-
Notifications
You must be signed in to change notification settings - Fork 2
/
lista10.tex
567 lines (532 loc) · 26.4 KB
/
lista10.tex
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
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
% Filename: lista10.tex
%
% This code is part of 'Solutions for MT402, Matrizes'
%
% Description: This file corresponds to the solutions of homework sheet 10.
%
% Created: 24.04.12 05:54:57 PM
% Last Change: 04.06.12 05:12:27 PM
%
% Authors:
% - Raniere Silva (2012): initial version
%
% Copyright (c) 2012 Raniere Silva <[email protected]>
%
% This work is licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/3.0/ or send a letter to Creative Commons, 444 Castro Street, Suite 900, Mountain View, California, 94041, USA.
%
% This work is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
%
\begin{questions}
\question Considere $A : m \times n$, $m \geq n$ e $\bar{x} \in \mathbb{R}^n$. Demonstre que: $\| b - A \bar{x} \|_2 = \min_{y \in \mathbb{R}^n} \| b - A y \|_2$ se e somente se $(b - A \bar{x}) \in \text{Im}(A)^\perp$.
\begin{solution}
Na p\'{a}gina 438 do Meyer\nocite{Meyer:2000:matrix}, temos que o problema de m\'{i}nimos quadrados consiste em encontrar $x$ tal que $A x = P_{\text{Im}(A)} b$. Mas este sistema \'{e} tal que
\begin{align*}
A x = P_{\text{Im}(A)} b &\leftrightarrow P_{\text{Im}(R)} A x = P_{\text{Im}(A)} P_{\text{Im}(A)} b = P_{\text{Im}(A)} b \\
&\leftrightarrow P_{\text{Im}(A)} \left( A x - b \right) = 0 \\
&\leftrightarrow \left( A x - b \right) \in \text{N}(P_{\text{Im}(A)}) = \text{Im}(A)^\perp = \text{N}(A^t).
\end{align*}
\end{solution}
\question Seja $\bar{x} \in \mathbb{R}^n$. Ent\~{a}o, $\bar{x}$ resolve o problema de quadrados m\'{i}nimos para $A x = b$, se e somente se $A^t A \bar{x} = A^t b$.
\begin{solution}
Se $\bar{x}$ \'{e} a solu\c{c}\~{a}o do problema de quadrados m\'{i}nimos de $A x = b$ ent\~{a}o
\begin{align*}
\bar{x} \mid \| A \bar{x} - b \| = \min \| A x - b \| &\leftrightarrow Q^t A \bar{x} = Q^t b \\
&\leftrightarrow R \bar{x} = Q^t b \\
&\leftrightarrow R^t R \bar{x} = R^t Q^t b \\
&\leftrightarrow R^t Q^t Q R \bar{x} = R^t Q^t b \\
&\leftrightarrow A^t A \bar{x} = A^t b.
\end{align*}
\end{solution}
\question Considere $A : m \times n$, $m \geq n$ e sua fatora\c{c}\~{a}o ortogonal dada por $A = Q R$ com $Q : m \times m$ ortogonal e $R : m \times n$, com coeficientes $r_{ij} = 0$, $i > j$. Reescreva a solu\c{c}\~{a}o de quadrados m\'{i}nimos $\bar{x} = \left( A^t A \right)^{-1} A^t b$ e a proje\c{c}\~{a}o ortogonal de $b \in \text{Im}(A)$ usando a fatora\c{c}\~{a}o ortogonal de $A$. Relacione as express\~{o}es obtidas com a solu\c{c}\~{a}o de quadrados m\'{i}nimos obtida diretamente da fatora\c{c}\~{a}o ortogonal: $\| b - A x \|_2 = \| Q^t \left( b - A x \right) \|_2$ \ldots
\begin{solution}
Podemos reescrever a solu\c{c}\~{a}o da seguinte forma:
\begin{align*}
\bar{x} &= \left( A^t A \right)^{-1} A^t b \\
&= \left( R^t Q^t Q R \right)^{-1} R^t Q^t b \\
&= \left( R^t R \right)^{-1} R^t Q^t b \\
&= R^{-1} R^{-t} R^t Q^t b \\
&= R^{-1} Q^t b.
