Locally Nilpotent Derivations

Q: Why does testing the generators suffice?

Because a : D^k(a) = 0 for some k is a subalgebra. Closure under products comes from (4.8): if D^p(a) = 0 and D^q(b) = 0 then every term of D^p+q-1(ab) has either D^ ≥ p(a) or D^ ≥ q(b) as a factor, so it vanishes.

Q: Is every locally nilpotent derivation of K[x_1, … ,x_n] conjugate to a triangular one?

For n ≤ 2 , yes - that is Rentschler's theorem. For n ≥ 3 , no: Bass gave a locally nilpotent derivation of K[x,y,z] that is not conjugate to a triangular one. The three-variable case is where the subject stops being classifiable, and it is also where the non-tame Nagata automorphism lives.

Q: If D is locally nilpotent, must D be a polynomial ring?

No. For n ≤ 2 it is a polynomial ring in one variable, and for n = 3 it is at least finitely generated. For n ≥ 5 it need not even be finitely generated: Daigle and Freudenburg gave an explicit counterexample. Assuming a nice kernel is one of the standard ways to produce an incorrect proof in this area.

Q: What is the relation between a slice and a coordinate?

A slice t is automatically a coordinate in the sense that S = ( D)[t] , so t is a variable of a polynomial ring structure over D . It need not be a variable of S = K[x_1, … ,x_n] in the sense of being part of a system of polynomial generators over K - that additionally requires D to be a polynomial ring over K .

Q: Does the theorem apply if the D_i are locally nilpotent but do not commute?

No, and the failure is immediate: the induction needs D_i to preserve D_1 , which is exactly what commutativity supplies. In the Jacobian application commutativity is a theorem, Coutinho (4.4.1), proved by extending the derivations to the power series ring and using the local inversion theorem .

Q: Why does the exponential of the Euler derivation not exist?

Because ε(x) = x gives ε^k(x) = x , so ∑ _k ε^k(x)z^k/k! = x e^z , which is not in K[x][z] . The flow exists, but it is an action of the multiplicative group and lives in a completion, not an action of G_a by polynomial maps.

Overview

Among all derivations of a polynomial ring, the locally nilpotent ones are the tractable minority. The condition is that for each $a$ there is some $k$ , allowed to depend on $a$ , with $D^k(a) = 0$ . The partial derivatives $∂_i$ satisfy it; the Euler derivation $∑ _i x_i ∂_i$ does not, because $x_1$ is an eigenvector with non-zero eigenvalue and survives every power.

In characteristic zero, local nilpotence is exactly the condition that lets one write $(D) = ∑ _k D^k/k!$ as a finite sum on each element. The result is a ring automorphism, and the correspondence between locally nilpotent derivations and algebraic actions of the additive group $G_a$ is a dictionary that the whole subject runs on.

The main theorem on this page is the slice theorem. If $D$ is locally nilpotent on a $ℚ$ -algebra $S$ and there is a $t ∈ S$ with $D(t) = 1$ , then $S = R[t]$ where $R = D$ , $t$ is transcendental over $R$ , and $D$ is honestly $d/dt$ . It generalises to a commuting family: $n$ commuting locally nilpotent derivations $D_1, … ,D_n$ with elements $t_1, … ,t_n$ satisfying $D_i(t_j) = δ_ij$ force $S = R[t_1, … ,t_n]$ with $D_i = ∂/∂ t_i$ .

That is the engine of Chapter 4. The Jacobian derivations $D_i = ∂/∂ F_i$ automatically satisfy $D_i(F_j) = δ_ij$ and commute; if they were also known to be locally nilpotent, the theorem would give $K[x_1, … ,x_n] = K[F_1, … ,F_n]$ , which is the Jacobian conjecture. Supplying that missing local nilpotence is exactly what the Dixmier conjecture would do, as shown on the automorphisms page.

Definition

Locally nilpotent derivationCoutinho, Ch. 4 §3

Let $S$ be a commutative $K$ -algebra and $D$ a $K$ -derivation of $S$ . Then $D$ is locally nilpotent if for every $a ∈ S$ there exists $k ∈ N$ with $D^k(a) = 0$ . The exponent may depend on $a$ ; if a single $k$ worked for all $a$ , $D$ would be nilpotent as an operator, which for a non-zero derivation of a polynomial ring is impossible.

Write $LND(S)$ for the set of locally nilpotent $K$ -derivations. It is not a Lie subalgebra and not closed under addition; it is closed under multiplication by elements of the kernel.

Slice, and the degree function

An element $t ∈ S$ with $D(t) = 1$ is a slice for $D$ . For $a ≠ 0$ the $D$ -degree is $_D(a) = k : D^k(a) ≠ 0$ , finite precisely by local nilpotence, with the convention $_D(0) = - ∈ fty$ .

Local nilpotence can be tested on generators

Let $S$ be generated as a $K$ -algebra by $a_1, … ,a_m$ . If for each $i$ there is a $k_i$ with $D^k_i(a_i) = 0$ , then $D$ is locally nilpotent.

