CHAPTER 2

LINEAR TRANSFORMATIONS
1. Dening linear transformations Linear transformations are (mathematical abstractions of) very common types of function. Exercise 1 (Rotations in the plane). Consider the function which, given a vector v in the plane, produces as output the same vector rotated (anti-clockwise) through an angle : we write R v for this new vector. What is R (v + w), in terms of R v and R w ? What is R (v ) in terms of R v ? Exercise 2 (The simple harmonic oscillator from quantum mechanics). The simple harmonic oscillator is the quantum mechanical version of a mass oscillating in simple harmonic motion on a spring. The operator H is a function from functions to functions. Given a function f : R R, we dene the new function Hf : R to R by the rule d2 (Hf )(x) = 2 f (x) + x2 f (x). dx 2 2 For instance, if f (x) = ex /2 , then Hf (x) = ex /2 . What is H(f + g ) in terms of Hf and Hg ? What is H(f ) in terms of Hf ? These two examples enjoy the same basic algebraic propertiesthey respect the basic vector operations of addition and scalar multiplication. Definition 2. Let V and W be vector spaces. A linear transformation (or mapping or map) from V to W is a function T : V W such that T (v + w ) = T v + T w T (v ) = T (v ) for all vectors v and w and scalars . The aim of our study of linear transformations is two-fold: to understand linear transformations in R, R2 and R3 . to bring this understanding to bear on more complex examples. Let us try to nd the linear maps from R to R. Which of the following functions are linear? f (x) = 3x + 1 h(x) = ex + 2 g (x) = x k (x) = sin(x).

In general, the linear functions from R to R are of the form l(x) = cx, for some xed c R (including 0).
25

26

2. LINEAR TRANSFORMATIONS

2. Linear Maps and Matrices Suppose that A is an m n matrix. Dene TA : Rm Rn by the formula TA x = Ax for all x Rm . Then TA is a linear map. Indeed, TA (x + y ) = A(x + y ) = Ax + Ay = TA (x) + TA (y ), and TA (x) = TA (x) similarly. In some senses, matrices are the only examples of linear maps. Theorem 15. Suppose that V and W are vector spaces with bases {v 1 , . . . , v m } = A and {w 1 , . . . , wn } = B respectively, and suppose that T : V W is a linear map. Let ai denote the vector T vi , [ai ]B denote its coordinates relative to the basis B, and A denote the matrix with columns [ai ]B . Then for all x in V , [T x]B = A[x]A . Proof. For x in V , x = x1 v 1 + x2 v 2 + + xm v m , where (x1 , x2 , . . . , xm )T = [x]A . Then T x = T (x1 v 1 + x2 v 2 + + xm v m ) = x1 T v1 + x2 T v 2 + + xm T v m , [T x]B = x1 [T v 1 ]B + x2 [T v2 ]B + + xm [T v m ]B = A(x1 , x2 , . . . , xm )T , as required. Example. Let R : R2 R2 be the rotation (anti-clockwise) through the angle . Then 1 cos = R 0 sin R 1 0 = sin , cos cos sin sin cos x . y so

in the standard basis for R2 . Then x 1 0 = xR + yR R y 0 1

Example. Let Ds : R3 R3 be the dilation by a factor of s in R+ . The Ds x = sx for all x in R3 . Thus s 0 0 Ds x = 0 s 0 x. 0 0 s Exercise 3. What is the geometric eect of multiplying by the matrix 1 0 0 0 1 0 0 0 103

3. KERNELS AND RANGES

27

in R3 ? Exercise 4. What is the geometrical eect of multiplying by the matrix R2 ? Exercise 5. Let a be a unit vector in R3 . Dene the map L : R3 R3 by the formula Lx = (x a)a. (a) Show that L is linear. (b) Express L as multiplication by a matrix. (c) Describe L geometrically. Exercise 6. Let a be a unit vector in R3 . Dene the map X : R3 R3 by the formula X v = v a. (a) Show that X is linear. (b) Express X as multiplication by a matrix. (c) Describe X geometrically. 3. Kernels and Ranges Consider the linear system a11 x1 + a12 x2 + + a1m xn = b1 ... am1 x1 + am2 x2 + + amn xn = bm , or equivalently, in matrix form A x = b. What b can we solve this for? Is the solution unique? We have seen that we can solve the equation if and only if b is a linear combination of the columns of A; we write b col(A) for short. We also know that, if xpart is a particular solution of Ax = b, then every solution is of the form xpart + xhom , where xhom is a solution of the homogeneous equation Ax = 0. If the homogeneous equation has one solution, so does Ax = b; if the homogeneous equation has many solutions, so does Ax = b. Next, consider the dierential equation d2 u Hu(x) = 2 (x) + x2 u(x) = b(x), dx where b is a known function and u is unknown. What b can we solve this for? Is the solution unique? We can answer the rst question in a formal sense: b must lie in the range of H, also known as the image of H, and written range(H) or image(H). For the second question, 1 1 0 1 in

