Solving Problems Using Matrices
Date: 8/31/95 at 0:34:51 From: Anonymous Subject: rotations I am a data communications technician who programs in his spare time. My question is this. What is the formula to solve for x,y,z - given a current x,y,z coordinate (i.e. -60,-60,-60) and an angle (i.e. 90 degrees) and a plane to rotate about (i.e. z)? I am currently taking x,y and solving for its angle (with ATN) and radius, adding to the angle (i.e. 90 degrees), then using sine and cosine to re-get an x,y coordinate. (Does this make sense?) In my example I would get (60,-60,-60 x,y,z) using my method. But it's a tedious method. Why am I doing this? I am rotating objects in 3 dimensions. It's been years since I've taken math classes. I've been trying to find some good resources on the internet and referred to some old books I have. But no answer to my question. If my question is confusing, please contact me. -Thank you -Brian
Date: 9/22/95 at 15:5:9 From: Doctor Jeremy Subject: Re: rotations We ("Dr. Math") usually prefer to answer questions from K-12 students, as that is our mission. However, as a fellow programmer, I'll answer. Rotations of any kind can be accomplished by the use of matrices. I'm not sure if you know them, so I'll give a brief introduction to them in case you don't. If they are old hat to you, skip ahead. ------- BEGIN The dirt on matrices -------- Matrices are a kind of mathematical object. The theory of matrices is complicated (being essentially the subject of a semester-long college math class called "linear algebra"), but the practical deal, as far as rotation is concerned, is pretty simple. A matrix is usually represented as a rectangular array of values (usually numbers); you usually work with matrices of a particular size, such as 3x3 (first being rows, second being columns). Matrices can be multiplied; any mXn matrix can be multiplied by an nXp matrix to yield an mXp matrix. Note that this multiplication is not commutative; if A and B are matrices, AB is not necessarily the same as BA. One may even exist without the other existing. Matrices are multiplied like this: If A is a mXn matrix like a[1,1] a[1,2] ... a[1,n] a[2,1] a[2,2] ... a[2,n] . . a[m,1] a[m,2] ... a[m,n] (note the order: row, then column) and B is a nXp matrix, equivalently represented (with b's) Let C=AB; C is a mXp matrix, likewise with c's. So for any i,j, c[i,j] = a[i,1]*b[1,j] + a[i,2]*b[2,j] + ... + a[i,n]*b[n,j] You dig? If you have questions, write back with them. A vector (like the coordinates (-60,-60,-60) you gave) can be represented by a matrix; a 3-vector can either be a 3x1 matrix or a 1x3 one. When you multiply a matrix and a vector (of compatible sizes, on the appropriate side) you get another vector, which is the first vector after being transformed in various ways. --------------END The dirt on matrices -------- So, for a 3-vector, you could multiply it by a 3x3 matrix. This would allow you to rotate it, scale it, and do some other less useful operations. (In math-ese, you can do any linear transformation, but nothing that isn't a linear transformation) Unfortunately, rotation around any point other than the origin is not a linear transformation (it's an affine transformation). If you just want to rotate around the origin, then fine. If not... I've seen two ways to get more versatility. Both ways add two main capabilities - they let you rotate around arbitrary centers, and they let you translate (move) stuff. 1. You can work with 4-dimensional vectors and 4x4 matrices. What you do is you set the last entry of the vector to 1. What this does is that the last row (or column, depending on which side you multiply on) of the matrix multiplies with that 1, and essentially adds values to each component. (This provides translation; rotation around an arbitrary center is really rotation around the origin plus translation) 2. You can keep to sane old 3-d vectors and 3x3 matrices, and also keep a 3-d vector to add to the input vector. Both ways are basically equivalent, and should be chosen based on implementation ease. Since you're always using the same size vectors, you can just write the equations for the transformation, i.e. newx = oldx * a + oldy * b + oldz * c + movex newy = oldx * d + oldy * e + oldz * f + movey newz = oldx * g + oldy * h + oldz * i + movez Practically, the above is equivalent to the other methods. Its major drawback is that you don't get the real usefulness of matrices. Using various facts about matrices, you can make one transformation by combining several, or check a transformation to check to see if it is orthogonal (i.e. does not distort the object), etc. OK. So now you're asking, what kind of matrices can be used for rotation. Not all matrices work. So here are some examples: 1 0 0 1 -- take a 2d vector and don't change it. This is called the identity matrix of rank 2. 1 0 0 0 1 0 0 0 1 -- ditto for 3d vectors; it's the identity matrix of rank 3 2 0 0 2 -- don't rotate at all, but double the size of a 2d vector 0.707 -0.707 0.707 0.707 -- rotate 45 degrees around the origin, for a 2d vector (0.707 = sin(45) = cos(45)) 0.707 0.707 0 -0.707 0.707 0 0 0 1 ^^^-- rotate a 3d vector 45 degrees about the Z axis. Note that the 3rd row&column (the part corresponding with Z) is copied from the identity matrix of rank 3 (see above). If you wanted the X axis, you'd do the same for the 1st row and column. In general, to get the 2x2 matrix to rotate by an angle of 't' you do cos(t) sin(t) -sin(t) cos(t) (You can swap the signs on the sines to get the opposite rotation; I never bother to intuitively figure out which is which.) The 3d matrix is, to rotate around the X, Y or Z axes, you make a 3x3 matrix with the appropriate row & column copied from the identity matrix, and then take the remaining four entries from the 2x2 rotation matrix. (You want to put them in order of course; the key is make sure the cosine terms are all on a diagonal going from the upper left corner of the matrix ([1,1]) to the lower right corner ([3,3], or [n,n] in general). Hope I've helped... -Doctor Jeremy, The Geometry Forum
Search the Dr. Math Library:
Ask Dr. MathTM
© 1994- The Math Forum at NCTM. All rights reserved.