Why the Weyl Algebra Has No Finite-dimensional Representations

Q: Is the trace argument enough on its own?

Yes, for the statement as given: it is complete and takes two lines. The other proofs are included because they use different hypotheses and therefore generalise differently - the simplicity proof to any infinite-dimensional simple algebra, the nilpotency proof to normed algebras where no trace exists.

Q: Does this mean A_n has no simple modules?

No. It has many; they are just all infinite dimensional over K . The polynomial ring K[X] is one, the modules A_1/A_1(x∂ - λ) give others, and localisations such as K[x][1/f] produce more. Classifying the simple A_1 -modules is a hard problem addressed by Block in the 1980s and still not reducible to a short list.

Q: Why can I not just take an infinite matrix and truncate it?

Because the relation fails exactly at the boundary of the truncation, and the failure has trace V , independent of how the truncation is chosen. Equation (2.26) makes this quantitative for the natural truncation, and the trace identity shows it is not an artefact of that choice.

Q: Does a finitely generated A_n-module have to be infinite dimensional over K?

Yes, if it is non-zero - finite generation over A_n is a much weaker condition than finite dimension over K . K[X] is generated by the single element 1 and has countably infinite K -dimension. What is finite for a finitely generated module is each piece of a good filtration, not the module.

Q: How does this square with the Stone-von Neumann theorem?

Stone-von Neumann classifies the irreducible unitary representations of the exponentiated Weyl relations on a Hilbert space, and there is exactly one up to unitary equivalence - the Schrödinger representation on L^2(R^n) , which is infinite dimensional. The passage to exponentials is forced precisely because the commutator form has no bounded solutions, which is the analytic version of this page's theorem.

Q: Is there any sense in which finite-dimensional approximations are useful?

Yes, but as approximations only. Truncated models like V_d are used in numerical work and in the study of Toeplitz and Berezin quantisation, where the error term of (2.26) is tracked explicitly and shown to be small in a suitable operator norm as d → ∈ fty . The point is that the error never becomes zero.

Q: What replaces "dimension of a representation" for A_n-modules?

The pair (d(M), e(M)) : the degree and normalised leading coefficient of the Hilbert polynomial of a good filtration. These are finite, independent of the filtration, additive in short exact sequences in the right sense, and they play the role dimension plays for finite-dimensional algebras. Holonomic modules - those with d(M) = n - behave the most like finite-dimensional representations, having finite length with length bounded by e(M) .

Overview

The commutation relation $[∂_i, x_i] = 1$ looks harmless. It is not: over a field of characteristic zero it cannot be realised by matrices of any finite size. If $P$ and $X$ are endomorphisms of a finite-dimensional space $V$ with $PX - XP = id_V$ , then $V = 0$ .

The consequence for $A_n$ is immediate and total. Every non-zero $A_n$ -module is infinite dimensional over $K$ . There is no analogue of the finite-dimensional representation theory that organises the study of finite groups, of semisimple Lie algebras, or of finite-dimensional associative algebras. The smallest module in the theory, the polynomial ring $K[X]$ itself, already has countably infinite dimension.

Three proofs are given below and they are genuinely different. The trace argument is the shortest and the most famous. The simplicity argument is the most structural: a non-zero module is faithful, so $A_n$ would embed in $End_K(V)$ , and an infinite-dimensional algebra cannot embed in a finite-dimensional one. The nilpotency argument avoids traces altogether, which matters because it is the one that survives into settings where traces are unavailable, such as normed algebras.

Every proof uses characteristic zero, and each shows you where. In characteristic $p$ the theorem is false: the relation is realised on $k[x]/(x^p)$ , of dimension $p$ , and the trace obstruction reads $0 = p$ , which is true. That contrast is developed on the positive characteristic page; here it serves as the sanity check that the hypothesis is not decorative.

Definition

Throughout, $K$ is a field, $V$ a $K$ -vector space, and a representation of a $K$ -algebra $R$ on $V$ means a $K$ -algebra homomorphism $ρ : R → End_K(V)$ with $ρ(1) = id_V$ . Equivalently, $V$ is a unital left $R$ -module. We write $P = ρ(∂_1)$ and $X = ρ(x_1)$ .

