ZETA FUNCTIONS AND BLOW-NASH EQUIVALENCE GOULWEN FICHOU Abstract. We propose a refinement of the notion of blow-Nash equivalence between Nash function germs, which has been introduced in [2] as an analog in theNash setting of the blow-analytic equivalence defined by T.-C. Kuo [13]. The new definition is more natural and geometric. Moreover, this equivalence relation still does not admit moduli 5 foraNashfamilyofisolatedsingularities. Butifthezetafunctionsconstructedin[2]are 0 no longer invariants for thisnew relation, however, thanks to a Denef & Loeser formula 0 comingfrommotivicintegrationinaNashsetting,wemanagedtoderivenewinvariants 2 for this equivalence relation. n a J The classification of real analytic function germs is a difficult topic, especially in the 7 choice of a good equivalence relation between germs to study. Even in the particular case 1 when the analytic function germs are Nash, that is they are moreover semi-algebraic, the ] difficulty still remains. G In [2], we have defined the blow-Nash equivalence between Nash function germs, as an A approximation with algebraic data of the blow-analytic equivalence defined by T.-C. Kuo . h in [13]. This blow-analytic equivalence has already been studied with slightly different t definitions since the original definition of T.-C. Kuo appeared (see in particular S. Koike a m & A. Parusin´ski [11] and T. Fukui & L. Paunescu [7]). Nevertheless, roughly speaking, it [ states that two given real analytic function germs are equivalent if they are topologicaly equivalentandmoreover, aftersuitablemodifications,theybecomeanalytically equivalent. 1 For the case of Nash function germs, the definition of blow-Nash equivalence runs as v 2 follows. Let f,g : (Rd,0) −→ (R,0) be Nash function germs. They are said to be blow- 5 Nash equivalent if there exist two Nash modifications (we refer to definition 1.1 for the 2 precise definition) 1 0 π : M ,π−1(0) −→ (Rd,0) and π : M ,π−1(0) −→ (Rd,0), 5 f (cid:0) f f (cid:1) g (cid:0) g g (cid:1) 0 h/ and a Nash isomorphism φ : (cid:0)Mf,πf−1(0)(cid:1) −→ (cid:0)Mg,πg−1(0)(cid:1), that is φ is a real ana- lytic isomorphism with semi-algebraic graph, which induces a homeomorphism h between t a neighbourhoods of 0 in Rd such that f = g◦h. m Forastrongernotion ofblow-Nash equivalence, weknowncomputableinvariants, which : v seems to be efficient tools to distinguish blow-Nash type [2, 3]. These invariants, called i zeta functions (cf. section 2.2), are constructed in a similar way to the motivic zeta X functions of Denef & Loeser, using the virtual Poincar´e polynomial of arc-symmetric sets r a as a generalized Euler characteristic (cf. section 2.1). Nevertheless, the definition of theblow-Nash equivalence given in [2] is strong and tech- nical. In particular the modifications are asked to be algebraic, which is not natural in theNash setting. Theweaker definition of blow-Nash equivalence introduced inthis paper is more natural and geometric. It is closer to the definition of blow-analytic equivalence considered by S. Koike and A. Parusin´ski in [11]. This blow-Nash equivalence is an equiv- alence relation (proposition 1.3). For such an equivalence relation, it is a crucial fact to prove that it has a good behaviour with respect to family of Nash function germs. In this direction, theorem 1.5 states that a family with isolated singularities does not admit moduli. This result is more general that the one in [2], whereas the present proof is just 1991 Mathematics Subject Classification. 14B05, 14P20, 14P25, 32S15. 