Derivations of the Polynomial Ring

Q: Why is Der_K(K[X]) a module over K[X] but the bracket only K-bilinear?

Because [D, gE] = D(g)E + g[D,E] : scaling one argument by a ring element produces an extra term D(g)E . That failure is not a defect; it is the reason Der_K is a Lie-Rinehart algebra rather than a Lie algebra over K[X] .

Q: Can a derivation of K[X] be nilpotent as an operator, rather than only locally nilpotent?

Only if it is zero. If D ≠ 0 then D(x_i) ≠ 0 for some i , and by the graded structure one can produce elements on which arbitrarily high powers of D do not vanish. Local nilpotence - every element killed by some power, with the power depending on the element - is the strongest condition available.

Q: What is the relationship between Der_K(K[X]) and A_n?

A_n is generated as a K -algebra by K[X] and Der_K(K[X]) ; in fact A_n is the enveloping algebra of the Lie-Rinehart pair. Concretely, the elements of A_n of order at most one are exactly K[X] ⊕ Der_K(K[X]) , as described on the page about the Weyl algebra as differential operators .

Overview

Derivations of a commutative ring are defined by one axiom, the Leibniz rule. For the polynomial ring $K[X] = K[x_1, … ,x_n]$ that single axiom has a very strong consequence: a $K$ -derivation cannot be exotic. Choose the $n$ polynomials $D(x_1), … ,D(x_n)$ freely, and the derivation is fixed; choose them any other way and you get a different derivation. There is nothing else to know.

So $Der_K(K[X])$ is a free $K[X]$ -module of rank $n$ with basis $∂_1, … ,∂_n$ . Read geometrically, a $K$ -derivation is a polynomial vector field on affine $n$ -space, and the module identification is the statement that a vector field is a tuple of component functions. Read algebraically, it says that the degree-one part of the order filtration of the Weyl algebra is exactly $K[X] ⊕ Der_K(K[X])$ .

What is not determined by the definition is the behaviour of the kernel. The ring of constants $ker D$ is always a subalgebra, but its size varies wildly: it is $K[x_2, … ,x_n]$ for $∂_1$ , it is $K$ for the Euler derivation, and for a generic vector field it is $K$ as well. Deciding what $ker D$ is, for a given $D$ , is a hard problem with a long history.

This page is the toolkit section for Chapter 4. The derivations that matter there are built from a polynomial map $F$ with non-vanishing Jacobian determinant: they are the operators $∂/∂ F_i$ , defined by Cramer's rule, and the whole reduction of the Jacobian conjecture to the Dixmier conjecture turns on whether those derivations are locally nilpotent.

Definition

Throughout, $K$ is a field of characteristic zero, $X = K^n$ , and $K[X] = K[x_1, … ,x_n]$ .

Derivation and K-derivation

Let $S$ be a commutative $K$ -algebra. A map $D : S \to S$ is a derivation if it is additive and satisfies the Leibniz rule $D(ab) = D(a)b + aD(b)$ for all $a,b ∈ S$ . It is a $K$ -derivation if in addition it is $K$ -linear. The set of $K$ -derivations of $S$ is written $Der_K(S)$ .

Taking $a = b = 1$ in the Leibniz rule gives $D(1) = 2D(1)$ , so $D(1) = 0$ ; with $K$ -linearity this forces $D(λ) = 0$ for every $λ ∈ K$ .

Ring of constants

The ring of constants of a derivation $D$ of $S$ is $S^D = D = a ∈ S : D(a) = 0$ . The Leibniz rule makes it closed under multiplication and additivity makes it closed under addition, so $S^D$ is a subring of $S$ ; for a $K$ -derivation it is a $K$ -subalgebra containing $K$ .

Structure of the derivations of a polynomial ringCoutinho, Ch. 3 §1; standard

$Der_K(K[X])$ is a free $K[X]$ -module of rank $n$ with basis $∂_1, … ,∂_n$ . Explicitly, every $K$ -derivation $D$ satisfies

D = ∑ _i=1^n D(x_i),∂_i,

and conversely, for any choice of $f_1, … ,f_n ∈ K[X]$ the operator $∑ _i f_i ∂_i$ is a $K$ -derivation with $D(x_i) = f_i$ .

