-
Notifications
You must be signed in to change notification settings - Fork 3
Expand file tree
/
Copy pathmt234.tex
More file actions
1321 lines (1044 loc) · 49.5 KB
/
mt234.tex
File metadata and controls
1321 lines (1044 loc) · 49.5 KB
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
\frfilename{mt234.tex}
\versiondate{11.4.09}
\copyrightdate{2008}
\def\chaptername{The Radon-Nikod\'ym theorem}
\def\sectionname{Operations on measures}
\newsection{234}
I take a few pages to describe some standard constructions. The ideas are
straightforward, but a number of details need to be worked
out if they are to be securely integrated into the general framework I
employ. The first step is to formally introduce \imp\ functions
(234A-234B), the most important class of transformations between measure
spaces. For construction of new measures, we have the notions of
image measure (234C-234E), %234C 234D 234E
sum of measures (234G-234H) and indefinite-integral measure
(234I-234O). %234I 234J 234K 234L 234M 234N 234O
Finally I mention a way of ordering the measures on a given set
(234P-234Q).
\leader{234A}{Inverse-measure-preserving
functions}\dvAformerly{2{}35G}\cmmnt{ It is high
time that I introduced the nearest thing in measure theory to a
`morphism'.} If $(X,\Sigma,\mu)$ and $(Y,\Tau,\nu)$ are measure
spaces, a function $\phi:X\to Y$ is {\bf \imp} if
$\phi^{-1}[F]\in\Sigma$ and $\mu(\phi^{-1}[F])=\nu F$ for every
$F\in\Tau$.
\leader{234B}{Proposition} Let $(X,\Sigma,\mu)$ and $(Y,\Tau,\nu)$ be
measure spaces, and $\phi:X\to Y$ an \imp\ function.
(a)\dvAformerly{2{}35Hc} If $\hat\mu$, $\hat\nu$ are the completions of
$\mu$, $\nu$
respectively, $\phi$ is also \imp\ for $\hat\mu$ and $\hat\nu$.
(b)\dvAformerly{2{}35Xe} $\mu$ is a probability measure iff $\nu$ is a probability measure.
(c)\dvAformerly{2{}35Xe}
$\mu$ is totally finite iff $\nu$ is totally finite.
(d)\dvAformerly{2{}35Xe}(i)
If $\nu$ is $\sigma$-finite, then $\mu$ is $\sigma$-finite.
\quad(ii) If $\nu$ is semi-finite and $\mu$ is $\sigma$-finite, then $\nu$
is $\sigma$-finite.
(e)\dvAformerly{2{}35Xe}(i)
If $\nu$ is $\sigma$-finite and atomless, then $\mu$ is atomless.
\quad(ii) If $\nu$ is semi-finite and $\mu$ is purely atomic, then
$\nu$ is purely atomic.
(f)\dvAnew{2008}(i) $\mu^*\phi^{-1}[B]\le\nu^*B$ for
every $B\subseteq Y$.
\quad(ii) $\mu^*A\le\nu^*\phi[A]$ for every $A\subseteq X$.
(g)\dvAformerly{2{}35Hc}
If $(Z,\Lambda,\lambda)$ is another measure space, and
$\psi:Y\to Z$ is \imp, then $\psi\phi:X\to Z$ is \imp.
\proof{{\bf (a)} If $\hat\nu$ measures $F$, there are $F'$, $F''\in\Tau$
such that $F'\subseteq F\subseteq F''$ and $\nu(F''\setminus F')=0$.
Now
\Centerline{$\phi^{-1}[F']\subseteq\phi^{-1}[F]\subseteq\phi^{-1}[F'']$,
\quad$\mu(\phi^{-1}[F'']\setminus\phi^{-1}[F'])=\nu(F''\setminus F')=0$,}
\noindent so $\hat\mu$ measures $\phi^{-1}[F]$ and
\Centerline{$\hat\mu(\phi^{-1}[F])=\mu\phi^{-1}[F']=\nu F'=\hat\nu F$.}
\noindent As $F$ is arbitrary, $\phi$ is \imp\ for $\hat\mu$ and $\hat\nu$.
\medskip
{\bf (b)-(c)} are surely obvious.
\medskip
{\bf (d)(i)} If $\sequencen{F_n}$ is a cover of $Y$ by sets of finite
measure for $\nu$, then $\sequencen{\phi^{-1}[F_n]}$ is a cover of $X$ by
sets of finite measure for $\mu$.
\medskip
\quad{\bf (ii)} Let $\Cal F\subseteq\Tau$ be a disjoint family of
non-$\nu$-negligible sets. Then $\family{F}{\Cal F}{\phi^{-1}[F]}$ is a
disjoint family of non-$\mu$-negligible sets. By 215B(iii), $\Cal F$ is
countable. By 215B(iii) in the opposite direction, $\nu$ is
$\sigma$-finite.
\medskip
{\bf (e)(i)} Suppose that $E\in\Sigma$ and $\mu E>0$. Let
$\sequencen{F_n}$ be a cover of $Y$ by sets of finite measure for $\nu$.
Because $\nu$ is atomless, we can find, for each $n$, a
finite partition $\family{i}{I_n}{F_{ni}}$
of $F_n$ such that $\nu F_{ni}<\mu E$ for every $i\in I_n$ (use 215D
repeatedly). Now
$X=\bigcup_{n\in\Bbb N,i\in I_n}\phi^{-1}[F_{ni}]$, so there are
$n\in\Bbb N$ and $i\in I_n$ with
\Centerline{$0<\mu(E\cap\phi^{-1}[F_{ni})
\le\mu\phi^{-1}[F_{ni}]=\nu F_{ni}<\mu E$,}
\noindent and $E$ is not a $\mu$-atom. As $E$ is arbitrary, $\mu$ is
atomless.
\medskip
\quad{\bf (ii)} Suppose that $F\in\Tau$ and $\nu F>0$. Because $\nu$ is
semi-finite, there is an $F_1\subseteq F$ such that $0<\nu F_1<\infty$.
Now $\mu\phi^{-1}[F_1]>0$; because $\mu$ is purely atomic, there is a
$\mu$-atom $E\subseteq\phi^{-1}[F_1]$.
Let $\Cal G$ be the set of those $G\in\Tau$ such that $G\subseteq F_1$ and
$\mu(E\cap\phi^{-1}[G])=0$. Then the union of any sequence in $\Cal G$
belongs to $\Cal G$, so by 215Ac there is an $H\in\Cal G$ such that
$\nu(G\setminus H)=0$ whenever $G\in\Cal G$. Consider $F_1\setminus H$.