This is because the set $a : D^k(a) = 0 for some k$ is a $K$ -subalgebra: it is visibly closed under sums, and closure under products follows from the higher Leibniz rule (4.8) below. It contains $K$ and the generators, so it is all of $S$ . This is what makes local nilpotence a finite check on a polynomial ring.

Core Concepts

Local nilpotence is a flow condition

Think of $D$ as a vector field. Its flow is $(zD)$ , which is a formal power series in the time parameter $z$ . For a general field this series is an infinite object and lives only in a completion; local nilpotence is exactly the statement that the flow of every polynomial is again a polynomial in the time variable. The orbit of a point is then an algebraic copy of the affine line, and the flow is an algebraic action of $G_a = (K,+)$ .

The Euler derivation is the standard contrast. Its flow is $x_i ↦ e^z x_i$ , an action of the multiplicative group $G_m$ , not of $G_a$ , and $e^z$ is not a polynomial in $z$ . That is precisely why it is not locally nilpotent.

A slice is a coordinate the derivation creates

If $D(t) = 1$ then the flow moves $t$ at unit speed: $(zD)(t) = t + z$ . So $t$ is a global time coordinate along the orbits, and $ker D$ is the ring of functions constant along orbits. Together they should reconstruct everything, and the slice theorem says they do: $S = ( D)[t]$ as a polynomial ring in one variable. Geometrically, the action is a translation and $Spec(S) ≅ Spec(R) × A^1$ with the action on the second factor.

Why a slice is a strong hypothesis

Not every locally nilpotent derivation has one. $D = y^2∂_x$ on $K[x,y]$ takes every value inside the ideal $(y^2)$ , so $D(t) = 1$ has no solution. Geometrically the action degenerates along the line $y = 0$ , where it is trivial, so there is no uniform time coordinate. The slice hypothesis rules out exactly this degeneration, and in Chapter 4 it is supplied for free by the relations $D_i(F_j) = δ_ij$ .

Construction and Proof

Everything below assumes $Q ⊂ eq S$ , so that the factorials in (4.9) are invertible.

The exponential map is a homomorphism

$φ$ of (4.9) is a $K$ -algebra homomorphism, and $φ ∘ D = (d/dz) ∘ φ$ .

Proof

Additivity and $K$ -linearity are clear. For products, expand using (4.8):

φ(ab) = ∑ _m ≥ 0 (z^m)/(m!) ∑ _j=0^m mjD^j(a)D^m-j(b) = ∑ _m ≥ 0 ∑ _j=0^m D^j(a)z^jj! ⋅ D^m-j(b)z^m-j(m-j)! = φ(a)φ(b),

the last step being the Cauchy product of the two finite sums. For the intertwining relation, differentiate (4.9) termwise: $(d)/(dz)φ(a) = ∑ _k ≥ 1 D^k(a)z^k-1/(k-1)! = ∑ _j ≥ 0D^j+1(a)z^j/j! = φ(D(a))$ . Setting $z = c$ gives a homomorphism $S \to S$ , and $(cD) (-cD) = (0 ⋅ D) = id$ by the same computation applied with two parameters, so each $(cD)$ is an automorphism.

Slice theoremCoutinho (4.3.2); Wright 1981

Let $S$ be a commutative $ℚ$ -algebra, $D$ a locally nilpotent derivation of $S$ , and suppose $t ∈ S$ satisfies $D(t) = 1$ . Put $R = D$ . Then

$S = R[t]$ ;
$t$ is transcendental over $R$ , so $S$ is a polynomial ring in one variable over $R$ ;
under this identification $D = d/dt$ .

Proof of the slice theorem

(1) $S = R[t]$ . Induct on $_D(a)$ . If $_D(a) ≤ 0$ then $D(a) = 0$ and $a ∈ R$ . Let $_D(a) = m ≥ 1$ , so $c := D^m(a)$ is non-zero and lies in $R$ . Since $D(t) = 1$ we get $D^m(t^m) = m!$ , and because $c ∈ R$ ,

D^m!(a - (c)/(m!)t^m) = c - (c)/(m!) ⋅ m! = 0.

So $a - (c/m!)t^m$ has strictly smaller $D$ -degree and, by induction, lies in $R[t]$ . Hence $a$ does too.

(2) Transcendence. Suppose $r_0 + r_1 t + … + r_m t^m = 0$ with $r_j ∈ R$ and $r_m ≠ 0$ , with $m$ minimal among all such relations. Apply $D^m$ . Every term with $j < m$ dies, and $D^m(r_m t^m) = r_m, m!$ . So $r_m m! = 0$ , and since $Q ⊂ eq S$ this forces $r_m = 0$ , a contradiction.

(3) $D = d/dt$ . On $S = R[t]$ , $D( ∑ _j r_j t^j) = ∑ _j r_j, j, t^j-1$ by the Leibniz rule and $D(r_j) = 0$ , $D(t) = 1$ . That is $d/dt$ .