\end{align*}
E a proje\c{c}\~{a}o ortogonal de $b \in \text{Im}(A)$ corresponde a
\begin{align*}
P_{\text{Im}(A)} b &= (I - Q) b.
\end{align*}
\end{solution}
\question Considere
\begin{align*}
A = \begin{bmatrix}
1 & 2 \\
0 & 1 \\
1 & 0
\end{bmatrix}.
\end{align*}
\begin{parts}
\part Construa o projetor ortogonal $P$ sobre o subespa\c{c}o $\text{Im}(A)$. Obtenha a proje\c{c}\~{a}o do vetor $v = \left( 1, -1, 2 \right)^t$ em $\text{Im}(A)$. Idem para o vetor $w = \left( 1, 1, -1 \right)^t$.
\begin{solution}
Na p\'{a}gina 430 do Meyer\nocite{Meyer:2000:matrix}, temos:
\begin{quote}
Seja $\mathcal{M}$ um subespa\c{c}o $r$-dimensional de $\mathbb{R}$, e seja as colunas de $M_{n \times r}$ e $N_{n \times {n - r}}$ as bases de $\mathcal{M}$ e $\mathcal{M}^\perp$, respectivamente. Ent\~{a}o os projetores ortogonais de $\mathcal{M}$ e $\mathcal{M}^\perp$ s\~{a}o
\begin{align*}
P_{\mathcal{M}} &= M \left( M^t M \right)^{-1} M^t \text{ e } P_{\mathcal{M}^\perp} = N \left( N^t N \right)^{-1} N^t.
\end{align*}
Se $M$ e $N$ s\~{a}o bases ortonormais para $\mathcal{M}$ e $\mathcal{M}^\perp$, ent\~{a}o
\begin{align*}
P_{\mathcal{M}} &= M M^t \text{ e } P_{\mathcal{M}^\perp} = N N^t.
\end{align*}
\end{quote}
Ent\~{a}o,
\begin{align*}
P &= A A^t = \begin{bmatrix}
5 & 2 & 1 \\
2 & 1 & 0 \\
1 & 0 & 1
\end{bmatrix}.
\end{align*}
E portanto,
\begin{align*}
P v &= \begin{bmatrix}
5 & 2 & 1 \\
2 & 1 & 0 \\
1 & 0 & 1
\end{bmatrix} \begin{bmatrix}
1 \\
-1 \\
2
\end{bmatrix} = \begin{bmatrix}
5 \\
1 \\
3
\end{bmatrix},
P w &= \begin{bmatrix}
5 & 2 & 1 \\
2 & 1 & 0 \\
1 & 0 & 1
\end{bmatrix} \begin{bmatrix}
1 \\
1 \\
-1
\end{bmatrix} = \begin{bmatrix}
6 \\
3 \\
0
\end{bmatrix}
\end{align*}
Geometricamente temos
\begin{center}
\begin{tikzpicture}
\coordinate (O) at (0,0,0);
\draw[->] (0,0,0) -- (8,0,0) node[anchor=north east]{$x_1$};
\draw[->] (0,0,0) -- (0,4,0) node[anchor=north west]{$x_2$};
\draw[->] (0,0,0) -- (0,0,8) node[anchor=south]{$x_3$};
\fill[color=gray!50] (0, 0, 0) -- (8, 0, 8) -- (8, 4, 0) -- (0, 0, 0);
\draw[->, color=blue] (0, 0, 0) -- (1, -1, 2) node[below right] {$v$};
\draw[->, color=blue] (0, 0, 0) -- (5, 1, 3) node[below right] {$P v$};
\draw[->, color=red] (0, 0, 0) -- (1, 1, -1) node[below right] {$w$};
\draw[->, color=red] (0, 0, 0) -- (6, 3, 0) node[below right] {$P w$};
\draw[->] (0, 0, 0) -- (1, 0, 1) node[below right] {$A_{\mdot 1}$};
\draw[->] (0, 0, 0) -- (2, 1, 0) node[below right] {$A_{\mdot 2}$};
\end{tikzpicture}
\end{center}
\end{solution}
\part Obtenha a fatora\c{c}\~{a}o QR reduzida de $A$.
\begin{solution}
Para obter a fatora\c{c}\~{a}o QR reduzida podemos utilizar o M\'{e}todo de Gram-Schmidt Modificado.