We have
\Centerline{$\nu(F_1\setminus H)
=\mu(\phi^{-1}[F_1]\setminus\phi^{-1}[H])
\ge\mu(E\setminus\phi^{-1}[H])
=\mu E>0$.}
\noindent If $G\in\Tau$ and $G\subseteq F_1\setminus H$,
then one of $E\cap\phi^{-1}[G]$, $E\setminus\phi^{-1}[G]$ is
$\mu$-negligible.
In the former case, $G\in\Cal G$ and $G=G\setminus H$ is $\nu$-negligible.
In the
latter case, $F_1\setminus G\in\Cal G$ and $(F_1\setminus H)\setminus G$ is
$\nu$-negligible. As $G$ is arbitrary, $F_1\setminus H$ is a $\nu$-atom
included in $F$; as $F$ is arbitrary, $\nu$ is purely atomic.
\medskip
{\bf (f)(i)} Let $F\in\Tau$ be such that $B\subseteq F$ and $\nu^*B=\nu F$
(132Aa); then $\phi^{-1}[B]\subseteq\phi^{-1}[F]$ so
\Centerline{$\mu^*\phi^{-1}[B]\le\mu\phi^{-1}[F]=\nu F=\nu^*B$.}
\medskip
\quad{\bf (ii)} $\mu^*A\le\mu^*(\phi^{-1}[\phi[A]])\le\nu^*\phi[A]$ by (i).
\medskip
{\bf (g)} For any $W\in\Lambda$,
\Centerline{$\mu(\psi\phi)^{-1}[W]=\mu\phi^{-1}[\psi^{-1}[W]]
=\nu\psi^{-1}[W]=\lambda W$.}
}%end of proof of 234B
\leader{234C}{Image \dvrocolon{measures}}\dvAformerly{1{}12E}\cmmnt{ The
following
construction is one of the commonest ways in which new measure spaces
appear.
\medskip
\noindent}{\bf Proposition} Let $(X,\Sigma,\mu)$ be a measure space, $Y$
any set, and $\phi:X\to Y$ a function. Set
\Centerline{$\Tau=\{F:F\subseteq Y,\,\phi^{-1}[F]\in\Sigma\}$,
\quad$\nu F=\mu(\phi^{-1}[F])$ for every $F\in\Tau$.}
\noindent Then $(Y,\Tau,\nu)$ is a measure space.
\proof{{\bf (a)} $\emptyset=\phi^{-1}[\emptyset]\in\Sigma$ so
$\emptyset\in\Tau$.
\medskip
{\bf (b)} If $F\in\Tau$, then $\phi^{-1}[F]\in\Sigma$, so
$X\setminus\phi^{-1}[F]\in\Sigma$; but
$X\setminus\phi^{-1}[F]=\phi^{-1}[Y\setminus F]$, so $Y\setminus
F\in\Tau$.
\medskip
{\bf (c)} If $\sequencen{F_n}$ is a sequence in $\Tau$,
then $\phi^{-1}[F_n]\in\Sigma$ for every $n$, so
$\bigcup_{n\in\Bbb N}\phi^{-1}[F_n]\in\Sigma$; but
$\phi^{-1}[\bigcup_{n\in\Bbb N}F_n]=\bigcup_{n\in\Bbb N}\phi^{-1}[F_n]$,
so $\bigcup_{n\in\Bbb N}F_n\in\Tau$.
Thus $\Tau$ is a $\sigma$-algebra.
\medskip
{\bf (d)} $\nu\emptyset=\mu\phi^{-1}[\emptyset]=\mu\emptyset=0$.
\medskip
{\bf (e)} If $\sequencen{F_n}$ is a disjoint sequence in $\Tau$, then
$\sequencen{\phi^{-1}[F_n]}$ is a disjoint sequence in $\Sigma$, so
\Centerline{$\nu(\bigcup_{n\in\Bbb N}F_n)
=\mu\phi^{-1}[\bigcup_{n\in\Bbb N}F_n]
=\mu(\bigcup_{n\in\Bbb N}\phi^{-1}[F_n])
=\sum_{n=0}^{\infty}\mu\phi^{-1}[F_n]
=\sum_{n=0}^{\infty}\nu F_n$.}
\noindent So $\nu$ is a measure.
}%end of proof of 234C
\leader{234D}{Definition}\dvAformerly{1{}12F}
In the context of 234C, $\nu$ is called
the {\bf image measure}\cmmnt{ or {\bf push-forward measure}};
I will denote it $\mu\phi^{-1}$.
\cmmnt{\medskip
\noindent{\bf Remark} I ought perhaps to say that this construction does
not always produce exactly the `right' measure on $Y$; there are
circumstances in which some modification of the measure $\mu\phi^{-1}$
described here is more useful. But I will note these explicitly when
they occur; when I use the unadorned phrase `image measure' I shall
mean the measure constructed above.
}%end of comment
\leader{234E}{Proposition} Let $(X,\Sigma,\mu)$ be a measure space, $Y$ a
set and $\phi:X\to Y$ a function; let $\mu\phi^{-1}$ be the image measure
on $Y$.
(a) $\phi$ is \imp\ for $\mu$ and $\mu\phi^{-1}$.
(b)\dvAformerly{2{}12Bd} If $\mu$ is complete, so is $\mu\phi^{-1}$.
(c)\dvAformerly{1{}12Xd} If $Z$ is another set, and $\psi:Y\to Z$ a function,
then the image measures $\mu(\psi\phi)^{-1}$ and $(\mu\phi^{-1})\psi^{-1}$
on $Z$ are the same.
\proof{{\bf (a)} Immediate from the definitions.
\medskip
{\bf (b)} Write $\nu$ for $\mu\phi^{-1}$ and $\Tau$ for its domain. If
$\nu^*B=0$, then $\mu^*\phi^{-1}[B]=0$, by 234B(f-i); as $\mu$ is
complete, $\phi^{-1}[B]\in\Sigma$, so $B\in\Tau$. As $B$ is arbitrary,
$\nu$ is complete.
\medskip
{\bf (c)} For $G\subseteq Z$ and $u\in[0,\infty]$,
$$\eqalign{(\mu(\psi\phi)^{-1})(G)&\text{ is defined and equal to }u\cr
&\iff\mu((\psi\phi)^{-1}[G])\text{ is defined and equal to }u\cr
&\iff\mu(\phi^{-1}[\psi^{-1}[G]])\text{ is defined and equal to }u\cr
&\iff(\mu\phi^{-1})(\psi^{-1}[G]))\text{ is defined and equal to }u\cr
&\iff((\mu\phi^{-1})\psi^{-1})(G)\text{ is defined and equal to }u.\cr}$$
}%end of proof of 234E
\leader{*234F}{}\dvAformerly{1{}32G}\cmmnt{ In the opposite direction, the
following construction of a pull-back measure is sometimes useful.
\medskip
\noindent}{\bf Proposition} Let $X$ be a set, $(Y,\Tau,\nu)$ a measure
space, and
$\phi:X\to Y$ a function such that $\phi [X]$ has full outer measure in $Y$.
