The Weyl Algebra Is Noetherian

Q: Does the result need characteristic zero?

No. Every ingredient - the canonical basis of A_n , the identification gr^B A_n ≅ K[y_1, … ,y_2n] , Hilbert's basis theorem, and the lifting argument - is valid over any field. This is a useful contrast with simplicity of A_n , which genuinely requires char K = 0 .

Q: Why is a filtration needed at all - why not imitate Hilbert's proof directly in A_n?

Hilbert's proof compares leading coefficients in a commutative ring of coefficients. In A_n there is no distinguished commutative coefficient ring in which the leading data lives, because the variables do not commute past each other. The filtration manufactures one: S_n plays the role of the coefficient ring, and Bernstein degree plays the role of degree in z .

Q: Is the converse of the lifting theorem true?

Not for an arbitrary filtration. If gr^Γ M is finitely generated then M is finitely generated, but a finitely generated M can carry filtrations whose graded module is not finitely generated. The filtrations for which the correspondence is exact are the good filtrations.

Q: Does noetherian imply that every left ideal of A_1 is principal?

No. Noetherian only says finitely generated. A_1 has proper left ideals that are not principal; what is true, by Stafford's theorem, is that two elements always suffice.

Q: How does noetherianity interact with short exact sequences?

Given 0 → N → M → M/N → 0 , the module M is noetherian if and only if both N and M/N are. That equivalence is what makes the property propagate through constructions; it is treated on the exact sequences page .

Q: Does the same proof work with the order filtration?

Yes, with one adjustment. The pieces F_m of the order filtration are infinite dimensional over K but finitely generated as K[x_1, … ,x_n] -modules, and gr^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n] . The final step selects a finite generating set over K[x_1, … ,x_n] rather than a K -basis, using Hilbert's theorem a second time.

Q: What does noetherianity buy that finite generation alone does not?

Stability under passing to submodules. A finitely generated module can have a badly infinite submodule; over a noetherian ring it cannot. Every argument that constructs a submodule - kernels, annihilators, intersections, torsion - relies on this.

Q: Is A_n noetherian over a coefficient ring rather than a field?

Yes, if the coefficient ring R is commutative and noetherian. Then gr^B A_n(R) ≅ R[y_1, … ,y_2n] , which is noetherian by Hilbert's theorem, and the lifting argument goes through once finite dimensionality over K is replaced by finite generation over R .

Overview

A ring is left noetherian when every one of its left ideals can be generated by finitely many elements. For commutative polynomial rings this is Hilbert's basis theorem, proved in 1890, and it is the technical foundation on which algebraic geometry rests. This page settles the same question for the Weyl algebra $A_n$ : every left ideal of $A_n$ is finitely generated, and so is every right ideal.

Nothing about $A_n$ makes this obvious. It is not commutative, it has no division algorithm in any naive sense, and it is simple, so the usual commutative technique of quotienting by a prime ideal and inducting is unavailable. Noncommutative rings that fail to be noetherian are easy to write down - the free algebra $K ⟨ x, y ⟩$ is one - so the result has to be earned.

The proof is a transfer argument. The Bernstein filtration $B_m$ approximates $A_n$ by a commutative object: the associated graded ring $S_n = gr^B A_n$ is a polynomial ring in $2n$ variables, which Hilbert's theorem says is noetherian. A general lifting theorem, treated in full on its own page, then says that a filtered $A_n$ -module whose associated graded module is noetherian is itself noetherian. Applying it to $A_n$ filtered by $B$ gives the result in one line.

The payoff is immediate and used everywhere afterwards. Finitely generated $A_n$ -modules are noetherian, so submodules of finitely generated modules are finitely generated, syzygies terminate, presentations exist, and - the point of the next section of the source - good filtrations behave well enough to define a dimension theory.

Definition

Throughout, $K$ is a field and $A_n = A_n(K)$ is the $n$ -th Weyl algebra, with generators $x_1, … , x_n, ∂_1, … , ∂_n$ subject to $[∂_i, x_j] = δ_ij$ and all other pairs commuting. No hypothesis on $char K$ is needed anywhere on this page.

Left noetherian ring

A ring $R$ is left noetherian if $R$ , regarded as a left module over itself, is a noetherian module. Unwinding the definition of a noetherian module, this says exactly that every left ideal of $R$ is finitely generated, and equivalently that every ascending chain $I_1 ⊂ eq I_2 ⊂ eq ⋅ s$ of left ideals is eventually constant. Right noetherian is the same condition for right ideals; a ring satisfying both is called noetherian.

