-
Notifications
You must be signed in to change notification settings - Fork 118
/
pset2sol.lyx
1701 lines (1327 loc) · 31.7 KB
/
pset2sol.lyx
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
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#LyX 2.3 created this file. For more info see http://www.lyx.org/
\lyxformat 544
\begin_document
\begin_header
\save_transient_properties true
\origin unavailable
\textclass article
\begin_preamble
\renewcommand{\vec}[1]{\mathbf{#1}}
\renewcommand{\labelenumi}{(\alph{enumi})}
\renewcommand{\labelenumii}{(\roman{enumii})}
\newcommand{\fl}{\operatorname{fl}}
\newcommand{\sign}{\operatorname{sign}}
\newcommand{\argmax}{\operatorname{argmax}}
\end_preamble
\use_default_options false
\maintain_unincluded_children false
\language english
\language_package default
\inputencoding auto
\fontencoding global
\font_roman "default" "default"
\font_sans "default" "default"
\font_typewriter "default" "default"
\font_math "auto" "auto"
\font_default_family default
\use_non_tex_fonts false
\font_sc false
\font_osf false
\font_sf_scale 100 100
\font_tt_scale 100 100
\use_microtype false
\use_dash_ligatures true
\graphics default
\default_output_format default
\output_sync 0
\bibtex_command default
\index_command default
\paperfontsize default
\spacing single
\use_hyperref false
\papersize default
\use_geometry true
\use_package amsmath 2
\use_package amssymb 2
\use_package cancel 1
\use_package esint 0
\use_package mathdots 1
\use_package mathtools 1
\use_package mhchem 1
\use_package stackrel 1
\use_package stmaryrd 1
\use_package undertilde 1
\cite_engine basic
\cite_engine_type default
\biblio_style plain
\use_bibtopic false
\use_indices false
\paperorientation portrait
\suppress_date false
\justification true
\use_refstyle 0
\use_minted 0
\index Index
\shortcut idx
\color #008000
\end_index
\topmargin 1in
\secnumdepth 3
\tocdepth 3
\paragraph_separation indent
\paragraph_indentation default
\is_math_indent 0
\math_numbering_side default
\quotes_style english
\dynamic_quotes 0
\papercolumns 1
\papersides 2
\paperpagestyle default
\tracking_changes false
\output_changes false
\html_math_output 0
\html_css_as_file 0
\html_be_strict false
\end_header
\begin_body
\begin_layout Section*
18.335 Problem Set 2 Solutions
\end_layout
\begin_layout Subsection*
Problem 1: (14+(10+5) points)
\end_layout
\begin_layout Enumerate
Trefethen, exercise 15.1.
In the following, I abbreviate
\begin_inset Formula $\epsilon_{\mbox{machine}}=\epsilon_{m}$
\end_inset
, and I use the fact (which follows trivially from the definition of continuity)
that we can replace any Lipshitz-continuous
\begin_inset Formula $g(O(\epsilon))$
\end_inset
with
\begin_inset Formula $g(0)+g'(0)O(\epsilon)$
\end_inset
.
I also assume that
\begin_inset Formula $\fl(x)$
\end_inset
is deterministic—by a stretch of Trefethen's definitions, it could conceivably
be nondeterministic in which case one of the answers changes as noted below,
but this seems crazy to me (and doesn't correspond to any real machine).
Note also that, at the end of lecture 13, Trefethen points out that the
same axioms hold for complex floating-point arithmetic as for real floating-poi
nt arithmetic (possibly with
\begin_inset Formula $\epsilon_{m}$
\end_inset
increased by a constant factor), so we don't need to do anything special
here for
\begin_inset Formula $\mathbb{C}$
\end_inset
vs.
\begin_inset Formula $\mathbb{R}$
\end_inset
.
\begin_inset Separator latexpar
\end_inset
\end_layout
\begin_deeper
\begin_layout Enumerate
Backward stable.
