Rings of Differential Operators: The Inductive Definition

Q: Why start the induction at D^-1(R) = 0?

It makes the single clause (3.1) cover n = 0 as well: [P,a] ∈ 0 for all a says exactly that P is R -linear, so D^0(R) = R . Without the convention one has to state the base case separately, as Coutinho does.

Q: Is it enough to test the commutator against a set of algebra generators of R?

For R = K[x_1, … ,x_n] yes, and that is how computations are done. The reason is that D^n is an R -bimodule and [P, ab] = a[P,b] + [P,a]b , so the condition propagates from generators to products. For a general R presented as S/J the reduction is not automatic and the relations must be taken into account.

Q: How does this recover the usual definition on K[x_1, … ,x_n]?

It gives D^m = ∑ _|α| ≤ m f_α ∂^α and D(K[X]) = A_n(K) , but only in characteristic zero, and the proof is not formal: it needs two lemmas, one saying an operator commuting with all x_i is a polynomial, the other a discrete Poincaré lemma. See the Weyl algebra as a ring of differential operators .

Q: Does the definition depend on R being commutative?

Yes, in two places. The identification D^0(R) = R uses that R -linear endomorphisms of R are multiplications, and the symmetry of (3.4) uses that the operators ad_a commute. Non-commutative analogues exist but require choices.

Q: Why is it important that commutators drop the order?

Because it makes gr,D(R) commutative. Every dimension, multiplicity and characteristic variety argument in D-module theory is carried out in that commutative graded ring and then transferred back. Without the drop there would be no commutative shadow to work in.

Q: Can D^n(R) be infinite dimensional over K?

Almost always. Already D^0(K[x]) = K[x] is infinite dimensional. Finite dimensionality of filtration pieces is a feature of the Bernstein filtration on the Weyl algebra, not of the order filtration.

Overview

For the polynomial ring one can simply write down the operators one wants: finite sums $∑ f_α∂^α$ . That description depends on having coordinates, and it does not survive a change of ring. If $R$ is the coordinate ring of a singular variety there are no coordinates and no obvious partial derivatives, yet there is still a sensible notion of differentiating a function on that variety.

The way out is to characterise differential operators by a property they have rather than by a formula they satisfy. The property is this: a differential operator of order $n$ fails to be $R$ -linear, but it fails in a controlled way - its commutator with multiplication by any element of $R$ is a differential operator of order $n-1$ . That is a definition by induction, with $R$ -linear operators as the base case, and it uses nothing but the ring structure of $R$ .

The definition produces a chain $R = D^0(R) ⊂ eq D^1(R) ⊂ eq D^2(R) ⊂ eq ⋅ s$ inside $End_K(R)$ , and the union $D(R)$ is a subring. Two facts make it a filtered ring rather than a merely increasing chain: composition adds orders, and commutators drop the total order by one. The second fact is what makes the associated graded ring commutative and puts geometry within reach.

This page sets up the definition, proves the two structural propositions, records the equivalent formulation by iterated commutators and Grothendieck's version by principal parts, and shows on $K[x]$ that finite order is a genuine restriction: the algebra endomorphism $x ↦ 2x$ is an operator of infinite order.

Definition

Let $K$ be a field of characteristic zero and $R$ a commutative $K$ -algebra with $1$ . Identify $a ∈ R$ with the multiplication operator $r \mapsto a r$ in $End_K(R)$ ; this is injective. For $P ∈ End_K(R)$ and $a ∈ R$ write $[P,a] = Pa - aP$ , again an element of $End_K(R)$ .

Operators of order at most $n$ Coutinho, Ch. 3 §1

Set $D^-1(R) = 0$ . For $n ≥ 0$ define, by induction on $n$ ,

D^n(R) = , P ∈ End_K(R) ;:; [P,a] ∈ D^n-1(R) for every a ∈ R ,.

(3.1)

An operator has order $n$ if it lies in $D^n(R)$ but not in $D^n-1(R)$ . An operator has finite order if it lies in $D^n(R)$ for some $n$ .

The ring of differential operators

The ring of differential operators of $R$ over $K$ is

D(R) = ⋃ _n ≥ 0 D^n(R) ⊂ eq End_K(R),

(3.2)

with the addition and composition it inherits from $End_K(R)$ . That this is closed under composition is Proposition (3.1.2) below and is the only point requiring proof.