The Weyl algebra is noetherianCoutinho (8.2.4)

For every field $K$ and every $n ≥ 0$ , the Weyl algebra $A_n(K)$ is both left and right noetherian. Consequently every finitely generated left or right $A_n$ -module is a noetherian module.

Note

The two halves are not independent. The transposition anti-automorphism $τ$ of $A_n$ , defined by $τ(x_i) = x_i$ and $τ(∂_i) = -∂_i$ and reversing products, carries left ideals to right ideals and preserves Bernstein degree. Left noetherian therefore forces right noetherian here, which is why the source proves only one side and then says the other follows with the obvious changes.

Core Concepts

Finite generation does not pass to submodules for free

A quotient of a finitely generated module is finitely generated, but a submodule need not be. The standard warning example is the polynomial ring $R = K[x_1, x_2, x_3, … ]$ in countably many variables. As a module over itself $R$ is cyclic - generated by $1$ - yet the ideal $(x_1, x_2, … )$ is not finitely generated, because any finite list of polynomials mentions only finitely many variables. Noetherian is precisely the condition that rules this out.

The graded ring as a commutative shadow

The Bernstein filtration replaces the relation $∂_i x_i = x_i ∂_i + 1$ by the observation that the correction term $1$ has degree $0$ , two units below the product's degree $2$ . In the associated graded ring the correction is invisible, so $S_n = gr^B A_n$ is commutative, and in fact a polynomial ring $K[y_1, … , y_2n]$ with $y_i = σ_1(x_i)$ and $y_n+i = σ_1(∂_i)$ . Every question about $A_n$ that only sees leading terms can be asked in $S_n$ instead, where Hilbert's theorem applies.

Leading terms determine generators

The transfer works because of one structural fact: if the leading terms of a submodule are all accounted for by finitely many elements, then those elements generate. Given $v$ in the submodule, subtract a combination of the chosen elements matching its leading term; what is left has strictly smaller degree, and induction on degree finishes. This is the same idea as division against a Gröbner basis, and it is why the argument is constructive in practice.

Where the finite dimensionality is used

One extra ingredient is needed that has no commutative analogue in Hilbert's proof: each $B_m$ is a finite dimensional $K$ -vector space, of dimension $2n+m2n$ . So once the argument has reduced the problem to "generated by the elements of degree at most $m$ ", a finite generating set is produced simply by taking a $K$ -basis of a finite dimensional space. With the order filtration the pieces are infinite dimensional over $K$ , and one has to argue over $K[x_1, … ,x_n]$ instead - the source leaves that version as an exercise.

Construction and Proof

The proof has three stages: Hilbert's theorem for the commutative graded ring, the lifting theorem from graded to filtered, and the one-line combination.

Stage 1: Hilbert's basis theorem

Hilbert's basis theoremCoutinho (8.2.1) and (8.2.2)

If $R$ is a commutative noetherian ring, then $R[z]$ is noetherian. In particular $K[y_1, … , y_2n]$ is noetherian for any field $K$ .

Proof for one variable

Suppose some ideal $I ⊂ eq R[z]$ is not finitely generated. Build a sequence in $I$ greedily: let $f_1$ be an element of $I$ of least degree, and having chosen $f_1, … , f_k$ , let $f_k+1$ be an element of least degree in $I ∖ (f_1, … , f_k)$ . The set being removed is never all of $I$ , precisely because $I$ is not finitely generated, so the sequence is infinite. Write $n_k = f_k$ and $a_k$ for the leading coefficient of $f_k$ ; by construction $n_1 ≤ n_2 ≤ ⋅ s$ .

In $R$ the chain of ideals $(a_1) ⊂ eq (a_1, a_2) ⊂ eq ⋅ s$ must stop, so for some $k$ we can write $a_k+1 = ∑ _i=1^k b_i a_i$ with $b_i ∈ R$ . Set

g ;=; f_k+1 - ∑ _i=1^k b_i, z^,n_k+1 - n_i f_i ,

which is a genuine polynomial because $n_i ≤ n_k+1$ . The degree- $n_k+1$ terms cancel, so $g < n_k+1$ . Also $g ∈ I$ , and $g ∉ (f_1, … ,f_k)$ since otherwise $f_k+1$ would lie there too. That contradicts the minimality in the choice of $f_k+1$ . Hence $I$ is finitely generated. Induction on the number of variables gives the many-variable statement.

