34 RICHARDB.
MELROSE
6. Test functions
So far we have largely been dealing with integration. One thing we
haveseenisthat,byconsideringdualspaces,wecanthinkoffunctions
asfunctionals. Letme briey review this idea.
Consider the unit ball in R
n
,
n
B ={x R
n
; |x|1} .
Itaketheclosed unitballbecauseIwanttodealwithacompactmetric
space. We have dealt with several Banach spaces of functions on B
n
,
forexample
C(B
n
)= u :B
n
C;ucontinuous
2
L
2
(B
n
)= u :B
n
C;Borelmeasurable with |u| dx < .
Here,asalwaysbelow,dx isLebesguemeasureandfunctionsareiden-
tied ifthey areequal almost everywhere.
Since B
n
is compact we have anaturalinclusion
(6.1) C(B
n
)L
2
(B
n
).
Thisisalsoatopologicalinclusion, i.e.,isaboundedlinearmap,since
(6.2) u
L
2 Cu||
where C
2
is the volume of the unitball.
In general if we have such aset up then
Lemma 6.1. If V U is a subspace with a stronger norm,
U
C
V
V
then restriction gives a continuous linear map
L =L|
V
V
(6.3) U
V
, U
L
V
CL
U
.
If V is dense in U then the map (6.3) is injective.
Proof. By denition ofthe dual norm
L
V
=sup L(v) ; v
V
1, v V
sup L(v) ; v
U
C , v V
sup{|L(u)| ;u
U
C , u U}
=CL
U
.
If V U is dense then the vanishing of L : U C on V implies its
vanishing onU.
35 LECTURE NOTES FOR 18.155, FALL 2004
Goingbacktotheparticularcase(6.1)wedoindeedgetacontinuous
mapbetween the dual spaces
L
2
(B
n
)
= (L
2
(B
n
))
(C(B
n
))
=M(B
n
).
Here we use the Riesz representation theorem and duality for Hilbert
spaces. The map use here is supposed tobe linear notantilinear, i.e.,
(6.4) L
2
(B
n
)g g dx(C(B
n
))
.
Sotheideaistomakethespaceoftestfunctionsassmallasreasonably
possible, while still retainingdensity inreasonable spaces.
Recallthatafunctionu:R
n
Cisdierentiable atxR
n
ifthere
exists aC
n
such that
(6.5) |u(x)u(x)a (x x)|=o(|x x|).
The little oh notationhere means thatgiven >0there exists >0
s.t.
|x x|<|u(x)u(x)a(x x)|< |x x|.
The coecients of a = (a
1
, . . . , a
n
) are the partial derivations of u at
x,
u
a
i
= (x)
x
j
since
u(x +te
i
)u(x)
(6.6) a
i
=lim ,
t0
t
e
i
= (0, . . . , 1, 0, . . . , 0) being the ith basis vector. The function u is
saidtobecontinuouslydierentiableonR
n
ifitisdierentiableateach
pointxR
n
and each ofthe n partialderivatives arecontinuous,
u
(6.7) :R
n
C.
x
j
Denition 6.2. Let C
0
1
(R
n
) be the subspace of C
0
(R
n
) =C
0
0
(R
n
) such
u
that each element u C
0
1
(R
n
) is continuously dierentiable and
x
j
C
0
(R
n
), j =1, . . . , n.
Proposition 6.3. The function
n
u
u
C
1 =u
x
1
i=1
is a norm on C
0
1
(R
n
) with respect to which it is a Banach space.
36
C
RICHARDB. MELROSE
Proof. That
C
1 isanormfollowsfromthepropertiesof
. Namely
u
C
1 =0 certainly implies u=0, au
C
1 =|a| u
C
1 and the triangle
inequality follows from the same inequality for
.
1
0
(R
n
) follows from Similarly, the main part of the completeness of C
0
0
(R
n
).
1
0
(R
n
) the completeness of C If {u
n
} is a Cauchy sequence in C
then u
n
and the
un
x
j
0
0
(R
n
are Cauchy in C ). It follows that there are
limits ofthese sequences,
u
n
0
0
(R
n
). v, v
j
C u
n
x
j
However we do have to check that v is continuously dierentiable and
that
v
=v
j
.
x
j
One way to do this is to use the Fundamental Theorem of Calculus
ineach variable. Thus
t
u
n
u
n
(x+te
i
)= (x+se
i
)ds+u
n
(x).
x
j 0
Asn alltermsconvergeandso,bythecontinuityoftheintegral,
t
u(x+te
i
)= v
j
(x+se
i
)ds+u(x).
