Derivations of a Commutative Ring

Q: Why does a derivation have to kill 1?

Apply the Leibniz rule to 1 ⋅ 1 : it gives D(1) = 2D(1) , so D(1) = 0 . Since D is K -linear, D vanishes on all of K . This is not an extra axiom; it is forced.

Q: Is every derivation of K[x_1, … ,x_n] really of the form ∑ f_i ∂_i?

Yes, with f_i = D(x_i) , and the representation is unique. The proof is three lines: subtract ∑ D(x_i)∂_i from D and check the difference kills every monomial. See derivations of the polynomial ring for the same statement in the setting of the Jacobian conjecture.

Q: Is the composition of two derivations a derivation?

No. DE satisfies DE(ab) = aDE(b) + bDE(a) + D(a)E(b) + D(b)E(a) ; the two cross terms spoil it. They cancel in DE - ED , which is why the commutator is a derivation and composition is not.

Q: How do derivations relate to differential operators of higher order?

They are the order-one part. In the inductive definition, P has order at most m if [P,a] has order at most m-1 for every a ∈ R ; derivations are exactly the operators with [P,a] of order zero and P(1) = 0 . See the order filtration .

Q: What is the difference between Der_K(R) and the tangent space?

Der_K(R) consists of maps R → R - global vector fields. A tangent vector at a point p is a map R → K satisfying the Leibniz rule with respect to evaluation at p . Evaluating a global field at p gives a tangent vector, but not every tangent vector arises that way when X is singular.

Q: Why does characteristic zero keep appearing?

Not for the results on this page - freeness of Der_K(K[X]) and the description of order-one operators hold in any characteristic. It appears when one asks whether these operators generate everything. In characteristic p , D(a^p) = 0 for every derivation, so the divided-power operators such as ∂^p/p! are differential operators that no combination of derivations can reach.

Overview

The Weyl algebra was built by hand: take the polynomial ring $K[x_1, … ,x_n]$ , adjoin the partial derivatives, and see what the resulting subring of $End_K(K[x_1, … ,x_n])$ looks like. That construction is tied to one particular ring. To see the Weyl algebra as one member of a family, one needs a definition of differential operator that mentions only the ring being differentiated. The inductive definition of P2 does exactly that, and its first non-trivial layer is the layer of derivations.

A derivation of a commutative $K$ -algebra $R$ is nothing more than a $K$ -linear map $D : R \to R$ that obeys the product rule. That single identity is enough to force a large amount of structure. The derivations form a $K$ -vector space; they form a module over $R$ ; they are closed under commutator but not under composition, so they form a Lie algebra rather than an associative one; and they are exactly the operators of order one that kill the constant $1$ .

Geometrically a derivation of the coordinate ring of an affine variety is a vector field on that variety, and evaluating a derivation at a point gives a tangent vector. This is why $Der_K(R)$ detects singularities: on a smooth irreducible variety of dimension $n$ the module of derivations is projective of rank $n$ , while on the cusp it is a rank-one module that needs two generators. The worked example on this page computes that case in full.

The page states the definition, proves the structural facts, computes $Der_K$ for the polynomial ring and for two singular examples, and explains the exact sense in which derivations are the order-one part of $D(R)$ .

Definition

Throughout, $K$ is a field of characteristic zero and $R$ is a commutative $K$ -algebra with $1$ . All maps are $K$ -linear.

DerivationCoutinho, Ch. 3 §1

A derivation of $R$ over $K$ is a $K$ -linear map $D : R \to R$ such that

D(ab) = a,D(b) + b,D(a) for all a, b ∈ R.

(3.1)

The set of all such maps is written $Der_K(R)$ . It is a $K$ -subspace of $End_K(R)$ .

The $R$ -module structure

For $a ∈ R$ and $D ∈ Der_K(R)$ , define $a D$ by $(aD)(b) = a,D(b)$ . This is again a derivation, and the operation makes $Der_K(R)$ a left $R$ -module.

Remark

Two conventions are worth fixing. First, derivation here always means $K$ -derivation: $D$ vanishes on $K$ . Second, $R$ is embedded in $End_K(R)$ by sending $a$ to the multiplication operator $r \mapsto a r$ ; this embedding is injective because $a ⋅ 1 = a$ . Statements such as \" $D^1(R) = R ⊕ Der_K(R)$ \" are understood inside $End_K(R)$ with that identification.

Core Concepts

Why the Leibniz rule is the right axiom

Linearity says $D$ respects the additive structure; the Leibniz rule is the weakest useful compatibility with multiplication. It cannot be strengthened to $D(ab) = D(a)D(b)$ without collapsing to something uninteresting, and it cannot be weakened to nothing without losing all contact with the ring structure. What makes (3.1) the right axiom is that it is first order: it expresses $D$ on a product in terms of $D$ on the factors, with coefficients from $R$ itself.

