac{d}{dx}\sin(x+n)|\le1$ gives a common Lipschitz constant. Ascoli–Arzelà.
$\mathcal{F}_3$: yes $|f(x)|\le|f(0)|+|x|\le2$ and $|f(x)-f(y)|\le|x-y|$. (Its closure contains non-differentiable functions such as $|x-\frac12|$: the closure is compact, the set itself is not closed.)
$\mathcal{F}_4$: no Equicontinuous (all $1$-Lipschitz) but not bounded.
(b) $n\ge26$ By the moment identities $B_n(x^2)=x^2+\frac{x(1-x)}n$, so the error is $\frac{x(1-x)}n$, with maximum $\frac1{4n}$ at $x=\frac12$. $\frac1{4n}\lt0.01$ iff $n\gt25$. The error decays only like $\frac1n$ even for this very smooth function — Bernstein polynomials are a proof device, not an efficient approximation method.
(c) $[0,1]$: yes; $[-1,1]$: no On $[0,1]$ the even polynomials are the polynomials in $x^2$; they form an algebra with the constants, and $x^2$ separates the points of $[0,1]$ — Stone–Weierstrass. (Directly: approximate $g(t)=f(\sqrt t)$ by a polynomial $q(t)$ and use $q(x^2)$.) On $[-1,1]$ every even polynomial satisfies $p(-x)=p(x)$, a property preserved by uniform limits; $f(x)=x$ is not even, and indeed $\|x-p\|_\infty\ge1$ for every even $p$ (compare $x=\pm1$).
⚠️ Counterexamples and pitfalls
  1. A pointwise limit of continuous functions need not be continuous: $x^n$ on $[0,1]$; $\sum_{n\ge0}\frac{x^2}{(1+x^2)^n}$ on $\mathbb{R}$.
  2. A continuous limit does not make the convergence uniform: $\frac{nx}{1+n^2x^2}\to0$ with $M_n=\frac12$.
  3. $\lim\int\ne\int\lim$ without uniformity: $2nxe^{-nx^2}$ (integrals $\to1$, limit $0$); tents of height $n$. But uniform convergence is only sufficient: $\frac{nx}{1+n^2x^2}$ is not uniform and still $\int\to0$.
  4. Uniform convergence of $f_n$ says nothing about $f_n'$: $\frac{\sin nx}{\sqrt n}\rightrightarrows0$ with divergent derivatives; $\sqrt{x^2+\frac1n}\rightrightarrows|x|$, not differentiable.
  5. Dini needs all four hypotheses — compact domain, continuous $f_n$, continuous limit, monotone in $n$.
  6. The M-test is sufficient, not necessary: $\sum\frac{(-1)^n}{n+x^2}$ converges uniformly on $\mathbb{R}$ but not absolutely anywhere.
  7. Uniform on every bounded interval $\ne$ uniform on $\mathbb{R}$: $\frac xn$; $\sum\frac{x^n}{n!}$; $(1+\frac xn)^n\to e^x$.
  8. Products of uniformly convergent sequences need not converge uniformly unless the functions are bounded ($x+\frac1n$ times $x$).
  9. Radius of convergence: the ratio test can fail where the root test works ($\sum(2+(-1)^n)^nx^n$, $R=\frac13$); with gaps substitute ($\sum\frac{x^{2n}}{4^n}$ has $R=2$).
  10. The endpoints are not decided by $R$, and convergence on $(-R,R)$ need not be uniform ($\sum x^n$). Differentiation can lose an endpoint ($\sum\frac{x^n}{n^2}\to\sum\frac{x^{n-1}}n$ at $x=1$).
  11. The converse of Abel's theorem is false: $\sum(-1)^nx^n\to\frac12$ as $x\to1^-$, but $\sum(-1)^n$ diverges.
  12. Bounded is not enough in $C[0,1]$: $\{x^n\}$, $\{\sin nx\}$ are bounded with no uniformly convergent subsequence — equicontinuity fails.
  13. Weierstrass needs a compact interval: no sequence of polynomials converges uniformly to $e^x$ on $\mathbb{R}$ or to $\frac1x$ on $(0,1]$; and it is not a statement about Taylor series.
📚 Go deeper — full proofs in the CSIR NET course

