The Bernstein-Sato Polynomial

Q: Is b_p ever equal to 1?

Only when p is a non-zero constant. Then p^s+1 = p,p^s , so D = p^-1 works with b = 1 . For any non-constant p the divisibility by s+1 forces b_p ≥ 1 .

Q: How does finite length produce the equation?

The submodules A_n(K(s))p^kp^s for k = 0,1,2, … form a descending chain in a holonomic module. Finite length makes it stationary, so p^kp^s ∈ A_n(K(s))p^k+1p^s for some k , and an automorphism shifting s by one moves this back to k = 0 . Clearing denominators in s gives b and D . The details are on the existence page .

Q: Are the operator D and the polynomial b unique?

b_p is unique because it is the monic generator of an ideal. D is never unique: any operator annihilating p^s+1 can be added to it, and the annihilator of p^s+1 is a large left ideal.

Q: Can I compute b_p by finding some operator and reading off the scalar?

That gives an element of B_p , hence a multiple of b_p , which is often all one needs. Proving that a candidate is the generator requires a lower bound argument - a grading argument as in the worked example, a specialisation, or a computer algebra elimination.

Q: Does the definition need p to be reduced or irreducible?

No. Any non-zero p has a b-function. Non-reducedness shows up in the roots: b_x^2 = (s+ 12)(s+1) differs from b_x = s+1 even though the zero sets agree.

Overview

Differentiating a power lowers the exponent: $(d)/(dx)x^s+1 = (s+1)x^s$ . The Bernstein-Sato polynomial is the answer to the question of how far that phenomenon survives when $x$ is replaced by an arbitrary polynomial $p$ in several variables. One cannot expect a single derivative to do the job, and one cannot expect the scalar to stay as simple as $s+1$ . What is true, and far from obvious, is that some differential operator with polynomial coefficients takes $p^s+1$ to a scalar multiple of $p^s$ , the scalar being a polynomial in $s$ alone.

Write the statement out. There exist $0 ≠ b(s) ∈ K[s]$ and $D(s) ∈ A_n(K)[s]$ with $b(s),p^s = D(s) ⋅ p^s+1$ . The set of $b$ that occur is an ideal of $K[s]$ , and $K[s]$ is a principal ideal domain, so that ideal has a unique monic generator. That generator is the Bernstein-Sato polynomial $b_p(s)$ , also called the b-function of $p$ .

The existence of one non-zero $b$ is the whole content of the theorem, and it is proved by D-module theory rather than by calculation: the module generated by the formal symbol $p^s$ over the Weyl algebra with coefficients in $K(s)$ is holonomic, therefore of finite length, therefore a certain visible descending chain of submodules must stop, and the point where it stops is exactly the functional equation. That argument is carried out on the existence page.

What makes $b_p$ worth naming is that its roots are a genuine invariant of the singularities of $p$ . Bernstein introduced it to solve Gelfand's problem on the meromorphic continuation of P2; Kashiwara proved that all its roots are negative rational numbers; its largest root computes the log canonical threshold; and its integer roots tell you exactly which power $p^-N$ generates the localisation $K[X][p^-1]$ over the Weyl algebra. It is also one of the few deep invariants in this subject that a computer algebra system will actually compute for you.

Definition

Two definitions are needed: first a module in which the symbol $p^s$ lives, and then the ideal of which $b_p$ is the generator. Throughout, $K$ is a field of characteristic zero, $K[X] = K[x_1, … ,x_n]$ , and $p ∈ K[X]$ is non-zero.

The module of formal $s$ -th powers

Let $L$ be either $K(s)$ or $K[s]$ . Let $L[X,p^-1],p^s$ denote the free module of rank one over $L[X,p^-1]$ on a symbol written $p^s$ . Make it a left $A_n(L)$ -module by letting $x_i$ act by multiplication and setting

∂_i ⋅ (f,p^s) ;=; (∂_i f + s,(f,∂_i p)/(p)) p^s, f ∈ L[X,p^-1].

