Q
stringlengths
4
3.96k
A
stringlengths
1
3k
Result
stringclasses
4 values
Lemma 7. The map root : Unit \( {}_{n} \rightarrow {A}_{n} \) is continuous.
Proof. Let \( P \in {\mathbf{{Unit}}}_{n} \) be given. Let \( {a}_{1},\ldots ,{a}_{r} \) be the distinct roots of \( P,{m}_{j} \) their multiplicities, and \( \rho \) the minimum distance between them. We denote by \( {D}_{j} \) the open disk with center \( {a}_{j} \) and radius \( \rho /2 \), and \( {C}_{j} \) its bou...
Yes
Theorem 5.3 Let \( {\lambda }_{0} \) be an algebraically simple eigenvalue of a matrix \( {M}_{0} \in \) \( {\mathbf{M}}_{n}\left( \mathbb{C}\right) \). Then there exists an open neighbourhood \( \mathcal{M} \) of \( {M}_{0} \) in \( {\mathbf{M}}_{n}\left( \mathbb{C}\right) \), and two analytic functions\n\n\[ \nM \map...
Proof. Let \( {X}_{0} \) be an eigenvector of \( {M}_{0} \) associated with \( {\lambda }_{0} \). We know that \( {\lambda }_{0} \) is also a simple eigenvalue of \( {M}_{0}^{T} \). Thanks to Proposition 3.15, an eigenvector \( {Y}_{0} \) of \( {M}_{0}^{T} \) (associated with \( {\lambda }_{0} \)) satisfies \( {Y}_{0}^...
Yes
Theorem 5.4 In \( {\mathbf{M}}_{n}\left( \mathbb{C}\right) \), a matrix is normal if and only if it is unitarily diagonalizable:
Proof. A diagonal matrix is obviously normal. If \( U \) is unitary, a matrix \( M \) is normal if and only if \( {U}^{ * }{MU} \) is normal: we deduce that unitarily diagonalizable matrices are normal.\n\nWe now prove the converse. We proceed by induction on the size \( n \) of the matrix \( M \) . If \( n = 0 \), the...
Yes
Corollary 5.2 Unitary, Hermitian, and skew-Hermitian matrices are unitarily diagonalizable.
Proof. A diagonal matrix is obviously normal. If \( U \) is unitary, a matrix \( M \) is normal if and only if \( {U}^{ * }{MU} \) is normal: we deduce that unitarily diagonalizable matrices are normal.\n\nWe now prove the converse. We proceed by induction on the size \( n \) of the matrix \( M \) . If \( n = 0 \), the...
Yes
Theorem 5.5 Let \( M \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \) be a normal matrix. There exists an orthogonal matrix \( O \) such that \( {OM}{O}^{-1} \) is block-diagonal, the diagonal blocks being \( 1 \times 1 \) (those corresponding to the real eigenvalues of \( M \) ) or \( 2 \times 2 \), the latter being ma...
Proof. One again proceeds by induction on \( n \) . When \( n \geq 1 \), the proof is the same as in the previous section whenever \( M \) has at least one real eigenvalue.\n\nIf this is not the case, then \( n \) is even. Let us first consider the case \( n = 2 \) . Then\n\n\[ \nM = \left( \begin{array}{ll} a & b \\ c...
Yes
Real symmetric matrices are diagonalizable over \( \mathbb{R} \), through orthogonal conjugation. In other words, given \( M \in {\operatorname{Sym}}_{n}\left( \mathbb{R}\right) \), there exists an \( O \in {\mathbf{O}}_{n}\left( \mathbb{R}\right) \) such that \( {\mathrm{{OMO}}}^{-1} \) is diagonal.
In fact, because the eigenvalues of \( M \) are real, \( {OM}{O}^{-1} \) has only \( 1 \times 1 \) blocks. We say that real symmetric matrices are orthogonally diagonalizable.
No
Proposition 5.7 The functional calculus with holomorphic functions enjoys the following properties. Below, the domains of functions are such that the expressions make sense.\n\n- If \( f \) and \( g \) match at order \( n \) over \( \operatorname{Sp}A \), then \( f\left( A\right) = g\left( A\right) \) .\n\n- Conjugatio...
Proof. The first property follows directly from the definition, and the linearity is obvious. The conjugation formula is already true for polynomials.\n\nThe product formula is true for polynomials. Now, if \( f \) and \( g \) are interpolated by \( P \) and \( Q \), respectively, at order \( n \) at every point of \( ...
Yes
Proposition 5.8 Let \( f \) be holomorphic over a domain \( \mathcal{U} \) containing the spectrum of \( A \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) . Let \( \Gamma \) be a positively oriented contour around \( \mathrm{{Sp}}A \), contained in \( \mathcal{U} \) .\n\nThen we have\n\n\[ f\left( A\right) = \frac{1}{2...
Proof. Because of the conjugation property, and thanks to Proposition 3.20, it is enough to verify the formula when \( A = \lambda {I}_{n} + N \) where \( N \) is nilpotent. By translation, it suffices to treat the nilpotent case.\n\nSo we assume that \( A \) is nilpotent. Therefore \( \Gamma \) is a disjoint union of ...
Yes
Proposition 5.11 For every \( A \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \), we have\n\n\[ w\left( A\right) \leq \parallel A{\parallel }_{2} \leq {2w}\left( A\right) \]\n
Proof. Cauchy-Schwarz gives\n\n\[ \left| {{x}^{ * }{Ax}}\right| \leq \parallel x{\parallel }_{2}\parallel {Ax}{\parallel }_{2} \leq \parallel A{\parallel }_{2}\parallel x{\parallel }_{2}^{2} \]\n\nwhich yields \( w\left( A\right) \leq \parallel A{\parallel }_{2} \) .\n\nOn the other hand, let us majorize \( \left| {{y}...
Yes
Proposition 5.12 (Gershgorin) The spectrum of \( A \) is included in the Gershgorin domain \( \mathcal{G}\left( A\right) \), defined as the union of the Gershgorin disks\n\n\[ \n{D}_{i}\left( A\right) \mathrel{\text{:=}} D\left( {{a}_{ii};{r}_{i}}\right) ,\;{r}_{i} \mathrel{\text{:=}} \mathop{\sum }\limits_{{j \neq i}}...
Proof. For \( r \in \left\lbrack {0,1}\right\rbrack \), we define a matrix \( A\left( r\right) \) by the formula\n\n\[ \n{a}_{ij}\left( r\right) \mathrel{\text{:=}} \left\{ \begin{array}{ll} {a}_{ii}, & j = i \\ r{a}_{ij}, & j \neq i \end{array}\right. \n\]\n\nIt is clear that the Gershgorin domain \( {\mathcal{G}}_{r}...
Yes
Theorem 5.7 There are exactly p eigenvalues of \( A \) in \( G \), counted with their multiplicities.
Proof. For \( r \in \left\lbrack {0,1}\right\rbrack \), we define a matrix \( A\left( r\right) \) by the formula\n\n\[ \n{a}_{ij}\left( r\right) \mathrel{\text{:=}} \left\{ \begin{array}{ll} {a}_{ii}, & j = i \\ r{a}_{ij}, & j \neq i \end{array}\right. \n\]\n\nIt is clear that the Gershgorin domain \( {\mathcal{G}}_{r}...
Yes
Proposition 5.13 Let \( A \) be an irreducible matrix. If an eigenvalue of \( A \) does not belong to the interior of any Gershgorin disk, then it belongs to every circle \( S\left( {{a}_{ii};{r}_{i}}\right) \) .
Proof. Let \( \lambda \) be such an eigenvalue and \( x \) an associated eigenvector. By assumption, one has \( \left| {\lambda - {a}_{ii}}\right| \geq \mathop{\sum }\limits_{{j \neq i}}\left| {a}_{ij}\right| \) for every \( i \) . Let \( I \) be the set of indices for which \( \left| {x}_{i}\right| = \) \( \parallel x...
Yes
Corollary 5.4 Let \( A \) be a square matrix. If \( A \) is strictly diagonally dominant, or if \( A \) is irreducible and strongly diagonally dominant, then \( A \) is invertible.
In fact, either zero does not belong to the Gershgorin domain, or it is not interior to the disks. In the latter case, \( A \) is assumed to be irreducible, and there exists a disk \( {D}_{j} \) that does not contain zero.
No
Proposition 6.1 Let \( H \in {\mathbf{{HPD}}}_{n} \) and \( K \in {\mathbf{H}}_{n} \) be given. Then the product \( {HK} \) (or KH as well) is diagonalizable with real eigenvalues. The number of positive (respectively, negative) eigenvalues of \( {HK} \) equals that for \( K \) .
