Twisting a Weyl Module by an Automorphism

Q: Why is the inverse needed in (R/J)_σ ≅ R/σ^-1(J)?

The map R → (R/J)_σ that sends b to b ∙ (1+J) lands on σ(b) + J , so b is in the kernel when σ(b) ∈ J , that is when b ∈ σ^-1(J) . Any statement with σ(J) has read the map in the wrong direction.

Q: Are the twists K[X]_σ_r really different from each other?

Yes, for distinct positive r . An isomorphism K[X]_σ_r → K[X]_σ_t would send the generator 1 to some non-zero polynomial f satisfying ∂ f/∂ x_i = (x_i^,t - x_i^,r)f ; comparing degrees on the two sides for r the isomorphism page .

Q: Is twisting the same thing as tensoring with a rank-one module?

In spirit, in the transvection case: K[X]_σ behaves like " e^H ⊗ K[X] ". But e^H is not an A_n -module of the kind one can tensor with over A_n , and the general construction here uses no tensor product at all. Keep them separate; the twist is defined for any ring and any automorphism, with no assumptions.

Overview

There is a cheap way to manufacture new modules from old ones. If $σ$ is an automorphism of a ring $R$ and $M$ is a left $R$ -module, define $M_σ$ to be the same abelian group with the action $a ∙ u = σ(a)u$ . Nothing is constructed and nothing is chosen; only the way $R$ is allowed to act has been relabelled. Yet $M_σ$ is frequently not isomorphic to $M$ .

For the Weyl algebra this construction is unusually productive, because $A_n$ has a very large automorphism group and, since its only units are the non-zero scalars, no non-identity automorphism is inner. Twisting by an inner automorphism never changes anything; over $A_n$ that escape route is closed, so every automorphism is a candidate for producing something new.

The single most important instance is the Fourier automorphism $ℱ$ , with $F(x_i) = ∂_i$ and $F(∂_i) = -x_i$ . Twisting the polynomial module by it produces $K[∂] ≅ A_n/ ∑ _i A_n x_i$ : the roles of multiplication and differentiation are exchanged, exactly as the analytic Fourier transform turns a constant-coefficient operator into a polynomial.

The reason twisting is safe is that it does not disturb the submodule lattice at all. A subgroup is a submodule of $M$ if and only if it is a submodule of $M_σ$ , so simplicity, torsion, cyclicity, finite generation and length all transfer. What can change is the isomorphism class, and that is precisely what the isomorphism problem exploits to build an infinite family of pairwise non-isomorphic simple $A_n$ -modules out of the single module $K[X]$ .

Definition

The twisted moduleCoutinho, Ch. 5 §2

Let $R$ be a ring, $σ$ an automorphism of $R$ , and $M$ a left $R$ -module. The twist of $M$ by $σ$ , written $M_σ$ , is the abelian group $M$ with the operation

a ∙ u = σ(a),u (a ∈ R, u ∈ M).

This is a left $R$ -module: $∙$ is additive in each argument because $σ$ is additive and the original action is bilinear, $(ab) ∙ u = σ(ab)u = σ(a)σ(b)u = a ∙ (b ∙ u)$ , and $1 ∙ u = σ(1)u = u$ .

The Fourier automorphismCoutinho, Ch. 5 §2; Exercise 1.4.8

$F : A_n → A_n$ is the unique $K$ -algebra map with $F(x_i) = ∂_i$ and $F(∂_i) = -x_i$ . It is well defined because the images satisfy the defining relations, and it is bijective because $F^4 = id$ . For a left $A_n$ -module $M$ , the module $M_F$ is called the Fourier transform of $M$ .

Convention warning

Some authors define the twist by $a ∙ u = σ^-1(a)u$ . Every formula below then acquires an inverse: with that convention $(R/J)_σ ≅ R/σ(J)$ rather than $R/σ^-1(J)$ . The convention used here is the one in the source. Whichever you adopt, check it against the Fourier example, where the answer is known independently.

