NOTES ON N.

BOURBAKI’S

Integration

Notes by S. K. Berberian
Notes on text through Ch.V, §5, No. 4 (10-19-08)
 Sterling K. Berberian, Prof. Emer.
The University of Texas at Austin
(e-mail: [email protected])

Mathematics Subject Classification (2000): 28-01, 28Bxx, 43-xx, 46Exx

 Foreword

When I first studied the original French fascicles of N. Bourbaki’s Inté-

gration (in the 1970’s) I made a handwritten working translation. Whenever
I reached an assertion that I didn’t “see” after a few seconds of reflection
(which was often), I opened a pair of braces { } and worked out an explana-
tion to myself before proceeding further. This document is a transcription
of those explanations, indexed by the page and line numbers of the text in
the published translation to which they pertain (N. Bourbaki, Integration,
Vols. I, II, Springer–Verlag, 2004). On restudying the work while preparing
the new translation (starting in 1998), I often consulted the notes, occa-
sionally finding my explanations inadequate; the notes in their present form
aspire to repair the inadequacies.
All of the above is rather personal; what justification can there be for
putting the notes out in front of everybody? Some of the gaps I thought I
saw proved to be trivial (after I saw the light); others took several days to
work out—but there might have been shortcuts that I didn’t find. My mo-
tivation was high; I was determined to acquire the prerequisites for reading
the author’s Théories spectrales and J. Dixmier’s books on operator alge-
bras, so I ground at the obstacle until it yielded. In answer to the question
posed above: My hope is that the reader of Integration, who comes upon
a sticking point and does not have the leisure to grind it away, will find in
these notes an explanation that will enable him to keep going without loss
of momentum.
Paragraph 3 of the mode d’emploi (“To the Reader”) makes it plain that
the typical reader of Integration comes to the task well-equipped. We can
infer that the “gaps” are there on purpose; the missing details were doubtless
present in an initial draft of a proof, but (if we may be permitted to read
the author’s mind) they have been deliberately pruned away to highlight
the main outline of the proof while leaving enough details to enable the
determined reader to fill in the gaps. This is marvelous exercise for the
research muscles, surely one of the author’s didactic aims. That is good
news; the reader can be reassured that the gaps are well thought out and
the chances good that a moderate amount of effort will see him through and
render these notes superfluous.
 Errata for Integration (INT)

III.52, `. −3. For I read L .

IV.17, `. 2. After “belong to A ,” insert “are 6 f ,”.
IV.30, `. −2. Delete inverted comma at the end of the line.
IV.40, `. −15. For “step function” as translation of fonction en escalier,
see the Note for this line (on p. IV.x26 below) and especially the Note for
IV.66, `. −15, −14 (on p. IV.x88).
IV.49, Footnote. The period at the end should immediately follow the
parenthesis.
IV.51, Footnote. Vis-a-vis the translation of fonction étagée, see the
Notes mentioned above in the entry for IV.40, `. −15.
IV.73, `. −8. For “Cor.” read “Cor. 1”.
IV.84, `. −13. For |y − z| 6 δ read |y − z| < δ . (See the Note for this
line.)
IV.88, Running head. For §6 read §5.
IV.108, `. 20. For TVS read EVT. {See the Note for `. 20–24 on
p. IV.x248 below.}
IV.113, `. 1. The argument needed for (16 n ) is more complicated. {See
the Note for this line on p. IV.x265 below.}
IV.115, `. −5. In the formula (19), for f (x) read f(x) .
IV.117, `. 21. For h(x) = 1 read, say, h(z) = 1 . (The letter x is
reserved for an element of M in the statement of the Corollary, and y occurs
later in the proof as another element of M .)
V.12, `. −9. For “No. 3” read “No. 4”.
V.46, `. −10. For (gα ) read (gn ) .
V.49, `. 9. For “Lemma 1” read “Lemma 2”.
V.77, `. 1 and −9. Lines incompletely printed (‘washed out’).
VI.10, `. 9. For “ q is gauge” read “ q its gauge”.
VI.14, `. −9. For “subset” read “subsets”.
 Extended “asides”

III.16, `. 13–15. The discrete measure on X = N with unit mass at

every point. {III.x13–x18}
IV.66, `. −15, −14. “Fonction étagée” vs. “fonction en escalier” (“step
function” vs. “interval step function”). {IV.x88, x89}. See also IV.40, `. −15.
{IV.x26, x27}
IV.79, `. 3, 4. Conditions c 0 ) and c 00 ) for No. 10, Prop. 15. {IV.x127}
IV.79, `. 7. Condition d 0 ). {IV.x129}
IV.80, `. −17 to −14. Measurability of functions defined on a measur-
able subset. {IV.x133–x146}
IV.85, `. 19–27 and the “echo” that follows it. Measure as a set func-
tion, ‘comparative anatomy’ of the Bourbaki and Halmos formulations of
measure on a locally compact space. {IV.x170–x175 and IV.x176–179}
R
V.10, `. 13, 14. ‘Bilinearity’ of (f , µ) 7→ f dµ . {V.x15–x21, especially
the Theorem (V.x17) and its Corollary 3 (V.x19)}
V.17, `. −6 to −4. Solution of Exer. 7 for Ch. V, §3. {V.x33–x36}
V.17, `. −4. Solution of Exer. 8 for Ch. V, §3. {V.x38, x39}
V.25, `. 8–11. The analog of “Lebesgue’s theorem” (Ch. IV, §3, No. 7,
Th. 6) for essentially integrable functions is worked out in this Note. {V.x68,
Remark}
R
V.27, `. −14, −13. Revisiting the notation µ = εx dµ(x) of Ch. III,
§3, No. 1, Example 2, in the context of diffusions. {V.x73–x75, An example}
V.29, `. −11 to −9. A handy ‘scholium’ for bounded diffusions. {V.x80}
V.30, `. 6–28. Composition of diffusions when viewed as mappings
µ 7→ µΛ ( µ ∈ domain of Λ ), and the inclusion Λ ◦ H ⊂ ΛH . {V.x82–x85}
V.38, `. −10 to −3 and V.39, `. 22–24. Adjustments to the definition
of local integrability. {V.x106–x109 and V.x110, Proposition}
V.39, `. 26, 27 and V.39, `. −12 to −10. Review of the locally convex
topology generated by a set of semi-norms. {V.x111–x113 and V.x113, x114}
V.46, `. 16, 17. A useful criterion for measurability of a numerical
function ( f is measurable ⇔ f ϕK is measurable for every compact set K ).
{V.x138, Lemma}
V.47, `. 14–16. If f and g are measurable numerical functions > 0 ,
then fg is measurable. {V.x145, Lemma}
 Loose ends: some questions posed in the Notes

III.23, `. 16–19. A question about a direct limit of induced topologies.

{III.x28, `. −6 to −4}.
III.54, `. 11–13. Wanted: A simpler proof of the associative law for
the product of measures. {III.x135, `. 8–10}
IV.85, `. 19–27. Is ν = ρ always? {IV.x170, `. −2, −1}
V.32, `. 18–20. A question about the definition of vague measurability
of mappings Λ : t 7→ λt . {V.x90, `. −8, −7}
V.45, `. 10, 11. A question about moderated measures posed here is
answered in an addendum to the Note for V.48, `. 2, 3.
 CHAPTER I

Inequalities of convexity

I.2, `. 15–16.
“. . .therefore there exists a λ0 > 0 such that the relation λ > λ0
implies λt ∈ K ;”
For example, set λ0 = ϕ(t1 , t2 , . . . , tn )−1 and cite the positive homo-
geneity of ϕ : if λ > λ0 , then ϕ(λt1 , . . . , λtn ) = λλ−1
0 > 1 , whence λt ∈ K .

I.2, `. 17.
“. . .the relations ti > 0 (1 6 i 6 n) imply αι1 t1 + · · · + αιn tn > 0 ”
Suppose first that ti > 0 for all i , and choose λ0 as above. Then, for
all λ > λ0 , λt ∈ K and therefore, for each ι ,

αι1 (λt1 ) + · · · + αιn (λtn ) > βι ,

that is,
αι1 t1 + · · · + αιn tn > λ−1 βι ,
and letting λ → +∞ yields αι1 t1 + · · · + αιn tn > 0 .
If now t = (ti ) with ti > 0 for i = 1, . . . , n , then, for each ι , the
foregoing yields αι1 (t1 + ε) + · · · + αιn (tn + ε) > 0 for every ε > 0 , whence
αι1 t1 + · · · + αιn tn > 0 .
I.2, `. 18–19.
“. . .it is then clear that K is also the intersection of the half-spaces
ti > 0 (1 6 i 6 n) and the Uι such that βι > 0 ;”
If βι < 0 , the inequality (2) is trivially satisfied when the t i are > 0 .
I.2, `. 20.
“. . .since the origin does not belong to K ”
Writing 0 = (0, . . . , 0) , ϕ(0) = ϕ(2 · 0) = 2ϕ(0) implies ϕ(0) = 0 by
finiteness, whence 0 ∈/ K.
INT I.x2 inequalities of convexity Chap. 1

I.2, `. 20–21.
“. . .there exists at least one index ι such that β ι > 0 .”
If one had βι = 0 for all ι , then 0 = (0, . . . , 0) would satisfy all of
the inequalities defining K , that is, it would belong to all of the half-spaces
whose intersection is K , contrary to 0 ∈ / K.
I.2, `. −11 to −9.
“Now let C be the convex cone in Rn+1 defined by the relations ti > 0
(1 6 i 6 n + 1) , tn+1 6 ϕ(t1 , t2 , . . . , tn ) (the closure of the convex cone
generated in Rn+1 by the convex set K × {1} ) ”
It is clear from the continuity of ϕ that C is closed. Let D be the
convex cone generated by the convex set K × {1} , namely,
[
D= α(K × {1})
α>0
= {(αt1 , . . . , αtn , α) : α > 0 & (t1 , . . . , tn ) ∈ K } .

We are to show that C = D .

If α > 0 and (t1 , . . . , tn ) ∈ K then

ϕ(αt1 , . . . , αtn ) = αϕ(t1 , . . . , tn ) > α · 1 = α,

whence (αt1 , . . . , αtn , α) ∈ C . Thus D ⊂ C and, since C is closed, D ⊂ C .

To prove the reverse inclusion C ⊂ D , suppose (t 1 , . . . , tn , tn+1 ) ∈ C ,
that is, ti > 0 for 1 6 i 6 n + 1 and tn+1 6 ϕ(t1 , . . . , tn ) .
case 1: tn+1 > 0 .
Then 1 6 ϕ(t−1 −1 −1
n+1 · t1 , . . . , tn+1 · tn ), thus tn+1 (t1 , . . . , tn ) ∈ K ,
(t1 , . . . , tn ) ∈ tn+1 K , and so (t1 , . . . , tn , tn+1 ) ∈ tn+1 (K × {1}) ⊂ D ⊂ D .
case 2: tn+1 = 0 .
Thus 0 = tn+1 6 ϕ(t1 , . . . , tn ) . We consider two sub-cases.
If ϕ(t1 , . . . , tn ) > 0 , then for γ0 = ϕ(t1 , . . . , tn )−1 we have

γ0 ϕ(t1 , . . . , tn ) = 1 ,

whence γ ϕ(t1 , . . . , tn ) > 1 for all γ > γ0 ; thus, for all γ > γ0 ,
we have ϕ(γt1 , . . . , γtn ) > 1 , so that (γt1 , . . . , γtn ) ∈ K and therefore
(γtn , . . . , γtn , 1) ∈ K × 1 , whence (t1 , . . . , tn , γ −1 ) ∈ γ −1 (K × {1}) ⊂ D ;
letting γ → +∞ one sees that (t1 , . . . , tn , 0) ∈ D .
Finally, suppose ϕ(t1 , . . . , tn ) = 0 . Let t0i > ti for i = 1, . . . , n . In
particular, t0i > 0 for i = 1, . . . , n , hence ϕ(t01 , . . . , t0n ) > 0 = tn+1 ,
thus (t01 , . . . , t0n , 0) ∈ D by the preceding sub-case; letting t 0i → ti yields
(t1 , . . . , tn , 0) ∈ D as desired.
Chap. 1 inequalities of convexity INT I.x3

I.2, `. −9 to −7.
“. . .it is immediate that C is also defined by the relations t i > 0
(1 6 i 6 n + 1) and

(3) βι tn+1 6 αι1 t1 + · · · + αιn tn (ι ∈ I, βι > 0) .”

Write E for the set of (t1 , . . . , tn , tn+1 ) defined by these relations. It

is clear that E is a closed convex cone; we are to show that E = C , that is,
that E = D (where D is defined as in the preceding remarks).
To show that D ⊂ E , it will suffice to show that K × {1} ⊂ E (because
D is the closed convex cone generated by K × {1} ). If (t 1 , . . . , tn ) ∈ K
then ti > 0 for i = 1, . . . , n and by (2) we have βι 6 αι1 t1 + · · · + αιn tn
for all ι ∈ I , thus (t1 , . . . , tn , 1) satisfies (3). Thus K × {1} ⊂ E , and so
D ⊂ E.
Conversely, suppose t1 , . . . , tn , tn+1 are > 0 and satisfy (3); we are
to show that (t1 , . . . , tn , tn+1 ) ∈ D . We consider two cases, according as
tn+1 > 0 or tn+1 = 0 :
If tn+1 > 0 then (3) yields

1
1

βι 6 αι1 · t1 + · · · + αιn · tn for all ι ∈ I ,
tn+1 tn+1

1
thus (t1 , . . . , tn ) ∈ K , so that (t1 , . . . , tn ) ∈ tn+1 K , whence
tn+1

(t1 , . . . , tn , tn+1 ) ∈ (tn+1 K × {tn+1 }) = tn+1 (K × {1}) ⊂ D .

If tn+1 = 0 then, since ϕ(t1 , . . . , tn ) > 0 by property 1◦ and the

continuity of ϕ , we have tn+1 = 0 6 ϕ(t1 , . . . , tn ) , and the argument used
in “case 2” of the preceding remarks yields (t 1 , . . . , tn , 0) ∈ D .
I.2, `. −6 to −4.
“For every x ∈ X , we therefore have


(4) βι ϕ f1 (x), . . . , fn (x) 6 αι1 f1 (x) + · · · + αιn fn (x)

for all ι ∈ I .”
Set ti = fi (x) for i = 1, . . . , n and set tn+1 = ϕ(t1 , . . . , tn ) . Obviously

ti > 0 (1 6 i 6 n) and tn+1 6 ϕ(t1 , . . . , tn ) ,

thus (t1 , . . . , tn , tn+1 ) ∈ C , where C is the convex cone defined earlier in

the proof. But C = D = E (where D and E are defined as in the comments
INT I.x4 inequalities of convexity Chap. 1

on `. −10, −9, −8), thus (t1 , . . . , tn , tn+1 ) satisfies (3), which is precisely the
assertion (4).
I.2, `. −3, −2.
“ M (ϕ(f1 , f2 , . . . , fn )) is finite and


βι M ϕ(f1 , f2 , . . . , fn ) 6 αι1 M (f1 ) + αι2 M (f2 ) + · · · + αιn M (fn ) ”

We are assuming βι > 0 (such an ι exists, as noted earlier in

the proof). Write f =
ϕ(f1 , . . . , fn ) for the function on X defined by
x 7→ ϕ f1 (x1 ), . . . , fn (x) . Since (4) holds for all x ∈ X , we have

βι f 6 αι1 f1 + · · · + αιn fn ,

and since βι > 0 we have, by the properties of M ,

βι M (f ) = M (βι f ) 6 M (αι1 f1 + · · · + αιn fn ) 6 αι1 M (f1 ) + · · · + αιn M (fn ) ,

whence M (f ) < +∞ .
I.3, Proposition 2.
Note that the condition α + β = 1 assures that the function ϕ(t 1 , t2 ) =
α β
t1 t2 is positively homogeneous.
I.4, `. 10, 11.
“. . .if g is a function belonging to F p (X, M ) and if |f | 6 |g| , then
f also belongs to F p (X, M ) ;”
Since |f |p 6 |g|p , one has M (|f |p ) 6 M (|g|p ) , therefore M (|f |p )1/p 6
M (|g|p )1/p < +∞ , thus f ∈ F (X, M ) .
I.4, `. 12.
“. . .the sum of two functions in F p (X, M ) also belongs to this set;”

1/p
For any finite numerical function f on X , write N p (f ) = M (|f |p ) .
By Minkowski’s inequality (Prop. 3), for any numerical functions f, g on X
one has Np (|f | + |g|) 6 Np (f ) + Np (g) , and it follows from |f + g| 6 |f | + |g|
that
Np (f + g) 6 Np (|f | + |g|) 6 Np (f ) + Np (g) .
Therefore if f and g belong to F p (X, M ) , that is, if Np (f ) and Np (g)
are finite, then Np (f + g) is also finite, thus f + g ∈ F p (X, M ) .
I.5, `. 7.
“. . .the corollary is proved.”
Explicitly, if 0 < r < p one sees that N r (f ) 6 Np (f ) by choosing q so
1 1 1
that = − and citing (9) with g = 1 .
q r p
Chap. 1 inequalities of convexity INT I.x5

I.5, `. 8–12.
Proposition 5. — For every finite numerical function f defined on X ,
the set I of values of 1/p (p > 0) such that N p (f ) is finite is either
empty or is an interval; if I is not reduced to a point, then the mapping
α 7→ log N1/α (f ) is either convex on I or is equal to −∞ on the interior
of I .
Rearrangement of the proof. Fix a finite numerical function f on X and
define J = {p > 0 : Np (f ) < +∞ }; thus, in the notation of the statement
of the proposition, I = {1/p : p ∈ J } .
claim 1: Either J = ∅ or J is an interval (possibly degenerate, i.e.,
reduced to a point).
In other words, the assertion is that J is a convex subset of ]0, +∞[ .
Suppose r, s ∈ J and 0 < α < 1 ; writing p = αr + (1 − α)s , we are
to show that p ∈ J , i.e., that M (|f |p ) < +∞ . Since r, s ∈ J we have
M (|f |r ) < +∞ and M (|f |s ) < +∞ ; by Hölder’s inequality (Prop. 2),


α
1−α
M |f |r )α (|f |s )1−α 6 M (|f |r ) M (|f |s ) ,

that is,

α
1−α
M (|f |p ) 6 M (|f |r ) M (|f |s ) ,
which can also be written

p
rα
s(1−α)
(∗) Np (f ) 6 Nr (f ) Ns (f ) ,

whence Np (f ) < +∞ . It follows that either I = ∅ or I is an interval

(possibly degenerate).
claim 2: If J is a nondegenerate interval and if N r (f ) = 0 for some
interior point r of J , then Np (f ) = 0 for every interior point p of J .
Suppose, for example, that p > r . Choose s ∈ J such that r < p < s
s−p
and write p = λr + (1 − λ)s with 0 < λ < 1 (namely, λ = ). Since
s−r
Nr (f ) = 0 and Ns (f ) is finite, it is clear from (*) that N p (f ) = 0 .
Assume henceforth that J (hence also I ) is a nondegenerate interval,
◦
and that there exists a p ∈ J with Np (f ) > 0 (hence, by the foregoing, that
◦
Np (f ) > 0 for all interior points p of J ). Define ϕ : I → R as follows: if
◦ ◦
α ∈ I then 1/α ∈ J , hence 0 < N1/α (f ) < +∞ ; we define
◦
ϕ(α) = log N1/α (f ) (α ∈ I) .

The proof will be completed by showing:

INT I.x6 inequalities of convexity Chap. 1

claim 3: ϕ is convex.
◦ ◦
Suppose r, s ∈ J ; thus 1/r, 1/s are typical elements of I . Given
0 < t < 1 , define p > 0 by the formula
1 1 1
(†) = t · + (1 − t) · ,
p r s
◦
so that 1 = tp/r + (1 − t)p/s ; thus 1/p ∈ I and the problem is to show that
1
1
1

ϕ 6 tϕ + (1 − t)ϕ ,
p r s
i.e., that
log Np (f ) 6 t log Nr (f ) + (1 − t) log Ns (f ) ,
equivalenty, that

t
1−t
Np (f ) 6 Nr (f ) Ns (f ) ,
i.e., that

1/p
t/r
(1−t)/s
M (|f |p 6 M (|f |r M |f |s ,

tp/r
(1−t)p/s
i.e., that M (|f |p ) 6 M (|f |r M (|f |s .
Setting λ = tp/r , we have, by (†),

tp (1 − t)p
1−λ=1− = > 0,
r s
thus 0 < λ < 1 and our problem is to show that

λ
1−λ
(∗∗) M (|f |p ) 6 M (|f |r M (|f |s .

Note that
rλ + s(1 − λ) = tp + (1 − t)p = p ;
since M (|f |r ) < +∞ and M (|f |s ) < +∞ , it follows from Hölder’s inequal-
ity (Prop. 2) that


M (|f |p ) = M (|f |rλ+s(1−λ) ) = M (|f |r )λ · (|f |s )1−λ

λ
1−λ
6 M (|f |r ) M (|f |s ,

which is the desired relation (**).

{Note: In verifying (**), r and/or s can be endpoints of J ; it is for
taking log that one restricts to the interior.}
 CHAPTER II

Riesz spaces

§1. RIESZ SPACES AND FULLY LATTICE-ORDERED SPACES

II.2, `. −2, −1.

“...whence, in particular,

(6) sup(x, y) = x + (y − x)+ = 12 (x + y + |x − y|) .”

Setting z = −x in (5) yields sup(0, y − x) = −x + sup(x, y) , that is,

sup(x, y) = x + sup(0, y − x) = x + (y − x)+ . But

1 1
(y − x)+ = (|y − x| + y − x) = (|x − y| + y − x) ,
2 2
1
thus x + (y − x)+ = (2x + |x − y| + y − x) , whence (6).
2
II.3, `. 6–8.
“If A and B are two subsets of E each of which has a supremum, then
A + B also has a supremum and

(10) sup(A + B) = sup A + sup B .”

Let a = sup A , b = sup B . For all x ∈ A and y ∈ B , one has x 6 a

and y 6 b , hence x + y 6 a + b , thus A + B admits a + b as an upper
bound.
Assuming c is any upper bound for A + B , we need only show that
a + b 6 c . If y ∈ B is fixed then, for every x ∈ A , x + y 6 c , x 6 c − y ,
whence a 6 c − y . Thus y 6 c − a for all y ∈ B , whence b 6 c − a .
II.7, `. 6–7.
“However, if A is infinite, a subset of B(A) may be bounded above
in RA without being bounded above in B(A) ”
For example, let A be the set of positive integers, x the function on A
such that x(k) = k for all k , and, for each positive integer n , let x n be
the function on A such that xn (n) = n and xn (k) = 0 for all k 6= n . Then
xn 6 x for all n , but the sequence (xn ) has no bounded majorant.
INT II.x2 riesz spaces §2

II.8, `. −7.
“Corollary. — Let a be an element of a fully lattice-ordered space E ,
Ba the band generated by a , B0a the band of elements alien to a . For every
element x > 0 of E , the component of x in B a (for the decomposition

of E as the ordered direct sum of Ba and B0a ) is equal to sup inf(n|a|, x) .
n∈N
This follows from Proposition 6, applied to M = {a} , and Proposi-
tion 5.”
Adopt the notations of Prop. 6 and its proof. Thus (when M = {a} )
M1 = {t : 0 6 t 6 n|a| for some n ∈ N } ;
then M1 has the properties of the set A of Prop. 5, and M 2 is the set M
of Prop. 5. Let x ∈ E , x > 0 . By Prop. 5, one can write x = y + z with
y > 0 , z > 0 and
y = sup{v : v ∈ M1 and v 6 x }
= sup{v : 0 6 v 6 n|a| for some n ∈ N and v 6 x }
= sup{v : 0 6 v 6 inf(n|a|, x) for some n ∈ N }
= sup{v : v = inf(n|a|, x) for some n ∈ N }
= sup inf(n|a|, x) ,
n∈N

and where z ∈ M02 = M0 = M000 = B0a . Thus, as noted in the proof of Prop. 6,
x = y + z is the representation of x in the direct sum E = B a ⊕ B0a . In
particular, the component of x in Ba (namely y ) is given by the desired
formula.
§2. LINEAR FORMS ON A RIESZ SPACE
II.10, `. −5.
“Theorem 1. — 1◦ In order that a linear form L on a Riesz space E
be relatively bounded, it is necessary and sufficient that it be the difference
of two positive linear forms.
2◦ The ordered vector space Ω of relatively bounded linear forms on E
is a Riesz space that is fully lattice-ordered.”
The proof can perhaps be clarified by separating out a part of it as a
lemma:
Lemma 0. — Let L be a relatively bounded linear form on the Riesz
space E . Let P = {x ∈ E : x > 0 } and define M : P → [0, +∞[ by the
formula
M (x) = sup L(y) for all x ∈ P .
06y6x

Then M may be extended to a positive linear form on E that is > L .

§2 linear forms on a riesz space INT II.x3

Proof. Note first that the indicated supremum is finite (because L is

relatively bounded). Moreover, M (x) > 0 for every x > 0 ; for, y = 0
satisfies 0 6 y 6 x and so 0 = L(0) 6 M (x) . Also, M (x) > L(x) for every
x > 0 because y = x satisfies 0 6 y 6 x . Thus if we show that M can
be extended to a linear form on E , it will be a positive linear form > L ; in
view of Prop. 3, it will suffice to show that M is additive.
Suppose x, x0 ∈ P and define

α = M (x) = sup L(y)

06y6x
0
β = M (x ) = sup L(y 0 )
06y 0 6x0

γ = M (x + x0 ) = sup L(z) ;
06z6x+x0

we are to show that α + β = γ .

To see that α + β 6 γ , let 0 6 y 6 x and 0 6 y 0 6 x0 . Then
0 6 y + y 0 6 x + x0 , so

L(y) + L(y 0 ) = L(y + y 0 ) 6 γ ;

varying y and y 0 independently, one concludes that α + β 6 γ . In detail,

for each y 0 , L(y) 6 γ − L(y 0 ) for all y , whence

α = sup L(y) 6 γ − L(y 0 ) ;

06y6x

thus L(y 0 ) 6 γ − α for all y 0 , whence β 6 γ − α .

To see that γ 6 α + β , suppose 0 6 z 6 x + x0 . By the decomposition
theorem (A, VI, §1, No. 10, Th. 1) one can write z = y + y 0 with 0 6 y 6 x
and 0 6 y 0 6 x0 ; explicitly, writing z 0 = (x + x0 ) − z > 0 , one has z + z 0 =
x + x0 and the first part of the proof of the cited theorem (loc. cit.) shows
that the elements y = sup(0, z − x0 ) and y 0 = z − y meet the requirements.
Then L(z) = L(y) + L(y 0 ) 6 α + β ; varying z , γ 6 α + β . This completes
the proof of the lemma.
Proof of 1◦ . “Necessity”: Suppose L is relatively bounded. By the
above Lemma 0, there exists a positive linear form M on E such that
M > L , and the formula L = M − (M − L) exhibits L as the difference of
two positive linear forms on E .
“Sufficiency”: Suppose L = U − V , where U and V are positive
linear forms on E . Let x ∈ E , x > 0 . If |y| 6 x , i.e., sup(y, −y) 6 x ,
i.e., −x 6 y 6 x , then −U (x) 6 U (y) 6 U (x) , i.e., |U (y)| 6 U (x) , and
similarly |V (y)| 6 V (x) , whence

|L(y)| = |U (y) − V (y)| 6 |U (y)| + |V (y)| 6 U (x) + V (x) ,

INT II.x4 riesz spaces §2

thus the set {|L(y)| : |y| 6 x } is bounded. In other words, L is relatively

bounded.
Proof of 2◦ . Write Ω for the set of linear forms characterized in 1 ◦ . Let
L ∈ Ω and let M be the positive linear form constructed in Lemma 0, such
that
M (x) = sup L(y) for all x > 0 .
06y6x

We assert that in the ordered vector space Ω (with the positive linear forms
serving as the convex cone of positive elements), the elements L and 0 have
a supremum, namely M . For, on the one hand, M > 0 and M > L . On
the other hand, if N ∈ Ω satisfies N > 0 and N > L , then for every x > 0
one has
0 6 y 6 x ⇒ N (x) > N (y) > L(y) ,
whence
N (x) > sup L(y) = M (x) ,
06y6x

thus N > M and the assertion is proved, authorizing us to write M =

sup(L, 0) . It follows that Ω is a Riesz space; for, if L, M are any two
elements of Ω , the linear form sup(L − M, 0) + M clearly serves as a supre-
mum for L and M . It follows from the foregoing calculations that, for
every x > 0 ,


(1) L+ (x) = sup(L, 0) (x) = sup L(y) .
06y6x

It remains only to show that Ω is fully lattice-ordered. In view of §1,

Prop. 1 (on p. II.4) it suffices to show that if H is any nonempty subset
of Ω that is bounded above in Ω and is directed upward by 6 , then H
admits a supremum. By the Lemma on p. II.12, we know that H has a
supremum M in the algebraic dual E∗ of E , such that

M (x) = sup L(x) for all x > 0 .

L∈H

In particular, for any L ∈ H , the relations M > L and M = L + (M − L)

show that M is the sum of two elements of Ω , hence M ∈ Ω .
II.12, formulas (3).
“From the formula (1), one deduces immediately that if L and M are
two relatively bounded linear forms on E then, for every x > 0 ,



 sup(L, M )(x) = sup L(y) + M (z)
y>0, z>0, y+z=x
(3)


 inf(L, M )(x) = inf L(y) + M (z) .”
y>0, z>0, y+z=x
§2 linear forms on a riesz space INT II.x5

Write N = sup(L, M ) . Citing formula (6) of §1, No. 1, one has

N = (L − M )+ + M ,

thus, for every x > 0 ,

N (x) = (L − M )+ (x) + M (x)

= sup [L(y) − M (y)] + M (x)
06y6x

= sup [L(y) − M (y) + M (x)]

06y6x

= sup [L(y) + M (x − y)]

06y6x

= sup [L(y) + M (z)] .

y>0, z>0, y+z=x

Since inf(L, M ) = L+M −sup(L, M ) (formula (8) of §1, No. 1), for all x > 0
one has


inf(L, M ) (x) = L(x) + M (x) − sup [L(y) + M (z)]
y>0, z>0, y+z=x

= L(x) + M (x) + inf [−L(y) − M (z)]

y>0, z>0, y+z=x

= inf [L(x) + M (x) − L(y) − M (z)]

y>0, z>0, y+z=x

= inf [L(x − y) + M (x − z)]

y>0, z>0, y+z=x

= inf [L(z) + M (y)]

y>0, z>0, y+z=x

= inf [L(y) + M (z)]

y>0, z>0, y+z=x

(the latter equality, by the symmetry of the roles of y and z ).

II.12, `. -7.
“...if x = y + z , y > 0 and z > 0 , then −x 6 y − z 6 x ”
For, y = x − z 6 (x − z) + 2z = x + z and z = x − y 6 (x − y) + 2y =
x + y , thus y − z 6 x and −x 6 y − z , whence the assertion (equivalently,
|y − z| 6 x ).
II.12, `. −6.
“the relation |u| 6 x implies L(u) 6 |L|(|u|) 6 |L|(x) .”
The inequality L(|u|) 6 |L|(x) is immediate from |u| 6 x and the
positivity of |L| . To see that L(u) 6 |L|(|u|) , set y = u + , z = u− and
x = y + z = u+ + u− = |u| ; then y − z = u+ − u− = u and, by the formula
displayed in `. −8, L(u) = L(y − z) 6 |L|(x) = |L|(|u|) .
INT II.x6 riesz spaces §2

II.12, `. −4.

“(4) |L|(x) = sup L(y) for x > 0 ”

|y|6x

Let x > 0 and let s = sup L(y) . If |u| 6 x then L(u) 6 |L|(x)
|y|6x
by `. −6, hence s 6 |L|(x) .
On the other hand, if x = y + z with y > 0 and z > 0 , then (as noted
in connection with `. −7) the element u = y − z satisfies |u| 6 x , hence
L(y −z) = L(u) 6 s ; varying y and z (subject to y > 0 , z > 0 , y +z = x )
it follows from the formula displayed in `. −8 that |L|(x) 6 s .
II.14, `. 5.
“Then every continuous linear form x 0 ∈ E0 is relatively bounded”
Recall that the convex cone P is proper (TVS, II, §2, No. 5, Prop. 13).
One has

{y : |y| 6 x } = {y : − x 6 y 6 x }
= {y : x − y ∈ P and x + y ∈ P }
= {y : y − x ∈ −P and x + y ∈ P }
= (x − P) ∩ (−x + P) ;

this set is homeomorphic (via translation by x ) to the set (2x − P) ∩ P ,

which is compact for σ(E, E0 ) (TVS, II, §6, No. 8, Cor. 2 of Prop. 11).
Since x0 ∈ E0 is continuous for σ(E, E0 ) , it follows that the set
x ({y : |y| 6 x }) is compact, hence x0 is bounded on {y : |y| 6 x } ;
0

thus x0 is relatively bounded (Def. 2).

II.14, `. 11–12.
“(the latter being compatible with the ordered vector space structure
of E )”
The point is that P , which is assumed to be closed for the original
topology of E (TVS, II, §2, No. 7, axiom (TO)), is also closed for σ(E, E 0 )
on account of its convexity (TVS, II, §5, No. 3, Cor. 1 of Prop. 4); thus
σ(E, E0 ) is also compatible with the ordered vector space structure of E .
II.14, `. −6.
“By translation, we can suppose that H ⊂ P ”
Choose any x0 ∈ H , drop down to H0 = {x ∈ H : x > x0 } , and
consider the translate −x0 + H0 ⊂ P .
§2 linear forms on a riesz space INT II.x7

II.14, `. −5, −4.

“. . .or again that every continuous linear form x 0 ∈ E0 has a limit with
respect to F ”
Assume that this condition is satisfied. Then, given any x 0 ∈ E0
and any ε > 0 , there exists an F0 ∈ F such that |x0 (x) − x0 (y)| 6 ε
for all x, y ∈ F0 . Since F is closed under finite intersections, it follows
that if x01 , . . . , x0n ∈ E0 and ε > 0 , there exists a set F0 ∈ F such that
|x0i (x) − x0i (y)| 6 ε for i = 1, . . . , n and x, y ∈ F0 , that is, writing

U = {(x, y) ∈ H × H : |x0i (x) − x0i (y)| 6 ε for i = 1, . . . , n } ,

we have F0 × F0 ⊂ U , i.e., F0 is small of order U . Since such sets U are

basic entourages for the uniform structure on H , it follows that F is Cauchy
for that uniform structure.
II.14, `. −4 to −2.
“But this follows at once from the monotone limit theorem when x 0 is
a positive linear form”
For, x0 (H) is an increasing directed set in R that is bounded above,
hence x0 has a limit with respect to F , namely lim F x0 = sup x0 (H) .
 CHAPTER III

Measures on locally compact spaces

§1. MEASURES ON A LOCALLY COMPACT SPACE

III.1, `. −6, −5.

“(namely, the subspace of continuous mappings of K into E that are
zero on the boundary of K ).”
Recall that the boundary of a subset A of a topological space X is the
set B = A ∩ X --- A (which is also the boundary of X --- A ). Note that the
union of A with its boundary is the closure of A : A ∪ B = A ; for,

A ∪ B = A ∪ (A ∩ X --- A) = (A ∪ A) ∩ (A ∪ X --- A) = A ∩ X = A .

◦ ◦
Incidentally, X --- A = { A , where A is the interior of A (GT, I, §1, No. 6,
formulas (2)). And when A is a closed set, A contains its boundary, since
A = A = A ∪ B.
Suppose now that A is a closed subset of X and f : A → Y is a con-
tinuous mapping of A into a topological space Y , such that the restriction
of f to its boundary B is constant, say f (B) = {y 0 } for some y0 ∈ Y .
Define g : X → Y to be the function equal to f on A and to y 0 on X --- A .
Since B is also the boundary of X --- A , g has the constant value y 0 on
the set X --- A = (X --- A) ∪ B . Thus, g is continuous on each of the closed
sets A and X --- A , whose union is X , consequently g is continuous on X
(GT, I, §3, No. 2, Prop. 4).
In particular, if K is a compact subset of the locally compact space X ,
E is a topological vector space over R or C , and f : K → E is a continuous
function that is zero on the boundary of K , then the extension g of f
to X defined by setting g(x) = 0 for all x ∈ X --- K is continuous, and
its support, being a closed subset of K , is compact. Conversely, if K is a
compact subset of X and g ∈ K (X, K; E) , that is, g is continuous on X
with support contained in K , i.e., {x : g(x) 6= 0 } ⊂ K , equivalently g is
equal to 0 on X --- K , then g is also equal to 0 on X --- K , hence on the
§1 measures on a locally compact space INT III.x2

boundary K ∩ X --- K of K , thus the restriction f = g K is a continuous
function on K equal to 0 on the boundary of K .
III.1, `. −5, −4.
“When C (K; E) is equipped with the topology of uniform convergence
in K , K (X, K; E) is a closed subspace of C (K; E) .”
More generally, suppose (gα ) is a directed family in K (X, K; E) that
converges pointwise to a function g ∈ C (K; E) . Then (assuming E is Haus-
dorff) g is equal to 0 on the boundary of K , hence belongs to K (X, K; E) .
For, if there existed a point x of the boundary of K such that g(x) 6= 0 ,
one could choose a neighborhood V of 0 in E such that g(x) ∈ / V ; but
there exists an α0 such that α > α0 ⇒ g(x) − gα (x) ∈ V , a contradiction
since gα (x) = 0 .
III.1, `. −2.
“. . .if the topology of E is defined by the semi-norms p n ”
Recall that a Fréchet space is a complete, metrizable locally convex
space, and that the topology of a metrizable locally convex space can be
defined by a countable set of semi-norms (TVS, II, §4, No. 1, comments
following the Corollary of Prop. 1).
III.2, `. 7, 8.
“If E is locally convex, one can therefore define on K (X; E) the direct
limit of the locally convex topologies of the K (X, K; E) ”
Recall (TVS, II, §4, No. 4, Prop. 5) that this topology (call it T ) is
the finest locally convex topology on K (X; E) that renders continuous the
canonical injections

uK : K (X, K; E) → K (X; E) ,

where, for each compact subset K of X , the space K (X, K; E) bears the
topology of uniform convergence in K ; and, for a linear mapping
u : K (X; E) → G ( G a locally convex space), the following conditions
are equivalent:
(a) u is continuous for the topology T ;
(b) for every compact subset K of X , the linear mapping
u ◦ uK : K (X, K; E) → G is continuous; that is,
(b0 ) for every compact subset K of X and for every directed family (f α )
in K (X, K; E) that converges uniformly to a function f ∈ K (X, K; E) , one
has u(fα ) → u(f ) in G .
The mappings uK are continuous when K (X, K; E) is equipped with
the topology of uniform convergence in K and K (X; E) is equipped with
the topology Tu of uniform convergence in X , consequently T is finer
INT III.x3 measures §1

than Tu (TVS, loc. cit., Example. Caution: In the cited Example, TVS
p. II.29, in lines 4 and 5 read “Denote by T K the topology induced on EK
by the topology Tu of uniform convergence on X ”; the sentence is correct
in EVT, p. II.31).
Scholium: The direct limit topology on K (X; E) is finer than the
topology of uniform convergence.
III.2, `. −5.
“. . .hence belongs to K (X, K; E) .”
Here f ∈ K (X, K0 ; E) and there exists a family (fα ) in K (X, K; E)
such that fα → f uniformly on K0 , hence uniformly on K0 --- K , hence
pointwise on K0 − K . Since the fα are 0 on X --- K , it follows that
f = 0 on K0 --- K ; but also f = 0 on X --- K0 , hence f = 0 on
X --- K = (X --- K0 ) ∪ (K0 --- K) , that is, f ∈ K (X, K; E).
III.2, `. −4, −3.
“(ii) The criterion for continuity in a direct limit (TVS, II, §4, No. 4,
Prop. 5) shows at once that the mapping f 7→ (pri ◦ f ) is continuous”
It suffices to show that for each i , f 7→ pr i ◦ f is a continuous mapping
K (X; E) → K (X; Ei ) . Fix a compact set K ⊂ X ; by the cited result in
TVS, it is enough to show that the composite mapping

K (X, K; E) → K (X; E) → K (X; Ei )

defined by f 7→ f 7→ pri ◦ f is continuous. Indeed, if fα , f ∈ K (X, K; E)

and fα → f uniformly on K , then pri ◦ fα → pri ◦ f uniformly on K
(regard Ei ⊂ E in the canonical way); thus pri ◦ fα , pri ◦ f ∈ K (X, K; Ei )
and pri ◦fα → pri ◦f in the space K (X, K; Ei ) (equipped with the topology
of uniform convergence on K ) hence in the space K (X; E i ) (by (i)).
III.2, `. −2.
“. . .the same is true of the inverse mapping”
Q
If (fi )16i6n ∈ K (X; Ei ) , for each i define a function
16i6n
fi0 ∈ K (X; E) to be the mapping such that pr i ◦ fi0 = fi and prj ◦ fi0 = 0
0
Q f i 7→ fi (1 6 i 6 n) is obviously
when j 6= i . Since each of the mappings
continuous, so is the mapping Φ : K (X; Ei ) → K (X; E) defined by
16i6n

P 0


Φ (fi ) = fi , and prj Φ (fi ) = fj (1 6 j 6 n) shows that Φ
16i6n
is Q
the mapping inverse to the mapping f 7→ (pr i ◦ f ) of K (X; E) into
K (X; Ei ) defined above.
16i6n
§1 measures on a locally compact space INT III.x4

III.3, `. 8–9.
“. . .it is immediate that the mapping f λ 7→ fλ00 of K (Xλ ; E) into
K (X; E) is continuous.”
For each compact Kλ ⊂ Xλ , the composite mapping

K (Xλ , Kλ ; E) → K (Xλ ; E) → K (X; E)

defined by fλ 7→ fλ 7→ fλ00 is obviously continuous.

III.3, `. 9–11.
“The assertion (iii) follows from these remarks and the criterion for
continuity in direct limits (TVS, II, §4, No. 4, Prop. 5).”
L
Define ϕ : K (X; E) → K (Xλ ; E) as follows. Let f ∈ K (X; E)
λ∈L
and let K be the support of f . Since the X λ form an open covering of X ,
K intersects Xλ for at most finitely many 
 λ , therefore f Xλ = 0 for all
but finitely many λ ; thus the family (f Xλ )λ∈L qualifies for membership
L
in K (Xλ ; E) and we may define ϕ(f ) = (f Xλ )λ∈L . Conversely, if
λ∈L L
(gλ )λ∈L ∈ K (Xλ ; E) , in particular gλ = 0 for all but finitely many λ ,
λ∈L 
it is clear that the function f on X defined by f Xλ = gλ for all λ is
continuous and has compact support (equal to the union of the supports of
the nonzero gλ ’s), and that ϕ(f ) = (gλ )λ∈L . Clearly ϕ is a vector space
isomorphism.
To see that ϕ is continuous, suppose K ⊂ X compact and consider the
composite mapping
M
ϕK : K (X, K; E) → K (X; E) → K (Xλ ; E)
λ∈L

L 
defined by f 7→ f 7→ (f Xλ ) . By the preceding discussion, K inter-
λ∈L
sects Xλ only for λ in some finite subset
L of L , thus the the above composite
has range contained in the subspace K (Xλ ; E) for some finite subset H
λ∈H
of L , and it defines a bicontinuous isomorphism of K (X, K; E) onto that
subspace; in particular, ϕK is continuous for every compact K ⊂ X , whence
the continuity of ϕ .
To see that ϕ−1 is continuous, it is enough, by the definition of topo-
logical direct sum, to note that for each λ ∈ L , the composite
M
K (Xλ ; E) → K (Xµ ; E) → K (X; E)
µ∈L
INT III.x5 measures §1

of the canonical injection and ϕ−1 is continuous; indeed, this composite is

precisely the continuous mapping f λ 7→ fλ00 (fλ ∈ K (Xλ ; E) .
III.4, `. 5.
“1◦ it is closed;”
Note that since K (X, K; E) is closed in K (X; E) , a subset of K (X, K; E)
is closed in K (X, K; E) if and only if it is closed in K (X; E) .
III.4, `. 5–6.
“2◦ it is equicontinuous;”
Note that a subset H of K (X, K; E) is equicontinuous on K if and only
if it is equicontinuous on X , since every function in H is zero on X − K .
III.4, Proof of the Corollary..
“It suffices, by virtue of Proposition 2, (ii), to note that for every com-
pact subset K of X , K (X, K; E) is a closed subspace of C (K; E) , which is
quasi-complete since every bounded subset of C (K; E) consists of functions
taking values in a same bounded subset of E .”
The proof is a “Knight’s tour” of the prerequisites; let us review some
fundamental definitions.
A subset B of a topological vector space E is bounded if, for every
neighborhood V of 0 in E , there exists a nonzero scalar λ such that
A ⊂ λV (TVS, III, §1, No. 2, comments following Def. 3). In testing for
boundedness, one can obviously restrict attention to V ’s belonging to a
fundamental system of neighborhoods of 0 .
A locally convex topological vector space F is said to be quasi-complete
if every closed bounded subset B of F is complete for the uniform structure
induced on B by that of F .
Suppose H is a closed bounded subset of K (X; E) ; we are to show
that H is complete. By Prop. 2, (ii), there exists a compact subset K
of X such that H ⊂ K (X, K; E) . As noted in No. 1, K (X, K; E) may be
identified with a closed subspace of C (K; E) , namely, the subspace C 0 (K; E)
consisting of the functions f = g K , where g ∈ K (X, K;E) and g = 0 on
the boundary K∩X --- K of K in X ; the mapping g 7→ g K is a topological
vector space isomorphism of K (X, K; E) onto C 0 (K; E) .
Let us digress to describe the bounded subsets of C (K; E) , where
C (K; E) is equipped with the topology of uniform convergence on K (GT,
X, §1, No. 1, Def. 1). In detail, a fundamental system of entourages for the
uniform structure on E is formed by the sets
W = {(a, b) ∈ E × E : a − b ∈ V } ,
where V runs over any fundamental system of neighborhoods of 0 in E
(TVS, I, §1, No. 4 and GT, III, §3, No. 1, Def. 1). A fundamental system of
§1 measures on a locally compact space INT III.x6

entourages for the uniform structure on C (K; E) is formed by the sets

W = {(f, g) ∈ C (K; E) : f (x), g(x) ∈ W for all x ∈ K } ,

where W runs over any fundamental system of entourages for the uniform
structure on E (GT, X, §1, No. 1, Def. 1). Thus a fundamental system of
entourages for the uniform structure on C (K; E) is formed by the sets

W = {(f, g) ∈ C (K; E) : f (x) − g(x) ∈ V for all x ∈ K } ,

where V runs over any fundamental system of neighborhoods of 0 in E . In

particular, a fundamental system of neighborhoods of a function
g ∈ C (K; E) is formed by the sets

{f ∈ C (K; E) : f (x) − g(x) ∈ V for all x ∈ K } ,

where V runs over a fundamental system of neighborhoods of 0 in E (GT,

II, §1, No. 2, Prop. 1); and so a fundamental system of neighborhoods of the
zero function 0 ∈ C (K; E) is formed by the sets

NV = {f ∈ C (K; E) : f (x) − 0 ∈ V for all x ∈ K }

= {f ∈ C (K; E) : f (K) ⊂ V } ,

where V runs over a fundamental system of neighborhoods of 0 in E .

The bounded subsets of C (K; E) may now be described as follows: in
order that a subset B of C (K; E) be bounded, it is necessary and sufficient
that for every neighborhood V of 0 in E , there exist a scalar λ 6= 0 such
that B ⊂ λNV , this inclusion being equivalent to each of the following
conditions:
λ−1 B ⊂ NV
λ−1 B ⊂ {f ∈ C (K; E) : f (K) ⊂ V }
λ−1 u(K) ⊂ V for all u ∈ B
u(K) ⊂ λV for all u ∈ B
[
u(K) ⊂ λV .
u∈B
S
In brief, a subset B of C (K; E) is bounded if and only if the set u(K)
u∈B
is bounded in E .
Let us return to the closed bounded subset H of K (X; E) and let K be
a compact subset of X such that H ⊂ K (X, K; E) . Let H 0 =
{g K : g ∈ H } ; thus H0 ⊂ C0 (K; E) ⊂ C (K; E) . Since the mapping
INT III.x7 measures §1

g 7→ g K is an isomorphism of topological vector spaces, it is clear that
H0 is a closed and bounded subset of C0 (K;
SE) and hence of C (K; E) . As
shown in the preceding paragraph, the set f (K) is bounded in E , hence
f ∈H0
(TVS, III, §1, No. 2, Prop. 1, d)) so is its closure
[
A= f (K),
f ∈H0

and since E is quasi-complete, A is complete (for the induced uniform

structure). Thus H0 may be regarded as a subset of the space C (K; A)
equipped with the topology of uniform convergence. Since A is complete,
so is C (K; A) (GT, X, §1, No. 6, Cor. 1 of Th. 2). Regarding C (K; A) as
a uniform subspace of C (K; E) , we have
H0 ⊂ C (K; A) ⊂ C (K; E) ,
where H0 is closed in C (K; E) , hence in C (K; A) , and since C (K; A) is
complete we conclude that H0 is complete (GT, II, §3, No. 4, Prop. 8),
therefore so is H , and we have proved that K (X; E) is quasi-complete.
III.4, `. −12, −11.
“(GT, IX, §4, No. 3, Prop. 3)”
The cited proposition exploits Urysohn’s theorem (GT, IX, §4, No. 1,
Th. 1) via the normality of a compact space (loc. cit., Prop. 1). The cor-
responding result in the bound French edition is TG, IX, §4, No. 3, Th. 3.
The difference is due to the fact that the translation in GT is based on an
earlier French edition of Chapter IX.
III.6, `. −14.
“ . . . whence our assertion.”
Let us review the topology of compact convergence on C (X; E) (cf.,
GT, X, §1, No. 3, Example III), that is, the topology of uniform convergence
in the compact subsets of X . (See also the notes for III.4, `. 5–6 and III.39,
`. 8–11.)
Let (Nα ) be a fundamental system of neighborhoods of 0 in E . The
sets
Vα = {(x, y) ∈ E × E : x − y ∈ Nα }
then form a fundamental system of entourages for the uniformity on E (TVS,
I, §1, No. 4 and GT, III, §3, No. 1, remark following Def. 1). A fundamental
system of entourages for a uniformity on C = C (K; E) is then given by the
sets


WK,α = {(f , g) ∈ C × C : f (x), g(x) ∈ Vα for all x ∈ K }
= {(f , g) ∈ C × C : f (x) − g(x) ∈ Nα for all x ∈ K }
§1 measures on a locally compact space INT III.x8

where K runs over the set of all compact subsets of X . The deduced topol-
ogy is called the topology of compact convergence and will be denoted τ cc ;
it has as fundamental system of neighborhoods of 0 the sets

NK,α = {f ∈ C : f (K) ⊂ Nα }

(notably, given indices α1 , . . . , αn and compact sets K1 , . . . , Kn , if α is

chosen so that Vα1 ∩ · · · ∩ Vαn ⊃ Vα , then NK1 ,α1 ∩ · · · ∩ NKn ,αn ⊃
NK1 ∪···∪Kn ,α ); and, for fixed f ∈ C , the sets

VK,α (f ) = f +NK,α = {g ∈ C : g−f ∈ NK,α } = {g ∈ C : (g−f )(K) ⊂ Nα }

form a fundamental system of neighborhoods of f for τ cc . One writes

Cc (X; E) for C (X; E) equipped with the topology τ cc .
Given f ∈ Cc (X; E) , we are to show that f belongs to the closure of
K (X; E) ; it suffices to show that every neighborhood V K,α (f ) contains an
element of K (X; E) . Indeed, with h ∈ K (X; R) chosen as in the proof,
one has hf = f on K , therefore

(hf − f )(K) = {0} ⊂ Nα ,

thus hf ∈ VK,α (f ).
III.6, `. −7.
. . . the second assertion is an obvious consequence of the first ”
For definiteness, let us consider the case of real scalars (the same argu-
ment works in the complex case). The identification in question associates
Pn
with f = ϕk ⊗ yk (where ϕk ∈ K (X, K; R) and yk ∈ E ) the function
k=1

n
X
x 7→ ϕk (x)yk (x ∈ X) ,
k=1

clearly an element of K (X, K; E) . By definition,

[
(i) K (X; E) = K (X, K; E) ,
K

where K runs over the set of all compact subsets of X ; citing (i) for E = R ,
we infer that
[
(ii) K (X; R) ⊗ E = K (X, K; R) ⊗ E .
K
INT III.x9 measures §1

The left side of (ii), viewed as a set of (continuous) functions X → E , is

clearly contained in the left side of (i).
Recall (No. 1) that the topology on K (X, K; E) , for K compact in X ,
is the topology of uniform convergence in X , equivalently in K (it is de-
scribed in detail in the notes for III.39, `. 8–11), and the topology on
K (X; E) is the direct limit of the topologies on the K (X, K; E) as K
varies over the set of all compact subsets of X . The first assertion of the
proposition is that the closure of K (X, K; R)⊗E in K (X, K; E) is equal to
K (X, K; E) . Since K (X, K; E) is a closed topological subspace of K (X; E)
(No. 1, Prop. 1), K (X, K; E) is also the closure of K (X, K; R) ⊗ E in
K (X; E) (GT, I, §3, No. 1, Cor. of Prop. 1); it then follows from (ii) that
the closure of K (X; R) ⊗ E in K (X; E) contains every term on the right
side of (i), hence is equal to K (X; E) .
III.6, `. −2, −1.
“ . . . this proves the proposition, by the definition of the topology of
K (X, K; E) .”
The formulation of the topology of (hence convergence in) K (X, K; E)
in terms of continuous semi-norms on E is worked out in the note for III.39,
`. 8–11; explicitly, the argument given here shows that, in the notations of (7)
and (8) of that note, given any f ∈ K (X, K; E) and any neighborhood of 0
in K (X, K; E) of the form


Uα = {f ∈ K (X, K; E) : qα f (x) 6 1 for all x ∈ X } ,
P
n
with qα a continuous semi-norm on E , there exists an element ϕj ⊗f (xj )
j=1
of K (X, K; R) ⊗ E (interpreted as a function on X ) such that
n
X
f− ϕj ⊗ f (xj ) ∈ Uα ,
j=1

whence the second assertion of the proposition.

III.10, `. 13.
“. . .whence our assertion (TVS, II, §4, No. 4, Prop. 5).”
Write u : K (X; C) → K (X; C) , u(f ) = gf . The foregoing shows that
for each compact set K ⊂ X , f 7→ gf is a continuous linear mapping of
K (X, K; C) into itself. Then for each K , the restricted mapping

uK (X, K; C) : K (X, K; C) → K (X; C)
is the composite
K (X, K; C) → K (X, K; C) → K (X; C)
§1 measures on a locally compact space INT III.x10

of two continuous mappings ( f 7→ gf 7→ gf ), hence is continuous. There-

fore u is continuous by the cited proposition (with G = K (X; C) ).
More generally, suppose E, F are locally convex spaces that are the
direct limits of families (with same index set) of subspaces E α , Fα of E, F ,
respectively. If u : E → F is a linear mapping such that for every α ,
u(E 
 α ) ⊂ Fα and u Eα : Eα → F is continuous, then u is continuous; for,

u Eα : Eα → F is the composite Eα → Fα → F of two continuous mappings
( x 7→ u(x) 7→ u(x) ).
III.11, `. 4–6.
“. . .moreover, the mapping (f1 , f2 ) 7→ f1 + if2 is an isomorphism of
the product topological vector space K (X; R) × K (X; R) onto the real
topological vector space K (X; C) (No. 1, Prop. 1).”
By the cited proposition, one has an isomorphism of real topological
vector spaces
Φ : K (X; R2 ) → K (X; R) × K (X; R)
defined by Φ(g) = (pr1 ◦ g, pr2 ◦ g) , where pr1 (a1 , a2 ) = a1 , pr2 (a1 , a2 ) =
2 −1
a2 for (a1 , a2 ) ∈ R
; the inverse mapping 2 is Φ (f1 , f2 ) = g , where
g(x) = f1 (x), f2 (x) . Since the mapping θ : R → C defined by θ(a1 , a2 ) =
a1 + ia2 is an isomorphism of real topological vector spaces, the mapping

Ψ : K (X; R2 ) → K (X; C)

defined by Ψ(g) = θ ◦ g is also an isomorphism of real topological vector

spaces. Composing, we have an isomorphism

Ψ ◦ Φ−1 : K (X; R) × K (X; R) → K (X; C) ,

namely


(Ψ ◦ Φ−1 )(f1 , f2 ) = Ψ Φ−1 (f1 , f2 )
= Ψ(g) (where g(x) = (f1 (x), f2 (x) ))
= θ◦g,


where (θ ◦ g)(x) = θ(g(x)) = θ (f1 (x), f2 (x) = f1 (x) + if2 (x) , thus
(Ψ ◦ Φ−1 )(f1 , f2 ) = f1 + if2 is the desired isomorphism.
III.14, `. 1–3.
“Moreover, we can suppose ζ so chosen that

|µ(g1 ) + ζµ(g2 )| = |µ(g1 )| + |µ(g2 )| ;

since |µ(gi )| is arbitrarily close to L(fi ) (i = 1, 2) . . .”

INT III.x11 measures §1

First choose gi so that |µ(gi )| is near L(fi ) , then choose ζ so as to

rotate µ(g2 ) to point in the same direction as µ(g 1 ) in the complex plane.
III.14, `. 9–10.
“. . .which proves the continuity of g i at the points where f1 (x) +
f2 (x) = 0 (i = 1, 2) , since at these points we have also g(x) = 0 .”
Some preliminaries are in order. Given g ∈ K (X; C) with |g| 6 f 1 +f2 ,
let U = {x ∈ X : f1 (x) + f2 (x) > 0 } ; since the fi are > 0 , X --- U =
{x ∈ X : f1 (x) = f2 (x) = 0 } . Define gi : X → C (i = 1, 2) by the formulas

 gfi on U
gi = f1 + f 2

0 on X --- U .
The continuity of gi at the points of U is obvious. Note that g 1 + g2 = g ;
this is clear at the points of U , whereas if x ∈ X --- U then g 1 (x) = g2 (x) = 0
by definition, and g(x) = 0 follows from |g(x)| 6 f 1 (x) + f2 (x) = 0 . Also,
|gi | 6 g ; for, if x ∈ X --- U then gi (x) = g(x) = 0 , whereas if x ∈ U then
fi (x)
|gi (x)| = |g(x)| · 6 |g(x)| .
f1 (x) + f2 (x)
And |gi | 6 fi ; for, if x ∈ X --- U then fi (x) = gi (x) = 0, whereas if x ∈ U
then
|g(x)|
|gi (x)| = fi (x) · 6 fi (x)
f1 (x) + f2 (x)
because |g| 6 f1 + f2 . End of preliminaries.
 Consider the covering of X by the closed sets X --- U and U . Since
gi X --- U = 0 is continuous,
 to prove that g i is continuous on X we need

only show that gi U is continuous (GT, I, §3, No. 2, Prop. 4).
Given x0 ∈ U and a directed family (xα ) in U such that xα → x0 , we
are to show that gi (xα ) → gi (x0 ) for i = 1, 2 . If x0 ∈ U then xα ∈ U from
some index onward, whence gi (xα ) → gi (x0 ) by the continuity of gi on U .
Thus we can suppose that x0 ∈ U --- U = U ∩ (X --- U) . Since x0 ∈ X --- U
we have gi (x0 ) = 0 ; also f1 (x0 ) = f2 (x0 ) = 0 . By the continuity of fi ,
we know that fi (xα ) → fi (x0 ) = 0 ; but |gi (xα )| 6 fi (xα ) → 0 , whence
gi (xα ) → 0 = gi (x0 ) .
III.14, `. 16.
“When µ is a real measure, it follows from formula (9) that |µ| 6 L ”
R
Let f ∈ K+ (X) . If g ∈ K (X; R) and |g| 6 f then g dµ ∈ R and
Z Z 
 
g dµ 6  g dµ 6 L(f )
§1 measures on a locally compact space INT III.x12
R
by the definition of L ; taking the supremum over all such g , f d|µ| 6 L(f )
by (9); thus |µ|(f ) 6 L(f ) for all f ∈ K + (X) , that is, |µ| 6 L as positive
linear forms on K (X; R) .
III.14, `. −7, −6.
“consequently, for every function g ∈ K (X; C) ,
Z  Z
 
(13)  g dµ 6 |g| d|µ| .”

Putting f = |g| in (12),

|µ|(|g|) = sup |µ(h)| > |µ(g)|

|h|6|g|, h∈K (X;C)

(because |g| 6 |g| ).

III.15, `. 4, 5.
“therefore

(16) |µ| = |µ| .”

 
For all g ∈ K (X; C) with |g| 6 f ∈ K+ (X) , |µ(g)| = µ(g) = |µ(g)| ;
taking supremum over all such g , one has |µ|(f ) = |µ|(f ) for all
f ∈ K+ (X) , hence |µ| = |µ| on K (X; C) .
III.15, `. 14, 15.
“if V is a dense linear subspace of K (X; C) ,”
Caution: Recall that K (X; C) (without express mention to the con-
trary) bears the direct limit topology T , which is finer than the topology
Tu of uniform convergence in X , thus a linear subspace dense in the sense
of T is dense for Tu ; but the converse may fail (TVS, II, §4, No. 4, Example
on page TVS II.29).
III.15, `. 16–18.
“. . .every linear form on V that is continuous for the topology induced
by that of K (X; C) may be extended (in only one way) to a measure on X .”
Since K (X; C) is locally convex for the direct limit topology (§1,
No. 1), the Hahn–Banach theorem is applicable (TVS, II, §8, No. 3, Cor. 2
of Th. 1).
III.15, `. −2, −1.
“It then suffices to apply Th. 1 of No. 5 and Prop. 1 of TVS, II, §3,
No. 1.”
INT III.x13 measures §1

By the cited proposition (see TVS II.21), the property of V just derived
assures that every positive linear form µ 0 on V (for the relative order) may
be extended to a positive linear form on K (X; R) , that is, to a positive
measure µ on X (No. 5, Th. 1); if, in addition, V is dense (for the direct
limit topology), then the extension is unique by the remarks at the beginning
of the present subsection.
III.16, `. 6–8.
“Definition 3. — A measure on a locally compact space X is said
to be bounded if it is continuous on K (X; C) for the topology of uniform
convergence.”
Note that a linear form on K (X; C) continuous for the topology of uni-
form convergence Tu (i.e., the norm topology) is automatically continuous
for the direct limit topology T (because T u ⊂ T ), hence is a measure.
III.16, `. 13–15.
“To say that µ is a bounded measure thus signifies that µ belongs to
the dual of the space K (X; C) normed by kf k ; we shall denote this dual
by M 1 (X; C) (or simply M 1 (X) when no confusion can result).”
The case X = N . The direct limit topology on K (X; C) remains a
slippery concept to grasp; it is instructive to trace through the concepts on
an atypical but simple example: X = N = {0, 1, 2, 3, . . .} with the discrete
topology.
(1) F (X) , the set of all functions f : N → C , may be identified with
the vector space (s) (in the notation of Banach’s book) of all sequences
(ck )k>0 of complex numbers with the termwise linear operations (or, when
viewed as a product space Cℵ0 , with the coordinatewise linear operations).
(2) C (X) = F (X) , as all functions from a discrete space to a topological
space ( C in this instance) are continuous.
(3) C b (X) : This is the linear subspace of F (X) consisting of all
bounded complex functions on X ; it may be identified with the space (m)
(in the notation of Banach’s book) of bounded sequences of complex num-
bers.
(4) K (X) : As the compact subsets K of X = N are its finite sub-
sets, K (X) is the set of all complex functions on X with finite support;
it may be identified with the linear subspace of (m) consisting of all com-
plex sequences that terminate in zeros. In particular, K (X, K) is finite-
dimensional, K (X) being the union of its linear subspaces K (X, K n ) ,
where Kn = {0, 1, 2, . . . , n} (n = 0, 1, 2, . . .) .
(5) The topology on K (X, K) , K compact (i.e., finite).
The topology of uniform convergence on K (X, K) is simply the topol-
ogy of pointwise convergence; since K (X, K) is finite-dimensional, this is
§1 measures on a locally compact space INT III.x14

the unique topology on K (X, K) that makes it a Hausdorff topological

vector space (TVS, I, §2, No. 3, Th. 2). One can define a semi-norm p K
on K (X) by
pK (f ) = sup |f (k)| = max |f (k)|
k∈K k∈K

(also denoted kf kK ); the restriction of pK to K (X, K) is a norm that

generates the topology just described.
(6) The topology on K (X) of uniform convergence in the compact sub-
sets of X .
Since the compact sets are finite, this is just the topology of pointwise
convergence.
(7) The topology on K (X) of uniform convergence in X .
This is the topology on K (X) derived from the norm

kf k = sup |f (k)| = max |f (k)| = sup{pK (f ) : K ⊂ N finite};

k∈N k∈N

it is obviously finer than the topology of pointwise convergence. The se-

quence fn = nϕn , where ϕn is the characteristic function of {n} , converges
pointwise to 0 , but kfn k = n ; thus the topology of uniform convergence
in X is strictly finer than the topology of uniform convergence in the com-
pact subsets of X .

The normed space K (X), k · k is incomplete, i.e., is not a Banach
1
space: for example, if fn is the function defined by fn (k) = for
k+1
k = 0, 1, . . . , n and fn (k) = 0 for k > n , then kfm −fn k → 0 as m, n → ∞ ;
but the sequence (fn ) is not convergent in K (X) , indeed f n → f ∈ F (X)
1
pointwise implies that f (k) = for all k ∈ N , whence f ∈/ K (X) .
k+1
(8) The direct limit topology T on K (X) .
The key observation: If S is a locally convex (Hausdorff) topology
on K (X) , then S is coarser than T (i.e., S ⊂ T , that is, the open sets
for S are open for T ). For, for each finite subset K of X , the insertion
mapping (or “canonical injection”)

uK : K (X, K) → K (X) , uK (f ) = f

(where K (X, K) is equipped with its unique compatible Hausdorff topology,

and K (X) is equipped with S ) is continuous because K (X, K) is finite-
dimensional (TVS, I, §2, No. 3, Cor. 2 of Th. 2), whence S ⊂ T by the
definition of T . On the other hand, there exists on K (X) a finest locally
convex topology Tω (TVS, II, §4, No. 2, item 2) and loc. cit., §8, No. 2,
fifth paragraph). Putting S = Tω in the foregoing, Tω ⊂ T and so
INT III.x15 measures §1

T = Tω by the maximality of Tω . Thus, T is the topological vector space

topology on K (X) for which the set of all absorbent, balanced, convex
sets in K (X) forms a fundamental system of neighborhoods of 0 (TVS,
loc. cit.). A fortiori, every closed (for T ), absorbent, balanced convex set
in K (X) is a neighborhood of 0 , consequently K (X) is barreled (TVS,
III, §4, No. 1, Def. 2). Since discrete spaces are paracompact (GT, I, §9,
No. 10), K (X) is quasi-complete (for T ) (III, §1, No. 1, Cor. of Prop. 2);
indeed, a bounded (in the topological vector space sense) subset of K (X) is
contained in the finite-dimensional space K (X, K) for some finite K ⊂ X
(loc. cit., Prop. 2), thus a closed and bounded subset of K (X) is compact
(hence complete) by the Weierstrass–Bolzano theorem (for C n ).
(9) M (X) = K (X)∗ (the algebraic dual of K (X) ).
Every linear mapping of K (X) into a locally convex space is continuous
(for the topology T ). For, suppose v : K (X) → G is a linear mapping,
G a locally convex space. For every finite subset K of X , the composite
mapping v ◦ uK ,
K (X, K) → K (X) → G

is continuous by the finite-dimensionality of K (X, K) , whence the conti-

nuity of v (TVS, II, §4, No. 4, Prop. 5). In particular, every linear form
on K (X) is continuous (for T ), i.e., is a measure on X . Since a linear form
continuous for the topology Tu of uniform convergence must be bounded,
one infers that T ⊃ Tu properly.
{Since T = Tω , an alternative proof is available: If E is a vector space
equipped with its finest locally convex topology, then every linear mapping
of E into a locally convex space G is continuous; for, the inverse image of an
absorbent balanced convex subset of G has the same properties, hence is a
neighborhood of 0 in E , and G has a fundamental system of neighborhoods
of 0 (automatically absorbent) that are balanced and convex (TVS, II, §4,
No. 1, p. TVS II.23, `. −9 to −7, and loc. cit., §8, No. 2, p. TVS II.62).}
(10) The vector space M (X) = K (X)∗ may be identified with the space
(s) of all complex sequences.
For every n ∈ N let ϕn : N → C be the characteristic function of {n} :
ϕn (k) = δnk (k = 0, 1, 2, . . .) . Since every f ∈ K (X) has finite support,
it is clear that the ϕn form a basis of K (X) ; so if µ ∈ M (X) ,
µ is
characterized by its values cn = µ(ϕn ) (n ∈ N) , thus µ 7→ µ(ϕn ) n>0
is a (clearly injective) linear mapping of M (X) into the space (s) of all
complex sequences. This mapping is bijective; for, if c = (c n ) ∈ (s) , the
P
∞
linear form µ : K (X)) → C defined by µ(f ) = ck f (k) (actually a finite
k=0
sum) satisfies µ(ϕn ) = cn for all n ∈ N , so that µ 7→ c under the mapping.
§1 measures on a locally compact space INT III.x16

(11) The vague topology on M (X) .

In M (X) , “ µα → µ vaguely” means that µα (f ) → µ(f ) in C for
every f ∈ K (X) ; equivalently, µα (ϕn ) → µ(ϕn ) for every n ∈ N . Thus, in
the identification M (X) → (s) , the vague topology on M (X) corresponds
to the topology of coordinatewise convergence in (s) , viewed as a product
space Cℵ0 . The latter topology is metrizable (Banach, p. 10), by the metric
∞

X 1 |an − bn |
d (an ), (bn ) = n
· ,
n=0
2 1 + |an − bn |

for which (s) is complete. The corresponding metric D on M (X) is

X∞
1 |µ(ϕn ) − ν(ϕn )|
D(µ, ν) = n
·
n=0
2 1 + |µ(ϕn ) − ν(ϕn )|

( ϕn the characteristic function of {n} ). The metric D is invariant in

the sense that D(µ + ρ, ν + ρ) = D(µ, ν) ; consequently the additive group
of M (X) is a complete topological group for the vague topology (cf. Prop. 7
on p. III.8 above, with S the set of all singletons {ϕ n } , n ∈ N ). In other
words, M (X) is a complete topological vector space; being locally convex
and metrizable, it is a Fréchet space (TVS, II, §4, No. 1, p. TVS II.24), hence
is barreled (TVS, III, §4, No. 1, Cor. of Prop. 2).
(12) The space M 1 (X) of bounded measures on X = N .
Let µ be a bounded measure on X , that is, a linear form on K (X)
that is continuous for the topology of uniform convergence in X , i.e., for the
norm
kf k = sup |f (n)| = max |f (n)| .
n∈N n∈N

An equivalent condition on
µ is that it be a “bounded linear form” on the

normed space K (X), k · k in the sense that there exists a constant M > 0
such that
|µ(f )| 6 M kf k for all f ∈ K (X) .
The smallest possible value of M (the infimum of all such M ) is the number
sup |µ(f )| , which is denoted kµk .
f ∈K (X), kf k61
P
∞
Explicitly, kµk = |µ(ϕn )| , where ϕn is the characteristic function
n=0
P
∞
of {n} . The details are as follows. If f ∈ K (X) , say f = an ϕn
n=0
P
∞
(a finite sum), then µ(f ) = an µ(ϕn ), thus
n=0
INT III.x17 measures §1

 ∞ 
X 
 an µ(ϕn ) = |µ(f )| 6 kµ k · kf k = kµk · sup |an | .
 n∈N
n=0

Writing cn = µ(ϕn ) for all n , we have

X 
 ∞ 
 an cn  6 kµk · sup |an |
 n∈N
n=0

for all complex sequences (an ) that terminate in zeros. For every n ∈ N
choose a complex number an such that |an | = 1 and an cn = |cn | . By the
foregoing, applied to the sequence a 0 , a1 , a2 , . . . , an , 0, 0, 0, . . . ,
n X 
X  n 
|ck | =  ak ck  6 kµk · 1
k=0 k=0

P
∞
for every n ∈ N , thus the series ck is absolutely convergent with
n=0
P
∞
|cn | 6 kµk . On the other hand, for every finite sequence a 0 , a1 , . . . , an
n=0
of complex numbers, we have
X  X X  X 
 n  n n ∞
 
ak ck  6 |ak | · |ck | 6 |ck | · max |ak | 6 |ck | · max |ak |;
 06k6n 06k6n
k=0 k=0 k=0 k=0

it follows that for every f ∈ K (X) ,

X
∞ 
|µ(f )| 6 |cn | kf k ,
n=0

P
∞
whence kµk 6 |cn | .
n=0
The set of all such measures, denoted M 1 (X) , is a linear subspace
of M (X) = K (X)∗ ; equipped with the
norm just defined, it is the dual
space of the normed space K (X), k · k and is a Banach space (TVS, III,
§3, No. 8, Cor. 2 of Prop. 12).
The foregoing associates, to each bounded measure µ on X , the se-

P
∞
quence µ(ϕn ) satisfying |µ(ϕn )| < +∞ . The set of all complex se-
n=0
P
∞
quences c = (cn )n>0 satisfying |cn | < +∞ is a linear subspace of (s)
n=0
§1 measures on a locally compact space INT III.x18

P
∞
and is a Banach space for the norm kck = |cn | ; it is denoted (`(1) )
n=0

(Banach, p. 12). The correspondence µ 7→ µ(ϕn ) is an isometric linear
mapping of M 1 (X) onto (`(1) ) . The space (`(1) ) may be regarded as the
space of functions integrable with respect to the measure on X = N that
assigns measure 1 to every set {n} , namely, the linear form λ : K (X) → C
P
∞
defined by λ(f ) = f (n) (a finite sum) for every f ∈ K (X) .
n=0
Identifying K (X) with the space (c00 ) of sequences that terminate
in 0 ’s, its completion for the sup-norm k · k may be identified with the
space (c0 ) of sequences that converge to 0 ; the Banach space (c 0 ) has the
same dual as its dense linear subspace (c 00 ) , thus

0
(c0 )0 ∼
= (c00 )0 ∼
= K (X), k · k = M 1 (X) ∼
= (`(1) )

(cf. Hewitt–Stromberg (7.13) and (14.25), or Banach pp. 66, 67). Banach
spaces whose dual is isometrically isomorphic to the space of functions in-
tegrable for some measure are called L 1 -predual spaces (cf. the book of
H.E. Lacey, The isometric theory of classical Banach spaces, Grundlehren
math. Wiss. Bd. 208, Springer–Verlag, New York, 1974).
III.17, `. 9–13.
“Proposition 10. — For every measure µ on X ,

(22) kµk = sup |µ|(f ) .

06f 61, f ∈K (X;R)

“For, taking into account the formula (12) that defines the absolute
value of a measure, the second member of (22) may be written
!
sup sup |µ(g)| = sup |µ(g)| .”
06f 61, f ∈K (X;R) |g|6f, g∈K (X;C) kgk61, g∈K (X;C)

Write α for the expression on the left (where, in the right side of (22),
|µ|(f ) has been replaced by its formula given by (12)), β for the expression
on the right (which is the definition of kµk ).
Proof that α 6 β : Suppose f ∈ K (X; R) , 0 6 f 6 1 . If g ∈ K (X; C)
and |g| 6 f , then kgk 6 kf k 6 1 , hence |µ(g)| 6 β ; varying g —and
then f —yields α 6 β .
Proof that β 6 α : Suppose g ∈ K (X; C) , kgk 6 1 ; we are to show that
|µ(g)| 6 α . Say g ∈ K (X, K; C) . Choose f ∈ K + (X) with 0 6 f 6 1 and
f = 1 on K . Evidently |g| 6 f , whence, citing (12), |µ(g)| 6 |µ|(f ) 6 α .
INT III.x19 measures §1

III.18, `. 1–4.
“Corollary 3. — For every real measure µ on a locally compact
space X,

(24) kµk = sup |µ(f )| .

kf k61, f ∈K (X;R)

“It suffices to make use of the formula (22) and the expression (9) for
|µ|(f ) when µ is a real measure and f ∈ K + (X) .”
Citing (22), then (9),
!
kµk = sup |µ|(f ) = sup sup µ(g) ;
f ∈K (X;R), 06f 61 f ∈K (X;R), 06f 61 g∈K (X;R), |g|6f

the g ’s involved on the right side of this equality are precisely the functions
g ∈ K (X; R) such that kgk 6 1 , thus the right side may be rewritten as
sup µ(g) , which equals the right side of (24) (where the absolute
g∈K (X;R), kgk61
value signs are superfluous since −f satisfies the same conditions as f ).
III.18, `. 7, 8.
“The canonical injection M 1 (X, R) → M 1 (X; C) is an isometry by
virtue of (24).”
Let µ be a bounded real measure on X . The left side of (24) is kµk
as calculated in M 1 (X; C) ; the right side is the norm of µ regarded as an
element of M 1 (X, R) .
III.18, `. −3 to −1.
“Let X be a locally compact space. On the space M (X; C) , one can
consider the topology of pointwise convergence in K (X; C) , which we shall
call the vague topology on M (X; C) .”
Since M (X; C) is by definition the dual space of the locally convex
space K (X; C) (equipped with the direct limit topology),
the vague topol-
ogy is the “weak topology” σ M (X; C), K (X; C) (TVS, II, §6, No. 2,
Def. 2).
The space M (X; C) , equipped with the vague topology,
may also be
described as the locally convex space L S K (X; C); C , where K (X; C)
bears the direct limit topology and S is the set of all 1-element subsets {f }
of K (X; C) , or, equivalently, the set of all finite subsets of K (X; C) (TVS,
III, §3, No. 1, p. TVS III.13, `. −3 to −1; recall that every finite subset of a
topological vector space is bounded).
§1 measures on a locally compact space INT III.x20

III.19, `. 3–6.
“To say that a filter F on M (X; C) converges vaguely to a measure µ 0
signifies that
µ0 (f ) = limµ, F µ(f )
for every function f ∈ K (X; R).”
This is a property of initial topologies (GT, I, §7, No. 6, Prop. 10).
Incidentally, a filter F on M (X; C) is vaguely Cauchy (i.e., Cauchy for
the uniformity associated with the vague topology) if and only if for every
f ∈ K (X; C) (or every f ∈ K (X; R) ), the filter base

F(f ) = {V(f ) : V ∈ F }

(where V(f ) = {µ(f ) : µ ∈ V } ) is Cauchy (i.e., convergent) in C ; this is

a property of initial uniform structures (GT, II, §3, No. 1, Prop. 4).
III.19, `. 6–7.
“For every function f ∈ K (X; C), the mapping µ 7→ µ(f ) is a vaguely
continuous linear form on the space M (X; C).”
And these are the only vaguely continuous linear forms on M (X; C)
(TVS, II, §6, No. 2, Prop. 3).
III.20, `. 13.
“It is clear that d) implies a).”
Suppose H satisfies d). Given any function f ∈ K (X; C) , we assert
that the set H(f ) = {µ(f ) : µ ∈ H } is bounded in C ; for, there exists a
compact set K such that f ∈ K (X, K; C) , and choosing M K as in d), we
have in particular |µ(f )| 6 MK kf k for all µ ∈ H .
Thus the implication d) ⇒ a) is a consequence of the following propo-
sition: a subset H of M (X; C) is vaguely bounded (i.e., bounded for the
vague topology) if and only if, for every f ∈ K (X; C) , the set H(f ) =
{µ(f ) : µ ∈ H } is bounded in C . The details are as follows.
Suppose H is vaguely bounded and let f ∈ H . Since the set

V = {µ ∈ M (X; C) : |µ(f )| 6 1 }

is a vague neighborhood of 0 , there exists a scalar a 6= 0 such that H ⊂ aV

(TVS, III, §1, No. 2, remarks following Def. 3), thus the set {|µ(f )| : µ ∈ H }
is bounded above by |a| .
Conversely, suppose H has the stated property. Given a vague neigh-
borhood V of 0 in M (X; C) , we seek a scalar a 6= 0 such that H ⊂ aV .
We can suppose that

V = {µ ∈ M (X; C) : |µ(fi )| 6 εi for i = 1, . . . , n }

INT III.x21 measures §1

where fi ∈ K (X; C) and εi > 0 (i = 1, . . . , n) , since such sets V form

a fundamental system of neighborhoods of 0 for the vague topology. By
assumption, for each i , the set

Si = {µ(fi ) : µ ∈ H }

is bounded in C , hence so is the set S = S 1 ∪. . .∪Sn , say |µ(fi )| 6 M < +∞

for all µ ∈ H and for i = 1, . . . , n . If a > 0 is so chosen that a −1 M 6 εi
for i = 1, . . . , n , then, for all µ ∈ H and all i = 1, . . . , n ,

a−1 |µ(fi )| 6 a−1 M 6 εi

thus a−1 µ ∈ V for all µ ∈ H , that is, H ⊂ aV .

III.20, `. 13–15.
“Finally, if H is equicontinuous then the set of restrictions of the mea-
sures µ ∈ H to K (X, K; C) is also equicontinuous, . . .”
The point is that the relative topology on K (X, K; C) induced by
the direct limit topology of K (X; C) is equal to the norm topology on
K (X, K; C) , by No. 1, Prop. 1, (i).
III.20, `. 19, 20.
“Corollary 1. — Let ν be a positive measure on X ; the set of mea-
sures µ such that |µ| 6 ν is vaguely compact.”
Let H = {µ ∈ M (X; C) : |µ| 6 ν } ; it suffices to show that H is
(i) vaguely relatively compact, and (ii) vaguely closed.
Proof of (i). We verify criterion d) of Prop. 15. Let K be a compact
subset of X . Since ν is a measure, there exists a finite constant M K > 0
such that

(∗) |ν(f )| 6 MK kf k for all f ∈ K (X, K; C) .

Then, for every µ ∈ H and f ∈ K (X, K; C) we have

|µ(f )| 6 |µ|(|f |) 6 ν(|f |) 6 MK · k |f | k = MK kf k

by (13), the definition of H , and the relation (*), thus the criterion d) is
satisfied.
Proof of (ii). Let (µj ) be a directed family in H with µj → µ in
M (X; C) vaguely; we are to show that µ ∈ H . Thus, assuming f ∈ K + (X) ,
it is to be shown that |µ|(f ) 6 ν(f ) . Given any g ∈ K (X; C) such that
|g| 6 f , it will suffice by (12) to show that |µ(g)| 6 ν(f ) . For each index j ,
we have
|µj (g)| 6 |µj |(|g|) 6 ν(|g|) 6 ν(f )
§1 measures on a locally compact space INT III.x22

by the relations (13), µj ∈ H and |g| 6 f ; since µj (g) → µ(g) by the

definition of the vague topology, it follows that |µ(g)| 6 ν(f ) .
III.20, `. 21, 22.
“Corollary 2. — The set of measures µ such that kµk 6 a ( a any
finite number > 0 ) is vaguely compact.”
Proof #1. We are looking at a closed ball B in the dual M 1 (X; C) of
a normed space (the space K (X; C) equipped with the topology of uniform


convergence in X ) equipped with the topology σ M 1 (X; C), K (X; C) , for
which B is known to be compact

(TVS, III, §3, No. 4, Cor. 3 of Prop. 4).
Since σ M 1 (X; C), K (X; C) is the topology on M 1 (X; C) induced by the
vague topology on M (X; C) , B is also a vaguely compact subset of M (X; C) .
Proof #2. Write H = {µ ∈ M (X; C) : kµk 6 a } (thus H ⊂
M 1 (X; C) ). We verify compactness by showing that (i) H is vaguely rela-
tively compact, and (ii) H is vaguely closed.
(i) For all µ ∈ H and f ∈ K (X; C) we have (since µ ∈ M 1 (X; C) )

|µ(f )| 6 kµk kf k 6 akf k ;

thus H(f ) = {µ(f ) : µ ∈ H } is a bounded set in C for every f ∈ K (X; C) ,

therefore H is vaguely bounded (see the notes above for `. 13), hence is
vaguely relatively compact by Prop. 15.
(ii) Suppose (µj ) is a directed family in H with µj → µ ∈ M (X; C)
vaguely. For every f ∈ K (X; C) and every index j , we have

|µj (f )| 6 kµj k kf k 6 akf k ;

since µj (f ) → µ(f ) , we infer that |µ(f )| 6 akf k , and since f ∈ K (X; C)

is arbitrary, we conclude that µ is bounded and kµk 6 a , that is, µ ∈ H .
III.20, `. −14, −13.
“Corollary 3. — If X is compact, the set of positive measures µ
on X such that kµk = 1 is vaguely compact.”
Write H = {µ ∈ M (X; C) : µ > 0 and kµk = 1 } for the set of
measures in question. By No. 8, Cor. 2 of Prop. 10,

H = {µ ∈ M (X; C) : µ > 0 and µ(1) = 1 }

= M+ (X) ∩ {µ ∈ M (X; C) : µ(1) = 1 } .

The first factor of the intersection is vaguely closed as noted in Prop. 14,
and the second factor, being the inverse image of {1} under the vaguely
continuous mapping µ 7→ µ(1) , is also vaguely closed, thus H is vaguely
INT III.x23 measures §1

closed; being a subset of the vaguely compact set {µ ∈ M (X; C) : kµk 6 1 }

(Cor. 2), H is also vaguely compact.
III.20, `. −9, −8.
“Corollary 4. — In the space M (X; C) , the mapping µ 7→ kµk is
lower semi-continuous for the vague topology.”
For a ∈ R , the set {µ ∈ M (X; C) : kµk 6 a } is vaguely compact
if a > 0 (Cor. 2) and empty if a < 0 , and in either case is vaguely closed,
hence the mapping M (X; C) → R defined by µ 7→ kµk is vaguely lower
semi-continuous (GT, IV, §6, No. 2, Prop. 1).
III.21, `. −15.
“Each of these topologies is coarser than the next.”
Let
S1 = all singletons {f } , f ∈ T ;
S2 = all singletons {f } , f ∈ K (X; C) ;
S3 = all strictly compact subsets of K (X; C) ;
S4 = all compact subsets of K (X; C) .
The inclusions S1 ⊂ S2 and S3 ⊂ S4 are obvious, and S2 ⊂ S3 because
each f ∈ K (X; C) belongs to K (X, K; C) for some compact set K , whence
the strict compactness of {f } . Thus S 1 ⊂ S2 ⊂ S3 ⊂ S4 , whence clearly
T1 ⊂ T2 ⊂ T3 ⊂ T4 (convergence with respect to Ti+1 implies convergence
with respect to Ti (i = 1, 2, 3) ).
Note that every set in S4 is bounded in K (X; C) , therefore Ti is the
Si -topology on M (X; C) for every i (TVS, III, §3, No. 7).
III.21, `. −11 to −9.
“A vaguely bounded subset H of M (X; C) is equicontinuous (No. 9,
Prop. 15), thus the first assertion follows from TVS, III, §3, No. 7, Prop. 9,
and the second from GT, X, §2, No. 4, Th. 1.”
Proof of (i). Since i < j ⇒ Ti ⊂ Tj (1 6 i, j 6 4) , it is clear that
every set bounded for Tj is bounded for Ti ; thus, to prove that the bounded
sets for T2 , T3 , T4 are the same, it suffices to show that every set H that
is bounded for T2 is bounded for T4 . Let H be a vaguely bounded set.
Since H is equicontinuous (Prop. 15) it is bounded for every S-topology
(TVS, III, §3, No. 7, Prop. 9) and in particular for S 4 .
Proof of (ii). Let H be a vaguely bounded subset of M (X; C) ; we are
to show that the topologies Ti ∩ H induced on H by Ti (i = 1, 2, 3, 4) are
identical. For notational purposes let us write X = K (X; C) (equipped
with the direct limit topology) and Y = C (equipped with the usual uni-
formity). We know that
H ⊂ M (X; C) ⊂ C (X ; Y ) ;
§1 measures on a locally compact space INT III.x24

since H is an equicontinuous subset of C (X ; Y ) by Prop. 15, it follows

that on H , the topology of pointwise convergence in the total subset T of
X = K (X; C) (equivalently, in its linear span, which is dense) coincides
with the topology of compact convergence (GT, X, §2, No. 4, Th. 1), that
is, T1 ∩ H = T4 ∩ H , whence the coincidence of all four topologies.
III.21, `. −8 to −6.
“Recall that when X is paracompact, the topology of strictly compact
convergence coincides with the topology of compact convergence (No. 1,
Prop. 2).”
Because, for such a space X , every compact subset of K (X; C) , being
bounded, is strictly compact, i.e., is a compact subset of K (X, K; C) for
some compact set K in X (No. 1, Prop. 2, (ii)).
III.22, `. 3–6.
“Since every filter is the intersection of the ultrafilters finer than it (GT,
I, §6, No. 4, Prop. 7), it suffices to show that if U is an ultrafilter on M + (X)
that converges to a measure µ0 for the topology T1 , then it also converges
to µ0 for T3 .”
Clearly T1 ⊂ T2 ⊂ T3 , so we need only show that the topology
on M+ (X) induced by T1 is finer than that induced by T3 , i.e., that
T1 ∩ M+ (X) ⊃ T3 ∩ M+ (X) . Thus, given a filter G on M+ (X) that is
convergent to µ0 ∈ M+ (X) for T1 ∩ M+ (X) (in other words, G contains
the filter of neighborhoods of µ0 for T1 ∩ M+ (X) ), we are to show that
G → µ0 for the topology T3 ∩ M+ (X) . If U is any ultrafilter on M+ (X)
such that U ⊃ G , obviously U → µ0 for T1 ∩ M+ (X) ; since G is the inter-
section of all such U , it will suffice to show that U → µ 0 for T3 ∩ M+ (X)
(GT, I, §7, No. 1, Prop. 2).
III.22, `. 6.
“Let K be a compact subset of X . . .”
We are to show that U → µ0 uniformly on each compact subset of
K (X, K; C) .
III.22, `. 6, 7.
“. . .by hypothesis, there exists a function h ∈ V that is > 0 on X and
takes values > 0 on K ”
Namely, the hypothesis that V satisfies the condition (P) of No. 7,
Prop. 9.
III.22, `. 8, 9.
“. . .it follows that every function f ∈ K (X, K; C) may be written
f = gh with g ∈ K (X, K; C) ”
INT III.x25 measures §1

Let U = {x ∈ X : h(x) > 0 } ; in particular, K ⊂ U . Define g : X → C

by the formulas

 f (x) for x ∈ U
g(x) = h(x)

0 for x ∈ X --- K ;

g is defined unambiguously, since on U ∩ (X --- K) ⊂ X --- K one has also

f (x) = 0 —which also establishes the formula f = gh . Since U and X --- K
are open sets with union X , on each of which g has a continuous restriction,
it follows that g is continuous on X (GT, I, §3, No. 2, Prop. 4), whence
clearly g ∈ K (X, K; C) .
III.22, `. 9.
“. . . c = inf h(x) > 0 . . .”
x∈K
Since K is compact and h > 0 on K , c > 0 by the theorem of
Weierstrass (GT, IV, §6, No. 1, Th. 1).
III.22, `. 9–11.
“By hypothesis, there exists a set H 0 ∈ U such that, for every measure
µ ∈ H0 ,
0 6 µ(h) 6 µ0 (h) + 1 = b .”

Recall that h ∈ V and that U → µ0 pointwise in V .

III.22, `. 15, 16.
“If U0 is the ultrafilter induced by U on H0 ”
Namely U0 = UH0 = {H ∩ H0 : H ∈ U } (GT, I, §6, No. 5, Prop. 9).
III.22, `. 16, 17.
“. . .the image of U0 under the mapping µ 7→ h · µ is the base of an
ultrafilter F on H ”
See GT, I, §6, No. 6, Prop. 10.
III.22, `. 17, 18.
“. . .and since H is relatively compact for the topology of strictly com-
pact convergence (Prop. 17 and No. 9, Prop. 15)”
Being vaguely bounded, H is vaguely relatively compact by Prop. 15.
Thus, if H is the vague closure of H in M (X; C) (since M + (X) is vaguely
closed in M (X; C) by Prop. 14, H is also the closure of H in M + (X) for
the topology induced by the vague topology), we know that H is vaguely
compact, hence also vaguely bounded. Therefore by Prop. 17, (ii), the topol-
ogy on H induced by the topology of strictly compact convergence coincides
with the topology induced by the vague topology; since the latter is compact,
§1 measures on a locally compact space INT III.x26

so is the former, i.e., H is compact for the topology induced by T 3 , so its

subset H is relatively compact for that topology.
III.22, `. 18, 19.
“. . . F is convergent to a measure ν0 for this topology.”
As in the preceding note, let H be the vague closure of H in M (X; C) ,
which we know to be compact for the topology induced by T 3 ; in what
follows, it is understood that H bears this topology.
The filter F on H is a filter base on H ; by the compactness of H ,
F admits a cluster point ν0 ∈ H (GT, I, §9, No. 3, Prop. 7), hence there
exists a filter G on H such that G ⊃ F and G → ν 0 (GT, I, §7, No. 2,
Prop. 4).
On the way to proving that F → ν0 in H , we observe that if G ∈ G and
F ∈ F then G ∩ F ∈ F ; for, for every F0 ∈ F the intersection (G ∩ F) ∩ F0 is
nonempty (it is an element of G ), thus G∩F is a subset of H that intersects
every set in the ultrafilter F , whence G ∩ F ∈ F (because {G ∩ F} ∪ F is a
base of a filter on H that, by the maximality of F , must coincide with F ).
Now let U be any neighborhood of ν0 in H ; we are to show that U
contains some set in F . Since G → ν0 , there exists a set G ∈ G such
that G ⊂ U ; choose any set F ∈ F ; then G ∩ F ∈ F by the foregoing, and
G ∩F ⊂ G ⊂ U.
III.22, `. 19–22.
“In other words, for any ε > 0 and any compact subset L of K (X, K; C),
there exists a subset N of H0 belonging to U such that, for every func-
tion g ∈ L and every pair of measures µ, µ 0 belonging to N , one has
|hg, h · µi − hg, h · µ0 i| 6 ε ”
With notations and assumptions as in the preceding Note, the sets
N ∈ U such that N ⊂ H0 are precisely the sets of U0 , thus the sets
h·N = {h·µ : µ ∈ N } form the base of a filter F on H , and F , regarded as
the base of a filter on H , is convergent (to ν 0 ) for the topology on H induced
by T3 . The set V = {(ν, ν 0 ) ∈ H × H : |ν(g) − ν 0 (g)| 6 ε for all g ∈ L }
is an entourage for the uniformity associated with that topology; since
F is Cauchy for this uniformity, there exists a set h · N ∈ F such that
h · N × h · N ⊂ V , whence the assertion.
III.22, `. 24, 25.
“Now, we saw above that the mapping g 7→ gh is an automorphism of
the Banach space K (X, K; C) .”
From kghk 6 kgk khk we see that the (linear) mapping is continuous;
the f = gh factorization property proved above shows that it is surjective,
and the relation kgk 6 c−1 kf k shows that it is injective, with continuous
inverse mapping.
INT III.x27 measures §2

III.22, `. 25, 26.

“We have thus shown that U is a Cauchy filter on M + (X) for the
topology of strictly compact convergence.”
Given any ε > 0 and any compact subset L 0 of K (X, K; C) . By the
preceding remark, there exists a compact subset L of K (X, K; C) such that
g · L = L0 . Then choose N as in `. 19–23; the inequality there shows that
|hf, µi − hf, µ0 i| 6 ε for all f ∈ L0 = g · L and all µ, µ0 ∈ N , which is the
desired Cauchy condition.
III.22, `. 26, 27.
“A fortiori, it is a Cauchy filter for vague convergence”
Because T3 ⊃ T2 .
III.22, `. −5, −4.
“ . . . moreover, since V is dense in K (X; C) , the hypothesis implies
that µ1 = µ0 ”
For each f ∈ V , we have U(f ) → µ0 (f ) in C because U → µ0 for T1 ;
but U(f ) → µ1 (f ) in C because U → µ1 vaguely, whence µ0 (f ) = µ1 (f ) .
Thus µ0 = µ1 on V ; since V is dense in K (X; C) , µ0 = µ1 as claimed.
III.23, `. 1–3.
“Corollary. — If X is paracompact then the topologies induced on
M+ (X) by the vague topology and the topology of compact convergence co-
incide.”
Since X is paracompact, T3 = T4 (in the notation of Prop. 17), thus
T2 ⊂ T3 = T4 . But T2 ∩ M+ (X) = T3 ∩ M+ (X) by Prop. 18. Thus
T2 ∩ M+ (X) = T4 ∩ M+ (X) as claimed.

§2. SUPPORT OF A MEASURE

III.23, `. 11–13.
“. . .every continuous function with values in a topological vector space E ,
defined on Y and with compact support, may be extended by continuity to
all of X , by giving it the value 0 on { Y ”
Suppose g ∈ K (Y; E) and let f be the extension by 0 of g to X ,
that is, (
g(x) for x ∈ Y
f (x) =
0 for x ∈ X --- Y .
§2 support of a measure INT III.x28

Say g ∈ K (Y, K; E) where K is a compact subset of Y (hence also of X ).

Since g = 0 on Y --- K , we have also
(
g(x) for x ∈ Y
f (x) =
0 for x ∈ X --- K ,

where Y and X --- K form an open covering of X ; since the restrictions

of f to Y and to X --- K are continuous, f is continuous on X (GT, I, §3,
No. 2, Prop. 4) as claimed, and f ∈ K (X, K; E) ⊂ K (X; E) .
III.23, `. 14–16.
“. . .one can therefore identify in this way the space K (Y; E) with
the linear subspace of K (X; E) formed by the continuous functions with
compact support contained in Y .
Caution: If f ∈ K (X; E) and f = 0 on X --- Y , it does not follow
that f belongs to the subspace K (Y; E) just constructed, i.e., it does not
follow that Supp f ⊂ Y . For example, if f ∈ K (X; E) and if the (open) set
Y = {x ∈ X : f (x) 6= 0 } is not closed, then f ∈
/ K (Y; E) ; for, the support
of f is Y , and Y 6⊂ Y .
III.23, `. 16–19.
“If µ is a measure on X , it is clear that the restriction of µ to K (Y; C)
is a measure on Y , which is called the restriction of µ to the open sub-
space Y , or the measure induced on Y by µ , and is denoted µ|Y .”
Proof #1: If K is a compact subset of Y , then K is also compact
in X and the restriction of µ to K (X, K; C) is known to be continuous for
the topology of uniform convergence in X —in other words in K —or in Y ,
it comes to the same thing since K ⊂ Y ⊂ X and the functions in question
are 0 outside of K .
Proof #2: Applying Th. 2 of §1, No. 5 to the real and imaginary parts
of µ , we see that µ is a linear combination of four positive measures on X ,
therefore µK (Y; C) is a linear combination of four linear forms whose
restrictions to K (Y; R) are positive linear forms and so define real measures
on Y (loc. cit., Th. 1).
The question naturally arises, when E is locally convex does the topol-
ogy τ on K (Y; E) induced by the direct limit topology τ X on K (X; E)
coincide with the direct limit topology τ Y on K (Y; E) ? (Probably the
answer is known, but it is not known to me.)
At any rate, it is easy to see that τY ⊃ τ . For, let


i : K (Y; E) → K (X; E), τX
INT III.x29 measures §2

be the mapping i(g) = g 0 , where g 0 is the extension by 0 of g to X ;

i is the insertion mapping for identifying K (Y; E) as a linear subspace of
K (X; E) , and τ is the initial topology for i . For any compact subset K
of Y , consider the insertion mapping


iK : K (Y, K; E) → K (Y; E), τY ,
where K (Y, K; E) bears the topology of uniform convergence in Y (equiv-
alently in K )—which is also the topology on K (Y, K; E) induced by τ Y
(§1, No. 1, Prop.

1, (i)). The composite mapping i ◦ i K : K (Y, K; E) →
K (X; E), τX ,



K (Y, K; E) → K (Y; E), τY → K (X; E), τX ,
is continuous for the indicated topologies on K (Y, K; E) and K (X; E) : if
gj , g ∈ K (Y, K; E) with gj → g uniformly in K (equivalently in Y ) then
gj0 → g 0 uniformly in X (equivalently in K ), hence g j0 → g 0 in K (X, K; E)


and therefore in K (X; E), τX , whence the asserted continuity. Since this
is true for every compact set K ⊂ Y , i is continuous because τ Y is a direct
limit topology, i.e., it is the locally convex final topology for the family
of mappings iK (TVS, II, §4, No. 4, Example II). But τ is the coarsest
topology on K (Y; E) rendering i continuous, therefore τ Y ⊃ τ .
This inclusion is sufficient for proving that if µ : K (X; E) → F is
a continuous mapping of K (X; E) into a topological space F , then the
restriction µK (Y; E) = µ ◦ i ,
K (Y; E) → K (X; E) → F ,
is continuous; for, it is certainly continuous when K (Y; E) bears the initial
topology τ for i , hence a fortiori when K (Y; E) bears the (finer) direct
limit topology τY .
In particular, if µ is a continuous linear mapping of K (X; E) into a
topological vector space F , then its restriction to K (Y; E) is also continu-
ous. The special case E = F = C yields a third proof of the initial assertion
about measures.
III.23, `. 19, 20.  
“The restrictions to Y of |µ| , Rµ and I µ are, respectively, µ|Y,
R(µ|Y) and I (µ|Y) , by virtue of §1, Nos. 5 and 6.”

Write ν = µY . As in §1, No. 5, let µ be the conjugate of µ , that
is, µ(f ) = µ(f ) for all f ∈ K (X; C) .  In particular, µ(g) = µ(g) for all
g ∈ K (Y; C) , whence it is clear that µY = ν . From
1 1
R(µ) = (µ + µ) , I (µ) = (µ − µ) ,
2 2i
§2 support of a measure INT III.x30

restriction to K (Y; C) yields

 1  1
R(µ)Y = (ν + ν) = R(ν) , I (µ)Y = (ν − ν) = I (ν) .
2 2i
If f ∈ K+ (X) then by definition (§1, No. 6, formula (12))

(∗) |µ|(f ) = sup |µ(g)| .

|g|6f, g∈K (X;C)

If in particular f ∈ K+ (Y) then |g| 6 f implies that Supp g ⊂ Supp f ⊂ Y,

thus the right side of (*) may be written

sup |ν(g)| ,
|g|6f, g∈K (Y;C)


which is the formula for |ν|(f ) ; thus it follows from (*) that |µ| Y = |ν| on
K+ (Y) , hence on K (Y; C) .
III.23, `. 20–22.
“If µ is real then the restrictions of µ + and µ− to Y are, respectively,
(µ|Y)+ and (µ|Y)− , by virtue of formula (8) of § 1, No. 5.”
From the formulas (INT II.2)

1 1
µ+ = (|µ| + µ) , µ− = (|µ| − µ) ,
2 2

restriction to K (Y; C) yields (with ν = µ Y as in the preceding Note)
 1  1
µ+ Y = (|ν| + ν) = ν + , µ− Y = (|ν| − ν) = ν − .
2 2

More generally, if µ 1 , µ 2 are real measures on X , and if ν 1 = µ 1
Y , ν 2 =

µ2 Y , then the formulas (INT II.2)

1
sup(µ1 , µ2 ) = (µ1 + µ2 + |µ1 − µ2 |)
2
1
inf(µ1 , µ2 ) = (µ1 + µ2 − |µ1 − µ2 |)
2
yield, on restriction to K (Y; C) , the formulas
 1
sup(µ1 , µ2 )Y = (ν1 + ν2 + |ν1 − ν2 |) = sup(ν1 , ν2 )
2
 1
inf(µ1 , µ2 )Y = (ν1 + ν2 − |ν1 − ν2 |) = inf(ν1 , ν2 ) .
2
INT III.x31 measures §2

III.23, `. 23–26.
“One sees immediately
 that if Y and Z are two open sets in X such
that Y ⊃ Z , and if µY and µZ are the restrictions of µ to Y and Z ,
then µZ is also the restriction of µY to the open subspace Z of the locally
compact space Y .”
If h ∈ K (Z; C) and h* ∈ K (Y, Z; C) is the extension by 0 of h
0
to Y , and if h* ∈ K (X, Y; C) is the extension by 0 of h* to X , then in
0
fact h* ∈ K (X, Z; C) is the extension by 0 of h to X , and so
 0
  

(µZ)(h) = µ(h* ) = (µY)(h*) = (µY)Z (h) ,

whence the assertion.

Writing µY = µY (as in Ch. IV, §5, No. 7, Def. 4), the above assertion
is that (µY )Z = µZ when Y, Z are open sets in X with Y ⊃ Z . It follows
that if Y, Z are arbitrary open sets in X , then

(µY )Y∩Z = µY∩Z = (µZ )Y∩Z .

III.23, `. −2, −1, III.24, `. 1.

“. . .the mapping
Z 1
f (x)
f 7→ dx
0 x
is a measure on Y ”
If K is a nonempty compact subset of the topological subspace Y =
]0, 1[ of R then K has a smallest element a and a largest b (apply
the theorem of Weierstrass to the canonical injection of K into R ), thus
0 < a 6 x 6 b < 1 and 1 < 1/b 6 1/x 6 1/a for all x ∈ K . For every
f ∈ K (Y, K; C) , the function x 7→ f (x)/x (x ∈ Y) is equal to 0 outside
[a, b] , continuous and bounded, hence integrable on ]0, 1[ in the sense of
FRV; writing
Z 1 Z b
f (x) f (x)
ν(f ) = dx = dx
0 x a x
b−a
and M = , we have
a
Z b
1 1
|ν(f )| 6 |f (x)| · dx 6 kf k · · (b − a) = Mkf k
a x a

for all f ∈ K (Y, K; C). Thus f 7→ ν(f ), obviously a linear form on K (Y; C),
is a measure on Y (§1, No. 3, (4)).
§2 support of a measure INT III.x32

Note that since Y is the union of a sequence of compact intervals, it is

trivially ν-moderated, that is, ν is a moderated measure (Ch. V, §1, No. 2,
Def. 2), hence the concepts “ν-integrable” and “essentially ν-integrable”
coincide (loc. cit., No. 3, Cor. of Prop. 9).
III.24, `. 2–4.
“However, this measure cannot be extended to a measure on R because,
in the contrary case, its restriction to the set of functions f ∈ K (Y; C)
such that kf k 6 1 would be bounded”
To put the matter positively, if a measure ρ on the topological subspace
Y = ]0, 1[ of R can be extended to a measure µ on R , then ρ is bounded.
For, as noted above, Y is ρ-moderated and the concepts “ ρ-integrable” and
“essentially ρ-integrable” coincide. Since [0, 1] is compact, [0, 1] ∩ Y = Y
is ρ-integrable by Ch. V, §4, No. 2, Prop. 3) (specifically, the implication
d) ⇒ b)), therefore ρ is bounded (Ch. IV, §4, No. 7, Prop. 12). See the
next remark for the rest of the argument.
III.24, `. 4.
“. . .but this is false.”
In view of the preceding remark, to show that ν cannot be extended
to a measure on R we need only observe that ν is not a bounded measure.
To this end, for n = 2, 3, 4, . . . define f n ∈ K (Y) to be the piecewise linear
continuous function for which
 1
 0 on ]0, 2n ]

fn (x) = 1 on [ n1 , 1 − n1 ]

 1
0 on [1 − 2n , 1[ .

Then the sequence

Z 1 Z 1−1/n
fn (x) 1
ν(fn ) = dx > dx
0 x 1/n x

is unbounded, whereas kfn k = 1 for all n .

III.25, `. 1.
“whence. . .”
In the preceding display (III.24, `. −1) all terms are in K (Y αi ; C) ,
hence one can apply µαi then sum over i .
III.25, `. 8–10.
“. . .the conclusion therefore follows at once from the definition of µ and
from Prop. 6 of §1, No. 3.”
INT III.x33 measures §2

Let x ∈ X . By assumption there exists an index α x ∈ A such that

x ∈ Yαx . Let Kx be a compact neighborhood of x such that K x ⊂ Yαx .
Varying x , we obtain a family (Kx )x∈X whose interiors are a covering of X ,
and such that Kx ⊂ Yαx for all x ∈ X . Since Kx ⊂ Yαx and since the
restriction µK (Yαx ; C) = µαx is by hypothesis a measure on Yαx , there
exists a constant Mx such that

(∗) |µ(f )| = |µαx (f )| 6 Mx kf k for all f ∈ K (Yαx , Kx ; C)

(§1, No. 3, (4)); note that we are concerned here only with the norm topology
on K (Yαx , Kx ; C) —the direct limit topology on K (Y αx ; C) that induces
it is not at issue here. If f ∈ K (X, Kx ; C) , then Supp f ⊂ Kx ⊂ Yαx ,
so that f can be regarded as an element of K (Y αx , Kx ; C) , and hence
satisfies (*). Thus (*) may be written

(∗∗) |µ(f )| 6 Mx kf k for all f ∈ K (X, Kx ; C) .

Summarizing, we have a family (Kx )x∈X of compact subsets of X whose

interiors form a covering of X , µ is a linear form on K (X; C) , and for
each x ∈ X there exists a constant Mx satisfying (**), consequently µ is a
measure on X by Prop. 6 of §1, No. 3.
III.25, `. 11–14.
“Corollary (Principle of localization). — Let µ and ν be two mea-
sures on X , and let (Yα ) be a family of open sets of X such that, for
every α , the restrictions
S to Yα of µ and ν are equal; then the restrictions
of µ and ν to Y = Yα are equal.”
α

As in the note for III.23, `. 23–26, let us write µ Y for µY . Then (loc.
cit.) for every α ,

(µY )Yα = µYα = νYα = (νY )Yα ;

thus, dropping down from X to Y , we can suppose that Y = X .

For every index α write µα = µYα and να = νYα . Then (loc. cit.) for
every pair (α, β) of indices, we have

(µα )Yα ∩Yβ = (µYα )Yα ∩Yβ = µYα ∩Yβ = (µYβ )Yα ∩Yβ = (µβ )Yα ∩Yβ ;

it follows from Prop. 1 that µ is the unique measure on X such that µ Yα =
µα for all α . Similarly, ν is the unique measure on X such that ν Yα = να
for all α . But  
µα = µ Yα = ν Yα = ν α
§2 support of a measure INT III.x34

for all α , whence µ = ν by uniqueness.

III.25, `. −9 to −7.
“To say that a point x ∈ X does not belong to the support of µ means
that there exists an open neighborhood V of x such that the restriction of µ
to V is zero”
With the notations preceding Def. 1, if x ∈
/ Supp(µ) , that is, if
x ∈ U0 = { Supp(µ) , then V = U0 meets the requirement. Conversely,
if there exists an open neighborhood V of x such that V ∈ G then
x ∈ V ⊂ U0 = { Supp(µ) , so x ∈
/ Supp(µ) .
III.25, `. −7 to −5.
“to say that x belongs to the support of µ therefore means that for
every neighborhood V of x , there exists a function f ∈ K (X; C) , whose
support is contained in V , such that µ(f ) 6= 0 .”
This is essentially the contrapositive form of the preceding assertion;
marching through the details requires a review of some delicate definitions.
Suppose x ∈ Supp(µ) and let V be any neighborhood of x ; let U be
an open set such that x ∈ U ⊂ V . Since x ∈ / { Supp(µ) = U0 and x ∈ U ,
necessarily U 6⊂ U0 ; thus U ∈ / G , that is, µU 6= 0 , hence there exists a
function g ∈ K (U; C) such that µU (g) 6= 0 ; if f = g 0 is the extension
by 0 of g to X , then f ∈ K (X; C) , Supp f = Supp g ⊂ U ⊂ V , and
µ(f ) = µU (g) 6= 0 .
Conversely, suppose that for each neighborhood V of x , there exists a
function f ∈ K (X; C) such that Supp f ⊂ V and µ(f ) 6= 0 ; we assert that
x ∈ Supp(µ) . Assuming to the contrary that x ∈ { Supp(µ) = U0 , then
U0 is an open neighborhood of x with µU0 = 0 , therefore µU0 (g) = 0 for
all g ∈ K (U0 ;C) . For every f ∈ K (X; C) such that Supp f ⊂ U 0 , the
function g = f U0 belongs to K (U0 ; C) and f is the extension by 0 of g
to X , therefore µ(f ) = µU0 (g) = 0 ; thus the original assumption on the
point x is contradicted by its neighborhood V = U 0 .
III.26, `. 11, 12.
“For, if the restriction of µ to an open set U is zero, then so is that of
|µ| (resp. of µ+ and µ− when µ is real), and conversely.”
Let U be an open set in X . Since |µU | = |µ|U (No. 1) one has

µU = 0 ⇔ |µU | = 0 ⇔ |µ|U = 0 ,

whence Supp(µ) = Supp(|µ|) . If, moreover, µ is real, then

|µU | = |µ|U = (µ+ + µ− )U = (µ+ )U + (µ− )U ,

INT III.x35 measures §2

therefore, by posititivy,

µU = 0 ⇔ (µ+ )U = 0 & (µ− )U = 0 ,

thus
U ⊂ { Supp(µ) ⇔ U ⊂ { Supp(µ+ ) ∩ { Supp(µ− )


⇔ U ⊂ { Supp(µ+ ) ∪ Supp(µ−) ;

letting { Supp(µ) play the role of U yields the inclusion

{ Supp(µ) ⊂ { Supp(µ+) ∪ Supp(µ− ) ,


and letting { Supp(µ+) ∪ Supp(µ−) play the role of U yields the reverse
inclusion, whence equality.
III.26, `. −11.
“The proposition is obvious from the definitions.”
The set U --- Supp(µU) is an open set in U (hence in X ) such that

0 = (µU )U -- Supp(µU ) = µU -- Supp(µU ) ,

therefore U -- Supp(µU) ⊂ X − Supp(µ) , whence

U -- Supp(µU) ⊂ U − U ∩ Supp(µ) ,

and so Supp(µU) ⊃ U ∩ Supp(µ) .

On the other hand, let V = U --- Supp(µ) = U ∩ X --- Supp(µ) =
U --- U∩Supp(µ) , which is open in both X and U . Since V ⊂ X --- Supp(µ) ,
we have
0 = µV = µU∩V = (µU )V ,

therefore V ⊂ U --- Supp(µU ) , that is, U --- U ∩ Supp(µ) ⊂ U --- Supp(µ U) ,

and so
U ∩ Supp(µ) ⊃ Supp(µU) ,

whence equality.
III.26, `. −7 to −5.
“For, it is the intersection of the vaguely closed hyperplanes with equa-
tion µ(f ) = 0 , where f runs over the set of functions in K (X; C) whose
support does not intersect F .”
§2 support of a measure INT III.x36

For a measure µ ∈ M (X; C) ,

Supp(µ) ⊂ F ⇔ { F ⊂ { Supp(µ)
⇔ µ{ F = 0
⇔ µ { F (g) = 0 for all g ∈ K ( { F; C)
⇔ µ(f ) = 0 for all f ∈ K (X, { F; C)
\
⇔ µ∈ {ν ∈ M (X; C) : ν(f ) = 0 } ;
f ∈K (X, { F;C)

for each f , the set {ν ∈ M (X; C) : ν(f ) = 0 } is the kernel of the vaguely
continuous linear form ν 7→ ν(f ) .

III.26, `. −4 to −1.
“Suppose X is not compact: given a filter Φ on the space M (X; C)
of measures on X , we shall say that the support of a measure µ recedes
indefinitely along Φ if, for every compact subset K of X , there exists a set
M ∈ Φ such that Supp(µ) ∩ K = ∅ for every measure µ ∈ M .”
Some thoughts on the definition: Let X 0 = X ∪ {ω} be the one-point
compactification of X (GT, I, §9, No. 8), with X viewed as an open subspace
of X0 . The open sets in X0 are the open sets U in X together with the
sets {ω} ∪ (X --- K) with K compact in X (the latter are the only open sets
in X0 that contain ω ). Thus the open neighborhoods of ω in X 0 are the
sets of the form U0 = {ω} ∪ (X --- K) for some compact subset K of X .
Thus, the concept defined above says that given any open neighborhood
V0 of ω in X0 (equivalently, given any compact subset K of X ), there
exists a set M ∈ Φ such that Supp(µ) ⊂ V 0 (resp. Supp(µ) ∩ K = ∅ ) for all
µ ∈ M . It is the same to say that for every neighborhood
S V 0 of ω in X0 ,
such an M exists. Thus, writing Supp(M) = Supp(µ) , the concept
µ∈M
says that for every neighborhood V 0 of ω in X0 , there exists a set M ∈ Φ
such that Supp(M) ⊂ V0 ; this might be expressed suggestively as

lim Supp(M) = ω
M∈Φ

or that Supp(µ) → ω with respect to the filter Φ , symbolically

lim µ,Φ Supp(µ) = ω .

At any rate, it is a property of Φ (not of Supp).

INT III.x37 measures §2

III.27, `. 6, 7.
“. . .which proves the proposition.”
To say that µ converges vaguely to a measure µ 0 with respect to a fil-
ter Φ on M (X; C) means that each vague neighborhood of µ 0 in M (X; C)
contains some set M ∈ Φ , that is, Φ ⊃ V0 , where V0 is the filter
of neighborhoods of µ0 in M (X; C) for the vague topology. Concisely,
Φ → µ0 vaguely.
An equivalent condition (see the note for III.19, `. 3–6) is that for each
f ∈ K (X; C) , one has
µ0 (f ) = lim µ,Φ µ(f ) ;
the meaning of this is that for each f ∈ K (X; C) one considers the linear
form Lf : M (X; C) → C defined by Lf (µ) = µ(f ) , and that
lim µ,Φ Lf (µ) = Lf (µ0 ) ;
that is, the filter on C with base Lf (Φ) converges to Lf (µ0 ) in C ; here,
Lf (Φ) = {Lf (M) : M ∈ Φ } , where
Lf (M) = {Lf (µ) : µ ∈ L } = {µ(f ) : µ ∈ M },
concisely and suggestively denoted M(f ) .
Equivalently, for every f ∈ K (X; C) and every ε > 0 there exists
an M ∈ Φ such that |µ(f ) − µ0 (f )| < ε for all µ ∈ M .
In particular, Φ converges to the measure 0 if and only if, for every
f ∈ K (X; C) and every ε > 0, there exists a set M ∈ Φ such that |µ(f )| < ε
for all µ ∈ M ; concisely, Φ → 0 vaguely.
The assertion of Prop. 7 is that if Φ is a filter on M (X; C) such
that “the support of µ recedes indefinitely along Φ ” then Φ → 0 vaguely.
To prove this, suppose f ∈ K (X; C) and ε > 0 ( ε will play no role in
the argument). Let K be the support of f . By hypothesis, there exists
an M ∈ Φ such that Supp(µ) ∩ K = ∅ for every µ ∈ M . In view of the
preceding paragraph, it will obviously suffice to show that µ(f ) = 0 for
every µ ∈ M . Fix µ ∈ M . We know that Supp(µ) ∩ K = ∅ , that is,
K ⊂ { Supp(µ) . Writing U = { Supp(µ) , we have µU = 0 , that is, µ = 0
on K (X, U; C) ; since Supp f = K ⊂ U , we have f ∈ K (X, U; C) , whence
µ(f ) = 0 , which completes the proof.
III.27, `. 9, 10.
“By definition, if the support of a function f ∈ K (X; C) does not
intersect the support of a measure µ , then µ(f ) = 0 ”
To say that Supp(f )∩Supp(µ) = ∅ is to say that Supp(f ) ⊂ { Supp(µ) .
Writing U = { Supp(µ) , we have µU = 0 , that is, µ = 0 on K (X, U; C)
and in particular µ(f ) = 0 .
§2 support of a measure INT III.x38

The condition Supp(f ) ∩ Supp(µ) = ∅ can also be written

Supp(µ) ⊂ { Supp(f ) = { {x : f (x) 6= 0 }

= int { {x : f (x) 6= 0 }
= int {x : f (x) = 0 }

(where “int” means “interior”) thus the observation is that

(∗) Supp(µ) ⊂ int {x : f (x) = 0 } ⇒ µ(f ) = 0 .

III.27, `. 10, 11.

“. . .but the following more precise result is true:”
Prop. 8 is the proposition that, for a function f ∈ K (X; C) ,

(∗∗) Supp(µ) ⊂ {x : f (x) = 0 } ⇒ µ(f ) = 0 ;

it is stronger than (*), in that it yields the same conclusion with a weaker
hypothesis. It is more precise in the sense that it enlarges the set of functions
for which the conclusion is true.
III.27, `. 15, 16.
“. . . V is an open set containing S by hypothesis”
For, S = Supp(µ) ⊂ {x : f (x) = 0 } ⊂ V .
III.27, `. 19.
“Since the support of f h does not intersect S . . .”
Supp(f h) ⊂ Supp(h) ⊂ { S .
III.27, `. 20, 21.
“. . .therefore |f − f h| 6 2ε on X ”
On V , |f − f h| 6 |f | + |f h| 6 |f | + |f | < 2ε , whereas on { V ,
|f − f h| = |f − f · 1| = 0 .
III.27, `. −3.
“. . . |f (x)| 6 ah(x) on Supp(µ) ”
On Supp(f ) ∩ Supp(µ) , one has |f (x)| 6 a = ah(x) , whereas on
{ Supp(f ) ∩ Supp(µ) one has |f (x)| = 0 6 ah(x) .
III.27, `. −3, −2.
“. . .therefore
|µ|(|f |) 6 a|µ|(h) 6 akµk ”
The measure |µ| is positive, and on Supp(|µ|) = Supp(µ) one has
ah − |f | > 0 , therefore |µ|(ah − |f |) > 0 by Cor. 2; thus

|µ|(|f |) 6 |µ|(ah) = a|µ|(h) 6 a k |µ| k · khk = a kµk · khk 6 a kµk .

INT III.x39 measures §2

III.28, `. 7.
“. . .then g 6 bf /a , whence µ(g) 6 bµ(f )/a = 0 .”
f (y)
If y ∈ V then f (y) > a , whence g(y) 6 b = b · 1 6 b · ; whereas
a
if y ∈ { V then g(y) = 0 . Therefore 0 6 µ(g) 6 bµ(f )/a = 0 , and so
µ(g) = 0 .
The argument shows that for every g ∈ K + (X) with Supp g ⊂ V ,
one has µ(g) = 0 . If g ∈ K (X; C) with Supp g ⊂ V , then |g| ∈ K + (X)
and Supp |g| = Supp g ⊂ V , therefore µ(|g|) = 0 by the foregoing, and so
|µ(g)| 6 µ(|g|) = 0 . Thus µ(g) = 0 for all g ∈ K (X; C) with Supp g ⊂ V .
◦
A fortiori, µ(g) = 0 for all g ∈ K (X; C) with Supp g ⊂ V , in other
◦ ◦
words µ = 0 on K (X, V; C) , that is, V ⊂ { Supp(µ) , and in particular
x ∈ { Supp(µ) . Thus x ∈
/ Supp(µ) , as we wished to show.
III.28, `. 11-13.
“. . .there exists an open neighborhood V of x 0 such that at every point
of V ∩ S , g is zero”
Write N(g) = {x : g(x) 6= 0 } , Z(g) = { N(g) = {x : g(x) = 0 } ;
in particular, T = N(g) ∩ S , and the assertion of the proposition is that
T = Supp(g · µ) . By assumption, x0 is not in the closure of N(g) ∩ S , hence
there exists an open neighborhood V of x 0 such that V ∩ N(g) ∩ S = ∅ ,
whence V ∩ S ⊂ { N(g) = Z(g) —in other words, g = 0 on V ∩ S .
III.28, `. 13.
“. . .then f g is zero on S ”
We know that g = 0 on V ∩ S , and that f = 0 on { V hence on
{ V ∩ S , therefore f g = 0 on (V ∩ S) ∪ ( { V ∩ S) = S .
III.28, `. 14, 15.
“. . .in other words, the restriction of g · µ to V is zero.”
The argument shows that µ(f g) = 0 for all f ∈ K (X, V; C) , in other
words g ·µ = 0 on K (X, V; C) , whence V ⊂ { Supp(g ·µ) and in particular
x0 ∈ { Supp(g · µ) . We have shown that { T ⊂ { Supp(g · µ) , that is,
T ⊃ Supp(g · µ) . That’s half the battle.
III.28, `. 16–18.
“Conversely, assuming that the restriction of g · µ to an open neighbor-
hood W of a point x0 ∈ X is zero, let us show that there does not exist a
point of W ∩ S at which g is 6= 0 .”
One is assuming that x0 ∈ W , where W is open and W ⊂ { Supp(g·µ) ;
to put it another way, we are assuming that x 0 ∈
/ Supp(g · µ) (a closed set)
and that W is any open neighborhood of x 0 disjoint from Supp(g · µ) . If
§2 support of a measure INT III.x40

we show that g = 0 on W ∩ S , that is, W ∩ S ⊂ Z(g) = { N(g) , that is,

W ∩ S ∩ N(g) = ∅ , we will have shown that x 0 is not in the closure of
S ∩ N(g) , in other words x0 ∈
/ T . Varying x0 , we will then have shown that
{ Supp(g · µ) ⊂ { T , that is, Supp(g · µ) ⊃ T , whence equality in view of
the preceding note.
III.28, `. 20–22.
“. . .but then every function f ∈ K (X; C) with support contained
in U could be written f = gh , where h ∈ K (X; C) has support contained
in U ⊂ W ”
For such an f , write Z(f ) = {x : f (x) = 0 } and define

 f (x) for x ∈ U
h(x) = g(x)

0 for x ∈ Z(f ) .

Since both formulas yield 0 on U ∩ Z(f ) , h is well-defined; and since

{ Z(f ) = N(f ) ⊂ Supp f ⊂ U , h is defined everywhere on X . As the
restrictions of h to the closed sets U and Z(f ) are continuous, h is con-
tinuous on X (GT, I, §3, No. 2, Prop. 4), and Supp h ⊂ U ⊂ W .
III.28, `. 22.
“. . .it would then follow that µ(f ) = µ(gh) = 0 ”
We know that W ⊂ { Supp(g · µ) , that is, g · µ = 0 on K (X, W; C)
and in particular (g · µ)(h) = 0 , thus 0 = µ(gh) = µ(f ) .
III.28, `. 23.
“. . .contrary to the hypothesis y ∈ S .”
It has just been argued that µ(f ) = 0 for every f ∈ K (X; C) with
◦ ◦
Supp f ⊂ U ; a fortiori µ = 0 on K (X, U; C) , that is, U ⊂ { Supp(µ) =
◦
{ S , and in particular y ∈ U ⊂ { S , contrary to y ∈ W ∩ S . This completes
the proof of Prop. 8, as observed at the end of the note for `. 16–18.
III.28, `. 24, 25.
“Note that T is contained in the intersection of the support S of µ
and the support of g ”
Writing N(g) = {x : g(x) 6= 0 } , one has

T = S ∩ N(g) ⊂ S ∩ N(g) = S ∩ Supp g .

In the example that follows, Supp(µ) = {0}, Supp g = R and Supp(g·µ) = ∅.

INT III.x41 measures §2

III.28, `. −9, −8.

“Corollary. — In order that the measure g · µ be zero, it is necessary
and sufficient that g be zero on the support of µ .”
Writing A = Supp(µ) ∩ {x : g(x) 6= 0 } , we know from Prop. 10 that
Supp(g · µ) = A . Since

g · µ = 0 ⇔ Supp(g · µ) = ∅

and since
g = 0 on Supp(µ) ⇔ A = ∅ ,
the assertion of the corollary reduces to the triviality A = ∅ ⇔ A = ∅ .

III.29, `. 2–4.
“Proposition 12. — Let ai (1 6 i 6 n) be distinct points in a locally
compact space X . Every measure on X whose support is contained in the
set of the ai is a linear combination of the measures ε ai (1 6 i 6 n) .”
Pn
Conversely, if µ = ci εai then (No. 2, Prop. 4)
i=1

n
[ n
[
Supp(µ) ⊂ Supp(ci εai ) ⊂ Supp(εai ) = {a1 , . . . , an } ,
i=1 i=1

where the fact that Supp(εa ) = {a} is shown by the following argument:
if U = X --- {a} then for all f ∈ K (X, U; C) one has ε a (f ) = f (a) = 0 ,
thus U ⊂ { Supp(εa ) , that is, Supp(εa ) ⊂ { U = {a} ; since Supp(εa ) 6= ∅
(because εa 6= 0 ) necessarily Supp(εa ) = {a} (alternatively—cf. the note
for III.25, `. −7 to −5 —for every neighborhood V of a , there exists a
function f ∈ K (X, V; C) with εa (f ) = f (a) 6= 0 , whence a ∈ Supp(εa ) ).
III.29, `. 16–18.
“It suffices to prove that µ is orthogonal to the subspace V ◦ of K (X; C)
orthogonal to V (TVS, II, §6, No. 3, Cor. 2 of Th. 1), that is, that the rela-
tions hf, εa i = 0 , where a runs over the support of µ , imply hf, µi = 0 ”
In view of Prop. 12, V is the linear subspace of M (X; C) generated
by the measures εa with a ∈ Supp(µ) ; regarding K (X; C) and M (X; C)
as being in a (separating) duality via the bilinear form


(f, ν) 7→ hf, νi = ν(f ) f ∈ K (X; C), ν ∈ M (X; C) ,

it follows that V◦ is a linear subspace of K (X; C) and

V◦ = {f ∈ K (X; C) : hf, νi = 0 for all ν ∈ V }

= {f ∈ K (X; C) : hf, εa i = 0 for all a ∈ Supp(µ) }
§2 support of a measure INT III.x42

(TVS, II, §6, No. 3, Prop. 4, (iii) for the case of real scalars, loc. cit., §8,
No. 4 for the case of complex scalars). By the theorem on bipolars (TVS, II,
§6, No. 3, Th. 1 and §8, No. 4), the polar V ◦◦ of V◦ in M (X; C) is equal

to the closure of V in M (X; C) for the topology σ M (X; C), K (X; C) ,
that is,
V = V◦◦ ,
where V is the vague closure of V in M (X; C) . Thus the assertion that
µ is vaguely adherent to V is equivalent to the assertion that µ ∈ V ◦◦ .
III.29, `. 19.
“. . .but this is just Prop. 8 of No. 3.”
The cited proposition states that

f = 0 on Supp(µ) ⇒ µ(f ) = 0 ,

that is,
hf, εa i = 0 for all a ∈ Supp(µ) ⇒ hf, µi = 0 ,
equivalently
f ∈ V◦ ⇒ hf, µi = 0 ,
in other words µ ∈ V◦◦ .
Preparation for Corollary 1 of Theorem 1:
(I). Let A = {a1 , . . . , an } be a finite set of pairwise distinct points of a
locally compact space X . In order that a measure µ on X have support A ,
Pn
it is necessary and sufficient that µ = ck εak with the ck nonzero scalars,
k=1
in which case
n
X
(i) kµk = |ck | .
k=1

P
n
Proof. If µ has support A , we know from Prop. 12 that µ = ck εk
k=1
for suitable scalars ck . If a coefficient, say c1 , were equal to 0 , then one
would have Supp(µ) ⊂ {a2 , . . . , an } by the note for III.29, `. 2–4, contrary
to hypothesis.
P
n
Conversely, if µ = ck εk for suitable nonzero scalars ck , then
k=1

Supp(µ) ⊂ {a1 , . . . , an } = A
INT III.x43 measures §2

by the note for III.29, `. 2–4; we assert that Supp(µ) = A . Assume to the
contrary that some ak is absent from Supp(µ) . Then there exists an open
neighborhood U of ak such that U∩Supp(µ) = ∅ , whence U ⊂ { Supp(µ) ,
and so µ(f ) = 0 for all f ∈ K (X, U; C) . Choose f ∈ K (X, U; C) such
that f (ak ) 6= 0 . For all indices j 6= k we have aj ∈ { U and so f (aj ) = 0 ,
thus
n
X n
X X
0 = µ(f ) = cj εaj (f ) = cj f (aj ) = ck f (ak ) + cj · 0 = ck f (ak ),
j=1 j=1 j6=k

contrary to ck 6= 0 and f (ak ) 6= 0 .

Suppose now that µ has the indicated representation. For every
f ∈ K (X; C) ,
Xn  n
X
 
|µ(f )| =  ck f (ak ) 6 kf k |ck | ,
k=1 k=1

P
n
hence kµk 6 |ck | .
k=1
On the other hand, let V1 , . . . , Vn be pairwise disjoint neighborhoods
of a1 , . . . , an , respectively, and choose functions f k ∈ K (X; R) such that
0 6 fk 6 1 , fk (ak ) = 1 and Supp fk ⊂ Vk ; in particular,
[
fk = 0 on Vj for k = 1, . . . , n .
j6=k

For each k write |ck | = θk ck with |θk | = 1 , and set

n
X
g= θk fk ;
k=1

then g(ak ) = θk fk (ak ) = θk for all k . If x ∈ X then, since the fk have

disjoint supports, at most one term in the sum for g(x) can be nonzero,
therefore |g(x)| 6 1 ; but |g(a1 )| = 1 , and so kgk = 1 . Moreover,

n
X n
X n
X n
X
µ(g) = ck εak (g) = ck g(ak ) = ck θk = |ck | ,
k=1 k=1 k=1 k=1

P
n
thus kµk > |µ(g)| = |ck | , which completes the proof of (i).
k=1
§2 support of a measure INT III.x44

(II). Let A = {a1 , . . . , an } be a finite set of pairwise distinct points of a

locally compact space X , and let µ be a measure on X with Supp(µ) ⊂ A .
Let r be a real number > 0 . Then

(ii) kµk 6 r ⇔ µ ∈ bal conv {rεa1 , . . . , rεan } ,

where bal conv {rεa1 , . . . , rεan } denotes the balanced convex envelope of the
set {rεa1 , . . . , rεan } in M (X; C) .
Proof. Recall that the balanced convex envelope of a subset S of a
vector space (real or complex) is the set of all finite linear
P P combinations
cj uj , where uj ∈ S and the cj are scalars such that |cj | 6 1 (TVS,
II, §8, No. 2).
P
n
By (I), µ = ck εak , where ck is 0 when the point ak does not
k=1
belong to Supp(µ) , and
n
X |ck |
kµk 6 r ⇔ 6 1.
r
k=1

Thus the formula

n 
X ck 
µ= r ε ak
r
k=1

shows that if kµk 6 r then µ belongs to the balanced convex envelope of

{rεa1 , . . . , rεan } .
P
n P
n
Conversely, if µ = bk (rεak ) with |bk | 6 1 then, by (i), kµk =
k=1 k=1
P
n P
n
|bk r| = r |bk | 6 r .
k=1 k=1

The following general lemma prepares the way for item (III) below.
Lemma. — Let F, G be complex vector spaces in duality with respect
to a bilinear form (x, y) 7→ hx, yi (x ∈ F , y ∈ G ), and let M be a “circled”
( x ∈ M and |c| = 1 imply cx ∈ M ) subset of F . Then:
(1) The polar M◦ of M in G is given by the formula

M◦ = {y ∈ G : |hx, yi| 6 1 for all x ∈ M } .

(2) M◦ is convex and balanced.

(3) If N is the balanced convex envelope of M , then N ◦ = M◦ .
(4) M◦◦ = N (the closure of N for σ(F, G) ), which is the closed bal-
anced convex envelope of M .
INT III.x45 measures §2

Proof. Recall that a subset A of a complex vector space is said to be

balanced if cx ∈ A for all x ∈ A and all scalars c with |c| 6 1 . Balanced
sets are obviously circled. The polar of M is defined (for any subset M
of F ) by the formula

M◦ = {y ∈ G : R hx, yi > −1 for all x ∈ M }

(TVS, II, §8, No. 4, Def. 1).

(1) If y ∈ G and |hx, yi| 6 1 for all x ∈ M , then for all x ∈ M one
has
|R hx, yi| 6 |hx, yi| 6 1 ,
whence −1 6 R hx, yi 6 1 and in particular y ∈ M ◦ .
Conversely, suppose y ∈ M◦ and x ∈ M . By assumption,

R hcx, yi > −1

for all scalars c with |c| = 1 ; if c is chosen so that chx, yi = −|hx, yi| , then

−1 6 R hcx, yi = R chx, yi = −|hx, yi| ,

whence |hx, yi| 6 1 .
(2) If y = c1 y1 + c2 y2 , where y1 , y2 ∈ M◦ , c1 > 0 , c2 > 0 and
c1 + c2 = 1 then, for all x ∈ M ,

|hx, yi| = |c1 hx, y1 i + c2 hx, y2 i|

6 c1 |hx, y1 i| + c2 |hx, y2 i| 6 c1 · 1 + c2 · 1 = 1 ,

whence y ∈ M◦ ; thus M◦ is convex. If y ∈ M◦ and c is a scalar with

|c| 6 1 then, for all x ∈ M ,

|hx, cyi| = |c| · |hx, yi| 6 1 · 1 = 1 ,

whence cy ∈ M◦ ; thus M◦ is balanced.

(3) As noted at the beginning ofP the proof of (II), N is equalPto the set
of all finite linear combinations x = ck xk with xk ∈ M and |ck | 6 1 .
For all such x , and for y ∈ M◦ , one has
X  X  X
|hx, yi| =  ck hxk , yi 6 |ck | · |hxk , yi 6 |ck | · 1 6 1 ,

whence y ∈ N◦ (by (1) applied to N in place of M ); thus M ◦ ⊂ N◦ . But

M ⊂ N implies M◦ ⊃ N◦ , therefore M◦ = N◦ .
(4) Since N is balanced and convex, its closure N for σ(F, G) is also
balanced and convex (TVS, II, §2, No. 6 and §8, No. 2), thus N is the
§2 support of a measure INT III.x46

closed balanced convex envelope of M . Since N is convex, its closed convex

envelope is also equal to N , and since 0 ∈ N it follows from the theorem on
bipolars that N◦◦ = N . In view of (3), we have also M◦◦ = N .
(III). Let S be a subset of the locally compact space X , let r > 0 , let

B = {ν ∈ M (X; C) : kνk 6 r and Supp(ν) is a finite subset of S } ,

and let B ◦ be the polar set of B in K (X; C) for the canonical duality
hf, νi = ν(f ) . Then

1
(iii) B ◦ = {f ∈ K (X; C) : |hf, εa i| 6 for all a ∈ S} .
r
Proof. Let P = {rεa : a ∈ S } , and let Pbc be the balancedPconvex
envelope of P , thatPis, the set of all finite linear combinations ck νk ,
where νk ∈ P and |ck | 6 1 .
We assert that B = Pbc . For, if ν ∈ B , so that kνk 6 r and
Supp(ν) = {a1 , . . . , an } with the ak pairwise distinct elements of S , then
by (II),
ν ∈ bal conv{rεa1 , . . . , rεan } ⊂ Pbc .
P
n
Conversely, suppose ν ∈ Pbc , so that ν = ck (rεak ) for suitable points
k=1
P
n
ak ∈ S and scalars ck such that |ck | 6 1 ; we can suppose (after
k=1
gathering together coefficients of a same ε ak ) that the points ak are pair-
wise distinct (at work here is the triangle inequality for the partial sums
of the ck ), so that the εak are linearly independent; then, omitting terms
with zero coefficients, we can further suppose that the c k are nonzero. Then,
Supp(ν) = {a1 , . . . , an } ⊂ S by (I), and kνk 6 r by (II), thus ν ∈ B .
Now let {M = cν : |c| = 1 and ν ∈ P } (the “circled envelope”
of P ). Clearly M and P have the same balanced convex envelope, thus

bal conv M = Pbc = B ;

the vague closure B of B is therefore equal to the vaguely closed balanced

convex envelope of B as well as to the closed balanced convex envelope
of M , and since both B and M are circled, it follows from the preceding
Lemma that
M◦◦ = B ◦◦ = B .
Then (TVS, II, §6, No. 3)

M◦ = (M◦◦ )◦ = (B ◦◦ )◦ = B ◦ ,
INT III.x47 measures §2

thus

B ◦ = M◦ = {f ∈ K (X; C) : |hf, νi| 6 1 for all ν ∈ M }

= {f ∈ K (X; C) : |hf, c(rεa )i| 6 1 for all c ∈ T and a ∈ S }
1
= {f ∈ K (X; C) : |hf, εa i| 6 for all a ∈ S } ,
r

which establishes the formula (iii).

The argument shows incidentally that a measure µ on X is vaguely
adherent to B if and only if µ ∈ B ◦◦ ; in view of the formula (iii), µ is
vaguely adherent to B if and only if

1
f ∈ K (X; C) & |hf, εa i| 6 for all a ∈ S ⇒ |hf, µi| 6 1 ,
r

that is,
 1
f ∈ K (X; C) & kf Sk 6 ⇒ |µ(f )| 6 1 .
r
III.29, `. 24–26.
“To prove the first assertion. . .”
This is covered by the preceding remarks, with Supp(µ) in the role
of S , r = kµk , and A in the role of B .
III.29, `. −8 to −5.
“. . .we note that
lim inf kνk > kµk
ν→µ, ν∈A

since the function ν 7→ kνk is lower semi-continuous for the vague topology
(§1, No. 9, Cor. 4 of Prop. 15)”
One must adapt the cited Prop. 15 to the situation of a limit with respect
to the relative topology induced on A by the vague topology. Let V be
the set of neighborhoods of µ for the vague topology on M (X; C) . By
definition,  
lim inf kνk = sup inf kνk .
ν→µ, ν∈A V∈V ν∈V∩A

Given 0 < h < kµk , by the cited lower semi-continuity there exists a V ∈ V
such that ν ∈ V ⇒ kνk > h ; in particular, ν ∈ V ∩ A ⇒ kνk > h , hence

h6 inf kνk 6 lim inf kνk ,

ν∈V∩A ν→µ, ν∈A

and the desired inequality results from letting h → kµk .

§2 support of a measure INT III.x48

III.29, `. −5, −4.

“. . .and the conclusion follows from the fact that kνk 6 kµk for ν ∈ A
by definition.”
Since kνk 6 kµk for all ν ∈ A by the definition of A , we have (nota-
tions as in the preceding note)
 
lim sup kνk = inf sup kνk 6 kµk 6 lim inf kνk ,
ν→µ, ν∈A V∈V ν∈V∩A ν→µ, ν∈A

whence equality throughout (GT, IV, §5, No. 6, formula (11)); thus

lim kνk = kµk

ν→µ, ν∈A

by GT, IV, §5, No. 6, Cor. 1 of Th. 3 (the filter G in question being the
trace V ∩ A of V on A ).
Note: It follows that the mapping A ∪ {µ} → [0, +∞[ defined by
ν 7→ kνk is continuous at µ for the topology on A ∪ {µ} induced by the
vague topology.
III.30, `. 2–4.
“. . .for every ε such that 0 < ε < 1 , there exists, by virtue of Cor. 1,
a measure ν0 whose support is finite and contained in Supp(µ) and for
which ν0 − µ ∈ V and kµk > kν0 k > (1 − ε)kµk .”
With A as in Corollary 1, we know that

lim kνk = kµk ;

ν→µ, ν∈A

therefore, since kµk > (1 − ε)kµk , there exists a vague neighborhood V 0

of 0 such that

ν ∈ A , ν ∈ µ + V0 ⇒ kνk > (1 − ε)kµk .

Since µ is vaguely adherent to A , the vague neighborhood V ∩ V 0

of 0 contains ν0 − µ for some ν0 ∈ A , whence both ν0 − µ ∈ V and
kµk > kν0 k > (1 − ε)kµk .
III.30, `. 5.
“. . .and kν − ν0 k 6 kµk ”
Since kν0 k > (1 − ε)kµk = kµk − εkµk , one has kµk − kν 0 k 6 εkµk ,
therefore
kµk  kµk  kµk − kν0 k
ν − ν0 = · ν0 − ν0 = − 1 ν0 = · ν0 ,
kν0 k kν0 k kν0 k
INT III.x49 measures §2

whence kν − ν0 k = kµk − kν0 k 6 εkµk < kµk .

III.30, `. 5, 6.
“. . .for ε sufficiently small we therefore have ν − µ ∈ V + V , whence
the conclusion.”
If f ∈ K (X; C) and c > 0 , the set

W = {ρ ∈ M (X; C) : |ρ(f )| < c }

is an open neighborhood of 0 (for the vague topology); replacing f by c −1 f ,

one has
W = {ρ ∈ M (X; C) : |ρ(f )| < 1 } .
The set of finite intersections of such sets is a fundamental system of neigh-
borhoods of 0 . Thus we can suppose that

V = {ρ ∈ M (X; C) : |ρ(fk )| < 1 for k = 1, . . . , n } ,

where fk ∈ K (X; C) for k = 1, . . . , n .

Let us now impose a further condition on ε , namely, that it be suffi-
ciently small that

εkµk · kfk k < 1 for k = 1, . . . , n .

Then, the measure ν0 constructed above will satisfy

|(ν − ν0 )(fk )| 6 kν − ν0 k · kfk k 6 εkµk · kfk k < 1

for all k , whence ν − ν0 ∈ V . Thus

ν − µ = (ν − ν0 ) + (ν0 − µ) ∈ V + V ,

and so ν ∈ µ + V + V ; since the V + V form a fundamental system of

neighborhoods of 0 , we have shown that every neighborhood of µ contains
a measure ν ∈ A such that kνk = kµk .
III.30, `. 10–13.
“. . .The same reasoning as in Cor. 2 shows that we can limit ourselves
to proving that µ is in the vague closure of the convex set B formed by
the positive measures with finite support contained in Supp(µ) and with
norm 6 kµk .”
Recall that the real vector space M (X; R) may be identified with the
R-linear subspace of M (X; C) consisting of the complex measures ρ on X
such that ρ = ρ (§1, No. 5). The identification is a homeomorphism for the
§2 support of a measure INT III.x50

vague topologies (induced by K (X; R) and K (X; C) , respectively); for, if

ρj , ρ ∈ M (X; R) , then ρj (g) → ρ(g) for all g ∈ K (X; R) if and only if, for
all f ∈ K (X; C) ,

ρj (f ) = ρj (R f ) + iρj (I f ) → ρ(R f ) + iρ(I f ) = ρ(f ) .

Thus if S is a subset of M (X; C) consisting of real measures, the vague

closure of S in M (X; C) coincides with the vague closure of S in M (X; R) .
Moreover, it follows from §1, No. 8, Cor. 3 of Prop. 10 that kρk for a real
measure is the same whether calculated in M (X; C) or in M (X; R) .
Let us write D for the set of all ν ∈ M+ (X) such that Supp(ν) is a
finite subset of Supp(µ) and kνk = kµk , and let D be the vague closure
of D . The convexity of D is assured by the computation kcν + (1 − c)ν 0 k =
ckνk + (1 − c)kν 0 k for positive measures ν, ν 0 and for 0 6 c 6 1 (§1, No. 8,
Prop. 11). The assertion of Cor. 3 is that µ ∈ D .
Suppose we can show that µ ∈ B . Then (arguing as in the preceding
note) given an open neighborhood V of 0 in M (X; C) , say

V = {ρ ∈ M (X; C) : |ρ(fk )| < 1 for k = 1, . . . , n } ,

and given an ε > 0 such that εkµk · kfk k < 1 for all k , choose ν 0 ∈ B so
kµk
that ν0 −µ ∈ V and kµk > kν0 k > (1−ε)kµk . Setting ν = ·ν0 , we have
kν0 k
ν > 0 (because ν0 > 0 ), kνk = kµk (hence ν ∈ D ) and kν − ν0 k 6 kµk ,
whence ν − ν0 ∈ V and ν − µ ∈ V + V . This shows that µ ∈ D .
III.30, `. 13, 14.
“Again, it suffices to establish that µ belongs to the polar set of B ◦ ,
the polar set of B in the space K (X; R) ”
The point is that to exploit the argument of the preceding note, we
must show that µ ∈ B . It is clear that B is convex and that 0 ∈ B , thus
B = B◦◦ by the theorem on bipolars for the duality between M (X; R) and
K (X; R) (TVS, II, §6, No. 3, Th. 1).
III.30, `. 15–17.
“. . .but this means that for f ∈ K (X; R) the relations hf, ε a i >
−1/kµk for all a ∈ Supp(µ) imply hf, µi > −1 , which is a consequence
of No. 3, Cor. 2 of Prop. 8.”
We know, from item (I) of the Note for III.29, `. 19, that B consists
of 0 together with the set of all measures ν on X of the form
n
X
(∗) ν= c k ε ak ,
k=1
INT III.x51 measures §2

where a1 , . . . , an are distinct points of Supp(µ) , ck > 0 for k = 1, . . . , n ,

Pn
and ck = kνk 6 kµk .
k=1
Given f ∈ K (X; R) , we assert that

1
f ∈ B◦ ⇔ hf, εa i > − for all a ∈ Supp(µ) .
kµk

For, if f ∈ B◦ , that is, if hf, νi > −1 for all ν ∈ B , then in particular

hf, kµkεa i > −1 for all a ∈ Supp(µ) . Conversely, if f has the latter
property and if ν ∈ B is represented as in (*), then

X 1 X
hf, νi = ck hf, εak i > − ck > −1 ,
kµk

whence f ∈ B◦ .
Thus, in order that µ belong to B = B◦◦ , it is necessary and sufficient
that
f ∈ B◦ ⇒ hf, µi > −1 ,

that is,

1
hf, εa i > − for all a ∈ Supp(µ) ⇒ hf, µi > −1 ,
kµk

that is,

1
(∗∗) f >− on Supp(µ) ⇒ µ(f ) > −1 .
kµk

To prove (**), choose g ∈ K (X; R) such that 0 6 g 6 1 and g = 1 on

1
Supp(µ) , and in particular, kgk = 1 . Then if f > − on Supp(µ) ,
kµk
1
one has f + g > 0 on Supp(µ) , and since µ > 0 it follows from
kµk
1
1
Cor. 2 of No. 3, Prop. 8 that µ f + g > 0 , that is, µ(f ) > − µ(g) ;
kµk kµk
but µ(g) 6 kµk · kgk = kµk , whence −µ(g) > −kµk , and finally
1
µ(f ) > − µ(g) > −1 .
kµk
III.30, `. 23.
“. . .the conclusion therefore follows from Th. 1.”
§2 support of a measure INT III.x52

Let V be the linear subspace of M (X; C) generated by the point

measures, that is, the set of finite linear combinations of the Dirac meas-
ures εa (a ∈ X) ; it follows from Prop. 12 that V is equal to the set of
measures on X with finite support (see item (I) in the note for III.29, `. 19).
It is clear from Th. 1 that V is vaguely dense in M (X; C) ; we are
to show that it is also dense for the (stronger) topology of strictly compact
convergence; writing W for the closure of V for the topology of strictly
compact convergence, we are to show that W = M (X; C) .
The strategy of the proof is to show that (1) W contains every bounded
measure, and (2) the set of bounded measures is dense in M (X; C) for the
topology of strictly compact convergence.
(1) Let µ be a bounded measure. Then Rµ and I µ are also bounded,
hence µ is a linear combination of four bounded positive measures (§1, No. 8,
Cor. 2 of Prop. 11); thus, in showing that µ ∈ W , we can suppose that
µ > 0 . Let U be an open neighborhood of µ in M (X; C) for the topology
of strictly compact convergence; then U ∩ M + (X) is an open neighborhood
of µ in M+ (X) for that topology. By Prop. 18 of §1, No. 10, the topolo-
gies on M+ (X) induced by topology of strictly compact convergence and the
vague topology are identical, therefore U∩M + (X) is also an open neighbor-
hood of µ for the vague topology, hence there exists a vaguely open set V
in M (X; C) such that U ∩ M+ (X) = V ∩ M+ (X) ; by Cor. 3, V ∩ M+ (X)
contains a measure ν with finite support (contained in Supp(µ) ), thus
ν ∈ U ∩ V and we have shown that µ is in the closure of V for the
topology of strictly compact convergence, that is, µ ∈ W .
(2) It remains to show that every measure µ is the limit, for the topology
of strictly compact convergence, of bounded measures. Let K be the set of
compact subsets K of X , which is directed upward by inclusion. For each
K ∈ K , choose a function gK ∈ K (X; C) such that gK = 1 on K (no
other properties are needed). Then, for each K ∈ K , the measure g K · µ
has (compact) support contained in Supp(g K ) ∩ Supp(µ) (No. 3, Prop. 10),
hence is bounded (No. 3, Prop. 11), thus it will suffice to show that the
directed family (gK · µ)K∈K converges to µ for the topology of strictly
compact convergence, that is,

(∗) gK · µ → µ uniformly on each strictly compact subset of K (X; C) .

Let A be a strictly compact subset of K (X; C) , say K 0 ∈ K with A ⊂

K (X, K0 ; C) ; we are to show that gK · µ → µ uniformly on A . For every
K ∈ K with K ⊃ K0 , we have gK · µ − µ = 0 on K (X, K0 ; C) ; for, if
f ∈ K (X, K0 ; C) then gK f = f (because gK = 1 on K , hence on K0 ,
whereas f = 0 outside K0 ) and so

(gK · µ − µ)(f ) = µ(gK f ) − µ(f ) = µ(f ) − µ(f ) = 0 ;

INT III.x53 measures §2

in particular, gK · µ − µ = 0 on A for all K ⊃ K0 (because A ⊂

K (X, K0 ; C)), thus (∗) is established with a vengeance.
One will notice that item (2) proved above is a consequence of the Cor. 4
for which it has served as a lemma.
Note also that given a measure µ ∈ M (X; C) , the approximations in
both (1) and (2) are accomplished with measures whose support is contained
in Supp(µ) ; thus µ is in the closure, for the topology of strictly compact
convergence (call it T3 , as in Prop. 17 of §1, No. 10) of the set of measures
with finite support contained in Supp(µ) . In detail: Let U be an open
neighborhood of µ for T3 , and let ν ∈ U be a bounded measure with
support contained in Supp(µ) ; since U is also a neighborhood of ν for T 3 ,
it contains a measure ρ with finite support contained in Supp(ν) , and hence
in Supp(µ) .
III.30, `. −12.
“The condition is obviously sufficient by virtue of Prop. 6 of No. 2.”
With F = Supp(µ) in the cited Prop. 6, one concludes that the set

L = {ν ∈ M (X; C) : Supp(ν) ⊂ Supp(µ) }

is a vaguely closed linear subspace of M (X; C) . For every g ∈ K (X; C)

(indeed, for every g ∈ C (X; C) ) it follows from No. 3, Prop. 10 that
Supp(g · µ) ⊂ Supp(µ) , thus g · µ ∈ L . Since L is vaguely closed, it
contains the vague closure of the set of all such measures g · µ ; by hypoth-
esis, εx0 belongs to that vague closure hence ε x0 ∈ L , that is,

{x0 } = Supp(εx0 ) ⊂ Supp(µ) ,

thus x0 ∈ Supp(µ) .
III.30, `. −10 to −6.
“. . .we are to prove that there exists a function g ∈ K (X; C) such
that kg · µk 6 1 and such that

|fk (x0 ) − µ(gfk )| 6 δ

for 1 6 k 6 n .”
One has fk (x0 ) − µ(gfk ) = εx0 (fk ) − (g · µ)(fk ) = (εx0 − g · µ)(fk ) , thus
the indicated inequalities state that the vague neighborhood of ε x0 defined
by
{ν ∈ M (X; C) : |(εx0 − ν)(fk )| 6 δ (1 6 k 6 n) }
contains g · µ .
§2 support of a measure INT III.x54

III.30, `. −6, −5.

“Let U be a relatively compact open neighborhood of x 0 such that the
oscillation of each of the fk (1 6 k 6 n) on U is 6 δ/2 .”
One could take U to be a relatively compact open neighborhood of x 0
such that |fk (x)−fk (x0 )| 6 δ/4 for all x ∈ U and k = 1, . . . , n . But in fact
the proof requires only that U be a relatively compact open neighborhood
of x0 such that |fk (x) − fk (x0 )| 6 δ/2 for all x ∈ U and k = 1, . . . , n .
III.31, `. 5.
“Since ν has its support in U , we may identify it with its restriction
to U ”
The matter is delicate. 
Let us write simply νU for the restriction ν U of ν to ν (§2, No. 1);
thus, for g ∈ K (U; C) , νU (g) = ν(g 0 ) , where g 0 is the extension by 0
of g to X . One knows (loc. cit.) that g 7→ g 0 is a vector space isomor-
phism (indeed, an algebra isomorphism) of K (U; C) onto the vector space
K (X, U; C) of continuous functions on X with compact support contained
in U , and is clearly an isometry: kg 0 k = kgk . But this is true for every
measure ν and every open set U , and does not exploit the assumption that
Supp(ν) ⊂ U .
The text proposes to identify kνk and kν U k . These are defined by the
formulas

kνk = sup{|ν(f )| : f ∈ K (X; C) , kf k 6 1 }

kνU k = sup{|νU (g)| : g ∈ K (U; C) , kgk 6 1 }
= sup{|ν(f )| : f ∈ K (X, U; C) , kf k 6 1 } .

Since kνk is defined by a larger class of functions f than is kν U k , we

obviously have kνk > kνU k ; the problem is to exploit the hypothesis
Supp(ν) ⊂ U to prove that the two norms are equal. To this end, we
employ an auxiliary function f0 ∈ K (X; C) .
Let f0 ∈ K (X; C) be such that 0 6 f0 6 1 , f0 = 1 on Supp(ν) and
Supp(f0 ) ⊂ U (hence f0 = 0 on X --- U ). (The lemma at the end of this
note reviews the construction of such a function.)

Proposition A. For every f ∈ K (X; C) , one has f f 0 ∈ K (X, U; C)

and ν(f ) = ν(f f0 ) ; thus if g = f f0 U , then g 0 = f f0 and νU (g) = ν(g 0 ) =
ν(f f0 ) = ν(f ) .
Proof. Let f ∈ K (X; C) . Then Supp(f f0) ⊂ Supp(f )∩Supp(f0) , thus
Supp(f f0) ⊂ Supp(f0) ⊂ U , whence f f0 ∈ K (X, U; C) . Since f f0 = f on
Supp(ν) , one has ν(f f0 ) = ν(f ) (§2, No. 3, Cor. 1 of Prop. 8).
INT III.x55 measures §2

Proposition B. kνU k = kνk .

Proof. For every f ∈ K (X; C) , we have, with notations as in Prop. A,

kgk = kg 0 k = kf f0 k 6 kf k · kf0 k 6 kf k ,

therefore

|ν(f )| = |ν(f f0 )| = |νU (g)| 6 kνU k · kgk 6 kνU k · kf k ,

whence kνk 6 kνU k . The reverse inequality was noted earlier.

The foregoing may be summarized as follows:
Proposition C. If U is an open set in X and if ν is a measure on X
with compact support contained in U , then

kνk = sup{|ν(f )| : f ∈ K (X, U; C) , kf k 6 1 }

(where kf k = sup |f (x)| and kνk = sup{|ν(f )| : f ∈ K (X; C), kf k 6 1 ).

x∈X

This is the justification for “identifying ν with ν U ”. The lemma used

in the proof is as follows:
Lemma. If, in a locally compact space X , A is a compact set and U is
an open set such that A ⊂ U , then there exist (i) an open set V and a
compact set B such that
A⊂V⊂B⊂U
(briefly, every open neighborhood of the compact set A contains a compact
neighborhood of A ), and (ii) a continuous function f : X → [0, 1] such that
f = 1 on A and f = 0 on { B (thus f ∈ K (X; C) and Supp(f ) ⊂ B ,
so that f ∈ K (X, B; C) ⊂ K (X, U; C) ).
Proof. (i) By local compactness, U contains a compact neighborhood
of each of its points; a covering of A by finitely many such compact neigh-
borhoods yields both V (the union of their interiors) and B (their union).
(ii) Since the compact space B is normal (GT, IX, §4, No. 1, Prop. 1),
by Urysohn’s theorem there exists a continuous function g : B → [0, 1]
such that g = 1 on A and g = 0 on B --- V (loc. cit., Th. 1). Defining
f : X → [0, 1] by the formulas
(
g(x) for x ∈ B
f (x) =
0 for x ∈ X --- V

(both formulas yield 0 on B ∩ (X --- V) = B --- V ), f is continuous on X

(GT, I, §3, No. 2, Prop. 4).
§2 support of a measure INT III.x56

III.31, `. 6–8.
“. . .the hypothesis kνk = 1 then implies that there exists a function
h ∈ K (X; C) , with support contained in U , such that khk 6 1 and such
that |αk (1 − ν(h))| 6 δ/2 for 1 6 k 6 n .”
It is here that we need Prop. C of the preceding note: Since kνk = 1 ,
there exists a function h ∈ K (X, U; C) such that khk 6 1 and |ν(h)| is as
close to kνk = 1 as we like, in particular, close enough so that

|αk (1 − |ν(h)|)| 6 δ/2 for 1 6 k 6 n ;

replacing h by ch for suitable |c| = 1 , we can suppose that |ν(h)| = ν(h) ,

whence the desired inequalities.
III.31, `. 8, 9.
“The definition of U moreover shows that |(α k − fk (x))h(x)| 6 δ/2
for all x ∈ U ”
Recall that khk 6 1 . Since the inequalities hold trivially for x ∈ { U
(because Supp(h) ⊂ U ), we have in fact kα k h − fk hk 6 δ/2 .
III.31, `. 9, 10.
“. . .since kνk = 1 and Supp(ν) ⊂ U we therefore have
|ν((αk − fk )h)| 6 δ/2 ”
For,


ν (αk − fk )h  6 kνk · k(αk − fk )h)k = kαk h − fk hk 6 δ/2

by the preceding note.

III.31, `. 10–12.
“. . .setting g = g0 h ,

|fk (x0 ) − µ(gfk )| 6 δ for 1 6 k 6 n . ”

For,

|fk (x0 ) − µ(gfk )| = |fk (x0 ) − µ(g0 hfk )|

= |fk (x0 ) − (g0 · µ)(fk h)|
= |αk − ν(fk h)|

  
6 αk 1 − ν(h)  + ν(αk h − fk h) 6 δ/2 + δ/2 .

III.31, `. 13.
“This proves the proposition, since kg ·µk = k(g 0 h)·µk 6 kg0 ·µk = 1 .”
INT III.x57 measures §2

For,

kg · µk = k(g0 h) · µk = kh · (g0 · µ)k 6 khk · kg0 · µk 6 kg0 · µk = kνk = 1

(§1, No. 4 and No. 8, Prop. 12).

III.31, `. 19–21.
“Conversely, if Supp(ν) ⊂ Supp(µ) then ν is the vague limit of mea-
sures with finite support contained in Supp(µ) (Th. 1), hence is in the vague
closure of the set of measures g · µ by Prop. 13.”
Suppose first that ν has finite support contained in Supp(µ) ; say ν =
P
n
ck εak for suitable points ak ∈ Supp(µ) (Prop. 12). Note that the set
k=1

K (X; C) · µ = {g · µ : g ∈ K (X; C) }

is a linear subspace of M (X; C) , hence its vague closure

V = {g · µ : g ∈ K (X; C) }

is also linear subspace of M (X; C) . It follows from Prop. 13 that ε a ∈ V

for every a ∈ Supp(µ) , in particular for every a k (k = 1, . . . , n) , hence
ν∈V .
Now let ν be any measure with Supp(ν) ⊂ Supp(µ) . We know from
Th. 1 that there exists a directed family (ν j ) of measures, with finite support
contained in Supp(µ) , such that νj → ν vaguely; since νj ∈ V by the
foregoing, and V is vaguely closed, we conclude that ν ∈ V .
Remarks. M (X; C) is of course a module over the ring K (X; C) , via
the mapping (f, ρ) 7→ f · ρ ; thus V is the vague closure of the cyclic sub-
module K (X; C) · µ of M (X; C) . In fact, V is a submodule of M (X; C) ,
that is,
h ∈ K (X; C) , ν ∈ V ⇒ h · ν ∈ V .
For, if (gj ) is a directed family in K (X; C) such that g j · µ → ν vaguely,
then h · (gj · µ) → h · ν vaguely, since, for every f ∈ K (X; C) ,


h · (gj · µ) (f ) = µ(hgj f ) = (gj · µ)(hf ) → ν(hf ) = (h · ν)(f ) ;

thus (h · gj ) · µ → h · ν vaguely, which shows that h · ν ∈ V . Thus V is

the vaguely closed submodule of M (X; C) generated by µ ; the Corollary
provides a remarkable characterization of it.
In particular, if Supp(µ) = X , then V = M (X; C) , so to speak M (X; C)
is a “topologically cyclic” module with generator µ . (Example: X = R (or
§2 support of a measure INT III.x58

X = [0, 1] ), with µ the Lebesgue measure.) Thus the corollary yields a

striking characterization of “topologically cyclic” measures:

K (X; C) · µ = M (X; C) ⇔ Supp(µ) = X

(closure with respect to the vague topology on M (X; C) ).