A convention worth stating

Coutinho phrases the induction in terms of order exactly $n$ : $P$ has order zero if it commutes with every $a ∈ R$ , and order $n$ if it does not have smaller order but $[P,a]$ has order less than $n$ for all $a$ . The two formulations define the same sets $D^n(R)$ . The $≤$ form used here is more convenient because $D^n(R)$ is then visibly a subspace containing $D^n-1(R)$ ; the set of operators of order exactly $n$ is not a subspace, since a difference of two order- $n$ operators can have lower order.

Core Concepts

Order measures the failure of $R$ -linearity

An operator of order $0$ commutes with all multiplications, so it is $R$ -linear, and an $R$ -linear endomorphism of $R$ is determined by its value at $1$ : $D^0(R) = R$ . Order $1$ then says that the failure of $R$ -linearity, measured by $[P,a]$ , is itself $R$ -linear - which is the Leibniz rule in disguise. Order $n$ says the failure of the failure of ... is $R$ -linear, $n$ levels down. Order is a measure of how far an operator is from being multiplication by a function.

Why commutators and not something else

On $K[x]$ , differentiating a product introduces exactly one extra term: $∂(fg) = f∂(g) + f'g$ , so $[∂, f] = f'$ . The commutator with $f$ strips one derivative and leaves a function. Nothing about coordinates is used in that observation, only that multiplication by $f$ and the operator $∂$ nearly commute. The inductive definition promotes this to the defining property.

Two structural facts, and what they buy

Composition adds order, $D^n D^m ⊂ eq D^n+m$ , which is what makes $D(R)$ a ring and a filtered one. Commutators lose an order, $[D^n, D^m] ⊂ eq D^n+m-1$ , which is what makes the associated graded ring $gr,D(R)$ commutative. The second fact is the reason a non-commutative ring like $D(R)$ can be studied with commutative algebra at all: pass to symbols, work in a commutative ring, transfer conclusions back. Every dimension theory in this collection depends on it.

Finite order is a restriction

It is easy to forget that $D(R)$ is usually much smaller than $End_K(R)$ . For $R = K[x]$ the algebra endomorphism sending $x$ to $2x$ has infinite order, as the worked example verifies. Only the finite-order part is captured, and that is the point: finite order is the algebraic shadow of locality.

Construction and Proof

Each $D^n(R)$ is an $R$ -bimodule

$D^n(R)$ is a $K$ -subspace of $End_K(R)$ , it contains $D^n-1(R)$ , and it is stable under left and right multiplication by elements of $R$ .

Proof

Linearity of $P ↦ [P,a]$ gives the subspace claim by induction. For the inclusion, if $P ∈ D^n-1(R)$ then $[P,a] ∈ D^n-2(R) ⊂ eq D^n-1(R)$ by induction, so $P ∈ D^n(R)$ . For the bimodule claim, note that $b ∈ R$ commutes with $a ∈ R$ , so $[bP, a] = b[P,a]$ and $[Pb, a] = [P,a]b$ ; induction on $n$ finishes it.

Composition adds orderCoutinho (3.1.2)

If $P ∈ D^n(R)$ and $Q ∈ D^m(R)$ then $PQ ∈ D^n+m(R)$ . In particular $D(R)$ is a subring of $End_K(R)$ .

Proof

Induct on $n + m$ . If $n + m = 0$ both operators are multiplications, and so is their composite. Assume the statement for all pairs with smaller total. For $a ∈ R$ , the derivation property of $ad_a$ on the associative ring $End_K(R)$ gives

[PQ, a] = P[Q,a] + [P,a]Q.

Now $[Q,a] ∈ D^m-1(R)$ and $[P,a] ∈ D^n-1(R)$ , so by the inductive hypothesis both $P[Q,a]$ and $[P,a]Q$ lie in $D^n+m-1(R)$ . Hence $[PQ,a] ∈ D^n+m-1(R)$ for every $a$ , which is precisely the statement $PQ ∈ D^n+m(R)$ .

Commutators drop order

If $P ∈ D^n(R)$ and $Q ∈ D^m(R)$ with $n, m ≥ 0$ , then $[P,Q] ∈ D^n+m-1(R)$ .

Proof

Induct on $n+m$ . If $n = m = 0$ then $P$ and $Q$ are multiplications by elements of the commutative ring $R$ , so $[P,Q] = 0 ∈ D^-1(R)$ . For the step, use the Jacobi identity in $End_K(R)$ :

[[P,Q],a ] = [[P,a],Q ] + [P,[Q,a] ].

