退出

An elliptic curve for $x_1^5+x_2^5+x_3^5=y_1^5+2y_2^5$?

椭圆曲线 Math StackExchange 5 票 2 回答 101 浏览 提问者: Tito Piezas III 2026-06-08 13:45
polynomials diophantine-equations elliptic-curves

问题内容

In a prior post, the equation,

$$x_1^5+2x_2^5 = y_1^5+2y_2^5$$

was considered. It has only one known primitive solution. This present post considers the similar,

$$x_1^k+x_2^k+x_3^k = y_1^k+2y_2^k$$

valid for both $k = (1,5)$. Duncan Moore found only one primitive solution, namely,

$$85333^k + 72631^k + 36981^k = 89703^k + 2\times52621^k$$

We can encode this in the form,

$$(a + c)^k + (-a + c)^k + (b + d)^k = (-b + d)^k + 2(b + c)^k $$

already true for $k=1$. Expanding for $k=5$, one must solve the quartic in $d$,

$$b (b^2 + d^2)^2 + c (a^2 + c^2)^2 = (b + c )(b^2 + c^2)^2$$

where,

$$d = \frac1b\sqrt{-b^4\pm b\sqrt{\color{blue}{b^6 - (a^4 - b^4) b c + 2 b^4 c^2 - 2 (a^2 - b^2) bc^3 + b^2 c^4}\,}}$$

The only rational solution $(a,b,c,d)$ so far is Moore's,

\begin{align} a &=+87\times73\\ b &=-87\times101\times3\\ c &=782\times101\\ d &=782\times81\end{align}

where the shared factors I think are not coincidence. The quartic polynomial in $c$ must be made a square,

$$b^6 - (a^4 - b^4) b c + 2 b^4 c^2 - 2 (a^2 - b^2) bc^3 + b^2 c^4 = y^2$$

With one initial rational point, it can be treated as an elliptic curve. I've found four polynomial solutions for $c$,

$$-b,\quad -\frac{(a^2+b^2)^2+4b^4}{4b(a^2+b^2)},\quad \frac{(a^2+b^2)^2}{4b^3},\quad -\frac{(a^2+b^2)(a^2-3b^2)}{4a^2b}$$

This causes the quartic in $d$ to factor into two quadratics, but seem unsolvable in the rationals. For example, let,

$$c = -\frac{(a^2+b^2)(a^2-3b^2)}{4a^2b}$$

and the quartic in $d$ has one factor solvable as,

$$d = \frac{(a^2+b^2)\sqrt{3(a^2+b^2)(a^2-3b^2)}}{4a^2b}$$

but there is no non-zero rational $(a,b)$ such that $d$ is also rational.

Question: Is there a rational polynomial $c=\dfrac{P_1(a,b)}{P_2(a,b)}$ of small degree less than $4$ such that the quartic in $d$ factors into two quadratics, one of which for the right $(a,b)$ is solvable in the rationals?

回答 (2)

Srinivasa Raghava 1 票 2026-06-08 16:57 原文

Using the substitution, the fifth-power equation reduces to

$$b(b^2+d^2)^2+c(a^2+c^2)^2=(b+c)(b^2+c^2)^2.$$

This is the main equation to study.

Assume $b\neq 0$, and put

$$r=\frac{a}{b}, \qquad u=\frac{c}{b}, \qquad v=\frac{d}{b}.$$

After dividing by $b^5$, the equation becomes

$$(1+v^2)^2+u(r^2+u^2)^2=(1+u)(1+u^2)^2.$$

Therefore

$$(1+v^2)^2 = u^4-2(r^2-1)u^3+2u^2-(r^4-1)u+1.$$

Define

$$q(r,u)=u^4-2(r^2-1)u^3+2u^2-(r^4-1)u+1.$$

Then the condition is

$$q(r,u)=(1+v^2)^2.$$

This is the important point. It is not enough to make $q(r,u)$ a square. We need its square root to have the special form

$$1+v^2.$$

So the problem has two layers. First, we may study

$$z^2=q(r,u).$$

This is an elliptic curve in disguise. But the original problem also needs

$$z=1+v^2.$$

This second condition is essential.

If we take

$$u=-1.$$

This means

$$c=-b.$$

Substituting $u=-1$ into $q(r,u)$, we get

$$q(r,-1)=(r^2+1)^2.$$

Hence we may take

$$1+v^2=r^2+1.$$

So

$$v^2=r^2,$$

and therefore

$$v=\pm r.$$

Since $v=d/b$ and $r=a/b$, this gives

$$d=\pm a.$$

Therefore we have the family

$$\boxed{c=-b,\qquad d=\pm a.}$$

This works for every rational choice of $a$ and $b$, with $b\neq 0$.

Although the family above is valid, it is not a genuinely new solution. Indeed, if

$$c=-b,$$

then

$$y_2=b+c=0.$$

For example, if $d=a$, then

$$x_1=a-b, \qquad x_2=-a-b, \qquad x_3=a+b,$$

$$y_1=a-b, \qquad y_2=0.$$

The equation becomes

$$(a-b)^5+(-a-b)^5+(a+b)^5=(a-b)^5.$$

The two terms

$$(-a-b)^5 \qquad\text{and}\qquad (a+b)^5$$

cancel each other. Thus the family is correct, but it is trivial in spirit.