Core Concepts

Why generators decide everything

The reason a derivation is pinned down by its values on the variables is that the elements it kills form a subalgebra. If $D(a) = D(b) = 0$ then $D(a+b) = 0$ and $D(ab) = D(a)b + aD(b) = 0$ , and $D$ kills $K$ automatically. So $ker D$ is a $K$ -subalgebra; if it contains the generators $x_1, … ,x_n$ it contains everything they generate, which is all of $K[X]$ . Apply this to a difference of two derivations agreeing on the variables and the difference is zero.

The same argument shows why the definition of a derivation is so much more rigid than that of a $K$ -linear map. A $K$ -linear endomorphism of $K[X]$ has infinitely many degrees of freedom; a derivation has exactly $n$ polynomials' worth.

Derivations as vector fields

Write $D = ∑ _i f_i ∂_i$ . The tuple $(f_1, … ,f_n)$ is a polynomial vector field on $K^n$ , and $D(g)$ is the directional derivative of $g$ along it. Over $ℝ$ or $ℂ$ the associated flow is the solution of the system $x_i = f_i(x)$ , and $ker D$ is the algebra of polynomial first integrals - quantities conserved along every trajectory. That is the correct mental picture for the ring of constants: constants are conserved quantities, and a system can have many, one, or none beyond the trivial scalars.

Derivations inside the Weyl algebra

An element of $A_n$ of order at most one has the form $g + ∑ _i f_i ∂_i$ ; the multiplication operator $g$ is the zero-order part and the rest is a derivation of $K[X]$ . Inside $A_n$ the action of the derivation on a polynomial appears as a commutator:

[,D,, g,] = D(g) for D ∈ Der_K(K[X]), g ∈ K[X],

(4.1)

where the right-hand side means the operator "multiply by the polynomial $D(g)$ ". This identity is the bridge used in Chapter 4: local nilpotence of $D$ as a derivation of $K[X]$ is the same thing as local nilpotence of $ad_D$ on the commutative subalgebra $K[X] ⊂ A_n$ .

Construction and Proof

Proof of the structure theorem

Let $D$ be a $K$ -derivation of $K[X]$ and put $f_i = D(x_i)$ . Set $E = ∑ _i f_i ∂_i$ , which is a $K$ -derivation because each $∂_i$ is one and $Der_K(K[X])$ is closed under multiplication by ring elements. Then $D - E$ is a $K$ -derivation and $(D-E)(x_j) = f_j - f_j = 0$ for every $j$ .

As observed above, $(D-E)$ is a $K$ -subalgebra of $K[X]$ . It contains $K$ and it contains $x_1, … ,x_n$ , hence it contains the subalgebra they generate, which is $K[X]$ . So $D = E$ .

For freeness, suppose $∑ _i f_i ∂_i = 0$ . Evaluating at $x_j$ gives $f_j = 0$ for each $j$ . So $∂_1, … ,∂_n$ are $K[X]$ -linearly independent, and together with the previous paragraph they form a basis.

The commutator of derivationsCoutinho, Ch. 4, Ex. 5.8

If $D, E ∈ Der_K(S)$ then $[D,E] = DE - ED ∈ Der_K(S)$ .

Proof

$K$ -linearity is clear. Expanding twice,

DE(ab) = D (E(a)b + aE(b) ) = DE(a),b + E(a)D(b) + D(a)E(b) + a,DE(b).

Interchanging $D$ and $E$ produces the same two cross terms $E(a)D(b) + D(a)E(b)$ , so they cancel in the difference and $[D,E](ab) = [D,E](a),b + a,[D,E](b)$ . Note that $D E$ alone is not a derivation - the cross terms are exactly what obstructs it.

Formula (4.4) follows by evaluating $[D,E]$ at $x_j$ and applying the structure theorem.

Extension to localisations and to power series

Let $D ∈ Der_K(K[X])$ . Then (i) $D$ extends uniquely to a $K$ -derivation of any localisation of $K[X]$ , in particular of $K[X, Δ^-1]$ and of $K(X)$ , by formula (4.5); and (ii) $D$ extends uniquely to a $K$ -derivation of the formal power series ring $K[[x_1, … ,x_n]]$ by the same expression $∑ _i f_i ∂_i$ , which converges in the $(x)$ -adic topology because $∂_i$ lowers order by one.