Derivations kill constants

Put $a = b = 1$ in (3.1): $D(1) = 1 ⋅ D(1) + 1 ⋅ D(1) = 2D(1)$ , so $D(1) = 0$ . By $K$ -linearity $D(λ) = 0$ for every $λ ∈ K$ . This is the algebraic form of the statement that the derivative of a constant vanishes, and it is where the phrase \"over $K$ \" earns its keep: a derivation over $ℚ$ of $C[x]$ need not kill $ℂ$ .

Vector fields and tangent vectors

If $X$ is an affine variety with coordinate ring $R = O(X)$ , an element of $Der_K(R)$ is a regular vector field on $X$ : it assigns to every point a tangent direction, algebraically. Composing a derivation with evaluation at a point $p$ gives a map $R \to K$ satisfying $δ(ab) = a(p)δ(b) + b(p)δ(a)$ , a point derivation, and the space of these is the Zariski tangent space at $p$ . The $R$ -module $Der_K(R)$ is therefore the module of sections of the tangent sheaf.

A Lie algebra, not an algebra

Composition destroys the Leibniz rule: $D^2(ab) = aD^2(b) + 2D(a)D(b) + bD^2(a)$ , and the cross term $2D(a)D(b)$ is not allowed. The commutator, however, is exactly what cancels it, so $Der_K(R)$ is closed under $[D,E] = DE - ED$ . This is the first sign that the ring generated by $R$ and $Der_K(R)$ is genuinely non-commutative, and it is where the Weyl algebra's commutation relations come from.

Construction and Proof

Three facts carry the theory. Each has a short honest proof.

Derivations form an $R$ -module and a $K$ -Lie algebra

For $a ∈ R$ and $D, E ∈ Der_K(R)$ , both $a D$ and $[D,E]$ lie in $Der_K(R)$ . The bracket satisfies the Jacobi identity, but $Der_K(R)$ is not a Lie algebra over $R$ unless $Der_K(R) = 0$ , because of the correction term in (3.3).

Proof

For $a D$ : $(aD)(bc) = a(bD(c) + cD(b)) = b,(aD)(c) + c,(aD)(b)$ , which is (3.1). For the bracket, expand both composites on a product:

DE(bc) = b,DE(c) + c,DE(b) + D(b)E(c) + D(c)E(b),

and the same formula with $D$ and $E$ interchanged. The two cross terms are symmetric in $D$ and $E$ , so they cancel on subtraction, leaving $[D,E](bc) = b,[D,E](c) + c,[D,E](b)$ . The Jacobi identity holds because it holds for commutators in any associative ring.

Derivations of the polynomial ringCoutinho (3.1.3)

Every $D ∈ Der_K(K[x_1, … ,x_n])$ has the form $D = ∑ _i=1^n f_i ∂_i$ with $f_i = D(x_i)$ , and the $f_i$ are uniquely determined. Hence $Der_K(K[X])$ is a free module of rank $n$ with basis $∂_1, … , ∂_n$ .

Proof

Set $E = D - ∑ _i D(x_i)∂_i$ , a derivation with $E(x_i) = 0$ for every $i$ . By (3.2) and the Leibniz rule, $E$ vanishes on every monomial $x_1^a_1 ⋅ s x_n^a_n$ : differentiating a product of variables produces only terms containing some $E(x_i)$ . Monomials span $K[X]$ over $K$ and $E$ is $K$ -linear, so $E = 0$ . Uniqueness follows by evaluating $∑ _i f_i∂_i$ at $x_j$ , which returns $f_j$ .

The argument uses nothing about the characteristic; freeness of $Der_K(K[X])$ holds over any base field. It is the identification of $D(K[X])$ with the whole Weyl algebra, not this proposition, that needs characteristic zero.

Order one equals $R$ plus derivationsCoutinho (3.1.1)

Inside $End_K(R)$ , the operators of order at most one are exactly the sums $D + a$ with $D ∈ Der_K(R)$ and $a ∈ R$ , and the sum is direct:

D^1(R) = R ⊕ Der_K(R), D^0(R) = R.

Proof

Let $Q$ have order at most one and set $P = Q - Q(1)$ , meaning $Q$ minus multiplication by the element $Q(1) ∈ R$ . Then $P(1) = 0$ and $P$ still has order at most one, so $[P,a]$ has order zero for every $a$ , hence $[[P,a],b] = 0$ for all $a,b ∈ R$ . Expanding that double commutator as an operator and applying it to $1$ gives

P(ab) - a P(b) - b P(a) + ab,P(1) = 0,

and since $P(1) = 0$ this is precisely the Leibniz rule for $P$ . Thus $Q = P + Q(1) ∈ Der_K(R) + R$ . Conversely every such sum has order at most one because $[D + a, b] = D(b)$ is a multiplication operator. Directness: a derivation that equals multiplication by $a$ sends $1$ to $a$ and to $0$ , so $a = 0$ . For order zero, an operator commuting with all multiplications is $R$ -linear, and an $R$ -linear endomorphism of $R$ is multiplication by its value at $1$ .

