Let $f:\mathbb R\to \mathbb R,$ and suppose $f$ is differentiable at every point of a measurable $E\subset \mathbb R,$ with $f'>0$ on $E$.

Suppose also that $m(E)>0$ (where $m$ is Lebesgue measure).

Prove that $m(f(E))>0$.

My proof:

Since $f$ is differentiable then it's continuous and hence it preserves both compact sets and intervals.

Now since $m(E)>0$ we can find compact interval inside it (is this true or not?) If this is true so the proof is completed.

I know that if $E$ is measurable then $E$ is either Borel set or a set of measure zero.

So here $E$ is Borel, but still not necessarily to be an interval.

  • 1
    $\begingroup$ No, the set of irrational numbers has positive measure but doesn't contain any interval. $\endgroup$ – user384138 Jan 7 '17 at 16:28
  • 1
    $\begingroup$ It's not even clear that $f(E)$ is measurable from the given hypotheses. $\endgroup$ – zhw. Jan 10 '17 at 19:29
  • 1
    $\begingroup$ I want to ask the OP: Is this really the problem? is this the exact wording?where does it come from? $\endgroup$ – zhw. Jan 11 '17 at 20:23
  • 1
    $\begingroup$ @charMD can you put your solution ,thanks in advance $\endgroup$ – Sara Suradi Jan 11 '17 at 20:55
  • 1
    $\begingroup$ @zhw. Yes I am almost sure,It's a final exam question , I will ask my prof at the beginning of next semester $\endgroup$ – Sara Suradi Jan 12 '17 at 14:20

I will prove two points : first, in response to a comment of zhw, we check that $f(E)$ is measurable, and then that its measure is $>0$.

We recall that $m^*$ denotes here Lebesgue outer measure, defined for every subset $S \subset \mathbb{R}$. For $I$ an open interval of $\mathbb{R}$, we note $l(I) = \sup I - \inf I$. Then we define $$m^*(S) = \inf \left \{ \sum \limits_{k=1}^{+\infty} l(I_k),\ (I_k)_{k \ge 1} \textrm{ is a sequence of intervals with } S \subset \bigcup \limits_{k \in \mathbb{N}^*} I_k \right \}$$

For both results, the following lemma will be needed.

Lemma : Given $f : \mathbb{R} \rightarrow \mathbb{R}$ differentiable at any point of a set $S$, assuming that there exists $C \ge 0$ such that $\forall x \in S,\ |f'(x)| \le C$, then $m^* \big( f(S) \big) \le C \cdot m^*(S)$.

Proof : Let $\varepsilon > 0$. We define an increasing sequence of sets $(S_n)_{n \ge 1}$ by : $$S_n = \left \{ x \in S,\ \forall t \in S,\ |t-x| \le \frac{1}{n} \Rightarrow |f(t)-f(x)| \le (C+\varepsilon)|t-x| \right \}$$

Using the hypothesis $|f'|\le C$, we get that $S = \bigcup \limits_{n=1}^{\infty} S_n$. For every $n \ge 1$, we can take a sequence of open intervals $(I_{n,k})_{k \ge 1}$ covering $S_n$ and such that $\sum \limits_{k=1}^{+\infty} m^*(I_{n,k}) \le m^*(S_n)+\varepsilon.$

Without loss of generality, we can assume that for every $n$ and $k$, $m^*(I_{n,k}) \le \frac{1}{n}$.

Then, for $n \in \mathbb{N}^*$, for all $k \in \mathbb{N}^*$, for $x,y \in S_n \cap I_{n,k}$, we have $|y-x|\le \frac{1}{n}$ and $x,y \in S_n$, so we can write $|f(y)-f(x)| \le (C+\varepsilon)|y-x|\le (C+\varepsilon)\cdot m^*(I_{n,k})$. Thus, for $n \ge 1$, $$m^*(f(S_n)) \le \sum \limits_{k=1}^{+\infty} m^*\big(f(S_n \cap I_{n,k}) \big) \le \sum \limits_{k=1}^{+\infty} (C+\varepsilon)\cdot m^*(I_{n,k}) \le (C+\varepsilon)\cdot (m^*(S_n)+\varepsilon)$$

Letting $n \to +\infty$ and then $\varepsilon \to 0^+$, we get $m^* \big( f(S) \big) \le C \cdot m^* (S)$.

