Published for SISSA by Springer

Received: September 5, 2023

Revised: November 21, 2023
Accepted: January 11, 2024
Published: February 21, 2024

An explicit expression of generating function for

one-loop tensor reduction

JHEP02(2024)158
Chang Hu ,𝑎,𝑏,∗ Tingfei Li,𝑑,𝑒 Jiyuan Shen𝑎,𝑏,𝑐,∗ and Yongqun Xu 𝑓 ,𝑔
𝑎
School of Fundamental Physics and Mathematical Sciences,
Hangzhou Institute for Advanced Study, UCAS,
Hangzhou 310024, China
𝑏
University of Chinese Academy of Sciences,
Beijing 100049, China
𝑐
Institute of Theoretical Physics, Chinese Academy of Sciences,
Beijing 100190, China
𝑑
Zhejiang Institute of Modern Physics, School of Physics, Zhejiang University,
Hangzhou, Zhejiang 310058, China
𝑒
Joint Center for Quanta-to-Cosmos Physics, Zhejiang University,
Hangzhou, Zhejiang 310058, China
𝑓
Interdisciplinary Center for Theoretical Study, University of Science and Technology of China,
Hefei, Anhui 230026, China
𝑔
Department of Physics, South China University of Technology,
Guangzhou 510641, China
E-mail: isiahalbert@126.com, tfli@zju.edu.cn, isaiahyshen@gmail.com,
yongqunxu@mail.ustc.edu.cn

Abstract: This work introduces an explicit expression for the generating function for the
reduction of an 𝑛-gon to an (𝑛 − 𝑘)-gon. A novel recursive relation of generating function is
formulated based on Feynman Parametrization in projective space, involving a single ordinary
differential equation. The explicit formulation of generating functions provides crucial insights
into the complex analytic structure inherent in loop amplitudes.

Keywords: Higher-Order Perturbative Calculations, Field Theories in Higher Dimensions,

Scattering Amplitudes

ArXiv ePrint: 2308.13336

∗
Corresponding author.

Open Access, © The Authors.

https://doi.org/10.1007/JHEP02(2024)158
Article funded by SCOAP3 .
Contents
1 Background and motivation 1

2 Preparation 3
2.1 Notations 3
2.2 Recursive relation 4

3 𝒏-gon to 𝒏-gon 6
3.1 Analytical results 7

JHEP02(2024)158
4 𝒏-gon to (𝒏 − 1)-gon 9

5 𝒏-gon to (𝒏 − 2)-gon 11

6 Generating function of 𝒏-gon to (𝒏 − 𝒌)-gon 16

6.1 Analytic example 18
6.1.1 Triangle to Tadpole 18
6.1.2 Reduction coefficients of 𝑛-gon to (𝑛 − 𝑘)-gon 19

7 Proof 20

8 Conclusion 24

A Solving differential equations 25

B Numerical check 27
B.1 Tadpole 27
B.2 Bubble 28
B.3 Triangle 29
B.4 Box 33
B.5 Pentagon 36

1 Background and motivation

Scattering amplitudes is a fundamental concept in modern quantum field theory. They

serve as not only a source of precise predictions and explanations for experiments conducted
at the LHC [1] but also make a significant contribution to the analysis of the analytical
structure within perturbative quantum field theory. Moreover, they play a pivotal role in
our exploration of potential new phenomena in physics. Amplitudes, which result from
perturbative expansions, can be classified into two primary categories: tree diagrams and
loop diagrams, based on their respective orders.
The study of tree diagrams has a rich history, and various discoveries, such as the
Bern-Carrasco-Johansson (BCJ) color-kinematic duality [2–5], the Kawai-Lewellen-Tye (KLT)
relation [6], and on-shell techniques like the Britto-Cachazo-Feng-Witten (BCFW) recursion
relations [7, 8] have significantly deepened our comprehension of field theory. More recently,

–1–
there have been notable developments like the Cachazo-He-Yuan (CHY) formalism [9–13],
which strives to reframe field theory from twistor string [14].
At the level of loop diagrams, in the context of constructing Feynman integral expressions,
one of the most significant methods over the past few decades has been the utilization of
on-shell techniques [7, 8, 15–22]. On-shell methods are grounded in a fundamental physical
concept: when a propagator becomes on-shell, a physical process factorizes into the product of
two physical processes. Building upon this concept, the unitarity cut method [17, 18, 20, 22, 23]
has been developed, allowing the calculation of a one-loop diagram to be reduced to a
product of several tree-level diagrams. This approach significantly simplifies the computations
compared to the Feynman rules. In addition to this, within the CHY framework, there have

JHEP02(2024)158
been relevant developments in constructing Feynman integrands for loop diagrams. These
efforts primarily involve taking the forward limit of two external legs in the (𝑛 + 2)-point
tree-level amplitudes. This approach establishes a relationship between 𝑛-point one-loop
diagrams and (𝑛 + 2)-point tree-level amplitudes [24–33].
After providing the loop Feynman integrand, the usual approach for computing this
integral involves expressing it as a linear combination of a set of basis integrals, with the
coefficients being rational functions of external momenta, masses, and spacetime dimension [34].
These basis integrals, referred to as master integrals, can be computed once and for all. The
process of reduction can be divided into two categories: integrand-level reduction and
integral-level reduction. The integrand-level reduction can indeed be viewed as an algebraic
factorization of the loop integrand. It can be systematically resolved using computational
algebraic geometry [21, 35–39]. As for the integral-level reduction, the initial proposal was
the notable Passarino-Veltman reduction (PV-reduction) method [40]. By observing the
possible tensor structures and employing the method of undetermined coefficients, we can
recursively determine the coefficients for each tensor structure. Other proposals include the
Integration-by-Parts (IBP) method [41–48], the unitarity cut method [17, 18, 20, 23, 49–54],
and the Intersection number method [55–61]. Despite numerous advancements in integral-level
reduction, the growing complexity of computations warrants further improvements. There
are some mature software packages available for calculating higher-loop [47, 48, 62–85].
This article will primarily focus on the tensor reduction problem at one-loop level. This
problem has been under investigation for an extended period; however, a recent breakthrough
reveals that the introduction of a virtual auxiliary vector 𝑅, can streamline the conventional
Passarino-Veltman reduction process [86–90]. When an auxiliary vector 𝑅 is incorporated
into the IBP method, the improvement of efficiency of reduction has been shown in [91–93].
Another notable advancement is the introduction of the “generating function” [94–98]. In
fact, the concept of a generating function is well-established in both physics and mathematics.
The application of generating functions in higher-order QFT calculations was introduced
earlier in [94] and has been utilized up to the level of 3- and 4-loop calculations. In [95],
Kosower employed integration-by-parts (IBP) generating vectors to derive appropriate IBP
reduction relations in certain special 2-loop case. More recently, these functions have also
been discussed in [96], detailing how to directly proceed from these representations to the
physical result.
In the reduction process, the reduction coefficients for high tensor rank still present a
considerable challenge. However, when we sum up the reduction coefficients of different tensor
ranks, we might arrive at a simpler solution. For instance, in our method, the numerator

–2–
of the integral is (2𝑙 · 𝑅)𝑟 with rank 𝑟 . We can sum them up in two typical ways as shown
in the equations below:
∞ ∞
1 (2𝑙 ·𝑅)𝑟 ·𝓉𝑟
𝜓 1 (𝓉) = 𝓉𝑟 ·(2𝑙 ·𝑅)𝑟 = 𝜓 2 (𝓉) = = 𝑒 𝓉·(2𝑙·𝑅) . (1.1)
X X
,
𝑟 =0
1−𝓉(2𝑙 ·𝑅) 𝑟 =0
𝑟!
In recent research [97, 98], Bo Feng proposed a recursive method to calculate the generating
function, establishing several partial differential equations based on the auxiliary R method.
Subsequent work [98] has not only improved upon this at the one-loop level but also extended
it to the 2-loop level by setting up and solving differential equations of these generating
functions, utilizing the Integration-by-Parts (IBP) method. Both pieces of work underscore

JHEP02(2024)158
the considerable potential of generating functions. However, in both studies, authors provide
only an iterative approach to compute the generating functions, and it remains strenuous
to directly write generating functions explicitly, even at the one-loop level. Fortunately,
we discovered a new recursive relation of generating functions based on investigations into
Feynman parametrization in the projective space [90, 99, 100]. This new relation consists of
a single ordinary differential equation on 𝓉 instead of a complex set of partial differential
equations, allowing us to directly write an explicit expression of the generating function for
the reduction of 𝑛-gon to (𝑛 − 𝑘)-gon for universal 𝑘 without recursion.
The organization of this paper is as follows. Section 2 introduces essential notations used
throughout the paper and establishes our new recursive relation. In section 3, we compute
the generating function of 𝑛-gon to 𝑛-gon as a warm-up. Section 4 and section 5 detail
the derivation of the generating function of 𝑛-gon to (𝑛 − 1)-gon and 𝑛-gon to (𝑛 − 2)-gon,
respectively, as two non-trivial examples. After summarizing the previous results in section 6,
we provide an explicit expression of the generating function for 𝑛-gon to (𝑛 − 𝑘)-gon of
universal 𝑘. Section 7 offers an inductive proof of our results. Finally, section 8 provides
a brief summary and discussion. Beside, we present the solution of the typical differential
equation in appendix A. We also provide the numerical verification in appendix B.

2 Preparation

In this paper, our goal is to provide an explicit expression for the generating functions of
one-loop tensor reduction. As we know, by introducing an auxiliary vector 𝑅, the general
one loop integral in 𝐷 dimensions can be written as
𝑑𝐷𝑙 (2𝑅 · 𝑙)𝑟
Z
𝐼𝑛(𝑟 ) = . (2.1)
𝑖𝜋 𝐷/2 𝑛𝑗=1 (𝑙 − 𝑞 𝑗 )2 − 𝑀 𝑗2
Q

In this formula, 𝑛 represents the number of propagators, and 𝑟 denotes the tensor rank number.

2.1 Notations
Let us start with the introduction of some notations that we’ll be using subsequently.
• Some 𝑛-dimension vectors:
𝑳 : 𝐿𝑖 = 1,
𝑽 : 𝑉𝑖 = 𝑅 · 𝑞𝑖 , (2.2)
𝑯 𝑏 = {0, · · · , 0, 1, 0, · · · , 0},

–3–
 where 1 is in the 𝑏-th position.

• 𝑄 is an 𝑛 × 𝑛 matrix defined as
𝑀𝑖2 + 𝑀 𝑗2 − (𝑞𝑖 − 𝑞 𝑗 )2
𝑄𝑖 𝑗 = . (2.3)
2

• The notations (𝐴𝐵) and (𝐴𝐵)b , with a label list b = {𝑏 1, 𝑏 2, . . . , 𝑏𝑘 } for two vectors 𝐴
and 𝐵, are defined as follows:

(𝐴𝐵) = 𝐴 · 𝑄 −1 · 𝐵,
(2.4)
(𝐴𝐵)b = 𝐴b · (𝑄 b )−1 · 𝐵b

JHEP02(2024)158
b bb
b b
.

In this context, 𝐴b
b
and 𝐵b
b
are two vectors derived by removing all the 𝑏𝑖 -th components
from vectors 𝐴 and 𝐵. 𝑄 bbb
b
is a matrix obtained by eliminating all the 𝑏𝑖 −th rows and
𝑏𝑖 −th columns from matrix 𝑄. For example, with 𝑛 = 4 and b = {2, 3}, we have
 −1
𝑀12 +𝑀42 +(𝑞 1 −𝑞 4 )2
!
 
 2 2𝑀1
2 1
(𝑉 𝐿)2,3 = 𝑅 · 𝑞 1, 𝑅 · 𝑞 4 𝑀4 +𝑀1 +(𝑞 4 −𝑞 1 )2
2  . (2.5)
𝑀 4
2 1
2

• If Ω is an analytic expression composed of (𝐴𝐵) or (𝐴𝐵)b , then [Ω]a with a label set a
outside the square brackets represents the analytic expression obtained by appending
subscript a on each term of the form (𝐴𝐵) or (𝐴𝐵)b present in Ω. For example

[(𝑉 𝐿)]1,2 = (𝑉 𝐿)1,2, [(𝐿𝐿)2 ]3 = (𝐿𝐿)2,3 . (2.6)

If
(𝐻 1 𝐿)(𝑉𝑉 )3 + (𝐿𝐿)2𝑅 2
𝑃= , (2.7)
(𝑉 𝐿)2 (𝐻 3𝑉 )1
then
(𝐻 1 𝐿)4,5 (𝑉𝑉 )3,4,5 + (𝐿𝐿)2,4,5𝑅 2
[𝑃]4,5 = . (2.8)
(𝑉 𝐿)2,4,5 (𝐻 3𝑉 )1,4,5
(𝑟 ) (𝑟 )
• 𝐼 b represents a one-loop integral where all propagators of 𝐼𝑛 with index 𝑏𝑖 ∈ b
𝑛,b
are removed
𝑑𝐷𝑙 (2𝑅 · 𝑙)𝑟
Z
(𝑟 )
𝐼 b= 2. (2.9)
/ (𝑙 − 𝑞 𝑗 ) − 𝑀 𝑗
2
Q
𝑛,b 𝑖𝜋 𝐷/2 𝑛𝑗=1,𝑗 ∈b

2.2 Recursive relation

After introducing the above notations, we begin to derive the recursive relations for the
generating function of the reduction coefficients. As pointed out in [100] and [90], there exists
a non-trivial recursion relation for one-loop tensor integrals for 𝑟 ≥ 1,1
2(𝐷 + 2𝑟 − 𝑛 − 2) (𝑉 𝐿) (𝑟 −1) 4(𝑟 − 1) 𝑅 2 − (𝑉𝑉 ) (𝑟 −2)
𝐼𝑛(𝑟 ) = · · 𝐼𝑛 + · · 𝐼𝑛
𝐷 +𝑟 −𝑛 −1 (𝐿𝐿) 𝐷 +𝑟 −𝑛 −1 (𝐿𝐿)
(2.10)
𝑛
(𝑟 −1) 2(𝑟 − 1) · 𝑌 (𝑏) (𝑟 −2)
+ (𝑋 + ),
X
(𝑏)
·𝐼 b ·𝐼
𝑏=1
𝑛,𝑏 𝐷 + 𝑟 − 𝑛 − 1 𝑛,𝑏b
1
Where 𝑏 represents a single label 𝑏.

–4–
where the coefficients are
𝑋 (𝑏) = (𝐻𝑏 𝐿)(𝑉 𝐿)𝑏 − (𝐻𝑏𝑉 )(𝐿𝐿)𝑏 /(𝐿𝐿),


  (2.11)
𝑌 (𝑏) = (𝐻𝑏 𝐿)𝑅 2 + (𝐻𝑏𝑉 )(𝑉 𝐿)𝑏 − (𝐻𝑏 𝐿)(𝑉𝑉 )𝑏 /(𝐿𝐿).

We focus on the summation of the tensor integrals with form

∞
𝑑𝐷𝑙 1 1
Z
𝜙𝑛 (𝓉) = 𝓉𝑟 · 𝐼𝑛(𝑟 ) =
X
,
1 − 𝓉(2𝑅 · 𝑙) 𝑗=1 (𝑙 − 𝑞 𝑗 )2 − 𝑀 𝑗2
𝐷/2
Q𝑛
𝑟 =0
𝑖𝜋
∞ (2.12)
𝑑𝐷𝑙 1 1
Z
(𝑟 )
(𝓉) = 𝓉 ·𝐼 b =
X
𝑟
𝜙𝑛,b 2.
b 1 − 𝓉(2𝑅 · 𝑙) 𝑗=1,𝑗 ∈b
/ (𝑙 − 𝑞 𝑗 ) − 𝑀 𝑗
2
Q𝑛
𝑛,b 𝐷/2

JHEP02(2024)158
𝑟 =0
𝑖𝜋

Then, we multiply both sides of equation (2.10) by 𝓉𝑟 and sum over 𝑟 from 1 to ∞. The
following relations can be used:
∞
𝓉𝑟 · 𝐼𝑛(𝑟 ) = 𝜙𝑛 (𝓉) − 𝐼𝑛(0),
X

𝑟 =1
∞ ∞
𝑟 · 𝓉𝑟 · 𝐼𝑛(𝑟 ) = 𝓉 · 𝑟 · 𝓉 (𝑟 −1) · 𝐼𝑛(𝑟 ) = 𝓉 · 𝜙𝑛′ (𝓉),
X X

𝑟 =1 𝑟 =0
∞ ∞
𝓉𝑟 · 𝐼𝑛(𝑟 −1) = 𝓉 · 𝓉𝑟 · 𝐼𝑛(𝑟 ) = 𝓉 · 𝜙𝑛 (𝓉),
X X

𝑟 =1 𝑟 =0
∞ ∞ (2.13)
· 𝐼𝑛(𝑟 −1) = (𝑟 + 1) · 𝓉 · 𝓉 =𝓉 · 𝜙𝑛′ (𝓉) + 𝓉 · 𝜙𝑛 (𝓉),
X X
𝑟 ·𝓉 𝑟 𝑟
· 𝐼𝑛(𝑟 ) 2

𝑟 =1 𝑟 =0
∞ ∞
𝓉𝑟 · 𝐼𝑛(𝑟 −2) = 𝓉 (𝑟 +2) · 𝐼𝑛(𝑟 ) = 𝓉 2 · 𝜙𝑛 (𝓉),
X X

𝑟 =2 𝑟 =0
∞ ∞
𝑟 · 𝓉𝑟 · 𝐼𝑛(𝑟 −2) = (𝑟 + 2) · 𝓉 (𝑟 +2) · 𝐼𝑛(𝑟 ) = 𝓉 3𝜙𝑛′ (𝓉) + 2𝓉 2𝜙𝑛 (𝓉).
X X

𝑟 =2 𝑟 =0

We obtain a differential function for the generating function 𝜙𝑛 (𝓉):

(𝑉 𝐿) 𝑅 2 −(𝑉𝑉 ) 2
 
(𝐷 −𝑛−1)−2(𝐷 −𝑛)· 𝓉−4· 𝓉 𝜙𝑛 (𝓉)
(𝐿𝐿) (𝐿𝐿)
(𝑉 𝐿) 2 𝑅 2 −(𝑉𝑉 ) 3 ′
 
+ 𝓉−4· 𝓉 −4· 𝓉 𝜙𝑛 (𝓉)−(𝐷 −𝑛−1)𝐼𝑛(0) (2.14)
(𝐿𝐿) (𝐿𝐿)
𝑛
(    )
= 𝑋 (𝑏) (𝐷 −𝑛)𝓉·𝜙𝑛;𝑏b(𝓉)+𝓉 2 ·𝜙 ′ b(𝓉) +2𝑌 (𝑏) 𝓉 3 ·𝜙 ′ b(𝓉)+𝓉 2 ·𝜙𝑛;𝑏b(𝓉)
X
.
𝑛;𝑏 𝑛;𝑏
𝑏=1

We know that any integral can be written as the linear combination of certain irreducible scalar
integrals (which are the master integrals in arbitrary spacetime dimension) with coefficients
as the rational functions of external momenta, masses and space-time dimension 𝐷.
(𝑟 ) (0)
𝐼𝑛(𝑟 ) =
X
𝐶 ·𝐼 b,
b
𝑛→𝑛;b 𝑛;b
b⊆{1,2,...,𝑛}

∞
nX o (2.15)
(𝑟 ) (0) (0)
𝜙𝑛 (𝓉) = ·𝐼 b = (𝓉) · 𝐼 b .
X X
𝓉 ·𝐶
𝑟
GF𝑛→𝑛;b
b
b
𝑛→𝑛;b 𝑛;b 𝑛;b
b⊆{1,2,...,𝑛} 𝑟 =0 b⊆{1,2,...,𝑛}

–5–
Thus, equation (2.14) is transformed into a recursive formula for the generating functions
of the reduction coefficients.
(𝑉 𝐿) 𝑅 2 − (𝑉𝑉 ) 2
 
(𝐷 − 𝑛 − 1) − 2(𝐷 − 𝑛) · 𝓉−4· 𝓉 GF𝑛→𝑛;b (𝓉)
(𝐿𝐿) (𝐿𝐿) b

(𝑉 𝐿) 2 𝑅 2 − (𝑉𝑉 ) 3
 
+ 𝓉−4· 𝓉 −4· 𝓉 GF′ (𝓉)
(𝐿𝐿) (𝐿𝐿) 𝑛→𝑛;b b
(   (2.16)
′
= 𝑋 (𝑏𝑖 ) (𝐷 − 𝑛)𝓉 · GF𝑛;𝑏b →𝑛;b (𝓉) + (𝓉)
X
2
b
𝓉 · GF
𝑖 b b 𝑛;𝑏𝑖 →𝑛;b
𝑏𝑖 ∈b
 )
(0)

JHEP02(2024)158
+ 2𝑌 (𝑏𝑖 ) 𝓉 3 · GF′ b b (𝓉) + 𝓉 2 · GF𝑛;𝑏b →𝑛;b
b
(𝓉) + (𝐷 − 𝑛 − 1)𝐶 ,
𝑛;𝑏𝑖 →𝑛;b 𝑖 b
𝑛→𝑛;b

(0) (0) (0)

where 𝐶 is the reduction coefficient of scalar integral 𝐼𝑛
to Master Integral 𝐼 b . Because
b
𝑛→𝑛;b 𝑛;b
(0) (0)
𝐼𝑛 is an irreducible scalar integral, the reduction coefficient 𝐶 is 1 if b is the null
b 𝑛→𝑛;b
(0)
set, i.e, b = ∅ and 𝐶 = 0 otherwise. GF𝑛;𝑏b →𝑛;b
b
(𝓉) in the right hand side of (2.16) is
b
𝑛→𝑛;b 𝑖
(0)
the generating function for the reduction coefficient of 𝜙𝑛;𝑏b (𝓉) to Master Integral 𝐼 b . The
𝑖 𝑛;b
equation (2.16) serves as the cornerstone of this paper, drawing a connection between the
generating functions in the reduction of an 𝑛-gon to an (𝑛 − 𝑘)-gon and an (𝑛 − 1)-gon to
an ((𝑛 − 1) − (𝑘 − 1))-gon. At first, if we set b = ∅, the terms inside the curly braces on
the right-hand side of (2.16) vanish. The remaining part is a first-order non-homogeneous
ordinary differential equation with respect to the generating function GF𝑛→𝑛 (𝓉). By solving
this differential equation, we obtain analytic expression of the generating function for an
𝑛-gon to an 𝑛-gon. Once we have obtained the analytic expression for the generating function
for the reduction of an 𝑛-gon to (𝑛 − (𝑘 − 1))-gon, due to the arbitrariness of 𝑛, we can also
know the analytic expression for an (𝑛 − 1)-gon to ((𝑛 − 1) − (𝑘 − 1))-gon (which is what
appears on the right side of the equation). Then, the equation (2.16) becomes a first-order
non-homogeneous differential equation for the generating function of an 𝑛-gon to (𝑛 − 𝑘)-gon.
Continuously repeating this process, we can calculate all the generating functions.