No finite-dimensional representations

Let $K$ have characteristic zero and let $V$ be a $K$ -vector space of finite dimension. If $P, X ∈ End_K(V)$ satisfy

P X - X P = id_V,

then $V = 0$ . Consequently, for every $n ≥ 1$ the only finite-dimensional $A_n(K)$ -module is the zero module, and there is no $K$ -algebra homomorphism $A_n(K) → M_d(K)$ for any $d ≥ 1$ .

Attribution

Coutinho does not isolate this as a numbered theorem; it sits behind the remarks of Ch. 2 §2 on how peculiar $A_n$ is, and behind the Ch. 1 exercises that realise $A_1$ only inside an algebra of infinite matrices. The statement itself is classical, going back to Wintner and Wielandt in the 1940s in the analytic setting. We state and prove it here because everything about $A_n$ -modules in this collection depends on it.

$A_n$ satisfies no polynomial identity

By Kaplansky's theorem, a primitive PI algebra is finite dimensional over its centre. $A_n$ is simple with centre $K$ and is infinite dimensional over $K$ , so it satisfies no non-trivial polynomial identity. In particular $A_n$ is not a subalgebra of any matrix algebra over a commutative ring, which is a strengthening of the theorem above.

Core Concepts

It is worth separating what is being ruled out from what is not.

The obstruction is the identity operator, not the operators $P$ and $X$

Nothing forbids two matrices with a large, complicated commutator. What is forbidden is a commutator equal to the identity. Commutators of matrices are exactly the trace-zero matrices - a classical theorem of Shoda and Albert-Muckenhoupt - so the identity is a commutator if and only if $V = 0$ in $K$ . The whole difficulty is concentrated in one scalar equation.

Truncation almost works, and the error tells you why it cannot

Take polynomials of degree at most $d$ , a space of dimension $d+1$ , let $P$ be differentiation and let $X$ be multiplication by $x$ followed by discarding the degree- $(d+1)$ term. Then $[P,X]$ is the identity except in the last coordinate, where it is $-d$ . The error is a single rank-one operator of trace $d+1$ , sitting exactly at the top of the truncation. You cannot push it out of the way: any finite truncation of an infinite-dimensional module has a boundary, and the failure of the relation lives there.

Simplicity turns the question into a size comparison

A non-zero module over a simple ring is faithful, so a $d$ -dimensional module would give an injection $A_n ↪ M_d(K)$ . But $_K A_n$ is infinite: the canonical monomials $x^α∂^β$ are linearly independent. An injective linear map from an infinite-dimensional space into a $d^2$ -dimensional one does not exist. This is the argument that generalises to any infinite-dimensional simple algebra.

Construction and Proof

Three proofs. They use different hypotheses, and it is worth knowing which is which.

Proof 1: the trace argument

Suppose $_K V = d < ∈ fty$ and $[P,X] = id_V$ . For any two endomorphisms of a finite-dimensional space, $tr(PX) = tr(XP)$ , so $tr[P,X] = 0$ . On the other hand $tr(id_V) = d ⋅ 1_K$ . Hence $d ⋅ 1_K = 0$ in $K$ . Since $char,K = 0$ , the map $ℤ \to K$ is injective, so $d = 0$ and $V = 0$ .

Where the hypothesis is used: only in the final step. In characteristic $p$ the conclusion weakens to $p ∣ d$ , and that is satisfiable.

Proof 2: simplicity and a dimension count

Let $V ≠ 0$ be a finite-dimensional $A_n$ -module and let $ρ : A_n → End_K(V)$ be the corresponding homomorphism. Its kernel is $Ann(V)$ , a two-sided ideal of $A_n$ . It is not all of $A_n$ , because $1$ acts as $id_V ≠ 0$ . By simplicity, $Ann(V) = 0$ , so $ρ$ is injective.

But $A_n$ has infinite dimension over $K$ - the canonical monomials $x^α∂^β$ form a basis - while $End_K(V)$ has dimension $( _K V)^2 < ∈ fty$ . No injective $K$ -linear map exists between them. Contradiction, so $V = 0$ .

