Automorphisms of the Weyl Algebra

Q: Is the Dixmier conjecture known for n = 1?

No. The automorphism group of A_1 has been known since 1968, but that does not decide whether some endomorphism fails to be surjective. DC_1 is open.

Q: Where exactly does the proof use Δ F = 1 rather than Δ F ≠ 0?

In the very first step. The derivations D_i are defined by dividing by Δ , so a priori they act on K[X,Δ^-1] . They restrict to K[X] only when Δ is a unit of K[X] , that is a non-zero constant. And local nilpotence genuinely fails on localisations - ∂_x is not locally nilpotent on K[x,x^-1] - so one cannot simply work upstairs.

Q: Are all automorphisms of A_n of the form (ad_a) composed with symplectic linear maps?

Unknown for n ≥ 2 . For n = 1 Dixmier's generators are all of that shape, so the answer there is yes. For higher n the analogous statement is the tameness question, and by analogy with Aut(K^3) , where Shestakov and Umirbaev found non-tame automorphisms, one should not assume it.

Q: What happens to this circle of ideas in characteristic p?

It collapses, and productively so. A_n becomes an Azumaya algebra over its centre, a polynomial ring in 2n variables, and endomorphisms can have kernels. The Jacobian conjecture is false. Paradoxically, reduction modulo p is the main tool in the proof of the converse implication JC_2n ⇒ DC_n : the centre carries a Poisson bracket that remembers the characteristic-zero commutator.

Q: Does the theorem produce an explicit inverse for F?

Only implicitly. It concludes K[F_1, … ,F_n] = K[X] , so each x_i is some polynomial in the F_j , and those polynomials are the coordinate functions of F^-1 . Extracting them requires expressing x_i in terms of the F_j , which the slice theorem's induction does supply constructively once local nilpotence is known - but local nilpotence is what is missing.

Overview

The simplicity of $A_n$ has an immediate consequence for maps out of it: any $K$ -algebra homomorphism $φ : A_n → A_n$ has kernel a two-sided ideal, and the kernel cannot be everything because $φ(1) = 1$ . So $φ = 0$ and every endomorphism of $A_n$ is injective. Whether every endomorphism is also surjective is Dixmier's question, and it is open - not just for large $n$ , but for $n = 1$ .

That shape - injectivity free, surjectivity hard - is exactly the shape of the Jacobian conjecture. There too the comorphism $F^#$ of a polynomial map with non-vanishing Jacobian determinant is injective for elementary reasons, and the entire content of the conjecture is that $K[F_1, … ,F_n] = K[x_1, … ,x_n]$ , that is, surjectivity.

The theorem of this page connects the two. Given a polynomial map $F$ with $Δ F = 1$ , the Jacobian derivations $D_i = ∂/∂ F_i$ satisfy exactly the Weyl relations against the $F_j$ , so there is an endomorphism $φ$ of $A_n$ with $φ(x_i) = F_i$ and $φ(∂_i) = D_i$ . If $φ$ happens to be surjective, then local nilpotence of $ad_∂_i$ transports across $φ$ to local nilpotence of $ad_D_i$ , hence of $D_i$ as a derivation of $K[X]$ - and the slice theorem finishes the job. So the Dixmier conjecture for $A_n$ implies the Jacobian conjecture in $n$ variables.

What is genuinely known about $Aut(A_n)$ is confined to $n = 1$ , where Dixmier gave generators in 1968 and Makar-Limanov later identified the group with the Jacobian-one automorphisms of the affine plane. For $n ≥ 2$ the group is not known. It contains the symplectic linear group $Sp_2n(K)$ and all exponentials $(ad_a)$ with $ad_a$ locally nilpotent, and nobody can say whether that is everything.

Definition

Let $K$ have characteristic zero and let $A_n = A_n(K)$ be the $n$ -th Weyl algebra, with generators $x_1, … ,x_n,∂_1, … ,∂_n$ and defining relations

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

(4.13)

Endomorphisms and automorphisms

An endomorphism of $A_n$ is a $K$ -algebra homomorphism $φ : A_n → A_n$ with $φ(1) = 1$ . It is an automorphism if it is bijective; the automorphisms form a group $Aut_K(A_n)$ .

By the presentation of P0, an endomorphism is the same thing as a choice of $2n$ elements $X_1, … ,X_n,Y_1, … ,Y_n ∈ A_n$ satisfying (4.13) with $X$ in place of $x$ and $Y$ in place of $∂$ . No further condition is imposed and none is available: the relations are all there is.

Inner derivations and their exponentials

For $a ∈ A_n$ define $ad_a : A_n → A_n$ by $ad_a(b) = [a,b] = ab - ba$ . This is a $K$ -derivation of $A_n$ , but not an algebra homomorphism. If $ad_a$ is locally nilpotent, then

(ad_a) = ∑ _k ≥ 0 (ad_a)^kk!

is a well-defined $K$ -algebra automorphism of $A_n$ , with inverse $(-ad_a)$ .

