2 MATH 605, HW 6 SOLUTIONS

implies f 1 (0) is closed. To go the other way, suppose f is not bounded. Then there must
exist xn 2 X with |xn | ! 0 but f (xn ) = 1. Choose y with f (y) = 1. Then f (y xn ) = 0 so
(y xn ) 2 f 1 (0) but (y xn ) ! y 2 / f 1 (0).
18.) Let X be a normed vector space.
(a) If M is a closed subspace and x 2 / M , then M +Rx is closed: The Hahn-Banach theorem
gives (see theorem 5.8 a.) a bounded linear functional f with f (x) 6= 0 and f |M = 0. We
need to show that if a sequence yj 2 M + Rx converges to y 2 X, then y 2 M + Rx. Set
zj = yj (f (yj )/f (x)) x so that zj 2 M +Rx and f (zj ) = 0; therefore, we have zj 2 M . Since
f is continuous and M is closed, the zj ’s converge to z 2 M . Finally, y = z + (f (y)/f (x)) x
so it is in M + Rx.
(b) Every finite dimensional subspace of X is closed: This is done by induction. Obviously
the 0-dimensional subspace {0} is closed. Suppose now that every n-dimensional subspace
is closed and M is an (n + 1)-dimensional subspace. We can find a basis x1 , . . . , xn+1 . Then
the span N of x1 , . . . , xn is closed and so (a) implies that so is M = N + Rxn+1 .
Folland’s Real Analysis; Chapter 6:
1). When does equality hold in Minkowski’s inequality (6.5 in the text)? I’ll do this for
1 < p < 1. Assume that f , g, and f + g are not zero a.e. We had two “” in the proof so
we just have to see when these are “=.” The first was
|f + g|p  (|f | + |g|)|f + g|p 1
.
We get equality here when |f + g| = |f | + |g| a.e. - i.e., when f and g have the same sign
a.e. The second was Holder’s inequality applies to |f | |f + g|p 1 and to |g| |f + g|p 1 . We get
equality here when |f |p = c1 |f + g|p and |g|p = c2 |f + g|p a.e. for some constants c1 , c2 6= 0.
Finally, we see that equality holds i↵ f is a positive multiple of g. (The case p = 1 is quite
di↵erent, but I’ll leave it for you.)
5). Suppose 0 < p < q < 1. (The arguments are similar so we will just do (a).)
(a) Lp is not contained in Lq i↵ X has sets of arbitrarily small positive measure: If En ⇢ X
has 0 < µ(En ) < 1/n, then fn = [µ(En )] 1/p En has |fn |p = 1 but
Z
|fn |q = [µ(En )] q/p
µ(En ) = [µ(En )](p q)/p
n(q p)/p
.
In fact, we can assume that the sets En are disjoint. Since q > p, we can choose ↵ so that
P P
n ↵ < 1 but n(q p)/p P ↵
= 1. It is now easy to see (from the monotone convergence
theorem) that the function n ↵ fn is in Lp but not in Lq .
To go the other way, suppose f 2 Lp \ Lq . The same is true of f {|f | 1} since |f |q (x) 
|f |p (x) when |f |(x)  1, so we can assume |f | is never between 0 and 1. Define the sets
En = {x | 2n  |f |(x) < 2n+1 } for n 0. For any r > 0, we have
X Z X
2n r m(En )  fr  2(n+1)r m(En ) .
n n
In particular, f 2 Lp implies that m(En ) ! 0 but f 2
/ Lq implies that infinitely many of the
m(En )’s must not be 0.
6). Follow the hint.
 MATH 605, HW 7 SOLUTIONS

Folland’s Real Analysis; Chapter 6:

7.) If f 2 Lp \ L1 , then show: ||f ||1 = limq!1 ||f ||q . We can assume ||f ||p 6= 0 and
||f ||1 6= 0 (otherwise f = 0 a.e. so it’s obvious).

(a) Proposition 6.10 gives ||f ||q  ||f ||p/q 1 p/q

p ||f ||1 . Since limr!0 ar = 1 and limr!0 a1 r =
a for any a > 0, this gives “.”
(b) Suppose 0 < a < ||f ||1 so Ea = {x | |f |(x) > a} has m(Ea ) > 0. Therefore ||f ||q
a [m(Ea )]1/q which goes to a as q ! 1. This gives “ .”
9.) Suppose 1  p < 1.