Where the hypothesis is used: inside simplicity, which is a characteristic-zero theorem. This proof is the one that generalises: any simple $K$ -algebra of infinite $K$ -dimension has no non-zero finite-dimensional modules.

Proof 3: forced nilpotency, without traces

Assume $[P,X] = id_V$ on a finite-dimensional $V$ . Identity (2.24), $[X, P^k] = -kP^k-1$ , follows by induction: for $k = 1$ it is (2.22), and $[X,P^k+1] = [X,P^k]P + P^k[X,P] = -kP^k-1P - P^k = -(k+1)P^k$ .

Now claim that the powers $id_V, P, P^2, …$ are linearly independent over $K$ , unless $V = 0$ . Suppose not, and choose a non-trivial relation $∑ _k=0^m c_k P^k = 0$ with $c_m ≠ 0$ and $m$ as small as possible. If $m = 0$ the relation reads $c_0,id_V = 0$ with $c_0 ≠ 0$ , which forces $V = 0$ and we are finished. If $m ≥ 1$ , apply $ad X$ to the relation and use (2.24):

0 = [X,; ∑ _k=0^m c_k P^k ] = - ∑ _k=1^m k,c_k,P^k-1.

This is a relation of degree $m-1$ whose top coefficient is $m,c_m$ , and in characteristic zero $m,c_m ≠ 0$ . That contradicts the minimality of $m$ .

So either $V = 0$ , or $End_K(V)$ contains an infinite linearly independent family, forcing $_K V = ∈ fty$ . For finite-dimensional $V$ the only possibility is $V = 0$ .

Where the hypothesis is used: in dividing by $k$ . This is exactly the same place as in the simplicity descent, and it is the version of the argument that has an analytic analogue: in a normed algebra one replaces "some power vanishes" by a norm estimate and obtains Wielandt's theorem, that $[a,b] = 1$ is impossible for bounded operators.

Every non-zero $A_n$ -module is infinite dimensional

Immediate. In particular every simple $A_n$ -module is infinite dimensional over $K$ , and the classification of simple $A_1$ -modules - begun by Block in the 1980s - is correspondingly delicate. Contrast this with a finite-dimensional simple algebra, where there is exactly one simple module up to isomorphism.

Consistency with Bernstein's inequality

A finite-dimensional module has bounded Hilbert function under any good filtration, so its dimension would be $d(M) = 0$ . Bernstein's inequality gives $d(M) ≥ n ≥ 1$ for every non-zero finitely generated module, which rules this out. So Bernstein's inequality is a strict quantitative strengthening of the present theorem, and the present theorem is the $n$ -independent shadow of it.

Key Equations

The relation being tested, and the two identities each proof turns on:

[P, X] = P X - X P = id_V.

(2.22)

Traces annihilate commutators, so applying $tr$ to (2.22) gives

0 = tr(PX) - tr(XP) = tr(id_V) = _K V ⋅ 1_K.

(2.23)

Iterating (2.22) gives the identity used by the nilpotency proof, valid in any ring:

[,X,; P^,k,] = -,k,P^,k-1 (k ≥ 1).

(2.24)

And the dimension count behind the simplicity proof:

_K A_n = ∈ fty, _K End_K(V) = ( _K V)^2 < ∈ fty.

(2.25)

The truncation error, for $V_d$ the polynomials of degree at most $d$ with $X$ the truncated multiplication:

[,P,; X,] = id_V_d - (d+1),E_dd,

(2.26)

where $E_dd$ is the projection onto the top coefficient. The correction has trace $d+1$ , matching (2.23) exactly.

Variable Definitions

$K$: the ground field, of characteristic zero unless stated otherwise
$V$: a $K$ -vector space carrying the putative representation
$ρ$: the algebra homomorphism $A_n → End_K(V)$ , sending $1$ to $id_V$
$P, X$: the images $ρ(∂_1)$ and $ρ(x_1)$
$tr$: the trace of an endomorphism of a finite-dimensional space
$Ann(V)$: the annihilator of $V$ in $A_n$ , a two-sided ideal
$V_d$: the space of polynomials of degree at most $d$ , of dimension $d+1$
$E_dd$: the projection of $V_d$ onto the coefficient of $x^d$

Properties and Behaviour

No non-zero homomorphism to a finite-dimensional algebra

If $B$ is a finite-dimensional $K$ -algebra, there is no $K$ -algebra homomorphism $φ : A_n → B$ with $φ(1) = 1$ at all. For $B$ acts faithfully on itself by left multiplication, so such a $φ$ would make $B$ a non-zero finite-dimensional $A_n$ -module. This is consistent with, and slightly sharper than, simplicity: $A_n$ has no proper non-zero quotients, and in particular no finite-dimensional ones.