Core Concepts

What twisting really does

Nothing about $M$ changes except the labelling of the operators. A useful way to say it: the set $R ∙ u = σ(a)u : a ∈ R$ equals the set $R u$ , because $σ$ is a bijection of $R$ . So the subgroups of $M$ that are closed under the action are the same subgroups before and after twisting. Every property of $M$ that is a statement about its submodule lattice is therefore automatically inherited by $M_σ$ , with no proof required beyond that sentence.

What twisting can change

Isomorphism class. An isomorphism $M → M_σ$ would be an additive bijection $ψ$ with $ψ(au) = σ(a)ψ(u)$ — a $σ$ -semilinear automorphism of $M$ — and there is no reason for one to exist. Annihilators track the change exactly: $ann_R(u)$ computed in $M_σ$ is $σ^-1$ of the annihilator computed in $M$ . Since a cyclic module is determined by the annihilator of its generator, twisting moves a maximal left ideal to another maximal left ideal, and different maximal left ideals can give non-isomorphic modules.

Why the Weyl algebra is the right place for this

If $σ$ is conjugation by a unit $u$ , then $m \mapsto u m$ is an isomorphism $M → M_σ$ , so inner automorphisms are invisible. In a commutative ring every automorphism has a chance of being non-inner but the module theory is usually too rigid to profit; in a matrix ring almost every automorphism is inner. $A_n$ is the good case: it is a domain whose units are only the non-zero scalars, which are central, so the only inner automorphism is the identity, while the automorphism group itself is enormous.

The analytic picture

For the transvection $σ(x_i) = x_i$ , $σ(∂_i) = ∂_i + g_i$ with $∇ H = (g_1, … ,g_n)$ , the twisted action of $∂_i$ on a polynomial $f$ is $∂ f/∂ x_i + g_i f$ , which is exactly $e^-H,∂_i,(e^H f)$ . So $K[X]_σ$ is the module " $e^H ⋅ K[X]$ " of functions, written algebraically. That is the picture to keep in mind: twisting multiplies by an exponential factor that is not itself a polynomial, which is why the result can fail to be isomorphic to what you started with.

Construction and Proof

What twisting preservesCoutinho (5.2.1)

Let $R$ be a ring, $σ$ an automorphism of $R$ , $M$ a left $R$ -module.

A subgroup $N ⊂ eq M$ is an $R$ -submodule of $M$ if and only if it is an $R$ -submodule of $M_σ$ ; the two submodule lattices are equal.
$M_σ$ is simple if and only if $M$ is simple.
$M_σ$ is a torsion module if and only if $M$ is a torsion module.
For a submodule $N$ , $(M/N)_σ = M_σ/N_σ$ .
For a left ideal $J$ of $R$ , $σ^-1(J)$ is a left ideal and $(R/J)_σ ≅ R/σ^-1(J)$ .

Proof

(1) A subgroup $N$ is a submodule of $M_σ$ iff $σ(a)u ∈ N$ for all $a ∈ R$ , $u ∈ N$ . Since $σ$ is onto, $σ(a) : a ∈ R = R$ , so this says exactly $au ∈ N$ for all $a ∈ R$ , $u ∈ N$ .

(2) is immediate from (1), because simplicity is a statement about the lattice and $M_σ = M$ as a set, so one is non-zero exactly when the other is.

(3) By (5.12) the annihilator of $u$ in $M_σ$ is $σ^-1(ann_R(u))$ . An automorphism carries $0$ to $0$ and non-zero left ideals to non-zero left ideals, so one annihilator is non-zero exactly when the other is; this holds for each $u$ separately.

(4) The underlying groups of $(M/N)_σ$ and $M_σ/N_σ$ are both $M/N$ , and on a class $u + N$ both actions send $a$ to $σ(a)u + N$ .

(5) $σ^-1$ is a ring automorphism, so it carries the left ideal $J$ to a left ideal. Define $φ : R → (R/J)_σ$ by $φ(b) = b ∙ (1 + J) = σ(b) + J$ . It is $R$ -linear for the twisted structure: $φ(ab) = σ(a)σ(b) + J = a ∙ φ(b)$ . It is surjective because $σ$ is. Its kernel is $b : σ(b) ∈ J = σ^-1(J)$ . Now apply the first isomorphism theorem.

