PrecisionLauricella : package for numerical computation of Lauricella functions depending on a parameter M.A. Bezuglov, B.A. Kniehl, A.I. Onishchenko, O.L. Veretin Abstract We introduce the PrecisionLauricella package, a computational tool developed in Wolfram Mathematica for high-precision numerical evaluations of Lauricella functions with indices linearly dependent on a parameter, ε \varepsilon . The package leverages a method based on analytical continuation via Frobenius generalized power series, providing an efficient and accurate alternative to conventional approaches relying on multi-dimensional series expansions or Mellin–Barnes representations. This one-dimensional approach is particularly advantageous for high-precision calculations and facilitates further optimization through ε \varepsilon -dependent reconstruction from evaluations at specific numerical values, enabling efficient parallelization. The underlying mathematical framework for this method has been detailed in our previous work, while the current paper focuses on the design, implementation, and practical applications of the PrecisionLauricella package. PROGRAM SUMMARY Program Title: PrecisionLauricella Developer’s repository link: https://bitbucket.org/BezuglovMaxim/precisionlauricella-package/src/main/ Licensing provisions(please choose one): GPLv3 Programming language: Wolfram Mathematica Supplementary material: PrecisionLauricella_Examples.nb Nature of problem(approx. 50-250 words): Lauricella functions, generalizations of hypergeometric functions, appearing in physics and mathematics, including Feynman integrals and string theory. When their indices depend linearly on a parameter ε \varepsilon , their numerical evaluation becomes challenging due to the complexity of high-dimensional series and singularities. Traditional methods, like hypergeometric re-expansion or Mellin–Barnes integrals, often lack efficiency and precision. Managing multi-dimensional sums exacerbates computational costs, especially for high-precision requirements, making these approaches unsuitable for many practical applications. Thus, there is a pressing need for efficient, scalable methods capable of maintaining numerical accuracy and effectively handling parameter dependencies. Solution method(approx. 50-250 words): Our method uses the Frobenius approach to achieve analytical continuations of Lauricella functions through generalized power series. Representing the functions as one-dimensional series simplifies high-precision numerical evaluation compared to traditional methods relying on multi-dimensional expansions or Mellin–Barnes integrals. We further optimize calculations by reconstructing ε \varepsilon dependencies from evaluations at specific values, enabling efficient parallelization and reducing computational costs. A comprehensive mathematical exposition of the method is provided in our previous work [ 1 ] . Journal: Computer Physics Communications 1 Introduction Hypergeometric functions, both in one and several variables, play a fundamental role in physics and mathematics, particularly in quantum field theory and the computation of Feynman integrals. Numerous approaches have been devised to express Feynman integrals through hypergeometric functions, including the Mellin--Barnes method, 1 1 1 For an introduction and references to foundational works, see Refs. [ 1 , 2 , 3 ] . the DRA method [ 4 , 5 , 6 ] , based on dimensional recurrence relation and analytical properties, the method of functional equations [ 7 ] , the exact Frobenius method [ 8 , 9 , 10 ] , and the Gelfand--Kapranov--Zelevinsky (GKZ) approach to Feynman integrals 2 2 2 For a comprehensive overview, see Refs. [ 11 , 12 ] . [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ] . These hypergeometric functions often depend on parameters linearly related to the dimensional regularization parameter \varepsilon , and their Laurent expansions in \varepsilon are essential for practical applications. Several computational tools have been developed for these