13

Given that $a \ne b \ne c $ are non zero reals satisfying $$a^2-\frac{1}{b}=b^2-\frac{1}{c}=c^2-\frac{1}{a}=d \tag {*}$$ Then express $(a+b+c)abc$ purely as a function of $d$.

My hard effort:

Taking pair wise subtractions $$ a^{2}-\frac{1} {b}=b^{2}-\frac{1} {c}=c^{2}-\frac{1} {a} $$ $$ \Rightarrow( a-b ) ( a+b )=\frac{c-b} {b c} $$ $$ ( b-c ) ( b+c )=\frac{a-c} {a c} $$ $$ ( a-c ) ( a+c )=\frac{a-b} {a b} $$ $$ \Rightarrow( a+b ) ( b+c ) ( c+a )=\frac{-1} {a^{2} b^{2} c^{2}} \tag {1} $$ Now I assumed $a+b+c=p, ab+bc+ca=q, abc=r$. So $(1)$ becomes $$pq-r=\frac{-1}{r^2} \tag {2}$$ Also from $(*)$ adding all equations with $d$, we get $$ a^{2}+b^{2}+c^{2}-\frac{1} {a}-\frac{1} {b}-\frac{1} {c}=3 d $$ $$ \Rightarrow( a+b+c )^{2}-2 ( a b+b c+c a )-\frac{( a b+b c+c a )}{abc}=3 d $$ $$ \Rightarrow p^{2}-2 q-\frac{q} {r}=3 d \tag {3} $$ Now I eliminated $q$ from $(2),(3)$, so we get $$2r+1-\frac{2r+1}{r^3}=p^3-3pd \tag {4}$$ Any help from here?

CuteMath
  • 301
  • Did you try specific values of $d$ to see what $abc(a+b+c)$ equals to? That could help you guess the answer, and perhaps work back from there. – Calvin Lin Apr 02 '25 at 19:32
  • Note that solutions to (*) are also solutions to (4), but not necessarily vice versa. You potentially have introduced more values, which could make it hard to continue from (4). – Calvin Lin Apr 02 '25 at 19:35
  • 1
    I computed the solutions to the polynomial equations, and obtained $abc(a+b+c)+d^2=0$. – Dietrich Burde Apr 02 '25 at 20:22
  • @CalvinLin, How ? could you give your approach please – CuteMath Apr 03 '25 at 01:23
  • @CuteMath, your question is having a small typing mistake. Please edit it. You have mentioned $a\neq b\neq c\neq$ but I am confused that whether it is $a\neq b\neq c$ or $a\neq b\neq c\neq d$. Kindly edit it to make it Clear – Dev Apr 03 '25 at 03:08
  • I think it should be $-d^2$, I checked it using computer algebra. Basically, I consider the ideal $I = \langle a^2b - 1 - db, b^2c - 1 - dc, c^2a - 1 - da, (a + b + c)abc - t \rangle \leq \mathbb{Q}[a, b, c, d, t]$ and compute the primary decomposition of the elimination ideal $I \cap \mathbb{Q}[t, d]$. This will give one component for the case where $a, b, c$ are not different, and one component $\langle d^2 + t \rangle$. I guess this is not the solution you were looking for, but maybe it helps anyway. – Feanor Apr 03 '25 at 03:08
  • 1
    Ah yes, I did get $-d^2$. Typo on my part. – Calvin Lin Apr 03 '25 at 03:53
  • @CalvinLin, can you give some inputs please – CuteMath Apr 03 '25 at 06:13
  • @Feanor Actually, this was exactly the way I did it, but without computing a primary decomposition. It is obvious that a resultant of two of certain equations give $d^2+t=0$. – Dietrich Burde Apr 03 '25 at 10:03
  • @DietrichBurde You are correct, $abc(a+b+c) = -d^2$ but $(a,b,c)$ are roots of a sextic so there are two triplets . Kindly see answer below. – Tito Piezas III Apr 03 '25 at 10:38
  • 1
    @TitoPiezasIII Yes, I know. But you only need to derive one further "element" in the Gröbner basis, namely $d^2+t$. To find all solution over $\Bbb C$ and $\Bbb R$ is much more complicated (which you did!) – Dietrich Burde Apr 03 '25 at 10:45
  • @DietrichBurde I am quite fond of systems of equations and radicals, so couldn’t resist. :) – Tito Piezas III Apr 03 '25 at 11:15

4 Answers4

5

Here is a by-hand solution. As in the question body, let $p=a+b+c$, $q=ab+bc+ca$, and $r=abc$. Our aim is to find $pr$.

