Q
stringlengths
4
3.96k
A
stringlengths
1
3k
Result
stringclasses
4 values
Theorem 7.2.6. Let \( A \in {M}_{n} \) be Hermitian and positive semidefinite, let \( r = \operatorname{rank}A \) , and let \( k \in \{ 2,3,\ldots \} \).\n\n(a) There is a unique Hermitian positive semidefinite matrix \( B \) such that \( {B}^{k} = A \).\n\n(b) There is a polynomial \( p \) with real coefficients such ...
Proof. Represent \( A = {U\Lambda }{U}^{ * } \), in which \( U = \left\lbrack \begin{array}{ll} {U}_{1} & {U}_{2} \end{array}\right\rbrack \) is unitary, \( {U}_{1} \in {M}_{n, r},\Lambda = \) \( \operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{r}}\right) \oplus {0}_{n - r} \), and \( {\lambda }_{1},\ldots...
Yes
Theorem 7.2.7. Let \( A \in {M}_{n} \) be Hermitian.\n\n(a) \( A \) is positive semidefinite if and only if there is a \( B \in {M}_{m, n} \) such that \( A = {B}^{ * }B \) .
Proof. (a) If \( A = {B}^{ * }B \) for some \( B \in {M}_{m, n} \), then \( {x}^{ * }{Ax} = {x}^{ * }{B}^{ * }{Bx} = \parallel {Bx}{\parallel }_{2}^{2} \geq 0 \) . The asserted factorization can be achieved, for example, with \( B = {A}^{1/2} \) and \( m = n \) .
Yes
Corollary 7.2.8. A Hermitian matrix \( A \) is positive definite if and only if it is \( {}^{ * } \) congruent to the identity.
Proof. This is simply a restatement of (7.2.7).
No
Corollary 7.2.9 (Cholesky factorization). Let \( A \in {M}_{n} \) be Hermitian. Then \( A \) is positive semidefinite (respectively, positive definite) if and only if there is a lower triangular matrix \( L \in {M}_{n} \) with nonnegative (respectively, positive) diagonal entries such that \( A = L{L}^{ * } \) . If \( ...
Proof. Let \( {A}^{1/2} = {QR} \) be a \( {QR} \) factorization and let \( L = {R}^{ * } \) . Then \( A = {A}^{1/2}{A}^{1/2} = \) \( {R}^{ * }{Q}^{ * }{QR} = {R}^{ * }R = L{L}^{ * } \) . The asserted properties of \( L \) follow from the properties of \( R \) stated in (2.1.14).
No
Theorem 7.2.10. Let \( {v}_{1},\ldots ,{v}_{m} \) be vectors in an inner product space \( V \) with inner product \( \langle \cdot , \cdot \rangle \), and let \( G = {\left\lbrack \left\langle {v}_{j},{v}_{i}\right\rangle \right\rbrack }_{i, j = 1}^{m} \in {M}_{m} \). Then\n\n(a) \( G \) is Hermitian and positive semid...
Proof. (a) Let \( \parallel \cdot \parallel \) be the norm derived from the given inner product and let \( x = \left\lbrack {x}_{i}\right\rbrack \in \) \( {\mathbf{C}}^{m} \). The properties listed in (5.1.3) ensure that \( G \) is Hermitian and\n\n\[ \n{x}^{ * }{Gx} = \mathop{\sum }\limits_{{i, j = 1}}^{m}\left\langle...
Yes
Theorem 7.3.1 (Polar decomposition). Let \( A \in {M}_{n, m} \). (a) If \( n < m \), then \( A = {PU} \), in which \( P \in {M}_{n} \) is positive semidefinite and \( U \in {M}_{n, m} \) has orthonormal rows. The factor \( P = {\left( A{A}^{ * }\right) }^{1/2} \) is uniquely determined; it is a polynomial in \( A{A}^{ ...
Proof. We adopt the notation of (2.6.3), which ensures that there are unitary matrices \( V \in {M}_{n} \) and \( W \in {M}_{m} \), and a nonnegative diagonal matrix \( \sum \in {M}_{n, m} \) with a special structure, such that \( A = {V\sum }{W}^{ * } \). Let \( q = \min \{ n, m\} \) and let \( {\sum }_{q} \in {M}_{q}...
Yes
Theorem 7.3.2. Let \( A \in {M}_{n, m} \), let \( q = \min \{ n, m\} \), and let \( r = \operatorname{rank}A \) . Suppose that \( {A}^{ * }A = {W\Lambda }{W}^{ * } \), in which \( W \in {M}_{m} \) is unitary, \( \Lambda = \operatorname{diag}\left( {{\sigma }_{1}^{2},\ldots ,{\sigma }_{r}^{2}}\right) \oplus {0}_{m - r} ...
Proof. (a) Let \( D = {\sum }_{r} \oplus {I}_{m - r} \in {M}_{m} \) and partition \( X = {AW}{D}^{-1} = \left\lbrack {{V}_{1}Z}\right\rbrack \in {M}_{n, m} \), in which \( {V}_{1} \in {M}_{n, r} \) . Then \( {X}^{ * }X = {D}^{-1}{W}^{ * }{A}^{ * }{AW}{D}^{-1} = {D}^{-1}\Lambda {D}^{-1} = {I}_{r} \oplus {0}_{m - r} \), ...
Yes
Theorem 7.3.3. Let \( A \in {M}_{n, m} \), let \( q = \min \{ n, m\} \), let \( {\sigma }_{1} \geq \cdots \geq {\sigma }_{q} \) be the ordered singular values of \( A \), and define the Hermitian matrix\n\n\[ \mathcal{A} = \left\lbrack \begin{matrix} 0 & A \\ {A}^{ * } & 0 \end{matrix}\right\rbrack \]\n\nThe ordered ei...
Proof. Suppose that \( n \geq m \) and let \( A = {V\sum }{W}^{ * } \) be a singular value decomposition, in which \( \sum = {\left\lbrack {\sum }_{m}0\right\rbrack }^{T} \in {M}_{n, m} \) . Write the left unitary factor as \( V = \left\lbrack \begin{array}{ll} {V}_{1} & {V}_{2} \end{array}\right\rbrack \in {M}_{n} \),...
Yes
Corollary 7.3.5. Let \( A, B \in {M}_{n, m} \) and let \( q = \min \{ m, n\} \) . Let \( {\sigma }_{1}\left( A\right) \geq \cdots \geq {\sigma }_{q}\left( A\right) \) and \( {\sigma }_{1}\left( B\right) \geq \cdots \geq {\sigma }_{q}\left( B\right) \) be the nonincreasingly ordered singular values of \( A \) and \( B \...
Proof. (a) Let \( E = A - B \) and apply (6.3.4.1) to \( \mathcal{A} = \left\lbrack \begin{matrix} 0 & A \\ {A}^{ * } & 0 \end{matrix}\right\rbrack \) and \( \mathcal{E} = \left\lbrack \begin{matrix} 0 & E \\ {E}^{ * } & 0 \end{matrix}\right\rbrack \) .\n\n(b) Apply (6.3.9) to \( \mathcal{A} \) and \( \mathcal{E} \) ; ...
No
Corollary 7.3.6. Let \( A \in {M}_{n, m} \), let \( q = \min \{ m, n\} \), and let \( \widehat{A} \) be a matrix obtained from A by deleting any one of its columns or rows. Let \( {\sigma }_{1} \geq \cdots \geq {\sigma }_{q} \) and \( {\widehat{\sigma }}_{1} \geq \cdots \geq {\widehat{\sigma }}_{q} \) denote the respec...
Proof. Let \( \mathcal{A} = \left\lbrack \begin{matrix} 0 & A \\ {A}^{ * } & 0 \end{matrix}\right\rbrack \) . Deleting row \( i \) from \( A \) corresponds to deleting row \( i \) and column \( i \) from \( \widetilde{\mathcal{A}} \) ; deleting column \( j \) from \( A \) corresponds to deleting row \( n + j \) and col...
Yes
Theorem 7.3.8. Let \( A \in {M}_{n, m} \), let \( q = \min \{ n, m\} \), let \( {\sigma }_{1}\left( A\right) \geq \cdots \geq {\sigma }_{q}\left( A\right) \) be the ordered singular values of \( A \), and let \( k \in \{ 1,\ldots, q\} \) . Then\n\n\[ \n{\sigma }_{k}\left( A\right) = \mathop{\min }\limits_{{\{ S : \dim ...
Proof. These characterizations follow from (4.2.7) and (4.2.8). If \( {\lambda }_{1} \leq {\lambda }_{2} \leq \cdots \leq {\lambda }_{m} \) are the ordered eigenvalues of the positive semidefinite Hermitian matrix \( {A}^{ * }A \), then \( {\sigma }_{k}^{2}\left( A\right) = {\lambda }_{m - k + 1}\left( {{A}^{ * }A}\rig...
Yes
Theorem 7.3.11. Let \( n, p \), and \( q \) be positive integers with \( p \leq q \) . Let \( A \in {M}_{p, n} \) and \( B \in {M}_{q, n} \) . Then \( {A}^{ * }A = {B}^{ * }B \) if and only if there is a \( V \in {M}_{q, p} \) with orthonormal columns such that \( B = {VA} \) . If \( A \) and \( B \) are real, then \( ...
Proof. If \( B = {VA} \), then \( {B}^{ * }B = {A}^{ * }{V}^{ * }{VA} = {A}^{ * }A \) . Conversely, if \( {A}^{ * }A = {B}^{ * }B \), then use (7.3.2) and its notation to write \( A = {V}_{1}{\sum }_{r}{W}_{1}^{ * } \) and \( B = {V}_{2}{\sum }_{r}{W}_{1}^{ * } \), in which \( {V}_{1} \in {M}_{p, r} \) and \( {V}_{2} \...
