The Weyl Algebra and the Birth of Quantum Mechanics

Q: Did Heisenberg know he was using matrices?

Not at first. He derived a multiplication rule for his arrays from the correspondence principle and was troubled that it depended on the order of the factors. Born identified the rule as matrix multiplication, which he knew from his mathematical training, and that recognition is what turned the calculation into a formalism.

Q: Why is the constant on the right-hand side imaginary?

Because p and q are required to be self-adjoint, and the commutator of two self-adjoint operators is anti-self-adjoint. A real constant is therefore impossible on physical grounds. The algebraic theory drops self-adjointness, so it can and does normalise the constant to 1 .

Q: Is the Weyl algebra the same as the Heisenberg algebra?

No, though they are closely related. The Heisenberg Lie algebra is the (2n+1) -dimensional Lie algebra spanned by the p_i , q_i and a central element z . The Weyl algebra is the quotient of its universal enveloping algebra by the ideal setting z = 1 . The Lie algebra is finite dimensional; the Weyl algebra is not.

Q: Where does Weyl's own contribution actually enter?

In the exponentiated relations (Q.6) and their reading as a projective representation of a group. That form is what makes the Stone-von Neumann theorem provable, since it replaces unbounded operators by unitary ones, and it is the basis of the Heisenberg group approach to harmonic analysis.

Q: Can I see the classical limit inside the algebra?

Only after filtering. Assign degrees to the generators and pass to the associated graded algebra: the images of x_i and ∂_i commute there, so the graded object is a polynomial ring in 2n variables - the functions on classical phase space. The commutator's leading term descends to the Poisson bracket. That is the algebraic form of ℏ → 0 .

Q: Do the relations for several degrees of freedom add anything new?

Structurally, A_n is the tensor product of n copies of A_1 over K , so nothing new appears at the level of the ring's generators. What is new is the geometry: with n ≥ 2 there is room for modules of intermediate dimension between n and 2n , and the whole theory of characteristic varieties becomes non-trivial.

Q: Why did it take until the 1960s for the algebra to get a name?

It was studied sporadically - by Littlewood in 1933, and by others in the context of operator theory - but it became a standard object only when noncommutative Noetherian ring theory and the theory of enveloping algebras made it a natural example. Dixmier's papers of 1963 and 1968 are the point at which A_n became a fixed piece of algebraic vocabulary.

Q: Is there a version of this story in positive characteristic?

There is an algebra with the same presentation, but it behaves entirely differently: x_i^p and ∂_i^p become central, the algebra is a finite module over its centre, it is not simple, and it has finite-dimensional representations. It is a legitimate object of study with applications in arithmetic geometry, but essentially none of the theorems on this page survive.

Overview

In the summer of 1925 Werner Heisenberg set out to build a mechanics of the atom from quantities that could actually be measured. Bohr's orbits were not among them; what spectroscopists observed were transitions, each attached to a pair of states rather than to one. Writing down a quantity indexed by two states means writing down an array, and arrays multiply in an order-dependent way. Heisenberg found this feature of his own calculation disturbing. It was, in fact, the discovery.

Within months Max Born recognised the arrays as matrices, and with Pascual Jordan he wrote down the relation that would define the subject: for position $q$ and momentum $p$ of a system with one degree of freedom,

pq - qp = (h)/(2π i), 1 = -iℏ, 1.

Two mathematical consequences followed immediately and have never gone away. First, no finite matrices satisfy this relation, so quantum mechanics is committed to infinite-dimensional objects from the outset. Second, once the physical constant is scaled away the relation $pq - qp = 1$ defines an associative algebra in its own right, with a rich and entirely mathematical life. Dirac called it the quantum algebra; Dixmier, forty years later, named it the Weyl algebra and wrote it $A_n$ .

This page is the historical entrance to the collection. It follows the relation from physics into algebra, states precisely what the physicists proved about it, separates that from what is often loosely attributed to them, and shows where the algebraic theory parts company with the Hilbert-space theory - a parting that happens sooner, and more sharply, than most accounts admit.

Definition

The physical relation and the algebraic one differ only by a normalisation, and the normalisation is worth doing explicitly once.

Canonical commutation relations

Operators $p_1, … ,p_n$ and $q_1, … ,q_n$ on a complex vector space satisfy the canonical commutation relations if

[,p_i, p_j,] = 0, [,q_i, q_j,] = 0, [,p_i, q_j,] = -iℏ,δ_ij.