The inverse in (5) is not optional

It is easy to state part (5) as $(R/J)_σ ≅ R/σ(J)$ , and the source's statement is written that way even though its proof produces $σ^-1(J)$ . The proof is right and the abbreviated statement is not, under the convention $a ∙ u = σ(a)u$ . The Fourier example settles it: $K[X] = A_n/ ∑ _i A_n∂_i$ and its Fourier transform is $K[∂] = A_n/ ∑ _i A_n x_i$ ; since $F(∂_i) = -x_i$ and $F^-1(∂_i) = x_i$ , only the $σ^-1$ version gives the right answer up to the harmless scalar.

The Fourier transform of the polynomial moduleCoutinho (5.2.2)

$K[X]_F ≅ K[∂] = A_n/ ∑ _i=1^n A_n x_i$ .

Proof

$K[X] ≅ A_n/J$ with $J = ∑ _i A_n∂_i$ . From $F(x_i) = ∂_i$ we get $F^-1(∂_i) = x_i$ , and since $F^-1$ is a ring automorphism it carries the left ideal generated by the $∂_i$ to the left ideal generated by the $x_i$ : $F^-1(J) = ∑ _i A_n x_i$ . Apply (5.13).

As a by-product, $K[∂]$ is simple, by part (2) of the previous proposition together with simplicity of P1 — no separate argument needed.

Key Equations

The definition and its two immediate consequences:

a ∙ u = σ(a)u, (M_σ)_τ = M_σ ∘ τ, M_id = M .

(5.11)

Annihilators transform by the inverse automorphism, which is the source of every asymmetry below:

ann^ ∙ _R(u) = a : σ(a)u = 0 = σ^-1 (ann_R(u) ).

(5.12)

Hence the rule for cyclic modules, the workhorse of the whole section:

(R/J)_σ ; ≅ ; R/σ^-1(J) for every left ideal J ⊂ eq R .

(5.13)

Applied to $ℱ$ and to the presentation of the polynomial module:

K[X]_F ; ≅ ; A_n /F^-1 ( ∑ _i A_n∂_i ) ;=; A_n / ∑ _i A_n x_i ; ≅ ; K[∂].

(5.14)

And to the transvections, which give the deformed modules of the previous section:

σ(x_i) = x_i, σ(∂_i) = ∂_i + g_i ⟹ K[X]_σ ; ≅ ; A_n / ∑ _i=1^n A_n(∂_i - g_i),

(5.15)

valid whenever the $g_i ∈ K[X]$ satisfy the integrability condition $∂ g_i/∂ x_j = ∂ g_j/∂ x_i$ , which is what makes $σ$ an automorphism.

Variable Definitions

$R$: an arbitrary ring with identity in the general construction; $A_n$ in the applications
$σ, τ$: automorphisms of $R$
$M$: a left $R$ -module
$M_σ$: the twist of $M$ by $σ$ : same abelian group, action $a ∙ u = σ(a)u$
$∙$: the twisted action, written with a bullet to distinguish it from the original action
$ℱ$: the Fourier automorphism of $A_n$ : $x_i ↦ ∂_i$ , $∂_i ↦ -x_i$
$J$: a left ideal of $R$ ; typically $∑ _i A_n∂_i$ or $∑ _i A_n x_i$
$g_i$: polynomials defining a transvection $∂_i ↦ ∂_i + g_i$ , subject to $∂ g_i/∂ x_j = ∂ g_j/∂ x_i$
$σ_r$: the automorphism $x_i ↦ x_i$ , $∂_i ↦ ∂_i - x_i^,r$ , used to build an infinite family of twists

Properties and Behaviour

FunctorialityCoutinho, Exercise 5.4.6

$M ↦ M_σ$ is an exact, additive self-equivalence of the category of left $R$ -modules. In particular it is the identity on underlying groups and on maps: an additive map is $R$ -linear for the original actions if and only if it is $R$ -linear for the twisted ones, so $Hom_R(M,N) = Hom_R(M_σ, N_σ)$ , and $(M ⊕ M')_σ = M_σ ⊕ M'_σ$ .

