aboutsummaryrefslogtreecommitdiffstats
path: root/buch/chapters/10-vektorenmatrizen/hadamard.tex
blob: c6380dcdc8dfa1c417c58cd418d11a2f27c00587 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
%
% hadamard.tex
%
% (c) 2021 Prof Dr Andreas Müller, OST Ostschweizer Fachhochschule
%
\section{Hadamard-Algebra
\label{buch:section:hadamard-algebra}}
\rhead{Hadamard-Algebra}
Das Matrizenprodukt ist nicht die einzige Möglichkeit, ein Produkt auf
Vektoren oder Matrizen zu definieren.
In diesem Abschnitt soll das Hadamard-Produkt beschrieben werden,
welches zu einer kommutativen-Algebra-Struktur führt.

%
% Definition des Hadamard-Produktes
%
\subsection{Hadamard-Produkt
\label{buch:vektorenmatrizen:subsection:hadamard-produkt}}
Im Folgenden werden wir $\Bbbk^n =M_{n\times 1}(\Bbbk)$ setzen
und den Fall der Vektoren nicht mehr separat diskutieren.
Die Addition und Multiplikation mit Skalaren ist in 
$M_{m\times n}(\Bbbk)$ komponentenweise definiert.
Wir können natürlich auch ein Produkt von Matrizen komponentenweise definieren,
dies ist das Hadamard-Produkt.

\begin{definition}
Das {\em Hadamard-Produkt} zweier Matrizen
\index{Hadamard-Produkt}%
$A,B\in M_{m\times n}(\Bbbk)$ ist definiert als die Matrix
$A\odot B$ 
mit den Komponenten
\[
(A\odot B)_{i\!j} = (A)_{i\!j} (B)_{i\!j}.
\]
Wir nennen $M_{m\times n}(\Bbbk)$ mit der Multiplikation $\odot$ 
auch die {\em Hadamard-Algebra} $H_{m\times n}(\Bbbk)$.
\index{Hadamard-Algebra}%
\end{definition}

Dies ist jedoch nur interessant, wenn $M_{m\times n}(\Bbbk)$ mit diesem
Produkt eine interessante algebraische Struktur erhält.
Dazu müssen die üblichen Verträglichkeitsgesetze zwischen den
Vektorraumoperationen von $M_{m\times n}(\Bbbk)$ und dem neuen Produkt
gelten, erst so erhalten wir eine Algebra.
Da alle Operationen elementweise definiert sind, muss man auch alle
Rechengesetze nur elementweise prüfen.
Es gilt
\[
\begin{aligned}
A\odot(B\odot C) &= (A\odot B)\odot C
&&\Leftrightarrow&
a_{i\!j}(b_{i\!j}c_{i\!j}) &= (a_{i\!j}b_{i\!j})c_{i\!j}
\\
A\odot(B+C) &= A\odot B + A\odot C
&&\Leftrightarrow&
a_{i\!j}(b_{i\!j}+c_{i\!j}) &= a_{i\!j}b_{i\!j} + a_{i\!j}c_{i\!j}
\\
(A+B)\odot C&=A\odot C+B\odot C
&&\Leftrightarrow&
(a_{i\!j}+b_{i\!j})c_{i\!j}&=a_{i\!j}c_{i\!j} + b_{i\!j}c_{i\!j}
\\
(\lambda A)\odot B &= \lambda (A\odot B)
&&\Leftrightarrow&
(\lambda a_{i\!j})b_{i\!j}&=\lambda(a_{i\!j}b_{i\!j})
\\
A\odot(\lambda B)&=\lambda(A\odot B)
&&\Leftrightarrow&
a_{i\!j}(\lambda b_{i\!j})&=\lambda(a_{i\!j}b_{i\!j})
\end{aligned}
\]
für alle $i,j$.

Das Hadamard-Produkt ist kommutativ, da die Multiplikation in $\Bbbk$
kommuativ ist.
Das Hadamard-Produkt kann auch für Matrizen mit Einträgen in einem
Ring definiert werden, in diesem Fall ist es möglich, dass die entstehende
Algebra nicht kommutativ ist.