Replacing $p_i$ by $∂_i = (i/ℏ),p_i$ and $q_i$ by $x_i$ removes the constant and gives $[,∂_i, x_j,] = δ_ij$ .

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

For a field $K$ of characteristic zero, $A_n(K)$ is the associative $K$ -algebra generated by $x_1, … ,x_n,∂_1, … ,∂_n$ subject only to

[x_i,x_j] = 0, [∂_i,∂_j] = 0, [∂_i,x_j] = δ_ij.

Equivalently, it is the algebra of differential operators with polynomial coefficients on affine $n$ -space. The identification of the abstract presentation with the concrete operator algebra is exactly the canonical basis theorem.

Signs and orders

Physics texts differ over whether the relation is written $[q,p] = iℏ$ or $[p,q] = -iℏ$ , and whether $ℏ$ or $h/2π i$ is used. All are the same statement. The mathematical convention adopted here, $[∂_i, x_j] = δ_ij$ , is fixed by the product rule: $∂(xg) - x∂ g = g$ . Nothing in the algebra depends on the choice, but a sign error propagates silently through every commutator computation, so it is worth fixing once.

Core Concepts

Why noncommutativity was forced, not chosen

A classical observable is a function on phase space and functions multiply commutatively. A quantum observable in Heisenberg's scheme is indexed by a pair of states, and the natural product - sum over intermediate states - is matrix multiplication. The noncommutativity is a consequence of the indexing by pairs, not an extra postulate. Dirac's Poisson bracket correspondence then pins the size of the failure to commute: $[a,b]$ must be $i ℏ$ times the classical bracket, so it is small but never zero. The classical limit $ℏ → 0$ recovers commutativity, and this survives in the algebra as the passage to the associated graded algebra, where the symbols commute and the leading part of the commutator becomes a Poisson bracket.

Why the matrices had to be infinite

Take traces in $pq - qp = c,1$ with $c ≠ 0$ . For finite square matrices $tr(pq) = tr(qp)$ , so the left side has trace zero while the right side has trace $c ⋅ N$ with $N$ the size. In characteristic zero that forces $c = 0$ . The relation therefore has no finite-dimensional solutions at all, and the physicists' conclusion in 1925 was correct as stated. The algebraic version of this argument shows more: no non-zero $A_n$ -module is finite dimensional over $K$ , which is why the representation theory is infinite dimensional throughout.

Differentiating with commutators

Dirac's observation that one can differentiate with respect to $p$ and $q$ inside the algebra is the statement that the two obvious partial derivatives are inner. Writing an element in canonical form $D = ∑ c_αβx^α∂^β$ , one has $[∂_i, D] = ∂ D/∂ x_i$ and $[x_i, D] = -∂ D/∂ ξ_i$ , where $ξ_i$ is treated as a commuting symbol standing for $∂_i$ . Both are formal derivatives of the coefficient polynomial. This turns questions about the algebra into questions about polynomials, and it is the mechanism behind the proofs that $A_n$ is simple and that its centre is just $K$ .

One relation, two very different theories

The physicists were interested in unitary representations on Hilbert space, and there the situation is rigid: the Stone-von Neumann theorem says there is essentially only one. Algebraists study modules over $A_n$ with no topology and no adjoints, and there the situation is wild: $A_1$ has enormous families of pairwise non-isomorphic simple modules. The same relation supports a uniqueness theorem in one category and a classification problem in another, and confusing the two is the commonest error made when moving between the physics and mathematics literature.

Key Equations

The relation as Born and Jordan wrote it, with $𝟏$ the identity matrix:

pq - qp = (h)/(2π i),1,

(Q.1)

and the Schrödinger representation on functions of one real variable, which satisfies it:

(q,ψ)(x) = x,ψ(x), (p,ψ)(x) = -iℏ,(dψ)/(dx).

(Q.2)

Normalising the constant away gives the relation used throughout this collection,

∂ x - x∂ = 1 in A_1, [∂_i, x_j] = δ_ij in A_n.

(Q.3)

Iterating (Q.3) gives the identity that does most of the routine work in the algebra:

∂^k x = x,∂^k + k,∂^k-1, ∂, x^k = x^k ∂ + k, x^k-1.