Is M (X; C) a topological module over K (X; C) (for the direct limit
topology on K (X; C) and the vague topology on M (X; C) ), i.e., is the
bilinear mapping

K (X; C) × M (X; C) → M (X; C)

defined by (g, ρ) 7→ g·ρ continuous (jointly in the variables) for the indicated
topologies? I suspect that the answer is usually “no” and that the case of
X = N with the discrete topology may provide a counter-example (cf. TVS,
III, §5, Exer. 3, TVS, IV, §1, Exer. 7, and the note for III.16, `. 13–15).
A counterexample with X compact is proposed in §1, Exer. 12 d) (possible
candidate: the 1-point compactification of the discrete space N ?). The
following is a more elementary observation:
Proposition. The above mapping (g, ρ) 7→ g · ρ is always separately
continuous in the variables.
Proof. If ρj → ρ vaguely then g·ρj → g·ρ vaguely; for, if f ∈ K (X; C)
then
(g · ρj )(f ) = ρj (gf ) → ρ(gf ) = (g · ρ)(f ) .
On the other hand, if gj → g for the direct limit topology, the claim is
that gj · ρ → g · ρ vaguely for every measure ρ . This entails showing that
for every f ∈ K (X; C) and every measure ρ ,

ρ(gj f ) → ρ(gf ) ;

for this, we need only show that the mapping u : g 7→ gf is a continuous

endomorphism of K (X; C) . To this end, if K is any compact subset of X ,
and uK : K (X, K; C) → K (X; C) is the canonical injection, it suffices to
show that the composite mapping u ◦ u K ,
K (X, K; C) → K (X; C) → K (X; C)
g 7→g 7→ f g
is continuous. It is convenient to factor u ◦ u K another way: let K0 be
the support of f , let v : K (X, K; C) → K (X, K 0 ; C) be the mapping
v(g) = f g , and consider the factorization u ◦ u K = uK0 ◦ v :

K (X, K; C) → K (X, K0 ; C) → K (X; C)

g 7→f g 7→ f g
INT III.x59 measures §2

The continuity of v ( kf gk 6 kf k kgk ) and that of u K0 assures that of u◦uK

and completes the proof of the Proposition.
III.31, `. −9 to −7.
“For every a ∈ N and every neighborhood V of a , there exists a
function f ∈ K (X; C) with support contained in V , equal to 1 at the
point a and to 0 at the other points of N , whence µ(f ) = h(a) 6= 0 .”
Since N is discrete, {a} is an open subset of N , hence there exists
a neighborhood W of a in X such that W ∩ N = {a} ; replacing W
by V ∩ W , we can suppose that W ⊂ V . Choose f ∈ K (X; C) so that
f (a) = 1 and Supp(f ) ⊂ W ; then f = 0 on { W , hence on N ∩ { W =
N --- W = N --- W ∩ N = N --- {a} , that is, f = 0 at the points of N other
than a ; moreover, Supp(f ) ⊂ W ⊂ V as desired. Finally,
X
µ(f ) = h(x)f (x) (definition of µ )
x∈X
X
= h(x)f (x) ( N = {x ∈ X : h(x) 6= 0 )
x∈N

= h(a)f (a) ( f = 0 on N --- {a} )

= h(a) · 1 6= 0 ( f (a) = 1 and a ∈ N ) .

To summarize: Assuming a ∈ N , for every neighborhood V of a there

exists a function f ∈ K (X; C) such that Supp(f ) ⊂ V and µ(f ) 6= 0 ,
whence a ∈ Supp(µ) (No. 2); thus N ⊂ Supp(µ) .
III.31, `. −5, −4.
“. . .for every function g ∈ K (X; C) with support contained in W , we
therefore have µ(g) = 0 , which proves that b ∈ / Supp(µ) .”
Calculating as in the preceding note, we have
X
µ(g) = h(x)g(x)
x∈N

=0 (because g = 0 on { W ⊃ N ).

To summarize: Assuming b ∈
/ N , the existence of such a neighbor-
/ Supp(µ) (No. 2). Thus { N ⊂ { Supp(µ) ,
hood W of b assures that b ∈
i.e., Supp(µ) ⊂ N .
III.31, `. −1 to III.32, `. 1.
“. . .the restriction of µ to Va is therefore a point measure with support
{a} (No. 2, Prop. 5)”

Supp(µVa ) = Va ∩ Supp(µ) = Va ∩ N = {a} ; quote Prop. 12.
§3 integrals of continuous vectorial functions INT III.x60

III.32, `. 2.
“. . .is of the form h(a)εa . . .”
The scalar coefficient yielded by the cited Prop. 12 is here written h(a)
in anticipation of the function h being defined.
III.32, `. 2–4.
“Setting h(x) = 0 at the points of { N , and denoting by ν the measure
defined by the masses h(x) , the principle of localization shows that ν = µ .”
Let us first establish that the function h : X → C so defined is eligible
to define a discrete measure. Indeed, for every compact subset K of X ,
N ∩ K is finite; for, since N is closed, N ∩ K is a compact subset of the
discrete space N , hence is covered by finitely many of the open subsets {x}
(x ∈ N) of N . It follows from the discussion in §1, No. 3, Example I that
a discrete measure ν on X can be defined, for every f ∈ K (X; C) , by the
formula
X
ν(f ) = h(x)f (x)
x∈X
X
= h(a)f (a) (h = 0 on { N )
a∈N
X
= h(a)εa (f ) ( f = 0 on { Supp(f ) )
a∈N∩Supp(f )

(a finite sum, since Supp(f ) is compact).


Note that if a ∈ N then ν Va = h(a)εa ; for, if f ∈ K (X, Va ; C)
then f = 0 on { Va , hence on N --- Va , thus the above sum reduces to
h(a)εa (f ) , that is, ν(f ) = h(a)εa (f ) . Thus,
 
(i) ν Va = h(a)εa = µVa for all a ∈ N .

Now, N is defined to be equal to Supp(µ) , and the proof of “necessity”

shows that Supp(ν) = N . Thus

(ii) Supp(µ) = N = Supp(ν) .

For each x ∈ { N define Vx = { N = { Supp(µ) = { Supp(ν) ; then

 
(iii) ν Vx = 0 = µVx for all x ∈ { N .

From (i) and (iii), we see that (Vx )x∈X is an open covering of X such that
 
ν Vx = µ Vx for all x ∈ X ,
INT III.x61 measures §3

therefore µ = ν by the principle of localization (No. 1, Cor. of Prop. 1),

thus µ is indeed discrete.

§3. INTEGRALS OF CONTINUOUS VECTOR-VALUED FUNCTIONS

III.32, `. 14–16.
“. . . E0 * , equipped with the weak topology σ(E 0 *, E0 ) , may be canoni-
cally identified with the completion of E equipped with the weakened topology
σ(E, E0 ) .”
In TVS, II, §6, No. 7, Prop. 9, set F = E and G = E 0 , placed in duality
via hz, z0 i = z0 (z) .
{In particular, regarding E ⊂ E b (the completion of E for the weakened
topology σ(E, E ) ), if z ∈ E then the corresponding element of E 0 * is
0

the linear form z0 7→ z0 (z) . The extension of the correspondence to general

elements b z of Eb entails identifying E0 with Eb 0 (a continuous linear form z0
on E being extended to E b by continuity).}
The proof of the cited Prop. 9 is based on the fact that if G is any vector
space over C , and G* is its algebraic dual equipped with the topology
σ(G*, G) , then G* is a complete topological vector space, a consequence
of the observation that G* is isomorphic, as a topological vector space, to
the product space CI , where I is a set whose cardinality is equal to the
dimension of G (TVS, §6, No. 6, Cor. 2 of Prop. 8). An alternative proof
of the latter observation is as follows.
Let (eι )ι∈I be a basis of G indexed faithfully ( ι 6= κ ⇒ e ι 6= eκ ), and
define a mapping v : G* → CI by


v(y*) = y*(eι ) ι∈I for y* ∈ G* .

The linearity of v follows directly from the definitions: for y* , z* in G* ,

v(y* + z*) = (y* + z*)(eι ) ι∈I = y*(eι ) + z*(eι ) ι∈I



= y*(eι ) ι∈I + z*(eι ) ι∈I = v(y*) + v(z*) ,

and similarly v(cy*) = c v(y*) for every scalar c. If v(y*) = 0 then y*(e ι ) = 0
for all ι ∈ I , whence y* = 0 by linearity; thus v is injective. Given
any (cι )ι∈I ∈ CI , the linear form y* ∈ G* for which y*(e ι ) = cι for
all ι ∈ I satisfies v(y*) = (cι )ι∈I ; thus v is surjective, hence is a vector
space isomorphism.
§3 integrals of continuous vectorial functions INT III.x62

The bicontinuity of v is perhaps most easily seen when convergence is

cast in terms of directed families (“nets”). For each y ∈ G , let f y : G* → C
be the linear form defined by fy (y*) = y*(y) for all y* ∈ G* ; thus the
topology σ(G*, G) on G* is by definition the initial topology for the family
(fy )y∈G of linear forms. It follows that if y* ∈ G* and F is a filter on G* ,
then

F → y* in G* ⇔ fy (F) → fy (y*) = y*(y) for all y ∈ G

(GT, I, §7, No. 6, Prop. 10). Translated in terms of nets, this says that for
a directed family (yα∗ ) in G* ,

yα∗ → y* in G* ⇔ yα∗ (y) → y*(y) in C for every y ∈ G .

Since every y ∈ G is a finite linear combination of the basis vectors e ι , and

since the linear operations in C are continuous, an equivalent condition is
that
yα∗ (eι ) → y*(eι ) for all ι ∈ I ,
that is, for every ι ∈ I , the ι’th coordinate of v(y α *) converges to the
ι’th coordinate of v(y*) ; since the product topology on C I is the initial
topology for the family of coordinate projections, this is in turn equivalent
to the condition v(yα∗ ) → v(y*) in CI (cf. J.L. Kelley, General topology,
p. 91, Th. 4). Thus

yα∗ → y* in G* ⇔ v(yα∗ ) → v(y*) in CI ,

whence the asserted bicontinuity of v . Since continuous linear mappings

are automatically uniformly continuous (GT, III, §3, No. 1, Prop. 3), and
since CI is complete (GT, II, §3, No. 5, Prop. 10), one concludes that G*
is complete.
III.32, `. −16.
“. . .into C . . .”
More precisely, into R or C as the case may be (cf. the preamble to
the section).
III.32, `. −16, −15.
“. . . (in other words the mapping z0 ◦ f , also denoted hf , z0 i ) is contin-
uous.”
Since σ(E, E0 ) is the initial topology for the family of mappings
z0 : E → C ( z0 ∈ E ), the weak continuity of f can also be expressed by
saying that f : X → E is continuous when E is equipped with its weakened
topology σ(E, E0 ) .
INT III.x63 measures §3

III.32, `. −10, −9.

“. . . they are, however, equal when E is finite-dimensional.”
Assume E is a finite-dimensional Hausdorff topological vector space;
then the weakened topology σ(E, E0 ) on E coincides with the original topol-
ogy of E (TVS, I, §2, No. 3, Th. 2). Therefore every weakly continuous
function f : X → E is continuous (the converse is trivially true for any topo-
logical vector space E ). We are to show that K f (X; E) ⊂ K (X; E) (whence
equality).
Let f ∈ K f (X; E) . By the preceding remark, f is continuous. Let
z1 , . . . , zn be a finite separating set in E0 (e.g., a basis of E0 , which we
0 0

know to be finite-dimensional with same dimension as E , by A, II, §7, No. 5,

Th. 4 and TVS, I, §2, No. 3, Cor. 2 of Th. 2). Then, for x ∈ X ,


f (x) 6= 0 ⇔ z0k f (x) =6 0 for some k ,

thus
n
[
{x : f (x) 6= 0 } = {x : (z0k ◦ f )(x) 6= 0 } ,
k=1

whence
n
[
{x : f (x) 6= 0 } = {x : (z0k ◦ f )(x) 6= 0 } ,
k=1

that is,
n
[
Supp(f ) = Supp(z0k ◦ f ) ;
k=1

by hypothesis, each term of the union is compact, whence f ∈ K (X; E) .

III.33, `. 6–11.
“Definition 1. — For every function f
f ∈ K
R R (X; E) we call Rintegral
R f with respect to µ , and denote
of
0
by f dµ or f (x) dµ(x) , or f µ , or
f (x)µ(x) , the element of E * defined by
Z  Z
0
(1) f dµ, z = hf , z0 i dµ for all z0 ∈ E0 .

We note thatR even if E is Hausdorff and f ∈ K (X; E) , one does not

necessarily have f dµ ∈ E (Exer. 1; cf. No. 3).”
R
It is noteworthy that the notation µ(f ) is not proposed for f dµ (a
function,
R more precisely, a linear form on E 0 ), nor for the generalization of
f dµ given in Ch. VI, §1, No. 1, Def. 1; yet it is authorized for the case
§3 integrals of continuous vectorial functions INT III.x64

of a vectorial measure Rm : K f (X) → E ( E a Banach space) (Ch. VI, §2,

No. 2, Def. 2), where f dm ∈ E0 ∗ is denoted m(f ) for (in particular)
f ∈ K (X)R.
For f dµ to belong to E (after identifying each element z of E with
the linear form on E0 defined by z0 7→ z0 (z) for all z0 ∈ E0 ) means that
there exists a vector zf ∈ E such that
Z



f dµ, z0 = zf , z0 for all z0 ∈ E0 .

Conditions under which this occurs are discussed in No. 3, Cor. 2 of Prop. 7
(assuming E to be a quasi-complete Hausdorff locally Rconvex space), but
only for functions f in K (X; E) . Criteria assuring that f dµ belongs to E
for functions f ∈ K f (X; E) are given in Ch. VI, §1, No. 2, Cor. of Prop. 7
(with extra conditions on E and f ) and Prop. 8 (with extra conditions on µ
and f ).
When E is a Banach space, so Rthat the quasi-completeness condition
cited above is trivially satisfied and f dµ ∈ E for every f ∈ K (X; E) , the
notation µ(f ) is authorized by Ch. IV, §4, No. 1, Def. 1, as well as by Ch. V,
§1, No. 3, Def. 3; one would suppose that it would also be tolerated in the
case that E is a quasi-complete Hausdorff locally convex space.
The issue does not come up in the case of a vector-valued measure m
(Ch. VI, §2, No. 1, Def. 1), where m is given at the outset R to be a continuous
linear mapping of K (X) into a Banach space E and f dm is defined to
be m(f ) for f ∈ K (X) (loc. cit., No. 2, Def. 2); similarly for the case
of integration of a vector-valued function f with respect to a vector-valued
measure m (loc. cit., No. 7, Prop. 11 and the remark following its proof).
A final instance of the use of µ(f ) occurs in the extension of the the-
ory to Hausdorff topological spaces X , with E a Banach space (remarks
following Prop. 16 in Ch. IX, §1, No. 10).
Summary.
R I see no clear pattern in the foregoing, but sense
R a tendency:
One
R writes f dµ = µ(f ) either when this is the definition of f dµ or when
f dµ turns out to be an element of the ambient locally convex space RE ; in
the two cases cited above where the notation µ(f ) is not proposed, f dµ
is an element of an algebraic dual E0 ∗ .
At any rate, the issue seems unimportant; but the above tour of defini-
tions may serve as a useful preview of forthcoming concepts.
III.33, `. −8 to −6.
P
n
“ . . . we then have f (x) = fi (x)ei for all x ∈ X , and
i=1
Z n
X
f dµ = µ(fi )ei . ”
i=1
INT III.x65 measures §3

The first equality follows from the computation (for k = 1, . . . , n and

x ∈ X)
n

X 

fi (x)ei , e0k = fk (x) = (e0k ◦ f )(x) = e0k f (x) = hf (x), e0k i .
i=1

P
n
Thus f = fi ei . To prove the second equality, it suffices to show that for
i=1
each i , the function gi = fi ei has integral µ(fi )ei . Indeed,



(z0 ◦ gi )(x) = z0 gi (x) = z0 fi (x)ei
= fi (x)z0 (ei ) = hei , z0 ifi (x) ,

thus z0 ◦ gi = hei , z0 ifi , whence

Z Z Z

0

gi dµ, z = hgi , z i dµ = (z0 ◦ gi ) dµ
0

Z
= hei , z i fi dµ = µ(fi )hei , z0 i
0

= hµ(fi )ei , z0 i ,
R
therefore gi dµ = µ(fi )ei .
III.34, `. 8–10.
“ . . . we have Z
f dεy = f (y)
R
because hf , z0 i dεy = hf (y), z0 i by definition.”
For all z0 ∈ E0 ,
Z Z Z


f dεy , z = hf , z i dεy = (z0 ◦ f ) dεy
0 0

= (z0 ◦ f )(y) = z0 f (y) = hf (y), z0 i .

III.34, `. 12, 13.

“. . . the duals E0 and E01 are identical . . .”
If π : E → E1 = E/N is the canonical mapping and z01 ∈ E01 then
0
z1 ◦ π ∈ E0 , whence a linear mapping u : E01 → E0 , u(z01 ) = z01 ◦ π , which
−1
is injective since π is surjective. If z 0 ∈ E0 then its kernel z0 (0) is a closed
−1
subset of E containing 0 , therefore z0 (0) ⊃ N ; it follows that there is
§3 integrals of continuous vectorial functions INT III.x66

a unique linear form z01 on E1 such that z0 = z01 ◦ π . Since E1 carries

the final topology for π (TVS, II, §4, No. 4, Example I), the continuity of
z01 ◦ π = z0 implies that of z01 . Thus z01 ∈ E01 and u(z01 ) = z0 , so u is
surjective (hence is a vector space isomorphism).
III.34, `. 13–15.
f (X; E) , it is necessary and sufficient
“. . . for a function f to belong to K
that f1 = π◦f (where π : E → E1 is the canonical homomorphism) belong
R R
to Kf (X; E1 ) , in which case f dµ = f1 dµ .”
In the notations of the preceding note, if z 01 ∈ E01 and z0 = z01 ◦ π , and
if f : X → E and f1 = π ◦ f : X → E1 , one has

z0 ◦ f = (z01 ◦ π) ◦ f = z01 ◦ (π ◦ f ) = z01 ◦ f1 ;

thus z0 ◦ f = z01 ◦ f1 , whence trivially z0 ◦ f ∈ K (X; C) if and only if

z01 ◦ f1 ∈ K (X; C) . It follows that

f (X; E)
f ∈K ⇔ f (X; E1 ) ,
f1 ∈ K

and in this case, for all such pairings z 0 ↔ z01 , f ↔ f1 , one has

µ(z0 ◦ f ) = µ(z01 ◦ f1 ) ,

that is, Z Z



f dµ, z0 = f1 dµ, z01 ;
R
thus,
R when E0 and E01 are identified via the pairing z0 ↔ z01 , f dµ and
f1 dµ define the same element of the algebraic dual (E 0 )∗ = (E01 )∗ .
III.34, `. −14 to −11.
“. . .the mapping z0 7→ Rz0 which, to every continuous (complex) linear
form z0 on E , makes correspond the continuous (real) linear form z 7→
Rhz, z0 i on E0 , is an R-isomorphism of the dual E0 onto the dual E00
of E0 (TVS, II, §8, No. 1).”
To establish the notation for the next note, we review the construction
of the isomorphism (TVS II.61). Most of what is going on is pure vector
space theory; to keep the notation simple, it is best to leave the topology to
the end.
To set the stage: by E00 the author means (E0 )0 , the dual of the topo-
logical vector space E0 , namely, the real vector space E0 underlying E ,
equipped with the given topology on E . The assertion is that (E 0 )0 is
isomorphic, as a real vector space, to (E 0 )0 , that is, the real vector space
INT III.x67 measures §3

underlying the complex vector space E 0 ; identifying (E0 )0 and (E0 )0 via
this isomorphism then renders the notation E 00 unambiguous (alternatively,
it allows E00 to denote either of the two spaces according to the context).
Suppose first that E is any complex vector space (no topology). If f
is a (complex)

linear form on E (that is, f ∈ E ∗ ) then the formula g(z) =
R f (z) defines an R-linear form on E (concisely denoted Rf —or R ◦ f ,
where R : C → R is the R-linear mapping R(a + ib) = a ), that is, a linear
form on E0 , and so g ∈ (E0 )∗ (the algebraic dual of E0 ). Since




g(iz) = R f (iz) = R if (z) = −I f (z) ,

one has



(∗) f (z) = R f (z) + iI f (z) = g(z) − ig(iz) .

The correspondence f 7→ g = Rf defines a mapping E ∗ → (E0 )∗ that is

clearly R-linear ; writing u(f ) = g = Rf , the R-linear mapping u : E ∗ →
(E0 )∗ can also be regarded as a linear mapping (E ∗ )0 → (E0 )∗ .
From (*) it is clear that g = 0 ⇒ f = 0 , therefore u is injective. To see
that u is surjective, suppose g is any element of (E 0 )∗ ; define f : E → C
by the formula

(**) f (z) = g(z) − i g(iz) (z ∈ E) .

It is routine to check that f : E → C is R-linear; to see that it is C-linear

(i.e., that f ∈ E∗ ) it suffices to check that f (iz) = i f (z) , and this is shown
by the computation

f (iz) = g(iz) − i g(i · iz) = g(iz) + i g(z)

= i g(z) − i g(iz) = i f (z) .

Thus u : (E∗ )0 → (E0 )∗ is an isomorphism of real vector spaces, but we will

write u : E∗ → (E0 )∗ when we want to regard u as an R-linear bijection
defined on a complex vector space.
Now suppose E is a topological vector space over C . If f is continuous
(i.e., if f ∈ E0 ) it is clear from the formula g = Rf that g is continuous (i.e.,
g ∈ (E0 )0 ), thus u defines a mapping (E0 )0 → (E0 )0 , linear and injective.
Whereas if g ∈ (E0 )0 is given, and if f is defined by (**), it follows from
the continuity of z 7→ iz that f is continuous, that is, f ∈ E 0 . Thus the
mapping (E0 )0 → (E0 )0 is surjective, hence is an isomorphism of real vector
spaces. In the notation of the cited assertion, z 0 7→ Rz0 ( z0 ∈ E0 ) defines
§3 integrals of continuous vectorial functions INT III.x68

an isomorphism of real vector spaces (E 0 )0 → E00 . The inverse mapping

v = u−1 : (E0 )0 → (E0 )0 is given, for g ∈ (E0 )0 , by v(g) = z0 ∈ E0 , where

z0 (z) = g(z) − i g(iz) for all z ∈ E ;

in a somewhat more cluttered notation, for z 00 ∈ (E0 )0 , v(z00 ) ∈ (E0 )0 is

given by

v(z00 ) (z) = z00 (z) − i z00 (iz) for all z ∈ E ,
that is,
hz, v(z00 )i = hz, z00 i − ihiz, z00 i
for all z00 ∈ (E0 )0 and z ∈ E .
III.34, `. −11, −10.
“. . . Similarly, the algebraic dual E 00 * of the real vector space E00 may be
canonically identified with the real space underlying the algebraic dual E 0 *
of E0 .”
As observed in the preceding note, E 00 is to be interpreted as (E0 )0 ;
the assertion is that there is a canonical isomorphism (E 0 )0 ∗ → (E0 ∗ )0 of
real vector spaces.
The crux of the matter is purely algebraic: as shown in the preceding
note, if F is any complex vector space, then the real vector spaces (F ∗ )0 and
(F0 )∗ are canonically isomorphic (where * denotes algebraic dual and 0 indi-
cates the underlying real vector space structure), via the mapping f 7→ Rf
( f ∈ F∗ ).
Now let E be the given locally convex space over C . By the preceding
note, there is a vector space isomorphism v : (E 0 )0 → (E0 )0 that assigns to
z00 ∈ (E0 )0 the element v(z00 ) ∈ E0 defined by


v(z00 ) (z) = z00 (z) − i z00 (iz) for all z ∈ E .

It follows that the transpose t v of v is an isomorphism

∗
∗
t
v : (E0 )0 → (E0 )0 = (E0 )0 ∗

of the algebraic duals (A, II, §2, No. 5), defined by t v(h) = h ◦ v for all
linear forms h : (E0 )0 → R .
On the other hand, by the second paragraph above (with F replaced by
the complex vector space E0 ) there exists an isomorphism w of real vector
spaces,

∗
w : (E0 ∗ )0 → (E0 )0
INT III.x69 measures §3

defined by w(f ) = Rf for every linear form f on E 0 . Thus the composite

mapping ϕ = t v ◦ w

(E0 ∗ )0 w - (E0 )
∗ t
v - (E )0 ∗
0 0

is the sought-for isomorphism of real vector spaces

ϕ : (E0 ∗ )0 → (E0 )0 ∗ .

The foregoing proof is efficient, but one yearns for a more direct con-
struction that, given an element of E 0 ∗ , exhibits the corresponding element
of (E0 )0 ∗ . Such a construction will now be given; the following (commuta-
tive) diagram will be helpful in keeping track of the argument.
f -C
v(z00 ) = z0 ∈ E0
6
v = u−1 R

?
z00 ∈ (E0 )0 -R
g
Given f ∈ E0 ∗ , the formula




g(z00 ) = Rf v(z00 ) z00 ∈ (E0 )0

defines a linear form on the real vector space (E 0 )0 , that is, g ∈ (E0 )0 ∗ .
We define ψ : E0 ∗ → (E0 )0 ∗ by ψ(f ) = g . It is routine to verify that ψ is
R-linear.


ψ is injective: If g = 0 then Rf v(z00 ) = 0 for all z00 ∈ (E0 )0 , that
is, (Rf )(z0 ) = 0 for all z0 ∈ E0 ( v is surjective); thus the real part of the
complex linear form f is 0 , whence f = 0 by a now-familiar argument.
ψ is surjective: Given g ∈ (E0 )0 ∗ , that is, a linear form g : (E0 )0 → R ,
we seek a linear form f : E0 → C such that ψ(f ) = g . For all z0 ∈ E0
define

f (z0 ) = g(Rz0 ) − i g R(iz0 )



(where R(iz0 ) denotes the R-linear form z → R iz0 (z) = −I z0 (z)
on E ) It is routine to show that f is R-linear and that
f (iz 0 ) = i f (z0 ) ,
hence that f is C-linear, that is, f ∈ E 0 ∗ . Now, R f (z0 ) = g(Rz0 ) for all
z0 ∈ E0 ; in other words, for all z00 ∈ (E0 )0 ,




R f (v(z00 )) = g R(v(z00 ) = g u(v(z00 )) = g(z00 ) ,
§3 integrals of continuous vectorial functions INT III.x70


that is, Rf (v(z00 )) = g(z00 ) , which says that ψ(f ) = g .
Inasmuch as f ◦ v = t v(f ) figures in both constructions, one suspects
that ϕ = ψ . Indeed, for f ∈ E0 ∗ and z00 ∈ (E0 )0 , one has



ϕ(f ) (z00 ) = (t v ◦ w)(f ) (z00 )



= t v w(f ) (z00 ) = t v(Rf ) (z00 )





= (Rf ) ◦ v (z00 ) = Rf v(z00 ) = ψ(f ) (z00 ) ,

whence ϕ(f ) = ψ(f ) for all f ∈ E0 ∗ .

III.34, `. −9 to −6.
“It follows that if µ is a real measure and f a mapping in K f (X; E) ,
the formula (1) is again valid when f is regarded as taking its values in E 0
and the canonical bilinear forms figuring in the two members as being, re-
spectively, relative to the duality between E 00 and E00 * for the first member
and the duality between E0 and E00 for the second.”
In the swarm of identifications, it is easy to lose the ball in the sun.
Here is what is going on. We are given a locally convex space E over C ,
and a measure µ ∈ M (X; C) such that the restriction of µ to K (X; R)
is real-valued, hence defines a measure µ 0 ∈ M (X; R) . If f ∈ K f (X; E) ,
R 0 0
then f dµ is defined to be a suitable element z * of E * (Def. 1). On the
other hand, one can regard f as belonging to K f (X; E0 ) , where E0 is the
locally convex space over R underlying E ; denotingR it f 0 when so regarded,
one can define (analogously to Def. 1) its integral f0 dµ0 to be a suitable
element z00 * of (E00 )* . The assertion, stated simply, is that

ψ(z0 *) = z00 * ,

where ψ : E0 * → (E0 )0 * is the R-linear bijection described in the preceding

note;
R so to speak,
R the identification of (E 0 *)0 and (E0 )0 * via ψ transforms
f dµ into f0 dµ0 .
The first task is to paraphrase Definition 1 for the case of real mea-
sures. Let ν : K (X; R) → R be a real measure, let F be a locally con-
vex space over R (not necessarily equal to E 0 for some locally convex
space E over C ), and let K f (X; F) be the real vector space of all mappings
g : X → F that are weakly continuous and scalarly of compact support, that
is, such for every w0 ∈ F0 the function


x 7→ hg(x), w0 i = w0 g(x) = (w0 ◦ g)(x) (x ∈ X)
INT III.x71 measures §3

f (X; F) , the function

belongs to K (X; R) . For each g ∈ K

Z Z
0 0
w 7→ (w ◦ g)(x) dν(x) = hg(x), w0 i dν(x) (w0 ∈ F0 )

is evidently a linear form on F0 , that is, an element

R of F0 * ; it is called the
integral of g with respect to ν and is denoted g dν , thus

DZ E Z
0
0
(1 ) g dν, w = hg, w0 i dν for all w0 ∈ F0 .

We now apply the preceding paragraph to the case that F = E 0 and

ν = µ0 (as described in the first paragraph above). With f ∈ K f (X; E)
R
regarded as an element f0 of Kf (X; E0 ) , we write f dµ = z0 * ∈ E0 * , and
R
f0 dµ0 for the element of (E0 )0 * such that

DZ E Z
f0 dµ0 , z00 = hf0 , z00 i dµ0 for all z00 ∈ (E0 )0 .

In a sense, the preceding formula is all that the assertion in the text says:
when µ ∈ M (X; C) is a real measure, the formula (1) remains true R (by
definition!) with subscripts 0 installed. The proof that ψ(z 0 *) = f0 dµ0
will show the relevance of the formula to the identifications of the preceding
note.
Before proceeding, we note that

Z Z
(*) R f dµ = (Rf ) dµ for all f ∈ K (X; C)

R R
(because (Rf ) dµ and (I f ) dµ are real).
R
Let z00 * = ψ(z0 *) , where z0 * = f dµ and ψ : E0 * → (E0 )0 * is
the R-linear
R bijection of the preceding note. Our problem is to show that
z00 * = f0 dµ0 . By the definition of ψ , z00 * is the linear form on (E0 )0
such that



z00 *(z00 ) = Rz0 * v(z00 ) for all z00 ∈ (E0 )0 ,

where v : (E0 )0 → (E0 )0 is the vector space isomorphism described in the

§3 integrals of continuous vectorial functions INT III.x72

preceding note. Thus, for all z00 ∈ (E0 )0 ,

hz00 *, z00 i = (z00 *)(z00 )

= Rz0 * v(z00 )



= R z0 * v(z00 )


= R z0 *, v(z00 )
DZ E
=R f dµ, v(z00 )
Z
 
=R f , v(z00 ) dµ
Z


= R f , v(z00 ) dµ (by (*))
Z


= R v(z00 ) ◦ f dµ



= R v(z00 ) ◦ f , µ ,

briefly,



(**) hz00 *, z00 i = R v(z00 ) ◦ f , µ .




But v(z00 ) ◦ f (x) = v(z00 ) f (x) for all x ∈ X , and, by the definition
of v , 



R v(z00 ) f (x) = z00 f (x) ,


thus R v(z00 ) ◦ f = z00 ◦ f ; substituting this in (**), we have, for all
z00 ∈ (E0 )0 ,

hz00 *, z00 i = hz00 ◦ f , µi

= hz00 ◦ f0 , µ0 i ( z00 ◦ f is real-valued)
D Z E
= f0 dµ0 , z00 ,

R R

thus f0 dµ0 = z00 * = ψ(z0 *) = ψ f dµ .
III.35, `. 6.
“. . . tt u extends the mapping u .”
b ∈ E0 * defined by
To each y ∈ E corresponds the element y

b (y0 ) = y0 (y) = hy, y0 i

y for all y0 ∈ E0 .
INT III.x73 measures §3

z ∈ F0 * by
Similarly, each z ∈ F defines b

z(z0 ) = z0 (z) = hz, z0 i

b for all z0 ∈ F0 .

The assertion is that tt

u(b [ for all y ∈ E ; indeed,
y) = u(y) tt
u(b b ◦ tu
y) = y
and, for all z0 ∈ F0 ,


y ◦ t u)(z0 ) = y
(b b t u(z0 ) = y
b(z0 ◦ u)


[ 0) .
= (z0 ◦ u)(y) = z0 u(y) = u(y)(z

This observation is not needed for the Proposition 2 that follows it.
III.36, `. 5, 6.
“If E is complex, we equip E with its underlying real vector space
structure, which, as we have seen, does not modify the formula (1).”
Working through the identifications entailed in deriving the case for
complex E from the case for the real space E 0 underlying E promises to
be daunting. Instead, a more direct argument, patterned after the proof of
the real case, is given in the Note for III.36, `. 9–11.
III.36, `. 7.
“(i) We know . . .”
We know it from TVS, II, §5, No. 3, Cor. 5 of Prop. 4, p. TVS II.39.
Strictly speaking, by f (S) is here meant the canonical image

T = { fd
(x) : x ∈ S }

of f (S) in E0 * .
III.36, `. 9–11.
“ . . . it therefore suffices to prove that, for z 0 ∈ E0 , the relation
hf (x), z0 i> 0 for all x ∈ S implies
Z 
0
f dµ, z > 0 ”

The argument assumes that E is a locally convex space over R (after

which, an argument for the complex case is given). Here the subset f (S)
of E is being regarded as a subset of E 0 * , by identifying z ∈ E with the
linear form bz on E0 defined by b z(z0 ) = z0 (z) = hz, z0 i (z0 ∈ E0 ). Rather
than make the identification, let us instead work directly with the canonical
image of f (S) in E0 * , namely the set

T = { fd
(x) : x ∈ S } ⊂ E0 * .
§3 integrals of continuous vectorial functions INT III.x74

Since E0 * is equipped with the weak topology σ(E 0 *, E0 ) for the duality
defined by the bilinear form (f, z0 ) 7→ hf, z0 i = f (z0 ) ( f ∈ E0 *, z0 ∈ E0 ), its
dual for this topology is E0 , i.e., consists of the linear forms

f 7→ f (z0 ) = hf, z0 i (f ∈ E0 *) ,

where z0 varies over E0 . Thus the half-spaces whose intersection is C are

the half-spaces

Hz0 = {f ∈ E0 * : hf, z0 i = f (z0 ) > 0 }

defined by those z0 ∈ E0 for which T ⊂ Hz0 , that is, for which

d 0

f (x), z = z0 f (x) > 0 for all x ∈ S ,

in other words z0 ◦ f > 0 on S . Writing A for the set of all such z 0 , we

have \
C= H z0 .
z0 ∈A
R R
To prove that f dµ ∈ C , we need only show that f dµ ∈ Hz0 for every
z0 ∈ A , that is, Z
0
z ∈A ⇒ f dµ ∈ Hz0 ,

in other words, Z
T ⊂ H z0 ⇒ f dµ ∈ Hz0 ,

i.e., that
DZ E
0
z ◦ f > 0 on S ⇒ f dµ, z0 > 0 .

By definition (see the formula (1)),

DZ E Z Z
0 0
f dµ, z = hf , z i dµ = (z0 ◦ f ) dµ ,

thus the implication to be proved may be written

Z
0
z ◦ f > 0 on S ⇒ (z0 ◦ f ) dµ > 0 ;

since µ is positive and z0 ◦ f ∈ K (X; R) , the implication follows at once

from §2, No. 3, Cor. 2 of Prop. 8.
INT III.x75 measures §3

The complex case. The structure of the proof is the same as that of the
real case. Assuming E is a locally convex space over C , equip E 0 * with
the topology σ(E0 *, E0 ) relative to the duality defined by the bilinear form

(f, z0 ) 7→ hf, z0 i = f (z0 ) (f ∈ E0 *, z0 ∈ E0 ) .

We again write
T = { fd
(x) : x ∈ S } ⊂ E0 * ,


where fd (x)(z0 ) = z0 f (x) for z0 ∈ E0 , and define C to be the closure
in E0 * of the convex cone generated by T . The definition of C takes place
in the context of the real vector space (E 0 *)0 equipped with the topology
σ(E0 *, E0 ) .
In the duality between E0 and E0 * , the continuous linear forms on E0 *
are the forms
f 7→ f (z0 ) (f ∈ E0 *) ,

where z0 varies over E0 , thus the forms

(∗) f 7→ R f (z0 ) (f ∈ E0 *)

are continuous R-linear forms on E0 * , hence are continuous linear forms on

real space (E0 *)0 underlying E0 * . Conversely, if θ is a continuous linear
form on (E0 *)0 , then the formula

η(f ) = θ(f ) − i θ(if ) (f ∈ E0 *)

defines a continuous linear form on E 0 * , hence there exists

z0 ∈ E0 such that
0 0 0
η(f ) = f (z ) for all f ∈ E * , and so θ(f ) = R f (z ) . Thus (*) describes
all of the continuous linear forms on (E 0 *)0 . The closed half-spaces whose
intersection is C (TVS, II, §5, No. 3, Cor. 5 of Prop. 4) are therefore the
half-spaces

Hz0 = {f ∈ E0 * : R f (z0 ) > 0 }

defined by those z0 ∈ E0 for which T ⊂ Hz0 , that is, for which

R fd
(x)(z0 ) > 0 for all x ∈ S ,

in other words R(z0 ◦ f ) > 0 on S . Writing A for the set of all such z 0 ,
we have \
C= H z0
z0 ∈A
§3 integrals of continuous vectorial functions INT III.x76
R
and, as argued in the real case, to prove that f dµ ∈ C we need only prove
the implication
Z
0
R(z ◦ f ) > 0 on S ⇒ f dµ ∈ Hz0 ,

that is,
DZ E
0
R(z ◦ f ) > 0 on S ⇒ R f dµ, z0 > 0 ;

since
DZ E Z Z Z
0 0 0
R f dµ, z =R hf , z i dµ = R (z ◦ f ) dµ = R(z0 ◦ f ) dµ ,

and since µ is positive and R(z0 ◦ f ) ∈ K (X; C) , the last implication is a

consequence of §2, No. 3, Cor. 2 of Prop. 8.
Postlude. How to derive the complex case from the real case R more effi-
ciently? Given E complex, the real case applied to E 0 yields f0 dµ0 ∈ C0 ,
where C0 is the closed convex cone in (E0 )0 * generated by the canoni-
0
cal image of f0 (S) ( µ and µ0 have the same
support), where (E0 ) * is
0 0
equippedR with the topology Rσ (E0 ) *, (E0 ) . How to transform the re-
lation f0 dµ0 ∈ C0 into f dµ ∈ C ? In the note for III.34, `. −9
to R−6, an

R-linear
R bijection ψ : E0 * → (E0 )0 * was exhibited such that
ψ f dµ = f0 dµ0 . Presumably ψ(C) = C0 ; one needs to check that ψ
transforms the canonical image of f
(S) into the canonical image of f 0 (S) ,
and σ(E0 *, E0 ) into σ (E0 )0 *, (E0 )0 . . . but that’s where I lose the ball in
the sun.
III.36, `. 15.
“(ii) We know . . . ”
We know it from TVS, II, §5, No. 3, Cor. 1 of Prop. 4, p. TVS II.38.
III.36, `. 19.
“ . . . but this follows from §2, No. 3, Cor. 3 of Prop. 8.”
I don’t see how to apply the cited Cor. 3, where the hypothesis |f (x)| 6 a
(on Supp(µ) ) imposes a constraint on |f | with a > 0 , whereas here the
constraint is on f and a need not be > 0 . However, the proof of Cor. 3
provides the necessary hint.
The real case. It is assumed that E is a locally convex space over R
and that µ is a bounded positive measure on X . As in part (i), we argue
directly with the canonical image

T = { fd
(x) : x ∈ S } ⊂ E0 *
INT III.x77 measures §3

of f (S) in E0 * rather than identifying E with its canonical image. The

closed half-spaces in E0 * (for the topology σ(E0 *, E0 ) ) are the sets

Hz0 ,a = {f ∈ E0 * : f (z0 ) 6 a } ,

where z0 ∈ E0 and a ∈ R . Thus D is the intersection of the set of H z0 ,a

defined by those pairs (z0 , a) for which T ⊂ Hz0 ,a (TVS, II, §5, No. 3, Cor. 1
of Prop. 4, p. TVS II.38); writing A for the set of all such pairs, we have
\
D= Hz0 ,a .
(z0 ,a)∈A
R R
To prove that f dµ ∈ kµk · D , we need only show that f dµ ∈ kµk · Hz0 ,a
for every (z0 , a) ∈ A , i.e., that
Z
(z0 , a) ∈ A ⇒ f dµ ∈ kµk · Hz0 ,a ,

in other words that

Z
−1
(∗) T ⊂ Hz0 ,a ⇒ kµk f dµ ∈ Hz0 ,a .

The condition on the left side of (*) says that

d 0
f (x), z 6 a for all x ∈ S ,

that is,

(a) z0 ◦ f 6 a on S .

and the condition on the right side says that

Z
(b) (z0 ◦ f ) dµ 6 akµk ,

thus our problem is to show that (a) ⇒ (b).

Suppose z0 ◦ f 6 a on S . Let g ∈ K (X; R) be such that 0 6 g 6 1
on X and g(x) = 1 for x ∈ Supp(z0 ◦ f ) (§1, No. 2, Lemma 1). Then

z0 ◦ f = (z0 ◦ f )g 6 ag on S ,
R
that is, ag − z0 ◦ f > 0 on S , therefore (ag − z0 ◦ f ) dµ > 0 by §2, No. 3,
Cor. 2 of Prop. 8, thus
Z
(∗∗) (z0 ◦ f ) dµ 6 aµ(g) .
§3 integrals of continuous vectorial functions INT III.x78

If we show that aµ(g) 6 akµk , the implication (a) ⇒ (b) will be proved. At
any rate, 0 6 µ(g) = |µ(g)| 6 kµk kgk = kµk .
case 1: If a > 0 then aµ(g) 6 akµk by the preceding inequality, there-
fore (b) follows from (**).
case 2: If a < 0 , then z0 ◦ f 6 a < 0 on S , therefore S ⊂ Supp(z0 ◦ f )
(in particular, µ has compact support), and since g = 1 on Supp(z 0 ◦ f ) ,
we infer that g = 1 on S . If we show that µ(g) = kµk then (**) will again
yield (b). Since g > 0 , for every h ∈ K (X; R) we have

−khkg 6 hg 6 khkg ,

therefore
−khkg 6 h 6 khkg on S ,
whence (since µ > 0 )

−khkµ(g) 6 µ(h) 6 khkµ(g) ;

thus |µ(h)| 6 khkµ(g) for all h ∈ K (X; R) , consequently kµk 6 µ(g) . But

µ(g) = |µ(g)| 6 kµk kgk = kµk ,

therefore µ(g) = kµk , which completes the proof.

The complex case. Assume E is a locally convex space over C . As ob-
served in the complex case for (i), the continuous (for σ(E 0 *, E0 ) )
R-linear forms on E0 * are the forms


f 7→ R f (z0 ) (f ∈ E0 *) ,

where z0 varies over E0 , thus the closed half-spaces of (E0 *)0 are the sets


Hz0 ,a = {f ∈ E0 * : R f (z0 ) 6 a } ,

where z0 ∈ E0 and a ∈ R , and D is the intersection of the set of all H z0 ,a

defined by those pairs (z0 , a) for which T ⊂ Hz0 ,a (TVS, §5, No. 3, Cor. 1
of Prop. 4, p. TVS II.38); writing A for the set of all such pairs, we have
\
D= Hz0 ,a .
(z0 ,a)∈A
R R
To prove that f dµ ∈ kµk·D , we need only show that kµk −1 f dµ ∈ Hz0 ,a
for every (z0 , a) ∈ A , i.e., that
Z
0 −1
(z , a) ∈ A ⇒ kµk f dµ ∈ Hz0 ,a ,
INT III.x79 measures §3

in other words that

Z
T ⊂ Hz0 ,a ⇒ kµk−1 f dµ ∈ Hz0 ,a .

0
R says0 that R(z ◦ f ) 6 a on S , and the condition
The condition on the left
on the right says that R(z ◦ f ) dµ 6 akµk , thus the problem is to show
that
Z
0
(†) R(z ◦ f ) 6 a on S ⇒ R(z0 ◦ f ) dµ 6 akµk .

Suppose R(z0 ◦ f ) 6 a on S . Let g ∈ K (X; C) with 0 6 g 6 1 on X

and g = 1 on Supp R(z0 ◦ f ) . Then


a − R(z0 ◦ f ) g > 0 on S ,

whence
R(z0 ◦ f ) = R(z0 ◦ f )g 6 ag on S ,
therefore (since µ is positive)
Z
(††) R(z0 ◦ f ) dµ 6 aµ(g) .

case 1: If a > 0 then aµ(g) 6 akµk as in the real case, and (†) then
follows from (††) .
case 2: If a < 0 then S ⊂ Supp R(z0 ◦ f ) as in the real case, therefore
g = 1 on S . Then, for every h ∈ K (X; C) , we have |h| 6 khkg on S ,
hence µ(|h|) 6 khkµ(g) ; therefore

|µ(h)| 6 |µ|(|h|) = µ(|h|) 6 khkµ(g) ,

which shows that kµk 6 µ(g) . But µ(g) 6 kµk kgk = kµk , thus µ(g) = kµk
and (†) again follows from (††).
III.36, `. −9.
“ . . . then ν is bounded”
Let U be an open set in X such that K ⊂ U and U is compact. By
Prop. 5 of §2, No. 2,

Supp (ν) = U ∩ Supp (µ) ⊂ U ⊂ U ,

thus ν has compact support, hence is bounded (§2, No. 3, Prop. 11).
§3 integrals of continuous vectorial functions INT III.x80

III.36, `. −9. R R
“ . . . and f dµ = f dν ∈ kνk · D by Prop. 4, (ii).”

Since Supp(f ) = K ⊂ U , we haveR f ∈ K (X, U; E) , hence f U ∈
R (U;
K  E) (see RIII.23,
 `. 14–16).
 Thus by f dν (abuse of notation)
 is meant
(f U) dν = (f U) d(µU) which, by the definition of µU , is equal to
R
f dµ since, for all z0 ∈ E0 ,
DZ  E 

D Z E
(f U) dν, z0 = ν z0 ◦ (f U) = ν (z0 ◦ f )U = µ(z0 ◦ f ) = f dµ, z0 .

R 
By the cited Prop. 4, (ii), (f U) dν ∈ kνk · D1
, where D
1 is the closed
convex envelope of (the canonical image of) f U Supp(ν) in E0 * . Now,






f U Supp(ν) = f U U∩Supp(µ) = f U∩Supp(µ) = f (U∩S) ⊂ f (S) ,

thus D1 ⊂ D , and so
Z Z

f dµ = (f U) dν ∈ kνk · D1 ⊂ a · D

with a = kνk .
III.36, `. −8, −7.
“The second result follows from this, since any complex measure may
be written as µ1 − µ2 + iµ3 − iµ4 , where the µj are positive.”
Attention must be paid to the supports of the positive measures µ j .
As noted in §2, No. 2, Prop. 2, if ν is a real measure (so that ν = ν + − ν − ),
then
Supp(ν) = Supp(|ν|) = Supp(ν +) ∪ Supp(ν −) ;
and if ν is a complex measure then |Rν| 6 |ν| and |I ν| 6 |ν| (§1, No. 6,
(17)) show that the supports of Rν and I ν —hence of their positive and
negative parts—are contained in Supp(|ν|) = Supp(ν) . Thus ν has a rep-
resentation of the indicated sort with Supp(µ j ) ⊂ Supp(ν) for 1 6 j 6 4 .
Then, by the previous result, for each j there exists a number a j > 0 such
that Z
f dµj ∈ aj · Dj ,


where Dj is the closed convex envelope of f Supp(µj ) ; but Supp(µj ) ⊂
Supp(ν) implies that Dj ⊂ D , thus
Z
f dµj ∈ aj · D for 1 6 j 6 4 ,
INT III.x81 measures §3

and the second result follows from the relation

Z Z Z Z Z
f dν = f dµ1 − f dµ2 + i f dµ3 − i f dµ4

(Prop. 1).
III.36, `. −1.
“ . . . E0 * is complete”
For the uniform structure on E0 * derived from the weak topology
σ(E *, E0 ) (TVS, II, §6, No. 7, Prop. 9).
0

III.37, `. R1, 2.
“ . . . f dµ ∈ C for every measure µ belonging to the convex set H of
positive measures on X of total mass equal to 1 .”
(As in previous notes, I regard C to be the closed convex envelope of
the canonical image {fd (x) : x ∈ X} of f (X) in E0 * .) The point is that for
µ ∈ H one has R kµk = µ(1) = 1 (§1, No. 8, Cor. 2 of Prop. 10). By
Prop. 4, (ii), f dµ belongs to the closed convex envelope D of (the canon-
ical image of) f (S) in E0 * , where S = Supp(µ) , and since f (S) ⊂ f (X)
implies D ⊂ C , one has
Z
f dµ ∈ kµk · D = D ⊂ C

as claimed.
III.37, `. 5–7. R
“ . . . the image of H0 under the mapping µ 7→ f dµ is the convex
envelope C0 of f (X) in E0 * .”
P
n
If µ is a measure with finite support, say µ = ci εxi , where the xi
i=1
are distinct points of X (§2, No. 4, Prop. 12), then µ is bounded, kµk =
Pn
|ci | (see item (12) in the note for III.16, `. 13–15) and µ > 0 ⇔ c i > 0
i=1
for all i , thus
n
X
µ ∈ H0 ⇔ ci > 0 for all i and ci = 1 ;
i=1

that is, µ ∈ H0 ⇔ µ is a convex combination of Dirac measures ε x (x ∈ X) ,

and, for such a measure µ ,
Z n
X n
X
f dµ = µ(f ) = ci εxi (f ) = ci f (xi ) ,
i=1 i=1
§3 integrals of continuous vectorial functions INT III.x82
R
whence f dµ : µ ∈ H0 = C0 .
III.37, `. 7, 8.
“ . . . this mapping is continuous for the vague topology on M (X; C)
and the topology σ(E0 *, E0 ) on E0 * . . . ”
If µj → µ vaguely, that is, if µj (g) → µ(g) for all g ∈ K (X; C) , then
in particular
µj (z0 ◦ f ) → µ(z0 ◦ f ) for all z0 ∈ E0 ,

R 
R  R
Rthat is, f dµj , z0 → f dµ, z0 for all z0 ∈ E0 , in other words f dµj →
f dµ in E0 * for σ(E0 *, E0 ) .
III.37, `. 9, 10.
“ . . . the image of H = H0 is a compact convex set containing C 0 and
contained in C ”
R
Writing Φ : M (X; C) → E0 * for the mapping µ 7→ f dµ , which we
know to be linear and continuous for the indicated topologies, we have

C0 = Φ(H0 ) ⊂ Φ(H) = Φ(H0 ) ⊂ Φ(H0 ) = C0 = C .

III.37, `. 17.
“ . . . therefore D = D◦◦ (TVS, II, §6, No. 3, Cor. 3 of Th. 1).”
Moreover, since D is balanced—hence symmetric or circled, according
as E is real or complex—one has

D◦ = {z0 ∈ E0 : |hz, z0 i| 6 1 for all z ∈ D } ,

D◦◦ = {z ∈ E : |hz, z0 i| 6 1 for all z0 ∈ D◦ }

(TVS, II, §6, No. 3 and §8, No. 4).

III.37, `. 18, 19.
“It therefore suffices to prove that for every z 0 ∈ D◦ ,
 Z  Z
 
 0 
f dµ, z  6 (q ◦ f ) d|µ| ”


R Let us first prove the inequality, then show why it is sufficient.

R By
f dµ ∈ E is meant that there exists a vector z 0 ∈ E such that f dµ = b
z0
(the canonical image of z0 in E0 * ), that is, such that
Z
0
hz0 , z i = (z0 ◦ f ) dµ for all z0 ∈ E0 .
INT III.x83 measures §3

Let z0 ∈ D◦ . This means that

z ∈ D ⇒ |hz, z0 i| 6 1 ,

that is,

(α) q(z) 6 1 ⇒ |hz, z0 i| 6 1 .

It follows that

(β) |hz, z0 i| 6 q(z) for all z ∈ E ;

for, if q(z) = 0 then q(nz) = nq(z) = 0 for n = 1, 2, 3, . . ., whence

|hnz, z0 i| 6 1 for all n by (α), therefore |hz, z 0 i| = 0 and (β) reduces to
0 6 0 ; whereas if q(z) > 0 then (β) results from applying (α) to the vector

−1
q(z) z . In particular,


|hf (x), z0 i| 6 q f (x) for all x ∈ X .

that is, |z0 ◦f | 6 q ◦f ; since both sides belong to K (X; R) and since |µ| > 0 ,
we infer that
|µ|(|z0 ◦ f |) 6 |µ|(q ◦ f ) ;
but |µ(z0 ◦ f )| 6 |µ|(|z0 ◦ f |) by the inequality (13) of §1, No. 6, therefore

|µ(z0 ◦ f )| 6 |µ|(q ◦ f ) ,

that is,
D Z E Z
 0 
 f dµ, z  6 (q ◦ f ) d|µ| ,

which is the asserted inequality; in other words,

Z
0
(γ) |hz0 , z i| 6 (q ◦ f ) d|µ| for all z0 ∈ D◦ .

R
Write a = (q ◦ f ) d|µ| ; the assertion of the proposition is that

(δ) q(z0 ) 6 a .

If a = 0 then, by (γ), |hz0 , z0 i| = 0 for all z0 ∈ D◦ , whence |hnz0 , z0 i| = 0

for all n and all z0 ∈ D◦ , therefore nz0 ∈ D◦◦ = D for all n , that is,
q(nz0 6 1 for all n and so q(z0 ) = 0 ; thus (δ) reduces to 0 6 0 . Whereas
if a > 0 then it follows from (γ) that |ha −1 z0 , z0 i| 6 1 for all z0 ∈ D◦ , thus
§3 integrals of continuous vectorial functions INT III.x84

a−1 z0 ∈ D◦◦ = D , that is, q(a−1 z0 ) 6 1 ; thus q(z0 ) 6 a and (δ) is again
verified.
R
We remark that if E is a Banach space, then the condition f dµ ∈ E
is automatically verified (by Cor. 1 of Prop. 7 in No. 3 below), therefore
Z Z

f dµ 6 kf (x)k d|µ|(x)

R
by the foregoing result; and if X is compact, f dµ 6 kf k∞ kµk .
III.38, `. 3–5.
“ . . . the closed convex envelope C of f (X) in E is then precompact (for
the uniform structure induced by that of E ) (TVS, II, §4, No. 1, Prop. 3).”
By the cited Prop. 3, the balanced convex envelope B 0 of f (X) in E is
precompact, therefore so is its closure B = B 0 (GT, II, §4, No. 2, Prop. 1);
since B is closed and convex (it is the closed balanced convex envelope
of f (X) by TVS, I, §1, No. 5, Prop. 2 and II, §2, No. 6, Prop. 14) it con-
tains C , therefore C is also precompact (GT, II, §4, No. 2, Prop. 1).
In fact, C is compact, as noted in the Corollary of the cited Prop. 3.
III.38, `. 8, 9.
“ . . . C is the closed convex envelope of f (X) in E 0 * for the topology
σ(E *, E0 ) ”
0

Let θ : E → E0 * be the canonical mapping θ(z) = b z(z0 ) =

z , where b
hz, z0 i = z0 (z) for all z0 ∈ E0 . Equip E with the weakened topology
σ(E, E0 ) , and E0 * with the topology σ(E0 *, E0 ) . Note that θ is contin-
uous for these topologies; for, since σ(E 0 *, E0 ) is the initial topology for the
family of linear forms
f 7→ f (z0 ) (f ∈ E0 *)

indexed by z0 ∈ E0 , it suffices to observe that the composite functions

z 7→ θ(z) = b z(z0 ) = z0 (z)

z 7→ b

are continuous for σ(E, E0 ) . The
assertion to be proved is that θ(C) is the
closed convex envelope of θ f (X) in E0 * .
Let C0 be the convex envelope of f (X) in E . The closure of C 0 in E
is the same for σ(E, E0 ) and the original topology of E (TVS, IV, §1, No. 2,
Prop. 2), therefore C is also the closed convex envelope of f (X) in E for
σ(E, E0 ) , that is, C = C0 (closure for either topology).


Let D0 be the convex envelope of θ f (X) in E0 * and let D = D0 be
its closure. Since θ is linear, θ(C0 ) = D0 . We know that C is compact
INT III.x85 measures §3

in E for σ(E, E0 ) , therefore θ(C) is compact in E0 * , hence closed in E0 * .

Thus
θ(C) = θ(C0 ) ⊂ θ(C0 ) = D0 = D ;


but θ(C) is a closed convex set in E0 * that contains θ f (X) , therefore
θ(C) ⊃ D , and finally θ(C) = D . In particular, D ⊂ θ(E) .
III.38, `. 9, 10.
“ . . . the proof is therefore concluded by the Corollary of Prop. 4 of
No. 2.”
By the cited corollary, there exist scalars a 1 , a2 , a3 , a4 such that
Z
f dµ ∈ a1 D − a2 D + ia3 D − ia4 D ,

where D is defined as in the precedingR note, and that note R shows that
ak D ⊂ θ(E) for 1 6 k 6 4 , therefore f dµ ∈ θ(E) ; that is, f dµ = b z0 for
some z0 ∈ ER . R
When f dµ = b z0 , z0 ∈ E , it is useful
R to redefine f dµ to be the
unique element z0 ∈ E such that hz0 , z0 i = hf , z0 i dµ for all z0 ∈ E0 .
III.38, `. 13–14.
“Since the duals of E and E b are identical, it suffices to apply Prop. 7
while regarding f as taking its values in E b .”
Strictly speaking, if E is regarded as a (dense) linear subspace of E b,
and if f is regarded as taking its values in E b , that is, as an element of
R
b
K (X; E) , then f dµ becomes (by Prop. 7) the unique element w 0 ∈ E b
R
0 0
such that hw0 , w i = hf , w i dµ for all w ∈ (E) 0 b . The fact that w0
0

extends an element of E0 is of no particular interest here.

However, to apply Prop. 7, it is necessary to note that E b is locally con-
vex. A roundabout proof using continuous semi-norms is indicated in TVS,
II, §4, No. 1, remarks following the Corollary of Prop. 1; a brief direct proof,
given in the book of J. Horváth (Topological vector spaces and distributions,
vol. 1, p. 134, Addison-Wesley, Reading, Mass., 1966), is as follows.
We can regard E as a linear subspace of E b , equipped with the topology
b
induced by that of E . Let N be any fundamental system of neighborhoods
of 0 in E , and let N1 be the set of closures in E b of the sets V ∈ N . It
will suffice to show that N1 is a fundamental system of neighborhoods of 0
in Eb ; for, the sets in N may be taken to be convex, and the closure of a
convex set, in any topological vector space, is convex (TVS, II, §2, No. 6,
Prop. 14).
At any rate, since E is dense in E b , the sets in N1 are neighborhoods
of 0 in E b (GT, I, §3, No. 1, Prop. 2). Let W be a closed neighborhood of 0
§3 integrals of continuous vectorial functions INT III.x86

b ; since such neighborhoods are fundamental (GT, II, §1, No. 2, Cor. 3
in E
of Prop. 2 and III, §3, No. 1), it will suffice to find a set V ∈ N such that
V ⊂ W (closure in E b ). Since E has the topology induced by E b , W ∩ E is
a neighborhood of 0 in E ; by assumption, there exists a set V ∈ N such
that V ⊂ W ∩ E , whence V ⊂ W = W .
III.38, `. 18.
“ . . . hence bounded . . . ”
Every precompact subset of a locally convex space is bounded (TVS,
III, §1, No. 2, Prop. 2).
An argument that does not appeal to the concept of total boundedness:
since f (X) is compact, it is bounded (by a simplification of the proof of
the Prop. 2 just cited, avoiding the concept of the completion of a uniform
space), therefore its closed convex envelope is also bounded (TVS, III, §1,
No. 2, Prop. 1).
III.38, `. −11 to −9.
“ 2◦ E is the dual of a barreled Hausdorff locally convex space G , and
E is equipped with an S-topology, where S is a covering of G by bounded
subsets (TVS, III, §4, No. 2, Cor. 4 of Th. 1).”
The adaptation of the cited Cor. 4 to the present notation is as follows:
Let G and F be two locally convex spaces, S a covering of G consisting
of bounded subsets. If G is barreled and F is Hausdorff and quasi-complete,
then the space LS (G; F) is Hausdorff and quasi-complete.
One takes F = K , where K = R or C according as the spaces are
real or complex, and one sets E = LS (G; K) . Recall that L (G; K) is the
vector space G0 of continuous linear forms on G , and E = L S (G; K) is
the vector space G0 equipped with the topology of uniform convergence in
the subsets M ∈ S of G .
III.38, `. −8, −7.
“For example, Cor. 2 of Prop. 7 can be applied when E is the weak
dual of a Banach space”
In the preceding note, one takes G to be a Banach space and S to
be the set of its 1-element subsets {y} (y ∈ G) (equivalently, the set of its
finite subsets).
III.38, `. −7, −6.
“ . . . or a space of measures M (Y; C) equipped with the vague topol-
ogy.”
Here one takes (in the foregoing) G = K (Y; C) equipped with the
direct limit topology (a barreled space, by §1, No. 1, Prop. 2) and S to be
the set of its 1-element subsets {f } (f ∈ K (Y; C) ) (§1, No. 9).
INT III.x87 measures §3

III.39, `. 2–5.
“Proposition 8. . . . ”
R
When f dµ ∈ E for all f ∈ K (X; E) (for example, when E is quasi-
complete, cf. No. 3, Cor. 2 of Prop. 7), it is a corollaryR of Prop. 8 that
the mapping Φ0 : K (X; E) → E defined by Φ0 (f ) = f dµ is the unique
continuous linear mapping K (X; E) → E such that g ·a 7→ µ(g)·a for every
vector a ∈ E and every function g ∈ K (X; C) . For, since E is a topological
subspace of Eb , equivalently, the topology of E is the initial topology for the
canonical injection ι : E → E b , the continuity of Φ0 is equivalent to that of
ι ◦ Φ0 = Φ (GT, I, §2, No. 3, Prop. 4).
III.39, `. 6–8. R
“To prove the continuity of the mapping f 7→ f dµ , it suffices to show
that its restriction to K (X, K; E) is continuous for every compact subset
K of X (TVS, II, §4, No. 4, Prop. 5).”
The issue has essentially been addressed in §1, No. 1 (see the note for
III.2, `. 7, 8); as some aspects are employed here for the first time, a review
is in order. R
In line with No. 3, Cor. 1 of Prop. 7, we regard f dµ as the unique
R
element w0 ∈ E b such that hw0 , w0 i = hf , w0 i dµ for all w0 ∈ (E)
b 0 , so that
R
f 7→ f dµ is a linear mapping of K (X; E) into the locally convex space E b.
Let Tu be the topology on K (X; E) of uniform convergence in X .
For every compact subset K of X , the topology on K (X, K; E) is by def-
inition the topology TK induced by Tu , that is, the topology of uniform
convergence (in X or in K , it comes to the same). On the other hand, the
(direct limit) topology T on K (X; E) (§1, No. 1) is by definition the “final
locally convex topology” for the family of mappings u K , where K runs over
the set of all compact subsets of X , and u K : K (X, K; E) → K (X; E) is
the canonical injection; that is, T is the finest locally convex topology on
K (X; E) for which all the uK are continuous (TVS, II, §4, No. 4, Prop. 5
and Example).
One knows that if G is a locally convex space and u : K (X; E) → G
is a linear mapping, then u is continuous (for T ) if and only if u ◦ u K is
continuous for every K (TVS, loc. cit.). The assertion at hand then follows
from the observation that u ◦ uK is the restriction of u to K (X, K; E) .
{In particular, when E = C , a linear form on K (X; E) is a measure
on X if and only if its restriction to every K (X, K; C) is continuous—as
noted following the definition of a measure (§1, No. 3)}.
To conclude the review, since Tu obviously renders the uK continuous,
T is finer than Tu , and the inclusion T ⊃ Tu may be proper (for example,
when X = N equipped with the discrete topology; see item (9) in the note
§3 integrals of continuous vectorial functions INT III.x88

for III.16, `. 13–15). Nevertheless, T and T u induce the same topology

TK on K (X, K; E) ; for, since uK is continuous when K (X; E) is equipped
with T , one has
T ∩ K (X, K; E) ⊂ TK ,
whereas, since Tu ⊂ T ,

TK = Tu ∩ K (X, K; E) ⊂ T ∩ K (X, K; E) ,

whence equality throughout. Also, as noted in §1, No. 1, Prop. 1, K (X, K; E)

is closed in K (X; E) for both T and Tu .
III.39, `. 8–11.
“We note that if the topology of E is defined by a family of semi-
norms (qα ) , that of K (X, K; E) is defined by the family of semi-norms


pα (f ) = sup qα f (x) .”
x∈K

The chain of reasoning is quite long; if the Corollary of Prop. 1 in

TVS, II, §4, No. 1, its version for complex spaces, and the definition of the
uniform structure of K (X, K; E) are well-digested, the following review can
be omitted.
(1) If (Nα ) is a fundamental system of neighborhoods of 0 in E , then
the sets
Vα = {(x.y) ∈ E × E : x − y ∈ Nα }
form a fundamental system of entourages for the uniform structure of E .
For, the uniform structure of E is by definition that of its underlying topo-
logical group structure under addition (TVS, I, §1, No. 4 and GT, III, §3,
No. 1, remark following Def. 1).
(2) Recall that in a complex topological vector space, the topology of
the underlying real topological vector space is by definition the same as
the original topology on the space (TVS, II, §8, No. 1; in particular, the
open sets, closed sets, closure, interior, neighborhoods, continuity, uniform
structure are the same–as well, of course, the convex sets).
The balanced sets for the complex structure are balanced for the real
structure, but not vice versa; those for the complex structure are circled,
whereas those for the real structure are symmetric. Sets absorbent for
the complex structure are absorbent for the real structure—so to speak,
C-absorbent ⇒ R-absorbent—but not vice versa (but neighborhoods of 0
are absorbent in both senses, since the R-neighborhoods and C-neighbor-
hoods are the same).
INT III.x89 measures §3

(3) When E is complex, the official definition of locally convexity is that

the underlying real topological vector space E 0 is locally convex, i.e., has a
fundamental system of convex neighborhoods of 0 (TVS, II, §8, No. 2). In
such a space, every neighborhood V of 0 in E contains a convex neighbor-
hood of 0 in E ; for, V is also a neighborhood of 0 in E 0 , hence contains
a convex neighborhood W of 0 in E0 , and W is also a neighborhood of 0
in E .
Thus one could have defined a complex topological vector space to be
locally convex if it has a fundamental system of convex neighborhoods, and
this would automatically have made E 0 locally convex. On the other hand, if
one starts with a real locally convex space F admitting an automorphism u
such that u2 = −1 ( 1 the identity mapping of F ), then F can be given the
structure of a complex vector space E such that, with the same topology,
E is a complex topological vector space and E 0 = F (TVS, II, §8, No. 1,
the paragraphs in fine print), and E is then automatically locally convex by
the author’s definition.
(4) A locally convex space (in particular, the given space E ) has a
fundamental system of closed, balanced, convex neighborhoods of 0 (TVS,
II, §8, No. 2).
For, let N be a closed neighborhood of 0 in E (such neighborhoods
are fundamental). Since N is also a neighborhood of 0 in E 0 , it contains
a convex neighborhood V of 0 in E0 . But V is also a neighborhood of 0
in E , hence it contains a balanced neighborhood W of 0 in E (TVS, I, §1,
No. 5, Prop. 4). Then V contains the convex envelope U of W , and U
is obviously balanced (cf. TVS, II, §2, No. 3, Cor. 1 of Prop. 8), therefore
its closure U is balanced (TVS, I, §1, No. 5, Prop. 2) and convex (TVS,
II, §2, No. 6, Prop. 14). Finally, the inclusions W ⊂ U ⊂ V ⊂ N yield
W ⊂ U ⊂ N = N , thus N contains a neighborhood U of 0 of the desired
type.
(5) If p is a continuous semi-norm on a topological vector space, then
the set A = {x : p(x) 6 1 } is a closed, balanced, convex neighborhood
of 0 .
For, by continuity, A is closed, {x : p(x) < 1 } is an open set contain-
ing 0 , and A is obviously balanced and convex. Conversely:
(6) If G is any real or complex topological vector space and A is a
closed, balanced, convex neighborhood of 0 in G , then:
◦
(i) the interior A of A is convex;
◦
(ii) the closure of A is equal to A ;
(iii) there exists a (unique) semi-norm p on G such that

A = {x : p(x) 6 1 } ;
§3 integrals of continuous vectorial functions INT III.x90

(iv) p is continuous (hence uniformly continuous) on G .

Items (i) and (ii) are consequences of TVS, II, §2, No. 6, Cor. 1 of
Prop. 16.
Proof of (iii): As in TVS, II, §2, No. 11, Prop. 22, define a function
p = pA > 0 (called the gauge of A ) on the space G by

p(x) = inf {ρ : ρ > 0 , x ∈ ρA } ;

since A is absorbent (it is a neighborhood of 0 ) one knows from the cited

Prop. 22 that p is finite, p(0) = 0 and

p(x + y) 6 p(x) + p(y) , p(λx) = λ p(x) for λ > 0 .

When G is real, A is symmetric, thus x ∈ ρA ⇔ −x ∈ ρA , whence

p(−x) = p(x) ; and if λ < 0 then p(λx) = p(−λx) = p(|λ|x) = |λ|p(x) ,
therefore
p(λx) = |λ| p(x) for all scalars λ .
The same formula holds when G is complex; for, writing |λ| = λγ with
|γ| = 1 , since A is circled we have γA = A , therefore

|λ| z = λγ z ∈ ρA ⇔ λz ∈ ρA ,

whence p(λz) = p(|λ|z) = |λ| p(z) . Thus p is a semi-norm in both cases,

and, by Prop. 23 (loc. cit.), A = {x : p(x) 6 1 } .
In general, a semi-norm p on a real or complex vector space is char-
acterized by the set {x : p(x) 6 1 } . More generally, if p and q are
semi-norms then

p 6 q ⇔ {x : p(x) 6 1 } ⊃ {x : q(x) 6 1 };

for, assuming that the inclusion on the right holds, given any vector x and,

−1
for any ε > 0 , setting yε = q(x) + ε x , one has

−1
q(yε ) = q(x) + ε q(x) < 1 ,

−1
therefore p(yε ) 6 1 by hypothesis, and so q(x) + ε p(x) 6 1 , that is,
p(x) 6 q(x) + ε .
(iv) Given any ε > 0 , one has

p(x) 6 ε ⇔ p(ε−1 x) 6 1 ⇔ ε−1 x ∈ A ⇔ x ∈ εA ;

INT III.x91 measures §3

in particular, since εA is a neighborhood of 0 in G , the implication

x ∈ εA ⇒ p(x) 6 ε shows that p is continuous at 0 . The uniform conti-
nuity of p follows from the implication

x − y ∈ εA ⇒ |p(x) − p(y)| 6 p(x − y) 6 ε ,

which says that if (x, y) belongs to the

entourage defined by the neighbor-
hood εA of 0 in G , then p(x), p(y) belongs to the entourage defined by
the neighborhood [ − ε, ε] of 0 in R .
Though it is not needed for the moment, this is a good place to record
the fact that
◦
(v) {x : p(x) < 1 } = A .

For, writing U for the set on the left, U is convex and it is open by the
◦
continuity of p , whence U ⊂ A ; and U ⊂ A = A . In fact, U = A ; for,
if x ∈ A and 0 < α < 1 then p(αx) = α p(x) 6 α < 1 , thus αx ∈ U ,
and, since αx → x as α → 1 , every neighborhood of x contains such a
vector αx , whence x ∈ U . Since U is a convex open set, it is equal to
the interior of its closure (TVS, II, §2, No. 7, Cor. 1 of Prop. 16), that is,
◦
U = A.
(7) Let (Nα ) be a fundamental system of neighborhoods of 0 in E , so
that, by (1), the sets

Vα = {(x, y) ∈ E × E : x − y ∈ Nα }

are a fundamental system of entourages for the uniform structure of E .

If X is any nonempty set and F = F (X; E) is the vector space of
all functions f : X → E (with the pointwise operations), then (GT, X, §1,
No. 1) the sets


Wα = {(f , g) ∈ F × F : f (x), g(x) ∈ Vα for all x ∈ X }

are a fundamental system of entourages for a uniformity on F that induces

on F the topology of uniform convergence in X , and the sets

Uα = {f ∈ F : f (x) ∈ Nα for all x ∈ X }

form a fundamental system of neighborhoods of 0 ∈ F . This topology

makes F a topological group under addition; for, given any index α , there
exist indices β and γ such that

Nβ + Nβ ⊂ Nα and Nγ ⊂ −Nα
§3 integrals of continuous vectorial functions INT III.x92

(GT, III, §1, No. 2), whence Uβ + Uβ ⊂ Uα and Uγ ⊂ −Uα . However,

F need not be a topological vector space: for example, if X is infinite and
f0 ∈ F (X; R) is unbounded, then λf0 does not converge to 0 as λ → 0 ,
indeed, sup |λf0 (x)| = +∞ for λ 6= 0 (TVS, I, §1, No. 1, Example 4).
x∈X
For every subset S of F , there is an induced uniformity: in the
foregoing, one replaces “ f ∈ F ” by “ f ∈ S ”; thus if g ∈ S then the sets

{f ∈ S : f (x) − g(x) ∈ Nα for all x ∈ X }

form a fundamental system of neighborhoods of g for the topology on S of

uniform convergence in X . When X is a compact space, the linear subspace
S = C (X; E) is a topological vector space for this topology (TVS, loc. cit.).
More generally, if X is a locally compact space and K is a compact subset
of X , then S = K (X, K; E) is a topological vector space for the topology
of uniform convergence in X , since K (X, K; E) may be identified with a
(closed) linear subspace of C (K; E) (§1, No. 1).
(8) Assume now that X is locally compact and K is a compact subset
of X . We know from (7) that K (X, K; E) is a topological vector space
for the topology of uniform convergence in X (equivalently, in K ). By (4)
and (6), we can suppose that the neighborhoods N α of 0 ∈ E are given by
continuous semi-norms qα ,

Nα = {x ∈ E : qα (x) 6 1 } ;

then (with Uα defined as in (7))

Uα = {f ∈ K (X, K; E) : qα f (x) 6 1 for all x ∈ X }.



For every f ∈ K (X, K; E) , qα f (X) ∪ {0} = qα f (K) ∪ {0} is a compact
subset of R , therefore the formula



pα (f ) = sup qα f (x) = sup qα f (x)
x∈X x∈K

defines a (finite) semi-norm pα on K (X, K; E) , such that

(∗) Uα = {f ∈ K (X, K; E) : pα (f ) 6 1 } .

The neighborhoods Uα of 0 in K (X, K; E) are thus convex, so K (X, K; E)

is a locally convex space; moreover, it is clear from (*) that p α is continuous
at 0 , hence uniformly continuous, by the argument in (6), (iv), therefore U α
is closed.
The formula (*) already establishes what we set out to prove: the topol-
ogy of K (X, K; E) is defined by the family of semi-norms (p α ) .
INT III.x93 measures §3

III.39, `. 13–16.
“ . . . by No. 2, Prop. 6 we have, for every function f ∈ K (X, K; E) ,
Z  Z  Z


qα f dµ = qα hf dµ 6 h(x)qα f (x) d|µ|(x) 6 |µ|(h) · pα (f )

(the Rqα being extended by continuity to E b ), which proves the continuity of

f 7→ f dµ .”
Superficially straightforward, the proof entails a subtle interplay be-
tween semi-norms on E , E b and the corresponding function spaces; the
details are as follows.
(1) It is more convenient to start from neighborhoods of 0 in E b . We
know that E b is locally convex (see the note for III.38, `. 13,14). By the pre-
ceding note, E b has a fundamental system (N∗α ) of closed, balanced, convex
neighborhoods of 0 , expressed by a family (q α∗ ) of continuous semi-norms,
b : qα∗ (x∗ ) 6 1 } .
N∗α = {x∗ ∈ E
b , equipped with the relative
We can regard E as a dense linear subspace of E
∗
topology; then the sets Nα = Nα ∩ E form a fundamental system of closed,
balanced, convex neighborhoods of 0 in E , and, setting q α = qα∗ E , the qα
are continuous semi-norms on E such that
Nα = {x ∈ E : qα (x) 6 1 } .
For f ∈ K (X, K; E) we write, as earlier in the proof,


pα (f ) = sup qα f (x) .
x∈K

(2) With the above notations, the displayed inequalities in the text
b ,
become, for all f * ∈ K (X, K; E)
Z  Z  Z
∗ ∗ ∗

qα f dµ = qα hf dµ 6 h(x)qα∗ f ∗ (x) d|µ|(x) 6 |µ|(h)·p∗α (f ∗ ) ,
∗

where


(i) p∗α (f ∗ ) = sup qα∗ f ∗ (x) ;
x∈K

the first inequality follows from the cited Prop. 6 and the fact that
qα∗ ◦ (h · f ∗ ) = h · (qα∗ ◦ f ∗ ) as functions on X , whereas the second inequality
follows from (i). From this, we need only retain
Z 
(ii) qα∗ f dµ 6 |µ|(h) · p∗α (f ∗ ) .
∗
§3 integrals of continuous vectorial functions INT III.x94

(3) Given f ∈ K (X, K; E) , we propose to apply the inequality (ii) to

the function f ∗ = ι ◦ f , where ι : E → E
b is the canonical injection. Since
(ι ◦ f )(x) = f (x) ∈ E , we have



p∗α (ι ◦ f ) = sup qα∗ f (x) = sup qα f (x) = pα (f ) ,
x∈K x∈K

thus the right side of (ii) becomes |µ|(h) · p α (f ) . We wish to prove that
Z 
(iii) q∗α f dµ 6 |µ|(h) · pα (f ) ,

which will establish

R the continuity of (the restriction to K (X, K; E) of) the
mapping f 7→ f dµ , since the pα define the topology of K (X, K; E) , and
R
the qα∗ that of Eb ; thus we need only show that the elements (ι ◦ f ) dµ and
R
b are equal, and this follows from the fact that, for every u ∈ E
f dµ of E b0 ,

DZ E Z
(ι ◦ f ) dµ, u = h(ι ◦ f )(x), ui dµ(x)
Z DZ E
= hf (x), ui dµ(x) = f dµ, u .

III.39, `. −13, −12.

“ . . . with the notations of the statement,
Z


g(x) · a dµ(x) = µ(g) · a ”

Since g R· a ∈ K (X; E) (by the continuity of scalar multiplication in E ),

the symbol (g · a) dµ is authorized by No. 1, Def. 1 for the element of E 0 ∗
such that, for all z0 ∈ E0 ,
DZ E Z
(g · a) dµ, z = h(g · a)(x), z0 i dµ(x)
0

Z
= g(x)ha, z0 i dµ(x)
Z
0
= ha, z i g(x) dµ(x)

= hµ(g) · a, z0 i ;
R
b , hence may
thus (g · a) dµ is equal to the canonical image of µ(g) · a in E
be identified with the element µ(g) · a .
INT III.x95 measures §3

III.39, `. −5, −4. R

b ...”
“ . . . the mapping µ 7→ f dµ of M (X; C) into E
Cor. 1 of Prop. 7 of No. 3 permits the definition of a mapping
Z
b , B(f , µ) =
B : K (X; E) × M (X; C) → E f dµ ,

that is clearly bilinear.

The message of Props. 8 and 9: B is separately continuous when
K (X; E) is equipped with the direct limit topology and M (X; C) with
the topology of strictly compact convergence, and, as such a mapping, is
characterized by its values B(g · a, ε x ) = g(x)a for g ∈ K (X; C) , a ∈ E
and x ∈ X .
III.40, `. 4—7.
“ . . . consider the linear mapping v : z 0 7→ hf , z0 i of E0 into K (X; C) ,
and let us show that the image under v of an equicontinuous subset H
of E0 is contained in a strictly compact subset of K (X; C) .”
From this point on, elements z0 ∈ E0 may be regarded as acting also
b
on E via their continuous extensions. It is simpler to replace E by E b:
as observed at the end of the note for III.39,R `. 13–16, if ι : E → E b
R
is the canonical injection, then (ι ◦ f ) dµ = f dµ , thus the two func-
tions f ∈ K (X; E) and ι ◦ f ∈ K (X; E) b produce the identical mapping
M (X; C) → E b.
To simplify the notation,
R we can therefore assume henceforth that E
is complete, so that f dµ ∈ E for all µ ∈ M (X; C) . We also make the
abbreviations K = K (X; C) and M = M (X; C) , while retaining the full
notations K (X; E) and K (X, K; E) , K (X, K; C) for K compact.
Since f ∈ K (X; E) , one has v(z0 ) = hf , z0 i = z0 ◦ f ∈ K for all
0
z ∈ E ; and if K is a compact set such that f ∈ K (X, K; E) , for instance
K = Supp f , then v(z0 ) ∈ K (X, K; C) for all z0 ∈ E0 . The mapping
v : E0 → K is linear by a straightforward computation. We are to show
that the set of functions

v(H) = {z0 ◦ f : z0 ∈ H } ⊂ K (X, K; C) ⊂ K

is contained in a strictly compact subset of K , in other words, that the

closure
v(H)
of v(H) in K (for the direct limit topology on K ) is strictly compact.
Now, v(H) ⊂ K (X, K; C) and we know that K (X, K; C) is a closed subset
§3 integrals of continuous vectorial functions INT III.x96

of K (§1, No. 1, Prop. 1), therefore v(H) ⊂ K (X, K; C) ; it follows (GT,

I, §3, No. 1, Prop. 1) that

v(H) = v(H) ∩ K (X, K; C)

is the closure of v(H) in K (X, K; C) for the topology induced by that

of K , and since that topology is the norm topology (§1, No. 1, Prop. 1),
v(H) is equal to the closure of v(H) for the norm topology. Thus it will
suffice to show that v(H) is compact for the norm topology.
Now, the norm topology on K (X, K; C) coincides with the topol-
ogy of uniform convergence in K , with the topology of uniform conver-
gence in X , and with the topology of compact convergence (the topology
of uniform convergence in the compact subsets of X ). One argues simi-
larly that K (X, K; C) is a closed subset of C c (X; C) (the space C (X; C)
equipped with the topology τcc of compact convergence), that τcc induces
on K (X, K; C) the norm topology, and that v(H) is also the closure of v(H)
τcc
in Cc (X; C) for τcc , that is, v(H) = v(H) .
Thus, the problem is to show that v(H) is a relatively compact subset
of Cc (X; C) ; by Ascoli’s theorem (GT, X, §2, No. 5, Cor. 3 of Th. 2), it
suffices to show that v(H) is (i) equicontinuous, and (ii) pointwise bounded:
(i) Fix x0 ∈ X . Given any ε > 0 , by equicontinuity of H there exists
a neighborhood V of 0 in E such that

z ∈ V ⇒ |z0 (z)| 6 ε for all z0 ∈ H ,

and since f is continuous at x0 there exists a neighborhood U of x0 in X

such that
x ∈ U ⇒ f (x) ∈ f (x0 ) + V ,
 0

whence x ∈ U ⇒ z f (x) − f (x0 )  6 ε for all z0 ∈ H , that is,
 
z0 ∈ H ⇒ (z0 ◦ f )(x) − (z0 ◦ f )(x0  6 ε for all x ∈ U ,

whence the equicontinuity of the functions z 0 ◦ f ∈ v(H) at x0 .

(ii) Fix x0 ∈ X . We are to show that the set



v(H) (x0 ) = z0 f (x0 ) : z0 ∈ H

is bounded in C . Indeed, since H ⊂ E0 = L (E; C) is equicontinuous on E ,

H is relatively compact for the weak topology σ(E 0 , E) on E0 (TVS, III, §3,
No. 4, Cor. 2 of Prop. 4). Since the mapping ϕ : E 0 → C defined by


ϕ(z0 ) = z0 f (x0 ) = hf (x0 ), z0 i (z0 ∈ E0 )
INT III.x97 measures §3

is continuous

for σ(E0 , E) , the relative compactness (i.e., boundedness) of
(v(H) (x0 ) = ϕ(H) follows from ϕ(H) ⊂ ϕ(H) and the compactness of H .
To summarize, we have shown that the closure of v(H) in K is con-
tained in K (X, K; C) and is compact, hence is a strictly compact subset
of K .
III.40, `. 11, 12.
“ . . . u is none other than the restriction to M (X; C) of the trans-
pose t v (in the algebraic sense)
The conventions of the preceding note are in force; in particular, E is
complete, and we abbreviate K = K (X; C) , M = M (X; C) , writing K *
and M * for their algebraic duals. The linear mappings

u : M → E , v : E0 → K
R
are defined by u(µ) = f dµ and v(z0 ) = z0 ◦ f ( µ ∈ M , z0 ∈ E0 ).
Equip E0 with the topology σ(E0 , E0 *) , and K with the topology
σ(K , K *) ; then v is continuous. For, if f ∈ K * then f ◦ v ∈ E 0 * is
continuous for σ(E0 , E0 *)
v-K f-
E0 C
and since σ(K , K *) is the initial topology for the linear forms f ∈ K *
(TVS, II, §6, No. 2, Def. 2), it follows that v is continuous (GT, I, §2, No. 3,
Prop. 4). The transposed linear mapping
t
v : K * → E0 *

is thus given by t v : f 7→ f ◦v (TVS, II, §6, No. 4, Prop. 5). {Moreover, t v is

continuous for the topologies σ(K *, K ) and σ(E 0 *, E0 ) , and t (t v) = v (loc.
cit., Cor. of Prop. 5), but this is not needed here.}
Finally, M = K 0 ⊂ K * , and if µ ∈ M then, for all z0 ∈ E0 ,

ht v(µ), z0 i = hµ, v(z0 )i = hµ, z0 ◦ f i

DZ E
= f dµ, z0 = hu(µ), z0 i ,

whence t v(µ) = u(µ) for all µ ∈ M . That is, t v M = u .
III.40, `. 12, 13.
“ . . . its continuity therefore follows from the foregoing (TVS, IV, §1,
No. 3, Prop. 6).”
The reference in the 1965 French original is inexplicit (Esp. vect. top.,
chap. IV, 2e éd.), to a work not yet published—the bound edition of EVT did
§3 integrals of continuous vectorial functions INT III.x98

not appear until 1981. Unable to find a suitable substitute reference in the
first edition at the time (1970’s) that I first studied Intégration, I improvised
a “direct proof” based on ingredients available in the first edition of Ch. III
of Esp. vect. top. When I was preparing (1998) the published translation,
Professor Jacques Dixmier guided me to the correct reference in TVS and
sketched the indicated proof of the continuity of u ; an exposition of his
proof (Proof #2) is given immediately after the following “direct proof”.
Proof #1. The conventions of the preceding two notes are in force.
−1
Assuming V is a neighborhood of 0 in E , let us show that u (V) is a
neighborhood of 0 in M for the topology τ scc of strictly compact conver-
gence on M . We can suppose that V is convex, balanced and closed in E .
For the canonical duality hz, z0 i = z0 (z) between E and E0 , the polar of V
in E0 is the set

V◦ = {z0 ∈ E0 : |z0 (z)| 6 1 for all z ∈ V }

(TVS, II, §8, No. 4); moreover, V ◦ is convex, balanced, and closed for the
weak topology σ(E0 , E) , and, by the theorem on bipolars (loc. cit., §6, No. 3,
Cor. 3 of Th. 1), V = V◦◦ (the polar of V◦ in E ).
Let H = V◦ ⊂ E0 ; since H◦ = V is a neighborhood of 0 in E , H is
equicontinuous on E (TVS, III, §3, No. 5, Prop. 7), therefore the closure
v(H) of v(H) in K is strictly compact (see (ii) of the Note for III.40,
`. 4–7).
Thus v(H) ∈ S , the set of all strictly compact subsets of K . The
topology τscc on M is the topology of uniform convergence in the sets
S ∈ S . It follows (GT, X, §1, No. 2) that if S ∈ S and D = {c ∈ C :
|c| 6 1 } is the closed unit disc in C , so that the set

D = {(c, d) ∈ C × C : c − d ∈ D }

is an entourage for the uniformity of C , then the set

{(µ, ν) ∈ M × M : µ(g), ν(g) ∈ D for all g ∈ S }
= {(µ, ν) : |µ(g) − ν(g)| 6 1 for all g ∈ S }
is an entourage for the uniformity on M of strictly compact convergence,
and the set

(*) {µ ∈ M : |µ(g)| 6 1 for all g ∈ S }

is a neighborhood of 0 in M for τscc . In the canonical duality between K

and K 0 = M , the set (*) is the polar S ◦ of S in M . In particular, the
set
v(H) ◦ = {µ ∈ M : |µ(g)| 6 1 for all g ∈ v(H) }
INT III.x99 measures §3

is a neighborhood of 0 in M for τscc , and, since the µ ∈ M are continuous

on K ,

v(H)◦ = {µ ∈ M : |µ(g)| 6 1 for all g ∈ v(H) } = v(H) ◦ .

Thus v(H)◦ is a neighborhood of 0 in M for τscc .

To conclude the proof, we need only show that
−1
(**) u (V) = v(H)◦ .

To that end, consider the linear mapping

v : E0 → K

in the context of the algebraic duals of E 0 and K : as observed in the pre-

ceding note, v is continuous for the topologies σ(E 0 , E0 *) and σ(K , K *) .
Moreover, if t v : K * → E0 * is its transpose, then (TVS, II, §6, No. 4,
Prop. 6),

(†) v(H)◦ = (t v)−1 (H◦ ) ,

where the polars are taken in K * and E 0 * , respectively; note that M =

K 0 ⊂ K * and the v(H)◦ in (**) is the intersection with M of the v(H) ◦
in (†). Finally,
−1
u (v) = {µ ∈ M : u(µ) ∈ V = H◦ } (polar in E )
t ◦

t 


= {µ ∈ M : ( v)(µ) ∈ H } v M =u
= {µ ∈ M : |h(t v)(µ), z0 i| 6 1 for all z0 ∈ H }
= M ∩ {f ∈ K * : |h(t v)(f ), z0 i| 6 1 for all z0 ∈ H }
= M ∩ {f ∈ K * : (t v)(f ) ∈ H◦ } (polar in E0 * )
= M ∩ (t v)−1 (H◦ )
= M ∩ v(H)◦ (polar in K * , by (†) )
= {µ ∈ M : |hµ, gi| 6 1 for all g ∈ v(H) }
= v(H)◦ (polar in M ) ,

that is, (**) holds.

Proof #2. {For consistency of this intricate proof with the notation of
the cited reference in TVS, IV, it is convenient to interchange the original
notations u and v in the proof ; otherwise, the conventions of the preceding
two notes are in force (in particular, E is complete).}
§3 integrals of continuous vectorial functions INT III.x100

Regard the pair of vector spaces

R (K , M ) as put in duality by the
bilinear form (g, µ) 7→ hg, µi = g dµ , and similarly for the pair (E, E 0 ) ,
put in duality by the bilinear form (z, z 0 ) 7→ hz, z0 i = z0 (z) . We have linear
mappings u : E0 → K and v : M → E , defined by the formulas (note the
reversed roles of u and v )
Z
0 0
u(z ) = hf , z i and v(µ) = f dµ ;

let us place them in the context of the foregoing dualities.

Let E1 be the vector space E0 equipped with the weak topology σ(E 0 , E) .
The dual of E1 may be identified with the vector space E , that is, E 01 = E
as vector spaces, and if E in turn is equipped with the weakened topology
σ(E, E0 ) , then the dual of E may be identified with the vector space E 0 ;
thus, with the first two primes interpreted in the sense of the dual pairing,
(E01 )0 = E0 = E1 as vector spaces.
Similarly, let E2 be the vector space K equipped with the weakened
topology σ(K , M ) . Then the dual E02 of E2 may be identified with the
vector space M , and if M is in turn equipped with the vague topology
σ(M , K ) , then the dual of M may be identified with the vector space K ;
interpreting the primes appropriately, (E 02 )0 = K = E2 as vector spaces.
With these notations, we have

(*) u : E1 → E2 , v : E02 → E01 .

For every z0 ∈ E0 = E1 and µ ∈ M = E02 we have (citing formula (1) of

No. 1 for the second equality)
Z DZ E
0 0
hu(z ), µi = hf , z i dµ = f dµ, z0 = hv(µ), z0 i ;

it follows (TVS, II, §6, No. 4, Prop. 5) that, in the notations of (*), u and v
are continuous for the topologies σ in question, and each is the transpose of
the other (in the sense of ibid., Remark ), concisely t u = v (and t v = u ).
Our problem is to show that the linear mapping v : M → E remains
continuous when M is equipped with the topology of strictly compact con-
vergence (in general stronger than σ(E 02 , E2 ) = σ(M , K ) , the vague topol-
ogy on M ), and E with its original topology (in general stronger than
σ(E01 , E1 ) = σ(E, E0 ) , the weakened topology on E ); this is fertile terrain
for the S-topologies associated with dualities (TVS, III, §3, No. 5). Indeed,
if S2 is the set of all strictly compact subsets of E 2 = K , then the topol-
ogy of (E02 )S2 is the topology of strictly compact convergence in E 02 = M ;
and if S1 is the set of all subsets of E1 = E0 that are equicontinuous with
INT III.x101 measures §4

respect to the original locally convex topology τ on E , then the topology

of (E01 )S1 coincides with the topology τ on E01 = E (loc.cit., Cor. 1 of
Prop. 7). Thus our problem is to show that the linear mapping

v : (E02 )S2 → (E01 )S1

is continuous for the indicated S-topologies. Now, v = t u ; in view of TVS,

IV, §1, No. 3, Prop. 6, we need only show that:
1◦ u : E1 → E2 is continuous for the weakened topologies of the E i
(i = 1, 2) ;
2◦ the sets in Si are bounded in Ei (i = 1, 2); and
3◦ if A ∈ S1 then there exist sets A1 , . . . , An in S2 and a real
number λ > 0 such that λ · u(A) is contained in the closed balanced
convex envelope of A1 ∪ . . . ∪ An .
As to 1◦ : Since Ei bears the topology σ(Ei , E0i ) , that topology co-
incides with its weakened topology, so the assertion is immediate from the
continuity property of u noted in connection with (*).
As to 2◦ : If A ∈ S1 , that is, if A ⊂ E0 is equicontinuous for the
original topology τ of E , then A is relatively compact for σ(E 0 , E) (TVS,
III, §3, No. 4, Cor. 2 of Prop. 4), therefore precompact and hence bounded
(TVS, III, §1, No. 2, Prop. 2). If A ∈ S 2 , so that A ⊂ K is (strictly)
compact for the inductive limit topology, then it is also compact (hence
bounded) for the weaker topology σ(K , M ) .
As to 3◦ : If A ∈ S1 then u(A) is contained in a strictly compact
subset A1 of K (see the Note for III.40, `. 4—7), thus condition 3 ◦ is
verified with n = 1 and λ = 1 (and no need to take envelopes).

§4. PRODUCTS OF MEASURES

III.41, `. 6, 7.
“It is immediate that the image under ω of K (X × Y, K × L; C) is
contained in K (X, K; K (Y, L; C)) .”
If f : X × Y →
C then ω(f ) : X → F (Y; C) is defined as follows: for
each x ∈ X , ω(f ) (x) : Y → C is defined by


ω(f ) (x) (y) = f (x, y) for all y ∈ Y .

If f is continuous then,

for each x ∈ X , the mapping y 7→ f (x, y) is con-
tinuous, hence ω(f ) (x) is continuous on Y , and so ω(f ) : X → C (Y; C) .
§4 products of measures INT III.x102

Suppose now that f ∈ K (X × Y, K × L; C) . For each x ∈ X ,



ω(f ) (x) (y) = f (x, y) = 0 for all y ∈ Y --- L ,

therefore ω(f ) (x) ∈ K (Y, L; C) for all x ∈ X . Moreover, if x ∈ X --- K ,
then 

ω(f ) (x) (y) = f (x, y) = 0 for all y ∈ Y ,


therefore ω(f ) (x) = 0 , the zero element of K (Y, L; C) ; thus


ω(f ) ∈ F X, K; K (Y, L; C) .


To show that ω(f ) ∈ K X, K; K (Y, L; C) , we need only show that ω(f )
is continuous for the given topology on X and the norm topology on
K (Y, L; C) . Since the norm topology on K (Y, L; C) coincides with the
topology of uniform convergence in the compact subsets of Y , K (Y, L; C)
is a topological subspace of Cc (Y; C) ; the desired continuity of ω(f ) is
therefore a consequence of GT, X, §3, No. 4, Th. 3 (with Z = C and
fe = ω(f ) ).
III.41, `. 7–10.
“ . . . if u is a continuous mapping of X into K (Y, L; C) , with support
contained in K , then the mapping (x, y) 7→ u(x)(y) of X × Y into C is
continuous and has support contained in K × L ”


Define f : X × Y → C by f (x, y) = u(x) (y) . For every x ∈ X ,



ω(f ) (x) (y) = f (x, y) = u(x) (y) for all y ∈ Y ,

hence ω(f ) (x) = u(x) for all x ∈ X ; thus ω(f ) = u . The continuity of
ω(f ) : X → K (Y, L; C) ⊂ Cc (Y; C) then implies that of f , by GT, X, §3,
No. 4, Th. 3 (the “conversely” part, with Z = C and fe = ω(f ) ), and it is
clear that f (x, y) 6= 0 implies (x, y) ∈ K × L .
III.41, `. 14–18.
“ . . . the image under ω , of

K (X, K; C) ⊗ C K (Y, L; C)

identified with a subspace of K (X × Y, K × L; C) , is again the space

K (X, K; C)⊗C K (Y, L; C) but this time identified canonically with a space
of mappings of X into K (Y, L; C) (A, II, §7, No. 7, Cor. of Prop. 15)”
INT III.x103 measures §4

P
n
An element f = gi ⊗hi of the displayed tensor product space, viewed
i=1
as a function on X × Y (the first identification referred to), is the function
n
X
f (x, y) = gi (x)hi (y) .
i=1

Let us calculate its image under ω : for all x ∈ X, y ∈ Y ,


 n
X
ω(f ) (x) (y) = f (x, y) = gi (x)hi (y) ,
i=1

thus n

X
ω(f ) (x) = gi (x) · hi ∈ K (Y, L; C) for all x ∈ X .
i=1

Writing E = K (Y, L; C) , the preceding formula shows that the image of

K (X, K; C) ⊗ C K (Y, L; C) = K (X, K; C) ⊗ C E

under ω coincides with its interpretation in § 1, No. 2, Prop. 5 as a dense

linear subspace of K (X, K; E) (the second identification referred to).
III.41, `. −13, −12.
“ . . . the conclusion of (ii) follows from the fact that the restriction of ω
is a topological isomorphism.”
Let ω0 be the indicated restriction; according to (i), the mapping


ω0 : K (X × Y, K × L; C) → K X, K; K (Y, L; C)

is an (isometric) isomorphism of topological vector spaces. Since ω 0 maps

the subset K (X, K; C)
⊗C K (Y, L; C) of its domain onto a dense subset of
K X, K; K (Y, L; C) , the isomorphism implies that

K (X, K; C) ⊗ C K (Y, L; C) = K (X × Y, K × L; C) .

III.41, `. −9, −8.

“ . . . the subspace K (X; C) ⊗ C K (Y; C) is dense in K (X × Y; C) . ”
Let f ∈ K (X × Y; C) . Since Supp(f ) is compact in X × Y , one
has Supp(f ) ⊂ K × L for suitable compact sets K ⊂ X , L ⊂ Y . Then
f ∈ K (X × Y, K × L; C) , so by Lemma 1, (ii), f belongs to the closure of
K (X, K; C)⊗C K (Y, L; C) in K (X×Y, K×L; C) ; but K (X×Y, K×L; C)
is a topological subspace of K (X × Y; C) (§1, No. 1, Prop. 1, (i)), therefore
§4 products of measures INT III.x104

f belongs to the closure of K (X, K; C) ⊗ C K (Y, L; C) in K (X × Y; C)

(GT, I, §3, No. 1, Prop. 1).
The rest of this note is for use in No. 4.
Another way of putting the matter is that the set of functions (let us
drop the “; C” for brevity)


f ⊗g f ∈ K (X), g ∈ K (Y)

is total in K (X × Y) (TVS, I, §2, No. 1, Def. 1); by the Hahn-Banach

theorem, a test for totality is that if ν is a continuous linear form on
K (X × Y) (that is, a measure on X × Y ) such that

(*) ν(f ⊗ g) = 0 for all f ∈ K (X), g ∈ K (Y) ,

then ν = 0 (TVS, II, §8, No. 3, Cor. of Prop. 1).

We wish to sharpen this result to the following: If ν is a measure on
X × Y such that ν(f ⊗ g) = 0 as f runs over a total subset of K (X) and
g runs over a total subset of K (Y) , then ν = 0 .
Lemma. — The bilinear function K (X) × K (Y) → K (X × Y) defined
by (f, g) 7→ f ⊗ g is separately continuous (for the direct limit topologies).
Proof. Fix g ∈ K (Y) , say g ∈ K (Y, L) , where L is a compact
subset of Y . Let us write u : K (X) → K (X × Y) for the linear mapping
u(f ) = f ⊗ g ; we are to show that u is continuous. Given any compact
subset K of X , it suffices to show that the composite mapping

K (X, K) → K (X) → K (X × Y)

defined by f 7→ f 7→ f ⊗ g (f ∈ K (X, K) ) is continuous (TVS,  II, §4,

No. 4, Prop. 5, (ii), Example II, and Remark ). Writing u K = uK (X, K) ,
the problem is to show that the mapping

uK : K (X, K) → K (X × Y)

is continuous. For all f ∈ K (X, K) one has f ⊗ g ∈ K (X × Y, K × L) ,

whence a factorization

K (X, K) → K (X × Y, K × L) → K (X × Y)

of uK . The second arrow being continuous by the definition of the direct

limit topology, we are reduced to proving the continuity of the mapping

K (X, K) → K (X × Y, K × L)
INT III.x105 measures §4

defined by f 7→ f ⊗ g , and this is immediate from kf ⊗ gk = kf k kgk .

Similarly for the mappings g 7→ f ⊗ g .
Proposition. — If A (resp. B ) is a total subset of K (X) (resp. K (Y) ),
then the set of functions

f ⊗g (f ∈ A, g ∈ B)

is total in K (X × Y) .
Proof. Let ν be a measure on X × Y such that ν(f ⊗ g) = 0 for all
f ∈ A and g ∈ B ; we are to show that ν = 0 . Fix g ∈ B . It follows
from the Lemma that the function f 7→ ν(f ⊗ g) is a continuous linear form
on K (X) ; since it vanishes on the total set A , it is identically zero. Thus
ν(f ⊗ g) = 0 for all f ∈ K (X) and g ∈ B . Similarly, fixing f ∈ K (X) ,
the validity of ν(f ⊗ g) = 0 for all g ∈ B implies that ν(f ⊗ g) = 0 for all
g ∈ K (Y) . Thus ν = 0 .
Corollary. — With notations as in the Proposition, if ν is a measure
on X × Y such that ν(f ⊗ g) = 0 for all f ∈ A and g ∈ B , then ν = 0 .
III.42, `. 3–5.
“ . . . for u = ω(f ) and for every y ∈ Y ,
Z  Z Z
u(x) dλ(x), εy = u(x)(y) dλ(x) = f (x, y) dλ(x) ,

whence the lemma.”

Write E = K (Y, L; C) ; thus u ∈ K (X; E) , where u(x) (y) = f (x, y)
for all x ∈ X , y ∈ Y . For y ∈ Y , the Dirac measure ε y on Y is a continuous
linear form on K (Y; C) ; for clarity, let us write η y for the restriction of εy
to K (Y, L; C) . Since K (Y, L; C) is a topological subspace of K (Y; C) ,
ηy is continuous on K (Y, L; C) = E , thus η y ∈ E0 . By the definition of
the vectorial integral,
DZ E Z
u dλ, ηy = (ηy ◦ u) dλ for all y ∈ Y ;

but (ηy ◦u)(x) = ηy u(x) = εy u(x) = u(x) (y) = f (x, y) , thus ηy ◦u =
f (·, y) and so
DZ E Z
u dλ, ηy = f (·, y) dλ
Z
= f (x, y) dλ(x) = h(y) ,
§4 products of measures INT III.x106

briefly
DZ E
(*) h(y) = u dλ, ηy for all y ∈ Y .

If y ∈/ L then f (·, y) is the zero element of K (X, K; C) and so h(y) = 0 ,

thus h ∈ FR (Y, L; C) ; it remains only to show that h is continuous on Y .
Now, u dλ ∈ E = K (Y, L; C) ⊂ K (Y; C) , and y 7→ εy is a con-
tinuous mapping Y → M (Y; C) for the vague topology on M (Y; C) (§1,
No. R9, Prop.

13), thus
R if
y → y0 in Y then εy → εy0 vaguely, in particular
εy u dλ = εy0 u dλ . But, for all y ∈ Y ,
Z  Z  DZ E
εy u dλ = ηy u dλ = u dλ, ηy = h(y)

by (*), thus h(y) → h(y0 ) .

Alternate proof. Clearly (2) defines a function h : Y → C such that
h = 0 on { L , so we need only show that h is continuous. Indeed, as
noted in the proof of Lemma 1 (with the roles of X and Y reversed), the
mapping Y → K (X, K; C) defined by y 7→ f (·, y) is continuous (GT, X,
§3, No. 4, Th. 3), K (X, K; C) is a topological subspace of K (X; C) , and
λ is a continuous linear form on K (X; C) , so the composite mapping


y 7→ f (·, y) 7→ λ f (·, y) = h(y)
is continuous.
III.42, `. 13.
“ . . . whence |ν(f )| 6 aK bL kf k .”
In the notation of Lemma 2, |h(y)| 6 a K kf k for all y ∈ Y , therefore
|µ(h)| 6 bL khk 6 bL aK kf k .
III.42, `. 13, 14.
“The linear form ν on K (X × Y; C) is thus a measure on X × Y ”
If A is any compact subset of X × Y , and K ⊂ X , L ⊂ Y are com-
pact sets such that A ⊂ K × L , there exists a constant M A = aK bL such
that |ν(f )| 6 MA kf k for every f ∈ K (X × Y, K × L; C) , hence for every
f ∈ K (X × Y, A; C) , whence the assertion (§1, No. 3, criterion following
Def. 2).
III.42, `. −2, −1.
Z Z Z
f (x, y) dν(x, y) = dλ(x) f (x, y) dµ(y)
“(3) Z Z
= dµ(y) f (x, y) dλ(x) . ”
INT III.x107 measures §4

By construction, ν is the unique measure on X × Y such that

Z
(i) g(x)h(y) dν(x, y) = λ(g)µ(h) for all g ∈ K (X; C), h ∈ K (Y; C) ,

and it satisfies
Z Z 
(ii) ν(f ) = f (x, y) dλ(x) dµ(y) for all f ∈ K (X × Y; C) .

Interchanging the roles of X and Y , there exists a unique measure ρ on

Y × X such that
Z
(iii) h(y)g(x) dρ(y, x) = µ(h)λ(g) for all h ∈ K (Y; C), g ∈ K (X; C) ,

and it satisfies
Z Z 
(iv) ρ(k) = k(y, x) dµ(y) dλ(x) for all k ∈ K (Y × X; C) .

For each f ∈ K (X × Y; C) let f 0 : Y × X → C be the function defined by

f 0 (y, x) = f (x, y) ; since (x, y) 7→ (y, x) is a homeomorphism of X × Y onto
Y × X , f 7→ f 0 defines a vector space isomorphism
K (X × Y; C) → K (Y × X; C)
that is a homeomorphism for the respective direct limit topologies. Define
ν 0 : K (X × Y; C) → C
by the formula ν 0 (f ) = ρ(f 0 ) . Then ν 0 is a continuous linear form, that is,
a measure on X × Y . In fact, ν 0 = ν by the uniqueness part of Th. 1, since,
for all g ∈ K (X; C) and h ∈ K (Y; C) , writing f (x, y) = g(x)h(y) one has
Z
g(x)h(y) dν 0 (x, y) = ν 0 (f )

= ρ(f 0 )
Z
= f 0 (y, x)dρ(y, x)
Z
= f (x, y)dρ(y, x)
Z
= h(y)g(x)dρ(y, x)

= µ(h)λ(g) by (iii)
Z
= g(x)h(y) dν(x, y) by (i) .
§4 products of measures INT III.x108

Then, for every f ∈ K (X × Y; C) , one has ν(f ) = ν 0 (f ) = ρ(f 0 ) , that is,

Z  
f (x, y) dλ(x) dµ(y) = ν(f ) by (ii)

= ρ(f 0 )
Z Z 
= f 0 (y, x) dµ(y) dλ(x) by (iv)
Z Z 
= f (x, y) dµ(y) dλ(x)

as claimed.
III.44, `. 1–4.
“Indeed, for every z0 ∈ E0 we have
Z Z  ZZ Z Z
0
f dλ dµ, z = hf , z i dλ dµ = dµ hf , z0 i dλ
0

Z Z  Z Z 
0 0
= f dλ, z dµ = dµ f dλ, z

by (4), whence (5).”

In slow motion: for all z0 ∈ E0 , one has
D ZZ E DZ E ZZ
f dλ dµ, z0 = f dν, z0 (definition of )
Z
= hf , z0 i dν (§3, No. 1, Def. 1)
Z Z 
= hf (x, y), z0 i dλ(x) dµ(y) (by (4))
Z Z 
= hf (·, y), z0 i dλ dµ(y) (cosmetics)
Z DZ E
= f (·, y) dλ, z0 dµ(y) (§3, No. 1, Def. 1)
Z
D  Z  E
= f (·, y) dλ dµ(y), z0 (§3, No. 1, Def. 1)

whence Z Z Z 
f dν = f (·, y) dλ dµ(y)
Z Z 
= f (x, y) dλ(x) dµ(y)
Z Z
= dµ(y) f (x, y) dλ(x) ;
INT III.x109 measures §4
R 0
Rthe argument shows that h dµ , a priori an element of E * , is equal to
f dν hence belongs to E . Similarly,
DZ E Z Z 
f dν, z0 = hf (x, y), z0 i dµ(y) dλ(x)
DZ Z  E
= f (x, ·) dµ dλ(x), z0 ,

whence Z Z Z 
f dν = f (x, ·) dµ dλ(x)
Z Z
= dλ(x) f (x, y) dµ(y) .

III.44, `. −5.
“ . . . which proves formula (6).”
For all f ∈ K (X × Y; C) ,
Z


hf, (g ⊗ h) · (λ ⊗ µ)i = f · (g ⊗ h) d(λ ⊗ µ)
Z


= f (x, y)g(x)h(y) d(λ ⊗ µ)(x, y)
Z Z 
= f (x, y)g(x)h(y) dµ(y) dλ(x) by (3)
Z Z 
= f (x, y)h(y) dµ(y) g(x) dλ(x)
Z Z 
= f (x, ·) d(h · µ) d(g · λ)(x)
Z


= f d (g · λ) ⊗ (h · µ) by (3)

= hf, (g · λ) ⊗ (h · µ)i ,
whence (g ⊗ h) · (λ ⊗ µ) = (g · λ) ⊗ (h · µ) .
III.45, `. 1–3.
“ . . . if U (resp. V ) is an open set in X (resp. Y ), then the restriction
of λ ⊗ µ to the product U × V is the product of the restrictions of λ to U
and of µ to V ”
We know from the first paragraph of §2, No. 1, that  K (X, U; C) may
be identified with K (U; C) via the mapping f 7→ f U , and the measure
λU on U is the result of transporting the restriction λ K (X, U; C) to a
linear form on K (U; C) via this identification; that is,
 
(λU)(f U) = λ(f ) for all f ∈ K (X, U; C) .
§4 products of measures INT III.x110

From another perspective, if, for g ∈ K (U; C) , g 0 denotes the extension

by 0 of g to X , then

(λU)(g) = λ(g 0 ) for all g ∈ K (U; C)

(see
 the note for III.23, `. 16–17). Similarly
 for the relation between µ and
µV , and between
 λ ⊗
 µ and (λ ⊗ µ)

U × V .
Both (λU) ⊗ (µV) and (λ ⊗ µ)U × V are measures on U × V ; if g ∈
K (U; C) and h ∈ K (V; C) , then g ⊗ h ∈ K (U × V; C) , g 0 ⊗ h0 = (g ⊗ h)0
on X × Y , and
   
[(λU) ⊗ (µV)](g ⊗ h) = (λU)(g) · (µV)(h)
= λ(g 0 )µ(h0 )
= (λ ⊗ µ)(g 0 ⊗ h0 )



= λ ⊗ µ (g ⊗ h)0

= [(λ ⊗ µ)U × V)](g ⊗ h) ,
  
therefore (λU) ⊗ (µV) = (λ ⊗ µ)U × V by the uniqueness part of No. 1,
Th. 1.
III.45, `. 6–8.
“ . . . which proves the proposition, on taking into account the definition
of the support of a measure (§2, No. 2).”

Let U = X --- Supp λ . Then λU = 0 and, writing ν = λ ⊗ µ , we have
  
ν U × Y = (λU) ⊗ (µY) = 0 ⊗ µ = 0 ,

therefore U × Y ⊂ { Supp ν , thus

Supp ν ⊂ { (U × Y) = ( { U) × Y = (Supp λ) × Y .

Similarly Supp ν ⊂ X × Supp µ , thus

Supp ν ⊂ [(Supp λ) × Y] ∩ [X × (Supp µ)] = (Supp λ) × (Supp µ) .

To prove the reverse inclusion, suppose (x, y) ∈ { Supp ν ; it will suffice

to show that either x ∈ { Supp λ or y ∈ { Supp µ . Let U ⊂ X and V ⊂ Y
be open sets such that

(x, y) ∈ U × V ⊂ { Supp ν .
   
Then ν U × V = 0 , thus (λU) ⊗ (µV) = (λ ⊗ µ)U × V = 0 , therefore
 
λU = 0 or µV = 0 , whence U ⊂ { Supp λ or V ⊂ { Supp µ , and so
x ∈ { Supp λ or y ∈ { Supp µ .
INT III.x111 measures §4

III.47, `. 3.
“There exist . . . ”
By §1, No. 8, Prop. 10.
III.47, `. 13–17.
“Proposition 6. — When M (X; C) , M (Y; C) and M (X × Y; C)
are equipped with the topology of strictly compact convergence (§1, No. 10),
the bilinear mapping (λ, µ) 7→ λ ⊗ µ of M (X; C) × M (Y; C) into
M (X × Y; C) is hypocontinuous for the set of vaguely bounded subsets of
M (X; C) and M (Y; C) (TVS, III, §5, No. 3).”
The assertion is that if S and T are the sets of vaguely bounded
subsets of M (X; C) and M (Y; C) , respectively, then the bilinear mapping
u : (λ, µ) 7→ λ ⊗ µ is (S, T)-hypocontinuous (in the sense of TVS, III, §5,
No. 3), that is, (i) it is separately continuous (for the topologies of strictly
compact convergence), (ii) for every M ∈ S and every neighborhood W
of 0 in M (X × Y; C) , there exists a neighborhood V of 0 in M (Y; C)
such that u(M × V) ⊂ W , and (ii0 ) for every N ∈ T and every neighbor-
hood W of 0 in M (X × Y; C) , there exists a neighborhood U of 0 in
M (X; C) such that u(U × N) ⊂ W .
(The conditions (i) and (ii) express that ϕ is S-hypocontinuous, whereas
(i) and (ii0 ) express that it is T-hypocontinuous.)
Remarks. — Suppose E, F, G are locally convex spaces, ϕ : E × F → G
is bilinear, and S (resp. T ) is a set of bounded subsets of E (resp. F ).
If ϕ is continuous (“jointly”) then it is (S, T)-hypocontinuous for all such
S and T (TVS, III, §5, No. 3, discussion following Def. 2). Thus (S, T)-
hypocontinuity is in general stronger than separate continuity and weaker
than continuity. According to Exercise 3 for §4, the bilinear mapping
u : (λ, µ) 7→ λ ⊗ µ is continuous (not just separately) for the topologies
of strictly compact convergence. It then follows from the foregoing that
u is (S, T)-hypocontinuous when S (resp. T ) is any set of sets that
are bounded for the topology τscc of strictly compact convergence; but
the sets bounded for τscc are the same as the vaguely bounded sets (§1,
No. 10, Prop. 17), thus one recovers Prop. 6, enhanced with joint continuity
(for τscc ).
III.47, `. −14 to −11.
“the mapping ϕ of K (X, K; K (Y, L; C)) × M (X; C) Rinto K (Y, L; C),
such that ϕ(g, λ) is the function h defined by h(y) = g(x, y) dλ(x) , is
separately continuous by virtue of §3, No. 4, Props. 8 and 9.”
There is here an “abuse of notation”: a function

g ∈ K (X × Y, K × L; C)
§4 products of measures INT III.x112

is being identified with the function

ω(g) ∈ K X, K; K (Y, L; C)


of No. 1,
Lemma 1; explicitly, ω(g) (x) is the function y 7→ g(x, y) , that
is, ω(g) (x) = g(x, ·) . For clarity, let us maintain the distinction.
Write E = K (Y, L; C) , a Banach space; thus ω(g) ∈ K (X, K; E) for
g ∈ K (X × Y, K × L; C) . Define a (bilinear) mapping

ϕ : K (X × Y, K × L; C) × M (X; C) → E
R R
by ϕ(g, λ) = ω(g) dλ = g(x, ·) dλ(x) ∈ E b = E (§3, No. 3, Cor. 2 of
Prop. 7).
For fixed g ∈ K (X × Y, K × L; C) , hence fixed ω(g) ∈ K (X, K; E) ,
the mapping
Z


λ 7→ ϕ(g, λ) = ω(g) dλ λ ∈ M (X; C)

is continuous for the topology of strictly compact convergence (§3, No. 4,

Prop. 9).
For fixed λ ∈ M (X; C) , the mapping
Z


ω(g) 7→ ω(g) dλ ω(g) ∈ K (X, K; E)

is continuous for the norm topology (§3, No. 4, Prop. 8), and since g 7→ ω(g)
is an isometry K (X × Y, K × L; C) → K (X, K; E) (No. 1, Lemma 1), the
mapping
Z


g 7→ ω(g) dλ = ϕ(g, λ) g ∈ K (X × Y, K × L; C)

is also continuous for the norm topology. Thus ϕ is separately continuous.

III.47, `. −4, −3.
“ . . . the conditions λ ∈ B and µ ∈ C◦ therefore imply λ ⊗ µ ∈ A◦ .”
Polars are being taken; what are the dualities involved?
As to C◦ : We have C ⊂ K (Y, L; C) ⊂ K (Y; C) ; as C is a compact
subset of K (Y, L; C) , it is a strictly compact subset of K (Y; C) (for the
direct limit topology). The duality in question is the canonical duality
Z
(h, µ) 7→ hh, µi = µ(h) = h dµ
INT III.x113 measures §4

between K (Y; C) and K (Y; C)0 = M (Y; C) , thus

(i) C◦ = {µ ∈ M (Y; C) : Rhh, µi > −1 for all h ∈ C }

(TVS, II, §8, No. 4, Def. 1).

As to A◦ : Similarly A ⊂ K (X × Y, K × L; C) is a strictly compact
subset of K (X × Y; C) , and

(ii) A◦ = {ν ∈ M (X × Y; C) : Rhf, νi > −1 for all f ∈ A } .

Let fR ∈ K (X × Y, K × L; C) , λ ∈ M (X; C) , µ ∈ M (Y; C) . Then

ϕ(f, λ) = f (x, ·) dλ(x) ∈ K (Y, L; C) , and
Z Z 
hf, λ ⊗ µi = f (x, y) dλ(x) dµ(y) (by (3) of No. 1)
Z


= f (x, ·) dλ(x) dµ

= hϕ(f, λ), µi .

If f ∈ A , λ ∈ B and µ ∈ C◦ , then ϕ(f, λ) ∈ C and, by (i),

(*) Rhf, λ ⊗ µi = Rhϕ(f, λ), µi > −1 ;

the validity of (*) for all f ∈ A says, by (ii), that λ ⊗ µ ∈ A ◦ . This proves
the assertion
(λ, µ) ∈ B × C◦ ⇒ λ ⊗ µ ∈ A◦ .
So to speak, B ⊗ C◦ ⊂ A◦ .
One notes that if µ ∈ M (Y; C) is such that λ ⊗ µ ∈ A ◦ for all λ ∈ B ,
then the validity of (*) for all f ∈ A and λ ∈ B shows, since ϕ(f, λ) then
runs over ϕ(A × B) = C , that µ ∈ C◦ . Thus if µ ∈ M (Y; C) , then

λ ⊗ µ ∈ A◦ for all λ ∈ B ⇒ µ ∈ C◦ ,

so to speak, B ⊗ µ ⊂ A◦ ⇒ µ ∈ C◦ . Combining this with the preceding

paragraph, we see that

C◦ = {µ ∈ M (Y; C) : B ⊗ µ ⊂ A◦ } .

Sets such as C and A figure in the topologies of strictly compact con-

vergence on M (Y; C) and M (X × Y; C) , respectively. Crucial for the rest
of the proof is that the sets C◦ (resp. A◦ ) form a fundamental system of
§4 products of measures INT III.x114

neighborhoods of 0 in M (Y; C) (resp. M (X × Y; C) ). The details are

given in the next note; in preparation, we give here the following:
Proposition . — If Z is a locally compact space and A is a strictly
compact subset of K (Z; C) then the closed balanced convex envelope of A is
strictly compact (therefore the closed balanced envelope and the closed convex
envelope of A are strictly compact).
Say A ⊂ K (Z, K; C) , K a compact subset of Z . Since K (Z, K; C)
is a closed topological subspace of K (Z; C) and is a Banach space, the
assertion is immediate from the following lemma:
Lemma. — If A is a compact subset of a complete Hausdorff locally
convex space G , then the closed balanced convex envelope of A is compact.
Proof. Let A1 be the balanced envelope of A , A2 the convex envelope
of A1 . Then A1 is compact (TVS, I, §1, No. 5, Prop. 3), and A 2 , being
the set of all convex combinations of elements of A 1 (TVS, II, §2, No. 3,
Cor. 1 of Prop. 8), is obviously balanced, thus A 2 is the balanced convex
envelope of A ; therefore A2 is precompact (TVS, II, §4, No. 1, Prop. 3),
b 2 of A 2 is compact. Since G is complete, A
that is, the completion A b 2 may
be identified with the closure A2 of A2 in G (GT, II, §3, No. 9, Cor. 1 of
Prop. 18), thus A2 is compact; since, moreover, A2 is balanced and convex
(TVS, I, §1, No. 5, Prop. 2 and II, §2, No. 6, Prop. 14), it is clear that A 2 is
the closed balanced convex envelope of A . (The lemma remains true in any
Hausdorff locally convex space G provided that the compact subset A is
contained in a complete balanced convex subset of G , cf. TVS, II, §4, No. 1,
Cor. of Prop. 3.)
III.47, `. −3 to −1.
“In view of the definition of the topology of strictly compact conver-
gence, this proves the proposition (TVS, III, §5, No. 3, Def. 2).”
The formula u(λ, µ) = λ ⊗ µ defines a bilinear mapping

u : M (X; C) × M (Y; C) → M (X × Y; C) .

Let us take on faith (details later) that the sets of the form A ◦ (resp. C◦ )
are a fundamental system of neighborhoods of 0 in M (X × Y; C) (resp.
M (Y; C) ) for the topology of strictly compact convergence. As in the note
for III.47, `. 13–17, let us write S (resp. T ) for the set of vaguely bounded
subsets of M (X; C) (resp. M (Y; C) ).
Given a neighborhood W = A◦ of 0 in M (X × Y; C) and a set
B ∈ S that is vaguely closed (not a restriction, since, in any real or complex
topological vector space, the closure of a bounded set is bounded—because
the closed neighborhoods of 0 are fundamental), we have produced a neigh-
borhood C◦ of 0 in M (Y; C) such that u(B × C◦ ) ⊂ A◦ , informally
INT III.x115 measures §4

B ⊗ C◦ ⊂ A◦ . In particular, if λ ∈ M (X; C) and B = {λ} , one sees that

µ 7→ λ ⊗ µ is continuous. A symmetric argument shows that λ 7→ λ ⊗ µ is
continuous for each µ (one begins with mappings

ω 0 : K (X × Y, K × L; C) → K (Y, L; K (X, K; C))

and ϕ0 : K (X × Y, K × L; C) × M (Y; C) → K (X, K; C) ), whence separate

continuity, and so the S-hypocontinuity of u is established. A symmetric
argument proves T-hypocontinuity.
The details. The following argument for X will apply as well to Y
and X × Y . To simplify the notations, we make the abbreviations K =
K (X; C) , M = M (X; C) ; K is equipped with the direct limit topology,
so that M = K 0 . A priori, the only obvious topology on M is the vague
topology, that is, the topology σ(M ; K ) derived from the canonical duality
Z
(f, λ) 7→ hf, λi = f dλ (f ∈ K , λ ∈ M ) .

But M is a set of functions on K , and “ λ j → λ vaguely” means simply

that λj (f ) → λ(f ) for each f ∈ K , whence λj → λ uniformly on every
finite subset of K ; thus vague convergence in M means uniform conver-
gence in the finite subsets of K . By modifying the sets of subsets of K
considered, one opens the door for other topologies on M . To wit:
Let A be the set of all strongly compact subsets A of K , and let S
be the set of all subsets of K that are contained in some A ∈ A ; since A
is closed under finite unions, the same is true of S , thus S is a bornology
on K (TVS, III, §1, No. 1, Def. 1). One knows (see the Proposition at the
end of the preceding note) that if A ∈ A , say A ⊂ K (X, K; C) , then the
closed balanced convex envelope of A in K is also compact, is contained
in K (X, K; C) , hence belongs to A ; the same is therefore true of every
subset of A , hence (varying A ) for every set S ∈ S . In particular, if
S ∈ S then the closure of S belongs to S (indeed, to A ). Moreover, if
S ∈ S and c ∈ C , then cS ∈ S . These properties may be summarized by
saying that the bornology S on K is adapted (loc. cit., No. 2, Def. 4), and,
in particular, it is convex (loc. cit., No. 1, Def. 2). {Another example: When
M is equipped with the vague topology, the set S of (vaguely) bounded
subsets of M is an adapted bornology on M (loc. cit., No. 2, Prop. 1).}
Now, M ⊂ F(K ; C) and S is a set of subsets of K ; the topol-
ogy of strictly compact convergence on M is by definition the S -topology,
that is, the topology of uniform convergence in the sets of S (GT, X, §1,
No. 2). Equipped with this topology, M = K 0 is also denoted Mscc =
LS (K ; C) = MS (X; C) , and is itself a locally convex space (TVS, III, §3,
No. 1, Cor. of Prop. 1).
§4 products of measures INT III.x116

We now describe a fundamental system of entourages for the uniformity

of S -convergence. A fundamental system of entourages for the uniformity
of C is given by the sets

Vε = {(c, d) ∈ C × C : |c − d| 6 ε } (ε > 0) ;

if S ∈ S , a fundamental system of entourages for the topology of uniform

convergence in S is given by the sets

Wε,S = {(λ, λ0 ) ∈ M × M : |λ(f ) − λ0 (f )| 6 ε for all f ∈ S } (ε > 0) ;

a fundamental system of entourages for the uniformity of S -convergence is

therefore given by the finite intersections

W ε1 , S 1 ∩ · · · ∩ W εn , S n

(GT, X, §1, No. 2, Remark 2). If ε = min {ε 1 , . . . , εn } and S = S1 ∪· · · ∪Sn ,

then
Wε, S ⊂ Wε1 , S1 ∩ · · · ∩ Wεn , Sn ;
it follows that the sets Wε, S form a fundamental system of entourages. But
if S ∈ S and A is the closed balanced convex envelope of S , then A ∈ S
and Wε, A ⊂ Wε, S ; thus the sets

Wε, A (ε > 0 , A ∈ A balanced and convex )

form a fundamental system of entourages. Now,

|λ(f ) − λ0 (f )| 6 ε ⇔ |λ(ε−1 f ) − λ0 (ε−1 f )| 6 1 ,

whence Wε, A = W1, ε−1 A ; since the ε−1 A run over the closed, balanced
convex elements of A as A does, the sets

W1, A ( A ∈ A balanced and convex)

form a fundamental system of entourages. Therefore the sets

UA = {λ ∈ M : |λ(f )| 6 1 for all f ∈ A } ,

as A runs over the set of balanced, convex, strictly compact subsets of K ,

form a fundamental system of neighborhoods of 0 in M for the topology
of S -convergence, that is, the topology of strictly compact convergence. In
the canonical duality between the vector spaces K and M ,

(*) U A = A◦
INT III.x117 measures §4

(TVS, II, §8, No. 4); thus the sets (*), as A runs over the set of balanced,
convex, strictly compact subsets of K , form a fundamental system of neigh-
borhoods of 0 ∈ M for the topology of strictly compact convergence.
Incidentally, by the theorem on bipolars (TVS, II, §6, No. 3, Th. 1),
◦◦
A is the closure of A in K for the weakened topology σ(K , M ) . But
the strongly closed convex set A is also closed for σ(K , M ) (loc. cit., Cor. 3
of Th. 1); thus A◦◦ = A .
III.48, `. 1, 2.
“The conclusion of Prop. 6 is no longer valid when the topology of
strictly compact convergence is replaced by the vague topology (Exer. 2 c)).”
A surprising result, since u is separately continuous for the vague topol-
ogy (Prop. 5).
In the cited exercise, X = Y = [0, 1] , and M (X; C) , M (X × X; C)
are equipped with the vague topology. Write u(λ, µ) = λ ⊗ µ as in the
preceding note. Let S be the set of vaguely bounded subsets of M (X; C) .
The set B = {µ ∈ M (X; C) : kµk 6 1 } is vaguely compact (§1, No. 9,
Cor. 2 of Prop. 15), hence vaguely bounded. If u were S-hypocontinuous
for the vague topology, then the restriction of u to B × M (X; C) would be
continuous (TVS, III, §5, No. 3, Prop. 4), contrary to part c) of the exercise.
III.48, `. 2–6.
“ . . . if B (resp. C ) is a vaguely bounded subset of M (X; C) (resp.
M (Y; C) ), then the image of B × C under the mapping (λ, µ) 7→ λ ⊗ µ
is vaguely bounded in M (X × Y; C) and therefore the restriction of this
mapping to B × C is vaguely continuous . . . ”
Let u : M (X)×M (Y) → M (X×Y) be the bilinear mapping u(λ, µ) =
λ ⊗ µ , and let S (resp. T ) be the set of vaguely bounded subsets of M (X)
(resp. M (Y) ); we know from Prop. 6 that u is (S, T)-hypcontinuous when
each of M (X) , M (Y) and M (X × Y) is equipped with the topology τ scc
of strictly compact convergence.
From §1, No. 10, Prop. 17, we know that S is also the set of subsets
of X that are bounded for τscc , and similarly for T and Y . By TVS, III,
§5, No. 3, Props. 4, 5, u(B × C) is bounded for τ scc —hence for the vague
topology—and the restriction of u to B × C is continuous when each of
B , C is equipped with the topology induced by τ scc , and B × C with the
product topology. Since B , C and u(B × C) are vaguely bounded, the
topologies induced on them by τscc and the vague topology are identical by
the Prop. 17 cited above; and the product topology on B × C coincides with
the topology induced by the product of the vague topologies on M (X)  and
M (Y) (GT, I, §4, No. 1, Cor. of Prop. 3); therefore the restriction u B × C
is continuous when all spaces in sight are equipped with the vague topology.
§4 products of measures INT III.x118

III.48, `. 10–14.
“The set of linear combinations of complex functions of the form
(x1 , x2 , . . . , xn ) 7→ f1 (x1 )f2 (x2 ) · · · fn (xn ) ,
where fi ∈ K (Xi ; C) (1 6 i 6 n) , may be identified canonically with the
N
n
tensor product K (Xi ; C) , and it follows from Lemma 1 of No. 1, by
i=1
induction on n , that this tensor product is dense in K (X; C) .”
The case n = 2 is proved in the remarks following the cited Lemma 1.
Assume inductively that all is well for for n − 1 and consider the
case of n . Write Y = X1 × · · · × Xn−1 . The canonical homeomorphism of
X1 ×· · ·×Xn with Y×Xn permits the identification of K (X1 ×· · ·×Xn ) with
K (Y×Xn ) . By the induction hypothesis, the set of functions f 1 ⊗· · ·⊗fn−1
is total in K (Y) , therefore the set of functions
f1 ⊗ · · · ⊗ fn = (f1 ⊗ · · · ⊗ fn−1 ) ⊗ fn ( fi ∈ K (Xi ) for i = 1, . . . , n)
is total in K (Y × Xn ) = K (X1 × · · · × Xn ) (see the Proposition in the note
for III.41, `. −9, −8).
III.49, `. −13 to −11.
“The integral notation and formula (14) may be extended in an obvious
way to functions f ∈ K (X; E) with values in a Hausdorff locally convex
space E , such that f (X) is contained in a complete convex subset of E .”
Preparatory to computing the case n = 3 , let us review the Remark in
No. 1 (which is the case n = 2 ). In the context of a measure ν = λ ⊗ µ
on X × Y , if f ∈ K (X × Y; E)R is such that f (X × Y) ⊂ C ⊂ E with C
complete and convex, so that f dν ∈ E by §3, No. 3, Prop. 7, then
Z Z Z Z Z
f dν = dµ(y) f (x, y) dλ(x) = dλ(x) f (x, y) dµ(y)

in the sense that R

(i) for each y ∈ Y , the integral f (x, y) dλ(x) exists and is an element
of E , R
(ii) the function h : Y → E defined by h(y) = f (x, y) dλ(x) belongs
to K (Y; RE) , and R
(iii) h dµ = f dν ;
thus, for each z0 ∈ E0 ,
Z DZ E DZ Z E
0 0
hf , z i dν = f dν, z = dµ(y) f (x, y) dλ(x), z0
Z DZ E
= f (x, y) dλ(x), z0 dµ(y)
Z Z 
= hf (x, y), z0 i dλ(x) dµ(y) ,
INT III.x119 measures §4

and similarly
DZ E DZ Z E
0
f dν, z = dλ(x) f (x, y) dµ(y), z0
Z DZ E
= f (x, y) dµ(y), z0 dλ(x)
Z Z 
= hf (x, y), z0 i dµ(y) dλ(x) .

Now consider locally compact spaces X 1 , X2 , X3 , and let Z = X1 ×

X2 × X3 , canonically identified with (X1 × X2 ) × X3 and X1 × (X2 × X3 ) .
Let µi be a measure on Xi (i = 1, 2, 3) and let f ∈ K (Z; E) be such that
f (Z) ⊂ C for some complete convex subset C of E R. Write ν = µ 1 ⊗µ2 ⊗µ3 =
(µ1 ⊗ µ2 ) ⊗ µ3 = µ1 ⊗ (µ2 ⊗ µ3 ) . We know that f dν ∈ E .
From the preliminary remarks, with λ = µ 1 ⊗ µ2 and µ = µ3 , we have
Z Z Z Z


f dν = f d (µ1 ⊗ µ2 ) × µ3 = dµ3 (x3 ) f (x1 , x2 , x3 ) d(µ1 ⊗ µ2 )(x1 , x2 )
Z Z Z 
= dµ3 (x3 ) dµ2 (x2 ) f (x1 , x2 , x3 ) dµ1 (x1 )
Z Z Z
= dµ3 (x3 ) dµ2 (x2 ) f (x1 , x2 , x3 ) dµ1 (x1 ) ,

as well as
Z Z Z Z 
f dν = dµ1 (x1 ) f (x1 , x2 , x3 ) dµ2 (x2 )
dµ3 (x3 )
Z Z Z
= dµ3 (x3 ) dµ1 (x1 ) f (x1 , x2 , x3 ) dµ2 (x2 ) .

This accounts for the permutations (3,2,1) and (3,1,2). On the other hand,
Z Z Z Z


f dν = f d (µ1 ⊗ µ2 ) ⊗ µ3 = d(µ1 ⊗ µ2 )(x1 , x2 ) f (x1 , x2 , x3 ) dµ3 (x3 )
Z Z Z 
= dµ1 (x1 ) dµ2 (x2 ) f (x1 , x2 , x3 ) dµ3 (x3 )
Z Z Z
= dµ1 (x1 ) dµ2 (x2 ) f (x1 , x2 , x3 ) dµ3 (x3 ) ,

as well as
Z Z Z Z 
f dν = dµ1 (x1 ) f (x1 , x2 , x3 ) dµ3 (x3 )
dµ2 (x2 )
Z Z Z
= dµ2 (x2 ) dµ1 (x1 ) f (x1 , x2 , x3 ) dµ3 (x3 ) ,
§4 products of measures INT III.x120

accounting for the permutations (1,2,3) and (2,1,3). Finally,

Z Z


f dν = f d µ1 ⊗ (µ2 ⊗ µ3 )

leads to formulas for the permutations (2,3,1) and (3,2,1).

III.50, `. −14, −13.
“It is clear that (M (Xα ; C), (pαβ )∗ ) is an inverse system of vector
spaces, and that ((pα )∗ ) is an inverse system of linear mappings . . . ”
If α 6 β , so that pαβ : Xβ → Xα , then, by the foregoing,

(pαβ )∗ : M (Xβ ; C) → M (Xα ; C)

is a linear mapping, continuous for the vague topology.

If α 6 β 6 γ then, from pαγ = pαβ ◦ pβγ (GT, I, §4, No. 4) we infer
that, for every f ∈ C (Xγ ; C) ,

pαγ 0 (f ) = f ◦ pαγ = (f ◦ pαβ ) ◦ pβγ

= pβγ 0 (f ◦ pαβ ) = pβγ 0 pαβ 0 (f ) ,

thus pαγ 0 = pβγ 0 ◦ pαβ 0 ; taking transpose,

(I) (pαγ )∗ = (pαβ )∗ ◦ (pβγ )∗ when α 6 β 6 γ .

Since pαα is the identity mapping of Xα ,

(II) (pαα )∗ is the identity mapping of M (Xα ; C) .

Thus, assuming

M (Xα ; C) equipped with the vague topology,
the family
M (Xα ; C) of topological vector spaces and the family (pαβ )∗ of contin-
uous linear mappings form an inverse system


(†) M (Xα ; C), (pαβ )∗

(S, III, §7, No. 1 and GT, loc. cit.).

Recall that the inverse limit

X = lim Xα
←−
Q
is a closed subspace of the compact space Xα (GT, I, §8, No. 2, Cor. 2 of
α
Prop. 7), the canonical mapping

pα : X → X α
INT III.x121 measures §4

of X into Xα is defined to be the restriction to X of the projection map-

ping pr α of the product space into Xα (GT, I, §4, No. 4). When Xα is
nonempty for every α , the compact space X is also nonempty (GT, I, §9,
No. 6, Prop. 8).
From the relation pα = pαβ ◦ pβ when α 6 β , one infers that

(*) (pα )∗ = (pαβ )∗ ◦ (pβ )∗ when α 6 β .

The sense in which the term “inverse system” is applied to the family of
mappings
(pα )∗ : M (X; C) → M (Xα ; C)
is explained in S, III, §7, No. 2, Remark 2 (reviewed at the end of this note),
but what the property (*) is called is not important; what is important is
the role that it plays in Definition 2 below, as follows.
Consider the inverse limit of the system (†), namely

lim M (Xα ; C) ;
←−

Q
its elements are the families (µα ) ∈ M (Xα ; C) such that µα = (pαβ )∗ (µβ )
α
when α 6 β .
For example, suppose µ is a measure on the compact space X = lim Xα .
←−
For each α , define a linear form µα on C (Xα ; C) by


µα (f ) = µ(f ◦ pα ) f ∈ C (Xα ; C) ;

since, for f ∈ C (Xα ; C) , f ◦ pα belongs to C (X; C) , one has

|µα (f )| 6 kµk · kf ◦ pα k 6 kµk · kf k ,

thus µα is a (bounded) measure on the compact space X α such that

kµα k 6 kµk . By definition, for all f ∈ C (Xα ; C) one has


µα (f ) = µ(f ◦ pα ) = µ pα 0 (f )


= (µ ◦ pα 0 )(f ) = [t (pα 0 )(µ)](f ) = (pα )∗ (µ) (f ) ,

whence (pα )∗ (µ) = µα for all α . It then follows from (*) that, for α 6 β ,


(pαβ )∗ (µβ ) = (pαβ )∗ (pβ )∗ (µ)


(pαβ )∗ ◦ (pβ )∗ (µ) = (pα )∗ (µ) = µα ,
§4 products of measures INT III.x122

thus the family (µα ) is (in the sense of Def. 2) an inverse system of measures
admitting µ as inverse limit. One can therefore define a (linear) mapping

Φ : M (X; C) → lim M (Xα ; C)

←−

by the formula



Φ(µ) = (pα )∗ (µ) µ ∈ M (X; C) .

By Def. 2 the range of Φ consists of the set of all inverse systems of measures
that admit an inverse limit. A preview of Prop. 8:
By (i) of Prop. 8, Φ is injective.
By (ii) of Prop. 8, if (µα ) is an inverse system admitting an inverse
limit µ , then the family (kµα k) is bounded (indeed, (µα ) = Φ(µ) satisfies
kµα k 6 kµk for all α as noted above).
By (iii) of Prop. 8, if the pαβ are surjective, then every inverse sys-
tem (µα ) for which (kµα k) is bounded has an inverse limit; in this case,
the range of Φ is precisely the set of elements (µ α ) of lim M (Xα ; C) for
←−
which (kµα k) is bounded, and Φ is a bijection of M (X; C) onto this set
(a linear subspace of lim M (Xα ; C) ).
←−
By (iv) of Prop. 8, if the pαβ are surjective, then Φ is a bijection
of M+ (X; C) onto the set of elements (µα ) of lim M (Xα ; C) such that
←−
µα > 0 for all α , and, when µ ∈ M+ (X; C) , kµα k
= kµk for all α .
To summarize, the family of mappings (pα )∗ and their property (*)
are the link between inverse systems of measures on the X α and possible
measures µ on the space X = lim Xα that may correspond to them.
←−


We now review the justification for calling the family (pα )∗ an “inverse
system” of mappings.
Let (Eα , fαβ ) be an inverse system, E = lim Eα , and fα : E → Eα the
←−
canonical mapping.
{In the application at hand, Eα = M (Xα ) (we drop the “ ;C” for
brevity), fαβ = (pαβ )∗ , E = lim M (Xα ) and fα the restriction of
Q ←−

prα : M (Xβ ) → M (Xα ) to lim M (Xβ ) . Thus fα (µβ ) = µα for every
β ←−
inverse system of measures µβ ∈ M (Xβ ) .}
Let (Fα , gαβ ) be a second inverse system, F = lim Fα , gα : F → Fα
←−
the canonical mapping.
{In the application at hand, Fα = M (X) for all α , gαβ is the identity
mapping on M (X) ; the elements (µα ) of lim Fα are then the “constant
←−
families” (µα ) where, for some µ ∈ M (X) , µα = µ for all α ; and the
INT III.x123 measures §4

canonical mapping gα : lim Fβ → Fα assigns to a constant family (µ) the

←− Q
value µ . In effect, lim Fα is the “diagonal” of the product space Fα ; one
←−
may identify lim Fα with F = M (X) , and gα with the identity mapping
←−
id on M (X) .}
Suppose that for each α we are given a mapping u α : Fα → Eα such
that, when α 6 β , the diagram
uβ -
Fβ Eβ

gαβ fαβ
? ?
Fα u - Eα
α

is commutative, that is, uα ◦ gαβ = fαβ ◦ uβ when α 6 β . One then says

that the family (uα ) is an inverse system of mappings of (F α , gαβ ) into
(Eα , fαβ ) (S, III, §7, No. 2).
{In the application at hand, uα = (pα )∗ , the diagram takes the form
(pβ )-
∗
M (X) M (Xβ )

id (pαβ )∗
? ?
M (X) - M (Xα )
(pα )∗
whose commutativity reduces to the relation

(*) (pα )∗ = (pαβ )∗ ◦ (pβ )∗

whenever α 6 β .}
By S, III, §7, No. 2, Cor. 1 of Prop. 1, there exists a unique mapping
u : F → E such that the diagram

F u- E

gα fα
? ?
Fα u - Eα
α

is commutative for all α . The mapping u is then called the inverse limit
of the family (uα ) , written u = lim uα (S, III, §7, No. 2).
←−
{In the application at hand, the diagram takes the form
§4 products of measures INT III.x124

M (X) u - lim M (X )
β
←−

id fα
? ?
M (X) - M (Xα )
(pα )∗
Q
where fα is the restriction of prα : M (Xβ ) → M (Xα ) to lim M (Xβ ) .

←−
The property required of u is that fα u(µ) = (pα )∗ (µ) for all α ; since


Φ(µ) = (pβ )∗ (µ) has this property, Φ = u by the uniqueness of u . Thus


(pα )∗ is an inverse system of mappings and Φ = lim (pα )∗ .}
←−

{Another application of inverse limits occurs in the theory of Souslin

spaces (GT, IX, §6, No. 5).}
III.51, `. 10–13.
“ . . . which shows that

g + h = (gβ ◦ pβα + hγ ◦ pγα ) ◦ pα

belongs to F . . . ”
In slow motion,

g + h = g β ◦ pβ + hγ ◦ pγ
(*) = gβ ◦ (pβα ◦ pα ) + hγ ◦ (gγα ◦ pα )
= (gβ ◦ pβα ) ◦ pα + (hγ ◦ pγα ) ◦ pα .

Now, A = gβ ◦ pβα and B = hγ ◦ pγα are continuous functions Xα → C ,

that is, elements of C (Xα ; C) ; therefore their sum is defined and belongs to
C (Xα ; C) , say A + B = fα ∈ C (Xα ; C) , and, by the definition of sum,

(A + B)(xα ) = A(xα ) + B(xα ) for all xα ∈ Xα .

In particular,





A + B pα (x) = A pα (x) + B pα (x) for all x ∈ X ,

that is, (A + B) ◦ pα = A ◦ pα + B ◦ pα as functions X → C . In view of (*),

g + h = A ◦ pα + B ◦ pα = (A + B) ◦ pα = fα ◦ pα ,

where fα = A + B ∈ C (Xα ; C) , consequently g + h ∈ F .

INT III.x125 measures §4

III.52, `. 10.
“ . . . (pα )∗ (µ) = µα for all α ∈ I . . . ”
For every fα ∈ C (Xα ; C) ,


µα (fα ) = λ(fα ◦ pα ) = µ(fα ◦ pα ) = (pα )∗ (µ) (fα ) ,
whence µα = (pα )∗ (µ) .
III.52, `. 13.
“ . . . the relation µα = (pαβ )∗ (µβ ) implies that µα (1) = µβ (1) . . . ”
Write 1α ∈ C (Xα ; C) for the constant function equal to 1 . The point
is that 1α ◦ pαβ = 1β , thus


µβ (1β ) = µβ (1α ◦ pαβ ) = (pαβ )∗ (µβ ) (1α ) = µα (1α ) .
III.52, `. 15, 16.
“ . . . the subspace F obviously satisfies the property (P) of §1, No. 7,
Prop. 9 . . . ”
The constant function 1 belongs to F because 1 = 1 α ◦ pα , where
1α ∈ C (Xα ; C) is the constant function equal to 1 , and f = 1 trivially
satisfies the requirements of (P) in the cited Prop. 9.
III.52, `. −13 to −10.
“We know that (XJ , prJK ) is an inverse system of compact spaces, and
that the inverse limit of the system of continuous mappings (pr J ) is a hom-
eomorphism of X onto the inverse limit space lim XJ , permitting one to
←−
identify these two spaces (GT, I, §4, No. 4 and S, III, §7, No. 2, Remark 3).”
I had great difficulty applying the cited references, making many mis-
takes along the way; be suspicious of every move in the following exposition.
To recapitulate the notations, we are given a family of nonempty com-
pact spaces, and Y
X= Xλ
λ∈L

is the product topological space, the topology being defined as the initial
topology for the family (prλ )λ∈L of projection mappings
prλ : X → Xλ .
No order relation on the index set L is assumed.
Let F be the set of all nonempty finite subsets J of L ; F is an in-
creasing directed set for the order relation J ⊂ K on F . For each J ∈ F
one writes Y
XJ = Xλ ,
λ∈J
§4 products of measures INT III.x126

and the mapping

prJ : X → XJ
is defined, for x = (xλ )λ∈L in X by

prJ x = (xλ )λ∈J = (prλ x)λ∈J ;

so to speak, prJ “masks out” the coordinates λ ∈ L --- J . If J, K ∈ F with

J ⊂ K , then
prJK : XK → XJ
is defined to be the mapping (xλ )λ∈K 7→ (xλ )λ∈J (thus masking out the
coordinates λ ∈ K --- J ); also prJJ = id on XJ and, when I ⊂ J ⊂ K ∈ F ,

prIJ ◦ prJK = prIK .

To summarize,
(XJ , prJK )
is an inverse system
Q of sets relative to the index set F (S, III, §7, No. 1).
Writing Z = XJ , the inverse limit
J∈F

lim XJ
←−

is the subset of the compact space Z consisting of the elements (z J )J∈F ∈ Z

such that
prJK zK = zJ when J ⊂ K ;
explicitly, if zK = (aµ )µ∈K , where aµ ∈ Xµ for all µ ∈ K , then prJK zK =
(aµ )µ∈J . In particular, if λ ∈ K and J = {λ} , then

pr{λ}K zK = (aµ )µ∈{λ} = aλ ;

thus, when λ ∈ K , the effect of pr{λ}K on an element zK of XK is to single

out the λ’th coordinate of zK .

For all x = (xλ )λ∈L ∈ X , prJK (prK x) = prJK (xλ )λ∈K = (xλ )λ∈J =
prJ x , thus

(*) prJK ◦ prK = prJ

as functions X → XJ . We write

Y = lim XJ
←−
INT III.x127 measures §4

and, for J ∈ F ,
pJ : Y → X J
for the restriction to Y of the J ’th coordinate projection mapping Z → X J .
By GT, I, §9, No. 6, Prop. 8, Y = lim XJ is compact and nonempty;
←−
our problem is to show that X is homeomorphic to Y . At any rate, there
is a natural mapping h : X → Y , defined by

h(x) = (prJ x)J∈F ,

where h(x) ∈ Y by (*); explicitly, if J ⊂ K then

prJK (prK x) = (prJK ◦ prK )x = prJ x .

For every K ∈ F and every λ ∈ K , we have the commutative diagram

Q h - Y = lim XJ
Xλ = X
λ∈L ←−

Q
Q
prλ Q pK
prK Q
Q
? sQ ?
Xλ  pr XK
K,λ

where prK,λ is the λ’th coordinate projection on the product space X K . The
equality pK ◦ h = prK expresses the definition of h , and pr K,λ ◦ prK = prλ
is clear from the definition of prK,λ .
We prove that h is a homeomorphism in three steps:
(i) h is injective;
(ii) h is continuous;
(iii) h is surjective.
Proof of (i): If h(x) = h(x0 ) then, for every J ∈ F , prJ x = prJ x0 ;
thus if λ ∈ L and if J is an element of F containing λ , then

prJ,λ (prJ x) = prJ,λ (prJ x0 ) ,

that is, prλ x = prλ x0 .

Proof of (ii): Y bears the initial topology for the family of mappings
(pK )K∈F , and pK ◦ h = prK is continuous for every K , therefore h is
continuous.
Proof of (iii): Given y = (yJ )J∈F ∈ Y , we seek an element x ∈ X such
that h(x) = y , that is, prJ x = yJ for all J ∈ F .
Given any λ ∈ L , we have {λ} ∈ F and y{λ} ∈ X{λ} = Xλ , therefore
y{λ} = xλ for a unique xλ ∈ Xλ . Define x = (xλ )λ∈L .
§4 products of measures INT III.x128

As noted earlier, if K is any element of F that contains λ , then the

effect of pr{λ}K on an element zK of XK is to single out its λ’th coordinate;
in particular, pr{λ}K yK = y{λ} = xλ shows that the λ ’th coordinate of yK
is xλ .
Now let J be any element of F and let us show that pr J x = yJ . Say
yJ = (aλ )λ∈J , where aλ ∈ Xλ for all λ ∈ J . For every λ ∈ J , we know
from the preceding paragraph that the λ’th coordinate of y J is xλ , that is,
aλ = xλ , whence yJ = (xλ )λ∈J = prJ x .
III.52, `. −9 to −6.
“Since the projections prJ,K are surjective, it follows from Prop. 8 that
the set M (X; C) (resp. M+ (X) ) may be identified with the set of inverse
systems (µJ ) such that the family of norms (kµJ k) is bounded (resp. such
that the µJ are all positive, and necessarily of the same total mass).”
Both notations prJK and prJ,K are employed in the Example to de-
note the canonical projection XK → XJ . The former, without a separating
comma, is consistent with the general notation for an inverse system, but
the latter is used here three times.
By the discussion at the beginning of No. 5, to a continuous map-
ping p : A → B between compact spaces there is associated a mapping
p∗ : M (A; C) → M (B; C) , p∗ = t (p0 ) . The correspondence evidently
satisfies (p ◦ q)∗ = p∗ ◦ q ∗ and (1A )∗ = 1M (A;C) . (In the language of
categories, p 7→ p∗ defines a covariant functor between the appropriate
categories.) When p is surjective, p 0 preserves norm,
Q hence so does p∗ .
As shown in the preceding note, if X = Xλ and Y = lim XJ
λ∈L ←−
( J ∈ F ) , then the mapping h : X → Y defined by h(x) = (pr J x )J∈F is
a homeomorphism. It follows that the mapping

(i) h∗ : M (X) → M (Y)

defined, for µ ∈ M (X)) and g ∈ C (Y) by

hh∗ (µ), gi = ht (h0 )(µ), gi = hµ, h0 (g)i = hµ, g ◦ hi

is an isomorphism of vector spaces such that µ > 0 ⇔ h ∗ (µ) > 0 , and
kh∗ (µ)k = kµk (because h0 preserves norm).
On the other hand, by the note for III.50, `. −14, −13, there is a linear
mapping

(ii) Φ : M (Y) → lim M (XJ ) ,

←−


defined by Φ(ν) = (pJ )∗ (ν) J∈F , that is injective (by (i) of Prop. 8), the


range Φ M (Y) of Φ is the set of inverse systems (µJ )J∈F of measures
INT III.x129 measures §4

on the XJ for which the norms kµJ k are bounded (by (iii) of Prop. 8; see

also the note for III. 50, `. −14, −13); and, by (iv) of Prop. 8, Φ M+ (X)
is the set of all inverse systems (µJ )J∈F for which µJ > 0 for all J , in
which case kµJ k is constant, specifically, if Φ(µ) = (µ J )J∈F then µ > 0
and kµJ k = kµk for all J ∈ F .
Combining the mappings (i) and (ii), we have a mapping

(iii) θ = Φ ◦ h∗ : M (X) → lim M (XJ )

←−


such that θ is linear, injective and preserves positivity, its range θ M (X)
is the set of all inverse systems (µJ )J∈F of measures

on the XJ for which the
norms kµJ k are bounded; and θ M+ (X) is the set of all inverse systems
(µJ )J∈F for which µJ > 0 for all J , in which case, if θ(µ) = (µ J )J∈F , one
has kµJ k = kµk for all J .
Let us calculate θ(µ) explicitly for µ ∈ M (X) . Writing ν = h ∗ (µ) ∈
M (Y) , we have




θ(µ) = Φ ◦ h∗ (µ) = Φ h∗ (µ) = Φ(ν) = (pJ )∗ (ν) J∈F ,

where (pJ )∗ (ν) ∈ M (XJ ) is the measure on XJ whose effect on a function

fJ ∈ C (XJ ) is given by

h(pJ )∗ (ν), fJ i = hν, fJ ◦ pJ i

= hh∗ (µ), fJ ◦ pJ i
= hµ, (fJ ◦ pJ ) ◦ hi
= hµ, fJ ◦ (pJ ◦ h)i
= h(pJ ◦ h)∗ (µ), fJ i ,

whence (pJ )∗ (ν) = (pJ ◦ h)∗ (µ) , and so

θ(µ) = (pJ ◦ h)∗ (µ) J∈F
;

thus the inverse family θ(µ) = (µJ )J∈F is given by

(iv) µJ = (pJ ◦ h)∗ (µ) for all J ∈ F .

Now, pJ ◦ h = prJ by the definitions of h and pJ , thus the meaning of (iv)

is that, for all J ∈ F and all fJ ∈ C (XJ ) , one has

hµJ , fJ i = h(pJ ◦ h)∗ (µ), fJ i

= hµ, fJ ◦ (pJ ◦ h)i = hµ, fJ ◦ prJ i ,
§4 products of measures INT III.x130

that is,

(v) µJ (fJ ) = µ(fJ ◦ prJ ) for all J ∈ F and fJ ∈ C (XJ ) .

We may regard (v) in two lights:

(a) given any µ ∈ M (X) , the formulas (v) define an inverse sys-
tem (µJ )J∈F of measures on the XJ , so to speak providing a represen-
tation of µ as the inverse limit of a family of measures on the X J , concisely
µ = (µJ )J∈F ;
(b) given any inverse system (µJ )J∈F of measures on the XJ , linked by
the formulas
(prJK )∗ (µK ) = µJ when J ⊂ K ∈ F ,
that is,

hµJ , fJ i = hµK , fJ ◦ prJK i when J ⊂ K ∈ F and fJ ∈ C (XJ ),

then there exists a unique measure µ ∈ M (X) satisfying (v), providing,

so to speak, a representation of the inverse system (µ J )J∈F of measures on
the XJ as a measure µ on X , concisely (µJ )J∈F = µ . It is in the sense
of this correspondence that one “identifies” measures on X with inverse
systems of measures on the XJ . In view of Prop. 8, the inverse systems
(µJ )J∈F on the XJ that correspond to measures on X are precisely those
for which the norms kµJ k are bounded; and an inverse system (µ J )J∈F
corresponds to a positive measure µ ∈ M + (X) if and only if µJ > 0 for
all J ∈ F , a condition that implies that the norms kµ J k = µJ (1XJ ) are all
equal to each other (if J, K ∈ F then J∪K ∈ F provides a link between kµ J k
and kµK k ) hence are bounded, so that the system (µ J )J∈F corresponds to
a measure µ > 0 on X and in fact kµJ k = µJ (1XJ ) = µ(1X ) = kµk for
all J ∈ F .
III.52, `. −3.
For “subsets of I ” read “subsets of L ”.
III.52, `. −3 to −1.
“ . . . we have, by virtue of formula (14) of No. 4,
Y
(17) µK (fJ ◦ prJ,K ) = µJ (fJ ) · µλ (1) . ”
λ∈K -- J

One is assuming that J ⊂ K , and fJ is a given function in C (XJ ; C) .

Define fK = fJ ◦ prJ,K . If J = K , the formula reduces to µJ (fJ ) = µJ (fJ ) .
Let us assume that J has j elements, K has k elements, and j < k . To
simplify the notation, let us suppose that

J = {1, . . . , j} , K = {1, . . . , j, j + 1, . . . , k} .
INT III.x131 measures §4

Then for each fixed (x1 , . . . , xj ) ∈ XJ and every (xj+1 , . . . , xk ) ∈ XK -- J ,

one has
fK (x1 , . . . , xj , xj+1 , . . . , xk ) = fJ (x1 , . . . , xj ) · 1 ;
by the cited formula (14), successive integration with respect to the variables
xj+1 , . . . , xk produces the factors µj+1 (1), . . . , µk (1) , thus
Z Y
(*) fK (x1 , . . . , xj , xj+1 , . . . , xk ) dµK -- J = fJ (x1 , . . . , xj ) · µi (1) .
i∈K -- J

Denote groups of variables by x = (x1 , . . . , xj , xj+1 , . . . , xk ) ∈ XK ,

y = (x1 , . . . , xj ) ∈ XJ , z = (xj+1 , . . . , xk ) ∈ XK -- J , making the identifica-
tion x = (y, z) . Then

fK (x) = fK (y, z) = fJ (y) · 1 ,

and the formula (*) may be written

Z Y
fK (y, z) dµK -- J (z) = fJ (y) · µi (1) .
i∈K -- J

Then, citing (14) for the case of two variables,

Z
µK (fJ ◦ prJ,K ) = µK (fK ) = fK d(µJ ⊗ µK -- J )
Z Z 
= dµJ (y) fK (y, z) dµK -- J (z)
Z  Y 
= dµJ (y) fJ (y) · µi (1)
i∈K -- J
Y
= µJ (fJ ) · µi (1) .
i∈K -- J

III.53, `. 1, 2.
“For (µJ ) to be an inverse system of measures, it is therefore necessary
and sufficient that either µλ = 0 for all λ ∈ L or µλ (1) = 1 for all
λ ∈ L .”
As in the preceding notes on the Example, F denotes the set of all
nonempty finite subsets J of L . The condition (17) may be written

Y
(prJ,K )∗ (µK ) (fJ ) = µJ (fJ ) · µλ (1)
λ∈K -- J
§4 products of measures INT III.x132

whenever J ⊂ K ∈ F and fJ ∈ C (XJ ; C) , that is,

 Y 
(prJ,K )∗ (µK ) = µλ (1) · µJ when J ⊂ K ∈ F ;
λ∈K -- J

in order that (µJ )J∈F be an inverse system of measures on the X J , it is

therefore necessary and sufficient that
 Y 
(*) µJ = µλ (1) · µJ when J ⊂ K ∈ F .
λ∈K -- J

If µλ (1) = 1 for all λ ∈ L , it is obvious that (*) holds, therefore

(µJ )J∈F is an inverse system. Whereas if µλ = 0 for all λ ∈ L , then
µJ = 0 for all J ∈ F and (µJ )J∈F is trivially an inverse system.
Conversely, suppose (µJ ) is an inverse system, that is, satisfies the
condition (*). We are to show that either µ λ = 0 for all λ ∈ L or µλ (1) = 1
for all λ ∈ L ; assuming there exists a λ 0 ∈ L such that µλ0 6= 0 , it will
suffice to show that µλ (1) = 1 for all λ ∈ L . Let λ be any element of L
such that λ 6= λ0 . Set J = {λ0 } , K = {λ0 , λ} . Then µJ = µλ0 and the
condition (*) reads µλ0 = µλ (1) · µλ0 , whence µλ (1) = 1 . In particular
µλ 6= 0 , so the roles of λ and λ0 can be interchanged, leading to µλ0 = 1 ,
consequently µλ (1) = 1 for all λ ∈ L .
III.53, `. 5, 6.
“We
N retain the notations of the Example of No. 5, so that in particular
µJ = µλ for every finite subset J of L .”
λ∈J
As in the preceding notes, we write F for the set of all nonempty finite
subsets J, K, . . . of L . The following remarks, or something like them,
should have been made at theQ outset of the Example of No. 5.
(i) What does XJ = Xλ mean? It is the set of all (finite) fami-
λ∈J
lies (xλ )λ∈J such that xλ ∈ Xλ for all λ ∈ J , equipped with the initial
topology for a family of canonical projections. Its elements are described
independently of any ordering that might exist in L or in J . If J ⊂ K ∈ F
then the formulas

XK = XJ × XK -- J = XK -- J × XJ

are a consequence of the associative law for products (GT, I, §4, No. 1,
Prop. 2), the equalities expressing canonical homeomorphisms. Thus if
y ∈ XJ and z ∈ XK -- J then (y, z) and (z, y) are representations of a
same element x of XK .
INT III.x133 measures §4
N
(ii) If fλ ∈ C (Xλ ; C) for all λ ∈ J , what does fλ mean? It is the
λ∈J
function f on XJ defined by

Y
f (xλ )λ∈J = fλ (xλ ) ,
λ∈J

the product being independent of any hypothetical ordering

N of the λ’s.
(iii) If µλ ∈ M (Xλ ; C) for all λ ∈ J , what does µλ mean? It is
λ∈J
the unique measure µJ on XJ characterized by the relations
O  Y
µJ fλ = µλ (fλ ) ( fλ ∈ C (Xλ ; C) for all λ ∈ J ) ,
λ∈J λ∈J

the product being independent of any consideration of ordering. If

J ⊂ K ∈ F then the formulas

µK = µJ ⊗ µK -- J = µK -- J ⊗ µJ

exhibit three representations of the same measure on X K , an instance of the

associative law for finite products of measures (No. 4, Prop. 7).
III.53, `. −5 to −2.
“It then follows from Prop. 8 of No. 5 that there exists a positive measure
µ0 on X of total mass 1 , such that µ0 (fJ ◦ prJ ) = µ0J (fJ ) for every finite
subset J of L and every function fJ ∈ C (XJ ; C) .”
As in previous notes, we write F for the set of all nonempty finite
subsets of L . Since µ0λ (1) = 1 for all λ , the measures
O
µ0J = µ0λ
λ∈J

form an inverse system of measures on the X J by the last sentenceQ 0 of No. 5.

It is clear that for every J ∈ F , one has µ 0J > 0 and µ0J (1) = µλ (1) = 1 .
λ∈J
By (iv) of the cited Prop. 8 there exists an inverse limit measure µ 0 = lim µ0J
←−
on X that is > 0 and satisfies kµ0 k = µ0 (1) =µ0J (1) = 1 for all J ∈ F . In
particular,

µ0 (fJ ◦ prJ ) = µ0J (fJ ) for all J ∈ F and fJ ∈ C (XJ ; C)

(see the formula (v) in the note for III.52, `. −13 to −10).
§4 products of measures INT III.x134

III.53, `. −2 to III.54, `. 3.
“The positive measure
!
Y
µ= µλ (1) µ0
λ∈L

then meets the requirements, since

Y
µJ (fJ ) = µ0J (fJ ) · µλ (1) ,
λ∈J
Y Y Y
µλ (1) = µλ (1) · µλ (1) . ”
λ∈L λ∈J λ∈L -- J

Q
The factorization of µλ (1) in the preceding line is a special case
λ∈L
of an associativity theorem for multipliable families (GT, IV, §7, No. 5,
Remark ).
For all J ∈ F we have
O O
−1
µ0J = µ0λ = µλ (1) µλ
λ∈J λ∈J
Y −1 O Y −1
= µλ (1) · µλ = µλ (1) · µJ ,
λ∈J λ∈J λ∈J

whence
Y 
(*) µJ = µλ (1) · µ0J ;
λ∈J

therefore, for all fJ ∈ C (XJ ; C) ,

Y 
µ(fJ ◦ prJ ) = µλ (1) µ0 (fJ ◦ prJ ) (definition of µ )
λ∈L
Y 
= µλ (1) µ0J (fJ ) (preceding note)
λ∈L
Y  Y
= µλ (1) · µ0J (fJ ) · µλ (1)
λ∈J λ∈L -- J
Y
= µJ (fJ ) · µλ (1) (by (*))
λ∈L -- J
INT III.x135 measures §4

whence the formula (18). And, since µ 0 (1) = 1 , the defining formula
Y 
µ= µλ (1) µ0
λ∈L

yields the formula (19).

III.54, `. 11–13.
“This follows at once from the formulas (18) and (19) and the associa-
tivity of the product for multipliable families in R + (GT, IV, §7, No. 5,
Remark ).”
(The term “at once” comes in various sizes. Am I overlooking a dramatic
simplification? The unpleasant length of the following argument carries with
it an increased risk of error.)
By the cited Remark, for every ρ ∈ R thesubfamily (µ λ (1))λ∈Lρ is
Q
multipliable, the family of products µλ (1) is multipliable, and
λ∈Lρ ρ∈R

Y Y  Y
(1) µλ (1) = µλ (1) .
ρ∈R λ∈Lρ λ∈L

Therefore, by Prop. 9, for each ρ ∈ R there exists the positive measure

O Y
(2) νρ = µλ on Yρ = Xλ
λ∈Lρ λ∈Lρ

satisfying

Y
(3) νρ (fJ ◦ prJ ) = µJ (fJ ) µλ (1)
λ∈Lρ -- J

for every finite J ⊂ Lρ and every fJ ∈ C (XJ ; C) , and one has

Y
(4) νρ (1) = µλ (1) ;
λ∈Lρ

moreover, since the family νρ (1) ρ∈R
is multipliable, Prop. 9 yields a pos-
itive measure
O Y
(5) ν= νρ on Yρ = X
ρ∈R ρ∈R
§4 products of measures INT III.x136

such that, writing

O
(6) νS = νρ for finite S ⊂ R ,
ρ∈S

one has
Y
(7) ν(gS ◦ prS ) = νS (gS ) νρ (1)
ρ∈R -- S

for finite S ⊂ R and all gS ∈ C (YS ; C) , where

Y Y Y  Y
(8) YS = Yρ = Xλ = Xλ
ρ∈S ρ∈S λ∈Lρ
S
λ∈ Lρ
ρ∈S

and, by (4) and (1),

Y Y
(9) ν(1) = νρ (1) = µλ (1) .
ρ∈R λ∈L

On the other hand, by Prop. 9 there exists the product measure

O Y
(10) µ= µλ on X = Xλ
λ∈L λ∈L

satisfying
Y
(11) µ(fJ ◦ prJ ) = µJ (fJ ) µλ (1)
λ∈L -- J

for all finite J ⊂ L and all fJ ∈ C (XJ ; C) , and one has

Y
(12) µ(1) = µλ (1) ( = ν(1) by (9) ) .
λ∈L

By (5) and (2),

O O O  Y
(13) ν= νρ = µλ on Yρ = X ;
ρ∈R ρ∈R λ∈Lρ ρ∈R

N
since µ = µλ on X , the problem is to show that ν = µ . So to speak,
Q λ∈L
Q
Xλ and Yρ are two representations of the same space X ; we are to
λ∈L ρ∈R
show that µ and ν are two representations of the same measure on X .
INT III.x137 measures §4
S
To this end, let J be a finite subset of L = Lρ and let fJ ∈
ρ∈R
C (XJ ; C) . The set J can intersect only finitely many of the mutually
S disjoint
sets Lρ ; let S be a nonempty finite subset of R such that J ⊂ Lρ , which
ρ∈S
we can suppose to be minimal, so that J ∩ L ρ 6= ∅ for every ρ ∈ S . Then
[ [
(14) J= J ∩ Lρ = Jρ ,
ρ∈S ρ∈S

where the sets Jρ = J ∩ Lρ ( ρ ∈ S ) areSpairwise disjoint andSnonempty.

In view of (8) and the fact J ⊂ Lρ , setting T = Lρ one can
ρ∈S ρ∈S
write Y Y
YS = Xλ = X J × Xλ ,
λ∈T λ∈T -- J
hence one can define a function gS ∈ C (YS ; C) by the formula

(15) gS = fJ ◦ prJ ,

where prJ is the J’th projection in the contextQ of the space Y S . Q

Writing
prS for the S’th projection in the context of Yρ = X = Y S × Yρ ,
ρ∈R ρ∈R -- S
one then has

(16) gS ◦ prS = (fJ ◦ prJ ) ◦ prS = fJ ◦ (prJ ◦ prS ) = fJ ◦ prJ ;

here prJ is the J’th projection in the context of

Y Y
X= Xλ = X J × Xλ ,
λ∈L λ∈L -- J

thus

(17) gS ◦ prS = fJ ◦ prJ

Q Q
on the space Yρ = X = Xλ (see (5) and (10)); this is the link
ρ∈R λ∈L
between the presentations ν and µ of the measure on X . Then

ν(fJ ◦ prJ ) = ν(gS ◦ prS ) (by (17))

Y
(18) = νS (gS ) νρ (1) (by (7)) .
ρ∈R -- S
§4 products of measures INT III.x138
S
Consider again the decomposition J = Jρ of (14). Let s = card S ,
ρ∈S
j = card J . Since each Jρ contains at least one element of J , one has s 6 j .
Let us index S as follows:

S = {ρ1 , . . . , ρs } ( s = card S ) .

Then J is the disjoint union

J = J ρ1 ∪ · · · ∪ J ρs ,

which yields a factorization

s
Y
XJ =XJρ1 × · · · × XJρs = X J ρi
i=1
(19) Y Y  Y
= Xλ = X Jρ .
ρ∈S λ∈Jρ ρ∈S

This is the crucial step; functionsQon X J ostensibly of j variables xλ ∈ Xλ

(λ ∈ J) in the context of X = Xλ , may be regarded as functions of s
λ∈L
 
Q Q
variables yρ ∈ XJρ = Xλ = Xλ = (Yρ )Jρ (ρ ∈ S) , where the
λ∈Jρ λ∈Lρ Jp
Q
last equality is in the context of Yρ = Xλ and Jρ ⊂ Lρ .
λ∈Lρ
Q 
Among the functions in C (XJ ; C) = C XJρ ; C are the “elemen-
ρ∈S
tary functions”
O

(20) fJ = f Jρ ρ ∈ S , fJρ ∈ C (XJρ ; C) ,
ρ∈S

which form a total set of continuous functions on X J (GT, X, §4, No. 3,

Th. 4); such a function acts on a point

Y
x = (xλ )λ∈J = (xλ )λ∈Jρ ρ∈S ∈ XJ = X Jρ
ρ∈S

as follows: Y

fJ (x) = fJρ (xλ )λ∈Jρ .
ρ∈S

We now construct a function gS ∈ C (YS ; C) such that (15) holds, that

is, fJ ◦ prJ = gS , where prJ is the projection YS → XJ described there,
and fJ is the function defined in (20).
INT III.x139 measures §4

For each ρ ∈ S define a function gρ ∈ C (Yρ ; C) by the formula

(21) gρ = fJρ ◦ prJρ ,

where prJρ is the projection Yρ → XJρ in the context of

Y Y Y Y
Yρ = Xλ = Xλ × X λ = X Jρ × Xλ .
λ∈Lρ λ∈Jρ λ∈Lρ -- Jρ λ∈Lρ -- Jρ

Q
Define gS ∈ C (YS ; C) = C ( Yρ ; C) by the formula
ρ∈S

O
(22) gS = gρ .
ρ∈S

We now show that fJ ◦ prJ = gS , where the projection prJ Q

: YS → XJ
has been described in connection with (15). Let y ∈ Y S = Yρ , say
ρ∈S
y = (yρ )ρ∈S , where, for ρ ∈ S , yρ ∈SYρ and yρ = (xλ )λ∈Lρ with xλ ∈ Xλ
for λ ∈ Lρ . Then (recall that J = Jρ with Jρ ⊂ Lρ )
ρ∈S

Y
(23) prJ y = (xλ )λ∈J = (xλ )λ∈Jρ ρ∈S
∈ X Jρ = X J ,
ρ∈S

therefore
O 
 Y

(24) fJ (prJ y) = f Jρ (xλ )λ∈Jρ ρ∈S
= fJρ (xλ )λ∈Jρ ;
ρ∈S ρ∈S

but, for ρ ∈ S , by (21) one has

gρ (yρ ) = fJρ (prJρ yρ ) = fJρ (xλ )λ∈Jρ ,

whence, by (24),
Y O 

fJ (prJ y) = gρ (yρ ) = gρ (yρ )ρ∈S = gS (y) ,
ρ∈S ρ∈S

and so

(25) gS = fJ ◦ prJ ∈ C (YS ; C)

§4 products of measures INT III.x140

(that is, (15) holds), in other words, taking into account (20) and (22),

O O 
(26) gρ = fJρ ◦ prJ .
ρ∈S ρ∈S

N Q
We now consider νS = νρ ∈ M (
Yρ ; C) = M (YS ; C) (see (6))
ρ∈S Q ρ∈S
and apply it to the function gS ∈ C (YS ; C) = C ( Yρ ; C) :
ρ∈S
 O  O 

νS (gS ) = νρ gρ by (6) and (22)
ρ∈S ρ∈S
Y

= νρ (gρ ) by (12) of No. 4
ρ∈S
Y


= νρ fJρ ◦ prJρ by (21)
ρ∈S
Yh
Y i

= µ Jρ f Jρ · µλ (1) by (3)
ρ∈S λ∈Lρ -- Jρ
hY
i h Y  Y i
= µ Jρ f Jρ · µλ (1)
ρ∈S ρ∈S λ∈Lρ -- Jρ
h O  O i h Y  Y i
= µ Jρ f Jρ · µλ (1)
ρ∈S ρ∈S ρ∈S λ∈Lρ -- Jρ
Y
= µJ (fJ ) µλ (1) ,
λ∈M -- J

N N N

where µ Jρ = µλ = µJ by No. 4, Prop. 7, and where M =
ρ∈S ρ∈S λ∈Jρ
S
Lρ , so that by disjointness one has
ρ∈S

[ [ [
(27) M --- J = Lρ --- Jρ = (Lρ --- Jρ ) ,
ρ∈S ρ∈S ρ∈S

and the previous display may be abbreviated

Y
(28) νS (gS ) = µJ (fJ ) µλ (1) .
λ∈M -- J

Noting that (15) has been verified, and substituting (28) into (18), we then
INT III.x141 measures §4

have (for the “elementary functions” f J )

Y
ν(fJ ◦ prJ ) = νS (gS ) νρ (1)
ρ∈R -- S
h Y i Y
(29) = µ J (f J ) µ λ (1) · νρ (1)
λ∈M -- J ρ∈R -- S
h Y i h Y  Y i
= µJ (fJ ) µλ (1) · µλ (1) ,
λ∈M -- J ρ∈R -- S λ∈Lρ

the second and third equalities by (28) and (4), respectively; the indices λ
for which µλ (1) occurs as a factor in (29) are those in
 [  [ [   [ 
(M --- J) ∪ Lρ = Lρ --- Jρ ∪ Lρ (by (27))
ρ∈R -- S ρ∈S ρ∈S ρ∈R -- S
[ [  [
= Lρ ∪ Lρ --- Jρ
ρ∈S ρ∈R -- S ρ∈R
[ 
= Lρ --- J
ρ∈R

= L --- J ,
whence (29) becomes
Y
(30) ν(fJ ◦ prJ ) = µJ (fJ ) µλ (1) = µ(fJ ◦ prJ ) (by (11)).
λ∈L -- J
We have thus shown that for every finite J ⊂ L , and, behind the scenes, the
finite S ⊂ R chosen preparatory to (14), one has
ν(fJ ◦ prJ ) = µ(fJ ◦ prJ )
for all fJ running over a suitable set of “elementary functions” total in
C (XJ ; C) , hence
(*) ν(f ◦ prJ ) = µ(f ◦ prJ )
for all finite linear combinations of such f J ’s. If now f ∈ C (XJ ; C) is
arbitrary, it is the uniform limit of a sequence f n of such linear combinations.
whence
kfn ◦ prJ − f ◦ prJ k = k(fn − f ) ◦ prJ k 6 kfn − f k → 0 ,
consequently (*) holds for all f ∈ C (X J ; C) by the continuity of ν and µ .
Such functions


f ◦ prJ J ⊂ L finite, f ∈ C (XJ ; C)
form a dense linear subspace of C (X; C) by Lemma 3 of No. 5, therefore
ν = µ on C (X; C) by continuity.
Ouf!
 CHAPTER IV

p
Extension of a measure. L spaces

§1. UPPER INTEGRAL OF A POSITIVE FUNCTION

IV.1, `. −6, −5.

“Lemma. — Every function f ∈ I+ is the upper envelope of the set
(directed for the relation 6 ) of all functions g ∈ K + such that g 6 f .”
Let G be the set of all g ∈ K+ such that g 6 f (for example, 0 ∈ G )
and let g0 = sup G be its upper envelope, g0 (x) = sup g(x) for all x ∈ X .
g∈G
Since g, g 0 ∈ G ⇒ sup(g, g 0 ) ∈ G , G is directed upward for 6 .
It is obvious that g0 6 f ; we are to show that g0 = f . Fix x ∈ X
and let us show that g0 (x) > f (x) . Given any a ∈ R with f (x) > a it
will suffice to find g ∈ G such that g(x) > a , for then g 0 (x) > g(x) > a
and letting a → f (x) yields g0 (x) > f (x) . If a 6 0 then g = 0 satisfies
g(x) = 0 > a as desired; so we can suppose that a > 0 .
Since 0 < a < f (x) it follows from lower semi-continuity that there
exists a neighborhood V of x such that f > a on V (GT, IV, §6, No. 2,
Def. 1) and we can suppose that V is compact. It follows from GT, IX,
§1, No. 5, Th. 2 that there exists a continuous function u : X → [0, 1]
such that u(x) = 0 and u = 1 on { V ; then g = a · (1 − u) satisfies
0 6 g(y) 6 a for all y ∈ X , g(x) = a , and g = 0 on { V . Since V is
closed, Supp g ⊂ V , and since V is compact, g ∈ K + . Finally g ∈ G ; for,
if y ∈ V then g(y) 6 a < f (y) (recall that f > a on V ) and if y ∈ { V
then g(y) = 0 6 f (y) . Thus the requirements g ∈ G and g(x) > a are
fulfilled.
Special case: Suppose f = ϕU is the characteristic function of an open
set U of X (lower semi-continuous by GT, IV, §6, No. 2, Cor. of Prop. 1).
In the foregoing proof, 0 < a < f (x) implies x ∈ U , so one can sup-
pose in addition that V ⊂ U ; then the constructed function g belongs to
K+ (X, U; C) , thus

ϕU = sup{g ∈ K+ : g 6 ϕU , Supp g ⊂ U } .
INT IV.x2 extension of a measure. lp spaces §1

IV.2, `. −13. P
“ . . . which again proves the relation µ*(f ) = α(x)f (x) .”
x∈X
Recall that µ ∈ M+ (X) , X a discrete space, K+ is equal to the
linear span of the characteristic functions ϕ {x} (x ∈ X) with finite real
coefficients > 0 , and α(x) = µ(ϕ{x} ) (x ∈ X) ; thus every g ∈ K+ has the
form X
g= g(x)ϕ{x}
x∈M

with M a finite subset of X , and one has

X
µ(g) = α(x)g(x) .
x∈M

For a fixed function f : X → R with f > 0 , we wish to show that

X
(*) µ*(f ) = α(x)f (x) ,
x∈X

where, by convention, α(x)f (x) = 0 when α(x) = 0 and f (x) = +∞ . Let

D = {x ∈ X : α(x) = 0 and f (x) = +∞ } ;

thus x ∈ D implies α(x)f (x) = 0 , and the right side S of (*) may be
written X
S= α(x)f (x) .
x∈X -- D

We are to show that µ*(f ) = S . By definition,

µ*(f ) = sup µ(g) .

g∈K+ , g6f

Let
A = {x ∈ X : f (x) = +∞ } .

case 1: There exists an x0 ∈ A such that α(x0 ) > 0 .

Then for every positive integer n one has 0 < n < +∞ = f (x 0 ) , so
that the function gn = n · ϕ{x0 } satisfies gn ∈ K+ , 0 6 gn 6 f , and the
sequence µ(gn ) = n · α(xn ) is unbounded, thus µ(gn ) 6 µ*(f ) shows that
µ*(f ) = +∞ . On the other hand, α(x0 )f (x0 ) = +∞ is a term on the right
side of (*), therefore S = +∞ and the equation (*) holds.
§1 upper integral of a positive function INT IV.x3

case 2: α(x) = 0 for all x ∈ A .

Then D = A , so that
X
S= α(x)f (x) ,
x∈X -- A

where, for every x ∈ X --- A , both α(x) and f (x) are finite.
Let usPfirst show that S 6 µ*(f ) . Let M be any finite subset of X --- A ;
then g = f (x)ϕ{x} belongs to K+ and g 6 f , therefore µ(g) 6 µ*(f ) ,
x∈M
that is, X
α(x)f (x) 6 µ*(f ) ;
x∈M

the validity of this inequality for every finite M ⊂ X --- A shows that
S 6 µ*(f ) .
To prove the reverse inequality, let g ∈ K + be such that g 6 f and let
us show that µ(g) 6 S . Let M be the support of g , so that
X X
g= g(x)ϕ{x} and µ(g) = α(x)g(x) .
x∈M x∈M

If x ∈ M ∩ A then f (x) = +∞ and α(x) = 0 by assumption, so that

α(x)f (x) = 0 by convention, consequently (recalling that g 6 f )
X X X
µ(g) = α(x)g(x) 6 α(x)f (x) = α(x)f (x) 6 S
x∈M x∈M x∈M -- A

(the last inequality because M --- A ⊂ X --- A ⊂ X --- D ).

IV.3, `. 11, 12.
“ . . . since H is directed, so is Φ , and f = sup ϕ .”
ϕ∈Φ
If g, g 0 ∈ H with g 6 g 0 , clearly Φg ⊂ Φg0 . Suppose ϕ1 , ϕ2 ∈ Φ , say
ϕ1 ∈ Φg1 , ϕ2 ∈ Φg2 . Choose g ∈ H so that g1 6 g and g2 6 g . Then
ϕk ∈ Φgk ⊂ Φg for k = 1, 2 , therefore sup(ϕ1 , ϕ2 ) ∈ Φg ⊂ Φ , thus Φ is
directed upward for 6 . The assertion that

sup g = sup ϕ ,
g∈H ϕ∈Φ

that is (in view of the Lemma)

 
sup sup ϕ = Ssup ϕ,
g∈H ϕ∈Φg ϕ∈ Φg
g∈H
INT IV.x4 extension of a measure. lp spaces §1

is an application of the following result (a straightforward consequence of

the definition of supremum):
Let E be an ordered (i.e., ‘partially ordered’) space in which every
T
nonempty subset has a supremum (for example E = R , [0, +∞] or R ,
T any set). If (xi )i∈I is a family of elements
S of E and if (Ij )j∈J is a family
of nonempty subsets of I such that I = Ij , then
j∈J
 
sup xi = sup sup xi
i∈I j∈J i∈Ji

(‘associativity of sups’; S, III, §1, No. 9, Prop. 7).

IV.3, `. 12, 13.
“Since ψ 6 f , ψ is the upper envelope of the set of functions inf(ψ, ϕ)
as ϕ runs over Φ . . . ”
Since inf(ψ, ϕ) 6 ψ for all ϕ ∈ Φ , obviously


(*) sup inf(ψ, ϕ) 6 ψ .
ϕ∈Φ

Given any x ∈ X , the problem is to show that



(†) sup min ψ(x), ϕ(x) = ψ(x) .
ϕ∈Φ

case 1: ψ(x) = f (x) .

Then min ψ(x), ϕ(x) = min f (x), ϕ(x) = ϕ(x) for all ϕ ∈ Φ , and
one knows that sup ϕ(x) = f (x) , thus (†) reduces to f (x) = ψ(x) .
ϕ∈Φ
case 2: ψ(x) < f (x) .
Since sup ϕ = f , there exists ϕ0 ∈ Φ such that ψ(x) < ϕ0 (x) 6 f (x) .
ϕ∈Φ


Then min ψ(x), ϕ0 (x) = ψ(x) , consequently



sup min ψ(x), ϕ(x) > min ψ(x), ϕ0 (x) = ψ(x) ,
ϕ∈Φ

whence equality by (*).

IV.4, `. −7, −6.
“Moreover, µ*(X) = kµk , as is shown by formula (22) of Ch. III, §1,
No. 8.”
The cited formula reads kµk = sup |µ|(f ) for any measure µ ;
06f 61, f ∈K (X)
when µ > 0 this simplifies to
kµk = sup µ(f ) = µ*(1) = µ*(ϕX ) = µ*(X)
06f 61, f ∈K (X)

by the definitions.
§2 negligible functions and sets INT IV.x5

IV.11, `. 5, 6.
“These propositions are the translations of Props. 10 and 13 and of
Th. 3 of No. 3 for characteristic functions of sets.”
S P
Let A = An . Then ϕA 6 ϕAn , hence
n n

X  X
µ*(ϕA ) 6 µ* ϕ An 6 µ*(ϕAn )
n n

by Props. 10 and 13, whence Prop. 18. If, moreover, (A n ) is increasing, then
(ϕAn ) is increasing and ϕA = sup ϕAn , therefore µ*(ϕA ) = sup µ*(ϕAn )
n n
by Th. 3.

§2. NEGLIGIBLE FUNCTIONS AND SETS

IV.12, `. −7, −6.

“For, by Prop. 3, in order that an open set G be negligible, it is neces-
sary and sufficient that G ∩ S = ∅ , that is, G ⊂ { S .”
For G open, write f = ϕG , which is lower semi-continuous and > 0 .
Then
|µ|*(f ) = 0 ⇔ f = 0 on S (Prop. 3)
⇔ S ⊂ {G
⇔ G ⊂ {S;
and, since { S is open, it is eligible to play the role of G .
IV.13, `. 5, 6.
“ . . . therefore ϕN is negligible (No. 1, Props. 1 and 2).”
Each nf is negligible (Prop. 1), hence sup nf is negligible (Prop. 2),
n
hence ϕN is negligible (Prop. 1), thus N is negligible. If P x is the
property hh f (x) = 0 ii , then N = {x ∈ X : not P x } ; thus the negligi-
bility of N means that P x almost everywhere, that is, f (x) = 0 almost
everywhere.
IV.13, `. 8, 9.
“ . . . therefore f is negligible (No. 1, Props. 1 and 2).”
Each nϕN is negligible (Prop. 1), hence sup nϕN is negligible (Prop. 2),
n
hence f is negligible (Prop. 1).
INT IV.x6 extension of a measure. lp spaces §2

IV.14, `. 10, 11. Q

“Corollary. — If ϕ is a mapping of Fn into a set G , then the


n
mappings ϕ (fn ) and ϕ (gn ) of X into G are equivalent.”

Q

If x ∈ X then fn (x) ∈ Fn . The mapping ϕ (fn ) : X → G of the
n Q
Corollary is the composite function X → Fn → G defined by
n



x 7→ fn (x) →7 ϕ fn (x) ;


let us abbreviate it u = ϕ (fn ) , thus


u(x) = ϕ fn (x) (x ∈ X) .


Similarly, let v = ϕ (gn ) be the function


v(x) = ϕ gn (x) (x ∈ X) .

Let H be a negligible subset of X as described
in Prop.

8. If x ∈ / H
then fn (x) = gn (x) for all n
, therefore
 fn (x) = gn (x) as elements of
Q

Fn , consequently ϕ fn (x) = ϕ gn (x) , that is, u(x) = v(x) ; thus
n

u = v almost

everywhere, that is, u and v are equivalent, i.e., ϕ (f n ) and
ϕ (gn ) are equivalent.
The assumption on the fn , gn is that fen
= gen for all n
; the conclusion
of the corollary is that the classes of ϕ (fn ) and ϕ (gn ) are equal. The




class of ϕ (fn ) is denoted ϕ (fen ) ; that is, ϕ (fen ) is defined to be the


class ϕ^ (fn ) . Thus, the message of the corollary is that if fen = gen for



all n , then ϕ (fen ) = ϕ (e gn ) .

Another way of viewing the conclusion is that the class of ϕ (fn ) is
unchanged if each fn is replaced by a function equivalent to it.
IV.14, `. −7 to −5.
“ . . . for, this is equivalent to saying that there exists a negligible set H
in X such that, for every x ∈ / H , ρn (f (x), g(x)) = 0 for all n , that is,
f (x) = g(x) .”
To put it another way, from
\

{x : f (x) = g(x) } = {x : ρn f (x), g(x) = 0 } ,
n
§3 lp spaces INT IV.x7
S

we have {x : f (x) 6= g(x) } = {x : ρn f (x), g(x) =6 0 } , and the set on
n
the left is negligible if and only if each term of the union is negligible.
IV.15, `. 5, 6.
“ . . . for it to be negligible, it is necessary and sufficient that it not
intersect the support of µ (No. 2, Prop. 5).”
Write G = {x : f (x) 6= g(x) } , an open subset of X since since F is
Hausdorff (GT, I, §8, No. 1, Prop. 2). Then
f (x) = g(x) a.e. ⇔ G negligible
⇔ G ⊂ { Supp(µ) (No. 2, Prop. 5)
⇔ Supp(µ) ⊂ { G = {x : f (x) = g(x) }
⇔ f (x) = g(x) for all x ∈ Supp(µ) .
IV.17, `. −3 to −1.
“For f to be negligible, it is necessary and sufficient that |f | be negli-
gible (or that both f + and f − be negligible).”
+


If f (x) is defined, one defines

|f |(x) = |f (x)|

, f (x) = sup f (x), 0 ,
− + −
and f (x) = sup − f (x), 0 = − inf f (x), 0 . Thus f , |f | , f , f
have the same domain of definition; f + (x) and f − (x) can’t both be equal
to +∞ , hence f = f + − f − ; and |f | = f + + f − . The assertions about
negligibility are then clear.

§3. Lp SPACES

IV.19, `. 20–24.
“We extend Def. 1 to the case of numerical functions, finite or not,
defined on X , by again setting
Z 1/p
∗ p
Np (f ) = |f | d|µ|

for such a function f . One sees immediately that the relations (3) and (4)
also hold for these functions when f + g is defined on X and α 6= 0 .”
It is convenient to simultaneously treat the proof of Prop. 2.
If f, g : X → R and |f | 6 |g| almost everywhere in X , then |f | p 6 |g|p
almost everywhere, therefore |µ|*(|f | p ) 6 |µ|*(|g|p ) by §2, No. 3, Prop. 6

1/p
1/p
and §1, No. 3, Prop. 10, hence |µ|*(|f |p ) 6 |µ|*(|g|p ) ; thus
(i) |f | 6 |g| a.e. ⇒ Np (f ) 6 Np (g) .
INT IV.x8 extension of a measure. lp spaces §3

Incidentally, Np (αf ) = |α| · Np (f ) even if α = 0 , with the convention

0 · (+∞) = 0 .
Note that if f : X → F ( F a Banach space, real or complex) then, by
definition,
(ii) Np (f ) = Np (|f |) .
It follows from (i) and (ii) that if f , g : X → F and |f | 6 |g| a.e., then
(iii) Np (f ) 6 Np (g) .
If f, g : X → R and f + g is defined, then |f + g| 6 |f | + |g| , hence
(by (i) and No. 1, Prop. 1)
Np (|f + g|) 6 Np (|f | + |g|) 6 Np (|f |) + Np (|g|) ,
thus
(iv) f, g : X → R , f + g defined ⇒ Np (f + g) 6 Np (f ) + Np (g) .
If f , g : X → F then |f + g| 6 |f | + |g| by the triangle inequality in F ,
therefore (by (i) and (iv))
Np (|f + g|) 6 Np (|f | + |g|) 6 Np (|f |) + Np (|g|) ,
thus
(v) f, g : X → F ⇒ Np (f + g) 6 Np (f ) + Np (g) .

IV.19, `. −1 to IV.20, `. 1.
“ . . . the definition of Np (f ) , and Th. 3 of §1, No. 3, show that N p (f ) =
sup Np (gn ) .”
n


By the definition of f , gn ↑ f , hence (gn )p ↑ f p , hence |µ|* (gn )p ↑

1/p
1/p
|µ|*(f p ) by Th.3 of §1, No. 3, hence |µ|* (gn )p ↑ |µ|*(f p ) as
claimed.
IV.20, `. 6.
“The proposition follows at once from Th. 1 of §2, No. 3.”

f , g equivalent ⇔ f − g = 0 a.e. (§2, No. 4)

p
⇔ |f − g| = 0 a.e. for some p > 1
⇔ |f − g|p = 0 a.e. for all p > 1
⇔ |µ|*(|f − g|p ) = 0 for some (all) p > 1 (§2, Th. 1)

1/p
⇔ |µ|*(|f − g|p ) = 0 for some (all) p > 1 ,
§3 lp spaces INT IV.x9

that is, f and g are equivalent if and only if N p (f − g) = 0 for some (hence
for all) p > 1 .
IV.20, `. 16–19.
“One can then define Np (f ) for a function with values in F (resp.
in R ) defined almost everywhere in X , by setting N p (f ) = Np (e f ) ; it is
then clear that the relations (3) and (4) again hold (assuming α 6= 0 and
f + g defined almost everywhere, in the case of numerical functions, finite
or not).”
Recall (§2, No. 5) that if f is any function defined almost everywhere
in X , with values in a set G (not necessarily vectorial or numerical), the
class fe is defined as follows: let f 0 be any extension of f to all of X , with
values in G , and let fe be the set of all functions g : X → G such that
g = f 0 almost everywhere in X , that is, fe is the equivalence class of f 0 in
F (X; G) , for the relation of equality almost everywhere; if f 00 is another
extension of f to X , then f 00 is equivalent to f 0 , whence it is clear that
fe is independent of the particular extension f 0 of f to X .
When G is equal to the Banach space F or to R , and f 0 is any
extension of f to all of X , then Np (f 0 ) is defined (Def. 1 and the remarks
following Prop. 2) and is independent of the particular extension f 0 (§2,
No. 3, Prop. 6), thus the definition N p (f) e = Np (fe0 ) ( = Np (f 0 ) ) depends
only on f . The author defines Np (f ) to be equal to Np (fe) ; but what it
all comes down to is that Np (f ) may unambiguously be defined to be equal
to Np (g) , where g is any function equivalent to any extension of f to X .
If f , g are functions defined almost everywhere in X with values in F ,
then f + g is defined almost everywhere and if f 0 and g0 are extension of f
and g to X , then f 0 + g0 extends f + g and the relation (4) for f 0 and g0
says precisely Np (f + g) 6 Np (f ) + Np (g) . Similarly for (3).
If f, g are defined almost everywhere in X and if f +g is defined almost
everywhere, then there exists a negligible set N such that f , g and f+ g
are defined 0
 on { N ; if 0f (resp. g 0 ) is a function on X that extends f  { N
(resp. g  { N ), then f , g 0 and f 0 + g 0 clearly represent the equivalence
classes fe, ge and f] + g , respectively, consequently the relation (4) for f 0
0
and g says precisely Np (f + g) 6 Np (f ) + Np (g) . Similarly for (3).
IV.21, `. 4–6.
“This terminology extends at once to the case that the functions f n and
the function f are only defined almost everywhere (or have values in R ,
and are defined and finite almost everywhere).”
When the functions fn , f have values in the Banach space F , there
exists a negligible set N on whose complement the f n and f are defined;
in particular (for a pair of such functions) linear operations on the classes of
INT IV.x10 extension of a measure. l p spaces §3

such functions are readily defined, and the notation N p (fn − f ) is available
(see the preceding Note).
When the functions take their values in R and are, moreover, finite
almost everywhere, then there exists a negligible set N on whose comple-
ment the functions are defined and finite-valued; one then proceeds as in
the preceding paragraph. Note in this connection that if f is a function de-
fined almost everywhere in X and taking values in R , and if N p (f ) < +∞
(where Np (f ) is defined as in the preceding Note), then f is finite-valued
almost everywhere (§2, No. 3, Prop. 7).
IV.21, `. 7–9.
“ . . . the closure of 0 in this space is the linear subspace N F of negligible
mappings of X into F (No. 2, Prop. 3).”
The closure in question is the set {f ∈ F Fp : Np (f ) = 0 } (TVS, II,
§1, No. 2, Prop. 2), and Np (f ) = 0 if and only if f is negligible (§2, No. 1,
Def. 1). It follows that NF is contained in each of the spaces LFp to be
defined in No. 4 (Def. 2).
IV.21, `. −13, −12.
“From this, it follows that Np (f − f0 ) 6 ε Np (h) , whence the proposi-
tion.”
Since, for every ε > 0 , there exists an M ∈ B such that

(*) |f (x) − f0 (x)| 6 ε for all f ∈ M and x ∈ X ,

and since every f ∈ M is 0 on { K , it is clear that f0 = 0 on { K , therefore

(*) yields

|f (x) − f0 (x)| 6 εh(x) for all f ∈ M and x ∈ X ,

whence

(**) Np (f − f0 ) 6 ε Np (h) for all f ∈ M ;

since Np (h) < +∞ one has Np (f − f0 ) < +∞ , therefore f − f0 ∈ FFp and

so f0 = f − (f − f0 ) ∈ FFp . Since, given any ε > 0 there exists an M ∈ B
satisfying (**), the convergence in mean of order p of B to f 0 is proved.
Incidentally, the role of h can be played equally well by ϕ K (§1, No. 4, Cor.
of Prop. 16).
IV.24, `. 5–7.
“Proposition 7. — For a function f to belong to L Fp , it is necessary
and sufficient that, for every ε > 0 , there exist a continuous function g
with compact support such that Np (f − g) 6 ε .”
§3 lp spaces INT IV.x11

If f has the indicated property, then for any such g one clearly has
f − g ∈ FFp , hence f = (f − g) + g ∈ FFp ; thus f belongs to the closure of
K (X; F) in FFp , that is, f ∈ LFp . The converse is immediate from Def. 2.
IV.24, `. 15, 16.
“For, f is finite almost everywhere and N p (f − g) 6 Np (h − g) 6 ε ;
Prop. 7 therefore shows that f is p -th power integrable.”
Since the functions g, h of the statement are finite almost everywhere,
so is f ; redefining f on a negligible set, we can suppose that it is everywhere
defined and finite. Then, given ε > 0 , choose g and h as in the statement
of the proposition; replacing them by equivalent functions, we can suppose
that g, h ∈ L p . Then all three functions are everywhere defined, finite, and
satisfy g 6 f 6 h almost everywhere in X ; and since 0 6 f − g 6 h − g
almost everywhere, Np (f − g) 6 Np (h − g) 6 ε . Choose g0 ∈ K (X; R)
with Np (g − g0 ) 6 ε (Def. 2); then


Np (f − g0 ) = Np (f − g) + (g − g0 ) 6 Np (f − g) + Np (g − g0 ) 6 ε + ε ,

therefore f ∈ L p by Prop. 7, hence the original function f (before redefi-

nition) is p-th power integrable.
IV.24, `. −6, −5.
“ . . . parts 2◦ and 3◦ then follow from Prop. 6 of No. 3 and the fact that
LF is closed in FFp .”
p

By 1◦ and the cited Prop. 6, the series with general term f nk+1 (x) −
fnk (x) is absolutely convergent for almost every x in X , and the function
g : X → F defined by
X
g(x) = [fnk+1 (x) − fnk (x)] = lim fnk (x) − fn1 (x)
k
k

at the points x where the series converges, and by g(x) = 0 at the remaining
points x , belongs to FFp , and, as k → ∞ ,

k−1
X
[fnj+1 − fnj ] = fnk − fn1 → g in FFp ,
j=1

that is, in mean of order p ; g ∈ LFp because the fnk − fn1 belong to LFp
and LFp is closed in FFp . Thus, setting f = g + fn1 , one has f ∈ LFp ,
fnk (x) → f (x) for almost every x in X , and f nk → fn1 + g = f in LFp ,
that is, in mean of order p . Since (fn ) is Cauchy in LFp and fnk → f
in LFp , it follows that fn → f in LFp (an easy application of the triangle
INT IV.x12 extension of a measure. l p spaces §3

inequality; for the abstract principle, see GT, II, §3, No. 2, Cor. 3 of Prop. 5).
Whence 3◦ .
IV.24, `. −2, −1.
“ . . . there exists a lower semi-continuous function g > h + |f n1 | such
that Np (g) < +∞ , which completes the proof.”
P
By definition, h(x) = |fnk+1 (x) − fnk (x)| for all x ∈ X , thus h =
P k
|fnk+1 − fnk | is the sum of the positive numerical functions |f nk+1 − fnk | .
k
By Th. 1 of No. 2,
X X
Np (h) 6 Np (|fnk+1 − fnk |) = Np (fnk+1 − fnk ) < +∞ ,
k k

thus h ∈ F p . For every positive integer k and every x ∈ X ,

k−1
X

fnj+1 (x) − fnj (x) = fnk (x) − fn1 (x) ,
j=1

P
k−1

thus fnk (x) = fn1 (x) + fnj+1 (x) − fnj (x) , therefore
j=1

k−1
X  
|fnk (x)| 6 |fn1 (x)| + fnj+1 (x) − fnj (x) 6 |fn1 (x)| + h(x) ,
j=1

briefly |fnk | 6 |fn1 | + h .

From |fn1 | ∈ L p and h ∈ F p we have |fn1 | + h ∈ F p , thus

Np (|fn1 | + h) < +∞ ,


whence |µ|* (|fn1 | + h)p < +∞ . By the definition of |µ|* (§1, No. 3,
Def. 3) there exists a lower semi-continuous function u > 0 such that

(|fn1 | + h)p 6 u and |µ|*(u) < +∞ .

Since t 7→ t1/p is continuous and increasing (GT, IV, §3, No. 3, with the
convention (+∞)1/p = +∞ ) it is easy to see that u1/p is also lower semi-
continuous (more generally, see GT, IV, §6, Exer. 4). Writing g = u 1/p , we
have |fn1 | + h 6 g and

1/p
1/p
Np (g) = |µ|*(g p ) = |µ|*(u) < +∞ ,
§3 lp spaces INT IV.x13

thus g ∈ F p . Finally, |fnk | 6 |fn1 | + h 6 g , thus |fnk (x)| 6 g(x) for

all x ∈ X . Incidentally, since g ∈ F p is lower semi-continuous, in fact
g ∈ L p (see below, §4, No. 4, Prop. 5).
IV.25, `. −3.
“ . . . the theorem is proved.”
Recall that the functions in LFp (in particular f ) are defined everywhere
in X ; and they are characterized as the functions in F Fp that are in the
closure of KF for the Np -topology. Since |(u ◦ f )(x)| 6 kuk · |f (x)| for all
x ∈ X , that is, |u ◦ f | 6 kuk · |f | , one has N p (u ◦ f ) 6 kuk · Np (f ) < +∞ ,
thus u ◦ f ∈ FGp ; and the argument in the text shows that u ◦ f is in the
closure of KG , whence u ◦ f ∈ LGp .
Theorem 4 extends to functions defined almost everywhere, as follows.
If f is p-th power integrable but is defined only almost everywhere in X ,
there exists an f 0 ∈ LFp such that f = f 0 almost everywhere (remarks
following No. 4, Def. 2). Then u ◦ f = u ◦ f 0 almost everywhere, where
u ◦ f 0 ∈ LGp by the theorem, hence u ◦ f is p-th power integrable.
IV.25, `. −2, −1.
“Corollary 1.”
In Theorem 4, take G to be the field of scalars of F , and u = a 0 ∈ F0 .
IV.26, `. 4, 5.
“This follows from the fact that the mapping t 7→ at of R into F is
continuous.”
The field of scalars K of F can be either R or C . We can suppose
that k = 1 . Thus, assuming a ∈ F and f ∈ L p , one defines f = af by
the formula f (x) = f (x)a . Define u : K → F by u(t) = ta ; then u is a
continuous linear mapping and


(u ◦ f )(x) = u f (x) = f (x)a = f (x) ,
whence f ∈ LFp by Theorem 4 (with the substitutions of K for F , and F
for G ).
IV.26, `. 10.
“This follows at once from Cors. 1 and 2 of Th. 4.”
Equip F with its unique compatible Hausdorff topology (TVS, I, §2,
No. 3, Th. 2) and let F0 be the dual space; since every linear form on F
is continuous (loc. cit., Cor. 2 of Th. 2), the topological dual coincides with
the algebraic dual. Let (e0k )16k6n be the basis of F0 dual to (ek )16k6n ;
then
hf (x), e0k i = fk (x) (x ∈ X) ,
thus fk = hf , e0k i . If f ∈ LFp then fk ∈ L p for all k by Cor. 1 of Th. 4;
the converse is immediate from Cor. 2.
INT IV.x14 extension of a measure. l p spaces §3

IV.27, `. −8 to −6.
“ . . . the image of the section filter F of H under this mapping is there-
fore a base of a Cauchy filter on R .”
The set of all ke f kp ( fe ∈ H ) has a finite supremum λ and is directed
upward by the relation 6 on H ; that is, the mapping Φ : H → R defined
by Φ(e f ) = ke f kp is increasing and has a finite supremum λ . The sets
g) = {e
H(e f ∈H: e
f >g
e} (e
g ∈ H)
are a base for the section filter F of the directed set H . The sets


Φ H(eg) = {ke f kp : e
f ∈ H, e
f >ge } (e g ∈ H)
form a base of a filter G on R (GT, I, §6, No. 6).
g k > λ − ε ; then
Given any ε > 0 , choose ge ∈ H so that ke
e
f ∈ H, e
f >g
e ⇒ gk 6 ke
λ − ε 6 ke fk 6 λ ,


thus Φ H(eg ) ⊂ [λ − ε, λ] ⊂ [λ − ε, λ + ε] . This shows that the filter G is
convergent (to λ ), hence is Cauchy.
IV.28, `. 10–12.
“ . . . every function f > 0 in L p is the limit (for convergence in mean
of order p ) of a sequence of continuous functions > 0 with compact support,
by the continuity of the mapping g 7→ |g| on L p (Prop. 11).”
By the definition of L p , there exists a sequence (fn ) in K such that
fn → f (in mean of order p ). Since |fn | → |f | = f by the cited Prop. 11,
one can suppose without loss of generality that f n > 0 for all n . Similarly,
since g − f ∈ L p and g − f > 0 , there exists a sequence (h n ) in K with
hn > 0 and hn → g − f . Then
fn + hn → f + (g − f ) = g ;
setting gn = fn + hn , we have 0 6 fn 6 gn , where fn , gn ∈ K+ and
fn → f , gn → g (in mean of order p ). By the special case considered
earlier,

p
p
p
Np (gn − fn ) 6 Np (gn ) − Np (fn ) for all n ,
and passage to the limit yields (9).
IV.31, `. 11, 12.
“ . . . limF Np (f − fα ) exists and is equal to the common limit 0 of all
of the sequences (Np (f − fαn )) .”
Explicitly: Given ε > 0 , we seek an index n such that N p (f − fα ) < ε
for all α ∈ An . Assume to the contrary that for every n there exists an
αn ∈ An such that Np (f − fαn ) > ε . This contradicts the fact, already
shown, that lim Np (f − fαn ) = 0 .
n→∞
§4 integrable functions and sets INT IV.x15

§4. INTEGRABLE FUNCTIONS AND SETS

IV.33, `. 13.
P
“ . . . we have |µ|*(|f |) = |α(x)| · |f (x)| < +∞ (§1, No. 3, Example)”
x∈X

In the case that F is replaced by R and f is a numerical function, this

implies that α(x) = 0 when f (x) = ±∞ . In any case, α(x) = µ(ϕ {x} ) is
finite since ϕ{x} is a continuous function with compact support ( X being
discrete).
IV.33, `. 16.
“ . . . at the points x ∈ M where |f | is finite”
In the vector-valued case, |f (x)| is finite for every x ∈ X . Apparently
the comment is intended for the case of numerical functions f ; similarly for
the subsequent remark “by the conventions that have been made”.
IV.33, `. −10 .
“ . . . whence the second assertion.”
P
The argument shows that as g → f in mean, µ(g) → α(x)f (x) ;
R R x∈X
but µ(g) → f dµ by the definition of f dµ .
IV.34, `. 14.
“ . . . whence the proposition.”
Since f ∈ L 1 there exists a sequence fn ∈ K such that fn → f
in mean, and since |fn | → |f | = f in mean (§3, No. 5, Prop. 11) we can
suppose that fn > 0 . From N1 (fn − f ) → 0 one infers N1 (fn ) → N1 (f )
( N1 is a semi-norm on L 1 ); but

N1 (fn ) = |µ|*(|fn |) = |µ|(fn ) → |µ|(f )

R∗
by the definition of |µ|(f ) , thus |µ|(f ) = lim N1 (fn ) = N1 (f ) = f d |µ| .
n

IV.34, `. 15–17.
“Corollary 1.”
The positive function f = |f | belongs to L 1 by §3, No. 5, Prop. 11,
and N1 (f ) = N1 (f ) .
INT IV.x16 extension of a measure. l p spaces §4

IV.35, `. 3, 4.
“This follows at once from the inequality (1) by passage to the limit, on
taking into account (3) and the continuity of N 1 (f ) on LF1 .”
By the definition of “ f integrable”, there exists a sequence f n ∈ KF
such that

(*) N1 (fn − f ) → 0 ,

and one defines

Z Z
(i) f dµ = lim fn dµ .
n
 
But also |fn | − |f | 6 |fn − f | , therefore N1 (|fn | − |f |) 6 N1 (fn − f ) → 0 ,
whence |f | ∈ L 1 and
Z Z
(ii) |f | d|µ| = lim |fn | d|µ| .
n

By (1),
Z  Z
 
(iii)  fn dµ 6 |fn | d|µ| = N1 (fn ) ;

since (*) implies N1 (fn ) → N1 (f ) , passage to the limit in (iii) yields

Z  Z
 
 f dµ 6 |f | d|µ| = N1 (f )

by (i) and (ii).

IV.35, `. 9–11.
“ . . . the relation (6), being valid for every f ∈ K F , extends to every
integrable function f by the principle of extension of identities”


From |(u ◦ f )(x)| = u f (x)  6 kuk · |f (x)| for all x ∈ X , we have
|u ◦ f | 6 kuk · |f | , whence N1 (u ◦ f ) 6 kuk N1 (f R) ; thus the linear mapping
f 7→ u ◦ f is continuous in mean, therefore f 7→ (u ◦ f ) dµ is a continuous
linear mapping LF1 → G . R
On the other hand, f 7→ fR dµ is
a continuous linear mapping L F1 → F,
hence the composite f 7→ u f dµ is also a continuous linear mapping
LF1 → G .
Summarizing, the mappings
Z Z 
f 7→ (u ◦ f ) dµ and f 7→ u f dµ
§4 integrable functions and sets INT IV.x17

are continuous mappings of LF1 into the Hausdorff space G ; since they agree
on the dense subspace KF (cf. Ch. III, §3, No. 2, Prop. 2 and No. 3, Cor. 2
of Prop. 7), they are identical by the “Principle of extension of identities”
(GT, I, §8, No. 1, Cor. 1 of Prop. 2).
IV.35, `. −1.

Z n
! n Z
X X
(8) ak fk dµ = ak fk dµ .
k=1 k=1

Let K ( = R or C ) be the field of scalars for the Banach space F .

Assume first that the fk ∈ L 1 are defined and finite-valued on all of X .
By linearity, we can suppose that n = 1 . Thus, given a ∈ F and
a function f ∈ L 1 defined and finite-valued on all of X , the formula
(af )(x) = f (x)a (x ∈ X) defines a function af : X → F . Let u : K → F
be the continuous linear mapping defined by u(t) = ta (t ∈ K) . Then


(u ◦ f )(x) = u f (x) = f (x)a = (af )(x) (x ∈ X) ,

hence by Th. 1, af ∈ LF1 and

Z Z 
af dµ = f dµ a .

(The notation af presumably suggests that a is a ‘constant coefficient’ for

the function f .)
When the fk are not necessarily finite-valued, let g k ∈ L 1 be finite-
valued with gk = fk almost everywhere, apply the foregoing to the g k , and
interpret (8) in the light of the comments at the end of No. 1.
IV.36, `. 7.
“For, B converges in mean to f0 (§3, No. 3, Prop. 4).”
By the cited Prop. 4, f0 ∈ FF1 and B → f0 in mean (of order 1). But
the functions f ∈ M ∈ B belong to LF1 , and LF1 is closed in FF1 (it is
by definition the closure of KF in FF1 for convergence
R in mean),
R therefore
f0 ∈ LF1 ; and since limB f = f0 in mean, R f0 dµ = limB f dµ by the
continuity (for convergence in mean) of g 7→ g dµ ( g ∈ LF1 ) .
IV.36, `. 14.
“ gn = fn + f1− ”
Caution . The functions fn are assumed to be integrable, but they
need not belong to L 1 (§3, No. 4) hence may have infinite values. For
INT IV.x18 extension of a measure. l p spaces §4

example, if f1 (x) = −∞ then f1− (x) = sup(−f1 (x), 0) = +∞ and the sum
f1 (x)+f1− (x) is not defined. (This problem did not arise in §3, No. 6, Cor. 2
of Th. 5, because the functions fn were assumed to belong to L p .)
The cure: If the fn are replaced by equivalent functions f n0 in L 1 then
the upper envelope f 0 of the fn0 is equivalent to f ; f 0 may admit +∞ as
a value, but there is no problem with g 0 = f 0 + f10 − since f10 is finite-valued.
To assure that the sequence (fn0 ) is increasing, let An be a negligible
S set on
whose complement fn0 = fn , and redefine the fn0 to be 0 on An . Finally,
n
gn0 = fn0 + f10 − > f10 + f10 − = f10 + , thus the functions gn0 belong to L 1 and
are > 0 . The proof of Prop. 4 carries through for the f n0 as indicated (see
the next note), and the validity of Prop. 4 for the f n follows at once from
the function equivalences.
IV.36, `. 15, 16.
“ . . . the proposition follows from Th. 5 of §3, No. 6.”
As observed in the preceding note, we can suppose that the f n are
finite-valued, that is, belong to L 1 . Since (recall that fn + f1− > 0 ) the
norms
Z Z Z
− −
N1 (fn + f1 ) = (fn + f1 ) d|µ| = fn d|µ| + f1− d|µ|

have a finite upper bound, it follows from the cited Th. 5 that the function
g = f + f1− is integrable, therefore f = g − f1− is integrable. Th. 5 also
yields
N1 (f + f1− ) = sup N1 (fn + f1− ) ,
n

that is, Z Z
(f + f1− ) d|µ| = sup (fn + f1− ) d|µ| ,
n

whence Z Z
f d|µ| = sup fn d|µ| ,
n
R R R
thus fn d|µ| → f d|µ| , therefore (f − fn ) d|µ| → 0 ; but f − fn > 0
and, citing Prop. 2 of No. 2, one has
Z  Z Z
 
 (f − fn ) dµ 6 |f − fn | d|µ| = (f − fn ) d|µ| → 0 ,

R R
whence fn dµ → f dµ .
§4 integrable functions and sets INT IV.x19

IV.36, `. −8.
“The theorem follows from Lebesgue’s theorem (§3, No. 7, Cor. of
Th. 6) . . .”
Here the functions fα belong to LF1 ( F a Banach space), and f : X → F;
one applies the cited Cor. with p = 1 , to conclude
R that Rf ∈ L F1 and that
fα → f in mean with respect to F , whence fα dµ → f dµ in F with
respect to F by the continuity of the integral.
IV.36, `. −5 to IV.37, `. 5.
“Corollary 1.”
Here Ω will play the role of A in Th. 2, and the role of F will be
played by the neighborhood filter of t 0 in Ω (which, by hypothesis, has a
countable base). For each t ∈ Ω define f t : X → F by the formula

ft (x) = f (x, t) (x ∈ X) .

By the hypothesis a), we have ft ∈ LF1 for every t ∈ Ω . The strategy

is to apply Th. 2 to the family (ft )t∈Ω , where Ω is filtered by F .
By the hypothesis b), for each x ∈ X we have

ft (x) = f (x, t) → f (x, t0 ) = ft0 (x) as t → t0 ;

in other words, the family (ft )t∈Ω converges pointwise in X to ft0 with
respect to the filter F .
By the hypothesis c), for each t ∈ U we have

|ft (x)| = |f (x, t)| 6 g(x) for all x ∈ X .

Since the neighborhoods V of t0 with V ⊂ U form a base for F , it follows

from Th. 2 that ft0 ∈ LF1 (which we already know from a)) and that
Z Z
ft0 dµ = lim ft dµ ,
t, F

in other words
Z Z
f (x, t) dµ(x) → f (x, t0 ) dµ(x) in F as t → t0 ,

which is the desired continuity property at t 0 .

Incidentally, the continuity in b) and the inequality in c) need only hold
for almost every x in X .
INT IV.x20 extension of a measure. l p spaces §4

IV.37, `. 6–13.
“Corollary 2.”
The points x ∈ X for which the series with general term f n (x) does not
converge form a negligible set; redefining all of the f n to be 0 at such x , we
can suppose that f is defined everywhere in X . For each positive integer n ,
define
Xn
gn = fk ;
k=1

then gn ∈ LF1 , gn → f pointwise in X , and

|gn (x)| 6 g(x) for almost every x in X (n = 1, 2, 3, . . .).

Taking F to be the Fréchet filter on the set A = {1, 2, 3, . . . } of indices, it

follows from Th. 2 that f ∈ LF1 and
Z Z n Z
X ∞ Z
X
f dµ = lim gn dµ = lim fk dµ = fn dµ
n, F n→∞
k=1 n=1

as claimed.
IV.38, `. 8.
“We may limit ourselves to the case of lower semi-continuous functions”
For, if H is a set of upper semi-continuous functions directed for > ,
then −H = {−f : f ∈ H } is a set of lower semi-continuous directed for 6 .
However, the case of lower semi-continuous functions of arbitrary sign,
directed for 6 , will be reduced to the case of Cor. 2 for sets of functions
> 0 , thus will entail the consideration of sets H of both types: lower semi-
continuous and directed for 6 as well as upper semi-continuous and directed
for > .
IV.38, `. 10–12.
“ . . . the upper (resp. lower) envelope of the f + (resp. f − ), for f ∈ H ,
is equal to g + (resp. g − ).”
Let h = sup f + . Since f 6 g implies f + = sup(f, 0) 6 sup(g, 0) = g + ,
f ∈H
it is obvious that h 6 g + . To prove that g + 6 h , let x ∈ X . Since g(x) =
sup f (x) , there exists a sequence fn ∈ H such that g(x) = sup fn (x) ,
f ∈H n
and since H is directed for 6 we can suppose that the sequence (f n )
is increasing. If g(x) = +∞ then fn (x) > 0 from some index onward,
therefore
g + (x) = +∞ = sup fn+ (x) 6 sup f + (x) = h(x) ;
n f ∈H
§4 integrable functions and sets INT IV.x21

if g(x) < +∞ then fn (x) → g(x) in R , whence fn+ (x) → g + (x) , so that
g + (x) = sup fn+ (x) 6 h(x) .
n

IV.38, `. 14. R R R
“ . . . then f + d|µ| 6 f d|µ| + f0− d|µ| ”
If f > f0 then f − = sup(−f, 0) 6 sup(−f0 , 0) = f0− ; since f =
f − f − , and f and f0 are finite almost everywhere, one has, for almost
+

every x ,
f + (x) = f (x) + f − (x) 6 f (x) + f0− (x) ,
R
whence the asserted inequality (which shows that sup f + d|µ| < +∞ ).
f ∈H

IV.38, `. 15, 16.

“ . . . we are reduced to proving the two assertions of the corollary when
H consists of positive functions.”
Suppose H consists R of lower semi-continuous functions and+ is directed
for 6 , satisfying sup f d|µ| < +∞ . Since f1 6 f2 implies f1 6 f2+ and
f ∈H
f1− > f2− , it is clear that the set

H+ = {f + : f ∈ H } (resp. H− = {f − : f ∈ H } )

consists of lower (resp. upper) semi-continuous functions (GT, IV, §6, No. 2,
Prop. 2 and its analog for upper semi-continuous functions) and is directed
for 6 (resp. > ). Moreover, from f > f0 one infers that, for almost every x ,

f − (x) = f + (x) − f (x) > f0+ (x) − f (x) ,

whence Z Z Z
f d|µ| > −
f0+ d|µ| − f d|µ| ,
R
which shows that inf f − d|µ| > −∞ .
f ∈H
Thus, the validity of both parts of Cor. 2 for functions > 0 will imply
that Z Z Z Z
+ +
g dµ = lim f dµ and g dµ = lim f − dµ ,
−
f ∈H f ∈H

as well as
Z Z Z Z
g + d|µ| = sup f + d|µ| and g − d|µ| = inf f − d|µ| ,
f ∈H f ∈H

whence
Z Z Z Z Z
+ − + −
g dµ = g dµ − g dµ = lim (f − f ) dµ = lim f dµ ,
f ∈H f ∈H
INT IV.x22 extension of a measure. l p spaces §4

as well as
Z Z Z Z
g d|µ| = (g + − g − ) d|µ| = g + d|µ| − g − d|µ|
Z Z
+
= sup f d|µ| − inf f − d|µ|
f ∈H f ∈H
Z + Z
+
= sup f d|µ| + sup (−f − ) d|µ|
f ∈H f ∈H
Z + Z
= lim +
f d|µ| + lim (−f − ) d|µ|
f ∈H f ∈H
Z Z 
= lim f + d|µ| + (−f − ) d|µ|
f ∈H
Z Z
= lim f d|µ| = sup f d|µ| .
f ∈H f ∈H

To summarize: The validity of Cor. 2 for functions > 0 implies the validity
of the ‘lower semi-continuous half’ of Cor. 2 for functions of arbitrary sign
(hence the validity of the corollary as stated).
IV.38, `. −9 to −7.
“If H is directed for > and consists of upper semi-continuous integrable
functions f such that 0 6 f 6 f1 with f1 ∈ H , then there exists a lower
semi-continuous integrable function h such that f 1 6 h ”
Given a set H Rof upper semi-continuous functions > 0 , directed for >
and such that inf f d|µ| > −∞ , one chooses any f1 ∈ H ; then the lower
f ∈H
envelope g of H is equal to the lower envelope of the cofinal set

H1 = {f ∈ H : 0 6 f 6 f1 } ;

also, Z Z
inf f d|µ| = inf f d|µ| > −∞ ,
f ∈H1 f ∈H

and if we can prove that g is integrable and that

Z Z Z Z
g dµ = lim f dµ and g d|µ| = inf f d|µ| ,
f ∈H1 f ∈H1

it will follow that

Z Z Z Z
g dµ = lim f dµ and g d|µ| = inf f d|µ| .
f ∈H f ∈H
§4 integrable functions and sets INT IV.x23

Replacing H by H1 , we can suppose that 0 6 f 6 f1 for all f ∈ H .

Since f1 is positive and integrable, |µ|*(f ) < +∞ (No. 2, Prop. 1),
hence there exists a lower semi-continuous function h such that f 1 6 h and
|µ|*(h) < +∞ (§1, No. 3, Def. 3), and h is integrable by Prop. 5.
IV.38, `. −6.
“ . . . we may write f = h − f 0 . . . ”
Not quite; if f (x) < +∞ but h(x) = +∞ , then by definition f 0 (x) =
h(x) − f (x) = +∞ , so that h − f 0 is not defined at x . However, since h is
integrable it is finite almost everywhere, so that f = h − f 0 almost every-
where (which is good enough for calculating integrals).
IV.38, `. −5, −4.
“ . . . the f 0 form a directed set, for 6 , of lower semi-continuous inte-
grable functions > 0 ”
It is clear that the f 0 are > 0 . Since H is directed for > , it suffices
to show that if u, v ∈ H and u 6 v 6 f1 , then u0 > v 0 . At any rate,
0 6 u 6 v 6 f1 6 h ; if v(x) < +∞ then also u(x) < +∞ , therefore

u0 (x) = h(x) − u(x) > h(x) − v(x) = v 0 (x) ,

whereas if v(x) = +∞ then v 0 (x) = 0 6 u0 (x) .

Finally, to show that f 0 is lower semi-continuous ( f ∈ H ), we revisit
the proof of a result on sums of lower semi-continuous functions (GT, IV,
§6, No. 2, Prop. 2). Recall that 0 6 f 6 f 1 6 h (where h depends
only on f1 ). The basic idea: h and −f are lower semi-continuous on X ,
h − f = h + (−f ) is defined almost everywhere and is lower semi-continuous
at every point x where it is defined, and f 0 = h − f at almost every point
of X ; but we have to show that f 0 is lower semi-continuous at every point
of X .
Fix x ∈ X and suppose that f 0 (x) > k ∈ R ; we seek a neighbor-
hood V of x in X such that f 0 > k on V . From f 0 (x) > k we know
that k < +∞ . If k < 0 (in particular, if k = −∞ ) then, since f 0 > 0
on X , one has f 0 > k on the neighborhood V = X of x . Suppose k > 0 ,
so that 0 6 k < +∞ and f 0 (x) > k > 0 . Since f 0 (x) > 0 , we know that
f (x) must be finite and f 0 (x) = h(x) − f (x) (possibly equal to +∞ ). Thus
h(x) − f (x) > k > 0 . It follows that

k − h(x) < −f (x)

(trivial if h(x) = +∞ , elementary algebra if h(x) is finite). As in the cited

Prop. 2, let r and s be (finite) real numbers such that

k = r + s , r < h(x) , s < −f (x) ;

INT IV.x24 extension of a measure. l p spaces §4

explicitly, choose s so that

k − h(x) < s < −f (x) ,

note that k − s < h(x) (trivial if h(x) = +∞ , elementary algebra if h(x) is

finite) and set r = k − s . If V1 and V2 are neighborhoods of x such that

h > r on V1 and − f > s on V2 ,

then both inequalities hold on V = V1 ∩ V2 ; and since f < −s < +∞

on V , one has f 0 = h − f on V and


f 0 (y) = h(y) + − f (y) > r + s = k for all y ∈ V ,

briefly f 0 > k on V .
IV.38, `. −2.
“ . . . we can apply to them what has been proved above . . . ”
“them” is the set H0 = {f 0 : f ∈ H } , consisting of lower semi-
continuous integrable functions > 0 , and H 0 is directed for the relation 6 .
Recall that the lower envelope of H is denoted g in the statement of Cor. 2.
Let us write gb for the upper envelope of H 0 . {The text uses the notation g 0 ,
but this is confusing since g 0 is not derived from g in the same way that
f 0 is derived from f ; more R about this later.}
As just shown, sup f 0 d|µ| < −∞ , thus, by the case of Cor. 2 for
f ∈H
positive lower semi-continuous functions, we know that gb is integrable and
that Z Z Z Z
0
gb dµ = lim f dµ and gb d|µ| = sup f 0 d|µ| .
f ∈H f ∈H

Recall that 0 6 f 6 f1 6 h for all f ∈ H . Note that

0 6 f0 6 h for all f ∈ H ;

for, if f (x) = +∞ then f 0 (x) = 0 6 h(x) , whereas if f (x) < +∞ then

f 0 (x) = h(x) − f (x) is 6 h because f > 0, and > 0 because f 6 h . It
follows that 0 6 gb 6 h everywhere in X . Since h is integrable, the set

A = {x ∈ X : h(x) = +∞ }

is negligible. If x ∈ { A then 0 6 f (x) 6 h(x) < +∞ (f ∈ H) , and

f 0 (x) = h(x) − f (x) , g(x) and gb(x) are finite, therefore


gb(x) = sup f 0 (x) = sup [h(x) − f (x)] = h(x) + sup − f (x)
f ∈H f ∈H f ∈H

= h(x) − inf f (x) = h(x) − g(x) ;

f ∈H
§4 integrable functions and sets INT IV.x25

thus g = h − gb almost everywhere, and since h and gb are integrable, so

is g . {Incidentally, the argument shows that if g 0 were to be defined to
be h(x) − g(x) when g(x) < +∞ and to be 0 otherwise, then g 0 = gb
almost everywhere.} Moreover, since, for every f ∈ H , f 0 = h − f almost
everywhere, we have
Z Z Z Z Z
g dµ = h dµ − gb dµ = h dµ − lim f 0 dµ
f ∈H
Z Z Z 
= h dµ − lim h dµ − f dµ ,
f ∈H
R R
therefore lim f dµ exists and is equal to g dµ ; and
f ∈H
Z Z Z Z Z
g d|µ| = gb d|µ| = h d|µ| − sup f 0 d|µ|
h d|µ| −
f ∈H
Z Z Z 
= h d|µ| − sup h d|µ| − f d|µ|
f ∈H
Z Z  Z  Z
= h d|µ| − h d|µ| − sup − f d|µ| = inf f d|µ| ,
f ∈H f ∈H

which completes the proof of Cor. 2.

IV.39, `. 18, 19.
“The condition is sufficient by a general criterion for integrability (§3,
No. 4, Prop. 8), Prop. 5 and its Corollary 1.”
Note that if 0 6 g 6 f 6 h with h lower semi-continuous and inte-
grable, and g is upper semi-continuous, then
Z ∗ Z ∗
g d|µ| = N1 (g) 6 N1 (h) = h d|µ| ;

R∗ R R∗
but h d|µ| = h d|µ| < +∞ (No. 2, Prop. 1), thus g d|µ| < +∞ ,
hence g is integrable by Cor. 1 Rof Prop. 5. By hypothesis, given any ε > 0
there exist such g and h with (h − g) d|µ| 6 ε , that is,
Z ∗ Z
N1 (h − g) = |h − g| d|µ| = (h − g) d|µ| 6 ε ,

whence f is integrable (§3, No. 4, Prop. 8).

IV.39, `. −16, −15.
“ . . . since u(x) is everywhere finite, it follows that (u(x) − v(x)) + 6
f (x) 6 u(x) + v(x) for all x ∈ X .”
INT IV.x26 extension of a measure. l p spaces §4

From −v(x) 6 f (x) − u(x) 6 v(x) , we have

u(x) − v(x) 6 f (x) 6 u(x) + v(x) ;

also 0 6 f (x) , therefore sup(u(x) − v(x), 0) 6 f (x) .

IV.39, `. −14.
“The functions g = (u − v)+ and h = u + v meet the requirements.”
At any rate, 0 6 g 6 f 6 h . Since u is continuous—hence lower semi-
continuous—and since v is lower semi-continuous, it follows that h = u+v is
also lower semi-continuous (GT, IV, §6, No. 2, Prop. 2); since v is integrable
by Prop. 5, h = u + v is integrable.
Since −v is upper semi-continuous and u is continuous, u − v and
(u − v)+ = sup(u − v, 0) are upper semi-continuous (by the upper semi-
continuous analog of the cited Prop. 2 from GT). Moreover, the function
g = (u − v)+ has compact support: if u(x) = 0 then from −v 6 0 we infer
that (u(x) − v(x))+ = 0 ; therefore

(u − v)+ (x) 6= 0 ⇒ u(x) 6= 0 ,

whence Supp (u − v)+ ⊂ Supp u . Also, g is finite-valued everywhere in X ;

for, if v(x) = +∞ then g(x) = 0 .
From h − g = (u + v) − (u − v)+ and u − v 6 (u − v)+ we see that

0 6 h − g = (u + v) − (u − v)+ 6 (u + v) − (u − v) = 2v ,

whence
ε
N1 (h − g) 6 2 N1 (v) 6 2 · = ε,
2
R
that is, (h − g) d|µ| 6 ε as desired.
IV.40, `. 2, 3.
“ . . . since g 6 f , g is integrable by Prop. 4 of No. 3”
The
R gn are integrable
R (§3, No. 5, Cor. of Prop. 12) and g n ↑ g 6 f ;
since gn d|µ| 6 f d|µ| < +∞ R for all n , itR follows from Prop. 4 of No. 3
that g is integrable and that g d|µ| = lim gn d|µ| .
n

IV.40, `. −15.
“step function”
The French original is “fonction en escalier ” (literally “function in
steps”), signifying a (finite) linear combination of characteristic functions
of subintervals of R (FVR, II, §1, No. 3). Later in the section (No. 9,
Def. 4) a more general concept is introduced, signifying a function X → F
§4 integrable functions and sets INT IV.x27
P
( F a Banach space) that is a finite sum ai ϕMi , where the ai belong to F
i
and the Mi belong to a given clan Φ of subsets of X —so to speak, a linear
combination of characteristic functions of sets of Φ , with ‘coefficients’ a i
in F ; this concept is translated as “step function with respect to Φ ”, or
“ Φ-step function” (in the original, “fonction Φ-étagée” or “fonction étagée
sur les ensembles de Φ ”). A fuller discussion of how this double usage of
“step function” came about is given in the note for IV.66, `. −15, −14; for
the present, suffice it to note that “step function” in the sense of FRV is
intended also at the following places in Integration: the footnote on p. xi of
Vol. I, and, in the present chapter, the hints for Exers. 29 d) and 30 for §5,
and Exer. 17 for §6.
IV.40, `. −12 to −8.
“It follows that if f is a regulated function on R with compact support
(FRV, II, §1, No. 3), then f is integrable, because it is the uniform limit of
a sequence of step functions gn Rwith support Rcontained in a fixed compact
set (No. 3, Prop. 3); moreover, f dµ = lim gn dµ .”
n→∞
From the boldface notation and the references, we can suppose that
f : R → F , F a Banach space.
Let [a, b] be a compact interval in R such that Supp f ⊂ [a, b] and let
gn : R → F be a sequence of step functions such that g n → f uniformly
in [a, b] (FRV, II, §1, No. 3, Def. 3). Multiplying g n by the characteristic
function of [a, b] , one can suppose that
R Supp g n ⊂ [a,
R b] for all n ; then by
Prop. 3 of No. 3, f is integrable and f dµ = lim gn dµ .
n→∞
IV.40, `. −6, −5. R R +∞
“ . . . the integral f dµ is equal to the integral −∞ f (x) dx defined
in FRV, II, §2, No. 1.”
The following remark will be useful in the proof: R +∞ If f : R → F is a
regulated function ( F a Banach space) such that −∞ f (x) dx exists, then,
R +∞
for every closed interval [a, b] of R , −∞ (ϕ[a,b] f )(x) dx exists and
Z +∞ Z b
(ϕ[a,b] f )(x) dx = f (x) dx .
−∞ a

Proof. One knows that f has a primitive F : R → F (FRV, II, §1,

Rb
No. 3, Th. 2), thus a f (x) dx exists and is equal to F(b) − F(a) (FVR, II,
§1, No. 4).
Now, ϕ[a,b] f is also regulated (FRV, II, §1, No. 3, Th. 3). The core
observation is that
Z b Z b
(ϕ[a,b] f )(x) dx = f (x) dx .
a a
INT IV.x28 extension of a measure. l p spaces §4

Indeed, let G : R → F be a primitive of ϕ [a,b] f , and consider the restric-

tions G0 , F0 , (ϕ[a,b] f )0 , f0 of G , F , ϕ[a,b] f , f to [a, b] . Then G0 , F0
are continuous and, for all but countably many x in ]a, b[ ,

G00 (x) = (ϕ[a,b] f )0 (x) = 1 · f0 (x) = f (x)

and
F00 (x) = f0 (x) = f (x) ,

therefore G0 and F0 differ by a constant vector (FRV, II, §1, Prop. 1) and
Z b
(ϕ[a,b] f )(x) dx = G(b) − G(a) = G0 (b) − G0 (a)
a
Z b
= F0 (b) − F0 (a) = F(b) − F(a) = f (x) dx .
a

Next, we note that if [r, s] is a closed interval of R that contains [a, b] ,

then
Z s Z b
(†) (ϕ[a,b] f )(x) dx = f (x) dx .
r a

For, by formula (6) in FRV, II, §1, No. 5, one has

Z s
(ϕ[a,b] f )(x) dx =
r
Z a Z b Z s
(ϕ[a,b] f )(x) dx + (ϕ[a,b] f )(x) dx + (ϕ[a,b] f )(x) dx ;
r a b

the first and third terms of the right member are 0 because ϕ [a,b] f is equal
to 0 on ]r, a[ and on ]b, s[ (then consider its restrictions to [r, a] and
[b, s] as in the foregoing discussion to argue that the zero function serves
as primitives for these restrictions), whereas the middle term is equal to
Rb
a
f (x) dx , thus (†) is verified.
Now let K be the set of all compact intervals [r, s] of R , which is
directed for ⊂ . For every [r, s] ∈ K satisfying [r, s] ⊃ [a, b] , the equa-
lity (†) holds; since such intervals [r, s] are cofinal in K , it follows that the
limit Z s
lim [r,s]∈K (ϕ[a,b] f )(x) dx
r
§4 integrable functions and sets INT IV.x29
Rb
exists and is equal to a
f (x) dx , consequently
Z +∞ Z b
(ϕ[a,b] f )(x) dx = f (x) dx
−∞ a

by the definition of the left member (FRV, II, §2, No. 1); if, moreover,
Supp f ⊂ [a, b] , then ϕ[a,b] f = f and we have the following:

Z +∞ Z b
(††) f (x) dx = f (x) dx when Supp f ⊂ [a, b] .
−∞ a

To return to the assertion of IV.40, `. −6, −5, suppose first that

f = af , where a ∈ F and f is the characteristic function of an interval I
in R with finite end-points c 6 d . Then, by No. 2, Cor. 2 of Th. 1,
Z Z
f dµ = a f dµ = (d − c)a .

R +∞
But also −∞ f (x) dx = (d − c)a . For, if F : R → R is the continuous
piecewise linear function


 0 for x 6 c
F (x) = x−c for c < x < d


d−c for x > d

then F 0 (x) = f (x) except at the points c, d , thus the function F : R → F

defined by F = aF is a primitive for f in R (FRV, II, §1, No. 1, Def. 1),
and, for every closed interval [r, s] containing I ,
Z s
f (x) dx = F(s) − F(r) = (d − c)a − 0
r

(FRV, II, §1, No. 4, paragraph preceding (2)), therefore

Z +∞ Z s
f (x) dx = lim f (x) dx = (d − c)a,
−∞ r→−∞, s→+∞ r

and so
Z Z +∞
(*) f dµ = f (x) dx
−∞
INT IV.x30 extension of a measure. l p spaces §4

for all such f (FRV, II, §2, No. 1, formula (1)). It follows by linearity that
(*) holds for every step function with compact support.
Now let f : R → F be a regulated function with compact support and
let (gn ) be a sequence of step functions with support contained in a fixed
compact interval [c, d] , such that g n → f uniformly in R (see the preceding
note). By (*),
Z Z +∞
(**) gn dµ = gn (x) dx for all n ;
−∞
R R
we know that f dµ = lim gn dµ , and it follows from FRV, II, §3, No. 1,
n→∞
Cor. 1 of Prop. 1 that
Z +∞ Z d Z d Z +∞
f (x) dx = f (x) dx = lim gn (x) dx = lim gn (x) dx
−∞ c n→∞ c n→∞ −∞

(the first and last equalities by (††)), so passage to the limit in (**) shows
that (*) holds for every regulated function f with compact support.
IV.40, `. −1 to IV.41, R `. 1. Rn
“ . . . therefore |f | dµ = lim −n |f (x)| dx by Th. 2 of No. 3”
R n→∞ R
By the cited Th. 2, |f | dµ = lim ϕIn |f | dµ . Since ϕIn |f | is regu-
n→∞
lated and has compact support,
Z Z +∞
(*) ϕIn |f | dµ = (ϕIn |f |)(x) dx
−∞

by the preceding Example. With primitive-function arguments, one shows,

as at the beginning of the preceding note, that
Z n Z n
(ϕIn |f |)(x) dx = |f |(x) dx ,
−n −n

next that
Z s Z n
(ϕIn |f |)(x) dx = |f |(x) dx when [r, s] ⊃ [ − n, n] ,
r −n

whence the limit Z

lim (ϕIn |f |)(x) dx
r→−∞, s→+∞

exists, and Z Z
+∞ n
(ϕIn |f |(x) dx = |f |(x) dx ;
−∞ −n
§4 integrable functions and sets INT IV.x31

therefore by (*), Z Z n
ϕIn |f | dµ = |f |(x) dx ,
−n

and passage to the limit yields the desired formula.

IV.41, `. 1, 2. R +∞
“ . . . thus, the integral −∞ f (x) dx is absolutely convergent (FRV, II,
§2, No. 3).”
Note that if f : R → R is a regulated function such that f > 0 , and
F is a primitive of f , then F is an increasing function; for, if a 6 b then
Z b
F (b) − f (a) = f (x) dx > 0
a

by the theorem of the mean (FRV, I, §1, No. 5, Prop. 6), and it follows that
Rb Rs
for intervals [a, b] ⊂ [r, s] one has a f (x) dx 6 r f (x) dx (FRV, II, §1,
No. 5, formula (6)).
In particular, if F is a primitive for the regulated function |f | then F
is increasing. Thus if K is the set of compact intervals [r, s] of R , directed
by ⊂ , then the function
Z s
[r, s] 7→ |f |(x) dx
r

is increasing. As shown in the preceding note,

Z Z n
|f | dµ = lim |f |(x) dx ;
n→∞ −n

since the intervals [ − n, n] are cofinal in K , it follows that

Z s Z n Z n Z
sup |f |(x) dx = sup |f |(x) dx = lim |f |(x) dx = |f | dµ ,
n n→∞ −n
[r,s]∈K r −n

in other words,
Z s Z
lim |f |(x) dx = |f | dµ < +∞ .
r→−∞, s→+∞ r

Thus, in the terminology following FRV, II, §2, No. 1, Def. 1, the integral
of |f | over the interval (−∞, +∞) is convergent and
Z +∞ Z
|f |(x) dx = |f | dµ ;
−∞
INT IV.x32 extension of a measure. l p spaces §4

that is, in the language of FRV, II, §2, No. 3, Def. 2, “the integral of f is ab-
R +∞
solutely convergent”. It then follows from loc. cit., Prop. 4, that −∞ f (x) dx
exists; more precisely, see the next note.
IV.41, `. 2. R R +∞
“Moreover, f dµ = −∞ f (x) dx by Th. 2 of No. 3.”
R +∞
As remarked at the end of the preceding note, −∞ f (x) dx exists (by
FRV, II, §2, No. 3, Prop. 4). Thus, writing K for the set of compact intervals
[a, b] of R , directed by ⊂ ,
Z +∞ Z b
f (x) dx = lim f (x) dx .
−∞ [a,b]∈K a

On the other hand, by the cited Th. 2 (with |f | playing the role of g ; the
intervals [ − n, n] form a countable base for K ),
Z Z
f dµ = lim ϕ[a,b] f dµ .
[a,b]∈K

Since ϕ[a,b] f is regulated and has compact support,

Z Z +∞
ϕ[a,b] f dµ = (ϕ[a,b] f )(x) dx
−∞

by the preceding Example, and, as shown at the beginning of the note for
IV.40, `. −6, −5,
Z +∞ Z b
(ϕ[a,b] f )(x) dx = f (x) dx ,
−∞ a

therefore
Z Z Z b Z +∞
f dµ = lim ϕ[a,b] f dµ = lim f (x) dx = f (x) dx .
[a,b]∈K [a,b]∈K a −∞

IV.41, `. 2–4. R +∞
“Conversely, suppose that −∞ f (x) dx is absolutely convergent; again,
R R +∞
by Th. 2 of No. 3, f dµ = −∞ f (x) dx .”
R +∞
Assuming f is a regulated function such that −∞ |f |(x) dx is conver-
Rn
gent (i.e., exists; equivalently, the integrals −n |f |(x) dx are bounded), the
problem is to show that f is µ-integrable. (We know from FRV, II, §2, No. 3,
R +∞
Prop. 4 that −∞ f (x) dx exists, but no use of this fact will be made.)
§4 integrable functions and sets INT IV.x33
R∗
We will show that |f | dµ < +∞ , so that the function |f | can play
the role of g in the cited Th. 2. The numerical function g = |f | is regulated
and > 0 . Let In = [ − n, n] (n = 1, 2, 3, . . .). For every n , the function
ϕIn g is regulated, positive, and has compact support; by the above Example,
ϕIn g is µ-integrable and
Z Z +∞
ϕIn g dµ = (ϕIn g)(x) dx .
−∞

R +∞
Since −∞
g(x) dx exists by hypothesis, we have
Z +∞ Z n
(ϕIn g)(x) dx = g(x) dx
−∞ −n

by
R the remarkR n at the beginning of the note for IV.40, `. −6, −5; thus
ϕIn g dµ = −n g(x) dx , and
Z Z n Z +∞
µ*(ϕIn g) = ϕIn g dµ = g(x) dx 6 g(x) dx < +∞ for all n
−n −∞

(the first equality by No. 2, Prop. 1); since ϕ In g ↑ g pointwise, by §1, No. 3,
Th. 3 one has
Z +∞
µ*(g) = sup µ*(ϕIn g) 6 g(x) dx < +∞ ,
n −∞

qualifying g for its role in the cited Th. 2: each ϕ In f is µ-integrable (it is reg-
ulated and has compact support), |ϕ In f | = ϕIn gR 6 g for all Rn, and ϕIn f → f
pointwise in R , hence f is µ-integrable (and f dµ = lim ϕIn f dµ ). The
n
R R +∞
formula f dµ = −∞ f (x) dx then follows from the preceding note.
IV.41, `. −10 to −8.
“ . . . for a set to be negligible, it is necessary and sufficient that it be of
measure zero with respect to |µ| .”
Every negligible function is integrable; for, the set N of negligible
functions (the functions f such that N 1 (f ) = 0 , equivalently Np (f ) = 0 for
1 6 p < +∞ ) is the closure of {0} in the topological vector space F p (§3,
No. 3), consequently N ⊂ L p for 1 6 p < +∞ (loc. cit., No. 4, Def. 2).
IV.42, `. 4, 5.
“1◦ If A and B are two integrable sets such that B ⊂ A , then the set
C = A --- B is integrable . . . ”
INT IV.x34 extension of a measure. l p spaces §4

It follows that the condition B ⊂ A can be omitted; for, A ∪ B is

integrable by Prop. 6, and A --- B = (A ∪ B) --- B .

IV.42, `. 10.
“ ϕA = inf ϕAn , therefore A is integrable (No. 3, Prop. 4).”
n

For the cited Prop. 4 to be applicable, the sequence (A n ) must be

decreasing; thus the crux of the matter is to show that if A and B are
T
n
integrable then A ∩ B is integrable, so that A n may be replaced by Ak .
k=1
Since
ϕA∩B = ϕA ϕB = inf(ϕA , ϕB ) ,

the integrability of A ∩ B is immediate from §3, No. 5, Cor. of Prop. 12.

Alternate proof : A ∪ B is integrable by Prop. 6, therefore A ∪ B --- B
is integrable by 1◦ of Prop. 7; but A ∪ B --- B = A --- B , thus A --- B is
integrable without the restriction B ⊂ A , and A ∩ B = A --- (A --- B) is
integrable. This shows that the set Φ of integrable sets is a (Boolean) ring
of sets, that is, a clan (No. 9, Prop. 17 below).
The property 2◦ of Φ (closure under countable intersections) makes it,
so to speak, a “ δ-ring” (“ δ ” as in “ Gδ ”). In general, Φ is not closed under
countable unions, hence is not a σ-ring: if (A n ) is a sequence in Φ such
S
∞
that |µ|(An ) is unbounded, then obviously the set An is not integrable.
n=1
On the other hand, if the function A 7→ |µ|(A) on Φ is bounded, then it
follows from Prop. 8 below that Φ is closed under countable unions, hence
is a σ-ring, and the formulas in this No. show that µ and |µ| are ‘measures’
in the set-function sense (see the 2nd Note for IV.85, `. 19–27).
One notes that, in a δ-ring R , if (An ) is a sequence that is ‘bounded
above’ in the sense that there exists a set B ∈ R such that A n ⊂ B
S
∞ S
∞
for all n , then An ∈ R ; for, B --- An ∈ R for all n , and An =
n=1 n=1
T
∞
B --- (B --- An ) .
n=1

IV.43, `. −15, −14.

“Corollary 1. — Every compact set is integrable; every relatively
compact open set is integrable.”
(i) If A is compact (hence closed) then |µ|*(A) < +∞ (§1, No. 4, Cor.
of Prop. 16), hence A is integrable by Prop. 10.
(ii) If A is relatively compact, then |µ|*(A) < +∞ by §1, same Cor.;
if, moreover, A is open, then A is integrable by Prop. 10.
§4 integrable functions and sets INT IV.x35

IV.43, `. −13, −12.

“Corollary 2. — For every positive measure µ on X , A 7→ µ*(A)
is a capacity on X (cf. GT, IX, §6, No. 9, Example).”
We have µ* : P(X) → R , where µ* is monotonic (§1, No. 4, Prop. 16),
An ↑ A implies µ*(An ) ↑ µ*(A) (loc. cit., Prop. 17), and if A n ↓ A with
the An (hence also A ) compact, then µ(An ) ↓ µ(A) (by Cor. 1 and No. 5,
Cor. of Prop. 7). These are the defining properties of a capacity (GT, IX, §6,
No. 9, Def. 8). {In TG (apparently published after GT), the axiom (CA III )
for a capacity f is changed to
f (K) = inf{f (U) : K ⊂ U , U open }
for every compact set K , such capacities being called ‘continuous on the
right’; this version of the axiom is also satisfied by µ* (§1, No. 4, Prop. 19).}
IV.43, `. −2, −1.
“ . . . the proposition therefore follows from Prop. 10.”
Alternate arrangement (with proof of the formulas). Let H = {ϕ G :
G ∈ G } ; then H is directed for the relation 6 , and ϕ A is the upper
envelope of H .
RThe ϕG are lower semi-continuous and integrable, and, by hypothesis,
sup ϕG d|µ| < +∞ ; therefore (No. 4, Cor. 2 of Prop. 5), ϕ A is integrable,
G∈G
R R R R
ϕA dµ = lim ϕG dµ and ϕA d|µ| = sup ϕG d|µ| .
G∈G G∈G
IV.44, `. 7.
“ . . . we are thus reduced to Prop. 11.”
As the sets H ∈ F with H ⊂ H0 are cofinal, we can suppose that
H ⊂ U for all H ∈ F . By Prop. 11, the set
[ [
U∩ {B = U∩ {H = (U --- H)
H∈F H∈F

is integrable, therefore so is B = U --- (U --- B) by No. 5, Prop. 7, and

µ(U) − µ(B) = µ(U --- B) = lim µ(U --- H)
H,F

= lim [µ(U) − µ(H)] = µ(U) − lim µ(H) ,

H,F H,F

whence µ(B) = lim µ(H) ; similarly,

H,F
 [ 
|µ|(U) − |µ|(B) = |µ|(U --- B) = |µ| (U --- H)
H∈F

= sup |µ|(U --- H) = sup [|µ|(U) − |µ|(H)]

H∈F H∈F

= |µ|(U) − inf |µ|(H) ,

H∈F
INT IV.x36 extension of a measure. l p spaces §4

whence |µ|(B) = inf |µ|(H) .

H∈F

IV.44, `. −11.
“ f 6 ϕK + δϕB ”
For f (x) = 0 , the inequality is trivial. Suppose f (x) > 0 . Then
ϕA > f yields ϕA (x) > 0 , so ϕA (x) = 1 , x ∈ A ; if x ∈ / K (that is,
f (x) < δ ) then x ∈ A --- K = B and the desired inequality reduces to
f (x) 6 0 + δ · 1 , whereas if x ∈ K then ϕK (x) = 1 and the desired
inequality results from f (x) 6 ϕA (x) = 1 = ϕK (x) .
IV.45, `. 1–3.
“The condition is sufficient, because it says that, for the topology of
convergence in mean, ϕA is in the closure of the set of integrable func-
tions ϕK ( K an arbitrary compact subset of A ).”
Observe first that ϕA ∈ F 1 : for ε = 1 choose a compact set K1 ⊂ A
with |µ|*(A --- K1 ) 6 1 ; then


|µ|*(A) = |µ|* (A --- K1 ) ∪ K1
6 |µ|*(A --- K1 ) + |µ|*(K1 ) 6 1 + |µ|*(K1 ) < +∞ ,

thus ϕA ∈ F 1 . The sufficiency of the condition then follows from the fact
that L 1 is closed in F 1 (§3, No. 4, Def. 2).
The set F of all compact sets K ⊂ A is directed for ⊂ , and from
|µ|(A) − |µ|(K) = |µ|(A --- K) and the monotonicity of |µ| , one infers that

|µ|(A) = lim |µ|(K) = sup |µ|(K) ;

K,F K∈F

and then (No. 5, Prop. 7) |µ(A) − µ(K)| = |µ(A --- K)| 6 |µ|(A --- K) yields
µ(A) = lim µ(K) .
K,F

IV.45, `. −11, −10.

“ ϕU is the upper envelope of the set H of functions f ∈ K + such
that f 6 ϕU and Supp (f ) ⊂ U (cf. the proof of §1, No. 1, Lemma) ”
Clearly H = {f ∈ K+ : 0 6 f 6 1 and Supp f ⊂ U }, and H is
directed for the relation ⊂ . Writing

g(x) = sup f (x) (x ∈ X) ,

f ∈H

we are to show that g = ϕU . Since f 6 ϕU for all f ∈ H , obviously

g 6 ϕU .
Conversely, assuming x ∈ X , let us show that ϕ U (x) 6 g(x) . If x ∈
/U
then ϕU (x) = 0 6 g(x) . Whereas if x ∈ U , then there exists an f ∈ K +
§4 integrable functions and sets INT IV.x37

such that 0 6 f 6 1 , f (x) = 1 and f = 0 on { U : for, let V be a

compact neighborhood of x such that V ⊂ U , and let f be a continuous
function X → [0, 1] such that f (x) = 1 and f = 0 on { V , whence
Supp f ⊂ V ⊂ U ; thus f ∈ H and so g(x) > f (x) = 1 = ϕ U (x) .
IV.45, `. −8, −7.
“ . . . the corollary is then immediate from the fact that if f ∈ H and
K = Supp (f ) , then f 6 ϕK 6 ϕU .”
From ϕU = g = supf ∈H f and Th. 1 of §1, No. 1, we have

(*) |µ|*(U) = |µ|*(ϕU ) = sup |µ|(f ) .

f ∈H

For f ∈ H , writing Kf = Supp f , we know that Kf ⊂ U , so that

f = ϕ Kf f 6 ϕ Kf · 1 6 ϕ U ;

in view of Cor. 1 of Prop. 10,

|µ|*(U) > |µ|*(Kf ) = |µ|(Kf ) > |µ|(f ) ,

and it follows from (*) that

|µ|*(U) > sup |µ|(Kf ) > sup |µ|(f ) = |µ|*(U) ,

f ∈H f ∈H

whence
|µ|*(U) = sup |µ|(Kf ) .
f ∈H

But |µ|(K) 6 |µ|*(U) for all compact K ⊂ U , so a fortiori

|µ|*(U) = sup{|µ|(K) : K ⊂ U compact}.

IV.46, `. 7, 8.
“For, we have seen that |µ|*(X) = kµk (§1, No. 2); the proposition
therefore follows from Prop. 10 of No. 6.”
By the remarks following §1, No. 2, Def. 2, |µ|*(X) = k |µ| k , and
k |µ| k = kµk (Ch. III, §1, No. 8, Cor. 1 of Prop. 10); µ is bounded if
and only if kµk < +∞ (loc. cit., III.16, `. −11, −10), hence if and only if
|µ|*(X) < +∞ , that is (No. 6, Prop. 10), X is µ-integrable. This Rmeans, in
turn, that ϕX = 1 is µ-integrableR (No. 5, Def. 2), in which case 1 d|µ| =
|µ|*(1) (No. 2, Prop. 1), that is, d|µ| = kµk , and one writes
Z Z
|µ|(X) = ϕX d|µ| = d|µ|

(No. 5, Def. 2).

INT IV.x38 extension of a measure. l p spaces §4

IV.46, `. −8, −7.

“ . . . if f ∈ M ∩ N then Np ((f − f0 )ϕK ) 6 ε and Np ((f − f0 )ϕ { K ) 6
2aε ”

−1/p
From f ∈ N we see that |f − f0 |ϕK 6 ε |µ|(K) ϕK , whence

−1/p
−1/p
1/p
Np (|f − f0 |ϕK ) 6 ε |µ|(K) Np (ϕK ) = ε |µ|(K) |µ|(K) = ε.

On the other hand, |f | 6 a (because f ∈ M ), and |f 0 | 6 a , thus

|f − f0 | 6 |f | + |f0 | 6 2a ; then |f − f0 |ϕ { K 6 2a ϕ { K , whence

1/p
Np (|f − f0 |ϕ { K ) 6 2a Np (ϕ { K ) = 2a (|µ|( { K) 6 2aε .

IV.46, `. −1 to IV.47, `. 3.
“It is clear that the sets MK form a filter base B on LFp , that the
functions belonging to MK are uniformly bounded, and that B converges
uniformly to f on every compact subset of X , whence the corollary.”
The functions hf belong to KF , hence to LFp (§3, No. 4, Def. 2), thus
MK ⊂ LFp for every compact set K ⊂ X . If K1 , K2 are compact and K is
a compact set such that K ⊃ K1 ∪ K2 , then MK ⊂ MK1 ∩ MK2 (because
h = 1 on K ⇒ h = 1 on K1 and h = 1 on K2 ), so that the MK form
a filter base on LFp . The uniform boundedness of the hf is clear from the
boundedness of f and the fact that khk = 1 .
Given any compact K0 ⊂ X , we are to show that B → f uniformly
on K0 (it will then follow from Prop. 13 that f ∈ L Fp and that B → f in
mean of order p ). Given any ε > 0 , suppose K ⊃ K 0 , K compact, and let
hf ∈ MK , where h ∈ K , 0 6 h 6 1 and h = 1 on K ; then

x ∈ K0 ⇒ x ∈ K ⇒ h(x) = 1 ⇒ (hf − f )(x) = 0 ,

thus sup |(hf − f )(x)| = 0 < ε and the claimed uniform convergence on K 0
x∈K0
holds trivially.
IV.47,R `. 4, 5. R
“ f dµ is the limit with respect to B of the integrals hf dµ .”
The integral is (by definition) a continuous linear mapping L F1 → F
(No. 1, Def. 1).
IV.47, `. 13–17. R
“ the mapping f 7→ f dµ is continuous on the Banach space C b (X; F) ;
its restriction to the closure C 0 (X; F) of K (X; F) in C b (X; F) , that is,
to the space of continuous functions tending to 0 at the point at infinity
§4 integrable functions and sets INT IV.x39

(Ch. III, §1, No. 2, Prop. 3), is therefore the extension by continuity of the
integral to C 0 (X; F) .”
The notations C b (X; F) and C 0 (X; F) are established in Ch. III, §1,
No. 2. Measures are defined in the following subsection (loc. cit., No. 3), the
set M (X; C) of all (complex) measures µ being defined to be the vector
space dual to the topological vector space K (X; C) (equipped with the
direct limit topology).
Bounded measures µ are defined loc. cit., No. 8, the definition of µ(f )
being limited to f ∈ K (X; C) ; it is observed that the linear subspace
M 1 (X; C) of M (X; C) consisting of the bounded measures µ may be re-
garded as the dual of the vector space K (X; C) equipped with the topol-
ogy defined by the norm kf k = sup |f (x)| , but the domain of µ remains
x∈X
K (X; C) . The possibility of extending µ to C b (X; C) , and in particu-
lar to C 0 (X; C) , is not taken up, but it is is latent in the observation
that if g ∈ C b (X; C) then g · µ is a bounded measure (ibid., Prop. 12),
whence ϕX = 1 is (g · µ)-integrable
R (Prop. 12 of the present No. 7), invit-
ing the definition µ(g) = 1 d(g · µ) (No. 1, Def. 1). This is accomplished
in the present No., more generally for f ∈ C b (X; F) (Cor. of Prop. 13):
C b (X; F) ⊂ LFp (X, µ) and, by the inequality (20), the mapping f 7→ µ(f ) so
defined is a linear mapping C b (X; F) → F continuous for the norm topology
on C b (X; F) . R
Write L(f ) = f dµ ( µ a bounded measure, f ∈ C b (X; F) ), which is
a linear mapping C b (X; F) → F continuous for the norm topology, and let
L0 = LC 0 (X; F) be the restriction of L to C 0 (X; F) . One has

K (X; F) ⊂ C 0 (X; F) ⊂ C b (X; F) ⊂ LF1 (X, µ)

(for µ bounded); since K (X; F) is by definition dense in C 0 (X; F) for the

norm topology, and since L0 is continuous for that topology,
R it is clear that
L0 is the extension by continuity of the integral f 7→ f dµ ( f ∈ K (X; F) )
to C 0 (X; F) .
IV.47, `. −11, −10.
“Let µ be a measure on X whose support S = Supp(µ) is compact;
the open set X --- S is negligible (§2, No. 2, Prop. 5).”
In particular, ϕX and ϕS are equivalent, whence

|µ|*(X) = |µ|*(S) = |µ|(S) < +∞ ,

thus X is integrable (No. 6, Prop. 10) and so µ is bounded (No. 7, Prop. 12),
as already observed in Ch. III, §2, No. 3, Prop. 11.
INT IV.x40 extension of a measure. l p spaces §4

IV.47, `. −5, −4.

“If, moreover, f is bounded on S , it follows from (20) that
Z 
 
 
(22)  f dµ 6 kµk · sup |f (x)| .”
x∈S

We know that µ is bounded (see the preceding note). Let M =

sup |f (x)| ; by assumption, M < +∞ . It is also assumed that f is µ-
x∈S
integrable, equivalently, f ϕS is µ-integrable. Then |f ϕS | 6 MϕS 6 MϕX
and
Z  Z  Z
   
 f dµ =  f ϕS dµ 6 |f ϕS | d|µ| 6 M |µ|(1) = M kµk ,

where the first inequality holds by the inequality (5) of No. 2, Prop. 5 (cited
also in the proof of (20), which, strictly speaking, is stated only for contin-
uous bounded functions).
IV.47, `. −3 to −1.
“In particular, if f is continuous on X then f is µ-integrable, since
f h ∈ K (X; F) for every function h ∈ K (X; R) equal to 1 on S (Ch. III,
§1, No. 2, Lemma 1).”
Let h be any such function (which exists by the cited lemma). Then
f −f h = 0 on S ; since { S is µ-negligible (§2, No. 2, Prop. 5), it follows that
|f − f h| = 0 almost everywhere, consequently N 1 (f − f h) = 0 (§2, No. 3,
Prop. 6). Thus, f −f h is negligible for µ (§2, No. 1, Def. 1) hence belongs to
LF1 (X, µ) (see the note for IV.21, `. 7–9); but f h ∈ K (X; F) ⊂ L F1 (X, µ)
(§3, No. 4, Def. 2), therefore f = (f − f h) + f h ∈ L F1 (X, µ) . (The same
argument shows that f ∈ LFp (X, µ) for 1 6 p < +∞ .)
Thus, when µ has compact support, every f ∈ C (X; F) is µ-integrable;
since such a function is bounded on every compact subset of X , and in
particular on S = Supp(µ) , the inequality (22) holds for every f ∈ C (X; F) .
IV.48, `. 9–11. R
“. . .the mapping f 7→ f dµ of C (X; F) into F is continuous for the
topology of compact convergence.”
The locally convex topology in question is defined by the family of semi-
norms pK (f ) = sup |f (x)| , where K ⊂ X is compact (GT, X, §1, No. 3,
x∈K
Example III and No. 6; TVS, II, §1, No. 2 and §4, No. 1, Cor. of Prop. 1;
and the note forR III.39, `. 8–11). The inequality (22) shows that the linear
mapping f 7→ f dµ ( f ∈ C (X; F) ) is continuous for the topology defined
by the semi-norm pS ( S the support of µ ), hence for the (finer) topology
defined by all the pK .
§4 integrable functions and sets INT IV.x41

IV.48, `. 12–14.
“Then, there is a compact set K ⊂ X and a number a > 0 such that
|µ(f )| 6 a · sup |f (x)| for every function f ∈ K (X; F) ”
x∈K
R 
 
It follows from the hypothesis that the formula q(f ) =  f dµ defines a
semi-norm on K (X; F) continuous for the topology of compact convergence;
as that topology is defined by the family of semi-norms p K (f ) = sup |f (x)|
x∈K
with K ⊂ X compact (see the preceding note), at issue is the character-
ization of the continuous semi-norms on K (X; F) (and C (X; F) ) for the
topology. As the topology of a topological vector space can be defined by
a set of semi-norms if and only if it is locally convex (TVS, II, §4, No. 1,
Cor. of Prop. 1), it is of interest (and notationally simpler) to consider the
general case.
Let E be a vector space (over R or C ) and let p be a semi-norm
on E . The sets
V(p, α) = {z ∈ E : p(z) 6 α } (α > 0)
form a fundamental system of neighborhoods of 0 for a locally convex topol-
ogy τp on E , and τp is the coarsest topology on E that makes p continu-
ous and is compatible with the additive group structure of E (TVS, II, §1,
No. 2).
Lemma 1. For a semi-norm q on E , the following conditions are equiv-
alent:
(a) q is continuous for τp ;
(b) τq ⊂ τp ;
(c) there exists a constant M > 0 such that q 6 Mp on E .
Proof. (b) ⇒ (a): q is continuous for τ q , hence a fortiori for the finer
topology τp .
(a) ⇒ (b): The topology τp is compatible with the additive structure
of E and makes q continuous, hence τq is coarser than τp .
(c) ⇒ (a): For every α > 0 , clearly p(z) 6 α/M ⇒ q(z) 6 α , thus
{z : q(z) 6 α } ⊃ {z : p(z) 6 α/M } ;
since the right side is a neighborhood of 0 for τ p (by the continuity of p
for τp ) so is the left side, consequently q is continuous at 0 for τ p . It
then follows from |q(z) − q(z0 )| 6 q(z − z0 ) that q is (uniformly) continuous
for τp .
(a) ⇒ (c): Let D = {λ : |λ| 6 1 } , a neighborhood of 0 in the field of
scalars. By hypothesis, q is continuous at 0 for τ p , hence the set
−1
V(q, 1) = {z : q(z) 6 1 } = q (D)
INT IV.x42 extension of a measure. l p spaces §4

is a neighborhood of 0 for τp ; it follows that there exists a scalar α > 0

such that
{z : q(z) 6 1 } ⊃ {z : p(z) 6 α } .
Let M = 1/α . If z ∈ E and ε > 0 is arbitrary, then the vector z 0 =

−1
α p(z) + ε z satisfies

−1
p(z0 ) = α p(z) + ε p(z) < α ,

0

−1

) 6 1 , that is, α p(z) + ε
therefore q(z q(z) 6 1 , in other words q(z) 6
M p(z) + ε ; since ε is arbitrary, q(z) 6 Mp(z) .
Lemma 2. Let Γ be a set of semi-norms on E and let τ be the (locally
convex ) topology on E defined by Γ , that is, the coarsest topology τ on E
that is compatible with the additive structure of E and is such that every
p ∈ Γ is continuous (TVS, II, §1, No. 2).
Assume, moreover, that if p1 , . . . , pn belong to Γ then there exists a
p ∈ Γ such that p > sup pi (the upper envelope of p1 , . . . , pn ). Then, for
16i6n
a semi-norm q on E , the following conditions are equivalent:
(i) q is continuous for τ ;
(ii) there exist a p ∈ Γ and a constant M > 0 such that q 6 Mp .
Proof. Clearly τ is the coarsest topology, compatible with the additive
structure of E , for which τ ⊃ τp for all p ∈ Γ .
(ii) ⇒ (i): By Lemma 1, q is continuous for τ p , hence a fortiori for
the finer topology τ .
(i) ⇒ (ii): The sets Dα = {λ : |λ| 6 α } (α > 0) form a base for
the neighborhoods of 0 in the field of scalars. By the definition of τ ,
a neighborhood base at 0 ∈ E for τ is given by the finite intersections

−1 −1
p 1 (Dα1 ) ∩ · · · ∩ p n (Dαn ) ,

where p1 , . . . , pn ∈ Γ and α1 , . . . , αn are > 0 (TVS, loc. cit.). If α =

min(α1 , . . . , αn ) , then Dα ⊂ Dαi for all i , therefore the sets

−1 −1
p 1 (Dα ) ∩ · · · ∩ p n (Dα ) ( α > 0 , pi ∈ Γ for 1 6 i 6 n )

also form a neighborhood base at 0 for τ . By assumption, there exists a

p ∈ Γ such that p > sup pi . Then
16i6n

p(z) 6 α ⇒ pi (z) 6 α for i = 1, . . . , n ,

§4 integrable functions and sets INT IV.x43
−1 −1
thus p (Dα ) ⊂ p i (Dα ) for i = 1, . . . , n , whence

−1 −1 −1
p (Dα ) ⊂ p 1 (Dα ) ∩ · · · ∩ p n (Dα ) ,

therefore the sets

−1
p (Dα ) (α > 0, p ∈ Γ)
also form a neighborhood base at 0 for τ .
−1
By assumption, q is continuous at 0 for τ , hence the set q (D1 ) =
{z : q(z) 6 1 } is a neighborhood of 0 for τ ; by the foregoing, there exist
−1 −1
an α > 0 and a p ∈ Γ such that q (D1 ) ⊃ p (Dα ) , that is,

{z : q(z) 6 1 } ⊃ {z : p(z) 6 α } ,

therefore, as argued in the proof of “(a) ⇒ (c)” of Lemma 1, q 6 Mp with

M = 1/α .
Consider now the space K (X; F) (or C (X; F) ) equipped with the
topology of compact convergence, that is, the topology defined by the family
of semi-norms

pK (f ) = sup |f (x)| ( K ⊂ X compact).

x∈K

S
n
If K1 , . . . , Kn are compact and K = Ki , then pK > pKi for i = 1, . . . , n ,
i=1
therefore pK > sup pKi ; in view of Lemma 2, if q is a semi-norm on the
16i6n
space, then q is continuous for the topology of compact convergence if and
only if q 6 a · pK for some a > 0 and compact set K , whence the original
assertion of the text.
IV.48, `. 16, 17.
“ µ(h) = 0 for every function h ∈ K (X; C) whose support does not
intersect K ”
Recall that µ(ha) = µ(h)a (Ch. III, §3, No. 4, Prop. 8).
IV.48, `. 17, 18.
“ . . . which proves that Supp(µ) ⊂ K .”
By the foregoing argument, µ(h) = 0 for every h ∈ K (X; C) such
that Supp(h) ⊂ { K ; this means that the measure induced by µ on the
open set { K is 0 (Ch. III, §2, No. 1), therefore { K ⊂ { Supp(µ) by the
definition of Supp(µ) (loc. cit., No. 2, Def. 1).
INT IV.x44 extension of a measure. l p spaces §4

IV.48, `. 22–24.
“The set of measures on X with compact support may therefore be
identified with the dual C 0 (X; C) of the Hausdorff locally convex space
C (X; C) .”
Write Mc for the set of all measures on X with compact support. Since
the support of the sum of two measures is the union of their supports (Ch. III,
§2, No. 2, Prop. 4), it is clear that M c is a linear subspace of the space
M (X; C) of all measures on X . According to Prop. 14, for each measure µ
with Rcompact support, every f ∈ C (X; C) is µ-integrable, the linear form
f 7→ f dµ is continuous for the topology τ cc of compact convergence on
C (X; C) (hence is an element of the dual space C 0 (X; C) ), and it is the
only element of C 0 (X; C) that extends the linear form µ : K (X; C) → C .
For each µ ∈ Mc let us write µ0 ∈ C 0 (X; C) for its unique τcc -
continuous extension to C (X; C) . The mapping M c → C 0 (X; C) defined
by µ 7→ µ0 is linear; for instance, if µ1 , µ2 ∈ Mc then µ01 + µ02 extends
µ1 + µ2 , therefore (µ1 + µ2 )0 = µ01 + µ02 . The mapping is injective; for, the
restriction of µ0 to K (X; C) is µ , so if µ0 = 0 then µ = 0 . Thus µ 7→ µ0
is an injective linear mapping Mc → C 0 (X; C) ; our problem is to show that
it is also surjective.
Thus, given ν ∈ C 0 (X; C) , we seek a measure µ on X with compact
support such that µ0 = ν . Let µ = ν K (X; C) and let us show that µ
is a measure on X with compact support. It suffices to show that µ is
a measure, for, by Prop. 14, compactness of its support is assured by the
existence of the extension ν ; and then µ 0 = ν , because ν extends µ .
Given any compact subset K of X , it suffices to show that there exists
a constant M > 0 such that

(∗) |µ(f )| 6 M · kf k for all f ∈ K (X, K; C) ,

where kf k = sup |f (x)| (Ch. III, §1, No. 3). Now, the topology τ cc on
x∈X
C (X; C) is defined by the set of semi-norms p L (f ) = sup |f (x)| ( L a com-
x∈L
pact subset of X ), and the semi-norm q(f ) = |ν(f )| ( f ∈ C (X; C) ) is
continuous for τcc ; as shown in the note for IV.48, `. 9–10, there exist a
compact set L and a constant M > 0 such that q 6 M · p L on C (X; C) ,
hence on K (X; C) . Thus, for f ∈ K (X; C) , we have

|µ(f )| = |ν(f )| = q(f ) and pL (f ) 6 kf k ,

therefore |µ(f )| 6 M · pL (f ) 6 M · kf k ; in particular, (*) holds.

The argument shows incidentally that a measure µ with compact sup-
port is continuous on K (X; C) for the norm topology—not news, since µ
is bounded (Ch. III, §2, No. 3, Prop. 11).
§4 integrable functions and sets INT IV.x45

IV.48, `. −13 to −11.

“ . . . the topology of C (X; C) can be defined by the countable family
of semi-norms pn (f ) = sup |f (x)| , therefore C (X; C) is a Fréchet space
x∈Kn
in this case.”
Following GT, X, §1, No. 6, we write C c (X; C) for C (X; C) equipped
with the topology of compact convergence (τ cc ). We know that τcc is defined
by the semi-norms pK (f ) = sup |f (x)| ( K ⊂ X compact), and the sets
x∈K

{f : pK (f ) 6 ε } ( K ⊂ X compact, ε > 0 }

form a fundamental system of neighborhoods of 0 in C c (X; C) (see the note

for IV.48, `. 12–14). To show that τcc is defined by the semi-norms pn , it
suffices to show that if K is any compact set in X then there exists an index
n such that pK 6 pn ; for, this will imply that

{f : pn (f ) 6 ε } ⊂ {f : pK (f ) 6 ε } ,

hence that the sets

{f : pn (f ) 6 ε } ( ε > 0 , n = 1, 2, 3, . . .)

form a fundamental system of neighborhoods of 0 for τ cc .

If K is any compact set in X , then
∞
[ ∞
[ ◦
K⊂X= Kn = Kn+1 ,
n=1 n=1

◦ ◦ ◦ ◦
thus the sets Kn form an open covering of K ; since K1 ⊂ K2 ⊂ K3 ⊂ · · ·
◦ ◦
and since K is covered by finitely many of the Kn , one has K ⊂ Kn for
some n , whence K ⊂ Kn and so pK 6 pKn = pn .
It follows that the sets

{f : pn (f ) 6 1/m } (m, n = 1, 2, 3, . . .)

form a countable fundamental system of neighborhoods of 0 for τ cc , and

their intersection is visibly equal to {0} , therefore C c (X; C) is metrizable
(TVS, I, §3, No. 1); being locally convex and complete, it is a Fréchet space
(TVS, II, §4, No. 1), hence is barreled (TVS, III, §4, No. 1, Cor. of Prop. 2).
IV.49, `. 11.
“ . . . and proves the proposition.”
INT IV.x46 extension of a measure. l p spaces §4

Write V = {µ ∈ C 0 (X; C) : |µ(f ) − εx0 (f )| 6 δ for all f ∈ H } ,

where H is a compact subset of Cc (X; C) , and δ > 0 ; such sets form a
fundamental system of neighborhoods of ε 0 in C 0 (X; C) for the topology
of uniform convergence in the compact subsets of C c (X; C) . The argument
produces a neighborhood U of x0 in X such that

x ∈ U ⇒ |εx (f ) − εx0 (f )| 6 δ for all f ∈ H ,

that is, x ∈ U ⇒ εx ∈ V , whence the continuity of x 7→ εx at x0 for the

indicated topology on C 0 (X; C) .
IV.49, `. 17, 18.
“It is clear that on L , the topology induced by T is finer than the
topology induced by T 0 .”
As in the note for IV.48, `. 22-24, we write M c for the set of all
measures on X with compact support, and µ 7→ µ 0 for the linear bijection
Mc → C 0 (X; C) described there.
Caution . There is defined on M (X; C) a topology of compact con-
vergence (uniform convergence in the subsets of K (X; C) that are compact
for the direct limit topology); the topology it induces on the subset M c
of M (X; C) is not to be confused with the topology T on the dual space
C 0 (X; C) = {µ0 : µ ∈ Mc } of C (X; C) ( T is the topology of uniform con-
vergence in the compact subsets of C (X; C) , where C (X; C) is equipped
with the topology of uniform convergence in the compact subsets of X ).
Thus, although it is possible to identify the vector space M c with the vector
space C 0 (X; C) , in the present discussion it is useful to maintain the distinc-
tion between them, that is, between µ ∈ M c (a linear form on K (X; C) )
and µ0 (a linear form on C (X; C) ).
{Incidentally, L is a vaguely closed linear subspace of M (X; C) (Ch. III,
§2, No. 2, Prop. 6), but I see no way to invoke Prop. 17 of Ch. III, §1,
No. 10).}
Since L ⊂ Mc ⊂ M (X; C) , we may write

L0 = {µ0 : µ ∈ L } ⊂ C 0 (X; C) .

We have L ⊂ M (X; C) and T 0 is a topology on M (X; C) ; let us

write T 0 ∩ L for the topology on L induced by T 0 .
On the other hand, L0 ⊂ C 0 (X; C) and T is a topology on C 0 (X; C) ;
we write T ∩ L0 for the topology on L0 induced by T .
Proposition 16 asserts, informally, that T ∩ L = T 0 ∩ L ; we interpret
this to mean that the mapping µ 7→ µ0 (µ ∈ L) is a homeomorphism L → L0
for the topologies T 0 ∩ L and T ∩ L0 .
§4 integrable functions and sets INT IV.x47

In the present note, the task is to show that “ T ∩ L is finer than

T ∩ L ”; we interpret this to mean that the inverse mapping L 0 → L
0

defined by µ0 7→ µ (µ ∈ L) is continuous for the topologies T ∩ L 0 and

T 0 ∩L . It will suffice to prove that the bijection C 0 (X; C) → Mc defined by
µ0 7→ µ is continuous for T and T 0 ∩ Mc (for, this will imply the asserted
continuity for an arbitrary subset L of M c , since T 0 ∩ L = (T 0 ∩ Mc ) ∩ L ).
Informally, we will show that if µ0 → µ00 in C 0 (X; C) for the topology
of uniform convergence in the compact subsets of C (X; C) , then µ → µ 0
in Mc for the topology of strictly compact convergence. {The argument
can be formalized using the concept of convergence of nets, as exposed in
S. Willard’s General topology [Addison–Wesley, Reading, 1970], p. 75, but
it is simpler to argue with neighborhoods of zero in locally convex spaces.}
We will need the following two lemmas.
Lemma 1. If J is a compact subset of X , then the topology of C (X; C)
induces on K (X, J; C) the norm topology.
Proof. C (X; C) bears the topology τcc of uniform convergence in the
compact subsets of X , thus the assertion is that τ cc ∩ K (X, J; C) is the
topology of uniform convergence in X (equivalently, in J ), in other words
the norm topology, which we shall denote by τ n . Thus, the assertion is that
τcc ∩ K (X, J; C) = τn .
Let G be a compact subset of X , let ε > 0 , and write

V(G, ε) = {f ∈ C (X; C) : |f (x)| 6 ε for all x ∈ G } ;

such sets form a fundamental system of neighborhoods of 0 for τ cc . Then

V(G, ε) ∩ K (X, J; C) = {f ∈ K (X, J; C) : |f (x)| 6 ε for all x ∈ G }

⊃ {f ∈ K (X, J; C) : |f (x)| 6 ε for all x ∈ X } ,

which shows that V(G, ε) ∩ K (X, J; C) is a neighborhood of 0 for τ n ,

whence τcc ∩ K (X, J; C) ⊂ τn .
On the other hand, for ε > 0 write

W(ε) = {f ∈ K (X, J; C) : |f (x)| 6 ε for all x ∈ X } ;

such sets form a fundamental system of neighborhoods of 0 for τ n . Then

W(ε) = {f ∈ C (X; C)) : |f (x)| 6 ε for all x ∈ J } ∩ K (X, J; C) ,

which shows that W(ε) is a neighborhood of 0 for τ cc ∩K (X, J; C) , whence

τn ⊂ τcc ∩ K (X, J; C) .
INT IV.x48 extension of a measure. l p spaces §4

Lemma 2. If S is a strictly compact subset of K (X; C) , then S is

compact in C (X; C) .
Proof. Let J be a compact subset of X such that S is a compact
subset of K (X, J; C) for the norm topology. By Lemma 1, S is compact
in K (X, J; C) for the topology τcc ∩ K (X, J; C) , whence it is elementary
that S is compact in C (X; C) for τcc .
We are ready to prove that the (bijective) linear mapping Φ : C 0 (X; C) →
Mc defined by Φ(µ0 ) = µ is continuous for the topologies T and T 0 ∩Mc .
Given a strictly compact subset S of K (X; C) , and ε > 0 , let

U(S, ε) = {µ ∈ M (X; C) : |µ(f )| 6 ε for all f ∈ S } ;

such sets form a fundamental system of neighborhoods of 0 in M (X; C)

for T 0 , hence the sets

U(S, ε) ∩ Mc = {µ ∈ Mc : |µ(f )| 6 ε for all f ∈ S }

form a fundamental system of neighborhoods of 0 in M c for T 0 ∩ Mc .

Since S is a compact subset of C (X; C) (Lemma 2), the set

V = {µ0 ∈ C 0 (X; C) : |µ0 (f )| 6 ε for all f ∈ S }


is a neighborhood of 0 in C 0 (X; C) for T ; but µ0 K (X; C) = µ , so

V = {µ0 ∈ C 0 (X; C) : |µ(f )| 6 ε for all f ∈ S }

= {µ0 ∈ C 0 (X; C) : µ ∈ U(S, ε) ∩ Mc }
−1

= Φ U(S, ε) ∩ Mc ,

−1

thus Φ U(S, ε) ∩ Mc is a neighborhood of 0 in C 0 (X; C) for T , whence
the asserted continuity of Φ .
IV.49, `. 19, 20.
“It is clear that the set H0 of functions f h , where f runs over H , is
strictly compact in K (X; C) ”
The formula Ψ(f ) = f h defines a linear mapping Ψ : C (X; C) →
C (X; C) , and Ψ(H) = H0 . Moreover, if G is the support of h , then Ψ
takes all of its values in K (X, G; C) and in particular H 0 ⊂ K (X, G; C) .
It will suffice to prove that Ψ is continuous (for the topology τ cc of compact
convergence); for, this will imply that H 0 = Ψ(H) is compact in C (X; C)
for τcc , whence H0 is compact in K (X, G; C) for the induced topology
§4 integrable functions and sets INT IV.x49

τcc ∩ K (X, G; C) , in other words, for the norm topology on K (X, G; C)

(Lemma 1 of the preceding note), whence the strict compactness of H 0 .
For each compact subset J of X , let pJ be the semi-norm on C (X; C)
defined by pJ (f ) = sup |f (x)| ; the topology τcc of C (X; C) is the topology
x∈J
defined by the semi-norms pJ . For J ⊂ X compact and ε > 0 , write

V(J, ε) = {f ∈ C (X; C) : pJ (f ) 6 ε } ;

the sets V(J, ε) form a fundamental system of neighborhoods of 0 for τ cc .

For every f ∈ C (X; C) one has pJ (f h) 6 pJ (f ) · khk , therefore

{f ∈ C (X; C) : pJ (f ) 6 ε/khk } ⊂ {f ∈ C (X; C) : pJ (f h) 6 ε } ,

in other words
−1

V(J, ε/khk) ⊂ {f ∈ C (X; C) : Ψ(f ) ∈ V(J, ε) } = Ψ V(J, ε) ;

−1

this shows that, for every J and ε , Ψ V(J, ε) is a neighborhood of 0 in
C (X; C) for τcc , whence the continuity of Ψ .
IV.49, `. 21.
“ for every measure µ ∈ L , µ(f ) = µ(f h) for every function f ∈ H ”
As f ∈ C (X; C) and f h ∈ K (X; C) , a review of the notations is in
order. In the expression µ(f h) , µ is the original linear form on K (X; C) .
As noted at the beginning of the No., since µ has compact support, f is
µ-integrable
in the sense of No. 1, Def. 1 ( f − f h = 0 on K ⊃ Supp(µ) and
{ Supp(µ) is negligible (§2, No. 2, Prop.R 5), therefore f = f h µ-almost
everywhere) and, by Prop. 14, f 7→ f dµ ( f ∈ C (X; C) ) is the unique
linear form on C (X; C) that is continuous for the topology of compact
convergence and extends theR original linear form f 7→ µ(f ) ( f ∈ K (X; C) ).
One also writes µ(f ) for f dµ ( f ∈ C (X; C) ), so that µ(f ) = µ(f h)
(because f = f h µ-almost everywhere).
In the note for IV.48, `. 22–24, we have introduced the ephemeral nota-
tion µ0 for the linear form in C (X; C) so defined, as an aid to understanding
the proofs; thus µ0 (f ) = µ(f ) ( f ∈ C (X; C) ) by the definition of µ 0 , and
µ0 (f ) = µ(f ) = µ(f h) = µ0 (f h) .
IV.49, `. 21, 22.
“ . . . whence the conclusion.”
The linear bijection Φ : C 0 (X; C) → Mc , Φ(µ0 ) = µ ( µ ∈ Mc ) was
introduced in the note for IV.49, `. 17–18, and shown to be continuous
for the topologies T and T 0 ∩ Mc , whence the continuity of the linear
INT IV.x50 extension of a measure. l p spaces §4

bijection ΦL0 : L0 → L for the topologies T ∩ L0 and T 0 ∩ L ; the desired
conclusion is that ΦL0 is a homeomorphism, thus we must show that the
inverse mapping (ΦL0 )−1 : L → L0 is continuous for T 0 ∩ L and T ∩ L0 .
Let us abbreviate Θ = (ΦL0 )−1 .
For every compact subset H of C (X; C) and every ε > 0 , write

W(H, ε) = {µ0 ∈ C 0 (X; C) : |µ0 (f )| 6 ε for all f ∈ H } ;

such sets form a fundamental system of neighborhoods of 0 in C 0 (X; C) for

the topology T , thus the sets

W(H, ε) ∩ L0 = {µ0 ∈ L0 : |µ0 (f )| 6 ε for all f ∈ H }

form a fundamental system of neighborhoods of 0 in L 0 for the topology

T ∩ L0 .
Given any compact set H in C (X; C) , choose h ∈ K (X; C) as in the
text, so that the set Hh = {f h : f ∈ H } is a strictly compact subset of
K (X; C) , and, when µ ∈ L , µ0 (f ) = µ0 (f h) = µ(f h) for all f ∈ H . Then
the set

V(Hh, ε) = {µ ∈ M (X; C) : |µ(f h)| 6 ε for all f ∈ H }

is a neighborhood of 0 in M (X; C) for the topology T 0 , so that the set

V(Hh, ε) ∩ L = {µ ∈ L : |µ(f h)| 6 ε for all f ∈ H }

is a neighborhood of 0 in L for T 0 ∩ L . But, when µ ∈ L , µ(f h) = µ0 (f )

for all f ∈ H , thus

V(Hh, ε) ∩ L = {µ ∈ L : |µ0 (f )| 6 ε for all f ∈ H }

= {µ ∈ L : µ0 ∈ W(H, ε) ∩ L0 }
= {Φ(µ0 ) : µ0 ∈ W(H, ε) ∩ L0 }

= {(ΦL0 )(µ0 ) : µ0 ∈ W(H, ε) ∩ L0 }



= (ΦL0 ) W(H, ε) ∩ L0 = Θ−1 W(H, ε) ∩ L0 ;


this shows that Θ−1 W(H, ε) ∩ L0 is a neighborhood of 0 in L for T 0 ∩ L ,
whence the continuity of Θ .
IV.49, `. −12 to −8.
“For, let H be a subset of C (X; C) consisting of functions that are
uniformly bounded on K ; there exists a number c > 0 such that |µ(f )| 6
c · kµk 6 ac for every function f ∈ H and every measure µ ∈ B , by virtue
§4 integrable functions and sets INT IV.x51

of (22); therefore B ⊂ acH◦ in the dual C 0 (X; C) of C (X; C) , which

proves the equicontinuity of B ”
We retain the notations introduced in the note for IV.48, `. 22–24,
thus maintaining the distinction between a measure µ on X with compact
support and its unique extension to a linear form µ 0 on C (X; C) continuous
for the topology τcc of compact convergence; as observed in the cited note,
µ 7→ µ0 is a linear bijection Mc → C 0 (X; C) , where Mc is the vector
space of all measures on X with compact support. In that light, since
B ⊂ Mc we interpret the assertions of the corollary to mean that the subset
B0 = {µ0 : µ ∈ B } of C 0 (X; C) acts equicontinuously on C (X; C) , and
that B0 is compact for the topology T .
The set H of the statement is a parameter at our disposal; we need
only one such set that “works”. For simplicity, let c > 0 (for even greater
simplicity let c = 1 ) and let

H = {f ∈ C (X; C) : |f (x)| 6 c for all x ∈ K }

= {f ∈ C (X; C) : pK (f ) 6 c } ,

where pK (f ) = sup |f (x)| defines a semi-norm on C (X; C) ; recalling that

x∈K
the topology of C (X; C) can be defined by such semi-norms, one knows that
pK is a continuous semi-norm and so H is a neighborhood of 0 in C (X; C) .
Moreover, H is a balanced subset of C (X; C) ; therefore, in the canonical
duality

(f, µ0 ) 7→ µ0 (f ) f ∈ C (X; C), µ0 ∈ C 0 (X; C) ,
the polar of H in C 0 (X; C) is given by

H◦ = {µ0 ∈ C 0 (X; C) : |µ0 (f )| 6 1 for all f ∈ H }

(TVS, II, §8, No. 4), H◦ is also balanced, and H ⊂ (H◦ )◦ (loc. cit., § 6,
No. 3).
Now, for every µ ∈ Mc and every f ∈ C (X; C) , f is µ-integrableR (that
is, f ∈ LC1 (X, µ) ), the inequality (22) holds, and µ 0 (f ) = µ(f ) = f dµ
defines, via Prop. 14, the element µ0 of C 0 (X; C) ; thus |µ0 (f )| = |µ(f )| 6
kµk · pS (f ) , where S is the support of µ . When µ ∈ B and f ∈ H , so that
S ⊂ K , kµk 6 a and pK (f ) 6 c , we have pS 6 pK and

|µ0 (f )| 6 kµk · pS (f ) 6 kµk · pK (f ) 6 ac ,

that is, |(ac)−1 µ0 (f )| 6 1 ; thus

(ac)−1 µ0 ⊂ H◦ for all µ0 ∈ B0 ,

INT IV.x52 extension of a measure. l p spaces §4

that is, (ac)−1 B0 ⊂ H◦ .

Since H is a neighborhood of 0 in C (X; C) and (H ◦ )◦ ⊃ H , (H◦ )◦
is also a neighborhood of 0 in C (X; C) , therefore H ◦ is equicontinuous on
C (X; C) (TVS, III, §3, No. 5, Prop. 7, (iii) ⇒ (i), with E = C (X; C) and
M = H◦ ), hence so is its subset (ac)−1 B0 , whence also B0 .
IV.49, `. −8 to −4.
“ . . . the fact that B is compact for T follows from the fact that, on B ,
T and the vague topology induce the same topology (Prop. 16 and Ch. III,
§1, No. 10, Prop. 17) and the fact that B is vaguely compact (Ch. III, §1,
No. 9, Cor. 2 of Prop. 15 and §2, No. 2, Prop. 6).”
As explained at the beginning of the preceding note, we interpret the
assertion to mean that B0 = {µ0 : µ ∈ B } is a compact subset of C 0 (X; C)
for the topology T . Let us write
L = {µ ∈ M (X; C) : Supp(µ) ⊂ K }
M = {µ ∈ M (X; C) : kµk 6 a } ;

thus B = L ∩ M . Since L is vaguely closed in M (X; C) (Ch. III, §2, No. 2,

Prop. 6) and M is vaguely compact (Ch. III, §1, No. 9, Cor. 2 of Prop. 15),
B = L ∩ M is vaguely compact, hence vaguely bounded (TVS, III, §1, No. 2,
Prop. 2). In the notations of Ch. III, §1, No. 10, Prop. 17, B is compact,
hence bounded, for the vague topology T 2 , therefore the induced topologies
T2 ∩ B and T3 ∩ B are identical, where T3 = T 0 is the topology of strictly
compact convergence; thus the topology T 0 ∩ B = T2 ∩ B is compact, that
is, B is compact for T 0 .
We know from Prop. 16 that the mapping µ 7→ µ 0 defines a homeomor-
phism L → L0 for T 0 ∩ L and T ∩ L0 (see the note for IV.49, `. 21, 22),
whence, since B ⊂ L , an induced homeomorphism B → B 0 for T 0 ∩ B
and T ∩ B0 . Since B is compact for T 0 , that is, the topology T 0 ∩ B is
compact, it follows that B0 is compact for T .
IV.50, `. 3–6.
“For, on the set B of measures ν such that Supp(ν) ⊂ Supp(µ) and
kνk 6 kµk , the topology induced by the vague topology is identical to the
topology induced by T , and the corollary therefore follows from Ch. III,
§2, No. 4, Cors. 2 and 3 of Th. 1.”
We retain the conventions of the preceding note. Setting K = Supp(µ)
and a = kµk , we are in the situation of Cor. 1; thus the set

B = {ν ∈ M (X; C) : Supp(ν) ⊂ Supp(µ) and kνk 6 kµk }

is compact for the topology T 0 on M (X; C) , and the induced (compact)

topology T 0 ∩ B coincides with T2 ∩ B , where T2 is the vague topology
§4 integrable functions and sets INT IV.x53

on M (X; C) . Let
C = {ν ∈ B : kνk = kµk }
C+ = {ν ∈ C : ν > 0 } .
By Cor. 2 of Th. 1 of Ch. III, §2, No. 4, µ belongs to the closure of C in
M (X; C) for the vague topology, hence for T 0 ; and when µ > 0 , µ belongs
to the closure of C+ in M (X; C) for the vague topology (loc. cit., Cor. 3),
hence for T 0 .
As observed in the preceding note, the mapping ν 7→ ν 0 defines a
homeomorphism B → B0 for the topologies T 0 ∩ B and T ∩ B0 , there-
fore µ0 belongs to the closure of C0 in C 0 (X; C) for T ; and when µ > 0 ,
µ0 belongs to the closure of (C+ )0 in C 0 (X; C) for T .
IV.53, `. −8, −7.
“ . . . for every compact subset S of X , µ and ν take on the same values
in K (X, S; C) .”
Let f ∈ K (X, S; C) . By Cor. 2, if ε > 0 there exists a finite linear
combination X
g= λ i ϕ Ki
i

such that kf − gk 6 ε , where the Ki are compact subsets of S , and it is

clear from the hypothesis that µ(g) = ν(g) . Then f − g is µ-integrable and
f − g = (f − g)ϕS , so

|f − g| = |f − g|ϕS 6 ε ϕS ,

therefore (No. 2, Prop. 2)

|µ(f ) − µ(g)| = |µ(f − g)| 6 |µ|(|f − g|) 6 ε |µ|(S) ,

where |µ|(S) < +∞ . Letting ε = 1/n , construct a sequence (g n ) of such

linear combinations satisfying kf − g n k 6 1/n ; then

1
|µ(f ) − µ(gn )| 6 |µ|(S) ,
n

thus µ(gn ) → µ(f ) . Similarly ν(gn ) → ν(f ) . But µ(gn ) = ν(gn ) for all n ,
whence µ(f ) = ν(f ) .
IV.55, `. 10.
“ . . . the I(K, U) form a base for the topology T ”
INT IV.x54 extension of a measure. l p spaces §4

The formula displayed in `. 7 shows that the set of all I(K, U) is closed
under finite intersections, therefore the set of unions of arbitrary families of
sets I(K, U) is the set of all open sets for T . In particular, for each M ⊂ X ,
the set of all I(K, U) for which K ⊂ M ⊂ U forms a fundamental system of
neighborhoods of M . Note that T exists independently of any function α .

IV.55, `. 12, 13.

“ . . . in P(X) , the set of compact subsets of X is dense.”
Let K be the set of all compact subsets of X . If O is any nonempty
open subset of P(X) for T , then O contains some set I(K, U) , whence
K ∈ I(K, U) ⊂ O ; thus, every nonempty open subset of P(X) contains an
element of K .

IV.55, `. 13.
“The condition (PCII ) expresses that Φ is dense in P(X) ”
If O is any nonempty open subset of P(X) then O contains some set
I(K, U) , and I(K, U) ∩ Φ 6= ∅ by (PCII ), whence O ∩ Φ 6= ∅ .

IV.55, `. 14, 15.

“ . . . condition (PMIV ) expresses that the function α is continuous on Φ
for the topology induced by T .”
Let M ∈ Φ . Given ε > 0 , choose K and U as in (PM IV ); then I(K, U)
is a neighborhood of M in P(X) for T , and

N ∈ I(K, U) ∩ Φ ⇒ |α(N) − α(M)| 6 ε ;

thus I(K, U) ∩ Φ is a neighborhood of M in Φ (for the induced topology

T ∩ Φ ) that is mapped by α into the ε-neighborhood of α(M) in R .

IV.55, `. 15–17.
“Th. 4 of No. 6 expresses that the function M 7→ µ(M) is continuous
on the clan of µ-integrable sets, for the topology induced by T .”
Here, µ can be any measure on X (not necessarily positive). Write Ψ
for the set of all µ-integrable subsets of X ; it follows from Props. 6 and 7
of No. 5 that Ψ is a clan, hence satisfies (PC I ). We are to show that the
function Ψ → C defined by A 7→ µ(A) (A ∈ Ψ) is continuous for the
topology T ∩ Ψ induced on Ψ by the topology T on P(X) .
Given M ∈ Ψ , let us show continuity at M . Given any ε > 0 , choose
K and G as in the cited Th. 4, so that M ∈ I(K, G) and |µ|(G --- K) 6 ε .
Then
N ∈ I(K, G) ∩ Ψ ⇒ |µ|(G --- N) 6 |µ|(G --- K) 6 ε ,
§4 integrable functions and sets INT IV.x55

and in particular |µ|(G --- M) 6 ε , whence

|µ(N) − µ(M)| 6 |µ(N) − µ(G)| + |µ(G) − µ(M)|

= |µ(G --- N)| + |µ(G --- M)|
6 |µ|(G --- N) + |µ|(G --- M) 6 2ε .

(This proves the asserted continuity, but Th. 4 says more.)

IV.55, `. 19, 20.
“We denote by Φ the set of subsets M ⊂ X such that α(N) tends
to a finite limit as N tends to M (for the topology T ) while remaining
in Φ ”
Φ is not to be confused with the closure of Φ in P(X) (it excludes the
elements of the closure where the limit is either infinite or fails to exist), the
closure of Φ being equal to P(X) by (PC II ).
IV.55, `. 20–22.
“ . . . we may then extend α in only one way to a continuous mapping α
of Φ into R (GT, I, §8, No. 5, Th. 1).”
To apply the cited Th. 1, we must check that Φ ⊂ Φ . One is assuming
here that Φ and α satisfy the conditions in the first sentence of Th. 5, and
that there exists a measure µ on X such that the sets of Φ are µ-integrable
and µ(M) = α(M) for all M ∈ Φ (since α is positive, it will turn out
that µ is necessarily positive). The problem is to show that µ is uniquely
determined by its values on Φ .
(1) One first observes that every µ-integrable set M ⊂ X belongs to Φ .
For, since Φ is dense in P(X) for T (by (PC II )), in particular M belongs
to the closure of Φ ; therefore if V is the filter of neighborhoods N of M
in P(X) , then its trace V ∩ Φ on Φ is a filter on Φ (GT, I, §6, No. 5,
Example). The assertion is that the filter base α(V ∩ Φ) on R converges
to an element of R (GT, I, §7, No. 3, Prop. 7).
Now, µ is continuous for T on the clan Ψ of µ-integrable sets (see
the note for `. 15–17); the continuity of µ at M ∈ Ψ means that the filter
base µ(V ∩ Ψ) converges to µ(M) ∈ R (GT, I, §7, No. 4, Prop. 9). Thus,
given any ε > 0 , there exists a neighborhood N of M in P(X) such that

|µ(N) − µ(M)| 6 ε for all N ∈ N ∩ Ψ ;

we are assuming that Φ ⊂ Ψ and that µ = α on Φ , therefore

|α(N) − µ(M)| 6 ε for all N ∈ N ∩ Φ ∈ V ∩ Φ ,

and we have shown that the filter base α(V ∩ Φ) converges to µ(M) ∈ R .
{Indeed, µ(M) ∈ R+ since α > 0 . Thus µ(M) > 0 for every µ-integrable
INT IV.x56 extension of a measure. l p spaces §4

set M . In particular, µ(K) > 0 for all compact K ⊂ X , therefore µ > 0

on E (Ψ) , where E (Ψ) is the R-linear span of the characteristic functions of
the sets in Ψ (No. 9, Prop. 18). It follows that if f ∈ K + (X) , there exists
a sequence (gn ) of functions in E (Ψ) such that 0 6 gn 6 f for all n and
gn → f uniformly (No. 10, Prop. 19); since Supp(g n ) ⊂ Supp(f ) , it follows
that µ(f ) = lim µ(gn ) > 0 (Ch. III, §1, No. 3), thus µ > 0 .}
n→∞
(2) By (1), we have Φ ⊂ Ψ ⊂ Φ ; in particular Φ ⊂ Φ , so we may speak
of extensions of α to Φ .
Since Φ is dense in P(X) for T , it is dense in Φ for T ∩ Φ ; and
since α : Φ → R is such that, for every M ∈ Φ , the limit

lim α(N)
N∈Φ, N→M

exists in R (by the definition of Φ ), it follows that α is uniquely extendible

to a continuous mapping α : Φ → R (GT, I, §8, No. 5, Th. 1), explicitly,

α(M) = lim α(N) for all M ∈ Φ .

N∈Φ, N→M

(3) If K is any compact set in X then, by (1), α(K) = µ(K) ; since µ

is characterized by its values for compact sets (No. 10, Cor. 3 of Prop. 19),
the uniqueness of α implies that of µ .
IV.55, `. −16, −15.
“Without assuming the existence of µ , we are now going to study the
set Φ and the extension α of α to Φ .”
For the rest of the proof, (PMIV ) is in force, consequently α is contin-
uous on Φ for the topology T ∩ Φ induced by T on Φ (see the note for
`. 14, 15). Then, for every M ∈ Φ , the limit

lim α(N)
N∈Φ, N→M

exists and is equal to α(M) , consequently Φ ⊂ Φ . The formula

α(M) = lim α(N) (M ∈ Φ)

N∈Φ, N→M

then provides a unique continuous extension of α to Φ for the topology

T ∩ Φ (GT, I, §8, No. 5, Th. 1).
IV.56, `. 8–10.
“For, let U be an open set belonging to Φ ; for every ε > 0 there
exists a compact set K ⊂ U such that, for every set M ∈ Φ satisfying
K ⊂ M ⊂ U , one has |α(U) − α(M)| 6 ε , whence |α(U) − α(K)| 6 ε ”
§4 integrable functions and sets INT IV.x57

Since U ∈ Φ , by the definition of Φ and α one has

lim α(N) = α(U) ∈ R .

N∈Φ, N→U

Thus, given any ε > 0 , there exists a neighborhood I(K, V) of U in P(X)

such that
M ∈ Φ ∩ I(K, V) ⇒ |α(M) − α(U)| 6 ε ;
since U ∈ I(K, V) , one has K ⊂ U ⊂ V , whence I(K, U) ⊂ I(K, V) , conse-
quently
M ∈ Φ ∩ I(K, U) ⇒ |α(M) − α(U)| 6 ε ,
that is,

(*) M ∈ Φ, K ⊂ M ⊂ U ⇒ |α(M) − α(U)| 6 ε .

By 3◦ , one has
α(K) = inf α(M) ;
M∈Φ, M⊃K

by the monotonicity of α established in 3 ◦ , the infimum is unchanged if we

restrict the M to be subsets of U —for, by (PC II ) there exists an N ∈ Φ
with K ⊂ N ⊂ U , and we may replace the M ’s by the M ∩ N ’s—therefore
the inequality |α(U) − α(K)| 6 ε is a consequence of (*).
IV.56, `. 10–12.
“ . . . if K0 is any compact set contained in U , then K ⊂ K ∪ K 0 ⊂ U ,
whence |α(U) − α(K ∪ K0 )| 6 ε and so α(U) > α(K ∪ K0 ) − ε > α(K0 ) − ε ”
With U , ε and K as in the preceding note, we also know that K∪K 0 ∈
Φ , hence
α(K ∪ K0 ) = lim 0
α(M) .
M∈Φ, M→K∪K

Given δ > 0 , choose a neighborhood I(K*, U*) of K ∪ K 0 such that

(**) M ∈ Φ ∩ I(K*, U*) ⇒ |α(M) − α(K ∪ K0 )| 6 δ .

In particular, K ∪ K0 ∈ I(K*, U*) , so

K* ⊂ K ∪ K0 ⊂ U* ;

replacing U* by U* ∩ U (which still contains K ∪ K 0 ), we can suppose that

U* ⊂ U .
By (PCII ) there exists a set M ∈ Φ with

K ∪ K0 ⊂ M ⊂ U* .
INT IV.x58 extension of a measure. l p spaces §4

Since K ⊂ K ∪ K0 ⊂ M ⊂ U* ⊂ U , by (*) of the preceding note one has

|α(M) − α(U)| 6 ε ,

and since K* ⊂ K ∪ K0 ⊂ M ⊂ U* , by (**) one has

|α(M) − α(K ∪ K0 )| 6 δ ,

therefore

|α(U) − α(K ∪ K0 )| 6 |α(U) − α(M)| + |α(M) − α(K ∪ K0 )| 6 ε + δ ,

concisely |α(U) − α(K ∪ K0 )| 6 ε + δ ; since δ was introduced after U , ε , K

and K0 were fixed, letting δ → 0 yields the first inequality of the statement.
It follows that α(U) > α(K ∪ K0 ) − ε > α(K0 ) − ε (the latter inequality
by the monotonicity of α for compact sets). In particular, α(K 0 ) 6 α(U)+ε ,
which is needed for the next note.
IV.56, `. 12, 13.
“ . . . α(U) is therefore indeed equal to the supremum of the num-
bers α(K) as K runs over the set of compact subsets of U .”
With notations as in the preceding note, the argument shows that if
an open set U belongs to Φ , then for every ε > 0 and every compact set
K0 ⊂ U , one has
α(K0 ) 6 α(U) + ε
(the K and M in the argument play auxiliary roles, and can be forgotten
here); fixing K0 and varying ε , one has α(K0 ) 6 α(U) , and since K0 is an
arbitrary compact subset of U , we conclude that

sup α(K0 ) 6 α(U) .

K0 compact, K0 ⊂U

But, by the note for `. 8–10, given any ε > 0 there exists a compact sub-
set K0 of U such that the difference between α(K 0 ) and α(U) is as small
as we like, therefore the preceding displayed inequality is in fact an equality.
IV.56, `. 17–19.
“ . . . by (PM00IV ), for every set M ∈ Φ such that K ⊂ M ⊂ U , there
exists a compact set K0 ⊂ M such that

α(M) 6 α(K0 ) + ε 6 b + ε ”

Let M be any set in Φ such that K ⊂ M ⊂ U (such M exist by (PC II )).

Then (PM00IV ) provides a compact set K0 ⊂ M such that α(N) > α(M) − ε
§4 integrable functions and sets INT IV.x59

for every N ∈ Φ with N ⊃ K0 ; but by 3◦ , K0 ∈ Φ and α(K0 ) is the infimum

of α(N) over all such N , therefore α(K 0 ) > α(M) − ε , whence the first
of the asserted inequalities—whereas the second follows from the definition
of b .
IV.56, `. 20.
“ . . . therefore b − ε 6 α(M) 6 b + ε , which proves that U ∈ Φ .”
We know that b − ε 6 α(K) by the choice of K ; but α(K) 6 α(M)
because K ⊂ M (see the definition of α(K) in 3 ◦ ), and α(M) 6 b + ε by
the preceding note.
Thus, given any ε > 0 one has found a neighborhood I(K, U) of U
in P(X) such that
M ∈ Φ ∩ I(K, U) ⇒ |α(M) − b| 6 ε ;
this shows that
lim α(M) = b ∈ R ,
M∈Φ, M→U

therefore U ∈ Φ , and α(U) = b (as one also knows from the proof of
“necessity” of the condition b < +∞ ).
IV.56, `. −8 to −3.
“The definition of Φ and α can now be transformed as follows (taking
into account (PCII )): in order that M ∈ Φ , it is necessary and sufficient
that, for every ε > 0 , there exist a compact set K and an open set U ∈ Φ
such that K ⊂ M ⊂ U and α(U) − α(K) 6 ε ; α(M) is, moreover, the
infimum of the α(U) for the open sets U ∈ Φ containing M , and the
supremum of the α(K) for the compact sets K ⊂ M .”
One is assuming that the hypotheses of Th. 5, including (PM IV ), are
fulfilled, and that, as shown in 3◦ :
(i) Every compact set K ⊂ X belongs to Φ , with α(K) ∈ R + defined
by the formula
α(K) = inf α(P) ;
P∈Φ, P⊃K

α is monotone, subadditive and additive on the set K of all compact subsets

of X , that is, α(K1 ) 6 α(K2 ) when K1 ⊂ K2 , and α(K1 ∪ K2 ) 6 α(K1 ) +
α(K2 ) , with equality when K1 , K2 are disjoint. Moreover, as shown in 4 ◦ :
(ii) Denoting by U the set of all open sets U in X for which
sup α(K) < +∞ ,
K∈K, K⊂U

an open set U in X belongs to Φ if and only if U ∈ U ; that is, writing U

for the set of all open sets in X , one has U ∩ Φ = U . Moreover,
α(U) = sup α(K) for all U ∈ U ,
K∈K, K⊂U
INT IV.x60 extension of a measure. l p spaces §4

and α is monotone and subadditive on U .

To summarize, K ⊂ Φ , U = U ∩ Φ , and α has the indicated properties
on K and U .
We are to show:
(A) A set M ⊂ X belongs to Φ if and only if it satisfies the following
condition: for every ε > 0 , there exist sets K ∈ K and U ∈ U such that
K ⊂ M ⊂ U and |α(U) − α(K)| 6 ε . Moreover:
For every M ∈ Φ , one has

(B) α(M) = sup α(K) = inf α(U) .

K∈K, K⊂M U∈U, U⊃M

Let us write Φ e for the set of all M ⊂ X that satisfy the condition
in (A); we are to show that Φe = Φ and that the formulas (B) hold for every
M ∈ Φ . The proof is organized into three parts: Φ ⊂ Φ e, Φe ⊂ Φ , and the
formulas (B).
Proof of Φ ⊂ Φe . Let M ∈ Φ and let ε > 0 . Since α(N) → α(M) as
N → M while remaining in Φ , there exists a neighborhood I(K, U) of M
in P(X) such that

(∗) N ∈ Φ ∩ I(K, U) ⇒ |α(N) − α(M)| 6 ε/2 .

We assert that U ∈ Φ (that is, U ∈ U ). To this end, let us show that, as K 0

runs over all compact subsets of U , the α(K 0 ) have a finite upper bound.
Let K0 ⊂ U be compact. Then K ∪ K0 ⊂ U , and by (PCII ) there exists a
set N ∈ Φ such that
K ∪ K0 ⊂ N ⊂ U .
Since K0 ⊂ N ∈ Φ , one has α(K0 ) 6 α(N) by 3◦ ; but K ⊂ N ⊂ U , therefore
α(N) 6 α(M)+ε/2 by (∗), whence α(K0 ) 6 α(M)+ε/2 , so that α(M)+ε/2
serves as an upper bound for the α(K0 ) (K0 ∈ K , K0 ⊂ U ), which proves
that U ∈ Φ . Then, by 4◦ , α(U) = sup α(K0 ) over all such K0 , whence
α(U) 6 α(M) + ε/2 .
Since K ⊂ U ∈ Φ , it follows from 3◦ that α(K) 6 α(U) . On the other
hand, it follows from (∗) that |α(K) − α(M)| 6 ε/2 ; for, one knows from 3 ◦
that α(K) is the infimum of the α(P) over all P ∈ Φ such that P ⊃ K , and
by the monotonicity of α one can restrict P to be a subset of any fixed N
occurring in (∗) (an argument made in the note for `. 8–10). Then

0 6 α(U) − α(K) = [α(U) − α(M)] + [α(M) − α(K)] 6 ε/2 + ε/2 ,

e , and so Φ ⊂ Φ
concisely α(U) − α(K) 6 ε , which proves that M ∈ Φ e.
§4 integrable functions and sets INT IV.x61

e ⊂ Φ . First, a useful lemma:

Proof of Φ
Lemma 1. If K ⊂ N ⊂ U , where N ∈ Φ , K is compact and U is an
open set with U ∈ Φ (that is, U ∈ U ), then α(K) 6 α(N) 6 α(U) .
Proof. That α(K) 6 α(N) was shown in 3 ◦ .
To prove that α(N) 6 α(U) we apply (PM IV ) to N : given any ε > 0 ,
there exists a neighborhood I(K0 , U0 ) of N in P(X) such that

N0 ∈ Φ ∩ I(K0 , U0 ) ⇒ |α(N0 ) − α(N)| 6 ε .

But I(K, U) is also a neighborhood of N , therefore so is

I(K ∪ K0 , U ∩ U0 ) = I(K, U) ∩ I(K0 , U0 ) ;

replacing I(K0 , U0 ) by I(K ∪ K0 , U ∩ U0 ) , we can suppose that

K ⊂ K0 ⊂ N ⊂ U 0 ⊂ U ,

and since U ∈ Φ it is clear from the criterion in 4 ◦ that also U0 ∈ Φ .

We assert that |α(U0 ) − α(N)| 6 ε . We know from 4◦ that
α(U0 ) = sup α(K0 ) , where K0 runs over all compact subsets of U0 ; since
also K0 ⊂ U0 , whence K0 ⊂ K0 ∪ K0 ⊂ U0 , and since α is monotone for
compact sets, α(U0 ) is also the supremum of the α(K0 ) where K0 is com-
pact and K0 ⊂ K0 ⊂ U , thus it will suffice to show that |α(K 0 ) − α(N)| 6 ε
for every such K0 .
Now, given such a K0 , we know from 3◦ that α(K0 ) = inf α(N0 ) , where
N ∈ Φ and N0 ⊃ K0 . But K0 ⊂ U0 , so by (PCII ) there exists an N00 ∈ Φ
0

with K0 ⊂ N00 ⊂ U0 . Then for every N0 ∈ Φ with N0 ⊃ K0 , one has

K0 ⊂ N0 ∩ N00 ⊂ U0 ; thus, by the monotonicity of α , it suffices to consider
only N0 ∈ Φ with K0 ⊂ N0 ⊂ U0 . For such an N0 , we need only show that
|α(N0 ) − α(N)| 6 ε . Indeed,

K ⊂ K 0 ⊂ K0 ⊂ N0 ⊂ U0 ,

thus N0 ∈ Φ ∩ I(K0 , U0 ) , therefore |α(N0 ) − α(N)| 6 ε by the choice of K0

and U0 , which completes the verification of |α(U 0 ) − α(N)| 6 ε .
To summarize: given any ε > 0 , we have found an open set U 0 ∈ Φ
such that
N ⊂ U0 ⊂ U and |α(U0 ) − α(N)| 6 ε .
With ε = 1/n (n = 1, 2, 3, . . .) choose an open set U n ∈ Φ such that
N ⊂ Un ⊂ U and |α(Un ) − α(N)| 6 1/n . Then α(Un ) → α(N) ; but
α(Un ) 6 α(U) because α is monotone on U , so passage to the limit yields
α(N) 6 α(U) and the lemma is proved. ♦
INT IV.x62 extension of a measure. l p spaces §4

Assuming now that M ∈ Φ e (that is, M satisfies the condition in (A)),

let us show that M ∈ Φ . By assumption, given any ε > 0 , there exists a
neighborhood I(K, U) of M with K compact and U an open set belonging
to Φ . We assert that if N, N0 is any pair of elements of Φ ∩ I(K, U) , then

|α(N) − α(N0 )| 6 ε ;

indeed, from K ⊂ N ⊂ U and K ⊂ N0 ⊂ U , the lemma yields

α(N) ∈ [α(K), α(U)] and α(N0 ) ∈ [α(K), α(U)] ,

whence |α(N) − α(N0 )| 6 α(U) − α(K) 6 ε .

This shows that the image, under α , of the trace on Φ of the neighbor-
hood filter of M , is Cauchy in R , hence is convergent to an element of R .
In other words M ∈ Φ , which completes the proof that Φ e ⊂ Φ , and hence
e
that Φ = Φ .
Proof of the formulas (B). The formulas are an easy consequence of the
following generalization of Lemma 1:
Lemma 2. If K ⊂ M ⊂ U , where K is compact, M ∈ Φ , and U ∈ Φ is
open, then α(K) 6 α(M) 6 α(U) .
Proof. If N ∈ Φ ∩ I(K, U) , then α(K) 6 α(N) 6 α(U) by Lemma 1,
thus


(∗∗) α Φ ∩ I(K, U) ⊂ [α(K), α(U)] .

Since Φ is dense in P(X) , every neighborhood of M in P(X) intersects Φ ,

and since I(K, U) is a neighborhood of M , clearly every neighborhood of M
intersects Φ∩I(K, U) ; it follows that if I(K 0 , U0 ) is any neighborhood of M ,
then Φ ∩ I(K0 , U0 ) (which is a basic neighborhood of M in Φ ) intersects
Φ ∩ I(K, U) , thus M belongs to the closure of Φ ∩ I(K, U) in Φ . Since
α : Φ → R is continuous and extends α , it follows from (∗∗) that (the long
overbars indicate closure in R )



α(M) ∈ α Φ ∩ I(K, U) = α Φ ∩ I(K, U) ⊂ [α(K), α(U)] ,

that is, α(K) 6 α(M) 6 α(U) . ♦

Given any M ∈ Φ , define

r = inf{α(U) : M ⊂ U ∈ Φ , U open} ,
s = sup{α(K) : K ⊂ M , K compact } ;

we are to show that r = s = α(M) .

§4 integrable functions and sets INT IV.x63

Since M ∈ Φ = Φ e , we know that M satisfies the condition in (A):

given any ε > 0 , there exist K ∈ K and U ∈ U such that K ⊂ M ⊂ U and
α(U) − α(K) 6 ε ; by the above lemma,

α(K) 6 α(M) 6 α(U) .

Obviously α(K) 6 s and r 6 α(U) .

For every compact subset K0 of M , one has K0 ⊂ M ⊂ U , therefore
α(K0 ) 6 α(M) by Lemma 2; consequently s 6 α(M) . On the other hand,
for every set U0 ∈ U such that M ⊂ U0 , one has K ⊂ M ⊂ U0 , therefore
α(M) 6 α(U0 ) by Lemma 2; consequently α(M) 6 r . Thus

α(K) 6 s 6 α(M) 6 r 6 α(U) ,

whence 0 6 r − s 6 α(U) − α(K) 6 ε . Since ε is arbitrary, r − s = 0 ; thus

s 6 α(M 6 r with s = r , whence equality throughout.
An immediate consequence of either of these formulas for α(M) : α is
monotone on Φ .
IV.57, `. 7, 8.
“By the foregoing, we have α(U) 6 α(K) + α(U --- K) .”
Since U --- K is open and U --- K ⊂ U ∈ Φ , U --- K ∈ Φ by the
criterion in 4◦ ; apply the preceding assertion (‘conditional subadditivity’) to
U = K ∪ (U --- K) .
IV.57, `. 11.
“ . . . α(K) + α(U --- K) 6 α(U) .”
By the previously displayed formula, α(K 0 ) 6 α(U) − α(K) for all
compact sets K0 ⊂ U --- K ⊂ Φ , whence α(U --- K) 6 α(U) − α(K) .
IV.57, `. −16.
“ . . . K00 ⊂ M ∩ { N ⊂ U00 . . . ”
From K ⊂ M ⊂ U and { U0 ⊂ { N ⊂ { K0 , one has

K00 = K ∩ { U0 ⊂ M ∩ { N ⊂ U ∩ { K0 = U00 .

IV.57, `. −16, −15.

“ . . . U00 --- K00 is contained in the union of U∩ { K and U0 ∩ { K0 . . . ”
For,

U00 --- K00 = U00 ∩ { (K ∩ { U0 ) = U00 ∩ ( { K ∪ U0 )

= (U00 ∩ { K) ∪ (U00 ∩ U0 )
= (U ∩ { K0 ∩ { K) ∪ (U ∩ { K0 ∩ U0 )
⊂ (U ∩ { K) ∪ (U0 ∩ { K0 ) ,
INT IV.x64 extension of a measure. l p spaces §4

whence α(U00 --- K00 ) 6 α(U --- K) + α(U0 --- K0 ) by the subadditivity of α
for open sets belonging to Φ .
IV.57, `. −12, −11.
“ . . . M ∪ N belongs to Φ .”
This completes the proof that Φ is a clan (No. 9, Prop. 17). Conse-
quently α is subadditive on Φ by the ‘conditional subadditivity’ (p. IV.56,
`. −2, −1) proved earlier.
IV.57, `. −9, −8.
“ . . . since ε is arbitrary, we have α(M ∪ N) = α(M) + α(N) .”
Since ε is arbitrary, we have α(M ∪ N) > α(M) + α(N) , and the reverse
inequality holds by the subadditivity of α .
IV.57, `. −3 to −1.
“Since β is positive, |β(f )| 6 α(K) · kf k for every function f ∈ E (Φ)
whose support is contained in K ”
For such a function f , −kf k ϕK 6 f 6 kf k ϕK , whence

−kf k β(ϕK ) 6 β(f ) 6 kf k β(ϕK ) ,

where β(ϕK ) = α(K) .

IV.58, `. 1, 2.
“ . . . it may therefore be extended to a positive continuous linear form
β K on G (K) .”
A detail to be checked: the positivity of β .
Recall that E (Φ) is a Riesz space for the pointwise ordering of functions
(No. 9, second paragraph following Def. 4); so is its linear subspace

E (Φ) ∩ F(X, K; R)

(the functions in E (Φ) whose support is contained in K ), whose closure G (K)

in the Banach space B(X; R) of bounded functions on X (TVS, I, §1, No. 4,
Examples) is also a Riesz space. Note that E (Φ)∩F(X, K; R) = E (Φ)∩G (K).
Let g ∈ G (K) and choose a sequence fn ∈ E (Φ) ∩ G (K) such that
kfn − gk → 0 . Then |g| ∈ G (K) and k |fn | − |g| k → 0 , therefore

β K (|g|) = lim β K (|fn |) = lim βK (|fn |) > 0 .

n n

In particular, if g > 0 , that is, g = |g| , then β K (g) > 0 .

IV.58, `. 3, 4.
“ . . . the restriction of β K1 to G (K) is identical to β K ”
§4 integrable functions and sets INT IV.x65

If g ∈ G (K) and fn ∈ E (Φ) ∩ G (K) with kfn − gk → 0 , then also

g ∈ G (K1 ) and fn ∈ E (Φ) ∩ F(X, K1 ; R) = E (Φ) ∩ G (K1 ) , therefore
β K1 (g) = lim β(fn ) = β K (g) .
n

IV.58, `. 8, 9.
“ . . . the restriction to K of the positive linear form β is therefore a
positive measure µ .”
Every positive linear form on K (X; R) is a positive measure (Ch. III,
§1, No. 5, Th. 1).
IV.58, `. 10.
“ there exists an open set U ∈ Φ such that K ⊂ U , µ(U) 6 µ(K) + ε ”
There exists an open set V with K ⊂ V ⊂ V ⊂ U and V compact;
replacing U by V , we can suppose that U is relatively compact, thereby
assuring that every function whose support is contained in U has compact
support.
IV.58, `. 13.
“Then µ(K) 6 µ(f ) 6 µ(U) 6 µ(K) + ε ”
To justify the notation µ(f ) at this stage, we must assume that f has
compact support; this is assured by arranging that U be relatively compact
(or by constructing f to have compact support). Moreover, the equality
µ(f ) = β(f ) will shortly be needed.
{Absent this precaution, we do not know that f is µ-integrable; of
course, after it has been shown that every open set belonging to Φ is
µ-integrable, a bounded continuous function with support contained in such
a set will be integrable (No. 4, Prop. 5).}
IV.58, `. 15.
“ . . . we see that |µ(K) − α(K)| 6 ε ”
From the above inequalities, we extract (noting that µ(f ) = β(f ) by
the definition of µ )
µ(K) 6 µ(f ) 6 µ(K) + ε
and α(K) 6 β(f ) = µ(f ) 6 α(K) + ε , that is,

−α(K) − ε 6 −µ(f ) 6 −α(K) .

