7

$f$ is a polynomial and suppose that $f(x)=\sin x$ has infinitely many solutions. Prove that $f$ is a constant between -1 and 1.

I get the problem intuitively, but how would I prove this formally?

Can anyone give me any hints?

zxcvber
  • 2,081
  • 1
    @newnewnewnewnenwwojpwjfpoergje but the solutions could be $f(1/n)=\sin(1/n)$ for every $n$, isn't it? – Surb Jun 16 '16 at 10:33
  • @Surb. If $f$ is analytic and $f(x)=\sin(x)$ at infinitely many $x$ on a compact set, then $f=\sin$. Polynomials are analytic (but there are no polynomials that coincide with $\sin$). – Batominovski Jun 16 '16 at 10:34
  • @Batominovski ? – Surb Jun 16 '16 at 10:34
  • I'm not sure it's correct. For example, $f(x)=x-\frac{x^2}{2}+\frac{x^3}{6}$, then $f(x)=\sin(x)$ has infinitely many solution, but $f$ is not constant. – user330587 Jun 16 '16 at 10:34
  • Yes I thought about polynomials with degree above 1 can diverge but I thought of a polynomial that has the exact same graph as sinx, at least for a interval. I couldn't determine whether that kind of polynomial existed or not. – zxcvber Jun 16 '16 at 10:35
  • @user330587 really? what are these solutions? – Surb Jun 16 '16 at 10:36
  • 1
    @user330587 Why does it have infinitely many solution? Isn't 0 the only solution? – zxcvber Jun 16 '16 at 10:36
  • 1
  • 1
    @GAVD Thanks, but I'm a high school student and I couldn't really understand the theorem and its usage... – zxcvber Jun 16 '16 at 10:39
  • @Surb: Maybe I'm wrong, but since $1-\frac{x^2}{2}+\frac{x^4}{4!}$ is an approximation of order 4 around 0 of $\sin(x)$, there is a neighborhood $]-\delta,\delta[$ such that $f$ and $\sin(x)$ coincident, and thus, there are infinitly many solution. No ? (my first polynomial was wrong, sorry) – user330587 Jun 16 '16 at 10:41
  • @zxcvber You should precise in your question that you're looking forward to a "high-school" level solution. – Surb Jun 16 '16 at 10:42
  • @user330587 Might be that I'm wrong, but I doubt it. We want more than an $\epsilon$ approximation. Moreover with this reasoning you can then simply consider $f(x)=x$... – Surb Jun 16 '16 at 10:44
  • @Surb I wasn't, but that theorem's usage was hard for me to understand. I'm sure I can understand it if given with explanations. – zxcvber Jun 16 '16 at 10:44
  • @Surb: But in these way, there are only one solution. But the principle of taylor, isn't to approximate the curve ? If we take a huge taylor polynomial, f and sin(x) will coincide on an interval, and thus f=sin will have infinitely many solution... no ? – user330587 Jun 16 '16 at 10:45
  • 1
    Considering that Talyor polynomial always has an error term, I think the only solution would be 0 since the error vanishes only when x=0... – zxcvber Jun 16 '16 at 10:45
  • good question. +1 But I don't think there is a solution at the level of high school (age 15-16 years max). I have provided an answer which uses Taylor's theorem and avoids complex analysis. – Paramanand Singh Jun 16 '16 at 19:11

4 Answers4

2

Not a High-School Level Solution: First, we show that $f(x)=\sin(x)$ must have a solution for arbitrarily large $x$. If not, then $f(x)=\sin(x)$ has infinitely many solutions on a compact subset of $\mathbb{R}$ (or $\mathbb{C}$). By the Identity Theorem for Analytic Functions, as polynomials are analytic, we conclude that $f=\sin$, but this contradicts the assumption that $f$ is a polynomial (because $f$ would have infinitely many roots, making it the zero polynomial). Hence, $f(x)=\sin(x)$ must have solutions with $|x|\to\infty$. Then, use @newnewnewnewnenwwojpwjfpoergje's now deleted hint (I don't know why he deleted it, as it was a good hint). If $f$ is nonconstant, then $\big|f(x)\big|\to\infty$ for $|x|\to\infty$, whence $f(x)=\sin(x)$ can't have arbitrarily large solutions $x$.

Batominovski
  • 49,629
2

This one is tricky to handle and does require the use of Taylor's Theorem, but strictly we don't need anything from complex analysis.


Let $g(x) = f(x) - \sin x$ where $f(x)$ is a polynomial. Clearly if we differentiate $g(x)$ multiple times, the part related to $f(x)$ will vanish ($(n + 1)^{\text{th}}$ derivative of a polynomial of degree $n$ is $0$) and the part related to $\sin x$ will take one of the four forms $\pm\cos x , \pm\sin x$ alternately. It thus follows that for any specific $x$ there will be a positive integer $n$ for which $g^{(n)}(x) \neq 0$.