(Q.4)

The two inner derivations that formalise Dirac's differentiation are

[∂_i, D] = (∂ D)/(∂ x_i), [x_i, D] = -(∂ D)/(∂ ξ_i),

(Q.5)

acting on the canonical form $D = ∑ c_αβx^α∂^β$ , with $ξ_i$ the commuting symbol of $∂_i$ .

Weyl's exponentiated form of the relations, valid for the unitary operators built from $p$ and $q$ , reads

e^i s p, e^i t q = e^i s t ℏ, e^i t q, e^i s p (s,t ∈ R).

(Q.6)

Variable Definitions

$p$ , $q$: the momentum and position operators of the physical theory
$h$ , $ℏ$: Planck's constant and the reduced constant $ℏ = h/2π$
$x_i$: the $i$ -th coordinate, acting on functions by multiplication; the normalised $q_i$
$∂_i$: the $i$ -th partial derivative; the normalised $p_i$ , equal to $(i/ℏ)p_i$
$δ_ij$: the Kronecker delta
$[a,b]$: the commutator $ab - ba$
$A_n$: the $n$ -th Weyl algebra over a field of characteristic zero
$ξ_i$: a commuting symbol standing for $∂_i$ in the associated graded algebra
$ψ$: a wavefunction, an element of the space carrying the Schrödinger representation

Properties and Behaviour

Three precise statements separate what the relation does and does not force.

No finite-dimensional solutions

Let $V$ be a non-zero finite-dimensional vector space over a field of characteristic zero. There are no endomorphisms $P, Q$ of $V$ with $PQ - QP = id_V$ .

Proof. Taking traces, $tr(PQ) - tr(QP) = 0$ because $tr(AB) = tr(BA)$ for square matrices, while $tr(id_V) = V$ . Hence $V = 0$ in $K$ , and in characteristic zero that means $V = 0$ .

Characteristic zero is essential

Over a field of characteristic $p > 0$ the trace argument only says that $dim V$ is divisible by $p$ , and finite-dimensional solutions genuinely exist: take $V$ of dimension $p$ with basis $e_0, … ,e_p-1$ , let $Q$ shift $e_k ↦ e_k+1$ for $k < p-1$ and $e_p-1 ↦ 0$ , and let $P$ send $e_k ↦ k,e_k-1$ . The same computation as in characteristic zero gives $PQ - QP = id$ on every basis vector, the boundary case working because $p = 0$ in $K$ . This is the first symptom of the very different behaviour described on the positive-characteristic page.

Stone-von Neumann

Let $U(s)$ and $V(t)$ be strongly continuous one-parameter unitary groups on a separable Hilbert space satisfying the Weyl relation (Q.6), and suppose the pair acts irreducibly. Then the representation is unitarily equivalent to the Schrödinger representation (Q.2) on $L^2(R)$ . The hypotheses are not decoration: the theorem is stated for the exponentiated relation and for irreducible, strongly continuous representations on a separable space. For unbounded operators satisfying (Q.1) on a common dense domain, without the exponentiated form, the conclusion is false, and explicit counterexamples are known.

The algebraic contrast

$A_1$ has an abundance of simple modules that are not isomorphic to each other. The polynomial module $K[x]$ , the delta module $A_1/A_1x$ , and the modules $A_1/A_1(x∂ - λ)$ for $λ ∉ Z$ are already three families, and the twisting construction produces many more. The classification of simple $A_1$ -modules is a genuinely hard problem. No uniqueness theorem in the style of Stone-von Neumann is available or expected, because the Hilbert space structure - unitarity, self-adjointness, continuity - is exactly what is dropped.

Examples and Special Cases

The Schrödinger representation, algebraically

Take $V = K[x]$ with $q$ acting by multiplication and $p$ by $d/dx$ . Then $pq - qp = 1$ , since $(xg)' - xg' = g$ . This is the polynomial module, and it is the algebraic skeleton of wave mechanics with $ℏ$ and the factor $i$ scaled away.

Creation and annihilation operators

Over $ℂ$ , set $a = (x + ∂)/ 2$ and $b = (x - ∂)/ 2$ . Then $[a,b] = 12 ([x,-∂] + [∂,x] ) = 12(1 + 1) = 1$ , so $(b,a)$ is another pair of generators satisfying the same relation, and $C[a,b]$ with that relation is all of $A_1$ . This is the physicists' harmonic-oscillator basis; algebraically it is an automorphism of $A_1$ , one of the linear symplectic ones described under automorphisms of the Weyl algebra.

