ScienceDirect
J. Differential Equations 260 (2016) 1993–2025
www.elsevier.com/locate/jde
On a classification of polynomial differential operators with respect to the type of first integrals
Jinzhi Lei
ZhouPei-YuanCenterforAppliedMathematics,TsinghuaUniversity,Beijing,100084,PRChina
Received 4November2010;revised 21September2012 Availableonline 21October2015
Abstract
ThispapergivesaclassificationofpolynomialdifferentialoperatorsX =X1(x1,x2)δ1+X2(x1,x2)δ2 (δi=∂/∂xi).TheclassificationisdefinedthroughanorderderivedfromX.LetX=Xybe theassociated differentialpolynomial,theorderisdefinedastheorderofadifferentialidealthatisanessentialextension of{X}.Themainresultshowstheordercanonlybefourpossiblevalues:0,1,2,3,or∞.Furthermore,when theorderisfinite,theessentialextension= {X,A},whereAisadifferentialpolynomialwithcoefficients obtainedthrougharationalsolutionofapartialdifferentialequationgivenexplicitlybycoefficientsofX. Whentheorder isinfinite,the extensionisidenticalwith{X}. Inaddition,if,andonlyif,theorder is0,1,or2,theassociatedpolynomialdifferentialequationhasLiouvillianfirstintegrals.Examplesand connectionswithGodbillon–Veysequencesarealsodiscussed.
©2015ElsevierInc.All rights reserved.
MSC:34A05;34A34;12H05
Keywords:Polynomialdifferentialoperator;Classification;Polynomialdifferentialequation;Differentialalgebra;
Liouvillianfirstintegral;Godbillon–Veysequence
E-mailaddress:jzlei@tsinghua.edu.cn.
http://dx.doi.org/10.1016/j.jde.2015.09.050 0022-0396/©2015ElsevierInc.All rights reserved.
1. Introduction 1.1. Background
This paper studies polynomial differential operators of form
X =X1(x1, x2)δ1+X2(x1, x2)δ2, (1) where δi =∂/∂xi (i=1, 2), and X1(x1, x2), X2(x1, x2)are polynomials of x1and x2. We fur- ther assume that X1≡0 without loss of generality.1 In this paper, we give a classification for all operators of form (1), according to which the solution of the first order partial differential equation
Xω=0 (2)
is discussed.
The operator (1)closely relates to the following polynomial differential equation dx1
dt =X1(x1, x2), dx2
dt =X2(x1, x2), (3) and any non-constant solutions of (2)give a first integral of (3). Therefore, our results also yield a classification of the polynomial systems (3)with respect to the function type of its first integral.
The current study was motivated by investigating integrating methods of a polynomial differ- ential equation of form (3). First, we take a look at a simple situation of how we can integrate(3).
If the equation (3)has an integrating factor ηthat is a rational function, a first integral ωis then obtained by an integral of a rational function, and satisfies
δ1ω−a=0, (4)
with a=ηX1a rational function. Therefore, there is a non-constant function ωthat satisfies both equations (2)and (4). In this case, the differential operator DAdefined by
DAω=δ1ω−a (5) is compatible with X. In other words, let two differential polynomials Xand Abe defined as
X=Xy, A=DAy,
then the differential ideal {X, A}is a nontrivial extension of the ideal {X} (refer to detailed definitions below). On the other hand, for any such differential polynomial A, we can always solve δ1ωand δ2ωfrom the equation {X =0, DAω=0}and therefore lead to the integration of the equation(3). This simple situation suggests that to integrate the equation (3) for first integrals, we need to find a differential polynomial Aso that {X, A}is a nontrivial extension
1 IfX1≡0,onecanexchangex1andx2(andalsothederivationsδ1andδ2)indiscussionsbelow.Ourclassification isanintrinsicpropertyoftheoperatorX,andisunchangedwhenweexchangex1andx2,andevenwhenweapplya lineartransformationin(x1,x2).SeeourdiscussionfollowingTheorem 24.
of the ideal{X}. Such a differential polynomial A, if exists, is not unique. However, the lowest order among these differential polynomials is uniquely determined by the differential operator X (called the order of X, to be detailed below), and therefore provides an index for a classification.
