What Algebraic D-modules Are

Q: What does the D stand for?

The ring of differential operators, written D or D by long habit. A D-module is a module over it. There is no separate object called a 'D'; the notation is a compression of 'module over the ring of differential operators'.

Q: Why replace an equation by a module at all? The equation is more concrete.

Because the module has structure the equation lacks. Modules can be added, quotiented, filtered, and transported along maps of the underlying space; they have numerical invariants; and they sit in exact sequences that let a hard case be reduced to easier ones. An equation on a page supports none of those operations. The payoff is theorems that hold for whole classes of systems at once.

Q: Is every D-module attached to a differential equation?

Every cyclic module A_n/I is, in the sense that I is generated by operators and those operators are the system. Since A_n is Noetherian, I is finitely generated, so the system is finite. Modules with several generators correspond to systems for a vector of unknowns, presented by a matrix of operators. Remarkably, every holonomic module turns out to be cyclic , so in the holonomic world a single equation always suffices.

Q: What is the difference between algebraic and analytic D-modules?

The coefficients and the base. Algebraic D-modules use polynomial or regular coefficients on an algebraic variety; analytic D-modules use holomorphic coefficients on a complex manifold and are sheaves rather than modules over a single ring. The analytic theory is more powerful and much heavier, requiring sheaf theory and derived categories. On smooth affine varieties the algebraic theory needs neither, because a module over the global ring already carries all the information.

Q: Does the theory work over the real numbers?

The algebra does: A_n(R) is a simple Noetherian domain and all the dimension theory applies, since only characteristic zero is used. What does not transfer well is the analytic interpretation. Solution spaces over R can behave badly, and statements about monodromy and analytic continuation need C . It is common to compute over Q and interpret over C .

Q: Is a D-module the same as a system of PDEs?

Not quite - it is the system modulo the choice of presentation. Two different generating sets for the same left ideal give the same module, and that is deliberate: the module remembers the ideal of all operators annihilating the unknown. Conversely, a module does not remember which generators of the ideal you started from, so information about the specific way the system was written is discarded.

Q: Where does holonomicity come from, intuitively?

From over-determination. Each independent equation cuts down how fast the module can grow. Bernstein's inequality says growth can never drop below order m^n , so there is a hard floor on how over-determined a consistent system can be, and holonomic systems are those that reach it. In one variable, a single non-trivial ODE already does; in n variables, roughly n independent conditions are needed.

Q: Do I need algebraic geometry to start?

No. The first ten chapters of the elementary theory need only linear algebra, basic ring and module theory, and comfort with filtrations. Algebraic geometry enters only when characteristic varieties are introduced, and symplectic geometry only for the involutivity theorem. See the roadmap for what depends on what.

Overview

The name is more forbidding than the object. In D-module, the letter $D$ stands for a ring of differential operators, and a D-module is simply a module over such a ring. Everything interesting is in the choice: which ring of operators, which modules, and what one does with them. In this collection the ring is the Weyl algebra $A_n$ , the algebra of differential operators with polynomial coefficients in $n$ variables, and the modules are the finitely generated left $A_n$ -modules.

The construction that gives the subject its purpose is a translation. A linear differential operator $P$ with polynomial coefficients generates a left ideal $A_n P$ , and the quotient $M = A_n/A_n P$ is a module. Nothing has been solved, but something has been gained: a solution of $Pu = 0$ with values in any $A_n$ -module $N$ is the same thing as an $A_n$ -homomorphism $M \to N$ . The equation has become an object, its solutions have become maps out of that object, and the choice of where the solutions live has become the choice of a second object. This is the module attached to a differential equation.

Once the equation is a module, the whole apparatus of ring theory applies to it. The module has a dimension $d(M)$ and a multiplicity $e(M)$ ; it has a characteristic variety that records where the equation degenerates; it can be pulled back and pushed forward along polynomial maps; and it can be assembled from, or decomposed into, other modules along exact sequences. None of these are properties of a single equation written on a page. They are properties of the module, and they turn out to control the analysis.

