An algorithm for annihilator and Bernstein-Sato polynomial of a rational function Thanks: JM is partially supported by the European Union NextGenerationEU/PRTR and by MICIU/AEI/10.13039/501100011033 (grant codes: RYC2021-034300-I and CNS2024-154271). MG, EL, and JM are partially supported by MICIU/AEI/10.13039/501100011033 (grant code: PID2020-114750GB-C31) and by Departamento de Ciencia, Universidad y Sociedad del Conocimiento del Gobierno de Aragón (grant code: E22_20R: “Álgebra y Geometría”). JM is also supported by Junta de Andalucía (FQM-333). MG and EL are also partially supported by the European Union NextGenerationEU/PRTR and UNIZAR via María Zambrano’s Program and by CONAHCYT project CF-2023-G33. Manuel González-Villa, Edwin León-Cardenal, Viktor Levandovskyy, Jorge Martín-Morales Abstract . The singularity theory of rational functions, i.e., the quotient of two polynomials, has been investigated in the past two decades. The Bernstein-Sato polynomial of a rational function has recently been introduced by Takeuchi. However, only trivial examples are known. We provide an algorithm for computing the Bernstein-Sato polynomial in this context. The strategy is to compute the annihilator of the rational function by using the annihilator of the pair consisting of the numerator and denominator of the quotient. In a natural way a non-vanishing condition on the Bernstein-Sato ideal of the pair appears. This method has been implemented in freely available computer algebra system Singular . It relies on Gröbner bases in noncommutative PBW algebras. The algorithm allows us to exhibit some explicit non-trivial examples and to support some existing conjectures. 2020 Mathematics Subject Classification 14F10, 16S32, 68W30, 32S40 1. Introduction The theory of b -functions was developed by M. Sato during the 60’s under the name of algebraic theory of linear differential systems [ 32 ] . The main purpose was to study prehomogeneous vector spaces and their zeta functions [ 31 , 30 ] . A similar development was taking form a few years later with the study of the properties of the ring of differential operators by I. N. Bernstein [ 7 , 8 ] . The whole mathematical construction is known today as \mathcal{D} -module theory and it has connections in many areas of mathematics such as algebraic geometry, singularity theory, topology of varieties, representation theory and differential equations, among others. For a nice introduction to \mathcal{D} -module theory we refer to [ 9 ] . Denote by D the n th Weyl algebra over \mathbb{C} , i.e., the ring of differential operators with polynomial coefficients in n variables over \mathbb{C} or \mathbb{C}[x]\langle\partial\rangle=\mathbb{C}[x_{1},\ldots,x_{n}]\langle\partial_{1},\ldots,\partial_{n}\rangle, where the variables obey the following noncommutative relations \partial_{i}x_{i}=x_{i}\partial_{i}+1 , i=1,\ldots,n . Pick another variable s and consider the algebra D[s]=D\otimes_{\mathbb{C}}\mathbb{C}[s] . Given a non-constant polynomial f\in\mathbb{C}[x] we denote by \mathbb{C}[x,s,\frac{1}{f}] the localization of \mathbb{C}[x,s] with respect to the multiplicative set \{f^{k}\}_{k\in\mathbb{Z}_{\geq 0}} . The free \mathbb{C}[x,s,\frac{1}{f}] -module of rank one generated by f^{s} , M=\mathbb{C}[x,s,\frac{1}{f}]f^{s} , has a natural structure of D[s] -module. A key result [ 8 ] in the theory of \mathcal{D} -modules asserts that there exists an element P(s)\in D[s] and a nonzero polynomial b(s)\in\mathbb{C}[s] verifying the following functional equation (1) P(s)f^{s+1}=b(s)f^{s}. The set of all polynomials b(s) of \mathbb{C}[s] satisfying ( 1 ) forms an ideal, and its unique monic generator is called the Bernstein-Sato polynomial b_{f}(s) of f . The roots of b_{f}(s) encode invariants of the singularities of f . Kashiwara