sm-transcript 0.0.6 → 0.0.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -0,0 +1,2614 @@
1
+ 00:07.121
2
+ I've been multiplying matrices
3
+ already, but certainly time for
4
+
5
+ 00:12.725
6
+ me to discuss the rules for
7
+ matrix multiplication.
8
+
9
+ 00:17.227
10
+ And the interesting part is the
11
+ many ways you can do it,
12
+
13
+ 00:22.281
14
+ and they all give the same
15
+ answer.
16
+
17
+ 00:25.313
18
+ And they're all important.
19
+
20
+ 00:28.529
21
+ So matrix multiplication,
22
+ and then, come inverses.
23
+
24
+ 00:32.488
25
+ So we mentioned the inverse of
26
+ a matrix.
27
+
28
+ 00:35.639
29
+ That's a big deal.
30
+
31
+ 00:37.094
32
+ Lots to do about inverses and
33
+ how to find them.
34
+
35
+ 00:40.811
36
+ Okay, so I'll begin with how to
37
+ multiply two matrices.
38
+
39
+ 00:45.821
40
+ First way, okay,
41
+ so suppose I have a matrix A
42
+
43
+ 00:52.515
44
+ multiplying a matrix B and --
45
+ giving me a result -- well,
46
+
47
+ 01:01.035
48
+ I could call it C.
49
+
50
+ 01:03.774
51
+ A times B.
52
+
53
+ 01:05.295
54
+ Okay.
55
+
56
+ 01:06.056
57
+ So, let me just review the rule
58
+ for this entry.
59
+
60
+ 01:13.054
61
+ That's the entry in row i and
62
+ column j.
63
+
64
+ 01:20.51
65
+ So that's the i j entry.
66
+
67
+ 01:22.766
68
+ Right there is C i j.
69
+
70
+ 01:24.74
71
+ We always write the row number
72
+ and then the column number.
73
+
74
+ 01:30.098
75
+ So I might -- I might -- maybe
76
+ I take it C 3 4,
77
+
78
+ 01:34.422
79
+ just to make it specific.
80
+
81
+ 01:36.772
82
+ So instead of i j,
83
+ let me use numbers.
84
+
85
+ 01:40.25
86
+ C 3 4.
87
+
88
+ 01:41.754
89
+ So where does that come from,
90
+ the three four entry?
91
+
92
+ 01:47.12
93
+ It comes from row three,
94
+ here, row three and column
95
+
96
+ 01:52.485
97
+ four, as you know.
98
+
99
+ 01:54.416
100
+ Column four.
101
+
102
+ 01:55.704
103
+ And can I just write down,
104
+ or can we write down the
105
+
106
+ 02:01.069
107
+ formula for it?
108
+
109
+ 02:03.644
110
+ If we look at the whole row and
111
+ the whole column,
112
+
113
+ 02:08.247
114
+ the quick way for me to say it
115
+ is row three of A -- I could use
116
+
117
+ 02:14.192
118
+ a dot for dot product.
119
+
120
+ 02:16.301
121
+ I won't often use that,
122
+ actually.
123
+
124
+ 02:19.369
125
+ Dot column four of B.
126
+
127
+ 02:21.383
128
+ But this gives us a chance to
129
+ just, like, use a little matrix
130
+
131
+ 02:27.136
132
+ notation.
133
+
134
+ 02:29.054
135
+ What are the entries?
136
+
137
+ 02:31.441
138
+ What's this first entry in row
139
+ three?
140
+
141
+ 02:35.534
142
+ That number that's sitting
143
+ right there is...
144
+
145
+ 02:40.422
146
+ A, so it's got two indices and
147
+ what are they?
148
+
149
+ 02:46.221
150
+ 3 1.
151
+
152
+ 02:46.746
153
+ So there's an a 3 1 there.
154
+
155
+ 02:50.165
156
+ Now what's the first guy at the
157
+ top of column four?
158
+
159
+ 02:56.739
160
+ So what's sitting up there?
161
+
162
+ 03:00.289
163
+ B 1 4, right.
164
+
165
+ 03:01.998
166
+ So that this dot product starts
167
+ with A 3 1 times B 1 4.
168
+
169
+ 03:10.15
170
+ And then what's the next -- so
171
+ this is like I'm accumulating
172
+
173
+ 03:16.156
174
+ this sum, then comes the next
175
+ guy, A 3 2, second column,
176
+
177
+ 03:21.755
178
+ times B 2 4,
179
+ second row.
180
+
181
+ 03:24.096
182
+ So it's b A 3 2,
183
+ B 2 4 and so on.
184
+
185
+ 03:27.354
186
+ Just practice with indices.
187
+
188
+ 03:30.102
189
+ Oh, let me even practice with a
190
+ summation formula.
191
+
192
+ 03:36.211
193
+ So this is -- most of the
194
+ course, I use whole vectors.
195
+
196
+ 03:42.102
197
+ I very seldom,
198
+ get down to the details of
199
+
200
+ 03:46.549
201
+ these particular entries,
202
+ but here we'd better do it.
203
+
204
+ 03:52.329
205
+ So it's some kind of a sum,
206
+ right?
207
+
208
+ 03:56.887
209
+ Of things in row three,
210
+ column K shall I say?
211
+
212
+ 04:01.555
213
+ Times things in row K,
214
+ column four.
215
+
216
+ 04:05.163
217
+ Do you see that that's what
218
+ we're seeing here?
219
+
220
+ 04:09.937
221
+ This is K is one,
222
+ here K is two,
223
+
224
+ 04:13.226
225
+ on along -- so the sum goes all
226
+ the way along the row and down
227
+
228
+ 04:19.699
229
+ the column, say,
230
+ one to N.
231
+
232
+ 04:23.625
233
+ So that's what the C three four
234
+ entry looks like.
235
+
236
+ 04:28.399
237
+ A sum of a three K b K four.
238
+
239
+ 04:31.185
240
+ Just takes a little practice to
241
+ do that.
242
+
243
+ 04:35.064
244
+ Okay.
245
+
246
+ 04:35.561
247
+ And -- well,
248
+ maybe I should say -- when are
249
+
250
+ 04:39.739
251
+ we allowed to multiply these
252
+ matrices?
253
+
254
+ 04:44.415
255
+ What are the shapes of these
256
+ things?
257
+
258
+ 04:47.647
259
+ The shapes are -- if we allow
260
+ them to be not necessarily
261
+
262
+ 04:52.726
263
+ square matrices.
264
+
265
+ 04:54.204
266
+ If they're square,
267
+ they've got to be the same
268
+
269
+ 04:58.267
270
+ size.
271
+
272
+ 04:58.729
273
+ If they're rectangular,
274
+ they're not the same size.
275
+
276
+ 05:04.177
277
+ If they're rectangular,
278
+ this might be -- well,
279
+
280
+ 05:07.18
281
+ I always think of A as m by n.
282
+ m rows, n columns.
283
+
284
+ 05:10.383
285
+ So that sum goes to n.
286
+
287
+ 05:11.85
288
+ Now what's the point -- how
289
+ many rows does B have to have?
290
+
291
+ 05:15.654
292
+ n.
293
+
294
+ 05:15.787
295
+ The number of rows in B,
296
+ the number of guys that we meet
297
+
298
+ 05:19.457
299
+ coming down has to match the
300
+ number of ones across.
301
+
302
+ 05:23.593
303
+ So B will have to be n by
304
+ something.
305
+
306
+ 05:26.903
307
+ Whatever.
308
+
309
+ 05:27.754
310
+ P.
311
+
312
+ 05:27.943
313
+ So the number of columns here
314
+ has to match the number of rows
315
+
316
+ 05:33.617
317
+ there, and then what's the
318
+ result?
319
+
320
+ 05:36.737
321
+ What's the shape of the result?
322
+
323
+ 05:39.668
324
+ What's the shape of C,
325
+ the output?
326
+
327
+ 05:42.789
328
+ Well, it's got these same m
329
+ rows -- it's got m rows.
330
+
331
+ 05:48.841
332
+ And how many columns?
333
+
334
+ 05:51.027
335
+ P.
336
+ m by P.
337
+
338
+ 05:51.964
339
+ Okay.
340
+
341
+ 05:52.485
342
+ So there are m times P little
343
+ numbers in there,
344
+
345
+ 05:57.275
346
+ entries, and each one,
347
+ looks like that.