The word algebraic fixes the setting. Coefficients are polynomials, or regular functions on an algebraic variety, rather than convergent power series; the base is affine space rather than a complex manifold; and because affine space has no interesting topology in the algebraic sense, no sheaves and no derived categories are needed. That restriction is what makes the theory teachable with only linear algebra and basic ring theory, and it is the setting of Coutinho's Primer which this collection follows.

Definition

Fix a field $K$ of characteristic zero. Two definitions do most of the work.

The Weyl algebraCoutinho, Ch. 1 §1

The $n$ -th Weyl algebra $A_n = A_n(K)$ is the subalgebra of $End_K(K[x_1, … ,x_n])$ generated by the multiplication operators $x_1, … ,x_n$ and the partial derivatives $∂_1 = ∂/∂ x_1, … , ∂_n = ∂/∂ x_n$ . Equivalently it is the associative $K$ -algebra on those $2n$ generators subject only to the relations

[x_i, x_j] = 0, [∂_i, ∂_j] = 0, [∂_i, x_j] = δ_ij,

where $[a,b] = ab - ba$ and $δ_ij$ is the Kronecker delta. Its elements are exactly the operators $∑ c_αβ, x^α ∂^β$ with finitely many non-zero coefficients $c_αβ ∈ K$ .

Algebraic D-module

An algebraic D-module is a module over a ring of differential operators with algebraic coefficients. In this collection that means a left module over $A_n(K)$ , or occasionally over the ring $D(X)$ of differential operators on a smooth affine variety $X$ . Unless said otherwise, module means finitely generated left module.

Left, right, and why it matters

$A_n$ is not commutative, so left and right modules are genuinely different. The conventions of the theory are set up for left modules, because the natural action of an operator on a function is on the left. Right modules appear as soon as one integrates rather than differentiates - they are the natural home for densities - and the two categories are exchanged by the transposition anti-automorphism.

Core Concepts

Four ideas carry the whole subject. Each is elementary on its own; the combination is what is powerful.

Equations become modules

A system of linear equations $P_1 u = … = P_r u = 0$ with $P_i ∈ A_n$ is encoded by the left ideal $I = A_n P_1 + … + A_n P_r$ and the quotient module $M = A_n/I$ . The generator of $M$ is the class of $1$ , playing the role of the unknown function $u$ ; the relations imposed on it are exactly the equations. Different systems with the same solution behaviour can give the same module, and that is a feature: the module remembers the ideal of all operators that annihilate the unknown, which is more information than an arbitrary chosen generating set.

Solutions become homomorphisms

Because $M$ is generated by one element subject to the relations $I$ , a homomorphism $M \to N$ is determined by the image $v$ of the generator, and the only constraint on $v$ is that $Iv = 0$ . So $Hom_A_n(M,N)$ is the space of solutions of the system inside $N$ . The classical question 'what are the solutions?' becomes 'what are the maps out of $M$ ?', and the classical question 'solutions of what kind?' becomes 'maps into which $N$ ?'. Polynomial solutions, formal solutions, holomorphic solutions and distributional solutions are all obtained by varying $N$ alone. See solutions as module homomorphisms.

Modules have invariants that equations do not

Filter $A_n$ by degree and filter $M$ compatibly. The dimensions $_K Γ_m$ of the filtration pieces eventually agree with a polynomial in $m$ , the Hilbert polynomial. Its degree is the dimension $d(M)$ and its leading coefficient gives the multiplicity $e(M)$ . These are the numerical invariants that measure how heavily constrained the unknown function is: the more relations, the slower the growth, the smaller $d(M)$ . Bernstein's inequality says the growth can never fall below $m^n$ for a non-zero module, and the modules that sit exactly at that floor are the holonomic ones - the maximally overdetermined systems.