The first term pairs $[P,a] ∈ D^n-1(R)$ with $Q ∈ D^m(R)$ , the second pairs $P ∈ D^n(R)$ with $[Q,a] ∈ D^m-1(R)$ ; by induction both lie in $D^n+m-2(R)$ . Hence $[[P,Q],a] ∈ D^n+m-2(R)$ for all $a$ , so $[P,Q] ∈ D^n+m-1(R)$ .

The symbol algebra is commutative

The associated graded ring $gr,D(R) = ⨁ _n ≥ 0 D^n(R)/D^n-1(R)$ is a commutative $R$ -algebra, because the class of $[P,Q]$ in degree $n+m$ is zero by the previous proposition. See the order filtration and symbols for the case $R = K[x_1, … ,x_n]$ , where the graded ring is a polynomial ring in $2n$ variables.

Key Equations

Write $ad_a(P) = [P,a]$ . Since $R$ is commutative, multiplication operators commute, and consequently the operators $ad_a$ commute with one another on $End_K(R)$ :

ad_a ad_b = ad_b ad_a (a, b ∈ R).

(3.3)

Unwinding (3.1) gives a non-inductive description, symmetric in its arguments by (3.3):

D^n(R) = , P : ad_a_0ad_a_1 ⋅ s ad_a_n(P) = 0 for all a_0, … ,a_n ∈ R ,.

(3.4)

The first two layers are known explicitly, the second by the lemma on derivations:

D^0(R) = R, D^1(R) = R ⊕ Der_K(R).

(3.5)

The two structural rules are

D^n(R),D^m(R) ⊂ eq D^n+m(R), [D^n(R), D^m(R)] ⊂ eq D^n+m-1(R).

(3.6)

Grothendieck's equivalent formulation replaces the induction by a single $Hom$ . Let $μ : R ⊗ _K R → R$ be the multiplication map, $I = μ$ , and $P^n_R/K = (R ⊗ _K R)/I^n+1$ the module of principal parts, regarded as an $R$ -module through the left factor. Then

D^n(R) ≅ Hom_R!(P^n_R/K,, R).

(3.7)

For $n = 1$ this specialises to $D^1(R) ≅ Hom_R(R ⊕ Ω_R/K, R) = R ⊕ Der_K(R)$ , recovering (3.5).

Variable Definitions

$K$: the ground field, of characteristic zero
$R$: a commutative $K$ -algebra with identity
$End_K(R)$: all $K$ -linear maps $R \to R$ , a ring under composition
$a$: an element of $R$ , and also the multiplication operator $r \mapsto a r$
$[P,a]$: the commutator $Pa - aP$ inside $End_K(R)$
$ad_a$: the map $P ↦ [P,a]$ on $End_K(R)$
$D^n(R)$: the $K$ -space of differential operators of order at most $n$
$D(R)$: the ring of differential operators, the union of the $D^n(R)$
$Der_K(R)$: the $R$ -module of $K$ -derivations of $R$
$P^n_R/K$: the module of principal parts of order $n$

Properties and Behaviour

Differential operators localise

Let $R$ be a finitely generated commutative $K$ -algebra and $S ⊂ eq R$ a multiplicative set. Then for every $n$

D^n(S^-1R) ≅ S^-1D^n(R), hence D(S^-1R) ≅ S^-1D(R).

The reason is (3.7): $R$ is Noetherian, so $P^n_R/K$ is a finitely presented $R$ -module, and $Hom$ out of a finitely presented module commutes with localisation. Concretely, an operator on $R$ extends uniquely to $S^-1R$ by the quotient rule, and clearing denominators brings any operator on $S^-1R$ back into $D(R)$ after multiplication by an element of $S$ .

Products and disjoint unions

$D(R_1 × R_2) ≅ D(R_1) × D(R_2)$ . Differential operators cannot mix the two factors: an operator of finite order commutes with the idempotents up to lower order, and the argument in the examples section shows it must preserve each factor. Geometrically, differential operators are local and cannot move mass between connected components.

Domains

If $R$ is a domain with fraction field $Q$ , then restriction gives an isomorphism

D(R) ≅ , P ∈ D(Q) ;:; P(R) ⊂ eq R ,.

Every operator on $R$ extends uniquely to $Q$ by the previous proposition, and the order conditions are inherited by restriction. This is the practical way to compute $D(R)$ for a singular affine curve: work with operators with rational coefficients and impose the condition that they preserve $R$ . Coutinho sets this out as Exercise 3.7; the version stated there identifies the set as a subring, and the sharper statement is that it is $D(R)$ .

