1

Find all continuous and strictly monotonic function $f:[0,\infty)\to \Bbb R$ such that:

  • If there is a pair $(x,y)\neq (0,0)$ such that $2f(x)=f(y)$ then $2f(tx)=f(ty)$ for all $t>0$;

  • There is at least one pair $(x,y)$ where the above condition hold.

What I got so far:

  • Check that $f(x)=ax^b$ is a solution. I want to discover if there is any other;

  • $t\to 0$ then $2f(0)=f(0) \Rightarrow f(0)=0$;

  • We can prove that $y> x$;

  • $t:=t/x$ and then $2f(t)=f(tk)$, for all $t>0$, with $k=y/x >1$;

  • $f(tk^n)=2^nf(t)$ and setting $t=1$ we get $f(k^n)=2^nf(1)$. It give us that, if $f(1)>0$ then $f$ is increasing, otherwise, $f$ is decreasing.

Arnaldo
  • 21,342
  • 1
    You need to also assume that there is some such pair with $x\neq y$, otherwise you could take $e^x-1$ or any continuous, monotone function with $f(0)=0$ as we could use the pair $(x,y)=(0,0)$. – lulu May 13 '19 at 16:13
  • @lulu, If I'm not wrong, if $x=y$ we can conclude that $f(t)=0$ for all $t \in [0,\infty)$ – Arnaldo May 13 '19 at 16:18
  • No, unless you also have $x=y\neq 0$. As my example illustrated, $f(x)=e^x-1$ is a counterexample, just using $(x,y)=(0,0)$. – lulu May 13 '19 at 16:20
  • So you need a new assumption to exclude my counterexamples. – lulu May 13 '19 at 16:21
  • @lulu, I included $(x,y) \neq (0,0)$, thanks – Arnaldo May 13 '19 at 16:24

4 Answers4

3

Actually, there are many such $f$.

Here I assume $f$ is positive, increasing. (If you have decreasing $f$, try to consider $-f$.)

Take $a<b$ s.t. $2f(a)=f(b)$. Let $d$ be $b/a$.

The restricted function $g=f|_{[a,b]}$ should be continuous, strictly increasing on $[a,b]\subset(0,\infty)$ and should satisfy $2g(a)=g(b)>0$.

Now, the main claim is that we can uniquely extend any such $(g,[a,b])$ to a function $f$ satisfying all the properties you asked.

(Uniqueness) As you mentioned above, $f(0)=0$. Choose any $t>0$. $\exists k\in\mathbb{Z},c\in[a,b]$ s.t. $t=d^kc$. Then, $f(t)=2^kf(c)$.

Therefore $f$ is unique on $[0,\infty)$.

(Existence) Define $f$ as follows.

$$f(t)=\begin{cases}0,t=0\\ 2^kg(c),t=d^kc,k\in\mathbb{Z},c\in[a,b]\end{cases}$$

Now it is left to check that such $f$ is positive, continuous and strictly increasing.

$f$ is positive by its definition.

$\forall p,q$ s.t. $0<p<q$, $f(p)<f(q)$ since $p=d^{k_1}c_1,q=d^{k_2}c_2,k_i\in\mathbb{Z},c\in[a,b)\Rightarrow k_1<k_2\,or\,k_1=k_2,c_1<c_2$, and if $k_1<k_2$, $f(p)=2^{k_1}f(c_1)<2^{k_1}f(b)\le \frac{1}{2}\cdot 2^{k_2}f(b)=2^{k_2}f(a)\le 2^{k_2}f(c_2)=f(q)$, and if $k_1=k_2,c_1<c_2$, $f(p)=2^{k_1}f(c_1)<2^{k_2}f(c_2)=f(q)$.

Thus $f$ is strictly increasing.

$f$ is continuous on each $(d^ka,d^kb)$ since $g$ is continuous. At $d^ka=d^{k-1}b$, $f$ is right continuous since so was $g$ at $b$, and is left continuous since so was $g$ at $a$.

Now we check that $f$ is continuous at $0$.

$\forall\epsilon>0$, we can take $N$ s.t. $2^{-N}\cdot f(a)<\epsilon$, then $\forall x\in(0,d^{-N}a)$, $f(x)<f(d^{-N}a)=2^{-N}f(a)<\epsilon$.

Then, since $f$ is positive, $\lim_{x\rightarrow 0}f(x)=0$ and this means $f$ is continuous at $0$.

-So we are done.

P.S. You can also rewrite this in the following more closed form.

$f$ is answer iff $$f(t)=\begin{cases}0,t=0\\ 2^kg(c),t=d^kc,k\in\mathbb{Z},c\in[1,d]\end{cases}$$ where $d>1$ and $g:[1,d]\rightarrow(0,\infty)$ is a continuous strictly monotonic function s.t. $g(d)=2g(1)$.