0
This shows thatthe limit in (6.6)exists, so v
i
(x) is the partialderiva-
tion of u with respect to x
i
. It remains only to show that u is indeed
dierentiable ateach pointand I leave this toyou inProblem 17.
So, almost by denition, we have an example ofLemma 6.1,
1
0
(R
n
) C
0
0
(R
n
). C
ItisinfactdensebutIwillnotbothershowingthis(yet). Soweknow
that
0
0
(R
n
))
(C
1
0
(R
n
))
(C
and we expect it to be injective. Thus there are more functionals on
1
0
(R
n
) including things thataremore singular thanmeasures.
An example is related to the Diracdelta
0
0
(R
n
), (x)(u)=u(x), u C
namely
u
(R
n
) u
1
0
C (x) C.
x
j
This is clearly a continuous linear functional which it is only just to
denote
(x).
x
j
Ofcourse, why stop atone derivative?
37 LECTURE NOTES FOR 18.155, FALL 2004
1
Denition 6.4. The space C
0
k
(R
n
) C
0
(R
n
)k 1 is dened induc-
tively by requiring that
u
C
0
k1
(R
n
), j =1, . . . , n.
x
j
The norm on C
0
k
(R
n
) is taken to be
n
u
(6.8) u
C
k =u
C
k1 +
C
k1 .
x
j
j=1
These areallBanachspaces, since if{u
n
} isCauchy inC
0
k
(R
n
),itis
CauchyandhenceconvergentinC
k1
(R
n
),asisu
n
/x
j
,j =1, . . . , n
0
1. Furthermorethelimitsoftheu
n
/x
j
arethederivativesofthelimits
by Proposition 6.3.
This gives us asequence ofspaces getting smoother andsmoother
1 k
C
0
0
(R
n
) C
0
(R
n
) C
0
(R
n
) ,
with norms getting larger and larger. The duals can also be expected
togetlarger andlarger as k increases.
As well as looking at functions getting smoother and smoother, we
need to think about innity, since R
n
is not compact. Observe that
anelement g L
1
(R
n
) (with respect to Lebesgue measure by default)
k
denesafunctionalonC
0
0
(R
n
)andhenceall theC
0
(R
n
)s. Howevera
functionsuchastheconstantfunction1isnot integrableonR
n
.Since
we certainly want to talk about this, and polynomials, we consider a
second condition ofsmallness at innity. Let us set
(6.9) x =(1+|x|
2
)
1/2
a function which is the size of |x| for |x| large, but has the virtue of
beingsmooth
10
Denition 6.5. For any k, l N={1, 2, } set
k
k k
x
l
C
0
(R
n
)= u C
0
(R
n
); u=x
l
v , v C
0
(R
n
)
,
with norm, u
k,l
=v
C
k , v=x
l
u.
k
Noticethatthedenitionjustsaysthatu=x
l
v,withv C
0
(R
n
).
It follows immediately that x
l
C
0
k
(R
n
) is a Banach space with this
norm.
Denition 6.6. Schwartz space
11
of test functions on R
n
is
k
S(R
n
)= u:R
n
C;u x
l
C
0
(R
n
) for all k and l N
.
10
See Problem 18.
11
Laurent Schwartz this one with a t.
38 RICHARDB. MELROSE
It is not immediately apparent that this space is non-empty (well 0
is in there but...); that
2
exp(|x| ) S(R
n
)
is Problem 19. There are lots of other functions in there as we shall
see.
Schwartz idea is that the dual of S(R
n
) should contain all the in-
teresting objects, at least those of polynomial growth. The problem
isthatwedonot haveagoodnormonS(R
n
).Ratherwehavealot of
them. Observe that
l
k
x
l
C
0
(R
n
) x C
0
(R
n
)if l l
and k k
.
Thus we see thatas alinear space
k
(6.10) S(R
n
)= x
k
C
0
(R
n
).
k
Since these spaces are getting smaller, we have a countably innite
number of norms. For this reason S(R
n
) is called a countably normed
space.
