MATH 211B HOMEWORK

ERIN PEARSE

1. (10/7) Find the eigenvalues and eigenfunctions of

(?1 )

L[u] = u00 + 2 u,

R+

with u(a) = u(b) = 0 on I = (a, b).

This equation is regular and already in normal form. We obtain the eigenvalues
from the characteristic equation:
2 + 2 = 0
2 = 2
= i
This gives {eix } as eigenfunctions. However, since the original equation has real coefficients, we would like a basis of real-valued eigenfunctions. Since Re(ei ) = cos x
and Im(ei ) = sin x, we take {cos x, sin x} as a basis. Thus the eigenfunctions of
(?1 ) are all of the form
u(x) = c1 cos x + c2 sin x.
Now we use the initial conditions:
u(a) = c1 cos a + c2 sin a = 0
u(b) = c1 cos b + c2 sin b = 0
From the first equation above,
sin a
c1 = c2 cos
.
a

Substituting into the second,

sin a
c2 cos
cos b + c2 sin b = 0
a
c2 sin a cos b + c2 cos a sin b = 0

c2 sin (b a) = 0.
Thus we must have (b a) = 2k for some k Z, and the eigenfunctions must look
like
2k
2k
{cos ba
x, sin ba
x}.

ERIN PEARSE

2. (10/12) The DE with constant coefficients

(?2 )

L[u] = u(n) + a1 u(n1) + a2 u(n2) + + an u,

(ai R)

is stable iff no root of its characteristic polynomial has positive real part and all
multiple roots have strictly negative real part.

(1)

We know from 146 that we can find a basis for the solution space which consists of
functions of the form {tk et } where k N = {0, 1, 2, . . . }. So every solution to (?2 )
can be written
n
X
u(t) =
c j t kj e j t .
j=0

(2)

() Assume that (?2 ) is stable, i.e., all solutions remain bounded as t . If

Re() were strictly positive, tk et would blow up as t , for any k, and hence so
would u. This contradiction shows Re() < 0 for every root of the characteristic
polynomial. Consider the case when is a repeated root. Then there is a term in
(1) of the form tk et for k 1. In this situation, Re() = 0 would imply that this
t
term is of the form tk ei . Since |tk ei | but u does not blow up, it must not
be the case that Re() = 0.
() Assume that no roots of the characteristic polynomial of (?2 ) have positive
real part and all multiple roots have strictly negative real part. Then we have


n
n
 X
X
 k t

kj j t 
c j t j e j 
cj t e 
|u(t)| = 


j=0

j=0

by the triangle inequality. If j is not a multiple root, then kj = 0 and we have

 k t  t
cj t j e j  = cj e j  |cj | .
Equality holds iff Re(j ) = 0. Otherwise,
 t  t
cj e j  0.
Re(j ) < 0
=

If j is a multiple root, then kj > 1 and the hypothesis gives Re(j ) < 0. In this
case we still have
 k t  t
cj t j e j  0

because et goes to zero faster than any polynomial for = j > 0. Returning
to (2), we see that the worst case scenario is when there are no multiple roots and
Re(j ) = 0. In this case, the largest u can get is
n
n
X
 k t X
j
j 

|u(t)|
cj t e
=
|cj | ,
j=0

which is clearly bounded. Hence u is stable.

j=0

MATH 211B HOMEWORK

3. (10/14) For solutions u, v of

(?3 )

L[u] = u00 + p(x)u0 + q(x)u = 0

p, q continuous on I,

{u, v} are linearly independent solutions of (?3 ) iff {u, v} is a fundamental set of
solutions.
() Assume {u, v} are linearly independent solutions. The solution space of a 2nd
order DE will be 2-dimensional, so two linearly independent solutions will span the
entire space. Hence {u, v} is a basis and thus also a fundamental set of solutions.
() Assume {u, v} is a fundamental set of solutions. Then it is a basis, and hence
its elements are linearly independent by definition.

4. (10/14) Consider the equation

(?4 )

y 00 + q(x)y = 0

q is piecewise continuous on R.