348
+
349
+ 06:01.231
350
+ Okay.
351
+
352
+ 06:01.752
353
+ So that's the standard rule.
354
+
355
+ 06:04.668
356
+ That's the way people think of
357
+ multiplying matrices.
358
+
359
+ 06:11.228
360
+ I do it too.
361
+
362
+ 06:12.922
363
+ But I want to talk about other
364
+ ways to look at that same
365
+
366
+ 06:20.689
367
+ calculation, looking at whole
368
+ columns and whole rows.
369
+
370
+ 06:28.031
371
+ Okay.
372
+
373
+ 06:28.737
374
+ So can I do A B C again?
375
+
376
+ 06:32.126
377
+ A B equaling C again?
378
+
379
+ 06:36.221
380
+ But now, tell me about...
381
+
382
+ 06:40.224
383
+ I'll put it up here.
384
+
385
+ 06:43.426
386
+ So here goes A,
387
+ again, times B producing C.
388
+
389
+ 06:50.149
390
+ And again, this is m by n.
391
+
392
+ 06:54.312
393
+ This is n by P and this is m by
394
+ P.
395
+
396
+ 06:59.595
397
+ Okay.
398
+
399
+ 07:00.395
400
+ Now I want to look at whole
401
+ columns.
402
+
403
+ 07:07.6
404
+ I want to look at the columns
405
+ of -- here's the second way to
406
+
407
+ 07:12.836
408
+ multiply matrices.
409
+
410
+ 07:14.434
411
+ Because I'm going to build on
412
+ what I know already.
413
+
414
+ 07:18.783
415
+ How do I multiply a matrix by a
416
+ column?
417
+
418
+ 07:22.156
419
+ I know how to multiply this
420
+ matrix by that column.
421
+
422
+ 07:27.303
423
+ Shall I call that column one?
424
+
425
+ 07:29.858
426
+ That tells me column one of the
427
+ answer.
428
+
429
+ 07:33.206
430
+ The matrix times the first
431
+ column is that first column.
432
+
433
+ 07:37.963
434
+ Because none of this stuff
435
+ entered that part of the answer.
436
+
437
+ 07:43.073
438
+ The matrix times the second
439
+ column is the second column of
440
+
441
+ 07:48.094
442
+ the answer.
443
+
444
+ 07:49.944
445
+ Do you see what I'm saying?
446
+
447
+ 07:52.773
448
+ That I could think of
449
+ multiplying a matrix by a
450
+
451
+ 07:57.591
452
+ vector, which I already knew how
453
+ to do, and I can think of just P
454
+
455
+ 08:04.294
456
+ columns sitting side by side,
457
+ just like resting next to each
458
+
459
+ 08:10.474
460
+ other.
461
+
462
+ 08:11.94
463
+ And I multiply A times each one
464
+ of those.
465
+
466
+ 08:14.775
467
+ And I get the P columns of the
468
+ answer.
469
+
470
+ 08:17.397
471
+ Do you see this as -- this is
472
+ quite nice, to be able to think,
473
+
474
+ 08:21.72
475
+ okay, matrix multiplication
476
+ works so that I can just think
477
+
478
+ 08:25.76
479
+ of having several columns,
480
+ multiplying by A and getting
481
+
482
+ 08:29.586
483
+ the columns of the answer.
484
+
485
+ 08:32.208
486
+ So, like, here's column one
487
+ shall I call that column one?
488
+
489
+ 08:41.236
490
+ And what's going in there is A
491
+ times column one.
492
+
493
+ 08:48.813
494
+ Okay.
495
+
496
+ 08:49.619
497
+ So that's the picture a column
498
+ at a time.
499
+
500
+ 08:57.197
501
+ So what does that tell me?
502
+
503
+ 09:00.769
504
+ What does that tell me about
505
+ these columns?
506
+
507
+ 09:06.541
508
+ These columns of C are
509
+ combinations,
510
+
511
+ 09:11.35
512
+ because we've seen that before,
513
+ of columns of A.
514
+
515
+ 09:17.809
516
+ Every one of these comes from A
517
+ times this, and A times a vector
518
+
519
+ 09:26.466
520
+ is a combination of the columns
521
+ of A.
522
+
523
+ 09:32.924
524
+ And it makes sense,
525
+ because the columns of A have
526
+
527
+ 09:37.446
528
+ length m and the columns of C
529
+ have length m.
530
+
531
+ 09:41.496
532
+ And every column of C is some
533
+ combination of the columns of A.
534
+
535
+ 09:47.241
536
+ And it's these numbers in here
537
+ that tell me what combination it
538
+
539
+ 09:53.081
540
+ is.
541
+
542
+ 09:54.211
543
+ Do you see that?
544
+
545
+ 09:56.013
546
+ That in that answer,
547
+ C, I'm seeing stuff that's
548
+
549
+ 10:01.194
550
+ combinations of these columns.
551
+
552
+ 10:04.572
553
+ Now, suppose I look at it --
554
+ that's two ways now.
555
+
556
+ 10:09.978
557
+ The third way is look at it by
558
+ rows.
559
+
560
+ 10:13.919
561
+ So now let me change to rows.
562
+
563
+ 10:17.185
564
+ Okay.
565
+
566
+ 10:18.875
567
+ So now I can think of a row of
568
+ A -- a row of A multiplying all
569
+
570
+ 10:28.038
571
+ these rows here and producing a
572
+ row of the product.
573
+
574
+ 10:35.55
575
+ So this row takes a combination
576
+ of these rows and that's the
577
+
578
+ 10:44.413
579
+ answer.
580
+
581
+ 10:45.464
582
+ So these rows of C are
583
+ combinations of what?
584
+
585
+ 10:51.924
586
+ Tell me how to finish that.
587
+
588
+ 10:57.482
589
+ The rows of C,
590
+ when I have a matrix B,
591
+
592
+ 11:00.883
593
+ it's got its rows and I
594
+ multiply by A,
595
+
596
+ 11:04.284
597
+ and what does that do?
598
+
599
+ 11:06.305
600
+ It mixes the rows up.
601
+
602
+ 11:08.235
603
+ It creates combinations of the
604
+ rows of B, thanks.
605
+
606
+ 11:12.647
607
+ Rows of B.
608
+
609
+ 11:13.566
610
+ That's what I wanted to see,
611
+ that this answer --
612
+
613
+ 11:18.897
614
+ I can see where the pieces are
615
+ coming from.
616
+
617
+ 11:23.292
618
+ The rows in the answer are
619
+ coming as combinations of these
620
+
621
+ 11:29.258
622
+ rows.
623
+
624
+ 11:29.782
625
+ The columns in the answer are
626
+ coming as combinations of those
627
+
628
+ 11:36.062
629
+ columns.
630
+
631
+ 11:36.899
632
+ And so that's three ways.
633
+
634
+ 11:40.457
635
+ Now you can say,
636
+ okay, what's the fourth way?
637
+
638
+ 11:46.462
639
+ The fourth way -- so that's --
640
+ now we've got,
641
+
642
+ 11:52.466
643
+ like, the regular way,
644
+ the column way,
645
+
646
+ 11:57.516
647
+ the row way and -- what's left?
648
+
649
+ 12:01.746
650
+ The one that I can -- well,
651
+ one way is columns times rows.
652
+
653
+ 12:09.524
654
+ What happens if I multiply --
655
+ So this was row times column,
656
+
657
+ 12:18.155
658
+ it gave a number.
659
+
660
+ 12:20.094
661
+ Okay.
662
+
663
+ 12:20.665
664
+ Now I want to ask you about
665
+ column times row.
666
+
667
+ 12:25.684
668
+ If I multiply a column of A
669
+ times a row of B,
670
+
671
+ 12:30.704
672
+ what shape I ending up with?
673
+
674
+ 12:33.898
675
+ So if I take a column times a
676
+ row, that's definitely different
677
+
678
+ 12:40.857
679
+ from taking a row times a
680
+ column.
681
+
682
+ 12:45.876
683
+ So a column of A was -- what's
684
+ the shape of a column of A?
685
+
686
+ 12:53.085
687
+ n by one.
688
+
689
+ 12:54.223
690
+ A column of A is a column.
691
+
692
+ 12:57.512
693
+ It's got m entries and one
694
+ column.
695
+
696
+ 13:01.685
697
+ And what's a row of B?