showed that they are negative rational numbers [ 15 ] . Malgrange introduced the V -filtration and proved that if s_{0} is a root of b_{f}(s) , then e^{-2\pi is_{0}} is an eigenvalue of the complex monodromy of f [ 19 , 20 ] . A more general functional equation that will be relevant in this work involves several functions. Given p polynomials f_{1},\ldots,f_{p} in \mathbb{C}[x_{1},\ldots,x_{n}] , Sabbah [ 28 ] showed that there exists a nonzero polynomial b(s_{1},\ldots,s_{p})\in\mathbb{C}[s_{1},\ldots,s_{p}] such that (2) P(s_{1},\ldots,s_{p})f_{1}^{s_{1}+1}\cdots f_{p}^{s_{p}+1}=b(s_{1},\ldots,s_{p})f_{1}^{s_{1}}\cdots f_{p}^{s_{p}}, with P(s_{1},\ldots,s_{p})\in D[s_{1},\ldots,s_{p}]=D\otimes_{\mathbb{C}}\mathbb{C}[s_{1},\ldots,s_{p}] . The set of polynomials satisfying ( 2 ) forms an ideal in \mathbb{C}[s_{1},\ldots,s_{p}] that we refer to as the Bernstein-Sato ideal of f_{1},\ldots,f_{p} and it is denoted by B_{f_{1},\ldots,f_{p}}(s_{1},\ldots,s_{p}) . For an extensive discussion of the connections between b -functions and the corresponding singularities, see for instance [ 26 , 2 ] . On the computational side, Oaku [ 23 ] designed the first general algorithm for computing the annihilator of f^{s} and obtain b_{f}(s) . The use of Gröbner deformations and initial forms to study Bernstein-Sato polynomials is explained by Saito, Sturmfels, and Takayama [ 29 ] . Then Noro [ 21 ] provided a modular and linear-algebra-driven algorithm for computing the global b -function. A family of algorithms for checking candidate roots of b_{f}(s) is presented by Levandovskyy and Martín-Morales [ 17 ] . For an overview of other computational contributions, see e.g. [ 34 ] . Some of these methods have been implemented in different computer algebra systems [ 25 , 18 , 16 , 3 ] . Recently Bernstein-Sato polynomial for rational functions was introduced by Takeuchi [ 33 ] . The Bernstein-Sato polynomial of \frac{f}{g} , related to the given nonnegative integer m , is the monic polynomial of smallest degree satisfying the relation b_{\frac{f}{g},m}(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}\in\sum_{k=1}^{\infty}D[s]\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s+k} in the free \mathbb{C}[x,s,\frac{1}{fg}] -module of rank one \mathbb{C}[x,s,\frac{1}{fg}]\left(\frac{f}{g}\right)^{s} . A similar approach was proposed by Àlvarez Montaner, González Villa, León-Cardenal and Núñez-Betancourt in [ 1 ] . The polynomial b_{\frac{f}{g},m}(s) is shown to satisfy some expected relations with respect to other invariants of the singularity of rational functions. However, only trivial examples are known. In this work we provide an algorithm to compute b_{\frac{f}{g},m}^{(N)}(s) defined by the functional equation (3) b_{\frac{f}{g},m}^{(N)}(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}=\sum_{k=1}^{N}P_{k}(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s+k}, where P_{k}(s)\in D[s] . Note that b_{\frac{f}{g},m}(s)=b_{\frac{f}{g},m}^{(N)}(s) when N is big enough. In Section 6 we provide explicit examples of the Bernstein-Sato polynomial of rational functions. Following Oaku’s original ideas [ 23 ] we first compute the annihilator in D[s] of \frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s} denoted by (4) \Ann_{D[s]}\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}. Towards the computation of ( 4 ), we first compute the annihilator in D[s_{1},s_{2}] of f^{s_{1}}g^{s_{2}} , see e.g. [ 22 ] , that we denote by I(s_{1},s_{2}) . After the substitution s_{1}=s and s_{2}=-s-m we obtain a new ideal I(s,-s-m) in D[s] that it is in general not \mathbb{C}[s] -saturated. Hence I(s,-s-m) does not agree with the one in ( 4 ). Theorem 4.5 gives a sufficient non-vanishing condition on the Bernstein-Sato ideal of the pair (f,g) to recover the annihilator