I need to find analytically the roots of the following polynomial equation:


for an arbitrary integer $n$ and an arbitrary real parameter $a$. The only trick I can think of is in the special case when $a\geq0$ and $n=2m$ (i.e. $n$ even). In this case the equation can be decomposed into two simpler and lower dimensional equations:

$$x^{m}\left(2-x\right)-\sqrt{a}=0$$ $$x^{m}\left(2-x\right)+\sqrt{a}=0$$

but then I don't know how to proceed (I was thinking that maybe through some change of coordinates it is possible to transform these new equations into trinomial equations of the form $x^n-x+t=0$, whose solution is known, see here). I'd also like to find the roots for any $n$ and $a$. Closed-form solutions are not required: I'd really appreciate also a solution written e.g. as a series expansion.

Many thanks in advance.

P.S.: I don't know if it may help, but in my case $x\in[-1,1]$ so that you may write $x=\cos y$ for $y\in [0,\pi]$.

  • 1
    $\begingroup$ I very much doubt you'll find a closed form solution. Your link to the trinomial equation solution suggests that you'd be happy with a series or other approximation method, perhaps numerical. If so, say so; maybe someone will step up to help you. $\endgroup$ – Ethan Bolker Oct 24 '16 at 13:45
  • $\begingroup$ You are right, thanks for the suggestion. I have just edited my question. $\endgroup$ – user2983638 Oct 24 '16 at 13:49
  • 1
    $\begingroup$ I suggest you to normalize the equation to the form $x^\alpha(1-x)=a$ for easier study. $\endgroup$ – Yves Daoust Oct 24 '16 at 14:16
  • 1
    $\begingroup$ Maybe you can try to find something using a $log$ $\endgroup$ – H.C. Lefevre Oct 24 '16 at 14:26
  • 1
    $\begingroup$ Notice that for $n$ even and $a < 0$ the equation has no real solution. $\endgroup$ – Alex Silva Oct 24 '16 at 15:10

See the following link (theorem 2) https://www.researchgate.net/publication/262973394_Solution_of_Polynomial_Equations_with_Nested_Radicals

He solves :


Your equation is :

$$2x^m-x^{m+1}-\sqrt{a}=0$$ If you make the following substitution :


We have :


So divide by $-\beta^{m+1}$ you have:


And make the last substitutions :

$$\beta^{1}=\frac{1}{m+1}$$ and $$\sqrt{a}=\beta^{m+1}$$

Finally we obtain :

$$(-2)(m+1)y^m+y^{m+1}+1=0$$ So you can apply the theorem 2 with $A=-2$ and $q=m+1$ Another way is to use the theorem 1 with $b=0$ but we obtain a nested radical... Ps:

It's a partial solution because there is some conditions on $\sqrt{a}$... If you want other details see this Solving 5th degree or higher equations

  • $\begingroup$ Wow, I was aware of that article, but I thought I had to use nested radicals, which I don't like too much. But I didn't realize it is possible to use theorem 2. So this solves the case with $a\geq0$ and $n$ even. I guess I have to understand how to get the remaining $m-1$ solutions, since eq. (13) basically gives only one solution, if I understand it correctly. Let me check the calculations before accepting your answer. By the way, the case with $n$ odd should still be an open problem. $\endgroup$ – user2983638 Oct 24 '16 at 15:33
  • $\begingroup$ I'm not sure that the final equation you get corresponds to the first one. It seems to me that the term $1$ has the wrong sign compared to the term $y^{m+1}$. But ok, maybe for the solution with the $+\sqrt a$ it works fine! $\endgroup$ – user2983638 Oct 24 '16 at 16:17

The first thing you can do is to study the trivial case $a=0$ which gives : $$x^n(2-x)^2=0$$ The roots are obviously $x=0$ with order $n$ and $x=2$ with order $2$. You can try to solve other cases to sketch the general form of the solution if it exists.

You can also try some particular values of $n$ :

For $n=0$ you have :


Which is an order $2$ root.

For $n=1$ I tried with Wolfram Alpha which provides only one ugly solution, which has to be order $3$

For $n=2$ you have : $$x^2(2-x)^2=a$$ $$x(2-x)=2x-x^2=\pm\sqrt{a}$$ Hence the solutions are : $$x=1\pm\sqrt{1\pm\sqrt{a}}$$ Depending on $a$ you can have up to four different solutions with order $1$ each.

It seems complicated to find a general expression as you can have different number/orders of solutions depending on the choice of $(a,n)$

  • $\begingroup$ Hi, thanks for your comments. I know it's not straightforward to find a solution for any $n$, but I was wondering if through some series expansion like the one I suggested (see en.wikipedia.org/wiki/Bring_radical#Glasser.27s_derivation) it is possible to solve the equation, at least approximately. Also the trinomial form of the link has arbitrary $n$, but a solution does exist. $\endgroup$ – user2983638 Oct 24 '16 at 15:00
  • $\begingroup$ It is easy to prove that this equation has always at least one solution $x\in\mathbb{R}$ as whatever $n$, $x^n(x-2)^2$ takes at least all the values in $\mathbb{R}_{+}^{*}$. The thing is there are many cases in some of them you have one only solution in some other you have $n$ different solutions and in some other you have two solutions etc. The intuition is that you cannot look for a methodology other than computer based to approach or explicitely compute this solution. $\endgroup$ – H.C. Lefevre Oct 24 '16 at 15:07
  • $\begingroup$ I see your point. Also observe that for $a<0$ and $n$ even there is no solution. $\endgroup$ – user2983638 Oct 24 '16 at 15:12
  • $\begingroup$ That is true, I considered $a\geq 0$ as you stated in your first post. $\endgroup$ – H.C. Lefevre Oct 25 '16 at 8:06

Excel can provide some intuition. Here are graphs of $x^n(2-x)^n$ on $[-1,1]$ for $n=6$ and $n=7$.

enter image description here

You can see the range in each case, and prove it (for $n$ odd and $n$ even). For large(r) $n$ than these modest values the shape remains the same, so for values of $a$ (not too close to $0$) you'll have one or two solutions. Those solutions approach $\pm 1$ as $n$ grows. Newton's method should work just fine.

  • $\begingroup$ Interesting. Indeed a very special case of the equation I need to consider (for reasons I don't explain here) is $n\rightarrow\infty$. $\endgroup$ – user2983638 Oct 24 '16 at 15:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.