Consequently a module and its twist have the same length, the same number of generators, and isomorphic lattices of submodules. What is not claimed is $M_σ ≅ M$ .

Inner automorphisms are invisible

If $σ(a) = uau^-1$ for a unit $u ∈ R$ , then $m \mapsto u m$ is an isomorphism $M → M_σ$ , since $a ∙ (um) = uau^-1um = u(am)$ . For $R = A_n$ the units are exactly $K^ ×$ , which is central, so the only inner automorphism is the identity and this escape clause never applies.

Dimension is preserved; multiplicity need not be

For finitely generated $A_n$ -modules the dimension $d(M)$ coincides with the Gelfand-Kirillov dimension, which is computed from any finite-dimensional generating subspace of the algebra. An automorphism carries one such subspace to another, so $d(M_σ) = d(M)$ and holonomicity is preserved.

The multiplicity $e(M)$ is defined through the Bernstein filtration specifically, and an automorphism need not respect that filtration — $σ_r$ sends $∂_i$ , of Bernstein degree $1$ , to an element of degree $r$ . So $e(M_σ) = e(M)$ should be checked, not assumed.

Every twist of the polynomial module is simpleCoutinho, Ch. 5 §2

For every automorphism $σ$ of $A_n$ , $K[X]_σ$ is a simple, cyclic, torsion $A_n$ -module with $d = n$ , isomorphic to $A_n/σ^-1( ∑ _i A_n∂_i)$ . This single sentence produces as many simple modules as $A_n$ has automorphisms; deciding how many of them are pairwise distinct is a separate question.

Examples and Special Cases

The Fourier automorphism has order four

$F^2(x_i) = F(∂_i) = -x_i$ and $F^2(∂_i) = F(-x_i) = -∂_i$ , so $F^2$ is the parity automorphism $x ↦ -x$ , $∂ ↦ -∂$ , and $F^4 = id$ . The parity twist changes nothing: $f(x) ↦ f(-x)$ is an isomorphism $K[X] → K[X]_F^2$ . Combining this with $(M_σ)_τ = M_σ ∘ τ$ gives $K[∂]_F = K[X]_F^2 ≅ K[X]$ : transforming twice returns the polynomial module, the algebraic counterpart of the classical inversion formula.

Transvections and the exponential factor

Take $g_i ∈ K[x_i]$ and $σ(x_i) = x_i$ , $σ(∂_i) = ∂_i + g_i$ . The relations survive: $[∂_i + g_i, x_j] = δ_ij$ , and $[∂_i + g_i, ∂_j + g_j] = ∂ g_j/∂ x_i - ∂ g_i/∂ x_j = 0$ because $g_i$ depends only on $x_i$ . Then $K[X]_σ ≅ A_n/ ∑ _i A_n(∂_i - g_i)$ , and as a space of functions it is $e^HK[X]$ with $∂ H/∂ x_i = g_i$ .

The family $σ_r$ Coutinho (5.2.3)

For each positive integer $r$ put $σ_r(x_i) = x_i$ and $σ_r(∂_i) = ∂_i - x_i^,r$ . Each is an automorphism, with inverse $∂_i ↦ ∂_i + x_i^,r$ , so

K[X]_σ_r ; ≅ ; A_n / ∑ _i=1^n A_n (∂_i + x_i^,r ),

a simple module for every $r$ . These are pairwise non-isomorphic; the proof compares degrees in the differential equation that any candidate isomorphism would have to satisfy, and is given on the isomorphism page. In function language, $K[X]_σ_r$ is $! (- ∑ _i x_i^,r+1/(r+1) )K[X]$ , and these subspaces of the function space are genuinely different for different $r$ .

Twisting by a polynomial change of variables

If $F = (F_1, … ,F_n)$ is a polynomial automorphism of affine $n$ -space, it induces an automorphism of $A_n$ sending $x_i ↦ F_i$ and $∂_i$ to the corresponding combination of partials given by the inverse Jacobian. Twisting $K[X]$ by it gives $K[X]$ back, since the substitution $f ↦ f ∘ F$ is an isomorphism. This is the standard example of a non-inner automorphism whose twist is nevertheless trivial — proving that $M_σ$ is not isomorphic to $M$ always needs an argument specific to $σ$ .