of \frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s} as the saturation of I(s,-s-m) . Even if this condition is not satisfied a careful analysis of the factors of the form s_{1}+s_{2}+\ell with \ell\in\mathbb{Z} allows us to compute the annihilator ( 4 ), see Section 4.1 . These algorithms have been implemented in Singular to exhibit explicit examples. The paper is organized as follows. In Section 2 , we give the necessary definitions and notations for the concepts to be dealt with in this paper such as Bernstein-Sato polynomial of rational functions and their related modules. In Section 3 , we recall a method for computing the saturation of an ideal in a noncommutative context. The main result of our work, namely Theorem 4.5 , is presented in Section 4 , and this allowed us to describe an algorithm to compute b_{\frac{f}{g}}^{(N)}(s) in Section 5 . Finally, Section 6 is devoted to showing some examples. 2. Settings Let f,g\in\mathbb{C}[x]=\mathbb{C}[x_{1},\ldots,x_{n}] be two non-constant polynomials and choose another variable s . We denote by \mathbb{C}[x,s,\frac{1}{fg}] the localization of \mathbb{C}[x,s] with respect to the multiplicatively closed set \{(fg)^{k}\}_{k\in\mathbb{Z}_{\geq 0}} . Consider the free \mathbb{C}[x,s,\frac{1}{fg}] -module of rank one M=\mathbb{C}\left[x,s,\frac{1}{fg}\right]\left(\frac{f}{g}\right)^{s}. Note that M has a natural structure of D[s] -module, where D[s]=D\otimes_{\mathbb{C}}\mathbb{C}[s] and D is the n th Weyl algebra, as follows \displaystyle x_{i}\bullet\frac{h(s)}{(fg)^{k}}\left(\frac{f}{g}\right)^{s} \displaystyle=\frac{x_{i}h(s)}{(fg)^{k}}\left(\frac{f}{g}\right)^{s}, \displaystyle s\bullet\frac{h(s)}{(fg)^{k}}\left(\frac{f}{g}\right)^{s} \displaystyle=\frac{sh(s)}{(fg)^{k}}\left(\frac{f}{g}\right)^{s}, \displaystyle\partial_{i}\bullet\frac{h(s)}{(fg)^{k}}\left(\frac{f}{g}\right)^{s} \displaystyle=\left(\frac{\partial}{\partial x_{i}}\left(\frac{h(s)}{(fg)^{k}}\right)+\frac{s\frac{\partial f}{\partial x_{i}}h(s)}{f(fg)^{k}}-\frac{s\frac{\partial g}{\partial x_{i}}h(s)}{g(fg)^{k}}\right)\left(\frac{f}{g}\right)^{s}, for h(s)\in\mathbb{C}[x,s] and k\in\mathbb{Z}_{\geq 0} , cf. [ 33 , Equation 1.4-1.5] and [ 1 , Definition 3.6] . In other words, polynomials in x,s act by the regular multiplication, while \partial_{i} acts as the partial differentiation. Theorem 2.1 . ( [ 33 , Theorem 1.1] ) Given a nonnegative integer m there exists a nonzero polynomial b_{\frac{f}{g},m}(s)\in\mathbb{C}[s] such that b_{\frac{f}{g},m}(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}\in\sum_{k=1}^{\infty}D[s]\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s+k}. This result is a global version of the original Theorem by Takeuchi. In particular, he defines a local version of the Bernstein-Sato polynomial, that is denoted by b_{\frac{f}{g},m,x_{0}}(s) for x_{0}\in\mathbb{C}^{n} . It is worth emphasizing the importance of m in relation to the local Milnor monodromy of the rational function [ 11 , 12 ] . More precisely, consider the set E_{\frac{f}{g},x_{0}} of eigenvalues of the local Milnor fibration at x_{0} and at points close to x_{0} . Moreover, assume that m\geq 2n for the ambient dimension n , then in [ 33 , Theorem 1.4] Takeuchi proves, that (5) \left\{\exp(2\pi i\alpha)\mid b_{\frac{f}{g},m,x_{0}}(\alpha)=0\right\}=E_{\frac{f}{g},x_{0}}. If m<2n , then only the inclusion \{\exp(2\pi i\alpha)\mid b_{\frac{f}{g},m,x_{0}}(\alpha)=0\}\subseteq E_{\frac{f}{g},x_{0}} holds in general. We end this preliminary section with the definition of the Bernstein-Sato ideal of the pair f,g that will be relevant for our purpose, see Theorem 4.5 . Its existence was first proven by Sabbah [ 28 ] . Definition 2.2 . Given two non-constant polynomials f,g\in\mathbb{C}[x_{1},\ldots,x_{n}] , there exists a nonzero polynomial b(s_{1},s_{2})\in\mathbb{C}[s_{1},s_{2}] such that b(s_{1},s_{2})f^{s_{1}}g^{s_{2}}=P(s_{1},s_{2})f^{s_{1}+1}g^{s_{2}+1}, where P(s_{1},s_{2})\in D[s_{1},s_{2}]=D\otimes_{\mathbb{C}}\mathbb{C}[s_{1},s_{2}] . The set of polynomials b(s_{1},s_{2}) satisfying this functional equation forms an ideal in \mathbb{C}[s_{1},s_{2}] , which is called the Bernstein-Sato ideal of f,g , and is denoted by B_{f,g} . 