The projection (4.12) gives an alternative proof of (1): $π_t$ lands in $R$ because $Dπ_t(a) = 0$ by a telescoping computation, and inverting the triangular substitution $z ↦ -t$ expresses $a$ as a polynomial in $t$ with coefficients in $R$ .

Several commuting derivations with a dual system of slicesCoutinho (4.3.1); Wright 1981

Let $S$ be a commutative $ℚ$ -algebra and let $D_1, … ,D_n$ be pairwise commuting locally nilpotent derivations of $S$ . Suppose there exist $t_1, … ,t_n ∈ S$ with $D_i(t_j) = δ_ij$ for all $i, j$ . Let $R = ⋂ _i D_i$ . Then $S = R[t_1, … ,t_n]$ , the $t_i$ are algebraically independent over $R$ , and $D_i = ∂/∂ t_i$ .

Proof by induction on $n$

For $n = 1$ this is the slice theorem. Let $n > 1$ and put $R_1 = D_1$ . The slice theorem applied to $D_1$ and $t_1$ gives $S = R_1[t_1]$ , with $t_1$ transcendental over $R_1$ and $D_1 = ∂/∂ t_1$ .

For $i ≥ 2$ , $D_i$ commutes with $D_1$ , so $D_1 D_i(a) = D_i D_1(a) = 0$ whenever $a ∈ R_1$ ; that is, $D_i(R_1) ⊂ eq R_1$ . Also $D_1(t_j) = δ_1j = 0$ for $j ≥ 2$ , so $t_2, … ,t_n ∈ R_1$ . Thus $D_2, … ,D_n$ restrict to commuting locally nilpotent derivations of $R_1$ with $D_i(t_j) = δ_ij$ for $i,j ≥ 2$ , and their common kernel inside $R_1$ is $R$ .

By induction $R_1 = R[t_2, … ,t_n]$ with the $t_j$ algebraically independent over $R$ and $D_i = ∂/∂ t_i$ there. Combining, $S = R_1[t_1] = R[t_1, … ,t_n]$ . Finally $D_i(t_1) = 0$ for $i ≥ 2$ , so each $D_i$ acts on $S = R_1[t_1]$ coefficientwise and is $∂/∂ t_i$ on the whole of $S$ .

What the theorem does not say

It concludes $S = R[t_1, … ,t_n]$ , not $S = K[t_1, … ,t_n]$ . To get the latter one must separately show $R = K$ . In the application to the Jacobian conjecture that step is genuine but short: $S = K[x_1, … ,x_n]$ has transcendence degree $n$ over $K$ , and $t_1, … ,t_n$ are already algebraically independent over $R ⊃ eq K$ , so $R$ is algebraic over $K$ ; but any element of $K[x_1, … ,x_n]$ of positive degree is transcendental over $K$ , since its powers have unbounded degree. Hence $R = K$ .

Key Equations

The defining condition, and the two constructions built on it:

∀ a ∈ S ∃ k ∈ N : D^k(a) = 0.

(4.7)

The higher Leibniz rule is what makes everything work; it is proved by induction on $m$ exactly as the binomial theorem:

D^m(ab) = ∑ _j=0^m mj, D^j(a), D^m-j(b).

(4.8)

The exponential homomorphism attaches to $D$ a map into the polynomial ring in one new variable $z$ :

φ : S ⟶ S[z], φ(a) = ∑ _k ≥ 0 D^k(a)k!, z^k,

(4.9)

a finite sum for each $a$ by (4.7). It is a $K$ -algebra homomorphism, precisely because of (4.8).

It intertwines $D$ with differentiation in the new variable, and specialising $z$ recovers the flow:

φ ∘ D = (d)/(dz) ∘ φ, (cD) := .φ|_z = c ∈ Aut_K(S) (c ∈ K),

(4.10)

with $(cD)^-1 = (-cD)$ , so $c ↦ (cD)$ is an action of $(K,+)$ by automorphisms.

When $S$ is a domain, the degree function is additive, which is the source of most structural results:

_D(ab) = _D(a) + _D(b), _D(a+b) ≤ _D(a), _D(b).

(4.11)

Finally, when a slice $t$ exists, substituting $z = -t$ into (4.9) gives a projection onto the constants:

π_t : S ⟶ S, π_t(a) = ∑ _k ≥ 0 (-1)^k, D^k(a)k!, t^k,

(4.12)

a $K$ -algebra homomorphism with image $R = D$ and $π_t|_R = id$ .