What can go wrong

None of the good properties of the Weyl algebra are automatic for a general $R$ . $D(R)$ need not be Noetherian, need not be finitely generated as a $K$ -algebra, and need not be a domain. It also need not be generated by $R$ and $Der_K(R)$ . All four failures occur already for coordinate rings of singular affine curves; see differential operators on an affine variety.

Examples and Special Cases

$R = K$

$End_K(K) = K$ and every operator is multiplication, so $D(K) = K$ with $D^n = K$ for all $n$ . The filtration is constant.

$R = K[x_1, … ,x_n]$

$D(R) = A_n(K)$ , the $n$ -th Weyl algebra, and $D^m(R)$ is the span of the $f_α ∂^α$ with $|α| ≤ m$ . This is the theorem of Ch. 3 §2 and it needs characteristic zero.

Laurent polynomials

$D(K[x,x^-1]) = K[x,x^-1]! ⟨ ∂ ⟩$ , the localisation of $A_1$ at the powers of $x$ , by the localisation proposition. This ring is again simple and Noetherian; it is the ring of operators on the punctured line.

$R = K × K$ : order does not stabilise upwards

Let $e = (1,0)$ . Suppose $P ∈ D^n(R)$ and, inductively, that $D^n-1(R) = R$ . Then $[P,e]$ is multiplication by some $w$ . Evaluating at $1$ gives $w = P(e) - eP(1)$ ; evaluating at $e$ gives $we = P(e) - eP(e) = (1-e)P(e)$ . Combining, $(1-e)P(e) = -e,P(1-e)$ , and since the left side lies in $(1-e)R$ and the right side in $e R$ , both vanish. Hence $P(e) ∈ eR$ and $P(1-e) ∈ (1-e)R$ , so $P$ preserves each factor and acts on each by a scalar: $P ∈ R$ . Therefore $D(K × K) = K × K$ , strictly smaller than the four-dimensional $End_K(R)$ .

$R = K[x]/(x^2)$ : order does not stabilise at all

Here $D^0 = R$ has dimension $2$ and $D^1 = R ⊕ Der_K(R)$ has dimension $3$ . The map $P$ with $P(1) = 0$ , $P(x) = 1$ satisfies $[P,x](1) = 1$ and $[P,x](x) = -x$ , which is the element $1 - 2, x∂$ of $D^1$ ; so $P ∈ D^2$ , and $P ∉ D^1$ . Since $End_K(R)$ has dimension $4$ , we get $D(R) = End_K(R)$ . For a finite-dimensional algebra the filtration exhausts everything, and the notion of order carries no geometric content.

Worked Example

Orders in $D(K[x])$ , computed from the definition alone

Step 1 - order zero and order one
$D^0(K[x]) = K[x]$ , since an operator commuting with multiplication by every polynomial is $K[x]$ -linear and hence multiplication by its value at $1$ . For $∂ = d/dx$ and $f ∈ K[x]$ , the product rule gives $∂ f = f∂ + f'$ , so $[∂, f] = f' ∈ D^0$ . Thus $∂ ∈ D^1$ . It is not in $D^0$ because $[∂, x] = 1 ≠ 0$ . So $∂$ has order exactly $1$ .
Step 2 - the order of $∂^2$
Apply $∂^2$ to a product: $(fg)'' = f''g + 2f'g' + fg''$ . As operators this reads $∂^2 f = f∂^2 + 2f'∂ + f''$ , so

$[∂^2, f] = 2f'∂ + f''.$

The right-hand side lies in $D^1$ by Step 1, so $∂^2 ∈ D^2$ . It is not in $D^1$ : taking $f = x$ gives $[∂^2, x] = 2∂$ , and $[2∂, x] = 2 ≠ 0$ , so the double commutator does not vanish. Hence $∂^2$ has order exactly $2$ , as (3.4) predicts.
Step 3 - a mixed operator
Let $θ = x∂$ , the Euler operator. Then $[θ, f] = x[∂, f] = xf'$ , an element of $D^0$ , so $θ$ has order $1$ ; it is not order $0$ because $[θ, x] = x ≠ 0$ . The operator $x^2∂^2 + x∂$ has order $2$ by the composition rule, and its double commutator with $x$ is $[[x^2∂^2 + x∂, x], x] = [2x^2∂ + x, x] = 2x^2 ≠ 0$ , confirming the order is exactly $2$ and not less.
Step 4 - an endomorphism of infinite order
Let $T$ be the $K$ -algebra endomorphism of $K[x]$ with $T(x) = 2x$ , so $T(x^k) = 2^k x^k$ . Set $U_m = ad_x^,m(T)$ . Then $U_0 = T$ and

