Generalized power series solutions to
linear partial differential equations

by Joris van der Hoeven

Université Paris-Sud

Département de Mathématiques

Bâtiment 425

91405 Orsay Cedex

France

March 9, 2006

Abstract

Let and , where is an effective field and and are given a suitable asymptotic ordering . Consider the mapping , where . For , it is natural to ask how to solve the system . In this paper, we will effectively characterize and show how to compute a so called distinguished right inverse of . We will also characterize the solution space of the homogeneous equation .

1Introduction

A well-known theorem [Fab85] states that any linear differential equation over admits a basis of formal solutions of the form

with , , and . This theorem naturally generalizes to the case when is replaced by an effective algebraically closed field of coefficients . If we also replace the coefficients by polynomials in , then several algorithms exist for the computation of a basis of solutions [Mal79, DDT82, vH97].

There are several directions in which the above theorem may be generalized. In [vdH97, vdH01, vdH06], it is shown how to deal with so called transseries coefficients (a transseries is an object which is constructed from or and an infinitely large variable using exponentiation, logarithm and infinite summation). In collaboration with M. Aschenbrenner and L. van den Dries, we are currently working on a generalization to arbitrary asymptotic fields (an asymptotic field is a differential field with a total asymptotic ordering which is “naturally compatible” with the derivation).

In this paper, we will be concerned with the generalization to the case of linear partial differential equations. The asymptotic resolution of systems of such equations can be decomposed into two subproblems: the computation of analogues of the exponential parts and the computation of the corresponding coefficients. We intend to deal with the first subproblem in a forthcoming paper and focus on the second subproblem in what follows.

In the case of holonomic systems of linear differential equations, algorithms are known for the computation of formal and convergent generalized series solutions [SST00, Chapter 2] in what the authors call “Nilsson rings” [Nil65]. On the other extreme, there exists a method [AC01] to find “fractional power series solutions” to a single p.d.e. with coefficients in . In this paper, we will search for formal series solutions to consistent systems of linear differential equations in variants of Nilsson rings of the form . One of the major difficulties is to cope with the integrability constraints which arise when considering more than one equation.

In fact, in the continuation of our previous work on transseries, we will rather work with infinitely large variables and series in . In this equivalent setting, our linear differential operators belong to and we consider series in where . More precisely, we assume a total asymptotic ordering on and consider so called grid-based series [vdH97, vdH06] with monomials in and coefficients in .

In sections 2 and 3 we first recall classical algorithms for the computation of “standard bases”, which are used to reduce a system of equations like with and to suitable normal forms. The first algorithm is a variant of the skew version [Cas84, Cas87, Gal85, Tak91] of Buchberger's algorithm [Buc65, Buc85], although we rather compute coherent autoreduced sets in the sense of differential algebra [Ros59]. We also recall Mora's standard cone algorithm [Mor83, MPT92]. However, we will systematically present them in the setting of p.d.e.s with second members, so the reader might at least want to take a look at the notations. Also, corollaries 7 and 12 characterize when a system of equations with second members satisfies the necessary integrability constraints which ensure the existence of a solution.

In section 4, we will start with the study of linear p.d.e.s with constant coefficients in . It is classical that the resolution of such equations in is equivalent to finding the roots of a set of polynomial equations in . In particular, solution sets in correspond to radical ideals in . More generally, we will show that there exists a correspondence between the solution sets in and arbitrary ideals in .

An important technique that we will use is the computation of so called “distinguished solutions” to systems of equations with second members. More precisely, given , we may consider as an operator . Denoting , we will effectively construct a right-inverse of . This right-inverse is unique with the property that the coefficient of any in any vanishes, where denotes the set of dominant monomials of solutions to . Having constructed , we will also show how the space of solutions to can be obtained from .

In the last section 5, we will study the case of linear p.d.e.s with coefficients in (for effective purposes) and (for theoretical purposes). We will first show how to reduce systems of such equations to suitable asymptotic normal forms. Given a system in normal form, we will next show how to compute a distinguished right inverse in a coefficientwise manner. We will also characterize the set in this context and give an explicit “strong basis” for .

Remark 1. Section 5.2 in particular contains a skew version of Mora's tangent cone algorithm. One of the referees pointed us to another such algorithm, which appeared recently [GOT05]. Besides the fact this alternative algorithm is applied to another problem (ideal membership and the computation of sygyzies), it is also a bit different in spirit: whereas our algorithm uses a twisted version of reduction (which enforces good properties for the ecart), the algorithm in [GOT05] is based on homogenization.