698
+
699
+ 13:04.467
700
+ It's got one row and P columns.
701
+
702
+ 13:08.388
703
+ So what's the shape --
704
+ what do I get if I multiply a
705
+
706
+ 13:16.017
707
+ column by a row?
708
+
709
+ 13:18.062
710
+ I get a big matrix.
711
+
712
+ 13:20.492
713
+ I get a full-sized matrix.
714
+
715
+ 13:23.816
716
+ If I multiply a column by a row
717
+ -- should we just do one?
718
+
719
+ 13:30.977
720
+ Let me take the column two
721
+ three four times the row one
722
+
723
+ 13:37.881
724
+ six.
725
+
726
+ 13:38.393
727
+ That product there --
728
+ I mean, when I'm just following
729
+
730
+ 13:44.359
731
+ the rules of matrix
732
+ multiplication,
733
+
734
+ 13:46.555
735
+ those rules are just looking
736
+ like -- kind of petite,
737
+
738
+ 13:49.849
739
+ kind of small,
740
+ because the rows here are so
741
+
742
+ 13:52.562
743
+ short and the columns there are
744
+ so short, but they're the same
745
+
746
+ 13:56.503
747
+ length, one entry.
748
+
749
+ 13:57.665
750
+ So what's the answer?
751
+
752
+ 13:59.732
753
+ What's the answer if I do two
754
+ three four times one six,
755
+
756
+ 14:05.35
757
+ just for practice?
758
+
759
+ 14:07.222
760
+ Well, what's the first row of
761
+ the answer?
762
+
763
+ 14:11.383
764
+ Two twelve.
765
+
766
+ 14:12.527
767
+ And the second row of the
768
+ answer is three eighteen.
769
+
770
+ 14:17.728
771
+ And the third row of the answer
772
+ is four twenty four.
773
+
774
+ 14:24.073
775
+ That's a very special matrix,
776
+ there.
777
+
778
+ 14:27.372
779
+ Very special matrix.
780
+
781
+ 14:29.257
782
+ What can you tell me about its
783
+ columns, the columns of that
784
+
785
+ 14:34.723
786
+ matrix?
787
+
788
+ 14:35.383
789
+ They're multiples of this guy,
790
+ right?
791
+
792
+ 14:38.776
793
+ They're multiples of that one.
794
+
795
+ 14:41.603
796
+ Which follows our rule.
797
+
798
+ 14:44.714
799
+ We said that the columns of the
800
+ answer were combinations,
801
+
802
+ 14:49.15
803
+ but there's only -- to take a
804
+ combination of one guy,
805
+
806
+ 14:53.269
807
+ it's just a multiple.
808
+
809
+ 14:54.933
810
+ The rows of the answer,
811
+ what can you tell me about
812
+
813
+ 14:58.814
814
+ those three rows?
815
+
816
+ 15:00.161
817
+ They're all multiples of this
818
+ row.
819
+
820
+ 15:03.567
821
+ They're all multiples of one
822
+ six, as we expected.
823
+
824
+ 15:11.05
825
+ But I'm getting a full-sized
826
+ matrix.
827
+
828
+ 15:16.506
829
+ And now, just to complete this
830
+ thought, if I have -- let me
831
+
832
+ 15:25.547
833
+ write down the fourth way.
834
+
835
+ 15:29.6
836
+ A B is a sum of columns of A
837
+ times rows of B.
838
+
839
+ 15:37.861
840
+ So that, for example,
841
+ if my matrix was two three four
842
+
843
+ 15:43.219
844
+ and then had another column,
845
+ say, seven eight nine,
846
+
847
+ 15:48.371
848
+ and my matrix here has -- say,
849
+ started with one six and then
850
+
851
+ 15:54.449
852
+ had another column like zero
853
+ zero, then -- here's the fourth
854
+
855
+ 16:00.528
856
+ way, okay?
857
+
858
+ 16:01.559
859
+ I've got two columns there,
860
+ I've got two rows there.
861
+
862
+ 16:07.947
863
+ So the beautiful rule is --
864
+ see, the whole thing by columns
865
+
866
+ 16:14.595
867
+ and rows is that I can take the
868
+ first column times the first row
869
+
870
+ 16:21.817
871
+ and add the second column times
872
+ the second row.
873
+
874
+ 16:27.089
875
+ So that's the fourth way --
876
+ that I can take columns times
877
+
878
+ 16:33.509
879
+ rows, first column times first
880
+ row, second column times second
881
+
882
+ 16:40.501
883
+ row and add.
884
+
885
+ 16:43.138
886
+ Actually, what will I get?
887
+
888
+ 16:45.561
889
+ What will the answer be for
890
+ that matrix multiplication?
891
+
892
+ 16:50.594
893
+ Well, this one it's just going
894
+ to give us zero,
895
+
896
+ 16:54.881
897
+ so in fact I'm back to this --
898
+ that's the answer,
899
+
900
+ 16:59.355
901
+ for that matrix multiplication.
902
+
903
+ 17:02.99
904
+ I'm happy to put up here these
905
+ facts about matrix
906
+
907
+ 17:06.69
908
+ multiplication,
909
+ because it gives me a chance to
910
+
911
+ 17:10.237
912
+ write down special matrices like
913
+ this.
914
+
915
+ 17:13.089
916
+ This is a special matrix.
917
+
918
+ 17:15.017
919
+ All those rows lie on the same
920
+ line.
921
+
922
+ 17:17.715
923
+ All those rows lie on the line
924
+ through one six.
925
+
926
+ 17:22.109
927
+ If I draw a picture of all
928
+ these row vectors,
929
+
930
+ 17:24.973
931
+ they're all the same direction.
932
+
933
+ 17:26.991
934
+ If I draw a picture of these
935
+ two column vectors,
936
+
937
+ 17:30.05
938
+ they're in the same direction.
939
+
940
+ 17:32.002
941
+ Later, I would use this
942
+ language.
943
+
944
+ 17:34.085
945
+ Not too much later,
946
+ either.
947
+
948
+ 17:35.777
949
+ I would say the row space,
950
+ which is like all the
951
+
952
+ 17:38.836
953
+ combinations of the rows,
954
+ is just a line for this matrix.
955
+
956
+ 17:43.392
957
+ The row space is the line
958
+ through the vector one six.
959
+
960
+ 17:48.836
961
+ All the rows lie on that line.
962
+
963
+ 17:51.978
964
+ And the column space is also a
965
+ line.
966
+
967
+ 17:55.642
968
+ All the columns lie on the line
969
+ through the vector two three
970
+
971
+ 18:01.82
972
+ four.
973
+
974
+ 18:02.344
975
+ So this is like a really
976
+ minimal matrix.
977
+
978
+ 18:07.474
979
+ And it's because of these ones.
980
+
981
+ 18:11.157
982
+ Okay.
983
+
984
+ 18:11.751
985
+ So that's a third way.
986
+
987
+ 18:14.364
988
+ Now I want to say one more
989
+ thing about matrix
990
+
991
+ 18:19.591
992
+ multiplication while we're on
993
+ the subject.
994
+
995
+ 18:24.461
996
+ And it's this.
997
+
998
+ 18:26.124
999
+ You could also multiply --
1000
+ You could also cut the matrix
1001
+
1002
+ 18:33.121
1003
+ into blocks and do the
1004
+ multiplication by blocks.
1005
+
1006
+ 18:37.723
1007
+ Yet that's actually so,
1008
+ useful that I want to mention
1009
+
1010
+ 18:42.815
1011
+ it.
1012
+
1013
+ 18:43.108
1014
+ Block multiplication.
1015
+
1016
+ 18:45.165
1017
+ So I could take my matrix A and
1018
+ I could chop it up,
1019
+
1020
+ 18:50.06
1021
+ like, maybe just for
1022
+ simplicity, let me chop it into
1023
+
1024
+ 18:55.054
1025
+ two --
1026
+ into four square blocks.
1027
+
1028
+ 18:59.191
1029
+ Suppose it's square.
1030
+
1031
+ 19:01.17
1032
+ Let's just take a nice case.
1033
+
1034
+ 19:03.94
1035
+ And B, suppose it's square
1036
+ also, same size.
1037
+
1038
+ 19:08.095
1039
+ So these sizes don't have to be
1040
+ the same.
1041
+
1042
+ 19:12.053
1043
+ What they have to do is match
1044
+ properly.
