4

I am looking for the roots (or basically any information regarding them) of the sixth order polynomial $$p(x):=ax^6+(a+1)x^4+2bx^3-b^2$$ for positive, real constants $a,b$. Since $p(0)=-b^2<0$ and $\lim_{x\to\pm\infty}p(x)=+\infty$ we have at least two real roots $x_0,x_1$. Moreover, plotting the function for several parameters suggests, that the remaining four roots are non-real (i.e. two pairs of complex conjugate solutions).

Unfortunately this is as far as I got. I am especially interested in the complex solutions and their behavior for large parameters $a,b$. Any idea on how to tackle this problem is very welcome!

weee
  • 507
  • Probably the first thing is to use WA, Mathematica, Maple as soon as see this equation :) Because it's the first and the fastest way to find out if this equation is solvable with radicals.. – lone student Mar 05 '21 at 15:34
  • Ah yes, obviously I forgot to mention that I tried sympy and WA but they gave me nothing useful. Of course, for specific parameters $a,b$ the roots can be approximated. – weee Mar 05 '21 at 15:36
  • For specific parameters $a,b$ if WA doesn't give a exact solution, then most likely , exact solution doesn't exist. This means, the equation is not solvable. – lone student Mar 05 '21 at 15:40
  • Certainly by the fundamental theorem of algebra 6 (counted by their multiplicity) roots exist. What do you mean by "the equation is not solvable"? – weee Mar 05 '21 at 15:48
  • 2
    I think they mean that it is not solvable in closed-form. – Moo Mar 05 '21 at 15:52
  • @weee: Are you expecting a closed form solution? – Moo Mar 08 '21 at 15:47
  • Actually I expect that a closed form solution is very hard to find. But maybe some bounds can be found. – weee Mar 08 '21 at 20:24
  • @weee Is $\frac{b^2}{a}$ large? – River Li Mar 09 '21 at 03:19
  • @RiverLi possibly yes. – weee Mar 09 '21 at 10:57
  • @weee When $a, b$ is large, it seems the roots are approximately $A, - A, B + C\mathrm{i}, B - C\mathrm{i}, -B + C \mathrm{i}, -B - C\mathrm{i}$ for some real numbers $A, B, C$. If $b^2/a$ is large, $A$ is near $\sqrt[6]{b^2/a}$. – River Li Mar 09 '21 at 11:00

4 Answers4

3

$\color{brown}{\textbf{Parametrization.}}$

Searching the given polynomial's representation in the form of $$P(a,b,x) =(a+2p)(x^6+t^6)+p(x^2+t^2)^3-3px^2(x^2+t^2)^2 + 2bx^3,\tag1$$ easily to get $$ax^6 -3pt^2 x^4 + 2b x^3 + (3p+a) t^6=ax^6+(a+1)x^4+2bx^3-b^2,$$

$$t^2=-\dfrac{a+1}{3p},\tag2$$ $$\quad (3p+a)(a+1)^3=27b^2p^3,\quad p^3=\dfrac{c^2}4(3p+a),$$ where $$c= \dfrac2b\sqrt{\dfrac{(a+1)^3}{27}}.\tag3$$

Then $$4\left(\dfrac pc\right)^3-3\,\dfrac pc = \dfrac ac.$$ Taking in account identities $$4s^3-3s = \cos(3\arccos s),$$ one can obtain $$p=c\cos\left(\dfrac23k\pi+\dfrac13\arccos\dfrac ac\right),\tag4$$ where $\;k\in\{-1,0,1\}\;$ provides the real positive result.

$\color{brown}{\textbf{Substitution.}}$

Assume $$y=x+\dfrac {t^2}x,\tag5$$ then $$x^3+\dfrac{t^6}{x^3}=\left(x+\dfrac {t^2}x\right)^3-3t^2\left(x+\dfrac {t^2}x\right)=y^3-3t^2y,$$ Therefore, the given polynomial $(1)$ presented in the form of $$P(a,b,x) = x^3Q\left(a,b,x-\dfrac{a+1}{3px}\right),\tag6$$ where $$Q(a,b,y) = (a+3p)y^3-3py^2+(a+2p)\dfrac{a+1}p y+2b.\tag7$$ Therefore, the given polynomial is reduced the cubic polynomial of the general form with the known constant cofficients.