expansions, both numerically and analytically 3 3 3 See also [ 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 ] and references therein for additional works on \varepsilon -expansion of hypergeometric functions. [ 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 ] . This paper introduces PrecisionLauricella , a software package written in Wolfram Mathematica, designed for the high-precision numerical evaluation of Lauricella functions with indices linearly dependent on the parameter \varepsilon . The mathematical foundations of the algorithm implemented in this package have been detailed in our previous work [ 40 ] . The package enables computations of these functions’ Laurent expansions about \varepsilon=0 for arbitrary argument values. The primary challenge lies in the analytical continuation of the defining series representations of Lauricella functions from their convergence domains to the entire \mathbb{C}^{n} space, where n is the number of arguments. While classical integral representations such as those of Euler [ 41 ] and Mellin–Barnes [ 41 , 42 ] can, in principle, address this continuation, they are less practical for high-precision numerical evaluations. Instead, series representations that solve the corresponding partial differential equations and are valid in subdomains covering the entire \mathbb{C}^{n} space are more effective. These representations not only facilitate accurate numerical computations, but also align with the classical approaches to analytic continuation developed by Kummer and Riemann [ 43 , 44 ] , which emphasize the monodromy group. Existing methods for the analytic continuation of Lauricella functions can be categorized into two main approaches: re-expansion of hypergeometric series using known continuations of simpler functions [ 42 , 45 , 46 , 41 , 47 , 48 ] and Mellin–Barnes integral representations [ 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 ] . In this work, we address the analytic continuation problem by leveraging the Frobenius method to construct generalized power series solutions of the Pfaffian systems satisfied by Lauricella functions along chosen analytic-continuation paths. This approach produces one-dimensional series, which are computationally more efficient than the multi-dimensional series arising in other methods, such as Mellin–Barnes representations. PrecisionLauricella automates the entire process, offering an accessible and efficient tool for researchers dealing with Lauricella functions and their applications. The remainder of the paper is organized as follows. In Section 2 , we briefly recall the definition of Lauricella functions and associated systems of differential equations. Section 3 , we provide a brief description of all the steps of the algorithm and present the program’s flowchart. Section 4 contains details on the usage of the PrecisionLauricella package and its performance. Finally, in Section 5 , we present our conclusion. 2 Lauricella Functions and associated differential equations In this section, we discuss the class of functions that are implemented in the presented package. Specifically, we focus on three Lauricella functions: F_{A}^{(n)} , F_{B}^{(n)} , and F_{D}^{(n)} for n\leq 3 . Generally, these functions are defined as hypergeometric series of several variables [ 60 , 61 ] : \displaystyle F_{A}^{(n)}(\alpha;\beta_{1},\dots,\beta_{n};\gamma_{1},\dots,\gamma_{n};x_{1},\dots,x_{n}) \displaystyle= \displaystyle\sum\limits_{m_{1},\dots,m_{n}=0}^{\infty}\frac{(\alpha)_{m_{1}+\dots+m_{n}}(\beta_{1})_{m_{1}}\dots(\beta_{n})_{m_{n}}}{(\gamma_{1})_{m_{1}}\dots(\gamma_{n})_{m_{n}}\,m_{1}!\dots m_{n}!}\,x_{1}^{m_{1}}\dots x_{n}^{m_{n}}\,, \displaystyle F_{B}^{(n)}(\alpha_{1},\dots,\alpha_{n};\beta_{1},\dots,\beta_{n};\gamma;x_{1},\dots,x_{n}) \displaystyle= \displaystyle\sum\limits_{m_{1},\dots,m_{n}=0}^{\infty}\frac{(\alpha_{1})_{m_{1}}\dots(\alpha_{n})_{m_{n}}(\beta_{1})_{m_{1}}\dots(\beta_{n})_{m_{n}}}{(\gamma)_{m_{1}+\dots+m_{n}}\,m_{1}!\dots m_{n}!