7

If $x^3 - \dfrac{1}{x^3} = 108+76\sqrt{2}$, find the value of $x-\dfrac{1}{x}$.

Here's what I've tried so far.

$$\begin{align} \left(x-\dfrac{1}{x}\right)^3&=x^3-\dfrac{1}{x^3}-3\left(x-\dfrac{1}{x}\right) \\ \rightarrow \quad \left(x-\dfrac{1}{x}\right)^3&=108+76\sqrt{2}-3\left(x-\dfrac{1}{x}\right) \\u:=x-\dfrac{1}{x} \quad\rightarrow \quad u^3+3u-108-76\sqrt{2}&=0 \end{align}$$

Got stuck here since I didn't know how to solve this cubic equation.

I also tried factorizing $x^3-\dfrac{1}{x^3}$.

$$\begin{align}x^3-\dfrac{1}{x^3}&=\left(x-\dfrac{1}{x}\right)\left(x^2+\dfrac{1}{x^2}+1\right) \\ &= \left(x-\dfrac{1}{x}\right)\left(x^2+\dfrac{1}{x^2}+2-1^2\right) \\ &= \left(x-\dfrac{1}{x}\right)\left(\left(x+\dfrac{1}{x}\right)^2-1^2\right) \\ &= \left(x-\dfrac{1}{x}\right)\left(x+\dfrac{1}{x}+1\right)\left(x+\dfrac{1}{x}-1\right) \end{align}$$

Again, I didn't know what I could do with this.

Blue
  • 83,939
avighnac
  • 415
  • 4
    Hint. $x^3-y^3=(x-y)(x^2+xy+y^2)=(x-y)^3-3xy(x-y)$. – RDK Oct 18 '22 at 15:42
  • 1
    You should be able to figure this out with the hints. However, MSE expects that you show your own efforts. This is not a portal to dump questions. – Andreas Oct 18 '22 at 15:44
  • @Andreas ok, adding what I've tried so far. – avighnac Oct 18 '22 at 15:46
  • @RDK, what after that? I've already tried doing this to no avail, although this could be something I missed. – avighnac Oct 18 '22 at 15:54
  • 2
    Alright, you may think this is not worth doing because of $xy$. Then, try to "eliminate" $xy$. What's $y$ which makes $xy=1$? (With a bit of calculus, you can easily know the critical point, but with precalculus, just try solving this.) – RDK Oct 18 '22 at 16:05
  • 1
    Try making the variable substitution $u = x^3$, solve for $u$, and take the cube root. – Dan Oct 18 '22 at 16:18
  • @Dan which method are you referring to, the first or second one? – avighnac Oct 18 '22 at 17:00
  • @RDK $y = \frac{1}{x}$. but I'm still getting a cubic, namely $u^3+3u-108-76\sqrt{2}=0$ – avighnac Oct 18 '22 at 17:01
  • @avighnac Nice. Now, let's think about a remainder theorem in $\Bbb Q$. This will help you solve this. (Or, just use Cubic Formula, which exactly tells you the answer.) – RDK Oct 18 '22 at 22:12
  • 1
    Cubic Formula with $ax^3+cx+d=0$: $x=\sqrt[3]{\left( - \frac d {2a} \right)+\sqrt{\frac{d^2}{4a^2}+\frac{c^3}{27a^3}}} + \sqrt[3]{\left( -\frac {d}{2a} \right)-\sqrt{\frac{d^2}{4a^2}+\frac{c^3}{27a^3}}}$ – RDK Oct 18 '22 at 22:20
  • @RDK you mean $u$ is probably a multiple of $\sqrt{2}$? It’d really help if add a part to your answer solving this using the remainder theorem. Also, using the cubic formula poses the problem of the ‘nested cube root’. – avighnac Oct 19 '22 at 07:39
  • @RDK, actually, if $c=\pm 3$, there’s an even simpler cubic formula. $x^3+3x=c : x= \sqrt[3]{\frac{c\pm \sqrt{c^2+4}}{2}} + \sqrt[3]{\frac{2}{c\pm \sqrt{c^2+4}}}$ – avighnac Oct 19 '22 at 07:43

5 Answers5

8

Solve for $x$ first. You can use the quadratic formula.

$$x^3 - \frac1{x^3} = 108 + 76\sqrt2 \implies x^6 - (108+76\sqrt2)x^3 - 1 = 0$$

$$\implies x^3 = \frac{108+76\sqrt2 \pm \sqrt{(108+76\sqrt2)^2 + 4}}2 = 54 + 38\sqrt2 + 3\sqrt{645 + 456 \sqrt2}$$

$$\implies x = \sqrt[3]{54 + 38\sqrt2 + 3\sqrt{645 + 456 \sqrt2}} = \frac{3+2\sqrt2}2 + \frac12 \sqrt{21 + 12\sqrt2}$$

The hardest part is de-nesting the cube root (see below). Otherwise take WA at its word.