$\color{brown}{\textbf{Example.}}$

Let $\;a=7,b=5,\;$ then $\;c\approx1.741\,859\,372\,646,\; p\approx2.172\,466\,644\,575.\;$

Then WA gives $\;y\approx-0.227\,483\,364\,391,\;$ with the right root $\;x_1\approx 1.000\,000\,000\,000.\;$

$\color{brown}{\textbf{Conclusions.}}$

From $(2)-(7)$ should the next:

  • The roots of the given polynomial are solvable in the closed form via elementary functions;
  • Identity $\;P\left(a,b,-\dfrac{a+1}{3px}\right)=-\dfrac{(a+1)^6}{(3px)^6}P\left(a,b,x\right)\;$ defines the correspondence between the values on the positive and negative arguments.
  • Typos: It should be $x^3+\dfrac{t^6}{x^3}=\left(x+\dfrac {t^2}x\right)^3-3t^2\left(x+\dfrac {t^2}x\right)=y^3-3t^2y$. Also, in (4), it is $c$ rather than $\frac{1}{c}$. – River Li Mar 13 '21 at 16:26
  • 1
    Does this mean all sextic in the shape $$x^6+c_{1}x^4+c_{2}x^3+c_{3} = 0$$ are all solvable – Aderinsola Joshua Mar 14 '21 at 18:10
  • @AderinsolaJoshua If you are interested in (6), you can check it. I can not get (6). – River Li Mar 15 '21 at 01:03
  • @AderinsolaJoshua You nice answer also supports my thoughts: we can not reduce this sextic to cubic in general. I think (6) is incorrect. – River Li Mar 15 '21 at 02:28
  • Typos: $ax^6 +3pt^2 x^4 + 2b x^3 - (6p+a) t^6$ is incorrect, it should be $ax^6 +3pt^2 x^4 + 2b x^3 - (3p+a) t^6$. Then $(6p+a)(a+1)^3=27b^2p^3$ is incorrect, etc. – River Li Mar 15 '21 at 04:54
  • 1
    (7) is incorrect. By $y = x + \frac{t^2}{x} = x - \frac{a + 1}{3px}$, we have $$P(a, b, x) = x^3\left[(a + 3p)y^3 - 3px y^2 - 3t^2(a + 2p)y + 2b\right].$$ You see, we can not eliminate $x$. – River Li Mar 15 '21 at 09:38
  • 1
    Please check: It should be $$\frac{P(a,b,x)}{x^3} = (a + 2p)\left(x^3 + \frac{t^6}{x^3}\right) + p\left(x + \frac{t^2}{x}\right)^3
    • 3px\left(x + \frac{t^2}{x}\right)^2 + 2b.$$ If you substitution $y = x + \frac{t^2}{x} = x - \frac{a+1}{3px}$, you can not eliminate $x$.
    – River Li Mar 15 '21 at 09:39
  • 1
    As pointed out by River Li, the first assumption (writing $p(x)/x^3$ as third order polynomial in $y=x+t^2/x$) does not work out. This can be verified simply by deriving a system of equations for the coefficients of the unknown third-order polynomial which is not solvable. – weee Mar 15 '21 at 10:00
  • @Yuri Negometyanov: I noticed someone downvoted you. I want to say, you should not assume that someone specific downvoted your answer, only because they criticised your answer in the comment. (Perhaps some other users rather than weee and I downvoted you.) – River Li Mar 15 '21 at 10:54
  • @RiverLi Please, look at (2) and fix your own error. Upvoter belived to your critic, despite the correct example. – Yuri Negometyanov Mar 15 '21 at 12:20
  • @YuriNegometyanov I did not downvote you. According to (2), $t^2 = -\frac{a + 1}{3p}$, so $x + \frac{t^2}{x} = x - \frac{a + 1}{3px}$. Can you check (6) by inserting (7) into it, taking into account (1)? – River Li Mar 15 '21 at 12:29
  • @YuriNegometyanov According to Aderinsola Joshua's example, $ax^6+(a+1)x^4+2bx^3-b^2$ is not solvable for general $a, b$. – River Li Mar 15 '21 at 12:35
  • @YuriNegometyanov You can check $a = 6, b = 5$. – River Li Mar 15 '21 at 12:43
  • @RiverLi You can do that too. The answer is detalized. – Yuri Negometyanov Mar 15 '21 at 14:04
  • @weee Tnanks for a comment! I've tried to simplify the calculations. – Yuri Negometyanov Mar 15 '21 at 14:36
  • @YuriNegometyanov OK. Let us take a simple example $a = 2, b = 1$. From (3), $c=2$. From (4), $p = -1, -1, 2$ for $k = -1, 1, 0$. From (2), $t^2 = - \frac{1}{p}$. (Continue in next comment) – River Li Mar 15 '21 at 14:41
  • 1
    @YuriNegometyanov If $p = -1$, then $t^2 = 1$, $Q(2, 1, y) = -y^3 + 3y^2 + 2t^3$, $$x^3Q(2, 1, x + \frac{1}{x}) = -{x}^{6}+3,{x}^{5}-3,{x}^{4}+ \left( 2,{t}^{3}+6 \right) {x}^{3}-3 ,{x}^{2}+3,x-1.$$ If $p = 2$, then $t^2 = - \frac{1}{2}$, $Q(2, 1, y) = 8y^3 - 6y^2 + 3y + 2t^3$, $$x^3Q(2, 1, x - \frac{1}{2x})= 8,{x}^{6}-6,{x}^{5}-9,{x}^{4}+ \left( 2,{t}^{3}+6 \right) {x}^{3}+ 9/2,{x}^{2}-3/2,x-1.$$ However, $P(2, 1, x) = 2x^6 + 3x^4 + 2x^3 - 1$. For both cases, we have $P(a, b, x) \ne x^3Q(a, b, x - \frac{a+1}{3px})$. (7) is incorrect. – River Li Mar 15 '21 at 14:42
  • 1
    A quick comparison of the $x^6$-term in $P(x)$ (which is $ax^6$) and in $x^3Q(y(x))$ (which is $(a+2p)x^6$) shows that they can only be equal for $p=0$. – weee Mar 15 '21 at 14:46
  • 1
    @YuriNegometyanov You can not get (6) by (7). The reason why you make mistake is the following: $$\frac{P(a,b,x)}{x^3} = (a + 2p)\left(x^3 + \frac{t^6}{x^3}\right) + p\left(x + \frac{t^2}{x}\right)^3 - 3px\left(x + \frac{t^2}{x}\right)^2 + 2b.$$ However, you made a mistake $$\frac{P(a,b,x)}{x^3} = (a + 2p)\left(x^3 + \frac{t^6}{x^3}\right) + p\left(x + \frac{t^2}{x}\right)^3 - 3p\left(x + \frac{t^2}{x}\right)^2 + 2b.$$ The latter is just your (7) if you substitute $y = x + \frac{t^2}{x}$. – River Li Mar 15 '21 at 16:40
1
  1. By Descartes rule of signs, the polynomial $$p(x):=ax^6+(a+1)x^4+2bx^3-b^2$$ has exactly one positive root. It is so because there is exactly one change of sign in $p(x).$
    Similarly, there is exactly one change of sign in $p(-x).$ Our polynomial has one negative root.
    Some bounds of real roots are found below.