Variable Definitions

$K$: the ground field, of characteristic zero
$S$: a commutative $K$ -algebra, usually $K[x_1, … ,x_n]$
$D$ , $D_i$: locally nilpotent $K$ -derivations of $S$
$R$: the ring of constants; for a family, $R = ⋂ _i D_i$
$t$ , $t_i$: slices: elements with $D(t) = 1$ , respectively $D_i(t_j) = δ_ij$
$_D(a)$: the largest $k$ with $D^k(a) ≠ 0$
$φ$: the exponential homomorphism $S → S[z]$ of (4.9)
$(cD)$: the automorphism of $S$ obtained by setting $z = c$ in $φ$
$π_t$: the projection $S \to R$ of (4.12), defined when a slice exists
$G_a$: the additive group scheme, whose algebraic actions correspond to locally nilpotent derivations

Properties and Behaviour

Additivity of the degree function

Let $S$ be a domain containing $ℚ$ and $D$ locally nilpotent. If $a,b ≠ 0$ with $_D(a) = p$ and $_D(b) = q$ , then by (4.8) the term $p+qpD^p(a)D^q(b)$ is the only surviving one in $D^p+q(ab)$ , and it is non-zero because $S$ is a domain and the binomial coefficient is invertible. So $_D(ab) = p + q$ .

The kernel is factorially closed

With $S$ a domain over $ℚ$ : if $a,b ≠ 0$ and $ab ∈ D$ , then $_D(a) + _D(b) = 0$ , so both are zero and $a,b ∈ D$ . Consequently $ker D$ contains every unit of $S$ , is algebraically closed in $S$ , and if $S$ is a UFD then $ker D$ contains all irreducible factors of any of its elements.

Kernels are large but not always tame

For a non-zero locally nilpotent derivation of $K[x_1, … ,x_n]$ the kernel has transcendence degree exactly $n-1$ over $K$ . It is a finitely generated $K$ -algebra when $n ≤ 3$ . Daigle and Freudenburg constructed a locally nilpotent derivation of $K[x_1, … ,x_5]$ whose kernel is not finitely generated, giving a counterexample to Hilbert's fourteenth problem in dimension five; the case $n = 4$ has no known counterexample.

Low-dimensional classification

Every locally nilpotent derivation of $K[x]$ is $c,∂_x$ with $c ∈ K$ . Rentschler's theorem says that every locally nilpotent derivation of $K[x,y]$ is, after conjugation by a polynomial automorphism of the plane, of the form $f(x),∂_y$ . No such normal form is known for $n ≥ 3$ : Bass constructed in 1984 a locally nilpotent derivation of $K[x_1,x_2,x_3]$ that is not conjugate to a triangular one.

Closure properties, and their failure

If $D$ is locally nilpotent and $r ∈ D$ then $r D$ is locally nilpotent.
If $D$ is locally nilpotent and $α$ is an automorphism then $α D α^-1$ is locally nilpotent.
The sum of two locally nilpotent derivations need not be locally nilpotent: $y∂_x$ and $x∂_y$ are each locally nilpotent, but their sum sends $x \mapsto y \mapsto x$ and so survives every power.
The bracket of two locally nilpotent derivations need not be locally nilpotent: $[y∂_x, x∂_y] = y∂_y - x∂_x$ , which is a semisimple derivation.
A locally nilpotent derivation of $S$ need not stay locally nilpotent on a localisation of $S$ , unless one inverts only elements of $ker D$ .

Examples and Special Cases

Triangular derivations

If $D(x_1) ∈ K$ and $D(x_i) ∈ K[x_1, … ,x_i-1]$ for every $i$ , then $D$ is locally nilpotent: each $D(x_i)$ has strictly fewer variables, so iterating drives everything to zero. These are the triangular derivations, and their exponentials are the triangular (de Jonquières) automorphisms.

The Weyl algebra analogue

On the non-commutative ring $A_n$ , the inner derivation $ad_∂_i(b) = [∂_i, b]$ lowers total degree by one, so it is locally nilpotent. This is the observation that drives the proof on the automorphisms page; note that $ad_x_i$ is locally nilpotent for the same reason, while $ad_x_i∂_i$ is not.

The Nagata automorphism

On $K[x,y,z]$ put $D = z,∂_y - 2y,∂_x$ , which is locally nilpotent (it is triangular in the order $z, y, x$ ), and set $δ = xz + y^2$ . Then $D(δ) = -2yz + 2yz = 0$ , so $δ ∈ D$ and $δ D$ is again locally nilpotent. Computing iterates: $(δ D)(z) = 0$ , $(δ D)(y) = δ z$ , $(δ D)(x) = -2yδ$ , $(δ D)^2(x) = -2δ^2 z$ , and everything higher vanishes. Hence

(δ D) : x ↦ x - 2yδ - zδ^2, y ↦ y + zδ, z ↦ z.

This is Nagata's automorphism of $K[x,y,z]$ . Shestakov and Umirbaev proved in 2004 that it is not tame - it is not a composite of affine and triangular automorphisms - which shows how much wilder the three-variable theory is than Rentschler's picture in two variables.

A derivation with a large kernel

$D = ∂_x + y,∂_z$ on $K[x,y,z]$ is triangular after reordering, hence locally nilpotent, and $x$ is a slice. The slice theorem gives $K[x,y,z] = D,[x]$ with $D = K[y,, z - xy]$ : indeed $D(z - xy) = y - y = 0$ , and counting transcendence degrees shows nothing else is needed.