1045
+
1046
+ 19:16.702
1047
+ Here they certainly will match.
1048
+
1049
+ 19:20.23
1050
+ So here's the rule for block
1051
+ multiplication,
1052
+
1053
+ 19:25.123
1054
+ that if this has blocks like,
1055
+ A -- so maybe A1,
1056
+
1057
+ 19:30.357
1058
+ A2, A3, A4 are the blocks here,
1059
+ and these blocks are B1,
1060
+
1061
+ 19:36.615
1062
+ B2,3 and B4?
1063
+
1064
+ 19:37.98
1065
+ Then the answer I can find
1066
+ block.
1067
+
1068
+ 19:42.759
1069
+ And if you tell me what's in
1070
+ that block, then I'm going to be
1071
+
1072
+ 19:46.042
1073
+ quiet about matrix
1074
+ multiplication for the rest of
1075
+
1076
+ 19:48.667
1077
+ the day.
1078
+
1079
+ 19:49.105
1080
+ What goes into that block?
1081
+
1082
+ 19:50.527
1083
+ You see, these might be -- this
1084
+ matrix might be -- these
1085
+
1086
+ 19:53.536
1087
+ matrices might be,
1088
+ like, twenty by twenty with
1089
+
1090
+ 19:55.998
1091
+ blocks that are ten by ten,
1092
+ to take the easy case where all
1093
+
1094
+ 19:59.17
1095
+ the blocks are the same shape.
1096
+
1097
+ 20:01.522
1098
+ And the point is that I could
1099
+ multiply those by blocks.
1100
+
1101
+ 20:08.591
1102
+ And what goes in here?
1103
+
1104
+ 20:11.47
1105
+ What's that block in the
1106
+ answer?
1107
+
1108
+ 20:15.528
1109
+ A1 B1, that's a matrix times a
1110
+ matrix, it's the right size,
1111
+
1112
+ 20:23.12
1113
+ ten by ten.
1114
+
1115
+ 20:25.606
1116
+ Any more?
1117
+
1118
+ 20:26.735
1119
+ Plus, what else goes in there?
1120
+
1121
+ 20:30.498
1122
+ A2 B3, right?
1123
+
1124
+ 20:32.128
1125
+ It's just like block rows times
1126
+ block columns.
1127
+
1128
+ 20:37.772
1129
+ Nobody, I think,
1130
+ not even Gauss could see
1131
+
1132
+ 20:42.789
1133
+ instantly that it works.
1134
+
1135
+ 20:46.803
1136
+ But somehow,
1137
+ if we check it through,
1138
+
1139
+ 20:50.29
1140
+ all five ways we're doing the
1141
+ same multiplications.
1142
+
1143
+ 20:55.271
1144
+ So this familiar multiplication
1145
+ is what we're really doing when
1146
+
1147
+ 21:01.448
1148
+ we do it by columns,
1149
+ by rows by columns times rows
1150
+
1151
+ 21:06.33
1152
+ and by blocks.
1153
+
1154
+ 21:08.622
1155
+ Okay.
1156
+
1157
+ 21:09.357
1158
+ I just have to,
1159
+ like, get the rules straight
1160
+
1161
+ 21:15.681
1162
+ for matrix multiplication.
1163
+
1164
+ 21:19.504
1165
+ Okay.
1166
+
1167
+ 21:20.24
1168
+ All right, I'm ready for the
1169
+ second topic,
1170
+
1171
+ 21:26.269
1172
+ which is inverses.
1173
+
1174
+ 21:28.916
1175
+ Okay.
1176
+
1177
+ 21:29.652
1178
+ Ready for inverses.
1179
+
1180
+ 21:32.446
1181
+ And let me do it for square
1182
+ matrices first.
1183
+
1184
+ 21:40.387
1185
+ Okay.
1186
+
1187
+ 21:40.848
1188
+ So I've got a square matrix A.
1189
+
1190
+ 21:43.612
1191
+ And it may or may not have an
1192
+ inverse, right?
1193
+
1194
+ 21:47.665
1195
+ Not all matrices have inverses.
1196
+
1197
+ 21:50.52
1198
+ In fact, that's the most
1199
+ important question you can ask
1200
+
1201
+ 21:55.494
1202
+ about the matrix,
1203
+ is if it's -- if you know it's
1204
+
1205
+ 21:59.824
1206
+ square, is it invertible or not?
1207
+
1208
+ 22:03.692
1209
+ If it is invertible,
1210
+ then there is some other
1211
+
1212
+ 22:10.081
1213
+ matrix, shall I call it A
1214
+ inverse?
1215
+
1216
+ 22:14.873
1217
+ And what's the -- if A inverse
1218
+ exists -- there's a big "if"
1219
+
1220
+ 22:23.295
1221
+ here.
1222
+
1223
+ 22:25.038
1224
+ If this matrix exists,
1225
+ and it'll be really central to
1226
+
1227
+ 22:31.143
1228
+ figure out when does it exist?
1229
+
1230
+ 22:34.665
1231
+ And then if it does exist,
1232
+ how would you find it?
1233
+
1234
+ 22:40.3
1235
+ But what's the equation here
1236
+ that I haven't -- that I have to
1237
+
1238
+ 22:47.345
1239
+ finish now?
1240
+
1241
+ 22:49.576
1242
+ This matrix,
1243
+ if it exists multiplies A and
1244
+
1245
+ 23:00.786
1246
+ produces, I think,
1247
+ the identity.
1248
+
1249
+ 23:10.356
1250
+ But a real -- an inverse for a
1251
+ square matrix could be on the
1252
+
1253
+ 23:17.481
1254
+ right as well -- this is true,
1255
+ too, that it's -- if I have a
1256
+
1257
+ 23:24.605
1258
+ -- yeah in fact,
1259
+ this is not -- this is probably
1260
+
1261
+ 23:30.28
1262
+ the --
1263
+ this is something that's not
1264
+
1265
+ 23:34.311
1266
+ easy to prove,
1267
+ but it works.
1268
+
1269
+ 23:36.684
1270
+ That a left -- square matrices,
1271
+ a left inverse is also a right
1272
+
1273
+ 23:42.046
1274
+ inverse.
1275
+
1276
+ 23:42.749
1277
+ If I can find a matrix on the
1278
+ left that gets the identity,
1279
+
1280
+ 23:47.759
1281
+ then also that matrix on the
1282
+ right will produce that
1283
+
1284
+ 23:52.242
1285
+ identity.
1286
+
1287
+ 23:54
1288
+ For rectangular matrices,
1289
+ we'll see a left inverse that
1290
+
1291
+ 23:58.691
1292
+ isn't a right inverse.
1293
+
1294
+ 24:00.602
1295
+ In fact, the shapes wouldn't
1296
+ allow it.
1297
+
1298
+ 24:03.817
1299
+ But for square matrices,
1300
+ the shapes allow it and it
1301
+
1302
+ 24:08.161
1303
+ happens, if A has an inverse.
1304
+
1305
+ 24:10.68
1306
+ Okay, so give me some cases --
1307
+ let's see.
1308
+
1309
+ 24:15.025
1310
+ I hate to be negative here,
1311
+ but let's talk about the case
1312
+
1313
+ 24:21.574
1314
+ with no inverse.
1315
+
1316
+ 24:23.445
1317
+ So -- these matrices are called
1318
+ invertible or non-singular --
1319
+
1320
+ 24:30.463
1321
+ those are the good ones.
1322
+
1323
+ 24:33.27
1324
+ And we want to be able to
1325
+ identify how -- if we're given a
1326
+
1327
+ 24:39.936
1328
+ matrix, has it got an inverse?
1329
+
1330
+ 24:44.498
1331
+ Can I talk about the singular
1332
+ case?
1333
+
1334
+ 24:49.132
1335
+ No inverse.
1336
+
1337
+ 24:50.631
1338
+ All right.
1339
+
1340
+ 24:51.994
1341
+ Best to start with an example.
1342
+
1343
+ 24:56.083
1344
+ Tell me an example -- let's get
1345
+ an example up here.
1346
+
1347
+ 25:02.898
1348
+ Let's make it two by two -- of
1349
+ a matrix that has not got an
1350
+
1351
+ 25:10.803
1352
+ inverse.
1353
+
1354
+ 25:11.894
1355
+ And let's see why.
1356
+
1357
+ 25:14.347
1358
+ Let me write one up.
