The Weyl Algebra Defined as a Ring of Operators

Q: Is the polynomial ring the only module one could have used?

No. Any faithful A_n -module realises A_n inside its endomorphism algebra, and in characteristic zero every non-zero module is faithful because A_n is simple. Formal power series, convergent power series and C^ ∈ fty functions all work. K[X] is chosen because it is the smallest and the most computable.

Q: Why does the definition insist on characteristic zero?

In characteristic p the operator ∂_i^,p acts as zero on K[X] , so the operator algebra satisfies a relation that the abstractly presented Weyl algebra does not. The operator algebra is then a proper quotient, it is not simple, and it is finitely generated as a module over its centre. Essentially every theorem in this collection fails or changes.

Q: Is A_n a finitely generated algebra?

Yes — by 2n elements, by definition. It is infinite dimensional as a vector space but finitely generated as an algebra, and moreover Noetherian on both sides. Finite generation as an algebra and finite dimension as a vector space are different conditions, and only the first holds here.

Q: Where does the identity operator come from in [∂, x] = 1?

From the product rule. ∂(xf) = x,∂ f + f , and the extra f is the value of the identity operator at f . It is not a normalisation choice: the constant is forced to be 1 by the definition of the derivative.

Q: Is ∂_i a ring homomorphism of K[X]?

No. It is K -linear but satisfies the Leibniz rule ∂_i(fg) = f,∂_i g + g,∂_i f rather than ∂_i(fg) = ∂_i(f)∂_i(g) . Maps of this kind are derivations , and the general theory of differential operators is built from them.

Overview

There are two standard ways to introduce the Weyl algebra. One writes down $2n$ symbols and a list of relations they are declared to satisfy; the other exhibits $2n$ concrete operators on a concrete vector space and takes the algebra they generate. This page takes the second route, which is the one Coutinho opens with, and which has a decisive pedagogical advantage: nothing has to be assumed consistent. The operators exist, so the algebra exists, and every relation between them is a fact that can be checked by differentiating a polynomial.

The vector space acted on is the polynomial ring $K[X] = K[x_1, … ,x_n]$ over a field $K$ of characteristic zero. The operators are the $n$ multiplications $f ↦ x_i f$ and the $n$ partial derivatives $f ↦ ∂ f / ∂ x_i$ . The $n$ -th Weyl algebra $A_n$ is the subalgebra of $End_K(K[X])$ that these $2n$ operators generate. Multiplication in $A_n$ is composition of operators, and it is not commutative: differentiating after multiplying is not the same as multiplying after differentiating, and the discrepancy is exactly the identity operator.

Two consequences are worth flagging immediately. First, $K[X]$ is by construction a faithful module over $A_n$ : an element of $A_n$ is an operator, so it is zero precisely when it kills every polynomial. Second, $A_n$ is genuinely infinite dimensional over $K$ , because $K[X]$ is, and this is not an accident that a cleverer construction could avoid — the commutation relations make finite-dimensional realisations impossible.

The companion page The Weyl Algebra by Generators and Relations gives the abstract presentation and proves the two definitions agree in characteristic zero. They do not agree in characteristic $p$ , a point taken up under Limits of Validity below.

Definition

Throughout, $K$ is a field of characteristic zero and $K[X] = K[x_1, … ,x_n]$ is the polynomial ring in $n$ commuting indeterminates. As a $K$ -vector space $K[X]$ has the countable basis $x^α : α ∈ N^n$ of monomials, so it is infinite dimensional. Write $End_K(K[X])$ for the $K$ -algebra of all $K$ -linear maps $K[X] → K[X]$ , with addition defined pointwise and multiplication given by composition.

The Weyl algebra $A_n$ Coutinho, Ch. 1 §1

For each $i$ with $1 ≤ i ≤ n$ define two elements of $End_K(K[X])$ by their effect on a polynomial $f$ :

x_i(f) = x_i f, ∂_i(f) = (∂ f)/(∂ x_i).

Both maps are $K$ -linear: multiplication by a fixed polynomial is linear, and partial differentiation is linear because $K$ has characteristic zero and the derivative of a sum is the sum of the derivatives. The $n$ -th Weyl algebra $A_n = A_n(K)$ is the $K$ -subalgebra of $End_K(K[X])$ generated by $x_1, … , x_n,∂_1, … ,∂_n$ — that is, the set of all finite $K$ -linear combinations of finite composites of these $2n$ operators, together with the identity operator. By convention $A_0 = K$ .

