The Weyl Algebra by Generators and Relations

Q: Why present A_n at all, when it already has a perfectly good definition?

Because the operator definition makes homomorphisms out of A_n awkward. Every module, every automorphism and every comparison with another algebra is built by choosing images of the generators and checking relations, which is exactly what the presentation licenses.

Q: Is the list of relations minimal?

As stated it is redundant in the trivial way: [z_i,z_j] and [z_j,z_i] generate the same ideal, and the cases i = j are automatically zero. What matters is that no family can be dropped. Removing the relations among the ∂ 's, say, gives a strictly larger algebra.

Q: Does the theorem say A_n has no other relations?

Yes, in the precise sense that ϕ = J : any identity satisfied by the operators x_i, ∂_i is a formal consequence of (3.2). That is exactly what makes computation in canonical form complete rather than merely sound.

Q: What is the difference between Q and A_n in characteristic p?

The presented algebra Q still has basis the ordered monomials, and its centre is generated by z_i^,p and z_n+i^,p . The operator algebra is the quotient of Q by the ideal generated by the z_n+i^,p , since ∂_i^,p annihilates every polynomial. Calling both 'the Weyl algebra' in characteristic p is a genuine ambiguity, and authors differ.

Q: Does the universal property produce automorphisms directly?

It produces endomorphisms. To get an automorphism you must exhibit an inverse, as in Step 4 of the worked example. There is no shortcut: injectivity is free from simplicity, but surjectivity is exactly the difficulty behind the Dixmier conjecture.

Q: How does this relate to defining a module?

An A_n -module structure on a K -vector space M is the same thing as a homomorphism A_n → End_K(M) , so by the universal property it is the same as 2n linear operators on M satisfying the relations. That is the systematic version of the ad hoc constructions used throughout Chapter 5.

Q: Can one present A_n with fewer generators?

A_1 is generated as a K -algebra by x and ∂ and by no single element, since it is non-commutative. There are presentations using different generating sets - for instance x^2, ∂^2 and x∂ , which span a copy of the Lie algebra sl_2 inside A_1 - but these generate a proper subalgebra, not all of A_1 , so they do not give an alternative presentation of the whole algebra.

Overview

There are two ways to say what the Weyl algebra is. The first is concrete: $A_n$ is the subalgebra of $End_K(K[X])$ generated by the multiplication operators $x_i$ and the partial derivatives $∂_i$ . The second is abstract: $A_n$ is the associative $K$ -algebra generated by $2n$ symbols subject only to the canonical commutation relations. This page proves the two descriptions agree, and explains why the second is worth having.

The concrete definition guarantees that the algebra exists and comes with a faithful module. What it does not do is make homomorphisms easy. To define a map out of $A_n$ from the operator definition you must check that your rule is consistent on every expression that happens to represent the same operator - an unbounded amount of work. The presentation replaces that with a finite check: choose images for the $2n$ generators, verify the relations hold, and the homomorphism exists and is unique.

That is the universal property, and almost every construction in the theory uses it. Automorphisms of $A_n$ are built by listing images of generators; a module structure on a vector space is specified by giving $2n$ operators that satisfy the relations, which is the content of defining a module action from generators and relations; and the isomorphism $A_1 ≅$ the algebra of creation and annihilation operators, familiar from quantum mechanics, is a one-line verification rather than a computation.

The proof that the presentation really does present $A_n$ rests on the canonical basis theorem and therefore, in the form given here, on $char K = 0$ . What follows states the presentation precisely, proves the theorem, derives the universal property, and applies it to construct a family of automorphisms that reappears in the chapters on embeddings and direct images.

Definition

The free algebra

Let $Kz_1, … ,z_2n$ denote the free associative $K$ -algebra on the symbols $z_1, … ,z_2n$ : its elements are finite $K$ -linear combinations of words in the $z_i$ , including the empty word, which is the identity. Multiplication is concatenation of words extended bilinearly. Nothing is assumed about the $z_i$ beyond associativity and $K$ -linearity; in particular they do not commute.

Its defining property is that a $K$ -algebra homomorphism $Kz_1, … ,z_2n → R$ is exactly a choice of $2n$ arbitrary elements of $R$ .