Define a soln of (?4 ) to be a function y = f (x) which is C 1 (but not C 2 ) and

satisfies the DE (?4 ) at all points where q is continuous.
(a) Describe explicitly a basis of solutions of (?4 ) where
(
1, x > 0
q(x) =
.
1 x < 0
For x > 0, a basis of solutions would be {cos x, sin x}, and for x < 0 a basis of
solutions would be {ex , ex }. Thus, a general solution would be
(
a cos x + b sin x,
x>0
u(x) =
.
x
x
e + e , x < 0
This is clearly C 1 on (, 0) and (0, ); we only need to worry about 0. To be
C 1 at 0, we require
(a cos x + b sin x)|x=0 = (ex + ex )|x=0
0

and

x 0

(a cos x + b sin x) |x=0 = (e + e ) |x=0 .

In other words,
(a cos 0 + b sin 0) = a = + = (e0 + e0 )

and

(a sin 0 + b cos 0) = b = = (e0 e0 ).

Thus, solve this system for and and get
=

a+b
2

and =

ab
,
2

and note that

ex + ex =

a+b x
e
2

ab x
e
2

= a2 (ex + ex ) + a2 (ex ex ).

ERIN PEARSE

We can write the basis of solutions as {u1 , u2 } where

( x x
( x x
e e
e +e
,
x
<
0
, x<0
2
2
u1 (x) =
and u2 (x) =
.
cos x,
x>0
sin x,
x>0
(b) State & prove an existence and uniqueness theorem for the corresponding IVP
and examine the DE satisfied by the Wronskian, if any.
Given any initial conditions y(x0 ) = y0 , y 0 (x0 ) = v0 , there exists a unique solution
to equation (?4 ). This follows essentially from the more basic case when q is
continuous. For example, take x0 < 0. Then
y 00 y = 0,

y(x0 ) = y0 , y 0 (x0 ) = v0

has a unique solution u(x) on (, 0) by the basic theory. However, this determines u(0), u0 (0) by continuity, which may then be used as the initial conds for
the other half interval. Now
y 00 + y = 0,

y(0) = u(0), y 0 (0) = u(0)

has a unique solution on R+ . Combining, we obtain a unique solution on R.

Examining the Wronskian of this system, we have
W (x) = u1 u02 u01 u2 .
Differentiating gives
0

W (x) =

u1 u002

u001 u2

(
u1 (u2 ) (u1 )u2 , x < 0
=
,
u1 (u2 ) (u1 )u2 ,
x>0

which is 0 in either case. The central equality follows because the given q makes
(?4 ) into
(
y, x < 0
00
y =
.
y,
x>0
Hence, the Wronskian satisfies W 0 = 0, i.e., W is constant.
(c) Relate this to
(i) the Laplace transform method for solving such equations, and
(ii) the distributional approach to ODEs.

5. (10/19)
(?5 )

L[u] = u00 + u = f

(f L1loc )

Show that if u is a classical solution (i.e., u C 2 ) of L[u] = f , then u is a distributional solution also.

MATH 211B HOMEWORK

Take a solution u C 2 of (?5 ) and fix Cc (R). Then

Z
Z
L[]u = (00 + )u
Z
Z
00
= u + u
Z
Z
2
00
= (1)
u + u
ibp twice
Z
= (u00 + u)
Z
= f
u is a soln
R
R
shows L[]u = f , i.e., u is also a distributional solution.
6. (10/29) The Bessel DE:
(?6 )

(3)

L[u] = x2 u00 + xu0 + (x2 n2 )u = 0

Show that the self-adjoint form of this equation is


0
n2
d
[xu
]
+
x

u = 0.
dx
x
We obtain (?6 ) from (3):


0
d
n2
[xu
]
+
x

u=0
dx
x


2
xu00 + u0 + x nx u = 0


x2 u00 + xu0 + x2 n2 u = 0.

product rule
mult through by x

Then note that (3) is self-adjoint by the theorem which states that the 2nd order DE
L[u] = p0 (x)u00 + p1 (x)u0 + p2 (x)u = 0
is self-adjoint iff it is of the form


d
p(x) du
+ q(x)u = 0.
dx
dx
7. (10/29) The Legendre DE of order n, self-adjoint form:


2 du
d
(?7 )
(1

x
)
+ u = 0
x R.
dx
dx

Show that for = n(n + 1), n N, this DE has polynomial solutions.

By differentiating the first term and plugging in , we have the equation
(1 x2 )u00 2xu0 + n(n + 1)u = 0.

