0
$\begingroup$

Suppose I have a ray in 3d space with an origin at $(x_0, y_0, z_0)$ and direction $(a, b, c)$. This can be represented with the following equation

$\frac{x-x_0}{a} = \frac{y-y_0}{b} = \frac{z-z_0}{c}$

And suppose I have another function in 3d space,

$y=sin(x)+cos(z)$

Assuming the intersection points between the straight line and the surface, how do I find the point (or a close approximation) closest to $(x_0, y_0, z_0)$ that these two entities intersect?

My understanding is I can rework the ray into the following:

$x = \frac{a}{b}(y-y_0)+x_0$

$z = \frac{c}{b}(y-y_0)+z_0$

Then plug those equations for x and z into the sinusoidal function,

$y = sin(\frac{a}{b}(y-y_0)+x_0) + cos(\frac{c}{b}(y-y_0)+z_0)$

Rearrange it and set it equal to 0:

$f(y) = sin(\frac{a}{b}(y-y_0)+x_0) + cos(\frac{c}{b}(y-y_0)+z_0) - y $

And this new function $f(y)$ represents the distance of a point on the ray from the original sinusoidal function. So I'd want to find $y$ where $f(y) = 0$. Finally I could plug the discovered $y$ value back into the ray equations to get $x$ and $z$.

Am I on the right track, and how would I find $f(y)=0$?

$\endgroup$
1
$\begingroup$

In general (see remark below), the straight line intersects the surface in a finite number of points.

enter image description here

Fig. 1 : A case with 3 intersection points. Point $(x_0;y_0;z_0)$ is figured as a little circle.

You are right when you look for solutions to your equation

$$y = \sin\left(\frac{a}{b}(y-y_0)+x_0\right) + \cos\left(\frac{c}{b}(y-y_0)+z_0\right)$$

But solutions $y_k $ of (1) w haven't to be plugged into parametric equations $x=... ; z=...$

It suffices

1) To consider them as abscissas on the straight line (variable $y$ can be thought as a sort of time variable). Then select, among all $y_k$s, the smallest one say $y_{min}$ in absolute value (avoiding in this way any real distance computation with formulas such as $\sqrt{(...)^2+(...)^2+(...)^2}$).

2) Then to plug this unique value of $y_{min}$ into the parametric equations in order to get $x_{min}$ and $z_{min}$.

Remark : In the exceptional case where the straight line is parallel to the $x-z$ plane :

  • either there is an infinite number of solutions (but the previous method can still be applied)

  • or there is no solution.

$\endgroup$
  • $\begingroup$ Thank you for your detailed answer! So if I understand correctly, to find $y_min$ I test the combined equation to find a f($y_k$)=0 closest to 0 by basically trying successively larger values of $y_k$ (and smaller values $y_k$ < 0). At least, that would be a brute force way of doing it. I recently read about Newton's Method and have been experimenting with it. I'm wondering if that is the most appropriate way to efficiently find a solution to this particular type of problem? $\endgroup$ – JasonCG Jun 5 at 1:15
  • 1
    $\begingroup$ The issue is that for a transcendental equation of this type : $y=\sin\left(\alpha y+\beta\right) +\cos\left(\gamma y+\delta\right)$, you don't know "a priori" how many roots you have. I would not advise you to use Newton's method (which is excellent when you are already close to a given root and you want to "refine" it and get quickly say an error $<10^{-10}$ ). Here, I would recommand the following strategy : as the RHS of the equation never takes values outside $[-2;2]$, it means that the roots are all in this interval, therefore if you want say an error $< 10^{-3}$, divide your (ctd...) $\endgroup$ – Jean Marie Jun 5 at 14:24
  • 1
    $\begingroup$ (...ctd) interval $[-2,2]$ into 4000 intervals $[a,b]$ of width $10^{-3}$ and test for each of them whether g(a) has the same sign or not as g(b) where $g(y):=(\sin\left(\alpha y+\beta\right) +\cos\left(\gamma y+\delta\right)) - y$. If it has not the same sign, it means that there is (at least) a root between $a$ and $b$. In this way you are sure not to miss any root, even if there are many of them. $\endgroup$ – Jean Marie Jun 5 at 14:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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