Twisting a non-simple module

Twisting says nothing about simplicity that was not already true. Applying $ℱ$ to $A_1$ itself gives $(A_1)_F ≅ A_1/F^-1(0) = A_1$ , so the regular module is fixed. Applying $ℱ$ to $K[x][1/x]$ gives a module with the same composition length $2$ , whose factors are the Fourier transforms of $K[x]$ and of the delta module — that is, $K[∂]$ and $A_1/A_1∂ ≅ K[x]$ , with the roles exchanged.

Worked Example

The twist $σ_1$ on $A_1$ , from all four sides

Step 1 - the automorphism
Let $n = 1$ , $K = Q$ , and $σ(x) = x$ , $σ(∂) = ∂ - x$ . To see that $σ$ extends to an automorphism, check the single defining relation on the images:

$[σ(∂), σ(x)] = [∂ - x, x] = [∂, x] - [x,x] = 1 .$

So $σ$ is an algebra endomorphism; it is bijective because $∂ ↦ ∂ + x$ , $x \mapsto x$ is a two-sided inverse. Note $σ^-1(∂) = ∂ + x$ , which is what (5.13) will need.
Step 2 - the twisted action, computed directly
On $K[x]_σ$ the variable acts as before and $∂ ∙ f = σ(∂)f = f' - xf$ . Concretely:

$u$ $∂ ∙ u$
$1$ $-x$
$x$ $1 - x^2$
$x^2$ $2x - x^3$
$-x$ $-1 + x^2$

In particular $∂ ∙ (∂ ∙ 1) = ∂ ∙ (-x) = x^2 - 1$ , so $∂^2 ∙ 1 = x^2 - 1$ : the twisted action raises degree rather than lowering it.
Step 3 - the presentation, via (5.13)
$K[x] ≅ A_1/A_1∂$ , so $K[x]_σ ≅ A_1/σ^-1(A_1∂) = A_1/A_1(∂ + x)$ . Check it against Step 2: in $A_1/A_1(∂+x)$ the generator $\bar{1}$ satisfies $∂ 1 = -x 1$ , matching the first row of the table. Using $σ$ instead of $σ^-1$ would have given $A_1/A_1(∂ - x)$ , whose generator satisfies $∂ 1 = +x 1$ — the wrong sign, and a concrete demonstration that the inverse matters.
Step 4 - simplicity, made explicit
By part (2) of the preservation proposition, $K[x]_σ$ is simple. It is worth seeing the operator that does the work. The element $σ^-1(∂) = ∂ + x$ of $A_1$ acts on $K[x]_σ$ by $(∂ + x) ∙ f = σ(∂ + x)f = ∂ f = f'$ , that is, as ordinary differentiation. So for $f = x^3 - 2x + 5$ ,

$(∂ + x)^3 ∙ f = f''' = 6,$

and $16 g,(∂+x)^3 ∙ f = g$ for any target $g$ . The generating operator of the untwisted module has simply been pulled back through $σ$ .
Step 5 - the analytic identification
Here $g_1 = -x$ , so $H = -x^2/2$ and the prediction is $K[x]_σ ≅ e^-x^2/2K[x]$ . Verify the intertwining directly:

$(d)/(dx) ( e^-x^2/2 f ) = e^-x^2/2 ( f' - x f ) = e^-x^2/2,(∂ ∙ f),$

so multiplication by $e^-x^2/2$ carries the twisted action to the ordinary one on the space of functions $e^-x^2/2Q[x]$ . Since $e^-x^2/2$ is not a polynomial, this identification lives outside $K[x]$ — which is exactly why the twist can fail to be isomorphic to $K[x]$ for other choices of $σ$ .

