February 8, 2008 1 0 A differential analog of a theorem of Chevalley 0 2 Victor G. Kac∗† n a J 5 2 ] G Abstract A InthisnoteaproofofadifferentialanalogofChevalley’stheorem[C]onhomomorphism h. extensions is given. An immediate corollary is a condition of finitenes of extensions of dif- t ferential algebras and severalequivalent definitions of a differentially closed field, including a m Kolchin’s Nullstellensatz. [ In this note I give a proof of the following differential analog of Chevalley’s theorem [C] on 1 homomorphism extensions. v 0 Theorem 1. Let S be a differential algebra over Q with no zero divisors and let b be a non- 1 zero element of S. Let R be a differential subalgebra of S over which S is differentially finitely 2 1 generated. Let F be a differentially closed field of characteristic 0. Then there exists a non-zero 0 element a of R such that any homomorphism ϕ : R → F which does not annihilate a extends 1 0 to a homomorphism ψ :S → F which does not annihilate b. / h An almost immediate consequence of the proof of Theorem 1 is t a m Theorem 2. Let F be a differentially closed field of characteristic 0 and let S ⊃ R be differ- : entially finitely generated differential algebra and subalgebra over F. Suppose that there exists v i a non-zero element b of S such that any homomorphism ϕ : R → F has only finitely many X extensions ψ : S → F satisfying ψ(b) 6= 0. Then the field extension Fract S ⊃ Fract R is finite. r a In particular, if any homomorphism ϕ : R → F has at most d extensions ψ : S → F with ψ(b) 6= 0, then the degree of Fract S over Fract R is at most d. An immediate corollary of Theorem 1 is Kolchin’s Nullstellensatz [K] and its earlier weaker versions by Ritt [R], Cohn [Cohn] and Seidenberg [S]. AsfarasIcanunderstandit,Theorem1iscloselyrelatedtoBlum’seliminationofquantifiers theorem [Blum], [M] in the model theory of differentially closed fields. In Section 1 I explain the necessary background on Differential Algebra, in Sections 2 and 3 give proofs of Theorems 1 and 2 and in Section 4 give about a dozen of equivalent definitions of ∗Department of Mathematics, M.I.T., Cambridge, MA 02139, [email protected] †Supported in part by NSFgrant DMS-9970007. 1 a differentially closed field of characteristic 0 (Theorem 1 and Kolchin’s Nullstellensatz being among them). This note is an offshoot of a course in Differential Algebra that I gave at M.I.T. in the fall of 2000. The motivation for teaching this course came from close connections of the theory of conformal algebras todifferential algebras [BDK]. Namely, each Lie conformalalgebra defines a functor from thecategory of differential algebras to thecategory of Lie algebras (very much like a group scheme defines a functor from the category of algebras to the category of groups). I am gratefulto my studentsfortheir enthusiasm, especially to A.DeSolefor several corrections and improvements. I would like to thank A. Buium for very useful and enlightening correspondence and J. Young for bringing Marker’s paper [M] to my attention and explaining some parts of it. 1. Here I recall some terminology and facts from Differential Algebra. All of this can be easily found in two excellent books [K] and [B], the primary source of both being Ritt’s foundational book [R]. By an algebra we always mean a commutative associative unital algebra. A differential algebra R is analgebra over