The Dixmier conjectureDixmier 1968, Problème 11.1

$DC_n$ : every $K$ -algebra endomorphism of $A_n(K)$ , $char,K = 0$ , is an automorphism. Equivalently, since injectivity is automatic, every endomorphism is surjective. The conjecture is open for every $n ≥ 1$ .

Core Concepts

Why injectivity is free and surjectivity is not

$A_n$ is a simple ring in characteristic zero: its only two-sided ideals are $0$ and $A_n$ . A ring homomorphism's kernel is a two-sided ideal, and $φ(1) = 1 ≠ 0$ rules out the whole ring. So $φ = 0$ . Nothing comparable constrains the image: $φ(A_n)$ is only a subalgebra, and subalgebras of $A_n$ can be small - $K[x_1]$ is one.

It is worth appreciating how little the analogous commutative statement gives. The endomorphism $x ↦ x^2$ of $K[x]$ is injective and not surjective. Simplicity is doing real work in the non-commutative case, but only half the work.

Why $ad_∂_i$ is locally nilpotent

Give $A_n$ the total degree in which every $x_i$ and every $∂_i$ has degree one - the Bernstein degree. Then commuting with $∂_i$ lowers degree by one:

(ad_∂_i(b) ) ≤ b - 1.

(4.14)

It is enough to check this on a monomial $x^α∂^β$ , where $[∂_i, x^α∂^β] = α_i x^α - e_i∂^β$ . Hence $(ad_∂_i)^k(b) = 0$ for $k > b$ , and $ad_∂_i$ is locally nilpotent. The same holds for $ad_x_i$ ; it fails badly for $ad_x_i∂_i$ , which preserves degree and has $x_i$ as an eigenvector.

Transport of local nilpotence

The single identity that makes the reduction work is that $φ$ intertwines inner derivations:

φ (ad_a(b) ) = φ(ab - ba) = ad_φ(a) (φ(b) ).

(4.15)

So $φ$ carries the locally nilpotent operator $ad_∂_i$ to $ad_φ(∂_i)$ on the image of $φ$ . If the image is everything - that is, if $φ$ is surjective - then $ad_φ(∂_i)$ is locally nilpotent on all of $A_n$ . Surjectivity is precisely what is needed and precisely what is missing.

Construction and Proof

Endomorphisms are injectiveCoutinho (2.2.2)

Every $K$ -algebra endomorphism of $A_n$ , $char,K = 0$ , is injective.

Proof

$ker φ$ is a two-sided ideal of $A_n$ . Since $φ(1) = 1 ≠ 0$ , we have $1 ∉ φ$ , so $φ ≠ A_n$ . By simplicity of $A_n$ in characteristic zero the only remaining possibility is $φ = 0$ .

The Dixmier conjecture implies the Jacobian conjectureCoutinho (4.4.2); Vaserstein-Katz, in Bass-Connell-Wright 1982

Fix $n ≥ 1$ and let $K$ have characteristic zero. If every endomorphism of $A_n(K)$ is an automorphism, then every polynomial map $F : K^n → K^n$ with $Δ F = 1$ satisfies $K[F_1, … ,F_n] = K[x_1, … ,x_n]$ , and hence has a polynomial inverse.

Proof

Step 1: build the endomorphism. Since $Δ F = 1$ is a unit, the derivations $D_i$ of (4.16) are derivations of $K[X]$ itself, not merely of $K[X,Δ^-1]$ . Coutinho's Lemma (4.4.1) gives $D_i(F_j) = δ_ij$ and $[D_i,D_j] = 0$ . Viewing $F_j$ and $D_i$ inside $A_n$ , and using $[D_i, F_j] = D_i(F_j)$ from (4.17), the $2n$ elements $F_1, … ,F_n,D_1, … ,D_n$ satisfy the relations (4.13). By the presentation of $A_n$ there is a unique endomorphism $φ$ with $φ(x_i) = F_i$ , $φ(∂_i) = D_i$ .

Step 2: use the hypothesis. By assumption $φ$ is an automorphism, in particular surjective.

Step 3: transport local nilpotence. Let $c ∈ A_n$ be arbitrary and write $c = φ(b)$ . By (4.14) there is a $k$ , namely any $k > b$ , with $(ad_∂_i)^k(b) = 0$ . Applying $φ$ and using (4.15) $k$ times,

(ad_D_i)^k(c) = (ad_φ(∂_i))^k (φ(b) ) = φ ((ad_∂_i)^k(b) ) = 0.

So $ad_D_i$ is locally nilpotent on $A_n$ .