The Heisenberg Lie algebra does have finite-dimensional representations

Let $h_n$ be the $(2n+1)$ -dimensional Heisenberg Lie algebra, with basis $p_i, q_i, z$ and brackets $[p_i,q_j] = δ_ijz$ , $z$ central. It has plenty of finite-dimensional representations - it is nilpotent, so by Lie's theorem it acts by upper triangular matrices over an algebraically closed field of characteristic zero. But in every such representation the central element $z$ acts by a nilpotent scalar, hence by $0$ .

$A_n$ is the quotient $U(h_n)/(z-1)$ . Requiring $z$ to act as $1$ rather than $0$ is precisely what kills all the finite-dimensional representations. This is the cleanest way to see that the theorem is about the normalisation of the commutator, not about the algebra being large.

Consequences for module theory

Since no non-zero module is finite dimensional, the coarse invariants of finite-dimensional representation theory - dimension, character, composition multiplicities as integers weighted by dimension - are unavailable. They are replaced by the graded invariants built from a good filtration: the Hilbert polynomial, its degree $d(M)$ and its normalised leading coefficient $e(M)$ . Those are the correct finite substitutes, and constructing them is the business of Chapters 7 to 9.

Worked Example

Truncating $K[x]$ to three dimensions and measuring the error

Step 1 - set up the truncation
Let $V_2 ⊂ Q[x]$ be the polynomials of degree at most $2$ , with basis $1, x, x^2$ , so $V_2 = 3$ . Differentiation preserves $V_2$ ; multiplication by $x$ does not, so define $X$ to be multiplication by $x$ followed by deleting the $x^3$ term. In the given basis, reading columns as images of the basis vectors,

$X = pmatrix 0 & 0 & 0 1 & 0 & 0 0 & 1 & 0 pmatrix, P = pmatrix 0 & 1 & 0 0 & 0 & 2 0 & 0 & 0 pmatrix,$

since $X: 1 ↦ x ↦ x^2 ↦ 0$ and $P: 1 ↦ 0$ , $x ↦ 1$ , $x^2 ↦ 2x$ .
Step 2 - compute the commutator exactly
Multiplying out the two products:

$PX = pmatrix 1 & 0 & 0 0 & 2 & 0 0 & 0 & 0 pmatrix, XP = pmatrix 0 & 0 & 0 0 & 1 & 0 0 & 0 & 2 pmatrix,$

$[P, X] = pmatrix 1 & 0 & 0 0 & 1 & 0 0 & 0 & -2 pmatrix = id_V_2 - 3E_22.$

The relation holds on $1$ and on $x$ and fails only on $x^2$ , where it is off by $-3$ . This is (2.26) with $d = 2$ : the correction is rank one with trace $3 = V_2$ .
Step 3 - confirm the trace bookkeeping
$tr[P,X] = 1 + 1 - 2 = 0$ , as it must be for any commutator. Meanwhile $tr(id_V_2) = 3$ . The discrepancy of $3$ is precisely the trace of the error term, and no adjustment of $X$ can remove it: whatever finite matrices you write down, the trace of the commutator is $0$ and the trace of the identity is $dim V$ .
Step 4 - run the nilpotency argument on this example
The powers of $P$ satisfy the relation $P^3 = 0$ , and $P^2 ≠ 0$ : indeed $P^2$ sends $x^2 ↦ 2$ and everything else to $0$ . If $[P,X] = id$ held, applying $ad X$ to the relation $P^3 = 0$ would give $0 = [X,P^3] = -3P^2$ , hence $P^2 = 0$ over $ℚ$ - false. That is the minimal-relation argument of Proof 3 in a single concrete instance, and it confirms independently that no such $X$ exists on $V_2$ .
Step 5 - change the field and watch the obstruction vanish
Read the same two matrices over $Z_3$ . Then $-2 = 1$ , so $[P,X] = id$ exactly, and the error term $3E_22$ is zero. The trace equation reads $0 = 3 = 0$ , which is satisfied. So over $Z_3$ these matrices give an honest $3$ -dimensional representation of the Weyl relations - the module $Z_3[x]/(x^3)$ of the positive characteristic page. Nothing about the matrices changed; only the arithmetic of the field did.

