The Degree of a Differential Operator

Q: Why is (0) set to - ∈ fty rather than 0 or left undefined?

So that (2.1) and (2.2) hold with no case distinctions. With (0) = - ∈ fty we get (0 ⋅ D) = - ∈ fty = - ∈ fty + (D) , and (D + (-D)) = - ∈ fty ≤ D, D . Setting (0) = 0 would break additivity, and leaving it undefined would force an exception into every statement.

Q: Does the degree depend on the order in which I write x's and ∂'s in a monomial?

No, for a single monomial. The reordering lemma shows (∂^β x^α) = |α|+|β| = (x^α∂^β) , and the same holds for any interleaving. What does depend on the presentation is the degree of a sum , because cancellation of top terms can only be detected in canonical form.

Q: Is the bound [D,D'] ≤ (D)+ (D')-2 ever an equality?

Yes, often. The worked example above achieves it: [x^2∂, ∂^2 x] = -3x^2∂^2 - 6x∂ has degree 4 = 3+3-2 . The general criterion is that equality holds precisely when the Poisson bracket σ(D), σ(D') of the leading symbols is non-zero, since that bracket is the symbol of the top-degree part of the commutator.

Q: Does additivity of degree require characteristic zero?

No. The proof uses only that K[x,ξ] has no zero divisors, which holds over any field. The Weyl algebra defined by generators and relations over Z_p is still a domain with additive degree. Characteristic zero is needed for simplicity and for the centre to be K , not for the degree calculus. The other presentation of A_n in positive characteristic, as operators on Z_p[x] , is a different and badly behaved ring.

Q: How does the degree relate to the Bernstein filtration and to dimension theory?

They are the same object seen twice. B_m is the set of operators of degree at most m ; (2.2) says B_iB_j ⊂ eq B_i+j and (2.3) says the associated graded ring is commutative. Once modules are filtered compatibly, _K of the pieces becomes a Hilbert function, its degree is d(M) and Bernstein's inequality bounds it below by n .

Q: Why can I not just define the degree as the smallest m with D ∈ B_m?

You can - that is an equivalent definition, and it is arguably the better one because it makes no reference to a presentation. The two agree exactly because the canonical monomials form a basis, so the span of those with |α|+|β| ≤ m contains D if and only if all higher coefficients of D vanish.

Q: Does every operator of degree k generate a left ideal that meets every degree below k?

No, and this is a useful check on intuition. The left ideal A_1∂ contains no non-zero operator of degree 0 , since a non-zero constant in A_1∂ would make ∂ invertible. What is true is the two-sided statement used for simplicity: a non-zero two-sided ideal contains elements of every sufficiently large degree and, by iterated bracketing, a non-zero constant.

Overview

The whole of the ideal theory of the Weyl algebra rests on a single numerical invariant: the degree of an operator. Because every $D ∈ A_n$ has a unique canonical form $D = ∑ c_αβ, x^α ∂^β$ , we may define $(D)$ to be the largest value of $|α| + |β|$ over the monomials that actually occur. The definition is unambiguous exactly because the canonical monomials are a basis.

What makes the invariant powerful is not the definition but two estimates. First, degree is additive on products: $(DD') = (D) + (D')$ , exactly as for polynomials over a field. Second, and this is the genuinely noncommutative statement, the commutator loses two degrees: $[D, D'] ≤ (D) + (D') - 2$ .

Together these say that $A_n$ is a polynomial ring to leading order. The failure of commutativity is real, but it is always hidden two degrees below the top. That single sentence is the engine behind almost everything in this chapter: additivity gives that P1 is a domain and that its only units are the non-zero scalars, while the commutator estimate lets you take an element of a two-sided ideal and manufacture a strictly smaller one, which is how simplicity is proved.

Later in the collection the same degree is rebranded as the Bernstein filtration, and the two estimates become the statements that $gr, A_n$ is a commutative polynomial ring in $2n$ variables and that the induced bracket on it is a Poisson bracket. This page proves the estimates from scratch; nothing beyond the canonical form is assumed.