Key Equations

Iterating the Leibniz rule on a product of $k$ equal factors gives the power rule, valid in any characteristic:

D(a^k) = k,a^k-1 D(a), k ≥ 1.

(3.2)

The commutator of two derivations is a derivation, and the bracket is $R$ -bilinear only up to a correction term:

[D, aE] = a,[D,E] + D(a),E, a ∈ R,; D, E ∈ Der_K(R).

(3.3)

Inside $End_K(R)$ , a derivation is characterised by a commutator identity with multiplication operators. Writing $a$ for the multiplication operator by $a$ ,

D ∈ Der_K(R) ⟺ [D, a] = D(a) for every a ∈ R, and D(1) = 0.

(3.4)

This is the bridge to the order filtration: $[D,a]$ is again a multiplication operator, hence of order zero, so $D$ has order at most one. For the polynomial ring the module of derivations is free:

Der_K(K[x_1, … ,x_n]) = ⨁ _i=1^n K[x_1, … ,x_n], ∂_i, ∂_i = (∂)/(∂ x_i).

(3.5)

Finally, derivations are representable: they are the $R$ -linear functionals on the module of Kaehler differentials,

Der_K(R) ≅ Hom_R(Ω_R/K, R).

(3.6)

Variable Definitions

$K$: the ground field, of characteristic zero
$R$: a commutative $K$ -algebra with identity
$End_K(R)$: the ring of $K$ -linear maps $R \to R$ under composition
$D, E$: derivations of $R$ over $K$
$Der_K(R)$: the $R$ -module of all $K$ -derivations of $R$
$[D,E]$: the commutator $DE - ED$ , computed in $End_K(R)$
$D^m(R)$: the space of differential operators on $R$ of order at most $m$
$∂_i$: partial differentiation with respect to $x_i$
$Ω_R/K$: the module of Kaehler differentials of $R$ over $K$
$O(X)$: the coordinate ring of an affine variety $X$

Properties and Behaviour

Extension to localisations

Let $S ⊂ eq R$ be a multiplicative set. Every $D ∈ Der_K(R)$ extends uniquely to a derivation of $S^-1R$ , by the quotient rule

D!((a)/(s)) = (s,D(a) - a,D(s))/(s^2).

Uniqueness is forced: applying $D$ to $s ⋅ (a/s) = a$ determines $D(a/s)$ . In particular $Der_K(S^-1R) ≅ S^-1Der_K(R)$ when $R$ is Noetherian, so derivations localise.

Derivations of a quotientCoutinho, Ch. 3, Exercises 3.2 and 3.5

Let $S = K[x_1, … ,x_n]$ , let $J ⊂ eq S$ be an ideal, and put $Der_J(S) = D ∈ Der_K(S) : D(J) ⊂ eq J$ . Reduction mod $J$ gives a surjection $Der_J(S) ↠ Der_K(S/J)$ with kernel $JDer_K(S)$ , hence an isomorphism of $S/J$ -modules

Der_K(S/J) ≅ Der_J(S) / JDer_K(S).

Surjectivity uses that $S$ is a polynomial ring: given $\bar{D}$ , lift each $D( x_i)$ to some $g_i ∈ S$ and take $D = ∑ g_i ∂_i$ ; the chain rule then shows $D(J) ⊂ eq J$ automatically.

Smoothness criterion

If $X$ is a smooth irreducible affine variety of dimension $n$ over $K$ , then $Der_K(O(X))$ is a projective $O(X)$ -module of rank $n$ , and it generates $D(X)$ together with $O(X)$ . Conversely, on a singular variety the module of derivations typically fails to be projective, as the worked example shows. Projectivity of $Der_K(R)$ is one of several equivalent formulations of regularity for finitely generated reduced $K$ -algebras in characteristic zero.

Locally nilpotent derivations

A derivation $D$ is locally nilpotent if every element of $R$ is killed by some power of $D$ . These are precisely the derivations that exponentiate to an algebraic action of the additive group, via $(tD) = ∑ _k ≥ 0 t^k D^k/k!$ , a finite sum on each element. They are the main tool for constructing automorphisms of the Weyl algebra; see locally nilpotent derivations.

Examples and Special Cases

$R = K$

$Der_K(K) = 0$ , since a derivation kills all of $K$ . More generally $Der_K(L) = 0$ for any finite separable field extension $L/K$ ; in characteristic zero that is every finite extension.