28

2. LINEAR TRANSFORMATIONS

we can show that, if upart is a particular solution of Hu = b, and uhom is any solution of the homogeneous equation Hu = 0, then H(upart + uhom ) = b + 0 = b, i.e., upart + uhom is a solution of Hu = b. In fact, every solution of Hu = b arises in this way. Thus the uniqueness or nonuniqueness of the solutions of the homogeneous equation controls the uniqueness or nonuniqueness of the original equation. We unify these (and other examples) in the following observations. Let T : V W be a linear map. The set of vectors T V = {T v : v V } is called the range or image of T , and written range(T ) or image(T ). Then range(T ) is the collection of vectors w in W for which the equation T x = w can be solved. Let xpart be a particular solution of the equation T x = w , and let xhom be any solution of the homogeneous equation T xhom = 0. Then T (xpart + xhom ) = T (xpart ) + T (xhom ) = w + 0 = w , so that xpart + xhom is a solution of T x = w . Further, every solution of T x = w is of this form. Indeed, if T x = w and T xpart = w , then T (x xpart ) = T (x) T (xpart ) = w w = 0, so x xpart is a solution of the homogeneous equation. Given a linear map T : V W , we dene the kernel of T , written ker(T ), to be the subspace of V consisting of all solutions of the homogeneous equation T x = 0; in symbols ker(T ) = {x V : T x = 0}. 4. Rank and nullity We dene the nullity of T , written nullity(T ), to be the dimension of ker(T ). This is equal to the number of parameters in the solution of T x = w. Example. The general solution of d2 u(x) u(x) = x dx2 is u(x) = x + A sin x + B cos x. This has two parameters, because the nullity of the dierential operator T : T f (x) = f (x) f (x), is 2. Suppose that V and W are vector spaces, and that T : V W is a linear map. The dimension of T (V ) (also known as image(T ) or range(T )) is called the rank of T . Sometimes the number dim(W ) dim(T (V )) is called the co-rank of T . Obviously the rank determines the co-rank, and vice versa. Proposition 16. Suppose that T : V W is a linear map. Then (i) T is one-to-one if and only if nullity(T ) = 0. (ii) T is onto if and only if rank(T ) = dim(W ), i.e., if and only if co-rank(T ) = 0.

4. RANK AND NULLITY

29

Theorem 17. If U is a subspace of W , then the smallest number of equations needed to describe U is dim(W ) dim(U ). Example. Consider the line with parametric equation x = d, where d = (2, 0, 1)T . This line is a one-dimensional subspace. It may also be described by the equations y = 0 and x = 2z . It is of codimension 2. Challenge Problem. Find equations which dene span (1, 2, 0, 1)T , (1, 0, 2, 0)T in R4 . What is the minimal number of equations? We conclude that the co-rank of a linear transformation T : Rm Rn is the number of equations needed to describe the image of T . Theorem 18 (The rank-nullity theorem for matrices). Suppose that A Mm,n . Then rank(A) + nullity(A) = m. Proof. Suppose that A is row-reduced to row-echelon form. Then the columns of A corresponding to the leading columns of the reduced matrix form a basis for range(A), hence rank(A) is equal to the number of leading columns. The nonleading columns of the reduced matrix correspond to the parameters of the solution, i.e., nullity(A) is equal to the number of nonleading columns. These numbers add to give the total number of columns, i.e., m. Corollary 19. If A is a square matrix, then nullity(A) = co-rank(A). Consequently, the linear transformations associated to square matrices are one-to-one if and only if they are onto. Example. Consider D : Pn Pn1 , given by D (an tn + + a0 ) = nan tn1 + + a1 (i.e., D corresponds to dierentiation). By choosing the coecients an , an1 , . . . , a1 correctly, we can arrange that the right hand side is any polynomial of degree n 1. Thus range(D ) = Pn1 , and hence rank(D ) = dim(Pn1 ) = n. Also, the kernel of D is the set of all constant polynomials: indeed, D (an tn + + a0 ) = 0 if and only if an = an1 = = a1 = 0, and the polynomial is constant. Then nullity(D ) = 1. Finally, rank(D ) + nullity(D ) = n + 1 = dim(Pn ). Exercise 7. Find the nullity of the 1 1 1 2 3 4 0 4 2 . 1 0 0 The reduced matrix is of the form . matrix

Answer. Reduce to row-echelon form. 1 0 0 0

Then the rank of the matrix is 3 and the nullity is 1. Note that we do not have to nd ker(T ) explicitly to show that it is one-dimensional.