Stage 2: lifting from the graded module

Noetherianity lifts across the filtrationCoutinho (8.2.3)

Let $M$ be a left $A_n$ -module with a filtration $Γ$ compatible with the Bernstein filtration. If $gr^Γ M$ is a noetherian $S_n$ -module, then $M$ is a noetherian $A_n$ -module.

Proof

Let $N ⊂ eq M$ be a submodule with the induced filtration $Γ^{'}$ . By (8.3), $gr^Γ' N$ is an $S_n$ -submodule of the noetherian module $gr^Γ M$ , hence finitely generated. Choose finitely many homogeneous generators and let $m$ be the largest of their degrees.

Claim: $N$ is generated as an $A_n$ -module by $Γ'_m = N ∩ Γ_m$ . If not, choose the least $k$ for which some $v ∈ Γ'_k$ fails to lie in $A_n Γ'_m$ ; necessarily $k > m$ . The symbol $μ_k(v)$ lies in $gr^Γ' N$ , so it is an $S_n$ -combination of the chosen generators:

μ_k(v) ;=; ∑ _i=1^s σ_k - r_i(a_i), μ_r_i(u_i), a_i ∈ B_k-r_i,; u_i ∈ Γ'_r_i,; r_i ≤ m .

The element $∑ _i a_i u_i$ lies in $Γ'_k$ and has the same symbol as $v$ , so $v - ∑ _i a_i u_i ∈ Γ'_k-1$ . By minimality of $k$ that difference lies in $A_n Γ'_m$ , and since each $u_i ∈ Γ'_m$ so does $v$ - a contradiction. The claim follows.

Finally $Γ'_m$ is a subspace of the finite dimensional space $Γ_m$ , so it has a finite $K$ -basis, and that basis generates $N$ over $A_n$ . Every submodule of $M$ is therefore finitely generated.

Stage 3: the combination

$A_n$ is left noetherianCoutinho (8.2.4)

Take $M = A_n$ with $Γ = B$ . Then $gr^B A_n = S_n$ is a polynomial ring in $2n$ variables by Coutinho (7.3.1), hence noetherian by Hilbert's theorem, hence noetherian as a module over itself. Stage 2 applies and $A_n$ is a noetherian left $A_n$ -module, that is, a left noetherian ring.

Finitely generated modules are noetherianCoutinho (8.2.5)

If $R$ is a left noetherian ring, every finitely generated left $R$ -module $M$ is noetherian. Indeed a generating set of size $k$ gives a surjection $R^k ↠ M$ ; the free module $R^k$ is noetherian because a sum of two noetherian submodules is noetherian, applied $k-1$ times; and a quotient of a noetherian module is noetherian.

Key Equations

The identification of the graded ring is the input that makes everything work:

S_n = gr^B A_n ; ≅ ; K[y_1, … , y_2n], y_i = σ_1(x_i), y_n+i = σ_1(∂_i).

(8.1)

Each filtration piece is finite dimensional, with dimension equal to the number of monomials $x^α ∂^β$ of total degree at most $m$ in $2n$ letters:

_K B_m = 2n+m2n.

(8.2)

For a submodule $N ⊂ eq M$ of a filtered module, the induced filtration and the resulting inclusion of graded modules are

Γ'_k = N ∩ Γ_k, gr^Γ' N ; ↪ ; gr^Γ M.

(8.3)

The conclusion of the lifting theorem is a statement about where the generators sit: if the generators of $gr^Γ' N$ have degree at most $m$ , then

N = A_n ⋅ (N ∩ Γ_m),

(8.4)

and $N ∩ Γ_m$ is a finite dimensional $K$ -vector space, so any of its bases is a finite generating set for $N$ .

For a principal left ideal the graded picture is completely explicit. If $d ∈ A_n$ has Bernstein degree $r$ , then the induced filtration on $A_n d$ is $B_k-r d$ , and

gr^B(A_n d) ; ≅ ; S_n, σ_r(d), gr^B!(A_n / A_n d) ; ≅ ; S_n / S_n σ_r(d).

