A holonomic function is a smooth function of one or more variables that satisfies a linear homogeneous partial differential equation (or system thereof) with polynomial coefficients, meaning its values and derivatives are linearly related through such equations.[1] This class encompasses solutions to holonomic systems in the sense of D-module theory, where the annihilator ideal in the Weyl algebra is holonomic, implying finite-dimensional solution spaces over the constants. Holonomic functions arise prominently in the study of linear differential equations with rational coefficients, as their solutions remain holonomic after clearing denominators to obtain polynomial coefficients.[2]Key properties of holonomic functions include closure under fundamental operations such as addition, multiplication, differentiation, indefinite integration, and composition with rational functions, making them a stable class for algebraic manipulations.[1] For instance, the vector space spanned by a holonomic function and its derivatives is finite-dimensional, and the function admits asymptotic expansions at regular singular points.[3] Examples abound in special functions, including the exponential function e^x, which satisfies f' - f = 0; the Bessel function J_\nu(x), solving x^2 y'' + x y' + (x^2 - \nu^2) y = 0; and the generalized hypergeometric function {}_pF_q, which obeys a linear ODE of order q+1.[1][4] In the discrete setting, q-holonomic functions generalize this concept, satisfying linear recurrences with coefficients that are polynomials in q and q^n, and they share analogous closure properties under summation and product.[5]Holonomic functions play a central role in symbolic computation and the automated proof of identities for special functions, facilitated by algorithms like creative telescoping introduced by Zeilberger in the 1990s.[1] Their theory underpins advances in D-module cohomology and microlocal analysis, with applications extending to quantum invariants, such as the colored Jones polynomial being q-holonomic.[6] This framework also aids in evaluating definite integrals and sums involving parameters, as these yield holonomic systems of PDEs.[2]
Fundamental Concepts
Definition for Functions
In mathematics, a holonomic function of one complex variable is a smooth or analytic function f(z) that satisfies a linear homogeneous ordinary differential equation with polynomial coefficients, specifically of the formP_0(z) f^{(n)}(z) + P_1(z) f^{(n-1)}(z) + \cdots + P_n(z) f(z) = 0,where the P_i(z) are polynomials in z, P_0(z) is nonzero, and n denotes the order of the equation.[7] This equation relates the function and its derivatives up to order n through polynomial dependencies, distinguishing holonomic functions from more general classes like arbitrary analytic functions.[8]The order n of such an equation is minimal if no linear homogeneous differential equation of lower order with polynomial coefficients is satisfied by f(z). The minimal order characterizes the function's "complexity" within this class, corresponding to the dimension of the solution space spanned by a fundamental set of solutions near regular points.[7] For instance, exponential functions like e^z satisfy a first-orderequation, making their minimal order 1.[9]The term "holonomic," originating from concepts in control theory and mechanics referring to integrable constraints, was adapted by Doron Zeilberger in 1990 to describe these special functions and systems in the context of proving identities via holonomic systems.[10] The collection of all holonomic functions forms a vector space over the field of rational functions \mathbb{C}(z), meaning that linear combinations with rational coefficients remain holonomic, with the minimal order of the combination at most the maximum of the individual orders.[7]
Definition for Sequences
A sequence (a_n)_{n \geq 0} of complex numbers is holonomic if it satisfies a linear homogeneous recurrence relation with polynomial coefficients, specifically, there exist polynomials P_0(n), P_1(n), \dots, P_k(n) in \mathbb{C}, where k is a positive integer and P_0(n) is not the zero polynomial, such that for all sufficiently large n,P_0(n) a_{n+k} + P_1(n) a_{n+k-1} + \cdots + P_k(n) a_n = 0.The minimal such k is called the order of the recurrence (or the order of the sequence). Holonomic sequences are also known as P-recursive sequences, emphasizing the polynomial nature of the coefficients in the defining recurrence.[11]The ordinary generating function A(x) = \sum_{n=0}^\infty a_n x^n of a holonomic sequence is itself a holonomic function, meaning it satisfies a linear homogeneous differential equation with polynomial coefficients.Given the recurrence of order k and initial values a_0, a_1, \dots, a_{k-1}, the holonomic sequence is uniquely determined for all n \geq k.[11]