Adding term-by-term the displayed inequalities, we get

[µ(K) − α(K)] − ε 6 0 6 [µ(K) − α(K)] + ε ;

adding µ(K) − α(K) throughout then yields −ε 6 µ(K) − α(K) 6 ε .

INT IV.x66 extension of a measure. l p spaces §4

IV.58, `. 18, 19.

“ . . . the open sets belonging to Φ are none other than the µ-integrable
open sets”
Write K for the set of all compact subsets of X . One knows that
µ(K) = α(K) for all K ∈ K . Let U be an open subset of X and let

b= sup α(K) = sup µ(K) .

K∈K, K⊂U K∈K, K⊂U

By 4◦ ,
U∈Φ ⇔ b < +∞ ,
in which case α(U) = b . On the other hand, by Prop. 10 of No. 6,

U is µ-integrable ⇔ µ*(U) < +∞ ,

in which case µ(U) = µ*(U) by definition. But µ*(U) = b (No. 6, Cor. 4

of Th. 4), thus

U is µ-integrable ⇔ b < +∞ ⇔ U ∈ Φ,

in which case µ(U) = b = α(U) .

IV.58, `. 21.
“ . . . the µ-integrable sets are the sets of Φ ”
Write K for the set of all compact subsets of X , and let M be any
subset of X .
As shown in 5◦ , M ∈ Φ if and only if it satisfies the following condition:
(A) For every ε > 0 , there exist a compact set K and an open set
U ∈ Φ such that K ⊂ M ⊂ U and α(U --- K) 6 ε .
By the foregoing discussion, we know that an open set U belongs to Φ
if and only if it is µ-integrable, and that µ(U) = α(U) for all open sets
U ∈ Φ . Thus the condition (A) may be expressed in terms of µ as follows:
(A0 ) For every ε > 0 , there exist a compact set K and an integrable
open set U such that K ⊂ M ⊂ U and µ(U --- K) 6 ε .
But (A0 ) holds if and only if M is µ-integrable (No. 6, Th. 4), thus

M∈Φ ⇔ M is µ-integrable.

Moreover, when M ∈ Φ one has, as shown in 5 ◦ ,

α(M) = sup α(K) ;

K∈K, K⊂M
§4 integrable functions and sets INT IV.x67

whereas when M is µ-integrable,

µ(M) = sup µ(K) ,

K∈K, K⊂M

as observed in No. 6, Cor. 1 of Th. 4 (see the note for IV.45, `. 1–3). Since
µ = α on K , one concludes that µ(M) = α(M) .
IV.58, `. 22.
“ µ*(U) = sup α(M) for every open set U ”
M∈Φ, M⊂U
Every M ∈ Φ is µ-integrable and α(M) = α(M) = µ(M) = µ*(M) ;
if, moreover, M ⊂ U , then α(M) = µ*(M) 6 µ*(U) , so the supremum in
question is 6 µ*(U) .
Write K for the set of all compact subsets of X . By Cor. 4 of Th. 4
of No. 6,

(∗) µ*(U) = sup µ(K) .

K∈K, K⊂U

But if K ∈ K and K ⊂ U , by (PCII ) there exists a set M ∈ Φ such that

K ⊂ M ⊂ U , therefore µ*(K) 6 µ*(M) 6 µ*(U) ; since K and M are
µ-integrable, and µ(M) = α(M) , one has

µ(K) 6 α(M) 6 µ*(U) ,

whence it is clear from (∗) that µ*(U) = sup α(M) .

M∈Φ, M⊂U

IV.58, `. −6, −5.

“The conditions (PCI ), (PCII ), (PMI ), (PMII ), (PMIII ) and (PM00IV ) are
then satisfied.”
Recall that Ψ is a tribe, namely, the tribe generated by the set Φ of
all compact subsets of X (GT, IX, §6, No. 3, Def. 4). The property (i) of β
is called complete additivity (last sentence of No. 5); one notes the following
consequences of (i) and (ii):
S∞

a) β(∅) = 0 . For, β(∅) < +∞ by (ii), and β(∅) = β ∅ =
k=1
P
∞
β(∅) by (i).
k=1
b) β is (finitely) additive (in the sense of No. 9, Def. 5), by a) and (i).
c) β is monotone. For, if B1 ⊂ B2 then B2 = B1 ∪ (B2 --- B1 ) is
a disjoint union, whence, by b), β(B 2 ) = β(B1 ) + β(B2 --- B1 ) > β(B1 ) ;
if, moreover, β(B1 ) is finite, then β(B2 --- B1 ) = β(B2 ) − β(B1 ) ( β is
‘conditionally subtractive’).
INT IV.x68 extension of a measure. l p spaces §4

d) β is subadditive. For,

β(B1 ∪ B2 ) = β(B1 ) + β(B2 --- B1 ) 6 β(B1 ) + β(B2 )

by b) and c).
S
∞
e) If (Bk )k>1 is an increasing sequence in Ψ and B = Bk , then
k=1

β(B) = sup β(Bk ) = lim β(Bk ) in R .

k k

For, setting C1 = B1 and Cn = Bn --- Bn−1 for n > 2 , the Cn are pairwise
S
∞
disjoint and Cn = B , therefore
n=1

∞
X n
X [
n 
β(B) = β(Cn ) = sup β(Ck ) = sup β Ck
n n
n=1 k=1 k=1
= sup β(Bn ) = lim β(Bn )
n n

by the theorem on monotone limits (GT, IV, §5, No. 2, Th. 2).
 S
∞  P∞
f ) For any sequence (Bk )k>1 in Ψ , β Bk 6 β(Bk ) (‘complete
k=1 k=1
S
n
subadditivity’). For, writing Cn = Bk for n = 1, 2, 3, . . . and C =
k=1
S
∞
Bk , by e) and d) one has
k=1

n
X ∞
X
β(C) = sup β(Cn ) 6 sup β(Bk ) = β(Bk ) .
n n
k=1 k=1

g) If (Cn )n>1 is a decreasing sequence in Ψ such that β(C 1 ) < +∞ ,

 T∞  T
∞
then β Cn = inf β(Cn ) = lim β(Cn ) . For, writing C = Cn
n=1 n n n=1
and Bn = C1 --- Cn for all n , the sequence (Bn ) in Ψ is increasing and
S
∞
C1 --- C = Bn , so by c) and e) one has
n=1

β(C1 ) − β(C) = β(C1 --- C) = sup β(Bn )

n
= sup[β(C1 ) − β(Cn )] = β(C1 ) − inf β(Cn ) .
n n
§4 integrable functions and sets INT IV.x69

As for the assertions at hand concerning the restriction α of β to the

set Φ of compact subsets of X :
(PCI ) If K1 , K2 are compact, then so are K1 ∪ K2 and K1 ∩ K2 .
(PCII ) Let M = K .
(PMI ) Follows from c) above.
(PMII ) Follows from d) above.
(PMIII ) Follows from b) above.
(PM00IV ) Let K = M .
Thus, to apply Theorem 5, it remains only to verify the condition
(PM0IV ); this is the condition whose verification requires that X have a
countable base.
IV.58, `. −5 to −3.
“Then K is the intersection of a decreasing sequence (U 1 , U2 , . . .) of
relatively compact open sets of X (GT, IX, §2, No. 5, Prop. 7).”
Since X is locally compact and has a countable base, it is metrizable
(GT, IX, §2, No. 9, Cor. of Prop. 16). Let (W n )n>1 be a sequence of open
T
∞
sets such that K = Wn (loc. cit., No. 5, Prop. 7). For each n , there
n=1
exists a relatively compact open set V n such that K ⊂ Vn ⊂ Wn , and the
open sets Un = V1 ∩ · · · ∩ Vn meet the requirements.
Let us proceed directly to the objective: the verification of the con-
dition (PM0IV ). Given any element of Φ —in other words any compact
set K —and any ε > 0 , we seek an open set U such that for every set
N ∈ Φ contained in U —that is, for every compact subset N of U —one has
β(N) 6 β(K) + ε .
T
∞
With K = Un as above, one knows that β(Un ) < ∞ ; for, all open
n=1
sets and closed sets are Borel sets, and β(U n ) 6 β(Un ) < +∞ by the
hypothesis (ii) of the Corollary. Therefore β(U n ) → β(K) by item g) in the
note for IV.58, `. −6, −5. Choose an index m such that β(U m ) 6 β(K)+ε .
Then, for every compact N ⊂ Um , one has

β(N) 6 β(Um ) 6 β(K) + ε ,

thus U = Um meets the requirements of (PM0IV ).

IV.59, `. 1.
“This proves that the condition (PM 0IV ) is satisfied.”
See the preceding note.
IV.59, `. 3, 4.
“Since every open set U of X is the union of an increasing sequence of
compact subsets, we have µ*(U) = β(U) .”
INT IV.x70 extension of a measure. l p spaces §4

One is assuming that X is locally compact and has a countable base,

therefore X is countable at infinity (GT, IX, §2, No. 9, Cor. of Prop. 16).
A subset of U is compact in X if and only if it is compact in U for the
induced topology. {This is immediate from the definition of compact subset
(GT, I, §9, No. 3, Def. 2) and the transitivity of induced topologies (GT, I,
§3, No. 1).} Since U is a neighborhood of each of its points, it follows that
the subspace U is itself locally compact and has a countable base, hence is
S
∞
countable at infinity; therefore U = Kn with the Kn compact in U ,
n=1
hence also in X . Replacing Kn by K1 ∪ · · · ∪ Kn , one can suppose that the
sequence (Kn ) is increasing. Then

µ*(U) = sup µ(Kn ) = sup β(Kn ) = β(U) ,

n n

the first equality by §1, No. 2, Prop. 7, and the last by item e) of the note
for IV.58, `. −6, −5.
IV.59, `. 6, 7.
“ . . . if B is an element of Ψ contained in L , then B is µ-integrable”
Let S be the tribe of µ-integrable subsets of L . The set of all A ⊂ X
such that A∩L ∈ S is easily seen to be a tribe that contains every compact
subset of X , hence contains the tribe Ψ of all Borel sets in X ; that is,
B ∩ L is µ-integrable for every B ∈ Ψ . In particular, if B ∈ Ψ is contained
in L , then B = B ∩ L is µ-integrable.
IV.59, `. 8, 9.
“Since β(U) = µ*(U) and β(K) = µ(K) , we see that |µ*(B)−β(B)| 6
2ε .”
For,

(∗) µ*(B) − β(B) = [µ*(B) − µ(K)] + [µ(K) − β(B)] ,

where

0 6 µ*(B) − µ*(K) = µ*(B) − µ(K) 6 µ*(U) − µ(K) 6 ε

and, since β(B) − µ(K) = β(B) − β(K) = β(B --- K) > 0 ,

0 6 β(B) − µ(K) = β(B --- K) 6 β(U --- K)

= β(U) − β(K) = µ*(U) − µ(K) 6 ε ,

therefore, taking absolute values in (∗) yields |µ*(B) − β(B)| 6 ε + ε ; since

ε is arbitrary, µ*(B) = β(B) .
§4 integrable functions and sets INT IV.x71

Since L is an arbitrary compact subset of X , the argument shows that

µ*(B) = β(B) for every relatively compact Borel set B .
IV.59, `. 10, 11.
“ . . . every Borel set C of X is the union of a sequence of pairwise
disjoint, relatively compact Borel sets”
S
∞
Write X = Kn with (Kn ) an increasing sequence of compact sets.
n=1
Then ∞
[

X = K1 ∪ Kn --- Kn−1
n=2

expresses the Borel set X as such a union, so the decomposition

∞
[  
C = (C ∩ K1 ) ∪ C ∩ Kn --- Kn−1
n=2

meets the requirements.

IV.59, `. 11.
“ . . . whence β(C) = µ*(C) .”
S
∞
Write C = Bn , with (Bn ) a sequence of pairwise disjoint, relatively
n=1
compact Borel sets. As shown above, the B n are µ-integrable and β(Bn ) =
µ*(Bn ) = µ(Bn ) , so the assertion follows from the complete additivity of µ
(No. 5, Prop. 6) and β .
IV.59, `. 11, 12.
“The uniqueness of µ follows at once from Th. 5.”
Suppose also ν is a measure on X such that ν*(B) = β(B) for all
B ∈ Ψ . In particular, for every compact set K one has

ν(K) = ν*(K) = β(K) = µ*(K) = µ(K) ,

 
thus ν Φ = µΦ = α , so µ = ν by the uniqueness part of Th. 5.
INT IV.x72 extension of a measure. l p spaces §5

§5. MEASURABLE FUNCTIONS AND SETS

IV.60, `. 6, 7.
“ . . . therefore the restriction of f to H is continuous.”
GT, I, §3, No. 2, Prop. 4.
IV.60, `. 17.
P
∞
“ |µ|(K --- K0 ) 6 ε/2n ”
n=1
Write An = K --- Kn and A = K --- K0 , and cite §4, No. 5, Cor. of
Prop. 8.
IV.60, `. −14 to −11.
“ . . . it comes to the same to say that a measurable set A is a set such
that, for every compact set K , there exist a negligible set N ⊂ K and a
partition (Kn ) of K --- N formed by a sequence of compact sets each of
which is contained either in K ∩ A or in K ∩ { A .”
Suppose A is measurable, that is, ϕA is measurable, and let K be
any compact set in X . By Def. 1, there exist a negligible set N ⊂ K and
a partition
 (Kn ) of K --- N into a sequence of compact sets K n such that
ϕA Kn is continuous for each n . But 0, 1 are the only possible values
of ϕA ; let
−1 −1
K0n = Kn ∩ ϕA (1) , K00n = Kn ∩ ϕA (0) .

Then K0n , K00n partition Kn into subsets that are closed (by the continuity
of ϕA Kn ) hence compact, with K0n ⊂ A and K00n ⊂ { A . Thus N , together
with the compact sets (K0n ), (K00n ) satisfy the condition in the assertion.
The converse is immediate from Def. 1, since ϕ A is constant (hence
continuous) on each of the sets K ∩ A and K ∩ { A .
IV.60, `. −7.
“The condition is necessary . . . ”
Suppose A is measurable and let K be any compact set in X . As
noted following Def. 2, there exists a partition

∞
[ ∞
[
K=N∪ K0n ∪ K00n ,
n=1 n=1
§5 measurable functions and sets INT IV.x73

where N is negligible, the K0n and K00n are compact, and K0n ⊂ A ,
K00n ⊂ { A for all n . Then A ∩ K00n = ∅ for all n , so
∞
[
(∗) A ∩ K = (A ∩ N) ∪ K0n ;
n=1

A ∩ N is negligible, hence integrable with |µ|(A ∩ N) = 0 ; and, for all n ,

n
X [
n 
|µ|(K0i ) = |µ| K0i 6 |µ|(K) < +∞ ,
i=1 i=1

S
∞
therefore K0n is integrable (§4, No. 5, Cor. of Prop. 8). Thus each of the
n=1
two terms on the right side of (∗) is integrable, hence so is A ∩ K .
IV.60, `. −5.
“The condition is sufficient . . . ”
Let K be any compact set in X . By hypothesis A ∩ K is integrable,
hence there exists a partition
∞
[
0
A∩K=N ∪ K0n ,
n=1

where N0 is negligible and the K0n are compact (§4, No. 6, Cor. 2 of Th. 4).
But ( { A) ∩ K = K --- A ∩ K is also integrable (§4, No. 5, Prop. 7), so there
is also a partition
∞
[
00
{A∩K = N ∪ K00n
n=1
00
with N negligible and the K00n compact. Then N = N0 ∪ N00 is negligible
and
∞
[ ∞
[
K = (A ∩ K) ∪ ({ A ∩ K) = N ∪ K0n ∪ K00n ,
n=1 n=1

where the K0n , K00n

(n = 1, 2, 3, . . .) partition K --- N into compact sets such
that K0n ⊂ A and K00n ⊂ { A for all n , therefore A is measurable by the
remark following Def. 2.
It follows from Prop. 3 that every integrable set A is measurable; for,
by Prop. 7, 2◦ of §4, No. 5, A ∩ K is integrable for every compact set K .
IV.60, `. −2.
“The open sets and the closed sets are measurable.”
INT IV.x74 extension of a measure. l p spaces §5

The measurability of closed sets is obvious from Prop. 3; as for open

sets, it suffices to prove the following:

A measurable ⇒ { A measurable.

For, if A is measurable and K is any compact set in X , then { A∩K =

K --- A ∩ K is the difference of two integrable sets, hence is integrable (§4,
No. 5, Prop. 7).
Prop. 3 yields a brief proof of the following:
“Corollary 3.” —If µ is a measure on the locally compact space X ,
then:
(i) The set of µ-measurable subsets of X is a tribe.
(ii) Every Borel set in X is µ-measurable.
These assertions are proved in No. 4 below as corollaries of Egoroff’s
theorem (Cors. 2 and 3 of Th. 2). For the definition of a tribe, see the
footnote on p. IV.50. T
(i) If (An ) is a sequence of measurable sets, then the set A = An
n
is
T measurable. For, if K is any compact subset of X , the set A ∩ K =
(An ∩ K) is the intersection of a sequence of integrable sets, hence is
n
integrable (§4, No. 5, Prop. 7). It follows by complementation that the set
S
An is also measurable.
n
(ii) The tribe of measurable sets contains every open set in X by Cor. 1,
hence it contains the tribe they generate (GT, IX, §6, No. 3, Def. 4).
IV.61, `. 3, 4.
“ . . . it suffices to verify that every relatively compact Souslin set A is
µ-integrable.”
Let A be a Souslin subset of the metrizable locally compact space X ,
and let µ be any measure on X . We are to show that A is µ-measurable;
given any compact subset K of X , it suffices by Prop. 3 to show that A ∩ K
is µ-integrable. Since K is a metrizable compact subspace of X , it is Polish
(GT, IX, §6, No. 1, Cor. of Prop. 2) hence Souslin (loc. cit., No. 2), therefore
A ∩ K is a Souslin subset of X (loc. cit., Prop. 8) and is obviously relatively
compact; thus, one is reduced to showing that a relatively compact Souslin
subset of X is µ-integrable.
IV.61, `. 4, 5.
“ . . . such a set A is capacitable for |µ|* (GT, IX, §6, No. 9, Th. 5).”
That |µ|* is a capacity on X is noted in §4, No. 6, Cor. 2 of Prop. 10.
(See the note for IV.43, `. −13, −12, where it is noted that the definition of
§5 measurable functions and sets INT IV.x75

‘capacity’ is different in GT and TG, but that with either definition, |µ|* has
the desired properties. In both cases, the property asserted for A is

|µ|*(A) = sup{|µ|*(K) : K ⊂ A , K compact} ,

established in GT, IX, §6, No. 9, Th. 5 and TG, IX, §6, No. 10, Th. 6,
respectively.
For the purposes of the proof at hand, one need read no further; the rest
of this note is devoted to the comparative analysis of the two definitions.
In GT, IX, §6, No. 9, Def. 8, a capacity on (a Hausdorff space) X is
defined to be a function f : P(X) → R satisfying the following axioms:
(CAI ) If A ⊂ B , then f (A) 6 f (B) .
(CAII ) If (An ) is any increasing sequence of subsets of X , then
[ 
f An = sup f (An ) .
n
n

(CAIII ) If (Kn ) is any decreasing sequence of compact subsets of X ,

then \ 
f Kn = inf f (Kn ) .
n
n

Whereas in TG, IX, §6, No. 10, Def. 9, a function f : P(X) → R is

called a ‘capacity’ if it merely satisfies (CA I ) and (CAII ); to compensate
for the absence of (CAIII ), a capacity continuous on the right (for short,
a CR-capacity is required in addition to satisfy the axiom
(CA0III ) For every compact subset K of X and every number a > f (K) ,
there exists an open set U containing K such that f (U) < a .
The prime 0 in (CA0III ) does not appear in TG but is added here to avoid
confusion. It is shown (TG, loc. cit., Remark ) that every CR-capacity is a
capacity in the sense of GT (for short, a GT-capacity); so to speak, axiom
(CA0III ) is (in the presence of (CAI )) stronger than (CAIII ).
Note that (CA0III ) is, in the presence of (CAI ), equivalent to the following
condition:
(CA00III ) For every compact subset K of X ,

f (K) = inf{f (U) : K ⊂ U , U open} .

For, let α be the infimum on the right side; if U is any open set containing K
(at least U = X qualifies) then f (K) 6 f (U) , whence f (K) 6 α .
(CA0III ) ⇒ (CA00III ): If f (K) = +∞ then also α = +∞ . If f (K) < +∞ ,
let a be any real number such that f (K) < a < +∞ ; since a > f (K) ,
by (CA0III ) there exists an open set U such that K ⊂ U and f (U) < a ,
INT IV.x76 extension of a measure. l p spaces §5

whence, by the definition of α , α 6 f (U) < a , and varying a yields

α 6 f (K) .
(CA00III ) ⇒ (CA0III ): Let K be a compact subset of X and let a > f (K) .
By assumption, f (K) = α , that is, f (K) is the greatest lower bound of
the f (U) for U open and K ⊂ U ; since a > f (K) , a is not a lower bound
for the f (U) , so there must exist an open set U ⊃ K such that f (U) < a .
Already noted above is that the function |µ|* : P(X) → R (in other
words, µ* for any positive measure µ ) is a GT-capacity; in view of the
preceding equivalence, it is also a CR-capacity, indeed, the formula

|µ|*(A) = inf{|µ|*(U) : A ⊂ U , U open}

holds for every subset A of X (§2, No. 4, Prop. 19).

For f either a GT-capacity or a CR-capacity, a subset A of X is said
to be capacitable if it satisfies the condition

f (A) = sup{f (K) : K ⊂ A , K compact}

(GT, IX, §6, No. 9, Def. 9) or (TG, IX, §6, No. 10, Def. 10). For f = |µ|*
the condition reads

|µ|*(A) = sup{|µ|(K) : K ⊂ A , K compact} .

If f is a GT-capacity on a metrizable space X , then every relatively

compact Souslin subset A of X is capacitable for f (GT, loc. cit., Th. 5),
whereas if f is a CR-capacity on a Hausdorff space X , then every Souslin
subset A of X is capacitable for f (TG, loc. cit., Th. 6).
IV.61, `. 5, 6.
“Therefore, for every ε > 0 there exists a compact subset K of A such
that |µ|*(A) 6 |µ|*(K) + ε ”
Since |µ|*(A) < +∞ ( A is relatively compact), this is immediate from
the definition of capacitability (GT, IX, §6, No. 9, Def. 9; see the discussion
in the preceding note).
IV.61, `. 6–8.
“Let U be a relatively compact open set in X containing A such that

|µ|(U) = |µ|*(U) 6 |µ|*(A) + ε . ”

Since A is compact, |µ|*(A) 6 |µ|*(A) < +∞ , hence there exists an

open set V with A ⊂ V and |µ|*(V) 6 |µ|*(A)+ε (§1, No. 4, Prop. 19). On
the other hand, there exists a relatively compact open set W with A ⊂ W
§5 measurable functions and sets INT IV.x77

(every point of A has a relatively compact open neighborhood; extract a

finite subcovering of A ). Let U = V ∩ W . Then A = A ∩ A ⊂ V ∩ W = U ,
where U is relatively compact (because W is) and |µ|*(U) 6 |µ|*(V) 6
|µ|*(A) + ε .
IV.61, `. 17–21.
“It follows at once (§4, No. 9, Lemma) that there exist a negligible set
N ⊂ K and a finite partition of K --- N formed of integrable sets M j such
that each of the sets K ∩ Vxi is the union of a subset of N and a certain
number of the Mj , and such that on each of the Mj , f is equal to one of
the functions gxi .”
Say i runs from 1 to r . For brevity, write V i = Vxi and gi = gxi for
i = 1, . . . , r . By hypothesis, for each i one has f = g i almost everywhere
in Vi , so there exists a negligible set Ni ⊂ Vi such that f = gi on Vi --- Ni .
Write Ai = Vi --- Ni ; as the difference of integrable sets, A i is integrable.
Thus

(∗) V i = Ni ∪ Ai (1 6 i 6 r) ,

a disjoint union with Ni negligible and Ai integrable.

For every r-tuple ε = (ε1 , . . . , εr ) with εi = ±1 , write

Vε = V1ε1 ∩ · · · ∩ Vrεr ,

with the conventions Vi1 = Vi and Vi−1 = X --- Vi . The 2r sets Vε

partition X :
[
X = (V1 ∪ { V1 ) ∩ (V2 ∪ { V2 ) ∩ · · · ∩ (Vr ∪ { Vr ) = Vε .
ε

Let ε0 = (−1, . . . , −1) ; since

Vε0 = { V1 ∩ · · · ∩ { Vr = { (V1 ∪ · · · ∪ Vr ) ,

the 2r − 1 sets Vε (ε 6= ε0 ) are integrable (each being contained in at least

one Vi ), pairwise disjoint, and
r
[ [ [
Vi = Vε , Vi = Vε .
i=1 ε6=ε0 εi =1

S
r
Since K ⊂ Vi , one has
i=1

r
[ [ [
(†) K=K∩ Vi = K ∩ Vε , K ∩ Vi = K ∩ Vε .
i=1 ε6=ε0 εi =1
INT IV.x78 extension of a measure. l p spaces §5

The plan is to decompose each of the sets K ∩ V ε (ε 6= ε0 ) in the form

of (∗)—the disjoint union of a negligible part and an integrable part—then
decompose each integrable part as the disjoint union of a negligible set and
a sequence of compact sets on which f is continuous.
To begin with, consider Vε1 , where ε1 = (1, . . . , 1) ; citing (∗),
r
\ r
\
V ε1 = V 1 ∩ · · · ∩ V r = (Ni ∪ Ai ) = Nε1 ∪ Ai ,
i=1 i=1

where Nε1 is the union of 2r − 1 sets of the form B1 ∩ · · · ∩ Br , each Bi

being equal to either Ni or Ai , and at least one of the Bi being equal
Tr
to Ni . Writing Aε1 = Ai , one has
i=1

V ε1 = N ε1 ∪ A ε1 ,

where Nε1 is negligible, Aε1 is integrable, and Aε1 ⊂ Ai for at least one
value of i (in this instance, for every i ).
Consider now any ε such that ε 6= ε0 and ε 6= ε1 (that is, εi is equal
to 1 for at least one value of i and to −1 for at least one value). Then
 \   \   \   [ 
Vε = Vi ∩ { Vi = Vi ∩ { Vi = R ∩ { S ,
εi =1 εi =−1 εi =1 εi =−1
T S
where R = Vε and S = Vε . By the argument used for Vε1 , one
εi =1 εi =−1
can write R as a disjoint union R = N0 ∪A
T with N0 negligible, A integrable
and A ⊂ Ai for some i (indeed, A = Ai ). Then
εi =1

Vε = R ∩ { S = (N0 ∪ A) ∩ { S
= (N0 ∩ { S) ∪ (A ∩ { S) = (N0 ∩ { S) ∪ (A --- A ∩ S) ,

where N0 ∩ { S is negligible,
T A --- A ∩ S is integrable and is contained in
some Ai (indeed, in Ai ). We write Nε = N0 ∩ { S and Aε = A --- A∩S .
εi =1
To summarize: for every ε 6= ε0 we may write

Vε = N ε ∪ A ε

with Nε negligible, ATε integrable, Nε ∩ Aε = ∅ and Aε contained in

some Ai (indeed, in Ai ), so that, for such i , f = gi on Ai . Then
εi =1

(††) (∀ ε 6= ε0 ) K ∩ Vε = (K ∩ Nε ) ∪ (K ∩ Aε ) ,
§5 measurable functions and sets INT IV.x79

a disjoint union with K ∩ Nε negligible,

T K ∩ Aε integrable, and K ∩ Aε
contained in some Ai (indeed, in Ai ). Setting
εi =1

[
N= K ∩ Nε ,
ε6=ε0

we have a disjoint union

[ [
K= K ∩ Vε = N ∪ K ∩ Aε ,
ε6=ε0 ε6=ε0

where N is negligible, every K ∩ Aε (ε 6= ε0 ) is integrable and is contained

in some Ai .
Let Mj (1 6 j 6 2r − 1) be any enumeration of the sets K ∩ A ε
(ε 6= ε0 ) . We then have a disjoint union
[
(†††) K=N∪ Mj ,
j

where N is negligible, the Mj are integrable and, for each j , there exists
an index i such that f = gi on Mj .
{Though it is not needed for the rest of the proof of Prop. 4, we verify
that for each i (1 6 i 6 r), K ∩ Vi = K ∩ Vxi has the form
[
K ∩ Vi = N0i ∪ Mj ,
j∈Ji

where N0i ⊂ N and Ji is some set of the indices j (1 6 j 6 2r − 1). Indeed,

as observed in (†), [
K ∩ Vi = K ∩ Vε ;
εi =1

in view of (††) this may be written

 [   [ 
K ∩ Vi = K ∩ Nε ∪ K ∩ Aε ,
εi =1 εi =1

where the K ∩ Nε are all subsets of N , and each K ∩ Aε is equal to one of

the Mj . (The set Ji of indices in question depends on how the K ∩ A ε are
enumerated.)}
IV.61, `. −9.
“. . . f is measurable.”
INT IV.x80 extension of a measure. l p spaces §5

The conclusion of the proof of Prop. 4 is a matter of refining (twice) the

partition (†††) of K of the preceding note, so as to exhibit the measurability
of f . Since each Mj is integrable, there exists a partition
[
Mj = N j ∪ Knj
n∈N

with Nj negligible1 and (Knj )n∈N a sequence of pairwise disjoint compact

sets (§4, No. 6, Cor. 2 of Th. 4). Also, for each j , there exists an index i
such that f = gi on Mj , hence f = gi on Knj for all n ∈ N ; since gi is
measurable and Knj is compact, there exists a further partition
[
Knj = Pnj ∪ Kmnj
m∈N

with Pnj negligible and (Kmnj )m∈N a sequence of pairwise disjoint compact
sets such that the restriction to Kmnj of gi —hence also of f —is continuous
(No. 1, Def. 1). The partition (†††) doubly refined now takes the form
[ [ 
K=N∪ Nj ∪ Knj
j n∈N
 [  [ [  [ 
= N∪ Nj ∪ Pnj ∪ Kmnj
j j n∈N m∈N
 [ [  [
= N∪ Nj ∪ Pnj ∪ Kmnj
j j,n j,n,m
[
= N0 ∪ Kmnj
j,n,m

S S
with N0 = N ∪ Nj ∪ Pnj negligible and the Kmnj compact sets such
 j j,n
that f Kmnj is continuous; since K is an arbitrary compact subset of X ,
f is measurable (No. 1, Def. 1).
IV.61, `. −4.
“By the principle of localization, every locally negligible set is measur-
able.”
Let A ⊂ X be locally measurable. Given any compact set K in X , it
suffices to show that A ∩ K is integrable (No. 1, Prop. 3). Indeed, A ∩ K is
negligible by Prop. 5 below (which is derived directly from Def. 3 without the
1 The N ’s here are different from the N ’s occurring in the preceding note (an
j i
ephemeral notation on the way to the partition (†††)).
§5 measurable functions and sets INT IV.x81

intervention of Prop. 4); but every negligible set N is integrable: ϕ N = 0

µ-almost everywhere, therefore ϕN ∈ L p for all p > 1 (§3, No. 4, remarks
following Def. 2) and so ϕN is integrable (§4, No. 1, Def. 1) with integral 0
(loc. cit.).
The proof via Prop. 4 is overkill, but here it is: Let A be a locally
negligible set and let f = ϕA . For every x ∈ X , there exists a neighbor-
hood Vx of x such that Vx ∩ A is negligible; dropping down to a compact
neighborhood of x contained in Vx , we can suppose that Vx is compact,

hence integrable. Set gx = 0 (the zero function on X ); then ϕA Vx = 0
except on the negligible set Vx ∩ A , thus f = gx almost everywhere in Vx ,
and so f is measurable by Prop. 4.
IV.62, `. 10, 11.
“Conversely, if it is satisfied then A is contained in an integrable open
set G ”
Since |µ|*(A) < +∞ , there exists an open set G with A ⊂ G and
|µ|*(G) < +∞ (§1, No. 4, Prop. 19), and an open set with finite outer
measure is integrable (§4, No. 6, Prop. 10).
IV.62, `. 18–20.
“ . . . if G is locally negligible then |µ|(K) = 0 for every compact set
K contained in G ”
For, K = K ∩ G is negligible by Prop. 5.
Thus, |µ|*(G) = sup{|µ|(K) : K ⊂ G , K compact} = 0 (§4, No. 6,
Cor. 4 of Th. 4), whence G ⊂ { Supp(µ) (§2, No. 2, Prop. 5).
IV.62, `. −12.
“If the set N of points of discontinuity of f is locally negligible . . . ”
The proof
 shows, more generally, that if N is any locally negligible set
such that f X --- N is continuous, then f is measurable.
IV.62, `. −9, −8.
“ . . . for every ε > 0 , there exists a compact set K 1 ⊂ K --- (K ∩ N)
such that |µ|(K --- K1 ) 6 ε ”
Let ε > 0 . Since |µ|*(K ∩ N) = 0 , there exists an open set U with
K ∩ N ⊂ U and |µ|(U) 6 ε (§1, No. 4, Prop. 19). Let

K1 = K ∩ { U = K --- U = K --- K ∩ U ;

then K1 is compact, K1 ⊂ K , and K --- K1 = K ∩ U ⊂ U , whence

|µ|(K --- K1 ) 6 |µ|(U) 6 ε . Since K∩N
 ⊂ K∩U , so that K 1 = K --- K∩U ⊂
K --- K ∩ N = K --- N , and since f X --- N is continuous, f K1 is also con-
tinuous; thus f is measurable by No. 1, Prop. 1.
INT IV.x82 extension of a measure. l p spaces §5

Note that it suffices to

 assume the weaker condition that N is a locally
negligible set such that f X --- N is continuous (i.e., f need not be contin-
uous as a function on X at the points of X --- N ). Consider, for example,
X = R , µ the Lebesgue measure on R , and f = ϕ Q  the characteristic
function of the rationals. Since Q is negligible and f R --- Q = 0 is con-
tinuous, f is measurable by the foregoing argument, despite being nowhere
continuous.
IV.62, `. −6, −5.
“ hh P x locally almost everywhere (with respect to µ ) ii ”
Let N = {x : x ∈ X and non P x } ; then Prop. 5 yields the criterion

P x locally almost everywhere in X

⇔ N is locally negligible
⇔ (∀ compact K ) K ∩ N is negligible
⇔ (∀ compact K ) {x : x ∈ K and non P x } is negligible
⇔ (∀ compact K ) P x almost everywhere in K .

IV.63, `. 11–13.
“Proposition 6.”
It is convenient to change the notation: assuming g : X → F is measur-
able and f : X → F is a function equal locally almost everywhere to g , let us
show that f is also measurable. Let N = {x ∈ X : f (x) 6= g(x) } , a locally
negligible set. The plan is to cite the principle of localization (Prop. 4).
For each x ∈ X let Vx be any compact (hence integrable) neighborhood
of x and let gx = g . Then Vx ∩ N is negligible (Prop. 5) and, on the subset
Vx --- Vx ∩ N of X --- N , f = g = gx , thus f = gx almost everywhere
in Vx . Quote Prop. 4.
IV.64, `. 6–9.
“Corollary 4.”
One will observe that the same proof works with R replaced by C . It
is tacit that F has its unique compatible Hausdorff topology (TVS, I, §2,
No. 3, Th. 2).
Sufficiency. Assume each fk : X → F is measurable. For each k ,
ek fk : X → F is the composite X → R → F , where

x 7→ fk (x) 7→ fk (x) · ek (x ∈ X) ;

since x 7→ fk (x) is measurable, and c 7→ c · ek is continuous, the composite

Pn
ek fk is measurable by Th. 1, so f = ek fk is measurable by Cor. 3.
k=1
§5 measurable functions and sets INT IV.x83

(Here, e1 , . . . , en can be any finite list of vectors, not necessarily linear

independent, in any topological vector space.)
Pn
Necessity. Assume f = ek fk is measurable. Let u : F → Rn be
k=1
the linear mapping such that u(e1 ), . . . , u(en ) is the canonical basis of Rn .
Since F is Hausdorff, u is continuous (even bicontinuous, by Th. 2 of TVS,
§2, No. 3). For all x ∈ X ,

X
n 
u f (x) = u ek fk (x) = (f1 (x), . . . , fn (x)) .
k=1

If prk : Rn → R is the k ’th coordinate projection, then pr k u(f (x) =
fk (x) for all x , thus
fk = (prk ◦ u) ◦ f ;
since f is measurable and prk ◦ u is continuous, fk is measurable by Th. 1.
IV.64, `. 10–13.
“Corollary 5.”
The mapping [f · g] : F × G → H is the composite X → F × G → H
defined by

x 7→ f (x), g(x) 7→ [f (x) · g(x)] ,
where f and g are measurable and (u, v) → [u · v] is continuous, so the
composite is measurable by Th. 1.
The notation [f · g] is suggestive of a bilinear operation, but in fact
bilinearity plays no role in the proof and can be omitted from the hypothesis.
IV.64, `. −5.
“The first assertion obviously follows from the second ”
With notations as in 2◦ , f is continuous on K1 ; cite No. 1, Prop. 1.
IV.65, `. 4, 5.
“ Bn,r is a countable union of compact sets contained in K 0 , hence is
integrable ”
For any pair (α, β) ∈ A × A , write Fα,β : K0 → R for the function


Fα,β (x) = d fα (x), fβ (x) (x ∈ K0 ) .

Then Fα,β is continuous, so the set

−1
{x ∈ K0 : d fα (x), fβ (x) > 1/r } = Fα,β ([1/r, +∞[)
INT IV.x84 extension of a measure. l p spaces §5

is a closed, hence compact, subset of K 0 ; by definition,

[ −1
(∗) Bn,r = Fα,β ([1/r, +∞[) ,
(α,β)∈An ×An

a countable union of compact subsets of K 0 . If (Sn )n>1 is any enumeration

S
m
of these compact subsets then, for each index m , the set T m = Sk is
k=1
compact, hence integrable (§4, No. 6, Cor. 1 of Prop. 10), with

|µ|(Tm ) 6 |µ|(K0 ) < +∞ for all m ,

S
∞
therefore Bn,r = Tm is integrable (§4, No. 5, Prop. 8).
m=1
Note that since An is a decreasing function of n , it is clear from (∗)
that, for fixed r , Bn,r is also a decreasing function of n ; and for fixed n ,
Bn,r is an increasing function of r .
IV.65, `. 6–8.
“If r is fixed, the intersection of the decreasing sequence of sets B n,r
(n = 1, 2, . . .) has measure zero, since f α (x) tends to f (x) almost every-
where in K0 with respect to the filter F ”
T
∞
Set Br = Bn,r . Since (Bn,r )n>1 is a decreasing sequence of in-
n=1
tegrable sets, we know that Br is integrable and |µ|(Bn,r ) → |µ|(Br )
as n → ∞ (§4, No. 5, Cor. of Prop. 7).
We are to show that |µ|(Br ) = 0 ; since K0 ∩ N is negligible (No. 2,
Prop. 5) and Br ⊂ K0 , it will suffice to show that Br ⊂ N .
Let x ∈ Br and assume to the contrary that x ∈
X --- N . Then,
fα (x) → f (x) with respect to F , hence the family fα (x) α∈A is Cauchy
with respect to F ; since the An form a base for F , given any ε > 0 there
exists an index n such that


(α, β) ∈ An × An ⇒ d fα (x), fβ (x) < ε .

In particular, let ε = 1/r and let n be such an index, so that

d fα (x), fβ (x) < 1/r for all (α, β) ∈ An × An ;

the existence of such an n means that x does not belong to B n,r , contrary
to x ∈ Br ⊂ Bn,r .
Thus lim |µ|(Bn,r ) = |µ|(Br ) = 0 .
n→∞
§5 measurable functions and sets INT IV.x85

IV.65, `. 8.
“ . . . thus lim |µ|(Bn,r ) = 0 ”
n→∞
See the preceding note.
IV.65, `. 13–16.
“Let C be the complement of B in K0 ; by construction, fα (x) con-
verges uniformly to f (x) in C with respect to the filter F , and since the
restrictions of the fα to C are continuous, so is the restriction of f to C .”
Of course B ⊂ K0 . Recall that K0 ∩ N is negligible; for reasons that
will be clear in the following argument, it is convenient to take C to be the
complement of B ∪ (K0 ∩ N) in K0 . Writing N0 = K0 ∩ N , an elementary
calculation yields

C = K0 --- (B ∪ N0 ) = (K0 --- B) ∩ (X --- N) .

Note that C is integrable; for, B is integrable and N 0 is negligible (hence

integrable), whence the integrability of B∪N 0 and then of K0 --- (B∪N0 ) =
C (§4, No. 5, Props. 6 and 7). Moreover, K 0 --- C = B ∪ N0 , whence
|µ|(K0 --- C) 6 |µ|(B) + |µ|(N0 ) = |µ|(B) 6 ε/4 .
Let x ∈ C . Since x ∈ X --- N , we know that f α (x) → f (x) with
respect to F . Also,
[
∞ 
x ∈ K0 --- B = K0 ∩ { B = K0 ∩ { Bnr ,r
r=1
∞
\ ∞
\
= K0 ∩ { Bnr ,r = (K0 --- Bnr ,r ) .
r=1 r=1

Thus, for all x ∈ C and all r , we have x ∈ K 0 --- Bnr ,r , therefore (by the
definition of Bnr ,r )


d fα (x), fβ (x) < 1/r for all α, β ∈ Anr ,r .

In other words: given any integer r > 1 , for every (α, β) ∈ A nr ,r × Anr ,r
we have

d fα (x), fβ (x) < 1/r for all x ∈ C ,


whence sup d fα (x), fβ (x) 6 1/r ; this shows that the family of restric-
x∈C

tions fα C is uniformly Cauchy with respect
 to F , and since f α → f
pointwise in C it follows that fα C → f C uniformly with  respect to F
(GT, X, §1, No. 5, Prop. 5). 
 And since C ⊂ K 0 and the fα K0 are contin-
uous, we conclude that f C is continuous.
INT IV.x86 extension of a measure. l p spaces §5

Since C is integrable, there exists a compact set K 1 ⊂ C such that

|µ|(C --- K1 ) 6 ε/4 (§4, No. 6, Cor. 1 of Th. 4), and since f C is continuous,
so is f K1 . Now, K1 ⊂ C ⊂ K0 ⊂ K , so it remains only to check that
|µ|(K --- K1 ) 6 ε . Indeed,

K --- K1 = (K --- K0 ) ∪ (K0 --- C) ∪ (C --- K1 ) ,

therefore

|µ|(K --- K1 ) = |µ|(K --- K0 )+|µ|(K0 --- C)+|µ|(C --- K1 ) 6 ε/2+ε/4+ε/4 .

IV.65, `. −15, −14.

“ f is the limit of the sequence (fαn ) locally almost everywhere, hence
is measurable”
We can suppose that the sequence (An )n∈N is decreasing; then

n > m ⇒ α n ∈ An ⊂ Am .

Let x ∈ X --- N . By hypothesis, fα (x) → f (x) with respect to F ; thus,

given any ε > 0 , there exists an index m such that


α ∈ Am ⇒ d fα (x), f (x) 6 ε

(GT, I, §7, No. 3, Prop. 7), consequently

n > m ⇒ αn ∈ Am ⇒ d fαn (x), f (x) 6 ε .


This shows that the sequence fαn (x) n∈N converges in F to f (x) for every
x ∈ X --- N . Switching to N as index set and to the Fréchet filter on N (GT,

I, §6, No. 1, Example 3), Th. 2 becomes applicable: f αn X --- N → f X --- N
pointwise with respect to the Fréchet filter (that is, for
each x ∈ X --- N the
elementary filter associated with the sequence fαn (x) n∈N converges in F
to f (x) ), therefore f extended in any manner to all of X is measurable
by 1◦ of Th. 2. In this context, 2◦ of Th. 2 holds, but restricted to the
sequence of functions fαn (the classical form of Egoroff’s theorem).
As for condition 2◦ as stated in the text, recall that its proof entails the
countability of the sets An (see (∗) in the note for `. 4–5).
It is perhaps useful to state the classical sequential form explicitly:
Theorem 20 . — Let X be a locally compact space, µ a measure
on X , and (fn ) a sequence of measurable mappings of X into a metriz-
able space F .
§5 measurable functions and sets INT IV.x87

Assume that there exists a locally negligible subset N of X such that

lim fn (x) exists in F for every x ∈ X --- N . Then:
n→∞
1◦ If f : X → F is any mapping such that

f (x) = lim fn (x) for all x ∈ X --- N ,

n→∞

then f is measurable.
2◦ For every compact subset K of X and every ε > 0 , there exists
a compact set K1 ⊂ K such that |µ|(K --- K1 ) 6 ε and such that the
restrictions of the fn to K1 are continuous and converge uniformly to f
on K1 .
In the language of Th. 2 and its proof, A = N and F is the filter on N
generated by the sets An = {k ∈ N : k > n } . Another way of packaging
the result:

Theorem 200 . — Let X be a locally compact space, µ a measure

on X , and (fn ) a sequence of measurable mappings of X into a metrizable
space F .
If f : X → F is a mapping such that fn (x) → f (x) locally almost
everywhere in X (i.e., for x in the complement of a locally negligible set),
then:
1◦ f is measurable.
2◦ Same as in Th. 20 above.
IV.66, `. −15.
“ . . . f is measurable.”
Recall that Ai ∩ K is integrable (No. 1, Prop. 3), whence (§4, No. 6,
Cor. 2 of Th. 4) the existence of the partitions

∞
[
Ai ∩ K = N i ∪ Kin (i = 1, . . . , m) .
n=1

S
m
The Ai ∩ K are pairwise disjoint, with union K (because Ai = X ),
i=1
therefore
m
[ [
m  [ ∞
m [
K= Ai ∩ K = Ni ∪ Kin ,
i=1 i=1 i=1 n=1

where the first union is negligible and the K in are pairwise disjoint compact
sets with f = ai on Kin , thus f is measurable by No. 1, Def. 1.
INT IV.x88 extension of a measure. l p spaces §5

IV.66, `. −15, −14.

“By an abuse of language . . . ”
Heretofore, the term “step function” (fonction étagée) was used only
for functions with values in a Banach space (§4, No. 9, Def. 4), whose addi-
tive structure admits representing step functions by means of characteristic
functions and confers algebraic structure on the set of step functions.
There is another sense in which the language is abused: if f : X → F is
a step function (with respect to some clan) and f is a measurable function
(with respect to the given measure on X ) then f need not be a ‘measurable
step function’ in the sense of the text. For example, if F = {a, b} is the
2-point space whose only open sets are F and ∅, then every function X → F
is continuous, hence measurable; but if X has a non-measurable subset A ,
then the function f : X → F that assigns the value a to the points of A
and the value b to the points of X --- A is a step function with respect to
the clan {∅, A, X --- A, X} but is not a measurable step function.
However, if F is a T1 -space (all 1-point subsets closed) then all is well:
every measurable function f : X → F with finite range f (X) is a measurable
−1
step function, because f (a) is measurable for all a ∈ F by Prop. 7 below.
The term used for a linear combination of characteristic functions of
intervals is fonction en escalier (FVR, II, §1, No. 3, Def. 2); having translated
it as “step function” for the footnote on p. ix of Vol. I and in the Example
in §4, No. 4 (p. IV.40), when I arrived at fonction étagée, I did not notice
the difference between the French terms and translated it as “step function”
too. There’s a real abuse of language, somewhat mitigated by the fact that
“fonction en escalier” is a special case of “fonction étagée”; alas, I noticed
my abuse of language only when exploring, for these notes, why the author
thought he had abused it.
For the record, I can find just 5 places in Integration where “step func-
tion” is to be interpreted in the FRV sense:
page xi of Vol. 1: footnote.
IV.40: Ch. IV, §4, No. 4, Example (once in italics, once in roman).
IV.133: Ch. IV, §5, the hint for Exer. 29 d). (Regrettably, in part c)
of the same exercise, “step function” is to be interpreted as fonction étagée;
part of the exercise is to guess the clan, the tribe of measurable functions
being a promising suspect.)
IV.134: Ch. IV, §5, the hint for Exer. 30 (twice).
IV.139: Ch. IV, §6, Exer. 17.
The following comments should help prevent misunderstandings:
a) a “step function” in the sense of FRV is defined on an interval I
of R and is a step function with respect to the clan Φ generated by the set
of subintervals of I ;
§5 measurable functions and sets INT IV.x89

b) for a step function whose domain is not an interval of R , the clan Φ

in question must be either signaled explicitly or be inferable from the context;
c) ambiguity can occur only when the domain of the function is an
interval of R ; if such a function does not fall under case a), the clan Φ in
question should be explicitly stated or be inferable from the context.

If not “step function”, what were the alternatives for “fonction étagée”?
Harrap’s Unabridged gives “tiered” or “terraced” as possibilities for “étagée”.
The frequent translation of “ét. . .” as “st. . .” suggests “staged”. In the ter-
minology of P.R. Halmos’ Measure theory, a “fonction Φ-étagée” (Φ a clan)
is a “simple function” with respect to a ring Φ of subsets of a set (op. cit.,
p. 84), whereas T.H. Hildebrandt’s Introduction to the theory of integration
employs “finite-step function”, with “step function” reserved for functions
with a countable number of values (op. cit., p. 208).
After consulting twenty or so textbooks on Integration and Real Anal-
ysis, I can report the following usage:
(1) “step function” (or “step map”), for “fonction en escalier”: Asplund
& Bungart, Burrill & Knudsen, Dieudonné, Hewitt & Stromberg, K. Hoff-
man, Lang, McShane & Botts, K.A. Ross, Royden, Stromberg, Zaanen. The
overwhelming choice. {Not all books treated the topic, and Hildebrandt em-
ploys the interesting term “staircase function”, with “interval step function”
as possible alternative (op. cit., p. 218).}
(2) “step function” (or “step map”) with respect to a class of sets,
for “fonction étagée”: Dinculeanu, L.M. Graves, Lang, McShane, Segal &
Kunze, Zaanen.
(20 ) “simple function” for “fonction étagée”: Berberian, Burrill & Knud-
sen, Halmos, Hewitt & Stromberg, Munroe, McShane & Botts, Royden,
Rudin, Saks, Stromberg. Somewhat in the lead, but it translates “étagée”
poorly.

Two possible solutions seem to me to stand out for the usage in Inte-
gration:

(A) “step function” for “étagée”, and “interval step function” for “en
escalier”;

(B) “simple function” for “étagée” and “step function” for “en escalier”.

Were Integration to receive a second edition, (A) would be easy to

implement—insert ‘interval’ at the 5 places noted above (twice on p. IV.40
and on p. IV.134). This would highlight, not contradict, the usage in FRV,
which clearly has first claim on “step function”.
INT IV.x90 extension of a measure. l p spaces §5

IV.66, `. −7, −6.

“The condition is sufficient by Egoroff’s theorem and the principle of
localization.”
Assuming the condition, we are to show that f is measurable. Let
x ∈ X and let K be a compact (hence integrable) neighborhood of x .
In order to invoke the principle of localization (No. 2, Prop. 4), we seek a
measurable function g on X such that f = g almost everywhere in K .
By assumption, there exists a sequence of measurable step functions
gn : X → F such that gn → f almost everywhere in K . Let N be a
negligible subset of K such that gn → f on the set A = K --- N , which
is measurable (indeed, integrable; see the comment at the end of the note
for IV.60, `. −5).
We will define a sequence of measurable step functions h n : X → F
that converges to the desired function g . Fix a point a ∈ F . For each n ,
define hn by (
gn (y) if y ∈ A
hn (y) =
a if y ∈ X --- A .

Since gn (X) is finite, so is hn (X) . To show that hn is a measurable step

−1
function, we need only show that if b is any value of h n , then hn (b) is a
measurable set; indeed, if b 6= a then
−1 −1
hn (b) = {y ∈ A : gn (y) = b } = A ∩ g n (b) ,

whereas if b = a then
−1 −1  −1

hn (b) = hn (a) = {y ∈ A : gn (y) = b }∪(X --- A) = A∩ g n (a) ∪(X --- A) .

If y ∈ A then hn (y) = gn (y) → f (y), and if y ∈ X --- A then hn (y) = a

for all n ; defining
(
f (y) if y ∈ A = K --- N
g(y) =
a if y ∈ X --- A ,

we have hn → g pointwise in X ; since the hn are measurable, g is mea-

surable by Egoroff’s theorem (No. 4, Th. 2) and the assertion is proved.
IV.66, `. −2 to IV.67, `. 2.
“ . . . for each Ki with index i 6 n , there exists a finite partition of K i
into integrable sets Aij (1 6 j 6 qi ) sufficiently small that the oscillation
of f on each of the Aij is 6 1/n (§4, No. 9, Lemma)”
§5 measurable functions and sets INT IV.x91

To accommodate special cases, it is convenient to rearrange slightly the

notations. We have a partition
[
K=N∪ Ki
i

with N negligible and (Ki ) a countable

 family of pairwise disjoint compact
sets such that the restrictions f Ki are all continuous. The problem is to
construct a sequence (gn ) of measurable step functions gn : X → F such
that, for every i , gn → f almost everywhere in Ki . For technical purposes,
fix a point a ∈ F .
(i) If K is negligible then every sequence (g n ) of measurable step func-
tions meets the requirements: Ki is a negligible subset of Ki , and gn → f
vacuously on Ki --- Ki = ∅ . (For example, let every gn be the constant
mapping gn (x) = a for all x ∈ X .)
Assuming K is not negligible, we can assume henceforth that every
negligible Ki has been assimilated in N , so that every remaining K i is
non-negligible, hence nonempty. We are going to construct a sequence (g n )
of measurable step functions such that g n → f at every point of K --- N
(hence almost everywhere in K ).
(ii) Suppose there are only finitely many K i , say K1 , . . . , Kr . Then
S
r 
K = N ∪ K0 , where K0 = Ki is compact and f K0 is continuous (GT, I,
i=1
§3, No. 2, Prop. 4), thus we are reduced to the case r = 1 .
Suppose then that K = N ∪ K0 is a partition
 of K with N negligible
and K0 a nonempty compact set such that f K0 is continuous.
Given any integer n > 1 let us construct a measurable step function g n
as follows. If x ∈ K0 then, by the continuity of f at x in K0 , there exists
a compact neighborhood Vx of x in X such that


d f (y), f (x) 6 1/2n for all y ∈ Vx ∩ K0 ;

since K0 is compact, K0 ⊂ Vx1 ∪ · · · ∪ Vxr for suitable x1 , . . . , xr in K0 .

Then
[r
K0 = V xk ∩ K 0 ,
k=1

where each of the sets Vxk ∩ K0 (1 6 k 6 r) is nonempty (it contains at

least xk ) and compact (hence integrable) and on which the oscillation of f
is 6 1/n ; for, if y, z ∈ Vxk ∩ K0 then




d f (y), f (z) 6 d f (y), f (xk ) + d f (xk ), f (z) 6 1/2n + 1/2n .
INT IV.x92 extension of a measure. l p spaces §5

By the Lemma of §4, No. 9 there exists in the clan generated by the V xk ∩K0
(hence in the clan of integrable sets and in the tribe of measurable sets)
a finite family A of pairwise disjoint sets such that each V xk ∩ K0 may be
expressed as the union of a subfamily of A . Let A 1 , A2 , . . . , Aq be a listing
(without repetitions) of those nonempty members of A that occur in the
expressions of Vx1 ∩ K0 , . . . , Vxr ∩ K0 ; then
q r
[ [
Aj = V xk ∩ K 0 = K 0 ,
j=1 k=1

where the Aj are nonempty measurable (and integrable) sets on each of

which f has oscillation 6 1/n . For j = 1, . . . , q choose a j ∈ f (Aj ) and
define a function gn : X → F with finite range by
(
aj if y ∈ Aj for some j
gn (y) =
a if y ∈ X --- K0 ;

gn is a measurable step function since (keeping in mind that, although

the Aj are pairwise disjoint, the aj need not be distinct, and any or all of
them may be equal to a )

−1 [
g n (a) = (X --- K0 ) ∪ Ak
ak =a

−1
(that is, the union of X --- K0 and those Ak for which ak = a ), and g n (aj )
is equal to [ [
Ak or (X --- K0 ) ∪ Ak
ak =aj ak =aj

according as aj 6= a or aj = a , respectively.
Given any y ∈ K0 , say y ∈ Aj , then

gn (y) = aj ∈ f (Aj ) and f (y) ∈ f (Aj ) ;

since f has oscillation 6 1/n on Aj , we have d gn (y), f (y) 6 1/n .
To summarize: for each integer n > 1 we have constructed a measurable
step function gn : X → F such that


d gn (y), f (y) 6 1/n for all y ∈ K0 ;

fixing y ∈ K0 and letting n → ∞ , we have gn (y) → f (y) , thus gn → f

pointwise on K0 = K --- N , whence gn → f almost everywhere in K .
§5 measurable functions and sets INT IV.x93

S
∞
(iii) Finally, suppose K = N ∪ Ki is a partition of K with N negli-
i=1
gible and
 (Ki ) an infinite sequence of non-negligible compact sets such that
every f Ki is continuous.
Sn
For every n > 1 let Jn = Ki , so that (Jn ) is an increasing sequence
i=1
S
∞ 
of compact sets such that Jn = K --- N and f Jn is continuous for
n=1
every n .
For each n , the technique of the preceding case shows that there exists
a measurable step function gn such that


(∗) d gn (y), f (y) 6 1/n for all y ∈ Jn ,
S
∞
and, say, gn (y) = a for all y ∈ Ki = (K --- N) --- Jn . Given any
i=n+1
S
∞
y ∈ K --- N = Jn , we assert that gn (y) → f (y) . Say y ∈ Jn0 . Given
n=1
any ε > 0 , choose an integer n1 such that 1/n1 < ε ; increasing n1 if
necessary, we can suppose that n1 > n0 . Then, for all n > n1 we have
y ∈ J n0 ⊂ J n1 ⊂ J n ,


therefore by (∗), d gn (y), f (y) 6 1/n 6 1/n1 < ε . Thus gn → f pointwise
on K --- N , hence almost everywhere in K .
IV.67, `. 5, 6.

“It is clear that the sequence gn (x) converges to f (x) at every point
of K not belonging to N .”
See the preceding note.
IV.67, `. 12.
“With notations as in the proof of Th. 3 . . . ”
The notations in the note before the last are readily adapted to the
present corollary, following the analysis into cases (i)–(iii).
(i) If K is negligible, define gn = 0 for all n .
 0 (ii) If K = N∪K0 is a partition of K with N negligible, K 0 compact and
f K continuous, given any integer n > 1 construct a function g n : X → F
as follows. As in the previous note, let K 0 = A1 ∪ · · · ∪ Aq be a partition
of K0 such that the Aj are nonempty measurable sets on each of which f
has oscillation 6 1/n . For each j = 1, . . . , q choose a point x j ∈ Aj and
let aj = f (xj ) . If j is an index for which |aj | > 1/n , that is, n|aj | − 1 > 0 ,
define
1

a0j = 1 − aj ∈ F ,
n|aj |
INT IV.x94 extension of a measure. l p spaces §5

1
and note that |a0j | = |aj | − 1/n , a0j − aj = − · aj and |a0j − aj | = 1/n .
n|aj |
Define 
0
 if x ∈ X --- K0 = (X --- K) ∪ N
gn (x) = 0 if x ∈ Aj and |aj | 6 1/n

 0
aj if x ∈ Aj and |aj | > 1/n .
Each of the finitely many values of gn is assumed on a measurable subset
of X : [
−1
gn (0) = (X --- K0 ) ∪ Aj ,
|aj |61/n

and if |aj | > 1/n then

−1 [
gn (a0j ) = Ak ,
ak =aj

thus gn is a measurable step function. We assert that

(∗) |gn (x) − f (x)| 6 2/n for all x ∈ K0 ;

for, if x ∈ Aj and |aj | 6 1/n then gn (x) = 0 and

|gn (x) − f (x)| = |f (x)| 6 |f (x) − f (xj )| + |f (xj )| 6 1/n + |aj | 6 1/n + 1/n ,

whereas if |aj | > 1/n then

|gn (x) − f (x)| = |a0j − f (x)| 6 |a0j − aj | + |aj − f (x)|

= 1/n + |f (xj ) − f (x)| 6 1/n + 1/n .

Having constructed such a gn for every n , it is clear from (∗) that

gn (x) → f (x) for every x ∈ K0 .
Fix an index n > 1 and assume the foregoing notations for g n ; let us
show that

(∗∗) |gn (x)| 6 |f (x)| for all x ∈ X .

If gn (x) = 0 this is trivial, so we can suppose that x ∈ A j for some index j

with |aj | > 1/n . Then

|gn (x)| = |a0j | = |aj | − 1/n = |f (xj )| − 1/n ;

but
|f (xj )| − |f (x)| 6 |f (xj ) − f (x)| 6 1/n ,
§5 measurable functions and sets INT IV.x95

thus |f (xj )| 6 |f (x)| + 1/n , therefore

|gn (x)| = |f (xj )| − 1/n 6 (|f (x)| + 1/n) − 1/n = |f (x) .

S
∞
(iii) Finally, suppose K = N ∪ Ki is a partition of K , with N
i=1
negligible and (Ki ) an infinite sequence of non-negligible compact sets such
 S
n
that f Ki is continuous for all i . For every n > 1 let J n = Ki , so that
i=1
S
∞
(Jn ) is an increasing sequence of compact sets such that Jn = K --- N
 n=1
and f Jn is continuous for every n .
For each n , the technique of the preceding case (with K 0 replaced
by Jn ) shows that there is a measurable step function g n : X → F such
that

(∗) |gn (x) − f (x)| 6 2/n for all x ∈ Jn ,

S
∞
gn = 0 on (X --- K) ∪ N ∪ Ki = (X --- K) ∪ (K --- Jn ) = X --- Jn , and
i=n+1
|gn (x)| 6 |f (x)| for all x ∈ X . It follows from (∗), as in the previous note,
S
∞
that gn (x) → f (x) for every x in Jn = K --- N .
n=1

IV.67, `. −13, −12.

“ . . . the proof can then be concluded as in Th. 3 without modification.”
Sn
Except that, writing Jn = Ki as in the note for IV.66, `. −2ff, we
i=1
S
∞
now have Jn = X --- N (i.e., “ K ” is out of the picture). When F is a
n=1
Banach space, one can arrange, as in Cor. 1, that |g n (x)| 6 |f (x)| for all
x ∈ X (see the preceding note).
IV.67, `. −1.
−1
“ . . . which proves that f (A) is measurable.”
−1
With B = f (A) in mind, the proof depends on the following charac-
terization of measurability:
Lemma. For a subset B of X , the following conditions are equivalent:
(a) B is measurable;
(b) for every compact subset K of X , there exists a partition
[
K ∩ B = N0 ∪ K0n
n
INT IV.x96 extension of a measure. l p spaces §5

of K ∩ B with N0 negligible and (K0n ) a sequence of (pairwise disjoint)

compact sets.
Proof. (a) ⇒ (b): Assuming B measurable and K ⊂ X compact, by
the remarks following No. 1, Def. 2 there exists a partition
[
K=N∪ Kn
n

with N negligible and (Kn ) a sequence of (pairwise disjoint) compact sets

such that, for each n , either Kn ⊂ B or Kn ⊂ { B . Writing N0 = N ∩ B
and K0n = Kn ∩ B we have a partition
[
K ∩ B = N0 ∪ K0n
n

with N0 negligible and K0n ( = Kn or ∅ ) compact.

(b) ⇒ (a): To show that B is measurable, it will suffice to show that,
given any compact set K ⊂ X , the set K ∩ B is integrable (No. 1, Prop. 3).
By assumption, we have a partition
[
K ∩ B = N0 ∪ K0n
n

with N0 negligible (hence integrable) and the K 0n compact (hence inte-

grable). For every n we have (§4, No. 5, Prop. 6)
n
X [
n 
|µ|(K0i ) = |µ| K0i 6 |µ|(K) < +∞ ,
i=1 i=1
S
therefore K0n is integrable (loc. cit., Cor. of Prop. 8), hence so is K ∩ B .
n

IV.68, `. 8–10.
“ . . . in the notations of Th. 3, condition b) is satisfied by taking H to
be the countable set formed by the values of all of the functions g n .”
A proof can be based directly on the definition of ‘measurable function’.
Given a compact set K ⊂ X , we seek a negligible set N ⊂ K and a countable
set H ⊂ F such that f (K --- N) ⊂ H . Since f is measurable, there exist
a negligibleS set N ⊂ K and a sequence (K n ) of compact sets such that
K --- N = Kn and f Kn is continuous for every n (No. 1, Def. 1; that
n
the Kn can be taken to be pairwise disjoint is not needed here). For each n ,
f (Kn ) is a compact metrizable subspace of F , hence has a countable base
(GT, IX, §2, No. 9, Prop. 16) and therefore a countable subset H n such that
§5 measurable functions and sets INT IV.x97

Hn = f (Kn ) (loc. cit., No. 8, Prop. 12; since f (K n ) is a closed subset

S of F ,
closure in f (Kn ) means the same as closure in F ). Then H = Hn is a
n
countable subset of F such that (with closure in the sense of F )
[ [ [

H⊃ Hn = f (Kn ) = f Kn = f (K --- N) .
n n n

One can arrange that H ⊂ f (K --- N) . For, f (K --- N) ⊂ H , where H is

a metrizable subspace of F containing a countable dense subset, therefore
H has a countable base (GT, IX, §2, No. 8, Prop. 12); then the subspace
f (K --- N) of F also has a countable base (transitivity of induced topologies),
hence has a countable dense subset H 0 (loc. cit., No. 8, Prop. 12 again). Thus
H0 ⊂ f (K --- N) ⊂ H0 .
A tiny supplement to Th. 4: If f : X → F is a measurable function,
where F is a Souslin space in the sense of TG, IX, §6, No. 2, Def. 2,( ∗ ) one
can again prove that b) holds, and one can arrange that H ⊂ f (K --- N) .
For, f (Kn ) is a compact subset of F , hence is closed in F , hence is a Souslin
subspace of F (loc. cit., Prop. 5), hence is metrizable (TG, IX, App. 1, Cor. 2
of Prop. 3); thus the earlier arguments are applicable.
The condition b) can be formulated in terms of neighborhoods:
Remark. — If X is a locally compact space equipped with a mea-
sure, and F is any topological space, the following conditions on a function
f : X → F are equivalent:
b) for every compact set K ⊂ X , there exist a negligible set N ⊂ K
and a countable set H ⊂ F such that f (K --- N) ⊂ H ;
b0 ) for every point x ∈ X , there exist a neighborhood V x of x , a neg-
ligible set Nx ⊂ Vx , and a countable set Hx ⊂ F such that f (Vx --- Nx ) ⊂
Hx .
When F is metrizable, one can arrange that H ⊂ f (K --- N) (or that
Hx ⊂ f (Vx --- Nx ) ).
Proof. b) ⇒ b0 ): Given any x ∈ X , let Vx be a compact neighborhood
of x and apply b) to K = Vx .
b0 ) ⇒ b): Let K be a compact set in X . For each x ∈ K , choose V x ,
Nx and Hx as in b0 ); say
K ⊂ V x1 ∪ · · · ∪ V xn .
Writing N = K∩(Nx1 ∪· · ·∪Nxn ) and H = Hx1 ∪· · ·∪Hxn , N is a negligible
subset of K such that
K --- N ⊂ (Vx1 --- Nx1 ) ∪ · · · ∪ (Vxn --- Nxn ) ,
(∗ ) In TG (resp. GT), a Souslin space is defined to be a Hausdorff space (resp.
metrizable space) that is a continuous image of a Polish space.
INT IV.x98 extension of a measure. l p spaces §5

H is a countable subset of F , and

f (K --- N) ⊂ f (Vx1 --- Nx1 ) ∪ · · · ∪ f (Vxn --- Nxn ) ⊂ H1 ∪ · · · ∪ Hn = H .

When F is metrizable, by an earlier argument we can arrange that

H ⊂ f (K --- N) .
IV.68, `. 13, 14.
“ . . . which we arrange in a sequence (a n ) .”
Before having chosen a compact set K , fix a point b ∈ F for use in all
choices of K .
−1
Note that for every a ∈ X , f (a) is a measurable set; for, writing B r (a)
−1

for the closed ball of radius r and center a , f Br (a) is measurable by the
assumption a), therefore the set

−1 
∞
\ ∞
\
−1 −1

f (a) = f B1/p (a) = f B1/p (a)
p=1 p=1

is measurable (No. 4, Cor. 2 of Th. 2).

Given any compact set K ⊂ X , it will suffice to show that there exists a
measurable function gK : X → F such that gK = f almost everywhere in K ;
it will then follow from the local compactness of X that f is measurable
(see the note for `. 23–25). If K is negligible, the constant function x 7→ b
will serve for gK . Assume K is non-negligible.
By b), there exist a negligible set N ⊂ K and a countable set H ⊂ X
such that f (K --- N) ⊂ H . By assumption, K --- N 6= ∅ . As observed in
the preceding note, we can suppose that H ⊂ f (K --- N) ⊂ H .
If H is finite then H = H = f (K --- N) . Write f (K --- N) = {a 1 , . . . , an }
−1
with the ai distinct, and let Ai = (K --- N) ∩ f (ai ) , which is a measurable
set. Then the Ai form a partition of K --- N and the function g K : X → F
defined by
(
ai if x ∈ Ai for some (unique) i ∈ {1, , . . . , n}
gK (x) =
b if x ∈ (X --- K) ∪ N

is a measurable step function such that g K = f on K --- N .

Suppose henceforth that H is infinite, and write H = {a 1 , a2 , a3 , . . .}
with an ∈ f (K --- N) for all n (distinct, if we like, but it is not necessary).
For later use, for each n choose xn ∈ K --- N with f (xn ) = an .
§5 measurable functions and sets INT IV.x99

IV.68, `. 15.
“It follows from condition a) that A n,p is measurable.”
We continue with the notations established in the preceding note. Fix
an integer p > 1 . For every integer n > 1 ,


An,p = {x ∈ K --- N : d f (x), an 6 1/p}
−1

= {x ∈ K --- N : f (x) ∈ B1/p (an ) } = (K --- N) ∩ f B1/p (an ) ,

which is a measurable set. For later use, we observe that

∞
[
K --- N = An,p .
n=1

For, if x ∈ K --- N , so that f (x) ∈ f (K
--- N) ⊂ H , there exists an ele-

ment of H , say an , such that d f (x), an 6 1/p , whence x ∈ An,p . Thus
S
∞
K --- N ⊂ An,p , and the reverse inclusion is immediate from the defini-
n=1
tion of the An,p .
Incidentally, with xn ∈ K --- N chosen in the preceding note so that
f (xn ) = an , one obviously has xn ∈ An,p , thus every An,p is nonempty.
IV.68, `. 18, 19.
“ . . . the sets Bn,p are measurable, and those that are nonempty form
a partition of K --- N .”
S
∞
For fixed p , we have K --- N = An,p by the preceding note; the Bn,p
k=1
are the result of ‘disjointifying’ the A n,p in the usual way: B1,p = A1,p and

n+1
[ n
[ n
[
Bn+1,p = Ak,p --- Ak,p = An+1,p --- Ak,p .
k=1 k=1 k=1

At least B1,p = A1,p is guaranteed to be nonempty: x1 ∈ A1,p (see the

preceding note).
IV.68, `. 19–21.
“Let gm,p be the function equal to ai on the set Bi,p for 1 6 i 6 m
and equal to a constant b ∈ F on the complement of the union of these
sets ”
S
m
For every integer m > 1 , write Cm,p = Bi,p ; some of the Bi,p may
i=1
be empty, but at least B1,p = A1,p 6= ∅ (see the preceding notes), therefore
INT IV.x100 extension of a measure. l p spaces §5

Cm,p is nonempty hence eligible to be part of the domain of a function.

One has
∞
[ ∞
[
K --- N = Bi,p = Cm,p ∪ Bi,p .
i=1 i=m+1

Define gm,p : X → F by
 ∞
[

b if x ∈ X --- Cm,p = (X --- K) ∪ N ∪ Bi,p
gm,p (x) = i=m+1


ai if x ∈ Bi,p for some (unique) i ∈ {1, , . . . , m} .

IV.68, `. 21–23.
“ . . . as m tends to infinity, gm,p converges pointwise to the function f p
equal to an on Bn,p (n > 1) and to b on N ∪ { K ”
Continuing the notations of the preceding note, the sets C m,p (m =
1, 2, 3, . . .) form an increasing sequence with union K --- N .
As Bn,p may be empty for some n , the definition of f p must be
adjusted. Let’s just calculate lim gm,p (x) . If x ∈ (X --- K) ∪ N then
m→∞
gm,p (x) = b for every m , so the limit is b . On the other hand, if x ∈ K --- N ,
then x ∈ Bn,p for a unique n ; clearly gm,p (x) = gn,p (x) = an for all m > n
(by the uniqueness mentioned in the definition of g m,p ), thus the limit is an .
So the pointwise limit of the gm,p as m → ∞ is the function fp : X → F
defined by
(
b if x ∈ (X --- K) ∪ N
fp (x) =
an if x ∈ Bn,p for some (unique) n .

Thus only the nonempty Bn,p figure in the definition of fp (but their union
is still K --- N ); if Bn,p = ∅ then an does not get used in the definition
(but, being an element of f (K --- N) , it will necessarily show up as a limit
of values of the fp ).
At any rate, fp is measurable, being the pointwise limit of a sequence
of measurable step functions (No. 4, Th. 2).
IV.68, `. 23–25.
“As p tends to infinity, fp (x) tends to f (x) for every x ∈ K --- N ,
and to b for x ∈ N ∪ { K ”
Let x ∈ K --- N . Given any integer p > 1 , x belongs to B n,p for
some n . We know that for all m > n , gm,p (x) = gn,p (x) = an and so



d f (x), gm,p (x) = d f (x), an 6 1/p ;
§5 measurable functions and sets INT IV.x101


letting m → ∞ , we have d f (x), fp (x) 6 1/p , whence fp (x) → f (x) as
p → ∞ . On the other hand, if x ∈ (X --- K) ∪ N then f p (x) = b for all p ,
therefore fp (x) → b trivially. Thus the function g K : X → F defined by
(
b if x ∈ (X --- K) ∪ N
gK (x) =
f (x) if x ∈ K --- N

is the pointwise limit of the sequence of measurable functions f p , hence is

measurable.
To summarize, given any compact set K ⊂ X , we have found a mea-
surable function gK : X → F such that f = gK almost everywhere in K .
Applying this to K = Vx , where Vx is a compact (hence integrable) neigh-
borhood of any given point x ∈ X , f is measurable by the Principle of
localization (No. 2, Prop. 4).
IV.68, `. 25, 26.
“ . . . the limit of the fp is therefore measurable, and the principle of
localization proves that f itself is measurable.”
See the preceding note.
IV.69, `. 10.
“ . . . and Th. 4 may be applied.”
Some reflections on the topology of R are in order. By definition, it
is the topology generated by the open intervals of R (GT, IV, §4, No. 2).
Then R is homeomorphic to every nondegenerate closed interval [a, b] of R
(loc. cit., Prop. 2), hence is a metrizable compact space. Explicitly, let
u : R → ] − 1, 1[ be the order-preserving homeomorphism defined (for
example) by
t
u(t) = (t ∈ R) ;
1 + |t|
defining u(−∞) = −1 and u(+∞) = +1 , one obtains an order-preserving
homeomorphism u : R → [− 1, 1] . It follows that the metric on R defined
by
d(a, b) = |u(a) − u(b)| (a, b ∈ R)
is compatible with its topology. The ‘closed balls’ in [−1, 1] (for the absolute
value metric) are its closed subintervals I , therefore the closed balls in R for
−1
the metric d are the sets u (I) , that is—since u is an order isomorphism—
the closed subintervals of R .
−1
Thus, to verify condition a) of Th. 4, it suffices to show that f ([a, b])
is measurable for every closed interval [a, b] of R .
INT IV.x102 extension of a measure. l p spaces §5

Condition b) is trivially verified: since D is dense in R (because R

is dense in R ), for every compact set K ⊂ X one may take N = ∅ and
H = D.
IV.69, `. 13.
“Corollary.”
If f is lower semi-continuous then, for every a ∈ R the set

{x : − f (x) > a } = {x : f (x) 6 −a }

is closed (GT, IV, §6, No. 2, Prop. 1), therefore −f is measurable (cite
Prop. 8 with D = Q ) hence so is f (No. 3, Th. 1 and the continuity of
c 7→ −c in R ).
IV.69, `. −9, −8.

“It is clear that d f (x), gn (x) 6 2/n for all x ∈ X .”
Say x ∈ Ak , so that f (x) ∈ Bk . Say x ∈ Ci ⊂ Ak , so that

gn (x) = f (ci ) ∈ f (Ak ) ⊂ Bk .

Thus f (x), gn (x) ∈ Bk , where diam Bk 6 2/n .

IV.69, `. −2.
“The condition being obviously necessary (No. 3, Th. 1) . . . ”
The scalar function in question is a 0n ◦f , so its measurability is a special
case of No. 3, Th. 1; the following argument is perhaps more transparent:
Proposition. If f : X → F is measurable and u : F → G is continuous,
then u ◦ f : X → G is measurable (where X is a locally compact space
equipped with a measure, and F, G are any topological spaces).
Proof. Immediate from the definition of measurability (No. 1, Def. 1).
But beware: if u : X → X is continuous and f : X → F is measurable,
then the composite function f ◦ u : X → F need not be measurable (No. 3,
Remark 2).
IV.70, `. 1–3.
“ . . . it will suffice by the principle of localization to prove that, for every
compact subset K of X and every closed ball B ⊂ F , with center a and
−1
radius r , the set A = K ∩ f (B) is measurable ”
−1
The objective is to show that f (B) is measurable. More generally,
Proposition. For a subset S ⊂ X , the following conditions are equiva-
lent:
§5 measurable functions and sets INT IV.x103

a) S is measurable;
b) K ∩ S is measurable for every compact set K ⊂ X ;
c) K ∩ S is integrable for every compact set K ⊂ X .
Proof. a) ⇔ c): No. 1, Prop. 3.
c) ⇒ b): Suppose S satisfies c). Given K ⊂ X compact. To prove that
K ∩ S is measurable, it suffices by c) ⇒ a) to show that for every compact
K0 ⊂ X , the set K0 ∩(K∩S) is integrable; indeed, K0 ∩(K∩S) = (K0 ∩K)∩S
is integrable by c).
b) ⇒ c): Suppose S satisfies b). Given K ⊂ X compact. We know that
K ∩ S is measurable, hence K ∩ S = K ∩ (K ∩ S) is integrable by a) ⇒ c).
The following corollary is sharpened in No. 6, Th. 5 below.
Corollary. For a subset S ⊂ X ,

S integrable ⇒ S measurable.
Proof. Suppose S is integrable. Given any compact set K ⊂ X , one
knows that K is integrable (§4, No. 6, Cor. 1 of Prop. 10), hence so is K ∩ S
(§4, No. 5, Prop. 7, 2◦ ), thus S is measurable by the above proposition.
IV.70, `. 3, 4.
“ . . . for every z ∈ F ,

|z| = sup |hz, a0n i|/|a0n | ”

n

Since F is separable, the unit ball B 0 = {z0 ∈ F0 : |a0 | 6 1 } of F0

is a metrizable compact space for the weak topology σ(F 0 , F) , hence has a
countable subset D0 that is dense in B0 for the weak topology (TVS, III,
§3, No. 4, Cor. 2 of Prop. 6).
We assume given such a countable subset, indexed as a sequence, D 0 =
0
{an : n = 1, 2, 3, . . . } , having the further property that for every n , the
function

(∗) x 7→ hf (x), a0n i (x ∈ X)

is measurable. The asserted formula suggests replacing D 0 by D = D0 −{0} ;

to be sure that the weak closure of D includes 0 (hence is equal to B 0 ),
one can add to the original subset D0 the vectors (1/n)a0m (n = 2, 3, . . .) ,
where a0m is a nonzero element of D0 , noting that the functions (∗) for the
added vectors are also measurable.
In other words, we can suppose that the vectors a 0n are all nonzero.
The asserted formula is then the formula (5) of TVS, IV, §1, No. 3, Cor. of
Prop. 8; as the D of that corollary is part of the conclusion rather than part
INT IV.x104 extension of a measure. l p spaces §5

of the hypothesis, it may be useful to review the proof. For every z ∈ F

define
ρ(z) = sup |hz, a0n i|/|a0n | = sup |hz, a0 i|/|a0 | ;
n a0 ∈D

we are to show that ρ(z) = |z| . Fix z ∈ F . For every n ,

|hz, a0n i|/|a0n | = |a0n (z)|/|a0n | 6 (|a0n | · |z|)/|a0n | = |z| ,
therefore ρ(z) 6 |z| . On the other hand,
|z| = sup |hz, a0 i|
a0 ∈B0

(loc. cit., Prop. 8, (i)); to show that |z| 6 ρ(z) , it suffices to show that
|hz, a0 i| 6 ρ(z) for every a0 ∈ B0 . Indeed, if a0 ∈ B0 then a0 belongs
to the weak closure of D , hence there exists a sequence in D , say a 0nk
(k = 1, 2, 3, . . . ) such that a0nk → a0 weakly as k → ∞ . In particular
hz, a0nk i → hz, a0 i , hence |hz, a0nk i| → |hz, a0 i| ; but
|hz, a0nk i| = (|hz, a0nk i|/|a0nk |) · |a0nk | 6 |hz, a0nk i|/|a0nk | 6 ρ(z) ,
and passage to the limit yields |hz, a 0 i| 6 ρ(z) .
Thus, writing b0n = a0n /|a0n | (a unit vector in F0 ), one has
|z| = sup |hz, b0n i|
n

for all z ∈ F .
IV.70, `. 5, 6.
“ A is thus the intersection of K and the sets defined by
|hf (x), a0n i − ha, a0n i| 6 r |a0n | ”

To simplify the notation slightly, write b 0n = a0n /|a0n | as in the preceding

note. Recalling that B is the closed ball B r (a) ,
−1
A = K ∩ f (B) = {x ∈ K : f (x) ∈ B }
= {x ∈ K : |f (x) − a| 6 r }
= {x ∈ K : sup |hf (x) − a, b0n i| 6 r }
n
= {x ∈ K : |hf (x), b0n i − ha, b0n i| 6 r for all n }
\∞
=K∩ {x : hf (x), b0n i 6 r + ha, b0n i}
n=1
∞
\
∩ {x : hf (x), b0n i > −r + ha, b0n i};
n=1
§5 measurable functions and sets INT IV.x105

since each function x 7→ hf (x), b0n i = hf (x), a0n i/|a0n | is by assumption

measurable, A is the intersection of (countably many) measurable sets by
Prop. 7, hence is measurable (No. 4, Cor. 2 of Th. 2).
IV.70, `. 16–18.
“ . . . we may (on account of b)) suppose, after modifying f if necessary
on a negligible set, that f (K) ⊂ H , where H is a countable subset of F . ”
−1
The objective is to show that K∩ f (B) is measurable (where B = Br (a)
is the ball with center a and radius r ). By b), there exists a negligible set
N ⊂ K and a countable set H ⊂ F such that f (K --- N) ⊂ H . Fix an
element b of H and define g : X → F by
(
b for x ∈ N
g(x) =
f (x) for x ∈ X --- N .

Then g = f almost everywhere, therefore g also satisfies the condition a)

(No. 2, Prop. 6), and it satisfies condition b) for every compact set
K0 ⊂ X (not just for K ) because the union of two negligible sets is neg-
ligible. Moreover, for the particular compact set K , we have
g(K) = g(N) ∪ g(K --- N) = {b} ∪ f (K --- N) ⊂ H .
−1
If we succeed in showing that K ∩ g (B) is measurable, it will follow from
−1 −1 −1
K ∩ f (B) = [(K --- N) ∩ f (B)] ∪ [N ∩ f (B)]
−1 −1
= [(K --- N) ∩ g (B)] ∪ [N ∩ f (B)]
 −1
 −1
= (K --- N) ∩ K ∩ g (B) ∪ [N ∩ f (B)]
−1
that K ∩ f (B) is also measurable (being the union of two measurable sets).
IV.70, `. 20, 21.
“ . . . every continuous linear form on V is the restriction of a form
a ∈ F0 ”
0

The assertion is immediate from the Hahn–Banach theorem; it follows

that when F is replaced by V , the condition a) continues to hold (with g
as in the preceding note) for the function g 0 : X → V defined by x 7→ g(x)
(as of course does condition b)).
The argument in the ‘sufficiency’ part of Prop. 10 then shows that the
−1
set A = K ∩ g0 (B) is measurable. But
−1
A = {x ∈ K : g0 (x) ∈ B } = {x ∈ K : g(x) ∈ B } = K ∩ g (B) ,
INT IV.x106 extension of a measure. l p spaces §5

−1 −1
thus K ∩ g (B) is measurable, hence so is K ∩ f (B) as argued in the
preceding note.
IV.70, `. 21, 22.
−1
“ . . . the same reasoning as in Prop. 10 then shows that K ∩ f (B) is
measurable.”
See the preceding note.
IV.70, `. −10, −9. Q
“We may regard F as a subspace of a countable product En of
n
Banach spaces (TVS, II, §4, No. 3)”
As the proof of ‘necessity’ is immediate (see the note for IV.69, `. −2),
one is concerned here with the proof of ‘sufficiency’.
The cited reference refers back to an earlier section (TVS, II, §1, No. 3);
let us review the argument, which entails some subtleties.
If F is any locally convex space (but not necessarily Hausdorff), its
topology can be generated by a family of continuous semi-norms (TVS, II,
§4, No. 1, Cor. of Prop. 1), say (pι )ι∈I . For each ι , the set

Nι = {a ∈ F : pι (a) = 0 }

is a closed linear subspace of F , the quotient topology on F ι = F/Nι (TVS,

II, §4, No. 4, Example I) coincides with the norm topology for the norm
on Fι deduced from pι (TVS, II, §1, No. 3); thus the quotient mapping
uι : F → F Qι is a continuous linear mapping of F into the normed space F ι .
Writing Fι for the product locally convex space (TVS, II, §4, No. 3),
Qι
Q
u : F → Fι for the mapping u(a) = uι (a) ι∈I and prι : Fχ → Fι for
ι χ∈I
the ι’th projection mapping, so that pr ι ◦ u = uι for all ι , the continuity of
the uι implies that of u (GT, I, §4, No. 1, Prop. 1).
For F to be Hausdorff, it is necessary and sufficient thatTthe family
(pι ) be separating ( pι (a) = 0 for all ι ⇒ a = 0 ), equivalently Nι = {0} ,
ι
which is equivalent to the
Qinjectivity of u . Writing M = u(F) , regarded as a
topological subspace of Fι , one sees that u implements a topological vec-
ι
tor space isomorphism F → M (TVS, II, §1, No. 3 and I, §1, No. 7, Prop. 7);
the crux of the matter is that the topology of F is the coarsest locally convex
topology that renders continuous the semi-norms p ι (equivalently, the lin-
ear mappings uι ; equivalently, the linear mapping u ). {Thus the topology
generated by a separating family of semi-norms is an ‘initial topology’ with
respect to locally convex topologies (TVS, II, §4, No. 3, Prop. 4).}
§5 measurable functions and sets INT IV.x107

Assuming F Hausdorff,
Q let Eι be the Banach space
Q obtained by com-
pleting Fι . Since Fι is a topological subspace of Eι (GT, I, §4, No. 1,
ι ι Q
Cor. of Prop. 3), M is also a topological subspace of Eι (transitivity of
ι
induced topologies); thus, every Hausdorff locally convex space is isomor-
phic as a topological vector space to a (topological) subspace of a product
of Banach spaces (TVS, II, §1, No. 3, Prop. 3). {Since a continuous linear
mapping of topological vector spaces is uniformly continuous (GT, III, §3,
No. 1, Prop. 3) it follows that if F is Hausdorff then the uniform structures
of F and M are isomorphic; if, moreover, FQis complete, then so is M , con-
sequently M is a closed linear subspace of Fι (GT, II, §3, No. 4, Prop. 8)
Q ι
hence also of Eι (TVS, II, §4, last paragraph of No. 3).
ι
Consider now the space F of the present corollary. Since F is metriz-
able, it has a countable fundamental system (V n ) of neighborhoods of 0
(TVS, I, §3, No. 1), and since F is locally convex, the V n may be taken
to be closed, balanced and convex (TVS, II, §4, No. 1). If p n is the gauge
of Vn , the topology of F is generated by the family of semi-norms (p n ) (loc.
cit., Cor. of Prop. 1; see also the note for III.39,
Q `. 8–11). With notations
as in the foregoing discussion, if u : F → Fn is the mapping defined by
n

−1
u(a) = un (a) , where Fn = F/ pn (0) (a normed space) and un : F → Fn is
the quotient mapping, then u defines a topological vector spaceQisomorphism
of F onto M = u(F) regarded as a topological subspace of Fn . If En
n
is the Banach space
Q completion of Fn , then u may be regarded as linear
mapping F → En that defines a topological vector space isomorphism
n Q
of F onto the topological subspace M = u(F) of En . Since F has a
n
countable dense subset, so does the subspace pr n M of En (n = 1, 2, 3, . . .);
replacing En by the closure of prn M in En , we can suppose that the En
are separable Banach spaces. Identifying F with M , we have shown that
F is a topological linear subspace of the product of a sequence of separable
Banach spaces.

IV.70, `. −8, −7.

“For every n , the mapping prn ◦ f is then measurable by Prop. 10 ”
Q
With notations as in the preceding note, write E = En and regard
n
F as a topological linear subspace of E .
Since f : X → F , F ⊂ E and prn : E → En , the composition that is
possible is prn ◦ θ ◦ f , where θ : F → E is the canonical injection; since
prn ◦ θ = prn F , it is the composition (prn F) ◦ f that we are to show is
INT IV.x108 extension of a measure. l p spaces §5

measurable. Let us write un = prn F , which is a continuous linear mapping
F → En , so that un ◦ f : X → En . To show that un ◦ f is measurable, it
suffices by Prop. 10 to show that, given any a 0n ∈ E0n , the function

(∗) x 7→ h(un ◦ f )(x), a0n i (x ∈ X)

is measurable. Now,

h(un ◦ f )(x), a0n i = hun (f (x)), a0n i = hf (x), t un a0n i = hf (x), a0n ◦ un i ,

where t un : E0n → F0 is the transpose of un ; since t un a0n ∈ F0 , by hypoth-

esis the function
x 7→ hf (x), t un a0n i (x ∈ X)
is measurable, in other words the function (∗) is indeed measurable.

To summarize, for every index n , the function (pr n F) ◦ f : X → En is
measurable.
IV.70, `. −7.
“ . . . therefore f is measurable by No. 3, Th. 1.”
Conserving the notations of the preceding note, we have a sequence of
measurable functions Q un ◦ f : X → En , where un = prn F . For all x ∈ X,
f (x) ∈ F ⊂ E = En , hence
n




f (x) = prn (f (x)) = (prn F)(f (x)) = (un ◦ f )(x) ;

it follows from the cited Th. 1 that, taking u to be the canonical injection
f (X) → E , the mapping u ◦ f : X → E is measurable. But f takes its values
in F , and we wish to show that f : X → F is measurable.
Let K ⊂ X be compact. By the measurability of u ◦ f , there exist
a negligible set N ⊂ K and a partition (K i )i∈I of K --- N into compact
sets Ki on each of which u ◦ f is continuous, that is, the function


x 7→ (u ◦ f )(x) = u f (x) = f (x) ∈ F (x ∈ Ki )

is continuous, whence the measurability of f .

IV.71, `. 4, 5.
“ . . . the polar sets Vn◦ are equicontinuous and their union is all of F 0 . ”
The inclusion Vn◦◦ ⊃ Vn shows that Vn◦◦ is a neighborhood of 0 in F ,
hence Vn◦ is equicontinuous (TVS, III, §3, No. 5, Prop. 7).
Let a0 ∈ F0 . Since a0 is continuous at 0 and the Vn are basic, there
exists an index m such that a0 (Vm ) ⊂ {λ : |λ| 6 1 } , that is, |a0 (x)| 6 1
for all x ∈ Vm ; since Vm is circled, this means that a0 ∈ Vm ◦
(TVS, II, §8,
No. 4).
§5 measurable functions and sets INT IV.x109

IV.71, `. 9.
“ . . . each of the Sy is measurable ”
Sy is the inverse image, under the measurable mapping x 7→ hy, f (x)i ,
of the closed set {c : |c| 6 1 } in the field of scalars, hence is measurable by
No. 5, Prop. 7.
IV.71, `. 9, 10.
“ Xn is the intersection of the countable family of S y for y ∈ D ∩ Vn .”
Since D = F and Vn is open in F , one has Vn = Vn ∩ D ⊂ Vn ∩ D
(GT, I, §1, No. 6, Prop. 5), whence V n = Vn ∩ D . The following computa-
tion then depends on a formula for the polar of a balanced set (cf. TVS, II,
§8, No. 4) and the continuity of the linear forms f (x) :

x ∈ Xn ⇔ f (x) ∈ Vn◦
⇔ |hy, f (x)i| 6 1 for all y ∈ Vn
⇔ |hy, f (x)i| 6 1 for all y ∈ V n
⇔ |hy, f (x)i| 6 1 for all y ∈ Vn ∩ D
⇔ |hy, f (x)i| 6 1 for all y ∈ Vn ∩ D
⇔ x ∈ Sy for all y ∈ Vn ∩ D ,
T
that is, Xn = Sy .
y∈Vn ∩D

IV.71, `. 10–13.
“ . . . for every compact subset K of
X and every ε > 0 , there exists an
integer n such that |µ| K --- (K ∩ Xn ) 6 ε/4
, and then a compact subset
K1 of K ∩ Xn such that |µ| (K ∩ Xn ) --- K1 6 ε/4 ”
Since the Xk are measurable, the K ∩ Xk form an increasing sequence
of integrable (No. 1, Prop. 3) sets with union K , therefore


|µ| K --- (K ∩ Xk ) → 0

(§4, No. 5, Cor. of Prop. 7), whence the existence of n . The existence of K 1
then follows from the ‘inner regularity’ of |µ| (§4, No. 6, Cor. 1 of Th. 4).
IV.71, `. 18, 19.
“ . . . the restriction of f to K2 is therefore continuous ”
Let (xj ) be a directed family in K2 such that xj → x ∈ K2 ; we are
to show that f (xj ) → f (x) in F0 for σ(F0 , F) , that is, for the topology
of pointwise convergence in F . We know that hy, f (x j )i → hy, f (x)i for
each y ∈ D , by the choice of K2 ; thus f (xj ) → f (x) for the topology of
INT IV.x110 extension of a measure. l p spaces §5

pointwise convergence in D . But f (x j ) and f (x) belong to the equicontin-

uous set f (K2 ) ⊂ Vn◦ , therefore f (xj ) → f (x) for the topology of pointwise
convergence in F (GT, X, §2, No. 4, Th. 1).
IV.71, `. 20, 21.
“If z0 is a continuous linear form on F , its restriction z 0n to Fn is
continuous ”
This is where the hypothesis F = lim Fn gets used: the canonical injec-
−→
tion in : Fn → F is continuous
 (TVS, II, §4, No. 4, Example II), therefore
the composite z0 ◦ in = z0 Fn is continuous. But it is not assumed that F n
is a topological subspace of F (TVS, loc. cit., Remark ).
IV.71, `. 21–23.
“ . . . the dual F0 of Q
F may be identified (algebraically) with a linear
subspace of the product F0n , and then prn (z0 ) = z0n .”
n
Q 0
Define a linear mapping u : F0 → Fn by u(z0 ) = (z0n ) , where
 n
z0n = Sz0 Fn ∈ F0n (see the preceding note). The injectivity of u follows
from Fn = F . Identifying F0 with the linear subspace u(F0 ) of the
n Q
product space via u , u becomes the canonical injection F 0 → F0n , z0 be-
n
comes (z0n ) , and prn z0 = z0n .
IV.71, `. 24, 25.
“ . . . the topology σ(F0 , F) is none other than the topology induced by
the product topology of the topologies σ(F 0n , Fn ) .”
S
Recalling that F = Fn , if (z0j ) is a directed family in F0n and if
n
z0 ∈ F0 , then

z0j → z0 in F0 for σ(F0 , F) ⇔ ha, z0j i → ha, z0 i for all a ∈ F

⇔ (∀ n) ha, z0j i → ha, z0 i for all a ∈ Fn
⇔ (∀ n) (∀ a ∈ Fn ) ha, (z0j )n i → ha, (z0 )n i
⇔ (∀ n) prn z0j → prn z0 for σ(F0n , Fn )
Y
⇔ z0j → z0 in F0n ,
n
Q
where F0n is equipped with the indicated product topology (GT, I, §7,
n
No. 6, Cor. 1 of Prop. 10).
IV.71, `. −11.
“ han , prn ◦ f i = han , f i ”
§5 measurable functions and sets INT IV.x111

For all an ∈ Fn and x ∈ X , one has

han , prn ◦ f i(x) = han , (prn ◦ f )(x)i


= han , prn f (x) i

= han , f (x)Fn i
= han , f (x)i = han , f i(x)
by the definition of the notations.
IV.71, `. −4 to −2.
“ . . . if f ∈ LFp then there exists a sequence (gn ) of continuous func-
tions with compact support that converges almost everywhere to f (§3,
No. 4, Cor. 2 of Th. 3) ”
Recall that LFp is defined to be the closure of the subset K (X; F) of
p
FF (X, µ) for the topology of convergence in mean of order p (§3, No. 4,
Def. 2).
IV.72, `. −12, −11.
Corollary 1. — For a set to be integrable, it is necessary and sufficient
that it be measurable and have finite outer measure.
By definition, A is measurable (resp. integrable) if and only if ϕ A is
measurable (resp. integrable); and |µ|*(A) = |µ|*(ϕ A ) . Quote Th. 5.
IV.73, `. −2.
“ g 0 is upper semi-continuous on X ”
It is the same to show that −g 0 is lower semi-continuous. Let k ∈ R
and let A = {x ∈ X : − g 0 (x) 6 k } = {x ∈ X : g 0 (x) > −k } ; the
problem is to show that A is a closed subset of X (GT, IV, §6, No. 2,
Prop. 1). If −k 6 0 then A = X (because g 0 > 0 ). If −k > 0 then, since
g 0 = 0 on X --- Y , one has A = {y ∈ Y : g(y) > −k } ⊂ Supp(g) ; since
g ∈ K (Y) , Supp(g) is compact, so its closed subset A is compact in Y ,
hence also in X , therefore A is closed in X (GT, I, §9, No. 3, Prop. 4).
Addendum. One observes that the proof, that g 0 is upper semi-continuous
on X when g ∈ K+ (Y) , does not require that Y be locally compact—it
can be any subspace of X (and X can be any Hausdorff space).
When Y is locally compact, it contains enough compact sets for every
point to have a fundamental system of compact neighborhoods, and K + (Y)
contains enough functions to separate the points of Y (III, §1, No. 2, Lem-
ma 1). But if Y is an arbitrary subset of X , does it contain enough compact
sets, and K (Y; C) enough functions, to be useful?
At any rate, K (Y; C) is, for arbitrary Y , a complex vector space
closed under complex conjugation, and K (Y; R) is a Riesz space, so that
every g ∈ K (Y; C) is a linear combination of four elements of K + (Y) .
INT IV.x112 extension of a measure. l p spaces §5

More about such matters in the next note.

IV.74, `. 3–6. 
“ . . . denote by µY or µY, the measure defined by the formula
Z Z
(1) g dµY = g 0 dµ

for every function g ∈ K (Y; C) , where g 0 denotes the function equal to g

on Y and to 0 on X --- Y . ”
The mapping K (Y; C) → F(X; C) defined by g 7→ g 0 is linear and
positive ( g > 0 ⇒ g 0 > 0 ) ; writing g = g1 −g2 +ig3 −ig4 with gj ∈ K+ (Y) ,
it follows from the foregoing that g 0 = g10 − g20 + ig30 − ig40 is µ-integrable, so
that g 7→ g 0 is a linear mapping K (Y; C) → LC1 (Y, µ) R . 0One may therefore
define a linear form µY : K (Y; C) → R C by µ Y (g) = g dµ .
If, moreoever, µ > 0 , then g 7→ g 0 dµ defines a positive linear form on
K (Y; R) , that is, a positive measure on Y (III, §1, No. 5, Th. 1); we set
aside, for the moment, the question of whether µ Y is a (complex) measure.
What if µ is not positive? Of course µ = µ 1 − µ2 R+ iµ3 − iµ4 with
the µj positive measures 6 |µ| , but, absent an analysis of d(µ+ν) (see the
Note for V.10, `. 13, 14), it is not clear how to exploit this expression for µ
to show that µY is a linear combination of (positive) measures. Instead, let
us show that µY is a measure by verifying the criterion of Ch. III, §1, No. 3:
given any compact subset K of Y , there exists a constant M K > 0 such
that
|µY (g)| 6 MK kgk for all g ∈ K (Y, K; C) ;
indeed, K is also a compact subset of X , its characteristic function ϕ K
is µ-integrable, and, for every g ∈ K (Y; C) , one has g 0 = ϕK g 0 and
|g 0 | 6 kgkϕK , whence (§4, No. 2, Prop. 2)
Z  Z Z
 
|µY (g)| =  ϕK g 0 dµ 6 |ϕK g 0 | d|µ| 6 ϕK kgk d|µ| = kgk · |µ|(K) ,

thus MK = |µ|(K) meets the requirements (§4, No. 6, Cor. 1 of Prop. 10) .
Addendum. Pursuing the addendum to the preceding note, suppose Y
is an arbitrary subspace of X , and let g ∈ K + (Y) . As observed there, g 0 is
lower semi-continuous on X ; it is also bounded and has compact support.
The argument in the preceding note shows that for every k ∈ R , the set
{x ∈ X : g 0 (x) > k } is either equal to X or is a compact subset of X ,
hence is measurable for µ (No. 1, Cor. 1 of Prop. 3); therefore g 0 is measur-
able (No. 5, Prop. 8). Since every g ∈ K (Y; C) is a linear combination of
four elements of K+ (Y) , it follows from the linearity of g 7→ g 0 that g 0 is
§5 measurable functions and sets INT IV.x113

bounded, measurable, and has compact support; therefore |µ|*(g 0 ) < +∞ ,

g 0 is µ-integrable (No. 6, Th. 5), and one can define a linear form
µY : K (Y; C) → C by µY (g) = µ(g 0 ) ; and when µ > 0 , g 7→ µY (g 0 )
defines a positive linear form on K (Y; R) .
But . . . we can’t call µY a measure (even if µ > 0 ), because the term
‘measure’ has been defined, and its properties developed, only for locally
compact spaces. At any rate, the argument used for locally compact Y
shows that for every compact set K in Y , there exists a constant M K > 0
such that |µY (g)| 6 MK kgk for all g ∈ K (Y, K; C) .
A theory of measures on Hausdorff spaces is developed in Chapter IX,
permitting induced measures on measurable subsets (IX, §2, No. 1, Def. 1).
The correspondence g 7→ g 0 there seems to be the analogue of g 7→ g 0 , but
I have not explored further the relationship between the two contexts.
IV.74, `. −13.
“ (i) For every compact (resp. open) subset H of K , µ K (H) = µ(H) . ”
We remark that this will imply that

(∗) µK (B) = µ(B)

for every Borel set B in K , in other words, for every Borel set B in X such
that B ⊂ K (GT, IX, §6, No. 3, Remark 2).
For, writing B(T) for the set of Borel sets in a topological space T
(the tribe generated by its open sets, equivalently by its closed sets), we have

B(K) = {B ∈ B(X) : B ⊂ K } = {B ∩ K : B ∈ B(X) } ,

concisely B(K) = B(X) ∩ K . Every set in B(X) (resp. B(K) ) is

µ-measurable (resp. µK -measurable) by No. 4, Cor. 3 of Th. 2, thus every
set in B(K) is both µK -measurable and µ-measurable; and since µ K (K)
and µ(K) are finite (moreover equal by (i)), every B ∈ B(K) is both
µK -integrable and µ-integrable (No. 6, Cor. 1 of Th. 5).
The set function B 7→ µK (B) ( B ∈ B(K) ) is a ‘finite measure’ in the
sense of Halmos (Measure theory, Van Nostrand, New York, 1950; reprinted
by Springer-Verlag); in particular, if ( B n ) is any sequence of pairwise disjoint
sets in B(K) then
[  X
µK Bn = µK (Bn )
n n

(§4, No. 5, Prop. 9), and

µK (A --- B) = µK (A --- A ∩ B) = µK (A) − µK (A ∩ B)

INT IV.x114 extension of a measure. l p spaces §5

for all A, B ∈ B(K) (loc. cit., Prop. 7). The same is true of the set function
B 7→ µ(B) ( B ∈ B(K) ). Let C be the set of all compact (i.e., closed)
subsets of K . By (i), (∗) holds for every set in C ; since C is closed under
finite unions and intersections, and since the tribe of subsets of K generated
by C is B(K) , it follows that (∗) holds for every set in B(K) (Th. 51.F in
Halmos’ book, p. 223; or S.K. Berberian, Measure and integration, Th. 2 on
p. 185, Macmillan, New York, 1965, reprinted by Chelsea). The heart of the
matter is that the clan generated by C is the set of all finite disjoint unions
of sets C --- D , where C, D ∈ C and D ⊂ C .
IV.74, `. −9.
“ (i) We can restrict ourselves to the case that H is compact. ”
Suppose the assertion proved for compact sets, so that, in particular,
µK (K) = µ(K) , and let J be an open set in K (that is, a set of the form
U ∩ K , where U is an open set in X ); then the set H = K --- J is compact,
therefore by assumption

µK (K --- J) = µ(K --- J) ,

that is (§4, No. 5, Prop. 7),

µK (K) --- µK (J) = µ(K) − µ(J) ,

whence µK (J) = µ(J) .

{It follows that µK (H) = µ(H) for every Borel set H in K (see the
preceding note).}
IV.74, `. −5, −4.
“ gα0 is upper semi-continuous ”
The argument is given in the note for IV.73, `. −2.
IV.75, `. 1, 2.
“(ii) If N is µ-negligible then, for every ε > 0 , there exists a relatively
compact open neighborhood U of N in X such that µ(U) 6 ε ”
Since µ*(N) = 0 , there exists an open set V in X such that N ⊂ V
and µ*(V) 6 ε (§1, No. 4, Prop. 19). But N ⊂ K , where K is compact;
let W be a relatively compact open set in X such that K ⊂ W . Set
U = V ∩ W ; U is an open neighborhood of N , U ⊂ W shows that it is
relatively compact, and µ*(U) 6 µ*(V) 6 ε . Since U is integrable (§4,
No. 6, Cor. 1 of Prop. 10), we may write µ(U) for µ*(U) .
IV.75, `. 7, 8.
“ (iii) For every open set U in K that intersects S , by hypothesis
µK (U ∩ S) 6= 0 ”
§5 measurable functions and sets INT IV.x115

Arguing contrapositively, assuming µ K (U ∩ S) = 0 let us show that

U ∩ S = ∅ . Since µ (hence also µK ) is positive, U ∩ S is µK -negligible (§2,
No. 2, Def. 2). On the other hand, since S is closed in K , U ∩ (K --- S)
is an open set in K that is contained in K --- S , therefore U ∩ (K --- S) is
µK -negligible (loc. cit., Prop. 5). It follows that the union


(U ∩ S) ∪ U ∩ (K --- S) = U ∩ K = U

is a µK -negligible open set in K , consequently U ⊂ K --- S by the cited

Prop. 5, that is, U ∩ S = ∅ .
IV.75, `. 9.
“ µS (U ∩ S) 6= 0 by (i) ”
By the transitivity of induced topologies, U ∩ S is also an open set in
the compact subset S of X , therefore µ S (U ∩ S) = µ(U ∩ S) by (i) with S
playing the role of K .
IV.75, `. 10.
“ . . . this proves that Supp(µS ) = S . ”
By the foregoing argument, if U is an open set in K such that U∩S 6= ∅,
one has µS (U ∩ S) 6= 0 . The sets U ∩ S ( U open in K ) are the open sets
in the subspace S of X (by the transitivity of induced topologies). Thus,
the only µS -negligible open set in S is the empty set, therefore S --- ∅ = S
is the support of µS (§2, No. 2, Prop. 5).
IV.75, `. 13–15.
“ . . . by definition . . . ”
Ch. III, §1, No. 6, formula (12).
IV.75, `. −13, −12.
“ . . . let K be the support of f and let U be a compact neighborhood
of K in X such that |µ|(U --- K) 6 ε ”
Since |µ|(K) = |µ|*(K) < +∞ (§4, No. 6, Cor. 1 of Prop. 10), there
exists an open set G such that K ⊂ G and |µ|*(G) 6 |µ|(K) + ε (§1,
No. 4, Prop. 19). For each x ∈ K , choose an open neighborhood V x
of x with Vx compact and Vx ⊂ G ; cover K by a finite number of such
neighborhoods, and let V be their union. Thus K ⊂ V ⊂ V ⊂ G ; since all
sets in sight are integrable (§4, No. 6, Prop. 10), |µ|(V --- K) 6 |µ|(G --- K) =
|µ|(G) − |µ|(K) 6 ε (§4, No. 5, Prop. 7) and U = V meets the requirements.
IV.75, `. −12 to −10.
“ . . . by Urysohn’s theorem, there exists a function f 1 ∈ K+ (X) , ex-
tending f , with support contained in U and such that kf 1 k = kf k . ”
INT IV.x116 extension of a measure. l p spaces §5

It is straightforward to define a continuous function on X , with compact

support contained in U , that agrees with f on K ; the trick is to ensure
the the extension agrees with f on all of Y , i.e., is equal to 0 on Y --- K .
The following argument is terribly long-winded. There must be a shortcut;
I did not find it. (On the plus side, the argument works for any subspace Y
of X .)
Theorem. Let X be a locally compact space, Y any subspace of X
(not necessarily locally compact), and let f ∈ K (Y; R) . Then f may be
extended to a function f1 ∈ K (X; R) , such that if [a, b] is the smallest
closed interval in R containing 0 and the compact set f (Y) , then also
f1 (X) ⊂ [a, b] ; consequently kf1 k = kf k , and if f > 0 then also f1 > 0 .
Moreover, if K is the (compact) support of f in Y , and U is a neigh-
borhood of K in X , one can arrange that the support of f 1 is contained
in U .
Proof. If A ⊂ Y , we write ClY (A) for the closure of A in Y; thus
ClY (A) = A ∩ Y (GT, I, §3, No. 1, Prop. 1). Since the support K of f is
compact in Y , it is also compact in X , hence is closed in X .
From the local compactness of X , we know that the compact neighbor-
hoods of K in X are basic, so we can suppose that U is compact. Note
that K ⊂ U ∩ Y . Define a subset Z ⊂ X by

Z = K ∪ U ∩ Y --- K ;

since K = K , we have

Z = K ∪ U ∩ Y --- K = K ∪ (U ∩ Y --- K) = U ∩ Y ⊂ U = U ,

thus Z = U ∩ Y is a compact subset of the compact subspace U of X .

Let us analyze the intersection of the two terms defining Z :

K ∩ U ∩ Y --- K = (K ∩ Y) ∩ U ∩ Y --- K


= K ∩ Y ∩ U ∩ Y --- K
= K ∩ ClY (U ∩ Y --- K) ;

since f = 0 on Y --- K ⊃ U ∩ Y --- K and f is continuous on Y , it

follows that f = 0 on ClY (U ∩ Y --- K) , hence on K ∩ ClY (U ∩ Y --- K) =
K ∩ U ∩ Y --- K , therefore we may define a function f 0 : Z → R by the
formula (
f (x) if x ∈ K
f0 (x) =
0 if x ∈ U ∩ Y --- K .
Since f0 is continuous on each of its closed subsets K and U ∩ Y --- K ,
it is continuous on their union Z (GT, I, §3, No. 2, Prop. 4). One has
§5 measurable functions and sets INT IV.x117

K ⊂ Z ⊂ U ; the domain Z of f0 neither visibly contains nor is contained

in Y , so f0 may not extend nor be extended by f —all we’re sure of is that
they agree on K .
From the definition of Z , it is clear that

Z --- K ⊂ U ∩ Y --- K ,

therefore f0 = 0 on Z --- K ; one can therefore redescribe the continuous

function f0 : Z → R by the formula
(
f (x) if x ∈ K
f0 (x) =
0 if x ∈ Z --- K ,

where Z = U ∩ Y , equipped with the topology induced by that of X (or

of U ).
The sets f (Y) and f0 (Z) can differ only by 0 ; for, they both contain
f (K) and both are contained in f (K) ∪ {0} . Let [a, b] be the smallest
closed interval of R containing 0 and f (K) ; then [a, b] contains f (Y) and
f0 (Z) , kf k = kf0 k = max{|a|, |b|} , and f > 0 ⇔ f0 > 0 ⇔ a > 0 . Let us
regard f0 as a function Z → [a, b] .
Now, the compact space Z is a closed subset of the normal space U
(GT, IX, §4, No. 1, Prop. 1); by the proof of Urysohn’s theorem (loc. cit.,
No. 2, Th.
 2), there exists a continuous function F 0 : U → [a, b] such

that F0 Z = f0 (a result also known as the ‘Tietze extension theorem’). In
particular, F0 K = f0 K = f K .
Let V be an open set in X such that K ⊂ V ⊂ U . There exists a
continuous function h : X → [0, 1] such that h(x) = 1 for x ∈ K , and
h(x) = 0 for x ∈ X --- V (III, §1, No. 2, Lemma 1); the cited Lemma 1
provides such a function with compact support contained in V , but in any
case Supp(h) ⊂ V ⊂ U = U also shows that h has compact support.
Define a function f1 : X → R by
(
h(x)F0 (x) if x ∈ U
f1 (x) =
0 if x ∈ X --- U .

Since h = 0 on X --- V ⊃ U --- V , we have f1 = 0 on U --- V , hence

f1 = 0 on
(X --- U) ∪ (U --- V) = X --- V ;
  
thus f1 X --- V = 0 and f1 U = (hU)F0 are continuous, where X --- V
and U are closed sets in X with union X , therefore f 1 is continuous on X .
Since 0 ∈ [a, b] , [a, b] contains the segment joining 0 with any of its points;
it follows that f1 (X) ⊂ [a, b] .
INT IV.x118 extension of a measure. l p spaces §5
  
Since   
 K ⊂ U and h = 1 on K , we have f1 K = F0 K = f K ; to show

that f1 Y = f , we need only show that f1 = 0 on Y --- K .
Let x ∈ Y --- K . If x ∈ X --- U , then f1 (x) = 0 by the definition
of f1 ; whereas if x ∈ U then x ∈ U ∩ Y --- K ⊂ Z , therefore

f1 (x) = h(x)F0 (x) = h(x)f0 (x) = h(x) · 0 = 0

by the definitions of f1 and f0 .

IV.75, `. −8.
“ . . . then h ∈ K (Y; C) ”
From |h| 6 f one infers Supp(h) ⊂ Supp(f ) = K , thus Supp(h) is a
closed subset of the compact subset K of Y .
IV.75, `. −8.
“ . . . and µ(h1 ) − µY (h) = µ(h1 ϕU -- K ) ”
From |h1 | 6 f1 one infers Supp(h1 ) ⊂ Supp(f1 ) ⊂ U , whence

h1 ϕU -- K = h1 ϕU − h1 ϕK = h1 − h1 ϕK ,

where h1 ∈ K (X; C) and h1 ϕU -- K , h1 ϕK are µ-integrable (No. 6, Cor. 3

of Th. 5); thus
µ(h1 ϕU -- K ) = µ(h1 ) − µ(h1 ϕK ) ,
R
where the first and third µ ’s are abbreviations for dµ . The assertion is
0
that
 µ(h ϕ
1 K ) = µ
 Y (h) . Recall that by definition µ Y (h) = µ(h ) , where
0 0
h Y = h and h X --- Y = 0 ; thus it will suffice to show that h 1 ϕK = h0 .
Indeed, since f = 0 on Y --- K and |h| 6 f , one has h = 0 on Y --- K ;
therefore
x ∈ Y --- K ⇒ h1 (x)ϕK (x) = h(x) · 0 = 0 · 0 = h(x)
x∈K ⇒ h1 (x)ϕK (x) = h(x) · 1 = h(x)
x ∈ X --- Y ⇒ h1 (x)ϕK (x) = h1 (x) · 0 = 0 ,

thus h1 ϕK is equal to h on Y = (Y --- K) ∪ K , and to 0 on X --- Y , in

other words h1 ϕK = h0 .
IV.75, `. −6.
“ |µ(h1 ) − µY (h)| 6 kf k · |µ|(U --- K) 6 ε kf k ”
Note that |h1 ϕU -- K | 6 |f1 | ϕU -- K 6 kf k1 ϕU -- K = kf k ϕU -- K , whence
|µ(h1 ) − µY (h)| = |µ(h1 ϕU -- K )| 6 |µ|(|h1 ϕU -- K |) 6 kf k · |µ|(U --- K) 6
ε kf k .
IV.75, `. −4.  
“ |µ|(f1 ) − |µ|Y (f ) = |µ|(f1 ϕU -- K ) and |µ|(f1 ) − |µ|Y (f ) 6 ε kf k ”
§5 measurable functions and sets INT IV.x119

We know that f1 ∈ K+ (X) and Supp(f1) ⊂ U , therefore

f1 ϕU -- K = f1 ϕU − f1 ϕK = f1 − f1 ϕK ,

whence |µ|(f1 ϕU -- K ) = |µ|(f1 ) − |µ|(f1 ϕK ) . To prove

 the first assertion, we
need only show that f1 ϕK = f . Indeed, since f1 Y = f and f = 0 on
0

Y --- K , one has

x ∈ Y --- K ⇒ f1 (x)ϕK (x) = f (x) · 0 = 0 · 0 = f (x)

x∈K ⇒ f1 (x)ϕK (x) = f (x) · 1 = f (x)
x ∈ X --- Y ⇒ f1 (x)ϕK (x) = f1 (x) · 0 = 0 ,

thus f1 ϕK is equal to f on Y, and to 0 on X --- Y, in other words f 1 ϕK = f 0 .

The first assertion and f1 ϕU -- K > 0 imply that |µ|(f1 ) − |µ|Y (f ) > 0 ;
finally, from 0 6 f1 ϕU -- K 6 kf1 kϕU -- K = kf kϕU -- K , we infer that
 
|µ|(f1 ) − |µ|Y (f ) = |µ|(f1 ϕU -- K ) 6 kf k · |µ|(U --- K) 6 ε kf k .

IV.75, `. −2.
“ |µ|Y (f ) 6 |µY (h)| + ε(1 + 2kf k) 6 |µY |(f ) + ε(1 + 2kf k) ”
In slow motion: the inequalities
 
|µ|(f1 ) − |µ|Y (f ) 6 εkf k ,
|µ|(f1 ) 6 |µ(h1 )| + ε ,
|µ(h1 )| 6 |µ(h1 ) − µY (h)| + |µY (h)| 6 ε kf k + |µY (h)| ,
|µY (h)| 6 |µY |(|h|) 6 |µY |(f )

yield, respectively,

|µ|Y (f ) 6 |µ|(f1 ) + εkf k 6 |µ|(h1 ) + ε + εkf k

6 (εkf k + |µY (h)|) + ε + εkf k
= |µY (h)| + ε(1 + 2kf k)
6 |µY |(f ) + ε(1 + 2kf k) .

IV.76, `. 18.
“It is immediate (No. 2, Prop. 5) that d) implies a) ”
Assuming d), we are to show that for a set B ⊂ A ,

B is locally negligible ⇔ B ∩ K is negligible for every K ∈ K .

INT IV.x120 extension of a measure. l p spaces §5

Proof of ⇒: Indeed, if B is any locally negligible set in X , then B ∩ K

is negligible for every compact set K in X by the cited Prop. 5.
Proof of ⇐: Assume B has the indicated property; according to the
criterion of the cited Prop. 5, we are to show that B ∩ K is negligible for
every compact set K in X .
Suppose
S first that K is a compact subset of A . By d) we may write
K = N ∪ Hn with N negligible and (Hn ) a sequence of sets in K ; then
n
[
B ∩ K = (B ∩ N) ∪ (B ∩ Hn ) ,
n

where B ∩ N is negligible and the B ∩ Hn are negligible by the assumption

on B , therefore B ∩ K is negligible.
Now let K beSany compact set in X . Since A is measurable, one can
write A ∩ K = N ∪ Kn with N negligible and (Kn ) a sequence of compact
n
sets (see the Lemma in the note for IV.67, `. −1); since B ⊂ A , one has
[
B ∩ K = (B ∩ A) ∩ K = B ∩ (A ∩ K) = (B ∩ N) ∪ (B ∩ Kn ) ,
n

where the B∩Kn are negligible by the preceding paragraph, therefore B∩K
is negligible.
IV.76, `. −19, −18.
“ . . . one defines
S recursively a sequence
S (H p ) of sets of K such that
Hn+1 ⊂ B --- Hp and |µ|(B --- Hp ) 6 1/n (§4, No. 6, Th. 4). ”
p6n p6n
By b), choose H1 ∈ K with H1 ⊂ B and |µ|(B --- H1 ) 6 1 . Since
B --- H1 is integrable,

there exists a compact set B 1 ⊂ B --- H1 such that
|µ| (B --- H1 ) --- B1 6 1/4 by Cor. 1 of the cited Th. 4; if, by b), H 2 ∈ K is
chosen so that H2 ⊂ B1 and |µ|(B1 --- H2 ) 6 1/4 , then since H2 ⊂ B1 ⊂
B --- H1 , so that


(B --- H1 ) --- H2 = (B --- H1 ) --- B1 ∪ (B1 --- H2 ) ,

it follows that

|µ| (B --- H1 ) --- H2 6 1/4 + 1/4 ,


that is, |µ| B --- (H1 ∪ H2 ) 6 1/2 .
Suppose H1 , H2 , . . . , Hn in K already constructed satisfying the rele-
vant inclusions and inequalities, in particular
[  [  1
Hn ⊂ B --- Hp and |µ| B --- Hp 6 .
n
p6n−1 p6n
§5 measurable functions and sets INT IV.x121
S S
Then B --- Hp is integrable; choose a compact set B n ⊂ B --- Hp
p6n p6n
such that  [ 

1
|µ| B --- Hp --- Bn 6 ,
2(n + 1)
p6n

then, by b), a set Hn+1 ∈ K such that Hn+1 ⊂ Bn and

1
|µ|(Bn --- Hn+1 ) 6 .
2(n + 1)
S
Then Hn+1 ⊂ Bn ⊂ B --- Hp , thus
p6n

[
 [


B --- Hp --- Hn+1 = B --- Hp --- Bn ∪ Bn --- Hn+1 ,
p6n p6n

therefore
 [
 1 1
|µ| B --- Hp --- Hn+1 6 + ,
2(n + 1) 2(n + 1)
p6n

S 1

that is, |µ| B --- Hp 6
, which completes the induction.
p6n+1 n + 1
S
∞ S
Set N = B --- Hp ; then N ⊂ B --- Hp for all n , therefore
p=1 p6n
|µ|*(N) 6 1/n for all n , thus N is negligible.
IV.76, `. −17.
“It remains to prove that a) implies b).”
In view of |µ|(K0 --- K) = |µ|(K0 ) − |µ|(K) (§4, No. 5, Prop. 7), the
meaning of b) is that, for every compact subset K 0 of A ,

|µ|(K0 ) = sup |µ|(K) .

K∈K, K⊂K0

The proof by contradiction ultimately rests on the fact that the measure
(for |µ| ) of an integrable set is the supremum of the measures of its compact
subsets (§4, No. 6, Cor. 1 of Th. 4).
IV.76, `. −13, −12.
“ B is integrable and |µ|(B) = α ”
Because (Ln ) is an increasing sequence of integrable sets such that the
sequence |µ|(Ln ) has a finite supremum α (§4, No. 5, Prop. 8).
INT IV.x122 extension of a measure. l p spaces §5

IV.76, `. −11, −10.

“ |µ| K ∩ (K0 --- B) = 0 , which, by virtue of a), will imply a contra-
diction.”
It will imply, by a), that K0 --- B is locally negligible; but |µ|*(K 0 --- B)
is finite, therefore K0 --- B is negligible (No. 2, Cor. 1 of Prop. 5).
IV.77, `. 1.
“The set of all compact subsets of A is µ-dense in A .”
The set K of all compact subsets of A clearly satisfies (PL I ) and (PLII );
let us verify that K satisfies the condition c) of Prop. 12. Since A is
measurable, that is, ϕA is a measurable function on X (No. 1, Def. 2),
given any compact subset K of X there exists a sequenceS(K n ) of pairwise
disjoint compact subsets of K such that the set N = K --- Kn is negligible
n
(No. 1, Def. 1; the continuity properties of ϕ A are of no interest here); if,
in particular, K ⊂ A , that is, K ∈ K , then also K n ∈ K for all n .
In this sense, a measurable subspace of a locally compact space does
have an abundance of compact subsets (cf. the Addendum to the note for
IV.73, `. −2).
IV.77, `. 3, 4.
“If X --- A is locally µ-negligible, then every set of compact subsets
of A that is µ-dense in A is also µ-dense in X .”
Since X --- A is measurable (see the note for IV.61, `. −4), so is A .
Let K be a set of compact subsets of A that is µ-dense in A ; in particular,
the condition a) of Prop. 12 is valid for A . We will show that K is µ-dense
in X by verifying that X satisfies the criterion a) of Prop. 12: given any
subset B of X such that B ∩ K is negligible for every K ∈ K , we are to
show that B is locally negligible. Now,

B = (B --- A) ∪ (B ∩ A) ;

we show that both terms of the union are locally negligible. Since X --- A
is locally negligible, so is its subset B --- A . On the other hand, for every
K ∈ K , the set
(B ∩ A) ∩ K = B ∩ (A ∩ K) = B ∩ K
is by assumption negligible, therefore B ∩ A is locally negligible by the
µ-density of K in A and its property a) in Prop. 12.
IV.77, `. 10–12.
“If K is a compact subset of X , it comes to the same to say that a set
of compact subsets of K is µ-dense in K or that it is µ K -dense in K ; this
follows from Lemmas 2 and 3 of No. 7 and condition b) of Prop. 12.”
§5 measurable functions and sets INT IV.x123

If K0 ⊂ K is compact in the subspace K of X (hence is compact in X )

and if J ∈ K , J ⊂ K0 , then

|µ|(K0 --- J) = |µ|(K0 ) − |µ|(J)

= |µ|K (K0 ) − |µ|K (J) (Lemma 2, (i))
= |µK |(K0 ) − |µK |(J) (Lemma 3)
= |µK |(K0 --- J) ;

in particular, |µ|(K0 --- J) 6 ε ⇔ |µK |(K0 --- J) 6 ε and the assertion is

now clear from the criterion b) of Lemma 12.
IV.77, `. −8 to −6.
“Def. 7 shows that the union of a locally countable set of µ-measurable
(resp. locally µ-negligible) subsets of a locally compact space is µ-measurable
(resp. locally µ-negligible) (No. 1, Prop. 3 and No. 2, Prop. 5).”
Let A be a locally countable set of subsets of the locally compact
space X equipped with a measure µ ; present A as a family (A α )α∈I ,
faithfully indexed byS I , that is, α 7→ A α is an injection I → P(X) with
range A . Let A = Aα .
α∈I
For each x ∈ X , choose an open neighborhood U x of x in X such that
the set
Ix = {α ∈ I : Ux ∩ Aα 6= ∅ }
is countable; in particular, Ux ∩ Aα = ∅ when α ∈ / Ix , thus
[ [ [
Ux ∩ A = U x ∩ Aα = Ux ∩ A α = Ux ∩ A α ,
α∈I α∈I α∈Ix

where the index set Ix is countable.

Assuming the Aα are measurable (resp. locally negligible), we are to
show that A is measurable (resp. locally negligible). Note that when the A α
are locally negligible, they are also measurable (see the note for IV.61, `. −4).
Given any compact subset K of X , it will suffice by No. 1, Prop. 3
(resp. No. 2, Prop. 5) to show that K ∩ A is integrable (resp. negligible).
With the open sets Ux chosen as above, the Ux (x ∈ K) form an open
covering of K , hence
K ⊂ U x1 ∪ · · · ∪ U xn
S
n
for a suitable finite set of points x1 , . . . , xn of K . Then K = K ∩ U xi ,
i=1
therefore n
[
K∩A= K ∩ U xi ∩ A .
i=1
INT IV.x124 extension of a measure. l p spaces §5

For each i the set,

[ [
K ∩ U xi ∩ A = K ∩ Aα = K ∩ Aα
α∈Ixi α∈Ixi

is a countable union of measurable sets, hence is measurable; being contained

in K , it is integrable (No. 6, Cor. 1 of Th. 5). Thus K ∩ A is a finite union
of integrable sets, hence is integrable. This shows that A is measurable.
If, moreover, the Aα are locally negligible, then every K∩A α is negligi-
ble, hence each K ∩ Uxi ∩ A is a countable union of negligible sets, therefore
so is K∩A , thus K∩A is negligible. This shows that A is locally negligible.
IV.77, `. −5 to −1.
“Proposition 14.”
For an application, see Théories spectrales, Ch. II, §3, No. 3, Lemma 3
(p. 146, item 4)) and Prop. 5, (iii).
A nontrivial special case ( A = X and K the set of all compact subsets
of X ; cf. IV.77, `. 1): There exists a Slocally countable set H of pairwise
disjoint compact sets such that X --- K is locally negligible.
K∈H

IV.78, `. 13.
“ |µ|(K ∩ V) > 0 for every K ∈ HV ”
The set K ∩ V is a nonempty open set in the subspace K of X , and
Supp(µK) = K , therefore |µK |(K ∩ V) > 0 by §2, No. 2, Prop. 5; and
|µK |(K ∩ V) = |µ|K (K ∩ V) = |µ|(K ∩ V) by Lemmas 3 and 2 of No. 7.
IV.78, `. 15.
“ N is µ-measurable ”
S
The set K is measurable (see the note for IV.77, `. −8 to −6), and
K∈H
the difference of two measurable sets is measurable.
IV.78, `. 15, 16.
“If N were not locally negligible, it would contain a non-negligible com-
pact set L0 ”
If N is not locally negligible then there exists a compact set K in X
such that K∩N is not negligible (No. 2, Prop. 5). Since K∩N is measurable
and has finite exterior measure ( 6 |µ|(K) ), it follows that K∩N is integrable
(No. 6, Cor. 1 of Th. 5); thus |µ|(K ∩ N) = |µ|*(K ∩ N) > 0 , therefore there
exists a compact set L0 ⊂ K ∩ N such that |µ|(L0 ) > 0 by ‘inner regularity’
of |µ| (§4, No. 6, Cor. 1 of Th. 4).
IV.79, `. 3.
“ d) implies c) ”
§5 measurable functions and sets INT IV.x125

Suppose f : A → F satisfies d); let y ∈ F be a point such that the

function f : X → F defined by
(
f (x) for x ∈ A
f(x) =
y for x ∈ X --- A

is µ-measurable. (By hypothesis, y can be any point of F ; the existence of

a single such point is sufficient to prove that c) is satisfied.)
Let G = F , g = f : X → F = G , and let j be the identity homeomor-
phism F → F = G . For every x ∈ A ,


g(x) = f(x) = f (x) = j f (x) = (j ◦ f )(x) ,

thus g A = j ◦ f as mappings A → G .
IV.79, `. 3, 4.
“The fact that c) implies a) follows from condition c) of Prop. 12 of
No. 8.”
One assumes that there exist a topological space G , ‘a homeomorphism
j of F onto a subspace of G ,’ and a µ-measurable mapping g : X → G
such that g A = j ◦ f .
What the author means by the phrase in quotes is that j : F → G is
a mapping such that, if the range j(F) of j is equipped with the topology
induced by G , and if j0 : F → j(F) is the function having the graph
of j , then j0 is a homeomorphism. It follows that j is injective (obvious)
and that j is continuous; for, if i : f (F) → G is the canonical injection,
then i is continuous ( j(F) has the initial topology for i ) hence so is the
composite function j = i ◦ j0 . Since j0 is a homeomorphism, it is trivial
that F has the initial topology for j 0 ; it then follows that F has the initial
topology for j = i ◦ j0 (‘transitivity of initial topologies’, GT, I, §2, No. 3,
Prop. 5). Indeed, for a mapping j : F → G to have the property that j 0
is a homeomorphism, it is necessary and sufficient that j be injective and
that F have the initial topology for j ; for, if the condition in italics holds
then j0 is bijective and, by the cited Prop. 5, F has the initial topology
for
 j0 , consequently j0 is a homeomorphism. The further assumption that
g A = j ◦ f is that the diagram

f-
A F

iA j
? ?
X g-G
INT IV.x126 extension of a measure. l p spaces §5

is commutative, where iA : A → X is the canonical injection.

Consider now any subset K of A , and let u K : K → A and iK : K → X
be the canonical injections, so that i K = iA ◦uK as mappings K → X . Then
the diagram

uK- f-
K A F
Q
Q
Q iA j
iK Q
Q
sQ ? ?
X g-G

is commutative; in particular,

(j ◦ f ) ◦ uK = (g ◦ iA ) ◦ uK = g ◦ (iA ◦ uK ) = g ◦ iK .

Note that if K is given the topology induced by X (equivalently by A ),

then  
f K is continuous ⇔ g K is continuous;
for, since F has the initial topology for j , the following conditions are
equivalent:

f K = f ◦ uK is a continuous mapping K → F
j ◦ (f ◦ uK ) is a continuous mapping K → G

and since j ◦ (f ◦ uK ) = (j ◦ f ) ◦ uK = g ◦ iK = g K , the assertion is proved.
{One observes that the injectivity of j is not needed for this argument; it
is sufficient that F have the initial topology for j .}

Now let H be the set of all compact subsets K of A such that f K
is continuous (equivalently, g K is continuous); we are to show that H is
µ-dense in A .
It is clear the H is closed under finite unions (GT, I, §3, No. 2, Prop. 4),
thus satisfies (PLII ) of Prop. 12, and (PLI ) is obviously satisfied; to complete
the proof, it will suffice to show that H satisfies the condition c) of Prop. 12
(No. 8, Def. 6) stated for H instead of K .
To that end, let B be a compact subset of A . Since g : X → G is
µ-measurable, one can write
[
B =N∪ Kn ,
n

where (KnS) is a sequence of pairwise

 disjoint compact sets such that
N = B --- Kn is µ-negligible and g Kn is continuous for all n (No. 1,
n
§5 measurable functions and sets INT IV.x127

Def. 1); therefore f Kn is continuous for all n , thus Kn ∈ H for all n ,
which completes the proof that H is µ-dense in A , and hence that the
condition a) of Prop. 15 is implied by its condition c).
{One observes that the condition a) of Prop. 15 is implied by the fol-
lowing condition weaker than c):
c0 ) There exist a topological space G , a mapping j : F → G such that F
 topology for j , and a µ-measurable mapping g of X into G ,
has the initial
such that g A = j ◦ f .
In particular, condition a) of Prop. 15 is implied by the following special
case of c), namely, that f may be extended to a µ-measurable mapping
X → F:

c00 ) There exists a µ-measurable mapping g : X → F such that g A = f .
For, if c00 ) is satisfied, then c0 ) is satisfied with G = F and j : F → F
the identity mapping.
Thus, after the proof of Prop. 15 is completed, one can add c 0 ) and c00 )
to the list of equivalent conditions. See also the last paragraph of the note
for IV.79, `. 7 for a condition d 0 ) weaker than d) that can be added to the
list.}
IV.79, `. 5.
“ b) implies a) ”
By assumption, K is a set of compact subsets  of A that is µ-dense
in A , such that for every K ∈ K , the function f K is µK -measurable.
Let H be the set of all compact subsets H of A such that f H is
continuous; we are to show that H is µ-dense in A . We know that H
satisfies (PLI ) and (PLII ) (cf. the preceding note); by Prop. 13, it will suffice
to show that for every K ∈ K , the set HK = {H ∈ H : H ⊂ K } is µK -dense
in K . (Incidentally, one sees easily that H K = {H ∩ K : H ∈ H } .)
Let K ∈ K . Since H satisfies the conditions (PL I ) and (PLII ), it is
immediate that HK also satisfies them. Consider Prop. 12 with K playing
the role of both X and A , and µK playing the role of µ ; in this context,
let us show that HK satisfies the property c) of Prop. 12 (hence is µ K -dense 
in K ). To this end, let B be any compact subset of K . Since K ∈ K , f K
is by assumption µK -measurable, therefore (No. 1, Def. 1) one can write
[
B =N∪ Hn ,
n

where
 (Hn ) is a sequence of pairwise disjoint compact sets in S
B such that
f Hn = (f K)Hn is continuous for every n , and N = B --- Hn is µK -
n
negligible; in particular, Hn ∈ HK , thus the desired property c) is verified.
INT IV.x128 extension of a measure. l p spaces §5

IV.79, `. 7.
“ a) implies d). ” 
Assuming that the set H = {K ⊂ A : K compact, f K continuous }
is µ-dense in A , we are to show that, given any y ∈ F , the extension of f
to X by y , that is, the function g : X → F defined by
(
f (x) for x ∈ A
g(x) =
y for x ∈ X --- A ,

is µ-measurable. By No. 1, Prop. 1 it will suffice to show that, given any

compact set L ⊂ X and any ε > 0 , there exists a compact set K ⊂ L such
that |µ|(L --- K) 6 ε and g K is continuous.
The sets L∩A and L∩(X --- A) are µ-integrable, since the intersection
of an integrable set and a measurable set is integrable (No. 6, Cor. 3 of Th. 5).
Therefore there exist compact sets P ⊂ L ∩ A and Q ⊂ L ∩ (X --- A) such
that



|µ| (L ∩ A) --- P 6 ε/4 and |µ| L ∩ (X --- A) --- Q 6 ε/4

(§4, No. 6, Cor. 1 of Th. 4). Since P is a compact subset of A , and H is

µ-dense in A , by b) of Prop. 12 there exists a set H ∈ H such that H ⊂ P
and
|µ|(P --- H) 6 ε/2 .
   
Set K = H∪Q , which is a compact subset    
 of L ; since g H = (g A) H = f H
is continuous (because 
H ∈ H ) and g Q is continuous (even constant), it

follows that g K is continuous (GT, I, §3, No. 2, Prop. 4). Finally, one has


L --- K = (L ∩ A) ∪ L ∩ (X --- A) --- (H ∪ Q)


⊂ (L ∩ A --- H) ∪ L ∩ (X --- A) --- Q


= (L ∩ A --- P) ∪ (P --- H) ∪ L ∩ (X --- A) --- Q ,

whence |µ|(L --- K) 6 ε/4 + ε/2 + ε/4 by the subadditivity of |µ| (§4, No. 5,
Prop. 6).
An outline of the proof:

a)  - b)
Q
k
Q
Q
Q
Q
? Q
d) - c)
§5 measurable functions and sets INT IV.x129

(where the arrows signify implication). The argument for ‘ d) ⇒ c) ’ shows

that if there exists at least one element y ∈ F such that the extension g of f
to X by y is measurable, then c) holds, whence a) and therefore d) hold;
thus, the existence of an element y ∈ F for which g is measurable implies
that for every y 0 ∈ F the extension g 0 of f to X by y 0 is measurable.
{This is obvious when F is a topological vector space, since g 0 − g is then
a measurable step function (No. 5) and g 0 = (g 0 − g) + g (No. 3, Cor. 3
of Th. 1), but it is not obvious when F is merely a topological space.} Thus,
with the proof of Prop. 15 completed, one can add to its list of equivalent
conditions the following:
d 0 ) There exists an element y0 ∈ F such that the extension of f to X
by y0 is µ-measurable.
IV.79, `. −17, −16.
“If A is locally µ-negligible, then every mapping of A into F is there-
fore µ-measurable.”
Let f : A → F , where A is locally µ-negligible. Then A is
µ-measurable (see the note for IV.61, `. −4), and it will suffice to show
that the condition a) of Prop. 15is satisfied. Let H be the set of all com-
pact subsets K of A such that f K is continuous (for example, every finite
subset of A belongs to H ). One knows that H satisfies (PL I ) and (PLII )
of Prop. 12; to show that H is µ-dense in A , it will suffice to verify that H
satisfies the condition c) of Prop. 12. Let B be a compact subset of A ; then
B = A ∩ B is negligible (No. 2, Prop. 5), so setting N = B and K 1 = ∅ one
has a partition B = N ∪ K1 with K1 ∈ H trivially. (The same argument
shows that if A ⊂ X is locally µ-negligible then the set H 0 = {∅} is µ-dense
in A , so a fortiori H is µ-dense in A .)
IV.79, `. −4.
“ . . . the conclusion therefore follows from Prop. 12 of No. 8.”

Recall that H = {K ∈ K : K ⊂ A and f K is continuous } ; the con-
clusion in question is that H is µ-dense in A . Clearly H satisfies (PL I ) and
(PLII ) of Prop. 12; it will suffice to show that it satisfies the condition b) of
that proposition. {The proof requires only that K be µ-dense in A .}
Thus, given any compact set K0 ⊂ A and any ε > 0 , we seek a set
K ∈ H such that K ⊂ K0 and |µ|(K0 --- K) 6 ε . Since K is µ-dense
in A , by b) of Prop. 12 there exists a set K 1 ∈ K such that K1 ⊂ K0 and
|µ|(K0 --- K1 ) 6 ε/2 .
On the other hand since f : A → F is measurable, by Def. 8 and d)
of Prop. 15, the function g : X → F obtained by extending f to X by
an (any) element of F is measurable. Then (No. 1, Prop. 1) there exists a
compact set K ⊂ K1 such that |µ|(K1 --- K) 6 ε/2 and g K is continuous;
INT IV.x130 extension of a measure. l p spaces §5
   
since K ⊂ K1 ∈ K one has K ∈ K , and since f K = (g A)K = g K is
continuous, one has K ∈ H . Finally, since K ⊂ K 1 ⊂ K0 one infers from
K0 --- K = (K0 --- K1 ) ∪ (K1 --- K) that |µ|(K0 --- K) 6 ε/2 + ε/2 .
IV.80, `. 1, 2.
“In view of Lemma 2 of No. 7, this follows at once from Prop. 1 of No. 1
and condition a) of Prop. 15.”

Let H be the set of compact subsets H of K such that f H is contin-
uous. One knows that

H is µ-dense in K ⇔ H is µK -dense in K

(see the note for IV.77, `. 10–12). We are to show that

f is µ-measurable ⇔ f is µK -measurable

(the statement on the left, in the sense of Def. 8; the statement on the right,
in the sense of No. 1, Def. 1 applied in the compact space K ).
Proof of ⇒ : Suppose f is µ-measurable. By criterion a) of Prop. 15,
we know that H is µ-dense in K , hence is µ K -dense in K ; from the charac-
terization of density in b) of Prop. 12 (applied to the measurable subset K
of the compact space K ), it follows that f is µ K -measurable (No. 1, Prop. 1,
applied in the space K ).
Proof of ⇐ : Suppose f is µK -measurable. By No. 1, Prop. 1 and b) of
Prop. 12 (both applied in the space K ) we see that H is µ K -dense in K ,
hence µ-dense in K , therefore f is µ-measurable by a) of Prop. 15.
IV.80, `. 7, 8.
“The condition being obviously necessary . . . ”
Suppose f : B → F is µ-measurable
 (in the sense of Def. 8). Fix
A ∈ A ; we are to show that f A is µ-measurable. Applying to B and f
the criterion c) of Prop. 15, there exist a topological space G , an injective
mapping j : F → G such that F has the initial  topology for j , and a
µ-measurable mapping g : X → G such that g B = j ◦ f (see the note for
IV.79, `. 3,4); thus the diagram

f-
B F

iB j
? ?
X g-G
§5 measurable functions and sets INT IV.x131

is commutative, where iB : B → X is the canonical injection. It follows that

the diagram

uA- f-
A B F
Q
Q
Q iB j
iA Q
Q
sQ ? ?
X g-G

is also commutative, where uA and iA are the canonical injections A → B

and A → X . In particular,
 
g A = g ◦ iA = j ◦ f ◦ uA = j ◦ (f A) ,

therefore 
 f A is µ-measurable by criterion c) of Prop. 15 applied to A
and f A in place of B and f .
IV.80, `. 9,10.
“By hypothesis, there exists a sequence (A n ) of sets belonging to A
such that the K ∩ An form a covering of K .”
Since K ⊂ B one has
[
K = K∩B = K ∩ A,
A∈A

and K ∩ A = ∅ for all but countably many A in A (No. 9, sentence after

Def. 7).
IV.80, `. 14.
“ . . . the restriction of f to Cn is µ-measurable ”

Since f An is µ-measurable by hypothesis, and C n ⊂ An is measur-
  

able, f Cn = (f An )Cn is µ-measurable by the argument in the proof of
“necessity” (see the note for `. 7,8).
IV.80, `. 14–16.
“ . . . there exists a partition of Cn formed by a µ-negligible
 set Nn and

a sequence (Lmn )m>0 of compact sets such that f Lmn is continuous.”
Recall that K is a compact subset of the µ-measurable set B , and that
Cn is a µ-integrable (hence µ-measurable) subset of K .   
Let Kn be the set of compact sets H ⊂ Cn such that (f Cn )H = f H
is continuous; since f Cn is µ-measurable, Kn is µ-dense in Cn by a) of
Prop. 15. Then by No. 9, Prop. 14, there exists a locally countable set
Hn ⊂ Kn , consisting of pairwise disjoint (compact) subsets of C n , such
INT IV.x132 extension of a measure. l p spaces §5
S
that the set Nn = Cn --- H is locally negligible. Since Nn ⊂ K ,
H∈Hn
Nn = Nn ∩ K is negligible (No. 2, Prop. 5). Since H n is locally countable
(because Kn is), K ∩ H = ∅ for all but countably many H in H n ; that is,
since H ⊂ Cn ⊂ K for all H ∈ Hn , H = K ∩ H = ∅ for all but countably
many H in Hn , in other words Hn is countable.
IV.80, `. 16–18.
S
“Since N = Nn is µ-negligible, we see that condition a) of Prop. 15
n
is satisfied, whence the proposition.”
Note that
[ [ [  [[ 
K= Cn = Nn ∪ Lmn = N ∪ Lmn
n n m n m

S
where all unions are ‘disjoint’, therefore K --- N = Lmn .
m,n
Thus it has been shown that, given any compact subset K of B , there
exists a partition of K formed of a µ-negligible
 set N and a countable family
(Lmn ) of compact sets such that the f Lmn are continuous.
 This shows that

the set H of all compact sets H ⊂ B such that f H is continuous satisfies
condition c) of Prop. 12 (as well, of course, the conditions (PL I ) and (PLII )),
therefore H is µ-dense in B (No. 8, Def. 6); thus condition a) of Prop. 15
is satisfied, consequently f : B → F is indeed µ-measurable (Def. 8).
IV.80, `. −15, −14.
“ . . . these generalizations are left to the reader.”
The proofs are sketched in the note after the next.
IV.80, `. −14 to −11.
“ . . . the principle of localization (No. 2, Prop. 4) remains valid when it
is assumed that each of the functions g x is only defined in Vx (or almost
everywhere in Vx ) and is measurable.”
The key to interpreting this statement is the phrase “and is measurable”.
Two interpretations of ‘measurable function’ are available: Def. 1 or Def. 8;
it is clearly Def. 8 that is intended here.
Let us call Prop. 40 the proposed generalization of Prop. 4. One is given:
(i) a topological space F and a mapping f : X → F ;
(ii) for each x ∈ X , a neighborhood Vx of x in X , a negligible sub-
set Nx of Vx , and a function gx , measurable in the sense of Def. 8, whose
(measurable) domain of definition B x contains
 the (integrable, hence mea-
surable) set Ax = Vx --- Nx , such that gx Ax = f Ax .
§5 measurable functions and sets INT IV.x133

We are to prove that f is measurable. By d) of Prop. 15, the function

gx : Bx → F may be extended to a function hx : X → F that is measurable
in the sense of Def. 1. Since
    
hx Ax = (hx Bx )Ax = gx Ax = f Ax ,
 
that is, hx (Vx --- Nx ) = f (Vx --- Nx ) , one has hx (y) = f (y) almost ev-
erywhere in Vx , therefore f is measurable by No. 2, Prop. 4.
IV.80, `. −17 to −14.
“Property d) of Prop. 15 makes it possible to immediately generalize the
properties of measurable functions defined on all of X , observed in Nos. 2
to 5, to measurable functions defined on a measurable subset A of X ; these
generalizations are left to the reader.”
We write, for example, Th. 10 for the contemplated generalization of Th. 1,
skipping over results not susceptible to generalization via property d) of
Prop. 15 (for instance, No. 2, Prop. 5, to which Prop. 15 can add nothing).
No. 2, Cor. 4 0 of Prop. 5. Not amenable to generalization. If A is
a measurable subset of X (equipped with the induced topology), and if
f : A → F is a function such that the set

N = {x ∈ A : f is not continuous at x }

is locally negligible, one would like to show that the extension f 0 of f

to X by an element of F is measurable, presumably by applying Cor. 4
(without the prime) to f 0 . But the extension to f 0 might introduce a set
of discontinuities not locally negligible, rendering Cor. 4 inapplicable.
For example, let X = R equipped with Lebesgue measure, F = R ,
A = Q equipped with the induced topology, and f :Q → R the constant
function equal to 1 . Then N = ∅ , the extension f 0 of f to R by 0
(the characteristic function of Q in R ) is a measurable function, but it is
nowhere continuous, hence owes nothing to Cor. 4 for its measurability.
No. 2, Prop. 6 0 . Let A ⊂ X be measurable, f : A → F measurable,
and g : A → F a function such that g = f locally almost everywhere in A ;
we are to show that g is measurable.
Let f 0 , g 0 be the extensions of f and g to X by some point y 0 in F .
Then f 0 is measurable by d) of Prop. 15, and g 0 = f 0 locally almost every-
where in X , therefore g 0 is measurable by the original Prop. 6, and finally
g is measurable by d) of Prop. 15.
No. 3, Th. 1 0 . With X , µ , (Fn ) and F as in Th. 1, let A be a
measurable subset of X , fn : A → Fn a sequence of mappings µ-measurable
INT IV.x134 extension of a measure. l p spaces §5

in the sense of Def. 8, f = (fn ) : A → F , and u : f (A) → G a continuous

mapping into a topological space G . We are to show that the composite
mapping u ◦ f : A → G is µ-measurable.
Fix a point

a ∈ A , define yn = fn (a) ∈ Fn for all n , and let y =
(yn ) = fn (a) = f (a) ∈ F . Write f 0 for the extension of f to X by y ;
clearly f 0 = (fn0 ) , where, for each n , fn0 is the extension of fn to X by yn .
Note that f 0 (X) = f (A) ; for, f 0 A = f , whereas if x ∈ X --- A
then f 0 (x) = y = f (a) ∈ f (A) . It follows that  the composite
 function
u ◦ f 0 : X → G is defined. Moreover, (u ◦ f 0 )A
= u ◦ (f 0 A) = u ◦ f ,
whereas if x ∈ X --- A then (u ◦ f 0 )(x) = u f
0 (x) = u(y) ; thus u ◦ f 0 is
the extension of u ◦ f to X by u(y) = u f (a) = (u ◦ f )(a) .
So much for the notation. For each n , f n is µ-measurable, therefore
fn0 is µ-measurable by Def. 8 and d) of Prop. 15. Since f 0 = (fn0 ) and u is
continuous on f (A) = f 0 (X) , it follows from the original Th. 1 that u ◦ f 0 is
µ-measurable, that is, the extension of u ◦ f to X by u(y) is µ-measurable;
thus u ◦ f satisfies the condition d0 ) at the end of the note for IV.79, `. 7
hence is µ-measurable by Def. 8.
In the present situation, it was convenient to extend the f n to X
‘uniformly’ by the particular elements f n (a) of fn (A) rather than by general
elements of the Fn ; this led to an extension of u ◦ f to X by a particular
element of G , whence the utility of adding the weaker condition d 0 ) to the
list in Prop. 15.
To avoid tedious repetitions, henceforth a function f : A → F will be
said to be µ-measurable in the sense of Def. 8 if it satisfies the conditions of
Prop. 15 to which have been added the equivalent conditions d 0 ), c 0 ), c 00 )
described at the end of the notes for IV.79, `. 7 and IV.79, `. 3,4. When there
is only a single measure µ in the picture, ‘ µ-measurable’ may be abbreviated
to ‘measurable’.
No. 3, Cor. 1 0 of Th. 1. Suppose A is a measurable subset of X , and
f, g : A → R are functions µ-measurable in the sense of Def. 8; we are to
show that sup(f, g) and inf(f, g) are also such functions.
With 0 signifying ‘extension to X by 0 ’, we know from Def. 8 that
f 0 , g 0 are measurable functions X → R ; since

0
0
sup(f, g) = sup(f 0 , g 0 ) and inf(f, g) = inf(f 0 , g 0 )

are measurable by the original Cor. 1, it follows from Def. 8 that sup(f, g)
and inf(f, g) are measurable functions A → R .
No. 3, Cor. 2 0 of Th. 1. Given a measurable subset A of X and a
function f : A → R , we are to show that f is measurable if and only
if f + and f − are measurable. Writing 0 for the function on A (or on X )
§5 measurable functions and sets INT IV.x135

identically equal to 0 ∈ R , and f 0 for the extension of f to X by 0 , one

has

0
(f + )0 = sup(f, 0) = sup(f 0 , 00 ) = sup(f 0 , 0) = (f 0 )+ ,

so it is clear from Def. 8 that f + is measurable if and only if (f 0 )+ is

measurable; similarly, f − is measurable if and only if (f 0 )− is measurable.
Therefore

f is measurable ⇔ f 0 is measurable (Def. 8)

0 + 0 −
⇔ (f ) and (f ) are measurable (Cor. 2)
+ −
⇔ f and f are measurable.

No. 3, Cor. 3 0 of Th. 1. Assuming A a measurable subset of X , and

f , g measurable mappings A → F , we are to show that f + g and αf ( α a
scalar) are also measurable. If f 0 , g0 are the extensions of f , g to X by 0 ,
then
(f + g)0 = f 0 + g0 and (αf )0 = αf 0 ,

therefore f + g and αf are measurable by Def. 8 and the original Cor. 3.

No. 3, Cor. 4 0 of Th. 1. With notations as in Cor. 4, assume that
the fk and f are defined on a measurable subset A of X . Then f 0 =
P
n
ek fk0 , where f 0 and fk0 are the extensions to X by 0 ∈ Kn and 0 ∈ K ,
k=1
respectively ( K = R or C ), whence

f is measurable ⇔ f 0 is measurable (Def. 8)

⇔ f10 , . . . , fn0 are measurable (original Cor. 4)
⇔ f1 , . . . , fn are measurable (Def. 8) .

No. 3, Cor. 5 0 of Th. 1. With F, G, H and (u, v) 7→ [u · v] as in

the original Cor. 5, suppose f : A → F and g : A → G are measurable
mappings; we are to show that the mapping [f · g] : A → H defined by
x 7→ [f (x), g(x)] (x ∈ A) is measurable.
To simplify (at least visually) the notation, let us write h(u, v) = [u · v]
(u ∈ F, v ∈ G) ; by hypothesis, h : F × G → H is a continuous mapping, but
it need not be bilinear (see the note for IV.64, `. 10–13). We are to show

that the mapping k = h ◦ (f , g) : A → H , defined by k(x) = h f (x), g(x)
(x ∈ A) is measurable.
Let f 0 and g0 be the extensions of f and g to X by 0 ∈ F and
0 ∈ G , respectively, and let k 0 be the extension of k to X by the element
INT IV.x136 extension of a measure. l p spaces §5

h(0, 0) of H . (Of course if h is bilinear, then h(0, 0) = 0 ∈ H .) Then

k 0 = h ◦ (f 0 , g0 ) , that is,


k 0 (x) = h f 0 (x), g0 (x) for all x ∈ X .

Since f 0 and g0 are measurable (Def. 8), k 0 is measurable by the original

Cor. 5, therefore k is measurable (Def. 8).
No. 3, Cor. 6 0 of Th. 1. If f : A → F and f 0 is the extension of f
to X by 0 , then |f |0 = |f 0 | , etc.
No. 4, Th. 2 0 (Egoroff). With X , µ , A , F as in the original Th. 2,
let M be a measurable subset of X , and (f α )α∈A a family measurable
mappings of M into a metrizable space F . (Why “ M ”? Our favorite
letter A has been pre-empted by the index set, and the letters B, C figure
in the proof of Th. 2.) Assume that there exists a locally negligible set
N ⊂ M such that limF fα (x) exists in F for every x ∈ M --- N , and let
f : M → F be any function such that

lim F fα (x) = f (x) for all x ∈ M --- N .

We are to show that:

1◦ f is measurable; and
2◦ for every compact subset K of M and every ε > 0 , there exists  a
compact set K1 ⊂ K such that |µ|(K --- K1 ) 6 ε and such that the fα K1
are continuous and converge uniformly to f K1 .
Fix a point y0 ∈ F and let fα0 , f 0 be the extensions of fα , f to X
by y0 . The fα0 are measurable by Def. 8, and fα0 → f 0 (x) for all
x ∈ X --- N ; for, if x ∈ M --- N then fα0 (x) = fα (x) → f (x) = f 0 (x) ,
and if x ∈ X --- M then fα0 (x) = y0 → y0 = f 0 (x) .
Thus the fα0 and f 0 satisfy the hypotheses of the original Th. 2, hence
satisfy the corresponding conditions 1 ◦ and 2◦ : f 0 is measurable—hence so
is f , by Def. 8—and fα0 , f 0 satisfy 2◦ for every compact set K ⊂ X , in
particular for every compact set K ⊂ M . But this is not the end of the
story: given a compact set K ⊂ M and an ε > 0 , we must find a compact
set K1 ⊂ K that avoids the points of N , i.e., is contained in K --- N , so
that fα0 K1 = fα K1 and f 0 K1 = f K1 .
Let K be a compact set in M and let ε > 0 . Since N is measurable
(IV.61, `. −4), the set K --- N is integrable, hence

there exists a compact
set K0 ⊂ K --- N such that |µ| (K --- N) --- K0 6 ε/2 (§4, No. 6, Cor. 1 of
Th. 4). Note that


K --- K0 ⊂ (K --- N) --- K0 ∪ (K ∩ N) ,
§5 measurable functions and sets INT IV.x137

where K ∩ N is negligible (No. 2, Prop. 5), therefore |µ|(K --- K 0 ) 6 ε/2 .

Now choose (by 2◦ ) a compact set K1 ⊂ K0 such that |µ|(K0 --- K1 ) 6 ε/2
and such that the fα0 K1 are continuous and converge uniformly
 to f 0 K1 .
Finally,
 from 0
 K1 ⊂ K ⊂ K --- N we
0
infer that the fα K1 = fα K1 and
0  0
f K1 = f K1 , and from K1 ⊂ K ⊂ K we infer that |µ|(K --- K1 ) 6
ε/2 + ε/2 .

No. 4, Cor. 1 0 of Th. 2. Let A ⊂ X be measurable, and let (f n ) be

a sequence of measurable functions A → R . Writing f n0 for the extension

0
of fn to X by 0 , one has sup fn = sup fn0 , etc.
n n

No. 5, Th. 3 0 . With X , µ and F as in Th. 3 (in particular, F is metriz-

able), let A be a measurable subset of X. A function g : A → F will be called
a measurable step function if g has only finitely many values, each assumed
on a measurable subset of X . It follows that g is measurable in the sense
of Def. 8; for, X --- A is a measurable set, so if f 0 : X → F is the extension
of f to X by some element y0 ∈ F , then f 0 is a measurable step function
in the sense of the first paragraph of No. 5, hence is measurable (as noted
−1
there). {There is no harm if y0 is already a value of g ; in this case g 0 (y0 ) is
−1 −1 −1
the disjoint union g (y0 ) ∪ (X --- A) , so that g (y0 ) = g 0 (y0 ) --- (X --- A) =
−1 −1
0 (y
g 0 )∩A will be measurable if and only if g 0 (y0 ) is measurable.} Note that
if g : A → F is a measurable
 step function, then for every measurable set

B ⊂ A , the restriction g B is a measurable step function on B because, for
 −1
y ∈ F , (g B)−1 (y) = B ∩ g (y) is the intersection of measurable sets.
The proposed Th. 3 0 can then be stated as follows.
If f : A → F is a function such that, for every compact set K ⊂ A ,
there exists a sequence gn : A → F of measurable step functions on A such
that gn (x) → f (x) for almost every x in K , then f is measurable.
It may be helpful to outline the long-winded proof. We will show:
1) For every compact set K ⊂ A , the restriction f K is measurable
(proof based on Th. 20 above).
2) For every
 compact set K ⊂ A , the set H K of all compact sets H ⊂ K
such that f H is continuous is µ-dense in A (proved by verifying criterion c)
of Prop. 12 in the context of the measurable subset K of X  ).
3) The set H of all compact sets H ⊂ A such that f H is continuous
is µ-dense in A (proof based on Prop. 13, with K the set of all compact
subsets of A ).
4) f is measurable (proof by verifying that H satisfies criterion a) of
Prop. 15).
INT IV.x138 extension of a measure. l p spaces §5

The details are as follows.

1) Let K ⊂ A be compact. By assumption there exist a sequence
gn : A → F of measurable step functions and a negligible set N ⊂ K such
that gn (x) → f (x) for all x ∈ K --- N . Then (g n
K) is a sequence of mea-
 
surable step functions on K such that g n K → f K pointwise on K --- N ,
therefore f K is measurable by Th. 20 (the generalization of Egoroff’s the-
orem proved above; metrizability of F is needed here).
2) Let K ⊂ A be  compact, and let HK be the set of all compact sets
H ⊂ K such that f H is continuous. We know that HK satisfies (PLI )
and (PLII ) of Prop. 12 (in the context of the measurable subset K of X );
it also satisfies criterion
 c) of Prop. 12, for, if B is a compact subset of K
(hence of A ), then f B is measurable by the preceding paragraph, therefore
 
f B has a measurable extension (f B)0 to X . By No. 1, Def. 1, there
exists
 a sequence (Hn ) of pairwise disjoint compact subsets S of B such that
(f B)0 Hn is continuous for all n , and such that B --- Hn is negligible.
 0    n
But (f B) Hn = (f B)Hn = f Hn , therefore Hn ∈ HK . Thus HK is
µ-dense in K (No. 8, Def. 6).
3) Let K be the set of all compact sets K ⊂ A , and let H be the set of
all compact sets H ⊂ A such that f H is continuous; thus, in the notation
of 2), HK = {H ∈ H : H ⊂ K } for every K ∈ K . We know that H satisfies
(PLI ) and (PLII ) of Prop. 12. By the preceding paragraph, for every K ∈ K
the set HK is µ-dense in K ; since K is µ-dense in A (IV.77, `. 1), it follows
from Prop. 13 that H is µ-dense in A .
4) By the preceding paragraph, H satisfies condition a) of Prop. 15,
therefore f is measurable by Def. 8.
A simplification of the above argument yields an extension of the Prin-
ciple of localization (No. 2, Prop. 4):
Proposition. Let X be a locally compact space, µ a measure on X , A a
measurable subset of X , and f : A → F a mapping of A into a topological
space F (not necessarily metrizable). Let K be the set of all compact subsets
of A . Then

f is measurable ⇔ f K is measurable for all K ∈ K .
Proof. ⇒ : Fix y0 ∈ F and let f 0 be the extension
 of f to X by y0 .
0
Then f 0 is measurable by Prop. 15, therefore
 f B is measurable for every
0
measurable set B in X (because f B has a measurable
  extension
 to
 X,
namely f 0 ). In particular, if K ∈ K , then f K = (f 0 A)K = f 0 K is
measurable. 
⇐ : Let K ∈ K and let H K be the set of all H ∈ K such that f H is
continuous. By assumption, f K is measurable, so by a part of the argument
§5 measurable functions and sets INT IV.x139

in 2) above, HK is µ-dense
 in K . By the argument in 3) above, the

set H = {H ∈ K : f H is continuous } is µ-dense in A , therefore f is
µ-measurable by the argument in 4) above.
Remark. A simpler result in this vein: If A , B are measurable sets
in X such that B ⊂ A , and if f : A → F is measurable ( F any topological
space), then f B is also measurable. For, f has a measurable extension
to X , therefore so does f B .