Proof. Recall that \( H \) is unitary diagonalizable with real eigenvalues: there is a \( U \in \) \( {\mathbf{U}}_{n} \) such that \( H = {U}^{ * }\operatorname{diag}\left( {{\mu }_{1},\ldots ,{\mu }_{n}}\right) U \) . Because \( H \) is positive-definite, we have \( {\mu }_{j} > 0 \) . Setting \( h = {U}^{ * }\operat...
Yes
Theorem 6.1 Given \( H \in {\mathbf{{HPD}}}_{n} \), there exists one and only one \( h \in {\mathbf{{HPD}}}_{n} \) such that \( {h}^{2} = H \) . The matrix \( h \) is denoted \( \sqrt{H} \), and called the square root of \( H \) .
Proof. Such a square root was constructed in the proof of Proposition 6.1. There remains to prove uniqueness. So let \( h,{h}^{\prime } \in {HP}{D}_{n} \) be such that \( {h}^{\prime 2} = {h}^{2} \) . Set \( U \mathrel{\text{:=}} \) \( {h}^{\prime }{h}^{-1} \) . We have \( {U}^{ * }U = {h}^{-1}{h}^{\prime 2}{h}^{-1} = ...
Yes
Theorem 6.2 Let \( M \) be an \( n \times n \) Hermitian matrix and \( {\lambda }_{1},\ldots ,{\lambda }_{n} \) its eigenvalues arranged in increasing order, counted with multiplicity. Then\n\n\[{\lambda }_{k} = \mathop{\min }\limits_{{\dim F = {kx}}}\mathop{\max }\limits_{{x \in F\smallsetminus \{ 0\} }}\frac{{x}^{ * ...
Proof. There remains to prove (6.4). There are two possible strategies. Either we start from nothing and argue again about the intersection of subspaces whose dimensions sum up to \( n + 1 \), or we remark that \( {\lambda }_{k}\left( M\right) = - {\lambda }_{n - k + 1}\left( {-M}\right) \), because \( M \) and \( - M ...
Yes
Theorem 6.3 Let \( A \) and \( B \) be \( n \times n \) Hermitian matrices. Let \( 1 \leq i, j, k \leq n \) be indices.\n\n- If \( i + j = k + 1 \), we have\n\n\[ \n{\lambda }_{k}\left( {A + B}\right) \geq {\lambda }_{i}\left( A\right) + {\lambda }_{j}\left( B\right) \n\]\n\n- If \( i + j = k + n \), we have\n\n\[ \n{\...
Proof. Once again, any inequality can be deduced from the other ones by means of \( \left( {A, B}\right) \leftrightarrow \left( {-A, - B}\right) \) . Thus it is sufficient to treat the case where \( i + j = k + n \) .\n\nFrom (6.4), we know that there exists an \( \left( {n - k + 1}\right) \) -dimensional subspace \( H...
Yes
Proposition 6.2 The eigenvalues \( {\lambda }_{k} \) over \( {\mathbf{H}}_{n} \) are Lipschitz functions, with Lipschitz ratio equal to one:\n\n\[ \left| {{\lambda }_{k}\left( M\right) - {\lambda }_{k}\left( N\right) }\right| \leq \left\lbrack {M - N}\right\rbrack \]
Proof. Taking \( i = k \) and either \( j = 1 \) or \( j = n \), then \( A = N \) and \( B = M - N \), we obtain\n\n\[ {\lambda }_{1}\left( {M - N}\right) \leq {\lambda }_{k}\left( M\right) - {\lambda }_{k}\left( N\right) \leq {\lambda }_{n}\left( {M - N}\right) \]
Yes
Proposition 6.3 The square root is Hölderian with exponent \( \frac{1}{2} \) over \( {\mathbf{{HPD}}}_{n} \) :\n\n\[ \parallel \sqrt{H} - \sqrt{K}{\parallel }_{2} \leq \sqrt{\parallel H - K{\parallel }_{2}},\;\forall H, K \in {\mathbf{{HPD}}}_{n}. \]
Proof. Let \( A, B \in {\mathbf{{HPD}}}_{n} \) be given. Let us develop\n\n\[ {B}^{2} - {A}^{2} = {\left( B - A\right) }^{2} + \left( {B - A}\right) A + A\left( {B - A}\right) . \]\n\n(6.8)\n\nUp to exchanging the roles of \( A \) and \( B \), we may assume that \( {\lambda }_{n}\left( {B - A}\right) \geq {\lambda }_{n...
Yes
Theorem 6.4 The square root is operator monotone over the positive-semidefinite Hermitian matrices: if \( {0}_{n} \leq H \leq K \), then \( \sqrt{H} \leq \sqrt{K} \) .
Proof. Let \( B \) and \( A \) be Hermitian positive-semidefinite. If \( {B}^{2} \leq {A}^{2} \), then (6.8) yields\n\n\[ \n{\left( B - A\right) }^{2} + \left( {B - A}\right) A + A\left( {B - A}\right) \leq {0}_{n}.\n\]\n\nLet \( {\lambda }_{1} \) be the smallest eigenvalue of \( A - B \), and \( x \) an associate eige...
Yes
Theorem 6.5 Let \( H \in {\mathbf{H}}_{n - 1}, x \in {\mathbb{C}}^{n - 1} \), and \( a \in \mathbb{R} \) be given. Let \( {\lambda }_{1} \leq \cdots \leq {\lambda }_{n - 1} \) be the eigenvalues of \( H \) and \( {\mu }_{1} \leq \cdots \leq {\mu }_{n} \) those of the Hermitian matrix \[ {H}^{\prime } = \left( \begin{ar...
Proof. The inequality \( {\mu }_{j} \leq {\lambda }_{j} \) follows from (6.3), because the infimum concerns the same quantity \[ \mathop{\max }\limits_{{x \in F, x \neq 0}}\frac{{x}^{ * }{H}^{\prime }x}{\parallel x{\parallel }_{2}^{2}} \] but is taken over a smaller set in the case of \( {\lambda }_{j} \) : that of sub...
Yes
Theorem 6.6 Let \( {\lambda }_{1} \leq \cdots \leq {\lambda }_{n - 1} \) and \( {\mu }_{1} \leq \cdots \leq {\mu }_{n} \) be real numbers satisfying \( {\mu }_{1} \leq {\lambda }_{1} \leq \cdots \leq {\mu }_{j} \leq {\lambda }_{j} \leq {\mu }_{j + 1} \leq \cdots \) . Then there exist a vector \( x \in {\mathbb{R}}^{n} ...
Proof. Let us compute the characteristic polynomial of \( H \) from Schur’s complement formula \( {}^{1} \) (see Proposition 3.9):\n\n\[ \n{p}_{n}\left( X\right) = \left( {X - a - {x}^{T}{\left( X{I}_{n - 1} - \Lambda \right) }^{-1}x}\right) \det \left( {X{I}_{n - 1} - \Lambda }\right)\n\]\n\n\[ \n= \left( {X - a - \ma...
No
Corollary 6.1 Let \( H \in {\mathbf{H}}_{n - 1}\left( \mathbb{R}\right) \) be given, with eigenvalues \( {\lambda }_{1} \leq \cdots \leq {\lambda }_{n - 1} \) . Let \( {\mu }_{1},\ldots ,{\mu }_{n} \) be real numbers satisfying \( {\mu }_{1} \leq {\lambda }_{1} \leq \cdots \leq {\mu }_{j} \leq {\lambda }_{j} \leq {\mu ...
The proof consists in diagonalizing \( H \) through a unitary matrix \( U \in {\mathbf{U}}_{n - 1} \), then applying Theorem 6.6, and conjugating the resulting matrix by \( \operatorname{diag}\left( {{U}^{ * },1}\right) \) .
No
Proposition 6.4 Let \( x, y \in {\mathbb{R}}^{n} \) be given. Then \( x \prec y \) if and only if for every real number \( t \) ,
Proof. We may assume that \( x \) and \( y \) are nondecreasing. If the inequality (6.9) holds, we write it first for \( t \) outside the interval \( I \) containing the \( {x}_{j}\mathrm{\;s} \) and the \( {y}_{j}\mathrm{\;s} \) . This gives \( {s}_{n}\left( x\right) = {s}_{n}\left( y\right) \) . Then we write it for ...
Yes
Theorem 6.7 (Schur) Let \( H \) be an Hermitian matrix with diagonal a and spectrum \( \lambda \) . Then \( a \succ \lambda \) .
Proof. Let \( n \) be the size of \( H \) . We argue by induction on \( n \) . We may assume that \( {a}_{n} \) is the largest component of \( a \) . Because \( {s}_{n}\left( \lambda \right) = \operatorname{Tr}A \), one has \( {s}_{n}\left( \lambda \right) = {s}_{n}\left( a\right) \) . In particular, the theorem holds ...
