This question arose from my thinking about order-preserving isomorphisms on the Real numbers. These functions must be injections (“one-to-one”) $$a\ne b \implies f(a)\ne f(b)$$ and preserve order $$a\le b \implies f(a)\le f(b)$$

In other words, I knew that these functions must be strictly increasing, perhaps with a finite number of points with zero slope. For if there were any intervals $(a,b)$ on which the function had zero or decreasing slope, it would no longer be an injection or preserve order: two different inputs would have the same output.

But then I realized that even an infinite, albeit countable, number of points would suffice, for example, a function that is increasing everywhere except at every integer; something like $f(x)=x+sin(x)$. This function’s graph looks like a smoothed-out staircase, and its derivative $f^{\prime}(x)=1+cos(x)$ is tangent to the $x$-axis at those points.

So then taking a step further, I wondered if perhaps an uncountable number of points would suffice, provided that these points weren’t “together in a straight line”, or put more precisely, they didn’t make an interval.

My first thought was an uncountable set of irrational numbers within an interval, say, $[1,2]$. This function would be increasing at all rational points, but have zero slope at all irrational points. (There are uncountably many irrational numbers within $[1,2]$.) If such a function exists, and is continuous and differentiable, its derivative would be something like the Dirichlet function: $$ g^{\prime}(x) = \begin{cases} 1 &: &x < 1\\ 1 &: &x \in [1,2]\cap\mathbb{Q}\\ 0 &: &x \in [1,2]-\mathbb{Q}\\ 1 &: &x > 2\\ \end{cases} $$

I'm not sure if that’s possible. If it isn’t, then the points in my uncountable set must all be isolated, that is, there must exist an $\epsilon$-neighborhood around each point that doesn’t contain any other points in the set.

Is it possible to select uncountably many isolated points?

  • $\begingroup$ No. It's not possible to have uncountably many isolated points on $\mathbb{R}$. $\endgroup$ – Jacky Chong Oct 3 '16 at 4:41

For $x\in\mathbb R$ and $\epsilon\gt0$ let $B(x,\epsilon)$ denote the $\epsilon$-neighborhood $(x-\epsilon,x+\epsilon).$ A set $X\subseteq\mathbb R$ is discrete if, for each point $x\in X,$ there is a number $\epsilon\gt0$ such that $B(x,\epsilon)\cap X=\{x\}.$

Every discrete subset of $\mathbb R$ is countable.

Proof. Assume for a contradiction that there is an uncountable discrete set $X\subseteq\mathbb R.$ For $n\in\mathbb N$ let $X_n=\{x\in X:B(x,\frac1n)\cap X=\{x\}\}.$ Then $X=\bigcup_{n\in\mathbb N}X_n,$ so $X_n$ is uncountable for some $n.$

Choose $n\in\mathbb N$ so that $X_n$ is uncountable. Then the sets $B(x,\frac1{2n}),\ x\in X_n$ constitute an uncountable family of pairwise disjoint open intervals, which is impossible.

However, the fact that discrete subsets of $\mathbb R$ are countable seems unrelated to your question about slopes of order-isomorphisms of $\mathbb R'$

There is a strictly increasing continuously differentiable bijection $f:\mathbb R\to\mathbb R$ such that $\{x:f'(x)=0\}$ is uncountable and even has positive Lebesgue measure.

Proof. Let $C$ be a compact nowhere dense subset of $\mathbb R$ with positive Lebesgue measure, e.g., a so-called fat Cantor set. Define $$g(x)=d(x,C)=\min\{|x-y|:y\in C\},$$ the distance from $x$ to $C,$ and define $$f(x)=\int_0^xg(t)dt.$$ Since $g$ is continuous and nonnegative, $f$ is continuously differentiable and nondecreasing. In fact, since $C$ is nowhere dense, $\{x:f'(x)\gt0\}$ is everywhere dense, whence $f$ is strictly increasing. Since $f(-\infty)=-\infty$ and $f(+\infty)=+\infty,\ $ $f$ is a bijection. Finally, the set $\{x:f'(x)=0\}=\{x:g(x)=0\}=C$ has positive Lebesgue measure.

| cite | improve this answer | |
  • $\begingroup$ interesting. so one would think that discrete sets are necessarily countable, but not all discrete sets. For example, the set $\omega_{1}$ of all countable ordinal numbers is uncountable yet discrete. (in the sense that there are successive elements with nothing in-between) $\endgroup$ – chharvey Oct 8 '16 at 23:44
  • 1
    $\begingroup$ (1) In the spaces we're most familiar with, i.e. the Euclidean spaces $\mathbb R^n$ with the usual topology, discrete sets are necessarily countable. (2) The ordinal space $\omega_1$ (with the order topology) is not discrete, as it has non-isolated points; namely, all the countable limit ordinals, such as $ \omega,\ \omega2=\omega+\omega, \omega3=\omega+\omega+\omega,\dots,\omega^2,$ etc. The space $\omega_1$ does, however, have an uncountable discreta subset, namely, the set of all countable non-limit ordinals. $\endgroup$ – bof Oct 9 '16 at 0:02
  • 1
    $\begingroup$ ah gotcha. any limit ordinal is not isolated because for every ordinal less than that limit ordinal, there's another one between them, correct? $\endgroup$ – chharvey Oct 9 '16 at 2:02
  • $\begingroup$ Yes, that's right. $\endgroup$ – bof Oct 9 '16 at 2:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.