Step 4: descend to $K[X]$ . Restrict to $c = g ∈ K[X] ⊂ A_n$ . By (4.17), $(ad_D_i)^k(g) = D_i^k(g)$ , so $D_i^k(g) = 0$ for some $k$ . Hence each $D_i$ is a locally nilpotent derivation of $K[X]$ .

Step 5: apply the structure theorem. The $D_i$ are commuting locally nilpotent derivations of $K[X]$ with $D_i(F_j) = δ_ij$ , so the slice theorem gives $K[X] = R[F_1, … ,F_n]$ , where $R = ⋂ _i D_i$ . Finally $R = K$ : the $F_i$ are algebraically independent over $R$ and $K[X]$ has transcendence degree $n$ over $K$ , so $R$ is algebraic over $K$ ; but an element of $K[X]$ of positive degree has powers of unbounded degree and is therefore transcendental over $K$ . So $R = K$ and $K[F_1, … ,F_n] = K[X]$ , which is the Jacobian conjecture in the form Coutinho (4.2.3). Bijectivity of $F$ then follows because $F^#$ is an isomorphism of rings.

Where the hypothesis is actually used

Only in Step 2, and only surjectivity is used - injectivity of $φ$ is free by the lemma. So the theorem may be sharpened: if the particular endomorphism $φ_F$ of (4.16) is surjective, then $F$ is invertible. One does not need the full Dixmier conjecture, only its instance at $φ_F$ .

Key Equations

The endomorphism attached to a polynomial map $F$ with $Δ F = 1$ is defined on generators by

φ_F(x_i) = F_i, φ_F(∂_i) = D_i, D_i(g) = J(F_1, … ,F_i-1,g,F_i+1, … ,F_n),

(4.16)

legitimate because the right-hand sides satisfy the relations (4.13), by Coutinho (4.4.1).

Inside $A_n$ the derivation $D_i$ acts on polynomials by commutator, which is the bridge between the non-commutative and the commutative statements:

ad_D_i(g) = [D_i, g] = D_i(g), hence (ad_D_i)^k(g) = D_i^k(g) (g ∈ K[X]).

(4.17)

Two standard families of automorphisms. For $f ∈ K[x_1, … ,x_n]$ and $h ∈ K[∂_1, … ,∂_n]$ ,

σ_f : x_i ↦ x_i, ∂_i ↦ ∂_i + (∂ f)/(∂ x_i); τ_h : ∂_i ↦ ∂_i, x_i ↦ x_i - (∂ h)/(∂ ∂_i).

(4.18)

Both are of the form $(ad_a)$ : take $a = -f$ for $σ_f$ and $a = -h$ for $τ_h$ . The images must be gradients - that is what makes the two relations $[∂_i,∂_j] = 0$ and $[x_i,x_j] = 0$ survive.

The linear automorphisms are governed by the symplectic form. Writing $u = (x_1, … ,x_n,∂_1, … ,∂_n)$ , a linear substitution $u \mapsto M u$ preserves (4.13) exactly when

M^T Ω, M = Ω, Ω = pmatrix 0 & I_n -I_n & 0 pmatrix,

(4.19)

that is, exactly when $M ∈ Sp_2n(K)$ .

Variable Definitions

$K$: the ground field, of characteristic zero
$A_n$: the $n$ -th Weyl algebra over $K$
$φ$ , $ψ$: $K$ -algebra endomorphisms of $A_n$
$Aut_K(A_n)$: the group of $K$ -algebra automorphisms of $A_n$
$ad_a$: the inner derivation $b ↦ [a,b]$
$deg b$: the total (Bernstein) degree of $b ∈ A_n$ , with all $x_i$ and $∂_i$ of degree one
$F$: a polynomial map $K^n → K^n$ with coordinate functions $F_1, … ,F_n$
$Δ F$: the Jacobian determinant $J(F)$
$D_i$: the derivation $∂/∂ F_i$ of $K[X]$ built from $F$ by Cramer's rule
$DC_n$: the Dixmier conjecture for $A_n$
$JC_n$: the Jacobian conjecture in $n$ variables
$Sp_2n(K)$: the symplectic group, the linear automorphisms of $A_n$

Properties and Behaviour

No non-trivial inner automorphisms

The units of $A_n$ are exactly $K^ ×$ . Indeed the associated graded ring of $A_n$ for the Bernstein filtration is a polynomial ring, hence a domain, so $(ab) = a + b$ ; if $ab = 1$ then $a = b = 0$ . Consequently conjugation by a unit is trivial, and every non-identity automorphism of $A_n$ is outer. The automorphisms $(ad_a)$ are outer even though they are built from a commutator.