Remark

The hat on $x_i$ distinguishes the operator \"multiply by $x_i$ \" from the polynomial $x_i$ on which operators act. Keeping them apart is useful for exactly one paragraph. From here on both are written $x_i$ , following the universal convention; the context always makes clear which is meant, and for $n = 1$ the generators are written simply $x$ and $∂$ .

Nested Weyl algebras

If $m ≤ n$ then the operators generating $A_m$ act on $K[x_1, … ,x_n]$ as well, by ignoring the extra variables, and the resulting embedding realises $A_m$ as a subalgebra of $A_n$ . The Weyl algebras therefore form a chain

K = A_0 ⊂ eq A_1 ⊂ eq A_2 ⊂ eq ⋅ s ⊂ eq A_n ⊂ eq ⋅ s

Each inclusion is strict, and each $A_n$ with $n ≥ 1$ is non-commutative.

Note

Being a subalgebra of an endomorphism algebra, $A_n$ comes with a distinguished module for free: $K[X]$ itself. This is the fundamental or natural representation. It is faithful by construction, and it turns out to be simple — it has no submodules other than $0$ and itself.

Core Concepts

Why start with operators

Presenting an algebra by generators and relations always raises a question of consistency: do the relations collapse the algebra further than intended? A presentation with contradictory relations defines the zero ring, and nothing in the list of relations announces this. Defining $A_n$ as a set of honest operators sidesteps the question entirely. The algebra is a subset of something that already exists, its elements are distinguishable because they act differently on polynomials, and every relation among the generators is discovered by computation rather than imposed.

The price is that the size of $A_n$ is not obvious; that is settled by the canonical form theorem, which says the operators $x^α∂^β$ are linearly independent. Under the abstract presentation the roles are reversed: the size is transparent and the consistency needs proof.

Multiplication and differentiation do not commute

The single computation that generates all the structure is the product rule. Applying $∂_i$ after multiplying by $x_i$ gives

∂_i (x_i f) = x_i (∂ f)/(∂ x_i) + f,

whereas multiplying by $x_i$ after differentiating gives only the first term. The two composites therefore differ by the identity operator. Writing $[P,Q] = PQ - QP$ for the commutator, this reads $[∂_i, x_i] = 1$ . The full list of relations, and what follows from them, is the subject of the commutation relations page.

Differentiation is a derivation, not a ring map

The operator $∂_i$ does not respect the multiplication of $K[X]$ ; it satisfies the Leibniz rule instead. The general statement is that $∂_i$ is a derivation of $K[X]$ over $K$ , and the Weyl algebra can be described as the algebra generated inside $End_K(K[X])$ by $K[X]$ acting on itself by multiplication together with all $K$ -derivations of $K[X]$ . That formulation is the one that generalises: replacing $K[X]$ by the coordinate ring of a smooth affine variety produces the ring of differential operators on that variety.

Infinite dimension is forced

One might hope to model the same relations on a finite-dimensional space, where linear algebra is easier. It is impossible in characteristic zero: taking traces in $[∂, x] = 1$ would give $0 = V$ , since the trace of a commutator vanishes while the trace of the identity on a $d$ -dimensional space is $d$ . The polynomial ring is the smallest natural home for these operators, and no finite-dimensional home exists.

Key Equations

The definition of the generating operators:

x_i : K[X] → K[X], f ↦ x_i f; ∂_i : K[X] → K[X], f ↦ (∂ f)/(∂ x_i).

(1.1)

The algebra they generate:

A_n = A_n(K) = K-subalgebra of End_K(K[X]) generated by x_1, … , x_n,∂_1, … ,∂_n.

(1.2)

The product rule, written as a statement about composites of operators:

(∂_i ∘ x_j)(f) = x_j (∂ f)/(∂ x_i) + δ_ij f = ( x_j ∘ ∂_i)(f) + δ_ij f,

(1.3)

valid for every $f ∈ K[X]$ , where $δ_ij$ is the Kronecker delta.

Equivalently, in commutator form, and this is the defining relation of the algebra:

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

(1.4)

More generally, commuting a derivative past multiplication by any polynomial produces its derivative, which is the operator form of the Leibniz rule:

[∂_i, f] = (∂ f)/(∂ x_i) (f ∈ K[X]).