Definition

Throughout, $K$ is a field of characteristic zero and $A_n = A_n(K)$ is the $n$ -th Weyl algebra, generated by $x_1, … ,x_n$ and $∂_1, … ,∂_n$ subject to $[∂_i, x_j] = δ_ij$ and $[x_i,x_j] = [∂_i,∂_j] = 0$ . We use multi-index notation: $x^α = x_1^α_1 ⋅ s x_n^α_n$ , $∂^β = ∂_1^β_1 ⋅ s∂_n^β_n$ , and $|α| = α_1 + … + α_n$ .

Degree of an operatorCoutinho, Ch. 2 §1

Let $D ∈ A_n$ have canonical form $D = ∑ _α,β c_αβ, x^α ∂^β$ with $c_αβ ∈ K$ , only finitely many non-zero. The degree of $D$ is

(D) = ,|α| + |β| ;:; c_αβ ≠ 0,,

with the convention $(0) = - ∈ fty$ . Equivalently, $(D) ≤ m$ if and only if $D$ lies in the span of the monomials $x^α∂^β$ with $|α|+|β| ≤ m$ .

Why the pair $(α,β)$ is one multi-index

A pair $(α,β) ∈ N^n × N^n$ is itself a multi-index in $N^2n$ , and $|α|+|β|$ is its length. So the degree treats the $2n$ generators $x_1, … ,x_n,∂_1, … ,∂_n$ perfectly symmetrically, each contributing $1$ . This is the choice that distinguishes the degree from the order of an operator, which gives the $x_i$ weight $0$ and the $∂_i$ weight $1$ .

A first computation

In $A_2$ , the operator $D = x_1^2∂_1 + x_1x_2^3∂_2^2$ is already in canonical form. Its first summand has $|α|+|β| = 2+1 = 3$ and its second has $4+2 = 6$ , so $(D) = 6$ . The convention $(0) = - ∈ fty$ is what makes the two estimates below hold without exceptions.

Core Concepts

Three ideas explain why the degree behaves as it does, and it is worth separating them before the proofs.

The canonical form is what makes the degree well defined

There is no such thing as "the degree of an expression". The operator $∂ x - x∂$ looks like a difference of two degree- $2$ terms, but it equals $1$ and has degree $0$ . The definition is only meaningful because the canonical monomials really are linearly independent; you must normalise before you read a degree off.

Reordering costs exactly two degrees

The single relation $∂_i x_i = x_i∂_i + 1$ replaces a degree- $2$ monomial by another degree- $2$ monomial plus a degree- $0$ correction. Every reordering in the Weyl algebra is built from repeated use of that relation, so every reordering produces a leading term of the same degree plus corrections of degree at least two lower. This is the whole content of the commutator estimate, and it is why the constant $2$ appears rather than $1$ .

To leading order the algebra is commutative

Assign to each $D$ of degree $m$ its leading symbol: replace $∂_i$ by a commuting variable $ξ_i$ and keep only the terms with $|α|+|β| = m$ . The result is a non-zero homogeneous polynomial $σ_m(D) ∈ K[x_1, … ,x_n,ξ_1, … ,ξ_n]$ . The two estimates combine into the single statement $σ_m+m'(DD') = σ_m(D),σ_m'(D')$ : taking symbols is multiplicative, and multiplicativity of symbols plus the fact that a polynomial ring is a domain is precisely additivity of degree.

Construction and Proof

Statement (2.1) is immediate: adding two canonical forms is adding coefficients, so no new monomial can appear, and if the two degrees differ the top monomials of the larger one survive untouched. The real work is (2.2) and (2.3), and they are proved together.

The commutator identities

In any ring, $[ab, c] = a[b,c] + [a,c]b$ and $[a, bc] = [a,b]c + b[a,c]$ . Both are one-line expansions: $a(bc - cb) + (ac - ca)b = abc - cab = [ab,c]$ .

Reordering estimateCoutinho (2.1.1), proof

For all multi-indices $α, β ∈ N^n$ , $[∂^β, x^α] ≤ |α| + |β| - 2$ . Consequently $(∂^β x^α) = |α| + |β|$ .