Two Labs

Interactive 1 — The ε-Tube and the Sup-Norm Gap

The red curve is $f_n$, the green curve is the pointwise limit $f$, and the shaded band is the $\varepsilon$-tube around $f$. The amber segment marks where $|f_n-f|$ is largest; its length is $M_n=\sup|f_n-f|$. Uniform convergence means: for every $\varepsilon$, from some $n$ on the whole red graph stays inside the band. Press Play to increase $n$, and shrink $\varepsilon$. The last readout compares $\int f_n$ with $\int f$.

n, ε—
Mₙ = sup |fₙ − f|—
attained at x =—
∫fₙ vs ∫f—
Interactive 2 — Weierstrass by Bernstein Polynomials

The black curve is $f$, the red curve is its Bernstein polynomial $B_nf$, and (for small $n$) the blue dots are the sampled values $f(\frac kn)$ that $B_nf$ averages with binomial weights. The shaded area shows the error $|B_nf-f|$. Press Play to raise the degree and watch the sup error fall — fast for smooth functions, slowly at a corner or a steep rise.

degree n—
sup |Bₙf − f|—
error at x = 1/2—
1/√n for scale—

Exercises

I
Pointwise and uniform convergence of sequences.
  • 1 $f_n(x)=\dfrac x{n+x}$. Is the convergence uniform on $[0,1]$? On $[0,\infty)$?
  • 2 $f_n(x)=\dfrac{nx}{1+nx}$ on $[0,1]$: find the pointwise limit and decide uniformity on $[0,1]$ and on $[a,1]$, $0\lt a\lt1$.
  • 3 Show that $f_n(x)=x^n-x^{2n}$ converges pointwise but not uniformly on $[0,1]$.
  • 4 Show that $\sqrt{x^2+\frac1{n^2}}\rightrightarrows|x|$ on $\mathbb{R}$. What does this say about uniform limits of differentiable functions?
  • 5 Show that $\left(1+\frac xn\right)^n\to e^x$ uniformly on $[0,1]$ but not on $[0,\infty)$.
  • 6 $f_n(x)=n^2x(1-x^2)^n$ on $[0,1]$. Find the pointwise limit and $\int_0^1f_n$. Is the convergence uniform?
  • 7 Show that if $f_n\rightrightarrows f$ on $D$ and $g$ is bounded on $D$, then $gf_n\rightrightarrows gf$. Show that boundedness of $g$ is needed.
  • 8 Give a sequence that converges uniformly on every bounded interval but not on $\mathbb{R}$.
  • 9 Show that $x^{1/n}$ converges uniformly on $[a,1]$ for every $0\lt a\lt1$, but not on $[0,1]$.
1 Limit $0$. On $[0,1]$, $f_n$ is increasing, so $M_n=f_n(1)=\frac1{n+1}\to0$: uniform. On $[0,\infty)$, $f_n(x)\to1$ as $x\to\infty$, so $M_n=1$ (or take $x_n=n$: $f_n(n)=\frac12$): not uniform.
2 $f_n(0)=0$, $f_n(x)\to1$ for $x\gt0$: the limit is discontinuous at $0$, so not uniform on $[0,1]$. On $[a,1]$, $|1-f_n(x)|=\frac1{1+nx}\le\frac1{1+na}\to0$: uniform.
3 The limit is $0$ on $[0,1]$ (both terms $\to0$ for $x\lt1$, and $1-1=0$ at $x=1$). With $t=x^n\in[0,1]$, $f_n=t-t^2$, maximal at $t=\frac12$: $M_n=\frac14$ for every $n$.
4 $0\le\sqrt{x^2+\frac1{n^2}}-|x|=\dfrac{1/n^2}{\sqrt{x^2+1/n^2}+|x|}\le\dfrac{1/n^2}{1/n}=\dfrac1n$. Each $f_n$ is differentiable, the uniform limit $|x|$ is not: uniform convergence does not preserve differentiability (Theorem 4.17 needs the derivatives to converge uniformly; here $f_n'(x)\to\operatorname{sgn}x$, not uniformly).
5 For $x\ge0$, $(1+\frac xn)^n$ increases with $n$ (AM–GM) to $e^x$; on the compact $[0,1]$ the functions and the limit are continuous, so Dini's theorem gives uniform convergence. On $[0,\infty)$, at $x=n$: $e^n-2^n\to\infty$.