Yes
Lemma 7.5.2. Let \( A, B \in {M}_{n} \) and \( x, y \in {\mathbf{C}}^{n} \) be given. Let \( \operatorname{diag}x \) and \( \operatorname{diag}y \) be the \( n \) -by- \( n \) diagonal matrices whose respective main diagonal entries are the respective entries of \( x \) and \( y \) ; see (0.9.1). Then\n\n\[ \n{x}^{ * }...
Proof. Let \( A = \left\lbrack {a}_{ij}\right\rbrack, B = \left\lbrack {b}_{ij}\right\rbrack, x = \left\lbrack {x}_{i}\right\rbrack \), and \( y = \left\lbrack {y}_{i}\right\rbrack \) . Then (diag \( \bar{x} \) ) \( A = \left\lbrack {{\bar{x}}_{i}{a}_{ij}}\right\rbrack \) and \( B \) diag \( y = \left\lbrack {{b}_{ij}{...
Yes
Theorem 7.5.3. Let \( A, B \in {M}_{n} \) be positive semidefinite.\n\n(a) \( A \circ B \) is positive semidefinite.\n\n(b) If \( A \) is positive definite and every main diagonal entry of \( B \) is positive, then \( A \circ B \) is positive definite.\n\n(c) If both \( A \) and \( B \) are positive definite, then \( A...
Proof. Let \( A = \left\lbrack {a}_{ij}\right\rbrack, B = \left\lbrack {b}_{ij}\right\rbrack \), and \( x = \left\lbrack {x}_{i}\right\rbrack \) .\n\n(a) Let \( C = \left( {\operatorname{diag}x}\right) {\bar{B}}^{1/2} \) and use the preceding lemma to compute\n\n\[{x}^{ * }\left( {A \circ B}\right) x = \operatorname{tr...
Yes
Theorem 7.5.4 (Moutard). Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) . Then \( A \) is positive semidefinite if and only if \( \operatorname{tr}\left( {A{B}^{T}}\right) = \mathop{\sum }\limits_{{i, j = 1}}^{n}{a}_{ij}{b}_{ij} \geq 0 \) for every positive semidefinite \( B = \left\lbrack {b}_{ij}\right\...
Proof. Suppose that \( A \) and \( B \) are positive semidefinite, and let \( e \in {\mathbf{C}}^{n} \) be the all-ones vector. Then \( \operatorname{diag}\left( e\right) = I \) and \( \operatorname{tr}\left( {A{B}^{T}}\right) = \operatorname{tr}\left( {\left( {\operatorname{diag}e}\right) A\left( {\operatorname{diag}e...
Yes
Theorem 7.5.9. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be positive semidefinite.\n\n(a) The Hadamard powers \( {A}^{\left( k\right) } = \left\lbrack {a}_{ij}^{k}\right\rbrack \) are positive semidefinite for all \( k = 1,2,\ldots \) ; they are positive definite if \( A \) is positive definite.\n\n(...
Proof. Only the assertions about positive definiteness require justification. The assertion in (a) follows from (7.5.3) and an induction. The assertion in (b) follows from (a). See (7.5.P18 to P21) for a proof of the assertion in (c).
No
Theorem 7.6.1. Let \( A, B \in {M}_{n} \) be Hermitian.\n\n(a) If \( A \) is positive definite, then there is a nonsingular \( S \in {M}_{n} \) such that \( A = {SI}{S}^{ * } \) and \( B = {S}^{- * }\Lambda {S}^{-1} \), in which \( \Lambda \) is real diagonal. The inertias of \( B \) and \( \Lambda \) are the same, so ...
Proof. (a) Theorem 4.5.7 ensures that there is a nonsingular \( T \in {M}_{n} \) such that \( {T}^{-1}A{T}^{- * } = I \) . The matrix \( {T}^{ * }{BT} \) is Hermitian, so there is a unitary \( U \in {M}_{n} \) such that \( {U}^{ * }\left( {{T}^{ * }{BT}}\right) U = \Lambda \) is diagonal. Let \( S = {TU} \) . Then \( {...
Yes
Corollary 7.6.2. Let \( A, B \in {M}_{n} \) be Hermitian.\n\n(a) If \( A \) is positive definite, then \( {AB} \) is diagonalizable and has real eigenvalues. If, in addition, \( B \) is positive definite or positive semidefinite, then \( \Lambda \) has positive or nonnegative eigenvalues, respectively.\n\n(b) If \( A \...
Proof. (a) Use part (a) of the preceding theorem to represent \( A = S{S}^{ * } \) and \( B = \) \( {S}^{- * }\Lambda {S}^{-1} \) . Then \( {AB} = S{S}^{ * }{S}^{- * }\Lambda {S}^{-1} = {S\Lambda }{S}^{-1} \).\n\n(b) Use part (b) of the preceding theorem to represent \( A = S\left( {{I}_{r} \oplus {0}_{n - r}}\right) {...
Yes
Theorem 7.6.3. Let \( A, B \in {M}_{n} \) be Hermitian, and assume that \( A \) is positive semidefinite and singular. Then \( {AB} \) is similar to \( \Lambda \oplus N \), in which \( \Lambda \) is real diagonal and \( N = {J}_{2}\left( 0\right) \oplus \cdots \oplus {J}_{2}\left( 0\right) \) is a direct sum of 2-by-2 ...
Proof. Choose a nonsingular \( S \) such that \( {S}^{-1}A{S}^{- * } = {I}_{r} \oplus {0}_{n - r} \) and partition \( {S}^{ * }{BS} = \) \( \left\lbrack {B}_{ij}\right\rbrack \) conformally to \( {I}_{r} \oplus {0}_{n - r} \) . Then \( {S}^{-1}{ABS} = \left( {{S}^{-1}A{S}^{- * }}\right) \left( {{S}^{ * }{BS}}\right) = ...
Yes
Theorem 7.6.4. Let \( A, B \in {M}_{n} \) be Hermitian.\n\n(a) If \( A \) is positive definite, then there is a nonsingular \( S \in {M}_{n} \) such that \( A = {SI}{S}^{ * } \) and \( B = {S\Lambda }{S}^{ * } \), in which \( \Lambda \) is real diagonal. The inertias of \( B \) and \( \Lambda \) are the same, so \( \La...
Proof. (a) Choose a nonsingular \( T \in {M}_{n} \) such that \( {T}^{-1}A{T}^{- * } = I \) . Choose a unitary \( U \in {M}_{n} \) such that \( {U}^{ * }\left( {{T}^{-1}B{T}^{- * }}\right) U = \Lambda \) is diagonal. Let \( S = {TU} \) . Then \( {S}^{-1}A{S}^{- * } = \) \( {U}^{ * }{T}^{-1}A{T}^{- * }U = {U}^{ * }{IU} ...
Yes
Theorem 7.6.5. Let \( A, B \in {M}_{n} \) . If \( A \) is positive definite and \( B \) is complex symmetric, then there is a nonsingular \( S \in {M}_{n} \) such that \( A = {SI}{S}^{ * } \) and \( B = {S\Lambda }{S}^{T} \), in which \( \Lambda \) is nonnegative diagonal. The main diagonal entries of \( {\Lambda }^{2}...
Proof. Choose a nonsingular matrix \( R \in {M}_{n} \) such that \( {R}^{-1}A{R}^{- * } = I \) . Use (4.4.4(c)) to choose a unitary \( U \in {M}_{n} \) such that \( {U}^{ * }{R}^{-1}B{R}^{-T}\bar{U} = \Lambda \) is nonnegative diagonal. Let \( S = {RU} \) . Then \( {S}^{-1}A{S}^{- * } = {U}^{ * }{R}^{-1}A{R}^{- * }U = ...
Yes
Theorem 7.6.6. The function \( f\left( A\right) = \log \det A \) is a strictly concave function on the convex set of positive definite Hermitian matrices in \( {M}_{n} \) .
Proof. Let \( A, B \in {M}_{n} \) be positive definite. We must show that\n\n\[ \log \det \left( {{\alpha A} + \left( {1 - \alpha }\right) B}\right) \geq \alpha \log \det A + \left( {1 - \alpha }\right) \log \det B \]\n\n(7.6.7)\n\nfor all \( \alpha \in \left( {0,1}\right) \), with equality if and only if \( A = B \) ....
Yes
Theorem 7.6.10. The function \( f\left( A\right) = \operatorname{tr}{A}^{-1} \) is a strictly convex function on the convex set of positive definite Hermitian matrices in \( {M}_{n} \) .
Proof. Let \( A, B \in {M}_{n} \) be positive definite. We must show that\n\n\[ \operatorname{tr}{\left( \alpha A + \left( 1 - \alpha \right) B\right) }^{-1} \leq \alpha \operatorname{tr}{A}^{-1} + \left( {1 - \alpha }\right) \operatorname{tr}{B}^{-1} \]\n\nfor all \( \alpha \in \left( {0,1}\right) \), with equality if...
Yes
Theorem 7.7.2. Let \( A, B \in {M}_{n} \) be Hermitian and let \( S \in {M}_{n, m} \) . Then\n\n(a) if \( A \succcurlyeq B \), then \( {S}^{ * }{AS} \succcurlyeq {S}^{ * }{BS} \)\n\n(b) if \( \operatorname{rank}S = m \), then \( A \succ B \) implies \( {S}^{ * }{AS} \succ {S}^{ * }{BS} \)\n\n(c) if \( m = n \) and \( S...
