So, I was just asking myself can something like this happen? I was thinking about some everywhere continuous but nowhere differentiable functions $f$ and $g$ and the natural question arose on can the composition $f \circ g$ be differentiable, in other words, can the operation of composition somehow "smoothen" the irregularities of $f$ and $g$ which make them non-differentiable in such a way that composition becomes differentiable?

So here is the question again:

Suppose that $f$ and $g$ are everywhere continuous but nowhere differentiable functions. Can $f \circ g$ be differentiable?

If such an example exists it would be interesting because the rule $(f(g(x))'=f'(g(x)) \cdot g'(x)$ would not hold, and not only that it would not hold, it would not make any sense because $f$ and $g$ are not differentiable.

  • 1
    $\begingroup$ monotone functions must be somewhere differentiable. $\endgroup$
    – Farewell
    Dec 12 '15 at 15:04

No, the composition of two continuous nowhere differentiable functions cannot be everywhere differentiable. To prove this, we need two facts:

  1. A function of bounded variation is differentiable almost everywhere
  2. A differentiable function has bounded variation on some subinterval

Suppose that $f\circ g$ is differentiable. Then it has bounded variation on some interval $[c,d]$. Since $g$ is nowhere differentiable, it is nowhere monotone; thus, we can arrange that $g(c) < g(d)$ by shrinking the interval further. Let $a=g(c)$ and $b=g(d)$.

The function $f$, being nowhere differentiable, has infinite variation on $[a,b]$. That is, for every $M$ we can find a partition $a=x_0<x_1<\dots<x_n=b$ such that $$ \sum_{i=1}^n |f(x_i)-f(x_{i-1})| > M $$ Let $t_i=\inf\{t\in [c,d] : g(t) \ge x_i\}$. The continuity of $g$ implies that $g(t_i)=x_i$ and $c=t_0< t_1<\dots<t_n\le d$. (It may happen that $t_n<d$.) Then $$ \sum_{i=1}^n |f(g(t_i))-f(g(t_{i-1}))| > M $$ and since $M$ was arbitrary, this contradicts $f\circ g$ having bounded variation on $[c,d]$.


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.