1 a refinement of the former one. We mention also in section 1.2 various criteria to ensure the blow-Nash triviality of a given family. Recently, invariants for this kind of equivalence relations have been introduced (see [4] for a survey). In particular, we defined in [2] zeta functions, following ideas coming from motivic integration [1], which are defined via the virtual Poincar´e polynomial [15]. Unfortunately, if this definition of the blow-Nash equivalence in this paper is more naturalandgeometric, thezetafunctionsarenolongerinvariantsingeneral. However, one can derived new invariants from these zeta functions by evaluating its coefficients, which are rational functions in the indeterminacy u with coefficients in Z at convenient values (cf. theorem 3.4). As a key ingredient, we generalize the Denef & Loeser formulae, that express the zeta functions in terms of a modification, in the setting of Nash modifications (see part 2.3). As a application, we manage to distinguish the blow-Nash type of some Brieskorn polynomials whose blow-analytic type is not even known! Acknowledgements. The author wish to thank T. Fukui, S. Koike and A. Parusin´ski for valuable discussions on the subject. 1. Blow-Nash equivalence 1.1. Let us begin by stating the definition of the blow-Nash equivalence between Nash function germs that we consider in this paper. It consists of a natural adaptation of the blow-analytic equivalence defined by T.-C. Kuo ([13]) to the Nash framework. Definition 1.1. (1) Let f : (Rd,0) −→ (R,0) be a Nash function germ. A Nash modification of f is a proper surjective Nash map π : M,π−1(0) −→ (Rd,0) whose complexification (cid:0) (cid:1) π∗ is an isomorphism except on some thin subset of M∗, and such that f ◦π has only normal crossings. (2) Two given germs of Nash functions f,g : (Rd,0) −→ (R,0) are said to be blow- Nash equivalent if there exist two Nash modifications σ : M ,σ−1(0) −→ (Rd,0) and σ : M ,σ−1(0) −→ (Rd,0), f (cid:0) f f (cid:1) g (cid:0) g g (cid:1) and a Nash isomorphism Φ between semi-algebraic neighbourhoods M ,σ−1(0) (cid:0) f f (cid:1) and M ,σ−1(0) which induces a homeomorphism φ : (Rd,0) −→ (Rd,0) such g g (cid:0) (cid:1) that the diagram M ,σ−1(0) Φ // M ,σ−1(0) (cid:0) f f (cid:1) (cid:0) g g (cid:1) σf σg (cid:15)(cid:15) (cid:15)(cid:15) (Rd,0) φ // (Rd,0) MMMMfMMMMMM&& xxqqqqqqgqqqq (R,0) is commutative. Remark 1.2. (1) Let us specify some classical terminology (see [4] for example). Such a homeomor- phism φ is called a blow-Nash homeomorphism. If, as in [2], we ask moreover Φ to preserve the multiplicities of the jacobian determinant along the exceptionnal divisors of the Nash modifications σ , σ , then Φ is called a blow-Nash isomor- f g phism. 2 Nota that there exist blow-Nash homeomorphisms which are not blow-Nash isomorphisms (see [4]). (2) In[2], we considera moreparticular notion of blow-Nash equivalence. Namely, the Nashmodificationswerereplacedbyproperalgebraicbirationalmorphismsandthe blow-Nash homeomorphism was moreover asked to be a blow-Nash isomorphism. The definition 1.1 is more natural since all the data are of Nash class. The proof of the following result is the direct analog of the corresponding one in [13]. Proposition 1.3. The blow-Nash equivalence is an equivalence relation between Nash function germs. Proof. The point is the transitivity property. Let f ,f ,f : (Rd,0) −→ (R,0) be Nash 1 2 3 functiongermssuchthatf ∼ f andf ∼ f . Letσ ,σ andσ′,σ′ beNashmodifications, 1 2 2 3 1 2 2 3 and φ,φ′ be homeomorphisms like in definition 1.1 for f ,f and f ,f respectively. The 1 2 2 3 fiberproductM (respectivelyM′)ofφ◦σ andσ (respectivelyφ′◦σ′ andσ′)givessuitable 1 2 2 3 Nash modifications of(Rd,0). TakingoncemorethefiberproductM′′ ofM andM′ solves theproblemsince thecompositions of theprojections withthe initial modificationsσ and 1 σ′ remain Nash modifications for f and f . 