Teoria das Distribuições (ENG) - Capítulo 4 by Casa Ciências

JOSE SEBASTIAO E SILVA

TEXTOS DIDAcTICOS

Volume III

SERVI<;O DE EDUCA<;AO E BOLSAS

FUNDA<;AO CALOUSTE GULBENKIAN

LISBOA

Reservados todos os direitos de acordo corn a lei Edic;ao da FVNDA<;AO CALOUSTE GULBENKIAN Av. de Berna - Lisboa

1999 ISBN 972-31-0971-9 Dep6sito Legal

n.O

148805100

111.1 THEORY OF DISTRIBUTIONS路

Este texto tern por base apontamentos coligidos por diversos alunos de Jose SebastHio e

Silva na sequencia de urn curso que realizou em

1958 na Universidade de Maryland, e que

posteriormente foram utilizados, e por ele revistos, na Faculdade de Ciencias de Lisboa.

C H A PTE R I V M U LTI P L I CATI O N A N D C H AN G E O F VA RIABLES

4.1 . Multiplication of a C n function by a Cn distribution

As we have seen, the concept of distribution was introduced in order to render the operation D always possible, though in a formal generalized sense. B ut as in the case of number theory, any advantage we gain in this direction is counterbalanced by the loss of some good properties. Multiplication of functions on the same interval is always fea sible; in particular the product of two continuous functions is again a continuous function uniquely defined. But it is not possible to define the product of two completely arbitrary distributions, as to guarantee a minimum of properties giving some interest to such a definition. We shall try to define the product of two distributions f and g on an interval I in IR, as to guarantee, at least, the three following conditions :

M 1. The product of two distributions f and g on I, when it exists, is

again a distribution on I (which can be denoted by fg or f · g ). M2. If f, g E C(/), then fg is the product of the functions f, g in the ordinary sense. M3. If the product f g and Df g exist, then f · D g exists and D (f g) = = f · D g + Df · g· .

We shall see next that in the case when f E C n and g E Cn ' the product is implicitly defined by these conditions . More precisely :

For any integer n· � 0, it is possible in one single way to assign to each couple (f, g) where f E C n (/ ) and g E CnC/), a distribution fg E Cn C/ ) not depending on n and satisfying the follow ing conditions: (i) If f, g E C(/ ), thenfg is the product of the functions f, g in the ordinary sense, (ii) If f E c n+l (/ ) and g E C CI ) . then D ( fg ) = f· Dg + Df · g . n By these conditions, if f E c n (/ ) and g = D n G with G E C(/ ) the dis tribution fg is, for each n, uniquely defined by: 4. 1 . 1 . THEOREM.

PROOF. a) We shall first prove, by induction on

n, that in order

for conditions (i) and (ii) to be satisfied, the product of f E C n by g E Cn is necessarily given by 4. 1 .2. This statement is obviously true for n = 0. Suppose it is true for n � 0, we prove it is also true for n +1. Let f E C n+l, g = D n+ 1 G, with G E C ; then by condition (ii) :

D ( f · D n G ) = f · D n+1 G + Df · D n G . Hence

4. 1.3.

f · D n+ 1 G = D ( f · D n G ) - f ' D n G .

Now, by the induction hypothesis, we have:

and

Hence by substitution in property

4 . 1.3 . and by applying the well-know

(:) + (k 1) (n ; ) n

we find:

So the statement is true for n + 1 and consequently for all n � O. n b) We now prove that for each n the product fg with f E C and g E Cn , is uniquely defined by 4. 1 .2. Suppose g = D nG = D nG* ; then G - G* is a polynomial function P of degree < n and:

c) Now we prove that fg does not depend on

= D n+ ! G , with DG

----

= G (in the ordinary

n. Suppose g = D rtG =

sense) . Then:

and since f (k) G = f (k)D G = D (f (k) G ) - f (k+ l) G , we find as we did in a) : ----

,.-....,

d) Condition (i) is obviously satisfied if we define fg by 4. 1 . 2 . , n = O . As to condition (ii), it is also implied by 4. 1 .2. , as can easily be proved by applying the property of the binomial coefficients as we did in a) . •

Therefore, in the case f E c n (l) and g E Cn (I), the natural defini tion of the product is given by 4. 1 .2.