O algoritmo a seguir, retirado do Golub\nocite{Golub:1996:matrix} (p\'{a}gina 231), tem como entrada $A \in \mathbb{R}^{m \times n}$ de posto completo e calcula a fatora\c{c}\~{a}o QR reduzida de $A$ onde $Q \in \mathbb{R}^{m \times n}$ tem colunas ortonormais e $R \in \mathbb{R}^{n \times n}$ \'{e} triangular superior.
\begin{algorithmic}
\For{$k = 1, \ldots, n$}
\State $R(k, k) \leftarrow \| A(1:m, k) \|_2$
\State $Q(1:m, k) \leftarrow A(1:m, k) / R(k,k)$
\For{$j = k + 1 : n$}
\State $R(k, j) \leftarrow Q(1:m, k)^t A(1:m, j)$
\State $A(1:m, j) \leftarrow A(1:m, j) - Q(1:m, k) R(k, j)$
\EndFor
\EndFor
\end{algorithmic}
Aplicando ao algoritmo acima, temos
\begin{enumerate}
\item $k = 1$:
\begin{align*}
R &\leftarrow \begin{bmatrix}
\sqrt{2} & \left( 1 / \sqrt{2} \right) 2 + 0 \left( 1 \right) + \left( 1 / \sqrt{2} \right) 0 \\
0 & 0
\end{bmatrix} = \begin{bmatrix}
\sqrt{2} & 2 / \sqrt{2} \\
0 & 0
\end{bmatrix}, \\
Q &\leftarrow \begin{bmatrix}
1 / \sqrt{2} & 0 \\
0 & 0 \\
1 / \sqrt{2} & 0
\end{bmatrix}, \\
A &\leftarrow \begin{bmatrix}
1 & 2 - \left( 1 / \sqrt{2} \right) \left( 2 / \sqrt{2} \right) \\
0 & 1 - 0 \left( 2 / \sqrt{2} \right) \\
1 & 0 - \left( 1 / \sqrt{2} \right) \left( 2 / \sqrt{2} \right)
\end{bmatrix} = \begin{bmatrix}
1 & 1 \\
0 & 1 \\
1 & -1
\end{bmatrix}.
\end{align*}
\item $k = 2$:
\begin{align*}
R &\leftarrow \begin{bmatrix}
\sqrt{2} & 2 / \sqrt{2} \\
0 & \sqrt{3}
\end{bmatrix}, \\
Q &\leftarrow \begin{bmatrix}
1 / \sqrt{2} & 1 / \sqrt{3} \\
0 & 1 / \sqrt{3} \\
1 / \sqrt{2} & -1 / \sqrt{3}
\end{bmatrix}, \\
A &\leftarrow \begin{bmatrix}
1 & 1 \\
0 & 1 \\
1 & -1
\end{bmatrix}.
\end{align*}
\end{enumerate}
\end{solution}
\part Obtenha a fatora\c{c}\~{a}o QR de $A$ atrav\'{e}s dos m\'{e}todos de Householder e Givens.
\begin{solution}
Primeiro vamos obter a fatora\c{c}\~{a}o QR atrav\'{e}es do m\'{e}todo de Householder.
O algoritmo a seguir, retirado do Golub\nocite{Golub:1996:matrix} (p\'{a}gina 224), tem como entrada $A \in \mathbb{R}^{m \times n}$ com $m \geq n$ e encontra as matrizes de Householder $H_1, \ldots, H_n$ tal que se $Q = H_1 \ldots H_n$ ent\~{a}o $Q^t A = R$ que \'{e} triangular superior. A parte triangular superior da matriz $A$ \'{e} sobrescrita pela parte triangular superior de $R$ e as componentes $j + 1 : n$ do $j$-\'{e}simo vetor de Householder \'{e} armazenado em $A(j + 1:m, j)$, $j < m$.