1359
+
1360
+ 25:18.709
1361
+ No inverse.
1362
+
1363
+ 25:20.69
1364
+ Let's see why.
1365
+
1366
+ 25:23.211
1367
+ Let me write up -- one three
1368
+ two six.
1369
+
1370
+ 25:29.695
1371
+ Why does that matrix have no
1372
+ inverse?
1373
+
1374
+ 25:36.178
1375
+ You could answer that various
1376
+ ways.
1377
+
1378
+ 25:42.302
1379
+ Give me one reason.
1380
+
1381
+ 25:45.724
1382
+ Well, you could --
1383
+ if you know about determinants,
1384
+
1385
+ 25:54.11
1386
+ which you're not supposed to,
1387
+ you could take its determinant
1388
+
1389
+ 26:00.474
1390
+ and you would get -- Zero.
1391
+
1392
+ 26:03.278
1393
+ Okay.
1394
+
1395
+ 26:03.818
1396
+ Now -- all right.
1397
+
1398
+ 26:05.651
1399
+ Let me ask you other reasons.
1400
+
1401
+ 26:08.779
1402
+ I mean, as for other reasons
1403
+ that that matrix isn't
1404
+
1405
+ 26:14.172
1406
+ invertible.
1407
+
1408
+ 26:16.438
1409
+ Here, I could use what I'm
1410
+ saying here.
1411
+
1412
+ 26:21.318
1413
+ Suppose A times other matrix
1414
+ gave the identity.
1415
+
1416
+ 26:27.226
1417
+ Why is that not possible?
1418
+
1419
+ 26:30.437
1420
+ Because -- oh,
1421
+ yeah -- I'm thinking about
1422
+
1423
+ 26:35.574
1424
+ columns here.
1425
+
1426
+ 26:37.244
1427
+ If I multiply this matrix A by
1428
+ some other matrix,
1429
+
1430
+ 26:43.409
1431
+ then the --
1432
+ the result -- what can you tell
1433
+
1434
+ 26:48.883
1435
+ me about the columns?
1436
+
1437
+ 26:50.677
1438
+ They're all multiples of those
1439
+ columns, right?
1440
+
1441
+ 26:54.521
1442
+ If I multiply A by another
1443
+ matrix that -- the product has
1444
+
1445
+ 26:59.304
1446
+ columns that come from those
1447
+ columns.
1448
+
1449
+ 27:02.38
1450
+ So can I get the identity
1451
+ matrix?
1452
+
1453
+ 27:05.968
1454
+ No way.
1455
+
1456
+ 27:06.499
1457
+ The columns of the identity
1458
+ matrix, like one zero -- it's
1459
+
1460
+ 27:10.752
1461
+ not a combination of those
1462
+ columns, because those two
1463
+
1464
+ 27:14.702
1465
+ columns lie on the -- both lie
1466
+ on the same line.
1467
+
1468
+ 27:18.272
1469
+ Every combination is just going
1470
+ to be on that line and I can't
1471
+
1472
+ 27:22.905
1473
+ get one zero.
1474
+
1475
+ 27:24.652
1476
+ So, do you see that sort of
1477
+ column picture of the matrix not
1478
+
1479
+ 27:30.497
1480
+ being invertible.
1481
+
1482
+ 27:32.181
1483
+ In fact, here's another reason.
1484
+
1485
+ 27:35.252
1486
+ This is even a more important
1487
+ reason.
1488
+
1489
+ 27:38.819
1490
+ Well, how can I say more
1491
+ important?
1492
+
1493
+ 27:42.187
1494
+ All those are important.
1495
+
1496
+ 27:44.565
1497
+ This is another way to see it.
1498
+
1499
+ 27:48.528
1500
+ A matrix has no inverse -- yeah
1501
+ -- here -- now this is
1502
+
1503
+ 27:55.417
1504
+ important.
1505
+
1506
+ 27:56.718
1507
+ A matrix has no -- a square
1508
+ matrix won't have an inverse if
1509
+
1510
+ 28:04.258
1511
+ there's no inverse because I can
1512
+ solve -- I can find an X of -- a
1513
+
1514
+ 28:12.577
1515
+ vector X with A times -- this A
1516
+ times X giving zero.
1517
+
1518
+ 28:20.377
1519
+ This is the reason I like best.
1520
+
1521
+ 28:24.973
1522
+ That matrix won't have an
1523
+ inverse.
1524
+
1525
+ 28:29.866
1526
+ Can you -- well,
1527
+ let me change I to U.
1528
+
1529
+ 28:35.351
1530
+ So tell me a vector X that,
1531
+ solves A X equals zero.
1532
+
1533
+ 28:42.763
1534
+ I mean, this is,
1535
+ like, the key equation.
1536
+
1537
+ 28:49.88
1538
+ In mathematics,
1539
+ all the key equations have zero
1540
+
1541
+ 28:53.485
1542
+ on the right-hand side.
1543
+
1544
+ 28:55.287
1545
+ So what's the X?
1546
+
1547
+ 28:56.541
1548
+ Tell me an X here -- so now I'm
1549
+ going to put -- slip in the X
1550
+
1551
+ 29:01.244
1552
+ that you tell me and I'm going
1553
+ to get zero.
1554
+
1555
+ 29:04.536
1556
+ What X would do that job?
1557
+
1558
+ 29:07.201
1559
+ Three and negative one?
1560
+
1561
+ 29:09.088
1562
+ Is that the one you picked,
1563
+ or -- yeah.
1564
+
1565
+ 29:12.208
1566
+ Or another -- well,
1567
+ if you picked zero with zero,
1568
+
1569
+ 29:16.148
1570
+ I'm not so excited,
1571
+ right?
1572
+
1573
+ 29:18.2
1574
+ Because that would always work.
1575
+
1576
+ 29:20.745
1577
+ So it's really the fact that
1578
+ this vector isn't zero that's
1579
+
1580
+ 29:25.424
1581
+ important.
1582
+
1583
+ 29:27.147
1584
+ It's a non-zero vector and
1585
+ three negative one would do it.
1586
+
1587
+ 29:32.196
1588
+ That just says three of this
1589
+ column minus one of that column
1590
+
1591
+ 29:37.422
1592
+ is the zero column.
1593
+
1594
+ 29:39.105
1595
+ Okay.
1596
+
1597
+ 29:39.548
1598
+ So now I know that A couldn't
1599
+ be invertible.
1600
+
1601
+ 29:43.357
1602
+ But what's the reasoning?
1603
+
1604
+ 29:45.571
1605
+ If A X is zero,
1606
+ suppose I multiplied by A
1607
+
1608
+ 29:49.114
1609
+ inverse.
1610
+
1611
+ 29:50.885
1612
+ Yeah, well here's the reason.
1613
+
1614
+ 29:53.826
1615
+ Here -- this is why this spells
1616
+ disaster for an inverse.
1617
+
1618
+ 29:59.403
1619
+ The matrix can't have an
1620
+ inverse if some combination of
1621
+
1622
+ 30:04.878
1623
+ the columns gives z- it gives
1624
+ nothing.
1625
+
1626
+ 30:09.339
1627
+ Because, I could take A X
1628
+ equals zero, I could multiply by
1629
+
1630
+ 30:12.837
1631
+ A inverse and what would I
1632
+ discover?
1633
+
1634
+ 30:14.984
1635
+ Suppose I take that equation
1636
+ and I multiply by -- if A
1637
+
1638
+ 30:18.236
1639
+ inverse existed,
1640
+ which of course I'm going to
1641
+
1642
+ 30:20.935
1643
+ come to the conclusion it can't
1644
+ because if it existed,
1645
+
1646
+ 30:24.187
1647
+ if there was an A inverse to
1648
+ this dopey matrix,
1649
+
1650
+ 30:27.009
1651
+ I would multiply that equation
1652
+ by that inverse and I would
1653
+
1654
+ 30:30.506
1655
+ discover X is zero.
1656
+
1657
+ 30:32.593
1658
+ If I multiply A by A inverse on
1659
+ the left, I get X.
1660
+
1661
+ 30:38.802
1662
+ If I multiply by A inverse on
1663
+ the right, I get zero.
1664
+
1665
+ 30:45.265
1666
+ So I would discover X was zero.
1667
+
1668
+ 30:49.194
1669
+ But it -- X is not zero.
1670
+
1671
+ 30:52.996
1672
+ X -- this guy wasn't zero.