afieldwith afixedderivation δ. By ahomomorphismof differential algebras we always mean a homomorphism commuting with derivations. Likewise subalgebras and ideals are assumed to be δ-invariant, though to emphasize this we shall often call them differential subalgebras and ideals. The first important observation of Ritt is that in a differential algebra over Q the radical of a differential ideal is a differential ideal (see [R], [Ra], [K] Lemma 1.8). (A counterexample in characteristic p is the zero ideal in R = F[x]/(xp), δ = d/dx.) Another important fact is the differential Krull theorem: any differential radical ideal is an intersection of differential prime ideals (see [Ra], [K] Theorem 2.1). These facts imply, in particular, that maximal (among) differential ideals are prime. If a differential algebra R has no zero divisors, we can form the field of fractions Fract R and extend the derivation δ to this field. One says that a differential algebra S is differentially generated by elements x ,... ,x over 1 n a differential subalgebra R, and writes S = R{x ,... ,x }, if the algebra S is generated by all 1 n elementsfromRandallderivativesx(k),i = 1,... ,n,k ∈ Z = {0,1,2,...},oftheelementsx . i + i LetS = R{x}beadifferentialalgebrawithnozerodivisors. Onesaysthatxisdifferentially transcendental over R if all elements x(k), k ∈ Z , are algebraically independent over Fract R; + otherwise x is called differentially algebraic over R. The algebra of differential polynomials over a differential algebra R in the differential inde- terminatesy ,... ,y isthealgebraofpolynomialsR[y(k),i = 1,... ,n, k ∈ Z ]withthederiva- 1 n i + tion δ extended from R by the rule δ(y(k))= y(k+1). This algebra is denoted by R{y ,... ,y }. i 1 n Consider the differential algebra R{y} of differential polynomials over R in one differential indeterminate y. Let A(y) ∈R{y}\R, be a “non-constant” differential polynomial. The largest r for which y(r) is present in A(y) is called the order of A(y) and is denoted by ord A. One can write in a unique way: A(y) = I (y)y(r)d+I (y)y(r)d−1 +···+I (y), A 1 d where ord I (y) < r, ord I (y) < r and I (y) 6= 0. Then d is called the degree of A(y) and is A j A denotedbydegA(y),andI (y)iscalledtheinitial ofA(y). ThedifferentialpolynomialS (y)= A A 2 ∂A(y)/∂y(r) is called the separant of A(y). The important property of the characteristic 0 case is that S (y) 6= 0 if A∈/ R. A For A,B ∈ R{y}\R we write A < B if either ord A < ord B, or ord A = ord B and degA< degB. We also write A< B if A ∈R, B ∈/ R. Note that I < A and S < A. A A Thebasicresultof Ritt’s theory [R](see [K]Lemma7.3or[B](2.3)) is thefollowing division algorithm: Given a non-constant differential polynomial A(y), for any F(y) ∈ R{y} there exists G(y) ∈ R{y} with G < A and m,n ∈ Z such that: + I (y)mS (y)nF(y) ≡ G(y) mod [A(y)], A A where [A(y)] is the ideal of R{y} generated by A(y)(k), k ∈ Z . Furthermore, one can take + m = 0 if only ord G ≤ ord A is required, which is called the weak division algorithm. The important notion of a differentially closed field was introduced by Robinson [Rob]. His axioms have been considerably simplified by Blum [Blum], and it is her definition, given below, that is commonly used. A differential field F is called differentially closed if for any two differential polynomials A(y),B(y) ∈ F{y} with B 6= 0 and ord B < ord A there exists α ∈ F such that