Suppose $f$ is a non-decreasing continuous function from $[a,b]$ to $\mathbb{R}$, and $\lambda$ is the Lebesgue measure in $\mathbb{R^1}$. Also, $f$ satisfies the property that $f$ maps Lebesgue measurable set to Lebesgue measurable set. Prove or disprove $\nu(E)=\lambda(f(E))$ is a measure.

Based on the hypothesis above, if $f$ also maps zero measurable set to zero measurable set, prove or disprove $\nu(E)=\lambda(f(E))$ is a measure.

For the first question, I don't believe $\nu$ is a measure, but I don't have a good example at hand. For the second question, I learned from somewhere that $f$ is actually absolutely continuous, thus $\nu$ has to be a measure. But I don't know how to prove this directly without proving $f$ is absolutely continous. Any ideas would be appreciated.

  • $\begingroup$ With "measurable set", do you mean Lebesgue measurable or Borel measurable? $\endgroup$ – PhoemueX Jan 26 '15 at 23:06
  • $\begingroup$ @PhoemueX, I mean Lebesgue measurable sets. $\endgroup$ – student Jan 26 '15 at 23:26

We first note that a continuous function $f : [a,b] \to \Bbb{R}$ maps Lebesgue measurable sets to Lebesgue measurable sets iff it maps null-sets to null-sets.

For one implication note that (by regularity) every measurable set $M \subset [a,b]$ can be expressed as a $\sigma$-compact set plus a null-set, i.e. $M = N \cup \bigcup_n K_n$ with $K_n$ compact and $N$ a null-set. Hence,

$$ f(M) = f(N) \cup \bigcup_n f(K_n) $$

is measurable, since $f(N)$ is a null-set by assumption and $f(K_n)$ is compact and hence measurable.

For the converse, assume that there is a null-set $N \subset [a,b]$ such that $f(N)$ is not a null-set, i.e. of positive measure. It is known (see Positive outer measure set and nonmeasurable subset) that every set of positive (outer) measure contains nonmeasurable sets. Hence, there is a nonmeasurable $V \subset f(N)$. But since $V$ is contained in $f(N)$, we have

$$ V = f(f^{-1}(V) \cap N), $$ where $f^{-1}(V) \cap N$ is a subset of a null-set and hence measurable. By assumption, this yields that $V$ is measurable as the image of a measurable set under $f$, contradiction.

Next, let us show that $\nu$ is always a measure as long as $f(E)$ is measurable for all measurable $E$. This even holds without continuity assumptions on $f$.

First, since $f$ is nondecreasing, we know that $I_y := f^{-1}(\{y\}) \subset [a,b]$ is an interval for all $y \in \Bbb{R}$. Since the $(I_y)_y$ are pairwise disjoint, there are at most countably many $(y_n)_n$ for which $I_{y_n}$ contains more than one point (why exactly?).

Note that $L := \bigcup_n I_{y_n}$ is measurable as a countable union of intervals.

Now $f$ is injective on $M := [a,b] \setminus L$ (which might be the empty set), because if there were $x,y \in M$ with $x<y$ but $z := f(x) = f(y)$, this would imply that $x,y \in I_z$ and hence $z = y_n$ for some $n$. But $x,y \in I_z = I_{y_n}$ in contradiction to $x,y \in M$.

Finally, we have

$$f(L) = \bigcup_n f(I_{y_n}) \subset \bigcup_n \{y_n\},$$ which is a countable set. But this implies

$$ \nu(E) = \lambda(f(E)) = \lambda(f(E \setminus L)) = \lambda(f(E \cap M)) =: \gamma(E) $$ for each measurable $E \subset [a,b]$.

Using the fact that $f$ is injective on $M$, it is easy to see that $\gamma$ is a measure, since if $(E_n)_n$ is a pairwise disjoint sequence, then so is $(E_n \cap M)_n $ and thus also $(f(E_n \cap M))_n$, which implies

$$ \gamma(\bigcup_n E_n) = \lambda(\biguplus_n f(E_n \cap M)) = \sum_n \lambda(f(E_n \cap M)) = \sum_n \gamma(E_n). $$ Hence, $\nu = \gamma$ is always a measure.

As observed at the beginning, we must always have that $f(E)$ is a null-set if $E$ is. Hence, $\nu \ll \lambda$, where $\lambda$ is Lebesgue measure.

By the Lebesgue-Radon-Nikodym theorem, we get $f \in L^1 ([a,b])$ with

$$ \nu(E) = \int_E f \, dt. $$

In particular, if $f$ is continuous, then

$$ \lambda([f(x), f(y)]) = \nu([x,y]) = \int_x^y f(t) \, dt $$ holds for $a \leq x < y \leq b$, so that $f$ is absolutely continuous.

  • $\begingroup$ Great answer! Great job! $\endgroup$ – student Jan 28 '15 at 1:32
  • $\begingroup$ I'll try to set a bounty tomorrow and give you more points. This answer deserves this. $\endgroup$ – student Jan 28 '15 at 1:38

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.