22

Approximation Theorems and Convolutions

22.1 Density Theorems

In this section, (X, M, µ) will be a measure space A will be a subalgebra of
M.
Notation 22.1 Suppose (X, M, µ) is a measure space and A ⊂ M is a sub-
algebra of M. Let S(A) denote those simple functions φ : X → C such that
φ−1 ({λ}) ∈ A for all λ ∈ C and let Sf (A, µ) denote those φ ∈ S(A) such that
µ(φ 6= 0) < ∞.
p P
Remark 22.2. For φ ∈ Sf (A, µ) and p ∈ [1, ∞), |φ| = z6=0 |z|p 1{φ=z} and
hence Z
p
X
|φ| dµ = |z|p µ(φ = z) < ∞ (22.1)
z6=0

so that Sf (A, µ) ⊂ L (µ). Conversely if φ ∈ S(A)∩Lp (µ), then from Eq. (22.1)
p

it follows that µ (φ = z) < ∞ for all z 6= 0 and therefore µ (φ 6= 0) < ∞. Hence

we have shown, for any 1 ≤ p < ∞,

Sf (A, µ) = S(A) ∩ Lp (µ).

Lemma 22.3 (Simple Functions are Dense). The simple functions,

Sf (M, µ), form a dense subspace of Lp (µ) for all 1 ≤ p < ∞.

Proof. Let {φn }∞ n=1 be the simple functions in the approximation Theo-
rem 18.42. Since |φn | ≤ |f | for all n, φn ∈ Sf (M, µ) and
p p
|f − φn |p ≤ (|f | + |φn |) ≤ 2p |f | ∈ L1 (µ) .

Therefore, by the dominated convergence theorem,

Z Z
lim |f − φn |p dµ = lim |f − φn |p dµ = 0.
n→∞ n→∞
420 22 Approximation Theorems and Convolutions

The goal of this section is to find a number of other dense subspaces of

Lp (µ) for p ∈ [1, ∞). The next theorem is the key result of this section.

Theorem 22.4 (Density Theorem). Let p ∈ [1, ∞), (X, M, µ) be a mea-

sure space and M be an algebra of bounded F — valued (F = R or F = C)
measurable functions such that
1. M ⊂ Lp (µ, F) and σ (M ) = M.
2. There exists ψk ∈ M such that ψk → 1 boundedly.
3. If F = C we further assume that M is closed under complex conjugation.
Then to every function f ∈ Lp (µ, F) , there exists φn ∈ M such that
limn→∞ kf − φn kLp (µ) = 0, i.e. M is dense in Lp (µ, F) .

Proof. Fix k ∈ N for the moment and let H denote those bounded M —
measurable functions, f : X → F, for which there exists {φn }∞n=1 ⊂ M such
that limn→∞ kψk f − φn kLp (µ) = 0. A routine check shows H is a subspace
of ∞ (M, F) such that 1 ∈ H, M ⊂ H and H is closed under complex
conjugation if F = C. Moreover, H is closed under bounded convergence.
To see this suppose fn ∈ H and fn → f boundedly. Then, by the dominated
convergence theorem, limn→∞ kψk (f − fn )kLp (µ) = 0.1 (Take the dominating
function to be g = [2C |ψk |]p where C is a constant bounding all of the
{|fn |}∞
n=1 .) We may now choose φn ∈ M such that kφn − ψk fn kLp (µ) ≤ n
1

then

lim sup kψk f − φn kLp (µ) ≤ lim sup kψk (f − fn )kLp (µ)
n→∞ n→∞
+ lim sup kψk fn − φn kLp (µ) = 0 (22.2)
n→∞

which implies f ∈ H. An application of Dynkin’s Multiplicative System The-

orem 18.51 if F = R or Theorem 18.52 if F = C now shows H contains all
bounded measurable functions on X.
Let f° ∈ Lp (µ) be given.
° The dominated convergence theorem implies
limk→∞ °ψk 1{|f |≤k} f − f °Lp (µ) = 0. (Take the dominating function to be
g = [2C |f |]p where C is a bound on all of the |ψk | .) Using this and what we
have just proved, there exists φk ∈ M such that
° °
°ψk 1{|f |≤k} f − φk ° p ≤ 1 .
L (µ) k
The same line of reasoning used in Eq. (22.2) now implies limk→∞ kf − φk kLp (µ) =
0.
1
It is at this point that the proof would break down if p = ∞.
 22.1 Density Theorems 421

Definition 22.5. Let (X, τ ) be a topological space and µ be a measure on

BX = σ (τ ) . A locally R integrable function is a Borel measurable function
f : X → C such that K |f | dµ < ∞ for all compact subsets K ⊂ X. We will
write L1loc (µ) for the space of locally integrable functions. More generally we
say f ∈ Lploc (µ) iff k1K f kLp (µ) < ∞ for all compact subsets K ⊂ X.

Definition 22.6. Let (X, τ ) be a topological space. A K-finite measure on

X is Borel measure µ such that µ (K) < ∞ for all compact subsets K ⊂ X.

Lebesgue measure on R is an example of a K-finite measure while counting

measure on R is not a K-finite measure.

Example 22.7. Suppose that µ is a K-finite measure on BRd . An application of

p d
Theorem 22.4 shows ¡ Cdc (R,
¢ C) is dense in L (R , BRd , µ; C). To apply
¡ Theorem
¢
22.4, let M := Cc R , C and ψk (x) := ψ (x/k) where ψ ∈ Cc Rd , C with
ψ ¡(x) =
¡ 1 in ¢¢
a neighborhood of 0. The proof is completed by showing σ (M ) =
σ Cc Rd , C = BRd , which follows directly from Lemma 18.57.
We may also give a more down to earth proof as follows. Let x0 ∈ Rd , R >
1/n
0, A := B (x0 , R)c and fn (x) := dA (x) . Then fn ∈ M and fn → 1B(x0 ,R)
as n → ∞ which shows 1B(x0 ,R) is σ (M )-measurable, i.e. B (x0 , R) ∈ σ (M ) .
Since x0 ∈ Rd and R > 0 were arbitrary, σ (M ) = BRd .

More generally we have the following result.

Theorem 22.8. Let (X, τ ) be a second countable locally compact Hausdorff

space and µ : BX → [0, ∞] be a K-finite measure. Then Cc (X) (the space
of continuous functions with compact support) is dense in Lp (µ) for all p ∈
[1, ∞). (See also Proposition 25.23 below.)

Proof. Let M := Cc (X) and use Item 3. of Lemma 18.57 to find functions
ψk ∈ M such that ψk → 1 to boundedly as k → ∞. The result now follows
from an application of Theorem 22.4 along with the aid of item 4. of Lemma
18.57.

