Please help me to solve this question:

Suppose $f:[a,b] \to \Bbb R$ satisfies:

  • $f''(x)+f(x)>0$ and $f(x)>0$ for all $x\in(a ,b)$;
  • $f(a)=f(b)=0$.

Prove that $b-a>\pi$.

Thanks in advance.

  • 2
    $\begingroup$ Just a remark to point out that the example $f(t) = \sin \alpha t$ on $[0,\pi/\alpha]$, for any $0<\alpha<1$, demonstrates that the inequality $b-a>\pi$ is best possible. $\endgroup$ – Greg Martin Jan 14 '13 at 21:08
  • $\begingroup$ Is $f''$ assumed continuous? $\endgroup$ – Davide Giraudo Jan 14 '13 at 21:45
  • $\begingroup$ this question is from shrif university contest math (1995) and didn't assume f" is continuous i think question is true we cant conclude f" is continuous by these hypothesis $\endgroup$ – M.H Jan 15 '13 at 2:44

I think I have a solution when $f''$ is assumed to be continuous.

When $x\in (a,b)$, we have that $\sin\left(\frac{x-a}{b-a}\pi\right)>0$ so using the first assumption, $$0<\int_a^bf(x)\sin\left(\frac{x-a}{b-a}\pi\right)dx+\int_a^bf''(x)\sin\left(\frac{x-a}{b-a}\pi\right)dx=I_1+I_2\tag{1}.$$ We have, integrating by parts and using $f(a)=f(b)=0$ that $$I_1=-\frac{\pi}{b-a}\left(\int_a^b-f'(x)\cos\left(\frac{x-a}{b-a}\pi\right)dx.\right)$$ Doing the same for $I_2$, we finally get $$\small 0<\left(\frac{(b-a)^2}{\pi^2}-1\right)\int_a^bf'(x)\cos\left(\frac{x-a}{b-a}\pi\right)=\left(\frac{(b-a)^2}{\pi^2}-1\right)\int_a^bf(x)\sin\left(\frac{x-a}{b-a}\pi\right)dx.$$ As $\int_a^bf(x)\sin\left(\frac{x-a}{b-a}\pi\right)dx>0$, we are done.

| cite | improve this answer | |

My answer is under the assumption that $f$ and $f'$ are continuous on $[a,b]$ and $f''$ exists on $(a,b)$.

Define $$g:[a,b]\to \mathbb{R}, \quad x\mapsto f'(x)\sin(x-a)-f(x)\cos(x-a).$$ By definition, $g$ is continuous on $[a,b]$ and differentiable on $(a,b)$. Moreover, since $f(a)=f(b)=0$, $g(a)=0$ and $g(b)=f'(b)\sin(b-a)$. Then by mean value theorem, there exists $c\in(a,b)$, such that $$\frac{f'(b)\sin(b-a)}{b-a}=\frac{g(b)-g(a)}{b-a}=g'(c)=(f''(c)+f(c))\sin(c-a).\tag{1}$$ On the one hand, beacause $f(x)>0$ on $(a,b)$ and $f(b)=0$, $f'(b)\le 0$; on the other hand, $f''(c)+f(c)>0$. Combing these facts with $(1)$, we can conclude that $b-a>\pi$.

Edit: After reading Sanchez's answer, I realized that without assuming that $f$ is differentiable at $a,b$, the proof can be modified as follows.

Assume that $b-a\le \pi$. Then $g$ is differentiable on $(a,b)$ and $$g'(x)=(f''(x)+f(x))\sin(x-a)>0.$$ Therefore, $g$ is strictly increasing on $(a,b)$. As a result, both $$g(a^+):=\lim_{x\to a^+} g(x)=\lim_{x\to a^+}f'(x)\sin(x-a)$$ and $$g(b^-):=\lim_{x\to b^-} g(x)=\lim_{x\to b^-}f'(x)\sin(x-a)$$ exist, and $g(b^-)>g(a^+)$. However, since $f(x)>0$ on $(a,b)$ and $f(a)=f(b)=0$, by mean value theorem, $$\limsup_{x\to a^+}f'(x)\ge\liminf_{x\to a^+}\frac{f(x)-f(a)}{x-a}\ge 0 $$ and $$\liminf_{x\to b^-}f'(x)\le\limsup_{x\to b^-}\frac{f(x)-f(b)}{x-b}\le 0 .$$ It follows that $g(a^+)\ge 0\ge g(b^-)$, a contradiction.

