Absolute Uniform Convergence

Uniform Control of Absolute Tails

For a function series, convergence at every input does not by itself control how quickly its tails become small across the domain. Uniform convergence strengthens that control for the signed tails. Absolute uniform convergence asks for uniform control even after every term has been replaced by its absolute value. This stronger condition is useful because cancellation is no longer needed to make the tails small.

Let $E$ be a nonempty set and let $f_n:E\to\mathbb{R}$. The series of absolute values is $\sum_{n=1}^{\infty}|f_n(x)|$. Its partial sums are nondecreasing at each input, but their limit may depend on $x$, and convergence at each input need not be uniform across $E$.

Definition: The series $\sum_{n=1}^{\infty}f_n(x)$ is absolutely uniformly convergent on $E$ if the series of nonnegative functions $\sum_{n=1}^{\infty}|f_n(x)|$ converges uniformly on $E$. Equivalently, the sums of its absolute tails tend uniformly to zero: for every $\varepsilon>0$, there is an integer $N$ such that $$ \sum_{n=N+1}^{\infty}|f_n(x)|<\varepsilon \qquad\text{for every }x\in E. $$

The equivalence follows from the Uniform Cauchy Criterion for a Series. Applied to the nonnegative terms $|f_n|$, that criterion says their finite tails are uniformly small. Taking the limit of the finite tails as the upper index tends to infinity gives the displayed bound on the infinite tail. Conversely, a bound on each infinite tail bounds every finite tail, so the criterion gives uniform convergence of the series of absolute values.

The definition involves the sum of absolute values, not merely the fact that the original series converges absolutely at each input. In particular, it requires one index $N$ to control the absolute tail for every $x\in E$. This is the feature that makes absolute uniform convergence stronger than pointwise absolute convergence.

Consequences for the Original Series

The absolute tail also bounds the signed tail: for every $x\in E$ and integers $q>N$, $$ \left|\sum_{n=N+1}^{q}f_n(x)\right| \leq \sum_{n=N+1}^{q}|f_n(x)| \leq \sum_{n=N+1}^{\infty}|f_n(x)|. $$ Thus absolute uniform convergence gives the uniform Cauchy condition for the original series. The next result records the resulting convergence and its error estimate.

Theorem (Absolute Uniform Convergence Controls the Sum): Suppose $\sum_{n=1}^{\infty}f_n(x)$ is absolutely uniformly convergent on $E$. Then it converges uniformly to a sum $S:E\to\mathbb{R}$, and for every $N\geq0$, $$ |S(x)-S_N(x)|\leq\sum_{n=N+1}^{\infty}|f_n(x)| \qquad\text{for every }x\in E, $$ where $S_N(x)=\sum_{n=1}^{N}f_n(x)$ and $S_0(x)=0$.

Proof. Fix $x\in E$. Absolute uniform convergence implies convergence of $\sum_{n=1}^{\infty}|f_n(x)|$, so the numerical series $\sum_{n=1}^{\infty}f_n(x)$ converges absolutely and hence converges. Let its sum be $S(x)$. Given $\varepsilon>0$, choose $N$ such that $\sum_{n=N+1}^{\infty}|f_n(x)|<\varepsilon$ for every $x\in E$. For any $m>N$, the triangle inequality gives $$ |S_m(x)-S_N(x)| =\left|\sum_{n=N+1}^{m}f_n(x)\right| \leq\sum_{n=N+1}^{m}|f_n(x)| \leq\sum_{n=N+1}^{\infty}|f_n(x)| <\varepsilon. $$ Letting $m$ tend to infinity at each fixed $x$ yields $$ |S(x)-S_N(x)|\leq\sum_{n=N+1}^{\infty}|f_n(x)|. $$ The right-hand side is less than $\varepsilon$ for every $x$, so the error tends uniformly to zero. Therefore $S_N$ converges uniformly to $S$, as claimed. $\square$

The estimate is often more useful than the bare conclusion of uniform convergence: it gives a direct bound for the remainder after $N$ terms. The Weierstrass M-Test, established earlier in this course, is a common way to obtain such a bound. If $|f_n(x)|\leq M_n$ for all $x$ and $\sum M_n$ converges, then its uniform remainder bound controls the absolute tails as well.

