4

I'm trying to find the three distinct and real roots of $$-x^3+3x+\left(2-\frac{4}{n}\right)=0,$$

where $n>0$ (we could say $n\geq 2$ if that helps), but I'm not able to get very far:

Using the notation of the Wikipedia-page, I find that the discriminant is $$\Delta=27\cdot 4\left(\frac{1}{n}-\frac{1}{n^2}\right),$$ which gives

$$C=3\left( 1-\frac{2}{n} \pm 2 i \sqrt{\frac{1}{n}-\frac{1}{n^2}} \;\right)^{1/3},$$

which is then used to find $x$, as $$x=\frac{C}{3}+\frac{3}{C}.$$

Two things confuse me:

  1. $C$ looks like a non-real number, but $x$ should be real (since $\Delta >0$). How can one further reduce the expression for $x(C)$ to show that the imaginary part is zero? I'm having trouble evaluating that cube root.
  2. When I use my expression for $x(C)$ in Mathematica and evaluate numerically for some $n$, I only find one of the three solutions that Mathematica finds if I just ask it to give the roots of the original equation (the change of sign in $C$, i.e. the $\pm$, doesn't even give two different solutions). What have I done to exclude the two other solutions (or is it just Mathematica excluding them somehow)?

Context:

Actually, I'm only trying to find the root where $-1\leq x\leq 1$. I'm trying solve (the first) part of system of equations that I solved numerically in order to make this answer analytically, so that I can play with the limit of the expression (as $n\rightarrow \infty$).

Thank you.

3 Answers3

3

For the calculation of the roots of the depressed cubic $$ y^{\,3} + p\,y + q = 0 $$ where $p$ and $q$ are real or complex, I personally adopt a method indicated in this work by A. Cauli, by which putting $$ u = \sqrt[{3\,}]{{ - \frac{q} {2} + \sqrt {\frac{{q^{\,2} }} {4} + \frac{{p^{\,3} }} {{27}}} }}\quad v = - \frac{p} {{3\,u}}\quad \omega = e^{\,i\,\frac{{2\pi }} {3}} $$ where for the radicals you take one value, the real or the first complex one (but does not matter which) then you compute the three solutions as: $$ y_{\,1} = u + v\quad y_{\,2} = \omega \,u + \frac{1} {\omega }\,v\quad y_{\,3} = \frac{1} {\omega }\,u + \omega \,v $$ In your case: $$ y^{\,3} - 3\,y - 2\left( {\frac{{n - 2}} {n}} \right) = 0 $$ we obtain $$ \frac{{q^{\,2} }} {4} + \frac{{p^{\,3} }} {{27}} = \left( {\frac{{n - 2}} {n}} \right)^{\,2} - 1 = - 4\frac{{\left( {n - 1} \right)}} {{n^{\,2} }} < 0 $$ which confirms that there are three real solutions, and $$ \begin{gathered} u = \sqrt[{3\,}]{{\frac{{n - 2}} {n} + i\,\frac{2} {n}\sqrt {\left( {n - 1} \right)} }} = \frac{1} {{\sqrt[{3\,}]{n}}}\;\sqrt[{3\,}]{{n - 2 + i\,2\sqrt {\left( {n - 1} \right)} }} = \hfill \\ = \frac{1} {{\sqrt[{3\,}]{n}}}\;\sqrt[{3\,}]{{n\,e^{\,i\,\alpha } }} = e^{\,i\,\alpha /3} \quad \left| {\,\alpha = \arctan \left( {\frac{{2\sqrt {\left( {n - 1} \right)} }} {{n - 2}}} \right)} \right. \hfill \\ v = - \frac{p} {{3\,u}} = \frac{1} {u} = e^{\, - \,i\,\alpha /3} \hfill \\ \end{gathered} $$ with the understanding that for $n=1,\; 2$, $\alpha= \pi , \; \pi /2$, i.e. that we use the 4-quadrant $arctan$.
So that in conclusion, for $0<n$, we have: $$ \left\{ \begin{gathered} y_{\,1} = e^{\,i\,\alpha /3} + e^{\, - \,i\,\alpha /3} = 2\cos \left( {\frac{\alpha } {3}} \right) \hfill \\ y_{\,2} = e^{\,i\,\alpha /3 + 2\pi /3} + e^{\, - \,i\,\alpha /3 - 2\pi /3} = 2\cos \left( {\frac{{\alpha + 2\pi }} {3}} \right) \hfill \\ y_{\,3} = e^{\,i\,\alpha /3 - 2\pi /3} + e^{\, - \,i\,\alpha /3 + 2\pi /3} = 2\cos \left( {\frac{{\alpha - 2\pi }} {3}} \right) \hfill \\ \end{gathered} \right. $$ Concerning the range spanned by the solutions, apart for $n=1$ where we get the solutions (1,-2,1), then for $2 \le\; n$ we have $$ \frac{{\alpha (n)}} {3}\quad \left| {\;2 \leqslant n} \right.\quad = \frac{1} {3}\arctan _{\,4\,Q} \left( {n - 2,\;2\sqrt {\left( {n - 1} \right)} } \right) = \left\{ {\frac{\pi } {6},\frac{\pi } {{7.66}},\; \cdots } \right\} $$ which means: $$ \left\{ \begin{gathered} \quad \quad 2 \leqslant n \hfill \\ 0 < \frac{{\alpha (n)}} {3} \leqslant \frac{\pi } {6}\quad \Rightarrow \quad \sqrt 3 \leqslant y_{\,1} < 2 \hfill \\ 2\frac{\pi } {3} < \frac{{\alpha (n)}} {3} + 2\frac{\pi } {3} \leqslant \frac{5} {6}\pi \quad \quad \Rightarrow \quad - 2 < y_{\,2} \leqslant - \sqrt 3 \hfill \\ - 2\frac{\pi } {3} < \frac{{\alpha (n)}} {3} - 2\frac{\pi } {3} \leqslant - \frac{\pi } {2}\quad \quad \Rightarrow \quad - 1 < y_{\,3} \leqslant 0 \hfill \\ \end{gathered} \right. $$