This is the Leibniz rule one would get if $p^s$ were an actual function, and a direct check confirms that the operators so defined satisfy $[∂_i, x_j] = δ_ij$ and commute among themselves, so the action is well defined. For an integer value $s = k$ the module specialises to the honest submodule $K[X,p^-1]p^k = K[X,p^-1]$ of the rational function field.

The Bernstein-Sato polynomialCoutinho (10.3.3)

Let $B_p ⊂ eq K[s]$ be the set of all $b(s)$ for which there exists an operator $D(s) ∈ A_n(K)[s]$ with

b(s),p^s ;=; D(s) ⋅ p,p^s

(10.1)

as an identity in $K(s)[X,p^-1]p^s$ . Then $B_p$ is an ideal of $K[s]$ : it is closed under sums, and multiplying (10.1) by any $c(s) ∈ K[s]$ replaces $D(s)$ by $c(s)D(s)$ , since $s$ is central. The existence theorem says $B_p ≠ 0$ . Its unique monic generator $b_p(s)$ is the Bernstein-Sato polynomial of $p$ .

Remark

Neither $b$ nor $D$ in (10.1) is unique; only the monic generator of the ideal is. Once $b_p$ is fixed, an operator $D$ realising it is still far from unique, since anything annihilating $p^s+1$ may be added to it.

Core Concepts

Why a formal symbol

For a general polynomial there is no function $p^s$ over an abstract field $K$ , and even over $ℂ$ the function $p(x)^s$ is multivalued. The construction sidesteps this: $p^s$ is declared to be a free generator, and the only thing that carries over from calculus is the differentiation rule. Everything that follows is algebra about a module, not analysis about a function. The analysis is recovered afterwards, by specialising $s$ to a complex number in a region where $p^s$ does define a function; that is the subject of the continuation page.

The functional equation as a lowering operator

Read (10.1) from right to left. It says that $p^s$ lies in the $A_n(K(s))$ -module generated by $p^s+1$ , since $b(s)$ is invertible in $K(s)$ . So the chain of submodules

A_n(K(s)),p^s ; ⊃ eq; A_n(K(s)),p,p^s ; ⊃ eq; A_n(K(s)),p^2p^s ; ⊃ eq; ⋅ s

is in fact constant. Conversely, if that chain is eventually constant then some functional equation holds. This equivalence is the bridge from finite length to the existence of $b_p$ , and it is why the b-function is a theorem in D-module theory rather than in calculus.

What the roots measure

Specialise $s$ to an integer $k$ in (10.1). If $b_p(k) ≠ 0$ then $p^k$ lies in $A_n ⋅ p^k+1$ , so nothing new is created when the exponent drops from $k+1$ to $k$ . The negative integers at which $b_p$ vanishes are therefore exactly the places where the descent can fail, and non-integer roots play the analogous role for the poles of complex-power integrals. Roots of $b_p$ mark the exponents where $p$ behaves badly; the largest one measures how badly.

Scale of the invariant

The b-function of $p$ is a global algebraic object: it sees every singular point of the hypersurface $p = 0$ at once, and also the behaviour at infinity to the extent that the coefficients of $D(s)$ are polynomial. Analysts usually want the local b-function at a point, defined with germs of analytic functions in place of $K[X]$ . The global one is divisible by every local one, and over $ℂ$ it is their least common multiple over the points of $V(p)$ .

Key Equations

The action defining the module, written on the generator itself:

∂_i ⋅ p^s ;=; s,(∂_i p)/(p),p^s ;=; s,(∂_i p),p^s-1.

(10.2)

Only the chain rule is postulated; everything else is forced.

The ideal and its generator:

B_p ;=; b_p(s),K[s], b_p monic.

(10.3)

Specialising $s$ to an integer $k$ turns the formal identity into a statement about honest rational functions:

b_p(k),p^k ;=; D(k) ⋅ p^k+1 in K[X,p^-1].

(10.4)

When the hypersurface $p = 0$ is smooth, so that $p$ and its partials generate the unit ideal, say $1 = a,p + ∑ _i a_i,∂_i p$ , an operator can be written down by hand:

( ∑ _i=1^n a_i,∂_i + (s+1),a) ⋅ p^s+1 ;=; (s+1),p^s.