6 $f_n(0)=f_n(1)=0$ and for $0\lt x\lt1$, $n^2(1-x^2)^n\to0$: limit $0$. But $\int_0^1f_n=n^2\Big[-\frac{(1-x^2)^{n+1}}{2(n+1)}\Big]_0^1=\frac{n^2}{2(n+1)}\to\infty$. Not uniform (otherwise the integrals would tend to $0$).
7 $|gf_n-gf|\le\|g\|\,\|f_n-f\|$. Needed: $f_n\equiv\frac1n\rightrightarrows0$ on $\mathbb{R}$ and $g(x)=x$; $gf_n=\frac xn$ is not uniformly convergent on $\mathbb{R}$.
8 $f_n(x)=\frac xn$: $\sup_{[-a,a]}|f_n|=\frac an\to0$, but $\sup_{\mathbb{R}}|f_n|=\infty$. (Also the partial sums of $e^x$.)
9 For $x\in(0,1]$, $x^{1/n}=e^{(\log x)/n}\to1$, increasing in $n$; on $[a,1]$ the limit $1$ is continuous and the interval compact: Dini (or $M_n=1-a^{1/n}\to0$). On $[0,1]$ the limit is $0$ at $x=0$ and $1$ elsewhere — discontinuous, so not uniform.
II
Series of functions; interchanging limits.
  • 1 Show that $\sum_{n\ge1}\frac{\cos nx}{n^2+1}$ converges uniformly on $\mathbb{R}$ to a continuous function.
  • 2 Find the sum of $\sum_{n\ge0}x^n(1-x)$ on $[0,1]$. Is the convergence uniform on $[0,1]$? On $[0,a]$, $a\lt1$?
  • 3 Show that $\sum_{n\ge1}\frac{(-1)^{n+1}x^n}n$ converges uniformly on $[0,1]$. Can the M-test prove this?
  • 4 Show that $F(x)=\sum_{n\ge1}\frac1{n^2+x^2}$ is differentiable on $\mathbb{R}$ and give $F'$ as a series.
  • 5 Show that $\sum_{n\ge1}n^{-x}$ converges uniformly on $[a,\infty)$ for every $a\gt1$, but not on $(1,\infty)$.
  • 6 Show that $\sum_{n\ge1}\frac{x}{n(1+nx^2)}$ converges uniformly on $\mathbb{R}$.
  • 7 Show that $\sum_{n\ge1}e^{-nx}$ converges uniformly on $[a,\infty)$ for $a\gt0$ but not on $(0,\infty)$, and find its sum.
  • 8 Evaluate $\displaystyle\int_0^1\sum_{n\ge1}\frac{x^n}{n^2}\,dx$ in closed form, justifying the interchange.
1 M-test with $M_n=\frac1{n^2+1}$; each term is continuous, so the sum is continuous (Theorem 4.16(1)).
2 The partial sums telescope: $\sum_{n=0}^Nx^n(1-x)=1-x^{N+1}$. Sum: $1$ on $[0,1)$ and $0$ at $x=1$ — discontinuous, so not uniform on $[0,1]$. On $[0,a]$, $\sup|1-S_N|=a^{N+1}\to0$: uniform.
3 $b_n(x)=\frac{x^n}n$ decreases in $n$ for $x\in[0,1]$ and $\|b_n\|=\frac1n\to0$: by the alternating form of Theorem 4.14(2) the tail after $N$ terms is at most $\frac1{N+1}$ for every $x$. The M-test fails, since $\sup_{[0,1]}|u_n|=\frac1n$ and $\sum\frac1n$ diverges.
4 M-test with $\frac1{n^2}$ for the series. Differentiated series: $\sum\frac{-2x}{(n^2+x^2)^2}$; the maximum of $\frac{2|x|}{(n^2+x^2)^2}$ is at $x^2=\frac{n^2}3$ and equals $\frac{9}{8\sqrt3\,n^3}$, summable. By Theorem 4.17, $F'(x)=-\sum_{n\ge1}\frac{2x}{(n^2+x^2)^2}$.
5 On $[a,\infty)$: $n^{-x}\le n^{-a}$ and $\sum n^{-a}\lt\infty$. On $(1,\infty)$: the terms are continuous on $[1,\infty)$; uniform convergence on $(1,\infty)$ would give uniform convergence on $[1,\infty)$ (Theorem 4.13(3c)), hence convergence of $\sum\frac1n$ — false.
6 $\frac{|x|}{1+nx^2}$ is maximal at $x=\frac1{\sqrt n}$ with value $\frac1{2\sqrt n}$, so $|u_n(x)|\le\frac1{2n^{3/2}}$: M-test.
7 On $[a,\infty)$: $e^{-nx}\le e^{-na}$, a convergent geometric series. On $(0,\infty)$: $\sup e^{-nx}=1\not\to0$ (Theorem 4.13(3b)). Sum: $\frac{e^{-x}}{1-e^{-x}}=\frac1{e^x-1}$.
8 On $[0,1]$, $\left|\frac{x^n}{n^2}\right|\le\frac1{n^2}$: uniform, integrate term by term: $\sum_{n\ge1}\frac1{n^2(n+1)}$. Since $\frac1{n^2(n+1)}=\frac1{n^2}-\frac1n+\frac1{n+1}$ and $\sum(\frac1{n+1}-\frac1n)=-1$, the value is $\frac{\pi^2}6-1\approx0.6449$.