3 1 3 uulllllllllllllllllM′′ RRRRRRRRRRRRRRRRR)) M M′ {{xxxxxxxxx BBBBBBBB }}{{{{{{{{ GGGGGGGGG## M M M′ M′ (σR1d(cid:15)(cid:15),10) φ 2FFFFFFσF2F// F(""Rd,0)||xxσx2′xxxxx 2 φ′ // (Rd(cid:15)(cid:15),3σ03′) VVVVVVVVVVVfV1VVVVVVVVVVVV++ (R,(cid:15)(cid:15) f02) sshhhhhhhhhhhhhfh3hhhhhhhhhhh (cid:3) Remark 1.4. Note that for the blow-Nash equivalence considered in [2], we had to con- sidertheequivalencerelationgenerated byasimilarrelation. Thisdifficultycamefromthe fact that the fiber productof an algebraic map and a Nash map needs not to be algebraic. The point here is that the fiber product of Nash maps still remains in the Nash class. The question of moduli is a natural and crucial issue when one studies an equivalence relation between germs. The following theorem states that the blow-Nash equivalence has a good behaviour with respect to family of Nash function germs. More precisely, the blow-Nash equivalence does not admit moduli for a Nash family of Nash function germs with an isolated singularity. Let’s P denote the cuboid [0,1]k for an integer k. Theorem 1.5. Let F : (Rd,0)×P −→ (R,0) be a Nash map and assume that F(.,p) : (Rd,0) −→ R has an isolated singularity at 0 for each p ∈P. Then the family F(.,p), for p ∈P, consists of a finite number of blow-Nash equivalence classes. Remark 1.6. The proof of theorem 1.5 can be performed in a similar way to the one in [2], even if the result is more general here. Indeed, we had to restrist the study in [2] to particular Nash families, that is falilies which admit, a resolution of the singularities, an 3 algebraic modification. But, if we allow the modifications to become Nash, the Hironaka’s resolution of singularities provides us suitable Nash modifications [9]. 1.2. Blow-Nash triviality. In view of classification problems, a worthwhile issue is to give criteria for a Nash family to consist of a unique blow-Nash class. In particular, one says that a Nash family F : (Rd,0)×P −→ (R,0) is blow-Nash trivial if there exist aNash modification σ : (M,E) −→ (Rd,0), a t-level preserving homeomorphism φ : (Rd,0)× P −→ (Rd,0)×P andat-levelpreservingNashisomorphismΦ :(M,E)×P −→ (M,E)×P such that the diagram (M,E)×P σ×id // (Rd,0)×P(x,p)7→F(x,0) // (R,0) Φ φ id (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) (M,E)×P σ×id // (Rd,0)×P(x,p)7→F(x,p) // (R,0) is commutative. Below, we mention sufficient conditions to ensure the blow-Nash triviality of a given family, that areanalogs of correspondingresults concerningblow-analytic equivalence ([7], [8]). Moreover their proof (cf. remark 1.10) is a direct consequence of the one of theorem 1.5. Let us introduce some terminology before stating the first result, which is inspired by the main theorem of [8]. For an analytic function germ f : (Rd,0) −→ (R,0), denote by c xI its Taylor expansion at the origin, where xI = xi1...xid, I = (i ,...,i ). The PI I 1 d 1 d Newton polygon of f is the convex hull of the union of the sets I+Rd, for those |I| such + that c 6= 0. For a face γ of this polyhedron, we put f (x) = c xI. The germ f is I γ PI∈γ I said to be non-degenerate, with respect to its Newton polygon, if the only singularities of f are concentrated in the coordinate hyperplanes, for any compact face γ of the Newton γ polygon. Finally, one says that a given face is a coordinate face if it is parallel to some coordinate hyperplane. Proposition 1.7. Assume that the Newton polygon of F(.,p) is independent of p ∈ P, non-degenerate for each p ∈ P, and moreover assume that F(.,p) in independent of (cid:0) (cid:1)γ p ∈ P for any non-compact and non-coordinate face γ of the Newton polygon. Then the family {F(.,p)} is blow-Nash trivial. p∈P Thesecond resultis inspiredby the main theorem in [7]. ConsidertheTaylor expansion F(x,p) = c (p)xI of F at the origin of Rd. For an d-uple of positive integers w = I I P (w ,...,w ), we set H(w)(x,p) = c xI, where |I| = i w +···+i w . Denote 1 d i PI: |I|w=i I w 1 1 d d (w) by m the smallest integer i such that H (x,p) is not identically equal to 0. i (w) Proposition 1.8. If there exists an d-uple of positive integers w such that H (x,p) has m an isolated singularity at the origin of Rd for any p ∈ P, then the family {F(.,p)} is p∈P blow-Nash trivial. Example 1.9. ([7]) Let F :(R3,0)×R −→ (R,0) be the Brianc¸on-Speder family, namely F(x,y,z,p) = z5+py6z+xy7+x15. Thisfamilyisweightedhomogenouswithweight(1,2,3)andweighteddegree15. Moreover 1517745 it defines and isolated singularity at the origin for p 6= p = − 2 . Therefore the 0 3 Brianc¸on-Speder family is blow-Nash trivial over all interval that does not contain p . 0 Remark 1.10. The proof of these triviality results in the blow-analytic case is based on the integration along an analytic vector field defined on the parameter space, and 4 that can be lifted through the modification. The flow of the lifted vector field gives the trivialisationupstairs. Moreovertheassumptionsmadeenabletochoose,asamodification, a toric modification that has an unique critical value at the origin of Rd. Therefore the trivialisation upstairs induces a trivialisation at the level of the parameter space. However, by integration along a Nash vector field, one needs not keep Nash data, and therefore the same method as in the blow-analytic case does not apply in the situation of propositions 1.7, 1.8. Nevertheless, one can replace this integration by the following argument (exposed with details in [2]). First, resolve the singularities of the family via the relevanttoricmodificationasin[7],[8]. Then,trivialisethezerolevelofthefunctiongerms with the Nash Isotopy Lemma [6]. Finally, trivialise the t-levels, t 6= 0, via well-choosen projections that can be proven to be of blow-Nash class. 2. Zeta functions In this section, we recall the definition of the naive zeta function of a Nash function germ (as it is defined in [2]). Then we prove the so-called Denef & Loeser formula for such a zeta function in terms of a Nash modification. This result is new and requires to generalize the change of variables formula in the theory of motivic integration to the Nash setting. 2.1. Virtual Poincar´e polynomial of arc-symmetric sets. Arc-symmetric sets have been introduced by K. Kurdyka [14] in 1988 in order to study “rigid components” of real algebraic varieties. The category of arc-symmetric sets contains the real algebraic varieties and, in some sense, this category has a better behaviour that the one of real algebraic varieties, maybe closer to complex algebraic varieties. For a detailed treatment of arc-symmetric sets, we refer to [2]. Nevertheless, let us precise the definition of such sets. We fix a compactification of Rn, for instance Rn ⊂ Pn. Definition 2.1. Let A ⊂ Pn be a semi-algebraic set. We say that A is arc-symmetric if, for every real analytic arc γ :]−1,1[−→ Pn such that γ(]−1,0[) ⊂ A, there exists ǫ > 0 such that γ(]0,ǫ[) ⊂ A. One can think about arc-symmetric sets as the biggest category, denoted AS, stable under boolean operations and containing the compact real algebraic varieties and their connected components. In particular, the following lemma specifies what the nonsingular arc-symmetric sets are. Note that by an isomorphism between arc-symmetric