(10.5)

And for the one-variable model case $p = x^m$ , the operator is a pure power of $∂$ :

1m^m,∂^m ⋅ x^m(s+1) ;=; ∏ _i=1^m(s + (i)/(m)) x^ms.

(10.6)

Variable Definitions

$K$: the ground field, of characteristic zero
$K[X]$: the polynomial ring $K[x_1, … ,x_n]$
$A_n(K)$: the $n$ -th Weyl algebra over $K$
$p$: a fixed non-zero polynomial in $K[X]$ , of degree $m$
$s$: an indeterminate, central in every ring in which it appears
$p^s$: the free generator of the rank-one module $K(s)[X,p^-1]p^s$
$D(s)$: an operator in $A_n(K)[s]$ realising a functional equation
$B_p$: the ideal of $K[s]$ of all admissible scalars $b(s)$
$b_p(s)$: the monic generator of $B_p$ , the Bernstein-Sato polynomial
$e(M)$: the multiplicity of a finitely generated module with respect to the Bernstein filtration

Properties and Behaviour

$s+1$ always divides $b_p$

If $p$ is not a unit of $K[X]$ , that is not a non-zero constant, then $b_p(-1) = 0$ .

Proof. Put $k = -1$ in (10.4). The right-hand side becomes $D(-1) ⋅ p^0 = D(-1) ⋅ 1$ , which is a polynomial. So $b_p(-1),p^-1 ∈ K[X]$ . If $b_p(-1)$ were non-zero this would make $p^-1$ a polynomial, forcing $p$ to be a unit. Hence $b_p(-1) = 0$ . $□$

The smooth case

If the ideal generated by $p, ∂_1 p, … , ∂_n p$ is the whole of $K[X]$ , then $b_p(s) = s+1$ .

Proof. Write $1 = a,p + ∑ _i a_i ∂_i p$ with $a, a_i ∈ K[X]$ . Applying $∑ _i a_i∂_i$ to $p^s+1$ and using (10.2) gives $(s+1) ( ∑ _i a_i ∂_i p )p^s = (s+1)(1 - a,p)p^s$ . Adding $(s+1)a,p^s+1 = (s+1)a,p,p^s$ recovers $(s+1)p^s$ , which is (10.5). So $s+1 ∈ B_p$ and $b_p ∣ s+1$ ; the previous proposition gives the reverse divisibility. $□$

Over an algebraically closed field the hypothesis says exactly that the hypersurface $p = 0$ is smooth and $p$ is reduced. The converse also holds, so $b_p(s) = s+1$ is a criterion for smoothness, but the converse is not proved by this argument.

Kashiwara's rationality theorem

Every root of $b_p(s)$ is a negative rational number. Moreover all roots lie in the interval $[-n, 0)$ , and the largest root is $-lct(p)$ , where $lct$ denotes the log canonical threshold.

Kashiwara proved rationality and negativity in 1976 by resolution of singularities; the identification of the largest root with the log canonical threshold and the lower bound $-n$ are later refinements. None of this is visible from the existence proof, which produces a $b$ with no control at all over its roots.

Products in disjoint variables

Let $p ∈ K[x_1, … ,x_n]$ and $q ∈ K[y_1, … ,y_r]$ . Then $b_pq ∣ b_p,b_q$ in $K[s]$ .

Proof. The operators $D_p(s)$ and $D_q(s)$ involve disjoint sets of variables, so they commute, and applying their product to $(pq)^s+1 = p^s+1q^s+1$ gives $b_p(s)b_q(s)(pq)^s$ . $□$ Equality is known to hold, but the reverse divisibility is not formal.

Integer roots and the localisation

Let $N$ be the largest positive integer with $b_p(-N) = 0$ (there is at least one, namely $N = 1$ ). Then the localisation is generated over the Weyl algebra by a single negative power:

K[X][p^-1] ;=; A_n ⋅ p^-N.