The relation ideal

Let $J$ be the two-sided ideal of $Kz_1, … ,z_2n$ generated by the elements

[z_n+i, z_j] - δ_ij, [z_i, z_j], [z_n+i, z_n+j] (1 ≤ i, j ≤ n),

(3.1)

where $[a,b] = ab - ba$ . Coutinho lists the same ideal slightly differently, as $[z_i+n,z_i] - 1$ together with $[z_i,z_j]$ for every pair that is not a matching variable-derivative pair; the two lists generate the same ideal.

The comparison map

Define $ϕ : Kz_1, … ,z_2n → A_n$ by $ϕ(z_i) = x_i$ and $ϕ(z_n+i) = ∂_i$ for $1 ≤ i ≤ n$ . It is surjective, because the $x_i$ and $∂_i$ generate $A_n$ . The relations of $A_n$ give $J ⊂ eq ϕ$ , so $ϕ$ descends to

ϕ : Kz_1, … ,z_2n/J ⟶ A_n.

The presentation is faithfulCoutinho (1.3.1)

Let $K$ be a field of characteristic zero. Then $\bar{ϕ}$ is an isomorphism of $K$ -algebras. Equivalently, $J = ϕ$ : every relation that holds among the generators of $A_n$ is a consequence of the canonical commutation relations.

Core Concepts

What a presentation buys you

A presentation is a trade. You give up the concrete realisation - the elements are now equivalence classes of formal words, not operators - and in exchange you get a clean statement about maps out of the algebra. That statement is the universal property, and it is the reason presentations are used even when a concrete model is available.

The asymmetry is worth stating plainly. Maps out of a presented algebra are easy: pick images, check relations. Maps into it are hard, and so is deciding whether a given element is zero. The operator model is the mirror image: deciding whether an element is zero is easy, because you can apply it to a polynomial, but building maps out is awkward. Having both descriptions, and knowing they agree, means each question can be attacked in whichever model makes it easy.

Why the theorem is not automatic

The inclusion $J ⊂ eq ϕ$ is a calculation. The reverse inclusion is a genuine theorem, and it can fail for a badly chosen relation list. If $J$ were too small, the quotient would be bigger than $A_n$ and $\bar{ϕ}$ would have a kernel; if the relations were inconsistent, the quotient could collapse. What the theorem says is that the relations (3.1) are exactly enough: enough to force every element into canonical form, and not so many as to identify distinct operators.

The shape of the proof

Both halves reduce to counting. The relations let any word be rewritten as a combination of the ordered monomials $z^α z_n+ ∙ ^β$ , so the quotient is spanned by their classes; hence $_K$ of the degree- $≤ m$ part of the quotient is at most $m+2n2n$ . Meanwhile $\bar{ϕ}$ carries those classes onto the canonical basis of $A_n$ , which is independent and has exactly $m+2n2n$ elements in that range. A surjection between spaces of the same finite dimension is bijective, in each degree, hence overall.

Construction and Proof

Step 1: the relations are satisfied in the Weyl algebra

That $J ⊂ eq ϕ$

Applying $ϕ$ to the listed generators of $J$ gives $[∂_i, x_j] - δ_ij$ , $[x_i,x_j]$ and $[∂_i,∂_j]$ , all of which vanish in $A_n$ by the commutation relations. Since $ker ϕ$ is a two-sided ideal and $J$ is the smallest two-sided ideal containing those elements, $J ⊂ eq ϕ$ .

Step 2: ordered monomials span the quotient

Lemma

Write $z_i$ for the class of $z_i$ in $Q = Kz_1, … ,z_2n/J$ . Then $Q$ is spanned over $K$ by the ordered monomials

z^,α,β = z_1^α_1 ⋅ s z_n^α_n, z_n+1^β_1 ⋅ s z_2n^β_n, α,β ∈ N^n.

Proof

$Q$ is spanned by classes of words. In $Q$ the relations read $z_n+i z_j = z_j z_n+i + δ_ij$ , together with commutativity within each block. Any word containing a factor $z_n+i z_j$ can therefore be replaced by the swapped word plus, when $i = j$ , a word of length two less. As in the spanning half of the canonical basis proof, this rewriting terminates: each step decreases the number of inversions at fixed length, or decreases the length. What is left is a combination of ordered monomials.

Consequently the span $Q_ ≤ m$ of the classes of words of length at most $m$ satisfies $_K Q_ ≤ m ≤ #(α,β) : |α|+|β| ≤ m = m+2n2n$ .