A non-trivial example

A much more interesting solution is obtained from

$$a=6351, \qquad b=-26361, \qquad c=78982, \qquad d=63342.$$

Then

$$x_1=a+c=85333,$$

$$x_2=-a+c=72631,$$

$$x_3=b+d=36981,$$

$$y_1=-b+d=89703,$$

$$y_2=b+c=52621.$$

Therefore

$$85333^5+72631^5+36981^5=89703^5+2\cdot 52621^5.$$

Also,

$$85333+72631+36981=89703+2\cdot 52621.$$

In normalized form, this solution is

$$r=\frac{a}{b}=-\frac{73}{303}, \qquad u=\frac{c}{b}=-\frac{782}{261}, \qquad v=\frac{d}{b}=-\frac{7038}{2929}.$$

For these values,

$$q(r,u)=(1+v^2)^2.$$

So this example satisfies both requirements: the elliptic-curve square condition and the extra condition needed to recover $d$.

Here the elliptic curve appears from

$$z^2=q(r,u).$$

But this is only the first part of the problem. The original fifth-power equation requires the stronger condition

$$z=1+v^2.$$

So the correct equation to study is

$$u^4-2(r^2-1)u^3+2u^2-(r^4-1)u+1=(1+v^2)^2.$$

The simple family

$$\boxed{c=-b, \qquad d=\pm a}$$

gives rational solutions of low degree, but these solutions are degenerate because $y_2=0$. For non-trivial solutions, one should search for rational points on

$$q(r,u)=(1+v^2)^2.$$

The example

$$(85333,72631,36981;89703,52621)$$

is a non-trivial solution of this type.

Hope this is helpful!

Srinivasa Raghava 0 票 2026-06-08 17:53 原文

@Ataulfo

You are right that a quartic model

$$y^2=f(x)$$

gives an elliptic curve only when $f(x)$ is square-free, and one should check this condition. In the present case, the quartic is

$$q_r(u)=u^4-2(r^2-1)u^3+2u^2-(r^4-1)u+1.$$

Its discriminant is

$$-(r-1)^4(r+1)^4\left(32r^{10}+59r^8-28r^6-110r^4-348r^2-117\right).$$

Hence, for rational $r$, the quartic is square-free for

$$r\neq \pm 1.$$

Also, the curve has the rational point

$$(u,z)=(-1,r^2+1).$$

Therefore, for rational $r\neq \pm 1$, it is indeed birationally equivalent to an elliptic curve. The cases

$$r=\pm 1$$

are degenerate and should be excluded.


@David

Thank you. That numerical identity is interesting, but it belongs to a related, slightly different equation.

The identity should be written with the correct sign as

$$1355^5+2(526)^5=1685^5+2(-1349)^5.$$

Equivalently,

$$526^5+1355^5+526^5=1685^5-2(1349)^5.$$

So it gives a solution of the equation

$$a^5+2b^5=c^5+2d^5.$$

However, the present problem requires more than a fifth-power identity. It also requires the corresponding first-power identity. In other words, the same variables must work for both powers

$$k=1 \quad \text{and} \quad k=5.$$

For the suggested numerical identity, we would have

$$x_1=526, \qquad x_2=1355, \qquad x_3=526,$$

$$y_1=1685, \qquad y_2=-1349.$$

The fifth-power equation is true:

$$526^5+1355^5+526^5=1685^5+2(-1349)^5.$$

But the first-power equation is not true, because

$$526+1355+526=2407,$$

whereas

$$1685+2(-1349)=-1013.$$

Thus

$$2407\neq -1013.$$

This is why the example cannot be used directly in the parametrization

$$x_1=a+c, \qquad x_2=-a+c, \qquad x_3=b+d,$$

$$y_1=-b+d, \qquad y_2=b+c.$$

This parametrization automatically makes the first-power equation true.

Indeed, if one tries to force the first four values by setting

$$x_1=526, \qquad x_2=1355, \qquad x_3=526, \qquad y_1=1685,$$

then the parametrization gives

$$a=\frac{x_1-x_2}{2}=-\frac{829}{2},$$

$$c=\frac{x_1+x_2}{2}=\frac{1881}{2},$$

$$b=\frac{x_3-y_1}{2}=-\frac{1159}{2},$$

$$d=\frac{x_3+y_1}{2}=\frac{2211}{2}.$$

Therefore

$$y_2=b+c=-\frac{1159}{2}+\frac{1881}{2}=361.$$

So the parametrization would force

$$y_2=361,$$

not

$$y_2=-1349.$$

Hence the smaller numerical identity is useful for the different equation

$$a^5+2b^5=c^5+2d^5,$$

but it does not directly help with the simultaneous problem for

$$k=1,5.$$

For the present elliptic-curve construction, Duncan Moore's larger example remains the relevant one because it satisfies both powers.