Relation to Differential Equations and Recurrences
Holonomic functions are defined as the smooth solutions to systems of linear homogeneous partial differential equations with polynomial coefficients. This characterization establishes a direct correspondence between such functions and the ideals in the Weyl algebra generated by the differential operators that annihilate them. Specifically, for a function f of one variable, the Weyl algebra D = \mathbb{C}\langle \partial \rangle, where \partial = d/dx, acts on functions via the relation [\partial, x] = 1. The annihilator ideal \mathrm{Ann}_D(f) = \{ P \in D \mid P \cdot f = 0 \} determines the module D / \mathrm{Ann}_D(f), and f is holonomic if and only if this module is holonomic, meaning its characteristic variety has dimension equal to the dimension of the space (here, 1 for univariate functions). This framework, rooted in algebraic analysis, provides the theoretical bijection between holonomic functions and linear differential equations with polynomial coefficients, often studied through Ore extensions of polynomial rings to incorporate non-commutative operators.[12]In the discrete setting, holonomic sequences satisfy linear homogeneous recurrence relations with polynomial coefficients in the independent variable. For a sequence (a_n)_{n \geq 0}, this means there exist polynomials p_0(n), \dots, p_r(n) \in \mathbb{C}, not all zero, such that \sum_{k=0}^r p_k(n) a_{n+k} = 0 for all sufficiently large n. This equivalence mirrors the continuous case but uses shift operators in an Ore algebra \mathbb{C}\langle S \rangle, where S a_n = a_{n+1}, instead of differentiation. The annihilator ideal in this algebra plays an analogous role, capturing the relations that the sequence obeys. Seminal work by Richard Stanley introduced this notion for P-recursive sequences, later termed holonomic, emphasizing their closure under shifts and rational operations.[13]A keybridge between the continuous and discrete realms is provided by generating functions. The ordinarygenerating function f(x) = \sum_{n=0}^\infty a_n x^n of a holonomic sequence (a_n) is itself a holonomic function, satisfying a linear differential equation with polynomial coefficients derived from the recurrence via operator transformations in the Ore algebra. Conversely, if f(x) is holonomic, applying the shift to the coefficients transforms the differential equation into a recurrence, establishing the equivalence. This connection enables algorithmic translations between differential equations and recurrences, facilitating proofs of identities and asymptotic analyses.A fundamental theorem underscores this duality: a function f(x) = \sum_{n=0}^\infty a_n x^n / n! (or ordinary power series) is holonomic if and only if its Taylor series coefficients (a_n) form a holonomic sequence. This result follows from the fact that differentiation on f corresponds to shifts and multiplications by n on the coefficients, preserving the structure of the annihilator ideal across the continuous-discrete boundary. Leonard Lipshitz's work on D-finite functions formalized this for power series, proving closure properties that reinforce the bijection.[14]
Properties and Characterizations
Closure Properties
Holonomic functions, also known as D-finite functions, exhibit remarkable stability under a variety of algebraic and differential operations. The class is closed under addition and scalar multiplication: if f(x) and g(x) are holonomic, then so are af(x) + bg(x) for any constants a, b \in \mathbb{C}. Similarly, the product f(x) g(x) is holonomic, as the annihilating differential operator for the product can be derived from those of f and g using elimination techniques.[15] These properties extend to differentiation and integration; the derivative f'(x) satisfies a differential equation obtained by applying the annihilator to the differentiated form, and the indefinite integral \int f(x) \, dx (or definite integral from 0 to x) is holonomic, with the operator adjusted via integration by parts.[15]For holonomic sequences, which satisfy linear recurrences with polynomial coefficients (P-finite sequences), closure holds under addition and scalar multiplication by rational functions. If (a_n) and (b_n) are holonomic, then (a_n + b_n) is holonomic, and multiplication by a rational function r(n), which itself satisfies a first-order recurrence, yields another holonomic sequence via the product rule for recurrences. The class is also closed under shifts: if (a_n) is holonomic, then so is (a_{n+1}), as shifting preserves the recurrence structure.