Yes
Theorem 6.8 Let \( a \) and \( \lambda \) be two sequences of \( n \) real numbers such that \( a \succ \lambda \) . Then there exists a real symmetric matrix of size \( n \times n \) whose diagonal is a and spectrum is \( \lambda \) .
Proof. We proceed by induction on \( n \) . The statement is trivial if \( n = 1 \) . If \( n \geq 2 \), we use the following lemma, which is proved afterwards.\n\nLemma 9. Let \( n \geq 2 \) and \( \alpha ,\beta \) be two nondecreasing sequences of \( n \) real numbers, satisfying \( \alpha \prec \beta \) . Then there...
Yes
Lemma 9. Let \( n \geq 2 \) and \( \alpha ,\beta \) be two nondecreasing sequences of \( n \) real numbers, satisfying \( \alpha \prec \beta \) . Then there exists a sequence \( \gamma \) of \( n - 1 \) real numbers such that\n\n\[{\alpha }_{1} \leq {\gamma }_{1} \leq {\alpha }_{2} \leq \cdots \leq {\gamma }_{n - 1} \l...
We now prove Lemma 9. Let \( \Delta \) be the set of sequences \( \delta \) of \( n - 1 \) real numbers satisfying\n\n\[{\alpha }_{1} \leq {\delta }_{1} \leq {\alpha }_{2} \leq \cdots \leq {\delta }_{n - 1} \leq {\alpha }_{n}\]\n\n(6.10)\n\ntogether with\n\n\[\mathop{\sum }\limits_{{j = 1}}^{k}{\delta }_{j} \leq \matho...
Yes
Proposition 6.5 Let \( H \in {\mathbf{H}}_{n} \) be a positive-semidefinite Hermitian matrix. Then\n\n\[ \det H \leq \mathop{\prod }\limits_{{j = 1}}^{n}{h}_{jj} \]\n\nIf \( H \in {\mathbf{{HPD}}}_{n} \), the equality holds only if \( H \) is diagonal.
Proof. If \( \det H = 0 \), there is nothing to prove, because the \( {h}_{jj} \) are nonnegative (these are numbers \( {\left( {\mathbf{e}}^{j}\right) }^{ * }H{\mathbf{e}}^{j} \) ). Otherwise, \( H \) is positive-definite and one has \( {h}_{jj} > 0 \) . We restrict our attention to the case with a constant diagonal b...
Yes
Theorem 6.10 The map\n\n\\[ \nH \mapsto {\\left( \\det H\\right) }^{1/n}\n\\]\n\nis concave over the cone of positive semidefinite \\( n \\times n \\) Hermitian matrices.
Proof. Let \\( H, K \\in {\\mathbf{{HPD}}}_{n} \\) be given. From Proposition 6.1, the eigenvalues \\( {\\mu }_{1},\\ldots ,{\\mu }_{n} \\) of \\( {HK} \\) are real and positive, even though \\( {HK} \\) is not Hermitian. We thus have\n\n\\[ \n{\\left( \\det H\\right) }^{1/n}{\\left( \\det K\\right) }^{1/n} = {\\left( ...
Yes
Proposition 7.2 For conjugate exponents \( p,{p}^{\prime } \), one has\n\n\[ \parallel x{\parallel }_{p} = \mathop{\sup }\limits_{{y \neq 0}}\frac{\Re \langle x, y\rangle }{\parallel y{\parallel }_{{p}^{\prime }}} = \mathop{\sup }\limits_{{y \neq 0}}\frac{\left| \langle x, y\rangle \right| }{\parallel y{\parallel }_{{p...
Proof. The inequality \( \geq \) is a consequence of Hölder’s. The reverse inequality is obtained by taking \( {y}_{j} = {\bar{x}}_{j}{\left| {x}_{j}\right| }^{p - 2} \) if \( p < \infty \) . If \( p = \infty \), choose \( {y}_{j} = {\bar{x}}_{j} \) for an index \( j \) such that \( \left| {x}_{j}\right| = \parallel x{...
Yes
Proposition 7.3 All norms on \( E = {K}^{n} \) are equivalent. For example,\n\n\[ \parallel x{\parallel }_{\infty } \leq \parallel x{\parallel }_{p} \leq {n}^{1/p}\parallel x{\parallel }_{\infty } \]
Proof. It is sufficient to show that every norm is equivalent to \( \parallel \cdot {\parallel }_{1} \) .\n\nLet \( N \) be a norm on \( E \) . If \( x \in E \), the triangle inequality gives\n\n\[ N\left( x\right) \leq \mathop{\sum }\limits_{i}\left| {x}_{i}\right| N\left( {\mathbf{e}}^{i}\right) \]\n\nwhere \( \left(...
Yes
Proposition 7.4 The bi-dual (dual of the dual norm) of a norm is this norm itself:
Proof. From (7.3), one has \( {\left( \parallel \cdot {\parallel }^{\prime }\right) }^{\prime } \leq \parallel \cdot \parallel \) . The converse is a consequence of the Hahn-Banach theorem: the unit ball \( B \) of \( \parallel \cdot \parallel \) is convex and compact. If \( x \) is a point of its boundary (i.e., \( \p...
Yes
Proposition 7.5 For an induced norm, the condition \( \parallel B\parallel < 1 \) implies that \( {I}_{n} - B \) is invertible, with the inverse given by the sum of the series\n\n\[ \mathop{\sum }\limits_{{k = 0}}^{\infty }{B}^{k} \]
Proof. The series \( \mathop{\sum }\limits_{k}{B}^{k} \) is normally convergent, because \( \mathop{\sum }\limits_{k}\begin{Vmatrix}{B}^{k}\end{Vmatrix} \leq \mathop{\sum }\limits_{k}\parallel B{\parallel }^{k} \), where the latter series converges because \( \parallel B\parallel < 1 \) . Because \( {\mathbf{M}}_{n}\le...
Yes
Proposition 7.6 For every induced norm, one has\n\n\[ \rho \left( A\right) \leq \parallel A\parallel \]
Proof. The case \( K = \mathbb{C} \) is easy, because there exists an eigenvector \( X \in E \) associated with an eigenvalue of modulus \( \rho \left( A\right) \) :\n\n\[ \rho \left( A\right) \parallel X\parallel = \parallel {\lambda X}\parallel = \parallel {AX}\parallel \leq \parallel A\parallel \parallel X\parallel ...
Yes
Proposition 7.7 Let \( \parallel \cdot \parallel \) be a norm on \( {K}^{n} \) and \( P \in {\mathbf{{GL}}}_{n}\left( K\right) \) . Hence, \( N\left( x\right) \mathrel{\text{:=}} \parallel {Px}\parallel \) defines a norm on \( {K}^{n} \) . Denoting still by \( \parallel \cdot \parallel \) and \( N \) the induced norms ...
Proof. Using the change of dummy variable \( y = {Px} \), we have\n\n\[ N\left( A\right) = \mathop{\sup }\limits_{{x \neq 0}}\frac{\parallel {PAx}\parallel }{\parallel {Px}\parallel } = \mathop{\sup }\limits_{{y \neq 0}}\frac{\begin{Vmatrix}PA{P}^{-1}y\end{Vmatrix}}{\parallel y\parallel } = \begin{Vmatrix}{{PA}{P}^{-1}...
Yes
Theorem 7.1 For every \( B \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) and all \( \varepsilon > 0 \), there exists a norm on \( {\mathbb{C}}^{n} \) such that for the induced norm,\n\n\[ \parallel B\parallel \leq \rho \left( B\right) + \varepsilon \]\n\nIn other words, \( \rho \left( B\right) \) is the infimum of \(...
Proof. From Theorem 3.5 there exists \( P \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) such that \( T \mathrel{\text{:=}} {PB}{P}^{-1} \) is upper-triangular. From Proposition 7.7, one has\n\n\[ \inf \parallel B\parallel = \inf \begin{Vmatrix}{{PB}{P}^{-1}}\end{Vmatrix} = \inf \parallel T\parallel \]\n\nwhere the...
Yes
Proposition 7.8 If \( A \in {\mathbf{M}}_{n}\left( k\right) \) (with \( k = \mathbb{R} \) or \( \mathbb{C} \) ), then\n\n\[ \rho \left( A\right) = \mathop{\lim }\limits_{{m \rightarrow \infty }}{\begin{Vmatrix}{A}^{m}\end{Vmatrix}}^{1/m} \]\n\nfor every matrix norm.
Proof. Let \( \parallel \cdot \parallel \) be a matrix norm over \( {\mathbf{M}}_{n}\left( k\right) \) . From Proposition 7.6 and the fact that\n\n\[ \operatorname{Sp}\left( {A}^{m}\right) = \left\{ {{\lambda }^{m} \mid \lambda \in \operatorname{Sp}A}\right\} \]\n\nwe have\n\n\[ \rho \left( A\right) = \rho {\left( {A}^...