The classification presented in this study is given through an order of the operator X de- fined below. This order is essential for understanding integrating methods of the polynomial differential equation (3)in different classes, and also the classification of un-integrable systems.
Furthermore, for a given equation (3), the differential polynomial Ain defining the nontrivial extension {X, A}provides additional informations for the first integral, which are important for further investigations of the structure of integrating curves (or foliations) of the polynomial dif- ferential equation.
1.2. Preliminary definitions
Before stating the main results, we give some preliminary concepts from differential algebra.
For details, refer to[1]and [2].
Let K be the field of all rational functions of (x1, x2) with complex number coefficients, δ1, δ2 are two derivationsof K, and are commutable to each other. Then K together with the two derivations form a differential field, with Cas the constant field. For a differential indetermi- natey, there is a usual way to add yto the differential field K, by adding an infinite sequence of symbols
y, δ1y, δ2y, δ1δ2y,· · ·, δ1i1δi22y,· · · (6) to K[1]. This procedure results in a differential ring, denoted as K{y}. Each element in K{y}is a polynomial of finite numbers of the symbols in (6), and therefore is a differential polynomial in y with coefficients in K.
We say an algebra ideal in K{y}to be a differential idealif a∈implies δia∈ (i= 1, 2). Let be any set of differential polynomials. The intersection of all differential ideals containing is called the differential ideal generated by , and is denoted as {}. A differential polynomial Ais in {}if, and only if, Ais a linear combination of differential polynomials in and of derivatives, of various orders, of such differential polynomials.
For the operator X given by (1), we have
X=Xy=X1δ1y+X2δ2y∈K{y}, (7) and {X}denotes the differential ideal generated by X.
Definition 1. Let
w1=δ1i1δ2i2y, w2=δj11δ2j2y,
be two derivatives of y, w1islowerthan w2(or w2ishigherthan w1) if either i1< j1, or i1=j1
and i2< j2. Any element in Kis lower than the indeterminate y.
The infinite sequence (6)can be organized from low to high (here w1< w2means that w1is lower than w2) as
a < y < δ2y < δ22y <· · ·
< δ1y < δ1δ2y < δ1δ22y <· · ·
< δ12y < δ21δ2y < δ12δ22y <· · ·
<· · ·. (8)
Here ameans any element in K.
Definition 2. Let Abe a differential polynomial, if Acontains y (or its derivatives) effectively, by theleaderof A, we mean the highest of those derivatives of ywhich are involved in A.
Definition 3. Let A1, A2be two differential polynomials, we say A2to be of higherrankthanA1, if either
(1) A2has higher leader than A1; or
(2) A1and A2have the same leader, and the degree of A2(as a polynomial of the leader) in the leader exceeds that of A1.
A differential polynomial which effectively involves the intermediate y is of higher rank than one which does not. Two differential polynomials of which no difference in the rank as created above is said to be of the same rank.
The following fact is basic [2, p. 3]:
Proposition 4. Any non-empty subset of differential polynomials ⊆K{y}contains a differen- tial polynomial which is not higher than any other differential polynomials in the subset .
For a differential polynomial A ∈K{y}, we associate with Aa differential operatorDAon analytic functions A(), where is an open subset of C2, such that
DAu=A|y=u, ∀u∈A(). (9) By S(DA), we denote the singularity set of DA, which contains all singular points (i.e., the poles) in the coefficients of the differential polynomial A. Because all coefficients of Aare rational functions, the singularity set DAis a closed subset in C2. Thus, for any u ∈A(), DAuis well defined in the open subset \S(DA).