2Standard bases for admissible monomial orderings

2.1Monomial orderings

Consider the “monomial monoid” , whose elements are of the form , with . A total ordering on is called a monomial ordering, if it is compatible with the multiplication, i.e. . It is classical [Rob85] that any such an ordering is non uniquely determined by a finite sequence of vectors and

(1)

Here denotes the scalar product. Clearly, the relation (1) allows to extend to and even . Moreover, this extension is unique so as to preserve the compatibility with the multiplication.

We say that is admissible if for all . In that case, extends the (partial) divisibility ordering on . In particular, from Dickson's lemma, it follows that is well-ordered. Given a subset , we will denote by the final segment of generated by for the divisibility relation. We recall that each final segment is finitely generated.

Let be a constant field of characteristic zero. Given a monomial ordering on , a non-zero polynomial and a monomial , we denote by the coefficient of in . We also denote by the highest monomial for occurring in and by the corresponding coefficient. We call the dominant monomial of , its dominant coefficient and its dominant term. The relation naturally extends to by . We denote by the equivalence relation associated to , so that . Similarly, we write if , which is equivalent to .

2.2Differential polynomials

Let be a differential field with derivations and field of constants . Given formal variables , we denote by the differential algebra of differential polynomials in over . Any admits a unique decomposition

where each is homogeneous of degree . We denote by the space of homogeneous polynomials of degree and .

Given and a tuple , the substitution of for () in yields a new tuple of differential polynomials in , called the composition of and , and denoted by . If and , then . In particular, is an algebra for and is a subalgebra of whenever and are subalgebras of with . If , we will denote by the unique elements of , such that .

Example 2. If

then

Example 3. If are differential operators, then

In other words, is isomorphic to .

In the remainder of this paper, we will only study differential polynomials with second members with and as above (and often ). Formally speaking, the monomial monoid for is isomorphic to . Consequently, the sets and are isomorphic as vector spaces (but not necessarily as algebras, except when ). This isomorphism induces natural definitions of , and for and of , , and on . These definitions naturally extend to , by taking for all . For instance, , if .

In our context of linear differential polynomials with second members, a differential ideal of is a -subvector space which is stable under (i.e. left composition with ). Moreover, if , then we require that . Any tuple naturally generates a differential ideal . If , then . When seeing as a system of equations , where belongs to any differential -algebra , then these equations are equivalent to . In particular, we say that a second system is equivalent to , if .

In the sequel, it will be convenient to extend notation for sets to tuples. For instance, and if is smallest with and if no such exists.

2.3Ritt reduction for linear equations with second members

Assume that we fixed an admissible monomial ordering on and denote . Let with , so that for some . The partial reduction of w.r.t. is defined by

(2)

Given a system , let . A normal form for modulo is an , with or , and such that

for certain (where ) and with for all . In that case, we write . We say that is autoreduced if for all . By using partial reductions of w.r.t. members of as long as possible, one obtains a normal form with :

Algorithm

Input: and

Output: a normal form of modulo

while do

return

Given , let be such that and . Setting , and , the -polynomial of and is defined by

(3)

By construction, be have . We also notice that , whenever . We say that a system is coherent if for all . A coherent and autoreduced system will also be called a standard basis. Given an arbitrary system , the following classical algorithm computes a standard basis which is equivalent to .

Algorithm

Input:

Output: a standard basis which is equivalent to

while

if then return

while do

if then

return

Remark 4. If , then under the natural isomorphism of with , the notions of partial reduction and -polynomials correspond to reduction and -polynomials in Buchberger's algorithm (up to details: Buchberger rather takes . Also, he not only reduces the dominant term of in , but all terms). Consequently, Buchberger's algorithm for computing a Gröbner basis [Buc65, Buc85] corresponds to the above algorithm for computing a coherent autoreduced set. Coherent autoreduced sets were first introduced by Rosenfeld [Ros59] and they are similar (although more effective) to the characteristic sets introduced by Ritt [Rit50]. We opted for Hironaka's name standard bases here [Hir64] in view of the generalization in the next section.