Exercise 22.1. Show that BC (R, C) is not dense in L∞ (R, BR , m; C). Hence
the hypothesis that p < ∞ in Theorem 22.4 can not be removed.

Corollary 22.9. Suppose X ⊂ Rn is an open set, BX is the Borel σ — algebra

on X and µ be a K-finite measure on (X, BX ) . Then Cc (X) is dense in Lp (µ)
for all p ∈ [1, ∞).

Corollary 22.10. Suppose that X is a compact subset of Rn and µ is a finite

measure on (X, BX ), then polynomials are dense in Lp (X, µ) for all 1 ≤ p <
∞.

Proof. Consider X to be a metric space with usual metric induced

from Rn . Then X is a locally compact separable metric space and therefore
422 22 Approximation Theorems and Convolutions

Cc (X, C) = C(X, C) is dense in Lp (µ) for all p ∈ [1, ∞). Since, by the domi-
nated convergence theorem, uniform convergence implies Lp (µ) — convergence,
it follows from the Weierstrass approximation theorem (see Theorem 8.34 and
Corollary 8.36 or Theorem 12.31 and Corollary 12.32) that polynomials are
also dense in Lp (µ).

Lemma 22.11. Let (X, τ ) be a second countable locally compact Hausdorff

space and µ : BX → [0, ∞] be a K-finite measure on X. If h ∈ L1loc (µ) is a
function such that Z
f hdµ = 0 for all f ∈ Cc (X) (22.3)
X

then h(x) = 0 for µ — a.e. x. (See also Corollary 25.26 below.)

Proof. Let dν(x) = |h(x)| dx, then ν is a K-finite measure on X and hence
Cc (X) is dense in L1 (ν) by Theorem 22.8. Notice that
Z Z
f · sgn(h)dν = f hdµ = 0 for all f ∈ Cc (X). (22.4)
X X

Let {Kk }∞
k=1 be a sequence of compact sets such that Kk ↑ X as in Lemma
11.23. Then 1Kk sgn(h) ∈ L1 (ν) and therefore there exists fm ∈ Cc (X) such
that fm → 1Kk sgn(h) in L1 (ν). So by Eq. (22.4),
Z Z
ν(Kk ) = 1Kk dν = lim fm sgn(h)dν = 0.
X m→∞ X
R
Since Kk ↑ X as k → ∞, 0 = ν(X) = X |h| dµ, i.e. h(x) = 0 for µ — a.e. x.
As an application of Lemma 22.11 and Example 12.34, we will show that
the Laplace transform is injective.

Theorem 22.12 (Injectivity of the Laplace Transform). For f ∈

L1 ([0, ∞), dx), the Laplace transform of f is defined by
Z ∞
Lf (λ) := e−λx f (x)dx for all λ > 0.
0

If Lf (λ) := 0 then f (x) = 0 for m -a.e. x.

Proof. Suppose that f ∈ L1 ([0, ∞), dx) such that Lf (λ) ≡ 0. Let g ∈
C0 ([0, ∞), R) and ε > 0 be given. By Example 12.34 we may choose {aλ }λ>0
such that # ({λ > 0 : aλ 6= 0}) < ∞ and
X
|g(x) − aλ e−λx | < ε for all x ≥ 0.
λ>0

Then
 22.1 Density Theorems 423
¯Z ∞ ¯ ¯¯Z ∞ à X
! ¯
¯
¯ ¯ ¯ ¯
¯ g(x)f (x)dx¯=¯ g(x) − a e−λx
f (x)dx¯
¯ ¯ ¯ λ
¯
0 0 λ>0
Z ∞¯¯ ¯
X ¯
¯ ¯
≤ ¯g(x) − aλ e−λx ¯ |f (x)| dx ≤ εkf k1 .
0 ¯ ¯
λ>0
R∞
Since ε > 0 is arbitrary, it follows that 0 g(x)f (x)dx = 0 for all g ∈
C0 ([0, ∞), R). The proof is finished by an application of Lemma 22.11.
Here is another variant of Theorem 22.8.

Theorem 22.13. Let (X, d) be a metric space, τd be the topology on X gen-

erated by d and BX = σ(τd ) be the Borel σ — algebra. Suppose µ : BX → [0, ∞]
is a measure which is σ — finite on τd and let BCf (X) denote the bounded
continuous functions on X such that µ(f 6= 0) < ∞. Then BCf (X) is a dense
subspace of Lp (µ) for any p ∈ [1, ∞).

Proof. Let Xk ∈ τd be open sets such that Xk ↑ X and µ(Xk ) < ∞ and
let
ψk (x) = min(1, k · dXkc (x)) = φk (dXkc (x)),
see Figure 22.1 below. It is easily verified that M := BCf (X) is an algebra,

y 1

0.75

0.5

0.25

0
0 0.5 1 1.5 2

Fig. 22.1. The plot of φn for n = 1, 2, and 4. Notice that φn → 1(0,∞) .

ψk ∈ M for all k and ψk → 1 boundedly as k → ∞. Given V ∈ τ and

k, n ∈ N,let
fk,n (x) := min(1, n · d(V ∩Xk )c (x)).
Then {fk,n 6= 0} = V ∩ Xk so fk,n ∈ BCf (X). Moreover

lim lim fk,n = lim 1V ∩Xk = 1V

k→∞ n→∞ k→∞

which shows V ∈ σ (M ) and hence σ (M ) = BX . The proof is now completed

by an application of Theorem 22.4.
424 22 Approximation Theorems and Convolutions

Exercise 22.2. (BRUCE: Should drop this exercise.) Suppose that (X, d) is
a metric space, µ is a measure on BX := σ(τd ) which is finite on bounded
measurable subsets of X. Show BCb (X, R), defined in Eq. (19.26), is dense in
Lp (µ) . Hints: let ψk be as defined in Eq. (19.27) which incidentally may be
used to show σ (BCb (X, R)) = σ (BC(X, R)) . Then use the argument in the
proof of Corollary 18.55 to show σ (BC(X, R)) = BX .
Theorem 22.14. Suppose p ∈ [1, ∞), A ⊂ M is an algebra such that σ(A) =
M and µ is σ — finite on A. Then Sf (A, µ) is dense in Lp (µ). (See also Remark
25.7 below.)
Proof. Let M := Sf (A, µ). By assumption there exits Xk ∈ A such that
µ(Xk ) < ∞ and Xk ↑ X as k → ∞. If A ∈ A, then Xk ∩ A ∈ A and
µ (Xk ∩ A) < ∞ so that 1Xk ∩A ∈ M. Therefore 1A = limk→∞ 1Xk ∩A is σ (M )
— measurable for every A ∈ A. So we have shown that A ⊂ σ (M ) ⊂ M
and therefore M = σ (A) ⊂ σ (M ) ⊂ M, i.e. σ (M ) = M. The theorem
now follows from Theorem 22.4 after observing ψk := 1Xk ∈ M and ψk → 1
boundedly.
Theorem 22.15 (Separability of Lp — Spaces). Suppose, p ∈ [1, ∞), A ⊂
M is a countable algebra such that σ(A) = M and µ is σ — finite on A. Then
Lp (µ) is separable and
X
D={ aj 1Aj : aj ∈ Q + iQ, Aj ∈ A with µ(Aj ) < ∞}

is a countable dense subset.

