Let $f:\mathbb{R}\to \mathbb{R}$ be differentiable, such that $f'(x) \ge 1_{\mathbb{Q}}(x)$. Is it true that $\lim_{x\to\infty}f(x) = \infty$?
-
$\begingroup$ This seems like it is on the border of being equivalent to the question: If $f : [0, 1] \to \mathbb R$ is continuous and differentiable on $(0, 1)$, and satisfies $f(0) = 0$ and $f'(x) \ge [\mathbb Q](x)$ for all $x \in (0, 1)$, then $f(1) \ge 1$. Do you know whether that statement is true? $\endgroup$– LSpiceCommented Dec 1, 2019 at 21:05
-
1$\begingroup$ @LSpice They don't appear to be equivalent to me. The condition $f(1)\geq 1$ seems too strict for your statement. If you have some $f$ such that your statement is true, you can build $f_1$ such that OP's is true by tiling $\mathbb{R}$ by $[0,1]$. But for each tile a bound much smaller than $1$ suffices. In particular for the region $[k, k+1]$ (and $k\in\mathbb{Z}$) you can choose the bound $f_1(1)\geq 1/|k|$, which will suffice by the divergence of the harmonic sum. So the bound on the growth within each individual tile will $\to 0$ while still yielding a "globally unbounded" function. $\endgroup$– Mark Schultz-WuCommented Dec 1, 2019 at 23:55
-
$\begingroup$ @Mark, I didn't literally mean that the only way to get an infinite sum was to sum a bunch of 1's, but rather that it seemed to me that, if this apparently stronger conjecture failed, then one could also build a counterexample to the weaker conjecture. Since this is vacuously true by virtue of the weaker conjecture (and probably, by the same construction, the stronger conjecture) being false, my intuition is vacuously vindicated. :-) $\endgroup$– LSpiceCommented Dec 2, 2019 at 16:38
2 Answers
The Pompeiu function $g:[0,1]\to \mathbb{R}$ is strictly increasing, everywhere differentiable, with derivative vanishing on a dense $G_\delta$ set. With minor modifications in the construction, one can make it instead a homeo $G:[0,1) \to [0,+\infty)$. (Alternatively, such modification $G$ can be made starting from the original $g$, see below).
In this case, the function $h(x):=x+G(x)$ is therefore a differentiable, strictly increasing homeo $[0,1)\to [0,+\infty)$ everywhere differentiable with $h'(x)=1$ in a dense $G_\delta$ set. Hence the inverse map $f$ is an increasing homeo $[0,+\infty)\to[0,1)$ with derivative $1$ on a dense $G_\delta$ set. (It may be a bit technical to make this set contain the rationals as required).
How to make an unbounded Pompeiu function: start from the original Pompeiu function $g:[0,1]\to[0,1]$, where we can assume $g(0)=g'(0)=0$, so that we can extend it on the left to a differentiable function vanishing identically for $x\le0$. Define, for any $0\le x<1$, and with coefficients $c_n>0$ large enough to make the sum unbounded $$G(x):=\sum_{n=1}^\infty c_n g\big(x-1+{1\over n}\big) $$ This sum is locally finite for all $x\in[0,1)$, therefore defines a differentiable function on $[0,1)$, which has a derivative vanishig on a dense set, due to the funny properties of the Pompeiu derivatives: Since the derivative of each term of the sum vanishes on a dense $G_\delta$ set, their sum vanishes on the intersection, which is a dense $G_\delta$ set by Baire's theorem.
-
-
$\begingroup$ I changed the construction; now $f'(x)>0$ for all $x$ $\endgroup$ Commented Dec 2, 2019 at 0:58
-
1$\begingroup$ It now suffices to compose your $f$ (a $C^1$ function with $f' = 1$ on a dense set) with a diffeomorphism which maps rationals into any countable dense subset of $\{x : f'(x) = 1\}$. The existence of such a homeomorphism follows from Theorem 5.2 in R.D. Anderson, D.W. Curtis, J. van Mill. A fake topological Hilbert space. Trans. Amer. Math. Soc. 272(1) (1982): 311–321. $\endgroup$ Commented Dec 2, 2019 at 7:06
-
2$\begingroup$ Yes, to biject two dense countable sets I think there is an entire analytic diffeo, like in the answer below. But composing doesn't it change the value 1 of the derivative? mathoverflow.net/q/42464 $\endgroup$ Commented Dec 2, 2019 at 8:17
-
$\begingroup$ Ah, of course, sorry. Need another coffee. $\endgroup$ Commented Dec 2, 2019 at 8:49
The answer is "no", and here is how one can adapt Pietro Majer's construction from the other answer to work with the set of rational numbers.
Step 1. We begin with the original construction of Pompeiu function. Let $q_n$ be an enumeration of rational numbers in $(0, 1)$, and define, as in the original construction, $$ u(x) = a + b \sum_{n = 1}^\infty \frac{\sqrt[3]{x - q_n}}{2^n} \, , $$ where $a$ and $b > 0$ are chosen in such a way that $u(0) = 0$ and $u(1) = 1$. Then $u$ is increasing and differentiable in an extended sense: $u'(x) \in (0, \infty]$ exists for every $x \in [0, 1]$, and $u'(x) = \infty$ for every rational $x \in (0, 1)$.