We can compute $$qd=\sum_{\mathrm{cyc}}bcd=\sum_{\mathrm{cyc}}bc\left(a^2-\frac1b\right)=\sum_{\mathrm{cyc}}(a^2bc-c)=p(r-1).\tag{$5$}$$ We first assume $r\not\in\{1,-1/2\}$; we will fix this up later. Combining this with identity (2) in the question body, i.e. $pq=r-1/r^2$, gives $$d(r^3-1)=pqr^2d=pr^2p(r-1),$$ so $(pr)^2=d(r^2+r+1)$. We now use the other identity (3) discovered in the question body: $p^2r-q(2r+1)=3dr$. Multiplying by $r$ gives $$qr(2r+1) = p^2r^2 - 3dr^2 = d(-2r^2+r+1)=-d(2r+1)(r-1).$$ Since $r\neq-1/2$, we conclude $$-d^2(r-1) = dqr(2r+1)=p(r-1)r;$$ since $r\neq 1$, we get $pr=-d^2$, as desired.


It remains to treat the degenerate cases $r=1$ and $r=-1/2$. For this we will use an additional sum: $$(p^2-2q)d=\sum_{\mathrm{cyc}}b^2d=\sum_{\mathrm{cyc}}b^2\left(a^2-\frac1b\right)=q^2-2pr-p.$$ (The last equality is by expanding out $q^2$.) Adding twice (5) gives $$p^2d=q^2-3p.\tag{$6$}$$

  • If $r=1$, then we have by (5) that $qd=0$. So either $q=0$ or $p=d=0$. If $d=0$ then $a^8=b^4=c^2=a$, and so $a=1$, meaning $b=c=1$, contradicting distinctness. So $q=0$. Then (3) gives $p^2=3d$ and (6) gives $p^2d=-3p$. These are enough to give $pr=p=-d^2$.

  • If $r=-1/2$, then (2) gives $pq=-9/2$, (3) gives $p^2=3d$, and (6) gives $p^2d+3p=q^2$. Using (2) to solve for $q$, (3) to solve for $d$, and plugging them into (6) gives $$\frac{p^4}3+3p=p^2d+3p=q^2=\frac{81}{4p^2}.$$ Simplifying gives $p^6+9p^3=243/4$, which factors as $(p^3+27/2)(p^3-9/2)=0$. If $p^3=9/2$, then $$-18pr=9p=2p^4=18d^2,$$ and so $pr=-d^2$ holds. Otherwise, $p=-3/\sqrt[3]{2}$. Then $q=3/\sqrt[3]{4}$, and so the cubic with roots $a$, $b$, and $c$ is $$x^3-\frac3{\sqrt[3]2}x^2+\frac3{\sqrt[3]4}x-\frac12.$$ This factors as $(x-\sqrt[3]{1/2})^3$, and so $a=b=c$ in this case. This finishes the proof.

  • +1 very nice! I had the same idea to start with $(5)$ but seems like I made algebra errors when combining it with $pr = r-1/r^2$ – dezdichado Apr 04 '25 at 13:52
3

You have three equations in three unknowns $(a,b,c)$, call this system $S$,

$$a^2-\frac{1}{b}=d\\ b^2-\frac{1}{c}=d\\ c^2-\frac{1}{a}=d$$

so can be solved where $a\neq b\neq c$. However, since this is a non-linear system, one can expect that $(a,b,c)$ are generally roots of non-linear equations as well.