Then there is a measure $\mu$ on $X$, with domain
$\Sigma=\{\phi^{-1}[F]:F\in\Tau\}$, such that $\phi$ is \imp\ for $\mu$ and
$\nu$.
\proof{ The check that $\Sigma$ is a $\sigma$-algebra of subsets of
$X$ is straightforward; all we need to know is that
$\phi^{-1}[\emptyset]=\emptyset$,
$X\setminus\phi^{-1}[F]=\phi^{-1}[Y\setminus F]$ for every $F\subseteq Y$,
and that
$\phi^{-1}[\bigcup_{n\in\Bbb N}F_n]=\bigcup_{n\in\Bbb N}\phi^{-1}[F_n]$ for
every sequence $\sequencen{F_n}$ of subsets of $Y$. The key fact is
that if $F_1$, $F_2\in\Tau$ and $\phi^{-1}[F_1]=\phi^{-1}[F_2]$,
then $\phi[X]$ does not meet $F_1\symmdiff F_2$; because $\phi[X]$ has full
outer measure, $F_1\symmdiff F_2$ is $\nu$-negligible and
$\nu F_1=\nu F_2$. Accordingly the formula $\mu\phi^{-1}[F]=\nu F$
does define a function $\mu:\Sigma\to[0,\infty]$. Now
\Centerline{$\mu\emptyset=\mu\phi^{-1}[\emptyset]=\nu\emptyset=0$.}
\noindent Next, if $\sequencen{E_n}$ is a disjoint sequence in $\Sigma$,
choose $F_n\in\Tau$ such that $E_n=\phi^{-1}[F_n]$ for each $n\in\Bbb N$.
The sequence $\sequencen{F_n}$ need not be disjoint, but if we set
$F'_n=F_n\setminus\bigcup_{i<n}F_i$ for each $n\in\Bbb N$, then
$\sequencen{F'_n}$ is disjoint and
\Centerline{$E_n=E_n\setminus\bigcup_{i<n}E_i=\phi^{-1}[F'_n]$}
\noindent for each $n$; so
\Centerline{$\mu(\bigcup_{n\in\Bbb N}E_n)
=\nu(\bigcup_{n\in\Bbb N}F'_n)
=\sum_{n=0}^{\infty}\nu F'_n
=\sum_{n=0}^{\infty}\mu E_n$.}
\noindent As $\sequencen{E_n}$ is arbitrary, $\mu$ is a measure on
$X$, as required.
}%end of proof of 234F
\leader{234G}{Sums of \dvrocolon{measures}}\dvAformerly{1{}12Ya;
1{}12Xe for sum of two measures}\cmmnt{ I come now to a quite
different way of building measures. The idea is an obvious one, but the
technical details, in the general case I wish to examine, need watching.
\medskip
\noindent}{\bf Proposition}
Let $X$ be a set, and $\familyiI{\mu_i}$ a family of measures on $X$.
For each $i\in I$, let $\Sigma_i$ be the
domain of $\mu_i$. Set $\Sigma=\Cal PX\cap\bigcap_{i\in I}\Sigma_i$ and
define $\mu:\Sigma\to[0,\infty]$ by setting $\mu E=\sum_{i\in I}\mu_iE$ for
every $E\in\Sigma$. Then $\mu$ is a measure on $X$.
\proof{ $\Sigma$ is a $\sigma$-algebra of subsets of $X$ because every
$\Sigma_i$ is. (Apply 111Ga with
$\frak S=\{\Sigma_i:i\in I\}\cup\{\Cal PX\}$.)
Of course $\mu$ takes values in $[0,\infty]$ (226A).
$\mu\emptyset=0$ because $\mu_i\emptyset=0$ for every $i$. If
$\sequencen{E_n}$ is a disjoint sequence in $\Sigma$ with union $E$, then
$$\eqalignno{\mu E
&=\sum_{i\in I}\mu_iE
=\sum_{i\in I}\sum_{n=0}^{\infty}\mu_iE_n
=\sum_{n=0}^{\infty}\sum_{i\in I}\mu_iE_n\cr
\displaycause{226Af}
&=\sum_{n=0}^{\infty}\mu E_n.\cr}$$
\noindent So $\mu$ is a measure.
}%end of proof of 234G
\medskip
\noindent{\bf Remark} In this context, I will call $\mu$ the {\bf sum} of
the family $\familyiI{\mu_i}$.
\leader{234H}{Proposition}\dvAnew{2008}
Let $X$ be a set and $\familyiI{\mu_i}$ a family
of complete measures on $X$ with sum $\mu$.
(a) $\mu$ is complete.
(b)(i) A subset of $X$ is $\mu$-negligible iff it is $\mu_i$-negligible for
every $i\in I$.
\quad(ii) A subset of $X$ is $\mu$-conegligible iff it is
$\mu_i$-conegligible for every $i\in I$.
(c)\dvAformerly{2{}12Xh, for sum of two measures}
Let $f$ be a function from a subset of $X$ to $[-\infty,\infty]$.
Then $\int fd\mu$ is defined in $[-\infty,\infty]$ iff $\int fd\mu_i$ is
defined in $[-\infty,\infty]$ for every $i$ and one of
$\sum_{i\in I}f^+d\mu_i$, $\sum_{i\in I}f^-d\mu_i$ is finite, and in this
case $\int fd\mu=\sum_{i\in I}\int fd\mu_i$.
\proof{ Write $\Sigma_i=\dom\mu_i$ for $i\in I$,
$\Sigma=\Cal PX\cap\bigcap_{i\in I}\Sigma_i=\dom\mu$.
\medskip
{\bf (a)} If $E\subseteq F\in\Sigma$ and $\mu F=0$, then $\mu_iF=0$ for
every $i\in I$; because $\mu_i$ is complete, $E_i\in\Sigma_i$ for
every $i\in I$, and $E\in\Sigma$.
\medskip
{\bf (b)} This now follows at once, since a set $A\subseteq X$ is
$\mu$-negligible iff $\mu A=0$.
\medskip
{\bf (c)(i)} Note first that (b-ii) tells us that, under either hypothesis,
$\dom f$ is conegligible both for $\mu$ and for every $\mu_i$, so that
if we extend $f$ to $X$ by giving it the value $0$ on $X\setminus\dom f$
then neither $\int fd\mu$ nor $\sum_{i\in I}\int fd\mu_i$ is affected. So
let us assume from now on that $f$ is defined everywhere on $X$. Now it
is plain that
either hypothesis ensures that $f$ is $\Sigma$-measurable, that is, is
$\Sigma_i$-measurable for every $i\in I$.
\medskip
\quad{\bf (ii)} Suppose that $f$ is non-negative.