The case $n = 1$ Dixmier 1968; Makar-Limanov 1984

$Aut_K(A_1)$ is generated by the two families of (4.18): $σ_f : (x,∂) ↦ (x, ∂ + f'(x))$ and $τ_h : (x,∂) ↦ (x - h'(∂), ∂)$ . Makar-Limanov proved that $Aut_K(A_1)$ is isomorphic to the group of polynomial automorphisms of the affine plane with Jacobian determinant $1$ , and that it decomposes as an amalgamated free product in the same way that group does.

This is a description of the automorphisms, not of the endomorphisms. It does not settle $DC_1$ , because it says nothing about an endomorphism that fails to be surjective.

What is known for $n ≥ 2$

$Aut_K(A_n)$ contains $Sp_2n(K)$ by (4.19), and all exponentials $(ad_a)$ with $ad_a$ locally nilpotent.
No generating set is known, and it is not known whether every automorphism preserves the Bernstein filtration up to a shift.
Belov-Kanel and Kontsevich conjectured that $Aut_K(A_n)$ is isomorphic to the group of polynomial symplectomorphisms of $K^2n$ , and constructed a canonical map relating the two using reduction modulo $p$ . The conjecture is open for $n ≥ 2$ .
Tsuchimoto, and independently Belov-Kanel and Kontsevich, proved that $JC_2n$ implies $DC_n$ . Together with the theorem above this makes the two families of conjectures stably equivalent: all Jacobian conjectures hold if and only if all Dixmier conjectures hold.

Automorphisms act on modules

Any $σ ∈ Aut_K(A_n)$ turns a left $A_n$ -module $M$ into a new one $M^σ$ by $a ⋅ m := σ(a)m$ ; see twisting a Weyl module. Twisting preserves simplicity, finite generation, dimension and multiplicity, so the group $Aut_K(A_n)$ acts on every invariant of the module category. This is how the largely unknown group makes itself felt in the rest of the theory.

Examples and Special Cases

The Fourier transform

$x_i ↦ ∂_i$ , $∂_i ↦ -x_i$ preserves (4.13): $[-x_i, ∂_j] = δ_ij$ . This is the algebraic Fourier transform, an automorphism of order four, and it is the element $pmatrix 0 & I -I & 0 pmatrix$ of $Sp_2n(K)$ . It exchanges the roles of position and momentum, and on modules it exchanges the polynomial module with the module generated by the Dirac delta.

Scaling

For $λ ∈ K^ ×$ , $x_i ↦ λ x_i$ , $∂_i ↦ λ^-1∂_i$ is an automorphism. It is not of the form $(ad_a)$ , because $ad_x_i∂_i$ is not locally nilpotent; the corresponding one-parameter group is a torus, not a $G_a$ .

A triangular exponential

Take $a = x^m ∈ A_1$ . Then $ad_a(∂) = [x^m,∂] = -m x^m-1$ and $ad_a(x) = 0$ , so $(ad_a)^2 = 0$ on both generators and $(ad_a)$ is $x \mapsto x$ , $∂ ↦ ∂ - mx^m-1$ . This is the automorphism $σ_f$ of (4.18) with $f = -x^m$ , consistent with the rule $a = -f$ .

An injective endomorphism of a commutative ring that is not surjective

$K[x] → K[x]$ , $x ↦ x^2$ , is injective with image $K[x^2]$ . Nothing in the commutative world forces injective endomorphisms to be surjective, so no soft argument can prove the Dixmier conjecture; whatever proof exists must use the Weyl relations specifically.

A composite of two shears

$F = (x_1 - x_2^2, x_2 + (x_1 - x_2^2)^2)$ is the composite of $(x_1,x_2) ↦ (x_1 - x_2^2, x_2)$ with $(y_1,y_2) ↦ (y_1, y_2 + y_1^2)$ . Each factor is triangular, so $F$ is invertible with polynomial inverse, and $Δ F = 1 ⋅ (1 - 4x_2(x_1-x_2^2)) - (-2x_2) ⋅ 2(x_1-x_2^2) = 1$ . The induced endomorphism of $A_2$ is the composite of two exponentials of the kind in the worked example. Every map with $Δ = 1$ that has been written down explicitly turns out to be invertible, which is what makes the conjecture plausible and searches for counterexamples unrewarding.

Worked Example

The endomorphism of $A_2$ attached to a shear, and its inverse

Step 1 - the map and its derivations
Take $n = 2$ and $F = (x_1 + x_2^2, x_2)$ on $K^2$ . Then

