Counterexample to implications of mean value theorem over the rationals that can be extended to a differentiable function on the reals

176 Views Asked by At

This is a follow-up question to a previous question of mine. The previous one was answered (in fact it was a duplicate), but I still didn't feel like my inquiry was over.

The original question was about the mean value theorem on a space which is not complete, in this case $\mathbb{Q}$. We can easily find a function $f\colon\mathbb{Q}\to\mathbb{R}$ which doesn't satisfy the mean value theorem by choosing a function with $f(a)=f(b),a,b\in\mathbb{Q}$ and the only maximum or minimum $x_m\in (a,b)\cap (\mathbb{R}\setminus\mathbb{Q})$.

My question now was: Is there a function $f\colon\mathbb{Q}\to\mathbb{R}$, differentiable everywhere (on $\mathbb{Q}$) with a nonnegative derivative $f'(x)\ge 0$ for every $x\in\mathbb{Q}$, but which is not monotone increasing.

It turns out that finding a such a function is a bit more difficult, but conceptually not much different - we simply take a function $f\colon\mathbb{R}\to\mathbb{R}$ that has a jump discontinuity at some irrational point but is constant everywhere else, then take its restriction $f\mid_{\mathbb{Q}}$.

At this point I wasn't satisfied, because it felt pathological to force a function into discontinuity in its (mostly continuous) extension to the reals. I started wondering whether I could construct a function that could be continuously extended to the reals and maybe even be differentiable everywhere.

Here I figured that, if it was differentiable everywhere, the derivative could not be continuous, since the derivative was nonnegative on a dense subset, so would have to be nonnegative everywhere. I also noticed that as soon as $f'(x)<0$ for some irrational $x\not\in\mathbb{Q}$, the derivative would have to take every value between $f'(x)$ and $0$ between $x$ and any rational point $p/q$ by the theorem of Darboux, which states that even non continuous derivatives satisfy the intermediate value theorem.

This feels impossible, but I don't know. So I approached the idea from a different angle. How about constructing the derivative and then taking its primitive?

My question now became: Is there a function $f\colon\mathbb{R}\to\mathbb{R}$, such that $f(p/q)=0$ for every rational $p/q$ but $f(x)<0$ for some irrational $x$? This in itself can easily be constructed, but is it still possible if we force the regularity condition that $f$ is integrable and the continuous $F\colon\mathbb{R}\to\mathbb{R}, x\mapsto\int_0^x f(t)\mathrm{d}t$ is differentiable everywhere with derivative $F'=f$ or at least $F'(y)=f(y)<0$ for some irrational $y$? Or maybe that $f$ satisfies the intermediate value theorem?

If you have input or an answer to any of my question, I'd much appreciate the ideas. I myself am stuck here. The idea seems impossible, but I am still missing the argument as to why. Maybe someone here can help. Cheers!

Edit: José Carlos Santos reply showed that my previous conditions on $f$ would not be enough to give a primitive that is not monotone increasing. So I needed to impose more conditions.

2

There are 2 best solutions below

1
On

Take$$\begin{array}{rccc}f\colon&\Bbb R&\longrightarrow&\Bbb R\\&x&\mapsto&\begin{cases}0&\text{ if }x\ne\sqrt2\\-1&\text{ otherwise.}\end{cases}\end{array}$$If you define $F\colon\Bbb R\longrightarrow\Bbb R$ by $F(x)=\int_0^xf(t)\,\mathrm dt$, then $F$ is differentiable everywhere, since it is the null function.

2
On

Update: I'm not sure anymore that this answer is even relevant to the question, let alone helpful. I would have deleted it already except for the value of the comments. I may delete it anyway eventually.


There is a basic conceptual problem with what you are trying to do. You apparently know that there are far more irrational numbers than rational numbers, which is a correct understanding. But you have a misconception that it is possible to find several irrational numbers between one rational number and the "next" one.

The rational numbers are dense in the reals. If you take any two real numbers $x,$ $y$ such that $x<y,$ $f(x)< 0,$ and $f(y)<0$, there is some rational number $p/q$ such that $x < p/q < y.$