Step 2. We now define $$ v(x) = u\biggl(\frac{1}{2} + \frac{x}{1 + 2 |x|}\biggr) . $$ Observe that $x \mapsto 1/2 + x / (1 + 2 |x|)$ is differtentiable with everywhere positive derivative, and it maps rational numbers to rational numbers. Thus, $v : \mathbb{R} \to (0, 1)$ is an increasing homeomorphism, $v$ is everywhere differentiable in the extended sense, $v'(x) \in (0, \infty]$ for every $x$, and $v'(x) = \infty$ whenever $x$ is rational.
Step 3. We now find an increasing homeomorphism $\phi : (0, 1) \to (0, 1)$ which is differentiable with everywhere positive derivative, and which has the following property: for every $s \in \mathbb{Q}$ there is $x \in \mathbb{Q}$ such that $$x + \phi(v(x)) = x.$$ Such a function $\phi$ can be constructed recursively, as follows.
We begin with $\phi_0(x) = s$. Let $r_n$ be an enumeration of all rational numbers. In step $n$ we suppose that we have already constructed a function $\phi_{n-1}$ with the following properties: $$|\phi_{n-1}' - 1| < 3^{-1} + \ldots + 3^{-(n-1)},$$ and there are rational numbers $x_1, \ldots, x_{n-1}$ such that $$x_j + \phi_{n-1}(v(x_j)) = r_j$$ for $j = 1, \ldots, n-1$. We now define $\phi_n$ to be an appropriate "tiny" modification of $\phi_{n-1}$.
Since $\phi_{n-1}' > 0$, there is a unique number $\tilde{x}_n$ such that $$\tilde{x}_n + \phi_{n-1}(v(\tilde{x}_n)) = r_n.$$ We choose a non-negative, compactly supported, smooth function $\psi_n$ on $(0, 1)$ such that $\psi_n(v(x_j)) = 0$ for $j = 1, \ldots, n - 1$, $\psi_n(v(\tilde{x}_n)) > 0$, and $\|\psi_n'\|_\infty < 3^{-n}$. For every $\epsilon \in [0, 1)$ we have $$\phi_{n-1}' + \epsilon \psi_n' \ge 1 - (3^{-1} + \ldots + 3^{-(n-1)} + \epsilon 3^{-n}) > 0 ,$$ and hence there is a unique solution $x$ of $$x + \phi_{n-1}(v(x)) + \epsilon \psi_n(v(x)) = r_n .$$ This solution depends continuously on $\epsilon$, and since $\psi_n(v(\tilde{x}_n)) \ne 0$, it indeed changes with $\epsilon$. By the intermediate value property, there is an $\epsilon$ such that the corresponding solution $x$ is a rational number. For this $\epsilon$ and $x$, we set $$ \phi_n = \phi_{n-1} + \epsilon \psi_n $$ and $x_n = x$. This way, we have constructed $\phi_n$ with all the desired properties.
By construction, $\psi_n$ converges in $C^1$ to an increasing homeomorphism $\psi : (0, 1) \to (0, 1)$ with the desired properties; namely, $|\psi' - 1| < 1/2$, and $x_n + \psi(v(x_n)) = r_n$.
Step 4. We set $w(x) = \psi(v(x))$. Then $w : \mathbb{R} \to (0, 1)$ is differentiable everywhere in the extended sense, $w'(x) = \psi'(v(x)) v'(x) \in (0, \infty]$, $w'(x) = \infty$ for every rational $x$, and $$x_n + w(x_n) = r_n.$$ Now we follow Pietro Majer's argument: we set $g : (0, 1) \to \mathbb{R}$ to be the inverse function of $w$, $h(x) = g(x) + x$, and $f : \mathbb{R} \to (0, 1)$ to be the inverse function of $h$.
Clearly, $g$, $h$ and $f$ are everywhere differentiable, with $g' \geqslant 0$, $h' \geqslant 1$ and $0 < f' \leqslant 1$, respectively.
Observe that $h(w(x_n)) = g(w(x_n)) + w(x_n) = x_n + w(x_n) = r_n$. Since $w'(x_n) = \infty$ for every $n$, we have $g'(w(x_n)) = 0$, and hence $h'(w(x_n)) = 1$. Thus, $f'(r_n) = f'(h(w(x_n)) = 1 / h'(w(x_n)) = 1$. Since $r_n$ exhaust all rational numbers, $f'(x) = 1$ for every rational $x$.
-
-
$\begingroup$ I haven't read carefully, but can we arrange $f(0)=0$ and $\lim_{+\infty}f$ arbitrarily small positive number? $\endgroup$– YCorCommented Dec 2, 2019 at 16:51
-
$\begingroup$ @YCor: Yes, the interval $(0, 1)$ for the range of $f$ is completely arbitrary, and $x \mapsto f(x) - f(0)$ is an equally good example. $\endgroup$ Commented Dec 2, 2019 at 21:35