Proof. (a) If \( \left( {A - B}\right) \succcurlyeq 0 \), then (7.1.8(a)) ensures that \( {S}^{ * }\left( {A - B}\right) S = {S}^{ * }{AS} - {S}^{ * }{BS} \) \( \succcurlyeq 0 \).\n\n(b) This assertion follows in the same way from (7.1.8(b)).\n\n(c) If \( {S}^{ * }{AS} \succ {S}^{ * }{BS} \), then \( {S}^{- * }\left( {...
Yes
Theorem 7.7.3. Let \( A, B \in {M}_{n} \) be Hermitian and suppose that \( A \) is positive definite.\n\n(a) If \( B \) is positive semidefinite, then \( A \succcurlyeq B \) (respectively, \( A \succ B \) ) if and only if \( \rho \left( {{A}^{-1}B}\right) \leq 1 \) (respectively, \( \rho \left( {{A}^{-1}B}\right) < 1 \...
Proof. (a) It follows from (a) and (d) of the preceding theorem that \( A \succcurlyeq B \) if and only if \( I = {A}^{-1/2}A{A}^{-1/2} \succcurlyeq {A}^{-1/2}B{A}^{-1/2} \) if and only if \( 1 \geq {\sigma }_{1}\left( {{A}^{-1/2}B{A}^{-1/2}}\right) \) . But \( {A}^{-1/2}B{A}^{-1/2} \) is positive semidefinite, so\n\n\...
Yes
Corollary 7.7.4. Let \( A, B \in {M}_{n} \) be Hermitian. Let \( {\lambda }_{1}\left( A\right) \leq \cdots \leq {\lambda }_{n}\left( A\right) \) and \( {\lambda }_{1}\left( B\right) \leq \cdots \leq {\lambda }_{n}\left( B\right) \) be the ordered eigenvalues of \( A \) and \( B \), respectively.\n\n(a) If \( A \succ 0 ...
Proof. (a) The preceding theorem ensures that \( A \succcurlyeq B \) if and only if \( \rho \left( {{A}^{-1}B}\right) = \rho \left( {B{A}^{-1}}\right) \leq 1 \) if and only if \( {B}^{-1} \succcurlyeq {A}^{-1} \) .
Yes
Lemma 7.7.6. Let \( X \in {M}_{p, q} \) and let \( K = \left\lbrack \begin{matrix} {I}_{p} & X \\ {X}^{ * } & {I}_{q} \end{matrix}\right\rbrack \in {M}_{p + q} \) . Then\n\n(a) \( K \) is positive definite if and only if \( X \) is a strict contraction\n\n(b) \( K \) is positive semidefinite if and only if \( X \) is a...
Proof. The identity (7.7.5) and (7.7.2(d)) ensure that \( \left\lbrack \begin{matrix} {I}_{p} & X \\ {X}^{ * } & {I}_{q} \end{matrix}\right\rbrack \succ 0 \) if and only if \( {I}_{q} - \) \( {X}^{ * }X \succ 0 \) if and only if \( {\sigma }_{1}\left( X\right) < 1 \) . The assertion in (b) follows in a similar fashion....
Yes
Theorem 7.7.7. Let \( H = \left\lbrack \begin{matrix} A & B \\ {B}^{ * } & C \end{matrix}\right\rbrack \in {M}_{p + q} \) be Hermitian with \( A \in {M}_{p} \) and \( C \in {M}_{q} \) . The following are equivalent:\n\n(a) \( H \) is positive definite.\n\n(b) \( A \) is positive definite and \( C - {B}^{ * }{A}^{-1}B \...
Proof. (a) \( \Leftrightarrow \) (b): Demonstrated in our discussion of (7.7.5).\n\n(b) \( \Rightarrow \) (c): Follows from (7.7.3(a)).\n\n(c) \( \Leftrightarrow \) (d): Let \( X = {A}^{-1/2}B{C}^{-1/2} \) . Then \( 1 > \rho \left( {{B}^{ * }{A}^{-1}B{C}^{-1}}\right) = \rho \left( {{C}^{-1/2}{B}^{ * }{A}^{-1}}\right. \...
Yes
Lemma 7.7.8. Let \( A \in {M}_{n} \) be positive semidefinite and singular, and let \( {A}_{k} = A + \) \( {k}^{-1}{I}_{n} \) for each \( k = 1,2,\ldots \) . Let \( {X}_{k} \in {M}_{m, n} \) be a contraction for each \( k = 1,2,\ldots \) . Then\n\n(a) each \( {A}_{k} \) is positive definite and \( \mathop{\lim }\limits...
Proof. (a) Let \( r = \operatorname{rank}A \), let \( {\lambda }_{1},\ldots ,{\lambda }_{r} \) be the positive eigenvalues of \( A \), and let \( A = \) \( U\left( {\operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{r}}\right) \oplus {0}_{n - r}}\right) {U}^{ * } \) be a spectral decomposition. Then continui...
Yes
Theorem 7.7.9. Let \( H = \left\lbrack \begin{matrix} A & B \\ {B}^{ * } & C \end{matrix}\right\rbrack \in {M}_{p + q} \) be Hermitian, with \( A \in {M}_{p} \) and \( C \in {M}_{q} \) . The following two statements are equivalent:\n\n(a) \( H \) is positive semidefinite.\n\n(b) A and \( C \) are positive semidefinite,...
Proof. (a) \( \Rightarrow \) (b): Consider \( {H}_{k} = H + {k}^{-1}{I}_{n} \) for each \( k = 1,2,\ldots \) Then \( {H}_{k} \) , \( {A}_{k} = A + {k}^{-1}{I}_{p} \), and \( {C}_{k} = C + {k}^{-1}{I}_{q} \) are positive definite for each \( k = 1,2,\ldots \) , so (7.7.7(e)) ensures that there is a contraction \( {X}_{k...
Yes
Corollary 7.7.10. Let \( A, C \in {M}_{p} \) be Hermitian.\n\n(a) If \( \left\lbrack \begin{array}{ll} A & {I}_{p} \\ {I}_{p} & C \end{array}\right\rbrack \succcurlyeq 0 \), then \( A \succ 0, C \succ 0, A \succcurlyeq {C}^{-1} \), and \( C \succcurlyeq {A}^{-1} \) .
Proof. (a) The hypothesis ensures that \( A \succcurlyeq 0 \) and \( C \succcurlyeq 0 \) . Theorem 7.7.9(b) ensures that there is a contraction \( X \) such that \( I = {A}^{1/2}X{C}^{1/2} \), so both \( {A}^{1/2} \) and \( {C}^{1/2} \) (and hence both \( A \) and \( C \) ) are nonsingular. Then \( A \succcurlyeq {C}^{...
Yes
Theorem 7.7.11. Let \( A \in {M}_{p} \) and \( C \in {M}_{q} \) be positive semidefinite and let \( B \in {M}_{p, q} \) . The following four statements are equivalent:\n\n(a) \( \left( {{x}^{ * }{Ax}}\right) \left( {{y}^{ * }{Cy}}\right) \geq {\left| {x}^{ * }By\right| }^{2} \) for all \( x \in {\mathbf{C}}^{p} \) and ...
Proof. (a) \( \Rightarrow \) (b): This implication follows from the arithmetic-geometric mean inequality: \( \frac{1}{2}\left( {{x}^{ * }{Ax} + {y}^{ * }{Cy}}\right) \geq {\left( {x}^{ * }Ax\right) }^{1/2}{\left( {y}^{ * }Cy\right) }^{1/2} \geq \left| {{x}^{ * }{By}}\right| \) . (b) \( \Rightarrow \) (c): Let \( z = {\...
Yes
Corollary 7.7.12. Let \( A \in {M}_{n} \) be positive semidefinite and let \( B \in {M}_{n} \) be Hermitian. The following four statements are equivalent:\n\n(a) \( {x}^{ * }{Ax} \geq \left| {{x}^{ * }{Bx}}\right| \) for all \( x \in {\mathbf{C}}^{n} \) .\n\n(b) \( {x}^{ * }{Ax} + {y}^{ * }{Ay} \geq 2\left| {{x}^{ * }{...
Proof. (a) \( \Rightarrow \) (b): Let \( x, y \in {\mathbf{C}}^{n} \) and use the inequality in (a), the triangle inequality, and Hermicity of \( B \) to compute\n\n\[ 2\left( {{x}^{ * }{Ax} + {y}^{ * }{Ay}}\right) = {\left( x + y\right) }^{ * }A\left( {x + y}\right) + {\left( x - y\right) }^{ * }A\left( {x - y}\right)...
Yes
Corollary 7.7.13. Let \( A, B \in {M}_{n} \) be positive semidefinite. The following statements are equivalent:\n\n(a) \( A \succcurlyeq B \) .\n\n(b) \( \left\lbrack \begin{array}{ll} A & B \\ B & A \end{array}\right\rbrack \succcurlyeq 0 \) .\n\n(c) There is a positive semidefinite contraction \( X \in {M}_{n} \) suc...
Proof. Since \( A \succcurlyeq B \) if and only if \( {x}^{ * }{Ax} \geq {x}^{ * }{Bx} \) for all \( x \in {\mathbf{C}}^{n} \), the asserted equivalences follow from (7.7.12) in the special case in which the Hermitian matrix \( B \) is positive semidefinite, provided we can choose \( X \) to be positive semidefinite in...
Yes
Corollary 7.7.14. Let \( A, B, C, D \in {M}_{n} \) be Hermitian and suppose that \( A \) and \( C \) are positive semidefinite. If \( {x}^{ * }{Ax} \geq \left| {{x}^{ * }{Bx}}\right| \) and \( {x}^{ * }{Cx} \geq \left| {{x}^{ * }{Dx}}\right| \) for all \( x \in {\mathbf{C}}^{n} \), then \( {x}^{ * }\left( {A \circ C}\r...