\begin{algorithmic}
\For{$j = 1, \ldots, n$}
\State $[v, \beta] \leftarrow \text{\bf householder}(A(j:m, j))$
\State $A(j:m, j:n) \leftarrow \left( I_{m - j + 1} - \beta v v^t \right) A(j:m, j:n)$
\If{$j < m$}
\State $A(j + 1:m, j) \leftarrow v(2:m - j + 1)$
\EndIf
\EndFor
\end{algorithmic}
Aplicando uma vers\~{a}o modificada do algoritmo acima, temos
\begin{enumerate}
\item $j = 1$:
\begin{align*}
x &\leftarrow \begin{bmatrix}
1 \\
0 \\
1
\end{bmatrix}, v \leftarrow \begin{bmatrix}
1 + \sqrt{2} \\
0 \\
1
\end{bmatrix} \approx \begin{bmatrix}
2.41 \\
0 \\
1
\end{bmatrix}\\
A &\leftarrow A - (2 / v^t v) v (v^t A) = \begin{bmatrix}
1 & 2 \\
0 & 1 \\
1 & 0
\end{bmatrix} - (1 / 3.41) \begin{bmatrix}
8.23 & 11.64 \\
0.00 & 0.00 \\
3.41 & 4.82
\end{bmatrix} \\
&\quad = \begin{bmatrix}
1 & 2 \\
0 & 1 \\
1 & 0
\end{bmatrix} - \begin{bmatrix}
2.41 & 3.41 \\
0.00 & 0.00 \\
1.00 & 1.41
\end{bmatrix} = \begin{bmatrix}
-1.41 & - 1.41 \\
0 & 1 \\
0 & - 1.41
\end{bmatrix}
\end{align*}
\item $j = 2$:
\begin{align*}
x &\leftarrow \begin{bmatrix}
1 \\
-1.41
\end{bmatrix}, v \leftarrow \begin{bmatrix}
2.72 \\
-1.41
\end{bmatrix} \\
A(2:3, 2:2) &\leftarrow A(2:3, 2:2) - (2 / v^t v) v (v^t A(2:3, 2:2) \\
&\quad = \begin{bmatrix}
1 \\
-1.41
\end{bmatrix} - \begin{bmatrix}
2.72 \\
-1.41
\end{bmatrix} = \begin{bmatrix}
-1.72 \\
0
\end{bmatrix}, \\
A &\leftarrow \begin{bmatrix}
-1.41 & -1.41 \\
0 & -1.72 \\
0 & 0
\end{bmatrix}.
\end{align*}
\end{enumerate}
Agora vamos obter a fatora\c{c}\~{a}o QR atrav\'{e}es do m\'{e}todo de Givens.
O algoritmo a seguir, retirado do Golub\nocite{Golub:1996:matrix} (p\'{a}gina 227), tem como entrada $A \in \mathbb{R}^{m \times n}$ com $m \geq n$ e sobrescreve a matriz $A$ tal que $Q^t A = R$, onde $R$ \'{e} triangular superior e $Q$ \'{e} ortogonal.
\begin{algorithmic}
\For{$j = 1, \ldots, n$}
\For{$i = m : - 1 : j + 1$}
\State $[c, s] \leftarrow \text{\bf givens}(A(i - 1, j), A(i, j))$
\State $A(i - 1:i, j:n) \leftarrow \begin{bmatrix}
c & s \\
-s & c
\end{bmatrix}^t A(i - 1:i, j:n)$
\EndFor
\EndFor
\end{algorithmic}
Aplicando uma vers\~{a}o modificada do algoritmo acima com detec\c{c}\~{a}o de zero, temos
\begin{enumerate}
\item $i = 1$, $k = 3$ e $j = 1$:
$c = 1 / \sqrt{2}$ e $s = 1 / \sqrt{2}$. Portanto
\begin{align*}
A &\leftarrow \begin{bmatrix}
2 / \sqrt{2} & 2 / \sqrt{2} \\
0 & 1 \\
0 & - 2 / \sqrt{2}
\end{bmatrix}.
\end{align*}
\item $i = 1$, $k = 2$ e $j = 1$:
N\~{a}o precisa ser feito nada, pois $a_{21} = 0$.
\item $i = 2$, $k = 3$ e $j = 2$:
$c = 1 / \sqrt{2}$ e $s = -2 / \sqrt{6}$. Portanto
\begin{align*}
A &\leftarrow \begin{bmatrix}
2 / \sqrt{2} & 2 / \sqrt{2} \\
0 & 3 / \sqrt{3} \\
0 & 0
\end{bmatrix}.