$J(F) = pmatrix 1 & 2x_2 0 & 1 pmatrix, Δ F = 1.$

Cramer's rule (4.16) gives $D_1(g) = pmatrix g_1 & g_2 0 & 1 pmatrix = g_1$ and $D_2(g) = pmatrix 1 & 2x_2 g_1 & g_2 pmatrix = g_2 - 2x_2 g_1$ , where $g_i = ∂ g/∂ x_i$ . So $D_1 = ∂_1$ and $D_2 = ∂_2 - 2x_2∂_1$ .
Step 2 - verify the Weyl relations
Write $X_1 = x_1 + x_2^2$ , $X_2 = x_2$ , $Y_1 = ∂_1$ , $Y_2 = ∂_2 - 2x_2∂_1$ . Then:
- $[Y_1, X_1] = [∂_1, x_1 + x_2^2] = 1$ , and $[Y_1,X_2] = [∂_1,x_2] = 0$ .
- $[Y_2, X_1] = [∂_2, x_2^2] - 2x_2[∂_1, x_1] = 2x_2 - 2x_2 = 0$ .
- $[Y_2, X_2] = [∂_2, x_2] - 2x_2[∂_1,x_2] = 1 - 0 = 1$ .
- $[X_1,X_2] = 0$ since both lie in $K[x_1,x_2]$ , and $[Y_1,Y_2] = [∂_1, ∂_2] - 2[∂_1, x_2∂_1] = 0$ because $∂_1$ commutes with $x_2$ .
So $[Y_i,X_j] = δ_ij$ and the two families commute internally: there is an endomorphism $φ$ of $A_2$ with $φ(x_i) = X_i$ , $φ(∂_i) = Y_i$ .
Step 3 - exhibit the inverse
The inverse polynomial map is $G = (x_1 - x_2^2, x_2)$ , again with $Δ G = 1$ , and its Cramer derivations are $E_1 = ∂_1$ , $E_2 = ∂_2 + 2x_2∂_1$ . Let $ψ$ be the corresponding endomorphism. Then on generators:

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

and $φ ψ$ fixes $x_2$ and $∂_1$ trivially. The same computation with the signs reversed gives $ψφ = id$ . So $φ ∈ Aut_K(A_2)$ , with $φ^-1 = ψ$ .
Step 4 - recognise $φ$ as an exponential
Put $a = x_2^2,∂_1 ∈ A_2$ . Compute the inner derivation on generators, using $[∂_1, x_1] = 1$ and $[∂_1, x_2] = 0$ :

$ad_a(x_1) = x_2^2, ad_a(x_2) = 0, ad_a(∂_1) = 0, ad_a(∂_2) = [x_2^2,∂_2],∂_1 = -2x_2∂_1.$

Each of these lies in the kernel of $ad_a$ : $ad_a(x_2^2) = 0$ and $ad_a(x_2∂_1) = 0$ . So $(ad_a)^2$ kills all four generators, $ad_a$ is locally nilpotent, and the exponential terminates after two terms:

$(ad_a) : x_1 ↦ x_1 + x_2^2, x_2 ↦ x_2, ∂_1 ↦ ∂_1, ∂_2 ↦ ∂_2 - 2x_2∂_1,$

which is exactly $φ$ . Note $φ^-1 = (-ad_a)$ , matching $ψ$ from Step 3. Note also that $φ$ is not an inner automorphism: $A_n$ has no units beyond $K^ ×$ , so it has no non-trivial inner automorphisms at all; the element $(a)$ that would conjugate does not exist in $A_2$ .
Step 5 - run the theorem's argument on this example
Since $φ$ is surjective, Step 3 of the proof applies. Take $b = x_2$ : then $ad_∂_2(x_2) = 1$ and $(ad_∂_2)^2(x_2) = 0$ , so $(ad_D_2)^2(φ(x_2)) = (ad_D_2)^2(x_2) = 0$ . Directly: $ad_D_2(x_2) = D_2(x_2) = 1$ and $ad_D_2(1) = 0$ . Similarly $D_2(x_1) = -2x_2$ , $D_2^2(x_1) = -2$ , $D_2^3(x_1) = 0$ , so $D_2$ is locally nilpotent on $K[x_1,x_2]$ .

The slice theorem then gives $K[x_1,x_2] = R[F_1,F_2]$ with $R = ∂_1 ∩ D_2 = K[x_2] ∩ D_2 = K$ . So $K[x_1 + x_2^2, x_2] = K[x_1,x_2]$ , which is confirmed directly by $x_1 = F_1 - F_2^2$ .

Result

$F = (x_1 + x_2^2, x_2)$ induces the automorphism $φ = (ad_a)$ of $A_2$ with $a = x_2^2∂_1$ , given on generators by $x_1 ↦ x_1 + x_2^2$ , $x_2 ↦ x_2$ , $∂_1 ↦ ∂_1$ , $∂_2 ↦ ∂_2 - 2x_2∂_1$ ; its inverse is $(-ad_a)$ , the endomorphism attached to $G = (x_1 - x_2^2, x_2)$ . Running the proof of the theorem on this example recovers $K[F_1,F_2] = K[x_1,x_2]$ , as it must, since $F$ was invertible to begin with.