$u$	$∂ ∙ u$
$1$	$-x$
$x$	$1 - x^2$
$x^2$	$2x - x^3$
$-x$	$-1 + x^2$

Result

$K[x]_σ$ for $σ(∂) = ∂ - x$ is the simple cyclic module $A_1/A_1(∂ + x)$ , with $∂$ acting as $f ↦ f' - xf$ , realised concretely as the space of functions $e^-x^2/2Q[x]$ . The operator $∂ + x$ acts as ordinary differentiation, so every generating computation on $K[x]$ transfers verbatim.

Applications and Industry Use

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

Getting a second theorem free. Every result proved for $K[X]$ transfers to $K[∂]$ and to every other twist without a new proof. Simplicity of $K[∂]$ , its cyclicity and its dimension all come from the corresponding facts for $K[X]$ via the Fourier automorphism.
Constant-coefficient equations. Under $ℱ$ a differential operator with constant coefficients becomes a polynomial, so questions about solvability of $P(∂)f = 0$ become questions about the polynomial $P$ acting on $K[∂]$ . This is the algebraic shadow of turning a differential equation into an algebraic one by Fourier transform, and it is why the automorphism carries that name.
Building simple modules. Since $A_n$ has an enormous automorphism group and its simple modules resist classification, twisting a known simple module is the main elementary source of new ones; the family $K[X]_σ_r$ is the standard illustration.
Exponential twists in D-module practice. The modules $A_n/ ∑ _i A_n(∂_i - g_i)$ encode $e^H$ with $∇ H = g$ , and they are what appear when one studies exponential integrals, irregular singularities and the Fourier transform of a $D$ -module in the wider theory.
Side changing. The same idea with an anti-automorphism in place of an automorphism converts left modules into right modules; that is the content of the transposition and the side-changing functors.

Design Considerations

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

Twist or change rings

Twisting is the special case of restriction of scalars along a ring map where the map happens to be an automorphism of the ring itself. Phrasing it as change of rings makes the functoriality obvious but hides the fact that the underlying group is unchanged; phrasing it as a twist keeps the concrete picture but must carry the inverse in (5.13) explicitly. For hand computation the twist language is better; for functorial arguments the change-of-rings language is.

Which presentation to record

A twisted module can be recorded either as the pair $(M, σ)$ or as the single quotient $A_n/σ^-1(J)$ . The pair keeps the provenance and makes composition easy, since $(M_σ)_τ = M_σ ∘ τ$ ; the quotient is what a computer algebra system can accept as input. Convert once, at the point where a computation begins, and record which convention was used.

Choosing the automorphism

The useful automorphisms of $A_n$ fall into recognisable families: the Fourier automorphism and its powers, the transvections $∂_i ↦ ∂_i + g_i$ with $g$ a gradient, the symplectic linear substitutions, and the automorphisms induced by polynomial automorphisms of affine space. The last of these never produce anything new from $K[X]$ ; the transvections do.

Failure Modes and Common Mistakes

Dropping the inverse in the cyclic-module formula

$(R/J)_σ ≅ R/σ^-1(J)$ , not $R/σ(J)$ , under the convention $a ∙ u = σ(a)u$ . The two agree only when $σ^2$ fixes $J$ . Step 3 of the worked example gives the smallest case where the difference is visible: $A_1/A_1(∂+x)$ versus $A_1/A_1(∂-x)$ , which are non-isomorphic modules.

Assuming a twist is a new module

Twisting may change the isomorphism class; it often does not. Twisting by the parity automorphism, by any inner automorphism, or by the automorphism induced by a polynomial change of coordinates all return the polynomial module up to isomorphism. Showing $M_σ$ is not isomorphic to $M$ requires an actual obstruction — in the family $σ_r$ it is a degree count in a first-order differential equation.

Twisting by an endomorphism that is not an automorphism

The construction needs surjectivity of $σ$ in two places: to identify the submodule lattices, and to invert $σ$ in (5.13). Composing the action with a non-surjective endomorphism gives a module, but the preservation results fail. For $A_n$ this distinction is delicate: every endomorphism of $A_n$ is injective, since the kernel is a two-sided ideal of a simple ring, and whether every endomorphism is surjective is exactly the Dixmier conjecture — still open.