}\,x_{1}^{m_{1}}\dots x_{n}^{m_{n}}\,, \displaystyle F_{D}^{(n)}(\alpha;\beta_{1},\dots,\beta_{n};\gamma;x_{1},\dots,x_{n}) (1) \displaystyle= \displaystyle\sum\limits_{m_{1},\dots,m_{n}=0}^{\infty}\frac{(\alpha)_{m_{1}+\dots+m_{n}}(\beta_{1})_{m_{1}}\dots(\beta_{n})_{m_{n}}}{(\gamma)_{m_{1}+\dots+m_{n}}\,m_{1}!\dots m_{n}!}\,x_{1}^{m_{1}}\dots x_{n}^{m_{n}}\,. All these series converge absolutely within specific regions. For the functions F_{B}^{(n)} and F_{D}^{(n)} , the region of convergence is given by |x_{1}|<1 , |x_{2}|<1 , …, |x_{n}|<1 . For the function F_{A}^{(n)} , the region is given by |x_{1}|+|x_{2}|+\dots+|x_{n}|<1 . When reduced to a single variable, these series correspond to the well-known hypergeometric function: F_{A}^{(1)}=F_{B}^{(1)}=F_{D}^{(1)}={}_{2}F_{1}\,. (2) Calculating these functions within their convergence regions is straightforward and does not pose significant challenges. However, our primary interest lies in their analytic continuation. To achieve this, we utilize partial differential equations associated with these functions. These partial differential equations can be derived directly from the series definitions. Consider a general series of the form f=\sum A_{i_{1},i_{2},\dots,i_{n}}x_{1}^{i_{1}}x_{2}^{i_{2}}\dots x_{n}^{i_{n}}\,, (3) where \frac{A_{i_{1},\dots,i_{k}+1,\dots,i_{n}}}{A_{i_{1},\dots,i_{k},\dots,i_{n}}}=\frac{P_{k}(i_{1},i_{2},\dots,i_{n})}{P^{\prime}_{k}(i_{1},i_{2},\dots,i_{n})}\,, (4) and P_{k} , P^{\prime}_{k} are polynomials in their variables. The function f then satisfies the following system of partial differential equations: \left[P^{\prime}_{k}\left(\theta_{x_{1}},\theta_{x_{2}},\dots,\theta_{x_{n}}\right)\frac{1}{x_{k}}-P_{k}\left(\theta_{x_{1}},\theta_{x_{2}},\dots,\theta_{x_{n}}\right)\right]f=0\,,\qquad k=1,\dots,n\,, (5) where \theta_{a}=a\partial/\partial a . For example, for the function F_{B}^{(2)}(\alpha_{1},\alpha_{2};\beta_{1},\beta_{2};\gamma;x,y)=F_{3} , the partial differential equations are: \displaystyle\small\left[x(1-x)\frac{\partial^{2}}{\partial x^{2}}+y\frac{\partial^{2}}{\partial x\partial y}+\left(\gamma-(\alpha_{1}+\beta_{1}+1)x\right)\frac{\partial}{\partial x}-\alpha_{1}\beta_{1}y\frac{\partial}{\partial y}-\alpha\beta_{1}\right]F_{3} \displaystyle= \displaystyle 0\,, \displaystyle\left[y(1-y)\frac{\partial^{2}}{\partial y^{2}}+x\frac{\partial^{2}}{\partial x\partial y}+\left(\gamma-(\alpha_{2}+\beta_{2}+1)y\right)\frac{\partial}{\partial y}-\alpha_{2}\beta_{2}x\frac{\partial}{\partial x}-\alpha\beta_{2}\right]F_{3} \displaystyle= \displaystyle 0\,. While obtaining such partial differential systems is straightforward, applying the Frobenius method to them can be challenging. To facilitate this, we require a complete system of equations in Pfaffian form: dJ=\left(M_{x}\,dx+M_{y}\,dy\right)J\,. (7) If we choose a basis as J=\left\{F_{3},~\theta_{x}F_{3},~\theta_{y}F_{3},~\theta_{x}\theta_{y}F_{3}\right\}\,, (8) then the matrices will have the form \displaystyle\scriptsize M_{x} \displaystyle= \displaystyle\left(\begin{array}[]{cccc}0&\frac{1}{x}&0&0\\ -\frac{\alpha_{1}\beta_{1}}{x-1}&-\frac{-\gamma+\alpha_{1}x+\beta_{1}x+1}{(x-1)x}&0&\frac{1}{(x-1)x}\\ 0&0&0&\frac{1}{x}\\ 0&-\frac{\alpha_{2}\beta_{2}y}{x(xy-x-y)}&-\frac{\alpha_{1}\beta_{1}(y-1)}{xy-x-y}&-\frac{-\alpha_{1}x-\beta_{1}x+\alpha_{1}xy+\beta_{1}xy+\alpha_{2}y+\beta_{2}y-\gamma y+y}{x(xy-x-y)}\\ \end{array}\right)\,, \displaystyle M_{y} \displaystyle= \displaystyle\left(\begin{array}[]{cccc}0&0&\frac{1}{y}&0\\ 0&0&0&\frac{1}{y}\\ -\frac{\alpha_{2}\beta_{2}}{y-1}&0&-\frac{-\gamma+\alpha_{2}y+\beta_{2}y+1}{(y-1)y}&\frac{1}{(y-1)y}\\ 0&-\frac{\alpha_{2}\beta_{2}(x-1)}{xy-x-y}&-\frac{\alpha_{1}\beta_{1}x}{y(xy-x-y)}&-\frac{\alpha_{1}x+\beta_{1}x-\gamma x+\alpha_{2}xy+\beta_{2}xy+x-\alpha_{2}y-\beta_{2}y}{y(xy-x-y)}\\ \end{array}\right)\,. In our work, for the functions under consideration, we utilize pre-derived Pfaffian systems from Ref. [ 39 ] with the following function bases: \displaystyle\left\{\theta_{x_{j_{1}}}\dots\theta_{x_{j_{k}}}F_{N}\,\Big|\,0\leq k\leq n,\,j_{1}