\end{align*}
\end{enumerate}
\end{solution}
\part Obtenha a solu\c{c}\~{a}o de Quadrados M\'{i}nimos para $A x = v$, atrav\'{e}s da resolu\c{c}\~{a}o do sistema linear com equa\c{c}\~{o}es normais e atrav\'{e}s do processo de fatora\c{c}\~{o}es ortogonais.
\begin{solution}
Primeiro vamos resolver o sistema utilizando equa\c{c}\~{o}es normais, i.e., $A^t A x = A^t v$. Ent\~{a}o,
\begin{align*}
\begin{bmatrix}
1 & 0 & 1 \\
2 & 1 & 0
\end{bmatrix} \begin{bmatrix}
1 & 2 \\
0 & 1 \\
1 & 0
\end{bmatrix} x &= \begin{bmatrix}
1 & 0 & 1 \\
2 & 1 & 0
\end{bmatrix} \begin{bmatrix}
1 \\
-1 \\
2
\end{bmatrix} \\
\begin{bmatrix}
2 & 2 \\
2 & 5
\end{bmatrix} x &= \begin{bmatrix}
3 \\
1
\end{bmatrix} \\
\begin{bmatrix}
2 & 2 \\
0 & 3
\end{bmatrix} x &= \begin{bmatrix}
3 \\
-1
\end{bmatrix}.
\end{align*}
Portanto, $x_1 = 13/ 6$ e $x_2 = -2 / 3$.
Agora vamos resolver o problema com fatora\c{c}\~{o}es ortogonais (utilizando Givens).
Antes de resolvermos o sistema triangular, precisamos adequar o lado direito.
\begin{align*}
\begin{bmatrix}
1 & 0 & 0 \\
0 & 1 / \sqrt{3} & -2 / \sqrt{6} \\
0 & 2 / \sqrt{6} & 1 / \sqrt{3}
\end{bmatrix} \begin{bmatrix}
1 / \sqrt{2} & 0 & 1 / \sqrt{2} \\
0 & 1 & 0 \\
- 1 / \sqrt{2} & 0 & 1 \sqrt{2}
\end{bmatrix} \begin{bmatrix}
1 \\
-1 \\
2
\end{bmatrix} &= \begin{bmatrix}
1 & 0 & 0 \\
0 & 1 / \sqrt{3} & -2 / \sqrt{6} \\
0 & 2 / \sqrt{6} & 1 / \sqrt{3}
\end{bmatrix} \begin{bmatrix}
3 / \sqrt{2} \\
-1 \\
1 / \sqrt{2}
\end{bmatrix} \\
&= \begin{bmatrix}
3 / \sqrt{2} \\
-2 / \sqrt{3} \\
- 1 / \sqrt{6}
\end{bmatrix}
\end{align*}
Ent\~{a}o temos seguinte sistema triangular
\begin{align*}
\begin{bmatrix}
2 / \sqrt{2} & 2 / \sqrt{2} \\
0 & 3 / \sqrt{3}
\end{bmatrix} x &= \begin{bmatrix}
3 / \sqrt{2} \\
-2 / \sqrt{3}
\end{bmatrix}
\end{align*}
cuja solu\c{c}\~{a}o \'{e} $x_1 = 13 / 6$ e $x_2 = -2 / 3$.
\end{solution}
\end{parts}
\question Seja o sistema linear $A x = b$, $A : m \times n$, $m > n$, $\text{posto}(A) = r$. Fa\c{c}a um resumo sobre a solu\c{c}\~{a}o de Quadrados M\'{i}nimos deste sistema. Analise a exist\^{e}ncia e unicidade desta solu\c{c}\~{a}o usando as equa\c{c}\~{o}es normais e as fatora\c{c}\~{o}es ortogonais. Considere os casos $r = n$ e $r < n$ e inclua em sua an\'{a}lise o caso em que $b \in \text{Im}(A)$.
\begin{solution}
A solu\c{c}\~{a}o de Quadrados M\'{i}nimos \'{e} representada na tabela abaixo.