$U_m+1(x^k) = U_m(x^k+1) - x,U_m(x^k).$

Claim: $U_m(x^k) = 2^k x^k+m$ for every $m ≥ 0$ . It holds for $m = 0$ ; and if it holds for $m$ , then $U_m+1(x^k) = 2^k+1x^k+1+m - 2^kx^k+m+1 = 2^kx^k+m+1$ , which is the claim for $m+1$ .
Step 5 - conclude
Since $U_m(1) = x^m ≠ 0$ for every $m$ , no iterated commutator of $T$ with $x$ vanishes. By (3.4), $T ∉ D^n(K[x])$ for any $n$ . The check at $m = 3$ , say, is $U_3(1) = x^3$ and $U_3(x) = 2x^4$ , both non-zero.

Result

$∂$ has order $1$ , $∂^2$ order $2$ , $x∂$ order $1$ , $x^2∂^2 + x∂$ order $2$ ; and the dilation $x ↦ 2x$ , although a perfectly good $K$ -algebra automorphism of $K[x]$ , has infinite order and is therefore not a differential operator. So $D(K[x])$ is a proper subring of $End_K(K[x])$ , and by the theorem of Ch. 3 §2 it is exactly the Weyl algebra $A_1(K)$ .

Applications and Industry Use

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

A uniform definition across geometry. The same induction defines $D_X$ on any scheme by sheafifying, which is what makes D-module theory available on varieties rather than just on affine space.
Singular spaces. For quotient singularities, toric varieties and cones, $D(R)$ computed from this definition is the right object; there is no coordinate description to fall back on.
Invariant differential operators. If a group $G$ acts on $R$ , the invariants $D(R)^G$ and the operators on $R^G$ are both defined by this induction, and comparing them is the Levasseur-Stafford circle of results in representation theory.
Rings of differential operators as test objects. Because $D(R)$ can fail to be Noetherian or finitely generated, it supplies natural counterexamples in non-commutative ring theory; the Bernstein-Gelfand-Gelfand cubic cone is the standard one.
Deformation quantisation. The filtration with commutative associated graded ring realises $D(R)$ as a deformation of functions on the cotangent space, with the Poisson bracket induced by $[, ⋅ ,, ⋅ ,]$ ; see the Poisson bracket on the symbol algebra.

Design Considerations

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

Inductive definition or principal parts

Definitions (3.1) and (3.7) agree, and each is better for a different job. The inductive definition is what one uses to prove things about operators, because every proof is an induction on order and each step is a commutator identity. The principal parts definition is what one uses to construct things: it is manifestly functorial, it sheafifies without effort, and it exhibits $D^n(R)$ as a $Hom$ , hence as something computable by linear algebra over $R$ once $P^n$ is presented.

Which filtration to carry

The order filtration $D^n$ is intrinsic: it exists for every $R$ and needs no coordinates. The Bernstein filtration, which gives $x$ and $∂$ equal weight, is available only for the Weyl algebra and depends on the choice of coordinates, but its pieces are finite dimensional over $K$ , which the order filtration's are not. For general $R$ there is no analogue of the Bernstein filtration, which is one reason the elementary proof of Bernstein's inequality does not transport to singular varieties.

Left modules or right modules

$R$ is a left $D(R)$ -module by construction. The definition is symmetric enough that right modules are equally natural, and on a smooth variety the two categories are equivalent via twisting by the canonical module. On a singular $R$ that equivalence is unavailable, so the choice of side has to be made explicitly and kept.

Standards and Codes

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

ISO 80000-2 governs the symbols $∂$ , $dim$ , $deg$ and the set-theoretic notation used above.
ISO/IEC 40314 (MathML 3.0) is the encoding of every formula on this page.
Notation for the operator ring is not standardised. $D(R)$ , $D(R)$ , $Diff(R)$ and $D_R/K$ all appear; the superscript convention $D^n$ for order at most $n$ is near-universal, but some authors write $D_n$ or $F_nD$ for the same thing. This collection uses $D^n(R)$ , and $F_n$ when the same filtration is regarded abstractly.
Grothendieck's definition in EGA IV §16 is the reference formulation for algebraic geometry, and is the one implemented in schemes-oriented software.

