Math 317: Linear Algebra Homework 8 Solutions The following problems are for additional practice and are not to be turned in: (All problems come from Linear Algebra: A Geometric Approach, 2nd Edition by ShifrinAdams.) Exercises: Section 3.6: 1–6 Section 4.1: 6,7,8 Turn in the following problems. 1. Section 3.6, Problem 8 (a) Proof : Let V = A ∈ Rn×n | A = AT . We claim that V is a subspace of Rn . We begin by observing that 0n×n ∈ V , since 0T = 0. That is, the zero matrix is symmetric. Let A, B ∈ V . Then A = AT and B = B T (wts. A + B is symmetric). Then (A + B)T = AT + B T = A + B so A + B is symmetric and hence is an element of V . Let c ∈ R and suppose that A ∈ V . Then (cA)T = cAT = cA and so cA is symmetric and hence an element of V . Since all three conditions of a subspace were met, we may conclude that V is a subspace of Rn×n . The dimension is given by 2 n + n 2−n (b) Proof : Let V = A ∈ Rn×n | A = −AT . We claim that V is a subspace of Rn . We begin by observing that 0n×n ∈ V , since 0T = −0 = 0. That is, the zero matrix is skew symmetric. Let A, B ∈ V . Then A = −AT and B = −B T (wts. A + B is skew symmetric). Then (A + B)T = AT + B T = −A − B = −(A + B) so A + B is skew-symmetric and hence is an element of V . Let c ∈ R and suppose that A ∈ V . Then (cA)T = cAT = −cA and so cA is skew-symmetric and hence an element of V . Since all three conditions of a subspace were met, we may conclude that V is a subspace 2 of Rn×n . The dimension is given by n 2−n . (c) Proof : From a previous homework, we recall that every square matrix A can be uniquely written as the sum of a symmetric matrix and a skewsymmetric matrix. This is equivalent to saying that Mn×n = S + K where S is the set of symmetric matrices, K is the set of skew-symmetric matrices, and Mn×n is the set of all n × n matrices. 2. Section 3.6, Problem 9 (Hint: The dimension of Rn×n is n2 . Why? ) (a) Proof : By definition of the transpose of A, we know that if A = [Aij ], then AT = [Aji ] and diagonal entries of A and AT are the same. Pn so the P Thus trace(A) = i=1 aii = ni=1 aTii = trace(AT ). (b) Proof : For any two matrices A and B, have that P (A + B)ii P = aii + bii . Pn Pwe n n Thus trace(A+B) = i=1 [A+B]ii = i=1 aii +bii = i=1 aii + ni=1 bii = trace(A) + trace(B).PFor any realPnumber c, we P have that [cA]ii = caii and so trace(cA) = ni=1 [cA]ii = ni=1 caii = c ni=1 aii = ctrace(A). 1 Math 317: Linear Algebra Homework 8 Solutions (c) Proof : Using the definition of matrix multiplication, we call that [AB]ii = ith row of A · ith column Pn of B [a a . . . , a ] · [b b . . . , b ] =⇒ trace(AB) = in 1i 2i ni Pi1n i2 Pn Pn i=1 [AB]ii ] · [b 1i b2i . . . , bni ]P = i=1 j=1 aij bji Pni=1 [a Pi1nai2 . . . , ainP P n n n b a = b a = [BA] = trace(BA). jj i=1 j=1 ij ji j=1 i=1 ij ji j=1 re= = = (d) Proof : Define < A, B >= trace(AT B). We verify that < A, B > is an inner product. We begin by observing (from part (a)) that < A, B >= trace(AT B) = trace((AT B)T ) = trace(B T A) =< B, A >. Next we find that < A + B, C >= trace((A + B)T C) = trace((AT + B T )C) = trace(AT C +B T C) = trace(AT C)+trace(B T C) =< A, C > + < B, C > . For any real number c, we have that < cA, B >= trace((cA)T B) = trace(cAT B) = ctrace(AT B) c < A, B >. PFinally, we have that Pn= P n n Pn T T < A, A >= trace(A A) = i=1 j=1 aij aji = i=1 j=1 a2ji ≥ 0, and P P < A, A >= 0 if and only if ni=1 nj=1 a2ji = 0 which occurs if and only if aji = 0 for all i, j, which is true if and only if A = 0. Thus < A, B > is an inner product. (e) Proof : Suppose that A is symmetric and B is skew-symmetric. Then A = AT and B = −B T . Thus we have that < A, B >= trace(AT B) = trace(BAT ) = trace(−B T A) = −trace(B T A) = − < B, A >. But < A, B >=< B, A > so < B, A >= − < B, A > =⇒ < B, A >= 0 =⇒ < A, B >= 0. (f) Proof : From the proof above, we have that A and B are orthogonal if A is symmetric and B is skew symmetric (since < A, B >= 0.) Since this is true for all symmetric and skew-symmetric matrices, then the set of all symmetric matrices and the set of all skew symmetric matrices are orthogonal complements of each other. 