10

$a,b,c$ are the (real) roots of $x^3-16x^2-57x+1=0$. Prove that $\sqrt[5]{a} + \sqrt[5]{b} +\sqrt[5]{c} = 1 $

................................................................................

edit : my answer to this question on another forum

Let $p=\sqrt[5]{a},q=\sqrt[5]{b}, r=\sqrt[5]{c}, u=p+q+r, v=qr+rp+pq, w=pqr$

From given cubic p⁵+q⁵+r⁵=16, q⁵r⁵+r⁵p⁵+p⁵q⁵=-57, p⁵q⁵r⁵=-1 (so w=-1)
Thus results at end gives the following simultaneous equations for u,v

p⁵+q⁵+r⁵ = 16 = u⁵-5u³v+5uv²-5u²+5v q⁵r⁵+r⁵p⁵+p⁵q⁵ = -57 = v⁵+5uv³+5u²v+5v²+5u

I solved these via Mathematica to give u=1, v=-2 (only real solution) so your answer is 1

====================

p,q,r are roots of z³-uz²+vz-w=0

Multiply by zⁿ and let s(n)=pⁿ+qⁿ+rⁿ

Set z=p,q,r in turn and sum for the recurrence s(n+3) = us(n+2)-vs(n+1)+ws(n)

Noting that s(0)=3, s(1)=u, s(2)=u²−2v generate s(3), s(4), s(5) as follows …

n=0 : s(3) = us(2)-vs(1)+ws(0) = u³-3uv+3w

n=1 : s(4) = us(3)-vs(2)+ws(1) = u⁴-4u²v+2v²+4uw

n=2 : s(5) = us(4)-vs(3)+ws(2) = u⁵-5u³v+5uv²+5u²w-5vw

p⁵+q⁵+r⁵ = u⁵-5u³v+5uv²+5u²w-5vw

Replace p by qr, q by rp, r by pq and get

q⁵r⁵+r⁵p⁵+p⁵q⁵ = v⁵-5uv³w+5u²vw²+5v²w²-5uw³

====================

URCHIN
  • 119
  • @molarmass OP made an edit. –  Aug 16 '18 at 08:47
  • 2
    $x^{15}-16x^{10}-57x^5+1$ factors as

    Hint: $(x^3-x^2-2x+1)(x^{12}+x^{11}+3x^{10}+4x^9+9x^8-2x^7+12x^6-x^5+25x^4+11x^3+5x^2+2x+1)$ and the sum of the roots of the first factor is $1$.

    –  Aug 16 '18 at 09:05
  • Why is this closed ? The OP did show some non-trivial effort, even though presented via an uncommon medium. –  Aug 16 '18 at 09:08
  • @YvesDaoust In that link I only see efforts that others have done to answer his question. – Jaap Scherphuis Aug 16 '18 at 09:11
  • @JaapScherphuis: the first answer is by the OP himself. Check the signatures. –  Aug 16 '18 at 09:12
  • @YvesDaoust Oh, right. The person who asked the question there is not the person who asked it here. Confusing. – Jaap Scherphuis Aug 16 '18 at 09:14
  • @urchin You should reproduce your solution here, as this creates confusion. –  Aug 16 '18 at 09:15
  • @YvesDaoust - thanks - I never thought I'd get anywhere with trying to factor a 15th degree polynomial ! However, the question remains of how to do this for a cubic with real roots that leads to a 15th degree equation without a cubic factor with rational coefficients. – URCHIN Aug 16 '18 at 09:33
  • @URCHIN: if your question gets unlocked, I can explain a simple solution. –  Aug 16 '18 at 09:34

2 Answers2

4

Let $A,B,C$ be the roots of the cubic polynomial $$ g(x) = x^3 - x^2 - 2x + 1\ . $$ So $A,B,C$ satisfy (Vieta): $$ \begin{aligned} e_1 &=A + B + C &&=+1\ ,\\ e_2 &= AB + BC + CA &&=-2\ ,\\ e_3 &= A B C &&= -1\ . \end{aligned} $$ (Computer support motivating the above is postponed.) These values are numerically:

sage: R.<x> = PolynomialRing(QQ)
sage: g = x^3 - x^2 - 2*x + 1
sage: g(x).roots(ring=QQbar, multiplicities=False)
[-1.246979603717467?, 0.4450418679126288?, 1.801937735804839?]

so they are real. We compute the polynomial with roots $a'=A^5$, $b'=B^5$, $c'=C^3$.