Not locally nilpotent, though every generator looks harmless

$D = y∂_x + x∂_y$ on $K[x,y]$ satisfies $D(x) = y$ , $D(y) = x$ , so $D^2(x) = x$ and no power kills $x$ . Local nilpotence is a statement about iterates, not about the size of $D(x_i)$ : small values on the generators guarantee nothing if the values feed back into each other.

Worked Example

The two Jacobian derivations of the shear $F = (x + y^2,, y)$

Step 1 - write down the derivations
Let $S = K[x,y]$ and $F_1 = x + y^2$ , $F_2 = y$ , so that $J(F) = pmatrix 1 & 2y 0 & 1 pmatrix$ and $Δ = J(F) = 1$ . Cramer's rule gives the derivations $∂/∂ F_i$ :

$D_1 = ∂_x, D_2 = ∂_y - 2y,∂_x.$

Check the duality relations: $D_1(F_1) = 1$ , $D_1(F_2) = 0$ , $D_2(F_1) = 2y - 2y ⋅ 1 = 0$ , $D_2(F_2) = 1$ . So $D_i(F_j) = δ_ij$ as required.
Step 2 - check commutativity and local nilpotence
Bracket: $[D_1,D_2](x) = D_1(-2y) - D_2(1) = 0$ and $[D_1,D_2](y) = D_1(1) - D_2(0) = 0$ , so $[D_1,D_2] = 0$ on the generators, hence identically.

Local nilpotence needs only the generators, by the lemma above. For $D_1 = ∂_x$ : $D_1(x) = 1$ , $D_1^2(x) = 0$ , $D_1(y) = 0$ . For $D_2$ : $D_2(y) = 1$ and $D_2^2(y) = 0$ ; $D_2(x) = -2y$ , $D_2^2(x) = -2$ , $D_2^3(x) = 0$ . Both are locally nilpotent.
Step 3 - apply the structure theorem
The hypotheses of the theorem hold with $t_i = F_i$ , so $S = R[F_1,F_2]$ with $R = D_1 ∩ D_2$ . Compute $R$ : $∂_x = K[y]$ , and for $g(y) ∈ K[y]$ we have $D_2(g) = g'(y)$ , which vanishes only for $g ∈ K$ . So $R = K$ and

$K[x,y] = K[F_1, F_2] = K[x + y^2,, y].$

Verify independently: $x = F_1 - F_2^2$ and $y = F_2$ , so the two generate. The theorem has recovered surjectivity of the comorphism - which is the conclusion the Jacobian conjecture asks for - from local nilpotence alone.
Step 4 - exponentiate
By (4.10), $(D_2)(y) = y + 1$ and $(D_2)(x) = x + D_2(x) + 12D_2^2(x) = x - 2y - 1$ . So

$(D_2) : x ↦ x - 2y - 1, y ↦ y + 1.$

As a check, $D_2(F_1) = 0$ means $(D_2)$ must fix $F_1$ . Indeed $(x - 2y - 1) + (y+1)^2 = x - 2y - 1 + y^2 + 2y + 1 = x + y^2 = F_1$ . Similarly $(D_1)$ is the translation $x ↦ x+1$ , which fixes $F_2$ and sends $F_1$ to $F_1 + 1$ .
Step 5 - two derivations that fail
$E = y^2∂_x$ is locally nilpotent with $E = K[y]$ , but $E(g) = y^2 g_x$ always lies in the ideal $(y^2)$ , so $E$ has no slice and the structure theorem does not apply. And $∂_x$ , locally nilpotent on $K[x]$ , is not locally nilpotent on the localisation $K[x, x^-1]$ , since $∂_x^k(x^-1) = (-1)^k k!, x^-k-1 ≠ 0$ for all $k$ . Local nilpotence is destroyed by inverting elements outside the kernel.

Result

$D_1 = ∂_x$ and $D_2 = ∂_y - 2y∂_x$ are commuting locally nilpotent derivations of $K[x,y]$ with $D_i(F_j) = δ_ij$ for $F = (x+y^2, y)$ . The structure theorem gives $K[x,y] = K[F_1,F_2]$ with $D_i = ∂/∂ F_i$ , confirming that $F$ is a polynomial isomorphism. The corresponding exponentials are the affine automorphisms $x ↦ x+1$ and $(x,y) ↦ (x - 2y - 1,, y+1)$ .