Step 3: comparing dimensions

Proof of the theorem

The map $\bar{ϕ}$ is surjective because $ϕ$ is. It sends $z^,α,β$ to $x^α∂^β$ and does not raise word length, so it restricts to a surjection $Q_ ≤ m ↠ (A_n)_ ≤ m$ , where $(A_n)_ ≤ m$ is the span of the operators of degree at most $m$ .

By the canonical basis theorem, $_K (A_n)_ ≤ m = m+2n2n$ exactly. By Step 2, $_K Q_ ≤ m ≤ m+2n2n$ . A surjective linear map from a space of dimension at most $N$ onto a space of dimension $N$ is an isomorphism, so $\bar{ϕ}$ is bijective on each $Q_ ≤ m$ . Since every element of $Q$ lies in some $Q_ ≤ m$ and the $(A_n)_ ≤ m$ exhaust $A_n$ , the map $\bar{ϕ}$ is bijective. A bijective algebra homomorphism is an isomorphism.

What was used where

Only Step 3 uses characteristic zero, and it uses it only through the canonical basis theorem. Steps 1 and 2 are valid over any commutative ring. The abstract quotient $Q$ has the ordered monomials as a basis in every characteristic, by Bergman's diamond lemma; what fails in characteristic $p$ is not the presentation but the identification of $Q$ with the operator algebra.

Key Equations

The relations, written in the algebra rather than in the free algebra, are

[∂_i, x_j] = δ_ij, [x_i, x_j] = 0, [∂_i, ∂_j] = 0 (1 ≤ i,j ≤ n).

(3.2)

The presentation itself is the statement

A_n ; ≅ ; Kz_1, … ,z_2n,/,J, J = ( [z_n+i,z_j] - δ_ij,; [z_i,z_j],; [z_n+i,z_n+j] ).

(3.3)

The universal property it yields is the assertion that for every $K$ -algebra $R$ ,

Hom_K-alg(A_n, R) ; ≅ ; ,(a_1, … ,a_n,b_1, … ,b_n) ∈ R^2n : [b_i,a_j] = δ_ij, [a_i,a_j] = [b_i,b_j] = 0 ,.

(3.4)

The family of automorphisms constructed below is given by

σ(x_i) = x_i + f_i, σ(∂_i) = ∂_i - ∑ _j=1^n (∂ f_j)/(∂ x_i), ∂_j,

(3.5)

where the $f_i$ are chosen as in the corollary below; its inverse is obtained by replacing every $f_i$ by $-f_i$ .

Variable Definitions

$K$: the ground field, of characteristic zero
$Kz_1, … ,z_2n$: the free associative $K$ -algebra on $2n$ non-commuting symbols
$J$: the two-sided ideal generated by the commutation relations (3.1)
$Q$: the quotient $Kz_1, … ,z_2n/J$ , the presented algebra
$ϕ, \bar{ϕ}$: the surjection onto $A_n$ and the induced map on the quotient
$δ_ij$: the Kronecker delta
$[a,b]$: the commutator $ab - ba$
$R$: an arbitrary $K$ -algebra, the target of a homomorphism built by the universal property
$σ, τ$: the automorphism of (3.5) and its inverse
$f_i$: the polynomials defining $σ$ , subject to the triangularity hypothesis

Properties and Behaviour

Universal property

Let $R$ be a $K$ -algebra containing elements $a_1, … ,a_n,b_1, … ,b_n$ with $[b_i,a_j] = δ_ij$ and $[a_i,a_j] = [b_i,b_j] = 0$ . Then there is a unique $K$ -algebra homomorphism $ψ : A_n → R$ with $ψ(x_i) = a_i$ and $ψ(∂_i) = b_i$ .

Indeed the assignment defines a homomorphism from the free algebra, which kills the generators of $J$ by hypothesis, hence factors through $Q ≅ A_n$ . Uniqueness holds because the $x_i, ∂_i$ generate.

Any such $R$ contains a copy of $A_n$

If in addition $R ≠ 0$ and $1 ≠ 0$ in $R$ , then $ψ$ is injective, because $A_n$ is simple and $ψ(1) = 1 ≠ 0$ forces $ψ ≠ A_n$ , hence $ψ = 0$ . So a non-zero algebra carrying $2n$ elements obeying the relations contains $A_n$ as a subalgebra. This is why the creation and annihilation operators of the quantum harmonic oscillator generate a copy of $A_1$ .