[1][15]Further closures include the Hadamard product for sequences, where the termwise product (a_n b_n) is holonomic, mirroring the multiplication closure. For functions, the Hadamard product of power series representations is also holonomic under suitable convergence conditions. Composition is preserved under restrictions: if f(x) is holonomic and g(x) is algebraic with g(0) = 0, then f(g(x)) is holonomic, computable via resultant-based elimination of the intermediate variable.[15][1]The class of holonomic functions and sequences is closed under taking limits of solutions to their defining equations, ensuring that pointwise or uniform limits of holonomic objects, when they satisfy the appropriate regularity, remain within the class. This structural stability under limits underscores the robustness of holonomic representations in asymptotic and computational contexts.[16]
Uniqueness and Characterization Theorems
One key characterization of holonomic functions arises from the structure of the annihilator ideal in the Weyl algebra. Specifically, every holonomic function satisfies a linear homogeneous differential equation of minimal order with polynomial coefficients, and this minimal-order operator is unique up to multiplication by a nonzero scalar in the polynomial ring. This uniqueness follows from the fact that the left ideal generated by the annihilators in the Weyl algebra admits a canonical generator of minimal degree, providing a normal form for the differential equation satisfied by the function.[1]A fundamental characterization theorem states that a smooth function f of several variables is holonomic if and only if the \mathbb{C}(x_1, \dots, x_n)-vector space spanned by f and all its partial derivatives has finite dimension. This finite-dimensionality condition captures the algebraic constraints imposed by the holonomic system and distinguishes holonomic functions from more general analytic functions, whose derivative spans would be infinite-dimensional.[17]For sequences, Zeilberger's theorem on creative telescoping provides a powerful characterization and construction tool. It asserts that if a bivariate function F(n, k) is holonomic—meaning it satisfies a linear recurrence with polynomial coefficients jointly in n and k—then the definite sum S(n) = \sum_k F(n, k) (over a suitable finite range for k) is also holonomic in n. The proof involves algorithmically producing a telescoping relation that yields a recurrence for S(n) directly from the annihilator of F, enabling the certification of holonomicity for sums without explicit computation.[18]
Examples
Holonomic Functions and Sequences
Holonomic functions and sequences are ubiquitous in mathematics, appearing in special functions, combinatorics, and their generating functions. These examples illustrate how linear differential equations or recurrences with polynomial coefficients capture a wide range of classical objects.A fundamental example is the exponential function e^z, which satisfies the first-order linear ordinary differential equation\frac{d}{dz} f(z) - f(z) = 0,with polynomial coefficients (of degree 0 in the leading term). This makes e^z holonomic of order 1.[19]Bessel functions of the first kind, J_\nu(z), provide another classical instance. They satisfy Bessel's differential equation,z^2 \frac{d^2}{dz^2} y(z) + z \frac{d}{dz} y(z) + (z^2 - \nu^2) y(z) = 0,a second-order linear equation with polynomial coefficients in z, confirming their holonomicity.[20][19]The generalized hypergeometric functions {}_p F_q \left( \begin{matrix} a_1, \dots, a_p \\ b_1, \dots, b_q \end{matrix} ; z \right) are also holonomic, as they obey a linear differential equation of order q+1 with polynomial coefficients derived from their series definition.[20][19]Turning to sequences, the factorial n! is holonomic, satisfying the first-order linear recurrences(n+1) = (n+1) s(n), \quad s(0) = 1,where the coefficient is a polynomial in n.[21]Binomial coefficients \binom{n}{k}, for fixed k, form a holonomicsequence as polynomials in n of degree k, satisfying a linear recurrence of order k+1 with polynomial coefficients; this follows from identities like Pascal's recurrence \binom{n}{k} = \binom{n-1}{k} + \binom{n-1}{k-1}, which extends to a univariate form via creative telescoping.[22]Generating functions for binomial sequences, such as (1 - z)^{-k} = \sum_{n=0}^\infty \binom{n+k-1}{k-1} z^n, are holonomic as they coincide with the hypergeometric function {}_1 F_0 (k ; ; z), satisfying an appropriate differential equation.[20]
Non-Holonomic Functions and Sequences