Die Hadamard-Algebra hat auch ein Eins-Elemente, nämlich die Matrix,
die aus lauter Einsen besteht.

\begin{definition}
Die sogenannte {\em Einsmatrix} $U$ ist die Matrix
\index{Einsmatrix}
\[
U=\begin{pmatrix}
1&1&\dots&1\\
\vdots&\vdots&\ddots&\vdots\\
1&1&\dots&1
\end{pmatrix}
\in
M_{m\times n}(\Bbbk)
\]
mit lauter Einträgen $1\in\Bbbk$.
\end{definition}

Die Hadamard-Algebra ist ein Spezialfall der Algebra der Funktionen
$\Bbbk^X$.
Ordnet man dem Vektor $v\in \Bbbk^n$ mit den Komponenten $v_i$ 
die Abbildung
\[
v\colon [n] \to \Bbbk: i \mapsto v_i
\]
zu, dann geht die Addition von Vektoren in die Addition von
Funktionen über, die Multiplikation von Skalaren mit Vektoren
geht in die Multiplikation von Funktionen mit Skalaren über 
und die Hadamard-Multiplikation geht über in das Produkt von
Funktionen.

Auch die Hadamard-Algebra $H_{m\times n}(\Bbbk)$ kann als Funktionenalgebra
betrachtet werden.
Einer Matrix $A\in H_{m\times n}(\Bbbk)$ ordnet man die Funktion
\[
a\colon [m]\times [n] : (i,j) \mapsto a_{i\!j}
\]
zu.
Dabei gehen die Algebraoperationen von $H_{m\times n}(\Bbbk)$ über
in die Algebraoperationen der Funktionenalgebra $\Bbbk^{[m]\times [n]}$.
Die Einsmatrix der Hadamard-Algebra wird dabei zur konstanten
Funktion $1$ auf $[m]\times[n]$.

\subsection{Hadamard-Produkt und Matrizenalgebra
\label{buch:vektorenmatrizen:subsection:vertraeglichkeit}}
Es ist nur in Ausnahmefällen sinnvoll, Hadamard-Produkt und Matrizen-Produkt
gleichzeitig zu verwenden.
Das liegt daran, dass die beiden Produkte sich überhaupt nicht 
vertragen.

\subsubsection{Unverträglichkeit von Hadamard- und Matrizen-Produkt}
Das Hadamard-Produkt und das gewöhnliche Matrizenprodukt sind
in keiner Weise kompatibel.
Die beiden Matrizen 
\[
A=\begin{pmatrix}3&4\\4&5\end{pmatrix}
\qquad\text{und}\qquad
B=\begin{pmatrix}-5&4\\4&-3\end{pmatrix}
\]
sind inverse Matrizen bezüglich des Matrizenproduktes, also
$AB=I$.
Für das Hadamard-Produkt gilt dagegen
\[
A\odot B
=
\begin{pmatrix}
-15& 16\\
 16&-15
\end{pmatrix}.
\]
Die Inverse einer Matrix $A$ bezüglich des Hadamard-Produktes hat
die Einträge $a_{i\!j}^{-1}$.
Die Matrix $I$ ist bezüglich des gewöhnlichen Matrizenproduktes
invertierbar, aber sie ist bezüglich des Hadamard-Produktes nicht
invertierbar.
Umgekehrt ist die Einsmatrix $U$ invertierbar bezüglich des
Hadamard-Produktes, aber für $n>1$ nicht für das Matrizenprodukt.

\subsubsection{Einbettung der Hadamard-Algebra in eine Matrizenalgebra}
Hadamard-Algebren können als Unteralgebren einer Matrizenalgebra
betrachtet werden.
Der Operator $\operatorname{diag}$ bildet Vektoren ab in Diagonalmatrizen
nach der Regel
\[
\operatorname{diag}
\colon
\Bbbk^n \to M_n(\Bbbk)
:
\begin{pmatrix}
v_1\\
\vdots\\
v_n
\end{pmatrix}
\mapsto
\begin{pmatrix}
v_1&\dots&0\\
\vdots&\ddots&\vdots\\
0&\dots&v_n
\end{pmatrix}.
\]
Das Produkt von Diagonalmatrizen ist besonders einfach.
Für zwei Vektoren $a,b\in\Bbbk^n$ 
\[
a\odot b
=
\begin{pmatrix}
a_1b_1\\
\vdots\\
a_nb_n
\end{pmatrix}
\mapsto
\begin{pmatrix}
a_1b_1&\dots&0\\
\vdots&\ddots&\vdots\\
0&\dots&a_nb_n
\end{pmatrix}
=
\begin{pmatrix}
a_1&\dots&0\\
\vdots&\ddots&\vdots\\
0&\dots&a_n
\end{pmatrix}
\begin{pmatrix}
b_1&\dots&0\\
\vdots&\ddots&\vdots\\
0&\dots&b_n
\end{pmatrix}.
\]
Das Hadamard-Produkt der Vektoren geht also über in das gewöhnliche
Matrizenprodukt der Diagonalmatrizen.