A(α) = 0 and B(α) 6= 0. Anydifferentiallyclosedfieldisalgebraicallyclosed(i.e.,hasnonon-trivialfiniteextensions), but it always has differentially algebraic extensions. Nevertheless, it turned out to be the right substitute for Differential Algebra of the notion of an algebraically closed field. The existence of a differentially closed field containing a given differential field of charac- teristic 0 is easy to establish in the framework of Model Theory (see e.g. [M]). An elementary proof (i.e., without reference to model theory) may be found in [B] (5.2). 2. Proof of Theorem 1. The general scheme of the proof is the same as Chevalley’s [C]. We have: (1) S = R{x ,... ,x }{x } ⊃ R{x ,... ,x } ⊃R. 1 n−1 n 1 n−1 By induction on n we reduce the proof to the case n = 1. Indeed, from being true for n = 1, we conclude that there exists a non-zero b ∈ R{x ,... ,x } such that any homomorphism 1 1 n−1 ψ : R{x ,... ,x } → F with ψ (b ) 6= 0 extends to ψ : R{x ,... ,x } → F with ψ(b) 6= 0, 1 1 n−1 1 1 1 n and by the inductive assumption, there exists a non-zero a ∈ R such that any homomorphism ϕ :R → F with ϕ(a) 6= 0 extends to ψ :R{x ,... ,x } → F with ψ (b ) 6= 0. 1 1 n−1 1 1 Thus, we may assume that S = R{x}. Given a homomorphism ϕ : R → F, we may extend it to the algebras of differential polynomials A(y) 7→ Aϕ(y) by applying ϕ to coefficients. Let B(y)∈ R{y} be such that B(x)= b. We consider separately two cases. Case 1: x is differentially transcendental over R. Let a ∈ R be any non-zero coefficient of B(y). If ϕ(a) 6= 0, then Bϕ(y) ∈ F{y} is a non-zero differential polynomial, hence there exists α ∈ F which is not a root of Bϕ(y) (we take B = Bϕ(y) and A with ord A > ord B in the definition of a differentially closed field). Then ψ(Q(x)) = Qϕ(α) is a well defined homomorphism S → F with ψ(b) 6= 0. Case 2: x is differentially algebraic over R. Let A(y) ∈ R{y} be a minimal in the partial ordering<irreducible(intheusualsense)overFract RdifferentialpolynomialsuchthatA(x) = 0. If F(y) ∈ R{y} is such that F(x) = 0, apply the division algorithm: S (y)mI (y)nF(y) ≡ G(y) mod [A(y)], A A 3 where G < A. Since F(x) = 0 and A(x) = A′(x) = ··· = 0, we see that G(x) = 0, hence, due to minimality of A, G(y) = 0, and we have (2) S (y)mI (y)nF(y) ∈ [A(y)]. A A Let a ∈ R be a non-zero coefficient of I (y). Let D(y) ∈ R{y} be the discriminant of A(y) 1 A viewed as a polynomial in y(ordA). Note that D(y) 6= 0 since A(y) is an irreducible polynomial, and that ord D(y)< ord A(y). Let a ∈ R be a non-zero coefficient of D(y). 2 Suppose that ϕ(a a ) 6= 0. Then ord Aϕ(y) = ord A(y) > ord Iϕ(y)Dϕ(y) and 1 2 A Iϕ(y)Dϕ(y) 6= 0. Since F is differentially closed, there exists α ∈ F which is a root of Aϕ(y) A but not a root of Iϕ(y)Dϕ(y). Since Dϕ(α) 6= 0, Aϕ(y) and Sϕ(y) have no common roots, A hence Sϕ(α) 6= 0. Since also Iϕ(α) 6= 0, but Aϕ(α) = 0, we conclude from (2) that Fϕ(α) = 0. A A A Therefore ψ(Q(x)) = Qϕ(α) is a well defined homomorphism S → F which extends ϕ :R → F. It remains to take care of the condition ψ(b) 6= 0 by an appropriate choice of α (satisfying the above conditions as well). By the weak division algorithm we have: (3) Sn(y)B(y) = B (y) mod [A(y)], A 1 where n ∈ Z and ord B (y)≤ ord A(y). Letting y = x in (3) we get + 1 (4) B (x) = Sn(x)b. 1 A ϕ Since S (α) 6= 0, we conclude that S (x) 6= 0, hence B (x) 6= 0, due to (3). Let r(y) ∈ R{y} A A 1 be the resultant of the polynomials B (y) and A(y) viewed as polynomials in y(ordA). Since 1 A(x) = 0 and B (x) 6= 0, we conclude that r(y) 6= 0 (otherwise A(y) and B (y) would have a 1 1 commonrootinFract RandthereforeB (y)wouldbedivisiblebyA(y)duetoitsirreducibility). 