A while ago I asked this question on MSE here. After placing a bounty it got quite a bit of attention but unfortunately it has yet to be resolved. After getting some advice from MO Meta I have decided to post the question here (note this is the same as area just multiplied by two and $a_n,b_n > 0$).


$$\frac{a_1b_1}{c_1^2}+\frac{a_2b_2}{c_2^2} = \frac{a_3b_3}{c^2_3}$$

for any three different primitive Pythagorean triples $(a_n,b_n,c_n)$?

My personal belief, for what it's worth, is that this does not occur and I am actively trying to disprove it. However, I would appreciate a counter example just as much.

Some results so far:

User @mathlove on MSE has found the following necessary condition,

The following is a necessary condition for $c_i.$

It is necessary that for every prime $p$, $$ \nu_p(c_1) \leq \nu_p(c_2)+\nu_p(c_3)$$ $$\nu_p(c_2)\le \nu_p(c_3)+\nu_p(c_1)$$ $$\nu_p(c_3)\le \nu_p(c_1)+\nu_p(c_2)$$ where $\nu_p(c_i)$ is the exponent of $p$ in the prime factorization of $c_i$.

(Proof can be found here)

In order to search for these values I created an exhaustive algorithm to search for these triples with help from here. This is what I've found,

For $c^2 < 10^{14}$

$$\frac{a_1b_1}{c_1^2}+\frac{a_2b_2}{c_2^2} \neq \frac{a_3b_3}{c^2_3}$$


$$\frac{1}{c_1^2}+\frac{1}{c_2^2} \neq \frac{1}{c^2_3}$$

Note that the trouble finding these triples appears to come from dividing by the square of the hypotenuse as there are many solutions to $a_1b_1 + a_2b_2 = a_3b_3$. It seemed at first that these solutions may just be extremely unlikely to occur (ratios lining up perfectly) hence why nothing was found but it looks like there is a little more to this. A bug in my initial code made me accidentally search for solutions to this,

$$\frac{a_1b_1}{c_1}+\frac{a_2b_2}{c_2} = \frac{a_3b_3}{c_3}$$

Which yielded these very interesting values for $c < 10^7$,

$$\frac{3*4}{5} + \frac{20*21}{29} = \frac{17*144}{145}$$ $$\frac{20*21}{29} + \frac{119*120}{169} = \frac{99*4900}{4901}$$ $$\frac{119*120}{169} + \frac{696*697}{985} = \frac{577*166464}{166465}$$ $$\frac{696*697}{985} + \frac{4059* 4060}{5741} = \frac{3363*5654884}{5654885}$$

This pattern has a well defined structure to it. Notice the recursive nature where one of the terms always comes from the sum of the previous. Additionally the LHS numerators are both one apart and on the RHS the $b_3$ and $c_3$ are also a distance one from each other.

Background and motivation A resolution one way or another to the original question could help to resolve a couple of (presumably not so important) but pesky open problems in number theory. I am preparing a website that I will link to eventually that will give the full background. However, it is too lengthy for this post and in accordance with advice from META MO I will omit it to keep this as brief as possible. Additionally I am not a research level mathematician so please forgive any unintended ignorance when responding to comments.

Edit for bounty Joe Silverman and Constantin-Nicolae Beli have already given some good insight into the problem, I am putting a bounty on this with the hope that it will get more attention. I don't have much reputation so doing anything, even just commenting would go a long way for me. Looking at the problem as it stands I see one main problem and a subproblem that may help answer the main problem.

Main problem

Prove the original statement or find a counter example.


Is the solution set mentioned by Constantin-Nicolae Beli the only solutions to $\frac{a_1b_1}{c_1}+\frac{a_2b_2}{c_2} = \frac{a_3b_3}{c_3}$ and would that also be the same solution set for when the hypotenuse is squared?

Important update

Going back and looking at the background of where this comes from. I found that what I am asking for is equivalent to this,


Where $c_n$ is the associated hypotenuse of the primitive triple and $x_n$ is an integer solution to the circle,


I wanted to mention this connection as it is related to the solution set mentioned by Constantin-Nicolae Beli.

  • 1
    $\begingroup$ It may be worth noting that any version where some of the $a_n, b_n < 0$ can be rearranged so all are positive - so it's equivalent to ask if there are any nontrivial answers (where "trivial" means one of the $a_n, b_n = 0$) $\endgroup$
    – user44191
    Dec 27, 2019 at 5:00
  • 1
    $\begingroup$ For subproblem 1, the answer seems like it should trivially be "no", just by parity considerations. $c$ is always odd in a primitive Pythagorean triple. $\endgroup$
    – user44191
    Dec 29, 2019 at 22:27

2 Answers 2


So you are looking for solutions $$ \bigl( [a_1,b_1,c_1],[a_2,b_2,c_2],[a_3,b_3,c_3]\bigr) \in (\mathbb P^2)^3(\mathbb Q) $$ to the equations $$ a_1^2+b_1^2=c_1^2,\quad a_2^2+b_2^2=c_2^2,\quad a_3^2+b_3^2=c_3^2,\quad a_1b_1c_2^2c_3^2+a_2b_2c_1^2c_3^2+a_3b_3c_1^2c_2^2=0. $$ These equations define a surface $S$ in $(\mathbb P^2)^3$. In general, we don't have very many tools for determining the rational points on algebraic surfaces, but a first question, which you should be able to answer, is to determine the geometry of $S$. For example, it has a singularity at $([0,0,1],[0,0,1],[0,0,1])$. You should find the others, resolve the singularities, and determine the Kodaira dimension of $S$. That will at least suggest what the set of solutions might look like. (A program such as Magma might be able to do this computation for you.)