For $n\in\Bbb N$ set
$f_n(x)=\sum_{k=1}^{4^n}2^{-n}\chi\{x:f(x)\ge 2^{-n}k\}$, so that
$\sequencen{f_n}$ is a non-decreasing sequence with supremum $f$. We have
$$\eqalign{\int f_nd\mu
&=\sum_{k=1}^{4^n}2^{-n}\mu\{x:f(x)\ge 2^{-n}k\}
=\sum_{k=1}^{4^n}\sum_{i\in I}2^{-n}\mu_i\{x:f(x)\ge 2^{-n}k\}\cr
&=\sum_{i\in I}\sum_{k=1}^{4^n}2^{-n}\mu_i\{x:f(x)\ge 2^{-n}k\}
=\sum_{i\in I}\int f_nd\mu_i\cr}$$
\noindent for every $n$, so
$$\eqalign{\int fd\mu
&=\sup_{n\in\Bbb N}\int f_nd\mu
=\sup_{n\in\Bbb N}\sup_{J\subseteq I\text{ is finite}}
\sum_{i\in J}\int f_nd\mu_i
=\sup_{J\subseteq I\text{ is finite}}\sup_{n\in\Bbb N}
\sum_{i\in J}\int f_nd\mu_i\cr
&=\sup_{J\subseteq I\text{ is finite}}
\lim_{n\to\infty}\sum_{i\in J}\int f_nd\mu_i
=\sup_{J\subseteq I\text{ is finite}}
\sum_{i\in J}\lim_{n\to\infty}\int f_nd\mu_i
=\sum_{i\in I}\int fd\mu_i.\cr}$$
\medskip
\quad{\bf (iii)} Generally,
$$\eqalign{\int fd\mu&\text{ is defined in }[\infty,\infty]\cr
&\iff\int f^+d\mu\text{ and }\int f^-d\mu
\text{ are defined and at most one is infinite}\cr
&\iff\sum_{i\in I}\int f^+d\mu_i\text{ and }\sum_{i\in I}\int f^-d\mu_i
\text{ are defined and at most one is infinite}\cr
&\iff\int fd\mu_i\text{ is defined for every }i
\text{ and at most one of }\sum_{i\in I}\int f^+d\mu_i,\cr
&\mskip400mu
\sum_{i\in I}\int f^-d\mu_i\text{ is infinite},\cr}$$
\noindent and in this case
\Centerline{$\int fd\mu
=\int f^+d\mu-\int f^-d\mu
=\sum_{i\in I}\int f^+d\mu_i-\sum_{i\in I}\int f^-d\mu_i
=\sum_{i\in I}\int fd\mu_i$.}
}%end of proof of 234H
\leader{234I}{Indefinite-integral
\dvrocolon{measures}}\dvAformerly{2{}34A}\cmmnt{ Extending an
idea already used in 232D, we are led to the following construction; once
again, we need to take care over the formal details if we want
to get full value from it.
\medskip
\noindent}{\bf Theorem} Let $(X,\Sigma,\mu)$ be a measure
space, and $f$ a non-negative $\mu$-virtually measurable real-valued
function defined on a conegligible subset of $X$. Write
$\nu F=\int f\times\chi F\,d\mu$ whenever $F\subseteq X$ is such that
the integral is
defined in $[0,\infty]$\cmmnt{ according to the conventions of 133A}.
Then $\nu$ is a complete measure on $X$, and its domain includes
$\Sigma$.
%the trouble with taking [0,\infty]-valued functions is that if
%I allow f=\infty on a non-negligible set then \nu becomes
%non-semi-finite for a silly reason
\proof{{\bf (a)} Write $\Tau$ for the domain of $\nu$, that is, the
family of sets $F\subseteq X$ such that $\int f\times\chi F\,d\mu$ is
defined in
$[0,\infty]$, that is, $f\times\chi F$ is $\mu$-virtually measurable
(133A). Then $\Tau$ is a $\sigma$-algebra of subsets of $X$. \Prf\
For each $F\in\Tau$ let $H_F\subseteq X$ be a $\mu$-conegligible set
such that $f\times\chi F\restr H_F$ is $\Sigma$-measurable. Because
$f$ itself is $\mu$-virtually measurable, $X\in\Tau$. If $F\in\Tau$,
then
\Centerline{$f\times\chi(X\setminus F)\restr(H_X\cap H_F)
=f\restr(H_X\cap H_F)-(f\times\chi F)\restr(H_X\cap H_F)$}
\noindent is $\Sigma$-measurable, while $H_X\cap H_F$ is
$\mu$-conegligible, so $X\setminus F\in\Tau$. If $\sequencen{F_n}$ is
a sequence in $\Tau$ with union $F$, set
$H=\bigcap_{n\in\Bbb N}H_{F_n}$;
then $H$ is conegligible, $f\times\chi F_n\restr H$ is $\Sigma$-measurable
for every $n\in\Bbb N$, and
$f\times\chi F=\sup_{n\in\Bbb N}f\times\chi F_n$, so
$f\times\chi F\restr H$ is $\Sigma$-measurable, and $F\in\Tau$. Thus
$\Tau$ is a $\sigma$-algebra. If $F\in\Sigma$, then
$f\times\chi F\restr H_X$ is $\Sigma$-measurable, so $F\in\Tau$.\ \Qed
\medskip
{\bf (b)} Next, $\nu$ is a measure. \Prf\ Of course
$\nu F\in[0,\infty]$
for every $F\in\Tau$. $f\times\chi\emptyset=0$ wherever it is defined,
so $\nu\emptyset=0$. If $\sequencen{F_n}$ is a disjoint sequence in
$\Tau$ with union $F$, then
$f\times\chi F=\sum_{n=0}^{\infty}f\times\chi F_n$.
If $\nu F_m=\infty$ for some $m$, then we surely have
$\nu F=\infty=\sum_{n=0}^{\infty}\nu F_n$. If $\nu F_m<\infty$ for
each $m$ but $\sum_{n=0}^{\infty}\nu F_n=\infty$, then
\Centerline{$\int f\times\chi(\bigcup_{n\le m}F_n)
=\sum_{n=0}^m\int f\times\chi F_n\to\infty$}
\noindent as $m\to\infty$, so again
$\nu F=\infty=\sum_{n=0}^{\infty}\nu F_n$. If
$\sum_{n=0}^{\infty}\nu F_n<\infty$ then by B.Levi's theorem
\Centerline{$\nu F=\int\sum_{n=0}^{\infty}f\times\chi F_n
=\sum_{n=0}^{\infty}\int f\times\chi F_n
=\sum_{n=0}^{\infty}\nu F_n$.