\begin_inset Formula $x\oplus x=\fl(x)\oplus\fl(x)=[x(1+\epsilon_{1})+x(1+\epsilon_{1})](1+\epsilon_{2})=2\tilde{x}$
\end_inset
for
\begin_inset Formula $|\epsilon_{i}|\leq\epsilon_{m}$
\end_inset
and
\begin_inset Formula $\tilde{x}=x(1+\epsilon_{1}+\epsilon_{2}+2\epsilon_{1}\epsilon_{2})=x[1+O(\epsilon_{m})]$
\end_inset
.
\end_layout
\begin_layout Enumerate
Backward stable.
\begin_inset Formula $x\otimes x=\fl(x)\otimes\fl(x)=[x(1+\epsilon_{1})\times x(1+\epsilon_{1})](1+\epsilon_{2})=\tilde{x}^{2}$
\end_inset
for
\begin_inset Formula $|\epsilon_{i}|\leq\epsilon_{m}$
\end_inset
and
\begin_inset Formula $\tilde{x}=x(1+\epsilon_{1})\sqrt{1+\epsilon_{2}}=x[1+O(\epsilon_{m})]$
\end_inset
.
\end_layout
\begin_layout Enumerate
Stable but not backwards stable.
\begin_inset Formula $x\oslash x=[\fl(x)/\fl(x)](1+\epsilon)=1+\epsilon$
\end_inset
(not including
\begin_inset Formula $x=0$
\end_inset
or
\begin_inset Formula $\infty$
\end_inset
, which give NaN).
This is actually forwards stable, but there is no
\begin_inset Formula $\tilde{x}$
\end_inset
such that
\begin_inset Formula $\tilde{x}/\tilde{x}\neq1$
\end_inset
so it is not backwards stable.
(Under the stronger assumption of correctly rounded arithmetic, this will
give exactly 1, however.)
\end_layout
\begin_layout Enumerate
Backwards stable.
\begin_inset Formula $x\ominus x=[\fl(x)-\fl(x)](1+\epsilon)=0$
\end_inset
.
This is the correct answer for
\begin_inset Formula $\tilde{x}=x$
\end_inset
.
(In the crazy case where
\begin_inset Formula $\fl$
\end_inset
is not deterministic, then it might give a nonzero answer, in which case
it is unstable.)
\end_layout
\begin_layout Enumerate
Unstable.
It is definitely not backwards stable, because there is no data (and hence
no way to choose
\begin_inset Formula $\tilde{x}$
\end_inset
to match the output).
To be stable, it would have to be forwards stable, but it isn't because
the errors decrease more slowly than
\begin_inset Formula $O(\epsilon_{m})$
\end_inset
.
More explicitly,
\begin_inset Formula $1\oplus\frac{1}{2}\oplus\frac{1}{6}\oplus\cdots$
\end_inset
summed from left to right will give
\begin_inset Formula $((1+\frac{1}{2})(1+\epsilon_{1})+\frac{1}{6})(1+\epsilon_{2})\cdots=e+\frac{3}{2}\epsilon_{1}+\frac{10}{6}\epsilon_{2}+\cdots$
\end_inset
\size normal
\noun off
\color none
\family roman
\series medium
\shape up
\emph off
\bar no
dropping terms of
\family default
\series default
\shape default
\emph default
\bar default
\size default
\noun default
\begin_inset Formula $O(\epsilon^{2})$
\end_inset
\family roman
\series medium
\shape up
\size normal
\emph off
\bar no
\noun off
, where the coefficients of the
\family default
\series default
\shape default
\emph default
\bar default
\size default
\noun default
\begin_inset Formula $\epsilon_{k}$
\end_inset
\family roman
\series medium
\shape up
\size normal
\emph off
\bar no
\noun off
factors converge to
\family default
\series default
\shape default
\emph default
\bar default
\size default
\noun default
\begin_inset Formula $e$
\end_inset
\family roman
\series medium
\shape up
\size normal
\emph off
\bar no
\noun off
.
\family default
\series default
\shape default
\size default
\emph default
\bar default
\noun default
The number of terms is
\begin_inset Formula $n$
\end_inset
where
\begin_inset Formula $n$
\end_inset
satisfies
\begin_inset Formula $n!\approx1/\epsilon_{m}$
\end_inset
, which is a function that grows very slowly with
\begin_inset Formula $1/\epsilon_{m}$
\end_inset
, and hence the error from the additions alone is bounded above by
\begin_inset Formula $\approx n\epsilon_{m}$
\end_inset
.