Applications and Industry Use

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

Reduction of the Jacobian conjecture. The theorem of this page is the reason a book on $D$ -modules discusses a problem in affine algebraic geometry at all.
Twisting modules. Automorphisms give new modules from old with the same dimension and multiplicity; see twisting a Weyl module and when two twists are isomorphic. This is the main source of examples of non-isomorphic modules with identical invariants.
Quantum mechanics. $Sp_2n(K)$ acting on $A_n$ is the algebraic shadow of the metaplectic representation, and the Fourier transform automorphism is the algebraic form of the position-momentum duality that motivated the algebra in the first place.
Noncommutative geometry. Belov-Kanel and Kontsevich's symplectomorphism conjecture places $Aut(A_n)$ in a dictionary with symplectic geometry, and the tools developed there - reduction modulo $p$ , Poisson brackets on the centre - are now standard.
Computer algebra. Automorphisms are used to normalise presentations of $D$ -modules before an expensive Gröbner computation: a good change of generators can lower degrees dramatically.

Design Considerations

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

Building an endomorphism: what must be checked

Exactly the relations (4.13) on the $2n$ chosen images, and nothing more. This is the practical value of the presentation of $A_n$ by generators and relations: it converts "define a map on an infinite-dimensional algebra" into a finite list of commutator identities. Getting a surjective map is a different problem entirely, and no finite check is known for it.

Which filtration to reason with

The Bernstein degree, where $x_i$ and $∂_i$ both count one, is the right one here, because it makes (4.14) true: commuting with a generator drops degree. With the order filtration, where $x_i$ has degree zero, $ad_∂_i$ does not lower the filtration level and the local nilpotence argument does not run in the same way.

Exponentials versus explicit substitutions

Writing an automorphism as $(ad_a)$ is more useful than writing it as a substitution table when the goal is to compose, invert, or interpolate: the inverse is $(-ad_a)$ for free, and one obtains a whole one-parameter family $(c,ad_a)$ . It is less useful when the goal is to compute images of specific large elements, where the substitution table is direct.

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.

Checking a candidate endomorphism. Compute the $2n^2 + n(n-1)$ commutators of (4.13) in canonical form. Each is a normal-form computation in $A_n$ ; cost is polynomial in the degrees of the images.
Computing $(ad_a)$ . Iterate $ad_a$ on each generator until zero. If it does not terminate quickly, $ad_a$ is probably not locally nilpotent, but there is no bound that certifies this.
Deciding surjectivity of a given endomorphism. No algorithm is known in general. Individual membership questions can sometimes be settled by a non-commutative Gröbner basis computation in the subalgebra generated by the $φ(x_i)$ and $φ(∂_i)$ , but there is no general procedure that decides membership in an arbitrary subalgebra of $A_n$ .
Inverting a known automorphism. If $φ = (ad_a)$ , use $(-ad_a)$ . Otherwise, solve for images of generators, which is again a subalgebra membership problem.

Weyl algebra arithmetic in canonical form is available in Singular (dmod.lib, nctools.lib), Macaulay2 (Dmodules), SageMath (ore_algebra) and Mathematica (HolonomicFunctions.m). None of them offers a surjectivity test, for the good reason that nobody knows one.

Limits of Validity

Characteristic zero throughout. In characteristic $p$ the algebra $A_n$ is not simple - $x_i^p$ and $∂_i^p$ are central - so endomorphisms need not be injective, and the Jacobian conjecture itself is false: $F(x_1, … ,x_n) = (x_1 - x_1^p,x_2, … ,x_n)$ has identity Jacobian matrix but is not invertible. See the positive characteristic page.
$Δ F = 1$ , not $Δ F ≠ 0$ . The construction of $φ_F$ needs $D_i$ to be derivations of $K[X]$ , which needs $Δ$ to be a unit of $K[X]$ , that is a non-zero constant. With only $Δ ≠ 0$ pointwise, the $D_i$ live on $K[X,Δ^-1]$ and local nilpotence is not available there.
The theorem gives one direction only. Coutinho (4.4.2) proves $DC_n ⇒ JC_n$ . The converse is a much later and much harder result, and it is not a converse at the same $n$ : what is proved is $JC_2n ⇒ DC_n$ .
Nothing here proves either conjecture. Both remain open. In particular the reduction does not make the Jacobian conjecture easier in any known sense; it relocates it.

Failure Modes and Common Mistakes

Concluding that an injective endomorphism of $A_n$ is an automorphism

Injectivity is a theorem; bijectivity is the conjecture. The temptation comes from finite-dimensional linear algebra, where injective implies surjective, and $A_n$ is infinite dimensional. There is a filtered version of the rank-nullity intuition, but an endomorphism need not respect the filtration, so it does not apply.

Assuming an endomorphism preserves degree

$φ_F$ maps $x_i$ to $F_i$ , which can have any degree. Nothing forces $φ(b) = b$ , or even $φ(b) ≥ b$ in a useful way. In the proof, local nilpotence is transported through $φ$ by (4.15), never by a degree estimate on the image side - that is the delicate point of the argument.