Definition 5. A differential ideal in K{y}is anextensionof {X}, or anextensionof X, if {X} ⊆. The extension is anontrivial extensionif there exists an open subset ⊂C2and a non-constant function ω∈A(), such that DAω=0 in \S(DA)for any A ∈. Otherwise, is atrivial extension.
It is obvious that {X}itself is a nontrivial extension of X, and K{y}is a trivial extension of X. Main results in this paper are to show that nontrivial extensions of X with the lowest order (to be defined below) are essential to provide the classification of X.
Proposition 6. Let be an extension of {X}, then either = {X}, or contains a differential polynomial Aof lower rank than X.
Proof. We only need to show that if {X} , then contains a differential polynomial Aof lower rank than X. Let 0=\{X}, then 0is a non-empty subset of . Proposition 4yields that there is a differential polynomial A ∈0that has the lowest rank.
We claim that Ahas lower rank than X. If otherwise, the leader of A, denoted as δi1δ2jy, is not lower than δ1y (the leader of X), i.e., i≥1. Hence, G =δ1i−1δj2Xis an element contained in {X}and has the same leader δ1iδj2y as in A, and the degree of δ1iδ2jy in Gis 1. Thus, there is a differential polynomial Q ∈K{y}so that
A1=A−QG
has lower rank than A. It is obvious A1∈0, but with lower rank than A, which is contradiction to our choice of A, and the proposition is proved. 2
Let be a nontrivial extension of {X}, and A ∈with the lowest rank. Proposition 6yields that the leader of Aeither has form δ2ry (r≥0), or equals δ1y. This fact leads to the order of given below.
Definition 7. Let be a nontrivial extension of {X}, and A ∈ with the lowest rank. If the leader of Ais δ2ry (r≥0), ris called theorderof , denoted by ord(). If otherwise, = {X} and the order of is infinity (ord() = ∞).
For a nontrivial extension , a differential polynomial with the lowest rank can only take one of the following forms:
• a polynomial of y, with at least one coefficient that is non-constant (ord() =0); or
• a differential polynomial of yeffectively involves derivatives (1 ≤ord() <∞); or
• the differential polynomial X, and therefore = {X}(ord() = ∞).
Definition 8. An essential extensionof X is a nontrivial extension of X that has order not higher than any other nontrivial extensions. By theorderof X, denoted as ord(X), we mean the order of an essential extension of X.
We note that for a given differential operator X, essential extensions of X may not be unique, but all essential extensions must have the same order. Thus, the order of X is well defined. Main results presented in the next subsection show that ord(X)provides a classification of polynomial differential operators.
1.3. Main results
Here we state the main results in this paper, with proofs given in the next section.
Theorem 9. Let the polynomial differential operator X be given by (1), with coefficients X1, X2∈K, then either 0 ≤ord(X) ≤3, or ord(X) = ∞. Furthermore, when 0≤ord(X)≤3,
we can always select an essential extension of X, such that = {X, A}, with A ∈K{y}given below
(1) if ord(X) =0, then
A=y−a (a∈K\R); (10)
(2) if ord(X) =1, then
A=(δ2y)n−a (n∈N, a∈K); (11) (3) if ord(X) =2, then
A=δ22y−aδ2y (a∈K); (12) (4) if ord(X) =3, then
A=2(δ2y)(δ23y)−3(δ22y)2−a(δ2y)2 (a∈K). (13) From Theorem 9, when the order of a differential operator X is finite, an essential extension of X is given by = {X, A}, with A ∈K{y}given by (10)–(13). Discussions in [2, Chapter 2]
have proved that the system of equations
Xy=0, DAy=0 (14) has solution in some extension field of K. It is easy to see that this solution gives a first integral of the polynomial differential equation (3). The following result for the classification of (3)is straightforward from Theorem 9.