Proof. For $k > N$ we have $b_p(-k) ≠ 0$ , so (10.4) at $s = -k$ gives $p^-k ∈ A_n,p^-k+1$ , whence $A_n p^-k = A_n p^-k+1$ . Induction downwards from any $k$ shows $A_n p^-k = A_n p^-N$ for all $k ≥ N$ , and $K[X][p^-1]$ is the union of the $K[X]p^-k$ . $□$ This is the cleanest explanation of why the localisation is a finitely generated, indeed cyclic, module - a fact also forced by the general theorem that holonomic modules are cyclic.

Invariance

If $φ$ is an automorphism of $K^n$ given by polynomials with polynomial inverse, then $b_p ∘ φ = b_p$ ; and $b_p$ is unchanged by extension of the ground field within characteristic zero. Both follow because the corresponding automorphism of $A_n$ , respectively the flat base change, carries a functional equation to a functional equation.

Examples and Special Cases

A coordinate, and a power of one

For $p = x_1$ in $A_n$ , the hypersurface is smooth and $b_p(s) = s+1$ with $D = ∂_1$ . For $p = x^m$ in $A_1$ , equation (10.6) gives $b_p(s) = ∏ _i=1^m(s + i/m)$ , with $D(s) = m^-m∂^m$ independent of $s$ . The roots $-1/m, -2/m, … , -1$ are rational and negative, as Kashiwara's theorem requires, and the largest is $-1/m$ , which is indeed the log canonical threshold of $x^m$ .

Normal crossings

For $p = x_1x_2 ⋅ s x_n$ take $D = ∂_1∂_2 ⋅ s∂_n$ . Since the variables separate, $D ⋅ p^s+1 = ∏ _i ∂_i x_i^s+1 = (s+1)^np^s$ , so $b_p ∣ (s+1)^n$ ; in fact $b_p(s) = (s+1)^n$ . The multiplicity $n$ of the root $-1$ records that $n$ smooth branches meet.

A quadratic form

For $p = x_1^2 + ⋅ s + x_n^2$ and $Δ = ∂_1^2 + ⋅ s + ∂_n^2$ , differentiating twice gives $∂_i^2p^s+1 = 2(s+1)p^s + 4s(s+1)x_i^2p^s-1$ , and summing over $i$ collapses the second term because $∑ _i x_i^2 = p$ :

Δ ⋅ p^s+1 ;=; (2n(s+1) + 4s(s+1) )p^s ;=; 4(s+1)(s+(n)/(2))p^s.

Hence $b_p(s) = (s+1)(s + n/2)$ , the operator being $14Δ$ . For $n = 2$ this is $(s+1)^2$ , agreeing with the normal-crossings answer, as it must, since $x^2+y^2$ factors as $(x+iy)(x-iy)$ over an extension.

The cusp

For the cuspidal cubic $p = x^2 - y^3$ one has $b_p(s) = (s+1)(s+ 56)(s+ 76)$ . The roots are still negative rationals, but they are no longer of the shape suggested by the smooth or normal-crossings cases, and the largest, $-5/6$ , is the log canonical threshold of the cusp. This is the smallest example where the b-function is genuinely hard to guess.

The b-function is not an invariant of the zero set

Take $p = x$ and $q = x^2$ in $A_1$ . They have the same zero set, but $b_p(s) = s+1$ while $b_q(s) = (s+ 12)(s+1)$ . The b-function depends on the polynomial, including its multiplicities, not merely on the variety it cuts out.

Worked Example

Computing $b_p$ exactly for $p = x^3$ in $A_1$

Step 1 - produce a functional equation
Take $D = ∂^3$ and apply it to $x^3(s+1) = x^3s+3$ . Each differentiation drops the exponent by one and multiplies by it:

$∂^3,x^3s+3 ;=; (3s+3)(3s+2)(3s+1),x^3s.$

Divide by $27$ . Since $(3s+3)(3s+2)(3s+1) = 27(s+1)(s+ 23)(s+ 13)$ , the operator $D = 127∂^3$ realises the scalar $b(s) = (s+1)(s+ 23)(s+ 13)$ . Therefore $b_p$ divides this cubic.
Step 2 - grade the Weyl algebra by weight
Give $x$ weight $+1$ and $∂$ weight $-1$ . In the canonical form $∑ c_abx^a∂^b$ every monomial is weight-homogeneous of weight $a-b$ , so $A_1(K)[s] = ⨁ _w ∈ Z A_1[s]_w$ , and the decomposition of an operator into weight components keeps its coefficients in $K[s]$ .