Reading $(ad_a)$ as conjugation by $(a)$

Formally $(ad_a)(b) = e^abe^-a$ , but $e^a$ does not exist in $A_n$ : the algebra has no units except non-zero scalars. The automorphism is genuinely outer. The exponential of the derivation is the object that exists; the exponential of the element is not.

Believing the reduction makes the Jacobian conjecture a corollary

It makes it a corollary of an unproved statement that is, if anything, harder. The value of the reduction is structural: it says the obstruction in the Jacobian conjecture is exactly a failure of surjectivity for one specific endomorphism, and it makes non-commutative tools available. It is not progress towards a proof by itself.

Assuming $R = K$ in the structure theorem

The slice theorem outputs $K[X] = R[F_1, … ,F_n]$ with $R$ the common ring of constants, not $K$ . The identification $R = K$ is short but is a separate argument using that $K[X]$ has transcendence degree $n$ and no non-constant polynomial is algebraic over $K$ . Sources sometimes leave it implicit.

Historical Notes

O. H. Keller asked in 1939 whether a polynomial map of the plane over the integers with determinant one is invertible; the question generalised into what is now the Jacobian conjecture. J. Dixmier, in his 1968 paper Sur les algèbres de Weyl, determined $Aut(A_1)$ and closed the paper with a list of problems, of which Problème 11.1 asks whether every endomorphism of $A_n$ is an automorphism.

The link between the two was noticed by L. Vaserstein and V. Katz and recorded in the 1982 survey of Bass, Connell and Wright. Its ingredients were already in place: Wright's 1981 structure theorem for commuting locally nilpotent derivations, and the elementary observation that $ad_∂_i$ lowers degree. Coutinho's Chapter 4 is a compact exposition of that argument, and it is the reason a primer on $D$ -modules opens with affine geometry.

L. Makar-Limanov's 1984 identification of $Aut(A_1)$ with the Jacobian-one automorphism group of the plane made the analogy between the two conjectures structural rather than accidental. Two decades later, Y. Tsuchimoto (2005) and, independently, A. Belov-Kanel and M. Kontsevich (2007) proved the implication in the other direction, $JC_2n ⇒ DC_n$ , using reduction modulo $p$ and the Poisson structure on the centre of $A_n$ in positive characteristic. Both families of conjectures remain open.

Comparison

The two conjectures side by side. The rows show how tightly the reduction matches them.
	Jacobian conjecture $JC_n$	Dixmier conjecture $DC_n$
Object	polynomial map $F : K^n → K^n$	endomorphism $φ$ of $A_n$
Hypothesis	$Δ F = 1$	none beyond being an endomorphism
Injectivity of the algebra map	free, from $Δ F ≠ 0$ (Coutinho (4.2.2))	free, from simplicity of $A_n$
What is conjectured	$K[F_1, … ,F_n] = K[X]$	$φ(A_n) = A_n$
Known cases	$n = 1$ ; degree $≤ 2$ maps	none
Implication proved here	implied by $DC_n$	implies $JC_n$
Converse	$JC_2n$ implies $DC_n$ (Tsuchimoto; Belov-Kanel-Kontsevich)	stably equivalent to the Jacobian family

Key Takeaways

Key points

Every endomorphism of $A_n$ in characteristic zero is injective, because $A_n$ is simple; the Dixmier conjecture is therefore purely a surjectivity statement.
An endomorphism is exactly a choice of $2n$ elements satisfying the Weyl relations, and nothing more needs to be checked.
$ad_∂_i$ lowers the Bernstein degree by one and is therefore locally nilpotent; surjectivity of $φ$ transports that property to $ad_φ(∂_i)$ .
For a polynomial map with $Δ F = 1$ , the elements $F_i$ and $D_i = ∂/∂ F_i$ satisfy the Weyl relations, giving an endomorphism $φ_F$ of $A_n$ .
If $φ_F$ is surjective then the $D_i$ are locally nilpotent, and the slice theorem gives $K[F_1, … ,F_n] = K[X]$ : the Dixmier conjecture for $A_n$ implies the Jacobian conjecture in $n$ variables.
$A_n$ has no non-trivial inner automorphisms, since its only units are the non-zero scalars; the exponentials $(ad_a)$ are outer.
$Aut(A_1)$ is known; $Aut(A_n)$ for $n ≥ 2$ is not, and the two conjecture families are now known to be stably equivalent.

FAQs

Why does simplicity give injectivity but not surjectivity?

Simplicity constrains two-sided ideals, and the kernel of a ring map is a two-sided ideal. The image, by contrast, is only a subalgebra, and simplicity says nothing about subalgebras. $A_n$ has plenty of proper subalgebras - $K[x_1]$ , $K[x_1, … ,x_n]$ , and $A_n-1$ among them.