Applications and Industry Use

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

The Jacobian conjecture. The reduction on the automorphisms page uses the theorem of this page as its final step, once local nilpotence has been extracted from a hypothetical automorphism of $A_n$ .
Polynomial automorphism groups. Exponentials of locally nilpotent derivations generate the tame subgroup of $Aut(K^n)$ together with the affine maps; Rentschler's classification in the plane yields the Jung-van der Kulk theorem that all plane automorphisms are tame.
Invariant theory and Hilbert's fourteenth problem. Rings of constants are rings of $G_a$ -invariants, and the known counterexamples to finite generation are constructed this way.
The cancellation problem. Makar-Limanov's invariant, the intersection of the kernels of all locally nilpotent derivations of a ring, distinguishes rings that are otherwise hard to tell apart, and settles instances of the cancellation problem.
Weyl algebra endomorphisms. Local nilpotence of $ad_a$ is the standard mechanism for producing automorphisms of $A_n$ : $(ad_a)$ is an automorphism whenever $ad_a$ is locally nilpotent, and building the endomorphism at all uses the presentation of P4 by generators and relations.
Open problems. Whether the kernel of a locally nilpotent derivation of $K[x_1, … ,x_4]$ is always finitely generated, and whether the Jacobian derivations are locally nilpotent, both sit on the list of open problems around this circle of ideas.

Design Considerations

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

Where to verify local nilpotence

Always on generators, never on a general element - the subalgebra lemma makes the check finite. But choose the generating set with care: for a triangular derivation the natural variables make the check immediate, while an arbitrary linear change of coordinates can hide triangularity completely.

Which ring to carry the derivation on

Localisations are dangerous, as the pitfall above records. If a construction naturally produces a derivation of $K[X,Δ^-1]$ , invest the effort to descend it to $K[X]$ before invoking any structure theory. Passing to $K[[X]]$ is safe for verifying identities but destroys local nilpotence in the same way localisation does, since the power series ring has units of positive order.

Slice or no slice

If a slice is available, use it: the theory becomes trivial, since $S$ is a polynomial ring over the kernel and $D$ is a partial derivative. If not, the invariants to reach for are the degree function, the plinth ideal $D( D^2) ∩ D$ , and the Makar-Limanov invariant, none of which need a slice.

Material Selection

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

Ground field. Must contain $ℚ$ for the exponential to exist. $ℚ$ itself is the right choice for computation; $ℂ$ for the geometric interpretation as a $G_a$ -action.
Base ring. The slice theorem is stated for an arbitrary commutative $ℚ$ -algebra, not just a polynomial ring, and that generality is used in the induction: the ring $R_1 = D_1$ appearing at the second stage is not a polynomial ring a priori.
Domain or not. The degree function and factorial closedness need a domain. The slice theorem itself does not.
Representation. Store $D$ as the tuple $(D(x_1), … ,D(x_n))$ and cache the iterates $D^k(x_i)$ ; almost every algorithm on this page consumes that table.

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.

Testing local nilpotence. Compute $D^k(x_i)$ for $k = 1,2, …$ and each $i$ , stopping when all vanish. Correct by the subalgebra lemma, but only a semi-decision procedure: nothing bounds how far to iterate. For triangular or homogeneous derivations sharp bounds are available.
Computing $(D)$ . Once local nilpotence is known, $(D)$ is a finite sum on each generator; the cost is that of computing $D^k(x_i)$ up to the nilpotency index, and the degrees grow quickly.
Computing $ker D$ with a slice. Immediate: apply $π_t$ of (4.12) to each generator and take the resulting algebra.
Computing $ker D$ without a slice. Use van den Essen's algorithm, which computes the kernel as an intersection with a localisation and terminates precisely when the kernel is finitely generated. Implemented in Singular; a Gröbner basis computation sits at its core.
Deciding whether a slice exists. Equivalent to solving $D(t) = 1$ , a linear system over $K$ once degrees are bounded, but the bound on $deg t$ is the hard part.

In practice these computations are done in a general commutative algebra system - Singular or Macaulay2 - since every step reduces to Gröbner bases, subalgebra membership and elimination. Packages for invariant rings of algebraic group actions are the natural home for kernel computations. All of it becomes impractical well before the dimensions where the interesting open questions live.

Limits of Validity

Characteristic zero is essential. In characteristic $p$ the factorials in (4.9) are not invertible and the exponential does not exist. Worse, the notion degenerates: $∂_x^p = 0$ on $K[x]$ , so $∂_x$ is nilpotent, and every derivation whose $p$ -th power vanishes is trivially locally nilpotent. The correspondence with $G_a$ -actions is replaced by the theory of iterative higher derivations; the same phenomenon is what breaks the Weyl algebra in characteristic P6.
The slice hypothesis cannot be dropped. Without $D(t) = 1$ the conclusion $S = R[t]$ is false: $y^2∂_x$ on $K[x,y]$ has $= K[y]$ and $K[x,y] = K[y][x]$ happens to hold, but $D ≠ d/dx$ ; and in general $S$ need not even be a polynomial ring over $ker D$ .
Commutativity of the family cannot be dropped. If the $D_i$ do not commute, $D_i(R_1) ⊂ eq R_1$ fails and the induction collapses.
The theorem gives $R$ , not $K$ . As noted in the remark above, identifying $R$ with $K$ is a separate argument that uses properties of the specific ring $S$ .
Domain hypothesis. Additivity of $_D$ and factorial closedness of the kernel both need $S$ to be a domain; on a ring with zero divisors they fail.