Proof. The hypotheses and the implication (a) \( \Rightarrow \) (c) in (7.7.12) ensure that \( \left\lbrack \begin{array}{ll} A & B \\ B & A \end{array}\right\rbrack \succcurlyeq \) 0 and \( \left\lbrack \begin{array}{ll} C & D \\ D & C \end{array}\right\rbrack \succcurlyeq 0 \), so (7.5.3) tells us that \( \left\lbrac...
Yes
Theorem 7.7.15. Let \( H \in {M}_{n} \) be positive definite and let \( \alpha \subset \{ 1,\ldots, n\} \) . Then \( {H}^{-1}\left\lbrack \alpha \right\rbrack \succcurlyeq {\left( H\left\lbrack \alpha \right\rbrack \right) }^{-1}. \)
Proof. Since a permutation congruence of a positive definite matrix is positive definite, we may assume that \( H = \left\lbrack \begin{matrix} A & B \\ {B}^{ * } & C \end{matrix}\right\rbrack ,\alpha = \{ 1,\ldots, k\} \), and \( H\left\lbrack \alpha \right\rbrack = A \) . The identity (0.7.3.1) ensures that \( {H}^{-...
Yes
Theorem 7.7.16. Let \( A \in {M}_{n} \) be positive definite. The following matrices are positive semidefinite and singular:\n\n(a) \( \left\lbrack \begin{matrix} A & X \\ {X}^{ * } & {X}^{ * }{A}^{-1}X \end{matrix}\right\rbrack \) for any \( X \in {M}_{n, m} \).\n\n(b) \( \left\lbrack \begin{matrix} A & {I}_{n} \\ {I}...
Proof. Use (7.7.9). In each case, we verify that for \( \left\lbrack \begin{matrix} A & B \\ {B}^{ * } & C \end{matrix}\right\rbrack \) we have \( A \) nonsingular and \( C = {B}^{ * }{A}^{-1}B \).\n\n(a) \( {X}^{ * }{A}^{-1}X - {X}^{ * }{A}^{-1}X = 0 \).\n\n(b) Take \( X = {I}_{n} \) in (a).\n\n(c) Take \( X = A \) in...
Yes
Theorem 7.7.17. Let \( A, B \in {M}_{n} \) be positive definite. Then\n\n(a) \( {A}^{-1} \circ {B}^{-1} \succcurlyeq {\left( A \circ B\right) }^{-1} \)
Proof. (a) The preceding theorem and the Schur product theorem ensure that\n\n\[ \left\lbrack \begin{matrix} A & {I}_{n} \\ {I}_{n} & {A}^{-1} \end{matrix}\right\rbrack \circ \left\lbrack \begin{matrix} B & {I}_{n} \\ {I}_{n} & {B}^{-1} \end{matrix}\right\rbrack = \left\lbrack \begin{matrix} A \circ B & {I}_{n} \\ {I}_...
Yes
Theorem 7.7.18. Let \( A, B \in {M}_{n} \) be positive definite. Then \( {\lambda }_{\min }\left( {A \circ B}\right) \geq \) \( \max \left\{ {{\lambda }_{\min }\left( {AB}\right) ,{\lambda }_{\min }\left( {A{B}^{T}}\right) }\right\} \) .
Proof. Since \( {\lambda }_{\min }\left( {{B}^{1/2}A{B}^{1/2}}\right) = {\lambda }_{\min }\left( {AB}\right) \), we have \( {B}^{1/2}A{B}^{1/2} \succcurlyeq {\lambda }_{\min }\left( {AB}\right) I \), which is equivalent to \( A \succcurlyeq {\lambda }_{\min }\left( {AB}\right) {B}^{-1} \) . It follows from (7.7.17(c)) ...
Yes
Theorem 7.8.1 (Hadamard’s inequality). Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be positive definite. Then\n\n\[ \det A \leq {a}_{11}\cdots {a}_{nn} \]\n\nwith equality if and only if \( A \) is diagonal.
Proof. Since \( A \) is positive definite, it has positive main diagonal entries and is diagonally *congruent to a correlation matrix. Let \( D = \operatorname{diag}\left( {{a}_{11}^{1/2},\ldots ,{a}_{nn}^{1/2}}\right) \) and define \( C = {D}^{-1}A{D}^{-1} \), which is also positive definite; it has unit diagonal entr...
Yes
Corollary 7.8.3 (Hadamard’s inequality). Let \( B \in {M}_{n} \) be nonsingular and partition \( B = \left\lbrack \begin{array}{lll} {b}_{1} & \ldots & {b}_{n} \end{array}\right\rbrack \) and \( {B}^{ * } = \left\lbrack \begin{array}{lll} {\beta }_{1} & \ldots & {\beta }_{n} \end{array}\right\rbrack \) according to the...
Proof. Apply (7.8.2) to the positive definite matrix \( A = {B}^{ * }B : \det A = {\left| \det B\right| }^{2} \), and the main diagonal entries of \( A \) are \( {\begin{Vmatrix}{b}_{1}\end{Vmatrix}}_{2}^{2},\ldots ,{\begin{Vmatrix}{b}_{n}\end{Vmatrix}}_{2}^{2} \) . The columns of \( B \) are orthogonal if and only if ...
Yes
Theorem 7.8.5 (Fischer's inequality). Suppose that the partitioned Hermitian matrix \[ H = \left\lbrack \begin{matrix} A & B \\ {B}^{ * } & C \end{matrix}\right\rbrack \in {M}_{p + q},\;A \in {M}_{p}\text{ and }C \in {M}_{q} \] is positive definite. Then \[ \det H \leq \left( {\det A}\right) \left( {\det C}\right) \]
Proof. Let \( A = {U\Lambda }{U}^{ * } \) and \( C = {V\Gamma }{V}^{ * } \) be spectral decompositions, in which \( U \) and \( V \) are unitary and \( \Lambda = \operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{p}}\right) \) and \( \Gamma = \operatorname{diag}\left( {{\gamma }_{1},\ldots ,{\gamma }_{q}}\ri...
Yes
Lemma 7.8.8. Let \( B \in {M}_{m} \) be positive definite. Let \( \alpha ,\beta \subset \{ 1,\ldots, m\} \) . Suppose that \( {\alpha }^{c} \) and \( {\beta }^{c} \) are nonempty and disjoint, and \( \alpha \cup \beta = \{ 1,\ldots, m\} \) . Then \( \det B\left\lbrack {{\alpha }^{c} \cup {\beta }^{c}}\right\rbrack \leq...
Proof. There is no loss of generality to assume that \( {\beta }^{c} = \{ 1,\ldots, k\} ,{\alpha }^{c} = \{ j,\ldots, m\} \) , and \( 1 < k < j < m \) . Then \( A\left\lbrack {\alpha }^{c}\right\rbrack \) and \( A\left\lbrack {\beta }^{c}\right\rbrack \) are complementary principal submatrices of \( A\left\lbrack {{\al...
Yes
Theorem 7.8.9 (Koteljanskii’s inequality). Let \( A \in {M}_{n} \) be positive definite and let \( \alpha ,\beta \subset \{ 1,\ldots, n\} \) . Then\n\n\[ \left( {\det A\left\lbrack {\alpha \cup \beta }\right\rbrack }\right) \left( {\det A\left\lbrack {\alpha \cap \beta }\right\rbrack }\right) \leq \left( {\det A\left\l...
Proof. There is no loss of generality to assume that \( \alpha \cup \beta = \{ 1,\ldots, n\} \) (if not, work within the principal submatrix \( A\left\lbrack {\alpha \cup \beta }\right\rbrack \) )). We may also assume that \( \alpha \cap \beta \) is nonempty (if it is empty, then \( \beta = {\alpha }^{c} \) and (7.8.10...
Yes
Theorem 7.8.11 (Szász’s inequality). Let \( A \in {M}_{n} \) be positive definite. Then\n\n\[ \n{P}_{k + 1}{\left( A\right) }^{{\left( \begin{matrix} n - 1 \\ k \end{matrix}\right) }^{-1}} \leq {P}_{k}{\left( A\right) }^{{\left( \begin{matrix} n - 1 \\ k - 1 \end{matrix}\right) }^{-1}}\;\text{ for each }\;k = 1,\ldots,...
Proof. The identity \( {A}^{-1} = {\left( \det A\right) }^{-1} \) adj \( A \) reminds us that each diagonal entry of \( {A}^{-1} \) is the ratio of a principal minor of \( A \) of size \( n - 1 \) and det \( A \) . Thus, an application of (7.8.2) to the positive definite matrix \( {A}^{-1} \) gives the inequality\n\n\[...
Yes
Lemma 7.8.15. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be positive semidefinite and partitioned as \( A = \) \( \left\lbrack \begin{matrix} {a}_{11} & {x}^{ * } \\ x & {A}_{22} \end{matrix}\right\rbrack \), in which \( {A}_{22} \in {M}_{n - 1} \) . Define\n\n\[ \alpha \left( A\right) = \left\{ \begi...
Proof. There is nothing to prove if \( A \) is singular, so assume that \( A \) is positive definite. Apply Sylvester's criterion (7.2.5) to the trailing principal minors. The trailing minors \( \det \left( {\widetilde{A}\left\lbrack {\{ k,\ldots, n\} }\right\rbrack }\right) = \det A\left\lbrack {\{ k,\ldots, n\} }\rig...
Yes
Theorem 7.8.19 (Ostrowski-Taussky inequality). Let \( H, K \in {M}_{n} \) be Hermitian and let \( A = H + {iK} \) . If \( H \) is positive definite, then\n\n\[ \det H \leq \left| {\det \left( {H + {iK}}\right) }\right| = \left| {\det A}\right| \]
Proof. We have \( A = H\left( {I + i{H}^{-1}K}\right) \), so (7.8.20) is equivalent to the inequality \( \left| {\det \left( {I + i{H}^{-1}K}\right) }\right| \geq 1 \) . Corollary 7.6.2(a) ensures that \( {H}^{-1}K \) is diagonalizable and has real eigenvalues \( {\lambda }_{1},\ldots ,{\lambda }_{n} \) . Then\n\n\[ \l...