The key point is that the errors grow at least as fast as
\begin_inset Formula $n\epsilon_{m}$
\end_inset
(not even counting errors from truncation of the series, approximation
of
\begin_inset Formula $1/k!$
\end_inset
, etcetera), which is
\emph on
not
\emph default
\begin_inset Formula $O(\epsilon_{m})$
\end_inset
because
\begin_inset Formula $n$
\end_inset
grows slowly with decreasing
\begin_inset Formula $\epsilon_{m}$
\end_inset
.
\end_layout
\begin_layout Enumerate
Stable.
As in (e), it is not backwards stable, so the only thing is to check forwards
stability.
Again, there will be
\begin_inset Formula $n$
\end_inset
terms in the series, where
\begin_inset Formula $n$
\end_inset
is a slowly growing function of
\begin_inset Formula $1/\epsilon_{m}$
\end_inset
(
\begin_inset Formula $n!\approx1/\epsilon_{m}$
\end_inset
).
However, the summation errors no longer grow as
\begin_inset Formula $n$
\end_inset
.
From right to left, we are summing
\begin_inset Formula $\frac{1}{n!}\oplus\frac{1}{(n-1)!}\oplus\cdots\oplus1$
\end_inset
.
But this gives
\begin_inset Formula $((\frac{1}{n!}+\frac{1}{(n-1)!})(1+\epsilon_{n-1})+\frac{1}{(n-2)!})(1+\epsilon_{n-2})\cdots,$
\end_inset
and the linear terms in the
\begin_inset Formula $\epsilon_{k}$
\end_inset
are then bounded by
\begin_inset Formula
\[
\left|\sum_{k=1}^{n-1}\epsilon_{k}\sum_{j=k}^{n}\frac{1}{j!}\right|\leq\epsilon_{m}\sum_{k=1}^{n-1}\sum_{j=k}^{n}\frac{1}{j!}=\epsilon_{m}\left[\frac{n-1}{n!}+\sum_{j=1}^{n-1}\frac{j}{j!}\right]\approx\epsilon_{m}e=O(\epsilon_{m}).
\]
\end_inset
The key point is that the coefficients of the
\begin_inset Formula $\epsilon_{k}$
\end_inset
coefficients grow smaller and smaller with
\begin_inset Formula $k$
\end_inset
, rather than approaching
\begin_inset Formula $e$
\end_inset
as for left-to-right summation, and the sum of the coefficients converges.
The truncation error is of
\begin_inset Formula $O(\epsilon_{m})$
\end_inset
, and we assume
\begin_inset Formula $1/k!$
\end_inset
can also be calculated to within
\begin_inset Formula $O(\epsilon_{m})$
\end_inset
, e.g.
via Stirling's approximation for large
\begin_inset Formula $k$
\end_inset
, so the overall error is
\begin_inset Formula $O(\epsilon_{m})$
\end_inset
and the algorithm is forwards stable.
\end_layout
\begin_layout Enumerate
Forwards stable.
Not backwards stable since no data, but what about forwards stability?
Supposing
\begin_inset Formula $\sin(x)$
\end_inset
is computed in a stable manner, then
\begin_inset Formula $\widetilde{\sin}(x)=\sin(x+\delta)\cdot[1+O(\epsilon_{m})]$
\end_inset
for
\begin_inset Formula $|\delta|=|x|O(\epsilon_{m})$
\end_inset
.
It follows that, in the vicinity of
\begin_inset Formula $x=\pi$
\end_inset
, the
\begin_inset Formula $\widetilde{\sin}$
\end_inset
function can only change sign within
\begin_inset Formula $|\delta|=\pi O(\epsilon_{m})$
\end_inset
of
\begin_inset Formula $x=\pi$
\end_inset
.
Hence, checking for
\begin_inset Formula $\widetilde{\sin}(x)\otimes\widetilde{\sin}(x')\le0$
\end_inset
, where
\begin_inset Formula $x'$
\end_inset
is the floating-point successor to
\begin_inset Formula $x$
\end_inset
(
\family typewriter
nextfloat(x)
\family default
in Julia) yields
\begin_inset Formula $\pi[1+O(\epsilon_{m})]$
\end_inset
, a forwards-stable result.