\Qed}
\medskip
{\bf (c)} Finally, $\nu$ is complete. \Prf\ If $A\subseteq F\in\Tau$
and $\nu F=0$, then $f\times\chi F=0$ a.e., so $f\times\chi A=0$ a.e.\
and $\nu A$ is defined and equal to zero.\ \Qed
}%end of proof of 234I
\leader{234J}{Definition}\dvAformerly{2{}34B}
Let $(X,\Sigma,\mu)$ be a measure space, and
$\nu$ another measure on $X$ with domain $\Tau$. I will call $\nu$ an
{\bf indefinite-integral measure} over $\mu$, or sometimes a {\bf
completed indefinite-integral measure}, if it can be obtained by the
method of 234I
from some non-negative virtually measurable function $f$ defined almost
everywhere on $X$. In this case, $f$ is a Radon-Nikod\'ym derivative
of $\nu$ with respect to $\mu$ in the sense of 232Hf.
\cmmnt{As in 232Hf, the phrase {\bf density function} is also used in
this context.}
\leader{234K}{Remarks}\dvAformerly{2{}34C}
Let $(X,\Sigma,\mu)$ be a measure space, and $f$
a $\mu$-virtually measurable non-negative real-valued function defined
almost everywhere
on $X$; let $\nu$ be the associated indefinite-integral measure.
\spheader 234Ka\cmmnt{ There is a $\Sigma$-measurable function
$g:X\to\coint{0,\infty}$ such that $f=g\,\,\mu$-a.e. \Prf\ Let
$H\subseteq\dom f$ be a measurable conegligible set such that
$f\restr H$ is measurable, and set $g(x)=f(x)$ for $x\in H$, $g(x)=0$
for $x\in X\setminus H$.\ \QeD\ In
this case, $\int f\times\chi E\,d\mu=\int g\times\chi E\,d\mu$ if either
is defined. So $g$ is a Radon-Nikod\'ym derivative of $\nu$, and}
$\nu$ has a
Radon-Nikod\'ym derivative which is $\Sigma$-measurable and defined
everywhere.
%includes old 2{}34Ce
\spheader 234Kb If $E$ is $\mu$-negligible,
then\cmmnt{ $f\times\chi E=0\,\,\mu$-a.e., so} $\nu E=0$.
\cmmnt{Many authors are prepared to say `$\nu$ is
absolutely continuous with respect to $\mu$' in this context. But if
$\nu$ is not totally finite, it need not be
absolutely continuous in the
$\epsilon$-$\delta$ sense of 232Aa (234Xh), and
further difficulties can arise if $\mu$ or $\nu$ is not $\sigma$-finite
(see 234Yk, 234Ym).
}%end of comment
\cmmnt{\spheader 234Kc I have defined `indefinite-integral measure' in
such a way as to produce a complete measure. In my view this is what
makes best sense in most applications. There are occasions on which it
seems more appropriate to use the measure $\nu_0:\Sigma\to[0,\infty]$
defined by
setting $\nu_0 E=\int_Efd\mu=\int f\times\chi E\,d\mu$ for $E\in\Sigma$.
I suppose I would call this the {\bf uncompleted indefinite-integral
measure} over $\mu$ defined by $f$. ($\nu$ is always the
completion of $\nu_0$; see 234Lb.)
}%end of comment
\spheader 234Kd\cmmnt{ Note the way in which I formulated the
definition of $\nu$: `$\nu E=\int f\times\chi E\,d\mu$ if the integral
is defined',
rather than `$\nu E=\int_Efd\mu$'. The point is that the longer
formula gives a rule for deciding what the domain of $\nu$ must be. Of
course it is the case that} $\nu E=\int_Efd\mu$ for every
$E\in\dom\nu$\cmmnt{ (apply 214F to $f\times\chi E$)}.
\spheader 234Ke\dvAnew{2008}
\cmmnt{Because }$\mu$ and its completion define the
same\cmmnt{ virtually measurable functions, the same null ideals and the same
integrals\cmmnt{ (212Eb, 212F)},
they define the same} indefinite-integral measures.
\leader{234L}{The domain of an indefinite-integral
\dvrocolon{measure}}\dvAformerly{2{}34D}\cmmnt{ It is sometimes useful
to have an explicit description of the domain of a measure constructed
in this way.
\medskip
\noindent}{\bf Proposition} Let $(X,\Sigma,\mu)$ be a measure space, $f$
a non-negative $\mu$-virtually measurable function defined almost
everywhere
in $X$, and $\nu$ the associated indefinite-integral measure. Set
$G=\{x:x\in\dom f$, $f(x)>0\}$, and let $\hat\Sigma$ be the domain of the
completion $\hat\mu$ of $\mu$.
(a) The domain $\Tau$ of $\nu$ is
$\{E:E\subseteq X,\,E\cap G\in\hat\Sigma\}$; \cmmnt{in particular,}
$\Tau\supseteq\hat\Sigma\supseteq\Sigma$.
(b) $\nu$ is the completion of its restriction to $\Sigma$.
(c) A set $A\subseteq X$ is $\nu$-negligible iff $A\cap G$ is
$\mu$-negligible.
(d) In particular, if $\mu$\cmmnt{ itself} is complete, then
$\Tau=\{E:E\subseteq X,\,E\cap G\in\Sigma\}$ and $\nu A=0$ iff
$\mu(A\cap G)=0$.
\proof{{\bf (a)(i)} If $E\in\Tau$, then $f\times\chi E$ is virtually
measurable, so there is a conegligible measurable set $H\subseteq\dom f$
such that $f\times\chi E\restr H$ is measurable. Now $E\cap G\cap
H=\{x:x\in H,\,(f\times\chi E)(x)>0\}$ must belong to $\Sigma$, while
$E\cap G\setminus H$ is negligible, so belongs to $\hat\Sigma$, and
$E\cap G\in\hat\Sigma$.
\medskip
\quad{\bf (ii)} If $E\cap G\in\hat\Sigma$, let $F_1$, $F_2\in\Sigma$ be
such that $F_1\subseteq E\cap G\subseteq F_2$ and $F_2\setminus F_1$ is
negligible. Let $H\subseteq\dom f$ be a conegligible set such that
$f\restr H$ is measurable. Then $H'=H\setminus(F_2\setminus F_1)$ is
conegligible and $f\times\chi E\restr H'=f\times\chi F_1\restr H'$ is
measurable, so $f\times\chi E$ is virtually measurable and $E\in\Tau$.
\medskip
{\bf (b)} Thus the given formula does indeed describe $\Tau$. If
$E\in\Tau$, let $F_1$, $F_2\in\Sigma$ be such that
$F_1\subseteq E\cap G\subseteq F_2$ and $\mu(F_2\setminus F_1)=0$.