$R = K[x]$

$Der_K(K[x]) = K[x],∂$ , free of rank one. The bracket is $[f∂, g∂] = (fg' - gf')∂$ , the Lie algebra of polynomial vector fields on the line. Note $[x∂, ∂] = -∂$ : the bracket is genuinely non-zero even in one variable.

Dual numbers $R = K[x]/(x^2)$

A derivation is determined by $D(x) = a$ , and $0 = D(x^2) = 2xa$ forces $a ∈ (x)$ . So $Der_K(R) = R, x∂$ , one-dimensional over $K$ , spanned by the derivation sending $x \mapsto x$ . The ring is not reduced, and $Der$ notices only the nilpotent direction.

The node $xy = 0$

For $R = K[x,y]/(xy)$ the derivations preserving $(xy)$ are generated by $x∂_x$ and $y∂_y$ modulo the ideal: $Der_K(R)$ is generated by $x∂_x$ and $y∂_y$ . Again the module is not free; a node, like a cusp, is visible in $Der$ .

Two points: $R = K × K$

Here $Der_K(R) = 0$ . The idempotent $e = (1,0)$ satisfies $e^2 = e$ , so $D(e) = 2eD(e)$ ; multiplying by $e$ gives $eD(e) = 0$ and hence $D(e) = 0$ , and likewise $D(1-e) = 0$ .

It is worth checking that $D(R)$ is also as small as possible, namely $R$ itself, even though $End_K(R)$ is four-dimensional. If $[P,e]$ is multiplication by $w$ , comparing values at $1$ and at $e$ gives $(1-e)P(e) = -eP(1-e)$ ; the left side lies in $(1-e)R$ and the right side in $e R$ , so both vanish. Thus $P$ preserves each factor and acts on each by a scalar. So $D(K × K) = K × K$ , and the operator swapping the two factors is a $K$ -endomorphism of infinite order. Finite order is a real restriction.

Worked Example

Derivations of the cusp $y^2 = x^3$

Step 1 - two models of the same ring
Let $J = (y^2 - x^3) ⊂ K[x,y]$ and $R = K[x,y]/J$ , the coordinate ring of the cuspidal cubic. The map $x ↦ t^2$ , $y ↦ t^3$ identifies $R$ with the subring $K[t^2,t^3] ⊂ eq K[t]$ : it is surjective onto that subring by construction, and injective because $K[t^2,t^3]$ has the same Hilbert series as $K[x,y]/(y^2-x^3)$ under the weighting $x = 2$ , $y = 3$ . Concretely $R = K ⊕ span_Kt^k : k ≥ 2$ : every power of $t$ except $t^1$ appears.
Step 2 - move to the fraction field
$R$ is a domain with fraction field $K(t)$ , since $t = t^3/t^2$ . Every derivation of $R$ extends uniquely to $K(t)$ by the quotient rule, and every $K$ -derivation of $K(t)$ is $f,∂_t$ for a single $f ∈ K(t)$ . Hence

$Der_K(R) = , f,∂_t ;:; f ∈ K(t), f,∂_t(R) ⊂ eq R ,.$

Because $R$ is generated by $t^2$ and $t^3$ , the condition $f∂_t(R) ⊂ eq R$ is equivalent to the two conditions $2tf ∈ R$ and $3t^2 f ∈ R$ , that is (char $K = 0$ ) to $tf ∈ R$ and $t^2 f ∈ R$ .
Step 3 - solve the two conditions
From $tf ∈ R ⊂ eq K[t]$ we get $f ∈ t^-1K[t]$ , so write $f = ∑ _j ≥ -1 c_j t^j$ . Then $tf = ∑ _j c_j t^j+1$ lies in $R$ if and only if its $t^1$ coefficient vanishes, i.e. $c_0 = 0$ . And $t^2 f = ∑ _j c_j t^j+2$ lies in $R$ if and only if $c_-1 = 0$ . Therefore $c_-1 = c_0 = 0$ and

$Der_K(R) = , f,∂_t : f ∈ t,K[t] ,.$
Step 4 - generators, and the failure of freeness
As an $R$ -module, $tK[t] = R,t + R,t^2$ : indeed $Rt = spant, t^3, t^4, …$ and $Rt^2 = spant^2, t^4, t^5, …$ , whose sum is $spant,t^2,t^3, …$ . So $Der_K(R)$ is generated by $δ_1 = t,∂_t$ and $δ_2 = t^2,∂_t$ .