A classic example of a non-holonomic function is e^{1/z}, which exhibits an essential singularity at z = 0. Unlike holonomic functions, which satisfy linear ordinarydifferential equations (ODEs) of finite order with polynomial coefficients and thus possess only regular singularities, e^{1/z} cannot be annihilated by any such ODE because its Laurent series expansion around 0 involves infinitely many negative powers without a finite principal part, requiring either infinite order or non-polynomial coefficients for description.[23] This irregular behavior at the singularity prevents it from belonging to the class of holonomic functions.For sequences, the partition function p(n), which enumerates the number of ways to write the positive integer n as a sum of positive integers disregarding order, serves as a prominent non-holonomic example. The ordinary generating function \sum_{n=0}^\infty p(n) z^n does not satisfy any linear ODE with polynomial coefficients, as proven by showing that assuming such an equation leads to a contradiction with known modular properties of the function.[24] Consequently, the sequence p(n) fails to obey a linear recurrence relation with polynomial coefficients, highlighting its irregular growth pattern that defies the structured constraints of holonomic sequences.These non-holonomic cases are distinguished from holonomic ones partly by their asymptotic behaviors; for instance, while holonomic sequences typically admit controlled growth via recurrences, p(n) displays super-exponential asymptotics roughly on the order of \exp(c \sqrt{n}) for some constant c > 0, underscoring the absence of finite-order polynomial relations. Similarly, the rapid, direction-dependent growth of e^{1/z} near its essential singularity further illustrates the failure to meet holonomic criteria.
Advanced Extensions
Multivariable Holonomic Functions
In the multivariable setting, a function f(x_1, \dots, x_m) of several complex variables is holonomic if it is annihilated by a holonomic left ideal in the Weyl algebra D = \mathbb{C}[x_1, \dots, x_m]\langle \partial_1, \dots, \partial_m \rangle, where the partial derivatives \partial_i = \frac{\partial}{\partial x_i}.[25] This means f satisfies a system of linear partial differential equations (PDEs) with polynomial coefficients, and the associated D-module has a characteristic variety of dimension m in \mathbb{C}^{2m}.[25] Such functions arise naturally in algebraic analysis, where the holonomicity ensures that the solution space is finite-dimensional over generic points, generalizing the finite-order ordinary differential equation condition from the single-variable case.[25]A key tool for characterizing holonomicity in the multivariable context is the Bernstein-Sato polynomial, also known as the b-function, which is the monic polynomial b_f(s) \in \mathbb{C} of minimal degree satisfying a functional equation P \cdot f^{s+1} = b_f(s) \cdot f^s for some non-zero differential operator P \in D.[25] The roots of b_f(s) are negative rational numbers, and their properties reflect the algebraic structure of the annihilatorideal, distinguishing holonomic functions by ensuring the D-module's regularity and integrability along hypersurfaces.[25] Unlike simpler cases, the multivariable b-function captures interactions across variables, aiding in the verification of holonomicity through computational algebraic methods.[25]Prominent examples of multivariable holonomic functions are the A-hypergeometric functions introduced by Gel'fand, Kapranov, and Zelevinsky, which solve the GKZ system of PDEs defined by an integer matrix A \in \mathbb{Z}^{d \times n} and parameter vector \boldsymbol{\beta} \in \mathbb{C}^d. These functions generate holonomic ideals in the Weyl algebra, with their solution spaces encoding combinatorial data from toric varieties and appearing in statistical models like the Fisher-Bingham distribution. The holonomic nature of GKZ systems follows from the finite rank of the associated D-module, typically equal to the normalized volume of the toric polytope.The multivariable framework connects to the one-variable case through restrictions: if f(x_1, \dots, x_m) is holonomic, then its restriction to a linear subspace, such as setting x_{k+1} = \dots = x_m = c for constants c, yields a holonomic function in the remaining k variables.[25] This preservation of holonomicity under non-characteristic restrictions allows reduction to lower-dimensional problems while maintaining the D-module's structural properties.[25]
Connection to D-Finite Functions