Worked Example: A Power Series with a Uniform Absolute Tail

On $E=[0,1]$, let $f_n(x)=x^n/n^2$, for $n\geq1$. Since $0\leq x\leq1$, $$ |f_n(x)|=\frac{x^n}{n^2}\leq\frac{1}{n^2}. $$ The numerical series $\sum_{n=1}^{\infty}1/n^2$ converges. The Weierstrass M-Test therefore gives uniform convergence of $\sum |f_n(x)|$, so the original series is absolutely uniformly convergent.

A concrete estimate is available for the tail. Since $t\mapsto 1/t^2$ is decreasing and positive, $$ \sum_{n=N+1}^{\infty}\frac{1}{n^2} \leq\int_N^\infty\frac{1}{t^2}\,dt =\frac{1}{N}\qquad(N\geq1). $$ Consequently, for every $x\in[0,1]$, $$ \sum_{n=N+1}^{\infty}|f_n(x)|\leq\frac{1}{N}. $$ The theorem then gives $|S(x)-S_N(x)|\leq1/N$ uniformly on the whole interval. This estimate does not require knowing a closed form for the sum.

Pointwise Absolute Convergence Is Not Enough

A series can converge absolutely at every input while its absolute tails fail to become uniformly small. The following example makes the difference visible without relying on cancellation.

Worked Example: Pointwise Absolute but Not Absolutely Uniform

Take $E=\mathbb{N}$, and for each positive integer $n$, define $f_n(k)=1$ when $k=n$, and $f_n(k)=0$ when $k\ne n$. Fix $k\in E$. Exactly one term of the series $\sum_{n=1}^{\infty}f_n(k)$ is nonzero, namely $f_k(k)=1$. Thus the series converges absolutely at every $k$, and its pointwise sum is $S(k)=1$.

For any $N\geq0$, if $k>N$, then $$ \sum_{n=N+1}^{\infty}|f_n(k)|=1, $$ because the tail contains the single nonzero term $f_k(k)$. If $k\leq N$, that tail is zero. Hence $$ \sup_{k\in\mathbb{N}}\sum_{n=N+1}^{\infty}|f_n(k)|=1 $$ for every $N$. The absolute tails do not tend uniformly to zero, so the series is not absolutely uniformly convergent. Indeed, its partial sum $S_N(k)$ is $1$ for $k\leq N$ and $0$ for $k>N$, giving $\sup_k|S(k)-S_N(k)|=1$ for every $N$. Pointwise absolute convergence does not supply a uniform choice of $N$.

The reverse distinction also matters: uniform convergence of the original series does not imply absolute uniform convergence. Uniform convergence controls signed tails, which may be small because positive and negative terms cancel. Absolute uniform convergence controls the sum of the magnitudes, so it cannot use that cancellation.

Worked Example: Uniform Convergence Without Absolute Uniform Convergence

On $E=[0,1]$, define the constant functions $$ f_n(x)=\frac{(-1)^{n+1}}{n}. $$ The partial sums are independent of $x$. The alternating-series estimate gives, for every $N\geq1$, $$ \left|\sum_{n=N+1}^{\infty}\frac{(-1)^{n+1}}{n}\right| \leq\frac{1}{N+1}. $$ This bound tends to zero and does not depend on $x$, so the function series converges uniformly.

However, at every $x\in[0,1]$, the series of absolute values is the harmonic series: $$ \sum_{n=1}^{\infty}|f_n(x)|=\sum_{n=1}^{\infty}\frac{1}{n}, $$ which diverges. Thus the series is not absolutely uniformly convergent; it is not even absolutely convergent at a single input. Its uniform convergence is made possible by alternating cancellation.

Rearranging the Terms

One important benefit of absolute uniform convergence is that the order of the terms does not affect the sum. A permutation is a one-to-one correspondence $\pi:\mathbb{N}\to\mathbb{N}$; the rearranged series lists the terms as $f_{\pi(1)},f_{\pi(2)},\ldots$. For an absolutely uniformly convergent series, the uniform control of absolute tails prevents any rearrangement from changing the limit.

Theorem (Rearrangement of an Absolutely Uniformly Convergent Series): Suppose $\sum_{n=1}^{\infty}f_n(x)$ is absolutely uniformly convergent on $E$, with sum $S$. For every permutation $\pi$ of $\mathbb{N}$, the rearranged series $\sum_{j=1}^{\infty}f_{\pi(j)}(x)$ converges uniformly on $E$ to the same function $S$.