Yes
Theorem 7.3 (Riesz-Thorin) Let \( \Omega \) be an open set in \( {\mathbb{R}}^{D} \) and \( \omega \) an open set in \( {\mathbb{R}}^{d} \) . Let \( {p}_{0},{p}_{1},{q}_{0},{q}_{1} \) be four numbers in \( \left\lbrack {1, + \infty }\right\rbrack \) . Let \( \theta \in \left\lbrack {0,1}\right\rbrack \) and \( p, q \) ...
Proof. (Due to Riesz) Let us fix \( x \) and \( y \) in \( {K}^{n} \) . We have to bound \[ \left| {\langle y,{Ax}\rangle }\right| = \left| {\mathop{\sum }\limits_{{j, k}}{a}_{jk}{x}_{j}{\bar{y}}_{k}}\right| . \] Let \( B \) be the strip in the complex plane defined by \( \Re z \in \left\lbrack {0,1}\right\rbrack \) . ...
Yes
Proposition 8.1 A matrix is nonnegative if and only if \( x \geq 0 \) implies \( {Ax} \geq 0 \) . It is positive if and only if \( x \geq 0 \) and \( x \neq 0 \) imply \( {Ax} > 0 \) .
Proof. Let us assume that \( {Ax} \geq 0 \) (respectively, \( > 0 \) ) for every \( x \geq 0 \) (respectively, \( \geq 0 \) and \( \neq 0 \) ). Then the \( i \) th column \( {A}^{\left( i\right) } \) is nonnegative (respectively, positive), since it is the image of the \( i \) th vector of the canonical basis. Hence \(...
Yes
Proposition 8.2 If \( A \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \) is nonnegative and irreducible, then \( {\left( I + A\right) }^{n - 1} > 0 \) .
Proof. Let \( x \neq 0 \) be nonnegative, and define \( {x}^{m} = {\left( I + A\right) }^{m}x \), which is nonnegative too. Let us denote by \( {P}_{m} \) the set of indices of the nonzero components of \( {x}^{m} \) : \( {P}_{0} \) is nonempty. Because \( {x}_{i}^{m + 1} \geq {x}_{i}^{m} \), one has \( {P}_{m} \subset...
Yes
Theorem 8.1 Let \( A \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \) be a nonnegative matrix. Then \( \rho \left( A\right) \) is an eigenvalue of \( A \) associated with a nonnegative eigenvector.
Proof. Let \( \lambda \) be an eigenvalue of maximal modulus and \( v \) an eigenvector, normalized by \( \parallel v{\parallel }_{1} = 1 \) . Then\n\n\[ \rho \left( A\right) \left| v\right| = \left| {\lambda v}\right| = \left| {Av}\right| \leq A\left| v\right| \]\n\nLet us denote by \( C \) the subset of \( {\mathbb{R...
Yes
Theorem 8.2 Let \( A \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \) be a nonnegative irreducible matrix. Then \( \rho \left( A\right) \) is a simple eigenvalue of \( A \), associated with a positive eigenvector: Moreover, \( \rho \left( A\right) > 0 \) .
Proof. For \( r \geq 0 \), we denote by \( {C}_{r} \) the set of vectors of \( {\mathbb{R}}^{n} \) defined by the conditions\n\n\[ x \geq 0,\;\parallel x{\parallel }_{1} = 1,\;{Ax} \geq {rx}. \]\n\nEach \( {C}_{r} \) is a convex compact set. We saw in the previous section that if \( \lambda \) is an eigenvalue associat...
Yes
Lemma 10. Let \( r \geq 0 \) and \( x \geq 0 \) such that \( {Ax} \geq {rx} \) and \( {Ax} \neq {rx} \) . Then there exists \( {r}^{\prime } > r \) such that \( {C}_{{r}^{\prime }} \) is nonempty.
Proof. Set \( y \mathrel{\text{:=}} {\left( {I}_{n} + A\right) }^{n - 1}x \) . Because \( A \) is irreducible and \( x \geq 0 \) is nonzero, one has \( y > 0 \) . Likewise, \( {Ay} - {ry} = {\left( {I}_{n} + A\right) }^{n - 1}\left( {{Ax} - {rx}}\right) > 0 \) . Let us define \( {r}^{\prime } \mathrel{\text{:=}} \) \( ...
Yes
Lemma 11. The nonnegative eigenvectors of \( A \) are positive. The corresponding eigenvalue is positive too.
Proof. Given such a vector \( x \) with \( {Ax} = {\lambda x} \), we observe that \( \lambda \in {\mathbb{R}}^{ + } \) . Then\n\n\[ x = \frac{1}{{\left( 1 + \lambda \right) }^{n - 1}}{\left( {I}_{n} + A\right) }^{n - 1}x \]\n\nand the right-hand side is strictly positive, from Proposition 8.2.\n\nInasmuch as \( A \) is...
Yes
Lemma 12. Let \( M, B \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) be matrices, with \( M \) irreducible and \( \left| B\right| \leq M \) . Then \( \rho \left( B\right) \leq \rho \left( M\right) \) .
In order to establish the inequality, we proceed as above. If \( \lambda \) is an eigenvalue of \( B \), of modulus \( \rho \left( B\right) \), and if \( x \) is a normalized eigenvector, then \( \rho \left( B\right) \left| x\right| \leq \left| B\right| \cdot \left| x\right| \leq \) \( M\left| x\right| \), so that \( {...
Yes
Theorem 8.3 Under the assumptions of Theorem 8.2, let \( p \) be the cardinality of the set \( {\operatorname{Sp}}_{\max }\left( A\right) \) of eigenvalues of \( A \) of maximal modulus \( \rho \left( A\right) \) . Then we have \( {\operatorname{Sp}}_{\max }\left( A\right) = \rho \left( A\right) {\mathcal{U}}_{p} \), w...
Proof. Let us denote by \( X \) the unique nonnegative eigenvector of \( A \) normalized by \( \parallel X{\parallel }_{1} = 1 \) . If \( Y \) is a unitary eigenvector, associated with an eigenvalue \( \mu \) of maximal modulus \( \rho \left( A\right) \), the inequality \( \rho \left( A\right) \left| Y\right| = \left| ...
Yes
Theorem 8.4 (Birkhoff) A matrix \( M \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \) is bistochastic if and only if it is a center of mass (i.e., a barycenter with nonnegative weights) of permutation matrices.
Proof. Let \( M \in {\mathbf{{DS}}}_{n} \) be given. If \( M \) is not a permutation matrix, there exists an entry \( {m}_{{i}_{1}{j}_{1}} \in \left( {0,1}\right) \) . Inasmuch as \( M \) is stochastic, there also exists \( {j}_{2} \neq {j}_{1} \) such that \( {m}_{{i}_{1}{j}_{2}} \in \left( {0,1}\right) \) . Because \...
Yes
Corollary 8.1 Let \( \parallel \cdot \parallel \) be a norm on \( {\mathbb{R}}^{n} \), invariant under permutation of the coordinates. Then \( \parallel M\parallel = 1 \) for every bistochastic matrix (where as usual we have denoted \( \parallel \cdot \parallel \) the induced norm on \( {\mathbf{M}}_{n}\left( \mathbb{R...
Proof. To begin with, \( \parallel P\parallel = 1 \) for every permutation matrix, by assumption. Because the induced norm is convex (true for every norm), one deduces from Birkhoff's theorem that \( \parallel M\parallel \leq 1 \) for every bistochastic matrix. Furthermore, \( M\mathbf{e} = \mathbf{e} \) implies \( \pa...
Yes
Theorem 8.5 A matrix \( A \) is bistochastic if and only if \( {Ax} \succ x \) for every \( x \in {\mathbb{R}}^{n} \) .
Proof. If \( A \) is bistochastic, then \( \parallel {Ax}{\parallel }_{1} \leq \parallel A{\parallel }_{1}\parallel x{\parallel }_{1} = \parallel x{\parallel }_{1} \), because \( {A}^{T} \) is stochastic. Because \( A \) is stochastic, \( A\mathbf{e} = \mathbf{e} \). Applying the inequality to \( x - t\mathbf{e} \), on...
Yes
Theorem 8.6 Let \( x, y \in {\mathbb{R}}^{n} \) . Then \( x \prec y \) if and only if there exists a bistochastic matrix \( A \) such that \( y = {Ax} \) .
Proof. From the previous theorem, it is enough to show that if \( x \prec y \), there exists \( A \), a bistochastic matrix, such that \( y = {Ax} \) . To do so, one applies Theorem 6.8: there exists an Hermitian matrix \( H \) whose diagonal and spectrum are \( y \) and \( x \), respectively. Let us diagonalize \( H \...