Proposition 6.7. For u S(R
n
), set
(6.11) u
(k)
=x
k
u
C
k
and dene
(6.12) d(u,v)=
2
k
uv
(k)
,
1+uv
(k)
k=0
then d is a distance function in S(R
n
) with respect to which it is a
complete metric space.
Proof. The series in (6.12)certainly converges, since
uv
(k)
1.
1+uv
(k)
The rst two conditions on ametric are clear,
d(u,v)=0 uv
C
0
=0 u=v,
and symmetry is immediate. The triangle inequality is perhaps more
mysterious!
Certainly it is enough toshow that
uv
(6.13) d
(u,v)=
1+uv
39 LECTURE NOTES FOR 18.155, FALL 2004
isametric onanynormed space, since thenwe maysumover [Link]
we consider
u v v w
+
1+u v 1+vw
u v(1+v w)+vw(1+u v)
= .
(1+u v)(1+v w)
Comparing this to d
(v, w)we must show that
(1+u v)(1+vw)u w
(u v(1+v w)+vw(1+u v))(1+u w).
Startingfromthe LHS and using the triangle inequality,
LHS u w+(u v+vw+u vv w)u w
(u v+v w+u vvw)(1+u w)
RHS.
Thus, dis ametric.
Suppose u
n
is a Cauchy sequence. Thus, d(u
n
, u
m
) 0 as n, m
.In particular, given
>0 N s.t. n, m>N implies
d(u
n
, u
m
)<2
k
n, m>N.
The terms in(6.12)are allpositive, so this implies
u
n
u
m
(k)
< n, m>N.
1+u
n
u
m
(k)
If<1/2this in turnimplies that
u
n
u
m
(k)
<2,
so the sequence is Cauchy in x
k
C
0
k
(R
n
) for each k. From the com-
pleteness of these spaces it follows that u
n
u in x
k
C
0
k
(R
n
)
j
for
each k. Given > 0 choose k so large that 2
k
< /2. Then N s.t.
n>N
u u
n
(j)
</2n>N, j k.
40 RICHARDB. MELROSE
Hence
d(u
n
, u)= 2
j
u u
n
(j)
1+u u
n
(j)
jk
+ 2
j
u u
n
(j)
1+u u
n
(j)
j>k
/4+2
k
<.
This u
n
uin S(R
n
).
AswellastheSchwartz space, S(R
n
),offunctions ofrapiddecrease
withallderivatives,thereisasmallerstandardspaceoftestfunctions,
namely
(6.14) C
(R
n
)={u S(R
n
);supp(u)R
n
} ,
c
the space of smooth functions of compact support. Again, it is not
quite obvious that this has any non-trivial elements, but it does as
we shall see. If we x a compact subset of R
n
and look at functions
with support in that set, for instance the closed ball of radius R >0,
thenwegetaclosedsubspaceofS(R
n
),henceacompletemetricspace.
OneproblemwithC
(R
n
)isthatitdoesnothaveacompletemetric
c
topology which restricts to this topology on the subsets. Rather we
must use aninductive limit procedure toget adecent topology.
Just to show that this is not really hard, I will discuss it briey
here, but it is not used in the sequel. In particular I will not do this
in the lectures themselves. By denition our space C
(R
n
) (denoted
c
traditionally asD(R
n
))is a countable union ofsubspaces
(6.15)
C
(R
n
)= C
(B(n)), C
(B(n))={u S(R
n
);u=0 in |x| >n}.
c c c
nN
Consider
(6.16)
T ={U C
(R
n
);U C
(B(n)) is open in C
(B(n)) foreach n}.
c c
(R
n
) This is a topology on C contains the empty set and the whole
c
space and is closed under nite intersections and arbitrary unions
simplybecausethesameistruefortheopensetsinC
(B(n))foreach
n. This is in fact the inductive limit topology. One obvious question
is:- what does it mean for a linear functional u:C
(R
n
) C to be
c
continuous? Thisjustmeansthatu
1
(O)isopenforeachopensetinC.
Directlyfromthedenitionthisinturnmeansthatu
1
(O)C
(B(n))
41 LECTURE NOTES FOR 18.155, FALL 2004
shouldbeopeninC
(B(n))[Link],
restricted to each of these subspaces u is continuous. If you now go
forwards toLemma 7.3you cansee what this means; see Problem 74.
Ofcourse there is alotmoretobesaid aboutthese spaces; you can
nd plenty ofit in the references.