III
Power series.
  • 1 Find the radius of convergence of $\sum\left(\frac n{n+1}\right)^{n^2}x^n$.
  • 2 Find the interval of convergence of $\sum_{n\ge1}\frac{x^n}{n\,3^n}$.
  • 3 Find the interval of convergence of $\sum_{n\ge1}\frac{(3x-2)^n}{n^2}$.
  • 4 Find the radii of $\sum n!\,x^n$ and $\sum\frac{x^n}{n!}$.
  • 5 Find the radius of convergence of $\sum_{n\ge0}2^nx^{3n}$.
  • 6 Show that $\sum_{n\ge1}nx^n=\frac x{(1-x)^2}$ for $|x|\lt1$ and evaluate $\sum_{n\ge1}\frac n{3^n}$.
  • 7 Show that $\log(1+x)=\sum_{n\ge1}\frac{(-1)^{n+1}x^n}n$ for $|x|\lt1$, and deduce $1-\frac12+\frac13-\cdots=\log2$.
  • 8 Let $f(x)=\sum_{n\ge1}\frac{x^n}{n^2}$. Show that $f$ is continuous on $[-1,1]$ and differentiable on $(-1,1)$, and that $f'$ is unbounded on $(0,1)$.
  • 9 Find the Maclaurin series of $\arctan x$ and its radius. Why is the radius $1$ although $\arctan$ is infinitely differentiable on $\mathbb{R}$?
1 $|a_n|^{1/n}=\left(\frac n{n+1}\right)^n=\left(1+\frac1n\right)^{-n}\to\frac1e$, so $R=e$.
2 $R=3$. At $x=3$: $\sum\frac1n$ diverges; at $x=-3$: $\sum\frac{(-1)^n}n$ converges. Interval $[-3,3)$.
3 Converges iff $|3x-2|\le1$ (absolutely at the endpoints, by $\sum\frac1{n^2}$): $x\in[\frac13,1]$. Centre $\frac23$, radius $\frac13$.
4 $\frac{(n+1)!}{n!}=n+1\to\infty$: $R=0$ (converges only at $0$). $\frac{1/(n+1)!}{1/n!}\to0$: $R=\infty$.
5 As a series in $y=x^3$, $\sum(2y)^n$ converges iff $|y|\lt\frac12$, i.e. $|x|\lt2^{-1/3}\approx0.794$. (Root test: the nonzero coefficients are $a_{3n}=2^n$, and $(2^n)^{1/(3n)}=2^{1/3}$.)
6 Differentiate $\sum x^n=\frac1{1-x}$ term by term (Theorem 4.18(3)) and multiply by $x$. At $x=\frac13$: $\frac{1/3}{(2/3)^2}=\frac34$.
7 Integrate $\frac1{1+t}=\sum_{n\ge0}(-t)^n$ from $0$ to $x$, $|x|\lt1$. At $x=1$ the series converges (alternating), so Abel's theorem gives its sum as $\lim_{x\to1^-}\log(1+x)=\log2$. (Exercise II.3 shows the convergence is even uniform on $[0,1]$.)
8 $|x^n/n^2|\le\frac1{n^2}$ on $[-1,1]$: uniform, so $f$ is continuous there. On $(-1,1)$, $f'(x)=\sum_{n\ge1}\frac{x^{n-1}}n=-\frac{\log(1-x)}x$ for $x\ne0$, which tends to $+\infty$ as $x\to1^-$. The differentiated series has the same radius $1$ but diverges at $x=1$.
9 $\arctan x=\int_0^x\frac{dt}{1+t^2}=\sum_{n\ge0}\frac{(-1)^nx^{2n+1}}{2n+1}$, radius $1$ (the terms grow for $|x|\gt1$). The obstruction is complex: $\frac1{1+z^2}$ has poles at $z=\pm i$, at distance $1$ from $0$, and a power series converges in a disc (Chapter 5).
IV
Ascoli–Arzelà and Weierstrass approximation.
  • 1 Is $\{x^n:n\ge1\}$ equicontinuous on $[0,1]$? On $[0,\frac12]$?
  • 2 Show that $(\sin nx)$ has no uniformly convergent subsequence on $[0,\pi]$. (Hint: $\int_0^\pi(\sin mx-\sin nx)^2dx$.)
  • 3 Show that $\mathcal{F}=\{f\in C[0,1]:f(0)=0,\ |f(x)-f(y)|\le|x-y|\}$ is compact in $(C[0,1],d_\infty)$.
  • 4 Let $f_n\in C^1[0,1]$ with $|f_n(0)|\le1$ and $\int_0^1|f_n'|^2\le1$. Show that $(f_n)$ has a uniformly convergent subsequence.
  • 5 Compute $B_n(x)$ and $B_n(x^2)$ directly from the definition for $n=2$, and check them against the moment identities.