The module $K(s)[x,x^-1]x^3s$ is graded compatibly: put $x^3s+k$ in degree $k$ . An operator of weight $w$ moves degree $k$ to degree $k+w$ .
Step 3 - only one weight can contribute
Suppose $b(s)x^3s = D(s) ⋅ x^3s+3$ with $D(s) ∈ A_1(K)[s]$ . The left side sits in degree $0$ and $x^3s+3$ in degree $3$ , so only the weight $-3$ component of $D$ contributes; discard the rest. A basis of that component is $x^a∂^a+3 : a ≥ 0$ , so write $D(s) = ∑ _a ≥ 0 c_a(s),x^a∂^a+3$ with $c_a ∈ K[s]$ , almost all zero.
Step 4 - evaluate each basis operator
Applying $x^a∂^a+3$ to $x^3s+3$ multiplies by the falling factorial with $a+3$ factors starting at $3s+3$ :

$x^a∂^a+3,x^3s+3 ;=; ∏ _i=0^a+2(3s+3-i);x^3s.$

Every one of these products, whatever $a ≥ 0$ is, begins with the three factors $(3s+3)(3s+2)(3s+1)$ . Hence each term of $b(s) = ∑ _a c_a(s) ∏ _i=0^a+2(3s+3-i)$ is divisible in $K[s]$ by $(3s+3)(3s+2)(3s+1)$ .
Step 5 - conclude
So every element of $B_p$ is a multiple of $(s+1)(s+ 23)(s+ 13)$ , and Step 1 shows that this cubic lies in $B_p$ . Being monic, it is the generator.

Sanity check on the integer roots: the only negative integer root is $-1$ , so the proposition on localisations predicts $K[x][x^-3] = A_1 ⋅ x^-1$ . That is visibly correct, since $∂^k-1x^-1 = (-1)^k-1(k-1)!,x^-k$ produces every negative power from $x^-1$ .

Result

$b_x^3(s) = (s+ 13)(s+ 23)(s+1)$ , realised by $D = 127∂^3$ . The same weight argument gives $b_x^m(s) = ∏ _i=1^m(s+i/m)$ for every $m ≥ 1$ .

Applications and Industry Use

In a mathematics topic, this section covers downstream use inside mathematics, computing and engineering rather than a manufactured product.

Meromorphic continuation. Bernstein's original application: the distribution-valued function $s ↦ ∈ t p^sφ$ extends meromorphically to the whole $s$ -plane, with poles confined to the roots of $b_p$ shifted by non-positive integers. See the continuation page.
Fundamental solutions. A constant-coefficient partial differential operator with symbol $P$ has a tempered fundamental solution; one route to this, due to Bernstein, is to divide by $P$ using the functional equation for $P$ regarded as a polynomial, which is exactly what the b-function makes legitimate.
Generating the localisation. The integer roots of $b_p$ give the explicit generator $p^-N$ of $K[X][p^-1]$ over $A_n$ , and hence an explicit presentation of a standard holonomic module.
Local cohomology and de Rham cohomology. Algorithms that compute the local cohomology of a hypersurface complement, or the de Rham cohomology of an affine variety, start by computing $b_p$ to bound the exponents that can occur.
Singularity theory. The roots of the local b-function are related to the eigenvalues of the monodromy of the Milnor fibration through $e^2π i α$ , a theorem of Malgrange; the b-function also determines the log canonical threshold and appears in the theory of multiplier ideals.
Representation theory. For a prehomogeneous vector space with relative invariant $p$ , the b-function is Sato's original object and enters the functional equation of the associated zeta function.

Computational Notes

Read this as the manufacturing section of the template: how the object is actually built by machine, at what cost, and where the computation stops being decidable.