It is not generated by one element. A single generator $g,∂_t$ would have $Rg = tK[t]$ ; but $R$ contains no element of $t$ -degree $1$ , so $Rg ⊂ eq spant^v, t^v+2, t^v+3, …$ where $v$ is the order of vanishing of $g$ , and that set omits $t^v+1$ . Hence $Der_K(R)$ has rank $1$ but minimal number of generators $2$ : it is not free, and not projective.
Step 5 - translate back to $x$ and $y$
Lift $δ_1$ and $δ_2$ to $K[x,y]$ . Set $D_1 = 2x,∂_x + 3y,∂_y$ and $D_2 = 2y,∂_x + 3x^2,∂_y$ . Both preserve $J$ :

$D_1(y^2 - x^3) = 6y^2 - 6x^3 = 6,(y^2-x^3), D_2(y^2 - x^3) = -6x^2 y + 6x^2 y = 0,$

so each induces a derivation of $R$ . Under $x = t^2$ , $y = t^3$ : $D_1(x) = 2t^2$ forces the corresponding $f$ to satisfy $2tf = 2t^2$ , so $f = t$ and $D_1 ↦ t∂_t = δ_1$ ; likewise $D_2(x) = 2y = 2t^3$ gives $f = t^2$ and $D_2 ↦ t^2∂_t = δ_2$ . The two pictures agree.

Result

$Der_K(K[t^2,t^3]) = R,(t∂_t) + R,(t^2∂_t)$ , a rank-one $R$ -module requiring two generators, hence not free. In the $(x,y)$ model the generators are $2x∂_x + 3y∂_y$ (the weighted Euler derivation) and $2y∂_x + 3x^2∂_y$ . The failure of freeness is the algebraic signature of the cusp at the origin, where the Zariski tangent space is two-dimensional on a one-dimensional curve.

Applications and Industry Use

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

Differential Galois theory and symbolic integration. A differential field is a field with a distinguished derivation; the Risch algorithm for integration in closed form is a statement about extensions of differential fields.
Invariant theory and group actions. Locally nilpotent derivations correspond to additive group actions, and their rings of constants are the invariant rings; Nagata's counterexample to Hilbert's fourteenth problem is phrased this way.
The Jacobian conjecture. The Jacobian matrix of a polynomial map is the matrix of the induced action on $Der_K(K[X])$ ; the Jacobian conjecture and the Dixmier conjecture are both statements about the interaction of derivations with endomorphisms.
Deformation theory. $Der_K(R)$ computes first-order deformations of the $K$ -algebra structure, and its Lie bracket is the first bracket of the Hochschild complex.
Control and dynamics. A polynomial vector field on $K^n$ is a derivation of $K[x_1, … ,x_n]$ ; the D-module treatment of global asymptotic stability in this collection starts from that identification.
Computer algebra. Derivations are how differential systems are represented internally: an Ore algebra is specified by a commutative ring together with the commutation rule of a derivation with its coefficients.

Design Considerations

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

Which ring to differentiate

The definition of $Der_K(R)$ is uniform, but its quality depends entirely on $R$ . On a polynomial ring it is free with an obvious basis; on a smooth affine variety it is projective and still behaves like a vector bundle; on a singular ring it can be non-projective, and the ring generated by $R$ and $Der_K(R)$ then falls short of $D(R)$ . If a construction needs $D(R)$ to be generated by $R$ and $Der_K(R)$ , smoothness has to be part of the hypotheses, not an afterthought.

Which base to take derivations over

Taking derivations over $K$ rather than over $ℤ$ or over a subring is a modelling choice that fixes what counts as constant. Over $ℤ$ , $Der_Z(C[x])$ is enormous, containing everything coming from derivations of $ℂ$ over $ℚ$ . Pinning the base field down at the start avoids that.

Module or Lie algebra

$Der_K(R)$ carries two structures that do not combine into one: an $R$ -module structure and a $K$ -Lie bracket, linked by (3.3). Objects with both are Lie–Rinehart algebras, or algebroids in the geometric language. Choosing to remember only one of the two structures loses information: the module structure alone forgets non-commutativity, and the bracket alone forgets which vector fields are $R$ -multiples of which.

Standards and Codes

For mathematics, the relevant standards are notation, numeric and markup standards together with reference implementations.

ISO 80000-2 fixes $∂$ for partial differentiation and the upright roman convention for operator names, which is why $Der$ and $dim$ are set upright here while $D$ , a variable operator, is italic.
ISO/IEC 40314 (MathML 3.0) is the encoding used for every formula on this page.
Notation for the module of derivations is not standardised: $Der_K(R)$ , $Der(R/K)$ , $T_R$ and $Θ_R$ all appear in the literature for the same object, the last two in geometric contexts. The dual object $Ω_R/K$ is more uniformly named.
Software names differ as well: Macaulay2's Der, Singular's derivations in sing.lib, and the ore_algebra conventions in SageMath all describe derivations but with different input formats. None of them is a standard in the ISO sense.