A relation that does collapse

Suppose one tries to impose $x∂ = λ$ for a scalar $λ$ on top of the Weyl relation. Then $∂ x = x∂ + 1 = λ + 1$ , and both $x∂$ and $∂ x$ are central. But the two-sided ideal they generate must be $0$ or the whole algebra by simplicity, and neither is compatible with the relation for any $λ$ . Littlewood's discovery that no further relation is possible is exactly this phenomenon, in general form.

Worked Example

Littlewood's infinite matrices, written out and checked

Step 1 - choose a basis
Work in the space $V$ with countable basis $e_0, e_1, e_2, …$ over a field $K$ of characteristic zero, consisting of finite linear combinations. Define two linear maps by their action on the basis:

$Q,e_k = e_k+1 (k ≥ 0), P,e_k = k,e_k-1 (k ≥ 1), P,e_0 = 0.$

As infinite matrices, $Q$ has $1$ 's on the subdiagonal and $P$ has the entries $1,2,3, …$ on the superdiagonal, everything else zero.
Step 2 - compute the two products on a basis vector
Apply $Q$ then $P$ : $PQ,e_k = P e_k+1 = (k+1),e_k$ . Apply $P$ then $Q$ : $QP,e_k = Q(k,e_k-1) = k,e_k$ for $k ≥ 1$ , and $QP,e_0 = Q ⋅ 0 = 0 = 0 ⋅ e_0$ , so the formula $QP,e_k = k,e_k$ holds for every $k ≥ 0$ .
Step 3 - subtract
$(PQ - QP),e_k = (k+1),e_k - k,e_k = e_k for all k ≥ 0,$

so $PQ - QP = id_V$ . Check the first three cases by hand: $k = 0$ gives $1 - 0 = 1$ ; $k = 1$ gives $2 - 1 = 1$ ; $k = 2$ gives $3 - 2 = 1$ . The relation holds with no adjustable constant left over.
Step 4 - see where the trace argument breaks
The diagonal entries are $(PQ)_kk = k+1$ and $(QP)_kk = k$ . Each individual difference is $1$ , but the formal traces are $∑ _k ≥ 0(k+1)$ and $∑ _k ≥ 0 k$ , both divergent. The identity $tr(AB) = tr(BA)$ used in the finite case is the statement that two convergent double sums may be reordered; here neither sum converges, so the obstruction simply does not apply. Infinite dimension is not a technicality dodged - it is the precise place where the obstruction fails.
Step 5 - identify what has been built
Send $e_k ↦ x^k$ . Then $Q$ is multiplication by $x$ and $P$ is $d/dx$ , since $d(x^k)/dx = k x^k-1$ . Littlewood's matrices are the polynomial module in disguise, written in the monomial basis. The physicists' 'infinite matrices' and the mathematicians' 'differential operators on $K[x]$ ' are the same object viewed in two coordinate systems.

Result

The pair $Q,e_k = e_k+1$ , $P,e_k = k,e_k-1$ satisfies $PQ - QP = id$ on a space of countable dimension, and under $e_k ↦ x^k$ it is exactly multiplication by $x$ together with $d/dx$ on $K[x]$ . Finite matrices cannot do this because $tr(PQ) = tr(QP)$ ; the infinite version escapes because the traces diverge.

Applications and Industry Use

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

The relation is not a historical curiosity; it continues to be the load-bearing identity in several distinct areas.

Quantum theory. The canonical commutation relations remain the definition of a quantum degree of freedom, and their exponentiated form defines the Heisenberg group, whose representation theory underlies the harmonic analysis of quantum mechanics and of signal processing.
Deformation quantisation. The Weyl algebra is the model case of a deformation of a commutative algebra in the direction of a Poisson bracket, and the general theory takes it as the local picture.
Lie theory. Quotients of enveloping algebras of nilpotent Lie algebras by primitive ideals are Weyl algebras, which is how $A_n$ entered mainstream algebra in the 1960s.
Signal processing and optics. The Weyl relations govern the time-frequency uncertainty principle, the Wigner distribution and the fractional Fourier transform, all of which are representations of the same one-parameter groups.
Computer algebra. The relation is the rewrite rule implemented by every symbolic package that manipulates differential operators, from Gröbner bases in the Weyl algebra to the holonomic-function machinery used for definite integration.