G Cab
  • 35,272
  • Beautiful, thank you! It seems $y_3$ is always between $-1$ and $0$ (for $n\in\mathbb{N}$) - how can we show this? – Bobson Dugnutt Sep 16 '16 at 00:16
  • @Lovsovs, glad you appreciate, I learnt much myself in developing it. concerning your question, it is not difficult to demonstrate that it so, except for $n=1$, as added to my post. – G Cab Sep 16 '16 at 14:00
  • Great, thanks! But shouldn't it be $n \geq 2?$ – Bobson Dugnutt Sep 16 '16 at 14:02
  • @Lovsovs, it is in fact, only that I prefer the notation in which the least term is put on the left (it has many advantages ..) – G Cab Sep 16 '16 at 14:06
  • Wauw, I don't know why I read it like $n \leq 2$, but I did (I guess one disadvantage of your notation is that it confuses tired people :P). Thanks again. – Bobson Dugnutt Sep 16 '16 at 14:08
  • sorry, I understand it might be confusing for those accustomed to put the variable first, as I was in the past. Later, specially in combinatorics and when you have chains of inequalities, I got used to arrange them as "along the $x$ axis". – G Cab Sep 16 '16 at 14:18
  • No problem! And yeah, I can definitely see it making sense! – Bobson Dugnutt Sep 16 '16 at 14:31
2

This is an approach on how to solve the DEPRESSED cubic equation. To show the roots are real just find the discriminant of the cubic, (it can be shown easily)

To find the roots of your query follow these steps:

Multiply orginal equation by -1.