Arnaldo
  • 21,342
Moonshine
  • 2,088
  • 3
    I am highly suspicious of all those proofs that contain the phrase "it is easy to check that". If it is so easy, do it yourself! After all, questions with bounties offered impose a higher level of quality for their answers than the other, regular questions. To me, it isn't easy at all. In particular, it is not clear at all why your construction gives continuity in $0$. – Alex M. May 16 '19 at 17:25
  • Sorry for being unclear. I've edited what you pointed out... so is it clear now? – Moonshine May 16 '19 at 18:28
  • @Arnaldo: No, that is not possible. Assume to be so. Since $p<q$, $d^{k_1}c_1<d^{k_2}c_2$. Then, $dc_1<c_2$. This makes contradiction since $c_1,c_2\in[a,b)$. – Moonshine May 17 '19 at 16:50
  • @C.Park: Your solution looks fine to me. It is a nice job. If no one post a better solution I'll chose yours. – Arnaldo May 17 '19 at 22:11
1

We assume $f$ is increasing (take $-f$ if it is not.) By the strictly increasing condition, $f(x)>0$ for all positive $x$. We now define a new function $g:$ $$g(\ln(x)) = \ln (f(x)),$$ for all positive $x$. Let $a= \ln(x)$, $b= \ln(y)$, $c= \ln(2)$. The condition becomes $g(t+a) +c = g(t+b) \: \forall t \in \mathbb{R}.$ Furthermore, note that $g$ is still increasing and continuous. Let $S$ be the set of functions $h: \; [0, b-a] \rightarrow [0,c]$ such that $h$ is strictly increasing, continuous, and satisfies $h(0)=0, h(b-a)=c$. It is clear that all solutions to $g$ are given by $g(n(b-a)+t) = h(t) + cn + d$ for any $h \in S$, all integers $n$, all $0 \leq t < b-a$ and any constant $d$. Using $f(x) = e^{g(\ln (x))}$, and then remembering that $f$ was possibly negated at the start, yields all possible $f$.

auscrypt
  • 8,186
  • Nice idea but there is some points: First, $f$ is monotonic, not increasing, but that is easy, just work with $-f$ when it is decreasing. Second, $g: \Bbb R \to \Bbb R$, so $h$ should have $\Bbb R$ as a domain. Isn't it? – Arnaldo May 20 '19 at 20:49
  • @Arnaldo With the second point, $h$ does not need to have $\mathbb{R}$ as a domain; we have $n(b-a)+t$ for all integers $n$ as the input of $g$ so this ensures that the range of $g$ is still $\mathbb{R}$. I’ve edited the solution for the first point you raised. – auscrypt May 20 '19 at 20:58
  • Oh, that is right, I was thinking $n \in \Bbb N$. – Arnaldo May 20 '19 at 21:03
0

Calling $y = x+\delta$ with $\delta > 0$ we have if

$$ 2f(x)-f(x+\delta) = 0\Rightarrow 2f\left(\frac{x}{\delta}\right)-f\left(\frac{x}{\delta}+1\right) = 0 $$

now calling $u = \frac{x}{\delta}$ we have the recurrence functional equation

$$ 2f(u) - f(u+1) = 0 $$

with solution

$$ f(u) = 2^u \Phi(u) $$

where $\Phi(u)$ is any periodic function with period $1$,

and also choosing $\Phi(u) = C_0$

$$ f\left(\frac{x}{y-x}\right)=C_02^{\frac{x}{y-x}} $$

for $x \ne y$ and such that $2f(x) = f(y)$

Cesareo
  • 33,252
  • 1
    If $\Phi(x) = \sin (2\pi x)$, then $2^x \Phi(x)$ has infinitely many zeros, and therefore cannot be strictly monotonic as required. In fact, it is not even monotonic, let alone strictly so. Also, you need a "$=0$" at the end of the first formula. – Alex M. May 16 '19 at 17:14
  • 1
    Furthermore, $x,y,\delta$ are fixed, so $u$ is a constant and then $2f(u)-f(u+1)=0$ is not a classic recurrence. It works only for $u$. – Arnaldo May 16 '19 at 17:47
  • $\Phi(x) = C_0$ is also periodic.
  • Yes. $x$ and $y$ are fixed for values such that $2f(x) = f(y)$ It is a functional recurrence.
  • – Cesareo May 16 '19 at 18:42
  • @Cesareo, do you claim that $f(x)=C_02^x$ is a solution? It doesn't seem so. – Arnaldo May 16 '19 at 19:47
  • @Arnaldo I claim that $$ f\left(\frac{x}{y-x}\right)=C_0 2^{\frac{x}{y-x}} $$

    for $x \ne y$ and such that $2f(x) = f(y)$

    – Cesareo May 16 '19 at 19:50