I. Sextic solution

Using resultants, your $(a,b,c)=(x_1,\,x_2,\,x_3)$ are three roots (correctly ordered) of the sextic with a solvable Galois group,

$$(1 - d^3) x^6 - d^2 x^5 - 3d(1- d^3) x^4 + (1 + 2 d^3) x^3 + d^2(4 - 3 d^3) x^2 - d^4 x + (1- d^3)^2 = 0$$

with discriminant $D = ( \color{blue}{4 d^3 - 3})^3 (16 d^6 - 36 d^3 + 27)^2$. However, the sextic factors over a square root extension and we get two cubics,

$$x^3+\frac{d^2\big(1-\sqrt\beta\big)}{2(d^3-1)}x^2+\frac{d\big(3-2d^3-\sqrt\beta\big)}{2(d^3-1)}x+\frac{\big(1+\sqrt\beta\big)}{2}=0$$

for $\pm\sqrt{\beta}$ and where $\beta = \color{blue}{4 d^3 - 3}$. By the elementary symmetric polynomials, we have,

\begin{align} x_1+x_2+x_3 &= -\frac{d^2\big(1-\sqrt\beta\big)}{2(d^3-1)}\\[5pt] x_1x_2x_3 &=-\frac{\big(1+\sqrt\beta\big)}{2}\end{align}

where the product of both removes the square root and simplifies to,

$$abc(a +b + c) = x_1x_2x_3(x_1+x_2+x_3) = x_4x_5x_6(x_4+x_5+x_6) =\color{red}{-d^2}$$

which is essentially the same relation found by cybcat, Calvin Lin, and Dietrich Burde in a deleted answer. However, since the sextic has six roots, then for a given $d$, there are generally two triplets $(a,b,c)$.


II. Example

Let $\color{red}{d=2}$ and we get the sextic,

$$-7 x^6 - 4 x^5 + 42 x^4 + 17 x^3 - 80 x^2 - 16 x + 49=0$$

with one factor as,

$$14x^3+4(1+\sqrt{29})x^2-2(13-\sqrt{29})+7(1-\sqrt{29})=0$$

and the other the negative case of $\pm\sqrt{29}$. We arrange the roots of the first cubic as,

\begin{align} x_1 &\approx -1.711333828766285769348878\\ x_2 &\approx 1.0768163371956621036573432\\ x_3 &\approx -1.189815310467806057522953\end{align}

and the second cubic as,

\begin{align} x_4 &\approx 1.1742622838169903974214464\\ x_5 &\approx -1.610025723419787135772065\\ x_6 &\approx 1.6886676702126550329936787 \end{align}

Then,

$$abc(a +b + c) = x_1x_2x_3(x_1+x_2+x_3) = x_4x_5x_6(x_4+x_5+x_6) =\color{red}{-2^2}$$

Of course, these two triplets $x_k$ also satisfy the system $S$,

$$a^2-\frac{1}{b}=b^2-\frac{1}{c}=c^2-\frac{1}{a}=d=2$$

and so on for other $d$.


III. Cardano's Method

Using Cardano's method of solving cubics, one will have to use the negative square,

$$\gamma=\sqrt{-z^2\,}=\sqrt{-\left(8d^6-16d^3+9+(2d^2-3)\sqrt{4d^3-3\,}\right)^2\,}$$

Thus, for real $d$ such that $\sqrt{4d^3-3}\,$ is real, then both cubics are casus irreducibilis and the sextic will have six real roots as in the above example $d=2$. However, if $\sqrt{4d^3-3}\,$ is complex, then the sextic will have six complex roots like when $d=-2$.

A special case is when $d=1$ as the sextic will have a zero root. However, the other cubic is still usable and quite familiar as,

$$x^3-x^2-2x+1=0$$

the roots of which, correctly ordered are $(a,b,c)=(-2\cos \frac{2\pi}7,\, -2\cos \frac{6\pi}7,\, -2\cos \frac{4\pi}7)$ implying the nice trigonometric identities,

