Return of the rescaling
Intro
One of the previous posts touched upon the importance of rescaling in simple perturbation problems. As the example discussed in that post involved a straightforward polynomial equation that contained explicitly a small parameter, a question remains: is rescaling relevant only in situations that involve such small parameters? The answer might come as a surprise to the “uninitiated”, but this is definitely not true. Employing rescaling strategies proves advantageous across a range of different scenarios that initially appear unrelated to perturbation theory. One such situation is considered next.
A simple approximation
We are interested in finding (without a calculator or Cardano’s formulae) reasonable approximations for the roots of the cubic
\[x^3- 21x^2 + 35x-7 = 0. \tag{1}\]The crucial step is to conveniently re-scale this equation in order to get a tight bracketing interval for one of its roots. By making the substitution $y=10x$, we are led to the new equation
\[P(y)\equiv y^3-210y^2+3500y-7000 = 0. \tag{2}\]It might seem that not much progress was made in passing from (1) to (2). However, this is not entirely true. Note that $P(2)<0$ and $P(3)>0$, so by the intermediate value theorem we know that there is a root $y_0$ of (2) such that $y_0\in(2,3)$. In terms of the original equation, there is a root $x_0=y_0/10\in (0.2,0.3)$.
Set $x = 0.2 + \varepsilon$, with $\varepsilon$ a small number (we already know that $0 < \varepsilon < 0.1$). Substituting this into (1) we find another cubic in $\varepsilon$:
\[(\varepsilon+0.2)^3 -21(\varepsilon+0.2)^2+35(\varepsilon+0.2) - 7 = 0, \tag{3}\]and since $\varepsilon^3\simeq 10^{-3}=0.001$ we can safely ignore the cubic term when expanding (3). This leaves us with the quadratic
\[-20.4{\varepsilon^2} + 26.72{\varepsilon}- 0.832 = 0.\]The root of interest is
\[\varepsilon_1\simeq \dfrac{13.36-12.71}{20.4}\simeq 0.032\quad\Longrightarrow\quad x_1\simeq 0.2 + \varepsilon_1 = 0.232.\]To find approximations for the remaining roots, we use Viete’s relations for the original cubic
\[x_1+x_2+x_3 = 21\quad\mbox{and}\quad x_1x_2x_3 = 0.\]We already have an idea about $x_1$ (see above), so
\[\begin{aligned}[t] &a\equiv x_2+x_3 = 21-x_1\simeq 20.768\\ &{}\\ &b\equiv x_2x_3 = 7/x_1\simeq 30.17. \end{aligned}\]The quadratic that has roots $x_2$ and $x_3$ is simply $x^2-ax+b=0$ or $x^2-20.768x+30.17=0$, whence $x_2\simeq 1.57$ and $x_3 \simeq 19.20$.
To summarise, we have found that the roots of the original cubic are
\[x_1\simeq 0.232,\quad x_2\simeq 1.57,\quad x_3\simeq 19.20.\]Using one of the many online calculators we discover the following values (rounded to 4 d.p. accuracy)
\[x_1\simeq 0.2319,\quad x_2\simeq 1.5724,\quad x_3\simeq 19.1957.\]Clearly, our approximations are pretty decent.
Further remarks
Equation (1) is related to trigonometry, in particular the value of $\tan(2\pi/7)$. To see the link with the above developments, we need some basic trigonometric identities. Let’s start with the well-known result
\[\tan(\alpha+\beta) = \dfrac{\tan\alpha + \tan\beta}{1-\tan\alpha\tan\beta}\,\]which is valid for any $\alpha$, $\beta\in\mathbb{R}$ such that all the expressions in this formula exist.
