Consider the following problem:

Suppose $f:[0,1] \to \mathbb{R}$ is a differentiable function with $f'(0)<0$, $f(0)=1$ and for all $x\in(0,1]:0<f(x)<1$. Calculate $$\lim_{n\to\infty}\int_0^nf\left(\frac{x}{n}\right)^ndx.$$

My attempt so far:

Let $$f_n:x\mapsto\chi_{[0,n]}(x)\cdot f\left(\dfrac{x}{n}\right)^n,$$ where $\chi_{[0,n]}$ is the indicator function on $[0,n]$. We wish to do two things:

  • Show that $f_n\to h$ pointwise for some integrable function $h:\mathbb{R}^+\to\mathbb{R}$ and
  • show that $|f_n|\leq k$ for some positively integrable $k:[0,1]\to \mathbb{R}^+$ so that we can use the dominated convergence theorem to show that $$\lim_{n\to\infty}\int_{[0,\infty)}f_n(x) \,dx = \int_{[0,\infty)}h(x)dx.$$

To do this, we first construct the function $$g:[0,1]\to\mathbb{R}:y\mapsto\begin{cases}\dfrac{\ln f(y)}{y}& y\in(0,1],\\ f'(0) & y=0.\end{cases}$$ This is a continuous extention of $\ln f(y)/y$ since $$\lim_{y\to0}\dfrac{\ln f(y)}{y}=f'(0)$$ by l'Hôpital's rule. The idea is to show that $$f\left(\dfrac{x}{n}\right)^n\leq e^{f'(0)x}$$ for all $x\in[0,n]$. To do this we can use the function $g$ as constructed above. Indeed, taking the natural log of this inequality gives us

$$n\ln f\left(\dfrac{x}{n}\right)\leq f'(0)x$$ which is equivalent to $$g(y)\leq f'(0),$$ for all $y\in[0,1]$. The problem is that I have trouble showing that this last inequality is true. Although I'm convinced it is. I've tried calculating $g'$ to see if it is negative everywhere, but have failed to do so.

Suppose it is true. Then, since $f'(0)<0$, the function $x\mapsto e^{f'(0)x}$ is integrable on $x>0$, so we have found a dominating function for the $f_n$.

Furthermore, we need to show that there is an integrable function $h$ to which the $f_n$ converge. I cannot see how to show this pointwise convergence.

Edit: I have found conclusive proofs for my two problems and have added them as an answer.

To conclude the problem, the limit is $$\lim_{n\to\infty}\int_0^nf\left(\frac{x}{n}\right)^ndx = \int_0^\infty e^{-|f'(0)|x}dx = \dfrac{1}{|f'(0)|}.$$

  • $\begingroup$ @LightYagami I agree that the answer is 1. However, plugging in a function $f$ is hardly a proof ;). I've found a proof and am writing an answer as we speak. $\endgroup$ Jan 7, 2021 at 14:08
  • $\begingroup$ That's nice. I will be happy to read it. $\endgroup$
    – V.G
    Jan 7, 2021 at 14:10
  • $\begingroup$ @LightYagami sorry, the limit is actually $|1/f'(0)|$. Since the limiting function is $e^{f'(0)x}$. $\endgroup$ Jan 7, 2021 at 14:15

2 Answers 2


For all $x\ge 0$: $$f\left(\dfrac{x}{n}\right)^n\to e^{f'(0)x}.$$


For $x=0$, this is true since $f(0)=1$. Let $x>0$. We whish to show that $$\left|f\left(\dfrac{x}{n}\right)^n - e^{f'(0)x} \right|\to 0.$$ This is equivalent to $$\dfrac{f\left(\dfrac{x}{n}\right)^n }{e^{f'(0)x}}\to 1.$$ (Note that both denominator and enumerator are positive) Which on its turn is equivalent to $$d_n =\ln \dfrac{f\left(\dfrac{x}{n}\right)^n }{e^{f'(0)x}} \to 0.$$ We will show that this is indeed true. Note that $$d_n = n\ln f\left(\dfrac{x}{n}\right)-f'(0)x$$ and hence that (we took $x>0$) $$\dfrac{d_n}{x} = g\left(\dfrac{x}{n}\right)-f'(0).$$

We already know that $g(0) = f'(0)$, so then $d_n \to 0$.