Now we prove that $f(E)$ is measurable. Classically (see Problem about $G_{\delta}$ and $F_{\delta}$ sets),

Claim 1 : There exists a subset $H \subset E$ which is $F_{\delta}$ (i.e. a countable union of closed sets) such that $N=E \backslash H$ is a null set.

Write $H = \bigcup \limits_{k=1}^{+\infty} F_k$ where the $F_k$ are closed. For $k \ge 1$, for all $M>0$, $[-M,M] \cap F_k$ is compact and $f$ is continuous (because it is differentiable) on this set, so $f([-M,M]\cap F_k)$ is closed (it is a compact set). So for all $k$, $f(F_k) = \bigcup \limits_{M \in \mathbb{N}^*} f\big([-M,M]\cap F_k\big)$ is a Borel set, so $f(H)$ is measurable.

Now we prove that $f(N)$ is a null set. For $k \in \mathbb{N}^*$, we denote $N_k = \{ x \in N,\ f'(x)<k \}$. $N_k \subset N$ so $N_k$ is a null set, so $m^*(N_k)=0$ for $k \ge 1$. Plus, we can use the previous lemma on $N_k$, because $0 \le f' \le k$ on $N_k$, so $m^* \big(f(N_k) \big)\le 0$. Thus $f(N_k)$ is a null set. As $f(E) = f(H) \cup f(N)$, we can conclude that $$f(E) \textrm{ is measurable}.$$

Now back to the original problem : we have some measurable set $E$ with positive measure, $f$ differentiable on $E$, $f'>0$ on E. We suppose that $m \big( f(E) \big)=0$.

For $x \in E$, $f'(x)>0$ so $\frac{f(x)-f(y)}{x-y}>0$ for all $y \in E \backslash \{x \}$ in some neighborhood of $x$. Thus \begin{align*} E & = \bigcup \limits_{q \in \mathbb{Q}} \left \{ x \in E\ | \ \ x > q \ \textrm{ and }\ \forall y \in ]q,x[,\ \frac{f(x)-f(y)}{x-y}>0 \right \}\\ & = \bigcup \limits_{q \in \mathbb{Q}} \left \{ x \in E\ | \ \ x > q \ \textrm{ and }\ \forall y \in ]q,x[,\ f(x)>f(y) \right \} \end{align*}

because $\mathbb{Q}$ is dense. Moreover, $E$ has positive measure and $\mathbb{Q}$ is countable. Hence, there exists $q_0 \in \mathbb{Q}$ such that $B = \left \{ x \in E\ | \ \ x > q_0 \ \textrm{ and }\ \forall y \in ]q_0,x[,\ f(x)>f(y) \right \}$ has positive measure.

Plus, for $(x,y) \in B^2$ with $x<y$, we have $q_0 < x <y$, so $f(y)>f(x)$. Hence $f_{|B}$ is increasing.

Finally it is a well-known fact (see Can we have an uncountable number of isolated points) that $B$ has countably many isolated points, and thus we have a measurable subset $A \subset B$ such that $m(A)=m(B)>0$ and $A$ has no isolated points. Note that we also have $f_{|A}$ increasing, $f'>0$ on A, and $m \big( f(A) \big)=0$.

Now we just need a stronger version of our lemma :

Lemma (bis) : Given $A \subset \mathbb{R}$ with no isolated points, and $f : A \rightarrow \mathbb{R}$, we say that $f$ is differentiable at $x \in A$ whenever $\lim \limits_{t \to x^{\neq}} \frac{f(x)-f(t)}{x-t}$ exists, and we note $f'(x)$ the limit. Assuming that $f$ is differentiable over $A$, and that there exists $C \ge 0$ such that $|f'| \le C$, we have $$m^* \big( f(A) \big) \le C \cdot m^*(A)$$

Proof : the proof is exactly the same as the one we gave for the first lemma.

