MATH 2030: MATRICES Introduction to Linear Transformations We Have

MATH 2030: MATRICES

Introduction to Linear Transformations We have seen that we may describe matrices as symbol with simple algebraic properties like matrix multiplication, addition and scalar addition. In the particular case of matrix-vector multiplication, i.e., Ax = b where A is an m × n matrix and x, b are n×1 matrices (column vectors) we may represent this as a transformation on the space of column vectors, that is a function F (x) = b , where x is the independent variable and b the dependent variable. In this section we will give a more rigorous description of this idea and provide examples of such matrix transformations, which will lead to the idea of a linear transformation. To begin we look at a matrix-vector multiplication to give an idea of what sort of functions we are working with 1 0  1 A = 2 −1 , v = .   −1 3 4 then matrix-vector multiplication yields  1  Av =  3  −1 We have taken a 2 × 1 matrix and produced a 3 × 1 matrix. More generally for any x we may describe this transformation as a matrix equation y 1 0   x  x 2 −1 = 2x − y .   y   3 4 3x + 4y From this product we have found a formula describing how A transforms an arbi- 2 3 trary vector in R into a new vector in R . Expressing this as a transformation TA we have  x  x T = 2x − y . A y   3x + 4y From this example we can deﬁne some helpful terminology. A transformation1 T from Rn to Rm is a rule that assigns to each each vector v ∈ Rn a unique vector T (v) ∈ Rm. The domain of T is Rn and the codomain is Rm, and we write this as T : Rn → Rm . For a vector v in the domain of T, the vector in the codomain T (v) is called the image of v under T . The set of all possible images T (v) for all vinRn 2 is called the range of T. In the previous example the domain of TA is R and the

1or mapping or function 1 2 MATH 2030: MATRICES

 1  1 codomain is 3, so T : 2 → 3. The image of v = is w = T (v) = 3 . R A R R −1   −1 The image of TA consists of all vectors in the codmain of the form 1  0  x T = x 2 + y −1 A y     3 4 this describes an arbitrary linear combination of the column vectors of A. We conclude that the image consists of the column space of A. Geometrically we may see this as a plane in R3 through the origin with the column vectors of A as direction 3 2 vectors. Notice that TA(x) ⊂ R where x is any vector in R

Linear Transformations. The previous example TA is a special case of a more general type of transformation called a linear transformation. We provide a less rigorous deﬁnition, that summarizes the key ideas that the transformation ”respect” vector operations of addition and scalar multiplication.

Deﬁnition 0.1. A transformation T : Rm → Rm is called a linear transformation if (1) T (u + v) = T (u) + T (v) for all u and v in Rn. (2) T (cv) = cT (v) for all v in Rn and all scalars c. Example 0.2. Consider once again the transformation T : R2 → R3 deﬁned by  x  x T = 2x − y y   3x + 4y

x w we will show this is indeed a linear transformation. Deﬁne u = and v = y z then compute T (u + v),  x + w   x + w  x w x + w T + = T = 2(x + w) − 3(y + z) = (2x − 3y) + (2w − 3z) y z y + z     3(x + w) + 4(y + z) (3x + 4y) + (3w + 4z) Looking at the far-right hand side we may write this as  x   w  x w (2x − 3y) + (2w − 3z) = T + T = T (u) + T (v)     y z (3x + 4y) (3w + 4z) To show the second property, consider T (cv) for some scalar c:  cx   cx   x  x cx x T c = T = 2cx − cy = c(2x − y) = c 2x − y = c . y cy       y 3cx + 4cy c(3x + 4y) 3x + 4y the second property holds, this is indeed a linear transformation. Although the linear transformation T in the previous example arose as a matrix transformation TA, one may go backwards and recover the matrix A from the MATH 2030: MATRICES 3 deﬁnition of T given in the example. Notice that  x  1  0  1 0  x x T = 2x − y = x 2 + y −1 = 2 −1 y         y 3x + 4y 3 4 3 4 where this is just the matrix-vector multiplication of A with an arbitrary vector in the domain. In general a matrix transformation is equivalent to a linear transformation, according to the next theorem

Theorem 0.3. Let A be an m × n matrix. Then the matrix transformation TA : Rn → Rm deﬁned by n TA(x) = Ax, x ∈ R is a linear transformation.

Proof. Let u and v be vectors in the domain, and c a scalar, then TA(u + v) = Au + Av = TA(u) + TA(v) and TA(cv) = cAv = cTA(v). Thus TA is a linear transformation.