The conditions formulated in 4. 1 . 1 . (equivalent to M 1, M2 and M3 in this particular case) were taken as a minimal request, in order to define implicity the product in this case. This product has most of the properties of the ordinary product except that it does not exist for all couples of distributions .

4. 1 .4. THEOREM. Given any integer n � 0, any two functions

qJ, ljf E C n (I ) and any two distributions f, g E C (I ) we have: n qJ f (j ) qJ + ljf f f + = ljf ( ) ( jj ) qJ ( f + g ) = qJ f + qJ g (jjj ) qJ ( ljff ) = ( qJ ljf) f ·

PROOF. For (j ) and (jj ) the proof is immediate. As for ( jjj ) observe that, to each pair of functions qJ, ljf E c n and to each distri bution f E C there is assigned the product qJ ( ljff ) E C is such a way n n that: (i) If f E C, then qJ ( ljff ) is the product ( qJ ljf) f in the ordinary sense. (ii) If qJ, ljf E c n+ 1 then:

D [ qJ ( ljff )] = ( qJ ljf) Df+D ( qJ ljf) . f · By theorem 4. 1 .4., this is possible only if qJ ( ljff ) = ( qJ ljf) f . So the proof is finished .•

Observe that, for any n > 0, the product qJ ljf is defined, for every couple qJ, ljf E C n and belongs again to C n; moreover, this operation

is associative and distributive (with respect to addition) and commu tative. Thus for n = O, 1 , . . . , Cn is a commutative ring. But C n is also a vector space over the field C, and multiplication of vectors f E C n by scalars A E C is related with multiplication of two vectors f, g E C n according to the rules :

A (fg) = ( A f )g = f ( Ag ) . All these facts can be expressed by saying that the ring C n is a commutative algebra over C. On the contrary, C , for n > 0, is not a ring, since the product fg n does not exist for all couples of distributions f, g E C • But C is a n n complex vector space and, on the other hand, there exists one and only one product qJf for each qJ E c n and f E C , with properties (j), n (jj) and (jjj). The conjunction of all these facts can be expressed by sayIng :

4.1.5.

For each n, C/I) is a module over the complex algebra c n (I) .

We denote by C OO (I), or simply C oo , the set of all infinitely diffe rentiable functions on I. Then C OO is the intersection of the C n and it

is again a complex algebra. On the other hand, we have adopted the symbol C (I), or simply oo Coo , as an alternative notation for the set !!iJ(I) of all distributions on I. So Coo is the union of all vector spaces Cn • As it follows from 4. 1 .5 .":

4. 1.6. COROLLARY. Coo (I) is a module over the complex algebra

C OO (I) .

Observe now that multiplication by complex numbers can be interpreted as a particular case of multiplication by C oo functions . In fact, to each A E C corresponds a constant function X E C oo , defined on any interval I by : -

A(X) = A for all x E I .

It is obvious that the correspondence A ---;;. A is a one-to-one map' ping of C onto a subset C of C oo such that if A , j.1, v E C, then:

A = j.1 + V � X = ji + v A = j.1V � X = ji v .

Af = Af for every f E !!iJ. Thus �e can identify each number A E cC with the corresponding function A E C oo so that the field C becomes a subalgebra of C oo. The number 1 is identified with the constant function 1 which is the unity element of C oo. Moreover

4.2. Extensions of the preceding concept of product. Examples We have seen previously (3 .6) how the product of a continuous function by a measure is defined. We have defined the vector space 91l n (l) of all distributions of order s n on l.

Then we can replace condition (i) of the theorem 4. 1 . 1 . by the stronger one: (i ' ) If f E C(l), g E 'IJTL (I ) . then fg is the product of the continuous

function f by the measure g as previously defined.

That being so, it is readily seen that theorems 4. 1 . 1 . and 4. 1 .4. can be immediately extended, by replacing C (I) by 'IJll (I), C(I) by n n 'IJTL(I) and (i ) by (i ' ). Thus :

4.2. 1. For each n, 'IJTL (I) is a module over the complex algebra c n (I).