Proof. It is left to reader to check D is dense in Sf (A, µ) relative to the
Lp (µ) — norm. The proof is then complete since Sf (A, µ) is a dense subspace
of Lp (µ) by Theorem 22.14.
Pn
Example 22.16. The collection of functions of the form φ = k=1 ck 1(ak ,bk ]
with ak , bk ∈ Q and ak < bk are dense in L (R, BR , m; C) and Lp (R, BR , m; C)
p

is separable for any p ∈ [1, ∞). To prove this simply apply Theorem 22.14 with
A being the algebra on R generated by the half open intervals `n(a, b] ∩ R with
a < b and a, b ∈ Q∪ {±∞} , i.e. A consists of sets of the form k=1 (ak , bk ]∩R,
where ak , bk ∈ Q∪ {±∞} .
Exercise 22.3. Show L∞ ([0, 1] , BR , m; C) is not separable. Hint: Suppose
Γ is a dense subset of L∞ ([0, 1] , BR , m; C) and for λ ∈ (0, 1) , let fλ (x) :=
1[0,λ] (x) . For each λ ∈ (0, 1) , choose gλ ∈ Γ such that kfλ − gλ k∞ < 1/2 and
then show the map λ ∈ (0, 1) → gλ ∈ Γ is injective. Use this to conclude that
Γ must be uncountable.
Corollary 22.17 (Riemann Lebesgue Lemma). Suppose that f ∈ L1 (R, m),
then Z
lim f (x)eiλx dm(x) = 0.
λ→±∞ R
 22.1 Density Theorems 425
Pn
Proof. By Example 22.16, given ε > 0 there exists φ = k=1 ck 1(ak ,bk ]
with ak , bk ∈ R such that
Z
|f − φ|dm < ε.
R

Notice that
Z Z X
n
φ(x)eiλx dm(x) = ck 1(ak ,bk ] (x)eiλx dm(x)
R R k=1

X Z bk
n n
X
iλx
= ck e dm(x) = ck λ−1 eiλx |bakk
k=1 ak k=1
n
X ¡ ¢
= λ−1 ck eiλbk − eiλak → 0 as |λ| → ∞.
k=1

Combining these two equations with

¯Z ¯ ¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯ ¯ ¯
¯ f (x)eiλx dm(x)¯ ≤ ¯ (f (x) − φ(x)) eiλx dm(x)¯ + ¯ φ(x)eiλx dm(x)¯
¯ ¯ ¯ ¯ ¯ ¯
R R R
Z ¯Z ¯
¯ ¯
≤ |f − φ|dm + ¯¯ φ(x)eiλx dm(x)¯¯
R R
¯Z ¯
¯ ¯
≤ ε + ¯¯ φ(x)e dm(x)¯¯
iλx
R

we learn that
¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
¯
lim sup ¯ f (x)e dm(x)¯¯ ≤ ε + lim sup
iλx ¯ φ(x)eiλx dm(x)¯ = ε.
¯ ¯
|λ|→∞ R |λ|→∞ R

Since ε > 0 is arbitrary, this completes the proof of the Riemann Lebesgue
lemma.

Corollary 22.18. Suppose A ⊂ M is an algebra such that σ(A) = M and µ

is σ — finite on A. Then for every B ∈ M such that µ(B) < ∞ and ε > 0
there exists D ∈ A such that µ(B4D) < ε. (See also Remark 25.7 below.)

Proof. By Theorem 22.14, there exists a collection,

R {Ai }ni=1 , of pairwise
disjoint
Pn subsets of A and λi ∈ R such that X |1B − f | dµ < ε where f =
n
i=1 i Ai . Let A0 := X \ ∪i=1 Ai ∈ A then
λ 1
426 22 Approximation Theorems and Convolutions
Z n Z
X
|1B − f | dµ = |1B − f | dµ
X i=0 Ai
n
"Z Z #
X
= µ (A0 ∩ B) + |1B − λi | dµ + |1B − λi | dµ
i=1 Ai ∩B Ai \B

Xn
= µ (A0 ∩ B) + [|1 − λi | µ (B ∩ Ai ) + |λi | µ (Ai \ B)] (22.5)
i=1
Xn
≥ µ (A0 ∩ B) + min {µ (B ∩ Ai ) , µ (Ai \ B)} (22.6)
i=1

where the last equality is a consequence of the fact that 1 ≤ |λi | + |1 − λi | .

Let ½
0 if µ (B ∩ Ai ) < µ (Ai \ B)
αi =
1 if µ (B ∩ Ai ) ≥ µ (Ai \ B)
Pn
and g = i=1 αi 1Ai = 1D where

D := ∪ {Ai : i > 0 & αi = 1} ∈ A.

Equation (22.5) with λi replaced by αi and f by g implies

Z n
X
|1B − 1D | dµ = µ (A0 ∩ B) + min {µ (B ∩ Ai ) , µ (Ai \ B)} .
X i=1
R
The latter expression, by Eq. (22.6), is bounded by X |1B − f | dµ < ε and
therefore, Z
µ(B4D) = |1B − 1D | dµ < ε.
X

Remark 22.19. We have to assume that µ(B) < ∞ as the following example
shows. Let X = R, M = B, µ = m, A be the algebra generated by half open
intervals of the form (a, b], and B = ∪∞
n=1 (2n, 2n + 1]. It is easily checked that
for every D ∈ A, that m(B∆D) = ∞.

22.2 Convolution and Young’s Inequalities

Throughout this section we will be solely concerned with
¡ d — ¢dimensional
Lebesgue measure, m, and we will simply write Lp for Lp Rd , m .