(We "hope" that $a=a'$, $b=b'$, $c=c'$.)

For this we have to start computations involving symmetrical polynomials, trying to get $e_1(a',b',c')$, $e_2(a',b',c')$, $e_3(a',b',c')$, in terms of $e_1(A,B,C)$, $e_2(A,B,C)$, $e_3(A,B,C)$. Here, $e_1$, $e_2$, $e_3$ are the first three symmetric polynomials. (We have formally $e_4=e_5=\dots=0$.)

One result is immediate: $$ e_3(a',b',c') =a'b'c' =A^5B^5C^5=(ABC)^5=e_3(A,B,C)^5=-1\ . $$ We need now the Newton polynomial $p_5$ in Newton's identities of fifth degree... $$ \begin{aligned} e_1(a',b',c') &= a'+b'+c' \\ &=A^5+B^5+C^5 \\ &=p_5(A,B,C)\ ,\\ &\qquad\text{so we compute successively}\\ p_1(A,B,C) &= e_1(A,B,C)=1\ ,\\ p_2(A,B,C) &= (e_1p_1-2e_2)(A,B,C)=1\cdot 1-2\cdot(-2)=5\ ,\\ p_3(A,B,C) &= (e_1p_2-e_2p_1+3e_2)(A,B,C)=1\cdot 5-(-2)\cdot1+3\cdot(-1)=4\ ,\\ p_4(A,B,C) &= (e_1p_3-e_2p_2+e_3p_1)(A,B,C)=1\cdot 4-(-2)\cdot5+(-1)\cdot1=13\ ,\\ e_1(a',b',c') &=p_5(A,B,C) \\&=(e_1p_4-e_2p_3+e_3p_2)(A,B,C)=1\cdot13-(-2)\cdot4+(-1)\cdot5=16\ . \end{aligned} $$ We need now finally $a'b'+b'c'+c'a'$. For this, we repeat the same procedure as above, but not for $A,B,C$, but for $s,t,u$, which are respectively $AB$, $BC$, $CA$, with $$ \begin{aligned} e_1(s,t,u) &=s+t+u=AB + BC + CA &&=-2\ ,\\ e_2(s,t,u) &=st+tu+us= ABC(A+B+C) &&=-1\ ,\\ e_3(s,t,u) &=stu= A^2 B^2 C^2 &&= +1\ . \end{aligned} $$ So we have: $$ \begin{aligned} p_1(s,t,u)&=e_1(s,t,u)=-2\ ,\\ p_2(s,t,u)&=(e_1p_1-2e_2)(s,t,u)=(-2)\cdot (-2)-2\cdot(-1)=6\ ,\\ p_3(s,t,u)&=(e_1p_2-e_2p_1+3e_2)(s,t,u)=(-2)\cdot6-(-1)\cdot(-2)+3\cdot1=-11\ ,\\ p_4(s,t,u)&=(e_1p_3-e_2p_2+e_3p_1)(s,t,u) =(-2)\cdot(-11)-(-1)\cdot6+1\cdot(-2)=26\ ,\\ e_2(a',b',c')&=p_5(s,t,u) \\ &=(e_1p_4-e_2p_3+e_3p_2)(s,t,u) =(-2)\cdot26-(-1)\cdot(-11)+1\cdot6=-57\ . \end{aligned} $$ So $a',b',c'$ are the roots of the given polynomial $$ x^3-E_1x^2+E_2x-E_3 = x^3 - 16x^2 -57 x +1\ . $$ So $a',b',c'$ are (up to reordering) the "given" values $a,b,c$. We conclude: $$ a^{1/5} + b^{1/5} + c^{1/5} =A+B+C=1\ . $$

$\square$

Numerical support for the computations done so far, and the motivation for the abrupt start with the roots $A,B,C$ of the polynomial $x^3 - x^2 - 2x + 1$:

Sage code:

R.<x> = PolynomialRing(QQ)
f = x^3  - 16*x^2 - 57*x + 1
a, b, c = f.roots( ring=QQbar, multiplicities=False )
a, b, c
a^(1/5), b^(1/5), c^(1/5)

This gives:

(-3.015065490237851?, 0.01745839634379104?, 18.99760709389406?)
(1.008827691046369? + 0.7329562209746396?*I,
 0.4450418679126288?,
 1.801937735804839?)