Theorem 10. Consider the polynomial differential equation (3), and let X be the corresponding differential operator given by (1); we have the following
(1) if ord(X) =0, then (3)has a first integral ω∈K;
(2) if ord(X) =1, then (3)has a first integral ω, such that (δ2ω)n∈K for some n ∈N;
(3) if ord(X) =2, then (3)has a first integral ω, such that δ22ω/δ2ω∈K; (4) if ord(X) =3, then (3)has a first integral ω, such that
2(δ2ω)(δ32ω)−3(δ22ω)2 (δ2ω)2 ∈K;
(5) if ord(X) = ∞, then any first integral of (3) does not satisfy any differential equation of form
DAy=0 with A ∈K{y}\{X}.
In 1992, Singer has proved that the cases (1)–(3) in Theorem 10(also refer toTheorem 24 below) are the only cases to have Liouvillian integrals, i.e., the first integral obtained from ra- tional functions using finite steps of exponentiation, integration, an algebraic functions [3](also refer to[4]). In the cases (4)–(5), however, there is no first integral of (3)that can be obtained from rational functions in finite steps operation given above (refer to[3]or [4]). From the proof of Lemma 15given below, when ord(X) =3, a first integral of (3)can be expressed through finite step operations from rational functions and a solution of the partial differential equation of form(35). This result provides further classification of polynomial differential equations that are not Liouvillian integrable.
In the rest of this paper, we first give the proof of Theorem 9in Section2, then show examples for each type of equations in Section3, and a discussion for connections between our results and Godbillon–Vey sequence in Section4.
2. Proof of the main result 2.1. Outline of the proof
Hereinafter, we denote δ2iy by yi (y0=y). For any essential extension of X, let A ∈ with the lowest rank. From the above definitions, if ord(X) =r(<∞), then Ais a polynomial of y0, y1, · · ·, yr, with coefficients in K. Write
A=
m
amym00y1m1· · ·yrmr, (15)
where m =(m0, m1, · · ·, mr) ∈Zr+1, and am∈K. To prove Theorem 9, we only need to deter- mine all possible non-zero coefficients in A. Let
IA= {m∈Zr+1|am=0}. (16) We only need to specify the finite set IA. The process is outlined below.
Let m =(m0, m1, · · ·, mr) ∈Zr+1; we define an operator i,j :Zr+1→Zr+1for 0 < i <
j≤rsuch that i,j(m) ∈Zr+1is given by
i,j(m)=m+ej−i−ej (17)
where
ek=( 0
↓ 0,· · ·,0,
k
↓
1,0,· · ·,0).
Fig. 1. Flow chart of the proof ofTheorem 9.
Therefore
−1
i,j(m)=m−ej−i+ej. (18)
For any m, n ∈Zr+1, we say m nif there exist 0 < i < j≤r, such that
i,j(m)=n.
The proof below is done by showing that if r=ord(X) <∞, then IAcan only be one of the following cases:
(1) r=0, and IA= {(1), (0)}; or (2) r=1, and IA= {(0, n), (0, 0)}; or (3) r=2, and IA= {(0, 0, 1), (0, 1, 0)}, with
(0,0,1)(0,1,0); or
(4) r=3, and IA= {(0, 1, 0, 1), (0, 0, 2, 0), (0, 2, 0, 0)}, with relations
(0,2,0,0) (0,1,1,0) (0,1,0,1)
(0,0,2,0)
≺
Here (0, 1, 1, 0)is an auxiliary index with a(0,1,1,0)=0.
The final proof is complete following fourteen preliminary lemmas as shown by the flow chart in Fig. 1.
2.2. Preliminary notations
Before proving Theorem 9, we introduce some notations as follows. Hereinafter, we note X1≡0, and assume X1(x10, x20) =0.
Let
[δ2,X] =δ2X −Xδ2=(δ2X1)δ1+(δ2X2)δ2, b0= −X1(δ2X2
X1
), bi=X1(δ2bi−1
X1
)= −X1(δi2+1X2
X1
), i=1,2,· · ·
For F ∈K{y}, and {X}the differential ideal generated by X=Xy, we write
F∼R (19)
if R∈K{y}such that F −R∈ {X}.