sets, we mean a birational map containing the arc-symmetric sets in the support. Moreover, a nonsingular arc-symmetric set is an arc-symmetric whose intersection with the singular locus of its Zariski closure is empty. Lemma 2.2. ([2]) Compact nonsingular arc-symmetric sets are isomorphic to unions of connected components of compact nonsingular real algebraic varieties. A Nash isomorphism between arc-symmetric sets A ,A ∈ AS is the restriction of an 1 2 analytic and semi-algebraic isomorphism between compact semi-algebraic and real an- alytic sets B ,B containing A ,A respectively. Generalized Euler characteristics for 1 2 1 2 arc-symmetric sets are the invariants, under Nash isomorphisms, which enable to give concrete measures in the theory of motivic integration. A generalized Euler characteristic is defined as follows. An additive map on AS with values in an abelian group is a map χ defined on AS such that (1) for arc-symmetric sets A and B which are Nash isomorphic, χ(A) = χ(B), (2) for a closed arc-symmetric subset B of A, χ(A) = χ(B)+χ(A\B). 5 If moreover χ takes its values in a commutative ring and satisfies χ(A×B)= χ(A)·χ(B) for arc-symmetric sets A,B, then we say that χ is a generalized Euler characteristic on AS. In [2] we proved: Proposition 2.3. There exist additive maps on AS with values in Z, denoted β and i Z called virtual Betti numbers, that coincide with the classical Betti numbers dimH (·, ) i 2Z on the connected component of compact nonsingular real algebraic varieties. Moreover β(·) = β (·)ui is a generalized Euler characteristic on AS, with values in Z[u]. Pi≥0 i Example 2.4. IfPk denotes therealprojective spaceofdimensionk, whichisnonsingular and compact, then β(Pk) = 1+u+···+uk. Now, compactify the affine line A1 in P1 R by adding one point at the infinity. By additivity β(A1) = β(P1)−β(point) = u, and so R β(Ak) = uk. R Remark 2.5. (1) The virtual Poincar´e polynomial is not a topological invariant (cf [15]). (2) The virtual Poincar´e polynomial β respects the dimension of arc-symmetric sets: for A ∈ AS, dim(A) = deg β(A) . In particular, it assures us that a nonempty (cid:0) (cid:1) arc-symmetric set has a nonzero value under the virtual Poincar´e polynomial. (3) By evaluating u at −1, one recover the classical Euler characteristic with compact supports ([2, 15]). 2.2. Zeta functions. The zeta functions of a Nash function germ are defined by taking the value, under the virtual Poincar´e polynomial, of certain sets of arcs related to the germ. Denote by L the space of formal arcs at the origin 0∈ Rd, defined by: L = L(Rd,0) = {γ : (R,0) −→ (Rd,0) :γ formal}, and by L , for an integer n, the space of arcs truncated at the order n+1: n L = {γ(t) = a t+a t2+···a tn, a ∈ Rd}. n 1 2 n i Let π : L−→ L be the truncation morphism. n n Consider a Nash function germ f : (Rd,0) −→ (R,0). We define the naive zeta function Z (u,T) of f as the following element of Z[u,u−1][[T]]: f Z (u,T) = β(X )u−ndTn, f X n n≥1 where X is composed of those arcs that, composed with f, give a series with order n: n X = {γ ∈ L :ord(f ◦γ) = n}= {γ ∈ L : f ◦γ(t) = btn+··· ,b6= 0}. n n n Similarly, we define zeta functions with signs by Z+(u,T) = β(X+)u−ndTn, Z−(u,T) = β(X−)u−ndTn f X n f X n n≥1 n≥1 where X± = {γ ∈ L : f ◦γ(t) = ±tn+···}. n n Remark that X , X±, for n ≥ 1, are constructible subsets of Rnd, hence belong to AS. n n In [2], we prove that these zeta functions are invariants for the stronger notion of blow- Nash equivalence (with blow-Nash isomorphism). Adapted to the present case, what we will prove is: 6 Proposition 2.6. Let f,g :(Rd,0) −→ (R,0) be germs of Nash functions. If f and g are