Computing $b_p$ is a Gröbner basis calculation in a Weyl algebra with one extra variable. Oaku's algorithm (1997) computes the annihilator of $p^s$ in $A_n[s]$ by an elimination in $A_n+1$ using the graph embedding $x ↦ (x, p(x))$ , and then obtains $b_p$ as the monic generator of the ideal $(Ann_A_n[s]p^s + A_n[s],p ) ∩ K[s]$ .

Macaulay2, package Dmodules: globalBFunction, bFunction, and annFs for the annihilator of $p^s$ .
Singular, library dmod.lib: bfct, bfctAnn, annfs.
Risa/Asir has long had a fast implementation of the Oaku-Takayama algorithms; SageMath reaches the same functionality through its interfaces.
For a hypersurface with isolated singularities, specialised methods based on the Brieskorn lattice are usually much faster than the general elimination.

Cost is the practical obstacle. The elimination step involves Gröbner bases in a non-commutative ring in $2n+2$ generators with a weight order, and worst-case behaviour is doubly exponential in $n$ . Degree $4$ or $5$ polynomials in three or four variables are routine; general polynomials in six variables are not. Choosing the ground field to be $ℚ$ rather than a number field, and exploiting weight-homogeneity as in the worked example above, are the two cheapest sources of speed.

Limits of Validity

Characteristic zero is essential. The proof of existence runs through Bernstein's inequality and the simplicity of $A_n$ , both of which fail in characteristic $p$ . In characteristic $p$ there is no b-function in this sense; the analogue is the theory of $F$ -thresholds and Bernstein-Sato polynomials defined by Mustata and others, and they behave differently.
$p$ must be non-zero. For $p = 0$ the module $K(s)[X,p^-1]$ is not defined. For $p$ a non-zero constant, $b_p = 1$ .
The existence proof is not effective. It bounds the length of a chain of submodules by a multiplicity, which bounds how many steps the descent needs, but it does not bound $b_p$ in terms of $n$ and $deg p$ in any usable way. Effective bounds are a separate and much harder subject.
Global versus local. The polynomial defined here is the global algebraic b-function. Statements about monodromy eigenvalues or about the Milnor fibre are properly statements about the local analytic b-function at a point of $V(p)$ , and the two coincide only after taking a least common multiple over the singular points.
One polynomial only. For an ideal generated by several polynomials there is a b-function too, but its definition (Budur-Mustata-Saito) is not the naive generalisation of (10.1).

Failure Modes and Common Mistakes

The normalisation constant

For $p = x_1^2+ ⋅ s+x_n^2$ the Laplacian gives $Δ p^s+1 = 4(s+1)(s+n/2)p^s$ , and it is tempting to record $4(s+1)(s+n/2)$ as the answer - Coutinho's text does exactly this at (10.3), immediately after defining $b_p$ as the monic generator. The two statements are inconsistent: the monic generator is $(s+1)(s+n/2)$ , obtained with $D = 14Δ$ . Nothing mathematical is at stake, but a scalar factor must be divided out before the polynomial deserves the name.

Letting $D$ depend rationally on $s$

The definition requires $D(s) ∈ A_n(K)[s]$ , polynomial in $s$ . If $D$ is allowed to lie in $A_n(K(s))$ then the ideal $B_p$ becomes all of $K[s]$ and the definition collapses, because denominators in $s$ can absorb any scalar. The existence proof genuinely produces $D$ over $K(s)$ first, and the last step - clearing denominators - is what makes the resulting $b$ a polynomial in $s$ rather than a rational function.

Treating $p^s$ as a function

Over $ℝ$ or $ℂ$ the expression $p(x)^s$ is not single-valued and is not defined where $p$ is negative or zero. Every algebraic statement on this page is about the free generator of a module. Passing to functions requires choosing a branch, restricting to $p > 0$ , and controlling convergence, and it is precisely at that transition that the roots of $b_p$ turn into poles.

Assuming small degree

There is no reason for $b_p$ to be bounded by $deg p$ , and no reason for the roots to be simple. Normal crossings already give $(s+1)^n$ , a root of multiplicity $n$ from a polynomial of degree $n$ ; other examples have degrees much larger than one might guess from the equations. Guessing $b_p$ from a lucky operator only ever gives a multiple of the answer, and proving minimality - as in the worked example - is the harder half.

