The Associated Graded Algebra

Q: Why is ^B A_n commutative when A_n is not?

Because the failure of commutativity is small in the filtration. For a ∈ B_k and b ∈ B_l the commutator [a,b] lies in B_k+l-2 , so it is zero in the degree- (k+l) component B_k+l/B_k+l-1 . Noncommutativity is not cancelled; it is pushed two degrees down, where it reappears as the Poisson bracket.

Q: Is the symbol map a ring homomorphism?

No. Each σ_k is linear but only defined on B_k , and the principal symbol map is multiplicative but not additive. There is in fact no algebra map A_n → S_n at all sending generators to generators, because such a map would have to send the relation [∂_i,x_i]=1 to 0 = 1 .

Q: Does the isomorphism ^B A_n ≅ K[z_1, … ,z_2n] need characteristic zero?

No. The proof uses only that the monomials x^α∂^β form a K -basis and that [∂_i,x_j] = δ_ij , both of which hold over any field. In characteristic p the algebra A_n behaves very differently — it is not simple and is finite over its centre — but its Bernstein-associated graded algebra is the same polynomial ring.

Q: Do I get the same answer with the order filtration?

You get a polynomial ring in 2n variables again, but with a different grading: ^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n] where the x_i sit in degree 0 . As ungraded rings the two agree; as graded rings they do not, and only the Bernstein version has finite-dimensional homogeneous components, which is what the counting arguments of Chapter 9 need.

Q: If two filtered algebras have the same associated graded algebra, are they isomorphic?

No, and the standard counterexample is right here: A_n and the commutative polynomial ring in 2n variables have isomorphic associated graded algebras but are not isomorphic as rings — one is simple and noncommutative, the other is neither. The associated graded algebra is an invariant of the pair (algebra, filtration), and it is a lossy one.

Q: Which properties actually lift from R to R?

Being a domain, being left or right Noetherian, having finite global dimension, and various finiteness conditions on modules. Roughly, properties expressed by the existence or non-existence of certain elements or chains lift; properties expressed by the absence of structure, such as simplicity, do not.

Q: What is the principal symbol good for concretely?

It is the leading-term data of an operator. It detects the degree, it multiplies correctly, and it is the object a Gröbner basis computation manipulates. Geometrically, for the order filtration, the zero set of the principal symbol is the characteristic variety of a single equation, which controls where solutions can be singular.

Q: Why does the construction use Ω_k/Ω_k-1 rather than a complement of Ω_k-1 in Ω_k?

A complement exists — these are vector spaces — but no complement is canonical, and choosing one destroys the multiplicative structure: the product of two chosen complements need not land in the third. The quotient is canonical and (7.3) is well defined precisely because the ambiguity has been quotiented away.

Overview

The Weyl algebra is simple, so the standard tools of ring theory, which run on two-sided ideals, have nothing to grip. The way around this is to build a commutative ring out of $A_n$ that is easier to study and that retains enough information to be useful. That ring is the associated graded algebra.

The construction takes as input a filtration $Ω = Ω_i_i ≥ 0$ of a $K$ -algebra $R$ and returns a graded algebra $^Ω R$ whose degree- $i$ component is the quotient $Ω_i/Ω_i-1$ . Multiplication is defined on top-degree parts: the product of the class of $a$ and the class of $b$ is the class of $a b$ . The whole point is that the correction terms which make $R$ noncommutative are of strictly lower degree, so they are invisible in the quotient and the resulting ring can be commutative even when $R$ is not.

For $A_n$ with the Bernstein filtration the answer is as good as it could be: $^B A_n$ is a polynomial ring in $2n$ variables over $K$ . Everything the theory does with $A_n$ afterwards — proving it is Noetherian, defining the dimension and multiplicity of a module, attaching a characteristic variety to a module — happens by moving a question to this polynomial ring, solving it there with commutative algebra, and lifting the answer back.

The price is that information is lost. Two very different filtered algebras can have the same associated graded algebra: $A_n$ and the commutative polynomial ring in $2n$ variables are a case in point. So $gr$ is a shadow, not a photograph. Knowing which properties survive the passage to the shadow, and which do not, is the real content of this page.

Definition

Throughout, $K$ is a field, $R$ is a $K$ -algebra, and $Ω = Ω_i_i ≥ 0$ is a filtration of $R$ : an increasing chain $Ω_0 ⊂ eq Ω_1 ⊂ eq ⋅ s$ of $K$ -subspaces with $⋃ _i Ω_i = R$ and $Ω_i Ω_j ⊂ eq Ω_i+j$ . As always we set $Ω_i = 0$ for $i < 0$ .