Result

On the three-dimensional space $V_2$ over $ℚ$ , the natural candidates give $[P,X] = id - 3E_22$ , not $id$ . The defect has trace $3 = V_2$ , which is the obstruction (2.23) made visible. Over $Z_3$ the same matrices satisfy $[P,X] = id$ exactly, because $3 = 0$ . The theorem is therefore sharp: the only thing standing between the Weyl relations and a finite-dimensional model is the characteristic of the field.

Applications and Industry Use

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

The theorem is the reason several standard tools are unavailable, and it therefore shapes the design of the whole theory:

Why filtrations exist at all. With no finite dimension to count, one counts the dimensions of the pieces of a filtration instead. The Bernstein filtration and the Hilbert polynomial are the substitutes, leading to dimension and multiplicity.
Why $K[X]$ is the fundamental example. It is the smallest natural module, and it is already infinite dimensional. Its simplicity as an P1-module then makes it the base case for most inductions.
Why solution spaces are finite dimensional but modules are not. The polynomial solution space of a single non-zero operator in $A_1$ is finite dimensional, and that finiteness is a statement about $Hom_A_1(M, K[x])$ , not about $M$ ; see solutions as homomorphisms.
Quantum mechanics. The unboundedness of position and momentum operators is the analytic shadow of this theorem, and it is why the canonical commutation relations are usually handled in Weyl's exponentiated form.

In computation, the theorem explains why there is no "matrix representation" data structure for elements of $A_n$ : implementations must work with normal forms in the algebra itself, or with the action on truncated polynomial spaces together with an explicit degree bound - the truncation error of (2.26) then has to be tracked, not ignored.

Limits of Validity

The theorem is narrower than it is sometimes quoted to be.

It is about characteristic zero. In characteristic $p$ every simple module has dimension exactly $p^n$ over an algebraically closed field. The theorem is not merely unproved there; it is false.
It is about unital modules. A ring homomorphism $A_n → End_K(V)$ not sending $1$ to $id_V$ is a different object, and the zero map always exists. All modules in this collection are unital by convention.
It says nothing about finitely generated modules being small. $K[X]$ is generated by one element and is infinite dimensional. Finite generation over $A_n$ and finite dimension over $K$ are unrelated conditions, and conflating them is a frequent source of error.
It does not say every module is faithful for elementary reasons. Faithfulness comes from simplicity, which is a genuine theorem. Over $A_n(Z)$ , for instance, modules are still infinite dimensional in the relevant sense but need not be faithful.

What the analytic version costs

The same argument in functional analysis says that $[P,X] = I$ has no solution with $P, X$ bounded operators on a Banach space. Since quantum mechanics requires the relation, position and momentum must be unbounded, defined only on dense subspaces, and the resulting domain questions are the reason the Stone-von Neumann uniqueness theorem is stated for the exponentiated Weyl relations rather than for the commutator relation directly.

Failure Modes and Common Mistakes

Concluding "no representations" rather than "no finite-dimensional representations"

$A_n$ has a huge supply of representations. The defining one on $K[X]$ is faithful, and there are uncountably many pairwise non-isomorphic simple $A_1$ -modules over $ℂ$ . What fails is finite dimensionality, which is a statement about the modules, not about their existence.

Trying to fix the truncation by choosing better matrices

The truncation error (2.26) is rank one and looks removable. It is not: the trace of a commutator is zero for every pair of finite matrices, so the defect can be moved around but its trace, $dim V$ , is an invariant. Time spent searching for clever $d \times d$ matrices is time wasted; the obstruction is one linear functional and it never vanishes in characteristic zero.