Yes
Theorem 7.8.21 (Minkowski’s determinant inequality). Let \( A, B \in {M}_{n} \) be positive definite. Then\n\n\[ \n{\left( \det A\right) }^{1/n} + {\left( \det B\right) }^{1/n} \leq {\left( \det \left( A + B\right) \right) }^{1/n} \n\]\n\n(7.8.22)\n\nwith equality if and only if \( A = {cB} \) for some \( c > 0 \) .
Proof. Theorem 7.6.4 ensures that there is a nonsingular \( S \in {M}_{n} \) such that \( A = {SI}{S}^{ * } \) and \( B = {S\Lambda }{S}^{ * } \), in which \( \Lambda = \operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{n}}\right) \) is positive diagonal. The assertion (7.8.22) is that\n\n\[ \n{\left( \det S...
Yes
Theorem 7.8.27 (Fan’s determinant inequality). Let \( H, K \in {M}_{n} \) be Hermitian and let \( A = H + {iK} \) . If \( H \) is positive definite, then\n\n\[ \n{\left( \det H\right) }^{2/n} + {\left| \det K\right| }^{2/n} \leq {\left| \det \left( H + iK\right) \right| }^{2/n} = {\left| \det A\right| }^{2/n} \n\]\n\n(...
Proof. Theorem 7.6.4 ensures that there is a nonsingular \( S \in {M}_{n} \) such that \( H = {SI}{S}^{ * } \) and \( K = {S\Lambda }{S}^{ * } \), in which \( \Lambda = \operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{n}}\right) \) is real diagonal; its diagonal entries are the eigenvalues of the diagonali...
Yes
Proposition 8.1.8. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) and \( x = \left\lbrack {x}_{i}\right\rbrack \in {\mathbf{C}}^{n} \) be given.\n\n(a) \( \left| {Ax}\right| \leq \left| A\right| \left| x\right| \) .
Proof. (a) The assertion follows from the triangle inequality:\n\n\[ \n{\left| Ax\right| }_{k} = \left| {\mathop{\sum }\limits_{j}{a}_{kj}{x}_{j}}\right| \leq \mathop{\sum }\limits_{j}\left| {{a}_{kj}{x}_{j}}\right| = \mathop{\sum }\limits_{j}\left| {a}_{kj}\right| \left| {x}_{j}\right| = {\left( \left| A\right| \left|...
Yes
Theorem 8.1.18. Let \( A, B \in {M}_{n} \) and suppose that \( B \) is nonnegative. If \( \left| A\right| \leq \) \( B \), then \( \rho \left( A\right) \leq \rho \left( \left| A\right| \right) \leq \rho \left( B\right) \) .
Proof. Invoking (8.1.10) and (8.1.12), we have \( \left| {A}^{m}\right| \leq {\left| A\right| }^{m} \leq {B}^{m} \) for each \( m = \) \( 1,2,\ldots \) . Thus,(8.1.16) and (8.1.17) ensure that\n\n\[ \n{\begin{Vmatrix}{A}^{m}\end{Vmatrix}}_{2} \leq {\begin{Vmatrix}{\left| A\right| }^{m}\end{Vmatrix}}_{2} \leq {\begin{Vm...
Yes
Corollary 8.1.20. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative.\n\n(a) If \( \widetilde{A} \) is principal submatrix of \( A \), then \( \rho \left( \widetilde{A}\right) \leq \rho \left( A\right) \) .
Proof. (a) If \( r = n \), there is nothing to prove. Suppose that \( 1 \leq r < n \), let \( \widetilde{A} \) be an \( r \) -by- \( r \) principal square submatrix of \( A \), and let \( P \) be a permutation matrix such that \( {PA}{P}^{T} = \left\lbrack \begin{array}{ll} \widehat{A} & B \\ C & D \end{array}\right\rb...
Yes
Lemma 8.1.21. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative. Then \( \rho \left( A\right) \leq \parallel A{\parallel }_{\infty } = \) \( \mathop{\max }\limits_{{1 \leq i \leq n}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij} \) and \( \rho \left( A\right) \leq \parallel A{\parallel }_{1} = \m...
Proof. We know that \( \left| \lambda \right| \leq \rho \left( A\right) \leq \parallel A\parallel \) for any eigenvalue \( \lambda \) of \( A \) and any matrix norm \( \parallel \mid \cdot \parallel \mid \) . If all the row sums of \( A \) are equal, then \( e = {\left\lbrack \begin{array}{lll} 1 & \ldots & 1 \end{arra...
Yes
Theorem 8.1.22. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative. Then\n\n\[ \mathop{\min }\limits_{{1 \leq i \leq n}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij} \leq \rho \left( A\right) \leq \mathop{\max }\limits_{{1 \leq i \leq n}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij} \]\n\n(8.1.23)\...
Proof. Let \( \alpha = \mathop{\min }\limits_{{1 \leq i \leq n}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij} \) . If \( \alpha = 0 \), let \( B = 0 \) . If \( \alpha > 0 \), define \( B = \left\lbrack {b}_{ij}\right\rbrack \) by letting each \( {b}_{ij} = \alpha {a}_{ij}{\left( \mathop{\sum }\limits_{{k = 1}}^{n}{a}_{ik...
Yes
Corollary 8.1.29. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative and let \( x = \left\lbrack {x}_{i}\right\rbrack \in {\mathbf{R}}^{n} \) be a positive vector. If \( \alpha ,\beta \geq 0 \) are such that \( {\alpha x} \leq {Ax} \leq {\beta x} \), then \( \alpha \leq \rho \left( A\right) \le...
Proof. If \( {\alpha x} \leq {Ax} \), then \( \alpha {x}_{i} \leq {\left( Ax\right) }_{i} \) and \( \alpha \leq \mathop{\min }\limits_{{1 \leq i \leq n}}{x}_{i}^{-1}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij}{x}_{j} \), so the preceding theorem ensures that \( \alpha \leq \rho \left( A\right) \) . If \( {\alpha x} < {A...
Yes
Corollary 8.1.30. Let \( A \in {M}_{n} \) be nonnegative. If \( x \) is a positive eigenvector of \( A \), then \( \rho \left( A\right), x \) is an eigenpair for \( A \) ; that is, if \( A \geq 0, x > 0 \), and \( {Ax} = {\lambda x} \), then \( \lambda = \rho \left( A\right) \) .
Proof. If \( x > 0 \) and \( {Ax} = {\lambda x} \), then \( \lambda \geq 0 \) and \( {\lambda x} \leq {Ax} \leq {\lambda x} \) . But then (8.1.29) ensures that \( \lambda \leq \rho \left( A\right) \leq \lambda \) .
Yes
Corollary 8.1.31. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative. If \( A \) has a positive eigenvector, then\n\n\[ \rho \left( A\right) = \mathop{\max }\limits_{{x > 0}}\mathop{\min }\limits_{{1 \leq i \leq n}}\frac{1}{{x}_{i}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij}{x}_{j} = \mathop{\m...
Exercise. Prove the preceding corollary. Hint: Use the positive eigenvector \( x \) in (8.1.27).
No
Corollary 8.1.33. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be nonnegative and write \( {A}^{m} = \left\lbrack {a}_{ij}^{\left( m\right) }\right\rbrack \) . If \( A \) has a positive eigenvector \( x = \left\lbrack {x}_{i}\right\rbrack \), then for all \( m = 1,2,\ldots \) and for all \( i = 1,\ldots...
Proof. Let \( x = \left\lbrack {x}_{i}\right\rbrack \) be a positive eigenvector of \( A \) . Then (8.1.30) ensures that \( {Ax} = \) \( \rho \left( A\right) x \), so \( {A}^{m}x = \rho {\left( A\right) }^{m}x \) for each \( m = 1,2,\ldots \) . Since \( {A}^{m} \geq 0 \), for any \( i = 1,\ldots, n \) we have\n\n\[ \rh...
Yes
Lemma 8.2.1. Let \( A \in {M}_{n} \) be positive. If \( \lambda, x \) is an eigenpair of \( A \) and \( \left| \lambda \right| = \rho \left( A\right) \) , then \( \left| x\right| > 0 \) and \( A\left| x\right| = \rho \left( A\right) \left| x\right| \) .
Proof. The hypotheses ensure that \( z = A\left| x\right| > 0 \) (8.1.14). We have \( z = A\left| x\right| \geq \left| {Ax}\right| = \) \( \left| {\lambda x}\right| = \left| \lambda \right| \left| x\right| = \rho \left( A\right) \left| x\right| \), so \( y = z - \rho \left( A\right) \left| x\right| \geq 0 \) . If \( y ...
Yes
Theorem 8.2.2. If \( A \in {\mathbf{M}}_{n} \) is positive, there are positive vectors \( x \) and \( y \) such that \( {Ax} = \rho \left( A\right) x \) and \( {y}^{T}A = \rho \left( A\right) {y}^{T}. \)
Proof. There is an eigenpair \( \lambda, x \) of \( A \) with \( \left| \lambda \right| = \rho \left( A\right) \) . The preceding lemma ensures that \( \rho \left( A\right) ,\left| x\right| \) is also an eigenpair of \( A \) and \( \left| x\right| > 0 \) . The assertion about \( y \) follows from considering \( {A}^{T}...