Symbol mapCoutinho Ch. 7 §3

For each $k ≥ 0$ the symbol map of order $k$ is the canonical projection of $K$ -vector spaces

σ_k : Ω_k ⟶ Ω_k/Ω_k-1.

(7.1)

It is linear and surjective, and $σ_k(a) = 0$ precisely when $a ∈ Ω_k-1$ . If $a ≠ 0$ has degree exactly $k$ , meaning $a ∈ Ω_k ∖ Ω_k-1$ , then $σ_k(a)$ is called the principal symbol of $a$ and written $σ(a)$ .

The associated graded algebraCoutinho Ch. 7 §3

As a $K$ -vector space,

^Ω R = ⨁ _i ≥ 0 Ω_i/Ω_i-1.

(7.2)

Multiplication is defined on homogeneous elements and extended bilinearly. Every homogeneous element of degree $k$ is $σ_k(a)$ for some $a ∈ Ω_k$ , and the product is

σ_k(a),σ_l(b) = σ_k+l(ab).

(7.3)

With this multiplication $^Ω R$ is a graded $K$ -algebra whose degree- $i$ homogeneous component is $Ω_i/Ω_i-1$ . It is called the graded algebra of $R$ associated with the filtration $Ω$ .

Notation used on this page

For $A_n$ with the Bernstein filtration $B$ we write $S_n = ^B A_n$ and put $x_i = σ_1(x_i)$ and $ξ_i = σ_1(∂_i)$ . Coutinho calls these $y_1, … ,y_2n$ ; the barred and Greek names are used here because they survive unchanged into the geometry of characteristic varieties, where $ξ_i$ is the cotangent coordinate dual to $x_i$ .

Core Concepts

Only the top degree survives

Passing to $Ω_k/Ω_k-1$ throws away everything of degree below $k$ . So the symbol of an element is its leading part, in exactly the sense in which the leading term of a polynomial is its leading part. The definition (7.3) says that leading parts multiply, provided you keep track of the degrees you are working at.

Why the shadow is commutative

In $A_n$ the operators $x_i$ and $∂_i$ fail to commute, but only by a scalar: $∂_i x_i - x_i ∂_i = 1$ . Both $x_i$ and $∂_i$ have Bernstein degree $1$ , so the product $x_i∂_i$ has degree $2$ , while the discrepancy $1$ has degree $0$ . Two degrees below the top, the discrepancy is invisible. Hence $x_i ξ_i = ξ_i x_i$ in $S_n$ , and the same reasoning applied to arbitrary operators gives commutativity throughout.

This is not an accident of the generators. For the Bernstein filtration the general statement is that commutators drop degree by two:

[B_k, B_l] ⊂ eq B_k+l-2,

(7.4)

and any filtration satisfying $[Ω_k,Ω_l] ⊂ eq Ω_k+l-1$ already has commutative associated graded algebra. The order filtration satisfies the weaker inclusion with $k+l-1$ , which is exactly enough.

The lost degree is not thrown away, it is demoted

Commutativity of $S_n$ does not mean the commutator has vanished from the theory. It means the commutator has moved down two degrees, where it defines a new operation on $S_n$ : the Poisson bracket $σ_k(a),σ_l(b) = σ_k+l-2([a,b])$ . The symbol algebra is therefore not merely a commutative ring; it is a commutative ring carrying the fossil of the noncommutativity it replaced. Gabber's involutivity theorem, and with it the geometric half of D-module theory, is a statement about that fossil.

A deformation picture

One useful way to hold all this in mind: $A_n$ is a deformation of the commutative polynomial ring $K[ x,ξ]$ , with the Planck-constant-like parameter absorbed into the relation $[∂_i,x_i]=1$ . Taking $^B$ is the classical limit — it forgets the deformation and returns the underlying commutative algebra of functions on the phase space $K^2n$ . The Poisson bracket is what remains of the commutator at first order, precisely as in the quantum-mechanical origin of the algebra.

Construction and Proof

Step 0: the multiplication is well defined

Formula (7.3) defines a product in terms of representatives, so it has to be checked. Suppose $a, a' ∈ Ω_k$ have the same symbol, that is $a - a' ∈ Ω_k-1$ , and let $b ∈ Ω_l$ . Then $(a-a')b ∈ Ω_k-1Ω_l ⊂ eq Ω_k+l-1$ , so $σ_k+l(ab) = σ_k+l(a'b)$ . The same argument applies in the second variable. Bilinearity, associativity and the identity element $σ_0(1)$ are then inherited directly from $R$ .