Example 0.4. Q: Let F : R2 → R2 be the transformation that sends each point to its reﬂection in the x-axis. Show that F is a linear transformation. A: This transformations send each point (x, y) to a new coordinate (x, −y), and so x x we may write F = To show this is linear notice that y −y x 1 0 1 0 x = x + y = −y 0 −1 0 −1 y Thus F x = Ax showing that this is a matrix transformation and hence a linear transformation by the previous theorem.

Example 0.5. Q:Let R : R2 → R2 be the transformation that rotates each point by an angle of π/4 (90 degrees) counterclockwise about the origin. Show that F is a linear transformation. A: Plotting this on the plane, we see that R takes any point (x, y) in the plane and sends it too (−y, x), and so as a transformation x −y 0 −1 0 −1 x R = = x + y = y x 1 0 1 0 y So R is described by a matrix transformation and therefore is a linear transformation. Recalling that if we multiply a matrix by standard basis vectors we ﬁnd the columns of the original matrix, we can use this fact to show that every linear transformation from Rn to Rm arises as a matrix transformation. Theorem 0.6. Let T: Rn → Rm be a linear transformation. Then T is a matrix transformation, and more speciﬁcally T = TA where A is the m × n matrix

A = [T (bfe1)|T (e2)| · · · |T (en)]. n Proof. Let e1, e2, ..., en be the standard basis vectors in R and let x be a vector n 1 n in R , so that x = x e1 + ... + x en. Noting that T (ei) for i = 1, ..., n are column 4 MATH 2030: MATRICES

m vectors in R , we denote A = [T (bfe1)|T (e2)| · · · |T (en)] be the m × n matrix with these vectors as its columns, then x1  1 n  .  T (x) = T (x e1 + ... + x en) = [T (bfe1)|T (e2)| · · · |T (en)]  .  = Ax. xn The matrix in the proof of the last theorem is called the standard matrix of the linear transformation T. Example 0.7. Q: Show that a rotation about the origin through an angle θ defines a 2 2 linear transformation from R to R and find its standard matrix. A: Let Rθ be the rotation, we will prove this geometrically. Let u and v be vectors in the plane, then the parallelogram rule determines the new vector u + v . If we now apply Rθ the parallelogram is rotated by an angle of θ and so the diagonal of the parallelogram defined by Rθ(u) + Rθ(v) Hence Rθ(u + v) = Rθ(u) + Rθ(v). Similarly if we apply a rotation to v and cv by a fixed angle of θ we find Rθ(v) and Rθ(cv), however as rotations do not affect lengths we must have Rθ(cv) = cRθ(v). We conclude that Rθ is a linear transformation, and we may apply the standard basis vectors of R2 to this transformation to determine its standard matrix. Using trigonometry we find that 1 cosθ R = . 0 sinθ Equivalently we find that the second standard basis vector is mapped to 0 −sinθ R = . 1 cosθ

Thus the standard matrix for Rθ will be cosθ −sinθ sinθ cosθ

Example 0.8. • Show that the transformation P:R2 → R2 that projects a point onto the x-axis is a linear transformation and ﬁnd its standard matrix. • More generally, if ` is a line through the origin in R2, show that the transfor- 2 2 mation P` : R → R that projects a point onto ` is a linear transformation and ﬁnd its standard matrix.

A: • P sends the point (x, y) to the point (x, 0) and so x x 1 0 1 0 x P = = x + y = y 0 0 0 0 0 y 1 0 Thus the transformation matrix for P is just . 0 0 • The line ` has direction vector d, then for any vector v, the transformation P` is given by projd(v) - the projection of v onto d, d · v proj (v) = d. d v · v MATH 2030: MATRICES 5

To show P` is linear consider the sum d · (u + bv) P (u + v) = d. ` v · v d · u + d · v = d. v · v d · u d · v = d + d. v · v v · v .

the last line is just P`(u) + P`(v). Similarly P`(cv) = cP`(v), proving that P` is indeed a linear transformation. d To determine its standard matrix, we denote d = 1 , the projection d2 onto the standard basis is just d1 d1 P`(e1) = 2 2 d1 + d2 d2 d2 d1 P`(e2) = 2 2 d1 + d2 d2 implying that the standard basis is of the form 2 1 d1 d1d2 A = 2 2 2 . d1 + d2 d1d2 d2 New Linear Transformations from Old. If T:Rm → Rn and S:Rn → Rp are linear transformations, then we may follow T by S to form the composition of the two transformations, denoted S ◦ T . Notice that in order for S ◦ T to make sense, the codomain of T and the domain of S must be the same, and the resulting transformation S ◦ T goes from Rm to Rp, that is it maps from the domain of T to the codomain of S. The formal definition of this new function is given as S ◦ T (v) = S(T (v)) We would like to have this new function be a linear transformation, which it is, and we may demonstrate this by showing that S ◦ T satisfies the definition of a linear transformation. We will do this by showing that it is a matrix transformation.