(8.5)

Variable Definitions

$K$: the ground field; arbitrary characteristic is allowed on this page
$A_n$: the $n$ -th Weyl algebra over $K$
$B_m$: the $m$ -th piece of the Bernstein filtration of $A_n$ , spanned by the $x^α ∂^β$ with $|α| + |β| ≤ m$
$S_n$: the associated graded ring $gr^B A_n$ , a commutative polynomial ring in $2n$ variables
$σ_m$: the symbol map $B_m → B_m / B_m-1$ , sending an operator of degree $m$ to its leading form
$y_1, … ,y_2n$: the images $σ_1(x_i)$ and $σ_1(∂_i)$ , which are the polynomial generators of $S_n$
$M$: a left $A_n$ -module carrying a filtration $Γ$ compatible with $B$
$Γ_k$: the $k$ -th piece of that filtration; $B_i Γ_k ⊂ eq Γ_i+k$ and $_K Γ_k < ∈ fty$
$μ_k$: the symbol map $Γ_k → Γ_k / Γ_k-1$ of the module filtration
$N$: a submodule of $M$ , carrying the induced filtration $Γ'_k = N ∩ Γ_k$

Properties and Behaviour

Consequences for modules

Let $M$ be a finitely generated left $A_n$ -module. Then: every submodule of $M$ is finitely generated; every ascending chain of submodules of $M$ stops; every set of submodules of $M$ has a maximal element; and the kernel of any homomorphism $A_n^r → M$ is finitely generated, so $M$ has a presentation by a finite matrix over $A_n$ and, iterating, a free resolution by finitely generated free modules.

$A_n$ is a noetherian domain

$A_n$ has no zero divisors, because $gr^B A_n$ is a polynomial ring and hence a domain; see the domain page. A noetherian domain satisfies the Ore condition, so $A_n$ embeds in a division ring of fractions $D_n$ , the $n$ -th Weyl skew field. Neither statement is available for the free algebra, which is a domain but not noetherian, nor for a commutative ring with nilpotents, which may be noetherian but not a domain.

A bound on the number of generatorsStafford 1978

Noetherian rings need not admit any uniform bound on the number of generators of an ideal: in $K[x_1,x_2]$ the ideal $I_k$ generated by all monomials of degree $k$ needs $k+1$ generators, since $I_k / (x_1,x_2)I_k$ has dimension $k+1$ over $K$ . For $A_n$ the situation is better - Stafford proved that every left ideal of $A_n$ can be generated by two elements. That result is much harder than noetherianity and is quoted, not proved, in the source.

Both sides, and both filtrations

$A_n$ is right noetherian as well, by transposition. The proof also runs with the order filtration $F$ in place of $B$ : there $gr^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n]$ , again a polynomial ring in $2n$ variables, and the finite dimensionality of the filtration pieces is replaced by finite generation over $K[x_1, … ,x_n]$ . The source sets this out as Exercise 8.4.3.

Examples and Special Cases

A noncommutative ring that is not noetherian

Let $F = K ⟨ x, y ⟩$ be the free algebra on two letters, and set $I_k = ∑ _j=0^k F, y x^j$ . A word in the spanning set of $I_k$ has the form $w, y, x^j$ with $j ≤ k$ , so its last occurrence of $y$ is followed by at most $k$ letters $x$ . The word $y x^k+1$ fails that test, so $y x^k+1 ∉ I_k$ and

I_0 ⊂ neq I_1 ⊂ neq I_2 ⊂ neq ⋅ s

is a strictly increasing chain of left ideals. Hence $F$ is not left noetherian, although it is a domain and is finitely generated as a $K$ -algebra. Being a finitely generated algebra is not enough; the Weyl algebra needs its filtration.

Rings of differential operators that are not noetherian

For a smooth affine variety $X$ over a field of characteristic zero, the ring $D(X)$ of differential operators is noetherian, by essentially the argument on this page applied to a suitable filtration. Smoothness cannot be dropped: Bernstein, Gelfand and Gelfand showed in 1972 that for the cubic cone $x^3 + y^3 + z^3 = 0$ the ring $D(X)$ is neither finitely generated as an algebra nor noetherian. The Weyl algebra is the smooth case $X = A^n$ .

Companions in the noetherian club

$K[x_1, … ,x_n]$ , by Hilbert's theorem; and every quotient and every localisation of it.
$ℤ$ and every principal ideal domain, trivially.
The enveloping algebra $U(g)$ of a finite dimensional Lie algebra: its Poincare-Birkhoff-Witt filtration has a polynomial associated graded ring, so the same lifting argument applies verbatim.
$A_n(R)$ for any commutative noetherian ring $R$ of coefficients, since the graded ring is then $R[y_1, … ,y_2n]$ .