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.

Two choices decide how well-behaved $Der_K(R)$ is.

Characteristic. In characteristic $p$ the power rule (3.2) makes $D(a^p) = 0$ for every derivation, so $R^p$ lies in the constants and $Der$ is a module over $R^p$ rather than a faithful invariant. The set of derivations is then closed under $D ↦ D^p$ , giving a restricted Lie algebra. None of the results on this page about recovering $D(R)$ from $R$ and $Der_K(R)$ survive; see the positive characteristic page.
Reducedness and smoothness. For $R$ reduced and finitely generated over a perfect field, $Der_K(R)$ has rank equal to $dim R$ at every generic point, and is projective exactly on the smooth locus. For non-reduced $R$ , as in the dual numbers, the rank statement fails.
Representation in software. A derivation of $K[x_1, … ,x_n]$ is stored as the vector $(D(x_1), … ,D(x_n))$ of polynomials. That is a complete and canonical encoding by (3.5); nothing else needs to be recorded. For a quotient ring, the same vector plus the ideal $J$ suffices, but the representation is no longer unique.

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.

Computing $Der_K(R)$ for $R = K[x_1, … ,x_n]/J$ with $J = (f_1, … ,f_r)$ is a syzygy computation.

Write the unknown derivation as $D = ∑ _i g_i ∂_i$ with $g_i ∈ S = K[x_1, … ,x_n]$ unknown.
Impose $D(f_j) ∈ J$ for each $j$ , that is $∑ _i g_i ,∂ f_j/∂ x_i ∈ J$ .
This is a linear system over $S$ whose solution module is computed as a syzygy module of the Jacobian matrix of $(f_1, … ,f_r)$ together with generators of $J$ .
Reduce modulo $JDer_K(S)$ to obtain $Der_K(S/J)$ .

Every step is a Groebner basis computation over a commutative polynomial ring, so the cost is the usual one: worst-case doubly exponential in $n$ , but tractable for small $n$ and low degrees. Macaulay2 exposes this directly as Der or via Hom(module of differentials, R); Singular provides derivations in sing.lib for hypersurfaces, and SageMath reaches it through its interface to those systems.

A practical check: for a reduced hypersurface $J = (f)$ the module $Der_J(S)$ always contains the $n2$ trivial derivations $∂_j(f)∂_i - ∂_i(f)∂_j$ and the multiples $f∂_i$ . A computation returning fewer generators than these has gone wrong. For $f = y^2 - x^3$ the trivial derivation is $-3x^2∂_y - 2y∂_x$ , which is $-D_2$ from the worked example.

Limits of Validity

The clean statements above have hypotheses that cannot be dropped.

Freeness is special to polynomial rings. (3.5) fails as soon as $R$ is singular. The cusp gives a rank-one module needing two generators, and higher-dimensional singularities are worse.
Derivations do not determine $D(R)$ . The subring of $End_K(R)$ generated by $R$ and $Der_K(R)$ can be strictly smaller than $D(R)$ . The cusp is again the standard witness; see differential operators on an affine variety.
Commutativity of $R$ is assumed. For non-commutative $R$ the Leibniz rule (3.1) is not the right axiom; one writes $D(ab) = D(a)b + aD(b)$ with the order of factors preserved, and the inner derivations $a ↦ [r,a]$ appear as a distinguished submodule with no commutative analogue.
Extension to localisations needs the quotient rule, not just linearity. A $K$ -linear map on $R$ need not extend to $S^-1R$ at all; it is the Leibniz rule that forces both existence and uniqueness.

Failure Modes and Common Mistakes

Treating $Der_K(R)$ as a ring

It is closed under addition, under multiplication by elements of $R$ , and under commutator, but not under composition. $∂^2$ is not a derivation: $∂^2(x ⋅ x) = 2 ≠ x∂^2(x) + x∂^2(x) = 0$ . The associative ring generated by $Der_K(R)$ and $R$ is a different and larger object - for $R = K[x_1, … ,x_n]$ it is the whole Weyl algebra.

Assuming the bracket is $R$ -bilinear

Equation (3.3) says otherwise: $[D, aE] = a[D,E] + D(a)E$ . Concretely $[∂, x∂] = ∂$ , which is not $x[∂,∂] = 0$ . Any argument that pulls a coefficient out of a bracket without the correction term is wrong.

Confusing $Der_K(R)$ with the tangent space

A derivation $R \to R$ is a global vector field; a tangent vector at a point $p$ is a point derivation $R \to K$ relative to evaluation at $p$ . Evaluation gives a map from the first to the second, but on a singular variety it is neither injective nor surjective in general. At the cusp point of $y^2 = x^3$ the Zariski tangent space is two-dimensional while every global vector field vanishes there.