A trap hiding in the notation

The level $k$ is part of the data. If $a$ has degree $3$ but you form $σ_5(a)$ , you get $0$ , and (7.3) then says $0 ⋅ σ_l(b) = σ_5+l(ab)$ , which is consistent only because $a b$ has degree at most $3+l < 5+l$ . The formula is correct at every level; it is informative only when both symbols are principal and the degrees add.

Commutators drop two degrees

Degree of a commutatorCoutinho (2.1.1)

For all $k,l ≥ 0$ , $[B_k,B_l] ⊂ eq B_k+l-2$ in $A_n$ .

Proof

Both sides are spanned by monomials, so it suffices to treat $a = x^α∂^β$ and $b = x^γ∂^δ$ with $|α|+|β| ≤ k$ and $|γ|+|δ| ≤ l$ . Use the derivation identities $[ab,c] = a[b,c] + [a,c]b$ and $[a,bc] = [a,b]c + b[a,c]$ to reduce to commutators of generators. The only non-zero case is $[∂_i, x_i] = 1$ , which removes one factor from each of the two monomials and contributes nothing in their place: the resulting term has degree at most $(k-1)+(l-1) = k+l-2$ . Induction on $k+l$ completes the argument.

Consequently, for $a ∈ B_k$ and $b ∈ B_l$ , $σ_k(a)σ_l(b) - σ_l(b)σ_k(a) = σ_k+l([a,b]) = 0$ , because $[a,b]$ lies in $B_k+l-2 ⊂ eq B_k+l-1$ . So $S_n$ is commutative.

The structure theorem

The symbol algebra of the Weyl algebraCoutinho (7.3.1)

Let $K$ be a field and $A_n = A_n(K)$ . Then $S_n = ^B A_n$ is isomorphic, as a graded $K$ -algebra, to the polynomial ring $K[z_1, … ,z_2n]$ with every $z_j$ of degree $1$ . An isomorphism is given by $z_i ↦ x_i$ and $z_n+i ↦ ξ_i$ for $i = 1, … ,n$ .

Proof, in three steps

Generation. A homogeneous element of $S_n$ of degree $k$ is $σ_k(d)$ for some $d ∈ B_k$ . Write $d$ in canonical form as $∑ c_αβx^α∂^β$ with $|α|+|β| ≤ k$ . Every term with $|α|+|β| < k$ lies in $B_k-1$ and is killed by $σ_k$ , so

σ_k(d) = ∑ _|α|+|β| = k c_αβ, x^α ξ^β ,

a homogeneous polynomial of degree $k$ in the $2n$ elements $x_i,ξ_i$ . Hence these generate $S_n$ as a $K$ -algebra.

Commutativity. Established above. Together with generation, this produces a surjective $K$ -algebra map $φ : K[z_1, … ,z_2n] → S_n$ with $φ(z_i) = x_i$ , $φ(z_n+i) = ξ_i$ . Since the $z_j$ and their images all have degree $1$ , $φ$ is a graded homomorphism.

Injectivity. Because $φ$ is graded it is enough to show that a homogeneous $G$ of degree $k$ with $φ(G)=0$ is zero. Write $G = ∑ _|α|+|β|=k c_αβ z^(α,β)$ and lift it to the operator $d = ∑ _|α|+|β|=k c_αβ x^α∂^β ∈ B_k$ , so that $σ_k(d) = φ(G) = 0$ . That means $d ∈ B_k-1$ , so $d$ is also a $K$ -combination of monomials $x^γ∂^δ$ with $|γ|+|δ| ≤ k-1$ . But the monomials $x^α∂^β$ are a $K$ -basis of $A_n$ , and the two expressions for $d$ involve disjoint sets of basis monomials. Comparing coefficients forces every $c_αβ$ to vanish, so $G = 0$ .

Where the hypotheses are used

The proof uses only that the monomials $x^α∂^β$ form a basis — the canonical basis theorem — and the commutation relation. Neither requires characteristic zero, so (7.5) holds over any field, including characteristic $p$ . This is worth noting because most theorems in this collection do need characteristic zero; the graded structure is not one of them.

Key Equations

The theorem that makes the construction worth doing:

S_n = ^B A_n ; ≅ ; K[ x_1, … , x_n,ξ_1, … ,ξ_n],

(7.5)

a commutative polynomial ring in $2n$ variables, graded so that every $x_i$ and every $ξ_i$ has degree $1$ .