No. 5, Cor. 1 0 of Th. 3. As in the original Cor. 1, Let X be a locally

compact space equipped with a measure, and let F be a Banach space.
Let A be a measurable subset of X and let f : A → F be a mapping
measurable in the sense of Def. 8; we are to show that, given any compact
set K ⊂ A , there exists a sequence gn : A → F of measurable step functions
(as defined in the note for Th. 30 ) such that (i) Supp(gn ) ⊂ K , (ii) |gn | 6 |f |
on A , and (iii) gn (x) → f (x) for almost every x in K .
Let K be a compact subset of A. By Def. 8, f may be extended to a
measurable mapping h : A → F, so by the original Cor. 1 there exists a se-
quence hn : X → F of measurable step functions such that Supp(h n) ⊂ K,
|hn | 6 |h| on X , and hn (x) → h(x) for almost every x ∈ K . The restric-
tions gn = hn A meet all the requirements.

No. 5, Cor. 2 0 of Th. 3. As in the original Cor. 2, let X be a locally

compact space countable at infinity, equipped with a measure, and let F be
a metrizable space. Let A be a measurable subset of X , and f : A → F
a mapping measurable in the sense of Def. 8; we are to show that there exists
a sequence gn : A → F of measurable step functions (as defined in the note
for Th. 30 ) such that gn (x) → f (x) for almost every x ∈ A .
By Def. 8, f may be extended to a measurable mapping h : X → F , so
by the original Cor. 2 there exists a sequence of measurable step functions
hn : X → F such that  hn (x) → h(x) for almost every x ∈ X. Then
the restrictions gn = hn A are measurable step functions on A such that
gn (x) → f (x) for almost every x ∈ A .

No. 5, Prop. 7 0 . Let X be a locally compact space equipped with a

measure (say µ ), and F any topological space. By the original Prop. 7,
−1
if h : X → F is measurable and U is an open set in F , then h (U) is
a measurable set. Since the measurable subsets of X form a tribe (No. 4,
−1
Cor. 2 of Th. 2), it follows that h (B) is measurable for every Borel set B
−1
in F (the set S of sets S ⊂ F such that h (S) is measurable in X is a
tribe that contains every open set, hence S contains every Borel set). Now,
INT IV.x140 extension of a measure. l p spaces §5

if B ⊂ F then
−1 −1
A ∩ h (B) = {x ∈ A : h(x) ∈ B } = {x ∈ A : f (x) ∈ B } = f (B) ,
−1
thus if B is a Borel set in F then f (B) is the intersection of two measurable
sets, hence is measurable.
−1
In particular, if B is any open set or closed set in F , then f (B) is a
measurable set of X that is contained in A , that is, a µ-measurable subset
of A .
Remark. The term ‘measurable subset of A ’ was avoided, as it begs the
question of whether A is a locally compact subspace of X equipped with a
measure of its own.
Suppose, indeed, that A is a locally compact subspace of X , hence is
a measurable
 subset of X (No. 7, first paragraph). One has the R measure
µA = µA induced on A (No. 7, Def. 4), defined by µ A (g) = g 0 dµ for
g ∈ K (A; C) , where g 0 is the extension of g to X by 0 . Borrowing from
the future (Ch. V, §7, No. 1, Cor. of Prop. 2), we know that a set C ⊂ A
is µA -measurable if and only if it is µ-measurable. And (loc. cit., Prop. 2),
a mapping f : A → F is µA -measurable if and only if it is µ-measurable in
the sense of Def. 8, hence has a µ-measurable extension h : X → F ; in this
−1 −1
case, if C is a Borel set in F , then f (C) = A ∩ h (C) is a µ-measurable
set contained in A by the earlier discussion, hence (by the foregoing) is a
µA -measurable subset of A .
The bottom line: when A is a locally compact subspace of X , the term
‘measurable subset of A ’ is unambiguous.
No. 5, Th. 4 0 . As in the original Th. 4, let X be a locally compact
space equipped with a measure µ , F a metrizable topological space, and d
a metric on F compatible with its topology.
The original Th. 4 asserts that a mapping h : X → F is measurable if
and only if it satisfies the following two conditions (slightly reworded):
−1
a) For every closed ball B in F , h (B) is a measurable set in X ;
b) for every compact set K ⊂ X , there exist a countable set H ⊂ F
and a negligible set N ⊂ K such that h(K --- N) ⊂ H .
An equivalent formulation of b) is that for every compact set K ⊂
X , there exists a negligible set N ⊂ K such that the topological subspace
h(K --- N) of F has a countable dense subset, equivalently (GT, IX, §2,
No. 8, Prop. 12), has a countable base for open sets, in other words (loc.
cit., Def. 4) is a separable metrizable space. For, writing S = h(K --- N) , if
there exists a countable set H ⊂ F such that S ⊂ H , then the subspace H
of F has a countable base for open sets, therefore so does the subspace S ,
§5 measurable functions and sets INT IV.x141

therefore S has a countable subset H 0 whose closure in S is equal to S ,

that is (GT, I, §3, No. 1, Prop. 1), S∩H0 = S , in other words S ⊂ H0 . (The
same argument shows that in a separable metrizable space, every subspace
is separable.) Condition b) can be rephrased succinctly:
b) for every compact set K ⊂ X , there exists a negligible set N ⊂ K
such that h(K --- N) is a separable (topological) subspace of F .
The proposed Th. 4 0 can then be formulated as follows: If A is a
measurable subset of X then a mapping f : A → F is measurable in the
sense of Def. 8 if and only if it satisfies the following two conditions:
−1
a0 ) For every closed ball B in F , the set f (B) ⊂ A is measurable;
b 0 ) for every compact set K ⊂ A , there exists a negligible set N ⊂ K
such that f (K --- N) is a separable subspace of F .
For, suppose f is measurable. By Def. 8, f has an extension h : X → F
that is measurable. Given any closed ball B in F and any compact set
−1
K ⊂ A , we know from the original Th. 4 that h (B) is measurable and
that there exists a negligible set N ⊂ K such that h(K --- N) is a separable
subspace of F . Then

−1 −1
A ∩ h (B) = {x ∈ A : h(x) ∈ B } = {x ∈ A : f (x) ∈ B } = f (B) ,

−1
thus f (B) is the intersection of two measurable sets, whence a 0 ) is satisfied.
And 
f (K --- N) = (hA)(K --- N) ⊂ h(K --- N) ;
thus f (K --- N) is a subspace of the separable subspace h(K --- N) , whence b 0 )
is satisfied.
Conversely, suppose f satisfies a 0 ) and b 0 ). Fix a point y0 ∈ F and let
h : X → F be the extension of f to X by y0 ; to prove that f is measurable,
it will suffice by Def. 8 to show that h is measurable, and by the original
Th. 4 we need only verify that h satisfies the conditions a) and b).
a) If B is a closed ball in F , then
 −1
−1  f (B) if y0 ∈
/B
h (B) =
 −1
f (B) ∪ (X --- A) if y0 ∈ B ;
−1 −1
since f (B) is measurable by a 0 ), we see that h (B) is also measurable.
b) Let K ⊂ X be compact; we seek a negligible set N ⊂ K such that
h(K --- N) is separable. Since A is measurable, K ∩ A is integrable (No. 2,
Prop. 3), hence there exists a sequence (K n ) of pairwise disjoint compact
INT IV.x142 extension of a measure. l p spaces §5
S
subsets of K ∩ A such that the set N0 = K ∩ A --- Kn is negligible (§4,
n
No. 6, Cor. 2, 2◦ of Th. 4). For each n , since Kn ⊂ A is compact, by b 0 )
there exists a negligible
S set Nn ⊂ Kn such that f (Kn --- Nn ) is separable.
0
The set N = N ∪ Nn is negligible, N ⊂ K ∩ A ⊂ K , and
n
[ [  [
K ∩ A = N0 ∪ Kn = N 0 ∪ Nn ∪ (Kn --- Nn ) = N ∪ (Kn --- Nn ) ,
n n n
S
therefore K ∩ A --- N = (Kn --- Nn ) ; then
n
[
f (K ∩ A --- N) = f (Kn --- Nn ) ,
n

thus f (K ∩ A --- N) is the union of a sequence of separable subspaces of F ,

therefore

(∗) f (K ∩ A --- N) is separable.

We assert that h(K --- N) is separable. For,

K --- N = [(K --- N) ∩ (X --- A)] ∪ [(K --- N) ∩ A] ⊂ (X --- A) ∪ [(K ∩ A --- N)] ,

whence

h(K --- N) ⊂ h(X --- A) ∪ h(K ∩ A --- N) = {y 0 } ∪ f (K ∩ A --- N) ;

in view of (∗), h(K --- N) is a subspace of a separable subspace of F , hence is

separable. Thus h satisfies condition b), which completes the proof that f
is measurable.
No. 5, Prop. 8 0 . Let X be a locally compact space equipped with a
measure, A a measurable subset of X , f : X → R , and D a countable
dense subset of R . Then:
1) If f is measurable, then the set {x ∈ A : f (x) > a } is measurable
for every a ∈ R .
2) If, for every a ∈ D , the set {x ∈ A : f (x) > a } is measurable,
then f is measurable.
Proof. Let h : X → R be the extension of f to X by 0 .
1) If f is measurable, then h is measurable (Def. 8). For every a ∈ R ,
[a, +∞] is a closed subset of R , therefore the set
−1
{x ∈ A : h(x) > a } = h ([a, +∞])
§5 measurable functions and sets INT IV.x143

is measurable by Prop. 7; consequently the set

−1
{x ∈ A : f (x) > a } = {x ∈ A : h(x) > a } = A ∩ h ([a, +∞])

is measurable.
2) For every a ∈ D , the set
(
{x ∈ A : f (x) > a } if a > 0
{x ∈ X : h(x) > a } =
{x ∈ A : f (x) > a } ∪ (X --- A) if a 6 0

is measurable, therefore h is measurable by the original Prop. 8, and so f

is measurable by Def. 8.
It may be useful to review here the proof of Prop. 8. Suppose
h : X → R is a function such that {x ∈ X : h(x) > a } is measurable
for every a ∈ D . As observed in the note for IV.69, `. 10, the topological
space R can be metrized in such a way that the closed balls are precisely its
closed intervals [a, b] . Since R has a countable dense subset, condition b)
of Th. 4 is trivially satisfied by h , so to prove that h is measurable, it
−1
suffices to show that h ([a, b]) is measurable for every closed interval [a, b]
of R .
For every a ∈ R , the set {x ∈ X : h(x) > a } is measurable: this is
trivial if a = −∞ , and if a > −∞ then there exists a sequence a n ∈ D
such that an < a and an → a , whence
\
{x ∈ X : h(x) > a } = {x ∈ X : h(x) > an }
n

is the intersection of countably many sets that are measurable by hypothesis.

Its complement {x ∈ X : h(x) < a } is therefore also measurable (empty
when a = −∞ ).
For every b ∈ R the set {x ∈ X : h(x) 6 b } is measurable. This is
trivial if b = +∞ . Assuming b < +∞ , there exists a sequence a n ∈ D
such that b < an and an → b . Then
\ \
[ − ∞, b] = [ − ∞, an [ = { [an , +∞] ,
n

−1 T −1
therefore {x ∈ X : h(x) 6 b = h ([ − ∞, b]) = { h ([an , +∞]) is the
n
intersection of a sequence of measurable sets.
INT IV.x144 extension of a measure. l p spaces §5

Finally, if [a, b] is any closed interval in R then the set

−1
h ([a, b]) = {x ∈ X : h(x) > a } ∩ {x ∈ X : h(x) 6 b }

is measurable, therefore h is measurable by Th. 4.

Incidentally, since the measurable subsets of X form a tribe (No. 4,
−1
Cor. 2 of Th. 2), it is clear that the set S = {S ⊂ R : h (S) is measurable}
is also a tribe, and since S contains all closed intervals, it contains the tribe
−1
they generate, namely, the tribe of Borel sets. Thus h (B) is measurable for
every Borel set B in R .
No. 5, Prop. 9 0 . Let X be a locally compact space equipped with a
measure, F a metrizable compact space, A a measurable subset of X , and
f : A → F a mapping measurable in the sense of Def. 8. Then f is the
uniform limit of a sequence gn : A → F of measurable step functions.
See the earliler note for Th. 30 for the definition of ‘measurable step
function’. Let h : X → F be a measurable extension of f (Def. 8). By
the original Prop. 9, there exists a sequence of measurable step functions
hn : X → F such that hn → h uniformly in X ; the sequence of restrictions
gn = hn A meets the requirements.
No. 5, Prop. 10 0 . Let X be a locally compact space equipped with a
measure, F a separable Banach space (over K = R or C ), F 0 its dual,
and let (a0n ) be a weakly dense sequence in the unit ball of F 0 .
Let A be a measurable subset of X , f : A → F a mapping. Then the
following conditions are equivalent:
a) f is measurable.
b) For every n , the scalar-valued function x 7→ hf (x), a 0n i (x ∈ A) is
measurable.
Write fn : A → K for the function fn (x) = hf (x), a0n i (x ∈ A) . Let
h : X → F be the extension of f to X by 0 ∈ F , and, for each n , let
hn : X → F be the extension of fn to X by 0 ∈ K . Note that

(∗) hh(x), a0n i = hn (x) for all x ∈ X .

Then:
f is measurable ⇔ h is measurable
⇔ hn is measurable for all n
⇔ fn is measurable for all n
(the first equivalence by Def. 8; the second by (∗) and the original Prop. 10;
the third by Def. 8), thus a) ⇔ b).
§5 measurable functions and sets INT IV.x145

No. 5, Cor. 1 0 of Prop. 10. Let X be a locally compact space equipped

with a measure, F any Banach space, and A a measurable subset of X . In
order that a mapping f : A → F be measurable, it is necessary and sufficient
that it satisfy the following two conditions:
a 0 ) For every a0 ∈ F0 , the scalar function x 7→ hf (x), a0 i (x ∈ A) is
measurable;
b 0 ) for every compact set K ⊂ A , there exist a countable set H ⊂ F
and a negligible set N ⊂ K such that f (K --- N) ⊂ H .
“Necessity”: Assuming f measurable, let h : X → F be a measurable
extension of f to X (Def. 8). By the original
 Cor. 1, h satisfies conditions a)
and b) of that corollary; then f = hA satisfies a 0 ) by a) and Def. 8. If K
is a compact subset of A , and if H ⊂ F and N ⊂ K are chosen as in b) of
Cor. 1, then
f (K --- N) = h(K --- N) ⊂ H ,
thus f satisfies b 0 ).
“Sufficiency”: Suppose f satisfies a 0 ) and b 0 ). Fix a point y0 ∈ F and
let h be the extension of f to X by y0 . To show that f is measurable,
equivalently (Def. 8) that h is measurable, it will suffice to show that h
satisfies the conditions a) and b) of Cor. 1.
Now, h satisfies b) of Cor. 1 by the argument for the proof of b) in the
earlier proof of Th. 4 0 . Also, h satisfies a) of Cor. 1: for, if a 0 ∈ F0 then
the function x 7→ hh(x), a0 i (x ∈ X) is the extension to X of the function
x 7→ hf (x), a0 i (x ∈ A) by hy0 , a0 i , hence is measurable by a 0 ) and Def. 8.
No. 5, Cor. 2 0 of Prop. 10. Let X and F be as in the original Cor. 2,
and let A be a measurable subset of X . In order that a mapping f : A → F
be measurable, it is necessary and sufficient that, for every a 0 ∈ F0 , the scalar
function x 7→ hf (x), a0 i (x ∈ A) be measurable.
For, let h : X → F be the extension of f to X by 0 ∈ F . For each
a0 ∈ F0 , the function

(1) x 7→ hh(x), a0 i (x ∈ X)

is the extension to X by 0 of the function

(2) x 7→ hf (x), a0 i (x ∈ A) .

Then:

f is measurable ⇔ h is measurable
⇔ the functions (1) are all measurable
⇔ the functions (2) are all measurable
INT IV.x146 extension of a measure. l p spaces §5

(the first equivalence, by Def. 8; the second, by the original Cor. 2; the third,
by Def. 8).
No. 5, Prop. 11 0 . Statement and proof as in the preceding, with “Cor. 2”
replaced by Prop. 11” (but note that the conditions on F are different in
the two statements).
IV.81, `. 2, 3.
“ . . . if V0 ⊂ V , B0 ⊃ B and δ 0 6 δ , then

W(V0 , B0 , δ 0 ) ⊂ W(V, B, δ) ”

Let (f, g) ∈ left side, and write

M = {x ∈ B : f (x), g(x) ∈/ V } , M0 = {x ∈ B0 : f (x), g(x) ∈/ V0 } ;

the assumptions B ⊂ B0 , V ⊃ V0 clearly imply M ⊂ M0 , therefore

|µ|*(M) 6 |µ|*(M0 ) 6 δ 0 6 δ ,

whence (f, g) ∈ right side.

Remark. f (x), g(x) ∈ V means that f (x) and g(x) are ‘near of
order V ’ (or ‘ V-close’). Then (f, g) ∈ W(V, B, δ) means, so to speak, that
f and g are uniformly within V of each other on the complement, relative
to B , of a subset of B of exterior measure 6 δ . Since A has an abundance
of integrable (indeed, compact) subsets B , as B expands and V and δ
shrink, it is clear that one has here the makings of a concept of uniform
convergence for measurable functions f : A → F .
IV.81, `. 4.
“ . . . it therefore suffices to verify the axiom (U 0III ) (GT, II, §1, No. 1).”
One disposes of (U0I ) and (U0II ) as follows. Since every entourage V
for F contains the diagonal of F × F , the set


M = {x ∈ B : f (x), f (x ∈/ V} = ∅,

thus |µ|*(M) = 0 < δ , whence (f, f ) ∈ W(V, B, δ) for all functions

−1
f : A → F . And f (x), g(x) ∈ V ⇔ g(x), f (x) ∈ V , therefore

−1
(f, g) ∈ W(V, B, δ) ⇔ (g, f ) ∈ W( V, B, δ) .

−1 −1
Writing W = W(V, B, δ) and W 0 = W( V, B, δ) , we then have W0 = W .
§5 measurable functions and sets INT IV.x147

IV.81, `. 4–6.
2
“Now, if V0 is an entourage such that V0 ⊂ V , then
W(V0 , B, δ/2) ◦ W(V0 , B, δ/2) ⊂ W(V, B, δ) .”
If δ1 , δ2 are > 0 , then
W(V0 , B, δ1 ) ◦ W(V0 , B, δ2 ) ⊂ W(V, B, δ1 + δ2 ) .
For, let (f, g) ∈ left side; then, for a suitable function h , one has (f, h) ∈
second factor and (h, g) ∈ first factor (GT, II, §1, No. 1, footnote). Since




f (x), h(x) ∈ V0 & h(x), g(x) ∈ V0 ⇒ f (x), g(x) ∈ V0 ◦ V0 ⊂ V ,
one has




f (x), g(x) ∈/V ⇒ h(x), g(x) ∈/ V0 or f (x), h(x) ∈/ V0 ;
thus, writing

M = {x ∈ B : f (x), g(x) ∈/ V}


M1 = {x ∈ B : h(x), g(x) ∈ / V0 }


M2 = {x ∈ B : f (x), h(x) ∈/ V0 } ,
we see that M ⊂ M1 ∪ M2 , therefore
|µ|*(M) 6 |µ|*(M1 ) + |µ|*(M2 ) 6 δ1 + δ2 ,
whence (f, g) ∈ right side.
IV.81, `. 9–11.
“ . . . there exists a compact set K ∈ K contained in B such that
|µ|(B --- K) 6 δ , and therefore W(V, K, δ) ⊂ W(V, B, 2δ) .”
Since B is integrable, there exists a compact set K 1 ⊂ B such that
|µ|(B --- K1 ) 6 δ/2 (§4, No. 6, Cor. 1 of Th. 4). In turn, since K is µ-dense
in A (such a K exists—cf. IV.77, `. 1), K 1 has a subset K ∈ K such that
|µ|(K1 --- K) 6 δ/2 (by b) of Prop. 12), therefore |µ|(B --- K) 6 δ/2 + δ/2 .
Suppose (f, g) ∈ W(V, K, δ) . Write


M1 = {x ∈ K : f (x), g(x) ∈ / V}


M2 = {x ∈ B : f (x), g(x) ∈ / V}.
By assumption |µ|*(M1 ) 6 δ ; we wish to show that |µ|*(M2 ) 6 2δ .
From K ⊂ B we see that M1 ⊂ M2 . Moreover, M2 --- M1 ⊂
B --- K :
for, if x ∈ M2 --- M1 then x ∈ M2 ⊂ B , therefore f (x), g(x) ∈ /
V ; if
one had x ∈ K , then from x ∈ / M1 we would infer that f (x), g(x) ∈ V ,
a contradiction. From M2 --- M1 ⊂ B --- K we infer that
M2 ⊂ M1 ∪ (B --- K) ,
whence |µ|*(M2 ) 6 |µ|*(M1 ) + |µ|*(B --- K) 6 δ + δ .
INT IV.x148 extension of a measure. l p spaces §5

IV.81, `. −14 to −11.

“ . . . the set M of x ∈ B such that f (x) 6= g(x) is µ-integrable, be-
cause it is the inverse image, under the µ-measurable mapping
x 7→ (f (x), g(x)) , of the complement of the diagonal in F × F , which is
open (No. 5, Prop. 7) ”
The mapping is measurable by Th. 10 (the generalization of No. 3, Th. 1
proved in the note for IV.80, `. −17 to −14). The measurability of M then
follows from Prop. 70 (the generalization of No. 5, Prop. 7 proved in the same
note), and M is then integrable by No. 6, Cor. 1 of Th. 5.
IV.81, `. −11 to −9.
“ . . . if |µ|(M) = α > 0 , there exists a compact subset K ⊂ M such
that |µ|(M --- K) < α/2 and such that the restrictions of f and g to K
are continuous ”
Since M is integrable, there exists a compact set K 0 ⊂ M such that
|µ|(M --- K0 ) < α/4 (§4, No. 6, Cor. 1 of Th. 4). 
The set K of compact sets K ⊂ M such that (f, g) K is continuous
 
(equivalently, both f K and g K are continuous) is µ-dense in M by the
condition a) of No. 10, Prop. 15; applying b) of No. 8, Prop. 12 (with M
playing the role of A ), there exists a set K ∈ K such that K ⊂ K 0 and
|µ|(K0 --- K) < α/4 .
Finally, |µ|(M --- K) 6 |µ|(M --- K0 ) + |µ|(K0 --- K) < α/4 + α/4 .
IV.81, `. −9, −8.
“ . . . there exists an entourage V0 of F such that (f (x), g(x)) ∈ / V0
for all x ∈ K ”
The mapping x 7→ (f (x), g(x)) (x ∈ K) is a continuous mapping of the
compact space K onto a compact subset S of F × F such that S ∩ ∆ = ∅ ,
where ∆ is the diagonal of F × F (recall that K ⊂ M and f (x) 6= g(x) for
all x ∈ M ). The closed entourages form a fundamental system of entourages
for F T(GT, II, §1, No. 2, Cor. 2 of Prop. 2); since F is Hausdorff, it follows
that V = ∆ , where V runs over the set of all closed entourages for F
(loc. cit., Prop. 3). Thus, since S ∩ ∆ = ∅ , for every p ∈ S there exists a
closed entourage Vp such that p ∈ / Vp , that is, p ∈ { Vp ; the sets { Vp
thus form an open covering of the compact set S . Let p 1 , . . . , pn be a finite
set of points of S such that
S ⊂ { Vp1 ∪ · · · ∪ { Vpn = { (Vp1 ∩ · · · ∩ Vpn ) ;
then V0 = Vp1 ∩ · · · ∩ Vpn is a closed entourage such that S ∩ V0 = ∅ ,
therefore



x ∈ K ⇒ f (x), g(x) ∈ S ⇒ f (x), g(x) ∈ / V0 ,
thus V0 meets the requirements.
§5 measurable functions and sets INT IV.x149

IV.81, `. −8.
“ . . . consequently (f, g) ∈ / W(V0 , B, α/2) .”


Let M0 = {x ∈ B : f (x), g(x) ∈ / V0 } ; we are to show that
|µ|*(M0 ) > α/2 .
Clearly K ⊂ M0 by the construction of V0 , and M0 ⊂ M
because V0
contains the
diagonal ∆ of F × F (if x ∈ M 0 then f (x), g(x) ∈ / V0 , hence
f (x), g(x) ∈ / ∆ , that is, f (x) 6= g(x) , and so x ∈ M ); thus K ⊂ M 0 ⊂ M .
Then
α/2 > |µ|(M --- K) = |µ|(M) − |µ|(K) = α − |µ|(K) ,
whence |µ|(K) > α − α/2 = α/2 and finally |µ|*(M 0 ) > |µ|(K) > α/2 .
To summarize: Write S = S (A, µ; F) ( F a Hausdorff uniform space)
and let B be a fundamental system of entourages for F . For every integrable
set B ⊂ A ,
\
{(f, g) ∈ S × S : f = g a.e. in B } = W(V, B, δ) .
V∈B, δ>0

Proof of ⊂ : Suppose (f, g) ∈ left side, so that the set

N = {x ∈ B : f (x) 6= g(x) }

is negligible. Let V be any entourage for F .
Since V contains the diago-

nal ∆ of F × F , one has {x ∈ B : f (x), g(x) ∈ / V } ⊂ N , therefore


|µ|*({x ∈ B : f (x), g(x) ∈/ V }) 6 |µ|*(N) = 0 < δ

for all δ > 0 ; thus (f, g) ∈ W(V, B, δ) for every entourage V (in particular
for every V ∈ B ) and every δ > 0 , whence (f, g) ∈ right side.
Proof of ⊃ : Assuming (f, g) ∈ / left side, we are to show that
(f, g) ∈
/ right side.
Writing M = {x ∈ B : f (x) 6= g(x) } (an integrable set), by as-
sumption |µ|(M) = α > 0 , and the earlier argument shows that there
exists a closed entourage V0 such that (f, g) ∈ / W(V0 , B, α/2) . Choose
V ∈ B so that V ⊂ V0 ; then W(V, B, α/2) ⊂ W(V0 , B, α/2) , therefore
(f, g) ∈
/ W(V, B, α/2) and so (f, g) ∈ / right side.
IV.81, `. −7 to −5.
“ . . . if F is Hausdorff, then the intersection of all the entourages of
S (A, µ; F) is the set of pairs (f, g) such that f (x) = g(x) locally almost
everywhere in A .”
(It is then a triviality that the same is true of the intersection of any
fundamental system of entourages of S (A, µ; F) .)
INT IV.x150 extension of a measure. l p spaces §5

Recall that the entourages of S (A, µ; F) are the supersets of the sets
W(V, B, δ) , where δ > 0 , V runs over the set of entourages for the uni-
formity of F , and B runs over the set of integrable subsets of A . A fun-
damental system of entourages is also given by the sets W(V, K, δ) , where
V runs over a fundamental system of entourages for the uniformity of F ,
K runs over any set K of compact subsets of A that is µ-dense in A , and δ
runs over, say, the values 1/n ( n = 1, 2, 3, . . . ).
Write S = S (A, µ; F) and let N be the intersection of all the en-
tourages for S . To be specific, let K be the set of all compact subsets
of A , and let B be a fundamental system of entourages for the uniformity
of F , so that
\
N = W(V, K, δ)
K∈K, V∈B, δ>0
\  \ 
= W(V, K, δ)
K∈K V∈B, δ>0
\
= {(f, g) ∈ S × S : f = g a.e. in K }
K∈K
= {(f, g) ∈ S × S : for every compact set K ⊂ A , f = g a.e. in K }

(the third equality, by the formula proved in the preceding note). For each
pair (f, g) ∈ S × S , write

M(f, g) = {x ∈ A : f (x) 6= g(x) } ;

M(f, g) is a measurable set (by the argument in the note for IV.81, `. −14
to −11). The assertion to be proved is that

(f, g) ∈ N ⇔ M(f, g) is locally negligible,

in other words (No. 2, Prop. 5) that

(f, g) ∈ N ⇔ K ∩ M(f, g) is negligible for every compact set K ⊂ X .

Proof of ⇐ : In particular, for every compact set K ⊂ A , the set

K ∩ M(f, g) is negligible; since

K --- K ∩ M(f, g) = {x ∈ K : f (x) = g(x) } ,

this means that f = g almost everywhere in K . Thus (f, g) ∈ N by the

formula for N proved above.
§5 measurable functions and sets INT IV.x151

Proof of ⇒ : Assume (f, g) ∈ N and let K ⊂ X be compact; we are

to show that K ∩ M(f, g) is negligible. Since M(f, g) is measurable, there
exists a partition
[
(∗) K ∩ M(f, g) = N ∪ Kn
n
with N negligible and (Kn ) a sequence of compact sets (see the note for
IV.67, `. −1).
Now, Kn ⊂ M(f, g) ⊂ A . Since (f, g) ∈ N , we know that f = g
almost everywhere in Kn (by the formula for N proved above), that is,
the set
{x ∈ Kn : f (x) 6= g(x) } = Kn ∩ M(f, g) = Kn
is negligible. It then follows from (∗) that K ∩ M(f, g) is negligible.
IV.81, `. −5 to −1.
“The Hausdorff uniform space associated with S (A, µ; F) , which we
shall denote S(A, µ; F) . . . therefore consists of the equivalence classes for
the relation hh f (x) = g(x) locally almost everywhere in A ii in the set
S (A, µ; F) .”
It is implicit in the wording that the notation S(A, µ; F) is reserved for
the case that F is Hausdorff, as the following argument will show.
Consider the uniform space S = S (A, µ; F) , with F not necessarily
Hausdorff. The associated Hausdorff space S can be obtained as follows.
As in the preceding note, let N be the intersection of all the entourages
for S . The condition hh (f, g) ∈ N ii clearly defines an equivalence relation
f ∼ g in S ; write f˙ for the equivalence class of f , and
S = S / ∼ = {f˙ : f ∈ S }
for the quotient set. For each entourage W for S , write
Ẇ = {(f˙, ġ) : (f, g) ∈ W } ;
it is routine to check that the sets Ẇ form a fundamental system of en-
tourages for a uniform structure on S , and this structure is Hausdorff since
\ \
Ẇ = {(f˙, ġ) : (f, g) ∈ W } = {(f˙, ġ) : (f, g) ∈ W for all W }
W W
\
= {(f˙, ġ) : (f, g) ∈ N = W } = {(f˙, ġ) : f˙ = ġ }
(GT, II, §1, No. 2, Prop. 3).
In brief, S is derived from S by passing to quotients for the equivalence
relation hh (f, g) ∈ N ii . For this to be the same relation as
hh f (x) = g(x) locally almost everywhere in A ii ,
one requires the formula for N in the preceding note, whose proof assumes
that F is Hausdorff.
INT IV.x152 extension of a measure. l p spaces §5

IV.82, `. 7.
“It follows from No. 10, Prop. 16 that ψ is bijective.”

S It is tacit that F is Hausdorff (see the preceding note). Let A 0 =

Aλ and N = A --- A0 ; thus A0 is measurable (IV.77, `. −8 to −6) and
λ∈L
N is locally negligible.
ψ is injective. Suppose f, g ∈ S ˙
 (A, µ; F)
 with ψ( f ) = ψ(ġ) , that is,
fλ = ġλ for all λ , in other words f Aλ = g Aλ locally almost everywhere
˙
for every λ ; thus, for each λ there exists a locally negligible set N λ ⊂ Aλ
such that f = g on Aλ --- Nλ . Then (pairwise disjointness)
[ [ [
f = g on (Aλ --- Nλ ) = Aλ --- Nλ = A0 --- N0 ,
λ λ λ

S
where N0 = Nλ ; since the Aλ are locally countable, so are the Nλ ,
λ
therefore N0 is also locally negligible (loc. cit.).
Since N∪N0 is locally negligible and A --- (N∪N 0 ) = (A --- N) --- N0 =
A0 --- N0 , it follows that f = g locally almost everywhere (on A ), that is,
f˙ = ġ .
Q
ψ is surjective. Let u = (uλ ) ∈ S(Aλ , µ; F) ; we seek a function
λ
f ∈ S (A, µ; F) such that ψ(f˙) = u , that is, f˙λ = uλ for all λ .
For each λ , let g λ be any mapping in S (Aλ , µ; F) whose equivalence
class is uλ (it is not assumed that g λ is the restriction to Aλ of a mapping g
on A ). Let f0 = A0 → F be the mapping such that f0 Aλ = g λ for all λ
(disjointness); since g λ is µ-measurable in Aλ for every λ , it follows that f0
is µ-measurable in A0 (No. 10, Prop. 16). Let f1 : X → F be a measurable
extension of f0 to X (No. 10, Def. 8), and define f = f1 A ; then f is
µ-measurable (Def. 8), that is, f ∈ S (A, µ; F) , and, for all λ ,
      
fλ = f Aλ = (f1 A)Aλ = f1 Aλ = (f1 A0 )Aλ = f0 Aλ = g λ

(the first equality defines fλ , as in `. 4, 5), whence f˙λ = uλ for all λ .

IV.82, `. 7–9.
“Consider an entourage T of S(A, µ; F) that is the canonical image of
a W(V, B, δ) , where B is a compact subset of A ”
The sets W = W(V, B, δ) form a fundamental system of entourages
for S (A, µ; F) (see the note for IV.81, `. 9–11), therefore the sets T =
Ẇ = {(f˙, ġ) : (f, g) ∈ W } form a fundamental system of entourages for
S(A, µ; F) (see the Note for IV.81, `. −5 to −1).
§5 measurable functions and sets INT IV.x153

IV.82, `. 9, 10.
PJ of λ ∈ L such that B ∩ Aλ 6= ∅ is countable (No. 9),
“ . . . the set
and |µ|(B) = |µ|(B ∩ Aλ ) ”
λ∈J
Since the family (Aλ )λ∈J is locally countable, J is countable (para-
graph following No. 9, Def. 7). With notations for A 0 and N as in the note
for `. 7, one then has
[ 
B = B ∩ A = (B ∩ A0 ) ∪ (B ∩ N) = B ∩ Aλ ∪ (B ∩ N) ,
λ∈J

where B ∩ N is negligible (No. 2, Prop. 5), whence the formula for |µ|(B)
(§4, No. 5, Prop. 9).
IV.82, `. 13–16.
“The image of T under ψ × ψ is . . . which proves the proposition.”
Let us pause to review product
Q uniform spaces. Let (X λ )λ∈L be a
family of uniform spaces, X = Xλ the product set, and prλ : X → Xλ
λ∈L
(λ ∈ L) the family of projection mappings. The product uniformity of X is
the initial uniform structure for the family (pr λ )λ∈L , that is, the coarsest
uniform structure on X that renders uniformly continuous every pr λ (GT,
II, §2, No. 6, Def. 4). If λ0 ∈ L and Wλ0 ⊂ Xλ0 × Xλ0 , then
Y
(prλ0 × prλ0 )−1 (Wλ0 ) = Wλ0 × (Xλ × Xλ ) .
λ6=λ0

More generally, if H is a finite subset of L , and if, for each λ ∈ H , W λ is

a subset of Xλ × Xλ , then
\ Y Y
(∗) (prλ × prλ )−1 (Wλ ) = Wλ × (Xλ × Xλ ) .
λ∈H λ∈H λ∈L -- H

If H runs over the set of finite subsets of L and if, for each λ ∈ H , W λ runs
over a fundamental system of entourages for X λ , then the sets (∗) form a
fundamental system of entourages for the product uniformity on X (GT, loc.
cit., No. 3, Prop. 4). {See also J.L. Kelley, General topology (Van Nostrand,
New York, 1955), p. 182.}
ψ −1 is uniformly continuous. Given a basic entourage W for S =
S (A, µ; F) , so that the set

Ẇ = {(f˙, ġ) : (f, g) ∈ W }

INT IV.x154 extension of a measure. l p spaces §5

is a basic entourage for S = S(A, µ; F) (see the note for IV.81, `. −5 to −1),
itQwill suffice to show that (ψ × ψ)(Ẇ) ⊃ V for some entourage V for
Sλ , where Sλ = S(Aλ , µ; F) for all λ ∈ L ; for then, writing
λ∈L

Y Y
Φ = ψ −1 × ψ −1 = (ψ × ψ)−1 : Sλ × Sλ → S × S ,
λ∈L λ∈L

one has Φ−1 (Ẇ)

Q = (ψ × ψ)(Ẇ) ⊃ V , thus Φ
−1
(Ẇ) contains—hence is—an
−1
entourage for Sλ , which will show that ψ is uniformly continuous (GT,
λ
II, §2, No. 1, remarks following Def. 1).
For the basic entourage W for S , take W = W(V, B, δ) , with V
an entourage for F , B a compact subset of A , and δ > Q
0 , so that Ẇ
is the set T of the text; we seek an entourage V for Sλ such that
λ
(ψ × ψ)(Ẇ) ⊃ V .
Let H ⊂ J ⊂ L ( J countable, H finite) be chosen as in the text
and let m be the number of elements of H . For λ ∈ H write W λ =
W(V, Aλ , δ/2m) , and

Ẇλ = {(f˙λ , ġ λ ) : f λ , g λ ∈ Sλ = S (Aλ , µ; F) } = {(uλ , vλ ) : uλ , vλ ∈ Sλ } ;

then Ẇλ is an entourage for Sλ , and the set

Y   Y  Y Y
V = Ẇλ × (Sλ × Sλ ) ⊂ Sλ × Sλ
λ∈H λ∈L -- H λ∈L λ∈L

Q
is an entourage for the product uniformity on Sλ . Let us show that
λ∈L
V ⊂ (ψ × ψ)(Ẇ) .
Let (u, v) ∈ V . Say u = (uλ )λ∈L , v = (vλ )λ∈L , so that (uλ , vλ ) ∈ Ẇλ
for all λ ∈ H (but Q no restriction on the u λ , vλ ∈ Sλ for λ ∈ L --- H ).
Since ψ : S → Sλ is bijective, there exist functions f, g ∈ S such
λ∈L


that ψ(f˙) = u and ψ(ġ) = v , and so (ψ × ψ) (f, ˙ ġ) = (u, v) . Thus, for

every λ ∈ L , uλ (resp. vλ ) is the equivalence class of fλ = f Aλ (resp.

gλ = g Aλ ). By the definition of V , for every λ ∈ H one has (u λ , vλ ) ∈ Ẇλ
and so

|µ|*({x ∈ Aλ : f (x), g(x) ∈ / V } 6 δ/2m .
Write

Mλ = {x ∈ Aλ : f (x), g(x) ∈ / V } for λ ∈ H ,


M = {x ∈ B : f (x), g(x) ∈ / V};
§5 measurable functions and sets INT IV.x155

for all λ ∈ H we have M ∩ Aλ ⊂ Mλ and |µ|*(Mλ ) 6 δ/2m .

If we can show that |µ|*(M) 6 δ , it will follow that (f, g) ∈ W(V, B, δ) ,
whence (f˙, ġ) ∈ Ẇ , and so


(u, v) = (ψ × ψ) (f˙, ġ) ∈ (ψ × ψ)(Ẇ) ,

as we wish to show. S
Now, M ⊂ B ⊂ A , and we have a partition A = N ∪ Aλ with N
λ∈L
locally negligible. Then
 [ 
M=M∩B∩A=M∩B∩ N∪ Aλ
λ∈L
h [ i
= (M ∩ B ∩ N) ∪ M ∩ B ∩ Aλ
λ∈L
h [ i
= (M ∩ N) ∪ M ∩ B ∩ Aλ
λ∈J
h[ i
= (M ∩ N) ∪ M ∩ Aλ
λ∈J
h [ i h [ i
= (M ∩ N) ∪ M ∩ Aλ ∪ M ∩ Aλ ;
λ∈H λ∈J -- H

but M ∩ Aλ ⊂ Mλ for all λ ∈ H , and M ∩ Aλ ⊂ B ∩ Aλ for all λ ∈ J --- H ,

therefore h [ i h [ i
M ⊂ (M ∩ N) ∪ Mλ ∪ B ∩ Aλ .
λ∈H λ∈J -- H

The first term on the right side is negligible; in the middle term, |µ|*(M λ ) 6
δ/2m for all λ ∈ H ; and the third term has outer measure 6 δ/2 by the
choice of H ; therefore

|µ|*(M) 6 0 + m · (δ/2m) + δ/2 = δ ,

as we wished to show.
ψ is uniformly Qcontinuous. Given a basic entourage V for the prod-
uct uniformity on Sλ , it will suffice to find an entourage W for S =
λ∈L
S (A, µ; F) such that (ψ × ψ)(Ẇ) ⊂ V (GT, loc. cit.).
We can suppose that
Y Y
V = Ẇλ × Sλ × S λ ,
λ∈H λ∈L -- H
INT IV.x156 extension of a measure. l p spaces §5

where H is a finite subset of L and, for each λ ∈ H , W λ = W(V, Bλ , δ)

in the context of Sλ = S (Aλ , µ; F) , with V a basic entourage for F ,
Bλ a compact subset
ofQ Aλ , and
Q δ > 0 ; that is, V is the set of pairs of
families (uλ ), (vλ ) ∈ Sλ × Sλ such that, for λ ∈ H , (uλ , vλ ) ∈ Ẇλ .
λ∈L λ∈L S
Construct W as follows: let B = Bλ (a compact subset of A such
λ∈H
that B ∩ Aλ = Bλ for λ ∈ H , and B ∩ Aλ = ∅ for λ ∈ L --- H ), and form
the entourage W = W(V, B, δ) for S ; let us show that the entourage Ẇ
for S = S(A, µ; F) satisfies (ψ × ψ)(Ẇ) ⊂ V .


Let (f, g) ∈ W . Then (ψ × ψ) (f, ˙ ġ) = (uλ ), (vλ ) , where uλ

(resp. vλ ) is the equivalence class of the function f λ = f Aλ (resp. gλ =


g Aλ ), so that (uλ , vλ ) = (f˙λ , ġλ ) ; to show that (ψ × ψ) (f, ˙ ġ) ∈ V we
need only show that (uλ , vλ ) ∈ Ẇλ for all λ ∈ H .
Let λ ∈ H . Then



{x ∈ Bλ : fλ (x), gλ (x) ∈ / V } ⊂ {x ∈ B : f (x), g(x) ∈ / V};

since the outer measure of the right side is 6 δ (because (f, g) ∈ W ), the
outer measure of the left side is also 6 δ , in other words, (f λ , gλ ) ∈ Wλ ,
and so (uλ , vλ ) ∈ Ẇλ .
IV.82, `. −16 to −14.
“Since each An is the union of a negligible set and a sequence of compact
sets, we can suppose that the An are already compact and pairwise disjoint.”
S S
Say A = N∪ An with N locally negligible; replacing N by N --- An
n S Sn
we can suppose that N ∩ An = ∅ . Then, replacing An by An --- Ak
n k<n
(§4, No. 5) we can suppose that theSAn are pairwise disjoint. For each n ,
there exists a partition An = Nn ∪ Ank with Nn negligible and (Ank ) a
k
sequence of pairwise disjoint compact sets (§4, No. 6, Cor. 2 of Th. 4). Then
 [  [
A = N ∪ Nn ∪ Ank ,
n n,k
S S
where N ∪ Nn is locally negligible, disjoint from Ank , and the Ank
n n,k
are pairwise disjoint compact sets.
IV.82, `. −14.
“Prop. 17 then allows us to suppose that A is compact.”
Writing S = S(A, µ; F) Q
and Sn = S(An , µ; F) , we know from Prop. 17
that the mapping ψ : S → Sn is an isomorphism of uniform spaces. If
n
§5 measurable functions and sets INT IV.x157

the compact
Q case is established, so that every S n is metrizable, it will follow
that Sn , hence S , is metrizable (GT, IX, §2, No. 4, Cor. 2 of Th. 1).
n

IV.82, `. −14 to −11.

“If (Vn ) is a countable fundamental system of entourages of F , it is
clear that the W(Vn , A, 1/n) form a fundamental system of entourages of
S (A, µ; F) as n runs over N ”
Our objective is to prove that S = S(A, µ; F) has a countable funda-
mental system of entourages (GT, IX, §2, No. 4, Th. 1), and to this end we
are free to choose any countable fundamental system of entourages (V n )
for F ; replacing Vn by V1 ∩ · · · ∩ Vn , we can suppose that the sequence
(Vn ) is decreasing. It will suffice to show that S = S (A, µ; F) has a
countable fundamental system of entourages.
Let W be any entourage for S . We know (IV.81, `. 7–9) that there
exist an entourage V of F , a compact subset K of A , and a number δ > 0
such that W(V, K, δ) ⊂ W . Choose an index m with V m ⊂ V ; since the
sequence (Vn ) is decreasing, we can choose m as large as we like, so we can
suppose that 1/m < δ . Then (IV.81, `. 1–3)

W(Vm , A, 1/m) ⊂ W(V, K, δ) ⊂ W ,

thus the entourages W(Vn , A, 1/n) form a fundamental system of entourages

for S .
IV.82, `. −11.
“ . . . whence the proposition.”
GT, IX, §2, No. 4, Th. 1.
IV.83, `. 1, 2.
“ . . . the possibility of such a definition follows from the fact that (f n )
is a Cauchy sequence in S (A, µ; F) .”
The crux of the matter is the following construction: If (δ k )k>0 is a
decreasing sequence of numbers > 0 such that δ k → 0 , then there exists a
subsequence (fnk )k>0 (also Cauchy) of (fn ) such that



|µ| {x ∈ B : d fnk (x), fnk+1 (x) > δk } 6 δk (k = 0, 1, 2, . . .).

For, set

Vk = {(y, y 0 ) ∈ F × F : d(y, y 0 ) 6 δk } (k = 0, 1, 2, . . .)

(since d is compatible with the uniformity of F , the V k form a fundamental

system of entourages for F ). Since B is integrable, we may form the sets

Wk = W(Vk , B, δk ) ;
INT IV.x158 extension of a measure. l p spaces §5

Wk is the set of all pairs (f, g) ∈ S = S (A, µ; F) such that

|µ| {x ∈ B : f (x), g(x) ∈/ V k } 6 δk ,

that is,



|µ| {x ∈ B : d f (x), g(x) > δk } 6 δk .

It is clear that the Wk form a fundamental system of entourages for the

uniformity of S (IV.81, `. 1–3). Construct, by induction on k , a sequence
(nk )k>0 of indices n0 < n1 < n2 < . . . such that

(∗) n, n0 > nk ⇒ (fn , fn0 ) ∈ Wk ,

as follows: since (fn ) is Cauchy in S , we may choose n0 so that

n, n0 > n0 ⇒ (fn , fn0 ) ∈ W0

(GT, II, §3, No. 1, second paragraph following Def. 2). Having chosen
n0 < . . . < nr−1 satisfying (∗) for k = 0, . . . , r − 1 , choose an index n r
such that (∗) is satisfied for k = r ; since (∗) continues to hold if n r is re-
placed by a larger integer, we can suppose that n r > nr−1 , which completes
the induction. In particular, for every k one has n k+1 > nk > nk and so
(fnk , fnk+1 ) ∈ Wk , thus



(∗∗) |µ| {x ∈ B : d fnk (x), fnk+1 (x) > δk } 6 δk

by the definition of Wk .
For each m > 0 , the sequence (fmn )n>0 is constructed inductively, as
follows.
m = 0 : Already defined is f0n = fn (n = 0, 1, 2, . . . ).
m = 1 : Set δ1,k = 1/2m+k+1 = 1/21+k+1 . Since (fn ) is Cauchy
in S , the argument for (∗∗) shows that (f n ) has a subsequence (f1n )n>0
(of course Cauchy) such that, writing


M1n = {x ∈ B : d f1n (x), f1,n+1 (x) > 1/21+n+1 }

(a measurable set, by No. 3, Th. 1 and No. 5, Prop. 7), one has |µ|(M 1n ) 6
1/21+n+1 . Note that

X X 1 1X 1 1
|µ|(M1n ) 6 = = · 1 = 1/21 .
21+n+1 2 2n+1 2
n>0 n>0 n>0
§5 measurable functions and sets INT IV.x159

m > 1 : Assuming the (Cauchy) sequence (f m−1,n )n>0 already defined,

set δm,k = 1/2m+k+1 and, by the argument for (∗∗), choose a subsequence
(fmn )n>0 of (fm−1,n )n>0 such that, writing


Mmn = {x ∈ B : d fmn (x), fm,n+1 (x) > 1/2m+n+1 } ,

one has |µ|(Mmn ) 6 1/2m+n+1 (n = 0, 1, 2, . . . ), which completes the in-

duction. Note that
X 1 X 1
(∗∗∗) |µ|(Mmn ) 6 m = 1/2m (m = 1, 2, 3, . . . )
2 2n+1
n>0 n>0

For notational completeness, and convenience later in the proof, we define

M0n = B for all n > 0 .
IV.83, `. 2, 3. S
“Set Mm = Mmn ; then
n>0

∞
X
|µ|(Mm ) 6 |µ|(Mmn ) 6 1/2m ”
n=0

It is intended here that m > 1 ; the inequality then follows from (∗∗∗)
of the preceding note.
By the convention established at the end of theTpreceding note,
M0 = B ; the above inequalities assure that the set N = Mm is negligi-
m>0
ble, irrespective of the value of |µ|(M 0 ) .
IV.83, `. 4, 5.
“ . . . and, for every x ∈ B --- Mm , we have d(fmn (x), fm,n+p (x)) 6
1/2m+n for all n > 0 and all p > 0 ”
It is assumed here that m > 1 . Thus,
[ \
x ∈ B --- Mm = B --- Mmn = (B --- Mmn )
n>0 n>0


means that d fmn (x), fm,n+1 (x) 6 1/2m+n+1 for all n > 0 . It then follows
from the triangle inequality that
p−1 p−1

X
X 1
d fmn (x), fm,n+p (x) 6 d fm,n+k (x), fm,n+k+1 (x) 6
2m+n+k+1
k=0 k=0
p−1
X
1 1 1
= < · 1.
2m+n 2k+1 2m+n
k=0
INT IV.x160 extension of a measure. l p spaces §5

IV.83, `. 8–10.
“ . . . if x ∈ B --- N , there exists an index m such that x ∈
/ M m , which
proves that the sequence (gn (x)) is a Cauchy sequence in F .”
/ N we know that there exists an index m > 0 such that
Since x ∈
x∈ / Mm . But x ∈ B = M0 , so m > 1 ; by the preceding note, (fmn (x))n>0
is Cauchy, hence so its its subsequence (g n (x))n>m .
IV.83, `. 11–15.
“If now B is the union of a sequence (B m ) . . . where Pm is negligi-
ble.”
The notations for m = 0 can be handled as follows. We can suppose
that B0 = ∅ ; setting P0 = ∅ and g0n = fn for all n > 0 , the sequence
(gn0 )n>0 is vacuously pointwise Cauchy in B0 --- P0 . The recursive defini-
tion then proceeds as in the text.
IV.83, `. −11, −10.
“ . . . we are thus reduced to proving the proposition when A is integrable ”
A product of nonempty uniform spaces is complete if and only if every
factor is complete (GT, II, §3, No. 5, Prop. 10).
IV.83, `. −9, −8.
“ . . . for every Cauchy sequence (fn ) in S (A, µ; F) there exists a sub-
sequence (fnk ) that is convergent in A --- N , where N is negligible ”
This is immediate from Lemma 4 and the completeness of F .
But . . . where does the Cauchy sequence (f n ) come from? We are to
show that S = S(A, µ; F) is complete, thus one begins with a Cauchy se-
quence (un ) in S . Say fn ∈ S = S (A, µ; F) with un = f˙n (in the
notation of Prop. 17); we need to know that (f n ) is Cauchy in S .
Lemma. Let A be any measurable subset of X , and W = W(V, B, δ) ,
where V is an entourage for F , B is an integrable subset of A , and δ > 0 ,
and let
Ẇ = {(f˙, ġ) : (f, g) ∈ W } .
Then, for a pair of functions f, g in S , one has

˙ ġ) ∈ Ẇ .
(f, g) ∈ W ⇔ (f,

⇒ : By the definition of Ẇ .
⇐ : Suppose (f˙, ġ) ∈ Ẇ. This means that there exists a pair (h, k) ∈ W
such that (f˙, ġ) = (ḣ, k̇), that is, locally almost everywhere, f = h and g = k.
Let M be a locally negligible subset of A such that f = h and g = k
everywhere in A --- M . Since B is integrable, B ∩ M is negligible (No. 2,
§5 measurable functions and sets INT IV.x161

Cor. 1 of Prop. 5), and f = h and g = k everywhere in B ∩ (A --- M) =

B --- B ∩ M ; it follows that the sets



{x ∈ B : f (x), g(x) ∈ / V } and {x ∈ B : h(x), k(x) ∈/ V}

differ at most by the negligible set B ∩ M , therefore

|µ|* {x ∈ B : f (x), g(x) ∈ / V } = |µ|* {x ∈ B : h(x), k(x) ∈/ V} .



Since (h, k) ∈ W , we know that |µ|*

{x
∈ B : h(x), k(x) ∈
/ V } 6 δ,
therefore |µ|* {x ∈ B : f (x), g(x) ∈/ V } 6 δ and so (f, g) ∈ W . ♦
With notations as in the Lemma, if (u n ) is a sequence in S and if
un = f˙n with fn ∈ S , then

(un ) is Cauchy in S ⇔ (fn ) is Cauchy in S .

For, if W = W(V, B, δ) as in the Lemma, so that the sets W (resp. Ẇ )

form a fundamental system of entourages for S (resp. S ), one has

(um , un ) ∈ Ẇ ⇔ (fm , fn ) ∈ W ,

thus the assertion follows from GT, II, §3, No. 1, second paragraph following
Def. 2.
IV.83, `. −7, −6.
“ . . . is then µ-measurable ”
By 1◦ of the generalization of Egoroff’s Theorem proved in the note for
IV.80, `. −17 to −14 (item “No. 4, Th. 2 0 in that note).
IV.83, `. −6, −5.
“ . . . it follows from the extension of Egoroff’s theorem mentioned in
No. 10 that the sequence (fnk ) converges in measure to f in A .”
By 2◦ of the generalization cited in the preceding note, if K ⊂ A
is compact and δ > 0 , there exists a compact set K 1 ⊂ K such that
|µ|(K --- K1 ) 6 δ and fnk → f uniformly on K1 (we do not need the
continuity of the restrictions to K 1 established there). We are to show that
(fnk ) converges to f in the uniform space S = S (A, µ; F) .
Given a basic entourage W for S , it will suffice to find an index k 0
such that
k > k0 ⇒ (fnk , f ) ∈ W
(GT, II, §3, No. 1, second paragraph following Def. 2). We can suppose that

W = W(V, K, δ) ,
INT IV.x162 extension of a measure. l p spaces §5

where δ > 0 , V = {(y, y 0 ) ∈ F × F : d(y, y 0 ) 6 ε } ( d a metric on F

compatible with its uniformity, and ε > 0 ), and K is a compact subset
of A (IV.81, `. 7–9); thus W is the set of all pairs (g, h) ∈ S × S such
that


|µ| {x ∈ K : d g(x), h(x) > ε } 6 δ .
Let K1 ⊂ K be a compact set such that |µ|(K --- K 1 ) 6 δ and fnk → f
uniformly on K1 , and choose an index k0 such that


k > k0 ⇒ d fnk (x), f (x) 6 ε for all x ∈ K1 ;

it follows that if k > k0 then

{x ∈ K : d fnk (x), f (x) > ε } ⊂ K --- K1 ,

whence



|µ| {x ∈ K : d fnk (x), f (x) > ε } 6 |µ|(K --- K1 ) 6 δ

and so (fnk , f ) ∈ W by the preceding paragraph.

Note that, with W as above, W(f ) = {g ∈ S : (g, f ) ∈ W } is a basic
neighborhood of f in S , and the foregoing shows that for every index n ,
there exists an index k such that nk > n and fnk ∈ W(f ) ; thus f is a
cluster point of the sequence (fn ) (GT, I, §7, No. 3, second paragraph of
Example 1).
IV.83, `. −4.
“ . . . f is a cluster point of the sequence (f n ) in S (A, µ; F) ”
Observed at the end of the preceding note.
IV.83, `. −3, −2.
“ . . . since the sequence (fn ) is by hypothesis a Cauchy sequence, it
converges to f . ”
GT, II, §3, No. 2, Cor. 2 of Prop. 5.
To recapitulate the argument: After reducing to the case that A is
integrable, one shows that every Cauchy sequence (f n ) in the uniform space
S = S (A, µ; F) is convergent (in measure) to an f ∈ S . If, then, (u n ) is
a Cauchy sequence in the (metrizable) uniform space S = S(A, µ; F) , say
un = f˙n with fn ∈ S , one knows that (fn ) is Cauchy in S (see the note
for `. −9, −8) hence is convergent in measure to an f ∈ S ; as (f n , f ) ∈ W
is equivalent to (f˙n , f)
˙ ∈ Ẇ for an entourage W = W(V, B, δ) (loc. cit.),
it follows that (un ) converges to u = f˙ in S .
§5 measurable functions and sets INT IV.x163

IV.84, `. 1.
“Corollary.”
This is a corollary of Lemma 4, using parts of the proof of Prop. 19.
IV.84, `. 10, 11.
“The assertion follows at once from the extension of Egoroff’s theorem
mentioned in No. 10.”
One is assuming that there exists a locally negligible set N ⊂ A such
that fn (x) → f (x) for all x ∈ A --- N . The argument that f is measurable
is given in the note for IV.83, `. −7, −6 (completeness of F is not needed
since the limits f (x) for x ∈ A --- N are assumed to exist); and the argument
that fn → f in the uniform space S (A, µ; F) (that is, in measure) is given
in the note for IV.83, `. −6, −5 (valid for A measurable but not necessarily
integrable; the role of integrability arises in the proof of Prop. 19 because of
the citation of Lemma 4 in order to obtain a limit function f , whereas f
is given here in advance).
IV.84, `. 15.
“ . . . f 0 is a µ-measurable mapping of B --- N into Fb”

Write gk = fnk B --- N ; we know that gk : B --- N → F is µ-measurable
( fnk has a measurable extension to X , therefore so does g k ; see item c00 )
in the note for IV.79, `. 3,4). Let i : F → F b be the canonical injection,
b
and gk = i ◦ gk : B --- N → F ; since i is continuous, gk0 is µ-measurable by
0

the generalization of No. 3, Th. 1 (item No. 3, Th. 1 0 in the note for IV.80,
`. −17 to −14). One has gk0 → f 0 pointwise (on all of B --- N ), therefore f 0
is µ-measurable by 1◦ of the generalization of Egoroff’s theorem (loc. cit.,
item No. 4, Th. 2 0 ).
IV.84, `. 15, 16.
“ . . . the sequence (fn ) converges in measure to f 0 in B --- N by (i) ”

Lemma. If C is a measurable subset of A , then the mapping f 7→ f C
of S (A, µ; F) into S (C, µ; F) is uniformly continuous.

For, f C is measurable (since f has a measurable extension to X , so


does f C ). If K is a compact subset of C then it is also a compact subset
of A ; let V be a basic entourage for F , let δ > 0 , and write

WA = W(V, K, δ) and WC = W(V, K, δ)

for the corresponding entourages for S A = S (A, µ; F) and SC = S (C, µ; F) .

Then, for f, g ∈ S (A, µ; F) ,

 

{x ∈ K : f (x), g(x) ∈/ V } = {x ∈ K : (f C)(x), (g C)(x) ∈/ V},
INT IV.x164 extension of a measure. l p spaces §5

whence obviously
 

(f, g) ∈ WA ⇔ f C, g C ∈ WC .

Since WC describes a basic entourage for SC , and WA is an entourage

for SA that is mapped into WC by f 7→ f C , the mapping is indeed
uniformly continuous.
In particular, if the
 sequence (fn ) is Cauchy (in measure) in the uniform
space 
 SA, then (fn C) is Cauchy in SC ; and if fn → f in measure then
fn C → f C in measure. ♦
Consider now the given convergent (in measure) sequence f n → f
in SA , the ‘countably integrable’ set B ⊂ A , the negligible set N ⊂ B ,
and the subsequence (fnk ) of (fn ) such that fnk (x) → f 0 (x) ∈ F b for all
x ∈ B --- N . Write C = B --- N . 
Writing i : F → F b for the canonical injection, and f 0 = i ◦ (fn C) ,
nk k
we have
fn0 k (x) → f 0 (x) for all x ∈ C = B --- N ;

b .
by (i) of the present Corollary, fn0 k → f 0 (in measure) in SC0 = S (C, µ; F)
On the other hand, since  fn  → f in measure, hence fnk → f in
measure, we know that fnk C → f C in measure by the Lemma; and since
 
i:F→F b is uniformly continuous, it follows that i ◦ (f n C) → i ◦ (f C) in
 k

measure, that is, fn0 k → i ◦ (f C) in SC0 .

Thus, if uk ∈ S0C = S(C, µ; F) b is the equivalence class of fn0 , u is
k

the class of i ◦ (f C) , and u is the class of f 0 , we have uk → u and
0

uk → u0 in S0C ; since
 S0C is Hausdorff (IV.81, `. −7 to −1), it follows that
u = u0 , thus i◦(f C) = f 0 locally almost everywhere
 in C , and since C is a
countable union of integrable sets, in fact i ◦ (f C) = f 0 almost everywhere
in C . But fnk (x) → f (x) for every x ∈ C , and f 0 (x) = f (x) for almost
every x ∈ C , therefore fnk (x) → f (x) for almost every x ∈ C . Since
B = C ∪ N and N is negligible, we conclude that f nk (x) → f (x) for almost
every x ∈ B ; this proves the assertion (ii) of the Corollary.
As for the original assertion, since f n → f in
 measure, the
 argument in
the paragraph before the last shows that i ◦ (f n C) → i ◦ (f C) in measure;
but we now know that i ◦ (f C) = f 0 almost everywhere in C , therefore

i ◦ (fn C) → f 0 in measure.

This is the sense in which the original assertion holds, avoiding the abuse of
notation where the fn take values in F whereas f 0 takes its values in F b.
§5 measurable functions and sets INT IV.x165

IV.84, `. 16, 17.

“ . . . f 0 is therefore equal to f almost everywhere in B .”
Writing C = B --- N as in the preceding note, we know that f 0 : C → F b
b
is measurable (IV.84, `. 15); its extension to X by a point y 0 ∈ F is then a
measurable function on X (No. 10, Def. 8), whose restriction to A is also
measurable (No. 10, Def. 8); denote this function A → F b also by f 0 , so
0
that f ∈ S (A, µ; F) b . Since f (x) = f (x) for almost every x ∈ C (by the
0

next-to-last paragraph of the preceding note) and N is negligible, it follows

that f 0 (x) = (i◦f )(x) for almost every x ∈ B = C∪N , that is, the functions
f 0 , i ◦ f of S (A, µ; F) b are equal almost everywhere in B .
The assertion (ii) of the Corollary is proved in the next-to-last paragraph
of the preceding note.
One observes that the statement of the Corollary makes no complete-
ness assumptions; the complete space F b is merely auxiliary to the proof.
Assuming F to be complete would simplify the proof of the Corollary, but
would needlessly restrict its generality.
IV.84, `. −14 to −11.
“ . . . if Vδ is the entourage of F formed by the pairs (y, z) such that
|y − z| 6 δ , the entourage W(Vδ , B, δ) is the set of pairs (f , g) of measur-
able mappings of A into F such that f − g ∈ T(B, δ) .”
Write S = S (A, µ; F) . By definition, W(V δ , B, δ) is the set of pairs
(f , g) ∈ S × S such that



|µ| {x ∈ B : f (x), g(x) ∈/ Vδ } 6 δ ,

whereas the condition f − g ∈ T(B, δ) means that

|µ| {x ∈ B : |(f − g)(x)| > δ } 6 δ ;

so the assertion will be true if

(†) (f (x), g(x) ∈ / Vδ ⇔ |f (x) − g(x)| > δ .

We have a choice: if the definition of T(B, δ) in the text is retained, then

one should define

(∗) Vδ = {(y, z) ∈ F × F : |y − z| < δ } .

On the other hand, if the proposed definition of V δ (with 6 instead of < )

is retained, then T(B, δ) should have been defined as the set of f ∈ S such
that

|µ| {x ∈ B : |f (x)| > δ } 6 δ .
INT IV.x166 extension of a measure. l p spaces §5

I propose to leave the definitions of T(B, δ) and W(V δ , B, δ) as in the

text, and to make them compatible by redefining V δ as in (∗).
Note that entourages of the form W(V δ , B, δ) are basic for the unifor-
mity of S ; for, if V is any entourage for F , B is any integrable subset
of A , and δ is any number > 0 , one can choose δ 0 > 0 sufficiently small
that both Vδ0 ⊂ V and δ 0 < δ , and so W(Vδ0 , B, δ 0 ) ⊂ W(V, B, δ) . Note
also that since the Vδ are symmetric, so are the W(V, B, δ) , that is,

(f , g) ∈ W(Vδ , B, δ) ⇔ (g, f ) ∈ W(Vδ , B, δ) .

IV.84, `. −10.
“ T(B, δ) + T(B, δ) ⊂ T(B, 2δ) ”
Let f , g ∈ S (A, µ; F) . One has

{x ∈ B : |(f + g)(x)| > 2δ } ⊂ {x ∈ B : |f (x)| > δ } ∪ {x ∈ B : |g(x)| > δ }

by the triangle inequality for the norm on F ; if f , g ∈ T(B, δ) then the

|µ|-measure of each set on the right is 6 δ , so the set on the left has
measure 6 2δ , therefore f + g ∈ T(B, 2δ) .
IV.84, `. −10, −9.
“ T(B, |α|δ) ⊂ αT(B, δ) for every nonzero scalar α such that |α| 6 1 ”
f ∈ T(B, |α|δ) means that


(∗) |µ| {x ∈ B : |f (x)| > |α|δ } 6 |α|δ ,

whereas f ∈ αT(B, δ) means that α−1 f ∈ T(B, δ) , that is,

|µ| {x ∈ B : |α−1 f (x)| > δ } 6 δ ,

in other words


(∗∗) |µ| {x ∈ B : |f (x)| > |α|δ } 6 δ ;

since |α| 6 1 , (∗) implies (∗∗), whence the asserted inclusion.

This shows that αT(B, δ) is a neighborhood of 0 in S (A, µ; F) when
0 < |α| 6 1 ; a subtler argument is needed for arbitrary α 6= 0 (see the note
for `. −2, −1).
IV.84, `. −6.
“ No. 3, Cor. 6 of Th. 1 ”
More precisely, its generalization (item No. 3, Cor. 6 0 of Th. 1 in the
note for IV.80, `. −17 to −14).
§5 measurable functions and sets INT IV.x167

IV.84, `. −2, −1.

“ . . . which completes the proof of assertion (i).”
Write S = S (A, µ; F) . Recall that

W(Vδ , B, δ) = {(f , g) ∈ S × S : f − g ∈ T(B, δ) } .

Since the sets W(Vδ , B, δ) form a fundamental system of entourages for the
uniformity of S , the sets

{f ∈ S : (f , g) ∈ W(Vδ , B, δ) }

form a fundamental system of neighborhoods of g ∈ S for the topology of

convergence in measure. In particular, the sets

{f ∈ S : (f , 0) ∈ W(Vδ , B, δ) } = {f ∈ S : f − 0 ∈ T(B, δ) } = T(B, δ)

form a fundamental system of neighborhoods of 0 in S ; and their translates

g + T(B, δ) = {f ∈ S : f − g ∈ T(B, δ) } = {f ∈ S : (f , g) ∈ W(V δ , B, δ) }

form a fundamental system of neighborhoods of g in S .

Let V be the set of all balanced neighborhoods of 0 in S , that is, the
set of all balanced subsets of S that contain some set T(B, δ) ; it will suffice
to show that V satisfies the conditions (EV I ), (EVII ), (EVIII ) of TVS, I,
§1, No. 5, Prop. 4.
(EVI ): Every T(B, δ) is balanced. For, if f ∈ T(B, δ) and |α| 6 1
then
{x ∈ B : |αf (x)| > δ } ⊂ {x ∈ B : |f (x)| > δ } ;
since the |µ|-measure of the set on the right is 6 δ , so is that of the set on
the left, in other words αf ∈ T(B, δ) . Thus every T(B, δ) belongs to V .
Every T(B, δ) is absorbent. For, given any f ∈ S , the argument
in the text shows that there exists a positive number λ = n 2 such that
f ∈ λT(B, δ) , and since T(B, δ) is balanced, it is absorbent (TVS, loc. cit.,
remarks following Def. 4).
Every set in V is by assumption balanced, and since it contains some
T(B, δ) it is absorbent. Thus V satisfies (EV I ).
(EVIII ): For every T(B, δ) one has T(B, δ/2) + T(B, δ/2) ⊂ T(B, δ) .
(EVII ): This follows from (EVI ) and (EVIII ) (TVS, loc. cit., Remark 2).
For, let U ∈ V and let λ be a scalar 6= 0 ; we are to show that λU ∈ V . At
any rate, since U is balanced, so is λU ; we must show that T(B, δ 0 ) ⊂ λU
INT IV.x168 extension of a measure. l p spaces §5

for some integrable set B ⊂ A and some number δ 0 > 0 . Say T(B, δ) ⊂ U .
Then
2T(B, δ/2) ⊂ T(B, δ/2) + T(B, δ/2) ⊂ T(B, δ) ;
thus 2T(B, δ/2) ⊂ T(B, δ) for every δ > 0 , whence, by induction,

2n T(B, δ/2n ) ⊂ T(B, δ) for every integer n > 1 .

Choose n so large that 2n |λ| > 1 ; since T(B, δ/2n ) is balanced, one has
(2n |λ|)−1 T(B, δ/2n ) ⊂ T(B, δ/2n ) , whence

T(B, δ/2n ) ⊂ |λ|2n T(B, δ/2n ) ⊂ |λ|T(B, δ) ⊂ |λ|U = λU

( |λ|U = λU because U is balanced, hence circled).

IV.85, `. 1–4. R p
“The relation |f | d|µ| 6 δ p+1 implies that if C is the set of x ∈ X
such that |f (x)| > δ , then
Z
δ |µ|*(C) 6 |f |p d|µ| 6 δ p+1 ,
p

whence |µ|*(C) 6 δ , which proves (iii).”

Recall that every p-th power integrable function f : X → F is mea-
surable, that is, LFp (X, µ) ⊂ S (X, µ; F) (No. 6, Th. 5). Let us abbreviate
L p = LFp (X, µ) and S = S (X, µ; F) .
We are to show that the norm topology on L p is finer than the topology
on L p induced by that of S ; that is, for p-th power integrable functions,
convergence in mean of order p implies convergence in measure; in other
words, the canonical injection L p → S is continuous. Since both L p
and S are topological vector spaces and the mapping is linear, it suffices
to establish continuity at 0 .
Thus, given a basic neighborhood U of 0 in S , it suffices to show that
U ∩ L p is a neighborhood of 0 in L p . We can suppose that U = T(B, δ) ,
where B is an integrable (or even compact) set in X , and δ > 0 . It will
suffice to show that T(B, δ) contains a closed ball in L p centered at 0 ;
thus we seek a real number ρ > 0 such that

(∗) {f ∈ L p : Np (f ) 6 ρ } ⊂ T(B, δ) .

Suppose, provisionally, that 0 < ρ < +∞ and f ∈ L p with Np (f ) 6 ρ.

For f to belong to T(B, δ) , one must have


|µ| {x ∈ B : |f (x)| > δ 6 δ .
§5 measurable functions and sets INT IV.x169

Let C = {x ∈ X : |f (x)| > δ } . Since |f | is measurable (No. 3, Cor. 6 of

Th. 1), so is the set C (No. 5, Prop. 8). In fact, C is integrable; for, if ϕ C
is its characteristic function, we have δ p ϕC 6 |f |p ϕC 6 |f |p , whence
Z
δ |µ|*(C) 6 |f |p d|µ| 6 ρp < +∞
p

and we may write |µ|(C) . To assure that f ∈ T(B, δ) , in other words that
|µ|(B ∩ C) 6 δ , it suffices to assure that |µ|(C) 6 δ . But δ p |µ|(C) 6 ρp ,
that is, |µ|(C) 6 (ρ/δ)p ; so f ∈ T(B, δ) is assured if (ρ/δ)p = δ , that is,
ρp = δ p+1 , ρ = δ 1+1/p .
To recapitulate: given a neighborhood T(B, δ) of 0 in S , where
B ⊂ X is integrable and δ > 0 , setting ρ = δ 1+1/p assures that (∗) is
satisfied.
IV.85, `. 5–7.
“In view of (iii), it suffices to show for example that L F1 is dense in SF ,
since by definition K (X; F) is dense in L F1 for the topology of convergence
in mean.”
The argument is easily modified to cover the case of L Fp ; borrowing
from the next note, let us assume already established that L Fp is dense
in SF .
Let U be a nonempty open set in SF ; we are to show that
U ∩ K (X; F) 6= ∅ . By the foregoing assumption, U ∩ L Fp is nonempty,
and by (iii) it is an open set in LFp , therefore


U ∩ K (X; F) = U ∩ LFp ∩ K (X; F) = (U ∩ LFp ) ∩ K (X; F) 6= ∅
by the density of K (X; F) in LFp (§3, No. 4, Def. 2).
IV.85, `. 7–12.
“Now, let f be any element of SF . . . and obviously f −g ∈ T(B, δ) . ”
The objective is to prove that LFp is dense in SF . Given f ∈ SF and
a set T(B, δ) , so that f + T(B,

δ) is a basic neighborhood of f in S F , we
are to show that f + T(B, δ) ∩ LFp 6= ∅ , establishing that f belongs to the
closure of LFp in SF ; thus we seek a function g ∈ LFp such g ∈ f +T(B, δ) .
As shown in the proof of (i), there exists a number n > 0 such that
the (integrable) set C = {x ∈ B : |f (x)| > n } satisfies |µ|(C) 6 δ . Then
|f | 6 1/n on B --- C , thus the function g : X → F defined by g = ϕ B -- C f is
measurable (No. 3, remarks following Cor. 5 of Th. 1) and |g| 6 (1/n)ϕ B -- C .
It follows that g ∈ LFp , since

1/p
Np (g) = Np (|g|) 6 (1/n)Np (ϕB -- C ) = (1/n) |µ|(B --- C) < +∞
(No. 6, Th. 5).
INT IV.x170 extension of a measure. l p spaces §5

Finally, g ∈ f + T(B, δ) . For, g − f = 0 on B --- C , therefore if

|g(x) − f (x)| > δ then x ∈ X --- (B --- C) ; if, moreover, x ∈ B , then
x ∈ B --- (B --- C) = C , thus {x ∈ B : |g(x) − f (x)| > δ } ⊂ C , whence


|µ| {x ∈ B : |g(x) − f (x)| > δ} 6 |µ|(C) 6 δ
and so g − f ∈ T(B, δ) .
IV.85, `. 19–27.
“Definition 10.”
The definition is self-explanatory and needs no notes—but to the reader
familiar with measures as set functions (let us call them ‘set-measures’, re-
serving ‘measures’ for the Bourbaki concept) as in the book of P.R. Halmos
(Measure theory, Springer), the definition invites interpretation; so to speak,
the numerical measures |f |p ·|µ| to be defined shortly are (i) equi-‘absolutely
continuous’ with respect to |µ| and (ii) equi-‘inner regular’.
To get a feeling for the definition, we consider the case that H consists
of a single function. The essence of the definition concerns a positive mea-
sure ( |µ| ) and positive numerical functions ( |f | p ) that are integrable with
respect to the measure; let us simplify the notation by assuming that µ
is a positive measure and H = {f } , where f ∈ L 1 (µ) and f > 0 . The
conditions then read:
(i) for every ε > 0 there exists a δ > 0 such that, for a µ-integrable
set A , Z
µ(A) 6 δ ⇒ ϕA f dµ 6 ε ;

(ii) for every ε > 0 there exists a compact set K ⊂ X such that
Z
ϕX -- K f dµ 6 ε .

Our goal is to establish these properties by interpreting the symbol f ·µ

as a measure on X . We will in fact construct two candidate measures ν and
ρ (both positive and bounded). The precedent of Ch. III, §1, No. 4 ( g · µ ,
g continuous) suggests strongly that the notation f · µ be awarded to ν ,
which has the easiest and most natural definition, but it is ρ that yields
the properties needed for item (i). For item (ii) there is a direct proof not
making use of any new measures (Prop. A below). The two measures will
be shown to be equal if every compact set in X is a G δ (the intersection of
a sequence of open sets), as is the case when X is metrizable (GT, IX, §2,
No. 5, Prop. 7). My guess is that ν = ρ always, but I have not succeeded
in putting the pieces together.
§5 measurable functions and sets INT IV.x171

Proposition A. Condition (ii) always holds.

Proof. Recall that ϕA f is µ-integrable for every µ-measurable set A
(No. 6, Cor. 3 of Th. 5). Writing K for the set of all compact subsets of X ,
the assertion of (ii) is equivalent to
Z Z
sup ϕ K f dµ = f dµ .
K∈K

By §4, No. 4, Cor. of Th. 3, there exists an increasing sequence (g n ) of

µ-integrable functions with compact support such that g n (x) → f (x) for
µ-almost every x . Consider the doubly indexed family of numbers
Z
ϕK gn dµ ;

by the ‘associativity of sups’ we may compute the supremum of the family

in two ways (a special case of GT, IV, §5, No. 4, Prop. 8):
 Z   Z 
(∗) sup sup ϕK gn dµ = sup sup ϕK gn dµ .
K∈K n n K∈K

Now, for each K ∈ K ,

Z Z
sup ϕK gn dµ = ϕK f dµ
n

because ϕK gn ↑ ϕK f µ-almost everywhere (§3, No. 6, Th. 5; or §4, No. 3,

Prop. 4 or Th. 2); whereas for each n ,
Z Z
sup ϕK gn dµ = gn dµ
K∈K

since ϕK gn = gn for K = Supp gn . Thus (∗) yields

Z Z Z
sup ϕK f dµ = sup gn dµ = f dµ ,
K∈K n

whence (ii). {We will see later that this number is ρ(X) .} ♦
Definition of the measure ν = f · µ .
For each h ∈ K = K (X; C) the function hf is µ-measurable (No. 3,
remark following Cor. 5 of Th. 1) and |hf | 6 khk f , whence

N1 (hf ) 6 khk N1 (f ) < +∞ ,

INT IV.x172 extension of a measure. l p spaces §5

thus hf is µ-integrable (No. 6, Th. 5); the formula

Z
ν(h) = hf dµ (h ∈ K )

defines a linear form on K such that h > 0 ⇒ ν(h) > 0 , that is, a positive
measure ν on X (Ch. III, §1, No. 5, Th. 1). We write ν = f · µ , a notation
consistent with Ch. III, §1, No. 4 when f is continuous. Moreover,
Z
|ν(h)| 6 ν(|h|) 6 khk f dµ ,

thus ν is a bounded measure on X (loc. cit., No. 8, remark following Def. 3).
It follows that X is ν-integrable (§4, No. 7, Prop. 12) and that kνk = ν(X) =
sup{ν(K) : K ⊂ X compact } (§4, No. 6, Cor. 1 of Th. 4).
Note that a compact set K in X is a Gδ if and only if there exists a
decreasing sequence (hn ) in K+ (X)T such that hn ↓ ϕK : for, if (Un ) is a
sequence of open sets such that K = Un , let kn ∈ K+ (X) with kn = 1 on K
n
and Supp(kn ) ⊂ Un (III, §1, No. 2, Lemma 1) and let hn = inf(k1 , . . . , kn ) ;
conversely, if hn ↓ ϕK then for each n the set Kn = {x : hn (x) > 1 } =
T
∞ T
∞
{x : hn (x) > 1 − 1/m } is a compact Gδ , and Kn = K .
m=1 n=1
DeferringR any hypothesis on X , we observe that if K is a compact G δ
then ν(K) = ϕK f dµ ; for, ifR(hn ) is a decreasing sequence in K+ (X) such
that hn ↓ ϕK , thenR ν(hn ) ↓ RϕK dν = ν(K) (§4, No. 3, Prop. 4) whereas
hn f ↓ ϕK f implies hn f dµ ↓ ϕK f dµ , thus
Z Z
ν(K) = lim ν(hn ) = lim hn f dµ = ϕK f dµ .
n n

This gives the clue for defining a second measure.

Definition of the measure ρ .
Let Φ be the set of all µ-integrable subsets of X , and define a positive
real-valued function α on Φ by the formula
Z
α(A) = ϕA f dµ ( A µ-integrable).

We propose to use Th. 5 of §4, No. 11 to construct a second measure on X ;

let us verify that its hypotheses are satisfied.
(PCI ): §4, No. 5, Props. 6 and 7.
(PCII ): If K ⊂ U with K compact and U open, then M = K ∈ Φ
satisfies K ⊂ M ⊂ U .
§5 measurable functions and sets INT IV.x173

(PMI ): If M, N are sets in Φ with M ⊂ N , then ϕ M f 6 ϕN f , whence

α(M) 6 α(N) .
(PMII ): For any sets M, N ∈ Φ one has ϕM∪N 6 ϕM + ϕN ,
whence α(M ∪ N) 6 α(M) + α(N) .
(PMIII ): If M, N ∈ Φ are disjoint, then ϕM∪N = ϕM + ϕN , whence
α(M ∪ N) = α(M) + α(N) .
(PMIV ): Given M ∈ Φ and ε > 0 , we seek a compact set K ⊂ M and
an open set U ⊃ M such that, for every N ∈ Φ satisfying K ⊂ N ⊂ U , one
has |α(N) − α(M)| 6 ε .
Since M is µ-integrable, there exist a sequence of compact sets (K n )
and a sequence (Un ) of µ-integrable open sets such that K n ⊂ M ⊂ Un and
µ(Un --- Kn ) → 0 (§4, No. 6, Th. 4). We can suppose that (K n ) is increasing
and (Un ) is decreasing (contemplate K1 ∪ ·S· · ∪ Kn and U1 T ∩ · · · ∩ Un ).
Then (Un --- Kn ) is decreasing. Let A = Kn and B = Un , which
n n
are µ-integrable sets such that Kn ⊂ A ⊂ M ⊂ B ⊂ Un for all n . Since
Un --- Kn ↓ B --- A , we have

µ(B --- A) = lim µ(Un --- Kn ) = 0 ,

n

that is, B --- A is µ-negligible. Then

Z Z
α(Un --- Kn ) = ϕUn -- Kn f dµ ↓ ϕB -- A f dµ = 0 ;

choose n so that α(Un --- Kn ) 6 ε/2 , and set U = Un , K = Kn . For any

N ∈ Φ such that K ⊂ N ⊂ U , we have

0 6 α(N) − α(K) = α(N --- K) 6 α(U --- K) 6 ε/2 ,

and in particular 0 6 α(M) − α(K) 6 ε/2 , whence |α(M) − α(N)| 6 ε .

Thus the hypotheses of §4, Th. 5 are fulfilled: there exists a (unique)
measure ρ on X such that, for every A ∈ Φ , A is ρ-integrable and
ρ(A) = α(A) for all A ∈ Φ , that is,
Z
(1) ρ(A) = ϕA f dµ for every µ-integrable set A ⊂ X ,

R
whence
R ρ(A) 6 f dµ < +∞ for all such A . In particular,
R ρ(K) =
ϕK f dµ for every compact set K ⊂ X , whence ρ(K) 6 f dµ . For every
open set U in X we know that

ρ*(U) = sup{ρ(K) : K ⊂ U , K compact}

INT IV.x174 extension of a measure. l p spaces §5

(§4, No. 6, Cor. 4 of Th. 4). In particular,

Z
ρ*(X) 6 f dµ < +∞ ,

therefore X is ρ-integrable (loc. cit., Prop. 10), that is, ρ is a bounded

measure and
Z
kρk = ρ(X) = sup{ρ(K) : K ⊂ X compact} 6 f dµ

(§4, No. 7, Prop. 12); we therefore have

Z Z
(2) ϕX dρ = ρ(X) = sup{ ϕK f dµ : K ⊂ X compact} .

R
Thus the condition (ii) is clearly equivalent to ρ(X) = f dµ . I do not see
how to derive the condition (ii) from (2); but we established (ii) at the outset
(Prop. A), therefore
Z Z
(3) ϕX dρ = ρ(X) = f dµ .

On the other hand, the condition (i) does follow from the properties
of ρ :
Proposition B. — The condition (i) is satisfied without restrictions on
the locally compact space X .
Proof. Every µ-negligible set A in X is ν-negligible; for, such a set
A is µ-integrable, hence it is ρ-integrable and ρ(A) = 0 by (1). Since ρ
is a finite measure, it follows that, given any ε > 0 , there exists a δ > 0
such thatR if A is a µ-integrable set with µ(A) 6 δ then ρ(A) 6 ε ; since
ρ(A) = ϕA f dµ , this will prove (i).
The argument for set-measures (cf. Halmos, op. cit., p. 125, §30, Th. B)
can be adapted as follows. Assume to the contrary that for some ε > 0 ,
no such δ exists; that is, for every δ > 0 there exists a µ-integrable set A
such that µ(A) 6 δ but ρ(A) > ε . For every positive integer n , choose a
µ-integrable set An such that µ(An ) 6 1/2n and ρ(An ) > ε , and let A be
the set of all x ∈ X that belong to An for infinitely many n , that is,
∞
\ [

A= Ak ,
n=1 k>n
§5 measurable functions and sets INT IV.x175

S
∞
which is a µ-measurable set. Writing B n = Ak , Bn is µ-measurable
k=n
and
∞
X ∞
X 1 1
µ*(Bn ) 6 µ(Ak ) 6 k
= n−1 ,
2 2
k=n k=n
therefore Bn is µ-integrable (No. 6, Cor. 1 of Th. 5); since B n ⊃ A , it
follows that A is µ-integrable and µ(A) 6 µ(B n ) → 0 , whence A is
µ-negligible, so by assumption ρ(A) = 0 . But B n ↓ A and ρ is finite,
therefore ρ(Bn ) → ρ(A) ; since Bn ⊃ An , one has ρ(Bn ) > ε for all n ,
whence ρ(A) > ε > 0 , a contradiction. ♦
Finally,
Proposition C. If every compact set in X is a G δ , then ν = ρ .
Proof. As observed at the end of the discussion of ν = f · µ ,
Z
(4) ν(K) = ϕK f dµ for all compact sets K ⊂ X .

In view of the property of ρ established in (1) above, we have ν(K) = ρ(K)

for every compact set K in X , therefore ν = ρ (§4, No. 10, Cor. 3 of
Prop. 19). ♦
When ν = ρ we gain information on ρ , the definition of ν being
transformed into
Z
(5) ρ(h) = hf dµ for h ∈ K ,

and on ν , the property (1) of ρ being transformed into

Z
(6) ν(A) = ϕA f dµ for all µ-integrable sets A ⊂ X ;

in terms of the notation f · µ = ν = ρ , we thus have

Z Z
(f · µ)(h) = h d(f · µ) = hf dµ ,
Z Z
(f · µ)(A) = ϕA d(f · µ) = ϕA f dµ

for functions h ∈ K and µ-integrable sets A .

Remark. — Since every µ-negligible set is ρ-negligible, it follows that
every µ-measurable set A is ρ-measurable (remarks following No. 1, Def. 1),
hence is ρ-integrable by the boundedness of ρ , but it is conceivable that the
formula of (1) may fail. In particular, we are not authorized to substitute
A = X in formula (1) as a means of obtaining (ii).
For a discussion of the relations with the book of Halmos (op. cit.), see
the next note.
INT IV.x176 extension of a measure. l p spaces §5

IV.85, `. 19–27. Echo.

“Definition 10.”
The materials are at hand for a comparative anatomy of ‘measure’
as in Integration and in the book of Halmos (Measure theory, Springer;
briefly, [MT]).
In brief, measure in the sense of Bourbaki corresponds to the regular
Borel measures of [MT]. As in the preceding note, ‘measures’ in the sense
of [MT] will be called set-measures. As vectorial matters are not treated
in [MT], the discussion concerns only numerical measures, and for simplicity
we restrict attention to positive measures.
To review some terminology of [MT], a set S of subsets of X is
called: a ring ( = clan) if it is closed under finite unions and differences, and
an algebra if, moreover, it is closed under complementation (equivalently,
X ∈ S ); a σ-ring (resp. σ-algebra) is a ring (resp. algebra) that is closed
under countable unions. Thus S is a σ-algebra if and only if it is a tribe
(in other words, a tribe is a clan that is closed under complementation and
countable unions). The set S is said to be hereditary if it contains the
subsets of any of its sets, that is, B ⊂ A ∈ S ⇒ B ∈ S .
Let µ be a positive measure on X , and write M for the set of all
µ-measurable sets A ⊂ X ; M is a tribe (No. 4, Cor. 2 of Th. 2), that is,
a σ-algebra of subsets of X . Let λ = µ* M be the restriction of µ* to M :

λ(A) = µ*(A) (A ∈ M) .

Note that λ(A) < +∞ ⇔ A is µ-integrable (No. 6, Cor. 1 of Th. 5), in

which case λ(A) = µ(A) .
Obviously λ(∅) = 0 and 0 6 λ(A) 6 +∞ for all A ∈ M , so to verify
that λ is a set-measure, it remains only to check that it is countably additive
[MT, §7, p. 30], established at the end of the following list of properties.
monotonicity: A ⊂ B ⇒ λ(A) 6 λ(B) (§1, No. 4, Prop. 16).
finite additivity: A ∩ B = ∅ ⇒ λ(A ∪ B) = λ(A) + λ(B) .
For, if A and B are both µ-integrable, the equality holds by §4, No. 5,
Prop. 6, whereas if one of A, B is not integrable then A∪B is not integrable,
whence both sides are equal to +∞ .
S
∞
continuity from below : If A = An , where (An ) is an increasing
n=1
sequence of sets in M , then A ∈ M and

λ(A) = sup λ(An )

(§1, No. 4, Prop. 17).

§5 measurable functions and sets INT IV.x177

countable subadditivity: If (An ) is any sequence of sets in M and A =

S
∞
An , then A ∈ M and
n=1

∞
X
λ(A) 6 λ(An )
n=1

(§1, No. 4, Prop. 18).

S
∞
countable additivity: If A = An , where (An ) is a sequence of pair-
n=1
wise disjoint sets in M , then A ∈ M and
∞
X
λ(A) = λ(An ) .
n=1

S
n
For, writing Bn = Ak , one has
k=1

n
X ∞
X
λ(A) = sup λ(Bn ) = sup λ(Ak ) = λ(An )
n n
k=1 n=1

by continuity from below and finite additivity.

In particular, λ is a set-measure.
Let us write R for the set of all µ-integrable sets A ⊂ X , that is,

R = {A ∈ M : λ(A) < +∞ } ;

R is a ring (clan) of sets: A, B ∈ R ⇒ A ∪ B, A --- B ∈ R (§4, No. 5,

S
∞
Prop. 7). If A = An , where (An ) is an increasing sequence of sets
n=1
in R , then
A∈R ⇔ sup λ(An ) + ∞ ,
n

in which case λ(A) = sup λ(An ) (§4, No. 5, Prop. 8). The restriction of λ
n
to R is a finite set-measure on the ring R [MT, §5, pp. 30, 31].
Borel sets and H-Borel sets.
Let B be the tribe (σ-algebra) of Borel sets in X , that is, the tribe
generated by the closed subsets of X (GT, IX, §6, No. 3,
 Def. 4). Of course
B ⊂ M (No. 4, Cor. 3 of Th. 2), and the restriction λ B is a set-measure
on B . However, the set of ‘Borel sets’ in the sense of Halmos [MT, §51,
INT IV.x178 extension of a measure. l p spaces §5

p. 219] is the σ-ring generated by the compact subsets of X ; let us denote

this σ-ring by B1 , and call its elements H-Borel sets.
Of course B1 ⊂ B . In order that B1 = B it is necessary and S sufficient
that X be countable at infinity (i.e., σ-compact), that is, X = Kn for
n
some sequence (Kn ) of compact sets (in such a space, every closed set C
is the union of the sequence of compact sets C ∩ K n ), in which case B1
is a tribe ( X ∈ B1 ). Indeed, B1 is a tribe if and only if X is countable
at infinity. {For, every element of B 1 is contained in a countable union of
compact sets, since the hereditary σ-ring [MT, §10, p. 41] generated by a
collection of sets must contain the σ-ring they generate; whence X ∈ B 1 if
and only if X is σ-compact}.

Let us write λ1 = λB1 , a set-measure on the σ-ring of H-Borel subsets
of X . Since λ1 (K) = µ(K) < +∞ for all compact sets K ⊂ X , λ 1 is a
Borel measure in the sense of MT, §52, p. 223. In fact, λ 1 is a regular Borel
measure [MT, §52, p. 224] in the sense that, for every A ∈ B 1 ,

λ(A) = sup{λ(K) : K ⊂ A , K compact}

(inner regularity) and

λ(A) = inf{λ(U) : A ⊂ U , U open and U ∈ B1 }

(outer regularity). It suffices [MT, §52, Th. F, p. 228] to show that if U ∈ B 1

is an open set and U ⊂ K1 for some compact set K1 , then

λ(U) = sup{λ(K) : K ⊂ U , K compact},

and this follows from the µ-integrability of U (§4, No. 6, Cor. 1 of Th. 4).
{Incidentally, if U ⊂ K1 with U open and K1 compact, then necessarily
U ∈ B1 ; indeed, it suffices that U can be covered by a sequence of compact
sets [MT, §51, Th. A, p. 219].}
To summarize:
Proposition A. If µ is a positive measure on the locally compact space X ,
then the restriction of µ* to the σ-ring generated by the compact subsets
of X is a regular Borel measure on X in the sense of Halmos [MT].
Conversely:
Proposition B. If λ is a regular Borel measure in the sense of [MT]
(defined on the σ-ring generated by the compact subsets of X ), then there
exists a unique measure µ on X (in the sense of Integration) such that
λ(K) = µ(K) for every compact set K in X .
§5 measurable functions and sets INT IV.x179

This follows from

 §4, No. 11, Th. 5, with Φ the set of all compact sets
K ⊂ X , and α = λΦ : for, the condition (PMIV ) of Th. 5 follows from the
regularity of λ , and the other conditions are obvious from the properties
of λ .
In effect, the cited Th. 5 corresponds to the theory of regular contents
in the book of Halmos [MT, §54].
IV.85, `. −6 to −4.
“Remark. — Suppose µ is bounded. For every a > 0 , the set of
measurable mappings of X into F such that |f (x)| 6 a almost everywhere
is equi-integrable of order p , and this is true for any p ∈ [1, +∞ [ .”
Let H be the set of all such f ∈ LFp . For each f ∈ H , write ρf for the
bounded measure on X determined by the condition that
Z
(∗) ρf (A) = ϕA |f |p d|µ|

for every µ-integrable set A ⊂ X (see the first note for `. 19-27). Write
M for the tribe of all µ-measurable sets A ⊂ X , and K for the set of all
compact sets K ⊂ X ; since µ is assumed to be bounded, every A ∈ M
is µ-integrable, hence (∗) holds for all A ∈ M . Moreover, since |f | 6 a
µ-almost everywhere for each f ∈ H , one has

ρf (A) 6 ap |µ|(A) 6 ap |µ|(X) = ap sup |µ|(K)

K∈K

for every f ∈ H and A ∈ M .

Given any ε > 0 , choose K ∈ K so that ap |µ|(X --- K) 6 ε and set
δ = ε/ap . Then, for every A ∈ M with |µ|(A) 6 δ one has
Z
|f |p ϕA d|µ| = ρf (A) 6 ap |µ|(A) 6 ap δ = ε ,

and Z
|f |p ϕX -- K d|µ| = ρf (X --- K) 6 ap |µ|(X --- K) 6 ε,

thus H satisfies the conditions (i) and (ii) of Def. 10.

IV.86, `. 2–4.
“Let f , g in H be such that
 1/p
ε
|f (x) − g(x)| 6
|µ|(K)

for x ∈ K , except on a set M of measure 6 δ .”

INT IV.x180 extension of a measure. l p spaces §5

Note first of all that only compact sets K of measure > 0 are in play:
assuming Np (f ) > 0 , for ε sufficiently small the K in property (ii) of Def. 10
will have to be non-negligible.
Where are we going? In the notations of Prop. 20, we have

H ⊂ LFp (X, µ) ⊂ S (X, µ; F) ;

by (iii) of Prop. 20, the uniformity U of S = S (X, µ; F) (the uniformity of

convergence in measure) induces on L Fp = LFp (X, µ) a uniformity coarser
than the uniformity Up of convergence in mean of order p , in other words,
for functions in LFp , convergence in mean of order p implies convergence
in measure. Succinctly, the trace of U on L Fp × LFp is contained in Up :

U ∩ (LFp × LFp ) ⊂ Up ,

whence U ∩ (H × H) ⊂ Up ∩ (H × H) . We are embarked on a proof of the

reverse inclusion.
Thus, given a basic entourage U ∈ Up , we seek an entourage W ∈ U
such that

(∗) W ∩ (H × H) ⊂ U ∩ (H × H) ,

which will show that U ∩ (H × H) ∈ U ∩ (H × H) and so U p ∩ (H × H) ⊂

U ∩ (H × H) .
We can suppose that

U = {(f , g) ∈ LFp × LFp : Np (f − g) 6 [(2p+1 + 1)ε]1/p } .

Let V be the entourage for F defined by

1/p
V = {(y, z) ∈ F × F : |y − z| 6 ε/|µ|(K) },

and form the entourage W ∈ U defined by

W = W(V, K, δ) ,

consisting of all pairs (f , g) ∈ S × S such that

|µ|* {x ∈ K : f (x), g(x) ∈/ V} 6 δ.

Writing, for any pair (f , g) ∈ S × S ,

M(f , g) = {x ∈ K : f (x), g(x) ∈/ V}

1/p
= {x ∈ K : |f (x) − g(x)| > ε/|µ|(K) },
§5 measurable functions and sets INT IV.x181

we thus have


W = {(f , g) ∈ S × S : |µ|* M(f , g) 6 δ } .

To establish (∗), we wish to show that
if (f , g) ∈ W ∩ (H × H) , that is,

if f , g ∈ H are such that |µ|* M(f , g) 6 δ , then (f , g) ∈ U , that is,
Np (f − g) 6 [(2p+1 + 1)ε]1/p .
The statement of `. 2–4 starts the ball rolling by saying: let (f , g) ∈
W(V, K, δ) ∩ (H × H) . The rest of the computation is straightforward.
IV.86, `. 5, 6.
Based on property (ii) in Def. 10.
IV.86, `. 8.
Based on property (i) in Def. 10.
IV.86, `. 11.
Follows from `. 10 because
ε
|f − g|p 6 on K --- M
|µ|(K)

by the definition of M in `. 4.
IV.86, `. −12 to −6.
“Lemma 5.”
One notes that the condition (i) reveals the Lemma as an analog, in lo-
cally compact spaces, of Lebesgue’s criterion for Riemann-integrability (see,
for example, §11.4, pp. 211–214 of my book A first course in real analysis,
Springer). Specialized to the case of a bounded function f : [a, b] → R
and Lebesgue measure µ on [a, b] , the lemma says (assuming Lebesgue’s
criterion) that f is Riemann-integrable if and only if, for every ε > 0 , there
exist continuous functions g, h : [a, b] → R , with h > 0 , such that
Z
|f − g| 6 h 6 2 sup |f | and h dµ 6 ε .

In effect, approximation of f by step functions (say, in the sense of op. cit.,

p. 174, Th. 9.8.2) is replaced by an approximation by continuous functions.
IV.87, `. 1–4.
“Multiplying g1 , . . . , gn by a suitable same element of K (X) , we can
suppose in addition that

|g1 a1 + · · · + gn an | 6 M = sup |f |
INT IV.x182 extension of a measure. l p spaces §5

on X , whence k 6 2M .”
I am indebted to Professor Dixmier for crucial help with the proof of
Lemma 5.
For technical reasons, let us assume the g i and R ai so chosen that,
on writing k = |f − g1 a1 − · · · − gn an | , one has k dµ 6 ε/4 (instead
of ε/2 ). When the gi are replaced by Rsuitable functions g i0 , the function
k 0 = |f − gi0 a1 − · · · − gn0 an | will satisfy k 0 dµ 6 ε/2 .
R Write p = g1 a1 + · · · + gn an . Thus p : X → F , k = |f − p| and
k dµ 6 ε/4 . We seek a continuous function ϕ : X → R , 0 6 ϕ 6 1 , such
that, on defining
gi0 = ϕgi , p0 = ϕp = g10 a1 + · · · + gn0 an , k 0 = |f − p0 | ,
R
one has |p0 | 6 M and k 0 dµ 6 ε/2 . We define

 M if |p(x)| > M
ϕ(x) = |p(x)|

1 if |p(x)| 6 M .
The continuity of ϕ is a special case of the following proposition (with
g = |p| ):
Proposition. Let X be any topological space, g : X → R a continuous
function > 0 , and M a real number > 0 . Let
U = {x ∈ X : g(x) > M }
and define ϕ : X → [0, 1] by

 M if x ∈ U
ϕ(x) = g(x)

1 if x ∈ X --- U .
Then ϕ is continuous.
Proof. By the continuity of g , U is open and g > M on U . On the
other hand, X --- U = {x ∈ X : g(x) 6 M } , therefore
(∗) g(x) = M on U ∩ (X --- U) = U --- U .
(Incidentally, U --- U = U ∩ X --- U = ∂U (the boundary of U .)
Now, U , X --- U is a closed covering of X , so it will suffice to show
that the restrictions ϕU and ϕX --- U are continuous (GT, I, §3, No. 2,
Prop. 4). Indeed, ϕX --- U is the constant function 1 . On the other hand,
M
ϕ(x) = on U = U ∪ (U --- U) ;
g(x)
§5 measurable functions and sets INT IV.x183

for, if x ∈ U then ϕ(x) = M/g(x) by definition, whereas if x ∈ U --- U =

U∩(X --- U) then g(x) = M (by (∗)) and ϕ(x) = 1 (by definition), so again
ϕ(x) = M/g(x) . ♦
Returning to the proof of Lemma 5, with U = {x ∈ X : |p(x)| > M } ,
gi0 = ϕgi and p0 = ϕp , we have

M
0
· |p(x)| = M if x ∈ U
|p (x)| = ϕ(x)|p(x)| = |p(x)|

1 · |p(x)| 6 M if x ∈ X --- U ,

thus |p0 (x)| 6 M on X .

claim: |f − p0 | 6 2|f − p| .
For, if x ∈ X --- U , that is, if |p(x)| 6 M , then ϕ(x) = 1 and so
0
p (x) = p(x) , whence

|f (x) − p0 (x)| = |f (x) − p(x)| 6 2|f (x) − p(x)| .

On the other hand, if x ∈ U , that is, if |p(x)| > M , then

M M
ϕ(x) = < 1 and p0 (x) = ϕ(x)p(x) = · p(x) ,
|p(x)| |p(x)|

thus  
 M 
|p(x) − p0 (x)| = p(x) − p(x)
|p(x)|
 M
 M

 
= 1− p(x) = 1 − |p(x)|
|p(x)| |p(x)|
= |p(x)| − M = |[p(x) − f (x)] + f (x)| − M
6 |p(x) − f (x)| + |f (x)| − M
6 |p(x) − f (x)| + M − M = |p(x) − f (x)| ,
therefore

|f (x) − p0 (x)| 6 |f (x) − p(x)| + |p(x) − p0 (x)|

6 |f (x) − p(x)| + |p(x) − f (x)| = 2|f (x) − p(x)| ,

and the claimed inequality is established.

{For the purposes of Lemma 5, the continuous function ϕ is adequate;
but the assertion promises a ϕ with compact support. This is easily ar-
ranged: let g ∈ K (X) be such that 0 6 g 6 1 and g = 1 on the
union of the supports of the gi ; then ggi = gi so (ϕg)gi = ϕgi = gi0
and (ϕg)p = ϕp = p0 , thus, replacing ϕ by ϕg ∈ K (X) , the functions g i0
INT IV.x184 extension of a measure. l p spaces §5

and p0 are unchanged. In particular, the inequality |f − p 0 | 6 2|f − p| is

invariant under the modification of ϕ (no need to reconsider its proof with
a modified ϕ ). From this point on, the argument makes no reference to ϕ ,
hence is valid whether or not ϕ was modified.}
Writing k 0 = |f − p0 | , we thus have k 0 6 2|f − p| = 2k , and so
Z Z
0
k dµ 6 2 k dµ 6 2(ε/4) = ε/2 ;

and of course |p0 | 6 M .

Thus, replacing the notations gi0 , p0 , k 0 by gi , p, k , we can suppose that
|p| 6 M ; moreover, k 6 2M since

k(x) = |f (x) − p(x)| 6 |f (x)| + |p(x)| 6 M + M

for all x ∈ X .
IV.87, `. 6.
“Then 2M > l > k on X , and l = k on X --- N0 ”
We know that k : X → [0, 2M] . By definition (GT, IV, §5, No. 6)
 
l(x) = inf sup k(y) ,
V y∈V

where V runs over the filter of neighborhoods of x (or any base thereof),
therefore l : X → [0, 2M] , thus l 6 2M ; and l(x) > k(x) because x ∈ V
for all V .
If x ∈ X --- N0 then k is continuous at x , hence is upper semi-continuous
at x (GT, IV, §6, No. 2, comment following Def. 1), hence

k(x) = lim sup k(y) = l(x)

y→x

(loc. cit., ‘dual’ of Prop. 3). The function l is upper semi-continuous (loc.
cit., the ‘dual’ of Prop. 4), and is called the upper semi-continuous regular-
ization of k .
IV.87, `. 7. R
“ . . . therefore l dµ 6 ε/2 .”
Since l = k almost everywhere, so that µ*(l) = µ*(k) < +∞ , it follows
1 1
that l ∈ FR R(§3, No.R3), hence l ∈ LR (§3, No. 4, third paragraph following
Def. 2) and l dµ = k dµ 6 ε/2 .
§5 measurable functions and sets INT IV.x185

IV.87, `. 7–9.
“ l is bounded and upper semi-continuous, hence is the lower envelope
of the set of bounded continuous functions > l .”
Of course 0 6 l 6 2M , and the upper semi-continuity of l was observed
in the note before the last. Then 2M − l is > 0 and lower semi-continuous,
that is, in the notation of §1, No. 1, 2M − l ∈ I + , therefore 2M − l is the
upper envelope of the functions g ∈ K + such that g 6 2M − l (loc. cit.,
Lemma); consequently l − 2M is the lower envelope of the functions −g ,
and so
l = (l − 2M) + 2M
is the lower envelope of the functions −g + 2M = 2M − g , which are indeed
bounded and continuous. Write

H = {h ∈ C (X) : h is bounded and l 6 h } .

Of course l 6 inf H , where (inf H)(x) = inf {h(x) : h ∈ H } ; but we have

just seen that the functions 2M − g described above form a subset H 0 of H
such that l = inf H0 , therefore

l 6 inf H 6 inf H0 = l ,

whence l = inf H as claimed.

IV.87, `. 9–11.
“Therefore there exists
R a bounded
R continuous function h > l on X
such that h 6 2M and h dµ 6 l dµ+ε/2 (§4, No. 4, Cor. 2 of Prop. 5).”
Maintaining the notations of the preceding note, since the set of func-
tions g ∈ K+ (X) with g 6 2M − l is directed upward, with upper envelope
2M − l , the set H0 of functions 2M − g is directed downward, with lower
envelope l , therefore Z Z
l dµ = inf
0 0
h0 dµ
h ∈H

by the cited Cor. 2. Since the h ∈ H0 satisfy 0 6 l 6 h0 6 2M , therefore

0

H0 ⊂ H , it follows from the preceding formula that there exists an h ∈ H

such that h 6 2M and
Z Z Z
h dµ 6 l dµ + ε/2 = k dµ + ε/2 6 ε/2 + ε/2 ;

moreover,

|f − g1 a1 − · · · − gn an | = |f − p| = k 6 l 6 h 6 2M ,

thus the proof of (i) ⇒ (ii) is complete.

INT IV.x186 extension of a measure. l p spaces §5

IV.87, `. 14–15.
“For every x ∈ X , ω(x) is the oscillation of f − g 1 a1 − · · · − gn an
at x , therefore ω(x) 6 2h(x) .”
Let us write ωf (x) = ω(x) for the oscillation of f at x , that is (GT,
IX, §2, No. 3),

ωf (x) = inf δ f (V) ,
V

where V runs over the filter of neighborhoods of x , and

δ f (V) = sup |f (y) − f (y 0 )|
y,y 0 ∈V

is the diameter of f (V) . Note that V 7→ δ f (V) is a decreasing function
of V , and, since |f | 6 M , one has 0 6 ωf 6 2M .
Why are we considering ω ?; because f is continuous at x if and only if
ω(x) = 0 (GT, loc. cit., comment following the proof of Prop. 4), therefore

{x ∈ X : ω(x) > 0 } = N

(the set of discontinuities of f ). Our objective is to show that µ*(N) = 0 ,

that is, ω(x) = 0 for almostR every x ; since ω > 0 , it will suffice to show
that ω is integrable and ω dµ = 0 (§2, No. 3, Th. 1).
Thus, given any ε > 0 (the ai , gi andR h being chosen as in (ii)), it will
suffice to show that ω is integrable and ω dµ 6 2ε . The proof exploits
the invariance of ωf under perturbation of f by a continuous function:
Proposition. (1) For any functions f , g : X → F , one has ω f +g 6
ωf + ω g .
(2) If, moreover, g is continuous, then ω f +g = ωf .
(3) ωf is upper semi-continuous.
Proof. (1) Let V be a neighborhood of x ∈ X . For all y, y 0 ∈ V one
has
 
|(f + g)(y) − (f + g)(y 0 )| = [f (y) − f (y 0 )] + [g(y) − g(y 0 )]



6 |f (y) − f (y 0 )| + |g(y) − g(y 0 )| 6 δ f (V) + δ g(V) ,

whence




(∗) δ (f + g)(V) 6 δ f (V) + δ g(V) .

We are to show that ωf +g (x) 6 ωf (x) + ωg (x) . This is trivial if either of

ωf (x) or ωg (x) is infinite; if both are finite, then (∗) shows that ‘ultimately’
(with respect to the neighborhood filter at x ) all terms of (∗) are finite,
§5 measurable functions and sets INT IV.x187

and since they are decreasing functions of V , passage to the limit yields the
desired inequality.
(2) If, moreover, g is continuous, then ω g = 0 and so ωf +g 6 ωf ;
similarly ωf = ω(f +g)+(−g) 6 ωf +g .
(3) See GT, loc. cit., Prop. 4. ♦
Returning to the proof of (ii) ⇒ (i), writing g = −g 1 a1 − · · · − gn an
and w = f + g , we have |w| 6 h . It follows that ω w 6 2h . For, if V is a
neighborhood of x ∈ X then, for all y, y 0 ∈ V one has

|w(y) − w(y 0 )| 6 |w(y)| + |w(y 0 )| 6 h(y) + h(y 0 ) ;

but


h(y) = [h(y) − h(x)] + h(x) 6 |h(y) − h(x)| + h(x) 6 δ h(V) + h(x)


and similarly h(y 0 ) 6 δ h(V) + h(x) . Thus

 

|w(y) − w(y 0 )| 6 δ h(V) + h(x) + δ h(V) + h(x)

for all y, y 0 ∈ V , therefore

δ w(V) 6 2 δ h(V) + 2 h(x) ;

since h is continuous at x , passing to the limit with respect to the neigh-

borhood filter at x , one has

ωw (x) 6 2 · ωh (x) + 2 h(x) = 2 · 0 + 2 h(x) ,

thus ωw 6 2h .
Finally, ωf = ωf +g = ωw 6 2h ; since ωf is upper semi-continuous,
hence measurable (No. 5, Cor. of Prop. 8), and h is integrable, it follows
that ωf is integrable and
Z Z
ωf dµ 6 2 h dµ 6 2ε .

This completes the proof of Lemma 5.

IV.87, `. −3.
“ . . . we can suppose in addition that h 0 > h .”
In slow motion: If K0 is a compact neighborhood of K (GT, I, §9, No. 7,
Prop. 10) and V is the interior of K0 , there exists a function f ∈ K (X) ,
with Supp f ⊂ V ⊂ K0 , such that 0 6 f 6 1 and f = 1 on K (Ch. III, §1,
INT IV.x188 extension of a measure. l p spaces §5

No. 2, Lemma 1); in particular, f = 0 on X --- K 0 . The continuous function

k = 2M(1 − f ) satisfies

0 6 k 6 2M , k = 0 on K , k = 2M on X --- K0 .

Setting h0 = sup(h, k) , it follows that

0 6 h 6 h0 6 2M , 0 6 h0 − h 6 2M − h 6 2M ,

h0 = h on K , and h0 = 2M on X --- K0 .
IV.87, `. −3,
R −2.
“Then (h0 − h) dµ 6 2Mµ*(X --- K) 6 2Mε .”
With notations as in the preceding note,

0 6 h0 − h = (h0 − h)ϕX -- K 6 2M · ϕX -- K ,
R
whence 0 6 (h0 − h) dµ 6 2Mµ(X --- K) 6 2Mε .
IV.87, `. −2, −1.
“ . . . h0 = h1 + 2M , where h1 ∈ K (X) .”
With the foregoing notations, set h 1 = h0 − 2M ; since h0 − 2M = 0 on
X --- K0 , one has h1 ∈ K (X) (and h1 6 0 ).
IV.88,R`. 1, 2. R R R
“ h0 dν = h1 dν +2Mkνk tends to h1 dµ+2Mkµk = h0 dµ with
respect to B .”
R R
h1 dν → h1 dµ because h1 ∈ K (X) and ν → µ vaguely, whereas
kνk → kµk by hypothesis, and
Z Z Z Z
0
h dν = h1 dν + 2M 1 dν = h1 dν + 2Mkνk

by §4, No. 7, Prop. 12.

IV.88, `. 2–4.
“There then exists an A ∈ B such that, for every ν ∈ A ,
Z Z 
 
 (g1 a1 + · · · + gn an ) dν − (g1 a1 + · · · + gn an ) dµ 6 ε ,
Z Z Z Z
0 0
h dν 6 h dν 6 h dµ + ε 6 h dµ + 2Mε + ε 6 2(M + 1)ε .”
§5 measurable functions and sets INT IV.x189

By Ch. III, §3, No. 1, Example 1 following Def. 1,

Z Z 
 
 (g1 a1 + · · · + gn an ) dν − (g1 a1 + · · · + gn an ) dµ
 Z Z  Z Z  
 
= g1 dν − g1 dµ a1 + · · · + gn dν − gn dµ an  ;

since ν → µ vaguely, the coefficients of the a i tend to 0 , hence there exists

an A1 ∈ B such that the inequality in the first row holds for every ν ∈ A 1 .
In the Rsecond rowR of the display, the crucial inequality is the second
one: since h0 dν → h0 dµ , there exists an A2 ∈ B such that for every
ν ∈ A2 one has Z Z
h0 dν 6 h0 dµ + ε .
R 0
RThe third inequality restates (h − h) dµ 6 2Mε , and the fourth restates
h dµ 6 ε .
Finally, choose A ∈ B such that A ⊂ A1 ∩ A2 .
IV.88, `. 5–8.
“These inequalities imply
Z Z 
 
 f dν − f dµ 6
Z Z Z  Z
 
h dν +  (g1 a1 + · · · +gn an ) dν − (g1 a1 + · · · + gn an ) dµ + h dµ

6 2(M + 2)ε ”

Writing p = g1 a1 + · · · + gn an ∈ K (X; F) , one has

Z Z Z Z Z Z
f dν − f dµ = (f − p) dν + p dν − p dµ − (f − p) dµ ,

R R 
 
where |f −p| 6 h and  p dν − p dµ 6 ε by the preceding note, therefore

Z Z  Z Z
 
 f dν − f dµ 6 h dν + ε + h dµ 6 2(M + 1)ε + ε + ε .

IV.88, `. 10, 11.

“Remark.”
By §4, No. 7, Cor. of Prop. 13, f is integrable with respect to every
bounded measure.
INT IV.x190 extension of a measure. l p spaces §6

IV.89, `. 3.
“ µn (A) tends to µ(A) .”
One proposes to apply Prop. 22 to the function f = ϕ A . The space X
is the disjoint union of the interior, exterior and boundary (‘frontier’) of A ,
◦
X = A ∪ ( { A)◦ ∪ ∂A
◦
(GT, I, §1, No. 6). Since f is constant on each of the open sets A and
( { A)◦ , it is continuous at those points, whereas ∂A = A ∩ { A shows that
it is discontinuous at the points of ∂A , therefore the set of discontinuities
of f is precisely ∂A , which is by assumption R negligible.
R Thus f satisfies
the conditions of Prop. 22, whence µ n (A) = f dµn → f dµ = µ(A) .
IV.89, `. 4, 5.
“ if pn denotes the number of integers k ∈ [0, n] such that e 2iπkθ ∈ A ,
then n−1 pn tends to µ(A ) as n tends to +∞ .”
For every integer k > 0 ,
Z (
1 if e2iπkθ ∈ A
νk (A) = ϕA dνk =
0 otherwise,

therefore
1
pn
µn (A) = ν0 (A) + · · · + νn (A) = ,
n+1 n+1
thus
pn n+1 pn n+1
= · = · µn (A) → µ(A) .
n n n+1 n
A stunning application of Prop. 22.

§6. CONVEXITY INEQUALITIES

IV.89, `. −4 to −1.
“Corollary.”
A variation on the argument shifts the burden from the measure to the
function:
Proposition. Let µ be a positive measure, f : X → F an integrable
function that is equal to zero outside some integrable
R subset A of X . If C
is the closed convex envelope of f (X) , then f dµ ∈ µ(A) · C .
§6 convexity inequalities INT IV.x191
R
Proof. If µ(A) = 0 then f is negligible and f dµ = 0 ∈ 0 · C .
If µ(A) > 0 then g = ϕA meets the requirements of Th. 1, f g = f , and
R R
f dµ f g dµ
= R ∈ C. ♦
µ(A) g dµ

The Corollary is a special case: If µ is a bounded positive measure and

f :X→ R F is any integrable function, then X qualifies to play the role of A ,
thus f dµ ∈ µ(X) · C .
For generalizations with F a Hausdorff locally convex space, see Ch. III,
§3, No. 2, Prop. 4, and, below, §7, No. 1, Cor. of Prop. 1.
IV.90, `. 13, 14.
“For every α > M∞ (f ) , the set of x ∈ X such that f (x) > α is
locally negligible ”
By the definition of M∞ (f ) as the infimum of all “locally almost every-
where majorants” of f , there exists a number β such that M ∞ (f ) 6 β < α
and f (x) 6 β locally almost everywhere; the set {x : f (x) > β } is thus
locally negligible, therefore so is its subset {x : f (x) > α } .
IV.90, `. 18.
“ m∞ (f ) 6 M∞ (f ) if the measure µ is nonzero ”
It suffices that m∞ (f ) 6 f (x) 6 M∞ (f ) for some point x ∈ X . We
know that there exist locally negligible sets A and B such that f 6 M ∞ (f )
on X --- A and f > m∞ (f ) on X --- B ; if no such x exists, then

(X --- A) ∩ (X --- B) = ∅ ,

so X = A ∪ B is locally negligible. Then K = K ∩ X is negligible for all

compact sets K (§5, No. 2, Prop. 5), therefore |µ|*(X) = 0 (§4, No. 6,
Cor. 4 of Th. 4), and so kµk = 0 (p. IV.4, `. −7, −6).
IV.90, `. −8.
“ . . . provided the second member is defined ”
That is, M∞ (f ) and M∞ (g) are not infinite of opposite signs. For
locally almost every x , f (x) + g(x) is defined, f (x) 6 M ∞ (f ) and g(x) 6
M∞ (g) , and the desired inequality

f (x) + g(x) 6 M∞ (f ) + M∞ (g)

follows from a property of addition in R (GT, IV, §4, No. 3): if α 6 β ,

γ 6 δ and α + γ , β + δ are both defined, then α + γ 6 β + δ .
{If both members of the inequality are finite, then all numbers in sight
are finite, and one is reduced to a property of addition in R . If α+γ = −∞
INT IV.x192 extension of a measure. l p spaces §6

or β + δ = +∞ , the inequality is trivial. If α + γ = +∞ then one of α , γ is

equal to +∞ , whence one of β , δ is equal to +∞ , therefore β + δ = +∞ .
One argues similarly if β + δ = −∞ .}
IV.90, `. −4.
“ . . . provided the second member is defined.”
For multiplication in R , only the four products 0 · (±∞) and (±∞) · 0
are forbidden (GT, IV, §4, No. 3).
However, if one makes the convention 0 · (+∞) = (+∞) · 0 = 0 , then
all products in R+ are admitted and

α 6 β & γ 6 δ ⇒ αγ 6 βδ .

For, if one of α, γ is 0 then αγ = 0 6 βδ . Suppose α > 0 and γ > 0 : if

one of β, δ is infinite then αγ 6 +∞ = βδ , whereas if they are both finite,
one is reduced to a property of multiplication in R + .
With this convention, the formula (2) is then valid: for locally almost
every x ,
0 6 f (x) 6 M∞ (f ) and 0 6 g(x) 6 M∞ (g) ,
whence f (x)g(x) 6 M∞ (f ) M∞ (g) .
IV.91, `. 3–10.
“Proposition 1 (Inequality of the mean).”
As the proof entails arithmetical operations on functions not necessarily
everywhere defined, nor necessarily finite where they are defined, it is helpful
to review the definitions and make some preliminary simplifications.
To say that g is integrable means that there exists a function h ∈ L R1
such that g = h almost everywhere (§4, No. 1, last sentence). Let D be
the set of x ∈ X such that g(x) is defined, finite and > 0 , and define
g 0 : X → R to be equal to g at the points of D , and to 0 at the points
of X --- D . Then X --- D is negligible, so g 0 = g almost everywhere; and
g 0 = h almost everywhere, so g 0 is integrable (§4, loc. cit.). The sets
{x : g 0 (x) 6= 0 } and {x : g(x) 6= 0 } differ at most by a negligible set;
replacing g by g 0 , we can suppose that g is everywhere defined, finite
and > 0 : none of the conclusions of Prop. 1 is thereby affected.
To say that f is bounded in measure implies that there exists a neg-
ligible set N such that, for every x ∈ X --- N , f (x) is defined and finite.
In particular, f need not be defined on all of X . Let A be the domain
of f ; thus X --- N ⊂ A ⊂ X . Then the set X --- A ⊂ N is negligible, hence
measurable, therefore A is measurable; to say that f is measurable means
that the extension f 0 of f to X by 0 is measurable
 (§5, No. 10, Def. 8).
Let f 00 : X → R be the extension by 0 of f X --- N to X , so that f 00 = f 0
§6 convexity inequalities INT IV.x193

except possibly on the negligible set A --- (X --- N) = A ∩ N ; thus f 00 = f 0

almost everywhere and so f 00 is measurable (§5, No. 2, Prop. 6) and every-
where finite-valued. Moreover, f 00 = f almost everywhere; it follows that
f 00 is also bounded in measure, with same M ∞ and m∞ as f . Moreover,
f 00 g = f g almost everywhere. To say that f g is integrable means that there
exists a k ∈ LR1 such that f g = k almost everywhere; then f 00 g = k almost
everywhere, hence f 00 g is also integrable. Replacing f by f 00 alters none
of the conclusions of Prop. 1.
To summarize: we can suppose that g : X → R + and that f : X → R .
IV.91, `. 13–15.
“ . . . because the set of points x ∈ X where g(x) 6= 0 is a countable
union of integrable sets (§5, No. 6, Lemma 1).”
And a locally negligible integrable set is negligible (§5, No. 2, Cor. 1 of
Prop. 5).
IV.91, `. 15.
“It follows that f g is integrable (§5, No. 6, Th. 5) ”
For, f g is measurable (§5, No. 3, remark following Cor. 5) and, writing
c = max{|m∞ (f )|, |M∞ (f )|} , one has −cg 6 f g 6 cg almost everywhere;
thus |f g| 6 cg almost everywhere, whence N 1 (f g) 6 c · N1 (g) < +∞ .
IV.91, `. −9 to −5.
“Definition 2.”
Note that |f | is defined and finite everywere on X , thus, for a function
f : X → R , the terminology here is consistent with that of No. 2.

IV.91, `. −4, −3.

“A function f in LF∞ may thus be characterized by the fact that there
exists a bounded measurable function equal locally almost everywhere to f .”
Suppose f ∈ LF∞ . In particular, M∞ (|f |) < +∞ , and the set

N = {x ∈ X : |f (x)| > M∞ (|f |) }

is locally negligible. Define g : X → F by g = ϕ { N f . Since g = f

locally almost everywhere, g is measurable (§5, No. 2, Prop. 6). Moreover,
|g| 6 M∞ (|f |) everywhere on X , thus g is bounded.
Conversely, suppose g : X → F is a bounded measurable function
such that f = g locally almost everywhere; say |g(x)| 6 M < +∞ for all
x ∈ X . Then f is measurable (§5, No. 2, Prop. 6) and N ∞ (f ) = M∞ (|f |) =
M∞ (|g|) 6 M < +∞ , therefore f ∈ LF∞ .
INT IV.x194 extension of a measure. l p spaces §6

IV.92, `. 5–7.
“ . . . for every integer m , there exist a locally negligible set H m and an
integer n0 such that |f (x) − fn (x)| 6 1/m for every integer n > n0 and
every x ∈ / Hm ”
For each m , let n0 (m) be an index such that

n > n0 (m) ⇒ N∞ (f − fn ) 6 1/m ;

thus, for each n > n0 (m) , there exists a locally negligible set H mn such
that

|f (x) − fn (x)| 6 M∞ (|f − fn |) 6 1/m for all x ∈ X --- Hmn .

S
The set Hm = Hmn is then locally negligible and
n>n0 (m)

n > n0 (m) ⇒ |f (x) − fn (x)| 6 1/m for all x ∈ X --- Hm .

IV.92, `. 10.
“ . . . the converse is immediate.”
That is, if there exists a locally negligible set H such that f n → f
uniformly on X --- H , then N∞ (fn − f ) → 0 . For,

N∞ (fn − f ) = M∞ (|fn − f |) 6 sup |fn (x) − f (x)| → 0 .

x∈X -- H

IV.92, `. 16, 17.

“ . . . its topology is defined by the norm deduced from N ∞ by passage
to the quotient ”
TVS, II, §1, No. 3 (p. TVS II.5, `. 5–6).
IV.92, `. −9, −8.
“ . . . g is bounded on the set of x ∈ X where |g k1 (x)| 6 N∞ (gk1 ) ”
Since g = 0 on A , it is the points of X --- A in which we are interested.
One has \
X --- A ⊂ X --- A1 = (X --- Ars ) .
r,s>k1

By definition, gn = fn on X --- A for all n , so for all n > k1 one has

gn − gk1 = fn − fk1 on X --- A ⊂ X --- Ank1 ;

therefore, by the definition of Ank1 ,

|gn (x) − gk1 (x)| = |fn (x) − fk1 (x)| 6 1 on X --- A

for all n > k1 .

§6 convexity inequalities INT IV.x195

Suppose now that x ∈ X --- A and |gk1 (x)| 6 N∞ (gk1 ) . Then, for all
n > k1 , one has

|g(x)| 6 |g(x) − gn (x)| + |gn (x) − gk1 (x)| + |gk1 (x)|

6 |g(x) − gn (x)| + 1 + N∞ (gk1 ) ,

and passage to the limit yields |g(x)| 6 0+1+N ∞ (gk1 ) . Thus g is bounded
by 1 + N∞ (gk1 ) on the indicated set.
IV.92, `. −7.
“ . . . g belongs to LF∞ . ”
The set N = {x ∈ X : |gk1 (x)| > N∞ (gk1 ) } is locally negligible, so
g = ϕ { N g locally almost everywhere, and the measurable function ϕ { N g
is bounded by the preceding note.
IV.93, `. 2–4.
“ . . . if there exists a continuous function f with negligible compact
support and not identically zero ”
Such a function exists if and only if Supp µ 6= X :
If : Suppose Supp µ 6= X . Then U = X --- Supp µ is a nonempty
negligible open set (§2, No. 2, Prop. 5). Choose a nonzero function f ∈
K+ (X) such that Supp f ⊂ U , and a nonzero vector a ∈ F , and define
f = f a ; then f ∈ K (X; F) , f 6= 0 , and the set Supp f = Supp f ⊂ U is
negligible. Moreover, |f | = |f | · |a| 6 kf k · |a| < +∞ , thus f ∈ C b (X; F)
(Ch. III, §1, No. 2).
Only if : Suppose f ∈ K (X; F) with f 6= 0 and Supp f negligible.
Then the open set V = {x : f (x) 6= 0 } ⊂ Supp f is nonempty and negligible,
hence V ⊂ X --- Supp µ by the cited Prop. 5, thus Supp µ 6= X . ♦
For such a function f , one has |f | = 0 almost everywhere, hence locally
almost everywhere, therefore M∞ (|f |) = 0 ; thus N∞ (f ) = 0 < kf k . This
proves the implication

N∞ (f ) = kf k for all f ∈ C b (X; F) ⇒ Supp µ = X .

IV.93, `. 7.
“ . . . which shows that N∞ (f ) = kf k . ”
We are assuming that Supp µ = X and f ∈ C b (X; F) . At any rate,
N∞ (f ) 6 kf k . Supposing to the contrary that M ∞ (|f |) = N∞ (f ) < kf k ,
choose α so that M∞ (|f |) < α < kf k and let U = {x : |f (x)| > α } ,
a nonempty open set. We know from the definition of M ∞ that |f (x)| 6 α
locally almost everywhere, therefore U is locally negligible. Let V be a
nonempty open set with V compact and V ⊂ U ; then V = V ∩ U is
INT IV.x196 extension of a measure. l p spaces §6

negligible (§5, No. 2, Prop. 5), thus V is a nonempty negligible open set,
contradicting X --- Supp µ = ∅ (§2, No. 2, Prop. 5).
IV.93, `. 11–13.
“ . . . but its canonical image in L∞ ∞
F is a closed subspace of LF (which
can moreover be identified with C b (X; F) in the case contemplated).”
For any measure µ (with no asssumption about its support) every f ∈
C b (X; F) belongs to LF∞ and N∞ (f ) 6 kf k , thus the canonical injection
C b (X; F) → LF∞ is continuous for the norm topology on C b (X; F) and the
topology on LF∞ defined by the semi-norm N∞ . In turn, the quotient
mapping LF∞ → L∞ ∞
F defined by f 7→ ḟ is continuous, where LF is a
Banach space equipped with the norm k ḟ k∞ = N∞ (f ) (also written N∞ (ḟ ) )
(Prop. 2). The composite mapping C b (X; F) → L∞ F , defined by f 7→ ḟ ,
satisfies kḟ k∞ 6 kf k hence is continuous but in general not injective. Its
image in L∞ b
F may not be complete: if fn ∈ C (X; F) is a sequence such that
∞
(ḟn ) is Cauchy in LF , there exists an f ∈ LF∞ with kḟn − ḟ k∞ → 0 , but
it is not assured that there exists a g ∈ C b (X; F) such that f = g locally
almost everywhere.
However, when Supp µ = X (the case contemplated), one has kf k =
N∞ (f ) = kḟ k∞ , therefore the image of C b (X; F) in L∞ F is a complete,
∞
hence closed, linear subspace of LF . Thus, the mapping C b (X; F) → L∞ F
is a linear injection such that the norm (and topology) induced by L ∞ F on the
image coincide with those of C b (X; F) , whence the proposed identification
C b (X; F) ⊂ L∞ F .
IV.93, `. 15–17.
“This implies that the space K (X; F) of mappings of X into F , con-
tinuous with compact support, is in general not dense in L ∞ F ”
If the space K (X; F) ⊂ C b (X; F) were dense in L∞F then, since C b (X; F)
is closed in L∞ b ∞
F , it would follow that C (X; F) = LF .

IV.93, `. 19, 20.

“ . . . the topology defined by the semi-norm N ∞ is finer than the topol-
ogy induced on LF∞ by the topology of convergence in measure ”
Let S = S (X, µ; F) be the set of all µ-measurable functions f : X → F,
equipped with the uniform structure of convergence in measure (§5, No. 11).
Then LF∞ ⊂ S ; if W is a basic entourage for S , then the set
W = (LF∞ × LF∞ ) ∩ W
is a basic entourge for the induced uniformity on L F∞ . It will suffice to show
that W is also an entourage for the uniformity on L F∞ defined by the semi-
norm N∞ ; it is enough to exhibit an entourage V for the N ∞ -uniformity
such that V ⊂ W .
§6 convexity inequalities INT IV.x197

We can suppose that

W = W(V, K, δ) ,

where K is a compact subset of X , δ > 0 , and, for some ε > 0 , V is the

entourage of F given by

V = {(a, b) ∈ F × F : |a − b| 6 ε }

(§5, No. 11, paragraph preceding Def. 9). Recall that



/ V} 6 δ
W(V, K, δ) = (f , g) ∈ S × S : |µ|* {x ∈ K : (f (x), g(x)) ∈


= (f , g) ∈ S × S : |µ|* {x ∈ K : |f (x) − g(x)| > ε} 6 δ ,

therefore


W = (f , g) ∈ LF∞ × LF∞ : |µ|* {x ∈ K : |f (x) − g(x)| > ε} 6 δ .

Let
V = {(f , g) ∈ LF∞ × LF∞ : N∞ (f − g) < ε }
(the same ε ); it suffices to show that V ⊂ W , that is, for f , g ∈ L F∞ ,


N∞ (f − g) < ε ⇒ |µ|* {x ∈ K : |f (x) − g(x)| > ε} 6 δ .

Suppose N∞ (f − g) < ε , that is, M∞ (|f − g|) <ε ; then |f − g| 6 ε

locally almost everywhere, that is, the set

A = {x ∈ X : |f (x) − g(x)| > ε}

is locally negligible (see the note for IV.90, `. 13, 14), therefore the set

K ∩ A = {x ∈ K : |f (x) − g(x)| > ε}

is negligible (§5, No. 2, Prop. 5) and so |µ|*(K ∩ A) = 0 < δ .

IV.94, `. 3–7.
“Corollary 1.”
The hypothesis of bilinearity can be omitted.
IV.94, `. 8.
“ . . . Φ(f , g) is measurable (§5, No. 3, Cor. 5 of Th. 1) ”
As in the cited Cor. 5, the hypothesis of bilinearity can be omitted; it is
in the criteria for continuity that bilinearity plays a role (GT, IX, §3, No. 5,
INT IV.x198 extension of a measure. l p spaces §6

Th. 1 or TVS, I, §1, No. 6, Prop. 5). This observation is useful for the proof
of Cor. 3 given below.
IV.94, `. −13.
“For, |hz, z0 i| 6 |z| · |z0 | . ”
The inequality, noted in the proof of TVS, IV, §1, No. 3, Prop. 8, assures
the continuity of the bilinear form (z, z 0 ) 7→ hz, z0 i (GT, IX, §3, No. 5, Th. 1).
For the definition of |z0 | , see GT, X, §3, No. 2, discussion preceding
Prop. 6; TVS, IV, §2, No. 4; or the final section of TVS (“Summary of some
important properties of Banach spaces”, TVS p. 355).
IV.94, `. −9 to −5.
“Corollary 3. — Let µ be a positive measure on X , F a real (resp.
complex ) Hilbert space. On the space L 2F , the symmetric (resp. Hermitian)
form Z
e e) 7→ hf , gi dµ
(f , g

defines a Hilbert space structure, for which the norm is equal to ke

f k2 . ”
A minor technical point: one could suppose that µ is an arbitrary
measure, provided the integral is taken with respect to |µ| , indispensable
for the form to be positive; for, LF2 (X, µ) = LF2 (X, |µ|) (§3, No. 4).
In the complex case (the real case being an obvious simplification) the
form Φ : F × F → C defined by Φ(a, b) = ha, bi (the scalar product,
or ‘inner product’ of a and b ) is sesquilinear ( hca, dbi = cdha, bi ), so
a slight adjustment is needed to make Cor. 1 applicable. Let F be the
Hilbert space conjugate to F (TVS, V, §1, No. 3,): F is the set F equipped
with its original additive structure; if scalar multiples in F are denoted ca
( c ∈ C , a ∈ F ), then scalar multiples c · a in F are defined by c · a =
ca ; and the Hermitian form on F is the conjugate of that on F (we need
not introduce a notation for it). The norm on F (hence also the norm
topology on F ) is identical to that on F . Defining Ψ : F × F → C by
Ψ(a, b) = Φ(a, b) = ha, bi , Ψ is a bilinear form on F × F (for instance,
Ψ(c · a, b) = hc · a, bi = hca, bi = cha, bi = cΨ(a, b) ) such that

|Ψ(a, b)| = |ha, bi| 6 |a| · |b| ,

hence Ψ is continuous (GT, IX, §3, No. 5, Th. 1), therefore so is Φ , and
|Φ(a, b)| 6 |a| · |b| . Since the proof of Cor. 1 does not require that Φ be
bilinear, we conclude that if f , g ∈ L F2 then Φ(f , g) ∈ LC1 and
Z 
 
 Φ(f , g) dµ 6 N2 (f )N2 (g) .
§6 convexity inequalities INT IV.x199

Now,

Φ(f , g) (x) = hf , gi(x) = hf (x), g(x)i ;

and if ef ∈ L2F is the class of f for equality almost everywhere (§3, No. 4,
Def. 2), then
Z 1/2
N2 (f ) = N2 (e
f) = |f (x)|2 dµ = ke
f k2 .

Thus hf , gi ∈ LC1 and

Z 
 
(∗)  hf , gi dµ 6 ke
f k2 · ke
g k2 .

The integral in (∗) depends only on the classes of f and g , so we may define
a function L2F × L2F → C by
Z
(e e) 7→
f, g hf , gi dµ

that is clearly sesquilinear, Hermitian and positive, by the properties of the

inner product in F and the linearity of integration. Writing
Z
he ei = hf , gi dµ ,
f, g

one has
Z Z

2
2
he
f,e
fi = hf , f i dµ = |f |2 dµ = N2 (f ) = ke
f k2 ,

thus he
f,e
f i > 0 when e
f 6= 0 . The norm e
f 7→ ke
f k derived from this form is

ke
f k = he
f,e
f i1/2 = ke
f k2 ;

as L2F is complete for this norm, we conclude that it is a Hilbert space for
the inner product he ei .
f, g
A shortcut, based on the characterization of a Hilbert space as a Banach
space whose norm satisfies the “parallelogram law ”:
By the parallelogram law in F , |f + g| 2 + |f − g|2 = 2|f |2 + 2|g|2 ;
integration term-by-term yields

(ke ek2 )2 + (ke

f +g ek2 )2 = 2(ke
f −g f k2 )2 + 2(ke
g k 2 )2 ,
INT IV.x200 extension of a measure. l p spaces §6

thus the norm of the Banach space L2F satisfies the parallelogram law, hence
L2F is a Hilbert space, with inner product

1 e
he
f ,e
gi = ek2 )2 − (ke
(kf + g f −gek2 )2 + i(ke
f + iegk2 )2 − i(ke g k 2 )2
f − ie
4Z Z Z Z
1n o
= |f + g|2 dµ − |f − g|2 dµ + i |f + ig|2 dµ − i |f − ig|2 dµ
4
Z
1
= |f + g|2 − |f − g|2 + i|f + ig|2 − i|f − ig|2 dµ
4
Z Z
= hf (x), g(x)i dµ(x) = hf , gi dµ .

IV.94, `. −4 to −1.
“Corollary 4.”
Immediate from Cor. 1, with Φ(λ, a) = λa ( λ scalar, a ∈ F ).
IV.95, `. 7, 8.
“ . . . since the inequality (8) is true for upper integrals (Ch. I, No. 2,
Cor. of Prop. 2) ”
One can suppose that the fi are everywhere finite and > 0 (§5, No. 6,
Cor. 3 of Th. 5), as required by the cited Cor. from Ch. I.
IV.95, `. −5 to −3.
P
n
“Suppose first that f is an integrable step function, f = a k ϕ Ak ,
k=1
where the Ak are pairwise disjoint (§4, No. 9, Lemma).”
The term ‘measurable step function’ is defined in the first paragraph
of §5, No. 5, but the term ‘integrable step function’ is nowhere explicitly
defined; we can infer its meaning from an earlier approximation theorem
(§4, No. 10, Cor. 1 of Prop. 19) cited later in the present proof. Proposed
definition: An integrable step function is an element of E F (Φ) (§4, No. 9,
Def. 4), where Φ is the clan (loc. cit., Prop. 17) of integrable subsets of X
(§4, No. 5, Props. 6 and 7).
Proposition. For a function f : X → F , the following conditions are
equivalent:
a) f is an integrable step function (in the above sense);
b) f is integrable and has only finitely many values.
Proof. a) ⇒ b): Obvious.
b) ⇒ a): Let a1 , . . . , an be the distinct nonzero values of f and, for
−1
each index k , let Ak = f (ak ) . Then every Ak is measurable (§5, No. 5,
§6 convexity inequalities INT IV.x201

Prop. 7) and
n
X
f= a k ϕ Ak .
k=1
P
n
The function |f | = |ak |ϕAk is also integrable (§4, No. 2, Cor. 1 of Prop. 1)
k=1
and, for each k , ϕAk 6 |ak |−1 |f | , whence Ak is integrable (§5, No. 6, Th. 5);
thus f is a step function with respect to the clan Φ of integrable sets. ♦
IV.95, `. −3 to IV.96, `. 1.
“For every ε > 0 , there exists (for every index k ) a vector a 0k ∈ F0
such that |a0k |q = |ak |p if p > 1 (resp. |a0k | = 1 if p = 1 ) and hak , a0k i >
(1 − ε)|ak | · |a0k | (TVS, IV, §1, No. 3, Prop. 8).”
Fix k . By the cited Prop. 8,
|ak | = sup |hak , a0 i| ,
a0 ∈F0 , |a0 |=1

so there exists an a0 with |a0 | = 1 and |hak , a0 i| > (1 − ε)|ak | ; multi-

plying a0 by a suitable scalar of absolute value 1 , we can suppose that
|hak , a0 i| = hak , a0 i and so
(∗) hak , a0 i > (1 − ε)|ak | .
{Indeed, by the Hahn-Banach theorem, there exists an a 0 ∈ F0 such
that |a0 | = 1 and hak , a0 i = |ak | (TVS, II, §8, No. 3, Cor. 1 of Th. 1).
However, in the proof of 2◦ , for F not reflexive, one must make do with the
analog of (∗).}
If p = 1 , define a0k = a0 ; then |a0k | = 1 and (∗) is precisely the desired
inequality.
If p > 1 (and < +∞ ), define a0k = |ak |p/q a0 ; then |a0k | = |ak |p/q , so
|ak | = |ak |p ; and multiplication of the inequality (∗) by |a k |p/q yields
0 q

hak , |ak |p/q a0 i > (1 − ε)|ak | · |ak |p/q ,

that is, hak , a0k i > (1 − ε)|ak | · |a0k | as desired.
IV.96, `. 7.
“ . . . which proves the relation (9) in this case.”
The case is 1 6 p < +∞ , with the normalization N p (f ) = 1 . The
argument shows that
Z 
 
 
sup  hf , gi dµ > Np (f ) ;
g∈LFq0 , Np (g)61

the reverse inequality is immediate from Hölder’s inequality (Th. 2).

INT IV.x202 extension of a measure. l p spaces §6

IV.96, `. 12.
Z
hf1 , gi dµ > Np (f1 ) − ε > 1 − 2ε .

1 = Np (f ) 6 Np (f − f1 ) + Np (f1 ) 6 ε + Np (f1 ) , whence the second

inequality of the display.
IV.96, `. −10.
Z 
 
 hf , gi dµ > 1 − 3ε
 

Abbreviate the relation

Z Z Z
hf , gi dµ = hf1 , gi dµ + hf − f1 , gi dµ

as α = β + γ ; we know that β is real, but α and γ need not be. Citing

inequalities proved earlier, one has
 
|α| − |β| 6 |α − β| = |γ| 6 Np (f − f1 ) Nq (g) 6 ε · 1 ,

whence −ε 6 |α| − |β| 6 ε and so

|α| > |β| − ε > β − ε > (1 − 2ε) − ε .

IV.96, `. −6.
“ . . . the set of x ∈ X such that |f (x)| > α is measurable ”
For, f is measurable (§5, No. 6, Th. 5), therefore so is |f | (§5, No. 3,
Cor. 6 of Th. 1), therefore so is the set in question (§5, No. 5, Prop. 8).
IV.96, `. −5.
“ . . . therefore it contains a compact set K of measure > 0 . ”
Let B = {x ∈ X : |f (x)| > α } . Since B is not locally negligible, there
exists a compact set H such that B∩H is not negligible (§5, No. 2, Prop. 5).
Since B is measurable, B ∩ H is integrable (§5, No. 6, Cor. 3 of Th. 5), and
since B ∩ H is not negligible it contains a compact set K of measure > 0
(§4, No. 6, Cor. 1 of Th. 4).
IV.96, `. −3 to −1.
“ . . . for every ε > 0 , there exists a partition of K 1 into a finite number
of integrable sets, in each of which the oscillation of f is 6 ε ”

Since f K1 is continuous, for every x ∈ K1 there exists an open neigh-
borhood Ux of x in X such that |f (y)−f (x)| 6 ε/2 on U x ∩K1 . Cover K1
§6 convexity inequalities INT IV.x203

with a finite number of open sets U1 , . . . , Un in X such that f K1 has os-
cillation 6 ε on each Ui ∩ K1 . The sets Ui ∩ K1 are integrable, and since
the integrable sets form a clan, the sets
[
U1 ∩ K1 , Ui ∩ K1 --- Uj ∩ K1 (i = 2, . . . , n)
j<i

form a partition of K1 with the desired properties.

IV.97, `. 3, 4.
“ . . . the function g = ϕA · a0 /µ(A) is integrable and N1 (g) = 1 ”
1 R
Writing g = ϕA , one has g ∈ LR1 , g dµ = 1 and g = g · a0 ;
µ(A)
thus g is an integrable step function X → F 0 (seeR the note for IV.95, `. −5
0
to −3) and |g| = |a | · g = g , whence N1 (g) = g dµ = 1 .
IV.97, `. 5.
Z Z
1
hf , gi dµ = hf , a0 iϕA dµ .
µ(A)
Since f ∈ LF∞ and g ∈ LF10 , we know from Cor. 2 of Th. 2 that
hf , gi ∈ LC1 . Explicitly, writing g = g · a0 as in the preceding note, hf , gi
is the function
x 7→ hf (x), g(x) · a0 i = hf (x), a0 ig(x) ,
where the function x 7→ hf (x), a0 i belongs to L ∞ and g ∈ L 1 .
IV.97, `. 7.
Z Z
0 0
hf , a iϕA dµ = ha, a iµ(A) + hf − a, a0 iϕA dµ

Here a also represents the constant function X → F with value a .

Thus f − a ∈ LF∞ whereas ϕA · a0 ∈ LF10 , and

hf , a0 iϕA = ha + (f − a), a0 iϕA = ha, a0 iϕA + hf − a, a0 iϕA ,

whence the asserted equation. Note that hf , a 0 iϕA = µ(A)hf , gi .

IV.97, `. 11.
Z 
 
 hf , gi dµ > |ha, a0 i| − ε > |a| − 2ε > α − 2ε
 

Since hf , a0 iϕA = µ(A)hf , gi , the equation of `. 7 may be written

Z Z
µ(A) hf , gi dµ = ha, a iµ(A) + hf − a, a0 iϕA dµ ;
0
INT IV.x204 extension of a measure. l p spaces §6

abbreviate the equation as a = b + c . Then

Z
 
|a| − |b| 6 |a − b| = |c| 6 ε ϕA dµ = ε µ(A) ,

therefore −ε µ(A) 6 |a| − |b| 6 ε µ(A) , whence

|a| > |b| − ε µ(A) = |ha, a0 i| · µ(A) − ε µ(A) = (|ha, a0 i| − ε)µ(A) ,

and so
Z 
 
 hf , gi dµ = |a| > |ha, a0 i| − ε > (|a| − ε) − ε > α − 2ε .
  µ(A)

IV.97, `. 12, 13.

“ . . . the relation (9) is also verified in this case.”
The foregoing shows that given any α with 0 < α < N ∞ (f ) and
1
any
R ε > 0 , there exists a function g ∈ L F0 such that N1 (g) = 1 and
 hf , gi dµ > α−2ε . This shows that the supremum in (9) is > α−2ε hence
(letting α → N∞ (f ) and ε → 0 ) it is > N∞ (f ) . The reverse inequality is
immediate from Cor. 2 of Th. 2.
IV.97, `. −15, −14.
◦
“ . . . it suffices to observe that the interior B of B is dense in B and
◦ ◦ ◦
that B∩ E is dense in B, since B is open.”
It is not necessary that the dense subset E of L Fq0 be a linear subspace.
By GT, I, §1, No. 6, Prop. 5,

◦ ◦ ◦
B ∩E ⊃ B∩E = B,

◦ ◦
therefore B ∩ E ⊃ B = B (TVS, II, §2, No. 6, Cor. 1 of Prop. 16), whereas
◦ ◦
B ∩ E ⊂ B = B , thus B ∩ E = B , whence obviously B ∩ E = B . Let
f ∈ LFp and let
s = sup |hf , gi| ;
g∈B∩E

we are to show that s = Np (f ) . The inequality s 6 Np (f ) is immediate

from Cor. 2 of Th. 2. To prove the reverse inequality it will suffice, by (9),
to show that |hf , gi| 6 s for every g ∈ B . Given g ∈ B , choose a sequence
gn ∈ B ∩ E such that Nq (gn − g) → 0 . Then hf , gn i → hf , gi , and since
|hf , gn i| 6 s for all n , it follows that |hf , gi| 6 s .
§6 convexity inequalities INT IV.x205

IV.97, `. −12, −11.

“But in this case, the formula (9) is true as g runs over B ∩ K (X; F 0 ) ,
even for p = 1 .”
The strategy is to refine the argument in the first paragraph of (i) in
the proof of Prop. 3 restricted to the case p = 1 , q = +∞ ; the second
paragraph then completes the argument without change.
We are given an integrable step function f : X → F with N 1 (f ) = 1 .
Let
B = {g ∈ LF∞0 : N∞ (g) 6 1 } ;
we know from Prop. 3 that
Z 
 
sup  hf , gi dµ = N1 (f ) = 1 .
g∈B

Inspection of the proof shows that B can be replaced by its subset consisting
of all integrable step functions g such that |g| 6 1 . Of course |g| 6 1 ⇒
N∞ (g) 6 1 . In the reverse direction, if g : X → F 0 is a measurable function
such that N∞ (g) 6 1 then |g(x)| 6 1 on the complement of a locally
negligible set N ; the function g1 = ϕ { N g is then measurable, |g1 | 6 1
and g = g1 locally almost everywhere. Since the set A = {x : f (x) 6= 0 }
is integrable, its intersection with N is negligible; for, A ∩ N is integrable
(§5, No. 6, Cor. 3 of Th. 5), and every compact set K ⊂ A ∩ N is negligible
(§5, No. 2, Prop. 5), therefore A∩N is negligible (§4, No. 6, Cor. 1 of Th. 4).
It follows that hf , gi = hf , g1 i almost everywhere; thus, writing B 1 for the
set of measurable functions g : X → F0 such that |g| 6 1 , that is, in the
notation of §5, No. 11,

B1 = {g ∈ S (X, µ; F0 ) : |g| 6 1 } ,

one has
Z 
 
(∗) sup  hf , gi dµ = N1 (f ) .

g∈B 1

Consider now the set

B2 = B ∩ K (X; F0 ) = {g ∈ K (X; F0 ) : N∞ (g) 6 1 } ;

we are to show that

Z 
 
(∗∗) sup  hf , gi dµ = N1 (f ) .
g∈B 2
INT IV.x206 extension of a measure. l p spaces §6

In fact, defining
B3 = {g ∈ K (X; F0 ) : |g| 6 1 }
(a subset of B1 ∩ B2 ) the argument will show that
Z 
 
(***) sup  hf , gi dµ = N1 (f ) .

g∈B 3

{Incidentally, if Supp µ = X then B2 = B3 . For, suppose g ∈ B2 and

let U = {x ∈ X : |g(x)| > 1 } . Then U is open, locally negligible (because
N∞ (g) 6 1 ) and has compact closure, therefore U = U ∩ U is negligible (§5,
No. 2, Prop. 5), hence U = ∅ (§2, No. 2, Prop. 5); in other words |g| 6 1 ,
that is, g ∈ B3 .}
IV.97, `. −8 to − 6.
“There exists a finite number of pairwise disjoint compact sets K i such
that g has a constant
R value a0i on each Ki and such that, if K is the union
of the Ki , then |f |ϕ { K dµ 6 ε .”
Pn
As in (i) of the proof of Prop. 3, we can suppose that f = a i ϕ Ai ,
i=1
where the ai are the distinct nonzero values of f , and the A i are pairwise
disjoint integrable sets. Given ε > 0 , one constructs elements a 0i ∈ F0 such
P
n
that |a0i | = 1 and hai , a0i i > (1 − ε)|ai | , and one defines g = a0i ϕAi .
i=1
For each i , choose a compact set Ki ⊂ Ai such that µ(Ai --- Ki ) 6
S
n
ε/n|ai | (§4, No. 6, Cor. 1 of Th. 4) and let K = Ki . Then Ai ∩ { K =
i=1
P
n
Ai --- Ai ∩ K = Ai --- Ki , therefore f ϕ { K = ai ϕAi -- Ki and
i=1

Z n
X n
X
|f |ϕ { K dµ = |ai |µ(Ai --- Ki ) 6 ε/n = ε .
i=1 i=1

IV.97, `. −3. P 0
“Setting h = ai hi ”
We require that h , hence the hi , have compact support: Ki ⊂ Vi ⊂
Hi ⊂ Ui for suitable open Vi and compact Hi ; a suitable hi then exists
by Lemma 1 of Ch. III, §1, No. 2.
IV.97, `. −1.
Z
|hf , hi|ϕ { K dµ 6 ε
§6 convexity inequalities INT IV.x207

For, hf , hiϕ { K = hf ϕ { K , hi , whence, by Cor. 2 of Th. 2,

Z
|hf , hi|ϕ { K dµ 6 N1 (f ϕ { K )N∞ (h) 6 ε · 1 .

IV.98, `. 1. R
“ . . . consequently | hf , hi dµ| > 1 − 3ε , which proves our assertion.”
Abbreviate the equation
Z Z Z
hf , hi dµ = hf , hiϕK dµ + hf , hiϕ { K dµ

by α = β + γ . Then
Z
 
|α| − |β| 6 |α − β| = |γ| 6 |hf , hi|ϕ { K dµ 6 ε ,

thus −ε 6 |α| − |β| 6 ε , whence |α| > |β| − ε . Now,

hf , hiϕK = hf , hϕK i = hf , gϕK i = hf , giϕK = hf , gi − hf , giϕ { K ,

thus Z Z
β= hf , gi dµ − hf , giϕ { K dµ ;

abbreviate this equation as β = σ − τ , noting that |σ| > 1 − ε by the choice

of g . Then
 
|β| − |σ| 6 |β − σ| = |τ | 6 N1 (f ϕ )N∞ (g) 6 ε · 1 ,
{K

therefore −ε 6 |β| − |σ| 6 ε , whence

|β| > |σ| − ε > (1 − ε) − ε ,

and finally |α| > |β| − ε > (1 − 2ε) − ε , which verifies the asserted inequality.
As in the note for IV.97, `. −12, −11, writing B 3 = {k ∈ K (X; F0 ) :
|k| 6 1 } , the foregoing shows that h ∈ B 3 and
Z  Z 
   
sup  hf , ki dµ >  hf , hi dµ = |α| > 1 − 3ε .
k∈B3

Since ε is arbitrary,
Z 
 
sup  hf , ki dµ > 1 = N1 (f ) ,
k∈B3

and the reverse inequality is immediate from Cor. 2 of Th. 2.

INT IV.x208 extension of a measure. l p spaces §6

IV.98, `. 4.
“ whose support is contained in a countable union of compact sets K n .”
Here, by “support” is meant the set {x : f (x) 6= 0 } ; it is not necessary
that its closure be contained in the union.
IV.98, `. 8.
“ . . . a special case of (9) ”
That is, (9) as generalized in Remark 1.
IV.98, `. 8, 9.
“ f is then equivalent to a function in L p (§5, No. 6, Th. 5).”
Since Np (f ) < +∞ , the set N = {x : f (x) = +∞ } is negligible (§2,
No. 3, Prop. 7). Then f ϕ { N is a finite-valued measurable function such
that f ϕ { N = f almost everywhere, whence Np (f ϕ { N ) = Np (f ) < +∞ ,
thus f ϕ { N ∈ L p by the cited Th. 5. In other words, f is p-th power
integrable (§3, No. 4, second paragraph after Def. 2).
Incidentally, the set {x : (f ϕ { N )(x) 6= 0 } is contained in the union
of a negligible set and a sequence of compact sets (§5, No. 6, Lemma 1),
hence the same is true of {x : f (x) 6= 0 } ; thus the assumption about the
‘support’ of f , not used in the case N p (f ) < +∞ , is in a sense redundant.
IV.98, `. 10.
“ . . . set fn = inf(n, f ϕKn ) . ”
It is understood that if f (x) = +∞ and x ∈ / K n then fn (x) =
inf(n, 0) = 0 by the convention +∞ · 0 = 0 . One can suppose that the
sequence (Kn ) is increasing.
Note that fn is measurable and 0 6 fn 6 nϕKn , hence fn is integrable.
If p < +∞ then Np (fn ) 6 n Np (ϕKn ) < +∞ , whereas N∞ (fn ) 6 n < +∞ ,
therefore (11) holds for fn for 1 6 p 6 +∞ .
The sequence (fn ) is increasing, with upper envelope equal to f , that
is, fn (x) ↑ f (x) for all x ∈ X . For, since ϕKn 6 ϕKn+1 , fn 6 fn+1 is
assured by f (x)ϕKn (x) 6 f (x)ϕn+1 (x) (even if f (x) = +∞ and x ∈ / Kn ).
If f (x) = +∞ then x ∈ Km for some index m ; then, for all n > m ,
f (x)ϕKn (x) = +∞ , therefore fn (x) = n , thus fn (x) ↑ +∞ = f (x) . If
0 < f (x) < +∞ then x ∈ Km for some m , and one can suppose that
m > f (x) ; then fn (x) = f (x) for all n > m , whence fn (x) ↑ f (x) . And if
f (x) = 0 then fn (x) = 0 for all n , fn (x) ↑ 0 = f (x) .
IV.98, `. 12, 13.
“ . . . passing to the limit (assuming, as we may, that the sequence (K n )
R∗
is increasing), we have sup |f g| dµ = +∞ (§1, No. 3, Th. 3).”
By assumption, Np (f ) = +∞ . We know (see the preceding note)
that the sequence (fn ) is increasing, with upper envelope f , and that
§6 convexity inequalities INT IV.x209

Np (fn ) < +∞ for all n . Thus every fn satisfies (11): writing B0 =

{g ∈ K (X; R) : Nq (g) 6 1 } , we have
Z ∗ Z ∗
Np (fn ) = sup |fn g| dµ 6 sup |f g| dµ 6 +∞ = Np (f ) ;
g∈B0 g∈B0

to verify that f satisfies (11), it suffices to show that sup Np (fn ) = Np (f ) =

n
+∞ . When p < +∞ , this follows from the cited Th. 3.
There remains the case that p = +∞ , q = 1 , B 0 = {g ∈ K (X; R) :
Nq (g) 6 1 } . We are to show that sup N∞ (fn ) = +∞ . Assume to the
n
contrary that the sequence N∞ (fn ) is bounded, say N∞ (fn ) 6 α < +∞
for all n . Then fn 6 α locally almost everywhere; let M n be a locally
S
∞
negligible set such that fn 6 α on X --- Mn . The set M = Mn is
n=1
then locally negligible and, for x ∈ { M , fn (x) 6 α for all n , therefore
f (x) 6 α . Thus f (x) 6 α locally almost everywhere; but then N ∞ (f ) 6 α ,
contrary to N∞ (f ) = +∞ .
IV.98, `. 14–17.
“Corollary. — Let µ be a positive measure on X , F a Banach space,
q
F0 its strong dual, and g any function in R L F0 . Then, the linear form
on LF , deduced from the linear form f 7→ hf , gi dµ on LFp by passage to
p

the quotient, is continuous and has norm N q (g) .”

When 1 6 p < +∞ , LpF is defined to be the set of elements e f
p e
( f ∈ LF ), where f is the class of f for the relation of equality almost
everywhere (§2, No. 6, §3, No. 4, Def. 2). But q = +∞ when p = 1 , and
L∞ ∞
F0 is defined to be the set of elements ġ ( g ∈ LF0 ), where ġ is the
class of g for the relation of equality locally almost everywhere (No. 3); we
are then in the situation of “mixing apples and oranges”. The purpose of
this note is to clarify this situation (of which the case of 1 < p < +∞ ,
q = (p − 1)/p is a straightforward simplification).
Lemma. If f , f1 ∈ LF1 and g, g1 : X → F0 are functions such that
f = f1 almost everywhere and g = g1 locally almost everywhere, then
hf , gi = hf1 , g1 i almost everywhere.
Proof. It suffice to show that hf , gi = hf 1 , gi almost everywhere and
hf1 , gi = hf1 , g1 i almost everywhere. By bilinearity (the note for IV.94, `. 8
is pertinent here!) this reduces to showing that, for f ∈ L F1 and g : X → F0 ,

f = 0 a.e. ⇒ hf , gi = 0 a.e.
g = 0 l.a.e. ⇒ hf , gi = 0 a.e.
INT IV.x210 extension of a measure. l p spaces §6

The first implication is obvious. Suppose g = 0 locally almost everywhere,

and let M = {x : g(x) 6= 0 } , A = {x : f (x) 6= 0 } ; then f = f ϕ A ,
g = gϕM . We know that M is locally negligible; and, since f ∈ L F1 ,
S
∞
A ⊂ N∪ Kn with N negligible and the Kn compact (§5, No. 6, Lem-
n=1
ma 1). Then the set
∞
[
A ∩ M ⊂ (N ∩ M) ∪ Kn ∩ M
n=1

is negligible (§5, No. 2, Prop. 5), therefore ϕ A ϕM = ϕA∩M = 0 almost

everywhere, thus hf , gi = hf ϕA , gϕM i = hf , giϕA ϕM = 0 almost every-
where. ♦
Fix g ∈ LF∞0 . If f ∈ LF1 then by Hölder’s inequality (Cor. 2 of Th. 2)
hf , gi is integrable and
Z 
 
 hf , gi dµ 6 N1 (f )N∞ (g) .

By the Lemma, we may define λ : L1F → R by

Z
e
λ(f ) = hf , gi dµ (e
f ∈ L1F ) ;

λ is a linear form, and |λ(e f )| 6 N1 (f )N∞ (g) shows that it is continuous

and that its norm, defined by
Z  Z 
   
kλk = sup  hf , gi dµ = sup  hf , gi dµ ,
ef ∈L1F , N1 (ef )61 1
f ∈LF , N1 (f )61

satisfies kλk 6 N∞ (g) . Indeed, kλk = N∞ (g) by formula (10) of Prop. 3.