Setting $\alpha\to{\theta}$ and $\beta\to{\theta}$ gives
\[\tan(2\theta) = \dfrac{2\tan \theta}{1-\tan^2\theta}, \tag{4}\]while for $\alpha\to{2\theta}$ and $\beta\to{\theta}$ we find
\[\tan(3\theta) = \dfrac{\tan(2\theta)+\tan \theta}{1-\tan(2\theta)\tan \theta}=\dots =\dfrac{3\tan \theta - \tan^3 \theta}{1-3\tan^2 \theta}. \tag{5}\]Furthermore, with $\theta\to 2\theta$ in (4),
\[\tan(4\theta) = \dfrac{2\tan(2\theta)}{1-\tan^2(2\theta)} = \dots =\dfrac{4\tan\theta(1-\tan^2\theta)}{1-6\tan^2\theta + \tan^4\theta}. \tag{6}\]We are now set to go. If $\theta=2\pi/7$ then $7\theta=2\pi$; taking the tangent of both sides,
\[\begin{aligned}[1] &0 = \tan(2\pi) = \tan(7\theta)=\tan(4\theta + 3\theta)\\ &{}\\ &{}\qquad\Longrightarrow\quad 0 = \dfrac{\tan(4\theta)+\tan(3\theta)}{1-\tan(4\theta)\tan(3\theta)}\\ &{}\\ &{}\qquad\Longrightarrow\quad \tan(4\theta)+\tan(3\theta)= 0. \end{aligned}\]The last equation above can be transformed with the help of (5) and (6), so that everything is expressed in terms of $\tan\theta$. Carrying out the relevant algebraic manipulations, it transpires that
\[\tan^6\theta-21\tan^4\theta+35\tan^2\theta - 7 = 0. \tag{7}\]Finally, by making the substitution $x=\tan^2\theta$ in (7) results in the original cubic (1). We remark in passing that our calculations in the previous section have shown that
\[\tan\dfrac{2\pi}{7}\simeq\sqrt{1.57}\simeq 1.25.\]The value $\pi/7$ is the hero of many other trigonometric stories; for example, the following interesting identity
\[\cos\dfrac{2\pi}{7} + \cos\dfrac{4\pi}{7} + \cos\dfrac{6\pi}{7}= -\dfrac{1}{2}. \tag{8}\]This is easily proved with the help of the standard formula
\[\sin\alpha\cos\beta =\dfrac{1}{2}\left[ \sin(\alpha+\beta)-\sin(\beta-\alpha) \right]. \tag{9}\]Let $S$ be the left-hand side of (8), and then multiply $S$ by $1\equiv \sin(\pi/7)/\sin(\pi/7)$. The result can be re-arranged in the more convenient form
\[S\times \dfrac{\sin(\pi/7)}{\sin(\pi/7)} = \dfrac{1}{\sin(\pi/7)} \left[ \sin\dfrac{\pi}{7}\cos\dfrac{2\pi}{7} + \sin\dfrac{\pi}{7}\cos\dfrac{4\pi}{7}+\sin\dfrac{\pi}{7}\cos\dfrac{6\pi}{7} \right].\]To evaluate the terms in the square bracket in this last equation, we need to apply (9) for suitably chosen values for $\alpha$ and $\beta$:
\[\begin{aligned}[t] &\sin\dfrac{\pi}{7}\cos\dfrac{2\pi}{7} = \dfrac{1}{2}\left[ \sin\dfrac{3\pi}{7}-\sin\dfrac{\pi}{7} \right],\\ &{}\\ &\sin\dfrac{\pi}{7}\cos\dfrac{4\pi}{7} = \dfrac{1}{2}\left[ \sin\dfrac{5\pi}{7}-\sin\dfrac{3\pi}{7} \right],\\ &{}\\ & \sin\dfrac{\pi}{7}\cos\dfrac{6\pi}{7} = \dfrac{1}{2}\left[ \sin\dfrac{7\pi}{7}-\sin\dfrac{5\pi}{7} \right]. \end{aligned}\]Adding together the last three results and taking into account that $\sin(7\pi/7)=\sin(\pi)=0$, we eventually recover the identity (8).
Epilogue
Here are a few other interesting polynomial equations related to trigonometry:
- The cubic $x^3-21x^2+35x-7=0$ has the roots: $\tan^2\dfrac{\pi}{7}$, $\tan^2\dfrac{2\pi}{7}$, $\tan^2\dfrac{3\pi}{7}$.
- $x=\tan\dfrac{\pi}{20}$ is one of the roots of the quartic $x^4-4x^3-14x^2-4x+1=0$.
- The quartic $x^4-6x^2+1=0$ has the roots: $\tan\dfrac{\pi}{8}$, $\tan\dfrac{3\pi}{8}$, $\tan\dfrac{5\pi}{8}$, $\tan\dfrac{7\pi}{8}$.
- The cubic $x^3-5x^2+6x-1=0$ has the roots: $4\cos^2\dfrac{\pi}{7}$, $4\cos^2\dfrac{2\pi}{7}$, $4\cos^2\dfrac{3\pi}{7}$.
- A regular heptagon $A_1A_2A_3A_4A_5A_6A_7$ is inscribed in a circle of radius $R=1$. Then, the lengths of the chords $A_1A_2$, $A_1A_3$, $A_1A_4$ are the roots of the cubic $x^3-7x^2+14x-7=0$.
- $x=2\sin\dfrac{\pi}{18}$ is one of the roots of the cubic $x^3-3x-1=0$.