Because $G$ itself also
belongs to $\hat\Sigma$, there are $G_1$, $G_2\in\Sigma$ such that
$G_1\subseteq G\subseteq G_2$ and $\mu(G_2\setminus G_1)=0$. Set
$F_2'=F_2\cup(X\setminus G_1)$. Then $F_2'\in\Sigma$ and
$F_1\subseteq E\subseteq F'_2$. But
$(F'_2\setminus F_1)\cap G
\subseteq(G_2\setminus G_1)\cup(F_2\setminus F_1)$ is $\mu$-negligible, so
$\nu(F'_2\setminus F_1)=0$.
This shows that if $\nuprime$ is the completion of $\nu\restr\Sigma$
and $\Tau'$ is its domain, then $\Tau\subseteq\Tau'$. But as
$\nu$ is complete, it surely extends $\nuprime$, so $\nu=\nuprime$, as
claimed.
\medskip
{\bf (c)} Now take any $A\subseteq X$. Because $\nu$ is complete,
$$\eqalign{A\text{ is }\nu\text{-negligible}
&\iff\nu A=0\cr
&\iff\int f\times\chi A\,d\mu=0\cr
&\iff f\times\chi A=0\,\,\mu\text{-a.e.}\cr
&\iff A\cap G\text{ is }\mu\text{-negligible.}\cr}$$
\medskip
{\bf (d)} This is just a restatement of (a) and (c) when $\mu=\hat\mu$.
}%end of proof of 234L
\leader{234M}{Corollary}\dvAformerly{2{}34E}
If $(X,\Sigma,\mu)$ is a complete measure
space and $G\in\Sigma$, then the indefinite-integral measure over $\mu$
defined by $\chi G$ is just the measure $\mu\LLcorner G$ defined by
setting
\Centerline{$(\mu\LLcorner G)(F)=\mu(F\cap G)$ whenever $F\subseteq X$
and $F\cap G\in\Sigma$.}
\proof{ 234Ld.
}%end of proof of 234M
\leader{*234N}{}\cmmnt{ The next two results will not be relied on in
this
volume, but I include them for future reference, and to give an idea of
the scope of these ideas.
\medskip
\noindent}{\bf Proposition}\dvAformerly{2{}34F}
Let $(X,\Sigma,\mu)$ be a measure space, and
$\nu$ an indefinite-integral measure over $\mu$.
(a) If $\mu$ is semi-finite, so is $\nu$.
(b) If $\mu$ is complete and locally determined, so is $\nu$.
(c) If $\mu$ is localizable, so is $\nu$.
(d) If $\mu$ is strictly localizable, so is $\nu$.
(e) If $\mu$ is $\sigma$-finite, so is $\nu$.
(f) If $\mu$ is atomless, so is $\nu$.
\proof{ By 234Ka, we may express $\nu$ as the indefinite integral of a
$\Sigma$-measurable function $f:X\to\coint{0,\infty}$. Let $\Tau$ be
the domain of $\nu$, and $\hat\Sigma$ the domain of the completion
$\hat\mu$ of
$\mu$; set $G=\{x:x\in X,\,f(x)>0\}\in\Sigma$.
\medskip
{\bf (a)} Suppose that $E\in\Tau$ and that $\nu E=\infty$. Then
$E\cap G$ cannot be $\mu$-negligible.
Because $\mu$ is semi-finite, there is a non-negligible $F\in\Sigma$
such that $F\subseteq E\cap G$ and $\mu F<\infty$. Now
$F=\bigcup_{n\in\Bbb N}\{x:x\in F$, $2^{-n}\le f(x)\le n\}$, so there
is an $n\in\Bbb N$ such
that $F'=\{x:x\in F$, $2^{-n}\le f(x)\le n\}$ is non-negligible.
Because $f$ is measurable, $F'\in\Sigma\subseteq\Tau$ and
$2^{-n}\mu F'\le\nu F'\le n\mu F'$. Thus we have found an
$F'\subseteq E$ such
that $0<\nu F'<\infty$. As $E$ is arbitrary, $\nu$ is semi-finite.
\medskip
{\bf (b)} We already know that $\nu$ is complete (234Lb) and semi-finite. Now
suppose that $E\subseteq X$ is such that $E\cap F\in\Tau$, that is,
$E\cap F\cap G\in\Sigma$ (234Ld), whenever $F\in\Tau$ and $\nu F<\infty$. Then
$E\cap G\cap F\in\Sigma$ whenever
$F\in\Sigma$ and $\mu F<\infty$. \Prf\ Set
$F_n=\{x:x\in F\cap G,\,f(x)\le n\}$. Then
$\nu F_n\le n\mu F<\infty$, so $E\cap G\cap F_n\in\Sigma$ for every $n$.
But this means that
$E\cap G\cap F=\bigcup_{n\in\Bbb N}E\cap G\cap F_n\in\Sigma$.\ \QeD\
Because $\mu$ is locally determined, $E\cap G\in\Sigma$ and $E\in\Tau$.
As $E$ is arbitrary, $\nu$ is locally determined.
\medskip
{\bf (c)} Let $\Cal F\subseteq\Tau$ be any set. Set
$\Cal E=\{F\cap G:F\in\Cal F\}$, so that $\Cal E\subseteq\hat\Sigma$. By
212Ga, $\hat\mu$
is localizable, so $\Cal E$ has an essential supremum $H\in\hat\Sigma$.
But now, for any $H'\in\Tau$,
$H'\cup(X\setminus G)=(H'\cap G)\cup(X\setminus G)$ belongs to
$\hat\Sigma$, so
$$\eqalign{\nu(F\setminus H')&=0\text{ for every }F\in\Cal F\cr
&\iff\hat\mu(F\cap G\setminus H')=0\text{ for every }F\in\Cal F\cr
&\iff\hat\mu(E\setminus H')=0\text{ for every }E\in\Cal E\cr
&\iff\hat\mu(E\setminus(H'\cup(X\setminus G)))=0
\text{ for every }E\in\Cal E\cr
&\iff\hat\mu(H\setminus((H'\cup(X\setminus G)))=0\cr
&\iff\hat\mu(H\cap G\setminus H')=0\cr
&\iff\nu(H\setminus H')=0.\cr}$$
\noindent Thus $H$ is also an essential supremum of $\Cal F$ in $\Tau$.
As $\Cal F$ is arbitrary, $\nu$ is localizable.
\medskip
{\bf (d)} Let
$\langle X_i\rangle_{i\in I}$ be a decomposition of $X$ for $\mu$;
then it is also a decomposition for $\hat\mu$ (212Gb).
Set $F_0=X\setminus G$, $F_n=\{x:x\in G,\,n-1<f(x)\le n\}$ for $n\ge 1$.
Then $\langle X_i\cap F_n\rangle_{i\in I,n\in\Bbb N}$ is a decomposition
for $\nu$. \Prf\ (i) $\langle X_i\rangle_{i\in I}$ and
$\langle F_n\rangle_{n\in\Bbb N}$ are partitions of $X$ into members of $\Sigma\subseteq\Tau$, so
$\langle X_i\cap F_n\rangle_{i\in I,n\in\Bbb N}$ also is.
