Find all $f\in C^1(\mathbb R,\mathbb R)$ such that $f^2+(1+f')^2\leq 1$

It's quite likely the answer is $f=0$.

Note that $|f|\leq 1$ and $-2\leq f'\leq 0$.

Therefore $f$ is decreasing and bounded.

What then ? I tried contradiction, without success.

  • $\begingroup$ Have you thought about the limits as $x$ approaches $\pm \infty$? Remember that its values are decreasing and lie in a bounded region. $\endgroup$ – Kyle Jun 12 '14 at 18:53
  • 1
    $\begingroup$ well $-k\arctan(x)$ with $k$ small enough it seems for me satisfies your condition. Say $-1/\pi$. Then function is from $[-1/2,1/2]$ with negative derivative $>-2$. Something like that. $\endgroup$ – Alexander Vigodner Jun 12 '14 at 18:55
  • 2
    $\begingroup$ @AlexanderVigodner with $k=0.01$ it fails at $x=13$. $\endgroup$ – Gabriel Romon Jun 12 '14 at 19:17

The equation is equivalent to $$ f^2+2f'+f'^2\le0\tag{1} $$ Since $f^2+2f'\le0$, where $f\ne0$, we have $$ (1/f)'\ge\color{#C00000}{1/2}\tag{2} $$ If $f(x_0)=a\gt0$, then $\dfrac1f(x_0)=\dfrac1a\gt0$ and $(2)$ says that $$ \frac1f\left(x_0-\frac3a\right)\le\frac1f(x_0)-\color{#C00000}{\frac12}\frac3a\lt0\tag{3} $$ as long as $\dfrac1f$ doesn't pass to $-\infty$ in $\left[x_0-\frac3a,x_0\right]$.

In any case, on $\left[x_0-\frac3a,x_0\right]$, $\dfrac1f$ must pass through $0$, which is impossible because $f\in C^1(\mathbb{R})$.

If $f(x_0)=a\lt0$, then $\dfrac1f(x_0)=\dfrac1a\lt0$ and $(2)$ says that $$ \frac1f\left(x_0-\frac3a\right)\ge\frac1f(x_0)-\color{#C00000}{\frac12}\frac3a\gt0\tag{4} $$ as long as $\dfrac1f$ doesn't pass to $\infty$ in $\left[x_0,x_0-\frac3a\right]$.

In any case, on $\left[x_0,x_0-\frac3a\right]$, $\dfrac1f$ must pass through $0$, which is impossible because $f\in C^1(\mathbb{R})$.

Therefore, $f(x)=0$ for all $x\in\mathbb{R}$.

  • $\begingroup$ Can you elaborate a bit on your argument following $(2)$? $\endgroup$ – Alex Schiff Jun 13 '14 at 1:30
  • 1
    $\begingroup$ @AlexSchiff: Any function with a slope at least $\color{#C00000}{1/2}$ (e.g. $1/f$) must pass through the $x$-axis at a finite point. $\endgroup$ – robjohn Jun 13 '14 at 1:40
  • $\begingroup$ What is $f=0$ how you can use (2) then ? $\endgroup$ – Alexander Vigodner Jun 13 '14 at 5:18
  • $\begingroup$ @AlexanderVigodner: $f=0$ satisfies $(1)$. However, $(2)$ obviously holds wherever $f(x)\ne0$. $\endgroup$ – robjohn Jun 13 '14 at 6:12
  • 2
    $\begingroup$ @JackD'Aurizio: I have edited my answer to account for the case that $1/f$ may blow up inside an interval. $\endgroup$ – robjohn Jun 13 '14 at 14:28

Since $f(x)$ is bounded and decreasing both $\lim_{x \rightarrow \infty} f(x)$ and $\lim_{x \rightarrow -\infty} f(x) $ exist. If $f(x)$ were not identically zero, then at least one of these limits is nonzero. Say it is the first one, and call the limit $L$.

By the mean value theorem, $f(n+1) - f(n) = f'(x_n)$ for some $x_n$ between $n$ and $n + 1$. The left-hand side of this equation converges to $L - L = 0$ as $n$ goes to infinity, so we have $$\lim_{n \rightarrow \infty} f'(x_n) = 0$$ But we also have $$\lim_{n \rightarrow \infty} f(x_n) = L$$ Plugging $x_n$ into $f(x)^2 + (1 + f'(x))^2 \leq 1$ and taking limits as $n$ goes to infinity gives $L^2 \leq 0$, a contradiction.

A similar argument works if $\lim_{x \rightarrow -\infty} f(x) \neq 0$.

  • $\begingroup$ This is intuitive, the simplest argument so far. $\endgroup$ – Gabriel Romon Jun 13 '14 at 14:41
  • $\begingroup$ And this doesn't use the continuity of the derivative. Nice. $\endgroup$ – Gabriel Romon Jun 13 '14 at 14:46
  • 1
    $\begingroup$ You don't have to reason by contradiction, because you end up with $L^2 \leq 0 \implies L = 0$ anyway. $\endgroup$ – Najib Idrissi Jun 13 '14 at 14:55

Hints: As you mentioned, $f$ is decreasing and bounded. Think about $\lim_{n \to \infty} f(n)$. Must this limit exist? What does this imply for the limit of the derivative $f'$?