Finally, we denote $g = f_{|A}$. As $g$ is stricly increasing, $g^{-1}$ is well defined. Moreover, as $f'>0$ on $A$, it is classical (see Inverse functions and differentiation) to show that $g^{-1}$ is differentiable in the sense of the Lemma bis on $g(A)$. As $g(A)$ is a null set, we can use our lemma (as we did with the set $N$ - see above) to conclude that $g^{-1}\big( g(A) \big)$ is a null set, so $A$ is a null set, which is absurd.

$$\textrm{Hence we have }\ m \big (f(E) \big) > 0.$$

| cite | improve this answer | |
  • 1
    $\begingroup$ Thank you CharMD for your valuable answer .I follow up with your answer until the last step where you said that f^-1 is differentiable (where the result follow immediately as you proved above) so how can we guarantee differentiability of f^-1 ? $\endgroup$ – Sara Suradi Jan 12 '17 at 14:10
  • $\begingroup$ Also you mean at the end the set A is it self the set E :)? $\endgroup$ – Sara Suradi Jan 12 '17 at 14:11
  • $\begingroup$ If $f$ is stricly monotone and that $f'(x)$ exists, with $0<f'(x)<+\infty$, then $f^{-1}$ is differentiable at $f(x)$ with derivative $\frac{1}{f'(x)}$. This can be found in various course of calculus (see en.wikipedia.org/wiki/Inverse_functions_and_differentiation) $\endgroup$ – charmd Jan 12 '17 at 14:26
  • $\begingroup$ In the end, I managed to use only classic tools, but still, this is not an easy question for an exam $\endgroup$ – charmd Jan 12 '17 at 16:34
  • $\begingroup$ @charMD It is not necessarily the case that $f|_A$ is increasing. Think of $f(x)=x-\lfloor x\rfloor$ and $A=(0,1)\cup (1,2)$. However, as can be inferred following the proof in the link to the Bogachev book, there exists a subset of $A$ with positive measure on which $f$ is increasing, and from here you can apply the rest of the proof. $\endgroup$ – Del Jan 13 '17 at 23:54

You can prove this by using the following result, which you can find in V. I. Bogachev's Measure Theory book (Springer, 2007). This is Lemma 5.8.13., which I quote almost verbatim but with adapted notation:

Proposition: Let $f$ be a function on $[a,b]$ and let $A$ be the set of all points at which $f$ has a nonzero derivative. Then, for every set $Z$ of measure zero, the set $f^{-1}(Z) \cap A$ has measure zero. In other words, $\lambda \circ f^{-1}|_A \ll \lambda|_A$, where $\lambda$ is Lebesgue measure.

Here is a link to the proof given by Bogachev. It is quite unwieldy (to me, at least) and I must say I haven't gone through the details. The proof relies on Vitali's covering theorem.

Note that the proposition remains true for a function $f$ defined on all of $\mathbb{R}$: just write $\mathbb{R}$ as a countable almost-disjoint union of intervals and apply the proposition to the restriction of $f$ to each interval.

To solve the problem at hand, we argue by contradiction and suppose that $m(f(E)) =0$. Since $E \subset f^{-1}(f(E))$ and $E \subset A$, we have $E \subset f^{-1}(f(E)) \cap A$. Since $f(E)$ has measure $0$, the proposition implies that $f^{-1}(f(E)) \cap A$ has measure $0$. This is a contradiction since we assumed that $m(E) >0$.

Remark: In the above we assumed that $f(E)$ is measurable. This does follow from the hypotheses. See CharMD's great answer for a proof, or alternatively Proposition 5.5.4. in Bogachev's book.

| cite | improve this answer | |
  • $\begingroup$ Have you read Bogachev in its entirety ? :D $\endgroup$ – Gabriel Romon Jan 13 '17 at 12:25
  • $\begingroup$ I haven't; just some sections here and there. I mostly use it as a reference and for the abundant exercises (about a hundred per chapter!). Why do you ask? $\endgroup$ – Dominique R.F. Jan 13 '17 at 16:27

From $f'>0$, there exists $c>0$ such that $f'>c$. Then $$m(f(E))=\int_{E} f' dm>\int_{E} c dm=c\int_{E} 1 dm=cm(E)>0.$$

| cite | improve this answer | |
  • 2
    $\begingroup$ How does m(f(E))=∫f'. ?can you give me a reference? $\endgroup$ – Sara Suradi Jan 8 '17 at 16:30
  • $\begingroup$ @user283366 $m(f(E))=\int_{y\in f(E)}1\ dy.$Substitute $y=f(x).$Then, $ m(f(E))=\int_{x\in E}f'(x)\ dx$ $\endgroup$ – kayak Jan 15 '17 at 13:08
  • $\begingroup$ Isn't the monotonicity of $f$ usually required for such a substitution ? $\endgroup$ – charmd Mar 12 '17 at 12:12

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.