(1.5)

Repeated use of (1.5) pushes every multiplication operator to the left of every derivative, which yields the canonical form; every element of $A_n$ can be written uniquely as

D = ∑ _α, β ∈ N^n c_αβ, x^α ∂^β, c_αβ ∈ K, almost all zero.

(1.6)

Variable Definitions

$K$: the ground field, always of characteristic zero on this page
$K[X]$: the polynomial ring $K[x_1, … ,x_n]$ , viewed as a $K$ -vector space of infinite dimension
$End_K(K[X])$: the $K$ -algebra of all $K$ -linear maps $K[X] → K[X]$ under addition and composition
$x_i$: the operator of multiplication by the variable $x_i$ ; written $x_i$ once the notation is settled
$∂_i$: the operator of partial differentiation with respect to $x_i$
$A_n$ , $A_n(K)$: the $n$ -th Weyl algebra over $K$ ; the subalgebra of $End_K(K[X])$ generated by the $2n$ operators above
$[P,Q]$: the commutator $PQ - QP$ of two operators
$δ_ij$: the Kronecker delta: $1$ when $i = j$ and $0$ otherwise
$α, β$: multi-indices in $N^n$ , with $x^α = x_1^α_1 ⋅ s x_n^α_n$ and $∂^β = ∂_1^β_1 ⋅ s ∂_n^β_n$
$1$: the identity operator, which is the multiplicative identity of the algebra

Properties and Behaviour

Everything below is a property of $A_n$ as it has just been defined; each is established on its own page, and the list is here so that the shape of the object is visible at the outset.

Canonical formCoutinho (2.1)

Every element of $A_n$ can be written in exactly one way as a finite sum $∑ c_αβ x^α ∂^β$ . In particular the set $x^α ∂^β : α,β ∈ N^n$ is a $K$ -basis of $A_n$ , so $_K A_n$ is countably infinite.

Ring-theoretic properties

$A_n$ has no zero divisors, and its centre is exactly the field of scalars $K$ . It is simple: its only two-sided ideals are $0$ and $A_n$ . It is left and right Noetherian, and it is a domain but not a division ring — the operator $∂_1$ , for instance, has no inverse in $A_n$ .

Every non-zero module is faithful

Simplicity has an immediate consequence for the definition given here: because a two-sided ideal is either $0$ or everything, every non-zero $A_n$ -module is faithful. So the faithfulness of $K[X]$ , which came free from the construction, was never going to be a restrictive condition in characteristic zero.

$A_n$ is an $n$ -fold external product

Composing the natural inclusions gives an isomorphism of $K$ -algebras

A_n ≅ A_1 ⊗ _K A_1 ⊗ _K ⋅ s ⊗ _K A_1 (n factors),

so all of the structure of $A_n$ is generated by the one-variable case together with the external product. Correspondingly $K[x_1, … ,x_n] ≅ K[x_1] ⊗ _K ⋅ s ⊗ _K K[x_n]$ as a module.

Agreement with the intrinsic definition

$A_n$ coincides with the ring $D(K[X])$ of differential operators on affine $n$ -space in the sense of Grothendieck. So the concrete definition given here, which looks like a choice of convenient generators, in fact produces the whole of an intrinsically defined object; see the comparison page.

Examples and Special Cases

The degenerate case $n = 0$

$A_0 = K$ by convention: there are no variables, $K[X] = K$ , and the only operators are the scalars.

The first Weyl algebra

$A_1$ is generated by $x$ and $∂$ acting on $K[x]$ , with the single relation $∂ x - x ∂ = 1$ . Its canonical basis is $x^a ∂^b : a, b ≥ 0$ . Every ordinary linear differential operator with polynomial coefficients, such as the Bessel operator $x^2∂^2 + x∂ + (x^2 - ν^2)$ for $ν ∈ K$ , is an element of $A_1$ .

The Euler operator

The Euler operator $θ = x∂ ∈ A_1$ acts diagonally on the monomial basis: $θ(x^k) = k x^k$ . It is the simplest interesting element of $A_1$ that is neither a multiplication nor a derivative, and it satisfies $[θ, x] = x$ and $[θ, ∂] = -∂$ , so conjugation by it grades $A_1$ . In $n$ variables, $θ = ∑ _i x_i ∂_i$ measures total degree.