Definition 22.20 (Convolution). Let f, g : Rd → C be measurable func-

tions. We define Z
f ∗ g(x) = f (x − y)g(y)dy (22.7)
Rd
 22.2 Convolution and Young’s Inequalities 427

whenever the integral is defined, i.e. either f (x − ·) g (·) ∈ L1 (Rd , m) or

f (x − ·) g (·) ≥ 0. Notice that the condition that f (x − ·) g (·) ∈ L1 (Rd , m)
is equivalent to writing |f | ∗ |g| (x) < ∞. By convention, if the integral in Eq.
(22.7) is not defined, let f ∗ g(x) := 0.
Notation 22.21 Given a multi-index α ∈ Zd+ , let |α| = α1 + · · · + αd ,
Yd µ ¶α Yd µ ¶αj
α ∂ ∂
xα := xj j , and ∂xα = := .
j=1
∂x j=1
∂xj

For z ∈ Rd and f : Rd → C, let τz f : Rd → C be defined by τz f (x) = f (x − z).

Remark 22.22 (The Significance of Convolution).
1. Suppose that f, g ∈ L1 (m) are positive functions and let µ be the measure
¡ ¢2
on Rd defined by
dµ (x, y) := f (x) g (y) dm (x) dm (y) .
Then if h : R → [0, ∞] is a measurable function we have
Z Z
h (x + y) dµ (x, y) = h (x + y) f (x) g (y) dm (x) dm (y)
(Rd )2 (Rd )2
Z
= h (x) f (x − y) g (y) dm (x) dm (y)
(Rd )2
Z
= h (x) f ∗ g (x) dm (x) .
Rd

In other words, this shows the measure (f ∗ g) m is the same as S∗ µ where

S (x, y) := x + y. In probability lingo, the distribution of a sum of two “in-
dependent” (i.e. product measure) random variables is the the convolution
of the individual distributions.
P
2. Suppose that L = |α|≤k aα ∂ α is a constant coefficient differential oper-
ator and suppose that we can solve (uniquely) the equation Lu = g in the
form Z
u(x) = Kg(x) := k(x, y)g(y)dy
Rd
where k(x, y) is an “integral kernel.” (This is a natural sort of assumption
since, in view of the fundamental theorem of calculus, integration is the
inverse operation to differentiation.) Since τz L = Lτz for all z ∈ Rd , (this
is another way to characterize constant coefficient differential operators)
and L−1 = K we should have τz K = Kτz . Writing out this equation then
says
Z
k(x − z, y)g(y)dy = (Kg) (x − z) = τz Kg(x) = (Kτz g) (x)
Rd
Z Z
= k(x, y)g(y − z)dy = k(x, y + z)g(y)dy.
Rd Rd
428 22 Approximation Theorems and Convolutions

Since g is arbitrary we conclude that k(x − z, y) = k(x, y + z). Taking

y = 0 then gives
k(x, z) = k(x − z, 0) =: ρ(x − z).
We thus find that Kg = ρ ∗ g. Hence we expect the convolution operation
to appear naturally when solving constant coefficient partial differential
equations. More about this point later.
Proposition 22.23. Suppose p ∈ [1, ∞], f ∈ L1 and g ∈ Lp , then f ∗ g(x)
exists for almost every x, f ∗ g ∈ Lp and
kf ∗ gkp ≤ kf k1 kgkp .
Proof. This follows directly from Minkowski’s inequality for integrals,
Theorem 21.27.
Proposition 22.24. Suppose that p ∈ [1, ∞), then τz : Lp → Lp is an iso-
metric isomorphism and for f ∈ Lp , z ∈ Rd → τz f ∈ Lp is continuous.
Proof. The assertion that τz : Lp → Lp is an isometric isomorphism
follows from translation invariance of Lebesgue measure and the fact that
τ−z ◦ τz = id. For the continuity assertion, observe that
kτz f − τy f kp = kτ−y (τz f − τy f )kp = kτz−y f − f kp

from which it follows that it is enough to show τz f → f in Lp as z → 0 ∈ Rd .

When f ∈ Cc (Rd ), τz f → f uniformly and since the K := ∪|z|≤1 supp(τz f ) is
compact, it follows by the dominated convergence theorem that τz f → f in
Lp as z → 0 ∈ Rd . For general g ∈ Lp and f ∈ Cc (Rd ),
kτz g − gkp ≤ kτz g − τz f kp + kτz f − f kp + kf − gkp
= kτz f − f kp + 2 kf − gkp
and thus
lim sup kτz g − gkp ≤ lim sup kτz f − f kp + 2 kf − gkp = 2 kf − gkp .
z→0 z→0

Because Cc (R ) is dense in Lp , the term kf − gkp may be made as small as

d

we please.
Exercise 22.4. Compute the operator norm, kτz − IkL(Lp (m)) , of τz − I and
use this to show z ∈ Rd → τz ∈ L (Lp (m)) is not continuous.
Definition 22.25. Suppose that (X, τ ) is a topological space and µ is a mea-
sure on BX = σ(τ ). For a measurable function f : X → C we define the
essential support of f by
suppµ (f ) = {x ∈ X : µ({y ∈ V : f (y) 6= 0}}) > 0 ∀ neighborhoods V of x}.
(22.8)
Equivalently, x ∈/ suppµ (f ) iff there exists an open neighborhood V of x such
that 1V f = 0 a.e.
 22.2 Convolution and Young’s Inequalities 429

It is not hard to show that if supp(µ) = X (see Definition 21.41) and

f ∈ C(X) then suppµ (f ) = supp(f ) := {f 6= 0} , see Exercise 22.7.

Lemma 22.26. Suppose (X, τ ) is second countable and f : X → C is a mea-

surable function and µ is a measure on BX . Then X := U \ suppµ (f ) may
be described as the largest open set W such that f 1W (x) = 0 for µ — a.e. x.
Equivalently put, C := suppµ (f ) is the smallest closed subset of X such that
f = f 1C a.e.

Proof. To verify that the two descriptions of suppµ (f ) are equivalent,

suppose suppµ (f ) is defined as in Eq. (22.8) and W := X \ suppµ (f ). Then

W = {x ∈ X : ∃ τ 3 V 3 x such that µ({y ∈ V : f (y) 6= 0}}) = 0}

= ∪ {V ⊂o X : µ (f 1V 6= 0) = 0}
= ∪ {V ⊂o X : f 1V = 0 for µ — a.e.} .

So to finish the argument it suffices to show µ (f 1W 6= 0) = 0. To to this let

U be a countable base for τ and set

Uf := {V ∈ U : f 1V = 0 a.e.}.

Then it is easily seen that W = ∪Uf and since Uf is countable

X
µ (f 1W 6= 0) ≤ µ (f 1V 6= 0) = 0.
V ∈Uf

Lemma 22.27. Suppose f, g, h : Rd → C are measurable functions and as-

sume that x is a point in Rd such that |f |∗|g| (x) < ∞ and |f |∗(|g| ∗ |h|) (x) <
∞, then
1. f ∗ g(x) = g ∗ f (x)
2. f ∗ (g ∗ h)(x) = (f ∗ g) ∗ h(x)
3. If z ∈ Rd and τz (|f | ∗ |g|)(x) = |f | ∗ |g| (x − z) < ∞, then

τz (f ∗ g)(x) = τz f ∗ g(x) = f ∗ τz g(x)

/ suppm (f ) + suppm (g) then f ∗ g(x) = 0 and in particular,

4. If x ∈

suppm (f ∗ g) ⊂ suppm (f ) + suppm (g)

where in defining suppm (f ∗ g) we will use the convention that “f ∗ g(x) 6=

0” when |f | ∗ |g| (x) = ∞.