Proof. Let $T_M(x)=\sum_{j=1}^{M}f_{\pi(j)}(x)$. Given $\varepsilon>0$, absolute uniform convergence supplies $K$ such that $$ \sum_{n=K+1}^{\infty}|f_n(x)|<\frac{\varepsilon}{2} \qquad\text{for every }x\in E. $$ Because $\pi$ is a permutation, each of the finitely many indices $1,\ldots,K$ occurs among $\pi(1),\ldots,\pi(J)$ for some sufficiently large $J$. Choose such a $J$. For every $M\geq J$, the first $M$ rearranged terms include all terms with original indices at most $K$. All additional terms have original indices greater than $K$. Therefore $$ T_M(x)=\sum_{n=1}^{K}f_n(x)+ \sum_{\substack{1\leq j\leq M\\\pi(j)>K}}f_{\pi(j)}(x). $$ The absolute value of the second sum is at most $\sum_{n=K+1}^{\infty}|f_n(x)|$. The original sum has the corresponding decomposition $$ S(x)=\sum_{n=1}^{K}f_n(x)+\sum_{n=K+1}^{\infty}f_n(x), $$ whose final tail has absolute value at most the same bound. Subtracting these expressions and using the triangle inequality gives $$ |T_M(x)-S(x)| \leq 2\sum_{n=K+1}^{\infty}|f_n(x)| <\varepsilon $$ for every $x\in E$ and every $M\geq J$. This proves uniform convergence of the rearranged series to $S$. $\square$

Worked Example: Swapping Adjacent Terms

On $E=[0,1]$, set $f_n(x)=x/2^n$. Its absolute tail satisfies $$ \sum_{n=N+1}^{\infty}|f_n(x)| =x\sum_{n=N+1}^{\infty}\frac{1}{2^n} =\frac{x}{2^N} \leq\frac{1}{2^N}. $$ Thus the series is absolutely uniformly convergent. Its sum is $S(x)=x$, since $\sum_{n=1}^{\infty}2^{-n}=1$.

Now swap each adjacent pair, so that $\pi(1)=2,\ \pi(2)=1,\ \pi(3)=4,\ \pi(4)=3$, and so on. The beginning of the rearranged series is $$ \frac{x}{4},\ \frac{x}{2},\ \frac{x}{16},\ \frac{x}{8},\ \frac{x}{64},\ \frac{x}{32},\ \ldots. $$ For example, its first four terms sum to $x/4+x/2+x/16+x/8=15x/16$, which is the sum of the original terms with indices $1,2,3,4$. After each complete pair, the rearranged partial sum agrees with the corresponding original partial sum. The rearrangement theorem guarantees that the full rearranged series converges uniformly to $x$, just as the original series does.

Using the Condition Effectively

When checking absolute uniform convergence, it is usually efficient to estimate absolute tails directly. A pointwise convergence calculation alone does not establish the required uniform bound. The Weierstrass M-Test gives a convenient sufficient condition: find numbers $M_n\geq0$ with $|f_n(x)|\leq M_n$ for all $x$, and verify that $\sum M_n$ converges. The M-Test then controls the absolute series uniformly. This is sufficient, but it is not necessary; an estimate tailored to the domain may work even when no useful numerical majorant is apparent.

Take absolute values.
Estimate $|f_n(x)|$ rather than relying on cancellation in the original series.

Control the whole tail.
Seek a bound for $\sum_{n=N+1}^{\infty}|f_n(x)|$ that tends to zero independently of $x$.

Use the tail estimate.
It bounds the error in the original sum and permits any permutation of the terms without changing the uniform limit.

A common pitfall is to show that $\sum |f_n(x)|$ converges for each fixed $x$ and then call the convergence absolute uniform. The example on $\mathbb{N}$ shows why this fails: every fixed input has only one nonzero term, while no single tail index controls all inputs. Another pitfall is to infer absolute uniform convergence from uniform convergence of the signed series. The alternating harmonic example shows that cancellation can give uniform convergence even when the absolute series diverges.

Takeaway: Absolute uniform convergence means that the sums of absolute tails become small uniformly over the domain. It implies uniform convergence, provides a direct remainder bound, and ensures that rearranging the terms leaves the uniform sum unchanged.

Check Your Understanding

Use the definitions, estimates, and results in this tutorial to answer the following questions.

What must tend uniformly to zero for a function series to be absolutely uniformly convergent?
How does the absolute tail bound control the error between a partial sum and the full sum?
Why is pointwise absolute convergence insufficient in the example on $\mathbb{N}$?
Why does the alternating harmonic function series converge uniformly but fail to converge absolutely?
What feature of an absolutely uniformly convergent series allows its terms to be rearranged without changing the sum?

What You'll Learn