Failure Modes and Common Mistakes

Confusing nilpotent with locally nilpotent

A non-zero derivation of $K[x_1, … ,x_n]$ in characteristic zero is never nilpotent as an operator: $∂_x^k(x^k) = k! ≠ 0$ for every $k$ . The word "locally" is load-bearing - the exponent depends on the element. Any argument that fixes a single $k$ in advance is wrong.

Assuming local nilpotence survives localisation

This is the trap that dictates the shape of Chapter 4. $∂_x$ is locally nilpotent on $K[x]$ but not on $K[x,x^-1]$ . Inverting an element outside $ker D$ destroys the property. It is exactly why the Jacobian derivations must be shown to live on $K[X]$ - which needs $Δ$ to be a non-zero constant - rather than merely on $K[X,Δ^-1]$ .

Thinking the set of locally nilpotent derivations is a Lie algebra

It is not closed under addition or bracket. $y∂_x$ and $x∂_y$ are locally nilpotent; their sum is not, and their bracket $y∂_y - x∂_x$ is semisimple. $LND(S)$ is a cone, closed under multiplication by kernel elements and under conjugation, and nothing more.

Expecting the kernel to be a polynomial ring

For $n ≤ 3$ the kernel of a locally nilpotent derivation of $K[x_1, … ,x_n]$ is finitely generated, and for $n ≤ 2$ it is a polynomial ring, but in general it need not be finitely generated at all. Writing down a presumed set of generators for $ker D$ and proceeding is a common source of false proofs in this area.

Reading the slice theorem as a proof of the Jacobian conjecture

The theorem is a genuine theorem, and its hypotheses are genuinely unverified in the Jacobian setting. The derivations $∂/∂ F_i$ commute and satisfy $D_i(F_j) = δ_ij$ unconditionally when $Δ = 1$ ; what is unknown is whether they are locally nilpotent. Everything difficult about the conjecture sits in that one word.

Historical Notes

The correspondence between locally nilpotent derivations and $G_a$ -actions is classical folklore, made precise in the algebraic-group literature of the 1950s and 1960s. Rentschler's 1968 classification in two variables was the first serious structural theorem, and it immediately reproved the Jung-van der Kulk description of $Aut(K^2)$ .

The version used in the primer - a commuting family with a dual system of slices - is due to D. Wright, in his 1981 paper on the Jacobian conjecture, and it was Vaserstein and Katz who saw that it converts Dixmier's problem about endomorphisms of $A_n$ into Keller's conjecture about polynomial maps. Bass, Connell and Wright's 1982 survey is where that observation was recorded and where Coutinho's Chapter 4 takes it from.

Since then the subject has become a field of its own. Nagata's counterexample to Hilbert's fourteenth problem was reinterpreted as a non-finitely generated kernel; Makar-Limanov introduced his invariant in the 1990s and used it on the cancellation problem; Daigle and Freudenburg produced a five-variable counterexample in 1999; and Shestakov and Umirbaev proved in 2004 that the Nagata automorphism of $K[x,y,z]$ is not tame, closing a question that had been open since the 1970s.

Comparison

Locally nilpotent derivations against the two nearest neighbours.
	Locally nilpotent	Semisimple (diagonalisable)	General derivation
Model	$∂_x$	$x∂_x$	arbitrary $∑ f_i∂_i$
Group it generates	$G_a$ action	$G_m$ action	no algebraic action in general
Exponential	polynomial automorphism	only formal, or a torus action	not defined
Kernel	factorially closed, transcendence degree $n-1$	spanned by weight-zero monomials	no general description
Slice	may or may not exist	never: $D$ scales each weight- $w$ monomial by $w$ , so $1$ is not in the image	may or may not exist
Decidable from generators	yes, but with no a priori bound	yes, by linear algebra on the degree-zero part	not applicable

Key Takeaways

Key points

$D$ is locally nilpotent if every element is killed by some power of $D$ ; the exponent depends on the element, and it suffices to test the algebra generators.
In characteristic zero, $φ(a) = ∑ _k D^k(a)z^k/k!$ is a ring homomorphism $S → S[z]$ , and specialising $z$ gives an action of $(K,+)$ by automorphisms.
Slice theorem: if $D(t) = 1$ then $S = ( D)[t]$ with $t$ transcendental and $D = d/dt$ .
For $n$ commuting locally nilpotent derivations with $D_i(t_j) = δ_ij$ , the ring is $R[t_1, … ,t_n]$ and $D_i = ∂/∂ t_i$ ; identifying $R$ with $K$ is a separate step.
In a domain over $ℚ$ the degree function is additive, so the kernel is factorially closed; kernels have transcendence degree $n-1$ but need not be finitely generated for $n ≥ 5$ .
$LND(S)$ is not closed under sums or brackets, and local nilpotence is destroyed by inverting elements outside the kernel.
In the Jacobian setting, everything except local nilpotence of $∂/∂ F_i$ is known; that single missing property is the conjecture.