(a) Show ||f fn ||p ! 0 implies fn ! f in measure: This is immediate since

||f fn ||pp
m({x | |f (x) fn (x)| > a) 
.
ap
(b) Show that fn ! f in measure and |fn |  g 2 Lp for all n implies ||f fn ||p ! 0:
Argue as for problem 2.34 (hmwk 4).
10.) See problem 2.20 from earlier problem set (hmwk 4).
Folland’s Real Analysis; Chapter 3:
2.) If ⌫ is a signed measure, show E is ⌫-null i↵ |⌫|(E) = 0: |⌫(E)| = ⌫+ (E) + ⌫ (E) = 0
implies that if F ⇢ E then |⌫(F )|  |⌫|(F ) = 0 (since ⌫+ and ⌫ are monotone). To go
the other way, let P [ N be Hahn decomposition for ⌫ so ⌫+ (E) = ⌫(P \ E) = 0 and
⌫ (E) = ⌫(N \ E) = 0 (since E is null) so |⌫|(E) = 0.
3.) (a) and (b) are trivial.
R
(c) Show Rthat if ⌫ is a signed measure, then |⌫|(E) = sup{ E f d⌫ | |f |  1}: Well, it’s
trivial that E f d⌫  (sup |f |) |⌫|(E). To get the other inequality, just choose f = P N
where P [ N is the Hahn decomposition for ⌫. Then
Z
( P N ) d⌫ = ⌫(E \ P ) ( 1)⌫(E \ N ) = ⌫+ (E) + ⌫ (E) = |⌫|(E) .
E

4.) Suppose ⌫ is a signed measure and , µ are (positive) measures with ⌫ = µ. Show
that ⌫+ and µ ⌫ : The arguments are very similar so I’ll just show one. Let P [ N
be the Hahn decomposition for and E any measurable set. Then
⌫+ (E) = ⌫(P \ E) = (P \ E) µ(P \ E)  (P \ E)  (E) .
Here the first equality is the definition of ⌫+ and the second is just that ⌫ = µ. The first
 used µ 0 and the last was just monotonicity for measures.
Professor Minicozzi, Fall 2002.
1
2 MATH 605, HW 7 SOLUTIONS

5.) If ⌫1 , ⌫2 are signed measures that both miss 1 or 1, then |⌫1 + ⌫2 |  |⌫1 | + |⌫2 |:
Just write ⌫1 + ⌫2 = (⌫1+ + ⌫2+ ) (⌫1 + ⌫2 ) and then use the previous problem.
8.) Show that ⌫ << µ i↵ |⌫| << µ: Problem 2 gives one direction and the other way is
obvious.
11.) (a) follows immediately from Corollary 3.6. To get (b), apply Corollary 3.6 to f then
use the convergence to get some > 0 which works for all but finitely many of the fn ’s; use
part (a) to take care of these finitely many.
 MATH 605, HW 8 SOLUTIONS

Folland’s Real Analysis; Chapter 3:

22.) Suppose f 2 L1 (Rn ) and ||f ||L1 6= 0. Show there exist C, R > 0 so (the maxi-
mal
R
function) Hf (x) C |x| n for |x| > R: Since ||f ||L1 6= 0, we can choose R > 0 so
BR/2 (0) |f | > ↵ > 0 for some ↵. For |x| > R, we get
R
|f |
BR (0) n n
Hf (x) = sup Ar |f |(x) A2|x| |f |(x) > ↵2 |x| /C1 .
r>0 C1 (2|x|)n
23.) It’s obvious that Hf  H ⇤ f since we are taking the supremum over a larger set of
balls. To get H ⇤ f (x)  2n Hf (x): Note that if x 2 BR (y), then BR (y) ⇢ B2R (x) and the
volume of B2R (x) is 2n times the volume of BR (y).
24.) Suppose f 2 L1 is continuous at x; show that x 2 Lf (the Lebesgue set): Continuity
implies that for ✏ > 0, we get > 0 so that |x y| < implies |f (x) f (y)| < ✏. Hence, if
r < , then
Ar |f f (x)|(x) < ✏ .
It follows that Ar |f f (x)|(x) goes to 0 as r goes to zero; i.e., x 2 Lf .
25.) For (a), just apply 3.22 with f = E.

28.) Just follow the hints!

30.) Construct an increasing function on R which is discontinuous at the rationals: First,
P
enumerate the rationals as q1 , q2 , . . . and define fj = 2 j [qj ,1) . Then set f = fj (this
exists by the monotone convergence theorem). It’s easy to see that this works.
b 0 R 0
33.) Show that if F is increasing, then F (b) F (a) a F (t) dt: Well, we know that F
exists a.e. and F 0 0. By L-R-N and the Lebesgue di↵erentation theorems, we know that
Z b
F (b) F (a) = µF ([a, b]) = ([a, b]) + F 0 (t) dt ,
a
where is singular. We just have to show that ([a, b]) 0. This is easy since there is a set
A with m(A) = 0 so that ([a, b]) = ([a, b] \ A) but this implies
([a, b] \ A) = µF ([a, b] \ A) m([a, b] \ A) = µF ([a, b] \ A) 0.

37.) Show that |F (x) F (y)|  M |x y| for all x, y i↵ F is a.c. and |F 0 |  M :

(a) Theorem 3.35(c) gives the “if.”
(b) Suppose now that |F (x) F (y)|  M |x y| for all x, y. It’s obvious that |F 0 |  M
whenever it P
is defined (since all di↵erence
P
quotients are 
P
M ). To see that F is a.c., just
note that if (bj aj ) < ✏/M , then |F (bj ) F (aj )|  M (bj aj ) < ✏.

Professor Minicozzi, Fall 2002.

1