3. Central saturation The goal of this section is to recall an algorithm to compute the saturation with respect to \mathbb{C}[s] of a left ideal in D[s] starting from a system of generators of the ideal. The formal definition is given below. We refer to e.g. [ 6 , §8.7] for a commutative version of this method and to [ 14 ] for a more general approach including the noncommutative case. Note, that the center of D[s] is precisely the ring \mathbb{C}[s] , and this holds for an arbitrary number of variables s_{1},\ldots,s_{\ell} . Hence, the only two-sided ideals in D[s] are those, extended from \mathbb{C}[s] . Moreover, the important technique of saturation , with which we deal below, can be treated by the much simpler central saturation . Definition 3.1 . Given a left ideal I\subset D[s] , we define the \mathbb{C}[s] - saturation of I as the left ideal \Sat_{\mathbb{C}[s]}(I):=\{P(s)\in D[s]\mid\exists q(s)\in\mathbb{C}[s]\setminus\{0\},\ q(s)P(s)\in I\}. This ideal contains I , i.e. I\subseteq\Sat_{\mathbb{C}[s]}(I) , and we say that I is \mathbb{C}[s] - saturated if the equality I=\Sat_{\mathbb{C}[s]}(I) takes place. Analogously, we can define this notion for a left ideal I\subset D[s_{1},\ldots,s_{\ell}] for \ell\geq 2 . Remark 3.2 . In a more general setting, saturation appears in the context of Ore localizations of modules over a Noetherian domain. It is taken with respect to a multiplicatively closed Ore set (also called a denominator set). Strictly speaking, we should say (\mathbb{C}[s]\setminus\{0\}) - saturated , but since we only deal with domains, we use the simpler term \mathbb{C}[s] - saturated . The localization of D[s] with respect to \mathbb{C}[s]\setminus\{0\} is denoted by D(s) and can be seen being naturally isomorphic to D\otimes_{\mathbb{C}}\mathbb{C}(s) via the extension of scalars. Thus the kernel of the homomorphism D[s]/D[s]I\to D(s)/D(s)I of D[s] -modules is precisely \Sat_{\mathbb{C}[s]}(I) . The following result, namely Proposition 3.3 , collects some properties of the saturation that will allow us to describe Algorithm 1 for computing it. Before that let us fix some notation. We focus on the case of the single variable {\ell}=1 . Let \iota:D[s]\hookrightarrow D(s) be the natural inclusion of D[s] -modules. By abuse of notation the image of P(s)\in D[s] is also denoted by P(s)\in D(s) . If I is a left ideal in D[s] , then I^{e}:=D(s)I denotes the extension of I . If J is a left ideal in D(s) , then J^{c}=J\cap D[s]=\iota^{-1}(J) denotes the contraction of J . In particular, I^{ec}:=(I^{e})^{c} denotes the contraction of the extension of I . We can assume that a generating set of a left ideal in D(s) we work with consists of elements from D[s] : this can always be achieved by clearing the denominators, which are polynomials in s . Let us fix a monomial well-ordering <_{D} on D . There are two natural orderings associated with <_{D} . The first one is the ordering <_{D(s)} on D(s) that is defined as <_{D} , since s belongs to the ground field of D(s) . The second one is the ordering <_{D[s]} on D[s] defined as follows: (6) s^{i_{1}}x^{\alpha_{1}}\partial^{\beta_{1}}<_{D[s]}s^{i_{2}}x^{\alpha_{2}}\partial^{\beta_{2}}\quad\Longleftrightarrow\quad\begin{cases}x^{\alpha_{1}}\partial^{\beta_{1}}<_{D}x^{\alpha_{2}}\partial^{\beta_{2}},\\ \quad\text{or}\\[3.0pt] x^{\alpha_{1}}\partial^{\beta_{1}}=x^{\alpha_{2}}\partial^{\beta_{2}}\ \text{and}\ i_{1}0 and k=m+i to obtain I_{m}(s)=\pi(\syz_{D[s]}(f^{e_{b(s_{1},s_{2})}-1-m},I_{e_{b(s_{1},s_{2})}-1}(s-e_{b(s_{1},s_{2})}+1+m))). Note that the computation of the previous ideal I_{e_{b(s_{1},s_{2})-1}}(s) is treated in ( b ). Remark 4.10 . Lemma 4.9 with i=1 and k=e_{b(s_{1},s_{2})}-j allows us to compute I_{e_{b(s_{1},s_{2})}-j-1}(s) from I_{e_{b(s_{1},s_{2})}-j}(s-1) for all j\geq 1 . This gives an alternative method to case ( c ) when e_{b(s_{1},s_{2})}-1-m is too large. Remark 4.11 . The choice of b(s_{1},s_{2}) from the system of generators of B_{f,g} with the smallest e_{b(s_{1},s_{2})} improves the efficiency of the computation of I_{m}(s) along the previous discussion. We summarize this method in the following algorithm. Input : f,g two polynomials in \mathbb{C}[x_{1},\ldots,x_{n}] ; m\geq 0 an integer. Output : \{P_{1}(s),\ldots,P_{r}(s)\}\subset D[s] a system of generators of I_{m}(s)=\Ann_{D[s]}\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s} . begin 1 I(s_{1},s_{2}):= Annihilator of f^{s_{1}}g^{s_{2}} in D[s_{1},s_{2}] ; 2 B_{f,g}:=\langle b_{1}(s_{1},s_{2}),\ldots,b_{r}(s_{1},s_{1})\rangle , Bernstein-Sato ideal of f,g in \mathbb{C}[s_{1},s_{2}] ; 3 \varepsilon:=\min\left\{e_{b_{i}(s_{1},s_{2})}\mid i=1,\ldots,r\right\}-1 ; /* see Remark 4.11 */ \gamma:=\varepsilon-m ; 4 if \gamma\leq 0 then 5 return \Sat_{\mathbb{C}[s]}(I(s,-s-m)) ; /* cases ( a ) and ( b ) above */ else 6 J(s):=\Sat_{\mathbb{C}[s]}(I(s,-s-\varepsilon)) ; /* equals I_{\varepsilon}(s) due to case ( b ) */ return \pi(\syz_{\mathbb{C}[s]}(f^{\gamma},J(s-\gamma))) ; /* case ( c ) */ end if 7 end 8 9 Algorithm 2 Annihilator ( computes the annihilator of \frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s} in D[s] ) Example 4.12 . Let us consider f=x^{2}+y^{2} and g=xy . The purpose of this example is to compute I_{m}(s) for all m\in\mathbb{Z} , m\geq 0 . The annihilator of f^{s_{1}}g^{s_{2}} in D[s_{1},s_{2}] is generated by two operators (12) \displaystyle P(s_{1},s_{2}) \displaystyle=x\partial_{x}+y\partial_{y}-2s_{1}-2s_{2}, \displaystyle Q(s_{1},s_{2}) \displaystyle=xy^{2}\partial_{x}-x^{2}y\partial_{y}+x^{2}s_{2}-y^{2}s_{2}, that is, I(s_{1},s_{2})=D[s_{1},s_{2}]\left\langle P(s_{1},s_{2}),Q(s_{1},s_{2})\right\rangle . The Bernstein-Sato ideal of f,g is generated by a single element (13) \displaystyle b(s_{1},s_{2})= \displaystyle(s_{1}+1)(s_{2}+1)\cdot \displaystyle(s_{1}+s_{2}+1)(s_{1}+s_{2}+2)(s_{1}+s_{2}+3)\cdot \displaystyle(2s_{1}+2s_{2}+3)(2s_{1}+2s_{2}+5). Using the algorithm described in Section 3 one checks that I(s,-s-m) is \mathbb{C}[s] -saturated for all m\in\mathbb{Z} , m\geq 0 . Following the notation in Section 4.1 (see Notation 4.8 and Algorithm 2 ) one obtains L_{b(s_{1},s_{2})}=\{1,2,3\},\qquad e_{b(s_{1},s_{2})}=3,\qquad\varepsilon=2,\qquad\gamma=2-m. Hence I_{m}(s)=\Sat_{\mathbb{C}[s]}(I(s,-s-m))=I(s,-s-m) for all m\geq 2 . It remains to compute I_{0}(s) and I_{1}(s) . For m=1 we use I_{1}(s)=\pi(\syz_{D[s]}(f,I_{2}(s-1))) and obtain three generators P(s,-s-1) , Q(s,-s-1) , R(s) , where R(s)=y^{2}\partial_{x}\partial_{y}-xy\partial_{y}^{2}-y\partial_{x}s-x\partial_{y}s+y\partial_{x}-2x\partial_{y}. Note that the new operator R(s) is not in I(s,-s-1) and then I_{1}(s)\supsetneq I(s,-s-1) . For m=0 there are two ways to compute I_{0}(s) , namely, either \pi(\syz_{D[s]}(f^{2},I_{2}(s-2))) or \pi(\syz_{D[s]}(f,I_{1}(s-1))) , see Remark 4.10 . The ideal I_{0}(s) is generated by three operators P(s,-s) , Q(s,-s) , T(s) , where T(s)=y\partial_{x}^{4}\partial_{y}-y\partial_{y}^{5}-\partial_{x}^{4}s+2\partial_{x}^{2}\partial_{y}^{2}s-\partial_{y}^{4}s-4\partial_{x}^{2}\partial_{y}^{2}-4\partial_{y}^{4}. Once again one verifies that I_{0}(s)\supsetneq I(s,-s) . 