  2. Consider a monic polynomial $$q(x):=x^6+\frac{a+1}{a}x^4+2\frac{b}{a}x^3-\frac{b^2}{a}.\tag 1$$ Clearly, $q(x)$ and $p(x)$ share the roots.
    I assume that $a+1<2b$ and $b>2,$ from where $2\frac{b}{a}<\frac{b^2}{a}.$
    By Lagrange theorem 1.3., an upper bound for the positive real root is $\left(\frac{b^2}{a}\right)^{1/6}.$ Replacing $x$ by $-x$ in $(1)$ we get a lower bound for negative real root $-\left(2\frac{b}{a}\right)^{1/3}-\left(\frac{b^2}{a}\right)^{1/6}.$ Putting together, real roots $r_i$ satisfy $$-\left(2\frac{b}{a}\right)^{1/3}-\left(\frac{b^2}{a}\right)^{1/6}<r_i<\left(\frac{b^2}{a}\right)^{1/6}.$$

  3. The remaining four roots are in conjugate pairs, $\;(\alpha\pm i\beta)\;$ and $\;(\gamma\pm i\delta).$ From the coefficients of $(1)$ we deduce that the sum of real parts of all six roots is $0,$ and the product of the roots is $-b^2\over a.$ The latest can be formulated as "the product of modules of the roots equals $b^2 \over a$".
    "Lagrange over $\mathbb{C}$" (Theorem 1.5 in the above mentionned article) gives an upper bound for the absolute values of the roots $w_i,$ real or complex: $$|w_i|<\left(2\frac{b}{a}\right)^{1/3}+\left(\frac{b^2}{a}\right)^{1/6}.$$

  • For a < 4, (b/2)^(1/3) is a better upper bound. And if a > b^2 then (b/(a+1)) ^(1/4) will be better. – gnasher729 Mar 14 '21 at 19:39
  • 1
    @gnasher729 I do not say no. With my assumption, the bounds are good. Also, my solution gives an idea on how to proceed in a different situation. – Viera Čerňanová Mar 14 '21 at 21:39
1

One thing to note is that there is no closed form solution in terms of radicals (as hinted in a comment).