Yes
Proposition 9.1 In a principal ideal domain, every pair of elements has a greatest common divisor. The gcd satisfies the Bézout identity: for every \( a, b \in A \), there exist \( u, v \in A \) such that\n\n\[ \gcd \left( {a, b}\right) = {ua} + {vb}. \]\n\nSuch \( u \) and \( v \) are coprime.
Proof. Let \( A \) be a principal ideal domain. If \( a, b \in A \), the ideal \( \mathcal{I} = : \left( {a, b}\right) \) is principal: \( \mathcal{I} = \left( d\right) \). Because \( a, b \in \mathcal{I}, d \) divides \( a \) and \( b \). Furthermore, \( d = {ua} + {vb} \) because \( d \in \mathcal{I} \). If \( c \) d...
Yes
Proposition 9.2 The principal ideal domains are Noetherian.
Proof. Let \( A \) be a principal ideal domain and let \( {\left( {I}_{j}\right) }_{j \geq 0} \) be a nondecreasing sequence of ideals in \( A \) . Let \( \mathcal{I} \) be their union, which happens to be an ideal. Let \( a \) be a generator: \( \mathcal{I} = \left( a\right) \) . Then \( a \) belongs to one of the ide...
Yes
Proposition 9.3 Euclidean domains are principal ideal domains.
Proof. Let \( \mathcal{I} \) be an ideal of a Euclidean domain \( A \) . If \( \mathcal{I} = \left( 0\right) \), there is nothing to show. Otherwise, let us select an element \( a \) for which \( N\left( a\right) \) is minimal in \( \mathcal{I} \smallsetminus \{ 0\} \) . If \( b \in \mathcal{I} \), the remainder \( r \...
Yes
Theorem 9.1 A square invertible matrix of size \( n \) with entries in a Euclidean domain \( A \) is a product of elementary matrices with entries in \( A \) .
Proof. We prove the theorem for \( n = 2 \) . The general case is deduced from that particular one and from the proof of Theorem 9.2 below, which uses multiplications by block-diagonal matrices with \( 1 \times 1 \) and \( 2 \times 2 \) diagonal blocks.\n\nLet\n\n\[ M = \left( \begin{matrix} a & {a}_{1} \\ c & d \end{m...
Yes
Lemma 13. There exists a map \( T : {\mathbf{M}}_{n \times m}\left( A\right) \rightarrow {\mathbf{M}}_{n \times m}\left( A\right) \) with the following properties.\n\n- \( {N}^{\prime } \mathrel{\text{:=}} T\left( N\right) \) is equivalent to \( N \) .\n\n- \( {n}_{11}^{\prime } \mid {n}_{11} \),\n\n- This divisibility...
Proof. (of the lemma).\n\nWe intentionally write this proof in the algorithmic style. Given the matrix \( N \in \) \( {\mathbf{M}}_{n \times m}\left( A\right) \), we distinguish four cases.\n\n1. IF \( {n}_{11} \) does not divide some \( {n}_{1j} \), THEN take the smallest such index \( j
No
Theorem 9.5 Two matrices in \( {\mathbf{M}}_{n}\left( k\right) \) are similar if and only if they have the same list of invariant polynomials (counted with their multiplicities).
Proof. We prove Theorem 9.6. The condition is clearly sufficient.\n\nConversely, if \( X{A}_{0} + {B}_{0} \) and \( X{A}_{1} + {B}_{1} \) are equivalent, there exist matrices \( P, Q \in \) \( {\mathbf{{GL}}}_{n}\left( A\right) \), such that \( P\left( {X{A}_{0} + {B}_{0}}\right) = \left( {X{A}_{1} + {B}_{1}}\right) Q ...
Yes
Theorem 9.6 Let \( {A}_{0},{A}_{1},{B}_{0},{B}_{1} \) be matrices in \( {\mathbf{M}}_{n}\left( k\right) \), with \( {A}_{0},{A}_{1} \in {\mathbf{{GL}}}_{n}\left( k\right) \). Then the matrices \( X{A}_{0} + {B}_{0} \) and \( X{A}_{1} + {B}_{1} \) are equivalent (in \( {\mathbf{M}}_{n}\left( A\right) \)) if and only if ...
Proof. We prove Theorem 9.6. The condition is clearly sufficient.\n\nConversely, if \( X{A}_{0} + {B}_{0} \) and \( X{A}_{1} + {B}_{1} \) are equivalent, there exist matrices \( P, Q \in \) \( {\mathbf{{GL}}}_{n}\left( A\right) \), such that \( P\left( {X{A}_{0} + {B}_{0}}\right) = \left( {X{A}_{1} + {B}_{1}}\right) Q ...
Yes
Theorem 9.7 If \( B \in {\mathbf{M}}_{n}\left( k\right) \), then \( B \) and \( {B}^{T} \) are similar.
Indeed, \( X{I}_{n} - B \) and \( X{I}_{n} - {B}^{T} \) are transposes of each other, and hence have the same list of minors, and the same invariant factors.
Yes
Theorem 9.9 Let \( k \) be a field, \( M \in {\mathbf{M}}_{n}\left( k\right) \) a square matrix, and \( {p}_{1},\ldots ,{p}_{n} \) its similarity invariants. Then \( {p}_{n} \) is the minimal polynomial of \( M \) . In particular, the minimal polynomial does not depend on the field under consideration, as long as it co...
Proof. We use the first canonical form \( {M}^{\prime } \) of \( M \) . Because \( {M}^{\prime } \) and \( M \) are similar, they have the same minimal polynomial. One thus can assume that \( M \) is in the canonical form \( M = \operatorname{diag}\left( {{M}_{1},\ldots ,{M}_{n}}\right) \), where \( {M}_{j} \) is the c...
Yes
Theorem 10.1 For every \( M \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \), there exists a unique pair\n\n\[ \n\left( {H, Q}\right) \in {\mathbf{{HPD}}}_{n} \times {\mathbf{U}}_{n}\n\]\n\nsuch that \( M = {HQ} \) . If \( M \in {\mathbf{{GL}}}_{n}\left( \mathbb{R}\right) \), then \( \left( {H, Q}\right) \in {\mathbf...
Proof. Existence. Because \( M{M}^{ * } \in {\mathbf{{HPD}}}_{n} \), we can set \( H \mathrel{\text{:=}} \sqrt{M{M}^{ * }} \) (the square root was defined in Section 6.1). Then \( Q \mathrel{\text{:=}} {H}^{-1}M \) satisfies \( {Q}^{ * }Q = {M}^{ * }{H}^{-2}M = \) \( {M}^{ * }{\left( M{M}^{ * }\right) }^{-1}M = {I}_{n}...
Yes
Proposition 10.1 Let \( A, B \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) be commuting matrices; that is, \( {AB} = {BA} \) . Then \( \exp \left( {A + B}\right) = \left( {\exp A}\right) \left( {\exp B}\right) .
Proof. The proof is exactly the same as for the exponential of complex numbers. We observe that because the series defining the exponential of a matrix is normally convergent, we may compute the product \( \left( {\exp A}\right) \left( {\exp B}\right) \) by multiplying term by term the series\n\n\[ \left( {\exp A}\righ...
Yes
For every \( A \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \), exp \( A \) is invertible, and its inverse is \( \exp \left( {-A}\right) \) .
Given two conjugate matrices \( B = {P}^{-1}{AP} \), we have \( {B}^{k} = {P}^{-1}{A}^{k}P \) for each integer \( k \) and thus\n\n\[ \exp \left( {{P}^{-1}{AP}}\right) = {P}^{-1}\left( {\exp A}\right) P. \]\n\n(10.1)
No
Proposition 10.2 For every \( A \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) , \[ \det \exp A = \exp \operatorname{Tr}A. \]
Proof. We begin with a reduction of \( A \) of the form \( A = {P}^{-1}{TP} \), where \( T \) is upper-triangular. Because \( {T}^{k} \) is still triangular, with diagonal entries equal to \( {t}_{jj}^{k} \) , \( \exp T \) is triangular too, with diagonal entries equal to \( \exp {t}_{jj} \) . Hence \[ \det \exp T = \m...
Yes
Proposition 10.3 The map \( \exp : {\mathbf{H}}_{n} \rightarrow {\mathbf{{HPD}}}_{n} \) is a homeomorphism.
Proof. Injectivity: Let \( A, B \in {\mathbf{H}}_{n} \) with \( \exp A = \exp B = : H \) . Then\n\n\[ \exp \frac{1}{2}A = \sqrt{H} = \exp \frac{1}{2}B \]\n\nBy induction, we have\n\n\[ \exp {2}^{-m}A = \exp {2}^{-m}B,\;m \in \mathbb{Z}. \]\n\nSubtracting \( {I}_{n} \), multiplying by \( {2}^{m} \), and passing to the l...
