Here
http://mathworld.wolfram.com/DiophantineEquation2ndPowers.html
it is claimed that a solution of the diophantine equation $$ax^2+bxy+cy^2=1\ \ \ \ (1)$$ is always a convergent of a root of $ax^2+bx+c$.
I found out that a solution of $(1)$ can exist even if $ax^2+bx+c$ has no real root, but in the example $$3x^2+9xy+7y^2=1$$ , a solution is $(3/-2)$ and $-\frac{3}{2}$ is the real part of both roots of $3x^2+9x+7$.
Is this a coincidence ?
Moreover, if the discriminant is $0$, for example $$4x^2+20xy+25y^2=1$$ has solution $(8/-3)$ , but $4x^2+20x+25=0$ has solution $-\frac{5}{2}$ , which does not have convergent $-\frac{8}{3}$.
And the case $a=0$ seems to work only if $b>2$, but this case is not very interesting anyway.
My work so far to prove the claim :
If $(s/t)$ is a solution of $(1)$ with $t\ne 0$ and $\alpha$ is a root of $ax^2+bx+c$, then we have
$$a(\frac{s}{t})^2+b\frac{s}{t}+c=\frac{1}{t^2}$$
and $$a\alpha^2+b\alpha+c=0$$
This implies $$a(\frac{s}{t}-\alpha)(\frac{s}{t}+\alpha)+b(\frac{s}{t}-\alpha)=\frac{1}{t^2}$$ implying $$\frac{s}{t}-\alpha=\frac{1}{t^2(a(\frac{s}{t}+\alpha)+b)}$$
This looks promising because we only have to estimate $a(\frac{s}{t}+\alpha)+b$.
Who has an idea how to complete the proof and find out the conditions when the solution of $(1)$ must be a convergent of one of the roots ?