2.4Theoretical properties of standard bases

Consider a standard basis . Then the reduction of each -polynomial with to zero yields a relation

with for all . This relation may be rewritten as

(4)

with . We call (4) the critical relation for the pair . Notice that we may regard the set of all critical relations as a tuple .

Lemma 5. Let be a standard basis. Then the generate the space of all with . In other words, given with , there exists a with .

Proof. Assume for contradiction that there exists a relation which is not generated by the . We may choose such that is minimal, as well as the number of with . Since , there must be at least two indices and with . Using the fact that divides , let be such that , and . By construction, and , so and . For all , we also have , so . It follows that the relation is smaller than the original relation in the sense of the minimality hypothesis. This contradiction completes the proof.

Consider a system of linear differential polynomials in . Given a tuple , we say that is compatible with , if for every relation with , we have . The set of such tuples forms a subvector space of , which we denote by .

Corollary 6. The system with and is a standard basis if and only if is a standard basis and is compatible with .

Proof. Assume that is a standard basis. Consider with and . Then if and . It follows that is a standard basis with critical relation for all . Given a relation , lemma 5 now implies for a certain . We conclude that , whence .

Assume now that is a standard basis and that is compatible with . Then is autoreduced, since for all . Furthermore, for all , the relation implies . But the relation precisely proves that reduces to zero modulo . Hence is coherent.

Corollary 7. Given a standard basis and , we have if and only if .

2.5Canonical forms

Let and . A canonical form for modulo is an with

for certain and with for all , and such that for each term occurring in . It is easy to modify so that it computes a canonical form of modulo with :

Algorithm

Input: and

Output: a canonical form of modulo

while do

Choose highest for

return

Lemma 8. Let be a standard basis. Then we have

where and .

Proof. Assume for contradiction that is such that . Replacing by , we may assume without loss of generality that is a canonical form w.r.t. . Now choose with such that is minimal, in the same sense as in the proof of lemma 5. Since , we have , so there must be at least two indices and with . Setting with the notations from the proof of lemma 5, then yields a more minimal representation for . This contradiction proves that for all .

3Standard bases for tangent cone orderings

In classical polynomial elimination theory, the use of non-admissible monomial orderings allows for the computation in localized rings and completions, such as rings of power series. However, additional care is needed in order to ensure termination. For instance, the naive reduction of modulo would yield an infinite sequence . The tangent cone algorithm [Mor83, MPT92] allows for the computation of standard bases in the case of localizations of polynomial rings.

In this section, we will present the tangent cone algorithm in the differential setting. In all what follows, is a differential field with constant field . Geometrically speaking, elements of or localizations of can still be thought of as operators. For instance, naturally operates on .

3.1Definition and properties of the ecart

Let and let be a monomial ordering on . Given , we define , and as in (3). As a special case, is given by (2) if . Now let be the opposite ordering of . Given , we denote the dominant monomial of for by for and we define , , , , etc. in a similar way. We will also write for the element of with . If is admissible, and for all , then we notice that (i.e. for the natural extension of the ordering to ). Moreover, if , then .

In the sequel, we will assume that the vectors which determine using (1) are all in . In that case is called a tangent cone ordering. Notice that it is possible to consider more general tangent cone orderings [MPT92], but we have chosen to keep the exposition as simple as possible. Given , let

Given with , we denote

Notice that and (for a dummy ). Now let and be such that and . Then we have and we define the -th ecart of by . We call the ecart of and recall that is well-ordered by the lexicographical ordering. The definition extends to the case when by taking for all .

Given , some easy properties of the ecart are

(5)

Moreover, if , then

where the inequality is strict whenever . It follows that

(6)

In particular, if , then

(7)

The following lemma will guarantee the termination of the tangent cone algorithm.

Lemma 9. Let and be such that for all , we have

  1. for some .

  2. Whenever for some , then .

Then the sequence is finite.

Proof. Assume for contradiction that there exist infinitely many with . By Dickson's lemma, we may find two such indices with and . But then

which contradicts our assumption (b). It follows that is strictly decreasing for sufficiently large . We conclude by the fact that is well-ordered.