Proof. For item 1.,

Z Z
|f | ∗ |g| (x) = |f | (x − y) |g| (y)dy = |f | (y) |g| (y − x)dy = |g| ∗ |f | (x)
Rd Rd
430 22 Approximation Theorems and Convolutions

where in the second equality we made use of the fact that Lebesgue measure
invariant under the transformation y → x − y. Similar computations prove
all of the remaining assertions of the first three items of the lemma. Item
4. Since f ∗ g(x) = f˜ ∗ g̃(x) if f = f˜ and g = g̃ a.e. we may, by replacing
f by f 1suppm (f ) and g by g1suppm (g) if necessary, assume that {f 6= 0} ⊂
suppm (f ) and {g 6= 0} ⊂ suppm (g). So if x ∈ / (suppm (f ) + suppm (g)) then
x∈ / ({f 6= 0} + {g 6= 0}) and for all y ∈ Rd , either x − y ∈/ {f 6= 0} or y ∈/
{g 6= 0} . That is to say either x − y ∈ {f = 0} or y ∈ {g = 0} and hence
f (x − y)g(y)
³ = 0 for all y and therefore
´ f ∗ g(x) = 0. This shows that f ∗ g = 0
d
on R \ suppm (f ) + suppm (g) and therefore
³ ´
Rd \ suppm (f ) + suppm (g) ⊂ Rd \ suppm (f ∗ g),

i.e. suppm (f ∗ g) ⊂ suppm (f ) + suppm (g).

Remark 22.28. Let A, B be closed sets of Rd , it is not necessarily true that

A + B is still closed. For example, take

A = {(x, y) : x > 0 and y ≥ 1/x} and B = {(x, y) : x < 0 and y ≥ 1/|x|} ,

then every point of A + B has a positive y - component and hence is not zero.
On the other hand, for x > 0 we have (x, 1/x) + (−x, 1/x) = (0, 2/x) ∈ A + B
for all x and hence 0 ∈ A + B showing A + B is not closed. Nevertheless if
one of the sets A or B is compact, then A + B is closed again. Indeed, if A is
compact and xn = an + bn ∈ A + B and xn → x ∈ Rd , then by passing to a
subsequence if necessary we may assume limn→∞ an = a ∈ A exists. In this
case
lim bn = lim (xn − an ) = x − a ∈ B
n→∞ n→∞

exists as well, showing x = a + b ∈ A + B.

Proposition 22.29. Suppose that p, q ∈ [1, ∞] and p and q are conjugate

exponents, f ∈ Lp and g ∈ Lq , then f ∗ g ∈ BC(Rd ), kf ∗ gk∞ ≤ kf kp kgkq
and if p, q ∈ (1, ∞) then f ∗ g ∈ C0 (Rd ).

Proof. The existence of f ∗g(x) and the estimate |f ∗ g| (x) ≤ kf kp kgkq for
all x ∈ Rd is a simple consequence of Holders inequality and the translation in-
variance of Lebesgue measure. In particular this shows kf ∗ gk∞ ≤ kf kp kgkq .
By relabeling p and q if necessary we may assume that p ∈ [1, ∞). Since

kτz (f ∗ g) − f ∗ gku = kτz f ∗ g − f ∗ gku

≤ kτz f − f kp kgkq → 0 as z → 0

it follows that f ∗ g is uniformly continuous. Finally if p, q ∈ (1, ∞), we learn

from Lemma 22.27 and what we have just proved that fm ∗ gm ∈ Cc (Rd )
where fm = f 1|f |≤m and gm = g1|g|≤m . Moreover,
 22.2 Convolution and Young’s Inequalities 431

kf ∗ g − fm ∗ gm k∞ ≤ kf ∗ g − fm ∗ gk∞ + kfm ∗ g − fm ∗ gm k∞
≤ kf − fm kp kgkq + kfm kp kg − gm kq
≤ kf − fm kp kgkq + kf kp kg − gm kq → 0 as m → ∞

showing, with the aid of Proposition 12.23, f ∗ g ∈ C0 (Rd ).

Theorem 22.30 (Young’s Inequality). Let p, q, r ∈ [1, ∞] satisfy

1 1 1
+ =1+ . (22.9)
p q r

If f ∈ Lp and g ∈ Lq then |f | ∗ |g| (x) < ∞ for m — a.e. x and

kf ∗ gkr ≤ kf kp kgkq . (22.10)

In particular L1 is closed under convolution. (The space (L1 , ∗) is an example

of a “Banach algebra” without unit.)

Remark 22.31. Before going to the formal proof, let us first understand Eq.
(22.9) by the following scaling argument. For λ > 0, let fλ (x) := f (λx), then
after a few simple change of variables we find

kfλ kp = λ−d/p kf k and (f ∗ g)λ = λd fλ ∗ gλ .

Therefore if Eq. (22.10) holds for some p, q, r ∈ [1, ∞], we would also have

kf ∗ gkr = λd/r k(f ∗ g)λ kr ≤ λd/r λ kfλ kp kgλ kq = λ(d+d/r−d/p−d/q) kf kp kgkq

for all λ > 0. This is only possible if Eq. (22.9) holds.

Proof. By the usual sorts of arguments, we may assume f and g are

positive functions. Let α, β ∈ [0, 1] and p1 , p2 ∈ (0, ∞] satisfy p−1 −1
1 +p2 +r
−1
=
1. Then by Hölder’s inequality, Corollary 21.3,
Z h i
f ∗ g(x) = f (x − y)(1−α) g(y)(1−β) f (x − y)α g(y)β dy
Rd
µZ ¶1/r µZ ¶1/p1
≤ f (x − y)(1−α)r g(y)(1−β)r dy f (x − y)αp1 dy ×
Rd Rd
µZ ¶1/p2
βp2
× g(y) dy
Rd
µZ ¶1/r
= f (x − y) (1−α)r
g(y) (1−β)r
dy kf kα β
αp1 kgkβp2 .
Rd

Taking the rth power of this equation and integrating on x gives

432 22 Approximation Theorems and Convolutions
Z µZ ¶
kf ∗ gkrr ≤ f (x − y)(1−α)r g(y)(1−β)r dy dx · kf kα β
αp1 kgkβp2
Rd Rd
(1−α)r (1−β)r αr βr
= kf k(1−α)r kgk(1−β)r kf kαp1 kgkβp2 . (22.11)