Operations follow maps of spaces

A polynomial map $f : K^n → K^m$ induces functors between $A_m$ -modules and $A_n$ -modules: the inverse image $f^*$ , which generalises substituting $f$ into a function, and the direct image $f_+$ , which generalises integrating along the fibres. The theory's central finiteness statement is that both preserve holonomicity. That is what allows a solvable class of systems to be closed under the operations one actually performs on differential equations.

Key Equations

The relation that generates everything is the commutator of a derivative with the variable it differentiates:

∂_i x_j - x_j ∂_i = δ_ij,

(I.1)

equivalently $∂ x = x∂ + 1$ in $A_1$ ; see the commutation relations.

It is forced by the product rule: applying both sides to a function $g$ gives $(xg)' - x g' = g$ . Every element of $A_n$ can be written in one and only one way in the canonical form

D = ∑ _α,β ∈ N^n c_αβ, x^α ∂^β, c_αβ ∈ K,

(I.2)

with all the $x$ 's pushed to the left. A system of equations is packaged as a module by

M = A_n / (A_n P_1 + ⋅ s + A_n P_r),

(I.3)

and the solutions of that system with values in an $A_n$ -module $N$ are recovered as

Hom_A_n(M, N) ; ≅ ; , v ∈ N ;:; P_1 v = ⋅ s = P_r v = 0 ,.

(I.4)

Finally, the numerical constraint that organises the classification of modules is

n ≤ d(M) ≤ 2n for M ≠ 0 finitely generated,

(I.5)

with equality on the left defining the holonomic modules.

Variable Definitions

$K$: the ground field, always of characteristic zero here
$A_n$: the $n$ -th Weyl algebra over $K$
$x_1, … ,x_n$: the coordinate functions, acting on a module by multiplication
$∂_1, … ,∂_n$: the partial derivatives, acting as differentiation
$δ_ij$: the Kronecker delta, equal to $1$ if $i = j$ and $0$ otherwise
$M$: a finitely generated left $A_n$ -module, the D-module under study
$N$: a second $A_n$ -module, serving as the space in which solutions are sought
$d(M)$ , $e(M)$: the dimension and multiplicity of $M$ , read off its Hilbert polynomial
$D(X)$: the ring of differential operators on a smooth affine variety $X$

Properties and Behaviour

The reason the translation is worth making is that $A_n$ is an unusually well-behaved noncommutative ring. Three facts, each proved in this collection, set the frame.

$A_n$ is a simple Noetherian domainCoutinho, Ch. 2 §1-§2, Ch. 8 §3

For $K$ of characteristic zero, $A_n(K)$ has no zero divisors, is left and right Noetherian, and its only two-sided ideals are $0$ and $A_n$ (simplicity). Simplicity is what makes every non-zero module faithful, and it is used in essentially every dimension estimate in the theory.

Bernstein's inequalityCoutinho (9.4.2)

Every non-zero finitely generated $A_n$ -module satisfies $n ≤ d(M) ≤ 2n$ . The lower bound is the non-trivial half, and it is what makes 'minimal dimension' a meaningful condition. See Bernstein's inequality.

Holonomic modules have finite lengthCoutinho, Ch. 10 §2

If $d(M) = n$ then $M$ has a composition series of length at most $e(M)$ . Finite length is the finiteness statement that the analysis needs: it is why a holonomic system has a finite-dimensional space of solutions in the appropriate sense, and why the class behaves like a category of finite objects.

What no D-module can be

No non-zero $A_n$ -module is finite dimensional over $K$ . If it were, the relation $[∂_i, x_i] = 1$ would give $tr(∂_i x_i) - tr(x_i ∂_i) = _K M$ , and the left side is zero while the right side is not, in characteristic zero. This is the algebraic shadow of the physical fact that Heisenberg's relation has no finite matrix solutions; see the no-go argument.