3 𝒏-gon to 𝒏-gon
(0)
For the case of reducing an 𝑛-gon to an 𝑛-gon, that is, b = ∅, we have, 𝐶𝑛→𝑛 = 1 and the
terms inside the curly braces on the right-hand side of equation (2.16) vanish. Consequently,
the differential equation for the generating function GF𝑛→𝑛 (𝓉) simplifies to:
(𝑉 𝐿) 𝑅 2 − (𝑉𝑉 ) 2
!
(𝐷 − 𝑛 − 1) − 2(𝐷 − 𝑛) · 𝓉−4· 𝓉 GF𝑛→𝑛 (𝓉)
(𝐿𝐿) (𝐿𝐿)
(3.1)
(𝑉 𝐿) 2 𝑅 2 − (𝑉𝑉 ) 3
!
′
+ 𝓉−4· 𝓉 −4· 𝓉 GF𝑛→𝑛 (𝓉) = (𝐷 − 𝑛 − 1).
(𝐿𝐿) (𝐿𝐿)
The general solution to this differential equation is:2
1 (𝑥 − −𝑥 + )·𝓉
!
𝐷−𝑛−2 𝐷−𝑛−2 1, 𝐷−𝑛
𝓉 1−𝐷+𝑛
(1−𝑥 + ·𝓉) 2 (1−𝑥 − ·𝓉) 2 ·𝐶 1 + · 2 𝐹1 2 , (3.2)
1−𝑥 + ·𝓉 𝐷 −𝑛 1−𝑥 + ·𝓉
2
See appendix A.

–6–
where
 q 
2 (𝑉 𝐿) ± (𝐿𝐿)𝑅 2 + (𝑉 𝐿)2 − (𝐿𝐿)(𝑉𝑉 )
(3.3)
𝑥± = .
(𝐿𝐿)

The Generalized Hypergeometric Function is defined as:

∞
(𝑎 1 )𝑛 · · · (𝑎𝑝 )𝑛
!
𝑎 1, 𝑎 2, · · · , 𝑎𝑝 𝑧𝑛
= (3.4)
X
𝑝 𝐹𝑞 𝑧 · ,
𝑏 1 , 𝑏 2 , · · · , 𝑏𝑞 𝑛=0
(𝑏 1 )𝑛 · · · (𝑏𝑞 )𝑛 𝑛!

JHEP02(2024)158
where we have used the Pochhammer symbol

Γ(𝑥 + 𝑛) Y 𝑛
(𝑥)𝑛 ≡ = (𝑥 + (𝑖 − 1)). (3.5)
Γ(𝑥) 𝑖=1

At first glance, one might expect the undetermined constant to be resolved by the initial
(0)
condition GF𝑛→𝑛 (0) = 𝐶𝑛→𝑛 = 1. However, this doesn’t work in our case. So, how should
we determine 𝐶 1 ? The definition of GF𝑛→𝑛 (0) specifies that the generating functions should
be represented as a Taylor series of 𝓉. Nevertheless, the term

𝐷−𝑛−2 𝐷−𝑛−2
𝓉 1−𝐷+𝑛 (1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2 , (3.6)

cannot be expanded into a Taylor series3 of 𝓉. Thus, the constant 𝐶 1 = 0. Ultimately, we

obtain the generating function of the reduction from an 𝑛-gon to an 𝑛-gon as follows:

1 (𝑥 − − 𝑥 + )𝓉
!
1, 𝐷−𝑛
GF𝑛→𝑛 (𝓉) = · 2 𝐹1 2 . (3.7)
1 − 𝑥+ · 𝓉 𝐷 − 𝑛 1 − 𝑥+ · 𝓉

3.1 Analytical results

Upon deriving the formula for the generating function pertaining to the reduction from an 𝑛-
gon to an 𝑛-gon, as delineated in equation (3.7), we subsequently elucidate the corresponding
analytical formalism.
Substituting (3.3) into (3.7), we can obtain:

1
!
1, 𝐷−𝑛
GF𝑛→𝑛 (𝓉) = √ · 2 𝐹1 2 𝓍(𝓉 , (3.8)
2[(𝑉 𝐿)+ (𝐿𝐿)𝑅 2 +(𝑉 𝐿)2 −(𝐿𝐿)(𝑉𝑉 )] 𝐷 −𝑛
1− (𝐿𝐿)
·𝓉

where
q
4 (𝐿𝐿)𝑅 2 + (𝑉 𝐿)2 − (𝐿𝐿)(𝑉𝑉 ) · 𝓉
𝓍(𝓉) = − q , (3.9)
(𝐿𝐿) − 2[(𝑉 𝐿) + (𝐿𝐿)𝑅 2 + (𝑉 𝐿)2 − (𝐿𝐿)(𝑉𝑉 )] · 𝓉

3
Typically, 𝐷 is not selected as an integer due to dimensional regularization.

–7–
and

(𝐿𝐿) = 𝐿·𝑄 −1 ·𝐿
−1  
𝑀22 +𝑀12 −(𝑞 2 −𝑞 1 )2 𝑀12 +𝑀𝑛2 −(𝑞 1 −𝑞𝑛 )2

𝑀12 2 ··· 2 1
 2 2
  𝑀2 +𝑀1 −(𝑞2 −𝑞1 )
2
𝑀22 +𝑀𝑛2 −(𝑞 2 −𝑞𝑛 )2 
𝑀22 ···  1
   
2 2
= 1 1 · · · 1 · .. .. .. ..  · .. 
 
,

 . . . .  .
 
𝑀𝑛2 +𝑀12 −(𝑞𝑛 −𝑞 1 )2 𝑀𝑛2 +𝑀22 −(𝑞𝑛 −𝑞 2 )2
2 2 ··· 𝑀𝑛2 1

(𝑉 𝐿) = 𝑉 ·𝑄 −1 ·𝐿

JHEP02(2024)158
−1  
𝑀22 +𝑀12 −(𝑞 2 −𝑞 1 )2 𝑀12 +𝑀𝑛2 −(𝑞 1 −𝑞𝑛 )2

𝑀12 2 ··· 2 1
 2 2
  𝑀2 +𝑀1 −(𝑞2 −𝑞1 )
2
𝑀22 +𝑀𝑛2 −(𝑞 2 −𝑞𝑛 )2 
𝑀22 ···  1
   
2 2
= 𝑅 ·𝑞 1 𝑅 ·𝑞 2 · · · 𝑅 ·𝑞𝑛 ·  .. .. .. ..  · .. 
 
,

 . . . .  .
 
𝑀𝑛2 +𝑀12 −(𝑞𝑛 −𝑞 1 )2 𝑀𝑛2 +𝑀22 −(𝑞𝑛 −𝑞 2 )2
2 2 ··· 𝑀𝑛2 1

(𝑉𝑉 ) = 𝑉 ·𝑄 −1 ·𝑉
−1 
𝑀22 +𝑀12 −(𝑞 2 −𝑞 1 )2 𝑀12 +𝑀𝑛2 −(𝑞 1 −𝑞𝑛 )2

𝑀12

2 ··· 2 𝑅 ·𝑞 1
 2 2
  𝑀2 +𝑀1 −(𝑞2 −𝑞1 )
2
𝑀22 +𝑀𝑛2 −(𝑞 2 −𝑞𝑛 )2 
𝑀22 ···  𝑅 ·𝑞 2 
   
2 2
= 𝑅 ·𝑞 1 𝑅 ·𝑞 2 · · · 𝑅 ·𝑞𝑛 ·  .. .. .. ..  ..  .
·
  
.  . 


 . . . 

𝑀𝑛2 +𝑀12 −(𝑞𝑛 −𝑞 1 )2 𝑀𝑛2 +𝑀22 −(𝑞𝑛 −𝑞 2 )2
2 2 ··· 𝑀𝑛2 𝑅 ·𝑞𝑛 .
(3.10)
By expanding the generating function as a series in terms of 𝓉, we obtain:

∞
GF𝑛→𝑛 (𝓉) = (3.11)
X
(𝑟 )
𝓉𝑟 · 𝐶𝑛→𝑛 ,
𝑟 =0

(𝑟 )
where 𝐶𝑛→𝑛 are the reduction coefficients of tensor integral

𝑑𝐷𝑙 (2𝑅 · 𝑙)𝑟

Z
𝐼𝑛(𝑟 ) = , (3.12)
𝑖𝜋 𝐷/2 𝑛𝑗=1 (𝑙 − 𝑞 𝑗 )2 − 𝑀 𝑗2
Q

(0)
to the irreducible scalar integral (Master Integral) 𝐼𝑛 . Furthermore, using the Taylor
series expansion,

𝑘−1 (𝑎 1 )𝑖 (𝑎 2 )𝑖 · · · (𝑎𝑝 )𝑖 𝑚 𝑛
∞
! !
𝑎 1, 𝑎 2, · · · , 𝑎𝑝 𝑚 · 𝑧
𝑘
C𝑖−1 𝑖 𝑘−𝑖
=1+ (3.13)
X X
𝑝 𝐹𝑞 𝑧𝑘 ,
𝑏 1 , 𝑏 2 , · · · , 𝑏𝑞 1 − 𝑛 · 𝑧 𝑘=1 𝑖=1
𝑖!(𝑏 1 )𝑖 (𝑏 2 )𝑖 · · · (𝑏𝑞 )𝑖

(𝑘−1)!
where C𝑖−1
𝑘−1 are binomial coefficients defined as C𝑘−1 =
𝑖−1
(𝑖−1)!(𝑘−1−(𝑖−1))! , we can obtain the
coefficients for each order in the series expansion:

−1 X
( 𝐷−𝑛
2 )𝑖
𝑟X 𝑗  

𝑖 𝑟 −𝑖 𝑖
(𝑟 )
𝐶𝑛→𝑛 = 𝑥+
𝑟
+ 𝑥− − 𝑥+ 𝑥+ C𝑗 . (3.14)
𝑗=0 𝑖=0
(𝐷 − 𝑛)𝑖

–8–
4 𝒏-gon to (𝒏 − 1)-gon

Next, let us consider the case of the (𝑛 − 1)-gon. Without loss of generality, we select the
Master Integral as follows:
𝑑𝐷𝑙 1
Z
(0)
𝐼𝑛,𝑎b = 2. (4.1)
𝑗=1,𝑗̸=𝑎 1 (𝑙 − 𝑞 𝑗 ) − 𝑀 𝑗
2
𝐷/2
Q𝑛
1 𝑖𝜋
(0)
In this case, 𝐶𝑛→𝑛;𝑎b = 0. Then equation (2.16) transforms into:
1

(𝑉 𝐿) 𝑅 2 − (𝑉𝑉 ) 2
!
(𝐷 − 𝑛 − 1) − 2(𝐷 − 𝑛) · 𝓉−4· 𝓉 GF𝑛→𝑛;𝑎b1 (𝓉)
(𝐿𝐿) (𝐿𝐿)

JHEP02(2024)158
(𝑉 𝐿) 2 𝑅 2 − (𝑉𝑉 ) 3
!
′
+ 𝓉−4· 𝓉 −4· 𝓉 GF𝑛→𝑛; (𝓉)
(𝐿𝐿) (𝐿𝐿) 𝑎b1
(4.2)
 
′
= 𝑋 (𝑎1 ) (𝐷 − 𝑛)𝓉 · GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓉) + 𝓉 2 · GF𝑛;𝑎b1 →𝑛;𝑎b1
(𝓉)
 
′
+ 2𝑌 (𝑎1 ) 𝓉 3 · GF𝑛;𝑎b1 →𝑛;𝑎b1
(𝓉) + 𝓉 2 · GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓉) ,

where GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓉) can be obtained from (3.7). All we need to do is replace all instances
of 𝑛 with (𝑛 − 1) in expression (3.7), and replace all instances of (𝐿𝐿), (𝑉 𝐿), (𝑉𝑉 ) with
(𝐿𝐿)𝑎1 , (𝑉 𝐿)𝑎1 , (𝑉𝑉 )𝑎1 respectively. That is, GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓉) = [GF𝑛−1→𝑛−1 (𝓉)]𝑎1 ,

1 [𝑥 − − 𝑥 + ]𝑎1 · 𝓉
!
1, 𝐷−𝑛+1
GF𝑛,𝑎b1 →𝑛,𝑎b1 (𝓉) = · 2 𝐹1 2 , (4.3)
1 − [𝑥 + ]𝑎1 · 𝓉 𝐷 − 𝑛 + 1 1 − [𝑥 + ]𝑎1 · 𝓉
where the notation
 q 
2 (𝑉 𝐿)𝑎1 ± (𝐿𝐿)𝑎1 𝑅 2 + (𝑉 𝐿)𝑎21 − (𝐿𝐿)𝑎1 (𝑉𝑉 )𝑎1
(4.4)
[𝑥 ± ]𝑎1 = ,
(𝐿𝐿)𝑎1
as defined in section 2. The general solution to equation (4.2) is4
−2+𝐷−𝑛 −2+𝐷−𝑛
𝓉 1−𝐷+𝑛 (1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2
( Z 𝓉
𝑛−𝐷 𝑛−𝐷
× 𝐶1 + 𝓊−1+𝐷−𝑛 (1 − 𝑥 + · 𝓊) 2 (1 − 𝑥 − · 𝓊) 2
0

(4.5)
′
× 2𝑌 (𝑎1 ) 𝓊2 · GF𝑛; (𝓊) + 𝓊 · GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓊)


𝑎b1 →𝑛;𝑎b1
! )
′
+𝑋 (𝑎 1 )
(𝐷 − 𝑛)GF𝑛;𝑎b1 →𝑛;𝑎b1 (𝓊) + 𝓊 · (𝓊)


GF𝑛;𝑎b1 →𝑛;𝑎b1
𝑑𝓊 .

Clearly, the initial condition GF𝑛→𝑛,𝑎b1 (0) = 0 is insufficient to determine the undetermined
coefficient 𝐶 1 . However, since we require the generating function to be a Taylor series of
𝓉, we must have 𝐶 1 = 0.5
4
See appendix A.
𝑛−𝐷 𝑛−𝐷
5
(1 − 𝑥 + · 𝓊) 2 (1 − 𝑥 − · 𝓊) 2 and the terms in the large square brackets can be expanded as a Taylor
series of 𝓊. The factor 𝓊−1+𝐷−𝑛 in the integrand will cancel the factor 𝓉 1−𝐷+𝑛 outside the integrand. This
makes the result a Taylor series of 𝓉 after integrating.

–9–
 The integral in (4.5) cannot be evaluated directly. An intuitive idea would be to first
expand the integrand into a series in terms of 𝓊, and then integrate. However, fully expanding
a hypergeometric function into a series would render the expression overly complex. Therefore,
we decide to select a suitable integral formula and perform a series expansion only on part
of the integrand. The integral formula we select is
! !
𝑧𝑚+1 𝑎 1, . . . , 𝑎 𝑝 , 1 + 𝑚
Z
𝑎 1, . . . , 𝑎 𝑝
𝑑𝑧 𝑝 𝐹𝑞 𝑘 · 𝑧 · 𝑧𝑚 = ·
(𝑝+1) 𝐹 (𝑞+1) 𝑘 · 𝑧 + 𝐶𝑜𝑛𝑠𝑡 .
𝑏 1, . . . , 𝑏𝑞 𝑚+1 𝑏 1, . . . , 𝑏𝑞 , 2 + 𝑚
(4.6)
Then in order to evaluate the integral explicitly, we need to change the integration variable

JHEP02(2024)158
𝓊
𝓊 → 𝓍𝑎1 (𝓊) = . (4.7)
1 − [𝑥 + ]𝑎1 · 𝓊
Note that in (4.5) there are terms involving the derivative of the generating function
GF𝑛;𝑎b1 →𝑛;𝑎b1 , which requires the use of the following derivative formula for the general-
ized hypergeometric function
! !
1 + 𝑎 1, . . . , 1 + 𝑎𝑝
Q𝑝
𝑑 𝑎 1, . . . , 𝑎 𝑝 𝑎𝑖
𝑝 𝐹𝑞 𝑧 = Q𝑞𝑖=1 · 𝑝 𝐹𝑞 𝑧 . (4.8)
𝑑𝑧 𝑏 1, . . . , 𝑏𝑞 𝑗=1 𝑏 𝑗 1 + 𝑏 1 , . . . , 1 + 𝑏𝑞

Substituting (4.3) in (4.5), after altering the integration variable, the integral part becomes
( !
1, 𝐷−𝑛+1
Z 𝓍𝑎 (𝓉)
1 𝑛−𝐷
𝑥 𝐷−𝑛−1
(1 + A𝑎1 𝑥 + B𝑎1 𝑥 ) 2 2 (P𝑎(0) + P𝑎(1) 𝑥) · 2 𝐹1 2 [𝑥 − − 𝑥 + ]𝑎1 · 𝑥
0
1 1
𝐷 −𝑛 +1
!)
𝑛−𝐷 2, 𝐷−𝑛+3
+ (1 + A𝑎1 𝑥 + B𝑎1 𝑥 ) 2 2 (Q𝑎(1) 𝑥 + Q𝑎(2) 𝑥 2) · 2 𝐹1 2 [𝑥 − − 𝑥 + ]𝑎1 · 𝑥 𝑑𝑥,
1 1
𝐷 −𝑛 +2
(4.9)
where
A𝑎1 = −(𝑥 + + 𝑥 − ) + 2[𝑥 + ]𝑎1 ,
B𝑎1 = 𝑥 + · 𝑥 − − (𝑥 + + 𝑥 − ) · [𝑥 + ]𝑎1 + [𝑥 +
2
]𝑎1 ,
P𝑎(0)
1
(𝑛) = (𝐷 − 𝑛) · 𝑋 (𝑎1 ),
(4.10)
P𝑎(1)
1
= 2𝑌 (𝑎1 ) + [𝑥 + ]𝑎1 · 𝑋 (𝑎1 ),
Q𝑎(1)
1
= −[𝑥 + − 𝑥 − ]𝑎1 · 𝑋 (𝑎1 ) /2,
1
Q𝑎(2) = − · (2𝑌 (𝑎1 ) + [𝑥 + ]𝑎1 · 𝑋 (𝑎1 ) ) · [𝑥 + − 𝑥 − ]𝑎1 .
1
2
We expand
∞
𝑛−𝐷
(1 + A𝑎1 𝑥 + B𝑎1 𝑥 2 ) (P𝑎(0) + P𝑎(1) 𝑥) = (𝑛; 𝑚) · 𝑥 𝑚 ,
X
2
1 1
C𝑎(1)
1

(4.11)
𝑚=0
∞
𝑛−𝐷
(1 + A𝑎1 𝑥 + B𝑎1 𝑥 ) (Q𝑎(1) + 𝑥 2) = (𝑛; 𝑚)
X
2 2
1
𝑥 Q𝑎(2)
1
C𝑎(2)
1
·𝑥 ,𝑚

𝑚=0

as a Taylor series of 𝑥. By formula

∞ n ⌊X ⌋ 𝑚
2
(𝑐 − 𝑚 + 𝑖 + 1)(𝑚−𝑖) o
(1 + A𝑥 + B𝑥 2 )𝑐 = (4.12)
X
A𝑚−2𝑖 B𝑖 𝑥 𝑚 ,
𝑚=0 𝑖=0
(𝑚 − 2𝑖)! · 𝑖!