\begin{center}
\begin{tabular}{|p{.15\textwidth}|p{.15\textwidth}|p{.15\textwidth}|p{.15\textwidth}|p{.15\textwidth}|}
\cline{2-5}
\multicolumn{1}{c|}{} & \multicolumn{2}{|c|}{$r = n$} & \multicolumn{2}{|c|}{$r < n$} \\ \cline{2-5}
\multicolumn{1}{c|}{} & $b \in \text{Im}(A)$ & $b \not\in \text{Im}(A)$ & $b \in \text{Im}(A)$ & $b \not\in \text{Im}(A)$ \\ \hline
Quadrados M\'{i}nimos & Solu\c{c}\~{a}o \'{u}nica com res\'{i}duo igual de zero & Solu\c{c}\~{a}o \'{u}nica com res\'{i}duo diferente de zero & Infinitas solu\c{c}\~{o}es com res\'{i}duo igual a zero & Infinitas solu\c{c}\~{o}es com res\'{i}duo diferente de zero \\ \hline
Equa\c{c}\~{o}es Normais & Solu\c{c}\~{a}o \'{u}nica & Solu\c{c}\~{a}o \'{u}nica & $A^t A$ \'{e} singular & $A^t A$ \'{e} singular \\ \hline
Fatora\c{c}\~{o}es Ortogonais & Solu\c{c}\~{a}o \'{u}nica & Solu\c{c}\~{a}o \'{u}nica & $R$ \'{e} trapezoidal & $R$ \'{e} trapezoidal \\ \hline
\end{tabular}
\end{center}
\end{solution}
\question Sistema Linear Aumentado: Considere $A : m \times n$, $m > n$, $\text{posto}(A) = n$ e o sistema linear $A x = b$. O sistema linear aumentado \'{e} dado por $B y = c$, onde
\begin{align*}
B &= \begin{bmatrix}
I & A \\
A^t & 0
\end{bmatrix}, & y &= \begin{bmatrix}
r \\
x
\end{bmatrix}, & c &= \begin{bmatrix}
b \\
0
\end{bmatrix},
\end{align*}
onde $I : m \times m$ \'{e} a matriz identidade.
Este sistema linear admite solu\c{c}\~{a}o \'{u}nica (por qu\^{e}?). Relacione a solu\c{c}\~{a}o deste sistema linear dada pelos vetores: $r^*$ e $x^*$ com a solu\c{c}\~{a}o de problema de Quadrados M\'{i}nimos para $A x = b$. Quais m\'{e}todos podem ser aplicados para resolver $B y = c$?
\begin{solution}
Considerando o sistema linear
\begin{align*}
\begin{bmatrix}
I & A \\
A^t & 0
\end{bmatrix} \begin{bmatrix}
r \\
x
\end{bmatrix} &= \begin{bmatrix}
b \\
0
\end{bmatrix}
\end{align*}
temos que $A^t r = 0$ e $I r + A x = b$. Portanto $r = b - A x$ e substituindo na outra equa\c{c}\~{a}o obtemos
\begin{align*}
A^t \left( b - A x \right) &= 0 \leftrightarrow A^t A x = A^tb
\end{align*}
que corresponde ao sistema normal. Como $A$ tem posto completo sabemos que o sistema de equa\c{c}\~{o}es normais possue solu\c{c}\~{a}o \'{u}nica.
\end{solution}
\question Considere o sistema linear $A x = b$, $A : m \times n$, $m > n$ e a solu\c{c}\~{a}o de Quadrados M\'{i}nimos para este sistema, $\bar{x}$. Se multiplicarmos a equa\c{c}\~{a}o $j$ por $\alpha \neq 0$, obtemos assim o sistema $B x = d$, a solu\c{c}\~{a}o de Quadrados M\'{i}nimos permanece a mesma? Justifique.
