• Aucun résultat trouvé

Solving linear partial differential equations with constant coefficients

Dans le document manuscript (Page 162-165)

7.5 Higher-Order models with Distributions

8.1.3 Solving linear partial differential equations with constant coefficients

8.1.3.1 Fundamental solutions

Among the solutions to LPDE’s, some are particularly studied: these are the fundamental solutions, that is solutions Eto the equation:

DE“δ0

where the parameter is a Dirac distribution. Because partial differential linear operators with constant coefficients behave particularly well with respect to convolution, the answer to this particular input is enough to compute the answers to any inputf.

Definition 8.1.11. A fundamental solution for the LPDODconsists in a distributionψPD1pRnq1to the equation:

Dψ“δ0.

Remark8.1.12. Notice that such a fundamental solution is in general not unique: ifφis such thatDφ“0, then ED`φis also a fundamental solution toD.

Example8.1.13. Because of linear partial differential operators, we are working with distributions whose support is not necessarily compact. Indeed, the existence of a fundamental solution is not ensured when distributions must apply to any smooth function. The typical example is

D:f PC8pR,Rq ÞÑf1,

wheref1is the derivative of the real, one-variable functionf. Iff has compact support, one can define:

ED:f ÞÑ ż0

8

f

and one has indeedDEDpfq “fp0q. This however is not possible in full generality whenf PC8pRnq. Observe that it would be enough for the partial derivative off to be bounded by an integrable function forxbig enough.

Then the derivative of f would be integrable, and ED could be defined. Extended to any partial differential operator, this says that if any partial derivativeBαf,αPNn off are integrable onRn, thenEDcan be defined.

This idea is implemented in the construction of a tamed fundamental solutionEDPS1pRRq[43].

Notation 8.1.14. ConsiderD : EpRnq //EpRnqa LPDO. When a fundamental solution forD exists and it is fixed, it is denoted byEDPD1pRnq.

Remark8.1.15. We will recall later in theorem8.1.18that whenDis a LPDO with constant coefficients there is always of at least one fundamental solution forD.

The resolution of LPDOs with constant coefficients is always possible, and particularly elegant, due to the behaviour of convolution with respect to partial differentiation.

Proposition 8.1.16. ConsiderD “ ř

αaαBα a LPDO with constant coefficients. Suppose that D admits a fundamental solutionED. Then for anyφPE1pRnqwe have:

DpED˚φq “φ. (8.3)

Proof. Thanks to the bilinearity of the convolution product DpED˚φq “ÿ

α

aαrBαpED˚φqs

“ pÿ

α

aαBαEDq ˚φthanks to equation8.1and the bilinearity of˚,

“δ0˚φby definition ofED,

“φ

Remark8.1.17. Beware that equation8.3is valid inD1pRnq, but not inE1pRnq. Indeed, the convolution between φPE1pRnqandψPD1pRnqonly yields a distributionψ˚φPD1pRnq. This is by construction of the convolution following proposition7.3.16. It is also a particular case of the theorem of support [42, Thm 4.2.4] which says that the support of convolutionφ˚ψof two distributionsφPD1pRnqandψPE1pRnqis contained is the sum of the respective supports ofψandφ. We do not emphasize on support and singularities of distributions here.

Theorem 8.1.18. Every LPDEcc admits a fundamental solutionEDPD1pRnq.

1note that we do not ask forψPE1pRnq, as this is not possible in general, even for the LPDO with constant coefficients

Comments on the proof. Several constructions of a fundamental solution exist: we refer to the one of Mal-grange, refined and exposed by Hormander [41, Thm 3.1.1] or to [42, 7.3.10] which gives a fundamental solution with optimal local growth. Others construct a fundamental solution which is temperate [43]. The proof starts with a technical Lemma [41, 3.1.1], which says that for any smooth function withcompact supportf P DpRnq, we have that

|fp0q| ďCkpcoshp|x|qDpfqk1, 1

degP (8.4)

whereCdoes not depend off. Thus ifDpfq “ Dpgq, one hasDpf ´gq “ 0andfp0q “gp0q. Then defines the functionEˇ onDpDpRnqqas EpDpfˇ qq “ fp0q. This function is well defined, linear and continuous. The above majoration allows then to extend this function toEPD1pRnqthrough Hahn-Banach3.3.6. The theorem is in fact much more precise as we have information about the local growth ofED. We do not have in general that EDPE1pRnq.