Part (ii) is what makes the power series ring available as a workspace: identities between derivations can be checked there, where the local inversion theorem is available, and then restricted back to $K[X]$ .

Where (4.6) comes from

If $F$ were an isomorphism, $F_1, … ,F_n$ would be a new coordinate system and $∂/∂ F_i$ would be the corresponding partial derivative. The chain rule $∂ g/∂ x_j = ∑ _i (∂ g/∂ F_i)(∂ F_i / ∂ x_j)$ is a linear system for the unknowns $∂ g/∂ F_i$ with coefficient matrix $J(F)$ ; Cramer's rule solves it and gives (4.6). The point of the definition is that it makes sense whenever $Δ$ is invertible, without assuming $F$ is an isomorphism - which is precisely the situation the Jacobian conjecture is about.

Key Equations

The Leibniz rule and its immediate consequences:

D(ab) = D(a)b + aD(b), D(a^k) = k,a^k-1D(a), D(λ) = 0 (λ ∈ K).

(4.2)

On a monomial $x^α = x_1^α_1 ⋅ s x_n^α_n$ the structure theorem reads

D(x^α) = ∑ _i=1^n α_i, x^α - e_i, D(x_i),

(4.3)

where $e_i$ is the $i$ -th standard multi-index; the sum runs over the $i$ with $α_i > 0$ .

The commutator of two derivations is again a derivation, so $Der_K(K[X])$ is a Lie algebra over $K$ . In coordinates, if $D = ∑ _i f_i∂_i$ and $E = ∑ _j g_j ∂_j$ then

[D,E] = ∑ _j=1^n ( D(g_j) - E(f_j) ),∂_j.

(4.4)

A derivation extends uniquely to any localisation, by the quotient rule; this is what allows the derivations of Chapter 4 to be defined first on $K(X)$ and then restricted:

D!((a)/(b)) = D(a),b - a,D(b)b^2, D(b^-k) = -k,b^-k-1D(b).

(4.5)

Finally, the derivations attached to a polynomial map $F = (F_1, … ,F_n)$ with $Δ = J(F)$ nowhere zero are defined by Cramer's rule:

D_i(g) = Δ^-1, J (F_1, … ,F_i-1,, g,, F_i+1, … ,F_n ),

(4.6)

the operator that deserves the name $∂/∂ F_i$ . It is a derivation of $K[X, Δ^-1]$ , and of $K[X]$ itself when $Δ$ is a non-zero constant.

Variable Definitions

$K$: the ground field, of characteristic zero
$X$: the affine space $K^n$
$K[X]$: the polynomial ring $K[x_1, … ,x_n]$ , the coordinate ring of $X$
$K(X)$: the field of rational functions in $x_1, … ,x_n$
$D$ , $E$: $K$ -derivations of $K[X]$
$∂_i$: the partial derivative $∂/∂ x_i$ , a $K$ -derivation of $K[X]$
$Der_K(S)$: the set of $K$ -derivations of the $K$ -algebra $S$
$S^D$: the ring of constants, that is $ker D$
$F$: a polynomial map $X \to X$ with coordinate functions $F_1, … ,F_n$
$J(F)$: the Jacobian matrix of $F$ , with $i j$ entry $∂ F_i/∂ x_j$
$Δ$: the Jacobian determinant $J(F)$
$ad_a$: the map $b ↦ [a,b]$ on the Weyl algebra

Properties and Behaviour

The module and Lie algebra structure

$Der_K(K[X])$ is simultaneously (i) a free $K[X]$ -module of rank $n$ and (ii) an infinite-dimensional Lie algebra over $K$ . The two structures are compatible in the sense that $[D, gE] = D(g)E + g[D,E]$ for $g ∈ K[X]$ - the bracket is not $K[X]$ -bilinear, only $K$ -bilinear.

The degree grading

Assign to $x^α ∂_i$ the degree $|α| - 1$ . This makes $Der_K(K[X]) = ⨁ _k ≥ -1 L_k$ a graded Lie algebra with $[L_k, L_ℓ] ⊂ eq L_k+ℓ$ . The piece $L_-1$ is spanned by $∂_1, … ,∂_n$ (translations), and $L_0$ is spanned by the $x_j∂_i$ and is a copy of $gl_n$ . This graded Lie algebra is the Witt algebra $W_n$ of Cartan type.