$A_n$ is not all of $End_K(K[X])$

$A_n$ is a very small part of $End_K(K[X])$ . Every $D ∈ A_n$ raises degrees by a bounded amount: if $D = ∑ c_αβ x^α ∂^β$ and $c = |α| : c_αβ ≠ 0$ , then $D(f) ≤ f + c$ for every $f$ . So the $K$ -linear map $T$ defined on the monomial basis of $K[x]$ by $T(x^k) = x^2k$ lies in $End_K(K[x])$ but not in $A_1$ : it raises the degree of $x^k$ by $k$ , which is unbounded. The inclusion $A_n ⊂ eq End_K(K[X])$ is very far from an equality — indeed $A_n$ has countable dimension over $K$ while $End_K(K[X])$ does not.

Worked Example

Composing operators in $A_1$ and reducing to canonical form

Step 1 - fix the action on the monomial basis
Work in $A_1$ over $K = Q$ , acting on $K[x]$ . On the basis $1, x, x^2, …$ the two generators act by

$x ⋅ x^k = x^k+1, ∂ ⋅ x^k = k,x^k-1 (k ≥ 1), ∂ ⋅ 1 = 0.$

Every computation below is carried out on this basis and then extended by linearity, which is legitimate because all the operators involved are $K$ -linear.
Step 2 - verify the commutation relation directly
Apply $∂ x$ and $x∂$ to $x^k$ :

$∂(x ⋅ x^k) = ∂(x^k+1) = (k+1)x^k, x(∂ ⋅ x^k) = x ⋅ k x^k-1 = k x^k.$

Subtracting, $(∂ x - x∂)(x^k) = (k+1)x^k - k x^k = x^k$ for every $k ≥ 0$ (for $k = 0$ the second term is $0$ and the first is $1$ ). The two operators therefore differ by the identity on a basis, hence everywhere, so $[∂, x] = 1$ . Note that this holds on $x^0 = 1$ as well, which is where a careless computation usually goes wrong.
Step 3 - reduce a composite to canonical form
Take $D = ∂^2 x^2$ , a composite written with the derivatives on the left. Push the multiplications leftwards using $∂ x = x∂ + 1$ twice. First,

$∂ x^2 = (∂ x) x = (x∂ + 1)x = x(∂ x) + x = x(x ∂ + 1) + x = x^2 ∂ + 2x.$

Now apply $∂$ on the left of that identity:

$∂^2 x^2 = ∂(x^2∂ + 2x) = (∂ x^2)∂ + 2(∂ x) = (x^2∂ + 2x)∂ + 2(x∂ + 1),$

which collects to

$∂^2 x^2 = x^2 ∂^2 + 4x∂ + 2.$
(1.7)

The right-hand side is in canonical form: every term is a scalar times $x^a ∂^b$ with the multiplications to the left.
Step 4 - check the answer on test polynomials
Apply both sides of (1.7) to $f = x^3$ . Left-hand side: $x^2 f = x^5$ , and $∂^2 x^5 = 5 ⋅ 4, x^3 = 20x^3$ . Right-hand side:

$x^2 ∂^2 (x^3) = x^2 ⋅ 6x = 6x^3, 4x∂(x^3) = 4x ⋅ 3x^2 = 12 x^3, 2 ⋅ x^3 = 2x^3,$

and $6 + 12 + 2 = 20$ . Now test $f = 1$ : the left-hand side gives $∂^2(x^2) = 2$ , while the right-hand side gives $0 + 0 + 2 = 2$ . And $f = x$ : left-hand side $∂^2(x^3) = 6x$ ; right-hand side $x^2 ⋅ 0 + 4x ⋅ 1 + 2x = 6x$ . All three agree.
Step 5 - upgrade the check to a proof
Finitely many monomials are not enough, but the general one is: both sides of (1.7) applied to $x^k$ give $∂^2 x^k+2 = (k+2)(k+1)x^k$ on the left and $k(k-1)x^k + 4kx^k + 2x^k = (k^2+3k+2)x^k$ on the right, and these agree identically in $k$ . Since elements of $A_1$ are operators, agreement on a basis of $K[x]$ is equality.