1 Note that ord r(y)< ord A(y). Let a ∈ R be a non-zero coefficient of r(y). 3 We let a = a a a and suppose that ϕ(a) 6= 0. Choose α ∈ F such that, as before, 1 2 3 Aϕ(α) 6= 0, Iϕ(α)Dϕ(α) 6= 0, and, in addition, rϕ(α) 6= 0. As before, define ψ(Q(x)) = Qϕ(α). A ϕ ϕ As before, this is a well defined homomorphism S → F, and, by (4): S (α)ψ(b) = B (α). But A 1 Bϕ(α) 6= 0, since rϕ(α) 6= 0 and therefore Aϕ(y) and Bϕ(y) have no common roots. Hence 1 1 ψ(b) 6= 0. Corollary 1. If S is a differentially finitely generated differential algebra over F, then for any non-zero b ∈ S there exists an F-algebra homomorphism ψ : S → F such that ψ(b) 6= 0. 3. In order to prove Theorem 2, we need the following lemma. Lemma 1. Let F be a differentially closed filed. Then (a) For any non-zero A(y) ∈ F{y} there exists infinitely many α∈ F such that A(α) 6= 0. (b) IfA(y) ∈ F{y}\F andthere existsB(y)∈ F{y}suchthatB(y) 6= 0, andB(y)< ord A(y) and only finitely many roots of A(y) are not roots of B(y), then ord A(y) = 0. Proof. (a) Take a sequence A (y) ∈ F{y}, j ∈ Z , of increasing order, such that A (y) = j + 0 A(y). For each j = 1,2,... there exists α ∈ F which is a root of A , but not of A ...A . j j 0 j−1 Hence all α ,α ,... are not roots of A(y). 1 2 4 (b) Let α ,... ,α ∈ F be all roots of A(y) which are not roots of B(y). Note that n ≥ 1 1 n n since F is differentially closed. Let B1(y) = Y(y−αi)B(y) and suppose that ord A(y) > 0. i=1 Then ord B (y) < ord A(y), hence there exists β ∈ F which is a root of A(y), butnot of B (y). 1 1 But then β 6= α for all i, a contradiction. (cid:3) i Proof of Theorem 2. Again, using(1), we reducethe proofto thecase S = R{x}. Certainly any homomorphism R{x ,... ,x } → F extends in only finitely many ways to S → F 1 n−1 that does not annihilate b. Hence from n = 1 case we conclude that Fract S is finite over Fract R{x ,... ,x }. By Theorem 1, there exists b′ 6= 0 in R{x ,... ,x } such that any 1 n−1 1 n−1 homomorphism ϕ to F with ϕ(b′) 6= 0 extends to ψ : S → F with ψ(b) 6= 0. Hence there exists only finitely many homomorphisms ψ : R{x ,... ,x } → F which extend ϕ : R → F 1 n−1 such that ψ(b′) 6= 0. Hence, by the inductive assumption, Fract R{x ,... ,x } is finite over 1 n−1 Fract R and Fract S is finite over Fract R. Since R is differentially finitely generated over F, by Corollary 1, for any non-zero a∈ R we can find a homomorphism ϕ : R → F with ϕ(a) 6= 0. Again we consider separately two cases of S = R{x}, keeping notations of the proof of Theorem 1. Case 1: x is differentially transcendental over R. Let a ∈ R be a non-zero coefficient of B(y), so that Bϕ(y) 6= 0. By Lemma 1a, Bϕ(y) has infinitely many non-roots α, and for each of them the homomorphism ψ(B(x)) = Bϕ(α) extends ϕ such that ψ(b) 6= 0. Thus, this case is impossible. Case 2: x is differentially algebraic over R. Take a= a a a from the proof of Theorem 1 1 2 3 and ϕ : R → F with ϕ(a) 6= 0. From the proof of Theorem 1 we know that ϕ extends to ψ :S → F with ψ(b) 6= 0 by letting ψ(Q(x)) = Qϕ(α), where α is a root of Aϕ(y) and is not a root of Iϕ(y)Dϕ(y)rϕ(y). Since, by conditions of the theorem, there exists