Constants of a non-zero derivation

If $D ≠ 0$ then $D ≠ K[X]$ , and $ker D$ has transcendence degree at most $n-1$ over $K$ . In particular a non-zero derivation of $K[x]$ in one variable has $D = K$ .

For the one-variable case: $D = f∂_x$ with $f ≠ 0$ , and $D(g) = f g'$ , which vanishes only when $g' = 0$ , that is when $g ∈ K$ - here characteristic zero is used.

Constants need not be finitely generated

For $n ≤ 3$ the ring of constants of a derivation of $K[x_1, … ,x_n]$ is a finitely generated $K$ -algebra. In higher dimension this can fail: Nagata's counterexamples to Hilbert's fourteenth problem produce rings of invariants, and hence rings of constants, that are not finitely generated. Do not assume $ker D$ is a polynomial ring, or even Noetherian, without proof.

Derivations and endomorphisms of $A_n$

Given polynomials $F_1, … ,F_n$ and derivations $D_1, … ,D_n$ of $K[X]$ with $D_i(F_j) = δ_ij$ and $[D_i,D_j] = 0$ , the assignments $x_i ↦ F_i$ , $∂_i ↦ D_i$ satisfy the defining relations of the Weyl algebra and therefore extend to an algebra endomorphism of $A_n$ . This is the mechanism behind the Weyl algebra route to the Jacobian conjecture; see the presentation of P8.

Examples and Special Cases

The partial derivatives

$∂_i$ has $∂_i = K[x_1, … , x_i, … ,x_n]$ , the polynomials not involving $x_i$ . Each $∂_i$ is locally nilpotent, since applying it more than $deg g$ times to $g$ gives zero.

The Euler derivation

$ε = ∑ _i x_i ∂_i$ satisfies $ε(g) = ( g),g$ for homogeneous $g$ . Its constants are exactly $K$ in characteristic zero, and it is as far from locally nilpotent as possible: no non-constant polynomial is killed by any power.

A triangular derivation with a slice

$D = ∂_x + y,∂_z$ on $K[x,y,z]$ is locally nilpotent and $D(x) = 1$ , so $x$ is a slice. The structure theorem for such derivations (see the next page) then gives $K[x,y,z] = D,[x]$ with $D = K[y, z - xy]$ ; note $D(z - xy) = y - y = 0$ .

A derivation with no slice

$D = y^2∂_x$ on $K[x,y]$ is locally nilpotent, with $D = K[y]$ . But every value $D(g) = y^2 ∂_x g$ lies in the ideal $(y^2)$ , so no $t$ with $D(t) = 1$ exists. Local nilpotence alone does not supply a slice.

The Jacobian derivations for a shear

Let $F = (x + y^2,, y)$ on $K^2$ , so $J(F) = pmatrix 1 & 2y 0 & 1 pmatrix$ and $Δ = 1$ . Formula (4.6) gives $D_1(g) = pmatrix g_x & g_y 0 & 1 pmatrix = g_x$ and $D_2(g) = pmatrix 1 & 2y g_x & g_y pmatrix = g_y - 2y,g_x$ . So $D_1 = ∂_x$ and $D_2 = ∂_y - 2y,∂_x$ . One checks $D_1(F_1) = 1$ , $D_1(F_2) = 0$ , $D_2(F_1) = 2y - 2y = 0$ , $D_2(F_2) = 1$ , and $[D_1,D_2] = 0$ - exactly the relations required to build an endomorphism of $A_2$ .