Using the trace argument in a setting where the trace does not exist

In infinite dimensions the trace is undefined for general operators, so (2.23) is unavailable, and indeed the relation is realisable there. Attempting a "formal trace" or a regularised trace to rule out infinite-dimensional representations is a category error. If a trace-like argument is wanted in infinite dimensions, one needs a genuine trace class condition, and the correct statement then is Wielandt's, using the norm rather than the trace.

Expecting the theorem to survive to characteristic $p$

It does not, and the failure is not a technicality. Over $Z_p$ the algebra has finite-dimensional simple modules of dimension $p^n$ , it is finite over its centre, and it satisfies polynomial identities. Every statement in this section - no finite-dimensional modules, no polynomial identity, faithful modules - reverses.

Historical Notes

The impossibility of the canonical commutation relation in finite dimensions was noticed almost as soon as matrix mechanics was written down in 1925: Born, Heisenberg and Jordan's relation $pq - qp = ℏ/(2π i)$ cannot hold for finite matrices, and this was understood at the time as the reason infinite matrices were unavoidable.

The functional-analytic version - that no bounded operators on a Banach space satisfy $[a,b] = 1$ - was proved by Wintner in 1947 for Hilbert space and in general by Wielandt in 1949, with the short norm argument that is now standard. Wielandt's proof is the analytic twin of Proof 3 above: repeated commutation forces a growth condition that a bounded operator cannot meet.

On the algebraic side the statement is a footnote to the structure theory of $A_n$ that Dixmier assembled in 1968. Coutinho's Ch. 1 exercises, which realise $A_1$ using infinite matrices with entries $1, 2, 3, …$ along a diagonal, are the constructive counterpart: they show what an unavoidably infinite model looks like.

The characteristic- $p$ picture, in which finite-dimensional representations of dimension $p^n$ exist and are the generic simple modules, was developed from the 1970s and is now central to modular representation theory; see the positive characteristic page.

Comparison

The same relation $[P,X] = 1$ behaves differently in each of the standard settings.

Where $[P,X] = 1$ can and cannot be realised.
Setting	Realisable?	Reason / example
$M_d(K)$ , $char,K = 0$ , $d ≥ 1$	no	$tr$ of a commutator is $0$ , $tr(id) = d$
$M_d(k)$ , $char,k = p$	yes when $p ∣ d$	$k[x]/(x^p)$ , dimension $p$
$End_K(V)$ , $dim V$ infinite	yes	$V = K[x]$ , $P = ∂$ , $X = x$
Bounded operators on a Banach space	no	Wielandt's theorem, via a norm estimate
Unbounded operators on a Hilbert space	yes	position and momentum; Stone-von Neumann
Any Banach algebra	no	Wintner-Wielandt
Commutative ring	no	commutators vanish, so $1 = 0$

The pattern is that the relation needs either infinite dimension or characteristic $p$ . In the analytic world the same tension is what forces the position and momentum operators of quantum mechanics to be unbounded, with all the domain difficulties that entails.

Key Takeaways

Key takeaways

Over a field of characteristic zero, $[P,X] = id_V$ forces $V = 0$ ; hence $A_n$ has no non-zero finite-dimensional modules.
Trace proof: $tr$ kills commutators, so $_K V = 0$ in $K$ .
Structural proof: a non-zero module over a simple ring is faithful, and an infinite-dimensional algebra cannot embed in $End_K(V)$ for finite-dimensional $V$ .
Trace-free proof: $[X,P^k] = -kP^k-1$ forces $P$ nilpotent and then forces $id_V = 0$ ; this is the version with an analytic analogue (Wielandt).
Truncating $K[x]$ to degree $≤ d$ gives $[P,X] = id - (d+1)E_dd$ : the defect is rank one with trace $dim V$ , and it cannot be removed.
In characteristic $p$ the theorem is false: the same matrices work on $k[x]/(x^p)$ , of dimension $p^n$ in $n$ variables.
Because dimension over $K$ is useless as an invariant, the theory replaces it with Hilbert polynomials of good filtrations - which is where $d(M)$ and $e(M)$ come from.

FAQs

Is the trace argument enough on its own?

Yes, for the statement as given: it is complete and takes two lines. The other proofs are included because they use different hypotheses and therefore generalise differently - the simplicity proof to any infinite-dimensional simple algebra, the nilpotency proof to normed algebras where no trace exists.