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.

Characteristic zero. The definition (3.1) makes sense over any commutative base, but its content changes. In characteristic $p$ , $D(K[x])$ contains the divided-power operators $∂^[k] : x^m ↦ mkx^m-k$ for all $k$ , and these are not polynomials in $∂$ once $k ≥ p$ . So $D(K[x])$ is strictly bigger than the Weyl algebra, and the order filtration has infinite-dimensional pieces even over $K[x]$ .
Base ring. Everything above works with $K$ replaced by a commutative base ring $k$ and $R$ a commutative $k$ -algebra, giving $D_R/k$ . The relative version is what one needs for families and for the parameter $s$ in the Bernstein-Sato construction.
Reduced versus non-reduced. For finite-dimensional algebras the filtration exhausts $End_K(R)$ , so order carries no information. Interesting behaviour begins when $R$ has positive Krull dimension.
Storage. An operator on $R = S/J$ is normally stored as a representative in the Weyl algebra $A_n$ over $S$ , together with $J$ ; the representative is not unique, and comparison requires normal-form computation modulo the idealiser of $J A_n$ .

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 $D(R)$ for $R = K[x_1, … ,x_n]/J$ is harder than computing $Der_K(R)$ , and is not always a finite problem, since $D(R)$ need not be finitely generated. What can be computed is each filtration piece.

Compute the ideal $J$ and a presentation of $P^n_R/K$ from generators and relations, or equivalently set up the linear conditions defining $D^n$ .
Solve $Hom_R(P^n_R/K, R)$ as a syzygy computation; this returns $D^n(R)$ as an $R$ -module with explicit generators.
For $R$ a domain, use the fraction-field description instead: parametrise operators of order at most $n$ with coefficients in $Q$ and impose $P(g_i) ∈ R$ for algebra generators $g_i$ of $R$ , together with the conditions coming from $P$ applied to products.
Increase $n$ and look for stabilisation of the generating set. Stabilisation is not guaranteed, and its failure is exactly the statement that $D(R)$ is not finitely generated.

Macaulay2 provides Dmodules for the smooth polynomial case and Hom computations for principal parts; Singular's dmod.lib and dmodapp.lib cover Weyl-algebra computations. There is no general-purpose algorithm returning $D(R)$ for arbitrary singular $R$ , and there cannot be one that always terminates with a finite presentation.

A useful sanity check while implementing: $D^0$ must come back as $R$ itself, and $D^1/D^0$ must come back as $Der_K(R)$ . If either fails, the linear conditions have been set up wrongly, most often by forgetting that the commutator is taken with all of $R$ and not only with the algebra generators.

Limits of Validity

The definition is completely general; the useful theorems about it are not.

$D(R)$ need not be generated by $R$ and $Der_K(R)$ . This holds for regular $R$ in characteristic zero and fails on singular rings. Assuming it is the single most common error in this area.
$D(R)$ need not be Noetherian or finitely generated. For the cone over a plane cubic, Bernstein, Gelfand and Gelfand showed $D(R)$ is neither.
Commutativity of $R$ is used. The proof that $ad_a$ and $ad_b$ commute, and hence the symmetry in (3.4), uses $[a,b] = 0$ . For non-commutative $R$ one must decide first what a differential operator should mean.
Finite dimensionality of the filtration pieces fails. $D^n(R)$ is a finitely generated $R$ -module in good cases but is essentially never finite dimensional over $K$ . Any counting argument in the style of Bernstein's inequality must therefore use a different filtration.

Failure Modes and Common Mistakes

Checking the commutator only against algebra generators

Definition (3.1) quantifies over all $a ∈ R$ . It is true that for $R = K[x_1, … ,x_n]$ it suffices to test $a = x_1, … ,x_n$ , but that is a small lemma resting on the identity $[P, ab] = a[P,b] + [P,a]b$ and on $D^n$ being an $R$ -bimodule; it is not part of the definition. On a quotient ring, testing only lifted generators can give a wrong answer if the relations are not respected.

Assuming $D(R) = End_K(R)$ , or that it is much smaller

Both extremes occur. For $R$ finite dimensional over $K$ the two coincide. For $R = K × K$ the operator ring collapses to $R$ . For $R = K[x]$ it is the Weyl algebra, strictly between $R$ and $End_K(R)$ . Nothing can be inferred about the size of $D(R)$ without looking at $R$ .