Worked Example

Two derivations of $K[x,y]$ , their constants, and their bracket

Step 1 - write the derivations down
Take $n = 2$ and write $K[x,y]$ . Define $H$ and $R$ by their values on the variables:

$H(x) = x, H(y) = -y ⟹ H = x∂_x - y∂_y; R(x) = -y, R(y) = x ⟹ R = x∂_y - y∂_x.$

By the structure theorem no further checking is needed: these prescriptions define unique $K$ -derivations. Geometrically $H$ is the hyperbolic scaling field and $R$ is the rotation field.
Step 2 - the constants of $H$
By (4.3), $H(x^a y^b) = a,x^a y^b - b,x^a y^b = (a-b),x^a y^b$ . So $H$ is diagonal in the monomial basis with eigenvalue $a-b$ on $x^a y^b$ . A polynomial is killed by $H$ exactly when every monomial occurring in it has $a = b$ , that is when it is a polynomial in $x y$ :

$H = K[xy].$

Sanity check: $H(xy) = H(x)y + xH(y) = xy - xy = 0$ . The ring of constants is therefore a polynomial ring in one variable, strictly bigger than $K$ but far smaller than $K[x,y]$ .
Step 3 - the constants of $R$
First check $R(x^2 + y^2) = -y ⋅ 2x + x ⋅ 2y = 0$ , so $K[x^2+y^2] ⊂ eq R$ . For the reverse inclusion, work over a field containing $i$ and set $u = x + iy$ , $v = x - iy$ , so that $K[x,y] = K[u,v]$ . Then

$R(u) = R(x) + iR(y) = -y + ix = i(x+iy) = iu, R(v) = -y - ix = -i(x - iy) = -iv.$

So $R = i(u∂_u - v∂_v)$ , which is $i$ times a derivation of the same shape as $H$ . By Step 2, $R = K[uv] = K[(x+iy)(x-iy)] = K[x^2+y^2]$ .
Step 4 - the bracket, and a copy of $sl_2$
Use (4.4), or evaluate directly. $[H,R](x) = H(R(x)) - R(H(x)) = H(-y) - R(x) = y - (-y) = 2y$ , and $[H,R](y) = H(x) - R(-y) = x + x = 2x$ . Hence

$[H,R] = 2y,∂_x + 2x,∂_y,$

which is neither a multiple of $H$ nor of $R$ : the Lie algebra is genuinely non-abelian. Setting $E = x∂_y$ and $F = y∂_x$ , the same method gives $[H,E] = 2E$ , $[H,F] = -2F$ and $[E,F] = H$ . So ${E, F, H}$ spans a copy of $sl_2$ inside $Der_K(K[x,y])$ .
Step 5 - which of these are locally nilpotent
$E = x∂_y$ satisfies $E(y) = x$ and $E(x) = 0$ , so $E^2$ kills both variables and $E$ is locally nilpotent with $E = K[x]$ . But $H^k(x) = x$ for every $k$ , so $H$ is not locally nilpotent; and $R^2(x) = R(-y) = -x$ , so $R$ is not either. Local nilpotence is a genuinely restrictive extra condition, not a formality.

Result

$H = x∂_x - y∂_y$ has $H = K[xy]$ ; $R = x∂_y - y∂_x$ has $R = K[x^2+y^2]$ ; $[H,R] = 2(y∂_x + x∂_y)$ . Neither $H$ nor $R$ is locally nilpotent, while $x∂_y$ is. The derivations $x∂_y$ , $y∂_x$ , $x∂_x - y∂_y$ span a copy of $sl_2$ .

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 derivations (4.6) are the input to the reduction described on the automorphisms page; the conjecture becomes a statement about their local nilpotence.
Invariant theory. Rings of constants of locally nilpotent derivations are exactly the rings of invariants of algebraic actions of the additive group $G_a$ , which is how Nagata's counterexample to Hilbert's fourteenth problem is usually presented.
Dynamical systems. A polynomial vector field is a derivation, and polynomial first integrals are its constants. The global asymptotic stability problem is a question about such fields, and Chapter 19 of the primer attacks it with $D$ -module methods.
Differential Galois theory. A differential field is a field with a derivation, and Picard-Vessiot theory studies extensions of it; polynomial derivations are the simplest concrete examples.
Computer algebra. Derivations are how symbolic differentiation is implemented: a rule table on generators plus the Leibniz rule, which is precisely the structure theorem in software form.