Für die Hadamard-Matrix ist die Einbettung etwas komplizierter.
Wir machen aus einer Matrix erst einen Vektor, den wir dann mit
dem Operator $\operatorname{diag}$ in eine Diagonalmatrix umwandeln:
\[
\begin{pmatrix}
a_{11}&\dots &a_{1n}\\
\vdots&\ddots&\vdots\\
a_{m1}&\dots &a_{mn}
\end{pmatrix}
\mapsto
\begin{pmatrix}
a_{11}\\
\vdots\\
a_{1n}\\
a_{21}\\
\vdots\\
a_{2n}\\
\vdots\\
a_{nn}
\end{pmatrix}.
\]
Bei dieser Abbildung geht die Hadamard-Multiplikation wieder in
das gewöhnliche Matrizenprodukt über.

\subsubsection{Transposition}
Das Hadamard-Produkt verträgt sich mit der Transposition:
\index{Transposition}%
\[
(A\odot B)^t = A^t \odot B^t.
\]
Insbesondere ist das Hadamard-Produkt zweier symmetrischer Matrizen auch 
wieder symmetrisch.

\subsubsection{Frobenius-Norm}
Das Hadamard-Produkt in der Hadamard-Algebra $H_{m\times n}(\mathbb{R})$
nimmt keine Rücksicht auf die Dimensionen einer Matrix und ist nicht
unterscheidbar von $\mathbb{R}^{m\times n}$ mit dem Hadamard-Produkt.
Daher darf auch der Begriff einer mit den algebraischen Operationen
verträglichen Norm nicht von den spezifischen Dimensionen $m$ und $n$ abhängen.
Dies führt auf die folgende Definition einer Norm.

\begin{definition}
Die {\em Frobenius-Norm} einer Matrix $A\in H_{m\times n}(\mathbb{R})$
\index{Frobenius-Norm}%
mit den Einträgen $(a_{i\!j})=A$ ist
\[
\| A\|_F
=
\sqrt{
\sum_{i,j} a_{i\!j}^2
}.
\]
Das {\em Frobenius-Skalarprodukt} zweier Matrizen
$A,B\in H_{m\times n}(\mathbb{R})$
ist
\[
\langle A,B\rangle_F
=
\sum_{i,j} a_{i\!j} b_{i\!j}
=
\operatorname{Spur} A^t B
\]
und es gilt $\|A\|_F = \sqrt{\langle A,A\rangle}$.
\end{definition}

Für komplexe Matrizen muss die Definition angepasst werden, damit 
das Skalarprodukt sesquilinear und positiv definit wird.

\begin{definition}
Die {\em komplexe Frobenius-Norm} einer Matrix $A\in H_{m\times n}(\mathbb{C})$
ist
\[
\| A\|
=
\sqrt{
\sum_{i,j} |a_{i\!j}|^2
}
=
\sqrt{
\sum_{i,u} \overline{a}_{i\!j} a_{i\!j}
}
\]
das {\em komplexe Frobenius-Skalarprodukt} zweier Matrizen
$A,B\in H_{m\times n}(\mathbb{C})$ ist das Produkt
\[
\langle A,B\rangle_F
=
\sum_{i,j}\overline{a}_{i\!j} b_{i\!j}
=
\operatorname{Spur} (A^* B)
\]
und es gilt $\|A\|_F = \sqrt{\langle A,A\rangle}$.
\end{definition}