– 10 –
we have the expansion coefficients as follows:

⌊𝑚 ⌋
X ( 𝑛−𝐷 − 𝑚 + 𝑖 + 1)(𝑚−𝑖)
2
C𝑎(1) (𝑛; 𝑚) = P𝑎(0) (𝑛) · 2
A𝑚−2𝑖 B𝑖𝑎1
1 1
𝑖=0
(𝑚 − 2𝑖)! · 𝑖! 𝑎1

⌊ 𝑚−1 ⌋
2X ( 𝑛−𝐷 − 𝑚 + 𝑖 + 2)(𝑚−1−𝑖)
+ P𝑎(1) · 2
A𝑚−1−2𝑖 B𝑖𝑎1 ,
1
(𝑚 − 1 − 2𝑖)! · 𝑖!
𝑎1
(4.13)
𝑖=0
⌊ 𝑚−1 ⌋
X ( 𝑛−𝐷 − 𝑚 + 𝑖 + 2)(𝑚−1−𝑖)
2
C𝑎(2) (𝑛; 𝑚) = Q (1)
· 2
A𝑚−1−2𝑖 B𝑖𝑎1
1 𝑎1
(𝑚 − 1 − 2𝑖)! · 𝑖! 𝑎1

JHEP02(2024)158
𝑖=0
⌊ 𝑚−2 ⌋
2X ( 𝑛−𝐷 − 𝑚 + 𝑖 + 3)(𝑚−2−𝑖)
+ Q𝑎(2) · 2
A𝑚−2−2𝑖 B𝑖𝑎1 .
1
𝑖=0
(𝑚 − 2 − 2𝑖)! · 𝑖!
𝑎1

Then the integral part (excluding the constant) becomes:

∞
!
1, 𝐷−𝑛+1
Z 𝓍𝑎 (𝓉)
1
(𝑛; 𝑚) [𝑥 − − 𝑥 + ]𝑎1 · 𝑥
X
C𝑎(1) 𝑑𝑥 𝑥 𝐷−𝑛−1+𝑚
· 2 𝐹1 2
1
0 𝐷 −𝑛 +1
(4.14)
𝑚=0
∞
!
2, 𝐷−𝑛+3
Z 𝓍𝑎 (𝓉)
1
+ (𝑛; 𝑚) [𝑥 − − 𝑥 + ]𝑎1 · 𝑥 .
X
C𝑎(2) 𝑑𝑥 𝑥 𝐷−𝑛−1+𝑚
· 2 𝐹1 2
𝑚=0
1
0 𝐷 −𝑛 +2

By using integral formula of generalized hypergeometric function (4.6), we can explicitly solve
for the generating function of the reduction from an 𝑛-gon to an (𝑛 − 1)-gon,

GF𝑛→𝑛;𝑎b1 (𝓉)
∞
(
−2+𝐷−𝑛 𝓉𝑚+1
= ((1 − 𝑥 + · 𝓉)(1 − 𝑥 − · 𝓉))
X
2 ×
𝑚=0
(𝑚 + 𝐷 − 𝑛)(1 − [𝑥 + ]𝑎1 · 𝓉)𝑚+𝐷−𝑛

[𝑥 − − 𝑥 + ]𝑎1 · 𝓉 (4.15)
( !
1, 𝐷−𝑛+1
,𝑚+𝐷 −𝑛
× C𝑎(1) (𝑛; 𝑚) · 3 𝐹2 2
1
𝐷 − 𝑛 + 1, 𝑚 + 𝐷 − 𝑛 + 1 1 − [𝑥 + ]𝑎1 · 𝓉
[𝑥 − − 𝑥 + ]𝑎1 · 𝓉
! ))
2, 𝐷−𝑛+3 ,𝑚 + 𝐷 − 𝑛
+ C𝑎(2) (𝑛; 𝑚) · 3 𝐹2 2 .
1
𝐷 − 𝑛 + 2, 𝑚 + 𝐷 − 𝑛 + 1 1 − [𝑥 + ]𝑎1 · 𝓉

5 𝒏-gon to (𝒏 − 2)-gon

Now, let us discuss the reduction from an 𝑛-gon to an (𝑛 − 2)-gon. Without loss of generality,
we select the Master Integral:

𝑑𝐷𝑙 1
Z
(0)
𝐼𝑛,\ = 2. (5.1)
𝑗=1,𝑗̸=𝑎 1 ,𝑎 2 (𝑙 − 𝑞 𝑗 ) − 𝑀 𝑗
2
𝑎 ,𝑎 𝐷 Q𝑛
1 2
𝑖𝜋 2

– 11 –
 (0)
In this case, 𝐶𝑛→𝑛;\
𝑎 ,𝑎
= 0. Then recursive relation (2.16) transforms into
1 2

(𝑉 𝐿) 𝑅 2 − (𝑉𝑉 ) 2
!
(𝐷 − 𝑛 − 1) − 2(𝐷 − 𝑛) · 𝓉−4· 𝑎 1 ,𝑎 2 (𝓉)
𝓉 GF𝑛→𝑛;\
(𝐿𝐿) (𝐿𝐿)
(𝑉 𝐿) 2 𝑅 2 − (𝑉𝑉 ) 3
!
′
+ 𝓉−4· 𝓉 −4· 𝑎 1 ,𝑎 2 (𝓉)
𝓉 GF𝑛→𝑛;\
(𝐿𝐿) (𝐿𝐿)
(
 
′
= 2𝑌 (𝑎1 ) 𝓉 3 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2
(𝓉) + 𝓉 2 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉)

)
 
′
+𝑋 (𝑎 1 )
(𝐷 − 𝑛)𝓉 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉) + 𝓉 ·
2
(𝓉)

JHEP02(2024)158
GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2
(
 
′
+ 2𝑌 (𝑎2 ) 𝓉 3 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2
(𝓉) + 𝓉 2 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉)

)
 
′
+𝑋 (𝑎 2 )
(𝐷 − 𝑛)𝓉 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉) + 𝓉 ·
2
GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2
(𝓉) . (5.2)

The general solution of (5.2) is

−2+𝐷−𝑛 −2+𝐷−𝑛
𝓉 1−𝐷+𝑛 (1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2
( Z 𝓉
𝑛−𝐷 𝑛−𝐷
× 𝐶1 + 𝓊−1+𝐷−𝑛 (1 − 𝑥 + · 𝓊) 2 (1 − 𝑥 − · 𝓊) 2
0
n  
′
× 2𝑌 (𝑎1 ) 𝓊2 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2
(𝓊) + 𝓊 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2 (𝓊)
 
′
+ 𝑋 (𝑎1 ) (𝐷 − 𝑛) · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2 (𝓊) + 𝓊 · GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2
(𝓊)
 
′
+ 2𝑌 (𝑎2 ) 𝓊2 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2
(𝓊) + 𝓊 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓊)
)
 o
′
+ 𝑋 (𝑎2 ) (𝐷 − 𝑛) · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓊) + 𝓊 · GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2
(𝓊) 𝑑𝓊 , (5.3)

where GF𝑛;𝑎b1 →𝑛;\

𝑎 1 ,𝑎 2 (𝓉) and GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉) can be obtained from (4.15) by

𝑎 1 ,𝑎 2 (𝓉) = [GF𝑛−1→𝑛−1;𝑎b2 (𝓉)]𝑎 1 ,

GF𝑛;𝑎b1 →𝑛;\

𝑎 1 ,𝑎 2 (𝓉) = [GF𝑛−1→𝑛−1;𝑎b1 (𝓉)]𝑎 2 .

GF𝑛;𝑎b2 →𝑛;\ (5.4)
We select the undetermined constant 𝐶 1 = 0 since the generating function should be a
Taylor series of 𝓉. We can see that in the integrand of expression (5.3), 𝑎 1 and 𝑎 2 are
completely symmetric. We can calculate only one half and obtain the other half by exploiting
this symmetry. Moreover, since we are going to use integral formula (4.6) to compute the
integral, we can divide GF𝑛;𝑎b1 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉) and GF𝑛;𝑎b2 →𝑛;\
𝑎 1 ,𝑎 2 (𝓉) according to the generalized
hypergeometric function as follows:
∞ X
(𝑙)(𝑚 )
𝑎 1 ,𝑎 2 (𝓉) = [𝐶𝑎(𝑙)2 (𝑛 − 1; 𝑚 1 )]𝑎1 GF𝑛;𝑎b →𝑛;\ (𝓉),
X
1
GF𝑛;𝑎b1 →𝑛;\ 𝑎 ,𝑎1 1 2
𝑚 1 𝑙=1,2
∞ X (5.5)
(𝑙)(𝑚 1 )
𝑎 1 ,𝑎 2 (𝓉) = [𝐶𝑎(𝑙)1 (𝑛 1; 𝑚 1 )]𝑎2 GF𝑛;𝑎b →𝑛;\ (𝓉),
X
GF𝑛;𝑎b2 →𝑛;\ − 𝑎 1 ,𝑎 2
2
𝑚 1 𝑙=1,2

– 12 –
where

−1+𝐷 −𝑛 −1+𝐷 −𝑛
(1)(𝑚 ) 1
1
GF𝑛;𝑎b →𝑛; (𝓉) = (1−[𝑥 + ]𝑎1 ·𝓉) 2 (1−[𝑥 − ]𝑎1 ·𝓉) 2
1 𝑎[
,𝑎 1 2 𝑚 1 +𝐷 −𝑛+1
( 𝐷 −𝑛+2 !)
𝓉𝑚1 +1 1, ,𝑚 1 +𝐷 −𝑛+1 [𝑥 − −𝑥 + ]𝑎1 ,𝑎2 ·𝓉
× · 𝐹 2 ,
(1+[𝑥 + ]𝑎1 ,𝑎2 ·𝓉)𝑚1 +𝐷−𝑛+1 3 2 𝐷 −𝑛+2,𝑚 +𝐷 −𝑛+2 1−[𝑥 + ]𝑎1 ,𝑎2 ·𝓉
1

−1+𝐷 −𝑛 −1+𝐷 −𝑛
(2)(𝑚 ) 1
1
GF𝑛;𝑎b →𝑛; (𝓉) = (1−[𝑥 + ]𝑎1 ·𝓉) 2 (1−[𝑥 − ]𝑎1 ·𝓉) 2
1 𝑎[
,𝑎 1 2 𝑚 1 +𝐷 −𝑛+1
( 𝐷 −𝑛+4 !)
𝓉𝑚1 +1 2, ,𝑚 1 +𝐷 −𝑛+1 [𝑥 − −𝑥 + ]𝑎1 ,𝑎2 ·𝓉

JHEP02(2024)158
× · 𝐹 2 ,
(1+[𝑥 + ]𝑎1 ,𝑎2 ·𝓉)𝑚1 +𝐷−𝑛+1 3 2 𝐷 −𝑛+3,𝑚 +𝐷 −𝑛+2 1−[𝑥 + ]𝑎1 ,𝑎2 ·𝓉
1

(1)(𝑚 ) (1)(𝑚 )
1
GF𝑛;𝑎b →𝑛;𝑎[
(𝓉) = GF𝑛;𝑎b →𝑛;
1
𝑎[
(𝓉) ,
2 ,𝑎 1 2 ,𝑎 1 1 2 𝑎 1 ↔𝑎 2

(2)(𝑚 ) (2)(𝑚 )
1
GF𝑛;𝑎b →𝑛;𝑎[
(𝓉) = GF𝑛;𝑎b →𝑛;
1
𝑎[
(𝓉) . (5.6)
2 ,𝑎 1 2 ,𝑎 1 1 2 𝑎 1 ↔𝑎 2

(1)
Each part involves only one generalized hypergeometric function. The constants [𝐶𝑎2 (𝑛 −
(1) (2) (2)
1; 𝑚 1 )]𝑎1 , [𝐶𝑎1 (𝑛 − 1; 𝑚 1 )]𝑎2 , [𝐶𝑎2 (𝑛 − 1; 𝑚 1 )]𝑎1 , and [𝐶𝑎1 (𝑛 − 1; 𝑚 1 )]𝑎2 can be factored out
of the integral sign. By this separation, the integral in equation (5.3) can be divided into
several parts. We can calculate each part separately and add it up. Next we will use the
following part as an example to illustrate the process:

Z 𝓉
𝑛−𝐷 𝑛−𝐷
𝓊−1+𝐷−𝑛 (1 − 𝑥 + · 𝓊) 2 (1 − 𝑥 − · 𝓊) 2
0

′(1)(𝑚 ) (5.7)
n  
(1)(𝑚 )
× 2𝑌 (𝑎1 ) 𝓊2 · GF𝑛;𝑎b →𝑛;\
1
𝑎 ,𝑎
(𝓊) + 𝓊 · GF𝑛;𝑎b →𝑛;\
1
𝑎 ,𝑎
(𝓊)
1 1 2 1 1 2

′(1)(𝑚 )
 o
(1)(𝑚 )
+ 𝑋 (𝑎1 ) (𝐷 − 𝑛) · GF𝑛;𝑎b →𝑛;\
1
𝑎 ,𝑎
(𝓊) + 𝓊 · GF𝑛;𝑎b →𝑛;\
1
𝑎 ,𝑎
(𝓊) 𝑑𝓊.
1 1 2 1 1 2

The main difference from what we did in section 4 is that we use another derivative
formula of the generalized hypergeometric function

!
𝑑 𝑎 1, 𝑎 2, · · · , 𝑎𝑝−1, 𝛼
𝑝 𝐹𝑞 𝑧
𝑑𝑧 𝑏 1, 𝑏 2, · · · , 𝑏𝑞−1, 𝛼 + 1
! !! (5.8)
𝛼 𝑎 1, 𝑎 2, · · · , 𝑎𝑝−1 𝑎 1, 𝑎 2, · · · , 𝑎𝑝−1, 𝛼
= · (𝑝−1) 𝐹 (𝑞−1) 𝑧 − 𝑝 𝐹𝑞 𝑧 .
𝑧 𝑏 1, 𝑏 2, · · · , 𝑏𝑞−1 𝑏 1, 𝑏 2, · · · , 𝑏𝑞−1, 𝛼 + 1

After changing the integration variable from 𝓊 to

𝓊
𝓊 → 𝓍𝑎1,𝑎2 (𝓊) = , (5.9)
1 − [𝑥 + ]𝑎1,𝑎2 · 𝓊

– 13 –
the integral (5.7) becomes

Z 𝓍𝑎 ,𝑎 (𝓉)
1 2
𝑑𝑥 𝑥 𝐷−𝑛+𝑚1
0
((
𝐷−𝑛 𝐷−𝑛−1
(1+A1 (𝑎 1 ;𝑎 2 )𝑥 +A2 (𝑎 1 ;𝑎 2 )𝑥 2 )− 2 (1+B1 (𝑎 1 ;𝑎 2 )𝑥 +B2 (𝑎 1 ;𝑎 2 )𝑥 2 ) 2

!)
1, 𝐷−𝑛+2
×(Q0 (𝑎 1 ;𝑎 2 )+Q1 (𝑎 1 ;𝑎 2 )𝑥)· 2 𝐹 1 2 [𝑥 − −𝑥 + ]𝑎1,𝑎2 ·𝑥
𝐷 −𝑛+2
(

JHEP02(2024)158
𝐷−𝑛 𝐷−𝑛−3
+ (1+A1 (𝑎 1 ;𝑎 2 )𝑥 +A2 (𝑎 1 ;𝑎 2 )𝑥 2 )− 2 (1+B1 (𝑎 1 ;𝑎 2 )𝑥 +B2 (𝑎 1 ;𝑎 2 )𝑥 2 ) 2

 
(𝐷 −𝑛−1) P1 (𝑎 1 ;𝑎 2 )𝑥 +P2 (𝑎 1 ;𝑎 2 )𝑥 2
!))
1, 𝐷−𝑛+2 ,𝑚 1 +𝐷 −𝑛+1
× · 3 𝐹2 2 [𝑥 − −𝑥 + ]𝑎1,𝑎2 ·𝑥 ,
2(𝐷 −𝑛+1+𝑚 1 ) 𝐷 −𝑛+2,𝑚 1 +𝐷 −𝑛+2
(5.10)
where coefficients are

A1 (𝑎 1 ; 𝑎 2 ) = −(𝑥 + + 𝑥 − ) + 2[𝑥 + ]𝑎1,𝑎2 ,

A2 (𝑎 1 ; 𝑎 2 ) = [𝑥 +
2
]𝑎1,𝑎2 + 𝑥 + · 𝑥 − − 2(𝑥 + + 𝑥 − )[𝑥 + ]𝑎1,𝑎2 ,
B1 (𝑎 1 ; 𝑎 2 ) = −[𝑥 + + 𝑥 − ]𝑎2 + 2[𝑥 + ]𝑎1,𝑎2 ,
B2 (𝑎 1 ; 𝑎 2 ) = [𝑥 +
2
]𝑎1,𝑎2 + [𝑥 + · 𝑥 − ]𝑎2 − 2[𝑥 + + 𝑥 − ]𝑎2 · [𝑥 + ]𝑎1,𝑎2 ,
 
P1 (𝑎 1 ; 𝑎 2 ) = − [𝑥 + + 𝑥 − ]𝑎2 𝑋 (𝑎2 ) + 4𝑌 (𝑎2 ) ,
P2 (𝑎 1 ; 𝑎 2 ) = 2𝑌 (𝑎2 ) ([𝑥 + + 𝑥 − ]𝑎2 − 2[𝑥 + ]𝑎1,𝑎2 ) + 𝑋 (𝑎2 ) (2[𝑥 + · 𝑥 − ]𝑎2 − [𝑥 + + 𝑥 − ]𝑎2 [𝑥 + ]𝑎1,𝑎2 ),
Q0 (𝑎 1 ; 𝑎 2 ) = 𝑋 (𝑎2 ),
Q1 (𝑎 1 ; 𝑎 2 ) = 2𝑌 (𝑎2 ) + [𝑥 + ]𝑎1,𝑎2 · 𝑋 (𝑎2 ) .
(5.11)
Note that labels 𝑎 1, 𝑎 2 are no longer symmetric in the above coefficients. By equation (4.12),
we can expand the integral (5.10) as follows:

∞ Z 𝓍𝑎 ,𝑎 (𝓉)
!
1 2
n 1, 𝐷−𝑛+2
(0)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 ) · 2 𝐹 1 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝑥
X
𝑚 1 +𝑚 2 +𝐷−𝑛 2
𝑥
𝑚 2 =0 0
𝐷 −𝑛 +2
! (5.12)
1, 𝐷−𝑛+2 , 𝑚1 + 𝐷 − 𝑛 + 1 o
+ (1)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 ) · 3 𝐹 2 2 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝑥 𝑑𝑥,
𝐷 − 𝑛 + 2, 𝑚 1 + 𝐷 − 𝑛 + 2

where the expansion coefficients are

 
(1) (1)
(𝐷 − 𝑛 − 1) · P1 (𝑎 1 ; 𝑎 2 ) · N𝑚2 −1 (𝑎 1 ; 𝑎 2 ) + P2 (𝑎 1 ; 𝑎 2 ) · N𝑚2 −2 (𝑎 1 ; 𝑎 2 )
(1)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2, 𝑚 1 ) = ,
2(𝐷 − 𝑛 + 1 + 𝑚 1 )
(0)
(0)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 ) = Q0 (𝑎 1 ; 𝑎 2 ) · N𝑚
(0)
2
(𝑎 1 ; 𝑎 2 ) + Q1 (𝑎 1 ; 𝑎 2 ) · N𝑚2 −1 (𝑎 1 ; 𝑎 2 ).
(5.13)
(1) (2)
The N𝑚′ (𝑎 1 ; 𝑎 2 ) and N𝑚′ (𝑎 1 ; 𝑎 2 ) come from the Taylor expansion of terms like (1 + 𝐴1𝑥 +

– 14 –
𝐴2𝑥 2 )𝑐 1 (1 + 𝐵 1𝑥 + 𝐵 2𝑥 2 )𝑐 2 ,which are
𝐷−𝑛 𝐷−𝑛−3
(1 + A1 (𝑎 1 ; 𝑎 2 )𝑥 + A2 (𝑎 1 ; 𝑎 2 )𝑥 2 )− 2 (1 + B1 (𝑎 1 ; 𝑎 2 )𝑥 + B2 (𝑎 1 ; 𝑎 2 )𝑥 2 ) 2

∞
(1) ′
= N𝑚′ (𝑎 1 ; 𝑎 2 ) · 𝑥 𝑚 ,
X

𝑚 ′ =0
𝐷−𝑛 𝐷−𝑛−1
(1 + A1 (𝑎 1 ; 𝑎 2 )𝑥 + A2 (𝑎 1 ; 𝑎 2 )𝑥 2 )− 2 (1 + B1 (𝑎 1 ; 𝑎 2 )𝑥 + B2 (𝑎 1 ; 𝑎 2 )𝑥 2 ) 2

∞
(0) ′
= N𝑚′ (𝑎 1 ; 𝑎 2 ) · 𝑥 𝑚 . (5.14)
X

𝑚 ′ =0

By formula (4.12), the Taylor expansion coefficients are

JHEP02(2024)158
′ −𝑙
′ ⌊ 𝑙 ⌋ ⌊𝑚 ⌋
(1)
𝑚 X
2 2
(− 𝐷−𝑛
2 − 𝑙 + 𝑖 + 1)(𝑙−𝑖) (
𝐷−𝑛−3
− 𝑚 ′ + 𝑙 + 𝑗 + 1)(𝑚′ −𝑙−𝑗)
N𝑚′ (𝑎 1 ; 𝑎 2 ) =
X X
2