Yes
Lemma 8.2.3. Let \( A \in {M}_{n} \) be positive. If \( \lambda, x \) is an eigenpair of \( A \) and \( \left| \lambda \right| = \rho \left( A\right) \) , then there is a \( \theta \in \mathbf{R} \) such that \( {e}^{-{i\theta }}x = \left| x\right| > 0 \) .
Proof. The hypothesis is that \( x \in {\mathbf{C}}^{n} \) is nonzero and \( \left| {Ax}\right| = \left| {\lambda x}\right| = \rho \left( A\right) \left| x\right| \) ; (8.2.1) ensures that \( A\left| x\right| = \rho \left( A\right) \left| x\right| \) and \( \left| x\right| > 0 \) . Since \( \left| {Ax}\right| = \rho \l...
Yes
Theorem 8.2.4. Let \( A \in {M}_{n} \) be positive. If \( \lambda \) is an eigenvalue of \( A \) and \( \lambda \neq \rho \left( A\right) \), then \( \left| \lambda \right| < \rho \left( A\right) \)
Proof. Let \( \lambda, x \) be an eigenpair of \( A \), so \( \left| \lambda \right| \leq \rho \left( A\right) \) . If \( \left| \lambda \right| = \rho \left( A\right) \) ,(8.2.3) ensures that \( w = {e}^{-{i\theta }}x > 0 \) for some \( \theta \in \mathbf{R} \) . Since \( {Aw} = {\lambda w} \) and \( w > 0 \), it foll...
Yes
Theorem 8.2.5. If \( A \in {M}_{n} \) is positive, then the geometric multiplicity of \( \rho \left( A\right) \) as an eigenvalue of \( A \) is 1 .
Proof. Suppose that \( w, z \in {\mathbf{C}}^{n} \) are nonzero vectors such that \( {Aw} = \rho \left( A\right) w \) and \( {Az} = \rho \left( A\right) z \) . Then \( w = {\alpha z} \) for some \( \alpha \in \mathbf{C} \) . Lemma 8.2.3 ensures that there are real numbers \( {\theta }_{1} \) and \( {\theta }_{2} \) suc...
Yes
Theorem 8.2.7. Let \( A \in {M}_{n} \) be positive. The algebraic multiplicity of \( \rho \left( A\right) \) as an eigenvalue of \( A \) is \( 1 \) . If \( x \) and \( y \) are the right and left Perron vectors of \( A \), then \( \mathop{\lim }\limits_{{m \rightarrow \infty }}{\left( \rho {\left( A\right) }^{-1}A\righ...
Proof. We know that \( \rho \left( A\right) > 0 \), and that \( x \) and \( y \) are positive vectors such that \( {Ax} = \) \( \rho \left( A\right) x,{y}^{T}A = \rho \left( A\right) {y}^{T} \), and \( {y}^{ * }x = {y}^{T}x = 1 \) . Theorem 1.4.12b ensures that \( \rho \left( A\right) \) has algebraic multiplicity 1, a...
Yes
Theorem 8.2.9 (Fan). Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) . Suppose that \( B = \left\lbrack {b}_{ij}\right\rbrack \in {M}_{n} \) is nonnegative and \( {b}_{ij} \geq \left| {a}_{ij}\right| \) for all \( i \neq j \) . Then every eigenvalue of \( A \) is in the union of \( n \) discs\n\n\[ \mathop...
Proof. First, assume that \( B > 0 \) . Theorem 8.2.8 ensures that there is a positive vector \( x \) such that \( {Bx} = \rho \left( B\right) x \), and hence\n\n\[ \mathop{\sum }\limits_{{j \neq i}}\left| {a}_{ij}\right| {x}_{j} \leq \mathop{\sum }\limits_{{j \neq i}}{b}_{ij}{x}_{j} = \rho \left( B\right) {x}_{i} - {b...
Yes
Theorem 8.3.1. If \( A \in {M}_{n} \) is nonnegative, then \( \rho \left( A\right) \) is an eigenvalue of \( A \) and there is a nonnegative nonzero vector \( x \) such that \( {Ax} = \rho \left( A\right) x \) .
Proof. For any \( \epsilon > 0 \), define \( A\left( \epsilon \right) = A + \epsilon {J}_{n} \) . Let \( x\left( \epsilon \right) = \left\lbrack {x{\left( \epsilon \right) }_{i}}\right\rbrack \) be the Perron vector of \( A\left( \epsilon \right) \), so \( x\left( \epsilon \right) > 0 \) and \( \mathop{\sum }\limits_{{...
Yes
Theorem 8.3.2. Let \( A \in {M}_{n} \) be nonnegative, and let \( x \in {\mathbf{R}}^{n} \) be nonnegative and nonzero. If \( \alpha \in \mathbf{R} \) and \( {Ax} \geq {\alpha x} \), then \( \rho \left( A\right) \geq \alpha \) .
Proof. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \), let \( \epsilon > 0 \), and define \( A\left( \epsilon \right) = A + \epsilon {J}_{n} > 0 \) . Then \( A\left( \epsilon \right) \) has a positive left Perron vector \( y\left( \epsilon \right) : y{\left( \epsilon \right) }^{T}A\left( \epsilon \right) = \rho \left...
Yes
Corollary 8.3.3. If \( A \in {M}_{n} \) is nonnegative, then\n\n\[ \rho \left( A\right) = \mathop{\max }\limits_{\substack{{x \geq 0} \\ {x \neq 0} }}\mathop{\min }\limits_{\substack{{1 \leq i \leq n} \\ {{x}_{i} \neq 0} }}\frac{1}{{x}_{i}}\mathop{\sum }\limits_{{j = 1}}^{n}{a}_{ij}{x}_{j} \]\n\n(8.3.3a)
Proof. Let \( x \) be any nonzero nonnegative vector and let \( \alpha = \mathop{\min }\limits_{{{x}_{i} \neq 0}}\mathop{\sum }\limits_{j}{a}_{ij}{x}_{j}/{x}_{i} \) . Then \( {Ax} \geq {\alpha x} \), so the preceding theorem ensures that \( \rho \left( A\right) \geq a \), and hence\n\n\[ \rho \left( A\right) \geq \math...
Yes
Theorem 8.3.4. Let \( A \in {M}_{n} \) be nonnegative. Suppose that there is a positive vector \( x \) and a nonnegative real number \( \lambda \) such that either \( {Ax} = {\lambda x} \) or \( {x}^{T}A = \lambda {x}^{T} \) . Then \( \lambda = \rho \left( A\right) \) .
Proof. Suppose that \( x = \left\lbrack {x}_{i}\right\rbrack \in {\mathbf{R}}^{n} \) and \( {Ax} = {\lambda x} \) . Let \( D = \operatorname{diag}\left( {{x}_{1},\ldots ,{x}_{n}}\right) \) and define \( B = {D}^{-1}{AD} \), which has the same eigenvalues as \( A \) . Then \( {Be} = {D}^{-1}{ADe} = \) \( {D}^{-1}{Ax} = ...
Yes
Theorem 8.3.5. Suppose that \( A \in {M}_{n} \) is nonnegative and has a positive left eigenvector.\n\n(a) If \( x \in {\mathbf{R}}^{n} \) is nonzero and \( {Ax} \geq \rho \left( A\right) x \), then \( x \) is an eigenvector of \( A \) corresponding to the eigenvalue \( \rho \left( A\right) \) .
Proof. Let \( y \) be a positive left eigenvector of \( A \) . The preceding theorem ensures that \( {A}^{T}y = \rho \left( A\right) y. \) (a) We know that \( x \neq 0 \) and \( {Ax} - \rho \left( A\right) x \geq 0 \) . We need to show that \( {Ax} - \rho \left( A\right) x = \) 0 . If \( {Ax} - \rho \left( A\right) x \...
Yes
Lemma 8.4.2. Let \( {\lambda }_{1},\ldots ,{\lambda }_{n} \) be the eigenvalues of \( A \in {M}_{n} \) . Then \( {\lambda }_{1} + 1,\ldots ,{\lambda }_{n} + 1 \) are the eigenvalues of \( I + A \) and \( \rho \left( {I + A}\right) \leq \rho \left( A\right) + 1 \) . If \( A \) is nonnegative, then \( \rho \left( {I + A}...
Proof. The first assertion is a consequence of (2.4.2). We have \( \rho \left( {I + A}\right) = \) \( \mathop{\max }\limits_{{1 \leq i \leq n}}\left| {{\lambda }_{i} + 1}\right| \leq \mathop{\max }\limits_{{1 \leq i \leq n}}\left| {\lambda }_{i}\right| + 1 = \rho \left( A\right) + 1 \) . However,(8.3.1) ensures that \(...
Yes
Lemma 8.4.3. If \( A \in {M}_{n} \) is nonnegative and \( {A}^{m} \) is positive for some \( m \geq 1 \), then \( \rho \left( A\right) \) is the only maximum-modulus eigenvalue of \( A \) ; it is positive and algebraically simple.
Proof. Let \( {\lambda }_{1},\ldots ,{\lambda }_{n} \) be the eigenvalues of \( A \) . Then \( {\lambda }_{1}^{m},\ldots ,{\lambda }_{n}^{m} \) are the eigenvalues of \( {A}^{m} \) . Theorem 8.2.8 ensures that exactly one of \( {\lambda }_{1}^{m},\ldots ,{\lambda }_{n}^{m} \) is equal to \( \rho \left( {A}^{m}\right) =...
Yes
Theorem 8.4.4 (Perron-Frobenius). Let \( A \in {M}_{n} \) be irreducible and nonnegative, and suppose that \( n \geq 2 \) . Then\n\n(a) \( \rho \left( A\right) > 0 \)\n\n(b) \( \rho \left( A\right) \) is an algebraically simple eigenvalue of \( A \)\n\n(c) there is a unique real vector \( x = \left\lbrack {x}_{i}\right...