Design Considerations

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

Which ring to work over

The derivations (4.6) exist a priori only on $K(X)$ , because Cramer's rule divides by $Δ$ . Restricting to $K[X,Δ^-1]$ keeps a Noetherian ring and loses nothing; restricting further to $K[X]$ is legitimate only when $Δ$ is a unit, that is a non-zero constant. Chapter 4 makes exactly this choice, which is why the hypothesis $Δ = 1$ rather than $Δ ≠ 0$ is what is needed to build an endomorphism of $A_n$ .

Commutator or composite

When a computation needs to stay inside the world of vector fields, use the bracket; when it needs the full algebra of operators, move to $A_n$ and use composition. Moving from $Der_K(K[X])$ to $A_n$ is a genuine enlargement - it is the passage from a Lie algebra to its enveloping algebra - and it is worth doing only when higher-order operators are actually required.

Formal power series as a scratch space

Identities among derivations that are hard to verify polynomially are often easy in $K[[X]]$ , where the local inversion theorem makes any map with invertible Jacobian at the origin a genuine change of coordinates. The proof that the $D_i$ of (4.6) commute is carried out this way: extend, verify, restrict.

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. Characteristic zero is required for the statements about kernels; $ℚ$ for exact computation, $ℂ$ for the geometry and for the flow interpretation.
Coefficient ring. The structure theorem holds over any commutative ring $R$ in place of $K$ , giving $Der_R(R[X])$ free of rank $n$ . This matters when a parameter is carried along, as with $A_n[s]$ in the theory of the b-function.
Representation. A derivation is stored as the $n$ -tuple $(D(x_1), … ,D(x_n))$ - nothing else is needed, and any other representation is redundant.
Localised ring. $K[X,Δ^-1]$ is stored as pairs (numerator, exponent of $Δ$ ); the quotient rule (4.5) keeps the denominator a power of $Δ$ automatically.

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.

Concrete tasks and their costs:

Applying $D$ to a polynomial. Linear in the number of terms of $g$ times $n$ ; a table lookup of $D(x_i)$ and formula (4.3).
Deciding local nilpotence. It is enough to test $D$ on the variables: the elements killed by some power of $D$ form a subalgebra, by the Leibniz formula for $D^m(ab)$ . So compute $D(x_i), D^2(x_i), …$ for each $i$ and look for zero. The catch is that there is no a priori bound on how long to keep going, so this is a semi-decision procedure unless extra structure is known; see locally nilpotent derivations.
Computing $ker D$ . This is a hard problem. For locally nilpotent $D$ with a slice the answer is immediate; in general one uses the van den Essen algorithm, which terminates when the kernel is finitely generated but need not terminate otherwise.
Computing the $D_i$ of (4.6). $n$ determinants of $n \times n$ polynomial matrices, so $O(n ⋅ n!)$ naive or $O(n^4)$ by fraction-free Gaussian elimination, with coefficient growth the practical bottleneck.

Implementations: Singular's derham.lib and jacobson.lib, the ore_algebra package in SageMath for the operator side, and Macaulay2 for the commutative kernel computations. Van den Essen's kernel algorithm is available in Singular. For symbolic vector fields and their brackets, any general computer algebra system suffices.

Limits of Validity

The structure theorem $D = ∑ _i D(x_i)∂_i$ is robust: it uses only the Leibniz rule and the fact that $x_1, … ,x_n$ generate $K[X]$ as a $K$ -algebra, so it holds over any commutative base ring and in any characteristic. What is fragile is everything about the kernel.

Characteristic zero. $∂_x = K[y]$ in $K[x,y]$ only in characteristic zero. In characteristic $p$ , $∂_x(x^p) = p x^p-1 = 0$ , so $∂_x = K[x^p, y]$ , and the whole theory of constants changes. See the positive characteristic page.
Smoothness. For a singular affine variety the derivations of the coordinate ring need not form a free module, and $Der_K$ can fail to generate the ring of differential operators. The clean picture on this page is a statement about affine space.
Finite generation of the kernel. Guaranteed only in low dimension; see the proposition above.
Derivations of $K[X]$ that are not $K$ -linear. If $K$ is not perfect or one allows derivations non-trivial on $K$ , the module $Der(K[X])$ is larger. Everything here assumes $D|_K = 0$ .