To get $$x^3 -3x+(\frac{4}{n}-2) =0$$

Subsitute $$x= ucosa$$

Substite in the original equation to get

$$(ucosa)^3 -3ucosa+(\frac{4}{n}-2) =0$$

Recall the identity $$cos(3a)=4cos^3a-3cosa$$

Manipulate the equation to get

$$u^3(\frac{cos3a +3cosa}{4}) -3ucosa+(\frac{4}{n}-2) =0$$

Distribute to obtain: $$u^3(\frac{cos3a}{4})+ u^3(\frac{3cosa}{4}) -3ucosa+(\frac{4}{n}-2) =0$$

Take the common factor out of the middle two terms to obtain:

$$u^3(\frac{cos3a}{4})+ 3ucosa(\frac{u^2}{4}-1)+(\frac{4}{n}-2) =0$$

"u" is a parameter chosen by us. We are in search of such a parameter "$u$"; so that the middle term cancels out so that we can solve the trigonometric equation. In order for this to occur: $$u=±2$$

Substituting $u=2$ we get

$$2cos3a+(\frac{4}{n}-2) =0$$

Solving for $cos3a$ we get

$$cos3a= 1- \frac{2}{n}$$

Therefore

$$3a= arccos(1- \frac{2}{n})$$

$$a = \frac{arccos(1- \frac{2}{n})}{3}$$

$$x= 2cos(\frac{arccos(1- \frac{2}{n})}{3})$$

Similarly we can subsitute $u=-2$ and find another root.

You will get $$x=-2cos(\frac{arccos(\frac{2}{n}-1)}{3})$$

Here is a link Wiki

In fact an general solution is for the cubic equation you have demanded is:

$$x=2cos(\frac{arccos(1- \frac{2}{n})-2πk}{3})$$

Where k = 0,1,2.

Hope this helped.

Sid
  • 1,234
  • interesting method (+1). Just correct the typo $2 \cos(3a)$ in $8 \cos(3a)$ after substituting $u=2$. – G Cab Mar 25 '17 at 17:53
  • @GCab. I am new to this website. I have asked my query, but it has been receiving a bad reputation from the community. Please see here: http://math.stackexchange.com/questions/2201633/logarithmic-inequality. The question was put on hold. When I tried to ask the question again, I was filed "duplicate." This stops ALL users from answering my query. http://math.stackexchange.com/questions/2202400/logarithmic-inequality. – Sid Mar 25 '17 at 19:16
  • well, I am also not so much acquainted with the "mechanisms" and the "sociology" of this site: It happened also to me to ask questions I thought to be interesting and not to get any answer at all. Concerning yours, I saw that Lovsov gave you good indications on how to edit it. The most important is that you should include what you know about logs and where you get stuck. Time allowing I'll try and help you. – G Cab Mar 26 '17 at 01:24
  • @GCab It is ok now, the problem has been re-opened. Now I have to wait for a response for the questions. Thanks for the tip. – Sid Mar 27 '17 at 09:13
1

$C/3$ has absolute value $1$, so $3/C=(C/3)^{-1}=\overline{C/3}$, therefore $x=(C/3)+\overline{C/3}=2\operatorname{Re}(C/3)$ is real.


Since $(C/3)^3=(1-\frac{2}{n})\pm i\cdot(2\sqrt{\frac{1}{n}-\frac{1}{n^2}})$, we use the definition of absolute values to obtain

$$|C/3|^6=|(C/3)^3|^2=(1-\frac{2}{n})^2+4\left(\frac{1}{n}-\frac{1}{n^2}\right)=1$$ and therefore $|C/3|=1$.

As to how to calculate $C$ and $x$ more directly: In the definition of square root we need there is a cubic root, so we need to write $C$ in polar form: Say $C/3=e^i\theta$. The sign of $\pm$ in the definition of $C$ will be the same as the sign of $\theta$.

Then $x=C/3+3/C=2\operatorname{C/3}=2\cos(\theta)$. Since $\cos$ is even, the sign of $\theta$ (which is the sign we choose when calculating $C$) doesn't matter.