Let m, n ∈Zr+1, the degreeof nis higher than that of m, denoted by n >m, if there exists 0 ≤k≤rsuch that nk> mk and
ni=mi, i=k+1,· · ·, r.
It is easy to verify that the relation implies >, and for any m ∈Zr+1and 0 < i < j≤r,
−1
i,j(m)m i,j(m), (20)
and
−1
i,j(m) >m> i,j(m). (21) In the following discussion, by m∗ we always denote the element in IA with the highest degree, and always assume Am∗=1 without loss of generality. This is possible as the coefficients in Aare rational functions in K.
For any m ∈Zr+1, define
P(m)= {p∈IA|pm} (22) and #(m) = |P(m)|, the number of elements in P(m).
We define a function C:Zr+1→Zby
C(m)= r j=1
j mj, (23)
where m =(m0, m1, · · ·, mr) ∈Zr+1. It is easy to verify that if m p, then C(m) > C(p). In particular,
C(m)−C( i,j(m))=i (0< i < j≤r). (24)
2.3. Preliminary lemmas
Now, we start the proof process. First, Lemma 11below is straightforward from the definition of nontrivial extension.
Lemma 11. Let A ∈K{y}\{X}, the differential ideal = {A, X}is a nontrivial extension of X if, and only if, the equation
Xy = 0
DAy = 0 (25)
has a non-constant solution in A(), with an open subset of C2. The following result is a direct conclusion from Lemma 11:
Lemma 12. If there existsa∈K, non-constant, such that Xa=0, then let A=y−a,
the differential ideal = {X, A}is a nontrivial extension of X. Lemma 13. If there exists a∈K, a=0, such that
Xa=nb0a, (26) where nis non-zero integer, let
A=(δ2y)|n|−a|n|/n, (27) then = {X, A}is a nontrivial extension of X.
Proof. From Lemma 11, we only need to show that there is a non-constant solution for the differential equation
X1δ1y+X2δ2y = 0
(δ2y)|n|−a|n|/n = 0. (28) Let
u=a1/n, v= −X2 X1u, and taking account of (26), direct calculations show that
δ1u= 1 X1
(b0u−X2δ2u)=δ2v.
Thus, the 1-form vdx1+udx2is closed, and therefore the function of form
ω=
(x1,x2) (x10,x02)
vdx1+udx2
is well defined and analytic on a neighborhood of some (x10, x20) ∈C2. Further, δ1ω=v, δ2ω=u.
It is easy to verify that ωsatisfies (28), and the lemma is proved. 2 Lemma 14. If there exists a∈Kthat satisfies
Xa=b0a+b1, (29) let
A=δ22y−aδ2y, (30)
then = {X, A}is a nontrivial extension of X.
Proof. Let
b= −X2
X1
a+ b0
X1
.
From (29), we have
δ1a= −X2
X1δ2a−(δ2X2
X1)a+δ2b0 X1=δ2b.
Thus, the 1-form bdx1+adx2 is closed, and there exists a function μ that is analytic on a neighborhood of some (x10, x20) ∈C2, such that
δ1μ=b, δ2μ=a.
Let u =exp(μ), then uis a non-zero function, and
Xu=u(X1δ1μ+X2δ2μ)=u(X1b+X2a)=b0u.
Thus, following the proof of Lemma 13, let v= −X2
X1
u,
then vdx1+udx2is a closed 1-form, and the function
ω=
(x1,x2) (x10,x20)
vdx1+udx2
is well defined in a neighborhood of (x10, x20), non-constant, and satisfies X1δ1ω+X2δ2ω=0, δ2ω−u=0.
Therefore,
X1δ1ω+X2δ2ω=0, δ22ω−aδ2u=0.