Is the Dixmier conjecture known for

n = 1

No. The automorphism group of $A_1$ has been known since 1968, but that does not decide whether some endomorphism fails to be surjective. $DC_1$ is open.

Does the Jacobian conjecture imply the Dixmier conjecture?

In a shifted form, yes: Tsuchimoto and, independently, Belov-Kanel and Kontsevich proved that $JC_2n$ implies $DC_n$ . Combined with the theorem on this page, the two families are equivalent when quantified over all $n$ . There is no known equivalence at a fixed $n$ .

Where exactly does the proof use

Δ F = 1

rather than

Δ F ≠ 0

In the very first step. The derivations $D_i$ are defined by dividing by $Δ$ , so a priori they act on $K[X,Δ^-1]$ . They restrict to $K[X]$ only when $Δ$ is a unit of $K[X]$ , that is a non-zero constant. And local nilpotence genuinely fails on localisations - $∂_x$ is not locally nilpotent on $K[x,x^-1]$ - so one cannot simply work upstairs.

Could one prove

DC_n

by a dimension or filtration count?

Not obviously. An endomorphism need not preserve the Bernstein filtration or any shift of it, so there is no target space in which to compare dimensions. The counting arguments that work so well for Bernstein's inequality operate on modules with good filtrations, and an endomorphism of the ring supplies no such filtration.

Are all automorphisms of

A_n

of the form

(ad_a)

composed with symplectic linear maps?

Unknown for $n ≥ 2$ . For $n = 1$ Dixmier's generators are all of that shape, so the answer there is yes. For higher $n$ the analogous statement is the tameness question, and by analogy with $Aut(K^3)$ , where Shestakov and Umirbaev found non-tame automorphisms, one should not assume it.

What happens to this circle of ideas in characteristic

p

It collapses, and productively so. $A_n$ becomes an Azumaya algebra over its centre, a polynomial ring in $2n$ variables, and endomorphisms can have kernels. The Jacobian conjecture is false. Paradoxically, reduction modulo $p$ is the main tool in the proof of the converse implication $JC_2n ⇒ DC_n$ : the centre carries a Poisson bracket that remembers the characteristic-zero commutator.

Does the theorem produce an explicit inverse for

F

Only implicitly. It concludes $K[F_1, … ,F_n] = K[X]$ , so each $x_i$ is some polynomial in the $F_j$ , and those polynomials are the coordinate functions of $F^-1$ . Extracting them requires expressing $x_i$ in terms of the $F_j$ , which the slice theorem's induction does supply constructively once local nilpotence is known - but local nilpotence is what is missing.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 4 §4, Lemma (4.4.1) and Theorem (4.4.2); injectivity of endomorphisms is (2.2.2).
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - the determination of $Aut(A_1)$ and Problème 11.1.
H. Bass, E. H. Connell and D. Wright, The Jacobian conjecture: reduction of degree and formal expansion of the inverse, Bulletin of the American Mathematical Society 7 (1982), 287-330 - the Vaserstein-Katz observation is recorded here.
D. Wright, On the Jacobian conjecture, Illinois Journal of Mathematics 25 (1981), 423-440.
L. Makar-Limanov, On automorphisms of Weyl algebra, Bulletin de la Société Mathématique de France 112 (1984), 359-363.
Y. Tsuchimoto, Endomorphisms of Weyl algebra and p-curvatures, Osaka Journal of Mathematics 42 (2005), 435-452.
A. Belov-Kanel and M. Kontsevich, The Jacobian conjecture is stably equivalent to the Dixmier conjecture, Moscow Mathematical Journal 7 (2007), 209-218.
A. van den Essen, Polynomial Automorphisms and the Jacobian Conjecture, Progress in Mathematics 190, Birkhäuser, 2000.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Graduate Studies in Mathematics 30, revised edition, American Mathematical Society, 2001 - Ch. 1 and 15 for the Weyl algebra and its units.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Verify that $ad_∂$ lowers the Bernstein degree on $A_1$ by computing it on $x^a∂^b$ .
Show directly that the units of $A_n$ are exactly $K^ ×$ , using the associated graded ring of the Bernstein filtration.
Write the Fourier transform automorphism as a matrix in $Sp_2(K)$ and compute its order.
For $F = (x_1 + x_2^3, x_2)$ build the endomorphism of $A_2$ , verify the relations, and identify it as an exponential.
Explain why the argument of the main theorem breaks if $Δ F$ is a non-constant polynomial that never vanishes on $K^n$ .
Give an endomorphism of $A_1$ that is not obviously surjective and investigate whether $x$ lies in its image.
Compare the twisted module $M^σ$ for $σ$ the Fourier transform and $M = K[x_1, … ,x_n]$ with the untwisted module.