We can analogously define the product o f a functi on f such that f (n) E 'IJTL, by a distribution g E C . Then fg E 'IJTL , but property Ujj ) in n n 4. 1 .4. requires , in the present case, the additional assumption that ( cp ljf) (n) E 'IJTL in addition to the hypothesis cp(n), ljf(n) E 'IJTL (which replaces the hypothesis qJ, Ijf E c n ) . Another similar possibility concerns the product of a function f s uc h that f (n) E L2 by a distribution g = D nG, with G E L2. Then fg is of the form D n (!J, with (!J E L l . Other variations can be imagined in a similar way. We have considered the product of a function by a distribution as though it were not commutative, but it is obvious that we did so o nl y for the sake of convenience.

4.2.2. In the preceding definitions of products, the order does not

matter; i. e., the product is commutative. We reach another natural extension of the concept of product by trying to satisfy the following supplementary condition, which is of course satisfied in the preceding cases:

M4. If f and g are two distributions on an interval I and if fg exists, then the product of their restrictions to every subinterval J of I exists

and:

Suppose I is represented as the union of a system (la) of open intervals. Denote by fa and ga the restrictions of f and g respectively

to la ' and suppos e that fa g a exists according to one of the preceding definitions, (regardless of order) . Then, placing fa g a = ha ' one easily se es that: 1 ) h a hp on la n Ip ; 2) there exists an integer v such that the rank of h a is less than v for all a . Therefore by the Collecting principle (2. 8.) there exists one (and only one) distribution h such that h = h a on each la . It is natural to plac e h = fg and it is readily seen th at this new concept of product s atis fies M 4 as well as the preceding properties of the product hold足 in g for each interval la 揃 We can, of course, consider any open subset Q of IR instead of the interval I, and even two global distributions instead of simple distributions . We have already seen (3 .6. 1 .) that f8= f(0) 8 for any continuous function f on IR . Suppose now that f E e n (lR). Then by formula 4. 1.2. (with G E 011 ) and 3 .6. 1 . , we obtain : =

4.2.3.

More generally, for any a E IR:

Observe now that condition f E e n (lR) is not necessary for the existence of f8(1l) ( x- a) ; for that is sufficient, according to the last ex足 tension, that the restriction of f to some neighborhood of 0 be a e n function . (This condition can even be enlarged by using the concept of value of a distribution at a point defined later on) . In particular:

and for a = b, this expression has no meaning according to the pre足 ceding definitions . However physici sts frequently consider such products as 88, 88' , etc .

4.3. Impossibility of defining an associative multiplication for arbitrary distributions

We are going to show that: 4.3. 1. It is

impossible to assign to every couple ( f, g) of distributions a distribution fg as to satisfy the following: (i) If f, g E C(I) then fg is the ordinary product, (ii) D ( fg) = Df · g + f · Dg , Vf, g E §"(I), (iii) ( fg)h = f(gh) , Vf, g, h E §"(I), PROOF. Suppose we have a multiplication satisfying ( i) and (ii), and consider the distribution

(PI

Pf � = D log I x I (cf. 3 . 3 .) � Then x

! ) x = (D log I x l) x = D(x log I x I ) - log I x 1 = .

= D(x log l x l) - D(x log l x l -x) = 1 .

Hence:

(P I

On the other hand, we have x .

�) (xO) = 0.

Consequently, conditions '"

( PI

�) (xO)

0= 0 . 0= 0 and therefore

(iJ and (iiJ imply

, which contradicts

(iiiJ.

This argument works for any interval sider the restrictions of Pf

[( P I

I containing 0, if we con

� and 0 to I, and can be extended to any

x interval in IR by a suitable translation of Pf � and x

proved • .

� ) . x ] 0 ", 0. So 4. 3 . 1 . is

It is clear that 4 . 3 . 1 . continues to be true if we consider only the sp ace of all distributions of the form f =DF, with F E L (/) instead of th e space !W(/) . It sh ould be observed that difficulties connected with the concept of product are already found in the space L (I) since the product of two locally summable functions would not be locally summable. For example, the square of the locally summable function

which corresponds to infinitely many distributions on fR.