Examples and Special Cases

The polynomial ring

$K[x_1, … ,x_n]$ is a module over $A_n$ by the defining action: $x_i$ multiplies, $∂_i$ differentiates. It is the module attached to the system $∂_1 u = ⋅ s = ∂_n u = 0$ shifted by nothing at all - more precisely $K[x] ≅ A_n/(A_n∂_1 + ⋅ s + A_n∂_n)$ , the class of $1$ corresponding to the constant function. It is simple and holonomic, with $d = n$ and $e = 1$ .

The algebra acting on itself

$A_n$ is a module over itself: the system with no equations at all. It is the largest possible module in the dimension theory, $d(A_n) = 2n$ , and it is never holonomic for $n ≥ 1$ . Imposing no constraints on the unknown function is the opposite extreme from being holonomic.

The delta module

$A_1/A_1 x$ is the module attached to the single equation $x u = 0$ , whose classical solution is the Dirac delta. As a vector space it is $K[∂] 1$ , with $∂^k 1$ standing for the $k$ -th derivative of $δ$ . It is simple and holonomic, and it is the standard example of a module supported at a point. See the delta module.

Localisation

For a non-zero $f ∈ K[x_1, … ,x_n]$ , the ring $K[x][1/f]$ of rational functions with poles only along $f = 0$ is an $A_n$ -module, since the quotient rule keeps derivatives inside it. That it is finitely generated is already a theorem, and that it is holonomic is the technical heart of the Bernstein-Sato theory.

A first-order equation with a parameter

$A_1/A_1(x∂ - λ)$ encodes $x u' = λ u$ , whose classical solution is $x^λ$ . For $λ ∉ Z$ this module is simple; for integer $λ$ it is not, and the failure detects exactly the difference between the well-behaved power $x^λ$ and the rational or polynomial cases. Small examples like this one are how the theory is normally tested.

Worked Example

Turning $u' = u$ into a module, and reading its solutions off

Step 1 - write the module
Take $n = 1$ and the equation $u' = u$ , that is $Pu = 0$ with $P = ∂ - 1 ∈ A_1$ . The associated module is

$M = A_1/A_1(∂ - 1).$

Write $\bar{1}$ for the class of $1$ . The single relation says $∂, 1 = 1$ .
Step 2 - find a $K$ -basis
Any $D ∈ A_1$ has canonical form $∑ c_ab x^a ∂^b$ . Since $∂^b 1 = 1$ for every $b ≥ 0$ , we get $D 1 = ( ∑ _a,b c_ab x^a ) 1$ , so $M$ is spanned over $K$ by $x^a 1 : a ≥ 0$ . These are linearly independent: division on the right by the monic first-order operator $∂ - 1$ writes every element of $A_1$ uniquely as $Q(∂-1) + g(x)$ with $g ∈ K[x]$ , so $M ≅ K[x]$ as a $K$ -vector space, and in fact as a $K[x]$ -module of rank $1$ .
Step 3 - compute the invariants
Filter $M$ by $Γ_m = B_m 1$ , where $B_m$ is the span of the $x^a∂^b$ with $a + b ≤ m$ . By Step 2, $Γ_m$ is the span of $x^a 1$ for $a ≤ m$ , so

$_K Γ_m = m + 1.$

Check the first values by hand: $Γ_0 = K 1$ has dimension $1$ ; $Γ_1$ is spanned by $1, x 1, ∂ 1 = 1$ , so dimension $2$ ; $Γ_2$ adds $x^2 1$ , giving $3$ . The Hilbert polynomial is $χ(m) = m+1$ , hence $d(M) = 1 = n$ and $e(M) = 1! ⋅ 1 = 1$ . The module is holonomic, and it sits at the lower end of (I.5).
Step 4 - vary the target and watch the solutions change
By (I.4), $Hom_A_1(M,N) = v ∈ N : v' = v$ . Three targets:
- $N = K[x]$ : if $v ≠ 0$ has degree $k$ , then $v^{'}$ has degree $k - 1 < k$ , so $v' = v$ is impossible. The solution space is $0$ .
- $N = O(C)$ , the entire functions, with $K = C$ : the solutions are $v = c e^x$ , a space of dimension $1$ .
- $N = M$ itself: the generator $\bar{1}$ is a solution, and the space of solutions is $K 1$ , again of dimension $1$ .
The module has not changed between the three lines. Only the place where solutions are allowed to live has changed, and that is precisely what the second argument of $Hom$ controls.