Proof of the reordering estimate

Induct on $|α| + |β|$ . If $α = 0$ or $β = 0$ the commutator is zero and there is nothing to prove. Otherwise pick $i$ with $β_i ≥ 1$ and write $∂^β = ∂_i,∂^β - e_i$ . The first commutator identity gives

[∂^β, x^α] = ∂_i,[∂^β - e_i, x^α] + [∂_i, x^α],∂^β - e_i.

For the second summand, $[∂_i, x^α] = ∂ x^α/∂ x_i$ is a polynomial of degree $|α| - 1$ (or zero), so that summand has degree at most $(|α| - 1) + (|β| - 1) = |α| + |β| - 2$ .

For the first summand, the inductive hypothesis bounds $[∂^β-e_i, x^α]$ by $|α| + |β| - 3$ , and left multiplication by $∂_i$ raises degree by at most $1$ - which needs only the trivial half of (2.2), that a product of a degree- $1$ and a degree- $d$ element lies in degree $≤ d+1$ , and that is clear from writing out the canonical form. So this summand has degree at most $|α| + |β| - 2$ as well.

Now $∂^β x^α = x^α∂^β + [∂^β, x^α]$ . The first term is a canonical monomial of degree exactly $|α| + |β|$ and the second has strictly smaller degree, so by the equality case of (2.1) the sum has degree exactly $|α| + |β|$ .

Degree of a product and of a commutatorCoutinho (2.1.1)

For all $D, D' ∈ A_n$ : $(DD') = (D) + (D')$ and $[D,D'] ≤ (D) + (D') - 2$ .

Proof

First take monomials $D = x^α∂^β$ and $D' = x^γ∂^δ$ . Substituting (2.4) for $∂^β x^γ$ ,

D D' = x^α,∂^β x^γ,∂^δ = x^α+γ∂^β+δ + x^α P ∂^δ, (P) ≤ |β| + |γ| - 2.

The correction term $x^α P ∂^δ$ has degree at most $|α| + |β| + |γ| + |δ| - 2 = (D) + (D') - 2$ , so the canonical monomial $x^α+γ∂^β+δ$ is the unique top-degree contribution and $(DD') = (D) + (D')$ .

For general $D, D^{'}$ , write $D = D_k + R$ and $D' = D'_l + R'$ , where $D_k$ collects the terms of top degree $k = (D)$ , $D'_l$ those of top degree $l = (D')$ , and $(R) < k$ , $(R') < l$ . Expanding, every product except $D_kD'_l$ has degree $< k + l$ . By the monomial case, the degree- $(k+l)$ part of $D_kD'_l$ is obtained by multiplying the symbols: it is the image of $σ_k(D)σ_l(D')$ under $ξ_i ↦ ∂_i$ . Since $K[x,ξ]$ is a polynomial ring over a field, hence a domain, that product is non-zero. Therefore $(DD') = k + l$ .

The same expansion applied to $D^{'} D$ produces the same top-degree part, because $σ_l(D')σ_k(D) = σ_k(D)σ_l(D')$ in the commutative ring $K[x,ξ]$ . Subtracting, the top-degree parts of $D D^{'}$ and $D^{'} D$ cancel, leaving

[D, D'] = (DD' - x-ξ top part) - (D'D - x-ξ top part),

a difference of two operators of degree at most $k + l - 2$ . Hence $[D,D'] ≤ (D) + (D') - 2$ .

Where the exponent $2$ comes from

Coutinho proves (2.2) and (2.3) by a single simultaneous induction on $(D) + (D')$ . The route above splits the induction off into the reordering lemma and then finishes with the symbol map, which is the same argument reorganised so that the role of the polynomial ring $K[x,ξ]$ is visible. Either way the drop of exactly $2$ traces back to $[∂_i, x_i] = 1$ : two degree- $1$ generators produce a degree- $0$ commutator.

Key Equations

The three properties of the degree, in the form in which they are used:

(D + D') ≤ (D), (D'),

(2.1)

with equality whenever $(D) ≠ (D')$ .