\end_layout
\end_deeper
\begin_layout Enumerate
Trefethen, exercise 16.1.
Note that we are free to switch norms as needed, by norm equivalence.
\emph on
Notation:
\emph default
the floating-point algorithm for computing
\begin_inset Formula $f(A)=QA$
\end_inset
will be denoted
\begin_inset Formula $\tilde{f}(A)=\widetilde{QA}$
\end_inset
; I will assume that we simply use the obvious three-loop algorithm, i.e.
computing the row–column dot products with in-order (
\begin_inset Quotes eld
\end_inset
recursive
\begin_inset Quotes erd
\end_inset
) summation, allowing us to re-use the summation error analysis from pset
1.
\begin_inset Separator latexpar
\end_inset
\end_layout
\begin_deeper
\begin_layout Enumerate
We will proceed by induction on
\begin_inset Formula $k$
\end_inset
: first, we will prove the base case, that multiplying
\begin_inset Formula $A$
\end_inset
by a
\emph on
single
\emph default
\begin_inset Formula $Q$
\end_inset
is backwards stable, and then we will do the inductive step (assume it
is true for
\begin_inset Formula $k$
\end_inset
, prove it for
\begin_inset Formula $k+1$
\end_inset
).
\begin_inset Newline newline
\end_inset
\begin_inset Newline newline
\end_inset
First, the base case: we need to find a
\begin_inset Formula $\delta A$
\end_inset
with
\begin_inset Formula $\Vert\delta A\Vert=\Vert A\Vert O(\epsilon_{\mbox{machine}})$
\end_inset
such that
\begin_inset Formula $\widetilde{QA}=Q(A+\delta A)$
\end_inset
.
Since
\begin_inset Formula $\Vert\delta A\Vert=\Vert Q^{*}\widetilde{QA}-A\Vert=\Vert Q(Q^{*}\widetilde{QA}-A)\Vert=\Vert\widetilde{QA}-QA\Vert$
\end_inset
in the
\begin_inset Formula $L_{2}$
\end_inset
norm, however, this is equivalent to showing
\begin_inset Formula $\Vert\widetilde{QA}-QA\Vert=\Vert A\Vert O(\epsilon_{\mbox{machine}})$
\end_inset
; that is, we can look at the
\emph on
forwards
\emph default
error, which is a bit easier.
It is sufficient to look at the error in the
\begin_inset Formula $ij$
\end_inset
-th element of
\begin_inset Formula $QA$
\end_inset
, i.e.
the error in computing
\begin_inset Formula $\sum_{k}q_{ik}a_{kj}$
\end_inset
.
Assuming we do this sum by a straightforward loop, the analysis is exactly
the same as in problem
\begin_inset space ~
\end_inset
2, except that there is an additional
\begin_inset Formula $(1+\epsilon)$
\end_inset
factor in each term for the error in the product
\begin_inset Formula $q_{ik}a_{kj}$
\end_inset
[or
\begin_inset Formula $(1+2\epsilon)$
\end_inset
if we include the rounding of
\begin_inset Formula $q_{ik}$
\end_inset
to
\begin_inset Formula $\tilde{q}_{ik}=\fl(q_{ik})$
\end_inset
].
Hence, the error in the
\begin_inset Formula $ij$
\end_inset
-th element is bounded by
\begin_inset Formula $mO(\epsilon_{\mbox{machine}})\sum_{k}|q_{ik}a_{kj}|$
\end_inset
, and (using the unitarity of
\begin_inset Formula $Q$
\end_inset
, which implies that
\begin_inset Formula $|q_{ik}|\leq1$
\end_inset
) this in turn is bounded by
\begin_inset Formula $mO(\epsilon_{\mbox{machine}})\sum_{k}|a_{kj}|\leq mO(\epsilon_{\mbox{machine}})\sum_{kj}|a_{kj}|\leq mO(\epsilon_{\mbox{machine}})\Vert A\Vert$
\end_inset
(since
\begin_inset Formula $\sum_{kj}|a_{kj}|$
\end_inset
is just an
\begin_inset Formula $L_{1}$
\end_inset
Frobenius norm of
\begin_inset Formula $A$
\end_inset
, which is within a constant factor of any other norm).