Let us now suppose, (1 − α)r = αp1 and (1 − β)r = βp2 , in which case Eq.
(22.11) becomes,
kf ∗ gkrr ≤ kf krαp1 kgkrβp2
which is Eq. (22.10) with

p := (1 − α)r = αp1 and q := (1 − β)r = βp2 . (22.12)

So to finish the proof, it suffices to show p and q are arbitrary indices in

[1, ∞] satisfying p−1 + q −1 = 1 + r−1 . If α, β, p1 , p2 satisfy the relations above,
then
r r
α= and β =
r + p1 r + p2
and
1 1 1 1 1 r + p1 1 r + p2
+ = + = +
p q αp1 αp2 p1 r p2 r
1 1 2 1
= + + =1+ .
p1 p2 r r
Conversely, if p, q, r satisfy Eq. (22.9), then let α and β satisfy p = (1 − α)r
and q = (1 − β)r, i.e.
r−p p r−q q
α := = 1 − ≤ 1 and β = = 1 − ≤ 1.
r r r r
Using Eq. (22.9) we may also express α and β as
1 1
α = p(1 − ) ≥ 0 and β = q(1 − ) ≥ 0
q p
and in particular we have shown α, β ∈ [0, 1]. If we now define p1 := p/α ∈
(0, ∞] and p2 := q/β ∈ (0, ∞], then
1 1 1 1 1 1
+ + =β +α +
p1 p2 r q p r
1 1 1
= (1 − ) + (1 − ) +
q p r
µ ¶
1 1
=2− 1+ + =1
r r
as desired.
Theorem 22.32 R (Approximate δ — functions). Let p ∈ [1, ∞], φ ∈
L1 (Rd ), a := Rd f (x)dx, and for t > 0 let φt (x) = t−d φ(x/t). Then
 22.2 Convolution and Young’s Inequalities 433

1. If f ∈ Lp with p < ∞ then φt ∗ f → af in Lp as t ↓ 0.

2. If f ∈ BC(Rd ) and f is uniformly continuous then kφt ∗ f − af k∞ → 0
as t ↓ 0.
3. If f ∈ L∞ and f is continuous on U ⊂o Rd then φt ∗ f → af uniformly
on compact subsets of U as t ↓ 0.
Proof. Making the change of variables y = tz implies
Z Z
φt ∗ f (x) = f (x − y)φt (y)dy = f (x − tz)φ(z)dz
Rd Rd

so that
Z
φt ∗ f (x) − af (x) = [f (x − tz) − f (x)] φ(z)dz
d
ZR
= [τtz f (x) − f (x)] φ(z)dz. (22.13)
Rd

Hence by Minkowski’s inequality for integrals (Theorem 21.27), Proposition

22.24 and the dominated convergence theorem,
Z
kφt ∗ f − af kp ≤ kτtz f − f kp |φ(z)| dz → 0 as t ↓ 0.
Rd

Item 2. is proved similarly. Indeed, form Eq. (22.13)

Z
kφt ∗ f − af k∞ ≤ kτtz f − f k∞ |φ(z)| dz
Rd

which again tends to zero by the dominated convergence theorem because

limt↓0 kτtz f − f k∞ = 0 uniformly in z by the uniform continuity of f.
Item 3. Let BR = B(0, R) be a large ball in Rd and K @@ U, then
sup |φt ∗ f (x) − af (x)|
x∈K
¯Z ¯ ¯¯Z ¯
¯
¯ ¯ ¯ ¯
≤ ¯¯ [f (x − tz) − f (x)] φ(z)dz ¯¯ + ¯ [f (x − tz) − f (x)] φ(z)dz ¯
BR ¯ c
BR ¯
Z Z
≤ |φ(z)| dz · sup |f (x − tz) − f (x)| + 2 kf k∞ |φ(z)| dz
BR x∈K,z∈BR c
BR
Z
≤ kφk1 · sup |f (x − tz) − f (x)| + 2 kf k∞ |φ(z)| dz
x∈K,z∈BR |z|>R

so that using the uniform continuity of f on compact subsets of U,

Z
lim sup sup |φt ∗ f (x) − af (x)| ≤ 2 kf k∞ |φ(z)| dz → 0 as R → ∞.
t↓0 x∈K |z|>R

See Theorem 8.15 if Folland for a statement about almost everywhere

convergence.
434 22 Approximation Theorems and Convolutions

Exercise 22.5. Let ½

e−1/t if t > 0
f (t) =
0 if t ≤ 0.
Show f ∈ C ∞ (R, [0, 1]).

Lemma 22.33. There Rexists φ ∈ Cc∞ (Rd , [0, ∞)) such that φ(0) > 0,
supp(φ) ⊂ B̄(0, 1) and Rd φ(x)dx = 1.

Proof. Define h(t) = f (1 − t)f (t + 1) where f is as in Exercise 22.5.

Then
R h ∈ Cc∞ (R, [0, 1]), supp(h) ⊂ [−1, 1] and h(0) = e−2 > 0. Define c =
2
Rd
h(|x|2 )dx. Then φ(x) = c−1 h(|x| ) is the desired function.
The reader asked to prove the following proposition in Exercise 22.9 below.

Proposition 22.34. Suppose that f ∈ L1loc (Rd , m) and φ ∈ Cc1 (Rd ), then
f ∗ φ ∈ C 1 (Rd ) and ∂i (f ∗ φ) = f ∗ ∂i φ. Moreover if φ ∈ Cc∞ (Rd ) then
f ∗ φ ∈ C ∞ (Rd ).

Corollary 22.35 (C ∞ — Uryhson’s Lemma). Given K @@ U ⊂o Rd , there

exists f ∈ Cc∞ (Rd , [0, 1]) such that supp(f ) ⊂ U and f = 1 on K.

Proof. Let φ be as in Lemma 22.33, φt (x) = t−d φ(x/t) be as in Theorem

d
22.32, d be the standard metric on
© R and ε = d(K, U cª). Since K is compact
c d
and U is closed, ε > 0. Let Vδ = x ∈ R : d(x, K) < δ and f = φε/3 ∗ 1Vε/3 ,
then
supp(f ) ⊂ supp(φε/3 ) + Vε/3 ⊂ V̄2ε/3 ⊂ U.
Since V̄2ε/3 is closed and bounded, f ∈ Cc∞ (U ) and for x ∈ K,
Z Z
f (x) = 1d(y,K)<ε/3 · φε/3 (x − y)dy = φε/3 (x − y)dy = 1.
Rd Rd

The proof will be finished after the reader (easily) verifies 0 ≤ f ≤ 1.