(D D') = (D) + (D'),

(2.2)

,[D, D'] ≤ (D) + (D') - 2.

(2.3)

The computational heart of the proof is the commutation of a pure derivative past a pure monomial:

∂^β x^α = x^α ∂^β + P, (P) ≤ |α| + |β| - 2.

(2.4)

In the one-variable case (2.4) can be made completely explicit by the Leibniz rule; for $f ∈ K[x]$ ,

∂^k f = ∑ _j=0^k kj d^,j fdx^,j, ∂^,k-j,

(2.5)

in which the $j = 0$ term is $f∂^k$ and every later term has dropped $j$ in derivative order while dropping at least $j$ in polynomial degree - a loss of at least $2$ once $j ≥ 1$ .

Counting the canonical monomials of degree at most $m$ gives the dimension of the corresponding filtration piece:

_K D ∈ A_n : (D) ≤ m = 2n + m2n.

(2.6)

Variable Definitions

$K$: the ground field, of characteristic zero
$A_n$: the $n$ -th Weyl algebra over $K$
$x_i, ∂_i$: the generators of $A_n$ , satisfying $[∂_i, x_j] = δ_ij$
$α, β$: multi-indices in $N^n$ , with $|α| = α_1 + … + α_n$
$x^α ∂^β$: a canonical monomial; the set of all of them is a $K$ -basis of $A_n$
$(D)$: the degree of $D$ : the largest $|α| + |β|$ occurring in its canonical form
$e_i$: the multi-index with $1$ in position $i$ and $0$ elsewhere
$[D, D']$: the commutator $DD' - D'D$
$σ_m(D)$: the leading symbol: the degree- $m$ part of $D$ with $∂_i$ replaced by a commuting variable $ξ_i$
$B_m$: the space of operators of degree at most $m$ , the $m$ -th piece of the Bernstein filtration

Properties and Behaviour

The consequences below are all immediate from (2.2), and they are the reason the degree is introduced at this point in the theory rather than later.

The Weyl algebra is a domainCoutinho (2.1.2)

If $D, D' ≠ 0$ then $(D), (D') ≥ 0$ , so $(DD') = (D) + (D') ≥ 0 > - ∈ fty$ and $DD' ≠ 0$ . The details, including what fails in characteristic $p$ , are on the domain page.

The units of $A_n$ are the non-zero scalars

If $DD' = 1$ then $(D) + (D') = 0$ ; since neither degree is negative for a non-zero operator, both are $0$ , so $D ∈ K^ ×$ . In particular every non-constant operator generates a proper non-zero left ideal, so $A_n$ is very far from a division ring even though it has no proper non-zero two-sided ideals.

Lowering degree inside a two-sided ideal

Let $D ∈ A_n$ with $(D) = k > 0$ . Then at least one of the $2n$ commutators $[x_i, D]$ , $[∂_i, D]$ is non-zero, and every non-zero one among them has degree at most $k - 1$ . Indeed, if all of them vanished, $D$ would lie in the centre of $A_n$ , which is $K$ , forcing $k = 0$ . This is exactly the step that drives the proof of simplicity: from a minimal-degree element of a non-zero two-sided ideal you produce a smaller one.

The degree defines a filtration

Writing $B_m = D : (D) ≤ m$ , the estimates say $B_iB_j ⊂ eq B_i+j$ and $[B_i, B_j] ⊂ eq B_i+j-2$ , with $⋃ _m B_m = A_n$ and $_K B_m = 2n+m2n < ∈ fty$ . This is the Bernstein filtration, and the second inclusion is precisely the statement that the associated graded algebra $gr,A_n ≅ K[x_1, … ,x_n,ξ_1, … ,ξ_n]$ is commutative.