Triangular automorphismsCoutinho (1.3.2)

Let $1 ≤ m ≤ n$ . Choose $f_1, … ,f_n ∈ K[X]$ so that $f_i$ is a polynomial in $x_m+1, … ,x_n$ alone when $i ≤ m$ , and $f_i = 0$ when $i > m$ . Then the assignment (3.5) defines an automorphism $σ$ of $A_n$ , with inverse given by the same formulas with $f_i$ replaced by $-f_i$ .

Why the relations hold

Write $X_j = x_j + f_j$ and $D_i = ∂_i - ∑ _k (∂ f_k/∂ x_i)∂_k$ . Since all the $X_j$ are multiplication operators, $[X_i, X_j] = 0$ . For the mixed relation, the coefficient polynomials commute with $X_j$ , so

[D_i, X_j] = δ_ij + (∂ f_j)/(∂ x_i) - ∑ _k (∂ f_k)/(∂ x_i)( δ_kj + (∂ f_j)/(∂ x_k) ) = δ_ij - ∑ _k (∂ f_k)/(∂ x_i), (∂ f_j)/(∂ x_k).

The hypothesis makes every term of the last sum vanish: $∂ f_k/∂ x_i ≠ 0$ forces $k ≤ m$ and $i > m$ , while $∂ f_j/∂ x_k ≠ 0$ forces $k > m$ . So $[D_i, X_j] = δ_ij$ . For $[D_i, D_j]$ , the terms quadratic in the $f$ 's vanish for the same reason, and what remains is $∑ _k ( ∂^2 f_k/∂ x_i ∂ x_j - ∂^2 f_k/∂ x_j ∂ x_i )∂_k = 0$ by equality of mixed partial derivatives. The universal property now produces $σ$ , and the composite of $σ$ with its candidate inverse is the identity on generators because $f_i$ does not involve any variable that $σ$ moves.

Where the hypothesis comes from

Any automorphism $θ$ of $K[X]$ induces an automorphism $D ↦ θ D θ^-1$ of $A_n$ , and (3.5) is exactly that construction for the substitution $x_i ↦ x_i + f_i$ . The triangularity hypothesis is what guarantees that this substitution is invertible without any further check. For a general choice of $f_i$ the substitution need not be invertible, and deciding when it is, is the Jacobian conjecture.

Examples and Special Cases

The Fourier automorphism

Set $a_i = ∂_i$ and $b_i = -x_i$ . Then $[b_i, a_j] = [-x_i, ∂_j] = δ_ij$ , and the other two families of relations are clear. The universal property gives an automorphism $F$ of $A_n$ with $F(x_i) = ∂_i$ and $F(∂_i) = -x_i$ ; $F^2$ is the automorphism negating every generator, so $F$ has order $4$ in $Aut(A_n)$ . It is the algebraic shadow of the Fourier transform, which exchanges multiplication and differentiation.

The action of $SL_2$ on $A_1$

For a matrix with entries $a, b, c, d$ and $ad - bc = 1$ , put $σ(∂) = a∂ + bx$ and $σ(x) = c∂ + dx$ . Then $[σ(∂), σ(x)] = ad[∂,x] + bc[x,∂] = ad - bc = 1$ , so $σ$ is an automorphism of $A_1$ . This gives an action of $SL_2(K)$ on $A_1$ by algebra automorphisms; the Fourier automorphism is the case $a = 0, b = -1, c = 1, d = 0$ .

Creation and annihilation operators

Let $V$ be a vector space with basis $u_0, u_1, u_2, …$ over a field of characteristic zero in which square roots of positive integers exist, and set $ξ(u_k) = k+1, u_k+1$ and $η(u_k) = k, u_k-1$ with $η(u_0) = 0$ . Then $ηξ(u_k) = (k+1)u_k$ and $ξη(u_k) = k,u_k$ , so $[η,ξ] = 1$ . By the universal property the subalgebra generated by $ξ$ and $η$ is a homomorphic image of $A_1$ , and by simplicity it is a copy of $A_1$ . This is the harmonic oscillator algebra of quantum mechanics.