Worked Example

A two-generator left ideal of $A_1$ that needs only one generator

Step 1 - set up the graded picture
Work in $A_1$ with generators $x, ∂$ and the Bernstein filtration. The graded ring is $S_1 = K[y_1, y_2]$ with $y_1 = σ_1(x)$ and $y_2 = σ_1(∂)$ . Let

$J ;=; A_1 ∂^2 ;+; A_1 x∂ .$

Both generators have Bernstein degree $2$ , with symbols $σ_2(∂^2) = y_2^2$ and $σ_2(x∂) = y_1 y_2$ .
Step 2 - the naive guess for the graded ideal
The obvious guess is $gr(J) = (y_1 y_2,, y_2^2)$ , the ideal of $S_1$ generated by the symbols of the two given generators. That ideal contains no non-zero element of degree $1$ , since it is generated in degree $2$ .
Step 3 - the guess is wrong
Compute inside $A_1$ , using $∂ x = x∂ + 1$ :

$∂ ⋅ (x∂) ; - ; x ⋅ ∂^2 ;=; (x∂ + 1)∂ - x∂^2 ;=; ∂ .$

So $∂ ∈ J$ , and $σ_1(∂) = y_2$ is a degree- $1$ element of $gr(J)$ . The guess of Step 2 is therefore strictly too small: symbols of a generating set need not generate the associated graded ideal. The cancellation of the leading terms $x y_1 y_2$ against $x y_1 y_2$ is exactly what lets a lower-degree element appear.
Step 4 - identify $J$ and its graded ideal
Since $∂ ∈ J$ we have $A_1 ∂ ⊂ eq J$ ; conversely $∂^2 = ∂ ⋅ ∂$ and $x∂ = x ⋅ ∂$ both lie in $A_1∂$ , so $J = A_1 ∂$ . Because $S_1$ is a domain, Bernstein degree is additive: $(a∂) = a + 1$ . Hence $J ∩ B_k = B_k-1∂$ , and by (8.5)

$gr(J) ;=; S_1, y_2 ;=; (y_2),$

an ideal generated in degree $1$ .
Step 5 - run the lifting argument
The generator of $gr(J)$ has degree $m = 1$ , so by (8.4) the ideal $J$ is generated by $J ∩ B_1$ . An element $a∂$ has degree $a + 1 ≤ 1$ only when $a ∈ K$ , so $J ∩ B_1 = K∂$ , a one-dimensional space. Its basis $∂$ generates $J$ , in agreement with Step 4.
Step 6 - sanity check the quotient
By (8.5), $gr^B(A_1/J) ≅ K[y_1,y_2]/(y_2) ≅ K[y_1]$ , whose $k$ -th homogeneous piece is one dimensional. So $_K Γ_k = k+1$ for the induced filtration on $A_1/J ≅ K[x]$ - and indeed $Γ_k$ is the space of polynomials of degree at most $k$ , of dimension $k+1$ . The two computations agree.

Result

$J = A_1∂^2 + A_1 x∂$ equals the principal left ideal $A_1∂$ , its graded ideal is $(y_2)$ rather than the naive $(y_1y_2, y_2^2)$ , and the lifting theorem correctly predicts that a single element of $B_1$ generates it. The example also shows why the theorem is stated with the graded module and not with the symbols of a chosen generating set.

Applications and Industry Use

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

Dimension theory. Every finitely generated $A_n$ -module admits a good filtration, and the noetherian property is what makes the induced filtration on a submodule good as well. Without that, the Hilbert polynomial machinery of Chapter 9 would not get off the ground.
Presentations and syzygies. A finitely generated module can be written as $A_n^r / N$ with $N$ finitely generated; iterating gives finite free resolutions, the object every homological computation in the theory manipulates.
Termination of algorithms. Buchberger's algorithm in the Weyl algebra terminates because the chain of leading-term ideals it produces in $S_n$ must stabilise. Noetherianity of $S_n$ is the termination proof.
Localisation. $A_n$ being a noetherian domain gives the Ore condition and so a well-behaved skew field of fractions, used when one wants to invert operators or work generically.
Transfer to other algebras. The same three-stage template - identify the graded ring, quote Hilbert, lift - proves noetherianity for enveloping algebras, for rings of differential operators on smooth affine varieties, and for many quantum algebras with a polynomial associated graded ring.