� is -.Lx ,

"v x

H. KONIG proved that it is possible to construct in infinitely many ways an extension � (l) of .2?(/), � ith the linear operator D, so that to each pair ( f, g) of elements of !W(/) , is assigned an element fg of � (/) as to satisfy some conditions like M2, M3 and M4. However it must be observed that: (I) in such an extension the product of two elements of !W(/) is not necessarily in !W(/), ...... (11) the product of two elements of !W(/) does not in general

exist, (Ill) multiplication is not associative, (IV) it is not possible to find, among all extensions, one that is "minimal " up to an isomorphism.

Hence there is an irreducible indetermination in defining the product of two distributions.

4.4. Linear differential operators

Let ao ' at " . . , an be n + 1 functions on 0, an open set in fR. The

� n

linear differential operator the formula: 4.4.1.

ajDj (of order

n) is usually defined by

where f i s any function having a derivative on 0, in the ordinary sense, of order n. It is obvious that the same operator can be extend ed to any distribution f on 0 for which all terms exist. In particular, whenever the ajE C oo ; Then, the sum and product of two operators A and B of this form, defined by

(A + B )f = A f + Bf, (A B )f = A (Bf), 'Vf E !?lJ(O ) , are again operators of the same form. Moreover, it is easily seen that:

�.4.2. The set Q, o/all opera:ors o/the/orm IS

a complex non-commutative algebra. The nth power,

� ajD j with ajE C OO (O), n

All, of an operator A E Q is defined by :

As a commutative sub-algebra of Q, it should be mentioned the algebra Q * of all linear differential operators of finite order with constant coefficients, i.e. , the algebra of all operators of the form

�J ajD j, n

where

ajE C.

As we did for functions (4. 1 . 6 . ) , we can identify each A E C with the operator A Do , where DO is the identity operator, so that C becomes a sub algebra of Q* . For example:

(D 2 + 3)f = D2f+ 3 f , 'Vf E !?lJ D2 + 3 = (D + i V3 )(D - t V3 ) = (D - i V3 )(D + i V3 ), etc . If the coefficients

aj of an operator A =

�J ajD j n

are not all in

C OO(O), then A is not defined (in any sense defined until now) on every distribution on O. In this case, instead of the space of distributions,

we can conceive other extensions of the space of continuous func tions by introducing new entities (which we shall call generically para-distributions), as to render the operator A always defined.

This method is quite similar to the one for distributions in 2.1. Finally, w e can consider linear differential operators whose coefficients are either distributions, like g and its derivatives, or functions which are not distributions, such as

-.L , e ! , etc . . x

4.5. Change of variable in distributions

Let us consider a complex-valued function f defined on an open set O C IR and a real-valued function h, which maps O * C IR into O. Then h assigns to each point t E 0* , a point x=h (t) in 0; in turn, f assigns to each x the complex number f(x)=f(h (t)). The corre spondence t � f(h (t)) is a complex-valued function defined on 0* which is called .the composition of f and h and denoted by f h. Thus ( f o h)(t) = f (h (t)) . The final operation f � f h is said to be the change of variable (or the substitution) defined by h. In particular, this operation is feasible for all continuous functions f on O. Moreover if f and h are both C l functions, we .c an apply the chain rule : 0

� f(h(t))=f' (h(t))h' (t) dt or 4.5.1.

( f o h) ' = h' ( f ' o h) .

It should be observed that in this formula, two derivation operat0rs are involved; one operating on functions f E e l ( 0) and the other on functions h E C l ( 0 * ) . For the sake of convenience, we denote the first by Dx and the second by Dt • Thus we can write 4.5 . 1 . as follows : 0

Df ( f h) = h' (Dx f h) whence, supposing

�

h' (t) 0 for all t E 0* ,

(Dx f ) o h =

4.5.2.

D f o h) . --;;;- r C

This formula can be expressed by saying:

4.5.3. The change of variable defined by h transforms Dx into the

differential operator 1 Dt h' __

•

More generally, let f =DnF, FE C( 0), be .a distribution. By 4.5.2., we are induced to write formally:

(D; F o h) = (

�, D, )"cFO h) .