Moreover, the Lemma shows that λ depends only on the class ġ of g . Thus,

1 0
to each u = ġ ∈ L∞F0 there corresponds a continuous linear form λ u ∈ LF
such that kλu k = N∞ (g) = N∞ (u) ; and u 7→ λu defines a linear isometry


1 0
of L∞F0 into the dual space LF (shown in Ch. V, §5, No. 8, Th. 4 to be

0
surjective when F = R ). The dual L1F , for F a separable Banach space,
is calculated in Ch. VI, §2, No. 6, Prop. 10.
The ‘dual’ result also holds: for fixed f ∈ L F1 , the correspondence
Z
ġ 7→ hf , gi dµ (g ∈ LF∞0 )

defines a continuous linear form on L ∞ F0 , with norm equal to N1 (f ) . The

proof is formally the same, citing the formula (9) of Prop. 3 in the calculation
§6 convexity inequalities INT IV.x211

0
of the norm. Whence a linear isometry L 1F → L∞F0 ; but this mapping need
not be surjective, even when F = R (Ch. V, §5, No. 8, Prop. 14).
When 1 < p < +∞ e e , obtaining linear

0 , one does business

0 with f and g
isometries LpF → LqF0 and LqF → LpF0 , both of which are surjective when
F = R (Ch. V, loc. cit., Th. 4).
IV.98, `. −10 to −7.
“We already know (Ch. I, No. 3, Prop. 5) that the set J of finite num-
bers p > 1 such that Np (f ) < +∞ is either empty or is an interval, and
that log Np (f ) is a convex function of 1/p on J (when f is not negligible);
this of course implies the continuity of p 7→ N p (f ) on J . ”


The cited Prop. 5 contemplates f 7→ Np (f ) f ∈ F (X; R for
0 < p < +∞ . In the present context, a measurable function f : X → F
is fixed, |f | plays the role of f , and we are interested in the function
R 1/p
∗ p
p 7→ Np (|f |) = |f | d|µ| for 1 6 p < +∞ (§3, No. 2, Def. 1);
the definition can be stretched to 0 < p < +∞ by means of the same for-
mula, as follows. One observes (§3, No. 1) that the function M defined on
R∗
the set of all positive functions f : X → R by M(f ) = f d|µ| satisfies
◦ ◦ ◦
the axioms 1 , 2 , 3 of Ch. I, No. 1, Def. 1. In Ch. I, No. 3 one defines, for
every p with 1 6 p < +∞ , the vector space F p = F p (X, M) to be the set
of all functions f : X → R such that M(|f | p ) < +∞ . Then, for every p
with 0 < p < +∞ , one defines

1/p

Np (f ) = M(|f |p ) f ∈ F (X; R) ,

accepting that Np (f ) = +∞ when M(|f |p ) = +∞ . The notation F p

remains reserved for the vector space obtained when 1 6 p < +∞ , in which
case Np F p is a semi-norm. Thus, for f : X → F , the proposed definition

Z ∗ 1/p
Np (f ) = |f |p d|µ| (0 < p < +∞)

is consistent with the notations of Ch. I, No. 3.

Fix a function f : X → F . Let us dispose of a trivial special case:
R∗
Suppose f is negligible, by which we mean that N 1 (f ) = |f | d|µ| = 0 .
{This is the definition suggested by the paragraph before the Remark in
§3, No. 3 (see §3, No. 2, Def. 1 and Prop. 3); equivalently, f = 0 almost
everywhere (§2, No. 3, Th. 1); equivalently, |f | p = 0 almost everywhere,
for 0 < p < +∞ ; equivalently, Np (f ) = 0 for 0 < p < +∞ . } Then
also N∞ (f ) = 0 : for every α > 0 , |f | 6 α almost everywhere, hence
locally almost everywhere, therefore N ∞ (f ) = inf α = 0 . Thus I = [1, +∞]
INT IV.x212 extension of a measure. l p spaces §6

and the function p 7→ Np (f ) (p ∈ I) is identically 0 , as its restriction to

J = [1, +∞[ .
Assume henceforth that f is not negligible. By Prop. 5 of Ch. I, No. 3,
the set
S = {1/p : 0 < p < +∞ , Np (f ) < +∞ }
is either empty or is an interval (possibly degenerate). {This set is called I
in the cited Prop. 5, but the letter I has been preempted here for other use.}
Now,

J = {p : 1 6 p < +∞, Np (f ) < +∞ } = {p : 1/p ∈ S and 1/p 6 1 } ;

thus J is the image of S ∩ ]0, 1] under the mapping s 7→ 1/s , that is,

J = {s−1 : s ∈ S ∩ ]0, 1] ,

consequently J is either empty or is an interval (possibly degenerate).

If J = ∅ then S ∩ ]0, 1] = ∅ , therefore

0<α61 ⇒ α∈
/ S ⇒ N1/α (f ) = +∞ ,

thus Np (f ) = +∞ for all p ∈ [1, +∞[ , whence the set

I = {p ∈ [1, +∞] : Np (f ) < +∞ }

is equal to either ∅ or {+∞} , and the assertions of Prop. 4 hold trivially.

If J = {p} for some 1/p ∈ ]0, 1] , then 1 6 p < +∞ , 0 < N p (f ) < +∞ ,
and the continuity condition is trivially satisfied.
Finally, suppose J is a nondegenerate interval. If α ∈ S , then α = 1/p
with 0 < Np (f ) < +∞ , in particular log Np (f ) 6= −∞ (i.e., Np (f ) 6= 0 ); so
by the cited Prop. 5, we know that the function

(∗) α 7→ log N1/α (f ) (α ∈ S)

is convex on S , hence is continuous at every interior point of S (FRV, I,

§4, No. 3, Prop. 6). Indeed, writing g(α) = log N 1/α (f ) (α ∈ S) , the cited
Prop. 6 shows that at every interior point of S , g has finite left and right
derivatives (whence continuity at the point); and the proof shows that if an
endpoint of S belongs to S , then g has a finite 1-sided derivative at that
point, hence is continuous on the right or on the left at the point, as the case
may be. Thus the mapping (∗) is continuous everywhere on S . A fortiori,
its restriction

(∗∗) α 7→ log N1/α (f ) (α ∈ J−1 = S ∩ ]0, 1])

§6 convexity inequalities INT IV.x213

is continuous; composing it with the mapping p 7→ 1/p (p ∈ J) , we see that

the mapping

(***) p 7→ log Np (f ) (p ∈ J)

is continuous; finally, composing with the exponential function, the desired

mapping

(****) p 7→ Np (f ) (p ∈ J)

is continuous.
IV.98, `. −3 to −1.
“If s ∈ J then, for every finite number p > s , |f | p = |f |s |f |p−s , and
the inequality of the mean shows that

s/p
(p−s)/p
(12) Np (f ) 6 Ns (f ) N∞ (f ) .”

s/p
Since s ∈ J and f is not negligible, 0 < Ns (f ) < +∞ . If N∞ (f ) =
+∞ the right side of (12) is equal to +∞ and the inequalilty is trivial.
Suppose N∞ (f ) < +∞ , that is, +∞ ∈ I , thus f is bounded in measure.
Note that N∞ (f ) > 0 . {For, N∞ (f ) = 0 would imply that the set A = {x :
S
∞
f (x) 6= 0 } is locally negligible; but |f | p is integrable, so A ⊂ N ∪ Kn
n=1
for a suitable negligible set N and compact sets K n (§5, No. 6, Lemma 1),
S
∞
whence A = (A ∩ N) ∪ A ∩ Kn is negligible (§5, No. 2, Prop. 5), contrary
n=1
to the assumption that f is not negligible.} Thus 0 < N ∞ (f ) < +∞ . Since
p − s > 0 , one has

p−s
p−s
N∞ (|f |p−s ) = M∞ (|f |p−s ) = M∞ (|f |) = N∞ (f ) < +∞

thus the measurable function |f |p−s is bounded in measure, whereas

|f |s ∈ L 1 ; by the inequality of the mean (No. 2, Prop. 1), |f | p = |f |s |f |p−s
is integrable—therefore p ∈ J —and
Z Z
|f | d|µ| 6 M∞ (|f | ) |f |s d|µ| ,
p p−s

p
p−s
s
that is, Np (f ) 6 N∞ (f ) Ns (f ) , whence the asserted inequality.
The argument shows that p ∈ J for all p ∈ (s, +∞) , therefore J must be
of the form ]a, +∞[ or [a, +∞[ , and then I = ]a, +∞] or [a, +∞] .
INT IV.x214 extension of a measure. l p spaces §6

A corollary of the argument is that if J consists of a single point then

+∞ ∈ / I , therefore I = J .

s/p
For use in the next note, we observe that as p → +∞ , Ns (f ) → 1;

(p−s)/p
and if N∞ (f ) < +∞ then N∞ (f ) → N∞ (f ) as p → +∞ .
IV.99, `. 1, 2.
“Letting p tend to +∞ , it follows that

(13) lim sup Np (f ) 6 N∞ (f ) . ”

p→+∞

The presumption is that J is of the form ]a, +∞[ or [a, +∞[ , so that
p → +∞ makes sense. If N∞ (f ) = +∞ , the inequality holds trivially.
If N∞ (f ) < +∞ then, applying Prop. 11 of GT, IV, §5, No. 6 to the in-
equality (12), one has
h
s/p
(p−s)/p i
lim sup Np (f ) 6 lim sup Ns (f ) N∞ (f )
p→+∞ p→+∞
h
s/p
(p−s)/p i
= lim Ns (f ) N∞ (f ) = 1 · N∞ (f )
p→+∞

(by the preceding note).

If the right endpoint of J is < +∞ , then N ∞ (f ) = +∞ by the pre-
ceding note, so +∞ ∈ / I and I = J .
IV.99, `. 4.
“ . . . thus I is indeed an interval of R , and I = J . ”
If +∞ ∈ / I then I = J , whereas if +∞ ∈ I then, by the foregoing,
J has right endpoint equal to +∞ and I = J ∪ {+∞} ; in either case, I is
an interval of R and I = J .
IV.99, `. 4–6.
“The proposition will be proved if we show that p 7→ N p (f ) is contin-
uous on J , and it suffices to establish continuity at the end-points of J . ”
After continuity is proved, the conclusion about convexity on J will be
a trivial consequence of the convexity on J already established.
We already know that p 7→ Np (f ) is continuous on J ; in particular
one-sided continuity holds at an endpoint that belongs to J . Assuming the
interval J to be nondegenerate, let r, s be its endpoints, 1 6 r < s 6 +∞ .
There are three situations where continuity in J remains to be established:
(i) r ∈/ J ; (ii) s < +∞ and s ∈ / J ; (iii) s = +∞ . In the following analysis,
p always signifies an element of J .
(i) If r ∈ / J , then Nr (f ) = +∞ and the problem is to show that
Np (f ) → +∞ as p ↓ r . The function 1/p 7→ log N p (f ) is already known
§6 convexity inequalities INT IV.x215

to be convex on J−1 (whose endpoints are 1/s < 1/r , where 1/s = 0 if
s = +∞ ); with the convention that log(+∞) = +∞ , the addition of the
point (1/r, +∞) to its graph creates no new chords between pairs of points
to challenge convexity.
(ii) If s < +∞ and s ∈ / J , then Ns (f ) = +∞ and the problem is
to show that Np (f ) → +∞ as p ↑ s , with the same conclusion concerning
convexity.
(iii) When s = +∞ , of course s ∈ / J ; and J −1 has endpoints 0 < 1/r .
Whether +∞ ∈ I or +∞ ∈ / I , that is, N∞ (f ) < +∞ or N∞ (f ) = +∞ , the
problem is to show that Np (f ) → N∞ (f ) as p ↑ +∞ . It has already been
shown that
lim sup Np (f ) 6 N∞ (f ) ;
p→+∞

showing that
lim inf Np (f ) > N∞ (f )
p→+∞

will establish that

N∞ (f ) 6 lim inf Np (f ) 6 lim sup Np (f ) 6 N∞ (f ) ,

p→+∞ p→+∞

whence the continuity of p 7→ Np (f ) at +∞ . There remains the question of

convexity in this case. We have J = ]r, +∞[ or [r, +∞[ , therefore J −1 =
]0, 1/r[ or ]0, 1/r] . If N∞ (f ) = +∞ then log N∞ (f ) = +∞ and the addi-
tion of the point (0, +∞) to the graph of the function σ(1/p) = log N p (f )
( 1/p ∈ J−1 ) produces no new chords, so all is well with convexity. Other-
wise, 0 < N∞ (f ) < +∞ and one adds the ‘finite’ point (0, log N ∞ (f )) to
the graph of σ , which does introduce new chords, namely the ones with left
endpoint (0, log N∞ (f )) . Suppose 0 < 1/t < 1/p and consider the chord C
joining (0, log N∞ (f )) and (1/p, log Np (f )) ; we wish to show that the point
(1/t, log Nt (f )) lies on or below C . Let t0 be such that 0 < 1/t0 < 1/t < 1/p
and let C0 be the chord joining (1/t0 , log Nt0 (f )) and (1/p, log Np (f )) . By
the convexity of σ on J−1 , we know that (1/t, log Nt (f )) lies on or be-
low C0 ; as t0 converges to t , the chord C0 ‘converges’ to C , so in the
limit (1/t, log Nt (f )) lies on or below C . (One can, tediously, say it with
inequalities.)
To summarize (assuming J a nondegenerate interval and f non-negligi-
ble): once lim Np (f ) has been calculated at the points of the set J --- J (a set
with at most two elements, at least one of them finite), the issue of convexity
of the mapping 1/p 7→ log Np (f ) ( p ∈ I = J )—in other words the convexity
of t 7→ log N1/t (f ) on the closure of J−1 —is straightforward, and adds to
what we know about J only in the case that J has right endpoint +∞ and
N∞ (f ) < +∞ .
INT IV.x216 extension of a measure. l p spaces §6

IV.99,R`. 14. R r
“ |f |p ϕA d|µ| tends to |f | ϕA d|µ| (§4, No. 3, Prop. 4).”
When p > r , |f | ϕA 6 |f | ϕA ∈ L 1 assures that |f |r ϕA ∈ L 1 .
r p

IV.99, `. 15.
R p R∗ r
“ Therefore |f | d|µ| tends to |f | d|µ| ”
R∗ r
We need only consider the case that r ∈
/ J , so that |f | d|µ| = +∞ .
Then
Z Z Z Z Z ∗
|f | d|µ| = |f | ϕA d|µ|+ |f | ϕ { A d|µ| → |f | ϕA d|µ|+ |f |r ϕ { A d|µ|
p p p r

by the preceding Note, Th. 3 of §1, No. 3, and

R the continuity of addition in
[0, +∞] (GT, IV, §4, No. 3, Prop. 7), and ∗ |f |r ϕ { A d|µ| = +∞ follows
from the fact that |f |r is not integrable.
IV.99, `. 15, 16.
“ . . . which proves the continuity of p 7→ N p (f ) at r .”
When r ∈ J this is not news; assuming r ∈ / J , so that N r (f ) = +∞ ,
we are to show that Np (f ) → +∞ as p → r . To simplify the notation, let
R R∗
us write ap = |f |p d|µ| (p ∈ J) . We know that ap → |f |r d|µ| = +∞ . It
follows that log ap → +∞ (given any real number M > 0 , ap > eM for p
sufficiently near r ). As p → r one has 1/p → 1/r , therefore

1 1
log ap → · (+∞) = +∞
p r

(GT, IV, §4, No. 3, Prop. 8), whence (a p )1/p → +∞ , that is, Np (f ) → +∞ .
IV.99, `. 17.
“The same reasoning may be applied at the point s if s < +∞ . ”
In particular, the roles of A and { A are reversed: when p < s ,
|f |s ϕ { A 6 |f |p ϕ { A ∈ L 1 assures that |f |s ϕ { A ∈ L 1 .
IV.99, `. −12.
“ . . . let a be a number such that 0 < a < N ∞ (f ) . ”
Since J 6= ∅ we know that N∞ (f ) > 0 (see the note for IV.98, `. −3
to −1). {It is not enough to note that f is not negligible; it might still be
locally negligible—that is, N∞ (f ) = 0 —but not when J 6= ∅ .}
IV.99, `. −10.
“ . . . non-negligible ”
Since N∞ (f ) is the infimum of the numbers M > 0 such that |f | 6 M
locally almost everywhere, it is not the case that |f | < a locally almost
§6 convexity inequalities INT IV.x217

everywhere. That is, the set A = {f : |f (x)| > a } is not locally negligible,
hence is not negligible.
IV.99, `. −7.
“ . . . which completes the proof.”
Letting a → N∞ (f ) , one has lim inf Np (f ) > N∞ (f ) . This completes
p→+∞
the proof of lim Np (f ) = N∞ (f ) and hence of the Proposition (see the
p→+∞
note for `. 4–6).
IV.99, `. −6 to −4.
“Corollary.”
By assumption, 1 6 r < p < s 6 +∞ and f ∈ LFr ∩ LFs ; we are to
show that f ∈ LFp , and can suppose that f is not negligible.
From f ∈ LFr we know that Nr (f ) < +∞ , so that r belongs to the
interval J of the proof of Prop. 4; if s < +∞ then similarly s ∈ J , therefore
Np (f ) < +∞ for all p ∈ [r, s] . On the other hand, if s = +∞ then f ∈ L F∞
is bounded in measure, N∞ (f ) < +∞ ; since J is nonempty, the proof of
Prop. 4 shows that J is an interval with right endpoint +∞ , consequently
Np (f ) < +∞ for every p ∈ [r, +∞[ .
IV.100, `. 9, 10.
“This is an immediate consequence of Prop. 4 above and of the Cor. of
Prop. 4 of Ch. I, No. 3.”
R
We may write simply kµk = |µ|(X) = d|µ| (§4, No. 7, Prop. 12). Since
µ is bounded, every measurable set in X is integrable (§5, No. 6, Cor. 1 of
Th. 5). It follows that every locally negligible set A in X is negligible; for,
A is measurable (§5, No. 2, sentence after Def. 3), hence integrable, therefore
S
∞
A = N∪ Kn with N negligible and the Kn compact (§4, No. 6, Cor. 2
n=1
of Th. 4), and the Kn = A ∩ Kn are also negligible (§5, No. 2, Prop. 5).
If N∞ (f ) < +∞ , then N∞ (f ) may be described as the infimum of all real
numbers α > 0 such that |f | 6 α almost everywhere, and N ∞ (f ) = 0
means that f = 0 almost everywhere
p (Nos. 2, 3). If N ∞ (f ) < +∞ then, for
p
all p ∈ [1, +∞[ , |f | 6 N∞ (f ) almost everywhere,
Z Z
p

p
|f | d|µ| 6 N∞ (f ) d|µ| ,

whence Np (f ) 6 N∞ (f ) · kµk1/p < +∞ .

As in Ch. I, No. 1, write P for the set of all functions f : X → R such
that f > 0 , and define

M (f ) = kµk−1 |µ|*(f ) (f ∈ P) ;
INT IV.x218 extension of a measure. l p spaces §6

then M satisfies the conditions 1◦ , 2◦ , 3◦ of Ch. I, No. 1 (§1, No. 3, Props. 10,
11, 12), and M (1) = kµk−1 |µ|*(1) = 1 . For 0 < p < +∞ and any function
f : X → F , define

1/p
N0p (f ) = M (|f |p
(cf. the note for IV.98, `. −10 to −7); since M (αf ) = αM (f ) for f ∈ P
and scalars α > 0 , we have

1/p
(∗) N0p (f ) = kµk−1 |µ|*(|f |p ) = kµk−1/p Np (f ) ,

which suggests the definition N0∞ (f ) = N∞ (f ) (formally ‘let p → ∞ in (∗)’).

Since M (1) = 1 , for every function f : X → F the mapping p 7→ N 0p (f )
(0 < p < +∞) is increasing (Ch. I, No. 3, Cor. of Prop. 4). {Review: The
argument given in Ch. I shows that if 0 < r < p and N 0p (f ) < +∞ , then
N0r (f ) 6 N0p (f ) , whence p 7→ N0p (f ) is finite-valued and increasing on the
interval [r, p] . It follows that if 0 < r < p < +∞ then N 0r (f ) 6 N0p (f ) :
trivial if N0p (f ) = +∞ , and true by the foregoing if N 0p (f ) < +∞ .}
Consider now the assertion of Prop. 5 (in particular, f is measurable):
let I be the set of all p ∈ [1, +∞] such that N p (f ) < +∞ , equivalently,
N0p (f ) < +∞ , and assume that I 6= ∅ ; we are to show that I is an interval
and the function p 7→ N0p (f ) (p ∈ I) is increasing. We can suppose that f is
not negligible, equivalently, N∞ (f ) > 0 .
If N∞ (f ) < +∞ , that is, f ∈ LF∞ , then, as observed in the first
paragraph of the note, f ∈ LFp for all p ∈ [1, +∞[ . Thus I = [1, +∞] .
We know that p 7→ N0p (f ) is finite-valued and increasing on [1, +∞[ ; it
remains to show that if 1 6 p < +∞ then N 0p (f ) 6 N0∞ (f ) . Indeed,

p
|f |p 6 N∞ (f ) · 1 almost everywhere, therefore
Z

p
|f |p d|µ| 6 N∞ (f ) · kµk ,

whence Np (f ) 6 N∞ (f ) · kµk1/p and so

N0p (f ) = kµk−1/p Np (f ) 6 N∞ (f ) = N0∞ (f ) .

Finally, suppose N∞ (f ) =+∞ , so that

I = {p ∈ [1, +∞[ : N0p (f ) < +∞ } .

By assumption, I 6= ∅ ; let p ∈ I . We know from the discussion above that

[1, p[ ⊂ I , thus I is an interval (it is the interval J of the proof of Prop. 4)
with left endpoint 1 , conceivably degenerating to the single point 1 , on
which the function p 7→ N0p (f ) is increasing (Ch. I, No. 3, Cor. of Prop. 4).
§6 convexity inequalities INT IV.x219

IV.100, `. 11–13.
“Corollary.”
The corollary is valid for 1 6 r < s 6 +∞ , except that when s = +∞
the topology on LF∞ defined by the semi-norm N∞ is the topology of
‘uniform convergence locally almost everywhere’ (No. 3).
For a measurable function f : X → F let us write I f for the set of all
p ∈ [1, +∞] such that Np (f ) < +∞ . Suppose f ∈ LFs , that is, s ∈ If . By
Prop. 5, If is an interval with left endpoint 1 , therefore [1, s] ⊂ I f and in
particular r ∈ If , that is, f ∈ LFr . Thus LFs ⊂ LFr . Moreover, for a fixed
measurable function f , if If 6= ∅ then, with notation as in the preceding
note, the function

p 7→ N0p (f ) = kµk−1/p Np (f ) ( p ∈ If )

is increasing. In particular (for the given r and s ), N 0r (f ) 6 N0s (f ) when

f ∈ LFs , that is,

(∗) kµk−1/r Nr (f ) 6 kµk−1/s Ns (f )

(when s = +∞ , the convention is that 1/s = 0 and the inequality reads

kµk−1/r Nr (f ) 6 1 · N∞ (f ) = N0∞ (f ) ). Writing the inequality (∗) as
 
kµk−1/s
Nr (f ) 6 Ns (f ) for all f ∈ LFs ,
kµk−1/r

we see that the canonical injection L Fs → LFr is continuous for the respec-
tive semi-norm topologies; in other words, the topology on L Fs induced by
the Nr -topology on LFr is coarser than the Ns -topology.
Incidentally, KF ⊂ LFp for all p ∈ [1, +∞] , and the normed space L pF
derived from the semi-norm Np is a Banach space (No. 3, Prop. 2 and §3,
No. 4, Th. 2).
IV.100, `. −13.
“ N∞ (f ) = kf k = sup |f (x)| ”
x∈X
Every function f : X → F is continuous, hence measurable, and the
empty set is the only negligible set, hence the only locally negligible set.
IV.100, `. −13 to −11.
“ . . . if there exists a number α > 0 such that |f (x)| > α for infinitely
many values of x ∈ X , then Np (f ) = +∞ for every finite p ”
In this case I = ∅ or {+∞} according as |f | is unbounded or bounded.
INT IV.x220 extension of a measure. l p spaces §6

IV.100, `. −11 to −8.

“ in the contrary case, there exists an x 0 ∈ X such that |f (x0 )| = kf k ,
whence
N∞ (f ) = |f (x0 )| 6 Np (f )
for every finite p . ”
The contrary case: for every α > 0 , |f (x)| < α for all but finitely
many x ; since the compact sets are the finite sets, this means that the
(continuous) function f “tends to 0 at infinity”, that is, f ∈ C 0 (X; F)
(cf. Ch. III, §1, No. 2, Prop. 3). In particular, f is bounded.
Choose y ∈ X with f (y) 6= 0 . The set A = {x : |f (x)| > |f (y)| } is
finite; if x0 ∈ A is chosen so that |f (x0 )| = max{|f (x)| : x ∈ A } , then
N∞ (f ) = |f (x0 )| .
p
p
For every finite p > 1 , N∞ (f ) = |f (x0 )|p 6 Np (f ) , whence
N∞ (f ) 6 Np (f ) .
If Np (f ) = +∞ for all finite p > 1 , then I = {+∞} (an example is
given at the end of this note).
Otherwise, there exists a finite p > 1 such that N p (f ) < +∞ , in which
case I is a nondegenerate interval with right end-point +∞ and left end-
−1
point r , 1 6 r < +∞ (Prop. 4). Then I = [r, +∞] , and I = [0, 1/r] .
We know from Prop. 4 that the function

g(α) = log N1/α (f ) ( α ∈ [0, 1/r] )

is convex. Note that g(0) = log N∞ (f ) < +∞ , whereas g(1/r) = log Nr (f )

is finite or equal to +∞ according as r ∈ I or r ∈ / I.
(i) Consider first the case that r ∈ I , so that I = I = [r, +∞] , and g is
a finite-valued convex function on [0, 1/r] . We know that for r 6 p < +∞ ,
0 < N∞ (f ) 6 Np (f ) , whence

g(0) = log N∞ (f ) 6 log Np (f ) = g(1/p) ,

that is, g(0) 6 g(α) for all α ∈ ]0, 1/r] ; thus g takes its smallest value
at the left end-point 0 . Let us show that g is an increasing function. For
0 6 α, β 6 1/r , α 6= β , write

g(β) − g(α)
Mαβ = = Mβα
β −α


for the slope

of the chord of the graph of g joining the points α, g(α) and
β, g(β) . Since g is convex, for every α ∈ [0, 1/r] the function

(∗) γ 7→ Mαγ ( γ ∈ [0, 1/r] --- {α} )

§6 convexity inequalities INT IV.x221

is increasing (FRV, I, §4, No. 3, Prop. 5). Suppose 0 6 α < β 6 1/r . If

α = 0 , then
g(β) − g(0)
M0β = >0
β
by the minimality of g(0) , thus g(0) 6 g(β) for all β ∈ ]0, 1/r] . Whereas,
if 0 < α < β 6 1/r , necessarily g(α) 6 g(β) ; otherwise, g(α) > g(β) would
imply that Mαβ < 0 , which, along with Mα0 > 0 , would contradict the fact
that the mapping (∗) is increasing.
(ii) Consider now the general case. If r ∈ I then g is increasing on I −1
by the foregoing. Suppose r ∈ / I , that is, g(1/r) = +∞ . Fix a number γ ,
0 < γ < 1/r ; the restriction of g to [0, γ] is convex, therefore g is increasing
by the argument of case (i). But for every such γ , g(γ) < +∞ = g(1/r) ;
thus g is increasing on [0, 1/r] = I −1 .
We have shown that if I 6= ∅ and I 6= {+∞} (equivalently, there exists
a finite p > 1 such that Np (f ) < +∞ ; see the Proposition below), then
f → 0 at infinity, I is a nondegenerate interval with right end-point equal
to +∞ , I = [r, +∞] with 1 6 r < +∞ , and the function

α 7→ log N1/α (f ) ( α ∈ I −1 = [0, 1/r] )

(convex by Prop. 4) is increasing. In particular, as p ∈ I increases,

α = 1/p decreases, therefore log Np (f ) decreases, therefore Np (f ) decreases;
thus Np (f ) is a decreasing function of p ∈ I , attaining its minimum at
p = +∞ . ♦
Proposition. Let X be a discrete space, µ a measure on X , F a Ban-
ach space, f : X → F , f 6= 0 , and I the set of all p ∈ [1, +∞] such that
Np (f ) < +∞ . The following conditions are equivalent:
(a) I 6= ∅ and I 6= {+∞} ;
(b) there exists a finite p > 1 such that f ∈ L Fp ;
(c) f → 0 at infinity and I 6= {+∞} .
Proof. (a) ⇒ (b): Obvious.
(b) ⇒ (c): At any rate p ∈ I , so I 6= {+∞} . If one did not have f → 0
at infinity, there would existPan ε > 0 such that |f (x)| > ε for infinitely
many values of x , whence |f (x)|p = +∞ , contrary to f ∈ LFp .
x∈X
(c) ⇒ (a): Since f → 0 at infinity |f | is bounded, that is, +∞ ∈ I ;
thus I 6= ∅ , whereas I 6= {+∞} by assumption. ♦
The following example shows that the condition I 6= {+∞} in (c) can-
not be omitted:
A positive-term sequence (cn ) such that cn → 0 but (cn ) is not
p-summable for any finite p > 1 . Here X is the set of positive integers
INT IV.x222 extension of a measure. l p spaces §7

with the discrete topology, µ is the atomic (hence discrete) measure defined
by a mass +1 at every point of X (Ch. III, §1, No. 3, Example I), and a
function on X is displayed as a sequence.
1 1
Let cn = . If 0 < p < +∞ , then cpn > for all sufficiently
log n n
P∞
large n , therefore cpn = +∞ . Thus I = {+∞} for this sequence. {This
i=1
example is given in Konrad Knopp’s Infinite sequences and series (Dover,
New York, 1956), p. 60, Example 7 of 3.2.1; and Theory and application of
infinite series (Hafner, New York, 1951), p. 119, Example g).}
A sequence for which I = ∅ : Interlace (c n ) with any unbounded se-
quence.
A sequence for which I = [1, +∞] : Any absolutely summable sequence.
IV.100, `. −4 to −1.
“Corollary.” (of Prop. 6)
The corollary is valid for 1 6 r < s 6 +∞ , except that when s = +∞
the topology on LF∞ defined by the semi-norm N∞ is the topology of
uniform convergence.
Let 1 6 r < s 6 +∞ and let f ∈ LFr . In the notation of Prop. 6,
r ∈ I . By Prop. 6, [r, +∞] ⊂ I , hence also s ∈ I , that is, f ∈ L Fs ;
thus LFr ⊂ LFs . Moreover, r < s implies by Prop. 6 that N r (f ) > Ns (f ) ,
therefore the canonical injection L Fr → LFs is continuous for the respective
semi-norm topologies; in other words, the topology on L Fr induced by the
Ns -topology of LFs is coarser than the Nr -topology.

§7. BARYCENTERS

IV.101, `. 7, 8. R
“ . . . the integral x dµ(x) is therefore defined and is an element of E 0 *
(Ch. III, §3, No. 1).”
Write iK : K → E for the canonical injection; then i K ∈ K (K; E) ⊂
R R R
f
K (K; E) and, by definition, iK dµ = iK (x) dµ(x) = x dµ(x) is the
unique element z ∈ E0 * such that, for all x0 ∈ E0 ,
Z Z
hz, x i = hiK (x), x i dµ(x) = hx, x0 i dµ(x)
0 0

Z
 
= (x0 K) dµ = µ(x0 K) = µ(x0 ◦ iK ) .
See also the note for III.33, `. 6–11.
§7 barycenters INT IV.x223

IV.101, `. 8, 9.
“Moreover, on K , the topology induced by the weak topology σ(E 0 *, E0 )
is identical with the original topology.”
As K ⊂ E , the assertion entails the identification of E as a linear
subspace of E0 * : for x ∈ E write x* for the linear form on E 0 defined by
x*(x0 ) = x0 (x) = hx, x0 i for all x0 ∈ E0 , and define the mapping (linear and
injective) θ : E → E0 * by θ(x) = x* (x ∈ E) .
The topology σ(E0 *, E0 ) on E0 * is the initial topology for the family
of ‘evaluation mappings’

(i) f 7→ f (x0 ) (f ∈ E0 *)

indexed by x0 ∈ E0 . The topology on θ(E) induced by σ(E 0 *, E0 ) is the

initial topology for the canonical injection θ(E) → E 0 * , which is, by the
‘transitivity of initial topologies’ (GT, I, §2, No. 3, Prop. 5), the initial
topology for the family of mappings

(ii) x* 7→ x*(x0 ) = hx, x0 i (x ∈ E)

indexed by x0 ∈ E0 —which is, in turn, the topology σ θ(E), E0 ; concisely,


σ(E0 *, E0 ) ∩ θ(E) = σ θ(E), E0 .

By ‘topology on E induced by σ(E0 *, E0 ) is meant the initial topology for the

mapping θ : E → E0 * , where E0 * is equipped with the topology σ(E0 *, E0 ) ,
that is (transitivity again), the initial topology for the family of composed
mappings

(iii) x 7→ x* 7→ x*(x0 ) = hx, x0 i (x ∈ E)

indexed by x0 ∈ E0 , which is precisely the weakened topology σ(E, E 0 )

on E ; after identifying E and θ(E) as vector spaces, we may write

σ(E0 *, E0 ) ∩ E = σ(E, E0 ) .

Thus the topology induced on E by σ(E 0 *, E0 ) is coarser (i.e., ‘weaker’) than

the original topology, therefore the topology on K induced by σ(E 0 *, E0 )
is coarser than its original topology; since K is compact for its original
topology, and Hausdorff for σ(E0 *, E0 ) ∩ K , the two topologies are equal
(GT, I, §9, No. 4, Cor. 3 of Th. 2).
{See also the note for III.38, `. 8.}
INT IV.x224 extension of a measure. l p spaces §7

IV.101, `. 9–12.
“ . . . if C is the closed convex envelope of K in E 0 * equipped with
σ(E *, E0 ) , then C ∩ E is the closed convex envelope of K in E for the
0

original topology (or for the weakened topology σ(E, E 0 ) ).”

Write T for the original topology on E , and T * for the topology
σ(E0 *, E0 ) . Let:
1) A be the convex envelope of K (in E ), and
2) B the closure of A in E for T ; thus B is the closed convex
envelope of K in E for T (TVS, II, §2, No. 6, Prop. 14), equivalently, for
the weakened topology σ(E, E0 ) (TVS, IV, §1, No. 2, Prop. 2).
As shown in the preceding note, viewing E as a linear subspace of E 0 * ,
the induced topology T * ∩ E coincides with σ(E, E 0 ) .
Now, C is the closure of A in E0 * for T * ; we are to show that
C ∩ E = B . From A ⊂ E ⊂ E0 * , we know that the closure of A in E for
T * ∩ E = σ(E, E0 ) is equal to C ∩ E (GT, I, §3, No. 1, Prop. 1), that is,
B = C∩ E.
IV.101, `. −15 to −13.
“Let µ be a discrete measure on K , positive and of total mass 1 ; it is
P
n
thus of the form µ = λi εxi , where xi ∈ K , and the λi are real numbers
i=1
P
n
such that λi > 0 for all i and λi = 1 . ”
i=1
Here, K can be any compact space (not necessarily a subspace of a
Banach space); let us write x ∈ K instead of x ∈ K .
To say that µ is discrete means that it is atomic (Ch.
P III, §1, No. 3,
Example I), given by a function α : K → C such that |α(x)| < +∞ ,
x∈K
and that, morever, the set

N = {x ∈ K : α(x) 6= 0 }

is finite (because K is compact and N = N ∩ K ). Say N = {x 1 , . . . , xn } ,

with the xi distinct. By the definition of µ , for f ∈ K (K; C) = C (K; C)
one has
X n
X n
X
µ(f ) = α(x)f (x) = α(xi )f (xi ) = α(xi )εxi (f ) ,
x∈K i=1 i=1

P
n
thus µ = λi εxi , where λi = α(xi ) .
i=1
There exist functions fi ∈ C (K) , 0 6 fi 6 1 , such that fi (xj ) = δij
for all i, j = 1, . . . , n , whence µ(fi ) = α(xi ) . The assumption that µ > 0
§7 barycenters INT IV.x225

is clearly equivalent to the condition λ i > 0 for all i (i.e., α(x) > 0 for all
x ∈ K ). Moreover, from

X
n 

|µ(f )| 6 λi kf k f ∈ C (K)
i=1

P
n
we know that µ is bounded, with kµk 6 λi ; since K is compact, we
i=1
have in fact
n
X n
X
kµk = |µ|(1) = µ(1) = λi εxi (1) = λi
i=1 i=1

(Ch. III, §1, No. 8, Cor. 2 of Prop. 10), and to say that µ has total mass
P
n
equal to 1 means that λi = 1 , so that µ is a convex combination of
i=1
Dirac measures.
IV.101, `. −12.
R P
“ bµ = x dµ(x) = λi xi .”
i
0 0
For all x ∈ E ,
Z

 R 
hbµ , x0 i = x dµ(x), x0 = hx, x0 i dµ(x) = µ(x0 K)
X  X 
= λi εxi (x0 K) = λi (x0 K)(xi )
i i
X X
X 
0
= λi x (xi ) = λi hxi , x0 i = λi xi , x 0 .
i i i

IV.101, `. −11.
“ x is the barycenter of the measure ε x .”
The discrete measure defined by α = ϕ{x} is µ = 1 · εx .
IV.101, `. −6, −5.
“This is nothing more than Prop. 5 of Ch. III, §3, No. 2 applied to the
canonical injection of K into E . ”
As in the note for `. 9–12, let us instead write B for the closed convex
envelope of K in E for the original topology T on E , reserving the letter C
for the closed convex envelope of K in E 0 * for topology T * = σ(E0 *, E0 ) as
in the sentence preceding Def. 1, where it is shown that C ∩ E = B (with E
canonically identified as a linear subspace of E 0 * , as in the note for `. 8,9).
INT IV.x226 extension of a measure. l p spaces §7

Write M1 for the set positive measures on K of total mass 1 , and set

S = {bµ : µ ∈ M1 } ,

which is a subset of E0 * (Def. 1); we are to show that E ∩ S = B .

Let f : K → E be the canonical injection; E and K being equipped
with their original topologies, we have f ∈ C (K; E) , f (x) = x for all x ∈ K ,
and, by Def. 1, R
S = { f (x) dµ(x) : µ ∈ M1 }.
By the cited Prop. 5, S is the closed convex envelope of f (K) = K in E 0 *
for the topology T * , that is, S = C ; thus E ∩ S = E ∩ C = B .
IV.101, `. −4 to −2.
“Corollary. — If the closed convex envelope C of K in E is compact,
then the barycenter of every positive measure of total mass 1 on K belongs
to E . ”
The statement can (should) be sharpened from “ . . . belongs to E ” to
“ . . . belongs to C ”. For, with notations and hypotheses of the Corollary as
stated in the text, if µ is a positive measure on K of total mass 1 , so that
bµ ∈ E by the Corollary, then bµ is a point of E that is the barycenter of
at least one positive measure on K of total mass 1 , therefore b µ belongs
to C by Prop. 1.
IV.101, `. −1 to IV.102, `. 2.
“For, C is then also the closed convex envelope of K in E 0 * equipped
with the weak topology σ(E0 *, E0 ) , and it suffices to apply, to the canonical
injection of K into E , Prop. 4 of Ch. III, §3, No. 2.”
As in the note for IV.101, `. −6, −5, let us instead write B for the closed
convex envelope of K in E for the original topology T on E , and let C
and S have the meanings in that note. Thus, writing A for the convex
envelope of K , B is the closure of A in E for T , and C is the closure
of A in E0 * for T * = σ(E0 *, E0 ) .
We are to show that B = C and S ⊂ E . (We show in fact that S = B .)
By assumption, B is compact for T , hence it is compact for the coarser
topology σ(E, E0 ) ; but σ(E, E0 ) coincides with the topology on E induced
by T * (see the note for IV.101, `. 8–9), therefore B is a compact subset
of E0 * for T * , and so B is closed in E0 * . Now, A ⊂ B and C is the
closure of A in E0 * , therefore C ⊂ B ; but B = C ∩ E ⊂ C , and so B = C .
Finally, S = C was shown in the preceding note, thus S = B ⊂ E .
IV.102, `. 4.
“ . . . or when E is quasi-complete.”
TVS, III, §1, No. 6, third paragraph after Def. 6.
§7 barycenters INT IV.x227

IV.102, `. 16. R
“Therefore sup hα (x) dµ(x) = sup hα (bµ ) = f (bµ ) ”
α α
R
In the expression hα (x) dµ(x) , by hα is meant the restriction

of hα to K ; and the second equality holds because 1) sup (hα K) = f ,
α
and 2) bµ ∈ K by the preceding Remark.
IV.102, `. 17, 18.
“When µ is a discrete positive measure on K of total mass 1 , Prop. 2
yields anew the inequality that defines the convex functions on K . ”
R+ is convex means that ifPx1 , . . . , xn are distinct
To say that f : K → P
elements of K , and x = λi xi with λi > 0 and λi = 1 , then
i i
X X

f (x) 6 λi f (xi ) = λi εxi (f ) ;
i i
P
one knows from the Example following Def. 1 that λi εxi is the general
i
discrete positive measure on K of total mass 1 , and that its barycenter
is x . To formulate this result in terms of Prop. 2 we will make use of the
following:
Remark. If K is a compact space and µ is a positive discrete measure
on K , then every lower semi-continuous function f : K → R + is integrable;
Pn R
and if µ = λi εxi as in the note for IV.101, `. −15 to −13, then f dµ =
i=1
P
n
λi f (xi ) .
i=1
Since f is measurable for any measure (§5, No. 5, Cor. of Prop. 8)
Proof. P
and µ∗ = λi (εxi )* (§1, No. 3, Prop. 15), one is reduced to the case that
i
µ = εx (§5, No. 6, Th. 5). Let H be the set of all functions g ∈ C + (K)
such that g 6 f ; by the Lemma of §1, No. 1,

f (x) = sup g(x) = sup εx (g) ,

g∈H g∈H

thus f (x) = (εx )*(f ) in the sense of loc. cit., Def. 1, hence also in the sense
R∗ R
of §1, No. 3, Def. 3. Thus f dεx = f (x) < +∞ , whence f dεx exists
(§4, No. 4, Cor. 1 of Prop. 5) and is equal to f (x) . ♦
Finally, with K as in Prop. 2, for a lower semi-continuous function
f > 0 to be convex, it is necessary and sufficient that it satisfy the inequality
of Prop. 2 for every positive discrete measure µ on K of total mass 1 .
Proof. Necessity: A special case of Prop. 2.
INT IV.x228 extension of a measure. l p spaces §7
R∗
Sufficiency: The assumption is that f (b µ ) 6 f dµ for every positive
discrete measure P µ on K of total mass 1 . If P x 1 , . . . , xn are distinct points
of K and if xP= λi xi with λi > 0 and λi = 1 , then x = bµ for the
measure µ = λi εxi by the Example following Def. 1, so by assumption
Z ∗
f (x) = f (bµ ) 6 f dµ ,

R P
whence, citing the above Remark, f (x) 6 f dµ = λi f (xi ) .
IV.102, `. −13.
“It suffices to note that inf g(x) = a is finite and apply Prop. 2 to
x∈K
g − a .”
The values of g are finite, and a = g(y) for some y ∈ K (GT, IV, §6,
No. 2, Th. 3).
The function f = g − a satisfies the hypotheses of Prop. 2, so
R∗
(g − a)(bµ ) 6 (g − a) dµ ; moreover, g − a is bounded and measurable,
and µ is bounded, therefore g − a is integrable (§5, No. 6, Th. 5) and
Z Z
g(bµ ) − a = (g − a)(bµ ) 6 (g − a) dµ = g dµ − a .

IV.102, `. −3 to IV.103, `. 1.
“For every point a ∈ K , there exists a closed convex neighborhood V a
of 0 in E such that

(2) |fi (y) − fi (a)| 6 δ/2

for 1 6 i 6 p and for every y ∈ Wa = K ∩ (a + Va ) . ”

Let T be the given topology on the locally convex space E . As V
varies over the set of all closed convex neighborhoods of 0 in E , the sets
W = K ∩ (a + V) vary over a fundamental system of neighborhoods of a
in K for the topology on K induced by T (TVS, II, §4, No. 1). Fix a ∈ K .
For each i = 1, . . . , p there exists, by the continuity of f i , a closed convex
neighborhood Vi of 0 in E such that (2) holds for all y ∈ W i = K∩(a+Vi ) .
T
p
Setting Va = Vi and
i=1

p
\ p
\  p
\ 
Wa = Wi = K ∩ (a + Vi ) = K ∩ a + Vi = K ∩ (a + Va ) ,
i=1 i=1 i=1

the inequality (2) holds for every i and every y ∈ W a .

§7 barycenters INT IV.x229

IV.103, `. 5–7.
“Each of the measures µj is positive, of total mass 1 , and its support
is contained in the compact convex set W aj ”
6 0 then µj (1) = α−1
If αj = −1
j µ(gj · 1) = αj αj = 1 , and, citing Ch. III,
§2, No. 3, Prop. 10,

Supp(µj ) = Supp(gj · µ) = {y ∈ Supp(µ) : gj (y) 6= 0 }

⊂ Supp(µ) ∩ Supp(gj ) ⊂ Supp(gj ) ⊂ Waj ;

whereas if αj = 0 then µj (1) = εaj (1) = 1 and

Supp(µj ) = Supp(εaj ) = {aj } ⊂ Waj .

IV.103, `. 7–9.
“ . . . by definition,

r
X
(3) µ= αj µj
j=1

since gj · µ = 0 if µ(gj ) = 0 ”
If αj = 0 , that is, if µ(gj ) = 0 , then, for every f ∈ C+ (K) with
0 6 f 6 1 , one has 0 6 gj f 6 gj , therefore 0 6 µ(gj f ) 6 µ(gj ) = 0 , thus
gj · µ = 0 , whence αj µj = 0 · εaj = 0 = gj · µ ; whereas if αj > 0 then
αj µj = gj · µ by the definition of µj . Thus

gj · µ = α j µj (j = 1, . . . , r) .

P
r P
r
Since gj = 1 , for every f ∈ C (K; C) one has f = gj f , therefore
j=1 j=1

r
X r
X r
X
µ(f ) = µ(gj f ) = (gj · µ)(f ) = (αj · µj )(f ) ,
j=1 j=1 j=1

whence (3).
IV.103, `. 11.
“Let xj be the barycenter of µj , which belongs to Waj (No. 1, Prop. 1) ”
At any rate, µj is a positive measure on K of total mass 1 , its barycen-
ter xj is defined, and xj ∈ K by the corollary of the cited Prop. 1 (since
the closed convex envelope of K in E is K itself).
INT IV.x230 extension of a measure. l p spaces §7

Let Kj = Supp µj . We know that Kj ⊂ Waj = K ∩ (aj + Vaj ) . Since

Kj is a compact subset of K , the induced measure ρ j = (µj )Kj on Kj is
defined: if h ∈ C (Kj ; C) and h0 is the extension
R 0 by 0 of h to K , then h0
is µj -integrable and one defines ρj (h) = h dµj . In particular, 1 ∈ C (Kj )
and 10 = ϕKj (the characteristic function of Kj in K ); since µj has total
mass 1 and K --- Kj is µj -negligible, one has
Z Z
ρj (1) = 10 dµj = ϕKj dµj = µj (Kj ) = µj (K) = 1 ,

thus ρj is a (positive) measure on Kj of total mass 1 . The barycenter bρj

of ρj is, a priori, an element of E0 * ; but Kj ⊂ Waj , where Waj is a
compact convex subset of E , thus the closed convex envelope C j of Kj
in E is a compact convex subset of Waj , hence of E . It then follows from
the Cor. of Prop. 1 (with Kj playing the role of K ) that bρj ∈ Cj ⊂ Waj ,
so it will suffice to show that bρj = xj , that is,

barycenter of (µj )Supp µj = barycenter of µj ;

and for this, it suffices to show that

(∗) hbρj , z0 i = hxj , z0 i for all z0 ∈ E0 .

Let z0 ∈ E0 . As the letter x is frozen in the statement of Prop. 3 (namely,

x = bµ ), we employ y for the variable of integration in the following:
DZ E
hxj , z0 i = y dµj (y), z0
Z Z
= hy, z0 i dµj (y) = hy, z0 iϕKj (y) dµj (y)

(all integrations are over K ), whereas if y 7→ hy, z 0 i0 (y ∈ K) denotes the

extension by 0 to K of the function y 7→ hy, z 0 i (y ∈ Kj ) , then
DZ E Z
0
hbρj , z i = 0
y dρj (y), z = hy, z0 i dρj (y)
Z Z
= hy, z0 i d(µj )Kj (y) = hy, z0 i0 dµj (y)
Z
= hy, z0 iϕKj (y) dµj (y)

(the first three integrations are over K j , the last two over K ), thus (∗) is
verified.
§7 barycenters INT IV.x231

IV.103, `. 13, 14.

P
r
“ . . . its barycenter is αj xj , which is also the barycenter of µ by
j=1
virtue of (3), thus is equal to x .”

For each z0 ∈ E0 , the function z0 K belongs to C (K; C) = K (K; C) ,
therefore, citing (3) at the third equality, we have
Z

hbµ , z i = hy, z0 i dµ(y) = µ(z0 K)
0

r
X r
X Z

0
= αj µj (z K) = αj hy, z0 i dµj (y)
j=1 j=1
Xr DX
r E
= αj hbµj , z0 i = α j b µj , z 0 ,
j=1 j=1

P
r
whence bµ = αj bµj . But x = bµ and xj = bµj by definition, thus
j=1
P
r P
r
x= αj xj . On the other hand bν = αj xj by the Example following
j=1 j=1
Def. 1 of No. 1, thus bν = x = bµ .
IV.103, `. 15, 16.
“ . . . since Supp(µj ) ⊂ Waj , |µj (fi ) − fi (aj )| 6 δ/2 for 1 6 i 6 p . ”
Writing Kj = Supp(µj ) , we have Kj ⊂ Waj ⊂ K , and K --- Kj is a
µj -negligible open set in K . Let ϕKj be the characteristic function of K j
in K . Since Kj ⊂ Waj , we know from (2) that the function f i − fi (aj ) · 1
in C (K; C) = K (K; C) satisfies

ϕKj |fi − fi (aj ) · 1| 6 δ/2 ,

whereas |fi − fi (aj ) · 1| = ϕKj |fi − f (aj ) · 1| µj -almost everywhere in K ,

therefore
Z
|µj (fi − fi (aj ) · 1)| 6 µj (|fi − fi (aj ) · 1|) = |fi − fi (aj ) · 1| dµj
Z
= ϕKj |fi − fi (aj ) · 1| dµj 6 (δ/2)µj (1) = δ/2 ,

that is, |µj (fi ) − fi (aj )| 6 δ/2 .

IV.103, `. 19–21.
“Since the αj are > 0 and have sum 1 , it follows from (3) and the
definition of ν that ν satisfies the inequality (1). ”
INT IV.x232 extension of a measure. l p spaces §7

Fix an index i (1 6 i 6 p) . We know that |µ j (fi ) − εxj (fi )| 6 δ for

all j = 1, . . . , r . Citing (3) and the definition of ν , we have

Xr r
X 
 
|µ(fi ) − ν(fi )| =  αj µj (fi ) − αj εxj (fi )
j=1 j=1
Xr 
 
= αj [µj (fi ) − εxj (fi )]
j=1
r
X r
X
6 αj |µj (fi ) − εxj (fi )| 6 αj δ = δ .
j=1 j=1

Thus ν satisfies the inequality (1).

Summarizing: If µ is a positive measure on K of total mass 1 , and x
is its barycenter, then every vague neighborhood of µ contains a discrete
positive measure ν of total mass 1 and barycenter x ; thus Prop. 3 is
verified.
IV.103, `. −5.
“Suppose x is an extremal point of K ”
The supposition would have implied that x belongs to K 0 even if this
had not been assumed in the statement of the Corollary (TVS, II, §7, No. 1,
Cor. of Prop. 2). However, the condition x ∈ K 0 is essential to stating the
Corollary: by εx is meant the measure on K0 defined by f 7→ f (x) for all
f ∈ C (K0 ) , whence the need for x ∈ K0 ; it then follows that the positive
measure εx on K0 of total mass 1 has barycenter x (No. 1, Example fol-
lowing Def. 1).
IV.103, `. −5 to −2.
“ . . . to prove that εx is the only positive measure on K0 , of total
mass 1 , having x as barycenter, it suffices, by Prop. 3, to see that the
set of discrete measures ν on K0 that are positive, of total mass 1 , and
have x as barycenter, reduces to εx . ”
Let us write M (x) for the set of all positive measures on K 0 , of total
mass 1 , having barycenter x ; we know that ε x ∈ M (x) (No. 1, Example
following Def. 1), and we are to show (assuming x is an extremal point
of K ) that M (x) = {εx } .
Let Md (x) be the set of measures in M (x) that are discrete (for
example εx ∈ Md (x) , as noted in the cited example). It will suffice to show
that Md (x) ⊂ {εx } ; for, since {εx } is vaguely closed, it will then follow
from Prop. 3 that M (x) ⊂ Md (x) ⊂ {εx } = {εx } (the overbar signifying
vague closure in M (K0 ; C) ), whence M (x) = {εx } .
§7 barycenters INT IV.x233

IV.104, `. 1, 2.
“ . . . the hypothesis that x is the barycenter of ν may be written
Pr
x= λi xi . ”
i=1
Recalling that the xi belong to K0 , the computation in the note for
P
r P
r
IV.103, `. 13, 14 yields bν = λi bεxi , that is, x = λi xi .
i=1 i=1

IV.104, `. 9.
“Then x is the barycenter of λµ0 + (1 − λ)µ00 . ”
Arguing as in the preceding note, bλµ0 +(1−λ)µ00 = λbµ0 + (1 − λ)bµ00 =
λx0 + (1 − λ)x00 = x .
IV.104, `. 9, 10.
“Therefore λµ0 + (1 − λ)µ00 = εx . ”
By the assumption on εx .
IV.104, `. 10.
“Therefore µ0 and µ00 are proportional to εx ”
Note first that Supp(εx ) = {x} . {At any rate, Supp(εx ) 6= ∅ since
εx 6= 0 , so it suffices to show that Supp(εx ) ⊂ {x} , that is, K0 − {x} ⊂
K0 − Supp(εx ) , in other words, the restriction of ε x to the open subset Y =
K0 --- {x} of K0 is zero (Ch. III, §2, No. 2, Def. 1). Indeed, if g ∈ K (Y; C)
and g 0 is the extension by 0 of g to K0 (i.e., by setting g 0 (x) = 0 ), then
(εx )Y (g) = εx (g 0 ) = g 0 (x) = 0 .}
From 0 6 λµ0 6 εx one then infers (loc. cit., Prop. 3)

Supp(µ0 ) = Supp(λµ0 ) ⊂ Supp(εx ) = {x} ,

therefore µ0 is proportional to εx (loc. cit., No. 4, Prop. 12), hence so is µ 00 .

And from µ0 (1) = εx (1) = µ00 (1) = 1 one infers that µ0 = εx = µ00 .
IV.104, `. 10, 11.
“ . . . whence x0 = x00 = x ”
For, x0 = bµ0 = bεx = x and similarly x00 = x .
IV.104, `. 18–20.
“ . . . the subset U of K × K × I formed by the triples (x, y, λ) such
that x 6= y and 0 < λ < 1 is open in K × K × I ”
If ∆ is the diagonal of K × K then U = (K × K --- ∆) × (0, 1) .
IV.104, `. −16.
“ K --- M = q(U) ”
INT IV.x234 extension of a measure. l p spaces §7

Let z ∈ K . Then

z is not extremal in K ⇔ z = λx + (1 − λ)y for some (x, y, λ) ∈ U ,

that is, z ∈
/ M ⇔ z ∈ q(U) .
IV.104, `. −6 to −4.
“By the Hahn-Banach theorem, for (a, b) to belong to S , it is necessary
and sufficient that h(a, b) > 0 for every continuous affine linear function h
on E × R such that h(x, u(x)) > 0 for x ∈ K .”
Review. A function f : E → R is affine linear if there exist a linear
form f0 on E and a real number α such that f (x) = f 0 (x) + α for all
x ∈ E (A, II, §9, No. 4, Prop. 6); f is continuous when f 0 ∈ E0 .
Thus, a function h : E × R → R is a continuous affine linear function
if there exist a continuous linear form h 0 ∈ (E × R)0 and an α ∈ R such
that h(x, t) = h0 (x, t) + α for all (x, t) ∈ E × R . But we can make the
identifications (E × R)0 = E0 ⊕ R0 = E0 ⊕ R (TVS, IV, §1, No. 5), so there
exist f0 ∈ E0 and β ∈ R such that h0 (x, t) = f0 (x) + βt , whence



h(x, t) = f0 (x) + βt + α = f0 (x) + α + βt ;

thus f (x) = f0 (x) + α defines a continuous affine linear function on E such

that h(x, t) = f (x) + βt for all (x, t) ∈ E × R . ♦
Necessity. Suppose (a, b) ∈ S (the closed convex envelope in E × R of
the graph G of u ). If h is a continuous affine linear function on E × R
such that h > 0 on G , then clearly h > 0 on S ; in particular, h(a, b) > 0 .
Sufficiency. Arguing contrapositively, assuming (a, b) ∈ / S let us show
that there exists a continuous affine linear function h on E × R such that
h(a, b) < 0 .
Since S is the intersection of the closed half-spaces that contain it (TVS,
II, §5, No. 3, Cor. 1 of Prop. 4), there exist a continuous linear form h 0 on
E × R and a real number α such that h0 (x, t) > α for (x, t) ∈ S and
h0 (a, b) < α . Then h = h0 − α is a continuous affine linear function on
E × R such that h > 0 on S (equivalently, h(x, u(x)) > 0 for all x ∈ K )
and h(a, b) < 0 .
IV.104, `. −3 to IV.105, `. 2.
“ . . . the relation (a, b) ∈ S is equivalent to the following property: the
relation

(4) f (x) > λ u(x) for all x ∈ K

implies

(5) f (a) > λb . ”

§7 barycenters INT IV.x235

Here f stands for any continuous affine linear function on E , and

λ ∈ R ; thus h(x, t) = f (x) − λt defines the most general continuous affine
linear function h on E × R , (4) signifies h(x, u(x)) > 0 for all x ∈ K ,
and (5) signifies h(a, b) > 0 . Thus the proposition


(a, b) ∈ S ⇔ (4) ⇒ (5)

is a restatement of the assertion just proved.

IV.105, `. 11.
“This proves the lemma . . . ”
The organization of the proof is somewhat complicated. With the no-
tations established in the lemma, we are to show:

(a, b) ∈ S ⇔ a ∈ K and u(a) 6 b 6 u(a) .

Proof of ⇒: Suppose (a, b) ∈ S . Since K is convex and closed, clearly

a ∈ K . Let h be any continuous affine linear function on E × R . We know
that

(∗) h > 0 on G ⇒ h(a, b) > 0 .

Say h is given by h(x, t) = f (x) − λt , where f is a continuous affine linear

function on E , and λ ∈ R .
case 1. λ = 0 .
We know from (∗) that

(i) f > 0 on K ⇒ f (a) > 0 .

Now, the closed half-spaces in E are the sets {x : f (x) > 0 } , where f is
a continuous affine linear function on E . Therefore the validity of (i) says
that a belongs to every closed half-space that contains K , so a belongs to
their intersection, which is equal to K (TVS, II, §5, No. 3, Cor. 1 of Prop. 4).
Conversely, if a ∈ K , then a belongs to every closed half-space con-
taining K , thus (i) is satisfied and so (∗) holds for every h with λ = 0 .
Conclusion: Assuming (a, b) ∈ S ,

(∗) holds when λ = 0 ⇔ a ∈ K .

case 2. λ = −1 .
For such h , the meaning of (∗) is

f (x) + t > 0 on G ⇒ f (a) + b > 0 ,

INT IV.x236 extension of a measure. l p spaces §7

that is,
f (x) + u(x) > 0 on K ⇒ f (a) + b > 0 ,
that is,
−f (x) 6 u(x) on K ⇒ − f (a) 6 b ;
since f 7→ −f is a permutation of the set of all continuous affine linear
functions on E , we may write this implication as

(ii) f 6 u on K ⇒ f (a) 6 b .

Now, u is the supremum of the f K  as f runs over the set of all continuous
affine linear functions such that f K 6 u (TVS, II, §5, No. 4, Prop. 5; see
the Remark at the end of this note); by (ii), f (a) 6 b for every such f ,
therefore u(a) 6 b .
Conversely, if a ∈ K (as is the case when (a, b) ∈ S , as noted above)
and u(a) 6 b , then f 6 u on K implies that f (a) 6 u(a) 6 b , so the
implication (ii) holds.
Conclusion: Assuming (a, b) ∈ S ,

(∗) holds when λ = −1 ⇔ u(a) 6 b .

case 3. λ = 1 .
For such h , the meaning of (∗) is

f (x) − t > 0 on G ⇒ f (a) − b > 0 ,

that is,
f (x) − u(x) > 0 on K ⇒ f (a) − b > 0 ,
that is,

(iii) u(x) 6 f (x) on K ⇒ b 6 f (a) .

Now, u is, by definition, the lower envelope of the f , as f runs

 over the
set of all continuous affine linear functions such that u 6 f K ; by (iii),
b 6 f (a) for every such f , therefore b 6 u(a) .
Conversely, if a ∈ K (as is the case when (a, b) ∈ S ) and b 6 u(a) ,
then
u 6 f on K ⇒ u 6 f on K
by the definition of u ; in particular, b 6 u(a) 6 f (a) , thus the implica-
tion (iii) holds.
§7 barycenters INT IV.x237

Conclusion: Assuming (a, b) ∈ S ,

(∗) holds when λ = 1 ⇔ b 6 u(a) .

Summary. If (a, b) ∈ S then a ∈ K and u(a) 6 b 6 u(a) . This

completes the “Proof of ⇒”. Moreover, the “Conversely” assertions show
that if a ∈ K and u(a) 6 b 6 u(a) , then the implications (i), (ii), (iii) hold.
Proof of ⇐: Assuming a ∈ K and u(a) 6 b 6 u(a) , we know from the
foregoing “Summary” that the implication (∗) holds in the special cases that
λ = 0, −1 or 1 in the formula for h . To infer that (a, b) ∈ S , it suffices to
show that (∗) holds for every h (i.e., with no restriction on λ ).
Let us write H(λ) for the set of all continuous affine linear functions h
on E × R of the form h(x, t) = f (x) − λt . Then
(
λ · H(1) when λ > 0
H(λ) =
|λ| · H(−1) when λ < 0

(because, for λ 6= 0 , f 7→ |λ|f is a permutation of the set of all continuous

affine linear functions on E ); thus the validity of (∗) for all λ ∈ R is a
consequence of its validity for λ = 0, 1 , and −1 . ♦
The proof of the cited Prop. 5 from TVS, II, §5, No. 4 is based on the
following:
Remark. If X is a topological space and f : X → R is lower semi-
continuous, then the function ϕ : X×R → R defined by ϕ(x, t) = f (x) − t
is lower semi-continuous.
Proof. Given a point (x0 , t0 ) ∈ X × R and a real number h ∈ R such
that ϕ(x0 , t0 ) > h , we seek a neighborhood of (x0 , t0 ) , say of the form
U × V , such that ϕ(x, t) > h on U × V (GT, IV, §6, No. 2, Def. 1).
We have f (x0 ) − t0 = ϕ(x0 , t0 ) > h , that is, f (x0 ) − (t0 + h) > 0 ;
choose ε so that 0 < ε < f (x0 ) − (t0 + h) , thus f (x0 ) > t0 + h + ε . Since f
is lower semi-continuous at x0 , there exists a neighborhood U of x0 such
that f (x) > t0 + h + ε for all x ∈ U .
Let V = ]t0 − ε, t0 + ε[ , so that for t ∈ V one has t < t0 + ε , hence
−t > −t0 − ε .
Then U×V is a neighborhood of (x0 , t0 ) and, for every (x, t) ∈ U×V ,

ϕ(x, t) = f (x) − t > (t0 + h + ε) − t > (t0 + h + ε) − t0 − ε = h ,

that is, ϕ > h on U × V . ♦

{The argument shows that if f is lower semi-continuous at a point
a ∈ X , then ϕ is lower semi-continuous at (a, b) for every b ∈ R .}
INT IV.x238 extension of a measure. l p spaces §7

IV.105, `. −14.

“ u(x) > u(y) + u(z) /2 ”
Proof #1. If f is a continuous affine linear function on E such that
f > u on K , then f > u on K by the definition of u , therefore

1 1 1 1
u(y) + u(z) 6 f (y) + f (z) = f (x) ,
2 2 2 2

and the assertion follows on taking the infimum of f (x) over all such f .

Proof #2. Since uK is the lower envelope of a family of concave (even
affine-linear) functions
 f K minorized by the finite-valued function u , it
follows that uK is concave (TVS, II, §2, No. 9). That is, if y, z ∈ K and
x = αy + (1 − α)z with 0 6 α 6 1 , then u(x) > αu(y) + (1 − α)u(z) .
IV.105, `. −11, −10.
“ . . . there exists, by Prop. 1 of No. 1, a positive measure ν on G , of
total mass 1 , having (a, u(a)) as barycenter.”
Recall that S is the closed convex envelope of the compact set G ; note
that Prop. 1 does not require that the closed convex envelope be compact.
IV.105, `. −6.

Z Z
(6) a= x dµ(x) and u(a) = u(x) dµ(x) .

The mapping ϕ : K → E × R defined by ϕ(x) = (x, u(x)) (x ∈ K)

is continuous and injective, and ϕ(K) = G (the graph of  u ); defining
p : G → K by p(x, u(x)) = x (x ∈ K) (that is, p = pr 1 G ), one sees
that ϕ effects a homeomorphism of K onto the subspace G of E × R , with
inverse mapping p .
If g ∈ C (K; R) , then g ◦ p ∈ C (G; R) and one defines µ(g) = ν(g ◦ p) ;
formally,
Z Z Z


(∗) g(x) dµ(x) = g ◦ p (x, u(x)) dν(x, u(x)) = g(x) dν(x, u(x)) .

On the other hand, if F is any Hausdorff locally convex space over R

and if g ∈ C (K; F) , then
Z Z
(∗∗) g dµ = (g ◦ p) dν (as elements of F0 * );
§7 barycenters INT IV.x239

for, if w0 ∈ F0 then
DZ E Z Z Z


0 0 0
g dµ, w = hg, w i dµ = (w ◦ g) dµ = (w0 ◦ g) ◦ p dν
Z Z DZ E
0

0
= w ◦ (g ◦ p) dν = hg ◦ p, w i dν = (g ◦ p) dν, w0 .

In particular, consider F = E × R and let g : K → E × R be the function

defined by
g(x) = (x, u(x)) (x ∈ K) ;
then (g◦p)(x, u(x)) = g(x) = (x, u(x)) , thus g◦p is the canonical injection
G → E × R , and (∗∗) yields the formula
Z Z
(†) (x, u(x)) dµ(x) = (x, u(x)) dν(x, u(x)) = bν = (a, u(a))

by the choice of ν . Let us calculate the left side of (†) as an element of

(E × R)0 * : for z0 + λ ∈ (E × R)0 = E0 ⊕ R one has
DZ E Z
(x, u(x)) dµ(x), z + λ = h(x, u(x)), z0 + λi dµ(x)
0

Z
 
hx, z0 i + λu(x) dµ(x)
Z Z
0
= hx, z i dµ(x) + λ u(x) dµ(x)
DZ E Z
= x dµ(x), z0 + λ u(x) dµ(x)
D Z Z  E
= x dµ(x), u(x) dµ(x) , z0 + λ ,

thus
Z Z Z 
(††) (x, u(x)) dµ(x) = x dµ(x), u(x) dµ(x) ;

R R 
from (†) and (††) we have x dµ(x), u(x) dµ(x) = (a, u(a)) , whence (6).

IV.105, `. −5, −4.

“The function u is upper semi-continuous and bounded on K ”
If f is a continuous
 affine linear
 function on E such that f > u on K ,
then f K > uK > u , thus uK is sandwiched between two continuous
functions on the compact space K , hence is bounded on K . Since the
INT IV.x240 extension of a measure. l p spaces §7

functions fK are upper semi-continuous (even continuous) so is their lower
envelope uK (GT, IV, §6, No. 2, Th. 4).
IV.105, `. −4, −3.
“ . . . hence is µ-integrable (§4, No. 4, Cor. 1 of Prop. 5) ”
Let c = inf u(x) . For every continuous affine linear function f on E
x∈K  
such that f > u on K , one has 0 6 u − c1K 6 f K − c1K , and uK − c1K
is the lower envelope of the positive continuous functions f K −  c1K hence
is upper semi-continuous (GT, IV, §6, No. 2, Th. 4); since 
 u K − c1K is
bounded and µ is a bounded measure,  it follows that µ*(u K − c1K ) < +∞
(§1, No. 3, Prop. 10), therefore 
u K − c1K is integrable (§4, No. 4, Cor. 1 of

Prop. 5), hence so is uK .
IV.105, `. −3, −2.
“ . . . the function −u is by definition convex ”
 
The convexity of −uK is immediate from the concavity of u K (see
the note for `. −14).
IV.106, `. 3, 4.
“ u(x) = u(x) almost everywhere for µ .”
 R R  R R 
From u 6 uK we have u dµ 6
R  (u K) dµ, whereas u dµ > (uK) dµ

by (8); thus uK − u > 0 and (uK − u) dµ = 0 , therefore uK − u = 0
µ-almost everywhere in K (§2, No. 3, Th. 1).
IV.106, `. 5.
“Theorem 1 will be proved once the following lemma has been estab-
lished:”
Let E and K be as in the statement of Th. 1 and assume Lemma 3
established, whose proof shows that there exists a continuous strictly convex
function u : K → R . Construct u K as in Lemma 1.
Fix any point a ∈ K and construct the measure µ on K as in the
discussion
 following the proof of Lemma 2; thus µ has barycenter a , and
u = uK µ-almost everywhere, that is, the set {x ∈ K : u(x) < u(x) } ,
which is µ-measurable (§5, No. 5, Prop. 8) hence µ-integrable, is µ-negligible:

µ({x ∈ K : u(x) < u(x) }) = 0 .

But, if M is the set of extremal points of K , we know from Lemma 2 that

K --- M ⊂ {x ∈ K : u(x) < u(x) } , therefore K --- M is µ-negligible.
Summarizing: Given any a ∈ K , there exists a positive measure µ
on K with total mass 1 and barycenter a , such that µ(K --- M) = 0 .
{In the language of Ch. V, §5, No. 7, Def. 4, µ is concentrated on the
set M of extremal points of K .}
§7 barycenters INT IV.x241

IV.106, `. 10.
“ . . . subspace A ”
Clearly a linear subspace of C (K; R) .
IV.106, `. 18.
“ . . . for this it suffices that hn (x) 6= hn (x0 ) ”
Let h be an affine linear function on K , and let x, x 0 be points of K
such that h(x) 6= h(x0 ) ; we are to show that h2 is strictly convex on the
segment [x, x0 ] = {λx + (1 − λ)x0 : 0 6 λ 6 1 } . That h2 is convex has
already been noted (TVS, II, §2, No. 8, Examples), and the computation in
the cited Examples show that if 0 < λ < 1 then


h2 λx + (1 − λ)x0 < λh2 (x) + (1 − λ)h2 (x0 ) .

Suppose y, y0 ∈ [x, 0 0

x ] 2with y 6= y , 2and0 let 0 < λ < 1 ; to prove that
2 0
h λy + (1 − λ)y < λh (y) + (1 − λ)h (y ) , it suffices by the foregoing to
show that h(y) 6= h(y0 ) . Say

y = ρx + (1 − ρ)x0 , y0 = σx + (1 − σ)x0 (0 6 ρ, σ 6 1) .

Assuming to the contrary that h(y) = h(y 0 ) , one has

ρh(x) + (1 − ρ)h(x0 ) = σh(x) + (1 − σ)h(x0 ) ,

whence (ρ − σ)h(x) = (ρ − σ)h(x0 ); since h(x) 6= h(x0 ), necessarily ρ − σ = 0,

contrary to y 6= y0 .
Implicit in the foregoing: Suppose E is a vector space over R , x and
x0 are distinct elements of E , [x, x0 ] = {λx + (1 − λ)x0 : 0 6 λ 6 1 } , and
h : [x, x0 ] → R is affine-linear. Then h(x) = h(x0 ) ⇔ h is constant on
[x, x0 ] , and h(x) 6= h(x0 ) ⇔ h is injective.
IV.106, `. −13 to `. −7.
“Corollary.”
Review of terminology. To say that C is a cone in E with ver-
tex 0 means that for every x ∈ C , x 6= 0 , C contains the set R + x =
{rx : r > 0 } , called a closed half-line originating at 0 (TVS, II, §2,
No. 4, paragraph following Def. 3); in other words, C is the union of a set
of half-lines originating at 0 . A convex cone C with vertex 0 is said to be
proper if it contains no line passing through 0 (loc. cit., sentence before
Prop. 9), in other words, if x ∈ C , x 6= 0 , then −x ∈ / C . A convex cone C
with vertex 0 is proper if and only if 0 is an extremal point of C (TVS, II,
§7, No. 2); for, to say that 0 is internal to a segment with distinct end-points
in C is equivalent to saying that C contains a line through 0 .
INT IV.x242 extension of a measure. l p spaces §7

Let C be a convex cone in E with vertex 0 , and let D ⊂ C be a closed

half-line originating at 0 . One says that D is an extremal generator
of C if, whenever I is an open segment in C that does not contain 0 but
intersects D , necessarily I ⊂ D .
Examples. (i) In E = R3 let
C1 = {(x, y, z) : x ∈ R, y ∈ R, z 6 0 } ,
C2 = {(x, y, z) : x ∈ R, y > 0, z 6 0 } ,
C3 = {(x, y, z) : x > 0, y > 0, z > 0 } ,
respectively: the points on or below the xy-plane; the points that are on or
below the xy-plane and are on or to the right of the xz-plane; the points of
the ‘first octant’.
C1 is a convex cone with vertex 0 (not proper) with no extremal gener-
ators; C2 is a convex cone with vertex 0 (not proper) whose only extremal
generators are the positive x-axis and the negative x-axis; C 3 is a proper
convex cone with vertex 0 , whose three extremal generators are the positive
coordinate axes, and whose only extremal point is (0, 0, 0) .
(ii) In a plane in R3 that does not pass through (0, 0, 0) , let K be
a circular disk, and let C be the set of all closed half-lines originating at
(0, 0, 0) and passing through some point of K ; then C is a proper convex
cone with vertex (0, 0, 0) whose only extremal generators are the half-lines
originating at (0, 0, 0) that pass through some point of the circumference
of K (i.e., some extremal point of K ), and whose only extremal point is
(0, 0, 0) .
If instead K is a convex polygon, then C is polyhedral (TVS, II, §7,
Exer. 24), its extremal generators are the closed half-lines originating at
(0, 0, 0) that pass through a vertex (i.e., extremal point) of the polygon, and
its only extremal point is (0, 0, 0) .
Let C be a convex set in a Hausdorff topological vector space E .
A nonempty compact convex subset A of C is called a cap of C if C --- K
is convex (TVS, II, §7, No. 2, Def. 3).
If a is an extremal point of C (if there are any!) then A = {a} is a
cap of C .
In the above examples: C1 contains no caps. For, assume to the con-
trary that A is a cap of C1 . If A contained a point P of the xy-plane, then
C1 --- A would contain two points Q and R on the xy-plane such that P is
internal to the segment QR , contradicting the convexity of C 1 --- A . Thus
A is contained in the interior z < 0 of C 1 . Let P be any point of the xy-
plane and let Q be a point of A . Prolong PQ to a point R in the interior
of C1 that does not belong to A . Then Q belongs to the segment PR in
the convex set C1 --- A , contrary to Q ∈ A .
§7 barycenters INT IV.x243

Similarly C2 contains no caps.

Slice C3 into two parts by a plane, not containing (0, 0, 0) , that inter-
sects all three positive coordinate axes—a bounded closed convex part A and
an unbounded convex part C3 --- A (so to speak, A is a compact ‘truncate’
of C3 ). Then A is a cap of C3 , and C3 is obviously the union of all of its
caps (see also TVS, II, §7, No. 2, Prop. 5). Other caps of C 3 : the segments
OP , where O = (0, 0, 0) and P is a point on one of the positive coordi-
nate axes (so to speak, a bounded truncate of a positive coordinate axis);
a triangular area OPQ , where P and Q are points on different positive
coordinate axes (a truncate of the ‘first quadrant’ of one of the coordinate
planes).
IV.106, `. −6.
“For, x belongs to a cap K of C (TVS, II, §7, No. 2, Prop. 5) ”
Write τw = σ(E, E0 ) for the weakened topology on E . The cited
Prop. 5, applied in the weak space (E, τ w ) , asserts that C is the union
of its caps K , which are convex sets compact for τ w . In particular, x be-
longs to such a K .
IV.106, `. −5.
“ M ∩ K contains the set of extremal points of K (loc. cit., Cor. 1 of
Prop. 4).”
Let B be the set of extremal points of K . Since C is proper, 0 is
extremal in C , hence also in K ; thus 0 ∈ B . By the cited Cor. 1 (applied
in the weak space (E, τw ) of the preceding note),

B = {0} ∪ { certain points of M } ;

since 0 ∈ M , we see that B ⊂ M , thus B ⊂ M ∩ K .

IV.106, `. −4.
“It then suffices to apply Th. 1.”
As noted above, K is a convex subset of C that is compact for the
weakened topology τw = σ(E, E0 ) , and by assumption C is metrizable for
the topology induced by τw , therefore so is K ; applying Th. 1 to the metriz-
able compact convex subset K of (E, τ w ) , there exists a positive measure
λ on (K, τw ∩ K) , of total mass 1 , such that x is the barycenter of λ and
λ(K --- B) = 0 , where B is the set of extremal points of K . As observed in
the preceding note, B ⊂ M ∩ K , therefore also λ*(K --- M ∩ K) = 0 .
IV.107, `. 9, 10.
“ . . . the second assertion follows at once from the definitions and (9) ”
INT IV.x244 extension of a measure. l p spaces §7

At issue is the continuity of the mapping i H : X → H 0 for the topology

σ(H 0 , H ) on H 0 . For every f ∈ H , the mapping

x 7→ hf, iH (x)i = f (x)

is continuous, and σ(H 0 , H ) is the initial topology for the family of linear
forms
α 7→ α(f ) = hf, αi (α ∈ H 0 )
indexed by the functions f ∈ H (TVS, II, §6, No. 2, Def. 2), whence the
assertion (GT, I, §2, No. 3, Prop. 4).
In more direct terms, given any x0 ∈ X , the continuity of iH at x0
may be seen as follows. A basic neighborhood V of i H (x0 ) in H 0 is given
by


V = {α ∈ H 0 : |α(fk ) − iH (x0 ) (fk )| < ε for k = 1, . . . , n } ,

where f1 , . . . , fn ∈ H and ε > 0 . Then

−1
iH (V) = {x ∈ X : iH (x) ∈ V }
= {x ∈ X : |(iH (x))(fk ) − (iH (x0 ))(fk )| < ε for k = 1, . . . , n }
= {x ∈ X : |fk (x) − fk (x0 )| < ε for k = 1, . . . , n }
\n
= {x ∈ X : |fk (x) − fk (x0 )| < ε } ,
k=1

which is a neighborhood of x0 in X by the continuity of the fk .

IV.107, `. −11 to −9.
“ . . . and the assertion follows from the fact that H 0 , equipped with the
weak topology σ(H 0 , H ) , is quasi-complete (TVS, III, §4, No. 2, Cor. 5 of
Th. 1).”
In such a space, the closed convex envelope of a compact set is compact
(TVS, III, §1, No. 6, sentence before Prop. 9; see also TVS, IV, §5, No. 5,
Th. 3). For a simpler proof of the compactness of C , one may cite the weak
compactness of the closed unit ball of the dual of a normed space (TVS, III,
§3, No. 4, Cor. 3 of Prop. 4).
IV.107, `. −6.
“ . . . the Dirac measure εiH (x) arising from εx . ”



The generic function g ∈ C iH (X) = K iH (X) has the form
g = f 0 , where f ∈ C (X) = K (X) and


f 0 iH (z) = f (z) for all z ∈ X .
§7 barycenters INT IV.x245

Let λ be a measure on X . The corresponding measure µ on i H (X) is

defined by
µ(f 0 ) = λ(f ) for all f ∈ C (X) .
In particular, if λ = εx then, for all f ∈ C (X) ,


µ(f 0 ) = εx (f ) = f (x) = f 0 iH (x) = εiH (x) (f 0 ) ,
thus µ = εiH (x) .
IV.107, `. −6 to −3.
“To say that µ admits iH (x) as barycenter means, by definition, that
Z
hh, iH (z)i dλ(z) = hh, iH (x)i
X

for every function h ∈ H . ”

I did not succeed in working through the assertion as stated, but have
persuaded myself that the following

reformulation is correct: assuming
λ ∈ M (X) and µ ∈ M iH (X) are paired as in the preceding notes,
Z
(∗) bµ = iH (x) ⇔ h(x) = h dλ for all h ∈ H ,

that is, µ has barycenter iH (x) if and only if λ satisfies (10) for every
h∈H .
The setting for discussing the barycenter of µ is as follows. We have
a Hausdorff locally convex space E = H 0 , where H ⊂ C (X) is a normed
space, and E is equipped with the weak topology σ(H 0 , H ) ; then E0 =
(H 0 )0 = H (TVS, II, §6, No. 2, Prop. 3). We have a compact subspace
K = iH (X) of E , whose closed convex envelope C in E is compact by
part (i) of the present Proposition. Thus the measure µ on K is eligible to
have a barycenter bµ ∈ E0 * = (H 0 )0 * = H * (No. 1, Def. 1), and in fact
bµ ∈ C ⊂ E = H 0 (No. 1, Prop. 1 and its Corollary), characterized by the
property
DZ E Z Z
bµ (h) = hbµ , hi = α dµ(α), h = hα, hi dµ(α) = α(h) dµ(α)
K K K

for all h ∈ H = E0 . Defining ĥ : K → R by ĥ(α) = α(h) for all

α ∈ K ⊂ E = H 0 , one has ĥ ∈ C (K) because α 7→ α(h) = hα, hi is the
restriction to K of one of the linear forms defining the topology σ(H 0 , H )
on E . The corresponding function ĥ0 ∈ C (X) is defined by



ĥ0 (z) = ĥ iH (z) = iH (z) (h) = h(z) (by (9))
for all z ∈ X . Continuing the above calculation,
INT IV.x246 extension of a measure. l p spaces §7

Z Z
hbµ , hi = α(h) dµ(α) = ĥ(α) dµ(α)
K K
Z Z Z
0
= ĥ (z) dλ(z) = h(z) dλ(z) = h dλ .
X X
Therefore
bµ = iH (x) ⇔ hbµ , hi = hiH (x), hi for all h ∈ H
Z


⇔ h dλ = iH (x) (h) = h(x) for all h ∈ H ,

whence (∗).
While the notations are at hand, let us complete the proof of (ii). Con-
sider the statements
(a) iH (x) is an extremal point of of C ;
(b) εx is
the only positive measure λ on X satisfying the condition (10)
λ(h) = h(x) for all h ∈ H .
We are to show that (a) ⇔ (b).
Proof of (a) ⇒ (b). The measure λ = εx trivially satisfies (10) for
every h ∈ H . Assuming (a), suppose λ is any positive measure on X
satisfying (10) for every h ∈ H (in particular, 1 X ∈ H , and λ(1X ) =
1X (x) = 1 , so λ has total mass 1 ). Let µ be the corresponding measure on
iH (X) , which is also positive and of total mass 1 . We know from (∗) that
bµ = iH (x) ; but iH (x) is extremal in C by the assumption (a), therefore
µ = εiH (x) by the Corollary of Prop. 3 of No. 2. Since the measure on X
corresponding to εiH (x) is εx , we conclude that λ = εx , whence (b).
Proof of (b) ⇒ (a). Assume (b). To prove (a), it suffices to verify that
iH (x) satisfies the conditions of the above-cited Corollary. At any rate,
εiH (x) is a positive measure on iH (X) , of total mass 1 , whose barycenter
is iH (x) (No. 1, Example). On the other hand, suppose µ is a positive
measure on iH (X) , of total mass 1 , such that bµ = iH (x) , and let λ
be the corresponding measure on X . By (∗), λ satisfies (10) for every
h ∈ H , therefore, by the assumption (b), λ = ε x . As the measure on
iH (X) corresponding to εx is εiH (x) , we conclude that µ = εiH (x) . Thus
the conditions of the Corollary are satisfied, whence (a).
Remarks. 1. If λ is any measure on the compact space X , its
 restriction
to the normed space H is a continuous linear form, and kλ H k 6 kλk .
The positive measures on X are the positive linear forms on K (X) = C (X)
(Ch. III, §1, No. 5, Th. 1), and kλk = λ(1) for all such measures λ (§4,
No. 7, Prop. 12); since 1 ∈ H and k1k = 1 , one has
 
kλH k > |(λH )(1)| = λ(1) = kλk ,

thus kλH k = kλk = λ(1) when λ > 0 .
§7 barycenters INT IV.x247

0
 Let us write Φ : M (X) → H for the mapping defined by Φ(λ) =
λH . From kΦ(λ)k 6 kλk we see that Φ is a continuous linear mapping
between Banach spaces, and kΦk 6 1 ; in fact kΦk = 1 since, if λ is
any nonzero positive measure on X of  norm 1 (for example, any Dirac
measure εx ) then kΦk > kΦ(λ)k = kλH k = kλk = 1 .
2. If x ∈ X , λ is a positive measure on X of total mass 1 , and µ is
the corresponding measure on iH (X) , the equivalence (∗) can be expressed
as  
λH = εx H ⇔ bµ = iH (x) .
3. Writing M+ (X) for the set of all positive measures on X , the
assertion (ii) of Prop. 4, for a point x ∈ X , can be expressed as

(ii) {λ ∈ M+ (X) : Φ(λ) = εx H } = {εx } ⇔ iH (x) is extremal in C .

If Φ is injective then the condition on the left in (ii) holds for every x ∈ X ,
and so iH (X) is precisely the set of all extremal points of C ; this is the case
if H is dense in C (X) —for example, if f ∈ H ⇒ |f | ∈ H (M.H. Stone’s
theorem, GT, X, §4, No. 1, Th. 2) or if f, g ∈ H ⇒ f g ∈ H (loc. cit.,
No. 4, Prop. 6).
The case that H = C (X) has been taken up in §4, No. 8, Prop. 15
and is continued in Ch. VI, §1, No. 6, Remark 1). A (real) character of
a commutative algebra A over R with unity is an algebra epimorphism
A → R ; the characters of C (X) are precisely the ε x (Gillman and Jerison,
Rings of continuous functions, p. 57, item 4.9, Van Nostrand, Princeton,
N.J., 1960).
5. In a vague sense, iH (X) is a ‘linearization’ of the compact space X ,
‘tailored’ to the linear subspace H of C (X) , placing X in a structurally
richer context (topological vector spaces) than that of topological spaces.
Prop. 4 reformulates a property of a point x ∈ X with respect to H in
terms of measures on X ; measures on i H (X) play only an auxiliary role
in the proof. The property is reformulated in Prop. 6 in topological terms,
with measures playing a role only in the proof; this theme culminates in the
theorem of Errett Bishop (No. 5, Th. 2), where measures are nowhere in
sight. In Prop. 8 and in Choquet’s theorem (No. 6, Th. 3), it is measures
that are in the forefront.
All in all, Prop. 4 is a subtle, far-reaching result whose cunning remains
a mystery to me.
IV.107, `. −3 to −1.
“ . . . the assertion (ii) is just the translation of the criterion of No. 2,
Cor. of Prop. 3 for iH (x) to be an extremal point of C .”
See the preceding note.
INT IV.x248 extension of a measure. l p spaces §7

IV.108, `. 8, 9.
“ the weakly closed hyperplane of H 0 with equation hh, t0 i = hh, iH (x)i
is a support hyperplane of iH (X) .”
Order of events: Fix h ∈ H and let x be a point of X where h
attains its supremum. The linear form t 0 7→ hh, t0 i (t0 ∈ H 0 ) is continuous
for σ(H 0 , H ) ; the set H = {t0 ∈ H 0 : hh, t0 i = hh, iH (x)i } is a closed
hyperplane in H 0 , the relation

hh hh, iH (z) 6 hh, iH (x)i for all z ∈ X ii

expresses that the points of iH (X) lie on the same side of H , and since
H contains at least the point iH (x) of iH (X) , it is a support hyperplane
of iH (X) (TVS, II, §5, No. 2, Def. 3). It follows that H is also a support
hyperplane of the closed convex envelope C of i H (X) in H 0 , therefore H
contains some extremal point t00 of C (loc. cit., §7, No. 1, Cor. of Prop. 1).
Necessarily t00 ∈ iH (X) (loc. cit., Cor. of Prop. 2), say t 00 = iH (y) ; then y
is by definition H -extremal, and

h(y) = hh, iH (y)i = hh, t00 i = hh, iH (x)i = h(x)

is the supremum of h (the third equality, because t 00 ∈ H ).

IV.108, `. 20–24.
“ . . . it is known (TVS, II, §3, No. 1, Prop. 1) that the infimum of the
numbers λ(f ) , for all the positive measures on X such that λ(h) = h(x)
for every function h ∈ H , is equal to the supremum of the numbers h(x) ,
where h runs over the set of functions h ∈ H such that h 6 f . ”
I should have said EVT instead of TVS (explanation below).
In the language of the cited Prop. 1, we are talking about the num-
ber α0 ; the notational ‘table of concordance’ between Prop. 1 and its appli-
cation here is as follows:
E ↔ C (X; R)
V ↔ H

f ↔ εx H (a positive linear form on H )
 
Sf ↔ {λ ∈ M+ (X) : λH = εx H } (Ch. III, §1, No. 5, Th. 1)
 
h ∈ Sf ↔ λ ∈ M+ (X) and λH = εx H
a ∈ E ↔ f ∈ C (X; R)

α0 ↔ sup (εx H )(h) = sup h(x)
h∈H , h6f h∈H , h6f

α00 ↔ inf (εx H )(h) = inf h(x) .
h∈H , f 6h h∈H , f 6h
§7 barycenters INT IV.x249

The assertion of the cited Prop. 1 in EVT (p. EVT II.22) is that

(∗) {h(a) : h ∈ Sf } = [α0 , α00 ] ,

whence
α0 = inf [α0 , α00 ] = inf{h(a) : h ∈ Sf } .
Translated to the present context, this says that
 
sup h(x) = inf{λ(f ) : λ ∈ M+ (X) and λH = εx H } ,
h∈H , h6f

as asserted.
Incidentally, the infimum in the assertion is attained: there exists a
λ0 ∈ S  such that λ0 (f ) = sup h(x) , but λ0 depends on f (be-
εx H
h∈H , h6f
cause α0 does) as well as on x . Similarly, there exists a λ 00 ∈ S  such
εx H
00
that λ (f ) = inf h(x) .
h∈H , h>f
{This argument was perfectly clear to me when I first studied it in 1974;
it is lifted verbatim from my notes at the time. On rereading the argument
when preparing these notes, with TVS at my elbow, I found that I no longer
understood it. After several days of struggle, it occurred to me to look at
the French original, and all was clear again.
The problem: In TVS, instead of the conclusion (∗), one finds the con-
clusion
{h(a) : h ∈ Sf } ⊂ [α0 , α00 ] ,
which (i) follows at once from the fact that if y, z ∈ V with z 6 a 6 y , and
h ∈ Sf , then
f (z) = h(z) 6 h(a) 6 h(y) = f (y) ,
whence α0 6 h(a) 6 α00 , and (ii) is of no help in proving (∗). The problem
did not arise in 1974, as I was working from the 2nd French edition of Chs. I
and II of EVT.}
IV.108, `. 24–26.
“Suppose that x is H -extremal; it then follows from Prop. 4, (ii) that
for every function f ∈ C (X; R) ,

(11) f (x) = sup h(x) . ”

h∈H , h6f

Since x is H -extremal, by
 the cited Prop. 4 the only positive mea-
sure λ on X that extends εx H is λ = εx ; thus, in the notation of the
INT IV.x250 extension of a measure. l p spaces §7

preceding note, S  = {εx } . It follows that, for every f ∈ C (X; R) ,

εx H
[α0 , α00 ] = {εx (f )} , whence α0 = α00 = f (x) . Thus, if x is H -extremal,
then
f (x) = sup h(x) = inf h(x)
h∈H , h6f h∈H , h>f

for every f ∈ C (X; R) .

IV.108, `. −1.
“ λ({x}) = inf λ(U) > 1 − ε ”
U
By §1, No. 4, Prop. 19.
IV.109, `. 1.
“ . . . therefore λ({x}) = 1 . ”
Obviously λ({x}) > 1 ; but λ has total mass 1 since it satisfies (10),
therefore λ({x}) 6 1 .
IV.108, `. 2.
“ . . . necessarily λ = εx ”
From λ(X --- {x}) = λ(X) --- λ({x}) = 1−1 = 0, one has Supp(λ) ⊂ {x}
(§2, No. 2, Prop. 5), therefore Supp(λ) = {x} (because λ 6= 0 ); since
λ(1) = 1 , it follows that λ = εx (Ch. III, §2, No. 4, Prop. 12).
IV.109, `. 12, 13.
“The condition a) signifies that G contains the set of extremal points
of C . ”
Recall that C is defined in Prop. 4 to be the closed convex envelope of
iH (X) in H 0 (the dual of the normed subspace H of C (X) , equipped
with the weak topology σ(H 0 , H ) ), and shown to be a compact subset
of H 0 . It follows that the set Cep of extremal points of C is contained in
the compact subset iH (X) of H 0 (TVS, II, §7, No. 1, Cor. of Prop. 2).
By definition, ChH (X) is the set of all x ∈ X that satisfy the condition
−1
in (ii) of Prop. 4, that is, such that i H (x) ∈ Cep ; thus ChH (X) = iH (Cep ) .
Since F is closed in X and ŠH (X) is the closure of ChH (X) in X , one
has


F ⊃ ŠH (X) ⇔ F ⊃ ChH (X) ⇔ iH (F) ⊃ iH ChH (X) ,

that is, ⇔ G ⊃ Cep .

IV.109, `. 13, 14.
“The condition b) signifies that G meets the intersection of i H (X)
with each of the closed support hyperplanes of i H (X) . ”
§7 barycenters INT IV.x251

A (weakly) continuous linear form on H 0 is a function z0 7→ hh, z0 i

( z0 ∈ H 0 ) defined by an element h ∈ H , and if α ∈ R the set

{z0 ∈ H 0 : hh, z0 i = α }

is a typical closed hyperplane in H 0 ; since X is compact, there exists a

point y ∈ X such that sup h(x) = h(y) , that is,
x∈X

suphh, iH (x)i = hh, iH (y)i ,

x∈X

and since hh, iH (x)i 6 hh, iH (y)i for all x ∈ X , the set

(∗) H = {z0 ∈ H 0 : hh, z0 i = hh, iH (y)i }

is a typical closed support hyperplane of i H (X) .

Note that a closed hyperplane H in H 0 is supporting for iH (X) if
and only if it is supporting for its closed convex envelope C . For, it is clear
that iH (X) lies to one side of H if and only if C does, so it suffices to
observe that H ∩ iH (X) 6= ∅ ⇔ H ∩ C 6= ∅ , and this is clear from the fact
that if H is supporting for C then H ∩ C contains at least one extremal
point of C (TVS, II, §7, No. 1, Cor. of Prop. 1) and every extremal point
of C belongs to iH (X) (loc. cit., Cor. of Prop. 2).
Condition b) says: If h ∈ H and if

α = suphh, iH (x)i = sup h(x) ,

x∈X x∈X

then there exists a point y ∈ F such that hh, i H (y)i = α ; that is, every
h ∈ H attains its supremum on iH (X) at some point of iH (F) = G .
This means: If H is any closed support hyperplane of i H (X) , then
there exists a point y ∈ F such that iH (y) ∈ H , in other words G∩H 6= ∅ .
Since G ⊂ iH (X) , G ∩ H = G ∩ iH (X) ∩ H , so it is equivalent to say (as
in the text) that G intersects iH (X) ∩ H .
IV.109, `. 15, 16.
“ . . . the condition c) signifies that every point of i H (X) is the barycen-
ter of a measure with support contained in G ”
Reviewing the notations employed in the proof of Prop. 4, let us write
λ for a measure on the compact space X , and µ for the corresponding
measure on the subspace iH (X) of H 0 homeomorphic to X , defined by

µ(f 0 ) = λ(f ) for all f ∈ C (X) ,

INT IV.x252 extension of a measure. l p spaces §7


where f 0 iH (x) = f (x) for all x ∈ X (see the note for p. IV.107, `. −6).
Writing Z(f ) = {x ∈ X : f (x) = 0 } (the ‘zero-set’ of f ), from

x ∈ Z(f ) ⇔ iH (x) ∈ Z(f 0 )

we see that
x ∈ Supp(f ) ⇔ iH (x) ∈ Supp(f 0 ) , whence Supp(f 0) =

iH Supp(f ) . It follows easily that Supp µ = iH (Supp λ) (remarks fol-
lowing Def. 1 of Ch. III, §2, No. 2).
Now, iH (X) is a compact subset of the Hausdorff locally convex space
H 0 ; if λ and µ are paired as above, the barycenter of µ is the element b µ
of (H 0 )0 * = H * such that
Z
(∗) hbµ , hi = h dλ for all h ∈ H

(see the note for p. IV.107, `. −6 to −3).

For consistency with these notations, let us restate condition c) as fol-
lows (replace the letter λ by µ ):
c*) For every point x ∈ X , there exists a positive
R measure λ of total
mass 1 on X , such that Supp(λ) ⊂ F and h(x) = h dλ for every func-
tion h ∈ H .
Then, if µ is the measure on iH (X) corresponding to λ , the condition
Supp λ ⊂ F is equivalent to iH (Supp λ) ⊂ iH (F) , that is, to Supp µ ⊂ G .
Thus, the condition c*) says that for every point x ∈ X , there exists a
positive measure µ on iH (X) of total mass 1 , such that Supp µ ⊂ G and
Z
hiH (x), hi = h dλ for all h ∈ H ,

that is, in view of (*), bµ = iH (x) .

To summarize, condition c) is equivalent to the following condition: for
every point iH (x) of iH (X) , there exists a positive measure µ on i H (X)
of total mass 1 , such that Supp µ ⊂ G and b µ = iH (x) .
IV.109, `. 16–18.
“ . . . by No. 1, Prop. 1, this is also equivalent to saying that the closed
convex envelope of iH (X) is equal to the closed convex envelope of G .”
The closed convex envelope of iH (X) in H 0 is C ; denote by D the
closed convex envelope of G = iH (F) . Since iH (X) is compact, C is equal
to the set of all barycenters bµ of positive measures µ on iH (X) of total
mass 1 (No. 1, Prop. 1); similarly, D is the set of all barycenters b ν of
positive measures ν on G of total mass 1 . We are to show that

condition c) holds ⇔ C = D .
§7 barycenters INT IV.x253

Proof of ⇐ : Given a point iH (x) ∈ iH (X) it suffices, by the last

paragraph of the preceding note, to show that there exists a positive measure
µ on iH (X) of total mass 1 , such that Supp µ ⊂ G and b µ = iH (x) .
Since iH (x) ∈ C = D , there exists a positive measure ν on G of total
mass 1 such that bν = iH (x) . Let µ be the measure on iH (X) defined
by

(†) µ(f ) = ν(f G) for f ∈ C (iH (X)) ,

which is also positive and, since µ(1) = ν(1 G) = 1 , of total mass 1 .
If f ∈ C (iH (X)) and Supp(f ) ⊂ iH (X) --- G , then f G = 0 , hence
µ(f ) = 0 ; this is true in particular for the functions f ∈ C (i H (X)) that are
extensions by 0 of functions g ∈ K (iH (X) --- G) , therefore the restriction
of µ to the open subset iH (X) --- G of iH (X) is equal to 0 (Ch. III, §2,
No. 1), and so
iH (X) --- G ⊂ iH (X) --- Supp µ
(loc. cit., No. 2, Def. 1), that is, Supp µ ⊂ G .
It remains to show that bµ = iH (x) , that is, bµ = bν ; given h ∈ H ,
it suffices to show that hbµ , hi = hbν , hi .
Let f ∈ C (iH (X)) be the function

f (iH (y)) = hiH (y), hi (y ∈ X) .


Then f G ∈ C (G) is the function

iH (y) 7→ hiH (y), hi (y ∈ F) .

By the definition of barycenter (No. 1, Def. 1),

Z
hbµ , hi = hz0 , hi dµ(z0 ) = µ(f ) ,
iH (X)

whereas Z

hbν , hi = hz0 , hi dν(z0 ) = ν(f G) ,
G

whence hbµ , hi = µ(f ) = ν(f G) = hbν , hi .
Proof of ⇒ : The foregoing computations show that if ν is a positive
measure on G of total mass 1 , then the measure µ on i H (X) defined
by (†) is a positive measure of total mass 1 such that b µ = bν ; in particular,
bν = bµ ∈ C (No. 1, Prop. 1), and since D is equal to the set of all such b ν
(same Prop. 1), one has D ⊂ C .
INT IV.x254 extension of a measure. l p spaces §7

Assuming that c) holds, we are to show that C ⊂ D . Since D is a

closed convex set and C is the closed convex envelope of i H (X) , it will
suffice to show that iH (X) ⊂ D .
Given x ∈ X , we are to show that iH (x) ∈ D . By c), iH (x) = bµ for
some positive measure µ on iH (X) of total mass 1 such that Supp µ ⊂ G .
Let ν = µG be the restriction of µ to the compact subspace G of i H (X) ,
in the sense of §5, No. 7, Def. 4:
Z Z
g dν = g 0 dµ for all g ∈ C (G) ,
G iH (X)

where g 0 is the extension by 0 of g to iH (X) . Note that µ(G) = 1 ; for,

iH (X) --- G ⊂ iH (X) --- Supp µ ,
thus the open subset iH (X) --- G of iH (X) is µ-negligible, therefore µ(G) =
µ(iH (X)) = 1 . It follows that if g = 1 (the function on G identically equal
to 1 ) then g 0 = ϕG , therefore
Z
ν(1) = ϕG dµ = µ(G) = 1 ,
iH (X)

thus ν is a positive measure on G of total mass 1 ; consequently b ν ∈ D

(see the Note for No. 1, Cor. of Prop. 1).
To show that bµ ∈ D , it will suffice to show that bµ = bν (the argument
in the proof of D ⊂ C does not apply here, for the present measure µ is
not defined by the formula (†)); given h ∈ H , it suffices to show that
hbµ , hi = hbν , hi . By the definition of bµ ,
Z
hbµ , hi = hz0 , hi dµ(z0 ) ;
iH (X)

the integrand is the function f ∈ C (i H (X)) defined by

f (iH (y)) = hiH (y), hi for y ∈ X ,
thus, since ϕG = 1 µ-almost everywhere,
Z Z
hbµ , hi = f dµ = ϕG f dµ .
iH (X) iH (X)

But (ϕG f )(iH (y)) is equal to f (iH (y)) when iH (y) ∈ G , and to 0 when
iH (y) ∈
/ G , thus ϕG f is the extension by 0 of the function g ∈ C (G)
defined by
g(iH (y)) = hiH (y), hi for y ∈ F ,
that is, ϕG f = g 0 . Thus
§7 barycenters INT IV.x255

Z Z Z
0
hbµ , hi = ϕG f dµ = g dµ = g dν
iH (X) iH (X) G
Z
= hz0 , hi dν(z0 ) = hbν , hi .
G

We extract from the foregoing argument the following observation (a

similar argument will be employed below in the proof of Prop. 8):
If G is any closed subset of iH (X) , and µ is a positive measure on
iH (X) of total mass 1 such that Supp µ ⊂ G , then the restriction ν = µ G
of ν to G is also a positive measure of total mass 1 and b ν = bµ .
IV.109, `. 18, 19.
“The equivalence of the conditions a), b) and c) therefore follows from
TVS, II, §7, No. 1, Cor. of Prop. 2.”
In the cited Corollary, E is a Hausdorff locally convex space, A is
a compact convex subset of E , and K is a compact subset of A . The
dictionary for translating the Corollary to the present setting is as follows:
0
E ↔ H (equipped with σ(H 0 , H ) )
A ↔ C (the closed convex envelope of i H (X) )
K ↔ G ( = iH (F) , F a closed set in X )
K ⊂ A ↔ G ⊂ C.

Let us write a0 ), b0 ), c0 ) for the (equivalent) conditions of the cited Corollary:

a0 ) the closed convex envelope of K is equal to A ;
b0 ) K meets the intersection of A with any closed support hyperplane
of A ;
c0 ) K contains every extremal point of A .
The translation of a0 ): the closed convex envelope of G is equal to C ,
i.e, in the notation of the preceding Notes, D = C ; in other words, c) holds.
The translation of b0 ): G meets the intersection of C with any closed
support hyperplane of C ; in other words, b) holds.
The translation of c0 ): G contains every extremal point of C ; equiva-
lently, a) holds.
Thus, the equivalence of a0 ), b0 ), c0 ) implies the equivalence of a), b), c),
whence Prop. 7.
IV.109, `. −15, −14.
“This is the translation of Th. 1 of No. 2, by transport of structure by
means of the homeomorphism x 7→ iH (x) , as in Prop. 5.”
Probably intended: “as in Prop. 4.”
INT IV.x256 extension of a measure. l p spaces §7

The dictionary between Choquet’s theorem (No. 2, Th. 1) and its ap-
plication here is as follows:
0
E ↔ H (equipped with σ(H 0 , H ) )
K ↔ C (the closed convex envelope of i H (X) )
M ↔ Cep (the set of extremal points of C ) .


One knows that iH ChH (X) = Cep ⊂ iH (X) ⊂ C (see the note for
IV.109, `. 12, 13).
To apply Choquet’s theorem, we must verify that C is a metrizable
subspace of H 0 . As noted in the proof of Prop. 4, (i), i H (X) is contained
in the set
B = {z0 ∈ H 0 : kz0 k 6 1 } ,
which is compact for σ(H 0 , H ) (TVS, III, §3, No. 4, Cor. 3 of Prop. 4),
whence C is contained in B and is compact. Moreover, since the com-
pact space X is assumed here to be metrizable, the Banach space C (X)
is separable, i.e., of countable type (GT, X, §3, No. 3, Th. 1), therefore its
linear subspace H is a normed space of countable type; it follows that B
is metrizable for σ(H 0 , H ) (TVS, loc. cit., Cor. 2 of Prop. 6, read “second
axiom of countability”), therefore so is C .
For consistency with the notations in the proof of Prop. 4, it is useful
to replace µ by λ in the statement of Prop. 8: Given any x ∈ X , we
seek
a positive measure
R λ on X of total mass 1 such that λ X --- ChH (X) = 0
and h(x) = h dλ for all h ∈ H .
By Choquet’s theorem, Cep is the intersection of a sequence of open
sets in C (i.e., is a Gδ in C ), hence is a Borel set in C (§5, No. 4, Cor. 3
of Th. 2); and, since iH (x) ∈ C , there exists a positive measure ρ on C of
total mass 1 such that bρ = iH (x) and ρ(C --- Cep ) = 0 . From
Cep ⊂ iH (X) ⊂ C


we have ρ C --- iH (X) 6 ρ(C --- Cep ) = 0 , thus the open set C --- iH (X)
in C is ρ-negligible, whence C --- iH (X) ⊂ C --- Supp ρ (Ch. III, §2, No. 2,
Def. 1), that is, Supp ρ ⊂ iH (X) .
Let µ = ρiH (X) be the restriction of ρ to the compact subset i H (X)
of C ; thus,
µ(g) = ρ(g 0 ) for all g ∈ C (iH (X)) ,
where g 0 is the extension by 0 of g to C (§5, No. 7, Def. 4). In particular,
if g = 1 then g 0 = ϕiH (X) (the characteristic function of the subset i H (X)
of C ), and since g 0 = 1 ρ-almost everywhere, one has
Z
µ(1) = ϕiH (X) dρ = ρ(1) = 1 ,
C
§7 barycenters INT IV.x257

whence µ is a positive measure on iH (X) of total mass 1 . Note that, since

iH (X) --- Cep ⊂ C --- Cep is ρ-negligible,

it is also µ-negligible (§5, No. 7,
Lemma 2), that is, µ iH (X) --- Cep = 0 .
We assert that bµ = bρ ( = iH (x) ); given h ∈ H , it suffices to show
that hbµ , hi = hbρ , hi . By the definition of bρ ,
Z
hbρ , hi = hz0 , hi dρ(z0 ) .
C

The integrand is the function f ∈ C (C) defined by

f (z0 ) = hz0 , hi (z0 ∈ C) ;
since ϕiH (X) = 1 ρ-almost everywhere, one has
Z Z
hbρ , hi = f dρ = ϕiH (X) f dρ ,
C C


where ϕiH (X) f (z0 ) is equal to f (z0 ) = hz0 , hi if z0 ∈ iH (X) , and to 0 if
z0 ∈
/ iH (X) . On the other hand,
Z Z
hbµ , hi = hz0 , hi dµ(z0 ) = g dµ ,
iH (X) iH (X)

where g ∈ C (iH (X)) is defined by

g(iH (y)) = hiH (y), hi = h(y) for y ∈ X ;
since g 0 is the ρ-integrable function such that g 0 (z0 ) is equal to hz0 , hi if
z0 ∈ iH (X) , and to 0 if z0 ∈/ iH (X) , one has g 0 = ϕiH (X) f . Therefore
Z Z Z
0
hbρ , hi = ϕiH (X) f dρ = g dρ = g dµ = hbµ , hi .
C C iH (X)

Thus bµ = bρ = iH (x) .
Let λ be the positive measure on X of total mass 1 that is paired
with µ as in the proof of Prop. 4, (ii), so that µ is derived from λ via the
homoemorphism iH of X onto iH (X) . Since bµ = iH (x) , we know that
Z
h(x) = h dλ for all h ∈ H
X

(see the note for

IV.107, `. −6 to −3). It remains only to show
that
λ X --- ChH (X) = 0 , and this is immediate from µ iH (X) --- Cep = 0
and the fact that



iH X --- ChH (X) = iH (X) --- iH ChH (X) = iH (X) --- Cep .
INT IV.x258 extension of a measure. l p spaces §7

IV.109, `. −5, −4.

“It then follows easily from Props. 5 and 7 that Ch H (X) = ŠH (X) =
S2 .”
The dictionary between Prop. 7 and its application here:

E ↔ R3
X ↔ X = {x ∈ R3 : kxk 6 1 }
F ↔ S2 = {x ∈ R3 : kxk = 1 }
H ↔ H ⊂ C (X) as described here.

Recall that, on a finite-dimensional real vector space, all compatible

topologies coincide (TVS, I, §2, No. 3, Th. 2) and all hyperplanes are closed
(loc. cit., Cor. 1 of Th. 2); in particular, the word “weak” can be omitted
from the present discussion, and (R 3 )0 may be identified with R3 equipped
with its norm topology. Thus X is a compact convex subset of R 3 , with
boundary S2 , and interior equal to {x ∈ R3 : kxk < 1 } .
{‘Intuitively’, S2 is the set of extremal points of X , but let us play by
the rules—no pictures, and only internal references.}
For h ∈ H , write Mh = {x ∈ X : h(x) = sup h(y) } . If h is
y∈X
nonconstant, then by hypothesis Mh ⊂ S2 , whereas if h is constant then
Mh = X ; in either case Mh ∩ S2 6= ∅ , therefore

(∗) ŠH (X) ⊂ S2

by Prop. 7 (specifically, b) ⇒ a) ).
On the other hand, if x ∈ S2 then there exists a support (i.e., ‘tangent’)
hyperplane H of X that contains x (TVS, §5, No. 2, Prop. 3). Say

H = {y ∈ R3 : f (y) = c } ,

where c ∈ R and f ∈ (R3 )0 ⊂ C (R3 ) , and f (y) 6 c for all y ∈ X . In

particular, x ∈ H ∩ S2 . In fact, H ∩ S2 = {x} . {For, if also y ∈ H ∩ S2 ,
1 1
then the point z = x + y also belongs to the convex set H ∩ X . Since X
2 2
lies to one side of H , H ∩ X cannot contain an interior point of X , therefore
kzk = 1 . Then
1 1 1 
1 = kzk2 = + + hxyi ,
4 4 2

whence hxyi = 1 = kxk kyk , and so y = ±x . As y = −x would imply
z = 0 , we are left with y = x .}
§7 barycenters INT IV.x259

By assumption, H contains the affine linear function h = f X , which
attains its supremum c at x ; and since h can attain its supremum only at
points of H , necessarily Mh = H ∩ X . By Prop. 5, Mh ∩ ChH (X) 6= ∅ ; but
ChH (X) ⊂ ŠH (X) ⊂ S2 by (*), therefore

Mh ∩ ChH (X) ⊂ H ∩ S2 = {x} ,

whence Mh ∩ ChH (X) = {x} and so x ∈ ChH (X) . Thus S2 ⊂ ChH (X) ,
and finally
S2 ⊂ ChH (X) ⊂ ŠH (X) ⊂ S2 ,
whence equality throughout.
IV.110, `. 13–15.
“The Hr -extremal points in X are again called H -extremal, the set
of them is denoted ChH (X) , and the closure of the latter set is denoted
ŠH (X) .”
Thus ChH (X) = ChHr (X) and

ŠH (X) = ChH (X) = ChHr (X) = ŠHr (X) .

Note that the concept of H -extremal point does not depend on the existence
of measures, but is characterized in terms of measures in Prop. 4. If x ∈ X
and if λ is a positive measure on X of total mass 1 , the conditions
Z
(10) h(x) = h dλ for all h ∈ Hr

and
Z
0
(10) f (x) = f dλ for all f ∈ H

are obviously equivalent, and if µ is the corresponding measure on i Hr (X)

then (see the note for IV.107, `. −6 to −3)

bµ = iHr (x) ⇔ the condition (10) holds

⇔ the condition (10)0 holds,

and
x ∈ ChH (X) ⇔ x ∈ ChHr (X) (by definition)
⇔ (10) holds only for λ = εx (Prop. 4)
⇔ (10)0 holds only for λ = εx .
INT IV.x260 extension of a measure. l p spaces §7

Inasmuch as the concept of H -extremal point refers back to H r , it is not

necessary here to define and develop the properties of i H (X) .
IV.110, `. −12.
“ . . . whence |g(b)| = 1 since |g(b)| 6 1 ”



In fact, g(b) = 1 ; for, 1 = R g(b) 6 |g(b)| 6 1 , whence I g(b) = 0 .
The Proposition is remarkable in that |f | is not required to belong
to H . In particular, with notations as in Prop. 5, if h ∈ H then |h|
attains its supremum at at least one H -extremal point.
IV.111, `. 3.
“The fact that a) implies b) follows from Prop. 9.”
If f ∈ H then, by Prop. 9, |f | attains its supremum at some point
b ∈ ChH (X) ⊂ ŠH (X) , and b ∈ F by a), therefore b) holds.
IV.111, `. 4, 5.
“this is a matter of seeing that if b) is verified, then, for every h ∈ H r ,
F intersects the set of points where h attains its infimum in X . ”
Suppose established that every h ∈ H r attains its infimum at some
point of F . If h ∈ Hr then also −h ∈ Hr ; by supposition, −h attains its
infimum at some point a ∈ F , whence


sup h = − inf(−h) = − (−h)(a) = h(a) .

Then F ⊃ ŠHr (X) = ŠH (X) by “b) ⇒ a)” of Prop. 7, thus a) of the present
proposition holds.
IV.111, `. 14, 15.
“Since g − b ∈ H , the hypothesis on F implies that |g − b| 6 b ”
By the hypothesis on F , |g − b| attains its supremum at some point
a ∈ F ; but |g − b| is 6 b at every point of F , therefore for every x ∈ X
one has
|g(x) − b| 6 sup |g − b| = |g(a) − b| 6 b ,

that is, |g − b| 6 b .
Indeed, since g − b ∈ Hr , one can replace b) by the weaker condition
b)r For every g ∈ Hr , F intersects the set of points of X where |g|
attains its supremum.
IV.111, `. −13 to −11.
“ . . . a point where |f | attains its supremum is a point where one of the
functions f, −f attains its supremum.”
§7 barycenters INT IV.x261

If f is a real function (continuous or not) such that |f | attains its

supremum at a point a , then either f or −f attains its supremum at a .
For, if f (a) > 0 then sup f = f (a) because, for all x ∈ X ,

f (x) 6 |f (x)| 6 |f |(a) = f (a) ;

whereas if −f (a) > 0 then, since | − f | = |f | attains its supremum at a ,

the foregoing yields sup(−f ) = (−f )(a) .
On the other hand, if f attains its supremum at a , and −f attains its
supremum at b , then |f | attains its supremum at one of these points. For,
let M = sup f = f (a) and m = inf f = − sup(−f ) = −(−f )(b) = f (b) . If
f is constant, then |f | assumes its supremum at every point. Assume f
nonconstant, i.e., m < M .
If m < M 6 0 then sup |f | = −m = −f (b) .
If m 6 0 6 M then sup |f | = max{−m, M } = max{−f (b), f (a)} .
If 0 6 m < M then sup |f | = M = f (a) .
In all cases, sup |f | = max{|m|, |M |} = max{|f (b)|, |f (a)|} , thus |f |
attains its supremum at either a or b .
IV.111, `. −11 to −9.
“For a vector space H of continuous real functions satisfying the hy-
potheses of No. 3, the Props. 9 and 10 are thus trivial corollaries of Props. 5
and 7, respectively.”
Call ‘Prop. 10r ’ the result of stating Prop. 10 with H consisting of
real functions, and denote its three conditions by a) r , b)r , c)r . Let a), b),
c) have their meanings as in Prop. 7.
Condition a)r is identical to condition a) of Prop. 7.
Condition c)r is identical to condition c) of Prop. 7.
b) ⇒ b)r : If f ∈ H then, by b), f attains its supremum at some point
of F ; but also −f ∈ H , so −f attains its supremum at some point of F ,
therefore |f | attains its supremum at one of these points (see the preceding
note), whence b)r .
b)r ⇒ b): As observed in the note for IV.111, `. 14, 15, b) r implies a),
and a) implies b) by Prop. 7.
Thus b) ⇔ b)r .
The equivalence of a), b), c) (Prop. 7) then assures the equivalence of
a)r , b)r , c)r , thus Prop. 10r is proved.
Call Prop. 9r the result of stating Prop. 9 with H consisting of real
functions. Let f ∈ H . By Prop. 5, f attains its supremum at some point
a ∈ ChH (X) . Since also −f ∈ H , −f attains its supremum at some point
b ∈ ChH (X) . As noted above, |f | attains its supremum at one of a, b , thus
Prop. 9r is proved.
INT IV.x262 extension of a measure. l p spaces §7

IV.111, `. −7 to −1.
“Lemma 4.”
So to speak, if a ∈ X is an “approximate peak point” and has a count-
able neighborhood base, then a is a “peak point” (relative to H ).
{For the terminology, cf. the book of G.M. Leibowitz (Lectures on com-
plex function algebras, p. 54, Scott, Foresman, Glenview, IL, 1970).}
As indicated by the “resp. C (X; R) ” in the statement of the lemma,
the proof works for either the real or the complex case.
IV.112, `. 2, 3.
“ . . . let λ, µ, ε be numbers such that

0 < λ < 1, 1 < µ < µ + ε 6 1 + λ.”

The order of events: Fix λ with 0 < λ < 1 . Fix µ with 1 < µ < 1+λ .
Then choose any ε > 0 (which will also remain fixed) such that µ+ε 6 1+λ ,
for example ε = (1 + λ) − µ .
IV.112, `. 4–6.
“We are going to define, by induction on n (n > 1), a decreasing
sequence (Un ) of open neighborhoods of a such that U n ⊂ Vn for all n ,
and a sequence (hn ) of functions in H ”
To get a feeling for the intricate argument, I found it necessary to look
at n = 1, 2, 3 .
One is assuming that

(∗) 0 < λ < 1, 1 < µ < µ+ε 6 1+λ ,

whence
λ 1
(†) 0< < < 1.
µ µ
◦
The argument for n = 1. Define U1 = V1 , set U = U1 , and apply
the hypothesis to the inequalities (†), that is, with c = λ/µ , d = 1/µ and
U = U1 : there exists a function f ∈ H such that
1 λ
(121 ) |f | 6 1 , |f (a)| > , |f (x)| 6 for all x ∈ X --- U1 .
µ µ
1
Define h1 = f . Then
f (a)

|f | 1
(131 ) |h1 | = 6 6 µ;
|f (a)| |f (a)|
§7 barycenters INT IV.x263

f (a)
(141 ) h1 (a) = = 1;
f (a)

|f (x)| λ 1 λ
(151 ) for all x ∈ X --- U1 , |h1 (x)| = 6 · 6 · µ = λ;
|f (a)| µ |f (a)| µ
2
X
1
(161 ) for all y ∈ X , |λ h1 (y)| = λ|h1 (y)| 6 λµ < λ(1 + λ) = λj .
j=1

With the convention U0 = X , the condition U1 ⊂ U0 ∩ V1 is fulfilled.

P
2−1
The argument for n = 2. The function λh 1 (formally equal to λj hj )
j=1
is continuous and takes the value λ at a , therefore there exists an open
neighborhood U2 of a , which we can suppose to be contained in U 1 ∩ V2 ,
such that

(172 ) |λh1 (y)| < λ + ελ2 for all y ∈ U2 .

Apply the hypothesis to (†) with U = U 2 to obtain a (new) function f ∈ H

such that
1 λ
(122 ) |f | 6 1 , |f (a)| > , |f (x)| 6 for all x ∈ X --- U2 .
µ µ
1
The function h2 = f then satisfies (132 ), (142 ) and (152 ). Set
f (a)
2
X
g= λj hj = λh1 + λ2 h2 ;
j=1

we have |g| 6 λ|h1 | + λ2 |h2 | , and to prove (162 ) we must show that |g(y)| <
P3
λ3 for all y ∈ X . At any rate, by (161 ) we know that
j=1

2
X
|g| < λj + λ2 |h2 | on X ;
j=1

if y ∈ X --- U2 then |h2 (y)| 6 λ by (152 ), so

2
X 3
X
j 2
|g(y)| < λ +λ ·λ = λj ;
j=1 j=1
INT IV.x264 extension of a measure. l p spaces §7

whereas if y ∈ U2 then |h2 (y)| 6 µ by (132 ) and, by (172 ), |λh1 (y)| <
λ + ελ2 , therefore

|g(y)| < (λ + ελ2 ) + µλ2

= λ + (µ + ε)λ2
3
X
6 λ + (1 + λ)λ2 = λj (by (∗))
j=1

thus (162 ) is verified.

P
2
The argument for n = 3. The function λj hj is continuous and
j=1
P
2
takes the value λj at a , therefore there exists an open neighborhood
j=1
U3 of a , which we can suppose to be contained in U 2 ∩ V3 , such that
X2 
 
(173 )  λj hj (y) < λ + λ2 + ελ3 for all y ∈ U3 .
j=1

Apply the hypothesis to (†) with U = U 3 to obtain a function f ∈ H such

that
1 λ
(123 ) |f | 6 1 , |f (a)| > , |f (x)| 6 for all x ∈ X --- U3 .
µ µ
1
The function h3 = f then satisfies (133 ), (143 ) and (153 ). Set
f (a)
3
X 2
X
j
g= λ hj = λj hj + λ 3 h3 .
j=1 j=1

P
4
To prove (163 ) we must show that |g(y)| < λj for all y ∈ X . At any
j=1
rate, by (162 ) we know that
3
X
|g| < λj + λ3 |h3 | on X ;
j=1

if y ∈ X --- U3 then |h3 (y)| 6 λ by (153 ), so

3
X 4
X
j 3
|g(y)| < λ +λ ·λ= λj ;
j=1 j=1
§7 barycenters INT IV.x265

whereas if y ∈ U3 then |h3 (y)| 6 µ by (133 ) and, by (173 ),

X2 
 j 
 λ h j (y)  < λ + λ2 + ελ3 ,
j=1

therefore

|g(y)| = (λ + λ2 + ελ3 ) + λ3 µ
= λ + λ2 + (µ + ε)λ3
4
X
< λ + λ2 + (1 + λ)λ3 = λj (by (∗))
j=1

thus (163 ) is verified.

IV.113, `. 1.
“ . . . for x ∈ X --- Un , we have |hp (x)| 6 λ for 1 6 p 6 n ”
A mirage, not justified by the n inequalities

(15p ) |hp (x)| 6 λ for x ∈ X --- Up

(1 6 p 6 n), the direction of the inclusions

X --- Un ⊃ X --- Un−1 ⊃ · · · ⊃ X --- U1

being unfavorable for x ∈ X --- Un . An alternate path to (16n ) is indicated

in the preceding note.
IV.113, `. 8, 9.
“ . . . if x 6= a , there exists an integer n such that x ∈ / U n+1 ; therefore
|hn+k (x)| 6 λ for all k > 1 ”
T
∞ T
∞
Because Uj ⊂ Vj = {a} ; and, for all k > 1 ,
j=1 j=1

x ∈ X --- Un+1 ⊂ X --- Un+k ,

whence |hn+k (x)| 6 λ by (15n+k ).

IV.113, `. 13–20.
“Theorem 2”
With notations as in No. 4, the theorem says that, when H is a norm-
closed subalgebra of C (X; C) that contains the constants and separates
INT IV.x266 extension of a measure. l p spaces §7

the points of X , a is a ‘peak point’ of X (relative to H ) if and only if

a ∈ ChH (X) and a has a countable neighborhood base.
The analogous theorem for H ⊂ C (X; R) is also true, as observed in
subsequent notes.
{Caution: Hr need not be a subalgebra of C (X; C) . For example,
if X is the closed unit disk {z : |z| 6 1 } of C , and H is the norm-
closure in C (X; C) of the algebra of complex polynomial functions on X ,
then every f ∈ H is differentiable in the open disk {z : |z| < 1 } , hence
the function u(x, y) = Rf (x + iy) is harmonic there ( u xx + uyy = 0 ); the
function z 7→ x = R(z) ( z ∈ X ) belongs to H r , but its square z 7→ x2
cannot be the real part of a function in H . This example of X and H is
worked out in detail in the book of Leibowitz (pp. 55-64) cited in the note
for IV.111, `. −7 to −1.}
IV.113, `. −10, −9.
“ . . . by Prop. 9 of No. 4, a is an A -extremal point.”
Because a is the only point where |f | attains its supremum.
{The cited Prop. 9 is also valid when H ⊂ C (X; R) (No. 4, Remark ).}
IV.113, `. −3.
“ . . . set ε = log d/ log c ”
Not to be confused with the ε in the proof of the cited lemma.
IV.113, `. −2, −1.
“Since a is an Ar -extremal point, there exists a function g ∈ A such
that



R(g) > 0 , R g(a) 6 ε , R g(x) > 1 for x ∈ X --- U

(No. 3, Prop. 6, b)).”

We are given 0 < c < d < 1 and an open neighborhood U of a ; the
objective is to construct a function f satisfying the condition (12) of No. 5,
Lemma 4.
The cited Prop. 6—which does not require H ( = A r ) to be norm-
closed in C (X; R) nor that it be an algebra—proves the existence of a
function h ∈ Ar having the indicated properties, and h = Rg for some
g∈A .
{The argument remains valid when A ⊂ C (X; R) , with the obvious
simplifications.}
IV.114, `. 1, 2.
“Set f = cg ; since f is the sum of the normally convergent series
P
∞
(log c)n g n /n! , we have f ∈ A ”
n=0
§7 barycenters INT IV.x267

Rather than explain the meaning of c g , it is simpler to define the

function f by the formula
X∞
[(log c)g(x)]n
f (x) = exp[(log c)g(x)] = ;
n=0
n!

1
the series is normally convergent since kgk < +∞ , | log c| = | − log | =
c
1
log and
c

n  1
n
X∞
| log c| kgk
∞
X log kgk 1  1 kgk
= c = e(log c ) kgk = < +∞,
n=0
n! n=0
n! c

Pm [(log c)g]n
and f ∈ A because the partial sums belong to A and A is
n=0 n!
norm-closed.

IV.114, `. 3.
“ |f | 6 1 , |f (a)| > cε = d , |f (x)| 6 c for x ∈ X --- U . ”
Write g = h + ik , where h, k ∈ Ar ; thus h = Rg > 0 . For every
x ∈ X,

f (x) = exp[(log c)g(x)] = exp[(log c)h(x) + i(log c)k(x)]

= exp[(log c)h(x)] · exp[i(log c)k(x)] ;

the first factor is positive and 6 1 since (log c)h(x) 6 0 , and the second
factor has absolute value 1 since (log c)k(x) is real, therefore |f (x)| 6 1 .
Since Rg > 0 and h(a) = Rg(a) 6 ε , one has 0 6 h(a) 6 ε , whereas
log c < 0 , therefore (log c)h(a) > (log c)ε ; citing the formula for f (x) as a
product,
 log d 
|f (a)| = exp[(log c)h(a)] > exp[(log c)ε] = exp (log c) · = d.
log c
Finally, if x ∈ X --- U then h(x) = Rg(x) > 1 , whereas log c < 0 ,
therefore (log c)h(x) 6 log c and

|f (x)| = exp[(log c)h(x)] 6 exp(log c) = c .

{The argument remains valid when A ⊂ C (X; R) , with simplifications:

Ar = A , h = g , k disappears from the stage, and f (x) = exp[(log c)g(x)] =
cg(x) . Thus Theorem 2 remains valid when A ⊂ C (X; R) .}
INT IV.x268 extension of a measure. l p spaces §7

IV.114, `. 9.
“ c) Let M be the set of subsets M of X such that . . . ”
If M ∈ M and M0 ⊃ M then obviously M0 ∈ M ; thus minimal sets M
are of interest. By No. 4, Prop. 9, Ch A (X) ∈ M . The implication a)T⇒ c)
says that ChA (X) ⊂ M for every M ∈ M , therefore ChA (X) ⊂ M,
M∈M
and
T the reverse inclusion is immediate from Ch A (X) ∈ M ; thus ChA (X) =
M , ChA (X) is the smallest element of M , and
M∈M

M = {A ⊂ X : A ⊃ ChA (X) } .
In particular, ŠA (X) ∈ M .
IV.114, `. 12.
“ d) Let N be the set of subsets N of X such that . . . ”
Such a set N is called a boundary for the algebra A (cf. G.M. Leib-
owitz, op. cit., p. 53, Exer. 4). If N ∈ N and N 0 ⊃ N , then N0 ∈ N . By
Prop. 5 of No. 3 applied to Ar , ChA (X) = ChAr (X) ∈ N ; arguing as in
the preceding note, Tone sees that the meaning of the implication a) ⇒ d) is
that ChA (X) = N , ChA (X) is the smallest element of N , and
N∈N

N = {A ⊂ X : A ⊃ ChA (X) } ,
which is also equal to M . In particular, ŠA (X) ∈ N .
One calls ChA (X) the Choquet boundary, and ŠA (X) the Shilov bound-
ary, for A (Leibowitz, op. cit., p. 49). Another valuable reference for §7 is
the book of Robert R. Phelps, Lectures on Choquet’s theorem, Van Nostrand,
Princeton, NJ, 1966.
IV.114, `. −15, −14.
“ . . . we can restrict ourselves to the case that X does not reduce to the
single point a ”
When X = {a} , each of a), c), d) (resp. b)) is trivially (resp. vacuously)
true.
IV.115, `. 8, 9.
“ . . . the spaces X1 and X2 are homeomorphic, both being bounded
convex sets in R4 with nonempty interior.”
A convex body in a topological vector space over R or C is a closed
convex set with nonempty interior. The theorem “Any two compact convex
bodies in Rn are homeomorphic” is cited in the prerequisites of the book
of E.H. Spanier (Algebraic topology, p. 10, McGraw-Hill, New York, 1966;
reprinted by Springer-Verlag, New York); presumably a proof can be found
in one of the 29 books listed there, but I have not tracked it down.
§7 barycenters INT IV.x269

IV.115, `. 12.
“ . . . pointed convex cone in E .”
With vertex 0 (TVS, II, §2, No. 4).
IV.115, `. 12.
“One knows . . . ”
TVS, II, §2, No. 5, Prop. 13.
IV.115, `. −9.
“The Ck are disjoint convex cones with union C .”
The proof is delicate. Note that, since the f λ are affine, hence convex,
it follows that f is convex (FRV, I, §4, No. 2, Prop. 3). Since the f λ are
positively homogeneous, so is f : f (tx) = tf (x) for x ∈ C , t > 0 .
For every y ∈ C there is at least one index j such that f (y) = f j (y) ,
and y ∈ Ck if and only if k is the first such index; thus C is the union of
the Ck , and the Ck to which y belongs is unique, whence disjointness.
Consider C1 = {y ∈ C : f1 (y) = f (y) } . If y, y 0 ∈ C1 and 0 < t < 1 ,
we are to show that ty + (1 − t)y 0 ∈ C1 . At any rate, ty + (1 − t)y 0 ∈ C ,
and


f1 ty + (1 − t)y 0 = tf1 (y) + (1 − t)f1 (y 0 ) ( f1 is affine)
= tf (y) + (1 − t)f (y 0 ) (because y, y 0 ∈ C1 ) .
Assume to the contrary that ty + (1 − t)y 0 ∈ Ck for some k > 1 . Then, by
the definition of Ck ,



f ty + (1 − t)y 0 > f1 ty + (1 − t)y 0 = tf (y) + (1 − t)f (y 0 ) ,

which contradicts the convexity of f . Thus C 1 is convex; moreover, f C1 =

f1 C1 is affine. Since tC1 ⊂ C1 for t > 0 by the positive homogeneity of f 1
and f , C1 is a convex cone.
“At the other end”, consider
Cp = {y ∈ C : fk (y) < f (y) for 1 6 k < p } .
Of course y ∈ Cp ⇒ fp (y) = f (y) . Suppose y, y 0 ∈ Cp and 0 < t < 1 .
If ty + (1 − t)y 0 did not belong to Cp , then ty + (1 − t)y 0 ∈ Ck for some
k < p , therefore



f ty+(1 − t)y 0 = fk ty + (1 − t)y 0 (because ty + (1 − t)y 0 ∈ Ck )
= tfk (y) + (1 − t)fk (y 0 )
< tfp (y) + (1 − t)fp (y 0 ) (because y, y 0 ∈ Cp and k < p )


= fp ty + (1 − t)y 0


6 f ty + (1 − t)y 0 (by the definition of f ) ,



whence the absurdity f ty + (1 − t)y 0 < f ty + (1 − t)y 0 .
INT IV.x270 extension of a measure. l p spaces §7

 Thus Cp is convex, hence is a convex cone (positive homogeneity), and

f Cp = fp Cp is affine.
“In between”, consider 1 < m < p and

Cm = {y ∈ C : fk (y) < f (y) for 1 6 k < m , and fm (y) = f (y) } .

If y, y 0 ∈ Cm and 0 < t < 1 , then

fm ty + (1 − t)y 0 = tfm (y) + (1 − t)fm (y 0 )
(∗)
= tf (y) + (1 − t)f (y 0 ) (because y, y 0 ∈ Cm ).

If ty + (1 − t)y 0 did not belong to Cm , one would have ty + (1 − t)y 0 ∈ Ck

for some k 6= m , therefore



(∗∗) f ty + (1 − t)y 0 = fk ty + (1 − t)y 0 = tfk (y) + (1 − t)fk (y 0 ) .

case 1. k < m. Then

f ty+(1 − t)y 0 = tfk (y) + (1 − t)fk (y 0 ) by (∗∗)
< tf (y) + (1 − t)f (y 0 ) (because y, y 0 ∈ Cm and k < m )



= fm ty + (1 − t)y 0 by (∗)


6 f ty + (1 − t)y 0 (by the definition of f ) ,



whence the absurdity f ty + (1 − t)y 0 < f ty + (1 − t)y 0 .
case 2. m < k < p . Then



tf (y)+(1 − t)f (y 0 ) = fm ty + (1 − t)y 0 by (∗)


< fk ty + (1 − t)y 0 (because ty + (1 − t)y 0 ∈ Ck and m < k )



= f ty + (1 − t)y 0 because ty + (1 − t)y 0 ∈ Ck ,

contradicting the convexity of f .  

Thus Cm is convex, indeed is a convex cone, and f Cm = fm Cm is
affine.
IV.115, `. −7.
“Then y1 + y2 + · · · + yp = x .”
Recall that a convex cone with vertex 0 is closed under addition (TVS,
II, §2, No. 4, Prop. 10); thus its operations are x + y and tx ( t > 0 ; or
t > 0 when the cone is pointed). In particular, y k ∈ C for k = 1, . . . , p .
Thus
y1 + · · · + y p = x 1 + · · · + x n
is immediate from the associativity theorem (A, I, §1, No. 3, Th. 1).
§7 barycenters INT IV.x271

For use below, we note that a positively homogeneous affine function

1 1

g : C → R is additive: x + y = 2 · x + y , therefore
2 2
1 1  h1 1 i
g(x + y) = 2g x + y = 2 g(x) + g(y) = g(x) + g(y) .
2 2 2 2
 
In particular, the functions f Ck = fk Ck are additive.
IV.115, `. −7, −6.
“Since f is affine on Ck , f (y1 ) + · · · + f (yp ) = f (x1 ) + · · · + f (xn ) .”
 P
In slow motion, let Ik = i ∈ {1, . . . , n} : xi ∈ Ck ; then yk = xi ,
  i∈I
P k

and f (yk ) = f (xi ) by the additivity of f Ck = fk Ck (see the preceding

i∈Ik
note), and
p p   Xn
X X X
f (yk ) = f (xi ) = f (xi )
k=1 k=1 i∈Ik i=1

by the associativity theorem. Resist writing f (x) for this sum; f is convex
on C but need not be additive.
For use below, we note that the convexity and positive homogeneity
of f implies that f is subadditive, that is, f (x + y) 6 f (x) + f (y) for
x, y ∈ C ; for,
1 1  h1 1 i
f (x + y) = 2f x + y 6 2 f (x) + f (y) = f (x) + f (y) .
2 2 2 2

IV.115, `. −5.


“ (19) f (x) = sup f (y1 ) + · · · + f (yp ) , ”

A misprint; read f(x) instead of f (x) .

IV.116, `. 8.
“ . . . therefore f is concave.”
For x, y ∈ C and 0 < t < 1 ,



f tx + (1 − t)y > f (tx) + f (1 − t)y = tf(x) + (1 − t)f(y) .

IV.116, `. 11.
“ . . . weakly complete”
Regarding the product space R × E as a weak locally convex space.
INT IV.x272 extension of a measure. l p spaces §7

IV.116, `. 12, 13.

“By Lemma 5, this sum is equal to L .”
Suppose (t, x) ∈ L , that is, x ∈ C and 0 6 t 6 f(x) . By Lemma 5 one
can write x = x1 +· · ·+xp with the xi in C and f(x) = f (x1 )+· · ·+f (xp ) .
Since
0 6 t 6 f(x) = f (x1 ) + · · · + f (xp )
and the f (xi ) are > 0 , one can write t = t1 + · · · + tp with 0 6 ti 6 f (xi )
for i = 1, . . . , p : for, if f (xi ) = 0 for all i , so that t = 0 , one sets ti = 0
for all i ; otherwise, one sets
f (xi ) t
ti = ·t= · f (xi ) 6 1 · f (xi ) .
f (x1 ) + · · · + f (xp ) f (x1 ) + · · · + f (xp )
Then (ti , xi ) ∈ Li and (t, x) = (t1 , x1 )+· · ·+(tp , xp ) ; thus L ⊂ L1 +· · ·+Lp .
On the other hand, if (t1 , x1 ) ∈ L1 , . . . , (tp , xp ) ∈ Lp are given, so that
xi ∈ C and 0 6 ti 6 f (xi ) for all i , then
0 6 t1 + · · · + tp 6 f (x1 ) + · · · + f (xp ) 6 f (x1 + · · · + xp ) ,
whence (t1 + · · · + tp , x1 + · · · + xp ) ∈ L , that is, (t1 , x1 ) + · · · + (tp , xp ) ∈ L .
Thus L1 + · · · + Lp ⊂ L .
Note that since the sets Li and L are closed in R+ × C , which is
complete hence closed in R+ × E , they are also closed in R+ × E .
IV.116, `. 13, 14.
“ . . . f is upper semi-continuous.”
Equivalently, we are to show that −f is lower semi-continuous. Thus,
given any t ∈ R , it is to be shown that the set A t = {x ∈ C : f(x) > t }
is closed in C (GT, IV, §6, No. 2, Prop. 1). If t < 0 then A t = C . If t > 0
then At is the inverse image, under the continuous mapping
C → R+ × E defined by x 7→ (t, x) (x ∈ C) , of the closed subset L =
{(s, x) : x ∈ C, 0 6 s 6 f(x) } of R+ × E .
IV.116, `. −11.
“ . . . f is indeed convex.”
The argument can be simplified slightly (no need to bring in ε ). By
Lemma 5 (with x + y in the role of x ) one can write
(∗) x + y = z 1 + · · · + zp
with the zi in C and f(x + y) = f (z1 ) + · · · + f (zp ) . Applying the decom-
position theorem to (∗) (A, VI, §1, No. 10, Th. 1) (with q = 2 ), x + y is
the sum of 2p elements zij of C (i = 1, . . . , p ; j = 1, 2) such that
p
X p
X
x= zi1 , y = zi2 , zi = zi1 + zi2 for i = 1, . . . , p .
i=1 i=1
§7 barycenters INT IV.x273

Setting xi = zi1 and yi = zi2 for i = 1, . . . , p , one has

x = x1 + · · · + xp , y = y1 + · · · + yp , zi = xi + yi for i = 1, . . . , p .

Since f (zi ) 6 f (xi ) + f (yi ) (see the note for IV.115, `. −7, −6) one has

p p p
X X X
f(x + y) = f (zi ) 6 f (xi ) + f (yi ) 6 f(x) + f(y) ;
i=1 i=1 i=1

therefore, for 0 < t < 1 ,

f tx + (1 − t)y 6 f (tx) + f (1 − t)y = tf(x) + (1 − t)f(y) ,

thus f is convex in C . Combined with concavity, this proves that f is affine.

It follows directly from the inequalities established here that f(x + y) =
f(x) + f(y) for all x, y ∈ C (see also the note for IV.115, `. −7, −6).
IV.116, `. −2.
“affine functions”
That is, the functions z 7→ hz, z 0 i + c (z ∈ E) , were z 0 is a continuous
linear form on E and c is a scalar; also called “affine linear functions” (A,
II, §9, No. 4, Def. 3).
IV.117, `. 1, 2.
“ . . . where f1 , . . . , fp belong to A , f1 > 0 , . . . , fp > 0 ”
Read: where f1 , . . . , fp belong to A , are 6 f , and f1 > 0 , . . . , fp > 0 .
IV.117, `. 5.
“ f(y) = f (y) if y ∈ G .”
By Lemma 5, there exist y1 , . . . , yp ∈ C with y = y1 + · · · + yp and
f(y) = f (y1 ) + · · · + f (yp ) . If p = 1 the assertion
P is trivial; assume p > 1 .
If y = 0 then yi = 0 for all i (because −yi = yj ∈ C and C is proper)
j6=i
and f(y) = 0 = f (y) .P
Assume y 6= 0 .
From y − yi = yi ∈ C we know that 0 6 yi 6 y for the order
j6=i
relation 6 on E with positive cone C , therefore y i = ri y with 0 < ri < 1
P
p
(TVS, II, §7, No. 2, Remark 1), whence ri = 1 and
i=1

p
X p
X p
X
f (y) = f (yi ) = f (ri y) = ri f (y) = 1 · f (y) .
i=1 i=1 i=1
INT IV.x274 extension of a measure. l p spaces §7

IV.117, `. 5, 6.  
“Since λ*(K --- (K ∩ G)) = 0 , we have λ(f K) = λ(f K) .”
We have just shown that f = f on G . Since K = (K∩G)∪(K--- K∩G)  ,
where f = f on K ∩ G and λ*(K --- K ∩ G) = 0 , we have f K = f K
λ-almost everywhere, therefore λ*(f K) = λ*(f K) (§2, No. 3, Prop. 6).
The function f (resp. f ) is lower (resp. upper) semi-continuous, by
hypothesis (resp. Lemma 6), hence so is its restriction to K (GT, IV, §6,
No. 2). It follows that f K and f K are measurable with respect to any
measure on K (§5, No. 5, Cor. of Prop. 8); being bounded, they are inte-
grable with respect to the (bounded)
 measure λ (§5, No. 6, Th. 5), so one
can drop the asterisks: λ(f K) = λ(f K) .
IV.117, `. 9.
“Let x ∈ K be the barycenter of λ .”
Since K is compact and convex, the barycenter of λ —a priori an ele-
ment of E0 * —may be viewed as an element of K (No. 1, Cor. of Prop. 1).
IV.117, `. 9, 10. 
“If g ∈ A then λ(g K) = g(x) .”
R  
Recall that x = bλ = K y dλ(y) ∈ E0 * . If g ∈ A then g K = z 0 K for
some z 0 ∈ E0 , thus
Z Z
 
g(x) = hx, z i = hy, z i dλ(y) = (g K)(y) dλ(y) = λ(g K) .
0 0

IV.117, `. 10. 
“Therefore λ(f K) = f(x) (§4, No. 4, Cor. 2 of Prop. 5).”

Since λ(g K) = g(x) for every g ∈ A , and since λ(1) = 1 , it is
immediate that the equality holds for every g ∈ A 0 . The functions in A 0
are continuous, f(K) is the lower envelope of a decreasing
 directed set D
of functions g K with g ∈ A , and the numbers λ(g K) ( = g(x) ) are
 0

bounded below (by f(x) ), therefore

Z Z
 
(f K) dλ = inf (g K) dλ

g∈D


by the cited Cor. 2, that is, λ(f K) = inf g(x) = f(x) .
g∈D

IV.117, `. 13.
“ . . . admitting a compact sole M ”
This means (TVS, II, §7, No. 3) that M = C ∩ H , where H is a closed
hyperplane in E that does not pass through the vertex 0 of C , the convex
§7 barycenters INT IV.x275

set M is compact, and M generates C in the sense that C is the smallest

pointed cone with vertex 0 that contains M .
Example: Imagine in R3 an ‘infinite pyramid’ resting on its ‘peak’ at
the origin ( 0, 0, 0) , extending indefinitely above the xy-plane. The extremal
generators of C are its edges. A cross-section of C by a plane not passing
through the origin is a convex polygonal area M that is a compact sole of C .
IV.117, `. 16.
“ . . . such that λ*(M --- (G ∩ M)) = 0 ”
In the language of Ch. V, §5, No. 7, Def. 4, λ is “concentrated on
G ∩ M ”, or “ G ∩ M carries λ ”. If G is closed (for example, if there are
only finitely many extremal generators, as in the example of the preceding
note) then M --- G ∩ M is open in M , so that Supp λ ⊂ G ∩ M . In the
example of the preceding note, G ∩ M is the set of vertices of the polygonal
boundary of M , and λ is a convex combination of the Dirac measures at
the vertices (Ch. III, §2, No. 4, Prop. 12).
IV.117, `. 17, 18.
“Replacing the topology of E by the weakened topology (which does
not change the topology of M ), . . . ”
Let T be the original topology on E , E 0 the dual space of E , and
Tw = σ(E, E0 ) the weakened topology on E . As M is compact for the
induced topology T ∩ M , and the coarser topology T w ∩ M is Hausdorff
(by the Hahn–Banach theorem), the two topologies on M coincide (GT, I,
§9, No. 4, Cor. 3 of Th. 2), whence the assertion. Thus M is a compact,
hence closed, convex subset of E for T w .
Also unchanged by the replacement: the set of measures on K , the
continuous linear forms on E (i.e., the dual space E 0 ), the closed hyper-
planes in E , the closed convex sets in E (TVS, IV, §1, No. 2), the weak
topology σ(E0 , E) on E0 , the algebraic dual (E0 )* of E0 , and the concept
of barycenter bλ ∈ (E0 )* of a positive measure λ on K of total mass 1 .
IV.117, `. 20, 21.
“ . . . let h be a continuous linear form on E such that M is the inter-
section of C and the hyperplane with equation h(x) = 1 .”
See the note for `. 13.
{“ h(y) = 1 ” would have been preferable, as the letter x is conscripted
for duty as a barycenter.}
IV.117, `. 21, 22.
“Let S be the subset of C (M) . . . ”
Clearly S is a pointed convex cone in C (M) with vertex 0 , hence is
closed under addition and under multiplication by scalars > 0 ; it follows
INT IV.x276 extension of a measure. l p spaces §7

that S − S is a linear subspace of C (M) . If f, g ∈ S then f ∪ g and

f ∩ g (the upper and lower envelopes) also belong to S . Other properties
will be noted below.
{The order relation in C (M) (or in S − S ) with positive cone S
(TVS, II, §2, No. 5, Prop. 13) is not at issue here.}

IV.117, `. −12.
“By Th. 3, λ(f ) = λ0 (f ) for every f ∈ S .”
Given f ∈ S , let f 0 : C → R be a continuous, positively homoge-
neous convex function such that f 0 M = f . Write M+1 (X) for the positive
measures of total mass 1 on a compact space X .
The dictionary between Th. 3 and its Corollary is as follows:

E ↔ E
C ↔ C
f ↔ f0
K ↔ M
 
f K ↔ f 0 M = f
λ, λ0 ∈ M1+ (K) ↔ λ, λ0 ∈ M1+ (M) .

The assertion is now clear.

IV.117, `. −11 to −9.

“If f1 , f2 , f3 , f4 belong to S , then

sup(f1 − f2 , f3 − f4 ) = sup(f1 + f4 , f3 + f2 ) − (f2 + f4 ) ∈ S − S

inf(f1 − f2 , f3 − f4 ) = − sup(f2 − f1 , f4 − f3 ) ∈ S − S . ”

Regard C (M) as a Riesz space in the usual way, with f 6 g the point-
wise order relation, and with f ∪ g and f ∩ g as the pointwise supremum
and infimum (Ch. III, §1, No. 5). As observed in the note for `. 21, 22, S is
closed under finite sups and infs. The first displayed relation is an applica-
tion, in the Riesz space C (M) , of the invariance of order under translation
by the element f2 + f4 of S (Ch. II, §1, No. 1, formula (5)), and shows
that S − S is closed under finite sups; and the second displayed relation is
a consequence of the first, showing that S − S is closed under finite infs.
(It follows that if f ∈ S − S then |f | = sup(f, −f ) ∈ S − S .)
Conclusion: S −S is itself a Riesz space for the order relation induced
by that of C (M) .
§7 barycenters INT IV.x277

IV.117, `. −8. 
“Since hM ∈ S , S − S contains the constant functions.”
The half-space H0 = {y ∈ E : h(y) > 0 } is a pointed cone (albeit
degenerate) with vertex 0 , and M ⊂ H0 (since h(y) = 1 for y ∈ M  ),
therefore C ⊂ H0 (see the note for `. 13), that is, h > 0 on C . Thus h M
qualifies for membership in S , that is, the constant function 1 M belongs
to S , whence the assertion.
IV.117, `. −6, −5.
“ . . . this form is the difference of two continuous linear forms that are
positive on C (TVS, II, §6, No. 8, Lemma 1).”
The weak completeness of C plays a role here.
IV.117, `. −5 to −3.
“It follows from the foregoing that for α, β real, there exists f ∈
S − S such that f (x) = α , f (y) = β .”
Say k ∈ E0 with k(x) 6= k(y) (Hahn–Banach), and write k = k 1 − k2
with
 ki ∈ E0 and ki > 0 on C for i = 1, 2 (see the preceding note). Then
k M = k1 M − k2 M ∈ S − S , thus S − S does contain a function that
distinguishes between x and y ; the passage to a function in the vector space
S − S (containing 1 ) taking on specified values at x and y is carried out
in the proof of Stone’s theorem (GT, X, §4, No. 1, Th. 2).
 CHAPTER V

Integration of measures

§1. ESSENTIAL UPPER INTEGRAL

V.2, `. 11, 12.

“ . . . the condition µ• (f ) = 0 means that f is locally negligible (Ch. IV,
§5, No. 2, Prop. 5) ”
The condition µ• (f ) = 0 means that for each compact set K in T ,
µ*(f ϕK ) = 0 , i.e., f ϕK is negligible (Ch. IV, §2, No. 2, Def. 1), i.e., the
set {t : f (t)ϕK (t) > 0 } is negligible (loc. cit., No. 3, Th. 1), that is,
writing A = {t : f (t) > 0 } , the set A ∩ K is negligible; this means that
A is locally negligible (by the cited Prop. 5), in other words f is locally
negligible (remarks following Cor. 4 of the cited Prop. 5).
V.2, `. 14.
“The mapping µ• of F+ (T) into R coincides with µ on K+ (T) .”
Let f ∈ K+ (T) . By (1), µ• (f ) 6 µ*(f ) ; but K = Supp f is compact
and f = f ϕK , therefore µ*(f ) = µ*(f ϕK ) 6 µ• (f ) , whence µ• (f ) =
µ*(f ) = µ(f ) (Ch. IV, §1, No. 1, remark following Def. 1).
V.2, `. −6, −5.
“ . . . a) from Proposition 6 of Ch. IV, §2, No. 3 and Proposition 5 of
Ch. IV, §5, No. 2 ”
By assumption, the set A = {t : f (t) 6= g(t) } is locally negligible.
For every compact set K in T , A ∩ K is negligible by the cited Prop. 5,
thus f ϕK = gϕK almost everywhere, so µ*(f ϕK ) = µ*(gϕK ) by the cited
Prop. 6.
V.2, `. −5, −4.
“ . . . b), c), d) from Propositions 10, 11, 12 of Ch. IV, §1, No. 3.”
As to c), the cited Prop. 11 assumes that α is finite. But c) is also true
for α = +∞ , as follows.
§1 essential upper integral INT V.x2

case 1. µ• (f ) = 0 .
Then f is locally negligible, and the convention (+∞)

· 0 in R assures
that (+∞)f is also locally negligible, whence µ • (+∞)f = 0 = (+∞)·0 =
(+∞) · µ• (f ) .
case 2. µ• (f ) > 0 .
Then (+∞) · µ• (f ) = +∞ , and f (t) > 0 ⇔ (+∞) · f (t) = +∞ (GT,
IV, §4, No. 3, Prop. 8). Write A = {t : f (t) > 0 } . By hypothesis, A is
not locally negligible, hence there exists a compact set K in T such that
A ∩ K is not negligible. But (+∞)f = +∞ on A ∩ K ,
thus (+∞)f · ϕ K is
not almost everywhere finite, therefore

µ* (+∞)f · ϕK = +∞ (Ch. IV, §2,
•
No. 3, Prop. 7), whence µ (+∞)f = +∞ and the equality in c) reduces
to +∞ = +∞ .
V.2, `. −2, −1.

lim µ• (fn ) = sup sup µ*(fn ϕK ) = sup sup µ*(fn ϕK )

n→∞ n∈N K∈K K∈K n∈N
•
= sup µ*(f ϕK ) = µ (f ) .
K∈K

By the theorem on monotone limits, f = sup fn (GT, IV, §5, No. 2,

n∈N
Th. 2 applied with X = R ). In view of b), the first displayed equality says
that lim µ• (fn ) = sup µ• (fn ) (the cited Th. 2, with X = F+ (T) and
n→∞ n∈N
•
f = µ ).
The second displayed equality follows from the ‘associativity of sups’
(S, III, §1, No. 9, Cor. of Prop. 7).
The third displayed equality follows from Ch. IV, §1, No. 3, Th. 3.
V.3, `. 6, 7.
“One is immediately reduced to the proof of the analogous formula for
the upper integral.”
Write K for the set of all compact subsets K of T . Assuming proven
R∗
the analogous formula for , for each K ∈ K one has
Z ∗ Z ∗ Z ∗ Z ∗
f (g + h)ϕK dµ = (f ϕK )(g + h) dµ = f ϕK g dµ + f ϕK h dµ ;

as K varies over K , the supremum of the left-most member is by definition

R∗ R∗
µ• f (g + h) . Write αK = f ϕK g dµ , βK = f ϕK h dµ , so that



µ• f (g + h) = sup αK + βK .
K∈K
INT V.x3 integration of measures §1

We know (GT, IV, §5, No. 7, formula (26)) that

sup αK1 + βK2 = sup αK + sup βK = µ• (f g) + µ• (f h) ,
K1 ,K2 ∈K K∈K K∈K

so it will suffice to show that

sup αK + βK = sup α K1 + β K2 .
K∈K K1 ,K2 ∈K

The left member γ is obviously 6 the right member δ ; on the other hand,
for all K1 , K2 ∈ K one has

αK1 + βK2 6 αK1 ∪K2 + βK1 ∪K2 6 γ ,

whence δ 6 γ . {Perhaps this is a special case of GT, loc. cit., Cor. 1 of

Prop. 13; I have not checked the details.}
V.3, `. 10.
“ . . . it remains to establish the reverse inequality.”
R∗ R∗ R∗
Let α = f g dµ , β = f h dµ , γ = f (g + h) dµ ; we are to show
that

(∗) α+β 6γ.

At any rate, 0 6 α, β 6 γ . If γ = +∞ then (∗) holds trivially, so we

can suppose that γ is finite; if one of α, β, γ is equal to 0 then (∗) holds
trivially, so we can suppose that α, β, γ are all finite and > 0 . It then follows
that, at almost every point of T , the functions f g, f h and f (g + h) are
finite-valued (Ch. IV, §2, No. 3, Prop. 7), none of the sets

A = {t : f (t)g(t) > 0}, B = {t : f (t)h(t) > 0}, C = {t : f (t)(g +h)(t) > 0}

is negligible, and A ∪ B = C .
Let N be a negligible set such that f g , f h , and f (g + h) are finite-
valued on T --- N . Replacement of f by f ϕ T -- N does not change the
numbers α, β, γ , so we can suppose that f g , gh , and f (g + h) are finite
at every point of T . This is useful for the next note.
Caution: The functions f and g + h are permitted to have infinite
values but, at a point where one of them is infinite, the other must be 0 .
V.3, `. 12, 13.
“ . . . then v > f and u > v(g + h) ”
We can suppose, after excluding special cases and modifying f on a
negligible set, that f (g + h) is finite at every point of T (see the preceding
§1 essential upper integral INT V.x4

note). It is tacit in the definition of v that if (g+h)(t) = +∞ then v(t) = 0 .

Thus: 

 u(t)

 (g + h)(t) if 0 < (g + h)(t) < +∞
v(t) =

 0 if (g + h)(t) = +∞


+∞ if (g + h)(t) = 0 .
Proof that v > f . If (g + h)(t) = 0 then v(t) = +∞ > f (t) ;
if (g + h)(t) = +∞ then f (t) = 0 (by the finiteness of f (g + h) ) and
v(t) = 0 = f (t) ; and if 0 < (g + h)(t) < +∞ then (since u > f (g + h) )
u(t) f (t)(g + h)(t)
v(t) = > = f (t) .
(g + h)(t) (g + h)(t)
Proof that u > v(g + h) . If (g + h)(t) = 0 , the inequality holds trivially
at t since u > f (g + h) > 0 ; if (g + h)(t) = +∞ then v(t) = 0 and
the inequality again holds trivially at t ; and if 0 < (g + h)(t) < +∞ then
u(t) = v(t)(g + h)(t) by the definition of v(t) .
V.3, `. 15.
“ . . . v being measurable (Ch. IV, §5, No. 6, Cor. 4 of Th. 5) ”
The cited Cor. 4 justifies the equality in the the display that follows,
once the measurability of v is established. We assume that the definition
of v has been fine-tuned as in the preceding note. 
Let A = {t : 0 < (g + h)(t) < +∞ } and let k = (g + h) A be the
restriction of g + h to A . Since g + h is measurable (Ch. IV, §5, No. 3,
Th. 1), the set A is measurable (loc. cit., No. 5, Prop. 7), therefore the
function k : A → R is measurable in the sense of Ch. IV, §5, No. 10, Def. 8
(see the criterion c00 ) on p. IV.x127 above). Since a 7→ 1/a (0 < a < +∞)
is continuous, the function 1/k : A → R is also measurable (see the item
No.3, Th. 1 0 on p. IV.x133). It follows that the extension by 0 of 1/k
to T is measurable (Ch. IV, §5, No. 10, Prop. 15, criterion d)), that is, the
function w defined by

 1
for t ∈ A
w(t) = (g + h)(t)

0 for t ∈ T --- A

is measurable. Since u is measurable (loc. cit., No. 5, Cor. of Prop. 8),

so is uw (see the note 
 for IV.64, `. 10–13), therefore uw A is measurable;
moreover,
 0 uw A = v A , and uw is the extension by 0 of v A to T , briefly

(v A) = uw . On the other hand, the sets

B = {t : (g + h)(t) = 0 } , C = {t : (g + h)(t) = +∞ }
INT V.x5 integration of measures §1

are also measurable (Ch. IV, §5, No. 5, Prop. 7), therefore the step function
(+∞) · ϕB on T is measurable (loc. cit.), so the function

v = (v A)0 + (+∞) · ϕB + 0 · ϕC = uw + (+∞) · ϕB

is the sum of measurable functions.

V.3, `. −7, −6.
“The case of an infinite sequence may be deduced from this by means
of Prop. 1, e).”
Pn P
Writing hn = gk , h = gk , one has hn ↑ h pointwise, therefore
k=0 k∈N
fR hn ↑ f h pointwise
R• (the convention 0 · (+∞) = 0 plays a role here), hence
•
f hn dµ ↑ f h dµ by the cited proposition.
V.4, `. 1, 2.
“The proof is immediate from the analogous statement in Ch. IV (§1,
No. 3, Prop. 15).”
It is understood that the measures µ and ν are positive. In slow
motion: for every f ∈ F+ (T) , as K varies over the set of all compact
subsets of T ,

(µ + ν)• (f ) = sup(µ + ν)*(f ϕK )

K
 
= sup µ*(f ϕK ) + ν*(f ϕK )
K
= sup µ*(f ϕK ) + sup ν*(f ϕK )
K K
• •
= µ (f ) + ν (f )
= (µ• + ν • )(f ) ,

where the second equality holds by the cited Prop. 15; the third, by the
theorem on monotone limits (GT, IV, §5, No. 2, Th. 2) and the continuity
of addition in R+ (GT, IV, §4, No. 3); and the other three, by definitions.
V.4, `. 6, 7.
“It follows, by the definition of upper integral, that µ*(f ) 6 µ • (f ) ”
In the notation of Ch. IV, §1, No. 1, f ∈ I + (T) , therefore µ*(f ) =
sup µ(g) as g varies over the set of all functions in K + (T) such that g 6 f
(loc. cit., Def. 1). For any such g , write K(g) for its (compact) support;
then g = gϕK(g) 6 f ϕK(g) and

µ(g) = µ*(g) 6 µ*(f ϕK(g) ) 6 µ• (f ) ,

whence the assertion.