Here is an application of this corollary whose proof is left to the reader,
Exercise 22.10.

Lemma 22.36 (Integration by Parts). Suppose f and g are measur-

able functions on Rd such that t → f (x1 , . . . , xi−1 , t, xi+1 , . . . , xd ) and t →
g(x1 , . . . , xi−1 , t, xi+1 , . . . , xd ) are continuously differentiable functions on R
∂f
for each fixed x = (x1 , . . . , xd ) ∈ Rd . Moreover assume f · g, ∂x i
· g and
∂g
f · ∂xi are in L1 (Rd , m). Then
Z Z
∂f ∂g
· gdm = − f· dm.
Rd ∂xi Rd ∂xi

With this result we may give another proof of the Riemann Lebesgue
Lemma.
 22.2 Convolution and Young’s Inequalities 435

Lemma 22.37 (Riemann Lebesgue Lemma). For f ∈ L1 (Rd , m) let

Z
fˆ(ξ) := (2π)−d/2 f (x)e−iξ·x dm(x)
Rd
° °
° °
be the Fourier transform of f. Then fˆ ∈ C0 (Rd ) and °fˆ° ≤ (2π)−d/2 kf k1 .
∞
(The choice of the normalization factor, (2π)−d/2 , in fˆ is for later conve-
nience.)

Proof. The fact that fˆ is continuous is a simple application of the domi-

nated convergence theorem. Moreover,
¯ ¯ Z
¯ˆ ¯
¯f (ξ)¯ ≤ |f (x)| dm(x) ≤ (2π)−d/2 kf k1
Rd

so it only remains to see that fˆ(ξ) → 0 as |ξ| → ∞. First suppose that

Pd ∂2
f ∈ Cc∞ (Rd ) and let ∆ = d
j=1 ∂x2 be the Laplacian on R . Notice that
j
∂ −iξ·x 2
∂xj e = −iξj e−iξ·x and ∆e−iξ·x = − |ξ| e−iξ·x . Using Lemma 22.36 re-
peatedly,
Z Z Z
2k
∆k f (x)e−iξ·x dm(x) = f (x)∆kx e−iξ·x dm(x) = − |ξ| f (x)e−iξ·x dm(x)
Rd Rd Rd
2k
= −(2π)d/2 |ξ| fˆ(ξ)
for any k ∈ N. Hence
¯ ¯
¯ ¯ −2k ° °
°∆k f ° → 0
(2π)d/2 ¯fˆ(ξ)¯ ≤ |ξ| 1

as |ξ| → ∞ and fˆ ∈ C0 (Rd ). Suppose that f ∈ L1 (m) and °fk ∈ Cc∞ d

° (R ) is
°ˆ ˆ°
a sequence such that limk→∞ kf − fk k1 = 0, then limk→∞ °f − fk ° = 0.
∞
Hence fˆ ∈ C0 (Rd ) by an application of Proposition 12.23.
Corollary 22.38. Let X ⊂ Rd be an open set and µ be a Radon measure on
BX .
1. Then Cc∞ (X) is dense in Lp (µ) for all 1 ≤ p < ∞.
2. If h ∈ L1loc (µ) satisfies
Z
f hdµ = 0 for all f ∈ Cc∞ (X) (22.14)
X

then h(x) = 0 for µ — a.e. x.

Proof. Let f ∈ Cc (X), φ be as in Lemma 22.33, φt be as in Theorem
22.32 and set ψt := φt ∗ (f 1X ) . Then by Proposition 22.34 ψt ∈ C ∞ (X) and
by Lemma 22.27 there exists a compact set K ⊂ X such that supp(ψt ) ⊂ K
for all t sufficiently small. By Theorem 22.32, ψt → f uniformly on X as t ↓ 0
436 22 Approximation Theorems and Convolutions

1. The dominated convergence theorem (with dominating function being

kf k∞ 1K ), shows ψt → f in Lp (µ) as t ↓ 0. This proves Item 1., since
Theorem 22.8 guarantees that Cc (X) is dense in Lp (µ).
2. Keeping the same notation as above, the dominated convergence theorem
(with dominating function being kf k∞ |h| 1K ) implies
Z Z Z
0 = lim ψt hdµ = lim ψt hdµ = f hdµ.
t↓0 X X t↓0 X

The proof is now finished by an application of Lemma 22.11.

22.2.1 Smooth Partitions of Unity

We have the following smooth variants of Proposition 12.16, Theorem 12.18

and Corollary 12.20. The proofs of these results are the same as their contin-
uous counterparts. One simply uses the smooth version of Urysohn’s Lemma
of Corollary 22.35 in place of Lemma 12.8.
Proposition 22.39 (Smooth Partitions of Unity for Compacts). Sup-
pose that X is an open subset of Rd , K ⊂ X is a compact set and U = {Uj }nj=1
is an open cover of K. Then there exists a smooth (i.e. hj ∈ C ∞ (X, [0, 1]))
n
partition of unity {hj }j=1 of K such that hj ≺ Uj for all j = 1, 2, . . . , n.

Theorem 22.40 (Locally Compact Partitions of Unity). Suppose that

X is an open subset of Rd and U is an open cover of X. Then there exists a
smooth partition of unity of {hi }N
i=1 (N = ∞ is allowed here) subordinate to
the cover U such that supp(hi ) is compact for all i.
Corollary 22.41. Suppose that X is an open subset of Rd and U =
{Uα }α∈A ⊂ τ is an open cover of X. Then there exists a smooth partition
of unity of {hα }α∈A subordinate to the cover U such that supp(hα ) ⊂ Uα for
all α ∈ A. Moreover if Ūα is compact for each α ∈ A we may choose hα so
that hα ≺ Uα .

22.3 Exercises
Exercise 22.6. Let (X, τ ) be a topological space, µ a measure on BX =
σ(τ ) and f : X → C be a measurable function. Letting ν be the measure,
dν = |f | dµ, show supp(ν) = suppµ (f ), where supp(ν) is defined in Definition
21.41).
Exercise 22.7. Let (X, τ ) be a topological space, µ a measure on BX = σ(τ )
such that supp(µ) = X (see Definition 21.41). Show suppµ (f ) = supp(f ) =
{f 6= 0} for all f ∈ C(X).
 22.3 Exercises 437

Exercise 22.8. Prove the following strong version of item 3. of Proposition

10.52, namely to every pair of points, x0 , x1 , in a connected open subset V
of Rd there exists σ ∈ C ∞ (R, V ) such that σ(0) = x0 and σ(1) = x1 . Hint:
First choose a continuous path γ : [0, 1] → V such that γ (t) = x0 for t near 0
and γ (t) = x1 for t near 1 and then use a convolution argument to smooth γ.

