16

Given a function $f: \mathbb{R}\to\mathbb{R}$, we define its left modulus of continuity, $L(f): \mathbb{R} \times (0, \infty)\to [0,\infty]$ by

$$L(f)(x, e) := \sup \{d \ge 0 \,:\, f((x, x+d)) \subseteq [f(x) - e, f(x) + e]\} $$

Similarly define the right modulus of continuity, $R(f): \mathbb{R} \times (0, \infty)\to [0,\infty]$ by

$$R(f) (x, e) := \sup \{d \ge 0 \,:\, f((x-d, x)) \subseteq [f(x) - e, f(x) + e]\}$$

Suppose $f$ and $g$ are continuous functions such that $L(f) = L(g)$, $R(f) = R(g)$, and $f(0) = g(0) = 0$. Does it follow that either $f = g$ or $f = -g$?

James Baxter
  • 2,029
  • 4
    If you were to replace the left/right modulus functions with a single two-sided modulus function, then the answer would be negative, in light of $f(x)=x$ and $g(x)=|x|$. – Joel David Hamkins Dec 21 '18 at 16:15
  • Yep, I realised that the single modulus was too messy, so I felt requiring both left and right was more natural. – James Baxter Dec 22 '18 at 04:41

1 Answers1

6

I think they are. Assume the contrary, then $f$ is not constant and therefore there exists a point $a$ such that $f$ is not constant neither on $[a,+\infty)$ nor on $(-\infty,a]$. Call such points admissible. The admissible points form an interval (possibly infinite). Without loss of generality $0$ is admissible point (else shift the argument and subtract the constant from $f$.)

Denote $L(f)(0,e)=A(e)$, $R(f)(0,e)=B(e)$. Clearly $A$ is decreasing function on $(0,\infty)$ and $|f(A(e))|=e$ (if $A(e)<\infty$), $|f|<e$ on $[0,A(e))$. The same holds for $g$, thus $g(A(e))=\pm f(A(e))$. I claim that the sign does not depend on $e$. Indeed, choose $e_1>0$ such that $B(e_1)<\infty$. Then analogously $f(-B(e_1))=\pm e_1, g(-B(e_1))=\pm e_1$. I claim that ${\rm sign}\, f(-B(e_1))/g(-B(e_1))={\rm sign}\, f(A(e))/g(A(e))$ for any positive $e$. Indeed, in the opposite case exactly one of the numbers $L(f)(-B(e_1),e+e_1)$, $L(g)(-B(e_1),e+e_1)$ equals to $B(e_1)+A(e)$. A contradiction. So, the sign is always the same.

We say that $0$ is a plus-point if this sign is always plus and 0 is minus-point if the sign is always minus. Analogously, any admissible point $a$ is a plus-point or minus-point. The next claim is that either all admissible points are plus-points or all of them are minus-points. It suffices to show that both sets of plus-points and minus-points are open. Assume the contrary: for example, 0 is a plus-point but there exists a sequence of minus-points $t_n\to 0$. Fix $e>0$ such that $A(e)>0$ and $A$ is continuous at $e$. This continuity yields that whenever $|f(s_n)| \to e$ or $|g(s_n)| \to e$ for a sequence $s_n\in [0,A(e)]$, we must have $s_n\to A(e)$.

Assume without loss of generality that $f(A(e))=g(A(e))=e$. Denote $b_n=L(f)(t_n,e-f(t_n))$. Then $t_n+b_n\leqslant A(e)$ and $f(t_n+b_n)=f(t_n)\pm (e-f(t_n))$, $g(t_n+b_n)=g(t_n)\mp (e-f(t_n))$ with opposite sign. In any case we have $|f(b_n+t_n)|\to e$, thus by aforementioned corollary of continuity of $A$ at $e$ we get $b_n\to A(e)$, but then one of functions $f, g$ becomes discontinuous at $A(e)$.

Now assume that all admissible points are plus-points but there exist points for which $f\ne g$. Denote by $\Omega$ the open set of admissible points for which $f\ne g$. $\Omega$ is non-empty but $0\notin \Omega$. Therefore $\Omega$ has a connected component $(\alpha,\beta)$ with $\alpha$ or $\beta$ admissible. Without loss of generality $\alpha=0$. Then for small $e$ we do not have $f(A(e))=g(A(e))$. A contradiction.

Fedor Petrov
  • 102,548
  • 1
    This will take awhile to digest, but i will get back to you eventually. – James Baxter Dec 22 '18 at 04:39
  • @James there were many bugs now fixed, but maybe some still remain. – Fedor Petrov Dec 22 '18 at 05:39
  • Hmm, everything seems to check out except the last line, Why is f(A(e)) =/= g(A(e)) for small e a contradiction? – James Baxter Dec 22 '18 at 07:33
  • Oh right, sorry. I misread something. It does work indeed. Thanks for the answer! – James Baxter Dec 22 '18 at 07:38
  • I’m having trouble understanding one last step actually, the part where there must be a connected component with admissible left endpoint of the set of points where f =/= g. More specifically the “such a component must exist, else f = g for all admissible points” part. – James Baxter Dec 22 '18 at 07:47
  • @JamesBaxter admissible rightpoint is equally appropriate. If there exist admissible points for which $f\ne g$, consider the set $\Omega$ of admissible points satisfying this inequality. It is open and different both from the empty set and from the whole set of admissible points (since $0\notin \Omega$). Therefore it has a connected component with admissible endpoint. – Fedor Petrov Dec 22 '18 at 08:06
  • Ah, that’s very nice. Btw, I won’t ask for any kind of rigorous answer, but do you have any guesses as to what would happen if we replaced “continuous” with just “measurable”, and asked the same question? Just off the cuff “gut feelings” kinda guesses. – James Baxter Dec 23 '18 at 03:24
  • 2
    @JamesBaxter there are many partitions of $\mathbb{R}$ onto two subsets each of which have positive measure in any interval. Characteristic function of such a subset have left/right modulus of continuity (and essential modulus of continuity) equal to 0 for $e<1$ and equal infinity for $e\geqslant 1$. – Fedor Petrov Dec 23 '18 at 06:50