I'm aware through a plethora of counterexamples that for two functions $f$, $g$ that are absolutely continuous, their composition is not necessarily so. However, before seeing these examples, I had written out what I thought was a proof confirming that it was true, and I'm not sure where my error is. My proof is as follows:

Assume that $f$ and $g$ are absolutely continuous. Let $\epsilon_f > 0$. Since $f$ is AC, there exists a $\delta_f$ such that for any finite disjoint collection $\{(a_k,b_k)\}$

$$ \sum_{k=1}^n [b_k - a_k] < \delta_f \Longrightarrow \sum_{k=1}^n |f(b_k) - f(a_k)| < \epsilon_f$$

Since $g$ is absolutely continuous as well, let $\epsilon_g = \delta_f$. Then there exists $\delta_g$ such that

$$ \sum_{k=1}^n [b_k - a_k] < \delta_g \Longrightarrow \sum_{k=1}^n |g(b_k) - g(a_k)| < \epsilon_g = \delta_f \Longrightarrow \sum_{k=1}^n |f(g(b_k)) - f(g(a_k)| < \epsilon_f $$

Thus, the composition is AC. I realize that the only difference between this is that in the composition, for the domain of $f$, I'm assuming that the sum $|g(b_k) - g(a_k)|$ in absolute value is less than $\delta_f$. However, I'm not quite sure why that absolute value should matter in the case of continuity. It's left out of the original definition because it's assumed that $b_k > a_k$. Other than that, though, I don't see any errors in logic. Thoughts?


First, note that in the definition we take $\sum_{k=1}^n |b_k-a_k|<\delta$ and not $>$. The other problem is that $(g(a_k),g(b_k))$ are not necessarily disjoint intervals, as required in the definition of absolute continuity.

  • $\begingroup$ Good catch on the inequality. A typo proliferated through copy and pasting. Ok, excellent, that is exactly what I was not seeing. Thanks for your answer! $\endgroup$
    – cnolte
    Aug 10 '16 at 3:53

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.