𝑙=0 𝑖=0 𝑗=0

(𝑙 − 2𝑖)! · 𝑖!(𝑚 ′ − 𝑙 − 2𝑗)! · 𝑗!
′
× A1 (𝑎 1 ; 𝑎 2 )𝑙−2𝑖 A2 (𝑎 1 ; 𝑎 2 )𝑖 B1 (𝑎 1 ; 𝑎 2 )𝑚 −𝑙−2𝑗 B2 (𝑎 1 ; 𝑎 2 ) 𝑗 ,
′ −𝑙
′ ⌊ 𝑙 ⌋ ⌊𝑚 ⌋
(0)
𝑚 X
2 2
(− 𝐷−𝑛
2 − 𝑙 + 𝑖 + 1)(𝑙−𝑖) (
𝐷−𝑛−1
− 𝑚 ′ + 𝑙 + 𝑗 + 1)(𝑚′ −𝑙−𝑗)
N𝑚′ (𝑎 1 ; 𝑎 2 ) =
X X
2

𝑙=0 𝑖=0 𝑗=0

′ (𝑙 − 2𝑖)! · 𝑖! · (𝑚 − 𝑙 − 2𝑗)! · 𝑗!
′
× A(𝑎 1 ; 𝑎 2 )𝑙−2𝑖
1 A2 (𝑎 1 ; 𝑎 2 )𝑖 B1 (𝑎 1 ; 𝑎 2 )𝑚 −𝑙−2𝑗 B2 (𝑎 1 ; 𝑎 2 ) 𝑗 . (5.15)
(1) (0)
Again, we need to emphasize label 𝑎 1, 𝑎 2 are not symmetric in N𝑚′ (𝑎 1 ; 𝑎 2 ), N𝑚′ (𝑎 1 ; 𝑎 2 )
and (1)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2, 𝑚 1 ), (0)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2, 𝑚 1 ). By using integral formula (4.6), we can calculate
the integral (5.7). Eventually, by summing everything up, we obtain the final generating
function of 𝑛-gon to (𝑛 − 2)-gon as

𝑎 1 ,𝑎 2 (𝓉)
GF𝑛→𝑛;\
−2+𝐷−𝑛 −2+𝐷−𝑛
= (1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2

∞
1
(
X 𝓉𝑚1 +𝑚2 +2
× ·
𝑚 1 ,𝑚 2 =0
𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 1 (1 − [𝑥 + ]𝑎1,𝑎2 · 𝓉)𝑚1 +𝑚2 +𝐷−𝑛+1
n 
× (1)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 )[C𝑎(1)
1
(𝑛 − 1; 𝑚 1 )]𝑎2 + (1)𝑚2 (𝑛; 𝑎 2 ; 𝑎 1 ; 𝑚 1 )[C𝑎(1)
2
(𝑛 − 1; 𝑚 1 )]𝑎1
, 𝑚 1 + 𝐷 − 𝑛 + 1, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 1 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝓉
!
1, 𝐷−𝑛+2
2
× 4 𝐹3
𝐷 − 𝑛 + 2, 𝑚 1 + 𝐷 − 𝑛 + 2, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 2 1 − [𝑥 + ]𝑎1,𝑎2 · 𝓉
 
+ (1)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 )[C𝑎(2)
1
(𝑛 − 1; 𝑚 1 )]𝑎2 + (1)𝑚2 (𝑛; 𝑎 2 ; 𝑎 1 ; 𝑚 1 )[C𝑎(2)
2
(𝑛 − 1; 𝑚 1 )]𝑎1
, 𝑚 1 + 𝐷 − 𝑛 + 1, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 1 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝓉
!
2, 𝐷−𝑛+4
2
× 4 𝐹3
𝐷 − 𝑛 + 3, 𝑚 1 + 𝐷 − 𝑛 + 2, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 2 1 − [𝑥 + ]𝑎1,𝑎2 · 𝓉
 
+ (0)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 )[C𝑎(1)
1
(𝑛 − 1; 𝑚 1 )]𝑎2 + (0)𝑚2 (𝑛; 𝑎 2 ; 𝑎 1 ; 𝑚 1 )[C𝑎(1)
2
(𝑛 − 1; 𝑚 1 )]𝑎1
, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 1 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝓉
!
1, 𝐷−𝑛+2
2
× 3 𝐹2
𝐷 − 𝑛 + 2, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 2 1 − [𝑥 + ]𝑎1,𝑎2 · 𝓉
 
+ (0)𝑚2 (𝑛; 𝑎 1 ; 𝑎 2 ; 𝑚 1 )[C𝑎(2)
1
(𝑛 − 1; 𝑚 1 )]𝑎2 + (0)𝑚2 (𝑛; 𝑎 2 ; 𝑎 1 ; 𝑚 1 )[C𝑎(2)
2
(𝑛 − 1; 𝑚 1 )]𝑎1

, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 1 [𝑥 − − 𝑥 + ]𝑎1,𝑎2 · 𝓉
! )
2, 𝐷−𝑛+4 o
× 3 𝐹2 2 . (5.16)
𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 2 1 − [𝑥 + ]𝑎1,𝑎2 · 𝓉

– 15 –
6 Generating function of 𝒏-gon to (𝒏 − 𝒌)-gon

From (4.15) and (5.16), we found GF𝑛→𝑛;\ 𝑎 1 ,𝑎 2 (𝓉) has an analogous functional form with
GF𝑛→𝑛;𝑎b1 (𝓉). This suggests that the generating function GF𝑛→𝑛;Ib (𝓉) (where the label list
𝑘
I𝑘 = {𝑎 1, 𝑎 2, . . . , 𝑎𝑘 }) should also have the same functional form. In this section, we will
provide the form of GF𝑛→𝑛;Ib (𝓉) directly. A proof by induction will be given in the next
𝑘
section to show it does satisfy the recursion relation (2.10).
Before that, we provide some notations to represent generalized hypergeometric functions.
First, we construct an array including 𝑘 elements{𝑏 1, 𝑏 2, . . . , 𝑏𝑘 }with each 𝑏𝑖 is either 0 or

JHEP02(2024)158
1. For example, for 𝑘 = 4, the array can be one of

{0, 0, 0, 0}, {0, 0, 0, 1}, {0, 0, 1, 0}, {0, 0, 1, 1}, {0, 1, 0, 0}, {0, 1, 0, 1}, {0, 1, 1, 0}, {0, 1, 1, 1},
(6.1)
{1, 0, 0, 0}, {1, 0, 0, 1}, {1, 0, 1, 0}, {1, 0, 1, 1}, {1, 1, 0, 0}, {1, 1, 0, 1}, {1, 1, 1, 0}, {1, 1, 1, 1}.

Now we can construct two types of generalized hypergeometric functions based on the array
as follows:
!
1, 𝐷−𝑛+𝑘 ,𝑆 1 −𝑏 1,𝑆 2 −𝑏 2, · · · ,𝑆𝑘 −𝑏𝑘
HG1 (𝑛,𝑘; {𝑏 1, · · · ,𝑏𝑘 };𝑧) =(2+𝑘) 𝐹 (1+𝑘) 2 𝑧 ,
𝐷 −𝑛+𝑘,𝑆 1,𝑆 2, · · · ,𝑆𝑘
! (6.2)
2, 𝐷−𝑛+𝑘+2 ,𝑆 1 −𝑏 1,𝑆 2 −𝑏 2, · · · ,𝑆𝑘 −𝑏𝑘
HG2 (𝑛,𝑘; {𝑏 1, · · · ,𝑏𝑘 };𝑧) =(2+𝑘) 𝐹 (1+𝑘) 2 𝑧 ,
𝐷 −𝑛+𝑘 +1,𝑆 1,𝑆 2, · · · ,𝑆𝑘

where
𝑖
𝑆𝑖 = 𝑚 𝑗 + 𝐷 − 𝑛 + 𝑘. (6.3)
X

𝑗=1

For example,

HG1 (𝑛, 3; {0, 0, 1}; 𝑧)

!
1, 𝐷−𝑛+3 , 𝑚 1 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 2
=5 𝐹 4 2 𝑧
𝐷 − 𝑛 + 3, 𝑚 1 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 3
!
1, 𝐷−𝑛+3 , 𝑚1 + 𝑚2 + 𝑚3 + 𝐷 − 𝑛 + 2
=3 𝐹 2 2 𝑧 ,
𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 3

HG2 (𝑛, 3; {1, 0, 1}; 𝑧)

!
2, 𝐷−𝑛+5 , 𝑚 1 + 𝐷 − 𝑛 + 2, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 2
=5 𝐹 4 2 𝑧
𝐷 − 𝑛 + 4, 𝑚 1 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 3
!
2, 𝐷−𝑛+5 , 𝑚 1 + 𝐷 − 𝑛 + 2, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 2
=4 𝐹 3 2 𝑧 .
𝐷 − 𝑛 + 4, 𝑚 1 + 𝐷 − 𝑛 + 3, 𝑚 1 + 𝑚 2 + 𝑚 3 + 𝐷 − 𝑛 + 3
(6.4)

– 16 –
 Then generating function GF𝑛→𝑛;Ib is
𝑘
−2+𝐷−𝑛 −2+𝐷−𝑛
GF𝑛→𝑛;Ib (𝓉) = (1 − 𝑥 + · 𝓉) 2 · (1 − 𝑥 − · 𝓉) 2
𝑘
P𝑘
∞
1
(
X 𝓉 𝑖=1 𝑚𝑖 +𝑘
× ·
𝑖=1 𝑚𝑖 + 𝐷 − 𝑛 + 𝑘 − 1 (1 − [𝑥 + ]I𝑘 · 𝓉) 𝑖=1 𝑖
P𝑘 P𝑘
𝑚 +𝐷−𝑛+𝑘−1
𝑚 1 ,...,𝑚𝑘 =0
1
(𝑎 ′ ,...,𝑎 ′ )
C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛)
X X
×
{𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘 ), 𝑏 1 ,𝑏 2 ,...,𝑏𝑘−1 =0
n
(1)
× [C𝑎′ (𝑛 − 𝑘 + 1; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ · HG1 (𝑛, 𝑘; {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1, 1}; 𝒲I𝑘 (𝓉))

JHEP02(2024)158
1
)
o
(2)
+ [C𝑎′ (𝑛 − 𝑘 + 1; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ · HG2 (𝑛, 𝑘; {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1, 1}; 𝒲I𝑘 (𝓉)) ,
1

(6.5)
where
[𝑥 − − 𝑥 + ]I𝑘 · 𝓉
𝒲I𝑘 (𝓉) =
. (6.6)
1 − [𝑥 + ]I𝑘 · 𝓉
The first summation in the penultimate line of (6.5) runs over all the permutations of
I𝑘 = {𝑎 1, 𝑎 2, . . . , 𝑎𝑘 }. For example for 𝑘 = 3, this summation should run over
n o
{𝑎 ′1, 𝑎 ′2, 𝑎 ′3 } = {𝑎 1, 𝑎 2, 𝑎 3 }, {𝑎 1, 𝑎 3, 𝑎 2 }, {𝑎 2, 𝑎 1, 𝑎 3 }, {𝑎 2, 𝑎 3, 𝑎 1 }, {𝑎 3, 𝑎 1, 𝑎 2 }, {𝑎 3, 𝑎 2, 𝑎 1 } .
(6.7)
It’s important to note that the second summation on that line does not include 𝑏𝑘 . For
example for 𝑘 = 4, this summation runs over without 𝑏 4 but
n o
{𝑏 1, 𝑏 2, 𝑏 3 } = {0, 0, 0}, {0, 0, 1}, {0, 1, 0}, {0, 1, 1}, {1, 0, 0}, {1, 0, 1}, {1, 1, 0}, {1, 1, 1} . (6.8)
(𝑎 ′ ,...,𝑎 ′ )
For a given permutation (𝑎 ′1, 𝑎 ′2 . . . , 𝑎𝑘′ ) and a fixed {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1 }, the C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛) in
the penultimate line is constructed as
(𝑎 ′ ,...,𝑎 ′ )
C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛)
= [(b1 )𝑚2 (𝑛 − 𝑘 + 2; 1; 𝑎 ′1 ; 𝑎 ′2 ; 𝑚 1 )]𝑎′3,...,𝑎𝑘′
· [(b2 )𝑚3 (𝑛 − 𝑘 + 3; 2; {𝑎 ′1, 𝑎 ′2 }; 𝑎 ′3 ; 𝑚 1 + 𝑚 2 )]𝑎′4,...,𝑎𝑘′
··· (6.9)
· [(b𝑖 )𝑚𝑖+1 (𝑛 − 𝑘 + 𝑖 + 1; 𝑖; {𝑎 ′1, . . . , 𝑎𝑖′ }; 𝑎𝑖+1
′
; 𝑚 1 + · · · + 𝑚𝑖 )]𝑎𝑖+2
′ ,...,𝑎𝑘′

···
· (b𝑘−1 )𝑚𝑘 (𝑛; 𝑘 − 1; {𝑎 ′1, . . . , 𝑎𝑘−1
′
}; 𝑎𝑘′ ; 𝑚 1 + . . . + 𝑚𝑘−1 ).
(𝑎 ′ ,...,𝑎 ′ )
For 𝑘 = 4, {𝑏 1, 𝑏 2, 𝑏 3 } = {0, 1, 0} as an example, the C{0,1,0}
1 4
(𝑛) is6
(𝑎 ′ ,...,𝑎 ′ )
1
C{0,1,0} 4
(𝑛) = [(0)𝑚2 (𝑛 − 2; 1; 𝑎 ′1 ; 𝑎 ′2 ; 𝑚 1 )]𝑎′3,𝑎′4 · [(1)𝑚3 (𝑛 − 1; 2; {𝑎 ′1, 𝑎 ′2 }; 𝑎 ′3 ; 𝑚 1 + 𝑚 2 )]𝑎′4
· (0)𝑚4 (𝑛; 3; {𝑎 ′1, 𝑎 ′2, 𝑎 ′3 }; 𝑎 ′4 ; 𝑚 1 + 𝑚 2 + 𝑚 3 ).
(6.10)
6
Remember notation [∆]b means we add a subscript b on each term in ∆ with form (𝐴𝐵) or (𝐴𝐵)a . As we
explained in second section.

– 17 –
There is a slight difference between the definitions of 0 and 1 when compared to the definitions
in formal section (5.13). Here, it includes one more variable 𝑘 ′ , defined as
 
(1) (1)
(𝐷 − 𝑛 ′ − 1) · P1 (a; 𝑏) · N𝑚′ −1 (a; 𝑏) + P2 (a; 𝑏) · N𝑚′ −2 (a; 𝑏)
(1)𝑚′ (𝑛 ′ ; 𝑘 ′ ; a; 𝑏; 𝑠𝑢𝑚) = ,
2(𝐷 − 𝑛 ′ + 𝑘 ′ + 𝑠𝑢𝑚)
(0) (0)
(0)𝑚′ (𝑛 ′ ; 𝑘 ′ ; a; 𝑏; 𝑠𝑢𝑚) = Q0 (a; 𝑏) · N𝑚′ (a; 𝑏) + Q1 (a; 𝑏) · N𝑚′ −1 (a; 𝑏).
(6.11)
The variables 𝑛′ , 𝑘 ′, 𝑚′ ,
and 𝑠𝑢𝑚 must be non-negative integers. The variable a can be
fixed either a single label or a list of labels, while 𝑏 solely represents a single label. Upon
setting 𝑚 ′ = 𝑚 2 , 𝑛 ′ = 𝑛, 𝑘 = 1, a = 𝑎 1 , 𝑏 = 𝑎 2 , and 𝑠𝑢𝑚 = 𝑚 1 , we revert to the form (5.13) as

JHEP02(2024)158
outlined in the previous section. The other coefficients P1 (a; 𝑏), P2 (a; 𝑏), Q0 (a; 𝑏), Q1 (a; 𝑏),
N𝑚′ (a; 𝑏) are defined by merely replacing all instances of the labels 𝑎 1 and 𝑎 2 in (5.11),
and (5.15) with variables a and 𝑏 respectively. We have thus completed the explanation of
every term in GF𝑛→𝑛;Ib (𝓉) (6.5). Furthermore, if we additionally define C𝑎∅1 ≡ 1 for 𝑘 = 1,
𝑘
we find that the function (6.5) aligns with GF𝑛→𝑛;𝑎b1 (𝓉)(4.15) and GF𝑛→𝑛;\ 𝑎 1 ,𝑎 2 (𝓉)(5.16).

6.1 Analytic example

6.1.1 Triangle to Tadpole
Now we present the reduction of an tensor triangle to tadpole GF3→3:2c ,3
as an example to
illustrate the analyticity. The generating function can be obtain from (6.5) by selecting
𝑛 = 3, 𝑘 = 2, 𝑎 1 = 2, 𝑎 2 = 3. Expanding the generating function as a series of 𝓉, we can obtain:
∞
(𝑟 )
(𝓉) = (6.12)
X
GF3→3;2c 𝓉𝑟 · 𝐶 ,
,3 3→3;2,3
c
𝑟 =0

(𝑟 )
where 𝐶 are the reduction coefficients of tensor triangle
3→3;2,3
c

𝑑𝐷𝑙 (2𝑅 · 𝑙)𝑟

Z
(𝑟 )
𝐼3 = , (6.13)
𝑖𝜋 𝐷/2 ((𝑙 − 𝑞 1 )2 − 𝑀12 )((𝑙 − 𝑞 2 )2 − 𝑀32 )((𝑙 − 𝑞 3 )2 − 𝑀32 )

to the master integral

𝑑𝐷𝑙 1
Z
(0)
𝐼1 = . (6.14)
𝑖𝜋 𝐷/2 (𝑙 − 𝑞 1 )2 − 𝑀12

Moreover, we list the first two non-zero orders in the expansion:

(2)
𝐶 = 𝑋 (2) · [𝑋 (3) ]2 + 𝑋 (3) · [𝑋 (2) ]3,
3→3;2
c,3

1
(
 
(3)
𝐶 = 𝐷 𝑋 (2) · [𝑋 (3) ]2 · [𝑥 + + 𝑥 − ]2 + 𝑋 (3) · [𝑋 (2) ]3 · [𝑥 + + 𝑥 − ]3
3→3;2
c,3 2(𝐷 − 1)
 
+ 𝑋 (2) · [𝑋 (3) ]2 + 𝑋 (3) · [𝑋 (2) ]3 (𝐷 − 1) · [𝑥 + + 𝑥 − ]2,3 + (𝐷 + 1) · (𝑥 + + 𝑥 − )

)
   
+8 𝑌 (2)
· [𝑋 (3)
]2 + 𝑌 (3)
· [𝑋 (2)
]3 + 4 𝑋 (2)
· [𝑌 (3)
]2 + 𝑋 (3)
· [𝑌 (2)
]3 .

(6.15)

– 18 –
6.1.2 Reduction coefficients of 𝒏-gon to (𝒏 − 𝒌)-gon

In addition, for the general reduction of an 𝑛-gon to an (𝑛 − 𝑘)-gon for 𝑘 ≥ 1, we expand

∞
(𝑟 )
GF𝑛→𝑛;Ib (𝓉) = (6.16)
X
𝓉𝑟 · 𝐶 .
𝑘 𝑛→𝑛;Ib𝑘
𝑟 =0

The reduction coefficients are

JHEP02(2024)158
P𝑘
( 𝑚 +𝑘≤𝑟
(𝑟 ) (𝑙 )
𝑖=1 X𝑖 (𝑙 )
= N2 2 (𝑚 1, . . . , 𝑚𝑘 )
X
𝐶 N1 1
𝑛→𝑛;Ib𝑘
𝑙 1 +𝑙 2 +𝑙 3 +
P𝑘
𝑚𝑖 +𝑘=𝑟 𝑚 1 ,...,𝑚𝑘 =0
𝑖=1
1
(𝑎 ′ ,...,𝑎 ′ )
C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛)
X X
×
(6.17)
{𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘 ), 𝑏 1 ,𝑏 2 ,...,𝑏𝑘−1 =0

(1) (𝑙 )
× [C𝑎′ (𝑛 − 𝑘 + 1; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ · M1;{𝑏
3
1 ,𝑏 2 ,...,𝑏𝑘−1 }
(𝑚 1, . . . , 𝑚𝑘 )
1
!)

(2) (𝑙 3 )
+ [C𝑎′ (𝑛 − 𝑘 + 1; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ · M2;{𝑏 1 ,𝑏 2 ,...,𝑏𝑘−1 }
(𝑚 1, . . . , 𝑚𝑘 ) .
1

The sources of those coefficients are shown as follows:

(𝑙 )
• N1 1 comes from

−2+𝐷−𝑛 −2+𝐷−𝑛
(1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2

∞
(−𝑥 + )𝑖 ( −2+𝐷−𝑛 − 𝑖 + 1)𝑖 (−𝑥 − )𝑙1 −𝑖 ( −2+𝐷−𝑛 − (𝑙 1 − 𝑖) + 1)(𝑙1 −𝑖) 𝑙1
𝑙1
!
=
X X
2 2
· 𝓉
𝑙 1 =0 𝑖=0
𝑖! (𝑙 1 − 𝑖)!
∞
(𝑙 )
=
X
N1 1 · 𝓉𝑙1 .
𝑙 1 =0
(6.18)

(𝑙 )
• N2 2 (𝑚 1, . . . , 𝑚𝑘 ) comes from

P𝑘
1 𝓉 𝑖=1 𝑚𝑖 +𝑘
𝑖=1 𝑚𝑖 + 𝐷 − 𝑛 + 𝑘 − 1 (1 − [𝑥 + ]I · 𝓉) 𝑖=1 𝑖
P𝑘 P𝑘
𝑚 +𝐷−𝑛+𝑘−1
𝑘
   
(−[𝑥 + ]I𝑘 )𝑙2 −( 𝑖=1 𝑚𝑖 + 𝐷 − 𝑛 + 𝑘 − 1) − 𝑙 2 + 1
P𝑘
∞ P𝑘
= 𝑙 + 𝑚𝑖 +𝑘
X 𝑙2 
·𝓉2
 P  𝑖=1
+ 𝐷 − 𝑛 + 𝑘 − 1 · 𝑙2 !