It follows that

$$\frac1x = \frac2{3 + 2\sqrt2 + \sqrt{21 + 12\sqrt2}} = -\frac{3+2\sqrt2}2 + \frac12 \sqrt{21 + 12\sqrt2}$$

and you can easily find $x-\frac1x$ from here.


De-nesting the cube root

Observe that

$$645 + 456 \sqrt2 = (21 + 12\sqrt2) (17 + 12\sqrt2) = (21 + 12\sqrt2) (3 + 2\sqrt2)^2 \\ \implies x^3 = 54 + 38\sqrt2 + 9\sqrt{21+12\sqrt2} + 6\sqrt2\sqrt{21+12\sqrt2}$$

Suppose we can decompose the cube root into the form

$$x = a + b \sqrt2 + c \sqrt{21+12\sqrt2}$$

where $a,b,c\in\Bbb Q$.

Taking cubes on both sides and matching up coefficients, we get the system of equations

$$\begin{cases} a^3 + 6 a b^2 + 63 a c^2 + 72 b c^2 = 54 \\ 2b^3 + 3 a^2 b + 36 a c^2 + 63 b c^2 = 38 \\ 7c^3 + a^2 c + 2 b^2 c = 3 \\ 2c^3 + a b c = 1 \end{cases}$$

Solving the last equation for $c^2 = 3ab-a^2-2b^2$ and substituting that into the first two equations yields yet another system,

$$\begin{cases} -62a^3 + 117a^2b + 96ab^2 - 144b^3 = 54 \\ -36a^3 + 48a^2b + 117 ab^2 - 124b^3 = 38 \end{cases}$$

Since both polynomials are homogeneous with degree $3$, suppose that $a,b$ are proportional. Let $b=ka$ where $k\in\Bbb Q$, so we have

$$\begin{cases} a^3(-62 + 117k + 96k^2 - 144k^3) = 54 \\ a^3(-36 + 48k + 117k^2 - 124k^3) = 38 \end{cases}$$

Eliminate $a^3$ by division to get a cubic equation in $k$.

$$\frac{a^3(-62 + 117k + 96k^2 - 144k^3)}{a^3(-36 + 48k + 117k^2 - 124k^3)} = \frac{27}{19} \implies 612k^3 - 1335k^2 + 927k - 206 = 0$$

Now apply the rational root theorem. If you start with the smallest divisors of $612$ and $206$, you'll soon find $k=\frac23$ (and we can show this is the only rational solution for $k$). It follows that

$$b=\frac23a \implies 16a^3 = 54 \implies a = \frac32 \implies b = 1 \implies c = \frac12$$