$$a^2-\frac{1}{b}=b^2-\frac{1}{c}=c^2-\frac{1}{a}=1$$ $$abc(a+b+c)=-1$$

  • Can you clarify please ? Does $a\neq b\neq c$ mean $a\neq b\wedge b\neq c\wedge a\neq c$ ? I would think no. Then does $2$ possible triples mean $abc(a+b+c)$ is not unique ? – nonuser Apr 03 '25 at 10:45
  • @nonuser It is unique. Also in the second case, for $d=2$, the result is $-d^2=-2^2$. But for $d=2$, we might have $a=b=c$. – Dietrich Burde Apr 03 '25 at 10:46
  • Thank you. You referenced the deleted answer, so I am confused. If correct why's deleted. – nonuser Apr 03 '25 at 10:48
  • @nonuser In general, the sextic yields two triples where $a\neq b\neq c$ with the same value $abc(a+b+c)$, as in the example above. I don't know why Burde deleted his answer, he may have found it incomplete. – Tito Piezas III Apr 03 '25 at 10:54
  • +1. I am wondering the original solution of the problem. Also I think that $a\neq b\neq c$ is problematic. Because,. e.g. $a=c$ is possible. $a\neq b\neq a$ which is true. – nonuser Apr 03 '25 at 10:56
  • @nonuser I deleted my answer, because this is a pre-calculus exercise, and I have used methods from commutative algebra, e.g., the resultant of two polynomials, and multivariate division. – Dietrich Burde Apr 03 '25 at 11:06
  • I do not agree with you. If the answer is correct, then we need to learn this answer anyway (your answer involves some advanced technique and it is very very nice to learn advanced techniques) – nonuser Apr 03 '25 at 11:07
  • @nonuser To messege someone, you have to use the @ symbol, otherwise the message goes to me. – Tito Piezas III Apr 03 '25 at 11:11
  • How does knowing the sextic and it's discriminant actually imply the desired equality ? – dezdichado Apr 03 '25 at 11:30
  • @dezdichado The discriminant refers more to the fact that the sextic has a solvable Galois group. The sextic arises from solving 3 equations in 3 unknowns. And the equation $abc(a+b+c)=z$ can then be solved using the sextic roots. – Tito Piezas III Apr 03 '25 at 12:15
  • 1
    Can you ask ? Why we cannot take $a=x_1,b=x_3, c=x_5$ ? – nonuser Apr 03 '25 at 14:56
  • @nonuser It’s the nature of non-linear systems. The final equation has degree greater than $1$, so you have to choose the correct combination of roots. – Tito Piezas III Apr 03 '25 at 15:29
  • It seems hard to me to understand this point really. – nonuser Apr 03 '25 at 15:32
  • @nonuser I've edited my answer and factored the sextic into two cubics over a square root extension. The three roots of one cubic form one triplet, while that of the second cubic form the second triplet. And that's why you can't mix up the roots. – Tito Piezas III Apr 03 '25 at 19:53
  • 1
    @dezdichado I've edited my answer and factored the sextic into two cubics. Using the elementary symmetric polynomials, it is now easy to find $x_1 x_2 x_3(x_1+x_2+x_3)$. – Tito Piezas III Apr 03 '25 at 19:56
1

Given

$$a^2-b^{-1}=b^2-c^{-1}=c^2-a^{-1}=d. $$

So we have

$$b a^2 - b d - 1=c b^2 - c d - 1=a c^2 - a d - 1=0. $$

Using Groebner basis computations, we have

$$x(b a^2 - b d - 1)+y(c b^2 - c d - 1)+z(a c^2 - a d - 1)\\ =(a-b) \left(a^2 b c+a b^2 c+a b c^2+d^2\right)=0. $$

By condition $a-b\ne 0$, then $a^2 b c+a b^2 c+a b c^2+d^2=0$, which is equivalent to