ERIN PEARSE

and

Two linearly independent solutions of this equation are

! m1
!

m1
X
Y
Y
x2m
u1 (x) = 1 +
(1)m
(n 2k)
(n + 1 + 2k)
(2m)!
m=1
k=0
k=0
u2 (x) = x +

(1)

m=1

m1
Y

(n 1 2k)

k=0

m
Y

(n + 2k)

k=0

x2m+1
(2m + 1)!

While these are ostensibly infinite series, n N may result in the disappearance of
many terms. For example, suppose n is
even and consider u1 . For m n2 + 1, the
coefficient will contain a factor n 2( n2 ) = 0, and will hence vanish. So for n even,
u1 (x) is a polynomial solution of degree n.
Alternatively, suppose n is odd and
consider u2 . For m n1
+ 1, the coefficient
2


will contain a factor n 1 2( n1
)
=
0,
and
will
hence
vanish.
So for n odd, u2 (x)
2
is a polynomial solution of degree n.
Either way, (?7 ) has a polynomial solution for n N.

8. (10/29) Show that the third order linear homogeneous DE

p0 (x)u000 + p1 (x)u00 + p2 (x)u0 + p3 (x)u = 0

(?8 )

00
0
is exact iff its coefficients satisfy p000
0 p1 + p2 p3 = 0.

We define a DE of the form (?8 ) to be exact iff

p0 (x)u000 + p1 (x)u00 + p2 (x)u0 + p3 (x)u =

d
dx

[A(x)u00 + B(x)u0 + C(x)u]

for some A, B, C C 1 . Differentiating the right-hand side above,

p0 u000 + p1 u00 + p2 u0 + p3 u = Au000 + A0 u00 + Bu00 + B 0 u0 + Cu0 + C 0 u.
Matching coefficients,
p0 = A,

p1 = A0 + B,

p2 = B 0 + C,

and p3 = C 0 .

Then using these equations we expand p3 as

p3 = C 0 = (p2 B 0 )0
= p02 (p1 A0 )00
= p02 p001 + p000
0.

9. (10/29)

n
(1 + n2 x2 )
R
Show that dn 0, R dn (x)dx = 1. Sketch the graph of dn , n = 1, 2, . . . and argue
that dn is a -sequence, i.e., dn .
dn (x) =

MATH 211B HOMEWORK

5
4
3
2
1

1
2

1 1
6 23

1
1
23 6

1
2

Figure 1. An sketch of dn for n = 1, 2, 3, 4, 5.

(4)

Clearly, dn 0 as both the numerator and denominator are positive for any real x
and n = 1, 2, . . . . We compute the integral and find
Z
Z
n
dx
dn (x)dx =
2 2
R
R (1 + n x )
Z
n
1
=
dx
R 1 + (nx)2


n 1
=
arctan(nx)
n

1
= (arctan() arctan())

1   

=
2
2
=1

for any n = 1, 2, . . . . A sketch of dn is depicted in Figure 1.

Note that
(5)

dn (x) =

1
n
= nd(xn) for d(x) :=
.
2
2
(1 + n x )
(1 + x2 )

Thus,
lim hdn , i = lim

dn (x)(x) dx
Z
= (0) + lim
dn (x) [(x) (0)] dx,
n

ERIN PEARSE

and we just need to show that the latter integral goes to 0. Fix > 0. Then
Z



 dn (x) [(x) (0)] dx


Z
Z

dn (x) |(x) (0)| dx +

dn (x) |(x) (0)| dx
|x|<r
|x|r
Z
max{|(x) (0)|}
dn (x) dx
|x|<r
|x|<r
Z
+ max{|(x) (0)|}
dn (x) dx
|x|r
|x|r
Z
max{|(x) (0)|} 1 + M
dn (x) dx,
|x|<r

|x|r

where M := max|x|r {|(x) (0)|}, and the last equality follows by (4). Now for
sufficiently small r, the continuity of gives
max{|(x) (0)|} < 2 .
|x|<r

Note that we dont let r go to 0, just pick r > 0 small enough that the inequality will
hold. Then deal with the second term as follows:
Z
Z
Z
dn (x) dx =
nd(nx) dx =
d(u) du
|x|r

|x|r

|x|rn

where the first equality comes by (5) and the second comes by the change of variables
u = nx. Then since
Z
d(u) du = 1

by (4) again, it must be possible to pick n so large that

Z

d(u) du <
.
2M
|x|rn
This is sufficient to get
Z



 dn (x) [(x) (0)] dx < .