user170231
  • 25,320
  • 1
    Wow. You did very difficult computations. – Bob Dobbs Oct 18 '22 at 17:26
  • 1
    Thank you, but I wouldn’t be able to decompose that cube root without WA! – avighnac Oct 19 '22 at 07:36
  • In the first part of ‘decomposing the cube root’, how did you manage to make that observation? Or was it just pure luck/intuition – avighnac Oct 20 '22 at 07:21
  • 1
    The second half of this answer is more an attempt at explaining how WA is getting its result. One could instead assume a decomposition of the form $a+b\sqrt2+c\sqrt{645+456\sqrt2}+d\sqrt2\sqrt{645+456\sqrt2}$ but then the system of equations that follows is a bit more complicated (again with a single solution over the rationals, namely $(a,b,c,d)=\left(\frac32,1,\frac32,-1\right)$). – user170231 Oct 20 '22 at 16:03
  • @user170231 why would you have the ‘d’ term when the original cube root doesn’t have it? – avighnac Oct 23 '22 at 06:48
  • 1
    My perhaps unsatisfying answer is that the algebra just works out that way. Consider a simpler example like$$(a + b\sqrt s + c\sqrt t + d\sqrt s\sqrt t)^2 = \ a^2 + b^2s + c^2t + d^2st + (2ab+2cdt)\sqrt s + (2ac+2bds)\sqrt t + (2ad + 2bc) \sqrt s\sqrt t$$Now make any choice of $a,b,c,d\in\Bbb Q$ such that $ad+bc=0$, and pick any two $s,t\in\Bbb N$ that are not perfect squares. For example, choosing $(a,b,c,d)=(1,1,-1,1)$ and $(s,t)=(3,5)$ gives$$1+\sqrt3-\sqrt5+\sqrt{15}=\sqrt{24-8\sqrt3+4\sqrt5}$$despite the nested form not explicitly involving $\sqrt{15}$. – user170231 Oct 23 '22 at 09:12
  • And in fact in an earlier version of this answer, I had assumed$$x = a + b \sqrt2 + c \sqrt{21+12\sqrt2} + d\sqrt2\sqrt{21+12\sqrt2}$$but since it turned out $d=0$, I was previously under the same wrong impression that the $\sqrt s\sqrt t$ term was not needed in the denested form. – user170231 Oct 23 '22 at 09:20
6

I thought it might be helpful with answering now.

\begin{align} \newcommand{w}{\omega} & x^3-\frac{1}{x^3}=108+76\sqrt{2}. \\ & 108+76\sqrt{2} = \left(x-\frac 1 x\right)\left(x^2+1+\frac 1 {x^2}\right) \\ & = \left(x-\frac 1 x \right)^3+3x\cdot \frac 1 x \left(x-\frac 1 x\right). \\ \ \\ & u^3+3u=108+76\sqrt{2}. \\ & 108+76\sqrt{2}= 4(27+19\sqrt{2})=4(5+3\sqrt{2})(1+\sqrt{2})^2. \\ & (5+3\sqrt{2})^2+3=46+30\sqrt{2} \neq 12+8\sqrt{2}. \\ & \vdots \\ & (3+2\sqrt{2})^2+3=20+12\sqrt{2}=4(5+3\sqrt{2}). \\ \therefore \; & u=3+2\sqrt{2}. \end{align}

Substituting $x-\dfrac 1 x=3+2\sqrt{2}$, it works.

RDK
  • 3,389
4

Note that

$$(x - \frac{1}{x})^3 = x^3 - 3x + \frac{3}{x} - \frac{1}{x^3}$$ $$(x - \frac{1}{x})^3 = x^3 - \frac{1}{x^3} - 3(x - \frac{1}{x})$$ $$x^3 - \frac{1}{x^3} = (x - \frac{1}{x})^3 + 3(x - \frac{1}{x})$$

So if you let $t = x - \frac{1}{x}$ (the value that you're ultimately planning to solve for), then your equation becomes:

$$t^3 + 3t = 108 + 76\sqrt{2}$$

This is a “depressed” cubic equation that you could solve with Cardano's Formula, but that gives you some ugly nested radicals. So instead, I'm going another route, by finding an equivalent polynomial equation with integer coefficients.

$$(t^3 + 3t - 108)^2 = (76\sqrt{2})^2$$ $$t^6 + 3t^4 - 108t^3 + 3t^4 + 9t^2 - 324t - 108t^3 - 324t + 11664 = 11552$$ $$t^6 + 6t^4 - 216t^3 + 9t^2 - 648t + 112 = 0$$

There's no general formula for a sixth-degree polynomial. And the Rational Root Theorem fails to provide us with any linear factors. But it turns out that a quadratic factor exists.

$$(t^2 - 6 t + 1) (t^4 + 6 t^3 + 41 t^2 + 24 t + 112) = 0$$

Solving the quadratic gives $t = 3 \pm 2\sqrt{2}$. Solving the remaining quartic gives 4 complex solutions:

$$t = -\frac{3}{2} + \sqrt{2} \pm \frac{3}{2} i \sqrt{7 - 4 \sqrt{2}}$$ $$t = -\frac{3}{2} - \sqrt{2} \pm \frac{3}{2} i \sqrt{7 + 4 \sqrt{2}}$$

But of these 6 roots, it turns out that only one of them is a valid solution to the original equation.

$$t = \boxed{3 + 2\sqrt{2}}$$

Dan
  • 18,262
3

Here is a suggestion to get an answer systematically - though not fully analytic.