Summing
\begin_inset Formula $m^{2}$
\end_inset
of these errors in the individual elements of
\begin_inset Formula $QA$
\end_inset
, again using norm equivalence, we obtain
\begin_inset Formula $\Vert\widetilde{QA}-QA\Vert=O(\sum_{ij}|(\widetilde{QA}-QA)_{ij}|)=m^{3}O(\epsilon_{\mbox{machine}})\Vert A\Vert$
\end_inset
.
Thus, we have proved backwards stability for multiplying by one unitary
matrix (with a overly pessimistic
\begin_inset Formula $m^{3}$
\end_inset
coefficient, but that doesn't matter here).
\begin_inset Newline newline
\end_inset
\begin_inset Newline newline
\end_inset
Now, we will show by induction that multiplying by
\begin_inset Formula $k$
\end_inset
unitary matrices is backwards stable.
Suppose we have proved it for
\begin_inset Formula $k$
\end_inset
, and want to prove for
\begin_inset Formula $k+1$
\end_inset
.
That, consider
\begin_inset Formula $QQ_{k}\cdots Q_{1}A$
\end_inset
.
By assumption,
\begin_inset Formula $Q_{k}\cdots Q_{1}A$
\end_inset
is backwards stable, and hence
\begin_inset Formula $\tilde{B}=\widetilde{Q_{k}\cdots Q_{1}A}=Q_{k}\cdots Q_{1}(A+\delta A_{k})$
\end_inset
for some
\begin_inset Formula $\Vert\delta A_{k}\Vert=O(\epsilon_{\mbox{machine}})\Vert A\Vert$
\end_inset
.
Also, from above,
\begin_inset Formula $\widetilde{Q\tilde{B}}=Q(\tilde{B}+\delta\tilde{B})$
\end_inset
for some
\begin_inset Formula $\Vert\delta\tilde{B}\Vert=O(\epsilon_{\mbox{machine}})\Vert\tilde{B}\Vert=\Vert Q_{k}\cdots Q_{1}(A+\delta A_{k})\Vert O(\epsilon_{\mbox{machine}})=\Vert A+\delta A_{k}\Vert O(\epsilon_{\mbox{machine}})\leq\Vert A\Vert O(\epsilon_{\mbox{machine}})+\Vert\delta A_{k}\Vert O(\epsilon_{\mbox{machine}})=\Vert A\Vert O(\epsilon_{\mbox{machine}})$
\end_inset
.
Hence,
\begin_inset Formula $\widetilde{QQ_{k}\cdots Q_{1}A}=\widetilde{Q\tilde{B}}=Q[Q_{k}\cdots Q_{1}(A+\delta A_{k})+\delta\tilde{B}]=QQ_{k}\cdots Q_{1}(A+\delta A)$
\end_inset
where
\begin_inset Formula $\delta A=\delta A_{k}+[Q_{1}^{*}\cdots Q_{k}^{*}]\delta\tilde{B}$
\end_inset
and
\begin_inset Formula $\Vert\delta A\Vert\leq\Vert\delta A_{k}\Vert+\Vert\delta\tilde{B}\Vert=O(\epsilon_{\mbox{machine}})\Vert A\Vert$
\end_inset
.
Q.E.D.
\end_layout
\begin_layout Enumerate
Consider
\begin_inset Formula $XA$
\end_inset
, where
\begin_inset Formula $X$
\end_inset
is some rank-1 matrix
\begin_inset Formula $xy^{*}$
\end_inset
and
\begin_inset Formula $A$
\end_inset
has rank
\begin_inset Formula $>1$
\end_inset
.
The product
\begin_inset Formula $XA$
\end_inset
has rank
\begin_inset space ~
\end_inset
1 in exact arithmetic, but after floating-point errors it is unlikely that
\begin_inset Formula $\widetilde{XA}$
\end_inset
will be exactly rank 1.