A module built from the relations

To make a vector space $M$ into an $A_n$ -module, it is enough to give $2n$ linear operators on $M$ obeying (3.2); a module is the same thing as a homomorphism $A_n → End_K(M)$ . This is how the polynomial module, the delta module and the twisted modules of Chapter 5 are all constructed. See the appendix page for the details.

Worked Example

A triangular automorphism of $A_2$ , verified from the relations

Step 1 - choose the data
Take $n = 2$ and $m = 1$ , so that $f_1$ may be any polynomial in $x_2$ alone and $f_2 = 0$ . Choose $f_1 = x_2^2$ . Formula (3.5) then reads

$σ(x_1) = x_1 + x_2^2, σ(x_2) = x_2, σ(∂_1) = ∂_1, σ(∂_2) = ∂_2 - 2x_2,∂_1,$

because $∂ f_1/∂ x_1 = 0$ and $∂ f_1 / ∂ x_2 = 2x_2$ , while $f_2 = 0$ contributes nothing.
Step 2 - check the four mixed relations
Each check is a two-line commutator computation.
- $[σ(∂_1), σ(x_1)] = [∂_1,, x_1 + x_2^2] = 1 + 0 = 1$ , as required.
- $[σ(∂_1), σ(x_2)] = [∂_1, x_2] = 0$ .
- $[σ(∂_2), σ(x_2)] = [∂_2 - 2x_2∂_1,, x_2] = 1 - 2x_2[∂_1,x_2] = 1$ .
- $[σ(∂_2), σ(x_1)] = [∂_2,, x_2^2] - 2x_2[∂_1,, x_1] = 2x_2 - 2x_2 = 0$ , as required, since $δ_21 = 0$ .
Step 3 - check the remaining relations
$[σ(x_1),σ(x_2)] = [x_1 + x_2^2, x_2] = 0$ , since both are multiplication operators. And

$[σ(∂_1), σ(∂_2)] = [∂_1,, ∂_2 - 2x_2∂_1] = -2,[∂_1, x_2],∂_1 = 0.$

All six relations hold, so by the universal property $σ$ extends uniquely to an endomorphism of $A_2$ .
Step 4 - produce the inverse
Let $τ$ be given by the same formulas with $f_1 = -x_2^2$ , that is $τ(x_1) = x_1 - x_2^2$ , $τ(x_2) = x_2$ , $τ(∂_1) = ∂_1$ , $τ(∂_2) = ∂_2 + 2x_2∂_1$ . Then
- $τσ(x_1) = τ(x_1) + τ(x_2)^2 = (x_1 - x_2^2) + x_2^2 = x_1$ ;
- $τσ(∂_2) = τ(∂_2) - 2τ(x_2)τ(∂_1) = (∂_2 + 2x_2∂_1) - 2x_2∂_1 = ∂_2$ ;
- $τσ(x_2) = x_2$ and $τσ(∂_1) = ∂_1$ trivially.
So $τ σ$ is the identity on generators, hence the identity; symmetrically $στ = id$ . Thus $σ$ is an automorphism.
Step 5 - see what it does to an operator
Apply $σ$ to $D = x_1∂_2$ . Then $σ(D) = (x_1 + x_2^2)(∂_2 - 2x_2∂_1) = x_1∂_2 - 2x_1x_2∂_1 + x_2^2∂_2 - 2x_2^3∂_1$ , already in canonical form and of degree $4$ , whereas $D = 2$ . Automorphisms of this kind do not preserve degree, which is exactly why they are useful for changing the shape of a module without changing its isomorphism class of algebra.

Result

$σ(x_1) = x_1 + x_2^2$ , $σ(x_2) = x_2$ , $σ(∂_1) = ∂_1$ , $σ(∂_2) = ∂_2 - 2x_2∂_1$ is an automorphism of $A_2$ with inverse obtained by negating $f_1$ . It is conjugation by the substitution $x_1 ↦ x_1 + x_2^2$ of $K[x_1,x_2]$ , and every step of the verification used only the relations - never the fact that these symbols are operators.