Another approach is to use some of the obvious symmetries to map to a simpler surface. For example, let $\pi:S\to S$ be the map that swaps $[a_1,b_1,c_1]$ and $[a_2,b_2,c_2]$, i.e., $\pi$ is the restriction to $S$ of the automorphism of $(\mathbb P^2)^3$ that swaps the first two factors. Compute the quotient variety $S/\pi$, see if you can determine its rational points, and then study which ones (if any) lift to $S$. Similarly for the automorphism that cyclically permutes the three factors of $(\mathbb P^2)^3$, and for the full $\mathcal S_3$ permutation group.

  • 2
    $\begingroup$ It may be easier to analyze the surface $\frac{(m_1^2 - n_1^2)m_1 n_1}{(m_1^2 + n_1^2)^2} + \frac{(m_2^2 - n_2^2)m_2 n_2}{(m_2^2 + n_2^2)^2} = \frac{(m_3^2 - n_3^2)m_3 n_3}{(m_3^2 + n_3)^2}$ as a subset of $(\mathbb{P}^1)^3$ (gotten using the usual parametrization of Pythagorean triples). $\endgroup$
    – user44191
    Dec 27, 2019 at 3:02
  • $\begingroup$ Thank you so much for pointing this out. I was a bit afraid of this and maybe I should have specified my background a bit more. I just got done with real analysis and am currently self studying abstract algebra so I'm afraid this is a bit out of my scope but I'll do my best. I'm going to try to pick up some algebraic geometry from the springer series in the next couple months or so. Hopefully I can pursue this route a bit further when I get there. Do you have any thoughts on the recursive nature of the solutions when c is not squared? It's pretty puzzling to me... $\endgroup$
    – PMaynard
    Dec 27, 2019 at 3:54
  • 4
    $\begingroup$ @PMaynard What better way to learn some algebraic/Diophantine geometry than to have an example like this in mind. There's a fantastic book called "Ideals, varieties, algorithms" by Cox, Little and O'Shea which might be a good place to start if you're interested in concrete examples like this (rather than certain other Springer books called, say, "Algebraic geometry"). $\endgroup$ Dec 27, 2019 at 6:17

Not really an answer, but a suggestion. Did you try to solve the "wrong" problem, where at the denominators you have $c_i$, not $c_i^2$? It seems quite interesting.

If we restrict ourselves to solutions of the type you found, then a Pythagorean triple with $b=a+1$ satisfies $a^2+(a+1)^2=c^2$, which writes as $(2a+1)^2-2c^2=-1$. The fundamental unit of ${\mathbb Z}[\sqrt 2]$ is $1+\sqrt 2$, of norm $-1$. Hence if $(x_n,y_n)$ are the solutions of the Pell equation $x^2-2y^2=\pm 1$, with $x_n+y_n\sqrt 2=(1+\sqrt 2)^n$, then $x_n^2-2y_n^2=(-1)^n$. So we have Pythagorean triplets $(a,a+1,c)$ if we take $2a+1=x_n$ and $c=y_n$, with $n$ odd. Such a triplet looks like $(\frac{x_n-1}2,\frac{x_n+1}2,y_n)$.

If we want a triplet with $c=b+1$ then we have the equation $a^2+b^2=(b+1)^2$ and we get $(a,b,c)=(a,\frac{a^2-1}2,\frac{a^2+1}2)$, with $a$ odd.

Then one can verify that $(a_1,b_1,c_1)=(\frac{x_{2k-1}-1}2,\frac{x_{2k-1}+1}2,y_{2k-1})$, $(a_2,b_2,c_2)=(\frac{x_{2k+1}-1}2,\frac{x_{2k+1}+1}2,y_{2k+1})$ and $(a_3,b_3,c_3)=(x_{2k},\frac{x_{2k}^2-1}2,\frac{x_{2k}^2+1}2)$, with $k\geq 2$, satisfy the equation $\frac{a_1b_1}{c_1}+\frac{a_2b_2}{c_2}=\frac{a_3b_3}{c_3}$. This checks straightforwardly if we write $x_n=\frac{\alpha^n+\beta^n}2$ and $y_n=\frac{\alpha^n-\beta^n}{2\sqrt 2}$, where $\alpha =1+\sqrt 2$ and $\beta =1-\sqrt 2=-\alpha^{-1}$.

A legitimate and nice problem you may think of is whether these are the only solutions or not. It also has some geometric meaning since $\frac{ab}c$ is the height perpendicular on $c$ in the triangle.

  • $\begingroup$ Very interesting. Unfortunately I haven't had time to study this so thank you for the insight. I want to believe that these are the only solutions because these are the only ones up to $10^7$, but then again there could always be more until shown otherwise. Honestly, there is just so much here to study so hopefully I can spend some more time on this as well. I was aware of the ratio being equivalent to the height. It is frustrating for $c^2$ because I can't apply any geometric meaning to that yet. What I'd also really like to know is if these solutions have any bearing on the $c^2$ case? $\endgroup$
    – PMaynard
    Dec 27, 2019 at 22:15
  • $\begingroup$ The pell equation above shows up in where this problem comes from. When I post the full background later this may help. So I'm wondering if maybe this form could also be a necessary condition for $c^2$ as well (although we still don't know if these are the only solutions). If this is the case it could make the problem easier to tackle. $\endgroup$
    – PMaynard
    Dec 27, 2019 at 22:27
  • $\begingroup$ I have updated the question to add some context that might be related to your findings. This is where the Pell equation shows up. $\endgroup$
    – PMaynard
    Dec 30, 2019 at 0:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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