Applied Linear Algebra , Final Exam 1. (20pts) Let A be a real symmetric square matrix. Show the followings. (a) Every eigenvalue of A is real. (b) Any two eigenvectors of A corresponding to distinct eigenvalues are orthogonal. (c) If A is positive definite, then every eigenvalue of A is positive. (d) If every eigenvalue of A is positive, then A is positive definite. Solution (a) Every eigenvalue of A is real . pf) Let v be an unit eigenvector corresponding to λ. 0 ≤ (Av)H Av = v H AH Av = |λ|2 v H v = |λ|2 (b) Any two eigenvectors of A corresponding to distinct eigenvalues are orthogonal. pf) Let v1 and v2 be two eigenvectors corresponding to λ1 and λ2 respectively. then λ1 v1T v2 = (Av1 )T v2 = v1T AT v2 = v1T Av2 = λ2 v1T v2 so v1 and v2 are orthogonal . (c) If A is positive definite, then every eigenvalue of A is positive. pf) Let v be an unit eigenvector of A corresponding to λ. then v T Av = λ is positive (d) If every eigenvalue of A is positive, then A is positive definite. pf) Since A can be diagonalizable, we can decompose A as A = QT ∆Q where Q is an orthogonal matrix and ∆ is a diagonal matrix whose entries on the main diagonal. for any x ∈ Rn xT Ax = xT QT ∆Qx = (Qx)T ∆Qx which is positive definite. 1 2. (10pts) Let A be a real skew-symmetric square matrix, i.e., AT = −A. P tn n (a) Show that the matrix etA = I + ∞ n=1 n! A is orthogonal for all t ∈ R. (b) Show that d dt ||u(t)|| = 0 if d dt u(t) = Au(t). Solution From e tA =I+ ∞ n X t n=1 we have tA T (e ) = I + ∞ n X t n=1 n! T n n! (A ) = I + An , ∞ X (−t)n n=1 n! An = e−tA Therefore, we have (etA )T etA = e−tA etA = I. The solution of the differential equation is u(t) = etA u(0). It follows that ||u(t)||2 = d u(0)T e−tA etA u(0) = u(0)T u(0) = ||u(0)||2 . Hence dt ||u(t)|| = 0. 2 3. (10pts) Let A, B be two 3 × 3 matrices given by 0 1 2 0 0 1 A = 0 0 1 , B = 0 0 0 . 0 0 0 0 0 0 Determine whether A and B are similar. Solution Please do it yourself. 3 4. (10pts) Let F0 , F1 , F2 , . . . be a sequence of real numbers defined by the recurrence relation: F0 = 0, F1 = a, Fk+2 = aFk+1 + Fk for k = 0, 1, 2, 3, . . . , where a ̸= 0. Find the general formula for Fk . Solution: a 1 Set A = to find that 1 0 Fk+2 a 1 Fk+1 aFk+1 + Fk = = Fk+1 1 0 Fk Fk+1 From Fk+1 k a =A 0 Fk ⇒ a−z 1 det(A − zI2 ) = det = z 2 − az − 1, 1 −z the eigenvalues of A are √ a2 + 4 , λ1 = 2 with the corresponding eigenvectors λ1 v1 = , 1 a+ λ2 = a− √ a2 + 4 , 2 λ2 v2 = 1 Diagonalize A A = SΛS −1 = λ1 λ2 1 1 1 λ1 0 1 −λ2 0 λ2 λ1 − λ2 −1 λ1 √ a2 + 4 ̸= 0. Compute k 1 λ1 λ2 λ1 0 1 −λ2 k −1 A = SΛS = 1 1 0 λk2 λ1 − λ2 −1 λ1 k+1 1 1 −λ2 λk+1 λ1 2 = −1 λ1 λk1 λk2 λ1 − λ2 k+1 1 λ1 − λk+1 λ1 λk2 − λ2 λk1 2 = λk1 − λk2 λ1 λk2 − λ2 λk1 λ1 − λ2 where λ1 − λ2 = Finally k+1 1 a Fk+2 a(λk+1 − λk+1 ) λ1 − λk+1 k a 1 2 2 =A =√ =√ Fk 0 a(λk1 − λk2 ) λk1 − λk2 a2 + 4 a2 + 4 to obtain that Fk = √ √ a2 a +4 a+ 2 +4 a2 4 !k − a− √ a2 2 +4 !k 5. (5pts) Use an appropriate method to show 1 A = 1 1 that the following matrix 1 1 5 3 3 11 is positive definite. Solution: The characteristic polynomial of A is 1−z 1 1 5−z 3 = −(z 3 − 17z 2 + 72z − 42) det(A − zI3 ) = 1 1 3 11 − z But it is not easy to find roots. You can try Cholesky factorization A = CC T 1 0 C= 1 2 1 1 to find that 0 0 3 and notice that C has positive diagonals so that A is positive definite. You can try A = LU = LDLT factorization to find that all pivots are positive. You can use Syvester’s criterion and find det 1 = 1 > 0, 1 1 det = 3 > 0, 1 5 1 1 1 det 1 5 3 = 1(55 − 9) − 1(11 − 3) + 1(3 − 5) = 36 > 0 1 3 11 to conclude that A is positive definite. Etc. 5 6. (20pts) Let A be a m × 3 real matrix, where m is a positive integer, satisfying 3 a −1 3 −1 and ab = 1, a > 0. AT A = b −1 −1 5 Answer the following questions and justify your answers. (a) What are the possible values of a and b? (3pts) (b) What is the rank of A? (3pts) (c) What are the possible values of m? (4pts) (d) What is the maximum value of ||Ax|| for all x ∈ R3 with ||x|| = √ 2. (10pts) Solution AT A should be a symmetric matrix since (AT A)T = AT (AT )T = AT A. Therefore a = b. From a2 = 1 and a > 0, we have a = 1. Then the eigenvalues of AT A are λ1 = 2 λ2 = 3 and λ3 = 6 with corresponding orthonormal eigenvectors are 1 1 −1 1 1 1 −1 , 1 , v3 = √ −1 , v2 = √ v1 = √ 2 3 6 0 1 2 It follows that A has non-zero singular values σ1 = the rank of A is 3. √ 2, σ2 = √ 3 and σ3 = √ 6 so that Since A is an m × 3 matrix with rank 3, m must be larger or equal to 3. We can decompose any x ∈ R3 as x = c1 v1 + c2 v2 + c3 v3 such√that ||x||2 = c21 + c22 + c23 , since {v1 , v2 , v3 } is an orthonormal basis of R3 . From ||x|| = 2, we have c21 +c22 +c23 = 2. From AT Ax = c1 λ1 v1 + c2 λ2 v2 + c3 λ3 v3 we obtain that ||Ax||2 = c21 λ1 + c22 λ2 + c23 λ3 ≤ λ3 (c21 + c22 + c23 ) = 2λ3 = 12 √ √ where we have used 0 < λ1 < λ2 < λ3 . Therefore 12 = 2 3 is an upper bound of ||Ax||. On the other hand √ √ √ √ √ A 2v3 = 2(Av3 ) = 2(σ3 u3 ) = 12u3 , ||A 2v3 ||2 = 12 T where u3 is the third column of U √ obtained from A = U ΣV . Therefore the maximum of ||Ax|| is 12 attained for x = 2v3 . 6 7. (25pts) Let A be a 2 × 3 real matrix given by 1 0 1 A= . −1 1 0 (a) Find a singular value decomposition A = U ΣV T . (10pts) (b) Find orthonormal bases of four fundamental subspaces associated to A and justify your answers. (5pts) (c) Find the pseudo (Moore-Penrose) inverse A+ of A and check that A(I3 −A+ A) = 0 and AT AA+ = AT . (5pts) 1 3 (d) Find every linear least square solution x̂ ∈ R minimizing ||Ax − b|| for b = 1 and justify your answer. (5pts). Solution. We have 2 −1 1 AT A = −1 1 0 1 0 1 and 2 − z −1 1 0 = (2 − z)(1 − z)(1 − z) − (1 − z) − (1 − z) det(AT A − zI3 ) = det −1 1 − z 1 0 1−z = (3 − z)(1 − z)z Therefore the eigenvalues of AT A are λ1 = 3, λ2 = 1 and λ3 = 0, all distinct, with the corresponding orthonormal eigenvectors 2 0 −1 1 1 1 −1 , v2 = √ 1 , v3 = √ −1 v1 = √ 6 2 1 3 1 1 So that we have the following 3 × 3 orthogonal matrix √ 2 −1 √1 2 √0 −√2 √ 1 1 0 V = √ −1 √3 −√ 2 ⇒ V T = √ √ √3 √3 6 6 − 2 − 2 2 1 3 2 The singular values of A are σ1 = √ 3, σ2 = 1 and σ3 = 0 so that √ 3 0 0 Σ= 0 1 0 7 From u1 = 1 1 Av1 = √ σ1 18 1 1 u2 = Av2 = √ σ2 2 1 0 −1 1 1 0 −1 1 2 1 1 1 3 1 −1 = √ =√ , 0 −3 −1 18 2 1 0 1 1 1 √ 1 = 0 2 1 1 we obtain the following 2 × 2 matrix 1 U=√ 2 1 1 −1 1 Finally we have a singular value decomposition 1 A = U ΣV T = √ 2 1 1 −1 1 √ 3 0 0 0 1 0 2 −1 √1 √ 1 √ √ 0 3 3 6 − 2 −√2 √2 Or in the compact form A= 2 X j=1 σj uj vjT √ 1 3 1 1 1 1 √ 2 −1 1 + √ √ 0 1 1 =√ 2 −1 6 2 1 2 Check if the student has correctly identified the column space of A, column space of AT , the null space of A and the null space of AT from the above. 8 The pseudo inverse A+ of A is 2 2 0 X 1 1 1 1 1 1 + T −1 √ 1 −1 + √ 1 √ 1 1 A = vj uj = √ √ σ 3 6 2 2 1 2 j=1 j 1 1 1 2 −2 0 0 3 −3 1 1 2 −1 1 + 1 1 = 13 = 3 6 2 2 1 1 −1 1 1 3 3 1 −1 1 = 1 2 3 2 1 so that 1 −1 2 −1 1 1 1 1 0 1 = −1 2 1 A+ A = 1 2 −1 1 0 3 3 2 1 1 1 2 Therefore we have the projector R3 → Null (A) 1 1 I3 − A+ A = 1 3 −1 given by 1 −1 1 −1 −1 1 and check A(I3 − A+ A) = 1 3 1 1 −1 1 0 0 0 1 0 1 1 1 −1 = −1 1 0 3 0 0 0 −1 −1 1 1 −1 2 −1 1 3 −3 1 1 AT AA+ = −1 1 0 1 2 = 0 3 = AT 3 3 2 1 1 0 1 3 0 A least square x̂ is a solution to the normal equation AT Ax = AT b and A+ b is a particular solution AT AA+ b = AT b and AT A(I3 − A+ A) = 0. Therefore every least square solution x̂ is given by x̂ = A+ b − (I3 − A+ A)w, 9 ∀w ∈ R3 .