| cite | improve this answer | |
  • $\begingroup$ There is some redundancy in your assumptions. If f'' exists everywhere on $(a,b)$, then $f$ and $f’$ are automatically continuous. $\endgroup$ – Ewan Delanoy Jan 17 '13 at 9:10
  • $\begingroup$ @EwanDelanoy: What I need additionally is the continuity of $f$ and $f'$ at the endpoints $a,b$. $\endgroup$ – 23rd Jan 17 '13 at 9:11
  • $\begingroup$ In fact, your additional assumptions are necessary, see the recent other "answer". $\endgroup$ – Ewan Delanoy Jan 17 '13 at 9:15
  • $\begingroup$ I don’t see how you deduce $b-a \gt \Pi$ from (1). $\endgroup$ – Ewan Delanoy Jan 17 '13 at 9:16
  • $\begingroup$ @EwanDelanoy: Yes, the continuity of $f$ is necessary, but I am uncertain whether the continuity of $f'$ is necessary. If $b-a\le\pi$, then $LHS\le 0$ but $RHS> 0$. $\endgroup$ – 23rd Jan 17 '13 at 9:18

Some information is missing in the question, otherwise there are some counterexamples:

  • Let $a = b = 0$ and $f(0) = 0$. This function has the desired properties (there is no $x \in (a,b)$, but $b - a = 0$. Note that this function is even continuous...

  • Let $f(a) = f(b) = 0$ and $f(x) = 1$ for $x \in (a,b)$. Of course, this function lacks continuity in $a$ and $b$, but it is smooth in $(a,b)$. Now, $a$ and $b$ can be chosen arbitrarily.

| cite | improve this answer | |

Edit: This answer requires that all integrals are well defined, which will be the case if $f''$ is continuous, as in Davide's answer.

Edit 2: Switched back to the original bounds $a$ and $b$ after comment of OP.

By partial integration (twice) we have $$\int_a^b f''(x)\sin(x - a)dx = f'(b) \sin(b-a) - \int_a^b f(x) \sin(x-a) dx$$ and so $$\int_a^b\left(f(x) + f''(x)\right) \sin(x-a) dx = f'(b) \sin(b-a).$$ Since $f'(b) \leq 0$ this is impossible if $b-a < \pi$.

| cite | improve this answer | |
  • $\begingroup$ thanks for your answer. a is arbitrary number but you assume a is zero $\endgroup$ – M.H Jan 15 '13 at 2:53
  • $\begingroup$ @MaisamHedyelloo The problem is translation invariant. That's what the "wlog" refers to. You can keep the bounds $a$ and $b$ unchanged and work with $\sin(x-a)$ instead if you want. $\endgroup$ – WimC Jan 15 '13 at 5:45
  • $\begingroup$ This answer is clever. However, as you pointed out, it requires that f′′ is integrable, or, more specifically, f′′(x)sin(x−a) is integrable. I think that this doesn't have to be true given the conditions above. Any thoughts? $\endgroup$ – thang Jan 15 '13 at 8:07
  • $\begingroup$ @thang: A continuous function on a compact interval is integrable. $\endgroup$ – JavaMan Jan 16 '13 at 20:53
  • $\begingroup$ @JavaMan, but f'' is not guaranteed to be continuous. f and f' are by virtual of being differentiable, but we don't know anything about f''. There are some examples of derivatives of functions that are not integrable. $\endgroup$ – thang Jan 16 '13 at 21:17

Clearly, the statement can only be true if $a < b$ and $f$ is continuous on $[a,b]$. We will work with this assumption.

Lemma 1 $\liminf_{x \to a^+} f'(x)$ is bounded below.

Lemma 2 $\liminf_{x \to b^-} f'(x) \leq 0$.

