14
$\begingroup$

Suppose I have two points on a unit sphere whose spherical coordinates are $(\theta_1, \varphi_1)$ and $(\theta_2, \varphi_2)$. What is the great arc distance between these two points?

I found something from Wiki here but it is written in terms of latitude and longitude instead. I wonder latitude is $\pi/2 - \theta$ and longitude is $\varphi$. Am I right?

spherical coordinates

| cite | improve this question | |
$\endgroup$
  • 3
    $\begingroup$ That is correct. $\endgroup$ – Robert Israel Nov 6 '12 at 6:34
  • $\begingroup$ @RobertIsrael So I can just plug these into the equation on Wiki and compute the great arc distance, right? $\endgroup$ – Patrick Li Nov 6 '12 at 6:36
  • $\begingroup$ latitude $\theta$ is measured from equator, not from the north pole as in your figure $\endgroup$ – Emanuele Paolini May 8 '13 at 13:44
16
$\begingroup$

Consider the two vectors to the points on the sphere, $${\bf v}_i=(r\sin\theta_i\cos\varphi_i,r\sin\theta_i\sin\varphi_i,r\cos\theta_i)$$ with $i=1,2$. Use the dot product to get the angle $\psi$ between them: $${\bf v}_1\cdot {\bf v}_2=r^2\left(\cos\theta_1\cos\theta_2+\sin\theta_1\sin\theta_2\cos\left(\varphi_1-\varphi_2\right)\right)=r^2\cos\psi.$$

Then the arclength is $$s=r\psi=r\cos^{-1}\left(\cos\theta_1\cos\theta_2+\sin\theta_1\sin\theta_2\cos\left(\varphi_1-\varphi_2\right)\right).$$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ how exactly you got the dot product ? $\endgroup$ – curious Feb 28 '14 at 17:25
  • $\begingroup$ What you get is not the formula from here: en.wikipedia.org/wiki/Great-circle_distance#Formulas . I can't understand how the differences in cos are derived since you compute the inner product $\endgroup$ – curious Feb 28 '14 at 17:47
  • $\begingroup$ @curious the Wikipedia coordinates are defined differently than those in the question posted here. $\endgroup$ – Jonathan Feb 28 '14 at 19:47
  • 2
    $\begingroup$ @curious $\sin^2\alpha+\cos^2\alpha=1$ and $\cos\alpha\cos\beta+\sin\alpha\sin\beta=\cos(\alpha-\beta)$. $\endgroup$ – Jonathan Mar 1 '14 at 17:15
  • 1
    $\begingroup$ I'm a bit confused, are \varphi_1 and \varphi_2 angles? If I use 0 for both \theta_1 and \theta_2, I get a 0 degree difference no matter what \varphi_1 and \varphi_2 are. That doesn't seem right to me. $\endgroup$ – B T Mar 8 '16 at 2:42
1
$\begingroup$

I compare two formulas in Matlab, in terms of precision and running time.

Consider the sphere, in Cartesian coordinates:

$x^2 + y^2 + z^2 = \rho^2$

or in spherical coordinates:

$x = \rho \sin(\theta) \cos(\varphi)\\ y = \rho \sin(\theta) \sin(\varphi)\\ z = \rho \cos(\theta)$

Choose a bunch of points on the sphere, for example:

N = 1e6 + 1;
rho_sph = rand;
phi_sph = rand(N, 1) * pi;
theta_sph = rand(N, 1) * pi;

xyz_sph = [rho_sph * sin(theta_sph) .* cos(phi_sph), ...
           rho_sph * sin(theta_sph) .* sin(phi_sph), ...
           rho_sph * cos(theta_sph)];

Now calculate the distance between each couple of points (1,2), (2,3), ..., (N-1, N).

1º formula

tic
aa = cross(xyz_sph(1:end-1, :), xyz_sph(2:end, :));
bb = dot(xyz_sph(1:end-1, :)', xyz_sph(2:end, :)');
aa = arrayfun(@(n) norm(aa(n, :)), 1:size(aa, 1));
d1 = rho_sph * atan2(aa, bb)';
toc
% Elapsed time is 12.376257 seconds.

2º formula

tic
d2 = rho_sph * acos(cos(theta_sph(1:end-1)) .* cos(theta_sph(2:end)) + ...
     sin(theta_sph(1:end-1)) .* sin(theta_sph(2:end)) .* ...
     cos(phi_sph(1:end-1) - phi_sph(2:end)));
toc
% Elapsed time is 0.264629 seconds.

The first formula is considered more precise, but it's slower; the second is less accurate at the poles, but it's much faster. Otherwise the two solutions are practically equivalent:

plot(d1 - d2, '.')
| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Does the expense of the first formula decrease if you store xyz_sph(2:end, :) as an intermediate value instead of computing it twice? $\endgroup$ – Brandon Rhodes Jun 16 '18 at 17:59
0
$\begingroup$

well, i see Wikipedia says there is serious rounding error in the formula of top answer. i found on google a new formula with the derivation. It seems slightly complicated but it is correct one. you can calculate values of distance & compare with percentage errors in both the formula.
i hope it will help you

| cite | improve this answer | |
$\endgroup$

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.