10. (11/02) We saw for f (x) = 1r , r := kxk, that f L1loc (R2 ). Extend this to d 2.
We want to show
f (x) := kxk1d L1loc (Rd )

for d 2.


Since this function is clearly in L1loc Rd \{0} , we only really need to check f in a
neighborhood of the origin. Let B = B(0, 1) be the ball of radius 1 centered at the
origin of Rd .

MATH 211B HOMEWORK

We convert to spherical coordinates with

r = kxk =

d
X

x2k

k=1

!1/2

so that
f (x) = r 1d .
The rectangular coordinates x1 , x2 , . . . , xd are related to the spherical coordinates
r, 1 , . . . , d1 by the equations:1
xk = rS k1 ck
xd = rS d1
r = (x21 + x22 + + x2d )1/2
 
Rk
k = arctan
xk

where

k
Sm

k
Y

sin j

j=m

cm = cos m
sm = sin m
Rk = (r 2 x21 x2k )1/2 = rS1k
Although we do not wish to work it out explicitly, the Jacobian for this change of
coordinates will be of the form





|J| = 




(rS 0 c1 )
r

(rS 1 c2 )
r

..
.

(rS d1 c2 )
r

(rS 0 c1 )
1

(rS 1 c2 )
1

...
...
..
.

..
.

(rS d1 cd ) . . .
1

(rS 0 c1 )
d1

(rS 1 c2 )
d1






.
..

.


d1

(rS
c
)
d
d1

As the determinant is expanded by cofactors and each cofactor is evaluated recur

sively, a factor of r will emerge for each term that does not stem from a r
. Thus,
there will be a common factor of r d1 in the final computed Jacobian |J| = r d1 |K|.

1Thanks

to Andrew Snowden of the University of Maryland for these conversion formulas.

10

ERIN PEARSE

Thus,
Z

f dx =
B

=
=

r 1d |J| dd1 . . . d1 dr

ZB

ZB

r 1d r d1 |K| dd1 . . . d1 dr
|K| dd1 . . . d1 dr

d/2

,
d2 + 1

the volume of the unit ball in Rd . Since this is finite, f L1loc (Rd ).

11. (11/02) Prove that for any regular distributions Tf , Tg ,

Tf = Tg = f =ae g.
First we note that T : D D 0 by f 7 Tf is a linear map. Since
hTf + Tg , i = hTf , i + hTg , i
Z
Z
= f dx + g dx
Z
= (f + g) dx
= hTf +g , i

for every D, we have Tf + Tg = Tf +g . Thus it suffices to show that

Tf = 0 = f =ae 0.
Let c (x) be the standard mollifier, i.e.,
(i) c C ,
(ii) spt(c ) = [ 1c , 1c ]
R
(iii) c dx = 1, c > 0.
Then

(f n )(x) =

f (y)n (x y)dy

= Tf (n ),
where (y) = x y for any fixed x. Since Tf = 0, this shows f n = 0 for any n.
Then by Proposition 1,
n

f n 0

f 0, ae.

MATH 211B HOMEWORK

11

Proposition 1. If n is the standard mollifer and f Lp , then

lim (f n ) = f, in Lp .

Proof. [Al-G, Example 2.23]

Cc0 is dense in Lp , so we can choose Cc0 such that kf kp < . Then
kf n n kp = k(f ) n kp kf kp < ,
where the central inequality comes by Lemma 2. Hence it suffices to prove
Lp

n .
unif

Since n on K := spt by Lemma 3, we can write

Z
1/p
p
k n kp =
|( n )(x) (x)| dx
K

sup {|( n )(x) (x)|}

xK

Z

dx
K

1/p

<
if n is large enough. Thus
kf n f kp kf n n kp + k n kp + k f kp < 3.

Lemma 2. For f Lp and 1 p < , we have kf n kp kf kp .
Proof. [Al-G, Example 2.23]
For p = 1,
kf n k1

Z Z

n (y) |f (x y)| dy dx
Z

Z
= n (y)
|f (x y)| dx dy
= kf k1 .