Worked Example

Product and commutator of two cubics in $A_1$

Step 1 - put both operators in canonical form
Take $D = x^2∂$ and $E = ∂^2 x$ . The first is already canonical, with $(D) = 3$ . The second is not. Using $[∂^2, x] = ∂[∂,x] + [∂,x]∂ = 2∂$ ,

$E = ∂^2 x = x∂^2 + 2∂,$

so $(E) = 3$ as well. Sanity check on a test function: $∂^2(xf) = xf'' + 2f'$ , which is exactly $(x∂^2 + 2∂)(f)$ .
Step 2 - compute the product $D E$
Only one reordering is needed, $∂ x = x∂ + 1$ :

$DE = x^2∂(x∂^2 + 2∂) = x^2(x∂ + 1)∂^2 + 2x^2∂^2 = x^3∂^3 + 3x^2∂^2.$

The top term is $x^3∂^3$ , of degree $6 = 3 + 3$ , confirming (2.2).
Step 3 - compute the product $E D$
Here two reorderings are needed. From $∂^2(x^2f) = x^2f'' + 4xf' + 2f$ we get $∂^2 x^2 = x^2∂^2 + 4x∂ + 2$ , and $∂ x^2 = x^2∂ + 2x$ . Hence

$ED = x∂^2 x^2∂ + 2∂ x^2 ∂ = x(x^2∂^2 + 4x∂ + 2)∂ + 2(x^2∂ + 2x)∂,$

$ED = x^3∂^3 + 4x^2∂^2 + 2x∂ + 2x^2∂^2 + 4x∂ = x^3∂^3 + 6x^2∂^2 + 6x∂.$

Again the degree is $6$ , and the top term $x^3∂^3$ is identical to the one in $D E$ , exactly as the proof predicts.
Step 4 - subtract and read the commutator degree
Subtracting the two canonical forms term by term:

$[D, E] = (x^3∂^3 + 3x^2∂^2) - (x^3∂^3 + 6x^2∂^2 + 6x∂) = -3x^2∂^2 - 6x∂.$

So $[D,E] = 4 = 3 + 3 - 2$ . The bound (2.3) is attained, not merely satisfied.
Step 5 - cross-check with the symbol calculus
The leading symbols are $σ_3(D) = x^2ξ$ and $σ_3(E) = xξ^2$ . The bracket induced on symbols by the commutator is the Poisson bracket $f,g = ∑ _i (∂ f/∂ ξ_i)(∂ g/∂ x_i) - (∂ f/∂ x_i)(∂ g/∂ ξ_i)$ . Here

$x^2ξ,; xξ^2 = x^2 ⋅ ξ^2 - 2xξ ⋅ 2xξ = -3x^2ξ^2,$

and the degree- $4$ part of $[D,E]$ computed in Step 4 is $-3x^2∂^2$ , whose symbol is $-3x^2ξ^2$ . The two agree, which is a genuine independent check on the arithmetic of Steps 2 and 3.

Result

$DE = x^3∂^3 + 3x^2∂^2$ and $ED = x^3∂^3 + 6x^2∂^2 + 6x∂$ , both of degree $6 = (D)+ (E)$ ; and $[D,E] = -3x^2∂^2 - 6x∂$ of degree $4 = (D)+ (E)-2$ . Both estimates hold, and both are sharp for this pair.

Applications and Industry Use

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

Within this collection the degree is used, in order of appearance:

to prove that P0 has no zero divisors and that its units are the scalars;
to prove simplicity, by descending induction on the degree of a minimal element of a two-sided ideal;
to show that $A_n$ has no non-zero finite-dimensional representations, where the trace argument is packaged by the degree bookkeeping;
to define the Bernstein filtration and hence Hilbert functions, dimension and multiplicity of modules;
to give the cheapest proof that P0 is Noetherian, by lifting the Hilbert basis theorem from $gr,A_n = K[x,ξ]$ .

Outside the collection, degree bounds are the standard complexity measure for algorithms over the Weyl algebra. In creative telescoping and in Gröbner basis computation for $D$ -ideals, the running time is controlled by the degree of the intermediate operators, and every practical implementation carries a degree bound as its main loop condition.

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 a degree is trivial once the canonical form is known; the cost is entirely in the normalisation. A dense operator of degree $m$ in $A_n$ has $2n+m2n$ coefficients, so representations are usually sparse dictionaries keyed by the pair $(α,β)$ .