Thus, the non-constant function ωsatisfies equations X1δ1y+X2δ2y=0
δ22y−aδ2y=0 (31)
and hence the lemma is concluded from Lemma 11. 2 Lemma 15. If there exists a∈Kthat satisfies
Xa=2b0a+b2, (32) let
A=2(δ2y)(δ23y)−3(δ22y)2−a(δ2y)2, (33) then = {X, A}is a nontrivial extension of X.
Proof. We only need to show that there is a function ωthat is analytic on an open subset of C2, non-constant, and satisfies
X1δ1ω+X2δ2ω=0
2(δ2ω)(δ32ω)−3(δ22ω)2−a(δ2ω)2=0. (34) We divide the proof into two steps. First, define
f (x1, x2, u)= −δ22X2
X1−X2
X1a−(δ2X2
X1)u−1 2(X2
X1)u2, g(x1, x2, u)=a+1
2u2.
Then f and gare analytic at some point (x10, x20, u0) ∈C3. Now, we prove that there is a function u(x1, x2)that is analytic on a neighborhood of (x10, x20), and u(x10, x20) =u0, such that
δ1u =f (x1, x2, u),
δ2u =g(x1, x2, u) (35)
is satisfied in a neighborhood of (x10, x20). To this end, we need to show ( ∂
∂x2+g(x1, x2, u)∂
∂u)f (x1, x2, u)=( ∂
∂x1+f (x1, x2, u)∂
∂u)g(x1, x2, u) (36) and apply Lemma 28in Appendix A.
From (32), we have
δ1a= −δ23X2
X1−δ2(X2
X1a)−(δ2
X2 X1)a.
Thus, from (35), we have ( ∂
∂x2 +g(x1, x2, u)∂
∂u)f (x1, x2, u)
= −δ23X2
X1 −δ2(X2
X1
a)−(δ22X2
X1
)u−1 2(δ2X2
X1
)u2
−g(x1, x2, u)(δ2X2 X1+X2
X1u)
= −δ23X2
X1 −δ2(X2
X1a)−(δ2
X2
X1)a−(δ22X2
X1)u−(X2 X1a)u
−(δ2
X2 X1)u2−1
2(X2 X1)u3,
= −δ23X2
X1 −δ2(X2
X1a)−(δ2X2
X1)a+uf (x1, x2, u)
=δ1a+uf (x1, x2, u)
=( ∂
∂x1+f (x1, x2, u)∂
∂u)g(x1, x2, u) and (36)is satisfied.
Now, assuming
u(x1, x2)=∞
i=0
∞ j=0
ui,j(x1−x10)i(x2−x20)j (u0,0=u0) (37)
and applying the Method of Majorants, we can obtain the coefficients ui,j by induction, and the power series (37)is convergent in a neighborhood of (x10, x20)(refer toLemma 28in Appendix A for detail). Thus, the function (37)gives an analytic solution of (35).
Next, we construct a solution ωof (34)from the above solution uof (35). Let v= −δ2
X2 X1 −X2
X1u.
It is easy to verify δ2v=δ1u, and hence the 1-form vdx1+udx2is closed. Let
μ=exp
⎡
⎢⎢
⎣
(x1,x2) (x10,x20)
vdx1+udx2
⎤
⎥⎥
⎦, (38)
then the function μis well defined, non-zero, and analytic on a neighborhood of (x10, x20)(here we note that X1(x10, x20) =0), and
Xμ=b0μ.
Following the proof of Lemma 13, there exists a non-constant function ω, analytic on a neigh- borhood of (x10, x20)(here we note that b0is analytic at (x10, x20)), such that
Xω=0, δ2ω=μ.
From (38)and (35), we have δ2μ =μu, and μδ22μ=μ
(δ2μ)u+μ(a+1 2u2)
=3
2(δ2μ)2+aμ2. Taking account ofμ =δ2ω, we have
(δ2ω)(δ23ω)−3
2(δ22ω)2−a(δ2ω)2=0.