1673
+
1674
+ 30:54.943
1675
+ There it is.
1676
+
1677
+ 30:55.842
1678
+ It's three minus one.
1679
+
1680
+ 30:57.416
1681
+ So, conclusion -- only,
1682
+ it takes us some time to really
1683
+
1684
+ 31:01.462
1685
+ work with that conclusion -- our
1686
+ conclusion will be that
1687
+
1688
+ 31:05.582
1689
+ non-invertible matrices,
1690
+ singular matrices,
1691
+
1692
+ 31:08.729
1693
+ some combinations of their
1694
+ columns gives the zero column.
1695
+
1696
+ 31:13.749
1697
+ They they take some vector X
1698
+ into zero.
1699
+
1700
+ 31:17.999
1701
+ And there's no way A inverse
1702
+ can recover, right?
1703
+
1704
+ 31:23.255
1705
+ That's what this equation says.
1706
+
1707
+ 31:26.723
1708
+ This equation says I take this
1709
+ vector X and multiplying by A
1710
+
1711
+ 31:33.321
1712
+ gives zero.
1713
+
1714
+ 31:35.446
1715
+ But then when I multiply by A
1716
+ inverse, I can never escape from
1717
+
1718
+ 31:41.664
1719
+ zero.
1720
+
1721
+ 31:42.174
1722
+ So there couldn't be an A
1723
+ inverse.
1724
+
1725
+ 31:45.538
1726
+ Where here -- okay,
1727
+ now fix -- all right.
1728
+
1729
+ 31:49.615
1730
+ Now let me take -- all right,
1731
+ back to the positive side.
1732
+
1733
+ 31:56.138
1734
+ Let's take a matrix that does
1735
+ have an inverse.
1736
+
1737
+ 32:00.35
1738
+ And why not invert it?
1739
+
1740
+ 32:02.409
1741
+ Okay.
1742
+
1743
+ 32:02.877
1744
+ Can I -- so let me take on this
1745
+ third board a matrix -- shall I
1746
+
1747
+ 32:08.679
1748
+ fix that up a little?
1749
+
1750
+ 32:10.644
1751
+ Tell me a matrix that has got
1752
+ an inverse.
1753
+
1754
+ 32:14.387
1755
+ Well, let me say one three two
1756
+ -- what shall I put there?
1757
+
1758
+ 32:20.657
1759
+ Well, don't put six,
1760
+ I guess is -- right?
1761
+
1762
+ 32:27.307
1763
+ Do I any favorites here?
1764
+
1765
+ 32:31.296
1766
+ One?
1767
+
1768
+ 32:31.961
1769
+ Or eight?
1770
+
1771
+ 32:33.457
1772
+ I don't care.
1773
+
1774
+ 32:35.618
1775
+ What, seven?
1776
+
1777
+ 32:37.613
1778
+ Seven.
1779
+
1780
+ 32:38.61
1781
+ Okay.
1782
+
1783
+ 32:39.442
1784
+ Seven is a lucky number.
1785
+
1786
+ 32:45.094
1787
+ All right, seven,
1788
+ okay.
1789
+
1790
+ 32:46.649
1791
+ Okay.
1792
+
1793
+ 32:47.002
1794
+ So -- now what's our idea?
1795
+
1796
+ 32:48.84
1797
+ We believe that this matrix is
1798
+ invertible.
1799
+
1800
+ 32:51.738
1801
+ Those who like determinants
1802
+ have quickly taken its
1803
+
1804
+ 32:55.202
1805
+ determinant and found it wasn't
1806
+ zero.
1807
+
1808
+ 32:57.747
1809
+ Those who like columns,
1810
+ and probably that --
1811
+
1812
+ 33:01.634
1813
+ that department is not totally
1814
+ popular yet -- but those who
1815
+
1816
+ 33:06.827
1817
+ like columns will look at those
1818
+ two columns and say,
1819
+
1820
+ 33:11.394
1821
+ hey, they point in different
1822
+ directions.
1823
+
1824
+ 33:14.885
1825
+ So I can get anything.
1826
+
1827
+ 33:16.855
1828
+ Now, let me see,
1829
+ what do I mean?
1830
+
1831
+ 33:20.436
1832
+ How I going to computer A
1833
+ inverse?
1834
+
1835
+ 33:24.119
1836
+ So A inverse -- here's A
1837
+ inverse, now,
1838
+
1839
+ 33:28.249
1840
+ and I have to find it.
1841
+
1842
+ 33:30.704
1843
+ And what do I get when I do
1844
+ this multiplication?
1845
+
1846
+ 33:35.949
1847
+ The identity.
1848
+
1849
+ 33:37.4
1850
+ You know, forgive me for taking
1851
+ two by two-s,
1852
+
1853
+ 33:42.31
1854
+ but --
1855
+ lt's good to keep the
1856
+
1857
+ 33:46.542
1858
+ computations manageable and let
1859
+ the ideas come out.
1860
+
1861
+ 33:52.099
1862
+ Okay, now what's the idea I
1863
+ want?
1864
+
1865
+ 33:55.656
1866
+ I'm looking for this matrix A
1867
+ inverse, how I going to find it?
1868
+
1869
+ 34:02.436
1870
+ Right now, I've got four
1871
+ numbers to find.
1872
+
1873
+ 34:07.883
1874
+ I'm going to look at the first
1875
+ column.
1876
+
1877
+ 34:12.001
1878
+ Let me take this first column,
1879
+ A B.
1880
+
1881
+ 34:15.787
1882
+ What's up there?
1883
+
1884
+ 34:17.568
1885
+ What -- tell me this.
1886
+
1887
+ 34:19.906
1888
+ What equation does the first
1889
+ column satisfy?
1890
+
1891
+ 34:24.692
1892
+ The first column satisfies A
1893
+ times that column is one zero.
1894
+
1895
+ 34:32.262
1896
+ The first column of the answer.
1897
+
1898
+ 34:34.862
1899
+ And the second column,
1900
+ C D, satisfies A times that
1901
+
1902
+ 34:38.97
1903
+ second column is zero one.
1904
+
1905
+ 34:41.15
1906
+ You see that finding the
1907
+ inverse is like solving two
1908
+
1909
+ 34:45.426
1910
+ systems.
1911
+
1912
+ 34:46.097
1913
+ One system, when the right-hand
1914
+ side is one zero -- I'm just
1915
+
1916
+ 34:51.043
1917
+ going to split it into two
1918
+ pieces.
1919
+
1920
+ 34:54.733
1921
+ I don't even need to rewrite
1922
+ it.
1923
+
1924
+ 34:59.927
1925
+ I can take A times -- so let me
1926
+ put it here.
1927
+
1928
+ 35:07.131
1929
+ A times column j of A inverse
1930
+ is column j of the identity.
1931
+
1932
+ 35:16.681
1933
+ I've got n equations.
1934
+
1935
+ 35:20.2
1936
+ I've got, well,
1937
+ two in this case.
1938
+
1939
+ 35:27.07
1940
+ And they have the same matrix,
1941
+ A, but they have different
1942
+
1943
+ 35:30.917
1944
+ right-hand sides.
1945
+
1946
+ 35:32.084
1947
+ The right-hand sides are just
1948
+ the columns of the identity,
1949
+
1950
+ 35:36
1951
+ this guy and this guy.
1952
+
1953
+ 35:37.511
1954
+ And these are the two
1955
+ solutions.
1956
+
1957
+ 35:39.641
1958
+ Do you see what I'm going --
1959
+ I'm looking at that equation by
1960
+
1961
+ 35:43.694
1962
+ columns.
1963
+
1964
+ 35:45
1965
+ I'm looking at A times this
1966
+ column, giving that guy,
1967
+
1968
+ 35:48.348
1969
+ and A times that column giving
1970
+ that guy.
1971
+
1972
+ 35:50.909
1973
+ So -- Essentially -- so this is
1974
+ like the Gauss -- we're back to
1975
+
1976
+ 35:54.979
1977
+ Gauss.
1978
+
1979
+ 35:55.373
1980
+ We're back to solving systems
1981
+ of equations,
1982
+
1983
+ 35:58.131
1984
+ but we're solving -- we've got
1985
+ two right-hand sides instead of
1986
+
1987
+ 36:02.136
1988
+ one.
1989
+
1990
+ 36:03.186
1991
+ That's where Jordan comes in.