Normalise. Rewrite the input as a linear combination of canonical monomials by repeatedly applying $∂_i x_j = x_j∂_i + δ_ij$ , always moving $x$ 's left. Termination follows because each application strictly decreases the number of inversions of the word.
Prune. Discard monomials whose coefficient has become zero. This step is essential: skipping it can only make the reported degree too large, and for exact rational coefficients the test is exact.
Read off. Take the maximum of $|α| + |β|$ over the surviving monomials.

In a computer algebra system the normalisation is done for you. Macaulay2's Dmodules package and Singular's dmod.lib both represent Weyl algebra elements in canonical form and expose the weight vector; setting all $2n$ weights to $1$ recovers the degree of this page, while setting the $x$ -weights to $0$ recovers the order. Sage's DifferentialWeylAlgebra uses the same normal form. Gröbner basis computations in $A_n$ are performed with respect to a chosen weight vector, so the choice made here is a genuine input to an algorithm and not merely a matter of exposition.

A practical note: because degree is additive, degree bounds propagate exactly through products but only weakly through sums and commutators. When bounding the degree of an expression built by repeated bracketing - as happens in the simplicity proof - the drop of $2$ per bracket is what guarantees termination, and an implementation that only records the weak bound $(D)+ (D')$ will fail to terminate.

Limits of Validity

Two hypotheses are doing work here, and one apparent hypothesis is not.

The canonical form must exist. The degree is defined via the basis $x^α∂^β$ . Everything on this page therefore depends on the proof that this set is a basis, which is where the infinite-dimensionality of $K[X]$ is used.
Additivity needs $K[x,ξ]$ to be a domain, i.e. $K$ must be a domain - a field is more than enough. But (2.2) does not need characteristic zero. In $A_1(Z_p)$ presented by generators and relations, degree is still additive and the algebra is still a domain; what fails in characteristic $p$ is simplicity, not the degree calculus. See the positive characteristic page.
Degree does not generalise to arbitrary rings of differential operators. For a general smooth affine variety $X$ there is no canonical way to give the coordinate functions weight $1$ , so $D(X)$ has an order filtration but no intrinsic Bernstein filtration. Statements about $A_n$ proved using the degree must be re-proved, not merely restated, in that setting.

What (2.3) does not say

The estimate is an inequality and is often strict by a wide margin. For instance $[x_1^k, x_2^k] = 0$ , so the left side is $- ∈ fty$ while the right side is $2k - 2$ . Equality holds exactly when the Poisson bracket of the two leading symbols is non-zero, which is a computable condition but not an automatic one.

Failure Modes and Common Mistakes

Reading a degree off a non-canonical expression

The commonest error. $∂ x - x∂$ presents as a difference of degree- $2$ terms but equals $1$ , of degree $0$ . Similarly $∂^2 x - x∂^2$ presents as degree $3$ but equals $2∂$ , of degree $1$ . A degree read before normalisation is only ever an upper bound. When an argument needs an exact degree - and the simplicity proof does - normalise first.

Confusing degree with order

The operator $x^100∂$ has degree $101$ and order $1$ . Statements such as "operators of degree $0$ are the constants" are true for degree and false for order, where the order-zero operators are all of $K[X]$ . If a source says "filtration" without qualification in this subject, check which weighting is meant before quoting a bound.

Expecting the commutator to drop by two in every filtration

The drop of $2$ in (2.3) is specific to the Bernstein weighting. With the order filtration the correct drop is $1$ , and with the mixed weightings used in Gröbner computations the drop can be $0$ , in which case the associated graded ring is not commutative. Do not import the constant $2$ into a different filtration.

Assuming a degree- $k$ operator has an inverse of degree $-k$

There is no such thing. Degrees are non-negative on non-zero operators, so (2.2) forces the only invertible elements to be the non-zero constants. In particular $x$ has no inverse in $A_1$ : to invert it you must pass to a localisation such as $A_1[x^-1]$ , which is a strictly larger ring. This is the point at which intuition from $K[x, x^-1]$ or from the ring $B_n(K)$ of operators with rational function coefficients breaks down.