Twisting the wrong side

For a right module the same recipe reads $u ∙ a = u,σ(a)$ , and the corresponding statement is $(R/J)_σ ≅ R/σ^-1(J)$ with $J$ a right ideal. Mixing the two conventions inside one argument silently replaces $σ$ by $σ^-1$ and produces statements that are off by exactly one inversion — the commonest error in this area, and one that the Fourier example will always detect.

Expecting numerical invariants to be untouched

Length, number of generators and dimension survive twisting. Anything defined through a specific filtration need not: multiplicity, the characteristic variety, and the order of a presenting operator can all change, because an automorphism of $A_n$ generally does not preserve the Bernstein or the order filtration. The automorphism $σ_r$ raises the Bernstein degree of $∂_i$ from $1$ to $r$ .

Historical Notes

Twisting a module along a ring automorphism is old and general — it is the reason one speaks of a module being defined "up to a semilinear change" — and it appears throughout the representation theory of rings with large automorphism groups. Its prominence for the Weyl algebra is due to two facts discovered relatively early: that $A_n$ has no non-trivial inner automorphisms, and that its automorphism group is large. Dixmier's 1968 paper determined generators for $Aut(A_1)$ , showing it is generated by the transvections in $x$ and in $∂$ , and posed the question, still open, of whether every endomorphism of $A_n$ is an automorphism.

The Fourier automorphism itself is much older in spirit: it is the algebraic residue of the classical Fourier transform, under which differentiation becomes multiplication by the dual variable. In the analytic theory this correspondence is the Fourier-Laplace transform of a $D$ -module, developed by Malgrange, Katz and others, where the same automorphism of $A_n$ is used to define the transform of an algebraic $D$ -module on affine space. The elementary version presented here is that theory stripped to the point where it is a two-line definition.

Comparison

What survives the passage from $M$ to $M_σ$ .
Feature	Behaviour under twisting
underlying abelian group	unchanged
lattice of submodules	unchanged as a set of subgroups
simple, torsion, cyclic, finitely generated	preserved in both directions
composition length	preserved
$Hom$ groups and exact sequences	preserved; the functor is exact and additive
annihilator of an element	replaced by $σ^-1$ of it
presenting left ideal $J$	replaced by $σ^-1(J)$
dimension $d(M)$	preserved
multiplicity $e(M)$ , characteristic variety	not preserved in general
isomorphism class of $M$	may change; this is the point

Key Takeaways

Key points

$M_σ$ has the same elements as $M$ and the action $a ∙ u = σ(a)u$ ; nothing is constructed, only relabelled.
Submodules of $M$ and of $M_σ$ coincide as subgroups, so simple, torsion, cyclic, finitely generated and length all transfer in both directions.
For cyclic modules the rule is $(R/J)_σ ≅ R/σ^-1(J)$ ; the inverse is essential and is where most errors occur.
Twisting by the Fourier automorphism $x_i ↦ ∂_i$ , $∂_i ↦ -x_i$ turns $K[X]$ into $K[∂] = A_n/ ∑ _i A_n x_i$ .
Twisting by inner automorphisms changes nothing; over $A_n$ the only inner automorphism is the identity, so every automorphism is a genuine candidate.
The transvections $∂_i ↦ ∂_i + g_i$ , with $g$ a gradient, realise $K[X]_σ$ as the function module $e^HK[X]$ .
The isomorphism class may change, and the family $K[X]_σ_r$ exploits that to give infinitely many pairwise non-isomorphic simple modules.

FAQs

M_σ

ever equal to

M

rather than merely isomorphic to it?

They are equal as abelian groups always, and equal as modules exactly when $σ$ acts trivially on the image of $R$ in $End(M)$ — for a faithful module, exactly when $σ = id$ . Being isomorphic is a weaker and more interesting condition, and it can hold for non-trivial $σ$ , as the parity automorphism shows.

Why is the inverse needed in

(R/J)_σ ≅ R/σ^-1(J)

The map $R → (R/J)_σ$ that sends $b$ to $b ∙ (1+J)$ lands on $σ(b) + J$ , so $b$ is in the kernel when $σ(b) ∈ J$ , that is when $b ∈ σ^-1(J)$ . Any statement with $σ(J)$ has read the map in the wrong direction.