Result

$M = A_1/A_1(∂ - 1)$ is a holonomic module with $d(M) = 1$ and $e(M) = 1$ , isomorphic to $K[x]$ as a vector space but not as an $A_1$ -module. Its solution space is zero in $K[x]$ and one-dimensional in the entire functions. The example shows both halves of the dictionary at work: the module carries the equation, and the target carries the notion of solution.

Applications and Industry Use

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

The applications are what justify the abstraction, and they are strikingly varied for a theory whose basic object is a quotient of a polynomial-like ring.

Analytic continuation and the b-function. Bernstein's proof that $f^s$ continues meromorphically in $s$ rests on the holonomicity of $K[x][1/f]$ and produces the Bernstein-Sato polynomial, now a standard invariant of a singularity.
Automatic proof of identities. Zeilberger's algorithm and creative telescoping turn the closure properties of holonomic functions into a decision procedure for binomial-sum and integral identities, implemented in every major computer algebra system.
Stability of differential equations. A D-module argument due to van den Essen supplies the key lemma in the study of global asymptotic stability of polynomial vector fields, an unexpected route from module theory to dynamics.
The Jacobian conjecture. The Dixmier conjecture about endomorphisms of the Weyl algebra is now known to be equivalent to the Jacobian conjecture, so a question about D-modules is a question about polynomial maps.
Representation theory. Localisation constructions realise representations of Lie algebras as D-modules on flag varieties, and the categories of holonomic modules that arise there carry the combinatorics of characters.
Singularity theory and mathematical physics. Vanishing cycles, monodromy, Feynman-integral reduction and hypergeometric systems are all organised by holonomic D-modules.

Design Considerations

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

Setting up a D-module problem requires three choices before any computation begins, and they are not interchangeable.

Which ring of operators

$A_n$ is the ring for affine space with polynomial coefficients. On a smooth affine variety $X$ one uses P2, which is again a simple Noetherian domain when $X$ is smooth and irreducible. On a singular variety $D(X)$ can fail to be Noetherian or even finitely generated, and nothing should be assumed. If solutions with essential singularities or convergent-series coefficients are needed, the algebraic ring is the wrong one and the analytic $D_X$ must be used instead.

Which filtration

The Bernstein filtration gives $x_i$ and $∂_i$ both degree $1$ , so every piece is finite dimensional over $K$ and counting arguments are elementary. The order filtration gives $x_i$ degree $0$ and $∂_i$ degree $1$ ; its pieces are infinite dimensional over $K$ but finitely generated over $K[x]$ , and its associated graded ring is the coordinate ring of the cotangent bundle, which is where the geometry lives. Dimension is the same for both, but that is a theorem, not a convention.

Which category to work in

For a single computation, the category of finitely generated $A_n$ -modules is enough. For anything involving the operations $f^*$ and $f_+$ systematically, the holonomic modules are the right category, because they are the ones closed under those operations. The full six-functor formalism needs derived categories, which is where the elementary treatment stops.

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.

D-modules are computable objects. A finitely generated module is stored as a presentation: a matrix over $A_n$ whose entries are in canonical form (I.2), so that a module is the cokernel of a map of free modules $A_n^s → A_n^r$ . Every structural question then becomes a question about left ideals and submodules, and the tool is the noncommutative Gröbner basis.

Encode operators in canonical form as exponent vectors with coefficients, so that products are computed by repeated application of the commutation relation.
Compute a Gröbner basis of the submodule of relations with respect to a term order refining the chosen filtration.
Take leading terms to obtain a graded module over a commutative polynomial ring in $2n$ variables.
Read off the Hilbert polynomial, hence the dimension and multiplicity, and hence whether the module is holonomic.
For solution questions, compute the restriction or integration modules that realise $f^*$ and $f_+$ .