Historical Notes

The degree in this form is the elementary shadow of a much older idea. The estimate $[D,D'] ≤ (D) + (D') - 2$ is the algebraic form of the fact, familiar from quantum mechanics since the 1920s, that the commutator of two operators corresponds to $i ℏ$ times the Poisson bracket of their classical symbols: one power of $ℏ$ is exactly the two degrees lost here. Hermann Weyl's 1928 book on group theory and quantum mechanics made the algebra generated by position and momentum a formal object, which is why it carries his name.

The systematic use of a total-degree filtration in which $x$ and $∂$ both have weight $1$ is due to Joseph Bernstein, in his 1971 and 1972 papers on the analytic continuation of complex powers. It is now universally called the Bernstein filtration, and its decisive advantage over the classical order filtration is (2.6): each piece is finite dimensional over $K$ , so Hilbert function arguments become available and lead to Bernstein's inequality.

That the two-sided ideal structure of $A_n$ can be settled by degree bookkeeping alone was known much earlier - it appears in Dirac's and Littlewood's work on the algebra of quantum observables in the 1930s, and in a fully algebraic form in Dixmier's 1968 study of $A_1$ , which classified the automorphisms and posed what is now the Dixmier conjecture.

Comparison

The Weyl algebra carries two standard size measures and they are not interchangeable. Both are used in this collection; the pages state which one is in force.

Degree (Bernstein weighting) versus order (order weighting) on $A_n$ .
	Degree $(D)$	Order $ord(D)$
Weight of $x_i$	$1$	$0$
Weight of $∂_i$	$1$	$1$
Value on $x^100∂$	$101$	$1$
Filtration pieces	finite dimensional over $K$	infinite dimensional over $K$ , finitely generated over $K[X]$
$_K$ of the $m$ -th piece	$2n+m2n$	infinite
Additive on products	yes	yes
Commutator drop	$2$	$1$
Associated graded ring	$K[x,ξ]$ , $2n$ variables	$K[x,ξ]$ , $2n$ variables
Generalises to $D(X)$	no (needs the grading of affine space)	yes

The commutator drop is the sharpest difference. Under the order filtration $[∂_i, x_i] = 1$ costs only one unit of order, so the estimate reads $ord[D,D'] ≤ ord(D) + ord(D') - 1$ . Proofs written for one filtration therefore do not transfer verbatim to the other; see the order filtration page.

Key Takeaways

Key takeaways

The degree of $D ∈ A_n$ is the largest $|α|+|β|$ occurring in its canonical form, with $(0) = - ∈ fty$ . It is only defined after normalisation.
Degree is additive on products: $(DD') = (D)+ (D')$ . This holds because leading symbols multiply in the polynomial ring $K[x,ξ]$ , which is a domain.
Commutators drop two degrees: $[D,D'] ≤ (D)+ (D')-2$ . The constant $2$ comes from $[∂_i,x_i] = 1$ , a degree- $0$ bracket of two degree- $1$ generators.
Immediate consequences: $A_n$ is a domain, and its only units are the non-zero elements of $K$ .
The commutator estimate is the tool that shrinks elements of a two-sided ideal, and hence proves simplicity.
Degree is not order. Degree gives $x_i$ and $∂_i$ equal weight $1$ ; order gives $x_i$ weight $0$ . Their filtration pieces, and their commutator drops, differ.
The sublevel sets $B_m$ form the Bernstein filtration, with $_K B_m = 2n+m2n$ and commutative associated graded ring.

FAQs

Why is

deg (0)

set to

- ∈ fty

rather than

0

or left undefined?

So that (2.1) and (2.2) hold with no case distinctions. With $(0) = - ∈ fty$ we get $(0 ⋅ D) = - ∈ fty = - ∈ fty + (D)$ , and $(D + (-D)) = - ∈ fty ≤ D, D$ . Setting $(0) = 0$ would break additivity, and leaving it undefined would force an exception into every statement.

Does the degree depend on the order in which I write

x

's and

∂

's in a monomial?