Does twisting preserve holonomicity?

Yes. Dimension equals Gelfand-Kirillov dimension for finitely generated $A_n$ -modules, and that invariant is computed from an arbitrary finite-dimensional generating subspace of the algebra, so it is unchanged by an automorphism. Multiplicity, being tied to the Bernstein filtration, is not protected by the same argument.

Can I twist by a mere endomorphism?

You can define the action, but the theory breaks. Identifying the submodule lattices uses surjectivity of $σ$ , and (5.13) uses $σ^-1$ . For $A_n$ whether the distinction is vacuous is the Dixmier conjecture: every endomorphism is injective, but surjectivity is unproved.

How does twisting interact with direct sums and exact sequences?

Perfectly. The functor is the identity on underlying groups and on maps, so it is additive and exact: $(M ⊕ M')_σ = M_σ ⊕ M'_σ$ , and a sequence is exact before twisting exactly when it is exact after.

Why is the Fourier automorphism called that?

Because it does algebraically what the Fourier transform does analytically: it converts differentiation into multiplication by the dual variable and back, up to sign. The sign convention $∂_i ↦ -x_i$ is chosen so that the commutation relations are preserved; it makes the automorphism have order $4$ , matching the classical fact that the Fourier transform applied four times is the identity.

Are the twists

K[X]_σ_r

really different from each other?

Yes, for distinct positive $r$ . An isomorphism $K[X]_σ_r → K[X]_σ_t$ would send the generator $1$ to some non-zero polynomial $f$ satisfying $∂ f/∂ x_i = (x_i^,t - x_i^,r)f$ ; comparing degrees on the two sides for $r < t$ gives a contradiction. The full argument is on the isomorphism page.

Is twisting the same thing as tensoring with a rank-one module?

In spirit, in the transvection case: $K[X]_σ$ behaves like " $e^H ⊗ K[X]$ ". But $e^H$ is not an $A_n$ -module of the kind one can tensor with over $A_n$ , and the general construction here uses no tensor product at all. Keep them separate; the twist is defined for any ring and any automorphism, with no assumptions.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 5 §2, Propositions (5.2.1) and (5.2.2), Theorem (5.2.3), and Exercise 5.4.6.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - generators for the automorphism group of $A_1$ and the endomorphism question.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, revised edition, Graduate Studies in Mathematics 30, American Mathematical Society, 2001 - Ch. 8, for Gelfand-Kirillov dimension and its invariance under automorphisms.
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, North-Holland, 1979 - Ch. 1, for filtrations and their behaviour under automorphisms.
B. Malgrange, Transformation de Fourier géométrique, Séminaire Bourbaki, exposé 692, Astérisque 176 (1989), 133-150 - the Fourier transform of a D-module in the geometric setting.
N. M. Katz, Exponential Sums and Differential Equations, Annals of Mathematics Studies 124, Princeton University Press, 1990 - Fourier transform and exponential twists of D-modules.
P. M. Cohn, Algebra, Volume 1, second edition, Wiley, 1982 - Ch. 10, for the module-theoretic background and the isomorphism theorems.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization - notation for maps, composition and partial derivatives.

AI Suggested Questions

Verify that $ℱ$ preserves all the defining relations of $A_2$ and compute $F(x_1∂_2 - x_2∂_1)$ .
Show that $(M_σ)_τ = M_σ ∘ τ$ directly from the definition, and deduce $(M_σ)_σ^-1 = M$ .
Compute $A_1/A_1(∂ + x)$ explicitly as a vector space with its action, and confirm it is simple.
Give an automorphism of $A_1$ for which the twist of $K[x]$ is isomorphic to $K[x]$ , and one for which it is not.
Work out the Fourier transform of the delta module $A_1/A_1x$ and identify the result.
Explain why $e(M_σ)$ can differ from $e(M)$ by computing a good filtration on $A_1/A_1(∂ + x^2)$ .
Prove that the only inner automorphism of $A_n$ is the identity, using the fact that the units of $A_n$ are the non-zero scalars.