Thus, ωsatisfies (34)and the lemma is concluded. 2 Lemma 16. Let [δ2, X]and yi be defined as previously, then (1) [δ2, X]=(δ2XX1
1 )X −b0δ2; (2) Xyj=δ2Xyj−1−(δ2XX1
1 )Xyj−1+b0yj. Proof. (1) is straightforward from
[δ2,X] =(δ2X1)δ1+(δ2X2)δ2
=δ2X1
X1 (X1δ1+X2δ2)−X2
X1(δ2X1)δ2+(δ2X2) δ2
=δ2X1
X1 X +X1X1δ2X2−X2δ2X1 X12 δ2
=δ2X1
X1 X −b0δ2.
(2) can be obtained by direct calculations below:
Xyj=Xδ2yi−1
=δ2Xyj−1− [δ2,X]yj−1
=δ2Xyj−1−(δ2X1
X1 X −b0δ2)yj−1
=δ2Xyj−1−δ2X1
X1 Xyj−1+b0δ2yj−1
=δ2Xyj−1−(δ2X1 X1
)Xyj−1+b0yj. 2
Lemma 17. We have
Xyj∼
j−1
i=0
ci,jbiyj−i (j≥1) (39)
where ci,j are positive integers, and c0,j=j. Proof. From Lemma 16, when j =1, we have
Xy1=δ2Xy0−(δ2X1
X1 )Xy0+b0y1∼b0y1. Thus (39)holds for j=1 with c0,1=1.
Assume that (39)is valid for j=kwith positive integer coefficients ci,k, and c0,k=k, apply- ing Lemma 16, we have
Xyk+1=δ2Xyk−(δ2X1 X1
)Xyk+b0yk+1
∼δ2(
k−1
i=0
ci,kbiyk−i)−(δ2X1
X1
)(
k−1
i=0
ci,kbiyk−i)+b0yk+1
=
k−1
i=0
ci,k((δ2bi)yk−i+biδ2yk−i)−
k−1
i=0
ci,kδ2X1
X1 biyk−i+b0yk+1
=
k−1
i=0
ci,k
(δ2bi−δ2X1
X1 bi)yk−i+biyk−i+1
+b0yk+1
=(c0,k+1)b0yk+1+
k−2
i=0
ci,kX1δ2(bi
X1)+ci+1,kbi+1
yk−i
+ck−1,kX1δ2(bk−1 X1 )y1
=(c0,k+1)b0yk+1+
k−2
i=0
(ci,k+ci+1,k)bi+1yk−i+ck−1,kbky1.
Thus, let
⎧⎪
⎨
⎪⎩
c0,k+1=c0,k+1=k+1, ci,k+1=ci−1,k+ci,k, ck,k+1=ck−1,k,
(1≤i≤k−1),
which are positive integers, we have
Xyk+1∼ k
i=0
ci,k+1biyk+1−i.
The lemma is proved by induction. 2 Lemma 18. We have
Xamym∼(Xam+C(m)b0am)ym+
r−1
i=1
r j=i+1
mjci,jbiamy i,j(m), (40)
where am∈K, ym=y0m0y1m1· · ·ymrr, and ci,j is defined as in Lemma 17.
Proof. It is easy to have
Xamym=(Xam)ym+am r j=0
∂ym
∂yj Xyj.
From Lemma 17, we have
Xamym=(Xam)ym+am
r j=0
mjym−ejXyi
∼(Xam)ym+am r j=1
mjym−ej(
j−1
i=0
ci,jbiyj−i)
=(Xam)ym+amb0( r j=1
c0,jmj)ym+am r j=1
j−1
i=1
mjci,jbiym+ej−i−ej
=(Xam+C(m) b0am)ym+
r−1
i=1
r j=i+1
mjci,jbiamy i,j(m),
and the lemma is concluded. 2
Lemma 19. Let be a nontrivial extension of X and = {X}, A ∈with the lowest rank and r=ord() (<∞). Let m∗∈IA with the highest degree and assume that am∗=1, then for any m ∈Zr+1, m <m∗, we have
Xam=(C(m∗)−C(m))b0am−
r−1
i=1
r j=i+1
(mj+1)ci,jbia −1
i,j(m). (41)
Here am=0whenever m /∈IA. Proof. We can write
A=
m∈IA
amym=
m≤m∗
amym.