Failure Modes and Common Mistakes

Composing derivations

$D E$ is almost never a derivation: $∂_x^2(x ⋅ x) = 2$ , but $∂_x^2(x) ⋅ x + x ⋅ ∂_x^2(x) = 0$ . Only the commutator $DE - ED$ survives, which is why $Der_K$ is a Lie algebra and not an associative algebra. The associative algebra it generates inside $End_K(K[X])$ is the Weyl algebra.

Assuming $D(x_i)$ can be prescribed with constraints

The map $D ↦ (D(x_1), … ,D(x_n))$ is a bijection onto $K[X]^n$ : any tuple occurs, with no compatibility condition. This is special to a polynomial ring. On a quotient $K[X]/I$ a candidate derivation must additionally satisfy $D(I) ⊂ eq I$ , and most tuples fail that test.

Confusing the ring of constants with the fixed ring of an automorphism

For a locally nilpotent $D$ over a field of characteristic zero the two do coincide: $ker D$ equals the fixed ring of $(D)$ . For a general derivation there is no automorphism $(D)$ to speak of, because the exponential series does not terminate. The Euler derivation is the standard warning: its kernel is $K$ , but it does not exponentiate inside the polynomial ring at all.

Reading (4.6) as if $F$ were invertible

The notation $∂/∂ F_i$ invites the reading "differentiate in the new coordinates $F$ ". That reading presupposes what the Jacobian conjecture asks. All that (4.6) actually gives, when $Δ$ is a non-zero constant, is $n$ commuting derivations of $K[X]$ with $D_i(F_j) = δ_ij$ ; whether they generate a coordinate system is exactly the open question.

Historical Notes

Derivations were isolated as an algebraic notion in the first half of the twentieth century, in the work of Ritt on differential algebra and of Zariski and others on algebraic geometry; the identification of $Der_K(K[X])$ with polynomial vector fields is essentially as old as the language of modules.

The specific use made of derivations in Chapter 4 of the primer comes from a much later strand. D. Wright's 1981 paper on the Jacobian conjecture supplies the structure theorem for a family of commuting locally nilpotent derivations with a dual system of slices, and L. Vaserstein and V. Katz are credited by Bass, Connell and Wright with the observation that this converts the Dixmier problem into the Jacobian conjecture. Coutinho's Chapter 4 is a presentation of that argument.

Meanwhile the study of locally nilpotent derivations became a subject in its own right, driven by Nagata's 1959 counterexample to Hilbert's fourteenth problem, by Makar-Limanov's invariant, and by the cancellation problem. Freudenburg's monograph is the standard reference for that literature.

Comparison

Five derivations of $K[x,y]$ compared. Only the first two are locally nilpotent.
Derivation	Locally nilpotent?	Ring of constants	Slice exists?	Geometric picture
$∂_x$	yes	$K[y]$	yes ( $t = x$ )	translation
$x∂_y$	yes	$K[x]$	no	shear
$y^2∂_x$	yes	$K[y]$	no	degenerate shear
$x∂_x - y∂_y$	no	$K[xy]$	no	hyperbolic scaling
$x∂_y - y∂_x$	no	$K[x^2+y^2]$	no	rotation

Key Takeaways

Key points

A $K$ -derivation of $K[x_1, … ,x_n]$ is determined by the $n$ polynomials $D(x_i)$ , and equals $∑ _i D(x_i)∂_i$ ; the proof is that $ker$ of a derivation is a subalgebra.
$Der_K(K[X])$ is a free $K[X]$ -module of rank $n$ and a Lie algebra over $K$ under the commutator; the composite $D E$ is not a derivation.
The kernel $ker D$ is a subalgebra, but its size is not controlled by anything in the definition and it need not be finitely generated in dimension $≥ 4$ .
Derivations extend uniquely to localisations by the quotient rule and to formal power series termwise.
For a polynomial map $F$ with invertible Jacobian determinant, Cramer's rule defines commuting derivations $D_i$ with $D_i(F_j) = δ_ij$ ; they live on $K[X]$ when $Δ$ is constant.
Inside $A_n$ , a derivation acts on polynomials by commutator: $[D,g] = D(g)$ , which is how local nilpotence transfers between the two settings.

FAQs

Is every

K

-linear map satisfying the Leibniz rule automatically continuous or bounded in some sense?