𝑘
𝑙 2 =0 𝑖=1 𝑚𝑖
∞ P𝑘
(𝑙 )
= N2 2 (𝑚 1, . . . , 𝑚𝑘 ) · 𝓉𝑙2 + 𝑚𝑖 +𝑘
(6.19)
X
𝑖=1 .
𝑙 2 =0

– 19 –
 (𝑙 ) (𝑙 )
• By (3.13), M1;{𝑏
3
1 ,𝑏 2 ,...,𝑏𝑘−1 }
(𝑚 1, . . . , 𝑚𝑘 ) and M2;{𝑏
3
1 ,𝑏 2 ,...,𝑏𝑘−1 }
(𝑚 1, . . . , 𝑚𝑘 ) come from

HG1 (𝑛, 𝑘; {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1, 1}; 𝒲I𝑘 (𝓉))

∞
C𝑙𝑖3 −1 ( 𝐷−𝑛+𝑘 )𝑖 ([𝑥 − − 𝑥 + ]I𝑘 )𝑖 ([𝑥 + ]I𝑘 )𝑙3 −𝑖
𝑙3
!
=1+ · 𝒮𝑖 ({𝑏 1, . . . , 𝑏𝑘−1 }) 𝓉𝑙3
X X
2

𝑙 3 =1 𝑖=1
(𝐷 − 𝑛 + 𝑘)𝑖
∞  
(𝑙 )
= (𝑚 1, . . . , 𝑚𝑘 ) · 𝓉𝑙3 , (6.20)
X
3
M1;{𝑏 1 ,𝑏 2 ,...,𝑏𝑘−1 }
𝑙 3 =0

HG2 (𝑛, 𝑘; {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1, 1}; 𝒲I𝑘 (𝓉))

JHEP02(2024)158
∞
C𝑙𝑖3 −1 ( 𝐷−𝑛+𝑘+2 )𝑖 ([𝑥 − − 𝑥 + ]I𝑘 )𝑖 ([𝑥 + ]I𝑘 )𝑙3 −𝑖
𝑙3
!
=1+ · 𝒮𝑖 ({𝑏 1, . . . , 𝑏𝑘−1 }) 𝓉𝑙3
X X
2

𝑙 3 =1 𝑖=1
(𝐷 − 𝑛 + 𝑘 + 1)𝑖
∞  
(𝑙 )
= (𝑚 1, . . . , 𝑚𝑘 ) · 𝓉𝑙3 , (6.21)
X
3
M2;{𝑏 1 ,𝑏 2 ,...,𝑏𝑘−1 }
𝑙 3 =0

where the factor 𝒮𝑖 ({𝑏 1, . . . , 𝑏𝑘−1 }) are defined as

𝒮𝑖 ({𝑏 1, . . . , 𝑏𝑘−1 })
Q  P 
𝛼=1 ( 𝛽=1 𝑚 𝛽 + 𝐷 − 𝑛 + 𝑘 − 𝑏𝛼 )𝑖 +𝐷 −𝑛 +𝑘 −1 (6.22)
𝑘−1 P𝛼 𝑘
𝛽=1 𝑚 𝛽
= Q  P  𝑖
.
𝛼=1 ( 𝛽=1 𝑚 𝛽 + 𝐷 − 𝑛 + 𝑘)𝑖 +𝐷 −𝑛 +𝑘
𝑘−1 P𝛼 𝑘
𝛽=1 𝑚 𝛽 𝑖

7 Proof

In this section, we will provide an inductive proof of parameter 𝑘 to verify the correctness
of (6.5). The main methodology echoes the computation of GF𝑛→𝑛;\ 𝑎 1 ,𝑎 2 (𝓉). Suppose that
expression (6.5) holds for 𝑘. We can evaluate the generating function for reduction of an
𝑛-gon to (𝑛 − (𝑘 + 1))-gon by solving the differential equation (2.16). After writing down the
general solution, we divided the integration into several parts according to the generalized
hypergeometric function. During the steps of integration, we firstly alter the integration
variable. Subsequently, we conduct the Taylor expansion on a part of the integrand function.
Lastly, by selecting the suitable integration formula (4.6), we affirm that the statement (6.5)
holds true for 𝑘 + 1. Readers who are already familiar with this process may choose to
skip this section. The derivation and integration rules that we employ during our proof are
outlined in (5.8) and (4.6). In the context of (6.2), these two rules transform into:
𝑑
HG𝑙 (𝑛−1;𝑘; {𝑏 1, . . .,𝑏𝑘−1, 1};𝑧)
𝑑𝑥
𝑗=1 𝑚 𝑗 +𝐷 −𝑛−𝑘
P𝑘  
= HG𝑙 (𝑛−1;𝑘; {𝑏 1, . . .,𝑏𝑘−1, 0};𝑧)−HG𝑙 (𝑛−1;𝑘; {𝑏 1, . . .,𝑏𝑘−1, 1};𝑧) ,
𝑧
for 𝑙 = 1, 2,
(7.1)
and
Z P𝑘+1
𝑚 𝑗 +𝐷−𝑛−𝑘−1
𝑑𝑧 𝑧 𝑗 =1 · HG𝑙 (𝑛 − 1; 𝑘; {𝑏 1, . . . , 𝑏𝑘−1, 𝑙}; ℎ · 𝑧)
P𝑘+1
𝑚 +𝐷−𝑛−𝑘 (7.2)
𝑧 𝑗 =1 𝑗
= P𝑘+1 · HG𝑙 (𝑛; 𝑘 + 1; {𝑏 1, . . . , 𝑏𝑘−1, 𝑙, 1}; ℎ · 𝑧), for 𝑙 = 1, 2.
𝑗=1 𝑚 𝑗 + 𝐷 − 𝑛 − 𝑘

– 20 –
 Supposing generating function of 𝑛-gon to (𝑛 − 𝑘)-gon GF𝑛→𝑛;I𝑘 (𝓉) satisfies (6.5), then
recursive relation of GF𝑛→𝑛;Id (𝓉) with I𝑘+1 = {𝑎 1, 𝑎 2, . . . , 𝑎𝑘+1 } is
𝑘+1

(𝑉 𝐿) 𝑅 2 − (𝑉𝑉 ) 2
!
(𝐷 − 𝑛 − 1) − 2(𝐷 − 𝑛) · 𝓉−4· 𝓉 GF𝑛→𝑛;Id (𝓉)
(𝐿𝐿) (𝐿𝐿) 𝑘+1

(𝑉 𝐿) 2 𝑅 2 − (𝑉𝑉 ) 3
!
+ 𝓉−4· 𝓉 −4· 𝓉 GF′ (𝓉)
(𝐿𝐿) (𝐿𝐿) 𝑛→𝑛;Id
𝑘+1

𝑘+1
(   (7.3)
= 2𝑌 GF′ (𝓉) + 𝓉 · GF𝑛;𝑎b →𝑛;Id (𝓉)
X
(𝑎𝑖 ) 3 2
𝓉 ·
𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1 𝑖 𝑘+1

JHEP02(2024)158
𝑎𝑖 ,𝑖=1
 )
+𝑋 (𝑎𝑖 )
(𝐷 − 𝑛)𝓉 · GF𝑛;𝑎b →𝑛;Id (𝓉) + 𝓉 · 2
GF′ (𝓉) .
𝑖 𝑘+1 𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1

The general solution is

−2+𝐷−𝑛 −2+𝐷−𝑛
𝓉 1−𝐷+𝑛 (1 − 𝑥 + · 𝓉) 2 (1 − 𝑥 − · 𝓉) 2
( Z 𝓉
𝑛−𝐷 𝑛−𝐷
× 𝐶1 + 𝓊−1+𝐷−𝑛 (1 − 𝑥 + · 𝓊) 2 (1 − 𝑥 − · 𝓊) 2
0
𝑘+1 n   (7.4)
2𝑌 (𝑎𝑖 ) 𝓊2 · GF′ d (𝓊) + 𝓊 · GF𝑛;𝑎b →𝑛;Id (𝓊)
X
×
𝑛;𝑎b𝑖 →𝑛;I𝑘+1 𝑖 𝑘+1
𝑎𝑖 ,𝑖=1
 o )
+𝑋 (𝑎𝑖 )
(𝐷 − 𝑛) · GF𝑛;𝑎b →𝑛;Id (𝓊) + 𝓊 · GF′ (𝓊) 𝑑𝓊 .
𝑖 𝑘+1 𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1

The undetermined constant 𝐶 1 is determined to be zero, as the generating function must

conform to the requirement of being a Taylor series in terms of 𝓉. Where GF𝑛;𝑎b →𝑛;Id (𝓉) =
𝑖 𝑘+1
[GF𝑛−1→𝑛−1;I \/𝑎
(𝓉)]𝑎𝑖 . We split it into
𝑘+1 𝑖

∞ 1
(𝑎 ′ ,...,𝑎 ′ )
GF𝑛;𝑎b →𝑛;Id (𝓉) = [C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛 − 1)]𝑎𝑖
X X X
𝑖 𝑘+1
𝑚 1 ,...,𝑚𝑘 =0, {𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘+1 /𝑎𝑖 ), 𝑏 1 ,...,𝑏𝑘−1 =0
  (7.5)
(𝑙) (𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1
[C𝑎′ (𝑛 − 𝑘; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ ,𝑎𝑖 · GF (𝓉) ,
X
×
1 𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1
𝑙=1,2

where
(𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1 −1+𝐷−𝑛 −1+𝐷−𝑛
GF (𝓉) = (1−[𝑥 + ]𝑎𝑖 ·𝓉) 2 (1−[𝑥 − ]𝑎𝑖 ·𝓉) 2
𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1
P𝑘
1
(
𝓉 𝑖=1 𝑚𝑖 +𝑘
× P𝑘 ·
𝑖=1 𝑚𝑖 +𝐷 −𝑛+𝑘 (1−[𝑥 + ]I𝑘+1 ·𝓉) 𝑖=1 𝑖
P𝑘
𝑚 +𝐷−𝑛+𝑘
)
×HG𝑙 (𝑛−1,𝑘; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1, 1}; 𝒲I𝑘+1 (𝓉)) , for 𝑙 = 1, 2.

(7.6)

– 21 –
 (𝑎 ′ ,...,𝑎 ′ ) (1)/(2)
Since the coefficients [C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛 − 1)]𝑎𝑖 and [𝐶𝑎′ (𝑛 − 𝑘; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ ,𝑎𝑖 in (7.5) are inde-
1
pendent of 𝓉, they can be pulled out of the integral. Then solution (7.4) can be separated into

−2+𝐷−𝑛 −2+𝐷−𝑛
(1−𝑥 + ·𝓉) 2 (1−𝑥 − ·𝓉) 2

2 𝑘+1 ∞ 1
(𝑎 ′ ,...,𝑎 ′ ) (𝑙)
[C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛−1)]𝑎𝑖 [C𝑎′ (𝑛−𝑘;𝑚 1 )]𝑎′2,...,𝑎𝑘′ ,𝑎𝑖
X X X X X
×
1
𝑙=1 𝑎𝑖 ,𝑖=1 𝑚 1 ,...,𝑚𝑘 =0 {𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘+1 /𝑎𝑖 ) 𝑏 1 ,...,𝑏𝑘−1 =0
( Z 𝓉
𝑛−𝐷 𝑛−𝐷
× 𝓉 1−𝐷+𝑛
𝓊−1+𝐷−𝑛 (1−𝑥 + ·𝓊) 2 (1−𝑥 − ·𝓊) 2
0

JHEP02(2024)158
(  
′(𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1 (𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1
× 2𝑌 (𝑎𝑖 ) 2
𝓊 ·GF (𝓊)+𝓊·GF (𝓊)
𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1 𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1

 ) )
(𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1 ′(𝑙);𝑚 1 ,...,𝑚𝑘 ;𝑏 1 ,...,𝑏𝑘−1
+𝑋 (𝑎𝑖 )
(𝐷 −𝑛)·GF (𝓊)+𝓊·GF (𝓊) 𝑑𝓊 .
𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1 𝑛;𝑎b𝑖 →𝑛;Id
𝑘+1

(7.7)
Next we focus on the part inside the curly braces. After changing the integration variable
𝓊 to 𝓍I𝑘+1 (𝓊):

𝓊
𝓍I𝑘+1 (𝓊) = , (7.8)
1 − [𝑥 + ]I𝑘+1 · 𝓊

and applying derivative formula (7.1), terms inside the big curly braces of (7.7) become

Z 𝓍I𝑘+1 (𝓉) P𝑘
1−𝐷+𝑛 𝑚 𝑗 +𝐷−𝑛+𝑘−1
𝓉 𝑑𝑥 𝑥 𝑗 =1

0
((
× (Q0 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )+Q1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )·𝑥)

− 𝐷−𝑛
× 1+A1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 +A2 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 2 2

𝐷−𝑛−1
× 1+B1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 +B2 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 2 2

o
×HG𝑙 (𝑛−1;𝑘; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1, 0}; [𝑥 − −𝑥 + ]I𝑘+1 ·𝑥)

(𝐷 −𝑛−1) P1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 +P2 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 2
(

+
2(𝐷 −𝑛+𝑘 +𝑚 1 +𝑚 2 +· · ·+𝑚𝑘 )

− 𝐷−𝑛
× 1+A1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 +A2 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 2 2

𝐷−𝑛−3
× 1+B1 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 +B2 ({𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 )𝑥 2 2

))
×HG𝑙 (𝑛−1;𝑘; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1, 1}; [𝑥 − −𝑥 + ]I𝑘+1 ·𝑥) . (7.9)

The coefficients P1 (a; 𝑏), P2 (a; 𝑏), Q0 (a; 𝑏), Q1 (a; 𝑏), A1 (a; 𝑏), A2 (a; 𝑏), B1 (a; 𝑏), and
B2 (a; 𝑏) are defined by simply replacing all instances of labels 𝑎 1 and 𝑎 2 in (5.11), (5.15)
with the label list {𝑎 ′1, . . . , 𝑎𝑘′ } and the single label 𝑎𝑖 , respectively. Then, upon carrying out

– 22 –
the series expansion, it transforms into
∞ Z 𝓍I (𝓉) P𝑘+1
𝑚 𝑗 +𝐷−𝑛+𝑘−1
(7.10)
X
𝓉 1−𝐷+𝑛
𝑘+1
𝑑𝑥 𝑥 𝑗 =1

𝑚𝑘+1 =0 0
( 𝑘
× (0)𝑚𝑘+1 (𝑛;𝑘; {𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 ; 𝑚 𝑗 )·HG𝑙 (𝑛−1;𝑘; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1, 0}; [𝑥 − −𝑥 + ]I𝑘+1 ·𝑥)
X

𝑗=1
𝑘
)
+(1)𝑚𝑘+1 (𝑛;𝑘; {𝑎 ′1, . . .,𝑎𝑘′ };𝑎𝑖 ; 𝑚 𝑗 )·HG𝑙 (𝑛−1;𝑘; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1, 1}; [𝑥 − −𝑥 + ]I𝑘+1 ·𝑥) .
X

𝑗=1

According to integration rule (7.2), above equation becomes

JHEP02(2024)158
P𝑘+1
∞ 𝑚 𝑗 +𝑘+1
(
1 𝓉 𝑗 =1
(7.11)
X
·
𝑗=1 𝑚 𝑗 +𝐷 −𝑛+𝑘
P𝑘+1 P𝑘+1
𝑚 𝑗 +𝐷−𝑛+𝑘
𝑚𝑘+1 =0 (1−[𝑥 + ]I𝑘+1 ·𝓉) 𝑗 =1

1 𝑘
)
X X
× (b𝑘 )𝑚𝑘+1 (𝑛;𝑘; {𝑎 ′1, . . .𝑎𝑘′ };𝑎𝑖 ; 𝑚 𝑗 )·HG𝑙 (𝑛;𝑘 +1; {𝑏 1,𝑏 2, . . .,𝑏𝑘−1,𝑏𝑘 , 1}; 𝒲I𝑘+1 (𝓉)) .
𝑏𝑘 =0 𝑗=1

When everything is added together, we obtain the generating function of an 𝑛-gon to an

(𝑛 − (𝑘 + 1))-gon as follows:
GF𝑛→𝑛;Id (𝓉)
𝑘+1
2 ∞ 1 𝑘+1
−2+𝐷−𝑛 −2+𝐷−𝑛
=(1 − 𝑥 + · 𝓉) (1 − 𝑥 − · 𝓉)
X X X X X
2 2 ×
𝑙=1 𝑚 1 ,...,𝑚𝑘+1 =0 𝑏 1 ,...,𝑏𝑘 =0 𝑎𝑖 ,𝑖=1 {𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘+1 /𝑎𝑖 )
P𝑘+1
𝑚 +𝑘+1
1
(
𝓉 𝑗 =1 𝑗
× P𝑘+1 ·
𝑗=1 𝑚 𝑗 + 𝐷 − 𝑛 + 𝑘 (1 − [𝑥 + ]I
P𝑘+1
𝑚 𝑗 +𝐷−𝑛+𝑘
𝑘+1
· 𝓉) 𝑗 =1

𝑘
(𝑎 ′ ,...,𝑎 ′ ) (𝑙)
×(b𝑘 )𝑚𝑘+1 (𝑛; 𝑘; {𝑎 ′1, . . . 𝑎𝑘′ }; 𝑎𝑖 ; 𝑚 𝑗 ) · [C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛 − 1)]𝑎𝑖 [C𝑎′ (𝑛 − (𝑘 + 1) − 1; 𝑚 1 )]𝑎′2,...,𝑎𝑘′ ,𝑎𝑖
X
1
𝑗=1
)
×HG𝑙 (𝑛; 𝑘 + 1; {𝑏 1, 𝑏 2, . . . , 𝑏𝑘−1, 𝑏𝑘 , 1}; 𝒲I𝑘+1 (𝓉)) . (7.12)

(𝑎 ′ ,...,𝑎 ′ )
From the definition of C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛) (6.9), we have
𝑘
(𝑎 ′ ,...,𝑎 ′ )
X
[C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛 − 1)]𝑎𝑖 · (b𝑘 )𝑚𝑘+1 (𝑛; 𝑘; {𝑎 ′1, . . . 𝑎𝑘′ }; 𝑎𝑖 ; 𝑚𝑗)
𝑗=1

= [(b1 )𝑚2 (𝑛 − 𝑘 + 1; 1; 𝑎 ′1 ; 𝑎 ′2 ; 𝑚 1 )]𝑎′3,...,𝑎𝑘′ ,𝑎𝑖

· [(b2 )𝑚3 (𝑛 − 𝑘 + 2; 2; {𝑎 ′1, 𝑎 ′2 }; 𝑎 ′3 ; 𝑚 1 + 𝑚 2 )]𝑎′4,...,𝑎𝑘′ ,𝑎𝑖
···
· [(b 𝑗 )𝑚 𝑗 +1 (𝑛 − 𝑘 + 𝑗; 𝑗; {𝑎 ′1, . . . , 𝑎 ′𝑗 }; 𝑎 ′𝑗+1 ; 𝑚 1 + · · · + 𝑚 𝑗 )]𝑎′𝑗 +2,...,𝑎𝑘′ ,𝑎𝑖
···
· [(b𝑘−1 )𝑚𝑘 (𝑛; 𝑘 − 1; {𝑎 ′1, . . . , 𝑎𝑘−1
′
}; 𝑎𝑘′ ; 𝑚 1 + . . . + 𝑚𝑘−1 )]𝑎𝑖
𝑘
X
· (b𝑘 )𝑚𝑘+1 (𝑛; 𝑘; {𝑎 ′1, . . . 𝑎𝑘′ }; 𝑎𝑖 ; 𝑚𝑗)
𝑗=1
(𝑎 ′ ,...,𝑎 ′ ,𝑎 )
= C{𝑏11,...,𝑏𝑘𝑘 } 𝑖 (𝑛). (7.13)

– 23 –
And obviously,
𝑘+1
(𝑎 ′ ,...,𝑎 ′ ,𝑎𝑖 ) (𝑎 ′ ,...,𝑎 ′ ,𝑎 ′ )
C{𝑏11,...,𝑏𝑘𝑘 } (𝑛) = C{𝑏11,...,𝑏𝑘𝑘 } 𝑘+1 (𝑛). (7.14)
X X X

𝑎𝑖 ,𝑖=1 {𝑎 ′1 ,...,𝑎𝑘′ }∈𝜎(I𝑘+1 /𝑎𝑖 ) {𝑎 ′1 ,...,𝑎𝑘+1

′ }∈𝜎(I𝑘+1 )

Then, we can see that (7.12) precisely aligns with our result (6.5) in the 𝑘 + 1 case.
Therefore, we have successfully validated the correctness of our result.

