Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Show that convergence of a sequence of power-series is equivalent to convergence of their respective coefficients.

More precisely, let $\mathbf{a}:=\left[a_{i,j}\right]_{i,j\in\mathbb{N_0}}$ be a real matrix, such as a stochastic matrix, such that

i) $\forall i\in\mathbb{N}_0,\space\mathbf{L}\leq\mathbf{a}_i$ for some real row vector $\mathbf{L}$ that sums to a real number ($\mathbf{a}_i$ being $\mathbf{a}$'s $i$th row),

ii) $\sup_{i\in\mathbb{N}_0} \sum \mathbf{a}_i<\infty$ (Given a real vector $\mathbf{v}=\left[v_m\right]_{m\in\mathbb{N}_0}$, we define $\sum\mathbf{v}:=\sum_{m=0}^\infty v_m$ whenever the series on the right converges, possibly to $\pm\infty$.)

Additionally, let $\mathbf{b}:=\left[b_j\right]_{j\in\mathbb{N}_0}$ be a real row vector that sums to a finite number and define $\mathbf{x}:=\left[x_j\right]_{j\in\mathbb{N}_0}$ to be the column vector of monomial functions: $$x_j:\left(-1,1\right)\rightarrow\left(-1,1\right),\space\space x_j\left(y\right):=y^j$$

Lastly, set $\mathbf{f}:=\mathbf{a}\mathbf{x}$, $F:=\mathbf{b}\mathbf{x}$ and $$\mathbf{S}:=\left\{S\subseteq\left(-1,1\right):\space S\mathrm{\, has\, an\, accumulation\, point\, }\in \left(-1,1\right)\right\}$$ So $\mathbf{f}$ is a column vector whose components are power series and $F$ is a power series, and the radii of convergence of all these power series are $\geq1$.

Show that $$\begin{align}\mathbf{a}_i\underset{i\rightarrow\infty}{\rightarrow}\mathbf{b} &\iff f_i\underset{i\rightarrow\infty}{\rightarrow}F\mathrm{\, on\, }\left(-1,1\right)\\ &\iff\exists S\in\mathbf{S},\space f_i\underset{i\rightarrow\infty}{\rightarrow}F\mathrm{\, on\, }S\end{align}$$


  1. We take $0^0$ to be $1$, so $x_0\equiv 1$ on $\left(-1, 1\right)$.

  2. Conditions $\left(i\right)$ and $\left(ii\right)$ above will be required for the application of Fatou's lemma in the proof. In $\left(ii\right)$, $\sup$ is used rather than the laxer $\limsup$ required by Fatou's lemma, to ensure that the $f_i$'s radii of convergence are $\geq1$.

  3. The claim is adapted from Klenke, where an equivalent claim is stated without proof (Lemma 3.6, $\left(i\right)\iff\left(iii\right)\iff\left(iv\right)$).

share|cite|improve this question
Convergence of a power series in what sense? – Qiaochu Yuan Sep 5 '12 at 16:38
@QiaochuYuan: Thanks for your interest. Pointwise convergence of a sequence of power series (in plural). – Evan Aad Sep 5 '12 at 16:44
Pointwise convergence where? – Qiaochu Yuan Sep 5 '12 at 18:49
@QiaochuYuan: Consider the following sequence of functions $$\mathbf{f}:=\left(f_i\right)_{i\in\mathbb{N}_0}$$ where $$f_i:\left[0,1\right]\rightarrow\left[0,1\right],\space\space f_i\left(y\right):=\sum_{j=0}^\infty a_{i,j}y^j$$ Each $f_i$ is a power series that is convergent on $\left[0,1\right]$. – Evan Aad Sep 5 '12 at 19:07
@QiaochuYuan: Assume that the sequence of functions $\mathbf{f}$ converges pointwise on $\left[0,1\right]$ to the function $$F:\left[0,1\right],\space\space F\left(y\right):=\sum_{k=0}^\infty b_k y^k$$ (which is also a power series convergent on $\left[0,1\right]$). Is it true that the coefficients of $f_0, f_1, \dots$ converge term-by-term to the corresponding coefficients of $F$: $$a_{i,0}\underset{i\rightarrow\infty}{\rightarrow}b_0, a_{i,1}\underset{i\rightarrow\infty}{\rightarrow}b_1, \dots$$ – Evan Aad Sep 5 '12 at 19:08
up vote 0 down vote accepted

The following proof was suggested to me by Prof. Klenke himself in an email. His original message follows the proof.

$\boxed{\mathbf{a}_i\underset{i\rightarrow\infty}{\rightarrow}\mathbf{b} \implies f_i\underset{i\rightarrow\infty}{\rightarrow}F\mathrm{\, on\, }\left(-1,1\right)}$ Suppose $\mathbf{a}_i\underset{i\rightarrow\infty}{\rightarrow}\mathbf{b}$. Let $r\in\left(0,1\right)$. We will show that $$f_i\rightarrow F\mathrm{\, on\,}(-r,r) \space\space\left(*\right)$$ Let $z\in\left(-r,r\right)$ and fix $\epsilon>0$. We proceed to define a number $N_{r,\epsilon}\in\mathbb{N}_{0}$, such that for all $N_{r,\epsilon}\leq i\in\mathbb{N}_{0}$, $|f_i(z)-F(z)|<\epsilon$, thus proving $\left(*\right)$. Choose $n_{r,\epsilon}\in\mathbb{N}_{0}$ such that $\sum_{j>n_{r,\epsilon}}r^j<\epsilon/2$. Choose $N_{r,\epsilon}\in\mathbb{N}_{0}$ such that for all $N_{r,\epsilon}\leq i$, $\left|a_{i,j}-b_i\right|<\frac{\epsilon/2}{n_{r,\epsilon}+1}$, $j\in\left\{ 0,1,2,\dots,n_{r,\epsilon}\right\} $. Then for all $N_{r,\epsilon}\leq i$, $|f_i(z)-F(z)|<\epsilon$.