Sign conventions

Some sources write the functional equation with $p^-s$ , or state it as $b(s)p^s+1 = …$ , which flips the sign of the roots or shifts them by one. Before quoting a root of a b-function from the literature, check which of $b(s)$ , $b(-s)$ or $b(s-1)$ the author means; the standard convention here makes all roots negative.

Historical Notes

The polynomial has two independent origins. Mikio Sato, working on prehomogeneous vector spaces in the 1960s, attached to a relative invariant $p$ of a group action a polynomial $b(s)$ satisfying essentially (10.1), and used it to derive functional equations for the associated zeta functions. In that setting the operator $D(s)$ can often be written down from the group action, so existence was not the issue.

Joseph Bernstein came to it from the other side. Gelfand had asked, at the 1954 International Congress, whether $s ↦ ∈ t f^sφ$ continues meromorphically in $s$ ; Atiyah and Hironaka-Bernstein answered it in 1970 using resolution of singularities. In 1971-72 Bernstein gave an entirely algebraic proof by inventing the dimension theory of modules over $A_n$ - what this collection calls Bernstein's inequality and holonomy - and deducing the functional equation from finite length. The b-function became a theorem about modules rather than a device.

Kashiwara proved in 1976 that the roots are negative rational numbers, using Hironaka's resolution of singularities and the behaviour of b-functions under the direct image; his methods work on any complex manifold. Malgrange, in the same period, connected the roots to the eigenvalues of the Milnor monodromy. Explicit computations followed: Yano, Cassou-Noguès and many others computed b-functions for families of singularities by hand, and Oaku's 1997 algorithm made the computation mechanical.

Comparison

Four objects that all get called a b-function.
	Global algebraic $b_p$	Local analytic $b_p,a$	b-function of an ideal	Indicial polynomial
Coefficients allowed in $D$	$A_n(K)[s]$	germs of analytic operators at $a$	several $s_j$ , one per generator	$K[x,∂]$ , one variable
Depends on	$p$ globally	the germ of $p$ at $a$	the ideal, not the generators	an operator and a point
Roots	negative rationals	negative rationals	a set of hyperplanes	the exponents of the solutions
Relation	least common multiple of the local ones over $V(p)$	divides the global one	specialises to $b_p$ for a principal ideal	unrelated in general
Computable	yes, by elimination	yes, locally	harder	immediately

The last column is included because the phrase indicial equation is sometimes used loosely for b-functions in texts on ordinary differential equations. The two agree in spirit - both record which exponents can appear - but they are different constructions.

Key Takeaways

Key points

For every non-zero $p ∈ K[X]$ there are $0 ≠ b(s) ∈ K[s]$ and $D(s) ∈ A_n(K)[s]$ with $b(s)p^s = D(s) ⋅ p^s+1$ .
Such $b$ form an ideal of $K[s]$ ; its monic generator $b_p(s)$ is the Bernstein-Sato polynomial.
Existence is a D-module theorem: $A_n(K(s))p^s$ is holonomic, hence of finite length, hence a visible descending chain stops.
$s+1$ divides $b_p$ whenever $p$ is non-constant, and $b_p(s) = s+1$ exactly when the hypersurface is smooth and reduced.
Model computations: $b_x^m = ∏ _i=1^m(s+i/m)$ , $b_x_1 ⋅ s x_n = (s+1)^n$ , $b_ ∑ x_i^2 = (s+1)(s+n/2)$ .
Kashiwara: all roots are negative rationals; the largest is minus the log canonical threshold.
The largest integer root $-N$ gives $K[X][p^-1] = A_n ⋅ p^-N$ .

FAQs

b_p

ever equal to

1

Only when $p$ is a non-zero constant. Then $p^s+1 = p,p^s$ , so $D = p^-1$ works with $b = 1$ . For any non-constant $p$ the divisibility by $s+1$ forces $b_p ≥ 1$ .

Does

b_p

determine

p

Not remotely. Every smooth reduced hypersurface, of any degree in any number of variables, has $b_p = s+1$ . The b-function is a measure of how singular $p$ is, not a description of it.