The dimension count that must match on both sides comes from the monomial basis $x^α∂^β$ with $|α| + |β| ≤ k$ of $B_k$ :

_K B_k = 2n+k2n, _K (B_k/B_k-1 ) = 2n+k-12n-1.

(7.6)

The right-hand quantity is exactly the number of monomials of total degree $k$ in $2n$ variables, which is the first evidence for (7.5).

For the order filtration $F$ , where $x_i$ has degree $0$ and $∂_i$ degree $1$ , the same construction gives a different graded ring:

^F A_n ; ≅ ; K[x_1, … ,x_n][ξ_1, … ,ξ_n],

(7.7)

again a polynomial ring in $2n$ variables, but graded with $x_i = 0$ , so its homogeneous components are infinite dimensional over $K$ .

Finally, the operation that survives the loss of noncommutativity:

σ_k(a), σ_l(b) ;=; σ_k+l-2 ([a,b] ) ;=; ∑ _i=1^n ( (∂ f)/(∂ ξ_i) ∂ g∂ x_i - ∂ f∂ x_i(∂ g)/(∂ ξ_i) ),

(7.8)

where $f = σ_k(a)$ and $g = σ_l(b)$ .

Variable Definitions

$K$: the ground field, of characteristic zero unless stated otherwise
$R$: an arbitrary $K$ -algebra carrying a filtration
$Ω = Ω_i$: a generic filtration of $R$ , increasing, exhaustive, with $Ω_iΩ_j ⊂ eq Ω_i+j$
$σ_k$: the symbol map of order $k$ , the projection $Ω_k → Ω_k/Ω_k-1$
$σ(a)$: the principal symbol of $a ≠ 0$ , that is $σ_k(a)$ where $k$ is the degree of $a$
$^Ω R$: the associated graded algebra of $R$ for the filtration $Ω$
$A_n$: the $n$ -th Weyl algebra over $K$ , with generators $x_1, … ,x_n,∂_1, … ,∂_n$
$B = B_k$: the Bernstein filtration: $B_k$ is spanned by the $x^α∂^β$ with $|α|+|β| ≤ k$
$F = F_k$: the order filtration: $F_k$ is the space of operators of order at most $k$ in the derivations
$S_n$: the symbol algebra $^B A_n$
$x_i, ξ_i$: the degree-one generators $σ_1(x_i)$ and $σ_1(∂_i)$ of $S_n$
$, ⋅ ,,, ⋅ ,$: the Poisson bracket on $S_n$ induced by the commutator of $A_n$

Properties and Behaviour

The value of $gr$ lies in the properties that travel back up from the graded ring to the filtered one. Here are the ones the rest of the theory uses.

Domains liftCoutinho Ch. 7, Ex. 6.6

Let $R$ be a filtered $K$ -algebra with $Ω_i = 0$ for $i<0$ . If $^Ω R$ is an integral domain, then so is $R$ , and $(ab) = a + b$ for all non-zero $a, b$ .

Proof. Let $a,b ≠ 0$ have degrees $k, l$ . Their principal symbols are non-zero by definition, so $σ_k(a)σ_l(b) ≠ 0$ in the domain $^Ω R$ . By (7.3) this element is $σ_k+l(ab)$ , so $ab ∉ Ω_k+l-1$ ; in particular $ab ≠ 0$ and its degree is $k+l$ . $□$

Noetherianity liftsCoutinho (8.3.1)

Let $R$ be filtered with $Ω_i = 0$ for $i<0$ and $⋃ _i Ω_i = R$ . If $^Ω R$ is left Noetherian, then $R$ is left Noetherian. Applied to (7.5) and the Hilbert basis theorem, this gives at once that P6 is left and right Noetherian. The argument is on the lifting page.

A graded algebra is its own shadow

If $G = ⨁ _i G_i$ is graded and filtered by $Ω_k = ⨁ _i ≤ k G_i$ , then $^Ω G ≅ G$ as graded algebras. So the construction does nothing to an algebra that was already graded — it is a genuine generalisation of the graded case, not a competitor to it.

Symbols of both standard filtrationsCoutinho (7.3.1) and Ch. 7, Ex. 6.5

$^B A_n ≅ K[ x_1, … , x_n,ξ_1, … ,ξ_n]$ with all generators in degree $1$ ; $^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n]$ with $x_i = 0$ and $ξ_i = 1$ . Both are polynomial rings in $2n$ variables; they differ as graded rings, and only the Bernstein version has finite-dimensional homogeneous components.