  • 6 Let $f\in C[0,1]$ with $\int_0^1f(x)x^n\,dx=0$ for all $n\ge0$. Prove $f=0$. What if only $\int_0^1f(x)x^{2n}dx=0$ for all $n$?
  • 7 Prove that if polynomials $p_n$ converge uniformly on $\mathbb{R}$ to $f$, then $f$ is a polynomial.
  • 8 Is there a sequence of polynomials converging uniformly to $\sqrt x$ on $[0,1]$? To $\frac1{1+x}$ on $[0,1]$? To $\frac1x$ on $(0,1]$?
  • 9 Modelling (computer-aided design). A car-body designer approximates a crease profile $y=|x-\frac12|$ metres, $0\le x\le1$, by the Bernstein (Bézier) curve $B_nf$. Popoviciu's estimate $\|B_nf-f\|_\infty\le\frac32\frac L{\sqrt n}$ for an $L$-Lipschitz $f$ may be assumed. Which degree does it guarantee for an error at most $1$ cm? The actual error is largest at the crease and equals $\mathbb{E}\left|\frac Kn-\frac12\right|$ with $K\sim\mathrm{Bin}(n,\frac12)$; it is about $\frac1{\sqrt{2\pi n}}$. Which degree is really needed?
1 On $[0,1]$: no — near $1$, $|1-x^n|\to1$ for $x=1-\frac\delta2$ however small $\delta$ is. On $[0,\frac12]$: yes, $|(x^n)'|=nx^{n-1}\le n2^{1-n}\le1$, a common Lipschitz constant (and in fact $x^n\rightrightarrows0$ there).
2 For integers $m\ne n\ge1$, $\int_0^\pi\sin^2mx\,dx=\frac\pi2$ and $\int_0^\pi\sin mx\sin nx\,dx=0$, so $\int_0^\pi(\sin mx-\sin nx)^2=\pi$. A uniformly convergent subsequence would be uniformly Cauchy, and then $\int_0^\pi(\sin n_kx-\sin n_lx)^2\le\pi\|\sin n_kx-\sin n_lx\|_\infty^2\to0$ — contradiction.
3 $|f(x)|=|f(x)-f(0)|\le1$: uniformly bounded; common Lipschitz constant $1$: equicontinuous; and closed, because $f(0)=0$ and the Lipschitz inequality survive uniform (even pointwise) limits. By Ascoli–Arzelà the closure, which is $\mathcal{F}$, is compact.
4 By Cauchy–Schwarz, $|f_n(x)-f_n(y)|=\left|\int_y^xf_n'\right|\le|x-y|^{1/2}\left(\int_0^1|f_n'|^2\right)^{1/2}\le|x-y|^{1/2}$: a common Hölder bound, so equicontinuous; and $|f_n(x)|\le1+1=2$. Ascoli–Arzelà.
5 $n=2$: weights $(1-x)^2,\ 2x(1-x),\ x^2$ at the nodes $0,\frac12,1$. $B_2(x)=0+\frac12\cdot2x(1-x)+x^2=x$. $B_2(x^2)=0+\frac14\cdot2x(1-x)+x^2=x^2+\frac{x(1-x)}2$, matching $x^2+\frac{x(1-x)}n$.
6 Theorem 4.20(3). With only even moments: $\int_0^1f(x)q(x^2)\,dx=0$ for every polynomial $q$, and the polynomials in $x^2$ are dense in $C[0,1]$ (Example 7(c)); the same argument gives $\int_0^1f^2=0$, so again $f=0$.
7 Take $N$ with $\|p_n-f\|\lt\frac12$ for $n\ge N$; then $p_n-p_N$ is a polynomial bounded by $1$ on $\mathbb{R}$, hence constant: $p_n=p_N+c_n$. The constants $c_n=p_n(0)-p_N(0)$ converge to some $c$, and $f=p_N+c$.
8 Yes and yes: both are continuous on the compact $[0,1]$ (Weierstrass). No for $\frac1x$: if $\|p-\frac1x\|_{(0,1]}\lt1$ then $\frac1x\lt|p(x)|+1\le\max_{[0,1]}|p|+1$ on $(0,1]$, impossible.
9 $L=1$: $\frac3{2\sqrt n}\le0.01$ needs $\sqrt n\ge150$, $n\ge22\,500$. The true error $\approx\frac1{\sqrt{2\pi n}}\le0.01$ needs $n\gtrsim\frac{10^4}{2\pi}\approx1592$; for example at $n=100$ it is $0.0398$ m $\approx4$ cm, against the guarantee $0.15$ m. The worst-case estimate is pessimistic by a factor of about $14$ in degree — and either way the corner forces a high degree, which is why designers use piecewise (spline) curves.
🎯 GATE corner GATE MA · MCQGATE MA · MSQGATE MA · NAT