Hereinafter am=0 if m /∈IA. First, it is easy to have
XA=X1δ1A+X2δ2A∈.
On the other hand, from Lemma 18, we have XA=
m≤m∗
Xamym
∼
m≤m∗
⎛
⎝(Xam+C(m)b0am)ym+
r−1
i=1
r j=i+1
mjci,jbiamy i,j(m)
⎞
⎠
=
m≤m∗
⎛
⎝Xam+C(m)b0am+
r−1
i=1
r j=i+1
(mj+1)ci,jbia −1 i,j(m)
⎞
⎠ym
Note that for any j > i, −i,j1(m∗) >m∗, and thus −i,j1(m∗) /∈IA, i.e., a −1
i,j(m∗)=0 for any j > i. Taking account of am∗=1, we have Xam∗=0, and hence
XA∼C(m∗)b0ym∗
+
m<m∗
⎛
⎝Xam+C(m)b0am+
r−1
i=1
r j=i+1
(mj+1)ci,jbia −1 i,j(m)
⎞
⎠ym.
Therefore,
XA−C(m∗)b0A∼R=
m<m∗
fmym, (42)
where the coefficients fmare
fm=Xam+(C(m)−C(m∗))b0am+
r−1
i=1 r−1
j=i+1
(mj+1)ci,jbia −1
i,j(m). (43)
Now, we obtain a differential polynomial Rthat has lower rank than Aand is contained in the differential ideal . But Ais an element in with the lowest rank. Thus, we must have R≡0.
Therefore the coefficients (43)are zero, from which (41)is concluded. 2 Note that am∗=1 and −i,j1(m∗) /∈IA, the equation (41)is also valid for am∗. The equation (41)can be rewritten in another form below.
Lemma 20. In Lemma 19, for any m ≤m∗, let k=#(m)and P(m) = {p1, · · ·, pk}, and assume
il,jl(pl) =m(l=1, 2, · · ·, k), then the coefficients apl, amsatisfy Xam=(C(m∗)−C(m))b0am−
#(m)
l=1
(mjl +1)cil,jlbilapl. (44)
Lemma 21. Let be a nontrivial extension of X and = {X}, A ∈with the lowest rank and r=ord() >1. Let m∗∈IA with the highest degree. Then for any m ∈IA, #(m) =0 if and only if C(m) =C(m∗). Furthermore, if #(m) =0, then amis a constant.
Proof. First, we prove that if #(m) =0, then C(m) =C(m∗).
If #(m) =0, Lemma 20yields
Xam=(C(m∗)−C(m))b0am.
If otherwise C(m) =C(m∗), then n =C(m∗) −C(m)is a non-zero integer, and am=0 such that
Xam=nb0am. From Lemma 13, let
A=(δ2y)|n|−am|n|/n,
then the differential ideal = {X, A}is a nontrivial extension of X and with order ≤1. This contradicts with the assumption that is an essential extension with order >1. Thus, we have concluded that C(m) =C(m∗).
Next, we prove that if C(m) =C(m∗), then #(m) =0.
If on the contrary, C(m) =C(m∗)but #(m) >0, there exists m1∈P(m). From (24), we have C(m1) > C(m) =C(m∗). Applying the previous part of the proof to m1, we have #(m1) >0.
Thus, we can repeat the above process, and obtain m2∈P(m1)such that C(m2) > C(m1) >
C(m∗)and #(m2) >0. This procedure can be continued to obtain an infinite sequence {mk}∞k=1⊆ IAsuch that #(mk) >0 and C(mk+1) > C(mk) > C(m∗). But IAis a finite set. Thus, we come to a contradiction, and therefore #(m) =0.