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.

In practice the theorem is used in its effective form: given a finite set of generators for a left ideal or submodule, compute a set whose symbols generate the graded object.

Fix a monomial order on $A_n$ that refines the Bernstein or the order filtration; the noncommutative multiplication is handled by normalising every product to the canonical form $∑ c_αβ x^α ∂^β$ .
Run Buchberger's algorithm: form S-pairs, reduce, and add non-zero remainders. Because $S_n$ is noetherian, the strictly increasing chain of leading-term ideals must stop, so the loop terminates.
The resulting Gröbner basis has the property the naive generating set lacks: its leading terms generate $gr(J)$ exactly.
Membership, intersection, quotients and syzygies are then all decidable by linear algebra over the leading-term ideal.

The cost is severe - worst-case doubly exponential in $2n$ - and coefficient growth over $ℚ$ is usually the practical bottleneck long before the theoretical bound bites. Implementations include the Dmodules package of Macaulay2, dmod.lib and dmodapp.lib in Singular, the ore_algebra package in SageMath, and HolonomicFunctions.m in Mathematica.

Limits of Validity

The hypotheses of the theorem are unusually cheap, and it is worth being precise about which ones matter.

Characteristic plays no role. Unlike simplicity, which fails in characteristic $p$ , noetherianity of $A_n(K)$ holds over any field. The proof only uses the canonical basis $x^α ∂^β$ and the identification of $S_n$ , both of which are characteristic-free.
Finitely many variables is essential. The infinite Weyl algebra $A_ ∈ fty = A_n$ is not noetherian, for the same reason as $K[x_1,x_2, … ]$ : the left ideal generated by all the $∂_i$ is not finitely generated.
The filtration pieces must be finite dimensional for the argument as stated. For the order filtration the correct replacement condition is that each piece be a finitely generated $K[x_1, … ,x_n]$ -module, and Hilbert's theorem is then used a second time.
Noetherian is not the same as "few generators". It bounds nothing about generator counts; the bound of two for $A_n$ is a separate and much deeper theorem.
Noetherian says nothing about the two-sided ideal structure here. $A_n$ in characteristic zero is simple, so its only two-sided ideals are $0$ and $A_n$ ; that is a different theorem with different hypotheses.

Failure Modes and Common Mistakes

Assuming the symbols of generators generate the graded ideal

If $J = A_n f_1 + ⋅ s + A_n f_s$ , it is always true that $gr(J) ⊃ eq (σ(f_1), … , σ(f_s))$ , and the inclusion is very often strict, as the worked example above shows. Leading terms can cancel in a combination and expose an element of lower degree. Computing $gr(J)$ correctly is exactly the problem a Gröbner basis computation solves; the source makes the same point in Exercise 8.4.5.

Expecting the lifting theorem to reverse

The theorem says: $gr^Γ M$ noetherian $\Rightarrow$ $M$ noetherian. The converse is false for an arbitrary filtration. A finitely generated module always admits a filtration whose associated graded module is not finitely generated - for instance $Ω_k = B_2k$ on $M = A_n$ . The filtrations for which the implication runs both ways are precisely the good ones.

Reading the ascending chain condition as applying to subspaces

Noetherian constrains chains of submodules, not chains of $K$ -subspaces. $A_n$ contains the strictly increasing chain of subspaces $B_0 ⊂ neq B_1 ⊂ neq B_2 ⊂ neq ⋅ s$ , which never stops and contradicts nothing. Only chains closed under multiplication by the ring are controlled.

Assuming left noetherian implies right noetherian in general

It does for $A_n$ , because of the transposition anti-automorphism, but not for rings in general: there are standard examples of rings that are right noetherian and not left noetherian. Whenever a ring lacks an anti-automorphism, the two sides must be checked separately.

Confusing noetherian with principal ideal domain

$A_1$ is noetherian but is not a principal left ideal domain: the left ideal $A_1 x + A_1 ∂$ is all of $A_1$ by simplicity, but for instance $A_1 x^2 + A_1 ∂$ also collapses to $A_1$ , while genuinely non-principal proper left ideals of $A_1$ exist. Finite generation is all that is claimed.

Historical Notes

Hilbert proved the basis theorem in 1890 in his work on invariant theory, replacing Gordan's constructive computations with a finiteness proof that Gordan famously described as theology. The condition itself was isolated and named by Emmy Noether in her 1921 paper on ideal theory in rings, where the ascending chain condition was made the organising axiom.