In the algebraic framework, D-finite functions are defined as those formal power series that are annihilated by a non-zero linear differential operator with polynomial coefficients in the Weyl algebra, which consists of differential operators over the polynomial ring. This definition aligns precisely with the notion of holonomic functions in the single-variable case, where both terms describe functions satisfying linear ordinarydifferential equations with polynomial coefficients, ensuring a finite-dimensional solution space spanned by the function and its derivatives.[26][16][27]For multivariable extensions, the concept generalizes to holonomic D-modules, which are left modules over the Weyl algebra that possess finite length in the category of D-modules; this finiteness implies a bounded composition series with simple subquotients, capturing the algebraic structure of solutions to systems of partial differential equations. Such modules maintain the holonomic property through their characteristic varieties, which have minimal dimension equal to that of the ambient space.[28][29]Chyzak's theorem establishes that D-finiteness is preserved under definite integration and summation of D-finite functions, providing an algorithmic framework via creative telescoping to compute the annihilating differential operators for the resulting expressions. This preservation extends Zeilberger's methods from hypergeometric to general holonomic settings, enabling efficient symbolic computation of closure properties.[30]Unlike global analytic functions, which emphasize convergence and holomorphy over entire domains, D-finite functions prioritize formal power series solutions to differential equations, focusing on algebraic and computational aspects without requiring analytic continuation.[31]
Computational Methods
Algorithms for Computation and Verification
Algorithms for determining whether a function or sequence is holonomic involve computing candidate annihilating operators, such as linear recurrences or differential equations, and verifying if they reduce the object to zero. For sequences, Zeilberger's algorithm provides a method to find linear recurrences with polynomial coefficients from generating functions, particularly for hypergeometric terms.[32] The algorithm, introduced in the context of the holonomic systems approach, constructs a certificate operator that proves identities and derives recurrences by solving for undetermined coefficients in a system of polynomial equations derived from shift operators. Specifically, for a generating function defined by a sum S(n) = \sum_k F(n,k), where F is holonomic, the method applies creative telescoping to find a recurrence of the form P(\mathbf{N}, n) S(n) = 0, where \mathbf{N} denotes forward shifts and P is a polynomial in the shift and variableoperators. The steps include assuming an order for the recurrence, setting up equations from the telescoping condition \mathcal{L}(n,k) F(n,k) = \mathbf{N}_k G(n,k) - G(n,k) for some operator \mathcal{L}, and solving the resulting linear system over the rationals. This approach is efficient for low orders, with complexity dominated by the solution of polynomial systems of size proportional to the assumed order and degree.[32]For continuous holonomic functions represented by power series, a linear algebra-based method, applied notably to the Ising model susceptibility, computes the minimal linear differential equation satisfied by the series.[33] This method assumes a differential operator of order r with polynomial coefficients of degree at most d, expands the operator applied to the truncated power series y(x) = \sum_{i=0}^N a_i x^i, and sets the coefficients of the resulting series to zero, yielding a linear system for the unknown polynomial coefficients. The system is overdetermined for sufficient N \gg r(d+1), and the minimal equation is obtained by finding the lowest r and d for which a non-trivial solution exists, solved via Gaussian elimination. The complexity is O((r(d+1))^3 + r^2 d N), polynomial in the series length N and parameters, allowing computation for series up to thousands of terms. Verification involves substituting the full series (or more terms) and confirming the result is identically zero within numerical precision or symbolically for rational coefficients.[33]Creative telescoping extends these ideas to compute recurrences or differential equations for definite sums and integrals of holonomic functions, enabling operations like indefinite summation or integration within the holonomic class. Originating from Zeilberger's work on hypergeometric sums, the general algorithm for holonomic functions, as extended by Chyzak, operates in the Weyl algebra by constructing an operator T that annihilates the sum or integral by eliminating the summation/integration variable through telescoping relations.