Hence it is not backwards stable, because
\begin_inset Formula $X\tilde{A}$
\end_inset
will be rank
\begin_inset space ~
\end_inset
1 regardless of
\begin_inset Formula $\tilde{A}$
\end_inset
, and thus is
\begin_inset Formula $\neq\widetilde{XA}$
\end_inset
.
(See also example 15.2 in the text.)
\end_layout
\end_deeper
\begin_layout Subsection*
Problem 2: (10+10 points)
\end_layout
\begin_layout Enumerate
Denote the rows of
\begin_inset Formula $A$
\end_inset
by
\begin_inset Formula $a_{1}^{T},\ldots,a_{m}^{T}$
\end_inset
.
Consider the unit ball in the
\begin_inset Formula $L_{\infty}$
\end_inset
norm, the set
\begin_inset Formula $\{x\in\mathbb{C}^{n}:\Vert x\Vert_{\infty}\leq1\}$
\end_inset
.
Any vector
\begin_inset Formula $Ax$
\end_inset
in the image of this set satisfies:
\begin_inset Formula
\[
\Vert Ax\Vert_{\infty}=\max_{j\in1:m}|a_{j}^{T}x|=\max_{j\in1:m}\left|\sum_{k\in1:n}a_{j,k}x_{k}\right|\leq\max_{j\in1:m}\sum_{k\in1:n}|a_{j,k}|=\max_{j\in1:m}\Vert a_{j}\Vert,
\]
\end_inset
since
\begin_inset Formula $|x_{k}|\leq1$
\end_inset
in the
\begin_inset Formula $L_{\infty}$
\end_inset
unit ball.
Furthermore, this bound is achieved when
\begin_inset Formula $x_{k}=\sign(a_{j,k})$
\end_inset
where
\begin_inset Formula $j=\argmax_{j}\Vert a_{j}\Vert$
\end_inset
.
Hence
\begin_inset Formula $\Vert A\Vert_{\infty}=\max_{j}\Vert a_{j}\Vert$
\end_inset
, corresponding to (3.10).
Q.E.D.
\begin_inset Newline newline
\end_inset
\begin_inset Newline newline
\end_inset
If we look in the Julia source code, we find that this norm is computed
by summing the absolute values of each row of
\family typewriter
A
\family default
and then takes the maximum, exactly as in (3.10).
\end_layout
\begin_layout Enumerate
To obtain
\begin_inset Formula $\mu\times\nu$
\end_inset
submatrix
\begin_inset Formula $B$
\end_inset
of the
\begin_inset Formula $m\times n$
\end_inset
matrix
\begin_inset Formula $A$
\end_inset
by selecting a subset of the rows and columns of
\begin_inset Formula $A$
\end_inset
, we simply multiply
\begin_inset Formula $A$
\end_inset
on the left and right by
\begin_inset Formula $\mu\times m$
\end_inset
and
\begin_inset Formula $n\times\nu$
\end_inset
matrices as follows:
\begin_inset Formula
\[
B=\left(\begin{array}{ccccc}
1\\
& & & 1\\
& & & & \ddots
\end{array}\right)A\left(\begin{array}{ccc}
\\
1\\
& 1\\
& & \ddots\\
\\
\end{array}\right)
\]
\end_inset
where there are
\begin_inset Formula $1$
\end_inset
's in the columns/rows to be selected.
More precisely, if we want a subset
\begin_inset Formula $\mathcal{R}$
\end_inset
of the rows of
\begin_inset Formula $A$
\end_inset
and a subset
\begin_inset Formula $\mathcal{C}$
\end_inset
of the columns of
\begin_inset Formula $A$
\end_inset
, then we compute
\begin_inset Formula $B=D_{\mathcal{R}}AD_{\mathcal{C}}^{T}$
\end_inset
, where the
\begin_inset Quotes eld
\end_inset
deletion matrix
\begin_inset Quotes erd
\end_inset
for an ordered set
\begin_inset Formula $\mathcal{S}$
\end_inset
of indices is given by
\begin_inset Formula $(D_{\mathcal{S}})_{ij}=1$
\end_inset