Notice that if $p_{a,b}(x)$ admits an expression of solutions by radicals, then specialising $a$ and $b$ yields a radical expression for the roots over $\mathbb{Q}$. In particular, consider $$p_{1,2}(x) = x^6+ 2 x^4 + 4 x^3 - 4 \in \mathbb{Q}[x]$$

Recall from galois theory that there is a radical expression for the roots of $p_{1,2}(x)$, then the galois group of $p_{1,2}(x)$ must be solvable.

We may verify (by Eisenstein at $2$) that $p_{1,2}$ is irreducible, and furthermore checking the reduction mod $p$ for a few $p$ can eventually see that $\operatorname{Gal}(p_{1,2}(x)) = S_6$ - alternatively use computer algebra (e.g., GaloisGroup in Magma). Indeed $S_6$ is nonsolvable.

As a remark, if you are wondering about a radical expression over $\mathbb{C}$, a similar argument works while considering $p_{a,2}(x) \in \mathbb{C}(a)$, showing this galois group is not solvable.

1

$$ p(x) = ax^6+(a+1)x^4+2bx^3-b^2$$ I earlier thought the polynomial was solvable, until I checked the Galois group of $6x^6+7x^4+10x^3-25 = 0$ $$ G \in [720,-1,1,"S6"]$$ If the Galois group of any polynomial above quartic is the Symmetric group, then the polynomial is unsolvable, and you don't need to ask for its root, but instead their approximation

Your polynomial is in two variable, so lucky enough some would be solvable, I wrote a program on pari/gp to check

for( a=-20, 20, for(b= -20, 20 , f= ax^6+(a+1)x^4+2bx^3-b^2; if(a!=0&&polisirreducible(f)&&polgalois(f)!=[720,-1,1,"S6"], print(a, " ", b, " " , polgalois(f)))))

The result is just

$$ 9 , -1 , [72, -1, 1, "F_36(6):2 = [S(3)^2]2 = S(3) wr 2"]$$ $$9 , 1 , [72, -1, 1, "F_36(6):2 = [S(3)^2]2 = S(3) wr 2"]$$

So if $a= 9$, $b= 1$ , then $9x^6+10x^4+4x^3-1 = $ is solvable and irreducible among range $ -20 \le a,b \le 20$

Now in case you still want to see a formula to $ ax^6+(a+1)x^4+2bx^3-b^2$ I can get it for you but would require a resolvent of degree $15$

EDIT:

R < x > := PolynomialRing (RationalField ()); f := 9x^6+10x^4+2*x^3-1; G := GaloisGroup (f); G; IsSolvable (G);

Working on magma, the result is

Permutation group $G$ acting on a set of cardinality $6$ Order = $72$ = $2^3 \cdot 3^2$

$(1, 2)(3, 4)(5, 6)$

$(1, 3, 6)$

$ (1, 3)$

true

  • It is a nice answer. I tried Sage code (copy from MSE) but it does not work on my Sage (some error). I wanted to try Maple, but I cannot find something like IsSolvable for Galois. You magma code works on http://magma.maths.usyd.edu.au/calc/. – River Li Mar 15 '21 at 02:42
  • Yes, can I see the sage code you tried... More specific, you should try to render the code directly on GAP – Aderinsola Joshua Mar 15 '21 at 14:56
  • Thanks. It is here: https://math.stackexchange.com/questions/4003336/is-this-sextic-equation-solvable. Copy from there: A. = QQ[] # defines A = Q[x] to be the polynomial ring in x over the rationals; poly = x^6 + 3x^5 + 3x^4 + 3x^3 + 2x^2 + 1; G = poly.galois_group(); G.is_solvable(); – River Li Mar 15 '21 at 15:16
  • G = poly.galois_group(); 【ValueError: Index n must be in {1,..,None}】 – River Li Mar 15 '21 at 15:22
  • Thanks, it worked fine for me... Check to see if you have that package of sage installed, or maybe you should take it step-by-step, A. = QQ[ ] ; poly = 9x^6+10x^4+2*x^3-1 ; G = poly.galois_group() ; print(G) ; G.is_solvable() ; – Aderinsola Joshua Mar 15 '21 at 16:19
  • Thanks. I looked into the error information. Perhaps some package is required: verbose 0 (2055: permgroup_named.py, cardinality) Warning: TransitiveGroups requ ires the GAP database package. Please install it with sage -i database_gap. – River Li Mar 15 '21 at 16:31
  • Hmm, you said "you should try to render the code directly on GAP", what do you mean? – River Li Mar 15 '21 at 16:32
  • Okay but I'm not very converse with gap, if you have gap installed you can check the manual(), if not... go for pari/gp – Aderinsola Joshua Mar 15 '21 at 17:46
  • 1
    Thank you very much. – River Li Mar 16 '21 at 00:39