Forgetting that the base field must be killed

$Der_Q(C[x])$ is vastly larger than $Der_C(C[x]) = C[x]∂$ , because $ℂ$ has non-zero $ℚ$ -derivations. Statements such as \"every derivation of the polynomial ring is $∑ f_i∂_i$ \" are false without the subscript $K$ and the assumption that $D$ vanishes on $K$ .

Using $D(a^k) = ka^k-1D(a)$ carelessly in characteristic $p$

The identity itself is valid in every characteristic, but its consequences are not. In characteristic $p$ it gives $D(a^p) = 0$ , so no derivation can be injective on $R$ , and arguments that recover $D$ from its values on generators of $R$ over $K$ break down when those generators involve $p$ -th powers.

Historical Notes

The Leibniz rule as an abstract axiom dates to the early twentieth century. Its systematic use in algebra begins with the work of Ritt in the 1930s on differential algebra, where differential fields and differential ideals were introduced to study algebraic differential equations, and with Kolchin's differential Galois theory in the following decades.

On the commutative-algebra side, the identification of $Der_K(R)$ with $Hom_R(Ω_R/K, R)$ and the use of $Ω$ as the algebraic replacement for the cotangent bundle belong to the Grothendieck school of the late 1950s and 1960s. Grothendieck's inductive definition of differential operators in EGA IV, which is the one used on the next page, places $Der_K(R)$ as the order-one layer of a filtration and thereby explains why derivations are so ubiquitous: they are simply the smallest non-trivial differential operators.

The failure of $R$ and $Der_K(R)$ to generate $D(R)$ on singular varieties was made explicit by Bernstein, Gelfand and Gelfand in 1972, using the cusp computed above; their example is the reason smoothness appears as a hypothesis throughout this part of the theory.

Comparison

How $Der_K(R)$ behaves for three sample rings.
Ring $R$	$Der_K(R)$	Rank	Free?	Does $R + Der$ generate $D(R)$ ?
$K[x_1, … ,x_n]$	free on $∂_1, … ,∂_n$	$n$	yes	yes, giving $A_n$
smooth affine $O(X)$ , $X = n$	projective	$n$	locally	yes
$K[t^2,t^3]$ (cusp)	$R,t∂_t + R,t^2∂_t$	$1$	no	no
$K[x]/(x^2)$	$R, x∂$	not defined (non-reduced)	no	no ( $D = End_K$ )
$K \times K$ (two points)	$0$	$0$	yes, trivially	yes, and both equal $R$

Key Takeaways

Key points

A derivation is a $K$ -linear map satisfying $D(ab) = aD(b) + bD(a)$ ; it automatically kills $K$ .
$Der_K(R)$ is an $R$ -module and a $K$ -Lie algebra, but not a ring and not an $R$ -Lie algebra: $[D, aE] = a[D,E] + D(a)E$ .
$Der_K(K[x_1, … ,x_n])$ is free with basis $∂_1, … ,∂_n$ ; every derivation is $∑ _i D(x_i)∂_i$ .
Inside $End_K(R)$ the operators of order at most one are exactly $R ⊕ Der_K(R)$ , and those of order zero are $R$ .
For the cusp $K[t^2,t^3]$ the module of derivations has rank one but needs two generators, so it is not free - singularities are visible in $Der$ .
Derivations extend uniquely to localisations and descend to quotients, with $Der_K(S/J) ≅ Der_J(S)/JDer_K(S)$ for $S$ a polynomial ring.

FAQs

Why does a derivation have to kill

1

Apply the Leibniz rule to $1 ⋅ 1$ : it gives $D(1) = 2D(1)$ , so $D(1) = 0$ . Since $D$ is $K$ -linear, $D$ vanishes on all of $K$ . This is not an extra axiom; it is forced.

Is every derivation of

K[x_1, … ,x_n]

really of the form

∑ f_i ∂_i

Yes, with $f_i = D(x_i)$ , and the representation is unique. The proof is three lines: subtract $∑ D(x_i)∂_i$ from $D$ and check the difference kills every monomial. See derivations of the polynomial ring for the same statement in the setting of the Jacobian conjecture.

Is the composition of two derivations a derivation?

No. $D E$ satisfies $DE(ab) = aDE(b) + bDE(a) + D(a)E(b) + D(b)E(a)$ ; the two cross terms spoil it. They cancel in $DE - ED$ , which is why the commutator is a derivation and composition is not.

How do derivations relate to differential operators of higher order?

They are the order-one part. In the inductive definition, $P$ has order at most $m$ if $[P,a]$ has order at most $m-1$ for every $a ∈ R$ ; derivations are exactly the operators with $[P,a]$ of order zero and $P(1) = 0$ . See the order filtration.

