Vector rotation

Unfortunately, the rotation from a to a' isn't unique, meaning there are any number of different rotation matrices that could perform that transformation. For instance, the rotation axis could have been parallel to a x a', or it could have been parallel to a + a'. Perhaps you could provide us with context to help you chose the best rotation.

There's probably a "right" way to do this, but I'll give it a go (despite this sounding like homework).

To find a matrix to rotate (x1,y1,z1) to (x1',y1',z1'), I'd do something like this:

1. Find (xr,yr,zr), the axis of rotation, by taking the cross product of (x1,y1,z1) and (x1',y1',z1').


2. Find the angle you need to rotate by taking the dot product of the normalized (1-length) original vectors (x1,y1,z1) and (x1',y1',z1'), then taking the acos of the result.


3. Google up "rotation arbitrary axis" or something to figure out how to make a matrix out of the information from step 1 and step 2.


4. Multiply (x2,y2,z2) by the resulting matrix.


5. Profit!

As always, the actual mathematicians are free to correct me [smile]

As Zipster said, the axis of rotation can be the median axis of any single cone which contains the start and end vectors — so any vector perpendicular to v' - v fits, which constitutes an infinite set.

To determine a rotation, you need at least two independent pairs of transformed vectors: a given pair determines a plane of possible axes, so two independent pairs intersect their planes and yield an axis.

This is not a homework. I am trying to build a pipe around a curve, the normals at curve points need to match so i can create circles around it and connect them with triangle strips.

BeanDog, your methods seems like it will work. I am not very good with linear algebra, so i'll try to implement this and let you know.

Thanks for replies