Yes
Proposition 10.4 Let \( G \) be a subgroup of \( {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \). We assume that \( G \) is stable under the map \( M \mapsto {M}^{ * } \) and that for every \( M \in G \cap {\mathbf{{HPD}}}_{n} \), the square root \( \sqrt{M} \) is an element of \( G \). Then \( G \) is stable under pola...
Proof. Let \( M \in G \) be given and let \( {HQ} \) be its polar decomposition. Because \( M{M}^{ * } \in \) \( G \cdot G = G \), we have \( {H}^{2} \in G \), hence \( H \in G \) by assumption. Finally, we have \( Q = \) \( {H}^{-1}M \in {G}^{-1} \cdot G = G \). An application of Theorem 10.1 finishes the proof.
Yes
Proposition 10.5 Let \( J \) be a complex \( n \times n \) matrix satisfying \( {J}^{2} = \pm {I}_{n} \) . The subgroup \( G \) of \( {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) defined by the equation \( {M}^{ * }{JM} = J \) is invariant under polar decomposition. If \( M \in G \), then \( \left| {\det M}\right| = 1 \...
Proof. Let \( M \in G \) . Then \( \det J = \det {M}^{ * }\det M\det J \) ; that is, \( {\left| \det M\right| }^{2} = 1 \) . In particular \( M \) is nonsingular. Then we have\n\n\[ \n{M}^{- * }J{M}^{-1} = {M}^{- * }\left( {{M}^{ * }{JM}}\right) {M}^{-1} = J, \n\] \n\nthus \( {M}^{-1} \in G \) . If \( M, N \in G \), we...
Yes
Theorem 10.2 Under the hypotheses of Proposition 10.5, the group \( G \) is homeomorphic to \( \left( {G \cap {\mathbf{U}}_{n}}\right) \times {\mathbb{R}}^{d} \), for a suitable integer \( d \) .
Proof. (of Theorem 10.2.)\n\nAccording to Proposition 10.4, the proof amounts to showing that the factor \( G \cap \) \( {\mathbf{{HPD}}}_{n} \) is homeomorphic to some \( {\mathbb{R}}^{d} \) . To do this, we define\n\n\[ \mathcal{G} \mathrel{\text{:=}} \left\{ {N \in {\mathbf{M}}_{n}\left( k\right) \mid \exp {tN} \in ...
Yes
Lemma 14. The set \( \mathcal{G} \) defined above satisfies\n\n\[ \mathcal{G} = \left\{ {N \in {\mathbf{M}}_{n}\left( k\right) \mid {N}^{ * }J + {JN} = {0}_{n}}\right\} . \]
Proof. If \( {N}^{ * }J + {JN} = {0}_{n} \), let us set \( M\left( t\right) = \exp {tN} \) . Then \( M\left( 0\right) = {I}_{n} \) and, thanks to (10.3)\n\n\[ \frac{d}{dt}M{\left( t\right) }^{ * }{JM}\left( t\right) = {M}^{ * }\left( t\right) \left( {{N}^{ * }J + {JN}}\right) M\left( t\right) = {0}_{n}, \]\n\nso that \...
Yes
Lemma 15. The map \( \exp : \mathcal{G} \cap {\mathbf{H}}_{n} \rightarrow G \cap {\mathbf{{HPD}}}_{n} \) is a homeomorphism.
Proof. We must show that \( \exp : \mathcal{G} \cap {\mathbf{H}}_{n} \rightarrow G \cap {\mathbf{{HPD}}}_{n} \) is onto. Let \( M \in G \cap {\mathbf{{HPD}}}_{n} \) and let \( N \) be the Hermitian matrix such that \( \exp N = M \) . Let \( p \in \mathbb{R}\left\lbrack X\right\rbrack \) be a polynomial with real entrie...
No
Proposition 10.6 The unitary group \( \mathbf{U}\left( {p, q}\right) \) is homeomorphic to \( {\mathbf{U}}_{p} \times {\mathbf{U}}_{q} \times {\mathbb{R}}^{2pq} \) . In particular, \( \mathbf{U}\left( {p, q}\right) \) is connected.
There remains to show connectedness. It is a straightforward consequence of the following lemma.\n\nLemma 16. The unitary group \( {\mathbf{U}}_{n} \) is connected.\n\nInasmuch as \( {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) is homeomorphic to \( {\mathbf{U}}_{n} \times {\mathbf{{HPD}}}_{n} \) (via polar decomposi...
No
Lemma 17. The linear group \( {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) is connected.
Proof. Let \( M \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) be given. Define \( A \mathrel{\text{:=}} \mathbb{C} \smallsetminus \left\{ {{\left( 1 - \lambda \right) }^{-1} \mid \lambda \in \operatorname{Sp}\left( M\right) }\right\} \), which is arcwise connected because its complement is finite. The set \( A \) ...
Yes
Lemma 18. Given \( M \in {\mathbf{O}}_{n} \), there exists \( Q \in {\mathbf{O}}_{n} \) such that the matrix \( {Q}^{-1}{MQ} \) has the form\n\n\[ \left( \begin{matrix} \left( \cdot \right) & 0 & \cdots & 0 \\ 0 & \ddots & \ddots & \vdots \\ \vdots & \ddots & \ddots & 0 \\ 0 & \cdots & 0 & \left( \cdot \right) \end{mat...
We now prove Lemma 18. As an orthogonal matrix, \( M \) is normal. From Theorem 5.5, it decomposes into a matrix of the form (10.8), the \( 1 \times 1 \) diagonal blocks being the real eigenvalues. These eigenvalues are \( \pm 1 \), inasmuch as \( M \) is orthogonal. The diagonal blocks \( 2 \times 2 \) are direct simi...
Yes
Proposition 10.8 \( \left( {p, q \geq 1}\right) \) The connected components of \( G = \mathbf{O}\left( {p, q}\right) \) are the sets \( {G}_{\alpha } \mathrel{\text{:=}} {\sigma }^{-1}\left( \alpha \right) \), defined by \( {\alpha }_{1}\det A > 0 \) and \( {\alpha }_{2}\det D > 0 \), when a matrix \( M \) is written b...
1. \( {G}_{\alpha }^{-1} = {G}_{\alpha } \). 2. \( {G}_{\alpha } \cdot {G}_{{\alpha }^{\prime }} = {G}_{\alpha {\alpha }^{\prime }} \).
No
Proposition 10.9 The symplectic group \( {\mathbf{{Sp}}}_{n} \) is homeomorphic to \( {\mathbf{U}}_{n} \times {\mathbb{R}}^{n\left( {n + 1}\right) } \) .
Corollary 10.2 In particular, every symplectic matrix has determinant +1 . Indeed, Proposition 10.9 implies that \( {\mathbf{{Sp}}}_{n} \) is connected. Because the determinant is continuous, with values in \( \{ - 1,1\} \), it is constant, equal to +1 .
No
Corollary 10.2 In particular, every symplectic matrix has determinant +1 .
Indeed, Proposition 10.9 implies that \( {\mathbf{{Sp}}}_{n} \) is connected. Because the determinant is continuous, with values in \( \{ - 1,1\} \), it is constant, equal to +1 .
Yes
Theorem 11.1 The matrix \( M \in {\mathbf{{GL}}}_{n}\left( k\right) \) admits an LU factorization if and only if its leading principal minors are nonzero. When this condition is fulfilled, the LU factorization is unique.
Proof. Let us begin with uniqueness: if \( {LU} = {L}^{\prime }{U}^{\prime } \), then \( {\left( {L}^{\prime }\right) }^{-1}L = {U}^{\prime }{U}^{-1} \), which reads \( {L}^{\prime \prime } = {U}^{\prime \prime } \), where \( {L}^{\prime \prime } \) and \( {U}^{\prime \prime } \) are triangular of opposite types, the d...
Yes
Corollary 11.1 Let \( M \in {\mathbf{{GL}}}_{n}\left( k\right) \), with \( n = {2m} \), read blockwise\n\n\[ M = \left( \begin{array}{ll} A & B \\ C & D \end{array}\right) ,\;A, B, C, D \in {\mathbf{{GL}}}_{m}\left( k\right) .\n\]\n\nThen\n\n\[ {M}^{-1} = \left( \begin{array}{ll} {\left( A - B{D}^{-1}C\right) }^{-1} & ...
Proof. We can verify the formula by multiplying by \( M \) . The only point to show is that the inverses are meaningful: \( A - B{D}^{-1}C,\ldots \) are invertible. Because of the symmetry of the formulæ, it is enough to check it for a single term, namely \( D - \) \( C{A}^{-1}B \) . Schur’s complement formula gives \(...