Reading "order $n$ " as a vector space condition

The set of operators of order exactly $n$ is not closed under addition: $∂^2$ and $-∂^2 + ∂$ both have order $2$ and their sum has order $1$ . Only $D^n(R)$ , the operators of order at most $n$ , is a subspace. Statements about symbols are statements about the quotients $D^n/D^n-1$ , not about order- $n$ operators as a set.

Carrying the characteristic-zero picture into characteristic $p$

In characteristic $p$ the operator $x^m ↦ mpx^m-p$ is a differential operator of order $p$ on $K[x]$ that is not in the subring generated by $x$ and $∂$ , because $∂^p = 0$ on $K[x]$ in that setting. The definition does not change; the answer does. See the Weyl algebra in positive characteristic.

Confusing the order filtration with the Bernstein filtration

$D^n(R)$ is defined for every $R$ and its pieces are $R$ -modules of infinite $K$ -dimension. The Bernstein filtration exists only for $A_n$ , gives $x_i$ and $∂_i$ equal weight, and has finite-dimensional pieces. Arguments that count $_K$ require the latter; arguments that need functoriality require the former.

Historical Notes

The idea of characterising differential operators by iterated commutators appears in the 1950s. Grothendieck gave the definitive algebraic formulation in EGA IV (1967), §16, defining $D^n$ as $Hom_R(P^n_R/K, R)$ and proving the composition and commutator rules in that setting. The equivalence with the inductive form is immediate once the ideal $I ⊂ eq R ⊗ _K R$ is introduced, and the inductive form has been the working definition in the D-module literature ever since.

The theorem that $D(K[x_1, … ,x_n]) = A_n(K)$ in characteristic zero, which is what ties the intrinsic definition back to the concrete one, is proved on its own page and is the whole point of Coutinho's Chapter 3.

That $D(R)$ can be badly behaved was settled quickly. In 1972 Bernstein, Gelfand and Gelfand computed $D(R)$ for the cone over a plane cubic curve and found a ring that is neither Noetherian nor finitely generated. In the opposite direction, the good behaviour for regular $R$ - generation by $R$ and $Der_K(R)$ , Noetherianity, simplicity in the irreducible case - is treated in McConnell and Robson, Ch. 15. The pattern that emerged, that regularity is exactly the dividing line, has organised the subject since.

Comparison

Three ways to say what a differential operator is.
	Inductive (commutators)	Grothendieck (principal parts)	Coordinates
Applies to	any commutative $K$ -algebra	any commutative $K$ -algebra	polynomial and regular rings only
Form	$[P,a] ∈ D^n-1$ for all $a$	$Hom_R(P^n, R)$	$∑ _\|α\| ≤ n f_α ∂^α$
Best for	proofs by induction on order	functoriality and sheafification	explicit computation
Sheafifies	yes, after localising	yes, immediately	only with a chosen atlas
Needs char $0$	no	no	yes, to be complete
Filtration pieces	$R$ -modules	$R$ -modules	free $R$ -modules

Key Takeaways

Key points

$D^-1(R) = 0$ and $P ∈ D^n(R)$ means $[P,a] ∈ D^n-1(R)$ for every $a ∈ R$ ; $D(R)$ is the union.
Equivalently, $P$ has order at most $n$ when every iterated commutator with $n+1$ elements of $R$ vanishes.
$D^0(R) = R$ and $D^1(R) = R ⊕ Der_K(R)$ .
Composition adds orders and commutators drop the total order by one, so $D(R)$ is a filtered ring with commutative associated graded ring.
Grothendieck's equivalent definition is $D^n(R) = Hom_R(P^n_R/K, R)$ , which sheafifies at once.
Finite order is a real restriction: the dilation $x ↦ 2x$ on $K[x]$ has infinite order.
Nothing about Noetherianity, finite generation or generation by derivations follows from the definition; those need regularity of $R$ .

FAQs

Why start the induction at

D^-1(R) = 0

It makes the single clause (3.1) cover $n = 0$ as well: $[P,a] ∈ 0$ for all $a$ says exactly that $P$ is $R$ -linear, so $D^0(R) = R$ . Without the convention one has to state the base case separately, as Coutinho does.

Is it enough to test the commutator against a set of algebra generators of

R

For $R = K[x_1, … ,x_n]$ yes, and that is how computations are done. The reason is that $D^n$ is an $R$ -bimodule and $[P, ab] = a[P,b] + [P,a]b$ , so the condition propagates from generators to products. For a general $R$ presented as $S/J$ the reduction is not automatic and the relations must be taken into account.