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

Here my $x,y,z$ is

$$ x=\\\tiny -\left(\left(c^2-d\right) \left(a^3 c (b d-1) \left(c^2-d\right)+a^2 c \left((b d-1) \left(c^2-d\right) \left(c^2 d+c-d^2\right)-b d\right)+a d \left(-\left(b^3-2\right) c^3+\left(b^3-2\right) c d-c^4 \left(b^2+2 b d^2-3 d\right)+c^6 (b d-1)+c^2 d \left(b \left(b+d^2\right)-2 d\right)-b d\right)-c^2 d (b d-1) \left(c^2-d\right)+1\right)\right), \\ y=\\\tiny a^3 \left(c^2-d\right) \left(c^4 d-c^3-2 c^2 d^2+d^3\right)+a^2 \left(-c^6+\left(6 c^3-1\right) c d^4-4 c^2 d^5+\left(-4 c^6+3 c^3-1\right) d^3+\left(c^6+c^3+1\right) c d+\left(c^6-3 c^3+1\right) c^2 d^2+d^6\right)+a \left(b^2 d \left(d \left(c^2-d\right)^2-1\right)+b (c d+1) \left(d \left(c^2-d\right)^2-1\right)+d \left(c^2-d\right) \left(\left(3 c^3-1\right) d^3+c^3+\left(c^3-1\right) c^4 d+\left(2-3 c^3\right) c^2 d^2-c d^4\right)\right)+c \left(c^6 \left(-d^2\right)+3 c^4 d^3-c^2 \left(3 d^4+d\right)+c+d^5\right),\\ z=\\\tiny a^4 b c (b d-1) \left(c^2-d\right)+a^3 b c \left((b d-1) \left(c^2-d\right) \left(c^2 d+c-d^2\right)-1\right)+a^2 \left(-c d \left(b c (b+c) \left(b^2 c+c^3-1\right)+1\right)-d^3 \left(b^2 \left(2 c^4+c\right)+2 b c^2+2 c^3-1\right)+c d^4 \left(b^2 c+1\right)+d^2 \left(b^4 c+b^3 c^2+b^2 \left(c^6+2 c^3-1\right)+b c \left(3 c^3-1\right)+c^2 \left(c^3-2\right)\right)+b c^2 \left(b c^2-1\right)\right)+a \left(-b^4 c d+b^3 \left(-c^2\right) d+b^2 \left(2 c^5 d^3+c^5-c^3 d^4+c^2+\left(-c^6+c^3+1\right) c d^2-\left(c^6+c^3-1\right) d\right)+b (c d+1)^2+d \left(c^3-c d-1\right) \left(c^2 d+c-d^2\right)^2\right)-b^2 c^3 \left(c d \left(d \left(c^2-d\right)^2+1\right)-1\right)+b c^2 d+d \left(c \left(c^5+\left(3 c^3-1\right) d^4+\left(c^3-1\right) c^3 d+\left(c^6-3 c^3+1\right) c d^2-3 \left(c^3-1\right) c^2 d^3-c d^5\right)-1\right). $$

Note that the $x,y,z$ above are given by Mathematica, so it is too difficult to check the proof manually. There should be shorter forms for $x,y,z$ that satisfy the equality. Attention is all you need. :)

btw: You can run the following code in Mathematica to check.

(*Output: (a-b) (a^2 b c+a b^2 c+a b c^2+d^2)*)