Yes
Lemma 19. If \( {P}_{n} \leq {c}_{\alpha }{n}^{\alpha } \) (with \( 2 \leq \alpha \leq 3 \) ), then \( {j}_{\ell } \leq {C}_{\alpha }{\pi }_{\ell } \), where \( {C}_{\alpha } = 1 + \) \( 3{c}_{\alpha }/\left( {{2}^{\alpha - 1} - 1}\right) \) .
Proof. It is enough to sum (11.3) from \( k = 1 \) to \( l \) and use the inequality \( 1 + q + \cdots + \) \( {q}^{l - 1} \leq {q}^{\ell }/\left( {q - 1}\right) \) for \( q > 1 \) .
No
Proposition 11.2 If the complexity \( {P}_{n} \) of the product in \( {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) is bounded by \( {c}_{\alpha }{n}^{\alpha } \) , then the complexity \( {J}_{n} \) of inversion in \( {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) is bounded by \( {d}_{\alpha }{n}^{\alpha } \), where\n\n...
Proof. There remains to prove the second part. We notice that if \( A, B \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) are given, then the matrix\n\n\[ \nM = \left( \begin{matrix} {I}_{n} & - A & {0}_{n} \\ {0}_{n} & {I}_{n} & - B \\ {0}_{n} & {0}_{n} & {I}_{n} \end{matrix}\right) \in {\mathbf{M}}_{3n}\left( \mathbb{...
Yes
Proposition 11.3 The complexity of the multiplication of \( n \times n \) matrices is \( O\left( {n}^{\alpha }\right) \), with \( \alpha = \log 7/\log 2 = {2.807}\ldots \) More precisely,\n\n\[ \n{P}_{n} \leq \frac{147}{2}{n}^{\log 7/\log 2}.\n\]
Here is Strassen’s formula [37]. Let \( M, N \in {\mathbf{M}}_{2}\left( A\right) \), with\n\n\[ \nM = \left( \begin{array}{ll} a & b \\ c & d \end{array}\right) ,\;N = \left( \begin{array}{ll} x & y \\ z & t \end{array}\right) .\n\]\n\nOne first forms the expressions \( {x}_{1} = \left( {a + d}\right) \left( {x + t}\ri...
Yes
Theorem 11.2 Let \( M \in {\mathbf{{SPD}}}_{n} \) . Then there exists a unique lower-triangular matrix \( L \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \), with strictly positive diagonal entries, satisfying \( M = L{L}^{T} \) .
Proof. Uniqueness. If \( {L}_{1} \) and \( {L}_{2} \) have the properties stated above, then \( {I}_{n} = L{L}^{T} \) , for \( L = {L}_{2}^{-1}{L}_{1} \), which still has the same form. In other words, \( L = {L}^{-T} \), where both sides are triangular matrices, but of opposite types (lower and upper). This equality s...
Yes
Proposition 11.4 Let \( M \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) be given. Then there exist a unitary matrix \( Q \) and an upper-triangular matrix \( R \) ; the diagonal entries of the latter real positive, such that \( M = {QR} \) . This factorization is unique.
Proof. Uniqueness. If \( \left( {{Q}_{1},{R}_{1}}\right) \) and \( \left( {{Q}_{2},{R}_{2}}\right) \) give two factorizations, then \( Q = R \) with \( Q \mathrel{\text{:=}} {Q}_{2}^{-1}{Q}_{1} \) and \( R \mathrel{\text{:=}} {R}_{2}{R}_{1}^{-1} \) . Because \( Q \) is unitary, that is, \( {Q}^{ * } = {Q}^{-1} \), this...
Yes
Theorem 11.5 Let \( A \in {M}_{n \times m}\left( \mathbb{C}\right) \) be given. There exists a unique matrix \( {A}^{ \dagger } \in \) \( {M}_{m \times n}\left( \mathbb{C}\right) \), called the Moore-Penrose generalized inverse, satisfying the following four properties.\n\n1. \( A{A}^{ \dagger }A = A \) .\n\n2. \( {A}^...
Proof. We first remark that if \( X \) satisfies these four properties, and if \( U \in {\mathbf{U}}_{n} \) , \( V \in {\mathbf{U}}_{m} \), then \( {V}^{ * }X{U}^{ * } \) is a generalized inverse of \( {UAV} \) . Therefore, existence and uniqueness need to be proved for only a single representative \( D \) of the equiv...
Yes
Proposition 11.5 The following equalities hold for the generalized inverse:\n\n\[ \n{\left( \lambda A\right) }^{ \dagger } = \frac{1}{\lambda }{A}^{ \dagger }\;\left( {\lambda \neq 0}\right) ,\;{\left( {A}^{ \dagger }\right) }^{ \dagger } = A,\;{\left( {A}^{ \dagger }\right) }^{ * } = {\left( {A}^{ * }\right) }^{ \dagg...
Because \( {\left( A{A}^{ \dagger }\right) }^{2} = A{A}^{ \dagger } \), the matrix \( A{A}^{ \dagger } \) is a projector, which can therefore be described in terms of its range and kernel. Because \( A{A}^{ \dagger } \) is Hermitian, these subspaces are orthogonal to each other. Obviously, \( R\left( {A{A}^{ \dagger }}...
Yes
Proposition 11.6 The system (11.6) is solvable if and only if \( b = M{M}^{ \dagger }b \) . When it is solvable, its general solution is \( x = {M}^{ \dagger }b + \left( {{I}_{m} - {M}^{ \dagger }M}\right) z \), where \( z \) ranges \( {\mathbb{C}}^{m} \) . Finally, the special solution \( {x}_{0} \mathrel{\text{:=}} {...
There remains to prove that \( {x}_{0} \) has the smallest norm among the solutions. That comes from the Pythagorean theorem and from the fact that \( R\left( {M}^{ \dagger }\right) = R\left( {{M}^{ \dagger }M}\right) = \) \( {\left( \ker M\right) }^{ \bot } \) .
No
Proposition 12.1 An iterative method is convergent if and only if \( \rho \left( {{M}^{-1}N}\right) < 1 \) .
Proof. If the method is convergent, then for \( b = 0 \) ,\n\n\[ \mathop{\lim }\limits_{{m \rightarrow + \infty }}{\left( {M}^{-1}N\right) }^{m}{x}^{0} = 0 \]\n\nfor every \( {x}^{0} \in {K}^{n} \) . In other words,\n\n\[ \mathop{\lim }\limits_{{m \rightarrow + \infty }}{\left( {M}^{-1}N\right) }^{m} = 0 \]\n\nFrom Pro...
Yes
Proposition 12.2 We have \( \rho \left( {\mathcal{L}}_{\omega }\right) \geq \left| {\omega - 1}\right| \) . In particular, if the relaxation method converges for a matrix \( A \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) and a parameter \( \omega \in \mathbb{C} \), then\n\n\[ \left| {\omega - 1}\right| < 1\text{.} \...
Proof. If the method is convergent, we have \( \rho \left( {\mathcal{L}}_{\omega }\right) < 1 \) . However,\n\n\[ \det {\mathcal{L}}_{\omega } = \frac{\det \left( {\left( {1 - \omega }\right) D + {\omega F}}\right) }{\det \left( {D - {\omega E}}\right) } = \frac{\det \left( {\left( {1 - \omega }\right) D}\right) }{\det...
Yes
Proposition 12.3 Under Hypothesis (1) or (2), the Jacobi method converges, as well as the relaxation method, for every \( \omega \in (0,1\rbrack \) .
Proof. Jacobi method: The matrix \( J = {D}^{-1}\left( {E + F}\right) \) is clearly irreducible when \( A \) is so. Furthermore, \[ \mathop{\sum }\limits_{{j = 1}}^{n}\left| {J}_{ij}\right| \leq 1,\;i = 1,\ldots, n \] in which all inequalities are strict if (1) holds, and at least one inequality is strict under the hyp...
Yes
Lemma 20. If \( A \) and \( {M}^{ * } + N \) are Hermitian positive-definite (in a decomposition \( A = M - N) \), then \( \rho \left( {{M}^{-1}N}\right) < 1 \) .
Proof. Let us remark first that \( {M}^{ * } + N = {M}^{ * } + M - A \) is necessarily Hermitian when \( A \) is so.\n\nIt is therefore enough to show that \( {\begin{Vmatrix}{M}^{-1}Nx\end{Vmatrix}}_{A} < \parallel x{\parallel }_{A} \) for every nonzero \( x \in {\mathbb{C}}^{n} \) , where \( \parallel \cdot {\paralle...
Yes
Theorem 12.1 If \( A \) is Hermitian positive-definite, then the relaxation method converges if and only if \( \left| {\omega - 1}\right| < 1 \) .