Exercise 22.9. Prove Proposition 22.34 by appealing to Corollary 19.43.

Exercise 22.10 (Integration by Parts). Suppose that (x, y) ∈ R × Rd−1 →

f (x, y) ∈ C and (x, y) ∈ R × Rd−1 → g(x, y) ∈ C are measurable functions
such that for each fixed y ∈ Rd , x → f (x, y) and x → g(x, y) are continuously
differentiable. Also assume f · g, ∂x f · g and f · ∂x g are integrable relative to
Lebesgue measure on R × Rd−1 , where ∂x f (x, y) := dt d
f (x + t, y)|t=0 . Show
Z Z
∂x f (x, y) · g(x, y)dxdy = − f (x, y) · ∂x g(x, y)dxdy. (22.15)
R×Rd−1 R×Rd−1

(Note: this result and Fubini’s theorem proves Lemma 22.36.)

Hints: Let ψ ∈ Cc∞ (R) be a function which is 1 in a neighborhood of
0 ∈ R and set ψε (x) = ψ(εx). First verify Eq. (22.15) with f (x, y) replaced
by ψε (x)f (x, y) by doing the x — integral first. Then use the dominated con-
vergence theorem to prove Eq. (22.15) by passing to the limit, ε ↓ 0.

Exercise 22.11. Let µ be a finite measure on BRd , then D := span{eiλ·x :

λ ∈ Rd } is a dense subspace of Lp (µ) for all 1 ≤ p < ∞. Hints: By Theorem
22.8, Cc (Rd ) is a dense subspace of Lp (µ). For f ∈ Cc (Rd ) and N ∈ N, let
X
fN (x) := f (x + 2πN n).
n∈Zd

Show fN ∈ BC(Rd ) and x → fN (N x) is 2π — periodic, so by Exercise 12.13,

x → fN (N x) can be approximated uniformly by trigonometric polynomials.
p
Use this fact to conclude that fN ∈ D̄L (µ) . After this show fN → f in Lp (µ).

Exercise 22.12. Suppose that µ and ν are two finite measures on Rd such
that Z Z
eiλ·x dµ(x) = eiλ·x dν(x) (22.16)
Rd Rd

for all λ ∈ Rd . Show µ = ν.

Hint: Perhaps the easiest way to do this is to use Exercise 22.11 with the
measure µ being replaced by µ + ν. Alternatively,
R use the Rmethod of proof
of Exercise 22.11 to show Eq. (22.16) implies Rd f dµ(x) = Rd f dν(x) for all
f ∈ Cc (Rd ) and then apply Corollary 18.58.

R 22.13. Again let µ be a finite measure on BRd . Further assume that

Exercise
CM := Rd eM |x| dµ(x) < ∞ for all M ∈ (0, ∞). Let P(Rd ) be the space of
P p
polynomials, ρ(x) = |α|≤N ρα xα with ρα ∈ C, on Rd . (Notice that |ρ(x)| ≤
438 22 Approximation Theorems and Convolutions

CeM |x| for some constant C = C(ρ, p, M ), so that P(Rd ) ⊂ Lp (µ) for all
1 ≤ p < ∞.) Show P(Rd ) is dense in Lp (µ) for all 1 ≤ p < ∞. Here is a
possible outline.
Outline: Fix a λ ∈ Rd and let fn (x) = (λ · x)n /n! for all n ∈ N.
1. Use calculus to verify supt≥0 tα e−Mt = (α/M )α e−α for all α ≥ 0 where
0
(0/M ) := 1. Use this estimate along with the identity
³ ´
pn pn pn pn pn
|λ · x| ≤ |λ| |x| = |x| e−M|x| |λ| eM|x|

to find an estimate on kfn kp .

P
2. Use your estimate on kfn kp to show ∞
n=0 kfn kp < ∞ and conclude
° °
° N °
° iλ·(·) X n °
lim °e − i fn ° = 0.
N→∞ ° °
n=0 p

3. Now finish by appealing to Exercise 22.11.

Exercise 22.14. Again let µ be a finiteR measure on BRd but now assume
ε|x|
there exists an ε > 0 such that C := R R d e dµ(x) < ∞. Also let q > 1 and
h ∈ Lq (µ) be a function such that Rd h(x)xα dµ(x) = 0 for all α ∈ Nd0 . (As
mentioned in Exercise 22.14, P(Rd ) ⊂ Lp (µ) for all 1 ≤ p < ∞, so x → h(x)xα
is in L1 (µ).) Show h(x) = 0 for µ— a.e. x using the following outline.
n
Outline: Fix a λ ∈ Rd , let fn (x) = (λ · x) /n! for all n ∈ N, and let
p = q/(q − 1) be the conjugate exponent to q.
1. Use calculus to verify supt≥0 tα e−εt = (α/ε)α e−α for all α ≥ 0 where
(0/ε)0 := 1. Use this estimate along with the identity
³ ´
pn pn pn pn pn
|λ · x| ≤ |λ| |x| = |x| e−ε|x| |λ| eε|x|

to find an estimate on kfn kp .

2. Use your estimate on kfn kp to show there exists δ > 0 such that
P∞
n=0 kf k < ∞ when |λ| ≤ δ and conclude for |λ| ≤ δ that eiλ·x =
p
Pn∞ p n
L (µ)- n=0 i fn (x). Conclude from this that
Z
h(x)eiλ·x dµ(x) = 0 when |λ| ≤ δ.
Rd
R
3. Let λ ∈ Rd (|λ| not necessarily small) and set g(t) := Rd eitλ·x h(x)dµ(x)
for t ∈ R. Show g ∈ C ∞ (R) and
Z
g (n) (t) = (iλ · x)n eitλ·x h(x)dµ(x) for all n ∈ N.
Rd
 22.3 Exercises 439

4. Let T = sup{τ ≥ 0 : g|[0,τ ] ≡ 0}. By Step 2., T ≥ δ. If T < ∞, then

Z
0 = g (n) (T ) = (iλ · x)n eiT λ·x h(x)dµ(x) for all n ∈ N.
Rd

Use Step 3. with h replaced by eiT λ·x h(x) to conclude

Z
g(T + t) = ei(T +t)λ·x h(x)dµ(x) = 0 for all t ≤ δ/ |λ| .
Rd

This violates the definition of T and therefore T = ∞ and in particular

we may take T = 1 to learn
Z
h(x)eiλ·x dµ(x) = 0 for all λ ∈ Rd .
Rd

5. Use Exercise 22.11 to conclude that

Z
h(x)g(x)dµ(x) = 0
Rd

for all g ∈ Lp (µ). Now choose g judiciously to finish the proof.