E1. GATE MA · MCQ Let $f_n(x)=\dfrac{x}{1+nx^2}$ for $x\in\mathbb{R}$. Then the sequence $(f_n)$ (A) converges uniformly to $0$ on $\mathbb{R}$; (B) converges pointwise to $0$ on $\mathbb{R}$, but not uniformly; (C) converges uniformly to $0$ on every bounded interval, but not on $\mathbb{R}$; (D) does not converge at $x=0$.

Limit $f_n(0)=0$ and $|f_n(x)|\le\frac1{n|x|}\to0$ for $x\ne0$; (D) is false.
Sup $f_n'(x)=\frac{1-nx^2}{(1+nx^2)^2}$ vanishes at $x=\pm\frac1{\sqrt n}$, so $\sup_{\mathbb{R}}|f_n|=\frac1{2\sqrt n}\to0$ (Example 1(c)).
Answer (A). (B) and (C) describe the trap in which the maximum point escapes to infinity — here it moves towards $0$ and the maximum value shrinks.

E2. GATE MA · MCQ The radius of convergence of $\displaystyle\sum_{n\ge1}\frac{(2n)!}{(n!)^2}x^n$ is (A) $\frac14$; (B) $4$; (C) $\frac12$; (D) $2$.

Ratio $\dfrac{a_{n+1}}{a_n}=\dfrac{(2n+2)(2n+1)}{(n+1)^2}\to4$.
Answer $R=\frac14$, option (A). (Choosing $4$ confuses the limit of the ratio with $R=\frac1L$.)

E3. GATE MA · MSQ Which of the following series converge(s) uniformly on the given set? (One or more options may be correct.) (A) $\sum_{n\ge1}\frac{\sin nx}{n^{3/2}}$ on $\mathbb{R}$; (B) $\sum_{n\ge0}x^n$ on $[0,1)$; (C) $\sum_{n\ge1}\frac{(-1)^n}{n+x^2}$ on $\mathbb{R}$; (D) $\sum_{n\ge0}\frac{x^2}{(1+x^2)^n}$ on $[-1,1]$.

(A) Yes M-test, $M_n=n^{-3/2}$.
(B) No $\sup_{[0,1)}x^n=1\not\to0$.
(C) Yes Alternating Dirichlet test: $\frac1{n+x^2}\downarrow0$ uniformly (bounded by $\frac1n$).
(D) No The sum is $1+x^2$ for $x\ne0$ and $0$ at $x=0$: a discontinuous sum of continuous terms (Example 2(e)).
Answer (A) and (C).