8 Conclusion

JHEP02(2024)158
In this paper, we present an explicit expression for the generating function for the reduction of
an 𝑛-gon to an (𝑛 − 𝑘)-gon, where 𝑘 is a general value. We formulate a novel recursive relation
of generating functions, which is based on our investigation into Feynman Parametrization
in projective space. This newly established relation comprises a single ordinary differential
equation in terms of variable 𝓉. To solve this equation, it is required to carry out the integral
part by the following steps: (1) Changing the integration variable. (2) Implementing the Taylor
expansion on the appropriate part of the integrand. (3) Selecting the suitable integration
formula. In the end, we unearthed the rule of the general term formula of generating functions.
In addition, there are several comments that we need to supplement. Firstly, it is
understood that the reduction coefficients should be rational functions of some Lorentz
invariants such as 𝑞𝑖 · 𝑞 𝑗 , 𝑀𝑖2 and spacetime dimension 𝐷. However, in the expression (3.7)
and (6.5), there exists irrational terms such as the square root term
q
[𝑥 + − 𝑥 − ]I𝑘 = 4 (𝐿𝐿)I𝑘 𝑅 2 + (𝑉 𝐿)2I𝑘 − (𝐿𝐿)I𝑘 (𝑉𝑉 )I𝑘 /(𝐿𝐿)I𝑘 . (8.1)

In fact, by the following quadratic transformation

! −𝑎 2 !
 𝑎 𝑎+1 
𝑎, 𝑏 𝑧 2, 2 𝑧
2 𝐹1 𝑧 = 1− · 2 𝐹1 , (8.2)
2𝑏 2 𝑏 + 12 2−𝑧

equation (3.7) can be changed to a rational form

2 (𝑥 + − 𝑥 − )2 · 𝓉 2
!!
2, 1
1
GF𝑛→𝑛 (𝓉) = · 2 𝐹1 . (8.3)
(𝑥 + + 𝑥 − ) · 𝓉 − 2 𝐷−𝑛+1
2 ((𝑥 + + 𝑥 − ) · 𝓉 − 2)2

We can observe that all terms in the above expression are rational terms. Therefore, even
though the generating function we obtain contains irrational terms, it does not alter the
fact that the final reduced coefficients remain rational.7
Expressing rational numbers with irrational numbers is not unusual in mathematics, with
the general term formula of the famous Fibonacci sequence being an example
√ !𝑛 √ !𝑛 !
1 1+ 5 1− 5
𝐹𝑛 = √ − . (8.4)
5 2 2
7
However, for the generating function from an 𝑛-gon to an (𝑛 − 𝑘)-gon (6.5), we have not yet identified an
appropriate transformation that would render it in the form of a rational function.

– 24 –
This may suggest that in the existing work, the study of the analytic structure of the master
integrals or reduction coefficients is perhaps not yet deep enough.
In numerical computations on computers, maintaining a function in rational form allows
us to preserve the accuracy of the entire numerical calculation through methods such as Finite
Field Reconstruction, ensuring complete precision. When irrational terms are present, we
are forced to use floating-point numbers, which can lead to increased computational time and
(𝑎 ′ ,...,𝑎 ′ )
a reduction in precision. However, based on the definition of coefficient C{𝑏11,...,𝑏𝑘𝑘−1 } (𝑛) (6.9)
and the formula of (2.11)(4.10)(4.13)(5.11), it can be seen that the only irrational term
appearing in the generating function (6.5) from an 𝑛-gon to an (𝑛 − 𝑘)-gon is (8.1). Therefore,
√
by extending the rational number field Q to the field of form Q + 𝑋 · Q, we can similarly

JHEP02(2024)158
employ finite field reconstruction methods to maintain computational accuracy and save
computational time. Nonetheless, we maintain our belief that there may exist a rationalized
analytical representation for the generating function of one-loop tensor integral. Investigating
methods to derive such a rationalized expression is a meaningful avenue for future research.
Secondly, the generating function we discuss in this paper is in any spacetime dimension
𝐷 where all scalar integrals with single pole are irreducible. When the spacetime dimension 𝐷
takes a finite value (such as 4−2𝜖), some scalar integrals will become reducible scalar integrals,
then what would the form of the generating functions be in this case? Additionally, this
paper focuses on one-loop tensor integrals that solely encompass single poles. As for one-loop
integrals involving higher poles, what would be the corresponding generating functions for
reduction coefficients? We have already computed some results, which will be presented
in future articles.
Thirdly, the simplicity of the method in this paper lies in the fact that the recursion
relation established is an ordinary differential equation rather than a set of partial differential
equations. The foundation of this relation is the study of the analytic structure of Feynman
Parameterization in projective space. Could this logic be extrapolated to two loops, thus
obtaining the general form of the two-loop generating function, or at least, obtain some
simpler recursion relations? In [95], Kosower focus on two-loop integrals, where he only
deriving generating functions for the reduction coefficients associated with the most intricate
master integral. His approach is applied to only a limited set of concrete examples. In
contrast, our goal is to offer a comprehensive and systematic treatment encompassing all
generating function for general two-loop integrals. For the case of two-loop diagrams, we have
undertaken some initial attempts and obtained preliminary results, which will be presented
in future work. Nevertheless, it remains a significant challenge.

Acknowledgments

We would like to thank Prof. Dr. Bo Feng, Dr. Jiaqi Chen and Mr. Xiang Li for useful
discussion. We are also grateful to Prof. Dr. Johannes Bluemlein for providing valuable
feedback on this paper.

A Solving differential equations

The standard form of the first-order linear ordinary differential equation is:
𝑦 ′ + 𝑝(𝑥)𝑦 = 𝑞(𝑥). (A.1)

– 25 –
The solution is
R Z R
−
𝑦 =𝑒 𝑝(𝑥) 𝑑𝑥
𝑞(𝑥)𝑒 𝑝(𝑥) 𝑑𝑥
𝑑𝑥 . (A.2)

This result can be used as a formula.

In our cases (2.16), the differential equation has the typical form with a non-homogeneous
term 𝑔(𝑥)
 
(𝑎 − 1) + (𝑎/2)𝐴1𝑥 + 𝐴2𝑥 2 𝑦(𝑥) + 𝑥(1 + 𝐴1𝑥 + 𝐴2𝑥 2 )𝑦 ′ (𝑥) = 𝑔(𝑥), (A.3)

JHEP02(2024)158
where 𝑎, 𝐴1, 𝐴2 are several coefficients which are independent of the variable 𝑥. Then 𝑝(𝑥),
𝑞(𝑥) in (A.2):

(𝑎 − 1) + (𝑎/2)𝐴1𝑥 + 𝐴2𝑥 2
𝑝(𝑥) = , (A.4)
𝑥(1 + 𝐴1𝑥 + 𝐴2𝑥 2 )

𝑔(𝑥)
𝑞(𝑥) = . (A.5)
𝑥(𝐴1 + 𝐴2𝑥 + 𝐴3𝑥 2 )
We split the integral of 𝑝(𝑥) into

(𝑎 − 1) + (𝑎/2)𝐴1𝑥 + 𝐴2𝑥 2
Z Z
𝑝(𝑥)𝑑𝑥 = 𝑑𝑥
𝑥(1 + 𝐴1𝑥 + 𝐴2𝑥 2 )
𝑎−1 (𝐴1 + 2𝐴2𝑥)
Z  
𝑎
= + (1 − ) 𝑑𝑥
𝑥 2 (1 + 𝐴1𝑥 + 𝐴2𝑥 2 ) (A.6)
𝑎−1 𝑎 (1 + 𝐴1𝑥 + 𝐴2𝑥 2 )′
Z  
= + (1 − ) 𝑑𝑥
𝑥 2 1 + 𝐴1𝑥 + 𝐴2𝑥 2
𝑎
= (𝑎 − 1) ln(𝑥) + (1 − ) ln(1 + 𝐴1𝑥 + 𝐴2𝑥 2 ) + 𝐶𝑜𝑛𝑠𝑡 .
2

Then the general solution of the homogeneous part is

R
𝑦𝐻 (𝑥) = 𝑒 −
𝑎−2
𝑝(𝑥) 𝑑𝑥
= 𝐶𝑜𝑛𝑠𝑡 · 𝑥 1−𝑎 (1 + 𝐴1𝑥 + 𝐴2𝑥 2 ) 2 . (A.7)

By (A.2), the solution of the original equation is

R Z R
𝑦(𝑥) = 𝑒 − 𝑝(𝑥)𝑑𝑥
𝑞(𝑥)𝑒 𝑝(𝑥)𝑑𝑥
𝑑𝑥
𝑎−2
Z 𝑥  (A.8)
2 − 𝑎2
=𝑥 1−𝑎
(1+𝐴1𝑥 +𝐴2𝑥 ) 2 2 𝑔(𝑢)𝑢 𝑎−2
(1+𝐴1𝑢 +𝐴2𝑢 ) 𝑑𝑢 +𝐶𝑜𝑛𝑠𝑡 .
0

In the case of the reduction of an 𝑛-gon to an 𝑛-gon, the non-homogeneous part is a constant
𝑔(𝑥) = (𝑎 − 1). By following integral formula:
2−𝑎
(𝑦 − 𝑥 1 )− 2 (𝑦 − 𝑥 2 )
!
1, 𝑎/2 𝑥 2 − 𝑥 1
Z 𝑎
2
− 𝑎2 − 𝑎2
(𝑦 − 𝑥 1 ) (𝑦 − 𝑥 2 ) 𝑑𝑦 = · 2 𝐹1 + 𝐶𝑜𝑛𝑠𝑡,
1−𝑎 𝑎 𝑦 − 𝑥1
(A.9)

– 26 –
we can solve
R Z R
𝑦(𝑥) = 𝑒 − 𝑝(𝑥) 𝑑𝑥
𝑞(𝑥)𝑒 𝑝(𝑥) 𝑑𝑥
𝑑𝑥
 Z 𝑥 
𝑎−2
= 𝑥 1−𝑎 (1 + 𝐴1𝑥 + 𝐴2𝑥 2 ) (𝑎 − 1) 𝑢 𝑎 (1 + 𝐴1𝑢 + 𝐴2𝑢 2 )− 2 𝑢 −2𝑑𝑢 + 𝐶𝑜𝑛𝑠𝑡
𝑎
2
0
1
1 1 1 1 1 1
  𝑎−2 Z  − 𝑎   !
2 2
= + 𝐴1 + 𝐴2 (𝑎 − 1) − 2 + 𝐴1 + 𝐴2 + 𝐶𝑜𝑛𝑠𝑡
𝑥
2
𝑑
𝑥 𝑥 𝑥 ∞ 𝑢 𝑢 𝑢
1
1 1 1
  𝑎−2   𝑎−2 Z !
2 2
− 𝑎2 − 𝑎2
= (1 − 𝑎) (𝑦 − 𝑥 + ) (𝑦 − 𝑥 − ) 𝑑𝑦 + 𝐶𝑜𝑛𝑠𝑡
𝑥
− 𝑥+ − 𝑥−
𝑥 𝑥 𝑥 ∞

JHEP02(2024)158
1 1 1
  𝑎−2   𝑎−2
2 2
= − 𝑥+ − 𝑥−
𝑥 𝑥 𝑥
 1

( !
1, 𝑎/2 𝑥 − − 𝑥 +
𝑥
2−𝑎
×  (𝑦 − 𝑥 + )− 2 (𝑦 − 𝑥 − ) 2 · 2 𝐹 1
𝑎
+ 𝐶𝑜𝑛𝑠𝑡 
 
𝑎 𝑦 − 𝑥+
𝑦=∞

1
1, 𝑎/2 (𝑥 − − 𝑥 + )𝑥
!
𝑎−2 𝑎−2
=𝑥 (1 − 𝑥 + · 𝑥) (1 − 𝑥 − · 𝑥)
1−𝑎 2 2 · 𝐶𝑜𝑛𝑠𝑡 + ,· 2 𝐹1
𝑎 1 − 𝑥+ · 𝑥
1 − 𝑥+ · 𝑥
(A.10)
where 𝑥 −, 𝑥 + are two roots of the quadratic equation 𝑥 + 𝐴1𝑥 + 𝐴2 = 0,
2 8
q q
−𝐴1 + 𝐴21 − 4𝐴2 −𝐴1 − 𝐴21 − 4𝐴2
𝑥+ = , 𝑥− = . (A.11)
2 2

B Numerical check

In this section we will verify our generating function numerically by comparing to the result
produced by the cpp version of FIRE6[48]. The tensor integral we reduce is defined as follows:
𝑑𝐷𝑙 (2𝑅 · 𝑙)𝑟
Z
. (B.1)
𝑖𝜋 𝐷/2 𝐷 1 𝐷 2 · · · 𝐷𝑛
B.1 Tadpole
Let us first begin with a trivial Tadpole case, where we have only one master integral. We
choose the numeric value of the kinetic variable as follows:
5 133 969
𝑀12 → , 𝑞 12 → 0, 𝑅 2 → , 𝑞 1 · 𝑅 → 0, 𝐷 → . (B.2)
4 10 50
• Tadpole to Tadpole 1 → 1
The numerical Taylor expansion of the generating function is given as follows:

GF1→1 (𝓉) = 1 + 3.4313725490196076 · 𝓉 2 + 32.0186540472129 · 𝓉 4

+ 455.355109952878 · 𝓉 6 + 8351.765314541557 · 𝓉 8
+ 182561.41492889417 · 𝓉 10 + 𝒪(𝓉 12 ), (B.3)

which match the result of FIRE6 perfectly.

√ √
8 −𝐴1 − 𝐴2 1
−4𝐴2 −𝐴1 + 𝐴21
−4𝐴2
We can also choose 𝑥 + = 2
, 𝑥− = 2
. In this way, we will have another form of
solution. Two forms are equivalent due to the properties of hypergeometric functions.

– 27 –
 Rank 0 1 2
Result 1. 0. 3.4313725490196076
Rank 3 4 5
Result 0. 32.0186540472129 0.
Rank 6 7 8
Result 455.355109952878 0. 8351.765314541557
Rank 9 10 11
Result 0. 182561.41492889417 0.

Table 1. Reduction Result of Bubble to Bubble by FIRE6.

JHEP02(2024)158
B.2 Bubble
We choose the numeric value of the kinetic variable as follows:
5 64 69
𝑀12 → , 𝑀22 → , 𝑞 12 → 0, 𝑞 22 → , 𝑞 1 · 𝑞 2 → 0,
4 25 50
133 367 969
𝑅2 → , 𝑞 1 · 𝑅 → 0, 𝑞 2 · 𝑅 → ,𝐷 → . (B.4)
10 100 50
For a tensor Bubble integral, we have totally 3 master integrals:

• Bubble to Bubble 2 → 2

Rank 0 1 2
Result 1. 0.18615942028985508 0.9969550236277754
Rank 3 4 5
Result 0.5438748755636772 2.706729663300785 2.394357906464815
Rank 6 7 8
Result 11.220093828863014 13.471511680729403 60.095504925096684
Rank 9 10 11
Result 89.67183068422794 384.30945689202724 675.8039722773336

Table 2. Reduction Result of Bubble to Bubble by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF2→2 (𝓉) = 1 + 0.18615942028985508 · 𝓉 + 0.9969550236277754 · 𝓉 2

+ 0.5438748755637103 · 𝓉 3 + 2.7067296633008473 · 𝓉 4
+ 2.394357906466195 · 𝓉 5 + 11.220093828873562 · 𝓉 6
+ 13.471511680934789 · 𝓉 7 + 60.09550492544111 · 𝓉 8
+ 89.67183069722255 · 𝓉 9 + 384.3094569884708 · 𝓉 10
+ 675.8039743685777 · 𝓉 11 + 𝒪(𝓉 12 ). (B.5)

• Bubble to Tadpole 2 → 1
There are two Tadpole master integrals in this case.

– 28 –
 Rank 0 1 2
Result 0. 2.6594202898550723 20.390645089351526
Rank 3 4 5
Result 175.8509958812531 1611.1537585017563 15713.736765104697
Rank 6 7 8
Result 161200.4597648606 1.726292310058 × 106 1.91794043011 × 107
Rank 9 10 11
Result 2.20010455418 × 108 2.59573008468 × 109 3.13986818427 × 1010

Table 3. Reduction Result of Bubble to 𝐷 2 Tadpole by FIRE6.

JHEP02(2024)158
The numerical Taylor expansion of the generating function is given as follows,

GF2→1;b1 (𝓉) = 2.6594202898·𝓉+20.390645089351·𝓉 2 +175.85099588125·𝓉 3

+1611.153758502·𝓉 4 +15713.736765105·𝓉 5 +161200.45976486·𝓉 6
+1.72629231×106 ·𝓉 7 +1.91794043×107 ·𝓉 8 +2.20010455×108 ·𝓉 9
+2.595730085×109 ·𝓉 10 +3.13986818×1010 ·𝓉 11 +𝒪(𝓉 12 ). (B.6)

Rank 0 1 2
Result 0. -2.6594202898550723 -0.4853067411154148
Rank 3 4 5
Result -14.069834047182031 -3.9250909295456946 -132.48553286333672
Rank 6 7 8
Result -40.82736337173687 -1820.708411154273 -554.9841273406652
Rank 9 10 11
Result -32477.51478825142 -9528.7663298311 -697318.0606970708

Table 4. Reduction Result of Bubble to 𝐷 1 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF2→1;b2 (𝓉) = −2.6594202898550·𝓉−0.4853067411154·𝓉 2 −14.0698340471820·𝓉 3

−3.9250909295456·𝓉 4 −132.4855328633367·𝓉 5 −40.8273633717368·𝓉 6
−1820.7084111542733·𝓉 7 −554.9841273406652·𝓉 8 −32477.5147882514220·𝓉 9
−9528.7663298311876·𝓉 10 −697318.0606970731579·𝓉 11 +𝒪(𝓉 12 ). (B.7)

B.3 Triangle
We choose the numeric value of the kinetic variable as follows:
5 64 357 2 69 129
𝑀12 → , 𝑀22 → , 𝑀32 → , 𝑞 1 → 0, 𝑞 22 → , 𝑞 32 → , 𝑞 1 · 𝑞 2 → 0, 𝑞 1 · 𝑞 3 → 0,
4 25 100 50 25
41 133 367 89 969
𝑞2 · 𝑞3 → , 𝑅2 → , 𝑞 1 · 𝑅 → 0, 𝑞 2 · 𝑅 → , 𝑞3 · 𝑅 → , 𝐷 → . (B.8)
20 10 100 10 50

– 29 –
 Rank 0 1 2
Result 1. 4.647429667957373 21.424063591070208
Rank 3 4 5
Result 97.94451396121778 443.9628092202622 1994.7300798765605
Rank 6 7 8
Result 8880.900093335815 39165.84906934403 171022.14771781594
Rank 9 10 11
Result 739055.1488881096 3.15884655856376 × 106 1.3344673752179299 × 107

Table 5. Reduction Result of Triangle to Triangle by FIRE6.

JHEP02(2024)158
• Triangle to Triangle,3 → 3
The numerical Taylor expansion of the generating function is given as follows:

GF3→3 (𝓉) = 1.+4.647429667957·𝓉+21.42406359107·𝓉 2 +97.9445139612·𝓉 3

+443.9628092202622·𝓉 4 +1994.7300798765605·𝓉 5 +8880.900093335815·𝓉 6
+39165.84906934403·𝓉 7 +171022.14771781594·𝓉 8 +739055.1488881096·𝓉 9
+3.15884655856376×106 ·𝓉 10 +1.3344673752179299×107 ·𝓉 11 +𝒪(𝓉 12 ).
(B.9)

• Triangle to Bubble,3 → 2
There are 3 Bubble master integrals in this case, we choose two typical integrals to
verify our generating function.

Rank 0 1 2
Result 0. 1.867765479902683 27.80579292856186
Rank 3 4 5
Result 326.58851259528217 3575.377043547329 38296.692771726164
Rank 6 7 8
Result 408802.874852242 4.3817950217888 × 106 4.730368560885 × 107
Rank 9 10 11
Result 5.148927898936 × 108 5.6522910727 × 109 6.2570409303155 × 1010

Table 6. Reduction Result of Triangle to 𝐷 2 𝐷 3 Bubble by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF3→2;b1 (𝓉) = 1.86776547990·𝓉+27.80579292856·𝓉 2 +326.58851259528·𝓉 3

+3575.377043547329·𝓉 4 +38296.692771726167·𝓉 5 +408802.87485224194·𝓉 6
+4.3817950217888×106 ·𝓉 7 +4.730368560885×107 ·𝓉 8 +5.1489278989×108 ·𝓉 9
+5.6522910727×109 ·𝓉 10 +6.2570409303×1010 ·𝓉 11 +𝒪(𝓉 12 ). (B.10)

– 30 –
 Rank 0 1 2
Result 0. -0.2371928862694034 -2.039991135700739
Rank 3 4 5
Result -12.812290548944913 -69.25281622582993 -336.83725868345635
Rank 6 7 8
Result -1487.8795033941478 -5880.3689489331 -19650.58517698222
Rank 9 10 11
Result -44502.51591828201 48219.21614887317 1.5017448726269251 × 106

Table 7. Reduction Result of Triangle to 𝐷 1 𝐷 3 Bubble by FIRE6.