blow-Nash equivalent via a blow-Nash isomorphism, then Z (u,T) = Z (u,T), Z±(u,T) = Z±(u,T). f g f g Remark 2.7. (1) We do not know whether or not the zeta functions are invariant for the blow-Nash equivalence. (2) This result is a step toward the resolution of the main issue of the paper (theorem 3.4): which informations can we preserve, at the level of zeta functions, with only a blow-Nash homeomorphism instead of a blow-Nash isomorphism. (3) Note that if the modifications appearing in the definition of the blow-Nash equiv- alence of f and g are moreover algebraic, the result is precisely the one in [2]. So whatwe have to justify hereis that Nash modifications are allowed. Thekey point is the Denef & Loeser formula (cf. next section). 2.3. Denef & Loeser formulae for a Nash modification. The key ingredient of the proof of proposition 2.6, and that will be crucial in section 3 also, is the following Denef & Loeser formulae which express the zeta functions of a Nash function germ in terms of a modification of its zero locus. First, we state the case of the naive zeta function. Proposition 2.8. (Denef & Loeser formula) Let σ : M,σ−1(0) −→ (Rd,0) be a Nash (cid:0) (cid:1) modification of Rd such that f ◦σ and the jacobian determinant jacσ have only normal crossings simultaneously, and assume moreover that σ is an isomorphism over the com- plement of the zero locus of f. Let (f ◦σ)−1(0) = ∪ E be the decomposition of (f ◦σ)−1(0) into irreducible compo- j∈J j nents, and assume that σ−1(0) = ∪ E for some K ⊂ J. k∈K k Put N = mult f ◦σ and ν = 1+mult jacσ, and, for I ⊂ J, denote by E0 the set i Ei i Ei I (∩ E )\(∪ E ). Then i∈I i j∈J\I j Z (u,T) = (u−1)|I|β E0∩σ−1(0) Φ (T) f X (cid:0) I (cid:1) I I6=∅ where Φ (T) = u−νiTNi . I Qi∈I 1−u−νiTNi In thecase with sign, let us definefirstcoverings of the exceptional strata E0 as follows. I Let U be an affine open subset of M such that f ◦σ = u yNi on U, where u is a Qi∈I i Nash function that does not vanish. Let us put 1 R± = {(x,t) ∈ (E0 ∩U)×R;tm = ± }, U I u(x) ] where m = gcd(N ). Then the R± glue together along the E0∩U to give E0,±. i U I I Proposition 2.9. With the assumptions and notations of proposition 2.8, one can express the zeta functions with sign in terms of a Nash modification as: ] u−νiTNi Z±(T) = (u−1)|I|−1β E0,±∩σ−1(0) . f X (cid:0) I (cid:1)Y 1−u−νiTNi I6=∅ i∈I Remark 2.10. The proof of propositions 2.8 and 2.9 in the Nash case run as in the algebraic one (cf. [2] for example, which is already an adaptation to the real case of [1]). In particular, in the remaining of this section, we prove that we can apply the same method. The main point is that we dispose of a Kontsevich change of variables formula in the Nash case. In order to prove this, the following lemma is crucial. 7 Lemma 2.11. Let h : M,h−1(0) −→ (Rd,0) be a proper surjective Nash map. (cid:0) (cid:1) Put ∆ = {γ ∈ L(M,E);ord jach γ(t) = e}, e t (cid:0) (cid:1) for an integer e ≥ 1, and ∆ = π (∆ ). e,n n e For e ≥ 1 and n ≥ 2n, then h (∆ ) is arc-symmetric and h is a piecewise trivial n e,n n fibration over ∆ , where the pieces are arc-symmetric sets, with fiber Re. e,n As an intermediate result, note the following elementary lemma whose proof is based on Taylor’s formula (cf. [1]). Lemma 2.12. Take e ≥ 1 and n ≥ 2e. Then, if γ ,γ ∈ L(M,E), then if γ ∈ ∆ and 1 2 1 e h(γ ) ≡ h(γ ) mod tn+1 then γ ∈ ∆ and γ ≡ γ mod tn−e+1. 