Proof. We have seen in Proposition 12.2 that the convergence implies \( \left| {\omega - 1}\right| < 1 \) . Let us see the converse. We have \( {E}^{ * } = F \) and \( {D}^{ * } = D \) . Thus\n\n\[ \n{M}^{ * } + N = \left( {\frac{1}{\omega } + \frac{1}{\bar{\omega }} - 1}\right) D = \frac{1 - {\left| \omega - 1\right| ...
Yes
Lemma 21. Let \( \mu \) be a nonzero complex number and \( C \) a tridiagonal matrix, of diagonal \( {C}_{0} \), of upper-triangular part \( {C}_{ + } \) and lower-triangular part \( {C}_{ - } \) . Then\n\n\[ \det C = \det \left( {{C}_{0} + \frac{1}{\mu }{C}_{ - } + \mu {C}_{ + }}\right) . \]
Proof. It is enough to observe that the matrix \( C \) is conjugate to\n\n\[ {C}_{0} + \frac{1}{\mu }{C}_{ - } + \mu {C}_{ + } \]\n\nby the nonsingular matrix\n\n\[ {Q}_{\mu } = \left( \begin{matrix} \mu & & & & \\ & {\mu }^{2} & & 0 & \\ & & \ddots & & \\ & 0 & & \ddots & \\ & & & & {\mu }^{n} \end{matrix}\right) . \]
Yes
Proposition 12.4 If \( A \) is tridiagonal and \( D \) invertible, then:\n\n1. \( {P}_{G}\left( {X}^{2}\right) = {X}^{n}{P}_{J}\left( X\right) \), where \( {P}_{G} \) is the characteristic polynomial of the Gauss-Seidel matrix \( G \) ,\n\n2. \( \rho \left( G\right) = \rho {\left( J\right) }^{2} \) ,\n\n3. The Gauss-Se...
Proof. Point 1 comes from Lemma 22. The spectrum of \( G \) is thus formed of \( \lambda = 0 \) (which is of multiplicity \( \left\lbrack {\left( {n + 1}\right) /2}\right\rbrack \) at least) and of squares of the eigenvalues of \( J \) , which proves Point 2. Point 3 follows immediately. Finally, if \( \mu \in \mathrm{...
Yes
Theorem 12.2 (See Figure 12.1) Suppose that \( A \) is tridiagonal, \( D \) is invertible, and that the eigenvalues of \( J \) are real and belong to \( \left( {-1,1}\right) \) . Assume also that \( \omega \in \mathbb{R} \) .\n\nThen the relaxation method converges if and only if \( \omega \in \left( {0,2}\right) \) . ...
![9d1472cf-1919-4f49-b930-ef2d2345b792_240_0.jpg](images/9d1472cf-1919-4f49-b930-ef2d2345b792_240_0.jpg)\n\nFig. 12.1 \( \rho \left( {\mathcal{L}}_{\omega }\right) \) in the tridiagonal case.
Yes
Lemma 23. Let us denote by \( {\lambda }_{n} \geq \cdots \geq {\lambda }_{1}\left( { > 0}\right) \) the eigenvalues of \( A \) . If \( k \leq \ell \), then\n\n\[ E\left( {{x}_{k} - \bar{x}}\right) \leq E\left( {e}_{0}\right) \cdot \mathop{\min }\limits_{{\deg Q \leq k - 1}}\mathop{\max }\limits_{j}{\left| 1 + {\lambda ...
Proof. Let us compute\n\n\[ E\left( {{x}_{k} - \bar{x}}\right) = \min \left\{ {E\left( {x - \bar{x}}\right) \mid x \in {x}_{0} + {\mathcal{H}}_{k}}\right\} \]\n\n\[ = \min \left\{ {E\left( {{e}_{0} + y}\right) \mid y \in {\mathcal{H}}_{k}}\right\} \]\n\n\[ = \min \left\{ {E\left( {\left( {{I}_{n} + {AQ}\left( A\right) ...
Yes
For every matrix \( M \in {\mathbf{M}}_{n}\left( \mathbb{C}\right) \) there exists a unitary transformation \( U \) such that \( {U}^{-1}{MU} \) is a Hessenberg matrix. If \( M \in {\mathbf{M}}_{n}\left( \mathbb{R}\right) \), one may take \( U \in {\mathbf{O}}_{n} \) .
Proof. Let \( X \in {\mathbb{C}}^{m} \) be a unit vector: \( {X}^{ * }X = 1 \) . The matrix of the unitary (orthogonal) symmetry with respect to the hyperplane \( {X}^{ \bot } \) is \( S = {I}_{m} - {2X}{X}^{ * } \) . In fact, \( {SX} = X - \) \( {2X} = - X \), and \( Y \in {X}^{ \bot } \) (i.e., \( {X}^{ * }Y = 0 \) )...
Yes
Lemma 24. Let \( A \in {\mathbf{{GL}}}_{n}\left( K\right) \) be given, with \( K = \mathbb{R} \) or \( \mathbb{C} \) . Let \( {A}_{k} = {Q}_{k}{R}_{k} \) be the sequence of matrices given by the QR algorithm. Let us define \( {P}_{k} = {Q}_{0}\cdots {Q}_{k - 1} \) and \( {U}_{k} = {R}_{k - 1}\cdots {R}_{0} \) . Then \(...
Proof. From (13.2), we have \( {A}_{k} = {P}_{k}^{-1}A{P}_{k} \) ; that is, \( {P}_{k}{A}_{k} = A{P}_{k} \) . Then\n\n\[ {P}_{k + 1}{U}_{k + 1} = {P}_{k}{Q}_{k}{R}_{k}{U}_{k} = {P}_{k}{A}_{k}{U}_{k} = A{P}_{k}{U}_{k}. \]\n\nBy induction, \( {P}_{k}{U}_{k} = {A}^{k} \) . However, \( {P}_{k} \in {\mathbf{U}}_{n} \) and \...
Yes
Theorem 13.2 Let \( A \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) be given. Assume that the moduli of the eigenvalues of \( A \) are distinct:\n\n\[ \n\left| {\lambda }_{1}\right| > \left| {\lambda }_{2}\right| > \cdots > \left| {\lambda }_{n}\right| \;\left( { > 0}\right) .\n\]\n\nIn particular, the eigenvalues...
Proof. Let \( Y = {LU} \) be the factorization of \( Y \) . We also make use of the \( {QR} \) factorization of \( {Y}^{-1} : {Y}^{-1} = {QR} \) . Because \( {A}^{k} = {Y}^{-1}{D}^{k}Y \), we have \( {P}_{k}{U}_{k} = {Y}^{-1}{D}^{k}Y = \) \( {QR}{D}^{k}{LU} \) .\n\nThe matrix \( {D}^{k}L{D}^{-k} \) is lower-triangular ...
Yes
Theorem 13.3 Let \( A \in {\mathbf{{GL}}}_{n}\left( \mathbb{C}\right) \) be an irreducible Hessenberg matrix whose eigenvalues are of distinct moduli:\n\n\[ \left| {\lambda }_{1}\right| > \cdots > \left| {\lambda }_{n}\right| \;\left( { > 0}\right) \]\n\nThen the QR method converges; that is, the lower-triangular part ...
Proof. In the light of Theorem 13.2, it is enough to show that the matrix \( Y \) in the previous proof admits an \( {LU} \) factorization. We have \( {YA} = \operatorname{diag}\left( {{\lambda }_{1},\ldots ,{\lambda }_{n}}\right) Y \) . The rows of \( Y \) are thus the left eigenvectors: \( {\ell }_{j}A = {\lambda }_{...
Yes
Corollary 13.1 If \( A \in {\mathbf{{HPD}}}_{n} \) and if \( {A}_{0} \) is a Hessenberg matrix, unitarily similar to \( A \) (e.g., a matrix obtained by Householder’s method), then the sequence \( {A}_{k} \) defined by the QR method converges to a diagonal matrix whose diagonal entries are the eigenvalues of \( A \) .
Indeed, the lower-triangular part converges, hence the whole matrix, because it is Hermitian.
No
Lemma 25. We have\n\n\[ \n{\begin{Vmatrix}{E}_{k + 1}\end{Vmatrix}}^{2} = {\begin{Vmatrix}{E}_{k}\end{Vmatrix}}^{2} - 2{\left( {a}_{pq}^{\left( k\right) }\right) }^{2}. \n\]
Proof. It suffices to redo the calculations of Section 13.4.1, noting that\n\n\[ \n{k}_{ip}^{2} + {k}_{iq}^{2} = {h}_{ip}^{2} + {h}_{iq}^{2} \n\]\n\nwhenever \( i \neq p, q \), whereas \( {k}_{pq}^{2} = 0 \) .
Yes