Theorem 0.9. Let T: Rm → Rn and S Rn → Rp be linear transformations. Then S ◦ T : Rm → Rp is a linear transformation. Moreover, their standard matrices are related by [S ◦ T ] = [S][T ]. Proof. Let [S] = A and [T ] = B, so that A is an m×n matrix and B a n×p matrix; if v is a vector in Rm we simply compute S ◦ T (v) = S(T (v)) = S(Bv) = A(Bv) = (AB)v Thus the effect of S ◦ T is to multiply vectors by AB, from which it follows im- mediately that S ◦ T is a matrix transformation and hence a linear transformation with the transformation rule [S ◦ T ] = [S][T ]. Example 0.10. Q:Consider the linear transformation T:R2 → R3 defined by  x  x T = 2x − y y   3x + 4y 6 MATH 2030: MATRICES and the linear transformation defined S R3 → R3 defined by     2y1 + y3 y1  3y2 − y3  S y2 =    y1 − y2  y3 y1 + y2 + y3 Find S ◦ T : R2 → R3. A: Calculating the matrices of each transformation and computing their product we find 2 0 1   5 4  1 0  0 3 −1 3 −7 [S ◦ T ] = [S][T ] =   2 −1 =   . 1 −1 0    −1 1    3 4   1 1 1 6 3 It follows that the corresponding transformation is then   5x1 + 4x2 x1 x1 3x1 − 7x2 (S ◦ T ) = [S ◦ T ] =   x2 x2 −x1 + x2  6x1 + 3x2 Example 0.11. Q: Find the standard matrix of the transformation that first rotates a point 90 degrees counterclockwise about the origin and then reflects the result in the x-axis. A: The rotation matrix [R] and reflection matrix [F ] were given in previous examples as 0 −1 1 0 [R] = , [F ] = 1 0 0 1 composing the two we find the desired transformation 0 −1 [R ◦ R] = [F ][R] = . −1 0 Inverse of Linear Transformations. Consider the effect of a 90 degree counterclockwise rotation about the origin followed by a 90 degree clockwise rotation about the origin. The cumulative effect of these two transformations is the identity transformation I, that is, no change at all (I(v) = v). If we denote R90 and R−90 for the respective transformations this means (R−90 ◦ R90(v) = v for any v in R2. Reversing the order geometrically gives the same result as well, i.e. R90 ◦ R−90(v) = v as well. Thus these two linear transformations are inverses of each other and we say that any two transformations related in this manner are called inverse transformations. Definition 0.12. Let S and T be linear transformations from Rn to Rn. Then S and T are inverse transformations if S ◦ T = In and T ◦ S = In. In terms of matrices, if S and T are inverse transformations then [S] = [T ]−1 since [S][T ] = [S ◦ T ] = I where the last matrix is the identity matrix.This show that [T ] and [S] are inverse matrices. Furthermore, if a linear transformation T is invertible, then its standard matrix [T ] must be invertible as well. As matrix inverses are unique, this means that the inverse of T is also unique, therefore we can use the notation T −1 to denote the unique inverse of each invertible linear transformation. MATH 2030: MATRICES 7

Theorem 0.13. Let T:Rn → Rn be an invertible linear transformation. Then its standard matrix [T ] is an invertible matrix and [T −1] = [T ]−1. Example 0.14. Q: Find the standard matrix of a 60 degree clockwise rotation about the origin in R2. A: Putting θ = π/3 in the sines and cosines in the matrix [Rθ] and using basic trig we ﬁnd that " √ # 1 − 3 √2 2 [R60] = 3 1 2 2

Using the fact that a 60 degree clockwise rotation is the inverse of R60 , and so we may ﬁnd that √ " 1 3 # −1 2√ 2 R−60] = [R60] = 3 1 − 2 2 by applying the last theorem. Example 0.15. Q:Determine whether projection onto the x-axis is an invertible transformation, and if it is, ﬁnd the inverse. A: We have seen that the standard matrix for this projection transformation P is 1 0 , this is not an invertible matrix as its determinant vanishes. We conclude 0 0 that P is not invertible as well.

References [1] D. Poole, Linear Algebra: A modern introduction - 3rd Edition, Brooks/Cole (2012).