\begin{solution}
A solu\c{c}\~{a}o do sistema $B x = d$ pode ser diferente da solu\c{c}\~{a}o do sistema $A x = b$. Esse fato pode ser observado ao comparar os sistemas de equa\c{c}\~{o}es normais de cada um dos sistemas pois
\begin{align*}
A &= \begin{bmatrix}
A_{1 \mdot} \\
A_{2 \mdot} \\
\vdots \\
A_{m \mdot}
\end{bmatrix}, \\
A^t A &= \begin{bmatrix}
A_{1 \mdot}^t & A_{2 \mdot}^t & \ldots & A_{m \mdot}^t
\end{bmatrix} \begin{bmatrix}
A_{1 \mdot} \\
A_{2 \mdot} \\
\vdots \\
A_{m \mdot}
\end{bmatrix} \\
&= A_{1 \mdot}^t A_{1 \mdot} + A_{2 \mdot}^t A_{2 \mdot} + \ldots + A_{m \mdot}^t A_{m \mdot}, \\
B &= \begin{bmatrix}
\alpha A_{1 \mdot} \\
A_{2 \mdot} \\
\vdots \\
A_{m \mdot}
\end{bmatrix}, \\
B^t B &= \begin{bmatrix}
\alpha A_{1 \mdot}^t & A_{2 \mdot}^t & \ldots & A_{m \mdot}^t
\end{bmatrix} \begin{bmatrix}
\alpha A_{1 \mdot} \\
A_{2 \mdot} \\
\vdots \\
A_{m \mdot}
\end{bmatrix} \\
&= \alpha^2 A_{1 \mdot}^t A_{1 \mdot} + A_{2 \mdot}^t A_{2 \mdot} + \ldots + A_{m \mdot}^t A_{m \mdot}.
\end{align*}
Logo, nota-se que $B^t B$ n\~{a}o \'{e} obtido atrav\'{e}s de nenhuma combina\c{c}\~{a}o de opera\c{c}\~{o}es elementrares sobre $A^t A$.
\end{solution}
\question Considere
\begin{align*}
A &= \begin{bmatrix}
R & w \\
0 & v
\end{bmatrix}, & b &= \begin{bmatrix}
c \\
d
\end{bmatrix},
\end{align*}
onde $A : m \times n$, $m \geq n$ com posto completo, $R : k \times k$, $w$ e $c$ s\~{a}o vetores $k \times 1$ e $v$ e $d$ s\~{a}o vetores $m - k \times 1$. Mostre que, se $k = n - 1$ ent\~{a}o
\begin{align*}
\min \| b - A x \|_2^2 &= \| d \|_2^2 - \left( v^t d / \| v \|_2 \right)^2.
\end{align*}
\begin{solution}
Pelo sistema de equa\c{c}\~{o}es normais, i.e., $A^t A x = A^t b$, temos que
\begin{align*}
\begin{bmatrix}
R^t R & R^t w \\
w^t R & w^t w + v^t v
\end{bmatrix} \begin{bmatrix}
x_1 \\
x_2
\end{bmatrix} &= \begin{bmatrix}
R^t c \\
w^t c + v^t d
\end{bmatrix}.
\end{align*}
Portanto,
\begin{align*}
R^t R x_1 + R^t w x_2 &= R^t c, \\
R^t \left( R x_1 + w x_2 - c \right) &= 0, \\
R x_1 + w x_2 &= c && \text{pois $R$ \'{e} uma matriz quadrada,} \\
w^t R x_1 + \left( w^t w + v^t v \right) x_2 &= w^t c + v^t d, \\
w^t \left( R x_1 + w x_2 - c \right) + v^t v x_2 &= v^t d, \\
v^t v x_2 &= v^t d, \\
x_2 &= \left( v^t d \right) / v^t v.
\end{align*}
Pela pri
Como $R$ \'{e} uma matriz quadrada,
\begin{align*}
R^t \left( R x_1 + w x_2 - c \right) &= 0 \rightarrow R x_1 + w x_2 = c.
\end{align*}
J\'{a} o res\'{i}duo do sistema linear corresponde a
\begin{align*}
\| b - A x \|_2^2 &= \left\| \begin{bmatrix}
c \\
d
\end{bmatrix} - \begin{bmatrix}
R & w \\
0 & v
\end{bmatrix} \begin{bmatrix}
x_1 \\
x_2
\end{bmatrix} \right\|_2^2 \\
&= \| d - v x_2 \|_2^2 \\
&= \| d \|_2^2 + x_2^2 \| v \|_2^2 - 2 (d^t v) x_2 \\
&= \| d \|_2^2 + \left( (v^t d)^2 / (\| v \|_2^2)^2 \right) \| v \|_2^2 - 2 (v^t d)^2 / \| v \|_2^2 \\
&= \| d \|_2^2 - \left( v^t d / \| v \|_2 \right)^2.
\end{align*}
\end{solution}
\end{questions}