Wierd problem with sin/cos.

Wierd problem with sin/cos.

Sebastian Mineur

Member #6,943

February 2006

Hey, guys!
I just recently started working on a project involving a lot of angles and vector math, but I seem to have hit a pretty nasty bump in the road...

It has to do with the standard 'sin' and 'cos' functions in 'math.h', and that they sometimes return really funky values.

Look here, this is a short program I've put together that reproduces the funkiness:

1#include <iostream>
2#include <math.h>
3 
4using namespace std;
5 
6int main() {
7    const float PI = 3.14159265f;
8 
9    cout << "acos(1.0)  = " << acos(1.0f) << "\n";
10    cout << "acos(0.0)  = " << acos(0.0f) << "\n";
11    cout << "acos(-1.0) = " << acos(-1.0f) << "\n";
12    cout << "\n";
13    cout << "cos(0.0)   = " << cos(0.0f) << "\n";
14    cout << "cos(PI)    = " << cos(PI) << "\n";
15    cout << "cos(-PI)   = " << cos(-PI) << "\n";
16    cout << "cos(PI/2)  = " << cos(PI/2.0f) << "\n";
17    cout << "cos(-PI/2) = " << cos(-PI/2.0f) << "\n";
18    cout << "\n";
19    cout << "sin(0.0)   = " << sin(0.0f) << "\n";
20    cout << "sin(PI)    = " << sin(PI) << "\n";
21    cout << "sin(-PI)   = " << sin(-PI) << "\n";
22    cout << "sin(PI/2)  = " << sin(PI/2.0f) << "\n";
23    cout << "sin(-PI/2) = " << sin(-PI/2.0f) << "\n";
24    cout << "\n";
25 
26    system("PAUSE");
27    return EXIT_SUCCESS;
28}

And when compiled, on my machine, it gives:

Quote:

acos(1.0) = 0
acos(0.0) = 1.5708
acos(-1.0) = 3.14159

cos(0.0) = 1
cos(PI) = -1
cos(-PI) = -1
cos(PI/2) = -4.37114e-08
cos(-PI/2) = -4.37114e-08

sin(0.0) = 0
sin(PI) = -8.74228e-08
sin(-PI) = 8.74228e-08
sin(PI/2) = 1
sin(-PI/2) = -1

You can see here that 'acos' works perfectly, but some of the return values from 'sin' and 'cos' look, outright, wrong!
More specifically, cos(PI/2) and cos(-PI/2) should return 0.0, right? And cos(-PI) should return positive 1.0, not negative.
Same thing with 'sin', 'cause sin(PI) and sin(-PI) should also return 0.0, shouldn't they?

Hmmm... I just now realised that even THAT's not right, because typically 0.0 degrees is straight to the right, and 'cos' gives the y-axis and 'sin' the x-axis. If I'm not mistaken.

Oh, well. I'm at a loss, here. Any help is greatly appreciated!

kdevil

Member #1,075

March 2001

The outputs look fine to me, but you have it backwards: cos gives the x-axis and sin gives the y axis.

cos(PI/2) and sin(PI) should return 0, but there's probably some inaccuracy due to floats not being able to exactly represent pi/2 (and also because your PI is just 9 digits long - try it again using Allegro's AL_PI or M_PI from math.h). So, because of that, you get something very close to 0, but not quite.

-----
"I am the Black Mage! I casts the spells that makes the peoples fall down!"

Sebastian Mineur

Member #6,943

February 2006

Ah, thanks.
I get the same results with M_PI, but what you're saying then is basically that -4.37114e-08 is actually just a really low number, like -0.0000000437... or something?

Goalie Ca

Member #2,579

July 2002

Remember that $<math>\pi = -\pi + 2\pi</math>$

And those other non-zero but approximately zero numbers are sadly just rounding errors. Software probably uses a modified taylor series to calculate it but obviously it doesn't calculate all infinite terms.

-------------
Bah weep granah weep nini bong!

Johan Halmén

Member #1,550

September 2001

acos(-1.0) = 3.14159

...is more off the right value than...

cos(PI/2) = -4.37114e-08

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Years of thorough research have revealed that the red "x" that closes a window, really isn't red, but white on red background.

Years of thorough research have revealed that what people find beautiful about the Mandelbrot set is not the set itself, but all the rest.

Sebastian Mineur

Member #6,943

February 2006

Mmm, then it was all just a matter of inaccuracy. It should still be accurate enough though, for the purposes of my project.

Goalie Ca said:

Remember that pi = -pi + 2pi

Yes, it looks obvious enough, but I guess I actually didn't really think about that. I'm not used to working with radians, and I was still thinking of it like degrees. Starting at 0 and going around clockwise to 360.
But, radians then, start at pi going counter-clockwise, full circle, to -pi.

That IS an important thing to remember.

Thanks for the help

EDIT:
Or... Maybe it's not counter-clockwise, I don't know. My point is that to make a full circle with radians, you would start at PI, straight to the left, go around to 0, straight to the right, and then on to -PI.

Goalie Ca

Member #2,579

July 2002

I think you should look at this
http://en.wikipedia.org/wiki/Unit_circle

-------------
Bah weep granah weep nini bong!

Matt Smith

Member #783

November 2000

..but bear in mind that mathematicians have y going up the page, while us computer people have it going down the screen.

Arthur Kalliokoski

Second in Command

February 2005

Quote:

mathematicians have y going up the page, while us computer people have it going down the screen.

To clarify, programmers were used to starting text at the upper right corner of the screen (as is normal) and continuing across, then going to the next line down. When they started fiddling with bitmap graphics, they continued the practice because they were used to it. Mathmaticians usually refer to the 1st quadrant as it has x and y positive, so y=0 is at the bottom.

[EDIT] As pointed out below, start at upper left Doh!

They all watch too much MSNBC... they get ideas.

Ron Novy

Member #6,982

March 2006

Quote:

Isn't that the upper left corner? ::)

----
Oh... Bieber! I thought everyone was chanting Beaver... Now it doesn't make any sense at all. :-/

Karadoc ~~

Member #2,749

September 2002

Quote:

Ah, thanks.
I get the same results with M_PI, but what you're saying then is basically that -4.37114e-08 is actually just a really low number, like -0.0000000437... or something?

Yes. That's exactly what -4.37114e-08 is. The "e-08" notation thing at the end means "times ten to the power of negative 8. So it is exactly what you said.

-----------

Matt Smith

Member #783

November 2000

-4.37114×10⁻⁸

Unicode FTW ✌☺✌

Slartibartfast

Member #8,789

June 2007

Quote:

My point is that to make a full circle with radians, you would start at PI, straight to the left, go around to 0, straight to the right, and then on to -PI.

1) When using radians, x = x + 2*Pi*k. So -Pi is equal to Pi which is equal to 3*Pi.
2) You can start wherever the hell you like (remember, a circle has no start nor an end), however for mathematical reasons it is usually more comfortable to use the range [-Pi,Pi].
3) The way you describe this doesn't make sense to me.

----
"No man who has at least one good hand has to rape someone if he's horny and can't get any." - Neil Black
"If God truly would send you to hell for trusting your senses and logic, I'd rather spend eternity alongside likeminded people than with a bunch of ignorant fools." - mEmO
"I'm going gay again" - weapon_S