Others have already noted that we need to solve for the desired $u = x- \frac{1}{x}$ in $$f(u) = u^3+3u- 108- 76\sqrt{2} = 0$$ Now observe that any expression of the form $y = a + b \sqrt{2}$, when taken to an integer power, reproduces this form, i.e. $y^n = a_n + b_n \sqrt{2}$. Since the constant part in $f(u)$ is also of this form, we can write $u = a + b \sqrt{2}$, which gives $$f(u) = a^3 + 6 a b^2 + 3 a -108 + \sqrt{2} (3 a^2 b +2 b^3 + 3 b - 76) = 0 $$ This enables us to solve for the rational and the irrational parts separately, i.e. $$a^3 + 6 a b^2 + 3 a -108 = 0\\ 3 a^2 b +2 b^3 + 3 b - 76 = 0 $$ where $a$ and $b$ are rational numbers. Note that it is not clear beforehand whether $a$ and $b$ are integers, as other answers have tested some integer combinations.

Now this is again two coupled third-degree equations so one might ask what has been won (maybe one of you has a nice idea how to solve this analytically).

I introduce now a way to manipulate these equations to come up with an iterative procedure with works without any knowledge or presupposition of the nature of $a$ and $b$. Also, it does not need any roots (of quadratic or cubic equations) which have been used earlier but which can become cumbersome.

Writing equivalently $$a^2 + 6 b^2 + 3 -\frac{108}{a} = 0\\ 3 a^2 +2 b^2 + 3 - \frac{76}{b} = 0 $$ allows to isolate the terms with $a^2$ and $b^2$ by weighted adding, which gives

$$4 a^2 + 54/a + 3 = 114/b \\ 8 b^2 + 38/b + 3 = 162/a$$ Now let $A = 1/a$ and $B = 1/b$ to end with $$B = \frac{9 A}{19} + \frac{1}{38} + \frac{2}{57 A^2} \\ A = \frac{19 B}{81} + \frac{1}{54} + \frac{4}{81 B^2}$$ One can make this an iterative pair of equations; start with some $A$, obtain directly $B$ from the first equation, plug into the second and obtain a new $A$, repeat ...

Both equations are of the same functional form. Since there must be a solution, the first one should fall faster with $A$ then the second one with $B$, so this iteration should converge. Here are some values which are obtained when starting with $A=1$:

$$A=1 , B = 0.535, A = 0.3165, B = 0.526, A = 0.320, B = 0.520, A = 0.323, \cdots $$

Convergence is not very fast, but letting this go for a while leads to values which are approximating $$A=\frac13 , B = \frac12$$ and indeed one can check that this solves the equations.

So the result is $u = 3 + 2 \sqrt{2}$, q.e.d. $\qquad \Box$

Andreas
  • 16,627
  • Hey, at the beginning of your answer, you mention that you need to solve for $u=x+\frac{1}{x}$, but that cubic equation only works if u is the conjugate of what you’ve assumed it to be. – avighnac Oct 20 '22 at 07:36
  • 1
    @avighnac Sorry, writing error -> corrected. Thank you for mentioning! – Andreas Oct 20 '22 at 10:13
2

$u^3+3u=108+76\sqrt{2}$

A few posts here assumed $u = a + b\sqrt{2}$, then search for rational $(a, b)$

Let "slope", $\displaystyle \;m_1 = b/a$

$\displaystyle u^2 = (a^2+2b^2) \;+\; (2ab)\,\sqrt{2}\qquad\; →m_2 = \frac{2\,m_1}{1+2\,m_1^2}$

Note that $m_2$ has same sign as $\displaystyle m_1 \quad →m_∞ = \frac{sgn(m1)}{\sqrt{2}}$

Slope of RHS = $\displaystyle \frac{76}{108} ≈ 0.7037 \;<\; \frac{1}{\sqrt{2}} ≈ 0.7071$

$u^3 ≈ RHS,\;m_1$ likely convergent of RHS slope.

C:\>spigot -C 76/108
0/1
1/1
2/3
5/7
7/10
19/27

Convergent 2/3 work! Upon checking, $(b,a) = (2,3)$, not just the ratio.

$u^3 + 3u = (99+70\sqrt{2}) + 3×(3+2\sqrt{2}) = 108+76\sqrt{2}$

albert chan
  • 2,204