SOLUTION KEY TO THE LINEAR ALGEBRA FINAL EXAM () We find a least squares solution to ( ) ( ) A x = y or 0 0 a b = c 4 0 0. 0 The normal equation is A T A x = A T y = y or 5 0 0 0 0 0 a b = 5 9. 0 0 4 7 c The least-squares solution is x = a b = 0 9 0 5 so the sought-after polynomial is p(t) = 9 t 0 t. () (a) c () rref(a) = 0 0 0 0 0 So a basis for V = Im(A) is given by the first two columns of A. A routine application of the Gram-Schmidt process to these two columns yields the orthonormal basis 4, 0. Another basis is,. Date: May 00. c Eduardo Dueñez 00.
SOLUTION KEY TO THE LINEAR ALGEBRA FINAL EXAM (b) Since A is a projection matrix onto V, Ker(A) = V. From (), the vector is a basis for Ker(A) so an orthonormal basis consists of the vector. (c) P = [ ] = 4 4. 9 4 4 On the other hand, it is geometrically obvious that x = proj V x + proj V x for any vector x R n and subspace V R n, which in our case can be read to say A + P = I, providing a second (and easier) way of computing P. () (a) The ellipse is q( x) = where q( x) = x T A x and [ ] 6 A =. We have p A (λ) = λ 9λ + 4 = (λ 7)(λ ) so the eigenvalues of A are λ = 7, λ =. The principal axes are c axis: E 7 = Ker(7I A) = span u, u = [ ] c axis: E = Ker(I A) = span u, u = [ ]. (b) In c -c coordinates: q( x) = λ c + λ c so the equation of the ellipse becomes 7c + c =. (c) The lengths of the semiaxes of the ellipse are / λ = / 7 and / λ = /. (4) (a) We need to prove that q( x) = x, x > 0 for any x 0. The determinants of the principal submatrices of A are det A () = det[] = > 0 and det A () = det A = 6 > 0 so q is a positive definite quadratic form and the property above holds. (b) Let us agree that v denotes not the Euclidean (usual) length of v but rather the length computed using the inner
SOLUTION KEY TO THE LINEAR ALGEBRA FINAL EXAM c x 0.5 c -0.4-0. 0. 0.4 x -0.5 - Figure. The ellipse 6x + 4x x + x = with its principal axes and the vectors u / 7 (black) and u / (blue). product: v = v, v = q( v). E is not an orthonormal basis of R since, for instance, e = q( e ) =. We apply the Gram-Schmidt process to the standard basis E and let v = e e = [ ] 0 ṽ = e v, e v = e + [ v = ] ṽ = q(ṽ ) = v = ṽ = [ ], { [ ] [ ]} so an orthonormal basis is U = { v, v } =,. 0 Another orthonormal [ ] basis consists of [ the semiaxis ] vectors λ / u = 5 and λ / u = 0 for the ellipse q( x) =. (5) (a) True. The equality of the eigenvalues follows from the equality of the characteristic polynomials. Since p M (λ) =
4 SOLUTION KEY TO THE LINEAR ALGEBRA FINAL EXAM λ (Trace M)λ+det M for any matrix M, it suffices to show that AB and BA have the same trace (we know this) and determinant. However, det(ab) = det(a) det(b) = det(b) det(a) = det(ba). (b) True. A will be a reflection in a line L R if A x = x for any x L and A x = x for any x L. Now, for our A we have p A (λ) = λ, so the eigenvalues of A are ± (each with multiplicity one). All we need to show now is that the eigenspaces E ± are perpendicular lines in R for then A will be the reflection in L = E +. There is no need to find the eigenvectors for A explicitly since, A being real and symmetric, its eigenspaces are necessarily orthogonal. (c) False. If there were such a basis B then the matrices A = [T ] E and I 5 = [T ] B would satisfy I 5 = S AS where S is the matrix of change of basis E B. However, it follows from the equation above that A = SI 5 S = SS = I 5, so A was the identity matrix to begin with. Hence the statement is false unless T is the identity transformation of R 5. (d) True. Take a basis (actually, any spanning set of vectors will do just as well) for V, say { v, v,..., v k }. Then Im(A) = V where A = v v v k. (e) Since A, B are positive definite, for any nonzero x R n, x T A x > 0 and x T B x > 0. Adding the two equations we obtain x T (A+B) x = x T A x+ x T B x > 0 so A+B is positive definite as well. (f) False. If A T b = 0 and A x = b is consistent then b is both in Ker(A T ) and in Im(A). Since these are mutually orthogonally complementary subspaces, this would imply b = 0. In other words, if A T b = 0 then system A x = b is necessarily inconsistent unless b = 0! (g) True. Under a suitable change of basis, the matrix B of the shear will be of the form [ ] t B = 0
SOLUTION KEY TO THE LINEAR ALGEBRA FINAL EXAM 5 for some number t. This is because, if { v, v } is a basis of R with v in the line L of the shear, then A v = v (hence the first column of B) and A v v must be some vector t v L. A quick calculation shows that () B + I = B. If S is the matrix of change of basis, then SBS = A. Multiplying equation () by S on the left and S on the right we obtain the desired result.