[34] Key steps include parameterizing candidate telescopers as undetermined coefficient polynomials in differential and shift operators, substituting into the annihilation condition derived from the input holonomic system's operators, and solving the resulting underdetermined system using Gröbner bases or linear algebra to find the minimal-degree solution. For an input holonomic function f(z) satisfying a differential equation and an integral I(z) = \int f(z,t) \, dt, the algorithm outputs a differential equation for I(z), preserving holonomicity. The approach applies similarly to sums, with complexity exponential in the number of variables due to ideal computations but practical for low-dimensional cases via heuristics for coefficient bounds.[34]Verification of holonomicity, or whether a given operator annihilates a function/sequence, relies on ideal membership testing in the corresponding Ore algebra. For a candidate recurrence or differential equation P, one computes a Gröbner basis of the annihilator ideal generated by known operators (if any) and reduces P with respect to the basis; if the remainder is zero, P annihilates the object. For power series or term sequences, an efficient numeric check substitutes the first M terms, where M exceeds the operator's degree, and confirms the output coefficients vanish, with error bounded by machine precision. Symbolically, for rational inputs, exact arithmetic verifies the identity. The complexity of Gröbner basis computation in the Weyl algebra is double-exponential in the number of variables and operator degrees, limiting practical verification to small systems, though for univariate cases, it reduces to polynomial time via linear algebra on Hankel-like matrices.[35]
Software Implementations
The gfun package in Maple facilitates the manipulation of holonomic sequences and functions defined by linear recurrences or differential equations with polynomial coefficients. Key features include guessing minimal recurrences or differential equations from a finite number of initial terms using the gfun[guess] command, computing formal power series expansions via gfun[series], and performing algebraic substitutions on holonomic functions with gfun[algebraicsubs]. These tools are particularly useful for rigorous computations in combinatorics and special functions, as detailed in the original implementation by Salvy and Zimmermann.[36][37][38]Complementing gfun, Maple's DEtools package supports computations for holonomic differential equations through functions like DEtools[polysols] and DEtools[ratsols], which identify polynomial and rational solutions of linear ordinary differential equations (ODEs) by solving associated systems in the Weyl algebra. The DEtools[FindODE] command determines the minimal-order homogeneous ODE satisfied by a given function, aiding verification of holonomicity. These capabilities enable efficient handling of solution spaces for holonomic systems.[39][40][41]Mathematica provides built-in support for hypergeometric functions, a prominent subclass of holonomic functions, via commands such as Hypergeometric2F1[a, b, c, z] for the Gauss hypergeometric function and HypergeometricPFQ[{a1, ..., ap}, {b1, ..., bq}, z] for generalized cases. These functions handle symbolic simplification, series expansions around specified points, and high-precision numerical evaluation, leveraging the closure properties of holonomic functions under differentiation and integration. For broader holonomic computations, the HolonomicFunctions package implements algorithms for multivariate cases, including creative telescoping to derive linear relations for integrals and sums.[42][43][44]FriCAS and its predecessor OpenAxiom offer domains for algebraic manipulations in D-modules, central to the theory of holonomic functions as solutions to systems of linear PDEs. The PartialDifferentialOperator domain constructs rings of partial differential operators over polynomial bases, supporting non-commutative arithmetic and Gröbner basis computations for ideals in the Weyl algebra. These features allow users to verify holonomicity by checking module dimensions and perform eliminations on differential ideals, with FriCAS emphasizing type-safe, category-based implementations for rigorous algebraic verification.[45][46]Recent developments in SageMath include enhancements to the ore_algebra package, with updates around 2019 integrated into versions 9.0 and later, extending support to multivariate Ore operators, enabling computations such as syzygy module calculations, operator factorization, and elimination ideals for verifying closure properties.[47] As of 2025, the package continues to be actively maintained, with recent enhancements for numerical evaluation methods in SageMath 10.x releases. This open-source implementation facilitates accessible Weyl algebra arithmetic, building on prior algorithms for differential and difference operators.