E4. GATE MA · MSQ Let $f_n:[0,1]\to\mathbb{R}$ be continuous. Which of the following is/are true? (One or more options may be correct.) (A) If $f_n\rightrightarrows f$ on $[0,1]$, then $\int_0^1f_n\to\int_0^1f$. (B) If $f_n\to f$ pointwise and $|f_n|\le1$ for all $n$, then $f$ is continuous. (C) If $|f_n|\le1$ and $|f_n(x)-f_n(y)|\le|x-y|$ for all $n,x,y$, then some subsequence of $(f_n)$ converges uniformly. (D) If $f_n\rightrightarrows f$ and every $f_n$ is differentiable on $(0,1)$, then $f$ is differentiable on $(0,1)$.

(A) True Theorem 4.16(3): $|\int f_n-\int f|\le\|f_n-f\|_\infty$.
(B) False $x^n$ is bounded by $1$ with a discontinuous limit.
(C) True Uniformly bounded and equicontinuous (common Lipschitz constant): Ascoli–Arzelà.
(D) False $f_n(x)=\sqrt{(x-\frac12)^2+\frac1n}\rightrightarrows|x-\frac12|$, not differentiable at $\frac12$.
Answer (A) and (C).

E5. GATE MA · NAT The value of $\displaystyle\lim_{n\to\infty}\int_0^1n\,x\,(1-x^2)^n\,dx$, correct to two decimal places, is ______.

Trap The integrand tends to $0$ at every $x\in[0,1]$, but the convergence is not uniform (the maximum, near $x=\frac1{\sqrt{2n}}$, grows like $\sqrt n$), so $\lim\int=\int\lim=0$ cannot be assumed.
Compute With $u=1-x^2$: $\int_0^1nx(1-x^2)^ndx=\frac n2\int_0^1u^ndu=\frac n{2(n+1)}\to\frac12$.
Answer $0.50$ (accept $0.49$ to $0.51$).

E6. GATE MA · NAT Let $f(x)=\left|x-\frac12\right|$ on $[0,1]$ and let $B_4f$ be its Bernstein polynomial of degree $4$. The value $B_4f(\frac12)$, correct to three decimal places, is ______.

Weights At $x=\frac12$, $p_{4,k}(\frac12)=\binom4k\frac1{16}$: $\frac1{16},\frac4{16},\frac6{16},\frac4{16},\frac1{16}$ for $k=0,\dots,4$.
Values $f(\frac k4)=\frac12,\frac14,0,\frac14,\frac12$.
Answer $B_4f(\frac12)=\frac{\frac12+1+0+1+\frac12}{16}=\frac3{16}=0.1875$, so $0.188$ (accept $0.187$ to $0.188$). This is also the sup error of $B_4f$, attained at the corner where $f(\frac12)=0$.
🧑‍🏫 For the teacher

Suggested plan — 3 lectures × 60 min.

  1. Uniform convergence. The three failures of pointwise convergence, Definitions 4.9–4.10, Theorem 4.13 and the strategy box; Example 1 at the board, then project Interactive 1 with the sliding bump $\frac{nx}{1+n^2x^2}$ and with $x^n(1-x)$ and ask the class to predict $M_n$ before it appears. Theorem 4.14 (prove the M-test, sketch summation by parts) and Dini's theorem with its proof; Example 2.
  2. What uniform convergence buys; power series. Theorems 4.16 and 4.17 with proofs; Examples 3 and 4 (vote on each part of Example 4 first). Definition 4.11, Theorem 4.18 with the Cauchy–Hadamard proof, the endpoint table, Examples 5 and 6. GATE corner E2 and E5 as a timed pair.
  3. Compactness and approximation in $C[a,b]$. Definition 4.12 and the Ascoli–Arzelà theorem (the diagonal argument on the board), Example 7(a). Definition 4.13, the moment identities, and Bernstein's proof of Theorem 4.20; project Interactive 2 with $x^2$ and with $|x-\frac12|$ and compare the error rates. Finish with E3, E4 and E6.

Discussion questions. (1) Uniform convergence is convergence in the sup metric. Which metric does "convergence of the integrals" correspond to, and why does it not see the sliding bumps of Example 1(b)? (2) The Bernstein error for $x^2$ is exactly $\frac{x(1-x)}n$. Is $\frac1n$ the best rate any sequence of degree-$n$ polynomials can achieve for $x^2$? (3) Why does the Ascoli–Arzelà theorem need equicontinuity but not differentiability, and which of the families in this section are compact for reasons of smoothness?