5. Bernstein-Sato polynomial of rational functions Let f,g\in\mathbb{C}[x_{1},\ldots,x_{n}] be two polynomials and m\in\mathbb{Z} , m\geq 0 . Takeuchi [ 33 ] showed that for a big enough N there exists a nonzero polynomial b(s)\in\mathbb{C}[s] and P_{1}(s),\ldots,P_{N}(s)\in D[s] such (14) b(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}=\sum_{k=1}^{N}P_{k}(s)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s+k}. The set of polynomials b(s) verifying a functional equation of the type ( 14 ) is an ideal in \mathbb{C}[s] . We denote by b_{\frac{f}{g},m}^{(N)}(s) its monic generator. In this section we present an algorithm to compute b_{\frac{f}{g},m}^{(N)}(s) for a fixed N\geq 1 . Note that expression ( 14 ) can alternatively be written as \left(b_{\frac{f}{g},m}^{(N)}(s)-\sum_{k=1}^{N}P_{k}(s)\frac{f^{k}}{g^{k}}\right)\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}=0. Cleaning the denominators in the first factor of the previous expression to get an element in D[s] we obtain an equivalent equation \left(b_{\frac{f}{g},m}^{(N)}(s)g^{N}-\sum_{k=1}^{N}P_{k}(s)f^{k}g^{N-k}\right)\frac{1}{g^{m+N}}\left(\frac{f}{g}\right)^{s}=0. In Section 4 , we have provided a method to compute the annihilator I_{m}(s):=\Ann_{D[s]}\frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s}. Using this notation we have (15) b_{\frac{f}{g},m}^{(N)}(s)g^{N}\in J\quad\mbox{where}\quad J=I_{m+N}(s)+\sum_{k=1}^{N}D[s]f^{k}g^{N-k}. Following the ideas in the proof of Lemma 4.9 and recalling Notation 4.8 , one finally gets the expression (16) \pi(\syz_{D[s]}(g^{N},J))\cap\mathbb{C}[s]=\Big\langle b_{\frac{f}{g},m}^{(N)}(s)\Big\rangle. Note that ( 16 ) provides a method for computing the Bernstein-Sato polynomial of \frac{f}{g} , once a system of generator of I_{m+N}(s)\subset D[s] has been obtained, see Section 4 . Example 5.1 . Consider f=x^{2}+y^{2} and g=xy as in Example 4.12 , where the annihilator of \frac{1}{g^{m}}\left(\frac{f}{g}\right)^{s} was computed for all m\geq 0 . Recall that I_{m}(s)=D[s]\langle P(s,-s-m),Q(s,-s-m)\rangle, for all m\geq 2 , where P(s_{1},s_{2}) and Q(s_{1},s_{2}) are the operators given in ( 12 ). Now we set m=1 and N=1 . Following ( 15 ) and ( 16 ), we first need to consider the ideal D[s]g+J , where J is generated by P(s,-s-2),Q(s,-s-2) and f . The syzygy module of the tuple (g,P(s,-s-2),Q(s,-s-2),f) is generated by the rows of the following matrix \begin{pmatrix}y\partial_{x}-x\partial_{y}&y^{2}&0&y\partial_{y}+1\\ x\partial_{x}+y\partial_{y}+2&xy&0&0\\ x^{2}+y^{2}&0&0&-xy\\ -y\partial_{x}+x\partial_{y}&x^{2}&0&x\partial_{x}+1\\ y^{2}\partial_{x}-xy\partial_{y}-2xs-2x&0&-y&ys+y\\ xy\partial_{x}+y^{2}\partial_{y}+2ys+4y&0&-x&-xy\partial_{y}-xs-2x\\ \end{pmatrix}. The ideal generated by the elements of the first column can also be generated by ys+y,\ xs+x,\ y\partial_{x}-x\partial_{y},\ x\partial_{x}+y\partial_{y}+2,\ x^{2}+y^{2}. and then the intersection with \mathbb{C}[s] is zero, that is, b_{\frac{f}{g},1}^{(1)}(s)=0 . This phenomenon has to do with the fact that the generator of B_{f,g} verifies that b(s,-s-m-2)=b(s,-s-3)=0 , see ( 13 ), cf. Oaku’s remark in [ 33 , p. 717] . In this case at least two terms are needed in Theorem 2.1 , i.e., two operators P_{1}(s) and P_{2}(s) in the functional equation ( 3 ) are nontrivial. Note that the number of operators needed to obtain the Bernstein-Sato polynomial can be very large, see Example 6.1 . For m=1 and N=2 the functional equation is given by (s+1)\frac{1}{g}\left(\frac{f}{g}\right)^{s}=P_{1}(s)\frac{1}{g}\left(\frac{f}{g}\right)^{s+1}+P_{2}(s)\frac{1}{g}\left(\frac{f}{g}\right)^{s+2}, where \displaystyle