Full Solution. The function $f(x)$ is decreasing and bounded, so $\lim_{x \to \infty} f(x)=L$ for some $L \in [-1,1]$. For the sake of contradiction, we suppose $|L|>0$. To set up the contradiction, we relate $|f(x)|$ and $f'(x)$: Let $\epsilon\in (0,1]$, and suppose that we have $0 \geq f'(x) \geq -\epsilon$ for some $x \in \mathbb{R}$. Then \begin{align*} f^2(x) & \leq 1-(1+f'(x))^2\\ &\leq -2f'(x) - (f'(x))^2 \\ &\leq -2f'(x) \\ & \leq 2\epsilon. \end{align*} Thus $|f(x)| \leq \sqrt{2\epsilon}$. Therefore we know that if $|f(x)| > \sqrt{2\epsilon}$, then $f'(x) <-\epsilon$. For sufficiently large $x$, we must have $|f(x)| > |L|/2=\sqrt{2(|L|^2/8)}$, hence $f'(x) <-|L|^2/8$. This contradicts the fact that $f(x)$ is bounded below. An entirely analogous argument shows that $\lim_{x \to -\infty} f(x)=0$. Monotonicity implies $f=0$.QED

  • $\begingroup$ Although it should be a cakewalk, I fail to prove that for a $C^1$ function with a limit at infinity, its derivative must have $0$ as limit at $\infty$... $\endgroup$ – Gabriel Romon Jun 12 '14 at 19:05
  • $\begingroup$ Have you first shown that the limit of the derivative exists? $\endgroup$ – Kyle Jun 12 '14 at 19:11
  • $\begingroup$ I don't think you can prove the existence of the limit for the derivative with usual theorems. The only information about it is boundedness. $\endgroup$ – Gabriel Romon Jun 12 '14 at 19:14
  • 2
    $\begingroup$ And now that I remember this, math.stackexchange.com/questions/788813/… this is definitely wrong. $\endgroup$ – Gabriel Romon Jun 12 '14 at 19:19
  • 1
    $\begingroup$ Another thought would be to compare the limits as $x$ approaches positive and negative infinity. There are some easy restrictions there. $\endgroup$ – Kyle Jun 12 '14 at 19:25

I think I have to completely rewrite the solution keeping the wrong one above untouched. As I said before I am sure we can build such function, and I think I did it using the ODE in the above solution in the end. I am building a counterexample function: Fistly, $f(x)= 0$ for $x\le 0$;

Now I'd like to build a simple function satisfying condition $$ f^2+(1+f^\prime)^2\le 1 $$ on some interval $[0,x^*]$.
I define $$ g(t)=-x^3/3-x^2/2\\ g^\prime(x)=-x^2-x $$ I is obvious that in some positive neighborhood $(0,\epsilon)$ $$ g^2+(1+g^\prime)^2 =O(\epsilon^4)+1-O(\epsilon) \le 1 $$ It is obvious also that staring from some $x^*$ $$ g^2+(1+g^\prime)^2\ge 1 $$ Besides we have this point $x^*$ is such that $$ g^2(x^*)+(1+g^\prime(x^*))^2= 1 $$ Let's check condition $g^\prime(x^*)>-1$. It is easy t estimate that $x^*$ is about $0.9$ and that then $g^\prime(x^*)< -1

Now again consider the differential equations $$ f^\prime=-1+\sqrt{1-f^2}\\ f^\prime=-1-\sqrt{1-f^2}\\ $$ and choose the second one in accordance with the sign $g^\prime(x^*)+1$. The solution of this equation with initial condition $f(x^*)=g(x^*)$ will extend our function on $R$. So the final function $f(x)$ is $$ f(x)=0 ~ if ~ x\le 0 \\ f(x)=g(x) ~ if~ 0<x\le x^* \\ solution~ of~ f^\prime=-1-\sqrt{1-f^2}, f(x^*)=g(x^*) ,~ x\ge x*\\ $$

Since I did not find any mistake in the proof that such function cannot exist I again probably made mistake somewhere. But I cannot find it. Any comments please. May be this ODE cannot have a solution on $R$? Lipschitz condition is not satisfied.


FIX Well the solution of differential equation with non zero initial condition will satisfy your property $$ f'=-1-\sqrt{1-f^2} $$ Edit Notice that equation $$ f'=-1+\sqrt{1-f^2} $$ is also OK. Now let's change $\tau = -t$ and rewrite the second equation as function of $\tau$ $$ f_\tau^\prime =1-\sqrt{1-f^2} $$ Let's build now the function on $R^+$ as a solution of the first eqaution and on $R^-$ as a solution of the third equation. To guarantee differentiability in $t=0$ let's make equal derivatives at time $t=\tau=0$. $$ f^\prime_t=-1-\sqrt{1-f^2}=-f^\prime_\tau=-1+\sqrt{1-f^2} $$ So with initial condition $f=1$ we can propagate this function on $R$. I changed signs here. Now the initial derivative is

  • $\begingroup$ Can you guarantee that it will exist on all of $\mathbb{R}$ if $f$ is not $\equiv 0$? $\endgroup$ – Daniel Fischer Jun 12 '14 at 19:24
  • $\begingroup$ I think we can play a bit with this equation to guarantee this. Let me think $\endgroup$ – Alexander Vigodner Jun 12 '14 at 19:27
  • $\begingroup$ You can't have $f(x) = 1$ for any $x$. That would imply $f'(x) = -1$, and thus $f(y) > 1$ for $x-\varepsilon < y < x$. $\endgroup$ – Daniel Fischer Jun 12 '14 at 19:41
  • $\begingroup$ Who said this. I said it is 1 in the inital point. After this you extend it in both direction via equations. Extention is built on the different equation it is not symmetric. $\endgroup$ – Alexander Vigodner Jun 12 '14 at 19:43
  • $\begingroup$ OK. I am wrong about negative direction. But something can be done, I am sure. I don't belieive f=0. $\endgroup$ – Alexander Vigodner Jun 12 '14 at 19:47

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.