7

In this age of electronic devices, sometimes I like to challenge myself to find clever ways to find solutions to math problems that seem to require a calculator, without a calculator. (Here is a favorite example.)

I recently came up with this:

Without a calculator, show that $$\int_0^1 \left(\sin{x}-\frac{4}{9}\right)^3dx<0$$

It's a pretty close shave: my computer says the LHS $\approx −0.0000050...$.

The exact value of the LHS is:

$$\frac{1}{729}\left(972(2+\cos{1})\sin^4{0.5}-243(2-\sin{2})+432(1-\cos{1})-64\right)$$

I tried to use Maclaurin series, but that seems to be futile.

I also tried to find some kind of useful symmetry of the graph of $y=\sin{x}-\frac{4}{9}$, but to no avail.

Any clever way to do this? Just curious.

Dan
  • 22,158
  • Perhaps another method would be to use approximation methods of an integral to show that the value is strictly negative. (considering that the value for sin(1) and sin(1/2) is known). I tried before to calculate using an approximation method, so I could see that the value is strictly negative. – Anamaria Oct 31 '22 at 09:57
  • By substitution,$$\int_0^1 \left(\sin(x) - \frac49\right)^3 , dx = \int_{-\frac49}^{\sin(1)-\frac49} \frac{y^3}{\sqrt{1 - \left(y + \frac49\right)^2}} , dy$$and you can apply the binomial series to the denominator. – user170231 Oct 31 '22 at 15:28
  • @user170231. This is spectacular (look at my second answer) – Claude Leibovici Nov 01 '22 at 06:21

3 Answers3

2

For a nicer looking expression $$I=\int_0^1 \left(\sin(x)-\frac{4}{9}\right)^3\,dx=\frac{1472+972 \sin (2)-3915 \cos (1)+243 \cos (3)}{2916}$$

Now, using the series expansion of the sine and the binomial theorem $$\left(\sin(x)-\frac{4}{9}\right)^3=-\frac{64}{729}+\frac{1}{216}\sum_{n=1}^\infty \frac{144\times 2^{n} \cos \left(\frac{\pi n}{2}\right)-2 \left(27\times 3^{n}-145\right) \sin \left(\frac{\pi n}{2}\right)}{n!} x^n$$ Now, integrating termwise, the partial sums $$S_p=-\frac{64}{729}+\frac{1}{108}\sum_{n=1}^p \frac{72\times 2^{n} \cos \left(\frac{\pi n}{2}\right)- \left(27\times 3^{n}-145\right) \sin \left(\frac{\pi n}{2}\right)}{(n+1)!}$$ Up to $S_3$, the values are easy to compute by hand. For $p>3$, I cheated and used my fifty plus years old non-programmable pocket calculator (four operations and a single memory) to generate the sequence $$\left\{\frac{152}{729},-\frac{172}{729},-\frac{31}{2916},\frac{1141 }{14580},-\frac{31}{7290},-\frac{649}{51030},\frac{1321}{1632960 },\frac{2089}{1632960},-\frac{1871}{24494400}\right\}$$ For $p \geq 9$, all $S_p$ are negative (checked with a calculator).

2

Using @user170231's suggestion in comments; letting $a=\frac 49$ and $s=\sin(1)$ $$I=\int_{-a}^{s-a} \frac{y^3}{\sqrt{1 - \left(y + a\right)^2}} \, dy=\frac{s}{2 \sqrt{\pi }}\sum_{n=0}^\infty u_n$$ $$u_n= \frac{\Gamma \left(n+\frac{1}{2}\right)}{\Gamma (n+1)} \left(-\frac{2 a^3}{2 n+1}+\frac{3 a^2 s}{n+1}-\frac{6 a s^2}{2 n+3}+\frac{s^3}{n+2}\right)s^{2 n}$$ $u_0$ is the only negative term. All other terms are positive, almost in geometric progression

$$\frac {u_{n+1}}{u_n}=s^2\,\left( 1-\frac{3}{2 n}+O\left(\frac{1}{n^2}\right)\right)$$ and their infinite sum does not exceed $u_0$.