Consequences drawn elsewhere in this collection

$A_n$ is a domain, and Bernstein degrees add across products.
$A_n$ is Noetherian on both sides, so submodules of finitely generated modules are finitely generated.
Finitely generated modules acquire a Hilbert polynomial, hence a dimension and a multiplicity.
A module acquires a characteristic ideal in $S_n$ and therefore a subvariety of $K^2n$ .
The Poisson bracket (7.8) makes $S_n$ a Poisson algebra, which is where involutivity of characteristic varieties is stated.

Examples and Special Cases

A filtration whose graded algebra is not commutative

Let $Q$ be the quantum plane, the $K$ -algebra on generators $u, v$ with $uv = λ vu$ for a fixed $λ ∈ K$ with $λ ≠ 1$ . It is already graded by total degree, so by the proposition above $Q ≅ Q$ , which is noncommutative. The commutator $[u,v] = (λ - 1)vu$ has degree $2$ , not $0$ : nothing drops, so nothing becomes commutative. Commutativity of $S_n$ is a property of the Weyl relations, not of the construction.

A commutative ring whose shadow has nilpotents

Filter $R = K[t]$ by $Ω_k = f : f ≤ 2k$ . This is a legitimate filtration, but $t$ and $t^2$ both have $Ω$ -degree $1$ . Writing $s_1 = σ_1(t)$ and $s_2 = σ_1(t^2)$ , we get $s_1^2 = σ_2(t^2) = 0$ , because $t^2$ already lies in $Ω_1$ . Counting dimensions ( $_K Ω_k/Ω_k-1 = 2$ for $k ≥ 1$ ) identifies the shadow as

^Ω K[t] ; ≅ ; K[s_1,s_2]/(s_1^2).

So a filtration of a polynomial ring can produce a non-reduced shadow, and degrees then fail to add: $t + t = 2$ but $t^2 = 1$ . A badly chosen filtration is worse than none.

Symbols of specific operators in $A_2$

In $A_2$ , with Bernstein degrees in brackets: $∂_1^3 x_2 + x_1^4$ has degree $4$ from both terms, hence principal symbol $ξ_1^3 x_2 + x_1^4$ . The operator $x_1∂_1 + x_2∂_2 + 7$ , the Euler operator plus a constant, has degree $2$ and principal symbol $x_1ξ_1 + x_2ξ_2$ — the constant is invisible. The Laplacian $∂_1^2 + ∂_2^2$ has principal symbol $ξ_1^2 + ξ_2^2$ , which is the classical symbol from the theory of partial differential equations.

Worked Example

Computing in $S_1 = ^B A_1$

Step 1 — check the dimensions agree
For $n=1$ , $B_k$ has basis $x^a∂^b : a+b ≤ k$ , so $_K B_k = k+22 = 12(k+1)(k+2)$ , giving $1, 3, 6, 10$ for $k = 0,1,2,3$ . The graded pieces therefore have dimensions

$_K(B_k/B_k-1) = 1,,2,,3,,4, … (k = 0,1,2,3, … ),$

which is exactly the number of monomials $x^aξ^b$ with $a+b=k$ . The Hilbert functions of $S_1$ and of $K[ x,ξ]$ match, as (7.5) requires.
Step 2 — two different operators, one symbol
In $A_1$ we have $∂ x = x∂ + 1$ , so $∂ x ≠ x ∂$ . Both have Bernstein degree $2$ , and their difference is $1 ∈ B_0 ⊂ eq B_1$ . Hence

$σ_2(∂ x) = σ_2(x∂) = x,ξ .$

The symbol cannot tell the two operators apart. This single computation is the whole reason $S_1$ is commutative, and also the whole reason the passage to $S_1$ loses information.
Step 3 — verify multiplicativity on a concrete product
Take $d = ∂^2 - x$ and $e = x∂$ . Their degrees are $2$ and $2$ , with $σ_2(d) = ξ^2$ (the term $-x$ has degree $1$ and is invisible) and $σ_2(e) = xξ$ . Now compute $d e$ in canonical form. From $∂ x = x∂ + 1$ we get $∂^2 x = ∂(x∂ + 1) = (x∂+1)∂ + ∂ = x∂^2 + 2∂$ , so