Limits of Validity

The step from the physics to the algebra discards several things, and the discarded structure is not recoverable from $A_n$ alone.

No inner product, no adjoints. In $A_n$ there is no notion of $p$ being self-adjoint, and no positivity. Statements about spectra, expectation values or unitarity have no algebraic counterpart.
No topology. Modules over $A_n$ are plain modules. Convergence, completeness and continuity are absent, which is why Stone-von Neumann has no algebraic analogue.
No $ℏ$ . The constant is scaled to $1$ , so the classical limit is not visible in $A_n$ itself. It reappears only when one filters the algebra and passes to the associated graded object, where the leading symbol behaves classically.
Characteristic zero only. Every statement above about the impossibility of finite-dimensional representations, and about simplicity, uses characteristic zero.
One relation is not one physical system. The algebra fixes kinematics only. Dynamics is a choice of Hamiltonian inside the algebra, and nothing in $A_n$ singles one out.

Failure Modes and Common Mistakes

Quoting Stone-von Neumann for the bare relation

The theorem is about the exponentiated Weyl form (Q.6), for strongly continuous irreducible representations by unitary operators on a separable Hilbert space. Applied to unbounded operators merely satisfying $PQ - QP = 1$ on a common dense domain, it is false; counterexamples have been known since the 1940s. In the purely algebraic category it fails badly, since $A_1$ has many pairwise non-isomorphic simple modules. 'The canonical commutation relation has a unique representation' is true only with all its hypotheses attached.

Believing the trace argument rules out infinite matrices too

It rules out only the finite case. As Step 4 of the worked example shows, in the infinite case the two traces are individually divergent and the identity $tr(AB) = tr(BA)$ is unavailable. The correct statement of the obstruction is that $P$ and $Q$ cannot both be trace-class, or more usefully that they cannot both be bounded - Wintner's and Wielandt's theorem - which is a genuinely different argument.

Treating the normalisation as a loss of information

Rescaling $p$ to remove $-iℏ$ is an isomorphism of algebras over $ℂ$ , so nothing algebraic is lost. What is lost is dimensional bookkeeping: after normalisation $x$ and $∂$ carry inverse units, and any physical statement must reinstate them. The algebra will not catch a dimensional error for you.

Attributing the algebra's theorems to Weyl

Weyl's contribution was the systematic use of the exponentiated relations and their group-theoretic reading, in his 1928 book. The canonical form, the absence of zero divisors and simplicity are due to Littlewood in 1933. The name Weyl algebra was proposed by Segal and adopted by Dixmier in 1968. Citing 'Weyl's theorem' for any of the structure results in Chapters 1 and 2 is a misattribution the literature has largely, but not entirely, corrected.

Assuming $p$ and $q$ generate a free algebra

They do not. The relation $∂ x = x∂ + 1$ lets every word in the generators be rewritten with all the $x$ 's to the left, so the monomials $x^a∂^b$ span, and the canonical basis theorem says they are independent. The algebra therefore has exactly the same size as a polynomial ring in $2n$ variables, not the much larger free algebra.

Historical Notes

1925: Heisenberg, Born and Jordan

Heisenberg's paper of July 1925 replaced the classical Fourier series describing an electron's motion by arrays of transition amplitudes, each indexed by two stationary states. He formed products of these arrays by a rule he derived from the correspondence principle, and observed that the rule was not symmetric in the factors. Born recognised the rule as matrix multiplication and, with Jordan, recast the theory in matrix form in a paper submitted in September 1925. There the commutation relation appears explicitly, together with the argument that it cannot be satisfied in finite dimensions. The 'three-man paper' of Born, Heisenberg and Jordan extended the scheme to systems with several degrees of freedom, producing the relations that in modern notation read $[,p_i, q_j,] = -iℏ,δ_ij$ with all other commutators zero.

1925-1926: Dirac's quantum algebra

Working independently in Cambridge, Dirac took a different route. He noticed that the commutator of two quantum quantities behaves like the classical Poisson bracket, and proposed the correspondence $[,a, b,] ≤ ftrightarrow iℏa,b$ as the basic principle of quantisation. That turns the relation $q,p = 1$ of classical mechanics into the commutation relation. Dirac then studied the associative algebra generated by $p$ and $q$ subject to that relation alone - his quantum algebra - and showed, among other things, how to differentiate a polynomial expression in $p$ and $q$ with respect to $p$ or to $q$ using only commutators. In modern language he had found that the partial derivatives of the algebra are inner derivations, a fact that is still the standard tool for computing in $A_n$ .