§1 essential upper integral INT V.x6

V.4, `. 20.
“Ch. IV, §1, No. 4, Prop. 19 ”
And Ch. IV, §4, No. 6, Prop. 10.
V.5, `. 14.
“ . . . f also has it.”
P
By assumption, f = fn . What does the notation mean? For
Pn∈N
every t ∈ T , the sum fn (t) exists in R+ as the supremum of the
n∈N
finite
P subsums (GT, IV, §7, No. 5, Prop. 2) and one defines a function
fn ∈ F+ (T) by
n∈N X  X
fn (t) = fn (t) .
n∈N n∈N

{I can’t put my finger on where this notation is explicitly defined; it appears

already in Ch. IV, §1, No. 1, Prop. 3 and No. 3, Prop. P13.}
Suppose that for each n ∈ N one has fn = hnk , where hnk ∈
k∈N
F+ (T) (k ∈ N), hn0 is µ-negligible, and,
 for every k > 1 , there exists
a compact set Knk in T such that hnk Knk is finite and continuous, and
hnk = 0 on T --- Knk . Thus, by assumption,
X XX 
f= fn = hnk .
n∈N n∈N k∈N

For every t ∈ T , one has

XX  X
hnk (t) = hnk (t) ;
n∈N k∈N (n,k)∈N×N

for, all of the sums exist in R+ , and the equality holds by associativity (GT,
IV, §7, No. 5, Remark ). In other words,
XX  X
hnk = hnk ,
n∈N k∈N (n,k)∈N×N

that is, X
f= hnk .
(n,k)∈N×N

Another application of associativity yields

X X
f= hnk + hn0 .
n∈N, k>1 n∈N
INT V.x7 integration of measures §1

Let (hn )n>1 be any rearrangement of the hnk (n ∈ N, k > 1) , and (K

Pn )n>1
the corresponding rearrangement of the K nk , and define h0 = hn0 ;
n∈N
then, by commutativity (GT, III, §5, No. 1, Remarks 2, 3),
X
f= hn ,
n∈N

where h0 is negligible and, for n > 1 , hn = 0 on T --- Kn and hn Kn is
finite-valued and continuous.
V.5, `. 16–18.
“ . . . since f is equal to the sum of the sequence (f n ) , it will thus suffice
to establish the proposition assuming f to be moderated and bounded.”
For every n , 0 6 inf(f, n) 6 n , in particular inf(f, 0) = 0 and
inf(f, n) is bounded; and 0 6 inf(f, n) 6 f , so inf(f, n) is moderated.
Also, inf(f, n) 6 inf(f, n + 1) . It follows that every f n is > 0 , bounded,
measurable and moderated, and (‘telescoping sum’)
n
X
fk = inf(f, n + 1) − inf(f, 0) = inf(f, n + 1) .
k=0

Moreover, the sequence of functions inf(f, n+1) is increasing

P and has upper
envelope f (even at points where f (t) = +∞ ), whence fn = f .
n∈N

V.5, `. −15 to −12. 

“Let K be the set of compact subsets K of T such that f K is con-
tinuous; since K is µ-dense (Ch. IV, §5, No. 10, Prop. 15), L is the union
of a negligible set N and a sequence (K n )n>1 of pairwise disjoint elements
of K (Ch. IV, §5, No. 8, Def. 6).”
The introduction of K and the reference to the cited Prop. 15 are con-
fusing, as that Proposition pertains to functions defined on a measurable
subset of T and taking values in a topological space.
Our situation is much simpler. We have a bounded, measurable nu-
merical function f > 0 defined on all of T , and we are given a compact
set L in T such that f = 0 on T --- L . Apply to f the definition of
measurability (Ch. IV, §5, No. 1, Def. 1): there exist a negligible set N ⊂ L
and a countable
S family
 (Kn ) of pairwise disjoint compact sets such that
L --- N = Kn and f Kn is continuous for every n . Then
n
X
f = f ϕ L = f ϕN + f ϕ Kn ,
n

and the functions h0 = f ϕN , hn = f ϕKn meet the requirements.

§1 essential upper integral INT V.x8

V.5, `. −2, −1.

“The first assertion follows immediately from Lemma 1 of Ch. IV, §5,
No. 6.”
Assuming µ*(f ) < +∞ we are to show that f is µ-moderated. Since f
is equal µ-almost everywhere to a finite-valued function (Ch. IV, §2, No. 3,
Prop. 7), we can suppose that f is finite-valued (Remark 4 and Ch. IV, loc.
cit., Prop. 6), i.e., takes its values in the Banach space R ; then, by the cited
Lemma 1, the set A = {t : f (t) 6= 0 } satisfies the condition c) of Prop. 5,
hence is µ-moderated, thus f = 0 on the complement of a µ-moderated set.
One notes that to say that f is µ-moderated, that is, f is zero on the
complement of a µ-moderated set—in other words {t : f (t) 6= 0 } ⊂ M for
some moderated set M —is equivalent to saying that the set {t : f (t) 6= 0 }
is moderated.
The following reformulation of 1) exhibits its parallelism with 3):
10 ) If µ*(f ) < +∞ then there exists a µ-moderated subset A , the union
of a sequence of compact subsets of T , such that f = f ϕ A almost every-
where.
For, if µ*(f ) < +∞ we know from the foregoing remark that the set
{t : f (t) 6= 0 } is moderated, hence
[
{t : f (t) 6= 0 } ⊂ N ∪ Kn
n
S
with N negligible and (Kn ) a sequence of compact sets. Let A = Kn . If
n
f (t) 6= f (t)ϕA (t) then f (t) 6= 0 and t ∈
/ A , whence t ∈ N ; thus f = f ϕ A
almost everywhere.
V.6, `. 14.
“ . . . from which 3) follows.”
µ• (f ϕ { A ) = 0 means that f ϕ { A is locally negligible (V.2, `. 11, 12),
thus the set

N = {t : f (t)ϕ { A (t) 6= 0 } = { A ∩ {t : f (t) 6= 0 }

is locally negligible; on its complement { N = A ∪ {t : f (t) = 0 } one clearly

has f = f ϕA .
V.6, `. 15, 16.
“Corollary 1.”
The assertion is that µ*(|f |) = 0 ⇔ µ• (|f |) = 0 and f is moderated.
⇐: Citing Prop. 7, 2), µ*(|f |) = µ• (|f |) = 0 .
⇒: The set N = {t : f (t) 6= 0 } is negligible, hence (trivially) moder-
ated; that is, f is moderated, whence µ • (|f |) = µ*(|f |) = 0 .
INT V.x9 integration of measures §1

V.6, `. 17, 18.

“Corollary 2.”
All functions on T are µ-moderated, so µ* = µ • by Prop. 7, 2).
V.6, `. −3 to −1.
“There exists a compact set K such that (Ch. IV, §4, No. 4, Cor. 1 of
Prop. 5):
µ• (h0 ) − a 6 µ*(h0 ϕK ) = µ(h0 ϕK ) 6 µ• (h0 ) . ”
The existence of K satisfying µ• (h0 ) − a 6 µ*(h0 ϕK ) 6 µ• (h0 ) follows
from the definition of µ• . Since h0 and ϕK are upper semi-continuous,
so is h0 ϕK (by the ‘dual’ of GT, IV, §6, No. 2, Prop. 2); but h 0 ϕK is
moderated, so µ*(h0 ϕK ) = µ• (h0 ϕK ) 6 µ• (h0 ) < +∞ , therefore h0 ϕK is
µ-integrable by the cited Cor. 1, whence µ*(h 0 ϕK ) = µ(h0 ϕK ) (Ch. IV, §4,
No. 2, Prop. 1).
V.7, `. 5.
“But (Ch. IV, §4, No. 4, Cor. 1 of Prop. 5) µ • (h0 ϕ { K ) 6 a ”
One has h0 = h0 ϕK + h0 ϕ { K , where ϕK and ϕ { K are measurable and
h0 > 0 ; by No. 1, Prop. 2,
µ• (h0 ) = µ• (h0 ϕK ) + µ• (h0 ϕ { K ) ;
but µ• (h0 ϕK ) = µ*(h0 ϕK ) = µ(h0 ϕK ) (preceding note), thus
µ• (h0 ϕ { K ) = µ• (h0 ) − µ• (h0 ϕK ) = µ• (h0 ) − µ(h0 ϕK ) 6 a .

V.7, `. 8.
“ inf µ*(hϕK ) > inf µ• (h) − a .”
h∈H h∈H
For all h ∈ H with h 6 h0 , one has µ• (hϕ { K ) 6 µ• (h0 ϕ { K ) 6 a and
so
µ• (h) = µ• (hϕK ) + µ• (hϕ { K ) 6 µ• (hϕK ) + a = µ*(hϕK ) + a ,
thus µ*(hϕK ) > µ• (h) − a , whence the asserted inequality.
V.7, `. −8, −7.
“The closure of 0 for this topology is the space N F∞ ”
p
The set in question is {f ∈ F F : Np (f ) = 0 } (TVS, II, §1, No. 2,
Prop. 2); to say that f : T → F is locally negligible means that µ • (|f |) = 0
(V.2, `. 11, 12). Thus the assertion is that
Np (f ) = 0 ⇔ f = 0 locally almost everywhere.
p
Given f ∈ F F let g ∈ FFp with f = g locally almost everywhere. By the

1/p
definitions, Np (f ) = Np (g) = µ*(|g|p ) .
§1 essential upper integral INT V.x10

Proof of ⇒: If Np (f ) = 0 , that is, Np (g) = 0 , then g = 0 almost ev-

erywhere (Ch. IV, §3, No. 2, Prop. 3); but f = g locally almost everywhere,
therefore f = 0 locally almost everywhere.
Proof of ⇐: If f = 0 locally almost everywhere, then also g = 0 locally
almost everywhere; but g is moderated, therefore g = 0 almost everywhere
by the Lemma, whence Np (f ) = Np (g) = 0 .
V.7, `. −6, −5.
p
“ . . . the normed space F F /NF∞ may be canonically identified with
FFp /NF ”
p
If f ∈ F F , and g ∈ FFp with f = g locally almost everywhere, the
class ge = g + NF of g for equality almost everywhere depends only on the
p
class f + NF∞ of f for equality locally almost everywhere; for, if f 0 ∈ L F
with f 0 = f locally almost everywhere, and if g 0 ∈ LFp with f 0 = g0
locally almost everywhere, then g 0 = g locally almost everywhere, therefore
g − g0 ∈ NF by the Lemma. The correspondence f + N F∞ 7→ g e = g + NF
(where g ∈ LFp and f = g locally almost everywhere) is therefore a well-
defined mapping
p
F F /NF∞ → FFp /NF ,

clearly linear and surjective. Moreover, by the definitions,

Np (f ) = Np (g) = Np (e
g)

(cf. Ch. IV, §3, No. 2), thus the mapping is an isometry. Since F Fp is
complete (loc. cit., No. 3, Prop. 5), F Fp /NF is a Banach space (given a
Cauchy sequence (e gn ) in FFp /NF , the sequence (gn ) is Cauchy in FFp ,
p p
etc.), therefore so is F F /NF∞ , consequently F F is complete.
Caution: Note the author’s avoidance of the notation e f for the class
f + NF∞ of f for equality locally almost everywhere, as this notation has
been pre-empted for f + NF in Ch. IV, §3, No. 2. {In another context, the
notation ḟ = f + NF∞ has been employed (Ch. IV, §6, No. 3).}
V.7, `. −3, −2.
p p p
“We shall similarly denote by L F (T, µ) (or L F (µ) , or L F ) the sub-
p
space LFp + NF∞ of F F ”
p p
As for the notation L F , note that LFp is dense in L F , that is, LFp =
p p p p
L F . For, {0} ⊂ L F ⊂ F F and the closure of {0} in F F is NF∞ ,
p p
therefore the closure of {0} in L F is NF∞ ∩ L F = NF∞ (GT, I, §3, No. 1,
p
Prop. 1); since LFp contains {0} , its closure in L F is a linear subspace
p
containing LFp and NF∞ , hence containing LFp + NF∞ = L F .
INT V.x11 integration of measures §1
p p
Better yet, KF is dense in L F . For, if f ∈ L F , say f = g locally
almost everywhere, where g ∈ LFp , there exists a directed family (gj )
in KF such that Np (gj − g) → 0 (Ch. IV, §3, No. 4, Def. 2); since g j − g =
gj − f locally almost everywhere, one has

Np (gj − f ) = Np (gj − g) → 0 ,
p
that is, gj → f in L F .
V.7, `. −2, −1.
p p
“ . . . one can also characterize L F as the subspace of F F constituted
by the measurable mappings (Ch. IV, §5, No. 6, Th. 5).”
Note that every function in NF∞ is measurable, since it is equal locally
almost everywhere to the measurable function 0 (Ch. IV, §5, No. 2, Prop. 6).
p
Let f ∈ F F , say f = g locally almost everywhere with g ∈ F Fp ; then
f − g ∈ NF∞ , so f − g is measurable.
If f is measurable then so is g = f − (f − g) ; but g ∈ F Fp , therefore
p
g ∈ LFp by the cited Th. 5, thus f = g + (f − g) ∈ L Fp + NF∞ = L F .
Conversely, if f ∈ LFp + NF∞ then f is the sum of two measurable
mappings, hence is measurable.
V.8, `. −15, −14.
“Corollary.”
p
Necessity: f ∈ LFp ⊂ L F , and |f |p (hence also f ) is moderated by
Prop. 7, 1).
p
Sufficiency: If f ∈ L F is moderated then µ*(|f |p ) = µ• (|f |p ) < +∞
(Prop. 7, 2) and Prop. 9); moreover, f is measurable (Prop. 9) therefore
f ∈ LFp (Ch. IV, §5, No. 6, Th. 5).
V.8, `. −13 to −7.
“Definition 3.”
1
In the present context, f denotes elements of L F ; thus the composition
indicated in Def. 3,
1
L F → L1F → F
e = g + NF ∈ L1F (where g ∈ LF1 and f = g locally
is the mapping f 7→ g
almost everywhere), followed by the mapping g e 7→ µ(g) . The continuity of
the composite is immediate from the definitions:

N1 (f ) = N1 (g) = N1 (e
g) = ke
gk1 = µ(|g|)

(cf. Ch. IV, §3, No. 4) and |µ(g)| 6 µ(|g|) (Ch. IV, §4, No. 2, Prop. 2).
§1 essential upper integral INT V.x12

V.8, `. −6, −5.

“Two essentially integrable functions that are equal locally almost ev-
erywhere have the same integral.”
1
More precisely, if f ∈ L F and f 0 : T → F is such that f 0 = f locally
1 R R
almost everywhere, then also f 0 ∈ L F and f 0 dµ = f dµ . This is implicit
1 1
in Def. 3. Indeed, f 0 −f ∈ NF∞ and so f 0 = f +(f 0 −f ) ∈ L F +NF∞ = L F ;
and if g ∈ LF1 with f = g locally almost everywhere, one knows that R the
e = g + NF of g for equality almost everywhere (hence also g dµ )
class g
is independent of the class fR + NF∞ = 0
R f + NF
∞
R of f for equality locally
0
almost everywhere, whence f dµ = g dµ = f dµ .
V.8, `. −5, −4.
R • “For every
R function f > 0 that is finite and essentially integrable,
f dµ = f dµ . ”
Note that a subset of T is negligible if and only if it is moderated and
locally negligible (No. 2, Cor. 1 of Prop. 7).
Let g ∈ LR1 with f = g locally almost everywhere. Then g > 0
locally almost everywhere, thus the set N = {t : g(t) < 0 } is locally
negligible. On the other hand, g is moderated (No. 2, Prop. 7, 1)), thus
the set A = {t : g(t) 6= 0 } is moderated. But N ⊂ A , so N = N ∩ A
is both locally negligible and moderated, hence negligible, that is, g > 0
almost everywhere. Replacing g by gϕ { N , we can suppose that g > 0
(everywhere). Then
Z • Z • Z Z
f dµ = g dµ = g dµ = f dµ ,

the equalities holding, respectively, by No. 1, Prop. 1, a), No. 2, Prop. 7, 2),
and Def. 3.
V.8, `. −4 to −2.
“If A is a set whose characteristic function is essentially integrable,
then A is said to be an essentially µ-integrable set ”
For a subset A of T , the following conditions are equivalent:
a) A is essentially integrable;
b) there exists an integrable set C such that ϕ A = ϕC locally almost
everywhere;
c) there exists an integrable set C such that the ‘symmetric difference’
of A and C , that is, the set (A --- C) ∪ (C --- A) , is locally negligible.
Proof. b) ⇔ c): For, (A --- C) ∪ (C --- A) = {t : ϕ A (t) 6= ϕC (t) } .
b) ⇒ a): Obvious from the definitions.
INT V.x13 integration of measures §1

a) ⇒ b): By assumption, there exists a function g ∈ L R1 such that

ϕA = g locally almost everywhere. Since (ϕ A )2 = ϕA we have g 2 = g
locally almost everywhere, hence g 2 = g almost everywhere (Lemma). Let

B = {t : g(t)2 = g(t) } = {t : g(t) = 0 or 1 } ;

we know that { B is negligible. Let

−1
C = {t : g(t) = 1 } = g (1) = {t ∈ B : g(t) 6= 0 } ,

which is a measurable set (Ch. IV, §5, No. 5, Prop. 7). If t ∈ C then
g(t) = 1 = ϕC (t) ; that is, g = ϕC on C . To show that ϕA = ϕC locally
almost everywhere, it will suffice to show that

{t : g(t) 6= ϕC (t) } ⊂ { B

(whence g = ϕC almost everywhere). Indeed, suppose g(t) 6= ϕ C (t) . By

the foregoing, t ∈
/ C ; therefore ϕC (t) = 0 and so g(t) 6= 0 . If one had t ∈ B
it would follow that g(t) = 1 , consequently t ∈ C , a contradiction. Finally,
since g is integrable and ϕC = g almost everywhere, one has µ*(ϕC ) =
µ*(g) < +∞ , consequently ϕC is integrable (Ch. IV, §5, No. 6, Th. 5), that
is, C is an integrable set.
V.9, `. 12–16.
“We note for example the inequality
Z  Z
 
 
(3)  f dµ 6 |f | dµ ,

valid for every essentially integrable function f with values in a Banach

space.”
Choose f1 ∈ LF1 with f = f1 locally almost everywhere. Then
|f1 | ∈ LR1 (Ch. IV, §3, No. 5, Prop.
R 11) andR |f | = |f 1 |R locally almost
R ev-
erywhere, therefore, by definition, f dµ = f1 dµ and |f | dµ = |f1 | dµ ,
thus (3) is immediate from the corresponding inequality for f 1 (Ch. IV, §4,
No. 2, Prop. 2).
Another example. The analogue of Th. 1 of Ch. IV, §4, No. 2: Suppose
F, G are Banach spaces, u : F → G is a continuous linear mapping, and
1
f ∈ L F . Choose f1 ∈ LF1 with f = f1 locally almost everywhere. Obvi-
ously u ◦ f = u ◦ f1 locally almost everywhere, therefore, by the cited Th. 1,
u ◦ f1 ∈ LF1 and Z Z 
(u ◦ f1 ) dµ = u f1 dµ ;
§1 essential upper integral INT V.x14
1
thus u ◦ f ∈ L F and
Z Z Z  Z 
(u ◦ f ) dµ = (u ◦ f1 ) dµ = u f1 dµ = u f dµ .

V.9, `. −9 to −7.
“To establish a), it suffices to show that for every compact subset L
R∗ R∗
of T , f ϕL dµ = sup f ϕK dµ , where K runs over the set of subsets
K
of L belonging to K .”
For, assuming this to be shown, it would follow that
Z ∗ Z ∗
f ϕL dµ 6 sup f ϕK dµ ,
K∈K

R∗
whence, varying L , µ• (f ) 6 sup f ϕK dµ , whereas the reverse inequality
K∈K
follows trivially from the definition of µ • .
V.10, `. 2.
Z Z  Z
 

“  f dµ − f ϕH dµ 6 |f |ϕ { H dµ ”

If f1 ∈ LF1 with f = f1 locally almost everywhere, then f1 ϕH , f1 ϕ { H

belong to LF1 by Ch. IV, §5, No. 6, Th. 5, therefore
Z Z Z
f1 dµ = f1 ϕH dµ + f1 ϕ { H dµ

by the additivity implicit in Ch. IV, §4, No. 1, Def. 1.

V.10, `. 5, 6.
p
“The Banach space L F (T, µ) may then be equipped with a natural
complex Banach space structure”
p
The notation L F (T, µ) has only been defined for functions with values
in a real Banach space, thus it is intended that, at the outset, F is regarded
p
as a real Banach space by restriction to real scalars. However, if f ∈ L F
p
and c ∈ C , the function cf defined by (cf )(t) = cf (t) belongs to L F ;
for, f is measurable, hence so is cf , and µ • (|cf |p ) = |c|p µ• (|f |p ) < +∞
p
(cf. Prop. 9). It is clear that L F becomes, thereby, a complex vector space,

1/p p
on which f 7→ Np (f ) = µ• (|f |p ) is a semi-norm that makes L F a
complete locally convex space over C , and for which N p (f ) = 0 if and only
if f ∈ NF∞ . The complex Banach space in question is the quotient space
p
L F /NF∞ , equipped with the norm kf + NF∞ kp = Np (f ) .
INT V.x15 integration of measures §1
p
To recapitulate: when F is a complex Banach space, L F is the same
set of functions whether F is regarded as a real or a complex Banach space;
all that changes is the field of scalars.
V.10, `. 13, 14.
“Assertion b) of Prop. 10 then extends at once to complex measures.”
Proof #1. Note that the inequality
Z  Z
 
(∗)  f dθ  6 |f | d|θ| ,
 

valid for f ∈ LF1 (θ) = LF1 (|θ|) (Ch. IV, §4, No. 2, Prop. 2), generalizes at
1
once for f ∈ L F (θ) .
1 1
Let f ∈ L F (θ) . Apply b) of Prop. 10 to µ = |θ| and |f | ∈ L R (µ) to
obtain Z Z
|f | d|θ| = lim |f |ϕK d|θ| ,
K

that is, Z
lim |f |ϕ { K d|θ| = 0 ;
K

then, citing (∗), one has

Z  Z
 
 fϕ 
 { K dθ  6 |f |ϕ { K d|θ| ,

R
whence lim f ϕ { K dθ = 0 , that is,
K
Z Z
f dθ = lim f ϕK dθ . ♦
K

Proof #2. The idea is to express θ as a linear combination of positive

measures (Ch. III, §1, No. 5)

(∗∗) θ = µ1 − µ2 + iµ3 − iµ4 , 0 6 µk 6 |θ| (k = 1, . . . , 4)

and apply b) of Prop. 10 to each of the µ k . However, to put together the

four limits, we need to know that the expression θ(f ) is the corresponding
linear combination of the µk (f ) ; when F = R and f ∈ K (T; R) , this is
immediate from the definition of the linear operations in M (T; R) (loc. cit.,
No. 3), and the case of f ∈ K (T; F) , F a Banach space, is implicit in Ch. III,
§3, No. 2, Prop. 1 and No. 3, Cor. 2 of Prop. 7, but for f ∈ L F1 (θ) = LF1 (|θ|) ,
a different approach is required.
§1 essential upper integral INT V.x16
R
Recall that for f ∈ LF1 (θ) , f dθ is defined as follows (Ch. IV, §4,
No. 1, Def. 1): KF is dense in LF1 (θ) for theR semi-norm N1 = |θ|* ; the
linear mapping θ : KF → F defined by θ(g) = g dθ being continuous with
1
respect to N1 , one extends θ by continuity to
R LF (θ) , and the value of the
1
extension at f ∈ LF (θ) is denoted θ(f ) or f dθ .
Thus, given any f ∈ LF1 (θ) , we may choose a sequence (gn ) in KF
such that N1 (f − gn ) → 0 (Ch. IV, §3, No. 4, Prop. 7), and we then have

θ(f ) = lim θ(gn ) in F .

n

If θ 0 is any measure on T such that |θ 0 | 6 |θ| , and if N01 = |θ 0 |* , then

(Ch. IV, §1, No. 3, Prop. 15)

N01 (f − gn ) 6 N1 (f − gn ) → 0 ,

consequently f ∈ LF1 (θ 0 ) and

θ 0 (f ) = lim θ 0 (gn ) in F .
n

Suppose now that θ is decomposed as in (∗∗). The foregoing shows

that, for k = 1, . . . , 4, f ∈ LF1 (µk ) and

µk (f ) = lim µk (gn ) in F ,
n

therefore


θ(f ) = lim θ(gn ) = lim µ1 (gn ) − µ2 (gn ) + iµ3 (gn ) − iµ4 (gn )
n n
= lim µ1 (gn ) − lim µ2 (gn ) + i lim µ3 (gn ) − i lim µ4 (gn )
n n n n
= µ1 (f ) − µ2 (f ) + iµ3 (f ) − iµ4 (f ) .

Applying b) to each of the four terms on the right, one obtains the desired
formula: for f ∈ LF1 (θ) ,

θ(f ) = lim µ1 (f ϕK )−lim µ2 (f ϕK )+i lim µ3 (f ϕK )−i lim µ4 (f ϕK ) = lim θ(f ϕK )

K K K K K

(the second equality, because θ(f ϕ K ) = µ1 (f ϕK ) − µ2 (f ϕK ) + iµ3 (f ϕK ) −

iµ4 (f ϕK ) ), that is, Z Z
f dθ = lim f ϕK dθ ;
K

1
its extension to f ∈ L F (θ) is then immediate. ♦
INT V.x17 integration of measures §1

The definition of a ‘universal’ bilinear mapping (θ, f ) 7→ θ(f ) remains

elusive, as the eligible functions f depend on θ ; but the foregoing arguments
yield the following:
Lemma. If µ is any positive measure on T , and if M µ is the vector
space of complex measures θ on T such that |θ| 6 cµ for some scalar c
(depending on θ ), then θ(f ) is defined for every f ∈ L F1 (µ) and every
θ ∈ Mµ , and the mapping


(θ, f ) 7→ θ(f ) θ ∈ Mµ , f ∈ LF1 (µ)

is bilinear.
Proof. It is clear that Mµ is a vector space. If f ∈ LF1 (µ) and (gn ) is
a sequence in KF such that µ(f − gn ) → 0 then, for every θ ∈ Mµ ,

|θ|*(f − gn ) → 0 ,

whence f ∈ LF1 (θ) and θ(f ) = limn θ(gn ) in F . Thus, if θ1 , θ2 ∈ Mµ then

(θ1 + θ2 )(f ) = lim(θ1 + θ2 )(gn ) = lim θ1 (gn ) + θ2 (gn )
n n
= lim θ1 (gn ) + lim θ2 (gn ) = θ1 (f ) + θ2 (f ) ,
n n

and similarly (aθ)(f ) = a · θ(f ) for all scalars a . ♦

It follows at once that if µ and µ0 are positive measures on T and if
f ∈ LF1 (µ + µ0 ) , then f ∈ LF1 (µ) ∩ LF1 (µ0 ) and (µ + µ0 )(f ) = µ(f ) + µ0 (f ) .
The converse is contained in the following:
Theorem. Let θ, θ 0 be any two complex measures on the locally compact
space T , let F be a Banach space, and suppose f ∈ L F1 (θ) ∩ LF1 (θ 0 ) . Then
f ∈ LF1 (θ + θ 0 ) and (θ + θ 0 )(f ) = θ(f ) + θ 0 (f ) .
Proof. Since LF1 (θ) = LF1 (|θ|) and LF1 (θ 0 ) = LF1 (|θ 0 |) (Ch. IV, §3,
No. 4, remark following Def. 2), one has |θ|*(f ) < +∞ and |θ 0 |*(f ) < +∞
(Ch. IV, §5, No. 6, Th. 5). Set µ = |θ|+|θ 0 | . Then µ* = |θ|*+|θ 0 |* (Ch. IV,
§1, No. 3, Prop. 15), therefore

µ*(f ) = |θ|*(f ) + |θ 0 |*(f ) < +∞ ;

by the cited Th. 5, to show that f is µ-integrable it will suffice to show that
it is µ-measurable.
To this end, we employ the criterion of Ch. IV, §5, No. 1, Prop. 1. Let
K be any compact subset of T , and let ε > 0 . Since f is |θ|-measurable,
there exists a compact set H ⊂ K such that

|θ|*(K --- H) 6 ε/2 and f H is continuous ;
§1 essential upper integral INT V.x18

similarly, there exists a compact set H 0 ⊂ K such that


|θ 0 |*(K --- H0 ) 6 ε/2 and f H0 is continuous .

Then H ∪ H0 is a compact subset of K and

K --- (H ∪ H0 ) = (K --- H) ∩ (K --- H0 ) ,

therefore




µ* K --- (H ∪ H0 ) = |θ|* K --- (H ∪ H0 ) + |θ 0 |* K --- (H ∪ H0 )
6 |θ|*(K --- H) + |θ 0 |*(K --- H0 ) 6 ε/2 + ε/2 ;

the coup de grace: f (H ∪ H0 ) is continuous (GT, Ch. I, §3, No. 2, Prop. 4),
thus f is µ-integrable.
Since |θ + θ 0 | 6 |θ| + |θ 0 | = µ (Ch. III, §1, No. 6) it follows from the
Lemma that f is (θ + θ 0 )-integrable and (θ + θ 0 )(f ) = θ(f ) + θ 0 (f ) . ♦
Remark. Recall that, for every measure ρ on T , the ρ-integrable sets
form a clan (Ch. IV, §4, No. 9, Example) that contains every compact set
(loc. cit., No. 6, Cor. 1 of Prop. 10), hence contains the clan R generated by
the compact sets; thus every set in R is ‘universally integrable’ in the sense
that it is ρ-integrable for every measure ρ on T . Thus, in hindsight, one
can omit the asterisks in the above computation (loc. cit., No. 5, Def. 2).
However, with the asterisks in place, the equality in the above display is
justified by the known property (|θ| + |θ 0 |)* = |θ|* + |θ 0 |* of outer measures
(Ch. IV, §1, No. 3, Prop. 15); to justify removing the asterisks would require
verifying the special case of the Theorem being proved, for the measures
|θ|, |θ 0 | and the (universally integrable) numerical function f = ϕ K -- (H∪H0 ) ,
which verification would bring back the asterisks.
The above Theorem extends, by induction, to finitely many complex
1 1
measures; but the analogous proposition for f ∈ L F (θ) ∩ L F (θ 0 ) , a special
case of §2, No. 2, Prop. 3, is more complicated (see (i) of Cor. 3 below for a
simpler proof).
Corollary 1. Let θ be any complex measure on T and suppose θ is
decomposed as in (∗∗) (for example, in the canonical way). Then
4
\
f∈ LF1 (θ) ⇔ f∈ LF1 (µK ) ,
k=1

in which case θ(f ) = µ1 (f ) − µ2 (f ) + iµ3 (f ) − iµ4 (f ) .

Proof. ⇒: The argument for this is given in “Proof #2” above.
⇐: This follows at once from the above Theorem. ♦
INT V.x19 integration of measures §1

Corollary 2. If θ is any real measure on T , then f ∈ L F1 (θ) if and

only if f ∈ LF1 (θ + ) ∩ LF1 (θ − ) , in which case

|θ|(f ) = θ + (f ) + θ − (f ) , θ(f ) = θ + (f ) − θ − (f ) .

Proof. By Cor. 1, LF1 (θ) = LF1 (θ + ) ∩ LF1 (θ − ) and the second formula
holds for f ∈ LF1 (θ) ; and, since LF1 (θ) = LF1 (|θ|) and |θ| = θ + + θ − , the
first formula is immediate from the Theorem. ♦

Corollary 3. Let θ1 , . . . , θn be complex measures on T .

(i) If µ = |θ1 | + · · · + |θn | then

n
\ 1 1
L F (θk ) = L F (µ) .
k=1

T
n 1 1
(ii) If θ = θ1 + · · · + θn then L F (θk ) ⊂ L F (θ) and, for every
k=1
T
n 1 R R R
f∈ L F (θk ) , one has f dθ = f dθ1 + · · · + f dθn .
k=1
T
n T
n
(iii) LF1 (θk ) = LF1 (µ) ⊂ LF1 (θ) and, for f ∈ LF1 (θk ) , one has
k=1 k=1
θ(f ) = θ1 (f ) + · · · + θn (f ) .
1 1
Proof. {Recall that for any complex measure θ , L F (θ) = L F (|θ|) by
definition.}
(i) For a function f : T → F , µ• (|f |) = |θ1 |• (|f ) + · · · + |θn |• (|f |) by
No. 1, Prop. 3, therefore µ• (|f |) < +∞ if and only if |θk |• (|f |) < +∞ for
1
every k ; in view of No. 3, Prop. 9, to show that f ∈ L F (µ) if and only if
1
f ∈ L F (θk ) for all k , it will suffice to show that

f is µ-measurable ⇔ f is θk -measurable for k = 1, . . . , n .

To this end, recall that µ* = |θ1 |* + · · · + |θn |* (Ch. IV, §1, No. 3, Prop. 15).
⇒: If f is µ-measurable, it is clear from |θ k |* 6 µ* and the definition
of measurability that f is θk -measurable.
⇐: Assuming f is θk -measurable for all k , we are to show that f is
µ-measurable. As in the proof of the Theorem, we employ the criterion of
Ch. IV, §5, No. 1, Prop. 1: if K ⊂ T is compact and  ε > 0 , there exist
compact sets Hk ⊂ K (k = 1, . . . , n) such that f Hk is continuous and
§1 essential upper integral INT V.x20

|θk |*(K --- Hk ) 6 ε/n ; then f (H1 ∪ · · · ∪ Hn ) is continuous and
n
[ n
X n
[



µ∗ K --- Hj = |θk |* K --- Hj
j=1 k=1 j=1
Xn n
\

= |θk |* (K --- Hj )
k=1 j=1
Xn
6 |θk |*(K --- Hk ) 6 ε ,
k=1

thus f is µ-measurable by the cited Prop. 1.

T
n 1 1
(ii) Let f ∈ L F (θk ) = L F (µ) . Since |θ| 6 |θ1 | + · · · + |θn | = µ ,
k=1
1
it follows that f ∈ L F (θ) ; for, f is µ-measurable and |θ|• (|f |) 6 µ• (|f |) <
+∞ (Prop. 9), and from |θ|* 6 µ* it follows from the definition of mea-
surability that f is θ-measurable, consequently f is essentially θ-integrable
(Prop. 9 again). Say g ∈ LF1 (µ) with f = g locally µ-almost every-
where, hence locally θ-almost everywhere. Since |θ k |• 6 µ• , f = g locally
θk -almost everywhere for every k . Moreover, from g ∈ L F1 (µ) and |θk | 6 µ
T
n
we know that g ∈ LF1 (θk ) , thus g ∈ LF1 (θk ) and θk (f ) = θk (g) ; by
k=1
recursion on the Theorem, g ∈ LF1 (θ) and
θ(g) = θ1 (g) + · · · + θn (g) ,
whence θ(f ) = θ1 (f ) + · · · + θn (f ) .
Tn P
n
(iii) If f ∈ LF1 (θk ) then f ∈ LF1 (θ) and θ(f ) = θk (f ) follow
k=1 k=1
recursively from the above Proposition. The inclusion L F1 (θ) ⊂ LF1 (µ)
follows from |θ| 6 µ and the Lemma to the Proposition.
One has LF1 (θ) = LF1 (|θ|) and similarly for the θk (Ch. IV, §3, No. 4,
line following Def. 2). Let f : T → F be any function. We know from
µ* = |θ1 |* + · · · + |θn |* that µ*(|f |) is finite if and only if |θ k |*(|f |) is finite
for every k ; and, by the proof of (i), f is µ-measurable if and only if f is
θk -measurable for every k ; thus f is µ-integrable if and only if it is θ k -inte-
Tn
grable for every k (Ch. IV, §5, No. 6, Th. 5), that is, LF1 (θk ) = LF1 (µ).
k=1
1
Incidentally, since f ∈ LF1 (θ)
if and only if f ∈ L F (θ) and f is
θ-moderated (Cor. of Prop. 9), one can derive (iii) from (i) and (ii) based on
the observation that f is µ-moderated if and only if it is θ k -moderated for
k = 1, . . . , n ; equivalently, writing A = {t : f (t) 6= 0 } ,
(†) A is µ-moderated ⇔ A is θk -moderated for k = 1, . . . , n .
INT V.x21 integration of measures §1

⇒: Immediate from |θk |* 6 µ* and criterion c) of No. 2, Prop. 5.

⇐: By the cited criterion c), for each k there exist a θ k -negligible set
Nk and a sequence (Kki ) of compact subsets of T such that
∞
[
A ⊂ Nk ∪ Kki ,
i=1

S
∞ T
n
that is, A --- Kki ⊂ Nk ; then the set N = Nk is θk -negligible for
i=1 k=1
every k , hence is µ-negligible, and
n [
[ ∞  n 
\ ∞
[  n
\
A --- Kki = A --- Kki ⊂ Nk = N ,
k=1 i=1 k=1 i=1 k=1

n  S
S ∞ 
thus A ⊂ N ∪ Kki and so A is µ-moderated.
k=1 i=1
Application of (†) to (iii): the equivalences
n
\ 1
f∈ LF1 (θk ) ⇔ f ∈ L F (θk ) and f is θk -moderated (k = 1, . . . , n)
k=1
1
⇔ f ∈ L F (µ) and f is µ-moderated
⇔ f ∈ LF1 (µ)

hold, respectively, by the Cor. of Prop. 9; by (i) and (†); and the Cor. of
Prop. 9. ♦
V.10, `. −7, −6.
“ ν*(f ) 6 ν(g) 6 ε , or λ∗α (f ) > λ*(f ) − ε (Ch. IV, §1, No. 3,
Prop. 15).”
Since 0 6 f 6 g , one has ν*(f ) 6 ν*(g) = ν(g) = λ(g) − λ α (g) 6 ε by
the choice of g . By the cited Prop. 15, λ* = (λ α + ν)* = λ∗α + ν* , therefore
λ*(f ) − λ∗α (f ) = ν*(f ) 6 ε , whence the second asserted inequality.
V.10, `. −1.
“ λ• (f ) = sup λ• (fn ) = sup sup λ•α (fn ) = sup sup λ•α (fn ) = sup λ•α (f ) . ”
n∈N n∈N α∈A α∈A n∈N α∈A

1st equality: No. 1, Prop. 1.

2nd equality: By the special case just proved.
3rd equality: Associativity of sups.
4th equality: No. 1, Prop. 1.
§1 essential upper integral INT V.x22

V.11, `. 3, 4.
λ• (f ) = sup λ• (f ϕK ) = sup sup λ•α (f ϕK )
K∈K K∈K α∈A
= sup sup λα (f ϕK ) = sup λ•α (f ) .
•
α∈A K∈K α∈A

1st equality: Definition of λ• , where λ• (f ϕK ) = λ*(f ϕK ) by No. 2,

Prop. 7.
2nd equality: By the special case just proved.
3rd equality: Associativity of sups.
4th equality: Definition of λ•α .
V.11, `. 5, 6.
“Corollary 1.”
By Prop. 11 applied to f = ϕN , λ• (N) = sup λ•α (N) . Thus (p. V.2,
α∈A
`. 11, 12)

N locally λ-negligible ⇔ λ• (N) = 0

⇔ λ•α (N) = 0 for all α
⇔ N locally λα -negligible for all α .

V.11, `. 10, 11.

“The condition is obviously necessary, since λ α 6 λ for every α
(Ch. IV, §1, No. 3, Prop. 15).”
By the cited Prop. 15, λ∗ 6 λ* , therefore
α

g λ-negligible ⇒ λα -negligible ,
whence g λ-measurable ⇒ g λα -measurable (Ch. IV, §5, No. 1, comment
following Def. 1).
V.11, `. 13.
“ . . . let L be a compact set such that . . . ”
The assumption that L be compact is unnecessary and confusing.
Assume g is λα -measurable for all  α , and let K be the set of all
compact subsets K of T such that g K is continuous. To show that g
is λ-measurable, it suffices to show that K is λ-dense (Ch. IV, §5, No. 10,
Prop. 15, criterion a), with A = X = T ); to this end, it will suffice to show
that K satisfies criterion a) of Ch. IV, §5, No. 8, Prop. 12. Thus, given an
arbitrary subset L of T , we are to show that

L locally λ-negligible ⇔ L ∩ K λ-negligible for every K ∈ K .

INT V.x23 integration of measures §2

⇒: Immediate from Ch. IV, §5, No. 2, Prop. 5.

⇐: By Cor. 1 it suffices to show that, given any α , L is locally
λα -negligible. Since g is λα -measurable, K is λα -dense by “d) ⇒ a)” of
the cited Prop. 15. For every K ∈ K , by assumption L ∩ K is λ-negligible,
hence λα -negligible (because λα 6 λ ); since K is λα -dense, it follows from
criterion a) of the cited Prop. 12 that L is locally λ α -negligible.

§2. SUMMABLE FAMILIES OF POSITIVE MEASURES

V.12, `. −9, −8.

“This follows at once from Remark 1, Prop. 11 of §1, No. 3 and Prop. 3
of §1, No. 1.”
P
For every finite set J ⊂ A , write νJ = λα . By the cited Remark 1,
α∈J
ν = sup νJ in M+ (T) , in the sense that νJ 6 ν for every J , and ν 6 ν 0 for
J
every measure ν 0 such that νJ 6 ν 0 for all J ; equivalently, ν(g) = sup νJ (g)
J
in R+ for every g ∈ K+ (T) . In other words, ν = sup νJ in the fully lattice-
J
ordered Riesz space M (T; R) (Ch. III, §1, No. 5, Th. 3).
Then, for every f ∈ F+ (T) , ν • (f ) = sup νJ• (f ) by the cited Prop. 11,
P • J P •
where νJ• (f ) = λα (f ) by the cited Prop. 3, thus ν • (f ) = λα (f ) .
α∈J α

V.12, `. −7 to −5.
“Corollary 1”
For every subset M of T ,
X
ν • (M) = λ•α (M)
α∈A

by Prop. 1 (with f = ϕM ). If M is ν-integrable then, since λ ∗α 6 ν*

(Ch. IV, §1, No. 3, Prop. 15), M is λα -integrable (Ch. IV, §4, No. 6, Cor. 1
of Th. 4) for every α ; since M is ν-moderated (first sentence of §1, No. 3),
ν • (M) = ν*(M) = ν(M) (§1, No. 2, Prop. 7), and similarly for λ α . Thus
X
ν(M) = λα (M)
α∈A

for every ν-integrable set M , and in particular for every compact set, or
relatively compact open set (Ch. IV, §4, No. 6, Cor. 1 of Prop. 10).
§2 summable families of positive measures INT V.x24

V.12, `. −4, −3.

“Corollary 2”
One knows that N is locally ν-negligible
P if and only if ν • (N) = 0
(p. V.2, `. 11, 12), and ν • (N) = λ•α (N) (put f = ϕN in Prop. 1),
α∈A
whence the assertion.
V.12, `. −2, −1.
“Corollary 3”
If f is not ν-moderated then ν*(f ) = +∞ (No. 2, Prop. 7), so (4)
holds trivially.
If f is ν-moderated then f is λα -moderated for every α . For, writing
A = {t : f (t) 6= 0 } , we know that
[
A ⊂N∪ Kn ,
n

where (Kn ) is a sequence of compact sets and ν*(N) = 0 (No. 2, Def. 2);
then, for every α , λα * 6 ν* , therefore λ∗α (N) = 0 and so f is
λα -moderated. By the cited Prop. 7, ν • (f ) = ν*(f ) and λ•α (f ) = λ∗α (f )
for all α , so it follows from (3) that (4) holds with equality.
V.13, `. 11.
“This follows at once from Cor. 2 of Prop. 11 of §1.”
It does, provided one has verified the proposition for finite sums of
the λα . The crux of the matter: If µ, µ0 are positive measures on T , and
f : T → G , then
f is (µ + µ0 )-measurable ⇔ f is measurable for µ and for µ 0 .
From (µ + µ0 )* = µ* + µ0 * one sees that a subset N of T is negligible
for µ + µ0 if and only if it is negligible for each of µ, µ 0 .
Proof of ⇒: Immediate from the preceding remark and the definition
of measurability (Ch. IV, §5, No. 1, Def. 1).
Proof of ⇐: This is shown in the proof of the Theorem in the Note for
p. V.10, `. 13, 14 (see also Cor. 3 of the cited Theorem).
V.13, `. −10, −9.
“The first part of the statement therefore follows at once from Props. 2
and 1.”
At any rate,
X
(∗) λ•α (|f |) = ν • (|f |)
α∈A

by Prop. 1.
INT V.x25 integration of measures §2

Suppose f is essentially ν-integrable. Thus f is ν-measurable and

•
ν (|f |) < +∞ (§1, No. 3, Prop. 9); then, for every α ∈ A , f is
λα -measurable (Prop. 2) and λ•α (|f |) < +∞ by (∗), therefore f is essentially
λα -integrable (§1, Prop. 9).
P Conversely,
•
suppose f is essentially λ α -integrable for all α , and that
λα (|f |) < +∞ . Then, for every α , f is λα -measurable (§1, Prop. 9),
α∈A
therefore f is ν-measurable (Prop. 2) and ν • (|f |) < +∞ by (∗), therefore f
isPessentially ν-integrable (§1, Prop. 9). We may then write (∗) as ν(|f |) =
λα (|f |) , or
α∈A
Z XZ
|f | dν = |f | dλα
α∈A

(§1, No. 3, Def. 3).

V.13, `. −6 to −4.
“The set of f ∈ LF1 (ν) that satisfy (6) is thus a closed linear subspace
H of LF1 (ν) ”
Note the absence of overbars; we are dealing with integrable functions.
For g ∈ LF1 (ν) write
Z XZ
Φ(g) = g dν , Ψ(g) = g dλα ,
α∈A

both of which are defined and are elements of F ; Φ and Ψ define linear
mappings LF1 (ν) → F .
Φ is continuous for convergence in mean: for, by the definition of inte-
gral, it is the extension to LF1 (ν) of the mapping g 7→ ν(g) (g ∈ KF (T) )
by continuity with respect to the semi-norm N 1 = ν* (Ch. IV, §4, No. 1,
Def. 1).
Ψ is likewise continuous for convergence in mean, by virtue of the com-
putation

X  Z  XZ
 X
|Ψ(g)| 6  g dλα  6 |g| dλα = λα (|g|)
α∈A α∈A α∈A
X
= λ•α (|g|) •
= ν (|g|) = ν(|g|) .
α∈A

Since F is Hausdorff, it is elementary that the linear subspace

H = {g ∈ LF1 (ν) : Φ(g) = Ψ(g) } is closed in LF1 (ν) (GT, I, §8, No. 1,
Prop. 2).
§2 summable families of positive measures INT V.x26

V.13, `. −4 to −2.
“ . . . it contains the functions of the form f · a , where a ∈ F and f
denotes a finite integrable positive function (Prop. 1).”
Assuming f ∈ LR1 (ν) , f > 0 , the assertion is that the function f = f ·a
satisfies (6). R R

One has f ·a ∈ LF1 (ν) and (f ·a) dν = f dν ·a (Ch. IV, §4, No. 2,
Cor.
R 2 of Th. 1);R from
λα 6 ν it follows that f ∈ LR1 (λα ) , consequently
(f · a) dλα = f dλα · a . By Prop. 1,

X
ν • (f ) = λ•α (f ) ,
α∈A

and since f is integrable (hence moderated) for ν and the λ α , this may be
written Z XZ
f dν = f dλα .
α∈A

Then Z Z  XZ 
(f · a) dν = f dν · a = f dλα · a
α∈A
X Z  XZ
= f dλα · a = (f · a) dλα
α∈A α∈A

(the third equality, by the continuity of scalar multiplication in F ), thus

f ·a∈H .
If f ∈ KR (T) then f · a = f + · a − f − · a ∈ H by the foregoing;
it follows that H is dense in LF1 (ν) for the topology of convergence in
mean (Ch. IV, §3, No. 5, Prop. 10), whence L F1 (ν) = H = H (the overbar
signifying closure for that topology).

V.13, `. −2, −1.

“Therefore H = LF1 (ν) and the proposition is established.”
1
For the equality, see the preceding note. Given f ∈ L F (ν)R, choose
g
R ∈ LF1 (ν) such that f = g locally ν-almost everywhere; thus f dν =
g dν . For each α ∈ A , one has λα 6 ν , therefore g ∈ LF1 (λα ) (because
∗
λα 6 ν* ) and f = g locally λα -almost everywhere (because λ•α 6 ν • );
1 R R 1
therefore
R f ∈ LP F (λ
Rα ) and f dλα = g dλα . Since R LF (ν) =
PH we know
that g dν = g dλα , which may be written f dν = f dλα , that
α∈A α∈A
is, f satisfies (6).
INT V.x27 integration of measures §2

V.14, `. 8–11.
“Conversely, if A is finite and if f is λ α -integrable for all α ∈ A ,
then f is essentially ν-integrable by Prop. 3, and it sufficesPto verify that
ν*(|f |) < +∞ ; this follows at once from the relation ν* = λ∗α (Ch. IV,
α∈A
§1, No. 3, Prop. 15).”
The proof cites Prop. 3, which cites Prop. 2, which cites Cor. 2 of
Prop. 11 of §1, whose proof employs the concept of a λ-dense set K of
compact sets.
For A finite, there is a simpler proof: in the Note for p. V.10, `. 13, 14,
see Cor. 3, (iii) of the Theorem proved there.
P •
But for arbitrary A , the additivity property ν • = λα and the
α∈A
concept of λ-dense sets K prove their worth.

V.14, `. 12–17.
“Corollary 2”
See also the note for p. V.10, `. 13, 14 (Cor. 1 of the Theorem proved
there).

V.15, `. 3, 4.
“ . . . the linear form µα on K (T) is positive, therefore is a positive
measure, with support contained in K α .”
That µα is a measure is shown in Ch. III, §1, No. 5, Th. 1. 
To show that Supp(µα ) ⊂ Kα , it suffices to show that µα  { Kα = 0
(Ch. III, §2, No. 2, Def. 1 and the sentence before it). Indeed, if f ∈ K (T)
and Supp(f ) ⊂ { Kα then f ϕKα = 0 and so µα (f ) = µ(f ϕKα ) = 0 ; thus
µα  { Kα = 0 (loc. cit., No. 1; for every g ∈ K ( { Kα ), the extension by 0
of g to T is such a function f ).
{The earlier-cited Prop. 14 (Ch. IV, §5, No. 9) also shows that one can
suppose that Supp(µKα ) = Kα ; an alternate proof of Prop. 4 based on the
measures µKα is given below in the Note for V.15, `. 11, 12.}

V.15, `. 8, 9.
“ . . . let A0 be the countable set formed by the α ∈ A such that
S ∩ Kα 6= ∅ .”
Remark following Ch. IV, §5, No. 9, Def. 7.

V.15, `. 9, 10.
“ . . . the set N ∩ S is µ-negligible ”
Ch. IV, §5, No. 2, Prop. 5.
§2 summable families of positive measures INT V.x28

V.15, `. 11, 12.

X X
“ µ(f ) = µ(f ϕS ) = µ(f ϕS∩Kα ) = µ(f ϕKα )
α∈A0 α∈A0
X X
= µ(f ϕKα ) = µα (f ) . ”
α∈A α∈A

1st equality: f = f ϕS by the definition of S .

2nd equality: One has
 [ 
S= S∩T = S∩ N∪ Kα
α∈A
[
= (S ∩ N) ∪ (S ∩ Kα )
α∈A
[
= (S ∩ N) ∪ (S ∩ Kα )
α∈A0

(because S ∩ Kα = ∅ when α ∈ / A0 ). It follows from the disjointness of the

terms of the union that
X
f = f ϕS = f ϕS∩N + f ϕS∩Kα ;
α∈A0

since S ∩ N is negligible, so is f ϕS∩N , thus all terms are µ-integrable, and

every sum of finitely many terms on the right is positive and 6 f , therefore
X
µ(f ) = 0 + µ(f ϕS∩Kα )
α∈A0

(Ch. IV, §1, No. 3, Th. 3 or Ch. IV, §4, No. 3, Cor. 2 of Th. 2).
3rd equality: f ϕS∩Kα = (f ϕS )ϕKα = f ϕKα .
4th equality: For α ∈ A --- A0 one has S ∩ Kα = ∅ , whence f ϕKα =
(f ϕS )ϕKα = f ϕS∩Kα = f · 0 = 0 .
Proof of Prop. 4 based on the measures µ Kα . For each α ∈ A , the
correspondence 
f 7→ µKα (f Kα ) (f ∈ K (T))
defines a positive linear form µ0α on K (T) , that is, a positive measure
on T . Then Supp(µ0α ) ⊂ Kα ; for, if f ∈ K (T) and Supp(f ) ⊂ { Kα , then
f Kα = 0 and µ0α (f ) = µKα (f Kα ) = 0.
Fix f ∈ K+ (T) and define A0 and S as above. One argues, as above,
that X
µ(f ) = µ(f ϕKα ) ;
α∈A0
INT V.x29 integration of measures §3
  
since f ϕKα = (f
Kα )0 , where
 (f Kα )0 is the extension by 0 of f Kα
to T , µ (f Kα )0 = µKα (f Kα ) by the definition of µKα (Ch. IV, §5,
No. 7, Def. 4), and f Kα = 0 for α ∈ A --- A0 , one has
X X
µ(f ) = µ0α (f ) = µ0α (f ) .
α∈A0 α∈A

Varying f ∈ K+ (T) (and, along with it, A0 ), this proves that

X
µ= µ0α .
α∈A

Finally, if α ∈ A and f ∈ K+ (T) then

 

µ0α (f ) = µKα (f Kα ) = µ (f Kα )0 = µ(f ϕKα ) = µα (f ) ,

therefore µ0α = µα ; the two procedures yield the same result.

V.15, `. 15, 16.
“ . . . let A0 be the countable set of α ∈ A such that K α intersects one
of the Ln .”
For each n , let A0n = {α ∈ A : Ln ∩Kα 6= ∅ } ; since Ln is compact
S and
(Kα )α∈A is locally countable, A0n is countable, therefore so is A0 = A0n .
n

V.15, `. 16.
/ A0 ”
“Then µα = 0 for α ∈
With A0 defined as in the preceding note,
S suppose α ∈ / SA 0 . Then
Kα ∩ Ln = ∅ for all n , thus Kα ⊂ T --- Ln ; but T --- Ln is by
n n
construction µ-negligible, therefore so is K α . Then, for
all f ∈ K (T) , the
function f ϕKα is µ-negligible, and so µα (f ) = µ f ϕKα = 0 .

§3. INTEGRATION OF POSITIVE MEASURES

V.17, `. 1–4.
“For, verifying that Λ is scalarly essentially integrable for a positive
measure η on T comes down to verifying that t 7→ λ t (g) is η-measurable
§3 integration of positive measures INT V.x30

and admits a finite essential upper integral, with respect to η , for every
function g ∈ K+ (X) .”
Prop. 9 of §1, No. 3. In what follows, the role of η is played by µ and
the µi .
V.17, `. 4, 5.
“The proposition therefore follows at once from Prop. 11 of §1, No. 4
and its Corollary 2.”
Suppose Λ is scalarly essentially µ-integrable. Given g ∈ K + (X) , let
gb : T → R+ be the function defined by gb(t) = λt (g) ; by assumption,
1
gb ∈ L R (µ) , thus gb is µ-measurable and µ• (b
g ) < +∞ (§1, No. 3, Prop. 9).
Since µ = sup µi , it follows from the cited Prop. 11 that
i∈I

sup µ•i (b
g ) = µ• (b
g ) < +∞ .
i∈I

Moreover, for every i ∈ I , gb is µi -measurable by the cited Cor. 2, and since

µ•i (b
g ) 6 µ• (b
g ) < +∞ , gb is essentially µi -integrable by the cited Prop. 9;
thus, since g ∈ K+ (X) is arbitrary, Λ is scalarly essentially µ i -integrable.
Writing Z Z
ν = λt dµ(t) , νi = λt dµi (t) ,

we have, for all g ∈ K+ (X) ,

Z Z
ν(g) = λt (g) dµ(t) = gb dµ
Z
•
= µ (b
g) = sup µ•i (b
= sup gb dµi
g)
i∈I i∈I
Z
= sup λt (g) dµi (t) = sup νi (g) ;
i∈I i∈I

thus ν = sup νi , which is the relation (2).

i∈I
Conversely, suppose that for
R every i ∈ I , Λ is scalarly essentially
µi -integrable and, writing νi = λt dµi (t) , suppose that

sup νi (g) < +∞ for every g ∈ K+ (X)

i∈I

(a condition equivalent to the existence of a measure ρ ∈ M (X) such that

νi 6 ρ for all i ∈ I ). Let g ∈ K+ (X) and define gb(t) = λt (g) (t ∈ T) .
By hypothesis, for every i ∈ I , gb is essentially µ i -integrable, hence gb is
INT V.x31 integration of measures §3

µi -measurable and µ•i (b

g ) < +∞ ; therefore gb is µ-measurable by the cited
Cor. 2, and since
Z
• •
µ (b g ) = sup gb dµi = sup νi (g) < +∞ ,
g ) = sup µi (b
i∈I i∈I i∈I

gb is essentially µ-integrable. Since g ∈ K + (X) is arbitrary, we conclude

that Λ is scalarly essentially µ-integrable.
V.17, `. 6–11.
“Corollary”
P 1
Consider first the case that µ = µα with A finite. Then L (µ) =
α∈A
T 1
L (µα ) and
α∈A
X 1
µ• (f ) = µ•α (f ) for all f ∈ L (µ)
α∈A

(a special case of §2, No. 2, Prop. 3; cf. Cor. 3 of the Theorem in the Note
for V.10, `. 13–14). Given a mapping Λ : t 7→ λ t ∈ M+ (X) , for g ∈ K+ (X)
define gb : T → R+ by gb(t) = λt (g) . The following conditions are equivalent:
a) Λ is scalarly essentially µ-integrable;
1
b) gb ∈ L (µ) for all g ∈ K+ (X) ;
1
c) gb ∈ L (µα ) for all α ∈ A and all g ∈ K+ (X) ;
d) Λ is scalarly essentially µα -integrable
R for all α ∈ A , R
in which case the positive measures ν = λt dµ(t) and να = λt dµα (t)
are defined, where, for all g ∈ K+ (X) ,
Z X XZ X
ν(g) = λt (g) dµ(t) = µ• (b g) = µ•α (b
g) = λt (g) dµα (t) = να (g),
α∈A α∈A α∈A
P
that is, ν = να . This proves the Corollary for the case that A is finite.
α∈A P
In the general case, for every finite set J ⊂ A write µ J = µα ;
α∈J
then (µJ ) is an increasing directed family of positive measures, and we
are assuming that the family admits µ as supremum. We know from the
foregoing that Λ is scalarly essentially µ α -integrable for all α ∈ A if and
only if it isP
scalarly essentially µ J -integrable for all finite J ⊂ A , in which
case νJ = να for every J , that is,
α∈J
Z XZ
λt dµJ (t) = λt dµα (t) .
α∈J
§3 integration of positive measures INT V.x32

By Prop. 1, the following conditions are then equivalent:

(i) Λ is scalarly essentially µ-integrable;
(ii) Λ is scalarly essentially
  P RµJ -integrable  for every finite J ⊂ A , and
R
the family λt dµJ (t) = λt dµα (t) indexed by the J ’s is bounded
α∈J
above in M+ (X) ;
 R (iii) Λ is scalarly essentially µα -integrable for all α , and the family
λt dµα (t) is summable in M+ (X) in the sense of §2, No. 1,
α∈A
in which case, for every g ∈ K+ (X) ,
Z X XZ X
ν(g) = λt (g) dµ(t) = µ• (bg) = µ•α (b
g) = λt (g) dµα (t) = να (g)
α∈A α∈A α∈A

(the third equality by §2, No.P2, Prop. 1), whence the family (ν α )α∈A is
summable in M+ (X) , with να = ν .
α∈A

V.17, `. 12, 13.

“It follows immediately that every scalarly essentially µ-integrable map-
ping is also scalarly essentially µ 0 -integrable for every measure µ0 6 µ .”
Overkill. Writing µ00 = µ − µ0 , the assertion follows from µ = µ0 + µ00
and the case of the Corollary for A finite.
A more direct proof: Regard Λ as a mapping F + (X) → F+ (T) , where,
for g ∈ F+ (X) , one writes Λ(g) for the function t 7→ λ •t (g) (t ∈ T); in
1
particular, when g ∈ K+ (X) , Λ(g) is the function gb ∈ L (µ) defined by
gb(t) = λt (g) as in the preceding note. To say that Λ is scalarly essentially
µ-integrable means that

1
Λ K+ (X) ⊂ L (µ) ;

thus the assertion is immediate from the fact that

1 1
0 6 µ0 6 µ ⇒ L (µ) ⊂ L (µ0 )

(see the Note for V.10, `. 13, 14, proof of (i) in Cor. 3 of the Theorem).
V.17, `. 17–25.
“Definition 1.”
As in the preceding note, regard the scalarly essentially
R µ-integrable
mapping Λ as a mapping F+ (X) → F+ (T) , let ν = λt dµ(t) and, as
in Ch. IV, §1, No. 1, write I+ (X) for the set of all lower semi-continuous
functions f > 0 on X . If f ∈ I+ (X) then f is universally measurable
(i.e., measurable for every measure on X ) by Ch. IV, §5, No. 5, Cor. of
INT V.x33 integration of measures §3

Prop. 8; moreover, ν • (f ) = ν*(f ) and λ•t (f ) = λ∗t (f ) for every t ∈ T (§1,

No. 1, Prop. 4), and
Z ∗

• ∗
Λ(f )(t) = λt (f ) = λt (f ) = f dλt .

Thus, to say that Λ is µ-pre-adequate means that for every f ∈ I + (X) the
function Λ(f ) is µ-measurable and


ν • (f ) = µ• Λ(f ) ,

which may also be written (when f ∈ I+ (X) ) as

Z ∗ Z • Z ∗
f (x) dν(x) = dµ(t) f (x) dλt (x) .

It follows that, for every positive measure µ 0 6 µ , Λ is scalarly R essentially

µ0 -integrable (by the preceding note)—hence the measure ν 0 = 0

λt dµ• (t) is
defined—and, for every f ∈ I+ (X) , the function t 7→ Λ(f ) (t) = λt (f ) =
λ∗t (f ) is µ0 -measurable (§2, No. 2, Prop. 2); thus, for the µ-pre-adequate
mapping Λ to be µ-adequate, it is necessary and sufficient that for every
positive measure µ0 6 µ and every f ∈ I+ (X) ,
Z • Z • Z •
0 0
f (x) dν (x) = dµ (t) f (x) dλt (x) ,

which may also be written (for f ∈ I+ (X) ) as

Z ∗ Z • Z ∗
f (x) dν 0 (x) = dµ0 (t) f (x) dλt (x) ,

or ν 0 *(f ) = µ0• Λ(f ) , where Λ(f ) (t) = λ•t (f ) = λ∗t (f ) for all t ∈ T .
V.17, `. −6 to −4.
R “It can be shown that if Λ is µ-pre-adequate and if the measure ν =
λt dµ(t) is moderated—in particular if X is countable at infinity—then
Λ is µ-adequate (Exer. 7) ”
The exercise (p. V.97):
7) a) Let Λ : t 7→ λt be a Rscalarly essentially integrable mapping
of T into M+ (X) , and let ν = λt dµ(t) . Show that for every lower
semi-continuous function f > 0 defined on X ,
Z • Z • Z •
f (x) dν(x) 6 dµ(t) f (x) dλt (x) .
§3 integration of positive measures INT V.x34

b) Suppose that Λ is µ-pre-adequate,R denote by µ 0 a positive meas-

ure 6 µ , and write µ = µ0 + µ00 , ν 0 = λt dµ0 (t) . Deduce from a) that,
for every lower semi-continuous function f > 0 that is ν-integrable,
Z • Z • Z •
0 0
f (x) dν (x) = dµ (t) f (x) dλt (x) .

Extend this result to a lower semi-continuous function f > 0 that is

ν-moderated. From this, deduce that if ν is a moderated measure (in par-
ticular, if X is countable at infinity), then Λ is µ-adequate.
Proof of a): Assuming f ∈ I+ (X) , we are to show that


ν • (f ) 6 µ• Λ(f ) ,


where Λ(f ) (t) = λ•t (f ) for all t ∈ T . Since f is lower semi-continuous,
we know that f is measurable with respect to every measure (Ch. IV, §5,
No. 5, Cor. of Prop. 8), and that ν • (f ) = ν*(f ) and λ•t (f ) = λ∗t (f ) for
every t ∈ T (§1, No. 1, Prop. 4). The key idea: since f is lower semi-
continuous,
ν*(f ) = sup ν(g)
g∈K+ (X), g6f

(Ch. IV, §1, No. 1, Def. 1 and the remark following Def. 3 of No. 3).
Let g ∈ K+ (X) , g 6 f . Then, for every t ∈ T
, λ•t (g) 6 λ•t
(f ) ; that
is, Λ(g) 6 Λ(f ) pointwise on T , therefore µ • Λ(g) 6 µ• Λ(f ) . On the
other hand, by the definition of ν ,



ν(g) = µ• (b
g ) = µ• Λ(g) 6 µ• Λ(f )

(when g ∈ K+ (X) we may use the notation gb in place of Λ(g) ) and finally


ν • (f ) = ν*(f ) = sup ν(g) 6 µ• Λ(f ) .
g∈K+ (X), g6f

Proof of b): Writing µ00 = µ − µ0 , from µ = µ0 + µ00 we know that Λ

is scalarly
R essentially integrable for both µ 0 and µ00 , so we can also define
ν = λt dµ00 (t) . The key observation is that
00

(†) ν 0 + ν 00 = ν .

For, if g ∈ K+ (X) we know from µ = µ0 + µ00 that

1 1 1
gb ∈ L (µ) = L (µ0 ) ∩ L (µ00 ) and µ• (b
g ) = µ0• (b
g ) + µ00• (b
g)
INT V.x35 integration of measures §3

(see, e.g., the Note for p. 17, `. 6–11), therefore, citing the definition of ν, ν 0
and ν 00 ,

ν(g) = µ• (b
g ) = (µ0 + µ00 )• (b
g ) = (µ0• + µ00• )(b
g)
0• 00•
= µ (b g ) + µ (b g ) = ν (g) + ν (g) = (ν + ν 00 )(g) ,
0 00 0

whence (†). {A trivial special case of the Cor. of Prop. 1.}

Let f ∈ I+ (X) . Since Λ is µ-pre-adequate, we know that


ν • (f ) = µ• Λ(f ) .

Assuming in addition that f is ν-integrable—equivalently, essentially

ν-integrable (§1, No. 1, Prop. 4)—hence also ν 0 -integrable and ν 00 -integrable,
we are to show that

ν 0• (f ) = µ0• Λ(f ) .
At any rate, by Part a) we have



(††) ν 0• (f ) 6 µ0• Λ(f ) and ν 00• (f ) 6 µ00• Λ(f ) ,

therefore




ν • (f ) = µ• Λ(f ) = (µ0 + µ00 )• Λ(f ) = (µ0• + µ00• ) Λ(f )



= µ0• Λ(f ) + µ00• Λ(f )
> ν 0• (f ) + ν 00• (f ) = (ν 0• + ν 00• )(f )
= (ν 0 + ν 00 )• (f ) = ν • (f )

(the inequality, by (††), and the last equality by (†)); we thus have equality
throughout, and since ν • (f ) = ν*(f ) < +∞ the inequalities in (††) must in
fact be equalities. This proves the first assertion of b).
Suppose now that f is merely a Rν-moderated lower semi-continuous
function > 0 . With µ0 6 µ and ν 0
= λt dµ0 (t) as above, we are to show
that the equality ν 0• (f ) = µ0• Λ(f ) again holds. The key idea (§1, No. 2,
Prop. 5, a)): every ν-moderated set is contained in the union of a sequence
of µ-integrable open sets.
Let A = {x ∈ X : f (x) 6= 0 } . By assumption there exists a
ν-moderated set B ⊂ X such that f = 0 on { B , in other words A ⊂ B ;
since subsets of moderated sets are moderated, it is the same to say that A
is ν-moderated. LetS(Un ) be a sequence of ν-integrable open sets in X
such that A ⊂ U = Un ; we can suppose that the sequence is increasing.
n
The functions f ϕUn are lower semi-continuous (GT, IV, §7, No. 2, Cor.
of Prop. 1 and Prop. 2) and the sequence (f ϕ Un ) is increasing; if x ∈ A
§3 integration of positive measures INT V.x36

then x ∈ U , so ϕUn (x) = 1 from some index onward, whereas if x ∈ / A

then (f ϕUn )(x) = 0 = f (x) for all n , consequently the sequence has upper
envelope f , concisely f ϕUn ↑ f pointwise on X . Define

fn = inf(n, f ϕUn ) ,

which is also lower semi-continuous (GT, loc. cit.), and f n ↑ f pointwise

in X .
Being lower semi-continuous, fn is ν-measurable, and, since ν*(fn ) 6
nν*(Un ) < +∞ , fn is ν-integrable, therefore, by the preceding case,


(∗) ν 0• (fn ) = µ0• Λ(fn ) (n = 1, 2, 3, . . .);

but from fn ↑ f it is immediate that λ•t (fn ) ↑ λ•t (f ) for every t ∈ T (§1,
No. 1, Prop. 1), that is, Λ(fn ) ↑ Λ(f
) pointwise on T , and passage to the
limit in (∗) yields ν 0• (f ) = µ0• Λ(f ) .
R
Finally, suppose Λ is µ-pre-adequate, ν = λt dµ(t) , and let µ0 be a
positive measure 6 Rµ ; one knows that Λ is scalarly essentially µ 0 -integrable
and, defining ν 0 = λt dµ0 (t) , obviously ν 0 6 ν . As observed in the Note
for V.17, `. 17–25, in order that
Λ be µ 0 -pre-adequate, it is necessary and
sufficient that ν 0• (f ) = µ0• Λ(f ) for every f ∈ I+ (X) , a condition that is
0
satisfied (as shown above) when S every f ∈ I + (X) is ν -moderated. If ν is
moderated, so that X = N ∪ Kn with (Kn ) a sequence of compact sets
n
and N ν-negligible, then N is also ν 0 -negligible, thus ν 0 is moderated and
0
so every function on
R X is ν -moderated. Conclusion: When Λ is µ-pre-
adequate and ν = λt dµ(t) is moderated, then Λ is µ0 -pre-adequate for
every positive measure µ0 6 µ , that is, Λ is µ-adequate.
V.17, `. −4.
“ . . . it is not known if these concepts are in general equivalent.”
The question implicitly posed: Is every µ-pre-adequate mapping
Λ : t 7→ λt automatically µ-adequate? {Perhaps the matter has been settled
by now? (5-14-2007).}
Exercise 7 (see the preceding note) and Prop. B below give an affirmative
answer for some special cases, and Exercise 8 (worked out later in this Note)
effectively reduces the problem to the case that µ has compact support.
First, some simple preliminaries:
Proposition A.PLet Λ : t 7→ λt be scalarly essentially µ-integrable, and
suppose that µ = µi for some summable family (µi )i∈I in M+ (T) . If Λ
i∈I
is µi -pre-adequate for all i ∈ I , then Λ is µ-pre-adequate.
INT V.x37 integration of measures §3

Proof. Since Λ is scalarly essentiallyR µ-integrable, it is

R scalarly essen-
tially µi -integrable for all i ; write νP= λt dµ(t) P, νi = λt dµi (t) , and,
for every finite J ⊂ I , write µJ = µi , ν J = νi . Let f ∈ I+ (X)
i∈J i∈J
and write Λ(f ) for the function t 7→ λ•t (f ) (t ∈ T) . For every i ∈ I ,
Λ is
µi -pre-adequate; by Def. 1, Λ(f ) is µi -measurable and νi• (f ) = µ•i Λ(f ) . It
follows (§2, No. 2, Prop. 2) that
Λ(f ) is µ-measurable

and µ J -measurable for
• • •
all J , and

J ν (f ) = µ J Λ(f ) 6 µ Λ(f ) ; in particular when f ∈ K+ (X) ,
µ• Λ(f )
< +∞ (since Λ is scalarly essentially µ-integrable), thus the fam-
ily νJ (f ) indexed by J is bounded, so Pthe family (ν•i )i∈IPis summable P(§2,•
No. 1). By the Cor. of Prop. 1, ν = νi . Then µ = µ•i , ν • = νi
i∈I i∈I i∈I
(§2, No. 2, Prop. 1) and, for every f ∈ I + (X) ,

X •
X •
µ• Λ(f ) = µi Λ(f ) = νi (f ) = ν • (f ) ,
i∈I i∈I

thus Λ is µ-pre-adequate (Def. 1). ♦

Corollary. Suppose that Λ is scalarly essentially µ-integrable and that Λ
is µ0 -pre-adequate for every positive measure µ 0 6 µ with compact support.
Then Λ is µ-adequate.
Proof. Given any positive measure µ 0 6 µ we are to show that Λ is
0
µ -pre-adequate. At any rate, Λ is scalarly
P essentially µ 0 -integrable (V.17,
`. 12, 13). By §2, No. 3, Prop. 4, µ0 = µα for a summable family (µα )α∈A
α∈A
of positive measures on T with compact support. By hypothesis, Λ is
µα -pre-adequate for all α , therefore Λ is µ 0 -pre-adequate by Prop. A. ♦
Proposition B. Let (µi )i∈I be an increasing directed family of positive
measures on T that is bounded above in M + (T) , and let µ = sup µi .
i∈I
Suppose that the mapping Λ : t 7→ λt is scalarly essentially µ-integrable,
and that Λ is µi -pre-adequate for every i ∈ I . Then Λ is µ-pre-adequate.
R
Proof. Let ν = λt dµ(t) . By Prop. 1 of No. 1 we know Rthat Λ is
scalarly essentially µi -integrable for all i ∈ I , and, writing ν i = λt dµi (t)
for i ∈ I , (νi )i∈I is an increasing directed family for which ν serves as an
upper bound in M+ (X) , indeed ν = sup νi .
i∈I
We now make use of the assumption that Λ is µ i -pre-adequate for
all i ∈ I . Let f ∈ I+ (X) . For every i
∈ I , the function Λ(f ) : t 7→ λ •t (f )
is µi -measurable and νi• (f ) = µ•i Λ(f ) , therefore f is µ-measurable (§1,
No. 4, Cor. 2 of Prop. 11) and, by the cited Prop. 11,



ν • (f ) = sup νi• (f ) = sup µ•i Λ(f ) = µ• Λ(f ) ,
i∈I i∈I

thus Λ is µ-pre-adequate. ♦
§3 integration of positive measures INT V.x38

Exercise 8 (p. V.97):

8) Let Λ : t 7→ λt be a mapping of T into M+ (X) .
a) Let µ1 , . . . , µn be measures on T , and µ = µ1 + · · · + µn . For Λ
to be µ-adequate, it is necessary and sufficient that Λ be µ i -adequate for
every i (make use of the decomposition lemma).
b) Suppose that µ is the supremum of an increasing directed family
(µi )i∈I . For Λ to be µ-adequate, it is necessary and sufficient that Λ be
µi -adequate for all i ∈ I and scalarly essentially µ-integrable.
c) Suppose that µ is the sum of a summable family (µ j )j∈J of positive
measures. For Λ to be µ-adequate, it is necessary and sufficient that Λ be
µj -adequate for all j ∈ J and that Λ be scalarly essentially µ-integrable.
d) For Λ to be µ-adequate, it is necessary and sufficient that Λ be
scalarly essentially µ-integrable and that Λ be µ 0 -pre-adequate for every
measure µ0 6 µ with compact support.
Proof of a): Sufficiency. Suppose that Λ is µ i -adequate for all i
(1 6 i 6 n). In particular, Λ is scalarly essentially µ i -integrable for all i ,
therefore Λ is scalarly essentially µ-integrable by No. 1, Cor. of Prop. 1 for
the case of finite sums.
Given a positive measure µ0 6 µ , we are to show that Λ is µ0 -pre-
adequate (No. 1, Def. 1). By the corollary of the “decomposition theorem”
(A, VI, §1, No. 10, Cor. of Th. 1), one can write

µ0 = µ01 + · · · + µ0n , where 0 6 µ0i 6 µi (1 6 i 6 n)

(the term “decomposition lemma” is introduced at the end of Ch. II, §1,
No. 1). For every i , Λ is µ0i -pre-adequate (No. 1, Def. 1), therefore Λ is
µ0 -pre-adequate by the above Prop. A.
Necessity. If Λ is µ-adequate (in particular, scalarly essentially
µ-integrable) then, for every positive measure µ 0 6 µ , it is obvious from
the definition of adequacy that Λ is also µ 0 -adequate. In particular, Λ is
µi -adequate for every i .
Proof of b): Sufficiency. Suppose that Λ is scalarly essentially
µ-integrable and µi -adequate for all i ∈ I . Given any positive measure
µ0 6 µ , we are to show that Λ is µ0 -pre-adequate.
Define µ0i = inf(µ0 , µi ) for all i ∈ I ; then (µ0i )i∈I is an increasing
directed family whose supremum in M + (T) is equal to inf(µ0 , µ) = µ0
(A, Ch. VI, §1, No. 12, Prop. 13). Since µ 0i 6 µi , Λ is µ0i -pre-adequate for
every i ∈ I , therefore Λ is µ0 -pre-adequate by Prop. B.
Necessity. Same argument as in Part a).
Proof of c): Sufficiency. Suppose Λ is scalarly essentially µ-integrable
and µj -adequate for all j ∈ J . Let A be the set of all finite subsets
INT V.x39 integration of measures §3
P
α of J and, for every α ∈ A , define µα = µj . By assumption,
j∈α
µ = sup µα . By Part a), Λ is µα -adequate for every α ∈ A , therefore
α∈A
it is µ-adequate by Part b).
Necessity. Same argument as in Part a).
Proof of d): Sufficiency. This is the Corollary of Prop. A.
Necessity. If Λ is µ-adequate (in particular, scalarly essentially
µ-integrable) then it is µ0 -pre-adequate for every µ0 6 µ regardless of sup-
port.
V.18, `. 3–11.
“Proposition 2.”
The ingenious argument in the text merges the proofs of Parts a) and b),
at some cost in clarity of the structure of the argument. The following
rearrangement highlights structure, at some cost in repetition.
Proof of a): We employ the notations Λ(f ) and gb as in the preceding
notes
(e.g., the Note for V.17, `. −6 to −4); in particular, when f ∈ I + (X) ,
Λ(f ) (t) = λ•t (f ) = λ∗t (f ) for all t ∈ T (§1, No. 1, Prop. 4), and when
g ∈ K+ (X) , gb abbreviates Λ(g) , so that gb(t) = λ •t (g) = λt (g) for
R all t ∈ T .
Since Λ is scalarly essentially µ-integrable, the measure ν = λt dµ(t) is
defined, by the formula ν(g) = µ• (b g ) for g ∈ K+ (X) (No. 1).
The vague continuity of Λ : T → M+ (T) means that for every g ∈
K+ (X) , the function t 7→ λt (g) is continuous on T , that is, gb is continuous.
Let f ∈ I+ (X) and define F = {g ∈ K+ (X) : g 6 f } . The functions
g ∈ F form an increasing directed family of functions whose upper envelope
is f (Ch. IV, §1, No. 1, Lemma), concisely g ↑ f (pointwise on X ). By
definition,
ν*(f ) = sup ν(g)
g∈F

(loc. cit., Def. 1); similarly

λ∗t (f ) = sup λt (g) = sup gb(t) for every t ∈ T ,

g∈F g∈F

thus Λ(f ) is the upper envelope of the functions gb (g ∈ F) , and since

the gb form an increasing directed family ( g 6 g 0 ⇒ λt (g) 6 λt (g 0 ) ), we
may write gb ↑ Λ(f ) (pointwise on T ). Since the gb are continuous, Λ(t) is
lower semi-continuous on T (GT, IV, §6, No. 2, Cor. of Th. 4).
To show that Λ is µ-adequate, it suffices to show that it is µ-pre-
adequate; for, Λ is scalarly essentially µ 0 -integrable for every positive mea-
sure µ0 6 µ and will therefore be µ0 -pre-adequate by the same argument. We
already know that Λ(f ) is lower semi-continuous, therefore µ-measurable
§3 integration of positive measures INT V.x40

(Ch. IV,
§5, No. 5, Cor. of Prop. 8), hence need only show that ν • (f ) =
µ• Λ(f ) . Indeed, for all g ∈ F one has

ν • (g) = ν*(g) = ν(g) = µ• (b

g ) = µ*(b
g) ,

whence



ν • (f ) = sup ν • (g) = sup µ*(b
g ) = µ* Λ(f ) = µ• Λ(f )
g∈F g∈F

(the 1st and 3rd equalities, by §1, No. 1, Prop. 4 and Ch. IV, §1, No. 1,
Th. 1; the 2nd, by the preceding display; the 4th by §1, No. 1, Prop. 4).
Proof of b): Suppose Λ is scalarly essentially µ-integrable and vaguely
µ-measurable. As in the text, let K be
 the set, µ-dense in T , of all compact
sets K ⊂ T such that the mapping ΛK : K → M+ (X) is vaguely continuous
(Ch. IV, §5, No. 10, Prop. 15), and let P(µ α )α∈A be a summable family of
positive measures on T such that µ = µα and Supp µα ∈ K (§2, No. 3,
α∈A
Prop. 4). In particular, the restriction of Λ to Supp µ α is vaguely continuous
for all α ∈ A . To show that Λ is µ-adequate, it will suffice to show that
it is µα -adequate for every α ∈ A (see assertion c) of Exer. 8, proved in
the preceding Note). We are thus reduced to proving the following: If Λ is
scalarly essentially µ-integrable and its restriction to S = Supp µ is vaguely
continuous, then Λ is µ-adequate; since, for every positive measure µ 0 6 µ ,
Λ is µ0 -scalarly essentially integrable and Supp µ 0 ⊂ Supp µ (Ch. III, §2,
No. 2, Prop. 3), it suffices to show that Λ is µ-pre-adequate. In the text,
the argument for this is given inR the proof of Part a), as follows (in slightly
different notation). Let ν = λt dµ(t) ; by definition, ν(g) = µ• (b g ) for
g ∈ K+ (X) (§1, No. 1).
Let f ∈ I+ (X) and let F = {g ∈ K+ (X) : g 6 f } . Given any
g ∈ K+ (X) , by the vague continuity assumption the  restriction to S of the
mapping t 7→ gb(t) = λt (g) is continuous, that is, gbS is continuous; write g
(instead of hg ) for the function on T defined by
(
gb(t) for t ∈ S
g(t) =
+∞ for t ∈ T --- S .

Then g is lower semi-continuous on T ; for, given any k ∈ R , the set

{t ∈ T : g(t) 6 k } = {t ∈ S : g(t) 6 k } = {t ∈ S : gb(t) 6 k }


is closed in S (by the continuity of gbS ), hence is closed in T (because S
is closed in T ), whence the assertion (GT, IV, §6, No. 2, Prop. 1). Since
INT V.x41 integration of measures §3

g(t) = gb(t) for t ∈ S , and T --- S is µ-negligible, one has g = gb µ-almost

everywhere, therefore
µ• (g) = µ• (b
g ) = ν(g)
(§1, No. 1, Prop. 1, a), and the definition of ν ).
The functions g (g ∈ F) form an increasing directed family; let
f = sup g be its upper envelope. Since the g are lower semi-continuous,
g∈F
so is f , and for t ∈ S one has


f (t) = sup g(t) = sup gb(t) = sup λt (g) = λ∗t (f ) = λ•t (f ) = Λ(f ) (t) .
g∈F g∈F g∈F

{The 1st equality, by the definition of f ; the 2nd equality, because t ∈ S ;

the 3rd, by the definition of gb ; the 4th, by Ch. IV, §1, No. 1, Def. 1; the 5th,
by §1, No. 1, Prop. 4; the 6th, by the definition of Λ(f ) }. Thus f = Λ(f )
µ-almost everywhere; since f is µ-measurable (Ch. IV, §5, No. 5, Cor. of
Prop. 8) so is Λ(f ) (loc. cit., No. 2, Prop. 6), and



µ• Λ(f ) = µ• (f) = µ• sup g = sup µ• (g) = sup ν(g) = ν*(f ) = ν • (f )
g∈F g∈F g∈F

(the 3rd equality, by §2, No. 2, Prop. 8; the 4th by the next-to-last display).
V.18, `. 15–17.
“Similarly, set

hf (t) = λ∗t (f ) = λ•t (f ) = sup hg (t)

g∈F

(§1, No. 1, Prop. 4).”

In the notations used in earlier Notes (for example, the preceding one),
hg (t)
= gb(t) for g ∈ K+ (X) , t ∈ T , and, for f ∈ I+ (X) , hf (t) =
Λ(f ) (t) = λ•t (f ) , where

λ•t (f ) = λ∗t (f ) = sup λt (g) = sup gb(t) = sup hg (t)

g∈F g∈F g∈F

(the 1st equality, by §1, No. 1, Prop. 4; the 2nd, by Ch. IV, §1, No. 1, Def. 1
and No. 3, Def. 3; the 3rd and 4th repeat definitions of the notations gb
and hg ).
V.18, `. −13.
“Set hf = sup hg ; then hf = hf on S .”
g∈F
§3 integration of positive measures INT V.x42

If t ∈ T --- S then hg (t) = +∞ for all g ∈ F , so

hf (t) = sup hg (t) = +∞ ,

g∈F

whereas if t ∈ S then

hf (t) = sup hg (t) = sup hg (t) = sup λt (g) = λ∗t (f ) = hf (t) ;

g∈F g∈F g∈F

thus hf is given by the same formula used for defining h g :

(
hf (t) for t ∈ S
hf (t) =
+∞ for t ∈ T --- S .

V.18, `. −13 to −11.

“For every g ∈ F , the function hg is lower semi-continuous; hf is
therefore lower semi-continuous . . . ”
Given any k ∈ R , the set

{t ∈ T : hg (t) 6 k} = {t ∈ S : hg (t) 6 k} = {t ∈ S : λt (g) 6 k}

is closed in S (by the vague continuity of t 7→ λ t on S ), hence is closed

in T (because S is closed in T ); thus h g is lower semi-continuous on T
(GT, §6, No. 2, Prop. 1), therefore so is h f (loc. cit., Th. 4).
V.18, `. −10.

“µ*(hf ) = sup µ*(hg ) = sup µ*(hg ) = sup ν(g) = ν*(f ) ”

g∈F g∈F g∈F

1st equality: Ch. IV, §1, No. 1, Th. 1.

2nd equality: Ch. IV, §2, No. 3, Prop. 6.
3rd equality: One has ν(g) = µ• (hg ) by the definition of ν , and
µ• (hg ) = µ• (hg ) = µ∗ (hg ) = µ∗ (hg ) because hg = hg µ-almost everywhere
and hg is lower semi-continuous.
4th equality: Ch. IV, §1, No. 1, Def. 1.
V.19, `. −12, −11.
“ . . . we identify K (X) with a subset of C (X 0 ) .”
The open sets in X0 are the open sets U of X together with the sets
(X --- K) ∪ {ω} , where K is a compact subset of X (GT, I, §9, No. 8);
thus the open neighborhoods of ω are the sets (X --- K) ∪ {ω} . For every
f ∈ K (X) , let f 0 be its extension to X0 by defining f 0 (ω) = 0 . One sees
easily that the functions f 0 are the functions in K (X0 ) that are equal to 0
on a neighborhood of ω .
INT V.x43 integration of measures §3

V.19, `. −7, −6.

“ . . . let ϕn be a function in K+ (X) equal to 1 on Un .”
Ch. III, §1, No. 2, Lemma 1, with n = 1 and V 1 = X .
V.19, `. −6, −5.
“ . . . denote by S the countable set of elements of K (X) of the form
ϕn g ( n ∈ N , g ∈ S0 ).”
The g ∈ S0 belong to C (X0 ) and the ϕn to K (X) , therefore the ϕn g
belong to K (X) ; thus S ⊂ K (X) . {K (X) is an ideal in the ring C (X 0 ).}
V.19, `. −5, −4.
“If f ∈ K (X) , let (gn ) be a sequence of elements of S0 that converges
uniformly to f ”
This approximation argument is the reason for embedding K (X) in
C (X0 ) ; although f can be viewed as an element of C (X 0 ) , it is notassured
that the gn are 0 in a neighborhood of ω , that is, the functions g n X need
not belong to K (X) .
V.19, `. −2, −1.
“The functions fn = ϕk gn belong to S and satisfy the statement, with
ϕ = mϕk . ”
Note that ϕ = (m · 1X0 )ϕk , where m · 1X0 ∈ S0 by assumption, thus
ϕ ∈ S . Since ϕk = 1 on Uk , hence on Supp f , one has ϕk f = f ; thus, for
every n ,
1
|fn − f | = |ϕk gn − f | = |ϕk gn − ϕk f | = ϕk |gn − f | =
ϕ · |gn − f | ,
m
where the vertical bars denote absolute value (not norm). Since kg n −f k → 0,
given any ε > 0 there exists an index n 0 such that |gn − f | 6 mε for
all n > n0 , whence
1

|fn − f | 6 mε · · ϕ = εϕ
m
for all n > n0 .
V.20, `. 7–9.
“ . . . the function t 7→ λt (f ) is then the uniform limit on K of the
continuous functions t 7→ λt (fn ) ; it is therefore continuous on K , and the
proposition is proved.”
Forget X0 : the functions f, fn , ϕ belong to K (X) ; thus we may
write fb = λt (f ) , etc., as in previous notes, and the problem is to show
that kfbn − fk b K → 0 , where

kfbn − fk
b K = sup |λt (fn ) − λt (f )| .
t∈K
§3 integration of positive measures INT V.x44

Given ε > 0 choose n0 so that |fn − f | 6 εϕ for all n > n0 ; then

−εϕ 6 fn − f 6 εϕ for all n > n0 ,

therefore

−ελt (ϕ) 6 λt (fn ) − λt (f ) 6 ελt (ϕ) for all n > n0 and t ∈ T ,

that is, −εϕ b 6 fbn − fb 6 εϕb , whence |fbn − b

f| 6 εϕ
b . We know that the
restriction to K of the mapping t 7→ λt (g) g∈S ∈ RS is continuous; there-
fore, for every ‘coordinate’ g ∈ S of the product space RS , the restriction
to K of the mapping t 7→ λt (g) is continuous, that is, gbK is continuous.
 
In particular, since ϕ and the fn belong to S , ϕ bK and the fbn K are
continuous, hence bounded; and from

kfbn − fk
b K 6 ε kϕk
b K for n > n0
 
we see that fbK is the uniform limit of the continuous functions fbn K ,
hence is continuous.
Since f ∈ K (X) is arbitrary, this shows that the restriction to K of
the mapping Λ : t 7→ λt is vaguely continuous; and since K ∈ K is arbitrary
and K is µ-dense in T , Λ is indeed vaguely µ-measurable (Ch. IV, §5,
No. 10, Prop. 15). A tour de force.
V.20, `. −4 to −2.
“The first of the inequalities (6) then follows from the definition of
R∗
f (x) dν(x) (Ch. IV, §1, No. 3, Def. 3), and the second follows immediately
from it.”
The new actor in Prop. 3 is the function
Z ∗
(∗) t 7→ f (x) dλt (x) = λ∗t (f ) (t ∈ T)

for an arbitrary function f ∈ F+ (X) . In analogy with the notation Λ(f )

for the function
Z •
t 7→ f (x) dλt (x) = λ•t (f ) (t ∈ T)

introduced in the above notes, it is an aid to grasping the formulas (and

saves much scribbling) to write Λ*(f ) for the function defined by (∗):


Λ*(f ) (t) = λ∗t (f ) (t ∈ T) ;
INT V.x45 integration of measures §3

the inequalities (6) may then be written succinctly as

ν*(f ) > µ• Λ*(f ) > µ• Λ(f ) ,

and the second inequality of (6) becomes obvious: λ ∗t (f ) > λ•t (f ) for all

(1)), thus Λ*(f ) > Λ(f ) pointwise on T , whence
t ∈ T (§1,
No. 1, formula
µ• Λ*(f ) > µ• Λ(f ) (loc. cit., Prop. 1). As for the first inequality, since

ν*(f ) = inf ν*(g)

g>f, g∈I+ (X)

(Ch. IV, §1, No. 3, Def. 3), it will suffice to show that


ν*(g) > µ• Λ*(f )

for every lower semi-continuous function g > f ; for such a function g , one
has (§1, No. 1, Prop. 4)

λ•t (g) = λ∗t (g) > λ∗t (f ) for all t ∈ T ,

whence Λ(g) = Λ*(g) > Λ*(f ) pointwise on T , and since Λ is µ-pre-

adequate,


ν*(g) = ν • (g) = µ• Λ(g) > µ• Λ*(f ) .
{Note that if, in addition, f is lower semi-continuous, then



ν • (f ) = ν*(f ) > µ• Λ*(f ) > µ• Λ(f ) = ν • (f )

by the foregoing and the µ-pre-adequacy of Λ ; thus for f not necessarily

lower semi-continuous, the possible

gap between the extreme members of (6)
•
is interpolated by µ Λ*(f ) .}
V.20, `. −2, −1.
“The inequality (7) is proved in an analogous way if Λ is vaguely con-
tinuous, using (5) instead of (4).”
Writing Λ*(f ) for the function t 7→ λ∗t (f ) we are to show that


ν*(f ) > µ* Λ*(f ) ,

and, as in the preceding note,

it suffices to show that if g ∈ I + (X) and
g > f then ν*(g) > µ* Λ(f ) ; indeed, by the monotonicity of outer measure
one has Λ*(g) > Λ*(f ) pointwise on T , and, citing (5) applied to the lower
semi-continuous function g ,



ν*(g) = µ* Λ*(g) > µ* Λ*(f ) .
§3 integration of positive measures INT V.x46

{Recall that if f is lower semi-continuous and if Λ is scalarly essen-

tially µ-integrable and vaguely continuous, the function Λ*(f ) = Λ(f ) is
lower semi-continuous and equality holds in (7) (Prop. 2, a)), whereas if f
is lower semi-continuous and Λ is only assumed to be scalarly essentially
µ-integrable then

ν*(f ) 6 µ• Λ*(f )
by item a) of Exer. 7 (see the note for V.17, `. −6 to −4); it is puzzling
(intriguing) that in two attempts to generalize the equality (4) (namely,
item b) of the present proposition, where f ∈ F + (X) is arbitrary but Λ is
µ-pre-adequate and vaguely continuous; and Exer. 7, where f is lower semi-
continuous but Λ is only assumed to be scalarly essentially µ-integrable),
the inequality in (7) “goes the other way” from the one in Exer. 7. That
equality holds when both sets of hypotheses are assumed teaches us nothing,
since (4) is then immediate from the definition of µ-pre-adequacy.}
V.21, `. 1.
“The mapping t 7→ λ∗t (1) is measurable ”
For, Λ is µ-pre-adequate and the constant function 1 is lower semi-
continuous, therefore the function t 7→ λ •t (1) is µ-measurable and λ•t (1) =
λ∗t (1) ; with notations as in
the preceding note, Λ(1)
= Λ*(1) pointwise
on T and ν • (1) = µ• Λ(1) , that is, ν*(1) = µ• Λ*(1) .
V.21, `. 2–4.
“The set K of compact subsets of T such the restriction of t 7→ λ ∗t (1)
to K is finite and continuous is therefore µ-dense ”
Write h(t) = λ∗t (1) for t ∈ T and let

A = {t ∈ T : h(t) < +∞ } .

By assumption, { A is locally µ-negligible; in particular, { A is µ-measurable

(p. IV.61, `. −4), therefore so is A . Since h is µ-measurable, it follows that
the (finite-valued) function hA is µ-measurable in the sense of Ch. IV, §5,
No. 10, Def. 8 (see c 00 ) in the Note for IV.79, `. 3,4).
 To say that
 a compact
 
set K belongs to K signifies that K ⊂ A and h K = (h A)K is contin-
uous; since hA is µ-measurable, it follows that K is µ-dense in A (loc.
cit., Prop. 15, criterion a)), and since T --- A is locally µ-negligible, K is
µ-dense in T (p. IV.77 `. 3, 4).
V.21, `. 6.
“The mapping Λ is µα -adequate for every α ∈ A ”
Recall the remark following No. 1, Def. 1, to the effect that in Nos. 2
and 3, in “a,b,c propositions”, the parts a) and b) will be valid for Λ µ-pre-
adequate, whereas part c) will require that Λ be µ-adequate. The present
INT V.x47 integration of measures §3

proposition illustrates the principle and its logic: in part c) arguments, it

is necessary to express µ as a sum of ‘better-behaved’ measures µ 0 6 µ ,
whence the need for µ-adequacy.
V.21, `. 8, 9.
“ να is a bounded measure (because λ∗t (1) is bounded on Supp(µα ) ).”
Since the constant function 1 is lower semi-continuous and Λ is
µα -pre-adequate, writing h(t) = λ•t (1) = λ∗t (1) for t ∈ T one has

να∗ (1) = µ•α (h)

by (4) of No. 1. Write Sα = Supp(µα ) , so that Sα ∈ K . By the defini-

tion of K , we know that the restriction of h to the compact space K α is
finite-valued and continuous, hence bounded, say h Sα 6 M < +∞ . Then
0 6 hϕSα 6 MϕSα , whence

µ•α (hϕSα ) 6 M µ•α (ϕSα ) 6 M µ∗α (ϕSα ) = M µα (Sα ) < +∞

(actually µ• (ϕSα ) = µ*(ϕSα ) by §1, No. 2, Prop. 7); but T --- Sα is

µα -negligible, therefore h = hϕSα µα -almost everywhere and so

να∗ (1) = µ•α (h) = µ•α (hϕSα ) < +∞ ,

thus να is indeed bounded (Ch. IV, §4, No. 7, Prop. 12).

V.21, `. 11–14.
“ . . . then
Z • Z • Z ∗ Z • Z •
f (x) dνα (x) > dµα (t) f (x) dλt (x) = dµα (t) f (x) dλt (x)

(the last equality due to the fact that λ t is bounded locally almost every-
where, and Prop. 7 of § 1).”
As in earlier notes, let us write Λ(f ) for the function t 7→ λ •t (f )
and Λ*(f ) for the function t 7→ λ∗t (f ) , so that Λ(f ) 6 Λ*(f ) pointwise
on T by (1) of §1, No. 1.
For locally µ-almost every t ∈ T one has λ ∗t (1) = λ•t (1) < +∞ and
so λt is bounded (Ch. IV, §4, No. 7, Prop. 12); but if λ t is bounded then
it is moderated (§1, No. 2, Remark 2), therefore f is λ t -moderated (loc.
cit., comment following Def. 2) and so λ ∗t (f ) = λ•t (f ) (loc. cit., Prop. 7).
It follows that Λ*(f ) = Λ(f ) locally µ-almost

everywhere,

therefore locally
µα -almost everywhere, and so µ•α Λ*(f ) = µ•α Λ(f ) .
§3 integration of positive measures INT V.x48

To summarize: Assuming λ•t (1) < +∞ locally µ-almost everywhere,

one has (by §1, No. 2, Prop. 7, and (6))



να• (f ) = να∗ (f ) > µ•α Λ*(f ) = µ•α Λ(f )

for every α and for every function f > 0 on X , and summing over α
yields (8):


ν • (f ) > µ• Λ*(f ) = µ• Λ(f )
(§2, No. 2, Prop. 1). Note that the first equality in (6) is sharpened in (8)
R∗ R•
(because > ), and the second inequality in (6) becomes an equality
in (8).
V.21, `. 18–24.
“Corollary 1”
{Cor. 1, Prop. 4 and Prop. 5 are variations on the same general pattern;
as the variations are numerous and subtle, it is helpful to have the pattern
in mind as one thrashes through the details.
One is given a mapping Λ : t 7→ R λt that is scalarly essentially
µ-integrable, so that the measure ν = λt dµ(t) is defined; certain con-
ditions are imposed on f and ν , and one shows that the same conditions
are satisfied for “most” of the λt . The progression from a) to c) tends to
be from general to special:
(i) In assertion a), Λ is assumed to be µ-pre-adequate, and “most”
means “for locally µ-almost every t ”.
(ii) To the hypotheses of a), assertion b) adds the assumption that Λ
is vaguely continuous, and “most” is sharpened to “for µ-almost every t ”.
(iii) In assertion c), Λ is assumed to be µ-adequate, and the function
t 7→ λ•t (1) = λ∗t (1) is assumed to be finite-valued locally µ-almost every-
where, in other words, the measure λ t is assumed to be bounded for locally
µ-almost every t ∈ T ; “most” means “for locally µ-almost every t ”.}
R
Let Λ : t 7→ λt be scalarly essentially µ-integrable and let ν = λt dµ(t) .
Assume in a) and b) that Λ is µ-pre-adequate, and in c) that Λ is
µ-adequate. Let f ∈ F+ (X) and set


H = {t ∈ T : λ∗t (f ) > 0 } = {t ∈ T : Λ*(f ) (t) > 0 } .

a) Assuming ν*(f ) = 0 , we are to show that µ • (H) = 0 , that is, f is

λt -negligible for locally µ-almost every t ∈ T . Citing a) of Prop. 3, we have


0 = ν*(f ) > µ• Λ*(f ) > 0 ,


therefore µ• Λ*(f ) = 0 , that is, Λ*(f ) = 0 locally µ-almost everywhere
in T ; in other words, µ• (H) = 0 .
INT V.x49 integration of measures §3

b) Assuming ν*(f ) = 0 and Λ vaguely continuous, we are to show that

µ*(H) = 0 , that is, f is λt -negligible for µ-almost every t ∈ T . Citing b)
of Prop. 3, we have


0 = ν*(f ) > µ* Λ*(f ) > 0 ,


therefore µ* Λ*(f ) = 0 , that is, Λ*(f ) = 0 µ-almost everywhere in T ; in
other words, µ*(H) = 0 .
c) Assuming ν • (f ) = 0 and λ•t (1) < +∞ for locally µ-almost every
t ∈ T , we are to show that µ• (H) = 0 , that is, f is λt -negligible for locally
µ-almost every t ∈ T . Citing c) of Prop. 3, we have


0 = ν • (f ) > µ• Λ*(f ) > 0 ,


therefore µ• Λ*(f ) = 0 , that is, Λ*(f ) = 0 locally µ-almost everywhere
in T ; in other words, µ• (H) = 0 .
V.21, `. −8 to −6.
“ f0 is then λt -negligible except for t forming a set that is locally
µ-negligible (and even µ-negligible, if Λ is vaguely continuous) by Cor. 1,
and the statement then follows at once.”
Write N = {t ∈ T : λ∗t (f0 ) > 0 } for the set of all t ∈ T such that f 0
is not λt -negligible. Since f0 is ν-negligible, it follows from a) of Cor. 1 that
µ• (N) = 0 ; and when Λ is vaguely continuous, µ*(N) = 0 by b) of Cor. 1.
Let M = {t ∈ T : f is not λt -moderated }; we are to show that
µ• (M) = 0 , and that µ*(M) = 0 when Λ is vaguely continuous; it will
suffice to show that M ⊂ N .
Assume t ∈ / N and let us show that t ∈ / M , i.e., that f is λ t -mode-
rated; by criterion c) of §1, No. 2, Prop. 5, it suffices to show that the set
{x ∈ X : f (x) > 0 } is contained in the union of a λ t -negligible set and a
sequence of compact sets. Indeed, f (x) > 0 if and only if f n (x) > 0 for
some index n > 0 , thus {x : f (x) > 0 } is equal to the set
[ [
{x : f0 (x) > 0} ∪ {x : fn (x) > 0} ⊂ {x : f0 (x) > 0} ∪ Kn ,
n>1 n>1

where {x : f0 (x) > 0 } is λt -negligible (because t ∈

/ N ).
V.22, `. 6, 7.
“ . . . the function f is constant on the complement B of a countable
union of ν-integrable open sets.”
Let A be a ν-moderated set such that f is constant S on X --- A , and
let (Ur ) be a sequence of ν-integrable sets such that A ⊂ Ur (§1, No. 2,
r
§3 integration of positive measures INT V.x50
S S
Prop. 5, a)). Then X --- Ur ⊂ X --- A and the set B = X --- Ur meets
r r
the requirements. The reason for choosing criterion a) of Prop. 5 is that the
set B is then closed, hence is a Borel set, hence is measurable with respect
to any measure.
V.22, `. 7–9.
“There exists a partition of X --- B formed by a ν-negligible set N and
a sequence (Kn ) of compact sets such that the restriction of f to each K n
is continuous.”
S
With X --- B = Ur as in the preceding note, let (Ar ) be a ‘disjoin-
r S
tification’ of the Ur , for example A1 = U1 , Ar+1 = Ur+1 --- Us , so
s6r
that X --- B is partitioned by the sequence (A r ) of ν-integrable sets. The
sets Ar need no longer be open, but their union is, and B remains a Borel
set.
Consider first the case of a single ν-integrable set A . By Ch. IV, §4,
No. 6, Cor. 2 of Th. 4, there exists a countable partition of A ,
[
A = N 0 ∪ Ci ,
i

with N0 ν-negligible and the Ci compact. Then, since f is ν-measurable,

for each i there exists a countable partition
[
Ci = Ni ∪ Kij
j

with Ni ν-negligible, the Kij compact and f Kij continuous (Ch. IV, §5,
S
No. 1, Def. 1); setting N = N0 ∪ Ni and enumerating the Kij into a
i
sequence (Kn ) , we have a partition
[
A=N∪ Kn
n

of the desired sort. S S

With X --- B = Ar as above, for each r let Ar = Nr ∪ Krj be
r j
S
such a partition of Ar . The negligible set N = Nr and an enumeration
r
(Kn ) of the compact sets Krj then yield a partition
[
X --- B = N ∪ Kn
n

that meets the requirements of the assertion.

INT V.x51 integration of measures §3

V.22, `. 9–11.
“Let S be the set of t ∈ T such that N is not λ t -negligible: S is
locally µ-negligible (resp. µ-negligible) by Cor. 1 of Prop. 3.”
Part a) (resp. Part b)) of Cor. 1 is being applied to the ν-negligible
function ϕN , the set S being denoted H there.
V.22, `. 11–13.
“The sets Kn , B, N are measurable for every measure on X , and the
restriction of f to each of them is λt -measurable for every t ∈
/ S .”
Since B and the Kn are closed sets, all sets in the formula
[
N = (X --- B) --- Kn
n

are Borel sets, hence are ‘universally measurable’ (Ch. IV, §5, No. 4, Cor. 3
of Th. 2). 
By construction, f B is a constant function, say equal to y 0 ∈ G at
every point of B ; its extension to X by y 0 is constant on X , hence contin-
uous, hence ‘universally measurable’, in particular measurable for every λ t
(t ∈/ S) . 
The function f Kn is continuous. Let K be the set of all compact
subsets of Kn . Obviously K is closed under finite unions, thus K satisfies
the conditions (PLI ) and (PLII ) of Ch. IV, §5, No. 8, Prop. 12; moreover,
condition b) of the cited Prop. 12 is satisfied with A = K n , K = K0 , and
any measure on X , therefore K is dense in K n for every measure on X
(loc. cit., Def. 6). Thus K and f Kn satisfy condition a) of Ch. IV,  §5,
No. 10, Prop. 15 with A = Kn , for every measure on X , therefore f Kn is
measurable for every measure on X (loc. cit., Def. 8) and in particular for
every λt (t ∈ / S) .
Let t ∈/ S . Then N is λt -negligible (hence also locally λt -negligible),
therefore f N is λt -measurable (see the Note for IV.79, `. −17, −16).
Thus, for every t ∈/ S , the restriction of f to B , N and every K n is
λt -measurable.
V.22, `. 13–14.
“The function f is therefore λt -measurable for every t ∈
/ S (Ch. IV,
§5, No. 10, Prop. 16).”
Continuing the preceding note: since the set consisting of B , N and
the Kn is countable, a fortiori locally countable, and since
[
X = B∪N∪ Kn ,
n
§3 integration of positive measures INT V.x52

it follows that f = f X is λt -measurable for every t ∈
/ S (Ch. IV, §5, No. 10,
Prop. 16).
V.22, `. −4 to −2.
“The assertions concerning the set N have already been established
(Prop. 4, and Cor. 2 of Prop. 3).”
Let
L = {t ∈ T : f is not λt -moderated }
M = {t ∈ T : f is not λt -measurable } ;
/ { L ∩ { M , thus
then t ∈ N means that t ∈

N = L ∪ M.

Since f is ν-measurable and ν-moderated, µ • (L) = 0 by the cited Cor. 2,

and µ• (M) = 0 by a) of Prop. 4, therefore µ• (N) = 0 ; if, moreover, Λ is
vaguely continuous, then µ*(L) = 0 by Cor. 2, and µ*(M) = 0 by b) of
Prop. 4, therefore µ*(N) = 0 .
V.22, `. −2 to V.23, `. 3.
“By Prop. 6 of §1, No. 2, we may limit ourselves to proving a) (resp. b))
in each of the following special cases:
1) The function f is ν-negligible.
2) There exists a compact set K such that f is zero outside K and
the restriction of f to K is continuous.”
*The argument is intricate and repetitious. To lighten the burden of
referencing, some basic results are bundled together here; any of the items
in (α) will be referenced simply by “(α)”, and similarly for (β).
Let ρ be a positive measure on a locally compact space, and let h and
hn (n ∈ N) be numerical functions > 0 on that space.
(α) If hn ↑ h pointwise then

ρ*(hn ) ↑ ρ*(h) and ρ• (hn ) ↑ ρ• (h)

(Ch. IV, §1, No. 3, Th. 3 and Ch. V, §1, No. 1, Prop. 1). If, moreover,
the hn are ρ-measurable, then so is h (Ch. IV, §5, No. 4, Cor. 1 of Th. 2).
If the hn are ρ-moderated, then so is h (Ch. V, §1, No. 2, Remark 3).
P
(β) If the hn are ρ-measurable and h = hn , then h is ρ-measurable:
n∈N
the measurability of h0 + h1 follows from
Ch. IV, §5, No. 3, Th. 1, ap-
plied to the mapping x 7→ h0 (x), h1 (x) ( x in the given space), with
u(a, b) = a + b for (a, b) ∈ A0 × A0 , where A0 = R ∪ {+∞} (GT, IV,
§4, No. 3, Prop. 7); the measurability of the finite sums h 0 + · · · + hn follows
INT V.x53 integration of measures §3

at once, and h = sup (h0 + · · · + hn ) is then measurable by (α). Moreover

n∈N
(the hn still measurable)
X X
ρ*(h) = ρ*(hn ) and ρ• (h) = ρ• (hn )
n∈N n∈N

by Ch. IV, §5, No. 6, Cor. 4 of Th. 5 and Ch. V, §1, No. 1, Cor. of Prop. 2).
If the hn are ρ-moderated, then so is h (Ch. V, §1, No. 3, Remark 3). ∗
By the cited Prop. 6, there exists a sequence f n (n = 0, 1, 2, . . .) of
numerical functions > 0 on X such that f 0 satisfies 1), and the fn for
n > 1 satisfy 2) and are finite-valued. Moreover, the f n for n > 1 are
ν-measurable and ν-moderated: for f 0 , this is obvious, and it is obvious
the fn for n > 1 are moderated (for any measure on X ). It remains to
check that if n > 1 then fn is measurable. Let Kn be a compact set in X
such that fn Kn is continuous and fn = 0 outside Kn , let a ∈ R and let

A = {x ∈ X : fn (x) > a } ;

if a 6 0 then A = X , while if a > 0 then A = {x ∈ K n : fn (x) > a } is a

closed set in Kn by the continuity of fn Kn , hence is closed in X . Thus fn
is upper semi-continuous (GT, IV, §6, No. 2, Prop. 1 applied to −f n ), hence
is measurable for every measure on X (Ch. IV, §5, No. 5, Cor. of Prop. 8).
Assuming it has been shown that for every n , the function
Z •
t 7→ fn (x) dλt (x)

(denoted Λ(fn ) in earlier notes) is µ-measurable, and that

(i) ν • (fn ) = µ• Λ(fn ) ,

and, when Λ is vaguely continuous, that the function Λ*(f n ) : t 7→ λ∗t (fn )
is µ-measurable and µ-moderated, and satisfies


(ii) ν*(fn ) = µ* Λ*(fn ) ,

we must verify that the same is true of f .

Verification of a) for f assuming it has been verified for the f n : For
every n ∈ N , one shows as in the preceding Note that f n is µ-measurable
and µ-moderated for locally µ-almost every t ; it follows that, for locally
§3 integration of positive measures INT V.x54

µ-almost every t , fn is µ-measurable and µ-moderated for every n . For

every n , define
Xn
gn = fk ;
k=0

since f0 , . . . , fn are λt -measurable and λt -moderated for locally µ-almost

every t , the same is true of gn and
n
X
λ•t (gn ) = λ•t (fk ) for locally µ-almost every t ∈ T
k=0

P
n
by (β). Thus Λ(gn ) = Λ(fk ) locally µ-almost everywhere on T , conse-
k=0
quently
n
X n
X
•

•


(iii) µ Λ(gn ) = µ Λ(fk ) = ν • (fk ) = ν • (gn ) for all n
k=0 k=0

by (β) and (i). Since gn ↑ f pointwise on T , for every t ∈ T one has

λ•t (gn ) ↑ λ•t (f ) by (α) ; thus Λ(gn ) ↑ Λ(f ) pointwise on T , therefore



(iv) µ• Λ(f ) = sup µ• Λ(gn )
n

by (α), and similarly ν • (f ) = sup ν • (gn ) . From (iii) and (iv) we see that
n



ν • (f ) = sup ν • (gn ) = sup µ• Λ(gn ) = µ• Λ(f ) .
n n

P
n
Since, by assumption, the Λ(fn ) are µ-measurable, so are the Λ(fk )
k=0
by (β), therefore so are the Λ(gn ) (Ch. IV, §5, No. 2, Prop. 6), therefore so
is Λ(f ) by (α). And since every gn is µ-measurable and µ-moderated, so
is f = sup gn by (α).
n

Verification of b) for f assuming it has been verified for the f n : As-

sume in addition that Λ is vaguely continuous. For every n , one shows as
in the preceding Note that for µ-almost every t , f n is λt -measurable and
λt -moderated; it follows that there exists a µ-negligible subset P of T such
that fn is λt -measurable and λt -moderated for all n and for all t ∈ T --- P .
In particular,
n
X
λ∗t (gn ) = λ∗t (fk ) for all n and for all t ∈ T --- P
k=0
INT V.x55 integration of measures §3

P
n
by (β), thus Λ*(gn ) = Λ*(fk ) µ-almost everywhere for all n ; since the
k=0
Λ*(fn ) are by assumption µ-measurable and µ-moderated, it follows that
Pn
Λ*(fk ) is µ-measurable and µ-moderated by (β), therefore Λ*(g n ) is
k=0
µ-measurable by Ch. IV, §5, No. 2, Prop. 6, and is obviously µ-moderated.
Moreover,
n

X

µ* Λ*(gn ) = µ* Λ*(fk ) for all n
k=0

by (β); but
n
X n
X


µ* Λ*(fk ) = ν*(fk ) = ν*(gn )
k=0 k=0

by (ii) and (β), therefore

(v) µ* Λ*(gn ) = ν*(gn ) for all n .

Since gn ↑ f pointwise on T , it follows from (α) that

λ∗t (gn ) ↑ λ∗t (f ) for every t ∈ T ,

thus Λ*(gn ) ↑ Λ*(f ) pointwise on T ; since the Λ*(g n ) are µ-measurable

and µ-moderated, by (α) so is Λ*(f ) , and



(vi) µ* Λ*(f ) = sup µ* Λ*(gn ) .
n

Finally,


ν*(f ) = sup ν*(gn ) = sup µ* Λ*(gn ) = µ* Λ*(f )
n n

by (α), (v) and (vi).

V.23, `. 4.
“The special case 1) has already been treated (Cor. 1 of Prop. 3).”
Assuming f is ν-negligible (hence ν-measurable and ν-moderated), we
are to show that a) and b) are satisfied. Let

N = {t ∈ T : f is not both λt -measurable and λt -moderated }.

a) The assumption that f is ν-moderated is redundant. Let

H = {t ∈ T : f is not λt -negligible };
§3 integration of positive measures INT V.x56

by part a) of Cor. 1 of Prop. 3, H is locally µ-negligible; since a negligible

function is both measurable and moderated, one has N ⊂ H , therefore N
is also locally µ-negligible. {The information about N is redundant here,
but N was not yet in the picture in the cited Cor. 1.}
If t ∈
/ H then f is λt -negligible and, a fortiori, locally λ t -negligible,
∗
thus λt (f ) and λ•t (f ) are both equal to 0 ; the functions

Λ(f ) : t 7→ λ•t (f ) and Λ*(f ) : t 7→ λ∗t (f )

are therefore both locally µ-negligible, hence are µ-measurable, and

µ• Λ(f ) = 0 = µ• Λ*(f ) .

But f is ν-negligible, therefore ν • (f ) = ν*(f ) = 0 . In particular,

ν • (f ) = 0 = µ• Λ(f ) ,

thus f satisfies (9).

b) If, moreover, Λ is vaguely continuous, then the set H is µ-negligible
by part b) of Cor. 1 of Prop. 3, hence so is N . Since f is λ t -negligible for
t∈/ H , Λ*(f ) = 0 µ-almost everywhere; thus Λ*(f ) is µ-negligible, hence
µ-moderated, and

ν*(f ) = 0 = µ* Λ*(f ) ,
thus f satisfies (10).
V.23, `. 8, 9.
“Moreover, f, g, h are ν-integrable.”
Recall that f can be assumed to be finite-valued (§1, No. 2, Prop. 6).
The functions f, g, h are all positive; being semi-continuous, they are (uni-
versally) measurable, and since 0 6 f 6 h = Mϕ G ,

ν*(f ) 6 M ν*(G) < +∞

shows that h and f are ν-integrable, therefore so is g = h − f .

V.23, `. 11–13.
“By subtraction, we see that the function
Z • Z ∗


t 7→ f (x) dλt (x) resp. f (x) dλt (x)

is µ-measurable and that the formula (9) (resp. (10)) holds.”

INT V.x57 integration of measures §3

Verification of a). Since g and h are lower semi-continuous func-

tions > 0 , and Λ is µ-pre-adequate, it follows from Def. 1 that the functions

Λ(g) : t 7→ λ•t (g) and Λ(h) : t 7→ λ•t (h)

are µ-measurable and satisfy (4):

ν • (g) = µ• Λ(g) and ν • (h) = µ• Λ(h) .


Since g is ν-integrable, µ• Λ(g) = ν*(g) < +∞ , thus the µ-measurable
function Λ(g) is essentially µ-integrable, and similarly for Λ(h) . Write

Ng = {t ∈ T : g is not both λt -measurable and µ-moderated } .

As shown at the outset, µ• (Ng ) = 0 , and similarly µ• (Nh ) = 0 . For

t ∈ { (Ng ∪ Nh ) , g and h are λt -measurable, hence so is f = h − g ,
therefore (§1, No. 1, Prop. 2)

λ•t (h) = λ•t (f + g) = λ•t (f ) + λ•t (g) ,

thus Λ(h) = Λ(f ) + Λ(g) locally µ-almost everywhere. Since Λ(g) and
Λ(h) are essentially µ-integrable, so is Λ(f ) ; for, if u, v ∈ L 1 (µ) are equal,
locally µ-almost everywhere, to Λ(g) and Λ(h) , respectively (§1, No. 3,
Def. 3), then, for locally µ-almost every t ∈ T ,

v(t) = λ•t (f ) + u(t) ,

therefore Λ(f ) = v − u locally µ-almost everywhere, where v − u ∈ L 1 (µ) ,

and so

1
Λ(f ) = (v − u) + Λ(f ) − (v − u) ∈ L 1 (µ) + N ∞ = L (µ) .

In particular Λ(f ), Λ(g), Λ(h) are µ-measurable functions > 0 , therefore

µ• Λ(h) = µ• Λ(f ) + µ• Λ(g)

(§1, No. 1, Prop. 2), whence

µ• Λ(f ) = µ• Λ(h) − µ• Λ(g) = ν • (h) − ν • (g) = ν(h) − ν(g) = ν(f ) ,

thus f satisfies (9).

Verification of b). If, moreover, Λ is vaguely continuous, it follows
from a) of No. 1, Prop. 2 that the functions

Λ*(g) : t 7→ λ∗t (g) and Λ*(h) : t 7→ λ∗t (h)

§3 integration of positive measures INT V.x58

are µ-measurable and satisfy (5):

ν*(g) = µ* Λ*(g) and ν*(h) = µ* Λ*(h) ,

and the ν-integrability of g and h implies that that Λ*(g) and Λ*(h) are
µ-integrable (Ch. IV, §5, No. 6, Th. 5).
As shown at the outset, g is λt -measurable and λt -moderated for
µ-almost every t , and similarly for h , so the same is true of f = h − g , and

λ∗t (h) = λ∗t (f ) + λ∗t (g) ;

thus Λ*(h) = Λ*(f ) + Λ*(g) µ-almost everywhere. Since Λ*(g) and Λ*(h)
are µ-integrable, they are finite µ-almost everywhere (Ch. IV, §2, No. 3,
Prop. 7), hence there exist functions u, v ∈ L 1 (µ) such that Λ(g) and Λ(h)
are equal, µ-almost everywhere, to u and v , respectively. Then, for
µ-almost every t ∈ T ,
v(t) = λ∗t (f ) + u(t) ,
therefore Λ*(f ) is equal µ-almost everywhere to the µ-measurable function
v − u , hence is µ-measurable (Ch. IV, §5, No. 2, Prop. 6). It follows that




µ* Λ*(h) = µ* Λ*(f ) + µ* Λ*(g)

(Ch. IV, §5, No. 6, Cor. 4 of Th. 5), therefore

µ* Λ*(f ) = µ* Λ*(h) − µ* Λ*(g) = ν*(h) − ν*(g) = ν(f ) < +∞ ,

thus Λ*(f ) is µ-integrable (Ch. IV, §5, No. 6, Th. 5), hence is µ-moderated
(§1, No. 2, Prop. 7), and satisfies (10).
Thus parts a) and b) of Prop. 5 are proved in full generality.
V.23, `. −12.
“ f is να -measurable and να -moderated ”
Because f is ν-measurable and 0 6 να 6 ν ; and because να is bounded
(§1, No. 2, Remark 2).
V.23, `. −9.
“It remains only to sum on α , applying Props. 1 and 2 of §2, No. 2.”
By the cited Prop. 1, f satisfies (9).
By the cited Prop. 2, the function Λ(f ) : t 7→ λ •t (f ) is µ-measurable.
By assertion a), N is locally µα -negligible for all α , therefore N is
locally µ-negligible by Cor. 2 of the cited Prop. 1.
INT V.x59 integration of measures §3

V.23, `. −2, −1.

“This follows at once from Prop. 5 and the criterion for integrability
(Ch. IV, §5, No. 6, Th. 5).”
Since f is ν-measurable and ν-moderated, so is |f | , and ν
• (|f |) =
ν*(|f |) (§1, No. 2, Prop. 7). By Prop. 5, a), ν • (|f |) = µ• Λ(|f |) , where
Λ(|f |) denotes the function t 7→ λ•t (|f |) on T . By the cited Th. 5,

f is ν-integrable ⇔ ν*(|f |) < +∞

⇔ ν • (|f |) < +∞


⇔ µ• Λ(|f |) < +∞ ,

whence the Corollary.

Another way of packaging the Corollary:
Corollary 0 . — Let f be a function defined on X , with values in a
Banach space F or in R , Rsuppose Λ : t 7→ λ t is a µ-pre-adequate mapping
T 7→ M+ (X) , and let ν = λt dµ(t) .
Then f is ν-integrable

if and only if it is ν-measurable, ν-moderated and
•
satisfies µ Λ(f ) < +∞ .
V.24, `. 13, 14.
“This statement is true when f is a positive numerical function (Prop. 5) ”
a) Assuming f > 0 and ν-integrable, let

N = {t ∈ T : f is not both λt -measurable and λt -moderated }.

Since f is both ν-measurable and ν-moderated (§1, No. 2, Prop. 7, 1)), by

part a) of Prop. 5, N is locally µ-negligible, the function Λ(f ) : t 7→ λ •t (f )
is µ-measurable, and


(∗) µ• Λ(f ) = ν • (f ) = ν*(f ) < +∞ ,

therefore Λ(f ) is essentially µ-integrable (§1, No. 3, Prop. 9). It follows that
Λ(f ) is finite locally µ-almost everywhere, that is, the set

P = {t ∈ T : λ•t (f ) = +∞ }

is locally µ-negligible. If t ∈/ P ∪ N then f is λ t -measurable, λt -moderated,

and λ∗t (f ) = λ•t (f ) < +∞ , therefore f is λt -integrable, that is, t ∈
/ H . Thus
H ⊂ P ∪ N , hence µ• (H) = 0 .
§3 integration of positive measures INT V.x60

Since Λ(f ) is essentially integrable, notational conventions permit writ-

ing (∗) as
Z Z Z Z
f (x) dν(x) = Λ(f ) dµ = dµ(t) f (x) dλt (x)

(§1, No. 3, Def. 3), that is, (11) holds.

Writing Λ*(f ) for the function t 7→ λ∗t (f ) , since f is λt -moderated for
locally µ-almost every t , one has Λ*(f ) = Λ(f ) locally µ-almost everywhere
(§1, No. 2, Prop. 7), therefore Λ*(f ) is also essentially µ-integrable, and



µ• Λ*(f ) = µ• Λ(f ) = ν • (f ) = ν*(f ) .

b) Suppose, in addition, that Λ is vaguely continuous. Then by b)

of Prop. 5, N is µ-negligible and Λ*(f ) is µ-moderated; since Λ*(f ) is
essentially µ-integrable and µ-moderated, it is µ-integrable (§1, No. 3, Cor.
of Prop. 9), hence is finite µ-almost everywhere (Ch. IV, §2, No. 3, Prop. 7).
Let
P* = {t ∈ T : λ∗t (f ) = +∞ } ,
which is a µ-negligible set (incidentally, the inclusion P ⊂ P* is useless
here). If t ∈/ P* ∪ N then f is λt -measurable and λ∗t (f ) < +∞ , hence f is
λt -integrable; therefore
R H ⊂ P* ∪ N , whence H is µ-negligible.
Finally, the signs in (11) may be interpreted
R as ordinary integrals
(rather than essential integrals)—in the case of f (x)dλt (x) , for µ-almost
every t (namely, for t ∈ / H ).
V.24, `. 16.
“ . . . extends at once to f by subtraction.”
At first glance ‘obvious’, it heralds a new phenomenon: heretofore, sym-
bols such as

Λ(f ) : t 7→ λ•t (f ) , Λ*(f ) : t 7→ λ∗t (f ) , Λ(f ) : t 7→ λt (f )

have (in these notes) denoted numerical functions that are defined every-
where on T , either because f > 0 or because f ∈ K (X) (No. 1); even in
the Cor. of Prop. 5 of No. 2, it is Λ(|f |) that figuresRin the proof, not Λ(f ) .
From now on, we have to deal with functions t 7→ f dλt that are defined
1
only on the set of t ∈ T such that f ∈ L (λt ) (or L 1 (λt ) ), and which
may have values in a Banach space.
A review of the core results on numerical functions that are involved:
(i) f is measurable ⇔ f + and f − are measurable, in which case
|f | = f + + f − is measurable (Ch. IV, §5, No. 3, Cors. 2 and 3 of Th. 1).
INT V.x61 integration of measures §3

(ii) f is integrable ⇔ f is measurable and |f | is integrable (by (i) and

loc. cit., No. 6, Th. 5).
R (iii)
R +f isR integrable ⇔ f + and f − are integrable, in which case
−
f= f − f .
The equivalence is clear from (i), (ii) and the fact that when f is mea-
R∗ R∗ + R∗ −
surable, |f | = f + f (loc. cit., No. 6, Cor. 4 of Th. 5). If f is
integrable, by definition there exists a (finite-valued) function g ∈ L 1 such
+ + − −
that f = g almost Reverywhere, R R whence R f− =R g + and
R −f = g almost
+
everywhere, and so f = g = g − g = f − f by the linearity
of integration on L 1 .
(iv) f is essentially integrable ⇔ f is measurable and |f | is essentially
integrable (§1, No. 3, Prop. 9).
+ −
(v) f is essentially
R R +integrable
R − ⇔ f and f are essentially integrable,
in which case f = f − f .
The equivalence
R• isR clear from
R • −(i), (iv) and the fact that when f is
• +
measurable,
R |f | = f + f (§1, No. 1, Prop. 2). The formula
for f is proved by the argument of (iii) with “almost everywhere” replaced
by “locally almost everywhere”.
Proof of a) for a numerical function f . Assuming f ν-integrable, let

H+ = {t ∈ T : f + is not λt -integrable },
H− = {t ∈ T : f − is not λt -integrable };

since f is λt -integrable if and only if f + and f − are λt -integrable, one has

{ H+ ∩ { H− = {t ∈ T : f + and f − are λt -integrable }

= {t ∈ T : f is λt -integrable } = { H ,

whence H = H+ ∪ H− . Since f + and f − are ν-integrable, we know from

the case of functions > 0 that H+ and H− are locally µ-negligible, hence
so is H ; that the functions

Λ(f + ) : t 7→ λ•t (f + ) , Λ(f − ) : t 7→ λ•t (f − ) ,

defined everywhere on T , are essentially µ-integrable; and that

ν(f + ) = µ• Λ(f + ) , ν(f − ) = µ• Λ(f − ) .
R
With denoting essential integral, define Λ(f ) : { H → R by
Z Z Z


Λ(f ) (t) = f dλt = f dλt − f − dλt = λ•t (f + ) − λ•t (f − )
+
(t ∈ { H) ,
§3 integration of positive measures INT V.x62
 
that is, Λ(f ) = Λ(f + ) { H − Λ(f − ) { H . We are to show that the function
Λ(f ) , defined locally µ-almost everywhere
R in T , is essentially µ-integrable
in the sense of §1, No. 3, and that Λ(f ) dµ = ν(f ) .
Since Λ(f + ) and Λ(f − ) are essentially µ-integrable, there exist func-
tions h, k ∈ L 1 (µ) such that Λ(f + ) = h and Λ(f − ) = k locally µ-almost
everywhere (§1, No. 3, Def. 3); since H is locally µ-negligible, it follows
that Λ(f ) = h − k locally µ-almost everywhere, therefore Λ(f ) is essentially
µ-integrable in the sense of §1, No. 3 (p. V.9, `. 1–5) and
Z Z Z Z
Λ(f ) dµ = (h − k) dµ = h dµ − k dµ
Z Z
= Λ(f ) dµ − Λ(f − ) dµ
+

= ν(f + ) − ν(f − ) = ν(f ) ,

thus (11) is satisfied.

{Incidentally, the functions
 u = ϕ { H Λ(f + ) and v = ϕ { H Λ(f − ) are
the extensions by 0 of Λ(f + ) { H and Λ(f − ) { H to T ; they are finite-
valued, µ-measurable, and, locally µ-almost everywhere, Λ(f + ) = u and
1
Λ(f − ) = v , thus u, v ∈ L (µ) (§1, No. 3, Prop. 9). Moreover, the extension
by 0 of Λ(f ) to T is equal pointwise to u − v . Thus, the extension by 0
1
of Λ(f ) to T belongs to L (µ) and may be substituted for Λ(f ) in (11).}
Proof of b) for a numerical function f . Assuming in addition that Λ
is vaguely continuous, the argument proceeds as in a), with “locally negli-
gible” replaced by “negligible”; “locally almost everywhere” by “almost ev-
1
erywhere”; “essentially integrable” by “integrable”; and L (µ) by L 1 (µ) .
V.24, `. 19, 20.
“ . . . the result pertaining to real functions implies at once the validity
of the statement for the elements of H . ”
By linearity it suffices to consider f ∈ H of the form f = af (the
function x 7→ f (x) · a ), where f ∈ K (X) and a ∈ F . One knows that f is
integrable for every measure on X (Ch. IV, §3, No. 5, Cor. 2 of Th. 4); in
particular, for every t ∈ T , f is λt -integrable and
Z
R

f dλt = f dλt · a = λt (f ) · a

(Ch. IV, §4, No. 2, Cor. 2 of Th. 1). Thus, forRsuch an f , H = ∅ and,
writing Λ(f ) for the vector-valued function t 7→ f dλt (t ∈ T), one has

Λ(f ) = aΛ(f )
INT V.x63 integration of measures §3

pointwise on T , where Λ(f ) is the real-valued function t 7→ λ t (f ) (t ∈ T) .

By the numerical case of a) (resp. b)) already treated, Λ(f ) is essentially
µ-integrable (resp. µ-integrable, when Λ is vaguely continuous) and
Z
ν(f ) = Λ(f ) dµ .

1
Being finite-valued, Λ(f ) belongs to L R (µ) (resp. LR1 (µ) ). Let
h ∈ LR1 (µ) with Λ(f ) = h locally µ-almost everywhere (resp. µ-almost
everywhere;
R Rwhen Λ is vaguely continuous, h = Λ(f ) will serve). Then
Λ(f ) dµ = h dµ , thus Z
ν(f ) = h dµ .

Now, ah is µ-integrable and

Z
R

ah dµ = h dµ · a

(loc. cit., Cor. 2 of Th. 1); since Λ(f ) = ah locally µ-almost everywhere
(resp. µ-almost everywhere), Λ(f ) is essentially µ-integrable (resp. µ-inte-
grable) and
Z Z
R

Λ(f ) dµ = ah dµ = h dµ · a = ν(f ) · a = ν(f ) ,

that is, (11) holds for f .

V.24, `. 20, 21.
“ H is dense in LF1 (ν) ”
Ch. IV, §3, No. 5, Prop. 10.
V.24, `. 21–26.
“ . . . for every f ∈ LF1 (ν) , there exists a sequence (fn ) of elements
of H that has the following properties:
1) the sequence (fn ) converges to f in mean in LF1 (ν) , and ν-almost
everywhere; P
2) the function g = |f0 | + |fn+1 − fn | is such that ν*(g) < +∞
n∈N
(Ch. IV, §3, No. 4, Th. 3).”
From a sequence in H converging in mean to f , by the cited Th. 3
select a subsequence (fn )n∈N
P such that fn converges to f ν-almost ev-
erywhere in X , the series ν*(|fn+1 − fn |) is convergent and the series
n∈N
§3 integration of positive measures INT V.x64
P
|fn+1 (x) − fn (x)| is convergent for ν-almost every x ∈ X . Let g be the
n∈N
numerical function on X defined by
X
g(x) = |f0 (x)| + |fn+1 (x) − fn (x)| for all x ∈ X ,
n∈N
P
that is, g = |f0 | + |fn+1 − fn | . We know that g is finite for ν-almost
n∈N
every x ∈ X ; moreover, g is measurable for every measure on X , and
X
ν*(g) = ν*(|f0 |) + ν*(|fn+1 − fn |) < +∞
n∈N

(see item (β) in the Note for p. V.22, `. −2 to V.23, `. 3), therefore g is
ν-integrable.
V.24, `. −11, −10.
“ . . . N1 is locally µ-negligible (resp. µ-negligible) by formula (6) (resp.
(7)).”
Proof #1. As observed in the preceding note, the function g is ‘univer-
sally measurable’ and ν-integrable. In particular, g is λ t -measurable for all
t ∈ T , thus g is λt -integrable if and only if λ∗t (g) < +∞ ; in other words,

N1 = {t ∈ T : g is not λt -integrable },

therefore N1 is locally µ-negligible (resp. µ-negligible) by a) (resp. b)) for

the case of numerical functions. ♦
Lemma. If h is a positive function on T such that µ • (h) < +∞ ,
then h is finite locally µ-almost everywhere in T .
Proof. By 3) of §1, No. 2, Prop. 7, there exists a µ-moderated pos-
itive function h0 such that h = h0 locally µ-almost everywhere, whence
µ• (h) = µ• (h0 ) . Let A be a locally µ-negligible set such that h = h 0
on { A . By 2) of the cited Prop. 7,

µ*(h0 ) = µ• (h0 ) = µ• (h) < +∞ ,

therefore h0 is finite µ-almost everywhere (Ch. IV, §2, No. 3, Prop. 7);
let B be a µ-negligible set such that h 0 is finite on { B . If t ∈ { (A ∪ B) =
{ A ∩ { B , then h(t) = h(t0 ) and h0 (t) is finite, thus h(t) is finite on the
complement of the locally µ-negligible set A ∪ B . ♦
Proof #2: By (6) (resp. (7)) of No. 2, Prop. 3,



µ• Λ*(g) 6 ν*(g) < +∞ (resp. µ* Λ*(g) 6 ν*(g) < +∞ ) ,
INT V.x65 integration of measures §3

therefore Λ*(g) is finite locally µ-almost everywhere (resp. µ-almost every-

where) by the Lemma (resp. Ch. IV, §2, No. 3, Prop. 7); thus µ • (N1 ) = 0
(resp. µ*(N1 ) = 0 ). ♦
V.24, `. −10 to −7.
/ N1 , the fn belong to LF1 (λt ) , the sequence (fn ) converges
“For t ∈
λt -almost everywhere, as well as for the topology of convergence in mean in
LF1 (λt ) (Ch. IV, §3, No. 3, Prop. 6).”
Let us write

f00 = f0 , fn0 = fn − fn−1 for n = 1, 2, 3, . . . ;

it is to the sequence (fn0 )n∈N and the measure λt that the cited Prop. 6 will
be applied.
As observed in the Notes for `. 19, 20 and 21–26, for every measure on X
the fn are integrable—hence so are the fn0 —and g is measurable. Since
t∈/ N1 , λ∗t (g) < +∞ , therefore g is λt -integrable and, by the argument in
the latter Note, X
λ∗t (|fn0 |) = λ∗t (g) < +∞ ,
n∈N

that is, the series with general term λ ∗t (|fn0 |) is summable. Therefore, by
the cited Prop. 6, for λt -almost every x ∈ X the series with general term
fn0 (x) is absolutely convergent in F ; the n ’th partial sum of the series is
equal to fn (x) , thus, writing Nt for the set of all such x , we may define a
function X → F by

 lim fn (x) for x ∈ { Nt
n→∞
ft (x) =
 0 for x ∈ Nt .

We assert that ft is λt -measurable. For, the fn are universally measurable

and { Nt is λt -measurable, therefore the functions ϕ { Nt fn are
λt -measurable for every n , hence so is their pointwise limit f t (Ch. IV,
§5, No. 4, Th. 2).
Also by the cited Prop. 6, ft ∈ FF1 (λt ) , that is, λ∗t (ft ) < +∞ ; and fn
converges to ft in mean for λt , that is, λ∗t (|fn − ft |) → 0 . Since ft is
λt -measurable, we thus have ft ∈ LF1 (λt ) and fn → ft in the semi-normed
space LF1 (λt ) , therefore
Z Z
ft dλt = lim fn dλt
n→∞

because the integral is by definition continuous on L F1 (λt ) (Ch. IV, §4,

No. 1, Def. 1).
§3 integration of positive measures INT V.x66

V.24, `. −3 to −1.
“Suppose that t does not belong to N1 ∪ N2 ; the sequence (fn ) con-
verges in mean in LF1 (λRt ) , and converges
R λt -almost everywhere to f . There-
fore f ∈ LF1 (λt ) and f dλt = lim fn dλt (Ch. IV, §4, No. 1).”
n→∞
/ N1 we may construct ft ∈ LF1 (λt ) as in the preceding Note;
Since t ∈
in particular, fn → ft λt -almost everywhere in X , and fn → ft in mean in
LF1 (λt ) .
Since t ∈
/ N2 , M is λt -negligible, that is, fn (x) → f (x) for λt -almost ev-
ery x ∈ X ; but already fn (x) → ft (x) for λt -almost every x ∈ X , therefore
f = ft λt -almost everywhere. Since ft ∈ LRF1 (λt ) we have Ralso f ∈ LF1 (λt )
and fn → f in mean in LF1 (λt ) , whence f dλt = lim fn dλt (Ch. IV,
n→∞
§1, No. 1, Def. 1).
In particular, since f is λt -integrable we have t ∈
/ H , and we have shown
that H ⊂ N1 ∪ N2 ; therefore H is locally µ-negligible (resp. µ-negligible).
V.25, `. 2–4. R
“ . . . the function t 7→ f dλt is equal locally µ-almost everywhere to
the limit of a sequence of µ-measurable functions; it is therefore µ-measur-
able.”
Since the fn belong to H Rwe know (see the Note for V.24, `. 19, 20)
that the functions Λ(fn ) : t 7→ fn dλt are defined everywhereR in T , are
essentially µ-integrable (resp. µ-integrable), and satisfy ν(f n ) = Λ(fn ) dµ .
Moreover (see the R precedingR Note) for every t ∈ { (N1 ∪ N2 ) one has
f ∈ LF1 (λt ) and fn dλt → f dλt .
Define a function Λ(f ) : T → F by the formula
( R

lim fn dλt for t ∈ { (N1 ∪ N2 )
n→∞
(∗) Λ(f ) (t) =
0 for t ∈ N1 ∪ N2 .

The functions ϕ { (N1 ∪N2 ) Λ(fn ) are λt -measurable and converge pointwise
on T to Λ(f ) , therefore Λ(f ) is λt -measurable (Ch. IV, §5, No. 4, Th. 2).
Moreover, Λ(fn ) → Λ(f ) locally µ-almost everywhere (resp. µ-almost
everywhere) in T by (∗).
V.25, `. 8–11.
R∗
“Now, the function t 7→ g(x) dλt (x) is essentially µ-integrable (resp.
µ-integrable) by Prop. 5. We may therefore apply Lebesgue’s theorem, which
yields
Z Z Z Z Z
dµ(t) f (x) dλt (x) = lim dµ(t) fn (x) dλt (x) = lim fn (x) dν(x). ”
n→∞ n→∞
INT V.x67 integration of measures §3

b) With notations as in the preceding Note, consider first the case that
Λ is vaguely continuous, the Λ(fn ) are µ-integrable, and Λ(fn ) → Λ(f )
µ-almost everywhere in T . By the case of a positive numerical function, we
know that the function Λ*(g) (defined everywhere on T ) is µ-integrable;
moreover, as just noted in the text, |Λ(f n )| 6 Λ*(g) on { (N1 ∪ N2 ) , that is,
µ-almost everywhere in T . It follows from Lebesgue’s theorem (Ch. IV, §3,
No. 7, Th. 6) that RΛ(f ) is µ-integrable
R and Λ(f n ) converges to Λ(f ) in mean
for µ , therefore Λ(fn ) dµ → Λ(f ) dµ . Finally since, by construction,
fn → f in mean for ν , we have
Z Z Z Z
f dν = lim fn dν = lim Λ(fn ) dµ = Λ(f ) dµ ,
n→∞ n→∞

thus f satisfies (11).

a) Assuming only that f is ν-integrable, we know that the (everywhere
defined) functions Λ(fn ) are essentially µ-integrable, Λ(f ) is µ-measurable,
and Λ(fn ) → Λ(f ) locally µ-almost everywhere in T . By the case of a posi-
tive numerical function, the function Λ(g) : t 7→ λ •t (g) (t ∈ T) is essentially
µ-integrable and Λ(g) = Λ*(g) locally µ-almost everywhere in T . And
|Λ(fn )| 6 Λ*(g) on { (N1 ∪ N2 ) , that is, locally µ-almost everywhere in T .
Choose functions hn , k ∈ LF1 (µ) such that Λ(fn ) = hn and Λ*(g) = k
locally µ-almost everywhere in T (§1, No. 3, Def. 3). Then h n → Λ(f ) and
|hn | 6 |k| locally µ-almost everywhere in T . Now, the sequence (h n (t)) is
Cauchy in F for locally µ-almost every t ∈ T ; but the set of t ∈ T such
that (hn (t)) is not Cauchy, namely, the set
∞ \
[ ∞ [ 1
Q= {t ∈ T : |hm (t) − hn (t)| > },
r=1 N =1 m,n>N, m6=n
r

is µ-moderated because the functions h m − hn are µ-moderated; thus Q is

both locally negligible and moderated for µ , hence is µ-negligible (§1, No. 2,
Cor. 1 of Prop. 7), that is, the sequence (h n (t)) is Cauchy in F for µ-almost
every t in T . Defining

 lim hn (t) for t ∈ T --- Q
n→∞
h(t) =
 0 for t ∈ Q ,

we have hn (t) → h(t) µ-almost everywhere in T , therefore h is µ-measurable.

Similarly, since hn and k are µ-moderated, we have |hn | 6 |k|
µ-almost everywhere. It now follows from Lebesgue’s theorem
R that
R h is
µ-integrable, hn converges to h in mean for µ , and so hn dµ → h dµ ,
§3 integration of positive measures INT V.x68
R R
that is, Λ(fn ) dµ → h dµ . Since hn → Λ(f ) locally µ-almost every-
where, we have Λ(f ) = h locally µ-almost everywhere, therefore Λ(f ) is
essentially µ-integrable and
Z Z Z Z Z
Λ(f ) dµ = h dµ = lim Λ(fn ) dµ = lim fn dν = f dν ,
n→∞ n→∞

thus f satisfies (11).

Remark. In case it is needed in the future: the foregoing argument shows
that Lebesgue’s theorem holds with “integrable” replaced by “essentially
integrable”, and “almost everywhere” by “locally almost everywhere”.
V.25, `. 17–19.
“Then g = f almost everywhere for λt , except for t forming a locally
µ-negligible set P (Cor. 1 c) of Prop. 3).”
We are assuming Λ is µ-adequate and λ •t (1) < +∞ for locally µ-almost
every t ∈ T . Given an essentially ν-integrable function g with values in F
or in R , let
Hg = {t ∈ T : g is not λt -integrable}
and write Λ(g) for the function on T defined by
Z

 g dλt for t ∈
/ Hg


Λ(g) (t) =


0 for t ∈ Hg ;

we are
R to show thatR Hg is locally µ-negligible, Λ(g) is essentially µ-integrable,
and Λ(g) dµ = g dν .
The letter f was reserved for a Rν-integrable
R function such that g = f
locally ν-almost everywhere, so that g dν = f dν (§1, No. 3, Def. 3), and
to which part a) of the theorem is to be applied; we can suppose that if g
is a numerical function then f ∈ LR1 (ν) , that is, all values of f are finite
(which facilitates subtraction). Writing

H = {t ∈ T : f is not λt -integrable} ,

we know from part a) that H is locally µ-negligible, that the function Λ(f )
defined by Z

 f dλt

for t ∈
/H
Λ(f ) (t) =


0 for t ∈ H
R R
is essentially µ-integrable, and that Λ(f ) dµ = f dν .
INT V.x69 integration of measures §3

The numerical function |g − f | is locally ν-negligible; by the cited

Cor. 1 c) of Prop. 3, the set

P = {t ∈ T : λ∗t (|g − f |) > 0 }

is locally µ-negligible. Thus the set P ∪ H is locally µ-negligible.

Suppose t ∈ { (P ∪ H) ; then f is λt -integrable (because t ∈ / H ) and
g = f λt -almostReverywhereR (because t ∈
/ P), therefore g is λ t -integrable, i.e.,
t ∈ { Hg , and g dλt = f dλt . In particular, Hg ⊂ P ∪ H , hence Hg is
locally µ-negligible. Moreover, Λ(g) = Λ(f ) on { (P∪H) , thus Λ(g) = Λ(f )
locally µ-almost everywhere, whence Λ(g) is essentially µ-integrable and
Z Z Z Z
Λ(g) dµ = Λ(f ) dµ = f dν = g dν ,

thus g satisfies (11).

V.25, `. −13, −12.
“ . . . it follows at once from the definitions that Λ 0 is also µ-adequate,
and that Λ and Λ0 have the same integral.”
Given two functions Λ : t 7→ λt , Λ0 : t 7→ λ0t of T into M+ (X) , such
that λt = λ0t for locally µ-almost every t ∈ T : R
(i) If Λ is scalarly essentially µ-integrable,R with ν = R λt dµ(t) , then
Λ0 is also scalarly essentially µ-integrable and λ0t dµ(t) = λt dµ(t) , that
is, ν 0 = ν .
For, if f ∈ K (X) then the functions

Λ(f ) : t 7→ λt (f ) , Λ0 (f ) : t 7→ λ0t (f )

are equal locally µ-almost everywhere in T ;
since Λ(f )
is essentially

µ-integrable so is Λ0 (f ) and ν 0 (f ) = µ• Λ0 (f ) = µ• Λ(f ) = ν(f ) for
all f ∈ K (X) , whence the assertion.
(ii) If Λ is µ-pre-adequate then so is Λ 0 .
For, for every lower semi-continuous function f > 0 , Λ 0 (f ) = Λ(f )
locally µ-almost everywhere in T and



µ• Λ0 (f ) = µ• Λ(f ) = ν • (f ) = ν 0• (f ) ,

thus Λ0 is µ-pre-adequate.
(iii) If Λ is µ-adequate then so is Λ0 . For, given any positive mea-
sure ρ 6 µ , Λ = Λ0 locally ρ-almost everywhere and Λ is ρ-pre-adequate,
therefore so is Λ0 .
§3 integration of positive measures INT V.x70

V.25, `. −6, −5.

“ . . . the propositions proved in the preceding Nos. extend to µ-adequate
functions defined locally µ-almost everywhere.”
I don’t see anything here except notation: if one comes across a function
H of the indicated type, one should do business with R a corresponding Λ ,
with the understanding
R that the measure denoted ηt dµ(t) is simply the
measure ν = λt dµ(t) and depends only on H , or on the equivalence class
of Λ under the relation of equality locally µ-almost everywhere.
The following theme seems to me more interesting: Start with a function
H : t 7→ ηt (t ∈ A) , where A is a subset of T such that µ • (T --- A) = 0
and ηt ∈ M+ (X) for all t ∈ A . For each f ∈ K (X) we have a numerical
function
H(f ) : t 7→ ηt (f ) (t ∈ A) .
Suppose that, by some miracle, for every f ∈ K (X) the function H(f )
is essentially µ-integrable, in other words that there exists a function g in
LR1 (µ) such that H(f ) = g locally µ-almost everywhere in T (§1, RNo. 3,
first paragraph on p. V.9). One can unambiguously define ν(f ) = g dµ
and it is clear that ν is a positive linear form on K (X) , i.e., is a positive
measure on X . If Λ : T → M+ (X) is any extension of H to T with values
in M+ (X) , then, for every f ∈ K (X) , the function

Λ(f ) : t 7→ λt (f ) (t ∈ T)

is equal locally µ-almost everywhere to H(fR) , hence to some

R g ∈ L R1 (µ) ,
thus Λ(f ) is essentially µ-integrable and
R Λ(f ) dµ = g dµ ; and since
H(f ) = g locally µ-almost everywhere, g dµ = ν(f ) , whence
Z
(∗) Λ(f ) dµ = ν(f ) .

R
Summarizing: Λ is scalarly essentially
R µ-integrable;

λt dµ(t) = ν , because
the left side of (∗) is equal to λt dµ(t) (f ) ; and Λ has the contemplated
relation to H .
Will any of this prove to be useful? (See also §4, No. 4, Remark.)
V.26, `. −6, −5.
“ N is a universally measurable set”
We are to show that the (numerical) function ϕ N is universally measur-
able. For every t ∈ T let Vt be a compact (hence ‘universally integrable’)
neighborhood of t in T . By the Principle of Localization (Ch. IV, §5, No. 2,
Prop. 4) it suffices to show that for every t ∈ T there exists a universally
INT V.x71 integration of measures §3
  
measurable function gt : T → R such that ϕN Vt = gt Vt . Now, ϕN Vt is
equal to 1 on Vt ∩ N and to 0 on Vt --- Vt ∩ N , therefore
 
ϕN Vt = ϕVt ∩N Vt ;

since ϕVt ∩N is universally measurable (because V t ∩ N is a Borel set), the

functions gt = ϕVt ∩N meet the requirements.
V.26, `. −5.
“ f 0 is a universally measurable function (Ch. IV, §5, No. 10, Prop. 16).”
The sets N and Ki (i ∈ I) are universally measurable and form a locally
countable family with union T , so it will suffice, by the cited Prop. 16, to
show that the restriction
 of f 0 to each of these sets is universally measurable.
The function f N is constant and the functions f 0 Ki are continuous; the
0

argument that they are all universally measurable is given in the Note for
V.22, `. 11–13.
V.26, `. −2.
“Definition 3.”
We record here some more-or-less immediate consequences of the defi-
nition.
(i) If Λ : t 7→ λt is pre-adequate for every positive measure on T with
compact support, then Λ is a diffusion.
For, if µ is a positive measure with compact support then every positive
measure µ0 6 µ also has compact support (Ch. III, §2, No. 2, Prop. 3).
(ii) If µ is a positive measure on T and Λ : t 7→ λ t is a mapping
T → M+ (X) such that Λ is µ-pre-adequate, it is implicit that Λ is scalarly
essentially µ-integrable (No. 1, Def. 1).
(iii) Let µ be any positive measure on T . One knows that µ is the sum
of a family (µα )α∈A of positive measures on T with compact support (§2,
No. 3, Prop. 4). Thus, if Λ : t 7→ λt is a diffusion then Λ is µα -adequate
(hence scalarly essentially µα -integrable) for every α , but one does not know
whether Λ is scalarly essentially µ-integrable.
However, if Λ is scalarly essentially µ-integrable and if µ is the sum of
a family (µα )α∈A of positive measures such that R Λ is µ α -pre-adequate for
every α , then Λ is µ-pre-adequate and (with signifying essential integral)
Z XZ
λt dµ(t) = λt dµα (t)
α∈A

(see Proposition A in the Note for V.17, `. −4).

§3 integration of positive measures INT V.x72

(iv) Let µ be a positive measure on T . If a mapping Λ : t 7→ λ t is

scalarly essentially µ-integrable and if Λ is µ-pre-adequate for every positive
measure µ0 6 µ with compact support, then Λ is µ-adequate (see part d)
of Exer. 8, worked out in the Note for V.17, `. −4).
(v) Let Λ : t 7→ λt be a diffusion T → M+ (X) and let µ be a positive
measure on T . In order that Λ be µ-adequate, it is necessary and sufficient
that it be scalarly essentially µ-integrable. (This is equivalent to Prop. 11
below.)
Necessity: (ii).
Sufficiency: By (iv) and the definition of a diffusion.
(vi) If Λ : t 7→ λt is a diffusion T → M+ (X) and if g ∈ K+ (X) , then
the function Λ(g) : t 7→ λt (g) is universally measurable, and is integrable
for every positive measure on T with compact support. P
For, let µ be any positive measure on T , and write µ = µα with
α∈A
(µα )α∈A a family of positive measures on T with compact support. Since Λ
is a diffusion, it is µα -adequate for every α , therefore Λ(g) is µ α -measurable
for all α (No. 1, Def. 1), whence Λ(g) is µ-measurable (§2, No. 2, Prop. 2).
Suppose now that µ is any positiveR measure with compact support.
Then Λ is µ-adequate, and if ν = λt dµ(t) and g ∈ K+ (X) , one has


µ• Λ(g) = ν • (g) = ν(g) < +∞ ,

thus Λ(g) is essentially µ-integrable (§1, No. 3, Prop. 9). But µ is bounded
(Ch. III, §2, No. 3, Prop. 11), hence moderated (§1, No. 2, Remark 2),
therefore every function on T is µ-moderated (loc. cit., Def. 2), whence
Λ(g) is µ-integrable (§1, No. 3, Cor. of Prop. 9).
V.27, `. −14, −13.
P 1
“ . . . then u is not integrable for the measure µ = ε
2 tn
with
n>1 n
compact support ”
1
(i) The family of measures εt (n > 1) is summable: for, if
n2 n
f ∈ K+ (T) then for every finite set J of integers > 1 one has
X 1
X 1 X 1
2
εtn (f ) = 2
f (tn ) 6 kf k < +∞ ,
n n n2
n∈J n∈J n>1

where kf k is the ‘sup-norm’ of f .

(ii) If h is a lower semi-continuous function > 0 on T then ε ∗t (h) =
h(t) for every t ∈ T .
INT V.x73 integration of measures §3

For, h is the upper envelope of the set A of all f ∈ K + (T) such that
f 6 h , and, by definition,

ε∗t (h) = sup εt (f ) = sup f (t) = h(t) .

f ∈A f ∈A

{Since h is universally measurable (Ch. IV, §5, No. 5, Cor. of Prop. 8), it
follows that h is εt -integrable ⇔ h(t) < +∞ (loc. cit., No. 6, Th. 5).}
(iii) µ*(u) = +∞ (whence the assertion).
By definition, µ*(u) is the infimum of µ*(h) for all lower semi-continuous
functions h > u , so it will suffice to show that µ*(h) = +∞ for all such h .
Indeed, by §2, No. 2, Cor. 3 of Prop. 1, one has
X 1
∗ X 1 X 1
µ*(h) > ε t (h) = h(t n ) > u(tn ) = +∞
n
n2 n n
n2 n
n2

1
because u(tn ) > 1 . ♦
n2
R example. The computation in (ii) throws light on the formula
An
µ = εx dµ(x) of Ch. III, §3, No. 1, Example 2, where it is restricted to
functions in K (X) and is merely a notation.
In the present context, consider the mapping Λ : x 7→ ε x of X into
M+ (X) , that is, T = X and λx = εx for all x ∈ X . Since εx is bounded,
ε•x = ε∗x (§1, No. 2,
Cor. 2 of Prop. 7), thus, for every function f > 0
on X , we have Λ(f ) (x) = ε•x (f ) = ε∗x (f ) for all x ∈ X . In particular, if
f ∈ K+ (X) the function

Λ(f ) : x 7→ ε∗x (f ) = εx (f ) = f (x)

is ‘universally integrable’; for every positive measure µ on X and every

f ∈ K+ (X) one has

µ• Λ(f ) = µ• (f ) = µ(f ) ,
R
thus Λ is scalarly essentially µ-integrable and ν = µ, that is, εx dµ(x) = µ.
Moreover, the continuity of Λ(f ) = f for every f ∈ K + (X) shows that Λ
is vaguely continuous (better yet, see Ch. III, §1, No. 9, Prop. 13). {In fact,
Λ(f ) = f for every function f > 0 on X ; see item (v) later in this note.}
If h is any lower semi-continuous function > 0 on X , by (ii) above one
has

Λ(h) (x) = ε•x (h) = ε∗x (h) = h(x) ,
thus Λ(h) is universally measurable (Ch. IV, §5, No. 5, Cor. of Prop. 8),
and, for every positive measure µ on X ,


µ• Λ(h) = µ• (h) = ν • (h) ;
§3 integration of positive measures INT V.x74

thus Λ is µ-adequate for every positive measure µ on X , and in particular

Λ is a diffusion, indeed, a bounded diffusion ( kε x k = 1 for all x ).
Fix a positive measure µ on X and consider a µ-integrable function
f > 0 . Let us apply No. 3, Th. 1, b): the set

H = {x ∈ X : f is not εx -integrable}

is µ-negligible, the function

Λ(f ) : x 7→ ε∗x (f )
R R
is µ-integrable, and f dν = Λ(f ) dµ , that is
Z Z
(∗) f dµ = Λ(f ) dµ .

As in the proof of the cited Th. 1, one extends this result, in particular (∗),
to an arbitrary µ-integrable function f by considering |f | = f + + f − and
f = f + −f − . We shall show that the underlying reason for the equality (∗) is
that Λ(f ) = f µ-almost everywhere; this entails a close look at the meaning
of εx -integrability.
Fix a point a ∈ X .
(i) The key fact is that Supp(εa ) = {a} . (See the Note for III.29,
`. 2–4.)
(ii) The negligible sets for εa are the subsets of X --- {a} , that is,
the subsets of X that do not contain a . For, X --- {a} = { Supp(εx ) is
εa -negligible; whereas if a ∈ A ⊂ X then ε ∗a (A) > ε∗a ({a}) = 1 because
ϕ{a} = ϕX = 1 εa -almost everywhere, therefore A is not ε a -negligible.
For something to happen εa -almost everywhere in X , it is necessary
and sufficient that it happen at a ; for, the set A where it does not happen
is εa -negligible if and only if a ∈
/ A.
For example, if f, g are functions on X then f = g ε a -almost every-
where if and only if f (a) = g(a) .
(iii) Every function f on X is εa -measurable (obvious from Ch. IV, §5,
No. 1, Def. 1, with N = K --- {a} ). It follows that f is ε a -integrable if and
only if ε∗a (|f |) < +∞ (loc. cit., No. 6, Th. 5).
(iv)
R A numerical function f is εa -integrable ⇔ f (a) is finite, in which
case f dεa = f (a) .
⇒: If f is εa -integrable, then the set A = {x ∈ X : f (x) = ±∞ } is
negligible for εa (Ch. IV, §2, No. 3, Prop. 7), thus a ∈
/ A.
INT V.x75 integration of measures §3

⇐: If f (a) is finite, let g ∈ K (X) with g(a) = f (a) . Then g is

εa -integrable and f = g almost everywhere for ε a , therefore f is εa -inte-
grable and Z Z
f d εa = g dεa = εa (g) = g(a) = f (a) .

(v) Summarizing (iii) and ∗

R (iv): f is ε a -integrable ⇔ εa (|f |) < +∞ ⇔
f (a) is finite, in which case f dεa = f (a) .
If f > 0 then ε∗a (f ) = f (a) always; for, if f (a) = +∞ then f is not
εa -integrable, so ε∗a (f ) = +∞ = f (a) . ♦
Returning to the discussion of a µ-integrable function f , in the context
of Th. 1 of No. 3 we have
H = {x ∈ X : f is not εa -integrable}
= {x ∈ X : |f | is not εa -integrable}
= {x ∈ X : |f (x)| = +∞ } ,

and we know from (v) above that for x ∈

/ H one has
Z


Λ(f ) (x) = f dεx = f (x) ;

thus Λ(f ) = f µ-almost everywhere, and the message of (∗) is that

Z Z Z
R

f dµ = f dν = f d εx dµ(x) ,
R
further justification for the notation µ = εx dµ(x) of Ch. III.
V.27, `. −13 to −11.
“ . . . contrary to the hypothesis on Λ , which implies that t 7→ λ t (g) is
integrable for every positive measure with compact support.”
See item (vi) of the Note for V.26, `. −2.
V.27, `. −10, −9.
“Conversely, the conditions 1) and 2) imply that Λ is scalarly essentially
µ-integrable for every measure µ with compact support.”
Conditions 1) and 3) imply that Λ is scalarly essentially µ-integrable
for every positive Rmeasure with compact support, but stating 3) requires
knowing that ν = λt dµ(t) exists; thus we must infer from 1) and 2) that
Λ is scalarly essentially µ-integrable.
Given g ∈ K+ (T) , we must show that the mapping Λ(g) : t 7→ λ t (g)
is scalarly essentially µ-integrable;

since it is µ-measurable by 1), it suffices
to show that µ* Λ(g) < +∞ (Ch. IV, §5, No. 6, Th. 5).
§3 integration of positive measures INT V.x76

By assumption the set K = Supp(µ) is compact. By 2), every point

of T has a neighborhood on which Λ(g) is bounded; therefore Λ(g) is
bounded on K (cover K with a finite number of such neighborhoods). Say
ϕK Λ(g) 6 M ϕK , where M < +∞ ; since µ( { K) = 0 , ϕK = 1 µ-almost
everywhere, therefore

Λ(g) = ϕK Λ(g) 6 M ϕK µ-almost everywhere,

whence µ* Λ(g) 6 M µ*(K) < +∞ .
Remark. Note that the above proof cites 1) only for f ∈ K + (X) .
V.28, `. 4.
“ Λ is scalarly essentially µ-integrable ”
See the Remark at the end of the preceding note. This property is
required for the application of No. 1, Prop. 2.
V.28, `. 10.
“ . . . we shall denote by Λf the mapping t 7→ λ •t (f ) . ”
The notation Λf is consistent with the notation Λ(f ) used earlier in
these notes (see, e.g., the note for V.20, `. −4 to −2), but is more restric-
tive. The restrictions: Λ must be a diffusion, and f must be a universally
measurable function > 0 .
V.28, `. 10, 11.
“If µ is a positive measure on T such that Λ is scalarly essentially
µ-integrable . . . ”
As is the case when Λ is µ-pre-adequate (No. 1, Def. 1), in particular
when µ has compact support (Def. 3).
V.28, `. 12–14.
“The definition of the integral then takes the form

hµΛ, f i = hµ, Λf i for f ∈ K+ (X) . ”

R
By “the integral” is meant the measure µΛ = λt dµ(t) . The displayed
condition is
(µΛ)• (f ) = µ• (Λf ) for all f ∈ K+ (X) ;
by employing the symbol µΛ , it presumes that the given diffusion Λ is
scalarly essentially
R µ-integrable and recapitulates the definition of the mea-
sure ν = λt dµ(t) ,


ν(f ) = µ• Λf for all f ∈ K+ (X) ,
INT V.x77 integration of measures §3

which is formula (1) of No. 1 written for a diffusion Λ .