$de = (∂^2 - x)(x∂) = (∂^2 x)∂ - x^2∂ = x∂^3 + 2∂^2 - x^2∂ .$

The three terms have degrees $4$ , $2$ and $3$ , so $(de) = 4$ and $σ_4(de) = xξ^3$ . Compare with $σ_2(d)σ_2(e) = ξ^2 ⋅ xξ = xξ^3$ . They agree, as (7.3) promises.
Step 4 — degrees add, and $A_1$ is a domain
In Step 3 the degrees added: $4 = 2+2$ . That is not a coincidence. Since $S_1 ≅ K[ x,ξ]$ is an integral domain, $σ_k(d)σ_l(e) = σ_k+l(de) ≠ 0$ whenever $d$ and $e$ are non-zero of degrees $k$ and $l$ ; so $de ∉ B_k+l-1$ , and in particular $de ≠ 0$ . This is the cleanest proof that P9 is a domain, and it also shows $(de) = d + e$ .
Step 5 — the same ring, a different shadow
Filter $A_1$ instead by order, so $x$ has degree $0$ and $∂$ degree $1$ . Then $d = ∂^2 - x$ has order $2$ with $F$ -symbol $ξ^2$ , while $e = x∂$ has order $1$ with $F$ -symbol $x ξ$ , and $^F A_1 ≅ K[x][ξ] = K[x,ξ]$ . The underlying ring is a polynomial ring again, but the grading is different: $^F$ has $F_0 = K[x]$ in degree $0$ , an infinite-dimensional space over $K$ . Same algebra $A_1$ , same construction, two genuinely different graded outputs.

Result

$S_1 = ^B A_1 ≅ K[ x,ξ]$ with graded dimensions $1,2,3,4, …$ ; $σ_2(x∂) = σ_2(∂ x) = xξ$ ; and $σ ((∂^2-x)(x∂) ) = xξ^3 = σ(∂^2-x),σ(x∂)$ . Because $K[ x,ξ]$ is a domain, Bernstein degrees add across products in $A_1$ .

Applications and Industry Use

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

Structure theory of $A_n$ . The two headline facts, that $A_n$ is a Noetherian domain, are both immediate from (7.5) once the lifting lemmas are in place. Direct proofs exist but are longer and less reusable.
Dimension theory. The Hilbert polynomial of a filtered module is computed in $S_n$ , where the graded module is finitely generated over a polynomial ring and classical commutative algebra applies. Dimension, multiplicity and Bernstein's inequality all live downstream of this.
Microlocal analysis and PDE. The principal symbol of a differential operator is exactly the classical symbol used to define characteristics, ellipticity and propagation of singularities. Ellipticity of $∂_1^2 + … + ∂_n^2$ is the statement that its symbol $ξ_1^2+ … +ξ_n^2$ vanishes only at $ξ = 0$ .
Computer algebra. Gröbner basis algorithms in the Weyl algebra work by tracking leading terms, which are symbols for a filtration refined to a monomial order. The correctness of Buchberger-style algorithms in this noncommutative setting rests on the commutativity of the associated graded ring.
Deformation theory and quantisation. $A_n$ is the standard first example of a deformation quantisation of a Poisson algebra, and (7.8) is the classical limit of its commutator. The same $gr$ construction defines the classical limit of a general filtered quantisation.

Design Considerations

For a mathematical object, design considerations are the modelling choices: which ring, which filtration, which category to work in.

Choosing a filtration

A filtration is a modelling decision. It should be coarse enough that the shadow is simple, fine enough that the shadow is informative, and adapted to the question. Assigning degree $1$ to all $2n$ generators gives finite-dimensional pieces and makes counting possible; assigning degree $0$ to the $x_i$ gives a shadow with a geometric meaning on any smooth variety. Weighted variants, where $x_i$ has weight $w_i$ and $∂_i$ weight $v_i$ with $w_i + v_i ≥ 2$ , are also used, most visibly in the theory of $b$ -functions, where the weights are tuned to a specific singularity.

Choosing what to carry

If the goal is a numerical invariant, carry the Hilbert function of the graded object. If the goal is geometric, carry the support of the graded module in $Spec S_n$ . If the goal is an algorithm, carry the ideal of leading terms, since that is what a Gröbner basis computes. These three are different amounts of information about the same shadow, in increasing order of cost.

When not to pass to the shadow

Questions about two-sided ideals, simplicity, automorphisms or finite-dimensional representations should not be routed through $gr$ : those are exactly the properties destroyed by the passage. The Dixmier conjecture, for instance, is a statement about $A_n$ that has no useful graded reflection, which is part of why it remains open.