Otherwise, for 1 < p < ,

kf

n kpp

p
Z Z



=  n (y)f (x y) dy  dx.

To use Holders inequality, we split the mollifier as

1/q
f n = (f 1/p
n )(n ),

where
Z

1
p

1
q

= 1. Then the inequality gives

Z
1/p Z
1/q
p
n (y) |f (x y)| dy
n (y) |f (x y)| dy
n (y) dy
.

12

ERIN PEARSE

Since

n (y) dy = 1, we have
Z Z
p
kf n kp
n (y) |f (x y)|p dy dx
Z

Z
p
= n (y)
|f (x y)| dx dy
Z
= n (y)kf kpp dy
= kukpp .

Thus kf n kp kf kp for all 1 p < .

Lemma 3. If f C 0 , then f n f uniformly on every compact subset.

Proof. [Al-G, Theorem 2.28(iii)]
Since f is continuous, it is uniformly continuous on any compact set E. I.e., given
> 0, there is a > 0 such that
|f (x y) f (x)| <
for all x E and for all y B(0, ). Then
Z



|(f n )(x) f (x)| =  [f (x y) f (x)] n (y) dy 
Z

|f (x y) f (x)| n (y) dy
B(0,1/n)

<
if we take n large enough that

1
n

< .

12. (11/04) Define T D 0 (R) by T () = (n) (0). Prove T is a distribution.

(i) well-defined. Since is C and has compact support, and all its derivatives
are bounded.
(ii) linearity. T is just composition of the evaluation operator and the differentiation
operator, both of which are linear. Since evaluation is a functional, this shows
T is a linear functional.
D

(iii) continuity. Let k 0. Part of the definition of convergence in D is that all

derivatives of the k also converge to 0. In particular,
(n)

k 0,
and hence

(n)

T (k ) = k (0) 0.

MATH 211B HOMEWORK

13

13. (11/04) Verify, via abstract nonsense, that if T : D() C is a distribution, then
.
K compact, the restriction of T to Dk () = { D() .. spt() K} is
continuous.
Define T |K to be the restriction of T to a compact set K , i.e.,
T |K () := T (), DK ().
Abstract nonsense approach: T |K is the composition of T with the inclusion map
: DK () D(), both of which are continuous.
k
Definitional approach: let {k } DK () with k 0. Then
k

T |K (k ) = T (k ) 0.

14. (11/09)
(a) Show D + T = D (D T ) for , Nd , N = {0, 1, 2, . . . }.
The equality of mixed partials gives the central equality in the following:
D + T =

||
||
=
= D (D T ).
1 +1 x1 . . . d +d xd
1 x1 . . . d xd 1 x1 . . . d xd

(b) For an open set Rd and T D 0 (), show that D T () = (1)|| T (D ).

Define Dj := D ej =

.
xj

Having already established the basis case

Dj T () = T (Dj )
in the notes, we induct on || by assuming D T () = (1)|| T (D ) for || = n.
Now consider || = n + 1 by taking = + ej where ej = [kj ] has a 1 in the j th
spot and 0s elsewhere. Then
D T () = Dj D T ()

by (a)

= Dj (1)|| T (D )

by inductive hypothesis

= (1)|| (1) T (Dj D )

by basis case

= (1)|| T (D ).

|| = || + 1

15. (11/09)
(a) Let f C () and T D 0 (). Then show f T D 0 () is well-defined by
hf T, i = hT, f i, D().
(i) The right-hand side shows hf T, i is well-defined, since f D.

14

ERIN PEARSE

(ii) Also, the right-hand side shows that hf T, i is a linear functional.

D

(iii) Let k 0. Then clearly

D

f k 0,
since f C () and spt(f k ) spt(k ). By the continuity of T (which
was given), this gives
k

hf T, k i = hT, f k i 0.
(b) Moreover, state and prove an analogue of Leibnitz rule for D (f T ).