R
{It can’t be the definition of λt dµ(t) , as the presence of the diffusion
Λ requires (via the definition of adequacy) that one already know how to
define the integral of a scalarly essentially µ-integrable mapping t 7→ λ t
(t ∈ T) . There is here a bit of potential incest that is to be avoided.}

V.28, `. 16–18.
“ . . . this amounts to saying (in view of Prop. 8) that Λ is scalarly
essentially µ-integrable and hµ0 Λ, f i = hµ0 , Λf i for every positive measure
µ0 6 µ and every lower semi-continuous positive function f . ”
(i) Suppose that the diffusion Λ is scalarly essentially µ-integrable and
satisfies the stated condition. Then, for every positive measure µ 0 6 µ ,
one knows that Λ is scalarly essentially µ 0 -integrable (see the Note for V.17,
`. 12, 13), so that the measure µ0 Λ = ν 0 is defined; and, for every lower semi-
continuous function f > 0 , Λf is universally measurable by 1) of Prop. 8,
so the assumption that
ν 0• (f ) = µ0• (Λf )

for all such f says that Λ is µ0 -pre-adequate, and this, for every positive
measure µ0 6 µ , thus Λ is µ-adequate (No. 1, Def. 1).
(ii) Conversely, if µ is a positive measure on T such that Λ is
µ-adequate, then it is clear from No. 1, Def. 1 that Λ is scalarly essen-
tially µ-integrable and satisfies the stated equality for every positive measure
µ0 6 µ and every lower semi-continuous function f > 0 (with the added
spice, via 1) of Prop. 8, that Λf is universally measurable).
It turns out (Prop. 11 below) that the stated condition is superfluous:
the diffusion Λ is µ-adequate for a measure µ > 0 if and only if it is scalarly
essentially µ-integrable. {See also item (v) in the Note for V.26, `. −2.}

V.28, `. 19–28.
“Proposition 10.”
See the next Note.

V.28, `. 28.
“See also the next proposition.”
An excellent suggestion. Prop. 11 has already been verified in an earlier
note (item (v) in the Note for V.26, `. −2): the proposition says that if Λ is
a diffusion and µ is a positive measure on T , then Λ is µ-adequate if and
only if it is scalarly essentially µ-integrable. The proof of Prop. 10 can then
be conducted as follows.
§3 integration of positive measures INT V.x78

a) Since f and g are universally measurable, so are f + g and af .

For every t ∈ T ,

λ•t (f + g) = λ•t (f ) + λ•t (g) (§1, No. 1, Prop. 2)

λ•t (af ) = aλ•t (f ) (loc. cit., Prop. 1)

thus
Λ(f + g) : t 7→ (Λf )(t) + (Λg)(t) = (Λf + Λg)(t)


Λ(af ) : t 7→ a(Λf )(t) = a(Λf ) (t) ,
whence Λ(f + g) = Λf + Λg and Λ(af ) = a(Λf ) .
b) Assuming that Λ is µ-adequate and ν-adequate, so that, in partic-
ular, Λ is scalarly essentially integrable for both µ and ν , we are to show
that Λ is (µ + ν)-adequate and that (µ + ν)Λ = µΛ + νΛ , that is,
Z Z Z
λt d(µ + ν)(t) = λt dµ(t) + λt dν(t) .

Let f ∈ K+ (X) . Since Λ is scalarly essentially integrable for both µ and ν ,

we know that Λf is essentially integrable for both µ and ν , that is,
1 1 1
Λf ∈ L (µ) ∩ L (ν) = L (µ + ν)

(in the Note for V.10, `. 13, 14, see item (i) of Cor. 3 of the Theorem there),
therefore Λf is essentially (µ + ν)-integrable and

(µ + ν)• (Λf ) = µ• (Λf ) + ν • (Λf )

(§1, No. 1, Prop. 3); thus Λ is (µ + ν)-scalarly essentially integrable—hence

(µ + ν)-adequate by Prop. 11—and, for every f ∈ K + (X) ,

•
(µ + ν)Λ (f ) = (µ + ν)• (Λf ) = µ• (Λf ) + ν • (Λf ) = (µΛ)• (f ) + (νΛ)• (f )

by the definition of adequacy (No. 1, Def. 1), that is,

•
(µ + ν)Λ (f ) = (µΛ + νΛ)• (f )

(§1, No. 1, Prop. 3), whence (µ + ν)Λ = µΛ + νΛ (§1, No. 1, comment

preceding Prop. 1).
Similarly for aµ .
V.28, `. −11 to −9.
“Proposition 11.”
See item (v) in the Note for V.26, `. −2.
INT V.x79 integration of measures §3

V.29, `. 6, 7.
“Corollary 1.”
Note that kλt k = λ∗t (1) < +∞ and kµk = µ*(1) < +∞ (Ch. IV,
§4, No. 7, Prop. 12); moreover, λ•t = λ∗t and µ• = µ* because bounded
measures are moderated (§1, No. 2, Remark 2 and Prop. 7).
To show that Λ is µ-adequate, it suffices (Prop. 11) to show that Λg is
µ-integrable (hence essentially µ-integrable) for every g ∈ K + (X) . At any
rate, Λg is universally measurable (Prop. 8). For all t ∈ T one has

0 6 λt (g) 6 kλt k kgk 6 kΛk kgk ,

thus Λg is bounded and

µ*(Λg) 6 kΛk kgk µ*(1) = kΛk kgk kµk < +∞ ,

therefore Λg is µ-integrable (Ch. IV, §5, No. 6, Th. 5). Thus Λ is scalarly
essentially µ-integrable and

(µΛ)(g) = µ*(Λg) 6 kΛk kgk kµk .

Then, for every g ∈ K (X) , one has

|(µΛ)(g)| 6 (µΛ)(|g|) 6 kΛk kgk kµk ,

thus the measure µΛ is bounded and kµΛk 6 kΛk kµk .

V.29, `. 12.
“It suffices to apply the Corollary of Prop. 1 of No. 1.”
P
To set the stage slightly differently, let µ = µα , where (µα )α∈A is
α∈A
any summable family of positive measures on T .
The cited Corollary says that Λ is scalarly essentially integrable for µ if
and only if it is scalarly essentially integrable for
P every µ α and the resulting
measures µα Λ are summable, in which case µα Λ = µΛ .
α∈A
In the light of Prop. 11, one can replace “scalarly essentially integrable”
by “adequate”.
V.29, `. 15–19.
“Proposition 12.”
Prop. 5 pertains to a mapping Λ : t 7→ λt and a measure µ such that
Λ is µ-adequate in part c) but may be µ-pre-adequate in parts a) and b) (see
the comments following No. 1, Def. 1); for simplicity, a global assumption of
µ-adequacy is imposed on No. 2, where Prop. 5 appears.
§3 integration of positive measures INT V.x80

In the context of a diffusion Λ , even if the statement of Prop. 5 only

assumes that Λ is scalarly essentially µ-integrable, Prop. 11 ensures that it
is µ-adequate. Prop. 5 remains more general than Prop. 12 in that f is only
required to be measurable for ν = µΛ ; the reason that f is assumed to be
universally measurable in Prop. 12 is that the notation Λf requires
R it.
At any rate, in the notations of Prop. 12, writing ν = µΛ = λt dµ(t) ,
formula (13) says that
ν • (f ) = µ• (Λf ) ;
when f is ν-moderated, this is the relation (9) given by part a) of Prop. 5,
and when the measures λt are bounded it is the (same) relation (9) given
by part c) of Prop. 5 (where the λt are only required to be bounded for
locally µ-almost every t ∈ T ).
V.29, `. −11 to −9.
“Corollary.”
One is assuming, as in Prop. 12, that µ is a positive measure on T such
that the diffusion Λ is µ-adequate, and that f is a universally measurable
function > 0 on X .
If X is countable at infinity then every numerical function on X is
moderated for every positive measure on X (in particular, f is moderated
for µΛ ). Thus the hypotheses in the Cor. fulfill the hypotheses in Prop. 12,
consequently Λf is µ-measurable and (13) holds. But, by the definition of
diffusion, the role of µ in Prop. 12 can be played by any positive measure
on T with compact support, therefore Λf is universally measurable by
Prop. 6 of No. 4.
From the foregoing we can extract the following, for application in No. 6:
Scholium. Let Λ be a bounded diffusion of T in X and let f > 0 be
a universally measurable function on X . Then:
(i) Λf is a universally measurable function on T ;
(ii) if µ is a positive measure on T that belongs to the domain of Λ ,
then hµΛ, f i = hµ, Λf i , that is, (µΛ)• (f ) = µ• (Λf ) .
(iii) if µ is a bounded positive measure on T , then µ belongs to the
domain of Λ (hence hµΛ, f i = hµ, Λf i ), µΛ is a bounded positive measure
on X , and kµΛk 6 kµk kΛk .
For, (i) and (ii) follow from the present Corollary, and (iii) is the con-
clusion of Cor. 1 of Prop. 11.
V.30, `. 4.
“Set γt = λt H ”
See item (iii) in the preceding Note: since H is a bounded diffusion of
X in Y and λt is a bounded measure on X , λt belongs to the domain
INT V.x81 integration of measures §3

of H and kλt Hk 6 kλt k kHk , whence kλt Hk 6 kΛk kHk ; moreover, if f is

any universally measurable function > 0 on Y , the function

Hf : x 7→ ηx• (f ) = ηx∗ (f )

on X is universally measurable and hλ t H, f i = hλt , Hf i , that is,

γt• (f ) = λ•t (Hf ) .

V.30, `. 4–6.
“ . . . we shall denote by Γ the mapping ΛH of T into M + (Y) , and
by Γf the function t 7→ hγt , f i (an abuse of notation, since we do not yet
know whether Γ is a diffusion).”
This No. is about the composition of diffusions, an operation that is
applied to mappings. How are we to regard a diffusion as a mapping? Let
me count the ways . . .
1◦ When mappings Λ : t 7→ λt were introduced in No. 1, clearly Λ was
being regarded as a mapping T → M+ (X) .
2◦ But Λ also induces a mapping F+ (X) → F+ (T) : if f ∈ F+ (X)
the corresponding element of F+ (T) is the function

t 7→ λ•t (f ) (t ∈ T) ,

which we have been denoting in these notes by Λ(f ) (with parentheses)

since the Note for V.17, `. 12, 13; in this sense, Λ : F + (X) → F+ (T) .
3◦ Finally, if Λ is a diffusion, we have a correspondence µ 7→ µΛ ,
where “ µ is in the domain of Λ ”. It is time to introduce a notation for “the
domain of Λ ”; I propose writing

DΛ = {µ ∈ M+ (T) : Λ is µ-adequate } .

Then Λ may be regarded as the mapping Λ : D Λ → M+ (X) defined by

µ 7→ µΛ (µ ∈ DΛ ) . It is this perspective that is appropriate for the compo-
sition of diffusions: given diffusions Λ of T in X , and H of X in Y , if a
measure µ on T belonging to DΛ is such that the measure µΛ on X be-
longs to DH , then we may form the measure (µΛ)H on Y ; such measures µ
form the domain (in the set-theoretic sense) of the composite mapping Λ◦H
(resp. H ◦ Λ ) according as mappings are written to the right (resp. left) of
the elements on which they act.
A mildly instructive example is the bounded diffusion t 7→ ε t of T
in T considered in the Note for V.27, `. −14, −13; let us denote it by E .
As shown there, every positive measure µ on T is in the domain of E and
§3 integration of positive measures INT V.x82

µE = µ ; that is, DE = M+ (T) , and E as a mapping of measures is the

identity mapping on M+ (T) . As functions acting on the right,

DEΛ = {µ ∈ M+ (T) : µE = µ ∈ DΛ } = DΛ

and µ(EΛ) = (µE)Λ = µΛ ; thus EΛ = Λ . Similarly D ΛE = DΛ and

ΛE = Λ . Of course
{εt : t ∈ T } ⊂ DΛ ,
and if g ∈ K+ (T) then the computation

hεt Λ, gi = hεt , Λgi = ε•t (Λgi = (Λg)(t) = λt (g)

shows that εt Λ = λt ; thus, when Λ is restricted to this subset of D Λ

(vaguely homeomorphic to T ), the result is in effect the presentation of Λ
as a function T → M+ (X) .
V.30, `. 6–28.
“Then hγt , f i = . . . in the course of the above proof.”
The main goal is to prove that the mapping Γ : t 7→ γ t of T into
M+ (Y) is a (bounded) diffusion of T in Y ; the notation Γ = ΛH is
proposed for it, but the relation of the notation with the set-theoretic com-
position of mappings remains to be developed (see item (viii) below).
The proof is essentially a repeated application of one or more parts
of the Scholium in the Note for V.29, `. −11 to −9 (referred to briefly as
Scholium).
All numerical functions occurring in the proof will, one way or another,
be universally measurable. It is helpful to ‘color-code’ the functions: the
letter f is reserved for a general universally measurable function > 0 on Y ;
g for an element of K+ (Y); and h for a lower semi-continuous function > 0
on Y ; they are eligible to be acted on by Γ or H , to produce functions on T
or X . Until it is established that Γ is a diffusion, we provisionally employ
the notation Γ(f ) (with parentheses) for the mapping t 7→ γ t (t ∈ T) .
(i) By the Scholium, if f > 0 is a universally measurable function on Y
then Hf is a universally measurable function on X and, since the bounded
measures λt belong to the domain of H , one has

hλt H, f i = hλt , Hf i ,

that is, γt• (f ) = λ•t (Hf ) for all t ∈ T ; in other words,

(∗) Γ(f ) = Λ(Hf )

for every universally measurable function f > 0 on Y .

INT V.x83 integration of measures §3

{The parentheses on the left side of (∗) are provisional, pending the
status of Γ ; on the right side, they group together the symbols that make
up the universally measurable function on which Λ is acting.}
(ii) Let f > 0 be a universally measurable function on Y . By (i),
Hf is universally measurable on X . Since Λ is a bounded diffusion of X
in Y , it follows from the Scholium that Λ(Hf ) is a universally measurable
function on T and, for every measure µ ∈ D Λ (the domain of Λ ),

hµΛ, Hf i = hµ, Λ(Hf )i ,

that is, (µΛ)• (Hf ) = µ• Λ(Hf ) . Combined with (∗), this shows that
Γ(f ) = Λ(Hf ) is a universally measurable function on T , and

(∗∗) hµ, Γ(f )i = hµ, Λ(Hf )i = hµΛ, Hf i

for every universally measurable function f > 0 on Y and every µ ∈ D Λ .

(iii) If µ is any bounded (positive) measure on T , then the mapping
Γ : t 7→ γt = λt H (t ∈ T) is scalarly essentially µ-integrable.
Given any g ∈ K+ (Y) we are to show that the function Γ(g) is essen-
tially µ-integrable; by (ii), Γ(g)

= Λ(Hg) is universally measurable, so it will
suffice to show that µ• Γ(g) < +∞ (§1, No. 3, Prop. 9). The computation

(Hg)(x) = ηx (g) 6 kηx k kgk (x ∈ X)

shows that 0 6 Hg 6 kHk kgk , therefore

λ∗t (Hg) 6 kHk kgk λ∗t (1) = kHk kgk kλt k 6 kHk kgk kΛk

for all t ∈ T , whence 0 6 Λ(Hg) 6 kHk kgk kΛk . Thus the universally
measurable function Γ(g) = Λ(Hg) is bounded, hence it is integrable for the
bounded measure µ .
Anticipating
R that Λ will prove to be a diffusion, we denote by µΓ the
integral γt dµ(t) of Γ (without relinquishing control to auto-pilot).
(iv) If µ is a bounded measure on T , then (µΛ)H = µΓ .
Here, µΓ is the measure on Y defined in (iii). Since µ is a bounded
measure on T , by the Scholium µ ∈ DΛ and µΛ is a bounded measure
on X , whence µΛ ∈ DH and (µΛ)H is a bounded measure on Y . For
every g ∈ K+ (Y) ,

hµΓ, gi = hµ, Γ(g)i = hµΛ, Hgi = h(µΛ)H, gi

by the definition of µΓ , the equality (∗∗), and the fact that µΛ ∈ D H ; thus
the measures µΓ and (µΛ)H on Y are equal.
§3 integration of positive measures INT V.x84

(v) Γ is a diffusion (hence is a bounded diffusion).

Assuming µ is a measure on T with compact support, we know from
(iii) thatR Γ is scalarly essentially µ-integrable, and we write µΓ for its
integral γt dµ(t) . We are to show that Γ is µ-adequate, and since the
positive measures µ0 6 µ also have compact support, it suffices to show
that Γ is µ-pre-adequate.
Given any lower semi-continuous function h > 0 on Y , we must show
that Γ(h) is µ-measurable (already noted in (ii)) and that


(µΓ)• (h) = µ• Γ(h) ,

that is, hµΓ, hi = hµ, Γ(h)i . Indeed, since µ and µΛ are bounded and h
is universally measurable, one has

hµΓ, hi = h(µΛ)H, hi = hµΛ, Hhi = hµ, Λ(Hh)i = hµ, Γ(h)i

(the first equality by (iv), the second and third by the Scholium, and the
fourth by (∗∗). Thus Γ is a diffusion.
Finally, kγt k = kλt Hk 6 kλt k kHk 6 kΛk kHk < +∞ by the Scholium,
thus the diffusion Γ is bounded.
(vi) If µ ∈ DΛ and µΛ ∈ DH (so that µΛ and (µΛ)H are defined)
then µ ∈ DΓ and µΓ = (µΛ)H .
Since Γ is a diffusion, it suffices by Prop. 11 to show that Γ is scalarly
essentially µ-integrable. Given g ∈ K + (Y) we are to show that Γg is
essentially µ-integrable.
At any rate, Γg and Hg are universally measurable (Scholium) and

hµ, Γgi = hµ, Λ(Hg)i = hµΛ, Hgi = h(µΛ)H, gi

(the first equality, by (∗∗); the second, by the Scholium applied to Λ ;

the third, because µΛ ∈ DH ), and the last expression on the right is
the result of applying the measure (µΛ)H to g ∈ K + (Y) hence is finite.
Thus hµ, Γgi < +∞ , that is, µ• (Γg) < +∞ , therefore Γg is essentially
µ-integrable (§1, No. 3, Prop. 9).
Thus µ ∈ DΓ . But then, for every g ∈ K+ (Y) one has

hµΓ, gi = hµ, Γgi = h(µΛ)H, gi ,

consequently µΓ = (µΛ)H .
(vii) By definition, ΛH is a notation for Γ ; thus, assuming that µ ∈ D Λ
and µΛ ∈ DH , we know from (vi) that µ ∈ DΛH ; we are to show that the
INT V.x85 integration of measures §3

following equalities hold for every universally measurable function f > 0

on Y :

hµΓ, f i = hµΛ, Hf i = hµ, Γf i ;

(150 )
(µΛ)H = µΓ ; Λ(Hf ) = Γf .

In the first line of the display, the first equality holds by (∗∗), while
hµΓ, f i = hµ, Γf i holds by the Scholium.
In the second line, the first equality holds by (vi), the second by (∗).
(viii) Let Λ : t 7→ λt and H : x 7→ ηx be bounded diffusions of T in X
and of X in Y , respectively. If µ is any bounded positive measure on T
then, by the Scholium, µ ∈ DΛ and µΛ is bounded, hence µΛ ∈ DH . Thus,
viewing Λ and H as mappings

Λ : DΛ → M+ (X) , H : DH → M+ (Y) ,

the domain of the composite mapping ( Λ followed by H ) is nonempty. In

this context, let us write diffusions on the right of elements of their domain,
so that the composite in question is Λ ◦ H (thus µ(Λ ◦ H) = (µΛ)H ). We
have shown in (vi) that

µ ∈ DΛ◦H ⇒ µ ∈ DΓ and µ(Λ ◦ H) = (µΛ)H = µΓ ;

thus, identifying mappings with their graphs, we have the inclusion

Λ ◦ H ⊂ Γ,

that is, Λ ◦ H ⊂ ΛH . The inclusion may be proper. {For example if η x = 0

for all x ∈ X , then Hg = 0 for all g ∈ K+ (Y) , and so

hγt , gi = hλt H, gi = hλt , Hgi = 0 for all t ∈ T ;

thus Γg = 0 for all g ∈ K+ (Y) . If µ is any positive measure on T then,

for every g ∈ K+ (Y) , Γg = 0 is essentially µ-integrable, therefore µ ∈ D Γ
(Prop. 11); thus DΓ = M+ (T) , whereas DΛ◦H ⊂ DΛ , which may be a
proper subset of M+ (T) .}
Thus Γ is an extension of Λ ◦ H , of which, in a sense, it is the ‘com-
pletion’.
V.30, `. −2, −1.
“It follows at once from Prop. 13 that

(Λ1 Λ2 )Λ3 = Λ1 (Λ2 Λ3 ) . ”

§3 integration of positive measures INT V.x86

Say Λ1 : x1 7→ λx1 ∈ M+ (X2 ) for x1 ∈ X1 . Let us write Γ = Λ1 Λ2 ,

a bounded diffusion of X1 in X3 ; for x1 ∈ X1 , λx1 is a bounded measure,
hence belongs to the domain of Λ2 , and, writing γx1 = λx1 Λ2 , one has
Γ : x1 7→ γx1 (x1 ∈ X1 ) . Then
(a) (Λ1 Λ2 )Λ3 = ΓΛ3 : x1 7→ γx1 Λ3 = (λx1 Λ2 )Λ3 (x1 ∈ X1 ) .
On the other hand,
(b) Λ1 (Λ2 Λ3 ) : x1 7→ λx1 (Λ2 Λ3 ) (x1 ∈ X1 ) .
Since (λx1 Λ2 )Λ3 = λx1 (Λ2 Λ3 ) by the equation in the southwest corner
of the array (15), the mappings (a) and (b) are equal.
V.31, `. 4–8.
“ . . . one defines diffusions Λ and H by the formulas
λt = εu(t) , ηx = εv(x) ;
the diffusion Γ = ΛH is then given by
γt = ε(v◦u)(t) . ”
If f ∈ K+ (X) then the function


t 7→ λt (f ) = εu(t) (f ) = f u(t) = (f ◦ u)(t)
is bounded and universally measurable, hence is integrable for every bounded
measure on T , therefore:
(i) Λ is scalarly essentially µ-integrable for every bounded positive mea-
sure µ on T ;
(ii) Λ is ‘vaguely universally measurable’, that is, vaguely measurable
for every measure µ on T ;
(iii) Λ is µ-adequate for every bounded positive measure µ on T ,
by (i), (ii) and Prop. 2, b) of No. 1; in particular,
(iv) Λ is µ-adequate for every positive measure µ on T with compact
support, hence is a diffusion. And kλ t k = 1 for all t , so Λ is a bounded
diffusion of T in X .
Similarly H is a bounded diffusion of X in Y .
Then Γ = ΛH is defined by
Γ : t 7→ λt H = εu(t) H ;
but, for every g ∈ K+ (Y) ,



hεu(t) H, gi = hεu(t) , Hgi = Hg u(t) = ηu(t) (g) = εv(u(t)) (g) ,
thus εu(t) H = ε(v◦u)(t) for all t ∈ T , whence
Γ : t 7→ λt H = εu(t) H = ε(v◦u)(t) .
INT V.x87 integration of measures §4

§4. INTEGRATION OF POSITIVE POINT MEASURES

V.31, `. −15.
R∗ R•

“ f (x) dλt (x) = f (x) dλt (x) = f π(t) g(t) ”


The assertion is that λ∗t (f ) = λ•t (f ) = f π(t) g(t) . Since λt is a
bounded measure, λ•t = λ∗t (§1, No. 2, Cor. 2 of Prop. 7).

If
g(t) = 0 then λt = 0 , so λ∗t (f ) = 0 , and f π(t) g(t) = 0 (even if
f π(t) = +∞ ).
Otherwise, 0 < g(t) < +∞ and λt = g(t)επ(t) , whence


λ∗t (f ) = g(t)ε∗π(t) (f ) = g(t)f π(t)

by item (v) of the Example in the Note for V.27, `. −14, −13.
V.31, `. −13, −12.
“Every function (with values in a topological space) defined on X is
λt -measurable for every t ∈ T . ”
See item (iii) of the Example cited above.
V.31, `. −12, −11.
“Every Rmapping f of X into
a Banach space F is λ t -integrable for all
t ∈ T , and f (x) dλt (x) = f π(t) g(t) . ”
If g(t) = 0 then λt = 0 and it is trivial that f is λt -integrable with
integral 0 . {The gory details are worked out at the end of this note.}
Otherwise 0 < g(t) < +∞ and λt = g(t)επ(t) . We know that f is
measurable for λt ; moreover,


λ∗t (|f |) = g(t)ε∗π(t) (|f |) = g(t) · |f | π(t) < +∞

(the second equality by item (v) of the above-cited Example), therefore f is

λt -integrable (Ch. IV, §5, No. 6, Th. 5).
It remains to show that
Z


f dλt = f π(t) g(t) = g(t) · (f ◦ π)(t) .

Given any continuous linear form u on F , it suffices to show that

Z 

(∗) u f dλt = g(t) · (u ◦ f ) π(t) .
§4 integration of positive point measures INT V.x88

By Th. 1 of Ch. IV, §4, No. 2, u ◦ f is λt -integrable (see also the next Note)
and Z  Z
u f dλt = (u ◦ f ) dλt ;

but Z Z


(u ◦ f ) dλt = g(t) (u ◦ f ) dεπ(t) = g(t)(u ◦ f ) π(t)

(the second equality, by item (iv) of the above-cited Example), whence the
asserted equality.
{Write θ for the zero measure on X . One has θ*(|f |) = 0 ; for, θ(g) = 0
for every g ∈ K+ (X) , therefore θ*(h) = 0 for every lower semi-continuous
function h > 0 (Ch. IV, §1, No. 2, Def. 2), whence θ*(f ) = 0 for every
numerical function f > 0 (loc. cit., No. 3, Def. 3). In particular |f | is
θ-negligible for every function f with values in a Banach space F or in R ,
thus f = 0 almost everywhere for θ , whence f is θ-integrable with inte-
gral 0 (p. IV.33, `. −3).}
V.31, `. −10 to −8.
“Finally, if f is an arbitrary numerical function defined
on X , for f
to be λt -integrable
R it is necessary and
sufficient that f π(t) g(t) be finite,
in which case f (x) dλt (x) = f π(t) g(t) . ”
As shown in item (iv) of
the above-cited Example, R f is integrable


for επ(t) if and only if f π(t) is finite, in which case f dεπ(t) = f π(t) .
Then,

f is λt -integrable ⇔ λt = 0 or f is επ(t) -integrable

⇔ g(t) = 0 or f π(t) is finite


⇔ g(t)f π(t) is finite ,

in which case the asserted equality is evident from the preliminary remark.
V.31, `. −4, −3.

“ 2◦ For every function f ∈ K (X) , the mapping t 7→ f π(t) g(t) is
essentially µ-integrable. ”
Defining λt = g(t)επ(t) for t ∈ T , the validity of 2◦ for every
f ∈ K+ (X) says that the mapping Λ : t 7→ λt ∈ M+ (X) is scalarly es-
sentially µ-integrable (§3, No. 1).
Conversely, if the assertion in 2◦ holds for every f ∈ K+ (X) , then it
holds for every f ∈ K (X) ; for,


(f ◦ π)g = (f + − f − ) ◦ π g = (f + ◦ π)g − (f − ◦ π)g
INT V.x89 integration of measures §4

is then the difference of finite-valued essentially integrable functions, hence

is essentially integrable.
It is important here that g is finite-valued; eventually (No. 4, Remark
below), g may be defined only locally µ-almost everywhere and may have
infinite values.
V.32, `. 9–11.
“In the general case, the set of compact subsets K of T such that the
restrictions of π and g to K are continuous is µ-dense (Ch. IV, §5, No. 10,
Prop. 15) ”


Proof #1: The mapping f (t) = π(t), g(t) of T into X × R is
µ-measurable (Ch. IV, §5, No. 3, Th. 1, with u the identity mapping on
X×R ), and its compositions with the coordinate projection mappings are π
and g ; if K is a compact set in T such that f K is continuous
 (the set of
 
such K is µ-dense by the cited Prop. 15), then π K and g K are continuous.
Proof #2: Writing K for the generic compact subset of T , we know
that the sets
 
Kπ = {K : π K is continuous }, Kg = {K : g K is continuous }

are µ-dense in T by criterion a) of the cited Prop. 15 (with A = T ); it will

clearly suffice to show that the set

K = {K ∩ K0 : K ∈ Kπ and K0 ∈ Kg }

is also µ-dense. To this end, we verify that K satisfies the conditions (PL I ),
(PLII ) and b) of Ch. IV, §5, No. 8, Prop. 12.
(PLI ): If K ∩ K0 ∈ K and B is a closed subset of K ∩ K0 , then B ⊂ K
and B ⊂ K0 , whence B ∈ Kπ ∩ Kg and so B = B ∩ B ∈ K .
(PLII ): Suppose K1 ∩ K01 and K2 ∩ K02 belong to K . Then the set
B = (K1 ∩ K01 ) ∪ (K2 ∩ K02 ) is a closed subset of the set

(K1 ∪ K2 ) ∩ (K01 ∪ K02 ) ∈ K ,

hence B ∈ K by the property (PLI ) just proved.

Incidentally, K = Kπ ∩ Kg . For, K ⊂ Kπ ∩ Kg since K ∩ K0 is a closed
subset of K and of K0 ; whereas if K ∈ Kπ ∩ Kg then K = K ∩ K ∈ K .
b): Let K0 be a compact subset of T , let ε > 0 , and choose K ∈ K π
and K0 ∈ Kg so that K ⊂ K0 , K0 ⊂ K0 and

µ(K0 --- K) 6 ε/2 , µ(K0 --- K0 ) 6 ε/2 ;

§4 integration of positive point measures INT V.x90

then K ∩ K0 ∈
K and K0 --- K ∩ K0 = (K0 --- K) ∪ (K0 --- K0 ) , whence

µ K0 --- K ∩ K0 6 ε .
Remark. From here on, the proof that Λ is µ-adequate is straightfor-
ward.
Example. If π and g are universally measurable, and g is bounded,
then Λ is a bounded diffusion and kΛk = kgk .


For, if f ∈ K (X) then t 7→ g(t)f π(t) is universally measurable and
bounded, hence µ-integrable for every bounded positive measure µ on T .
By the part of Prop. 1 already proved, the mapping Λ : t 7→ λ t = g(t)επ(t) is
µ-adequate for every bounded positive measure µ , hence for every positive
measure with compact support. Thus Λ is a diffusion (§3, No. 5, Def. 3)
and kλt k = g(t) for all t ∈ T .
V.32, `. 18–20. 
“The set K of compact sets K ⊂ S such that g K is continuous and


Λ K is vaguely continuous is µ-dense in S (Ch. IV, §5, No. 10, Prop. 15) ”
It is well to pause over the concept of vague measurability introduced
in §3, No. 1 (whose definition is subtly different from that of vague continu-
ity). To say that Λ is vaguely continuous means that for every f ∈ K (X) ,
the function Λf : t 7→ λt (f ) = hλt , f i is continuous (loc. cit.). Whereas:
To say that the mapping Λ : T → M (X) is vaguely µ-measurable means
that when M (X) is equipped with the vague topology, Λ is µ-measurable
in the sense of Ch. IV, §5, No. 1, Def. 1; this is equivalent to the condition
that
 the set KΛ of all compact sets K in T , such that the restriction
ΛK : K → M (X) is continuous, is µ-dense in T (loc. cit., No. 10, Prop. 15);
and, for a compact set K in T , K ∈ KΛ means that for every f ∈ K(X) ,
the numerical function t 7→ λt (f ) (t ∈ K) is continuous, that is, Λf K is
continuous.
{Caution. It is clear that if Λ is vaguely µ-measurable, then for
every f ∈ K (X) the numerical function Λf is µ-measurable since, in the
foregoing notations, KΛ ⊂ KΛf , so that the µ-density  of KΛ implies that
of KΛf (the set of all compact sets K such that Λf K is continuous; clearly
KΛf satisfies (PLI ) and (PLII ) of Ch. IV, §5, No. 8, Prop. 12, and KΛ ⊂ KΛf
ensures that it satisfies criterion a) of that Proposition). Problem: When is
the converse true?}

Since S is a µ-measurable subset of T and the restriction Λ S has a
(vaguely)
 µ-measurable extension to T (namely Λ itself), it follows that
ΛS is µ-measurable
 (see criterion c00 ) in the Note for IV.79, `. 3,4).

 Similarly, g S is µ-measurable. {Alternate proof: the extension by 0 of
g S to T is the µ-measurable function gϕ S , therefore g S is µ-measurable

by criterion d) of the cited Prop. 15.}
INT V.x91 integration of measures §4

With K denoting the generic compact subset of S , we know from cri-

terion a) of the cited Prop. 15 that the sets
 
KΛ|S = {K : ΛK is continuous }, Kg|S = {K : g K is continuous }

are µ-dense in S . Obviously KΛ|S ∩ Kg|S = K (the set in the assertion). One
proves, as in the preceding Note, that the set

{K ∩ K0 : K ∈ KΛ|S and K0 ∈ Kg|S }

is µ-dense in S and is equal to K (with S playing the role of A in Ch. IV,

§5, No. 8, Prop. 12).
V.32, `. 21, 22. 
“ . . . and this implies the continuity of π K (Ch. III, §1, No. 9, Prop. 13).”
By the cited Prop. 13, the mapping x 7→ ε x (x ∈ X) is a homeomor-
phism of X onto a subspace E of M (X) , whence the continuity of the
composite
1
t 7→ λt 7→ λt = επ(t) 7→ π(t) (t ∈ K) ,
g(t)
where the last mapping is the restriction of the inverse homeomorphism
E → X to the subspace {επ(t) : t ∈ K } of E .
V.32, `. −14 to −9.
“Lemma.”
The following graphics help to keep track of the numerous players in
the otherwise elementary proof.
(
inf{g(t) : π(t) = x } when x ∈ π(T)
f (x) =
+∞ when x ∈ X --- π(T) .

T π-X

g f
? ?
R R

The range of π determines a partitioning of T into the sets of constancy

of π : that is, the relation R t, t0 ⇔ π(t) = π(t0 ) defines a quotient set T/R
whose elements are indexed by the elements of the range of π , thus there is
a bijection π(T) → T/R , namely
−1
x 7→ π ({x}) (x ∈ π(T)) .
§4 integration of positive point measures INT V.x92

Then g defines a numerical function g R : T/R → R that assigns to an

equivalence class the infimum of the restriction of g to that class:

−1  −1
gR : π ({x}) 7→ inf g  π ({x}) (x ∈ π(T)) ;

the definition of the function f : X → R then takes the form

( −1

gR π ({x}) when x ∈ π(T)
f (x) =
+∞ when x ∈ X --- π(T) .

V.32, `. −1.
“ . . . and nonempty ”
−1
Since f (x) 6 a < +∞ , π (x) is nonempty by the definition of f .
V.32, `. −1 to V.33, `. 1.
−1
“ . . . there exists a t ∈ π (x) such that . . . ”
Recall that the restriction of g to any subspace S of T is lower semi-
continuous, since {t ∈ S : g(t) 6 a } = S ∩ {t ∈ T : g(t) 6 a } .
V.33, `. 10, 11.
“For every numerical function f > 0 defined on X ,
Z • Z •

(1) f (x) dν(x) = f π(t) g(t) dµ(t) . ”

We know that the mapping Λ : t 7→ λt = g(t)επ(t) is µ-adequate (No. 1,

Prop. 1), thus (1) is known to hold for every lower semi-continuous function
f > 0 (§3, No. 1, Def. 1); the miracle of (π, g) is that (1) holds for every
function f > 0 .
V.33, `. 13–15. R
“By formula (4) of §3, No. 1, ν • (1) = K g(t) dµ(t) < +∞ , so that all
of the measures that figure in formula (1) are bounded.”
Since 1 is lower semi-continuous, by the cited formula (4) one has
Z • Z •
•
(∗) ν (1) = dµ(t) 1 dλt .

The crux of the matter is the calculation

Z •
1 dλt = λ•t (1) = λ∗t (1) = kλt k = g(t)kεπ(t) k = g(t) < +∞
INT V.x93 integration of measures §4

(the second equality by §1, No. 1, Prop. 4; or by the fact that λ t is bounded,
hence moderated, cf. §1, No. 2, Prop. 7). Thus (∗) may be written ν*(1) =
µ• (g) , or, since µ is bounded,

ν*(1) = µ*(g) .

Next, we note that g is µ-integrable. Since µ has support K , ϕ K = 1

µ-almost everywhere, therefore g = ϕ K g µ-almost everywhere, so it suffices
to show that ϕK g is µ-integrable. Indeed, ϕK g is µ-measurable and is
bounded (because g K is continuous), hence it is integrable with respect to
the bounded measure µ ; therefore g is µ-integrable and
Z Z Z
µ*(g) = g dµ = ϕK g dµ = g dµ
K

(the last equality is merely a notation). Finally

Z
ν*(1) = µ*(g) = g dµ < +∞ ,

R
thus ν is bounded and kνk = g dµ .
V.33, `. −12.
“In view of formula (6) of §3, No. 2 . . . ”
The context here is the µ-adequate mapping Λ : t 7→ λ t = g(t)επ(t) , for
which the cited formula (6) yields
Z ∗ Z • Z ∗
f (x) dν(x) > dµ(t) f (x) dλt (x) ,

R∗

where f (x) dλt (x) = f π(t) g(t) (V.31, `. −15) and µ• = µ* (preceding
Note), thus Z ∗ Z ∗


f (x) dν(x) > f π(t) g(t) dµ(t) .

V.33, `. −5, −4.

“ . . . let u be the function (h + ε)/g , which is lower semi-continuous
in K . ”
The function u : T → R+ is defined by

 h(t) + ε when g(t) > 0
u(t) = g(t)

+∞ when g(t) = 0 .
§4 integration of positive point measures INT V.x94

The assertion is that uK is lower semi-continuous; thus, given any a ∈ R ,
we are to show that the set

Ka = {t ∈ K : u(t) 6 a }

is a closed set in K . From the definition of u , it is clear that

h(t) + ε
Ka = {t ∈ K : g(t) > 0 and 6 a}.
g(t)

If a < 0 then Ka = ∅ because u > 0 . If a = 0 then

h(t) + ε
K0 = {t ∈ K : g(t) > 0 and 6 0} = ∅
g(t)

because the fraction is > 0 . So we can suppose that a > 0 . Then

Ka = {t ∈ K : g(t) > 0 and h(t) + ε 6 ag(t) }

= {t ∈ K : h(t) + ε 6 ag(t) }
= {t ∈ K : h(t) + ε − ag(t) 6 0 } ;
 
since hK is lower semi-continuous, and (ε − ag) K is finite and continuous,
their sum is lower semi-continuous (GT, IV, §6, No. 2, Prop. 2), therefore
Ka is closed.
V.34, `. 1, 2.
−1
“ for every x ∈ X let v(x) be the infimum of u(t) for t ∈ π ({x})∩K . ”
−1 
Note that π
({x}) ∩ K = (π K)−1 ({x}) , thus v(x) = +∞ when
/ π(K) = π K (K) since (π K)−1 ({x}) = ∅ and inf = +∞ by conven-
x∈
t∈∅
tion.
V.34, `. 3, 4.
“ . . . it is lower semi-continuous on X by the Lemma (applied to the
restriction of π to K ) ”
The measure µ has compact support K , but it is a measure on T . The
Lemma is purely topological, thus we are concerned here exclusively with
topology. The picture is
π|K-
K X

u|K v
? ?
R R
INT V.x95 integration of measures §4

The meaning of the parenthetical remark is  that, in the Lemma,

K plays the role of T , π K plays the role of π , uK plays the role of g ,
and v plays the role of f ; thus the recipe for v is the one prescribed for f
in the Lemma. Since u is lower semi-continuous, so is itsrestriction to K .
Finally, it must be noted that the continuous mapping π K is proper; this
follows from the compactness of K (GT, I, §10, No. 2, Cor. 2 of Th. 1).
Thus the hypotheses of the Lemma are fulfilled.
V.34, `. 4, 5.

“ . . . and v π(t) g(t) 6 h(t) + ε for all t ∈ K (recall that the first
member is zero by convention if g(t) = 0 ).”
Let t ∈ K . By the parenthetical remark, we can suppose that g(t) > 0 .
Set x = π(t) . By definition,

−1
v(x) = inf{u(s) : s ∈ π ({x}) ∩ K } 6 u(t)

−1
because t ∈ π ({x}) ; then


v (π(t) g(t) = v(x)g(t) 6 u(t)g(t) = h(t) + ε

by the definition of u .
V.34, `. 7–9.

Z ∗ Z ∗
f (x) dν(x) 6 v(x) dν(x)
Z ∗ Z ∗



(4) = v π(t) g(t) dµ(t) 6 h(t) + ε dµ(t)
K
Z ∗
= h(t) dµ(t) + εµ(1) .

{Incidentally, the desired inequality (3) is trivial if µ*(h) = +∞ , so we

can suppose (to reduce the stress level) that h is µ-integrable.}
The first inequality follows from v > f , and the last equality, from
ϕK = 1 µ-almost everywhere (because Supp µ ⊂ K ). The second inequality
results from


v (π(t) g(t) 6 h(t) + ε for t ∈ K ,

equivalently (v ◦ π)gϕK 6 (h + ε)ϕK everywhere on T , and the fact that

(v ◦ π)g = (v ◦ π)gϕK µ-almost everywhere.
§4 integration of positive point measures INT V.x96

It remains to verify the first equality. Since the mapping Λ : t 7→ λ t

is µ-adequate (No. 1, Prop. 1) and v > 0 is lower semi-continuous on X ,
formula (4) of §3, No. 1 is applicable and yields
Z • Z • Z •
v(x) dν(x) = dµ(t) v(x) dλt (x) ,

that is, ν • (v) = µ• (v ◦ π)g (p. V.31, `. −15). Since v is lower semi-


continuous and µ is bounded, this may be written ν*(v) = µ* (v ◦ π)g ,
which is the asserted equality.
V.34, `. 13, 14.
“ . . . the set K of compact subsets K of T such that the restrictions of
π and g to K are continuous is µ-dense (Ch. IV, §5, No. 10, Prop. 15).”
See also the Note for V.32, `. 9–11.
V.34, `. 16, 17.
“ . . . the pair (π, g) being µα -adapted for every α ∈ A ”
If (π, g) is µ-adapted, then it is µ0 -adapted for every positive measure
0
µ 6 µ:
1◦ π and g are µ0 -measurable (because µ-negligible sets are µ 0 -negligi-
ble);
2◦ every essentially µ-integrable function is essentially µ 0 -integrable; in-
1 1 1
deed, writing µ00 = µ − µ0 , one has L (µ) = L (µ0 ) ∩ L (µ00 ) (see Corol-
lary 3 in the Note for V.10, `. 13, 14).
V.34, `. −8 to −6.
“Corollary.”
We propose to apply Th. 1 to the function f = ϕ N . Since ϕN ◦ π =
ϕπ−1 (N) , by Th. 1 one has



ν • (N) = ν • (ϕN ) = µ• (ϕN ◦ π)g = µ• ϕπ−1 (N) g ;

thus N is locally negligible for ν if and only if the set

−1
{t ∈ T : ϕπ−1 (N) g (t) 6= 0 } = π (N) ∩ {t ∈ T : g(t) 6= 0 }

is locally negligible for µ (§1, No. 1).

V.35, `. 3, 4.
“ . . . for every function ψ ∈ K (X) , ψ ◦ π is continuous with compact
support, since π is proper ”
INT V.x97 integration of measures §4


Let K be a compact set such that ψ = 0 on X --- K . If ψ π(t) =6 0
then π(t) ∈ K , thus

−1
{t : (ψ ◦ π)(t) 6= 0 } ⊂ π (K) ;

−1
since π is proper, π (K) is compact (GT, I, §10, No. 2, Prop. 6), therefore

−1
Supp(ψ ◦ π) = {t : (ψ ◦ π)(t) 6= 0} ⊂ π (K) ,

whence ψ ◦ π ∈ K (T) .
V.35, `. 4, 5.
“ . . . the pair (π, g) is therefore µ-adapted ”
The functions π and g are universally measurable and, for every
ψ ∈ K (X) , the function (ψ ◦ π)g is continuous with compact support,
hence is integrable with respect to every measure on T . Thus (π, g) is, so
to speak, ‘universally adapted’, that is, µ-adapted for every positive measure
µ on T .
Remark. It then follows from No. 1, Prop. 1 that the mapping
Λ : t 7→ λt = g(t)επ(t) is µ-adequate for every positive measure µ on T ,
in particular for every positive measure with compact support, thus Λ is a
diffusion of T in X (§3, No. 5, Def. 3) whose domain is all of M + (T) .
See also the Example in the Note for V.32, `. 9–11.
V.35, `. 5, 6.
“ . . . the mapping t 7→ g(t)επ(t) is vaguely continuous.”


For every ψ ∈ K (X) , the mapping t 7→ g(t)ε π(t) (ψ) = ψ (π(t) g(t) is
continuous (with compact support), whence the assertion (§3, No. 1).
Thus, for the diffusion Λ : t 7→ λt = g(t)επ(t) (see the preceding Note),
Λψ = (ψ ◦ π)g ∈ K+ (T) for all ψ ∈ K+ (X) .
V.35, `. −10.
“ 1◦ f(x) > f (x) for all x ∈ X (since g(t) > 0 for all t ∈ T ).”
If x ∈
/ π(T) , then f (x) = +∞ and the inequality is trivial. If x ∈ π(T)
−1 −1
then f(x) is defined as an infimum over π (x) ; now, for every t ∈ π (x)
one has

h(t)
f (x) = f π(t) 6
g(t)
−1
by the choice of h , and taking the infimum over t ∈ π (x) yields f (x) 6
f(x) .
§4 integration of positive point measures INT V.x98

V.35, `. −9.

“ 2◦ f π(t) g(t) 6 h(t) for all t ∈ T .”


Fix t ∈ T and set x = π(t) . Thus f π(t) = f(x) is defined as an
−1 −1
infimum over π (x) . Now, π (x) is compact since π is proper (GT, I, §10,
−1
No. 2, Th. 1), and h/g is lower semi-continuous on T , hence on π (x) ,
−1
therefore h/g attains its infimum on π (x) (GT, IV, §6, No. 2, Th. 3).
−1
Thus there exists a point t0 ∈ π (x) such that

h(t0 ) h(s) −1
f(x) = 6 for all s ∈ π (x) ,
g(t0 ) g(s)


and in particular f(x) 6 h(t)/g(t) , whence f π(t) g(t) 6 h(t) .
V.35, `. −2 to V.36, `. 4.
“ Proposition 3. ”
The main players:

T π-X

g f
? ?
R+ G

V.36, `. 7.
“ . . . is µ-dense in S (Ch. IV, §5, No. 10, Prop. 15).”
Since π is µ-measurable, its restriction to the µ-measurable
 set S is
µ-measurable in the sense of loc. cit., Def. 8 (because π S has an extension
to T that is µ-measurable, namely π itself; see the condition c 00 ) in the
Note for IV.79, `. 3, 4). Therefore K is µ-dense in S by the equivalent
condition a) of the cited Prop. 15.
V.36, `. 8–10.
“ . . . it therefore suffices to prove that for every K ∈ K , the set of com-
pact subsets H of K , such that the restriction of f ◦ π to H is continuous,
is µ-dense in K (Ch. IV, §5, No. 8, Prop. 13).”

Write H for the set of all compact subsets H of S such that (f ◦π) H is
continuous. One sees easily that H satisfies the conditions (PL I ) and (PLII )
of loc. cit., Prop. 12, that is: if H ∈ H then every closed subset of H belongs
to H ; and if H1 , H2 ∈ H then H1 ∪ H2 ∈ H (GT, I, §3, No. 2, Prop. 4).
Thus if, for every K ∈ K , the set of all H ∈ H such that H ⊂ K is
shown to be µ-dense in K , it will follow from the cited Prop. 13 that H
INT V.x99 integration of measures §4

is µ-dense in S , and this will establish that (f ◦ π) S is µ-measurable by
criterion a) of Ch. IV, §5, No. 10, Prop. 15.
Review of the cited Prop. 13: With the notations K and H as in that
Prop., given K ∈ K note that the set of H ∈ H such that H ⊂ K is the
trace H ∩ K of H on K , that is,

{H ∈ H : H ⊂ K } = {H ∩ K : H ∈ H } .

For, the inclusion ⊂ is immediate from H = H ∩ K ; whereas if H ∈ H then

H ∩ K ∈ H (because H satisfies (PLI )), whence the inclusion ⊃ .
The message of Prop. 13: If K and H are sets of compact subsets of the
µ-measurable set A , such that (i) K is µ-dense in A , (ii) H ∩ K is µ-dense
in K for every K ∈ K , then H is also µ-dense in A .
(A kind of ‘transitivity’ of µ-denseness.)
V.36, `. 14, 15.
−1
“ K ∩ π (N) is µ-negligible by virtue of the Cor. of Th. 1 of No. 2 ”
−1
By the cited Cor., the set S ∩ π (N) is locally µ-negligible, therefore
−1
K ∩ S ∩ π (N) is µ-negligible (Ch. IV, §5, No. 2, Prop. 5); and K ⊂ S .
V.36, `. 15.
−1
“ . . . the sets K ∩ π (Cn ) are compact ”
−1  
For, K ∩ π (Cn ) = (π K)−1 (Cn ) and π K is continuous.
V.36, `. 16.
“ . . . the restriction of f ◦ π to each of the latter sets is continuous ”
  −1 −1
Since π K is continuous, so is π K ∩ π (Cn ) . Thus π maps K ∩ π (Cn )
continuously into Cn ; but f Cn is continuous, therefore f ◦ π maps
−1
K ∩ π (Cn ) continuously into G .
V.36, `. 17.
“ . . . the restriction of f ◦ π to S is µ-measurable.”
As in the Note for `. 8–10, write H for the set of all compact sets
H ⊂ S such that (f ◦ π)H is continuous. We know that H satisfies 
(PLI ) and (PLII ) of Ch. IV, §5, No. 8, Prop. 12; to show that (f ◦ π) S is
µ-measurable, it will suffice to show that H is µ-dense in S (loc. cit., No. 10,
Prop. 15, criterion a)).
We know that the set K of all compact subsets of S on which π is
continuous is µ-dense in S . To show that H is µ-dense in S , it will suffice
to show that for every K0 ∈ K , the trace of H on K0 , namely, the set

HK0 = H ∩ K0 = {H ∈ H : H ⊂ K0 } ,
§4 integration of positive point measures INT V.x100

is µ-dense in K0 (loc. cit., No. 8, Prop. 13, with T playing the role of X
in Prop. 13, and S the role of A ). Since H satisfies (PL I ) and (PLII ), so
does HK0 .
Let K0 ∈ K . To show that HK0 is µ-dense in K0 , it will suffice to show
that HK0 satisfies the criterion c) of loc. cit., Prop. 12 (with T playing the
role of X , K0 the role of A , and HK0 the role K ). Thus, let K be any
compact subset of K0 . Then K ∈ K . Apply to K the arguments of the first
part of the proof to produce a partition
−1
[ −1
K = K ∩ π (N) ∪ K ∩ π (Cn ) ,
n
−1 −1
where K ∩ π (N) is µ-negligible (`. 14, 15) and K ∩ π (Cn ) ∈ H for all n
−1 −1
(`. 16); since K ∩ π (Cn ) ⊂ K ⊂ K0 , one has K ∩ π (Cn ) ∈ HK0 , and the
criterion c) is verified.
V.36, `. −17.
“ . . . let us show that N is locally ν-negligible.”
It will then follow from criterion a) of Ch. IV, §5, No. 8, Prop. 12 that
L is µ-dense (in X ).
V.36, `. −14.
“ . . . is by hypothesis µ-dense in S (Ch. IV, §5, No. 10, Prop. 15).”
The argument for this is given in the Note for V.32, `. 9–11.
V.36, `. −13.
−1
“It therefore suffices to prove that π (N) ∩ H is µ-negligible for every
H ∈ H .”
By criterion a) of Ch. IV, §5, No. 8, Prop. 12.
V.36, `. −12 to −10.
“ π(H) is compact and may be identified with the quotient space of H
by the equivalence relation π(t) = π(t 0 ) , π being identified with the canon-
ical mapping of H onto this quotient space (GT, I, §5, No. 2, Prop. 3).”

Since π H is a continuous mapping of the compact space H into the
Hausdorff space X , it is a closed mapping hence is eligible for application
of the cited Prop. 3. One has the diagram
p- h - π(H) i - X.
H H/R
0
where R is the equivalence relation in H defined  by π(t) = π(t ) . The
elements of H/R are the sets of constancy of π H . Let us write ṫ for the
equivalence class of t ∈ H , that is,
−1

−1

ṫ = H ∩ π π(t) = π H π(t) (t ∈ H) ;
INT V.x101 integration of measures §4

thus p(t) = ṫ for t ∈ H , h(ṫ) = π(t) for ṫ ∈ H/R , and i is the canonical
injection of π(H) into X .
By definition, H/R bears the quotient topology, that is, the final topol-
ogy for the mapping p , and the topology of the subspace π(H) is the initial
topology for the mapping i . By the cited Prop. 3, h is a homeomorphism,
therefore the topology on π(H) is also the final topology for h ; by transi-
tivity (GT, I, §2, No. 4, Prop. 7) the topology on π(H) is the
 final topology
for the mapping h ◦ p : H → π(H) , which is the mapping π H : H → π(H) .
Consequence: a mapping
 π(H) → G will be continuous if and only if

its composition with π H is continuous (loc. cit., Prop. 6).
V.36, `. −10 to −8.
“Since the restriction of f ◦ π to H is continuous, the restriction of f
to π(H) is therefore continuous ”
 

  
The continuity of (f ◦ π) H = f π(H) ◦ (π H) implies the continuity
of f π(H) by the preceding Note.
V.36, `. −2 to V.37, `. 2.
“Remark.”
One is contemplating Prop. 3 with G = F (a Banach space).

Suppose (f ◦π)S is µ-measurable. This is equivalent to the measurabil- 
ity of the function h = (f ◦π)ϕS , because h is the extension by 0 of (f ◦π) S
to T (Ch. IV, §5, No. 10, Prop. 15, criterion d)). Then hg is also measur-
able on T (loc. cit., No. 3, Cor. 5 of Th. 1). Since S = {t ∈ T : g(t) 6= 0 }
one has g = ϕS g , therefore (f ◦ π)g = (f ◦ π)ϕS g = hg is measurable. 
Conversely, assume that (f ◦ π)g is measurable. The function g S is
measurable (its extension by 0 to T is the measurable function g ) and takes
its values in ]0, +∞[ ; composing it with the continuous function u(a) = 1/a
on ]0, +∞[ , it follows that the function

1
 : t 7→ 1/g(t) (t ∈ S)
g S

is measurable by loc. cit., No. 3, Th. 1 as generalized to measurability on S

(see item No.3, Th. 1 0 in the Note for IV.80, `. −17 to −14). Let k be the
1
extension by 0 of  to T ; since k is measurable on T , so is (f ◦ π)gk ,
gS

therefore the function 
 (f ◦ π)gk S is measurable. In other words, since
gk= 1 on S , (f ◦ π)S is measurable.
Note: The argument is valid for the case that G = R ; in particular,
the equality of h and the extension by 0 of (f ◦ π) S to T is assured by
the convention (±∞) · 0 = 0 .
§4 integration of positive point measures INT V.x102

V.37, `. 15, 16.

“The measures µ and ν are then bounded ”
Let K = Supp µ . Then ϕK = 1 µ-almost everywhere.
Boundedness of µ : kµk = µ*(1) = µ*(ϕK ) = µ*(K) < +∞ .
•


Boundedness of ν : For every h ∈ K+ (X)

, one has ν(h) = µ (h◦π)g ,
which can be written ν(h) = µ* (h ◦ π)g since µ is bounded (§1, No. 2,
Prop. 7). By assumption, g K is bounded, equivalently, gϕK is bounded.
Thus
(h ◦ π)g = (h ◦ π)gϕK µ-almost everywhere,

where h ◦ π and gϕK are bounded and µ-measurable; the bounded

µ-measurable function (h ◦ π)gϕK is integrable for the bounded measure µ :



µ* (h ◦ π)g = µ* (h ◦ π)gϕK 6 khk kgϕK k µ*(1) < +∞ ,
 
whence ν(h) 6 khk kg  Kk kµk . Thus ν is bounded and kνk 6 kg Kk kµk .
V.37, `. 17, 18.

“ . . . the function f π(t) g(t) is then µ-integrable, and the relation (9)
is verified, by Th. 1 of §3, No. 3.”
By No. 1, Prop. 1, the mapping Λ : t 7→ λt = g(t)επ(t) is µ-adequate,
hence part a) of the cited Th. 1 is applicable.
V.37, `. 19, 20.
“ f is then ν-measurable (No. 3, Prop. 3 and Remark ) ”
Since (f ◦ π)g is µ-integrable, it is µ-measurable; it follows from the
cited Remark (or its analog for G = R ) that (f ◦ π) S is µ-measurable,
therefore f is ν-measurable by Prop. 3.
V.37, `. −5, −4.
“The pair (g, π) is obviously µα -adapted for every α ∈ A ”
See the Note for V.34, `. 16, 17 .
V.37, `. −2, −1.
“Since the argument of A) may be applied to the measures µ α , να , the
first part of the statement then follows from Prop. 3 of §2, No. 2.”
Suppose f is essentially ν-integrable. By the cited Prop. 3 (and the
boundedness of the να ), f is να -integrable for every α , and
Z XZ
(i) f dν = f dνα (absolutely summable) .
α∈A
INT V.x103 integration of measures §4

By Part A), the function h = (f ◦ π)g is µ α -integrable for every α , and

Z Z
(ii) f dνα = h dµα .

But |f | is also essentially ν-integrable, and (|f | ◦ π)g = |h| , therefore |h| is
µα -integrable and
Z • Z •
(ii• ) |f | dνα = |h| dµα .

The equality
Z • XZ •
•
(i ) |f | dν = |f | dνα
α∈A

holds by §2, No. 2, Prop. 1 (with no restrictions on f ). Now,

XZ • XZ Z •
|h| dµα = |f | dνα = |f | dν < +∞
α∈A α∈A

by (ii• ), (i• ) and the essential ν-integrability of f ; it then follows from the
cited Prop. 3 that h is essentially µ-integrable and
Z XZ
(iii) h dµ = h dµα (absolutely summable) .
α∈A

Combining (i), (ii) and (iii), we see that

Z XZ XZ Z
f dν = f dνα = h dµα = h dµ ,
α∈A α∈A

thus (9) is verified

Conversely, suppose h = (f ◦ π)g is essentially µ-integrable. For every
α ∈ A , h is µα -integrable, hence f is να -integrable by Part A). It follows
that f is ν-measurable (§2, No. 2, Prop. 2); and, since |h| is also essentially
µα -integrable, the equality (ii• ) holds for every α . Then
Z • XZ • XZ • Z •
|f | dν = |f | dνα = |h| dµα = |h| dµ < +∞
α∈A α∈A

(the 1st and 3rd equality, by §2, No. 2, Prop. 1, the 2nd equality by (ii • )),
therefore f is essentially ν-integrable, and the relation (9) holds by the first
part of the argument.
§4 integration of positive point measures INT V.x104

V.38, `. 4, 5.
“The second part of the statement therefore follows from the first part
and Proposition 2.”
By the cited Prop. 2 and

the inequality (1) of §1, No. 1, we have the
diagram where h = (f ◦ π)g :

ν*(|f |) > ν • (|f |)

..
k .

µ*(|h|) > µ• (|h|)

If f is ν-integrable, then the (equal) elements of the first column are

finite, f is ν-measurable, and ν • (|f |) < +∞ , therefore f is essentially
ν-integrable (§1, No. 3, Prop. 9); whence, by the first part of the statement,
h is µ-measurable (and more), and since µ*(|h|) = ν*(|f |) < +∞ , h is
µ-integrable (Ch. IV, §5, No. 6, Th. 5). The argument is symmetric in f
and h : If h is µ-integrable, then the (equal) elements etc.
In this case, we have equality in the last column (by the first part of the
statement) and, since integrable functions are moderated (§1, No. 3, Cor. of
Prop. 9), equality in the top row and in the bottom row (loc. cit., No. 2,
Prop. 7), thus all four corners are equal.
V.38, `. −15 to `. −13.
“The statements of Ths. 1 and 2 and of Prop. 3 remain valid when π
and g are only assumed to be defined locally almost everywhere.”
With the notations (π, g) and (π 0 , g 0 ) as in the immediately preceding
text ( π 0 and g 0 defined only locally µ-almost everywhere), write
Z Z
ν= λt dµ(t) = g(t)επ(t) dµ(t) .

We know that for f ∈ K (X) , the function (f ◦π)g is essentially µ-integrable

and
(f ◦ π 0 )g 0 = (f ◦ π)g locally µ-almost everywhere,

where (f ◦ π 0 )g 0 is only known to be defined locally µ-almost everywhere.

By definition (§1, No. 3, Def. 3) there exists a function h ∈ L 1 (µ) such
that (f ◦ π)g = h locally µ-almost everywhere, and one writes
Z Z
(f ◦ π)g dµ = h dµ ;
INT V.x105 integration of measures §4

then also (f ◦ π 0 )g 0 = h locally µ-almost everywhere, therefore, by the same

Def. 3, (f ◦ π 0 )g 0 is said to be essentially µ-integrable and one writes
Z Z Z
(f ◦ π 0 )g 0 dµ = h dµ = (f ◦ π)g dµ .
R 0
This motivates defining the symbol g (t)επ0 (t) dµ(t) to be equal to the
0 0
measure ν , with λt = g (t)επ0 (t) defined only locally µ-almost everywhere
(cf. V.25, `. −6, −5).
0 0
Re Theorem 1 : Assume R 0 f > 0 on T . The left •side of (1) for (π , g )
is, by the definition of g (t)επ0 (t) dµ(t) , equal to ν (f ) , thus is the
same
as for (π, g) . On the right side of (1), for (π, g) we have µ • (f ◦ π)g . The
function (f ◦ π 0 )g 0 , defined only locally µ-almost everywhere, satisfies
(f ◦ π 0 )g 0 = (f ◦ π)g locally µ-almost everywhere ;
by the extension of the notation µ• to functions > 0 defined only locally
µ-almost everywhere, one therefore has



µ• (f ◦ π 0 )g 0 = µ• (f ◦ π)g ,
that is, the right side of (1) for (π 0 , g 0 ) is by definition equal to the right
side for (π, g) . Thus the extension of Th. 1 to (π 0 , g 0 ) is simply a matter of
notation.
Re the first part of the statement of Theorem 2 : Simply a matter of
notation.
Re the second part of the statement of Theorem 2 : The terms “contin-
uous” and “proper” are not appropriate for functions defined only locally
almost everywhere. With (π, g) as in the statement, one assumes that π 0
and g 0 are functions, defined locally µ-almost everywhere in T , such that
π 0 = π and g 0 = g locally µ-almost everywhere. The rest is simply notation.
{For the record: “proper” is defined only for continuous mappings (GT,
I, §10, No. 1, Def. 1). Thus, the expression “continuous proper function”
would be redundant; but the expression “proper continuous function” is not,
since not every continuous function is proper.}
Re Proposition 3 : The new element here is S . With (π 0 , g 0 ) and (π, g)
related as above, suppose (π, g) satisfies the hypotheses of Prop. 3. Let
S0 = {t ∈ T : g 0 (t) is defined and 0 < g 0 (t) < +∞ } .
Since g = g 0 locally µ-almost everywhere, S and S 0 differ at most by a
locally µ-negligible set, thus ϕS = ϕS0 locally µ-almost
 everywhere
 in T ,
whence S0 is also µ-measurable. Moreover, (f ◦π 0 )S0 and (f ◦π)S are equal
locally µ-almost everywhere; since their extensions by 0 to T are then equal
0  0
locally µ-almost
 everywhere, it follows that (f ◦π ) S is µ-measurable if and
only if (f ◦ π)S is µ-measurable (Ch. IV, §5, No. 10, Prop. 15, criterion d)).
§5 measures defined by numerical densities INT V.x106

§5. MEASURES DEFINED BY NUMERICAL DENSITIES

V.38, `. −10 to −3.

“Proposition 1.”
There is a problem here: each of the conditions a), b), c) implies that
the function g is defined almost everywhere (not merely locally almost ev-
erywhere).
To set the stage, we note that, for a mapping g defined locally
µ-almost everywhere in T and taking values in a topological space F , the
domain of g , being the complement of a locally negligible set, is measurable
(Ch. IV, §5, No. 2, sentence following Def. 3); the following conditions are
therefore equivalent:
(i) g is measurable in the sense of loc. cit., No. 10, Def. 8;
(ii) g has a measurable extension to T (see the condition c 00 ) in the
Note for IV.79, `. 3, 4);
(iii) every extension of g to T is measurable (Ch. IV, §5, No. 2,
Prop. 6).
Re a): The assertion that gϕV is integrable means that the function
gϕV (hence also g ) is defined almost everywhere, and is equal almost ev-
erywhere to an everywhere-defined integrable function (Ch. IV, §4, No. 1,
last paragraph).
R∗
Re b): For the symbol |g|ϕK dµ to be defined, it is necessary that
the function |g|ϕK (hence also g ) be defined almost everywhere (Ch. IV,
§3, No. 2, next-to-last paragraph).
Re c): Similar to a).
What is the problem? In Def. 1 below, a function g is defined to be
locally integrable if it satisfies the conditions of Prop. 1. It follows from the
foregoing that if f is a function defined locally almost everywhere but not
almost everywhere, then f cannot be locally integrable—even if it is equal
locally almost everywhere to a locally integrable function g —contrary to
the assertion immediately following Def. 1.
Proposed remedy: In the statement of Prop. 1, replace ‘integrable’
R∗ R•
by ‘essentially integrable’, and by R.
•
{So far, the essential upper integral has only been applied to every-
where-defined positive functions (§1, No. 1). For a function f defined lo-
cally almost everywhere in T and > 0 locally almost everywhere, define
INT V.x107 integration of measures §5
R• R•
f dµ = f * dµ , where f * is any function, defined and > 0 at every
point of T , such that f = f * locally almost everywhere; for example, let
f 0 be the extension by 0 of f to T and let f * = f 0 ϕ { N , where N is the
set (locally negligible) of points t where f (t) is defined and < 0 .}
Prop. 1 then takes the form:
Proposition 1 0 . — Let g be a function defined locally almost every-
where in T (for the positive measure µ ), with values in a Banach space F
(resp. in R ). The following properties are equivalent:
a0 ) For every point t ∈ T , there exists a neighborhood V of t such that
the function gϕV is essentially µ-integrable.
0
R • b ) The function g is µ-measurable and, for every compact set K ⊂ T ,
|g|ϕK dµ < +∞ .
c0 ) For every numerical function h ∈ K (T) , gh is essentially
µ-integrable.
Proof. a0 ) ⇒ b0 ): Assume that g satisfies a0 ), and let g0 be its extension
by 0 to T . To say that gϕV is essentially integrable (with respect to µ )
means that there exists a function f ∈ L F1 (resp. f ∈ LR1 when the values
of g are in R ) such that gϕV = f locally almost everywhere (§1, No. 3,
first paragraph on INT V.9). Since g = g 0 locally almost everywhere, one
has g0 ϕV = f locally almost everywhere, therefore g 0 ϕV is also essentially
integrable, and in particular measurable (Ch. IV, §5, No. 2, Prop. 6).
To show that g is measurable, it suffices to show that g 0 is measurable.
For every t ∈ T , by assumption there exists a neighborhood V t of t such
that gϕVt is essentially integrable, and so g 0 ϕVt is essentially integrable,
say g0 ϕVt = ft locally almost everywhere, where f t ∈ LF1 (resp. ft ∈ LR1 ).
If Wt is a compact neighborhood of t such that W t ⊂ Vt , then
g0 ϕWt = (g0 ϕVt )ϕWt = ft ϕWt
locally almost everywhere, where f t ϕWt is integrable (Ch. IV, §5, No. 6,
Cor. 3 of Th. 5); replacing Vt by Wt , we can suppose that Vt is integrable,
hence moderated (§1, No. 2, Def. 2). Moreover g 0 ϕVt is integrable, since it
is essentially integrable and moderated (§1, No. 3, Cor. of Prop. 9). Define
gt = g0 ϕVt . For every t ∈ T , gt is a measurable function such that g 0 = gt
everywhere in Vt , consequently g0 is measurable by the cited Principle of
localization (Ch. IV, §5, No. 2, Prop. 4).
If, moreover, K is any compact set in T , then, by the argument of
R∗ 0
the text, the measurable function g 0 ϕK satisfies |g |ϕK dµ < +∞ , hence
g0 ϕK is integrable; since gϕK = g0 ϕK locally almost R • everywhere, gϕK is
essentially integrable and (by the convention about proposed above)
Z • Z • Z ∗
|g|ϕK dµ = |g0 |ϕK dµ = |g0 |ϕK dµ < +∞
§5 measures defined by numerical densities INT V.x108

(the second equality by §1, No. 2, Prop. 7, 2)).

b0 ) ⇒ c0 ): Assume that g satisfies b0 ). Since g is measurable, its
extension g0 to T by 0 is measurable. As noted in the preceding paragraph,
R∗ 0
g0 ϕK is integrable for every compact set K in T , hence |g |ϕK dµ < +∞ .
By the argument in the text, if h ∈ K (T) then g 0 h is integrable, and
since gh = g0 h locally almost everywhere, it follows that gh is essentially
integrable (§1, No. 3, second paragraph after Def. 3).
c0 ) ⇒ a0 ): Assume that g satisfies c0 ), and let g0 be its extension by 0
to T . Since g is defined locally almost everywhere, g = g 0 locally almost
everywhere. If h ∈ K (T) , by assumption gh is essentially integrable, and
since g0 h = gh locally almost everywhere, it follows that g 0 h is essentially
integrable, and, being moderated, it is integrable. By the argument in the
text, for every t ∈ T there exists a neighborhood V of t such that g 0 ϕV
is integrable, and since gϕV = g0 ϕV locally almost everywhere, gϕV is
essentially integrable. ♦
An equivalent form of Prop. 10 is as follows (my guess is that this is
what the author had in mind, and that the version in the text was intended
as a more memorable paraphrase):
Proposition 100 . — Let g be a function defined locally almost every-
where in T (for the positive measure µ ), with values in a Banach space F
(resp. in R ), and let g0 be the extension of g to T by 0 . The following
properties are equivalent:
a00 ) For every point t ∈ T , there exists a neighborhood V of t such that
the function g0 ϕV is µ-integrable.
b00 ) The function g is µ-measurable and, for every compact set K ⊂ T ,
R∗ 0
|g |ϕK dµ < +∞ .
c00 ) For every numerical function h ∈ K (T) , g 0 h is µ-integrable.
Proof. We will show: a0 ) ⇔ a00 ); b0 ) ⇔ b00 ); c0 ) ⇔ c00 ). Since g is defined
locally almost everywhere, we know that g = g 0 locally almost everywhere.
a0 ) ⇔ a00 ): Assuming g satisfies a0 ), it is shown in the proof of Prop. 10
that for every t ∈ T there exists a neighborhood V of t such that g 0 ϕV is
integrable.
Conversely, if g0 ϕV is integrable then, since gϕV = g0 ϕV locally almost
everywhere, gϕV is essentially integrable.
b0 ) ⇔ b00 ): At any rate, since g is defined on a measurable set, we
know that g is measurable if and only if g 0 is measurable. Moreover, for
every compact set K ⊂ T , one has gϕK = g0 ϕK = (gϕK )0 locally almost
everywhere, whence
Z • Z • Z ∗
0
|g|ϕK dµ = |g |ϕK dµ = |g0 |ϕK dµ ,
INT V.x109 integration of measures §5

the first equality by the definition of the left member, the second because
|g0 |ϕK is moderated. Thus
Z • Z ∗
|g|ϕK dµ < +∞ ⇔ |g0 |ϕK dµ < +∞ ;

since gϕK is measurable if and only if (gϕK )0 = g0 ϕK is measurable, this

can also be expressed by saying that gϕ K is essentially integrable if and
only if g0 ϕK is integrable (§1, No. 3, Prop. 9).
c0 ) ⇔ c00 ): Similarly, for h ∈ K (T) ,
Z • Z • Z ∗
0
|gh| dµ = |g h| dµ = |g0 h| dµ ,

thus Z • Z ∗
|gh| dµ < +∞ ⇔ |g0 h| dµ < +∞ ,

whence gh is essentially integrable if and only if g 0 h is integrable. ♦

It remains to see how the proposed modifications will play out in the
rest of the text.
V.39, `. 20, 21.
“If g is locally θ-integrable, then every function equal to g locally
almost everywhere is locally integrable.”
Let f and g be functions defined locally almost everywhere in T ,
with values in F (or R ), such that f = g locally almost everywhere, and
let f 0 and g0 be their extensions by 0 to T . Then f 0 = g0 locally almost
everywhere and, for every numerical function h ∈ K (T) , one has

f 0 h = g0 h almost everywhere

(because the set K = Supp h is compact). In detail, the set

N = {t : f 0 (t) 6= g0 (t) }

is locally negligible, and, if f 0 (t)h(t) 6= g0 (t)h(t) then h(t) = 1 (by the

convention ±∞ · 0 = 0 ) and f 0 (t) 6= g0 (t) , whence t ∈ N ∩ K , which is a
negligible set (Ch. IV, §5, No. 2, Prop. 5).
It follows that f 0 h is integrable if and only if g 0 h is integrable (Ch. IV,
§5, No. 6, Th. 5 and No. 2, Prop. 6, and §2, No. 3, Prop. 6); thus f is
locally integrable if and only if g is locally integrable, by the criterion c 00 )
of Prop. 100 in the preceding Note.
§5 measures defined by numerical densities INT V.x110

V.39, `. 21, 22.

“It is clear that the sum of two locally integrable functions is locally
integrable.”
See the next Note.
V.39, `. 22–24.
“The functions with values in F , everywhere defined and locally inte-
1
grable for θ , form a vector space denoted L loc (T, θ; F) ”
1
For, if f , g ∈ Lloc (T, θ; F) and α is a scalar, then, for every h ∈ K (T) ,
one has
(f + g)h = f h + gh and (αf )h = α(f h) ,
and the assertion is immediate from criterion c 00 ) of Prop. 100 in the Note for
Prop. 1.
The assertion of the preceding Note is an immediate consequence of the
following:
Proposition. In order that a function g with values in a Banach space F
(resp. in R ) be locally integrable for a complex measure θ on T , it is neces-
sary and sufficient that it be equal locally almost everywhere to a function f
1 1
in Lloc (T, θ; F) (resp. Lloc (T, θ; R) ).
Proof. Sufficiency: Immediate from the Note for `. 20, 21.
Necessity: Suppose g is locally integrable (in the sense of Def. 1),
consider first the case that g takes its values in F , and let g 0 be its extension
to T by 0 . Then g0 also takes its values in F , and since g 0 = g locally
almost everywhere, g0 is locally integrable by the Note for `. 20, 21, thus
f = g0 meets the requirements.
Suppose now that the locally integrable function g takes its values
in R . The foregoing argument is thwarted by the fact that g may have
infinite values; let us show that g is finite locally almost everywhere.
Let A be the domain of g , so that the set N = { A is locally negligible,
and let
M = {t ∈ A : g(t) = ±∞ } ;
to show that M is locally negligible, it will suffice to show that for every
compact set K ⊂ T , the set K ∩ M is negligible. By criterion b 00 ) of
R∗ 0
Prop. 100 in the Note for Prop. 1, |g |ϕK d|θ| < +∞ , therefore |g 0 |ϕK is
finite almost everywhere (Ch. IV, §2, No. 3, Prop. 7), thus the set

{t : (g0 ϕK )(t) = ±∞ } = {t ∈ K : t ∈ A and g(t) = ±∞ } = K ∩ M

is indeed negligible. Thus, g is finite on the set A --- M , whose complement

N ∪ M is locally negligible.
INT V.x111 integration of measures §5

Then the function f = g0 ϕA -- M on T is finite-valued and f = g

locally almost everywhere, therefore f is also locally integrable, whence
1
f ∈ Lloc (T, θ; R) .
V.39, `. 26, 27. R 
“ . . . the topology defined by the semi-norms g 7→ gϕK  d|θ| , where
K runs over the set of compact subsets of T .”
It may be useful to pause to get a feeling for the topology just defined.
For the (locally convex) topology defined by a set of semi-norms, see TVS,
II, §1, No. 2, or §37 of my book LFAOT. (1)
1
For each compact set K ⊂ T R , let pK be the semi-norm on Lloc =
1 1
Lloc (T, θ; F) defined by pK (g) = |g|ϕK d|θ| . Equip Lloc with the locally
convex topology generated by the semi-norms p K . Our objective is to prove
the following:
1
Proposition. If E is a topological vector space and u : E → L loc is a
linear mapping, the following conditions are equivalent:
a) u is continuous;
b) for every compact set K in T , the semi-norm p K ◦ u on E is
continuous.
The proof is accomplished in two lemmas.
Lemma 1. Let Γ be a set of semi-norms on a vector space G and let
τ (Γ) be the (locally convex) topology on G generated by Γ ; a semi-norm p
on G is continuous for τ (Γ) if and only if p 6 p 1 + · · · + pn for a suitable
finite set of semi-norms pi ∈ Γ (TVS, II, §1, or LFAOT, p. 151, 37.15 and
37.17).
1
In particular, a semi-norm p on Lloc is continuous if and only if
p 6 cpK for some scalar c > 0 and some compact set K ⊂ T . For, if p is
continuous, say p 6 pK1 + · · · + pKn (Lemma 1), then p 6 n · pK1 ∪···∪Kn ;
conversely, if p 6 cpK then the continuity of p is immediate from Lemma 1.
The Proposition is a special case of the following:
Lemma 2 (LFAOT, p. 153, 37.24). Let G be a vector space equipped
with the (locally convex) topology generated by a set Γ of semi-norms. If
u : E → G is a linear mapping of a topological vector space E into G ,
following conditions are equivalent:
(i) u is continuous;
(ii) for every p ∈ Γ , the semi-norm p ◦ u on E is continuous.
Proof. (i) ⇒ (ii): Because every p ∈ Γ is continuous.
(1) Lectures on functional analysis and operator theory (Springer, 1974), whose
treatment of topological vector spaces was based on the fascicles of the first edition of EVT.
§5 measures defined by numerical densities INT V.x112

(ii) ⇒ (i): Given any neighborhood V of 0 in G , it will suffice to

−1
show that u (V) is a neighborhood of 0 in E . We can suppose that V
is convex, balanced and closed, since such neighborhoods are fundamental
(TVS, II, §8, No. 2); there then exists a continuous semi-norm p on G such
that V = {y ∈ G : p(y) 6 1 } (TVS, II, §2, No. 11, Prop. 22, or LFAOT,
p. 146, 37.4).
By Lemma 1, p 6 p1 + · · · + pn for suitable pi ∈ Γ , whence p ◦ u 6
p1 ◦ u + · · · + pn ◦ u . Since the pi ◦ u are, by assumption, continuous on E ,
the topology σ 0 they generate is coarser than the given topology σ on E ;
then p ◦ u is continuous for σ 0 (Lemma 1), hence for the finer topology σ .
Then
−1
u (V) = {x ∈ E : u(x) ∈ V }


= {x ∈ E : p u(x) 6 1 }
= {x ∈ E : (p ◦ u)(x) 6 1 } ,
which is a neighborhood of 0 in E by the continuity of p ◦ u }. ♦
If G is a locally convex space, its topology is generated by the set of
all continuous semi-norms on G (TVS, II, §4, No. 1, remarks following the
Cor. of Prop. 1, or LFAOT, p. 151, 37.17); Lemma 2 then takes the form:
Remark 1. If E is a topological vector space, G is a locally convex
space, and u : E → G is a linear mapping, then u is continuous if and only
if, for every continuous semi-norm p on G , the semi-norm p ◦ u on E is
continuous.
With notations as in Lemma 2, the topology τ on G generated by
the set Γ of semi-norms is characterized by the validity of the equivalence
(i) ⇔ (ii):
Remark 2. Let G be a locally convex space whose topology τ is gener-
ated by a set Γ of semi-norms. If σ is a locally convex topology on G for
which the equivalence (i) ⇔ (ii) of Lemma 2 holds, then σ = τ . The proof
in three steps:
1) The semi-norms p ∈ Γ are continuous for σ . For, consider the iden-
tity mapping u : (G, σ) → (G, σ) ; since u is continuous, by the assumption
(i) ⇒ (ii) for σ one knows that every p ◦ u = p (p ∈ Γ) is continuous on G
for σ .
2) It follows that for every p ∈ Γ , the (locally convex) topology τ p
on G generated by p satisfies τp ⊂ σ (LFAOT, p. 147, item (4) of 37.5),
therefore τ ⊂ σ .
3) Consider the identity mapping v : (G, τ ) → (G, σ) . We know that
for every p ∈ Γ , the semi-norm p ◦ v = p on G is continuous for τ (because
τp ⊂ τ ), therefore v is continuous by the assumption (ii) ⇒ (i) for σ ; that
is, σ ⊂ τ .
INT V.x113 integration of measures §5

1
In particular, the locally convex topology on L loc generated by the
semi-norms pK is characterized by the validity of the equivalence a) ⇔ b)
of the Proposition.
V.39, `. −12 to −10.
1
“The associated Hausdorff space, the quotient of L loc (T, θ; F) by the
∞
subspace NF of mappings that are zero locally almost everywhere, is de-
noted L1loc (T, θ; F) .”
The space NF∞ wasR defined in Ch. IV, §6, No. 3. As in the preceding
1 1
Note, we write pK (g) = |g|ϕK d|θ| for every g ∈ Lloc = Lloc (T, θ; F) and
every compact set K ⊂ T .
Proof that NF∞ ⊂ Lloc 1
: Let g ∈ NF∞ . For every compact K ⊂ T , the
function gϕK is locally negligible and moderated, hence negligible (§1, No. 2,
1
Cor. 1 of Prop. 7), therefore integrable, whence g ∈ L loc and pK (g) = 0
for every K .
1
Conversely, if g ∈ Lloc and pK (g) = 0 for every compact K ⊂ T , then
the integrable function |g|ϕK is negligible, whence g is locally negligible
(Ch. IV, §5, No. 2, Prop. 5). Thus

NF∞ = {g ∈ Lloc
1
: pK (g) = 0 for every compact set K ⊂ T } .
1
Since the semi-norms pK define the topology of Lloc , it follows that
NF∞ 1
is the closure of {0} in Lloc (TVS, II, §1, No. 2, Prop. 2, (i)), there-
1
fore the quotient space Lloc /NF∞ is the Hausdorff topological vector space
associated with Lloc (loc. cit., No. 3). The space L1loc = Lloc
1 1
/NF∞ is also
locally convex (TVS, II, §4, No. 4, Example I), its topology being defined by
1
the family of semi-norms ġ 7→ pK (g) ( ġ the equivalence class of g in Lloc
for equality locally almost everywhere); the details are as follows.
As L1loc bears the final topology for the quotient mapping u : L loc 1
→
1 1
Lloc (loc. cit.), a semi-norm q on Lloc is continuous if and only if the semi-
1
norm q◦u : g 7→ q(ġ) on Lloc is continuous. Every continuous semi-norm p
1 ∞
on Lloc is zero on NF (because the pK are), so the formula q(ġ) = p(g)
defines a semi-norm q on L1loc such that p = q ◦ u , and q is continuous
by the foregoing. In particular, for every compact set K ⊂ T the formula
qK (ġ) = pK (g) defines a continuous semi-norm on L 1loc . On the other hand,
if q is any continuous semi-norm on L 1loc , then the semi-norm q ◦u on Lloc 1

is continuous, therefore q ◦ u 6 pK1 + · · · + pKn for suitable K1 , . . . , Kn

(Lemma 1 in the preceding Note), whence q 6 q K1 + · · · + qKn ; it follows
(Lemma 1 again) that the locally convex topology of L 1loc is generated by
the semi-norms qK .
1
{The argument carries over to the case that L loc and the pK are
replaced by a vector space E and a set Γ of semi-norms on E (cf., TVS,
II, §1, No. 3, `. 3–6 on p. TVS II.5).}
§5 measures defined by numerical densities INT V.x114

Remark. If T is countable at infinity, then L 1loc is metrizable.

Proof. One knows (GT, I, §1, No. 9, Prop. 15) that T is the union of
a sequence (Un ) of relatively compact open sets such that U n ⊂ Un+1 for
all n . Let Kn = Un ; (Kn ) is an increasing sequence of compact sets with
union T , and each compact set K is contained in K n for some n (loc. cit.,
Cor. 1). R
As in the preceding Note, write qK (ġ) = pK (g) = |g|ϕK d|θ| for
the semi-norms qK that generate the locally convex topology of L 1loc . For
every compact set K in T , one has qK 6 qKn for some n , consequently
the topology of L1loc is generated by the sequence of semi-norms q Kn (Lem-
ma 1 in the preceding Note); it follows that the uniform structure of L 1loc is
metrizable (GT, IX, §2, No. 4, Cor. 1 of Th. 1). ♦
{In general, a Hausdorff locally convex space is metrizable if and only
if its topology is generated by a sequence of semi-norms: (1) “If”, by the
cited Cor. 1; “Only if”, because a metrizable locally convex space has a
fundamental sequence of neighborhoods of 0 , which one can suppose are
closed, balanced and convex, whence the desired sequence of semi-norms
(TVS, II, §2, No. 11, Prop. 22).}
V.39, `. −9, −8.
“It can be shown that the topological vector spaces just defined are
complete (Exer. 31).”
If T is countable at infinity, then L1loc is metrizable (Remark in the
preceding Note), and the present assertion implies that L 1loc is a Fréchet
space (TVS, II, §8, No. 2, last sentence). The hint given in Exer. 31:
P
“Write |θ| in the form µα , where (µα ) is a summable family of
α
measures > 0 on T whose supports form a locally countable family of
pairwise disjoint compact
Q 1 subsets. This defines a continuous mapping of
1
Lloc (T, θ; F) into LF (T, µα ) . Let F be a Cauchy filter on L1loc (T, θ; F) .
Q 1 α
Its image in LF (T, µα ) converges to an element (fα ) . Show that the fα
α
define a θ-measurable function f on T , then that f is locally θ-integrable,
and that ḟ is the limit of F in L1loc (T, θ; F) .”
I did not succeed in showing that f is locally θ-integrable, but the
following construction of a measurable function f may be a step in the right
direction:
(1) Cf. H.H. Schaefer, Topological vector spaces, p. 48, Macmillan, 1966; W. Rudin,
Functional analysis, p. 27, Remark 1.38, c), McGraw-Hill, 1973; J. Horvath, Topological
vector spaces and distributions. I., p. 113, Addison-Wesley, 1966; R.E. Edwards, Func-
tional analysis: Theory and applications, p. 422, Holt, Rinehart and Winston, 1965.
INT V.x115 integration of measures §5
P
Write µ = |θ| . By §2, No. 3, Prop. 4, µ = µα , where (µα ) is a
α
summable family of positive measures on T whose supports K α = Supp µα
are compact, pairwise disjoint, and form a locally countable family (K α ) .
The natural mapping is
Y
u : L1loc (T, µ; F) → L1F (T, µα )
α


where u(ġ) = (gϕKα )˙ for g ∈ Lloc
1
(T, µ; F) , characterized by

prα ◦ u : ġ 7→ (gϕKα )˙ for all α and g .

In detail:
1
(i) Definition of u . Let g ∈ Lloc (T, µ; F) , and fix an index α . One
knows that g is µ-measurable and gϕKα is µα -integrable (Def. 1 and
Prop. 1). Since µα 6 µ , it follows that g is also µα -measurable (§1, No. 4,
Cor. 2 of Prop. 11) and gϕKα is µα -integrable. Since µ∗α (T --- Kα ) = 0
(Ch. IV, §2, No. 2, Prop. 5), it follows that ϕ Kα = 1 µα -almost everywhere,
whence gϕKα = g µα -almost everywhere; thus g ∈ LF1 (T, µα ) . Moreover,
if g = g0 locally µ-almost everywhere then g = g 0 µα -almost everywhere,
whence a well-defined mapping

uα : L1loc (T, µ; F) → L1F (T, µα )

1
such that uα (ġ) = ġ for g ∈ Lloc (T, µ; F) , where the over-dot signifies, in
1
turn, the class of g in Lloc (T, µ; F) for equality locally µ-almost everywhere
and the class of g in LF1 (T, µα ) for equality µα -almost everywhere. Thus
the proposed mapping u is defined by
1
u : ġ 7→ (ġ) for g ∈ Lloc (T, µ; F),

where the (ġ) on the right side is the family, indexed by α , whose α-th
coordinate is ġ ∈ L1F (T, µα ) ; in other words, prα ◦ u = uα .
(ii) Continuity of u . It suffices to show that for each α , the mapping
prα ◦ u = uα is continuous. The topology of L1F (T, µα ) is generated by the
norm function q α ,
Z Z
q α (ġ) = |g| dµα = |g|ϕKα dµα ;

to prove that uα is continuous, we need only show that the semi-norm

q α ◦ uα on L1loc (T, µ; F) is continuous (Lemma 2 in the Note for V.39,
`. 26, 27). Indeed,
§5 measures defined by numerical densities INT V.x116

(q α ◦ uα )(ġ) = q α uα (ġ) ġ ∈ L1loc (T, µ; F)


= q α (ġ) ġ ∈ L1F (T, µα )
Z Z
= |g| dµα = |g|ϕKα dµα
Z
6 |g|ϕKα dµ = pKα (g) = qKα (ġ)

( qKα is defined in the Note for V.39, `. −12 to −10), that is, q α ◦ uα 6 qKα ;
since qKα is a continuous semi-norm on L1loc (T, µ; F) , so is q α ◦ uα .
(iii) Since u is linear and continuous, it is uniformly continuous (GT, III,
§3, No. 1, Prop. 3); therefore, if FQis a Cauchy filter on L 1loc (T, µ; F) , then
the filter base u(F) is Cauchy on L1F (T, µα ) (GT, II, §3, No. 1, Prop. 3).
Q 1 α
Now, LF (T, µα ) is locally convex for the product topology (TVS, II, §4,
α
No. 3, Prop. 4), and, for the product uniformity (GT, II, §2, No. 6, Def. 4),
it is complete (GT, II, §3, No. 5, Prop. 10), therefore the Cauchy filter base
u(F) is convergent; say u(F) → (ḟα ) , where fα ∈ LF1 (T, µα ) for all α .
We can suppose that the fα are universally measurable, i.e., measurable
for every measure on T (§3, No. 4, Prop. 7).
(iv) Definition of f . Let (fα ) be the family of universally measurable
functions on T constructed in (iii), with f α µα -integrable for all α . Define
f : T → F by 
 fα (t) for t ∈ Kα
f (t) = [
 0 for t ∈ T --- Kα .
α

(iv) Universal measurability

S of f . Since the family (K α ) is locally
countable, the set B = Kα , is universally measurable (Ch. IV, §5, No. 9);
α  
moreover, for every α , the function f Kα = fα Kα is universally
 measur-

able (in the sense of loc. cit., No. 10, Def. 8), therefore f B is universally

measurable (loc. cit., Prop. 16). Finally, f is the extension by 0 of f B
to T , hence is universally measurable (loc. cit., Prop. 15).
V.39, `. −7, −6.
“Every measurable function g , that is essentially bounded on every
compact set, is locally integrable.”
The term “essentially bounded” is not indexed and seems nowhere to
be defined. My guess is that “ g is essentially bounded” means the same
thing as “ g is bounded in measure” (Ch. IV, §6, No. 3, Def. 2), i.e., there
exists a bounded function h such that g = h locally almost everywhere,
INT V.x117 integration of measures §5

equivalently, N∞ (g) < +∞ . If so, then a function g : T → F is measurable

and essentially bounded if and only if g ∈ L F∞ (T, θ) (ibid.); and, to say that
g is essentially bounded on a compact set K means that gϕ K is bounded
almost everywhere (Ch. IV, §5, No. 2, Prop. 5), that is, |g|ϕ K 6 c almost
everywhere for some constant c < +∞ . Thus, if g is measurable and is
essentially bounded on a compact set K , then gϕ K is integrable (loc. cit.,
No. 6, Th. 5); and if this is true for every compact set K , then g is locally
integrable.
V.39, `. −2 to V.40, `. 2.
“Let F, G, H be three Banach spaces, and (u, v) 7→ Φ(u, v) a contin-
uous bilinear mapping of F × G into H . If f is locally integrable and takes
its values in F , and if g ∈ LG∞ , then Φ(f , g) is locally integrable (Ch. IV,
§6, No. 4, Cor. 1 of Th. 2).”
The rôle of bilinearity is to assure the existence of a real constant c > 0
such that |Φ(f , g)| 6 c · N1 (f ) N∞ (g) for f ∈ LF1 and g ∈ LG∞ (GT, IX,
§3, No. 5, Th. 1).
1
Suppose f ∈ Lloc (T, θ; F) and g ∈ LG∞ (T, θ) . For every compact set
K ⊂ T , the function f ϕK is integrable, therefore Φ(f ϕK , g) ∈ LH1 (T, θ) by
the cited Cor. 1. But, for all ∈ T , by definition



Φ(f ϕK , g) (t) = Φ (f ϕK )(t), g(t)


= Φ ϕK (t)f (t), g(t)


= ϕK (t)Φ f (t), g(t)


= ϕK Φ(f , g) (t) ,

thus ϕK Φ(f , g) = Φ(f ϕK , g) is integrable. Since K is arbitrary, Φ(f , g) is

locally integrable—and, in particular, measurable (Prop. 1).
V.40, `. 5–9.
“ . . . the set of t such that g(t) = +∞ is then locally µ-negligible,
because gϕK is µ-integrable for every compact set K (Ch. IV, §2, No. 3,
Prop. 7). Now let g 0 be a locally integrable function that is positive and
finite, equal to g locally µ-almost everywhere ”
See the Proposition in the Note for V.39, `. 22–24. (Here, µ is a positive
measure on T —for example, µ = |θ| for some complex measure θ on T .)
V.40, `. 9–11.
“The mapping t 7→ λ0t of T into M+ (T) is vaguely µ-measurable
and scalarly essentially integrable (or again, the pair (I, g 0 ) , where I is the
identity mapping of T , is µ-adapted) ”
§5 measures defined by numerical densities INT V.x118

The mapping Λ : T → M+ (T) defined by Λ : t 7→ λ0t = g 0 (t)εt is

scalarly essentially µ-integrable (§3, No. 1); for, for every f ∈ K (T) , the
function
Λ(f ) : t 7→ λ0t (f ) = g 0 (t)εt (f ) = g 0 (t)f (t),
that is, the function g 0 f , is µ-integrable (Def. 1) hence R essentially µ-integrable;
one can therefore define the positive measure ν = λ0t dµ(t) .
Since the functions I and g 0 are µ-measurable and, for every f in
K (T) , the function


t 7→ f I(t) g 0 (t) = f (t)g 0 (t)

is µ-integrable, the pair (I, g 0 ) is µ-adapted (§4, No. 1, Def. 1); therefore Λ
is vaguely measurable (and µ-adequate) (loc. R cit., Prop. 1). R
The measure ν is defined by ν(f ) = λt (f ) dµ(t) = g 0 (t)f (t) dµ(t)
0

for all f ∈ K (T) (§3, No. 1); since gf = g 0 f µ-almost everywhere, one is
permitted to write
Z Z
ν(f ) = gf dµ = g(t)f (t) dµ(t)

(Ch. IV, §4, No. 1, last paragraph), to replace Λ by the mapping t 7→ λ t =

g(t)εt (defined
R at the points in the domain of g where g is finite), and to
write ν = λt dµ(t) (§3, No. 3, Remark ).
V.40, `. −15 to −12.
“ . . . one can write
u = g1 − g2 + i(g3 − g4 )
(2)
θ = µ1 − µ2 + i(µ3 − µ4 )

where µ1 = (Rθ)+ , . . .”
Understood, that when u takes its values in R , g 3 = g4 = 0 and g1 , g2
are not simultaneously equal to +∞ at any point; when u takes its values
in C , the gi are all finite-valued; when θ is a real measure, µ 3 = µ4 = 0 .
One need not take µ1 = (Rθ)+ , etc., but it is essential that the g i be
θ-measurable, 0 6 gi 6 |g| and 0 6 µj 6 µ for all i and j (the argument
of §2, No. 2, Cor. 2 of Prop. 3 is then applicable); thus the local integrability
of u for θ implies that gi is locally µj -integrable for all i and j .
It is clear from the earlier discussion that one can suppose u to be
1
defined and finite at every point of T , i.e., that u ∈ L loc (T, θ; C) .
{Conversely, if gi is µj -integrable for all i and j , then each g i is
integrable for θ (§2, No. 2, Cor. 2 of Prop. 3), therefore so is u ; it follows that
if gi is locally µj -integrable for all i and j , then u is locally θ-integrable.}
INT V.x119 integration of measures §5

V.40, `. −9 to −6.
“ . . . the mapping Z
f 7→ f (t)u(t) dθ(t)

on K (T) is a complex measure.”

By assumption, f u is |θ|-integrable for every f ∈ K (T) = K (T; C)
(No. 1, Def. 1). Regarding θ and the µ j as linear forms on K (T) , the
second equation of (2) means that

(i) θ(f ) = µ1 (f ) − µ2 (f ) + iµ3 (f ) − iµ4 (f )

for all f ∈ K (T) . It follows from the assumptions on the µ j that every
θ-integrable function h is µj -integrable for all j , and that
Z Z Z Z Z
(ii) h dθ = h dµ1 − h dµ2 + i h dµ3 − i h dµ4

(see the Note for V.10, `. 13, 14). In particular, for every f ∈ K (T) one
has
Z Z Z Z Z
(iii) f u dθ = f u dµ1 − f u dµ2 + i f u dµ3 − i f u dµ4 .

Also, for f ∈ K (T) one has

f u = f g1 − f g2 + if g3 − if g4 ;

substituting this expression for f u on the right side Rof (iii), it follows from
the linearity of integration that the linear form f 7→ f u dθ on K (T) is a
linear combination of the (sixteen) linear forms
Z


(iv) f 7→ f gi dµj f ∈ K (T) ,

each of which is known to be a positive linear form on K (T; R) , hence R is a

measure on T (Ch. III, §1, No. 5, Th. 1). Thus, the linear form f 7→ f u dθ
on K (T) , being a linear combination of measures, is itself a measure, i.e.,
is continuous for the inductive limit topology (loc. cit., No. 3, Def. 2).
V.41, `. 6, 7.
“ . . . if u1 and u2 are locally θ-integrable, then (u1 + u2 ) · θ = u1 · θ +
u2 · θ .”
§5 measures defined by numerical densities INT V.x120

The function u1 + u2 is defined and finite locally θ-almost everywhere

in T and, for every f ∈ K (T) , f · (u1 + u2 ) = f u1 + f u2 θ-almost
everywhere; since f u1 and f u2 are θ-integrable, so is f · (u1 + u2 ) , and
Z Z Z
f · (u1 + u2 ) dθ = f u1 dθ + f u2 dθ ,

1
whence the assertion. {Alternatively, one can suppose that u i ∈ Lloc etc.}
V.41, `. 7–9.
“ . . . if θ1 and θ2 are two measures on T , and if u is a function locally
integrable for θ1 and θ2 , then u is locally integrable for θ1 + θ2 and one
has u · (θ1 + θ2 ) = u · θ1 + u · θ2 .”
Let A = {t ∈ T : u(t) is defined and finite} , and let N = { A .
Since N is locally negligible for θ1 and θ2 , it is locally negligible for θ1 + θ2
( |θ1 + θ2 |• 6 |θ1 |• + |θ2 |• ); we can therefore suppose that u is defined and
finite everywhere on T .
By assumption, for every f ∈ K (T) one has f u ∈ L 1 (θ1 ) ∩ L 1 (θ2 ) ,
therefore f u ∈ L 1 (θ1 + θ2 ) and
Z Z Z
f u d(θ1 + θ2 ) = f u dθ1 + f u dθ2

(see the Theorem in the Note for V.10, `. 13, 14).

V.41, `. −13.

(4) h|θ|, |f |i = sup |hθ, cf i| = sup |hθ, cf i|

c∈K1 c∈B1

The brackets pertain to the duality between measures and integrable

functions (discussed at length in the Note for V.10, `. 13, 14). In the second
member of (4), cf is integrable because f is essentially integrable and
c ∈ K (T) is bounded, measurable
R and moderated (§1, No. 3, Cor. of
Prop. 9), and so hθ, cf i = (cf ) dθ . However, in the first and third mem-
bers, |f | and cf are only known to be essentially integrable; nevertheless,
Z Z
h|θ|, |f |i = |f | d|θ| , hθ, cf i = (cf ) dθ ,

because f = f 0 locally almost everywhere for some f 0 ∈ LC1 (T, θ) and, by

definition (loc. cit., Def. 3),
Z Z Z Z
0
|f | d|θ| = |f | d|θ| , (cf ) dθ = (cf 0 ) dθ

(the left members are ‘essential integrals’—merely a notation for the right
members, which are ‘honest’ integrals).
INT V.x121 integration of measures §5

V.41, `. −10 to −8.

“Obviously

sup |hθ, cf i| 6 sup |hθ, cf i| 6 h|θ|, |f |i

c∈K1 c∈B1

(Ch. IV, §4, No. 2, Prop. 2).”

The first inequality results from the fact that continuous functions are
Borel (GT, IX, §6, No. 3, Prop. 10).(1) Note that if X is any locally compact
space and Y is a metrizable space with a countable base for open sets
(e.g., a separable Banach space), then every Borel mapping c : X → Y is
universally measurable (§3, No. 4, Def. 2 and Ch. IV, §5, No. 5, Remark 2).
As for the second inequality: if c ∈ B 1 then c is (universally) measur-
able and |cf | 6 |f | , whence

|hθ, cf i| 6 h|θ|, |cf |i 6 h|θ|, |f |i

by the generalization of the cited Prop. 2 to essentially integrable functions

(signaled in the paragraph preceding §1, No. 3, Prop. 10).
V.41, `. −8 to −4.
“ . . . let g be an element of K (T; C) such that |g| 6 |f | ; g is the
uniform limit of a sequence (gn ) of elements of K (T; C) whose supports
are contained in the open set U formed by the t such that f (t) 6= 0 , and
one may clearly suppose that |gn | 6 |f | for every n .”
The condition |g| 6 |f | looks forward to a supremum over all such g .
For n = 1, 2, 3, . . . let Kn = {t ∈ T : |g(t)| > 1/n } ; then Kn is a closed
subset of the support of g , hence is compact, and K n ⊂ U . For each n ,
choose hn ∈ K+ (T) so that 0 6 hn 6 1 , hn = 1 on Kn , and Supp hn ⊂ U
(Ch. III, §1, No. 2, Lemma 1). Then hn g ∈ K (T; C) , |hn g| 6 |g| 6 |f | ,
and
Supp(hn g) ⊂ Supp hn ⊂ U ;
moreover, hn g − g = 0 on Kn , and, for t ∈ { Kn ,


|(hn g − g)(t)| = 1 − hn (t) · |g(t)| 6 |g(t)| < 1/n ,

whence khn g − gk 6 1/n . Thus gn = hn g meets the requirements of the

assertion (and, moreover, satisfies |g n | 6 |g| ).
(1) The Borel sets of a topological space X are the elements of the tribe generated
by the set of closed subsets of X (GT, IX, §6, No. 3, Def. 4). The proof of the cited
Prop. 10, and Exer. 16 for GT, IX, §6, suggest that a mapping f : X → Y between
topological spaces should be called a Borel mapping if the inverse image of every Borel
set (equivalently, of every closed set) in Y is a Borel set in X , but GT does not pass to
the act; TG passes to the act (TG, IX, §6, No. 3, Def. 5).
§5 measures defined by numerical densities INT V.x122

V.41, `. −3 to V.42, `. 2.
“ . . . then cn ∈ K1 , g = lim cn f , therefore |hθ, gi| = lim |hθ, cn f i| ,
n→∞ n→∞
and finally
sup |hθ, gi| 6 sup |hθ, cf i| .
|g|6|f |, g∈K (T;C) c∈K1
One concludes by observing that the first member of this inequality is equal
to h|θ|, |f |i (Ch. III, §1, No. 6, formula (12)).”
In slow motion,

 gn (t) for t ∈ U
cn (t) = f (t)

0 for t ∈ { U .
From Supp cn ⊂ Supp gn one knows that cn has compact support; at issue
is the continuity of cn . From Supp gn ⊂ U one knows that the open sets U
and { (Supp gn ) have union T , so it suffices to show that the restrictions
of cn to U and { (Supp  gn ) are continuous. Indeed, from the definition
of cn , it is clear that cn U is continuous; whereas
 if t ∈ { (Supp gn ) then

cn (t) = 0 whether t ∈ U or t ∈ { U , thus cn { (Supp gn ) = 0 .
Note that cn f = gn : the equality holds at the points of U by the
definition of cn , and both members are equal to 0 on the subset { U
of { (Supp gn ) . It follows that kg − cn f k = kg − gn k → 0 , thus cn f → g
uniformly. Let K = U = Supp f ; then Supp(c n f ) ⊂ Supp f = K and
Supp g ⊂ Supp f = K (because |g| 6 |f | ), thus c n f and g belong to
K (T, K; C) , consequently cn f → g in K (T, K; C) . Since the norm topol-
ogy on K (T, K; C) is equal to the topology induced by the (inductive limit)
topology of K (T; C) , it follows that c n f → g in K (T; C) ; by the conti-
nuity of θ , θ(cn f ) → θ(g) , whence
lim |hθ, cn f i| = |hθ, gi| .
n→∞
Passing to the limit in the inequality |hθ, c n f i| 6 sup |hθ, cf i| , one has
c∈K1

|hθ, gi| 6 sup |hθ, cf i| ,

c∈K1
and the validity of this inequality for all g yields
sup |hθ, gi| 6 sup |hθ, cf i| ,
|g|6|f |, g∈K (T;C) c∈K1

in other words
h|θ|, |f |i 6 sup |hθ, cf i|
c∈K1
by the cited formula (12). This implies that in the inequalities of `. −9,
equality holds throughout. Thus the equalities (4) are verified for the case
that f ∈ K (T; C) .
INT V.x123 integration of measures §5

V.42, `. 6.
“ . . . the dense subspace K (T; C) .”
1
Note that the semi-norm N 1 on L C coincides with N1 on LC1 (§1,
No. 2, Prop. 7, 2)).
1
The closure of K (T; C) in L C is a linear subspace that contains
LC1 (Ch. IV, §3, No. 4, Def. 2) and NC∞ (§1, No. 3) hence it contains
1
LC1 + NC∞ = L C .
V.42, `. 8, 9.
|h|θ|, |f |i − h|θ|, |f 0 |i| 6 h|θ|, |f − f 0 |i = N1 (f − f 0 )
|hθ, cf i − hθ, cf 0 i| 6 h|θ|, |c| |f − f 0 |i 6 N1 (f − f 0 )

If g ∈ LC1 with f = g locally almost everywhere, then cf = cg locally

almost everywhere and
Z Z
hθ, cf i = cf dθ = cg dθ = hθ, cgi

by the definitions, and similarly for the other five bracket expressions in the
displayed relations; one can therefore suppose that f and f 0 belong to LC1 ,
whence also N1 (f − f 0 ) = N1 (f − f 0 ) .
 The first
 line of the displayed relations then follows from the inequality
|f | − |f 0 | 6 |f − f 0 | and the computation
   
h|θ|, |f |i − h|θ|, |f 0 |i = h|θ|, |f | − |f 0 |i

 
6 |θ|, |f | − |f 0 | 6 h|θ|, |f − f 0 |i = N1 (f − f 0 )

(Ch. IV, §4, No. 2, Prop. 2); similarly for the second line of the display,
taking into account that c is (universally) measurable and |c| 6 1 .
From the first line of the display, it is clear that the first member of (4)
1
depends continuously on f ∈ L C . As to the second member, let us abbre-
viate by defining

1
α(f ) = sup |hθ, cf i| (f ∈ L C ) .
c∈K1

1
For f, f 0 ∈ L C and c ∈ K1 , one has

|hθ, cf i| 6 |hθ, cf i − hθ, cf 0 i| + |hθ, cf 0 i|

6 N1 (f − f 0 ) + |hθ, cf 0 i| 6 N1 (f − f 0 ) + α(f 0 )
§5 measures defined by numerical densities INT V.x124

(the first 6 , by the triangle inequality in C ; the second 6 , by the second

line of the display; the third 6 , by the definition of α ). Varying c ∈ K 1 ,
one has
α(f ) 6 N1 (f − f 0 ) + α(f 0 ) ;
by symmetry in f and f 0 , clearly

|α(f ) − α(f 0 )| 6 N1 (f − f 0 ) ,

whence the continuity of the second member of (4) as a function of f , and

a similar argument establishes the continuity of the third member of (4).
{For the purposes of Prop. 2, it is not necessary to introduce the pa-
rameter c ∈ B1 ; perhaps it is destined for a future application.}
V.42, `. −6, −5.
“This follows at once from Prop. 2 and the formulas (6) of Ch. II, §1,
No. 1.”
1
The equality sup(g1 , g2 ) = (g1 + g2 + |g1 − g2 |) holds on the set of
2
points where g1 and g2 are both defined and finite—hence locally almost
everywhere—therefore

1
sup(g1 , g2 ) · µ = (g1 · µ + g2 · µ + |g1 − g2 | · µ)
2
1
= (g1 · µ + g2 · µ + |(g1 − g2 ) · µ|)
2
1
= (g1 · µ + g2 · µ + |g1 · µ − g2 · µ|)
2
= sup(g1 · µ, g2 · µ)

(the second equality, by Prop. 2; the fourth, by the cited formula), and
similarly
inf(g1 , g2 ) · µ = inf(g1 · µ, g2 · µ) .
In particular, setting g1 = g and g2 = 0 , one has

(g · µ)+ = sup(g · µ, 0) = sup(g, 0) · µ = g + · µ

+
+
(g · µ)− = − (g · µ) = (−g) · µ = (−g)+ · µ = g − · µ .

V.42, `. −3 to −1.
“In the statements of this subsection, g denotes a positive numerical
function, defined everywhere and locally µ-integrable, θ denotes a complex
measure, and u a locally θ-integrable complex function.”
INT V.x125 integration of measures §5

Here µ denotes a positive measure on T . As announced in No. 2

(p. V.41, `. 10, 11), henceforth all locally integrable functions are assumed
to be defined everywhere on T . In particular, u(t) ∈ C for all t ∈ T ; the
function g > 0 may take the value +∞ , but, being locally integrable, it is
finite locally almost everywhere.
V.43, `. 1–3.
“The remarks in RNo. 2 show that the results of §4 are applicable to the
measure ν = g · µ = g(t)εt dµ(t) (even though the measure g(t)ε t is not
defined unless g(t) 6= +∞ ).”
Since g is locally µ-integrable, there exists a (everywhere finite) function
0 1
g ∈ Lloc (T, µ; R) such that g = g 0 locally µ-almost everywhere and g 0 (t) > 0
for all t ∈ T . Defining λ0t = g 0 (t)εt for all t ∈ T , the mapping Λ0 : t 7→ λ0t
is scalarly essentially µ-integrable: for f ∈ K (T) , the function t 7→ λ 0t (f ) =
g 0 (t)f (t) is the µ-integrable function g 0 f , and the formula
Z Z
ν(f ) = g 0 f dµ = λ0t (f ) dµ(t)

defines a measure ν > 0 on T (by the cited remarks). The functions gf

and g 0 f are defined everywhere in T and gf = g 0 f µ-almost everywhere
( f is µ-moderated),
R therefore gf is also µ-integrable, and one can also write
ν(f ) = gf dµ (p. IV.34, `. 1–5). By No. 2, Def. 2, g · µ = ν .
Another perspective on ν : let A = {t ∈ T : g(t) 6= +∞ } and consider
the mapping
Λ : t 7→ g(t)εt (t ∈ A) .
One knows (by the cited remarks) that (I, g 0 ) is µ-adapted, hence the map-
ping Λ0 : t 7→ g 0 (t)εt (t ∈ T) is µ-adequate (and vaguely µ-measurable)
by §4, No. 1, Prop. 1. Since Λ = Λ0 locally µ-almost everywhere, we are
authorized (§3, No. 3, Remark ) to say that Λ is µ-adequate, and to write
Z Z
g(t)εt dµ(t) = g 0 (t)εt dµ(t) ,

R
that is, g(t)εt dµ(t) = ν .
V.43, `. 6.
“This follows from Th. 1 of §4, No. 2.”
By assumption, g is defined and > 0 everywhere in T , but may have
values equal to +∞ . With notations as in the preceding Note, (I, g 0 ) is
µ-adapted;
R in particular, g 0 (t) is finite and > 0 for every t ∈ T , and one
0
has ν = g (t)εt dµ(t) .
§5 measures defined by numerical densities INT V.x126

If f is defined everywhere on T , the cited Th. 1 immediately yields

ν (f ) = µ• (f g 0 ) . As f g is also everywhere-defined and f g = f g 0 locally
•

µ-almost everywhere, one has µ• (f g) = µ• (f g 0 ) , whence ν • (f ) = µ• (f g) ,

that is,
Z • Z •
(∗) f (t) dν(t) = f (t)g(t) dµ(t) .

When f is only defined and > 0 locally µ-almost everywhere, the same
is true of f g and f g 0 , and f gR= f g 0 locally
R • µ-almost everywhere. Imitating
• 0
p. V.9, `. 1–5, if one defines f dν
R• = f dν
R•, where f 0 isR the extension
• 0
by 0 of f to T , and similarly f g dµ = (f g)0 dµ = f g dµ (the
convention 0 · (+∞) = 0 is at work in the last equality), then
Z • Z •
0
f (t) dν(t) = f 0 (t)g(t) dµ(t)

R• R•
by (∗); in other words f (t) dν(t) = f (t)g(t) dµ(t) , that is, f satis-
fies (∗) formally.
V.43, `. 13.
“ . . . the statement then follows at once from Prop. 3.”
In Prop. 3, put µ = |θ| , g = |u| , f = |f | ; then ν = g·µ = |u|·|θ| = |u·θ|
(No. 2, Prop. 2) and Prop. 3 yields
Z • Z • Z •
|u · θ|• (|f |) = |f | d(|u · θ|) = |f | |u| d|θ| = |uf | d|θ| = |θ|• (|uf |) ;

in particular, |u · θ|• (|f |) = 0 ⇔ |θ|• (|uf |) = 0 , thus f is locally negligible

for u · θ if and only if uf is locally negligible for θ (§1, No. 1).
V.43, `. 17.
“One is immediately reduced to showing that . . . ”
Suppose, more generally, that u1 , u2 are defined locally almost ev-
erywhere for θ , with values either in C or in R . For i = 1, 2 choose
1
vi ∈ Lloc (T, θ; C) with ui = vi locally almost everywhere for θ . For every
f ∈ K (T) , f vi is θ-integrable and f ui = f vi |θ|-almost everywhere (No. 1,
Def. 1 and IV.34, `. 1–5), whence ui · θ = vi · θ (No. 2, Def. 2).
We can therefore suppose that u1 , u2 are defined and complex-valued
everywhere in T , whence u2 · θ = (u2 − u1 ) · θ + u1 · θ (V.41, `. 6, 7). Then
u1 · θ = u2 · θ ⇔ (u2 − u1 ) · θ = 0 ⇔ (u2 − u1 )1 is locally negligible for θ
(Cor. 1, with 1 the constant function equal to 1 on T ), that is, if and only
if u1 = u2 locally almost everywhere for θ .
INT V.x127 integration of measures §5

V.43, `. −9, −8.

“When u and θ are positive, this follows at once from Prop. 3 of §4,
No. 3.”
When θ > 0 and the locally θ-integrable function u is > 0 everywhere
in T then, by the first paragraph of No. 2, R the pair (I, u) is θ-adapted and
it follows from (1) that the measure ν = u(t) εt dθ(t) is the measure u · θ
of Def. 2 (loc. cit.). Since f ◦ I = f , the assertion is indeed immediate from
the cited Prop. 3.
V.43, `. −8, −7.
“The result then extends to the case that u and θ are complex thanks
to Prop. 2.”
In slow motion,

f is measurable for u · θ ⇔ f is measurable for |u · θ|

⇔ f is measurable for |u| · |θ|

⇔ f S is measurable for |θ|

⇔ f S is measurable for θ

(the 1st and 4th equivalence, by Ch. IV, §5, No. 1, Def. 1), the 2nd by No. 2,
Prop. 2, and the 3rd by the ‘positive case’ already considered). This settles
the case that u is defined everywhere on T .
Suppose more generally that the locally θ-integrable function u (with
values in C or in R ) is defined only locally θ-almost everywhere, say
with domain A ⊂ T , where T --- A is locally θ-negligible. Choose u 0
1
in Lloc (T, θ; C) such that u = u0 locally θ-almost everywhere, and let N
be a locally θ-negligible set such that u(t) is defined and equal to u 0 (t) on
T --- N , that is, T --- N ⊂ A and u = u0 on T --- N . By Cor. 2, u·θ = u0 ·θ .
Let
S = {t ∈ A : u(t) 6= 0 } , S0 = {t ∈ T : u0 (t) 6= 0 } .

By the foregoing, we know that f S0 is θ-measurable
 if and only if f is
measurable for u0 · θ = u · θ ; to show that f S is θ-measurable if and only
if f is measurable for u · θ , we need only show that
 
(∗) f S is θ-measurable ⇔ f S0 is θ-measurable.

One knows that the set S0 is θ-measurable (Ch. IV, §5, No. 5, Prop. 7),
therefore so is S0 ∩ (T --- N) . Note that

S ∩ (T --- N) = S0 ∩ (T --- N) ;
§5 measures defined by numerical densities INT V.x128

for, if t ∈ T --- N then u(t) = u0 (t) , therefore u(t) 6= 0 ⇔ u(t) 6= 0 , that

is, t ∈ S ⇔ t ∈ S0 . In other words, ϕS ϕ { N = ϕS0 ϕ { N , whence ϕS = ϕS0
locally θ-almost everywhere, and so the θ-measurability of S 0 implies that
of S .
 Fix any point c ∈ G and let g (resp. g 0 ) be the extension by c of
f S (resp. f S0 ) to T . Observe that g = g 0 locally θ-almost everywhere;
indeed, on the set

T --- (S ∪ S0 ) = (T --- S) ∩ (T --- S0 )

one has g = g 0 = c , whereas on the set

(S ∪ S0 ) ∩ (T --- N) = [S ∩ (T --- N)] ∪ [S0 ∩ (T --- N)]

= S ∩ (T --- N) = S0 ∩ (T --- N)

one has g = f = g 0 , therefore g and g 0 can differ only on the locally

negligible set (S ∪ S0 ) ∩ N . Finally, with respect to θ ,

f S is measurable ⇔ g is measurable
⇔ g 0 is measurable

⇔ f S0 is measurable

(the 1st and 3rd equivalences, by Ch. IV, §5, No. 10, Prop. 15, more precisely,
criterion d 0 ) in the Note for IV.79, `. 7; the 2nd, by Ch. IV, §5, No. 2,
Prop. 6), that is, (∗) holds.
V.43, `. −3. 
“For, uf is the extension by 0 of (uf ) S to T .”
By assumption, u : T → C and f : T → G , where G = F (a Banach
space) or G = R . What does uf mean? The interpretation uf : t 7→
u(t)f (t) (t ∈ T) has consequences:
When G = F is a real Banach space, then u : T → R .
When G = F is a Banach space over C , then u : T → C .
When G = R , let us assume that u : T → R ; for, if f (t) = ±∞ ,
permitting u(t) to be a non-real complex number opens the door to too
many ‘infinities’.
In any case, if u(t) = 0 then u(t)f (t) = 0 (even if G = R and f (t) =
±∞ ), consequently
(
u(t)f (t) for t ∈ S
(uf )(t) =
0 for t ∈ T --- S .
INT V.x129 integration of measures §5

This means that uf is the extension by 0 of (uf ) S to T . Since S is
θ-measurable, it follows that

(i) uf is θ-measurable ⇔ uf S is θ-measurable
(Ch. IV, §5, No. 10, Prop. 15, more precisely, criterion d 0 ) in the Note for
 1
IV.79, `. 7). But uS and  are θ-measurable functions on S (Ch. IV,
uS
§5, No. 3, Th. 1, as generalized in the Note for IV.80, `. −17 to −14), so
 1   
from f S =  · uf S it follows that
uS
 
(ii) uf S is θ-measurable ⇔ f S is θ-measurable
(Ch. IV, §5, No. 3, Cor. 5 of Th. 1, as generalized in the Note just cited).
On the other hand, by Prop. 4,

(iii) f S is θ-measurable ⇔ f is (u · θ)-measurable .
From (i)–(iii) we infer that uf is θ-measurable if and only if f is (u · θ)-
measurable, as asserted.
Suppose, more generally, that the locally θ-integrable function u is
defined only locally θ-almost everywhere in T , and let u 0 be the extension
by 0 of u to T . By the case already considered, f is (u 0 · θ)-measurable
if and only if u0 f is θ-measurable; but u · θ = u0 · θ , and uf = u0 f locally
θ-almost everywhere, whence again f is (u · θ)-measurable if and only if
uf is θ-measurable.
V.44, `. 6, 7.
“The case that u and θ are positive follows at once from Th. 2 of §4,
No. 4.”
Since u is locally θ-integrable, there exists a function u 0 ∈ Lloc
1
(T, θ; C)
such that u = u locally θ-almost everywhere. Then u · θ = u 0 · θ and
0

uf = u0 f locally θ-almost everywhere; and when f is θ-integrable, uf = u 0 f

θ-almost everywhere. We can therefore suppose that u(t) is defined and > 0
at every point t ∈ T , and, in the third assertion, that 0 < u(t) < +∞
for all t ∈ T. Then (I, u) is θ-adapted by the first paragraph of No. 2,
hence all three
R assertions are immediate from the cited Th. 2, inasmuch as
η = u · θ = u(t)εt dθ(t) and f ◦ I = f . {The need for u to be everywhere
finite may be traced back to Prop. 2 of §4, No. 2.}
V.44, `. 10.
“Finally, . . .”
It remains to verify the second assertion, i.e., that when f is essentially
§5 measures defined by numerical densities INT V.x130
R
η-integrable, the equality (6) holds with signifying essential integral (it
will hold
R a fortiori when f is η-integrable and u(t) 6= 0 for all t ∈ T ,
with signifying integral).
V.44, `. 11-13.
“ . . . f is essentially integrable for each of the measures η ij = gi · µj
(i = 1, 2, 3, 4, j = 1, 2, 3, 4), because these measures are 6 |η| .”
Since u is everywhere finite, it is continuous at a point t ∈ T if and
only if every gi is continuous at t ; it follows that the θ-measurability of u
is equivalent to that of all the gi (Ch. IV, §5, No. 1, Def. 1). Similarly, the
θ-integrability of uf for every f ∈ K (T) is equivalent to that of the g i f for
all i (loc. cit., No. 6, Th. 5), which is in turn equivalent to that of the g i f
with respect to µj for all i and j (because 0 6 µ∗j 6 |θ|* ; see, e.g., the
Note for V.10, `. 13, 14). Thus the (positive) measures η ij = gi · µj exist
for all i and j ; moreover, gi · µj 6 |u| · |θ| (that is, ηij 6 |η| ) because, for
every f ∈ K+ (T) , one has
Z Z Z
(gi · µj )(f ) = gi f dµj 6 |u|f dµj 6 |u|f d|θ| = (|u| · |θ|)(f )

by No. 2, Def. 2.
Since f is essentially integrable for η , and 0 6 η ij 6 |η| , it follows
that f is essentially integrable for every η ij ; for, η-negligible sets are ηij -
negligible and so measurability of f for η implies its measurability for
•
the ηij , and, since ηij 6 |η|• , the essential integrability of f for the η ij
follows from ηij (|f |) 6 |η|• (|f |) < +∞ (§1, No. 3, Prop. 9 and Def. 3).
•

V.44, `. 15.
“The formula (6) follows immediately from this.”
Let us consider the case of a Banach space F (of which the case for R
is a simplification).
It will be convenient to write the decompositions (2) in the form

4
X
(2a) u= ai gi
i=1
4
X
(2b) θ= aj µj
j=1

where a1 = 1, a2 = −1, a3 = i, a4 = −i (forgive the double use of “i” as

index and as complex number).
INT V.x131 integration of measures §5

By the preceding Note, each gi is locally integrable for aj µj for all j ;

therefore, for every f ∈ K (T) , f gi is integrable for aj µj for all j , and so
f gi is integrable for θ and one has
Z 4
X Z
f gi dθ = aj f gi dµj
j=1

(by (2b) and the Theorem in the Note for V.10, `. 13, 14). Thus each g i is
locally integrable for θ , and
4
X
gi · θ = aj (gi · µj ) .
j=1

Similarly, since the ai gi are locally integrable for θ , it follows from (2a) and
the above-cited Theorem that
4
X 4
X
u·θ = ai (gi · θ) = ai aj (gi · µj ) ,
i=1 i,j=1

P
4
that is, η = ai aj ηij . Since f is essentially integrable for η and the η ij ,
i,j=1
it follows from Corollary 3 of the above-cited Theorem that
Z X4 Z
(∗) f dη = ai aj f dηij .
i,j=1

But, by the ‘positive case’ already treated, one has the 16 equalities
Z Z
(∗∗) f dηij = gi f dµj (1 6 i, j 6 4).

Finally,
Z Z X X Z


(uf )dθ = ai gi f dθ = ai gi f dθ
i i
X Z X X Z


= ai gi f dθ = ai aj gi f dµj
i i j
X Z X Z
= ai aj gi f dµj = ai aj f dηij
i,j i,j
Z
= f dη

(the 2nd equality, by the additivity of the essential integral; the 4th, by the
above-cited Corollary 3; the 6th, by (∗∗); and the last, by (∗).
§5 measures defined by numerical densities INT V.x132

The case that F = R . Given f : T → R essentially integrable for

η = u · θ , we know by the first assertion R of Th. R 1 that uf is essentially
integrable for θ ; we are to show that f dη = (uf ) dθ . By assumption,
u : T → R is locally integrable for θ (see the convention in the Note for
V.43, `. −3).
By the definition of “ f essentially integrable for η ” (V.9, `. 7–9) there
1
exists a function g ∈R L R (T, Rη) such that f = g locally almost everywhere
for η , and one sets f dη = g dη .
By the case F = R (a simplification of the R BanachR space case), we
know that ug is essentially
R integrable
R for θ and g dη = ug dθ . Thus we
need only show that ug dθ = uf dθ , and for this it suffices to show that
uf = ug locally almost everywhere for θ .
•
R • Let N = {t ∈ T : f (t) 6= g(t) } ; we know that |η| (N) = 0 , that is,
ϕN d|η| = 0 . Citing (5) of Prop. 3, one has
Z Z •
0 = ϕN d|η| = (ϕN |u|) d|θ| ,

thus ϕN |u| = 0 locally almost everywhere for |θ| . Let

P = {t ∈ T : (ϕN |u|)(t) 6= 0 }
= {t ∈ T : t ∈ N and u(t) 6= 0 }
= {t ∈ T : u(t) 6= 0 and f (t) 6= g(t) } ;
we know that P is locally negligible for |θ| . If u(t)f (t) 6= u(t)g(t) then
u(t) 6= 0 (recall the convention 0 · ±∞ = 0 ) and f (t) 6= g(t) , therefore
t ∈ P ; thus
{t ∈ T : (uf )(t) 6= (ug)(t) } ⊂ P ,
and since P is locally negligible for |θ| we conclude that uf = ug locally
almost everywhere for θ .
V.44, `. 16, 17.
“Corollary. — For the measure u · θ to be bounded, it is necessary
and sufficient that u be essentially θ-integrable.”
Taking f to be the constant function 1 (continuous, hence universally
measurable) in Th. 1, we know that u = u1 is essentially integrable for θ
if and only if 1 is essentially integrable for u · θ ; but
1 is essentially integrable for u · θ ⇔ |u · θ| • (1) < +∞
⇔ |u · θ|*(1) < +∞
⇔ u · θ is bounded
(the first equivalence, by §1, No. 3, Prop. 9, the second by §1, No. 1, Prop. 4,
and the third by Ch. IV, §4, No. 7, Prop. 12), whence the assertion.
INT V.x133 integration of measures §5

V.44, `. 18, 19.

“ . . . for ϕA to be locally µ-integrable, it is necessary and sufficient that
A be µ-measurable.”
As announced at the beginning of the chapter, µ denotes any positive
measure on the locally compact space T .
Necessity. If ϕA is locally µ-integrable then it is µ-measurable (No. 1,
Prop. 1), equivalently, A is µ-measurable (Ch. IV, §5, No. 1, Def. 2).
Sufficiency. If A is µ-measurable then, for every f ∈ K (T) , f ϕ A
is µ-integrable (Ch. IV, §5, No. 6, Cor. 3 of Th. 5), thus ϕ A is locally
µ-integrable.
In particular, if θ is a complex measure on T , and µ = |θ| , then
θ-measurability and µ-measurability are equivalent (Ch. IV, §5, No. 1, Def. 1),
as are local θ-integrability and local µ-integrability (No. 1, Def. 1); thus the
assertion is true with µ replaced by any complex measure θ .
V.44, `. −10 to −8.
“For a mapping g of T into a topological space G to be ν-measurable,
it is necessary and sufficient that the restriction of g to A be µ-measurable.”
The assertion holds with µ replaced by any complex measure θ, and ν
by η = ϕA · θ. For, writing u = ϕA in Prop. 4, one has S = A, thus
a mapping g : T → G is measurable for ϕA · θ if and only if g A is
measurable for θ . Writing µ = |θ| , this means that  g is measurable for

|ϕA · θ| = ϕA · |θ| (No. 2, Prop. 2) if and only if g A is measurable for |θ| ;

that is, g is measurable for ν = ϕA · µ if and only if g A is measurable
for µ .
V.44, `. −3 to V.45, `. 1.
“ . . . if two mappings of T into G (resp. F, R ) coincide on A , then for
one of them to be ν-measurable (resp. essentially ν-integrable), it is necessary
and sufficient that the other be so.”
As in the preceding Note, let θ be any complex measure on T , and let
µ = |θ| , η = ϕA · θ , ν = |η| = ϕA · µ .
Suppose, more generally, that f , g : T → G (resp. F , R ) are mappings
such that f (t) = g(t) locally η-almost everywhere in A , i.e., that there
exists a locally η-negligible set N ⊂ A such that f = g on A --- N . Note
that the set T --- A is locally η-negligible; for,
Z • Z •
•
|η| (T --- A) = ϕT -- A d|η| = ϕT --- A ϕA d|θ| = 0 .

Then N ∪ (T --- A) is locally η-negligible, and f = g on its complement

A --- N ; thus f = g locally η-almost everywhere in T .
§5 measures defined by numerical densities INT V.x134

It follows that if f is measurable for η , then so is g ; and

R (when G
R =F
or R ) if f is essentially integrable for η then so is g , and g dη = f dη .
Thus the assertion holds when µ is replaced by θ , and ν by η = ϕ A · θ .
V.45, `. 5, 6.
“ . . . if some extension f to T of the restriction of f to A is essentially
ν-integrable”

Then every extension of f A to T is essentially ν-integrable, and any
two extensions are equal locally ν-almost everywhere (see the preceding
Note).
V.45, `. 7, 8.
“ . . . one then sets
Z Z Z
f dµ = f dµ = fϕA dµ ”
A A

The first
R two members are being defined to be equal to the third, which
is equal to f dν .
V.45, `. 10, 11.
R∗ R•
“ . . . one defines similarly A f dµ and A f dµ .”
 
Let f be any function > 0 on T such that f A = f A , and define
Z •
f dµ = µ• (f ϕA ) ;
A
 
if also f 0 is a function > 0 on T such that f 0 A = f A , we know that
f 0 = f locally almost everywhere R • for ν = Rϕ A · µ (see the Note
R • 0for V.44,
0 •
R`.• −3 to V.45, `. 1), therefore f dν = f dν , whence f ϕA dµ =
R• fϕ A dµ by the preceding argument in the Example. Thus the symbol
A
f dµ is well-defined, i.e., is independent of the choice of f .
R∗
The situation for A f dµ is more delicate. One would like to define it
to be µ∗ (f ϕA ) . Assuming that f and f 0 are any two functions > 0 on T
that coincide with f on A , one knows that fϕ A = f 0 ϕA locally µ-almost
R∗
everywhere; for A f dµ to be well-defined, one must arrange that fϕ A =
f 0 ϕA µ-almost everywhere (Ch. IV, §2, No. 3, Prop. 6). This will be the
case, for example, if A is µ-moderated (§1, No. 3, Lemma). Caution: One
knows that µ• (fϕA ) = ν • (f ) by Prop. 3, but the corresponding equality
for outer measure is not justified. {Problem: If A is µ-moderated, is the
measure ϕA · µ moderated? (answered in the Note for V.48, `. 2, 3)}
Remarks. (i) If A is a µ-measurable set such that µ • (A) < +∞ —
in other words if the function ϕA is essentially µ-integrable (§1, No. 3,
INT V.x135 integration of measures §5

Prop. 9)—then the measure ν = ϕA · µ is bounded (Cor. of Th. 1), hence

moderated (§1, No. 2, Remark 2), and so ν* = ν • (loc. cit., Cor. 2 of
Prop. 7). Then, if f is a function >0 defined
 on a set B ⊃ A , and if f is
any function > 0 on T such that f A = f A , by the Example one has the
equality
ν*(f) = ν • (f ) = µ• (f ϕA ) ,

the last member of which is known to be independent of the particular

extension f of f A .
(ii) If the set A is µ-integrable, that is, if A is µ-measurable and
µ*(A) < +∞ , then A is µ-moderated; as we saw above, the symbol
Z ∗ Z ∗
(∗) f dµ = µ*(f ϕA ) = fϕA dµ
A

is well-defined ( f a function
 > 0 defined on B ⊃ A , and f any function
> 0 on T such that f A = f A ). But fϕA is also µ-moderated, therefore
(§1, No. 2, Prop. 7)

(∗∗) µ*(f ϕA ) = µ• (fϕA ) .

Moreover, µ• (A) = µ*(A) < +∞ , thus A is essentially µ-integrable and

(∗∗∗) ν*(f ) = µ• (f ϕA )

by (i). From (∗∗) and (∗∗∗) we have µ*(f ϕ A ) = ν*(f ) and (∗) may then
be written Z ∗ Z ∗ Z ∗
f dµ = f ϕA dµ = f dν .
A

V.45, `. 13, 14.

“ . . . this is equivalent to saying that, for every compact subset K of T ,
gϕK∩A is µ-integrable.”
The following conditions are equivalent:
a) g is locally ν-integrable;
b) g ϕK is ν-integrable for every compact K ⊂ T ;
c) g ϕK is essentially ν-integrable for every compact K ⊂ T ;
d) gϕK ϕA is essentially µ-integrable for every compact K ⊂ T ;
e) gϕK∩A is µ-integrable for every compact K ⊂ T ;
for, a) ⇔ b) by No. 1, Def. 1; b) ⇔ c) and d) ⇔ e) because compact sets
are (universally) moderated; and c) ⇔ d) by the Example.
§5 measures defined by numerical densities INT V.x136

V.45, `. 17.
“ . . . admitting in M (T) a supremum λ .”
Review: The criterion for this is that, for every f ∈ K + (T) , the (in-
creasing directed) family of real numbers λ α (f ) > 0 admit a finite upper
bound (Ch. II, §2, No. 2, Lemma). That the correspondence f 7→ sup λα (f )
α∈A
defines an additive function on K+ (T) follows from the theorem on mono-
tone limits (GT, Ch. IV, §5, No. 2, Th. 2) and the continuity of addition
in R ; that it is extendible to a linear form λ on K (T) then results from
the fact that every f ∈ K (T) is the difference of two functions in K + (T)
(Ch. II, §2, No. 1, Prop. 2), and, being a positive linear form on K (T) ,
λ is a measure on T (Ch. III, §1, No. 5, Th. 1).
This construction—as a result of which, the space M (T; R) of real
measures on T is a fully lattice-ordered Riesz space (loc. cit., Th. 3)—is
fundamental for the concept of essential integral (cf. §1, No. 4, Prop. 11).
V.45, `. −10.
“It is clear that the condition is necessary.”
Suppose g is locally integrable for λ . Given any h ∈ K + (T) , one
knows that gh is λ-integrable; since λ is the sum of the positive measures
λα and λ − λα , it follows that gh is integrable for every λ α and λα (gh) 6
λ(gh) < +∞ (see, e.g., the Note for V.10, `. 13, 14, the remark preceding
the Theorem), consequently g is locally integrable for λ α and g · λα 6 g · λ ,
whence the assertion.
Thus, by the preceding Note, the measure sup g · λα exists, and it is
α∈A
equal to g · λ by the following computation: for every h ∈ K + (T) ,


sup g · λα (h) = sup (g · λα )(h) = sup λα (gh) = λ(gh) = (g · λ)(h) ;
α∈A α∈A α∈A

the first equality, by Ch. II, §2, No. 2, Lemma; the third, because gh is
(universally) moderated and

λ(gh) = λ• (gh) = sup λ•α (gh) = sup λα (gh)

α∈A α∈A

by §1, No. 4, Prop. 11. {A striking application of the essential integral!}

V.46, `. 1–5.
Corollary.
Let F Pbe the set of all finite subsets J of A , and for each J ∈ F
write µJ = µα ; then (µJ )J∈F is an increasing directed family of positive
α∈J
INT V.x137 integration of measures §5

measures such that µ = sup µJ (§2, No. 1, Remark 1). If J ∈ F and

J∈F
h ∈ K+ (T) one knows (§2, No. 2, Cor. 1 of Prop. 3) that

gh is µJ -integrable ⇔ gh is µα -integrable for every α ∈ J ,

P
in which case µJ (gh) = µα (gh) . It follows that g is locally µJ -inte-
α∈J
grable if and
P only if it is locally µα -integrable for every α ∈ J , in which case
g · µJ = g · µα .
α∈J
Necessity. Suppose g is locally µ-integrable. Since µ J 6 µ for every
J ∈ F , it follows that g is locally integrable for the µ J and that g ·µJ 6 g ·µ
(see the preceding Note), consequently g · µ = sup g · µJ by Prop. 5. By the
J∈F
preceding remarks, g is locally µα -integrable for every α ∈ A , and
X
g · µ = sup g · µα ;
J∈F
α∈J

that is, for every h ∈ K+ (T) ,

X
X X
(g · µ)(h) = sup g · µα (h) = sup (g · µα )(h) = (g · µα )(h) ,
J∈F J∈F
α∈J α∈J α∈A
P
therefore the family (g · µα )α∈A is summable and g · µα = g · µ (§2,
α∈A
No. 1).
Sufficiency. Suppose g is locally µα -integrable
P for every α ∈ A and
the family (g · µα ) is summable, and let ρ = g · µα ; then, for every
α∈A
P
J ∈ F , g is locally integrable for µJ and g · µJ = g · µα . Moreover, the
α∈J
family (g · µJ )J∈F is increasing and
X
ρ = sup g · µJ = sup g · µα
J∈F J∈F
α∈J

(loc. cit., Remark 1). It then follows from P Prop. 5 that g is locally
µ-integrable and that g · µ = sup g · µJ = ρ = g · µα .
J∈F α∈A

V.46, `. 8, 9.
“ . . . if the family (Sα ) is locally countable (Ch. IV, §5, No. 9)”
The concept of local countability (loc. cit., Def. 7) has been defined
only for a set of subsets (not a family of subsets) of a topological space T .
§5 measures defined by numerical densities INT V.x138

Guided by the definition of a ‘locally finite family’ (GT, I, §1, No. 5, Def. 8),
the appropriate definition is as follows: call a family (S α )α∈A of subsets
of T locally countable if, for every t ∈ T , there exists a neighborhood V
of t such that the set of indices {α ∈ A : V ∩ S α 6= ∅ } is countable.
It is obvious that if the family (Sα )α∈A is locally countable in the
foregoing sense, then the set A = {Sα : α ∈ A } is locally countable in the
sense of the cited Def. 7 (Ch. IV, § 5, No. 9). However, the converse is false.
{For example, if the index set A is uncountable and S α = T for all α ∈ A ,
then the set A = {Sα : α ∈ A } = {T} is trivially locally countable, but
the family (Sα )α∈A is not.}
V.46, `. 9, 10.
“ . . . this amounts to saying
 that, for every compact set K in T , the
set of α ∈ A such that gα K is not zero is countable.”
Consider the conditions:
a) The family of functions (gα )α∈A is locally countable, that is, the
family (Sα )α∈A is locally countable;
b) for every compact set K in T , the set of α ∈ A such that K∩S α 6= ∅
is countable; 
c) for every compact set K in T , the set of α ∈ A such that g Kα 6= 0
is countable. 
Since K ∩ Sα 6= ∅ ⇔ gα K 6= 0 , the equivalence b) ⇔ c) is trivial.
a) ⇒ b): The finite covering argument for locally countable sets A ,
following Def. 7 of Ch. IV, §5, No. 9, is readily adapted to the case of
families.
b) ⇒ a): When T is locally compact, every point t ∈ T has a compact
neighborhood K .
Thus the assertion “amounts to saying” entails the local compactness
of T .
V.46, `. 16, 17.
“It is clear that g is µ-measurable (Ch. IV, §5, No. 2, Prop. 4 and No. 4,
Cor. 1 of Th. 2).”
Lemma. If f is a function on the locally compact space T such that
f ϕK is µ-measurable for every compact set K ⊂ T , then f is µ-measurable.
Proof. For every t ∈ T let Vt be a compact (hence µ-integrable) neigh-
borhood of T . Since f ϕVt is by assumption µ-measurable, and
f = f ϕVt on Vt , f is µ-measurable by the Principle of localization (Ch. IV,
§5, No. 2, Prop. 4). ♦
Consider now the given function g . Given any compact set K in T , it
will suffice to show that gϕK is µ-measurable. Let

AK = {α ∈ A : gK K 6= 0 }
INT V.x139 integration of measures §5

(by assumption, a countable set). Since g α ϕK = 0 when α ∈

/ AK , one has
X
gϕK = gα ϕK ;
α∈AK

for, if t ∈
/ K then both sides are equal to 0 at t , whereas if t ∈ K then
X X X
g(t)ϕK (t) = g(t) = gα (t) = gα (t) = gα (t)ϕK (t) .
α∈A α∈AK α∈AK

Since the gα are locally integrable, the gα ϕK are measurable by criterion b)

of No. 1, Prop. 1, thus gϕK is the sum of a countable family of measurable
functions; as the finite subsums are measurable (Ch. IV, §5, No. 3, Cor. 3 of
Th. 1) and gϕK is the supremum of a sequence of such subsums, it follows
that gϕK is indeed measurable (loc. cit., No. 4, Cor. 1 of Th. 2).
V.46, `. 17, 18.
“For g to be locally µ-integrable, it is therefore necessary and sufficient
that µ• (gf ) be finite for every f ∈ K+ (T) .”
Since gf is (universally) moderated, this will follows from criterion c)
of No. 1, Prop. 1 and §1, No. 3, Cor. of Prop. 9 provided gf is shown to be
µ-measurable. We know that the gα f are µ-measurable (even µ-integrable),
but the measurability of gf is not obvious, as application of Th. 1 of Ch. IV,
§5, No. 3 is thwarted by the fact that multiplication in R + is not continuous:
1
· (+∞) 6→ 0 · (+∞) . The following lemmas settle the matter:
n
P
∞
Lemma 1. If c ∈ R+ and ak ∈ R+ (k = 1, 2, 3, . . .) then c ak =
k=1
P
∞
cak .
k=1
Proof. If c = 0 or c = +∞ the equality is obvious. Suppose that
0 < c < +∞ . Write
n
X n
X ∞
X ∞
X
sn = ak , t n = cak , s = ak = sup sn , t = cak = sup tn ;
n n
k=1 k=1 k=1 k=1

we are to show that cs = t .

If ak = +∞ for some k , the assertion reduces to +∞ = +∞ . Assume
that ak < +∞ for all k ; then csn = tn for all n . If the sequence (sn )
is bounded then so is (tn ) and the assertion reduces to c lim s n = lim csn ;
whereas if (sn ) is unbounded then so is (tn ) and the assertion reduces to
+∞ = +∞ . ♦
§5 measures defined by numerical densities INT V.x140

Lemma 2. gf is µ-measurable for every f ∈ K + (T) .

Proof. Let K be a compact set such that f = 0 on { K , and let

AK = {α ∈ A : gα K 6= 0 }
P
(a countable set). As observed in the preceding Note, gϕ K = gα ϕK ,
α∈AK
therefore X
(gϕK )f = gα ϕK f
α∈AK

by Lemma 1, and since f ϕK = f one has

X X
(∗) gf = g(f ϕK ) = (gϕK )f = gα ϕK f = gα f ;
α∈AK α∈AK

thus gf is the sum of a sequence of measurable functions > 0 , hence is

measurable by the argument of the preceding Note. ♦
V.46, `. 18–20.
“ . . . since
P the set of α ∈ A such that gα f 6= 0 is countable, we have
µ• (gf ) = µ• (gα f ) (§1, No. 1, Cor. of Prop. 2).”
α∈A
Applying µ• to the equality (∗) of the preceding Note, one has
X
(∗∗) µ• (gf ) = µ• (gα f )
α∈AK

by the cited Cor.

P of Prop. 2; as gα f = 0 for α ∈
/ AK , this may be written
• •
as µ (gf ) = µ (gα f ) .
α∈A
Incidentally, since gf and the gα f are measurable and moderated, the
relation (∗∗) may also be deduced from §1, No. 2, Prop. 7 and Ch. IV, §5,
No. 6, Cor. 4 of Th. 5.
V.46, `. −4, −3.
“It suffices to apply Prop. 6 to the functions (positive locally almost
everywhere) gn0 = gn+1 − gn .”
{Misprint: In the statement of the Corollary, for ( g α ) read (gn ) .}
Regarding the gn as indexed by n ∈ N, it is not assumed that g 0 ·µ > 0 ,
i.e., that g0 > 0 locally µ-almost everywhere (No. 3, Cor. 3 of Prop. 3).
{Positivity is relevant for sums (GT, IV, §7, No. 5, Remark ) but not for
suprema of increasing families (loc. cit., § 4 No. 2, Prop. 4).}
To avoid dealing with (undefined) differences such as (+∞) − (+∞) ,
we can suppose, on redefining the gn on a countable union of locally µ-
negligible sets, that the gn are everywhere finite-valued (see the Proposition
INT V.x141 integration of measures §5

in the Note for V.39, `. 22–24); their supremum g (which may take on
infinite values) is then altered at most on a locally µ-negligible set—and the
measures gn · µ and g · µ , not at all (No. 3, Cor. 2 of Prop. 3). The relations
gn · µ 6 gn+1 · µ then imply that

0 6 gn+1 · µ − gn · µ = (gn+1 − gn ) · µ

S `. 6, 7), whence g n 6 gn+1 locally µ-almost every-

(see the Note for V.41,
where. Let H = {t ∈ T : gn (t) > gn+1 (t) } (a locally µ-negligible
n∈N
set); redefining the gn to be equal to 0 on H , we can suppose that, for
every n , gn 6 gn+1 at every point of T . Thus, for every t ∈ T , g(t) is
the supremum of the increasing sequence g n (t) of real numbers.
The functions gn0 = gn+1 −gn (n ∈ N) are positive, locally µ-integrable,
and
Xn
gk0 = gn+1 − g0 (n ∈ N) .
k=0

Let F be the set of all finite sets J ⊂ N . For J ∈ F write

X X
sJ = gk0 , νJ = gk0 · µ ;
k∈J k∈J

one knows that sJ · µ = νJ . In particular, writing Jn = {k : 0 6 k 6 n}

for n ∈ N , one has

sJn = gn+1 − g0 , νJn = gn+1 · µ − g0 · µ .

Since the sets Jn are cofinal in the set F ordered by ⊂ , one has
X
gk0 = sup sJ = sup sJn = sup (gn+1 − g0 )
J∈F n∈N n∈N
k∈N

= −g0 + sup gn+1 = −g0 + sup gn

n∈N n∈N

= −g0 + g ,
P
thus g = g0 + gk0 .
k∈N
Necessity. Assuming the sequence (g n · µ) has an upper bound in
M (T) , we are to show that g is locally µ-integrable. Since M (T) is fully
lattice-ordered (Ch. III, §1, No. 5, Th. 3), the measure ρ = sup (gn · µ)
n∈N
exists, as does
sup (gn · µ − g0 · µ) = ρ − g0 · µ .
n∈N
§5 measures defined by numerical densities INT V.x142

Then

ρ − g0 · µ = sup (gn · µ) − g0 · µ = sup (gn+1 · µ) − g0 · µ

n∈N n∈N

= sup (gn+1 · µ − g0 · µ) = sup νJn

n∈N n∈N
X 
= sup νJ = sup gk0 · µ ;
J∈F J∈F
k∈J

this means that the family (gk0 · µ)k∈N is summable, with P sum equal to
0
ρ − g0 · µ . It then follows from Prop. 6 that the function gk = g − g0 is
k∈N
locally µ-integrable, whence so is g , and
X
(g − g0 ) · µ = gk0 · µ = ρ − g0 · µ ;
k∈N

thus g · µ = ρ = sup (gn · µ) .

n∈N
Sufficiency. Conversely, suppose that P g is locally µ-integrable. Then,
from the local µ-integrability of g − g 0 = gk0 , one infers (Prop. 6) that
k∈N
the family of measures (gk0 · µ)k∈N is summable and that
X
gk0 · µ = (g − g0 ) · µ ;
k∈N

this means that

(g − g0 ) · µ = sup νJ = sup νJn = sup (gn+1 · µ − g0 · µ) ,

J∈F n∈N n∈N

whence gn+1 · µ − g0 · µ 6 (g − g0 ) · µ , that is, gn+1 · µ 6 g · µ for all n .

Thus the sequence (gn · µ)n∈N does indeed have an upper bound in M (T) .
V.47, `. 5.
Z
(9) g·ν = (g · λt ) dµ(t) .

When g is continuous the equality is elementary, X can be any locally

compact space and t 7→ λt any scalarly essentially µ-integrable mapping
of T into M+ (X) . For, the measure g · ν of No. 2, Def. 2 then coincides
with the measure
h 7→ ν(gh) (h ∈ K (T))
INT V.x143 integration of measures §5

defined in Ch. III, §1, No. 4. On the other hand, for h ∈ K (T) , the function

t 7→ (g · λt )(h) = λt (gh)

is essentially µ-integrable because gh ∈ K (T) and the mapping t 7→ λ t is

scalarly essentially µ-integrable; it follows that the mapping t 7→ g · λ t is
also scalarly essentially µ-integrable, and, for h ∈ K (T) ,
Z Z Z


h, (g·λt ) dµ(t) = (g·λt )(h) dµ(t) = λt (gh) dµ(t) = ν(gh) = (g·ν)(h) ,

whence (9).
The formula is more memorable when written as a ‘distributive law’
Z Z
0
(9 ) g · λt dµ(t) = (g · λt ) dµ(t) ,

but (9) has the merit that it forces one to check the stringent hypotheses
on X and the mapping t 7→ λt ∈ M+ (X) .
V.47, `. 7, 8.
“ . . . to say that g is locally η-integrable is equivalent to saying that
gϕKn is η-integrable for every n .”
If g is locally η-integrable, then gϕ K is η-integrable for every compact
set K in T by condition b) of No. 1, Prop. 1.
Conversely, suppose gϕKn is η-integrable for every n . If K is any com-
pact set in T then K ⊂ Kn for some n (GT, I, §9, No. 9, Cor. 1 of Prop. 15),
and the η-integrability of gϕKn implies that of gϕK = (gϕKn )ϕK ; in par-
ticular, gϕK is η-measurable for every K , hence g is η-measurable (Lemma
in the Note for V.46, `. 16, 17), therefore g satisfies the cited condition b),
hence is locally η-integrable.
V.47, `. 10. S
“ . . . let H = Hn ”
n
By the preceding Note, g is not λt -integrable if and only if gϕKn is
not λt -integrable for some n , therefore
[
{t ∈ T : g is not λt -integrable } = Hn = H .
n

V.47, `. 10, 11.

“ . . . since Hn is locally µ-negligible for all n (§3, No. 3, Th. 1), the
same is true of H , which establishes the first assertion of the statement.”
§5 measures defined by numerical densities INT V.x144

For each n the function gϕKn is ν-integrable, therefore, by Part a) of

the cited Th. 1, the set Hn is locally µ-negligible, the function

t 7→ λt (gϕKn ) (t ∈
/ Hn )

is essentially µ-integrable, and

Z Z •
gϕKn dν = λt (gϕKn ) dµ(t).

S
The local µ-negligibility of H = Hn then follows from Ch. IV, §5, No. 2,
n
Def. 3, and the first assertion of the statement follows from the fact that

{t ∈ T : g is not λt -integrable } = H

(preceding Note).
V.47, `. 14–16.
“ . . . we have, by Prop. 3 and by Prop. 5 of §3, No. 2,
Z • Z • Z • Z • Z • Z •
h d(g·ν) = (gh) dν = dµ(t) (gh) dλt = dµ(t) h d(g·λt ) . ”

R
Here, ν = λt dµ(t) . As the cited Prop. 3 employs the notation ν
in another sense, it is helpful to restate it: If η is a positive measure on
a locally compact space X and if g is a locally η-integrable function > 0
on X , then Z Z
• •
f d(g · η) = (f g) dη

for every function f > 0 on X .

The displayed equalities, in slow motion: For the first equality, apply
the cited Prop. 3 with η = ν , applicable because g is locally ν-integrable;
for the third equality, apply Prop. 3 with η = λ t , applicable because g is
locally integrable for every λt ; the second equality holds by Part a) of the
cited Prop. 5, applicable since the mapping

t 7→ λt ∈ M+ (X) (t ∈ T)
R
is µ-adequate, with integral ν = λt dµ(t) , and gh is ν-measurable (be-
cause g and h are; see the Lemma at the end of this Note) and ν-moderated
(because g is ν-integrable; in fact, since X is countable at infinity, all func-
tions on X are moderated for every measure on X ).
INT V.x145 integration of measures §5

The cited Prop. 5 shows, in addition, that the function

Z • Z •
t 7→ (gh) dλt = h d(g · λt ) (t ∈ T)

is µ-measurable; when h ∈ K+ (X) , the equality

Z • Z • Z •
dµ(t) h d(g · λt ) = h d(g · ν) = (g · ν)(h) < +∞

shows that the mapping t 7→ g · λt ∈ M+ (X) is scalarly essentially

µ-integrable, and
Z Z • Z •


h, (g · λt ) dµ(t) = hh, g · λt i dµ(t) = hgh, λt i dµ(t)
Z • Z •
= dµ(t) (gh) dλt
Z •
= h d(g · ν) = hh, g · νi ;

R
thus (g · λt ) dµ(t) = g · ν , which is the formula (9).
The following minor point, touched on in the Note for V.46, `. 17, 18,
seems not to have been established explicitly (if it has, I have forgotten
where):
Lemma. Let µ be a measure on a locally compact space X , and let
f, g : X → R be functions that are (with respect to µ ) measurable and are
> 0 locally almost everywhere on X . Then f g is measurable.
Proof. We can suppose that µ > 0 (Ch. IV, §2, No. 1, Def. 1 and §5,
No. 1, Def. 1). Redefining f (x) and g(x) to be 0 on the locally negligible
set where either number is < 0 , we can suppose that f and g are > 0
everywhere on X (Ch. IV, §5, No. 2, Prop. 6).
A function h > 0 on X is measurable if and only if the set

S = {x ∈ X : h(x) > a }

is measurable for every real number a > 0 (loc. cit., No. 5, comment pre-
ceding the Cor. of Prop. 8, plus the fact that S = X when a < 0 ); and since
(for the case a = 0 )
∞
[
{x : h(x) > 0 } = {x : h(x) > 1/n } ,
n=1
§5 measures defined by numerical densities INT V.x146

h is measurable if and only if S is measurable for every real number a > 0

(loc. cit., No. 4, Cor. 2 of Th. 2).
It will therefore suffice to show that the set

A = {x : (f g)(x) > a }

is measurable for every real number a > 0 . Now, (f g)(x) > a if and only
if one of the following three conditions holds:
f (x) = +∞ and g(x) > 0 ;
f (x) > 0 and g(x) = +∞ ;
0 < f (x) < +∞ , 0 < g(x) < +∞ and f (x)g(x) > a .
Thus, writing

B = {x : f (x) = +∞} ∩ {x : g(x) > 0}

C = {x : f (x) > 0} ∩ {x : g(x) = +∞}
D = {x : 0 < f (x) < +∞}, E = {x : 0 < g(x) < +∞} ,

one has
A = B ∪ C ∪ (D ∩ E ∩ {x : f (x)g(x) > a}) ;
the sets B, C, D, E are known to be measurable, so it suffices to show that
the set
D ∩ E ∩ {x : f (x)g(x) > a}
is measurable. This set can be written

D ∩ E ∩ {x : (ϕD f )(x) ϕE g)(x) > a} ,

so it will suffice to show that the set

{x : (ϕD f )(x) ϕE g)(x) > a}

is measurable for every real number a > 0 . Now, the finite-valued functions
ϕD f and ϕE g are measurable; for example, the set

{x : (ϕD g)(x) > a} = D ∩ {x : g(x) > a}

is measurable for every real number a > 0 . One is thus reduced to proving
that the product of measurable functions with values in R is measurable,
and this is known from the comments following Ch. IV, §5, No. 3, Cor. 5 of
Th. 1. ♦
INT V.x147 integration of measures §5

V.47, `. 21, 22.

“ . . . it follows at once from these relations that t 7→ g ·λ t is µ-adequate
(§3, No. 1, Def. 1).”
As shown in the preceding Note, the mapping
(∗) t 7→ g · λt (t ∈ T)
is scalarly essentially µ-integrable, and its integral (in the sense of §3, No. 1)
is given by the ‘distributive law’ (9),
Z
(g · λt ) dµ(t) = g · ν ,
R
where ν = λt dµ(t) . Moreover, for every ν-measurable function h > 0
on X , one has the equality
Z • Z • Z •
h d(g · ν) = dµ(t) h d(g · λt ) ;

in particular, its validity for every lower semi-continuous function h > 0

on X means that the mapping (∗) is µ-pre-adequate (loc. cit., Def. 1).
We are to show that the mapping (∗) is µ 0 -pre-adequate for every pos-
itive measure µ0 6 µ on T . Indeed, the original mapping Rt 7→ λ t is
a fortiori scalarly essentially µ0 -integrable and, writing ν 0 = λt dµ0 (t) ,
obviously ν 0 6 ν . It follows that g is also locally ν 0 -integrable, and the
foregoing argument shows that t 7→ g · λ t is µ0 -pre-adequate (with integral
g · ν 0 ).
Incidentally, if T and X are any locally compact spaces, µ is a positive
measure Ron T , and t 7→ ηt ∈ M+ (X) is a µ-pre-adequate mapping whose
integral ηt dµ(t) is a moderated measure on X, then t 7→ η t is in fact
µ-adequate (see the Note for V.17, `. −6 to −4). {This comment is not in-
tended as a generalization of the present Prop. 7, whose proof makes heavy
use of the countable compactness of X in arranging that g be locally inte-
grable for every λt .}
V.47, `. −2, −1.
“For every function f ∈ K+ (T) we have, by Propositions 2 and 3,
Z • Z • Z •
|g2 | f d|θ1 | = |g2 | f |g1 | d|θ| = |g2 g1 | f d|θ| . ”

In slow motion:
Z • Z • Z •
|g2 | f d|θ1 | = |g2 | f d|g1 · θ| = |g2 | f d(|g1 | · |θ|)
Z • Z •
= |g2 | f |g1 | d|θ| = |g2 g1 | f d|θ| .
§5 measures defined by numerical densities INT V.x148

For these equalities to be valid, g2 can be any complex function and f

any positive function on T ; it is only necessary that g 1 · θ be defined,
i.e., that g1 be locally θ-integrable, equivalently (No. 1, Def. 1) that g 1 be
locally |θ|-integrable, equivalently (since g 1 is θ-measurable) that |g1 | be
locally |θ|-integrable, equivalently that |g 1 | · |θ| be defined (this is implicit in
Prop. 2). The first and last equalities are cosmetic; the 2nd equality holds
by Prop. 2, and the 3rd by Prop. 3.
In particular, when f belongs to K+ (T) , hence is universally moder-
ated, the equality of the first and last members may be written (§1, No. 2,
Prop. 7)
Z ∗ Z ∗
|g2 | f d|θ1 | = |g2 g1 | f d|θ| ,

so that the two members are simultaneously finite or simultaneously in-

finite; thus when, in addition, g2 is θ1 -measurable (equivalently, g2 g1 is
θ-measurable), g2 f is θ1 -integrable (i.e., the left member is finite) if and
only if g2 g1 f is θ-integrable (i.e., the right member is finite); in other words,
assuming g1 locally θ-integrable, the following conditions are equivalent:
a) g2 is locally θ1 -integrable—which implies that g2 is θ1 -measurable
(No. 1, Prop. 1), hence g2 g1 is θ-measurable;
b) g2 g1 is locally θ-integrable—which implies that g 2 g1 is θ-measurable,
hence g2 is θ1 -measurable.
V.48, `. 2, 3.
“ . . . by Th. 1 we have ”
Since f is universally moderated,
R • in applyingR Th. 1 (twice) the word
“essentially” can be omitted, and replaced by .

Addendum: The answer to a question posed in the Note for V.45,

`. 10, 11 is as follows:
Example. There exist a locally compact space T , a positive measure µ
on T , and a µ-moderated subset A of T such that the measure ϕ A · µ is
not moderated.
Lemma. If µ a positive measure on a locally compact space T such that
µ = sup µn , where (µn ) is an increasing sequence of positive measures, and
n
if A is a µ-measurable
S subset of T such that Supp µ n ⊂ A for all n
(for example A = Supp µn , which is a Borel set, hence is universally
n
measurable), then ϕA · µ = µ .
Proof. Write Sn = Supp µn . Since µn 6 µ , one knows that A is
µn -measurable for all n , therefore ϕA is locally µn -integrable (No. 3, Ex-
ample). Thus, for every g ∈ K+ (T) , ϕA g is µn -integrable and, µn -almost
INT V.x149 integration of measures §5

everywhere, one has ϕA g = ϕSn ϕA g = ϕSn g = g , whence

(ϕA · µn )(g) = µ•n (ϕA g) = µ•n (g) = µn (g) ,

that is, ϕA · µn = µn . Since µ• = sup µ•n (§1, No. 4, Prop. 11) it follows
n
that, for all g ∈ K+ (T) ,

(ϕA · µ)(g) = µ• (ϕA g) = sup µ•n (ϕA g) = sup µ•n (g) = µ• (g) = µ(g) ,
n n

thus ϕA · µ = µ . ♦
The example. Let T and µ be as in Ch. V, §2, Exer. 4 (on p. V.95).
By part a) of the exercise, µ is not moderated but µ = sup µn , where
n
(µn ) is anSincreasing sequence of positive measures with finite support. The
set A = Supp µn is the union of a sequence of compact sets, hence is
n
universally moderated, but ϕA · µ = µ by the Lemma, thus ϕA · µ is not
moderated.

Coda. This is a good place to stop—the Notes are getting too fre-
quent and too long, the document a bloated 586 pages, and the complicated
arrangement of the proof of the Lebesgue-Nikodym theorem in No. 5 exacer-
bates the trend; the reader who has gotten this far will probably find filling
in the gaps easier than studying how I filled them in.
As epitome of the formal beauty of Bourbaki’s treatment of integration,
I nominate the formula
p
|µ + iν| = µ2 + ν 2

of No. 9 ( µ and ν real measures) with its audacious right member.