Limits of Validity

Three limits on how far the shadow can be trusted.

$gr$ does not determine $R$ . The commutative polynomial ring $P = K[z_1, … ,z_2n]$ with its own degree filtration has $P ≅ P ≅ ^B A_n$ . Yet $P$ is commutative and $A_n$ is simple with no finite-dimensional representations. Any argument of the form 'the graded rings agree, therefore the rings agree' is invalid.
Properties travel upwards, not downwards. 'If $gr R$ has property $𝒫$ then $R$ has it' is the useful direction and holds for being a domain, being Noetherian, and having finite global dimension. The converse routinely fails: $A_n$ is simple, $S_n$ is very far from simple.
The filtration must be exhaustive and bounded below. If $⋃ _i Ω_i ⊂ neq R$ , elements outside the union have no degree and no symbol, and the lifting arguments break down at the first step. Filtrations indexed by $ℤ$ need the separatedness condition $⋂ _i Ω_i = 0$ in its place.

What is genuinely lost

Everything below the top degree. The operators $∂$ and $∂ - 1$ in $A_1$ have the same principal symbol $ξ$ , so the modules $A_1/A_1∂ ≅ K[x]$ and $A_1/A_1(∂ - 1) ≅ K[x]e^x$ have the same associated graded module $S_1/(ξ)$ , hence identical dimension and multiplicity. They are nevertheless not isomorphic: $K[x]$ contains a non-zero element killed by $∂$ , while $fe^x$ is killed by $∂$ only if $f' + f = 0$ , which forces $f = 0$ for polynomial $f$ . Invariants coming from $gr$ are coarse by design.

Failure Modes and Common Mistakes

Treating $σ$ as a ring homomorphism

The principal symbol map $σ : A_n ∖ 0 → S_n$ is multiplicative, but it is emphatically not additive. In $A_1$ , $σ(∂^2) = ξ^2$ and $σ(-∂^2 + x) = -ξ^2$ , yet the sum is $x$ with $σ(x) = x$ , which is not $ξ^2 + (-ξ^2) = 0$ . Symbols of sums can jump down in degree without warning. Only $σ_k$ , at a fixed level $k$ , is linear.

Confusing $x_i$ with $x_i$

There is no ring homomorphism $A_n → S_n$ sending $x_i ↦ x_i$ and $∂_i ↦ ξ_i$ : such a map would force $1 = σ([∂_i,x_i])$ to be $ξ_i x_i - x_iξ_i = 0$ . The elements $x_i$ live in a different ring and only remember $x_i$ modulo $B_0 = K$ . Writing $x_i$ for $x_i$ is a common abuse of notation and a common source of wrong proofs.

Assuming commutativity of the shadow comes for free

$^Ω R$ is commutative if and only if $[Ω_k,Ω_l] ⊂ eq Ω_k+l-1$ for all $k, l$ . That inclusion is a real hypothesis about the filtration, and it fails for the degree filtration of the quantum plane, of a free algebra, or of any noncommutative graded ring. Check it before you use it.

Reading $σ_k(a) = 0$ as $a = 0$

$σ_k(a) = 0$ says only that $a$ lies in $Ω_k-1$ . It is entirely normal for a non-zero operator to have zero symbol at a level above its degree. When an argument needs $σ_k(a) ≠ 0$ , it must first establish that $a$ has degree exactly $k$ .

Expecting degrees to add in a general filtered ring

$(ab) = a + b$ requires the associated graded ring to have no zero divisors in the relevant degrees. For $A_n$ this holds because $S_n$ is a polynomial ring. In a filtered ring whose shadow has nilpotents — say $K[t]/(t^2)$ filtered by degree — two non-zero elements can multiply to something of strictly smaller degree, or to zero.

Historical Notes

The idea of replacing a filtered object by its associated graded object is older than D-modules. It appears in commutative algebra as the Rees construction and the associated graded ring of an ideal, systematised by Krull, Zariski and Samuel in the 1940s and 1950s as the technical basis for local intersection multiplicity and for the theory of dimension in local rings.

The symbol of a differential operator has a separate and older lineage, in the study of characteristics of partial differential equations. The two merged in the 1960s, when Sato's algebraic analysis treated the symbol as an honest element of a graded ring of functions on the cotangent bundle.

Bernstein's 1971 and 1972 papers introduced the filtration that now carries his name, precisely because assigning degree $1$ to the $x_i$ makes each $B_k$ finite dimensional and so makes the counting argument for his inequality possible. Björk's 1979 book placed the whole apparatus of filtered rings and their graded shadows on a systematic footing, and it is that treatment which Coutinho's primer condenses in this chapter.