Edit

Pushed by @Dan's comment to do it, assume the above ratio and write $$S_p==\frac{s}{2 \sqrt{\pi }}\Bigg[ \sum_{n=0}^p u_n+u_{p+1}\, \, _2F_1\left(1,p-\frac{1}{2};p+1;s^2\right)\Bigg]$$ we have (I used a calculator) $$\left( \begin{array}{cc} p & 10^6\,S_p \\ 0 & -2595.89 \\ 1 & -730.922 \\ 2 & -270.549 \\ 3 & -115.935 \\ 4 & -55.3259 \\ 5 & -29.1880 \\ 6 & -17.1416 \\ 7 & -11.3071 \\ 8 & -8.36943 \\ 9 & -6.84325 \\ 10 & -6.02958 \\ 20 & -5.01422 \\ 30 & -5.00872 \\ 40 & -5.00867 \\ \end{array} \right)$$

1

Using Maclaurin series, we find that for all positive real $x$, the Maclaurin series of $\sin x$ alternates between being larger and smaller than $\sin x$. This is true because: 1) the Maclaurin series is always tangent to $\sin x$ as it shares the same first and second derivative (and so the error behaves almost like $\pm x^3$ with even smaller higher-order terms) (2) look at the behaviour of the function to infinity, as given by the sign of the highest-order term. For example, with $n = 3$ the limit to infinity of the Taylor series is $-\infty$, and so $\sin x > x - x^3/3!$. Choosing $x = \pi/4$ for example, we can see this is true given the first 3 decimal places, if we know $1/\sqrt{2} \approx 0.707$ and overestimate $\pi$ as $22/7$.

Given $f(x) = (\sin x - 4/9)^3$ and $T_n (x)$ being the Taylor series of $\sin x - 4/9$ to $n$ terms, $|f(x)^3 - (T_n (x))^3|$ $< f(x) - (T_n (x))$ when both $f(x), T_n (x)$ are small and if $T_n(x) > f(x)$. The right hand side of this inequality is bounded upwards by the Lagrange error bound of $\sin x$. This is given by $\frac{M}{(n + 1)!} |x - c|^{n+1}$. With $M = 1$, $x = 1$ and $c = 0$ (to find the maximum error for all $x$ between $0$ and $1$), we get that $\frac{1}{(n + 1)!} < 5 \cdot 10^{-6}$.

Now to bound this integral upwards, the negative area must be smaller and the positive area must be larger. In other words, we need to find an upper bound of $f(x)$, which gives candidates $x = 1, 5, 9 \cdots$ (and $f(x) < T_n(x)$ also holds for these values of $n$). $n = 5$ does not satisfy the inequality and so the next candidate is $n = 9$.

The value of this integral is $-5.0079 \cdot 10^{-6}$ which is a remarkably good approximation.

So we need to show that:

$$\int_0^1 \left(x-\frac{x^{3}}{3!}+\frac{x^{5}}{5!}-\frac{x^{7}}{7!}+\frac{x^{9}}{9!}-\frac{4}{9}\right)^{3} \ dx < 0$$

which would be extremely tedious to do completely by hand, but you only need a pocket calculator to show this.

Toby Mak
  • 16,827
  • How did you get the value $5 \cdot 10^{-6}$ ? – Dan Oct 31 '22 at 13:51
  • 1
    I got the same by Wolphram Alpha, free version. – Ryszard Szwarc Oct 31 '22 at 18:24
  • @Dan This is going to be circular, but use a lot of terms for the Maclaurin series to approximate sin x, use a pocket calculator to find the value of the approximate integral, then we find the minimum number of terms such that we know it will be negative even with the maximum error. Otherwise we have to try $n = 5$ first before we settle on $n = 9$, which you might prefer. – Toby Mak Nov 01 '22 at 03:27