Proof. (a) Corollary 8.1.25 shows that \( \rho \left( A\right) > 0 \) under conditions even weaker than irreducibility.\n\n(b) If \( \rho \left( A\right) \) is a multiple eigenvalue of \( A \), then (8.4.2) ensures that \( \rho \left( A\right) + 1 = \rho (I + \) \( A) \) is a multiple eigenvalue of \( I + A \) and henc...
Yes
Theorem 8.4.5. Let \( A, B \in {M}_{n} \) . Suppose that \( A \) is nonnegative and irreducible, and \( A \geq \left| B\right| \) . Let \( \lambda = {e}^{i\varphi }\rho \left( B\right) \) be a given maximum-modulus eigenvalue of \( B \) . If \( \rho \left( A\right) = \) \( \rho \left( B\right) \), then there is a diago...
Proof. Let \( x \) be a nonzero vector such that \( {Bx} = {\lambda x} \), and let \( \rho = \rho \left( A\right) = \rho \left( B\right) \) . Then\n\n\[ \rho \left| x\right| = \left| {\lambda x}\right| = \left| {Bx}\right| \leq \left| B\right| \left| x\right| \overset{\left( \alpha \right) }{ \leq }A\left| x\right| \]\...
Yes
Corollary 8.4.7. Suppose that \( A \in {M}_{n} \) is irreducible and nonnegative. If \( A \) has \( k > 1 \) eigenvalues of maximum modulus, then every main diagonal entry of \( A \) is zero. Moreover, every main diagonal entry of \( {A}^{m} \) is zero for each positive integer \( m \) that is not divisible by \( k \) ...
Proof. Let \( \varphi = {2\pi }/k \) . Corollary 8.4.6a ensures that \( A \) is similar to \( {e}^{i\varphi }A \), so \( {A}^{m} \) is similar to \( {e}^{im\varphi }A \) for each \( m = 1,2,3,\ldots \) and \( \operatorname{tr}{A}^{m} = {e}^{im\varphi }\operatorname{tr}{A}^{m} \) . Since \( {e}^{im\varphi } \) is real a...
Yes
Theorem 8.5.1. If \( A \in {M}_{n} \) is nonnegative and primitive, and if \( x \) and \( y \) are, respectively, the right and left Perron vectors of \( A \), then \( \mathop{\lim }\limits_{{m \rightarrow \infty }}{\left( \rho {\left( A\right) }^{-1}A\right) }^{m} = x{y}^{T} \), which is a positive rank-one matrix.
Proof. We have in hand all the ingredients required in the proof of (8.2.7): \( \rho \left( A\right) \) is a simple eigenvalue with positive associated right and left eigenvectors \( x \) and \( y \) such that \( {x}^{T}y = 1 \) . We can perform the factorization (8.2.7a), in which every eigenvalue of \( B \) has modul...
Yes
Theorem 8.5.2. If \( A \in {M}_{n} \) is nonnegative, then \( A \) is primitive if and only if \( {A}^{m} > 0 \) for some \( m \geq 1 \) .
Proof. If \( {A}^{m} \) is positive, there is a directed path of length \( m \) between every pair of nodes of the directed graph \( \Gamma \left( A\right) \) of \( A \), so \( \Gamma \left( A\right) \) is strongly connected and \( A \) is irreducible. In addition,(8.4.3) ensures that there are no maximum-modulus eigen...
Yes
Theorem 8.5.3. Let \( A \in {M}_{n} \) be irreducible and nonnegative, and let \( {P}_{1},\ldots ,{P}_{n} \) be the nodes of the directed graph \( \Gamma \left( A\right) \) . Let \( {L}_{i} = \left\{ {{k}_{1}^{\left( i\right) },{k}_{2}^{\left( i\right) },\ldots }\right\} \) be the set of lengths of all directed paths i...
Proof. Irreducibility of \( A \) implies that no set \( {L}_{i} \) is empty: For each \( i \) and for any \( j \neq i \) , there is a path in \( \Gamma \left( A\right) \) that joins \( {P}_{i} \) to \( {P}_{j} \) ; there is also a path in \( \Gamma \left( A\right) \) that joins \( {P}_{j} \) to \( {P}_{i} \) . If \( A ...
Yes
Lemma 8.5.4. If \( A \in {M}_{n} \) is irreducible and nonnegative, and if all its main diagonal entries are positive, then \( {A}^{n - 1} > 0 \), so \( A \) is primitive.
Proof. If every main diagonal entry of \( A \) is positive, let \( \alpha = \min \left\{ {{a}_{11},\ldots ,{a}_{nn}}\right\} > 0 \) and define \( B = A - \operatorname{diag}\left( {{a}_{11},\ldots ,{a}_{nn}}\right) \) . Then \( B \) is nonnegative and irreducible (because \( A \) is irreducible), and \( A \geq {\alpha ...
Yes
Lemma 8.5.5. Let \( A \in {M}_{n} \) be nonnegative and primitive. Then \( {A}^{m} \) is nonnegative and primitive for every integer \( m \geq 1 \) .
Proof. Since all sufficiently large powers of \( A \) are positive, the same is true for \( {A}^{m} \) for any \( m \) . If \( {A}^{m} \) were reducible, then \( {A}^{mp} \) would be reducible for all \( p = 2,3,\ldots \), and hence these matrices cannot be positive. This contradiction shows that no power of \( A \) ca...
No
Theorem 8.5.6. Let \( A \in {M}_{n} \) be nonnegative. If \( A \) is primitive, then \( {A}^{k} > 0 \) for some positive integer \( k \leq \left( {n - 1}\right) {n}^{n} \) .
Proof. Because \( A \) is irreducible, there is a directed path from the node \( {P}_{1} \) in \( \Gamma \left( A\right) \) back to itself; let \( {k}_{1} \) be the shortest such path, so that \( {k}_{1} \leq n \) . The matrix \( {A}^{{k}_{1}} \) has a positive entry in its 1,1 position, and any power of \( {A}^{{k}_{1...
Yes
Corollary 8.5.8 (Wielandt). Let \( A \in {M}_{n} \) be nonnegative. Then \( A \) is primitive if and only if \( {A}^{{n}^{2} - {2n} + 2} > 0 \) .
Proof. If some power of \( A \) is positive, then \( A \) is primitive, so only the converse implication is of interest. If \( n = 1 \), the result is trivial, so assume that \( n > 1 \) . If \( A \) is primitive, then it is irreducible and there are cycles in \( \Gamma \left( A\right) \) . If the shortest cycle in \( ...
Yes
Theorem 8.5.9. Let \( A \in {M}_{n} \) be irreducible and nonnegative, and suppose that \( A \) has \( d \) positive main diagonal entries, \( 1 \leq d \leq n \) . Then \( {A}^{{2n} - d - 1} > 0 \) ; that is, \( \gamma \left( A\right) \leq \) \( {2n} - d - 1 \) .
Proof. Under the stated hypotheses, \( A \) must be primitive, and \( \Gamma \left( A\right) \) has \( d \) cycles with (minimum) length one. We may assume that \( {P}_{1},\ldots ,{P}_{d} \) are the nodes in \( \Gamma \left( A\right) \) that have loops. Consider \( {A}^{{2n} - d - 1} = {A}^{n - d}{\left( {A}^{1}\right)...
Yes
Theorem 8.6.1. Let \( A \in {M}_{n} \) be irreducible and nonnegative, let \( n \geq 2 \), and let \( x \) and \( y \), respectively, be the right and left Perron vectors of \( A \) . Then \[ \mathop{\lim }\limits_{{N \rightarrow \infty }}\frac{1}{N}\mathop{\sum }\limits_{{m = 1}}^{N}{\left( \rho {\left( A\right) }^{-1...
Proof. If \( A \) is primitive, \( \rho {\left( A\right) }^{-1}A \) can be factored as in (8.2.7a), in which \( x \) is the first column of \( S \) and \( y \) is the first column of \( {S}^{-1} \) . We have \[ \frac{1}{N}\mathop{\sum }\limits_{{m = 1}}^{N}{\left( \rho {\left( A\right) }^{-1}A\right) }^{m} = S\left\lbr...
Yes
Lemma 8.7.1. Let \( A = \left\lbrack {a}_{ij}\right\rbrack \in {M}_{n} \) be a doubly stochastic matrix that is not the identity matrix. There is a permutation \( \sigma \) of \( \{ 1,\ldots, n\} \) that is not the identity permutation and is such that \( {a}_{{1\sigma }\left( 1\right) }\cdots {a}_{{n\sigma }\left( n\r...
Proof. Suppose that every permutation \( \sigma \) of \( \{ 1,\ldots, n\} \) that is not the identity permutation \( {\sigma }_{0} \) has the property that \( {a}_{{1\sigma }\left( 1\right) }\cdots {a}_{{n\sigma }\left( n\right) } = 0 \) . This assumption and (0.3.2.1) permit us to compute the characteristic polynomial...
Yes
Theorem 8.7.2 (Birkhoff). A matrix \( A \in {M}_{n} \) is doubly stochastic if and only if there are permutation matrices \( {P}_{1},\ldots ,{P}_{N} \in {M}_{n} \) and positive scalars \( {t}_{1},\ldots ,{t}_{N} \in \mathbf{R} \) such that \( {t}_{1} + \cdots + {t}_{N} = 1 \) and\n\n\[ A = {t}_{1}{P}_{1} + \cdots + {t}...
Proof. The sufficiency of the representation (8.7.3) is clear; we prove its necessity by exhibiting an algorithm that constructs it in finitely many steps.\n\nIf \( A \) is a permutation matrix, there is nothing to prove. If not, the preceding lemma ensures that there is a nonidentity permutation \( \sigma \) of \( \{ ...