Why is the answer a polynomial in

s

alone, rather than something involving

x

Because that is what is being demanded: the scalar in (10.1) is required to lie in $K[s]$ . The theorem is that this demand can be met at all. If the scalar were allowed to involve $x$ , the equation would be trivially solvable with $b = p$ and $D = 1$ .

How does finite length produce the equation?

The submodules $A_n(K(s))p^kp^s$ for $k = 0,1,2, …$ form a descending chain in a holonomic module. Finite length makes it stationary, so $p^kp^s ∈ A_n(K(s))p^k+1p^s$ for some $k$ , and an automorphism shifting $s$ by one moves this back to $k = 0$ . Clearing denominators in $s$ gives $b$ and $D$ . The details are on the existence page.

Are the operator

D

and the polynomial

b

unique?

$b_p$ is unique because it is the monic generator of an ideal. $D$ is never unique: any operator annihilating $p^s+1$ can be added to it, and the annihilator of $p^s+1$ is a large left ideal.

What is the relation to the log canonical threshold?

Over a field of characteristic zero the log canonical threshold $lct(p)$ equals the negative of the largest root of $b_p$ . For $x^m$ this reads $lct = 1/m$ ; for a smooth hypersurface it reads $lct = 1$ ; for the cusp $x^2-y^3$ it gives $5/6$ .

Can I compute

b_p

by finding some operator and reading off the scalar?

That gives an element of $B_p$ , hence a multiple of $b_p$ , which is often all one needs. Proving that a candidate is the generator requires a lower bound argument - a grading argument as in the worked example, a specialisation, or a computer algebra elimination.

Does the definition need

p

to be reduced or irreducible?

No. Any non-zero $p$ has a b-function. Non-reducedness shows up in the roots: $b_x^2 = (s+ 12)(s+1)$ differs from $b_x = s+1$ even though the zero sets agree.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 10 §3, Theorem (10.3.3) and the discussion following it.
I. N. Bernstein, The analytic continuation of generalized functions with respect to a parameter, Functional Analysis and its Applications 6 (1972), 273-285.
I. N. Bernstein, Modules over a ring of differential operators. Study of the fundamental solutions of equations with constant coefficients, Functional Analysis and its Applications 5 (1971), 89-101.
M. Sato and T. Shintani, On zeta functions associated with prehomogeneous vector spaces, Annals of Mathematics 100 (1974), 131-170.
M. Kashiwara, B-functions and holonomic systems. Rationality of roots of b-functions, Inventiones Mathematicae 38 (1976), 33-53.
B. Malgrange, Le polynôme de Bernstein d'une singularité isolée, in Fourier Integral Operators and Partial Differential Equations, Lecture Notes in Mathematics 459, Springer, 1975, 98-119.
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, North-Holland, 1979 - Ch. 6.
T. Oaku, An algorithm of computing b-functions, Duke Mathematical Journal 87 (1997), 115-132.
M. Saito, On microlocal b-function, Bulletin de la Société Mathématique de France 122 (1994), 163-184 - for the location of the roots in $[-n,0)$ .
Macaulay2 Dmodules package documentation, globalBFunction and annFs; Singular library dmod.lib, bfct.

AI Suggested Questions

Verify by direct differentiation that $Δ,(x^2+y^2+z^2)^s+1 = 4(s+1)(s+3/2)(x^2+y^2+z^2)^s$ .
Run the weight-grading argument of the worked example for $p = x_1x_2$ and show that $b_p(s) = (s+1)^2$ .
Show directly that the annihilator of $p^s$ in $A_n(K(s))$ is a left ideal and compute it for $p = x^2$ .
Given $b_p(s) = (s+1)(s+2)(s+ 12)$ , which power $p^-N$ generates $K[X][p^-1]$ ?
Explain why allowing $D(s) ∈ A_n(K(s))$ would make every $b$ admissible.
Compute the log canonical threshold of $x^a + y^b$ and compare it with the largest root of its b-function.
Sketch how Oaku's algorithm reduces the computation of $b_p$ to an elimination in $A_n+1$ .