1992
+
1993
+ 36:08.371
1994
+ So at the very beginning of the
1995
+ lecture, I mentioned
1996
+
1997
+ 36:17.489
1998
+ Gauss-Jordan,
1999
+ let me write it up again.
2000
+
2001
+ 36:24.283
2002
+ Okay.
2003
+
2004
+ 36:25.177
2005
+ Here's the Gauss-Jordan idea.
2006
+
2007
+ 36:30.362
2008
+ Gauss-Jordan solve two
2009
+ equations at once.
2010
+
2011
+ 36:39.123
2012
+ Okay.
2013
+
2014
+ 36:39.793
2015
+ Let me show you how the
2016
+ mechanics go.
2017
+
2018
+ 36:44.624
2019
+ How do I solve a single
2020
+ equation?
2021
+
2022
+ 36:48.917
2023
+ So the two equations are one
2024
+ three two seven,
2025
+
2026
+ 36:54.821
2027
+ multiplying A B gives one zero.
2028
+
2029
+ 36:58.98
2030
+ And the other equation is the
2031
+ same one three two seven
2032
+
2033
+ 37:06.091
2034
+ multiplying C D gives zero one.
2035
+
2036
+ 37:11.726
2037
+ Okay.
2038
+
2039
+ 37:12.21
2040
+ That'll tell me the two columns
2041
+ of the inverse.
2042
+
2043
+ 37:16.66
2044
+ I'll have inverse.
2045
+
2046
+ 37:18.401
2047
+ In other words,
2048
+ if I can solve with this matrix
2049
+
2050
+ 37:22.85
2051
+ A, if I can solve with that
2052
+ right-hand side and that
2053
+
2054
+ 37:27.783
2055
+ right-hand side,
2056
+ I'm invertible.
2057
+
2058
+ 37:30.781
2059
+ I've got it.
2060
+
2061
+ 37:31.942
2062
+ Okay.
2063
+
2064
+ 37:33.586
2065
+ And Jordan sort of said to
2066
+ Gauss, solve them together,
2067
+
2068
+ 37:38.815
2069
+ look at the matrix -- if we
2070
+ just solve this one,
2071
+
2072
+ 37:43.452
2073
+ I would look at one three two
2074
+ seven, and how do I deal with
2075
+
2076
+ 37:49.174
2077
+ the right-hand side?
2078
+
2079
+ 37:51.147
2080
+ I stick it on as an extra
2081
+ column, right?
2082
+
2083
+ 37:54.995
2084
+ That's this augmented matrix.
2085
+
2086
+ 37:58.842
2087
+ That's the matrix when I'm
2088
+ watching the right-hand side at
2089
+
2090
+ 38:03.493
2091
+ the same time,
2092
+ doing the same thing to the
2093
+
2094
+ 38:06.838
2095
+ right side that I do to the
2096
+ left?
2097
+
2098
+ 38:09.449
2099
+ So I just carry it along as an
2100
+ extra column.
2101
+
2102
+ 38:12.957
2103
+ Now I'm going to carry along
2104
+ two extra columns.
2105
+
2106
+ 38:16.71
2107
+ And I'm going to do whatever
2108
+ Gauss wants, right?
2109
+
2110
+ 38:21.523
2111
+ I'm going to do elimination.
2112
+
2113
+ 38:23.891
2114
+ I'm going to get this to be
2115
+ simple and this thing will turn
2116
+
2117
+ 38:28.796
2118
+ into the inverse.
2119
+
2120
+ 38:30.233
2121
+ This is what's coming.
2122
+
2123
+ 38:32.093
2124
+ I'm going to do elimination
2125
+ steps to make this into the
2126
+
2127
+ 38:36.66
2128
+ identity, and lo and behold,
2129
+ the inverse will show up here.
2130
+
2131
+ 38:41.564
2132
+ K--- let's do it.
2133
+
2134
+ 38:43.002
2135
+ Okay.
2136
+
2137
+ 38:44.355
2138
+ So what are the elimination
2139
+ steps?
2140
+
2141
+ 38:46.541
2142
+ So you see -- here's my matrix
2143
+ A and here's the identity,
2144
+
2145
+ 38:50.252
2146
+ like, stuck on,
2147
+ augmented on.
2148
+
2149
+ 38:52.108
2150
+ STUDENT: I'm sorry...
2151
+
2152
+ 38:53.499
2153
+ STRANG: Yeah?
2154
+
2155
+ 38:54.361
2156
+ STUDENT: -- is the two and the
2157
+ three supposed to be switched?
2158
+
2159
+ 38:59
2160
+ STRANG: Did I -- oh,
2161
+ no, they weren't supposed to be
2162
+
2163
+ 39:03.775
2164
+ switched.
2165
+
2166
+ 39:04.618
2167
+ Sorry.
2168
+
2169
+ 39:05.179
2170
+ Thanks.
2171
+
2172
+ 39:05.835
2173
+ Okay.
2174
+
2175
+ 39:06.303
2176
+ Thank you very much.
2177
+
2178
+ 39:08.176
2179
+ And there -- I've got them
2180
+ right.
2181
+
2182
+ 39:11.172
2183
+ Okay, thanks.
2184
+
2185
+ 39:12.389
2186
+ Okay.
2187
+
2188
+ 39:12.857
2189
+ So let's do elimination.
2190
+
2191
+ 39:15.105
2192
+ All right, it's going to be
2193
+ simple, right?
2194
+
2195
+ 39:20.255
2196
+ So I take two of this row away
2197
+ from this row.
2198
+
2199
+ 39:23.616
2200
+ So this row stays the same and
2201
+ two of those come away from
2202
+
2203
+ 39:27.971
2204
+ this.
2205
+
2206
+ 39:28.353
2207
+ That leaves me with a zero and
2208
+ a one and two of these away from
2209
+
2210
+ 39:33.089
2211
+ this is that what you're getting
2212
+ -- after one elimination step --
2213
+
2214
+ 39:38.666
2215
+ Let me sort of separate the --
2216
+ the left half from the right
2217
+
2218
+ 39:43.815
2219
+ half.
2220
+
2221
+ 39:44.258
2222
+ So two of that first row got
2223
+ subtracted from the second row.
2224
+
2225
+ 39:49.495
2226
+ Now this is an upper triangular
2227
+ form.
2228
+
2229
+ 39:52.69
2230
+ Gauss would quit,
2231
+ but Jordan says keeps going.
2232
+
2233
+ 39:56.685
2234
+ Use elimination upwards.
2235
+
2236
+ 39:59.702
2237
+ Subtract a multiple of equation
2238
+ two from equation one to get rid
2239
+
2240
+ 40:04.293
2241
+ of the three.
2242
+
2243
+ 40:05.24
2244
+ So let's go the whole way.
2245
+
2246
+ 40:07.135
2247
+ So now I'm going to -- this guy
2248
+ is fine, but I'm going to --
2249
+
2250
+ 40:11.434
2251
+ what do I do now?
2252
+
2253
+ 40:12.673
2254
+ What's my final step that
2255
+ produces the inverse?
2256
+
2257
+ 40:16.025
2258
+ I multiply this by the right
2259
+ number to get up to ther to
2260
+
2261
+ 40:20.032
2262
+ remove that three.
2263
+
2264
+ 40:22.219
2265
+ So I guess, I -- since this is
2266
+ a one, there's the pivot sitting
2267
+
2268
+ 40:26.626
2269
+ there.
2270
+
2271
+ 40:27.053
2272
+ I multiply it by three and
2273
+ subtract from that,
2274
+
2275
+ 40:30.252
2276
+ so what do I get?
2277
+
2278
+ 40:31.46
2279
+ I'll have one zero -- oh,
2280
+ yeah that was my whole point.
2281
+
2282
+ 40:35.299
2283
+ I'll multiply this by three and
2284
+ subtract from that,
2285
+
2286
+ 40:38.853
2287
+ which will give me seven.
2288
+
2289
+ 40:41.413
2290
+ And I multiply this by three
2291
+ and subtract from that,
2292
+
2293
+ 40:50.784
2294
+ which gives me a minus three.
2295
+
2296
+ 40:56.113
2297
+ And what's my hope,
2298
+ belief?
2299
+
2300
+ 41:00.891
2301
+ Here I started with A and the
2302
+ identity, and I ended up with
2303
+
2304
+ 41:11.549
2305
+ the identity and who?