JHEP02(2024)158
The numerical Taylor expansion of the generating function is given as follows:

GF3→2;b2 (𝓉) = −0.2371928863·𝓉−2.0399911357·𝓉 2 −12.81229054894·𝓉 3

−69.25281622583·𝓉 4 −336.83725868346·𝓉 5 −1487.87950339415·𝓉 6
−5880.368948933·𝓉 7 −19650.5851769822·𝓉 8 −44502.51591828201·𝓉 9
+48219.2161488732·𝓉 10 +1.50174487263×106 ·𝓉 11 +𝒪(𝓉 12 ). (B.11)

• Triangle to Tadpole, 3 → 1

There are 3 Tadpole master integrals in this case.

Rank 0 1 2
Result 0. 0. 3.5943363780985007
Rank 3 4 5
Result 120.15405371902189 2856.4892399384603 60343.814251369666
Rank 6 7 8
Result 1.22027431911 × 106 2.43997498612 × 107 4.899382591712 × 108
Rank 9 10 11
Result 9.95303381377 × 109 2.052398465613 × 1011 4.300770011403 × 1012

Table 8. Reduction Result of Triangle to 𝐷 1 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = 3.594336378098·𝓉 +120.154053719·𝓉 +2856.489239938·𝓉

2 3 4
GF3→1;12
+60343.81425137·𝓉 5 +1.220274319×106 ·𝓉 6 +2.439974986×107 ·𝓉 7
+4.899382592×108 ·𝓉 8 +9.953033814×109 ·𝓉 9 +2.052398466×1011 ·𝓉 10
+4.30077001×1012 ·𝓉 11 +𝒪(𝓉 12 ). (B.12)

– 31 –
 Rank 0 1 2
Result 0. 0. 4.745489600791057
Rank 3 4 5
Result 24.912564026958766 142.5977420103585 671.4679047318388
Rank 6 7 8
Result 3235.5137500307833 14241.416319020253 63070.04241030086
Rank 9 10 11
Result 258765.71357130772 1.06056655995 × 106 3.96215910911 × 106

Table 9. Reduction Result of Triangle to 𝐷 1 Tadpole by FIRE6.

JHEP02(2024)158
The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = 4.74548960079·𝓉 +24.91256402696·𝓉 +142.59774201036·𝓉

2 3 4
GF3→1;23
+671.4679047318 𝑐𝑑𝑜𝑡𝓉 5 +3235.5137500308·𝓉 6 +14241.4163190203·𝓉 7
+63070.0424103009·𝓉 8 +258765.71357130772·𝓉 9 +1.0605665599514×106 ·𝓉 10
+3.9621591091×106 ·𝓉 11 +𝒪(𝓉 12 ). (B.13)

Rank 0 1 2
Result 0. 0. -8.339825978889557
Rank 3 4 5
Result -142.45218080322186 -1909.4499907718186 -23603.993924573977
Rank 6 7 8
Result -282767.93446188176 −3.347293948348 × 106 −3.9512933931 × 107
Rank 9 10 11
Result −4.673481474464 × 108 −5.55382934594 × 109 −6.64265408461 × 1010

Table 10. Reduction Result of Triangle to 𝐷 2 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = −8.33982597889·𝓉 −142.45218080322·𝓉 −1909.44999077182·𝓉

2 3 4
GF3→1;13
−23603.993924574·𝓉 5 −282767.9344618818·𝓉 6 −3.34729394835×106 ·𝓉 7
−3.95129339314×107 ·𝓉 8 −4.67348147446×108 ·𝓉 9
−5.55382934594×109 ·𝓉 10 −6.64265408461×1010 ·𝓉 11 +𝒪(𝓉 12 ).
(B.14)

– 32 –
B.4 Box
We choose the numeric value of the kinetic variable as follows:
5 64 357 2 339 2 69 129
𝑀12 → , 𝑀22 → , 𝑀32 → , 𝑀4 → , 𝑞 1 → 0, 𝑞 22 → , 𝑞 32 → ,
4 25 100 50 50 25
2069 41
𝑞 42 → , 𝑞 1 · 𝑞 2 → 0, 𝑞 1 · 𝑞 3 → 0, 𝑞 1 · 𝑞 4 → 0, 𝑞 2 · 𝑞 3 → ,
100 20
93 56 2 133
𝑞2 · 𝑞4 → , 𝑞3 · 𝑞4 → , 𝑅 → , 𝑞 1 · 𝑅 → 0,
25 5 10
367 89 2413 969
𝑞2 · 𝑅 → , 𝑞3 · 𝑅 → , 𝑞4 · 𝑅 → ,𝐷 → . (B.15)
100 10 100 50

JHEP02(2024)158
• Box to Box,4 → 4

Rank 0 1 2
Result 1. -3.6920841521541408 16.55377492166991
Rank 3 4 5
Result -82.69660780758952 447.6598650239185 -2578.2813404112676
Rank 6 7 8
Result 15614.703573346487 -98601.17919296624 645114.4523761669
Rank 9 10 11
Result −4.35187298694 × 106 3.0151991318039 × 107 −2.138897834607 × 108

Table 11. Reduction Result of Box to Box by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF4→4 (𝓉) = 1.−3.692084152154·𝓉+16.5537749217·𝓉 2 −82.6966078076·𝓉 3

+447.6598650239·𝓉 4 −2578.281340411·𝓉 5 +15614.70357335·𝓉 6
−98601.179192966·𝓉 7 +645114.4523762·𝓉 8 −4.351872986943×106 ·𝓉 9
+3.015199131804×107 ·𝓉 10 −2.1388978346×108 ·𝓉 11 +𝒪(𝓉 12 ). (B.16)

• Box to Triangle, 4 → 3
We choose two typical master integrals to verify our generating function.

Rank 0 1 2
Result 0. 2.297291695479869 -3.938379081474582
Rank 3 4 5
Result 60.596241405397464 -100.49970077717936 2298.7596822646256
Rank 6 7 8
Result -1488.7266429821102 114099.44338526356 128454.40427289286
Rank 9 10 11
Result 7.110492574192 × 106 2.395974442872 × 107 5.39796222751 × 108

Table 12. Reduction Result of Box to 𝐷 2 𝐷 3 𝐷 4 Triangle by FIRE6.

– 33 –
 The numerical Taylor expansion of the generating function is given as follows:

GF4→3;b1 (𝓉) = 2.29729169548·𝓉−3.93837908147·𝓉 2 +60.5962414054·𝓉 3

−100.499700777·𝓉 4 +2298.7596822646·𝓉 5 −1488.7266429821·𝓉 6
+114099.44338526357·𝓉 7 +128454.4042728928·𝓉 8 +7.11049257419×106 ·𝓉 9
+2.39597444287×107 ·𝓉 10 +5.3979622275×108 ·𝓉 11 +𝒪(𝓉 12 ). (B.17)

JHEP02(2024)158
Rank 0 1 2
Result 0. 1.1727646483063998 -13.437042633643701
Rank 3 4 5
Result 133.44969733467502 -1305.8124781550728 12995.470486241737
Rank 6 7 8
Result -132770.84174232915 1.395549705737 × 106 −1.508260416218 × 107
Rank 9 10 11
Result 1.673393577774 × 108 −1.902215247513 × 109 2.2110105646826 × 1010

Table 13. Reduction Result of Box to 𝐷 1 𝐷 3 𝐷 4 Triangle by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF4→3;b2 (𝓉) = 1.172764648·𝓉−13.437042634·𝓉 2 +133.449697335·𝓉 3

−1305.812478155·𝓉 4 +12995.470486242·𝓉 5 −132770.84174233·𝓉 6
+1.3955497057×106 ·𝓉 7 −1.508260416×107 ·𝓉 8 +1.673393578×108 ·𝓉 9
−1.9022152475×109 ·𝓉 10 +2.2110105647×1010 ·𝓉 11 +𝒪(𝓉 12 ). (B.18)

• Box to Bubble, 4 → 2

We choose two typical integrals to verify our generating function.

Rank 0 1 2
Result 0. 0. 14.283644489205553
Rank 3 4 5
Result 204.56675272706224 4109.787640764211 54424.224763862854
Rank 6 7 8
Result 614054.5523170701 1.9047172844007 × 106 −8.985743083129 × 107
Rank 9 10 11
Result −2.6301936458306 × 109 −3.119325555871 × 1010 1.6840989924241 × 1011

Table 14. Reduction Result of Box to 𝐷 3 𝐷 4 Bubble by FIRE6.

– 34 –
 The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = 14.283644489·𝓉 +204.566752727·𝓉 +4109.7876407642·𝓉

2 3 4
GF4→2;12
+54424.2247638628·∗𝓉 5 +614054.55231707·𝓉 6 +1.904717284×106 ·𝓉 7
−8.985743083×107 ·𝓉 8 −2.630193646×109 ·𝓉 9 −3.1193256×1010 ·𝓉 10
+1.6840989924×1011 ·𝓉 11 +𝒪(𝓉 12 ). (B.19)

JHEP02(2024)158
Rank 0 1 2
Result 0. 0. -9.695886285967024
Rank 3 4 5
Result -188.13277462466073 -6091.362674944455 -152379.17835453493
Rank 6 7 8
Result −4.04931789159 × 106 −1.034624729484 × 108 −2.669747661571 × 109
Rank 9 10 11
Result −6.863487054848 × 1010 −1.7752858181645 × 1012 −4.6131078626 × 1013

Table 15. Reduction Result of Box to 𝐷 1 𝐷 4 Bubble by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = −9.6958862859·𝓉 −188.1327746246·𝓉 −6091.3626749444·𝓉

2 3 4
GF4→2;23
−152379.1783545·𝓉 5 −4.04931789×106 ·𝓉 6 −1.034624729484×108 ·𝓉 7
−2.66974766×109 ·𝓉 8 −6.863487055×1010 ·𝓉 9 −1.77528582×1012 ·𝓉 10
−4.61310786×1013 𝓉 11 +𝒪(𝓉 12 ). (B.20)

• Box to Tadpole 4 → 1

We choose two typical integrals to verify our generating function.

Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5
Result -2.39789456182502 -26.096086956520185 986.8535842106996
Rank 6 7 8
Result 53824.37061112356 1.945611485647 × 106 5.97872946799 × 107
Rank 9 10 11
Result 1.726812691986 × 109 4.81514011073 × 1010 1.320524816629 × 1012

Table 16. Reduction Result of Box to 𝐷 1 Tadpole by FIRE6.

– 35 –
 The numerical Taylor expansion of the generating function is given as follows:

c (𝓉) = −2.397894561·𝓉 −26.096086956·𝓉 +986.853584210·𝓉

3 4 5
GF4→1;234
+53824.370611123·𝓉 6 +1.94561148×106 ·𝓉 7 +5.97872947×107 ·𝓉 8
+1.726813×109 ·𝓉 9 +4.81514×1010 ·𝓉 10 +1.320525×1012 ·𝓉 11 +𝒪(𝓉 12 ).
(B.21)

Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5

JHEP02(2024)158
Result 21.912939954259137 780.2207801551945 20884.548959274303
Rank 6 7 8
Result 490764.9683116434 1.078965830487 × 107 2.265411104988 × 108
Rank 9 10 11
Result 4.57159700668 × 109 8.82631887425 × 1010 1.6031881275156 × 1012

Table 17. Reduction Result of Box to 𝐷 2 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

c (𝓉) = 21.912939954·𝓉 +780.220780155·𝓉 +20884.54895927·𝓉

3 4 5
GF4→1;134
+490764.9683116·𝓉 6 +1.0789658×107 ·𝓉 7 +2.2654111×108 ·𝓉 8
+4.571597×109 ·𝓉 9 +8.826319×1010 ·𝓉 10 +1.6031881×1013 ·𝓉 11 +𝒪(𝓉 12 ).
(B.22)

B.5 Pentagon
We choose the numeric value of the kinetic variable as follows:
5 64 357 2 339 2 489 2 69 129
𝑀12 → , 𝑀22 → , 𝑀32 → , 𝑀4 → , 𝑀5 → , 𝑞 1 → 0, 𝑞 22 → , 𝑞 32 → ,
4 25 100 50 100 50 25
2069 2 1957 41
𝑞 42 → ,𝑞 → , 𝑞 1 · 𝑞 2 → 0, 𝑞 1 · 𝑞 3 → 0, 𝑞 1 · 𝑞 4 → 0, 𝑞 1 · 𝑞 5 → 0, 𝑞 2 · 𝑞 3 → ,
100 5 25 20
93 721 56 473 4437 2 133
𝑞2 · 𝑞4 → , 𝑞2 · 𝑞5 → , 𝑞3 · 𝑞4 → , 𝑞3 · 𝑞5 → , 𝑞4 · 𝑞5 → ,𝑅 → ,
25 100 5 20 100 10
367 89 2413 3737 969
𝑞 1 · 𝑅 → 0, 𝑞 2 · 𝑅 → , 𝑞3 · 𝑅 → , 𝑞4 · 𝑅 → , 𝑞5 · 𝑅 → ,𝐷 → . (B.23)
100 10 100 100 50
• Pentagon to Pentagon 5 → 5
Rank 0 1 2
Result 1. 28.14374534067907 891.8998182261959
Rank 3 4 5
Result 30720.54870031494 1.1282643383555224*∧ 6 4.363353048091063*∧ 7
Rank 6 7 8
Result 1.761204119860 × 109 7.370817943138 × 1010 3.1822242795255 × 1012
Rank 9 10 11
Result 1.41158438406 × 1014 6.412540767851 × 1015 2.975347813775 × 1017

– 36 –
 The numerical Taylor expansion of the generating function is given as follows:

GF5→5 (𝑡) = 1. + 28.14374534067907 · 𝓉 + 891.8998182261959 · 𝓉 2

+ 30720.548700314932 · 𝓉 3 + 1.1282643383555175 × 106 · 𝓉 4
+ 4.3633530480910465 × 107 · 𝓉 5 + 1.7612041198600552 × 109 · 𝓉 6
+ 7.370817943138432 × 1010 · 𝓉 7 + 3.1822242795256416 × 1012 · 𝓉 8
+ 1.411584384056077 × 1014 · 𝓉 9 + 6.412540767842304 × 1015 · 𝓉 10
+ 2.975347813764606 × 1017 · 𝓉 11 + 𝒪(𝓉 12 ). (B.24)

JHEP02(2024)158
• Pentagon to Box, 5 → 4
We choose two typical integrals to verify our generating function.

Rank 0 1 2
Result 0. 1.9794571126964369 117.0165516634012
Rank 3 4 5
Result 5816.790907215448 278858.4717048013 1.33561859165727 × 107
Rank 6 7 8
Result 6.467560554308 × 108 3.17978538594604 × 1010 1.5892825779136 × 1012
Rank 9 10 11
Result 8.07512917564463 × 1013 4.16873605287026 × 1015 2.18489096755686 × 101 7·

Table 18. Reduction Result of Pentagon to 𝐷 2 𝐷 3 𝐷 4 𝐷 5 Triangle by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

GF5→4;b1 (𝓉) = 1.979457112696436853508·𝓉+117.016551663401212089687·𝓉 2

+5816.790907215448086546878·𝓉 3 +278858.471704801335516720233·𝓉 4
+1.335618591657268161878×107 ·𝓉 5 +6.467560554307838030352×108 ·𝓉 6
+3.179785385946040065817×1010 ·𝓉 7 +1.589282577913656415226×1012 ·𝓉 8
+8.075129175644627972880×1013 ·𝓉 9 +4.168736052870256702551×1015 𝓉 10
+2.184890967556854672072×1017 𝓉 11 +𝒪(𝓉 12 ). (B.25)

The numerical Taylor expansion of the generating function is given as follows:

GF5→4;b3 (𝓉) = − 4.281411630475701002407 · 𝓉 − 344.283064611784617369902 · 𝓉 2

− 22151.543708665336258844567 · 𝓉 3 − 1.324001651820724484404 · 𝓉 4
− 7.657550600362864047 × 107 · 𝓉 5 − 4.35288324879994 × 109 · 𝓉 6
− 2.4492345879991 × 1011 · 𝓉 7 − 1.369070967619795958 × 1013 · 𝓉 8
− 7.61811824460816178 × 1014 · 𝓉 9 − 4.2250671939123482 × 1016 · 𝓉 10
− 2.337426727880305159 × 1018 · 𝓉 11 + 𝒪(𝓉 19 ). (B.26)

– 37 –
Rank 0 1 2
Result 0. -4.281411630475701 -344.28306461178465
Rank 3 4 5
Result -22151.543708665336 −1.3240016518207 × 106 −7.657550600363 × 107
Rank 6 7 8
Result −4.3528832488 × 109 −2.449234588 × 1011 −1.36907010 × 1013
Rank 9 10 11
Result −7.61811824461 × 1014 −4.225067193912 × 1016 −2.33742672788 × 1018

Table 19. Reduction Result of Pentagon to 𝐷 1 𝐷 2 𝐷 4 𝐷 5 Triangle by FIRE6.

JHEP02(2024)158
• Pentagon to Triangle, 5 → 3
We choose two typical integrals to verify our generating function.

Rank 0 1 2
Result 0. 0. 7.3203111013775555
Rank 3 4 5
Result 873.0823162422049 76037.34249486253 5.893055817313 × 106
Rank 6 7 8
Result 4.32079052040 × 108 3.074305412771 × 1010 2.14894585343427 × 1012
Rank 9 10 11
Result 1.485324405683 × 1014 1.0188953358844 × 1016 6.9518264510604 × 1017

Table 20. Reduction Result of Pentagon to 𝐷 3 𝐷 4 𝐷 5 Triangle by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = 7.32031110138 · 𝓉 + 873.082316242 · 𝓉 + 76037.34249486 · 𝓉

2 3 4
GF5→3;12
+ 5.89305581731 × 106 · 𝓉 5 + 4.3207905204 × 108 · 𝓉 6
+ 3.07430541277 × 1010 𝓉 7 + 2.14894585343 × 1012 · 𝓉 8
+ 1.485324405683 × 1014 𝓉 9 + 1.018895335884 × 1016 · 𝓉 10
+ 6.951826451060 × 1017 · 𝓉 11 + 𝒪(𝓉 12 ). (B.27)

The numerical Taylor expansion of the generating function is given as follows:

b (𝓉) = −1.47485132634·𝓉 −115.81178544·𝓉 −6395.13996956·𝓉

2 3 4
GF5→3;23
−288631.6787552·𝓉 5 −1.02042164191×107 ·𝓉 6 −1.69252211527×108 ·𝓉 7
+1.550496634133×1010 ·𝓉 8 +2.369360641667×1012 ·𝓉 9
+2.2181000022×1014 ·𝓉 10 +1.77060597057×1016 ·𝓉 11 +𝒪(𝓉 12 ).
(B.28)

– 38 –
Rank 0 1 2
Result 0. 0. -1.4748513263396505
Rank 3 4 5
Result -115.81178544031663 -6395.139969561789 -288631.67875521374
Rank 6 7 8
Result −1.0204216419115 × 107 −1.692522115271 × 108 1.550496634133 × 1010
Rank 9 10 11
Result 2.369360641667 × 1012 2.218100002242 × 1014 1.770605970571 × 1016

Table 21. Reduction Result of Pentagon to 𝐷 1 𝐷 4 𝐷 5 Triangle by FIRE6.

JHEP02(2024)158
• Pentagon to Bubble, 5 → 2

We choose two typical integrals to verify our generating function.

Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5
Result -3.45043209014488 -598.2368342483164 -68719.25792386582
Rank 6 7 8
Result −6.5930722347316 × 106 −5.722281197191 × 108 −4.662849011137 × 1010

Table 22. Reduction Result of Pentagon to 𝐷 4 𝐷 5 Bubble by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

c (𝓉) = −3.45043209014488·𝓉 −598.236834248316468·𝓉

3 4
GF5→2;123
−68719.2579238658296·𝓉 5 −6.593072234731597×106 ·𝓉 6
−5.722281197191451×108 ·𝓉 7 −4.662849011137064×1010 𝓉 8 +𝒪(𝓉 9 ).
(B.29)

Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5
Result 13.319303811986032 948.0673270255978 58591.314814083744
Rank 6 7 8
Result 3.3075884251460 × 106 1.7962165060926 × 108 9.51718360623 × 109

Table 23. Reduction Result of Pentagon to 𝐷 1 𝐷 4 Bubble by FIRE6.

– 39 –
 The numerical Taylor expansion of the generating function is given as follows:

c (𝓉) = 13.3193038119860332485·𝓉 +948.06732702559779935·𝓉

3 4
GF5→2;235
+58591.3148140837440432·𝓉 5 +3.307588425146047×106 ·𝓉 6
+1.796216506092603×108 ·𝓉 7 +9.5171836062295834×109 ·𝓉 8 +𝒪(𝓉 9 ).
(B.30)

• Pentagon to Tadpole 5 → 1
We choose two typical integrals to verify our generating function.

JHEP02(2024)158
Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5
Result 0. -61.3501671673956 -7960.61229768613
Rank 6 7 8
Result -747095.8262283974 −6.1530993935739 × 107 −4.746641454844 × 109

Table 24. Reduction Result of Pentagon to 𝐷 4 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

d (𝓉) = −61.3501671673955991·𝓉 −7960.6122976861288296·𝓉

4 5
GF5→1;1235
−747095.8262283975179056·𝓉 6 −6.1530993935738919×107 ·𝓉 7
−4.7466414548441339×109 ·𝓉 8 +𝒪(𝓉 9 ). (B.31)

Rank 0 1 2
Result 0. 0. 0.
Rank 3 4 5
Result 0. -0.4647775732011648 -76.65585039332824
Rank 6 7 8
Result -7394.867372623504 -504749.27267807984 −2.9542488797636 × 107

Table 25. Reduction Result of Pentagon to 𝐷 1 Tadpole by FIRE6.