Result

$∂^2 x^2 = x^2∂^2 + 4x∂ + 2$ in $A_1$ , verified as an identity of operators on the general monomial $x^k$ : both sides multiply $x^k$ by $(k+1)(k+2)$ . The reduction used only the relation $[∂, x] = 1$ , which was itself verified from the definition of the operators.

Applications and Industry Use

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

The operator definition is the one that connects most directly to uses outside pure algebra, because in every one of them the object of interest genuinely is an operator.

Quantum mechanics. The canonical commutation relation $[ p, q] = -iℏ$ is the relation $[∂, x] = 1$ after rescaling, and $A_n$ is the algebraic skeleton of the Heisenberg picture. The impossibility of a finite-dimensional model is the algebraic content of the statement that position and momentum cannot both be represented by finite matrices. See the Weyl algebra and the birth of quantum mechanics.
Linear differential equations with polynomial coefficients. A system of such equations is an element or a matrix of elements of $A_n$ , and its solution theory is the study of the module it presents. This is the entry point to the whole of D-module theory.
Computer algebra. Symbolic integration and summation algorithms manipulate elements of $A_n$ and its difference analogues (Ore algebras) directly. Canonical form is what makes the data structure and the equality test possible.
Combinatorial identity proving. Holonomic sequences and functions are those annihilated by large enough ideals of a Weyl or Ore algebra, and Zeilberger's method runs entirely inside such algebras.
Control and systems theory. Linear time-varying systems with polynomial coefficients are modules over $A_1$ or $B_1$ ; controllability and observability translate into module-theoretic properties such as freeness and torsion-freeness.

Design Considerations

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

Operators or symbols

The operator definition buys existence and faithfulness for free and costs a proof that the algebra is as large as expected; the abstract presentation buys the size for free and costs a proof that it is not smaller. Coutinho does both, reconciled by the canonical basis theorem. For everyday work the presentation is more flexible: constructing a homomorphism out of $A_n$ becomes a matter of checking finitely many relations, which is exactly how automorphisms of P1 are built.

Which module to act on

For algebra, $K[X]$ is the right choice: smallest faithful module, transparent structure. For analysis the natural modules are smooth or holomorphic functions, or distributions, each detecting different solutions of the same operator. The algebra is fixed; the module is where the information lives.

Left or right

Because $A_n$ is non-commutative, left and right modules are different categories. The convention throughout this collection is left modules, matching the way operators act on functions. The two theories are equivalent via the transposition anti-automorphism, which sends $x_i ↦ x_i$ and $∂_i ↦ -∂_i$ , so nothing is lost — but signs are, if the transposition is applied carelessly.

Standards and Codes

For mathematics, the relevant standards are notation, numeric and markup standards together with reference implementations.

ISO 80000-2 fixes the notation used here: $∂$ for partial differentiation, $ℕ$ and $ℚ$ for the number systems, upright roman type for operator names such as $dim$ and $deg$ , and $δ_ij$ for the Kronecker delta.
ISO/IEC 40314 (MathML 3.0) is the markup in which the expressions on this page are encoded, so the mathematics is machine-readable rather than an image.
There is no standard for the name: $A_n$ , $A_n(K)$ , $D(A^n)$ , $D_n$ and $W_n$ all appear in the literature for the same algebra. $W_n$ is the most common alternative and is standard in the Russian tradition; $D_n$ often denotes the formal or analytic version instead, so it is worth checking.
Software conventions: Macaulay2 constructs the algebra with makeWeylAlgebra, Singular with Weyl() in nctools.lib, and SageMath with DifferentialWeylAlgebra. Each normalises products to a fixed side and each documents which.
Coutinho's own numbering — results cited as section-and-item pairs within a chapter — is the citation convention used in the References below.

Material Selection

The material of a mathematical construction is its numeric substrate: the ground field, the coefficient ring and the representation used to store it.

The only genuine choice at this stage is the ground field, and characteristic is the property that matters.

$ℚ$ is the default for exact computation. Everything on this page is true over $ℚ$ , and results proved there transfer to any extension, since $A_n(K) ⊗ _K L ≅ A_n(L)$ .
$ℂ$ is the field to use when the module of interest is a space of holomorphic functions or distributions, or when monodromy is in play.
$ℝ$ behaves exactly like $ℚ$ for all structural purposes; it is chosen only when the application is real-analytic.
Fields of characteristic $p$ change the object. The operator algebra defined above is then not the Weyl algebra of the abstract presentation, and neither one is simple.
Rings that are not fields are not admissible here without care. Over a commutative ring $R$ containing $ℚ$ one can define $A_n(R)$ , but simplicity fails immediately: any ideal of $R$ generates a two-sided ideal.