3. Section 3.6, Problem 10a Proof : We show that V = {A ∈ Rn | trace(A) = 0} is a subspace of the space of We begin by noting that 0n×n ∈ V since trace(0n×n ) = Pnall n × n matrices. Pn 0 = 0. Now, suppose that A, B ∈ V and let c be any 0 = i=1 ii i=1 real number. Then trace(A) = 0 and trace(B) = 0. Using the properties for trace established in the previous problem, we have that trace(A + B) = trace(A) + trace(B) = 0 + 0 = 0 and trace(cA) = ctrace(A) = c(0) = 0. Thus A + B and cA ∈ V for any real number c. Therefore, V is a subspace of the space of all n × n matrices. 4. Let Pn denote the vector space of polynomials of degree n or less. Define V = {tn p0 (0) + t, p ∈ Pn , t ∈ R} ⊂ Pn where p0 (0) denotes the first derivative of p ∈ Pn evaluated at 0. Prove or disprove that V is a subspace of Pn . We claim that V is not a subspace of Pn because it is not closed under addition. 0 0 To see why, let s, q ∈ V . Then s(t) = p1 (0)tn + t and q(t) = p2 (0)tn + t, where 0 0 p1 and p2 ∈ Pn . Then s + q = (p1 (0) + p2 (0))tn + 2t 6= tn p0 (0) + t for some 2 Math 317: Linear Algebra Homework 8 Solutions polynomial p. (since we now have 2t rather than t). Thus s + q ∈ / V and hence V cannot be a subspace of Pn . 5. Section 4.1, Problem 7 To find the least squares solution associated system with the given of equations, 1 1 1 we find x̂ = (AT A)−1 AT b where A = 1 −3 and b = 4. Thus we have 2 1 3 that −1 1 1 1 11 0 1 1 2 1 −3 = , = 1 −3 1 66 0 6 2 1 (AT A)−1 and so 1 1 121 1 11 0 1 1 2 4 = x̂ = 66 0 6 1 −3 1 66 −48 3 73 1 265 . The point closest to b is given by projV (b) = Ax̂ = 66 194 6. Section 4.1, Problem 15 Proof: We show that A is a projection matrix onto its own column space, C(A). To show this, we must show that for all x ∈ Rn , we have Ax ∈ C(A) and x − Ax ∈ C(A)⊥ = N (AT ). Note that this comes from the definition of a projection with p = Ax and b = x. We observe that by definition of C(A), Ax ∈ C(A). To show that x − Ax ∈ N (AT ) we must show that AT (x − Ax) = 0. Using that fact that AT = A and A2 = A, we see that AT (x − Ax) = AT x − AT Ax = Ax − A2 x = Ax − Ax = 0. Thus, A is a projection matrix onto C(A). 7. A small company has been in business for three years and has recorded annual profits (in thousands of dollars) as follows. Year 1 Sales 7 2 4 3 3 Assuming that there is a linear trend in the declining profits, predict the year and the month in which the company begins to lose money. Since we are told that there is a linear trend, we will use a least squares approach to find the line that best fits the data. To this extent, using y = mx + b with the data points (1, 7), (2, 4), (3, 3) we obtain the following system of equations: 3 Math 317: Linear Algebra Homework 8 Solutions m+b=7 2m + b = 4 3m + b = 3, which in matrix form Ax = b becomes 1 1 7 2 1 m = 4 . b 3 1 3 Noting that this system is inconsistent (while A is full rank), we proceed by finding the least squares solution x̂ = (AT A)−1 AT b. We begin by calculating (AT A)−1 to obtain: −1 1 1 −1 1 3 −6 1 2 3 14 6 2 1 = . = = 1 1 1 6 3 6 −6 14 3 1 (AT A)−1 Thus, the least squares solution is 7 1 3 −6 1 2 3 1 −6 T −1 T 4 = . x̂ = (A A) A b = 6 −6 14 1 1 1 3 26 3 Thus, the best fit line is given by y = −2x+ 26 . To find out when the company 3 begins to lose money, we find the x such that y = 0. This is given by x = 13/3 months which translate into 4 years and 4 months. By the 5th month of the 4th year, the company will begin to lose money. 4