OK, the computer needs human assitance to get the right fifth root of $a$ in $\Bbb R$.

 A, B, C = -(-a)^(1/5), b^(1/5), c^(1/5)

giving

sage: A, B, C
(-1.246979603717467?, 0.4450418679126288?, 1.801937735804839?)

this is good, three real numbers. Let us compute the elementary symmetric functions, and the first Newton polynomials for them:

print "A + B + C    = %s" % (A+B+C)
print "AB + BC + CA = %s" % (A*B + B*C + C*A) 
print "A B C        = %s" % (A*B*C)

for k in [1..5]:
    print "A^%s + B^%s + C^%s = %s" % (k, k, k, A^k+B^k+C^k)

This gives so far:

A + B + C    = 1.000000000000000?
AB + BC + CA = -2.000000000000000?
A B C        = -1.000000000000000?
A^1 + B^1 + C^1 = 1.000000000000000?
A^2 + B^2 + C^2 = 5.000000000000000?
A^3 + B^3 + C^3 = 4.000000000000000?
A^4 + B^4 + C^4 = 13.00000000000000?
A^5 + B^5 + C^5 = 16.00000000000000?

Same for the values $s,t,u$:

s, t, u = A*B, B*C, C*A

print "s + t + u    = %s" % (s+t+u)
print "st + tu + us = %s" % (s*t + t*u + u*s) 
print "s t u        = %s" % (s*t*u)

for k in [1..5]:
    print "s^%s + t^%s + u^%s = %s" % (k, k, k, s^k+t^k+u^k)

This gives:

s + t + u    = -2.000000000000000?
st + tu + us = -1.000000000000000?
s t u        = 1.000000000000000?
s^1 + t^1 + u^1 = -2.000000000000000?
s^2 + t^2 + u^2 = 6.000000000000000?
s^3 + t^3 + u^3 = -11.00000000000000?
s^4 + t^4 + u^4 = 26.00000000000000?
s^5 + t^5 + u^5 = -57.00000000000000?
dan_fulea
  • 32,856
  • The auxiliary cubic polynomial is readily obtained as a factor of $p(x^5)$ where $p$ is the original cubic. –  Aug 16 '18 at 09:22
  • Yes, as in the comments to the OP. My problem was to see its roots and the fact that they are real. Else one has to take the suitable branch of the fifth root. to match this one "beautiful factor". People wanted computations explicitly, i did it explicitly. On a piece of paper with own notations the calculus is done in some five minutes, typing all the stuff... – dan_fulea Aug 16 '18 at 09:28
  • Computationally we can simply go with (-(-a)^(1/5) + b^(1/5) + c^(1/5)).minpoly(), which tells us that it's the root of x-1, i.e. it equals 1. – Max Alekseyev Jun 21 '23 at 17:31
3

The given polynomial $p(x)$ has three real roots so $p(x^5)$ also has three real roots as all others are complex fifth roots of reals.

So we need to factor $p(x^5)=q(x)r(x)$ where $q$ is cubic with real roots and $r$ is an unimportant polynomial of degree twelve with complex ones.

With a simple numerical solver or with Cardano's formulas, we estimate the coefficients of $q$ using Vieta's formulas on the fifth roots of $p$, giving

$$q(x)\approx (x-1.80194) (x-0.445042) (x+1.24698)\\\approx x^3-1.000002 x^2-2.000002633x+1.000001871.$$

So we can hypothetize $q(x)=x^3-x^2-2x+1$ and long division confirms this hypothesis exactly *.

Then the sum of the real roots of $q$ is $1$.


*$$\dfrac{x^{15}-16x^{10}-57x^5+1}{x^3-x^2-2x+1}\\=x^{12}+x^{11}+3x^{10}+4x^9+9x^8-2x^7+12x^6-x^5+25x^4+11x^3+5x^2+2x+1.$$

  • Yes, a useful trick. Wolfram will also factor p(x⁵) directly, which is something |I should have tried. – URCHIN Aug 16 '18 at 14:57
  • @URCHIN: the next challenge is to find a way to solve fully by hand. –  Aug 16 '18 at 21:27