Remark8.1.19. Notice that equation8.4ensures thatEDis well defined onDpDpRnqq, as forf PDpDpRnqqthe functiongsuch thatf “Dgis unique. This is not true for functions inDpEpRnqq

8.1.3.2 The space!D

This section studies the question of which space is the good interpretation of!DRnfor " the space of distributions which are solutions to a LPDEcc "

Let us sum up the situation: the computationEDpgqis well defined as soon asgPDpRnq, as Theorem8.1.18 shows thatEDPD1pRnq. Iff PDpRnq, then obviouslyDpfˇ q PDpRnqand we have:

EDpDpfˇ qq “DEDpfq “fp0q.

Iff PEpRnq(that is,fdoes not necessarily have compact support) is such thatˇpDpfqq PDpRnq, thenEDpDpfˇ qq is well defined, and we have alsoEDpDpfˇ qq “fp0q. However, whenf PEpRnqandDpfˇ qdoes not have compact support, thenEDpDpfˇ qqisnot defined, and thus we can’t computefp0qasEDapplied toDpfˇ q. Let us export this this reasoning to distributions.

Definition 8.1.20. We denote by!DRnthe sub-vector space ofD1pRnqconsisting of the distributionsφPD1pRnq such thatDφPE1pRnq.

!DRn :“ tφPD1pRnq |DφPE1pRnqu

Example8.1.21. A typical example of distribution in!DRnisED, asDED“δ0PE1pRnq.

Proposition 8.1.22. Endowed with the topology inherited fromD1pRnq, the space!DRn is a lcs. Moreover, we have the topological embeddings2E1pRnqãÑ!DRnãÑD1pRnq.

Proof. If we considerDas linear continuous endomorphism:

D:D1pRnq //D1pRnq

then!DRn is the inverse image ofE1pRnqbyD. AsEpRnqis closed inDpRnq, and asD is continuous, then

!D is a closed sub-locally convex topological vector space of D1pRnq. It is Hausdorff: considerφ ‰ φ1 are both distributions, and ifDφandDφ1both have compact support. ConsiderV andV1 disjoint neighbourhoods D1pRnqofφandφ1 respectively. Then asE1pRnqis dense inD1pRnq,V XE1pRnqandV XE1pRnqare two non-empty open sets contained in!DRn whose intersection is empty and containing respectivelyφandφ1. The topological embeddings follows from the fact that ifφhas compact support, thenDφhas compact support: thus E1pRnq Ă !DRn. This embedding is topological asD1pRnqandE1pRnqcarry the same topology. The other embedding follows from the fact that by definition,!DRnis a sub-lcs ofD1pRnq.

Notation 8.1.23. We denote by?DRnthe lcsp!DRnq1endowed with the strong topology on the dual.

AsDcommutes with convolution (proposition8.1.10) and as the convolution of two distributions with compact support has compact support, we have immediately:

Corollary 8.1.24. For anyφP!DRnand anyψPEpRnqwe haveφ˚ψP!DRn.

Corollary 8.1.25. Forf P?DRnandgPC8pRnq,Dpf˚gqis defined asDf˚g. It is coherent with the notation whenf PDpRnq, as partial differentiation commutes with convolution (Proposition8.1.10).

2that is, the linear continuous injections

Analogy with DiLL Defining!D0Rnas the inverse image ofE1pRnqbyD0inD1pRnqdoes not make any sense.

Indeed, D0f is a linear function: it has compact support if and only if it is null. Thus with the definition used before, we have?D0Rn “ t0uand!D0Rn“Rn.

Remark8.1.26. AlthoughD0is not a LPDOCC,δ0kind of behaves as a fundamental solution toD0. Indeed, if a functionf PC8pRn,Rqis such that there isgPC8pRn,Rqsuch that:

D0g“f, thenfis necessarily linear, and thus one can takeg“f “δ0˚f.

The good definition for “the space of distributions solutions toD0ψ“φ“ is

?D0Rn:“D´10 pEpRnqq, and thus

!D0Rn»LpRn,Rq1 »Rn. In that case dereliction and codereliction rules are interpreted in NUCLby:

dD0 : !Rn //!D0Rn, ψÞÑ p`ÞÑψp`qq d¯D0 : !D0Rn //!Rn, φ“evxÞÑ pf ÞÑφpD0pfqq.

However if we look for a fundamental solutionED0PE2, it must satisfy:

ED0pD0fq “fp0q

and thus in particular, for any ` P E1, we haveED0p`q “ ED0pD0`q “ `p0q “ 0. Thus the analogy with fundamental solutions does not hold betweenD0 and LPDOcc, outside of the interpretation of dereliction and co-dereliction.

Outlook 16. One should explore the similarities between the work of Ehrhard on anti-derivatives [20], and the proof theory of fundamental solutions as described in this chapter.

Dans le document manuscript (Page 162-165)