2.2-exam1-2007-johnson¶
Quiz 1, 2007 (johnson) · Section 2.2: Elimination Matrices and Inverse Matrices Source: exam1_1.pdf, pp. 178-179
Problem¶
3 (20 pts.) Determine whether the following statements are true or false, and explain your reasoning.
\((\clubsuit)\) If \(A^2 = A\), then \(A = 0\) or \(A = I\).
\((\diamondsuit)\) Ignoring row swaps, any invertible matrix \(A\) has a "UL" factorization (as an alternative to LU factorization): \(A\) can be written as \(A = UL\) where \(U\) and \(L\) are some upper and lower triangular matrices, respectively.
\((\spadesuit)\) All the \(2 \times 2\) matrices that commute with \(A = \begin{bmatrix} 1 & 3 \\ 2 & 0 \end{bmatrix}\) (i.e. all \(2 \times 2\) matrices \(B\) with \(AB = BA\)) form a vector space (with the usual formulas for addition of matrices and multiplication of matrices by numbers).
\((\heartsuit)\) There is no \(3 \times 3\) matrix whose column space equals its nullspace.
Solution
\((\clubsuit)\) False. Counterexample: if \(A = \begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix}\), then \(A^2 = A\) but \(A \neq I\) and \(A \neq 0\). Note that if we assume \(A\) is invertible, then the only solution is \(A = I\) (multiply both sides of \(A^2 = A\) by \(A^{-1}\)), but this assumption is not warranted here. \((\diamondsuit)\) True. Instead of "downwards" elimination we can also do "upwards" elimination to put \(A\) into lower-triangular form \(L\) (possibly with row swaps). In this procedure the corresponding elimination matrices are upper-triangular, but still multiply on the left (since they are still row operations), so we get a UL factorization. Alternatively, the "leftwards" elimination of problem 2 also leads to a UL factorization, because the (lower-triangular!) elimination matrices multiply on the *right* to give \(U = AL^{-1}\). Technically, however, this process may require *column* swaps if zeros are encountered in pivot positions. \((\spadesuit)\) True. We need to know that linear combinations of vectors stay in the vector space. If \(B\) is a matrix where \(AB = BA\), then clearly \(A(cB) = c(AB) = c(BA) = (cB)A\) for any \(c\). If \(B'\) is another matrix where \(AB' = B'A\), then \(A(B + B') = AB + AB' = BA + B'A = (B + B')A\). (The other properties of a vector space, associativity etcetera, need not be shown since they are automatic for the usual addition and multiplication operations.) \((\heartsuit)\) True. Suppose the rank of \(A\) is \(r\); then the dimension of the column space is \(r\), and the dimension of the null space is \(3 - r\). Obviously no matter \(r = 0, 1, 2, 3\), we always have \(r \neq 3 - r\). (Equivalently, \(r = 3 - r\) would imply a fractional rank \(r = 3/2\)!) This shows that the two spaces are not the same, since they must have different dimensions.Keywords: invertible matrix, UL factorization, vector space, nullspace, rank