The cost is dominated by the Gröbner step, which is doubly exponential in the number of variables in the worst case and involves $2n$ of them. In practice $n$ up to about four or five with modest degrees is routine, and beyond that the structure of the specific problem has to be exploited. Standard implementations are the Dmodules package in Macaulay2, dmod.lib in Singular, ore_algebra in SageMath and HolonomicFunctions.m in Mathematica.

Limits of Validity

The elementary theory buys its accessibility by restricting its scope. Four restrictions matter.

Characteristic zero. In characteristic $p$ the elements $x_i^p$ and $∂_i^p$ are central, $A_n$ is a finite module over its centre and is not simple, and the dimension theory collapses. See the positive-characteristic page.
Linearity. The dictionary applies only to linear systems. Nonlinear differential equations have no module attached to them by this construction; the relevant D-module methods there are indirect.
Affine and smooth. Everything is stated over affine space or a smooth affine variety. Projective bases and singular bases require sheaves of D-modules, and the pathologies on singular varieties are real.
No regularity theory. The distinction between regular and irregular singularities, and with it the Riemann-Hilbert correspondence, is a further layer. Holonomicity alone does not control the asymptotic behaviour of solutions; see the outlook page.

What algebraic D-modules do not tell you

The theory is very good at finiteness statements and very poor at explicit solutions. It will tell you that a solution space is finite dimensional, that a family of integrals satisfies some holonomic system, or that an operation preserves a good class. It will not hand you a closed form. That division of labour is exactly why the computational applications pair a D-module argument with a separate certificate-checking step.

Failure Modes and Common Mistakes

Confusing the module with its solution space

$Hom_A_n(M,N)$ depends on both arguments, and it is contravariant in the first. A non-zero module can have a zero solution space, as $A_1/A_1(∂ - 1)$ does in $K[x]$ ; two non-isomorphic modules can have isomorphic solution spaces in a given $N$ . The module is the equation, not the answer, and passing to solutions loses information unless a correspondence theorem says otherwise.

Assuming a D-module is a vector bundle with a connection

Modules that are finitely generated over $K[x]$ as well as over $A_n$ do correspond to vector bundles with a flat connection, and that is the picture most people carry from geometry. But it is a special case. The delta module $A_1/A_1x$ is not free, not even torsion-free, over $K[x]$ ; it is supported at a single point. The whole point of allowing general $A_n$ -modules is to admit such objects.

Expecting holonomic to mean finite dimensional

Holonomic means the Hilbert polynomial has the smallest possible degree $n$ , not that the module is small in absolute terms. No non-zero $A_n$ -module is finite dimensional over $K$ . $K[x]$ is holonomic and infinite dimensional; what is finite about a holonomic module is its length, and in good situations the dimension of its solution space.

Writing operators without respecting the order of factors

$x∂$ and $∂ x$ are different elements of $A_1$ ; they differ by $1$ . Every mechanical manipulation - comparing two operators, computing a degree, checking a relation - must first put both into canonical form (I.2). Nearly all errors in hand computation with the Weyl algebra come from silently commuting a derivative past a variable.

Reading 'algebraic' as a claim about the solutions

The adjective describes the coefficients of the operators and the base space, not the functions that solve the system. The module $A_1/A_1(∂ - 1)$ is as algebraic as anything in the theory, and its solution is $e^x$ , which is transcendental. Algebraic D-module theory studies transcendental functions routinely; it just refuses to let them into the coefficients.

Historical Notes

The ring came first and the modules came later, by nearly half a century. The relation $pq - qp = 1$ entered mathematics in 1925 through Heisenberg, Born and Jordan, and was taken up by Dirac as the defining relation of his 'quantum algebra'; that story is told on the quantum origins page. D. E. Littlewood in 1933 gave the algebra its first purely mathematical treatment, establishing the canonical form, the absence of zero divisors and the fact that no proper relation can be adjoined - simplicity, in modern language. The name Weyl algebra and the notation $A_n$ are due to Dixmier in the 1960s, after Weyl's book on group theory and quantum mechanics.