Justification of this is given by the following

4.5.4. THEOREM. IfF E C(O) and h E C maps 0 * into 0 in such a

n way that 1 E c n (o *), then the expression ( 1 Dt ) (F O h) denotes h' h' a distribution in Cf) (O *). Moreover, ijD;F = DxmG, GE C(O) we have __

again

( �, D}FO h) (t DJ(G h) =

PROOF. The first part is proved by induction. The statement is obviously true for n = O. Suppose it is true for n � 0 and assume 1 __

E c n+l(O *).

Then, since

+l 1 1 ( - D ) n (Fa h) = D h' t h'

[(- Dt )n(F o h)] 1

and the right side is the product of the function

�, E C " + I (0 *)

by a

C + ! (0 *), the statement is also true for n + l . Hence it n is true for every integer n � O. distribution in

The second part we can reduce to the case where 0 is an inter足 val. Suppose D;F = D'; G, with m ";! n and place l/J = r;sm-nF; then v m l/J = D m G and l/J - G= P is in f!J> ' By 4 . 5 . 2 . , we find

1 D ) n [ (Dxm - n l/J ) o h J = (_ 1 D ) m (l/J O h) = ( _ (_ 1 D ) n (F O h) . h' t h' t h' t

On the other hand:

Hence

DEFINITION. Under the hypothesis of theorem 4.5 .4 . , we shall write:

4.5.5.

and we say that the distribution f h , defined by this formula, is the composition of the distribution f with the function h. We sometimes use the notation f(h( f )). 0

( i)

(ii)

(iii)

Now the following are easy to verify:

Iff E C(O) and h E C(O *), then f o h is the compound off with h in the ordinary sense.

If f E C.(O) and (Chain rule).

+ E Cn +l(O *), then D, U o h) = h' (Dx f o h).

If f, g E C (O), A E C and _ 1 E cn(o *), then n h'

( f+g) o h = f o h + g o h and (Af ) o h = A( f o h) .

(iv)

If f E C/O) and h, k are C l mappings of 0 * into 0 and of 0 * * into 0 * respectively, such that 1 E cn(o *) and 1 E cn(o * *), h' k' then ( f o h) o k = f o ( h o k) . (Associative law) . __

Examples I. Translation operators are the most simple examples of change of variable . In fact the translation 'fa f of a distribution f is the distribution f(f - a) , composition of f(x ) with the function

x = t- a . 11. For any real k ;l! 0, and n > 0, we have: 4.5.6.

Indeed, putting x = kt, we find:

since H(kt)

H(t) or H(kt) = 1 - H(t ) , depending on whether k > 0 or

Ill . Let us see whether the change of variable defined by

k < O.

X = t 2- c2 , with c > O, is feasible on 8 (x). The function h (t) = t 2- c2 maps IR into

[-c2 , + 00 [. Also, h' (t) = 2t, and

h' (t)

= 1 is a __

C oo

on the open set 0 of all t ;l! 0 in JR. Hence the distribution 8(t 2- C2 ) of t is defined on 0 and:

function

Now it is easily seen that:

H(t 2- c2 ) = H(t-c) + H(- t - c)

if if

-c < t< c t < - c or t > c

Hence: 1 1 5: (t 2_ C2 ) = - [Dt H(t-c) + Dt H(- t-c)] = - [D H(t- c) -D H(- t-c)] = 2t 2t

= - [ 8(t- c) - 8(- t-c)] . 2t

But 8(- t- c) = 8(t + c) (cf. 4.5 .6.) and by 3 . 6. 1 . : 1 1 - 8(t- c) = - 8(t- c) , 2t 2c

_ 1 8(t + c) =

1- 8(t + c) .

_ _

Consequentely : 4.5.7.

1 8(t2- C2 ) = - [ 8(t- c) 2c

8(t + c)] for t ďż˝ O.

REFERENCES [ 1 ] H. KONIG. Multiplication und Variablentransformation in der Theorie der Distributionen . Arch. Math. 6 ( 1 955). [2] H. KONIG. Multiplication von Distributionen I. Math. Annalen 1 28 ( 1 955) 420-452 . (Maths . Reviews 1 9-935) .