(6)

We will use induction to show that

X  

D (f T ) =
(D f ) D T ,

!
where := !()!
, and the factorial of a multiindex is defined by ! =
1 ! . . . d !. Also, means that j j for j = 1, . . . , d.
The basis case is a straightforward calculation:
Dj (f T )() = f T (Dj )

def of T 0

= T (f Dj )

def of f T

= T (Dj (f ) Dj f )

product rule

= T (Dj (f )) + T (Dj f )

linearity

= Dj T (f ) + T (Dj f )

def of T 0

= f Dj T () + (Dj f )T ()

def of f T

shows Dj (f T ) = f Dj T + (Dj f )T .
Now we assume that (6) holds for || = n and let = + ej as in Problem 14b.
D (f T ) = Dj D (f T )
X  


= Dj
(D f ) D T

X   


=
Dj (D f ) D T


X 




=
(D f ) Dj D T + D T Dj (D f )


X 






=
(D f ) D +ej T + D T Dj D +ej f

X  
=
(D f )(D T )

inductive hypothesis
linearity
basis case
Problem 14a
reordering

MATH 211B HOMEWORK

15

16. (11/09) With {an } R and

Ha (x) := H(x a) =

1, x 0
,
0, x < a

we have seen that the step function

k
X

fk (x) =

wn H(x an )

n=1

is in L1loc (R) by the vector space properties of L1loc (R). Thus it induces a regular
distribution Tf which has derivative
d
T
dx fk

k
X

d
wn dx
H(x

an ) =

n=1

k
X

wn (x an ).

n=1

Hence, the distributional derivative of fk is

k

dfk X
=
w n an .
dx
n=1

Assuming that anyPbounded interval contains finitely many an s, extend this to the
case when f (x) =
n=1 wn H(x an ).
k

Denote the regular distribution associated with f by Tf . Since fk f in

L1loc (R), the continuity of the derivative allows us to say
Tf0k

k
X

wn (x an )

n=1

wn (x an ) = Tf0 .

n=1

Note that the right-hand side makes sense, since for any D, there can only be
finitely many an in spt(). If we denote them by {anj }, we have
!
Z X
m
Tf0 () =
wnj (x anj ) (x) dx
R

=
=

j=1
Z
m
X

wnj (x anj ) (x) dx

j=1
m
X

wnj (anj ) < .

j=1

This makes it easy to see that Tf0 is continuous. If q 0, then

Tf0 (q ) =

m
X
j=1

wnj q (anj ) 0,

because q (anj ) 0 for each j.

16

ERIN PEARSE

17. (11/16)
(a) Let T D 0 (R). Then
h T T
dT
= lim
dx h0
h

(limit in D 0 ),

where h T D 0 is defined by hh T, i = hT, h i for (h )(x) = (x + h).

The action of the right-hand side against a test function is given by




h T T
h T T
lim
, = lim
,
continuity of h, i
h0
h0
h
h

1
= lim
hh T, i hT, i
linearity of h, i
h0 h


1
= lim
hT, h i hT, i
defn of
h0 h


h
= lim T,
linearity of h, i
h0
h
= hT, 0 i

continuity of h, i

= hT, 0 i
= hT 0 , i

(b) Extend this to higher dimensions.

Let ej be the standard basis vector with 1 in the j th slot and 0 elsewhere.
Define hj T D 0 by hhj T, i = hT, hj i for (hj )(x) = (x + hej ). Then
hj T T
T
= lim
h0
xj
h
hj T T
lim
,
h0
h

h
= lim T, j
h0
h


= hT, Dj i

(limit in D 0 ).

linearity of h, i
continuity of h, i

= hDj T, i
via the same arguments as in (a), where we are using Dj := D ej =
14(b) extends this to the more general case of D T .

.
xj

MATH 211B HOMEWORK

18. (11/16) If the series

term:
More generally, D (
We have
*
X
n=0

Tn

!0

17

Tn converges in D 0 , then it can be differentiated term by

(

Tn ) =

lim

= lim

= lim

= lim

= lim

Tn ) 0 =

k
X

Tn , 0

defn ofT 0 ,

Tn , 0

continuity of h, i

n=0

* k
X
n=0

k
X

hTn , 0 i

linearity of h, i

hTn0 , i

defn of T 0

n=0

k
X
n=0

*
X

k
X

lim

Tn0 .

Tn0 ,

linearity of h, i

Tn0 ,

continuity of h, i

n=0

D Tn , for every multiindex .