P_{1}(s) \displaystyle=\frac{1}{4}y^{4}\partial_{x}^{3}\partial_{y}-\frac{1}{4}xy^{3}\partial_{x}^{2}\partial_{y}^{2}+\frac{1}{4}y^{4}\partial_{x}\partial_{y}^{3}-\frac{1}{4}xy^{3}\partial_{y}^{4}-\frac{1}{2}y^{3}\partial_{x}^{3}s+\frac{1}{2}xy^{2}\partial_{x}^{2}\partial_{y}s-\frac{1}{4}y^{3}\partial_{x}^{3} \displaystyle+\frac{1}{2}xy^{2}\partial_{x}^{2}\partial_{y}+\frac{3}{4}y^{3}\partial_{x}\partial_{y}^{2}-\frac{1}{2}xy^{2}\partial_{y}^{3}+2y^{2}\partial_{x}\partial_{y}s+\frac{15}{4}y^{2}\partial_{x}\partial_{y}+\frac{1}{4}xy\partial_{y}^{2}+\frac{1}{2}y\partial_{x}s+\frac{5}{4}y\partial_{x}, \displaystyle P_{2}(s) \displaystyle=-\frac{1}{4}y^{3}\partial_{x}^{2}\partial_{y}s-\frac{1}{4}y^{3}\partial_{y}^{3}s-\frac{1}{2}y^{3}\partial_{x}^{2}\partial_{y}-\frac{1}{2}y^{3}\partial_{y}^{3}-\frac{1}{2}y^{2}\partial_{y}^{2}s-y^{2}\partial_{y}^{2}+\frac{1}{4}y\partial_{y}s+\frac{1}{2}y\partial_{y}. Hence the Bernstein-Sato polynomial of \frac{f}{g} for m=1 is s+1 . Alternatively, if one knows in advance that b_{\frac{f}{g},m}^{(N)}(s) is not zero, there is a better approach for the computation of the Bernstein-Sato polynomial. Indeed, ( 15 ) says that \overline{b_{\frac{f}{g},m}^{(N)}(s)g^{N}}=\overline{0}\in\frac{D[s]}{J}. Then the set \left\{\overline{1\cdot g^{N}},\overline{s\cdot g^{N}},\overline{s^{2}\cdot g^{N}},\ldots\right\} is linearly dependent in \frac{D[s]}{J} and the linear relation with minimal degree in s is precisely the Bernstein-Sato polynomial of \frac{f}{g} . This approach is much more efficient that the one given by formula ( 16 ). By contrast, if b_{\frac{f}{g},m}^{(N)}(s) is zero, then the set \big\{\overline{s^{i}\cdot g^{N}}\big\}_{i\geq 0} is linearly independent. We finish this section presenting an algorithm for computing b_{\frac{f}{g},m}^{(N)}(s) assuming it is not zero. Its correctness follows from the previous discussion. Input : f,g two polynomials in \mathbb{C}[x_{1},\ldots,x_{n}] ; m\geq 0 and N\geq 1 integers; < a monomial order on D[s] . Output : b_{\frac{f}{g},m}^{(N)}(s)\in\mathbb{C}[s] , a monic generator of the ideal defined by ( 14 ). begin 1 I:=I_{m+N}(s) ; /* use Algorithm 2 */ J:=I+D[s]\langle fg^{N-1},f^{2}g^{N-2},\ldots,f^{N}\rangle ; /* see ( 15 ) */ G:= Gröbner basis of J with respect to < ; 2 P_{0}:=NF(g^{N},G) ; /* the remainder of the division of g^{N} by G */ if P_{0}=0 then 3 return b=1 ; 4 end if 5 k:=1 ; 6 while true do 7 P_{k}=NF(s^{k}g^{N},G) ; 8 if \exists a_{0},\ldots,a_{k}\in\mathbb{Q} (with a_{k}=1 ) such that a_{0}P_{0}+\ldots+a_{k}P_{k}=0 then 9 return b=a_{0}+a_{1}s+\ldots+a_{k}s^{k} ; /* by the assumption b_{\frac{f}{g},m}^{(N)}(s)\neq 0 */ end if 10 k:=k+1 ; 11 end while 12 end 13 14 Algorithm 3 Bernstein-Sato ( computes b_{\frac{f}{g},m}^{(N)}(s) assuming it is not zero ) 6. Examples Example 6.1 . Let us consider f=x^{2}+y^{3} and g=xy . For all m\geq 0 the ideal I(s,-s-m) is \mathbb{C}[s] -saturated and the Bernstein-Sato ideal satisfies the \mathcal{C}_{m} -conditions. Then, by Theorem 4.5 , I_{m}(s)=I(s,-s-m) for all m\geq 0 . Running Algorithm 3 for m=0 one obtains \displaystyle b_{\frac{f}{g},0}^{(1)}(s) \displaystyle=(s+1)(s+5)(s+7), \displaystyle b_{\frac{f}{g},0}^{(N)}(s) \displaystyle=(s+1)(s+5),\quad N=2,\ldots,5, \displaystyle b_{\frac{f}{g},0}^{(6)}(s) \displaystyle=s+1. In this case we need six differential operators to achieve the Bernstein-Sato polynomial of \frac{f}{g} which is s+1 , that is, b_{\frac{f}{g},0}(s)=b_{\frac{f}{g},0}^{(N)}(s) for all N\geq 6 . On the other hand, for m=1 Algorithm 3 yields \displaystyle b_{\frac{f}{g},1}^{(1)}(s) \displaystyle=s(s+1)(s+2), \displaystyle b_{\frac{f}{g},1}^{(N)}(s) \displaystyle=s(s+1),\quad N=2,\ldots,20. However, we could not verify whether b_{\frac{f}{g},1}^{(N)}(s)=s+1 for N big enough as in case m=0 . In the next example we will use the eigenvalues of the monodromy of a rational function [ 11 , 12 ] to determine the Bernstein-Sato polynomial. Example 6.2 . Let f=x^{2}+y^{3} , g=x , and m=4 . The Bernstein-Sato ideal satisfies the \mathcal{C}_{m} -condition and the ideal I(s,-s-m) is \mathbb{C}[s] -saturated for all m\geq 0 . According to Theorem 4.5 I_{m}(s)=I(s,-s-m) for all m\geq 0 . Algorithm 3 for N=1 produces (17) b_{\frac{f}{g},4}^{(1)}(s)=\Big(s+1\Big)\Big(s-\frac{7}{3}\Big)\Big(s-\frac{5}{3}\Big). Note that not all the roots are negative by contrast with the classical case [ 15 ] . We plan to use the equality ( 5 ) to show that the polynomial in ( 17 ) is in fact the Bernstein-Sato polynomial of \frac{f}{g} . A resolution \pi:X\to(\mathbb{C}^{2},0) of the singularity of the quotient can be achieved with four blow-ups. The final situation is illustrated in Figure 1 where the multiplicities of f and g along the exceptional divisors are shown. By [ 11 , Theorem 1] the local monodromy zeta functions of the rational function are Z_{{\rm mon},\frac{f}{g}}(t)_{0}=\frac{1-t}{1-t^{3}}=\frac{1}{t^{2}+t+1},\qquad Z_{{\rm mon},\frac{f}{g}}(t)_{a}=1-t,\quad a\in V(f)\setminus\{0\}, and then the eigenvalues are all the roots of unity of order 3 . Equality ( 5 ) implies that b_{\frac{f}{g},4}(s) has exactly 3 different roots modulo \mathbb{Z} . Since b_{\frac{f}{g},4}(s) divides b_{\frac{f}{g},4}^{(N)}(s) which in turn divides b_{\frac{f}{g},4}^{(1)}(s) for all N\geq 1 and the latter has already 3 different roots modulo \mathbb{Z} , it follows that they all must be equal. E_{3}(6,3) E_{1}(2,1) \hat{f} E_{2}(3,2) E_{4}(3,3) \hat{g} Figure 1. Resolution of \frac{x^{2}+y^{3}}{x} . The resolution help us also to get a bound for the roots of the Bernstein-Sato polynomial according to [ 33 , Theorem 4.1] . More precisely, all the roots of the polynomial ( 17 ) are contained in the set B_{\frac{f}{g},m}^{\pi}=\left\{m-\frac{1}{3}-\ell,\ m-\frac{2}{3}-\ell,\ 2m-1-\ell\ \Big|\ \ell\in\mathbb{Z},\,\ell\geq 0\right\}. Notice that the elements of the form m-\frac{1}{3}-\ell and m-\frac{2}{3}-\ell come from E_{3} while 2m-1-\ell comes from E_{2} . Heuristically we have observed that b_{\frac{f}{g},m}^{(N)}(s)=\Big(s+1\Big)\Big(s-m+\frac{5}{3}\Big)\Big(s-m+\frac{7}{3}\Big). However, for m=0,1,2,3 , we cannot use ( 5 ) and conclude that the Bernstein-Sato polynomial of the rational function has really three factors. It is worth mentioning that the weak monodromy conjecture for plane rational functions was proven in [ 10 ] . In this case the local topological zeta function at the origin of \frac{f}{g} is Z_{\textrm{top},\frac{f}{g}}(s)_{o}=\frac{11s+15}{12\big(s+1\big)\big(s+\frac{5}{3}\big)}. In the context of the monodromy conjecture this example suggests that it is natural to relate the poles of Z_{\textrm{top},\frac{f}{g}}(s)_{o} with the roots of b_{\frac{f}{g},0}(s) , that is, when m=0 . Example 6.3 . Consider f=x^{6}+y^{6}+2zx^{3}y^{3} and g=z^{2} . The Bernstein-Sato ideal of f,g was computed in [ 5 ] B_{f,g}=b(s_{1},s_{2})\langle 3s_{1}+5,2s_{2}+3\rangle, where \displaystyle b(s_{1},s_{2})= \displaystyle(s_{1}+1)^{2}(2s_{1}+1)(2s_{1}+3)(3s_{1}+1)(3s_{1}+2)\cdot \displaystyle(3s_{1}+4)(6s_{1}+5)(6s_{1}+7)(s_{2}+1)(2s_{2}+1) Then B_{f,g} satisfies the \mathcal{C}_{m} -condition for all m\geq 0 and by Theorem 4.5 , I_{m}(s)=\Sat_{\mathbb{C}[s]}(I(s,-s-m)),\ \forall m\geq 0. Using the methods from Section 3 , we have checked that the ideal I(s,-s-m) is \mathbb{C}[s] -saturated for all m\geq 2 while I(s,-s-1) is not. In fact, I_{1}(s)=\Sat_{\mathbb{C}[s]}(I(s,-s-1))=I(s,-s-1):\Big(s+\frac{1}{2}\Big)\supsetneq I(s,-s-1). Algorithm 3 for m=0 and N=1,\ldots,4 yields b_{\frac{f}{g},0}^{(N)}(s)=\Big(s+1\Big)^{2}\Big(s+\frac{1}{2}\Big)\Big(s+\frac{3}{2}\Big)\Big(s+\frac{1}{3}\Big)\Big(s+\frac{2}{3}\Big)\Big(s+\frac{4}{3}\Big)\Big(s+\frac{5}{6}\Big)\Big(s+\frac{7}{6}\Big). Note that s+\frac{1}{2} remains as a factor of b_{\frac{f}{g},0}^{(N)}(s) , although it plays a role in the saturation of I(s,-s-1) .