bas = {b a^2 - b d - 1, c b^2 - c d - 1, a c^2 - a d - 1}; s = {-((c^2 - d) (1 - a^3 c^3 - a^2 c^4 + a^3 c d - a^2 b c d + a^2 c^2 d + 2 a c^3 d + a^3 b c^3 d - a b^3 c^3 d + c^4 d + a^2 b c^4 d - a b^2 c^4 d - a^2 c^5 d - a c^6 d - a b d^2 - 2 a c d^2 - a^3 b c d^2 + a b^3 c d^2 - c^2 d^2 - a^2 b c^2 d^2 + a b^2 c^2 d^2 + 2 a^2 c^3 d^2 + 3 a c^4 d^2 - b c^4 d^2 + a^2 b c^5 d^2 + a b c^6 d^2 - a^2 c d^3 - 2 a c^2 d^3 + b c^2 d^3 - 2 a^2 b c^3 d^3 - 2 a b c^4 d^3 + a^2 b c d^4 + a b c^2 d^4)), -a b + c^2 - a^3 c^5 - a^2 c^6 - a b^2 d + a^2 c d - a b c d - c^3 d + a^3 c^3 d + a^2 c^4 d + a b c^4 d + a c^5 d + a^3 c^6 d + a^2 c^7 d + a^2 c^2 d^2 - 2 a b c^2 d^2 - a c^3 d^2 - 3 a^3 c^4 d^2 + a b^2 c^4 d^2 - 3 a^2 c^5 d^2 + a b c^5 d^2 - a c^6 d^2 - c^7 d^2 + a^2 c^8 d^2 + a c^9 d^2 - a^2 d^3 + a b d^3 + 3 a^3 c^2 d^3 - 2 a b^2 c^2 d^3 + 3 a^2 c^3 d^3 - 2 a b c^3 d^3 + 3 a c^4 d^3 + 3 c^5 d^3 - 4 a^2 c^6 d^3 - 4 a c^7 d^3 - a^3 d^4 + a b^2 d^4 - a^2 c d^4 + a b c d^4 - 3 a c^2 d^4 - 3 c^3 d^4 + 6 a^2 c^4 d^4 + 6 a c^5 d^4 + a d^5 + c d^5 - 4 a^2 c^2 d^5 - 4 a c^3 d^5 + a^2 d^6 + a c d^6, a b - a^3 b c - a^2 b c^2 + a b^2 c^2 - a^4 b c^3 + b^2 c^3 - a^3 b c^4 + a^2 b^2 c^4 + a b^2 c^5 - d + a b^2 d - a^2 c d + 2 a b c d + a^4 b c d - a b^4 c d - a c^2 d + b c^2 d + a^3 b c^2 d + a^2 b^2 c^2 d - a b^3 c^2 d + a^2 b c^3 d - a b^2 c^3 d + a^4 b^2 c^3 d - a^2 b^4 c^3 d - b^2 c^4 d + a^3 b^2 c^4 d - a^2 b^3 c^4 d + a c^5 d - a^3 b c^5 d - a^2 b^2 c^5 d + c^6 d - a^2 b c^6 d - a b^2 c^6 d - a^2 b^2 d^2 - a^2 b c d^2 + a b^2 c d^2 - a^4 b^2 c d^2 + a^2 b^4 c d^2 - 2 a^2 c^2 d^2 + a b c^2 d^2 - a^3 b^2 c^2 d^2 + a^2 b^3 c^2 d^2 - 3 a c^3 d^2 + 2 a^3 b c^3 d^2 + 2 a^2 b^2 c^3 d^2 - c^4 d^2 + 3 a^2 b c^4 d^2 + a b^2 c^4 d^2 + a^2 c^5 d^2 + a^3 b^2 c^5 d^2 + 2 a c^6 d^2 + a^2 b^2 c^6 d^2 + c^7 d^2 - a b^2 c^7 d^2 - b^2 c^8 d^2 + a^2 d^3 + 2 a c d^3 - a^3 b c d^3 - a^2 b^2 c d^3 + c^2 d^3 - 2 a^2 b c^2 d^3 - 2 a^2 c^3 d^3 - 2 a^3 b^2 c^3 d^3 - 5 a c^4 d^3 - 2 a^2 b^2 c^4 d^3 - 3 c^5 d^3 + 2 a b^2 c^5 d^3 + 2 b^2 c^6 d^3 + a c^7 d^3 + c^8 d^3 + a^2 c d^4 + a^3 b^2 c d^4 + 4 a c^2 d^4 + a^2 b^2 c^2 d^4 + 3 c^3 d^4 - a b^2 c^3 d^4 - b^2 c^4 d^4 - 3 a c^5 d^4 - 3 c^6 d^4 - a d^5 - c d^5 + 3 a c^3 d^5 + 3 c^4 d^5 - a c d^6 - c^2 d^6}; Factor[bas . s]

cybcat
  • 982
1

A way to solve it with the help of a symbolic processor and Groebner basis

solrd = Eliminate[{Numerator[Together[a^2 - 1/b - d]] == 0, 
                   Numerator[Together[b^2 - 1/c - d]] == 0, 
                   Numerator[Together[c^2 - 1/a - d]] == 0, 
                   (a + b + c) a b c ==  r}, {a, b, c}] // FullSimplify

(* (d^2 + r) (-27 + 9 d (-4 + d^3) r - 6 d^2 r^2 + r^3) == 0 *) Solve[solrd, r]

Cesareo
  • 36,341