1 2 2 e 1 2 Proof of lemma 2.11. It follows from lemma 2.12 that h is injective in restriction to n ∆ ∩π L(M,E) , and that h ∆ ∩π L(M,E) = h (∆ ). Then h (∆ ) e,n n−e n(cid:16) e,n n−e (cid:17) n e,n n e,n (cid:0) (cid:1) (cid:0) (cid:1) is arc-symmetric, as being the image by an injective Nash map of an arc-symmetric set (more precisely a constructible set). Now, the remaining of the proof can be carried on exactly as in [1]. (cid:3) To obtain the Kontsevich change of variables formula for a Nash modification, and therefore propositions 2.8 and 2.9, it suffices to follow the same computation as in [2]. Indeed, lemma 2.11 enables to apply word by word the method exposed in [2], just by replacing “constructible sets” by “arc-symmetric sets”. Now we can detail the proof of proposition 2.6. Proof of proposition 2.6. Let us prove the proposition in the case of the naive zeta func- tions. Letf,g :(Rd,0) −→ (R,0) beblow-Nash equivalent Nash function germs. By definition of the blow-Nash equivalence, there exist two Nash modifications, joined together by a commutative diagram as in definition 1.12. By a sequence of blowings-up with smooth Nash centres, one can make the jacobian determinantshavingonlynormalcrossings. Onecanassumemoreoverthattheexceptional divisors have also only normal crossings with the ones of the previous Nash modifications, so that we are in situation to apply the Denef & Loeser formula. Then,itis sufficient toprove thattheexpressions of thezeta functions of thegerms, ob- tainedviatheDenef&Loeserformula,coincide. Now,thetermsoftheformβ E0∩σ−1(0) (cid:0) I (cid:1) are equal since the virtual Poincar´e polynomial β is invariant under Nash isomorphisms (cf. proposition 2.3) and the N remain the same because of the commutativity of the i diagram (cf. definition 1.12). Finally, the ν coincide due to the additional assumption on i the blow-Nash homeomorphism to be a blow-Nash isomorphism. (cid:3) 3. Evaluating the zeta functions InordertoperformaclassificationofNashfunctiongermsunderblow-Nashequivalence, one needs invariants for this equivalence relation. The only ones known until now are the Fukui invariants [10] and the zeta functions of Koike-Parusin´ski defined with the Euler caracteristic with compact supports [11]. However, for the stronger notion of blow- Nash equivalence, the zeta functions obtained via the virtual Poincar´e polynomial are also invariants (cf. proposition 2.6). In this section, we define new invariants for the blow-Nash equivalence. These new invariants are derived from the zeta functions of a Nash function germ introduced in section 2.2. Recall that the zeta functions are formal power series in the indeterminacy T withcoefficientsinZ[u,u−1]. Thenthenewinvariantsareobtainedfromthezetafunctions by evaluating u in an appropriate way. 8 3.1. Evaluate u at −1. To begin with, let us note that we recover the zeta functions defined by S. Koike and A. Parusin´ski in [11], which has been proven to be invariants for the blow-analytic equivalence of real analytic function germs, by evaluating the zeta functions of section 2.2 at u= −1. Indeed, one recover the Euler characteristic with compact supports by evaluating the virtual Poincar´e polynomial at u = −1 (cf. remark 2.7.3). Remark 3.1. We recover also the zeta functions with sign in [11] of a Nash function germ f as −2Z±(−1,T). Indeed, their ones are defined by considering the value under f the Euler characteristic with compact supports χ of the set of arcs c Y± := {γ ∈ L : f ◦γ(t) = btn+··· ,±b> 0}. n n But X±×R∗ −→ Y±, (γ(t),a) 7→ γ(at) is a homeomorphism, therefore n + n χ (Y±)= χ (R∗)·χ (X±) = −2χ (X±). c n c + c n c n As a consequence: Proposition 3.2. Let f,g : (Rd,0) −→ (R,0) be blow-Nash equivalent germs of Nash functions. Then Z (−1,T) = Z (−1,T), f g and Z+(−1,T) = Z+(−1,T), Z−(−1,T) = Z−(−1,T). f g f g Remark 3.3. (1) This is also a direct consequence of the proof of proposition 2.6 because by a blow-Nash homeomorphism, just the parity of the ν are preserved. i (2) As an application, it follows from [11] that we can state the classification of the Brieskorn polynomials of two variables f = ±xp ± yq, p,q ∈ N under blow- p,q Nash equivalence, by using the zeta functions evaluated at u= −1 and the Fukui invariants. We will see another approach in section 3.3. 