Proof of the problem assuming the lemma

Assume the contrary that $b-a \leq \pi$. Consider $g(x) = \sin (x-a)f'(x) - f(x) \cos (x-a)$. Then $\lim_{x \to a^+} g(x) \ge 0$ by lemma 1, and for $x \in (a,b)$,

$$g'(x) = (f(x) + f''(x)) \sin(x-a) > 0$$

since $b-a \leq \pi$. Therefore $g$ is increasing on $(a,b)$, so $g > 0$ on this interval, which implies that $\displaystyle \lim_{u \to b^-} g(u) > 0$. But if $u_n \to b^-$, $lim_{n \to \infty} f'(u_n) \leq 0$, then

$$\lim_{n \to \infty} g(u_n) = \lim_{n \to \infty} \sin(u_n - a)f'(u_n) - \lim_{n \to \infty} f(u_n) \cos (u_n - a) \leq 0$$

by lemma 2. Contradiction.

Proof of Lemma 1 For the sake of contradiction assume that $\liminf_{x \to a^+} f'(x) = -\infty$.

Sublemma This implies that $\lim_{x \to a^+} f'(x) = -\infty$.

Proof Let $f(x) < M$ on $[a,b]$. The condition $f''(x) > -f(x)$ on $(a,b)$ gives that for $u < v \in (a,b)$, $$f'(v) - f'(u) = \int_u^v f''(x)dx > -\int_u^v f(x)dx > -M(v-u)$$ Let $1 > \epsilon > 0$ be suitably small. If the limit does not exist, then there exists $u_n \in (a,a+\epsilon)$, $u_n \to a^+$ satisfies $\lim_{n \to \infty} f'(u_n) > C$ for some constant $C$. Then for any $v \in (a,a+\epsilon)$, for large enough $n$ we have $u_n < v$, then $$f'(v) - f'(u_n) > -M(v - u_n) > -M\epsilon$$ which implies that $f'(v) > C - M\epsilon$ for $v$ close to $a$, contradicting $\liminf_{x \to a^+} f'(x) = -\infty$.

Now suppose that for $1 > \epsilon > 0$, and $x \in (a,a+\epsilon)$, $f'(x) < N \ll 0$. Then for $x < y \in (a,a+\epsilon)$, $$f(y) - f(x) = f'(c)(y-x) < N(y-x) < N\epsilon < 0$$ for some $c$ between $x,y$. By continuity at $a$, $f(y) = f(y) - f(a) \leq N\epsilon < 0$, contradicting $f(y) > 0$ for all $y \in (a,b)$.

Proof of Lemma 2 For the sake of contradiction, assume that for $u \in (b - \epsilon, b)$, $f'(u) > N > 0$ where $N$ is a constant, and $\epsilon > 0$ is sufficiently small. Then for $u < v \in (b-\epsilon, b)$, $$f(v) - f(u) = f'(c)(v-u) > N(v-u)$$ for some $c$ between $u$ and $v$. Fix $u$, and let $v \to b^-$, we then get $$0 > -f(u) = f(b) - f(u) \ge N(b-u) > 0$$ Contradiction.

| cite | improve this answer | |
  • 1
    $\begingroup$ I’m afraid there is a sign problem with your computations : I have computed that $\int_a^u g(x) dx =\sin(u-a)f(u)-2\int_a^u \cos(x-a)f(x) dx$. So your integration by parts doesn't work the way you intended. $\endgroup$ – Ewan Delanoy Jan 17 '13 at 16:50
  • $\begingroup$ @EwanDelanoy, sorry you are right. The proof is now fixed. $\endgroup$ – user27126 Jan 17 '13 at 18:07
  • $\begingroup$ I still see a sign problem with your new final argument. The number $b-a$ could be strictly between $\frac{\pi}{2}$ and $\pi$ ; in that case, $-f(b)\cos(b-a)$ is positive, and you cannot deduce that $\lim_{n\to +\infty} -f(u_n)\cos(u_n-a) \lt 0$. $\endgroup$ – Ewan Delanoy Jan 18 '13 at 7:24
  • $\begingroup$ @EwanDelanoy, $f(b) = 0$ is a given condition, so sign doesn't matter - the limit is 0. Here I am using the continuity of $f$ at $b$ - the statement is clearly false is $f$ isn't continuous at $b$, by modifying Greg Martin's functions to make $f(b) = 0$. $\endgroup$ – user27126 Jan 18 '13 at 7:37

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.