1926 onwards: wave mechanics and the naming

Schrödinger's wave mechanics arrived shortly afterwards and looked completely different: a partial differential equation, familiar territory for physicists, rather than infinite arrays. The two formalisms were quickly shown to be equivalent, and the reason is visible in the algebra. Setting $q = x$ (multiplication) and $p = -iℏ,d/dx$ on functions of one variable satisfies the same relation, so wave mechanics is a representation of the same algebra. Hermann Weyl's 1928 book on group theory and quantum mechanics gave Dirac's algebraic viewpoint its most influential presentation, working with the exponentiated form of the relations that now carry his name.

1933: Littlewood, and the algebra as a mathematical object

D. E. Littlewood observed in 1933 that while finite-dimensional algebras had been studied intensively, infinite-dimensional ones had not, and took Dirac's quantum algebra as a test case. He constructed explicit infinite matrices satisfying $pq - qp = 1$ ; established that every element has a normal form as a polynomial in $q$ with the $p$ 's collected on one side, which is the canonical form used throughout this collection; showed the algebra has no zero divisors; and proved that no further relation can be imposed without collapsing the algebra entirely - what is now called simplicity. Three of the four foundational theorems of Chapters 1 and 2 of the modern theory are already there.

1963-1968: the modern name

Interest revived when the algebra reappeared in Lie theory: the quotient of the enveloping algebra of a nilpotent Lie algebra over $ℂ$ by a primitive ideal is always a Weyl algebra. Dixmier introduced the notation $A_n$ for the algebra of a system with $n$ degrees of freedom in 1963, and in his 1968 paper adopted the name Weyl algebra, following a suggestion of I. Segal. The name honours Weyl's treatment rather than a theorem of his about the algebra, which is worth remembering when reading the older literature.

Comparison

Three 1925-26 formulations of the same algebra.
	Matrix mechanics	Wave mechanics	Dirac's quantum algebra
Principal authors	Heisenberg, Born, Jordan	Schrödinger	Dirac
Basic object	infinite matrices of transition amplitudes	wavefunctions and a differential equation	an abstract algebra with generators and relations
$q$ acts as	a matrix	multiplication by $x$	a generator
$p$ acts as	a matrix	$-iℏ,d/dx$	a generator
Status in modern terms	a representation	a representation	the algebra itself
Descendant in this collection	modules over $A_n$	the module $K[x]$ and its relatives	the presentation of $A_n$ by generators and relations

The distinction in the last two rows is the one that matters mathematically. Matrix and wave mechanics are two representations; Dirac's formulation is the algebra of which they are representations. The modern theory studies the algebra first and its representations - its modules - second, which is why a page like this collection's introduction to D-modules begins with the ring.

Key Takeaways

Key points

Heisenberg's 1925 arrays multiply noncommutatively because they are indexed by pairs of states; Born and Jordan recognised them as matrices and wrote $pq - qp = h/2π i$ .
Rescaling the constant gives $[∂_i, x_j] = δ_ij$ , the defining relation of the Weyl algebra $A_n$ .
No finite matrices satisfy the relation in characteristic zero, by a one-line trace argument; in characteristic $p$ they do exist.
Littlewood's explicit infinite matrices are the polynomial module $K[x]$ written in the monomial basis.
Dirac's differentiation with respect to $p$ and $q$ is the fact that the partial derivatives of $A_n$ are inner derivations.
Stone-von Neumann gives uniqueness for the exponentiated relation on Hilbert space; the algebraic category has no such theorem and many simple modules.
Littlewood proved canonical form, the domain property and simplicity in 1933; Dixmier introduced $A_n$ and, after Segal's suggestion, the name Weyl algebra.

FAQs

Did Heisenberg know he was using matrices?

Not at first. He derived a multiplication rule for his arrays from the correspondence principle and was troubled that it depended on the order of the factors. Born identified the rule as matrix multiplication, which he knew from his mathematical training, and that recognition is what turned the calculation into a formalism.

Why is the constant on the right-hand side imaginary?

Because $p$ and $q$ are required to be self-adjoint, and the commutator of two self-adjoint operators is anti-self-adjoint. A real constant is therefore impossible on physical grounds. The algebraic theory drops self-adjointness, so it can and does normalise the constant to $1$ .