The numerical Taylor expansion of the generating function is given as follows:

d (𝓉) = − 0.46477757320116485 · 𝓉 − 76.65585039332823 · 𝓉

4 5
GF5→1;2345
− 7394.8673726235035272 · 𝓉 6 − 504749.27267807984127 · 𝓉 7
− 2.95424887976361 × 107 · 𝓉 8 + 𝒪(𝓉 9 ). (B.32)

Open Access. This article is distributed under the terms of the Creative Commons
Attribution License (CC-BY4.0), which permits any use, distribution and reproduction in
any medium, provided the original author(s) and source are credited.

– 40 –
References

[1] NLO Multileg Working Group collaboration, The NLO multileg working group: Summary
report, in the proceedings of the 5th Les Houches Workshop on Physics at TeV Colliders, Les
Houches, France, June 11–29 (2007), p. 1–120 [arXiv:0803.0494] [INSPIRE].
[2] Z. Bern, J.J.M. Carrasco and H. Johansson, New Relations for Gauge-Theory Amplitudes, Phys.
Rev. D 78 (2008) 085011 [arXiv:0805.3993] [INSPIRE].
[3] M. Chiodaroli, M. Günaydin, H. Johansson and R. Roiban, Scattering amplitudes in 𝒩 = 2
Maxwell-Einstein and Yang-Mills/Einstein supergravity, JHEP 01 (2015) 081
[arXiv:1408.0764] [INSPIRE].

JHEP02(2024)158
[4] H. Johansson and A. Ochirov, Color-Kinematics Duality for QCD Amplitudes, JHEP 01 (2016)
170 [arXiv:1507.00332] [INSPIRE].
[5] H. Johansson and A. Ochirov, Double copy for massive quantum particles with spin, JHEP 09
(2019) 040 [arXiv:1906.12292] [INSPIRE].
[6] H. Kawai, D.C. Lewellen and S.H.H. Tye, A Relation Between Tree Amplitudes of Closed and
Open Strings, Nucl. Phys. B 269 (1986) 1 [INSPIRE].
[7] R. Britto, F. Cachazo and B. Feng, New recursion relations for tree amplitudes of gluons, Nucl.
Phys. B 715 (2005) 499 [hep-th/0412308] [INSPIRE].
[8] R. Britto, F. Cachazo, B. Feng and E. Witten, Direct proof of tree-level recursion relation in
Yang-Mills theory, Phys. Rev. Lett. 94 (2005) 181602 [hep-th/0501052] [INSPIRE].
[9] F. Cachazo, S. He and E.Y. Yuan, Scattering equations and Kawai-Lewellen-Tye orthogonality,
Phys. Rev. D 90 (2014) 065001 [arXiv:1306.6575] [INSPIRE].
[10] F. Cachazo, S. He and E.Y. Yuan, Scattering of Massless Particles in Arbitrary Dimensions,
Phys. Rev. Lett. 113 (2014) 171601 [arXiv:1307.2199] [INSPIRE].
[11] F. Cachazo, S. He and E.Y. Yuan, Scattering of Massless Particles: Scalars, Gluons and
Gravitons, JHEP 07 (2014) 033 [arXiv:1309.0885] [INSPIRE].
[12] F. Cachazo, S. He and E.Y. Yuan, Einstein-Yang-Mills Scattering Amplitudes From Scattering
Equations, JHEP 01 (2015) 121 [arXiv:1409.8256] [INSPIRE].
[13] F. Cachazo, S. He and E.Y. Yuan, Scattering Equations and Matrices: From Einstein To
Yang-Mills, DBI and NLSM, JHEP 07 (2015) 149 [arXiv:1412.3479] [INSPIRE].
[14] E. Witten, Perturbative gauge theory as a string theory in twistor space, Commun. Math. Phys.
252 (2004) 189 [hep-th/0312171] [INSPIRE].
[15] J.M. Henn and J.C. Plefka, Scattering Amplitudes in Gauge Theories, Springer (2014)
[DOI:10.1007/978-3-642-54022-6].
[16] H. Elvang and Y.-T. Huang, Scattering Amplitudes in Gauge Theory and Gravity, Cambridge
University Press (2015) [DOI:10.1017/cbo9781107706620].
[17] Z. Bern, L.J. Dixon, D.C. Dunbar and D.A. Kosower, One loop n point gauge theory amplitudes,
unitarity and collinear limits, Nucl. Phys. B 425 (1994) 217 [hep-ph/9403226] [INSPIRE].
[18] Z. Bern, L.J. Dixon, D.C. Dunbar and D.A. Kosower, Fusing gauge theory tree amplitudes into
loop amplitudes, Nucl. Phys. B 435 (1995) 59 [hep-ph/9409265] [INSPIRE].
[19] F. Cachazo, P. Svrcek and E. Witten, MHV vertices and tree amplitudes in gauge theory, JHEP
09 (2004) 006 [hep-th/0403047] [INSPIRE].

– 41 –
[20] R. Britto, F. Cachazo and B. Feng, Generalized unitarity and one-loop amplitudes in 𝑁 = 4
super-Yang-Mills, Nucl. Phys. B 725 (2005) 275 [hep-th/0412103] [INSPIRE].
[21] G. Ossola, C.G. Papadopoulos and R. Pittau, Reducing full one-loop amplitudes to scalar
integrals at the integrand level, Nucl. Phys. B 763 (2007) 147 [hep-ph/0609007] [INSPIRE].
[22] R.K. Ellis, Z. Kunszt, K. Melnikov and G. Zanderighi, One-loop calculations in quantum field
theory: from Feynman diagrams to unitarity cuts, Phys. Rept. 518 (2012) 141
[arXiv:1105.4319] [INSPIRE].
[23] R. Britto, E. Buchbinder, F. Cachazo and B. Feng, One-loop amplitudes of gluons in SQCD,
Phys. Rev. D 72 (2005) 065012 [hep-ph/0503132] [INSPIRE].

JHEP02(2024)158
[24] L. Mason and D. Skinner, Ambitwistor strings and the scattering equations, JHEP 07 (2014)
048 [arXiv:1311.2564] [INSPIRE].
[25] T. Adamo, E. Casali and D. Skinner, Ambitwistor strings and the scattering equations at one
loop, JHEP 04 (2014) 104 [arXiv:1312.3828] [INSPIRE].
[26] T. Adamo and E. Casali, Scattering equations, supergravity integrands, and pure spinors, JHEP
05 (2015) 120 [arXiv:1502.06826] [INSPIRE].
[27] E. Casali and P. Tourkine, Infrared behaviour of the one-loop scattering equations and
supergravity integrands, JHEP 04 (2015) 013 [arXiv:1412.3787] [INSPIRE].
[28] Y. Geyer, L. Mason, R. Monteiro and P. Tourkine, Loop Integrands for Scattering Amplitudes
from the Riemann Sphere, Phys. Rev. Lett. 115 (2015) 121603 [arXiv:1507.00321] [INSPIRE].
[29] Y. Geyer, L. Mason, R. Monteiro and P. Tourkine, One-loop amplitudes on the Riemann sphere,
JHEP 03 (2016) 114 [arXiv:1511.06315] [INSPIRE].
[30] Y. Geyer and R. Monteiro, Gluons and gravitons at one loop from ambitwistor strings, JHEP
03 (2018) 068 [arXiv:1711.09923] [INSPIRE].
[31] S. He and E.Y. Yuan, One-loop Scattering Equations and Amplitudes from Forward Limit, Phys.
Rev. D 92 (2015) 105004 [arXiv:1508.06027] [INSPIRE].
[32] F. Cachazo, S. He and E.Y. Yuan, One-Loop Corrections from Higher Dimensional Tree
Amplitudes, JHEP 08 (2016) 008 [arXiv:1512.05001] [INSPIRE].
[33] B. Feng and C. Hu, One-loop CHY-Integrand of Bi-adjoint Scalar Theory, JHEP 02 (2020) 187
[arXiv:1912.12960] [INSPIRE].
[34] S. Weinzierl, Feynman Integrals, arXiv:2201.03593 [DOI:10.1007/978-3-030-99558-4]
[INSPIRE].
[35] P. Mastrolia and G. Ossola, On the Integrand-Reduction Method for Two-Loop Scattering
Amplitudes, JHEP 11 (2011) 014 [arXiv:1107.6041] [INSPIRE].
[36] S. Badger, H. Frellesvig and Y. Zhang, Hepta-Cuts of Two-Loop Scattering Amplitudes, JHEP
04 (2012) 055 [arXiv:1202.2019] [INSPIRE].
[37] Y. Zhang, Integrand-Level Reduction of Loop Amplitudes by Computational Algebraic Geometry
Methods, JHEP 09 (2012) 042 [arXiv:1205.5707] [INSPIRE].
[38] K.J. Larsen and Y. Zhang, Integration-by-parts reductions from unitarity cuts and algebraic
geometry, Phys. Rev. D 93 (2016) 041701 [arXiv:1511.01071] [INSPIRE].
[39] Y. Zhang, Lecture Notes on Multi-loop Integral Reduction and Applied Algebraic Geometry,
arXiv:1612.02249 [INSPIRE].

– 42 –
[40] G. Passarino and M.J.G. Veltman, One Loop Corrections for 𝑒 +𝑒 − Annihilation Into 𝜇 + 𝜇 − in
the Weinberg Model, Nucl. Phys. B 160 (1979) 151 [INSPIRE].
[41] R.M. Schabinger, A New Algorithm For The Generation Of Unitarity-Compatible Integration
By Parts Relations, JHEP 01 (2012) 077 [arXiv:1111.4220] [INSPIRE].
[42] K.G. Chetyrkin and F.V. Tkachov, Integration by parts: The algorithm to calculate 𝛽-functions
in 4 loops, Nucl. Phys. B 192 (1981) 159 [INSPIRE].
[43] F.V. Tkachov, A theorem on analytical calculability of 4-loop renormalization group functions,
Phys. Lett. B 100 (1981) 65 [INSPIRE].
[44] S. Laporta, High-precision calculation of multiloop Feynman integrals by difference equations,
Int. J. Mod. Phys. A 15 (2000) 5087 [hep-ph/0102033] [INSPIRE].

JHEP02(2024)158
[45] A. von Manteuffel and C. Studerus, Reduze 2 — Distributed Feynman Integral Reduction,
arXiv:1201.4330 [INSPIRE].
[46] A. von Manteuffel and R.M. Schabinger, A novel approach to integration by parts reduction,
Phys. Lett. B 744 (2015) 101 [arXiv:1406.4513] [INSPIRE].
[47] P. Maierhöfer, J. Usovitsch and P. Uwer, Kira — A Feynman integral reduction program,
Comput. Phys. Commun. 230 (2018) 99 [arXiv:1705.05610] [INSPIRE].
[48] A.V. Smirnov and F.S. Chuharev, FIRE6: Feynman Integral REduction with Modular
Arithmetic, Comput. Phys. Commun. 247 (2020) 106877 [arXiv:1901.07808] [INSPIRE].
[49] R. Britto, B. Feng and P. Mastrolia, The Cut-constructible part of QCD amplitudes, Phys. Rev.
D 73 (2006) 105004 [hep-ph/0602178] [INSPIRE].
[50] C. Anastasiou et al., D-dimensional unitarity cut method, Phys. Lett. B 645 (2007) 213
[hep-ph/0609191] [INSPIRE].
[51] C. Anastasiou et al., Unitarity cuts and Reduction to master integrals in d dimensions for
one-loop amplitudes, JHEP 03 (2007) 111 [hep-ph/0612277] [INSPIRE].
[52] R. Britto and B. Feng, Unitarity cuts with massive propagators and algebraic expressions for
coefficients, Phys. Rev. D 75 (2007) 105006 [hep-ph/0612089] [INSPIRE].
[53] R. Britto and B. Feng, Integral coefficients for one-loop amplitudes, JHEP 02 (2008) 095
[arXiv:0711.4284] [INSPIRE].
[54] R. Britto and E. Mirabella, Single Cut Integration, JHEP 01 (2011) 135 [arXiv:1011.2344]
[INSPIRE].
[55] P. Mastrolia and S. Mizera, Feynman Integrals and Intersection Theory, JHEP 02 (2019) 139
[arXiv:1810.03818] [INSPIRE].
[56] H. Frellesvig et al., Vector Space of Feynman Integrals and Multivariate Intersection Numbers,
Phys. Rev. Lett. 123 (2019) 201602 [arXiv:1907.02000] [INSPIRE].
[57] S. Mizera and A. Pokraka, From Infinity to Four Dimensions: Higher Residue Pairings and
Feynman Integrals, JHEP 02 (2020) 159 [arXiv:1910.11852] [INSPIRE].
[58] H. Frellesvig et al., Decomposition of Feynman Integrals by Multivariate Intersection Numbers,
JHEP 03 (2021) 027 [arXiv:2008.04823] [INSPIRE].
[59] S. Caron-Huot and A. Pokraka, Duals of Feynman integrals. Part I. Differential equations,
JHEP 12 (2021) 045 [arXiv:2104.06898] [INSPIRE].
[60] S. Caron-Huot and A. Pokraka, Duals of Feynman Integrals. Part II. Generalized unitarity,
JHEP 04 (2022) 078 [arXiv:2112.00055] [INSPIRE].

– 43 –
[61] J. Chen, B. Feng and L.L. Yang, Intersection theory rules symbology, Sci. China Phys. Mech.
Astron. 67 (2024) 221011 [arXiv:2305.01283] [INSPIRE].
[62] Z. Wu et al., NeatIBP 1.0, a package generating small-size integration-by-parts relations for
Feynman integrals, Comput. Phys. Commun. 295 (2024) 108999 [arXiv:2305.08783] [INSPIRE].
[63] T. Peraro, FiniteFlow: multivariate functional reconstruction using finite fields and dataflow
graphs, JHEP 07 (2019) 031 [arXiv:1905.08019] [INSPIRE].
[64] J. Klappert and F. Lange, Reconstructing rational functions with FireFly, Comput. Phys.
Commun. 247 (2020) 106951 [arXiv:1904.00009] [INSPIRE].
[65] J. Klappert, S.Y. Klein and F. Lange, Interpolation of dense and sparse rational functions and
other improvements in FireFly, Comput. Phys. Commun. 264 (2021) 107968

JHEP02(2024)158
[arXiv:2004.01463] [INSPIRE].
[66] A.V. Belitsky, A.V. Smirnov and R.V. Yakovlev, Balancing act: Multivariate rational
reconstruction for IBP, Nucl. Phys. B 993 (2023) 116253 [arXiv:2303.02511] [INSPIRE].
[67] X. Liu and Y.-Q. Ma, Determining arbitrary Feynman integrals by vacuum integrals, Phys. Rev.
D 99 (2019) 071501 [arXiv:1801.10523] [INSPIRE].
[68] X. Guan, X. Liu and Y.-Q. Ma, Complete reduction of integrals in two-loop five-light-parton
scattering amplitudes, Chin. Phys. C 44 (2020) 093106 [arXiv:1912.09294] [INSPIRE].
[69] Blade, https://gitlab.com/multiloop-pku/blade.
[70] A.V. Smirnov and V.A. Smirnov, How to choose master integrals, Nucl. Phys. B 960 (2020)
115213 [arXiv:2002.08042] [INSPIRE].
[71] J. Usovitsch, Factorization of denominators in integration-by-parts reductions,
arXiv:2002.08173 [INSPIRE].
[72] C. Anastasiou and A. Lazopoulos, Automatic integral reduction for higher order perturbative
calculations, JHEP 07 (2004) 046 [hep-ph/0404258] [INSPIRE].
[73] A.V. Smirnov, Algorithm FIRE — Feynman Integral REduction, JHEP 10 (2008) 107
[arXiv:0807.3243] [INSPIRE].
[74] A.V. Smirnov and V.A. Smirnov, FIRE4, LiteRed and accompanying tools to solve integration
by parts relations, Comput. Phys. Commun. 184 (2013) 2820 [arXiv:1302.5885] [INSPIRE].
[75] A.V. Smirnov, FIRE5: a C++ implementation of Feynman Integral REduction, Comput. Phys.
Commun. 189 (2015) 182 [arXiv:1408.2372] [INSPIRE].
[76] P. Maierhöfer and J. Usovitsch, Kira 1.2 Release Notes, arXiv:1812.01491 [INSPIRE].
[77] J. Klappert, F. Lange, P. Maierhöfer and J. Usovitsch, Integral reduction with Kira 2.0 and
finite field methods, Comput. Phys. Commun. 266 (2021) 108024 [arXiv:2008.06494]
[INSPIRE].
[78] R.N. Lee, Presenting LiteRed: a tool for the Loop InTEgrals REDuction, arXiv:1212.2685
[INSPIRE].
[79] R.N. Lee, LiteRed 1.4: a powerful tool for reduction of multiloop integrals, J. Phys. Conf. Ser.
523 (2014) 012059 [arXiv:1310.1145] [INSPIRE].
[80] C. Studerus, Reduze-Feynman Integral Reduction in C++, Comput. Phys. Commun. 181 (2010)
1293 [arXiv:0912.2546] [INSPIRE].
[81] A.V. Smirnov and M.N. Tentyukov, Feynman Integral Evaluation by a Sector decomposiTion
Approach (FIESTA), Comput. Phys. Commun. 180 (2009) 735 [arXiv:0807.4129] [INSPIRE].

– 44 –
 [82] A.V. Smirnov, V.A. Smirnov and M. Tentyukov, FIESTA 2: Parallelizeable multiloop numerical
calculations, Comput. Phys. Commun. 182 (2011) 790 [arXiv:0912.0158] [INSPIRE].
[83] A.V. Smirnov, FIESTA 3: cluster-parallelizable multiloop numerical calculations in physical
regions, Comput. Phys. Commun. 185 (2014) 2090 [arXiv:1312.3186] [INSPIRE].
[84] A.V. Smirnov, FIESTA4: Optimized Feynman integral calculations with GPU support, Comput.
Phys. Commun. 204 (2016) 189 [arXiv:1511.03614] [INSPIRE].
[85] A.V. Smirnov, N.D. Shapurov and L.I. Vysotsky, FIESTA5: Numerical high-performance
Feynman integral evaluation, Comput. Phys. Commun. 277 (2022) 108386 [arXiv:2110.11660]
[INSPIRE].

JHEP02(2024)158
[86] B. Feng, T. Li and X. Li, Analytic tadpole coefficients of one-loop integrals, JHEP 09 (2021)
081 [arXiv:2107.03744] [INSPIRE].
[87] C. Hu, T. Li and X. Li, One-loop Feynman integral reduction by differential operators, Phys.
Rev. D 104 (2021) 116014 [arXiv:2108.00772] [INSPIRE].
[88] B. Feng, T. Li, H. Wang and Y. Zhang, Reduction of general one-loop integrals using auxiliary
vector, JHEP 05 (2022) 065 [arXiv:2203.14449] [INSPIRE].
[89] B. Feng and T. Li, PV-reduction of sunset topology with auxiliary vector, Commun. Theor.
Phys. 74 (2022) 095201 [arXiv:2203.16881] [INSPIRE].
[90] B. Feng, C. Hu, T. Li and Y. Song, Reduction with degenerate Gram matrix for one-loop
integrals, JHEP 08 (2022) 110 [arXiv:2205.03000] [INSPIRE].
[91] J. Chen and B. Feng, Module intersection and uniform formula for iterative reduction of
one-loop integrals, JHEP 02 (2023) 178 [arXiv:2207.03767] [INSPIRE].
[92] J. Chen, Iteratively Reduce Auxiliary Scalar Product in Multi-loop Integrals, arXiv:2208.14693
[INSPIRE].
[93] J. Chen et al., Baikov representations, intersection theory, and canonical Feynman integrals,
JHEP 07 (2022) 066 [arXiv:2202.08127] [INSPIRE].
[94] J. Ablinger et al., Calculating Massive 3-loop Graphs for Operator Matrix Elements by the
Method of Hyperlogarithms, Nucl. Phys. B 885 (2014) 409 [arXiv:1403.1137] [INSPIRE].
[95] D.A. Kosower, Direct Solution of Integration-by-Parts Systems, Phys. Rev. D 98 (2018) 025008
[arXiv:1804.00131] [INSPIRE].
[96] A. Behring, J. Blümlein and K. Schönwald, The inverse Mellin transform via analytic
continuation, JHEP 06 (2023) 062 [arXiv:2303.05943] [INSPIRE].
[97] B. Feng, Generation function for one-loop tensor reduction, Commun. Theor. Phys. 75 (2023)
025203 [arXiv:2209.09517] [INSPIRE].
[98] X. Guan, X. Li and Y.-Q. Ma, Exploring the linear space of Feynman integrals via generating
functions, Phys. Rev. D 108 (2023) 034027 [arXiv:2306.02927] [INSPIRE].
[99] B. Feng, J. Gong and T. Li, Universal treatment of the reduction for one-loop integrals in a
projective space, Phys. Rev. D 106 (2022) 056025 [arXiv:2204.03190] [INSPIRE].
[100] T. Li, Nontrivial one-loop recursive reduction relation, JHEP 07 (2023) 051
[arXiv:2209.11428] [INSPIRE].

– 45 –