3.2The tangent cone algorithm

Given , a normal form for modulo is an , with or , and such that

for certain and with and for all . Notice that this notion extends the previous notion of normal forms, since if is admissible. In our new context, we may use the following algorithm to compute a normal form:

Algorithm

Input: and

Output: a normal form of modulo

while do

Take with such that is minimal

return

Indeed, the sequence of successive values of during the algorithm fulfills the conditions of lemma 9, so this sequence is finite. Moreover, using induction, it is easily checked that there exist and and with and for all . So the last term of the sequence is indeed a normal form for modulo .

Defining the notions of autoreduced systems, coherent systems and standard bases as in section 2.3, the same algorithm may be used to compute an equivalent standard basis for a given system. Given a standard basis and , we have a relation

with and for all . As before, we may rewrite this relation as a critical relation of the form .

In order to generalize lemma 5, let be the set of series with well-ordered support . If is admissible, then coincides with . If is admissible then elements of are power series in applied to . The set is naturally stable under composition. We denote .

Lemma 10. Let be a standard basis. Then the generate the space of all with .

Proof. We have to construct with , where corresponds to the critical relation . For each , let . Writing , let us construct by transfinite induction over . Given an ordinal , the induction hypothesis is as follows:

If or is a limit ordinal, then we may take . If and , then we are done. So assume that and . Let and let be minimal such that . Let , with

let and take for all with . By construction,

for all . Since , it follows that as well. This proves the last induction hypothesis. By transfinite induction, we conclude that there exists an with , whence .

Consider a system of linear differential polynomials in . Assume also that naturally operates on a subring of (for instance, we may take ). Given a tuple , we say that is compatible with , if for every relation with , we have . The set of such tuples forms a (strong) subvector space of . The following consequences of the above lemma is proved in a similar way as corollaries 6 and 7.

Corollary 11.

The system with and is a standard basis if and only if is a standard basis and is compatible with .

Corollary 12.

Given a standard basis and , we have if and only if .

Let and . A canonical form for modulo is an with

for certain and with and for all , and such that for each term occurring in . Although we have no algorithm to compute canonical forms, like in section 2.5, the existence of canonical forms can be proved using a similar transfinite induction as in the proof of lemma 10. Using another transfinite induction, lemma 8 also generalizes to the current setting:

Lemma 13.

Let be a standard basis. Then .

4Linear differential equations with constant coefficients

In this section, we consider systems of linear partial differential equations in one unknown with coefficients in a field of constants of characteristic zero. We will consider the resolution of such systems in the algebras

where (Kronecker symbol). We will first consider homogeneous linear differential equations, but we will also study linear differential equations with second members. In the latter case, we will allow the second members to belong to or . Throughout this section stands for an admissible tangent cone ordering on .

4.1Solving in

In this section, we will only consider linear p.d.e.s without second members. Let be a homogeneous linear differential polynomial. We may represent as

where is a polynomial in . Inversely, each polynomial gives rise to a homogeneous linear differential polynomial . Denoting , we have

for all and in particular

Let denote the set of all with . We have

for all , where denotes the -vector space generated by . Given , we will denote .

More generally, given a set of homogeneous linear differential polynomials, a subset of , a subset of and a subset of , we denote

and

Because of the natural isomorphisms

all algebraic geometry properties of the correspondences and induce analogue properties for the correspondences and . In particular, Hilbert's Nullstellensatz implies

Theorem 14. Let be a coherent and autoreduced system with . If is algebraically closed, then admits a solution .

4.2Solving in

Recall that stands for an admissible tangent cone ordering on . Consider a standard basis for . We may regard as an operator from into , whose image is in . We denote by the set of monomials , such that for all . The aim of this section is to construct a right inverse of , which is “distinguished” in the sense that for all and .

The relation on induces a relation on by

Whenever are such that and , it follows that

Indeed, if , then for at least one with .

Proposition 15. Given a standard basis for and , let be such that is maximal for . Then does not depend on the choice of and .

Proof. We will first show that whenever is another index with . Let and be such that and consider the associated critical relation

with for all . Since is compatible with , it follows that

For each , we have

It follows that

(8)

Hence

It follows that

Now implies , so we conclude that .

It remains to be proved that , i.e. for all . If , then and for all with , we have . By strong linearity, it follows that . Furthermore implies , whence . If , then the relation (8) remains valid. Moreover, if is such that , then

and

Since , it follows that , whence . By construction, we therefore have .

Given , let be the term as in proposition 15. Now consider the sequence defined by and . This sequence is finite, since and is a well-ordering on . Consequently, is a solution to with for all . We set and call the distinguished right inverse of .