3.2. Evaluate u at 1. In a similar way, one can evaluate the zeta functions at 1. In the case of the naive zeta function, what we obtain is only zero! Nevertheless, one can obtain finer invariants. Actually, let us decompose the naive zeta function Z (u,T) of a Nash f function germ f in the following way: Z (u,T) = (u−1)lz (u,T), f X f,l l≥1 where z (u,T) is a formal power series in T with coefficient in Z[u,u−1] which is not f,l divisible by u−1. Similarly, decompose the zeta functions with sign: Z±(u,T) = (u−1)lz±(u,T). f X f,l l≥0 Note that here the index of the sum may begin at 0. By evaluating these series in Z[u,u−1][[T]] at u = 1, one finds new invariants for the blow-Nash equivalence. Theorem 3.4. Let f,g : (Rd,0) −→ (R,0) be blow-Nash equivalent germs of Nash func- tions. Then z (1,T) = z (1,T), f,1 g,1 z± (1,T) = z± (1,T), f,0 g,0 and z (1,T) ≡ z (1,T) mod 2, f,2 g,2 9 z± (1,T) ≡ z± (1,T) mod 2. f,1 g,1 Note that by mod 2 congruence we mean equality of the series considered as elements Z in [[T]]. 2Z Remark 3.5. For k ≥ 2, then the series z± (1,T) and z (1,T) are also invariant mod f,k f,k+1 2, but unfortunately they just vanish! Proof. This is a consequence of the Denef & Loeser formulae given in propositions 2.8 and 2.9. Let us concentrate firstly on the naive case. Actually, note that Z (u,T) Z (u,T) f g z (1,T) = lim and z (1,T) = lim , f,1 g,1 u→1 u−1 u→1 u−1 that is z (1,T) (respectively z (1,T)) is the derivative with respect to u of Z (u,T) f,1 g,1 f (respectively Z (u,T)) evaluated at u = 1. One can express these quotients via the Denef g & Loeser formula (proposition 2.8). As Z (u,T) and Z (u,T) are divisible by u − 1, f g these quotients coincide except the coefficients ν , which only have the same parity. By i evaluating u at 1, we obtain the equality z (1,T) = z (1,T). f,1 g,1 Similarly, z (1,T) is the derivative of Zf(u,T) evaluated at u = 1. However, the f,2 u−1 derivative of quotients of the type u−νTN arriving in the expression of the Denef & 1−u−νTN Loeser formula for Z (u,T) are of the form f uν−1TN −ν . (1−u−νTN)2 Therefore the mod 2 congruence of z (1,T) and z (1,T) comes from the mod 2 con- f,2 g,2 gruence of the different ν. One just have to repeat the same arguments with z± (1,T) and z± (1,T) in order to f,0 f,1 complete the proof of the theorem in the cases with sign. (cid:3) Example 3.6. Let f be the Brieskorn polynomial defined by p,k f = ±(xp+ykp+zkp), p even, k ∈ N. p,k It is not known whether two such polynomials are blow-analyticaly equivalent or not. However we prove below that for fixed p and different k, two such polynomials are not blow-Nash equivalent. Note that in [2], we established the analog result concerning the blow-Nash equivalence via blow-Nash isomorphism, by using the naive zeta functions. Actually, the naive zeta function Z of f looks like fp,k p,k Z = (u−1) u−1Tp+u−2T2p+···+u−(k−1)T(k−1)p +(u3−1)u−k−2Tkp fp,k (cid:0) (cid:1) +(u−1) u−(k+3)T(k+1)p+u−(k+4)T(k+2)p+···+u−(2k+1)T(2k−1)p (cid:0) (cid:1) +(u3−1)u−2(k−2)T2kp+··· . Now, for p fixed and k < k′, the pk-coefficient of Z is (u3 −1)u−k−2 whereas the one fp,k of Z is (u−1)u−k. Therefore, the pk-coefficient of z equals 2 whereas the one of fp,k′ fp,k,1 zfp,k′,1 is 1, and so fp,k and fp,k′ are not blow-Nash equivalent. 10