For the module, the monomial basis of $K[X]$ is the standard storage choice because it makes both generators sparse: $x_i$ shifts one exponent up, $∂_i$ shifts one down and multiplies by an integer.

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.

An element of $A_n$ is stored as its canonical form: a sparse list of pairs $(α, β) ∈ N^n × N^n$ with non-zero coefficients in $K$ . Addition is merging of sparse lists. Multiplication is the only non-trivial operation.

To multiply $x^α ∂^β$ by $x^γ ∂^δ$ , the block $∂^β x^γ$ must be normalised.
The rule is the Leibniz expansion $∂^β x^γ = ∑ _κ βκ γκ κ!; x^γ - κ ∂^β - κ$ , the sum running over multi-indices $κ ≤ (β,γ)$ componentwise.
Multiply through by $x^α$ on the left and $∂^δ$ on the right and merge terms.
Equality of two elements is then coefficient-by-coefficient comparison, which is what the canonical form is for.

One monomial product expands into $∏ _i( (β_i,γ_i)+1)$ terms, so the cost is exponential in $n$ in the worst case; for $n=1$ it is at most $(β,γ)+1$ terms. In practice coefficient growth over $ℚ$ bites before the term count does.

Implementations that expose $A_n$ directly include the Dmodules package and WeylAlgebra type in Macaulay2, Weyl algebras in Singular via dmod.lib and nctools.lib, the ore_algebra package in SageMath, and HolonomicFunctions.m in Mathematica. All of them normalise to a fixed side; consult the documentation before comparing output with a formula written elsewhere.

Limits of Validity

The operator definition is robust, but three hypotheses in it are load-bearing.

Characteristic zero. In characteristic $p > 0$ the operator $∂_i^,p$ annihilates $K[X]$ , because $∂^p(x^k) = k(k-1) ⋅ s(k-p+1)x^k-p$ and one of any $p$ consecutive integers is divisible by $p$ . So the algebra of operators generated by $x_i$ and $∂_i$ is a proper quotient of the algebra defined by generators and relations, and the two definitions of the Weyl algebra part company. See the positive characteristic page.
The module must be $K[X]$ , or something at least as large. The construction realises $A_n$ inside $End_K(V)$ for $V = K[X]$ . Any faithful $A_n$ -module would do equally well — formal power series $K[[X]]$ , or holomorphic functions on a domain when $K = C$ — and in characteristic zero every non-zero module is faithful, so the resulting algebra is the same. What is not available is a finite-dimensional $V$ .
Polynomial coefficients only. Enlarging the coefficients to rational functions, formal power series or convergent series gives different rings — $B_n$ , $D_n$ and so on — with genuinely different behaviour. The rational-coefficient ring, for instance, has a division-ring-like localisation theory that $A_n$ lacks.

What is not needed

Nothing above requires $K$ to be algebraically closed, and nothing requires it to be $ℂ$ . All the structural facts listed under Properties hold over any field of characteristic zero, including $ℚ$ . Extension of scalars is harmless: $A_n(K) ⊗ _K L ≅ A_n(L)$ for any field extension $L/K$ .

Failure Modes and Common Mistakes

Treating the canonical basis as a commuting monomial basis

$A_n$ is not the polynomial ring $K[x_1, … ,x_n,∂_1, … ,∂_n]$ , even though its canonical basis $x^α∂^β$ looks exactly like the monomial basis of that polynomial ring. The underlying vector spaces are isomorphic; the multiplications are not. In $A_1$ , $∂ x = x∂ + 1 ≠ x ∂$ . The polynomial ring is the associated graded algebra of $A_n$ , which is a genuinely weaker piece of data — see the associated graded algebra.

Dropping the order of composition

Writing $∂^2 x^2$ and $x^2∂^2$ for the same operator. They differ by $4x∂ + 2$ , as computed above. Order matters at every step of a composite, and the safest habit is to reduce every expression to canonical form immediately, so that two elements can be compared coefficient by coefficient.

