Let $\,f:[0,\infty)\to [0,\infty)$ be a function such that $\,f(x+y)=f(x)+f(y),\,$ for all $\,x,y\ge 0$. Prove that $\,f(x)=ax,\,$ for some constant $a$.

My proof :

We have , $\,f(0)=0$. Then , $$\displaystyle f'(x)=\lim_{h\to 0}\frac{f(x+h)-f(x)}{h}=\lim_{h\to 0}\frac{f(h)}{h}=\lim_{h\to 0}\frac{f(h)-f(0)}{h}=f'(0)=a\text{(constant)}.$$

Then, $\,f(x)=ax+b$. As, $\,f(0)=0$ so $b=0$ and $f(x)=ax.$

Is my proof correct?

  • 13
    $\begingroup$ It is not correct unless differentiability be an hypothesis $\endgroup$ – sinbadh Feb 15 '16 at 9:34
  • 3
    $\begingroup$ It's also true if $f:\mathbb{Q}\to\mathbb{Q}$, without assuming differentiability $\endgroup$ – vrugtehagel Feb 15 '16 at 9:35
  • 1
    $\begingroup$ Assuming AC, without giving that $\;f\;$ is continuous the claim is false. $\endgroup$ – DonAntonio Feb 15 '16 at 9:42
  • 7
    $\begingroup$ Please, note that the question is not about a function defined over $\mathbb{R}$, but a function defined on $[0,\infty)$ with values in $[0,\infty)$, which makes this a different problem from Cauchy functional equation. $\endgroup$ – egreg Feb 15 '16 at 9:54
  • 1
    $\begingroup$ I have added to the title domain and codomain, to stress the difference pointed out by @egreg. (And I have also retraced my close vote. I should have read the question more carefully.) $\endgroup$ – Martin Sleziak Feb 15 '16 at 11:34

In your proof you assume that $f$ is differentiable, which is not given.

Let me suggest how to obtain the formula of $f$:

Step I. Show that $\,f(px)=p\,f(x),\,$ when $p$ is a positive rational and $x$ a non-negative real. (At first show this for $p$ integer.) We obtain also that, $\,f(0)=0$.

Step II. Observe that $f$ is increasing, since, for $y>x$, we have $$ f(y)=f(x)+f(y-x)\ge f(x). $$

Step III. Since $f$ is increasing, then the limit $\,\lim_{x\to 0^+}f(x)\,$ exists. However $$ \lim_{x\to 0^+}f(x)=\lim_{n\to\infty}f\Big(\frac{1}{n}\Big) =\lim_{n\to\infty}\frac{1}{n}\,f(1)=0. $$

Step IV. Pick an arbitrary $x\in(0,\infty)$, and a decreasing sequence $\{q_n\}\subset\mathbb Q$ tending to $x$. Then $$ f(q_n)=q_n\,f(1) $$ and $$ x\,f(1)\longleftarrow q_n\,f(1)=f(q_n)=f(x)+f(q_n-x)\longrightarrow f(x), $$ since $\,\,q_n-x\to 0^+$, and thus $\,\,\lim_{n\to\infty}f(q_n-x)=0$.

Therefore, $\,f(x)=x\,f(1),\,$ for all $x\in\mathbb [0,\infty)$, and hence $\,f'(x)=f(1)$.

  • $\begingroup$ Why $\;q_nf(0)=f(x)+f(q_n-x)=f(q_n)\;$ ? Shouldn't it be $\;f(q_n)=q_nf(1)\;$ ? Also, why $\;q_n-x\,\to\, \implies\,f(q_n-x)\,\to f(0)\;?$ Isn't this assuming continuity? $\endgroup$ – DonAntonio Feb 15 '16 at 10:27
  • $\begingroup$ You are absolutely right! I am correcting it! $\endgroup$ – Yiorgos S. Smyrlis Feb 15 '16 at 10:28
  • $\begingroup$ For Step III: by Monotone Limit Theorem, $\ell=\lim\limits_{x\to0^+}f(x)$ exists; by composition of limits: $\ell=\lim\limits_{n\to+\infty}f(1/n)=\lim\limits_{n\to+\infty}f(1)/n$ (by Step I) hence $\ell=0$. I guess you don't need to use a reasoning by contradiction. $\endgroup$ – gniourf_gniourf Feb 15 '16 at 13:44
  • 1
    $\begingroup$ @gniourf_gniourf I simplified my answer according to your comment. Thanx! $\endgroup$ – Yiorgos S. Smyrlis Feb 15 '16 at 15:24

Note that $f$ is monotonic: if $y>0$, then $f(y)\ge0$ so $f(x+y)=f(x)+f(y)\ge f(x)$.

In particular, the function is continuous over $[0,\infty)$ except for an at most countable set.

You can extend $f$ to $f_e\colon\mathbb{R}\to\mathbb{R}$ by setting $f_e(x)=-f(-x)$, for $x<0$. Show that this function still has the property that $f_e(x+y)=f_e(x)+f_e(y)$. Then continuity at a point implies continuity at $0$.

The result now follows from methods in Overview of basic facts about Cauchy functional equation

Your attempt is not good, I'm afraid: you're assuming differentiability at $0$, which is not among the hypotheses.


By induction, $f(nx)=nf(x)$ for an integer $n$.

Now take any real $x$. From

$$\lfloor nx\rfloor\le nx\le\lceil nx\rceil,$$applying the non-decreasing function $f$, we deduce $$f\left(\lfloor nx\rfloor\right)\le f(nx)\le f\left(\lceil nx\rceil\right).$$

By the above induction property, $$\lfloor nx\rfloor f(1)\le nf(x)\le \lceil nx\rceil f(1),$$ and $$\frac{\lfloor nx\rfloor}nf(1)\le f(x)\le \frac{\lceil nx\rceil}nf(1).$$

As $n$ can be arbitrarily large, by squeezing


  • $\begingroup$ I don't think $\;f\;$ must be continuous: choosing a Hamel basis for linear space $\;\Bbb R_{\Bbb Q}\;$ any linear operator will do the trick, and not all are of the wanted form. $\endgroup$ – DonAntonio Feb 15 '16 at 11:07
  • $\begingroup$ @Joanpemo Monotone functions can only have at most countably many discontinuities in an interval, so $f$ is continuous somewhere, hence everywhere. $\endgroup$ – Qiyu Wen Feb 15 '16 at 11:13
  • $\begingroup$ Thank you@QiyuWen . Apparently the domain and codomain being the positive reals make the trick for monotony. $\endgroup$ – DonAntonio Feb 15 '16 at 11:17
  • 1
    $\begingroup$ @Joanpemo: I have completely rewritten, with a much simpler argument. $\endgroup$ – Yves Daoust Feb 15 '16 at 11:50
  • 1
    $\begingroup$ Only answer not needing a sledgehammer to crack a nut. $\endgroup$ – Bananach Feb 22 '16 at 8:48

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.