Let . Since for all , it follows that . Consequently is a solution to with . Inversely, implies , since otherwise for some and . We claim that the form a basis for the solution space of in . Indeed, given an arbitrary solution , consider the sequence defined by and as long as . This sequence is necessarily finite, since and is well-ordered. Hence, . We call the distinguished basis of .

We notice that , where , so that decomposes into an isomorphism on with left inverse and the zero map on . We also notice that the distinguished right inverse is uniquely determined by the fact that for all and . Indeed, assume that and and for all . Then for and for all . It follows that .

Let us now consider an arbitrary system . Using the tangent cone algorithm, may be rewritten into an equivalent system which is a standard basis. Then the sets and are independent from the particular choice of , since is precisely the set of elements which cannot occur as dominant monomials of elements in , by lemma 13. Consequently, the construction of the distinguished right-inverse and the distinguished basis do not depend on the choice of , and we may define , , etc.

4.3Solving in

Let us now consider a general system as an operator . Then acts “by spectral components” . More precisely, given , let be the unique operator such that

for all . Considering as an operator in , we obtain from by substituting for each . Given

with , it follows that

Hence, denoting by the solution space of for , the solution space of for is given by

Denoting by the distinguished inverse of as an operator on , the mapping

is a right-inverse of . Moreover, is unique with the property that , where

Remark 16. When extending the total ordering on to in any way which preserves spectral components (i.e. if , then for all ), the space coincides with the set of all such that for all ; see the next section.

Theorem 17. Let be the set of differential ideals of and let the set of subsets of which occur as zero-sets of systems . Then the correspondences

are mutually inverse bijections.

Proof. Let and be two differential ideals with the same set of solutions . Then the differential ideal generated by and still has the same set of solutions. Assuming for contradiction that , the set strictly contains or , say . Now consider the differential ideal . By theorem 14, there exists an with . Since and (here stands for , where is any system which generates ), it follows that . But then and .

Remark 18. Whereas Hilbert's Nullstellensatz establishes a correspondence between radical ideals and algebraic sets, theorem 17 yields a correspondence between any differential ideal of (which is necessarily radical and even prime) and “linear differentially algebraic” zero-sets in . Via the isomorphism , arbitrary ideals of are therefore also in a geometric correspondence with zero-sets of linear differential operators. This provides a geometrical reason why the existence of Ritt-Rosenfeld-Buchberger-type algorithms for the computation with ideals, and not merely radical ideals, is important.

5Equations with polynomial coefficients

The study of the linear p.d.e.s with coefficients in is equivalent to the study of equations with coefficients in modulo the substitutions , and multiplication with a suitable . Since the ordinary partial derivatives preserve the “valuation” in , it will be more convenient to work with coefficients in .

Assume that we have fixed an admissible ordering on , determined by . Assume also that we have fixed a total ordering on which gives the structure of a totally ordered -vector space. Then we also have a natural ordering on :

A subset of is said to be grid-based if there exist with and . Given a ring of coefficients the set of series with grid-based support forms an -algebra [vdH97, vdH06]. We denote this algebra by and its elements are called grid-based series. This still goes through for coefficients in , since such operators act by spectral components. In this section, we will consider systems of linear p.d.e.s in and study their solutions in .

5.1Skew standard bases

The admissible orderings on and on may be combined into a total admissible ordering on using

Consequently, an element can also be regarded as a series with anti-well-ordered support in (the support is not necessarily grid-based, although we might have required this). Similarly, elements in can be seen as series with monomials in . The ordering is extended to by understanding that for all . We will use in order to emphasis when a notation should be understood with respect to the relation .

Consider a system such that for all . Given with and , let , , and

We say that is a standard basis for if for each there exists a critical relation

(9)

where is such that for all .

Given with (or with ), let us denote (resp. ).

Lemma 19. Let be a standard basis and let be such that . Then is a standard basis and .

Proof. Since for all , the system is autoreduced. For all , the relation (9) implies

so is a standard basis for the relations . Now consider a relation . Then we have for some . Now implies . We conclude that , so .

Lemma 20. Let be a standard basis and . Then is again a standard basis.

Proof. Any satisfy the relations

Hence, any critical relation for induces a critical relation for . So we may take .