Misconceptions to watch for. Concluding uniform convergence from a continuous limit; forgetting that the maximum of $|f_n-f|$ may sit at a point that moves with $n$; using "uniform convergence of $f_n$" to differentiate term by term; taking $R=L$ instead of $R=\frac1L$ in the ratio formula; reading a power series with gaps ($x^{2n}$) as if its coefficients were $a_n$; believing that the Weierstrass theorem is about Taylor polynomials.

Suggested homework. I(2, 3, 5, 6, 9), II(2, 3, 4, 5, 8), III(1, 2, 3, 5, 7, 8), IV(1, 2, 3, 6, 7, 9); GATE corner E1, E3, E5.

🔬 Research–Teaching NexusResearch-tutored · optional, about 60 minutes

Research question. When a square wave is approximated by the partial sums of its Fourier series, the partial sums overshoot the jump by a fixed percentage no matter how many terms are taken (the Gibbs phenomenon). How large is the overshoot, why does it not go away, and how can averaging remove it?

Why it matters. The Gibbs phenomenon appears as "ringing" next to sharp edges in compressed images and audio, in MRI scans reconstructed from Fourier data, and in digital filters. It is a vivid example of pointwise convergence that is not uniform, and of how the right averaging restores uniform convergence.

What to do.

  1. Let $f(x)=1$ on $(0,\pi)$ and $-1$ on $(-\pi,0)$, extended $2\pi$-periodically. Its Fourier series is $\frac4\pi\sum_{k\text{ odd}}\frac{\sin kx}{k}$. Plot the partial sums $S_N$ (terms up to $k=N$) for $N=11,51,201$ on $(0,\frac\pi2)$.
  2. Locate the first maximum of $S_N$ to the right of $0$ and its height. Show that it occurs at $x=\frac\pi{N+1}$ (differentiate $S_N$ and sum the cosines) and that its height tends to $\frac2\pi\int_0^\pi\frac{\sin t}t\,dt$. Evaluate this number.
  3. Explain, using Theorem 4.13, why the convergence of $S_N$ to $f$ cannot be uniform on $(0,\pi)$ — without computing anything.
  4. Replace $S_N$ by the Fejér (Cesàro) means $\sigma_N=\frac1{N+1}\sum_{n=0}^NS_n$ and repeat step 2. What happens to the overshoot? (Fejér's theorem: $\sigma_Nf\rightrightarrows f$ for every continuous periodic $f$ — another proof that trigonometric polynomials are dense.)

Tools: Python (numpy, matplotlib) or any graphing tool; a table of the sine integral $\operatorname{Si}(\pi)=\int_0^\pi\frac{\sin t}t\,dt$, or numerical integration.

Hand in: the plots; a table of the peak positions and heights for three values of $N$; the limit value with its derivation; one paragraph comparing partial sums and Fejér means.

Going further. Read about the Lanczos $\sigma$-factors used in signal processing, and about why the Fejér kernel is positive while the Dirichlet kernel is not (§8.2 shows that Fourier partial sums of continuous functions can even diverge).

Instructor note: the peak of $S_N$ sits at $x=\frac\pi{N+1}$ (for $N$ odd) with height $\approx1.1813$ for $N=11$, $1.1791$ for $N=51$, $1.1790$ for $N=201$, tending to $\frac2\pi\operatorname{Si}(\pi)\approx1.17898$ — an overshoot of about $8.95\%$ of the jump $2$. Non-uniformity follows from Theorem 4.13(3c) or from Theorem 4.16(1): the partial sums are continuous at $0$ while the limit jumps. The Fejér means never exceed $1$ (for $N=51$ the maximum is about $0.988$) because the Fejér kernel is a non-negative average.

🧪 Self-check quiz

0/5 correct

Q1. On $[0,1]$ the sequence $f_n(x)=x^n$

Q2. The radius of convergence of $\sum_{n\ge1}\dfrac{n^3}{5^n}x^n$ is

Q3. For which series does the Weierstrass M-test prove uniform convergence on all of $\mathbb{R}$?

Q4. Which family has compact closure in $(C[0,1],d_\infty)$?

Q5. The sup error $\|B_n(x^2)-x^2\|_\infty$ of the Bernstein polynomial of degree $n=50$ for $f(x)=x^2$ on $[0,1]$ is