FAQs

Why does testing the generators suffice?

Because $a : D^k(a) = 0$ for some $k}$ is a subalgebra. Closure under products comes from (4.8): if $D^p(a) = 0$ and $D^q(b) = 0$ then every term of $D^p+q-1(ab)$ has either $D^ ≥ p(a)$ or $D^ ≥ q(b)$ as a factor, so it vanishes.

Is every locally nilpotent derivation of

K[x_1, … ,x_n]

conjugate to a triangular one?

For $n ≤ 2$ , yes - that is Rentschler's theorem. For $n ≥ 3$ , no: Bass gave a locally nilpotent derivation of $K[x,y,z]$ that is not conjugate to a triangular one. The three-variable case is where the subject stops being classifiable, and it is also where the non-tame Nagata automorphism lives.

D

is locally nilpotent, must

ker D

be a polynomial ring?

No. For $n ≤ 2$ it is a polynomial ring in one variable, and for $n = 3$ it is at least finitely generated. For $n ≥ 5$ it need not even be finitely generated: Daigle and Freudenburg gave an explicit counterexample. Assuming a nice kernel is one of the standard ways to produce an incorrect proof in this area.

What is the relation between a slice and a coordinate?

A slice $t$ is automatically a coordinate in the sense that $S = ( D)[t]$ , so $t$ is a variable of a polynomial ring structure over $ker D$ . It need not be a variable of $S = K[x_1, … ,x_n]$ in the sense of being part of a system of polynomial generators over $K$ - that additionally requires $ker D$ to be a polynomial ring over $K$ .

Does the theorem apply if the

D_i

are locally nilpotent but do not commute?

No, and the failure is immediate: the induction needs $D_i$ to preserve $D_1$ , which is exactly what commutativity supplies. In the Jacobian application commutativity is a theorem, Coutinho (4.4.1), proved by extending the derivations to the power series ring and using the local inversion theorem.

Why does the exponential of the Euler derivation not exist?

Because $ε(x) = x$ gives $ε^k(x) = x$ , so $∑ _k ε^k(x)z^k/k! = x e^z$ , which is not in $K[x][z]$ . The flow exists, but it is an action of the multiplicative group and lives in a completion, not an action of $G_a$ by polynomial maps.

How does this transfer to the Weyl algebra, which is not commutative?

The definition of local nilpotence makes sense for any derivation of any ring, and $ad_a : b ↦ [a,b]$ is a derivation of the Weyl algebra $A_n$ . When $ad_a$ is locally nilpotent, $(ad_a)$ is an algebra automorphism of $A_n$ . The bridge back to the commutative statement is the identity $[D, g] = D(g)$ for a derivation $D$ and a polynomial $g$ inside $A_n$ .

Is there an algorithm that decides local nilpotence?

The naive iteration is only a semi-decision procedure: it confirms local nilpotence when it holds but never terminates with a negative answer. For homogeneous or triangular derivations there are degree bounds that make it decidable. In general one usually proves non-nilpotence by exhibiting an eigenvector, as with the Euler derivation.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 4 §3, Proposition (4.3.1) and Lemma (4.3.2).
D. Wright, On the Jacobian conjecture, Illinois Journal of Mathematics 25 (1981), 423-440.
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.
R. Rentschler, Opérations du groupe additif sur le plan affine, Comptes Rendus de l'Académie des Sciences Paris 267 (1968), 384-387.
G. Freudenburg, Algebraic Theory of Locally Nilpotent Derivations, Encyclopaedia of Mathematical Sciences 136, Springer, 2nd edition, 2017 - the definitive reference; Ch. 1-4 cover everything on this page.
A. van den Essen, Polynomial Automorphisms and the Jacobian Conjecture, Progress in Mathematics 190, Birkhäuser, 2000 - Ch. 1-2, including the kernel algorithm.
D. Daigle and G. Freudenburg, A counterexample to Hilbert's fourteenth problem in dimension 5, Journal of Algebra 221 (1999), 528-535.
I. Shestakov and U. Umirbaev, The tame and the wild automorphisms of polynomial rings in three variables, Journal of the American Mathematical Society 17 (2004), 197-227.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Prove the higher Leibniz rule (4.8) by induction and use it to show that the locally nilpotent elements form a subalgebra.
Show that $(D)$ and $(-D)$ are mutually inverse, directly from the definition (4.9).
For $D = ∂_x + y∂_z$ on $K[x,y,z]$ , compute $π_x$ on each variable and confirm that its image is $K[y, z - xy]$ .
Find all locally nilpotent derivations of $K[x,y]$ that are homogeneous of degree zero for the standard grading.
Give an example of a locally nilpotent derivation whose kernel is not a polynomial ring over $K$ .
Explain why $ad_x∂$ is not locally nilpotent on $A_1$ , and identify its eigenvectors.
Show that if $D$ is locally nilpotent and $D(t) = 1$ then $(cD)(t) = t + c$ , and deduce that the action is free.