5.2Computation of skew standard bases

Given an arbitrary system , an equivalent standard basis can be “computed” by a variant of Hironaka's infinite division “algorithm”. If the dependency of in is only polynomial, then a fully effective method can be devised, by adapting the algorithms from section 2.3.

In this subsection and in this subsection only, let , , and . The set is formally isomorphic (as a vector space) to by sending each to and to . Moreover, the ordering on corresponds to a tangent cone ordering on . Consequently, the definition of ecart in section 3.2 transposes to elements in .

Unfortunately, we do not necessarily have for and (for instance ). Nevertheless, this relation does hold if . For this reason, we adapt the definition of partial reduction by setting

for all with . Because of the twist, we again have

We also notice that coincides with the usual partial reduction “up to lower order terms”, since . We obtain the following version of :

Algorithm

Input: and

Output: an “asymptotic normal form” of modulo

while do

Take with such that is minimal

return

The termination of the modified version of NF is proved in the same way as before. Again, the successive values of in the algorithm verify relations

for certain , and with and for all .

Example 21. Let and . Then

Hence divides , from the asymptotic point of view.

In a similar way, we may define the twisted -polynomial of by

Given a system , the corresponding algorithm now computes an equivalent system , such that for all we have a relation

where , and are such that and for all . But admits as inverse in , which leads to the relation

(10)

Moreover, each induces an element

with . When rewriting (10) in terms of and , we obtain a critical relation for and in the sense of section 5.1. Modulo this normalization of the result, SB therefore computes a skew standard basis.

5.3Theoretical properties of standard bases

Let again , and . Let . Using the isomorphism , we observe that , , etc. are well-defined. Given it is also convenient to extend the notation by setting if and only if .

Lemma 22. Let be a standard basis for and let be such that . Then there exists a with and . In particular, if , then there exists a with .

Proof. Let with be such that . Then is stable under composition. For each with , let us show how to construct , such that satisfies . We use weak induction over .

So let and assume that has been constructed for all . Let . Since for all with , and , we have . Setting , we have , so for some . Taking , it follows that .

By induction, we conclude that is well-defined and we have for all , so .

Corollary 23. Given a standard basis and , we have if and only if .

Proof. Similar to the proofs of corollaries 6 and 7.

Corollary 24. Assume that is a standard basis for and let be non-zero. Then .

Proof. Let be such that . Modulo division of by , we may assume without loss of generality that . Let be as in the above lemma, so that . In fact, , since implies . We conclude that .

5.4Solving in

Consider a standard basis for . Given , we may regard as an operator on . We denote

and write for the distinguished right inverse of .

Proposition 25. Let be a standard basis for . Then admits a unique right inverse such that for all .

Proof. Let with and be such that and . For any with , it follows that . Let us show by well-ordered induction over how to construct such that for .

Given , we assume that has been constructed for all with . Denoting , we also assume that for all with . By construction, we first observe that , whence . Now we take , which is well-defined by lemmas 20 and 19. Setting , it follows that . For all with , we also have . We infer that . By induction, we obtain a series with and for all . We conclude that . The uniqueness is proved as usual.

Proposition 26. Let be a standard basis for . For each , let . Then

for all solutions to .

Proof. Setting

we have

Now for all , by the distinguished property of and the fact that . Consequently, and . But this is only possible if .

Let us now consider an arbitrary system and let be an equivalent standard basis. By corollary 24, we notice that the differential ideal does not depend on the particular choice of , and similarly for the twisted differential ideals . Consequently, the spaces , and the operator are independent of the particular choice of . We may therefore define the distinguished right inverse of by .

Putting everything together from the effective point of view, we have:

Theorem 27. There exists an algorithm which, given and , computes the asymptotic expansion of .

Proof. Using the algorithm from section 5.2, we start by computing an equivalent standard basis for and make the corresponding change for . We next test whether is compatible with using corollary 23. If so, and assuming that , we determine the dominant term of and compute the dominant term of using the method from section 4.2. Setting and continuing the same procedure with instead of , we obtain the asymptotic expansion of .

Remark 28. The theorem still works if we take , where .

Remark 29. Using the technique of Cartesian representations [vdH97, vdH06], it is possible to compute the full expansion of and not merely the first terms (as done by the above algorithm).

Acknowledgment.
The author would like to thank N. Ratier and both referees for their helpful comments on earlier versions of this paper [vdH04].