Applications and Industry Use

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

Constructing modules. A representation of $A_n$ is $2n$ operators satisfying (3.2). This is how essentially every explicit module in the theory is defined, from $K[X]$ to modules of formal solutions.
Constructing automorphisms. The triangular automorphisms of (3.5) are used to move a hypersurface into coordinate position; they are the algebraic side of the graph construction in the factorisation of a polynomial map through its graph.
Comparing with other algebras. Recognising a copy of $A_n$ inside another algebra becomes a finite check. This identifies the algebra generated by creation and annihilation operators, and the algebra generated by certain infinite matrices, as $A_1$ .
Deformation and quantisation. The presentation exhibits $A_n$ as a flat deformation of the commutative polynomial ring in $2n$ variables, with the relations $[∂_i,x_j] = ℏδ_ij$ interpolating; that is the starting point for deformation quantisation.
Computer algebra. Systems such as Singular's Plural accept a $G$ -algebra by its generators and its commutation relations. The Weyl algebra is entered exactly as (3.1), and the software's correctness conditions are the algebraic content of the theorem above.

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.

Presentations are the input format for non-commutative Groebner basis engines. A $G$ -algebra is specified by generators $z_1, … ,z_N$ together with rules $z_j z_i = c_ij z_i z_j + d_ij$ for $i < j$ , where $c_ij ∈ K^ ×$ and $d_ij$ is a polynomial that is smaller in a fixed ordering. The Weyl algebra is the case $c_ij = 1$ with $d_ij$ equal to $1$ for matching pairs and $0$ otherwise.

Declare the $2n$ generators and the relations (3.1).
The engine checks the non-degeneracy conditions - overlap or diamond conditions on triples of generators - which guarantee that ordered monomials form a basis. For the Weyl algebra they hold, which is a computational restatement of the theorem on this page.
Ordered monomials then serve as normal forms, and Buchberger's algorithm runs with the usual reduction step.

Practically: Singular's Plural kernel and its dmod.lib library, Macaulay2's Dmodules package, and SageMath's ore_algebra all construct $A_n$ from the presentation and then work with canonical forms. Because the correction terms in a product have strictly smaller degree, the leading exponent of a product is the sum of the leading exponents, and the standard theory of term orders applies unchanged. Nothing in the algorithms requires characteristic zero, but the mathematical guarantees attached to the output usually do.

Limits of Validity

Characteristic zero, for the identification. Step 3 of the proof uses the canonical basis theorem. Over a field of characteristic $p$ the presented algebra $Q$ still has the ordered monomials as a basis, but the operator algebra does not: $∂_i^,p$ acts as zero on $K[X]$ , so $\bar{ϕ}$ has a non-zero kernel and the two algebras are genuinely different. See the positive characteristic page.
The relations must be complete. The theorem is about this specific ideal. Dropping any of the three families in (3.1) gives a strictly larger algebra; the resulting quotient is not $A_n$ and generally has very different behaviour.
A presentation does not by itself give a faithful module. That $A_n$ acts faithfully on $K[X]$ is an input to the proof, not an output of the presentation. Presented algebras can be zero, or can fail to have any interesting module, and one always needs a separate argument or model.
Homomorphisms into $A_n$ are not covered. The universal property describes maps out of $A_n$ . Deciding which subalgebras of $A_n$ exist, or whether a given endomorphism is surjective, is not addressed - and in the latter case is the open Dixmier conjecture.

Failure Modes and Common Mistakes

Forgetting to check the relations among images

The universal property is conditional. Writing down a rule on generators does not define a homomorphism until all three families of relations are verified for the images - including the ones that look trivial. In (3.5) the vanishing of $[D_i, X_j]$ for $i ≠ j$ is precisely where the triangularity hypothesis on the $f$ 's is consumed; a version of the corollary without that hypothesis is false.

Assuming an endomorphism is an automorphism

Every non-zero endomorphism of $A_n$ is injective, because $A_n$ is simple. Injectivity is not surjectivity here: $A_n$ is infinite dimensional, so the usual finite-dimensional argument is unavailable. Whether every endomorphism of $A_n$ is onto is the Dixmier conjecture, open for all $n ≥ 1$ and known to be equivalent to the Jacobian conjecture in $2n$ variables. Coutinho flags this at the end of Ch. 1 §3, and it is worth not overstating.

Reading the presentation as if the generators commuted

$J$ is a two-sided ideal of a non-commutative algebra. Its elements are sums of terms $u r v$ with $r$ one of the listed relations and $u, v$ arbitrary words - not just $K$ -multiples of the relations. Computing modulo $J$ by treating it as an ideal in a commutative ring gives wrong answers immediately: for instance $z_n+1z_1$ and $z_1 z_n+1$ are different in $Q$ .