No, for a single monomial. The reordering lemma shows $(∂^β x^α) = |α|+|β| = (x^α∂^β)$ , and the same holds for any interleaving. What does depend on the presentation is the degree of a sum, because cancellation of top terms can only be detected in canonical form.

Is the bound

[D,D'] ≤ (D)+ (D')-2

ever an equality?

Yes, often. The worked example above achieves it: $[x^2∂, ∂^2 x] = -3x^2∂^2 - 6x∂$ has degree $4 = 3+3-2$ . The general criterion is that equality holds precisely when the Poisson bracket $σ(D), σ(D')$ of the leading symbols is non-zero, since that bracket is the symbol of the top-degree part of the commutator.

Does additivity of degree require characteristic zero?

No. The proof uses only that $K[x,ξ]$ has no zero divisors, which holds over any field. The Weyl algebra defined by generators and relations over $Z_p$ is still a domain with additive degree. Characteristic zero is needed for simplicity and for the centre to be $K$ , not for the degree calculus. The other presentation of $A_n$ in positive characteristic, as operators on $Z_p[x]$ , is a different and badly behaved ring.

How does the degree relate to the Bernstein filtration and to dimension theory?

They are the same object seen twice. $B_m$ is the set of operators of degree at most $m$ ; (2.2) says $B_iB_j ⊂ eq B_i+j$ and (2.3) says the associated graded ring is commutative. Once modules are filtered compatibly, $_K$ of the pieces becomes a Hilbert function, its degree is $d(M)$ and Bernstein's inequality bounds it below by $n$ .

Why can I not just define the degree as the smallest

m

with

D ∈ B_m

You can - that is an equivalent definition, and it is arguably the better one because it makes no reference to a presentation. The two agree exactly because the canonical monomials form a basis, so the span of those with $|α|+|β| ≤ m$ contains $D$ if and only if all higher coefficients of $D$ vanish.

Does every operator of degree

k

generate a left ideal that meets every degree below

k

No, and this is a useful check on intuition. The left ideal $A_1∂$ contains no non-zero operator of degree $0$ , since a non-zero constant in $A_1∂$ would make $∂$ invertible. What is true is the two-sided statement used for simplicity: a non-zero two-sided ideal contains elements of every sufficiently large degree and, by iterated bracketing, a non-zero constant.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Ch. 2 §1, Theorem (2.1.1) and Corollary (2.1.2).
S. C. Coutinho, A Primer of Algebraic D-modules, Ch. 1 §2, Proposition (1.2.1) - the canonical basis, on which the definition of degree depends.
I. N. Bernstein, Modules over a ring of differential operators. Study of the fundamental solutions of equations with constant coefficients, Functional Analysis and Its Applications 5 (1971), 89-101 - introduces the total-degree filtration used here.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - degree arguments for $A_1$ , units and automorphisms.
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, 1979 - Ch. 1, filtrations and symbol calculus on $A_n$ .
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, revised edition, American Mathematical Society, 2001 - Ch. 1 and Ch. 8, filtered and graded rings, the Weyl algebra as the model example.
S. P. Smith, Differential operators on commutative algebras, in Ring Theory (Antwerp 1985), Lecture Notes in Mathematics 1197, Springer, 1986 - the positive characteristic picture.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics - notation for multi-indices, factorials and binomial coefficients.
Macaulay2 Dmodules package documentation, and Singular dmod.lib - weight vectors and normal forms for elements of $A_n$ .

AI Suggested Questions

Show me the full simultaneous induction that Coutinho uses to prove additivity of degree and the commutator estimate at the same time.
Work out the canonical form and degree of $(x∂)^k$ in $A_1$ for small $k$ .
Give an example of two operators in $A_2$ for which the commutator estimate is strict rather than sharp, and explain why.
How does the degree filtration compare with the order filtration on the ring of differential operators of a smooth affine curve?
Prove that the centre of $A_n$ is $K$ using only the degree and the commutator estimate.
What degree bounds control the cost of a Gröbner basis computation in $A_n$ ?
Explain the relation between the two-degree drop in the commutator and the appearance of $ℏ$ in the semiclassical limit.