(ii) $\nu(X_i\cap F_0)=0$,
$\nu(X_i\cap F_n)\le n\mu X_i<\infty$ for $i\in I$, $n\ge 1$.
(iii) If $E\subseteq X$ and $E\cap X_i\cap F_n\in\Tau$ for every
$i\in I$ and $n\in\Bbb N$ then
$E\cap X_i\cap G=\bigcup_{n\in\Bbb N}E\cap X_i\cap F_n\cap G$ belongs to
$\hat\Sigma$ for every $i$, so $E\cap G\in\hat\Sigma$ and
$E\in\Tau$.
(iv) If $E\in\Tau$, then of course
\Centerline{$\sum_{i\in I,n\in\Bbb N}\nu(E\cap X_i\cap F_n)
=\sup_{J\subseteq I\times\Bbb N\text{ is finite}}
\sum_{(i,n)\in J}\nu(E\cap X_i\cap F_n)\le\nu E$.}
\noindent So if $\sum_{i\in I,n\in\Bbb N}\nu(E\cap X_i\cap F_n)=\infty$
it is surely equal to $\nu E$. If the sum is finite, then
$K=\{i:i\in I,\,\nu(E\cap X_i)>0\}$ must be countable. But for
$i\in I\setminus K$,
$\int_{E\cap X_i}fd\mu=0$, so $f=0\,\,\mu$-a.e.\ on $E\cap X_i$, that
is, $\hat\mu(E\cap G\cap X_i)=0$. Because
$\langle X_i\rangle_{i\in I}$ is a decomposition for $\hat\mu$,
$\hat\mu(E\cap G\cap\bigcup_{i\in I\setminus K}X_i)=0$ and
$\nu(E\cap\bigcup_{i\in I\setminus K}X_i)=0$. But this
means that
\Centerline{$\nu E=\sum_{i\in K}\nu(E\cap X_i)
=\sum_{i\in K,n\in\Bbb N}\nu(E\cap X_i\cap F_n)
=\sum_{i\in I,n\in\Bbb N}\nu(E\cap X_i\cap F_n)$.}
\noindent As $E$ is arbitrary,
$\langle X_i\cap F_n\rangle_{i\in I,n\in\Bbb N}$ is a decomposition for
$\nu$.\ \QeD\ So $\nu$ is strictly localizable.
\medskip
{\bf (e)} If $\mu$ is $\sigma$-finite, then in (d) we can take $I$ to be
countable, so that $I\times\Bbb N$ also is countable, and $\nu$ will be
$\sigma$-finite.
\medskip
{\bf (f)} If $\mu$ is atomless, so is $\hat\mu$ (212Gd). If $E\in\Tau$
and $\nu E>0$, then $\hat\mu(E\cap G)>0$, so there is an $F\in\hat\Sigma$
such that $F\subseteq E\cap G$ and
neither $F$ nor $E\cap G\setminus F$ is $\hat\mu$-negligible. But in this case both $\nu F=\int_Ffd\mu$ and
$\nu(E\setminus F)=\int_{E\setminus F}fd\mu$ must be greater than $0$
(122Rc). As $E$ is arbitrary, $\nu$ is atomless.
}%end of proof of 234N
\leader{*234O}{}\cmmnt{ For localizable measures, there is a
straightforward description of the associated indefinite-integral
measures.
\medskip
\noindent}{\bf Theorem}\dvAformerly{2{}34G}
Let $(X,\Sigma,\mu)$ be a localizable measure
space. Then a measure $\nu$, with domain $\Tau\supseteq\Sigma$, is an
indefinite-integral measure over $\mu$ iff ($\alpha$) $\nu$ is
semi-finite
and zero on $\mu$-negligible sets ($\beta$) $\nu$ is the completion of
its restriction to $\Sigma$ ($\gamma$) whenever $\nu E>0$ there is an
$F\subseteq E$ such that $F\in\Sigma$, $\mu F<\infty$ and $\nu F>0$.
\proof{{\bf (a)} If $\nu$ is an indefinite-integral measure over $\nu$,
then by 234Na, 234Kb and 234Lb it is semi-finite, zero on
$\mu$-negligible sets and the completion of its restriction to $\Sigma$.
Now suppose that $E\in\Tau$ and $\nu E>0$. Then there is an
$E_0\in\Sigma$
such that $E_0\subseteq E$ and $\nu E_0=\nu E$, by 234Lb. If
$f:X\to\Bbb R$ is a $\Sigma$-measurable
Radon-Nikod\'ym derivative of $\nu$ (234Ka), and $G=\{x:f(x)>0\}$, then
$\mu(G\cap E_0)>0$; because $\mu$ is semi-finite, there is an
$F\in\Sigma$ such that
$F\subseteq G\cap E_0$ and $0<\mu F<\infty$, in which case $\nu F>0$.
\medskip
{\bf (b)} So now suppose that $\nu$ satisfies the conditions.
\medskip
\quad{\bf (i)} Set $\Cal E=\{E:E\in\Sigma,\,\nu E<\infty\}$. For
each $E\in\Cal E$, consider $\nu_E:\Sigma\to\Bbb R$, setting
$\nu_EG=\nu(G\cap E)$ for every $G\in\Sigma$. Then $\nu_E$ is
countably additive and truly continuous with respect to $\mu$. \Prf\
$\nu_E$ is countably additive, just as in 231De. Because $\nu$ is zero
on $\mu$-negligible sets, $\nu_E$ must
be absolutely continuous with respect to $\mu$, by 232Ba. Since
$\nu_E$ clearly satisfies condition ($\gamma$) of 232Bb,
it must be truly continuous.\ \Qed
By 232E, there is a $\mu$-integrable function $f_E$ such that
$\nu_EG=\int_Gf_Ed\mu$ for every $G\in\Sigma$, and we may suppose that
$f_E$ is $\Sigma$-measurable (232He). Because $\nu_E$ is non-negative, $f_E\ge
0\,\,\mu$-almost everywhere.
\medskip
\quad{\bf (ii)} If $E$, $F\in\Cal E$ then $f_E=f_F\,\,\mu$-a.e. on
$E\cap F$, because
\Centerline{$\int_Gf_Ed\mu=\nu G=\int_Gf_Fd\mu$}
\noindent whenever $G\in\Sigma$ and $G\subseteq E\cap F$. Because
$(X,\Sigma,\mu)$ is localizable, there is a measurable $f:X\to\Bbb R$
such that $f_E=f\,\,\mu$-a.e.\ on $E$ for every $E\in\Cal E$ (213N).
Because every $f_E$ is non-negative almost everywhere, we may suppose
that
$f$ is non-negative, since surely $f_E=f\vee\tbf{0}\,\,\mu$-a.e. on $E$
for
every $E\in\Cal E$.