Expecting automorphisms to preserve degree or filtration

The worked example sends an operator of degree $2$ to one of degree $4$ . Automorphisms of $A_n$ preserve everything defined intrinsically - simplicity, the centre, dimension of modules - but not the Bernstein filtration, which is a choice rather than an invariant. Any argument that transports a filtration through an automorphism has to say why.

Historical Notes

The relations came first. Born, Heisenberg and Jordan wrote $pq - qp = -iℏ$ in 1925 as the defining rule of matrix mechanics, and the question of which operators can satisfy it - and on what space - drove a decade of work, culminating in the Stone-von Neumann theorem on the essentially unique unitary representation of the relations on a Hilbert space. The algebraic version, with no topology and no Hilbert space, is what became $A_n$ .

Presenting an algebra by generators and relations is older still and belongs to the general theory of free objects; for enveloping algebras the corresponding basis statement is the Poincare-Birkhoff-Witt theorem, and $A_n$ is the quotient of the enveloping algebra of the $(2n+1)$ -dimensional Heisenberg Lie algebra by the ideal setting the central element equal to $1$ . Bergman's diamond lemma of 1978 gave the general rewriting criterion under which a presentation has the expected monomial basis.

Dixmier's 1968 paper is where the algebraic theory of $A_n$ was systematised. It determined the automorphism group of $A_1$ - generated by the triangular automorphisms of the two obvious kinds together with $SL_2$ - and posed the problem, still open, of whether every endomorphism of $A_n$ is an automorphism. Makar-Limanov later described $Aut(A_1)$ as an amalgamated free product, in close analogy with the automorphism group of the polynomial ring in two variables.

Comparison

The two descriptions of the Weyl algebra, and what each makes easy.
	As a ring of operators	By generators and relations
Existence of the algebra	Immediate: a subalgebra of an endomorphism ring	Immediate: a quotient of a free algebra
Deciding whether an element is zero	Easy: apply it to polynomials	Hard without a normal form theorem
Building a homomorphism out	Hard: needs consistency on all expressions	Easy: choose images, check (3.2)
Faithful module supplied	Yes, $K[X]$ by construction	No, must be supplied separately
Canonical basis	A theorem, needs $char K = 0$	A theorem, valid in any characteristic
Behaviour in characteristic $p$	$∂_i^,p = 0$ ; a proper quotient	Ordered monomials still a basis
Typical use	Examples, verification, intuition	Constructions, automorphisms, modules

Key Takeaways

Key points

$A_n$ is the free $K$ -algebra on $2n$ generators modulo the two-sided ideal generated by the canonical commutation relations.
The proof compares dimensions degree by degree: the relations force ordered monomials to span the quotient, and the canonical basis theorem says the image is exactly that big.
The payoff is the universal property: a homomorphism out of $A_n$ is exactly a choice of $2n$ elements of the target satisfying (3.2).
Because $A_n$ is simple, any non-zero algebra with such a family of elements contains a copy of $A_n$ - for instance the algebra of creation and annihilation operators.
The universal property produces the triangular automorphisms $σ(x_i) = x_i + f_i$ , $σ(∂_i) = ∂_i - ∑ _j (∂ f_j/∂ x_i)∂_j$ , whose triangularity hypothesis is exactly what makes the relations hold.
In characteristic $p$ the presented algebra and the operator algebra differ; the theorem as stated needs characteristic zero, and only in its last step.

FAQs

Why present

A_n

at all, when it already has a perfectly good definition?

Because the operator definition makes homomorphisms out of $A_n$ awkward. Every module, every automorphism and every comparison with another algebra is built by choosing images of the generators and checking relations, which is exactly what the presentation licenses.

Is the list of relations minimal?

As stated it is redundant in the trivial way: $[z_i,z_j]$ and $[z_j,z_i]$ generate the same ideal, and the cases $i = j$ are automatically zero. What matters is that no family can be dropped. Removing the relations among the $∂$ 's, say, gives a strictly larger algebra.

Does the theorem say

A_n

has no other relations?