R

and

Der_K(R)

always generate the full ring of differential operators?

No. They do for smooth affine varieties in characteristic zero, and in particular for $K[x_1, … ,x_n]$ , where they generate the Weyl algebra. They do not for the cusp $K[t^2,t^3]$ : the operator $∂_t^2 - 2t^-1∂_t$ maps $K[t^2,t^3]$ into itself but is not a polynomial in the derivations of that ring.

What is the difference between

Der_K(R)

and the tangent space?

$Der_K(R)$ consists of maps $R \to R$ - global vector fields. A tangent vector at a point $p$ is a map $R \to K$ satisfying the Leibniz rule with respect to evaluation at $p$ . Evaluating a global field at $p$ gives a tangent vector, but not every tangent vector arises that way when $X$ is singular.

Why does characteristic zero keep appearing?

Not for the results on this page - freeness of $Der_K(K[X])$ and the description of order-one operators hold in any characteristic. It appears when one asks whether these operators generate everything. In characteristic $p$ , $D(a^p) = 0$ for every derivation, so the divided-power operators such as $∂^p/p!$ are differential operators that no combination of derivations can reach.

Does

Der_K(R)

see the difference between a node and a cusp?

Both give non-free modules, so at that level of resolution they look similar. The finer invariants differ: for the node $K[x,y]/(xy)$ the module of derivations is generated by $x∂_x$ and $y∂_y$ with the relation coming from $xy = 0$ , while for the cusp the two generators $t∂_t$ and $t^2∂_t$ satisfy relations reflecting the missing degree-one element of $R$ .

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 3 §1, Lemma (3.1.1) and Proposition (3.1.3); Exercises 3.5, 3.6, 3.8.
H. Matsumura, Commutative Ring Theory, Cambridge Studies in Advanced Mathematics 8, Cambridge University Press, 1986 - Ch. 9, for derivations, differentials and the smoothness criterion.
R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 52, Springer, 1977 - Ch. II §8, for the sheaf of differentials and the tangent sheaf.
A. Grothendieck, Éléments de géométrie algébrique IV, Publications Mathématiques de l'IHÉS 32 (1967), §16 - the inductive definition of differential operators, with derivations as the order-one part.
I. N. Bernstein, I. M. Gelfand and S. I. Gelfand, Differential operators on a cubic cone, Russian Mathematical Surveys 27 (1972), 169-174 - the first example where $R$ and its derivations do not generate all differential operators.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Graduate Studies in Mathematics 30, revised edition, American Mathematical Society, 2001 - Ch. 15, for rings of differential operators over regular rings.
A. van den Essen, Polynomial Automorphisms and the Jacobian Conjecture, Progress in Mathematics 190, Birkhäuser, 2000 - Ch. 1, for locally nilpotent derivations.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.
D. R. Grayson and M. E. Stillman, Macaulay2, a software system for research in algebraic geometry - documentation for Der and modules of differentials.

AI Suggested Questions

Prove that $[D,E]$ is a derivation by expanding both composites on a product, and identify exactly which terms cancel.
Compute $Der_K(K[x,y]/(y^2 - x^3))$ directly by the syzygy method and compare with the answer obtained through the parametrisation $x = t^2$ , $y = t^3$ .
Show that $Der_K(R)$ is a free module of rank $n$ when $R$ is a Laurent polynomial ring in $n$ variables, and write down the basis.
Give an example of a $K$ -linear map on $K[x]$ that is not a derivation but agrees with one on all monomials of degree at most $3$ .
Determine which derivations of $K[x,y]$ preserve the ideal $(xy)$ , and deduce $Der_K(K[x,y]/(xy))$ .
Explain why $Der_K(R) ≅ Hom_R(Ω_R/K, R)$ and use it to compute the derivations of a hypersurface ring.
Show that a locally nilpotent derivation of $K[x_1, … ,x_n]$ exponentiates to a polynomial automorphism, and compute the automorphism for $D = y∂_x$ on $K[x,y]$ .

Derivations of a Commutative Ring

Overview

Definition

Core Concepts

Why the Leibniz rule is the right axiom

Derivations kill constants

Vector fields and tangent vectors

A Lie algebra, not an algebra

Construction and Proof

Key Equations

Variable Definitions

Properties and Behaviour

Examples and Special Cases

Worked Example

Applications and Industry Use

Design Considerations

Which ring to differentiate

Which base to take derivations over

Module or Lie algebra

Standards and Codes

Material Selection

Computational Notes

Limits of Validity

Failure Modes and Common Mistakes

Historical Notes

Comparison

Key Takeaways

Key points

FAQs

Related Engineering Topics

References

AI Suggested Questions

Continue learning