2306
+
2307
+ 41:16.878
2308
+ That better be A inverse.
2309
+
2310
+ 41:19.699
2311
+ That's the Gauss Jordan idea.
2312
+
2313
+ 41:22.972
2314
+ Start with this long matrix,
2315
+ double-length A I,
2316
+
2317
+ 41:28.164
2318
+ eliminate, eliminate until this
2319
+ part is down to I,
2320
+
2321
+ 41:33.695
2322
+ then this one will -- must be
2323
+ for some reason,
2324
+
2325
+ 41:38.773
2326
+ and we've got to find the
2327
+ reason -- must be A inverse.
2328
+
2329
+ 41:44.755
2330
+ Shall I just check that it
2331
+ works?
2332
+
2333
+ 41:49.722
2334
+ Let me just check that -- can I
2335
+ multiply this matrix this part
2336
+
2337
+ 41:54.113
2338
+ times A, I'll carry A over here
2339
+ and just do that multiplication.
2340
+
2341
+ 41:58.648
2342
+ You'll see I'll do it the old
2343
+ fashioned way.
2344
+
2345
+ 42:01.744
2346
+ Seven minus six is a one.
2347
+
2348
+ 42:03.544
2349
+ Twenty one minus twenty one is
2350
+ a zero, minus two plus two is a
2351
+
2352
+ 42:07.935
2353
+ zero, minus six plus seven is a
2354
+ one.
2355
+
2356
+ 42:11.246
2357
+ Check.
2358
+
2359
+ 42:11.871
2360
+ So that is the inverse.
2361
+
2362
+ 42:14.265
2363
+ That's the Gauss-Jordan idea.
2364
+
2365
+ 42:17.283
2366
+ So, you'll -- one of the
2367
+ homework problems or more than
2368
+
2369
+ 42:22.904
2370
+ one for Wednesday will ask you
2371
+ to go through those steps.
2372
+
2373
+ 42:28.733
2374
+ I think you just got to go
2375
+ through Gauss-Jordan a couple of
2376
+
2377
+ 42:34.769
2378
+ times, but I --
2379
+ yeah -- just to see the
2380
+
2381
+ 42:40.382
2382
+ mechanics.
2383
+
2384
+ 42:41.691
2385
+ But the, important thing is,
2386
+ why -- is, like,
2387
+
2388
+ 42:47.45
2389
+ what happened?
2390
+
2391
+ 42:49.283
2392
+ Why did we -- why did we get A
2393
+ inverse there?
2394
+
2395
+ 42:55.042
2396
+ Let me ask you that.
2397
+
2398
+ 42:57.66
2399
+ We got -- so we take --
2400
+ We do row reduction,
2401
+
2402
+ 43:04.155
2403
+ we do elimination on this long
2404
+ matrix A I until the first half
2405
+
2406
+ 43:11.19
2407
+ is up.
2408
+
2409
+ 43:11.882
2410
+ Then a second half is A
2411
+ inverse.
2412
+
2413
+ 43:15.458
2414
+ Well, how do I see that?
2415
+
2416
+ 43:18.226
2417
+ Let me put up here how I see
2418
+ that.
2419
+
2420
+ 43:22.032
2421
+ So here's my Gauss-Jordan
2422
+ thing, and I'm doing stuff to
2423
+
2424
+ 43:28.26
2425
+ it.
2426
+
2427
+ 43:29.989
2428
+ So I'm -- well,
2429
+ whole lot of E's.
2430
+
2431
+ 43:32.697
2432
+ Remember those are those
2433
+ elimination matrices.
2434
+
2435
+ 43:36.503
2436
+ Those are the -- those are the
2437
+ things that we figured out last
2438
+
2439
+ 43:41.663
2440
+ time.
2441
+
2442
+ 43:42.086
2443
+ Yes, that's what an elimination
2444
+ step is it's in matrix form,
2445
+
2446
+ 43:47.078
2447
+ I'm multiplying by some Es.
2448
+
2449
+ 43:50.208
2450
+ And the result -- well,
2451
+ so I'm multiplying by a whole
2452
+
2453
+ 43:54.728
2454
+ bunch of Es.
2455
+
2456
+ 43:55.772
2457
+ So, I get a -- can I call the
2458
+ overall matrix E?
2459
+
2460
+ 43:59.771
2461
+ That's the elimination matrix,
2462
+ the product of all those little
2463
+
2464
+ 44:05.074
2465
+ pieces.
2466
+
2467
+ 44:05.683
2468
+ What do I mean by little
2469
+ pieces?
2470
+
2471
+ 44:09.248
2472
+ Well, there was an elimination
2473
+ matrix that subtracted two of
2474
+
2475
+ 44:14.126
2476
+ that away from that.
2477
+
2478
+ 44:15.779
2479
+ Then there was an elimination
2480
+ matrix that subtracted three of
2481
+
2482
+ 44:20.74
2483
+ that away from that.
2484
+
2485
+ 44:22.394
2486
+ I guess in this case,
2487
+ that was all.
2488
+
2489
+ 44:25.867
2490
+ So there were just two Es in
2491
+ this case, one that did this
2492
+
2493
+ 44:30.872
2494
+ step and one that did this step
2495
+ and together they gave me an E
2496
+
2497
+ 44:36.324
2498
+ that does both steps.
2499
+
2500
+ 44:38.201
2501
+ And the net result was to get
2502
+ an I here.
2503
+
2504
+ 44:41.687
2505
+ And you can tell me what that
2506
+ has to be.
2507
+
2508
+ 44:45.173
2509
+ This is, like,
2510
+ the picture of what happened.
2511
+
2512
+ 44:50
2513
+ If E multiplied A,
2514
+ whatever that E is -- we never
2515
+
2516
+ 44:57.343
2517
+ figured it out in this way.
2518
+
2519
+ 45:01.474
2520
+ But whatever that E times that
2521
+ E is, E times A is -- What's E
2522
+
2523
+ 45:10.654
2524
+ times A?
2525
+
2526
+ 45:11.878
2527
+ It's I.
2528
+
2529
+ 45:14.019
2530
+ That E, whatever the heck it
2531
+ was, multiplied A and produced
2532
+
2533
+ 45:20.25
2534
+ I.
2535
+
2536
+ 45:20.465
2537
+ So E must be -- E A equaling I
2538
+ tells us what E is,
2539
+
2540
+ 45:25.729
2541
+ namely it is -- STUDENT:
2542
+ It's the inverse of A.
2543
+
2544
+ 45:30.67
2545
+ STRANG: It's the inverse of A.
2546
+
2547
+ 45:33.893
2548
+ Great.
2549
+
2550
+ 45:34.538
2551
+ And therefore,
2552
+ when the second half,
2553
+
2554
+ 45:38.298
2555
+ when E multiplies I,
2556
+ it's E --
2557
+
2558
+ 45:42.809
2559
+ Put this A inverse.
2560
+
2561
+ 45:44.09
2562
+ You see the picture looking
2563
+ that way?
2564
+
2565
+ 45:46.515
2566
+ E times A is the identity.
2567
+
2568
+ 45:48.266
2569
+ It tells us what E has to be.
2570
+
2571
+ 45:50.22
2572
+ It has to be the inverse,
2573
+ and therefore,
2574
+
2575
+ 45:52.848
2576
+ on the right-hand side,
2577
+ where E -- where we just
2578
+
2579
+ 45:56.014
2580
+ smartly tucked on the identity,
2581
+ it's turning in,
2582
+
2583
+ 45:59.181
2584
+ step by step --
2585
+ It's turning into A inverse.
2586
+
2587
+ 46:04.197
2588
+ There is the statement of
2589
+ Gauss-Jordan elimination.
2590
+
2591
+ 46:09.908
2592
+ That's how you find the
2593
+ inverse.
2594
+
2595
+ 46:13.449
2596
+ Where we can look at it as
2597
+ elimination, as solving n
2598
+
2599
+ 46:19.274
2600
+ equations at the same time --
2601
+ -- and tacking on n columns,
2602
+
2603
+ 46:26.682
2604
+ solving those equations and up
2605
+ goes the n columns of A inverse
2606
+
2607
+ 46:33.617
2608
+ .
2609
+
2610
+ 46:33.73
2611
+ Okay, thanks.
2612
+
2613
+ 46:35.208
2614
+ See you on Wednesday.