11

If $(f\circ g) (x)=(\sin\sqrt{x})^2$ and $(g\circ f) (x)=\lvert\sin x\rvert$, find $f$ and $g$.

My progress:

By trial and error I got one such pair $f(x) = \sin^2x$, $g(x)=\sqrt{x}$. But I can’t conclude that this is the only such pair or whether there exist other such pairs.

Ѕᴀᴀᴅ
  • 34,263
Makar
  • 2,267
  • The question is good, never thought about other possibility... – King Tut Mar 13 '18 at 16:28
  • Do you have specific domain and codomains in mind? – Arnaud D. Mar 13 '18 at 16:35
  • Let $a(x)=\sin^2\sqrt x$ and $b(x)=\lvert\sin x\rvert$. Then $a\circ f=f\circ b$ and $g\circ a=b\circ g$. Not sure where to go from there, but at least each equation contains only one unknown function. –  Mar 13 '18 at 16:37
  • I'm sure that there exist infinitely many of such functions but I will let it as an hypotesis since I can't prove it. – user Mar 13 '18 at 17:01
  • @ArnaudD. I don’t have any specific Co domain/ domain for f and g – Makar Mar 13 '18 at 17:01
  • @gimusi can you give some examples – Makar Mar 13 '18 at 17:03
  • @Makar RobertZ has just given an example! I had in mind something similar but I can't give suche example. – user Mar 13 '18 at 17:08

2 Answers2

5

No, the pair is not unique. There are infinite (uncountable) pairs! Consider $f(x)=\sin^2 (x)$ and
$$g_A(x)=\begin{cases} \sqrt{x} &\text{if $x\geq 0$ and $x\not=(k\pi)^2$ with $k\in A$,}\\ 0 &\text{if $x=(k\pi)^2$ with $k\in A$,} \end{cases}$$ where $A$ is any subset of $\mathbb{N}^+$.

Note that if $0\leq x\leq 1$ then $g_A(x)=\sqrt{x}$ and $$(g_A\circ f) (x)=g_A(\sin^2(x))=|\sin x|.$$ Moreover, if $x=(k\pi)^2$ with $k\in A$ then $$(f\circ g_A) (x)=\sin^2(0)=0=\sin^2\sqrt{(k\pi)^2}=\sin^2(\sqrt{x}).$$

Robert Z
  • 145,942
3

A different, more abstract approach:

Note that the given formulas make sense only if $x\ge0$. Note also that the functions $\Phi(x)=\sin^2(\sqrt x\,)$ and $\Psi(x)=\sin x$ are given as conjugate, via the function $\alpha(x)=x^2$, that is, $\Phi=\alpha\circ\Psi\circ\alpha^{-1}$. The functions you found were $f=\alpha\circ\Psi$ and $g=\alpha^{-1}$. Then $f\circ g=\alpha\circ\Psi\circ\alpha^{-1}=\Phi$, and $g\circ f=\alpha^{-1}\circ\alpha\circ\Psi=\Psi$.

Clearly, then, we may replace our $\alpha$ by $\bar\alpha=\alpha\circ\psi$ for any $\psi$ commuting with $\Psi$, for then $\bar f=\bar\alpha\circ\Psi$ and $\bar g={\bar\alpha}^{-1}$ will have $\bar f\circ\bar g=\alpha\circ\psi\circ\Psi\circ\psi^{-1}\circ\alpha^{-1}=\Phi$ and $\bar g\circ\bar f=\psi^{-1}\circ\alpha^{-1}\circ\alpha\circ\psi\circ\Psi=\Psi$.

The functions commuting with $\sin$ are rare, but among them are $\sin$ and $\arcsin$, so we have:

Example 1: $f=\sin^2(\sin(x))$, $g=(\alpha\circ\sin)^{-1}=\arcsin(\sqrt x\,)$.
Example 2: $f=x^2$, $g=\sin(\sqrt x\,)$. This is the answer we all should have seen, but it took me a lot of work.

Of course you can use $\psi=\sin^{\circ n}$, where that means iterating the sine $n$ times, or the $(-n)$-fold iteration of $\arcsin$ if $n$ is negative, giving another, different infinite family of solutions.

Lubin
  • 62,818
  • 2
    Example 2 does not work. For $x\geq 0$, $g(f(x))=\sin(\sqrt{x^2})=\sin(|x|)=\sin(x)$ which is not the same of $|\sin(x)|$. – Robert Z Mar 14 '18 at 05:32
  • 2
    @RobertZ: However that can be fixed by choosing $g=\left|\sin\sqrt{x}\right|$. The point being that for this construction to work, the image of $\alpha^{-1}$ must be in the domain of $\phi$, but the image of $\sin(x)$ is $[-1,1]\not\subset [0,\infty)$. – celtschk Mar 14 '18 at 06:02
  • @celtschk Yes, I agree with you! – Robert Z Mar 14 '18 at 06:18
  • Well, as I said at the top, the original data involve $\sin(\sqrt x,)$, so that we must restrict to $x\ge0$. I plead guilty, however, to restricting further to the domain in which the sine is one-to-one, thus allowing use of the arcsine function. Further, I presented the problem to myself as one in elementary group theory, not real analysis. After all, the choice of the sine function to base everything on was arbitrary and inconsequential. (Perhaps it would have been better to start with sinh and conjugate with the cubing function!) – Lubin Mar 14 '18 at 19:34