* k
X

n=0

Tn0 ,

n=0

To extend this to the general case,

*
+
! +
*
k

X
X
Tn , D
Tn , = (1)|| lim
D
k

n=0

= (1)|| lim

= (1)

||

*
X

lim

n=0

n=0

k
X

hTn , 0 i

k
X

as in (a)

n=0

k
X
lim
(1)|| hD Tn , i

defn of D T

defn of D T

n=0

D Tn ,

n=0

D Tn ,

as in (a)

18

ERIN PEARSE

19. (11/16) Suppose Tn D 0 is such that

T () := lim Tn ()
n

exists in C, D(). Show T D .

Since T is clearly a linear functional, it just remains to check the continuity of T .
Recall the Principle of Uniform Boundedness (for Frechet spaces):
Theorem. [Ru, Theorem 2.6]: If {Tn } is a sequence of continuous linear mappings
from D to C and if the sets
.
() := {|Tn | .. n N}
are bounded in C for each D, then is equicontinuous.
This shows that {Tn } is equicontinuous, i.e., to every neighborhood W of 0 in C,
there corresponds a neighborhood V of 0 in D such that
Tn (V ) W,
for all n [Ru, Definition 2.3]. Note that the concept of neighborhood makes sense
here: although a locally convex space need not be metrizable in general, every Frechet
space comes with a topology induced by a complete metric.
Let {k }
k=0 D with k 0. To see that T is continuous, we must show
k
hT, k i 0. Fix > 0 and consider B := B(0, ) C. By the above remarks,
there must be some > 0 such that Tn (k ) B for all k A := B(0, ), n N.
Since
D
E
hT, k i = lim Tn , k = lim hTn , k i,
n

we can pick N such that k > N = k A, and be sure that

|hT, k i| < .

20. (11/16) Let I = (a, b) R, and T D(I). Show that

dT
=0
=
T is constant.
dx
If dT
= 0, we have hT 0 , i = hT, 0 i = 0, D. This just means, = 0 =
dx
hT, i = 0. Since test functions are C , they are clearly absolute continuous, and
may thus be represented
Z x
(x) =
0 (t)dt + (a).
a
Rx
In particular, for D, we can define (x) = a (t)dt + c, so that 0 = . In
other words, every test function can be represented as the derivative of some other
test function, and we have hT, i = 0, .

MATH 211B HOMEWORK

19

21. (11/18) Show that W1,p () is separable for 1 p < .

Because of the isometric embedding I : W1,p () Lp (), we may think of W1,p ()
as a closed subspace of Lp ().

22. (11/18) = I = (1, 1) R. Show that u(x) = 12 (|u| + u) belongs to W1,p (I), for
1 p < , and that u0 = H. More generally, a continuous function on I = [1, 1]
that is piecewise C 1 belongs to W1,p (I), 1 p < .
u is clearly in Lp :
Z

u dx =
I
0

0 dx +
(1,0)
p

so it just remains to show u L (I).

(0,1)

x dx = 12 ,

hu0 , i = hu, 0 i
Z
= u0 dx
I

= M
<

where M := supI 0 (x).

23. (11/18) Prove that



Z
Z
..
1,p
p
p
0
1
W (I) = u L (I) . g L (I) such that u = g, Cc (I) .
I

1,p

Also, for u W (I) we have u = g. Note: g is ae-unique.

By definition,
u W1,p (I)

u Lp and u0 Lp ,

where u0 Lp means that u0 = Tg for some unique g Lp .

24. (11/18)
(a) Define u
(x) =

Rx
0

u0 (t) dt for u W1,p (I). Show u

is absolutely continuous on I.

(b) Using (a), show

T du =
dx

d
T.
dx u

20

ERIN PEARSE

References
[Al-G]
[ChZa]
[Ev]
[Ru]

M. A. Al-Gwaiz, Theory of Distributions, Pure and Applied Mathematics 159, Marcel

Dekker, Inc., New York, 1992.

Edwin K. P. Chong and Stanislaw H. Zak,

An Introduction to Optimization, 2nd ed.,
Wiley, Inc., New York, 2001.
Lawrence C. Evans, Partial Differential Equations, Graduate Studies in Mathematics
volume 19, American Mathematical Society, Providence, 1998.
Walter Rudin, Functional Analysis, 2nd ed., International Series in Pure and Applied
Mathematics, McGraw-Hill, Inc., 1973.