Now consider a root $\alpha$ of $g$ so that $g(\alpha) = 0$. From the preceding argument it is obvious that there is a least positive integer $n$ such that $g^{(n)}(\alpha) \neq 0$ (so that $g(\alpha) = g'(\alpha) = \cdots = g^{(n - 1)}(\alpha) = 0$. By Taylor's Theorem we have $$g(\alpha + h) = \frac{h^{n}}{n!}g^{(n)}(\alpha) + o(h^{n})$$ and hence $g(\alpha + h) \neq 0$ for all sufficiently small values of $h$. Thus there is a neighborhood of $\alpha$ in which $g$ does not have any root other than $\alpha$.

Next we observe that if $f(x)$ is a polynomial of positive degree then as $x \to \infty$ or $x \to -\infty$ the polynomial $f(x)$ will also tend to $\pm\infty$ and $\sin x$ will remain bounded. Thus if there are any roots of $g(x) = f(x) - \sin x$ they necessarily lie in some bounded interval of type $[a, b]$. If there are infinitely many roots of $g$ lying in $[a, b]$ then by Bolzano Weierstrass Theorem the set $A$ of roots of $g$ has an accumulation point $\beta \in [a, b]$. So there is a sequence of roots $\alpha_{n}$ of $g$ such that $\alpha_{n} \to \beta$ as $n \to \infty$. Then by continuity of $g$ we have $$g(\beta) = g(\lim_{n \to \infty}\alpha_{n}) = \lim_{n \to \infty}g(\alpha_{n}) = 0$$ so that $\beta$ is also a root of $g$. But then we reach a contradiction because we have proved earlier that each root of $g$ has a neighborhood in which there are no other roots of $g$ and $\beta$ being an accumulation point of roots of $g$ is such that every neighborhood of $\beta$ contains a root of $g$ other than $\beta$.

Hence the polynomial $f(x)$ must be of zero degree i.e $f$ is a constant. Now clearly $\sin x = f(x)$ implies that $f$ must be a constant between $-1$ and $1$ (both inclusive).

1

We prove the contrapositive. Suppose $f$ is not the desired constant polynomial. There are two possibilities. Firstly $f$ is a constant taking on a value outside of $[-1,1]$ in which case $f(x) = \sin x$ has $0$ solutions, i.e. finitely many solutions. The second possibility is that $f$ is polynomial of finite degree at least $1$. Then we can find $c$ such that $f'(x) \neq 0$ for all $|x|>c$ (otherwise $f$ does not have finite degree), in which case $f$ is either monotonically increasing or decreasing when $|x|>c$. Hence eventually we must have $|f(x)|>\sin(x)$. Moreover there can only be finitely many intersections on any finite interval, so we are done.

fosho
  • 6,334
  • How can I show that f has finitely many intersections on $\abs{x} \leq c$? Couldn't there be a interval of the polynomial that is exactly same as sin? – zxcvber Jun 16 '16 at 10:56
  • 1
    There is no polynomial of finite degree that is exactly $\sin(x)$ on an interval $[a,b]$ since $\sin(x)$ is analytic so if $f$ and $\sin(x)$ agree on some curve then they agree everywhere in the region of analyticity. – fosho Jun 16 '16 at 11:00
  • That's not quite an argument, though. There is a polynomial coinciding infinitely many times with $x\sin(1/x)$ on a compact set. You are using the fact that $\sin$ is analytic. – Batominovski Jun 16 '16 at 11:03
  • You are right, I updated the comment. – fosho Jun 16 '16 at 11:07
  • @Dman Why is that "If f and ~ everywhere in the region of analyticity"? Or Is there a theorem for this? – zxcvber Jun 16 '16 at 11:11
  • "Moreover there can only be finitely many intersections on $|x|≤c$, so we are done" Why ? – Dark Jun 16 '16 at 11:12
  • Why are we done or why are there only finitely many intersections? – fosho Jun 16 '16 at 11:13
  • Why are there only finitely many intersections? – Dark Jun 16 '16 at 11:13
  • See the previous comments – fosho Jun 16 '16 at 11:14
  • @zxcvber you could call this a theorem. – fosho Jun 16 '16 at 11:15
  • Can you tell me how to prove that? Or let me know the name of other theorems used to prove it? – zxcvber Jun 16 '16 at 11:16
  • I don't understand. If say $f(\frac{1}{n})=\sin(\frac{1}{n})$ for all $n \geq 1$, you have an infinite number of intersections even if $f$ and $\sin$ do not coincide on any interval. – Dark Jun 16 '16 at 11:16
  • But then $f$ must have infinitely many points where $f'(x) = 0$, i.e. not finite degree. – fosho Jun 16 '16 at 11:17
  • Why ? Sorry but I don't see a clear implication. – Dark Jun 16 '16 at 11:20
  • Since $\sin(x)$ is monotonic in $[0,\pm \epsilon]$ for $\epsilon <\pi/2$ – fosho Jun 16 '16 at 11:22
-2

Assume that the polynomial is not constant.
we know that
$\displaystyle{\lim_{x \rightarrow \pm \infty}}$f(x)=$\pm$ $\infty$ (f(x) is not constant,also it has a leading term)
as f(x) is a polynomial
for the equation f(x)=sinx to have infinitely many solutions f(x) must
lie below $\pm$ 1 when x $\rightarrow$$\pm$$\infty$. But this is a contradiction.
Therefore f(x) is a constant function lying between $\pm$1.

GAUTAM
  • 69