Yes, in the precise sense that $ϕ = J$ : any identity satisfied by the operators $x_i, ∂_i$ is a formal consequence of (3.2). That is exactly what makes computation in canonical form complete rather than merely sound.

What is the difference between

Q

and

A_n

in characteristic

p

The presented algebra $Q$ still has basis the ordered monomials, and its centre is generated by $z_i^,p$ and $z_n+i^,p$ . The operator algebra is the quotient of $Q$ by the ideal generated by the $z_n+i^,p$ , since $∂_i^,p$ annihilates every polynomial. Calling both 'the Weyl algebra' in characteristic $p$ is a genuine ambiguity, and authors differ.

Does the universal property produce automorphisms directly?

It produces endomorphisms. To get an automorphism you must exhibit an inverse, as in Step 4 of the worked example. There is no shortcut: injectivity is free from simplicity, but surjectivity is exactly the difficulty behind the Dixmier conjecture.

How does this relate to defining a module?

An $A_n$ -module structure on a $K$ -vector space $M$ is the same thing as a homomorphism $A_n → End_K(M)$ , so by the universal property it is the same as $2n$ linear operators on $M$ satisfying the relations. That is the systematic version of the ad hoc constructions used throughout Chapter 5.

Can one present

A_n

with fewer generators?

$A_1$ is generated as a $K$ -algebra by $x$ and $∂$ and by no single element, since it is non-commutative. There are presentations using different generating sets - for instance $x^2, ∂^2$ and $x∂$ , which span a copy of the Lie algebra $sl_2$ inside $A_1$ - but these generate a proper subalgebra, not all of $A_1$ , so they do not give an alternative presentation of the whole algebra.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 1 §3, Theorem (1.3.1) and Corollary (1.3.2), with the exercises of §4.
J. Dixmier, Sur les algebres de Weyl, Bulletin de la Societe Mathematique de France 96 (1968), 209-242 - automorphisms of $A_1$ and the endomorphism problem.
L. Makar-Limanov, On automorphisms of Weyl algebra, Bulletin de la Societe Mathematique de France 112 (1984), 359-363.
G. M. Bergman, The diamond lemma for ring theory, Advances in Mathematics 29 (1978), 178-218 - when a presentation has the expected monomial basis.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Graduate Studies in Mathematics 30, revised edition, American Mathematical Society, 2001 - Ch. 1, for presentations of the Weyl algebra and iterated Ore extensions.
M. Born, W. Heisenberg and P. Jordan, Zur Quantenmechanik II, Zeitschrift fur Physik 35 (1926), 557-615 - the origin of the relations.
V. Levandovskyy and H. Schonemann, Plural - a computer algebra system for noncommutative polynomial algebras, Proceedings of ISSAC 2003, ACM, 176-183 - $G$ -algebras and their non-degeneracy conditions.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Verify directly that the Fourier automorphism $F$ satisfies $F^4 = id$ on the generators of $A_n$ .
Write out the triangular automorphism of $A_3$ obtained from $m = 2$ , $f_1 = x_3$ , $f_2 = x_3^2$ , and check all nine relations.
Show that the map $SL_2(K) → Aut(A_1)$ of the example is an injective group homomorphism.
Explain why the composition of two triangular automorphisms need not be triangular, and give an example.
Prove that any non-zero $K$ -algebra homomorphism out of $A_n$ is injective, and identify exactly where simplicity is used.
Describe the kernel of the map from the presented algebra to the operator algebra over a field of characteristic $3$ , for $n = 1$ .
Use the universal property to construct the transposition anti-automorphism of $A_n$ sending $x_i$ to $x_i$ and $∂_i$ to $-∂_i$ , and say why it is an anti-homomorphism rather than a homomorphism.

The Weyl Algebra by Generators and Relations

Overview

Definition

Core Concepts

What a presentation buys you

Why the theorem is not automatic

The shape of the proof

Construction and Proof

Step 1: the relations are satisfied in the Weyl algebra

Step 2: ordered monomials span the quotient

Step 3: comparing dimensions

Key Equations

Variable Definitions

Properties and Behaviour

Examples and Special Cases

Worked Example

Applications and Industry Use

Computational Notes

Limits of Validity

Failure Modes and Common Mistakes

Historical Notes

Comparison

Key Takeaways

Key points

FAQs

Related Engineering Topics

References

AI Suggested Questions

Continue learning