Assuming everyone orders the canonical form the same way

The mathematical convention used here puts all multiplication operators to the left. Much of the physics literature, and some computer algebra systems, use the opposite (normal ordering, derivatives to the left) or the Weyl-symmetric ordering. The same operator then has different coefficients. Before comparing a formula with a source or a package, check which side the coefficients sit on: for example $x∂ = ∂ x - 1$ , so a symbol that reads $ξ η$ in one convention need not correspond to the same element in another.

Confusing $A_n$ with the ambient endomorphism algebra

$End_K(K[X])$ is enormous and badly behaved — not Noetherian, full of idempotents and zero divisors — while $A_n$ is a Noetherian simple domain. Facts about the ambient algebra do not transfer. In particular $A_n$ has no idempotents other than $0$ and $1$ , so there is no projection available to split $K[X]$ as an $A_n$ -module.

Carrying characteristic-zero statements into characteristic p

In characteristic $p$ the phrase \"the Weyl algebra\" is ambiguous and the two standard definitions give non-isomorphic answers. If a source states results \"over a field $K$ \" without saying characteristic zero, and uses the operator definition, treat every simplicity or faithfulness claim with suspicion until the characteristic is pinned down.

Historical Notes

The relations came before the algebra. In 1925 and 1926 Heisenberg, Born and Jordan, and then Dirac, arrived at the canonical commutation relation between position and momentum, and Dirac in particular framed it as an algebraic condition on operators. It was immediately clear that the operators could not be finite matrices.

Hermann Weyl, in Gruppentheorie und Quantenmechanik (1928), studied the exponentiated form of the relations and their unitary representations — the material that became the Stone–von Neumann theorem — and the algebra generated by the unexponentiated operators was eventually named after him. It appeared in a purely algebraic setting as early as Littlewood's 1933 work on non-commutative algebras, and its simplicity over fields of characteristic zero was established in that algebraic tradition.

The name \"Weyl algebra\" became standard after Dixmier's 1968 paper Sur les algèbres de Weyl, which set out the structure theory used ever since and posed the conjecture that every endomorphism of $A_n$ is an automorphism — still open, and now known to be equivalent to the Jacobian conjecture. Shortly afterwards Bernstein's work on the analytic continuation of $f^s$ turned $A_n$ from an interesting example in ring theory into the foundation of a theory of differential equations, which is the subject of the rest of this collection.

Comparison

Three algebras that share the same underlying vector space or the same generators, and how they differ.
	$A_n$ (Weyl algebra)	$K[x_1, … ,x_n,ξ_1, … ,ξ_n]$	$End_K(K[X])$
Definition	operators generated by multiplications and derivatives	commutative polynomials in $2n$ variables	all $K$ -linear maps of $K[X]$
$_K$	countably infinite	countably infinite	uncountable
Commutative?	no, $[∂_i,x_i]=1$	yes	no
Zero divisors	none	none	many
Two-sided ideals	only $0$ and the whole ring	many	many
Noetherian	yes, left and right	yes	no
Relation to the others	—	the associated graded ring of $A_n$	contains $A_n$ as a tiny subalgebra

Key Takeaways

Key points

$A_n$ is defined as the subalgebra of $End_K(K[x_1, … ,x_n])$ generated by the $n$ multiplications $f ↦ x_i f$ and the $n$ partial derivatives $f ↦ ∂ f/∂ x_i$ , with $K$ of characteristic zero.
Because the elements are operators, the algebra exists without any consistency check and $K[X]$ is automatically a faithful module.
The product rule gives $[∂_i, x_j] = δ_ij$ and $[x_i,x_j] = [∂_i,∂_j] = 0$ ; these relations generate all others.
Every element has a unique canonical form $∑ c_αβx^α∂^β$ , so $x^α∂^β$ is a $K$ -basis and $A_n$ is countably infinite dimensional.
$A_n$ is a Noetherian simple domain with centre $K$ ; it is very much smaller than the ambient $End_K(K[X])$ and very different from the commutative polynomial ring in $2n$ variables.
No finite-dimensional vector space supports these relations in characteristic zero, and in characteristic $p$ the operator definition and the abstract presentation give non-isomorphic algebras.

FAQs

Why define

A_n

inside

End_K(K[X])

instead of just listing relations?

Because then existence is free. A presentation by generators and relations might, for all one knows at the outset, define the zero ring; proving otherwise requires exhibiting a model, and the model is exactly the operator algebra. Coutinho gives the operator definition first for this reason and only afterwards proves the abstract presentation describes the same object.

