AMFlow: A Mathematica package for Feynman integrals computation via auxiliary mass flow

Xiao Liu, Yan-Qing MaComput. Phys. Commun 2023 · DOI

TL;DR

This paper introduces AMFlow, a Mathematica package that automates the computation of complex Feynman integrals using the auxiliary mass flow method. This new tool significantly simplifies the process of obtaining high-precision numerical results for calculations in particle physics, particularly for multi-loop and multi-scale problems.

What Problem Does This Paper Address?

Calculating Feynman integrals is a cornerstone of theoretical particle physics, essential for testing theories like the Standard Model and searching for new physics. However, these calculations become incredibly complex, especially for multi-loop processes with multiple energy scales. Traditional methods for computing these integrals, such as sector decomposition or Mellin-Barnes representations, are often inefficient for achieving high precision. While differential equation methods offer efficiency, they can be challenging to set up, particularly in determining boundary conditions. This paper addresses the need for a robust, automated, and efficient tool to tackle these complex Feynman integral calculations.

Methods

The core of AMFlow lies in its implementation of the auxiliary mass flow method. This technique introduces an auxiliary parameter, $\eta$, into the Feynman integrals, transforming them into functions of $\eta$. The integrals are then treated as solutions to a system of ordinary differential equations with respect to $\eta$. By solving these equations, typically through series expansions and numerical integration, the physical result can be recovered by taking the limit as $\eta$ approaches a specific value (often $\eta \to i0^-$).

Key aspects of the AMFlow package include:

  • Automated Differential Equation Solver: AMFlow provides a high-performance solver for these differential equations, capable of handling complex systems.
  • Iterative Strategy: For problems with a large number of master integrals, AMFlow employs an iterative approach to reduce the number of integrals that need to be solved, thereby decreasing computational cost.
  • Numerical Fit Strategy: To overcome potential difficulties with numerical solutions, AMFlow incorporates a "numerical fit" trick. This involves evaluating the function at several sample points and solving a system of linear equations to determine the coefficients of the series expansion. This method is particularly useful for handling complex kinematic parameters and avoiding issues with removable singularities.
  • Integration with Reduction Tools: AMFlow integrates with external programs like LiteRed, FiniteFlow, and Kira, which handle the initial reduction of Feynman integrals to a basis of master integrals.

Datasets

Not discussed in the paper. The paper focuses on the development and application of a computational tool rather than training on specific datasets.

Key Insights and Ideas

  • Auxiliary Mass Flow as a Unified Approach: The auxiliary mass flow method provides a systematic way to handle Feynman integrals, offering a unified framework for various types of integrals, including those with linear propagators and phase-space integrations.
  • Systematic Boundary Conditions: A significant advantage of the auxiliary mass flow method is the systematic way to obtain boundary conditions, often by considering the limit $\eta \to \infty$.
  • Efficiency of Ordinary Differential Equations: By transforming the problem into solving ordinary differential equations, high-precision results can be efficiently obtained, which is a significant improvement over multidimensional integration methods.
  • Numerical Fit for Robustness: The numerical fit strategy is a clever way to bypass potential numerical instabilities and achieve high precision, especially when dealing with complex kinematic configurations or removable singularities in the differential equations.
  • Code Simplification and Parallelization: The numerical fit approach simplifies code structure by treating integrals as pure numbers at sample points, and the computations at different sample points are independent, allowing for massive parallelization.

Key Results

The paper demonstrates the functionality of AMFlow through an explicit example of a two-loop double-box family involved in $t\bar{t}$ hadroproduction. The package successfully computes these integrals, providing results that agree with analytical calculations. The paper highlights the package's ability to handle complex scenarios, including:

  • High-Precision Calculations: AMFlow is designed to achieve high-precision numerical results.
  • Automation: The package automates the process of solving differential equations and performing numerical fits, significantly reducing manual effort.
  • Scalability: The iterative strategy and the numerical fit method are designed to handle increasingly complex problems with a large number of master integrals.

Limitations and Future Work

The authors acknowledge that while AMFlow is designed for general purposes, there might be "extreme cases" where the automatic computation function might not yield satisfactory results. In such instances, they introduce a "manual computation" approach where users can define their own parameters for numerical evaluation.

Future work mentioned includes:

  • Public Implementation of Reduction Methods: A public implementation of reduction methods developed in Refs. [11, 12] is planned, which is expected to significantly reduce computation time.
  • Enhanced Solver Access: In later versions, AMFlow will provide more convenient functions for users to access its differential equation solver.
  • Integration of More Reducers: The package is designed to be extensible, allowing for the integration of other integral reduction tools.

Summary updated

BibTeX
@article{liu2023amflow,
  author = {Xiao Liu and Yan-Qing Ma},
  title = {{AMFlow: A Mathematica package for Feynman integrals computation via auxiliary mass flow}},
  journal = {Comput. Phys. Commun.},
  year = {2023},
  doi = {10.1016/j.cpc.2022.108565},
}
Figure