I'd love your help proving the following claim:

If $f$ is continuous and $\int_a^\infty |f(x)|\;dx$ is finite then $\lim\limits_{ x \to \infty } f(x)=0$.

Here the counter example of all these sine functions won't do, since their infinite integral is not absolute converges.

Any hints for a proof?


The assertion is not true, even if $f(x)\geq0$ for all $x$: let $f(x)$ be $$ f(x)=\begin{cases}1-2n^2\left|x-\left(n+\frac1{2n^2}\right)\right|,&\ \text{ if } x\in[n,n+\frac1{n^2}),\\ \ \\ 0,&\ \text{ if }x\in [n+\frac1{n^2},n+1) \end{cases} $$ for $n\in\mathbb{N}$, $f(x)=0$ for $x<1$. Then $$ \int_0^\infty f(x)\,dx\leq\sum_{n=1}^\infty \frac1{2n^2}=\frac{\pi^2}{12}<\infty. $$ But $f(n+\frac1{2n^2})=1$ for all $n$.

Your assertion is true if you also require $f$ to be monotone.

One can also tweak this example to get $f(x)>0$ for all $x$, and one can also get $\limsup_{x\to\infty}f(x)=\infty$ (for instance, one could get $f(n)=n$ for all $n\in\mathbb N$).


The result is false, as the often presented example of spikes with vanishing bases shows.

For concreteness, introduce the elementary spike $s:x\mapsto(1-|x|)^+$ centered at $0$, with height $1$ and width $2$, and consider $$ f(x)=\sum\limits_{n\geqslant1}s(n^2x-n^3). $$ In words, one adds countably many spikes, the $n$th spike being centered at $n\geqslant1$, with height $1$ and width $\frac2{n^2}$. Then:

  1. The function $f$ is finite and nonnegative everywhere.
  2. The function $f$ is integrable since $\int\limits_{-\infty}^{+\infty} s(ax+b)\mathrm dx=\frac1a\int\limits_{-\infty}^{+\infty} s(x)\mathrm dx=\frac1a$, for every $a\gt0$ and every $b$, hence $\int\limits_{-\infty}^{+\infty} f(x)\mathrm dx=\sum\limits_{n\geqslant1}\frac1{n^2}$, which converges.
  3. The function $f$ has no limit at infinity since, for every $n\geqslant3$, $f(n)=1$ and $f\left(n+\frac12\right)=0$.
  • $\begingroup$ Wouldn't it rather be $f(x) = \sum_{n \geq 1} s(n^2 (x-n)) = \sum_{n \geq 1} s(n^2 x-n^3)$ ? $\endgroup$ – D. Thomine Feb 11 '12 at 18:13
  • $\begingroup$ @D.Thomine: Absolutely! Good mind reading... Thanks. $\endgroup$ – Did Feb 11 '12 at 18:17

this is not true

you can construct a function easily like this:function graph

so $\int_{a}^{\infty}|f(x)| = \sum{u_n}$ starting from n0 such that $u_{n0-1} < a \leq u_{n0}$

so take any $u_n$ such that $\sum{u_n}$ converges

  • $\begingroup$ Thanks for such a beautifully crafted answer. $\endgroup$ – StammeringMathematician Nov 28 '19 at 10:54

We do have the following closely related result: Suppose $f \in L^1 ([0,\infty))$, that is $f$ is measurable and $\int_{0}^\infty |f| < \infty$. Then $f$ "nearly" vanishes at infinity. More precisely, for every $\varepsilon > 0$ there exists a set $A \subset [0,\infty)$ with Lebesgue measure $m(A)<\varepsilon$ such that $\lim_{x \to \infty, x \notin A} f = 0$. In other words, the restriction of $f$ to the complement of $A$ vanishes at infinity.

Here's a proof. Clearly it suffices to show that there is a set $A$ of some finite measure which has the above property. First note that $\left\{ x\ge0:\left|f\left(x\right)\right|>1\right\} $ has finite measure (since $f$ is in $L^1$) so there exists $x_1 \ge 0 $ such that $A_1 := \left\{ x\in\left[x_{1},\infty\right):\left|f\left(x\right)\right|>1\right\} $ has measure smaller than $\frac{1}{2}$. Next note that $\left\{ x\ge0:\left|f\left(x\right)\right|>\frac{1}{2}\right\} $ also has finite measure, so there exists $x_2 \ge x_1$ such that $A_2 := \left\{ x\in\left[x_{2},\infty\right):\left|f\left(x\right)\right|>\frac{1}{2}\right\} $ has measure smaller than $\frac{1}{2^2}$. Continuing this way, we get a sequence of numbers $0 \le x_1 \le x_2 \le \cdots $ such that $A_{n}:=\left\{ x\in\left[x_{n},\infty\right):\left|f\left(x\right)\right|>\frac{1}{n}\right\} $ has measure smaller than $\frac{1}{2^n}$. I now claim that $A := \bigcup_{n=1}^{\infty} A_n$ is a set as desired.

First, suppose that $0 \le y_1 < y_2 < \cdots$ is a sequence of numbers disjoint from $A$ which tends to infinity. Let $n$ be a postiive integer. Then $f(y_k) > \frac{1}{n}$ for at most finitely many $k$, since otherwise there would be a $k$ such that $y_k \ge x_n$ and $f(y_k) > \frac{1}{n}$, contradicting the assumption that $y_k$ is not in $A$ (and therefore in particular not in $A_n$). It now follows that $\lim_{k \to \infty} f(y_k) = 0$.

Second, let us show that $A$ has finite measure. Indeed, $m\left(A\right)=m\left(\bigcup_{n=1}^{\infty}A_{n}\right)\le\sum_{n=1}^{\infty}m\left(A_{n}\right)\le\sum_{n=1}^{\infty}\frac{1}{2^{n}}=1$. Now we are done.

The conclusion cannot be strengthened to obtain an $A$ with measure zero, even if $f$ is continuous, as the other answers you got clearly show.


A general hint, and, hopefully, motivation for the process behind the other answers:

For $\delta>0$ and positive constants $a_1, a_2,\ldots$ with $\sum\limits_{i=1}^\infty a_i$ finite given, construct $f$ to satisfy each of the following conditions:

  • $f$ is nonnegative and integrable on any set $[a,b]$.
  • $f$ is identically 0 off the intervals $A_1$, $A_2$, $\cdots$; where, $A_i\cap A_j=\emptyset$ for $i\ne j$, and $j\in A_j$ for each $j$.
  • For each $j$, we have $\int_{A_j} f(x)\,dx =a_j$ and $\sup\limits_{x\in A_j} f(x)>\delta$.

In other words, the graph of $f$ consists of disjoint positive spikes of common amplitude $\delta$ whose widths converge to 0 fast enough so that the sum of their areas is finite.

You could even set things up so that the amplitudes tend to infinity.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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