The module-theoretic viewpoint arose independently on two fronts. In the 1960s Malgrange treated constant-coefficient equations as modules over a polynomial ring; Sato's school in Japan, and decisively Kashiwara's 1970 thesis, developed the systematic theory for analytic coefficients under the name algebraic analysis. In the Soviet Union, Bernstein arrived at modules over the Weyl algebra from a completely different direction: a question of Gelfand from the 1954 International Congress about the meromorphic continuation of $f^s$ . In 1971-72 he answered it with an elementary argument built on the dimension theory of $A_n$ -modules, replacing the resolution of singularities that earlier proofs had required.

The subsequent history belongs to the analytic branch: Kashiwara's constructibility and index theorems, Gabber's 1981 proof that characteristic varieties are involutive, and the Riemann-Hilbert correspondence of Kashiwara and Mebkhout in 1984. The algebraic theory over the Weyl algebra remained as the accessible core, and acquired a second life in the 1990s when Zeilberger turned holonomicity into an algorithm for proving combinatorial identities.

Comparison

Three ways to look at the same linear system.
	Classical analysis	Algebraic D-modules	Analytic D-modules
Primary object	the equation $Pu = 0$	the module $A_n/A_n P$	the sheaf $D_X/D_X P$
Coefficients	whatever the problem gives	polynomial or regular	holomorphic
Base	a domain in $R^n$ or $C^n$	affine space or a smooth affine variety	a complex manifold
Solutions	functions satisfying the equation	elements of $Hom_A_n(M,N)$	the solution complex $RHom(M,O_X)$
Machinery needed	analysis	linear algebra and ring theory	sheaves and derived categories
Central finiteness	existence and uniqueness theorems	holonomicity, preserved by the operations	regular holonomicity and Riemann-Hilbert

The middle column is the subject of this collection. It is the smallest setting in which the mechanism of the theory is visible without the technical overhead of the third, and Coutinho's Primer was written precisely to occupy it.

Key Takeaways

Key points

A D-module is a module over a ring of differential operators; in the algebraic theory that ring is the Weyl algebra $A_n$ .
A linear system $P_1 u = ⋅ s = P_r u = 0$ becomes the module $A_n/(A_n P_1 + ⋅ s + A_n P_r)$ , and its solutions in $N$ become $Hom_A_n(M,N)$ .
Varying $N$ - polynomials, holomorphic functions, distributions - varies the notion of solution while leaving the module fixed.
Modules carry invariants that equations do not: dimension $d(M)$ , multiplicity $e(M)$ and the characteristic variety.
Bernstein's inequality confines $d(M)$ to $[n, 2n]$ ; the modules at the lower end are holonomic and have finite length.
Polynomial maps induce inverse and direct image functors, and holonomicity is preserved by both - the finiteness that makes the theory usable.
The algebraic setting assumes characteristic zero, linear equations, polynomial coefficients and a smooth affine base.

FAQs

What does the D stand for?

The ring of differential operators, written $D$ or $𝒟$ by long habit. A D-module is a module over it. There is no separate object called a 'D'; the notation is a compression of 'module over the ring of differential operators'.

Why replace an equation by a module at all? The equation is more concrete.

Because the module has structure the equation lacks. Modules can be added, quotiented, filtered, and transported along maps of the underlying space; they have numerical invariants; and they sit in exact sequences that let a hard case be reduced to easier ones. An equation on a page supports none of those operations. The payoff is theorems that hold for whole classes of systems at once.

Is every D-module attached to a differential equation?

Every cyclic module $A_n/I$ is, in the sense that $I$ is generated by operators and those operators are the system. Since $A_n$ is Noetherian, $I$ is finitely generated, so the system is finite. Modules with several generators correspond to systems for a vector of unknowns, presented by a matrix of operators. Remarkably, every holonomic module turns out to be cyclic, so in the holonomic world a single equation always suffices.