For all $y\in[0,1]:$ $$g(y) \leq f'(0).$$


Note that from what we know about $f$, it must be true that for all $y\in [0,1]: f(y)\leq f(0)=1$. Then it is also true that for all $y\in[0,1]:$ $$\ln f(y) \leq \ln f(0).$$ Now, it is also true that for all $y\in (0,1]:$ $$\dfrac{\ln f(y)}{y} \leq \dfrac{\ln f(0)}{y}.$$ Now, note that $\ln f(0) = 0$, so for any $\varepsilon > 0$ we get that $$\dfrac{\ln f(y)}{y} \leq \dfrac{\ln f(0)}{y-\varepsilon}.$$ Taking the limit $\varepsilon \to y$ makes the right hand side $g(0) = f'(0)$.

  • $\begingroup$ Thats correct ! $\endgroup$
    – EDX
    Jan 7, 2021 at 14:14

Here is an alternative proof that doesn't require convergence theorems for integrals. Write $\ell=|f'(0)|$ and take $0<\epsilon<\frac12\ell$: the factor $1/2$ will come into play at the end and is purely here for convenience. Since $f$ is differentiable at $0$ there exists $\eta\in(0,1]$ such that $$ \forall x\in [0,\eta], \quad |f(x)-1+\ell x|\leq\epsilon x $$ Without loss of generality we further assume $\eta < \frac{1}{2\ell}$. Then for all $x\in[0,\eta]$, $$ 0<1-\ell x-\epsilon x\leq f(x)\leq 1-\ell x+\epsilon x $$ and so $$ \begin{array}{rcl} \displaystyle \int_0^nf(x/n)^n~dx &=& \displaystyle n\int_0^1f(x)^n~dx\\ &=& \displaystyle \underbrace{n\int_0^\eta f(x)^n~dx}_{I_n} + \underbrace{n\int_\eta^1f(x)^n~dx}_{II_n} \end{array} $$ We show that the first term $I_n$ converges to $\frac1{f'(0)}$ and the second term $II_n$ tends to $0$.

The second term. This is swiftly dealt with: by continuity of $f$ and $0<f<1$ on $(0,1]$ it follows that $M=\max_{[\eta,1]}f<1$. Then $$ 0\leq II_n\leq nM^n $$ so that $II$ indeed (converges and) tends to $0$.

The first term. Integrating the inequality from above yields $$ -n\bigg[\frac{(1-(\ell+\epsilon)x)^{n+1}}{(n+1)(\ell + \epsilon)}\bigg]^\eta_0 \leq I_n\leq -n\bigg[\frac{(1-(\ell - \epsilon)x)^{n+1}}{(n+1)(\ell - \epsilon)}\bigg]^\eta_0 $$ The LHS equals $$ \frac{n}{n+1}\frac{1}{\ell+\epsilon} - C_nq_1^n=\frac{1}{\ell+\epsilon}+o(1) $$ for $C_n=\frac{n}{(n+1)(\ell + \epsilon)}$ and $q_1=1-(\ell + \epsilon)\eta\in(0,1)$.

Similarly, the RHS equals $$ \frac{n}{n+1}\frac{1}{\ell-\epsilon} - D_nq_2^n=\frac{1}{\ell-\epsilon}+o(1) $$ for $D_n=\frac{n}{(n+1)(\ell - \epsilon)}$ and $q_2=1-(\ell - \epsilon)\eta\in(0,1)$.

Since $\frac{1}{\ell+2\epsilon}<\frac{1}{\ell+\epsilon}$ and $\frac{1}{\ell-\epsilon}<\frac{1}{\ell-2\epsilon}$, there exists an integer $n_0$ such that for all $n\geq n_0$ $$ \frac{1}{\ell+2\epsilon} \leq I_n \leq \frac{1}{\ell-2\epsilon} $$ This establishes the convergence of $I_n$ to $\frac1\ell=\frac1{|f'(0)|}$ and finishes the proof.

  • 1
    $\begingroup$ That's a lot of estimations and inequalities. Thanks for the proof! $\endgroup$ Jan 7, 2021 at 15:43

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.