Now to calculate $\theta$: Looking at the real and imaginary parts of $1-\frac{2}{n}+2i\sqrt{\frac{1}{n}-\frac{1}{n^2}}$, we obtain $$C/3=\left(e^{i\arctan\left(\frac{2\sqrt{\frac{1}{n}-\frac{1}{n^2}}}{1-\frac{2}{n}}\right)}\right)^{1/3}=\left(e^{i\arctan\left(\frac{2\sqrt{n-1}}{n-2}\right)}\right)^{1/3}=e^{i\frac{\arctan\left(\frac{2\sqrt{n-1}}{n-2}\right)}{3}}$$ so $\theta=\frac{\arctan\left(\frac{2\sqrt{n-1}}{n-2}\right)}{3}$ works, in which case $x=2\cos\frac{\arctan\left(\frac{2\sqrt{n-1}}{n-2}\right)}{3}$.


As for the other roots. Denote by $p(x)$ the initial polynomials. The derivative of the polynomials is $-3x^2+3$, so maximums and minimums are attained at $\pm 1$.

So $p(x)$ starts (for "very negative" numbers) large, and decreases until it gets to the first local minimum, on which $$p(-1)=-(-1)^3+3(-1)+\left(2-\frac{4}{n}\right)=-\frac{4}{n}$$ so it must have passed through a root at some point.

After $-1$, $p(x)$ starts increasing until it gets to the local maximum at $x=1$, on which $$p(1)=-1^3+3(1)+\left(2-\frac{4}{n}\right)=4-\frac{4}{n}\geq 0$$ so there will be another root in $(-1,1]$ (more precisely, $1$ will be a double root iff $n=1$, and if $n\neq 1$ there will be another root strictly in $(-1,1)$).

After that, $p(x)$ decreases all the way to $-\infty$, in which case it will necessarily pass through another root (if $n\neq 1$).

In the end, this shows that $p(x)$ has three real roots. One of them is found by the procedure above, and after we calculate it we can simply decompose $p(x)$ as a product of a linear and a quadratic, and solve the quadratic.

Luiz Cordeiro
  • 18,513
  • How would you evaluate $\operatorname{Re} (C/3)$ though, and how can it produce three different solutions? Also, how can you see it has length $1$? Could you perhaps expand your answer to include these questions? Thanks! – Bobson Dugnutt Sep 15 '16 at 19:11
  • Thank you very much! W.r.t. your last sentence, how would you actually do that? The one found by your (excellently explained) method is unfortunately not the one in $(-1,1)$ (for $n\not=1$), which is the one I need. How would you decompose the polynomial to find an expression for this root? (After this I promise not to bother you with further questions! :)) – Bobson Dugnutt Sep 15 '16 at 20:18
  • @Lovsovs I don't think there is a very nice formula for the other roots. Maybe you can put the initial root in simpler terms, but for that we'd need some formula for $\cos(x/3)$ in terms of $\cos(x)$ (which I couldn't find as well. The only formula I found is $\cos(3x)=4\cos^3x-3\cos x$, so this amounts so solving a polynomial of the form $4c^3-3c+a=0$, which might yield some horrible formulas. Anyway, if you manage to do that, it should be easy enough to find $b,c$ for which $x^3-3x-\left(2-\frac{4}{n}\right)=(x-x_1)(x^2+bx+c)$ (where $x_1$ is the root we already have). (cont...) – Luiz Cordeiro Sep 15 '16 at 23:54
  • After expanding the right-hand side, this will give you some linear equations on $b$ and $c$ which will determine then. Now you have a (probably grotesque) quadratic, so "simply" use the quadratic formula and find the other roots. I guess you understand why I preferred to stop there, and just argue that "it is possible to find the other roots". – Luiz Cordeiro Sep 15 '16 at 23:55