IV Image and Kernel of Linear Transformations Motivation: In the last class, we looked at the linearity of this function: F : R3 R2 F(x,y,z)=(x+y+z,2x-3y+4z) How does a 3 dimensional space get ‘mapped into’ a 2 dimensional space? At least one dimension ‘collapses’, or disappears. Image Let F : Rn Rm be a linear mapping. The image of the linear transformation F (Im F) is the set of image points in Rm. That is, Im F = u R m | there exists v R n such that F v u Examples non-linear example: f(x) = sin(x) the image is the interval (-1,1) Consider F: R2 R2 given by F(x,y)=(-y,x). (rotation by 90 degrees) Im F = entire plane, every point in R2. Choose an arbitrary point (a,b), then the point (b,-a) will map there. Consider F: R2 R2 given by F(x,y)=(x,x). This is the projection onto the line x=y. Im F = single line in R2 (the line x=y). Choose an arbitrary point (a,a) on the line, then any point (a,*) will map there. Consider F: R2 R2 given by F(x,y)=(0,0). This is a map to the zero vector. Im F = single point in R2 (the point (0,0)). Consider F: R2 R3 given by F(x,y)=(2x,2y,x-2y). Note that a two-dimensional space has been mapped into a 3 dimensional space. Theorem The image of a linear transformation T: Rn Rm is the span of the images of a set of basis vectors in Rn. In particular, if T(x)=Ax the image is the span of the column vectors of A. (Note the dimension of the span is equal to the rank of A). Proof: We need to prove that the span of the images of a set of basis vectors in Rn span the entire image of T. In other words that any uIm F is in their span. Suppose e1,e1,…,en form a standard basis for Rn and T(e1), T(e2) … ,T(en) denotes their images. Note that A is constructed as A= T (e1 ) T (e2 ) ... T (en ) , so the columns of A are the images of each basis vector. We choose a random vector uIm T. By definition of being in the image, there exists a vector u Rn such that T(v)=u. Since e1,e1,…,en form a standard basis, u=k1e1+k2e2+…+knen for some scalars ki. By linearity, T(v)= T(k1e1+k2e2+…+knen)= k1T(e1)+k2 T(e2)+…+kn T(en) and thus u=T(v) is in the span of T(e1), T(e2) … ,T(en). Interpretation: We need only look at the images of a set of basis vectors to determine the span of the image of a transformation. Each basis vector in Rn may generate a linearly independent image vector, so that the dimension of the image is always dimension of the preimage. Example F: R2 R3 given by F(x,y)=(2x,2y,x-2y). F(1,0)=(2,0,1). F(0,1)=(0,2,-2). These two vectors form a span for the image. The image is 2 dimensional, as expected. image is spanned by { (2,0,1),(0,2,-2)} after row reduction, image is spanned by { (1,0,.5),(0,1,-1)} and the image is 2D. Example (#10) F : R3 R2 F(x,y,z)=(x+y+z,2x-3y+4z) Let’s look at the map of the basis vectors: F(1,0,0)=(1,2) F(0,1,0)=(1,-3) F(0,0,1)=(1,4) The image = span { (1,2),(1,-3),(1,4)} After row reduction, the image = span ( (1,0),(0.1)) A basis of the image = (1,0),(0,1) and it has dimension 2. Analysis: The first two basis vectors map to linearly independent image vectors! By considering all linear combinations, we can reach every point in R2. But the third basis vector maps to a linearly dependent image vector, and thus overlays points on the same set. (For example F(2,2,0)=(4,-2) but also F(9,0,-5)=(4,-2).) Kernel Let F : Rn Rm be a linear mapping. The kernel of F, written Ker F, is the set of elements in Rn that map into the zero vector in Rm. In particular, if F(x)=Ax the kernel of F is the set of solutions to Ax=0. Example F : R3 R2 F(x,y,z)=(x+y+z,2x-3y+4z) The kernel will be the set of solutions to F(x,y,z)=0. Solve: (#10) F(x,y,z)=(x+y+z,2x-3y+4z) = (0,0,0) 1 1 1 0 1 0 7 5 0 ~ 2 3 4 0 0 1 2 5 0 x 0 7 y 0 s 2 7 The kernel is the set S= s and has dimension 1. 2 Note: The dimension of the image was 2, and the dimension of the kernel was 1, which adds up to 3. Theorem The image and the kernel of a linear transformation are subspaces and their dimensions sum to n. This is just another statement of the rank nullity theorem. In detail: A transformation can be described as Ax=T. Rank = num of linearly independent column vectors of A = dimension of the image (since the span of the columns of A is the image of T) nullity = num of columns that don’t have pivots = dimension of the kernel (since each column without a pivot generates a free variable in the solution to Ax=0)