The noncommutative filtration technique used here - approximate a ring by its associated graded ring and transfer finiteness properties back - became standard for enveloping algebras and rings of differential operators through the 1960s and 1970s, and is set out systematically in Bjork's 1979 book and in McConnell and Robson. The Weyl algebra is the model case: the graded ring is as simple as possible, a polynomial ring, so the transfer is immediate.

Two later results complete the picture. In 1972 Bernstein, Gelfand and Gelfand showed that on a singular variety the ring of differential operators can fail to be noetherian, marking the boundary of the method. In 1978 Stafford proved that every left ideal of $A_n$ is generated by two elements, a strengthening in a direction Hilbert's theorem gives no access to.

Comparison

How the Weyl algebra sits among nearby rings.
Ring	Noetherian?	Domain?	Simple?	Bound on generators of a left ideal
$A_n(K)$ , any $K$	yes, both sides	yes	yes if $char K = 0$	$2$ (Stafford)
$K[x_1, … ,x_2n]$	yes	yes	no	unbounded
$K ⟨ x, y ⟩$	no	yes	no	not applicable
$K[x_1,x_2, … ]$	no	yes	no	not applicable
$U(g)$ , $g < ∈ fty$	yes	yes	generally no	unbounded in general
$D(X)$ , $X$ smooth affine	yes	if $X$ is irreducible	yes if $X$ is smooth irreducible over $ℂ$	$2$
$D(X)$ , $X$ the cubic cone	no	yes	no	not applicable

Key Takeaways

Key points

$A_n(K)$ is left and right noetherian for every field $K$ ; no assumption on the characteristic is used.
The proof is a transfer: $gr^B A_n = S_n$ is a polynomial ring in $2n$ variables, noetherian by Hilbert's basis theorem, and noetherianity lifts back across the filtration.
The lifting step works because leading terms determine generators, and because each $B_m$ is finite dimensional over $K$ .
Every finitely generated $A_n$ -module is therefore noetherian, so submodules are finitely generated and finite presentations and resolutions exist.
The symbols of a chosen generating set of an ideal usually do not generate the graded ideal; producing a set that does is what a Gröbner basis is for.
Noetherianity is not automatic in this world: the free algebra and the ring of differential operators on a singular variety both fail it.

FAQs

Does the result need characteristic zero?

No. Every ingredient - the canonical basis of $A_n$ , the identification $gr^B A_n ≅ K[y_1, … ,y_2n]$ , Hilbert's basis theorem, and the lifting argument - is valid over any field. This is a useful contrast with simplicity of $A_n$ , which genuinely requires $char K = 0$ .

Why is a filtration needed at all - why not imitate Hilbert's proof directly in

A_n

Hilbert's proof compares leading coefficients in a commutative ring of coefficients. In $A_n$ there is no distinguished commutative coefficient ring in which the leading data lives, because the variables do not commute past each other. The filtration manufactures one: $S_n$ plays the role of the coefficient ring, and Bernstein degree plays the role of degree in $z$ .

Is the converse of the lifting theorem true?

Not for an arbitrary filtration. If $gr^Γ M$ is finitely generated then $M$ is finitely generated, but a finitely generated $M$ can carry filtrations whose graded module is not finitely generated. The filtrations for which the correspondence is exact are the good filtrations.

Does noetherian imply that every left ideal of

A_1

is principal?

No. Noetherian only says finitely generated. $A_1$ has proper left ideals that are not principal; what is true, by Stafford's theorem, is that two elements always suffice.

How does noetherianity interact with short exact sequences?

Given $0 → N → M → M/N → 0$ , the module $M$ is noetherian if and only if both $N$ and $M/N$ are. That equivalence is what makes the property propagate through constructions; it is treated on the exact sequences page.

Does the same proof work with the order filtration?

Yes, with one adjustment. The pieces $F_m$ of the order filtration are infinite dimensional over $K$ but finitely generated as $K[x_1, … ,x_n]$ -modules, and $gr^F A_n ≅ K[x_1, … ,x_n][ξ_1, … ,ξ_n]$ . The final step selects a finite generating set over $K[x_1, … ,x_n]$ rather than a $K$ -basis, using Hilbert's theorem a second time.

What does noetherianity buy that finite generation alone does not?