$\boxed{\exists S\in\mathbf{S},\space f_i\underset{i\rightarrow\infty}{\rightarrow}F\mathrm{\, on\, }S\implies\mathbf{a}_i\underset{i\rightarrow\infty}{\rightarrow}\mathbf{b}}$ Let $S\in\mathbf{S}$ and suppose $f_i\underset{i\rightarrow\infty}{\rightarrow}F$ on $S$. It is left for the reader to verify that it suffices to demonstrate that every strictly ascending sequence of indices $\left(i_k\right)_{k\in\mathbb{N}_{0}},\, i_k\in\mathbb{N}_{0}$ has a strictly ascending subsequence $\left(i_{k_n}\right)_{n\in\mathbb{N}_{0}}$ such that $$\mathbf{a}_{i_{k_n}}\underset{n\rightarrow\infty}{\rightarrow}\mathbf{b}\space\space\left(**\right)$$

Therefore let $\left(i_k\right)_{k\in\mathbb{N}_{0}}$ be strictly ascending. By the Bolzano-Weierstrass theorem there is a srtictly ascending subsequence $\left(i_{k_n}\right)_{n\in\mathbb{N}_{0}}$ such that $a_{i_{k_n}, 0}$ converges to some limit $c_0\in\mathbb{R}$. Appealing to the same theorem again we can further assume w.l.g. that also $\left(a_{i_{k_n}, 1}\right)_{n\in\mathbb{N}_{0}}$ converges to some $c_1\in\mathbb{R}$. Iterating the argument recursively, we may assume w.l.g. that $\left(a_{i_{k_n}, j}\right)_{n\in\mathbb{N}_{0}}$ converges to some $c_j\in\mathbb{R}$ for every $j\in\mathbb{N}_{0}$. Defining the row vector $\mathbf{c}:=\left[c_j\right]_{j\in\mathbb{N}_0}$, we have $\mathbf{a}_{i_{k_n}}\underset{n\rightarrow\infty}{\rightarrow}\mathbf{c}$. Since $\forall i\in\mathbb{N}_0,\space\mathbf{L}\leq\mathbf{a}_i$, $-\infty<\sum\mathbf{L}\leq\sum\mathbf{c}$ and by Fatou's lemma $\sum\mathbf{c}\leq\limsup_{i\in\mathbb{N}_0} \sum\mathbf{a}_i<\infty$. So $\sum\mathbf{c}\in\mathbb{R}$.

By the first part we see that $f_{i_{k_n}}\underset{n\rightarrow\infty}{\rightarrow}F_\mathbf{c}$ on $\left(-1,1\right)$, where $F_\mathbf{c}$ is the power series with coefficient vector $\mathbf{c}$. Now we use the underlying assumption that $f_i\underset{i\rightarrow\infty}{\rightarrow}F$ on $S$ to conclude that $F_\mathbf{c}=F$ on $S$. By the uniqueness theorem of power series (cf. Baby Rudin, Theorem 8.5) we get $\mathbf{c}=\mathbf{b}$, and so $\left(**\right)$ holds. $\square$

Note Here's Prof. Klenke's original message to me.

I do think that the equivalence of $(i)$ and $(iii)$ in Lemma 3.6 is standard, but I do not have a reference at hand at the moment. Let’s try the following argument.

$(i)\implies(iii)$. Fix $r\in(0,1)$ and $\epsilon>0$. Choose $N$ such that $\sum_{k>N} r^k < \epsilon/2$. Choose $n$ such that $\mu_n(k)<\epsilon/(N+1)$ for $k\leq N$. Then $|\psi_n(z)-\psi(z)|<\epsilon$ for all $0<z<r$. Hence $\psi_n$ converges to $\psi$ pointwise on $(0,r)$ for all $0<r<1$ and hence pointwise on $(0,1)$. Finally $\psi_n(1)=1=\psi(1)$ hence this is trivial.

$(iii)\implies(i)$ By the Bolzano-Weierstrass theorem, there is a subsequence such that $\mu_{n_k}(0)$ converges to some limit $\nu(0)$. Taking a further subsequence, also $\mu_{n_k}(1)$ converges to some $\nu(1)$. Iterating the argument, we get a subsequence such that $\mu_{n_k}(i)$ converges to some $\nu(i)$ for every $(i)$. By Fatou’s lemma, we have $\sum_i \nu(i)\leq 1$. By the argument of part $(i)\implies(iii)$, we see that $\psi_n(z)$ converges to $\phi(z)$ [the generating function of $\nu$] for all $0<z<1$ and hence $\phi_z=\psi(z)$ for all $0<z<1$. By the uniqueness theorem of power series, we get $\mu=\nu$. This construction shows that every subsequence of $\mu_n$ has a subsequence that converges to $\mu$. Hence $\mu_n$ converges to $\mu$.

I hope the argument is clear.

With the best regards, Achim.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.