Does this mean

A_n

has no simple modules?

No. It has many; they are just all infinite dimensional over $K$ . The polynomial ring $K[X]$ is one, the modules $A_1/A_1(x∂ - λ)$ give others, and localisations such as $K[x][1/f]$ produce more. Classifying the simple $A_1$ -modules is a hard problem addressed by Block in the 1980s and still not reducible to a short list.

Why can I not just take an infinite matrix and truncate it?

Because the relation fails exactly at the boundary of the truncation, and the failure has trace $dim V$ , independent of how the truncation is chosen. Equation (2.26) makes this quantitative for the natural truncation, and the trace identity shows it is not an artefact of that choice.

Does a finitely generated

A_n

-module have to be infinite dimensional over

K

Yes, if it is non-zero - finite generation over $A_n$ is a much weaker condition than finite dimension over $K$ . $K[X]$ is generated by the single element $1$ and has countably infinite $K$ -dimension. What is finite for a finitely generated module is each piece of a good filtration, not the module.

How does this square with the Stone-von Neumann theorem?

Stone-von Neumann classifies the irreducible unitary representations of the exponentiated Weyl relations on a Hilbert space, and there is exactly one up to unitary equivalence - the Schrödinger representation on $L^2(R^n)$ , which is infinite dimensional. The passage to exponentials is forced precisely because the commutator form has no bounded solutions, which is the analytic version of this page's theorem.

Is there any sense in which finite-dimensional approximations are useful?

Yes, but as approximations only. Truncated models like $V_d$ are used in numerical work and in the study of Toeplitz and Berezin quantisation, where the error term of (2.26) is tracked explicitly and shown to be small in a suitable operator norm as $d → ∈ fty$ . The point is that the error never becomes zero.

What replaces "dimension of a representation" for

A_n

-modules?

The pair $(d(M), e(M))$ : the degree and normalised leading coefficient of the Hilbert polynomial of a good filtration. These are finite, independent of the filtration, additive in short exact sequences in the right sense, and they play the role dimension plays for finite-dimensional algebras. Holonomic modules - those with $d(M) = n$ - behave the most like finite-dimensional representations, having finite length with length bounded by $e(M)$ .

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 2 §2 for the ideal structure and units, and Ch. 1 Exercises 4.9-4.10 for the realisation of $A_1$ by infinite matrices.
H. Wielandt, Über die Unbeschränktheit der Operatoren der Quantenmechanik, Mathematische Annalen 121 (1949), 21 - no bounded operators satisfy the canonical commutation relation.
A. Wintner, The unboundedness of quantum-mechanical matrices, Physical Review 71 (1947), 738-739.
M. Born, W. Heisenberg and P. Jordan, Zur Quantenmechanik II, Zeitschrift für Physik 35 (1926), 557-615 - the origin of the relation and of the need for infinite matrices.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242.
R. E. Block, The irreducible representations of the Lie algebra sl(2) and of the Weyl algebra, Advances in Mathematics 39 (1981), 69-110 - the simple $A_1$ -modules.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, revised edition, American Mathematical Society, 2001 - Ch. 13 for Kaplansky's theorem and PI algebras.
M. Reed and B. Simon, Methods of Modern Mathematical Physics I: Functional Analysis, revised edition, Academic Press, 1980 - the Stone-von Neumann theorem and the necessity of unbounded operators.

AI Suggested Questions

Prove that a square matrix over a field of characteristic zero is a commutator if and only if its trace is zero.
Give Wielandt's norm proof that $[a,b] = 1$ is impossible in a Banach algebra, and compare it line by line with the nilpotency proof here.
Compute the truncation error $[P,X] - id$ for the degree- $d$ truncation of $K[x]$ for general $d$ and confirm its trace.
Describe the finite-dimensional representations of the Heisenberg Lie algebra and show the centre always acts by zero.
Why does Bernstein's inequality imply this theorem, and what does it add beyond it?
Sketch Block's classification of simple modules over the first Weyl algebra.
How do Berezin-Toeplitz truncations approximate the Weyl algebra, and in what norm does the commutator error tend to zero?