Suppose f and g are two Borel function on [0, 1]. The push-forward of the Lebesgue measure on [0,1] by f and by g are the same. Then are there some Borel measurable function from [0,1] to [0,1], such that

f = g(h)?

  • $\begingroup$ I know If f and g are simple functions, the answer is yes. $\endgroup$ – Banan.SUN Jul 25 '18 at 2:00
  • $\begingroup$ Do you mean $f = g\circ h$ almost everywhere? I can give a simple counterexample if you demand $f(t) = g(h(t))$ for all $t$. $\endgroup$ – Nik Weaver Jul 25 '18 at 3:45

Assuming push-forward means this ...

Hint... Consider the two maps $f(x) = \{2x\}$ and $g(x) = \{3x\}$, where the brackets are the fractional part. What would $h$ be then?



  • $\begingroup$ I thought of this too, but then I tried to show no $h$ could exist and discovered that it's easy to find an $h$ that works. $\endgroup$ – Nik Weaver Jul 25 '18 at 14:04
  • $\begingroup$ Perhaps we have to distinguish between the problem in the title (measure preserving) and the problem in the text that does not say $h$ is measure preserving. $\endgroup$ – Gerald Edgar Jul 25 '18 at 14:46
  • $\begingroup$ Ah, I didn't notice that. Yes, you certainly have a counterexample for $h$ measure preserving. The problem in the text seems more subtle --- here I feel the existence of $h$ should follow from some measurable selection theorem which could only fail for really pathological $g$, if it fails at all. $\endgroup$ – Nik Weaver Jul 25 '18 at 15:56
  • $\begingroup$ For the case where $h$ need not be measure preserving, the hypothesis that $f, g$ have the same distribution is irrelevant: it is enough that $f$ and $g$ have the same range. But, as Nik notes, the selection $h$ may fail to be Borel measurable: but it may be chosen universally measurable. $\endgroup$ – Gerald Edgar Jul 25 '18 at 16:53

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.