I'd like to prove or disprove the following:
given a compact arc of a conic section $\mathcal{T} \subset \mathbb{R}^2$ such that for any $P \neq Q \in \mathcal{T}$ the tangent vectors $\tau_P, \tau_Q$ aren't parallel, then i can parametrize it using a quadratic Rational Bézier curve.
First of all I can find the three points: the endpoints of the arc and the intersection of the tangent lines at the endpoints (here i use the hypotesis making this intersection nonempty, as a counterexample we can pick a semicircumference). Then I can, without loss of generality, normalize the weights $w_0=w_2=1$. Now, how can I find $w_1$? I searched on NURBS,Farin, online and even on Stack Exchange. I didn't find a general case (only the one with circular arcs) and moreover it seems that you can only approximate an arc of conic section (why?).
Edit1: I found these slides https://public.vrac.iastate.edu/~oliver/courses/me625/week11.pdf but I don't understand nor the direct approach (because who grants me that of the three equations at least two are linearly independent) nor the geometric one (that one I just don't get it at all).
We can do this from first principals. From the given conic arc, we can compute the Bézier control points $P_0, P_1, P_2$. And we know we can assume that the weights for the first and last control points are both $1$. So, then, the equation of the Bézier curve is: $$ C(t) = \frac {(1-t)^2P_0 + 2t(1-t)wP_1 + t^2P_2} {(1-t)^2+ 2t(1-t)w+ t^2} $$ We just need to find the middle weight, $w$.
Let $M = \tfrac12(P_0 + P_2)$ be the mid-point of the chord, and let $Q$ be the point where the conic intersects the line segment $MP_1$. After computing $Q$, we define $k = |MQ|/|MP_1|$. Then $$ Q = (1-k)M + kP_1 = \tfrac12(1-k)P_0 + kP_1 + \tfrac12(1-k)P_2. $$ Our strategy is to choose $w$ so that $C(\tfrac12) = Q$. Then the Bézier curve will pass through the points $P_1, Q, P_2$ and it will have the same end tangents as the original conic. A conic is fully determined by three points and two tangents, so if we accept that our rational quadratic curve is a conic, then it must be the same conic as the original one.
But $C(\tfrac12) = Q$ gives us $$ \frac {\tfrac14P_0 + \tfrac12wP_1 + \tfrac14P_2} {\tfrac14 + \tfrac12w+ \tfrac14} = \tfrac12(1-k)P_0 + kP_1 + \tfrac12(1-k)P_2. $$ Equating coefficients of $P_1$. We get $$ \frac{w}{1+w}= k $$ And rearranging gives $$ w = \frac{k}{1-k} $$