\medskip
\quad{\bf (iii)} Let $\nuprime$ be the indefinite-integral measure
defined by $f$. If $E\in\Cal E$ then
\Centerline{$\nu E=\int_Ef_Ed\mu=\int_Efd\mu=\nuprime E$.}
\noindent For $E\in\Sigma\setminus\Cal E$, we have
\Centerline{$\nuprime E
\ge\sup\{\nuprime F:F\in\Cal E,\,F\subseteq E\}
=\sup\{\nu F:F\in\Cal E,\,F\subseteq E\}=\nu E=\infty$}
\noindent because $\nu$ is semi-finite. Thus $\nuprime$ and $\nu$
agree on $\Sigma$. But since each is the completion of its restriction
to $\Sigma$, they must be equal.
}%end of proof of 234O
\leader{234P}{Ordering \dvrocolon{measures}}\cmmnt{ There are many ways in
which one measure can dominate another. Here I will describe one of the
simplest.
\medskip
\noindent}{\bf Definition}\dvAnew{2008}
Let $\mu$, $\nu$ be two measures on a set $X$.
I will say that $\mu\le\nu$ if $\mu E$ is defined, and $\mu E\le\nu E$,
whenever $\nu$ measures $E$.
\leader{234Q}{Proposition}\dvAnew{2008}
Let $X$ be a set, and write $\Mu$ for the set of
all measures on $X$.
(a) Defining $\le$ as in 234P, $(\Mu,\le)$ is a partially ordered set.
(b) If $\mu$, $\nu\in\Mu$, then $\mu\le\nu$ iff there is a $\lambda\in\Mu$
such that $\mu+\lambda=\nu$.
(c) If $\mu\le\nu$ in $\Mu$ and $f$ is a $[-\infty,\infty]$-valued
function,
defined on a subset of $X$, such that $\int fd\nu$ is defined in
$[-\infty,\infty]$, then $\int fd\mu$ is defined; if $f$ is non-negative,
$\int fd\mu\le\int fd\nu$.
\proof{{\bf (a)} Of course $\mu\le\mu$ for every $\mu\in\Mu$.
If $\mu\le\nu$ and $\nu\le\lambda$ in $\Mu$, then
$\dom\lambda\subseteq\dom\nu\subseteq\dom\mu$, and
$\mu E\le\nu E\le\lambda E$ whenever $\lambda$ measures $E$.
If $\mu\le\nu$ and $\nu\le\mu$ then
$\dom\mu\subseteq\dom\nu\subseteq\dom\mu$ and $\mu E\le\nu E\le\mu E$ for
every $E$ in their common domain, so $\mu=\nu$.
\medskip
{\bf (b)(i)}
If $\mu+\lambda=\nu$, then the definitions in 234G and 234P make
it plain that $\mu\le\nu$.
\medskip
\quad{\bf (ii)}\grheada\ In the reverse direction, if $\mu\le\nu$,
write $\Tau$ for the domain of $\nu$.
Define $\lambda:\Tau\to[0,\infty]$ by setting
\Centerline{$\lambda G
=\sup\{\nu F-\mu F:F\in\Tau,\,F\subseteq G,\,\mu F<\infty\}$}
\noindent for $G\in\Tau$.
Then $\lambda\in\Mu$. \Prf\ Of course $\dom\lambda=\Tau$ is a
$\sigma$-algebra, and $\lambda\emptyset=0$. Suppose that
$\sequencen{G_n}$ is a
disjoint sequence in $\Tau$ with union $G$.
If $F\in\Tau$, $F\subseteq G$ and $\mu F<\infty$, then
$$\eqalign{\nu F-\mu F
&=\sum_{n=0}^{\infty}\nu(F\cap G_n)-\sum_{n=0}^{\infty}\mu(F\cap G_n)\cr
&=\sum_{n=0}^{\infty}\nu(F\cap G_n)-\mu(F\cap G_n)
\le\sum_{n=0}^{\infty}\lambda G_n;\cr}$$
\noindent as $F$ is arbitrary,
$\lambda G\le\sum_{n=0}^{\infty}\lambda G_n$. If
$\gamma<\sum_{n=0}^{\infty}\lambda G_n$, there are an $m\in\Bbb N$ such
that $\gamma<\sum_{n=0}^m\lambda G_n$, and $F_0,\ldots,F_m$ such that
$F_n\in\Tau$, $F_n\subseteq G_n$ and $\mu F_n<\infty$ for every $n\le m$,
while $\sum_{n=0}^m\nu F_n-\mu F_n\ge\gamma$. Set
$F=\bigcup_{n\le m}F_n$; then $F\in\Tau$, $F\subseteq G$ and
$\mu F<\infty$, so
\Centerline{$\lambda G\ge\nu F-\mu F
=\sum_{n=0}^m\nu F_n-\mu F_n\ge\gamma$.}
\noindent As $\gamma$ is arbitrary,
$\lambda G\ge\sum_{n=0}^{\infty}\lambda G_n$; as $\sequencen{G_n}$ is
arbitrary, $\lambda$ is countably additive.\ \Qed
\medskip
\qquad\grheadb\ Now $\mu+\lambda=\nu$. \Prf\ The domain of $\mu+\lambda$
is $\dom\mu\cap\dom\lambda=\Tau=\dom\nu$. Take $G\in\Tau$. If
$\mu G=\infty$, then $\nu G=\infty=(\mu+\lambda)G$. Otherwise,
\Centerline{$(\mu+\lambda)G\ge\mu G+\nu G-\mu G=\nu G$.}
\noindent So if $\nu G=\infty$ we shall certainly have
$\nu G=(\mu+\lambda)G$. Finally, if $\nu G<\infty$ then
$$\eqalign{(\mu+\lambda)G
&=\mu G+\sup\{\nu F-\mu F:F\in\Tau,\,F\subseteq G\}\cr
&=\sup\{\nu F+\mu(G\setminus F):F\in\Tau,\,F\subseteq G\}\cr
&\le\sup\{\nu F+\nu(G\setminus F):F\in\Tau,\,F\subseteq G\}
=\nu G,\cr}$$
\noindent so again we have equality.\ \Qed
Thus we have an appropriate expression of $\nu$ as a sum of measures.
\medskip
{\bf (c)(i)} If $f$ is non-negative, put (b) and 234Hc together.
\medskip
\quad{\bf (ii)} In general, if $\int fd\nu$ is defined, so are both
$\int f^+d\nu$ and $\int f^-d\nu$, and at most one is infinite;
so $\int f^+d\mu$ and $\int f^-d\mu$ are defined
and at most one is infinite.
}%end of proof of 234Q
\exercises{
\leader{234X}{Basic exercises (a)}
%\spheader 234Xa
\dvAformerly{1{}32Xk, 2{}14Xp}