There is no topology in play here - the statement is purely algebraic. What replaces continuity is the structure theorem: the derivation is determined by finitely many polynomials, so it is as "finite" a piece of data as one could want.

Why is

Der_K(K[X])

a module over

K[X]

but the bracket only

K

-bilinear?

Because $[D, gE] = D(g)E + g[D,E]$ : scaling one argument by a ring element produces an extra term $D(g)E$ . That failure is not a defect; it is the reason $Der_K$ is a Lie-Rinehart algebra rather than a Lie algebra over $K[X]$ .

Does

D = K

imply that

D

is "generic" in some sense?

It is the typical behaviour, but it carries no structural conclusion. The Euler derivation has $= K$ and is highly structured; a random field also has $= K$ . Conversely a large kernel does say something: it means the associated flow has conserved quantities.

How do I know the

D_i

of (4.6) really commute?

They do, and it is Coutinho (4.4.1). The argument extends everything to the power series ring, notes that $[D_i,D_j]$ is a derivation killing every $F_k$ , and then uses the local inversion theorem to conclude that the $F_k$ generate the power series ring after a translation, so the bracket kills everything. The step needing care is that $F_1, … ,F_n$ are algebraically independent, which follows from injectivity of the comorphism.

Can a derivation of

K[X]

be nilpotent as an operator, rather than only locally nilpotent?

Only if it is zero. If $D ≠ 0$ then $D(x_i) ≠ 0$ for some $i$ , and by the graded structure one can produce elements on which arbitrarily high powers of $D$ do not vanish. Local nilpotence - every element killed by some power, with the power depending on the element - is the strongest condition available.

What is the relationship between

Der_K(K[X])

and

A_n

$A_n$ is generated as a $K$ -algebra by $K[X]$ and $Der_K(K[X])$ ; in fact $A_n$ is the enveloping algebra of the Lie-Rinehart pair. Concretely, the elements of $A_n$ of order at most one are exactly $K[X] ⊕ Der_K(K[X])$ , as described on the page about the Weyl algebra as differential operators.

Do these results need

K

algebraically closed?

No. The structure theorem and the kernel computations here work over any field of characteristic zero. Algebraic closure is convenient in Step 3 of the worked example, where $i$ is used, but the conclusion $R = K[x^2+y^2]$ holds over $ℚ$ as well, since the kernel does not change under field extension.

Is the ring of constants always integrally closed in

K[X]

For a locally nilpotent derivation in characteristic zero the kernel is factorially closed, which is stronger. For a general derivation this can fail, so do not assume it; check the specific derivation.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 4 §3 and §4, with Lemma (4.4.1) and Exercises 5.7-5.9.
D. Wright, On the Jacobian conjecture, Illinois Journal of Mathematics 25 (1981), 423-440 - the source of the structure theorem for commuting locally nilpotent derivations with slices.
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.
G. Freudenburg, Algebraic Theory of Locally Nilpotent Derivations, Encyclopaedia of Mathematical Sciences 136, Springer, 2nd edition, 2017 - the standard modern reference for derivations of polynomial rings.
A. van den Essen, Polynomial Automorphisms and the Jacobian Conjecture, Progress in Mathematics 190, Birkhäuser, 2000 - Ch. 1 for derivations, Ch. 2 for kernel algorithms.
M. Nagata, On the fourteenth problem of Hilbert, Proceedings of the International Congress of Mathematicians 1958, Cambridge University Press, 1960, 459-462.
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 derivations and Ore extensions.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Compute the ring of constants of $x∂_x + 2y∂_y$ on $K[x,y]$ and explain the answer via the eigenvalue decomposition.
Show that $Der_K(K[x]/(x^2))$ is not a free module and identify it explicitly.
Verify formula (4.4) for the bracket by evaluating both sides on a monomial of degree three.
For $F = (x + y^3, y)$ compute the derivations $D_1, D_2$ of (4.6) and check that they commute.
Prove that a derivation of $K[X]$ vanishing on a set of algebra generators is zero, and find where characteristic zero is or is not used.
Describe the graded pieces $L_-1, L_0, L_1$ of $Der_K(K[x,y])$ and compute the brackets between them.
Give a derivation of $K[x,y,z]$ whose ring of constants has transcendence degree two but is not a polynomial ring.