Yes
Corollary 8.7.4. The maximum (respectively, minimum) of a convex (respectively, concave) real-valued function on the set of doubly stochastic \( n \) -by- \( n \) matrices is attained at a permutation matrix.
Proof. Let \( f \) be a convex real-valued function on the set of \( n \) -by- \( n \) doubly stochastic matrices, let \( A \) be a doubly stochastic matrix at which \( f \) attains its maximum value, represent \( A = {t}_{1}{P}_{1} + \cdots + {t}_{N}{P}_{N} \) as a convex combination of permutation matrices, and let \...
Yes
Lemma 8.7.5. Let \( A \in {M}_{n} \) be doubly substochastic. There is a doubly stochastic matrix \( S \in {M}_{n} \) such that \( A \leq S \) .
Proof. For any doubly substochastic matrix \( S \in {M}_{n} \), let \( N\left( S\right) \) denote the number of row sums and column sums of \( S \) that are less than one, that is, the number of entries of the vectors \( {Ae} \) and \( {A}^{T}e \) whose entries are less than one.\n\nLet \( A \in {M}_{n} \) be doubly su...
Yes
Theorem 1.1 If \( \left( {P, \leq }\right) \) is a partially ordered set, then the relation\n\n\[ a < b\;\text{ if }\;a \leq b, a \neq b \]\n\nis a strict order on \( P \) . Conversely, if \( < \) is a strict order on \( P \), then the relation\n\n\[ a \leq b\;\text{ if }\;a < b\text{ or }a = b \]\n\nis a partial order...
\[ ▱ \]
No
Theorem 1.4 Let \( P \) be a poset.\n\n1) P has the ACC if and only if it has the maximal condition.\n\n2) P has the DCC if and only if it has the minimal condition.
Proof. Suppose \( P \) has the ACC and let \( S \subseteq P \) be nonempty. Let \( {s}_{1} \in S \) . If \( {s}_{1} \) is maximal we are done. If not, then we can pick \( {s}_{2} \in S \) such that \( {s}_{2} > {s}_{1} \) . Continuing in this way, we either arrive at a maximal element in \( S \) or we get a strictly in...
Yes
Theorem 1.5 Let \( P \) be a poset.\n\n1) The following are equivalent:\n\na) P has no infinite chains.\n\nb) \( P \) has both chain conditions.\n\nIf these conditions hold, then for any \( a < b \) in \( P \), there is a maximal finite chain from a to \( b \) .
Proof. It is clear that 1a) implies 1b). For the converse, suppose that \( P \) has BCC and let \( \mathcal{C} \) be an infinite chain. The ACC implies that \( \mathcal{C} \) has a maximal element \( {x}_{1} \), which must be maximum in \( \mathcal{C} \) since \( \mathcal{C} \) is totally ordered. Then \( \mathcal{C} \...
Yes
Theorem 1.7 Suppose that \( \left( {P, \leq }\right) \) is a partially ordered set for which every subset of \( P \) has a meet. Then \( \left( {P, \leq }\right) \) is a complete lattice, where the join of a subset \( S \) of \( P \) is the meet of all upper bounds for \( S \) .
Proof. First, note that the meet of the empty set is the maximum element of \( P \) and the meet of \( P \) is the minimum element of \( P \) and so \( P \) is bounded. In particular, the join of \( \varnothing \) exists.\n\nLet \( S \) be a nonempty subset of \( P \) . The family \( U \) of upper bounds for \( S \) is...
Yes
Theorem 1.18 Let \( p \) be a prime and let \( d \geq 1 \) . Let \( {o}_{p}\left( n\right) \) denote the largest exponent \( e \) for which \( {p}^{e} \) divides \( n \) .\n\n1) For \( 1 \leq k \leq {p}^{d} \) ,\n\n\[ \n{o}_{p}\left\lbrack \left( \begin{matrix} {p}^{d} \\ k \end{matrix}\right) \right\rbrack = d - {o}_{...
Proof. For part 1), write\n\n\[ \n\left( \begin{matrix} {p}^{d} \\ k \end{matrix}\right) = \frac{{p}^{d}}{k}\frac{\left( {p}^{d} - 1\right) }{1}\cdots \frac{\left( {p}^{d} - u\right) }{u}\cdots \frac{{p}^{d} - \left( {k - 1}\right) }{k - 1} \n\]\n\nwhere \( 1 \leq u \leq k - 1 \) . Now, if \( 1 \leq i \leq d \), then \...
Yes
Theorem 2.1 Let \( G \) be a group. If \( a \in G \) has finite order \( o\left( a\right) \), then the exponents of a are precisely the integral multiples of \( o\left( a\right) \) .
Proof. Let \( n = o\left( a\right) \) . Any integral multiple of \( n \) is clearly an exponent of \( a \) . Conversely, if \( {a}^{m} = 1 \), then \( m = {qn} + r \) where \( 0 \leq r < n \) . Hence,\n\n\[ \n1 = {a}^{m} = {a}^{{qn} + r} = {a}^{qn}{a}^{r} = {a}^{r} \n\] \n\nand so the minimality of \( n \) implies that...
Yes
Theorem 2.10 Let \( G \) be a group and \( a, b \in G \). 1) If \( o\left( a\right) = n \), then for \( 1 \leq k < n \), \[ o\left( {a}^{k}\right) = \frac{n}{\gcd \left( {n, k}\right) } \] In particular, \[ \langle a\rangle = \left\langle {a}^{k}\right\rangle \; \Leftrightarrow \;\gcd \left( {o\left( a\right), k}\right...
Proof. For part 1), Theorem 2.1 implies that \( {\left( {a}^{k}\right) }^{m} = 1 \) if and only if \( n \mid {km} \). But \[ n \mid {km}\; \Leftrightarrow \;\frac{n}{\gcd \left( {n, k}\right) }\left| {\frac{km}{\gcd \left( {n, k}\right) }\; \Leftrightarrow \;\frac{n}{\gcd \left( {n, k}\right) }}\right| m \] and so the ...
Yes
Corollary 2.11 Let \( G \) be a group and let \( a \in G \) . If \( o\left( a\right) = {uv} \), where \( \left( {u, v}\right) = 1 \) , then\n\n\[ a = {a}_{1}{a}_{2} \]\n\nwhere \( {a}_{1},{a}_{2} \in \langle a\rangle \) and \( o\left( {a}_{1}\right) = u \) and \( o\left( {a}_{2}\right) = v \) .
Proof. Since \( \left( {u, v}\right) = 1 \), there exist integers \( s \) and \( t \) for which \( {su} + {tv} = 1 \) . Hence,\n\n\[ a = {a}^{{su} + {tv}} = {a}^{su}{a}^{tv} \]\n\nThen \( \left( {{su}, v}\right) = 1 \) implies that \( o\left( {a}^{su}\right) = v \) and similarly, \( o\left( {a}^{tv}\right) = u.▱
Yes
Let \( {S}_{n} \) be the symmetric group.\n\n1) Let \( \sigma \in {S}_{n} \) . For any \( k \) -cycle \( \left( {{a}_{1}\cdots {a}_{k}}\right) \), we have\n\n\[{\left( {a}_{1}\cdots {a}_{k}\right) }^{\sigma } = \left( {\sigma {a}_{1}\cdots \sigma {a}_{k}}\right)\]\n\nHence, if \( \tau = {c}_{1}\cdots {c}_{k} \) is a cy...
Proof. For part 1), we have\n\n\[{\left( {a}_{1}\cdots {a}_{k}\right) }^{\sigma }\left( {\sigma {a}_{i}}\right) = \left\{ \begin{array}{ll} \sigma {a}_{i + 1} & i < k \\ \sigma {a}_{1} & i = k \end{array}\right.\]\n\nAlso, if \( b \neq \sigma {a}_{i} \) for any \( i \), then \( {\sigma }^{-1}b \neq {a}_{i} \) and so\n\...
Yes
The set \( {A}_{n} \) of all even permutations in \( {S}_{n} \) is a subgroup of \( {S}_{n} \).
To see that \( {A}_{n} \) is closed under the product, if \( \sigma \) and \( \tau \) are even, then they can each be written as a product of an even number of transpositions. Hence, \( {\sigma \tau } \) is also a product of an even number of transpositions and so is in \( {A}_{n} \). The subgroup \( {A}_{n} \) is call...
No
Theorem 2.19 Let \( G \) be a group and let \( A, B, C \leq G \) with \( A \leq B \). 1) (Dedekind law) \[ A\left( {B \cap C}\right) = B \cap {AC} \] 2) \[ A \cap C = B \cap C\text{ and }{AC} = {BC}\; \Rightarrow \;A = B \]
Proof. We leave proof of the Dedekind law to the reader. For part 2), \[ A = A\left( {A \cap C}\right) = A\left( {B \cap C}\right) = B \cap {AC} = B \cap {BC} = B \]
No
Theorem 2.20 Let \( X \) be a nonempty subset of a group \( G \) and let \( {X}^{-1} = \) \( \left\{ {{x}^{-1} \mid x \in X}\right\} \) . Let\n\n\[ \nW = {\left( X \cup {X}^{-1}\right) }^{ * } \n\]\n\nbe the set of all words over the alphabet \( X \cup {X}^{-1} \) .\n\n1) If we interpret juxtaposition in \( W \) as the...
Proof. It is clear that \( W \subseteq \langle X\rangle \) . It is also clear that \( W \) is closed under product. As to inverses, if \( w = {x}_{1}^{{e}_{1}}\cdots {x}_{n}^{{e}_{n}} \in W \) then \( {w}^{-1} = {x}_{n}^{-{e}_{n}}\cdots {x}_{1}^{-{e}_{1}} \in W \) . Hence, \( W \leq \langle X\rangle \) . However, since...
Yes