Is the polynomial ring the only module one could have used?

No. Any faithful $A_n$ -module realises $A_n$ inside its endomorphism algebra, and in characteristic zero every non-zero module is faithful because $A_n$ is simple. Formal power series, convergent power series and $C^ ∈ fty$ functions all work. $K[X]$ is chosen because it is the smallest and the most computable.

Does

A_n

contain the polynomial ring

K[X]

Yes, as the subalgebra of multiplication operators, and that copy is commutative. It also contains the commutative subalgebra $K[∂_1, … ,∂_n]$ of constant-coefficient operators. What it does not do is decompose as a tensor product of these two as an algebra — only as a vector space, which is precisely the content of the canonical form.

Why does the definition insist on characteristic zero?

In characteristic $p$ the operator $∂_i^,p$ acts as zero on $K[X]$ , so the operator algebra satisfies a relation that the abstractly presented Weyl algebra does not. The operator algebra is then a proper quotient, it is not simple, and it is finitely generated as a module over its centre. Essentially every theorem in this collection fails or changes.

A_n

a finitely generated algebra?

Yes — by $2n$ elements, by definition. It is infinite dimensional as a vector space but finitely generated as an algebra, and moreover Noetherian on both sides. Finite generation as an algebra and finite dimension as a vector space are different conditions, and only the first holds here.

Where does the identity operator come from in

[∂, x] = 1

From the product rule. $∂(xf) = x,∂ f + f$ , and the extra $f$ is the value of the identity operator at $f$ . It is not a normalisation choice: the constant is forced to be $1$ by the definition of the derivative.

Can I think of

x

and

∂

as matrices?

Only as infinite matrices. With respect to the monomial basis of $K[x]$ , $x$ is the shift matrix with $1$ s on one subdiagonal and $∂$ is the shift matrix with $1, 2, 3, …$ on the other. These are honest infinite matrices with finitely many non-zero entries in each column, and their commutator is the identity. No finite truncation of them satisfies the relation.

∂_i

a ring homomorphism of

K[X]

No. It is $K$ -linear but satisfies the Leibniz rule $∂_i(fg) = f,∂_i g + g,∂_i f$ rather than $∂_i(fg) = ∂_i(f)∂_i(g)$ . Maps of this kind are derivations, and the general theory of differential operators is built from them.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 1 §1 for the definition as a ring of operators, §2 for the canonical form and §3 for the abstract presentation.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - the paper that fixed the name and the basic structure theory.
H. Weyl, Gruppentheorie und Quantenmechanik, Hirzel, Leipzig, 1928; English translation The Theory of Groups and Quantum Mechanics, Dover, 1950.
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, North-Holland, 1979 - Ch. 1, for the filtered and analytic versions of the same algebra.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Graduate Studies in Mathematics 30, revised edition, American Mathematical Society, 2001 - Ch. 1 and Ch. 8, for $A_n$ as the standard example of a simple Noetherian domain.
R. Hotta, K. Takeuchi and T. Tanisaki, D-modules, Perverse Sheaves, and Representation Theory, Progress in Mathematics 236, Birkhäuser, 2008 - Ch. 1, for the sheaf-theoretic version.
Dmodules: a Macaulay2 package for D-modules, and ore_algebra: a SageMath package for Ore algebras - package documentation, for the computational conventions cited above.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.
ISO/IEC 40314:2016, Information technology - Mathematical Markup Language (MathML) Version 3.0, International Organization for Standardization.

AI Suggested Questions

Reduce $∂^3 x^3$ to canonical form in $A_1$ and verify the answer on the general monomial $x^k$ .
Show that the operator $T(x^k) = x^k+1$ on $K[x]$ lies in $A_1$ but $T(x^k) = x^2k$ does not.
Prove that $A_1$ has no non-zero idempotents, using the canonical form.
Write the infinite matrices of $x$ and $∂$ in the monomial basis and check that their commutator is the identity matrix.
Explain why $∂^p$ acts as zero on $K[x]$ when $K$ has characteristic $p$ , and what that does to the definition.
Show that the Euler operator $θ = x∂$ satisfies $[θ, x^k] = k x^k$ and use it to grade $A_1$ .
Explain why the trace argument rules out finite-dimensional representations, and what changes in characteristic $p$ .