What is the difference between algebraic and analytic D-modules?

The coefficients and the base. Algebraic D-modules use polynomial or regular coefficients on an algebraic variety; analytic D-modules use holomorphic coefficients on a complex manifold and are sheaves rather than modules over a single ring. The analytic theory is more powerful and much heavier, requiring sheaf theory and derived categories. On smooth affine varieties the algebraic theory needs neither, because a module over the global ring already carries all the information.

Does the theory work over the real numbers?

The algebra does: $A_n(R)$ is a simple Noetherian domain and all the dimension theory applies, since only characteristic zero is used. What does not transfer well is the analytic interpretation. Solution spaces over $ℝ$ can behave badly, and statements about monodromy and analytic continuation need $ℂ$ . It is common to compute over $ℚ$ and interpret over $ℂ$ .

Is a D-module the same as a system of PDEs?

Not quite - it is the system modulo the choice of presentation. Two different generating sets for the same left ideal give the same module, and that is deliberate: the module remembers the ideal of all operators annihilating the unknown. Conversely, a module does not remember which generators of the ideal you started from, so information about the specific way the system was written is discarded.

Where does holonomicity come from, intuitively?

From over-determination. Each independent equation cuts down how fast the module can grow. Bernstein's inequality says growth can never drop below order $m^n$ , so there is a hard floor on how over-determined a consistent system can be, and holonomic systems are those that reach it. In one variable, a single non-trivial ODE already does; in $n$ variables, roughly $n$ independent conditions are needed.

Do I need algebraic geometry to start?

No. The first ten chapters of the elementary theory need only linear algebra, basic ring and module theory, and comfort with filtrations. Algebraic geometry enters only when characteristic varieties are introduced, and symplectic geometry only for the involutivity theorem. See the roadmap for what depends on what.

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 - Introduction §2, and Chs. 1, 5, 6 for the constructions described here.
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.
D. E. Littlewood, On the classification of algebras, Proceedings of the London Mathematical Society 35 (1933), 200-240 - the first mathematical study of Dirac's quantum algebra.
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209-242 - the source of the name and of the notation $A_n$ .
J.-E. Björk, Rings of Differential Operators, North-Holland Mathematical Library 21, North-Holland, 1979.
R. Hotta, K. Takeuchi and T. Tanisaki, D-modules, Perverse Sheaves, and Representation Theory, Progress in Mathematics 236, Birkhäuser, 2008 - the analytic and sheaf-theoretic branch.
M. Petkovšek, H. S. Wilf and D. Zeilberger, A = B, A. K. Peters, 1996 - the algorithmic applications of holonomicity.
A. Leykin and H. Tsai, Dmodules: functions for computations with D-modules, a package for Macaulay2.
ISO 80000-2:2019, Quantities and units - Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Write the system $∂_1 u = x_2 u$ , $∂_2 u = x_1 u$ as a module over $A_2$ and decide whether it is holonomic.
Show directly that $A_1/A_1 x$ has $K$ -basis the classes of $∂^k$ , and compute its Hilbert polynomial.
Explain why $Hom_A_n(M,N)$ is contravariant in $M$ and what that means for a surjection of modules.
Give two non-isomorphic $A_1$ -modules with the same space of polynomial solutions.
Compute the canonical form of $∂^2 x^2$ in $A_1$ and check it against the product rule.
Describe how the delta module changes if the base field has characteristic $p$ .
Compare the module attached to $u'' = 0$ with the module attached to $u' = 0$ and identify the exact sequence relating them.

Overview

Definition

Core Concepts

Equations become modules

Solutions become homomorphisms

Modules have invariants that equations do not

Operations follow maps of spaces

Key Equations

Variable Definitions

Properties and Behaviour

Examples and Special Cases

Worked Example

Applications and Industry Use

Design Considerations

Which ring of operators

Which filtration

Which category to work in

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