only finitely many A such α, we conclude by Lemma 1, that ord Aϕ(y) = 0 = ord A(y). In other words, A(y) is an ordinary (non-zero) polynomial over R such that A(x) = 0. Hence Fract S is finite over Fract R. (cid:3) 4. Theorem 3. The following properties of a differential filed F of characteristic 0 are equivalent: (a) F is differentially closed. (b) If A,B ∈ F{y} are such that A is irreducible, B is not divisible by A and ord A≥ ord B, then there exists α ∈ F such that A(α) = 0, B(α) 6= 0 and S (α) 6= 0. A (c) If J is a prime differential ideal of F{y} and B ∈F{y}\J, then there exists α ∈ F such that f(α) = 0 for all f ∈ J and B(α) 6= 0. (d) The same as (c), where J is a radical differential ideal. (e) The same as (c) (resp. d) where F{y} is replaced by F{y ,... ,y }, n ≥ 1. 1 n (f) If J is a proper prime differential ideal of F{y ,... ,y }, n ≥ 1 arbitrary, then there 1 n exists α ∈ F such that f(α)= 0 for all f ∈ J. 5 (g) The same as (f), where J is a proper differential ideal (resp. radical differential ideal). (h) The same as (f), where J is a proper maximal differential ideal. (i) Given any S, R and b as in Theorem 1, the conclusion of Theorem 1 holds for F. Proof. (a) implies (i) by Theorem 1. (i) implies (c) by taking S = F{y}/J, b = image of B in S and applying Corollary 1 (and the differential Krull theorem). (e) trivially implies (c) and (d). A standard argument of adding extra indeterminates shows that (f) is equivalent to (e). By the differential Krull theorem, (g) is equivalent to (f). Since every prime differential ideal can beincluded in a maximal one, which is primetoo, (h)is equivalent to (f). (b) trivially implies(a)(bytakingin(a)anirreduciblefactorofAofthesameorder). Finally,theimplication (c) ⇒ (b) is proved in the same way as 3) ⇒ 1) in Theorem 5.1 of [B]. Indeed, {A} : S is a A prime ideal by Ritt’s structure theorem [B] Theorem 2.5 or [K] Lemma 7.9, and this ideal does not contain S B by Ritt’s divisibility theorem [B] Theorem 2.4 or [K] Lemma 7.8. (cid:3) A Remark. All proofs can be extended without difficulty to the case of several commuting derivations δ ,... ,δ . 1 m References [BDK] B. Bakalov, A. D’Andrea and V.G. Kac, Theory of finite pseudoalgebras, preprint ESI 916, 2000. math.QA/0007121. [Blum] L. Blum, Generalized algebraic structures: a model theoretical approach, Ph.D., M.I.T., Cambridge, MA, 1968. [B] A. Buium, Differential algebra and Diophantine geometry, Chapter 2, Hermann, Paris, 1994. [C] C. Chevalley, An algebraic proof of a property of Lie groups, Amer. J. Math. 63 (1941), 785-793. [Cohn] R. Cohn, On the analogue for differential equations of the Hilbert–Netto theorem, Bull. AMS 47 (1941), 268–270. [Kap] I. Kaplansky, An introduction to differential algebra, Hermann, Paris, 1957. [K] E.R. Kolchin, Constrained extensions of differential fields, Adv. Math. 12 (1974), 141–170. [M] D.Marker, Modeltheory of differential fields,inLecture Notes in Logic 5, Springer- Verlag, 1996, pp. 38–113. [Ra] H.W.Raudenbush,Jr.,Idealtheoryandalgebraicdifferentialequations,Tans. AMS 36 (1934), 361–368. [R] J.F. Ritt, Differential algebra, Am. Math. Soc. Coll. Publ. 33, New York, 1950. 6 [Rob] A. Robinson, On the concept of differentially closed fields, Bull. Rec. Counc. Isr. Sect. F 8 (1959), 113–118. [S] A. Seidenberg, An elimination theory for differential algebra, Univ. of Calif. Publ in Math 3 (1956), 31–56. 7