Is the Weyl algebra the same as the Heisenberg algebra?

No, though they are closely related. The Heisenberg Lie algebra is the $(2n+1)$ -dimensional Lie algebra spanned by the $p_i$ , $q_i$ and a central element $z$ . The Weyl algebra is the quotient of its universal enveloping algebra by the ideal setting $z = 1$ . The Lie algebra is finite dimensional; the Weyl algebra is not.

Where does Weyl's own contribution actually enter?

In the exponentiated relations (Q.6) and their reading as a projective representation of a group. That form is what makes the Stone-von Neumann theorem provable, since it replaces unbounded operators by unitary ones, and it is the basis of the Heisenberg group approach to harmonic analysis.

Can I see the classical limit inside the algebra?

Only after filtering. Assign degrees to the generators and pass to the associated graded algebra: the images of $x_i$ and $∂_i$ commute there, so the graded object is a polynomial ring in $2n$ variables - the functions on classical phase space. The commutator's leading term descends to the Poisson bracket. That is the algebraic form of $ℏ → 0$ .

Do the relations for several degrees of freedom add anything new?

Structurally, $A_n$ is the tensor product of $n$ copies of $A_1$ over $K$ , so nothing new appears at the level of the ring's generators. What is new is the geometry: with $n ≥ 2$ there is room for modules of intermediate dimension between $n$ and $2n$ , and the whole theory of characteristic varieties becomes non-trivial.

Why did it take until the 1960s for the algebra to get a name?

It was studied sporadically - by Littlewood in 1933, and by others in the context of operator theory - but it became a standard object only when noncommutative Noetherian ring theory and the theory of enveloping algebras made it a natural example. Dixmier's papers of 1963 and 1968 are the point at which $A_n$ became a fixed piece of algebraic vocabulary.

Is there a version of this story in positive characteristic?

There is an algebra with the same presentation, but it behaves entirely differently: $x_i^p$ and $∂_i^p$ become central, the algebra is a finite module over its centre, it is not simple, and it has finite-dimensional representations. It is a legitimate object of study with applications in arithmetic geometry, but essentially none of the theorems on this page survive.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Introduction §1, and Exercise 1.4.10 for Littlewood's matrices.
W. Heisenberg, Über quantentheoretische Umdeutung kinematischer und mechanischer Beziehungen, Zeitschrift für Physik 33 (1925), 879-893.
M. Born and P. Jordan, Zur Quantenmechanik, Zeitschrift für Physik 34 (1925), 858-888 - the commutation relation and the argument against finite matrices.
M. Born, W. Heisenberg and P. Jordan, Zur Quantenmechanik II, Zeitschrift für Physik 35 (1926), 557-615 - several degrees of freedom.
P. A. M. Dirac, The fundamental equations of quantum mechanics, Proceedings of the Royal Society A 109 (1925), 642-653 - the quantum algebra and the Poisson bracket correspondence.
H. Weyl, The Theory of Groups and Quantum Mechanics, Dover reprint, 1950 - the exponentiated relations.
D. E. Littlewood, On the classification of algebras, Proceedings of the London Mathematical Society 35 (1933), 200-240 - canonical form, domain, simplicity.
J. von Neumann, Die Eindeutigkeit der Schrödingerschen Operatoren, Mathematische Annalen 104 (1931), 570-578.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - the name and the notation $A_n$ .
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, revised edition, American Mathematical Society, 2001 - Ch. 1, for the algebra in its ring-theoretic setting.

AI Suggested Questions

Verify by induction that $∂^k x = x∂^k + k∂^k-1$ in $A_1$ .
Show that $[∂, D] = ∂ D/∂ x$ for $D$ in canonical form, and deduce that the centre of $A_1$ is $K$ .
Write out the $5 × 5$ corner of Littlewood's matrices $P$ and $Q$ and multiply them both ways.
Exhibit two bounded operators on a Hilbert space and explain why their commutator can never be the identity.
Construct a finite-dimensional representation of the relation $PQ - QP = 1$ over a field of characteristic $3$ .
Explain how $A_n$ arises as a quotient of the enveloping algebra of the Heisenberg Lie algebra.
Compare the harmonic-oscillator generators $a, b$ with $x, ∂$ and describe the automorphism relating them.