Bibliography

[AC01]

F. Aroca and J. Cano. Formal solutions to linear pdes and convex polyhedra. JSC, 32:717–737, 2001.

[Bou94]

F. Boulier. Étude et implantation de quelques algorithmes en algèbre différentielle. PhD thesis, University of Lille I, 1994.

[Buc65]

B. Buchberger. Ein Algorithmus zum auffinden der Basiselemente des Restklassenringes nach einem null-dimensionalen Polynomideal. PhD thesis, University of Innsbruck, 1965.

[Buc85]

B. Buchberger. Multidimensional Systems Theory, chapter Gröbner bases: an algorithmic method in polynomial ideal theory, pages 184–232. Reidel, 1985. Chapter 6.

[Cas84]

F. Castro. Théorème de division pour les opérateurs différientiels et calcul des multiplicités. PhD thesis, Université Paris-VII, 1984.

[Cas87]

F. Castro. Calculs effectifs pour les idéaux d'opérateurs différentiels. In Travaux en cours, volume 24, pages 1–19. Hermann, Paris, 1987.

[DDT82]

J. Della Dora, C. Discrescenzo, and E. Tournier. An algorithm to obtain formal solutions of a linear homogeneous differential equation at an irregular singular point. In Eurocal '82, volume 174 of Lect. Notes in Comp. Science, pages 273–280, Marseille (France), 1982. Springer Verlag.

[Fab85]

E. Fabry. Sur les intégrales des équations différentielles linéaires à coefficients rationnels. PhD thesis, Paris, 1885.

[Gal85]

A. Galligo. Some algorithmic questions on ideals of differential operators. In Proc. EUROCAL '85, volume 204 of Lecture Notes in Computer Science, pages 413–421. Springer-Verlag, 1985.

[GOT05]

M. Granger, T. Oaku, and N. Takayama. Tangent cone algorithm for homogenenized differential operators. JSC, 39:417–431, 2005.

[Hir64]

H. Hironaka. Resolution of singularities of an algebraic variety over a field of characteristic zero. Annals of Math., 79:109–326, 1964.

[Mal79]

B. Malgrange. Sur la réduction formelle d'équations différentielles à singularités irrégulières. Manuscript, 1979.

[Mor83]

T. Mora. An algorithm to compute the equations of tangent cones. In Proc. EUROCAL '83, number 144 in Lect. Notes in Computer Sc., pages 158–165, 1983.

[MPT92]

T. Mora, G. Pfister, and C. Traverso. An introduction to the tangent cone algorithm. In Advances in Comp. Research, volume 6, pages 199–270. JAI Press, 1992.

[Nil65]

N. Nilsson. Some growth and ramification properties of certain integrals on algebraic manifolds. Arkiv für Mathematik, 5:463–476, 1965.

[Rit50]

J.F. Ritt. Differential algebra. Amer. Math. Soc., New York, 1950.

[Rob85]

Lorenzo Robbiano. Term orderings on the polynominal ring. In European Conference on Computer Algebra (2), pages 513–517, 1985.

[Ros59]

A. Rosenfeld. Specializations in differential algebra. Trans. Amer. Math. Soc., 90:394–407, 1959.

[SST00]

M. Saito, B. Sturmfels, and N. Takayama. Gröbner deformations of hypergeometric differential equations, volume 6 of Algorithms and Computation in Mathematics. Springer-Verlag, Berlin, Heidelberg, New York, 2000.

[Tak91]

N. Takayama. Kan: a system for computation in algebraic analysis. http://www.math.kobe-u.ac.jp/KAN/, 1991.

[vdH97]

J. van der Hoeven. Automatic asymptotics. PhD thesis, École polytechnique, France, 1997.

[vdH01]

J. van der Hoeven. Complex transseries solutions to algebraic differential equations. Technical Report 2001-34, Univ. d'Orsay, 2001.

[vdH04]

J. van der Hoeven. Generalized power series solutions to linear partial differential equations. Technical Report 2004-51, Université Paris-Sud, Orsay, France, 2004.

[vdH06]

J. van der Hoeven. Transseries and real differential algebra, volume 1888 of Lecture Notes in Mathematics. Springer-Verlag, 2006.

[vH97]

M. van Hoeij. Formal solutions and factorization of differential operators with power series coefficients. JSC, 24:1–30, 1997.