30

2. LINEAR TRANSFORMATIONS

Exercise 8. Find a basis for the kernel 1 2 1 0 1 1

of the matrix 3 4 4 2 . 0 0

Answer. We need to nd x1 , . . . , x4 such that x1 1 2 3 4 1 0 4 2 x2 = 0, x3 1 1 0 0 x4 i.e., to nd the solutions of the system represented by the augmented matrix 1 2 3 4 0 1 0 4 2 0 . 1 1 0 0 0 Row-reduced, this is of the form 1 0 0 0 0 0 . 0

The solution space has the parametric equation 10 10 x = 2 , 9 and is 1-dimensional. Then {(10, 10, 2, 9)T } is a basis for the kernel. Example. Find a basis for the image of 1 2 1 0 1 1 Answer. We row-reduce this matrix, 1 1 0 1 0 0 the matrix 3 4 4 2 . 0 0

and get

0 0 4 2 . 9 2

Thus the rst three columns are linearly independent, and the fourth column depends linearly on these. Hence the vectors (1, 1, 1)T , (2, 0, 1)T and (3, 4, 0)T are linearly independent; since R3 is 3-dimensional, they must form a basis. Of course, other sets of three of these vectors, such as {(1, 0, 0)T , (0, 1, 0)T , (0, 0, 1)T }, are also bases for R3 .

4. RANK AND NULLITY

31

Lemma 20. Suppose that V and W are vector spaces and T : V W is a function. Then T is a linear transformation if and only if T (u + v ) = T (u) + T (v) for all u, v V and all scalars , . Proof. If T is linear, then T (u + v) = T (u) + T (v ) = T (u) + T (v ). Conversely, if (2.1) holds, then taking = = 1 shows that T (u + v ) = T (u ) + T (v ) and taking = 0 shows that T (u) = T (u), so T is linear. Exercise 9. Dene T : P3 (R) R4 by T (a3 x3 + a2 x2 + a1 x + a0 ) = (a0 , a1 , a2 , a3 )T . Show that T is a linear mapping. Answer. Suppose that p(x) = a3 x3 + a2 x2 + a1 x + a0 q (x) = b3 x3 + b2 x2 + b1 x + b0 . Then T (p(x) + q (x)) = T ((a3 + b3 )x2 + (a2 + b2 )x2 + (a1 + b1 )x + (a0 + b0 )) = (c0 + b0 , a1 + b1 , a2 + b2 , a3 + b3 )T = (a0 , a1 , a2 , a3 )T + (b0 , b1 , b2 , b3 )T = T (p(x)) + T (q (x)), and further T (p(x)) = T (a3 x3 + a2 x2 + a1 x + a0 ) = (a0 , a1 , a2 , a3 ) = (a0 , a1 , a2 , a3 ) = T (p()), so T is linear. (2.1)

32

2. LINEAR TRANSFORMATIONS

Alternatively, it suces to write T (p(x) + q (x)) = T ((a3 + b3 )x3 + (a2 + b2 )x2 + (a1 + b1 )x + (a0 + b0 )) = (a0 + b0 , a1 + b1 , a2 + b2 , a3 + b3 )T = (a0 , a1 , a2 , a3 )T + (b0 , b1 , b2 , b3 )T = T (p(x)) + (T (q (x)). by Lemma 20. Exercise 10. Dene T : P3 (R) R4 by T (a3 x3 + a2 x2 + a1 x + a0 ) = (a0 , a1 + 1, a2 , a3 )T . Show that T is not a linear map. Answer. Now T (0) = (0, 1, 0, 0)T , so T (0) = 0T . This implies that T is not linear.

Exercise 11. Dene T : P3 (R) R by

Show that T is not a linear map. Answer. Suppose that p(x) = a3 x3 + a2 x2 + a1 x + a0 , where a2 = 0. Then T (p(x)) = T (a3 x3 + a2 x2 + a1 x + a0 ) = (a0 , a1 , (a2 )2 , a3 ) = (a0 , a1 , a2 2 , a3 ) = T (p(x)), unless = 0 or 1, or a2 = 0. Exercise 12. Dene T : Pn (R) Pn (R) by T (p(x)) = x2 Find rank(T ) and nullity(T ). Answer. We can represent T by a matrix: let B be the basis {1, x, x2 , , xn } for Pn (R). Then [T (p(x))]B = M[p(x)]B , d2 p(x) dp(x) + p(x). 2 x dx2 dx

4. RANK AND NULLITY

33

where the ith column of M is [T (xi1 )]B . Now T (xi1 ) = x2 (i 1)(i 2)xi3 2x(i 1)xi2 + xi1 = [(i2 3i + 2) 2(i 1) + 1]xi1 = (i2 5i + 5)xi1 . Then M is the diagonal matrix 0 0 . . . 0 0 0 . . . 0 0 0 . . . 0 .. . 0 0 0 , . . .

and mii = i2 5i + 5. For i = 1, 2, 3, . . . , this never vanishes. Thus rank(M) = n + 1 and nullity(M) = 0. It follows that rank(T ) = n + 1 and nullity(T ) = 0. Alternatively, we can avoid using matrices. By dierential equations, the general solution to T (f (x)) = 0 is f (x) = Aex + Bex , where , = (5 5)/2. The only such f (x) which is a polynomial is f (x) = 0. Thus when we consider T acting on polynomials, the kernel is {0}. By the rank-nullity theorem, the rank is (n + 1). This means that, for any q in Pn (R), we can nd a unique polynomial solution to T (p(x)) = q (x).