Comparison

The two standard filtrations of $A_n$ and the graded algebras they produce.
	Bernstein filtration $B$	Order filtration $F$
$x_i$	$1$	$0$
$∂_i$	$1$	$1$
Associated graded	$K[ x_1, … , x_n,ξ_1, … ,ξ_n]$	$K[x_1, … ,x_n][ξ_1, … ,ξ_n]$
Grading of the generators	all in degree $1$	$x_i$ in degree $0$ , $ξ_i$ in degree $1$
$_K$ of each piece	finite, $2n+k-12n-1$	infinite for every $k$
Commutator drop	two degrees	one degree
Defined for $D(X)$ , $X$ affine	no	yes
Used for	Hilbert polynomials, dimension, multiplicity	characteristic varieties, geometry, microlocal analysis

The practical rule: use $B$ when you need to count, because its pieces are finite dimensional over $K$ ; use $F$ when you need geometry, because its symbols are functions on the cotangent bundle of affine $n$ -space and generalise to other smooth varieties. The dimension of a module comes out the same either way, but that is a theorem, not a definition.

Key Takeaways

Key points

A filtration $Ω$ of $R$ yields $^Ω R = ⨁ _i Ω_i/Ω_i-1$ , with multiplication $σ_k(a)σ_l(b) = σ_k+l(ab)$ .
The product is well defined because lowering a representative by one degree lowers the product by one degree.
$^Ω R$ is commutative as soon as $[Ω_k,Ω_l] ⊂ eq Ω_k+l-1$ ; for the Bernstein filtration commutators actually drop two degrees.
$^B A_n ≅ K[ x_1, … , x_n,ξ_1, … ,ξ_n]$ , a polynomial ring in $2n$ variables with all generators in degree $1$ . The proof needs only the canonical basis and the commutation relations, so it holds in any characteristic.
Being a domain and being Noetherian lift from the graded ring to the filtered ring; this is how both facts are proved for $A_n$ .
The shadow does not determine the original: $A_n$ and the commutative polynomial ring have isomorphic associated graded algebras.
The commutator survives as a Poisson bracket of degree $-2$ on $S_n$ , which is where the geometry of characteristic varieties begins.

FAQs

Why is

^B A_n

commutative when

A_n

is not?

Because the failure of commutativity is small in the filtration. For $a ∈ B_k$ and $b ∈ B_l$ the commutator $[a,b]$ lies in $B_k+l-2$ , so it is zero in the degree- $(k+l)$ component $B_k+l/B_k+l-1$ . Noncommutativity is not cancelled; it is pushed two degrees down, where it reappears as the Poisson bracket.

Is the symbol map a ring homomorphism?

No. Each $σ_k$ is linear but only defined on $B_k$ , and the principal symbol map is multiplicative but not additive. There is in fact no algebra map $A_n → S_n$ at all sending generators to generators, because such a map would have to send the relation $[∂_i,x_i]=1$ to $0 = 1$ .

Does the isomorphism

^B A_n ≅ K[z_1, … ,z_2n]

need characteristic zero?

No. The proof uses only that the monomials $x^α∂^β$ form a $K$ -basis and that $[∂_i,x_j] = δ_ij$ , both of which hold over any field. In characteristic $p$ the algebra $A_n$ behaves very differently — it is not simple and is finite over its centre — but its Bernstein-associated graded algebra is the same polynomial ring.

Do I get the same answer with the order filtration?

You get a polynomial ring in $2n$ variables again, but with a different grading: $^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n]$ where the $x_i$ sit in degree $0$ . As ungraded rings the two agree; as graded rings they do not, and only the Bernstein version has finite-dimensional homogeneous components, which is what the counting arguments of Chapter 9 need.

If two filtered algebras have the same associated graded algebra, are they isomorphic?

No, and the standard counterexample is right here: $A_n$ and the commutative polynomial ring in $2n$ variables have isomorphic associated graded algebras but are not isomorphic as rings — one is simple and noncommutative, the other is neither. The associated graded algebra is an invariant of the pair (algebra, filtration), and it is a lossy one.

Which properties actually lift from

gr R

R

Being a domain, being left or right Noetherian, having finite global dimension, and various finiteness conditions on modules. Roughly, properties expressed by the existence or non-existence of certain elements or chains lift; properties expressed by the absence of structure, such as simplicity, do not.

What is the principal symbol good for concretely?