How does this recover the usual definition on

K[x_1, … ,x_n]

It gives $D^m = ∑ _|α| ≤ m f_α ∂^α$ and $D(K[X]) = A_n(K)$ , but only in characteristic zero, and the proof is not formal: it needs two lemmas, one saying an operator commuting with all $x_i$ is a polynomial, the other a discrete Poincaré lemma. See the Weyl algebra as a ring of differential operators.

D(R)

always bigger than the ring generated by

R

and its derivations?

It always contains it, and the containment is an equality exactly when $R$ is regular (in characteristic zero, for finitely generated reduced algebras). For singular $R$ the containment is usually strict; the cusp $K[t^2,t^3]$ is the smallest standard example.

Does the definition depend on

R

being commutative?

Yes, in two places. The identification $D^0(R) = R$ uses that $R$ -linear endomorphisms of $R$ are multiplications, and the symmetry of (3.4) uses that the operators $ad_a$ commute. Non-commutative analogues exist but require choices.

Why is it important that commutators drop the order?

Because it makes $gr,D(R)$ commutative. Every dimension, multiplicity and characteristic variety argument in D-module theory is carried out in that commutative graded ring and then transferred back. Without the drop there would be no commutative shadow to work in.

Can

D^n(R)

be infinite dimensional over

K

Almost always. Already $D^0(K[x]) = K[x]$ is infinite dimensional. Finite dimensionality of filtration pieces is a feature of the Bernstein filtration on the Weyl algebra, not of the order filtration.

Does the ring of differential operators determine

R

Not in general, and this is a delicate question. Non-isomorphic rings can have isomorphic rings of differential operators - for instance certain non-isomorphic affine curves. Recovering $R$ requires remembering more structure, such as the filtration together with the action on $R$ .

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 3 §1, with Lemma (3.1.1) and Proposition (3.1.2); Exercises 3.2, 3.3, 3.7.
A. Grothendieck, Éléments de géométrie algébrique IV, quatrième partie, Publications Mathématiques de l'IHÉS 32 (1967), §16 - the definition of $D^n$ by principal parts.
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 commutative rings.
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, North-Holland, 1979 - Ch. 1, for the filtration and the graded ring.
I. N. Bernstein, I. M. Gelfand and S. I. Gelfand, Differential operators on a cubic cone, Russian Mathematical Surveys 27 (1972), 169-174.
S. P. Smith and J. T. Stafford, Differential operators on an affine curve, Proceedings of the London Mathematical Society (3) 56 (1988), 229-259.
R. Hotta, K. Takeuchi and T. Tanisaki, D-modules, Perverse Sheaves, and Representation Theory, Progress in Mathematics 236, Birkhäuser, 2008 - Ch. 1, for the sheaf-theoretic version.
ISO/IEC 40314:2016, Information technology - Mathematical Markup Language (MathML) Version 3.0, International Organization for Standardization.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Prove that for $R = K[x_1, … ,x_n]$ it suffices to test the commutator condition against $x_1, … ,x_n$ , and identify where the bimodule property is used.
Verify directly from (3.1) that $x^2∂^2$ has order $2$ on $K[x]$ by computing two iterated commutators.
Deduce (3.5) from (3.7) by computing $P^1_R/K$ explicitly.
Show that $D(R_1 × R_2) ≅ D(R_1) × D(R_2)$ in general, following the argument given for $K \times K$ .
Find an operator of order $3$ on $K[t^2,t^3]$ that is not a polynomial in the derivations of that ring, and verify its order.
Show that in characteristic $p$ the divided-power operator $x^m ↦ mpx^m-p$ has order $p$ and is not in the subring generated by $x$ and $∂$ .
Prove that $D(R)$ equals $End_K(R)$ whenever $R$ is finite dimensional over $K$ .

Rings of Differential Operators: The Inductive Definition

Overview

Definition

Core Concepts

Order measures the failure of R-linearity

Why commutators and not something else

Two structural facts, and what they buy

Finite order is a restriction

Construction and Proof

Key Equations

Variable Definitions

Properties and Behaviour

Examples and Special Cases

Worked Example

Applications and Industry Use

Design Considerations

Inductive definition or principal parts

Which filtration to carry

Left modules or right modules

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

Order measures the failure of $R$ -linearity